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/mm.h>
26#include <linux/proc_fs.h>
27#include <linux/tty.h>
28#include <linux/binfmts.h>
29#include <linux/coredump.h>
30#include <linux/security.h>
31#include <linux/syscalls.h>
32#include <linux/ptrace.h>
33#include <linux/signal.h>
34#include <linux/signalfd.h>
35#include <linux/ratelimit.h>
36#include <linux/task_work.h>
37#include <linux/capability.h>
38#include <linux/freezer.h>
39#include <linux/pid_namespace.h>
40#include <linux/nsproxy.h>
41#include <linux/user_namespace.h>
42#include <linux/uprobes.h>
43#include <linux/compat.h>
44#include <linux/cn_proc.h>
45#include <linux/compiler.h>
46#include <linux/posix-timers.h>
47#include <linux/cgroup.h>
48#include <linux/audit.h>
49#include <linux/sysctl.h>
50#include <uapi/linux/pidfd.h>
51
52#define CREATE_TRACE_POINTS
53#include <trace/events/signal.h>
54
55#include <asm/param.h>
56#include <linux/uaccess.h>
57#include <asm/unistd.h>
58#include <asm/siginfo.h>
59#include <asm/cacheflush.h>
60#include <asm/syscall.h> /* for syscall_get_* */
61
62#include "time/posix-timers.h"
63
64/*
65 * SLAB caches for signal bits.
66 */
67
68static struct kmem_cache *sigqueue_cachep;
69
70int print_fatal_signals __read_mostly;
71
72static void __user *sig_handler(struct task_struct *t, int sig)
73{
74 return t->sighand->action[sig - 1].sa.sa_handler;
75}
76
77static inline bool sig_handler_ignored(void __user *handler, int sig)
78{
79 /* Is it explicitly or implicitly ignored? */
80 return handler == SIG_IGN ||
81 (handler == SIG_DFL && sig_kernel_ignore(sig));
82}
83
84static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
85{
86 void __user *handler;
87
88 handler = sig_handler(t, sig);
89
90 /* SIGKILL and SIGSTOP may not be sent to the global init */
91 if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
92 return true;
93
94 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
95 handler == SIG_DFL && !(force && sig_kernel_only(sig)))
96 return true;
97
98 /* Only allow kernel generated signals to this kthread */
99 if (unlikely((t->flags & PF_KTHREAD) &&
100 (handler == SIG_KTHREAD_KERNEL) && !force))
101 return true;
102
103 return sig_handler_ignored(handler, sig);
104}
105
106static bool sig_ignored(struct task_struct *t, int sig, bool force)
107{
108 /*
109 * Blocked signals are never ignored, since the
110 * signal handler may change by the time it is
111 * unblocked.
112 */
113 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
114 return false;
115
116 /*
117 * Tracers may want to know about even ignored signal unless it
118 * is SIGKILL which can't be reported anyway but can be ignored
119 * by SIGNAL_UNKILLABLE task.
120 */
121 if (t->ptrace && sig != SIGKILL)
122 return false;
123
124 return sig_task_ignored(t, sig, force);
125}
126
127/*
128 * Re-calculate pending state from the set of locally pending
129 * signals, globally pending signals, and blocked signals.
130 */
131static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
132{
133 unsigned long ready;
134 long i;
135
136 switch (_NSIG_WORDS) {
137 default:
138 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
139 ready |= signal->sig[i] &~ blocked->sig[i];
140 break;
141
142 case 4: ready = signal->sig[3] &~ blocked->sig[3];
143 ready |= signal->sig[2] &~ blocked->sig[2];
144 ready |= signal->sig[1] &~ blocked->sig[1];
145 ready |= signal->sig[0] &~ blocked->sig[0];
146 break;
147
148 case 2: ready = signal->sig[1] &~ blocked->sig[1];
149 ready |= signal->sig[0] &~ blocked->sig[0];
150 break;
151
152 case 1: ready = signal->sig[0] &~ blocked->sig[0];
153 }
154 return ready != 0;
155}
156
157#define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
158
159static bool recalc_sigpending_tsk(struct task_struct *t)
160{
161 if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) ||
162 PENDING(&t->pending, &t->blocked) ||
163 PENDING(&t->signal->shared_pending, &t->blocked) ||
164 cgroup_task_frozen(t)) {
165 set_tsk_thread_flag(t, TIF_SIGPENDING);
166 return true;
167 }
168
169 /*
170 * We must never clear the flag in another thread, or in current
171 * when it's possible the current syscall is returning -ERESTART*.
172 * So we don't clear it here, and only callers who know they should do.
173 */
174 return false;
175}
176
177void recalc_sigpending(void)
178{
179 if (!recalc_sigpending_tsk(current) && !freezing(current))
180 clear_thread_flag(TIF_SIGPENDING);
181
182}
183EXPORT_SYMBOL(recalc_sigpending);
184
185void calculate_sigpending(void)
186{
187 /* Have any signals or users of TIF_SIGPENDING been delayed
188 * until after fork?
189 */
190 spin_lock_irq(¤t->sighand->siglock);
191 set_tsk_thread_flag(current, TIF_SIGPENDING);
192 recalc_sigpending();
193 spin_unlock_irq(¤t->sighand->siglock);
194}
195
196/* Given the mask, find the first available signal that should be serviced. */
197
198#define SYNCHRONOUS_MASK \
199 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
200 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
201
202int next_signal(struct sigpending *pending, sigset_t *mask)
203{
204 unsigned long i, *s, *m, x;
205 int sig = 0;
206
207 s = pending->signal.sig;
208 m = mask->sig;
209
210 /*
211 * Handle the first word specially: it contains the
212 * synchronous signals that need to be dequeued first.
213 */
214 x = *s &~ *m;
215 if (x) {
216 if (x & SYNCHRONOUS_MASK)
217 x &= SYNCHRONOUS_MASK;
218 sig = ffz(~x) + 1;
219 return sig;
220 }
221
222 switch (_NSIG_WORDS) {
223 default:
224 for (i = 1; i < _NSIG_WORDS; ++i) {
225 x = *++s &~ *++m;
226 if (!x)
227 continue;
228 sig = ffz(~x) + i*_NSIG_BPW + 1;
229 break;
230 }
231 break;
232
233 case 2:
234 x = s[1] &~ m[1];
235 if (!x)
236 break;
237 sig = ffz(~x) + _NSIG_BPW + 1;
238 break;
239
240 case 1:
241 /* Nothing to do */
242 break;
243 }
244
245 return sig;
246}
247
248static inline void print_dropped_signal(int sig)
249{
250 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
251
252 if (!print_fatal_signals)
253 return;
254
255 if (!__ratelimit(&ratelimit_state))
256 return;
257
258 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
259 current->comm, current->pid, sig);
260}
261
262/**
263 * task_set_jobctl_pending - set jobctl pending bits
264 * @task: target task
265 * @mask: pending bits to set
266 *
267 * Clear @mask from @task->jobctl. @mask must be subset of
268 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
269 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
270 * cleared. If @task is already being killed or exiting, this function
271 * becomes noop.
272 *
273 * CONTEXT:
274 * Must be called with @task->sighand->siglock held.
275 *
276 * RETURNS:
277 * %true if @mask is set, %false if made noop because @task was dying.
278 */
279bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
280{
281 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
282 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
283 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
284
285 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
286 return false;
287
288 if (mask & JOBCTL_STOP_SIGMASK)
289 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
290
291 task->jobctl |= mask;
292 return true;
293}
294
295/**
296 * task_clear_jobctl_trapping - clear jobctl trapping bit
297 * @task: target task
298 *
299 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
300 * Clear it and wake up the ptracer. Note that we don't need any further
301 * locking. @task->siglock guarantees that @task->parent points to the
302 * ptracer.
303 *
304 * CONTEXT:
305 * Must be called with @task->sighand->siglock held.
306 */
307void task_clear_jobctl_trapping(struct task_struct *task)
308{
309 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
310 task->jobctl &= ~JOBCTL_TRAPPING;
311 smp_mb(); /* advised by wake_up_bit() */
312 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
313 }
314}
315
316/**
317 * task_clear_jobctl_pending - clear jobctl pending bits
318 * @task: target task
319 * @mask: pending bits to clear
320 *
321 * Clear @mask from @task->jobctl. @mask must be subset of
322 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
323 * STOP bits are cleared together.
324 *
325 * If clearing of @mask leaves no stop or trap pending, this function calls
326 * task_clear_jobctl_trapping().
327 *
328 * CONTEXT:
329 * Must be called with @task->sighand->siglock held.
330 */
331void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
332{
333 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
334
335 if (mask & JOBCTL_STOP_PENDING)
336 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
337
338 task->jobctl &= ~mask;
339
340 if (!(task->jobctl & JOBCTL_PENDING_MASK))
341 task_clear_jobctl_trapping(task);
342}
343
344/**
345 * task_participate_group_stop - participate in a group stop
346 * @task: task participating in a group stop
347 *
348 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
349 * Group stop states are cleared and the group stop count is consumed if
350 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
351 * stop, the appropriate `SIGNAL_*` flags are set.
352 *
353 * CONTEXT:
354 * Must be called with @task->sighand->siglock held.
355 *
356 * RETURNS:
357 * %true if group stop completion should be notified to the parent, %false
358 * otherwise.
359 */
360static bool task_participate_group_stop(struct task_struct *task)
361{
362 struct signal_struct *sig = task->signal;
363 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
364
365 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
366
367 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
368
369 if (!consume)
370 return false;
371
372 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
373 sig->group_stop_count--;
374
375 /*
376 * Tell the caller to notify completion iff we are entering into a
377 * fresh group stop. Read comment in do_signal_stop() for details.
378 */
379 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
380 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
381 return true;
382 }
383 return false;
384}
385
386void task_join_group_stop(struct task_struct *task)
387{
388 unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK;
389 struct signal_struct *sig = current->signal;
390
391 if (sig->group_stop_count) {
392 sig->group_stop_count++;
393 mask |= JOBCTL_STOP_CONSUME;
394 } else if (!(sig->flags & SIGNAL_STOP_STOPPED))
395 return;
396
397 /* Have the new thread join an on-going signal group stop */
398 task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING);
399}
400
401static struct ucounts *sig_get_ucounts(struct task_struct *t, int sig,
402 int override_rlimit)
403{
404 struct ucounts *ucounts;
405 long sigpending;
406
407 /*
408 * Protect access to @t credentials. This can go away when all
409 * callers hold rcu read lock.
410 *
411 * NOTE! A pending signal will hold on to the user refcount,
412 * and we get/put the refcount only when the sigpending count
413 * changes from/to zero.
414 */
415 rcu_read_lock();
416 ucounts = task_ucounts(t);
417 sigpending = inc_rlimit_get_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING,
418 override_rlimit);
419 rcu_read_unlock();
420 if (!sigpending)
421 return NULL;
422
423 if (unlikely(!override_rlimit && sigpending > task_rlimit(t, RLIMIT_SIGPENDING))) {
424 dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
425 print_dropped_signal(sig);
426 return NULL;
427 }
428
429 return ucounts;
430}
431
432static void __sigqueue_init(struct sigqueue *q, struct ucounts *ucounts,
433 const unsigned int sigqueue_flags)
434{
435 INIT_LIST_HEAD(&q->list);
436 q->flags = sigqueue_flags;
437 q->ucounts = ucounts;
438}
439
440/*
441 * allocate a new signal queue record
442 * - this may be called without locks if and only if t == current, otherwise an
443 * appropriate lock must be held to stop the target task from exiting
444 */
445static struct sigqueue *sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
446 int override_rlimit)
447{
448 struct ucounts *ucounts = sig_get_ucounts(t, sig, override_rlimit);
449 struct sigqueue *q;
450
451 if (!ucounts)
452 return NULL;
453
454 q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
455 if (!q) {
456 dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
457 return NULL;
458 }
459
460 __sigqueue_init(q, ucounts, 0);
461 return q;
462}
463
464static void __sigqueue_free(struct sigqueue *q)
465{
466 if (q->flags & SIGQUEUE_PREALLOC) {
467 posixtimer_sigqueue_putref(q);
468 return;
469 }
470 if (q->ucounts) {
471 dec_rlimit_put_ucounts(q->ucounts, UCOUNT_RLIMIT_SIGPENDING);
472 q->ucounts = NULL;
473 }
474 kmem_cache_free(sigqueue_cachep, q);
475}
476
477void flush_sigqueue(struct sigpending *queue)
478{
479 struct sigqueue *q;
480
481 sigemptyset(&queue->signal);
482 while (!list_empty(&queue->list)) {
483 q = list_entry(queue->list.next, struct sigqueue , list);
484 list_del_init(&q->list);
485 __sigqueue_free(q);
486 }
487}
488
489/*
490 * Flush all pending signals for this kthread.
491 */
492void flush_signals(struct task_struct *t)
493{
494 unsigned long flags;
495
496 spin_lock_irqsave(&t->sighand->siglock, flags);
497 clear_tsk_thread_flag(t, TIF_SIGPENDING);
498 flush_sigqueue(&t->pending);
499 flush_sigqueue(&t->signal->shared_pending);
500 spin_unlock_irqrestore(&t->sighand->siglock, flags);
501}
502EXPORT_SYMBOL(flush_signals);
503
504void ignore_signals(struct task_struct *t)
505{
506 int i;
507
508 for (i = 0; i < _NSIG; ++i)
509 t->sighand->action[i].sa.sa_handler = SIG_IGN;
510
511 flush_signals(t);
512}
513
514/*
515 * Flush all handlers for a task.
516 */
517
518void
519flush_signal_handlers(struct task_struct *t, int force_default)
520{
521 int i;
522 struct k_sigaction *ka = &t->sighand->action[0];
523 for (i = _NSIG ; i != 0 ; i--) {
524 if (force_default || ka->sa.sa_handler != SIG_IGN)
525 ka->sa.sa_handler = SIG_DFL;
526 ka->sa.sa_flags = 0;
527#ifdef __ARCH_HAS_SA_RESTORER
528 ka->sa.sa_restorer = NULL;
529#endif
530 sigemptyset(&ka->sa.sa_mask);
531 ka++;
532 }
533}
534
535bool unhandled_signal(struct task_struct *tsk, int sig)
536{
537 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
538 if (is_global_init(tsk))
539 return true;
540
541 if (handler != SIG_IGN && handler != SIG_DFL)
542 return false;
543
544 /* If dying, we handle all new signals by ignoring them */
545 if (fatal_signal_pending(tsk))
546 return false;
547
548 /* if ptraced, let the tracer determine */
549 return !tsk->ptrace;
550}
551
552static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
553 struct sigqueue **timer_sigq)
554{
555 struct sigqueue *q, *first = NULL;
556
557 /*
558 * Collect the siginfo appropriate to this signal. Check if
559 * there is another siginfo for the same signal.
560 */
561 list_for_each_entry(q, &list->list, list) {
562 if (q->info.si_signo == sig) {
563 if (first)
564 goto still_pending;
565 first = q;
566 }
567 }
568
569 sigdelset(&list->signal, sig);
570
571 if (first) {
572still_pending:
573 list_del_init(&first->list);
574 copy_siginfo(info, &first->info);
575
576 /*
577 * posix-timer signals are preallocated and freed when the last
578 * reference count is dropped in posixtimer_deliver_signal() or
579 * immediately on timer deletion when the signal is not pending.
580 * Spare the extra round through __sigqueue_free() which is
581 * ignoring preallocated signals.
582 */
583 if (unlikely((first->flags & SIGQUEUE_PREALLOC) && (info->si_code == SI_TIMER)))
584 *timer_sigq = first;
585 else
586 __sigqueue_free(first);
587 } else {
588 /*
589 * Ok, it wasn't in the queue. This must be
590 * a fast-pathed signal or we must have been
591 * out of queue space. So zero out the info.
592 */
593 clear_siginfo(info);
594 info->si_signo = sig;
595 info->si_errno = 0;
596 info->si_code = SI_USER;
597 info->si_pid = 0;
598 info->si_uid = 0;
599 }
600}
601
602static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
603 kernel_siginfo_t *info, struct sigqueue **timer_sigq)
604{
605 int sig = next_signal(pending, mask);
606
607 if (sig)
608 collect_signal(sig, pending, info, timer_sigq);
609 return sig;
610}
611
612/*
613 * Try to dequeue a signal. If a deliverable signal is found fill in the
614 * caller provided siginfo and return the signal number. Otherwise return
615 * 0.
616 */
617int dequeue_signal(sigset_t *mask, kernel_siginfo_t *info, enum pid_type *type)
618{
619 struct task_struct *tsk = current;
620 struct sigqueue *timer_sigq;
621 int signr;
622
623 lockdep_assert_held(&tsk->sighand->siglock);
624
625again:
626 *type = PIDTYPE_PID;
627 timer_sigq = NULL;
628 signr = __dequeue_signal(&tsk->pending, mask, info, &timer_sigq);
629 if (!signr) {
630 *type = PIDTYPE_TGID;
631 signr = __dequeue_signal(&tsk->signal->shared_pending,
632 mask, info, &timer_sigq);
633
634 if (unlikely(signr == SIGALRM))
635 posixtimer_rearm_itimer(tsk);
636 }
637
638 recalc_sigpending();
639 if (!signr)
640 return 0;
641
642 if (unlikely(sig_kernel_stop(signr))) {
643 /*
644 * Set a marker that we have dequeued a stop signal. Our
645 * caller might release the siglock and then the pending
646 * stop signal it is about to process is no longer in the
647 * pending bitmasks, but must still be cleared by a SIGCONT
648 * (and overruled by a SIGKILL). So those cases clear this
649 * shared flag after we've set it. Note that this flag may
650 * remain set after the signal we return is ignored or
651 * handled. That doesn't matter because its only purpose
652 * is to alert stop-signal processing code when another
653 * processor has come along and cleared the flag.
654 */
655 current->jobctl |= JOBCTL_STOP_DEQUEUED;
656 }
657
658 if (IS_ENABLED(CONFIG_POSIX_TIMERS) && unlikely(timer_sigq)) {
659 if (!posixtimer_deliver_signal(info, timer_sigq))
660 goto again;
661 }
662
663 return signr;
664}
665EXPORT_SYMBOL_GPL(dequeue_signal);
666
667static int dequeue_synchronous_signal(kernel_siginfo_t *info)
668{
669 struct task_struct *tsk = current;
670 struct sigpending *pending = &tsk->pending;
671 struct sigqueue *q, *sync = NULL;
672
673 /*
674 * Might a synchronous signal be in the queue?
675 */
676 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
677 return 0;
678
679 /*
680 * Return the first synchronous signal in the queue.
681 */
682 list_for_each_entry(q, &pending->list, list) {
683 /* Synchronous signals have a positive si_code */
684 if ((q->info.si_code > SI_USER) &&
685 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
686 sync = q;
687 goto next;
688 }
689 }
690 return 0;
691next:
692 /*
693 * Check if there is another siginfo for the same signal.
694 */
695 list_for_each_entry_continue(q, &pending->list, list) {
696 if (q->info.si_signo == sync->info.si_signo)
697 goto still_pending;
698 }
699
700 sigdelset(&pending->signal, sync->info.si_signo);
701 recalc_sigpending();
702still_pending:
703 list_del_init(&sync->list);
704 copy_siginfo(info, &sync->info);
705 __sigqueue_free(sync);
706 return info->si_signo;
707}
708
709/*
710 * Tell a process that it has a new active signal..
