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/*
2 * linux/kernel/signal.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * 1997-11-02 Modified for POSIX.1b signals by Richard Henderson
7 *
8 * 2003-06-02 Jim Houston - Concurrent Computer Corp.
9 * Changes to use preallocated sigqueue structures
10 * to allow signals to be sent reliably.
11 */
12
13#include <linux/slab.h>
14#include <linux/module.h>
15#include <linux/init.h>
16#include <linux/sched.h>
17#include <linux/fs.h>
18#include <linux/tty.h>
19#include <linux/binfmts.h>
20#include <linux/security.h>
21#include <linux/syscalls.h>
22#include <linux/ptrace.h>
23#include <linux/signal.h>
24#include <linux/signalfd.h>
25#include <linux/ratelimit.h>
26#include <linux/tracehook.h>
27#include <linux/capability.h>
28#include <linux/freezer.h>
29#include <linux/pid_namespace.h>
30#include <linux/nsproxy.h>
31#define CREATE_TRACE_POINTS
32#include <trace/events/signal.h>
33
34#include <asm/param.h>
35#include <asm/uaccess.h>
36#include <asm/unistd.h>
37#include <asm/siginfo.h>
38#include "audit.h" /* audit_signal_info() */
39
40/*
41 * SLAB caches for signal bits.
42 */
43
44static struct kmem_cache *sigqueue_cachep;
45
46int print_fatal_signals __read_mostly;
47
48static void __user *sig_handler(struct task_struct *t, int sig)
49{
50 return t->sighand->action[sig - 1].sa.sa_handler;
51}
52
53static int sig_handler_ignored(void __user *handler, int sig)
54{
55 /* Is it explicitly or implicitly ignored? */
56 return handler == SIG_IGN ||
57 (handler == SIG_DFL && sig_kernel_ignore(sig));
58}
59
60static int sig_task_ignored(struct task_struct *t, int sig,
61 int from_ancestor_ns)
62{
63 void __user *handler;
64
65 handler = sig_handler(t, sig);
66
67 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
68 handler == SIG_DFL && !from_ancestor_ns)
69 return 1;
70
71 return sig_handler_ignored(handler, sig);
72}
73
74static int sig_ignored(struct task_struct *t, int sig, int from_ancestor_ns)
75{
76 /*
77 * Blocked signals are never ignored, since the
78 * signal handler may change by the time it is
79 * unblocked.
80 */
81 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
82 return 0;
83
84 if (!sig_task_ignored(t, sig, from_ancestor_ns))
85 return 0;
86
87 /*
88 * Tracers may want to know about even ignored signals.
89 */
90 return !t->ptrace;
91}
92
93/*
94 * Re-calculate pending state from the set of locally pending
95 * signals, globally pending signals, and blocked signals.
96 */
97static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
98{
99 unsigned long ready;
100 long i;
101
102 switch (_NSIG_WORDS) {
103 default:
104 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
105 ready |= signal->sig[i] &~ blocked->sig[i];
106 break;
107
108 case 4: ready = signal->sig[3] &~ blocked->sig[3];
109 ready |= signal->sig[2] &~ blocked->sig[2];
110 ready |= signal->sig[1] &~ blocked->sig[1];
111 ready |= signal->sig[0] &~ blocked->sig[0];
112 break;
113
114 case 2: ready = signal->sig[1] &~ blocked->sig[1];
115 ready |= signal->sig[0] &~ blocked->sig[0];
116 break;
117
118 case 1: ready = signal->sig[0] &~ blocked->sig[0];
119 }
120 return ready != 0;
121}
122
123#define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
124
125static int recalc_sigpending_tsk(struct task_struct *t)
126{
127 if ((t->jobctl & JOBCTL_PENDING_MASK) ||
128 PENDING(&t->pending, &t->blocked) ||
129 PENDING(&t->signal->shared_pending, &t->blocked)) {
130 set_tsk_thread_flag(t, TIF_SIGPENDING);
131 return 1;
132 }
133 /*
134 * We must never clear the flag in another thread, or in current
135 * when it's possible the current syscall is returning -ERESTART*.
136 * So we don't clear it here, and only callers who know they should do.
137 */
138 return 0;
139}
140
141/*
142 * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
143 * This is superfluous when called on current, the wakeup is a harmless no-op.
144 */
145void recalc_sigpending_and_wake(struct task_struct *t)
146{
147 if (recalc_sigpending_tsk(t))
148 signal_wake_up(t, 0);
149}
150
151void recalc_sigpending(void)
152{
153 if (!recalc_sigpending_tsk(current) && !freezing(current))
154 clear_thread_flag(TIF_SIGPENDING);
155
156}
157
158/* Given the mask, find the first available signal that should be serviced. */
159
160#define SYNCHRONOUS_MASK \
161 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
162 sigmask(SIGTRAP) | sigmask(SIGFPE))
163
164int next_signal(struct sigpending *pending, sigset_t *mask)
165{
166 unsigned long i, *s, *m, x;
167 int sig = 0;
168
169 s = pending->signal.sig;
170 m = mask->sig;
171
172 /*
173 * Handle the first word specially: it contains the
174 * synchronous signals that need to be dequeued first.
175 */
176 x = *s &~ *m;
177 if (x) {
178 if (x & SYNCHRONOUS_MASK)
179 x &= SYNCHRONOUS_MASK;
180 sig = ffz(~x) + 1;
181 return sig;
182 }
183
184 switch (_NSIG_WORDS) {
185 default:
186 for (i = 1; i < _NSIG_WORDS; ++i) {
187 x = *++s &~ *++m;
188 if (!x)
189 continue;
190 sig = ffz(~x) + i*_NSIG_BPW + 1;
191 break;
192 }
193 break;
194
195 case 2:
196 x = s[1] &~ m[1];
197 if (!x)
198 break;
199 sig = ffz(~x) + _NSIG_BPW + 1;
200 break;
201
202 case 1:
203 /* Nothing to do */
204 break;
205 }
206
207 return sig;
208}
209
210static inline void print_dropped_signal(int sig)
211{
212 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
213
214 if (!print_fatal_signals)
215 return;
216
217 if (!__ratelimit(&ratelimit_state))
218 return;
219
220 printk(KERN_INFO "%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
221 current->comm, current->pid, sig);
222}
223
224/**
225 * task_set_jobctl_pending - set jobctl pending bits
226 * @task: target task
227 * @mask: pending bits to set
228 *
229 * Clear @mask from @task->jobctl. @mask must be subset of
230 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
231 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
232 * cleared. If @task is already being killed or exiting, this function
233 * becomes noop.
234 *
235 * CONTEXT:
236 * Must be called with @task->sighand->siglock held.
237 *
238 * RETURNS:
239 * %true if @mask is set, %false if made noop because @task was dying.
240 */
241bool task_set_jobctl_pending(struct task_struct *task, unsigned int mask)
242{
243 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
244 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
245 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
246
247 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
248 return false;
249
250 if (mask & JOBCTL_STOP_SIGMASK)
251 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
252
253 task->jobctl |= mask;
254 return true;
255}
256
257/**
258 * task_clear_jobctl_trapping - clear jobctl trapping bit
259 * @task: target task
260 *
261 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
262 * Clear it and wake up the ptracer. Note that we don't need any further
263 * locking. @task->siglock guarantees that @task->parent points to the
264 * ptracer.
265 *
266 * CONTEXT:
267 * Must be called with @task->sighand->siglock held.
268 */
269void task_clear_jobctl_trapping(struct task_struct *task)
270{
271 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
272 task->jobctl &= ~JOBCTL_TRAPPING;
273 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
274 }
275}
276
277/**
278 * task_clear_jobctl_pending - clear jobctl pending bits
279 * @task: target task
280 * @mask: pending bits to clear
281 *
282 * Clear @mask from @task->jobctl. @mask must be subset of
283 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
284 * STOP bits are cleared together.
285 *
286 * If clearing of @mask leaves no stop or trap pending, this function calls
287 * task_clear_jobctl_trapping().
288 *
289 * CONTEXT:
290 * Must be called with @task->sighand->siglock held.
291 */
292void task_clear_jobctl_pending(struct task_struct *task, unsigned int mask)
293{
294 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
295
296 if (mask & JOBCTL_STOP_PENDING)
297 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
298
299 task->jobctl &= ~mask;
300
301 if (!(task->jobctl & JOBCTL_PENDING_MASK))
302 task_clear_jobctl_trapping(task);
303}
304
305/**
306 * task_participate_group_stop - participate in a group stop
307 * @task: task participating in a group stop
308 *
309 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
310 * Group stop states are cleared and the group stop count is consumed if
311 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
312 * stop, the appropriate %SIGNAL_* flags are set.
313 *
314 * CONTEXT:
315 * Must be called with @task->sighand->siglock held.
316 *
317 * RETURNS:
318 * %true if group stop completion should be notified to the parent, %false
319 * otherwise.
320 */
321static bool task_participate_group_stop(struct task_struct *task)
322{
323 struct signal_struct *sig = task->signal;
324 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
325
326 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
327
328 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
329
330 if (!consume)
331 return false;
332
333 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
334 sig->group_stop_count--;
335
336 /*
337 * Tell the caller to notify completion iff we are entering into a
338 * fresh group stop. Read comment in do_signal_stop() for details.
339 */
340 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
341 sig->flags = SIGNAL_STOP_STOPPED;
342 return true;
343 }
344 return false;
345}
346
347/*
348 * allocate a new signal queue record
349 * - this may be called without locks if and only if t == current, otherwise an
350 * appropriate lock must be held to stop the target task from exiting
351 */
352static struct sigqueue *
353__sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimit)
354{
355 struct sigqueue *q = NULL;
356 struct user_struct *user;
357
358 /*
359 * Protect access to @t credentials. This can go away when all
360 * callers hold rcu read lock.
361 */
362 rcu_read_lock();
363 user = get_uid(__task_cred(t)->user);
364 atomic_inc(&user->sigpending);
365 rcu_read_unlock();
366
367 if (override_rlimit ||
368 atomic_read(&user->sigpending) <=
369 task_rlimit(t, RLIMIT_SIGPENDING)) {
370 q = kmem_cache_alloc(sigqueue_cachep, flags);
371 } else {
372 print_dropped_signal(sig);
373 }
374
375 if (unlikely(q == NULL)) {
376 atomic_dec(&user->sigpending);
377 free_uid(user);
378 } else {
379 INIT_LIST_HEAD(&q->list);
380 q->flags = 0;
381 q->user = user;
382 }
383
384 return q;
385}
386
387static void __sigqueue_free(struct sigqueue *q)
388{
389 if (q->flags & SIGQUEUE_PREALLOC)
390 return;
391 atomic_dec(&q->user->sigpending);
392 free_uid(q->user);
393 kmem_cache_free(sigqueue_cachep, q);
394}
395
396void flush_sigqueue(struct sigpending *queue)
397{
398 struct sigqueue *q;
399
400 sigemptyset(&queue->signal);
401 while (!list_empty(&queue->list)) {
402 q = list_entry(queue->list.next, struct sigqueue , list);
403 list_del_init(&q->list);
404 __sigqueue_free(q);
405 }
406}
407
408/*
409 * Flush all pending signals for a task.
410 */
411void __flush_signals(struct task_struct *t)
412{
413 clear_tsk_thread_flag(t, TIF_SIGPENDING);
414 flush_sigqueue(&t->pending);
415 flush_sigqueue(&t->signal->shared_pending);
416}
417
418void flush_signals(struct task_struct *t)
419{
420 unsigned long flags;
421
422 spin_lock_irqsave(&t->sighand->siglock, flags);
423 __flush_signals(t);
424 spin_unlock_irqrestore(&t->sighand->siglock, flags);
425}
426
427static void __flush_itimer_signals(struct sigpending *pending)
428{
429 sigset_t signal, retain;
430 struct sigqueue *q, *n;
431
432 signal = pending->signal;
433 sigemptyset(&retain);
434
435 list_for_each_entry_safe(q, n, &pending->list, list) {
436 int sig = q->info.si_signo;
437
438 if (likely(q->info.si_code != SI_TIMER)) {
439 sigaddset(&retain, sig);
440 } else {
441 sigdelset(&signal, sig);
442 list_del_init(&q->list);
443 __sigqueue_free(q);
444 }
445 }
446
447 sigorsets(&pending->signal, &signal, &retain);
448}
449
450void flush_itimer_signals(void)
451{
452 struct task_struct *tsk = current;
453 unsigned long flags;
454
455 spin_lock_irqsave(&tsk->sighand->siglock, flags);
456 __flush_itimer_signals(&tsk->pending);
457 __flush_itimer_signals(&tsk->signal->shared_pending);
458 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
459}
460
461void ignore_signals(struct task_struct *t)
462{
463 int i;
464
465 for (i = 0; i < _NSIG; ++i)
466 t->sighand->action[i].sa.sa_handler = SIG_IGN;
467
468 flush_signals(t);
469}
470
471/*
472 * Flush all handlers for a task.
