]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - kernel/signal.c
bpf: devmap: Implement devmap prog execution for generic XDP
[mirror_ubuntu-jammy-kernel.git] / kernel / signal.c
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
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
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
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
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 */
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
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 */
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
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
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
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
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 */
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 */
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 */
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 */
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
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 *
411 __sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
412 int override_rlimit, const unsigned int sigqueue_flags)
413 {
414 struct sigqueue *q = NULL;
415 struct ucounts *ucounts = NULL;
416 long sigpending;
417
418 /*
419 * Protect access to @t credentials. This can go away when all
420 * callers hold rcu read lock.
421 *
422 * NOTE! A pending signal will hold on to the user refcount,
423 * and we get/put the refcount only when the sigpending count
424 * changes from/to zero.
425 */
426 rcu_read_lock();
427 ucounts = task_ucounts(t);
428 sigpending = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING, 1);
429 if (sigpending == 1)
430 ucounts = get_ucounts(ucounts);
431 rcu_read_unlock();
432
433 if (override_rlimit || (sigpending < LONG_MAX && sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
434 q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
435 } else {
436 print_dropped_signal(sig);
437 }
438
439 if (unlikely(q == NULL)) {
440 if (ucounts && dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING, 1))
441 put_ucounts(ucounts);
442 } else {
443 INIT_LIST_HEAD(&q->list);
444 q->flags = sigqueue_flags;
445 q->ucounts = ucounts;
446 }
447 return q;
448 }
449
450 static void __sigqueue_free(struct sigqueue *q)
451 {
452 if (q->flags & SIGQUEUE_PREALLOC)
453 return;
454 if (q->ucounts && dec_rlimit_ucounts(q->ucounts, UCOUNT_RLIMIT_SIGPENDING, 1)) {
455 put_ucounts(q->ucounts);
456 q->ucounts = NULL;
457 }
458 kmem_cache_free(sigqueue_cachep, q);
459 }
460
461 void flush_sigqueue(struct sigpending *queue)
462 {
463 struct sigqueue *q;
464
465 sigemptyset(&queue->signal);
466 while (!list_empty(&queue->list)) {
467 q = list_entry(queue->list.next, struct sigqueue , list);
468 list_del_init(&q->list);
469 __sigqueue_free(q);
470 }
471 }
472
473 /*
474 * Flush all pending signals for this kthread.
475 */
476 void flush_signals(struct task_struct *t)
477 {
478 unsigned long flags;
479
480 spin_lock_irqsave(&t->sighand->siglock, flags);
481 clear_tsk_thread_flag(t, TIF_SIGPENDING);
482 flush_sigqueue(&t->pending);
483 flush_sigqueue(&t->signal->shared_pending);
484 spin_unlock_irqrestore(&t->sighand->siglock, flags);
485 }
486 EXPORT_SYMBOL(flush_signals);
487
488 #ifdef CONFIG_POSIX_TIMERS
489 static void __flush_itimer_signals(struct sigpending *pending)
490 {
491 sigset_t signal, retain;
492 struct sigqueue *q, *n;
493
494 signal = pending->signal;
495 sigemptyset(&retain);
496
497 list_for_each_entry_safe(q, n, &pending->list, list) {
498 int sig = q->info.si_signo;
499
500 if (likely(q->info.si_code != SI_TIMER)) {
501 sigaddset(&retain, sig);
502 } else {
503 sigdelset(&signal, sig);
504 list_del_init(&q->list);
505 __sigqueue_free(q);
506 }
507 }
508
509 sigorsets(&pending->signal, &signal, &retain);
510 }
511
512 void flush_itimer_signals(void)
513 {
514 struct task_struct *tsk = current;
515 unsigned long flags;
516
517 spin_lock_irqsave(&tsk->sighand->siglock, flags);
518 __flush_itimer_signals(&tsk->pending);
519 __flush_itimer_signals(&tsk->signal->shared_pending);
520 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
521 }
522 #endif
523
524 void ignore_signals(struct task_struct *t)
525 {
526 int i;
527
528 for (i = 0; i < _NSIG; ++i)
529 t->sighand->action[i].sa.sa_handler = SIG_IGN;
530
531 flush_signals(t);
532 }
533
534 /*
535 * Flush all handlers for a task.
536 */
537
538 void
539 flush_signal_handlers(struct task_struct *t, int force_default)
540 {
541 int i;
542 struct k_sigaction *ka = &t->sighand->action[0];
543 for (i = _NSIG ; i != 0 ; i--) {
544 if (force_default || ka->sa.sa_handler != SIG_IGN)
545 ka->sa.sa_handler = SIG_DFL;
546 ka->sa.sa_flags = 0;
547 #ifdef __ARCH_HAS_SA_RESTORER
548 ka->sa.sa_restorer = NULL;
549 #endif
550 sigemptyset(&ka->sa.sa_mask);
551 ka++;
552 }
553 }
554
555 bool unhandled_signal(struct task_struct *tsk, int sig)
556 {
557 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
558 if (is_global_init(tsk))
559 return true;
560
561 if (handler != SIG_IGN && handler != SIG_DFL)
562 return false;
563
564 /* if ptraced, let the tracer determine */
565 return !tsk->ptrace;
566 }
567
568 static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
569 bool *resched_timer)
570 {
571 struct sigqueue *q, *first = NULL;
572
573 /*
574 * Collect the siginfo appropriate to this signal. Check if
575 * there is another siginfo for the same signal.
576 */
577 list_for_each_entry(q, &list->list, list) {
578 if (q->info.si_signo == sig) {
579 if (first)
580 goto still_pending;
581 first = q;
582 }
583 }
584
585 sigdelset(&list->signal, sig);
586
587 if (first) {
588 still_pending:
589 list_del_init(&first->list);
590 copy_siginfo(info, &first->info);
591
592 *resched_timer =
593 (first->flags & SIGQUEUE_PREALLOC) &&
594 (info->si_code == SI_TIMER) &&
595 (info->si_sys_private);
596
597 __sigqueue_free(first);
598 } else {
599 /*
600 * Ok, it wasn't in the queue. This must be
601 * a fast-pathed signal or we must have been
602 * out of queue space. So zero out the info.
603 */
604 clear_siginfo(info);
605 info->si_signo = sig;
606 info->si_errno = 0;
607 info->si_code = SI_USER;
608 info->si_pid = 0;
609 info->si_uid = 0;
610 }
611 }
612
613 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
614 kernel_siginfo_t *info, bool *resched_timer)
615 {
616 int sig = next_signal(pending, mask);
617
618 if (sig)
619 collect_signal(sig, pending, info, resched_timer);
620 return sig;
621 }
622
623 /*
624 * Dequeue a signal and return the element to the caller, which is
625 * expected to free it.
626 *
627 * All callers have to hold the siglock.
628 */
629 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, kernel_siginfo_t *info)
630 {
631 bool resched_timer = false;
632 int signr;
633
634 /* We only dequeue private signals from ourselves, we don't let
635 * signalfd steal them
636 */
637 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
638 if (!signr) {
639 signr = __dequeue_signal(&tsk->signal->shared_pending,
640 mask, info, &resched_timer);
641 #ifdef CONFIG_POSIX_TIMERS
642 /*
643 * itimer signal ?
644 *
645 * itimers are process shared and we restart periodic
646 * itimers in the signal delivery path to prevent DoS
647 * attacks in the high resolution timer case. This is
648 * compliant with the old way of self-restarting
649 * itimers, as the SIGALRM is a legacy signal and only
650 * queued once. Changing the restart behaviour to
651 * restart the timer in the signal dequeue path is
652 * reducing the timer noise on heavy loaded !highres
653 * systems too.
654 */
655 if (unlikely(signr == SIGALRM)) {
656 struct hrtimer *tmr = &tsk->signal->real_timer;
657
658 if (!hrtimer_is_queued(tmr) &&
659 tsk->signal->it_real_incr != 0) {
660 hrtimer_forward(tmr, tmr->base->get_time(),
661 tsk->signal->it_real_incr);
662 hrtimer_restart(tmr);
663 }
664 }
665 #endif
666 }
667
668 recalc_sigpending();
669 if (!signr)
670 return 0;
671
672 if (unlikely(sig_kernel_stop(signr))) {
673 /*
674 * Set a marker that we have dequeued a stop signal. Our
675 * caller might release the siglock and then the pending
676 * stop signal it is about to process is no longer in the
677 * pending bitmasks, but must still be cleared by a SIGCONT
678 * (and overruled by a SIGKILL). So those cases clear this
679 * shared flag after we've set it. Note that this flag may
680 * remain set after the signal we return is ignored or
681 * handled. That doesn't matter because its only purpose
682 * is to alert stop-signal processing code when another
683 * processor has come along and cleared the flag.
684 */
685 current->jobctl |= JOBCTL_STOP_DEQUEUED;
686 }
687 #ifdef CONFIG_POSIX_TIMERS
688 if (resched_timer) {
689 /*
690 * Release the siglock to ensure proper locking order
691 * of timer locks outside of siglocks. Note, we leave
692 * irqs disabled here, since the posix-timers code is
693 * about to disable them again anyway.
694 */
695 spin_unlock(&tsk->sighand->siglock);
696 posixtimer_rearm(info);
697 spin_lock(&tsk->sighand->siglock);
698
699 /* Don't expose the si_sys_private value to userspace */
700 info->si_sys_private = 0;
701 }
702 #endif
703 return signr;
704 }
705 EXPORT_SYMBOL_GPL(dequeue_signal);
706
707 static int dequeue_synchronous_signal(kernel_siginfo_t *info)
708 {
709 struct task_struct *tsk = current;
710 struct sigpending *pending = &tsk->pending;
711 struct sigqueue *q, *sync = NULL;
712
713 /*
714 * Might a synchronous signal be in the queue?
715 */
716 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
717 return 0;
718
719 /*
720 * Return the first synchronous signal in the queue.
721 */
722 list_for_each_entry(q, &pending->list, list) {
723 /* Synchronous signals have a positive si_code */
724 if ((q->info.si_code > SI_USER) &&
725 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
726 sync = q;
727 goto next;
728 }
729 }
730 return 0;
731 next:
732 /*
733 * Check if there is another siginfo for the same signal.
734 */
735 list_for_each_entry_continue(q, &pending->list, list) {
736 if (q->info.si_signo == sync->info.si_signo)
737 goto still_pending;
738 }
739
740 sigdelset(&pending->signal, sync->info.si_signo);
741 recalc_sigpending();
742 still_pending:
743 list_del_init(&sync->list);
744 copy_siginfo(info, &sync->info);
745 __sigqueue_free(sync);
746 return info->si_signo;
747 }
748
749 /*
750 * Tell a process that it has a new active signal..
751 *
752 * NOTE! we rely on the previous spin_lock to
753 * lock interrupts for us! We can only be called with
754 * "siglock" held, and the local interrupt must
755 * have been disabled when that got acquired!
756 *
757 * No need to set need_resched since signal event passing
758 * goes through ->blocked
759 */
760 void signal_wake_up_state(struct task_struct *t, unsigned int state)
761 {
762 set_tsk_thread_flag(t, TIF_SIGPENDING);
763 /*
764 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
765 * case. We don't check t->state here because there is a race with it
766 * executing another processor and just now entering stopped state.
767 * By using wake_up_state, we ensure the process will wake up and
768 * handle its death signal.
769 */
770 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
771 kick_process(t);
772 }
773
774 /*
775 * Remove signals in mask from the pending set and queue.
776 * Returns 1 if any signals were found.
777 *
778 * All callers must be holding the siglock.
779 */
780 static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
781 {
782 struct sigqueue *q, *n;
783 sigset_t m;
784
785 sigandsets(&m, mask, &s->signal);
786 if (sigisemptyset(&m))
787 return;
788
789 sigandnsets(&s->signal, &s->signal, mask);
790 list_for_each_entry_safe(q, n, &s->list, list) {
791 if (sigismember(mask, q->info.si_signo)) {
792 list_del_init(&q->list);
793 __sigqueue_free(q);
794 }
795 }
796 }
797
798 static inline int is_si_special(const struct kernel_siginfo *info)
799 {
800 return info <= SEND_SIG_PRIV;
801 }
802
803 static inline bool si_fromuser(const struct kernel_siginfo *info)
804 {
805 return info == SEND_SIG_NOINFO ||
806 (!is_si_special(info) && SI_FROMUSER(info));
807 }
808
809 /*
810 * called with RCU read lock from check_kill_permission()
811 */
812 static bool kill_ok_by_cred(struct task_struct *t)
813 {
814 const struct cred *cred = current_cred();
815 const struct cred *tcred = __task_cred(t);
816
817 return uid_eq(cred->euid, tcred->suid) ||
818 uid_eq(cred->euid, tcred->uid) ||
819 uid_eq(cred->uid, tcred->suid) ||
820 uid_eq(cred->uid, tcred->uid) ||
821 ns_capable(tcred->user_ns, CAP_KILL);
822 }
823
824 /*
825 * Bad permissions for sending the signal
826 * - the caller must hold the RCU read lock
827 */
828 static int check_kill_permission(int sig, struct kernel_siginfo *info,
829 struct task_struct *t)
830 {
831 struct pid *sid;
832 int error;
833
834 if (!valid_signal(sig))
835 return -EINVAL;
836
837 if (!si_fromuser(info))
838 return 0;
839
840 error = audit_signal_info(sig, t); /* Let audit system see the signal */
841 if (error)
842 return error;
843
844 if (!same_thread_group(current, t) &&
845 !kill_ok_by_cred(t)) {
846 switch (sig) {
847 case SIGCONT:
848 sid = task_session(t);
849 /*
850 * We don't return the error if sid == NULL. The
851 * task was unhashed, the caller must notice this.
852 */
853 if (!sid || sid == task_session(current))
854 break;
855 fallthrough;
856 default:
857 return -EPERM;
858 }
859 }
860
861 return security_task_kill(t, info, sig, NULL);
862 }
863
864 /**
865 * ptrace_trap_notify - schedule trap to notify ptracer
866 * @t: tracee wanting to notify tracer
867 *
868 * This function schedules sticky ptrace trap which is cleared on the next
869 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
870 * ptracer.
871 *
872 * If @t is running, STOP trap will be taken. If trapped for STOP and
873 * ptracer is listening for events, tracee is woken up so that it can
874 * re-trap for the new event. If trapped otherwise, STOP trap will be
875 * eventually taken without returning to userland after the existing traps
876 * are finished by PTRACE_CONT.
877 *
878 * CONTEXT:
879 * Must be called with @task->sighand->siglock held.
880 */
881 static void ptrace_trap_notify(struct task_struct *t)
882 {
883 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
884 assert_spin_locked(&t->sighand->siglock);
885
886 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
887 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
888 }
889
890 /*
891 * Handle magic process-wide effects of stop/continue signals. Unlike
892 * the signal actions, these happen immediately at signal-generation
893 * time regardless of blocking, ignoring, or handling. This does the
894 * actual continuing for SIGCONT, but not the actual stopping for stop
895 * signals. The process stop is done as a signal action for SIG_DFL.
896 *
897 * Returns true if the signal should be actually delivered, otherwise
898 * it should be dropped.
899 */
900 static bool prepare_signal(int sig, struct task_struct *p, bool force)
901 {
902 struct signal_struct *signal = p->signal;
903 struct task_struct *t;
904 sigset_t flush;
905
906 if (signal->flags & (SIGNAL_GROUP_EXIT | SIGNAL_GROUP_COREDUMP)) {
907 if (!(signal->flags & SIGNAL_GROUP_EXIT))
908 return sig == SIGKILL;
909 /*
910 * The process is in the middle of dying, nothing to do.
911 */
912 } else if (sig_kernel_stop(sig)) {
913 /*
914 * This is a stop signal. Remove SIGCONT from all queues.
915 */
916 siginitset(&flush, sigmask(SIGCONT));
917 flush_sigqueue_mask(&flush, &signal->shared_pending);
918 for_each_thread(p, t)
919 flush_sigqueue_mask(&flush, &t->pending);
920 } else if (sig == SIGCONT) {
921 unsigned int why;
922 /*
923 * Remove all stop signals from all queues, wake all threads.
924 */
925 siginitset(&flush, SIG_KERNEL_STOP_MASK);
926 flush_sigqueue_mask(&flush, &signal->shared_pending);
927 for_each_thread(p, t) {
928 flush_sigqueue_mask(&flush, &t->pending);
929 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
930 if (likely(!(t->ptrace & PT_SEIZED)))
931 wake_up_state(t, __TASK_STOPPED);
932 else
933 ptrace_trap_notify(t);
934 }
935
936 /*
937 * Notify the parent with CLD_CONTINUED if we were stopped.