711 *
712 * NOTE! we rely on the previous spin_lock to
713 * lock interrupts for us! We can only be called with
714 * "siglock" held, and the local interrupt must
715 * have been disabled when that got acquired!
716 *
717 * No need to set need_resched since signal event passing
718 * goes through ->blocked
719 */
720void signal_wake_up_state(struct task_struct *t, unsigned int state)
721{
722 lockdep_assert_held(&t->sighand->siglock);
723
724 set_tsk_thread_flag(t, TIF_SIGPENDING);
725
726 /*
727 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
728 * case. We don't check t->state here because there is a race with it
729 * executing another processor and just now entering stopped state.
730 * By using wake_up_state, we ensure the process will wake up and
731 * handle its death signal.
732 */
733 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
734 kick_process(t);
735}
736
737static inline void posixtimer_sig_ignore(struct task_struct *tsk, struct sigqueue *q);
738
739static void sigqueue_free_ignored(struct task_struct *tsk, struct sigqueue *q)
740{
741 if (likely(!(q->flags & SIGQUEUE_PREALLOC) || q->info.si_code != SI_TIMER))
742 __sigqueue_free(q);
743 else
744 posixtimer_sig_ignore(tsk, q);
745}
746
747/* Remove signals in mask from the pending set and queue. */
748static void flush_sigqueue_mask(struct task_struct *p, sigset_t *mask, struct sigpending *s)
749{
750 struct sigqueue *q, *n;
751 sigset_t m;
752
753 lockdep_assert_held(&p->sighand->siglock);
754
755 sigandsets(&m, mask, &s->signal);
756 if (sigisemptyset(&m))
757 return;
758
759 sigandnsets(&s->signal, &s->signal, mask);
760 list_for_each_entry_safe(q, n, &s->list, list) {
761 if (sigismember(mask, q->info.si_signo)) {
762 list_del_init(&q->list);
763 sigqueue_free_ignored(p, q);
764 }
765 }
766}
767
768static inline int is_si_special(const struct kernel_siginfo *info)
769{
770 return info <= SEND_SIG_PRIV;
771}
772
773static inline bool si_fromuser(const struct kernel_siginfo *info)
774{
775 return info == SEND_SIG_NOINFO ||
776 (!is_si_special(info) && SI_FROMUSER(info));
777}
778
779/*
780 * called with RCU read lock from check_kill_permission()
781 */
782static bool kill_ok_by_cred(struct task_struct *t)
783{
784 const struct cred *cred = current_cred();
785 const struct cred *tcred = __task_cred(t);
786
787 return uid_eq(cred->euid, tcred->suid) ||
788 uid_eq(cred->euid, tcred->uid) ||
789 uid_eq(cred->uid, tcred->suid) ||
790 uid_eq(cred->uid, tcred->uid) ||
791 ns_capable(tcred->user_ns, CAP_KILL);
792}
793
794/*
795 * Bad permissions for sending the signal
796 * - the caller must hold the RCU read lock
797 */
798static int check_kill_permission(int sig, struct kernel_siginfo *info,
799 struct task_struct *t)
800{
801 struct pid *sid;
802 int error;
803
804 if (!valid_signal(sig))
805 return -EINVAL;
806
807 if (!si_fromuser(info))
808 return 0;
809
810 error = audit_signal_info(sig, t); /* Let audit system see the signal */
811 if (error)
812 return error;
813
814 if (!same_thread_group(current, t) &&
815 !kill_ok_by_cred(t)) {
816 switch (sig) {
817 case SIGCONT:
818 sid = task_session(t);
819 /*
820 * We don't return the error if sid == NULL. The
821 * task was unhashed, the caller must notice this.
822 */
823 if (!sid || sid == task_session(current))
824 break;
825 fallthrough;
826 default:
827 return -EPERM;
828 }
829 }
830
831 return security_task_kill(t, info, sig, NULL);
832}
833
834/**
835 * ptrace_trap_notify - schedule trap to notify ptracer
836 * @t: tracee wanting to notify tracer
837 *
838 * This function schedules sticky ptrace trap which is cleared on the next
839 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
840 * ptracer.
841 *
842 * If @t is running, STOP trap will be taken. If trapped for STOP and
843 * ptracer is listening for events, tracee is woken up so that it can
844 * re-trap for the new event. If trapped otherwise, STOP trap will be
845 * eventually taken without returning to userland after the existing traps
846 * are finished by PTRACE_CONT.
847 *
848 * CONTEXT:
849 * Must be called with @task->sighand->siglock held.
850 */
851static void ptrace_trap_notify(struct task_struct *t)
852{
853 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
854 lockdep_assert_held(&t->sighand->siglock);
855
856 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
857 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
858}
859
860/*
861 * Handle magic process-wide effects of stop/continue signals. Unlike
862 * the signal actions, these happen immediately at signal-generation
863 * time regardless of blocking, ignoring, or handling. This does the
864 * actual continuing for SIGCONT, but not the actual stopping for stop
865 * signals. The process stop is done as a signal action for SIG_DFL.
866 *
867 * Returns true if the signal should be actually delivered, otherwise
868 * it should be dropped.
869 */
870static bool prepare_signal(int sig, struct task_struct *p, bool force)
871{
872 struct signal_struct *signal = p->signal;
873 struct task_struct *t;
874 sigset_t flush;
875
876 if (signal->flags & SIGNAL_GROUP_EXIT) {
877 if (signal->core_state)
878 return sig == SIGKILL;
879 /*
880 * The process is in the middle of dying, drop the signal.
881 */
882 return false;
883 } else if (sig_kernel_stop(sig)) {
884 /*
885 * This is a stop signal. Remove SIGCONT from all queues.
886 */
887 siginitset(&flush, sigmask(SIGCONT));
888 flush_sigqueue_mask(p, &flush, &signal->shared_pending);
889 for_each_thread(p, t)
890 flush_sigqueue_mask(p, &flush, &t->pending);
891 } else if (sig == SIGCONT) {
892 unsigned int why;
893 /*
894 * Remove all stop signals from all queues, wake all threads.
895 */
896 siginitset(&flush, SIG_KERNEL_STOP_MASK);
897 flush_sigqueue_mask(p, &flush, &signal->shared_pending);
898 for_each_thread(p, t) {
899 flush_sigqueue_mask(p, &flush, &t->pending);
900 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
901 if (likely(!(t->ptrace & PT_SEIZED))) {
902 t->jobctl &= ~JOBCTL_STOPPED;
903 wake_up_state(t, __TASK_STOPPED);
904 } else
905 ptrace_trap_notify(t);
906 }
907
908 /*
909 * Notify the parent with CLD_CONTINUED if we were stopped.
910 *
911 * If we were in the middle of a group stop, we pretend it
912 * was already finished, and then continued. Since SIGCHLD
913 * doesn't queue we report only CLD_STOPPED, as if the next
914 * CLD_CONTINUED was dropped.
915 */
916 why = 0;
917 if (signal->flags & SIGNAL_STOP_STOPPED)
918 why |= SIGNAL_CLD_CONTINUED;
919 else if (signal->group_stop_count)
920 why |= SIGNAL_CLD_STOPPED;
921
922 if (why) {
923 /*
924 * The first thread which returns from do_signal_stop()
925 * will take ->siglock, notice SIGNAL_CLD_MASK, and
926 * notify its parent. See get_signal().
927 */
928 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
929 signal->group_stop_count = 0;
930 signal->group_exit_code = 0;
931 }
932 }
933
934 return !sig_ignored(p, sig, force);
935}
936
937/*
938 * Test if P wants to take SIG. After we've checked all threads with this,
939 * it's equivalent to finding no threads not blocking SIG. Any threads not
940 * blocking SIG were ruled out because they are not running and already
941 * have pending signals. Such threads will dequeue from the shared queue
942 * as soon as they're available, so putting the signal on the shared queue
943 * will be equivalent to sending it to one such thread.
944 */
945static inline bool wants_signal(int sig, struct task_struct *p)
946{
947 if (sigismember(&p->blocked, sig))
948 return false;
949
950 if (p->flags & PF_EXITING)
951 return false;
952
953 if (sig == SIGKILL)
954 return true;
955
956 if (task_is_stopped_or_traced(p))
957 return false;
958
959 return task_curr(p) || !task_sigpending(p);
960}
961
962static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
963{
964 struct signal_struct *signal = p->signal;
965 struct task_struct *t;
966
967 /*
968 * Now find a thread we can wake up to take the signal off the queue.
969 *
970 * Try the suggested task first (may or may not be the main thread).
971 */
972 if (wants_signal(sig, p))
973 t = p;
974 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
975 /*
976 * There is just one thread and it does not need to be woken.
977 * It will dequeue unblocked signals before it runs again.
978 */
979 return;
980 else {
981 /*
982 * Otherwise try to find a suitable thread.
983 */
984 t = signal->curr_target;
985 while (!wants_signal(sig, t)) {
986 t = next_thread(t);
987 if (t == signal->curr_target)
988 /*
989 * No thread needs to be woken.
990 * Any eligible threads will see
991 * the signal in the queue soon.
992 */
993 return;
994 }
995 signal->curr_target = t;
996 }
997
998 /*
999 * Found a killable thread. If the signal will be fatal,
1000 * then start taking the whole group down immediately.
1001 */
1002 if (sig_fatal(p, sig) &&
1003 (signal->core_state || !(signal->flags & SIGNAL_GROUP_EXIT)) &&
1004 !sigismember(&t->real_blocked, sig) &&
1005 (sig == SIGKILL || !p->ptrace)) {
1006 /*
1007 * This signal will be fatal to the whole group.
1008 */
1009 if (!sig_kernel_coredump(sig)) {
1010 /*
1011 * Start a group exit and wake everybody up.
1012 * This way we don't have other threads
1013 * running and doing things after a slower
1014 * thread has the fatal signal pending.
1015 */
1016 signal->flags = SIGNAL_GROUP_EXIT;
1017 signal->group_exit_code = sig;
1018 signal->group_stop_count = 0;
1019 __for_each_thread(signal, t) {
1020 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1021 sigaddset(&t->pending.signal, SIGKILL);
1022 signal_wake_up(t, 1);
1023 }
1024 return;
1025 }
1026 }
1027
1028 /*
1029 * The signal is already in the shared-pending queue.
1030 * Tell the chosen thread to wake up and dequeue it.
1031 */
1032 signal_wake_up(t, sig == SIGKILL);
1033 return;
1034}
1035
1036static inline bool legacy_queue(struct sigpending *signals, int sig)
1037{
1038 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1039}
1040
1041static int __send_signal_locked(int sig, struct kernel_siginfo *info,
1042 struct task_struct *t, enum pid_type type, bool force)
1043{
1044 struct sigpending *pending;
1045 struct sigqueue *q;
1046 int override_rlimit;
1047 int ret = 0, result;
1048
1049 lockdep_assert_held(&t->sighand->siglock);
1050
1051 result = TRACE_SIGNAL_IGNORED;
1052 if (!prepare_signal(sig, t, force))
1053 goto ret;
1054
1055 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1056 /*
1057 * Short-circuit ignored signals and support queuing
1058 * exactly one non-rt signal, so that we can get more
1059 * detailed information about the cause of the signal.
1060 */
1061 result = TRACE_SIGNAL_ALREADY_PENDING;
1062 if (legacy_queue(pending, sig))
1063 goto ret;
1064
1065 result = TRACE_SIGNAL_DELIVERED;
1066 /*
1067 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1068 */
1069 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1070 goto out_set;
1071
1072 /*
1073 * Real-time signals must be queued if sent by sigqueue, or
1074 * some other real-time mechanism. It is implementation
1075 * defined whether kill() does so. We attempt to do so, on
1076 * the principle of least surprise, but since kill is not
1077 * allowed to fail with EAGAIN when low on memory we just
1078 * make sure at least one signal gets delivered and don't
1079 * pass on the info struct.
1080 */
1081 if (sig < SIGRTMIN)
1082 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1083 else
1084 override_rlimit = 0;
1085
1086 q = sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit);
1087
1088 if (q) {
1089 list_add_tail(&q->list, &pending->list);
1090 switch ((unsigned long) info) {
1091 case (unsigned long) SEND_SIG_NOINFO:
1092 clear_siginfo(&q->info);
1093 q->info.si_signo = sig;
1094 q->info.si_errno = 0;
1095 q->info.si_code = SI_USER;
1096 q->info.si_pid = task_tgid_nr_ns(current,
1097 task_active_pid_ns(t));
1098 rcu_read_lock();
1099 q->info.si_uid =
1100 from_kuid_munged(task_cred_xxx(t, user_ns),
1101 current_uid());
1102 rcu_read_unlock();
1103 break;
1104 case (unsigned long) SEND_SIG_PRIV:
1105 clear_siginfo(&q->info);
1106 q->info.si_signo = sig;
1107 q->info.si_errno = 0;
1108 q->info.si_code = SI_KERNEL;
1109 q->info.si_pid = 0;
1110 q->info.si_uid = 0;
1111 break;
1112 default:
1113 copy_siginfo(&q->info, info);
1114 break;
1115 }
1116 } else if (!is_si_special(info) &&
1117 sig >= SIGRTMIN && info->si_code != SI_USER) {
1118 /*
1119 * Queue overflow, abort. We may abort if the
1120 * signal was rt and sent by user using something
1121 * other than kill().
1122 */
1123 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1124 ret = -EAGAIN;
1125 goto ret;
1126 } else {
1127 /*
1128 * This is a silent loss of information. We still
1129 * send the signal, but the *info bits are lost.
1130 */
1131 result = TRACE_SIGNAL_LOSE_INFO;
1132 }
1133
1134out_set:
1135 signalfd_notify(t, sig);
1136 sigaddset(&pending->signal, sig);
1137
1138 /* Let multiprocess signals appear after on-going forks */
1139 if (type > PIDTYPE_TGID) {
1140 struct multiprocess_signals *delayed;
1141 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1142 sigset_t *signal = &delayed->signal;
1143 /* Can't queue both a stop and a continue signal */
1144 if (sig == SIGCONT)
1145 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1146 else if (sig_kernel_stop(sig))
1147 sigdelset(signal, SIGCONT);
1148 sigaddset(signal, sig);
1149 }
1150 }
1151
1152 complete_signal(sig, t, type);
1153ret:
1154 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1155 return ret;
1156}
1157
1158static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1159{
1160 bool ret = false;
1161 switch (siginfo_layout(info->si_signo, info->si_code)) {
1162 case SIL_KILL:
1163 case SIL_CHLD:
1164 case SIL_RT:
1165 ret = true;
1166 break;
1167 case SIL_TIMER:
1168 case SIL_POLL:
1169 case SIL_FAULT:
1170 case SIL_FAULT_TRAPNO:
1171 case SIL_FAULT_MCEERR:
1172 case SIL_FAULT_BNDERR:
1173 case SIL_FAULT_PKUERR:
1174 case SIL_FAULT_PERF_EVENT:
1175 case SIL_SYS:
1176 ret = false;
1177 break;
1178 }
1179 return ret;
1180}
1181
1182int send_signal_locked(int sig, struct kernel_siginfo *info,
1183 struct task_struct *t, enum pid_type type)
1184{
1185 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1186 bool force = false;
1187
1188 if (info == SEND_SIG_NOINFO) {
1189 /* Force if sent from an ancestor pid namespace */
1190 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1191 } else if (info == SEND_SIG_PRIV) {
1192 /* Don't ignore kernel generated signals */
1193 force = true;
1194 } else if (has_si_pid_and_uid(info)) {
1195 /* SIGKILL and SIGSTOP is special or has ids */
1196 struct user_namespace *t_user_ns;
1197
1198 rcu_read_lock();
1199 t_user_ns = task_cred_xxx(t, user_ns);
1200 if (current_user_ns() != t_user_ns) {
1201 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1202 info->si_uid = from_kuid_munged(t_user_ns, uid);
1203 }
1204 rcu_read_unlock();
1205
1206 /* A kernel generated signal? */
1207 force = (info->si_code == SI_KERNEL);
1208
1209 /* From an ancestor pid namespace? */
1210 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1211 info->si_pid = 0;
1212 force = true;
1213 }
1214 }
1215 return __send_signal_locked(sig, info, t, type, force);
1216}
1217
1218static void print_fatal_signal(int signr)
1219{
1220 struct pt_regs *regs = task_pt_regs(current);
1221 struct file *exe_file;
1222
1223 exe_file = get_task_exe_file(current);
1224 if (exe_file) {
1225 pr_info("%pD: %s: potentially unexpected fatal signal %d.\n",
1226 exe_file, current->comm, signr);
1227 fput(exe_file);
1228 } else {
1229 pr_info("%s: potentially unexpected fatal signal %d.\n",
1230 current->comm, signr);
1231 }
1232
1233#if defined(__i386__) && !defined(__arch_um__)
1234 pr_info("code at %08lx: ", regs->ip);
1235 {
1236 int i;
1237 for (i = 0; i < 16; i++) {
1238 unsigned char insn;
1239
1240 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1241 break;
1242 pr_cont("%02x ", insn);
1243 }
1244 }
1245 pr_cont("\n");
1246#endif
1247 preempt_disable();
1248 show_regs(regs);
1249 preempt_enable();
1250}
1251
1252static int __init setup_print_fatal_signals(char *str)
1253{
1254 get_option (&str, &print_fatal_signals);
1255
1256 return 1;
1257}
1258
1259__setup("print-fatal-signals=", setup_print_fatal_signals);
1260
1261int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1262 enum pid_type type)
1263{
1264 unsigned long flags;
1265 int ret = -ESRCH;
1266
1267 if (lock_task_sighand(p, &flags)) {
1268 ret = send_signal_locked(sig, info, p, type);
1269 unlock_task_sighand(p, &flags);
1270 }
1271
1272 return ret;
1273}
1274
1275enum sig_handler {
1276 HANDLER_CURRENT, /* If reachable use the current handler */
1277 HANDLER_SIG_DFL, /* Always use SIG_DFL handler semantics */
1278 HANDLER_EXIT, /* Only visible as the process exit code */
1279};
1280
1281/*
1282 * Force a signal that the process can't ignore: if necessary
1283 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1284 *
1285 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1286 * since we do not want to have a signal handler that was blocked
1287 * be invoked when user space had explicitly blocked it.
1288 *
1289 * We don't want to have recursive SIGSEGV's etc, for example,
1290 * that is why we also clear SIGNAL_UNKILLABLE.
1291 */
1292static int
1293force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t,
1294 enum sig_handler handler)
1295{
1296 unsigned long int flags;
1297 int ret, blocked, ignored;
1298 struct k_sigaction *action;
1299 int sig = info->si_signo;
1300
1301 spin_lock_irqsave(&t->sighand->siglock, flags);
1302 action = &t->sighand->action[sig-1];
1303 ignored = action->sa.sa_handler == SIG_IGN;
1304 blocked = sigismember(&t->blocked, sig);
1305 if (blocked || ignored || (handler != HANDLER_CURRENT)) {
1306 action->sa.sa_handler = SIG_DFL;
1307 if (handler == HANDLER_EXIT)
1308 action->sa.sa_flags |= SA_IMMUTABLE;
1309 if (blocked)
1310 sigdelset(&t->blocked, sig);
1311 }
1312 /*
1313 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1314 * debugging to leave init killable. But HANDLER_EXIT is always fatal.