473 */
474
475void
476flush_signal_handlers(struct task_struct *t, int force_default)
477{
478 int i;
479 struct k_sigaction *ka = &t->sighand->action[0];
480 for (i = _NSIG ; i != 0 ; i--) {
481 if (force_default || ka->sa.sa_handler != SIG_IGN)
482 ka->sa.sa_handler = SIG_DFL;
483 ka->sa.sa_flags = 0;
484 sigemptyset(&ka->sa.sa_mask);
485 ka++;
486 }
487}
488
489int unhandled_signal(struct task_struct *tsk, int sig)
490{
491 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
492 if (is_global_init(tsk))
493 return 1;
494 if (handler != SIG_IGN && handler != SIG_DFL)
495 return 0;
496 /* if ptraced, let the tracer determine */
497 return !tsk->ptrace;
498}
499
500/*
501 * Notify the system that a driver wants to block all signals for this
502 * process, and wants to be notified if any signals at all were to be
503 * sent/acted upon. If the notifier routine returns non-zero, then the
504 * signal will be acted upon after all. If the notifier routine returns 0,
505 * then then signal will be blocked. Only one block per process is
506 * allowed. priv is a pointer to private data that the notifier routine
507 * can use to determine if the signal should be blocked or not.
508 */
509void
510block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
511{
512 unsigned long flags;
513
514 spin_lock_irqsave(¤t->sighand->siglock, flags);
515 current->notifier_mask = mask;
516 current->notifier_data = priv;
517 current->notifier = notifier;
518 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
519}
520
521/* Notify the system that blocking has ended. */
522
523void
524unblock_all_signals(void)
525{
526 unsigned long flags;
527
528 spin_lock_irqsave(¤t->sighand->siglock, flags);
529 current->notifier = NULL;
530 current->notifier_data = NULL;
531 recalc_sigpending();
532 spin_unlock_irqrestore(¤t->sighand->siglock, flags);
533}
534
535static void collect_signal(int sig, struct sigpending *list, siginfo_t *info)
536{
537 struct sigqueue *q, *first = NULL;
538
539 /*
540 * Collect the siginfo appropriate to this signal. Check if
541 * there is another siginfo for the same signal.
542 */
543 list_for_each_entry(q, &list->list, list) {
544 if (q->info.si_signo == sig) {
545 if (first)
546 goto still_pending;
547 first = q;
548 }
549 }
550
551 sigdelset(&list->signal, sig);
552
553 if (first) {
554still_pending:
555 list_del_init(&first->list);
556 copy_siginfo(info, &first->info);
557 __sigqueue_free(first);
558 } else {
559 /*
560 * Ok, it wasn't in the queue. This must be
561 * a fast-pathed signal or we must have been
562 * out of queue space. So zero out the info.
563 */
564 info->si_signo = sig;
565 info->si_errno = 0;
566 info->si_code = SI_USER;
567 info->si_pid = 0;
568 info->si_uid = 0;
569 }
570}
571
572static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
573 siginfo_t *info)
574{
575 int sig = next_signal(pending, mask);
576
577 if (sig) {
578 if (current->notifier) {
579 if (sigismember(current->notifier_mask, sig)) {
580 if (!(current->notifier)(current->notifier_data)) {
581 clear_thread_flag(TIF_SIGPENDING);
582 return 0;
583 }
584 }
585 }
586
587 collect_signal(sig, pending, info);
588 }
589
590 return sig;
591}
592
593/*
594 * Dequeue a signal and return the element to the caller, which is
595 * expected to free it.
596 *
597 * All callers have to hold the siglock.
598 */
599int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
600{
601 int signr;
602
603 /* We only dequeue private signals from ourselves, we don't let
604 * signalfd steal them
605 */
606 signr = __dequeue_signal(&tsk->pending, mask, info);
607 if (!signr) {
608 signr = __dequeue_signal(&tsk->signal->shared_pending,
609 mask, info);
610 /*
611 * itimer signal ?
612 *
613 * itimers are process shared and we restart periodic
614 * itimers in the signal delivery path to prevent DoS
615 * attacks in the high resolution timer case. This is
616 * compliant with the old way of self-restarting
617 * itimers, as the SIGALRM is a legacy signal and only
618 * queued once. Changing the restart behaviour to
619 * restart the timer in the signal dequeue path is
620 * reducing the timer noise on heavy loaded !highres
621 * systems too.
622 */
623 if (unlikely(signr == SIGALRM)) {
624 struct hrtimer *tmr = &tsk->signal->real_timer;
625
626 if (!hrtimer_is_queued(tmr) &&
627 tsk->signal->it_real_incr.tv64 != 0) {
628 hrtimer_forward(tmr, tmr->base->get_time(),
629 tsk->signal->it_real_incr);
630 hrtimer_restart(tmr);
631 }
632 }
633 }
634
635 recalc_sigpending();
636 if (!signr)
637 return 0;
638
639 if (unlikely(sig_kernel_stop(signr))) {
640 /*
641 * Set a marker that we have dequeued a stop signal. Our
642 * caller might release the siglock and then the pending
643 * stop signal it is about to process is no longer in the
644 * pending bitmasks, but must still be cleared by a SIGCONT
645 * (and overruled by a SIGKILL). So those cases clear this
646 * shared flag after we've set it. Note that this flag may
647 * remain set after the signal we return is ignored or
648 * handled. That doesn't matter because its only purpose
649 * is to alert stop-signal processing code when another
650 * processor has come along and cleared the flag.
651 */
652 current->jobctl |= JOBCTL_STOP_DEQUEUED;
653 }
654 if ((info->si_code & __SI_MASK) == __SI_TIMER && info->si_sys_private) {
655 /*
656 * Release the siglock to ensure proper locking order
657 * of timer locks outside of siglocks. Note, we leave
658 * irqs disabled here, since the posix-timers code is
659 * about to disable them again anyway.
660 */
661 spin_unlock(&tsk->sighand->siglock);
662 do_schedule_next_timer(info);
663 spin_lock(&tsk->sighand->siglock);
664 }
665 return signr;
666}
667
668/*
669 * Tell a process that it has a new active signal..
670 *
671 * NOTE! we rely on the previous spin_lock to
672 * lock interrupts for us! We can only be called with
673 * "siglock" held, and the local interrupt must
674 * have been disabled when that got acquired!
675 *
676 * No need to set need_resched since signal event passing
677 * goes through ->blocked
678 */
679void signal_wake_up(struct task_struct *t, int resume)
680{
681 unsigned int mask;
682
683 set_tsk_thread_flag(t, TIF_SIGPENDING);
684
685 /*
686 * For SIGKILL, we want to wake it up in the stopped/traced/killable
687 * case. We don't check t->state here because there is a race with it
688 * executing another processor and just now entering stopped state.
689 * By using wake_up_state, we ensure the process will wake up and
690 * handle its death signal.
691 */
692 mask = TASK_INTERRUPTIBLE;
693 if (resume)
694 mask |= TASK_WAKEKILL;
695 if (!wake_up_state(t, mask))
696 kick_process(t);
697}
698
699/*
700 * Remove signals in mask from the pending set and queue.
701 * Returns 1 if any signals were found.
702 *
703 * All callers must be holding the siglock.
704 *
705 * This version takes a sigset mask and looks at all signals,
706 * not just those in the first mask word.
707 */
708static int rm_from_queue_full(sigset_t *mask, struct sigpending *s)
709{
710 struct sigqueue *q, *n;
711 sigset_t m;
712
713 sigandsets(&m, mask, &s->signal);
714 if (sigisemptyset(&m))
715 return 0;
716
717 sigandnsets(&s->signal, &s->signal, mask);
718 list_for_each_entry_safe(q, n, &s->list, list) {
719 if (sigismember(mask, q->info.si_signo)) {
720 list_del_init(&q->list);
721 __sigqueue_free(q);
722 }
723 }
724 return 1;
725}
726/*
727 * Remove signals in mask from the pending set and queue.
728 * Returns 1 if any signals were found.
729 *
730 * All callers must be holding the siglock.
731 */
732static int rm_from_queue(unsigned long mask, struct sigpending *s)
733{
734 struct sigqueue *q, *n;
735
736 if (!sigtestsetmask(&s->signal, mask))
737 return 0;
738
739 sigdelsetmask(&s->signal, mask);
740 list_for_each_entry_safe(q, n, &s->list, list) {
741 if (q->info.si_signo < SIGRTMIN &&
742 (mask & sigmask(q->info.si_signo))) {
743 list_del_init(&q->list);
744 __sigqueue_free(q);
745 }
746 }
747 return 1;
748}
749
750static inline int is_si_special(const struct siginfo *info)
751{
752 return info <= SEND_SIG_FORCED;
753}
754
755static inline bool si_fromuser(const struct siginfo *info)
756{
757 return info == SEND_SIG_NOINFO ||
758 (!is_si_special(info) && SI_FROMUSER(info));
759}
760
761/*
762 * called with RCU read lock from check_kill_permission()
763 */
764static int kill_ok_by_cred(struct task_struct *t)
765{
766 const struct cred *cred = current_cred();
767 const struct cred *tcred = __task_cred(t);
768
769 if (cred->user->user_ns == tcred->user->user_ns &&
770 (cred->euid == tcred->suid ||
771 cred->euid == tcred->uid ||
772 cred->uid == tcred->suid ||
773 cred->uid == tcred->uid))
774 return 1;
775
776 if (ns_capable(tcred->user->user_ns, CAP_KILL))
777 return 1;
778
779 return 0;
780}
781
782/*
783 * Bad permissions for sending the signal
784 * - the caller must hold the RCU read lock
785 */
786static int check_kill_permission(int sig, struct siginfo *info,
787 struct task_struct *t)
788{
789 struct pid *sid;
790 int error;
791
792 if (!valid_signal(sig))
793 return -EINVAL;
794
795 if (!si_fromuser(info))
796 return 0;
797
798 error = audit_signal_info(sig, t); /* Let audit system see the signal */
799 if (error)
800 return error;
801
802 if (!same_thread_group(current, t) &&
803 !kill_ok_by_cred(t)) {
804 switch (sig) {
805 case SIGCONT:
806 sid = task_session(t);
807 /*
808 * We don't return the error if sid == NULL. The
809 * task was unhashed, the caller must notice this.
810 */
811 if (!sid || sid == task_session(current))
812 break;
813 default:
814 return -EPERM;
815 }
816 }
817
818 return security_task_kill(t, info, sig, 0);
819}
820
821/**
822 * ptrace_trap_notify - schedule trap to notify ptracer
823 * @t: tracee wanting to notify tracer
824 *
825 * This function schedules sticky ptrace trap which is cleared on the next
826 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
827 * ptracer.
828 *
829 * If @t is running, STOP trap will be taken. If trapped for STOP and
830 * ptracer is listening for events, tracee is woken up so that it can
831 * re-trap for the new event. If trapped otherwise, STOP trap will be
832 * eventually taken without returning to userland after the existing traps
833 * are finished by PTRACE_CONT.
834 *
835 * CONTEXT:
836 * Must be called with @task->sighand->siglock held.
837 */
838static void ptrace_trap_notify(struct task_struct *t)
839{
840 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
841 assert_spin_locked(&t->sighand->siglock);
842
843 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
844 signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
845}
846
847/*
848 * Handle magic process-wide effects of stop/continue signals. Unlike
849 * the signal actions, these happen immediately at signal-generation
850 * time regardless of blocking, ignoring, or handling. This does the
851 * actual continuing for SIGCONT, but not the actual stopping for stop
852 * signals. The process stop is done as a signal action for SIG_DFL.
853 *
854 * Returns true if the signal should be actually delivered, otherwise
855 * it should be dropped.
856 */
857static int prepare_signal(int sig, struct task_struct *p, int from_ancestor_ns)
858{
859 struct signal_struct *signal = p->signal;
860 struct task_struct *t;
861
862 if (unlikely(signal->flags & SIGNAL_GROUP_EXIT)) {
863 /*
864 * The process is in the middle of dying, nothing to do.
865 */
866 } else if (sig_kernel_stop(sig)) {
867 /*
868 * This is a stop signal. Remove SIGCONT from all queues.
