1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/kernel/signal.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  *
7  *  1997-11-02  Modified for POSIX.1b signals by Richard Henderson
8  *
9  *  2003-06-02  Jim Houston - Concurrent Computer Corp.
10  *		Changes to use preallocated sigqueue structures
11  *		to allow signals to be sent reliably.
12  */
13 
14 #include <linux/slab.h>
15 #include <linux/export.h>
16 #include <linux/init.h>
17 #include <linux/sched/mm.h>
18 #include <linux/sched/user.h>
19 #include <linux/sched/debug.h>
20 #include <linux/sched/task.h>
21 #include <linux/sched/task_stack.h>
22 #include <linux/sched/cputime.h>
23 #include <linux/file.h>
24 #include <linux/fs.h>
25 #include <linux/proc_fs.h>
26 #include <linux/tty.h>
27 #include <linux/binfmts.h>
28 #include <linux/coredump.h>
29 #include <linux/security.h>
30 #include <linux/syscalls.h>
31 #include <linux/ptrace.h>
32 #include <linux/signal.h>
33 #include <linux/signalfd.h>
34 #include <linux/ratelimit.h>
35 #include <linux/tracehook.h>
36 #include <linux/capability.h>
37 #include <linux/freezer.h>
38 #include <linux/pid_namespace.h>
39 #include <linux/nsproxy.h>
40 #include <linux/user_namespace.h>
41 #include <linux/uprobes.h>
42 #include <linux/compat.h>
43 #include <linux/cn_proc.h>
44 #include <linux/compiler.h>
45 #include <linux/posix-timers.h>
46 #include <linux/cgroup.h>
47 #include <linux/audit.h>
48 
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/signal.h>
51 
52 #include <asm/param.h>
53 #include <linux/uaccess.h>
54 #include <asm/unistd.h>
55 #include <asm/siginfo.h>
56 #include <asm/cacheflush.h>
57 
58 /*
59  * SLAB caches for signal bits.
60  */
61 
62 static struct kmem_cache *sigqueue_cachep;
63 
64 int print_fatal_signals __read_mostly;
65 
sig_handler(struct task_struct * t,int sig)66 static void __user *sig_handler(struct task_struct *t, int sig)
67 {
68 	return t->sighand->action[sig - 1].sa.sa_handler;
69 }
70 
sig_handler_ignored(void __user * handler,int sig)71 static inline bool sig_handler_ignored(void __user *handler, int sig)
72 {
73 	/* Is it explicitly or implicitly ignored? */
74 	return handler == SIG_IGN ||
75 	       (handler == SIG_DFL && sig_kernel_ignore(sig));
76 }
77 
sig_task_ignored(struct task_struct * t,int sig,bool force)78 static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
79 {
80 	void __user *handler;
81 
82 	handler = sig_handler(t, sig);
83 
84 	/* SIGKILL and SIGSTOP may not be sent to the global init */
85 	if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
86 		return true;
87 
88 	if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
89 	    handler == SIG_DFL && !(force && sig_kernel_only(sig)))
90 		return true;
91 
92 	/* Only allow kernel generated signals to this kthread */
93 	if (unlikely((t->flags & PF_KTHREAD) &&
94 		     (handler == SIG_KTHREAD_KERNEL) && !force))
95 		return true;
96 
97 	return sig_handler_ignored(handler, sig);
98 }
99 
sig_ignored(struct task_struct * t,int sig,bool force)100 static bool sig_ignored(struct task_struct *t, int sig, bool force)
101 {
102 	/*
103 	 * Blocked signals are never ignored, since the
104 	 * signal handler may change by the time it is
105 	 * unblocked.
106 	 */
107 	if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
108 		return false;
109 
110 	/*
111 	 * Tracers may want to know about even ignored signal unless it
112 	 * is SIGKILL which can't be reported anyway but can be ignored
113 	 * by SIGNAL_UNKILLABLE task.
114 	 */
115 	if (t->ptrace && sig != SIGKILL)
116 		return false;
117 
118 	return sig_task_ignored(t, sig, force);
119 }
120 
121 /*
122  * Re-calculate pending state from the set of locally pending
123  * signals, globally pending signals, and blocked signals.
124  */
has_pending_signals(sigset_t * signal,sigset_t * blocked)125 static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
126 {
127 	unsigned long ready;
128 	long i;
129 
130 	switch (_NSIG_WORDS) {
131 	default:
132 		for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
133 			ready |= signal->sig[i] &~ blocked->sig[i];
134 		break;
135 
136 	case 4: ready  = signal->sig[3] &~ blocked->sig[3];
137 		ready |= signal->sig[2] &~ blocked->sig[2];
138 		ready |= signal->sig[1] &~ blocked->sig[1];
139 		ready |= signal->sig[0] &~ blocked->sig[0];
140 		break;
141 
142 	case 2: ready  = signal->sig[1] &~ blocked->sig[1];
143 		ready |= signal->sig[0] &~ blocked->sig[0];
144 		break;
145 
146 	case 1: ready  = signal->sig[0] &~ blocked->sig[0];
147 	}
148 	return ready !=	0;
149 }
150 
151 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
152 
recalc_sigpending_tsk(struct task_struct * t)153 static bool recalc_sigpending_tsk(struct task_struct *t)
154 {
155 	if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) ||
156 	    PENDING(&t->pending, &t->blocked) ||
157 	    PENDING(&t->signal->shared_pending, &t->blocked) ||
158 	    cgroup_task_frozen(t)) {
159 		set_tsk_thread_flag(t, TIF_SIGPENDING);
160 		return true;
161 	}
162 
163 	/*
164 	 * We must never clear the flag in another thread, or in current
165 	 * when it's possible the current syscall is returning -ERESTART*.
166 	 * So we don't clear it here, and only callers who know they should do.
167 	 */
168 	return false;
169 }
170 
171 /*
172  * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
173  * This is superfluous when called on current, the wakeup is a harmless no-op.
174  */
recalc_sigpending_and_wake(struct task_struct * t)175 void recalc_sigpending_and_wake(struct task_struct *t)
176 {
177 	if (recalc_sigpending_tsk(t))
178 		signal_wake_up(t, 0);
179 }
180 
recalc_sigpending(void)181 void recalc_sigpending(void)
182 {
183 	if (!recalc_sigpending_tsk(current) && !freezing(current))
184 		clear_thread_flag(TIF_SIGPENDING);
185 
186 }
187 EXPORT_SYMBOL(recalc_sigpending);
188 
calculate_sigpending(void)189 void calculate_sigpending(void)
190 {
191 	/* Have any signals or users of TIF_SIGPENDING been delayed
192 	 * until after fork?
193 	 */
194 	spin_lock_irq(&current->sighand->siglock);
195 	set_tsk_thread_flag(current, TIF_SIGPENDING);
196 	recalc_sigpending();
197 	spin_unlock_irq(&current->sighand->siglock);
198 }
199 
200 /* Given the mask, find the first available signal that should be serviced. */
201 
202 #define SYNCHRONOUS_MASK \
203 	(sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
204 	 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
205 
next_signal(struct sigpending * pending,sigset_t * mask)206 int next_signal(struct sigpending *pending, sigset_t *mask)
207 {
208 	unsigned long i, *s, *m, x;
209 	int sig = 0;
210 
211 	s = pending->signal.sig;
212 	m = mask->sig;
213 
214 	/*
215 	 * Handle the first word specially: it contains the
216 	 * synchronous signals that need to be dequeued first.
217 	 */
218 	x = *s &~ *m;
219 	if (x) {
220 		if (x & SYNCHRONOUS_MASK)
221 			x &= SYNCHRONOUS_MASK;
222 		sig = ffz(~x) + 1;
223 		return sig;
224 	}
225 
226 	switch (_NSIG_WORDS) {
227 	default:
228 		for (i = 1; i < _NSIG_WORDS; ++i) {
229 			x = *++s &~ *++m;
230 			if (!x)
231 				continue;
232 			sig = ffz(~x) + i*_NSIG_BPW + 1;
233 			break;
234 		}
235 		break;
236 
237 	case 2:
238 		x = s[1] &~ m[1];
239 		if (!x)
240 			break;
241 		sig = ffz(~x) + _NSIG_BPW + 1;
242 		break;
243 
244 	case 1:
245 		/* Nothing to do */
246 		break;
247 	}
248 
249 	return sig;
250 }
251 
print_dropped_signal(int sig)252 static inline void print_dropped_signal(int sig)
253 {
254 	static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
255 
256 	if (!print_fatal_signals)
257 		return;
258 
259 	if (!__ratelimit(&ratelimit_state))
260 		return;
261 
262 	pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
263 				current->comm, current->pid, sig);
264 }
265 
266 /**
267  * task_set_jobctl_pending - set jobctl pending bits
268  * @task: target task
269  * @mask: pending bits to set
270  *
271  * Clear @mask from @task->jobctl.  @mask must be subset of
272  * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
273  * %JOBCTL_TRAPPING.  If stop signo is being set, the existing signo is
274  * cleared.  If @task is already being killed or exiting, this function
275  * becomes noop.
276  *
277  * CONTEXT:
278  * Must be called with @task->sighand->siglock held.
279  *
280  * RETURNS:
281  * %true if @mask is set, %false if made noop because @task was dying.
282  */
task_set_jobctl_pending(struct task_struct * task,unsigned long mask)283 bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
284 {
285 	BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
286 			JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
287 	BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
288 
289 	if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
290 		return false;
291 
292 	if (mask & JOBCTL_STOP_SIGMASK)
293 		task->jobctl &= ~JOBCTL_STOP_SIGMASK;
294 
295 	task->jobctl |= mask;
296 	return true;
297 }
298 
299 /**
300  * task_clear_jobctl_trapping - clear jobctl trapping bit
301  * @task: target task
302  *
303  * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
304  * Clear it and wake up the ptracer.  Note that we don't need any further
305  * locking.  @task->siglock guarantees that @task->parent points to the
306  * ptracer.
307  *
308  * CONTEXT:
309  * Must be called with @task->sighand->siglock held.
310  */
task_clear_jobctl_trapping(struct task_struct * task)311 void task_clear_jobctl_trapping(struct task_struct *task)
312 {
313 	if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
314 		task->jobctl &= ~JOBCTL_TRAPPING;
315 		smp_mb();	/* advised by wake_up_bit() */
316 		wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
317 	}
318 }
319 
320 /**
321  * task_clear_jobctl_pending - clear jobctl pending bits
322  * @task: target task
323  * @mask: pending bits to clear
324  *
325  * Clear @mask from @task->jobctl.  @mask must be subset of
326  * %JOBCTL_PENDING_MASK.  If %JOBCTL_STOP_PENDING is being cleared, other
327  * STOP bits are cleared together.
328  *
329  * If clearing of @mask leaves no stop or trap pending, this function calls
330  * task_clear_jobctl_trapping().
331  *
332  * CONTEXT:
333  * Must be called with @task->sighand->siglock held.
334  */
task_clear_jobctl_pending(struct task_struct * task,unsigned long mask)335 void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
336 {
337 	BUG_ON(mask & ~JOBCTL_PENDING_MASK);
338 
339 	if (mask & JOBCTL_STOP_PENDING)
340 		mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
341 
342 	task->jobctl &= ~mask;
343 
344 	if (!(task->jobctl & JOBCTL_PENDING_MASK))
345 		task_clear_jobctl_trapping(task);
346 }
347 
348 /**
349  * task_participate_group_stop - participate in a group stop
350  * @task: task participating in a group stop
351  *
352  * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
353  * Group stop states are cleared and the group stop count is consumed if
354  * %JOBCTL_STOP_CONSUME was set.  If the consumption completes the group
355  * stop, the appropriate `SIGNAL_*` flags are set.
356  *
357  * CONTEXT:
358  * Must be called with @task->sighand->siglock held.
359  *
360  * RETURNS:
361  * %true if group stop completion should be notified to the parent, %false
362  * otherwise.
363  */
task_participate_group_stop(struct task_struct * task)364 static bool task_participate_group_stop(struct task_struct *task)
365 {
366 	struct signal_struct *sig = task->signal;
367 	bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
368 
369 	WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
370 
371 	task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
372 
373 	if (!consume)
374 		return false;
375 
376 	if (!WARN_ON_ONCE(sig->group_stop_count == 0))
377 		sig->group_stop_count--;
378 
379 	/*
380 	 * Tell the caller to notify completion iff we are entering into a
381 	 * fresh group stop.  Read comment in do_signal_stop() for details.
382 	 */
383 	if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
384 		signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
385 		return true;
386 	}
387 	return false;
388 }
389 
task_join_group_stop(struct task_struct * task)390 void task_join_group_stop(struct task_struct *task)
391 {
392 	unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK;
393 	struct signal_struct *sig = current->signal;
394 
395 	if (sig->group_stop_count) {
396 		sig->group_stop_count++;
397 		mask |= JOBCTL_STOP_CONSUME;
398 	} else if (!(sig->flags & SIGNAL_STOP_STOPPED))
399 		return;
400 
401 	/* Have the new thread join an on-going signal group stop */
402 	task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING);
403 }
404 
405 /*
406  * allocate a new signal queue record
407  * - this may be called without locks if and only if t == current, otherwise an
408  *   appropriate lock must be held to stop the target task from exiting
409  */
410 static struct sigqueue *
__sigqueue_alloc(int sig,struct task_struct * t,gfp_t gfp_flags,int override_rlimit,const unsigned int sigqueue_flags)411 __sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
412 		 int override_rlimit, const unsigned int sigqueue_flags)
413 {
414 	struct sigqueue *q = NULL;
415 	struct user_struct *user;
416 	int sigpending;
417 
418 	/*
419 	 * Protect access to @t credentials. This can go away when all
420 	 * callers hold rcu read lock.
421 	 *
422 	 * NOTE! A pending signal will hold on to the user refcount,
423 	 * and we get/put the refcount only when the sigpending count
424 	 * changes from/to zero.
425 	 */
426 	rcu_read_lock();
427 	user = __task_cred(t)->user;
428 	sigpending = atomic_inc_return(&user->sigpending);
429 	if (sigpending == 1)
430 		get_uid(user);
431 	rcu_read_unlock();
432 
433 	if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
434 		/*
435 		 * Preallocation does not hold sighand::siglock so it can't
436 		 * use the cache. The lockless caching requires that only
437 		 * one consumer and only one producer run at a time.
438 		 */
439 		q = READ_ONCE(t->sigqueue_cache);
440 		if (!q || sigqueue_flags)
441 			q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
442 		else
443 			WRITE_ONCE(t->sigqueue_cache, NULL);
444 	} else {
445 		print_dropped_signal(sig);
446 	}
447 
448 	if (unlikely(q == NULL)) {
449 		if (atomic_dec_and_test(&user->sigpending))
450 			free_uid(user);
451 	} else {
452 		INIT_LIST_HEAD(&q->list);
453 		q->flags = sigqueue_flags;
454 		q->user = user;
455 	}
456 
457 	return q;
458 }
459 
exit_task_sigqueue_cache(struct task_struct * tsk)460 void exit_task_sigqueue_cache(struct task_struct *tsk)
461 {
462 	/* Race free because @tsk is mopped up */
463 	struct sigqueue *q = tsk->sigqueue_cache;
464 
465 	if (q) {
466 		tsk->sigqueue_cache = NULL;
467 		/*
468 		 * Hand it back to the cache as the task might
469 		 * be self reaping which would leak the object.
470 		 */
471 		 kmem_cache_free(sigqueue_cachep, q);
472 	}
473 }
474 
sigqueue_cache_or_free(struct sigqueue * q)475 static void sigqueue_cache_or_free(struct sigqueue *q)
476 {
477 	/*
478 	 * Cache one sigqueue per task. This pairs with the consumer side
479 	 * in __sigqueue_alloc() and needs READ/WRITE_ONCE() to prevent the
480 	 * compiler from store tearing and to tell KCSAN that the data race
481 	 * is intentional when run without holding current->sighand->siglock,
482 	 * which is fine as current obviously cannot run __sigqueue_free()
483 	 * concurrently.
484 	 */
485 	if (!READ_ONCE(current->sigqueue_cache))
486 		WRITE_ONCE(current->sigqueue_cache, q);
487 	else
488 		kmem_cache_free(sigqueue_cachep, q);
489 }
490 
__sigqueue_free(struct sigqueue * q)491 static void __sigqueue_free(struct sigqueue *q)
492 {
493 	if (q->flags & SIGQUEUE_PREALLOC)
494 		return;
495 	if (atomic_dec_and_test(&q->user->sigpending))
496 		free_uid(q->user);
497 	sigqueue_cache_or_free(q);
498 }
499 
flush_sigqueue(struct sigpending * queue)500 void flush_sigqueue(struct sigpending *queue)
501 {
502 	struct sigqueue *q;
503 
504 	sigemptyset(&queue->signal);
505 	while (!list_empty(&queue->list)) {
506 		q = list_entry(queue->list.next, struct sigqueue , list);
507 		list_del_init(&q->list);
508 		__sigqueue_free(q);
509 	}
510 }
511 
512 /*
513  * Flush all pending signals for this kthread.
514  */
flush_signals(struct task_struct * t)515 void flush_signals(struct task_struct *t)
516 {
517 	unsigned long flags;
518 
519 	spin_lock_irqsave(&t->sighand->siglock, flags);
520 	clear_tsk_thread_flag(t, TIF_SIGPENDING);
521 	flush_sigqueue(&t->pending);
522 	flush_sigqueue(&t->signal->shared_pending);
523 	spin_unlock_irqrestore(&t->sighand->siglock, flags);
524 }
525 EXPORT_SYMBOL(flush_signals);
526 
527 #ifdef CONFIG_POSIX_TIMERS
__flush_itimer_signals(struct sigpending * pending)528 static void __flush_itimer_signals(struct sigpending *pending)
529 {
530 	sigset_t signal, retain;
531 	struct sigqueue *q, *n;
532 
533 	signal = pending->signal;
534 	sigemptyset(&retain);
535 
536 	list_for_each_entry_safe(q, n, &pending->list, list) {
537 		int sig = q->info.si_signo;
538 
539 		if (likely(q->info.si_code != SI_TIMER)) {
540 			sigaddset(&retain, sig);
541 		} else {
542 			sigdelset(&signal, sig);
543 			list_del_init(&q->list);
544 			__sigqueue_free(q);
545 		}
546 	}
547 
548 	sigorsets(&pending->signal, &signal, &retain);
549 }
550 
flush_itimer_signals(void)551 void flush_itimer_signals(void)
552 {
553 	struct task_struct *tsk = current;
554 	unsigned long flags;
555 
556 	spin_lock_irqsave(&tsk->sighand->siglock, flags);
557 	__flush_itimer_signals(&tsk->pending);
558 	__flush_itimer_signals(&tsk->signal->shared_pending);
559 	spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
560 }
561 #endif
562 
ignore_signals(struct task_struct * t)563 void ignore_signals(struct task_struct *t)
564 {
565 	int i;
566 
567 	for (i = 0; i < _NSIG; ++i)
568 		t->sighand->action[i].sa.sa_handler = SIG_IGN;
569 
570 	flush_signals(t);
571 }
572 
573 /*
574  * Flush all handlers for a task.
575  */
576 
577 void
flush_signal_handlers(struct task_struct * t,int force_default)578 flush_signal_handlers(struct task_struct *t, int force_default)
579 {
580 	int i;
581 	struct k_sigaction *ka = &t->sighand->action[0];
582 	for (i = _NSIG ; i != 0 ; i--) {
583 		if (force_default || ka->sa.sa_handler != SIG_IGN)
584 			ka->sa.sa_handler = SIG_DFL;
585 		ka->sa.sa_flags = 0;
586 #ifdef __ARCH_HAS_SA_RESTORER
587 		ka->sa.sa_restorer = NULL;
588 #endif
589 		sigemptyset(&ka->sa.sa_mask);
590 		ka++;
591 	}
592 }
593 
unhandled_signal(struct task_struct * tsk,int sig)594 bool unhandled_signal(struct task_struct *tsk, int sig)
595 {
596 	void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
597 	if (is_global_init(tsk))
598 		return true;
599 
600 	if (handler != SIG_IGN && handler != SIG_DFL)
601 		return false;
602 
603 	/* if ptraced, let the tracer determine */
604 	return !tsk->ptrace;
605 }
606 
collect_signal(int sig,struct sigpending * list,kernel_siginfo_t * info,bool * resched_timer)607 static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
608 			   bool *resched_timer)
609 {
610 	struct sigqueue *q, *first = NULL;
611 
612 	/*
613 	 * Collect the siginfo appropriate to this signal.  Check if
614 	 * there is another siginfo for the same signal.
615 	*/
616 	list_for_each_entry(q, &list->list, list) {
617 		if (q->info.si_signo == sig) {
618 			if (first)
619 				goto still_pending;
620 			first = q;
621 		}
622 	}
623 
624 	sigdelset(&list->signal, sig);
625 
626 	if (first) {
627 still_pending:
628 		list_del_init(&first->list);
629 		copy_siginfo(info, &first->info);
630 
631 		*resched_timer =
632 			(first->flags & SIGQUEUE_PREALLOC) &&
633 			(info->si_code == SI_TIMER) &&
634 			(info->si_sys_private);
635 
636 		__sigqueue_free(first);
637 	} else {
638 		/*
639 		 * Ok, it wasn't in the queue.  This must be
640 		 * a fast-pathed signal or we must have been
641 		 * out of queue space.  So zero out the info.
