]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - kernel/signal.c
igb: Add counter to i21x doublecheck
[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 switch (sigpending) {
430 case 1:
431 if (likely(get_ucounts(ucounts)))
432 break;
433 fallthrough;
434 case LONG_MAX:
435 /*
436 * we need to decrease the ucount in the userns tree on any
437 * failure to avoid counts leaking.
438 */
439 dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING, 1);
440 rcu_read_unlock();
441 return NULL;
442 }
443 rcu_read_unlock();
444
445 if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
446 q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
447 } else {
448 print_dropped_signal(sig);
449 }
450
451 if (unlikely(q == NULL)) {
452 if (dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING, 1))
453 put_ucounts(ucounts);
454 } else {
455 INIT_LIST_HEAD(&q->list);
456 q->flags = sigqueue_flags;
457 q->ucounts = ucounts;
458 }
459 return q;
460 }
461
462 static void __sigqueue_free(struct sigqueue *q)
463 {
464 if (q->flags & SIGQUEUE_PREALLOC)
465 return;
466 if (q->ucounts && dec_rlimit_ucounts(q->ucounts, UCOUNT_RLIMIT_SIGPENDING, 1)) {
467 put_ucounts(q->ucounts);
468 q->ucounts = NULL;
469 }
470 kmem_cache_free(sigqueue_cachep, q);
471 }
472
473 void flush_sigqueue(struct sigpending *queue)
474 {
475 struct sigqueue *q;
476
477 sigemptyset(&queue->signal);
478 while (!list_empty(&queue->list)) {
479 q = list_entry(queue->list.next, struct sigqueue , list);
480 list_del_init(&q->list);
481 __sigqueue_free(q);
482 }
483 }
484
485 /*
486 * Flush all pending signals for this kthread.
487 */
488 void flush_signals(struct task_struct *t)
489 {
490 unsigned long flags;
491
492 spin_lock_irqsave(&t->sighand->siglock, flags);
493 clear_tsk_thread_flag(t, TIF_SIGPENDING);
494 flush_sigqueue(&t->pending);
495 flush_sigqueue(&t->signal->shared_pending);
496 spin_unlock_irqrestore(&t->sighand->siglock, flags);
497 }
498 EXPORT_SYMBOL(flush_signals);
499
500 #ifdef CONFIG_POSIX_TIMERS
501 static void __flush_itimer_signals(struct sigpending *pending)
502 {
503 sigset_t signal, retain;
504 struct sigqueue *q, *n;
505
506 signal = pending->signal;
507 sigemptyset(&retain);
508
509 list_for_each_entry_safe(q, n, &pending->list, list) {
510 int sig = q->info.si_signo;
511
512 if (likely(q->info.si_code != SI_TIMER)) {
513 sigaddset(&retain, sig);
514 } else {
515 sigdelset(&signal, sig);
516 list_del_init(&q->list);
517 __sigqueue_free(q);
518 }
519 }
520
521 sigorsets(&pending->signal, &signal, &retain);
522 }
523
524 void flush_itimer_signals(void)
525 {
526 struct task_struct *tsk = current;
527 unsigned long flags;
528
529 spin_lock_irqsave(&tsk->sighand->siglock, flags);
530 __flush_itimer_signals(&tsk->pending);
531 __flush_itimer_signals(&tsk->signal->shared_pending);
532 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
533 }
534 #endif
535
536 void ignore_signals(struct task_struct *t)
537 {
538 int i;
539
540 for (i = 0; i < _NSIG; ++i)
541 t->sighand->action[i].sa.sa_handler = SIG_IGN;
542
543 flush_signals(t);
544 }
545
546 /*
547 * Flush all handlers for a task.
548 */
549
550 void
551 flush_signal_handlers(struct task_struct *t, int force_default)
552 {
553 int i;
554 struct k_sigaction *ka = &t->sighand->action[0];
555 for (i = _NSIG ; i != 0 ; i--) {
556 if (force_default || ka->sa.sa_handler != SIG_IGN)
557 ka->sa.sa_handler = SIG_DFL;
558 ka->sa.sa_flags = 0;
559 #ifdef __ARCH_HAS_SA_RESTORER
560 ka->sa.sa_restorer = NULL;
561 #endif
562 sigemptyset(&ka->sa.sa_mask);
563 ka++;
564 }
565 }
566
567 bool unhandled_signal(struct task_struct *tsk, int sig)
568 {
569 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
570 if (is_global_init(tsk))
571 return true;
572
573 if (handler != SIG_IGN && handler != SIG_DFL)
574 return false;
575
576 /* if ptraced, let the tracer determine */
577 return !tsk->ptrace;
578 }
579
580 static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
581 bool *resched_timer)
582 {
583 struct sigqueue *q, *first = NULL;
584
585 /*
586 * Collect the siginfo appropriate to this signal. Check if
587 * there is another siginfo for the same signal.
588 */
589 list_for_each_entry(q, &list->list, list) {
590 if (q->info.si_signo == sig) {
591 if (first)
592 goto still_pending;
593 first = q;
594 }
595 }
596
597 sigdelset(&list->signal, sig);
598
599 if (first) {
600 still_pending:
601 list_del_init(&first->list);
602 copy_siginfo(info, &first->info);
603
604 *resched_timer =
605 (first->flags & SIGQUEUE_PREALLOC) &&
606 (info->si_code == SI_TIMER) &&
607 (info->si_sys_private);
608
609 __sigqueue_free(first);
610 } else {
611 /*
612 * Ok, it wasn't in the queue. This must be
613 * a fast-pathed signal or we must have been
614 * out of queue space. So zero out the info.
615 */
616 clear_siginfo(info);
617 info->si_signo = sig;
618 info->si_errno = 0;
619 info->si_code = SI_USER;
620 info->si_pid = 0;
621 info->si_uid = 0;
622 }
623 }
624
625 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
626 kernel_siginfo_t *info, bool *resched_timer)
627 {
628 int sig = next_signal(pending, mask);
629
630 if (sig)
631 collect_signal(sig, pending, info, resched_timer);
632 return sig;
633 }
634
635 /*
636 * Dequeue a signal and return the element to the caller, which is
637 * expected to free it.
638 *
639 * All callers have to hold the siglock.
640 */
641 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, kernel_siginfo_t *info)
642 {
643 bool resched_timer = false;
644 int signr;
645
646 /* We only dequeue private signals from ourselves, we don't let
647 * signalfd steal them
648 */
649 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
650 if (!signr) {
651 signr = __dequeue_signal(&tsk->signal->shared_pending,
652 mask, info, &resched_timer);
653 #ifdef CONFIG_POSIX_TIMERS
654 /*
655 * itimer signal ?
656 *
657 * itimers are process shared and we restart periodic
658 * itimers in the signal delivery path to prevent DoS
659 * attacks in the high resolution timer case. This is
660 * compliant with the old way of self-restarting
661 * itimers, as the SIGALRM is a legacy signal and only
662 * queued once. Changing the restart behaviour to
663 * restart the timer in the signal dequeue path is
664 * reducing the timer noise on heavy loaded !highres
665 * systems too.
666 */
667 if (unlikely(signr == SIGALRM)) {
668 struct hrtimer *tmr = &tsk->signal->real_timer;
669
670 if (!hrtimer_is_queued(tmr) &&
671 tsk->signal->it_real_incr != 0) {
672 hrtimer_forward(tmr, tmr->base->get_time(),
673 tsk->signal->it_real_incr);
674 hrtimer_restart(tmr);
675 }
676 }
677 #endif
678 }
679
680 recalc_sigpending();
681 if (!signr)
682 return 0;
683
684 if (unlikely(sig_kernel_stop(signr))) {
685 /*
686 * Set a marker that we have dequeued a stop signal. Our
687 * caller might release the siglock and then the pending
688 * stop signal it is about to process is no longer in the
689 * pending bitmasks, but must still be cleared by a SIGCONT
690 * (and overruled by a SIGKILL). So those cases clear this
691 * shared flag after we've set it. Note that this flag may
692 * remain set after the signal we return is ignored or
693 * handled. That doesn't matter because its only purpose
694 * is to alert stop-signal processing code when another
695 * processor has come along and cleared the flag.
696 */
697 current->jobctl |= JOBCTL_STOP_DEQUEUED;
698 }
699 #ifdef CONFIG_POSIX_TIMERS
700 if (resched_timer) {
701 /*
702 * Release the siglock to ensure proper locking order
703 * of timer locks outside of siglocks. Note, we leave
704 * irqs disabled here, since the posix-timers code is
705 * about to disable them again anyway.
706 */
707 spin_unlock(&tsk->sighand->siglock);
708 posixtimer_rearm(info);
709 spin_lock(&tsk->sighand->siglock);
710
711 /* Don't expose the si_sys_private value to userspace */
712 info->si_sys_private = 0;
713 }
714 #endif
715 return signr;
716 }
717 EXPORT_SYMBOL_GPL(dequeue_signal);
718
719 static int dequeue_synchronous_signal(kernel_siginfo_t *info)
720 {
721 struct task_struct *tsk = current;
722 struct sigpending *pending = &tsk->pending;
723 struct sigqueue *q, *sync = NULL;
724
725 /*
726 * Might a synchronous signal be in the queue?
727 */
728 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
729 return 0;
730
731 /*
732 * Return the first synchronous signal in the queue.
733 */
734 list_for_each_entry(q, &pending->list, list) {
735 /* Synchronous signals have a positive si_code */
736 if ((q->info.si_code > SI_USER) &&
737 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
738 sync = q;
739 goto next;
740 }
741 }
742 return 0;
743 next:
744 /*
745 * Check if there is another siginfo for the same signal.
746 */
747 list_for_each_entry_continue(q, &pending->list, list) {
748 if (q->info.si_signo == sync->info.si_signo)
749 goto still_pending;
750 }
751
752 sigdelset(&pending->signal, sync->info.si_signo);
753 recalc_sigpending();
754 still_pending:
755 list_del_init(&sync->list);
756 copy_siginfo(info, &sync->info);
757 __sigqueue_free(sync);
758 return info->si_signo;
759 }
760
761 /*
762 * Tell a process that it has a new active signal..
763 *
764 * NOTE! we rely on the previous spin_lock to
765 * lock interrupts for us! We can only be called with
766 * "siglock" held, and the local interrupt must
767 * have been disabled when that got acquired!
768 *
769 * No need to set need_resched since signal event passing
770 * goes through ->blocked
771 */
772 void signal_wake_up_state(struct task_struct *t, unsigned int state)
773 {
774 set_tsk_thread_flag(t, TIF_SIGPENDING);
775 /*
776 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
777 * case. We don't check t->state here because there is a race with it
778 * executing another processor and just now entering stopped state.
779 * By using wake_up_state, we ensure the process will wake up and
780 * handle its death signal.
781 */
782 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
783 kick_process(t);
784 }
785
786 /*
787 * Remove signals in mask from the pending set and queue.
788 * Returns 1 if any signals were found.
789 *
790 * All callers must be holding the siglock.
791 */
792 static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
793 {
794 struct sigqueue *q, *n;
795 sigset_t m;
796
797 sigandsets(&m, mask, &s->signal);
798 if (sigisemptyset(&m))
799 return;
800
801 sigandnsets(&s->signal, &s->signal, mask);
802 list_for_each_entry_safe(q, n, &s->list, list) {
803 if (sigismember(mask, q->info.si_signo)) {
804 list_del_init(&q->list);
805 __sigqueue_free(q);
806 }
807 }
808 }
809
810 static inline int is_si_special(const struct kernel_siginfo *info)
811 {
812 return info <= SEND_SIG_PRIV;
813 }
814
815 static inline bool si_fromuser(const struct kernel_siginfo *info)
816 {
817 return info == SEND_SIG_NOINFO ||
818 (!is_si_special(info) && SI_FROMUSER(info));
819 }
820
821 /*
822 * called with RCU read lock from check_kill_permission()
823 */
824 static bool kill_ok_by_cred(struct task_struct *t)
825 {
826 const struct cred *cred = current_cred();
827 const struct cred *tcred = __task_cred(t);
828
829 return uid_eq(cred->euid, tcred->suid) ||
830 uid_eq(cred->euid, tcred->uid) ||
831 uid_eq(cred->uid, tcred->suid) ||
832 uid_eq(cred->uid, tcred->uid) ||
833 ns_capable(tcred->user_ns, CAP_KILL);
834 }
835
836 /*
837 * Bad permissions for sending the signal
838 * - the caller must hold the RCU read lock
839 */
840 static int check_kill_permission(int sig, struct kernel_siginfo *info,
841 struct task_struct *t)
842 {
843 struct pid *sid;
844 int error;
845
846 if (!valid_signal(sig))
847 return -EINVAL;
848
849 if (!si_fromuser(info))
850 return 0;
851
852 error = audit_signal_info(sig, t); /* Let audit system see the signal */
853 if (error)
854 return error;
855
856 if (!same_thread_group(current, t) &&
857 !kill_ok_by_cred(t)) {
858 switch (sig) {
859 case SIGCONT:
860 sid = task_session(t);
861 /*
862 * We don't return the error if sid == NULL. The
863 * task was unhashed, the caller must notice this.
864 */
865 if (!sid || sid == task_session(current))
866 break;
867 fallthrough;
868 default:
869 return -EPERM;
870 }
871 }
872
873 return security_task_kill(t, info, sig, NULL);
874 }
875
876 /**
877 * ptrace_trap_notify - schedule trap to notify ptracer
878 * @t: tracee wanting to notify tracer
879 *
880 * This function schedules sticky ptrace trap which is cleared on the next
881 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
882 * ptracer.
883 *
884 * If @t is running, STOP trap will be taken. If trapped for STOP and
885 * ptracer is listening for events, tracee is woken up so that it can
886 * re-trap for the new event. If trapped otherwise, STOP trap will be
887 * eventually taken without returning to userland after the existing traps
888 * are finished by PTRACE_CONT.
889 *
890 * CONTEXT:
891 * Must be called with @task->sighand->siglock held.
892 */
893 static void ptrace_trap_notify(struct task_struct *t)
894 {
895 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
896 assert_spin_locked(&t->sighand->siglock);
897
898 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
899 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
900 }
901
902 /*
903 * Handle magic process-wide effects of stop/continue signals. Unlike
904 * the signal actions, these happen immediately at signal-generation
905 * time regardless of blocking, ignoring, or handling. This does the
906 * actual continuing for SIGCONT, but not the actual stopping for stop
907 * signals. The process stop is done as a signal action for SIG_DFL.
908 *
909 * Returns true if the signal should be actually delivered, otherwise
910 * it should be dropped.
911 */
912 static bool prepare_signal(int sig, struct task_struct *p, bool force)
913 {
914 struct signal_struct *signal = p->signal;
915 struct task_struct *t;
916 sigset_t flush;
917
918 if (signal->flags & (SIGNAL_GROUP_EXIT | SIGNAL_GROUP_COREDUMP)) {
919 if (!(signal->flags & SIGNAL_GROUP_EXIT))
920 return sig == SIGKILL;
921 /*
922 * The process is in the middle of dying, nothing to do.
923 */
924 } else if (sig_kernel_stop(sig)) {
925 /*
926 * This is a stop signal. Remove SIGCONT from all queues.
927 */
928 siginitset(&flush, sigmask(SIGCONT));
929 flush_sigqueue_mask(&flush, &signal->shared_pending);
930 for_each_thread(p, t)
931 flush_sigqueue_mask(&flush, &t->pending);
932 } else if (sig == SIGCONT) {
933 unsigned int why;
934 /*
935 * Remove all stop signals from all queues, wake all threads.
936 */
937 siginitset(&flush, SIG_KERNEL_STOP_MASK);
938 flush_sigqueue_mask(&flush, &signal->shared_pending);
939 for_each_thread(p, t) {
940 flush_sigqueue_mask(&flush, &t->pending);
941 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
942 if (likely(!(t->ptrace & PT_SEIZED)))
943 wake_up_state(t, __TASK_STOPPED);
944 else
945 ptrace_trap_notify(t);
946 }
947
948 /*
949 * Notify the parent with CLD_CONTINUED if we were stopped.
950 *
951 * If we were in the middle of a group stop, we pretend it
952 * was already finished, and then continued. Since SIGCHLD
953 * doesn't queue we report only CLD_STOPPED, as if the next
954 * CLD_CONTINUED was dropped.
