1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/signal.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * 1997-11-02 Modified for POSIX.1b signals by Richard Henderson
8 *
9 * 2003-06-02 Jim Houston - Concurrent Computer Corp.
10 * Changes to use preallocated sigqueue structures
11 * to allow signals to be sent reliably.
12 */
13
14#include <linux/slab.h>
15#include <linux/export.h>
16#include <linux/init.h>
17#include <linux/sched/mm.h>
18#include <linux/sched/user.h>
19#include <linux/sched/debug.h>
20#include <linux/sched/task.h>
21#include <linux/sched/task_stack.h>
22#include <linux/sched/cputime.h>
23#include <linux/file.h>
24#include <linux/fs.h>
25#include <linux/mm.h>
26#include <linux/proc_fs.h>
27#include <linux/tty.h>
28#include <linux/binfmts.h>
29#include <linux/coredump.h>
30#include <linux/security.h>
31#include <linux/syscalls.h>
32#include <linux/ptrace.h>
33#include <linux/signal.h>
34#include <linux/signalfd.h>
35#include <linux/ratelimit.h>
36#include <linux/task_work.h>
37#include <linux/capability.h>
38#include <linux/freezer.h>
39#include <linux/pid_namespace.h>
40#include <linux/nsproxy.h>
41#include <linux/user_namespace.h>
42#include <linux/uprobes.h>
43#include <linux/compat.h>
44#include <linux/cn_proc.h>
45#include <linux/compiler.h>
46#include <linux/posix-timers.h>
47#include <linux/cgroup.h>
48#include <linux/audit.h>
49#include <linux/sysctl.h>
50
51#define CREATE_TRACE_POINTS
52#include <trace/events/signal.h>
53
54#include <asm/param.h>
55#include <linux/uaccess.h>
56#include <asm/unistd.h>
57#include <asm/siginfo.h>
58#include <asm/cacheflush.h>
59#include <asm/syscall.h> /* for syscall_get_* */
60
61/*
62 * SLAB caches for signal bits.
63 */
64
65static struct kmem_cache *sigqueue_cachep;
66
67int print_fatal_signals __read_mostly;
68
69static void __user *sig_handler(struct task_struct *t, int sig)
70{
71 return t->sighand->action[sig - 1].sa.sa_handler;
72}
73
74static inline bool sig_handler_ignored(void __user *handler, int sig)
75{
76 /* Is it explicitly or implicitly ignored? */
77 return handler == SIG_IGN ||
78 (handler == SIG_DFL && sig_kernel_ignore(sig));
79}
80
81static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
82{
83 void __user *handler;
84
85 handler = sig_handler(t, sig);
86
87 /* SIGKILL and SIGSTOP may not be sent to the global init */
88 if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
89 return true;
90
91 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
92 handler == SIG_DFL && !(force && sig_kernel_only(sig)))
93 return true;
94
95 /* Only allow kernel generated signals to this kthread */
96 if (unlikely((t->flags & PF_KTHREAD) &&
97 (handler == SIG_KTHREAD_KERNEL) && !force))
98 return true;
99
100 return sig_handler_ignored(handler, sig);
101}
102
103static bool sig_ignored(struct task_struct *t, int sig, bool force)
104{
105 /*
106 * Blocked signals are never ignored, since the
107 * signal handler may change by the time it is
108 * unblocked.
109 */
110 if (sigismember(set: &t->blocked, sig: sig) || sigismember(set: &t->real_blocked, sig: sig))
111 return false;
112
113 /*
114 * Tracers may want to know about even ignored signal unless it
115 * is SIGKILL which can't be reported anyway but can be ignored
116 * by SIGNAL_UNKILLABLE task.
117 */
118 if (t->ptrace && sig != SIGKILL)
119 return false;
120
121 return sig_task_ignored(t, sig, force);
122}
123
124/*
125 * Re-calculate pending state from the set of locally pending
126 * signals, globally pending signals, and blocked signals.
127 */
128static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
129{
130 unsigned long ready;
131 long i;
132
133 switch (_NSIG_WORDS) {
134 default:
135 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
136 ready |= signal->sig[i] &~ blocked->sig[i];
137 break;
138
139 case 4: ready = signal->sig[3] &~ blocked->sig[3];
140 ready |= signal->sig[2] &~ blocked->sig[2];
141 ready |= signal->sig[1] &~ blocked->sig[1];
142 ready |= signal->sig[0] &~ blocked->sig[0];
143 break;
144
145 case 2: ready = signal->sig[1] &~ blocked->sig[1];
146 ready |= signal->sig[0] &~ blocked->sig[0];
147 break;
148
149 case 1: ready = signal->sig[0] &~ blocked->sig[0];
150 }
151 return ready != 0;
152}
153
154#define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
155
156static bool recalc_sigpending_tsk(struct task_struct *t)
157{
158 if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) ||
159 PENDING(&t->pending, &t->blocked) ||
160 PENDING(&t->signal->shared_pending, &t->blocked) ||
161 cgroup_task_frozen(task: t)) {
162 set_tsk_thread_flag(tsk: t, TIF_SIGPENDING);
163 return true;
164 }
165
166 /*
167 * We must never clear the flag in another thread, or in current
168 * when it's possible the current syscall is returning -ERESTART*.
169 * So we don't clear it here, and only callers who know they should do.
170 */
171 return false;
172}
173
174/*
175 * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
176 * This is superfluous when called on current, the wakeup is a harmless no-op.
177 */
178void recalc_sigpending_and_wake(struct task_struct *t)
179{
180 if (recalc_sigpending_tsk(t))
181 signal_wake_up(t, fatal: 0);
182}
183
184void recalc_sigpending(void)
185{
186 if (!recalc_sigpending_tsk(current) && !freezing(current))
187 clear_thread_flag(TIF_SIGPENDING);
188
189}
190EXPORT_SYMBOL(recalc_sigpending);
191
192void calculate_sigpending(void)
193{
194 /* Have any signals or users of TIF_SIGPENDING been delayed
195 * until after fork?
196 */
197 spin_lock_irq(lock: &current->sighand->siglock);
198 set_tsk_thread_flag(current, TIF_SIGPENDING);
199 recalc_sigpending();
200 spin_unlock_irq(lock: &current->sighand->siglock);
201}
202
203/* Given the mask, find the first available signal that should be serviced. */
204
205#define SYNCHRONOUS_MASK \
206 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
207 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
208
209int next_signal(struct sigpending *pending, sigset_t *mask)
210{
211 unsigned long i, *s, *m, x;
212 int sig = 0;
213
214 s = pending->signal.sig;
215 m = mask->sig;
216
217 /*
218 * Handle the first word specially: it contains the
219 * synchronous signals that need to be dequeued first.
220 */
221 x = *s &~ *m;
222 if (x) {
223 if (x & SYNCHRONOUS_MASK)
224 x &= SYNCHRONOUS_MASK;
225 sig = ffz(~x) + 1;
226 return sig;
227 }
228
229 switch (_NSIG_WORDS) {
230 default:
231 for (i = 1; i < _NSIG_WORDS; ++i) {
232 x = *++s &~ *++m;
233 if (!x)
234 continue;
235 sig = ffz(~x) + i*_NSIG_BPW + 1;
236 break;
237 }
238 break;
239
240 case 2:
241 x = s[1] &~ m[1];
242 if (!x)
243 break;
244 sig = ffz(~x) + _NSIG_BPW + 1;
245 break;
246
247 case 1:
248 /* Nothing to do */
249 break;
250 }
251
252 return sig;
253}
254
255static inline void print_dropped_signal(int sig)
256{
257 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
258
259 if (!print_fatal_signals)
260 return;
261
262 if (!__ratelimit(&ratelimit_state))
263 return;
264
265 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
266 current->comm, current->pid, sig);
267}
268
269/**
270 * task_set_jobctl_pending - set jobctl pending bits
271 * @task: target task
272 * @mask: pending bits to set
273 *
274 * Clear @mask from @task->jobctl. @mask must be subset of
275 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
276 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
277 * cleared. If @task is already being killed or exiting, this function
278 * becomes noop.
279 *
280 * CONTEXT:
281 * Must be called with @task->sighand->siglock held.
282 *
283 * RETURNS:
284 * %true if @mask is set, %false if made noop because @task was dying.
285 */
286bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
287{
288 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
289 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
290 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
291
292 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
293 return false;
294
295 if (mask & JOBCTL_STOP_SIGMASK)
296 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
297
298 task->jobctl |= mask;
299 return true;
300}
301
302/**
303 * task_clear_jobctl_trapping - clear jobctl trapping bit
304 * @task: target task
305 *
306 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
307 * Clear it and wake up the ptracer. Note that we don't need any further
308 * locking. @task->siglock guarantees that @task->parent points to the
309 * ptracer.
310 *
311 * CONTEXT:
312 * Must be called with @task->sighand->siglock held.
313 */
314void task_clear_jobctl_trapping(struct task_struct *task)
315{
316 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
317 task->jobctl &= ~JOBCTL_TRAPPING;
318 smp_mb(); /* advised by wake_up_bit() */
319 wake_up_bit(word: &task->jobctl, JOBCTL_TRAPPING_BIT);
320 }
321}
322
323/**
324 * task_clear_jobctl_pending - clear jobctl pending bits
325 * @task: target task
326 * @mask: pending bits to clear
327 *
328 * Clear @mask from @task->jobctl. @mask must be subset of
329 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
330 * STOP bits are cleared together.
331 *
332 * If clearing of @mask leaves no stop or trap pending, this function calls
333 * task_clear_jobctl_trapping().
334 *
335 * CONTEXT:
336 * Must be called with @task->sighand->siglock held.
337 */
338void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
339{
340 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
341
342 if (mask & JOBCTL_STOP_PENDING)
343 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
344
345 task->jobctl &= ~mask;
346
347 if (!(task->jobctl & JOBCTL_PENDING_MASK))
348 task_clear_jobctl_trapping(task);
349}
350
351/**
352 * task_participate_group_stop - participate in a group stop
353 * @task: task participating in a group stop
354 *
355 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
356 * Group stop states are cleared and the group stop count is consumed if
357 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
358 * stop, the appropriate `SIGNAL_*` flags are set.
359 *
360 * CONTEXT:
361 * Must be called with @task->sighand->siglock held.
362 *
363 * RETURNS:
364 * %true if group stop completion should be notified to the parent, %false
365 * otherwise.
366 */
367static bool task_participate_group_stop(struct task_struct *task)
368{
369 struct signal_struct *sig = task->signal;
370 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
371
372 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
373
374 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
375
376 if (!consume)
377 return false;
378
379 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
380 sig->group_stop_count--;
381
382 /*
383 * Tell the caller to notify completion iff we are entering into a
384 * fresh group stop. Read comment in do_signal_stop() for details.
385 */
386 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
387 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
388 return true;
389 }
390 return false;
391}
392
393void task_join_group_stop(struct task_struct *task)
394{
395 unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK;
396 struct signal_struct *sig = current->signal;
397
398 if (sig->group_stop_count) {
399 sig->group_stop_count++;
400 mask |= JOBCTL_STOP_CONSUME;
401 } else if (!(sig->flags & SIGNAL_STOP_STOPPED))
402 return;
403
404 /* Have the new thread join an on-going signal group stop */
405 task_set_jobctl_pending(task, mask: mask | JOBCTL_STOP_PENDING);
406}
407
408/*
409 * allocate a new signal queue record
410 * - this may be called without locks if and only if t == current, otherwise an
411 * appropriate lock must be held to stop the target task from exiting
412 */
413static struct sigqueue *
414__sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
415 int override_rlimit, const unsigned int sigqueue_flags)
416{
417 struct sigqueue *q = NULL;
418 struct ucounts *ucounts;
419 long sigpending;
420
421 /*
422 * Protect access to @t credentials. This can go away when all
423 * callers hold rcu read lock.
424 *
425 * NOTE! A pending signal will hold on to the user refcount,
426 * and we get/put the refcount only when the sigpending count
427 * changes from/to zero.
428 */
429 rcu_read_lock();
430 ucounts = task_ucounts(t);
431 sigpending = inc_rlimit_get_ucounts(ucounts, type: UCOUNT_RLIMIT_SIGPENDING);
432 rcu_read_unlock();
433 if (!sigpending)
434 return NULL;
435
436 if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
437 q = kmem_cache_alloc(cachep: sigqueue_cachep, flags: gfp_flags);
438 } else {
439 print_dropped_signal(sig);
440 }
441
442 if (unlikely(q == NULL)) {
443 dec_rlimit_put_ucounts(ucounts, type: UCOUNT_RLIMIT_SIGPENDING);
444 } else {
445 INIT_LIST_HEAD(list: &q->list);
446 q->flags = sigqueue_flags;
447 q->ucounts = ucounts;
448 }
449 return q;
450}
451
452static void __sigqueue_free(struct sigqueue *q)
453{
454 if (q->flags & SIGQUEUE_PREALLOC)
455 return;
456 if (q->ucounts) {
457 dec_rlimit_put_ucounts(ucounts: q->ucounts, type: UCOUNT_RLIMIT_SIGPENDING);
458 q->ucounts = NULL;
459 }
460 kmem_cache_free(s: sigqueue_cachep, objp: q);
461}
462
463void flush_sigqueue(struct sigpending *queue)
464{
465 struct sigqueue *q;
466
467 sigemptyset(set: &queue->signal);
468 while (!list_empty(head: &queue->list)) {
469 q = list_entry(queue->list.next, struct sigqueue , list);
470 list_del_init(entry: &q->list);
471 __sigqueue_free(q);
472 }
473}
474
475/*
476 * Flush all pending signals for this kthread.
477 */
478void flush_signals(struct task_struct *t)
479{
480 unsigned long flags;
481
482 spin_lock_irqsave(&t->sighand->siglock, flags);
483 clear_tsk_thread_flag(tsk: t, TIF_SIGPENDING);
484 flush_sigqueue(queue: &t->pending);
485 flush_sigqueue(queue: &t->signal->shared_pending);
486 spin_unlock_irqrestore(lock: &t->sighand->siglock, flags);
487}
488EXPORT_SYMBOL(flush_signals);
489
490#ifdef CONFIG_POSIX_TIMERS
491static void __flush_itimer_signals(struct sigpending *pending)
492{
493 sigset_t signal, retain;
494 struct sigqueue *q, *n;
495
496 signal = pending->signal;
497 sigemptyset(set: &retain);
498
499 list_for_each_entry_safe(q, n, &pending->list, list) {
500 int sig = q->info.si_signo;
501
502 if (likely(q->info.si_code != SI_TIMER)) {
503 sigaddset(set: &retain, sig: sig);
504 } else {
505 sigdelset(set: &signal, sig: sig);
506 list_del_init(entry: &q->list);
507 __sigqueue_free(q);
508 }
509 }
510
511 sigorsets(r: &pending->signal, a: &signal, b: &retain);
512}
513
514void flush_itimer_signals(void)
515{
516 struct task_struct *tsk = current;
517 unsigned long flags;
518
519 spin_lock_irqsave(&tsk->sighand->siglock, flags);
520 __flush_itimer_signals(pending: &tsk->pending);
521 __flush_itimer_signals(pending: &tsk->signal->shared_pending);
522 spin_unlock_irqrestore(lock: &tsk->sighand->siglock, flags);
523}
524#endif
525
526void ignore_signals(struct task_struct *t)
527{
528 int i;
529
530 for (i = 0; i < _NSIG; ++i)
531 t->sighand->action[i].sa.sa_handler = SIG_IGN;
532
533 flush_signals(t);
534}
535
536/*
537 * Flush all handlers for a task.