869 */
870 rm_from_queue(sigmask(SIGCONT), &signal->shared_pending);
871 t = p;
872 do {
873 rm_from_queue(sigmask(SIGCONT), &t->pending);
874 } while_each_thread(p, t);
875 } else if (sig == SIGCONT) {
876 unsigned int why;
877 /*
878 * Remove all stop signals from all queues, wake all threads.
879 */
880 rm_from_queue(SIG_KERNEL_STOP_MASK, &signal->shared_pending);
881 t = p;
882 do {
883 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
884 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
885 if (likely(!(t->ptrace & PT_SEIZED)))
886 wake_up_state(t, __TASK_STOPPED);
887 else
888 ptrace_trap_notify(t);
889 } while_each_thread(p, t);
890
891 /*
892 * Notify the parent with CLD_CONTINUED if we were stopped.
893 *
894 * If we were in the middle of a group stop, we pretend it
895 * was already finished, and then continued. Since SIGCHLD
896 * doesn't queue we report only CLD_STOPPED, as if the next
897 * CLD_CONTINUED was dropped.
898 */
899 why = 0;
900 if (signal->flags & SIGNAL_STOP_STOPPED)
901 why |= SIGNAL_CLD_CONTINUED;
902 else if (signal->group_stop_count)
903 why |= SIGNAL_CLD_STOPPED;
904
905 if (why) {
906 /*
907 * The first thread which returns from do_signal_stop()
908 * will take ->siglock, notice SIGNAL_CLD_MASK, and
909 * notify its parent. See get_signal_to_deliver().
910 */
911 signal->flags = why | SIGNAL_STOP_CONTINUED;
912 signal->group_stop_count = 0;
913 signal->group_exit_code = 0;
914 }
915 }
916
917 return !sig_ignored(p, sig, from_ancestor_ns);
918}
919
920/*
921 * Test if P wants to take SIG. After we've checked all threads with this,
922 * it's equivalent to finding no threads not blocking SIG. Any threads not
923 * blocking SIG were ruled out because they are not running and already
924 * have pending signals. Such threads will dequeue from the shared queue
925 * as soon as they're available, so putting the signal on the shared queue
926 * will be equivalent to sending it to one such thread.
927 */
928static inline int wants_signal(int sig, struct task_struct *p)
929{
930 if (sigismember(&p->blocked, sig))
931 return 0;
932 if (p->flags & PF_EXITING)
933 return 0;
934 if (sig == SIGKILL)
935 return 1;
936 if (task_is_stopped_or_traced(p))
937 return 0;
938 return task_curr(p) || !signal_pending(p);
939}
940
941static void complete_signal(int sig, struct task_struct *p, int group)
942{
943 struct signal_struct *signal = p->signal;
944 struct task_struct *t;
945
946 /*
947 * Now find a thread we can wake up to take the signal off the queue.
948 *
949 * If the main thread wants the signal, it gets first crack.
950 * Probably the least surprising to the average bear.
951 */
952 if (wants_signal(sig, p))
953 t = p;
954 else if (!group || thread_group_empty(p))
955 /*
956 * There is just one thread and it does not need to be woken.
957 * It will dequeue unblocked signals before it runs again.
958 */
959 return;
960 else {
961 /*
962 * Otherwise try to find a suitable thread.
963 */
964 t = signal->curr_target;
965 while (!wants_signal(sig, t)) {
966 t = next_thread(t);
967 if (t == signal->curr_target)
968 /*
969 * No thread needs to be woken.
970 * Any eligible threads will see
971 * the signal in the queue soon.
972 */
973 return;
974 }
975 signal->curr_target = t;
976 }
977
978 /*
979 * Found a killable thread. If the signal will be fatal,
980 * then start taking the whole group down immediately.
981 */
982 if (sig_fatal(p, sig) &&
983 !(signal->flags & (SIGNAL_UNKILLABLE | SIGNAL_GROUP_EXIT)) &&
984 !sigismember(&t->real_blocked, sig) &&
985 (sig == SIGKILL || !t->ptrace)) {
986 /*
987 * This signal will be fatal to the whole group.
988 */
989 if (!sig_kernel_coredump(sig)) {
990 /*
991 * Start a group exit and wake everybody up.
992 * This way we don't have other threads
993 * running and doing things after a slower
994 * thread has the fatal signal pending.
995 */
996 signal->flags = SIGNAL_GROUP_EXIT;
997 signal->group_exit_code = sig;
998 signal->group_stop_count = 0;
999 t = p;
1000 do {
1001 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1002 sigaddset(&t->pending.signal, SIGKILL);
1003 signal_wake_up(t, 1);
1004 } while_each_thread(p, t);
1005 return;
1006 }
1007 }
1008
1009 /*
1010 * The signal is already in the shared-pending queue.
1011 * Tell the chosen thread to wake up and dequeue it.
1012 */
1013 signal_wake_up(t, sig == SIGKILL);
1014 return;
1015}
1016
1017static inline int legacy_queue(struct sigpending *signals, int sig)
1018{
1019 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1020}
1021
1022static int __send_signal(int sig, struct siginfo *info, struct task_struct *t,
1023 int group, int from_ancestor_ns)
1024{
1025 struct sigpending *pending;
1026 struct sigqueue *q;
1027 int override_rlimit;
1028
1029 trace_signal_generate(sig, info, t);
1030
1031 assert_spin_locked(&t->sighand->siglock);
1032
1033 if (!prepare_signal(sig, t, from_ancestor_ns))
1034 return 0;
1035
1036 pending = group ? &t->signal->shared_pending : &t->pending;
1037 /*
1038 * Short-circuit ignored signals and support queuing
1039 * exactly one non-rt signal, so that we can get more
1040 * detailed information about the cause of the signal.
1041 */
1042 if (legacy_queue(pending, sig))
1043 return 0;
1044 /*
1045 * fast-pathed signals for kernel-internal things like SIGSTOP
1046 * or SIGKILL.
1047 */
1048 if (info == SEND_SIG_FORCED)
1049 goto out_set;
1050
1051 /*
1052 * Real-time signals must be queued if sent by sigqueue, or
1053 * some other real-time mechanism. It is implementation
1054 * defined whether kill() does so. We attempt to do so, on
1055 * the principle of least surprise, but since kill is not
1056 * allowed to fail with EAGAIN when low on memory we just
1057 * make sure at least one signal gets delivered and don't
1058 * pass on the info struct.
1059 */
1060 if (sig < SIGRTMIN)
1061 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1062 else
1063 override_rlimit = 0;
1064
1065 q = __sigqueue_alloc(sig, t, GFP_ATOMIC | __GFP_NOTRACK_FALSE_POSITIVE,
1066 override_rlimit);
1067 if (q) {
1068 list_add_tail(&q->list, &pending->list);
1069 switch ((unsigned long) info) {
1070 case (unsigned long) SEND_SIG_NOINFO:
1071 q->info.si_signo = sig;
1072 q->info.si_errno = 0;
1073 q->info.si_code = SI_USER;
1074 q->info.si_pid = task_tgid_nr_ns(current,
1075 task_active_pid_ns(t));
1076 q->info.si_uid = current_uid();
1077 break;
1078 case (unsigned long) SEND_SIG_PRIV:
1079 q->info.si_signo = sig;
1080 q->info.si_errno = 0;
1081 q->info.si_code = SI_KERNEL;
1082 q->info.si_pid = 0;
1083 q->info.si_uid = 0;
1084 break;
1085 default:
1086 copy_siginfo(&q->info, info);
1087 if (from_ancestor_ns)
1088 q->info.si_pid = 0;
1089 break;
1090 }
1091 } else if (!is_si_special(info)) {
1092 if (sig >= SIGRTMIN && info->si_code != SI_USER) {
1093 /*
1094 * Queue overflow, abort. We may abort if the
1095 * signal was rt and sent by user using something
1096 * other than kill().
1097 */
1098 trace_signal_overflow_fail(sig, group, info);
1099 return -EAGAIN;
1100 } else {
1101 /*
1102 * This is a silent loss of information. We still
1103 * send the signal, but the *info bits are lost.
1104 */
1105 trace_signal_lose_info(sig, group, info);
1106 }
1107 }
1108
1109out_set:
1110 signalfd_notify(t, sig);
1111 sigaddset(&pending->signal, sig);
1112 complete_signal(sig, t, group);
1113 return 0;
1114}
1115
1116static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
1117 int group)
1118{
1119 int from_ancestor_ns = 0;
1120
1121#ifdef CONFIG_PID_NS
1122 from_ancestor_ns = si_fromuser(info) &&
1123 !task_pid_nr_ns(current, task_active_pid_ns(t));
1124#endif
1125
1126 return __send_signal(sig, info, t, group, from_ancestor_ns);
1127}
1128
1129static void print_fatal_signal(struct pt_regs *regs, int signr)
1130{
1131 printk("%s/%d: potentially unexpected fatal signal %d.\n",
1132 current->comm, task_pid_nr(current), signr);
1133
1134#if defined(__i386__) && !defined(__arch_um__)
1135 printk("code at %08lx: ", regs->ip);
1136 {
1137 int i;
1138 for (i = 0; i < 16; i++) {
1139 unsigned char insn;
1140
1141 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1142 break;
1143 printk("%02x ", insn);
1144 }
1145 }
1146#endif
1147 printk("\n");
1148 preempt_disable();
1149 show_regs(regs);
1150 preempt_enable();
1151}
1152
1153static int __init setup_print_fatal_signals(char *str)
1154{
1155 get_option (&str, &print_fatal_signals);
1156
1157 return 1;
1158}
1159
1160__setup("print-fatal-signals=", setup_print_fatal_signals);
1161
1162int
1163__group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1164{
1165 return send_signal(sig, info, p, 1);
1166}
1167
1168static int
1169specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
1170{
1171 return send_signal(sig, info, t, 0);
1172}
1173
1174int do_send_sig_info(int sig, struct siginfo *info, struct task_struct *p,
1175 bool group)
1176{
1177 unsigned long flags;
1178 int ret = -ESRCH;
1179
1180 if (lock_task_sighand(p, &flags)) {
1181 ret = send_signal(sig, info, p, group);
1182 unlock_task_sighand(p, &flags);
1183 }
1184
1185 return ret;
1186}
1187
1188/*
1189 * Force a signal that the process can't ignore: if necessary
1190 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1191 *
1192 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1193 * since we do not want to have a signal handler that was blocked
1194 * be invoked when user space had explicitly blocked it.
1195 *
1196 * We don't want to have recursive SIGSEGV's etc, for example,
1197 * that is why we also clear SIGNAL_UNKILLABLE.
1198 */
1199int
1200force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
1201{
1202 unsigned long int flags;
1203 int ret, blocked, ignored;
1204 struct k_sigaction *action;
1205
1206 spin_lock_irqsave(&t->sighand->siglock, flags);
1207 action = &t->sighand->action[sig-1];
1208 ignored = action->sa.sa_handler == SIG_IGN;
1209 blocked = sigismember(&t->blocked, sig);
1210 if (blocked || ignored) {
1211 action->sa.sa_handler = SIG_DFL;
1212 if (blocked) {
1213 sigdelset(&t->blocked, sig);
1214 recalc_sigpending_and_wake(t);
1215 }
1216 }
1217 if (action->sa.sa_handler == SIG_DFL)
1218 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1219 ret = specific_send_sig_info(sig, info, t);
1220 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1221
1222 return ret;
1223}
1224
1225/*
1226 * Nuke all other threads in the group.