955 */
956 why = 0;
957 if (signal->flags & SIGNAL_STOP_STOPPED)
958 why |= SIGNAL_CLD_CONTINUED;
959 else if (signal->group_stop_count)
960 why |= SIGNAL_CLD_STOPPED;
961
962 if (why) {
963 /*
964 * The first thread which returns from do_signal_stop()
965 * will take ->siglock, notice SIGNAL_CLD_MASK, and
966 * notify its parent. See get_signal().
967 */
968 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
969 signal->group_stop_count = 0;
970 signal->group_exit_code = 0;
971 }
972 }
973
974 return !sig_ignored(p, sig, force);
975 }
976
977 /*
978 * Test if P wants to take SIG. After we've checked all threads with this,
979 * it's equivalent to finding no threads not blocking SIG. Any threads not
980 * blocking SIG were ruled out because they are not running and already
981 * have pending signals. Such threads will dequeue from the shared queue
982 * as soon as they're available, so putting the signal on the shared queue
983 * will be equivalent to sending it to one such thread.
984 */
985 static inline bool wants_signal(int sig, struct task_struct *p)
986 {
987 if (sigismember(&p->blocked, sig))
988 return false;
989
990 if (p->flags & PF_EXITING)
991 return false;
992
993 if (sig == SIGKILL)
994 return true;
995
996 if (task_is_stopped_or_traced(p))
997 return false;
998
999 return task_curr(p) || !task_sigpending(p);
1000 }
1001
1002 static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
1003 {
1004 struct signal_struct *signal = p->signal;
1005 struct task_struct *t;
1006
1007 /*
1008 * Now find a thread we can wake up to take the signal off the queue.
1009 *
1010 * If the main thread wants the signal, it gets first crack.
1011 * Probably the least surprising to the average bear.
1012 */
1013 if (wants_signal(sig, p))
1014 t = p;
1015 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1016 /*
1017 * There is just one thread and it does not need to be woken.
1018 * It will dequeue unblocked signals before it runs again.
1019 */
1020 return;
1021 else {
1022 /*
1023 * Otherwise try to find a suitable thread.
1024 */
1025 t = signal->curr_target;
1026 while (!wants_signal(sig, t)) {
1027 t = next_thread(t);
1028 if (t == signal->curr_target)
1029 /*
1030 * No thread needs to be woken.
1031 * Any eligible threads will see
1032 * the signal in the queue soon.
1033 */
1034 return;
1035 }
1036 signal->curr_target = t;
1037 }
1038
1039 /*
1040 * Found a killable thread. If the signal will be fatal,
1041 * then start taking the whole group down immediately.
1042 */
1043 if (sig_fatal(p, sig) &&
1044 !(signal->flags & SIGNAL_GROUP_EXIT) &&
1045 !sigismember(&t->real_blocked, sig) &&
1046 (sig == SIGKILL || !p->ptrace)) {
1047 /*
1048 * This signal will be fatal to the whole group.
1049 */
1050 if (!sig_kernel_coredump(sig)) {
1051 /*
1052 * Start a group exit and wake everybody up.
1053 * This way we don't have other threads
1054 * running and doing things after a slower
1055 * thread has the fatal signal pending.
1056 */
1057 signal->flags = SIGNAL_GROUP_EXIT;
1058 signal->group_exit_code = sig;
1059 signal->group_stop_count = 0;
1060 t = p;
1061 do {
1062 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1063 sigaddset(&t->pending.signal, SIGKILL);
1064 signal_wake_up(t, 1);
1065 } while_each_thread(p, t);
1066 return;
1067 }
1068 }
1069
1070 /*
1071 * The signal is already in the shared-pending queue.
1072 * Tell the chosen thread to wake up and dequeue it.
1073 */
1074 signal_wake_up(t, sig == SIGKILL);
1075 return;
1076 }
1077
1078 static inline bool legacy_queue(struct sigpending *signals, int sig)
1079 {
1080 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1081 }
1082
1083 static int __send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1084 enum pid_type type, bool force)
1085 {
1086 struct sigpending *pending;
1087 struct sigqueue *q;
1088 int override_rlimit;
1089 int ret = 0, result;
1090
1091 assert_spin_locked(&t->sighand->siglock);
1092
1093 result = TRACE_SIGNAL_IGNORED;
1094 if (!prepare_signal(sig, t, force))
1095 goto ret;
1096
1097 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1098 /*
1099 * Short-circuit ignored signals and support queuing
1100 * exactly one non-rt signal, so that we can get more
1101 * detailed information about the cause of the signal.
1102 */
1103 result = TRACE_SIGNAL_ALREADY_PENDING;
1104 if (legacy_queue(pending, sig))
1105 goto ret;
1106
1107 result = TRACE_SIGNAL_DELIVERED;
1108 /*
1109 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1110 */
1111 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1112 goto out_set;
1113
1114 /*
1115 * Real-time signals must be queued if sent by sigqueue, or
1116 * some other real-time mechanism. It is implementation
1117 * defined whether kill() does so. We attempt to do so, on
1118 * the principle of least surprise, but since kill is not
1119 * allowed to fail with EAGAIN when low on memory we just
1120 * make sure at least one signal gets delivered and don't
1121 * pass on the info struct.
1122 */
1123 if (sig < SIGRTMIN)
1124 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1125 else
1126 override_rlimit = 0;
1127
1128 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit, 0);
1129
1130 if (q) {
1131 list_add_tail(&q->list, &pending->list);
1132 switch ((unsigned long) info) {
1133 case (unsigned long) SEND_SIG_NOINFO:
1134 clear_siginfo(&q->info);
1135 q->info.si_signo = sig;
1136 q->info.si_errno = 0;
1137 q->info.si_code = SI_USER;
1138 q->info.si_pid = task_tgid_nr_ns(current,
1139 task_active_pid_ns(t));
1140 rcu_read_lock();
1141 q->info.si_uid =
1142 from_kuid_munged(task_cred_xxx(t, user_ns),
1143 current_uid());
1144 rcu_read_unlock();
1145 break;
1146 case (unsigned long) SEND_SIG_PRIV:
1147 clear_siginfo(&q->info);
1148 q->info.si_signo = sig;
1149 q->info.si_errno = 0;
1150 q->info.si_code = SI_KERNEL;
1151 q->info.si_pid = 0;
1152 q->info.si_uid = 0;
1153 break;
1154 default:
1155 copy_siginfo(&q->info, info);
1156 break;
1157 }
1158 } else if (!is_si_special(info) &&
1159 sig >= SIGRTMIN && info->si_code != SI_USER) {
1160 /*
1161 * Queue overflow, abort. We may abort if the
1162 * signal was rt and sent by user using something
1163 * other than kill().
1164 */
1165 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1166 ret = -EAGAIN;
1167 goto ret;
1168 } else {
1169 /*
1170 * This is a silent loss of information. We still
1171 * send the signal, but the *info bits are lost.
1172 */
1173 result = TRACE_SIGNAL_LOSE_INFO;
1174 }
1175
1176 out_set:
1177 signalfd_notify(t, sig);
1178 sigaddset(&pending->signal, sig);
1179
1180 /* Let multiprocess signals appear after on-going forks */
1181 if (type > PIDTYPE_TGID) {
1182 struct multiprocess_signals *delayed;
1183 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1184 sigset_t *signal = &delayed->signal;
1185 /* Can't queue both a stop and a continue signal */
1186 if (sig == SIGCONT)
1187 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1188 else if (sig_kernel_stop(sig))
1189 sigdelset(signal, SIGCONT);
1190 sigaddset(signal, sig);
1191 }
1192 }
1193
1194 complete_signal(sig, t, type);
1195 ret:
1196 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1197 return ret;
1198 }
1199
1200 static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1201 {
1202 bool ret = false;
1203 switch (siginfo_layout(info->si_signo, info->si_code)) {
1204 case SIL_KILL:
1205 case SIL_CHLD:
1206 case SIL_RT:
1207 ret = true;
1208 break;
1209 case SIL_TIMER:
1210 case SIL_POLL:
1211 case SIL_FAULT:
1212 case SIL_FAULT_TRAPNO:
1213 case SIL_FAULT_MCEERR:
1214 case SIL_FAULT_BNDERR:
1215 case SIL_FAULT_PKUERR:
1216 case SIL_PERF_EVENT:
1217 case SIL_SYS:
1218 ret = false;
1219 break;
1220 }
1221 return ret;
1222 }
1223
1224 static int send_signal(int sig, struct kernel_siginfo *info, struct task_struct *t,
1225 enum pid_type type)
1226 {
1227 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1228 bool force = false;
1229
1230 if (info == SEND_SIG_NOINFO) {
1231 /* Force if sent from an ancestor pid namespace */
1232 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1233 } else if (info == SEND_SIG_PRIV) {
1234 /* Don't ignore kernel generated signals */
1235 force = true;
1236 } else if (has_si_pid_and_uid(info)) {
1237 /* SIGKILL and SIGSTOP is special or has ids */
1238 struct user_namespace *t_user_ns;
1239
1240 rcu_read_lock();
1241 t_user_ns = task_cred_xxx(t, user_ns);
1242 if (current_user_ns() != t_user_ns) {
1243 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1244 info->si_uid = from_kuid_munged(t_user_ns, uid);
1245 }
1246 rcu_read_unlock();
1247
1248 /* A kernel generated signal? */
1249 force = (info->si_code == SI_KERNEL);
1250
1251 /* From an ancestor pid namespace? */
1252 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1253 info->si_pid = 0;
1254 force = true;
1255 }
1256 }
1257 return __send_signal(sig, info, t, type, force);
1258 }
1259
1260 static void print_fatal_signal(int signr)
1261 {
1262 struct pt_regs *regs = signal_pt_regs();
1263 pr_info("potentially unexpected fatal signal %d.\n", signr);
1264
1265 #if defined(__i386__) && !defined(__arch_um__)
1266 pr_info("code at %08lx: ", regs->ip);
1267 {
1268 int i;
1269 for (i = 0; i < 16; i++) {
1270 unsigned char insn;
1271
1272 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1273 break;
1274 pr_cont("%02x ", insn);
1275 }
1276 }
1277 pr_cont("\n");
1278 #endif
1279 preempt_disable();
1280 show_regs(regs);
1281 preempt_enable();
1282 }
1283
1284 static int __init setup_print_fatal_signals(char *str)
1285 {
1286 get_option (&str, &print_fatal_signals);
1287
1288 return 1;
1289 }
1290
1291 __setup("print-fatal-signals=", setup_print_fatal_signals);
1292
1293 int
1294 __group_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1295 {
1296 return send_signal(sig, info, p, PIDTYPE_TGID);
1297 }
1298
1299 int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1300 enum pid_type type)
1301 {
1302 unsigned long flags;
1303 int ret = -ESRCH;
1304
1305 if (lock_task_sighand(p, &flags)) {
1306 ret = send_signal(sig, info, p, type);
1307 unlock_task_sighand(p, &flags);
1308 }
1309
1310 return ret;
1311 }
1312
1313 /*
1314 * Force a signal that the process can't ignore: if necessary
1315 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1316 *
1317 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1318 * since we do not want to have a signal handler that was blocked
1319 * be invoked when user space had explicitly blocked it.
1320 *
1321 * We don't want to have recursive SIGSEGV's etc, for example,
1322 * that is why we also clear SIGNAL_UNKILLABLE.
1323 */
1324 static int
1325 force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t)
1326 {
1327 unsigned long int flags;
1328 int ret, blocked, ignored;
1329 struct k_sigaction *action;
1330 int sig = info->si_signo;
1331
1332 spin_lock_irqsave(&t->sighand->siglock, flags);
1333 action = &t->sighand->action[sig-1];
1334 ignored = action->sa.sa_handler == SIG_IGN;
1335 blocked = sigismember(&t->blocked, sig);
1336 if (blocked || ignored) {
1337 action->sa.sa_handler = SIG_DFL;
1338 if (blocked) {
1339 sigdelset(&t->blocked, sig);
1340 recalc_sigpending_and_wake(t);
1341 }
1342 }
1343 /*
1344 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1345 * debugging to leave init killable.
1346 */
1347 if (action->sa.sa_handler == SIG_DFL && !t->ptrace)
1348 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1349 ret = send_signal(sig, info, t, PIDTYPE_PID);
1350 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1351
1352 return ret;
1353 }
1354
1355 int force_sig_info(struct kernel_siginfo *info)
1356 {
1357 return force_sig_info_to_task(info, current);
1358 }
1359
1360 /*
1361 * Nuke all other threads in the group.
1362 */
1363 int zap_other_threads(struct task_struct *p)
1364 {
1365 struct task_struct *t = p;
1366 int count = 0;
1367
1368 p->signal->group_stop_count = 0;
1369
1370 while_each_thread(p, t) {
1371 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1372 count++;
1373
1374 /* Don't bother with already dead threads */
1375 if (t->exit_state)
1376 continue;
1377 sigaddset(&t->pending.signal, SIGKILL);
1378 signal_wake_up(t, 1);
1379 }
1380
1381 return count;
1382 }
1383
1384 struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1385 unsigned long *flags)
1386 {
1387 struct sighand_struct *sighand;
1388
1389 rcu_read_lock();
1390 for (;;) {
1391 sighand = rcu_dereference(tsk->sighand);
1392 if (unlikely(sighand == NULL))
1393 break;
1394
1395 /*
1396 * This sighand can be already freed and even reused, but
1397 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1398 * initializes ->siglock: this slab can't go away, it has
1399 * the same object type, ->siglock can't be reinitialized.
1400 *
1401 * We need to ensure that tsk->sighand is still the same
1402 * after we take the lock, we can race with de_thread() or
1403 * __exit_signal(). In the latter case the next iteration
1404 * must see ->sighand == NULL.
1405 */
1406 spin_lock_irqsave(&sighand->siglock, *flags);
1407 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1408 break;
1409 spin_unlock_irqrestore(&sighand->siglock, *flags);
1410 }
1411 rcu_read_unlock();
1412
1413 return sighand;
1414 }
1415
1416 /*
1417 * send signal info to all the members of a group
1418 */
1419 int group_send_sig_info(int sig, struct kernel_siginfo *info,
1420 struct task_struct *p, enum pid_type type)
1421 {
1422 int ret;
1423
1424 rcu_read_lock();
1425 ret = check_kill_permission(sig, info, p);
1426 rcu_read_unlock();
1427
1428 if (!ret && sig)
1429 ret = do_send_sig_info(sig, info, p, type);
1430
1431 return ret;
1432 }
1433
1434 /*
1435 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1436 * control characters do (^C, ^Z etc)
1437 * - the caller must hold at least a readlock on tasklist_lock
1438 */
1439 int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1440 {
1441 struct task_struct *p = NULL;
1442 int retval, success;
1443
1444 success = 0;
1445 retval = -ESRCH;
1446 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1447 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1448 success |= !err;
1449 retval = err;
1450 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1451 return success ? 0 : retval;
1452 }
1453
1454 int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1455 {
1456 int error = -ESRCH;
1457 struct task_struct *p;
1458
1459 for (;;) {
1460 rcu_read_lock();
1461 p = pid_task(pid, PIDTYPE_PID);
1462 if (p)
1463 error = group_send_sig_info(sig, info, p, PIDTYPE_TGID);
1464 rcu_read_unlock();
1465 if (likely(!p || error != -ESRCH))
1466 return error;
1467
1468 /*
1469 * The task was unhashed in between, try again. If it
1470 * is dead, pid_task() will return NULL, if we race with
1471 * de_thread() it will find the new leader.
1472 */
1473 }
1474 }
1475
1476 static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1477 {
1478 int error;
1479 rcu_read_lock();
1480 error = kill_pid_info(sig, info, find_vpid(pid));
1481 rcu_read_unlock();
1482 return error;
1483 }
1484
1485 static inline bool kill_as_cred_perm(const struct cred *cred,
1486 struct task_struct *target)
1487 {
1488 const struct cred *pcred = __task_cred(target);
1489
1490 return uid_eq(cred->euid, pcred->suid) ||
1491 uid_eq(cred->euid, pcred->uid) ||
1492 uid_eq(cred->uid, pcred->suid) ||
1493 uid_eq(cred->uid, pcred->uid);
1494 }
1495
1496 /*
1497 * The usb asyncio usage of siginfo is wrong. The glibc support
1498 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1499 * AKA after the generic fields:
1500 * kernel_pid_t si_pid;
1501 * kernel_uid32_t si_uid;
1502 * sigval_t si_value;
1503 *
1504 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1505 * after the generic fields is:
1506 * void __user *si_addr;
1507 *
1508 * This is a practical problem when there is a 64bit big endian kernel
1509 * and a 32bit userspace. As the 32bit address will encoded in the low
1510 * 32bits of the pointer. Those low 32bits will be stored at higher
1511 * address than appear in a 32 bit pointer. So userspace will not
1512 * see the address it was expecting for it's completions.