642 		 */
643 		clear_siginfo(info);
644 		info->si_signo = sig;
645 		info->si_errno = 0;
646 		info->si_code = SI_USER;
647 		info->si_pid = 0;
648 		info->si_uid = 0;
649 	}
650 }
651 
__dequeue_signal(struct sigpending * pending,sigset_t * mask,kernel_siginfo_t * info,bool * resched_timer)652 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
653 			kernel_siginfo_t *info, bool *resched_timer)
654 {
655 	int sig = next_signal(pending, mask);
656 
657 	if (sig)
658 		collect_signal(sig, pending, info, resched_timer);
659 	return sig;
660 }
661 
662 /*
663  * Dequeue a signal and return the element to the caller, which is
664  * expected to free it.
665  *
666  * All callers have to hold the siglock.
667  */
dequeue_signal(struct task_struct * tsk,sigset_t * mask,kernel_siginfo_t * info)668 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, kernel_siginfo_t *info)
669 {
670 	bool resched_timer = false;
671 	int signr;
672 
673 	/* We only dequeue private signals from ourselves, we don't let
674 	 * signalfd steal them
675 	 */
676 	signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
677 	if (!signr) {
678 		signr = __dequeue_signal(&tsk->signal->shared_pending,
679 					 mask, info, &resched_timer);
680 #ifdef CONFIG_POSIX_TIMERS
681 		/*
682 		 * itimer signal ?
683 		 *
684 		 * itimers are process shared and we restart periodic
685 		 * itimers in the signal delivery path to prevent DoS
686 		 * attacks in the high resolution timer case. This is
687 		 * compliant with the old way of self-restarting
688 		 * itimers, as the SIGALRM is a legacy signal and only
689 		 * queued once. Changing the restart behaviour to
690 		 * restart the timer in the signal dequeue path is
691 		 * reducing the timer noise on heavy loaded !highres
692 		 * systems too.
693 		 */
694 		if (unlikely(signr == SIGALRM)) {
695 			struct hrtimer *tmr = &tsk->signal->real_timer;
696 
697 			if (!hrtimer_is_queued(tmr) &&
698 			    tsk->signal->it_real_incr != 0) {
699 				hrtimer_forward(tmr, tmr->base->get_time(),
700 						tsk->signal->it_real_incr);
701 				hrtimer_restart(tmr);
702 			}
703 		}
704 #endif
705 	}
706 
707 	recalc_sigpending();
708 	if (!signr)
709 		return 0;
710 
711 	if (unlikely(sig_kernel_stop(signr))) {
712 		/*
713 		 * Set a marker that we have dequeued a stop signal.  Our
714 		 * caller might release the siglock and then the pending
715 		 * stop signal it is about to process is no longer in the
716 		 * pending bitmasks, but must still be cleared by a SIGCONT
717 		 * (and overruled by a SIGKILL).  So those cases clear this
718 		 * shared flag after we've set it.  Note that this flag may
719 		 * remain set after the signal we return is ignored or
720 		 * handled.  That doesn't matter because its only purpose
721 		 * is to alert stop-signal processing code when another
722 		 * processor has come along and cleared the flag.
723 		 */
724 		current->jobctl |= JOBCTL_STOP_DEQUEUED;
725 	}
726 #ifdef CONFIG_POSIX_TIMERS
727 	if (resched_timer) {
728 		/*
729 		 * Release the siglock to ensure proper locking order
730 		 * of timer locks outside of siglocks.  Note, we leave
731 		 * irqs disabled here, since the posix-timers code is
732 		 * about to disable them again anyway.
733 		 */
734 		spin_unlock(&tsk->sighand->siglock);
735 		posixtimer_rearm(info);
736 		spin_lock(&tsk->sighand->siglock);
737 
738 		/* Don't expose the si_sys_private value to userspace */
739 		info->si_sys_private = 0;
740 	}
741 #endif
742 	return signr;
743 }
744 EXPORT_SYMBOL_GPL(dequeue_signal);
745 
dequeue_synchronous_signal(kernel_siginfo_t * info)746 static int dequeue_synchronous_signal(kernel_siginfo_t *info)
747 {
748 	struct task_struct *tsk = current;
749 	struct sigpending *pending = &tsk->pending;
750 	struct sigqueue *q, *sync = NULL;
751 
752 	/*
753 	 * Might a synchronous signal be in the queue?
754 	 */
755 	if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
756 		return 0;
757 
758 	/*
759 	 * Return the first synchronous signal in the queue.
760 	 */
761 	list_for_each_entry(q, &pending->list, list) {
762 		/* Synchronous signals have a positive si_code */
763 		if ((q->info.si_code > SI_USER) &&
764 		    (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
765 			sync = q;
766 			goto next;
767 		}
768 	}
769 	return 0;
770 next:
771 	/*
772 	 * Check if there is another siginfo for the same signal.
773 	 */
774 	list_for_each_entry_continue(q, &pending->list, list) {
775 		if (q->info.si_signo == sync->info.si_signo)
776 			goto still_pending;
777 	}
778 
779 	sigdelset(&pending->signal, sync->info.si_signo);
780 	recalc_sigpending();
781 still_pending:
782 	list_del_init(&sync->list);
783 	copy_siginfo(info, &sync->info);
784 	__sigqueue_free(sync);
785 	return info->si_signo;
786 }
787 
788 /*
789  * Tell a process that it has a new active signal..
790  *
791  * NOTE! we rely on the previous spin_lock to
792  * lock interrupts for us! We can only be called with
793  * "siglock" held, and the local interrupt must
794  * have been disabled when that got acquired!
795  *
796  * No need to set need_resched since signal event passing
797  * goes through ->blocked
798  */
signal_wake_up_state(struct task_struct * t,unsigned int state)799 void signal_wake_up_state(struct task_struct *t, unsigned int state)
800 {
801 	set_tsk_thread_flag(t, TIF_SIGPENDING);
802 	/*
803 	 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
804 	 * case. We don't check t->state here because there is a race with it
805 	 * executing another processor and just now entering stopped state.
806 	 * By using wake_up_state, we ensure the process will wake up and
807 	 * handle its death signal.
808 	 */
809 	if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
810 		kick_process(t);
811 }
812 
813 /*
814  * Remove signals in mask from the pending set and queue.
815  * Returns 1 if any signals were found.
816  *
817  * All callers must be holding the siglock.
818  */
flush_sigqueue_mask(sigset_t * mask,struct sigpending * s)819 static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
820 {
821 	struct sigqueue *q, *n;
822 	sigset_t m;
823 
824 	sigandsets(&m, mask, &s->signal);
825 	if (sigisemptyset(&m))
826 		return;
827 
828 	sigandnsets(&s->signal, &s->signal, mask);
829 	list_for_each_entry_safe(q, n, &s->list, list) {
830 		if (sigismember(mask, q->info.si_signo)) {
831 			list_del_init(&q->list);
832 			__sigqueue_free(q);
833 		}
834 	}
835 }
836 
is_si_special(const struct kernel_siginfo * info)837 static inline int is_si_special(const struct kernel_siginfo *info)
838 {
839 	return info <= SEND_SIG_PRIV;
840 }
841 
si_fromuser(const struct kernel_siginfo * info)842 static inline bool si_fromuser(const struct kernel_siginfo *info)
843 {
844 	return info == SEND_SIG_NOINFO ||
845 		(!is_si_special(info) && SI_FROMUSER(info));
846 }
847 
848 /*
849  * called with RCU read lock from check_kill_permission()
850  */
kill_ok_by_cred(struct task_struct * t)851 static bool kill_ok_by_cred(struct task_struct *t)
852 {
853 	const struct cred *cred = current_cred();
854 	const struct cred *tcred = __task_cred(t);
855 
856 	return uid_eq(cred->euid, tcred->suid) ||
857 	       uid_eq(cred->euid, tcred->uid) ||
858 	       uid_eq(cred->uid, tcred->suid) ||
859 	       uid_eq(cred->uid, tcred->uid) ||
860 	       ns_capable(tcred->user_ns, CAP_KILL);
861 }
862 
863 /*
864  * Bad permissions for sending the signal
865  * - the caller must hold the RCU read lock
866  */
check_kill_permission(int sig,struct kernel_siginfo * info,struct task_struct * t)867 static int check_kill_permission(int sig, struct kernel_siginfo *info,
868 				 struct task_struct *t)
869 {
870 	struct pid *sid;
871 	int error;
872 
873 	if (!valid_signal(sig))
874 		return -EINVAL;
875 
876 	if (!si_fromuser(info))
877 		return 0;
878 
879 	error = audit_signal_info(sig, t); /* Let audit system see the signal */
880 	if (error)
881 		return error;
882 
883 	if (!same_thread_group(current, t) &&
884 	    !kill_ok_by_cred(t)) {
885 		switch (sig) {
886 		case SIGCONT:
887 			sid = task_session(t);
888 			/*
889 			 * We don't return the error if sid == NULL. The
890 			 * task was unhashed, the caller must notice this.
891 			 */
892 			if (!sid || sid == task_session(current))
893 				break;
894 			fallthrough;
895 		default:
896 			return -EPERM;
897 		}
898 	}
899 
900 	return security_task_kill(t, info, sig, NULL);
901 }
902 
903 /**
904  * ptrace_trap_notify - schedule trap to notify ptracer
905  * @t: tracee wanting to notify tracer
906  *
907  * This function schedules sticky ptrace trap which is cleared on the next
908  * TRAP_STOP to notify ptracer of an event.  @t must have been seized by
909  * ptracer.
910  *
911  * If @t is running, STOP trap will be taken.  If trapped for STOP and
912  * ptracer is listening for events, tracee is woken up so that it can
913  * re-trap for the new event.  If trapped otherwise, STOP trap will be
914  * eventually taken without returning to userland after the existing traps
915  * are finished by PTRACE_CONT.
916  *
917  * CONTEXT:
918  * Must be called with @task->sighand->siglock held.
919  */
ptrace_trap_notify(struct task_struct * t)920 static void ptrace_trap_notify(struct task_struct *t)
921 {
922 	WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
923 	assert_spin_locked(&t->sighand->siglock);
924 
925 	task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
926 	ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
927 }
928 
929 /*
930  * Handle magic process-wide effects of stop/continue signals. Unlike
931  * the signal actions, these happen immediately at signal-generation
932  * time regardless of blocking, ignoring, or handling.  This does the
933  * actual continuing for SIGCONT, but not the actual stopping for stop
934  * signals. The process stop is done as a signal action for SIG_DFL.
935  *
936  * Returns true if the signal should be actually delivered, otherwise
937  * it should be dropped.
938  */
prepare_signal(int sig,struct task_struct * p,bool force)939 static bool prepare_signal(int sig, struct task_struct *p, bool force)
940 {
941 	struct signal_struct *signal = p->signal;
942 	struct task_struct *t;
943 	sigset_t flush;
944 
945 	if (signal->flags & (SIGNAL_GROUP_EXIT | SIGNAL_GROUP_COREDUMP)) {
946 		if (!(signal->flags & SIGNAL_GROUP_EXIT))
947 			return sig == SIGKILL;
948 		/*
949 		 * The process is in the middle of dying, nothing to do.
950 		 */
951 	} else if (sig_kernel_stop(sig)) {
952 		/*
953 		 * This is a stop signal.  Remove SIGCONT from all queues.
954 		 */
955 		siginitset(&flush, sigmask(SIGCONT));
956 		flush_sigqueue_mask(&flush, &signal->shared_pending);
957 		for_each_thread(p, t)
958 			flush_sigqueue_mask(&flush, &t->pending);
959 	} else if (sig == SIGCONT) {
960 		unsigned int why;
961 		/*
962 		 * Remove all stop signals from all queues, wake all threads.
963 		 */
964 		siginitset(&flush, SIG_KERNEL_STOP_MASK);
965 		flush_sigqueue_mask(&flush, &signal->shared_pending);
966 		for_each_thread(p, t) {
967 			flush_sigqueue_mask(&flush, &t->pending);
968 			task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
969 			if (likely(!(t->ptrace & PT_SEIZED)))
970 				wake_up_state(t, __TASK_STOPPED);
971 			else
972 				ptrace_trap_notify(t);
973 		}
974 
975 		/*
976 		 * Notify the parent with CLD_CONTINUED if we were stopped.
977 		 *
978 		 * If we were in the middle of a group stop, we pretend it
979 		 * was already finished, and then continued. Since SIGCHLD
980 		 * doesn't queue we report only CLD_STOPPED, as if the next
981 		 * CLD_CONTINUED was dropped.
982 		 */
983 		why = 0;
984 		if (signal->flags & SIGNAL_STOP_STOPPED)
985 			why |= SIGNAL_CLD_CONTINUED;
986 		else if (signal->group_stop_count)
987 			why |= SIGNAL_CLD_STOPPED;
988 
989 		if (why) {
990 			/*
991 			 * The first thread which returns from do_signal_stop()
992 			 * will take ->siglock, notice SIGNAL_CLD_MASK, and
993 			 * notify its parent. See get_signal().
994 			 */
995 			signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
996 			signal->group_stop_count = 0;
997 			signal->group_exit_code = 0;
998 		}
999 	}
1000 
1001 	return !sig_ignored(p, sig, force);
1002 }
1003 
1004 /*
1005  * Test if P wants to take SIG.  After we've checked all threads with this,
1006  * it's equivalent to finding no threads not blocking SIG.  Any threads not
1007  * blocking SIG were ruled out because they are not running and already
1008  * have pending signals.  Such threads will dequeue from the shared queue
1009  * as soon as they're available, so putting the signal on the shared queue
1010  * will be equivalent to sending it to one such thread.
1011  */
wants_signal(int sig,struct task_struct * p)1012 static inline bool wants_signal(int sig, struct task_struct *p)
1013 {
1014 	if (sigismember(&p->blocked, sig))
1015 		return false;
1016 
1017 	if (p->flags & PF_EXITING)
1018 		return false;
1019 
1020 	if (sig == SIGKILL)
1021 		return true;
1022 
1023 	if (task_is_stopped_or_traced(p))
1024 		return false;
1025 
1026 	return task_curr(p) || !task_sigpending(p);
1027 }
1028 
complete_signal(int sig,struct task_struct * p,enum pid_type type)1029 static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
1030 {
1031 	struct signal_struct *signal = p->signal;
1032 	struct task_struct *t;
1033 
1034 	/*
1035 	 * Now find a thread we can wake up to take the signal off the queue.
1036 	 *
1037 	 * If the main thread wants the signal, it gets first crack.
1038 	 * Probably the least surprising to the average bear.
1039 	 */
1040 	if (wants_signal(sig, p))
1041 		t = p;
1042 	else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1043 		/*
1044 		 * There is just one thread and it does not need to be woken.
1045 		 * It will dequeue unblocked signals before it runs again.
1046 		 */
1047 		return;
1048 	else {
1049 		/*
1050 		 * Otherwise try to find a suitable thread.
1051 		 */
1052 		t = signal->curr_target;
1053 		while (!wants_signal(sig, t)) {
1054 			t = next_thread(t);
1055 			if (t == signal->curr_target)
1056 				/*
1057 				 * No thread needs to be woken.
1058 				 * Any eligible threads will see
1059 				 * the signal in the queue soon.
1060 				 */
1061 				return;
1062 		}
1063 		signal->curr_target = t;
1064 	}
1065 
1066 	/*
1067 	 * Found a killable thread.  If the signal will be fatal,
1068 	 * then start taking the whole group down immediately.
1069 	 */
1070 	if (sig_fatal(p, sig) &&
1071 	    !(signal->flags & SIGNAL_GROUP_EXIT) &&
1072 	    !sigismember(&t->real_blocked, sig) &&
1073 	    (sig == SIGKILL || !p->ptrace)) {
1074 		/*
1075 		 * This signal will be fatal to the whole group.
1076 		 */
1077 		if (!sig_kernel_coredump(sig)) {
1078 			/*
1079 			 * Start a group exit and wake everybody up.
1080 			 * This way we don't have other threads
1081 			 * running and doing things after a slower
1082 			 * thread has the fatal signal pending.
1083 			 */
1084 			signal->flags = SIGNAL_GROUP_EXIT;
1085 			signal->group_exit_code = sig;
1086 			signal->group_stop_count = 0;
1087 			t = p;
1088 			do {
1089 				task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1090 				sigaddset(&t->pending.signal, SIGKILL);
1091 				signal_wake_up(t, 1);
1092 			} while_each_thread(p, t);
1093 			return;
1094 		}
1095 	}
1096 
1097 	/*
1098 	 * The signal is already in the shared-pending queue.
1099 	 * Tell the chosen thread to wake up and dequeue it.
1100 	 */
1101 	signal_wake_up(t, sig == SIGKILL);
1102 	return;
1103 }
1104 
legacy_queue(struct sigpending * signals,int sig)1105 static inline bool legacy_queue(struct sigpending *signals, int sig)
1106 {
1107 	return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1108 }
1109 
__send_signal(int sig,struct kernel_siginfo * info,struct task_struct * t,enum pid_type type,bool force)1110 static int __send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1111 			enum pid_type type, bool force)
1112 {
1113 	struct sigpending *pending;
1114 	struct sigqueue *q;
1115 	int override_rlimit;
1116 	int ret = 0, result;
1117 
1118 	assert_spin_locked(&t->sighand->siglock);
1119 
1120 	result = TRACE_SIGNAL_IGNORED;
1121 	if (!prepare_signal(sig, t, force))
1122 		goto ret;
1123 
1124 	pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1125 	/*
1126 	 * Short-circuit ignored signals and support queuing
1127 	 * exactly one non-rt signal, so that we can get more
1128 	 * detailed information about the cause of the signal.
1129 	 */
1130 	result = TRACE_SIGNAL_ALREADY_PENDING;
1131 	if (legacy_queue(pending, sig))
1132 		goto ret;
1133 
1134 	result = TRACE_SIGNAL_DELIVERED;
1135 	/*
1136 	 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1137 	 */
1138 	if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1139 		goto out_set;
1140 
1141 	/*
1142 	 * Real-time signals must be queued if sent by sigqueue, or
1143 	 * some other real-time mechanism.  It is implementation
1144 	 * defined whether kill() does so.  We attempt to do so, on
1145 	 * the principle of least surprise, but since kill is not
1146 	 * allowed to fail with EAGAIN when low on memory we just
1147 	 * make sure at least one signal gets delivered and don't
1148 	 * pass on the info struct.
1149 	 */
1150 	if (sig < SIGRTMIN)
1151 		override_rlimit = (is_si_special(info) || info->si_code >= 0);
1152 	else
1153 		override_rlimit = 0;
1154 
1155 	q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit, 0);
1156 
1157 	if (q) {
1158 		list_add_tail(&q->list, &pending->list);
1159 		switch ((unsigned long) info) {
1160 		case (unsigned long) SEND_SIG_NOINFO:
1161 			clear_siginfo(&q->info);
1162 			q->info.si_signo = sig;
1163 			q->info.si_errno = 0;
1164 			q->info.si_code = SI_USER;
1165 			q->info.si_pid = task_tgid_nr_ns(current,
1166 							task_active_pid_ns(t));
1167 			rcu_read_lock();
1168 			q->info.si_uid =
1169 				from_kuid_munged(task_cred_xxx(t, user_ns),
1170 						 current_uid());
1171 			rcu_read_unlock();
1172 			break;
1173 		case (unsigned long) SEND_SIG_PRIV:
1174 			clear_siginfo(&q->info);
1175 			q->info.si_signo = sig;
1176 			q->info.si_errno = 0;
1177 			q->info.si_code = SI_KERNEL;
1178 			q->info.si_pid = 0;
1179 			q->info.si_uid = 0;
1180 			break;
1181 		default:
1182 			copy_siginfo(&q->info, info);
1183 			break;
1184 		}
1185 	} else if (!is_si_special(info) &&
1186 		   sig >= SIGRTMIN && info->si_code != SI_USER) {
1187 		/*
1188 		 * Queue overflow, abort.  We may abort if the
1189 		 * signal was rt and sent by user using something
1190 		 * other than kill().
1191 		 */
1192 		result = TRACE_SIGNAL_OVERFLOW_FAIL;
1193 		ret = -EAGAIN;
1194 		goto ret;
1195 	} else {
1196 		/*
1197 		 * This is a silent loss of information.  We still
1198 		 * send the signal, but the *info bits are lost.