538 */
539
540void
541flush_signal_handlers(struct task_struct *t, int force_default)
542{
543 int i;
544 struct k_sigaction *ka = &t->sighand->action[0];
545 for (i = _NSIG ; i != 0 ; i--) {
546 if (force_default || ka->sa.sa_handler != SIG_IGN)
547 ka->sa.sa_handler = SIG_DFL;
548 ka->sa.sa_flags = 0;
549#ifdef __ARCH_HAS_SA_RESTORER
550 ka->sa.sa_restorer = NULL;
551#endif
552 sigemptyset(set: &ka->sa.sa_mask);
553 ka++;
554 }
555}
556
557bool unhandled_signal(struct task_struct *tsk, int sig)
558{
559 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
560 if (is_global_init(tsk))
561 return true;
562
563 if (handler != SIG_IGN && handler != SIG_DFL)
564 return false;
565
566 /* If dying, we handle all new signals by ignoring them */
567 if (fatal_signal_pending(p: tsk))
568 return false;
569
570 /* if ptraced, let the tracer determine */
571 return !tsk->ptrace;
572}
573
574static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
575 bool *resched_timer)
576{
577 struct sigqueue *q, *first = NULL;
578
579 /*
580 * Collect the siginfo appropriate to this signal. Check if
581 * there is another siginfo for the same signal.
582 */
583 list_for_each_entry(q, &list->list, list) {
584 if (q->info.si_signo == sig) {
585 if (first)
586 goto still_pending;
587 first = q;
588 }
589 }
590
591 sigdelset(set: &list->signal, sig: sig);
592
593 if (first) {
594still_pending:
595 list_del_init(entry: &first->list);
596 copy_siginfo(to: info, from: &first->info);
597
598 *resched_timer =
599 (first->flags & SIGQUEUE_PREALLOC) &&
600 (info->si_code == SI_TIMER) &&
601 (info->si_sys_private);
602
603 __sigqueue_free(q: first);
604 } else {
605 /*
606 * Ok, it wasn't in the queue. This must be
607 * a fast-pathed signal or we must have been
608 * out of queue space. So zero out the info.
609 */
610 clear_siginfo(info);
611 info->si_signo = sig;
612 info->si_errno = 0;
613 info->si_code = SI_USER;
614 info->si_pid = 0;
615 info->si_uid = 0;
616 }
617}
618
619static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
620 kernel_siginfo_t *info, bool *resched_timer)
621{
622 int sig = next_signal(pending, mask);
623
624 if (sig)
625 collect_signal(sig, list: pending, info, resched_timer);
626 return sig;
627}
628
629/*
630 * Dequeue a signal and return the element to the caller, which is
631 * expected to free it.
632 *
633 * All callers have to hold the siglock.
634 */
635int dequeue_signal(struct task_struct *tsk, sigset_t *mask,
636 kernel_siginfo_t *info, enum pid_type *type)
637{
638 bool resched_timer = false;
639 int signr;
640
641 /* We only dequeue private signals from ourselves, we don't let
642 * signalfd steal them
643 */
644 *type = PIDTYPE_PID;
645 signr = __dequeue_signal(pending: &tsk->pending, mask, info, resched_timer: &resched_timer);
646 if (!signr) {
647 *type = PIDTYPE_TGID;
648 signr = __dequeue_signal(pending: &tsk->signal->shared_pending,
649 mask, info, resched_timer: &resched_timer);
650#ifdef CONFIG_POSIX_TIMERS
651 /*
652 * itimer signal ?
653 *
654 * itimers are process shared and we restart periodic
655 * itimers in the signal delivery path to prevent DoS
656 * attacks in the high resolution timer case. This is
657 * compliant with the old way of self-restarting
658 * itimers, as the SIGALRM is a legacy signal and only
659 * queued once. Changing the restart behaviour to
660 * restart the timer in the signal dequeue path is
661 * reducing the timer noise on heavy loaded !highres
662 * systems too.
663 */
664 if (unlikely(signr == SIGALRM)) {
665 struct hrtimer *tmr = &tsk->signal->real_timer;
666
667 if (!hrtimer_is_queued(timer: tmr) &&
668 tsk->signal->it_real_incr != 0) {
669 hrtimer_forward(timer: tmr, now: tmr->base->get_time(),
670 interval: tsk->signal->it_real_incr);
671 hrtimer_restart(timer: tmr);
672 }
673 }
674#endif
675 }
676
677 recalc_sigpending();
678 if (!signr)
679 return 0;
680
681 if (unlikely(sig_kernel_stop(signr))) {
682 /*
683 * Set a marker that we have dequeued a stop signal. Our
684 * caller might release the siglock and then the pending
685 * stop signal it is about to process is no longer in the
686 * pending bitmasks, but must still be cleared by a SIGCONT
687 * (and overruled by a SIGKILL). So those cases clear this
688 * shared flag after we've set it. Note that this flag may
689 * remain set after the signal we return is ignored or
690 * handled. That doesn't matter because its only purpose
691 * is to alert stop-signal processing code when another
692 * processor has come along and cleared the flag.
693 */
694 current->jobctl |= JOBCTL_STOP_DEQUEUED;
695 }
696#ifdef CONFIG_POSIX_TIMERS
697 if (resched_timer) {
698 /*
699 * Release the siglock to ensure proper locking order
700 * of timer locks outside of siglocks. Note, we leave
701 * irqs disabled here, since the posix-timers code is
702 * about to disable them again anyway.
703 */
704 spin_unlock(lock: &tsk->sighand->siglock);
705 posixtimer_rearm(info);
706 spin_lock(lock: &tsk->sighand->siglock);
707
708 /* Don't expose the si_sys_private value to userspace */
709 info->si_sys_private = 0;
710 }
711#endif
712 return signr;
713}
714EXPORT_SYMBOL_GPL(dequeue_signal);
715
716static int dequeue_synchronous_signal(kernel_siginfo_t *info)
717{
718 struct task_struct *tsk = current;
719 struct sigpending *pending = &tsk->pending;
720 struct sigqueue *q, *sync = NULL;
721
722 /*
723 * Might a synchronous signal be in the queue?
724 */
725 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
726 return 0;
727
728 /*
729 * Return the first synchronous signal in the queue.
730 */
731 list_for_each_entry(q, &pending->list, list) {
732 /* Synchronous signals have a positive si_code */
733 if ((q->info.si_code > SI_USER) &&
734 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
735 sync = q;
736 goto next;
737 }
738 }
739 return 0;
740next:
741 /*
742 * Check if there is another siginfo for the same signal.
743 */
744 list_for_each_entry_continue(q, &pending->list, list) {
745 if (q->info.si_signo == sync->info.si_signo)
746 goto still_pending;
747 }
748
749 sigdelset(set: &pending->signal, sig: sync->info.si_signo);
750 recalc_sigpending();
751still_pending:
752 list_del_init(entry: &sync->list);
753 copy_siginfo(to: info, from: &sync->info);
754 __sigqueue_free(q: sync);
755 return info->si_signo;
756}
757
758/*
759 * Tell a process that it has a new active signal..
760 *
761 * NOTE! we rely on the previous spin_lock to
762 * lock interrupts for us! We can only be called with
763 * "siglock" held, and the local interrupt must
764 * have been disabled when that got acquired!
765 *
766 * No need to set need_resched since signal event passing
767 * goes through ->blocked
768 */
769void signal_wake_up_state(struct task_struct *t, unsigned int state)
770{
771 lockdep_assert_held(&t->sighand->siglock);
772
773 set_tsk_thread_flag(tsk: t, TIF_SIGPENDING);
774
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(tsk: t, state: state | TASK_INTERRUPTIBLE))
783 kick_process(tsk: 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 */
792static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
793{
794 struct sigqueue *q, *n;
795 sigset_t m;
796
797 sigandsets(r: &m, a: mask, b: &s->signal);
798 if (sigisemptyset(set: &m))
799 return;
800
801 sigandnsets(r: &s->signal, a: &s->signal, b: mask);
802 list_for_each_entry_safe(q, n, &s->list, list) {
803 if (sigismember(set: mask, sig: q->info.si_signo)) {
804 list_del_init(entry: &q->list);
805 __sigqueue_free(q);
806 }
807 }
808}
809
810static inline int is_si_special(const struct kernel_siginfo *info)
811{
812 return info <= SEND_SIG_PRIV;
813}
814
815static 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 */
824static 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(left: cred->euid, right: tcred->suid) ||
830 uid_eq(left: cred->euid, right: tcred->uid) ||
831 uid_eq(left: cred->uid, right: tcred->suid) ||
832 uid_eq(left: cred->uid, right: tcred->uid) ||
833 ns_capable(ns: 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 */
840static 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, p2: t) &&
857 !kill_ok_by_cred(t)) {
858 switch (sig) {
859 case SIGCONT:
860 sid = task_session(task: 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(p: 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 */
893static void ptrace_trap_notify(struct task_struct *t)
894{
895 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
896 lockdep_assert_held(&t->sighand->siglock);
897
898 task_set_jobctl_pending(task: t, JOBCTL_TRAP_NOTIFY);
899 ptrace_signal_wake_up(t, resume: 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 */
912static 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) {
919 if (signal->core_state)
920 return sig == SIGKILL;
921 /*
922 * The process is in the middle of dying, drop the signal.
923 */
924 return false;
925 } else if (sig_kernel_stop(sig)) {
926 /*
927 * This is a stop signal. Remove SIGCONT from all queues.
928 */
929 siginitset(set: &flush, sigmask(SIGCONT));
930 flush_sigqueue_mask(mask: &flush, s: &signal->shared_pending);
931 for_each_thread(p, t)
932 flush_sigqueue_mask(mask: &flush, s: &t->pending);
933 } else if (sig == SIGCONT) {
934 unsigned int why;
935 /*
936 * Remove all stop signals from all queues, wake all threads.
937 */
938 siginitset(set: &flush, SIG_KERNEL_STOP_MASK);
939 flush_sigqueue_mask(mask: &flush, s: &signal->shared_pending);
940 for_each_thread(p, t) {
941 flush_sigqueue_mask(mask: &flush, s: &t->pending);
942 task_clear_jobctl_pending(task: t, JOBCTL_STOP_PENDING);
943 if (likely(!(t->ptrace & PT_SEIZED))) {
944 t->jobctl &= ~JOBCTL_STOPPED;
945 wake_up_state(tsk: t, __TASK_STOPPED);
946 } else
947 ptrace_trap_notify(t);
948 }
949
950 /*
951 * Notify the parent with CLD_CONTINUED if we were stopped.
952 *
953 * If we were in the middle of a group stop, we pretend it
954 * was already finished, and then continued. Since SIGCHLD
955 * doesn't queue we report only CLD_STOPPED, as if the next
956 * CLD_CONTINUED was dropped.
957 */
958 why = 0;
959 if (signal->flags & SIGNAL_STOP_STOPPED)
960 why |= SIGNAL_CLD_CONTINUED;
961 else if (signal->group_stop_count)
962 why |= SIGNAL_CLD_STOPPED;
963
964 if (why) {
965 /*
966 * The first thread which returns from do_signal_stop()
967 * will take ->siglock, notice SIGNAL_CLD_MASK, and
968 * notify its parent. See get_signal().
969 */
970 signal_set_stop_flags(sig: signal, flags: why | SIGNAL_STOP_CONTINUED);
971 signal->group_stop_count = 0;
972 signal->group_exit_code = 0;
973 }
974 }
975
976 return !sig_ignored(t: p, sig, force);
977}
978
979/*
980 * Test if P wants to take SIG. After we've checked all threads with this,
981 * it's equivalent to finding no threads not blocking SIG. Any threads not
982 * blocking SIG were ruled out because they are not running and already
983 * have pending signals. Such threads will dequeue from the shared queue
984 * as soon as they're available, so putting the signal on the shared queue
985 * will be equivalent to sending it to one such thread.
986 */
987static inline bool wants_signal(int sig, struct task_struct *p)
988{
989 if (sigismember(set: &p->blocked, sig: sig))
990 return false;
991
992 if (p->flags & PF_EXITING)
993 return false;
994
995 if (sig == SIGKILL)
996 return true;
997
998 if (task_is_stopped_or_traced(p))
999 return false;
1000
1001 return task_curr(p) || !task_sigpending(p);
1002}
1003
1004static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
1005{
1006 struct signal_struct *signal = p->signal;
1007 struct task_struct *t;
1008
1009 /*
1010 * Now find a thread we can wake up to take the signal off the queue.
1011 *
1012 * Try the suggested task first (may or may not be the main thread).
1013 */
1014 if (wants_signal(sig, p))
1015 t = p;
1016 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1017 /*
1018 * There is just one thread and it does not need to be woken.
1019 * It will dequeue unblocked signals before it runs again.
1020 */
1021 return;
1022 else {
1023 /*
1024 * Otherwise try to find a suitable thread.
1025 */
1026 t = signal->curr_target;
1027 while (!wants_signal(sig, p: t)) {
1028 t = next_thread(p: t);
1029 if (t == signal->curr_target)
1030 /*
1031 * No thread needs to be woken.
1032 * Any eligible threads will see
1033 * the signal in the queue soon.
1034 */
1035 return;
1036 }
1037 signal->curr_target = t;
1038 }
1039
1040 /*
1041 * Found a killable thread. If the signal will be fatal,
1042 * then start taking the whole group down immediately.
1043 */
1044 if (sig_fatal(p, sig) &&
1045 (signal->core_state || !(signal->flags & SIGNAL_GROUP_EXIT)) &&
1046 !sigismember(set: &t->real_blocked, sig: sig) &&
1047 (sig == SIGKILL || !p->ptrace)) {
1048 /*
1049 * This signal will be fatal to the whole group.
1050 */
1051 if (!sig_kernel_coredump(sig)) {
1052 /*
1053 * Start a group exit and wake everybody up.
1054 * This way we don't have other threads
1055 * running and doing things after a slower
1056 * thread has the fatal signal pending.