938 *
939 * If we were in the middle of a group stop, we pretend it
940 * was already finished, and then continued. Since SIGCHLD
941 * doesn't queue we report only CLD_STOPPED, as if the next
942 * CLD_CONTINUED was dropped.
943 */
944 why = 0;
945 if (signal->flags & SIGNAL_STOP_STOPPED)
946 why |= SIGNAL_CLD_CONTINUED;
947 else if (signal->group_stop_count)
948 why |= SIGNAL_CLD_STOPPED;
949
950 if (why) {
951 /*
952 * The first thread which returns from do_signal_stop()
953 * will take ->siglock, notice SIGNAL_CLD_MASK, and
954 * notify its parent. See get_signal().
955 */
956 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
957 signal->group_stop_count = 0;
958 signal->group_exit_code = 0;
959 }
960 }
961
962 return !sig_ignored(p, sig, force);
963 }
964
965 /*
966 * Test if P wants to take SIG. After we've checked all threads with this,
967 * it's equivalent to finding no threads not blocking SIG. Any threads not
968 * blocking SIG were ruled out because they are not running and already
969 * have pending signals. Such threads will dequeue from the shared queue
970 * as soon as they're available, so putting the signal on the shared queue
971 * will be equivalent to sending it to one such thread.
972 */
973 static inline bool wants_signal(int sig, struct task_struct *p)
974 {
975 if (sigismember(&p->blocked, sig))
976 return false;
977
978 if (p->flags & PF_EXITING)
979 return false;
980
981 if (sig == SIGKILL)
982 return true;
983
984 if (task_is_stopped_or_traced(p))
985 return false;
986
987 return task_curr(p) || !task_sigpending(p);
988 }
989
990 static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
991 {
992 struct signal_struct *signal = p->signal;
993 struct task_struct *t;
994
995 /*
996 * Now find a thread we can wake up to take the signal off the queue.
997 *
998 * If the main thread wants the signal, it gets first crack.
999 * Probably the least surprising to the average bear.
1000 */
1001 if (wants_signal(sig, p))
1002 t = p;
1003 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1004 /*
1005 * There is just one thread and it does not need to be woken.
1006 * It will dequeue unblocked signals before it runs again.
1007 */
1008 return;
1009 else {
1010 /*
1011 * Otherwise try to find a suitable thread.
1012 */
1013 t = signal->curr_target;
1014 while (!wants_signal(sig, t)) {
1015 t = next_thread(t);
1016 if (t == signal->curr_target)
1017 /*
1018 * No thread needs to be woken.
1019 * Any eligible threads will see
1020 * the signal in the queue soon.
1021 */
1022 return;
1023 }
1024 signal->curr_target = t;
1025 }
1026
1027 /*
1028 * Found a killable thread. If the signal will be fatal,
1029 * then start taking the whole group down immediately.
1030 */
1031 if (sig_fatal(p, sig) &&
1032 !(signal->flags & SIGNAL_GROUP_EXIT) &&
1033 !sigismember(&t->real_blocked, sig) &&
1034 (sig == SIGKILL || !p->ptrace)) {
1035 /*
1036 * This signal will be fatal to the whole group.
1037 */
1038 if (!sig_kernel_coredump(sig)) {
1039 /*
1040 * Start a group exit and wake everybody up.
1041 * This way we don't have other threads
1042 * running and doing things after a slower
1043 * thread has the fatal signal pending.
1044 */
1045 signal->flags = SIGNAL_GROUP_EXIT;
1046 signal->group_exit_code = sig;
1047 signal->group_stop_count = 0;
1048 t = p;
1049 do {
1050 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1051 sigaddset(&t->pending.signal, SIGKILL);
1052 signal_wake_up(t, 1);
1053 } while_each_thread(p, t);
1054 return;
1055 }
1056 }
1057
1058 /*
1059 * The signal is already in the shared-pending queue.
1060 * Tell the chosen thread to wake up and dequeue it.
1061 */
1062 signal_wake_up(t, sig == SIGKILL);
1063 return;
1064 }
1065
1066 static inline bool legacy_queue(struct sigpending *signals, int sig)
1067 {
1068 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1069 }
1070
1071 static int __send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1072 enum pid_type type, bool force)
1073 {
1074 struct sigpending *pending;
1075 struct sigqueue *q;
1076 int override_rlimit;
1077 int ret = 0, result;
1078
1079 assert_spin_locked(&t->sighand->siglock);
1080
1081 result = TRACE_SIGNAL_IGNORED;
1082 if (!prepare_signal(sig, t, force))
1083 goto ret;
1084
1085 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1086 /*
1087 * Short-circuit ignored signals and support queuing
1088 * exactly one non-rt signal, so that we can get more
1089 * detailed information about the cause of the signal.
1090 */
1091 result = TRACE_SIGNAL_ALREADY_PENDING;
1092 if (legacy_queue(pending, sig))
1093 goto ret;
1094
1095 result = TRACE_SIGNAL_DELIVERED;
1096 /*
1097 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1098 */
1099 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1100 goto out_set;
1101
1102 /*
1103 * Real-time signals must be queued if sent by sigqueue, or
1104 * some other real-time mechanism. It is implementation
1105 * defined whether kill() does so. We attempt to do so, on
1106 * the principle of least surprise, but since kill is not
1107 * allowed to fail with EAGAIN when low on memory we just
1108 * make sure at least one signal gets delivered and don't
1109 * pass on the info struct.
1110 */
1111 if (sig < SIGRTMIN)
1112 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1113 else
1114 override_rlimit = 0;
1115
1116 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit, 0);
1117
1118 if (q) {
1119 list_add_tail(&q->list, &pending->list);
1120 switch ((unsigned long) info) {
1121 case (unsigned long) SEND_SIG_NOINFO:
1122 clear_siginfo(&q->info);
1123 q->info.si_signo = sig;
1124 q->info.si_errno = 0;
1125 q->info.si_code = SI_USER;
1126 q->info.si_pid = task_tgid_nr_ns(current,
1127 task_active_pid_ns(t));
1128 rcu_read_lock();
1129 q->info.si_uid =
1130 from_kuid_munged(task_cred_xxx(t, user_ns),
1131 current_uid());
1132 rcu_read_unlock();
1133 break;
1134 case (unsigned long) SEND_SIG_PRIV:
1135 clear_siginfo(&q->info);
1136 q->info.si_signo = sig;
1137 q->info.si_errno = 0;
1138 q->info.si_code = SI_KERNEL;
1139 q->info.si_pid = 0;
1140 q->info.si_uid = 0;
1141 break;
1142 default:
1143 copy_siginfo(&q->info, info);
1144 break;
1145 }
1146 } else if (!is_si_special(info) &&
1147 sig >= SIGRTMIN && info->si_code != SI_USER) {
1148 /*
1149 * Queue overflow, abort. We may abort if the
1150 * signal was rt and sent by user using something
1151 * other than kill().
1152 */
1153 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1154 ret = -EAGAIN;
1155 goto ret;
1156 } else {
1157 /*
1158 * This is a silent loss of information. We still
1159 * send the signal, but the *info bits are lost.
1160 */
1161 result = TRACE_SIGNAL_LOSE_INFO;
1162 }
1163
1164 out_set:
1165 signalfd_notify(t, sig);
1166 sigaddset(&pending->signal, sig);
1167
1168 /* Let multiprocess signals appear after on-going forks */
1169 if (type > PIDTYPE_TGID) {
1170 struct multiprocess_signals *delayed;
1171 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1172 sigset_t *signal = &delayed->signal;
1173 /* Can't queue both a stop and a continue signal */
1174 if (sig == SIGCONT)
1175 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1176 else if (sig_kernel_stop(sig))
1177 sigdelset(signal, SIGCONT);
1178 sigaddset(signal, sig);
1179 }
1180 }
1181
1182 complete_signal(sig, t, type);
1183 ret:
1184 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1185 return ret;
1186 }
1187
1188 static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1189 {
1190 bool ret = false;
1191 switch (siginfo_layout(info->si_signo, info->si_code)) {
1192 case SIL_KILL:
1193 case SIL_CHLD:
1194 case SIL_RT:
1195 ret = true;
1196 break;
1197 case SIL_TIMER:
1198 case SIL_POLL:
1199 case SIL_FAULT:
1200 case SIL_FAULT_TRAPNO:
1201 case SIL_FAULT_MCEERR:
1202 case SIL_FAULT_BNDERR:
1203 case SIL_FAULT_PKUERR:
1204 case SIL_PERF_EVENT:
1205 case SIL_SYS:
1206 ret = false;
1207 break;
1208 }
1209 return ret;
1210 }
1211
1212 static int send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1213 enum pid_type type)
1214 {
1215 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1216 bool force = false;
1217
1218 if (info == SEND_SIG_NOINFO) {
1219 /* Force if sent from an ancestor pid namespace */
1220 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1221 } else if (info == SEND_SIG_PRIV) {
1222 /* Don't ignore kernel generated signals */
1223 force = true;
1224 } else if (has_si_pid_and_uid(info)) {
1225 /* SIGKILL and SIGSTOP is special or has ids */
1226 struct user_namespace *t_user_ns;
1227
1228 rcu_read_lock();
1229 t_user_ns = task_cred_xxx(t, user_ns);
1230 if (current_user_ns() != t_user_ns) {
1231 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1232 info->si_uid = from_kuid_munged(t_user_ns, uid);
1233 }
1234 rcu_read_unlock();
1235
1236 /* A kernel generated signal? */
1237 force = (info->si_code == SI_KERNEL);
1238
1239 /* From an ancestor pid namespace? */
1240 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1241 info->si_pid = 0;
1242 force = true;
1243 }
1244 }
1245 return __send_signal(sig, info, t, type, force);
1246 }
1247
1248 static void print_fatal_signal(int signr)
1249 {
1250 struct pt_regs *regs = signal_pt_regs();
1251 pr_info("potentially unexpected fatal signal %d.\n", signr);
1252
1253 #if defined(__i386__) && !defined(__arch_um__)
1254 pr_info("code at %08lx: ", regs->ip);
1255 {
1256 int i;
1257 for (i = 0; i < 16; i++) {
1258 unsigned char insn;
1259
1260 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1261 break;
1262 pr_cont("%02x ", insn);
1263 }
1264 }
1265 pr_cont("\n");
1266 #endif
1267 preempt_disable();
1268 show_regs(regs);
1269 preempt_enable();
1270 }
1271
1272 static int __init setup_print_fatal_signals(char *str)
1273 {
1274 get_option (&str, &print_fatal_signals);
1275
1276 return 1;
1277 }
1278
1279 __setup("print-fatal-signals=", setup_print_fatal_signals);
1280
1281 int
1282 __group_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1283 {
1284 return send_signal(sig, info, p, PIDTYPE_TGID);
1285 }
1286
1287 int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1288 enum pid_type type)
1289 {
1290 unsigned long flags;
1291 int ret = -ESRCH;
1292
1293 if (lock_task_sighand(p, &flags)) {
1294 ret = send_signal(sig, info, p, type);
1295 unlock_task_sighand(p, &flags);
1296 }
1297
1298 return ret;
1299 }
1300
1301 /*
1302 * Force a signal that the process can't ignore: if necessary
1303 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1304 *
1305 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1306 * since we do not want to have a signal handler that was blocked
1307 * be invoked when user space had explicitly blocked it.
1308 *
1309 * We don't want to have recursive SIGSEGV's etc, for example,
1310 * that is why we also clear SIGNAL_UNKILLABLE.
1311 */
1312 static int
1313 force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t)
1314 {
1315 unsigned long int flags;
1316 int ret, blocked, ignored;
1317 struct k_sigaction *action;
1318 int sig = info->si_signo;
1319
1320 spin_lock_irqsave(&t->sighand->siglock, flags);
1321 action = &t->sighand->action[sig-1];
1322 ignored = action->sa.sa_handler == SIG_IGN;
1323 blocked = sigismember(&t->blocked, sig);
1324 if (blocked || ignored) {
1325 action->sa.sa_handler = SIG_DFL;
1326 if (blocked) {
1327 sigdelset(&t->blocked, sig);
1328 recalc_sigpending_and_wake(t);
1329 }
1330 }
1331 /*
1332 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1333 * debugging to leave init killable.
1334 */
1335 if (action->sa.sa_handler == SIG_DFL && !t->ptrace)
1336 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1337 ret = send_signal(sig, info, t, PIDTYPE_PID);
1338 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1339
1340 return ret;
1341 }
1342
1343 int force_sig_info(struct kernel_siginfo *info)
1344 {
1345 return force_sig_info_to_task(info, current);
1346 }
1347
1348 /*
1349 * Nuke all other threads in the group.
1350 */
1351 int zap_other_threads(struct task_struct *p)
1352 {
1353 struct task_struct *t = p;
1354 int count = 0;
1355
1356 p->signal->group_stop_count = 0;
1357
1358 while_each_thread(p, t) {
1359 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1360 count++;
1361
1362 /* Don't bother with already dead threads */
1363 if (t->exit_state)
1364 continue;
1365 sigaddset(&t->pending.signal, SIGKILL);
1366 signal_wake_up(t, 1);
1367 }
1368
1369 return count;
1370 }
1371
1372 struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1373 unsigned long *flags)
1374 {
1375 struct sighand_struct *sighand;
1376
1377 rcu_read_lock();
1378 for (;;) {
1379 sighand = rcu_dereference(tsk->sighand);
1380 if (unlikely(sighand == NULL))
1381 break;
1382
1383 /*
1384 * This sighand can be already freed and even reused, but
1385 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1386 * initializes ->siglock: this slab can't go away, it has
1387 * the same object type, ->siglock can't be reinitialized.
1388 *
1389 * We need to ensure that tsk->sighand is still the same
1390 * after we take the lock, we can race with de_thread() or
1391 * __exit_signal(). In the latter case the next iteration
1392 * must see ->sighand == NULL.
1393 */
1394 spin_lock_irqsave(&sighand->siglock, *flags);
1395 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1396 break;
1397 spin_unlock_irqrestore(&sighand->siglock, *flags);
1398 }
1399 rcu_read_unlock();
1400
1401 return sighand;
1402 }
1403
1404 /*
1405 * send signal info to all the members of a group
1406 */
1407 int group_send_sig_info(int sig, struct kernel_siginfo *info,
1408 struct task_struct *p, enum pid_type type)
1409 {
1410 int ret;
1411
1412 rcu_read_lock();
1413 ret = check_kill_permission(sig, info, p);
1414 rcu_read_unlock();
1415
1416 if (!ret && sig)
1417 ret = do_send_sig_info(sig, info, p, type);
1418
1419 return ret;
1420 }
1421
1422 /*
1423 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1424 * control characters do (^C, ^Z etc)
1425 * - the caller must hold at least a readlock on tasklist_lock
1426 */
1427 int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1428 {
1429 struct task_struct *p = NULL;
1430 int retval, success;
1431
1432 success = 0;
1433 retval = -ESRCH;
1434 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1435 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1436 success |= !err;
1437 retval = err;
1438 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1439 return success ? 0 : retval;
1440 }
1441
1442 int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1443 {
1444 int error = -ESRCH;
1445 struct task_struct *p;
1446
1447 for (;;) {
1448 rcu_read_lock();
1449 p = pid_task(pid, PIDTYPE_PID);
1450 if (p)
1451 error = group_send_sig_info(sig, info, p, PIDTYPE_TGID);
1452 rcu_read_unlock();
1453 if (likely(!p || error != -ESRCH))
1454 return error;
1455
1456 /*
1457 * The task was unhashed in between, try again. If it
1458 * is dead, pid_task() will return NULL, if we race with
1459 * de_thread() it will find the new leader.
1460 */
1461 }
1462 }
1463
1464 static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1465 {
1466 int error;
1467 rcu_read_lock();
1468 error = kill_pid_info(sig, info, find_vpid(pid));
1469 rcu_read_unlock();
1470 return error;
1471 }
1472
1473 static inline bool kill_as_cred_perm(const struct cred *cred,
1474 struct task_struct *target)
1475 {
1476 const struct cred *pcred = __task_cred(target);
1477
1478 return uid_eq(cred->euid, pcred->suid) ||
1479 uid_eq(cred->euid, pcred->uid) ||
1480 uid_eq(cred->uid, pcred->suid) ||
1481 uid_eq(cred->uid, pcred->uid);
1482 }
1483
1484 /*
1485 * The usb asyncio usage of siginfo is wrong. The glibc support
1486 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1487 * AKA after the generic fields:
1488 * kernel_pid_t si_pid;
1489 * kernel_uid32_t si_uid;
1490 * sigval_t si_value;
1491 *
1492 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1493 * after the generic fields is:
1494 * void __user *si_addr;
1495 *
1496 * This is a practical problem when there is a 64bit big endian kernel
1497 * and a 32bit userspace. As the 32bit address will encoded in the low
1498 * 32bits of the pointer. Those low 32bits will be stored at higher
1499 * address than appear in a 32 bit pointer. So userspace will not
1500 * see the address it was expecting for it's completions.