1199 		 */
1200 		result = TRACE_SIGNAL_LOSE_INFO;
1201 	}
1202 
1203 out_set:
1204 	signalfd_notify(t, sig);
1205 	sigaddset(&pending->signal, sig);
1206 
1207 	/* Let multiprocess signals appear after on-going forks */
1208 	if (type > PIDTYPE_TGID) {
1209 		struct multiprocess_signals *delayed;
1210 		hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1211 			sigset_t *signal = &delayed->signal;
1212 			/* Can't queue both a stop and a continue signal */
1213 			if (sig == SIGCONT)
1214 				sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1215 			else if (sig_kernel_stop(sig))
1216 				sigdelset(signal, SIGCONT);
1217 			sigaddset(signal, sig);
1218 		}
1219 	}
1220 
1221 	complete_signal(sig, t, type);
1222 ret:
1223 	trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1224 	return ret;
1225 }
1226 
has_si_pid_and_uid(struct kernel_siginfo * info)1227 static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1228 {
1229 	bool ret = false;
1230 	switch (siginfo_layout(info->si_signo, info->si_code)) {
1231 	case SIL_KILL:
1232 	case SIL_CHLD:
1233 	case SIL_RT:
1234 		ret = true;
1235 		break;
1236 	case SIL_TIMER:
1237 	case SIL_POLL:
1238 	case SIL_FAULT:
1239 	case SIL_FAULT_MCEERR:
1240 	case SIL_FAULT_BNDERR:
1241 	case SIL_FAULT_PKUERR:
1242 	case SIL_PERF_EVENT:
1243 	case SIL_SYS:
1244 		ret = false;
1245 		break;
1246 	}
1247 	return ret;
1248 }
1249 
send_signal(int sig,struct kernel_siginfo * info,struct task_struct * t,enum pid_type type)1250 static int send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1251 			enum pid_type type)
1252 {
1253 	/* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1254 	bool force = false;
1255 
1256 	if (info == SEND_SIG_NOINFO) {
1257 		/* Force if sent from an ancestor pid namespace */
1258 		force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1259 	} else if (info == SEND_SIG_PRIV) {
1260 		/* Don't ignore kernel generated signals */
1261 		force = true;
1262 	} else if (has_si_pid_and_uid(info)) {
1263 		/* SIGKILL and SIGSTOP is special or has ids */
1264 		struct user_namespace *t_user_ns;
1265 
1266 		rcu_read_lock();
1267 		t_user_ns = task_cred_xxx(t, user_ns);
1268 		if (current_user_ns() != t_user_ns) {
1269 			kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1270 			info->si_uid = from_kuid_munged(t_user_ns, uid);
1271 		}
1272 		rcu_read_unlock();
1273 
1274 		/* A kernel generated signal? */
1275 		force = (info->si_code == SI_KERNEL);
1276 
1277 		/* From an ancestor pid namespace? */
1278 		if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1279 			info->si_pid = 0;
1280 			force = true;
1281 		}
1282 	}
1283 	return __send_signal(sig, info, t, type, force);
1284 }
1285 
print_fatal_signal(int signr)1286 static void print_fatal_signal(int signr)
1287 {
1288 	struct pt_regs *regs = signal_pt_regs();
1289 	pr_info("potentially unexpected fatal signal %d.\n", signr);
1290 
1291 #if defined(__i386__) && !defined(__arch_um__)
1292 	pr_info("code at %08lx: ", regs->ip);
1293 	{
1294 		int i;
1295 		for (i = 0; i < 16; i++) {
1296 			unsigned char insn;
1297 
1298 			if (get_user(insn, (unsigned char *)(regs->ip + i)))
1299 				break;
1300 			pr_cont("%02x ", insn);
1301 		}
1302 	}
1303 	pr_cont("\n");
1304 #endif
1305 	preempt_disable();
1306 	show_regs(regs);
1307 	preempt_enable();
1308 }
1309 
setup_print_fatal_signals(char * str)1310 static int __init setup_print_fatal_signals(char *str)
1311 {
1312 	get_option (&str, &print_fatal_signals);
1313 
1314 	return 1;
1315 }
1316 
1317 __setup("print-fatal-signals=", setup_print_fatal_signals);
1318 
1319 int
__group_send_sig_info(int sig,struct kernel_siginfo * info,struct task_struct * p)1320 __group_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1321 {
1322 	return send_signal(sig, info, p, PIDTYPE_TGID);
1323 }
1324 
do_send_sig_info(int sig,struct kernel_siginfo * info,struct task_struct * p,enum pid_type type)1325 int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1326 			enum pid_type type)
1327 {
1328 	unsigned long flags;
1329 	int ret = -ESRCH;
1330 
1331 	if (lock_task_sighand(p, &flags)) {
1332 		ret = send_signal(sig, info, p, type);
1333 		unlock_task_sighand(p, &flags);
1334 	}
1335 
1336 	return ret;
1337 }
1338 
1339 /*
1340  * Force a signal that the process can't ignore: if necessary
1341  * we unblock the signal and change any SIG_IGN to SIG_DFL.
1342  *
1343  * Note: If we unblock the signal, we always reset it to SIG_DFL,
1344  * since we do not want to have a signal handler that was blocked
1345  * be invoked when user space had explicitly blocked it.
1346  *
1347  * We don't want to have recursive SIGSEGV's etc, for example,
1348  * that is why we also clear SIGNAL_UNKILLABLE.
1349  */
1350 static int
force_sig_info_to_task(struct kernel_siginfo * info,struct task_struct * t)1351 force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t)
1352 {
1353 	unsigned long int flags;
1354 	int ret, blocked, ignored;
1355 	struct k_sigaction *action;
1356 	int sig = info->si_signo;
1357 
1358 	spin_lock_irqsave(&t->sighand->siglock, flags);
1359 	action = &t->sighand->action[sig-1];
1360 	ignored = action->sa.sa_handler == SIG_IGN;
1361 	blocked = sigismember(&t->blocked, sig);
1362 	if (blocked || ignored) {
1363 		action->sa.sa_handler = SIG_DFL;
1364 		if (blocked) {
1365 			sigdelset(&t->blocked, sig);
1366 			recalc_sigpending_and_wake(t);
1367 		}
1368 	}
1369 	/*
1370 	 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1371 	 * debugging to leave init killable.
1372 	 */
1373 	if (action->sa.sa_handler == SIG_DFL && !t->ptrace)
1374 		t->signal->flags &= ~SIGNAL_UNKILLABLE;
1375 	ret = send_signal(sig, info, t, PIDTYPE_PID);
1376 	spin_unlock_irqrestore(&t->sighand->siglock, flags);
1377 
1378 	return ret;
1379 }
1380 
force_sig_info(struct kernel_siginfo * info)1381 int force_sig_info(struct kernel_siginfo *info)
1382 {
1383 	return force_sig_info_to_task(info, current);
1384 }
1385 
1386 /*
1387  * Nuke all other threads in the group.
1388  */
zap_other_threads(struct task_struct * p)1389 int zap_other_threads(struct task_struct *p)
1390 {
1391 	struct task_struct *t = p;
1392 	int count = 0;
1393 
1394 	p->signal->group_stop_count = 0;
1395 
1396 	while_each_thread(p, t) {
1397 		task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1398 		count++;
1399 
1400 		/* Don't bother with already dead threads */
1401 		if (t->exit_state)
1402 			continue;
1403 		sigaddset(&t->pending.signal, SIGKILL);
1404 		signal_wake_up(t, 1);
1405 	}
1406 
1407 	return count;
1408 }
1409 
__lock_task_sighand(struct task_struct * tsk,unsigned long * flags)1410 struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1411 					   unsigned long *flags)
1412 {
1413 	struct sighand_struct *sighand;
1414 
1415 	rcu_read_lock();
1416 	for (;;) {
1417 		sighand = rcu_dereference(tsk->sighand);
1418 		if (unlikely(sighand == NULL))
1419 			break;
1420 
1421 		/*
1422 		 * This sighand can be already freed and even reused, but
1423 		 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1424 		 * initializes ->siglock: this slab can't go away, it has
1425 		 * the same object type, ->siglock can't be reinitialized.
1426 		 *
1427 		 * We need to ensure that tsk->sighand is still the same
1428 		 * after we take the lock, we can race with de_thread() or
1429 		 * __exit_signal(). In the latter case the next iteration
1430 		 * must see ->sighand == NULL.
1431 		 */
1432 		spin_lock_irqsave(&sighand->siglock, *flags);
1433 		if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1434 			break;
1435 		spin_unlock_irqrestore(&sighand->siglock, *flags);
1436 	}
1437 	rcu_read_unlock();
1438 
1439 	return sighand;
1440 }
1441 
1442 /*
1443  * send signal info to all the members of a group
1444  */
group_send_sig_info(int sig,struct kernel_siginfo * info,struct task_struct * p,enum pid_type type)1445 int group_send_sig_info(int sig, struct kernel_siginfo *info,
1446 			struct task_struct *p, enum pid_type type)
1447 {
1448 	int ret;
1449 
1450 	rcu_read_lock();
1451 	ret = check_kill_permission(sig, info, p);
1452 	rcu_read_unlock();
1453 
1454 	if (!ret && sig)
1455 		ret = do_send_sig_info(sig, info, p, type);
1456 
1457 	return ret;
1458 }
1459 
1460 /*
1461  * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1462  * control characters do (^C, ^Z etc)
1463  * - the caller must hold at least a readlock on tasklist_lock
1464  */
__kill_pgrp_info(int sig,struct kernel_siginfo * info,struct pid * pgrp)1465 int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1466 {
1467 	struct task_struct *p = NULL;
1468 	int retval, success;
1469 
1470 	success = 0;
1471 	retval = -ESRCH;
1472 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1473 		int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1474 		success |= !err;
1475 		retval = err;
1476 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1477 	return success ? 0 : retval;
1478 }
1479 
kill_pid_info(int sig,struct kernel_siginfo * info,struct pid * pid)1480 int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1481 {
1482 	int error = -ESRCH;
1483 	struct task_struct *p;
1484 
1485 	for (;;) {
1486 		rcu_read_lock();
1487 		p = pid_task(pid, PIDTYPE_PID);
1488 		if (p)
1489 			error = group_send_sig_info(sig, info, p, PIDTYPE_TGID);
1490 		rcu_read_unlock();
1491 		if (likely(!p || error != -ESRCH))
1492 			return error;
1493 
1494 		/*
1495 		 * The task was unhashed in between, try again.  If it
1496 		 * is dead, pid_task() will return NULL, if we race with
1497 		 * de_thread() it will find the new leader.
1498 		 */
1499 	}
1500 }
1501 
kill_proc_info(int sig,struct kernel_siginfo * info,pid_t pid)1502 static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1503 {
1504 	int error;
1505 	rcu_read_lock();
1506 	error = kill_pid_info(sig, info, find_vpid(pid));
1507 	rcu_read_unlock();
1508 	return error;
1509 }
1510 
kill_as_cred_perm(const struct cred * cred,struct task_struct * target)1511 static inline bool kill_as_cred_perm(const struct cred *cred,
1512 				     struct task_struct *target)
1513 {
1514 	const struct cred *pcred = __task_cred(target);
1515 
1516 	return uid_eq(cred->euid, pcred->suid) ||
1517 	       uid_eq(cred->euid, pcred->uid) ||
1518 	       uid_eq(cred->uid, pcred->suid) ||
1519 	       uid_eq(cred->uid, pcred->uid);
1520 }
1521 
1522 /*
1523  * The usb asyncio usage of siginfo is wrong.  The glibc support
1524  * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1525  * AKA after the generic fields:
1526  *	kernel_pid_t	si_pid;
1527  *	kernel_uid32_t	si_uid;
1528  *	sigval_t	si_value;
1529  *
1530  * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1531  * after the generic fields is:
1532  *	void __user 	*si_addr;
1533  *
1534  * This is a practical problem when there is a 64bit big endian kernel
1535  * and a 32bit userspace.  As the 32bit address will encoded in the low
1536  * 32bits of the pointer.  Those low 32bits will be stored at higher
1537  * address than appear in a 32 bit pointer.  So userspace will not
1538  * see the address it was expecting for it's completions.
1539  *
1540  * There is nothing in the encoding that can allow
1541  * copy_siginfo_to_user32 to detect this confusion of formats, so
1542  * handle this by requiring the caller of kill_pid_usb_asyncio to
1543  * notice when this situration takes place and to store the 32bit
1544  * pointer in sival_int, instead of sival_addr of the sigval_t addr
1545  * parameter.
1546  */
kill_pid_usb_asyncio(int sig,int errno,sigval_t addr,struct pid * pid,const struct cred * cred)1547 int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1548 			 struct pid *pid, const struct cred *cred)
1549 {
1550 	struct kernel_siginfo info;
1551 	struct task_struct *p;
1552 	unsigned long flags;
1553 	int ret = -EINVAL;
1554 
1555 	if (!valid_signal(sig))
1556 		return ret;
1557 
1558 	clear_siginfo(&info);
1559 	info.si_signo = sig;
1560 	info.si_errno = errno;
1561 	info.si_code = SI_ASYNCIO;
1562 	*((sigval_t *)&info.si_pid) = addr;
1563 
1564 	rcu_read_lock();
1565 	p = pid_task(pid, PIDTYPE_PID);
1566 	if (!p) {
1567 		ret = -ESRCH;
1568 		goto out_unlock;
1569 	}
1570 	if (!kill_as_cred_perm(cred, p)) {
1571 		ret = -EPERM;
1572 		goto out_unlock;
1573 	}
1574 	ret = security_task_kill(p, &info, sig, cred);
1575 	if (ret)
1576 		goto out_unlock;
1577 
1578 	if (sig) {
1579 		if (lock_task_sighand(p, &flags)) {
1580 			ret = __send_signal(sig, &info, p, PIDTYPE_TGID, false);
1581 			unlock_task_sighand(p, &flags);
1582 		} else
1583 			ret = -ESRCH;
1584 	}
1585 out_unlock:
1586 	rcu_read_unlock();
1587 	return ret;
1588 }
1589 EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1590 
1591 /*
1592  * kill_something_info() interprets pid in interesting ways just like kill(2).
1593  *
1594  * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1595  * is probably wrong.  Should make it like BSD or SYSV.
1596  */
1597 
kill_something_info(int sig,struct kernel_siginfo * info,pid_t pid)1598 static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1599 {
1600 	int ret;
1601 
1602 	if (pid > 0)
1603 		return kill_proc_info(sig, info, pid);
1604 
1605 	/* -INT_MIN is undefined.  Exclude this case to avoid a UBSAN warning */
1606 	if (pid == INT_MIN)
1607 		return -ESRCH;
1608 
1609 	read_lock(&tasklist_lock);
1610 	if (pid != -1) {
1611 		ret = __kill_pgrp_info(sig, info,
1612 				pid ? find_vpid(-pid) : task_pgrp(current));
1613 	} else {
1614 		int retval = 0, count = 0;
1615 		struct task_struct * p;
1616 
1617 		for_each_process(p) {
1618 			if (task_pid_vnr(p) > 1 &&
1619 					!same_thread_group(p, current)) {
1620 				int err = group_send_sig_info(sig, info, p,
1621 							      PIDTYPE_MAX);
1622 				++count;
1623 				if (err != -EPERM)
1624 					retval = err;
1625 			}
1626 		}
1627 		ret = count ? retval : -ESRCH;
1628 	}
1629 	read_unlock(&tasklist_lock);
1630 
1631 	return ret;
1632 }
1633 
1634 /*
1635  * These are for backward compatibility with the rest of the kernel source.
1636  */
1637 
send_sig_info(int sig,struct kernel_siginfo * info,struct task_struct * p)1638 int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1639 {
1640 	/*
1641 	 * Make sure legacy kernel users don't send in bad values
1642 	 * (normal paths check this in check_kill_permission).
1643 	 */
1644 	if (!valid_signal(sig))
1645 		return -EINVAL;
1646 
1647 	return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1648 }
1649 EXPORT_SYMBOL(send_sig_info);
1650 
1651 #define __si_special(priv) \
1652 	((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1653 
1654 int
send_sig(int sig,struct task_struct * p,int priv)1655 send_sig(int sig, struct task_struct *p, int priv)
1656 {
1657 	return send_sig_info(sig, __si_special(priv), p);
1658 }
1659 EXPORT_SYMBOL(send_sig);
1660 
force_sig(int sig)1661 void force_sig(int sig)
1662 {
1663 	struct kernel_siginfo info;
1664 
1665 	clear_siginfo(&info);
1666 	info.si_signo = sig;
1667 	info.si_errno = 0;
1668 	info.si_code = SI_KERNEL;
1669 	info.si_pid = 0;
1670 	info.si_uid = 0;
1671 	force_sig_info(&info);
1672 }
1673 EXPORT_SYMBOL(force_sig);
1674 
1675 /*
1676  * When things go south during signal handling, we
1677  * will force a SIGSEGV. And if the signal that caused
1678  * the problem was already a SIGSEGV, we'll want to
1679  * make sure we don't even try to deliver the signal..
1680  */
force_sigsegv(int sig)1681 void force_sigsegv(int sig)
1682 {
1683 	struct task_struct *p = current;
1684 
1685 	if (sig == SIGSEGV) {
1686 		unsigned long flags;
1687 		spin_lock_irqsave(&p->sighand->siglock, flags);
1688 		p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1689 		spin_unlock_irqrestore(&p->sighand->siglock, flags);
1690 	}
1691 	force_sig(SIGSEGV);
1692 }
1693 
force_sig_fault_to_task(int sig,int code,void __user * addr ___ARCH_SI_TRAPNO (int trapno)___ARCH_SI_IA64 (int imm,unsigned int flags,unsigned long isr),struct task_struct * t)1694 int force_sig_fault_to_task(int sig, int code, void __user *addr
1695 	___ARCH_SI_TRAPNO(int trapno)
1696 	___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1697 	, struct task_struct *t)
1698 {
1699 	struct kernel_siginfo info;
1700 
1701 	clear_siginfo(&info);
1702 	info.si_signo = sig;
1703 	info.si_errno = 0;
1704 	info.si_code  = code;
1705 	info.si_addr  = addr;
1706 #ifdef __ARCH_SI_TRAPNO
1707 	info.si_trapno = trapno;
1708 #endif
1709 #ifdef __ia64__
1710 	info.si_imm = imm;
1711 	info.si_flags = flags;
1712 	info.si_isr = isr;
1713 #endif
1714 	return force_sig_info_to_task(&info, t);
1715 }
1716 
force_sig_fault(int sig,int code,void __user * addr ___ARCH_SI_TRAPNO (int trapno)___ARCH_SI_IA64 (int imm,unsigned int flags,unsigned long isr))1717 int force_sig_fault(int sig, int code, void __user *addr
1718 	___ARCH_SI_TRAPNO(int trapno)
1719 	___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr))
1720 {
1721 	return force_sig_fault_to_task(sig, code, addr
1722 				       ___ARCH_SI_TRAPNO(trapno)
1723 				       ___ARCH_SI_IA64(imm, flags, isr), current);
1724 }
1725 
send_sig_fault(int sig,int code,void __user * addr ___ARCH_SI_TRAPNO (int trapno)___ARCH_SI_IA64 (int imm,unsigned int flags,unsigned long isr),struct task_struct * t)1726 int send_sig_fault(int sig, int code, void __user *addr
1727 	___ARCH_SI_TRAPNO(int trapno)
1728 	___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1729 	, struct task_struct *t)
1730 {
1731 	struct kernel_siginfo info;
1732 
1733 	clear_siginfo(&info);
1734 	info.si_signo = sig;
1735 	info.si_errno = 0;
1736 	info.si_code  = code;
1737 	info.si_addr  = addr;
1738 #ifdef __ARCH_SI_TRAPNO
1739 	info.si_trapno = trapno;
1740 #endif
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 
force_sig_mceerr(int code,void __user * addr,short lsb)1749 int 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 
send_sig_mceerr(int code,void __user * addr,short lsb,struct task_struct * t)1763 int 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 }
1776 EXPORT_SYMBOL(send_sig_mceerr);
1777 
force_sig_bnderr(void __user * addr,void __user * lower,void __user * upper)1778 int 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
force_sig_pkuerr(void __user * addr,u32 pkey)1793 int 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 
1807 /* For the crazy architectures that include trap information in
1808  * the errno field, instead of an actual errno value.
1809  */
force_sig_ptrace_errno_trap(int errno,void __user * addr)1810 int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1811 {
1812 	struct kernel_siginfo info;
1813 
1814 	clear_siginfo(&info);
1815 	info.si_signo = SIGTRAP;
1816 	info.si_errno = errno;
1817 	info.si_code  = TRAP_HWBKPT;
1818 	info.si_addr  = addr;
1819 	return force_sig_info(&info);
1820 }
1821 
kill_pgrp(struct pid * pid,int sig,int priv)1822 int kill_pgrp(struct pid *pid, int sig, int priv)
1823 {
1824 	int ret;
1825 
1826 	read_lock(&tasklist_lock);
1827 	ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1828 	read_unlock(&tasklist_lock);
1829 
1830 	return ret;
1831 }
1832 EXPORT_SYMBOL(kill_pgrp);
1833 
kill_pid(struct pid * pid,int sig,int priv)1834 int kill_pid(struct pid *pid, int sig, int priv)
1835 {
1836 	return kill_pid_info(sig, __si_special(priv), pid);
1837 }
1838 EXPORT_SYMBOL(kill_pid);
1839 
1840 /*
1841  * These functions support sending signals using preallocated sigqueue
1842  * structures.  This is needed "because realtime applications cannot
1843  * afford to lose notifications of asynchronous events, like timer
1844  * expirations or I/O completions".  In the case of POSIX Timers
1845  * we allocate the sigqueue structure from the timer_create.  If this
1846  * allocation fails we are able to report the failure to the application
1847  * with an EAGAIN error.