1057 */
1058 signal->flags = SIGNAL_GROUP_EXIT;
1059 signal->group_exit_code = sig;
1060 signal->group_stop_count = 0;
1061 __for_each_thread(signal, t) {
1062 task_clear_jobctl_pending(task: t, JOBCTL_PENDING_MASK);
1063 sigaddset(set: &t->pending.signal, SIGKILL);
1064 signal_wake_up(t, fatal: 1);
1065 }
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, fatal: sig == SIGKILL);
1075 return;
1076}
1077
1078static inline bool legacy_queue(struct sigpending *signals, int sig)
1079{
1080 return (sig < SIGRTMIN) && sigismember(set: &signals->signal, sig: sig);
1081}
1082
1083static int __send_signal_locked(int sig, struct kernel_siginfo *info,
1084 struct task_struct *t, 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 lockdep_assert_held(&t->sighand->siglock);
1092
1093 result = TRACE_SIGNAL_IGNORED;
1094 if (!prepare_signal(sig, p: 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(signals: 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, sigqueue_flags: 0);
1129
1130 if (q) {
1131 list_add_tail(new: &q->list, head: &pending->list);
1132 switch ((unsigned long) info) {
1133 case (unsigned long) SEND_SIG_NOINFO:
1134 clear_siginfo(info: &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 ns: task_active_pid_ns(tsk: 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(info: &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(to: &q->info, from: 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
1176out_set:
1177 signalfd_notify(tsk: t, sig);
1178 sigaddset(set: &pending->signal, sig: 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(set: signal, SIG_KERNEL_STOP_MASK);
1188 else if (sig_kernel_stop(sig))
1189 sigdelset(set: signal, SIGCONT);
1190 sigaddset(set: signal, sig: sig);
1191 }
1192 }
1193
1194 complete_signal(sig, p: t, type);
1195ret:
1196 trace_signal_generate(sig, info, task: t, group: type != PIDTYPE_PID, result);
1197 return ret;
1198}
1199
1200static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1201{
1202 bool ret = false;
1203 switch (siginfo_layout(sig: info->si_signo, si_code: 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_FAULT_PERF_EVENT:
1217 case SIL_SYS:
1218 ret = false;
1219 break;
1220 }
1221 return ret;
1222}
1223
1224int send_signal_locked(int sig, struct kernel_siginfo *info,
1225 struct task_struct *t, 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, ns: task_active_pid_ns(tsk: 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(), uid: info->si_uid);
1244 info->si_uid = from_kuid_munged(to: 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, ns: task_active_pid_ns(tsk: t))) {
1253 info->si_pid = 0;
1254 force = true;
1255 }
1256 }
1257 return __send_signal_locked(sig, info, t, type, force);
1258}
1259
1260static void print_fatal_signal(int signr)
1261{
1262 struct pt_regs *regs = task_pt_regs(current);
1263 struct file *exe_file;
1264
1265 exe_file = get_task_exe_file(current);
1266 if (exe_file) {
1267 pr_info("%pD: %s: potentially unexpected fatal signal %d.\n",
1268 exe_file, current->comm, signr);
1269 fput(exe_file);
1270 } else {
1271 pr_info("%s: potentially unexpected fatal signal %d.\n",
1272 current->comm, signr);
1273 }
1274
1275#if defined(__i386__) && !defined(__arch_um__)
1276 pr_info("code at %08lx: ", regs->ip);
1277 {
1278 int i;
1279 for (i = 0; i < 16; i++) {
1280 unsigned char insn;
1281
1282 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1283 break;
1284 pr_cont("%02x ", insn);
1285 }
1286 }
1287 pr_cont("\n");
1288#endif
1289 preempt_disable();
1290 show_regs(regs);
1291 preempt_enable();
1292}
1293
1294static int __init setup_print_fatal_signals(char *str)
1295{
1296 get_option (str: &str, pint: &print_fatal_signals);
1297
1298 return 1;
1299}
1300
1301__setup("print-fatal-signals=", setup_print_fatal_signals);
1302
1303int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1304 enum pid_type type)
1305{
1306 unsigned long flags;
1307 int ret = -ESRCH;
1308
1309 if (lock_task_sighand(task: p, flags: &flags)) {
1310 ret = send_signal_locked(sig, info, t: p, type);
1311 unlock_task_sighand(task: p, flags: &flags);
1312 }
1313
1314 return ret;
1315}
1316
1317enum sig_handler {
1318 HANDLER_CURRENT, /* If reachable use the current handler */
1319 HANDLER_SIG_DFL, /* Always use SIG_DFL handler semantics */
1320 HANDLER_EXIT, /* Only visible as the process exit code */
1321};
1322
1323/*
1324 * Force a signal that the process can't ignore: if necessary
1325 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1326 *
1327 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1328 * since we do not want to have a signal handler that was blocked
1329 * be invoked when user space had explicitly blocked it.
1330 *
1331 * We don't want to have recursive SIGSEGV's etc, for example,
1332 * that is why we also clear SIGNAL_UNKILLABLE.
1333 */
1334static int
1335force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t,
1336 enum sig_handler handler)
1337{
1338 unsigned long int flags;
1339 int ret, blocked, ignored;
1340 struct k_sigaction *action;
1341 int sig = info->si_signo;
1342
1343 spin_lock_irqsave(&t->sighand->siglock, flags);
1344 action = &t->sighand->action[sig-1];
1345 ignored = action->sa.sa_handler == SIG_IGN;
1346 blocked = sigismember(set: &t->blocked, sig: sig);
1347 if (blocked || ignored || (handler != HANDLER_CURRENT)) {
1348 action->sa.sa_handler = SIG_DFL;
1349 if (handler == HANDLER_EXIT)
1350 action->sa.sa_flags |= SA_IMMUTABLE;
1351 if (blocked) {
1352 sigdelset(set: &t->blocked, sig: sig);
1353 recalc_sigpending_and_wake(t);
1354 }
1355 }
1356 /*
1357 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1358 * debugging to leave init killable. But HANDLER_EXIT is always fatal.
1359 */
1360 if (action->sa.sa_handler == SIG_DFL &&
1361 (!t->ptrace || (handler == HANDLER_EXIT)))
1362 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1363 ret = send_signal_locked(sig, info, t, type: PIDTYPE_PID);
1364 spin_unlock_irqrestore(lock: &t->sighand->siglock, flags);
1365
1366 return ret;
1367}
1368
1369int force_sig_info(struct kernel_siginfo *info)
1370{
1371 return force_sig_info_to_task(info, current, handler: HANDLER_CURRENT);
1372}
1373
1374/*
1375 * Nuke all other threads in the group.
1376 */
1377int zap_other_threads(struct task_struct *p)
1378{
1379 struct task_struct *t = p;
1380 int count = 0;
1381
1382 p->signal->group_stop_count = 0;
1383
1384 while_each_thread(p, t) {
1385 task_clear_jobctl_pending(task: t, JOBCTL_PENDING_MASK);
1386 /* Don't require de_thread to wait for the vhost_worker */
1387 if ((t->flags & (PF_IO_WORKER | PF_USER_WORKER)) != PF_USER_WORKER)
1388 count++;
1389
1390 /* Don't bother with already dead threads */
1391 if (t->exit_state)
1392 continue;
1393 sigaddset(set: &t->pending.signal, SIGKILL);
1394 signal_wake_up(t, fatal: 1);
1395 }
1396
1397 return count;
1398}
1399
1400struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1401 unsigned long *flags)
1402{
1403 struct sighand_struct *sighand;
1404
1405 rcu_read_lock();
1406 for (;;) {
1407 sighand = rcu_dereference(tsk->sighand);
1408 if (unlikely(sighand == NULL))
1409 break;
1410
1411 /*
1412 * This sighand can be already freed and even reused, but
1413 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1414 * initializes ->siglock: this slab can't go away, it has
1415 * the same object type, ->siglock can't be reinitialized.
1416 *
1417 * We need to ensure that tsk->sighand is still the same
1418 * after we take the lock, we can race with de_thread() or
1419 * __exit_signal(). In the latter case the next iteration
1420 * must see ->sighand == NULL.
1421 */
1422 spin_lock_irqsave(&sighand->siglock, *flags);
1423 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1424 break;
1425 spin_unlock_irqrestore(lock: &sighand->siglock, flags: *flags);
1426 }
1427 rcu_read_unlock();
1428
1429 return sighand;
1430}
1431
1432#ifdef CONFIG_LOCKDEP
1433void lockdep_assert_task_sighand_held(struct task_struct *task)
1434{
1435 struct sighand_struct *sighand;
1436
1437 rcu_read_lock();
1438 sighand = rcu_dereference(task->sighand);
1439 if (sighand)
1440 lockdep_assert_held(&sighand->siglock);
1441 else
1442 WARN_ON_ONCE(1);
1443 rcu_read_unlock();
1444}
1445#endif
1446
1447/*
1448 * send signal info to all the members of a group
1449 */
1450int group_send_sig_info(int sig, struct kernel_siginfo *info,
1451 struct task_struct *p, enum pid_type type)
1452{
1453 int ret;
1454
1455 rcu_read_lock();
1456 ret = check_kill_permission(sig, info, t: p);
1457 rcu_read_unlock();
1458
1459 if (!ret && sig)
1460 ret = do_send_sig_info(sig, info, p, type);
1461
1462 return ret;
1463}
1464
1465/*
1466 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1467 * control characters do (^C, ^Z etc)
1468 * - the caller must hold at least a readlock on tasklist_lock
1469 */
1470int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1471{
1472 struct task_struct *p = NULL;
1473 int ret = -ESRCH;
1474
1475 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1476 int err = group_send_sig_info(sig, info, p, type: PIDTYPE_PGID);
1477 /*
1478 * If group_send_sig_info() succeeds at least once ret
1479 * becomes 0 and after that the code below has no effect.
1480 * Otherwise we return the last err or -ESRCH if this
1481 * process group is empty.
1482 */
1483 if (ret)
1484 ret = err;
1485 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1486
1487 return ret;
1488}
1489
1490int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1491{
1492 int error = -ESRCH;
1493 struct task_struct *p;
1494
1495 for (;;) {
1496 rcu_read_lock();
1497 p = pid_task(pid, PIDTYPE_PID);
1498 if (p)
1499 error = group_send_sig_info(sig, info, p, type: PIDTYPE_TGID);
1500 rcu_read_unlock();
1501 if (likely(!p || error != -ESRCH))
1502 return error;
1503
1504 /*
1505 * The task was unhashed in between, try again. If it
1506 * is dead, pid_task() will return NULL, if we race with
1507 * de_thread() it will find the new leader.
1508 */
1509 }
1510}
1511
1512static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1513{
1514 int error;
1515 rcu_read_lock();
1516 error = kill_pid_info(sig, info, pid: find_vpid(nr: pid));
1517 rcu_read_unlock();
1518 return error;
1519}
1520
1521static inline bool kill_as_cred_perm(const struct cred *cred,
1522 struct task_struct *target)
1523{
1524 const struct cred *pcred = __task_cred(target);
1525
1526 return uid_eq(left: cred->euid, right: pcred->suid) ||
1527 uid_eq(left: cred->euid, right: pcred->uid) ||
1528 uid_eq(left: cred->uid, right: pcred->suid) ||
1529 uid_eq(left: cred->uid, right: pcred->uid);
1530}
1531
1532/*
1533 * The usb asyncio usage of siginfo is wrong. The glibc support
1534 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1535 * AKA after the generic fields:
1536 * kernel_pid_t si_pid;
1537 * kernel_uid32_t si_uid;
1538 * sigval_t si_value;
1539 *
1540 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1541 * after the generic fields is:
1542 * void __user *si_addr;
1543 *
1544 * This is a practical problem when there is a 64bit big endian kernel
1545 * and a 32bit userspace. As the 32bit address will encoded in the low
1546 * 32bits of the pointer. Those low 32bits will be stored at higher
1547 * address than appear in a 32 bit pointer. So userspace will not
1548 * see the address it was expecting for it's completions.
1549 *
1550 * There is nothing in the encoding that can allow
1551 * copy_siginfo_to_user32 to detect this confusion of formats, so
1552 * handle this by requiring the caller of kill_pid_usb_asyncio to
1553 * notice when this situration takes place and to store the 32bit
1554 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1555 * parameter.
1556 */
1557int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1558 struct pid *pid, const struct cred *cred)
1559{
1560 struct kernel_siginfo info;
1561 struct task_struct *p;
1562 unsigned long flags;
1563 int ret = -EINVAL;
1564
1565 if (!valid_signal(sig))
1566 return ret;
1567
1568 clear_siginfo(info: &info);
1569 info.si_signo = sig;
1570 info.si_errno = errno;
1571 info.si_code = SI_ASYNCIO;
1572 *((sigval_t *)&info.si_pid) = addr;
1573
1574 rcu_read_lock();
1575 p = pid_task(pid, PIDTYPE_PID);
1576 if (!p) {
1577 ret = -ESRCH;
1578 goto out_unlock;
1579 }
1580 if (!kill_as_cred_perm(cred, target: p)) {
1581 ret = -EPERM;
1582 goto out_unlock;
1583 }
1584 ret = security_task_kill(p, info: &info, sig, cred);
1585 if (ret)
1586 goto out_unlock;
1587
1588 if (sig) {
1589 if (lock_task_sighand(task: p, flags: &flags)) {
1590 ret = __send_signal_locked(sig, info: &info, t: p, type: PIDTYPE_TGID, force: false);
1591 unlock_task_sighand(task: p, flags: &flags);
1592 } else
1593 ret = -ESRCH;
1594 }
1595out_unlock:
1596 rcu_read_unlock();
1597 return ret;
1598}
1599EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1600
1601/*
1602 * kill_something_info() interprets pid in interesting ways just like kill(2).
1603 *
1604 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1605 * is probably wrong. Should make it like BSD or SYSV.
1606 */
1607
1608static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1609{
1610 int ret;
1611
1612 if (pid > 0)
1613 return kill_proc_info(sig, info, pid);
1614
1615 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1616 if (pid == INT_MIN)
1617 return -ESRCH;
1618
1619 read_lock(&tasklist_lock);
1620 if (pid != -1) {
1621 ret = __kill_pgrp_info(sig, info,
1622 pgrp: pid ? find_vpid(nr: -pid) : task_pgrp(current));
1623 } else {
1624 int retval = 0, count = 0;
1625 struct task_struct * p;
1626
1627 for_each_process(p) {
1628 if (task_pid_vnr(tsk: p) > 1 &&
1629 !same_thread_group(p1: p, current)) {
1630 int err = group_send_sig_info(sig, info, p,
1631 type: PIDTYPE_MAX);
1632 ++count;
1633 if (err != -EPERM)
1634 retval = err;
1635 }
1636 }
1637 ret = count ? retval : -ESRCH;
1638 }
1639 read_unlock(&tasklist_lock);
1640
1641 return ret;
1642}
1643
1644/*
1645 * These are for backward compatibility with the rest of the kernel source.
1646 */
1647
1648int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1649{
1650 /*
1651 * Make sure legacy kernel users don't send in bad values
1652 * (normal paths check this in check_kill_permission).