1315 */
1316 if (action->sa.sa_handler == SIG_DFL &&
1317 (!t->ptrace || (handler == HANDLER_EXIT)))
1318 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1319 ret = send_signal_locked(sig, info, t, PIDTYPE_PID);
1320 /* This can happen if the signal was already pending and blocked */
1321 if (!task_sigpending(t))
1322 signal_wake_up(t, 0);
1323 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1324
1325 return ret;
1326}
1327
1328int force_sig_info(struct kernel_siginfo *info)
1329{
1330 return force_sig_info_to_task(info, current, HANDLER_CURRENT);
1331}
1332
1333/*
1334 * Nuke all other threads in the group.
1335 */
1336int zap_other_threads(struct task_struct *p)
1337{
1338 struct task_struct *t;
1339 int count = 0;
1340
1341 p->signal->group_stop_count = 0;
1342
1343 for_other_threads(p, t) {
1344 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1345 count++;
1346
1347 /* Don't bother with already dead threads */
1348 if (t->exit_state)
1349 continue;
1350 sigaddset(&t->pending.signal, SIGKILL);
1351 signal_wake_up(t, 1);
1352 }
1353
1354 return count;
1355}
1356
1357struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1358 unsigned long *flags)
1359{
1360 struct sighand_struct *sighand;
1361
1362 rcu_read_lock();
1363 for (;;) {
1364 sighand = rcu_dereference(tsk->sighand);
1365 if (unlikely(sighand == NULL))
1366 break;
1367
1368 /*
1369 * This sighand can be already freed and even reused, but
1370 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1371 * initializes ->siglock: this slab can't go away, it has
1372 * the same object type, ->siglock can't be reinitialized.
1373 *
1374 * We need to ensure that tsk->sighand is still the same
1375 * after we take the lock, we can race with de_thread() or
1376 * __exit_signal(). In the latter case the next iteration
1377 * must see ->sighand == NULL.
1378 */
1379 spin_lock_irqsave(&sighand->siglock, *flags);
1380 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1381 break;
1382 spin_unlock_irqrestore(&sighand->siglock, *flags);
1383 }
1384 rcu_read_unlock();
1385
1386 return sighand;
1387}
1388
1389#ifdef CONFIG_LOCKDEP
1390void lockdep_assert_task_sighand_held(struct task_struct *task)
1391{
1392 struct sighand_struct *sighand;
1393
1394 rcu_read_lock();
1395 sighand = rcu_dereference(task->sighand);
1396 if (sighand)
1397 lockdep_assert_held(&sighand->siglock);
1398 else
1399 WARN_ON_ONCE(1);
1400 rcu_read_unlock();
1401}
1402#endif
1403
1404/*
1405 * send signal info to all the members of a thread group or to the
1406 * individual thread if type == PIDTYPE_PID.
1407 */
1408int group_send_sig_info(int sig, struct kernel_siginfo *info,
1409 struct task_struct *p, enum pid_type type)
1410{
1411 int ret;
1412
1413 rcu_read_lock();
1414 ret = check_kill_permission(sig, info, p);
1415 rcu_read_unlock();
1416
1417 if (!ret && sig)
1418 ret = do_send_sig_info(sig, info, p, type);
1419
1420 return ret;
1421}
1422
1423/*
1424 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1425 * control characters do (^C, ^Z etc)
1426 * - the caller must hold at least a readlock on tasklist_lock
1427 */
1428int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1429{
1430 struct task_struct *p = NULL;
1431 int ret = -ESRCH;
1432
1433 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1434 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1435 /*
1436 * If group_send_sig_info() succeeds at least once ret
1437 * becomes 0 and after that the code below has no effect.
1438 * Otherwise we return the last err or -ESRCH if this
1439 * process group is empty.
1440 */
1441 if (ret)
1442 ret = err;
1443 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1444
1445 return ret;
1446}
1447
1448static int kill_pid_info_type(int sig, struct kernel_siginfo *info,
1449 struct pid *pid, enum pid_type type)
1450{
1451 int error = -ESRCH;
1452 struct task_struct *p;
1453
1454 for (;;) {
1455 rcu_read_lock();
1456 p = pid_task(pid, PIDTYPE_PID);
1457 if (p)
1458 error = group_send_sig_info(sig, info, p, type);
1459 rcu_read_unlock();
1460 if (likely(!p || error != -ESRCH))
1461 return error;
1462 /*
1463 * The task was unhashed in between, try again. If it
1464 * is dead, pid_task() will return NULL, if we race with
1465 * de_thread() it will find the new leader.
1466 */
1467 }
1468}
1469
1470int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1471{
1472 return kill_pid_info_type(sig, info, pid, PIDTYPE_TGID);
1473}
1474
1475static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1476{
1477 int error;
1478 rcu_read_lock();
1479 error = kill_pid_info(sig, info, find_vpid(pid));
1480 rcu_read_unlock();
1481 return error;
1482}
1483
1484static inline bool kill_as_cred_perm(const struct cred *cred,
1485 struct task_struct *target)
1486{
1487 const struct cred *pcred = __task_cred(target);
1488
1489 return uid_eq(cred->euid, pcred->suid) ||
1490 uid_eq(cred->euid, pcred->uid) ||
1491 uid_eq(cred->uid, pcred->suid) ||
1492 uid_eq(cred->uid, pcred->uid);
1493}
1494
1495/*
1496 * The usb asyncio usage of siginfo is wrong. The glibc support
1497 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1498 * AKA after the generic fields:
1499 * kernel_pid_t si_pid;
1500 * kernel_uid32_t si_uid;
1501 * sigval_t si_value;
1502 *
1503 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1504 * after the generic fields is:
1505 * void __user *si_addr;
1506 *
1507 * This is a practical problem when there is a 64bit big endian kernel
1508 * and a 32bit userspace. As the 32bit address will encoded in the low
1509 * 32bits of the pointer. Those low 32bits will be stored at higher
1510 * address than appear in a 32 bit pointer. So userspace will not
1511 * see the address it was expecting for it's completions.
1512 *
1513 * There is nothing in the encoding that can allow
1514 * copy_siginfo_to_user32 to detect this confusion of formats, so
1515 * handle this by requiring the caller of kill_pid_usb_asyncio to
1516 * notice when this situration takes place and to store the 32bit
1517 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1518 * parameter.
1519 */
1520int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1521 struct pid *pid, const struct cred *cred)
1522{
1523 struct kernel_siginfo info;
1524 struct task_struct *p;
1525 unsigned long flags;
1526 int ret = -EINVAL;
1527
1528 if (!valid_signal(sig))
1529 return ret;
1530
1531 clear_siginfo(&info);
1532 info.si_signo = sig;
1533 info.si_errno = errno;
1534 info.si_code = SI_ASYNCIO;
1535 *((sigval_t *)&info.si_pid) = addr;
1536
1537 rcu_read_lock();
1538 p = pid_task(pid, PIDTYPE_PID);
1539 if (!p) {
1540 ret = -ESRCH;
1541 goto out_unlock;
1542 }
1543 if (!kill_as_cred_perm(cred, p)) {
1544 ret = -EPERM;
1545 goto out_unlock;
1546 }
1547 ret = security_task_kill(p, &info, sig, cred);
1548 if (ret)
1549 goto out_unlock;
1550
1551 if (sig) {
1552 if (lock_task_sighand(p, &flags)) {
1553 ret = __send_signal_locked(sig, &info, p, PIDTYPE_TGID, false);
1554 unlock_task_sighand(p, &flags);
1555 } else
1556 ret = -ESRCH;
1557 }
1558out_unlock:
1559 rcu_read_unlock();
1560 return ret;
1561}
1562EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1563
1564/*
1565 * kill_something_info() interprets pid in interesting ways just like kill(2).
1566 *
1567 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1568 * is probably wrong. Should make it like BSD or SYSV.
1569 */
1570
1571static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1572{
1573 int ret;
1574
1575 if (pid > 0)
1576 return kill_proc_info(sig, info, pid);
1577
1578 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1579 if (pid == INT_MIN)
1580 return -ESRCH;
1581
1582 read_lock(&tasklist_lock);
1583 if (pid != -1) {
1584 ret = __kill_pgrp_info(sig, info,
1585 pid ? find_vpid(-pid) : task_pgrp(current));
1586 } else {
1587 int retval = 0, count = 0;
1588 struct task_struct * p;
1589
1590 for_each_process(p) {
1591 if (task_pid_vnr(p) > 1 &&
1592 !same_thread_group(p, current)) {
1593 int err = group_send_sig_info(sig, info, p,
1594 PIDTYPE_MAX);
1595 ++count;
1596 if (err != -EPERM)
1597 retval = err;
1598 }
1599 }
1600 ret = count ? retval : -ESRCH;
1601 }
1602 read_unlock(&tasklist_lock);
1603
1604 return ret;
1605}
1606
1607/*
1608 * These are for backward compatibility with the rest of the kernel source.
1609 */
1610
1611int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1612{
1613 /*
1614 * Make sure legacy kernel users don't send in bad values
1615 * (normal paths check this in check_kill_permission).
1616 */
1617 if (!valid_signal(sig))
1618 return -EINVAL;
1619
1620 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1621}
1622EXPORT_SYMBOL(send_sig_info);
1623
1624#define __si_special(priv) \
1625 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1626
1627int
1628send_sig(int sig, struct task_struct *p, int priv)
1629{
1630 return send_sig_info(sig, __si_special(priv), p);
1631}
1632EXPORT_SYMBOL(send_sig);
1633
1634void force_sig(int sig)
1635{
1636 struct kernel_siginfo info;
1637
1638 clear_siginfo(&info);
1639 info.si_signo = sig;
1640 info.si_errno = 0;
1641 info.si_code = SI_KERNEL;
1642 info.si_pid = 0;
1643 info.si_uid = 0;
1644 force_sig_info(&info);
1645}
1646EXPORT_SYMBOL(force_sig);
1647
1648void force_fatal_sig(int sig)
1649{
1650 struct kernel_siginfo info;
1651
1652 clear_siginfo(&info);
1653 info.si_signo = sig;
1654 info.si_errno = 0;
1655 info.si_code = SI_KERNEL;
1656 info.si_pid = 0;
1657 info.si_uid = 0;
1658 force_sig_info_to_task(&info, current, HANDLER_SIG_DFL);
1659}
1660
1661void force_exit_sig(int sig)
1662{
1663 struct kernel_siginfo info;
1664
1665 clear_siginfo(&info);
1666 info.si_signo = sig;
1667 info.si_errno = 0;
1668 info.si_code = SI_KERNEL;
1669 info.si_pid = 0;
1670 info.si_uid = 0;
1671 force_sig_info_to_task(&info, current, HANDLER_EXIT);
1672}
1673
1674/*
1675 * When things go south during signal handling, we
1676 * will force a SIGSEGV. And if the signal that caused
1677 * the problem was already a SIGSEGV, we'll want to
1678 * make sure we don't even try to deliver the signal..
1679 */
1680void force_sigsegv(int sig)
1681{
1682 if (sig == SIGSEGV)
1683 force_fatal_sig(SIGSEGV);
1684 else
1685 force_sig(SIGSEGV);
1686}
1687
1688int force_sig_fault_to_task(int sig, int code, void __user *addr,
1689 struct task_struct *t)
1690{
1691 struct kernel_siginfo info;
1692
1693 clear_siginfo(&info);
1694 info.si_signo = sig;
1695 info.si_errno = 0;
1696 info.si_code = code;
1697 info.si_addr = addr;
1698 return force_sig_info_to_task(&info, t, HANDLER_CURRENT);
1699}
1700
1701int force_sig_fault(int sig, int code, void __user *addr)
1702{
1703 return force_sig_fault_to_task(sig, code, addr, current);
1704}
1705
1706int send_sig_fault(int sig, int code, void __user *addr, 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 return send_sig_info(info.si_signo, &info, t);
1716}
1717
1718int force_sig_mceerr(int code, void __user *addr, short lsb)
1719{
1720 struct kernel_siginfo info;
1721
1722 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1723 clear_siginfo(&info);
1724 info.si_signo = SIGBUS;
1725 info.si_errno = 0;
1726 info.si_code = code;
1727 info.si_addr = addr;
1728 info.si_addr_lsb = lsb;
1729 return force_sig_info(&info);
1730}
1731
1732int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1733{
1734 struct kernel_siginfo info;
1735
1736 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1737 clear_siginfo(&info);
1738 info.si_signo = SIGBUS;
1739 info.si_errno = 0;
1740 info.si_code = code;
1741 info.si_addr = addr;
1742 info.si_addr_lsb = lsb;
1743 return send_sig_info(info.si_signo, &info, t);
1744}
1745EXPORT_SYMBOL(send_sig_mceerr);
1746
1747int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1748{
1749 struct kernel_siginfo info;
1750
1751 clear_siginfo(&info);
1752 info.si_signo = SIGSEGV;
1753 info.si_errno = 0;
1754 info.si_code = SEGV_BNDERR;
1755 info.si_addr = addr;
1756 info.si_lower = lower;
1757 info.si_upper = upper;
1758 return force_sig_info(&info);
1759}
1760
1761#ifdef SEGV_PKUERR
1762int force_sig_pkuerr(void __user *addr, u32 pkey)
1763{
1764 struct kernel_siginfo info;
1765
1766 clear_siginfo(&info);
1767 info.si_signo = SIGSEGV;
1768 info.si_errno = 0;
1769 info.si_code = SEGV_PKUERR;
1770 info.si_addr = addr;
1771 info.si_pkey = pkey;
1772 return force_sig_info(&info);
1773}
1774#endif
1775
1776int send_sig_perf(void __user *addr, u32 type, u64 sig_data)
1777{
1778 struct kernel_siginfo info;
1779
1780 clear_siginfo(&info);
1781 info.si_signo = SIGTRAP;
1782 info.si_errno = 0;
1783 info.si_code = TRAP_PERF;
1784 info.si_addr = addr;
1785 info.si_perf_data = sig_data;
1786 info.si_perf_type = type;
1787
1788 /*
1789 * Signals generated by perf events should not terminate the whole
1790 * process if SIGTRAP is blocked, however, delivering the signal
1791 * asynchronously is better than not delivering at all. But tell user
1792 * space if the signal was asynchronous, so it can clearly be
1793 * distinguished from normal synchronous ones.
1794 */
1795 info.si_perf_flags = sigismember(¤t->blocked, info.si_signo) ?
1796 TRAP_PERF_FLAG_ASYNC :
1797 0;
1798
1799 return send_sig_info(info.si_signo, &info, current);
1800}
1801
1802/**
1803 * force_sig_seccomp - signals the task to allow in-process syscall emulation
1804 * @syscall: syscall number to send to userland
1805 * @reason: filter-supplied reason code to send to userland (via si_errno)
1806 * @force_coredump: true to trigger a coredump
1807 *
1808 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
1809 */
1810int force_sig_seccomp(int syscall, int reason, bool force_coredump)
1811{
1812 struct kernel_siginfo info;
1813
1814 clear_siginfo(&info);
1815 info.si_signo = SIGSYS;
1816 info.si_code = SYS_SECCOMP;
1817 info.si_call_addr = (void __user *)KSTK_EIP(current);
1818 info.si_errno = reason;
1819 info.si_arch = syscall_get_arch(current);
1820 info.si_syscall = syscall;
1821 return force_sig_info_to_task(&info, current,
1822 force_coredump ? HANDLER_EXIT : HANDLER_CURRENT);
1823}
1824
1825/* For the crazy architectures that include trap information in
1826 * the errno field, instead of an actual errno value.
1827 */
1828int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1829{
1830 struct kernel_siginfo info;
1831
1832 clear_siginfo(&info);
1833 info.si_signo = SIGTRAP;
1834 info.si_errno = errno;
1835 info.si_code = TRAP_HWBKPT;
1836 info.si_addr = addr;
1837 return force_sig_info(&info);
1838}
1839
1840/* For the rare architectures that include trap information using
1841 * si_trapno.
1842 */
1843int force_sig_fault_trapno(int sig, int code, void __user *addr, int trapno)
1844{
1845 struct kernel_siginfo info;
1846
1847 clear_siginfo(&info);
1848 info.si_signo = sig;
1849 info.si_errno = 0;
1850 info.si_code = code;
1851 info.si_addr = addr;
1852 info.si_trapno = trapno;
1853 return force_sig_info(&info);
1854}
1855
1856/* For the rare architectures that include trap information using
1857 * si_trapno.
1858 */
1859int send_sig_fault_trapno(int sig, int code, void __user *addr, int trapno,
1860 struct task_struct *t)
1861{
1862 struct kernel_siginfo info;
1863
1864 clear_siginfo(&info);
1865 info.si_signo = sig;
1866 info.si_errno = 0;
1867 info.si_code = code;
1868 info.si_addr = addr;
1869 info.si_trapno = trapno;
1870 return send_sig_info(info.si_signo, &info, t);
1871}
1872
1873static int kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1874{
1875 int ret;
1876 read_lock(&tasklist_lock);
1877 ret = __kill_pgrp_info(sig, info, pgrp);
1878 read_unlock(&tasklist_lock);
1879 return ret;
1880}
1881
1882int kill_pgrp(struct pid *pid, int sig, int priv)
1883{
1884 return kill_pgrp_info(sig, __si_special(priv), pid);
1885}
1886EXPORT_SYMBOL(kill_pgrp);
1887
1888int kill_pid(struct pid *pid, int sig, int priv)
1889{
1890 return kill_pid_info(sig, __si_special(priv), pid);
1891}
1892EXPORT_SYMBOL(kill_pid);
1893
1894#ifdef CONFIG_POSIX_TIMERS
1895/*
1896 * These functions handle POSIX timer signals. POSIX timers use
1897 * preallocated sigqueue structs for sending signals.
1898 */
1899static void __flush_itimer_signals(struct sigpending *pending)
1900{
1901 sigset_t signal, retain;
1902 struct sigqueue *q, *n;
1903
1904 signal = pending->signal;
1905 sigemptyset(&retain);
1906
1907 list_for_each_entry_safe(q, n, &pending->list, list) {
1908 int sig = q->info.si_signo;
1909
1910 if (likely(q->info.si_code != SI_TIMER)) {
1911 sigaddset(&retain, sig);
1912 } else {
1913 sigdelset(&signal, sig);
1914 list_del_init(&q->list);
1915 __sigqueue_free(q);
1916 }
1917 }
1918
1919 sigorsets(&pending->signal, &signal, &retain);
1920}
1921
1922void flush_itimer_signals(void)
1923{
1924 struct task_struct *tsk = current;
1925
1926 guard(spinlock_irqsave)(&tsk->sighand->siglock);
1927 __flush_itimer_signals(&tsk->pending);
1928 __flush_itimer_signals(&tsk->signal->shared_pending);
1929}
1930
1931bool posixtimer_init_sigqueue(struct sigqueue *q)
1932{
1933 struct ucounts *ucounts = sig_get_ucounts(current, -1, 0);
1934
1935 if (!ucounts)
1936 return false;
1937 clear_siginfo(&q->info);
1938 __sigqueue_init(q, ucounts, SIGQUEUE_PREALLOC);
1939 return true;
1940}
1941
1942static void posixtimer_queue_sigqueue(struct sigqueue *q, struct task_struct *t, enum pid_type type)
1943{
1944 struct sigpending *pending;
1945 int sig = q->info.si_signo;
1946
1947 signalfd_notify(t, sig);
1948 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1949 list_add_tail(&q->list, &pending->list);
1950 sigaddset(&pending->signal, sig);
1951 complete_signal(sig, t, type);
1952}
1953
1954/*
1955 * This function is used by POSIX timers to deliver a timer signal.