1501 *
1502 * There is nothing in the encoding that can allow
1503 * copy_siginfo_to_user32 to detect this confusion of formats, so
1504 * handle this by requiring the caller of kill_pid_usb_asyncio to
1505 * notice when this situration takes place and to store the 32bit
1506 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1507 * parameter.
1508 */
1509 int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1510 struct pid *pid, const struct cred *cred)
1511 {
1512 struct kernel_siginfo info;
1513 struct task_struct *p;
1514 unsigned long flags;
1515 int ret = -EINVAL;
1516
1517 if (!valid_signal(sig))
1518 return ret;
1519
1520 clear_siginfo(&info);
1521 info.si_signo = sig;
1522 info.si_errno = errno;
1523 info.si_code = SI_ASYNCIO;
1524 *((sigval_t *)&info.si_pid) = addr;
1525
1526 rcu_read_lock();
1527 p = pid_task(pid, PIDTYPE_PID);
1528 if (!p) {
1529 ret = -ESRCH;
1530 goto out_unlock;
1531 }
1532 if (!kill_as_cred_perm(cred, p)) {
1533 ret = -EPERM;
1534 goto out_unlock;
1535 }
1536 ret = security_task_kill(p, &info, sig, cred);
1537 if (ret)
1538 goto out_unlock;
1539
1540 if (sig) {
1541 if (lock_task_sighand(p, &flags)) {
1542 ret = __send_signal(sig, &info, p, PIDTYPE_TGID, false);
1543 unlock_task_sighand(p, &flags);
1544 } else
1545 ret = -ESRCH;
1546 }
1547 out_unlock:
1548 rcu_read_unlock();
1549 return ret;
1550 }
1551 EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1552
1553 /*
1554 * kill_something_info() interprets pid in interesting ways just like kill(2).
1555 *
1556 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1557 * is probably wrong. Should make it like BSD or SYSV.
1558 */
1559
1560 static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1561 {
1562 int ret;
1563
1564 if (pid > 0)
1565 return kill_proc_info(sig, info, pid);
1566
1567 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1568 if (pid == INT_MIN)
1569 return -ESRCH;
1570
1571 read_lock(&tasklist_lock);
1572 if (pid != -1) {
1573 ret = __kill_pgrp_info(sig, info,
1574 pid ? find_vpid(-pid) : task_pgrp(current));
1575 } else {
1576 int retval = 0, count = 0;
1577 struct task_struct * p;
1578
1579 for_each_process(p) {
1580 if (task_pid_vnr(p) > 1 &&
1581 !same_thread_group(p, current)) {
1582 int err = group_send_sig_info(sig, info, p,
1583 PIDTYPE_MAX);
1584 ++count;
1585 if (err != -EPERM)
1586 retval = err;
1587 }
1588 }
1589 ret = count ? retval : -ESRCH;
1590 }
1591 read_unlock(&tasklist_lock);
1592
1593 return ret;
1594 }
1595
1596 /*
1597 * These are for backward compatibility with the rest of the kernel source.
1598 */
1599
1600 int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1601 {
1602 /*
1603 * Make sure legacy kernel users don't send in bad values
1604 * (normal paths check this in check_kill_permission).
1605 */
1606 if (!valid_signal(sig))
1607 return -EINVAL;
1608
1609 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1610 }
1611 EXPORT_SYMBOL(send_sig_info);
1612
1613 #define __si_special(priv) \
1614 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1615
1616 int
1617 send_sig(int sig, struct task_struct *p, int priv)
1618 {
1619 return send_sig_info(sig, __si_special(priv), p);
1620 }
1621 EXPORT_SYMBOL(send_sig);
1622
1623 void force_sig(int sig)
1624 {
1625 struct kernel_siginfo info;
1626
1627 clear_siginfo(&info);
1628 info.si_signo = sig;
1629 info.si_errno = 0;
1630 info.si_code = SI_KERNEL;
1631 info.si_pid = 0;
1632 info.si_uid = 0;
1633 force_sig_info(&info);
1634 }
1635 EXPORT_SYMBOL(force_sig);
1636
1637 /*
1638 * When things go south during signal handling, we
1639 * will force a SIGSEGV. And if the signal that caused
1640 * the problem was already a SIGSEGV, we'll want to
1641 * make sure we don't even try to deliver the signal..
1642 */
1643 void force_sigsegv(int sig)
1644 {
1645 struct task_struct *p = current;
1646
1647 if (sig == SIGSEGV) {
1648 unsigned long flags;
1649 spin_lock_irqsave(&p->sighand->siglock, flags);
1650 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1651 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1652 }
1653 force_sig(SIGSEGV);
1654 }
1655
1656 int force_sig_fault_to_task(int sig, int code, void __user *addr
1657 ___ARCH_SI_TRAPNO(int trapno)
1658 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1659 , struct task_struct *t)
1660 {
1661 struct kernel_siginfo info;
1662
1663 clear_siginfo(&info);
1664 info.si_signo = sig;
1665 info.si_errno = 0;
1666 info.si_code = code;
1667 info.si_addr = addr;
1668 #ifdef __ARCH_SI_TRAPNO
1669 info.si_trapno = trapno;
1670 #endif
1671 #ifdef __ia64__
1672 info.si_imm = imm;
1673 info.si_flags = flags;
1674 info.si_isr = isr;
1675 #endif
1676 return force_sig_info_to_task(&info, t);
1677 }
1678
1679 int force_sig_fault(int sig, int code, void __user *addr
1680 ___ARCH_SI_TRAPNO(int trapno)
1681 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr))
1682 {
1683 return force_sig_fault_to_task(sig, code, addr
1684 ___ARCH_SI_TRAPNO(trapno)
1685 ___ARCH_SI_IA64(imm, flags, isr), current);
1686 }
1687
1688 int send_sig_fault(int sig, int code, void __user *addr
1689 ___ARCH_SI_TRAPNO(int trapno)
1690 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1691 , struct task_struct *t)
1692 {
1693 struct kernel_siginfo info;
1694
1695 clear_siginfo(&info);
1696 info.si_signo = sig;
1697 info.si_errno = 0;
1698 info.si_code = code;
1699 info.si_addr = addr;
1700 #ifdef __ARCH_SI_TRAPNO
1701 info.si_trapno = trapno;
1702 #endif
1703 #ifdef __ia64__
1704 info.si_imm = imm;
1705 info.si_flags = flags;
1706 info.si_isr = isr;
1707 #endif
1708 return send_sig_info(info.si_signo, &info, t);
1709 }
1710
1711 int force_sig_mceerr(int code, void __user *addr, short lsb)
1712 {
1713 struct kernel_siginfo info;
1714
1715 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1716 clear_siginfo(&info);
1717 info.si_signo = SIGBUS;
1718 info.si_errno = 0;
1719 info.si_code = code;
1720 info.si_addr = addr;
1721 info.si_addr_lsb = lsb;
1722 return force_sig_info(&info);
1723 }
1724
1725 int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1726 {
1727 struct kernel_siginfo info;
1728
1729 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1730 clear_siginfo(&info);
1731 info.si_signo = SIGBUS;
1732 info.si_errno = 0;
1733 info.si_code = code;
1734 info.si_addr = addr;
1735 info.si_addr_lsb = lsb;
1736 return send_sig_info(info.si_signo, &info, t);
1737 }
1738 EXPORT_SYMBOL(send_sig_mceerr);
1739
1740 int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1741 {
1742 struct kernel_siginfo info;
1743
1744 clear_siginfo(&info);
1745 info.si_signo = SIGSEGV;
1746 info.si_errno = 0;
1747 info.si_code = SEGV_BNDERR;
1748 info.si_addr = addr;
1749 info.si_lower = lower;
1750 info.si_upper = upper;
1751 return force_sig_info(&info);
1752 }
1753
1754 #ifdef SEGV_PKUERR
1755 int force_sig_pkuerr(void __user *addr, u32 pkey)
1756 {
1757 struct kernel_siginfo info;
1758
1759 clear_siginfo(&info);
1760 info.si_signo = SIGSEGV;
1761 info.si_errno = 0;
1762 info.si_code = SEGV_PKUERR;
1763 info.si_addr = addr;
1764 info.si_pkey = pkey;
1765 return force_sig_info(&info);
1766 }
1767 #endif
1768
1769 int force_sig_perf(void __user *addr, u32 type, u64 sig_data)
1770 {
1771 struct kernel_siginfo info;
1772
1773 clear_siginfo(&info);
1774 info.si_signo = SIGTRAP;
1775 info.si_errno = 0;
1776 info.si_code = TRAP_PERF;
1777 info.si_addr = addr;
1778 info.si_perf_data = sig_data;
1779 info.si_perf_type = type;
1780
1781 return force_sig_info(&info);
1782 }
1783
1784 /* For the crazy architectures that include trap information in
1785 * the errno field, instead of an actual errno value.
1786 */
1787 int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1788 {
1789 struct kernel_siginfo info;
1790
1791 clear_siginfo(&info);
1792 info.si_signo = SIGTRAP;
1793 info.si_errno = errno;
1794 info.si_code = TRAP_HWBKPT;
1795 info.si_addr = addr;
1796 return force_sig_info(&info);
1797 }
1798
1799 int kill_pgrp(struct pid *pid, int sig, int priv)
1800 {
1801 int ret;
1802
1803 read_lock(&tasklist_lock);
1804 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1805 read_unlock(&tasklist_lock);
1806
1807 return ret;
1808 }
1809 EXPORT_SYMBOL(kill_pgrp);
1810
1811 int kill_pid(struct pid *pid, int sig, int priv)
1812 {
1813 return kill_pid_info(sig, __si_special(priv), pid);
1814 }
1815 EXPORT_SYMBOL(kill_pid);
1816
1817 /*
1818 * These functions support sending signals using preallocated sigqueue
1819 * structures. This is needed "because realtime applications cannot
1820 * afford to lose notifications of asynchronous events, like timer
1821 * expirations or I/O completions". In the case of POSIX Timers
1822 * we allocate the sigqueue structure from the timer_create. If this
1823 * allocation fails we are able to report the failure to the application
1824 * with an EAGAIN error.
1825 */
1826 struct sigqueue *sigqueue_alloc(void)
1827 {
1828 return __sigqueue_alloc(-1, current, GFP_KERNEL, 0, SIGQUEUE_PREALLOC);
1829 }
1830
1831 void sigqueue_free(struct sigqueue *q)
1832 {
1833 unsigned long flags;
1834 spinlock_t *lock = &current->sighand->siglock;
1835
1836 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1837 /*
1838 * We must hold ->siglock while testing q->list
1839 * to serialize with collect_signal() or with
1840 * __exit_signal()->flush_sigqueue().
1841 */
1842 spin_lock_irqsave(lock, flags);
1843 q->flags &= ~SIGQUEUE_PREALLOC;
1844 /*
1845 * If it is queued it will be freed when dequeued,
1846 * like the "regular" sigqueue.
1847 */
1848 if (!list_empty(&q->list))
1849 q = NULL;
1850 spin_unlock_irqrestore(lock, flags);
1851
1852 if (q)
1853 __sigqueue_free(q);
1854 }
1855
1856 int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1857 {
1858 int sig = q->info.si_signo;
1859 struct sigpending *pending;
1860 struct task_struct *t;
1861 unsigned long flags;
1862 int ret, result;
1863
1864 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1865
1866 ret = -1;
1867 rcu_read_lock();
1868 t = pid_task(pid, type);
1869 if (!t || !likely(lock_task_sighand(t, &flags)))
1870 goto ret;
1871
1872 ret = 1; /* the signal is ignored */
1873 result = TRACE_SIGNAL_IGNORED;
1874 if (!prepare_signal(sig, t, false))
1875 goto out;
1876
1877 ret = 0;
1878 if (unlikely(!list_empty(&q->list))) {
1879 /*
1880 * If an SI_TIMER entry is already queue just increment
1881 * the overrun count.
1882 */
1883 BUG_ON(q->info.si_code != SI_TIMER);
1884 q->info.si_overrun++;
1885 result = TRACE_SIGNAL_ALREADY_PENDING;
1886 goto out;
1887 }
1888 q->info.si_overrun = 0;
1889
1890 signalfd_notify(t, sig);
1891 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1892 list_add_tail(&q->list, &pending->list);
1893 sigaddset(&pending->signal, sig);
1894 complete_signal(sig, t, type);
1895 result = TRACE_SIGNAL_DELIVERED;
1896 out:
1897 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
1898 unlock_task_sighand(t, &flags);
1899 ret:
1900 rcu_read_unlock();
1901 return ret;
1902 }
1903
1904 static void do_notify_pidfd(struct task_struct *task)
1905 {
1906 struct pid *pid;
1907
1908 WARN_ON(task->exit_state == 0);
1909 pid = task_pid(task);
1910 wake_up_all(&pid->wait_pidfd);
1911 }
1912
1913 /*
1914 * Let a parent know about the death of a child.
1915 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1916 *
1917 * Returns true if our parent ignored us and so we've switched to
1918 * self-reaping.
1919 */
1920 bool do_notify_parent(struct task_struct *tsk, int sig)
1921 {
1922 struct kernel_siginfo info;
1923 unsigned long flags;
1924 struct sighand_struct *psig;
1925 bool autoreap = false;
1926 u64 utime, stime;
1927
1928 BUG_ON(sig == -1);
1929
1930 /* do_notify_parent_cldstop should have been called instead. */
1931 BUG_ON(task_is_stopped_or_traced(tsk));
1932
1933 BUG_ON(!tsk->ptrace &&
1934 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1935
1936 /* Wake up all pidfd waiters */
1937 do_notify_pidfd(tsk);
1938
1939 if (sig != SIGCHLD) {
1940 /*
1941 * This is only possible if parent == real_parent.
1942 * Check if it has changed security domain.
1943 */
1944 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
1945 sig = SIGCHLD;
1946 }
1947
1948 clear_siginfo(&info);
1949 info.si_signo = sig;
1950 info.si_errno = 0;
1951 /*
1952 * We are under tasklist_lock here so our parent is tied to
1953 * us and cannot change.
1954 *
1955 * task_active_pid_ns will always return the same pid namespace
1956 * until a task passes through release_task.
1957 *
1958 * write_lock() currently calls preempt_disable() which is the
1959 * same as rcu_read_lock(), but according to Oleg, this is not
1960 * correct to rely on this
1961 */
1962 rcu_read_lock();
1963 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
1964 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
1965 task_uid(tsk));
1966 rcu_read_unlock();
1967
1968 task_cputime(tsk, &utime, &stime);
1969 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
1970 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
1971
1972 info.si_status = tsk->exit_code & 0x7f;
1973 if (tsk->exit_code & 0x80)
1974 info.si_code = CLD_DUMPED;
1975 else if (tsk->exit_code & 0x7f)
1976 info.si_code = CLD_KILLED;
1977 else {
1978 info.si_code = CLD_EXITED;
1979 info.si_status = tsk->exit_code >> 8;
1980 }
1981
1982 psig = tsk->parent->sighand;
1983 spin_lock_irqsave(&psig->siglock, flags);
1984 if (!tsk->ptrace && sig == SIGCHLD &&
1985 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1986 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1987 /*
1988 * We are exiting and our parent doesn't care. POSIX.1
1989 * defines special semantics for setting SIGCHLD to SIG_IGN
1990 * or setting the SA_NOCLDWAIT flag: we should be reaped
1991 * automatically and not left for our parent's wait4 call.
1992 * Rather than having the parent do it as a magic kind of
1993 * signal handler, we just set this to tell do_exit that we
1994 * can be cleaned up without becoming a zombie. Note that
1995 * we still call __wake_up_parent in this case, because a
1996 * blocked sys_wait4 might now return -ECHILD.
1997 *
1998 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1999 * is implementation-defined: we do (if you don't want
2000 * it, just use SIG_IGN instead).
2001 */
2002 autoreap = true;
2003 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2004 sig = 0;
2005 }
2006 /*
2007 * Send with __send_signal as si_pid and si_uid are in the
2008 * parent's namespaces.