1653 */
1654 if (!valid_signal(sig))
1655 return -EINVAL;
1656
1657 return do_send_sig_info(sig, info, p, type: PIDTYPE_PID);
1658}
1659EXPORT_SYMBOL(send_sig_info);
1660
1661#define __si_special(priv) \
1662 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1663
1664int
1665send_sig(int sig, struct task_struct *p, int priv)
1666{
1667 return send_sig_info(sig, __si_special(priv), p);
1668}
1669EXPORT_SYMBOL(send_sig);
1670
1671void force_sig(int sig)
1672{
1673 struct kernel_siginfo info;
1674
1675 clear_siginfo(info: &info);
1676 info.si_signo = sig;
1677 info.si_errno = 0;
1678 info.si_code = SI_KERNEL;
1679 info.si_pid = 0;
1680 info.si_uid = 0;
1681 force_sig_info(info: &info);
1682}
1683EXPORT_SYMBOL(force_sig);
1684
1685void force_fatal_sig(int sig)
1686{
1687 struct kernel_siginfo info;
1688
1689 clear_siginfo(info: &info);
1690 info.si_signo = sig;
1691 info.si_errno = 0;
1692 info.si_code = SI_KERNEL;
1693 info.si_pid = 0;
1694 info.si_uid = 0;
1695 force_sig_info_to_task(info: &info, current, handler: HANDLER_SIG_DFL);
1696}
1697
1698void force_exit_sig(int sig)
1699{
1700 struct kernel_siginfo info;
1701
1702 clear_siginfo(info: &info);
1703 info.si_signo = sig;
1704 info.si_errno = 0;
1705 info.si_code = SI_KERNEL;
1706 info.si_pid = 0;
1707 info.si_uid = 0;
1708 force_sig_info_to_task(info: &info, current, handler: HANDLER_EXIT);
1709}
1710
1711/*
1712 * When things go south during signal handling, we
1713 * will force a SIGSEGV. And if the signal that caused
1714 * the problem was already a SIGSEGV, we'll want to
1715 * make sure we don't even try to deliver the signal..
1716 */
1717void force_sigsegv(int sig)
1718{
1719 if (sig == SIGSEGV)
1720 force_fatal_sig(SIGSEGV);
1721 else
1722 force_sig(SIGSEGV);
1723}
1724
1725int force_sig_fault_to_task(int sig, int code, void __user *addr,
1726 struct task_struct *t)
1727{
1728 struct kernel_siginfo info;
1729
1730 clear_siginfo(info: &info);
1731 info.si_signo = sig;
1732 info.si_errno = 0;
1733 info.si_code = code;
1734 info.si_addr = addr;
1735 return force_sig_info_to_task(info: &info, t, handler: HANDLER_CURRENT);
1736}
1737
1738int force_sig_fault(int sig, int code, void __user *addr)
1739{
1740 return force_sig_fault_to_task(sig, code, addr, current);
1741}
1742
1743int send_sig_fault(int sig, int code, void __user *addr, struct task_struct *t)
1744{
1745 struct kernel_siginfo info;
1746
1747 clear_siginfo(info: &info);
1748 info.si_signo = sig;
1749 info.si_errno = 0;
1750 info.si_code = code;
1751 info.si_addr = addr;
1752 return send_sig_info(info.si_signo, &info, t);
1753}
1754
1755int force_sig_mceerr(int code, void __user *addr, short lsb)
1756{
1757 struct kernel_siginfo info;
1758
1759 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1760 clear_siginfo(info: &info);
1761 info.si_signo = SIGBUS;
1762 info.si_errno = 0;
1763 info.si_code = code;
1764 info.si_addr = addr;
1765 info.si_addr_lsb = lsb;
1766 return force_sig_info(info: &info);
1767}
1768
1769int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1770{
1771 struct kernel_siginfo info;
1772
1773 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1774 clear_siginfo(info: &info);
1775 info.si_signo = SIGBUS;
1776 info.si_errno = 0;
1777 info.si_code = code;
1778 info.si_addr = addr;
1779 info.si_addr_lsb = lsb;
1780 return send_sig_info(info.si_signo, &info, t);
1781}
1782EXPORT_SYMBOL(send_sig_mceerr);
1783
1784int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1785{
1786 struct kernel_siginfo info;
1787
1788 clear_siginfo(info: &info);
1789 info.si_signo = SIGSEGV;
1790 info.si_errno = 0;
1791 info.si_code = SEGV_BNDERR;
1792 info.si_addr = addr;
1793 info.si_lower = lower;
1794 info.si_upper = upper;
1795 return force_sig_info(info: &info);
1796}
1797
1798#ifdef SEGV_PKUERR
1799int force_sig_pkuerr(void __user *addr, u32 pkey)
1800{
1801 struct kernel_siginfo info;
1802
1803 clear_siginfo(info: &info);
1804 info.si_signo = SIGSEGV;
1805 info.si_errno = 0;
1806 info.si_code = SEGV_PKUERR;
1807 info.si_addr = addr;
1808 info.si_pkey = pkey;
1809 return force_sig_info(info: &info);
1810}
1811#endif
1812
1813int send_sig_perf(void __user *addr, u32 type, u64 sig_data)
1814{
1815 struct kernel_siginfo info;
1816
1817 clear_siginfo(info: &info);
1818 info.si_signo = SIGTRAP;
1819 info.si_errno = 0;
1820 info.si_code = TRAP_PERF;
1821 info.si_addr = addr;
1822 info.si_perf_data = sig_data;
1823 info.si_perf_type = type;
1824
1825 /*
1826 * Signals generated by perf events should not terminate the whole
1827 * process if SIGTRAP is blocked, however, delivering the signal
1828 * asynchronously is better than not delivering at all. But tell user
1829 * space if the signal was asynchronous, so it can clearly be
1830 * distinguished from normal synchronous ones.
1831 */
1832 info.si_perf_flags = sigismember(set: &current->blocked, sig: info.si_signo) ?
1833 TRAP_PERF_FLAG_ASYNC :
1834 0;
1835
1836 return send_sig_info(info.si_signo, &info, current);
1837}
1838
1839/**
1840 * force_sig_seccomp - signals the task to allow in-process syscall emulation
1841 * @syscall: syscall number to send to userland
1842 * @reason: filter-supplied reason code to send to userland (via si_errno)
1843 * @force_coredump: true to trigger a coredump
1844 *
1845 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
1846 */
1847int force_sig_seccomp(int syscall, int reason, bool force_coredump)
1848{
1849 struct kernel_siginfo info;
1850
1851 clear_siginfo(info: &info);
1852 info.si_signo = SIGSYS;
1853 info.si_code = SYS_SECCOMP;
1854 info.si_call_addr = (void __user *)KSTK_EIP(current);
1855 info.si_errno = reason;
1856 info.si_arch = syscall_get_arch(current);
1857 info.si_syscall = syscall;
1858 return force_sig_info_to_task(info: &info, current,
1859 handler: force_coredump ? HANDLER_EXIT : HANDLER_CURRENT);
1860}
1861
1862/* For the crazy architectures that include trap information in
1863 * the errno field, instead of an actual errno value.
1864 */
1865int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1866{
1867 struct kernel_siginfo info;
1868
1869 clear_siginfo(info: &info);
1870 info.si_signo = SIGTRAP;
1871 info.si_errno = errno;
1872 info.si_code = TRAP_HWBKPT;
1873 info.si_addr = addr;
1874 return force_sig_info(info: &info);
1875}
1876
1877/* For the rare architectures that include trap information using
1878 * si_trapno.
1879 */
1880int force_sig_fault_trapno(int sig, int code, void __user *addr, int trapno)
1881{
1882 struct kernel_siginfo info;
1883
1884 clear_siginfo(info: &info);
1885 info.si_signo = sig;
1886 info.si_errno = 0;
1887 info.si_code = code;
1888 info.si_addr = addr;
1889 info.si_trapno = trapno;
1890 return force_sig_info(info: &info);
1891}
1892
1893/* For the rare architectures that include trap information using
1894 * si_trapno.
1895 */
1896int send_sig_fault_trapno(int sig, int code, void __user *addr, int trapno,
1897 struct task_struct *t)
1898{
1899 struct kernel_siginfo info;
1900
1901 clear_siginfo(info: &info);
1902 info.si_signo = sig;
1903 info.si_errno = 0;
1904 info.si_code = code;
1905 info.si_addr = addr;
1906 info.si_trapno = trapno;
1907 return send_sig_info(info.si_signo, &info, t);
1908}
1909
1910int kill_pgrp(struct pid *pid, int sig, int priv)
1911{
1912 int ret;
1913
1914 read_lock(&tasklist_lock);
1915 ret = __kill_pgrp_info(sig, __si_special(priv), pgrp: pid);
1916 read_unlock(&tasklist_lock);
1917
1918 return ret;
1919}
1920EXPORT_SYMBOL(kill_pgrp);
1921
1922int kill_pid(struct pid *pid, int sig, int priv)
1923{
1924 return kill_pid_info(sig, __si_special(priv), pid);
1925}
1926EXPORT_SYMBOL(kill_pid);
1927
1928/*
1929 * These functions support sending signals using preallocated sigqueue
1930 * structures. This is needed "because realtime applications cannot
1931 * afford to lose notifications of asynchronous events, like timer
1932 * expirations or I/O completions". In the case of POSIX Timers
1933 * we allocate the sigqueue structure from the timer_create. If this
1934 * allocation fails we are able to report the failure to the application
1935 * with an EAGAIN error.
1936 */
1937struct sigqueue *sigqueue_alloc(void)
1938{
1939 return __sigqueue_alloc(sig: -1, current, GFP_KERNEL, override_rlimit: 0, SIGQUEUE_PREALLOC);
1940}
1941
1942void sigqueue_free(struct sigqueue *q)
1943{
1944 unsigned long flags;
1945 spinlock_t *lock = &current->sighand->siglock;
1946
1947 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1948 /*
1949 * We must hold ->siglock while testing q->list
1950 * to serialize with collect_signal() or with
1951 * __exit_signal()->flush_sigqueue().
1952 */
1953 spin_lock_irqsave(lock, flags);
1954 q->flags &= ~SIGQUEUE_PREALLOC;
1955 /*
1956 * If it is queued it will be freed when dequeued,
1957 * like the "regular" sigqueue.
1958 */
1959 if (!list_empty(head: &q->list))
1960 q = NULL;
1961 spin_unlock_irqrestore(lock, flags);
1962
1963 if (q)
1964 __sigqueue_free(q);
1965}
1966
1967int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1968{
1969 int sig = q->info.si_signo;
1970 struct sigpending *pending;
1971 struct task_struct *t;
1972 unsigned long flags;
1973 int ret, result;
1974
1975 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1976
1977 ret = -1;
1978 rcu_read_lock();
1979
1980 /*
1981 * This function is used by POSIX timers to deliver a timer signal.
1982 * Where type is PIDTYPE_PID (such as for timers with SIGEV_THREAD_ID
1983 * set), the signal must be delivered to the specific thread (queues
1984 * into t->pending).
1985 *
1986 * Where type is not PIDTYPE_PID, signals must be delivered to the
1987 * process. In this case, prefer to deliver to current if it is in
1988 * the same thread group as the target process, which avoids
1989 * unnecessarily waking up a potentially idle task.
1990 */
1991 t = pid_task(pid, type);
1992 if (!t)
1993 goto ret;
1994 if (type != PIDTYPE_PID && same_thread_group(p1: t, current))
1995 t = current;
1996 if (!likely(lock_task_sighand(t, &flags)))
1997 goto ret;
1998
1999 ret = 1; /* the signal is ignored */
2000 result = TRACE_SIGNAL_IGNORED;
2001 if (!prepare_signal(sig, p: t, force: false))
2002 goto out;
2003
2004 ret = 0;
2005 if (unlikely(!list_empty(&q->list))) {
2006 /*
2007 * If an SI_TIMER entry is already queue just increment
2008 * the overrun count.
2009 */
2010 BUG_ON(q->info.si_code != SI_TIMER);
2011 q->info.si_overrun++;
2012 result = TRACE_SIGNAL_ALREADY_PENDING;
2013 goto out;
2014 }
2015 q->info.si_overrun = 0;
2016
2017 signalfd_notify(tsk: t, sig);
2018 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
2019 list_add_tail(new: &q->list, head: &pending->list);
2020 sigaddset(set: &pending->signal, sig: sig);
2021 complete_signal(sig, p: t, type);
2022 result = TRACE_SIGNAL_DELIVERED;
2023out:
2024 trace_signal_generate(sig, info: &q->info, task: t, group: type != PIDTYPE_PID, result);
2025 unlock_task_sighand(task: t, flags: &flags);
2026ret:
2027 rcu_read_unlock();
2028 return ret;
2029}
2030
2031static void do_notify_pidfd(struct task_struct *task)
2032{
2033 struct pid *pid;
2034
2035 WARN_ON(task->exit_state == 0);
2036 pid = task_pid(task);
2037 wake_up_all(&pid->wait_pidfd);
2038}
2039
2040/*
2041 * Let a parent know about the death of a child.
2042 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
2043 *
2044 * Returns true if our parent ignored us and so we've switched to
2045 * self-reaping.
2046 */
2047bool do_notify_parent(struct task_struct *tsk, int sig)
2048{
2049 struct kernel_siginfo info;
2050 unsigned long flags;
2051 struct sighand_struct *psig;
2052 bool autoreap = false;
2053 u64 utime, stime;
2054
2055 WARN_ON_ONCE(sig == -1);
2056
2057 /* do_notify_parent_cldstop should have been called instead. */
2058 WARN_ON_ONCE(task_is_stopped_or_traced(tsk));
2059
2060 WARN_ON_ONCE(!tsk->ptrace &&
2061 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
2062
2063 /* Wake up all pidfd waiters */
2064 do_notify_pidfd(task: tsk);
2065
2066 if (sig != SIGCHLD) {
2067 /*
2068 * This is only possible if parent == real_parent.
2069 * Check if it has changed security domain.
2070 */
2071 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
2072 sig = SIGCHLD;
2073 }
2074
2075 clear_siginfo(info: &info);
2076 info.si_signo = sig;
2077 info.si_errno = 0;
2078 /*
2079 * We are under tasklist_lock here so our parent is tied to
2080 * us and cannot change.
2081 *
2082 * task_active_pid_ns will always return the same pid namespace
2083 * until a task passes through release_task.
2084 *
2085 * write_lock() currently calls preempt_disable() which is the
2086 * same as rcu_read_lock(), but according to Oleg, this is not
2087 * correct to rely on this
2088 */
2089 rcu_read_lock();
2090 info.si_pid = task_pid_nr_ns(tsk, ns: task_active_pid_ns(tsk: tsk->parent));
2091 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
2092 task_uid(tsk));
2093 rcu_read_unlock();
2094
2095 task_cputime(t: tsk, utime: &utime, stime: &stime);
2096 info.si_utime = nsec_to_clock_t(x: utime + tsk->signal->utime);
2097 info.si_stime = nsec_to_clock_t(x: stime + tsk->signal->stime);
2098
2099 info.si_status = tsk->exit_code & 0x7f;
2100 if (tsk->exit_code & 0x80)
2101 info.si_code = CLD_DUMPED;
2102 else if (tsk->exit_code & 0x7f)
2103 info.si_code = CLD_KILLED;
2104 else {
2105 info.si_code = CLD_EXITED;
2106 info.si_status = tsk->exit_code >> 8;
2107 }
2108
2109 psig = tsk->parent->sighand;
2110 spin_lock_irqsave(&psig->siglock, flags);
2111 if (!tsk->ptrace && sig == SIGCHLD &&
2112 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
2113 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
2114 /*
2115 * We are exiting and our parent doesn't care. POSIX.1
2116 * defines special semantics for setting SIGCHLD to SIG_IGN
2117 * or setting the SA_NOCLDWAIT flag: we should be reaped
2118 * automatically and not left for our parent's wait4 call.