1848  */
sigqueue_alloc(void)1849 struct sigqueue *sigqueue_alloc(void)
1850 {
1851 	return __sigqueue_alloc(-1, current, GFP_KERNEL, 0, SIGQUEUE_PREALLOC);
1852 }
1853 
sigqueue_free(struct sigqueue * q)1854 void sigqueue_free(struct sigqueue *q)
1855 {
1856 	unsigned long flags;
1857 	spinlock_t *lock = &current->sighand->siglock;
1858 
1859 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1860 	/*
1861 	 * We must hold ->siglock while testing q->list
1862 	 * to serialize with collect_signal() or with
1863 	 * __exit_signal()->flush_sigqueue().
1864 	 */
1865 	spin_lock_irqsave(lock, flags);
1866 	q->flags &= ~SIGQUEUE_PREALLOC;
1867 	/*
1868 	 * If it is queued it will be freed when dequeued,
1869 	 * like the "regular" sigqueue.
1870 	 */
1871 	if (!list_empty(&q->list))
1872 		q = NULL;
1873 	spin_unlock_irqrestore(lock, flags);
1874 
1875 	if (q)
1876 		__sigqueue_free(q);
1877 }
1878 
send_sigqueue(struct sigqueue * q,struct pid * pid,enum pid_type type)1879 int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1880 {
1881 	int sig = q->info.si_signo;
1882 	struct sigpending *pending;
1883 	struct task_struct *t;
1884 	unsigned long flags;
1885 	int ret, result;
1886 
1887 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1888 
1889 	ret = -1;
1890 	rcu_read_lock();
1891 	t = pid_task(pid, type);
1892 	if (!t || !likely(lock_task_sighand(t, &flags)))
1893 		goto ret;
1894 
1895 	ret = 1; /* the signal is ignored */
1896 	result = TRACE_SIGNAL_IGNORED;
1897 	if (!prepare_signal(sig, t, false))
1898 		goto out;
1899 
1900 	ret = 0;
1901 	if (unlikely(!list_empty(&q->list))) {
1902 		/*
1903 		 * If an SI_TIMER entry is already queue just increment
1904 		 * the overrun count.
1905 		 */
1906 		BUG_ON(q->info.si_code != SI_TIMER);
1907 		q->info.si_overrun++;
1908 		result = TRACE_SIGNAL_ALREADY_PENDING;
1909 		goto out;
1910 	}
1911 	q->info.si_overrun = 0;
1912 
1913 	signalfd_notify(t, sig);
1914 	pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1915 	list_add_tail(&q->list, &pending->list);
1916 	sigaddset(&pending->signal, sig);
1917 	complete_signal(sig, t, type);
1918 	result = TRACE_SIGNAL_DELIVERED;
1919 out:
1920 	trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
1921 	unlock_task_sighand(t, &flags);
1922 ret:
1923 	rcu_read_unlock();
1924 	return ret;
1925 }
1926 
do_notify_pidfd(struct task_struct * task)1927 static void do_notify_pidfd(struct task_struct *task)
1928 {
1929 	struct pid *pid;
1930 
1931 	WARN_ON(task->exit_state == 0);
1932 	pid = task_pid(task);
1933 	wake_up_all(&pid->wait_pidfd);
1934 }
1935 
1936 /*
1937  * Let a parent know about the death of a child.
1938  * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1939  *
1940  * Returns true if our parent ignored us and so we've switched to
1941  * self-reaping.
1942  */
do_notify_parent(struct task_struct * tsk,int sig)1943 bool do_notify_parent(struct task_struct *tsk, int sig)
1944 {
1945 	struct kernel_siginfo info;
1946 	unsigned long flags;
1947 	struct sighand_struct *psig;
1948 	bool autoreap = false;
1949 	u64 utime, stime;
1950 
1951 	BUG_ON(sig == -1);
1952 
1953  	/* do_notify_parent_cldstop should have been called instead.  */
1954  	BUG_ON(task_is_stopped_or_traced(tsk));
1955 
1956 	BUG_ON(!tsk->ptrace &&
1957 	       (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1958 
1959 	/* Wake up all pidfd waiters */
1960 	do_notify_pidfd(tsk);
1961 
1962 	if (sig != SIGCHLD) {
1963 		/*
1964 		 * This is only possible if parent == real_parent.
1965 		 * Check if it has changed security domain.
1966 		 */
1967 		if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
1968 			sig = SIGCHLD;
1969 	}
1970 
1971 	clear_siginfo(&info);
1972 	info.si_signo = sig;
1973 	info.si_errno = 0;
1974 	/*
1975 	 * We are under tasklist_lock here so our parent is tied to
1976 	 * us and cannot change.
1977 	 *
1978 	 * task_active_pid_ns will always return the same pid namespace
1979 	 * until a task passes through release_task.
1980 	 *
1981 	 * write_lock() currently calls preempt_disable() which is the
1982 	 * same as rcu_read_lock(), but according to Oleg, this is not
1983 	 * correct to rely on this
1984 	 */
1985 	rcu_read_lock();
1986 	info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
1987 	info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
1988 				       task_uid(tsk));
1989 	rcu_read_unlock();
1990 
1991 	task_cputime(tsk, &utime, &stime);
1992 	info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
1993 	info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
1994 
1995 	info.si_status = tsk->exit_code & 0x7f;
1996 	if (tsk->exit_code & 0x80)
1997 		info.si_code = CLD_DUMPED;
1998 	else if (tsk->exit_code & 0x7f)
1999 		info.si_code = CLD_KILLED;
2000 	else {
2001 		info.si_code = CLD_EXITED;
2002 		info.si_status = tsk->exit_code >> 8;
2003 	}
2004 
2005 	psig = tsk->parent->sighand;
2006 	spin_lock_irqsave(&psig->siglock, flags);
2007 	if (!tsk->ptrace && sig == SIGCHLD &&
2008 	    (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
2009 	     (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
2010 		/*
2011 		 * We are exiting and our parent doesn't care.  POSIX.1
2012 		 * defines special semantics for setting SIGCHLD to SIG_IGN
2013 		 * or setting the SA_NOCLDWAIT flag: we should be reaped
2014 		 * automatically and not left for our parent's wait4 call.
2015 		 * Rather than having the parent do it as a magic kind of
2016 		 * signal handler, we just set this to tell do_exit that we
2017 		 * can be cleaned up without becoming a zombie.  Note that
2018 		 * we still call __wake_up_parent in this case, because a
2019 		 * blocked sys_wait4 might now return -ECHILD.
2020 		 *
2021 		 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
2022 		 * is implementation-defined: we do (if you don't want
2023 		 * it, just use SIG_IGN instead).
2024 		 */
2025 		autoreap = true;
2026 		if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2027 			sig = 0;
2028 	}
2029 	/*
2030 	 * Send with __send_signal as si_pid and si_uid are in the
2031 	 * parent's namespaces.
2032 	 */
2033 	if (valid_signal(sig) && sig)
2034 		__send_signal(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2035 	__wake_up_parent(tsk, tsk->parent);
2036 	spin_unlock_irqrestore(&psig->siglock, flags);
2037 
2038 	return autoreap;
2039 }
2040 
2041 /**
2042  * do_notify_parent_cldstop - notify parent of stopped/continued state change
2043  * @tsk: task reporting the state change
2044  * @for_ptracer: the notification is for ptracer
2045  * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2046  *
2047  * Notify @tsk's parent that the stopped/continued state has changed.  If
2048  * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2049  * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2050  *
2051  * CONTEXT:
2052  * Must be called with tasklist_lock at least read locked.
2053  */
do_notify_parent_cldstop(struct task_struct * tsk,bool for_ptracer,int why)2054 static void do_notify_parent_cldstop(struct task_struct *tsk,
2055 				     bool for_ptracer, int why)
2056 {
2057 	struct kernel_siginfo info;
2058 	unsigned long flags;
2059 	struct task_struct *parent;
2060 	struct sighand_struct *sighand;
2061 	u64 utime, stime;
2062 
2063 	if (for_ptracer) {
2064 		parent = tsk->parent;
2065 	} else {
2066 		tsk = tsk->group_leader;
2067 		parent = tsk->real_parent;
2068 	}
2069 
2070 	clear_siginfo(&info);
2071 	info.si_signo = SIGCHLD;
2072 	info.si_errno = 0;
2073 	/*
2074 	 * see comment in do_notify_parent() about the following 4 lines
2075 	 */
2076 	rcu_read_lock();
2077 	info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2078 	info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2079 	rcu_read_unlock();
2080 
2081 	task_cputime(tsk, &utime, &stime);
2082 	info.si_utime = nsec_to_clock_t(utime);
2083 	info.si_stime = nsec_to_clock_t(stime);
2084 
2085  	info.si_code = why;
2086  	switch (why) {
2087  	case CLD_CONTINUED:
2088  		info.si_status = SIGCONT;
2089  		break;
2090  	case CLD_STOPPED:
2091  		info.si_status = tsk->signal->group_exit_code & 0x7f;
2092  		break;
2093  	case CLD_TRAPPED:
2094  		info.si_status = tsk->exit_code & 0x7f;
2095  		break;
2096  	default:
2097  		BUG();
2098  	}
2099 
2100 	sighand = parent->sighand;
2101 	spin_lock_irqsave(&sighand->siglock, flags);
2102 	if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2103 	    !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2104 		__group_send_sig_info(SIGCHLD, &info, parent);
2105 	/*
2106 	 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2107 	 */
2108 	__wake_up_parent(tsk, parent);
2109 	spin_unlock_irqrestore(&sighand->siglock, flags);
2110 }
2111 
may_ptrace_stop(void)2112 static inline bool may_ptrace_stop(void)
2113 {
2114 	if (!likely(current->ptrace))
2115 		return false;
2116 	/*
2117 	 * Are we in the middle of do_coredump?
2118 	 * If so and our tracer is also part of the coredump stopping
2119 	 * is a deadlock situation, and pointless because our tracer
2120 	 * is dead so don't allow us to stop.
2121 	 * If SIGKILL was already sent before the caller unlocked
2122 	 * ->siglock we must see ->core_state != NULL. Otherwise it
2123 	 * is safe to enter schedule().
2124 	 *
2125 	 * This is almost outdated, a task with the pending SIGKILL can't
2126 	 * block in TASK_TRACED. But PTRACE_EVENT_EXIT can be reported
2127 	 * after SIGKILL was already dequeued.
2128 	 */
2129 	if (unlikely(current->mm->core_state) &&
2130 	    unlikely(current->mm == current->parent->mm))
2131 		return false;
2132 
2133 	return true;
2134 }
2135 
2136 /*
2137  * Return non-zero if there is a SIGKILL that should be waking us up.
2138  * Called with the siglock held.
2139  */
sigkill_pending(struct task_struct * tsk)2140 static bool sigkill_pending(struct task_struct *tsk)
2141 {
2142 	return sigismember(&tsk->pending.signal, SIGKILL) ||
2143 	       sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
2144 }
2145 
2146 /*
2147  * This must be called with current->sighand->siglock held.
2148  *
2149  * This should be the path for all ptrace stops.
2150  * We always set current->last_siginfo while stopped here.
2151  * That makes it a way to test a stopped process for
2152  * being ptrace-stopped vs being job-control-stopped.
2153  *
2154  * If we actually decide not to stop at all because the tracer
2155  * is gone, we keep current->exit_code unless clear_code.
2156  */
ptrace_stop(int exit_code,int why,int clear_code,kernel_siginfo_t * info)2157 static void ptrace_stop(int exit_code, int why, int clear_code, kernel_siginfo_t *info)
2158 	__releases(&current->sighand->siglock)
2159 	__acquires(&current->sighand->siglock)
2160 {
2161 	bool gstop_done = false;
2162 
2163 	if (arch_ptrace_stop_needed(exit_code, info)) {
2164 		/*
2165 		 * The arch code has something special to do before a
2166 		 * ptrace stop.  This is allowed to block, e.g. for faults
2167 		 * on user stack pages.  We can't keep the siglock while
2168 		 * calling arch_ptrace_stop, so we must release it now.
2169 		 * To preserve proper semantics, we must do this before
2170 		 * any signal bookkeeping like checking group_stop_count.
2171 		 * Meanwhile, a SIGKILL could come in before we retake the
2172 		 * siglock.  That must prevent us from sleeping in TASK_TRACED.
2173 		 * So after regaining the lock, we must check for SIGKILL.
2174 		 */
2175 		spin_unlock_irq(&current->sighand->siglock);
2176 		arch_ptrace_stop(exit_code, info);
2177 		spin_lock_irq(&current->sighand->siglock);
2178 		if (sigkill_pending(current))
2179 			return;
2180 	}
2181 
2182 	set_special_state(TASK_TRACED);
2183 
2184 	/*
2185 	 * We're committing to trapping.  TRACED should be visible before
2186 	 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2187 	 * Also, transition to TRACED and updates to ->jobctl should be
2188 	 * atomic with respect to siglock and should be done after the arch
2189 	 * hook as siglock is released and regrabbed across it.
2190 	 *
2191 	 *     TRACER				    TRACEE
2192 	 *
2193 	 *     ptrace_attach()
2194 	 * [L]   wait_on_bit(JOBCTL_TRAPPING)	[S] set_special_state(TRACED)
2195 	 *     do_wait()
2196 	 *       set_current_state()                smp_wmb();
2197 	 *       ptrace_do_wait()
2198 	 *         wait_task_stopped()
2199 	 *           task_stopped_code()
2200 	 * [L]         task_is_traced()		[S] task_clear_jobctl_trapping();
2201 	 */
2202 	smp_wmb();
2203 
2204 	current->last_siginfo = info;
2205 	current->exit_code = exit_code;
2206 
2207 	/*
2208 	 * If @why is CLD_STOPPED, we're trapping to participate in a group
2209 	 * stop.  Do the bookkeeping.  Note that if SIGCONT was delievered
2210 	 * across siglock relocks since INTERRUPT was scheduled, PENDING
2211 	 * could be clear now.  We act as if SIGCONT is received after
2212 	 * TASK_TRACED is entered - ignore it.
2213 	 */
2214 	if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2215 		gstop_done = task_participate_group_stop(current);
2216 
2217 	/* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2218 	task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2219 	if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2220 		task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2221 
2222 	/* entering a trap, clear TRAPPING */
2223 	task_clear_jobctl_trapping(current);
2224 
2225 	spin_unlock_irq(&current->sighand->siglock);
2226 	read_lock(&tasklist_lock);
2227 	if (may_ptrace_stop()) {
2228 		/*
2229 		 * Notify parents of the stop.
2230 		 *
2231 		 * While ptraced, there are two parents - the ptracer and
2232 		 * the real_parent of the group_leader.  The ptracer should
2233 		 * know about every stop while the real parent is only
2234 		 * interested in the completion of group stop.  The states
2235 		 * for the two don't interact with each other.  Notify
2236 		 * separately unless they're gonna be duplicates.
2237 		 */
2238 		do_notify_parent_cldstop(current, true, why);
2239 		if (gstop_done && ptrace_reparented(current))
2240 			do_notify_parent_cldstop(current, false, why);
2241 
2242 		/*
2243 		 * Don't want to allow preemption here, because
2244 		 * sys_ptrace() needs this task to be inactive.
2245 		 *
2246 		 * XXX: implement read_unlock_no_resched().
2247 		 */
2248 		preempt_disable();
2249 		read_unlock(&tasklist_lock);
2250 		cgroup_enter_frozen();
2251 		preempt_enable_no_resched();
2252 		freezable_schedule();
2253 		cgroup_leave_frozen(true);
2254 	} else {
2255 		/*
2256 		 * By the time we got the lock, our tracer went away.
2257 		 * Don't drop the lock yet, another tracer may come.
2258 		 *
2259 		 * If @gstop_done, the ptracer went away between group stop
2260 		 * completion and here.  During detach, it would have set
2261 		 * JOBCTL_STOP_PENDING on us and we'll re-enter
2262 		 * TASK_STOPPED in do_signal_stop() on return, so notifying
2263 		 * the real parent of the group stop completion is enough.
2264 		 */
2265 		if (gstop_done)
2266 			do_notify_parent_cldstop(current, false, why);
2267 
2268 		/* tasklist protects us from ptrace_freeze_traced() */
2269 		__set_current_state(TASK_RUNNING);
2270 		if (clear_code)
2271 			current->exit_code = 0;
2272 		read_unlock(&tasklist_lock);
2273 	}
2274 
2275 	/*
2276 	 * We are back.  Now reacquire the siglock before touching
2277 	 * last_siginfo, so that we are sure to have synchronized with
2278 	 * any signal-sending on another CPU that wants to examine it.
2279 	 */
2280 	spin_lock_irq(&current->sighand->siglock);
2281 	current->last_siginfo = NULL;
2282 
2283 	/* LISTENING can be set only during STOP traps, clear it */
2284 	current->jobctl &= ~JOBCTL_LISTENING;
2285 
2286 	/*
2287 	 * Queued signals ignored us while we were stopped for tracing.
2288 	 * So check for any that we should take before resuming user mode.
2289 	 * This sets TIF_SIGPENDING, but never clears it.
2290 	 */
2291 	recalc_sigpending_tsk(current);
2292 }
2293 
ptrace_do_notify(int signr,int exit_code,int why)2294 static void ptrace_do_notify(int signr, int exit_code, int why)
2295 {
2296 	kernel_siginfo_t info;
2297 
2298 	clear_siginfo(&info);
2299 	info.si_signo = signr;
2300 	info.si_code = exit_code;
2301 	info.si_pid = task_pid_vnr(current);
2302 	info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2303 
2304 	/* Let the debugger run.  */
2305 	ptrace_stop(exit_code, why, 1, &info);
2306 }
2307 
ptrace_notify(int exit_code)2308 void ptrace_notify(int exit_code)
2309 {
2310 	BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2311 	if (unlikely(current->task_works))
2312 		task_work_run();
2313 
2314 	spin_lock_irq(&current->sighand->siglock);
2315 	ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
2316 	spin_unlock_irq(&current->sighand->siglock);
2317 }
2318 
2319 /**
2320  * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2321  * @signr: signr causing group stop if initiating
2322  *
2323  * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2324  * and participate in it.  If already set, participate in the existing
2325  * group stop.  If participated in a group stop (and thus slept), %true is
2326  * returned with siglock released.
2327  *
2328  * If ptraced, this function doesn't handle stop itself.  Instead,
2329  * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2330  * untouched.  The caller must ensure that INTERRUPT trap handling takes
2331  * places afterwards.
2332  *
2333  * CONTEXT:
2334  * Must be called with @current->sighand->siglock held, which is released
2335  * on %true return.
2336  *
2337  * RETURNS:
2338  * %false if group stop is already cancelled or ptrace trap is scheduled.
2339  * %true if participated in group stop.
2340  */
do_signal_stop(int signr)2341 static bool do_signal_stop(int signr)
2342 	__releases(&current->sighand->siglock)
2343 {
2344 	struct signal_struct *sig = current->signal;
2345 
2346 	if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2347 		unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2348 		struct task_struct *t;
2349 
2350 		/* signr will be recorded in task->jobctl for retries */
2351 		WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2352 
2353 		if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2354 		    unlikely(signal_group_exit(sig)))
2355 			return false;
2356 		/*
2357 		 * There is no group stop already in progress.  We must
2358 		 * initiate one now.
2359 		 *
2360 		 * While ptraced, a task may be resumed while group stop is
2361 		 * still in effect and then receive a stop signal and
2362 		 * initiate another group stop.  This deviates from the
2363 		 * usual behavior as two consecutive stop signals can't
2364 		 * cause two group stops when !ptraced.  That is why we
2365 		 * also check !task_is_stopped(t) below.
2366 		 *
2367 		 * The condition can be distinguished by testing whether
2368 		 * SIGNAL_STOP_STOPPED is already set.  Don't generate
2369 		 * group_exit_code in such case.
2370 		 *
2371 		 * This is not necessary for SIGNAL_STOP_CONTINUED because
2372 		 * an intervening stop signal is required to cause two
2373 		 * continued events regardless of ptrace.
2374 		 */
2375 		if (!(sig->flags & SIGNAL_STOP_STOPPED))
2376 			sig->group_exit_code = signr;
2377 
2378 		sig->group_stop_count = 0;
2379 
2380 		if (task_set_jobctl_pending(current, signr | gstop))
2381 			sig->group_stop_count++;
2382 
2383 		t = current;
2384 		while_each_thread(current, t) {
2385 			/*
2386 			 * Setting state to TASK_STOPPED for a group
2387 			 * stop is always done with the siglock held,
2388 			 * so this check has no races.
2389 			 */
2390 			if (!task_is_stopped(t) &&
2391 			    task_set_jobctl_pending(t, signr | gstop)) {
2392 				sig->group_stop_count++;
2393 				if (likely(!(t->ptrace & PT_SEIZED)))
2394 					signal_wake_up(t, 0);
2395 				else
2396 					ptrace_trap_notify(t);
2397 			}
2398 		}
2399 	}
2400 
2401 	if (likely(!current->ptrace)) {
2402 		int notify = 0;
2403 
2404 		/*
2405 		 * If there are no other threads in the group, or if there
2406 		 * is a group stop in progress and we are the last to stop,
2407 		 * report to the parent.