1227 */
1228int zap_other_threads(struct task_struct *p)
1229{
1230 struct task_struct *t = p;
1231 int count = 0;
1232
1233 p->signal->group_stop_count = 0;
1234
1235 while_each_thread(p, t) {
1236 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1237 count++;
1238
1239 /* Don't bother with already dead threads */
1240 if (t->exit_state)
1241 continue;
1242 sigaddset(&t->pending.signal, SIGKILL);
1243 signal_wake_up(t, 1);
1244 }
1245
1246 return count;
1247}
1248
1249struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1250 unsigned long *flags)
1251{
1252 struct sighand_struct *sighand;
1253
1254 for (;;) {
1255 local_irq_save(*flags);
1256 rcu_read_lock();
1257 sighand = rcu_dereference(tsk->sighand);
1258 if (unlikely(sighand == NULL)) {
1259 rcu_read_unlock();
1260 local_irq_restore(*flags);
1261 break;
1262 }
1263
1264 spin_lock(&sighand->siglock);
1265 if (likely(sighand == tsk->sighand)) {
1266 rcu_read_unlock();
1267 break;
1268 }
1269 spin_unlock(&sighand->siglock);
1270 rcu_read_unlock();
1271 local_irq_restore(*flags);
1272 }
1273
1274 return sighand;
1275}
1276
1277/*
1278 * send signal info to all the members of a group
1279 */
1280int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1281{
1282 int ret;
1283
1284 rcu_read_lock();
1285 ret = check_kill_permission(sig, info, p);
1286 rcu_read_unlock();
1287
1288 if (!ret && sig)
1289 ret = do_send_sig_info(sig, info, p, true);
1290
1291 return ret;
1292}
1293
1294/*
1295 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1296 * control characters do (^C, ^Z etc)
1297 * - the caller must hold at least a readlock on tasklist_lock
1298 */
1299int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1300{
1301 struct task_struct *p = NULL;
1302 int retval, success;
1303
1304 success = 0;
1305 retval = -ESRCH;
1306 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1307 int err = group_send_sig_info(sig, info, p);
1308 success |= !err;
1309 retval = err;
1310 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1311 return success ? 0 : retval;
1312}
1313
1314int kill_pid_info(int sig, struct siginfo *info, struct pid *pid)
1315{
1316 int error = -ESRCH;
1317 struct task_struct *p;
1318
1319 rcu_read_lock();
1320retry:
1321 p = pid_task(pid, PIDTYPE_PID);
1322 if (p) {
1323 error = group_send_sig_info(sig, info, p);
1324 if (unlikely(error == -ESRCH))
1325 /*
1326 * The task was unhashed in between, try again.
1327 * If it is dead, pid_task() will return NULL,
1328 * if we race with de_thread() it will find the
1329 * new leader.
1330 */
1331 goto retry;
1332 }
1333 rcu_read_unlock();
1334
1335 return error;
1336}
1337
1338int kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1339{
1340 int error;
1341 rcu_read_lock();
1342 error = kill_pid_info(sig, info, find_vpid(pid));
1343 rcu_read_unlock();
1344 return error;
1345}
1346
1347/* like kill_pid_info(), but doesn't use uid/euid of "current" */
1348int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
1349 uid_t uid, uid_t euid, u32 secid)
1350{
1351 int ret = -EINVAL;
1352 struct task_struct *p;
1353 const struct cred *pcred;
1354 unsigned long flags;
1355
1356 if (!valid_signal(sig))
1357 return ret;
1358
1359 rcu_read_lock();
1360 p = pid_task(pid, PIDTYPE_PID);
1361 if (!p) {
1362 ret = -ESRCH;
1363 goto out_unlock;
1364 }
1365 pcred = __task_cred(p);
1366 if (si_fromuser(info) &&
1367 euid != pcred->suid && euid != pcred->uid &&
1368 uid != pcred->suid && uid != pcred->uid) {
1369 ret = -EPERM;
1370 goto out_unlock;
1371 }
1372 ret = security_task_kill(p, info, sig, secid);
1373 if (ret)
1374 goto out_unlock;
1375
1376 if (sig) {
1377 if (lock_task_sighand(p, &flags)) {
1378 ret = __send_signal(sig, info, p, 1, 0);
1379 unlock_task_sighand(p, &flags);
1380 } else
1381 ret = -ESRCH;
1382 }
1383out_unlock:
1384 rcu_read_unlock();
1385 return ret;
1386}
1387EXPORT_SYMBOL_GPL(kill_pid_info_as_uid);
1388
1389/*
1390 * kill_something_info() interprets pid in interesting ways just like kill(2).
1391 *
1392 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1393 * is probably wrong. Should make it like BSD or SYSV.
1394 */
1395
1396static int kill_something_info(int sig, struct siginfo *info, pid_t pid)
1397{
1398 int ret;
1399
1400 if (pid > 0) {
1401 rcu_read_lock();
1402 ret = kill_pid_info(sig, info, find_vpid(pid));
1403 rcu_read_unlock();
1404 return ret;
1405 }
1406
1407 read_lock(&tasklist_lock);
1408 if (pid != -1) {
1409 ret = __kill_pgrp_info(sig, info,
1410 pid ? find_vpid(-pid) : task_pgrp(current));
1411 } else {
1412 int retval = 0, count = 0;
1413 struct task_struct * p;
1414
1415 for_each_process(p) {
1416 if (task_pid_vnr(p) > 1 &&
1417 !same_thread_group(p, current)) {
1418 int err = group_send_sig_info(sig, info, p);
1419 ++count;
1420 if (err != -EPERM)
1421 retval = err;
1422 }
1423 }
1424 ret = count ? retval : -ESRCH;
1425 }
1426 read_unlock(&tasklist_lock);
1427
1428 return ret;
1429}
1430
1431/*
1432 * These are for backward compatibility with the rest of the kernel source.
1433 */
1434
1435int send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1436{
1437 /*
1438 * Make sure legacy kernel users don't send in bad values
1439 * (normal paths check this in check_kill_permission).
1440 */
1441 if (!valid_signal(sig))
1442 return -EINVAL;
1443
1444 return do_send_sig_info(sig, info, p, false);
1445}
1446
1447#define __si_special(priv) \
1448 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1449
1450int
1451send_sig(int sig, struct task_struct *p, int priv)
1452{
1453 return send_sig_info(sig, __si_special(priv), p);
1454}
1455
1456void
1457force_sig(int sig, struct task_struct *p)
1458{
1459 force_sig_info(sig, SEND_SIG_PRIV, p);
1460}
1461
1462/*
1463 * When things go south during signal handling, we
1464 * will force a SIGSEGV. And if the signal that caused
1465 * the problem was already a SIGSEGV, we'll want to
1466 * make sure we don't even try to deliver the signal..
1467 */
1468int
1469force_sigsegv(int sig, struct task_struct *p)
1470{
1471 if (sig == SIGSEGV) {
1472 unsigned long flags;
1473 spin_lock_irqsave(&p->sighand->siglock, flags);
1474 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1475 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1476 }
1477 force_sig(SIGSEGV, p);
1478 return 0;
1479}
1480
1481int kill_pgrp(struct pid *pid, int sig, int priv)
1482{
1483 int ret;
1484
1485 read_lock(&tasklist_lock);
1486 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1487 read_unlock(&tasklist_lock);
1488
1489 return ret;
1490}
1491EXPORT_SYMBOL(kill_pgrp);
1492
1493int kill_pid(struct pid *pid, int sig, int priv)
1494{
1495 return kill_pid_info(sig, __si_special(priv), pid);
1496}
1497EXPORT_SYMBOL(kill_pid);
1498
1499/*
1500 * These functions support sending signals using preallocated sigqueue
1501 * structures. This is needed "because realtime applications cannot
1502 * afford to lose notifications of asynchronous events, like timer
1503 * expirations or I/O completions". In the case of POSIX Timers
1504 * we allocate the sigqueue structure from the timer_create. If this
1505 * allocation fails we are able to report the failure to the application
1506 * with an EAGAIN error.
1507 */
1508struct sigqueue *sigqueue_alloc(void)
1509{
1510 struct sigqueue *q = __sigqueue_alloc(-1, current, GFP_KERNEL, 0);
1511
1512 if (q)
1513 q->flags |= SIGQUEUE_PREALLOC;
1514
1515 return q;
1516}
1517
1518void sigqueue_free(struct sigqueue *q)
1519{
1520 unsigned long flags;
1521 spinlock_t *lock = ¤t->sighand->siglock;
1522
1523 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1524 /*
1525 * We must hold ->siglock while testing q->list
1526 * to serialize with collect_signal() or with
1527 * __exit_signal()->flush_sigqueue().
1528 */
1529 spin_lock_irqsave(lock, flags);
1530 q->flags &= ~SIGQUEUE_PREALLOC;
1531 /*
1532 * If it is queued it will be freed when dequeued,
1533 * like the "regular" sigqueue.
1534 */
1535 if (!list_empty(&q->list))
1536 q = NULL;
1537 spin_unlock_irqrestore(lock, flags);
1538
1539 if (q)
1540 __sigqueue_free(q);
1541}
1542
1543int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
1544{
1545 int sig = q->info.si_signo;
1546 struct sigpending *pending;
1547 unsigned long flags;
1548 int ret;
1549
1550 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1551
1552 ret = -1;
1553 if (!likely(lock_task_sighand(t, &flags)))
1554 goto ret;
1555
1556 ret = 1; /* the signal is ignored */
1557 if (!prepare_signal(sig, t, 0))
1558 goto out;
1559
1560 ret = 0;
1561 if (unlikely(!list_empty(&q->list))) {
1562 /*
1563 * If an SI_TIMER entry is already queue just increment
1564 * the overrun count.
1565 */
1566 BUG_ON(q->info.si_code != SI_TIMER);
1567 q->info.si_overrun++;
1568 goto out;
1569 }
1570 q->info.si_overrun = 0;
1571
1572 signalfd_notify(t, sig);
1573 pending = group ? &t->signal->shared_pending : &t->pending;
1574 list_add_tail(&q->list, &pending->list);
1575 sigaddset(&pending->signal, sig);
1576 complete_signal(sig, t, group);
1577out:
1578 unlock_task_sighand(t, &flags);
1579ret:
1580 return ret;
1581}
1582
1583/*
1584 * Let a parent know about the death of a child.
1585 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1586 *
1587 * Returns true if our parent ignored us and so we've switched to
1588 * self-reaping.
1589 */
1590bool do_notify_parent(struct task_struct *tsk, int sig)
1591{
1592 struct siginfo info;
1593 unsigned long flags;
1594 struct sighand_struct *psig;
1595 bool autoreap = false;
1596
1597 BUG_ON(sig == -1);
1598
1599 /* do_notify_parent_cldstop should have been called instead. */
1600 BUG_ON(task_is_stopped_or_traced(tsk));
1601
1602 BUG_ON(!tsk->ptrace &&
1603 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1604
1605 info.si_signo = sig;
1606 info.si_errno = 0;
1607 /*
1608 * we are under tasklist_lock here so our parent is tied to
1609 * us and cannot exit and release its namespace.
1610 *
1611 * the only it can is to switch its nsproxy with sys_unshare,
1612 * bu uncharing pid namespaces is not allowed, so we'll always
1613 * see relevant namespace
1614 *
1615 * write_lock() currently calls preempt_disable() which is the
1616 * same as rcu_read_lock(), but according to Oleg, this is not
1617 * correct to rely on this
1618 */
1619 rcu_read_lock();
1620 info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1621 info.si_uid = __task_cred(tsk)->uid;
1622 rcu_read_unlock();
1623
1624 info.si_utime = cputime_to_clock_t(cputime_add(tsk->utime,
1625 tsk->signal->utime));
1626 info.si_stime = cputime_to_clock_t(cputime_add(tsk->stime,
1627 tsk->signal->stime));
1628
1629 info.si_status = tsk->exit_code & 0x7f;
1630 if (tsk->exit_code & 0x80)
1631 info.si_code = CLD_DUMPED;
1632 else if (tsk->exit_code & 0x7f)
1633 info.si_code = CLD_KILLED;
1634 else {
1635 info.si_code = CLD_EXITED;
1636 info.si_status = tsk->exit_code >> 8;
1637 }
1638
1639 psig = tsk->parent->sighand;
1640 spin_lock_irqsave(&psig->siglock, flags);
1641 if (!tsk->ptrace && sig == SIGCHLD &&
1642 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1643 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1644 /*
1645 * We are exiting and our parent doesn't care. POSIX.1
1646 * defines special semantics for setting SIGCHLD to SIG_IGN
1647 * or setting the SA_NOCLDWAIT flag: we should be reaped
1648 * automatically and not left for our parent's wait4 call.
1649 * Rather than having the parent do it as a magic kind of
1650 * signal handler, we just set this to tell do_exit that we
1651 * can be cleaned up without becoming a zombie. Note that
1652 * we still call __wake_up_parent in this case, because a
1653 * blocked sys_wait4 might now return -ECHILD.
1654 *
1655 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1656 * is implementation-defined: we do (if you don't want
1657 * it, just use SIG_IGN instead).
1658 */
1659 autoreap = true;
1660 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1661 sig = 0;
1662 }
1663 if (valid_signal(sig) && sig)
1664 __group_send_sig_info(sig, &info, tsk->parent);
1665 __wake_up_parent(tsk, tsk->parent);
1666 spin_unlock_irqrestore(&psig->siglock, flags);
1667
1668 return autoreap;
1669}
1670
1671/**
1672 * do_notify_parent_cldstop - notify parent of stopped/continued state change
1673 * @tsk: task reporting the state change
1674 * @for_ptracer: the notification is for ptracer
1675 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
1676 *
1677 * Notify @tsk's parent that the stopped/continued state has changed. If
1678 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
1679 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
1680 *
1681 * CONTEXT:
1682 * Must be called with tasklist_lock at least read locked.