1513 *
1514 * There is nothing in the encoding that can allow
1515 * copy_siginfo_to_user32 to detect this confusion of formats, so
1516 * handle this by requiring the caller of kill_pid_usb_asyncio to
1517 * notice when this situration takes place and to store the 32bit
1518 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1519 * parameter.
1520 */
1521 int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1522 struct pid *pid, const struct cred *cred)
1523 {
1524 struct kernel_siginfo info;
1525 struct task_struct *p;
1526 unsigned long flags;
1527 int ret = -EINVAL;
1528
1529 if (!valid_signal(sig))
1530 return ret;
1531
1532 clear_siginfo(&info);
1533 info.si_signo = sig;
1534 info.si_errno = errno;
1535 info.si_code = SI_ASYNCIO;
1536 *((sigval_t *)&info.si_pid) = addr;
1537
1538 rcu_read_lock();
1539 p = pid_task(pid, PIDTYPE_PID);
1540 if (!p) {
1541 ret = -ESRCH;
1542 goto out_unlock;
1543 }
1544 if (!kill_as_cred_perm(cred, p)) {
1545 ret = -EPERM;
1546 goto out_unlock;
1547 }
1548 ret = security_task_kill(p, &info, sig, cred);
1549 if (ret)
1550 goto out_unlock;
1551
1552 if (sig) {
1553 if (lock_task_sighand(p, &flags)) {
1554 ret = __send_signal(sig, &info, p, PIDTYPE_TGID, false);
1555 unlock_task_sighand(p, &flags);
1556 } else
1557 ret = -ESRCH;
1558 }
1559 out_unlock:
1560 rcu_read_unlock();
1561 return ret;
1562 }
1563 EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1564
1565 /*
1566 * kill_something_info() interprets pid in interesting ways just like kill(2).
1567 *
1568 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1569 * is probably wrong. Should make it like BSD or SYSV.
1570 */
1571
1572 static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1573 {
1574 int ret;
1575
1576 if (pid > 0)
1577 return kill_proc_info(sig, info, pid);
1578
1579 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1580 if (pid == INT_MIN)
1581 return -ESRCH;
1582
1583 read_lock(&tasklist_lock);
1584 if (pid != -1) {
1585 ret = __kill_pgrp_info(sig, info,
1586 pid ? find_vpid(-pid) : task_pgrp(current));
1587 } else {
1588 int retval = 0, count = 0;
1589 struct task_struct * p;
1590
1591 for_each_process(p) {
1592 if (task_pid_vnr(p) > 1 &&
1593 !same_thread_group(p, current)) {
1594 int err = group_send_sig_info(sig, info, p,
1595 PIDTYPE_MAX);
1596 ++count;
1597 if (err != -EPERM)
1598 retval = err;
1599 }
1600 }
1601 ret = count ? retval : -ESRCH;
1602 }
1603 read_unlock(&tasklist_lock);
1604
1605 return ret;
1606 }
1607
1608 /*
1609 * These are for backward compatibility with the rest of the kernel source.
1610 */
1611
1612 int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1613 {
1614 /*
1615 * Make sure legacy kernel users don't send in bad values
1616 * (normal paths check this in check_kill_permission).
1617 */
1618 if (!valid_signal(sig))
1619 return -EINVAL;
1620
1621 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1622 }
1623 EXPORT_SYMBOL(send_sig_info);
1624
1625 #define __si_special(priv) \
1626 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1627
1628 int
1629 send_sig(int sig, struct task_struct *p, int priv)
1630 {
1631 return send_sig_info(sig, __si_special(priv), p);
1632 }
1633 EXPORT_SYMBOL(send_sig);
1634
1635 void force_sig(int sig)
1636 {
1637 struct kernel_siginfo info;
1638
1639 clear_siginfo(&info);
1640 info.si_signo = sig;
1641 info.si_errno = 0;
1642 info.si_code = SI_KERNEL;
1643 info.si_pid = 0;
1644 info.si_uid = 0;
1645 force_sig_info(&info);
1646 }
1647 EXPORT_SYMBOL(force_sig);
1648
1649 /*
1650 * When things go south during signal handling, we
1651 * will force a SIGSEGV. And if the signal that caused
1652 * the problem was already a SIGSEGV, we'll want to
1653 * make sure we don't even try to deliver the signal..
1654 */
1655 void force_sigsegv(int sig)
1656 {
1657 struct task_struct *p = current;
1658
1659 if (sig == SIGSEGV) {
1660 unsigned long flags;
1661 spin_lock_irqsave(&p->sighand->siglock, flags);
1662 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1663 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1664 }
1665 force_sig(SIGSEGV);
1666 }
1667
1668 int force_sig_fault_to_task(int sig, int code, void __user *addr
1669 ___ARCH_SI_TRAPNO(int trapno)
1670 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1671 , struct task_struct *t)
1672 {
1673 struct kernel_siginfo info;
1674
1675 clear_siginfo(&info);
1676 info.si_signo = sig;
1677 info.si_errno = 0;
1678 info.si_code = code;
1679 info.si_addr = addr;
1680 #ifdef __ARCH_SI_TRAPNO
1681 info.si_trapno = trapno;
1682 #endif
1683 #ifdef __ia64__
1684 info.si_imm = imm;
1685 info.si_flags = flags;
1686 info.si_isr = isr;
1687 #endif
1688 return force_sig_info_to_task(&info, t);
1689 }
1690
1691 int force_sig_fault(int sig, int code, void __user *addr
1692 ___ARCH_SI_TRAPNO(int trapno)
1693 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr))
1694 {
1695 return force_sig_fault_to_task(sig, code, addr
1696 ___ARCH_SI_TRAPNO(trapno)
1697 ___ARCH_SI_IA64(imm, flags, isr), current);
1698 }
1699
1700 int send_sig_fault(int sig, int code, void __user *addr
1701 ___ARCH_SI_TRAPNO(int trapno)
1702 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1703 , struct task_struct *t)
1704 {
1705 struct kernel_siginfo info;
1706
1707 clear_siginfo(&info);
1708 info.si_signo = sig;
1709 info.si_errno = 0;
1710 info.si_code = code;
1711 info.si_addr = addr;
1712 #ifdef __ARCH_SI_TRAPNO
1713 info.si_trapno = trapno;
1714 #endif
1715 #ifdef __ia64__
1716 info.si_imm = imm;
1717 info.si_flags = flags;
1718 info.si_isr = isr;
1719 #endif
1720 return send_sig_info(info.si_signo, &info, t);
1721 }
1722
1723 int force_sig_mceerr(int code, void __user *addr, short lsb)
1724 {
1725 struct kernel_siginfo info;
1726
1727 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1728 clear_siginfo(&info);
1729 info.si_signo = SIGBUS;
1730 info.si_errno = 0;
1731 info.si_code = code;
1732 info.si_addr = addr;
1733 info.si_addr_lsb = lsb;
1734 return force_sig_info(&info);
1735 }
1736
1737 int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1738 {
1739 struct kernel_siginfo info;
1740
1741 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1742 clear_siginfo(&info);
1743 info.si_signo = SIGBUS;
1744 info.si_errno = 0;
1745 info.si_code = code;
1746 info.si_addr = addr;
1747 info.si_addr_lsb = lsb;
1748 return send_sig_info(info.si_signo, &info, t);
1749 }
1750 EXPORT_SYMBOL(send_sig_mceerr);
1751
1752 int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1753 {
1754 struct kernel_siginfo info;
1755
1756 clear_siginfo(&info);
1757 info.si_signo = SIGSEGV;
1758 info.si_errno = 0;
1759 info.si_code = SEGV_BNDERR;
1760 info.si_addr = addr;
1761 info.si_lower = lower;
1762 info.si_upper = upper;
1763 return force_sig_info(&info);
1764 }
1765
1766 #ifdef SEGV_PKUERR
1767 int force_sig_pkuerr(void __user *addr, u32 pkey)
1768 {
1769 struct kernel_siginfo info;
1770
1771 clear_siginfo(&info);
1772 info.si_signo = SIGSEGV;
1773 info.si_errno = 0;
1774 info.si_code = SEGV_PKUERR;
1775 info.si_addr = addr;
1776 info.si_pkey = pkey;
1777 return force_sig_info(&info);
1778 }
1779 #endif
1780
1781 int force_sig_perf(void __user *addr, u32 type, u64 sig_data)
1782 {
1783 struct kernel_siginfo info;
1784
1785 clear_siginfo(&info);
1786 info.si_signo = SIGTRAP;
1787 info.si_errno = 0;
1788 info.si_code = TRAP_PERF;
1789 info.si_addr = addr;
1790 info.si_perf_data = sig_data;
1791 info.si_perf_type = type;
1792
1793 return force_sig_info(&info);
1794 }
1795
1796 /* For the crazy architectures that include trap information in
1797 * the errno field, instead of an actual errno value.
1798 */
1799 int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1800 {
1801 struct kernel_siginfo info;
1802
1803 clear_siginfo(&info);
1804 info.si_signo = SIGTRAP;
1805 info.si_errno = errno;
1806 info.si_code = TRAP_HWBKPT;
1807 info.si_addr = addr;
1808 return force_sig_info(&info);
1809 }
1810
1811 int kill_pgrp(struct pid *pid, int sig, int priv)
1812 {
1813 int ret;
1814
1815 read_lock(&tasklist_lock);
1816 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1817 read_unlock(&tasklist_lock);
1818
1819 return ret;
1820 }
1821 EXPORT_SYMBOL(kill_pgrp);
1822
1823 int kill_pid(struct pid *pid, int sig, int priv)
1824 {
1825 return kill_pid_info(sig, __si_special(priv), pid);
1826 }
1827 EXPORT_SYMBOL(kill_pid);
1828
1829 /*
1830 * These functions support sending signals using preallocated sigqueue
1831 * structures. This is needed "because realtime applications cannot
1832 * afford to lose notifications of asynchronous events, like timer
1833 * expirations or I/O completions". In the case of POSIX Timers
1834 * we allocate the sigqueue structure from the timer_create. If this
1835 * allocation fails we are able to report the failure to the application
1836 * with an EAGAIN error.
1837 */
1838 struct sigqueue *sigqueue_alloc(void)
1839 {
1840 return __sigqueue_alloc(-1, current, GFP_KERNEL, 0, SIGQUEUE_PREALLOC);
1841 }
1842
1843 void sigqueue_free(struct sigqueue *q)
1844 {
1845 unsigned long flags;
1846 spinlock_t *lock = &current->sighand->siglock;
1847
1848 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1849 /*
1850 * We must hold ->siglock while testing q->list
1851 * to serialize with collect_signal() or with
1852 * __exit_signal()->flush_sigqueue().
1853 */
1854 spin_lock_irqsave(lock, flags);
1855 q->flags &= ~SIGQUEUE_PREALLOC;
1856 /*
1857 * If it is queued it will be freed when dequeued,
1858 * like the "regular" sigqueue.
1859 */
1860 if (!list_empty(&q->list))
1861 q = NULL;
1862 spin_unlock_irqrestore(lock, flags);
1863
1864 if (q)
1865 __sigqueue_free(q);
1866 }
1867
1868 int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1869 {
1870 int sig = q->info.si_signo;
1871 struct sigpending *pending;
1872 struct task_struct *t;
1873 unsigned long flags;
1874 int ret, result;
1875
1876 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1877
1878 ret = -1;
1879 rcu_read_lock();
1880 t = pid_task(pid, type);
1881 if (!t || !likely(lock_task_sighand(t, &flags)))
1882 goto ret;
1883
1884 ret = 1; /* the signal is ignored */
1885 result = TRACE_SIGNAL_IGNORED;
1886 if (!prepare_signal(sig, t, false))
1887 goto out;
1888
1889 ret = 0;
1890 if (unlikely(!list_empty(&q->list))) {
1891 /*
1892 * If an SI_TIMER entry is already queue just increment
1893 * the overrun count.
1894 */
1895 BUG_ON(q->info.si_code != SI_TIMER);
1896 q->info.si_overrun++;
1897 result = TRACE_SIGNAL_ALREADY_PENDING;
1898 goto out;
1899 }
1900 q->info.si_overrun = 0;
1901
1902 signalfd_notify(t, sig);
1903 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1904 list_add_tail(&q->list, &pending->list);
1905 sigaddset(&pending->signal, sig);
1906 complete_signal(sig, t, type);
1907 result = TRACE_SIGNAL_DELIVERED;
1908 out:
1909 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
1910 unlock_task_sighand(t, &flags);
1911 ret:
1912 rcu_read_unlock();
1913 return ret;
1914 }
1915
1916 static void do_notify_pidfd(struct task_struct *task)
1917 {
1918 struct pid *pid;
1919
1920 WARN_ON(task->exit_state == 0);
1921 pid = task_pid(task);
1922 wake_up_all(&pid->wait_pidfd);
1923 }
1924
1925 /*
1926 * Let a parent know about the death of a child.
1927 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1928 *
1929 * Returns true if our parent ignored us and so we've switched to
1930 * self-reaping.
1931 */
1932 bool do_notify_parent(struct task_struct *tsk, int sig)
1933 {
1934 struct kernel_siginfo info;
1935 unsigned long flags;
1936 struct sighand_struct *psig;
1937 bool autoreap = false;
1938 u64 utime, stime;
1939
1940 BUG_ON(sig == -1);
1941
1942 /* do_notify_parent_cldstop should have been called instead. */
1943 BUG_ON(task_is_stopped_or_traced(tsk));
1944
1945 BUG_ON(!tsk->ptrace &&
1946 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1947
1948 /* Wake up all pidfd waiters */
1949 do_notify_pidfd(tsk);
1950
1951 if (sig != SIGCHLD) {
1952 /*
1953 * This is only possible if parent == real_parent.
1954 * Check if it has changed security domain.
1955 */
1956 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
1957 sig = SIGCHLD;
1958 }
1959
1960 clear_siginfo(&info);
1961 info.si_signo = sig;
1962 info.si_errno = 0;
1963 /*
1964 * We are under tasklist_lock here so our parent is tied to
1965 * us and cannot change.
1966 *
1967 * task_active_pid_ns will always return the same pid namespace
1968 * until a task passes through release_task.
1969 *
1970 * write_lock() currently calls preempt_disable() which is the
1971 * same as rcu_read_lock(), but according to Oleg, this is not
1972 * correct to rely on this
1973 */
1974 rcu_read_lock();
1975 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
1976 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
1977 task_uid(tsk));
1978 rcu_read_unlock();
1979
1980 task_cputime(tsk, &utime, &stime);
1981 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
1982 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
1983
1984 info.si_status = tsk->exit_code & 0x7f;
1985 if (tsk->exit_code & 0x80)
1986 info.si_code = CLD_DUMPED;
1987 else if (tsk->exit_code & 0x7f)
1988 info.si_code = CLD_KILLED;
1989 else {
1990 info.si_code = CLD_EXITED;
1991 info.si_status = tsk->exit_code >> 8;
1992 }
1993
1994 psig = tsk->parent->sighand;
1995 spin_lock_irqsave(&psig->siglock, flags);
1996 if (!tsk->ptrace && sig == SIGCHLD &&
1997 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1998 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1999 /*
2000 * We are exiting and our parent doesn't care. POSIX.1
2001 * defines special semantics for setting SIGCHLD to SIG_IGN
2002 * or setting the SA_NOCLDWAIT flag: we should be reaped
2003 * automatically and not left for our parent's wait4 call.
2004 * Rather than having the parent do it as a magic kind of
2005 * signal handler, we just set this to tell do_exit that we
2006 * can be cleaned up without becoming a zombie. Note that
2007 * we still call __wake_up_parent in this case, because a
2008 * blocked sys_wait4 might now return -ECHILD.
2009 *
2010 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
2011 * is implementation-defined: we do (if you don't want
2012 * it, just use SIG_IGN instead).
2013 */
2014 autoreap = true;
2015 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2016 sig = 0;
2017 }
2018 /*
2019 * Send with __send_signal as si_pid and si_uid are in the
2020 * parent's namespaces.
2021 */
2022 if (valid_signal(sig) && sig)
2023 __send_signal(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2024 __wake_up_parent(tsk, tsk->parent);
2025 spin_unlock_irqrestore(&psig->siglock, flags);
2026
2027 return autoreap;
2028 }
2029
2030 /**
2031 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2032 * @tsk: task reporting the state change
2033 * @for_ptracer: the notification is for ptracer
2034 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2035 *
2036 * Notify @tsk's parent that the stopped/continued state has changed. If
2037 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2038 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2039 *
2040 * CONTEXT:
2041 * Must be called with tasklist_lock at least read locked.