2408 		 */
2409 		if (task_participate_group_stop(current))
2410 			notify = CLD_STOPPED;
2411 
2412 		set_special_state(TASK_STOPPED);
2413 		spin_unlock_irq(&current->sighand->siglock);
2414 
2415 		/*
2416 		 * Notify the parent of the group stop completion.  Because
2417 		 * we're not holding either the siglock or tasklist_lock
2418 		 * here, ptracer may attach inbetween; however, this is for
2419 		 * group stop and should always be delivered to the real
2420 		 * parent of the group leader.  The new ptracer will get
2421 		 * its notification when this task transitions into
2422 		 * TASK_TRACED.
2423 		 */
2424 		if (notify) {
2425 			read_lock(&tasklist_lock);
2426 			do_notify_parent_cldstop(current, false, notify);
2427 			read_unlock(&tasklist_lock);
2428 		}
2429 
2430 		/* Now we don't run again until woken by SIGCONT or SIGKILL */
2431 		cgroup_enter_frozen();
2432 		freezable_schedule();
2433 		return true;
2434 	} else {
2435 		/*
2436 		 * While ptraced, group stop is handled by STOP trap.
2437 		 * Schedule it and let the caller deal with it.
2438 		 */
2439 		task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2440 		return false;
2441 	}
2442 }
2443 
2444 /**
2445  * do_jobctl_trap - take care of ptrace jobctl traps
2446  *
2447  * When PT_SEIZED, it's used for both group stop and explicit
2448  * SEIZE/INTERRUPT traps.  Both generate PTRACE_EVENT_STOP trap with
2449  * accompanying siginfo.  If stopped, lower eight bits of exit_code contain
2450  * the stop signal; otherwise, %SIGTRAP.
2451  *
2452  * When !PT_SEIZED, it's used only for group stop trap with stop signal
2453  * number as exit_code and no siginfo.
2454  *
2455  * CONTEXT:
2456  * Must be called with @current->sighand->siglock held, which may be
2457  * released and re-acquired before returning with intervening sleep.
2458  */
do_jobctl_trap(void)2459 static void do_jobctl_trap(void)
2460 {
2461 	struct signal_struct *signal = current->signal;
2462 	int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2463 
2464 	if (current->ptrace & PT_SEIZED) {
2465 		if (!signal->group_stop_count &&
2466 		    !(signal->flags & SIGNAL_STOP_STOPPED))
2467 			signr = SIGTRAP;
2468 		WARN_ON_ONCE(!signr);
2469 		ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2470 				 CLD_STOPPED);
2471 	} else {
2472 		WARN_ON_ONCE(!signr);
2473 		ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2474 		current->exit_code = 0;
2475 	}
2476 }
2477 
2478 /**
2479  * do_freezer_trap - handle the freezer jobctl trap
2480  *
2481  * Puts the task into frozen state, if only the task is not about to quit.
2482  * In this case it drops JOBCTL_TRAP_FREEZE.
2483  *
2484  * CONTEXT:
2485  * Must be called with @current->sighand->siglock held,
2486  * which is always released before returning.
2487  */
do_freezer_trap(void)2488 static void do_freezer_trap(void)
2489 	__releases(&current->sighand->siglock)
2490 {
2491 	/*
2492 	 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2493 	 * let's make another loop to give it a chance to be handled.
2494 	 * In any case, we'll return back.
2495 	 */
2496 	if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2497 	     JOBCTL_TRAP_FREEZE) {
2498 		spin_unlock_irq(&current->sighand->siglock);
2499 		return;
2500 	}
2501 
2502 	/*
2503 	 * Now we're sure that there is no pending fatal signal and no
2504 	 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2505 	 * immediately (if there is a non-fatal signal pending), and
2506 	 * put the task into sleep.
2507 	 */
2508 	__set_current_state(TASK_INTERRUPTIBLE);
2509 	clear_thread_flag(TIF_SIGPENDING);
2510 	spin_unlock_irq(&current->sighand->siglock);
2511 	cgroup_enter_frozen();
2512 	freezable_schedule();
2513 }
2514 
ptrace_signal(int signr,kernel_siginfo_t * info)2515 static int ptrace_signal(int signr, kernel_siginfo_t *info)
2516 {
2517 	/*
2518 	 * We do not check sig_kernel_stop(signr) but set this marker
2519 	 * unconditionally because we do not know whether debugger will
2520 	 * change signr. This flag has no meaning unless we are going
2521 	 * to stop after return from ptrace_stop(). In this case it will
2522 	 * be checked in do_signal_stop(), we should only stop if it was
2523 	 * not cleared by SIGCONT while we were sleeping. See also the
2524 	 * comment in dequeue_signal().
2525 	 */
2526 	current->jobctl |= JOBCTL_STOP_DEQUEUED;
2527 	ptrace_stop(signr, CLD_TRAPPED, 0, info);
2528 
2529 	/* We're back.  Did the debugger cancel the sig?  */
2530 	signr = current->exit_code;
2531 	if (signr == 0)
2532 		return signr;
2533 
2534 	current->exit_code = 0;
2535 
2536 	/*
2537 	 * Update the siginfo structure if the signal has
2538 	 * changed.  If the debugger wanted something
2539 	 * specific in the siginfo structure then it should
2540 	 * have updated *info via PTRACE_SETSIGINFO.
2541 	 */
2542 	if (signr != info->si_signo) {
2543 		clear_siginfo(info);
2544 		info->si_signo = signr;
2545 		info->si_errno = 0;
2546 		info->si_code = SI_USER;
2547 		rcu_read_lock();
2548 		info->si_pid = task_pid_vnr(current->parent);
2549 		info->si_uid = from_kuid_munged(current_user_ns(),
2550 						task_uid(current->parent));
2551 		rcu_read_unlock();
2552 	}
2553 
2554 	/* If the (new) signal is now blocked, requeue it.  */
2555 	if (sigismember(&current->blocked, signr)) {
2556 		send_signal(signr, info, current, PIDTYPE_PID);
2557 		signr = 0;
2558 	}
2559 
2560 	return signr;
2561 }
2562 
hide_si_addr_tag_bits(struct ksignal * ksig)2563 static void hide_si_addr_tag_bits(struct ksignal *ksig)
2564 {
2565 	switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2566 	case SIL_FAULT:
2567 	case SIL_FAULT_MCEERR:
2568 	case SIL_FAULT_BNDERR:
2569 	case SIL_FAULT_PKUERR:
2570 	case SIL_PERF_EVENT:
2571 		ksig->info.si_addr = arch_untagged_si_addr(
2572 			ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2573 		break;
2574 	case SIL_KILL:
2575 	case SIL_TIMER:
2576 	case SIL_POLL:
2577 	case SIL_CHLD:
2578 	case SIL_RT:
2579 	case SIL_SYS:
2580 		break;
2581 	}
2582 }
2583 
get_signal(struct ksignal * ksig)2584 bool get_signal(struct ksignal *ksig)
2585 {
2586 	struct sighand_struct *sighand = current->sighand;
2587 	struct signal_struct *signal = current->signal;
2588 	int signr;
2589 
2590 	if (unlikely(current->task_works))
2591 		task_work_run();
2592 
2593 	/*
2594 	 * For non-generic architectures, check for TIF_NOTIFY_SIGNAL so
2595 	 * that the arch handlers don't all have to do it. If we get here
2596 	 * without TIF_SIGPENDING, just exit after running signal work.
2597 	 */
2598 	if (!IS_ENABLED(CONFIG_GENERIC_ENTRY)) {
2599 		if (test_thread_flag(TIF_NOTIFY_SIGNAL))
2600 			tracehook_notify_signal();
2601 		if (!task_sigpending(current))
2602 			return false;
2603 	}
2604 
2605 	if (unlikely(uprobe_deny_signal()))
2606 		return false;
2607 
2608 	/*
2609 	 * Do this once, we can't return to user-mode if freezing() == T.
2610 	 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2611 	 * thus do not need another check after return.
2612 	 */
2613 	try_to_freeze();
2614 
2615 relock:
2616 	spin_lock_irq(&sighand->siglock);
2617 
2618 	/*
2619 	 * Every stopped thread goes here after wakeup. Check to see if
2620 	 * we should notify the parent, prepare_signal(SIGCONT) encodes
2621 	 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2622 	 */
2623 	if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2624 		int why;
2625 
2626 		if (signal->flags & SIGNAL_CLD_CONTINUED)
2627 			why = CLD_CONTINUED;
2628 		else
2629 			why = CLD_STOPPED;
2630 
2631 		signal->flags &= ~SIGNAL_CLD_MASK;
2632 
2633 		spin_unlock_irq(&sighand->siglock);
2634 
2635 		/*
2636 		 * Notify the parent that we're continuing.  This event is
2637 		 * always per-process and doesn't make whole lot of sense
2638 		 * for ptracers, who shouldn't consume the state via
2639 		 * wait(2) either, but, for backward compatibility, notify
2640 		 * the ptracer of the group leader too unless it's gonna be
2641 		 * a duplicate.
2642 		 */
2643 		read_lock(&tasklist_lock);
2644 		do_notify_parent_cldstop(current, false, why);
2645 
2646 		if (ptrace_reparented(current->group_leader))
2647 			do_notify_parent_cldstop(current->group_leader,
2648 						true, why);
2649 		read_unlock(&tasklist_lock);
2650 
2651 		goto relock;
2652 	}
2653 
2654 	/* Has this task already been marked for death? */
2655 	if (signal_group_exit(signal)) {
2656 		ksig->info.si_signo = signr = SIGKILL;
2657 		sigdelset(&current->pending.signal, SIGKILL);
2658 		trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2659 				&sighand->action[SIGKILL - 1]);
2660 		recalc_sigpending();
2661 		goto fatal;
2662 	}
2663 
2664 	for (;;) {
2665 		struct k_sigaction *ka;
2666 
2667 		if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2668 		    do_signal_stop(0))
2669 			goto relock;
2670 
2671 		if (unlikely(current->jobctl &
2672 			     (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2673 			if (current->jobctl & JOBCTL_TRAP_MASK) {
2674 				do_jobctl_trap();
2675 				spin_unlock_irq(&sighand->siglock);
2676 			} else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2677 				do_freezer_trap();
2678 
2679 			goto relock;
2680 		}
2681 
2682 		/*
2683 		 * If the task is leaving the frozen state, let's update
2684 		 * cgroup counters and reset the frozen bit.
2685 		 */
2686 		if (unlikely(cgroup_task_frozen(current))) {
2687 			spin_unlock_irq(&sighand->siglock);
2688 			cgroup_leave_frozen(false);
2689 			goto relock;
2690 		}
2691 
2692 		/*
2693 		 * Signals generated by the execution of an instruction
2694 		 * need to be delivered before any other pending signals
2695 		 * so that the instruction pointer in the signal stack
2696 		 * frame points to the faulting instruction.
2697 		 */
2698 		signr = dequeue_synchronous_signal(&ksig->info);
2699 		if (!signr)
2700 			signr = dequeue_signal(current, &current->blocked, &ksig->info);
2701 
2702 		if (!signr)
2703 			break; /* will return 0 */
2704 
2705 		if (unlikely(current->ptrace) && signr != SIGKILL) {
2706 			signr = ptrace_signal(signr, &ksig->info);
2707 			if (!signr)
2708 				continue;
2709 		}
2710 
2711 		ka = &sighand->action[signr-1];
2712 
2713 		/* Trace actually delivered signals. */
2714 		trace_signal_deliver(signr, &ksig->info, ka);
2715 
2716 		if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
2717 			continue;
2718 		if (ka->sa.sa_handler != SIG_DFL) {
2719 			/* Run the handler.  */
2720 			ksig->ka = *ka;
2721 
2722 			if (ka->sa.sa_flags & SA_ONESHOT)
2723 				ka->sa.sa_handler = SIG_DFL;
2724 
2725 			break; /* will return non-zero "signr" value */
2726 		}
2727 
2728 		/*
2729 		 * Now we are doing the default action for this signal.
2730 		 */
2731 		if (sig_kernel_ignore(signr)) /* Default is nothing. */
2732 			continue;
2733 
2734 		/*
2735 		 * Global init gets no signals it doesn't want.
2736 		 * Container-init gets no signals it doesn't want from same
2737 		 * container.
2738 		 *
2739 		 * Note that if global/container-init sees a sig_kernel_only()
2740 		 * signal here, the signal must have been generated internally
2741 		 * or must have come from an ancestor namespace. In either
2742 		 * case, the signal cannot be dropped.
2743 		 */
2744 		if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2745 				!sig_kernel_only(signr))
2746 			continue;
2747 
2748 		if (sig_kernel_stop(signr)) {
2749 			/*
2750 			 * The default action is to stop all threads in
2751 			 * the thread group.  The job control signals
2752 			 * do nothing in an orphaned pgrp, but SIGSTOP
2753 			 * always works.  Note that siglock needs to be
2754 			 * dropped during the call to is_orphaned_pgrp()
2755 			 * because of lock ordering with tasklist_lock.
2756 			 * This allows an intervening SIGCONT to be posted.
2757 			 * We need to check for that and bail out if necessary.
2758 			 */
2759 			if (signr != SIGSTOP) {
2760 				spin_unlock_irq(&sighand->siglock);
2761 
2762 				/* signals can be posted during this window */
2763 
2764 				if (is_current_pgrp_orphaned())
2765 					goto relock;
2766 
2767 				spin_lock_irq(&sighand->siglock);
2768 			}
2769 
2770 			if (likely(do_signal_stop(ksig->info.si_signo))) {
2771 				/* It released the siglock.  */
2772 				goto relock;
2773 			}
2774 
2775 			/*
2776 			 * We didn't actually stop, due to a race
2777 			 * with SIGCONT or something like that.
2778 			 */
2779 			continue;
2780 		}
2781 
2782 	fatal:
2783 		spin_unlock_irq(&sighand->siglock);
2784 		if (unlikely(cgroup_task_frozen(current)))
2785 			cgroup_leave_frozen(true);
2786 
2787 		/*
2788 		 * Anything else is fatal, maybe with a core dump.
2789 		 */
2790 		current->flags |= PF_SIGNALED;
2791 
2792 		if (sig_kernel_coredump(signr)) {
2793 			if (print_fatal_signals)
2794 				print_fatal_signal(ksig->info.si_signo);
2795 			proc_coredump_connector(current);
2796 			/*
2797 			 * If it was able to dump core, this kills all
2798 			 * other threads in the group and synchronizes with
2799 			 * their demise.  If we lost the race with another
2800 			 * thread getting here, it set group_exit_code
2801 			 * first and our do_group_exit call below will use
2802 			 * that value and ignore the one we pass it.
2803 			 */
2804 			do_coredump(&ksig->info);
2805 		}
2806 
2807 		/*
2808 		 * PF_IO_WORKER threads will catch and exit on fatal signals
2809 		 * themselves. They have cleanup that must be performed, so
2810 		 * we cannot call do_exit() on their behalf.
2811 		 */
2812 		if (current->flags & PF_IO_WORKER)
2813 			goto out;
2814 
2815 		/*
2816 		 * Death signals, no core dump.
2817 		 */
2818 		do_group_exit(ksig->info.si_signo);
2819 		/* NOTREACHED */
2820 	}
2821 	spin_unlock_irq(&sighand->siglock);
2822 out:
2823 	ksig->sig = signr;
2824 
2825 	if (!(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
2826 		hide_si_addr_tag_bits(ksig);
2827 
2828 	return ksig->sig > 0;
2829 }
2830 
2831 /**
2832  * signal_delivered -
2833  * @ksig:		kernel signal struct
2834  * @stepping:		nonzero if debugger single-step or block-step in use
2835  *
2836  * This function should be called when a signal has successfully been
2837  * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2838  * is always blocked, and the signal itself is blocked unless %SA_NODEFER
2839  * is set in @ksig->ka.sa.sa_flags.  Tracing is notified.
2840  */
signal_delivered(struct ksignal * ksig,int stepping)2841 static void signal_delivered(struct ksignal *ksig, int stepping)
2842 {
2843 	sigset_t blocked;
2844 
2845 	/* A signal was successfully delivered, and the
2846 	   saved sigmask was stored on the signal frame,
2847 	   and will be restored by sigreturn.  So we can
2848 	   simply clear the restore sigmask flag.  */
2849 	clear_restore_sigmask();
2850 
2851 	sigorsets(&blocked, &current->blocked, &ksig->ka.sa.sa_mask);
2852 	if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2853 		sigaddset(&blocked, ksig->sig);
2854 	set_current_blocked(&blocked);
2855 	tracehook_signal_handler(stepping);
2856 }
2857 
signal_setup_done(int failed,struct ksignal * ksig,int stepping)2858 void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2859 {
2860 	if (failed)
2861 		force_sigsegv(ksig->sig);
2862 	else
2863 		signal_delivered(ksig, stepping);
2864 }
2865 
2866 /*
2867  * It could be that complete_signal() picked us to notify about the
2868  * group-wide signal. Other threads should be notified now to take
2869  * the shared signals in @which since we will not.
2870  */
retarget_shared_pending(struct task_struct * tsk,sigset_t * which)2871 static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2872 {
2873 	sigset_t retarget;
2874 	struct task_struct *t;
2875 
2876 	sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2877 	if (sigisemptyset(&retarget))
2878 		return;
2879 
2880 	t = tsk;
2881 	while_each_thread(tsk, t) {
2882 		if (t->flags & PF_EXITING)
2883 			continue;
2884 
2885 		if (!has_pending_signals(&retarget, &t->blocked))
2886 			continue;
2887 		/* Remove the signals this thread can handle. */
2888 		sigandsets(&retarget, &retarget, &t->blocked);
2889 
2890 		if (!task_sigpending(t))
2891 			signal_wake_up(t, 0);
2892 
2893 		if (sigisemptyset(&retarget))
2894 			break;
2895 	}
2896 }
2897 
exit_signals(struct task_struct * tsk)2898 void exit_signals(struct task_struct *tsk)
2899 {
2900 	int group_stop = 0;
2901 	sigset_t unblocked;
2902 
2903 	/*
2904 	 * @tsk is about to have PF_EXITING set - lock out users which
2905 	 * expect stable threadgroup.
2906 	 */
2907 	cgroup_threadgroup_change_begin(tsk);
2908 
2909 	if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2910 		tsk->flags |= PF_EXITING;
2911 		cgroup_threadgroup_change_end(tsk);
2912 		return;
2913 	}
2914 
2915 	spin_lock_irq(&tsk->sighand->siglock);
2916 	/*
2917 	 * From now this task is not visible for group-wide signals,
2918 	 * see wants_signal(), do_signal_stop().
2919 	 */
2920 	tsk->flags |= PF_EXITING;
2921 
2922 	cgroup_threadgroup_change_end(tsk);
2923 
2924 	if (!task_sigpending(tsk))
2925 		goto out;
2926 
2927 	unblocked = tsk->blocked;
2928 	signotset(&unblocked);
2929 	retarget_shared_pending(tsk, &unblocked);
2930 
2931 	if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2932 	    task_participate_group_stop(tsk))
2933 		group_stop = CLD_STOPPED;
2934 out:
2935 	spin_unlock_irq(&tsk->sighand->siglock);
2936 
2937 	/*
2938 	 * If group stop has completed, deliver the notification.  This
2939 	 * should always go to the real parent of the group leader.
2940 	 */
2941 	if (unlikely(group_stop)) {
2942 		read_lock(&tasklist_lock);
2943 		do_notify_parent_cldstop(tsk, false, group_stop);
2944 		read_unlock(&tasklist_lock);
2945 	}
2946 }
2947 
2948 /*
2949  * System call entry points.
2950  */
2951 
2952 /**
2953  *  sys_restart_syscall - restart a system call
2954  */
SYSCALL_DEFINE0(restart_syscall)2955 SYSCALL_DEFINE0(restart_syscall)
2956 {
2957 	struct restart_block *restart = &current->restart_block;
2958 	return restart->fn(restart);
2959 }
2960 
do_no_restart_syscall(struct restart_block * param)2961 long do_no_restart_syscall(struct restart_block *param)
2962 {
2963 	return -EINTR;
2964 }
2965 
__set_task_blocked(struct task_struct * tsk,const sigset_t * newset)2966 static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2967 {
2968 	if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
2969 		sigset_t newblocked;
2970 		/* A set of now blocked but previously unblocked signals. */
2971 		sigandnsets(&newblocked, newset, &current->blocked);
2972 		retarget_shared_pending(tsk, &newblocked);
2973 	}
2974 	tsk->blocked = *newset;
2975 	recalc_sigpending();
2976 }
2977 
2978 /**
2979  * set_current_blocked - change current->blocked mask
2980  * @newset: new mask
2981  *
2982  * It is wrong to change ->blocked directly, this helper should be used
2983  * to ensure the process can't miss a shared signal we are going to block.