1683 */
1684static void do_notify_parent_cldstop(struct task_struct *tsk,
1685 bool for_ptracer, int why)
1686{
1687 struct siginfo info;
1688 unsigned long flags;
1689 struct task_struct *parent;
1690 struct sighand_struct *sighand;
1691
1692 if (for_ptracer) {
1693 parent = tsk->parent;
1694 } else {
1695 tsk = tsk->group_leader;
1696 parent = tsk->real_parent;
1697 }
1698
1699 info.si_signo = SIGCHLD;
1700 info.si_errno = 0;
1701 /*
1702 * see comment in do_notify_parent() about the following 4 lines
1703 */
1704 rcu_read_lock();
1705 info.si_pid = task_pid_nr_ns(tsk, parent->nsproxy->pid_ns);
1706 info.si_uid = __task_cred(tsk)->uid;
1707 rcu_read_unlock();
1708
1709 info.si_utime = cputime_to_clock_t(tsk->utime);
1710 info.si_stime = cputime_to_clock_t(tsk->stime);
1711
1712 info.si_code = why;
1713 switch (why) {
1714 case CLD_CONTINUED:
1715 info.si_status = SIGCONT;
1716 break;
1717 case CLD_STOPPED:
1718 info.si_status = tsk->signal->group_exit_code & 0x7f;
1719 break;
1720 case CLD_TRAPPED:
1721 info.si_status = tsk->exit_code & 0x7f;
1722 break;
1723 default:
1724 BUG();
1725 }
1726
1727 sighand = parent->sighand;
1728 spin_lock_irqsave(&sighand->siglock, flags);
1729 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1730 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1731 __group_send_sig_info(SIGCHLD, &info, parent);
1732 /*
1733 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1734 */
1735 __wake_up_parent(tsk, parent);
1736 spin_unlock_irqrestore(&sighand->siglock, flags);
1737}
1738
1739static inline int may_ptrace_stop(void)
1740{
1741 if (!likely(current->ptrace))
1742 return 0;
1743 /*
1744 * Are we in the middle of do_coredump?
1745 * If so and our tracer is also part of the coredump stopping
1746 * is a deadlock situation, and pointless because our tracer
1747 * is dead so don't allow us to stop.
1748 * If SIGKILL was already sent before the caller unlocked
1749 * ->siglock we must see ->core_state != NULL. Otherwise it
1750 * is safe to enter schedule().
1751 */
1752 if (unlikely(current->mm->core_state) &&
1753 unlikely(current->mm == current->parent->mm))
1754 return 0;
1755
1756 return 1;
1757}
1758
1759/*
1760 * Return non-zero if there is a SIGKILL that should be waking us up.
1761 * Called with the siglock held.
1762 */
1763static int sigkill_pending(struct task_struct *tsk)
1764{
1765 return sigismember(&tsk->pending.signal, SIGKILL) ||
1766 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
1767}
1768
1769/*
1770 * This must be called with current->sighand->siglock held.
1771 *
1772 * This should be the path for all ptrace stops.
1773 * We always set current->last_siginfo while stopped here.
1774 * That makes it a way to test a stopped process for
1775 * being ptrace-stopped vs being job-control-stopped.
1776 *
1777 * If we actually decide not to stop at all because the tracer
1778 * is gone, we keep current->exit_code unless clear_code.
1779 */
1780static void ptrace_stop(int exit_code, int why, int clear_code, siginfo_t *info)
1781 __releases(¤t->sighand->siglock)
1782 __acquires(¤t->sighand->siglock)
1783{
1784 bool gstop_done = false;
1785
1786 if (arch_ptrace_stop_needed(exit_code, info)) {
1787 /*
1788 * The arch code has something special to do before a
1789 * ptrace stop. This is allowed to block, e.g. for faults
1790 * on user stack pages. We can't keep the siglock while
1791 * calling arch_ptrace_stop, so we must release it now.
1792 * To preserve proper semantics, we must do this before
1793 * any signal bookkeeping like checking group_stop_count.
1794 * Meanwhile, a SIGKILL could come in before we retake the
1795 * siglock. That must prevent us from sleeping in TASK_TRACED.
1796 * So after regaining the lock, we must check for SIGKILL.
1797 */
1798 spin_unlock_irq(¤t->sighand->siglock);
1799 arch_ptrace_stop(exit_code, info);
1800 spin_lock_irq(¤t->sighand->siglock);
1801 if (sigkill_pending(current))
1802 return;
1803 }
1804
1805 /*
1806 * We're committing to trapping. TRACED should be visible before
1807 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
1808 * Also, transition to TRACED and updates to ->jobctl should be
1809 * atomic with respect to siglock and should be done after the arch
1810 * hook as siglock is released and regrabbed across it.
1811 */
1812 set_current_state(TASK_TRACED);
1813
1814 current->last_siginfo = info;
1815 current->exit_code = exit_code;
1816
1817 /*
1818 * If @why is CLD_STOPPED, we're trapping to participate in a group
1819 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
1820 * across siglock relocks since INTERRUPT was scheduled, PENDING
1821 * could be clear now. We act as if SIGCONT is received after
1822 * TASK_TRACED is entered - ignore it.
1823 */
1824 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
1825 gstop_done = task_participate_group_stop(current);
1826
1827 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
1828 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
1829 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
1830 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
1831
1832 /* entering a trap, clear TRAPPING */
1833 task_clear_jobctl_trapping(current);
1834
1835 spin_unlock_irq(¤t->sighand->siglock);
1836 read_lock(&tasklist_lock);
1837 if (may_ptrace_stop()) {
1838 /*
1839 * Notify parents of the stop.
1840 *
1841 * While ptraced, there are two parents - the ptracer and
1842 * the real_parent of the group_leader. The ptracer should
1843 * know about every stop while the real parent is only
1844 * interested in the completion of group stop. The states
1845 * for the two don't interact with each other. Notify
1846 * separately unless they're gonna be duplicates.
1847 */
1848 do_notify_parent_cldstop(current, true, why);
1849 if (gstop_done && ptrace_reparented(current))
1850 do_notify_parent_cldstop(current, false, why);
1851
1852 /*
1853 * Don't want to allow preemption here, because
1854 * sys_ptrace() needs this task to be inactive.
1855 *
1856 * XXX: implement read_unlock_no_resched().
1857 */
1858 preempt_disable();
1859 read_unlock(&tasklist_lock);
1860 preempt_enable_no_resched();
1861 schedule();
1862 } else {
1863 /*
1864 * By the time we got the lock, our tracer went away.
1865 * Don't drop the lock yet, another tracer may come.
1866 *
1867 * If @gstop_done, the ptracer went away between group stop
1868 * completion and here. During detach, it would have set
1869 * JOBCTL_STOP_PENDING on us and we'll re-enter
1870 * TASK_STOPPED in do_signal_stop() on return, so notifying
1871 * the real parent of the group stop completion is enough.
1872 */
1873 if (gstop_done)
1874 do_notify_parent_cldstop(current, false, why);
1875
1876 __set_current_state(TASK_RUNNING);
1877 if (clear_code)
1878 current->exit_code = 0;
1879 read_unlock(&tasklist_lock);
1880 }
1881
1882 /*
1883 * While in TASK_TRACED, we were considered "frozen enough".
1884 * Now that we woke up, it's crucial if we're supposed to be
1885 * frozen that we freeze now before running anything substantial.
1886 */
1887 try_to_freeze();
1888
1889 /*
1890 * We are back. Now reacquire the siglock before touching
1891 * last_siginfo, so that we are sure to have synchronized with
1892 * any signal-sending on another CPU that wants to examine it.
1893 */
1894 spin_lock_irq(¤t->sighand->siglock);
1895 current->last_siginfo = NULL;
1896
1897 /* LISTENING can be set only during STOP traps, clear it */
1898 current->jobctl &= ~JOBCTL_LISTENING;
1899
1900 /*
1901 * Queued signals ignored us while we were stopped for tracing.
1902 * So check for any that we should take before resuming user mode.
1903 * This sets TIF_SIGPENDING, but never clears it.
1904 */
1905 recalc_sigpending_tsk(current);
1906}
1907
1908static void ptrace_do_notify(int signr, int exit_code, int why)
1909{
1910 siginfo_t info;
1911
1912 memset(&info, 0, sizeof info);
1913 info.si_signo = signr;
1914 info.si_code = exit_code;
1915 info.si_pid = task_pid_vnr(current);
1916 info.si_uid = current_uid();
1917
1918 /* Let the debugger run. */
1919 ptrace_stop(exit_code, why, 1, &info);
1920}
1921
1922void ptrace_notify(int exit_code)
1923{
1924 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1925
1926 spin_lock_irq(¤t->sighand->siglock);
1927 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
1928 spin_unlock_irq(¤t->sighand->siglock);
1929}
1930
1931/**
1932 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
1933 * @signr: signr causing group stop if initiating
1934 *
1935 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
1936 * and participate in it. If already set, participate in the existing
1937 * group stop. If participated in a group stop (and thus slept), %true is
1938 * returned with siglock released.
1939 *
1940 * If ptraced, this function doesn't handle stop itself. Instead,
1941 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
1942 * untouched. The caller must ensure that INTERRUPT trap handling takes
1943 * places afterwards.
1944 *
1945 * CONTEXT:
1946 * Must be called with @current->sighand->siglock held, which is released
1947 * on %true return.
1948 *
1949 * RETURNS:
1950 * %false if group stop is already cancelled or ptrace trap is scheduled.
1951 * %true if participated in group stop.
1952 */
1953static bool do_signal_stop(int signr)
1954 __releases(¤t->sighand->siglock)
1955{
1956 struct signal_struct *sig = current->signal;
1957
1958 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
1959 unsigned int gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
1960 struct task_struct *t;
1961
1962 /* signr will be recorded in task->jobctl for retries */
1963 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
1964
1965 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
1966 unlikely(signal_group_exit(sig)))
1967 return false;
1968 /*
1969 * There is no group stop already in progress. We must
1970 * initiate one now.
1971 *
1972 * While ptraced, a task may be resumed while group stop is
1973 * still in effect and then receive a stop signal and
1974 * initiate another group stop. This deviates from the
1975 * usual behavior as two consecutive stop signals can't
1976 * cause two group stops when !ptraced. That is why we
1977 * also check !task_is_stopped(t) below.
1978 *
1979 * The condition can be distinguished by testing whether
1980 * SIGNAL_STOP_STOPPED is already set. Don't generate
1981 * group_exit_code in such case.
1982 *
1983 * This is not necessary for SIGNAL_STOP_CONTINUED because
1984 * an intervening stop signal is required to cause two
1985 * continued events regardless of ptrace.
1986 */
1987 if (!(sig->flags & SIGNAL_STOP_STOPPED))
1988 sig->group_exit_code = signr;
1989 else
1990 WARN_ON_ONCE(!current->ptrace);
1991
1992 sig->group_stop_count = 0;
1993
1994 if (task_set_jobctl_pending(current, signr | gstop))
1995 sig->group_stop_count++;
1996
1997 for (t = next_thread(current); t != current;
1998 t = next_thread(t)) {
1999 /*
2000 * Setting state to TASK_STOPPED for a group
2001 * stop is always done with the siglock held,
2002 * so this check has no races.
2003 */
2004 if (!task_is_stopped(t) &&
2005 task_set_jobctl_pending(t, signr | gstop)) {
2006 sig->group_stop_count++;
2007 if (likely(!(t->ptrace & PT_SEIZED)))
2008 signal_wake_up(t, 0);
2009 else
2010 ptrace_trap_notify(t);
2011 }
2012 }
2013 }
2014
2015 if (likely(!current->ptrace)) {
2016 int notify = 0;
2017
2018 /*
2019 * If there are no other threads in the group, or if there
2020 * is a group stop in progress and we are the last to stop,
2021 * report to the parent.
2022 */
2023 if (task_participate_group_stop(current))
2024 notify = CLD_STOPPED;
2025
2026 __set_current_state(TASK_STOPPED);
2027 spin_unlock_irq(¤t->sighand->siglock);
2028
2029 /*
2030 * Notify the parent of the group stop completion. Because
2031 * we're not holding either the siglock or tasklist_lock
2032 * here, ptracer may attach inbetween; however, this is for
2033 * group stop and should always be delivered to the real
2034 * parent of the group leader. The new ptracer will get
2035 * its notification when this task transitions into
2036 * TASK_TRACED.