2042 */
2043 static void do_notify_parent_cldstop(struct task_struct *tsk,
2044 bool for_ptracer, int why)
2045 {
2046 struct kernel_siginfo info;
2047 unsigned long flags;
2048 struct task_struct *parent;
2049 struct sighand_struct *sighand;
2050 u64 utime, stime;
2051
2052 if (for_ptracer) {
2053 parent = tsk->parent;
2054 } else {
2055 tsk = tsk->group_leader;
2056 parent = tsk->real_parent;
2057 }
2058
2059 clear_siginfo(&info);
2060 info.si_signo = SIGCHLD;
2061 info.si_errno = 0;
2062 /*
2063 * see comment in do_notify_parent() about the following 4 lines
2064 */
2065 rcu_read_lock();
2066 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2067 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2068 rcu_read_unlock();
2069
2070 task_cputime(tsk, &utime, &stime);
2071 info.si_utime = nsec_to_clock_t(utime);
2072 info.si_stime = nsec_to_clock_t(stime);
2073
2074 info.si_code = why;
2075 switch (why) {
2076 case CLD_CONTINUED:
2077 info.si_status = SIGCONT;
2078 break;
2079 case CLD_STOPPED:
2080 info.si_status = tsk->signal->group_exit_code & 0x7f;
2081 break;
2082 case CLD_TRAPPED:
2083 info.si_status = tsk->exit_code & 0x7f;
2084 break;
2085 default:
2086 BUG();
2087 }
2088
2089 sighand = parent->sighand;
2090 spin_lock_irqsave(&sighand->siglock, flags);
2091 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2092 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2093 __group_send_sig_info(SIGCHLD, &info, parent);
2094 /*
2095 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2096 */
2097 __wake_up_parent(tsk, parent);
2098 spin_unlock_irqrestore(&sighand->siglock, flags);
2099 }
2100
2101 static inline bool may_ptrace_stop(void)
2102 {
2103 if (!likely(current->ptrace))
2104 return false;
2105 /*
2106 * Are we in the middle of do_coredump?
2107 * If so and our tracer is also part of the coredump stopping
2108 * is a deadlock situation, and pointless because our tracer
2109 * is dead so don't allow us to stop.
2110 * If SIGKILL was already sent before the caller unlocked
2111 * ->siglock we must see ->core_state != NULL. Otherwise it
2112 * is safe to enter schedule().
2113 *
2114 * This is almost outdated, a task with the pending SIGKILL can't
2115 * block in TASK_TRACED. But PTRACE_EVENT_EXIT can be reported
2116 * after SIGKILL was already dequeued.
2117 */
2118 if (unlikely(current->mm->core_state) &&
2119 unlikely(current->mm == current->parent->mm))
2120 return false;
2121
2122 return true;
2123 }
2124
2125 /*
2126 * Return non-zero if there is a SIGKILL that should be waking us up.
2127 * Called with the siglock held.
2128 */
2129 static bool sigkill_pending(struct task_struct *tsk)
2130 {
2131 return sigismember(&tsk->pending.signal, SIGKILL) ||
2132 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
2133 }
2134
2135 /*
2136 * This must be called with current->sighand->siglock held.
2137 *
2138 * This should be the path for all ptrace stops.
2139 * We always set current->last_siginfo while stopped here.
2140 * That makes it a way to test a stopped process for
2141 * being ptrace-stopped vs being job-control-stopped.
2142 *
2143 * If we actually decide not to stop at all because the tracer
2144 * is gone, we keep current->exit_code unless clear_code.
2145 */
2146 static void ptrace_stop(int exit_code, int why, int clear_code, kernel_siginfo_t *info)
2147 __releases(&current->sighand->siglock)
2148 __acquires(&current->sighand->siglock)
2149 {
2150 bool gstop_done = false;
2151
2152 if (arch_ptrace_stop_needed(exit_code, info)) {
2153 /*
2154 * The arch code has something special to do before a
2155 * ptrace stop. This is allowed to block, e.g. for faults
2156 * on user stack pages. We can't keep the siglock while
2157 * calling arch_ptrace_stop, so we must release it now.
2158 * To preserve proper semantics, we must do this before
2159 * any signal bookkeeping like checking group_stop_count.
2160 * Meanwhile, a SIGKILL could come in before we retake the
2161 * siglock. That must prevent us from sleeping in TASK_TRACED.
2162 * So after regaining the lock, we must check for SIGKILL.
2163 */
2164 spin_unlock_irq(&current->sighand->siglock);
2165 arch_ptrace_stop(exit_code, info);
2166 spin_lock_irq(&current->sighand->siglock);
2167 if (sigkill_pending(current))
2168 return;
2169 }
2170
2171 set_special_state(TASK_TRACED);
2172
2173 /*
2174 * We're committing to trapping. TRACED should be visible before
2175 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2176 * Also, transition to TRACED and updates to ->jobctl should be
2177 * atomic with respect to siglock and should be done after the arch
2178 * hook as siglock is released and regrabbed across it.
2179 *
2180 * TRACER TRACEE
2181 *
2182 * ptrace_attach()
2183 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2184 * do_wait()
2185 * set_current_state() smp_wmb();
2186 * ptrace_do_wait()
2187 * wait_task_stopped()
2188 * task_stopped_code()
2189 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2190 */
2191 smp_wmb();
2192
2193 current->last_siginfo = info;
2194 current->exit_code = exit_code;
2195
2196 /*
2197 * If @why is CLD_STOPPED, we're trapping to participate in a group
2198 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2199 * across siglock relocks since INTERRUPT was scheduled, PENDING
2200 * could be clear now. We act as if SIGCONT is received after
2201 * TASK_TRACED is entered - ignore it.
2202 */
2203 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2204 gstop_done = task_participate_group_stop(current);
2205
2206 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2207 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2208 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2209 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2210
2211 /* entering a trap, clear TRAPPING */
2212 task_clear_jobctl_trapping(current);
2213
2214 spin_unlock_irq(&current->sighand->siglock);
2215 read_lock(&tasklist_lock);
2216 if (may_ptrace_stop()) {
2217 /*
2218 * Notify parents of the stop.
2219 *
2220 * While ptraced, there are two parents - the ptracer and
2221 * the real_parent of the group_leader. The ptracer should
2222 * know about every stop while the real parent is only
2223 * interested in the completion of group stop. The states
2224 * for the two don't interact with each other. Notify
2225 * separately unless they're gonna be duplicates.
2226 */
2227 do_notify_parent_cldstop(current, true, why);
2228 if (gstop_done && ptrace_reparented(current))
2229 do_notify_parent_cldstop(current, false, why);
2230
2231 /*
2232 * Don't want to allow preemption here, because
2233 * sys_ptrace() needs this task to be inactive.
2234 *
2235 * XXX: implement read_unlock_no_resched().
2236 */
2237 preempt_disable();
2238 read_unlock(&tasklist_lock);
2239 cgroup_enter_frozen();
2240 preempt_enable_no_resched();
2241 freezable_schedule();
2242 cgroup_leave_frozen(true);
2243 } else {
2244 /*
2245 * By the time we got the lock, our tracer went away.
2246 * Don't drop the lock yet, another tracer may come.
2247 *
2248 * If @gstop_done, the ptracer went away between group stop
2249 * completion and here. During detach, it would have set
2250 * JOBCTL_STOP_PENDING on us and we'll re-enter
2251 * TASK_STOPPED in do_signal_stop() on return, so notifying
2252 * the real parent of the group stop completion is enough.
2253 */
2254 if (gstop_done)
2255 do_notify_parent_cldstop(current, false, why);
2256
2257 /* tasklist protects us from ptrace_freeze_traced() */
2258 __set_current_state(TASK_RUNNING);
2259 if (clear_code)
2260 current->exit_code = 0;
2261 read_unlock(&tasklist_lock);
2262 }
2263
2264 /*
2265 * We are back. Now reacquire the siglock before touching
2266 * last_siginfo, so that we are sure to have synchronized with
2267 * any signal-sending on another CPU that wants to examine it.
2268 */
2269 spin_lock_irq(&current->sighand->siglock);
2270 current->last_siginfo = NULL;
2271
2272 /* LISTENING can be set only during STOP traps, clear it */
2273 current->jobctl &= ~JOBCTL_LISTENING;
2274
2275 /*
2276 * Queued signals ignored us while we were stopped for tracing.
2277 * So check for any that we should take before resuming user mode.
2278 * This sets TIF_SIGPENDING, but never clears it.
2279 */
2280 recalc_sigpending_tsk(current);
2281 }
2282
2283 static void ptrace_do_notify(int signr, int exit_code, int why)
2284 {
2285 kernel_siginfo_t info;
2286
2287 clear_siginfo(&info);
2288 info.si_signo = signr;
2289 info.si_code = exit_code;
2290 info.si_pid = task_pid_vnr(current);
2291 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2292
2293 /* Let the debugger run. */
2294 ptrace_stop(exit_code, why, 1, &info);
2295 }
2296
2297 void ptrace_notify(int exit_code)
2298 {
2299 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2300 if (unlikely(current->task_works))
2301 task_work_run();
2302
2303 spin_lock_irq(&current->sighand->siglock);
2304 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
2305 spin_unlock_irq(&current->sighand->siglock);
2306 }
2307
2308 /**
2309 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2310 * @signr: signr causing group stop if initiating
2311 *
2312 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2313 * and participate in it. If already set, participate in the existing
2314 * group stop. If participated in a group stop (and thus slept), %true is
2315 * returned with siglock released.
2316 *
2317 * If ptraced, this function doesn't handle stop itself. Instead,
2318 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2319 * untouched. The caller must ensure that INTERRUPT trap handling takes
2320 * places afterwards.
2321 *
2322 * CONTEXT:
2323 * Must be called with @current->sighand->siglock held, which is released
2324 * on %true return.
2325 *
2326 * RETURNS:
2327 * %false if group stop is already cancelled or ptrace trap is scheduled.
2328 * %true if participated in group stop.
2329 */
2330 static bool do_signal_stop(int signr)
2331 __releases(&current->sighand->siglock)
2332 {
2333 struct signal_struct *sig = current->signal;
2334
2335 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2336 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2337 struct task_struct *t;
2338
2339 /* signr will be recorded in task->jobctl for retries */
2340 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2341
2342 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2343 unlikely(signal_group_exit(sig)))
2344 return false;
2345 /*
2346 * There is no group stop already in progress. We must
2347 * initiate one now.
2348 *
2349 * While ptraced, a task may be resumed while group stop is
2350 * still in effect and then receive a stop signal and
2351 * initiate another group stop. This deviates from the
2352 * usual behavior as two consecutive stop signals can't
2353 * cause two group stops when !ptraced. That is why we
2354 * also check !task_is_stopped(t) below.
2355 *
2356 * The condition can be distinguished by testing whether
2357 * SIGNAL_STOP_STOPPED is already set. Don't generate
2358 * group_exit_code in such case.
2359 *
2360 * This is not necessary for SIGNAL_STOP_CONTINUED because
2361 * an intervening stop signal is required to cause two
2362 * continued events regardless of ptrace.
2363 */
2364 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2365 sig->group_exit_code = signr;
2366
2367 sig->group_stop_count = 0;
2368
2369 if (task_set_jobctl_pending(current, signr | gstop))
2370 sig->group_stop_count++;
2371
2372 t = current;
2373 while_each_thread(current, t) {
2374 /*
2375 * Setting state to TASK_STOPPED for a group
2376 * stop is always done with the siglock held,
2377 * so this check has no races.
2378 */
2379 if (!task_is_stopped(t) &&
2380 task_set_jobctl_pending(t, signr | gstop)) {
2381 sig->group_stop_count++;
2382 if (likely(!(t->ptrace & PT_SEIZED)))
2383 signal_wake_up(t, 0);
2384 else
2385 ptrace_trap_notify(t);
2386 }
2387 }
2388 }
2389
2390 if (likely(!current->ptrace)) {
2391 int notify = 0;
2392
2393 /*
2394 * If there are no other threads in the group, or if there
2395 * is a group stop in progress and we are the last to stop,
2396 * report to the parent.
2397 */
2398 if (task_participate_group_stop(current))
2399 notify = CLD_STOPPED;
2400
2401 set_special_state(TASK_STOPPED);
2402 spin_unlock_irq(&current->sighand->siglock);
2403
2404 /*
2405 * Notify the parent of the group stop completion. Because
2406 * we're not holding either the siglock or tasklist_lock
2407 * here, ptracer may attach inbetween; however, this is for
2408 * group stop and should always be delivered to the real
2409 * parent of the group leader. The new ptracer will get
2410 * its notification when this task transitions into
2411 * TASK_TRACED.
2412 */
2413 if (notify) {
2414 read_lock(&tasklist_lock);
2415 do_notify_parent_cldstop(current, false, notify);
2416 read_unlock(&tasklist_lock);
2417 }
2418
2419 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2420 cgroup_enter_frozen();
2421 freezable_schedule();
2422 return true;
2423 } else {
2424 /*
2425 * While ptraced, group stop is handled by STOP trap.
2426 * Schedule it and let the caller deal with it.
2427 */
2428 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2429 return false;
2430 }
2431 }
2432
2433 /**
2434 * do_jobctl_trap - take care of ptrace jobctl traps
2435 *
2436 * When PT_SEIZED, it's used for both group stop and explicit
2437 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2438 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2439 * the stop signal; otherwise, %SIGTRAP.
2440 *
2441 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2442 * number as exit_code and no siginfo.
2443 *
2444 * CONTEXT:
2445 * Must be called with @current->sighand->siglock held, which may be
2446 * released and re-acquired before returning with intervening sleep.
2447 */
2448 static void do_jobctl_trap(void)
2449 {
2450 struct signal_struct *signal = current->signal;
2451 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2452
2453 if (current->ptrace & PT_SEIZED) {
2454 if (!signal->group_stop_count &&
2455 !(signal->flags & SIGNAL_STOP_STOPPED))
2456 signr = SIGTRAP;
2457 WARN_ON_ONCE(!signr);
2458 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2459 CLD_STOPPED);
2460 } else {
2461 WARN_ON_ONCE(!signr);
2462 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2463 current->exit_code = 0;
2464 }
2465 }
2466
2467 /**
2468 * do_freezer_trap - handle the freezer jobctl trap
2469 *
2470 * Puts the task into frozen state, if only the task is not about to quit.
2471 * In this case it drops JOBCTL_TRAP_FREEZE.
2472 *
2473 * CONTEXT:
2474 * Must be called with @current->sighand->siglock held,
2475 * which is always released before returning.
2476 */
2477 static void do_freezer_trap(void)
2478 __releases(&current->sighand->siglock)
2479 {
2480 /*
2481 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2482 * let's make another loop to give it a chance to be handled.
2483 * In any case, we'll return back.
2484 */
2485 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2486 JOBCTL_TRAP_FREEZE) {
2487 spin_unlock_irq(&current->sighand->siglock);
2488 return;
2489 }
2490
2491 /*
2492 * Now we're sure that there is no pending fatal signal and no
2493 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2494 * immediately (if there is a non-fatal signal pending), and
2495 * put the task into sleep.
2496 */
2497 __set_current_state(TASK_INTERRUPTIBLE);
2498 clear_thread_flag(TIF_SIGPENDING);
2499 spin_unlock_irq(&current->sighand->siglock);
2500 cgroup_enter_frozen();
2501 freezable_schedule();
2502 }
2503
2504 static int ptrace_signal(int signr, kernel_siginfo_t *info)
2505 {
2506 /*
2507 * We do not check sig_kernel_stop(signr) but set this marker
2508 * unconditionally because we do not know whether debugger will
2509 * change signr. This flag has no meaning unless we are going
2510 * to stop after return from ptrace_stop(). In this case it will
2511 * be checked in do_signal_stop(), we should only stop if it was
2512 * not cleared by SIGCONT while we were sleeping. See also the
2513 * comment in dequeue_signal().
2514 */
2515 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2516 ptrace_stop(signr, CLD_TRAPPED, 0, info);
2517
2518 /* We're back. Did the debugger cancel the sig? */
2519 signr = current->exit_code;
2520 if (signr == 0)
2521 return signr;
2522
2523 current->exit_code = 0;
2524
2525 /*
2526 * Update the siginfo structure if the signal has
2527 * changed. If the debugger wanted something
2528 * specific in the siginfo structure then it should
2529 * have updated *info via PTRACE_SETSIGINFO.