2984  */
set_current_blocked(sigset_t * newset)2985 void set_current_blocked(sigset_t *newset)
2986 {
2987 	sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
2988 	__set_current_blocked(newset);
2989 }
2990 
__set_current_blocked(const sigset_t * newset)2991 void __set_current_blocked(const sigset_t *newset)
2992 {
2993 	struct task_struct *tsk = current;
2994 
2995 	/*
2996 	 * In case the signal mask hasn't changed, there is nothing we need
2997 	 * to do. The current->blocked shouldn't be modified by other task.
2998 	 */
2999 	if (sigequalsets(&tsk->blocked, newset))
3000 		return;
3001 
3002 	spin_lock_irq(&tsk->sighand->siglock);
3003 	__set_task_blocked(tsk, newset);
3004 	spin_unlock_irq(&tsk->sighand->siglock);
3005 }
3006 
3007 /*
3008  * This is also useful for kernel threads that want to temporarily
3009  * (or permanently) block certain signals.
3010  *
3011  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
3012  * interface happily blocks "unblockable" signals like SIGKILL
3013  * and friends.
3014  */
sigprocmask(int how,sigset_t * set,sigset_t * oldset)3015 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
3016 {
3017 	struct task_struct *tsk = current;
3018 	sigset_t newset;
3019 
3020 	/* Lockless, only current can change ->blocked, never from irq */
3021 	if (oldset)
3022 		*oldset = tsk->blocked;
3023 
3024 	switch (how) {
3025 	case SIG_BLOCK:
3026 		sigorsets(&newset, &tsk->blocked, set);
3027 		break;
3028 	case SIG_UNBLOCK:
3029 		sigandnsets(&newset, &tsk->blocked, set);
3030 		break;
3031 	case SIG_SETMASK:
3032 		newset = *set;
3033 		break;
3034 	default:
3035 		return -EINVAL;
3036 	}
3037 
3038 	__set_current_blocked(&newset);
3039 	return 0;
3040 }
3041 EXPORT_SYMBOL(sigprocmask);
3042 
3043 /*
3044  * The api helps set app-provided sigmasks.
3045  *
3046  * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3047  * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3048  *
3049  * Note that it does set_restore_sigmask() in advance, so it must be always
3050  * paired with restore_saved_sigmask_unless() before return from syscall.
3051  */
set_user_sigmask(const sigset_t __user * umask,size_t sigsetsize)3052 int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3053 {
3054 	sigset_t kmask;
3055 
3056 	if (!umask)
3057 		return 0;
3058 	if (sigsetsize != sizeof(sigset_t))
3059 		return -EINVAL;
3060 	if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3061 		return -EFAULT;
3062 
3063 	set_restore_sigmask();
3064 	current->saved_sigmask = current->blocked;
3065 	set_current_blocked(&kmask);
3066 
3067 	return 0;
3068 }
3069 
3070 #ifdef CONFIG_COMPAT
set_compat_user_sigmask(const compat_sigset_t __user * umask,size_t sigsetsize)3071 int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3072 			    size_t sigsetsize)
3073 {
3074 	sigset_t kmask;
3075 
3076 	if (!umask)
3077 		return 0;
3078 	if (sigsetsize != sizeof(compat_sigset_t))
3079 		return -EINVAL;
3080 	if (get_compat_sigset(&kmask, umask))
3081 		return -EFAULT;
3082 
3083 	set_restore_sigmask();
3084 	current->saved_sigmask = current->blocked;
3085 	set_current_blocked(&kmask);
3086 
3087 	return 0;
3088 }
3089 #endif
3090 
3091 /**
3092  *  sys_rt_sigprocmask - change the list of currently blocked signals
3093  *  @how: whether to add, remove, or set signals
3094  *  @nset: stores pending signals
3095  *  @oset: previous value of signal mask if non-null
3096  *  @sigsetsize: size of sigset_t type
3097  */
SYSCALL_DEFINE4(rt_sigprocmask,int,how,sigset_t __user *,nset,sigset_t __user *,oset,size_t,sigsetsize)3098 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3099 		sigset_t __user *, oset, size_t, sigsetsize)
3100 {
3101 	sigset_t old_set, new_set;
3102 	int error;
3103 
3104 	/* XXX: Don't preclude handling different sized sigset_t's.  */
3105 	if (sigsetsize != sizeof(sigset_t))
3106 		return -EINVAL;
3107 
3108 	old_set = current->blocked;
3109 
3110 	if (nset) {
3111 		if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3112 			return -EFAULT;
3113 		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3114 
3115 		error = sigprocmask(how, &new_set, NULL);
3116 		if (error)
3117 			return error;
3118 	}
3119 
3120 	if (oset) {
3121 		if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3122 			return -EFAULT;
3123 	}
3124 
3125 	return 0;
3126 }
3127 
3128 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(rt_sigprocmask,int,how,compat_sigset_t __user *,nset,compat_sigset_t __user *,oset,compat_size_t,sigsetsize)3129 COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3130 		compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3131 {
3132 	sigset_t old_set = current->blocked;
3133 
3134 	/* XXX: Don't preclude handling different sized sigset_t's.  */
3135 	if (sigsetsize != sizeof(sigset_t))
3136 		return -EINVAL;
3137 
3138 	if (nset) {
3139 		sigset_t new_set;
3140 		int error;
3141 		if (get_compat_sigset(&new_set, nset))
3142 			return -EFAULT;
3143 		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3144 
3145 		error = sigprocmask(how, &new_set, NULL);
3146 		if (error)
3147 			return error;
3148 	}
3149 	return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3150 }
3151 #endif
3152 
do_sigpending(sigset_t * set)3153 static void do_sigpending(sigset_t *set)
3154 {
3155 	spin_lock_irq(&current->sighand->siglock);
3156 	sigorsets(set, &current->pending.signal,
3157 		  &current->signal->shared_pending.signal);
3158 	spin_unlock_irq(&current->sighand->siglock);
3159 
3160 	/* Outside the lock because only this thread touches it.  */
3161 	sigandsets(set, &current->blocked, set);
3162 }
3163 
3164 /**
3165  *  sys_rt_sigpending - examine a pending signal that has been raised
3166  *			while blocked
3167  *  @uset: stores pending signals
3168  *  @sigsetsize: size of sigset_t type or larger
3169  */
SYSCALL_DEFINE2(rt_sigpending,sigset_t __user *,uset,size_t,sigsetsize)3170 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3171 {
3172 	sigset_t set;
3173 
3174 	if (sigsetsize > sizeof(*uset))
3175 		return -EINVAL;
3176 
3177 	do_sigpending(&set);
3178 
3179 	if (copy_to_user(uset, &set, sigsetsize))
3180 		return -EFAULT;
3181 
3182 	return 0;
3183 }
3184 
3185 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(rt_sigpending,compat_sigset_t __user *,uset,compat_size_t,sigsetsize)3186 COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3187 		compat_size_t, sigsetsize)
3188 {
3189 	sigset_t set;
3190 
3191 	if (sigsetsize > sizeof(*uset))
3192 		return -EINVAL;
3193 
3194 	do_sigpending(&set);
3195 
3196 	return put_compat_sigset(uset, &set, sigsetsize);
3197 }
3198 #endif
3199 
3200 static const struct {
3201 	unsigned char limit, layout;
3202 } sig_sicodes[] = {
3203 	[SIGILL]  = { NSIGILL,  SIL_FAULT },
3204 	[SIGFPE]  = { NSIGFPE,  SIL_FAULT },
3205 	[SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3206 	[SIGBUS]  = { NSIGBUS,  SIL_FAULT },
3207 	[SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3208 #if defined(SIGEMT)
3209 	[SIGEMT]  = { NSIGEMT,  SIL_FAULT },
3210 #endif
3211 	[SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3212 	[SIGPOLL] = { NSIGPOLL, SIL_POLL },
3213 	[SIGSYS]  = { NSIGSYS,  SIL_SYS },
3214 };
3215 
known_siginfo_layout(unsigned sig,int si_code)3216 static bool known_siginfo_layout(unsigned sig, int si_code)
3217 {
3218 	if (si_code == SI_KERNEL)
3219 		return true;
3220 	else if ((si_code > SI_USER)) {
3221 		if (sig_specific_sicodes(sig)) {
3222 			if (si_code <= sig_sicodes[sig].limit)
3223 				return true;
3224 		}
3225 		else if (si_code <= NSIGPOLL)
3226 			return true;
3227 	}
3228 	else if (si_code >= SI_DETHREAD)
3229 		return true;
3230 	else if (si_code == SI_ASYNCNL)
3231 		return true;
3232 	return false;
3233 }
3234 
siginfo_layout(unsigned sig,int si_code)3235 enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3236 {
3237 	enum siginfo_layout layout = SIL_KILL;
3238 	if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3239 		if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3240 		    (si_code <= sig_sicodes[sig].limit)) {
3241 			layout = sig_sicodes[sig].layout;
3242 			/* Handle the exceptions */
3243 			if ((sig == SIGBUS) &&
3244 			    (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3245 				layout = SIL_FAULT_MCEERR;
3246 			else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3247 				layout = SIL_FAULT_BNDERR;
3248 #ifdef SEGV_PKUERR
3249 			else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3250 				layout = SIL_FAULT_PKUERR;
3251 #endif
3252 			else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3253 				layout = SIL_PERF_EVENT;
3254 		}
3255 		else if (si_code <= NSIGPOLL)
3256 			layout = SIL_POLL;
3257 	} else {
3258 		if (si_code == SI_TIMER)
3259 			layout = SIL_TIMER;
3260 		else if (si_code == SI_SIGIO)
3261 			layout = SIL_POLL;
3262 		else if (si_code < 0)
3263 			layout = SIL_RT;
3264 	}
3265 	return layout;
3266 }
3267 
si_expansion(const siginfo_t __user * info)3268 static inline char __user *si_expansion(const siginfo_t __user *info)
3269 {
3270 	return ((char __user *)info) + sizeof(struct kernel_siginfo);
3271 }
3272 
copy_siginfo_to_user(siginfo_t __user * to,const kernel_siginfo_t * from)3273 int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3274 {
3275 	char __user *expansion = si_expansion(to);
3276 	if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3277 		return -EFAULT;
3278 	if (clear_user(expansion, SI_EXPANSION_SIZE))
3279 		return -EFAULT;
3280 	return 0;
3281 }
3282 
post_copy_siginfo_from_user(kernel_siginfo_t * info,const siginfo_t __user * from)3283 static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3284 				       const siginfo_t __user *from)
3285 {
3286 	if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3287 		char __user *expansion = si_expansion(from);
3288 		char buf[SI_EXPANSION_SIZE];
3289 		int i;
3290 		/*
3291 		 * An unknown si_code might need more than
3292 		 * sizeof(struct kernel_siginfo) bytes.  Verify all of the
3293 		 * extra bytes are 0.  This guarantees copy_siginfo_to_user
3294 		 * will return this data to userspace exactly.
3295 		 */
3296 		if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3297 			return -EFAULT;
3298 		for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3299 			if (buf[i] != 0)
3300 				return -E2BIG;
3301 		}
3302 	}
3303 	return 0;
3304 }
3305 
__copy_siginfo_from_user(int signo,kernel_siginfo_t * to,const siginfo_t __user * from)3306 static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3307 				    const siginfo_t __user *from)
3308 {
3309 	if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3310 		return -EFAULT;
3311 	to->si_signo = signo;
3312 	return post_copy_siginfo_from_user(to, from);
3313 }
3314 
copy_siginfo_from_user(kernel_siginfo_t * to,const siginfo_t __user * from)3315 int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3316 {
3317 	if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3318 		return -EFAULT;
3319 	return post_copy_siginfo_from_user(to, from);
3320 }
3321 
3322 #ifdef CONFIG_COMPAT
3323 /**
3324  * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3325  * @to: compat siginfo destination
3326  * @from: kernel siginfo source
3327  *
3328  * Note: This function does not work properly for the SIGCHLD on x32, but
3329  * fortunately it doesn't have to.  The only valid callers for this function are
3330  * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3331  * The latter does not care because SIGCHLD will never cause a coredump.
3332  */
copy_siginfo_to_external32(struct compat_siginfo * to,const struct kernel_siginfo * from)3333 void copy_siginfo_to_external32(struct compat_siginfo *to,
3334 		const struct kernel_siginfo *from)
3335 {
3336 	memset(to, 0, sizeof(*to));
3337 
3338 	to->si_signo = from->si_signo;
3339 	to->si_errno = from->si_errno;
3340 	to->si_code  = from->si_code;
3341 	switch(siginfo_layout(from->si_signo, from->si_code)) {
3342 	case SIL_KILL:
3343 		to->si_pid = from->si_pid;
3344 		to->si_uid = from->si_uid;
3345 		break;
3346 	case SIL_TIMER:
3347 		to->si_tid     = from->si_tid;
3348 		to->si_overrun = from->si_overrun;
3349 		to->si_int     = from->si_int;
3350 		break;
3351 	case SIL_POLL:
3352 		to->si_band = from->si_band;
3353 		to->si_fd   = from->si_fd;
3354 		break;
3355 	case SIL_FAULT:
3356 		to->si_addr = ptr_to_compat(from->si_addr);
3357 #ifdef __ARCH_SI_TRAPNO
3358 		to->si_trapno = from->si_trapno;
3359 #endif
3360 		break;
3361 	case SIL_FAULT_MCEERR:
3362 		to->si_addr = ptr_to_compat(from->si_addr);
3363 #ifdef __ARCH_SI_TRAPNO
3364 		to->si_trapno = from->si_trapno;
3365 #endif
3366 		to->si_addr_lsb = from->si_addr_lsb;
3367 		break;
3368 	case SIL_FAULT_BNDERR:
3369 		to->si_addr = ptr_to_compat(from->si_addr);
3370 #ifdef __ARCH_SI_TRAPNO
3371 		to->si_trapno = from->si_trapno;
3372 #endif
3373 		to->si_lower = ptr_to_compat(from->si_lower);
3374 		to->si_upper = ptr_to_compat(from->si_upper);
3375 		break;
3376 	case SIL_FAULT_PKUERR:
3377 		to->si_addr = ptr_to_compat(from->si_addr);
3378 #ifdef __ARCH_SI_TRAPNO
3379 		to->si_trapno = from->si_trapno;
3380 #endif
3381 		to->si_pkey = from->si_pkey;
3382 		break;
3383 	case SIL_PERF_EVENT:
3384 		to->si_addr = ptr_to_compat(from->si_addr);
3385 		to->si_perf = from->si_perf;
3386 		break;
3387 	case SIL_CHLD:
3388 		to->si_pid = from->si_pid;
3389 		to->si_uid = from->si_uid;
3390 		to->si_status = from->si_status;
3391 		to->si_utime = from->si_utime;
3392 		to->si_stime = from->si_stime;
3393 		break;
3394 	case SIL_RT:
3395 		to->si_pid = from->si_pid;
3396 		to->si_uid = from->si_uid;
3397 		to->si_int = from->si_int;
3398 		break;
3399 	case SIL_SYS:
3400 		to->si_call_addr = ptr_to_compat(from->si_call_addr);
3401 		to->si_syscall   = from->si_syscall;
3402 		to->si_arch      = from->si_arch;
3403 		break;
3404 	}
3405 }
3406 
__copy_siginfo_to_user32(struct compat_siginfo __user * to,const struct kernel_siginfo * from)3407 int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3408 			   const struct kernel_siginfo *from)
3409 {
3410 	struct compat_siginfo new;
3411 
3412 	copy_siginfo_to_external32(&new, from);
3413 	if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3414 		return -EFAULT;
3415 	return 0;
3416 }
3417 
post_copy_siginfo_from_user32(kernel_siginfo_t * to,const struct compat_siginfo * from)3418 static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3419 					 const struct compat_siginfo *from)
3420 {
3421 	clear_siginfo(to);
3422 	to->si_signo = from->si_signo;
3423 	to->si_errno = from->si_errno;
3424 	to->si_code  = from->si_code;
3425 	switch(siginfo_layout(from->si_signo, from->si_code)) {
3426 	case SIL_KILL:
3427 		to->si_pid = from->si_pid;
3428 		to->si_uid = from->si_uid;
3429 		break;
3430 	case SIL_TIMER:
3431 		to->si_tid     = from->si_tid;
3432 		to->si_overrun = from->si_overrun;
3433 		to->si_int     = from->si_int;
3434 		break;
3435 	case SIL_POLL:
3436 		to->si_band = from->si_band;
3437 		to->si_fd   = from->si_fd;
3438 		break;
3439 	case SIL_FAULT:
3440 		to->si_addr = compat_ptr(from->si_addr);
3441 #ifdef __ARCH_SI_TRAPNO
3442 		to->si_trapno = from->si_trapno;
3443 #endif
3444 		break;
3445 	case SIL_FAULT_MCEERR:
3446 		to->si_addr = compat_ptr(from->si_addr);
3447 #ifdef __ARCH_SI_TRAPNO
3448 		to->si_trapno = from->si_trapno;
3449 #endif
3450 		to->si_addr_lsb = from->si_addr_lsb;
3451 		break;
3452 	case SIL_FAULT_BNDERR:
3453 		to->si_addr = compat_ptr(from->si_addr);
3454 #ifdef __ARCH_SI_TRAPNO
3455 		to->si_trapno = from->si_trapno;
3456 #endif
3457 		to->si_lower = compat_ptr(from->si_lower);
3458 		to->si_upper = compat_ptr(from->si_upper);
3459 		break;
3460 	case SIL_FAULT_PKUERR:
3461 		to->si_addr = compat_ptr(from->si_addr);
3462 #ifdef __ARCH_SI_TRAPNO
3463 		to->si_trapno = from->si_trapno;
3464 #endif
3465 		to->si_pkey = from->si_pkey;
3466 		break;
3467 	case SIL_PERF_EVENT:
3468 		to->si_addr = compat_ptr(from->si_addr);
3469 		to->si_perf = from->si_perf;
3470 		break;
3471 	case SIL_CHLD:
3472 		to->si_pid    = from->si_pid;
3473 		to->si_uid    = from->si_uid;
3474 		to->si_status = from->si_status;
3475 #ifdef CONFIG_X86_X32_ABI
3476 		if (in_x32_syscall()) {
3477 			to->si_utime = from->_sifields._sigchld_x32._utime;
3478 			to->si_stime = from->_sifields._sigchld_x32._stime;
3479 		} else
3480 #endif
3481 		{
3482 			to->si_utime = from->si_utime;
3483 			to->si_stime = from->si_stime;
3484 		}
3485 		break;
3486 	case SIL_RT:
3487 		to->si_pid = from->si_pid;
3488 		to->si_uid = from->si_uid;
3489 		to->si_int = from->si_int;
3490 		break;
3491 	case SIL_SYS:
3492 		to->si_call_addr = compat_ptr(from->si_call_addr);
3493 		to->si_syscall   = from->si_syscall;
3494 		to->si_arch      = from->si_arch;
3495 		break;
3496 	}
3497 	return 0;
3498 }
3499 
__copy_siginfo_from_user32(int signo,struct kernel_siginfo * to,const struct compat_siginfo __user * ufrom)3500 static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3501 				      const struct compat_siginfo __user *ufrom)
3502 {
3503 	struct compat_siginfo from;
3504 
3505 	if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3506 		return -EFAULT;
3507 
3508 	from.si_signo = signo;
3509 	return post_copy_siginfo_from_user32(to, &from);
3510 }
3511 
copy_siginfo_from_user32(struct kernel_siginfo * to,const struct compat_siginfo __user * ufrom)3512 int copy_siginfo_from_user32(struct kernel_siginfo *to,
3513 			     const struct compat_siginfo __user *ufrom)
3514 {
3515 	struct compat_siginfo from;
3516 
3517 	if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3518 		return -EFAULT;
3519 
3520 	return post_copy_siginfo_from_user32(to, &from);
3521 }
3522 #endif /* CONFIG_COMPAT */
3523 
3524 /**
3525  *  do_sigtimedwait - wait for queued signals specified in @which
3526  *  @which: queued signals to wait for
3527  *  @info: if non-null, the signal's siginfo is returned here
3528  *  @ts: upper bound on process time suspension
3529  */
do_sigtimedwait(const sigset_t * which,kernel_siginfo_t * info,const struct timespec64 * ts)3530 static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3531 		    const struct timespec64 *ts)
3532 {
3533 	ktime_t *to = NULL, timeout = KTIME_MAX;
3534 	struct task_struct *tsk = current;
3535 	sigset_t mask = *which;
3536 	int sig, ret = 0;
3537 
3538 	if (ts) {
3539 		if (!timespec64_valid(ts))
3540 			return -EINVAL;
3541 		timeout = timespec64_to_ktime(*ts);
3542 		to = &timeout;
3543 	}
3544 
3545 	/*
3546 	 * Invert the set of allowed signals to get those we want to block.
3547 	 */
3548 	sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3549 	signotset(&mask);
3550 
3551 	spin_lock_irq(&tsk->sighand->siglock);
3552 	sig = dequeue_signal(tsk, &mask, info);
3553 	if (!sig && timeout) {
3554 		/*
3555 		 * None ready, temporarily unblock those we're interested
3556 		 * while we are sleeping in so that we'll be awakened when
3557 		 * they arrive. Unblocking is always fine, we can avoid
3558 		 * set_current_blocked().