2119 * Rather than having the parent do it as a magic kind of
2120 * signal handler, we just set this to tell do_exit that we
2121 * can be cleaned up without becoming a zombie. Note that
2122 * we still call __wake_up_parent in this case, because a
2123 * blocked sys_wait4 might now return -ECHILD.
2124 *
2125 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
2126 * is implementation-defined: we do (if you don't want
2127 * it, just use SIG_IGN instead).
2128 */
2129 autoreap = true;
2130 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2131 sig = 0;
2132 }
2133 /*
2134 * Send with __send_signal as si_pid and si_uid are in the
2135 * parent's namespaces.
2136 */
2137 if (valid_signal(sig) && sig)
2138 __send_signal_locked(sig, info: &info, t: tsk->parent, type: PIDTYPE_TGID, force: false);
2139 __wake_up_parent(p: tsk, parent: tsk->parent);
2140 spin_unlock_irqrestore(lock: &psig->siglock, flags);
2141
2142 return autoreap;
2143}
2144
2145/**
2146 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2147 * @tsk: task reporting the state change
2148 * @for_ptracer: the notification is for ptracer
2149 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2150 *
2151 * Notify @tsk's parent that the stopped/continued state has changed. If
2152 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2153 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2154 *
2155 * CONTEXT:
2156 * Must be called with tasklist_lock at least read locked.
2157 */
2158static void do_notify_parent_cldstop(struct task_struct *tsk,
2159 bool for_ptracer, int why)
2160{
2161 struct kernel_siginfo info;
2162 unsigned long flags;
2163 struct task_struct *parent;
2164 struct sighand_struct *sighand;
2165 u64 utime, stime;
2166
2167 if (for_ptracer) {
2168 parent = tsk->parent;
2169 } else {
2170 tsk = tsk->group_leader;
2171 parent = tsk->real_parent;
2172 }
2173
2174 clear_siginfo(info: &info);
2175 info.si_signo = SIGCHLD;
2176 info.si_errno = 0;
2177 /*
2178 * see comment in do_notify_parent() about the following 4 lines
2179 */
2180 rcu_read_lock();
2181 info.si_pid = task_pid_nr_ns(tsk, ns: task_active_pid_ns(tsk: parent));
2182 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2183 rcu_read_unlock();
2184
2185 task_cputime(t: tsk, utime: &utime, stime: &stime);
2186 info.si_utime = nsec_to_clock_t(x: utime);
2187 info.si_stime = nsec_to_clock_t(x: stime);
2188
2189 info.si_code = why;
2190 switch (why) {
2191 case CLD_CONTINUED:
2192 info.si_status = SIGCONT;
2193 break;
2194 case CLD_STOPPED:
2195 info.si_status = tsk->signal->group_exit_code & 0x7f;
2196 break;
2197 case CLD_TRAPPED:
2198 info.si_status = tsk->exit_code & 0x7f;
2199 break;
2200 default:
2201 BUG();
2202 }
2203
2204 sighand = parent->sighand;
2205 spin_lock_irqsave(&sighand->siglock, flags);
2206 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2207 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2208 send_signal_locked(SIGCHLD, info: &info, t: parent, type: PIDTYPE_TGID);
2209 /*
2210 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2211 */
2212 __wake_up_parent(p: tsk, parent);
2213 spin_unlock_irqrestore(lock: &sighand->siglock, flags);
2214}
2215
2216/*
2217 * This must be called with current->sighand->siglock held.
2218 *
2219 * This should be the path for all ptrace stops.
2220 * We always set current->last_siginfo while stopped here.
2221 * That makes it a way to test a stopped process for
2222 * being ptrace-stopped vs being job-control-stopped.
2223 *
2224 * Returns the signal the ptracer requested the code resume
2225 * with. If the code did not stop because the tracer is gone,
2226 * the stop signal remains unchanged unless clear_code.
2227 */
2228static int ptrace_stop(int exit_code, int why, unsigned long message,
2229 kernel_siginfo_t *info)
2230 __releases(&current->sighand->siglock)
2231 __acquires(&current->sighand->siglock)
2232{
2233 bool gstop_done = false;
2234
2235 if (arch_ptrace_stop_needed()) {
2236 /*
2237 * The arch code has something special to do before a
2238 * ptrace stop. This is allowed to block, e.g. for faults
2239 * on user stack pages. We can't keep the siglock while
2240 * calling arch_ptrace_stop, so we must release it now.
2241 * To preserve proper semantics, we must do this before
2242 * any signal bookkeeping like checking group_stop_count.
2243 */
2244 spin_unlock_irq(lock: &current->sighand->siglock);
2245 arch_ptrace_stop();
2246 spin_lock_irq(lock: &current->sighand->siglock);
2247 }
2248
2249 /*
2250 * After this point ptrace_signal_wake_up or signal_wake_up
2251 * will clear TASK_TRACED if ptrace_unlink happens or a fatal
2252 * signal comes in. Handle previous ptrace_unlinks and fatal
2253 * signals here to prevent ptrace_stop sleeping in schedule.
2254 */
2255 if (!current->ptrace || __fatal_signal_pending(current))
2256 return exit_code;
2257
2258 set_special_state(TASK_TRACED);
2259 current->jobctl |= JOBCTL_TRACED;
2260
2261 /*
2262 * We're committing to trapping. TRACED should be visible before
2263 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2264 * Also, transition to TRACED and updates to ->jobctl should be
2265 * atomic with respect to siglock and should be done after the arch
2266 * hook as siglock is released and regrabbed across it.
2267 *
2268 * TRACER TRACEE
2269 *
2270 * ptrace_attach()
2271 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2272 * do_wait()
2273 * set_current_state() smp_wmb();
2274 * ptrace_do_wait()
2275 * wait_task_stopped()
2276 * task_stopped_code()
2277 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2278 */
2279 smp_wmb();
2280
2281 current->ptrace_message = message;
2282 current->last_siginfo = info;
2283 current->exit_code = exit_code;
2284
2285 /*
2286 * If @why is CLD_STOPPED, we're trapping to participate in a group
2287 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2288 * across siglock relocks since INTERRUPT was scheduled, PENDING
2289 * could be clear now. We act as if SIGCONT is received after
2290 * TASK_TRACED is entered - ignore it.
2291 */
2292 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2293 gstop_done = task_participate_group_stop(current);
2294
2295 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2296 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2297 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2298 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2299
2300 /* entering a trap, clear TRAPPING */
2301 task_clear_jobctl_trapping(current);
2302
2303 spin_unlock_irq(lock: &current->sighand->siglock);
2304 read_lock(&tasklist_lock);
2305 /*
2306 * Notify parents of the stop.
2307 *
2308 * While ptraced, there are two parents - the ptracer and
2309 * the real_parent of the group_leader. The ptracer should
2310 * know about every stop while the real parent is only
2311 * interested in the completion of group stop. The states
2312 * for the two don't interact with each other. Notify
2313 * separately unless they're gonna be duplicates.
2314 */
2315 if (current->ptrace)
2316 do_notify_parent_cldstop(current, for_ptracer: true, why);
2317 if (gstop_done && (!current->ptrace || ptrace_reparented(current)))
2318 do_notify_parent_cldstop(current, for_ptracer: false, why);
2319
2320 /*
2321 * The previous do_notify_parent_cldstop() invocation woke ptracer.
2322 * One a PREEMPTION kernel this can result in preemption requirement
2323 * which will be fulfilled after read_unlock() and the ptracer will be
2324 * put on the CPU.
2325 * The ptracer is in wait_task_inactive(, __TASK_TRACED) waiting for
2326 * this task wait in schedule(). If this task gets preempted then it
2327 * remains enqueued on the runqueue. The ptracer will observe this and
2328 * then sleep for a delay of one HZ tick. In the meantime this task
2329 * gets scheduled, enters schedule() and will wait for the ptracer.
2330 *
2331 * This preemption point is not bad from a correctness point of
2332 * view but extends the runtime by one HZ tick time due to the
2333 * ptracer's sleep. The preempt-disable section ensures that there
2334 * will be no preemption between unlock and schedule() and so
2335 * improving the performance since the ptracer will observe that
2336 * the tracee is scheduled out once it gets on the CPU.
2337 *
2338 * On PREEMPT_RT locking tasklist_lock does not disable preemption.
2339 * Therefore the task can be preempted after do_notify_parent_cldstop()
2340 * before unlocking tasklist_lock so there is no benefit in doing this.
2341 *
2342 * In fact disabling preemption is harmful on PREEMPT_RT because
2343 * the spinlock_t in cgroup_enter_frozen() must not be acquired
2344 * with preemption disabled due to the 'sleeping' spinlock
2345 * substitution of RT.
2346 */
2347 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2348 preempt_disable();
2349 read_unlock(&tasklist_lock);
2350 cgroup_enter_frozen();
2351 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2352 preempt_enable_no_resched();
2353 schedule();
2354 cgroup_leave_frozen(always_leave: true);
2355
2356 /*
2357 * We are back. Now reacquire the siglock before touching
2358 * last_siginfo, so that we are sure to have synchronized with
2359 * any signal-sending on another CPU that wants to examine it.
2360 */
2361 spin_lock_irq(lock: &current->sighand->siglock);
2362 exit_code = current->exit_code;
2363 current->last_siginfo = NULL;
2364 current->ptrace_message = 0;
2365 current->exit_code = 0;
2366
2367 /* LISTENING can be set only during STOP traps, clear it */
2368 current->jobctl &= ~(JOBCTL_LISTENING | JOBCTL_PTRACE_FROZEN);
2369
2370 /*
2371 * Queued signals ignored us while we were stopped for tracing.
2372 * So check for any that we should take before resuming user mode.
2373 * This sets TIF_SIGPENDING, but never clears it.
2374 */
2375 recalc_sigpending_tsk(current);
2376 return exit_code;
2377}
2378
2379static int ptrace_do_notify(int signr, int exit_code, int why, unsigned long message)
2380{
2381 kernel_siginfo_t info;
2382
2383 clear_siginfo(info: &info);
2384 info.si_signo = signr;
2385 info.si_code = exit_code;
2386 info.si_pid = task_pid_vnr(current);
2387 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2388
2389 /* Let the debugger run. */
2390 return ptrace_stop(exit_code, why, message, info: &info);
2391}
2392
2393int ptrace_notify(int exit_code, unsigned long message)
2394{
2395 int signr;
2396
2397 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2398 if (unlikely(task_work_pending(current)))
2399 task_work_run();
2400
2401 spin_lock_irq(lock: &current->sighand->siglock);
2402 signr = ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED, message);
2403 spin_unlock_irq(lock: &current->sighand->siglock);
2404 return signr;
2405}
2406
2407/**
2408 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2409 * @signr: signr causing group stop if initiating
2410 *
2411 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2412 * and participate in it. If already set, participate in the existing
2413 * group stop. If participated in a group stop (and thus slept), %true is
2414 * returned with siglock released.
2415 *
2416 * If ptraced, this function doesn't handle stop itself. Instead,
2417 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2418 * untouched. The caller must ensure that INTERRUPT trap handling takes
2419 * places afterwards.
2420 *
2421 * CONTEXT:
2422 * Must be called with @current->sighand->siglock held, which is released
2423 * on %true return.
2424 *
2425 * RETURNS:
2426 * %false if group stop is already cancelled or ptrace trap is scheduled.
2427 * %true if participated in group stop.
2428 */
2429static bool do_signal_stop(int signr)
2430 __releases(&current->sighand->siglock)
2431{
2432 struct signal_struct *sig = current->signal;
2433
2434 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2435 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2436 struct task_struct *t;
2437
2438 /* signr will be recorded in task->jobctl for retries */
2439 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2440
2441 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2442 unlikely(sig->flags & SIGNAL_GROUP_EXIT) ||
2443 unlikely(sig->group_exec_task))
2444 return false;
2445 /*
2446 * There is no group stop already in progress. We must
2447 * initiate one now.
2448 *
2449 * While ptraced, a task may be resumed while group stop is
2450 * still in effect and then receive a stop signal and
2451 * initiate another group stop. This deviates from the
2452 * usual behavior as two consecutive stop signals can't
2453 * cause two group stops when !ptraced. That is why we
2454 * also check !task_is_stopped(t) below.
2455 *
2456 * The condition can be distinguished by testing whether
2457 * SIGNAL_STOP_STOPPED is already set. Don't generate
2458 * group_exit_code in such case.
2459 *
2460 * This is not necessary for SIGNAL_STOP_CONTINUED because
2461 * an intervening stop signal is required to cause two
2462 * continued events regardless of ptrace.
2463 */
2464 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2465 sig->group_exit_code = signr;
2466
2467 sig->group_stop_count = 0;
2468
2469 if (task_set_jobctl_pending(current, mask: signr | gstop))
2470 sig->group_stop_count++;
2471
2472 t = current;
2473 while_each_thread(current, t) {
2474 /*
2475 * Setting state to TASK_STOPPED for a group
2476 * stop is always done with the siglock held,
2477 * so this check has no races.
2478 */
2479 if (!task_is_stopped(t) &&
2480 task_set_jobctl_pending(task: t, mask: signr | gstop)) {
2481 sig->group_stop_count++;
2482 if (likely(!(t->ptrace & PT_SEIZED)))
2483 signal_wake_up(t, fatal: 0);
2484 else
2485 ptrace_trap_notify(t);
2486 }
2487 }
2488 }
2489
2490 if (likely(!current->ptrace)) {
2491 int notify = 0;
2492
2493 /*
2494 * If there are no other threads in the group, or if there
2495 * is a group stop in progress and we are the last to stop,
2496 * report to the parent.
2497 */
2498 if (task_participate_group_stop(current))
2499 notify = CLD_STOPPED;
2500
2501 current->jobctl |= JOBCTL_STOPPED;
2502 set_special_state(TASK_STOPPED);
2503 spin_unlock_irq(lock: &current->sighand->siglock);
2504
2505 /*
2506 * Notify the parent of the group stop completion. Because
2507 * we're not holding either the siglock or tasklist_lock
2508 * here, ptracer may attach inbetween; however, this is for
2509 * group stop and should always be delivered to the real
2510 * parent of the group leader. The new ptracer will get
2511 * its notification when this task transitions into
2512 * TASK_TRACED.
2513 */
2514 if (notify) {
2515 read_lock(&tasklist_lock);
2516 do_notify_parent_cldstop(current, for_ptracer: false, why: notify);
2517 read_unlock(&tasklist_lock);
2518 }
2519
2520 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2521 cgroup_enter_frozen();
2522 schedule();
2523 return true;
2524 } else {
2525 /*
2526 * While ptraced, group stop is handled by STOP trap.