2530 */
2531 if (signr != info->si_signo) {
2532 clear_siginfo(info);
2533 info->si_signo = signr;
2534 info->si_errno = 0;
2535 info->si_code = SI_USER;
2536 rcu_read_lock();
2537 info->si_pid = task_pid_vnr(current->parent);
2538 info->si_uid = from_kuid_munged(current_user_ns(),
2539 task_uid(current->parent));
2540 rcu_read_unlock();
2541 }
2542
2543 /* If the (new) signal is now blocked, requeue it. */
2544 if (sigismember(&current->blocked, signr)) {
2545 send_signal(signr, info, current, PIDTYPE_PID);
2546 signr = 0;
2547 }
2548
2549 return signr;
2550 }
2551
2552 static void hide_si_addr_tag_bits(struct ksignal *ksig)
2553 {
2554 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2555 case SIL_FAULT:
2556 case SIL_FAULT_TRAPNO:
2557 case SIL_FAULT_MCEERR:
2558 case SIL_FAULT_BNDERR:
2559 case SIL_FAULT_PKUERR:
2560 case SIL_PERF_EVENT:
2561 ksig->info.si_addr = arch_untagged_si_addr(
2562 ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2563 break;
2564 case SIL_KILL:
2565 case SIL_TIMER:
2566 case SIL_POLL:
2567 case SIL_CHLD:
2568 case SIL_RT:
2569 case SIL_SYS:
2570 break;
2571 }
2572 }
2573
2574 bool get_signal(struct ksignal *ksig)
2575 {
2576 struct sighand_struct *sighand = current->sighand;
2577 struct signal_struct *signal = current->signal;
2578 int signr;
2579
2580 if (unlikely(current->task_works))
2581 task_work_run();
2582
2583 /*
2584 * For non-generic architectures, check for TIF_NOTIFY_SIGNAL so
2585 * that the arch handlers don't all have to do it. If we get here
2586 * without TIF_SIGPENDING, just exit after running signal work.
2587 */
2588 if (!IS_ENABLED(CONFIG_GENERIC_ENTRY)) {
2589 if (test_thread_flag(TIF_NOTIFY_SIGNAL))
2590 tracehook_notify_signal();
2591 if (!task_sigpending(current))
2592 return false;
2593 }
2594
2595 if (unlikely(uprobe_deny_signal()))
2596 return false;
2597
2598 /*
2599 * Do this once, we can't return to user-mode if freezing() == T.
2600 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2601 * thus do not need another check after return.
2602 */
2603 try_to_freeze();
2604
2605 relock:
2606 spin_lock_irq(&sighand->siglock);
2607
2608 /*
2609 * Every stopped thread goes here after wakeup. Check to see if
2610 * we should notify the parent, prepare_signal(SIGCONT) encodes
2611 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2612 */
2613 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2614 int why;
2615
2616 if (signal->flags & SIGNAL_CLD_CONTINUED)
2617 why = CLD_CONTINUED;
2618 else
2619 why = CLD_STOPPED;
2620
2621 signal->flags &= ~SIGNAL_CLD_MASK;
2622
2623 spin_unlock_irq(&sighand->siglock);
2624
2625 /*
2626 * Notify the parent that we're continuing. This event is
2627 * always per-process and doesn't make whole lot of sense
2628 * for ptracers, who shouldn't consume the state via
2629 * wait(2) either, but, for backward compatibility, notify
2630 * the ptracer of the group leader too unless it's gonna be
2631 * a duplicate.
2632 */
2633 read_lock(&tasklist_lock);
2634 do_notify_parent_cldstop(current, false, why);
2635
2636 if (ptrace_reparented(current->group_leader))
2637 do_notify_parent_cldstop(current->group_leader,
2638 true, why);
2639 read_unlock(&tasklist_lock);
2640
2641 goto relock;
2642 }
2643
2644 /* Has this task already been marked for death? */
2645 if (signal_group_exit(signal)) {
2646 ksig->info.si_signo = signr = SIGKILL;
2647 sigdelset(&current->pending.signal, SIGKILL);
2648 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2649 &sighand->action[SIGKILL - 1]);
2650 recalc_sigpending();
2651 goto fatal;
2652 }
2653
2654 for (;;) {
2655 struct k_sigaction *ka;
2656
2657 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2658 do_signal_stop(0))
2659 goto relock;
2660
2661 if (unlikely(current->jobctl &
2662 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2663 if (current->jobctl & JOBCTL_TRAP_MASK) {
2664 do_jobctl_trap();
2665 spin_unlock_irq(&sighand->siglock);
2666 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2667 do_freezer_trap();
2668
2669 goto relock;
2670 }
2671
2672 /*
2673 * If the task is leaving the frozen state, let's update
2674 * cgroup counters and reset the frozen bit.
2675 */
2676 if (unlikely(cgroup_task_frozen(current))) {
2677 spin_unlock_irq(&sighand->siglock);
2678 cgroup_leave_frozen(false);
2679 goto relock;
2680 }
2681
2682 /*
2683 * Signals generated by the execution of an instruction
2684 * need to be delivered before any other pending signals
2685 * so that the instruction pointer in the signal stack
2686 * frame points to the faulting instruction.
2687 */
2688 signr = dequeue_synchronous_signal(&ksig->info);
2689 if (!signr)
2690 signr = dequeue_signal(current, &current->blocked, &ksig->info);
2691
2692 if (!signr)
2693 break; /* will return 0 */
2694
2695 if (unlikely(current->ptrace) && signr != SIGKILL) {
2696 signr = ptrace_signal(signr, &ksig->info);
2697 if (!signr)
2698 continue;
2699 }
2700
2701 ka = &sighand->action[signr-1];
2702
2703 /* Trace actually delivered signals. */
2704 trace_signal_deliver(signr, &ksig->info, ka);
2705
2706 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2707 continue;
2708 if (ka->sa.sa_handler != SIG_DFL) {
2709 /* Run the handler. */
2710 ksig->ka = *ka;
2711
2712 if (ka->sa.sa_flags & SA_ONESHOT)
2713 ka->sa.sa_handler = SIG_DFL;
2714
2715 break; /* will return non-zero "signr" value */
2716 }
2717
2718 /*
2719 * Now we are doing the default action for this signal.
2720 */
2721 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2722 continue;
2723
2724 /*
2725 * Global init gets no signals it doesn't want.
2726 * Container-init gets no signals it doesn't want from same
2727 * container.
2728 *
2729 * Note that if global/container-init sees a sig_kernel_only()
2730 * signal here, the signal must have been generated internally
2731 * or must have come from an ancestor namespace. In either
2732 * case, the signal cannot be dropped.
2733 */
2734 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2735 !sig_kernel_only(signr))
2736 continue;
2737
2738 if (sig_kernel_stop(signr)) {
2739 /*
2740 * The default action is to stop all threads in
2741 * the thread group. The job control signals
2742 * do nothing in an orphaned pgrp, but SIGSTOP
2743 * always works. Note that siglock needs to be
2744 * dropped during the call to is_orphaned_pgrp()
2745 * because of lock ordering with tasklist_lock.
2746 * This allows an intervening SIGCONT to be posted.
2747 * We need to check for that and bail out if necessary.
2748 */
2749 if (signr != SIGSTOP) {
2750 spin_unlock_irq(&sighand->siglock);
2751
2752 /* signals can be posted during this window */
2753
2754 if (is_current_pgrp_orphaned())
2755 goto relock;
2756
2757 spin_lock_irq(&sighand->siglock);
2758 }
2759
2760 if (likely(do_signal_stop(ksig->info.si_signo))) {
2761 /* It released the siglock. */
2762 goto relock;
2763 }
2764
2765 /*
2766 * We didn't actually stop, due to a race
2767 * with SIGCONT or something like that.
2768 */
2769 continue;
2770 }
2771
2772 fatal:
2773 spin_unlock_irq(&sighand->siglock);
2774 if (unlikely(cgroup_task_frozen(current)))
2775 cgroup_leave_frozen(true);
2776
2777 /*
2778 * Anything else is fatal, maybe with a core dump.
2779 */
2780 current->flags |= PF_SIGNALED;
2781
2782 if (sig_kernel_coredump(signr)) {
2783 if (print_fatal_signals)
2784 print_fatal_signal(ksig->info.si_signo);
2785 proc_coredump_connector(current);
2786 /*
2787 * If it was able to dump core, this kills all
2788 * other threads in the group and synchronizes with
2789 * their demise. If we lost the race with another
2790 * thread getting here, it set group_exit_code
2791 * first and our do_group_exit call below will use
2792 * that value and ignore the one we pass it.
2793 */
2794 do_coredump(&ksig->info);
2795 }
2796
2797 /*
2798 * PF_IO_WORKER threads will catch and exit on fatal signals
2799 * themselves. They have cleanup that must be performed, so
2800 * we cannot call do_exit() on their behalf.
2801 */
2802 if (current->flags & PF_IO_WORKER)
2803 goto out;
2804
2805 /*
2806 * Death signals, no core dump.
2807 */
2808 do_group_exit(ksig->info.si_signo);
2809 /* NOTREACHED */
2810 }
2811 spin_unlock_irq(&sighand->siglock);
2812 out:
2813 ksig->sig = signr;
2814
2815 if (!(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
2816 hide_si_addr_tag_bits(ksig);
2817
2818 return ksig->sig > 0;
2819 }
2820
2821 /**
2822 * signal_delivered -
2823 * @ksig: kernel signal struct
2824 * @stepping: nonzero if debugger single-step or block-step in use
2825 *
2826 * This function should be called when a signal has successfully been
2827 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2828 * is always blocked, and the signal itself is blocked unless %SA_NODEFER
2829 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2830 */
2831 static void signal_delivered(struct ksignal *ksig, int stepping)
2832 {
2833 sigset_t blocked;
2834
2835 /* A signal was successfully delivered, and the
2836 saved sigmask was stored on the signal frame,
2837 and will be restored by sigreturn. So we can
2838 simply clear the restore sigmask flag. */
2839 clear_restore_sigmask();
2840
2841 sigorsets(&blocked, &current->blocked, &ksig->ka.sa.sa_mask);
2842 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2843 sigaddset(&blocked, ksig->sig);
2844 set_current_blocked(&blocked);
2845 if (current->sas_ss_flags & SS_AUTODISARM)
2846 sas_ss_reset(current);
2847 tracehook_signal_handler(stepping);
2848 }
2849
2850 void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2851 {
2852 if (failed)
2853 force_sigsegv(ksig->sig);
2854 else
2855 signal_delivered(ksig, stepping);
2856 }
2857
2858 /*
2859 * It could be that complete_signal() picked us to notify about the
2860 * group-wide signal. Other threads should be notified now to take
2861 * the shared signals in @which since we will not.
2862 */
2863 static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2864 {
2865 sigset_t retarget;
2866 struct task_struct *t;
2867
2868 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2869 if (sigisemptyset(&retarget))
2870 return;
2871
2872 t = tsk;
2873 while_each_thread(tsk, t) {
2874 if (t->flags & PF_EXITING)
2875 continue;
2876
2877 if (!has_pending_signals(&retarget, &t->blocked))
2878 continue;
2879 /* Remove the signals this thread can handle. */
2880 sigandsets(&retarget, &retarget, &t->blocked);
2881
2882 if (!task_sigpending(t))
2883 signal_wake_up(t, 0);
2884
2885 if (sigisemptyset(&retarget))
2886 break;
2887 }
2888 }
2889
2890 void exit_signals(struct task_struct *tsk)
2891 {
2892 int group_stop = 0;
2893 sigset_t unblocked;
2894
2895 /*
2896 * @tsk is about to have PF_EXITING set - lock out users which
2897 * expect stable threadgroup.
2898 */
2899 cgroup_threadgroup_change_begin(tsk);
2900
2901 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2902 tsk->flags |= PF_EXITING;
2903 cgroup_threadgroup_change_end(tsk);
2904 return;
2905 }
2906
2907 spin_lock_irq(&tsk->sighand->siglock);
2908 /*
2909 * From now this task is not visible for group-wide signals,
2910 * see wants_signal(), do_signal_stop().
2911 */
2912 tsk->flags |= PF_EXITING;
2913
2914 cgroup_threadgroup_change_end(tsk);
2915
2916 if (!task_sigpending(tsk))
2917 goto out;
2918
2919 unblocked = tsk->blocked;
2920 signotset(&unblocked);
2921 retarget_shared_pending(tsk, &unblocked);
2922
2923 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
2924 task_participate_group_stop(tsk))
2925 group_stop = CLD_STOPPED;
2926 out:
2927 spin_unlock_irq(&tsk->sighand->siglock);
2928
2929 /*
2930 * If group stop has completed, deliver the notification. This
2931 * should always go to the real parent of the group leader.
2932 */
2933 if (unlikely(group_stop)) {
2934 read_lock(&tasklist_lock);
2935 do_notify_parent_cldstop(tsk, false, group_stop);
2936 read_unlock(&tasklist_lock);
2937 }
2938 }
2939
2940 /*
2941 * System call entry points.
2942 */
2943
2944 /**
2945 * sys_restart_syscall - restart a system call
2946 */
2947 SYSCALL_DEFINE0(restart_syscall)
2948 {
2949 struct restart_block *restart = &current->restart_block;
2950 return restart->fn(restart);
2951 }
2952
2953 long do_no_restart_syscall(struct restart_block *param)
2954 {
2955 return -EINTR;
2956 }
2957
2958 static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
2959 {
2960 if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
2961 sigset_t newblocked;
2962 /* A set of now blocked but previously unblocked signals. */
2963 sigandnsets(&newblocked, newset, &current->blocked);
2964 retarget_shared_pending(tsk, &newblocked);
2965 }
2966 tsk->blocked = *newset;
2967 recalc_sigpending();
2968 }
2969
2970 /**
2971 * set_current_blocked - change current->blocked mask
2972 * @newset: new mask
2973 *
2974 * It is wrong to change ->blocked directly, this helper should be used
2975 * to ensure the process can't miss a shared signal we are going to block.
2976 */
2977 void set_current_blocked(sigset_t *newset)
2978 {
2979 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
2980 __set_current_blocked(newset);
2981 }
2982
2983 void __set_current_blocked(const sigset_t *newset)
2984 {
2985 struct task_struct *tsk = current;
2986
2987 /*
2988 * In case the signal mask hasn't changed, there is nothing we need
2989 * to do. The current->blocked shouldn't be modified by other task.
2990 */
2991 if (sigequalsets(&tsk->blocked, newset))
2992 return;
2993
2994 spin_lock_irq(&tsk->sighand->siglock);
2995 __set_task_blocked(tsk, newset);
2996 spin_unlock_irq(&tsk->sighand->siglock);
2997 }
2998
2999 /*
3000 * This is also useful for kernel threads that want to temporarily
3001 * (or permanently) block certain signals.
3002 *
3003 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
3004 * interface happily blocks "unblockable" signals like SIGKILL
3005 * and friends.
3006 */
3007 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
3008 {
3009 struct task_struct *tsk = current;
3010 sigset_t newset;
3011
3012 /* Lockless, only current can change ->blocked, never from irq */
3013 if (oldset)
3014 *oldset = tsk->blocked;
3015
3016 switch (how) {
3017 case SIG_BLOCK:
3018 sigorsets(&newset, &tsk->blocked, set);
3019 break;
3020 case SIG_UNBLOCK:
3021 sigandnsets(&newset, &tsk->blocked, set);
3022 break;
3023 case SIG_SETMASK:
3024 newset = *set;
3025 break;
3026 default:
3027 return -EINVAL;
3028 }
3029
3030 __set_current_blocked(&newset);
3031 return 0;
3032 }
3033 EXPORT_SYMBOL(sigprocmask);
3034
3035 /*
3036 * The api helps set app-provided sigmasks.
3037 *
3038 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3039 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3040 *
3041 * Note that it does set_restore_sigmask() in advance, so it must be always
3042 * paired with restore_saved_sigmask_unless() before return from syscall.