2009 */
2010 if (valid_signal(sig) && sig)
2011 __send_signal(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2012 __wake_up_parent(tsk, tsk->parent);
2013 spin_unlock_irqrestore(&psig->siglock, flags);
2014
2015 return autoreap;
2016 }
2017
2018 /**
2019 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2020 * @tsk: task reporting the state change
2021 * @for_ptracer: the notification is for ptracer
2022 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2023 *
2024 * Notify @tsk's parent that the stopped/continued state has changed. If
2025 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2026 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2027 *
2028 * CONTEXT:
2029 * Must be called with tasklist_lock at least read locked.
2030 */
2031 static void do_notify_parent_cldstop(struct task_struct *tsk,
2032 bool for_ptracer, int why)
2033 {
2034 struct kernel_siginfo info;
2035 unsigned long flags;
2036 struct task_struct *parent;
2037 struct sighand_struct *sighand;
2038 u64 utime, stime;
2039
2040 if (for_ptracer) {
2041 parent = tsk->parent;
2042 } else {
2043 tsk = tsk->group_leader;
2044 parent = tsk->real_parent;
2045 }
2046
2047 clear_siginfo(&info);
2048 info.si_signo = SIGCHLD;
2049 info.si_errno = 0;
2050 /*
2051 * see comment in do_notify_parent() about the following 4 lines
2052 */
2053 rcu_read_lock();
2054 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2055 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2056 rcu_read_unlock();
2057
2058 task_cputime(tsk, &utime, &stime);
2059 info.si_utime = nsec_to_clock_t(utime);
2060 info.si_stime = nsec_to_clock_t(stime);
2061
2062 info.si_code = why;
2063 switch (why) {
2064 case CLD_CONTINUED:
2065 info.si_status = SIGCONT;
2066 break;
2067 case CLD_STOPPED:
2068 info.si_status = tsk->signal->group_exit_code & 0x7f;
2069 break;
2070 case CLD_TRAPPED:
2071 info.si_status = tsk->exit_code & 0x7f;
2072 break;
2073 default:
2074 BUG();
2075 }
2076
2077 sighand = parent->sighand;
2078 spin_lock_irqsave(&sighand->siglock, flags);
2079 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2080 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2081 __group_send_sig_info(SIGCHLD, &info, parent);
2082 /*
2083 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2084 */
2085 __wake_up_parent(tsk, parent);
2086 spin_unlock_irqrestore(&sighand->siglock, flags);
2087 }
2088
2089 static inline bool may_ptrace_stop(void)
2090 {
2091 if (!likely(current->ptrace))
2092 return false;
2093 /*
2094 * Are we in the middle of do_coredump?
2095 * If so and our tracer is also part of the coredump stopping
2096 * is a deadlock situation, and pointless because our tracer
2097 * is dead so don't allow us to stop.
2098 * If SIGKILL was already sent before the caller unlocked
2099 * ->siglock we must see ->core_state != NULL. Otherwise it
2100 * is safe to enter schedule().
2101 *
2102 * This is almost outdated, a task with the pending SIGKILL can't
2103 * block in TASK_TRACED. But PTRACE_EVENT_EXIT can be reported
2104 * after SIGKILL was already dequeued.
2105 */
2106 if (unlikely(current->mm->core_state) &&
2107 unlikely(current->mm == current->parent->mm))
2108 return false;
2109
2110 return true;
2111 }
2112
2113 /*
2114 * Return non-zero if there is a SIGKILL that should be waking us up.
2115 * Called with the siglock held.
2116 */
2117 static bool sigkill_pending(struct task_struct *tsk)
2118 {
2119 return sigismember(&tsk->pending.signal, SIGKILL) ||
2120 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
2121 }
2122
2123 /*
2124 * This must be called with current->sighand->siglock held.
2125 *
2126 * This should be the path for all ptrace stops.
2127 * We always set current->last_siginfo while stopped here.
2128 * That makes it a way to test a stopped process for
2129 * being ptrace-stopped vs being job-control-stopped.
2130 *
2131 * If we actually decide not to stop at all because the tracer
2132 * is gone, we keep current->exit_code unless clear_code.
2133 */
2134 static void ptrace_stop(int exit_code, int why, int clear_code, kernel_siginfo_t *info)
2135 __releases(&current->sighand->siglock)
2136 __acquires(&current->sighand->siglock)
2137 {
2138 bool gstop_done = false;
2139
2140 if (arch_ptrace_stop_needed(exit_code, info)) {
2141 /*
2142 * The arch code has something special to do before a
2143 * ptrace stop. This is allowed to block, e.g. for faults
2144 * on user stack pages. We can't keep the siglock while
2145 * calling arch_ptrace_stop, so we must release it now.
2146 * To preserve proper semantics, we must do this before
2147 * any signal bookkeeping like checking group_stop_count.
2148 * Meanwhile, a SIGKILL could come in before we retake the
2149 * siglock. That must prevent us from sleeping in TASK_TRACED.
2150 * So after regaining the lock, we must check for SIGKILL.
2151 */
2152 spin_unlock_irq(&current->sighand->siglock);
2153 arch_ptrace_stop(exit_code, info);
2154 spin_lock_irq(&current->sighand->siglock);
2155 if (sigkill_pending(current))
2156 return;
2157 }
2158
2159 set_special_state(TASK_TRACED);
2160
2161 /*
2162 * We're committing to trapping. TRACED should be visible before
2163 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2164 * Also, transition to TRACED and updates to ->jobctl should be
2165 * atomic with respect to siglock and should be done after the arch
2166 * hook as siglock is released and regrabbed across it.
2167 *
2168 * TRACER TRACEE
2169 *
2170 * ptrace_attach()
2171 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2172 * do_wait()
2173 * set_current_state() smp_wmb();
2174 * ptrace_do_wait()
2175 * wait_task_stopped()
2176 * task_stopped_code()
2177 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2178 */
2179 smp_wmb();
2180
2181 current->last_siginfo = info;
2182 current->exit_code = exit_code;
2183
2184 /*
2185 * If @why is CLD_STOPPED, we're trapping to participate in a group
2186 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2187 * across siglock relocks since INTERRUPT was scheduled, PENDING
2188 * could be clear now. We act as if SIGCONT is received after
2189 * TASK_TRACED is entered - ignore it.
2190 */
2191 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2192 gstop_done = task_participate_group_stop(current);
2193
2194 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2195 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2196 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2197 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2198
2199 /* entering a trap, clear TRAPPING */
2200 task_clear_jobctl_trapping(current);
2201
2202 spin_unlock_irq(&current->sighand->siglock);
2203 read_lock(&tasklist_lock);
2204 if (may_ptrace_stop()) {
2205 /*
2206 * Notify parents of the stop.
2207 *
2208 * While ptraced, there are two parents - the ptracer and
2209 * the real_parent of the group_leader. The ptracer should
2210 * know about every stop while the real parent is only
2211 * interested in the completion of group stop. The states
2212 * for the two don't interact with each other. Notify
2213 * separately unless they're gonna be duplicates.
2214 */
2215 do_notify_parent_cldstop(current, true, why);
2216 if (gstop_done && ptrace_reparented(current))
2217 do_notify_parent_cldstop(current, false, why);
2218
2219 /*
2220 * Don't want to allow preemption here, because
2221 * sys_ptrace() needs this task to be inactive.
2222 *
2223 * XXX: implement read_unlock_no_resched().
2224 */
2225 preempt_disable();
2226 read_unlock(&tasklist_lock);
2227 cgroup_enter_frozen();
2228 preempt_enable_no_resched();
2229 freezable_schedule();
2230 cgroup_leave_frozen(true);
2231 } else {
2232 /*
2233 * By the time we got the lock, our tracer went away.
2234 * Don't drop the lock yet, another tracer may come.
2235 *
2236 * If @gstop_done, the ptracer went away between group stop
2237 * completion and here. During detach, it would have set
2238 * JOBCTL_STOP_PENDING on us and we'll re-enter
2239 * TASK_STOPPED in do_signal_stop() on return, so notifying
2240 * the real parent of the group stop completion is enough.
2241 */
2242 if (gstop_done)
2243 do_notify_parent_cldstop(current, false, why);
2244
2245 /* tasklist protects us from ptrace_freeze_traced() */
2246 __set_current_state(TASK_RUNNING);
2247 if (clear_code)
2248 current->exit_code = 0;
2249 read_unlock(&tasklist_lock);
2250 }
2251
2252 /*
2253 * We are back. Now reacquire the siglock before touching
2254 * last_siginfo, so that we are sure to have synchronized with
2255 * any signal-sending on another CPU that wants to examine it.
2256 */
2257 spin_lock_irq(&current->sighand->siglock);
2258 current->last_siginfo = NULL;
2259
2260 /* LISTENING can be set only during STOP traps, clear it */
2261 current->jobctl &= ~JOBCTL_LISTENING;
2262
2263 /*
2264 * Queued signals ignored us while we were stopped for tracing.
2265 * So check for any that we should take before resuming user mode.
2266 * This sets TIF_SIGPENDING, but never clears it.
2267 */
2268 recalc_sigpending_tsk(current);
2269 }
2270
2271 static void ptrace_do_notify(int signr, int exit_code, int why)
2272 {
2273 kernel_siginfo_t info;
2274
2275 clear_siginfo(&info);
2276 info.si_signo = signr;
2277 info.si_code = exit_code;
2278 info.si_pid = task_pid_vnr(current);
2279 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2280
2281 /* Let the debugger run. */
2282 ptrace_stop(exit_code, why, 1, &info);
2283 }
2284
2285 void ptrace_notify(int exit_code)
2286 {
2287 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2288 if (unlikely(current->task_works))
2289 task_work_run();
2290
2291 spin_lock_irq(&current->sighand->siglock);
2292 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
2293 spin_unlock_irq(&current->sighand->siglock);
2294 }
2295
2296 /**
2297 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2298 * @signr: signr causing group stop if initiating
2299 *
2300 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2301 * and participate in it. If already set, participate in the existing
2302 * group stop. If participated in a group stop (and thus slept), %true is
2303 * returned with siglock released.
2304 *
2305 * If ptraced, this function doesn't handle stop itself. Instead,
2306 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2307 * untouched. The caller must ensure that INTERRUPT trap handling takes
2308 * places afterwards.
2309 *
2310 * CONTEXT:
2311 * Must be called with @current->sighand->siglock held, which is released
2312 * on %true return.
2313 *
2314 * RETURNS:
2315 * %false if group stop is already cancelled or ptrace trap is scheduled.
2316 * %true if participated in group stop.
2317 */
2318 static bool do_signal_stop(int signr)
2319 __releases(&current->sighand->siglock)
2320 {
2321 struct signal_struct *sig = current->signal;
2322
2323 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2324 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2325 struct task_struct *t;
2326
2327 /* signr will be recorded in task->jobctl for retries */
2328 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2329
2330 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2331 unlikely(signal_group_exit(sig)))
2332 return false;
2333 /*
2334 * There is no group stop already in progress. We must
2335 * initiate one now.
2336 *
2337 * While ptraced, a task may be resumed while group stop is
2338 * still in effect and then receive a stop signal and
2339 * initiate another group stop. This deviates from the
2340 * usual behavior as two consecutive stop signals can't
2341 * cause two group stops when !ptraced. That is why we
2342 * also check !task_is_stopped(t) below.
2343 *
2344 * The condition can be distinguished by testing whether
2345 * SIGNAL_STOP_STOPPED is already set. Don't generate
2346 * group_exit_code in such case.
2347 *
2348 * This is not necessary for SIGNAL_STOP_CONTINUED because
2349 * an intervening stop signal is required to cause two
2350 * continued events regardless of ptrace.
2351 */
2352 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2353 sig->group_exit_code = signr;
2354
2355 sig->group_stop_count = 0;
2356
2357 if (task_set_jobctl_pending(current, signr | gstop))
2358 sig->group_stop_count++;
2359
2360 t = current;
2361 while_each_thread(current, t) {
2362 /*
2363 * Setting state to TASK_STOPPED for a group
2364 * stop is always done with the siglock held,
2365 * so this check has no races.
2366 */
2367 if (!task_is_stopped(t) &&
2368 task_set_jobctl_pending(t, signr | gstop)) {
2369 sig->group_stop_count++;
2370 if (likely(!(t->ptrace & PT_SEIZED)))
2371 signal_wake_up(t, 0);
2372 else
2373 ptrace_trap_notify(t);
2374 }
2375 }
2376 }
2377
2378 if (likely(!current->ptrace)) {
2379 int notify = 0;
2380
2381 /*
2382 * If there are no other threads in the group, or if there
2383 * is a group stop in progress and we are the last to stop,
2384 * report to the parent.
2385 */
2386 if (task_participate_group_stop(current))
2387 notify = CLD_STOPPED;
2388
2389 set_special_state(TASK_STOPPED);
2390 spin_unlock_irq(&current->sighand->siglock);
2391
2392 /*
2393 * Notify the parent of the group stop completion. Because
2394 * we're not holding either the siglock or tasklist_lock
2395 * here, ptracer may attach inbetween; however, this is for
2396 * group stop and should always be delivered to the real
2397 * parent of the group leader. The new ptracer will get
2398 * its notification when this task transitions into
2399 * TASK_TRACED.
2400 */
2401 if (notify) {
2402 read_lock(&tasklist_lock);
2403 do_notify_parent_cldstop(current, false, notify);
2404 read_unlock(&tasklist_lock);
2405 }
2406
2407 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2408 cgroup_enter_frozen();
2409 freezable_schedule();
2410 return true;
2411 } else {
2412 /*
2413 * While ptraced, group stop is handled by STOP trap.
2414 * Schedule it and let the caller deal with it.
2415 */
2416 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2417 return false;
2418 }
2419 }
2420
2421 /**
2422 * do_jobctl_trap - take care of ptrace jobctl traps
2423 *
2424 * When PT_SEIZED, it's used for both group stop and explicit
2425 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2426 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2427 * the stop signal; otherwise, %SIGTRAP.
2428 *
2429 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2430 * number as exit_code and no siginfo.
2431 *
2432 * CONTEXT:
2433 * Must be called with @current->sighand->siglock held, which may be
2434 * released and re-acquired before returning with intervening sleep.
2435 */
2436 static void do_jobctl_trap(void)
2437 {
2438 struct signal_struct *signal = current->signal;
2439 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2440
2441 if (current->ptrace & PT_SEIZED) {
2442 if (!signal->group_stop_count &&
2443 !(signal->flags & SIGNAL_STOP_STOPPED))
2444 signr = SIGTRAP;
2445 WARN_ON_ONCE(!signr);
2446 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2447 CLD_STOPPED);
2448 } else {
2449 WARN_ON_ONCE(!signr);
2450 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2451 current->exit_code = 0;
2452 }
2453 }
2454
2455 /**
2456 * do_freezer_trap - handle the freezer jobctl trap
2457 *
2458 * Puts the task into frozen state, if only the task is not about to quit.
2459 * In this case it drops JOBCTL_TRAP_FREEZE.
2460 *
2461 * CONTEXT:
2462 * Must be called with @current->sighand->siglock held,
2463 * which is always released before returning.
2464 */
2465 static void do_freezer_trap(void)
2466 __releases(&current->sighand->siglock)
2467 {
2468 /*
2469 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2470 * let's make another loop to give it a chance to be handled.
2471 * In any case, we'll return back.
2472 */
2473 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2474 JOBCTL_TRAP_FREEZE) {
2475 spin_unlock_irq(&current->sighand->siglock);
2476 return;
2477 }
2478
2479 /*
2480 * Now we're sure that there is no pending fatal signal and no
2481 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2482 * immediately (if there is a non-fatal signal pending), and
2483 * put the task into sleep.
2484 */
2485 __set_current_state(TASK_INTERRUPTIBLE);
2486 clear_thread_flag(TIF_SIGPENDING);
2487 spin_unlock_irq(&current->sighand->siglock);
2488 cgroup_enter_frozen();
2489 freezable_schedule();
2490 }
2491
2492 static int ptrace_signal(int signr, kernel_siginfo_t *info)
2493 {
2494 /*
2495 * We do not check sig_kernel_stop(signr) but set this marker
2496 * unconditionally because we do not know whether debugger will
2497 * change signr. This flag has no meaning unless we are going
2498 * to stop after return from ptrace_stop(). In this case it will
2499 * be checked in do_signal_stop(), we should only stop if it was
2500 * not cleared by SIGCONT while we were sleeping. See also the
2501 * comment in dequeue_signal().