1956 * Where type is PIDTYPE_PID (such as for timers with SIGEV_THREAD_ID
1957 * set), the signal must be delivered to the specific thread (queues
1958 * into t->pending).
1959 *
1960 * Where type is not PIDTYPE_PID, signals must be delivered to the
1961 * process. In this case, prefer to deliver to current if it is in
1962 * the same thread group as the target process and its sighand is
1963 * stable, which avoids unnecessarily waking up a potentially idle task.
1964 */
1965static inline struct task_struct *posixtimer_get_target(struct k_itimer *tmr)
1966{
1967 struct task_struct *t = pid_task(tmr->it_pid, tmr->it_pid_type);
1968
1969 if (t && tmr->it_pid_type != PIDTYPE_PID &&
1970 same_thread_group(t, current) && !current->exit_state)
1971 t = current;
1972 return t;
1973}
1974
1975void posixtimer_send_sigqueue(struct k_itimer *tmr)
1976{
1977 struct sigqueue *q = &tmr->sigq;
1978 int sig = q->info.si_signo;
1979 struct task_struct *t;
1980 unsigned long flags;
1981 int result;
1982
1983 guard(rcu)();
1984
1985 t = posixtimer_get_target(tmr);
1986 if (!t)
1987 return;
1988
1989 if (!likely(lock_task_sighand(t, &flags)))
1990 return;
1991
1992 /*
1993 * Update @tmr::sigqueue_seq for posix timer signals with sighand
1994 * locked to prevent a race against dequeue_signal().
1995 */
1996 tmr->it_sigqueue_seq = tmr->it_signal_seq;
1997
1998 /*
1999 * Set the signal delivery status under sighand lock, so that the
2000 * ignored signal handling can distinguish between a periodic and a
2001 * non-periodic timer.
2002 */
2003 tmr->it_sig_periodic = tmr->it_status == POSIX_TIMER_REQUEUE_PENDING;
2004
2005 if (!prepare_signal(sig, t, false)) {
2006 result = TRACE_SIGNAL_IGNORED;
2007
2008 if (!list_empty(&q->list)) {
2009 /*
2010 * The signal was ignored and blocked. The timer
2011 * expiry queued it because blocked signals are
2012 * queued independent of the ignored state.
2013 *
2014 * The unblocking set SIGPENDING, but the signal
2015 * was not yet dequeued from the pending list.
2016 * So prepare_signal() sees unblocked and ignored,
2017 * which ends up here. Leave it queued like a
2018 * regular signal.
2019 *
2020 * The same happens when the task group is exiting
2021 * and the signal is already queued.
2022 * prepare_signal() treats SIGNAL_GROUP_EXIT as
2023 * ignored independent of its queued state. This
2024 * gets cleaned up in __exit_signal().
2025 */
2026 goto out;
2027 }
2028
2029 /* Periodic timers with SIG_IGN are queued on the ignored list */
2030 if (tmr->it_sig_periodic) {
2031 /*
2032 * Already queued means the timer was rearmed after
2033 * the previous expiry got it on the ignore list.
2034 * Nothing to do for that case.
2035 */
2036 if (hlist_unhashed(&tmr->ignored_list)) {
2037 /*
2038 * Take a signal reference and queue it on
2039 * the ignored list.
2040 */
2041 posixtimer_sigqueue_getref(q);
2042 posixtimer_sig_ignore(t, q);
2043 }
2044 } else if (!hlist_unhashed(&tmr->ignored_list)) {
2045 /*
2046 * Covers the case where a timer was periodic and
2047 * then the signal was ignored. Later it was rearmed
2048 * as oneshot timer. The previous signal is invalid
2049 * now, and this oneshot signal has to be dropped.
2050 * Remove it from the ignored list and drop the
2051 * reference count as the signal is not longer
2052 * queued.
2053 */
2054 hlist_del_init(&tmr->ignored_list);
2055 posixtimer_putref(tmr);
2056 }
2057 goto out;
2058 }
2059
2060 if (unlikely(!list_empty(&q->list))) {
2061 /* This holds a reference count already */
2062 result = TRACE_SIGNAL_ALREADY_PENDING;
2063 goto out;
2064 }
2065
2066 /*
2067 * If the signal is on the ignore list, it got blocked after it was
2068 * ignored earlier. But nothing lifted the ignore. Move it back to
2069 * the pending list to be consistent with the regular signal
2070 * handling. This already holds a reference count.
2071 *
2072 * If it's not on the ignore list acquire a reference count.
2073 */
2074 if (likely(hlist_unhashed(&tmr->ignored_list)))
2075 posixtimer_sigqueue_getref(q);
2076 else
2077 hlist_del_init(&tmr->ignored_list);
2078
2079 posixtimer_queue_sigqueue(q, t, tmr->it_pid_type);
2080 result = TRACE_SIGNAL_DELIVERED;
2081out:
2082 trace_signal_generate(sig, &q->info, t, tmr->it_pid_type != PIDTYPE_PID, result);
2083 unlock_task_sighand(t, &flags);
2084}
2085
2086static inline void posixtimer_sig_ignore(struct task_struct *tsk, struct sigqueue *q)
2087{
2088 struct k_itimer *tmr = container_of(q, struct k_itimer, sigq);
2089
2090 /*
2091 * If the timer is marked deleted already or the signal originates
2092 * from a non-periodic timer, then just drop the reference
2093 * count. Otherwise queue it on the ignored list.
2094 */
2095 if (tmr->it_signal && tmr->it_sig_periodic)
2096 hlist_add_head(&tmr->ignored_list, &tsk->signal->ignored_posix_timers);
2097 else
2098 posixtimer_putref(tmr);
2099}
2100
2101static void posixtimer_sig_unignore(struct task_struct *tsk, int sig)
2102{
2103 struct hlist_head *head = &tsk->signal->ignored_posix_timers;
2104 struct hlist_node *tmp;
2105 struct k_itimer *tmr;
2106
2107 if (likely(hlist_empty(head)))
2108 return;
2109
2110 /*
2111 * Rearming a timer with sighand lock held is not possible due to
2112 * lock ordering vs. tmr::it_lock. Just stick the sigqueue back and
2113 * let the signal delivery path deal with it whether it needs to be
2114 * rearmed or not. This cannot be decided here w/o dropping sighand
2115 * lock and creating a loop retry horror show.
2116 */
2117 hlist_for_each_entry_safe(tmr, tmp , head, ignored_list) {
2118 struct task_struct *target;
2119
2120 /*
2121 * tmr::sigq.info.si_signo is immutable, so accessing it
2122 * without holding tmr::it_lock is safe.
2123 */
2124 if (tmr->sigq.info.si_signo != sig)
2125 continue;
2126
2127 hlist_del_init(&tmr->ignored_list);
2128
2129 /* This should never happen and leaks a reference count */
2130 if (WARN_ON_ONCE(!list_empty(&tmr->sigq.list)))
2131 continue;
2132
2133 /*
2134 * Get the target for the signal. If target is a thread and
2135 * has exited by now, drop the reference count.
2136 */
2137 guard(rcu)();
2138 target = posixtimer_get_target(tmr);
2139 if (target)
2140 posixtimer_queue_sigqueue(&tmr->sigq, target, tmr->it_pid_type);
2141 else
2142 posixtimer_putref(tmr);
2143 }
2144}
2145#else /* CONFIG_POSIX_TIMERS */
2146static inline void posixtimer_sig_ignore(struct task_struct *tsk, struct sigqueue *q) { }
2147static inline void posixtimer_sig_unignore(struct task_struct *tsk, int sig) { }
2148#endif /* !CONFIG_POSIX_TIMERS */
2149
2150void do_notify_pidfd(struct task_struct *task)
2151{
2152 struct pid *pid = task_pid(task);
2153
2154 WARN_ON(task->exit_state == 0);
2155
2156 __wake_up(&pid->wait_pidfd, TASK_NORMAL, 0,
2157 poll_to_key(EPOLLIN | EPOLLRDNORM));
2158}
2159
2160/*
2161 * Let a parent know about the death of a child.
2162 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
2163 *
2164 * Returns true if our parent ignored us and so we've switched to
2165 * self-reaping.
2166 */
2167bool do_notify_parent(struct task_struct *tsk, int sig)
2168{
2169 struct kernel_siginfo info;
2170 unsigned long flags;
2171 struct sighand_struct *psig;
2172 bool autoreap = false;
2173 u64 utime, stime;
2174
2175 WARN_ON_ONCE(sig == -1);
2176
2177 /* do_notify_parent_cldstop should have been called instead. */
2178 WARN_ON_ONCE(task_is_stopped_or_traced(tsk));
2179
2180 WARN_ON_ONCE(!tsk->ptrace &&
2181 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
2182 /*
2183 * tsk is a group leader and has no threads, wake up the
2184 * non-PIDFD_THREAD waiters.
2185 */
2186 if (thread_group_empty(tsk))
2187 do_notify_pidfd(tsk);
2188
2189 if (sig != SIGCHLD) {
2190 /*
2191 * This is only possible if parent == real_parent.
2192 * Check if it has changed security domain.
2193 */
2194 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
2195 sig = SIGCHLD;
2196 }
2197
2198 clear_siginfo(&info);
2199 info.si_signo = sig;
2200 info.si_errno = 0;
2201 /*
2202 * We are under tasklist_lock here so our parent is tied to
2203 * us and cannot change.
2204 *
2205 * task_active_pid_ns will always return the same pid namespace
2206 * until a task passes through release_task.
2207 *
2208 * write_lock() currently calls preempt_disable() which is the
2209 * same as rcu_read_lock(), but according to Oleg, this is not
2210 * correct to rely on this
2211 */
2212 rcu_read_lock();
2213 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
2214 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
2215 task_uid(tsk));
2216 rcu_read_unlock();
2217
2218 task_cputime(tsk, &utime, &stime);
2219 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
2220 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
2221
2222 info.si_status = tsk->exit_code & 0x7f;
2223 if (tsk->exit_code & 0x80)
2224 info.si_code = CLD_DUMPED;
2225 else if (tsk->exit_code & 0x7f)
2226 info.si_code = CLD_KILLED;
2227 else {
2228 info.si_code = CLD_EXITED;
2229 info.si_status = tsk->exit_code >> 8;
2230 }
2231
2232 psig = tsk->parent->sighand;
2233 spin_lock_irqsave(&psig->siglock, flags);
2234 if (!tsk->ptrace && sig == SIGCHLD &&
2235 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
2236 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
2237 /*
2238 * We are exiting and our parent doesn't care. POSIX.1
2239 * defines special semantics for setting SIGCHLD to SIG_IGN
2240 * or setting the SA_NOCLDWAIT flag: we should be reaped
2241 * automatically and not left for our parent's wait4 call.
2242 * Rather than having the parent do it as a magic kind of
2243 * signal handler, we just set this to tell do_exit that we
2244 * can be cleaned up without becoming a zombie. Note that
2245 * we still call __wake_up_parent in this case, because a
2246 * blocked sys_wait4 might now return -ECHILD.
2247 *
2248 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
2249 * is implementation-defined: we do (if you don't want
2250 * it, just use SIG_IGN instead).
2251 */
2252 autoreap = true;
2253 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2254 sig = 0;
2255 }
2256 /*
2257 * Send with __send_signal as si_pid and si_uid are in the
2258 * parent's namespaces.
2259 */
2260 if (valid_signal(sig) && sig)
2261 __send_signal_locked(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2262 __wake_up_parent(tsk, tsk->parent);
2263 spin_unlock_irqrestore(&psig->siglock, flags);
2264
2265 return autoreap;
2266}
2267
2268/**
2269 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2270 * @tsk: task reporting the state change
2271 * @for_ptracer: the notification is for ptracer
2272 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2273 *
2274 * Notify @tsk's parent that the stopped/continued state has changed. If
2275 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2276 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2277 *
2278 * CONTEXT:
2279 * Must be called with tasklist_lock at least read locked.
2280 */
2281static void do_notify_parent_cldstop(struct task_struct *tsk,
2282 bool for_ptracer, int why)
2283{
2284 struct kernel_siginfo info;
2285 unsigned long flags;
2286 struct task_struct *parent;
2287 struct sighand_struct *sighand;
2288 u64 utime, stime;
2289
2290 if (for_ptracer) {
2291 parent = tsk->parent;
2292 } else {
2293 tsk = tsk->group_leader;
2294 parent = tsk->real_parent;
2295 }
2296
2297 clear_siginfo(&info);
2298 info.si_signo = SIGCHLD;
2299 info.si_errno = 0;
2300 /*
2301 * see comment in do_notify_parent() about the following 4 lines
2302 */
2303 rcu_read_lock();
2304 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2305 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2306 rcu_read_unlock();
2307
2308 task_cputime(tsk, &utime, &stime);
2309 info.si_utime = nsec_to_clock_t(utime);
2310 info.si_stime = nsec_to_clock_t(stime);
2311
2312 info.si_code = why;
2313 switch (why) {
2314 case CLD_CONTINUED:
2315 info.si_status = SIGCONT;
2316 break;
2317 case CLD_STOPPED:
2318 info.si_status = tsk->signal->group_exit_code & 0x7f;
2319 break;
2320 case CLD_TRAPPED:
2321 info.si_status = tsk->exit_code & 0x7f;
2322 break;
2323 default:
2324 BUG();
2325 }
2326
2327 sighand = parent->sighand;
2328 spin_lock_irqsave(&sighand->siglock, flags);
2329 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2330 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2331 send_signal_locked(SIGCHLD, &info, parent, PIDTYPE_TGID);
2332 /*
2333 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2334 */
2335 __wake_up_parent(tsk, parent);
2336 spin_unlock_irqrestore(&sighand->siglock, flags);
2337}
2338
2339/*
2340 * This must be called with current->sighand->siglock held.
2341 *
2342 * This should be the path for all ptrace stops.
2343 * We always set current->last_siginfo while stopped here.
2344 * That makes it a way to test a stopped process for
2345 * being ptrace-stopped vs being job-control-stopped.
2346 *
2347 * Returns the signal the ptracer requested the code resume
2348 * with. If the code did not stop because the tracer is gone,
2349 * the stop signal remains unchanged unless clear_code.
2350 */
2351static int ptrace_stop(int exit_code, int why, unsigned long message,
2352 kernel_siginfo_t *info)
2353 __releases(¤t->sighand->siglock)
2354 __acquires(¤t->sighand->siglock)
2355{
2356 bool gstop_done = false;
2357
2358 if (arch_ptrace_stop_needed()) {
2359 /*
2360 * The arch code has something special to do before a
2361 * ptrace stop. This is allowed to block, e.g. for faults
2362 * on user stack pages. We can't keep the siglock while
2363 * calling arch_ptrace_stop, so we must release it now.
2364 * To preserve proper semantics, we must do this before
2365 * any signal bookkeeping like checking group_stop_count.
2366 */
2367 spin_unlock_irq(¤t->sighand->siglock);
2368 arch_ptrace_stop();
2369 spin_lock_irq(¤t->sighand->siglock);
2370 }
2371
2372 /*
2373 * After this point ptrace_signal_wake_up or signal_wake_up
2374 * will clear TASK_TRACED if ptrace_unlink happens or a fatal
2375 * signal comes in. Handle previous ptrace_unlinks and fatal
2376 * signals here to prevent ptrace_stop sleeping in schedule.
2377 */
2378 if (!current->ptrace || __fatal_signal_pending(current))
2379 return exit_code;
2380
2381 set_special_state(TASK_TRACED);
2382 current->jobctl |= JOBCTL_TRACED;
2383
2384 /*
2385 * We're committing to trapping. TRACED should be visible before
2386 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2387 * Also, transition to TRACED and updates to ->jobctl should be
2388 * atomic with respect to siglock and should be done after the arch
2389 * hook as siglock is released and regrabbed across it.
2390 *
2391 * TRACER TRACEE
2392 *
2393 * ptrace_attach()
2394 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2395 * do_wait()
2396 * set_current_state() smp_wmb();
2397 * ptrace_do_wait()
2398 * wait_task_stopped()
2399 * task_stopped_code()
2400 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2401 */
2402 smp_wmb();
2403
2404 current->ptrace_message = message;
2405 current->last_siginfo = info;
2406 current->exit_code = exit_code;
2407
2408 /*
2409 * If @why is CLD_STOPPED, we're trapping to participate in a group
2410 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2411 * across siglock relocks since INTERRUPT was scheduled, PENDING
2412 * could be clear now. We act as if SIGCONT is received after
2413 * TASK_TRACED is entered - ignore it.
2414 */
2415 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2416 gstop_done = task_participate_group_stop(current);
2417
2418 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2419 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2420 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2421 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2422
2423 /* entering a trap, clear TRAPPING */
2424 task_clear_jobctl_trapping(current);
2425
2426 spin_unlock_irq(¤t->sighand->siglock);
2427 read_lock(&tasklist_lock);
2428 /*
2429 * Notify parents of the stop.
2430 *
2431 * While ptraced, there are two parents - the ptracer and
2432 * the real_parent of the group_leader. The ptracer should
2433 * know about every stop while the real parent is only
2434 * interested in the completion of group stop. The states
2435 * for the two don't interact with each other. Notify
2436 * separately unless they're gonna be duplicates.
2437 */
2438 if (current->ptrace)
2439 do_notify_parent_cldstop(current, true, why);
2440 if (gstop_done && (!current->ptrace || ptrace_reparented(current)))
2441 do_notify_parent_cldstop(current, false, why);
2442
2443 /*
2444 * The previous do_notify_parent_cldstop() invocation woke ptracer.
2445 * One a PREEMPTION kernel this can result in preemption requirement
2446 * which will be fulfilled after read_unlock() and the ptracer will be
2447 * put on the CPU.
2448 * The ptracer is in wait_task_inactive(, __TASK_TRACED) waiting for
2449 * this task wait in schedule(). If this task gets preempted then it
2450 * remains enqueued on the runqueue. The ptracer will observe this and
2451 * then sleep for a delay of one HZ tick. In the meantime this task
2452 * gets scheduled, enters schedule() and will wait for the ptracer.
2453 *
2454 * This preemption point is not bad from a correctness point of
2455 * view but extends the runtime by one HZ tick time due to the
2456 * ptracer's sleep. The preempt-disable section ensures that there
2457 * will be no preemption between unlock and schedule() and so
2458 * improving the performance since the ptracer will observe that
2459 * the tracee is scheduled out once it gets on the CPU.
2460 *
2461 * On PREEMPT_RT locking tasklist_lock does not disable preemption.
2462 * Therefore the task can be preempted after do_notify_parent_cldstop()
2463 * before unlocking tasklist_lock so there is no benefit in doing this.
2464 *
2465 * In fact disabling preemption is harmful on PREEMPT_RT because
2466 * the spinlock_t in cgroup_enter_frozen() must not be acquired
2467 * with preemption disabled due to the 'sleeping' spinlock
2468 * substitution of RT.
2469 */
2470 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2471 preempt_disable();
2472 read_unlock(&tasklist_lock);
2473 cgroup_enter_frozen();
2474 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2475 preempt_enable_no_resched();
2476 schedule();
2477 cgroup_leave_frozen(true);
2478
2479 /*
2480 * We are back. Now reacquire the siglock before touching
2481 * last_siginfo, so that we are sure to have synchronized with
2482 * any signal-sending on another CPU that wants to examine it.