2527 * Schedule it and let the caller deal with it.
2528 */
2529 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2530 return false;
2531 }
2532}
2533
2534/**
2535 * do_jobctl_trap - take care of ptrace jobctl traps
2536 *
2537 * When PT_SEIZED, it's used for both group stop and explicit
2538 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2539 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2540 * the stop signal; otherwise, %SIGTRAP.
2541 *
2542 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2543 * number as exit_code and no siginfo.
2544 *
2545 * CONTEXT:
2546 * Must be called with @current->sighand->siglock held, which may be
2547 * released and re-acquired before returning with intervening sleep.
2548 */
2549static void do_jobctl_trap(void)
2550{
2551 struct signal_struct *signal = current->signal;
2552 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2553
2554 if (current->ptrace & PT_SEIZED) {
2555 if (!signal->group_stop_count &&
2556 !(signal->flags & SIGNAL_STOP_STOPPED))
2557 signr = SIGTRAP;
2558 WARN_ON_ONCE(!signr);
2559 ptrace_do_notify(signr, exit_code: signr | (PTRACE_EVENT_STOP << 8),
2560 CLD_STOPPED, message: 0);
2561 } else {
2562 WARN_ON_ONCE(!signr);
2563 ptrace_stop(exit_code: signr, CLD_STOPPED, message: 0, NULL);
2564 }
2565}
2566
2567/**
2568 * do_freezer_trap - handle the freezer jobctl trap
2569 *
2570 * Puts the task into frozen state, if only the task is not about to quit.
2571 * In this case it drops JOBCTL_TRAP_FREEZE.
2572 *
2573 * CONTEXT:
2574 * Must be called with @current->sighand->siglock held,
2575 * which is always released before returning.
2576 */
2577static void do_freezer_trap(void)
2578 __releases(&current->sighand->siglock)
2579{
2580 /*
2581 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2582 * let's make another loop to give it a chance to be handled.
2583 * In any case, we'll return back.
2584 */
2585 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2586 JOBCTL_TRAP_FREEZE) {
2587 spin_unlock_irq(lock: &current->sighand->siglock);
2588 return;
2589 }
2590
2591 /*
2592 * Now we're sure that there is no pending fatal signal and no
2593 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2594 * immediately (if there is a non-fatal signal pending), and
2595 * put the task into sleep.
2596 */
2597 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2598 clear_thread_flag(TIF_SIGPENDING);
2599 spin_unlock_irq(lock: &current->sighand->siglock);
2600 cgroup_enter_frozen();
2601 schedule();
2602}
2603
2604static int ptrace_signal(int signr, kernel_siginfo_t *info, enum pid_type type)
2605{
2606 /*
2607 * We do not check sig_kernel_stop(signr) but set this marker
2608 * unconditionally because we do not know whether debugger will
2609 * change signr. This flag has no meaning unless we are going
2610 * to stop after return from ptrace_stop(). In this case it will
2611 * be checked in do_signal_stop(), we should only stop if it was
2612 * not cleared by SIGCONT while we were sleeping. See also the
2613 * comment in dequeue_signal().
2614 */
2615 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2616 signr = ptrace_stop(exit_code: signr, CLD_TRAPPED, message: 0, info);
2617
2618 /* We're back. Did the debugger cancel the sig? */
2619 if (signr == 0)
2620 return signr;
2621
2622 /*
2623 * Update the siginfo structure if the signal has
2624 * changed. If the debugger wanted something
2625 * specific in the siginfo structure then it should
2626 * have updated *info via PTRACE_SETSIGINFO.
2627 */
2628 if (signr != info->si_signo) {
2629 clear_siginfo(info);
2630 info->si_signo = signr;
2631 info->si_errno = 0;
2632 info->si_code = SI_USER;
2633 rcu_read_lock();
2634 info->si_pid = task_pid_vnr(current->parent);
2635 info->si_uid = from_kuid_munged(current_user_ns(),
2636 task_uid(current->parent));
2637 rcu_read_unlock();
2638 }
2639
2640 /* If the (new) signal is now blocked, requeue it. */
2641 if (sigismember(set: &current->blocked, sig: signr) ||
2642 fatal_signal_pending(current)) {
2643 send_signal_locked(sig: signr, info, current, type);
2644 signr = 0;
2645 }
2646
2647 return signr;
2648}
2649
2650static void hide_si_addr_tag_bits(struct ksignal *ksig)
2651{
2652 switch (siginfo_layout(sig: ksig->sig, si_code: ksig->info.si_code)) {
2653 case SIL_FAULT:
2654 case SIL_FAULT_TRAPNO:
2655 case SIL_FAULT_MCEERR:
2656 case SIL_FAULT_BNDERR:
2657 case SIL_FAULT_PKUERR:
2658 case SIL_FAULT_PERF_EVENT:
2659 ksig->info.si_addr = arch_untagged_si_addr(
2660 addr: ksig->info.si_addr, sig: ksig->sig, si_code: ksig->info.si_code);
2661 break;
2662 case SIL_KILL:
2663 case SIL_TIMER:
2664 case SIL_POLL:
2665 case SIL_CHLD:
2666 case SIL_RT:
2667 case SIL_SYS:
2668 break;
2669 }
2670}
2671
2672bool get_signal(struct ksignal *ksig)
2673{
2674 struct sighand_struct *sighand = current->sighand;
2675 struct signal_struct *signal = current->signal;
2676 int signr;
2677
2678 clear_notify_signal();
2679 if (unlikely(task_work_pending(current)))
2680 task_work_run();
2681
2682 if (!task_sigpending(current))
2683 return false;
2684
2685 if (unlikely(uprobe_deny_signal()))
2686 return false;
2687
2688 /*
2689 * Do this once, we can't return to user-mode if freezing() == T.
2690 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2691 * thus do not need another check after return.
2692 */
2693 try_to_freeze();
2694
2695relock:
2696 spin_lock_irq(lock: &sighand->siglock);
2697
2698 /*
2699 * Every stopped thread goes here after wakeup. Check to see if
2700 * we should notify the parent, prepare_signal(SIGCONT) encodes
2701 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2702 */
2703 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2704 int why;
2705
2706 if (signal->flags & SIGNAL_CLD_CONTINUED)
2707 why = CLD_CONTINUED;
2708 else
2709 why = CLD_STOPPED;
2710
2711 signal->flags &= ~SIGNAL_CLD_MASK;
2712
2713 spin_unlock_irq(lock: &sighand->siglock);
2714
2715 /*
2716 * Notify the parent that we're continuing. This event is
2717 * always per-process and doesn't make whole lot of sense
2718 * for ptracers, who shouldn't consume the state via
2719 * wait(2) either, but, for backward compatibility, notify
2720 * the ptracer of the group leader too unless it's gonna be
2721 * a duplicate.
2722 */
2723 read_lock(&tasklist_lock);
2724 do_notify_parent_cldstop(current, for_ptracer: false, why);
2725
2726 if (ptrace_reparented(current->group_leader))
2727 do_notify_parent_cldstop(current->group_leader,
2728 for_ptracer: true, why);
2729 read_unlock(&tasklist_lock);
2730
2731 goto relock;
2732 }
2733
2734 for (;;) {
2735 struct k_sigaction *ka;
2736 enum pid_type type;
2737
2738 /* Has this task already been marked for death? */
2739 if ((signal->flags & SIGNAL_GROUP_EXIT) ||
2740 signal->group_exec_task) {
2741 clear_siginfo(info: &ksig->info);
2742 ksig->info.si_signo = signr = SIGKILL;
2743 sigdelset(set: &current->pending.signal, SIGKILL);
2744 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2745 ka: &sighand->action[SIGKILL - 1]);
2746 recalc_sigpending();
2747 goto fatal;
2748 }
2749
2750 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2751 do_signal_stop(signr: 0))
2752 goto relock;
2753
2754 if (unlikely(current->jobctl &
2755 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2756 if (current->jobctl & JOBCTL_TRAP_MASK) {
2757 do_jobctl_trap();
2758 spin_unlock_irq(lock: &sighand->siglock);
2759 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2760 do_freezer_trap();
2761
2762 goto relock;
2763 }
2764
2765 /*
2766 * If the task is leaving the frozen state, let's update
2767 * cgroup counters and reset the frozen bit.
2768 */
2769 if (unlikely(cgroup_task_frozen(current))) {
2770 spin_unlock_irq(lock: &sighand->siglock);
2771 cgroup_leave_frozen(always_leave: false);
2772 goto relock;
2773 }
2774
2775 /*
2776 * Signals generated by the execution of an instruction
2777 * need to be delivered before any other pending signals
2778 * so that the instruction pointer in the signal stack
2779 * frame points to the faulting instruction.
2780 */
2781 type = PIDTYPE_PID;
2782 signr = dequeue_synchronous_signal(info: &ksig->info);
2783 if (!signr)
2784 signr = dequeue_signal(current, &current->blocked,
2785 &ksig->info, &type);
2786
2787 if (!signr)
2788 break; /* will return 0 */
2789
2790 if (unlikely(current->ptrace) && (signr != SIGKILL) &&
2791 !(sighand->action[signr -1].sa.sa_flags & SA_IMMUTABLE)) {
2792 signr = ptrace_signal(signr, info: &ksig->info, type);
2793 if (!signr)
2794 continue;
2795 }
2796
2797 ka = &sighand->action[signr-1];
2798
2799 /* Trace actually delivered signals. */
2800 trace_signal_deliver(sig: signr, info: &ksig->info, ka);
2801
2802 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2803 continue;
2804 if (ka->sa.sa_handler != SIG_DFL) {
2805 /* Run the handler. */
2806 ksig->ka = *ka;
2807
2808 if (ka->sa.sa_flags & SA_ONESHOT)
2809 ka->sa.sa_handler = SIG_DFL;
2810
2811 break; /* will return non-zero "signr" value */
2812 }
2813
2814 /*
2815 * Now we are doing the default action for this signal.
2816 */
2817 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2818 continue;
2819
2820 /*
2821 * Global init gets no signals it doesn't want.
2822 * Container-init gets no signals it doesn't want from same
2823 * container.
2824 *
2825 * Note that if global/container-init sees a sig_kernel_only()
2826 * signal here, the signal must have been generated internally
2827 * or must have come from an ancestor namespace. In either
2828 * case, the signal cannot be dropped.
2829 */
2830 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2831 !sig_kernel_only(signr))
2832 continue;
2833
2834 if (sig_kernel_stop(signr)) {
2835 /*
2836 * The default action is to stop all threads in
2837 * the thread group. The job control signals
2838 * do nothing in an orphaned pgrp, but SIGSTOP
2839 * always works. Note that siglock needs to be
2840 * dropped during the call to is_orphaned_pgrp()
2841 * because of lock ordering with tasklist_lock.
2842 * This allows an intervening SIGCONT to be posted.
2843 * We need to check for that and bail out if necessary.
2844 */
2845 if (signr != SIGSTOP) {
2846 spin_unlock_irq(lock: &sighand->siglock);
2847
2848 /* signals can be posted during this window */
2849
2850 if (is_current_pgrp_orphaned())
2851 goto relock;
2852
2853 spin_lock_irq(lock: &sighand->siglock);
2854 }
2855
2856 if (likely(do_signal_stop(ksig->info.si_signo))) {
2857 /* It released the siglock. */
2858 goto relock;
2859 }
2860
2861 /*
2862 * We didn't actually stop, due to a race
2863 * with SIGCONT or something like that.
2864 */
2865 continue;
2866 }
2867
2868 fatal:
2869 spin_unlock_irq(lock: &sighand->siglock);
2870 if (unlikely(cgroup_task_frozen(current)))
2871 cgroup_leave_frozen(always_leave: true);
2872
2873 /*
2874 * Anything else is fatal, maybe with a core dump.
2875 */
2876 current->flags |= PF_SIGNALED;
2877
2878 if (sig_kernel_coredump(signr)) {
2879 if (print_fatal_signals)
2880 print_fatal_signal(signr: ksig->info.si_signo);
2881 proc_coredump_connector(current);
2882 /*
2883 * If it was able to dump core, this kills all
2884 * other threads in the group and synchronizes with
2885 * their demise. If we lost the race with another
2886 * thread getting here, it set group_exit_code
2887 * first and our do_group_exit call below will use
2888 * that value and ignore the one we pass it.
2889 */
2890 do_coredump(siginfo: &ksig->info);
2891 }
2892
2893 /*
2894 * PF_USER_WORKER threads will catch and exit on fatal signals
2895 * themselves. They have cleanup that must be performed, so
2896 * we cannot call do_exit() on their behalf.
2897 */
2898 if (current->flags & PF_USER_WORKER)
2899 goto out;
2900
2901 /*
2902 * Death signals, no core dump.
2903 */
2904 do_group_exit(ksig->info.si_signo);
2905 /* NOTREACHED */
2906 }
2907 spin_unlock_irq(lock: &sighand->siglock);
2908out:
2909 ksig->sig = signr;
2910
2911 if (!(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
2912 hide_si_addr_tag_bits(ksig);
2913
2914 return ksig->sig > 0;
2915}
2916
2917/**
2918 * signal_delivered - called after signal delivery to update blocked signals
2919 * @ksig: kernel signal struct
2920 * @stepping: nonzero if debugger single-step or block-step in use
2921 *
2922 * This function should be called when a signal has successfully been
2923 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2924 * is always blocked), and the signal itself is blocked unless %SA_NODEFER
2925 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2926 */
2927static void signal_delivered(struct ksignal *ksig, int stepping)
2928{
2929 sigset_t blocked;
2930
2931 /* A signal was successfully delivered, and the
2932 saved sigmask was stored on the signal frame,
2933 and will be restored by sigreturn. So we can
2934 simply clear the restore sigmask flag. */
2935 clear_restore_sigmask();
2936
2937 sigorsets(r: &blocked, a: &current->blocked, b: &ksig->ka.sa.sa_mask);
2938 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2939 sigaddset(set: &blocked, sig: ksig->sig);
2940 set_current_blocked(&blocked);
2941 if (current->sas_ss_flags & SS_AUTODISARM)
2942 sas_ss_reset(current);
2943 if (stepping)
2944 ptrace_notify(SIGTRAP, message: 0);
2945}
2946
2947void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2948{
2949 if (failed)
2950 force_sigsegv(sig: ksig->sig);
2951 else
2952 signal_delivered(ksig, stepping);
2953}
2954
2955/*
2956 * It could be that complete_signal() picked us to notify about the
2957 * group-wide signal. Other threads should be notified now to take
2958 * the shared signals in @which since we will not.