3559 		 */
3560 		tsk->real_blocked = tsk->blocked;
3561 		sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3562 		recalc_sigpending();
3563 		spin_unlock_irq(&tsk->sighand->siglock);
3564 
3565 		__set_current_state(TASK_INTERRUPTIBLE);
3566 		ret = freezable_schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3567 							 HRTIMER_MODE_REL);
3568 		spin_lock_irq(&tsk->sighand->siglock);
3569 		__set_task_blocked(tsk, &tsk->real_blocked);
3570 		sigemptyset(&tsk->real_blocked);
3571 		sig = dequeue_signal(tsk, &mask, info);
3572 	}
3573 	spin_unlock_irq(&tsk->sighand->siglock);
3574 
3575 	if (sig)
3576 		return sig;
3577 	return ret ? -EINTR : -EAGAIN;
3578 }
3579 
3580 /**
3581  *  sys_rt_sigtimedwait - synchronously wait for queued signals specified
3582  *			in @uthese
3583  *  @uthese: queued signals to wait for
3584  *  @uinfo: if non-null, the signal's siginfo is returned here
3585  *  @uts: upper bound on process time suspension
3586  *  @sigsetsize: size of sigset_t type
3587  */
SYSCALL_DEFINE4(rt_sigtimedwait,const sigset_t __user *,uthese,siginfo_t __user *,uinfo,const struct __kernel_timespec __user *,uts,size_t,sigsetsize)3588 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3589 		siginfo_t __user *, uinfo,
3590 		const struct __kernel_timespec __user *, uts,
3591 		size_t, sigsetsize)
3592 {
3593 	sigset_t these;
3594 	struct timespec64 ts;
3595 	kernel_siginfo_t info;
3596 	int ret;
3597 
3598 	/* XXX: Don't preclude handling different sized sigset_t's.  */
3599 	if (sigsetsize != sizeof(sigset_t))
3600 		return -EINVAL;
3601 
3602 	if (copy_from_user(&these, uthese, sizeof(these)))
3603 		return -EFAULT;
3604 
3605 	if (uts) {
3606 		if (get_timespec64(&ts, uts))
3607 			return -EFAULT;
3608 	}
3609 
3610 	ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3611 
3612 	if (ret > 0 && uinfo) {
3613 		if (copy_siginfo_to_user(uinfo, &info))
3614 			ret = -EFAULT;
3615 	}
3616 
3617 	return ret;
3618 }
3619 
3620 #ifdef CONFIG_COMPAT_32BIT_TIME
SYSCALL_DEFINE4(rt_sigtimedwait_time32,const sigset_t __user *,uthese,siginfo_t __user *,uinfo,const struct old_timespec32 __user *,uts,size_t,sigsetsize)3621 SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3622 		siginfo_t __user *, uinfo,
3623 		const struct old_timespec32 __user *, uts,
3624 		size_t, sigsetsize)
3625 {
3626 	sigset_t these;
3627 	struct timespec64 ts;
3628 	kernel_siginfo_t info;
3629 	int ret;
3630 
3631 	if (sigsetsize != sizeof(sigset_t))
3632 		return -EINVAL;
3633 
3634 	if (copy_from_user(&these, uthese, sizeof(these)))
3635 		return -EFAULT;
3636 
3637 	if (uts) {
3638 		if (get_old_timespec32(&ts, uts))
3639 			return -EFAULT;
3640 	}
3641 
3642 	ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3643 
3644 	if (ret > 0 && uinfo) {
3645 		if (copy_siginfo_to_user(uinfo, &info))
3646 			ret = -EFAULT;
3647 	}
3648 
3649 	return ret;
3650 }
3651 #endif
3652 
3653 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64,compat_sigset_t __user *,uthese,struct compat_siginfo __user *,uinfo,struct __kernel_timespec __user *,uts,compat_size_t,sigsetsize)3654 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3655 		struct compat_siginfo __user *, uinfo,
3656 		struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3657 {
3658 	sigset_t s;
3659 	struct timespec64 t;
3660 	kernel_siginfo_t info;
3661 	long ret;
3662 
3663 	if (sigsetsize != sizeof(sigset_t))
3664 		return -EINVAL;
3665 
3666 	if (get_compat_sigset(&s, uthese))
3667 		return -EFAULT;
3668 
3669 	if (uts) {
3670 		if (get_timespec64(&t, uts))
3671 			return -EFAULT;
3672 	}
3673 
3674 	ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3675 
3676 	if (ret > 0 && uinfo) {
3677 		if (copy_siginfo_to_user32(uinfo, &info))
3678 			ret = -EFAULT;
3679 	}
3680 
3681 	return ret;
3682 }
3683 
3684 #ifdef CONFIG_COMPAT_32BIT_TIME
COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32,compat_sigset_t __user *,uthese,struct compat_siginfo __user *,uinfo,struct old_timespec32 __user *,uts,compat_size_t,sigsetsize)3685 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3686 		struct compat_siginfo __user *, uinfo,
3687 		struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3688 {
3689 	sigset_t s;
3690 	struct timespec64 t;
3691 	kernel_siginfo_t info;
3692 	long ret;
3693 
3694 	if (sigsetsize != sizeof(sigset_t))
3695 		return -EINVAL;
3696 
3697 	if (get_compat_sigset(&s, uthese))
3698 		return -EFAULT;
3699 
3700 	if (uts) {
3701 		if (get_old_timespec32(&t, uts))
3702 			return -EFAULT;
3703 	}
3704 
3705 	ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3706 
3707 	if (ret > 0 && uinfo) {
3708 		if (copy_siginfo_to_user32(uinfo, &info))
3709 			ret = -EFAULT;
3710 	}
3711 
3712 	return ret;
3713 }
3714 #endif
3715 #endif
3716 
prepare_kill_siginfo(int sig,struct kernel_siginfo * info)3717 static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3718 {
3719 	clear_siginfo(info);
3720 	info->si_signo = sig;
3721 	info->si_errno = 0;
3722 	info->si_code = SI_USER;
3723 	info->si_pid = task_tgid_vnr(current);
3724 	info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3725 }
3726 
3727 /**
3728  *  sys_kill - send a signal to a process
3729  *  @pid: the PID of the process
3730  *  @sig: signal to be sent
3731  */
SYSCALL_DEFINE2(kill,pid_t,pid,int,sig)3732 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3733 {
3734 	struct kernel_siginfo info;
3735 
3736 	prepare_kill_siginfo(sig, &info);
3737 
3738 	return kill_something_info(sig, &info, pid);
3739 }
3740 
3741 /*
3742  * Verify that the signaler and signalee either are in the same pid namespace
3743  * or that the signaler's pid namespace is an ancestor of the signalee's pid
3744  * namespace.
3745  */
access_pidfd_pidns(struct pid * pid)3746 static bool access_pidfd_pidns(struct pid *pid)
3747 {
3748 	struct pid_namespace *active = task_active_pid_ns(current);
3749 	struct pid_namespace *p = ns_of_pid(pid);
3750 
3751 	for (;;) {
3752 		if (!p)
3753 			return false;
3754 		if (p == active)
3755 			break;
3756 		p = p->parent;
3757 	}
3758 
3759 	return true;
3760 }
3761 
copy_siginfo_from_user_any(kernel_siginfo_t * kinfo,siginfo_t __user * info)3762 static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3763 		siginfo_t __user *info)
3764 {
3765 #ifdef CONFIG_COMPAT
3766 	/*
3767 	 * Avoid hooking up compat syscalls and instead handle necessary
3768 	 * conversions here. Note, this is a stop-gap measure and should not be
3769 	 * considered a generic solution.
3770 	 */
3771 	if (in_compat_syscall())
3772 		return copy_siginfo_from_user32(
3773 			kinfo, (struct compat_siginfo __user *)info);
3774 #endif
3775 	return copy_siginfo_from_user(kinfo, info);
3776 }
3777 
pidfd_to_pid(const struct file * file)3778 static struct pid *pidfd_to_pid(const struct file *file)
3779 {
3780 	struct pid *pid;
3781 
3782 	pid = pidfd_pid(file);
3783 	if (!IS_ERR(pid))
3784 		return pid;
3785 
3786 	return tgid_pidfd_to_pid(file);
3787 }
3788 
3789 /**
3790  * sys_pidfd_send_signal - Signal a process through a pidfd
3791  * @pidfd:  file descriptor of the process
3792  * @sig:    signal to send
3793  * @info:   signal info
3794  * @flags:  future flags
3795  *
3796  * The syscall currently only signals via PIDTYPE_PID which covers
3797  * kill(<positive-pid>, <signal>. It does not signal threads or process
3798  * groups.
3799  * In order to extend the syscall to threads and process groups the @flags
3800  * argument should be used. In essence, the @flags argument will determine
3801  * what is signaled and not the file descriptor itself. Put in other words,
3802  * grouping is a property of the flags argument not a property of the file
3803  * descriptor.
3804  *
3805  * Return: 0 on success, negative errno on failure
3806  */
SYSCALL_DEFINE4(pidfd_send_signal,int,pidfd,int,sig,siginfo_t __user *,info,unsigned int,flags)3807 SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3808 		siginfo_t __user *, info, unsigned int, flags)
3809 {
3810 	int ret;
3811 	struct fd f;
3812 	struct pid *pid;
3813 	kernel_siginfo_t kinfo;
3814 
3815 	/* Enforce flags be set to 0 until we add an extension. */
3816 	if (flags)
3817 		return -EINVAL;
3818 
3819 	f = fdget(pidfd);
3820 	if (!f.file)
3821 		return -EBADF;
3822 
3823 	/* Is this a pidfd? */
3824 	pid = pidfd_to_pid(f.file);
3825 	if (IS_ERR(pid)) {
3826 		ret = PTR_ERR(pid);
3827 		goto err;
3828 	}
3829 
3830 	ret = -EINVAL;
3831 	if (!access_pidfd_pidns(pid))
3832 		goto err;
3833 
3834 	if (info) {
3835 		ret = copy_siginfo_from_user_any(&kinfo, info);
3836 		if (unlikely(ret))
3837 			goto err;
3838 
3839 		ret = -EINVAL;
3840 		if (unlikely(sig != kinfo.si_signo))
3841 			goto err;
3842 
3843 		/* Only allow sending arbitrary signals to yourself. */
3844 		ret = -EPERM;
3845 		if ((task_pid(current) != pid) &&
3846 		    (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3847 			goto err;
3848 	} else {
3849 		prepare_kill_siginfo(sig, &kinfo);
3850 	}
3851 
3852 	ret = kill_pid_info(sig, &kinfo, pid);
3853 
3854 err:
3855 	fdput(f);
3856 	return ret;
3857 }
3858 
3859 static int
do_send_specific(pid_t tgid,pid_t pid,int sig,struct kernel_siginfo * info)3860 do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3861 {
3862 	struct task_struct *p;
3863 	int error = -ESRCH;
3864 
3865 	rcu_read_lock();
3866 	p = find_task_by_vpid(pid);
3867 	if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3868 		error = check_kill_permission(sig, info, p);
3869 		/*
3870 		 * The null signal is a permissions and process existence
3871 		 * probe.  No signal is actually delivered.
3872 		 */
3873 		if (!error && sig) {
3874 			error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3875 			/*
3876 			 * If lock_task_sighand() failed we pretend the task
3877 			 * dies after receiving the signal. The window is tiny,
3878 			 * and the signal is private anyway.
3879 			 */
3880 			if (unlikely(error == -ESRCH))
3881 				error = 0;
3882 		}
3883 	}
3884 	rcu_read_unlock();
3885 
3886 	return error;
3887 }
3888 
do_tkill(pid_t tgid,pid_t pid,int sig)3889 static int do_tkill(pid_t tgid, pid_t pid, int sig)
3890 {
3891 	struct kernel_siginfo info;
3892 
3893 	clear_siginfo(&info);
3894 	info.si_signo = sig;
3895 	info.si_errno = 0;
3896 	info.si_code = SI_TKILL;
3897 	info.si_pid = task_tgid_vnr(current);
3898 	info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3899 
3900 	return do_send_specific(tgid, pid, sig, &info);
3901 }
3902 
3903 /**
3904  *  sys_tgkill - send signal to one specific thread
3905  *  @tgid: the thread group ID of the thread
3906  *  @pid: the PID of the thread
3907  *  @sig: signal to be sent
3908  *
3909  *  This syscall also checks the @tgid and returns -ESRCH even if the PID
3910  *  exists but it's not belonging to the target process anymore. This
3911  *  method solves the problem of threads exiting and PIDs getting reused.
3912  */
SYSCALL_DEFINE3(tgkill,pid_t,tgid,pid_t,pid,int,sig)3913 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
3914 {
3915 	/* This is only valid for single tasks */
3916 	if (pid <= 0 || tgid <= 0)
3917 		return -EINVAL;
3918 
3919 	return do_tkill(tgid, pid, sig);
3920 }
3921 
3922 /**
3923  *  sys_tkill - send signal to one specific task
3924  *  @pid: the PID of the task
3925  *  @sig: signal to be sent
3926  *
3927  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
3928  */
SYSCALL_DEFINE2(tkill,pid_t,pid,int,sig)3929 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
3930 {
3931 	/* This is only valid for single tasks */
3932 	if (pid <= 0)
3933 		return -EINVAL;
3934 
3935 	return do_tkill(0, pid, sig);
3936 }
3937 
do_rt_sigqueueinfo(pid_t pid,int sig,kernel_siginfo_t * info)3938 static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
3939 {
3940 	/* Not even root can pretend to send signals from the kernel.
3941 	 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3942 	 */
3943 	if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3944 	    (task_pid_vnr(current) != pid))
3945 		return -EPERM;
3946 
3947 	/* POSIX.1b doesn't mention process groups.  */
3948 	return kill_proc_info(sig, info, pid);
3949 }
3950 
3951 /**
3952  *  sys_rt_sigqueueinfo - send signal information to a signal
3953  *  @pid: the PID of the thread
3954  *  @sig: signal to be sent
3955  *  @uinfo: signal info to be sent
3956  */
SYSCALL_DEFINE3(rt_sigqueueinfo,pid_t,pid,int,sig,siginfo_t __user *,uinfo)3957 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
3958 		siginfo_t __user *, uinfo)
3959 {
3960 	kernel_siginfo_t info;
3961 	int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3962 	if (unlikely(ret))
3963 		return ret;
3964 	return do_rt_sigqueueinfo(pid, sig, &info);
3965 }
3966 
3967 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,compat_pid_t,pid,int,sig,struct compat_siginfo __user *,uinfo)3968 COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
3969 			compat_pid_t, pid,
3970 			int, sig,
3971 			struct compat_siginfo __user *, uinfo)
3972 {
3973 	kernel_siginfo_t info;
3974 	int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3975 	if (unlikely(ret))
3976 		return ret;
3977 	return do_rt_sigqueueinfo(pid, sig, &info);
3978 }
3979 #endif
3980 
do_rt_tgsigqueueinfo(pid_t tgid,pid_t pid,int sig,kernel_siginfo_t * info)3981 static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
3982 {
3983 	/* This is only valid for single tasks */
3984 	if (pid <= 0 || tgid <= 0)
3985 		return -EINVAL;
3986 
3987 	/* Not even root can pretend to send signals from the kernel.
3988 	 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3989 	 */
3990 	if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3991 	    (task_pid_vnr(current) != pid))
3992 		return -EPERM;
3993 
3994 	return do_send_specific(tgid, pid, sig, info);
3995 }
3996 
SYSCALL_DEFINE4(rt_tgsigqueueinfo,pid_t,tgid,pid_t,pid,int,sig,siginfo_t __user *,uinfo)3997 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, 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_tgsigqueueinfo(tgid, pid, sig, &info);
4005 }
4006 
4007 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,compat_pid_t,tgid,compat_pid_t,pid,int,sig,struct compat_siginfo __user *,uinfo)4008 COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
4009 			compat_pid_t, tgid,
4010 			compat_pid_t, pid,
4011 			int, sig,
4012 			struct compat_siginfo __user *, uinfo)
4013 {
4014 	kernel_siginfo_t info;
4015 	int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4016 	if (unlikely(ret))
4017 		return ret;
4018 	return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4019 }
4020 #endif
4021 
4022 /*
4023  * For kthreads only, must not be used if cloned with CLONE_SIGHAND
4024  */
kernel_sigaction(int sig,__sighandler_t action)4025 void kernel_sigaction(int sig, __sighandler_t action)
4026 {
4027 	spin_lock_irq(&current->sighand->siglock);
4028 	current->sighand->action[sig - 1].sa.sa_handler = action;
4029 	if (action == SIG_IGN) {
4030 		sigset_t mask;
4031 
4032 		sigemptyset(&mask);
4033 		sigaddset(&mask, sig);
4034 
4035 		flush_sigqueue_mask(&mask, &current->signal->shared_pending);
4036 		flush_sigqueue_mask(&mask, &current->pending);
4037 		recalc_sigpending();
4038 	}
4039 	spin_unlock_irq(&current->sighand->siglock);
4040 }
4041 EXPORT_SYMBOL(kernel_sigaction);
4042 
sigaction_compat_abi(struct k_sigaction * act,struct k_sigaction * oact)4043 void __weak sigaction_compat_abi(struct k_sigaction *act,
4044 		struct k_sigaction *oact)
4045 {
4046 }
4047 
do_sigaction(int sig,struct k_sigaction * act,struct k_sigaction * oact)4048 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4049 {
4050 	struct task_struct *p = current, *t;
4051 	struct k_sigaction *k;
4052 	sigset_t mask;
4053 
4054 	if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4055 		return -EINVAL;
4056 
4057 	k = &p->sighand->action[sig-1];
4058 
4059 	spin_lock_irq(&p->sighand->siglock);
4060 	if (oact)
4061 		*oact = *k;
4062 
4063 	/*
4064 	 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4065 	 * e.g. by having an architecture use the bit in their uapi.
4066 	 */
4067 	BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4068 
4069 	/*
4070 	 * Clear unknown flag bits in order to allow userspace to detect missing
4071 	 * support for flag bits and to allow the kernel to use non-uapi bits
4072 	 * internally.
4073 	 */
4074 	if (act)
4075 		act->sa.sa_flags &= UAPI_SA_FLAGS;
4076 	if (oact)
4077 		oact->sa.sa_flags &= UAPI_SA_FLAGS;
4078 
4079 	sigaction_compat_abi(act, oact);
4080 
4081 	if (act) {
4082 		sigdelsetmask(&act->sa.sa_mask,
4083 			      sigmask(SIGKILL) | sigmask(SIGSTOP));
4084 		*k = *act;
4085 		/*
4086 		 * POSIX 3.3.1.3:
4087 		 *  "Setting a signal action to SIG_IGN for a signal that is
4088 		 *   pending shall cause the pending signal to be discarded,
4089 		 *   whether or not it is blocked."