3043 */
3044 int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3045 {
3046 sigset_t kmask;
3047
3048 if (!umask)
3049 return 0;
3050 if (sigsetsize != sizeof(sigset_t))
3051 return -EINVAL;
3052 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3053 return -EFAULT;
3054
3055 set_restore_sigmask();
3056 current->saved_sigmask = current->blocked;
3057 set_current_blocked(&kmask);
3058
3059 return 0;
3060 }
3061
3062 #ifdef CONFIG_COMPAT
3063 int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3064 size_t sigsetsize)
3065 {
3066 sigset_t kmask;
3067
3068 if (!umask)
3069 return 0;
3070 if (sigsetsize != sizeof(compat_sigset_t))
3071 return -EINVAL;
3072 if (get_compat_sigset(&kmask, umask))
3073 return -EFAULT;
3074
3075 set_restore_sigmask();
3076 current->saved_sigmask = current->blocked;
3077 set_current_blocked(&kmask);
3078
3079 return 0;
3080 }
3081 #endif
3082
3083 /**
3084 * sys_rt_sigprocmask - change the list of currently blocked signals
3085 * @how: whether to add, remove, or set signals
3086 * @nset: stores pending signals
3087 * @oset: previous value of signal mask if non-null
3088 * @sigsetsize: size of sigset_t type
3089 */
3090 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3091 sigset_t __user *, oset, size_t, sigsetsize)
3092 {
3093 sigset_t old_set, new_set;
3094 int error;
3095
3096 /* XXX: Don't preclude handling different sized sigset_t's. */
3097 if (sigsetsize != sizeof(sigset_t))
3098 return -EINVAL;
3099
3100 old_set = current->blocked;
3101
3102 if (nset) {
3103 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3104 return -EFAULT;
3105 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3106
3107 error = sigprocmask(how, &new_set, NULL);
3108 if (error)
3109 return error;
3110 }
3111
3112 if (oset) {
3113 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3114 return -EFAULT;
3115 }
3116
3117 return 0;
3118 }
3119
3120 #ifdef CONFIG_COMPAT
3121 COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3122 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3123 {
3124 sigset_t old_set = current->blocked;
3125
3126 /* XXX: Don't preclude handling different sized sigset_t's. */
3127 if (sigsetsize != sizeof(sigset_t))
3128 return -EINVAL;
3129
3130 if (nset) {
3131 sigset_t new_set;
3132 int error;
3133 if (get_compat_sigset(&new_set, nset))
3134 return -EFAULT;
3135 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3136
3137 error = sigprocmask(how, &new_set, NULL);
3138 if (error)
3139 return error;
3140 }
3141 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3142 }
3143 #endif
3144
3145 static void do_sigpending(sigset_t *set)
3146 {
3147 spin_lock_irq(&current->sighand->siglock);
3148 sigorsets(set, &current->pending.signal,
3149 &current->signal->shared_pending.signal);
3150 spin_unlock_irq(&current->sighand->siglock);
3151
3152 /* Outside the lock because only this thread touches it. */
3153 sigandsets(set, &current->blocked, set);
3154 }
3155
3156 /**
3157 * sys_rt_sigpending - examine a pending signal that has been raised
3158 * while blocked
3159 * @uset: stores pending signals
3160 * @sigsetsize: size of sigset_t type or larger
3161 */
3162 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3163 {
3164 sigset_t set;
3165
3166 if (sigsetsize > sizeof(*uset))
3167 return -EINVAL;
3168
3169 do_sigpending(&set);
3170
3171 if (copy_to_user(uset, &set, sigsetsize))
3172 return -EFAULT;
3173
3174 return 0;
3175 }
3176
3177 #ifdef CONFIG_COMPAT
3178 COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3179 compat_size_t, sigsetsize)
3180 {
3181 sigset_t set;
3182
3183 if (sigsetsize > sizeof(*uset))
3184 return -EINVAL;
3185
3186 do_sigpending(&set);
3187
3188 return put_compat_sigset(uset, &set, sigsetsize);
3189 }
3190 #endif
3191
3192 static const struct {
3193 unsigned char limit, layout;
3194 } sig_sicodes[] = {
3195 [SIGILL] = { NSIGILL, SIL_FAULT },
3196 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3197 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3198 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3199 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3200 #if defined(SIGEMT)
3201 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3202 #endif
3203 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3204 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3205 [SIGSYS] = { NSIGSYS, SIL_SYS },
3206 };
3207
3208 static bool known_siginfo_layout(unsigned sig, int si_code)
3209 {
3210 if (si_code == SI_KERNEL)
3211 return true;
3212 else if ((si_code > SI_USER)) {
3213 if (sig_specific_sicodes(sig)) {
3214 if (si_code <= sig_sicodes[sig].limit)
3215 return true;
3216 }
3217 else if (si_code <= NSIGPOLL)
3218 return true;
3219 }
3220 else if (si_code >= SI_DETHREAD)
3221 return true;
3222 else if (si_code == SI_ASYNCNL)
3223 return true;
3224 return false;
3225 }
3226
3227 enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3228 {
3229 enum siginfo_layout layout = SIL_KILL;
3230 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3231 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3232 (si_code <= sig_sicodes[sig].limit)) {
3233 layout = sig_sicodes[sig].layout;
3234 /* Handle the exceptions */
3235 if ((sig == SIGBUS) &&
3236 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3237 layout = SIL_FAULT_MCEERR;
3238 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3239 layout = SIL_FAULT_BNDERR;
3240 #ifdef SEGV_PKUERR
3241 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3242 layout = SIL_FAULT_PKUERR;
3243 #endif
3244 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3245 layout = SIL_PERF_EVENT;
3246 #ifdef __ARCH_SI_TRAPNO
3247 else if (layout == SIL_FAULT)
3248 layout = SIL_FAULT_TRAPNO;
3249 #endif
3250 }
3251 else if (si_code <= NSIGPOLL)
3252 layout = SIL_POLL;
3253 } else {
3254 if (si_code == SI_TIMER)
3255 layout = SIL_TIMER;
3256 else if (si_code == SI_SIGIO)
3257 layout = SIL_POLL;
3258 else if (si_code < 0)
3259 layout = SIL_RT;
3260 }
3261 return layout;
3262 }
3263
3264 static inline char __user *si_expansion(const siginfo_t __user *info)
3265 {
3266 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3267 }
3268
3269 int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3270 {
3271 char __user *expansion = si_expansion(to);
3272 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3273 return -EFAULT;
3274 if (clear_user(expansion, SI_EXPANSION_SIZE))
3275 return -EFAULT;
3276 return 0;
3277 }
3278
3279 static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3280 const siginfo_t __user *from)
3281 {
3282 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3283 char __user *expansion = si_expansion(from);
3284 char buf[SI_EXPANSION_SIZE];
3285 int i;
3286 /*
3287 * An unknown si_code might need more than
3288 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3289 * extra bytes are 0. This guarantees copy_siginfo_to_user
3290 * will return this data to userspace exactly.
3291 */
3292 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3293 return -EFAULT;
3294 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3295 if (buf[i] != 0)
3296 return -E2BIG;
3297 }
3298 }
3299 return 0;
3300 }
3301
3302 static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3303 const siginfo_t __user *from)
3304 {
3305 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3306 return -EFAULT;
3307 to->si_signo = signo;
3308 return post_copy_siginfo_from_user(to, from);
3309 }
3310
3311 int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3312 {
3313 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3314 return -EFAULT;
3315 return post_copy_siginfo_from_user(to, from);
3316 }
3317
3318 #ifdef CONFIG_COMPAT
3319 /**
3320 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3321 * @to: compat siginfo destination
3322 * @from: kernel siginfo source
3323 *
3324 * Note: This function does not work properly for the SIGCHLD on x32, but
3325 * fortunately it doesn't have to. The only valid callers for this function are
3326 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3327 * The latter does not care because SIGCHLD will never cause a coredump.
3328 */
3329 void copy_siginfo_to_external32(struct compat_siginfo *to,
3330 const struct kernel_siginfo *from)
3331 {
3332 memset(to, 0, sizeof(*to));
3333
3334 to->si_signo = from->si_signo;
3335 to->si_errno = from->si_errno;
3336 to->si_code = from->si_code;
3337 switch(siginfo_layout(from->si_signo, from->si_code)) {
3338 case SIL_KILL:
3339 to->si_pid = from->si_pid;
3340 to->si_uid = from->si_uid;
3341 break;
3342 case SIL_TIMER:
3343 to->si_tid = from->si_tid;
3344 to->si_overrun = from->si_overrun;
3345 to->si_int = from->si_int;
3346 break;
3347 case SIL_POLL:
3348 to->si_band = from->si_band;
3349 to->si_fd = from->si_fd;
3350 break;
3351 case SIL_FAULT:
3352 to->si_addr = ptr_to_compat(from->si_addr);
3353 break;
3354 case SIL_FAULT_TRAPNO:
3355 to->si_addr = ptr_to_compat(from->si_addr);
3356 to->si_trapno = from->si_trapno;
3357 break;
3358 case SIL_FAULT_MCEERR:
3359 to->si_addr = ptr_to_compat(from->si_addr);
3360 to->si_addr_lsb = from->si_addr_lsb;
3361 break;
3362 case SIL_FAULT_BNDERR:
3363 to->si_addr = ptr_to_compat(from->si_addr);
3364 to->si_lower = ptr_to_compat(from->si_lower);
3365 to->si_upper = ptr_to_compat(from->si_upper);
3366 break;
3367 case SIL_FAULT_PKUERR:
3368 to->si_addr = ptr_to_compat(from->si_addr);
3369 to->si_pkey = from->si_pkey;
3370 break;
3371 case SIL_PERF_EVENT:
3372 to->si_addr = ptr_to_compat(from->si_addr);
3373 to->si_perf_data = from->si_perf_data;
3374 to->si_perf_type = from->si_perf_type;
3375 break;
3376 case SIL_CHLD:
3377 to->si_pid = from->si_pid;
3378 to->si_uid = from->si_uid;
3379 to->si_status = from->si_status;
3380 to->si_utime = from->si_utime;
3381 to->si_stime = from->si_stime;
3382 break;
3383 case SIL_RT:
3384 to->si_pid = from->si_pid;
3385 to->si_uid = from->si_uid;
3386 to->si_int = from->si_int;
3387 break;
3388 case SIL_SYS:
3389 to->si_call_addr = ptr_to_compat(from->si_call_addr);
3390 to->si_syscall = from->si_syscall;
3391 to->si_arch = from->si_arch;
3392 break;
3393 }
3394 }
3395
3396 int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3397 const struct kernel_siginfo *from)
3398 {
3399 struct compat_siginfo new;
3400
3401 copy_siginfo_to_external32(&new, from);
3402 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3403 return -EFAULT;
3404 return 0;
3405 }
3406
3407 static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3408 const struct compat_siginfo *from)
3409 {
3410 clear_siginfo(to);
3411 to->si_signo = from->si_signo;
3412 to->si_errno = from->si_errno;
3413 to->si_code = from->si_code;
3414 switch(siginfo_layout(from->si_signo, from->si_code)) {
3415 case SIL_KILL:
3416 to->si_pid = from->si_pid;
3417 to->si_uid = from->si_uid;
3418 break;
3419 case SIL_TIMER:
3420 to->si_tid = from->si_tid;
3421 to->si_overrun = from->si_overrun;
3422 to->si_int = from->si_int;
3423 break;
3424 case SIL_POLL:
3425 to->si_band = from->si_band;
3426 to->si_fd = from->si_fd;
3427 break;
3428 case SIL_FAULT:
3429 to->si_addr = compat_ptr(from->si_addr);
3430 break;
3431 case SIL_FAULT_TRAPNO:
3432 to->si_addr = compat_ptr(from->si_addr);
3433 to->si_trapno = from->si_trapno;
3434 break;
3435 case SIL_FAULT_MCEERR:
3436 to->si_addr = compat_ptr(from->si_addr);
3437 to->si_addr_lsb = from->si_addr_lsb;
3438 break;
3439 case SIL_FAULT_BNDERR:
3440 to->si_addr = compat_ptr(from->si_addr);
3441 to->si_lower = compat_ptr(from->si_lower);
3442 to->si_upper = compat_ptr(from->si_upper);
3443 break;
3444 case SIL_FAULT_PKUERR:
3445 to->si_addr = compat_ptr(from->si_addr);
3446 to->si_pkey = from->si_pkey;
3447 break;
3448 case SIL_PERF_EVENT:
3449 to->si_addr = compat_ptr(from->si_addr);
3450 to->si_perf_data = from->si_perf_data;
3451 to->si_perf_type = from->si_perf_type;
3452 break;
3453 case SIL_CHLD:
3454 to->si_pid = from->si_pid;
3455 to->si_uid = from->si_uid;
3456 to->si_status = from->si_status;
3457 #ifdef CONFIG_X86_X32_ABI
3458 if (in_x32_syscall()) {
3459 to->si_utime = from->_sifields._sigchld_x32._utime;
3460 to->si_stime = from->_sifields._sigchld_x32._stime;
3461 } else
3462 #endif
3463 {
3464 to->si_utime = from->si_utime;
3465 to->si_stime = from->si_stime;
3466 }
3467 break;
3468 case SIL_RT:
3469 to->si_pid = from->si_pid;
3470 to->si_uid = from->si_uid;
3471 to->si_int = from->si_int;
3472 break;
3473 case SIL_SYS:
3474 to->si_call_addr = compat_ptr(from->si_call_addr);
3475 to->si_syscall = from->si_syscall;
3476 to->si_arch = from->si_arch;
3477 break;
3478 }
3479 return 0;
3480 }
3481
3482 static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3483 const struct compat_siginfo __user *ufrom)
3484 {
3485 struct compat_siginfo from;
3486
3487 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3488 return -EFAULT;
3489
3490 from.si_signo = signo;
3491 return post_copy_siginfo_from_user32(to, &from);
3492 }
3493
3494 int copy_siginfo_from_user32(struct kernel_siginfo *to,
3495 const struct compat_siginfo __user *ufrom)
3496 {
3497 struct compat_siginfo from;
3498
3499 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3500 return -EFAULT;
3501
3502 return post_copy_siginfo_from_user32(to, &from);
3503 }
3504 #endif /* CONFIG_COMPAT */
3505
3506 /**
3507 * do_sigtimedwait - wait for queued signals specified in @which
3508 * @which: queued signals to wait for
3509 * @info: if non-null, the signal's siginfo is returned here
3510 * @ts: upper bound on process time suspension
3511 */
3512 static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3513 const struct timespec64 *ts)
3514 {
3515 ktime_t *to = NULL, timeout = KTIME_MAX;
3516 struct task_struct *tsk = current;
3517 sigset_t mask = *which;
3518 int sig, ret = 0;
3519
3520 if (ts) {
3521 if (!timespec64_valid(ts))
3522 return -EINVAL;
3523 timeout = timespec64_to_ktime(*ts);
3524 to = &timeout;
3525 }
3526
3527 /*
3528 * Invert the set of allowed signals to get those we want to block.
3529 */
3530 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3531 signotset(&mask);
3532
3533 spin_lock_irq(&tsk->sighand->siglock);
3534 sig = dequeue_signal(tsk, &mask, info);
3535 if (!sig && timeout) {
3536 /*
3537 * None ready, temporarily unblock those we're interested
3538 * while we are sleeping in so that we'll be awakened when
3539 * they arrive. Unblocking is always fine, we can avoid
3540 * set_current_blocked().