2502 */
2503 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2504 ptrace_stop(signr, CLD_TRAPPED, 0, info);
2505
2506 /* We're back. Did the debugger cancel the sig? */
2507 signr = current->exit_code;
2508 if (signr == 0)
2509 return signr;
2510
2511 current->exit_code = 0;
2512
2513 /*
2514 * Update the siginfo structure if the signal has
2515 * changed. If the debugger wanted something
2516 * specific in the siginfo structure then it should
2517 * have updated *info via PTRACE_SETSIGINFO.
2518 */
2519 if (signr != info->si_signo) {
2520 clear_siginfo(info);
2521 info->si_signo = signr;
2522 info->si_errno = 0;
2523 info->si_code = SI_USER;
2524 rcu_read_lock();
2525 info->si_pid = task_pid_vnr(current->parent);
2526 info->si_uid = from_kuid_munged(current_user_ns(),
2527 task_uid(current->parent));
2528 rcu_read_unlock();
2529 }
2530
2531 /* If the (new) signal is now blocked, requeue it. */
2532 if (sigismember(&current->blocked, signr)) {
2533 send_signal(signr, info, current, PIDTYPE_PID);
2534 signr = 0;
2535 }
2536
2537 return signr;
2538 }
2539
2540 static void hide_si_addr_tag_bits(struct ksignal *ksig)
2541 {
2542 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2543 case SIL_FAULT:
2544 case SIL_FAULT_TRAPNO:
2545 case SIL_FAULT_MCEERR:
2546 case SIL_FAULT_BNDERR:
2547 case SIL_FAULT_PKUERR:
2548 case SIL_PERF_EVENT:
2549 ksig->info.si_addr = arch_untagged_si_addr(
2550 ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2551 break;
2552 case SIL_KILL:
2553 case SIL_TIMER:
2554 case SIL_POLL:
2555 case SIL_CHLD:
2556 case SIL_RT:
2557 case SIL_SYS:
2558 break;
2559 }
2560 }
2561
2562 bool get_signal(struct ksignal *ksig)
2563 {
2564 struct sighand_struct *sighand = current->sighand;
2565 struct signal_struct *signal = current->signal;
2566 int signr;
2567
2568 if (unlikely(current->task_works))
2569 task_work_run();
2570
2571 /*
2572 * For non-generic architectures, check for TIF_NOTIFY_SIGNAL so
2573 * that the arch handlers don't all have to do it. If we get here
2574 * without TIF_SIGPENDING, just exit after running signal work.
2575 */
2576 if (!IS_ENABLED(CONFIG_GENERIC_ENTRY)) {
2577 if (test_thread_flag(TIF_NOTIFY_SIGNAL))
2578 tracehook_notify_signal();
2579 if (!task_sigpending(current))
2580 return false;
2581 }
2582
2583 if (unlikely(uprobe_deny_signal()))
2584 return false;
2585
2586 /*
2587 * Do this once, we can't return to user-mode if freezing() == T.
2588 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2589 * thus do not need another check after return.
2590 */
2591 try_to_freeze();
2592
2593 relock:
2594 spin_lock_irq(&sighand->siglock);
2595
2596 /*
2597 * Every stopped thread goes here after wakeup. Check to see if
2598 * we should notify the parent, prepare_signal(SIGCONT) encodes
2599 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2600 */
2601 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2602 int why;
2603
2604 if (signal->flags & SIGNAL_CLD_CONTINUED)
2605 why = CLD_CONTINUED;
2606 else
2607 why = CLD_STOPPED;
2608
2609 signal->flags &= ~SIGNAL_CLD_MASK;
2610
2611 spin_unlock_irq(&sighand->siglock);
2612
2613 /*
2614 * Notify the parent that we're continuing. This event is
2615 * always per-process and doesn't make whole lot of sense
2616 * for ptracers, who shouldn't consume the state via
2617 * wait(2) either, but, for backward compatibility, notify
2618 * the ptracer of the group leader too unless it's gonna be
2619 * a duplicate.
2620 */
2621 read_lock(&tasklist_lock);
2622 do_notify_parent_cldstop(current, false, why);
2623
2624 if (ptrace_reparented(current->group_leader))
2625 do_notify_parent_cldstop(current->group_leader,
2626 true, why);
2627 read_unlock(&tasklist_lock);
2628
2629 goto relock;
2630 }
2631
2632 /* Has this task already been marked for death? */
2633 if (signal_group_exit(signal)) {
2634 ksig->info.si_signo = signr = SIGKILL;
2635 sigdelset(&current->pending.signal, SIGKILL);
2636 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2637 &sighand->action[SIGKILL - 1]);
2638 recalc_sigpending();
2639 goto fatal;
2640 }
2641
2642 for (;;) {
2643 struct k_sigaction *ka;
2644
2645 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2646 do_signal_stop(0))
2647 goto relock;
2648
2649 if (unlikely(current->jobctl &
2650 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2651 if (current->jobctl & JOBCTL_TRAP_MASK) {
2652 do_jobctl_trap();
2653 spin_unlock_irq(&sighand->siglock);
2654 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2655 do_freezer_trap();
2656
2657 goto relock;
2658 }
2659
2660 /*
2661 * If the task is leaving the frozen state, let's update
2662 * cgroup counters and reset the frozen bit.
2663 */
2664 if (unlikely(cgroup_task_frozen(current))) {
2665 spin_unlock_irq(&sighand->siglock);
2666 cgroup_leave_frozen(false);
2667 goto relock;
2668 }
2669
2670 /*
2671 * Signals generated by the execution of an instruction
2672 * need to be delivered before any other pending signals
2673 * so that the instruction pointer in the signal stack
2674 * frame points to the faulting instruction.
2675 */
2676 signr = dequeue_synchronous_signal(&ksig->info);
2677 if (!signr)
2678 signr = dequeue_signal(current, &current->blocked, &ksig->info);
2679
2680 if (!signr)
2681 break; /* will return 0 */
2682
2683 if (unlikely(current->ptrace) && signr != SIGKILL) {
2684 signr = ptrace_signal(signr, &ksig->info);
2685 if (!signr)
2686 continue;
2687 }
2688
2689 ka = &sighand->action[signr-1];
2690
2691 /* Trace actually delivered signals. */
2692 trace_signal_deliver(signr, &ksig->info, ka);
2693
2694 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2695 continue;
2696 if (ka->sa.sa_handler != SIG_DFL) {
2697 /* Run the handler. */
2698 ksig->ka = *ka;
2699
2700 if (ka->sa.sa_flags & SA_ONESHOT)
2701 ka->sa.sa_handler = SIG_DFL;
2702
2703 break; /* will return non-zero "signr" value */
2704 }
2705
2706 /*
2707 * Now we are doing the default action for this signal.
2708 */
2709 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2710 continue;
2711
2712 /*
2713 * Global init gets no signals it doesn't want.
2714 * Container-init gets no signals it doesn't want from same
2715 * container.
2716 *
2717 * Note that if global/container-init sees a sig_kernel_only()
2718 * signal here, the signal must have been generated internally
2719 * or must have come from an ancestor namespace. In either
2720 * case, the signal cannot be dropped.
2721 */
2722 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2723 !sig_kernel_only(signr))
2724 continue;
2725
2726 if (sig_kernel_stop(signr)) {
2727 /*
2728 * The default action is to stop all threads in
2729 * the thread group. The job control signals
2730 * do nothing in an orphaned pgrp, but SIGSTOP
2731 * always works. Note that siglock needs to be
2732 * dropped during the call to is_orphaned_pgrp()
2733 * because of lock ordering with tasklist_lock.
2734 * This allows an intervening SIGCONT to be posted.
2735 * We need to check for that and bail out if necessary.
2736 */
2737 if (signr != SIGSTOP) {
2738 spin_unlock_irq(&sighand->siglock);
2739
2740 /* signals can be posted during this window */
2741
2742 if (is_current_pgrp_orphaned())
2743 goto relock;
2744
2745 spin_lock_irq(&sighand->siglock);
2746 }
2747
2748 if (likely(do_signal_stop(ksig->info.si_signo))) {
2749 /* It released the siglock. */
2750 goto relock;
2751 }
2752
2753 /*
2754 * We didn't actually stop, due to a race
2755 * with SIGCONT or something like that.
2756 */
2757 continue;
2758 }
2759
2760 fatal:
2761 spin_unlock_irq(&sighand->siglock);
2762 if (unlikely(cgroup_task_frozen(current)))
2763 cgroup_leave_frozen(true);
2764
2765 /*
2766 * Anything else is fatal, maybe with a core dump.
2767 */
2768 current->flags |= PF_SIGNALED;
2769
2770 if (sig_kernel_coredump(signr)) {
2771 if (print_fatal_signals)
2772 print_fatal_signal(ksig->info.si_signo);
2773 proc_coredump_connector(current);
2774 /*
2775 * If it was able to dump core, this kills all
2776 * other threads in the group and synchronizes with
2777 * their demise. If we lost the race with another
2778 * thread getting here, it set group_exit_code
2779 * first and our do_group_exit call below will use
2780 * that value and ignore the one we pass it.
2781 */
2782 do_coredump(&ksig->info);
2783 }
2784
2785 /*
2786 * PF_IO_WORKER threads will catch and exit on fatal signals
2787 * themselves. They have cleanup that must be performed, so
2788 * we cannot call do_exit() on their behalf.
2789 */
2790 if (current->flags & PF_IO_WORKER)
2791 goto out;
2792
2793 /*
2794 * Death signals, no core dump.
2795 */
2796 do_group_exit(ksig->info.si_signo);
2797 /* NOTREACHED */
2798 }
2799 spin_unlock_irq(&sighand->siglock);
2800 out:
2801 ksig->sig = signr;
2802
2803 if (!(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
2804 hide_si_addr_tag_bits(ksig);
2805
2806 return ksig->sig > 0;
2807 }
2808
2809 /**
2810 * signal_delivered -
2811 * @ksig: kernel signal struct
2812 * @stepping: nonzero if debugger single-step or block-step in use
2813 *
2814 * This function should be called when a signal has successfully been
2815 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2816 * is always blocked, and the signal itself is blocked unless %SA_NODEFER
2817 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2818 */
2819 static void signal_delivered(struct ksignal *ksig, int stepping)
2820 {
2821 sigset_t blocked;
2822
2823 /* A signal was successfully delivered, and the
2824 saved sigmask was stored on the signal frame,
2825 and will be restored by sigreturn. So we can
2826 simply clear the restore sigmask flag. */
2827 clear_restore_sigmask();
2828
2829 sigorsets(&blocked, &current->blocked, &ksig->ka.sa.sa_mask);
2830 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2831 sigaddset(&blocked, ksig->sig);
2832 set_current_blocked(&blocked);
2833 tracehook_signal_handler(stepping);
2834 }
2835
2836 void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2837 {
2838 if (failed)
2839 force_sigsegv(ksig->sig);
2840 else
2841 signal_delivered(ksig, stepping);
2842 }
2843
2844 /*
2845 * It could be that complete_signal() picked us to notify about the
2846 * group-wide signal. Other threads should be notified now to take
2847 * the shared signals in @which since we will not.
2848 */
2849 static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2850 {
2851 sigset_t retarget;
2852 struct task_struct *t;
2853
2854 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2855 if (sigisemptyset(&retarget))
2856 return;
2857
2858 t = tsk;
2859 while_each_thread(tsk, t) {
2860 if (t->flags & PF_EXITING)
2861 continue;
2862
2863 if (!has_pending_signals(&retarget, &t->blocked))
2864 continue;
2865 /* Remove the signals this thread can handle. */
2866 sigandsets(&retarget, &retarget, &t->blocked);
2867
2868 if (!task_sigpending(t))
2869 signal_wake_up(t, 0);
2870
2871 if (sigisemptyset(&retarget))
2872 break;
2873 }
2874 }
2875
2876 void exit_signals(struct task_struct *tsk)
2877 {
2878 int group_stop = 0;
2879 sigset_t unblocked;
2880
2881 /*
2882 * @tsk is about to have PF_EXITING set - lock out users which
2883 * expect stable threadgroup.
2884 */
2885 cgroup_threadgroup_change_begin(tsk);
2886
2887 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2888 tsk->flags |= PF_EXITING;
2889 cgroup_threadgroup_change_end(tsk);
2890 return;
2891 }
2892
2893 spin_lock_irq(&tsk->sighand->siglock);
2894 /*
2895 * From now this task is not visible for group-wide signals,
2896 * see wants_signal(), do_signal_stop().
2897 */
2898 tsk->flags |= PF_EXITING;
2899
2900 cgroup_threadgroup_change_end(tsk);
2901
2902 if (!task_sigpending(tsk))
2903 goto out;
2904
2905 unblocked = tsk->blocked;
2906 signotset(&unblocked);
2907 retarget_shared_pending(tsk, &unblocked);
2908
2909 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2910 task_participate_group_stop(tsk))
2911 group_stop = CLD_STOPPED;
2912 out:
2913 spin_unlock_irq(&tsk->sighand->siglock);
2914
2915 /*
2916 * If group stop has completed, deliver the notification. This
2917 * should always go to the real parent of the group leader.
2918 */
2919 if (unlikely(group_stop)) {
2920 read_lock(&tasklist_lock);
2921 do_notify_parent_cldstop(tsk, false, group_stop);
2922 read_unlock(&tasklist_lock);
2923 }
2924 }
2925
2926 /*
2927 * System call entry points.
2928 */
2929
2930 /**
2931 * sys_restart_syscall - restart a system call
2932 */
2933 SYSCALL_DEFINE0(restart_syscall)
2934 {
2935 struct restart_block *restart = &current->restart_block;
2936 return restart->fn(restart);
2937 }
2938
2939 long do_no_restart_syscall(struct restart_block *param)
2940 {
2941 return -EINTR;
2942 }
2943
2944 static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2945 {
2946 if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
2947 sigset_t newblocked;
2948 /* A set of now blocked but previously unblocked signals. */
2949 sigandnsets(&newblocked, newset, &current->blocked);
2950 retarget_shared_pending(tsk, &newblocked);
2951 }
2952 tsk->blocked = *newset;
2953 recalc_sigpending();
2954 }
2955
2956 /**
2957 * set_current_blocked - change current->blocked mask
2958 * @newset: new mask
2959 *
2960 * It is wrong to change ->blocked directly, this helper should be used
2961 * to ensure the process can't miss a shared signal we are going to block.
2962 */
2963 void set_current_blocked(sigset_t *newset)
2964 {
2965 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
2966 __set_current_blocked(newset);
2967 }
2968
2969 void __set_current_blocked(const sigset_t *newset)
2970 {
2971 struct task_struct *tsk = current;
2972
2973 /*
2974 * In case the signal mask hasn't changed, there is nothing we need
2975 * to do. The current->blocked shouldn't be modified by other task.
2976 */
2977 if (sigequalsets(&tsk->blocked, newset))
2978 return;
2979
2980 spin_lock_irq(&tsk->sighand->siglock);
2981 __set_task_blocked(tsk, newset);
2982 spin_unlock_irq(&tsk->sighand->siglock);
2983 }
2984
2985 /*
2986 * This is also useful for kernel threads that want to temporarily
2987 * (or permanently) block certain signals.
2988 *
2989 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2990 * interface happily blocks "unblockable" signals like SIGKILL
2991 * and friends.
2992 */
2993 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2994 {
2995 struct task_struct *tsk = current;
2996 sigset_t newset;
2997
2998 /* Lockless, only current can change ->blocked, never from irq */
2999 if (oldset)
3000 *oldset = tsk->blocked;
3001
3002 switch (how) {
3003 case SIG_BLOCK:
3004 sigorsets(&newset, &tsk->blocked, set);
3005 break;
3006 case SIG_UNBLOCK:
3007 sigandnsets(&newset, &tsk->blocked, set);
3008 break;
3009 case SIG_SETMASK:
3010 newset = *set;
3011 break;
3012 default:
3013 return -EINVAL;
3014 }
3015
3016 __set_current_blocked(&newset);
3017 return 0;
3018 }
3019 EXPORT_SYMBOL(sigprocmask);
3020
3021 /*
3022 * The api helps set app-provided sigmasks.
3023 *
3024 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3025 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3026 *
3027 * Note that it does set_restore_sigmask() in advance, so it must be always
3028 * paired with restore_saved_sigmask_unless() before return from syscall.