4090 		 *
4091 		 *  "Setting a signal action to SIG_DFL for a signal that is
4092 		 *   pending and whose default action is to ignore the signal
4093 		 *   (for example, SIGCHLD), shall cause the pending signal to
4094 		 *   be discarded, whether or not it is blocked"
4095 		 */
4096 		if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4097 			sigemptyset(&mask);
4098 			sigaddset(&mask, sig);
4099 			flush_sigqueue_mask(&mask, &p->signal->shared_pending);
4100 			for_each_thread(p, t)
4101 				flush_sigqueue_mask(&mask, &t->pending);
4102 		}
4103 	}
4104 
4105 	spin_unlock_irq(&p->sighand->siglock);
4106 	return 0;
4107 }
4108 
4109 static int
do_sigaltstack(const stack_t * ss,stack_t * oss,unsigned long sp,size_t min_ss_size)4110 do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4111 		size_t min_ss_size)
4112 {
4113 	struct task_struct *t = current;
4114 
4115 	if (oss) {
4116 		memset(oss, 0, sizeof(stack_t));
4117 		oss->ss_sp = (void __user *) t->sas_ss_sp;
4118 		oss->ss_size = t->sas_ss_size;
4119 		oss->ss_flags = sas_ss_flags(sp) |
4120 			(current->sas_ss_flags & SS_FLAG_BITS);
4121 	}
4122 
4123 	if (ss) {
4124 		void __user *ss_sp = ss->ss_sp;
4125 		size_t ss_size = ss->ss_size;
4126 		unsigned ss_flags = ss->ss_flags;
4127 		int ss_mode;
4128 
4129 		if (unlikely(on_sig_stack(sp)))
4130 			return -EPERM;
4131 
4132 		ss_mode = ss_flags & ~SS_FLAG_BITS;
4133 		if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4134 				ss_mode != 0))
4135 			return -EINVAL;
4136 
4137 		if (ss_mode == SS_DISABLE) {
4138 			ss_size = 0;
4139 			ss_sp = NULL;
4140 		} else {
4141 			if (unlikely(ss_size < min_ss_size))
4142 				return -ENOMEM;
4143 		}
4144 
4145 		t->sas_ss_sp = (unsigned long) ss_sp;
4146 		t->sas_ss_size = ss_size;
4147 		t->sas_ss_flags = ss_flags;
4148 	}
4149 	return 0;
4150 }
4151 
SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss,stack_t __user *,uoss)4152 SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4153 {
4154 	stack_t new, old;
4155 	int err;
4156 	if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4157 		return -EFAULT;
4158 	err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4159 			      current_user_stack_pointer(),
4160 			      MINSIGSTKSZ);
4161 	if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4162 		err = -EFAULT;
4163 	return err;
4164 }
4165 
restore_altstack(const stack_t __user * uss)4166 int restore_altstack(const stack_t __user *uss)
4167 {
4168 	stack_t new;
4169 	if (copy_from_user(&new, uss, sizeof(stack_t)))
4170 		return -EFAULT;
4171 	(void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4172 			     MINSIGSTKSZ);
4173 	/* squash all but EFAULT for now */
4174 	return 0;
4175 }
4176 
__save_altstack(stack_t __user * uss,unsigned long sp)4177 int __save_altstack(stack_t __user *uss, unsigned long sp)
4178 {
4179 	struct task_struct *t = current;
4180 	int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4181 		__put_user(t->sas_ss_flags, &uss->ss_flags) |
4182 		__put_user(t->sas_ss_size, &uss->ss_size);
4183 	if (err)
4184 		return err;
4185 	if (t->sas_ss_flags & SS_AUTODISARM)
4186 		sas_ss_reset(t);
4187 	return 0;
4188 }
4189 
4190 #ifdef CONFIG_COMPAT
do_compat_sigaltstack(const compat_stack_t __user * uss_ptr,compat_stack_t __user * uoss_ptr)4191 static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4192 				 compat_stack_t __user *uoss_ptr)
4193 {
4194 	stack_t uss, uoss;
4195 	int ret;
4196 
4197 	if (uss_ptr) {
4198 		compat_stack_t uss32;
4199 		if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4200 			return -EFAULT;
4201 		uss.ss_sp = compat_ptr(uss32.ss_sp);
4202 		uss.ss_flags = uss32.ss_flags;
4203 		uss.ss_size = uss32.ss_size;
4204 	}
4205 	ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4206 			     compat_user_stack_pointer(),
4207 			     COMPAT_MINSIGSTKSZ);
4208 	if (ret >= 0 && uoss_ptr)  {
4209 		compat_stack_t old;
4210 		memset(&old, 0, sizeof(old));
4211 		old.ss_sp = ptr_to_compat(uoss.ss_sp);
4212 		old.ss_flags = uoss.ss_flags;
4213 		old.ss_size = uoss.ss_size;
4214 		if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4215 			ret = -EFAULT;
4216 	}
4217 	return ret;
4218 }
4219 
COMPAT_SYSCALL_DEFINE2(sigaltstack,const compat_stack_t __user *,uss_ptr,compat_stack_t __user *,uoss_ptr)4220 COMPAT_SYSCALL_DEFINE2(sigaltstack,
4221 			const compat_stack_t __user *, uss_ptr,
4222 			compat_stack_t __user *, uoss_ptr)
4223 {
4224 	return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4225 }
4226 
compat_restore_altstack(const compat_stack_t __user * uss)4227 int compat_restore_altstack(const compat_stack_t __user *uss)
4228 {
4229 	int err = do_compat_sigaltstack(uss, NULL);
4230 	/* squash all but -EFAULT for now */
4231 	return err == -EFAULT ? err : 0;
4232 }
4233 
__compat_save_altstack(compat_stack_t __user * uss,unsigned long sp)4234 int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4235 {
4236 	int err;
4237 	struct task_struct *t = current;
4238 	err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4239 			 &uss->ss_sp) |
4240 		__put_user(t->sas_ss_flags, &uss->ss_flags) |
4241 		__put_user(t->sas_ss_size, &uss->ss_size);
4242 	if (err)
4243 		return err;
4244 	if (t->sas_ss_flags & SS_AUTODISARM)
4245 		sas_ss_reset(t);
4246 	return 0;
4247 }
4248 #endif
4249 
4250 #ifdef __ARCH_WANT_SYS_SIGPENDING
4251 
4252 /**
4253  *  sys_sigpending - examine pending signals
4254  *  @uset: where mask of pending signal is returned
4255  */
SYSCALL_DEFINE1(sigpending,old_sigset_t __user *,uset)4256 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4257 {
4258 	sigset_t set;
4259 
4260 	if (sizeof(old_sigset_t) > sizeof(*uset))
4261 		return -EINVAL;
4262 
4263 	do_sigpending(&set);
4264 
4265 	if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4266 		return -EFAULT;
4267 
4268 	return 0;
4269 }
4270 
4271 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE1(sigpending,compat_old_sigset_t __user *,set32)4272 COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4273 {
4274 	sigset_t set;
4275 
4276 	do_sigpending(&set);
4277 
4278 	return put_user(set.sig[0], set32);
4279 }
4280 #endif
4281 
4282 #endif
4283 
4284 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
4285 /**
4286  *  sys_sigprocmask - examine and change blocked signals
4287  *  @how: whether to add, remove, or set signals
4288  *  @nset: signals to add or remove (if non-null)
4289  *  @oset: previous value of signal mask if non-null
4290  *
4291  * Some platforms have their own version with special arguments;
4292  * others support only sys_rt_sigprocmask.
4293  */
4294 
SYSCALL_DEFINE3(sigprocmask,int,how,old_sigset_t __user *,nset,old_sigset_t __user *,oset)4295 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4296 		old_sigset_t __user *, oset)
4297 {
4298 	old_sigset_t old_set, new_set;
4299 	sigset_t new_blocked;
4300 
4301 	old_set = current->blocked.sig[0];
4302 
4303 	if (nset) {
4304 		if (copy_from_user(&new_set, nset, sizeof(*nset)))
4305 			return -EFAULT;
4306 
4307 		new_blocked = current->blocked;
4308 
4309 		switch (how) {
4310 		case SIG_BLOCK:
4311 			sigaddsetmask(&new_blocked, new_set);
4312 			break;
4313 		case SIG_UNBLOCK:
4314 			sigdelsetmask(&new_blocked, new_set);
4315 			break;
4316 		case SIG_SETMASK:
4317 			new_blocked.sig[0] = new_set;
4318 			break;
4319 		default:
4320 			return -EINVAL;
4321 		}
4322 
4323 		set_current_blocked(&new_blocked);
4324 	}
4325 
4326 	if (oset) {
4327 		if (copy_to_user(oset, &old_set, sizeof(*oset)))
4328 			return -EFAULT;
4329 	}
4330 
4331 	return 0;
4332 }
4333 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4334 
4335 #ifndef CONFIG_ODD_RT_SIGACTION
4336 /**
4337  *  sys_rt_sigaction - alter an action taken by a process
4338  *  @sig: signal to be sent
4339  *  @act: new sigaction
4340  *  @oact: used to save the previous sigaction
4341  *  @sigsetsize: size of sigset_t type
4342  */
SYSCALL_DEFINE4(rt_sigaction,int,sig,const struct sigaction __user *,act,struct sigaction __user *,oact,size_t,sigsetsize)4343 SYSCALL_DEFINE4(rt_sigaction, int, sig,
4344 		const struct sigaction __user *, act,
4345 		struct sigaction __user *, oact,
4346 		size_t, sigsetsize)
4347 {
4348 	struct k_sigaction new_sa, old_sa;
4349 	int ret;
4350 
4351 	/* XXX: Don't preclude handling different sized sigset_t's.  */
4352 	if (sigsetsize != sizeof(sigset_t))
4353 		return -EINVAL;
4354 
4355 	if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4356 		return -EFAULT;
4357 
4358 	ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4359 	if (ret)
4360 		return ret;
4361 
4362 	if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4363 		return -EFAULT;
4364 
4365 	return 0;
4366 }
4367 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(rt_sigaction,int,sig,const struct compat_sigaction __user *,act,struct compat_sigaction __user *,oact,compat_size_t,sigsetsize)4368 COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4369 		const struct compat_sigaction __user *, act,
4370 		struct compat_sigaction __user *, oact,
4371 		compat_size_t, sigsetsize)
4372 {
4373 	struct k_sigaction new_ka, old_ka;
4374 #ifdef __ARCH_HAS_SA_RESTORER
4375 	compat_uptr_t restorer;
4376 #endif
4377 	int ret;
4378 
4379 	/* XXX: Don't preclude handling different sized sigset_t's.  */
4380 	if (sigsetsize != sizeof(compat_sigset_t))
4381 		return -EINVAL;
4382 
4383 	if (act) {
4384 		compat_uptr_t handler;
4385 		ret = get_user(handler, &act->sa_handler);
4386 		new_ka.sa.sa_handler = compat_ptr(handler);
4387 #ifdef __ARCH_HAS_SA_RESTORER
4388 		ret |= get_user(restorer, &act->sa_restorer);
4389 		new_ka.sa.sa_restorer = compat_ptr(restorer);
4390 #endif
4391 		ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4392 		ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4393 		if (ret)
4394 			return -EFAULT;
4395 	}
4396 
4397 	ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4398 	if (!ret && oact) {
4399 		ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4400 			       &oact->sa_handler);
4401 		ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4402 					 sizeof(oact->sa_mask));
4403 		ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4404 #ifdef __ARCH_HAS_SA_RESTORER
4405 		ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4406 				&oact->sa_restorer);
4407 #endif
4408 	}
4409 	return ret;
4410 }
4411 #endif
4412 #endif /* !CONFIG_ODD_RT_SIGACTION */
4413 
4414 #ifdef CONFIG_OLD_SIGACTION
SYSCALL_DEFINE3(sigaction,int,sig,const struct old_sigaction __user *,act,struct old_sigaction __user *,oact)4415 SYSCALL_DEFINE3(sigaction, int, sig,
4416 		const struct old_sigaction __user *, act,
4417 	        struct old_sigaction __user *, oact)
4418 {
4419 	struct k_sigaction new_ka, old_ka;
4420 	int ret;
4421 
4422 	if (act) {
4423 		old_sigset_t mask;
4424 		if (!access_ok(act, sizeof(*act)) ||
4425 		    __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4426 		    __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4427 		    __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4428 		    __get_user(mask, &act->sa_mask))
4429 			return -EFAULT;
4430 #ifdef __ARCH_HAS_KA_RESTORER
4431 		new_ka.ka_restorer = NULL;
4432 #endif
4433 		siginitset(&new_ka.sa.sa_mask, mask);
4434 	}
4435 
4436 	ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4437 
4438 	if (!ret && oact) {
4439 		if (!access_ok(oact, sizeof(*oact)) ||
4440 		    __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4441 		    __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4442 		    __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4443 		    __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4444 			return -EFAULT;
4445 	}
4446 
4447 	return ret;
4448 }
4449 #endif
4450 #ifdef CONFIG_COMPAT_OLD_SIGACTION
COMPAT_SYSCALL_DEFINE3(sigaction,int,sig,const struct compat_old_sigaction __user *,act,struct compat_old_sigaction __user *,oact)4451 COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4452 		const struct compat_old_sigaction __user *, act,
4453 	        struct compat_old_sigaction __user *, oact)
4454 {
4455 	struct k_sigaction new_ka, old_ka;
4456 	int ret;
4457 	compat_old_sigset_t mask;
4458 	compat_uptr_t handler, restorer;
4459 
4460 	if (act) {
4461 		if (!access_ok(act, sizeof(*act)) ||
4462 		    __get_user(handler, &act->sa_handler) ||
4463 		    __get_user(restorer, &act->sa_restorer) ||
4464 		    __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4465 		    __get_user(mask, &act->sa_mask))
4466 			return -EFAULT;
4467 
4468 #ifdef __ARCH_HAS_KA_RESTORER
4469 		new_ka.ka_restorer = NULL;
4470 #endif
4471 		new_ka.sa.sa_handler = compat_ptr(handler);
4472 		new_ka.sa.sa_restorer = compat_ptr(restorer);
4473 		siginitset(&new_ka.sa.sa_mask, mask);
4474 	}
4475 
4476 	ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4477 
4478 	if (!ret && oact) {
4479 		if (!access_ok(oact, sizeof(*oact)) ||
4480 		    __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4481 			       &oact->sa_handler) ||
4482 		    __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4483 			       &oact->sa_restorer) ||
4484 		    __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4485 		    __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4486 			return -EFAULT;
4487 	}
4488 	return ret;
4489 }
4490 #endif
4491 
4492 #ifdef CONFIG_SGETMASK_SYSCALL
4493 
4494 /*
4495  * For backwards compatibility.  Functionality superseded by sigprocmask.
4496  */
SYSCALL_DEFINE0(sgetmask)4497 SYSCALL_DEFINE0(sgetmask)
4498 {
4499 	/* SMP safe */
4500 	return current->blocked.sig[0];
4501 }
4502 
SYSCALL_DEFINE1(ssetmask,int,newmask)4503 SYSCALL_DEFINE1(ssetmask, int, newmask)
4504 {
4505 	int old = current->blocked.sig[0];
4506 	sigset_t newset;
4507 
4508 	siginitset(&newset, newmask);
4509 	set_current_blocked(&newset);
4510 
4511 	return old;
4512 }
4513 #endif /* CONFIG_SGETMASK_SYSCALL */
4514 
4515 #ifdef __ARCH_WANT_SYS_SIGNAL
4516 /*
4517  * For backwards compatibility.  Functionality superseded by sigaction.
4518  */
SYSCALL_DEFINE2(signal,int,sig,__sighandler_t,handler)4519 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4520 {
4521 	struct k_sigaction new_sa, old_sa;
4522 	int ret;
4523 
4524 	new_sa.sa.sa_handler = handler;
4525 	new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4526 	sigemptyset(&new_sa.sa.sa_mask);
4527 
4528 	ret = do_sigaction(sig, &new_sa, &old_sa);
4529 
4530 	return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4531 }
4532 #endif /* __ARCH_WANT_SYS_SIGNAL */
4533 
4534 #ifdef __ARCH_WANT_SYS_PAUSE
4535 
SYSCALL_DEFINE0(pause)4536 SYSCALL_DEFINE0(pause)
4537 {
4538 	while (!signal_pending(current)) {
4539 		__set_current_state(TASK_INTERRUPTIBLE);
4540 		schedule();
4541 	}
4542 	return -ERESTARTNOHAND;
4543 }
4544 
4545 #endif
4546 
sigsuspend(sigset_t * set)4547 static int sigsuspend(sigset_t *set)
4548 {
4549 	current->saved_sigmask = current->blocked;
4550 	set_current_blocked(set);
4551 
4552 	while (!signal_pending(current)) {
4553 		__set_current_state(TASK_INTERRUPTIBLE);
4554 		schedule();
4555 	}
4556 	set_restore_sigmask();
4557 	return -ERESTARTNOHAND;
4558 }
4559 
4560 /**
4561  *  sys_rt_sigsuspend - replace the signal mask for a value with the
4562  *	@unewset value until a signal is received
4563  *  @unewset: new signal mask value
4564  *  @sigsetsize: size of sigset_t type
4565  */
SYSCALL_DEFINE2(rt_sigsuspend,sigset_t __user *,unewset,size_t,sigsetsize)4566 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4567 {
4568 	sigset_t newset;
4569 
4570 	/* XXX: Don't preclude handling different sized sigset_t's.  */
4571 	if (sigsetsize != sizeof(sigset_t))
4572 		return -EINVAL;
4573 
4574 	if (copy_from_user(&newset, unewset, sizeof(newset)))
4575 		return -EFAULT;
4576 	return sigsuspend(&newset);
4577 }
4578 
4579 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(rt_sigsuspend,compat_sigset_t __user *,unewset,compat_size_t,sigsetsize)4580 COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4581 {
4582 	sigset_t newset;
4583 
4584 	/* XXX: Don't preclude handling different sized sigset_t's.  */
4585 	if (sigsetsize != sizeof(sigset_t))
4586 		return -EINVAL;
4587 
4588 	if (get_compat_sigset(&newset, unewset))
4589 		return -EFAULT;
4590 	return sigsuspend(&newset);
4591 }
4592 #endif
4593 
4594 #ifdef CONFIG_OLD_SIGSUSPEND
SYSCALL_DEFINE1(sigsuspend,old_sigset_t,mask)4595 SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4596 {
4597 	sigset_t blocked;
4598 	siginitset(&blocked, mask);
4599 	return sigsuspend(&blocked);
4600 }
4601 #endif
4602 #ifdef CONFIG_OLD_SIGSUSPEND3
SYSCALL_DEFINE3(sigsuspend,int,unused1,int,unused2,old_sigset_t,mask)4603 SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4604 {
4605 	sigset_t blocked;
4606 	siginitset(&blocked, mask);
4607 	return sigsuspend(&blocked);
4608 }
4609 #endif
4610 
arch_vma_name(struct vm_area_struct * vma)4611 __weak const char *arch_vma_name(struct vm_area_struct *vma)
4612 {
4613 	return NULL;
4614 }
4615 
siginfo_buildtime_checks(void)4616 static inline void siginfo_buildtime_checks(void)
4617 {
4618 	BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4619 
4620 	/* Verify the offsets in the two siginfos match */
4621 #define CHECK_OFFSET(field) \
4622 	BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4623 
4624 	/* kill */
4625 	CHECK_OFFSET(si_pid);
4626 	CHECK_OFFSET(si_uid);
4627 
4628 	/* timer */
4629 	CHECK_OFFSET(si_tid);
4630 	CHECK_OFFSET(si_overrun);
4631 	CHECK_OFFSET(si_value);
4632 
4633 	/* rt */
4634 	CHECK_OFFSET(si_pid);
4635 	CHECK_OFFSET(si_uid);
4636 	CHECK_OFFSET(si_value);
4637 
4638 	/* sigchld */
4639 	CHECK_OFFSET(si_pid);
4640 	CHECK_OFFSET(si_uid);
4641 	CHECK_OFFSET(si_status);
4642 	CHECK_OFFSET(si_utime);
4643 	CHECK_OFFSET(si_stime);
4644 
4645 	/* sigfault */
4646 	CHECK_OFFSET(si_addr);
4647 	CHECK_OFFSET(si_addr_lsb);
4648 	CHECK_OFFSET(si_lower);
4649 	CHECK_OFFSET(si_upper);
4650 	CHECK_OFFSET(si_pkey);
4651 	CHECK_OFFSET(si_perf);
4652 
4653 	/* sigpoll */
4654 	CHECK_OFFSET(si_band);
4655 	CHECK_OFFSET(si_fd);
4656 
4657 	/* sigsys */
4658 	CHECK_OFFSET(si_call_addr);
4659 	CHECK_OFFSET(si_syscall);
4660 	CHECK_OFFSET(si_arch);
4661 #undef CHECK_OFFSET
4662 
4663 	/* usb asyncio */
4664 	BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4665 		     offsetof(struct siginfo, si_addr));
4666 	if (sizeof(int) == sizeof(void __user *)) {
4667 		BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4668 			     sizeof(void __user *));
4669 	} else {
4670 		BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4671 			      sizeof_field(struct siginfo, si_uid)) !=
4672 			     sizeof(void __user *));
4673 		BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4674 			     offsetof(struct siginfo, si_uid));
4675 	}
4676 #ifdef CONFIG_COMPAT
4677 	BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4678 		     offsetof(struct compat_siginfo, si_addr));
4679 	BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4680 		     sizeof(compat_uptr_t));
4681 	BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4682 		     sizeof_field(struct siginfo, si_pid));
4683 #endif
4684 }
4685 
signals_init(void)4686 void __init signals_init(void)
4687 {
4688 	siginfo_buildtime_checks();
4689 
4690 	sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
4691 }
4692 
4693 #ifdef CONFIG_KGDB_KDB
4694 #include <linux/kdb.h>
4695 /*
4696  * kdb_send_sig - Allows kdb to send signals without exposing
4697  * signal internals.  This function checks if the required locks are
4698  * available before calling the main signal code, to avoid kdb
4699  * deadlocks.
4700  */
kdb_send_sig(struct task_struct * t,int sig)4701 void kdb_send_sig(struct task_struct *t, int sig)
4702 {
4703 	static struct task_struct *kdb_prev_t;
4704 	int new_t, ret;
4705 	if (!spin_trylock(&t->sighand->siglock)) {
4706 		kdb_printf("Can't do kill command now.\n"
4707 			   "The sigmask lock is held somewhere else in "
4708 			   "kernel, try again later\n");
4709 		return;
4710 	}
4711 	new_t = kdb_prev_t != t;
4712 	kdb_prev_t = t;
4713 	if (t->state != TASK_RUNNING && new_t) {
4714 		spin_unlock(&t->sighand->siglock);
4715 		kdb_printf("Process is not RUNNING, sending a signal from "
4716 			   "kdb risks deadlock\n"
4717 			   "on the run queue locks. "
4718 			   "The signal has _not_ been sent.\n"
4719 			   "Reissue the kill command if you want to risk "
4720 			   "the deadlock.\n");
4721 		return;
4722 	}
4723 	ret = send_signal(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4724 	spin_unlock(&t->sighand->siglock);
4725 	if (ret)
4726 		kdb_printf("Fail to deliver Signal %d to process %d.\n",
4727 			   sig, t->pid);
4728 	else
4729 		kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4730 }
4731 #endif	/* CONFIG_KGDB_KDB */
4732