2037 */
2038 if (notify) {
2039 read_lock(&tasklist_lock);
2040 do_notify_parent_cldstop(current, false, notify);
2041 read_unlock(&tasklist_lock);
2042 }
2043
2044 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2045 schedule();
2046 return true;
2047 } else {
2048 /*
2049 * While ptraced, group stop is handled by STOP trap.
2050 * Schedule it and let the caller deal with it.
2051 */
2052 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2053 return false;
2054 }
2055}
2056
2057/**
2058 * do_jobctl_trap - take care of ptrace jobctl traps
2059 *
2060 * When PT_SEIZED, it's used for both group stop and explicit
2061 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2062 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2063 * the stop signal; otherwise, %SIGTRAP.
2064 *
2065 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2066 * number as exit_code and no siginfo.
2067 *
2068 * CONTEXT:
2069 * Must be called with @current->sighand->siglock held, which may be
2070 * released and re-acquired before returning with intervening sleep.
2071 */
2072static void do_jobctl_trap(void)
2073{
2074 struct signal_struct *signal = current->signal;
2075 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2076
2077 if (current->ptrace & PT_SEIZED) {
2078 if (!signal->group_stop_count &&
2079 !(signal->flags & SIGNAL_STOP_STOPPED))
2080 signr = SIGTRAP;
2081 WARN_ON_ONCE(!signr);
2082 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2083 CLD_STOPPED);
2084 } else {
2085 WARN_ON_ONCE(!signr);
2086 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2087 current->exit_code = 0;
2088 }
2089}
2090
2091static int ptrace_signal(int signr, siginfo_t *info,
2092 struct pt_regs *regs, void *cookie)
2093{
2094 ptrace_signal_deliver(regs, cookie);
2095 /*
2096 * We do not check sig_kernel_stop(signr) but set this marker
2097 * unconditionally because we do not know whether debugger will
2098 * change signr. This flag has no meaning unless we are going
2099 * to stop after return from ptrace_stop(). In this case it will
2100 * be checked in do_signal_stop(), we should only stop if it was
2101 * not cleared by SIGCONT while we were sleeping. See also the
2102 * comment in dequeue_signal().
2103 */
2104 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2105 ptrace_stop(signr, CLD_TRAPPED, 0, info);
2106
2107 /* We're back. Did the debugger cancel the sig? */
2108 signr = current->exit_code;
2109 if (signr == 0)
2110 return signr;
2111
2112 current->exit_code = 0;
2113
2114 /*
2115 * Update the siginfo structure if the signal has
2116 * changed. If the debugger wanted something
2117 * specific in the siginfo structure then it should
2118 * have updated *info via PTRACE_SETSIGINFO.
2119 */
2120 if (signr != info->si_signo) {
2121 info->si_signo = signr;
2122 info->si_errno = 0;
2123 info->si_code = SI_USER;
2124 info->si_pid = task_pid_vnr(current->parent);
2125 info->si_uid = task_uid(current->parent);
2126 }
2127
2128 /* If the (new) signal is now blocked, requeue it. */
2129 if (sigismember(¤t->blocked, signr)) {
2130 specific_send_sig_info(signr, info, current);
2131 signr = 0;
2132 }
2133
2134 return signr;
2135}
2136
2137int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
2138 struct pt_regs *regs, void *cookie)
2139{
2140 struct sighand_struct *sighand = current->sighand;
2141 struct signal_struct *signal = current->signal;
2142 int signr;
2143
2144relock:
2145 /*
2146 * We'll jump back here after any time we were stopped in TASK_STOPPED.
2147 * While in TASK_STOPPED, we were considered "frozen enough".
2148 * Now that we woke up, it's crucial if we're supposed to be
2149 * frozen that we freeze now before running anything substantial.
2150 */
2151 try_to_freeze();
2152
2153 spin_lock_irq(&sighand->siglock);
2154 /*
2155 * Every stopped thread goes here after wakeup. Check to see if
2156 * we should notify the parent, prepare_signal(SIGCONT) encodes
2157 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2158 */
2159 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2160 int why;
2161
2162 if (signal->flags & SIGNAL_CLD_CONTINUED)
2163 why = CLD_CONTINUED;
2164 else
2165 why = CLD_STOPPED;
2166
2167 signal->flags &= ~SIGNAL_CLD_MASK;
2168
2169 spin_unlock_irq(&sighand->siglock);
2170
2171 /*
2172 * Notify the parent that we're continuing. This event is
2173 * always per-process and doesn't make whole lot of sense
2174 * for ptracers, who shouldn't consume the state via
2175 * wait(2) either, but, for backward compatibility, notify
2176 * the ptracer of the group leader too unless it's gonna be
2177 * a duplicate.
2178 */
2179 read_lock(&tasklist_lock);
2180 do_notify_parent_cldstop(current, false, why);
2181
2182 if (ptrace_reparented(current->group_leader))
2183 do_notify_parent_cldstop(current->group_leader,
2184 true, why);
2185 read_unlock(&tasklist_lock);
2186
2187 goto relock;
2188 }
2189
2190 for (;;) {
2191 struct k_sigaction *ka;
2192
2193 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2194 do_signal_stop(0))
2195 goto relock;
2196
2197 if (unlikely(current->jobctl & JOBCTL_TRAP_MASK)) {
2198 do_jobctl_trap();
2199 spin_unlock_irq(&sighand->siglock);
2200 goto relock;
2201 }
2202
2203 signr = dequeue_signal(current, ¤t->blocked, info);
2204
2205 if (!signr)
2206 break; /* will return 0 */
2207
2208 if (unlikely(current->ptrace) && signr != SIGKILL) {
2209 signr = ptrace_signal(signr, info,
2210 regs, cookie);
2211 if (!signr)
2212 continue;
2213 }
2214
2215 ka = &sighand->action[signr-1];
2216
2217 /* Trace actually delivered signals. */
2218 trace_signal_deliver(signr, info, ka);
2219
2220 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2221 continue;
2222 if (ka->sa.sa_handler != SIG_DFL) {
2223 /* Run the handler. */
2224 *return_ka = *ka;
2225
2226 if (ka->sa.sa_flags & SA_ONESHOT)
2227 ka->sa.sa_handler = SIG_DFL;
2228
2229 break; /* will return non-zero "signr" value */
2230 }
2231
2232 /*
2233 * Now we are doing the default action for this signal.
2234 */
2235 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2236 continue;
2237
2238 /*
2239 * Global init gets no signals it doesn't want.
2240 * Container-init gets no signals it doesn't want from same
2241 * container.
2242 *
2243 * Note that if global/container-init sees a sig_kernel_only()
2244 * signal here, the signal must have been generated internally
2245 * or must have come from an ancestor namespace. In either
2246 * case, the signal cannot be dropped.
2247 */
2248 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2249 !sig_kernel_only(signr))
2250 continue;
2251
2252 if (sig_kernel_stop(signr)) {
2253 /*
2254 * The default action is to stop all threads in
2255 * the thread group. The job control signals
2256 * do nothing in an orphaned pgrp, but SIGSTOP
2257 * always works. Note that siglock needs to be
2258 * dropped during the call to is_orphaned_pgrp()
2259 * because of lock ordering with tasklist_lock.
2260 * This allows an intervening SIGCONT to be posted.
2261 * We need to check for that and bail out if necessary.
2262 */
2263 if (signr != SIGSTOP) {
2264 spin_unlock_irq(&sighand->siglock);
2265
2266 /* signals can be posted during this window */
2267
2268 if (is_current_pgrp_orphaned())
2269 goto relock;
2270
2271 spin_lock_irq(&sighand->siglock);
2272 }
2273
2274 if (likely(do_signal_stop(info->si_signo))) {
2275 /* It released the siglock. */
2276 goto relock;
2277 }
2278
2279 /*
2280 * We didn't actually stop, due to a race
2281 * with SIGCONT or something like that.
2282 */
2283 continue;
2284 }
2285
2286 spin_unlock_irq(&sighand->siglock);
2287
2288 /*
2289 * Anything else is fatal, maybe with a core dump.
2290 */
2291 current->flags |= PF_SIGNALED;
2292
2293 if (sig_kernel_coredump(signr)) {
2294 if (print_fatal_signals)
2295 print_fatal_signal(regs, info->si_signo);
2296 /*
2297 * If it was able to dump core, this kills all
2298 * other threads in the group and synchronizes with
2299 * their demise. If we lost the race with another
2300 * thread getting here, it set group_exit_code
2301 * first and our do_group_exit call below will use
2302 * that value and ignore the one we pass it.
2303 */
2304 do_coredump(info->si_signo, info->si_signo, regs);
2305 }
2306
2307 /*
2308 * Death signals, no core dump.
2309 */
2310 do_group_exit(info->si_signo);
2311 /* NOTREACHED */
2312 }
2313 spin_unlock_irq(&sighand->siglock);
2314 return signr;
2315}
2316
2317/*
2318 * It could be that complete_signal() picked us to notify about the
2319 * group-wide signal. Other threads should be notified now to take
2320 * the shared signals in @which since we will not.
2321 */
2322static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2323{
2324 sigset_t retarget;
2325 struct task_struct *t;
2326
2327 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2328 if (sigisemptyset(&retarget))
2329 return;
2330
2331 t = tsk;
2332 while_each_thread(tsk, t) {
2333 if (t->flags & PF_EXITING)
2334 continue;
2335
2336 if (!has_pending_signals(&retarget, &t->blocked))
2337 continue;
2338 /* Remove the signals this thread can handle. */
2339 sigandsets(&retarget, &retarget, &t->blocked);
2340
2341 if (!signal_pending(t))
2342 signal_wake_up(t, 0);
2343
2344 if (sigisemptyset(&retarget))
2345 break;
2346 }
2347}
2348
2349void exit_signals(struct task_struct *tsk)
2350{
2351 int group_stop = 0;
2352 sigset_t unblocked;
2353
2354 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2355 tsk->flags |= PF_EXITING;
2356 return;
2357 }
2358
2359 spin_lock_irq(&tsk->sighand->siglock);
2360 /*
2361 * From now this task is not visible for group-wide signals,
2362 * see wants_signal(), do_signal_stop().
2363 */
2364 tsk->flags |= PF_EXITING;
2365 if (!signal_pending(tsk))
2366 goto out;
2367
2368 unblocked = tsk->blocked;
2369 signotset(&unblocked);
2370 retarget_shared_pending(tsk, &unblocked);
2371
2372 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2373 task_participate_group_stop(tsk))
2374 group_stop = CLD_STOPPED;
2375out:
2376 spin_unlock_irq(&tsk->sighand->siglock);
2377
2378 /*
2379 * If group stop has completed, deliver the notification. This
2380 * should always go to the real parent of the group leader.
2381 */
2382 if (unlikely(group_stop)) {
2383 read_lock(&tasklist_lock);
2384 do_notify_parent_cldstop(tsk, false, group_stop);
2385 read_unlock(&tasklist_lock);
2386 }
2387}
2388
2389EXPORT_SYMBOL(recalc_sigpending);
2390EXPORT_SYMBOL_GPL(dequeue_signal);
2391EXPORT_SYMBOL(flush_signals);
2392EXPORT_SYMBOL(force_sig);
2393EXPORT_SYMBOL(send_sig);
2394EXPORT_SYMBOL(send_sig_info);
2395EXPORT_SYMBOL(sigprocmask);
2396EXPORT_SYMBOL(block_all_signals);
2397EXPORT_SYMBOL(unblock_all_signals);
2398
2399
2400/*
2401 * System call entry points.
2402 */
2403
2404/**
2405 * sys_restart_syscall - restart a system call
2406 */
2407SYSCALL_DEFINE0(restart_syscall)
2408{
2409 struct restart_block *restart = ¤t_thread_info()->restart_block;
2410 return restart->fn(restart);
2411}
2412
2413long do_no_restart_syscall(struct restart_block *param)
2414{
2415 return -EINTR;
2416}
2417
2418static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2419{
2420 if (signal_pending(tsk) && !thread_group_empty(tsk)) {
2421 sigset_t newblocked;
2422 /* A set of now blocked but previously unblocked signals. */
2423 sigandnsets(&newblocked, newset, ¤t->blocked);
2424 retarget_shared_pending(tsk, &newblocked);
2425 }
2426 tsk->blocked = *newset;
2427 recalc_sigpending();
2428}
2429
2430/**
2431 * set_current_blocked - change current->blocked mask
2432 * @newset: new mask
2433 *
2434 * It is wrong to change ->blocked directly, this helper should be used
2435 * to ensure the process can't miss a shared signal we are going to block.