2483 */
2484 spin_lock_irq(¤t->sighand->siglock);
2485 exit_code = current->exit_code;
2486 current->last_siginfo = NULL;
2487 current->ptrace_message = 0;
2488 current->exit_code = 0;
2489
2490 /* LISTENING can be set only during STOP traps, clear it */
2491 current->jobctl &= ~(JOBCTL_LISTENING | JOBCTL_PTRACE_FROZEN);
2492
2493 /*
2494 * Queued signals ignored us while we were stopped for tracing.
2495 * So check for any that we should take before resuming user mode.
2496 * This sets TIF_SIGPENDING, but never clears it.
2497 */
2498 recalc_sigpending_tsk(current);
2499 return exit_code;
2500}
2501
2502static int ptrace_do_notify(int signr, int exit_code, int why, unsigned long message)
2503{
2504 kernel_siginfo_t info;
2505
2506 clear_siginfo(&info);
2507 info.si_signo = signr;
2508 info.si_code = exit_code;
2509 info.si_pid = task_pid_vnr(current);
2510 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2511
2512 /* Let the debugger run. */
2513 return ptrace_stop(exit_code, why, message, &info);
2514}
2515
2516int ptrace_notify(int exit_code, unsigned long message)
2517{
2518 int signr;
2519
2520 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2521 if (unlikely(task_work_pending(current)))
2522 task_work_run();
2523
2524 spin_lock_irq(¤t->sighand->siglock);
2525 signr = ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED, message);
2526 spin_unlock_irq(¤t->sighand->siglock);
2527 return signr;
2528}
2529
2530/**
2531 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2532 * @signr: signr causing group stop if initiating
2533 *
2534 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2535 * and participate in it. If already set, participate in the existing
2536 * group stop. If participated in a group stop (and thus slept), %true is
2537 * returned with siglock released.
2538 *
2539 * If ptraced, this function doesn't handle stop itself. Instead,
2540 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2541 * untouched. The caller must ensure that INTERRUPT trap handling takes
2542 * places afterwards.
2543 *
2544 * CONTEXT:
2545 * Must be called with @current->sighand->siglock held, which is released
2546 * on %true return.
2547 *
2548 * RETURNS:
2549 * %false if group stop is already cancelled or ptrace trap is scheduled.
2550 * %true if participated in group stop.
2551 */
2552static bool do_signal_stop(int signr)
2553 __releases(¤t->sighand->siglock)
2554{
2555 struct signal_struct *sig = current->signal;
2556
2557 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2558 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2559 struct task_struct *t;
2560
2561 /* signr will be recorded in task->jobctl for retries */
2562 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2563
2564 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2565 unlikely(sig->flags & SIGNAL_GROUP_EXIT) ||
2566 unlikely(sig->group_exec_task))
2567 return false;
2568 /*
2569 * There is no group stop already in progress. We must
2570 * initiate one now.
2571 *
2572 * While ptraced, a task may be resumed while group stop is
2573 * still in effect and then receive a stop signal and
2574 * initiate another group stop. This deviates from the
2575 * usual behavior as two consecutive stop signals can't
2576 * cause two group stops when !ptraced. That is why we
2577 * also check !task_is_stopped(t) below.
2578 *
2579 * The condition can be distinguished by testing whether
2580 * SIGNAL_STOP_STOPPED is already set. Don't generate
2581 * group_exit_code in such case.
2582 *
2583 * This is not necessary for SIGNAL_STOP_CONTINUED because
2584 * an intervening stop signal is required to cause two
2585 * continued events regardless of ptrace.
2586 */
2587 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2588 sig->group_exit_code = signr;
2589
2590 sig->group_stop_count = 0;
2591 if (task_set_jobctl_pending(current, signr | gstop))
2592 sig->group_stop_count++;
2593
2594 for_other_threads(current, t) {
2595 /*
2596 * Setting state to TASK_STOPPED for a group
2597 * stop is always done with the siglock held,
2598 * so this check has no races.
2599 */
2600 if (!task_is_stopped(t) &&
2601 task_set_jobctl_pending(t, signr | gstop)) {
2602 sig->group_stop_count++;
2603 if (likely(!(t->ptrace & PT_SEIZED)))
2604 signal_wake_up(t, 0);
2605 else
2606 ptrace_trap_notify(t);
2607 }
2608 }
2609 }
2610
2611 if (likely(!current->ptrace)) {
2612 int notify = 0;
2613
2614 /*
2615 * If there are no other threads in the group, or if there
2616 * is a group stop in progress and we are the last to stop,
2617 * report to the parent.
2618 */
2619 if (task_participate_group_stop(current))
2620 notify = CLD_STOPPED;
2621
2622 current->jobctl |= JOBCTL_STOPPED;
2623 set_special_state(TASK_STOPPED);
2624 spin_unlock_irq(¤t->sighand->siglock);
2625
2626 /*
2627 * Notify the parent of the group stop completion. Because
2628 * we're not holding either the siglock or tasklist_lock
2629 * here, ptracer may attach inbetween; however, this is for
2630 * group stop and should always be delivered to the real
2631 * parent of the group leader. The new ptracer will get
2632 * its notification when this task transitions into
2633 * TASK_TRACED.
2634 */
2635 if (notify) {
2636 read_lock(&tasklist_lock);
2637 do_notify_parent_cldstop(current, false, notify);
2638 read_unlock(&tasklist_lock);
2639 }
2640
2641 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2642 cgroup_enter_frozen();
2643 schedule();
2644 return true;
2645 } else {
2646 /*
2647 * While ptraced, group stop is handled by STOP trap.
2648 * Schedule it and let the caller deal with it.
2649 */
2650 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2651 return false;
2652 }
2653}
2654
2655/**
2656 * do_jobctl_trap - take care of ptrace jobctl traps
2657 *
2658 * When PT_SEIZED, it's used for both group stop and explicit
2659 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2660 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2661 * the stop signal; otherwise, %SIGTRAP.
2662 *
2663 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2664 * number as exit_code and no siginfo.
2665 *
2666 * CONTEXT:
2667 * Must be called with @current->sighand->siglock held, which may be
2668 * released and re-acquired before returning with intervening sleep.
2669 */
2670static void do_jobctl_trap(void)
2671{
2672 struct signal_struct *signal = current->signal;
2673 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2674
2675 if (current->ptrace & PT_SEIZED) {
2676 if (!signal->group_stop_count &&
2677 !(signal->flags & SIGNAL_STOP_STOPPED))
2678 signr = SIGTRAP;
2679 WARN_ON_ONCE(!signr);
2680 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2681 CLD_STOPPED, 0);
2682 } else {
2683 WARN_ON_ONCE(!signr);
2684 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2685 }
2686}
2687
2688/**
2689 * do_freezer_trap - handle the freezer jobctl trap
2690 *
2691 * Puts the task into frozen state, if only the task is not about to quit.
2692 * In this case it drops JOBCTL_TRAP_FREEZE.
2693 *
2694 * CONTEXT:
2695 * Must be called with @current->sighand->siglock held,
2696 * which is always released before returning.
2697 */
2698static void do_freezer_trap(void)
2699 __releases(¤t->sighand->siglock)
2700{
2701 /*
2702 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2703 * let's make another loop to give it a chance to be handled.
2704 * In any case, we'll return back.
2705 */
2706 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2707 JOBCTL_TRAP_FREEZE) {
2708 spin_unlock_irq(¤t->sighand->siglock);
2709 return;
2710 }
2711
2712 /*
2713 * Now we're sure that there is no pending fatal signal and no
2714 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2715 * immediately (if there is a non-fatal signal pending), and
2716 * put the task into sleep.
2717 */
2718 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2719 clear_thread_flag(TIF_SIGPENDING);
2720 spin_unlock_irq(¤t->sighand->siglock);
2721 cgroup_enter_frozen();
2722 schedule();
2723
2724 /*
2725 * We could've been woken by task_work, run it to clear
2726 * TIF_NOTIFY_SIGNAL. The caller will retry if necessary.
2727 */
2728 clear_notify_signal();
2729 if (unlikely(task_work_pending(current)))
2730 task_work_run();
2731}
2732
2733static int ptrace_signal(int signr, kernel_siginfo_t *info, enum pid_type type)
2734{
2735 /*
2736 * We do not check sig_kernel_stop(signr) but set this marker
2737 * unconditionally because we do not know whether debugger will
2738 * change signr. This flag has no meaning unless we are going
2739 * to stop after return from ptrace_stop(). In this case it will
2740 * be checked in do_signal_stop(), we should only stop if it was
2741 * not cleared by SIGCONT while we were sleeping. See also the
2742 * comment in dequeue_signal().
2743 */
2744 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2745 signr = ptrace_stop(signr, CLD_TRAPPED, 0, info);
2746
2747 /* We're back. Did the debugger cancel the sig? */
2748 if (signr == 0)
2749 return signr;
2750
2751 /*
2752 * Update the siginfo structure if the signal has
2753 * changed. If the debugger wanted something
2754 * specific in the siginfo structure then it should
2755 * have updated *info via PTRACE_SETSIGINFO.
2756 */
2757 if (signr != info->si_signo) {
2758 clear_siginfo(info);
2759 info->si_signo = signr;
2760 info->si_errno = 0;
2761 info->si_code = SI_USER;
2762 rcu_read_lock();
2763 info->si_pid = task_pid_vnr(current->parent);
2764 info->si_uid = from_kuid_munged(current_user_ns(),
2765 task_uid(current->parent));
2766 rcu_read_unlock();
2767 }
2768
2769 /* If the (new) signal is now blocked, requeue it. */
2770 if (sigismember(¤t->blocked, signr) ||
2771 fatal_signal_pending(current)) {
2772 send_signal_locked(signr, info, current, type);
2773 signr = 0;
2774 }
2775
2776 return signr;
2777}
2778
2779static void hide_si_addr_tag_bits(struct ksignal *ksig)
2780{
2781 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2782 case SIL_FAULT:
2783 case SIL_FAULT_TRAPNO:
2784 case SIL_FAULT_MCEERR:
2785 case SIL_FAULT_BNDERR:
2786 case SIL_FAULT_PKUERR:
2787 case SIL_FAULT_PERF_EVENT:
2788 ksig->info.si_addr = arch_untagged_si_addr(
2789 ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2790 break;
2791 case SIL_KILL:
2792 case SIL_TIMER:
2793 case SIL_POLL:
2794 case SIL_CHLD:
2795 case SIL_RT:
2796 case SIL_SYS:
2797 break;
2798 }
2799}
2800
2801bool get_signal(struct ksignal *ksig)
2802{
2803 struct sighand_struct *sighand = current->sighand;
2804 struct signal_struct *signal = current->signal;
2805 int signr;
2806
2807 clear_notify_signal();
2808 if (unlikely(task_work_pending(current)))
2809 task_work_run();
2810
2811 if (!task_sigpending(current))
2812 return false;
2813
2814 if (unlikely(uprobe_deny_signal()))
2815 return false;
2816
2817 /*
2818 * Do this once, we can't return to user-mode if freezing() == T.
2819 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2820 * thus do not need another check after return.
2821 */
2822 try_to_freeze();
2823
2824relock:
2825 spin_lock_irq(&sighand->siglock);
2826
2827 /*
2828 * Every stopped thread goes here after wakeup. Check to see if
2829 * we should notify the parent, prepare_signal(SIGCONT) encodes
2830 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2831 */
2832 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2833 int why;
2834
2835 if (signal->flags & SIGNAL_CLD_CONTINUED)
2836 why = CLD_CONTINUED;
2837 else
2838 why = CLD_STOPPED;
2839
2840 signal->flags &= ~SIGNAL_CLD_MASK;
2841
2842 spin_unlock_irq(&sighand->siglock);
2843
2844 /*
2845 * Notify the parent that we're continuing. This event is
2846 * always per-process and doesn't make whole lot of sense
2847 * for ptracers, who shouldn't consume the state via
2848 * wait(2) either, but, for backward compatibility, notify
2849 * the ptracer of the group leader too unless it's gonna be
2850 * a duplicate.
2851 */
2852 read_lock(&tasklist_lock);
2853 do_notify_parent_cldstop(current, false, why);
2854
2855 if (ptrace_reparented(current->group_leader))
2856 do_notify_parent_cldstop(current->group_leader,
2857 true, why);
2858 read_unlock(&tasklist_lock);
2859
2860 goto relock;
2861 }
2862
2863 for (;;) {
2864 struct k_sigaction *ka;
2865 enum pid_type type;
2866
2867 /* Has this task already been marked for death? */
2868 if ((signal->flags & SIGNAL_GROUP_EXIT) ||
2869 signal->group_exec_task) {
2870 signr = SIGKILL;
2871 sigdelset(¤t->pending.signal, SIGKILL);
2872 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2873 &sighand->action[SIGKILL-1]);
2874 recalc_sigpending();
2875 /*
2876 * implies do_group_exit() or return to PF_USER_WORKER,
2877 * no need to initialize ksig->info/etc.
2878 */
2879 goto fatal;
2880 }
2881
2882 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2883 do_signal_stop(0))
2884 goto relock;
2885
2886 if (unlikely(current->jobctl &
2887 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2888 if (current->jobctl & JOBCTL_TRAP_MASK) {
2889 do_jobctl_trap();
2890 spin_unlock_irq(&sighand->siglock);
2891 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2892 do_freezer_trap();
2893
2894 goto relock;
2895 }
2896
2897 /*
2898 * If the task is leaving the frozen state, let's update
2899 * cgroup counters and reset the frozen bit.
2900 */
2901 if (unlikely(cgroup_task_frozen(current))) {
2902 spin_unlock_irq(&sighand->siglock);
2903 cgroup_leave_frozen(false);
2904 goto relock;
2905 }
2906
2907 /*
2908 * Signals generated by the execution of an instruction
2909 * need to be delivered before any other pending signals
2910 * so that the instruction pointer in the signal stack
2911 * frame points to the faulting instruction.
2912 */
2913 type = PIDTYPE_PID;
2914 signr = dequeue_synchronous_signal(&ksig->info);
2915 if (!signr)
2916 signr = dequeue_signal(¤t->blocked, &ksig->info, &type);
2917
2918 if (!signr)
2919 break; /* will return 0 */
2920
2921 if (unlikely(current->ptrace) && (signr != SIGKILL) &&
2922 !(sighand->action[signr -1].sa.sa_flags & SA_IMMUTABLE)) {
2923 signr = ptrace_signal(signr, &ksig->info, type);
2924 if (!signr)
2925 continue;
2926 }
2927
2928 ka = &sighand->action[signr-1];
2929
2930 /* Trace actually delivered signals. */
2931 trace_signal_deliver(signr, &ksig->info, ka);
2932
2933 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2934 continue;
2935 if (ka->sa.sa_handler != SIG_DFL) {
2936 /* Run the handler. */
2937 ksig->ka = *ka;
2938
2939 if (ka->sa.sa_flags & SA_ONESHOT)
2940 ka->sa.sa_handler = SIG_DFL;
2941
2942 break; /* will return non-zero "signr" value */
2943 }
2944
2945 /*
2946 * Now we are doing the default action for this signal.
2947 */
2948 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2949 continue;
2950
2951 /*
2952 * Global init gets no signals it doesn't want.
2953 * Container-init gets no signals it doesn't want from same
2954 * container.
2955 *
2956 * Note that if global/container-init sees a sig_kernel_only()
2957 * signal here, the signal must have been generated internally
2958 * or must have come from an ancestor namespace. In either
2959 * case, the signal cannot be dropped.
2960 */
2961 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2962 !sig_kernel_only(signr))
2963 continue;
2964
2965 if (sig_kernel_stop(signr)) {
2966 /*
2967 * The default action is to stop all threads in
2968 * the thread group. The job control signals
2969 * do nothing in an orphaned pgrp, but SIGSTOP
2970 * always works. Note that siglock needs to be
2971 * dropped during the call to is_orphaned_pgrp()
2972 * because of lock ordering with tasklist_lock.
2973 * This allows an intervening SIGCONT to be posted.
2974 * We need to check for that and bail out if necessary.
2975 */
2976 if (signr != SIGSTOP) {
2977 spin_unlock_irq(&sighand->siglock);
2978
2979 /* signals can be posted during this window */
2980
2981 if (is_current_pgrp_orphaned())
2982 goto relock;
2983
2984 spin_lock_irq(&sighand->siglock);
2985 }
2986
2987 if (likely(do_signal_stop(signr))) {
2988 /* It released the siglock. */
2989 goto relock;
2990 }
2991
2992 /*
2993 * We didn't actually stop, due to a race
2994 * with SIGCONT or something like that.
2995 */
2996 continue;
2997 }
2998
2999 fatal:
3000 spin_unlock_irq(&sighand->siglock);
3001 if (unlikely(cgroup_task_frozen(current)))
3002 cgroup_leave_frozen(true);
3003
3004 /*
3005 * Anything else is fatal, maybe with a core dump.
3006 */
3007 current->flags |= PF_SIGNALED;
3008
3009 if (sig_kernel_coredump(signr)) {
3010 if (print_fatal_signals)
3011 print_fatal_signal(signr);
3012 proc_coredump_connector(current);
3013 /*
3014 * If it was able to dump core, this kills all
3015 * other threads in the group and synchronizes with
3016 * their demise. If we lost the race with another
3017 * thread getting here, it set group_exit_code
3018 * first and our do_group_exit call below will use
3019 * that value and ignore the one we pass it.
3020 */
3021 do_coredump(&ksig->info);
3022 }
3023
3024 /*
3025 * PF_USER_WORKER threads will catch and exit on fatal signals
3026 * themselves. They have cleanup that must be performed, so we
3027 * cannot call do_exit() on their behalf. Note that ksig won't
3028 * be properly initialized, PF_USER_WORKER's shouldn't use it.
3029 */
3030 if (current->flags & PF_USER_WORKER)
3031 goto out;
3032
3033 /*
3034 * Death signals, no core dump.
3035 */
3036 do_group_exit(signr);
3037 /* NOTREACHED */
3038 }
3039 spin_unlock_irq(&sighand->siglock);
3040
3041 ksig->sig = signr;
3042
3043 if (signr && !(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
3044 hide_si_addr_tag_bits(ksig);
3045out:
3046 return signr > 0;
3047}
3048
3049/**
3050 * signal_delivered - called after signal delivery to update blocked signals
3051 * @ksig: kernel signal struct
3052 * @stepping: nonzero if debugger single-step or block-step in use
3053 *
3054 * This function should be called when a signal has successfully been
3055 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
3056 * is always blocked), and the signal itself is blocked unless %SA_NODEFER
3057 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
3058 */
3059static void signal_delivered(struct ksignal *ksig, int stepping)
3060{
3061 sigset_t blocked;
3062
3063 /* A signal was successfully delivered, and the
3064 saved sigmask was stored on the signal frame,
3065 and will be restored by sigreturn. So we can
3066 simply clear the restore sigmask flag. */
3067 clear_restore_sigmask();
3068
3069 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask);
3070 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
3071 sigaddset(&blocked, ksig->sig);
3072 set_current_blocked(&blocked);
3073 if (current->sas_ss_flags & SS_AUTODISARM)
3074 sas_ss_reset(current);
3075 if (stepping)
3076 ptrace_notify(SIGTRAP, 0);
3077}
3078
3079void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
3080{
3081 if (failed)
3082 force_sigsegv(ksig->sig);
3083 else
3084 signal_delivered(ksig, stepping);
3085}
3086
3087/*
3088 * It could be that complete_signal() picked us to notify about the
3089 * group-wide signal. Other threads should be notified now to take
3090 * the shared signals in @which since we will not.