2959 */
2960static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2961{
2962 sigset_t retarget;
2963 struct task_struct *t;
2964
2965 sigandsets(r: &retarget, a: &tsk->signal->shared_pending.signal, b: which);
2966 if (sigisemptyset(set: &retarget))
2967 return;
2968
2969 t = tsk;
2970 while_each_thread(tsk, t) {
2971 if (t->flags & PF_EXITING)
2972 continue;
2973
2974 if (!has_pending_signals(signal: &retarget, blocked: &t->blocked))
2975 continue;
2976 /* Remove the signals this thread can handle. */
2977 sigandsets(r: &retarget, a: &retarget, b: &t->blocked);
2978
2979 if (!task_sigpending(p: t))
2980 signal_wake_up(t, fatal: 0);
2981
2982 if (sigisemptyset(set: &retarget))
2983 break;
2984 }
2985}
2986
2987void exit_signals(struct task_struct *tsk)
2988{
2989 int group_stop = 0;
2990 sigset_t unblocked;
2991
2992 /*
2993 * @tsk is about to have PF_EXITING set - lock out users which
2994 * expect stable threadgroup.
2995 */
2996 cgroup_threadgroup_change_begin(tsk);
2997
2998 if (thread_group_empty(p: tsk) || (tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
2999 sched_mm_cid_exit_signals(t: tsk);
3000 tsk->flags |= PF_EXITING;
3001 cgroup_threadgroup_change_end(tsk);
3002 return;
3003 }
3004
3005 spin_lock_irq(lock: &tsk->sighand->siglock);
3006 /*
3007 * From now this task is not visible for group-wide signals,
3008 * see wants_signal(), do_signal_stop().
3009 */
3010 sched_mm_cid_exit_signals(t: tsk);
3011 tsk->flags |= PF_EXITING;
3012
3013 cgroup_threadgroup_change_end(tsk);
3014
3015 if (!task_sigpending(p: tsk))
3016 goto out;
3017
3018 unblocked = tsk->blocked;
3019 signotset(set: &unblocked);
3020 retarget_shared_pending(tsk, which: &unblocked);
3021
3022 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
3023 task_participate_group_stop(task: tsk))
3024 group_stop = CLD_STOPPED;
3025out:
3026 spin_unlock_irq(lock: &tsk->sighand->siglock);
3027
3028 /*
3029 * If group stop has completed, deliver the notification. This
3030 * should always go to the real parent of the group leader.
3031 */
3032 if (unlikely(group_stop)) {
3033 read_lock(&tasklist_lock);
3034 do_notify_parent_cldstop(tsk, for_ptracer: false, why: group_stop);
3035 read_unlock(&tasklist_lock);
3036 }
3037}
3038
3039/*
3040 * System call entry points.
3041 */
3042
3043/**
3044 * sys_restart_syscall - restart a system call
3045 */
3046SYSCALL_DEFINE0(restart_syscall)
3047{
3048 struct restart_block *restart = &current->restart_block;
3049 return restart->fn(restart);
3050}
3051
3052long do_no_restart_syscall(struct restart_block *param)
3053{
3054 return -EINTR;
3055}
3056
3057static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
3058{
3059 if (task_sigpending(p: tsk) && !thread_group_empty(p: tsk)) {
3060 sigset_t newblocked;
3061 /* A set of now blocked but previously unblocked signals. */
3062 sigandnsets(r: &newblocked, a: newset, b: &current->blocked);
3063 retarget_shared_pending(tsk, which: &newblocked);
3064 }
3065 tsk->blocked = *newset;
3066 recalc_sigpending();
3067}
3068
3069/**
3070 * set_current_blocked - change current->blocked mask
3071 * @newset: new mask
3072 *
3073 * It is wrong to change ->blocked directly, this helper should be used
3074 * to ensure the process can't miss a shared signal we are going to block.
3075 */
3076void set_current_blocked(sigset_t *newset)
3077{
3078 sigdelsetmask(set: newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
3079 __set_current_blocked(newset);
3080}
3081
3082void __set_current_blocked(const sigset_t *newset)
3083{
3084 struct task_struct *tsk = current;
3085
3086 /*
3087 * In case the signal mask hasn't changed, there is nothing we need
3088 * to do. The current->blocked shouldn't be modified by other task.
3089 */
3090 if (sigequalsets(set1: &tsk->blocked, set2: newset))
3091 return;
3092
3093 spin_lock_irq(lock: &tsk->sighand->siglock);
3094 __set_task_blocked(tsk, newset);
3095 spin_unlock_irq(lock: &tsk->sighand->siglock);
3096}
3097
3098/*
3099 * This is also useful for kernel threads that want to temporarily
3100 * (or permanently) block certain signals.
3101 *
3102 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
3103 * interface happily blocks "unblockable" signals like SIGKILL
3104 * and friends.
3105 */
3106int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
3107{
3108 struct task_struct *tsk = current;
3109 sigset_t newset;
3110
3111 /* Lockless, only current can change ->blocked, never from irq */
3112 if (oldset)
3113 *oldset = tsk->blocked;
3114
3115 switch (how) {
3116 case SIG_BLOCK:
3117 sigorsets(r: &newset, a: &tsk->blocked, b: set);
3118 break;
3119 case SIG_UNBLOCK:
3120 sigandnsets(r: &newset, a: &tsk->blocked, b: set);
3121 break;
3122 case SIG_SETMASK:
3123 newset = *set;
3124 break;
3125 default:
3126 return -EINVAL;
3127 }
3128
3129 __set_current_blocked(newset: &newset);
3130 return 0;
3131}
3132EXPORT_SYMBOL(sigprocmask);
3133
3134/*
3135 * The api helps set app-provided sigmasks.
3136 *
3137 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3138 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3139 *
3140 * Note that it does set_restore_sigmask() in advance, so it must be always
3141 * paired with restore_saved_sigmask_unless() before return from syscall.
3142 */
3143int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3144{
3145 sigset_t kmask;
3146
3147 if (!umask)
3148 return 0;
3149 if (sigsetsize != sizeof(sigset_t))
3150 return -EINVAL;
3151 if (copy_from_user(to: &kmask, from: umask, n: sizeof(sigset_t)))
3152 return -EFAULT;
3153
3154 set_restore_sigmask();
3155 current->saved_sigmask = current->blocked;
3156 set_current_blocked(&kmask);
3157
3158 return 0;
3159}
3160
3161#ifdef CONFIG_COMPAT
3162int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3163 size_t sigsetsize)
3164{
3165 sigset_t kmask;
3166
3167 if (!umask)
3168 return 0;
3169 if (sigsetsize != sizeof(compat_sigset_t))
3170 return -EINVAL;
3171 if (get_compat_sigset(set: &kmask, compat: umask))
3172 return -EFAULT;
3173
3174 set_restore_sigmask();
3175 current->saved_sigmask = current->blocked;
3176 set_current_blocked(&kmask);
3177
3178 return 0;
3179}
3180#endif
3181
3182/**
3183 * sys_rt_sigprocmask - change the list of currently blocked signals
3184 * @how: whether to add, remove, or set signals
3185 * @nset: stores pending signals
3186 * @oset: previous value of signal mask if non-null
3187 * @sigsetsize: size of sigset_t type
3188 */
3189SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3190 sigset_t __user *, oset, size_t, sigsetsize)
3191{
3192 sigset_t old_set, new_set;
3193 int error;
3194
3195 /* XXX: Don't preclude handling different sized sigset_t's. */
3196 if (sigsetsize != sizeof(sigset_t))
3197 return -EINVAL;
3198
3199 old_set = current->blocked;
3200
3201 if (nset) {
3202 if (copy_from_user(to: &new_set, from: nset, n: sizeof(sigset_t)))
3203 return -EFAULT;
3204 sigdelsetmask(set: &new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3205
3206 error = sigprocmask(how, &new_set, NULL);
3207 if (error)
3208 return error;
3209 }
3210
3211 if (oset) {
3212 if (copy_to_user(to: oset, from: &old_set, n: sizeof(sigset_t)))
3213 return -EFAULT;
3214 }
3215
3216 return 0;
3217}
3218
3219#ifdef CONFIG_COMPAT
3220COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3221 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3222{
3223 sigset_t old_set = current->blocked;
3224
3225 /* XXX: Don't preclude handling different sized sigset_t's. */
3226 if (sigsetsize != sizeof(sigset_t))
3227 return -EINVAL;
3228
3229 if (nset) {
3230 sigset_t new_set;
3231 int error;
3232 if (get_compat_sigset(set: &new_set, compat: nset))
3233 return -EFAULT;
3234 sigdelsetmask(set: &new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3235
3236 error = sigprocmask(how, &new_set, NULL);
3237 if (error)
3238 return error;
3239 }
3240 return oset ? put_compat_sigset(compat: oset, set: &old_set, size: sizeof(*oset)) : 0;
3241}
3242#endif
3243
3244static void do_sigpending(sigset_t *set)
3245{
3246 spin_lock_irq(lock: &current->sighand->siglock);
3247 sigorsets(r: set, a: &current->pending.signal,
3248 b: &current->signal->shared_pending.signal);
3249 spin_unlock_irq(lock: &current->sighand->siglock);
3250
3251 /* Outside the lock because only this thread touches it. */
3252 sigandsets(r: set, a: &current->blocked, b: set);
3253}
3254
3255/**
3256 * sys_rt_sigpending - examine a pending signal that has been raised
3257 * while blocked
3258 * @uset: stores pending signals
3259 * @sigsetsize: size of sigset_t type or larger
3260 */
3261SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3262{
3263 sigset_t set;
3264
3265 if (sigsetsize > sizeof(*uset))
3266 return -EINVAL;
3267
3268 do_sigpending(set: &set);
3269
3270 if (copy_to_user(to: uset, from: &set, n: sigsetsize))
3271 return -EFAULT;
3272
3273 return 0;
3274}
3275
3276#ifdef CONFIG_COMPAT
3277COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3278 compat_size_t, sigsetsize)
3279{
3280 sigset_t set;
3281
3282 if (sigsetsize > sizeof(*uset))
3283 return -EINVAL;
3284
3285 do_sigpending(set: &set);
3286
3287 return put_compat_sigset(compat: uset, set: &set, size: sigsetsize);
3288}
3289#endif
3290
3291static const struct {
3292 unsigned char limit, layout;
3293} sig_sicodes[] = {
3294 [SIGILL] = { NSIGILL, SIL_FAULT },
3295 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3296 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3297 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3298 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3299#if defined(SIGEMT)
3300 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3301#endif
3302 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3303 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3304 [SIGSYS] = { NSIGSYS, SIL_SYS },
3305};
3306
3307static bool known_siginfo_layout(unsigned sig, int si_code)
3308{
3309 if (si_code == SI_KERNEL)
3310 return true;
3311 else if ((si_code > SI_USER)) {
3312 if (sig_specific_sicodes(sig)) {
3313 if (si_code <= sig_sicodes[sig].limit)
3314 return true;
3315 }
3316 else if (si_code <= NSIGPOLL)
3317 return true;
3318 }
3319 else if (si_code >= SI_DETHREAD)
3320 return true;
3321 else if (si_code == SI_ASYNCNL)
3322 return true;
3323 return false;
3324}
3325
3326enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3327{
3328 enum siginfo_layout layout = SIL_KILL;
3329 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3330 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3331 (si_code <= sig_sicodes[sig].limit)) {
3332 layout = sig_sicodes[sig].layout;
3333 /* Handle the exceptions */
3334 if ((sig == SIGBUS) &&
3335 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3336 layout = SIL_FAULT_MCEERR;
3337 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3338 layout = SIL_FAULT_BNDERR;
3339#ifdef SEGV_PKUERR
3340 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3341 layout = SIL_FAULT_PKUERR;
3342#endif
3343 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3344 layout = SIL_FAULT_PERF_EVENT;
3345 else if (IS_ENABLED(CONFIG_SPARC) &&
3346 (sig == SIGILL) && (si_code == ILL_ILLTRP))
3347 layout = SIL_FAULT_TRAPNO;
3348 else if (IS_ENABLED(CONFIG_ALPHA) &&
3349 ((sig == SIGFPE) ||
3350 ((sig == SIGTRAP) && (si_code == TRAP_UNK))))
3351 layout = SIL_FAULT_TRAPNO;
3352 }
3353 else if (si_code <= NSIGPOLL)
3354 layout = SIL_POLL;
3355 } else {
3356 if (si_code == SI_TIMER)
3357 layout = SIL_TIMER;
3358 else if (si_code == SI_SIGIO)
3359 layout = SIL_POLL;
3360 else if (si_code < 0)
3361 layout = SIL_RT;
3362 }
3363 return layout;
3364}
3365
3366static inline char __user *si_expansion(const siginfo_t __user *info)
3367{
3368 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3369}
3370
3371int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3372{
3373 char __user *expansion = si_expansion(info: to);
3374 if (copy_to_user(to, from , n: sizeof(struct kernel_siginfo)))
3375 return -EFAULT;
3376 if (clear_user(to: expansion, SI_EXPANSION_SIZE))
3377 return -EFAULT;
3378 return 0;
3379}
3380
3381static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3382 const siginfo_t __user *from)
3383{
3384 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3385 char __user *expansion = si_expansion(info: from);
3386 char buf[SI_EXPANSION_SIZE];
3387 int i;
3388 /*
3389 * An unknown si_code might need more than
3390 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3391 * extra bytes are 0. This guarantees copy_siginfo_to_user
3392 * will return this data to userspace exactly.
3393 */
3394 if (copy_from_user(to: &buf, from: expansion, SI_EXPANSION_SIZE))
3395 return -EFAULT;
3396 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3397 if (buf[i] != 0)
3398 return -E2BIG;
3399 }
3400 }
3401 return 0;
3402}
3403
3404static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3405 const siginfo_t __user *from)
3406{
3407 if (copy_from_user(to, from, n: sizeof(struct kernel_siginfo)))
3408 return -EFAULT;
3409 to->si_signo = signo;
3410 return post_copy_siginfo_from_user(info: to, from);
3411}
3412
3413int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3414{
3415 if (copy_from_user(to, from, n: sizeof(struct kernel_siginfo)))
3416 return -EFAULT;
3417 return post_copy_siginfo_from_user(info: to, from);
3418}
3419
3420#ifdef CONFIG_COMPAT
3421/**
3422 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3423 * @to: compat siginfo destination
3424 * @from: kernel siginfo source
3425 *
3426 * Note: This function does not work properly for the SIGCHLD on x32, but
3427 * fortunately it doesn't have to. The only valid callers for this function are
3428 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3429 * The latter does not care because SIGCHLD will never cause a coredump.