2436 */
2437void set_current_blocked(const sigset_t *newset)
2438{
2439 struct task_struct *tsk = current;
2440
2441 spin_lock_irq(&tsk->sighand->siglock);
2442 __set_task_blocked(tsk, newset);
2443 spin_unlock_irq(&tsk->sighand->siglock);
2444}
2445
2446/*
2447 * This is also useful for kernel threads that want to temporarily
2448 * (or permanently) block certain signals.
2449 *
2450 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2451 * interface happily blocks "unblockable" signals like SIGKILL
2452 * and friends.
2453 */
2454int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2455{
2456 struct task_struct *tsk = current;
2457 sigset_t newset;
2458
2459 /* Lockless, only current can change ->blocked, never from irq */
2460 if (oldset)
2461 *oldset = tsk->blocked;
2462
2463 switch (how) {
2464 case SIG_BLOCK:
2465 sigorsets(&newset, &tsk->blocked, set);
2466 break;
2467 case SIG_UNBLOCK:
2468 sigandnsets(&newset, &tsk->blocked, set);
2469 break;
2470 case SIG_SETMASK:
2471 newset = *set;
2472 break;
2473 default:
2474 return -EINVAL;
2475 }
2476
2477 set_current_blocked(&newset);
2478 return 0;
2479}
2480
2481/**
2482 * sys_rt_sigprocmask - change the list of currently blocked signals
2483 * @how: whether to add, remove, or set signals
2484 * @nset: stores pending signals
2485 * @oset: previous value of signal mask if non-null
2486 * @sigsetsize: size of sigset_t type
2487 */
2488SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
2489 sigset_t __user *, oset, size_t, sigsetsize)
2490{
2491 sigset_t old_set, new_set;
2492 int error;
2493
2494 /* XXX: Don't preclude handling different sized sigset_t's. */
2495 if (sigsetsize != sizeof(sigset_t))
2496 return -EINVAL;
2497
2498 old_set = current->blocked;
2499
2500 if (nset) {
2501 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
2502 return -EFAULT;
2503 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2504
2505 error = sigprocmask(how, &new_set, NULL);
2506 if (error)
2507 return error;
2508 }
2509
2510 if (oset) {
2511 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
2512 return -EFAULT;
2513 }
2514
2515 return 0;
2516}
2517
2518long do_sigpending(void __user *set, unsigned long sigsetsize)
2519{
2520 long error = -EINVAL;
2521 sigset_t pending;
2522
2523 if (sigsetsize > sizeof(sigset_t))
2524 goto out;
2525
2526 spin_lock_irq(¤t->sighand->siglock);
2527 sigorsets(&pending, ¤t->pending.signal,
2528 ¤t->signal->shared_pending.signal);
2529 spin_unlock_irq(¤t->sighand->siglock);
2530
2531 /* Outside the lock because only this thread touches it. */
2532 sigandsets(&pending, ¤t->blocked, &pending);
2533
2534 error = -EFAULT;
2535 if (!copy_to_user(set, &pending, sigsetsize))
2536 error = 0;
2537
2538out:
2539 return error;
2540}
2541
2542/**
2543 * sys_rt_sigpending - examine a pending signal that has been raised
2544 * while blocked
2545 * @set: stores pending signals
2546 * @sigsetsize: size of sigset_t type or larger
2547 */
2548SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, set, size_t, sigsetsize)
2549{
2550 return do_sigpending(set, sigsetsize);
2551}
2552
2553#ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2554
2555int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2556{
2557 int err;
2558
2559 if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2560 return -EFAULT;
2561 if (from->si_code < 0)
2562 return __copy_to_user(to, from, sizeof(siginfo_t))
2563 ? -EFAULT : 0;
2564 /*
2565 * If you change siginfo_t structure, please be sure
2566 * this code is fixed accordingly.
2567 * Please remember to update the signalfd_copyinfo() function
2568 * inside fs/signalfd.c too, in case siginfo_t changes.
2569 * It should never copy any pad contained in the structure
2570 * to avoid security leaks, but must copy the generic
2571 * 3 ints plus the relevant union member.
2572 */
2573 err = __put_user(from->si_signo, &to->si_signo);
2574 err |= __put_user(from->si_errno, &to->si_errno);
2575 err |= __put_user((short)from->si_code, &to->si_code);
2576 switch (from->si_code & __SI_MASK) {
2577 case __SI_KILL:
2578 err |= __put_user(from->si_pid, &to->si_pid);
2579 err |= __put_user(from->si_uid, &to->si_uid);
2580 break;
2581 case __SI_TIMER:
2582 err |= __put_user(from->si_tid, &to->si_tid);
2583 err |= __put_user(from->si_overrun, &to->si_overrun);
2584 err |= __put_user(from->si_ptr, &to->si_ptr);
2585 break;
2586 case __SI_POLL:
2587 err |= __put_user(from->si_band, &to->si_band);
2588 err |= __put_user(from->si_fd, &to->si_fd);
2589 break;
2590 case __SI_FAULT:
2591 err |= __put_user(from->si_addr, &to->si_addr);
2592#ifdef __ARCH_SI_TRAPNO
2593 err |= __put_user(from->si_trapno, &to->si_trapno);
2594#endif
2595#ifdef BUS_MCEERR_AO
2596 /*
2597 * Other callers might not initialize the si_lsb field,
2598 * so check explicitly for the right codes here.
2599 */
2600 if (from->si_code == BUS_MCEERR_AR || from->si_code == BUS_MCEERR_AO)
2601 err |= __put_user(from->si_addr_lsb, &to->si_addr_lsb);
2602#endif
2603 break;
2604 case __SI_CHLD:
2605 err |= __put_user(from->si_pid, &to->si_pid);
2606 err |= __put_user(from->si_uid, &to->si_uid);
2607 err |= __put_user(from->si_status, &to->si_status);
2608 err |= __put_user(from->si_utime, &to->si_utime);
2609 err |= __put_user(from->si_stime, &to->si_stime);
2610 break;
2611 case __SI_RT: /* This is not generated by the kernel as of now. */
2612 case __SI_MESGQ: /* But this is */
2613 err |= __put_user(from->si_pid, &to->si_pid);
2614 err |= __put_user(from->si_uid, &to->si_uid);
2615 err |= __put_user(from->si_ptr, &to->si_ptr);
2616 break;
2617 default: /* this is just in case for now ... */
2618 err |= __put_user(from->si_pid, &to->si_pid);
2619 err |= __put_user(from->si_uid, &to->si_uid);
2620 break;
2621 }
2622 return err;
2623}
2624
2625#endif
2626
2627/**
2628 * do_sigtimedwait - wait for queued signals specified in @which
2629 * @which: queued signals to wait for
2630 * @info: if non-null, the signal's siginfo is returned here
2631 * @ts: upper bound on process time suspension
2632 */
2633int do_sigtimedwait(const sigset_t *which, siginfo_t *info,
2634 const struct timespec *ts)
2635{
2636 struct task_struct *tsk = current;
2637 long timeout = MAX_SCHEDULE_TIMEOUT;
2638 sigset_t mask = *which;
2639 int sig;
2640
2641 if (ts) {
2642 if (!timespec_valid(ts))
2643 return -EINVAL;
2644 timeout = timespec_to_jiffies(ts);
2645 /*
2646 * We can be close to the next tick, add another one
2647 * to ensure we will wait at least the time asked for.
2648 */
2649 if (ts->tv_sec || ts->tv_nsec)
2650 timeout++;
2651 }
2652
2653 /*
2654 * Invert the set of allowed signals to get those we want to block.
2655 */
2656 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
2657 signotset(&mask);
2658
2659 spin_lock_irq(&tsk->sighand->siglock);
2660 sig = dequeue_signal(tsk, &mask, info);
2661 if (!sig && timeout) {
2662 /*
2663 * None ready, temporarily unblock those we're interested
2664 * while we are sleeping in so that we'll be awakened when
2665 * they arrive. Unblocking is always fine, we can avoid
2666 * set_current_blocked().
2667 */
2668 tsk->real_blocked = tsk->blocked;
2669 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
2670 recalc_sigpending();
2671 spin_unlock_irq(&tsk->sighand->siglock);
2672
2673 timeout = schedule_timeout_interruptible(timeout);
2674
2675 spin_lock_irq(&tsk->sighand->siglock);
2676 __set_task_blocked(tsk, &tsk->real_blocked);
2677 siginitset(&tsk->real_blocked, 0);
2678 sig = dequeue_signal(tsk, &mask, info);
2679 }
2680 spin_unlock_irq(&tsk->sighand->siglock);
2681
2682 if (sig)
2683 return sig;
2684 return timeout ? -EINTR : -EAGAIN;
2685}
2686
2687/**
2688 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
2689 * in @uthese
2690 * @uthese: queued signals to wait for
2691 * @uinfo: if non-null, the signal's siginfo is returned here
2692 * @uts: upper bound on process time suspension
2693 * @sigsetsize: size of sigset_t type
2694 */
2695SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
2696 siginfo_t __user *, uinfo, const struct timespec __user *, uts,
2697 size_t, sigsetsize)
2698{
2699 sigset_t these;
2700 struct timespec ts;
2701 siginfo_t info;
2702 int ret;
2703
2704 /* XXX: Don't preclude handling different sized sigset_t's. */
2705 if (sigsetsize != sizeof(sigset_t))
2706 return -EINVAL;
2707
2708 if (copy_from_user(&these, uthese, sizeof(these)))
2709 return -EFAULT;
2710
2711 if (uts) {
2712 if (copy_from_user(&ts, uts, sizeof(ts)))
2713 return -EFAULT;
2714 }
2715
2716 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
2717
2718 if (ret > 0 && uinfo) {
2719 if (copy_siginfo_to_user(uinfo, &info))
2720 ret = -EFAULT;
2721 }
2722
2723 return ret;
2724}
2725
2726/**
2727 * sys_kill - send a signal to a process
2728 * @pid: the PID of the process
2729 * @sig: signal to be sent
2730 */
2731SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
2732{
2733 struct siginfo info;
2734
2735 info.si_signo = sig;
2736 info.si_errno = 0;
2737 info.si_code = SI_USER;
2738 info.si_pid = task_tgid_vnr(current);
2739 info.si_uid = current_uid();
2740
2741 return kill_something_info(sig, &info, pid);
2742}
2743
2744static int
2745do_send_specific(pid_t tgid, pid_t pid, int sig, struct siginfo *info)
2746{
2747 struct task_struct *p;
2748 int error = -ESRCH;
2749
2750 rcu_read_lock();
2751 p = find_task_by_vpid(pid);
2752 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
2753 error = check_kill_permission(sig, info, p);
2754 /*
2755 * The null signal is a permissions and process existence
2756 * probe. No signal is actually delivered.
2757 */
2758 if (!error && sig) {
2759 error = do_send_sig_info(sig, info, p, false);
2760 /*
2761 * If lock_task_sighand() failed we pretend the task
2762 * dies after receiving the signal. The window is tiny,
2763 * and the signal is private anyway.
2764 */
2765 if (unlikely(error == -ESRCH))
2766 error = 0;
2767 }
2768 }
2769 rcu_read_unlock();
2770
2771 return error;
2772}
2773
2774static int do_tkill(pid_t tgid, pid_t pid, int sig)
2775{
2776 struct siginfo info;
2777
2778 info.si_signo = sig;
2779 info.si_errno = 0;
2780 info.si_code = SI_TKILL;
2781 info.si_pid = task_tgid_vnr(current);
2782 info.si_uid = current_uid();
2783
2784 return do_send_specific(tgid, pid, sig, &info);
2785}
2786
2787/**
2788 * sys_tgkill - send signal to one specific thread
2789 * @tgid: the thread group ID of the thread
2790 * @pid: the PID of the thread
2791 * @sig: signal to be sent
2792 *
2793 * This syscall also checks the @tgid and returns -ESRCH even if the PID
2794 * exists but it's not belonging to the target process anymore. This
2795 * method solves the problem of threads exiting and PIDs getting reused.
2796 */
2797SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
2798{
2799 /* This is only valid for single tasks */
2800 if (pid <= 0 || tgid <= 0)
2801 return -EINVAL;
2802
2803 return do_tkill(tgid, pid, sig);
2804}
2805
2806/**
2807 * sys_tkill - send signal to one specific task
2808 * @pid: the PID of the task
2809 * @sig: signal to be sent
2810 *
2811 * Send a signal to only one task, even if it's a CLONE_THREAD task.
2812 */
2813SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
2814{
2815 /* This is only valid for single tasks */
2816 if (pid <= 0)
2817 return -EINVAL;
2818
2819 return do_tkill(0, pid, sig);
2820}
2821
2822/**
2823 * sys_rt_sigqueueinfo - send signal information to a signal
2824 * @pid: the PID of the thread
2825 * @sig: signal to be sent
2826 * @uinfo: signal info to be sent
2827 */
2828SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
2829 siginfo_t __user *, uinfo)
2830{
2831 siginfo_t info;
2832
2833 if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2834 return -EFAULT;
2835
2836 /* Not even root can pretend to send signals from the kernel.