3029 */
3030 int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3031 {
3032 sigset_t kmask;
3033
3034 if (!umask)
3035 return 0;
3036 if (sigsetsize != sizeof(sigset_t))
3037 return -EINVAL;
3038 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3039 return -EFAULT;
3040
3041 set_restore_sigmask();
3042 current->saved_sigmask = current->blocked;
3043 set_current_blocked(&kmask);
3044
3045 return 0;
3046 }
3047
3048 #ifdef CONFIG_COMPAT
3049 int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3050 size_t sigsetsize)
3051 {
3052 sigset_t kmask;
3053
3054 if (!umask)
3055 return 0;
3056 if (sigsetsize != sizeof(compat_sigset_t))
3057 return -EINVAL;
3058 if (get_compat_sigset(&kmask, umask))
3059 return -EFAULT;
3060
3061 set_restore_sigmask();
3062 current->saved_sigmask = current->blocked;
3063 set_current_blocked(&kmask);
3064
3065 return 0;
3066 }
3067 #endif
3068
3069 /**
3070 * sys_rt_sigprocmask - change the list of currently blocked signals
3071 * @how: whether to add, remove, or set signals
3072 * @nset: stores pending signals
3073 * @oset: previous value of signal mask if non-null
3074 * @sigsetsize: size of sigset_t type
3075 */
3076 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3077 sigset_t __user *, oset, size_t, sigsetsize)
3078 {
3079 sigset_t old_set, new_set;
3080 int error;
3081
3082 /* XXX: Don't preclude handling different sized sigset_t's. */
3083 if (sigsetsize != sizeof(sigset_t))
3084 return -EINVAL;
3085
3086 old_set = current->blocked;
3087
3088 if (nset) {
3089 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3090 return -EFAULT;
3091 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3092
3093 error = sigprocmask(how, &new_set, NULL);
3094 if (error)
3095 return error;
3096 }
3097
3098 if (oset) {
3099 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3100 return -EFAULT;
3101 }
3102
3103 return 0;
3104 }
3105
3106 #ifdef CONFIG_COMPAT
3107 COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3108 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3109 {
3110 sigset_t old_set = current->blocked;
3111
3112 /* XXX: Don't preclude handling different sized sigset_t's. */
3113 if (sigsetsize != sizeof(sigset_t))
3114 return -EINVAL;
3115
3116 if (nset) {
3117 sigset_t new_set;
3118 int error;
3119 if (get_compat_sigset(&new_set, nset))
3120 return -EFAULT;
3121 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3122
3123 error = sigprocmask(how, &new_set, NULL);
3124 if (error)
3125 return error;
3126 }
3127 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3128 }
3129 #endif
3130
3131 static void do_sigpending(sigset_t *set)
3132 {
3133 spin_lock_irq(&current->sighand->siglock);
3134 sigorsets(set, &current->pending.signal,
3135 &current->signal->shared_pending.signal);
3136 spin_unlock_irq(&current->sighand->siglock);
3137
3138 /* Outside the lock because only this thread touches it. */
3139 sigandsets(set, &current->blocked, set);
3140 }
3141
3142 /**
3143 * sys_rt_sigpending - examine a pending signal that has been raised
3144 * while blocked
3145 * @uset: stores pending signals
3146 * @sigsetsize: size of sigset_t type or larger
3147 */
3148 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3149 {
3150 sigset_t set;
3151
3152 if (sigsetsize > sizeof(*uset))
3153 return -EINVAL;
3154
3155 do_sigpending(&set);
3156
3157 if (copy_to_user(uset, &set, sigsetsize))
3158 return -EFAULT;
3159
3160 return 0;
3161 }
3162
3163 #ifdef CONFIG_COMPAT
3164 COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3165 compat_size_t, sigsetsize)
3166 {
3167 sigset_t set;
3168
3169 if (sigsetsize > sizeof(*uset))
3170 return -EINVAL;
3171
3172 do_sigpending(&set);
3173
3174 return put_compat_sigset(uset, &set, sigsetsize);
3175 }
3176 #endif
3177
3178 static const struct {
3179 unsigned char limit, layout;
3180 } sig_sicodes[] = {
3181 [SIGILL] = { NSIGILL, SIL_FAULT },
3182 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3183 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3184 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3185 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3186 #if defined(SIGEMT)
3187 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3188 #endif
3189 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3190 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3191 [SIGSYS] = { NSIGSYS, SIL_SYS },
3192 };
3193
3194 static bool known_siginfo_layout(unsigned sig, int si_code)
3195 {
3196 if (si_code == SI_KERNEL)
3197 return true;
3198 else if ((si_code > SI_USER)) {
3199 if (sig_specific_sicodes(sig)) {
3200 if (si_code <= sig_sicodes[sig].limit)
3201 return true;
3202 }
3203 else if (si_code <= NSIGPOLL)
3204 return true;
3205 }
3206 else if (si_code >= SI_DETHREAD)
3207 return true;
3208 else if (si_code == SI_ASYNCNL)
3209 return true;
3210 return false;
3211 }
3212
3213 enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3214 {
3215 enum siginfo_layout layout = SIL_KILL;
3216 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3217 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3218 (si_code <= sig_sicodes[sig].limit)) {
3219 layout = sig_sicodes[sig].layout;
3220 /* Handle the exceptions */
3221 if ((sig == SIGBUS) &&
3222 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3223 layout = SIL_FAULT_MCEERR;
3224 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3225 layout = SIL_FAULT_BNDERR;
3226 #ifdef SEGV_PKUERR
3227 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3228 layout = SIL_FAULT_PKUERR;
3229 #endif
3230 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3231 layout = SIL_PERF_EVENT;
3232 #ifdef __ARCH_SI_TRAPNO
3233 else if (layout == SIL_FAULT)
3234 layout = SIL_FAULT_TRAPNO;
3235 #endif
3236 }
3237 else if (si_code <= NSIGPOLL)
3238 layout = SIL_POLL;
3239 } else {
3240 if (si_code == SI_TIMER)
3241 layout = SIL_TIMER;
3242 else if (si_code == SI_SIGIO)
3243 layout = SIL_POLL;
3244 else if (si_code < 0)
3245 layout = SIL_RT;
3246 }
3247 return layout;
3248 }
3249
3250 static inline char __user *si_expansion(const siginfo_t __user *info)
3251 {
3252 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3253 }
3254
3255 int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3256 {
3257 char __user *expansion = si_expansion(to);
3258 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3259 return -EFAULT;
3260 if (clear_user(expansion, SI_EXPANSION_SIZE))
3261 return -EFAULT;
3262 return 0;
3263 }
3264
3265 static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3266 const siginfo_t __user *from)
3267 {
3268 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3269 char __user *expansion = si_expansion(from);
3270 char buf[SI_EXPANSION_SIZE];
3271 int i;
3272 /*
3273 * An unknown si_code might need more than
3274 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3275 * extra bytes are 0. This guarantees copy_siginfo_to_user
3276 * will return this data to userspace exactly.
3277 */
3278 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3279 return -EFAULT;
3280 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3281 if (buf[i] != 0)
3282 return -E2BIG;
3283 }
3284 }
3285 return 0;
3286 }
3287
3288 static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3289 const siginfo_t __user *from)
3290 {
3291 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3292 return -EFAULT;
3293 to->si_signo = signo;
3294 return post_copy_siginfo_from_user(to, from);
3295 }
3296
3297 int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3298 {
3299 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3300 return -EFAULT;
3301 return post_copy_siginfo_from_user(to, from);
3302 }
3303
3304 #ifdef CONFIG_COMPAT
3305 /**
3306 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3307 * @to: compat siginfo destination
3308 * @from: kernel siginfo source
3309 *
3310 * Note: This function does not work properly for the SIGCHLD on x32, but
3311 * fortunately it doesn't have to. The only valid callers for this function are
3312 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3313 * The latter does not care because SIGCHLD will never cause a coredump.
3314 */
3315 void copy_siginfo_to_external32(struct compat_siginfo *to,
3316 const struct kernel_siginfo *from)
3317 {
3318 memset(to, 0, sizeof(*to));
3319
3320 to->si_signo = from->si_signo;
3321 to->si_errno = from->si_errno;
3322 to->si_code = from->si_code;
3323 switch(siginfo_layout(from->si_signo, from->si_code)) {
3324 case SIL_KILL:
3325 to->si_pid = from->si_pid;
3326 to->si_uid = from->si_uid;
3327 break;
3328 case SIL_TIMER:
3329 to->si_tid = from->si_tid;
3330 to->si_overrun = from->si_overrun;
3331 to->si_int = from->si_int;
3332 break;
3333 case SIL_POLL:
3334 to->si_band = from->si_band;
3335 to->si_fd = from->si_fd;
3336 break;
3337 case SIL_FAULT:
3338 to->si_addr = ptr_to_compat(from->si_addr);
3339 break;
3340 case SIL_FAULT_TRAPNO:
3341 to->si_addr = ptr_to_compat(from->si_addr);
3342 to->si_trapno = from->si_trapno;
3343 break;
3344 case SIL_FAULT_MCEERR:
3345 to->si_addr = ptr_to_compat(from->si_addr);
3346 to->si_addr_lsb = from->si_addr_lsb;
3347 break;
3348 case SIL_FAULT_BNDERR:
3349 to->si_addr = ptr_to_compat(from->si_addr);
3350 to->si_lower = ptr_to_compat(from->si_lower);
3351 to->si_upper = ptr_to_compat(from->si_upper);
3352 break;
3353 case SIL_FAULT_PKUERR:
3354 to->si_addr = ptr_to_compat(from->si_addr);
3355 to->si_pkey = from->si_pkey;
3356 break;
3357 case SIL_PERF_EVENT:
3358 to->si_addr = ptr_to_compat(from->si_addr);
3359 to->si_perf_data = from->si_perf_data;
3360 to->si_perf_type = from->si_perf_type;
3361 break;
3362 case SIL_CHLD:
3363 to->si_pid = from->si_pid;
3364 to->si_uid = from->si_uid;
3365 to->si_status = from->si_status;
3366 to->si_utime = from->si_utime;
3367 to->si_stime = from->si_stime;
3368 break;
3369 case SIL_RT:
3370 to->si_pid = from->si_pid;
3371 to->si_uid = from->si_uid;
3372 to->si_int = from->si_int;
3373 break;
3374 case SIL_SYS:
3375 to->si_call_addr = ptr_to_compat(from->si_call_addr);
3376 to->si_syscall = from->si_syscall;
3377 to->si_arch = from->si_arch;
3378 break;
3379 }
3380 }
3381
3382 int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3383 const struct kernel_siginfo *from)
3384 {
3385 struct compat_siginfo new;
3386
3387 copy_siginfo_to_external32(&new, from);
3388 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3389 return -EFAULT;
3390 return 0;
3391 }
3392
3393 static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3394 const struct compat_siginfo *from)
3395 {
3396 clear_siginfo(to);
3397 to->si_signo = from->si_signo;
3398 to->si_errno = from->si_errno;
3399 to->si_code = from->si_code;
3400 switch(siginfo_layout(from->si_signo, from->si_code)) {
3401 case SIL_KILL:
3402 to->si_pid = from->si_pid;
3403 to->si_uid = from->si_uid;
3404 break;
3405 case SIL_TIMER:
3406 to->si_tid = from->si_tid;
3407 to->si_overrun = from->si_overrun;
3408 to->si_int = from->si_int;
3409 break;
3410 case SIL_POLL:
3411 to->si_band = from->si_band;
3412 to->si_fd = from->si_fd;
3413 break;
3414 case SIL_FAULT:
3415 to->si_addr = compat_ptr(from->si_addr);
3416 break;
3417 case SIL_FAULT_TRAPNO:
3418 to->si_addr = compat_ptr(from->si_addr);
3419 to->si_trapno = from->si_trapno;
3420 break;
3421 case SIL_FAULT_MCEERR:
3422 to->si_addr = compat_ptr(from->si_addr);
3423 to->si_addr_lsb = from->si_addr_lsb;
3424 break;
3425 case SIL_FAULT_BNDERR:
3426 to->si_addr = compat_ptr(from->si_addr);
3427 to->si_lower = compat_ptr(from->si_lower);
3428 to->si_upper = compat_ptr(from->si_upper);
3429 break;
3430 case SIL_FAULT_PKUERR:
3431 to->si_addr = compat_ptr(from->si_addr);
3432 to->si_pkey = from->si_pkey;
3433 break;
3434 case SIL_PERF_EVENT:
3435 to->si_addr = compat_ptr(from->si_addr);
3436 to->si_perf_data = from->si_perf_data;
3437 to->si_perf_type = from->si_perf_type;
3438 break;
3439 case SIL_CHLD:
3440 to->si_pid = from->si_pid;
3441 to->si_uid = from->si_uid;
3442 to->si_status = from->si_status;
3443 #ifdef CONFIG_X86_X32_ABI
3444 if (in_x32_syscall()) {
3445 to->si_utime = from->_sifields._sigchld_x32._utime;
3446 to->si_stime = from->_sifields._sigchld_x32._stime;
3447 } else
3448 #endif
3449 {
3450 to->si_utime = from->si_utime;
3451 to->si_stime = from->si_stime;
3452 }
3453 break;
3454 case SIL_RT:
3455 to->si_pid = from->si_pid;
3456 to->si_uid = from->si_uid;
3457 to->si_int = from->si_int;
3458 break;
3459 case SIL_SYS:
3460 to->si_call_addr = compat_ptr(from->si_call_addr);
3461 to->si_syscall = from->si_syscall;
3462 to->si_arch = from->si_arch;
3463 break;
3464 }
3465 return 0;
3466 }
3467
3468 static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3469 const struct compat_siginfo __user *ufrom)
3470 {
3471 struct compat_siginfo from;
3472
3473 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3474 return -EFAULT;
3475
3476 from.si_signo = signo;
3477 return post_copy_siginfo_from_user32(to, &from);
3478 }
3479
3480 int copy_siginfo_from_user32(struct kernel_siginfo *to,
3481 const struct compat_siginfo __user *ufrom)
3482 {
3483 struct compat_siginfo from;
3484
3485 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3486 return -EFAULT;
3487
3488 return post_copy_siginfo_from_user32(to, &from);
3489 }
3490 #endif /* CONFIG_COMPAT */
3491
3492 /**
3493 * do_sigtimedwait - wait for queued signals specified in @which
3494 * @which: queued signals to wait for
3495 * @info: if non-null, the signal's siginfo is returned here
3496 * @ts: upper bound on process time suspension
3497 */
3498 static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3499 const struct timespec64 *ts)
3500 {
3501 ktime_t *to = NULL, timeout = KTIME_MAX;
3502 struct task_struct *tsk = current;
3503 sigset_t mask = *which;
3504 int sig, ret = 0;
3505
3506 if (ts) {
3507 if (!timespec64_valid(ts))
3508 return -EINVAL;
3509 timeout = timespec64_to_ktime(*ts);
3510 to = &timeout;
3511 }
3512
3513 /*
3514 * Invert the set of allowed signals to get those we want to block.
3515 */
3516 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3517 signotset(&mask);
3518
3519 spin_lock_irq(&tsk->sighand->siglock);
3520 sig = dequeue_signal(tsk, &mask, info);
3521 if (!sig && timeout) {
3522 /*
3523 * None ready, temporarily unblock those we're interested
3524 * while we are sleeping in so that we'll be awakened when
3525 * they arrive. Unblocking is always fine, we can avoid
3526 * set_current_blocked().