3091 */
3092static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
3093{
3094 sigset_t retarget;
3095 struct task_struct *t;
3096
3097 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
3098 if (sigisemptyset(&retarget))
3099 return;
3100
3101 for_other_threads(tsk, t) {
3102 if (t->flags & PF_EXITING)
3103 continue;
3104
3105 if (!has_pending_signals(&retarget, &t->blocked))
3106 continue;
3107 /* Remove the signals this thread can handle. */
3108 sigandsets(&retarget, &retarget, &t->blocked);
3109
3110 if (!task_sigpending(t))
3111 signal_wake_up(t, 0);
3112
3113 if (sigisemptyset(&retarget))
3114 break;
3115 }
3116}
3117
3118void exit_signals(struct task_struct *tsk)
3119{
3120 int group_stop = 0;
3121 sigset_t unblocked;
3122
3123 /*
3124 * @tsk is about to have PF_EXITING set - lock out users which
3125 * expect stable threadgroup.
3126 */
3127 cgroup_threadgroup_change_begin(tsk);
3128
3129 if (thread_group_empty(tsk) || (tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
3130 sched_mm_cid_exit_signals(tsk);
3131 tsk->flags |= PF_EXITING;
3132 cgroup_threadgroup_change_end(tsk);
3133 return;
3134 }
3135
3136 spin_lock_irq(&tsk->sighand->siglock);
3137 /*
3138 * From now this task is not visible for group-wide signals,
3139 * see wants_signal(), do_signal_stop().
3140 */
3141 sched_mm_cid_exit_signals(tsk);
3142 tsk->flags |= PF_EXITING;
3143
3144 cgroup_threadgroup_change_end(tsk);
3145
3146 if (!task_sigpending(tsk))
3147 goto out;
3148
3149 unblocked = tsk->blocked;
3150 signotset(&unblocked);
3151 retarget_shared_pending(tsk, &unblocked);
3152
3153 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
3154 task_participate_group_stop(tsk))
3155 group_stop = CLD_STOPPED;
3156out:
3157 spin_unlock_irq(&tsk->sighand->siglock);
3158
3159 /*
3160 * If group stop has completed, deliver the notification. This
3161 * should always go to the real parent of the group leader.
3162 */
3163 if (unlikely(group_stop)) {
3164 read_lock(&tasklist_lock);
3165 do_notify_parent_cldstop(tsk, false, group_stop);
3166 read_unlock(&tasklist_lock);
3167 }
3168}
3169
3170/*
3171 * System call entry points.
3172 */
3173
3174/**
3175 * sys_restart_syscall - restart a system call
3176 */
3177SYSCALL_DEFINE0(restart_syscall)
3178{
3179 struct restart_block *restart = ¤t->restart_block;
3180 return restart->fn(restart);
3181}
3182
3183long do_no_restart_syscall(struct restart_block *param)
3184{
3185 return -EINTR;
3186}
3187
3188static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
3189{
3190 if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
3191 sigset_t newblocked;
3192 /* A set of now blocked but previously unblocked signals. */
3193 sigandnsets(&newblocked, newset, ¤t->blocked);
3194 retarget_shared_pending(tsk, &newblocked);
3195 }
3196 tsk->blocked = *newset;
3197 recalc_sigpending();
3198}
3199
3200/**
3201 * set_current_blocked - change current->blocked mask
3202 * @newset: new mask
3203 *
3204 * It is wrong to change ->blocked directly, this helper should be used
3205 * to ensure the process can't miss a shared signal we are going to block.
3206 */
3207void set_current_blocked(sigset_t *newset)
3208{
3209 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
3210 __set_current_blocked(newset);
3211}
3212
3213void __set_current_blocked(const sigset_t *newset)
3214{
3215 struct task_struct *tsk = current;
3216
3217 /*
3218 * In case the signal mask hasn't changed, there is nothing we need
3219 * to do. The current->blocked shouldn't be modified by other task.
3220 */
3221 if (sigequalsets(&tsk->blocked, newset))
3222 return;
3223
3224 spin_lock_irq(&tsk->sighand->siglock);
3225 __set_task_blocked(tsk, newset);
3226 spin_unlock_irq(&tsk->sighand->siglock);
3227}
3228
3229/*
3230 * This is also useful for kernel threads that want to temporarily
3231 * (or permanently) block certain signals.
3232 *
3233 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
3234 * interface happily blocks "unblockable" signals like SIGKILL
3235 * and friends.
3236 */
3237int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
3238{
3239 struct task_struct *tsk = current;
3240 sigset_t newset;
3241
3242 /* Lockless, only current can change ->blocked, never from irq */
3243 if (oldset)
3244 *oldset = tsk->blocked;
3245
3246 switch (how) {
3247 case SIG_BLOCK:
3248 sigorsets(&newset, &tsk->blocked, set);
3249 break;
3250 case SIG_UNBLOCK:
3251 sigandnsets(&newset, &tsk->blocked, set);
3252 break;
3253 case SIG_SETMASK:
3254 newset = *set;
3255 break;
3256 default:
3257 return -EINVAL;
3258 }
3259
3260 __set_current_blocked(&newset);
3261 return 0;
3262}
3263EXPORT_SYMBOL(sigprocmask);
3264
3265/*
3266 * The api helps set app-provided sigmasks.
3267 *
3268 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3269 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3270 *
3271 * Note that it does set_restore_sigmask() in advance, so it must be always
3272 * paired with restore_saved_sigmask_unless() before return from syscall.
3273 */
3274int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3275{
3276 sigset_t kmask;
3277
3278 if (!umask)
3279 return 0;
3280 if (sigsetsize != sizeof(sigset_t))
3281 return -EINVAL;
3282 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3283 return -EFAULT;
3284
3285 set_restore_sigmask();
3286 current->saved_sigmask = current->blocked;
3287 set_current_blocked(&kmask);
3288
3289 return 0;
3290}
3291
3292#ifdef CONFIG_COMPAT
3293int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3294 size_t sigsetsize)
3295{
3296 sigset_t kmask;
3297
3298 if (!umask)
3299 return 0;
3300 if (sigsetsize != sizeof(compat_sigset_t))
3301 return -EINVAL;
3302 if (get_compat_sigset(&kmask, umask))
3303 return -EFAULT;
3304
3305 set_restore_sigmask();
3306 current->saved_sigmask = current->blocked;
3307 set_current_blocked(&kmask);
3308
3309 return 0;
3310}
3311#endif
3312
3313/**
3314 * sys_rt_sigprocmask - change the list of currently blocked signals
3315 * @how: whether to add, remove, or set signals
3316 * @nset: stores pending signals
3317 * @oset: previous value of signal mask if non-null
3318 * @sigsetsize: size of sigset_t type
3319 */
3320SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3321 sigset_t __user *, oset, size_t, sigsetsize)
3322{
3323 sigset_t old_set, new_set;
3324 int error;
3325
3326 /* XXX: Don't preclude handling different sized sigset_t's. */
3327 if (sigsetsize != sizeof(sigset_t))
3328 return -EINVAL;
3329
3330 old_set = current->blocked;
3331
3332 if (nset) {
3333 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3334 return -EFAULT;
3335 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3336
3337 error = sigprocmask(how, &new_set, NULL);
3338 if (error)
3339 return error;
3340 }
3341
3342 if (oset) {
3343 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3344 return -EFAULT;
3345 }
3346
3347 return 0;
3348}
3349
3350#ifdef CONFIG_COMPAT
3351COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3352 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3353{
3354 sigset_t old_set = current->blocked;
3355
3356 /* XXX: Don't preclude handling different sized sigset_t's. */
3357 if (sigsetsize != sizeof(sigset_t))
3358 return -EINVAL;
3359
3360 if (nset) {
3361 sigset_t new_set;
3362 int error;
3363 if (get_compat_sigset(&new_set, nset))
3364 return -EFAULT;
3365 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3366
3367 error = sigprocmask(how, &new_set, NULL);
3368 if (error)
3369 return error;
3370 }
3371 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3372}
3373#endif
3374
3375static void do_sigpending(sigset_t *set)
3376{
3377 spin_lock_irq(¤t->sighand->siglock);
3378 sigorsets(set, ¤t->pending.signal,
3379 ¤t->signal->shared_pending.signal);
3380 spin_unlock_irq(¤t->sighand->siglock);
3381
3382 /* Outside the lock because only this thread touches it. */
3383 sigandsets(set, ¤t->blocked, set);
3384}
3385
3386/**
3387 * sys_rt_sigpending - examine a pending signal that has been raised
3388 * while blocked
3389 * @uset: stores pending signals
3390 * @sigsetsize: size of sigset_t type or larger
3391 */
3392SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3393{
3394 sigset_t set;
3395
3396 if (sigsetsize > sizeof(*uset))
3397 return -EINVAL;
3398
3399 do_sigpending(&set);
3400
3401 if (copy_to_user(uset, &set, sigsetsize))
3402 return -EFAULT;
3403
3404 return 0;
3405}
3406
3407#ifdef CONFIG_COMPAT
3408COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3409 compat_size_t, sigsetsize)
3410{
3411 sigset_t set;
3412
3413 if (sigsetsize > sizeof(*uset))
3414 return -EINVAL;
3415
3416 do_sigpending(&set);
3417
3418 return put_compat_sigset(uset, &set, sigsetsize);
3419}
3420#endif
3421
3422static const struct {
3423 unsigned char limit, layout;
3424} sig_sicodes[] = {
3425 [SIGILL] = { NSIGILL, SIL_FAULT },
3426 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3427 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3428 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3429 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3430#if defined(SIGEMT)
3431 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3432#endif
3433 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3434 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3435 [SIGSYS] = { NSIGSYS, SIL_SYS },
3436};
3437
3438static bool known_siginfo_layout(unsigned sig, int si_code)
3439{
3440 if (si_code == SI_KERNEL)
3441 return true;
3442 else if ((si_code > SI_USER)) {
3443 if (sig_specific_sicodes(sig)) {
3444 if (si_code <= sig_sicodes[sig].limit)
3445 return true;
3446 }
3447 else if (si_code <= NSIGPOLL)
3448 return true;
3449 }
3450 else if (si_code >= SI_DETHREAD)
3451 return true;
3452 else if (si_code == SI_ASYNCNL)
3453 return true;
3454 return false;
3455}
3456
3457enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3458{
3459 enum siginfo_layout layout = SIL_KILL;
3460 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3461 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3462 (si_code <= sig_sicodes[sig].limit)) {
3463 layout = sig_sicodes[sig].layout;
3464 /* Handle the exceptions */
3465 if ((sig == SIGBUS) &&
3466 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3467 layout = SIL_FAULT_MCEERR;
3468 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3469 layout = SIL_FAULT_BNDERR;
3470#ifdef SEGV_PKUERR
3471 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3472 layout = SIL_FAULT_PKUERR;
3473#endif
3474 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3475 layout = SIL_FAULT_PERF_EVENT;
3476 else if (IS_ENABLED(CONFIG_SPARC) &&
3477 (sig == SIGILL) && (si_code == ILL_ILLTRP))
3478 layout = SIL_FAULT_TRAPNO;
3479 else if (IS_ENABLED(CONFIG_ALPHA) &&
3480 ((sig == SIGFPE) ||
3481 ((sig == SIGTRAP) && (si_code == TRAP_UNK))))
3482 layout = SIL_FAULT_TRAPNO;
3483 }
3484 else if (si_code <= NSIGPOLL)
3485 layout = SIL_POLL;
3486 } else {
3487 if (si_code == SI_TIMER)
3488 layout = SIL_TIMER;
3489 else if (si_code == SI_SIGIO)
3490 layout = SIL_POLL;
3491 else if (si_code < 0)
3492 layout = SIL_RT;
3493 }
3494 return layout;
3495}
3496
3497static inline char __user *si_expansion(const siginfo_t __user *info)
3498{
3499 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3500}
3501
3502int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3503{
3504 char __user *expansion = si_expansion(to);
3505 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3506 return -EFAULT;
3507 if (clear_user(expansion, SI_EXPANSION_SIZE))
3508 return -EFAULT;
3509 return 0;
3510}
3511
3512static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3513 const siginfo_t __user *from)
3514{
3515 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3516 char __user *expansion = si_expansion(from);
3517 char buf[SI_EXPANSION_SIZE];
3518 int i;
3519 /*
3520 * An unknown si_code might need more than
3521 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3522 * extra bytes are 0. This guarantees copy_siginfo_to_user
3523 * will return this data to userspace exactly.
3524 */
3525 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3526 return -EFAULT;
3527 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3528 if (buf[i] != 0)
3529 return -E2BIG;
3530 }
3531 }
3532 return 0;
3533}
3534
3535static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3536 const siginfo_t __user *from)
3537{
3538 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3539 return -EFAULT;
3540 to->si_signo = signo;
3541 return post_copy_siginfo_from_user(to, from);
3542}
3543
3544int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3545{
3546 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3547 return -EFAULT;
3548 return post_copy_siginfo_from_user(to, from);
3549}
3550
3551#ifdef CONFIG_COMPAT
3552/**
3553 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3554 * @to: compat siginfo destination
3555 * @from: kernel siginfo source
3556 *
3557 * Note: This function does not work properly for the SIGCHLD on x32, but
3558 * fortunately it doesn't have to. The only valid callers for this function are
3559 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3560 * The latter does not care because SIGCHLD will never cause a coredump.
3561 */
3562void copy_siginfo_to_external32(struct compat_siginfo *to,
3563 const struct kernel_siginfo *from)
3564{
3565 memset(to, 0, sizeof(*to));
3566
3567 to->si_signo = from->si_signo;
3568 to->si_errno = from->si_errno;
3569 to->si_code = from->si_code;
3570 switch(siginfo_layout(from->si_signo, from->si_code)) {
3571 case SIL_KILL:
3572 to->si_pid = from->si_pid;
3573 to->si_uid = from->si_uid;
3574 break;
3575 case SIL_TIMER:
3576 to->si_tid = from->si_tid;
3577 to->si_overrun = from->si_overrun;
3578 to->si_int = from->si_int;
3579 break;
3580 case SIL_POLL:
3581 to->si_band = from->si_band;
3582 to->si_fd = from->si_fd;
3583 break;
3584 case SIL_FAULT:
3585 to->si_addr = ptr_to_compat(from->si_addr);
3586 break;
3587 case SIL_FAULT_TRAPNO:
3588 to->si_addr = ptr_to_compat(from->si_addr);
3589 to->si_trapno = from->si_trapno;
3590 break;
3591 case SIL_FAULT_MCEERR:
3592 to->si_addr = ptr_to_compat(from->si_addr);
3593 to->si_addr_lsb = from->si_addr_lsb;
3594 break;
3595 case SIL_FAULT_BNDERR:
3596 to->si_addr = ptr_to_compat(from->si_addr);
3597 to->si_lower = ptr_to_compat(from->si_lower);
3598 to->si_upper = ptr_to_compat(from->si_upper);
3599 break;
3600 case SIL_FAULT_PKUERR:
3601 to->si_addr = ptr_to_compat(from->si_addr);
3602 to->si_pkey = from->si_pkey;
3603 break;
3604 case SIL_FAULT_PERF_EVENT:
3605 to->si_addr = ptr_to_compat(from->si_addr);
3606 to->si_perf_data = from->si_perf_data;
3607 to->si_perf_type = from->si_perf_type;
3608 to->si_perf_flags = from->si_perf_flags;
3609 break;
3610 case SIL_CHLD:
3611 to->si_pid = from->si_pid;
3612 to->si_uid = from->si_uid;
3613 to->si_status = from->si_status;
3614 to->si_utime = from->si_utime;
3615 to->si_stime = from->si_stime;
3616 break;
3617 case SIL_RT:
3618 to->si_pid = from->si_pid;
3619 to->si_uid = from->si_uid;
3620 to->si_int = from->si_int;
3621 break;
3622 case SIL_SYS:
3623 to->si_call_addr = ptr_to_compat(from->si_call_addr);
3624 to->si_syscall = from->si_syscall;
3625 to->si_arch = from->si_arch;
3626 break;
3627 }
3628}
3629
3630int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3631 const struct kernel_siginfo *from)
3632{
3633 struct compat_siginfo new;
3634
3635 copy_siginfo_to_external32(&new, from);
3636 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3637 return -EFAULT;
3638 return 0;
3639}
3640
3641static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3642 const struct compat_siginfo *from)
3643{
3644 clear_siginfo(to);
3645 to->si_signo = from->si_signo;
3646 to->si_errno = from->si_errno;
3647 to->si_code = from->si_code;
3648 switch(siginfo_layout(from->si_signo, from->si_code)) {
3649 case SIL_KILL:
3650 to->si_pid = from->si_pid;
3651 to->si_uid = from->si_uid;
3652 break;
3653 case SIL_TIMER:
3654 to->si_tid = from->si_tid;
3655 to->si_overrun = from->si_overrun;
3656 to->si_int = from->si_int;
3657 break;
3658 case SIL_POLL:
3659 to->si_band = from->si_band;
3660 to->si_fd = from->si_fd;
3661 break;
3662 case SIL_FAULT:
3663 to->si_addr = compat_ptr(from->si_addr);
3664 break;
3665 case SIL_FAULT_TRAPNO:
3666 to->si_addr = compat_ptr(from->si_addr);
3667 to->si_trapno = from->si_trapno;
3668 break;
3669 case SIL_FAULT_MCEERR:
3670 to->si_addr = compat_ptr(from->si_addr);
3671 to->si_addr_lsb = from->si_addr_lsb;
3672 break;
3673 case SIL_FAULT_BNDERR:
3674 to->si_addr = compat_ptr(from->si_addr);
3675 to->si_lower = compat_ptr(from->si_lower);
3676 to->si_upper = compat_ptr(from->si_upper);
3677 break;
3678 case SIL_FAULT_PKUERR:
3679 to->si_addr = compat_ptr(from->si_addr);
3680 to->si_pkey = from->si_pkey;
3681 break;
3682 case SIL_FAULT_PERF_EVENT:
3683 to->si_addr = compat_ptr(from->si_addr);
3684 to->si_perf_data = from->si_perf_data;
3685 to->si_perf_type = from->si_perf_type;
3686 to->si_perf_flags = from->si_perf_flags;
3687 break;
3688 case SIL_CHLD:
3689 to->si_pid = from->si_pid;
3690 to->si_uid = from->si_uid;
3691 to->si_status = from->si_status;
3692#ifdef CONFIG_X86_X32_ABI
3693 if (in_x32_syscall()) {
3694 to->si_utime = from->_sifields._sigchld_x32._utime;
3695 to->si_stime = from->_sifields._sigchld_x32._stime;
3696 } else
3697#endif
3698 {
3699 to->si_utime = from->si_utime;
3700 to->si_stime = from->si_stime;
3701 }
3702 break;
3703 case SIL_RT:
3704 to->si_pid = from->si_pid;
3705 to->si_uid = from->si_uid;
3706 to->si_int = from->si_int;
3707 break;
3708 case SIL_SYS:
3709 to->si_call_addr = compat_ptr(from->si_call_addr);
3710 to->si_syscall = from->si_syscall;
3711 to->si_arch = from->si_arch;
3712 break;
3713 }
3714 return 0;
3715}
3716
3717static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3718 const struct compat_siginfo __user *ufrom)
3719{
3720 struct compat_siginfo from;
3721
3722 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3723 return -EFAULT;
3724
3725 from.si_signo = signo;
3726 return post_copy_siginfo_from_user32(to, &from);
3727}
3728
3729int copy_siginfo_from_user32(struct kernel_siginfo *to,
3730 const struct compat_siginfo __user *ufrom)
3731{
3732 struct compat_siginfo from;
3733
3734 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3735 return -EFAULT;
3736
3737 return post_copy_siginfo_from_user32(to, &from);
3738}
3739#endif /* CONFIG_COMPAT */
3740
3741/**
3742 * do_sigtimedwait - wait for queued signals specified in @which
3743 * @which: queued signals to wait for
3744 * @info: if non-null, the signal's siginfo is returned here
3745 * @ts: upper bound on process time suspension
3746 */
3747static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3748 const struct timespec64 *ts)
3749{
3750 ktime_t *to = NULL, timeout = KTIME_MAX;
3751 struct task_struct *tsk = current;
3752 sigset_t mask = *which;
3753 enum pid_type type;
3754 int sig, ret = 0;
3755
3756 if (ts) {
3757 if (!timespec64_valid(ts))
3758 return -EINVAL;
3759 timeout = timespec64_to_ktime(*ts);
3760 to = &timeout;
3761 }
3762
3763 /*
3764 * Invert the set of allowed signals to get those we want to block.