3541 */
3542 tsk->real_blocked = tsk->blocked;
3543 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3544 recalc_sigpending();
3545 spin_unlock_irq(&tsk->sighand->siglock);
3546
3547 __set_current_state(TASK_INTERRUPTIBLE);
3548 ret = freezable_schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3549 HRTIMER_MODE_REL);
3550 spin_lock_irq(&tsk->sighand->siglock);
3551 __set_task_blocked(tsk, &tsk->real_blocked);
3552 sigemptyset(&tsk->real_blocked);
3553 sig = dequeue_signal(tsk, &mask, info);
3554 }
3555 spin_unlock_irq(&tsk->sighand->siglock);
3556
3557 if (sig)
3558 return sig;
3559 return ret ? -EINTR : -EAGAIN;
3560 }
3561
3562 /**
3563 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3564 * in @uthese
3565 * @uthese: queued signals to wait for
3566 * @uinfo: if non-null, the signal's siginfo is returned here
3567 * @uts: upper bound on process time suspension
3568 * @sigsetsize: size of sigset_t type
3569 */
3570 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3571 siginfo_t __user *, uinfo,
3572 const struct __kernel_timespec __user *, uts,
3573 size_t, sigsetsize)
3574 {
3575 sigset_t these;
3576 struct timespec64 ts;
3577 kernel_siginfo_t info;
3578 int ret;
3579
3580 /* XXX: Don't preclude handling different sized sigset_t's. */
3581 if (sigsetsize != sizeof(sigset_t))
3582 return -EINVAL;
3583
3584 if (copy_from_user(&these, uthese, sizeof(these)))
3585 return -EFAULT;
3586
3587 if (uts) {
3588 if (get_timespec64(&ts, uts))
3589 return -EFAULT;
3590 }
3591
3592 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3593
3594 if (ret > 0 && uinfo) {
3595 if (copy_siginfo_to_user(uinfo, &info))
3596 ret = -EFAULT;
3597 }
3598
3599 return ret;
3600 }
3601
3602 #ifdef CONFIG_COMPAT_32BIT_TIME
3603 SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3604 siginfo_t __user *, uinfo,
3605 const struct old_timespec32 __user *, uts,
3606 size_t, sigsetsize)
3607 {
3608 sigset_t these;
3609 struct timespec64 ts;
3610 kernel_siginfo_t info;
3611 int ret;
3612
3613 if (sigsetsize != sizeof(sigset_t))
3614 return -EINVAL;
3615
3616 if (copy_from_user(&these, uthese, sizeof(these)))
3617 return -EFAULT;
3618
3619 if (uts) {
3620 if (get_old_timespec32(&ts, uts))
3621 return -EFAULT;
3622 }
3623
3624 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3625
3626 if (ret > 0 && uinfo) {
3627 if (copy_siginfo_to_user(uinfo, &info))
3628 ret = -EFAULT;
3629 }
3630
3631 return ret;
3632 }
3633 #endif
3634
3635 #ifdef CONFIG_COMPAT
3636 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3637 struct compat_siginfo __user *, uinfo,
3638 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3639 {
3640 sigset_t s;
3641 struct timespec64 t;
3642 kernel_siginfo_t info;
3643 long ret;
3644
3645 if (sigsetsize != sizeof(sigset_t))
3646 return -EINVAL;
3647
3648 if (get_compat_sigset(&s, uthese))
3649 return -EFAULT;
3650
3651 if (uts) {
3652 if (get_timespec64(&t, uts))
3653 return -EFAULT;
3654 }
3655
3656 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3657
3658 if (ret > 0 && uinfo) {
3659 if (copy_siginfo_to_user32(uinfo, &info))
3660 ret = -EFAULT;
3661 }
3662
3663 return ret;
3664 }
3665
3666 #ifdef CONFIG_COMPAT_32BIT_TIME
3667 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3668 struct compat_siginfo __user *, uinfo,
3669 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3670 {
3671 sigset_t s;
3672 struct timespec64 t;
3673 kernel_siginfo_t info;
3674 long ret;
3675
3676 if (sigsetsize != sizeof(sigset_t))
3677 return -EINVAL;
3678
3679 if (get_compat_sigset(&s, uthese))
3680 return -EFAULT;
3681
3682 if (uts) {
3683 if (get_old_timespec32(&t, uts))
3684 return -EFAULT;
3685 }
3686
3687 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3688
3689 if (ret > 0 && uinfo) {
3690 if (copy_siginfo_to_user32(uinfo, &info))
3691 ret = -EFAULT;
3692 }
3693
3694 return ret;
3695 }
3696 #endif
3697 #endif
3698
3699 static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3700 {
3701 clear_siginfo(info);
3702 info->si_signo = sig;
3703 info->si_errno = 0;
3704 info->si_code = SI_USER;
3705 info->si_pid = task_tgid_vnr(current);
3706 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3707 }
3708
3709 /**
3710 * sys_kill - send a signal to a process
3711 * @pid: the PID of the process
3712 * @sig: signal to be sent
3713 */
3714 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3715 {
3716 struct kernel_siginfo info;
3717
3718 prepare_kill_siginfo(sig, &info);
3719
3720 return kill_something_info(sig, &info, pid);
3721 }
3722
3723 /*
3724 * Verify that the signaler and signalee either are in the same pid namespace
3725 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3726 * namespace.
3727 */
3728 static bool access_pidfd_pidns(struct pid *pid)
3729 {
3730 struct pid_namespace *active = task_active_pid_ns(current);
3731 struct pid_namespace *p = ns_of_pid(pid);
3732
3733 for (;;) {
3734 if (!p)
3735 return false;
3736 if (p == active)
3737 break;
3738 p = p->parent;
3739 }
3740
3741 return true;
3742 }
3743
3744 static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3745 siginfo_t __user *info)
3746 {
3747 #ifdef CONFIG_COMPAT
3748 /*
3749 * Avoid hooking up compat syscalls and instead handle necessary
3750 * conversions here. Note, this is a stop-gap measure and should not be
3751 * considered a generic solution.
3752 */
3753 if (in_compat_syscall())
3754 return copy_siginfo_from_user32(
3755 kinfo, (struct compat_siginfo __user *)info);
3756 #endif
3757 return copy_siginfo_from_user(kinfo, info);
3758 }
3759
3760 static struct pid *pidfd_to_pid(const struct file *file)
3761 {
3762 struct pid *pid;
3763
3764 pid = pidfd_pid(file);
3765 if (!IS_ERR(pid))
3766 return pid;
3767
3768 return tgid_pidfd_to_pid(file);
3769 }
3770
3771 /**
3772 * sys_pidfd_send_signal - Signal a process through a pidfd
3773 * @pidfd: file descriptor of the process
3774 * @sig: signal to send
3775 * @info: signal info
3776 * @flags: future flags
3777 *
3778 * The syscall currently only signals via PIDTYPE_PID which covers
3779 * kill(<positive-pid>, <signal>. It does not signal threads or process
3780 * groups.
3781 * In order to extend the syscall to threads and process groups the @flags
3782 * argument should be used. In essence, the @flags argument will determine
3783 * what is signaled and not the file descriptor itself. Put in other words,
3784 * grouping is a property of the flags argument not a property of the file
3785 * descriptor.
3786 *
3787 * Return: 0 on success, negative errno on failure
3788 */
3789 SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3790 siginfo_t __user *, info, unsigned int, flags)
3791 {
3792 int ret;
3793 struct fd f;
3794 struct pid *pid;
3795 kernel_siginfo_t kinfo;
3796
3797 /* Enforce flags be set to 0 until we add an extension. */
3798 if (flags)
3799 return -EINVAL;
3800
3801 f = fdget(pidfd);
3802 if (!f.file)
3803 return -EBADF;
3804
3805 /* Is this a pidfd? */
3806 pid = pidfd_to_pid(f.file);
3807 if (IS_ERR(pid)) {
3808 ret = PTR_ERR(pid);
3809 goto err;
3810 }
3811
3812 ret = -EINVAL;
3813 if (!access_pidfd_pidns(pid))
3814 goto err;
3815
3816 if (info) {
3817 ret = copy_siginfo_from_user_any(&kinfo, info);
3818 if (unlikely(ret))
3819 goto err;
3820
3821 ret = -EINVAL;
3822 if (unlikely(sig != kinfo.si_signo))
3823 goto err;
3824
3825 /* Only allow sending arbitrary signals to yourself. */
3826 ret = -EPERM;
3827 if ((task_pid(current) != pid) &&
3828 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3829 goto err;
3830 } else {
3831 prepare_kill_siginfo(sig, &kinfo);
3832 }
3833
3834 ret = kill_pid_info(sig, &kinfo, pid);
3835
3836 err:
3837 fdput(f);
3838 return ret;
3839 }
3840
3841 static int
3842 do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3843 {
3844 struct task_struct *p;
3845 int error = -ESRCH;
3846
3847 rcu_read_lock();
3848 p = find_task_by_vpid(pid);
3849 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3850 error = check_kill_permission(sig, info, p);
3851 /*
3852 * The null signal is a permissions and process existence
3853 * probe. No signal is actually delivered.
3854 */
3855 if (!error && sig) {
3856 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3857 /*
3858 * If lock_task_sighand() failed we pretend the task
3859 * dies after receiving the signal. The window is tiny,
3860 * and the signal is private anyway.
3861 */
3862 if (unlikely(error == -ESRCH))
3863 error = 0;
3864 }
3865 }
3866 rcu_read_unlock();
3867
3868 return error;
3869 }
3870
3871 static int do_tkill(pid_t tgid, pid_t pid, int sig)
3872 {
3873 struct kernel_siginfo info;
3874
3875 clear_siginfo(&info);
3876 info.si_signo = sig;
3877 info.si_errno = 0;
3878 info.si_code = SI_TKILL;
3879 info.si_pid = task_tgid_vnr(current);
3880 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3881
3882 return do_send_specific(tgid, pid, sig, &info);
3883 }
3884
3885 /**
3886 * sys_tgkill - send signal to one specific thread
3887 * @tgid: the thread group ID of the thread
3888 * @pid: the PID of the thread
3889 * @sig: signal to be sent
3890 *
3891 * This syscall also checks the @tgid and returns -ESRCH even if the PID
3892 * exists but it's not belonging to the target process anymore. This
3893 * method solves the problem of threads exiting and PIDs getting reused.
3894 */
3895 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
3896 {
3897 /* This is only valid for single tasks */
3898 if (pid <= 0 || tgid <= 0)
3899 return -EINVAL;
3900
3901 return do_tkill(tgid, pid, sig);
3902 }
3903
3904 /**
3905 * sys_tkill - send signal to one specific task
3906 * @pid: the PID of the task
3907 * @sig: signal to be sent
3908 *
3909 * Send a signal to only one task, even if it's a CLONE_THREAD task.
3910 */
3911 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
3912 {
3913 /* This is only valid for single tasks */
3914 if (pid <= 0)
3915 return -EINVAL;
3916
3917 return do_tkill(0, pid, sig);
3918 }
3919
3920 static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
3921 {
3922 /* Not even root can pretend to send signals from the kernel.
3923 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3924 */
3925 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3926 (task_pid_vnr(current) != pid))
3927 return -EPERM;
3928
3929 /* POSIX.1b doesn't mention process groups. */
3930 return kill_proc_info(sig, info, pid);
3931 }
3932
3933 /**
3934 * sys_rt_sigqueueinfo - send signal information to a signal
3935 * @pid: the PID of the thread
3936 * @sig: signal to be sent
3937 * @uinfo: signal info to be sent
3938 */
3939 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
3940 siginfo_t __user *, uinfo)
3941 {
3942 kernel_siginfo_t info;
3943 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3944 if (unlikely(ret))
3945 return ret;
3946 return do_rt_sigqueueinfo(pid, sig, &info);
3947 }
3948
3949 #ifdef CONFIG_COMPAT
3950 COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
3951 compat_pid_t, pid,
3952 int, sig,
3953 struct compat_siginfo __user *, uinfo)
3954 {
3955 kernel_siginfo_t info;
3956 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3957 if (unlikely(ret))
3958 return ret;
3959 return do_rt_sigqueueinfo(pid, sig, &info);
3960 }
3961 #endif
3962
3963 static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
3964 {
3965 /* This is only valid for single tasks */
3966 if (pid <= 0 || tgid <= 0)
3967 return -EINVAL;
3968
3969 /* Not even root can pretend to send signals from the kernel.
3970 * Nor can they impersonate a kill()/tgkill(), which adds source info.
3971 */
3972 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
3973 (task_pid_vnr(current) != pid))
3974 return -EPERM;
3975
3976 return do_send_specific(tgid, pid, sig, info);
3977 }
3978
3979 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
3980 siginfo_t __user *, uinfo)
3981 {
3982 kernel_siginfo_t info;
3983 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
3984 if (unlikely(ret))
3985 return ret;
3986 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
3987 }
3988
3989 #ifdef CONFIG_COMPAT
3990 COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
3991 compat_pid_t, tgid,
3992 compat_pid_t, pid,
3993 int, sig,
3994 struct compat_siginfo __user *, uinfo)
3995 {
3996 kernel_siginfo_t info;
3997 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
3998 if (unlikely(ret))
3999 return ret;
4000 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4001 }
4002 #endif
4003
4004 /*
4005 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
4006 */
4007 void kernel_sigaction(int sig, __sighandler_t action)
4008 {
4009 spin_lock_irq(&current->sighand->siglock);
4010 current->sighand->action[sig - 1].sa.sa_handler = action;
4011 if (action == SIG_IGN) {
4012 sigset_t mask;
4013
4014 sigemptyset(&mask);
4015 sigaddset(&mask, sig);
4016
4017 flush_sigqueue_mask(&mask, &current->signal->shared_pending);
4018 flush_sigqueue_mask(&mask, &current->pending);
4019 recalc_sigpending();
4020 }
4021 spin_unlock_irq(&current->sighand->siglock);
4022 }
4023 EXPORT_SYMBOL(kernel_sigaction);
4024
4025 void __weak sigaction_compat_abi(struct k_sigaction *act,
4026 struct k_sigaction *oact)
4027 {
4028 }
4029
4030 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4031 {
4032 struct task_struct *p = current, *t;
4033 struct k_sigaction *k;
4034 sigset_t mask;
4035
4036 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4037 return -EINVAL;
4038
4039 k = &p->sighand->action[sig-1];
4040
4041 spin_lock_irq(&p->sighand->siglock);
4042 if (oact)
4043 *oact = *k;
4044
4045 /*
4046 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4047 * e.g. by having an architecture use the bit in their uapi.
4048 */
4049 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4050
4051 /*
4052 * Clear unknown flag bits in order to allow userspace to detect missing
4053 * support for flag bits and to allow the kernel to use non-uapi bits
4054 * internally.
4055 */
4056 if (act)
4057 act->sa.sa_flags &= UAPI_SA_FLAGS;
4058 if (oact)
4059 oact->sa.sa_flags &= UAPI_SA_FLAGS;
4060
4061 sigaction_compat_abi(act, oact);
4062
4063 if (act) {
4064 sigdelsetmask(&act->sa.sa_mask,
4065 sigmask(SIGKILL) | sigmask(SIGSTOP));
4066 *k = *act;
4067 /*
4068 * POSIX 3.3.1.3:
4069 * "Setting a signal action to SIG_IGN for a signal that is
4070 * pending shall cause the pending signal to be discarded,
4071 * whether or not it is blocked."