3527 */
3528 tsk->real_blocked = tsk->blocked;
3529 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3530 recalc_sigpending();
3531 spin_unlock_irq(&tsk->sighand->siglock);
3532
3533 __set_current_state(TASK_INTERRUPTIBLE);
3534 ret = freezable_schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3535 HRTIMER_MODE_REL);
3536 spin_lock_irq(&tsk->sighand->siglock);
3537 __set_task_blocked(tsk, &tsk->real_blocked);
3538 sigemptyset(&tsk->real_blocked);
3539 sig = dequeue_signal(tsk, &mask, info);
3540 }
3541 spin_unlock_irq(&tsk->sighand->siglock);
3542
3543 if (sig)
3544 return sig;
3545 return ret ? -EINTR : -EAGAIN;
3546 }
3547
3548 /**
3549 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3550 * in @uthese
3551 * @uthese: queued signals to wait for
3552 * @uinfo: if non-null, the signal's siginfo is returned here
3553 * @uts: upper bound on process time suspension
3554 * @sigsetsize: size of sigset_t type
3555 */
3556 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3557 siginfo_t __user *, uinfo,
3558 const struct __kernel_timespec __user *, uts,
3559 size_t, sigsetsize)
3560 {
3561 sigset_t these;
3562 struct timespec64 ts;
3563 kernel_siginfo_t info;
3564 int ret;
3565
3566 /* XXX: Don't preclude handling different sized sigset_t's. */
3567 if (sigsetsize != sizeof(sigset_t))
3568 return -EINVAL;
3569
3570 if (copy_from_user(&these, uthese, sizeof(these)))
3571 return -EFAULT;
3572
3573 if (uts) {
3574 if (get_timespec64(&ts, uts))
3575 return -EFAULT;
3576 }
3577
3578 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3579
3580 if (ret > 0 && uinfo) {
3581 if (copy_siginfo_to_user(uinfo, &info))
3582 ret = -EFAULT;
3583 }
3584
3585 return ret;
3586 }
3587
3588 #ifdef CONFIG_COMPAT_32BIT_TIME
3589 SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3590 siginfo_t __user *, uinfo,
3591 const struct old_timespec32 __user *, uts,
3592 size_t, sigsetsize)
3593 {
3594 sigset_t these;
3595 struct timespec64 ts;
3596 kernel_siginfo_t info;
3597 int ret;
3598
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_old_timespec32(&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 #endif
3620
3621 #ifdef CONFIG_COMPAT
3622 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3623 struct compat_siginfo __user *, uinfo,
3624 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3625 {
3626 sigset_t s;
3627 struct timespec64 t;
3628 kernel_siginfo_t info;
3629 long ret;
3630
3631 if (sigsetsize != sizeof(sigset_t))
3632 return -EINVAL;
3633
3634 if (get_compat_sigset(&s, uthese))
3635 return -EFAULT;
3636
3637 if (uts) {
3638 if (get_timespec64(&t, uts))
3639 return -EFAULT;
3640 }
3641
3642 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3643
3644 if (ret > 0 && uinfo) {
3645 if (copy_siginfo_to_user32(uinfo, &info))
3646 ret = -EFAULT;
3647 }
3648
3649 return ret;
3650 }
3651
3652 #ifdef CONFIG_COMPAT_32BIT_TIME
3653 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3654 struct compat_siginfo __user *, uinfo,
3655 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3656 {
3657 sigset_t s;
3658 struct timespec64 t;
3659 kernel_siginfo_t info;
3660 long ret;
3661
3662 if (sigsetsize != sizeof(sigset_t))
3663 return -EINVAL;
3664
3665 if (get_compat_sigset(&s, uthese))
3666 return -EFAULT;
3667
3668 if (uts) {
3669 if (get_old_timespec32(&t, uts))
3670 return -EFAULT;
3671 }
3672
3673 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3674
3675 if (ret > 0 && uinfo) {
3676 if (copy_siginfo_to_user32(uinfo, &info))
3677 ret = -EFAULT;
3678 }
3679
3680 return ret;
3681 }
3682 #endif
3683 #endif
3684
3685 static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3686 {
3687 clear_siginfo(info);
3688 info->si_signo = sig;
3689 info->si_errno = 0;
3690 info->si_code = SI_USER;
3691 info->si_pid = task_tgid_vnr(current);
3692 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3693 }
3694
3695 /**
3696 * sys_kill - send a signal to a process
3697 * @pid: the PID of the process
3698 * @sig: signal to be sent
3699 */
3700 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3701 {
3702 struct kernel_siginfo info;
3703
3704 prepare_kill_siginfo(sig, &info);
3705
3706 return kill_something_info(sig, &info, pid);
3707 }
3708
3709 /*
3710 * Verify that the signaler and signalee either are in the same pid namespace
3711 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3712 * namespace.
3713 */
3714 static bool access_pidfd_pidns(struct pid *pid)
3715 {
3716 struct pid_namespace *active = task_active_pid_ns(current);
3717 struct pid_namespace *p = ns_of_pid(pid);
3718
3719 for (;;) {
3720 if (!p)
3721 return false;
3722 if (p == active)
3723 break;
3724 p = p->parent;
3725 }
3726
3727 return true;
3728 }
3729
3730 static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3731 siginfo_t __user *info)
3732 {
3733 #ifdef CONFIG_COMPAT
3734 /*
3735 * Avoid hooking up compat syscalls and instead handle necessary
3736 * conversions here. Note, this is a stop-gap measure and should not be
3737 * considered a generic solution.
3738 */
3739 if (in_compat_syscall())
3740 return copy_siginfo_from_user32(
3741 kinfo, (struct compat_siginfo __user *)info);
3742 #endif
3743 return copy_siginfo_from_user(kinfo, info);
3744 }
3745
3746 static struct pid *pidfd_to_pid(const struct file *file)
3747 {
3748 struct pid *pid;
3749
3750 pid = pidfd_pid(file);
3751 if (!IS_ERR(pid))
3752 return pid;
3753
3754 return tgid_pidfd_to_pid(file);
3755 }
3756
3757 /**
3758 * sys_pidfd_send_signal - Signal a process through a pidfd
3759 * @pidfd: file descriptor of the process
3760 * @sig: signal to send
3761 * @info: signal info
3762 * @flags: future flags
3763 *
3764 * The syscall currently only signals via PIDTYPE_PID which covers
3765 * kill(<positive-pid>, <signal>. It does not signal threads or process
3766 * groups.
3767 * In order to extend the syscall to threads and process groups the @flags
3768 * argument should be used. In essence, the @flags argument will determine
3769 * what is signaled and not the file descriptor itself. Put in other words,
3770 * grouping is a property of the flags argument not a property of the file
3771 * descriptor.
3772 *
3773 * Return: 0 on success, negative errno on failure
3774 */
3775 SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3776 siginfo_t __user *, info, unsigned int, flags)
3777 {
3778 int ret;
3779 struct fd f;
3780 struct pid *pid;
3781 kernel_siginfo_t kinfo;
3782
3783 /* Enforce flags be set to 0 until we add an extension. */
3784 if (flags)
3785 return -EINVAL;
3786
3787 f = fdget(pidfd);
3788 if (!f.file)
3789 return -EBADF;
3790
3791 /* Is this a pidfd? */
3792 pid = pidfd_to_pid(f.file);
3793 if (IS_ERR(pid)) {
3794 ret = PTR_ERR(pid);
3795 goto err;
3796 }
3797
3798 ret = -EINVAL;
3799 if (!access_pidfd_pidns(pid))
3800 goto err;
3801
3802 if (info) {
3803 ret = copy_siginfo_from_user_any(&kinfo, info);
3804 if (unlikely(ret))
3805 goto err;
3806
3807 ret = -EINVAL;
3808 if (unlikely(sig != kinfo.si_signo))
3809 goto err;
3810
3811 /* Only allow sending arbitrary signals to yourself. */
3812 ret = -EPERM;
3813 if ((task_pid(current) != pid) &&
3814 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3815 goto err;
3816 } else {
3817 prepare_kill_siginfo(sig, &kinfo);
3818 }
3819
3820 ret = kill_pid_info(sig, &kinfo, pid);
3821
3822 err:
3823 fdput(f);
3824 return ret;
3825 }
3826
3827 static int
3828 do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3829 {
3830 struct task_struct *p;
3831 int error = -ESRCH;
3832
3833 rcu_read_lock();
3834 p = find_task_by_vpid(pid);
3835 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3836 error = check_kill_permission(sig, info, p);
3837 /*
3838 * The null signal is a permissions and process existence
3839 * probe. No signal is actually delivered.
3840 */
3841 if (!error && sig) {
3842 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3843 /*
3844 * If lock_task_sighand() failed we pretend the task
3845 * dies after receiving the signal. The window is tiny,
3846 * and the signal is private anyway.
3847 */
3848 if (unlikely(error == -ESRCH))
3849 error = 0;
3850 }
3851 }
3852 rcu_read_unlock();
3853
3854 return error;
3855 }
3856
3857 static int do_tkill(pid_t tgid, pid_t pid, int sig)
3858 {
3859 struct kernel_siginfo info;
3860
3861 clear_siginfo(&info);
3862 info.si_signo = sig;
3863 info.si_errno = 0;
3864 info.si_code = SI_TKILL;
3865 info.si_pid = task_tgid_vnr(current);
3866 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3867
3868 return do_send_specific(tgid, pid, sig, &info);
3869 }
3870
3871 /**
3872 * sys_tgkill - send signal to one specific thread
3873 * @tgid: the thread group ID of the thread
3874 * @pid: the PID of the thread
3875 * @sig: signal to be sent
3876 *
3877 * This syscall also checks the @tgid and returns -ESRCH even if the PID
3878 * exists but it's not belonging to the target process anymore. This
3879 * method solves the problem of threads exiting and PIDs getting reused.
3880 */
3881 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
3882 {
3883 /* This is only valid for single tasks */
3884 if (pid <= 0 || tgid <= 0)
3885 return -EINVAL;
3886
3887 return do_tkill(tgid, pid, sig);
3888 }
3889
3890 /**
3891 * sys_tkill - send signal to one specific task
3892 * @pid: the PID of the task
3893 * @sig: signal to be sent
3894 *
3895 * Send a signal to only one task, even if it's a CLONE_THREAD task.
3896 */
3897 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
3898 {
3899 /* This is only valid for single tasks */
3900 if (pid <= 0)
3901 return -EINVAL;
3902
3903 return do_tkill(0, pid, sig);
3904 }
3905
3906 static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
3907 {
3908 /* Not even root can pretend to send signals from the kernel.
3909 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3910 */
3911 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3912 (task_pid_vnr(current) != pid))
3913 return -EPERM;
3914
3915 /* POSIX.1b doesn't mention process groups. */
3916 return kill_proc_info(sig, info, pid);
3917 }
3918
3919 /**
3920 * sys_rt_sigqueueinfo - send signal information to a signal
3921 * @pid: the PID of the thread
3922 * @sig: signal to be sent
3923 * @uinfo: signal info to be sent
3924 */
3925 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
3926 siginfo_t __user *, uinfo)
3927 {
3928 kernel_siginfo_t info;
3929 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3930 if (unlikely(ret))
3931 return ret;
3932 return do_rt_sigqueueinfo(pid, sig, &info);
3933 }
3934
3935 #ifdef CONFIG_COMPAT
3936 COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
3937 compat_pid_t, pid,
3938 int, sig,
3939 struct compat_siginfo __user *, uinfo)
3940 {
3941 kernel_siginfo_t info;
3942 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3943 if (unlikely(ret))
3944 return ret;
3945 return do_rt_sigqueueinfo(pid, sig, &info);
3946 }
3947 #endif
3948
3949 static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
3950 {
3951 /* This is only valid for single tasks */
3952 if (pid <= 0 || tgid <= 0)
3953 return -EINVAL;
3954
3955 /* Not even root can pretend to send signals from the kernel.
3956 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3957 */
3958 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3959 (task_pid_vnr(current) != pid))
3960 return -EPERM;
3961
3962 return do_send_specific(tgid, pid, sig, info);
3963 }
3964
3965 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
3966 siginfo_t __user *, uinfo)
3967 {
3968 kernel_siginfo_t info;
3969 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3970 if (unlikely(ret))
3971 return ret;
3972 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3973 }
3974
3975 #ifdef CONFIG_COMPAT
3976 COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
3977 compat_pid_t, tgid,
3978 compat_pid_t, pid,
3979 int, sig,
3980 struct compat_siginfo __user *, uinfo)
3981 {
3982 kernel_siginfo_t info;
3983 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3984 if (unlikely(ret))
3985 return ret;
3986 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3987 }
3988 #endif
3989
3990 /*
3991 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
3992 */
3993 void kernel_sigaction(int sig, __sighandler_t action)
3994 {
3995 spin_lock_irq(&current->sighand->siglock);
3996 current->sighand->action[sig - 1].sa.sa_handler = action;
3997 if (action == SIG_IGN) {
3998 sigset_t mask;
3999
4000 sigemptyset(&mask);
4001 sigaddset(&mask, sig);
4002
4003 flush_sigqueue_mask(&mask, &current->signal->shared_pending);
4004 flush_sigqueue_mask(&mask, &current->pending);
4005 recalc_sigpending();
4006 }
4007 spin_unlock_irq(&current->sighand->siglock);
4008 }
4009 EXPORT_SYMBOL(kernel_sigaction);
4010
4011 void __weak sigaction_compat_abi(struct k_sigaction *act,
4012 struct k_sigaction *oact)
4013 {
4014 }
4015
4016 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4017 {
4018 struct task_struct *p = current, *t;
4019 struct k_sigaction *k;
4020 sigset_t mask;
4021
4022 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4023 return -EINVAL;
4024
4025 k = &p->sighand->action[sig-1];
4026
4027 spin_lock_irq(&p->sighand->siglock);
4028 if (oact)
4029 *oact = *k;
4030
4031 /*
4032 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4033 * e.g. by having an architecture use the bit in their uapi.
4034 */
4035 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4036
4037 /*
4038 * Clear unknown flag bits in order to allow userspace to detect missing
4039 * support for flag bits and to allow the kernel to use non-uapi bits
4040 * internally.
4041 */
4042 if (act)
4043 act->sa.sa_flags &= UAPI_SA_FLAGS;
4044 if (oact)
4045 oact->sa.sa_flags &= UAPI_SA_FLAGS;
4046
4047 sigaction_compat_abi(act, oact);
4048
4049 if (act) {
4050 sigdelsetmask(&act->sa.sa_mask,
4051 sigmask(SIGKILL) | sigmask(SIGSTOP));
4052 *k = *act;
4053 /*
4054 * POSIX 3.3.1.3:
4055 * "Setting a signal action to SIG_IGN for a signal that is
4056 * pending shall cause the pending signal to be discarded,
4057 * whether or not it is blocked."