2837 * Nor can they impersonate a kill()/tgkill(), which adds source info.
2838 */
2839 if (info.si_code >= 0 || info.si_code == SI_TKILL) {
2840 /* We used to allow any < 0 si_code */
2841 WARN_ON_ONCE(info.si_code < 0);
2842 return -EPERM;
2843 }
2844 info.si_signo = sig;
2845
2846 /* POSIX.1b doesn't mention process groups. */
2847 return kill_proc_info(sig, &info, pid);
2848}
2849
2850long do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, siginfo_t *info)
2851{
2852 /* This is only valid for single tasks */
2853 if (pid <= 0 || tgid <= 0)
2854 return -EINVAL;
2855
2856 /* Not even root can pretend to send signals from the kernel.
2857 * Nor can they impersonate a kill()/tgkill(), which adds source info.
2858 */
2859 if (info->si_code >= 0 || info->si_code == SI_TKILL) {
2860 /* We used to allow any < 0 si_code */
2861 WARN_ON_ONCE(info->si_code < 0);
2862 return -EPERM;
2863 }
2864 info->si_signo = sig;
2865
2866 return do_send_specific(tgid, pid, sig, info);
2867}
2868
2869SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
2870 siginfo_t __user *, uinfo)
2871{
2872 siginfo_t info;
2873
2874 if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2875 return -EFAULT;
2876
2877 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
2878}
2879
2880int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
2881{
2882 struct task_struct *t = current;
2883 struct k_sigaction *k;
2884 sigset_t mask;
2885
2886 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2887 return -EINVAL;
2888
2889 k = &t->sighand->action[sig-1];
2890
2891 spin_lock_irq(¤t->sighand->siglock);
2892 if (oact)
2893 *oact = *k;
2894
2895 if (act) {
2896 sigdelsetmask(&act->sa.sa_mask,
2897 sigmask(SIGKILL) | sigmask(SIGSTOP));
2898 *k = *act;
2899 /*
2900 * POSIX 3.3.1.3:
2901 * "Setting a signal action to SIG_IGN for a signal that is
2902 * pending shall cause the pending signal to be discarded,
2903 * whether or not it is blocked."
2904 *
2905 * "Setting a signal action to SIG_DFL for a signal that is
2906 * pending and whose default action is to ignore the signal
2907 * (for example, SIGCHLD), shall cause the pending signal to
2908 * be discarded, whether or not it is blocked"
2909 */
2910 if (sig_handler_ignored(sig_handler(t, sig), sig)) {
2911 sigemptyset(&mask);
2912 sigaddset(&mask, sig);
2913 rm_from_queue_full(&mask, &t->signal->shared_pending);
2914 do {
2915 rm_from_queue_full(&mask, &t->pending);
2916 t = next_thread(t);
2917 } while (t != current);
2918 }
2919 }
2920
2921 spin_unlock_irq(¤t->sighand->siglock);
2922 return 0;
2923}
2924
2925int
2926do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2927{
2928 stack_t oss;
2929 int error;
2930
2931 oss.ss_sp = (void __user *) current->sas_ss_sp;
2932 oss.ss_size = current->sas_ss_size;
2933 oss.ss_flags = sas_ss_flags(sp);
2934
2935 if (uss) {
2936 void __user *ss_sp;
2937 size_t ss_size;
2938 int ss_flags;
2939
2940 error = -EFAULT;
2941 if (!access_ok(VERIFY_READ, uss, sizeof(*uss)))
2942 goto out;
2943 error = __get_user(ss_sp, &uss->ss_sp) |
2944 __get_user(ss_flags, &uss->ss_flags) |
2945 __get_user(ss_size, &uss->ss_size);
2946 if (error)
2947 goto out;
2948
2949 error = -EPERM;
2950 if (on_sig_stack(sp))
2951 goto out;
2952
2953 error = -EINVAL;
2954 /*
2955 * Note - this code used to test ss_flags incorrectly:
2956 * old code may have been written using ss_flags==0
2957 * to mean ss_flags==SS_ONSTACK (as this was the only
2958 * way that worked) - this fix preserves that older
2959 * mechanism.
2960 */
2961 if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2962 goto out;
2963
2964 if (ss_flags == SS_DISABLE) {
2965 ss_size = 0;
2966 ss_sp = NULL;
2967 } else {
2968 error = -ENOMEM;
2969 if (ss_size < MINSIGSTKSZ)
2970 goto out;
2971 }
2972
2973 current->sas_ss_sp = (unsigned long) ss_sp;
2974 current->sas_ss_size = ss_size;
2975 }
2976
2977 error = 0;
2978 if (uoss) {
2979 error = -EFAULT;
2980 if (!access_ok(VERIFY_WRITE, uoss, sizeof(*uoss)))
2981 goto out;
2982 error = __put_user(oss.ss_sp, &uoss->ss_sp) |
2983 __put_user(oss.ss_size, &uoss->ss_size) |
2984 __put_user(oss.ss_flags, &uoss->ss_flags);
2985 }
2986
2987out:
2988 return error;
2989}
2990
2991#ifdef __ARCH_WANT_SYS_SIGPENDING
2992
2993/**
2994 * sys_sigpending - examine pending signals
2995 * @set: where mask of pending signal is returned
2996 */
2997SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, set)
2998{
2999 return do_sigpending(set, sizeof(*set));
3000}
3001
3002#endif
3003
3004#ifdef __ARCH_WANT_SYS_SIGPROCMASK
3005/**
3006 * sys_sigprocmask - examine and change blocked signals
3007 * @how: whether to add, remove, or set signals
3008 * @nset: signals to add or remove (if non-null)
3009 * @oset: previous value of signal mask if non-null
3010 *
3011 * Some platforms have their own version with special arguments;
3012 * others support only sys_rt_sigprocmask.
3013 */
3014
3015SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
3016 old_sigset_t __user *, oset)
3017{
3018 old_sigset_t old_set, new_set;
3019 sigset_t new_blocked;
3020
3021 old_set = current->blocked.sig[0];
3022
3023 if (nset) {
3024 if (copy_from_user(&new_set, nset, sizeof(*nset)))
3025 return -EFAULT;
3026 new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
3027
3028 new_blocked = current->blocked;
3029
3030 switch (how) {
3031 case SIG_BLOCK:
3032 sigaddsetmask(&new_blocked, new_set);
3033 break;
3034 case SIG_UNBLOCK:
3035 sigdelsetmask(&new_blocked, new_set);
3036 break;
3037 case SIG_SETMASK:
3038 new_blocked.sig[0] = new_set;
3039 break;
3040 default:
3041 return -EINVAL;
3042 }
3043
3044 set_current_blocked(&new_blocked);
3045 }
3046
3047 if (oset) {
3048 if (copy_to_user(oset, &old_set, sizeof(*oset)))
3049 return -EFAULT;
3050 }
3051
3052 return 0;
3053}
3054#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
3055
3056#ifdef __ARCH_WANT_SYS_RT_SIGACTION
3057/**
3058 * sys_rt_sigaction - alter an action taken by a process
3059 * @sig: signal to be sent
3060 * @act: new sigaction
3061 * @oact: used to save the previous sigaction
3062 * @sigsetsize: size of sigset_t type
3063 */
3064SYSCALL_DEFINE4(rt_sigaction, int, sig,
3065 const struct sigaction __user *, act,
3066 struct sigaction __user *, oact,
3067 size_t, sigsetsize)
3068{
3069 struct k_sigaction new_sa, old_sa;
3070 int ret = -EINVAL;
3071
3072 /* XXX: Don't preclude handling different sized sigset_t's. */
3073 if (sigsetsize != sizeof(sigset_t))
3074 goto out;
3075
3076 if (act) {
3077 if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
3078 return -EFAULT;
3079 }
3080
3081 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
3082
3083 if (!ret && oact) {
3084 if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
3085 return -EFAULT;
3086 }
3087out:
3088 return ret;
3089}
3090#endif /* __ARCH_WANT_SYS_RT_SIGACTION */
3091
3092#ifdef __ARCH_WANT_SYS_SGETMASK
3093
3094/*
3095 * For backwards compatibility. Functionality superseded by sigprocmask.
3096 */
3097SYSCALL_DEFINE0(sgetmask)
3098{
3099 /* SMP safe */
3100 return current->blocked.sig[0];
3101}
3102
3103SYSCALL_DEFINE1(ssetmask, int, newmask)
3104{
3105 int old = current->blocked.sig[0];
3106 sigset_t newset;
3107
3108 siginitset(&newset, newmask & ~(sigmask(SIGKILL) | sigmask(SIGSTOP)));
3109 set_current_blocked(&newset);
3110
3111 return old;
3112}
3113#endif /* __ARCH_WANT_SGETMASK */
3114
3115#ifdef __ARCH_WANT_SYS_SIGNAL
3116/*
3117 * For backwards compatibility. Functionality superseded by sigaction.
3118 */
3119SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
3120{
3121 struct k_sigaction new_sa, old_sa;
3122 int ret;
3123
3124 new_sa.sa.sa_handler = handler;
3125 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
3126 sigemptyset(&new_sa.sa.sa_mask);
3127
3128 ret = do_sigaction(sig, &new_sa, &old_sa);
3129
3130 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
3131}
3132#endif /* __ARCH_WANT_SYS_SIGNAL */
3133
3134#ifdef __ARCH_WANT_SYS_PAUSE
3135
3136SYSCALL_DEFINE0(pause)
3137{
3138 while (!signal_pending(current)) {
3139 current->state = TASK_INTERRUPTIBLE;
3140 schedule();
3141 }
3142 return -ERESTARTNOHAND;
3143}
3144
3145#endif
3146
3147#ifdef __ARCH_WANT_SYS_RT_SIGSUSPEND
3148/**
3149 * sys_rt_sigsuspend - replace the signal mask for a value with the
3150 * @unewset value until a signal is received
3151 * @unewset: new signal mask value
3152 * @sigsetsize: size of sigset_t type
3153 */
3154SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
3155{
3156 sigset_t newset;
3157
3158 /* XXX: Don't preclude handling different sized sigset_t's. */
3159 if (sigsetsize != sizeof(sigset_t))
3160 return -EINVAL;
3161
3162 if (copy_from_user(&newset, unewset, sizeof(newset)))
3163 return -EFAULT;
3164 sigdelsetmask(&newset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3165
3166 current->saved_sigmask = current->blocked;
3167 set_current_blocked(&newset);
3168
3169 current->state = TASK_INTERRUPTIBLE;
3170 schedule();
3171 set_restore_sigmask();
3172 return -ERESTARTNOHAND;
3173}
3174#endif /* __ARCH_WANT_SYS_RT_SIGSUSPEND */
3175
3176__attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
3177{
3178 return NULL;
3179}
3180
3181void __init signals_init(void)
3182{
3183 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
3184}
3185
3186#ifdef CONFIG_KGDB_KDB
3187#include <linux/kdb.h>
3188/*
3189 * kdb_send_sig_info - Allows kdb to send signals without exposing
3190 * signal internals. This function checks if the required locks are
3191 * available before calling the main signal code, to avoid kdb
3192 * deadlocks.
3193 */
3194void
3195kdb_send_sig_info(struct task_struct *t, struct siginfo *info)
3196{
3197 static struct task_struct *kdb_prev_t;
3198 int sig, new_t;
3199 if (!spin_trylock(&t->sighand->siglock)) {
3200 kdb_printf("Can't do kill command now.\n"
3201 "The sigmask lock is held somewhere else in "
3202 "kernel, try again later\n");
3203 return;
3204 }
3205 spin_unlock(&t->sighand->siglock);
3206 new_t = kdb_prev_t != t;
3207 kdb_prev_t = t;
3208 if (t->state != TASK_RUNNING && new_t) {
3209 kdb_printf("Process is not RUNNING, sending a signal from "
3210 "kdb risks deadlock\n"
3211 "on the run queue locks. "
3212 "The signal has _not_ been sent.\n"
3213 "Reissue the kill command if you want to risk "
3214 "the deadlock.\n");
3215 return;
3216 }
3217 sig = info->si_signo;
3218 if (send_sig_info(sig, info, t))
3219 kdb_printf("Fail to deliver Signal %d to process %d.\n",
3220 sig, t->pid);
3221 else
3222 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
3223}
3224#endif /* CONFIG_KGDB_KDB */