3765 */
3766 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3767 signotset(&mask);
3768
3769 spin_lock_irq(&tsk->sighand->siglock);
3770 sig = dequeue_signal(&mask, info, &type);
3771 if (!sig && timeout) {
3772 /*
3773 * None ready, temporarily unblock those we're interested
3774 * while we are sleeping in so that we'll be awakened when
3775 * they arrive. Unblocking is always fine, we can avoid
3776 * set_current_blocked().
3777 */
3778 tsk->real_blocked = tsk->blocked;
3779 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3780 recalc_sigpending();
3781 spin_unlock_irq(&tsk->sighand->siglock);
3782
3783 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
3784 ret = schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3785 HRTIMER_MODE_REL);
3786 spin_lock_irq(&tsk->sighand->siglock);
3787 __set_task_blocked(tsk, &tsk->real_blocked);
3788 sigemptyset(&tsk->real_blocked);
3789 sig = dequeue_signal(&mask, info, &type);
3790 }
3791 spin_unlock_irq(&tsk->sighand->siglock);
3792
3793 if (sig)
3794 return sig;
3795 return ret ? -EINTR : -EAGAIN;
3796}
3797
3798/**
3799 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3800 * in @uthese
3801 * @uthese: queued signals to wait for
3802 * @uinfo: if non-null, the signal's siginfo is returned here
3803 * @uts: upper bound on process time suspension
3804 * @sigsetsize: size of sigset_t type
3805 */
3806SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3807 siginfo_t __user *, uinfo,
3808 const struct __kernel_timespec __user *, uts,
3809 size_t, sigsetsize)
3810{
3811 sigset_t these;
3812 struct timespec64 ts;
3813 kernel_siginfo_t info;
3814 int ret;
3815
3816 /* XXX: Don't preclude handling different sized sigset_t's. */
3817 if (sigsetsize != sizeof(sigset_t))
3818 return -EINVAL;
3819
3820 if (copy_from_user(&these, uthese, sizeof(these)))
3821 return -EFAULT;
3822
3823 if (uts) {
3824 if (get_timespec64(&ts, uts))
3825 return -EFAULT;
3826 }
3827
3828 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3829
3830 if (ret > 0 && uinfo) {
3831 if (copy_siginfo_to_user(uinfo, &info))
3832 ret = -EFAULT;
3833 }
3834
3835 return ret;
3836}
3837
3838#ifdef CONFIG_COMPAT_32BIT_TIME
3839SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3840 siginfo_t __user *, uinfo,
3841 const struct old_timespec32 __user *, uts,
3842 size_t, sigsetsize)
3843{
3844 sigset_t these;
3845 struct timespec64 ts;
3846 kernel_siginfo_t info;
3847 int ret;
3848
3849 if (sigsetsize != sizeof(sigset_t))
3850 return -EINVAL;
3851
3852 if (copy_from_user(&these, uthese, sizeof(these)))
3853 return -EFAULT;
3854
3855 if (uts) {
3856 if (get_old_timespec32(&ts, uts))
3857 return -EFAULT;
3858 }
3859
3860 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3861
3862 if (ret > 0 && uinfo) {
3863 if (copy_siginfo_to_user(uinfo, &info))
3864 ret = -EFAULT;
3865 }
3866
3867 return ret;
3868}
3869#endif
3870
3871#ifdef CONFIG_COMPAT
3872COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3873 struct compat_siginfo __user *, uinfo,
3874 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3875{
3876 sigset_t s;
3877 struct timespec64 t;
3878 kernel_siginfo_t info;
3879 long ret;
3880
3881 if (sigsetsize != sizeof(sigset_t))
3882 return -EINVAL;
3883
3884 if (get_compat_sigset(&s, uthese))
3885 return -EFAULT;
3886
3887 if (uts) {
3888 if (get_timespec64(&t, uts))
3889 return -EFAULT;
3890 }
3891
3892 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3893
3894 if (ret > 0 && uinfo) {
3895 if (copy_siginfo_to_user32(uinfo, &info))
3896 ret = -EFAULT;
3897 }
3898
3899 return ret;
3900}
3901
3902#ifdef CONFIG_COMPAT_32BIT_TIME
3903COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3904 struct compat_siginfo __user *, uinfo,
3905 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3906{
3907 sigset_t s;
3908 struct timespec64 t;
3909 kernel_siginfo_t info;
3910 long ret;
3911
3912 if (sigsetsize != sizeof(sigset_t))
3913 return -EINVAL;
3914
3915 if (get_compat_sigset(&s, uthese))
3916 return -EFAULT;
3917
3918 if (uts) {
3919 if (get_old_timespec32(&t, uts))
3920 return -EFAULT;
3921 }
3922
3923 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3924
3925 if (ret > 0 && uinfo) {
3926 if (copy_siginfo_to_user32(uinfo, &info))
3927 ret = -EFAULT;
3928 }
3929
3930 return ret;
3931}
3932#endif
3933#endif
3934
3935static void prepare_kill_siginfo(int sig, struct kernel_siginfo *info,
3936 enum pid_type type)
3937{
3938 clear_siginfo(info);
3939 info->si_signo = sig;
3940 info->si_errno = 0;
3941 info->si_code = (type == PIDTYPE_PID) ? SI_TKILL : SI_USER;
3942 info->si_pid = task_tgid_vnr(current);
3943 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3944}
3945
3946/**
3947 * sys_kill - send a signal to a process
3948 * @pid: the PID of the process
3949 * @sig: signal to be sent
3950 */
3951SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3952{
3953 struct kernel_siginfo info;
3954
3955 prepare_kill_siginfo(sig, &info, PIDTYPE_TGID);
3956
3957 return kill_something_info(sig, &info, pid);
3958}
3959
3960/*
3961 * Verify that the signaler and signalee either are in the same pid namespace
3962 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3963 * namespace.
3964 */
3965static bool access_pidfd_pidns(struct pid *pid)
3966{
3967 struct pid_namespace *active = task_active_pid_ns(current);
3968 struct pid_namespace *p = ns_of_pid(pid);
3969
3970 for (;;) {
3971 if (!p)
3972 return false;
3973 if (p == active)
3974 break;
3975 p = p->parent;
3976 }
3977
3978 return true;
3979}
3980
3981static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3982 siginfo_t __user *info)
3983{
3984#ifdef CONFIG_COMPAT
3985 /*
3986 * Avoid hooking up compat syscalls and instead handle necessary
3987 * conversions here. Note, this is a stop-gap measure and should not be
3988 * considered a generic solution.
3989 */
3990 if (in_compat_syscall())
3991 return copy_siginfo_from_user32(
3992 kinfo, (struct compat_siginfo __user *)info);
3993#endif
3994 return copy_siginfo_from_user(kinfo, info);
3995}
3996
3997static struct pid *pidfd_to_pid(const struct file *file)
3998{
3999 struct pid *pid;
4000
4001 pid = pidfd_pid(file);
4002 if (!IS_ERR(pid))
4003 return pid;
4004
4005 return tgid_pidfd_to_pid(file);
4006}
4007
4008#define PIDFD_SEND_SIGNAL_FLAGS \
4009 (PIDFD_SIGNAL_THREAD | PIDFD_SIGNAL_THREAD_GROUP | \
4010 PIDFD_SIGNAL_PROCESS_GROUP)
4011
4012/**
4013 * sys_pidfd_send_signal - Signal a process through a pidfd
4014 * @pidfd: file descriptor of the process
4015 * @sig: signal to send
4016 * @info: signal info
4017 * @flags: future flags
4018 *
4019 * Send the signal to the thread group or to the individual thread depending
4020 * on PIDFD_THREAD.
4021 * In the future extension to @flags may be used to override the default scope
4022 * of @pidfd.
4023 *
4024 * Return: 0 on success, negative errno on failure
4025 */
4026SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
4027 siginfo_t __user *, info, unsigned int, flags)
4028{
4029 int ret;
4030 struct pid *pid;
4031 kernel_siginfo_t kinfo;
4032 enum pid_type type;
4033
4034 /* Enforce flags be set to 0 until we add an extension. */
4035 if (flags & ~PIDFD_SEND_SIGNAL_FLAGS)
4036 return -EINVAL;
4037
4038 /* Ensure that only a single signal scope determining flag is set. */
4039 if (hweight32(flags & PIDFD_SEND_SIGNAL_FLAGS) > 1)
4040 return -EINVAL;
4041
4042 CLASS(fd, f)(pidfd);
4043 if (fd_empty(f))
4044 return -EBADF;
4045
4046 /* Is this a pidfd? */
4047 pid = pidfd_to_pid(fd_file(f));
4048 if (IS_ERR(pid))
4049 return PTR_ERR(pid);
4050
4051 if (!access_pidfd_pidns(pid))
4052 return -EINVAL;
4053
4054 switch (flags) {
4055 case 0:
4056 /* Infer scope from the type of pidfd. */
4057 if (fd_file(f)->f_flags & PIDFD_THREAD)
4058 type = PIDTYPE_PID;
4059 else
4060 type = PIDTYPE_TGID;
4061 break;
4062 case PIDFD_SIGNAL_THREAD:
4063 type = PIDTYPE_PID;
4064 break;
4065 case PIDFD_SIGNAL_THREAD_GROUP:
4066 type = PIDTYPE_TGID;
4067 break;
4068 case PIDFD_SIGNAL_PROCESS_GROUP:
4069 type = PIDTYPE_PGID;
4070 break;
4071 }
4072
4073 if (info) {
4074 ret = copy_siginfo_from_user_any(&kinfo, info);
4075 if (unlikely(ret))
4076 return ret;
4077
4078 if (unlikely(sig != kinfo.si_signo))
4079 return -EINVAL;
4080
4081 /* Only allow sending arbitrary signals to yourself. */
4082 if ((task_pid(current) != pid || type > PIDTYPE_TGID) &&
4083 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
4084 return -EPERM;
4085 } else {
4086 prepare_kill_siginfo(sig, &kinfo, type);
4087 }
4088
4089 if (type == PIDTYPE_PGID)
4090 return kill_pgrp_info(sig, &kinfo, pid);
4091 else
4092 return kill_pid_info_type(sig, &kinfo, pid, type);
4093}
4094
4095static int
4096do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
4097{
4098 struct task_struct *p;
4099 int error = -ESRCH;
4100
4101 rcu_read_lock();
4102 p = find_task_by_vpid(pid);
4103 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
4104 error = check_kill_permission(sig, info, p);
4105 /*
4106 * The null signal is a permissions and process existence
4107 * probe. No signal is actually delivered.
4108 */
4109 if (!error && sig) {
4110 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
4111 /*
4112 * If lock_task_sighand() failed we pretend the task
4113 * dies after receiving the signal. The window is tiny,
4114 * and the signal is private anyway.
4115 */
4116 if (unlikely(error == -ESRCH))
4117 error = 0;
4118 }
4119 }
4120 rcu_read_unlock();
4121
4122 return error;
4123}
4124
4125static int do_tkill(pid_t tgid, pid_t pid, int sig)
4126{
4127 struct kernel_siginfo info;
4128
4129 prepare_kill_siginfo(sig, &info, PIDTYPE_PID);
4130
4131 return do_send_specific(tgid, pid, sig, &info);
4132}
4133
4134/**
4135 * sys_tgkill - send signal to one specific thread
4136 * @tgid: the thread group ID of the thread
4137 * @pid: the PID of the thread
4138 * @sig: signal to be sent
4139 *
4140 * This syscall also checks the @tgid and returns -ESRCH even if the PID
4141 * exists but it's not belonging to the target process anymore. This
4142 * method solves the problem of threads exiting and PIDs getting reused.
4143 */
4144SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
4145{
4146 /* This is only valid for single tasks */
4147 if (pid <= 0 || tgid <= 0)
4148 return -EINVAL;
4149
4150 return do_tkill(tgid, pid, sig);
4151}
4152
4153/**
4154 * sys_tkill - send signal to one specific task
4155 * @pid: the PID of the task
4156 * @sig: signal to be sent
4157 *
4158 * Send a signal to only one task, even if it's a CLONE_THREAD task.
4159 */
4160SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
4161{
4162 /* This is only valid for single tasks */
4163 if (pid <= 0)
4164 return -EINVAL;
4165
4166 return do_tkill(0, pid, sig);
4167}
4168
4169static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
4170{
4171 /* Not even root can pretend to send signals from the kernel.
4172 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4173 */
4174 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4175 (task_pid_vnr(current) != pid))
4176 return -EPERM;
4177
4178 /* POSIX.1b doesn't mention process groups. */
4179 return kill_proc_info(sig, info, pid);
4180}
4181
4182/**
4183 * sys_rt_sigqueueinfo - send signal information to a signal
4184 * @pid: the PID of the thread
4185 * @sig: signal to be sent
4186 * @uinfo: signal info to be sent
4187 */
4188SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
4189 siginfo_t __user *, uinfo)
4190{
4191 kernel_siginfo_t info;
4192 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4193 if (unlikely(ret))
4194 return ret;
4195 return do_rt_sigqueueinfo(pid, sig, &info);
4196}
4197
4198#ifdef CONFIG_COMPAT
4199COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
4200 compat_pid_t, pid,
4201 int, sig,
4202 struct compat_siginfo __user *, uinfo)
4203{
4204 kernel_siginfo_t info;
4205 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4206 if (unlikely(ret))
4207 return ret;
4208 return do_rt_sigqueueinfo(pid, sig, &info);
4209}
4210#endif
4211
4212static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
4213{
4214 /* This is only valid for single tasks */
4215 if (pid <= 0 || tgid <= 0)
4216 return -EINVAL;
4217
4218 /* Not even root can pretend to send signals from the kernel.
4219 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4220 */
4221 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4222 (task_pid_vnr(current) != pid))
4223 return -EPERM;
4224
4225 return do_send_specific(tgid, pid, sig, info);
4226}
4227
4228SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
4229 siginfo_t __user *, uinfo)
4230{
4231 kernel_siginfo_t info;
4232 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4233 if (unlikely(ret))
4234 return ret;
4235 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4236}
4237
4238#ifdef CONFIG_COMPAT
4239COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
4240 compat_pid_t, tgid,
4241 compat_pid_t, pid,
4242 int, sig,
4243 struct compat_siginfo __user *, uinfo)
4244{
4245 kernel_siginfo_t info;
4246 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4247 if (unlikely(ret))
4248 return ret;
4249 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4250}
4251#endif
4252
4253/*
4254 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
4255 */
4256void kernel_sigaction(int sig, __sighandler_t action)
4257{
4258 spin_lock_irq(¤t->sighand->siglock);
4259 current->sighand->action[sig - 1].sa.sa_handler = action;
4260 if (action == SIG_IGN) {
4261 sigset_t mask;
4262
4263 sigemptyset(&mask);
4264 sigaddset(&mask, sig);
4265
4266 flush_sigqueue_mask(current, &mask, ¤t->signal->shared_pending);
4267 flush_sigqueue_mask(current, &mask, ¤t->pending);
4268 recalc_sigpending();
4269 }
4270 spin_unlock_irq(¤t->sighand->siglock);
4271}
4272EXPORT_SYMBOL(kernel_sigaction);
4273
4274void __weak sigaction_compat_abi(struct k_sigaction *act,
4275 struct k_sigaction *oact)
4276{
4277}
4278
4279int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4280{
4281 struct task_struct *p = current, *t;
4282 struct k_sigaction *k;
4283 sigset_t mask;
4284
4285 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4286 return -EINVAL;
4287
4288 k = &p->sighand->action[sig-1];
4289
4290 spin_lock_irq(&p->sighand->siglock);
4291 if (k->sa.sa_flags & SA_IMMUTABLE) {
4292 spin_unlock_irq(&p->sighand->siglock);
4293 return -EINVAL;
4294 }
4295 if (oact)
4296 *oact = *k;
4297
4298 /*
4299 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4300 * e.g. by having an architecture use the bit in their uapi.
4301 */
4302 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4303
4304 /*
4305 * Clear unknown flag bits in order to allow userspace to detect missing
4306 * support for flag bits and to allow the kernel to use non-uapi bits
4307 * internally.
4308 */
4309 if (act)
4310 act->sa.sa_flags &= UAPI_SA_FLAGS;
4311 if (oact)
4312 oact->sa.sa_flags &= UAPI_SA_FLAGS;
4313
4314 sigaction_compat_abi(act, oact);
4315
4316 if (act) {
4317 bool was_ignored = k->sa.sa_handler == SIG_IGN;
4318
4319 sigdelsetmask(&act->sa.sa_mask,
4320 sigmask(SIGKILL) | sigmask(SIGSTOP));
4321 *k = *act;
4322 /*
4323 * POSIX 3.3.1.3:
4324 * "Setting a signal action to SIG_IGN for a signal that is
4325 * pending shall cause the pending signal to be discarded,
4326 * whether or not it is blocked."
4327 *
4328 * "Setting a signal action to SIG_DFL for a signal that is
4329 * pending and whose default action is to ignore the signal
4330 * (for example, SIGCHLD), shall cause the pending signal to
4331 * be discarded, whether or not it is blocked"
4332 */
4333 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4334 sigemptyset(&mask);
4335 sigaddset(&mask, sig);
4336 flush_sigqueue_mask(p, &mask, &p->signal->shared_pending);
4337 for_each_thread(p, t)
4338 flush_sigqueue_mask(p, &mask, &t->pending);
4339 } else if (was_ignored) {
4340 posixtimer_sig_unignore(p, sig);
4341 }
4342 }
4343
4344 spin_unlock_irq(&p->sighand->siglock);
4345 return 0;
4346}
4347
4348#ifdef CONFIG_DYNAMIC_SIGFRAME
4349static inline void sigaltstack_lock(void)
4350 __acquires(¤t->sighand->siglock)
4351{
4352 spin_lock_irq(¤t->sighand->siglock);
4353}
4354
4355static inline void sigaltstack_unlock(void)
4356 __releases(¤t->sighand->siglock)
4357{
4358 spin_unlock_irq(¤t->sighand->siglock);
4359}
4360#else
4361static inline void sigaltstack_lock(void) { }
4362static inline void sigaltstack_unlock(void) { }
4363#endif
4364
4365static int
4366do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4367 size_t min_ss_size)
4368{
4369 struct task_struct *t = current;
4370 int ret = 0;
4371
4372 if (oss) {
4373 memset(oss, 0, sizeof(stack_t));
4374 oss->ss_sp = (void __user *) t->sas_ss_sp;
4375 oss->ss_size = t->sas_ss_size;
4376 oss->ss_flags = sas_ss_flags(sp) |
4377 (current->sas_ss_flags & SS_FLAG_BITS);
4378 }
4379
4380 if (ss) {
4381 void __user *ss_sp = ss->ss_sp;
4382 size_t ss_size = ss->ss_size;
4383 unsigned ss_flags = ss->ss_flags;
4384 int ss_mode;
4385
4386 if (unlikely(on_sig_stack(sp)))
4387 return -EPERM;
4388
4389 ss_mode = ss_flags & ~SS_FLAG_BITS;
4390 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4391 ss_mode != 0))
4392 return -EINVAL;
4393
4394 /*
4395 * Return before taking any locks if no actual
4396 * sigaltstack changes were requested.