4058 *
4059 * "Setting a signal action to SIG_DFL for a signal that is
4060 * pending and whose default action is to ignore the signal
4061 * (for example, SIGCHLD), shall cause the pending signal to
4062 * be discarded, whether or not it is blocked"
4063 */
4064 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4065 sigemptyset(&mask);
4066 sigaddset(&mask, sig);
4067 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
4068 for_each_thread(p, t)
4069 flush_sigqueue_mask(&mask, &t->pending);
4070 }
4071 }
4072
4073 spin_unlock_irq(&p->sighand->siglock);
4074 return 0;
4075 }
4076
4077 static int
4078 do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4079 size_t min_ss_size)
4080 {
4081 struct task_struct *t = current;
4082
4083 if (oss) {
4084 memset(oss, 0, sizeof(stack_t));
4085 oss->ss_sp = (void __user *) t->sas_ss_sp;
4086 oss->ss_size = t->sas_ss_size;
4087 oss->ss_flags = sas_ss_flags(sp) |
4088 (current->sas_ss_flags & SS_FLAG_BITS);
4089 }
4090
4091 if (ss) {
4092 void __user *ss_sp = ss->ss_sp;
4093 size_t ss_size = ss->ss_size;
4094 unsigned ss_flags = ss->ss_flags;
4095 int ss_mode;
4096
4097 if (unlikely(on_sig_stack(sp)))
4098 return -EPERM;
4099
4100 ss_mode = ss_flags & ~SS_FLAG_BITS;
4101 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4102 ss_mode != 0))
4103 return -EINVAL;
4104
4105 if (ss_mode == SS_DISABLE) {
4106 ss_size = 0;
4107 ss_sp = NULL;
4108 } else {
4109 if (unlikely(ss_size < min_ss_size))
4110 return -ENOMEM;
4111 }
4112
4113 t->sas_ss_sp = (unsigned long) ss_sp;
4114 t->sas_ss_size = ss_size;
4115 t->sas_ss_flags = ss_flags;
4116 }
4117 return 0;
4118 }
4119
4120 SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4121 {
4122 stack_t new, old;
4123 int err;
4124 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4125 return -EFAULT;
4126 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4127 current_user_stack_pointer(),
4128 MINSIGSTKSZ);
4129 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4130 err = -EFAULT;
4131 return err;
4132 }
4133
4134 int restore_altstack(const stack_t __user *uss)
4135 {
4136 stack_t new;
4137 if (copy_from_user(&new, uss, sizeof(stack_t)))
4138 return -EFAULT;
4139 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4140 MINSIGSTKSZ);
4141 /* squash all but EFAULT for now */
4142 return 0;
4143 }
4144
4145 int __save_altstack(stack_t __user *uss, unsigned long sp)
4146 {
4147 struct task_struct *t = current;
4148 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4149 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4150 __put_user(t->sas_ss_size, &uss->ss_size);
4151 if (err)
4152 return err;
4153 if (t->sas_ss_flags & SS_AUTODISARM)
4154 sas_ss_reset(t);
4155 return 0;
4156 }
4157
4158 #ifdef CONFIG_COMPAT
4159 static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4160 compat_stack_t __user *uoss_ptr)
4161 {
4162 stack_t uss, uoss;
4163 int ret;
4164
4165 if (uss_ptr) {
4166 compat_stack_t uss32;
4167 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4168 return -EFAULT;
4169 uss.ss_sp = compat_ptr(uss32.ss_sp);
4170 uss.ss_flags = uss32.ss_flags;
4171 uss.ss_size = uss32.ss_size;
4172 }
4173 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4174 compat_user_stack_pointer(),
4175 COMPAT_MINSIGSTKSZ);
4176 if (ret >= 0 && uoss_ptr) {
4177 compat_stack_t old;
4178 memset(&old, 0, sizeof(old));
4179 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4180 old.ss_flags = uoss.ss_flags;
4181 old.ss_size = uoss.ss_size;
4182 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4183 ret = -EFAULT;
4184 }
4185 return ret;
4186 }
4187
4188 COMPAT_SYSCALL_DEFINE2(sigaltstack,
4189 const compat_stack_t __user *, uss_ptr,
4190 compat_stack_t __user *, uoss_ptr)
4191 {
4192 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4193 }
4194
4195 int compat_restore_altstack(const compat_stack_t __user *uss)
4196 {
4197 int err = do_compat_sigaltstack(uss, NULL);
4198 /* squash all but -EFAULT for now */
4199 return err == -EFAULT ? err : 0;
4200 }
4201
4202 int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4203 {
4204 int err;
4205 struct task_struct *t = current;
4206 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4207 &uss->ss_sp) |
4208 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4209 __put_user(t->sas_ss_size, &uss->ss_size);
4210 if (err)
4211 return err;
4212 if (t->sas_ss_flags & SS_AUTODISARM)
4213 sas_ss_reset(t);
4214 return 0;
4215 }
4216 #endif
4217
4218 #ifdef __ARCH_WANT_SYS_SIGPENDING
4219
4220 /**
4221 * sys_sigpending - examine pending signals
4222 * @uset: where mask of pending signal is returned
4223 */
4224 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4225 {
4226 sigset_t set;
4227
4228 if (sizeof(old_sigset_t) > sizeof(*uset))
4229 return -EINVAL;
4230
4231 do_sigpending(&set);
4232
4233 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4234 return -EFAULT;
4235
4236 return 0;
4237 }
4238
4239 #ifdef CONFIG_COMPAT
4240 COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4241 {
4242 sigset_t set;
4243
4244 do_sigpending(&set);
4245
4246 return put_user(set.sig[0], set32);
4247 }
4248 #endif
4249
4250 #endif
4251
4252 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
4253 /**
4254 * sys_sigprocmask - examine and change blocked signals
4255 * @how: whether to add, remove, or set signals
4256 * @nset: signals to add or remove (if non-null)
4257 * @oset: previous value of signal mask if non-null
4258 *
4259 * Some platforms have their own version with special arguments;
4260 * others support only sys_rt_sigprocmask.
4261 */
4262
4263 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4264 old_sigset_t __user *, oset)
4265 {
4266 old_sigset_t old_set, new_set;
4267 sigset_t new_blocked;
4268
4269 old_set = current->blocked.sig[0];
4270
4271 if (nset) {
4272 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4273 return -EFAULT;
4274
4275 new_blocked = current->blocked;
4276
4277 switch (how) {
4278 case SIG_BLOCK:
4279 sigaddsetmask(&new_blocked, new_set);
4280 break;
4281 case SIG_UNBLOCK:
4282 sigdelsetmask(&new_blocked, new_set);
4283 break;
4284 case SIG_SETMASK:
4285 new_blocked.sig[0] = new_set;
4286 break;
4287 default:
4288 return -EINVAL;
4289 }
4290
4291 set_current_blocked(&new_blocked);
4292 }
4293
4294 if (oset) {
4295 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4296 return -EFAULT;
4297 }
4298
4299 return 0;
4300 }
4301 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4302
4303 #ifndef CONFIG_ODD_RT_SIGACTION
4304 /**
4305 * sys_rt_sigaction - alter an action taken by a process
4306 * @sig: signal to be sent
4307 * @act: new sigaction
4308 * @oact: used to save the previous sigaction
4309 * @sigsetsize: size of sigset_t type
4310 */
4311 SYSCALL_DEFINE4(rt_sigaction, int, sig,
4312 const struct sigaction __user *, act,
4313 struct sigaction __user *, oact,
4314 size_t, sigsetsize)
4315 {
4316 struct k_sigaction new_sa, old_sa;
4317 int ret;
4318
4319 /* XXX: Don't preclude handling different sized sigset_t's. */
4320 if (sigsetsize != sizeof(sigset_t))
4321 return -EINVAL;
4322
4323 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4324 return -EFAULT;
4325
4326 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4327 if (ret)
4328 return ret;
4329
4330 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4331 return -EFAULT;
4332
4333 return 0;
4334 }
4335 #ifdef CONFIG_COMPAT
4336 COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4337 const struct compat_sigaction __user *, act,
4338 struct compat_sigaction __user *, oact,
4339 compat_size_t, sigsetsize)
4340 {
4341 struct k_sigaction new_ka, old_ka;
4342 #ifdef __ARCH_HAS_SA_RESTORER
4343 compat_uptr_t restorer;
4344 #endif
4345 int ret;
4346
4347 /* XXX: Don't preclude handling different sized sigset_t's. */
4348 if (sigsetsize != sizeof(compat_sigset_t))
4349 return -EINVAL;
4350
4351 if (act) {
4352 compat_uptr_t handler;
4353 ret = get_user(handler, &act->sa_handler);
4354 new_ka.sa.sa_handler = compat_ptr(handler);
4355 #ifdef __ARCH_HAS_SA_RESTORER
4356 ret |= get_user(restorer, &act->sa_restorer);
4357 new_ka.sa.sa_restorer = compat_ptr(restorer);
4358 #endif
4359 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4360 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4361 if (ret)
4362 return -EFAULT;
4363 }
4364
4365 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4366 if (!ret && oact) {
4367 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4368 &oact->sa_handler);
4369 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4370 sizeof(oact->sa_mask));
4371 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4372 #ifdef __ARCH_HAS_SA_RESTORER
4373 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4374 &oact->sa_restorer);
4375 #endif
4376 }
4377 return ret;
4378 }
4379 #endif
4380 #endif /* !CONFIG_ODD_RT_SIGACTION */
4381
4382 #ifdef CONFIG_OLD_SIGACTION
4383 SYSCALL_DEFINE3(sigaction, int, sig,
4384 const struct old_sigaction __user *, act,
4385 struct old_sigaction __user *, oact)
4386 {
4387 struct k_sigaction new_ka, old_ka;
4388 int ret;
4389
4390 if (act) {
4391 old_sigset_t mask;
4392 if (!access_ok(act, sizeof(*act)) ||
4393 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4394 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4395 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4396 __get_user(mask, &act->sa_mask))
4397 return -EFAULT;
4398 #ifdef __ARCH_HAS_KA_RESTORER
4399 new_ka.ka_restorer = NULL;
4400 #endif
4401 siginitset(&new_ka.sa.sa_mask, mask);
4402 }
4403
4404 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4405
4406 if (!ret && oact) {
4407 if (!access_ok(oact, sizeof(*oact)) ||
4408 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4409 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4410 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4411 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4412 return -EFAULT;
4413 }
4414
4415 return ret;
4416 }
4417 #endif
4418 #ifdef CONFIG_COMPAT_OLD_SIGACTION
4419 COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4420 const struct compat_old_sigaction __user *, act,
4421 struct compat_old_sigaction __user *, oact)
4422 {
4423 struct k_sigaction new_ka, old_ka;
4424 int ret;
4425 compat_old_sigset_t mask;
4426 compat_uptr_t handler, restorer;
4427
4428 if (act) {
4429 if (!access_ok(act, sizeof(*act)) ||
4430 __get_user(handler, &act->sa_handler) ||
4431 __get_user(restorer, &act->sa_restorer) ||
4432 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4433 __get_user(mask, &act->sa_mask))
4434 return -EFAULT;
4435
4436 #ifdef __ARCH_HAS_KA_RESTORER
4437 new_ka.ka_restorer = NULL;
4438 #endif
4439 new_ka.sa.sa_handler = compat_ptr(handler);
4440 new_ka.sa.sa_restorer = compat_ptr(restorer);
4441 siginitset(&new_ka.sa.sa_mask, mask);
4442 }
4443
4444 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4445
4446 if (!ret && oact) {
4447 if (!access_ok(oact, sizeof(*oact)) ||
4448 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4449 &oact->sa_handler) ||
4450 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4451 &oact->sa_restorer) ||
4452 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4453 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4454 return -EFAULT;
4455 }
4456 return ret;
4457 }
4458 #endif
4459
4460 #ifdef CONFIG_SGETMASK_SYSCALL
4461
4462 /*
4463 * For backwards compatibility. Functionality superseded by sigprocmask.
4464 */
4465 SYSCALL_DEFINE0(sgetmask)
4466 {
4467 /* SMP safe */
4468 return current->blocked.sig[0];
4469 }
4470
4471 SYSCALL_DEFINE1(ssetmask, int, newmask)
4472 {
4473 int old = current->blocked.sig[0];
4474 sigset_t newset;
4475
4476 siginitset(&newset, newmask);
4477 set_current_blocked(&newset);
4478
4479 return old;
4480 }
4481 #endif /* CONFIG_SGETMASK_SYSCALL */
4482
4483 #ifdef __ARCH_WANT_SYS_SIGNAL
4484 /*
4485 * For backwards compatibility. Functionality superseded by sigaction.
4486 */
4487 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4488 {
4489 struct k_sigaction new_sa, old_sa;
4490 int ret;
4491
4492 new_sa.sa.sa_handler = handler;
4493 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4494 sigemptyset(&new_sa.sa.sa_mask);
4495
4496 ret = do_sigaction(sig, &new_sa, &old_sa);
4497
4498 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4499 }
4500 #endif /* __ARCH_WANT_SYS_SIGNAL */
4501
4502 #ifdef __ARCH_WANT_SYS_PAUSE
4503
4504 SYSCALL_DEFINE0(pause)
4505 {
4506 while (!signal_pending(current)) {
4507 __set_current_state(TASK_INTERRUPTIBLE);
4508 schedule();
4509 }
4510 return -ERESTARTNOHAND;
4511 }
4512
4513 #endif
4514
4515 static int sigsuspend(sigset_t *set)
4516 {
4517 current->saved_sigmask = current->blocked;
4518 set_current_blocked(set);
4519
4520 while (!signal_pending(current)) {
4521 __set_current_state(TASK_INTERRUPTIBLE);
4522 schedule();
4523 }
4524 set_restore_sigmask();
4525 return -ERESTARTNOHAND;
4526 }
4527
4528 /**
4529 * sys_rt_sigsuspend - replace the signal mask for a value with the
4530 * @unewset value until a signal is received
4531 * @unewset: new signal mask value
4532 * @sigsetsize: size of sigset_t type
4533 */
4534 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4535 {
4536 sigset_t newset;
4537
4538 /* XXX: Don't preclude handling different sized sigset_t's. */
4539 if (sigsetsize != sizeof(sigset_t))
4540 return -EINVAL;
4541
4542 if (copy_from_user(&newset, unewset, sizeof(newset)))
4543 return -EFAULT;
4544 return sigsuspend(&newset);
4545 }
4546
4547 #ifdef CONFIG_COMPAT
4548 COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4549 {
4550 sigset_t newset;
4551
4552 /* XXX: Don't preclude handling different sized sigset_t's. */
4553 if (sigsetsize != sizeof(sigset_t))
4554 return -EINVAL;
4555
4556 if (get_compat_sigset(&newset, unewset))
4557 return -EFAULT;
4558 return sigsuspend(&newset);
4559 }
4560 #endif
4561
4562 #ifdef CONFIG_OLD_SIGSUSPEND
4563 SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4564 {
4565 sigset_t blocked;
4566 siginitset(&blocked, mask);
4567 return sigsuspend(&blocked);
4568 }
4569 #endif
4570 #ifdef CONFIG_OLD_SIGSUSPEND3
4571 SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4572 {
4573 sigset_t blocked;
4574 siginitset(&blocked, mask);
4575 return sigsuspend(&blocked);
4576 }
4577 #endif
4578
4579 __weak const char *arch_vma_name(struct vm_area_struct *vma)
4580 {
4581 return NULL;
4582 }
4583
4584 static inline void siginfo_buildtime_checks(void)
4585 {
4586 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4587
4588 /* Verify the offsets in the two siginfos match */
4589 #define CHECK_OFFSET(field) \
4590 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4591
4592 /* kill */
4593 CHECK_OFFSET(si_pid);
4594 CHECK_OFFSET(si_uid);
4595
4596 /* timer */
4597 CHECK_OFFSET(si_tid);
4598 CHECK_OFFSET(si_overrun);
4599 CHECK_OFFSET(si_value);
4600
4601 /* rt */
4602 CHECK_OFFSET(si_pid);
4603 CHECK_OFFSET(si_uid);
4604 CHECK_OFFSET(si_value);
4605
4606 /* sigchld */
4607 CHECK_OFFSET(si_pid);
4608 CHECK_OFFSET(si_uid);
4609 CHECK_OFFSET(si_status);
4610 CHECK_OFFSET(si_utime);
4611 CHECK_OFFSET(si_stime);
4612
4613 /* sigfault */
4614 CHECK_OFFSET(si_addr);
4615 CHECK_OFFSET(si_trapno);
4616 CHECK_OFFSET(si_addr_lsb);
4617 CHECK_OFFSET(si_lower);
4618 CHECK_OFFSET(si_upper);
4619 CHECK_OFFSET(si_pkey);
4620 CHECK_OFFSET(si_perf_data);
4621 CHECK_OFFSET(si_perf_type);
4622
4623 /* sigpoll */
4624 CHECK_OFFSET(si_band);
4625 CHECK_OFFSET(si_fd);
4626
4627 /* sigsys */
4628 CHECK_OFFSET(si_call_addr);
4629 CHECK_OFFSET(si_syscall);
4630 CHECK_OFFSET(si_arch);
4631 #undef CHECK_OFFSET
4632
4633 /* usb asyncio */
4634 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4635 offsetof(struct siginfo, si_addr));
4636 if (sizeof(int) == sizeof(void __user *)) {
4637 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4638 sizeof(void __user *));
4639 } else {
4640 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4641 sizeof_field(struct siginfo, si_uid)) !=
4642 sizeof(void __user *));
4643 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4644 offsetof(struct siginfo, si_uid));
4645 }
4646 #ifdef CONFIG_COMPAT
4647 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4648 offsetof(struct compat_siginfo, si_addr));
4649 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4650 sizeof(compat_uptr_t));
4651 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4652 sizeof_field(struct siginfo, si_pid));
4653 #endif
4654 }
4655
4656 void __init signals_init(void)
4657 {
4658 siginfo_buildtime_checks();
4659
4660 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
4661 }
4662
4663 #ifdef CONFIG_KGDB_KDB
4664 #include <linux/kdb.h>
4665 /*
4666 * kdb_send_sig - Allows kdb to send signals without exposing
4667 * signal internals. This function checks if the required locks are
4668 * available before calling the main signal code, to avoid kdb
4669 * deadlocks.
4670 */
4671 void kdb_send_sig(struct task_struct *t, int sig)
4672 {
4673 static struct task_struct *kdb_prev_t;
4674 int new_t, ret;
4675 if (!spin_trylock(&t->sighand->siglock)) {
4676 kdb_printf("Can't do kill command now.\n"
4677 "The sigmask lock is held somewhere else in "
4678 "kernel, try again later\n");
4679 return;
4680 }
4681 new_t = kdb_prev_t != t;
4682 kdb_prev_t = t;
4683 if (!task_is_running(t) && new_t) {
4684 spin_unlock(&t->sighand->siglock);
4685 kdb_printf("Process is not RUNNING, sending a signal from "
4686 "kdb risks deadlock\n"
4687 "on the run queue locks. "
4688 "The signal has _not_ been sent.\n"
4689 "Reissue the kill command if you want to risk "
4690 "the deadlock.\n");
4691 return;
4692 }
4693 ret = send_signal(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4694 spin_unlock(&t->sighand->siglock);
4695 if (ret)
4696 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4697 sig, t->pid);
4698 else
4699 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4700 }
4701 #endif /* CONFIG_KGDB_KDB */