4072 *
4073 * "Setting a signal action to SIG_DFL for a signal that is
4074 * pending and whose default action is to ignore the signal
4075 * (for example, SIGCHLD), shall cause the pending signal to
4076 * be discarded, whether or not it is blocked"
4077 */
4078 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4079 sigemptyset(&mask);
4080 sigaddset(&mask, sig);
4081 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
4082 for_each_thread(p, t)
4083 flush_sigqueue_mask(&mask, &t->pending);
4084 }
4085 }
4086
4087 spin_unlock_irq(&p->sighand->siglock);
4088 return 0;
4089 }
4090
4091 static int
4092 do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4093 size_t min_ss_size)
4094 {
4095 struct task_struct *t = current;
4096
4097 if (oss) {
4098 memset(oss, 0, sizeof(stack_t));
4099 oss->ss_sp = (void __user *) t->sas_ss_sp;
4100 oss->ss_size = t->sas_ss_size;
4101 oss->ss_flags = sas_ss_flags(sp) |
4102 (current->sas_ss_flags & SS_FLAG_BITS);
4103 }
4104
4105 if (ss) {
4106 void __user *ss_sp = ss->ss_sp;
4107 size_t ss_size = ss->ss_size;
4108 unsigned ss_flags = ss->ss_flags;
4109 int ss_mode;
4110
4111 if (unlikely(on_sig_stack(sp)))
4112 return -EPERM;
4113
4114 ss_mode = ss_flags & ~SS_FLAG_BITS;
4115 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4116 ss_mode != 0))
4117 return -EINVAL;
4118
4119 if (ss_mode == SS_DISABLE) {
4120 ss_size = 0;
4121 ss_sp = NULL;
4122 } else {
4123 if (unlikely(ss_size < min_ss_size))
4124 return -ENOMEM;
4125 }
4126
4127 t->sas_ss_sp = (unsigned long) ss_sp;
4128 t->sas_ss_size = ss_size;
4129 t->sas_ss_flags = ss_flags;
4130 }
4131 return 0;
4132 }
4133
4134 SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4135 {
4136 stack_t new, old;
4137 int err;
4138 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4139 return -EFAULT;
4140 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4141 current_user_stack_pointer(),
4142 MINSIGSTKSZ);
4143 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4144 err = -EFAULT;
4145 return err;
4146 }
4147
4148 int restore_altstack(const stack_t __user *uss)
4149 {
4150 stack_t new;
4151 if (copy_from_user(&new, uss, sizeof(stack_t)))
4152 return -EFAULT;
4153 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4154 MINSIGSTKSZ);
4155 /* squash all but EFAULT for now */
4156 return 0;
4157 }
4158
4159 int __save_altstack(stack_t __user *uss, unsigned long sp)
4160 {
4161 struct task_struct *t = current;
4162 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4163 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4164 __put_user(t->sas_ss_size, &uss->ss_size);
4165 return err;
4166 }
4167
4168 #ifdef CONFIG_COMPAT
4169 static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4170 compat_stack_t __user *uoss_ptr)
4171 {
4172 stack_t uss, uoss;
4173 int ret;
4174
4175 if (uss_ptr) {
4176 compat_stack_t uss32;
4177 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4178 return -EFAULT;
4179 uss.ss_sp = compat_ptr(uss32.ss_sp);
4180 uss.ss_flags = uss32.ss_flags;
4181 uss.ss_size = uss32.ss_size;
4182 }
4183 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4184 compat_user_stack_pointer(),
4185 COMPAT_MINSIGSTKSZ);
4186 if (ret >= 0 && uoss_ptr) {
4187 compat_stack_t old;
4188 memset(&old, 0, sizeof(old));
4189 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4190 old.ss_flags = uoss.ss_flags;
4191 old.ss_size = uoss.ss_size;
4192 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4193 ret = -EFAULT;
4194 }
4195 return ret;
4196 }
4197
4198 COMPAT_SYSCALL_DEFINE2(sigaltstack,
4199 const compat_stack_t __user *, uss_ptr,
4200 compat_stack_t __user *, uoss_ptr)
4201 {
4202 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4203 }
4204
4205 int compat_restore_altstack(const compat_stack_t __user *uss)
4206 {
4207 int err = do_compat_sigaltstack(uss, NULL);
4208 /* squash all but -EFAULT for now */
4209 return err == -EFAULT ? err : 0;
4210 }
4211
4212 int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4213 {
4214 int err;
4215 struct task_struct *t = current;
4216 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4217 &uss->ss_sp) |
4218 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4219 __put_user(t->sas_ss_size, &uss->ss_size);
4220 return err;
4221 }
4222 #endif
4223
4224 #ifdef __ARCH_WANT_SYS_SIGPENDING
4225
4226 /**
4227 * sys_sigpending - examine pending signals
4228 * @uset: where mask of pending signal is returned
4229 */
4230 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4231 {
4232 sigset_t set;
4233
4234 if (sizeof(old_sigset_t) > sizeof(*uset))
4235 return -EINVAL;
4236
4237 do_sigpending(&set);
4238
4239 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4240 return -EFAULT;
4241
4242 return 0;
4243 }
4244
4245 #ifdef CONFIG_COMPAT
4246 COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4247 {
4248 sigset_t set;
4249
4250 do_sigpending(&set);
4251
4252 return put_user(set.sig[0], set32);
4253 }
4254 #endif
4255
4256 #endif
4257
4258 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
4259 /**
4260 * sys_sigprocmask - examine and change blocked signals
4261 * @how: whether to add, remove, or set signals
4262 * @nset: signals to add or remove (if non-null)
4263 * @oset: previous value of signal mask if non-null
4264 *
4265 * Some platforms have their own version with special arguments;
4266 * others support only sys_rt_sigprocmask.
4267 */
4268
4269 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4270 old_sigset_t __user *, oset)
4271 {
4272 old_sigset_t old_set, new_set;
4273 sigset_t new_blocked;
4274
4275 old_set = current->blocked.sig[0];
4276
4277 if (nset) {
4278 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4279 return -EFAULT;
4280
4281 new_blocked = current->blocked;
4282
4283 switch (how) {
4284 case SIG_BLOCK:
4285 sigaddsetmask(&new_blocked, new_set);
4286 break;
4287 case SIG_UNBLOCK:
4288 sigdelsetmask(&new_blocked, new_set);
4289 break;
4290 case SIG_SETMASK:
4291 new_blocked.sig[0] = new_set;
4292 break;
4293 default:
4294 return -EINVAL;
4295 }
4296
4297 set_current_blocked(&new_blocked);
4298 }
4299
4300 if (oset) {
4301 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4302 return -EFAULT;
4303 }
4304
4305 return 0;
4306 }
4307 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4308
4309 #ifndef CONFIG_ODD_RT_SIGACTION
4310 /**
4311 * sys_rt_sigaction - alter an action taken by a process
4312 * @sig: signal to be sent
4313 * @act: new sigaction
4314 * @oact: used to save the previous sigaction
4315 * @sigsetsize: size of sigset_t type
4316 */
4317 SYSCALL_DEFINE4(rt_sigaction, int, sig,
4318 const struct sigaction __user *, act,
4319 struct sigaction __user *, oact,
4320 size_t, sigsetsize)
4321 {
4322 struct k_sigaction new_sa, old_sa;
4323 int ret;
4324
4325 /* XXX: Don't preclude handling different sized sigset_t's. */
4326 if (sigsetsize != sizeof(sigset_t))
4327 return -EINVAL;
4328
4329 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4330 return -EFAULT;
4331
4332 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4333 if (ret)
4334 return ret;
4335
4336 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4337 return -EFAULT;
4338
4339 return 0;
4340 }
4341 #ifdef CONFIG_COMPAT
4342 COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4343 const struct compat_sigaction __user *, act,
4344 struct compat_sigaction __user *, oact,
4345 compat_size_t, sigsetsize)
4346 {
4347 struct k_sigaction new_ka, old_ka;
4348 #ifdef __ARCH_HAS_SA_RESTORER
4349 compat_uptr_t restorer;
4350 #endif
4351 int ret;
4352
4353 /* XXX: Don't preclude handling different sized sigset_t's. */
4354 if (sigsetsize != sizeof(compat_sigset_t))
4355 return -EINVAL;
4356
4357 if (act) {
4358 compat_uptr_t handler;
4359 ret = get_user(handler, &act->sa_handler);
4360 new_ka.sa.sa_handler = compat_ptr(handler);
4361 #ifdef __ARCH_HAS_SA_RESTORER
4362 ret |= get_user(restorer, &act->sa_restorer);
4363 new_ka.sa.sa_restorer = compat_ptr(restorer);
4364 #endif
4365 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4366 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4367 if (ret)
4368 return -EFAULT;
4369 }
4370
4371 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4372 if (!ret && oact) {
4373 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4374 &oact->sa_handler);
4375 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4376 sizeof(oact->sa_mask));
4377 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4378 #ifdef __ARCH_HAS_SA_RESTORER
4379 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4380 &oact->sa_restorer);
4381 #endif
4382 }
4383 return ret;
4384 }
4385 #endif
4386 #endif /* !CONFIG_ODD_RT_SIGACTION */
4387
4388 #ifdef CONFIG_OLD_SIGACTION
4389 SYSCALL_DEFINE3(sigaction, int, sig,
4390 const struct old_sigaction __user *, act,
4391 struct old_sigaction __user *, oact)
4392 {
4393 struct k_sigaction new_ka, old_ka;
4394 int ret;
4395
4396 if (act) {
4397 old_sigset_t mask;
4398 if (!access_ok(act, sizeof(*act)) ||
4399 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4400 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4401 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4402 __get_user(mask, &act->sa_mask))
4403 return -EFAULT;
4404 #ifdef __ARCH_HAS_KA_RESTORER
4405 new_ka.ka_restorer = NULL;
4406 #endif
4407 siginitset(&new_ka.sa.sa_mask, mask);
4408 }
4409
4410 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4411
4412 if (!ret && oact) {
4413 if (!access_ok(oact, sizeof(*oact)) ||
4414 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4415 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4416 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4417 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4418 return -EFAULT;
4419 }
4420
4421 return ret;
4422 }
4423 #endif
4424 #ifdef CONFIG_COMPAT_OLD_SIGACTION
4425 COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4426 const struct compat_old_sigaction __user *, act,
4427 struct compat_old_sigaction __user *, oact)
4428 {
4429 struct k_sigaction new_ka, old_ka;
4430 int ret;
4431 compat_old_sigset_t mask;
4432 compat_uptr_t handler, restorer;
4433
4434 if (act) {
4435 if (!access_ok(act, sizeof(*act)) ||
4436 __get_user(handler, &act->sa_handler) ||
4437 __get_user(restorer, &act->sa_restorer) ||
4438 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4439 __get_user(mask, &act->sa_mask))
4440 return -EFAULT;
4441
4442 #ifdef __ARCH_HAS_KA_RESTORER
4443 new_ka.ka_restorer = NULL;
4444 #endif
4445 new_ka.sa.sa_handler = compat_ptr(handler);
4446 new_ka.sa.sa_restorer = compat_ptr(restorer);
4447 siginitset(&new_ka.sa.sa_mask, mask);
4448 }
4449
4450 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4451
4452 if (!ret && oact) {
4453 if (!access_ok(oact, sizeof(*oact)) ||
4454 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4455 &oact->sa_handler) ||
4456 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4457 &oact->sa_restorer) ||
4458 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4459 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4460 return -EFAULT;
4461 }
4462 return ret;
4463 }
4464 #endif
4465
4466 #ifdef CONFIG_SGETMASK_SYSCALL
4467
4468 /*
4469 * For backwards compatibility. Functionality superseded by sigprocmask.
4470 */
4471 SYSCALL_DEFINE0(sgetmask)
4472 {
4473 /* SMP safe */
4474 return current->blocked.sig[0];
4475 }
4476
4477 SYSCALL_DEFINE1(ssetmask, int, newmask)
4478 {
4479 int old = current->blocked.sig[0];
4480 sigset_t newset;
4481
4482 siginitset(&newset, newmask);
4483 set_current_blocked(&newset);
4484
4485 return old;
4486 }
4487 #endif /* CONFIG_SGETMASK_SYSCALL */
4488
4489 #ifdef __ARCH_WANT_SYS_SIGNAL
4490 /*
4491 * For backwards compatibility. Functionality superseded by sigaction.
4492 */
4493 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4494 {
4495 struct k_sigaction new_sa, old_sa;
4496 int ret;
4497
4498 new_sa.sa.sa_handler = handler;
4499 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4500 sigemptyset(&new_sa.sa.sa_mask);
4501
4502 ret = do_sigaction(sig, &new_sa, &old_sa);
4503
4504 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4505 }
4506 #endif /* __ARCH_WANT_SYS_SIGNAL */
4507
4508 #ifdef __ARCH_WANT_SYS_PAUSE
4509
4510 SYSCALL_DEFINE0(pause)
4511 {
4512 while (!signal_pending(current)) {
4513 __set_current_state(TASK_INTERRUPTIBLE);
4514 schedule();
4515 }
4516 return -ERESTARTNOHAND;
4517 }
4518
4519 #endif
4520
4521 static int sigsuspend(sigset_t *set)
4522 {
4523 current->saved_sigmask = current->blocked;
4524 set_current_blocked(set);
4525
4526 while (!signal_pending(current)) {
4527 __set_current_state(TASK_INTERRUPTIBLE);
4528 schedule();
4529 }
4530 set_restore_sigmask();
4531 return -ERESTARTNOHAND;
4532 }
4533
4534 /**
4535 * sys_rt_sigsuspend - replace the signal mask for a value with the
4536 * @unewset value until a signal is received
4537 * @unewset: new signal mask value
4538 * @sigsetsize: size of sigset_t type
4539 */
4540 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4541 {
4542 sigset_t newset;
4543
4544 /* XXX: Don't preclude handling different sized sigset_t's. */
4545 if (sigsetsize != sizeof(sigset_t))
4546 return -EINVAL;
4547
4548 if (copy_from_user(&newset, unewset, sizeof(newset)))
4549 return -EFAULT;
4550 return sigsuspend(&newset);
4551 }
4552
4553 #ifdef CONFIG_COMPAT
4554 COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4555 {
4556 sigset_t newset;
4557
4558 /* XXX: Don't preclude handling different sized sigset_t's. */
4559 if (sigsetsize != sizeof(sigset_t))
4560 return -EINVAL;
4561
4562 if (get_compat_sigset(&newset, unewset))
4563 return -EFAULT;
4564 return sigsuspend(&newset);
4565 }
4566 #endif
4567
4568 #ifdef CONFIG_OLD_SIGSUSPEND
4569 SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4570 {
4571 sigset_t blocked;
4572 siginitset(&blocked, mask);
4573 return sigsuspend(&blocked);
4574 }
4575 #endif
4576 #ifdef CONFIG_OLD_SIGSUSPEND3
4577 SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4578 {
4579 sigset_t blocked;
4580 siginitset(&blocked, mask);
4581 return sigsuspend(&blocked);
4582 }
4583 #endif
4584
4585 __weak const char *arch_vma_name(struct vm_area_struct *vma)
4586 {
4587 return NULL;
4588 }
4589
4590 static inline void siginfo_buildtime_checks(void)
4591 {
4592 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4593
4594 /* Verify the offsets in the two siginfos match */
4595 #define CHECK_OFFSET(field) \
4596 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4597
4598 /* kill */
4599 CHECK_OFFSET(si_pid);
4600 CHECK_OFFSET(si_uid);
4601
4602 /* timer */
4603 CHECK_OFFSET(si_tid);
4604 CHECK_OFFSET(si_overrun);
4605 CHECK_OFFSET(si_value);
4606
4607 /* rt */
4608 CHECK_OFFSET(si_pid);
4609 CHECK_OFFSET(si_uid);
4610 CHECK_OFFSET(si_value);
4611
4612 /* sigchld */
4613 CHECK_OFFSET(si_pid);
4614 CHECK_OFFSET(si_uid);
4615 CHECK_OFFSET(si_status);
4616 CHECK_OFFSET(si_utime);
4617 CHECK_OFFSET(si_stime);
4618
4619 /* sigfault */
4620 CHECK_OFFSET(si_addr);
4621 CHECK_OFFSET(si_trapno);
4622 CHECK_OFFSET(si_addr_lsb);
4623 CHECK_OFFSET(si_lower);
4624 CHECK_OFFSET(si_upper);
4625 CHECK_OFFSET(si_pkey);
4626 CHECK_OFFSET(si_perf_data);
4627 CHECK_OFFSET(si_perf_type);
4628
4629 /* sigpoll */
4630 CHECK_OFFSET(si_band);
4631 CHECK_OFFSET(si_fd);
4632
4633 /* sigsys */
4634 CHECK_OFFSET(si_call_addr);
4635 CHECK_OFFSET(si_syscall);
4636 CHECK_OFFSET(si_arch);
4637 #undef CHECK_OFFSET
4638
4639 /* usb asyncio */
4640 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4641 offsetof(struct siginfo, si_addr));
4642 if (sizeof(int) == sizeof(void __user *)) {
4643 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4644 sizeof(void __user *));
4645 } else {
4646 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4647 sizeof_field(struct siginfo, si_uid)) !=
4648 sizeof(void __user *));
4649 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4650 offsetof(struct siginfo, si_uid));
4651 }
4652 #ifdef CONFIG_COMPAT
4653 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4654 offsetof(struct compat_siginfo, si_addr));
4655 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4656 sizeof(compat_uptr_t));
4657 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4658 sizeof_field(struct siginfo, si_pid));
4659 #endif
4660 }
4661
4662 void __init signals_init(void)
4663 {
4664 siginfo_buildtime_checks();
4665
4666 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
4667 }
4668
4669 #ifdef CONFIG_KGDB_KDB
4670 #include <linux/kdb.h>
4671 /*
4672 * kdb_send_sig - Allows kdb to send signals without exposing
4673 * signal internals. This function checks if the required locks are
4674 * available before calling the main signal code, to avoid kdb
4675 * deadlocks.
4676 */
4677 void kdb_send_sig(struct task_struct *t, int sig)
4678 {
4679 static struct task_struct *kdb_prev_t;
4680 int new_t, ret;
4681 if (!spin_trylock(&t->sighand->siglock)) {
4682 kdb_printf("Can't do kill command now.\n"
4683 "The sigmask lock is held somewhere else in "
4684 "kernel, try again later\n");
4685 return;
4686 }
4687 new_t = kdb_prev_t != t;
4688 kdb_prev_t = t;
4689 if (!task_is_running(t) && new_t) {
4690 spin_unlock(&t->sighand->siglock);
4691 kdb_printf("Process is not RUNNING, sending a signal from "
4692 "kdb risks deadlock\n"
4693 "on the run queue locks. "
4694 "The signal has _not_ been sent.\n"
4695 "Reissue the kill command if you want to risk "
4696 "the deadlock.\n");
4697 return;
4698 }
4699 ret = send_signal(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4700 spin_unlock(&t->sighand->siglock);
4701 if (ret)
4702 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4703 sig, t->pid);
4704 else
4705 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4706 }
4707 #endif /* CONFIG_KGDB_KDB */