4397 */
4398 if (t->sas_ss_sp == (unsigned long)ss_sp &&
4399 t->sas_ss_size == ss_size &&
4400 t->sas_ss_flags == ss_flags)
4401 return 0;
4402
4403 sigaltstack_lock();
4404 if (ss_mode == SS_DISABLE) {
4405 ss_size = 0;
4406 ss_sp = NULL;
4407 } else {
4408 if (unlikely(ss_size < min_ss_size))
4409 ret = -ENOMEM;
4410 if (!sigaltstack_size_valid(ss_size))
4411 ret = -ENOMEM;
4412 }
4413 if (!ret) {
4414 t->sas_ss_sp = (unsigned long) ss_sp;
4415 t->sas_ss_size = ss_size;
4416 t->sas_ss_flags = ss_flags;
4417 }
4418 sigaltstack_unlock();
4419 }
4420 return ret;
4421}
4422
4423SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4424{
4425 stack_t new, old;
4426 int err;
4427 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4428 return -EFAULT;
4429 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4430 current_user_stack_pointer(),
4431 MINSIGSTKSZ);
4432 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4433 err = -EFAULT;
4434 return err;
4435}
4436
4437int restore_altstack(const stack_t __user *uss)
4438{
4439 stack_t new;
4440 if (copy_from_user(&new, uss, sizeof(stack_t)))
4441 return -EFAULT;
4442 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4443 MINSIGSTKSZ);
4444 /* squash all but EFAULT for now */
4445 return 0;
4446}
4447
4448int __save_altstack(stack_t __user *uss, unsigned long sp)
4449{
4450 struct task_struct *t = current;
4451 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4452 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4453 __put_user(t->sas_ss_size, &uss->ss_size);
4454 return err;
4455}
4456
4457#ifdef CONFIG_COMPAT
4458static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4459 compat_stack_t __user *uoss_ptr)
4460{
4461 stack_t uss, uoss;
4462 int ret;
4463
4464 if (uss_ptr) {
4465 compat_stack_t uss32;
4466 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4467 return -EFAULT;
4468 uss.ss_sp = compat_ptr(uss32.ss_sp);
4469 uss.ss_flags = uss32.ss_flags;
4470 uss.ss_size = uss32.ss_size;
4471 }
4472 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4473 compat_user_stack_pointer(),
4474 COMPAT_MINSIGSTKSZ);
4475 if (ret >= 0 && uoss_ptr) {
4476 compat_stack_t old;
4477 memset(&old, 0, sizeof(old));
4478 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4479 old.ss_flags = uoss.ss_flags;
4480 old.ss_size = uoss.ss_size;
4481 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4482 ret = -EFAULT;
4483 }
4484 return ret;
4485}
4486
4487COMPAT_SYSCALL_DEFINE2(sigaltstack,
4488 const compat_stack_t __user *, uss_ptr,
4489 compat_stack_t __user *, uoss_ptr)
4490{
4491 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4492}
4493
4494int compat_restore_altstack(const compat_stack_t __user *uss)
4495{
4496 int err = do_compat_sigaltstack(uss, NULL);
4497 /* squash all but -EFAULT for now */
4498 return err == -EFAULT ? err : 0;
4499}
4500
4501int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4502{
4503 int err;
4504 struct task_struct *t = current;
4505 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4506 &uss->ss_sp) |
4507 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4508 __put_user(t->sas_ss_size, &uss->ss_size);
4509 return err;
4510}
4511#endif
4512
4513#ifdef __ARCH_WANT_SYS_SIGPENDING
4514
4515/**
4516 * sys_sigpending - examine pending signals
4517 * @uset: where mask of pending signal is returned
4518 */
4519SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4520{
4521 sigset_t set;
4522
4523 if (sizeof(old_sigset_t) > sizeof(*uset))
4524 return -EINVAL;
4525
4526 do_sigpending(&set);
4527
4528 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4529 return -EFAULT;
4530
4531 return 0;
4532}
4533
4534#ifdef CONFIG_COMPAT
4535COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4536{
4537 sigset_t set;
4538
4539 do_sigpending(&set);
4540
4541 return put_user(set.sig[0], set32);
4542}
4543#endif
4544
4545#endif
4546
4547#ifdef __ARCH_WANT_SYS_SIGPROCMASK
4548/**
4549 * sys_sigprocmask - examine and change blocked signals
4550 * @how: whether to add, remove, or set signals
4551 * @nset: signals to add or remove (if non-null)
4552 * @oset: previous value of signal mask if non-null
4553 *
4554 * Some platforms have their own version with special arguments;
4555 * others support only sys_rt_sigprocmask.
4556 */
4557
4558SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4559 old_sigset_t __user *, oset)
4560{
4561 old_sigset_t old_set, new_set;
4562 sigset_t new_blocked;
4563
4564 old_set = current->blocked.sig[0];
4565
4566 if (nset) {
4567 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4568 return -EFAULT;
4569
4570 new_blocked = current->blocked;
4571
4572 switch (how) {
4573 case SIG_BLOCK:
4574 sigaddsetmask(&new_blocked, new_set);
4575 break;
4576 case SIG_UNBLOCK:
4577 sigdelsetmask(&new_blocked, new_set);
4578 break;
4579 case SIG_SETMASK:
4580 new_blocked.sig[0] = new_set;
4581 break;
4582 default:
4583 return -EINVAL;
4584 }
4585
4586 set_current_blocked(&new_blocked);
4587 }
4588
4589 if (oset) {
4590 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4591 return -EFAULT;
4592 }
4593
4594 return 0;
4595}
4596#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4597
4598#ifndef CONFIG_ODD_RT_SIGACTION
4599/**
4600 * sys_rt_sigaction - alter an action taken by a process
4601 * @sig: signal to be sent
4602 * @act: new sigaction
4603 * @oact: used to save the previous sigaction
4604 * @sigsetsize: size of sigset_t type
4605 */
4606SYSCALL_DEFINE4(rt_sigaction, int, sig,
4607 const struct sigaction __user *, act,
4608 struct sigaction __user *, oact,
4609 size_t, sigsetsize)
4610{
4611 struct k_sigaction new_sa, old_sa;
4612 int ret;
4613
4614 /* XXX: Don't preclude handling different sized sigset_t's. */
4615 if (sigsetsize != sizeof(sigset_t))
4616 return -EINVAL;
4617
4618 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4619 return -EFAULT;
4620
4621 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4622 if (ret)
4623 return ret;
4624
4625 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4626 return -EFAULT;
4627
4628 return 0;
4629}
4630#ifdef CONFIG_COMPAT
4631COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4632 const struct compat_sigaction __user *, act,
4633 struct compat_sigaction __user *, oact,
4634 compat_size_t, sigsetsize)
4635{
4636 struct k_sigaction new_ka, old_ka;
4637#ifdef __ARCH_HAS_SA_RESTORER
4638 compat_uptr_t restorer;
4639#endif
4640 int ret;
4641
4642 /* XXX: Don't preclude handling different sized sigset_t's. */
4643 if (sigsetsize != sizeof(compat_sigset_t))
4644 return -EINVAL;
4645
4646 if (act) {
4647 compat_uptr_t handler;
4648 ret = get_user(handler, &act->sa_handler);
4649 new_ka.sa.sa_handler = compat_ptr(handler);
4650#ifdef __ARCH_HAS_SA_RESTORER
4651 ret |= get_user(restorer, &act->sa_restorer);
4652 new_ka.sa.sa_restorer = compat_ptr(restorer);
4653#endif
4654 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4655 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4656 if (ret)
4657 return -EFAULT;
4658 }
4659
4660 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4661 if (!ret && oact) {
4662 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4663 &oact->sa_handler);
4664 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4665 sizeof(oact->sa_mask));
4666 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4667#ifdef __ARCH_HAS_SA_RESTORER
4668 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4669 &oact->sa_restorer);
4670#endif
4671 }
4672 return ret;
4673}
4674#endif
4675#endif /* !CONFIG_ODD_RT_SIGACTION */
4676
4677#ifdef CONFIG_OLD_SIGACTION
4678SYSCALL_DEFINE3(sigaction, int, sig,
4679 const struct old_sigaction __user *, act,
4680 struct old_sigaction __user *, oact)
4681{
4682 struct k_sigaction new_ka, old_ka;
4683 int ret;
4684
4685 if (act) {
4686 old_sigset_t mask;
4687 if (!access_ok(act, sizeof(*act)) ||
4688 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4689 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4690 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4691 __get_user(mask, &act->sa_mask))
4692 return -EFAULT;
4693#ifdef __ARCH_HAS_KA_RESTORER
4694 new_ka.ka_restorer = NULL;
4695#endif
4696 siginitset(&new_ka.sa.sa_mask, mask);
4697 }
4698
4699 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4700
4701 if (!ret && oact) {
4702 if (!access_ok(oact, sizeof(*oact)) ||
4703 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4704 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4705 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4706 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4707 return -EFAULT;
4708 }
4709
4710 return ret;
4711}
4712#endif
4713#ifdef CONFIG_COMPAT_OLD_SIGACTION
4714COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4715 const struct compat_old_sigaction __user *, act,
4716 struct compat_old_sigaction __user *, oact)
4717{
4718 struct k_sigaction new_ka, old_ka;
4719 int ret;
4720 compat_old_sigset_t mask;
4721 compat_uptr_t handler, restorer;
4722
4723 if (act) {
4724 if (!access_ok(act, sizeof(*act)) ||
4725 __get_user(handler, &act->sa_handler) ||
4726 __get_user(restorer, &act->sa_restorer) ||
4727 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4728 __get_user(mask, &act->sa_mask))
4729 return -EFAULT;
4730
4731#ifdef __ARCH_HAS_KA_RESTORER
4732 new_ka.ka_restorer = NULL;
4733#endif
4734 new_ka.sa.sa_handler = compat_ptr(handler);
4735 new_ka.sa.sa_restorer = compat_ptr(restorer);
4736 siginitset(&new_ka.sa.sa_mask, mask);
4737 }
4738
4739 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4740
4741 if (!ret && oact) {
4742 if (!access_ok(oact, sizeof(*oact)) ||
4743 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4744 &oact->sa_handler) ||
4745 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4746 &oact->sa_restorer) ||
4747 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4748 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4749 return -EFAULT;
4750 }
4751 return ret;
4752}
4753#endif
4754
4755#ifdef CONFIG_SGETMASK_SYSCALL
4756
4757/*
4758 * For backwards compatibility. Functionality superseded by sigprocmask.
4759 */
4760SYSCALL_DEFINE0(sgetmask)
4761{
4762 /* SMP safe */
4763 return current->blocked.sig[0];
4764}
4765
4766SYSCALL_DEFINE1(ssetmask, int, newmask)
4767{
4768 int old = current->blocked.sig[0];
4769 sigset_t newset;
4770
4771 siginitset(&newset, newmask);
4772 set_current_blocked(&newset);
4773
4774 return old;
4775}
4776#endif /* CONFIG_SGETMASK_SYSCALL */
4777
4778#ifdef __ARCH_WANT_SYS_SIGNAL
4779/*
4780 * For backwards compatibility. Functionality superseded by sigaction.
4781 */
4782SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4783{
4784 struct k_sigaction new_sa, old_sa;
4785 int ret;
4786
4787 new_sa.sa.sa_handler = handler;
4788 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4789 sigemptyset(&new_sa.sa.sa_mask);
4790
4791 ret = do_sigaction(sig, &new_sa, &old_sa);
4792
4793 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4794}
4795#endif /* __ARCH_WANT_SYS_SIGNAL */
4796
4797#ifdef __ARCH_WANT_SYS_PAUSE
4798
4799SYSCALL_DEFINE0(pause)
4800{
4801 while (!signal_pending(current)) {
4802 __set_current_state(TASK_INTERRUPTIBLE);
4803 schedule();
4804 }
4805 return -ERESTARTNOHAND;
4806}
4807
4808#endif
4809
4810static int sigsuspend(sigset_t *set)
4811{
4812 current->saved_sigmask = current->blocked;
4813 set_current_blocked(set);
4814
4815 while (!signal_pending(current)) {
4816 __set_current_state(TASK_INTERRUPTIBLE);
4817 schedule();
4818 }
4819 set_restore_sigmask();
4820 return -ERESTARTNOHAND;
4821}
4822
4823/**
4824 * sys_rt_sigsuspend - replace the signal mask for a value with the
4825 * @unewset value until a signal is received
4826 * @unewset: new signal mask value
4827 * @sigsetsize: size of sigset_t type
4828 */
4829SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4830{
4831 sigset_t newset;
4832
4833 /* XXX: Don't preclude handling different sized sigset_t's. */
4834 if (sigsetsize != sizeof(sigset_t))
4835 return -EINVAL;
4836
4837 if (copy_from_user(&newset, unewset, sizeof(newset)))
4838 return -EFAULT;
4839 return sigsuspend(&newset);
4840}
4841
4842#ifdef CONFIG_COMPAT
4843COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4844{
4845 sigset_t newset;
4846
4847 /* XXX: Don't preclude handling different sized sigset_t's. */
4848 if (sigsetsize != sizeof(sigset_t))
4849 return -EINVAL;
4850
4851 if (get_compat_sigset(&newset, unewset))
4852 return -EFAULT;
4853 return sigsuspend(&newset);
4854}
4855#endif
4856
4857#ifdef CONFIG_OLD_SIGSUSPEND
4858SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4859{
4860 sigset_t blocked;
4861 siginitset(&blocked, mask);
4862 return sigsuspend(&blocked);
4863}
4864#endif
4865#ifdef CONFIG_OLD_SIGSUSPEND3
4866SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4867{
4868 sigset_t blocked;
4869 siginitset(&blocked, mask);
4870 return sigsuspend(&blocked);
4871}
4872#endif
4873
4874__weak const char *arch_vma_name(struct vm_area_struct *vma)
4875{
4876 return NULL;
4877}
4878
4879static inline void siginfo_buildtime_checks(void)
4880{
4881 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4882
4883 /* Verify the offsets in the two siginfos match */
4884#define CHECK_OFFSET(field) \
4885 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4886
4887 /* kill */
4888 CHECK_OFFSET(si_pid);
4889 CHECK_OFFSET(si_uid);
4890
4891 /* timer */
4892 CHECK_OFFSET(si_tid);
4893 CHECK_OFFSET(si_overrun);
4894 CHECK_OFFSET(si_value);
4895
4896 /* rt */
4897 CHECK_OFFSET(si_pid);
4898 CHECK_OFFSET(si_uid);
4899 CHECK_OFFSET(si_value);
4900
4901 /* sigchld */
4902 CHECK_OFFSET(si_pid);
4903 CHECK_OFFSET(si_uid);
4904 CHECK_OFFSET(si_status);
4905 CHECK_OFFSET(si_utime);
4906 CHECK_OFFSET(si_stime);
4907
4908 /* sigfault */
4909 CHECK_OFFSET(si_addr);
4910 CHECK_OFFSET(si_trapno);
4911 CHECK_OFFSET(si_addr_lsb);
4912 CHECK_OFFSET(si_lower);
4913 CHECK_OFFSET(si_upper);
4914 CHECK_OFFSET(si_pkey);
4915 CHECK_OFFSET(si_perf_data);
4916 CHECK_OFFSET(si_perf_type);
4917 CHECK_OFFSET(si_perf_flags);
4918
4919 /* sigpoll */
4920 CHECK_OFFSET(si_band);
4921 CHECK_OFFSET(si_fd);
4922
4923 /* sigsys */
4924 CHECK_OFFSET(si_call_addr);
4925 CHECK_OFFSET(si_syscall);
4926 CHECK_OFFSET(si_arch);
4927#undef CHECK_OFFSET
4928
4929 /* usb asyncio */
4930 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4931 offsetof(struct siginfo, si_addr));
4932 if (sizeof(int) == sizeof(void __user *)) {
4933 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4934 sizeof(void __user *));
4935 } else {
4936 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4937 sizeof_field(struct siginfo, si_uid)) !=
4938 sizeof(void __user *));
4939 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4940 offsetof(struct siginfo, si_uid));
4941 }
4942#ifdef CONFIG_COMPAT
4943 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4944 offsetof(struct compat_siginfo, si_addr));
4945 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4946 sizeof(compat_uptr_t));
4947 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4948 sizeof_field(struct siginfo, si_pid));
4949#endif
4950}
4951
4952#if defined(CONFIG_SYSCTL)
4953static struct ctl_table signal_debug_table[] = {
4954#ifdef CONFIG_SYSCTL_EXCEPTION_TRACE
4955 {
4956 .procname = "exception-trace",
4957 .data = &show_unhandled_signals,
4958 .maxlen = sizeof(int),
4959 .mode = 0644,
4960 .proc_handler = proc_dointvec
4961 },
4962#endif
4963};
4964
4965static int __init init_signal_sysctls(void)
4966{
4967 register_sysctl_init("debug", signal_debug_table);
4968 return 0;
4969}
4970early_initcall(init_signal_sysctls);
4971#endif /* CONFIG_SYSCTL */
4972
4973void __init signals_init(void)
4974{
4975 siginfo_buildtime_checks();
4976
4977 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC | SLAB_ACCOUNT);
4978}
4979
4980#ifdef CONFIG_KGDB_KDB
4981#include <linux/kdb.h>
4982/*
4983 * kdb_send_sig - Allows kdb to send signals without exposing
4984 * signal internals. This function checks if the required locks are
4985 * available before calling the main signal code, to avoid kdb
4986 * deadlocks.
4987 */
4988void kdb_send_sig(struct task_struct *t, int sig)
4989{
4990 static struct task_struct *kdb_prev_t;
4991 int new_t, ret;
4992 if (!spin_trylock(&t->sighand->siglock)) {
4993 kdb_printf("Can't do kill command now.\n"
4994 "The sigmask lock is held somewhere else in "
4995 "kernel, try again later\n");
4996 return;
4997 }
4998 new_t = kdb_prev_t != t;
4999 kdb_prev_t = t;
5000 if (!task_is_running(t) && new_t) {
5001 spin_unlock(&t->sighand->siglock);
5002 kdb_printf("Process is not RUNNING, sending a signal from "
5003 "kdb risks deadlock\n"
5004 "on the run queue locks. "
5005 "The signal has _not_ been sent.\n"
5006 "Reissue the kill command if you want to risk "
5007 "the deadlock.\n");
5008 return;
5009 }
5010 ret = send_signal_locked(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
5011 spin_unlock(&t->sighand->siglock);
5012 if (ret)
5013 kdb_printf("Fail to deliver Signal %d to process %d.\n",
5014 sig, t->pid);
5015 else
5016 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
5017}
5018#endif /* CONFIG_KGDB_KDB */