3430 */
3431void copy_siginfo_to_external32(struct compat_siginfo *to,
3432 const struct kernel_siginfo *from)
3433{
3434 memset(to, 0, sizeof(*to));
3435
3436 to->si_signo = from->si_signo;
3437 to->si_errno = from->si_errno;
3438 to->si_code = from->si_code;
3439 switch(siginfo_layout(sig: from->si_signo, si_code: from->si_code)) {
3440 case SIL_KILL:
3441 to->si_pid = from->si_pid;
3442 to->si_uid = from->si_uid;
3443 break;
3444 case SIL_TIMER:
3445 to->si_tid = from->si_tid;
3446 to->si_overrun = from->si_overrun;
3447 to->si_int = from->si_int;
3448 break;
3449 case SIL_POLL:
3450 to->si_band = from->si_band;
3451 to->si_fd = from->si_fd;
3452 break;
3453 case SIL_FAULT:
3454 to->si_addr = ptr_to_compat(uptr: from->si_addr);
3455 break;
3456 case SIL_FAULT_TRAPNO:
3457 to->si_addr = ptr_to_compat(uptr: from->si_addr);
3458 to->si_trapno = from->si_trapno;
3459 break;
3460 case SIL_FAULT_MCEERR:
3461 to->si_addr = ptr_to_compat(uptr: from->si_addr);
3462 to->si_addr_lsb = from->si_addr_lsb;
3463 break;
3464 case SIL_FAULT_BNDERR:
3465 to->si_addr = ptr_to_compat(uptr: from->si_addr);
3466 to->si_lower = ptr_to_compat(uptr: from->si_lower);
3467 to->si_upper = ptr_to_compat(uptr: from->si_upper);
3468 break;
3469 case SIL_FAULT_PKUERR:
3470 to->si_addr = ptr_to_compat(uptr: from->si_addr);
3471 to->si_pkey = from->si_pkey;
3472 break;
3473 case SIL_FAULT_PERF_EVENT:
3474 to->si_addr = ptr_to_compat(uptr: from->si_addr);
3475 to->si_perf_data = from->si_perf_data;
3476 to->si_perf_type = from->si_perf_type;
3477 to->si_perf_flags = from->si_perf_flags;
3478 break;
3479 case SIL_CHLD:
3480 to->si_pid = from->si_pid;
3481 to->si_uid = from->si_uid;
3482 to->si_status = from->si_status;
3483 to->si_utime = from->si_utime;
3484 to->si_stime = from->si_stime;
3485 break;
3486 case SIL_RT:
3487 to->si_pid = from->si_pid;
3488 to->si_uid = from->si_uid;
3489 to->si_int = from->si_int;
3490 break;
3491 case SIL_SYS:
3492 to->si_call_addr = ptr_to_compat(uptr: from->si_call_addr);
3493 to->si_syscall = from->si_syscall;
3494 to->si_arch = from->si_arch;
3495 break;
3496 }
3497}
3498
3499int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3500 const struct kernel_siginfo *from)
3501{
3502 struct compat_siginfo new;
3503
3504 copy_siginfo_to_external32(to: &new, from);
3505 if (copy_to_user(to, from: &new, n: sizeof(struct compat_siginfo)))
3506 return -EFAULT;
3507 return 0;
3508}
3509
3510static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3511 const struct compat_siginfo *from)
3512{
3513 clear_siginfo(info: to);
3514 to->si_signo = from->si_signo;
3515 to->si_errno = from->si_errno;
3516 to->si_code = from->si_code;
3517 switch(siginfo_layout(sig: from->si_signo, si_code: from->si_code)) {
3518 case SIL_KILL:
3519 to->si_pid = from->si_pid;
3520 to->si_uid = from->si_uid;
3521 break;
3522 case SIL_TIMER:
3523 to->si_tid = from->si_tid;
3524 to->si_overrun = from->si_overrun;
3525 to->si_int = from->si_int;
3526 break;
3527 case SIL_POLL:
3528 to->si_band = from->si_band;
3529 to->si_fd = from->si_fd;
3530 break;
3531 case SIL_FAULT:
3532 to->si_addr = compat_ptr(uptr: from->si_addr);
3533 break;
3534 case SIL_FAULT_TRAPNO:
3535 to->si_addr = compat_ptr(uptr: from->si_addr);
3536 to->si_trapno = from->si_trapno;
3537 break;
3538 case SIL_FAULT_MCEERR:
3539 to->si_addr = compat_ptr(uptr: from->si_addr);
3540 to->si_addr_lsb = from->si_addr_lsb;
3541 break;
3542 case SIL_FAULT_BNDERR:
3543 to->si_addr = compat_ptr(uptr: from->si_addr);
3544 to->si_lower = compat_ptr(uptr: from->si_lower);
3545 to->si_upper = compat_ptr(uptr: from->si_upper);
3546 break;
3547 case SIL_FAULT_PKUERR:
3548 to->si_addr = compat_ptr(uptr: from->si_addr);
3549 to->si_pkey = from->si_pkey;
3550 break;
3551 case SIL_FAULT_PERF_EVENT:
3552 to->si_addr = compat_ptr(uptr: from->si_addr);
3553 to->si_perf_data = from->si_perf_data;
3554 to->si_perf_type = from->si_perf_type;
3555 to->si_perf_flags = from->si_perf_flags;
3556 break;
3557 case SIL_CHLD:
3558 to->si_pid = from->si_pid;
3559 to->si_uid = from->si_uid;
3560 to->si_status = from->si_status;
3561#ifdef CONFIG_X86_X32_ABI
3562 if (in_x32_syscall()) {
3563 to->si_utime = from->_sifields._sigchld_x32._utime;
3564 to->si_stime = from->_sifields._sigchld_x32._stime;
3565 } else
3566#endif
3567 {
3568 to->si_utime = from->si_utime;
3569 to->si_stime = from->si_stime;
3570 }
3571 break;
3572 case SIL_RT:
3573 to->si_pid = from->si_pid;
3574 to->si_uid = from->si_uid;
3575 to->si_int = from->si_int;
3576 break;
3577 case SIL_SYS:
3578 to->si_call_addr = compat_ptr(uptr: from->si_call_addr);
3579 to->si_syscall = from->si_syscall;
3580 to->si_arch = from->si_arch;
3581 break;
3582 }
3583 return 0;
3584}
3585
3586static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3587 const struct compat_siginfo __user *ufrom)
3588{
3589 struct compat_siginfo from;
3590
3591 if (copy_from_user(to: &from, from: ufrom, n: sizeof(struct compat_siginfo)))
3592 return -EFAULT;
3593
3594 from.si_signo = signo;
3595 return post_copy_siginfo_from_user32(to, from: &from);
3596}
3597
3598int copy_siginfo_from_user32(struct kernel_siginfo *to,
3599 const struct compat_siginfo __user *ufrom)
3600{
3601 struct compat_siginfo from;
3602
3603 if (copy_from_user(to: &from, from: ufrom, n: sizeof(struct compat_siginfo)))
3604 return -EFAULT;
3605
3606 return post_copy_siginfo_from_user32(to, from: &from);
3607}
3608#endif /* CONFIG_COMPAT */
3609
3610/**
3611 * do_sigtimedwait - wait for queued signals specified in @which
3612 * @which: queued signals to wait for
3613 * @info: if non-null, the signal's siginfo is returned here
3614 * @ts: upper bound on process time suspension
3615 */
3616static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3617 const struct timespec64 *ts)
3618{
3619 ktime_t *to = NULL, timeout = KTIME_MAX;
3620 struct task_struct *tsk = current;
3621 sigset_t mask = *which;
3622 enum pid_type type;
3623 int sig, ret = 0;
3624
3625 if (ts) {
3626 if (!timespec64_valid(ts))
3627 return -EINVAL;
3628 timeout = timespec64_to_ktime(ts: *ts);
3629 to = &timeout;
3630 }
3631
3632 /*
3633 * Invert the set of allowed signals to get those we want to block.
3634 */
3635 sigdelsetmask(set: &mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3636 signotset(set: &mask);
3637
3638 spin_lock_irq(lock: &tsk->sighand->siglock);
3639 sig = dequeue_signal(tsk, &mask, info, &type);
3640 if (!sig && timeout) {
3641 /*
3642 * None ready, temporarily unblock those we're interested
3643 * while we are sleeping in so that we'll be awakened when
3644 * they arrive. Unblocking is always fine, we can avoid
3645 * set_current_blocked().
3646 */
3647 tsk->real_blocked = tsk->blocked;
3648 sigandsets(r: &tsk->blocked, a: &tsk->blocked, b: &mask);
3649 recalc_sigpending();
3650 spin_unlock_irq(lock: &tsk->sighand->siglock);
3651
3652 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
3653 ret = schedule_hrtimeout_range(expires: to, delta: tsk->timer_slack_ns,
3654 mode: HRTIMER_MODE_REL);
3655 spin_lock_irq(lock: &tsk->sighand->siglock);
3656 __set_task_blocked(tsk, newset: &tsk->real_blocked);
3657 sigemptyset(set: &tsk->real_blocked);
3658 sig = dequeue_signal(tsk, &mask, info, &type);
3659 }
3660 spin_unlock_irq(lock: &tsk->sighand->siglock);
3661
3662 if (sig)
3663 return sig;
3664 return ret ? -EINTR : -EAGAIN;
3665}
3666
3667/**
3668 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3669 * in @uthese
3670 * @uthese: queued signals to wait for
3671 * @uinfo: if non-null, the signal's siginfo is returned here
3672 * @uts: upper bound on process time suspension
3673 * @sigsetsize: size of sigset_t type
3674 */
3675SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3676 siginfo_t __user *, uinfo,
3677 const struct __kernel_timespec __user *, uts,
3678 size_t, sigsetsize)
3679{
3680 sigset_t these;
3681 struct timespec64 ts;
3682 kernel_siginfo_t info;
3683 int ret;
3684
3685 /* XXX: Don't preclude handling different sized sigset_t's. */
3686 if (sigsetsize != sizeof(sigset_t))
3687 return -EINVAL;
3688
3689 if (copy_from_user(to: &these, from: uthese, n: sizeof(these)))
3690 return -EFAULT;
3691
3692 if (uts) {
3693 if (get_timespec64(ts: &ts, uts))
3694 return -EFAULT;
3695 }
3696
3697 ret = do_sigtimedwait(which: &these, info: &info, ts: uts ? &ts : NULL);
3698
3699 if (ret > 0 && uinfo) {
3700 if (copy_siginfo_to_user(to: uinfo, from: &info))
3701 ret = -EFAULT;
3702 }
3703
3704 return ret;
3705}
3706
3707#ifdef CONFIG_COMPAT_32BIT_TIME
3708SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3709 siginfo_t __user *, uinfo,
3710 const struct old_timespec32 __user *, uts,
3711 size_t, sigsetsize)
3712{
3713 sigset_t these;
3714 struct timespec64 ts;
3715 kernel_siginfo_t info;
3716 int ret;
3717
3718 if (sigsetsize != sizeof(sigset_t))
3719 return -EINVAL;
3720
3721 if (copy_from_user(to: &these, from: uthese, n: sizeof(these)))
3722 return -EFAULT;
3723
3724 if (uts) {
3725 if (get_old_timespec32(&ts, uts))
3726 return -EFAULT;
3727 }
3728
3729 ret = do_sigtimedwait(which: &these, info: &info, ts: uts ? &ts : NULL);
3730
3731 if (ret > 0 && uinfo) {
3732 if (copy_siginfo_to_user(to: uinfo, from: &info))
3733 ret = -EFAULT;
3734 }
3735
3736 return ret;
3737}
3738#endif
3739
3740#ifdef CONFIG_COMPAT
3741COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3742 struct compat_siginfo __user *, uinfo,
3743 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3744{
3745 sigset_t s;
3746 struct timespec64 t;
3747 kernel_siginfo_t info;
3748 long ret;
3749
3750 if (sigsetsize != sizeof(sigset_t))
3751 return -EINVAL;
3752
3753 if (get_compat_sigset(set: &s, compat: uthese))
3754 return -EFAULT;
3755
3756 if (uts) {
3757 if (get_timespec64(ts: &t, uts))
3758 return -EFAULT;
3759 }
3760
3761 ret = do_sigtimedwait(which: &s, info: &info, ts: uts ? &t : NULL);
3762
3763 if (ret > 0 && uinfo) {
3764 if (copy_siginfo_to_user32(to: uinfo, from: &info))
3765 ret = -EFAULT;
3766 }
3767
3768 return ret;
3769}
3770
3771#ifdef CONFIG_COMPAT_32BIT_TIME
3772COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3773 struct compat_siginfo __user *, uinfo,
3774 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3775{
3776 sigset_t s;
3777 struct timespec64 t;
3778 kernel_siginfo_t info;
3779 long ret;
3780
3781 if (sigsetsize != sizeof(sigset_t))
3782 return -EINVAL;
3783
3784 if (get_compat_sigset(set: &s, compat: uthese))
3785 return -EFAULT;
3786
3787 if (uts) {
3788 if (get_old_timespec32(&t, uts))
3789 return -EFAULT;
3790 }
3791
3792 ret = do_sigtimedwait(which: &s, info: &info, ts: uts ? &t : NULL);
3793
3794 if (ret > 0 && uinfo) {
3795 if (copy_siginfo_to_user32(to: uinfo, from: &info))
3796 ret = -EFAULT;
3797 }
3798
3799 return ret;
3800}
3801#endif
3802#endif
3803
3804static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3805{
3806 clear_siginfo(info);
3807 info->si_signo = sig;
3808 info->si_errno = 0;
3809 info->si_code = SI_USER;
3810 info->si_pid = task_tgid_vnr(current);
3811 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3812}
3813
3814/**
3815 * sys_kill - send a signal to a process
3816 * @pid: the PID of the process
3817 * @sig: signal to be sent
3818 */
3819SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3820{
3821 struct kernel_siginfo info;
3822
3823 prepare_kill_siginfo(sig, info: &info);
3824
3825 return kill_something_info(sig, info: &info, pid);
3826}
3827
3828/*
3829 * Verify that the signaler and signalee either are in the same pid namespace
3830 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3831 * namespace.
3832 */
3833static bool access_pidfd_pidns(struct pid *pid)
3834{
3835 struct pid_namespace *active = task_active_pid_ns(current);
3836 struct pid_namespace *p = ns_of_pid(pid);
3837
3838 for (;;) {
3839 if (!p)
3840 return false;
3841 if (p == active)
3842 break;
3843 p = p->parent;
3844 }
3845
3846 return true;
3847}
3848
3849static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3850 siginfo_t __user *info)
3851{
3852#ifdef CONFIG_COMPAT
3853 /*
3854 * Avoid hooking up compat syscalls and instead handle necessary
3855 * conversions here. Note, this is a stop-gap measure and should not be
3856 * considered a generic solution.
3857 */
3858 if (in_compat_syscall())
3859 return copy_siginfo_from_user32(
3860 to: kinfo, ufrom: (struct compat_siginfo __user *)info);
3861#endif
3862 return copy_siginfo_from_user(to: kinfo, from: info);
3863}
3864
3865static struct pid *pidfd_to_pid(const struct file *file)
3866{
3867 struct pid *pid;
3868
3869 pid = pidfd_pid(file);
3870 if (!IS_ERR(ptr: pid))
3871 return pid;
3872
3873 return tgid_pidfd_to_pid(file);
3874}
3875
3876/**
3877 * sys_pidfd_send_signal - Signal a process through a pidfd
3878 * @pidfd: file descriptor of the process
3879 * @sig: signal to send
3880 * @info: signal info
3881 * @flags: future flags
3882 *
3883 * The syscall currently only signals via PIDTYPE_PID which covers
3884 * kill(<positive-pid>, <signal>. It does not signal threads or process
3885 * groups.
3886 * In order to extend the syscall to threads and process groups the @flags
3887 * argument should be used. In essence, the @flags argument will determine
3888 * what is signaled and not the file descriptor itself. Put in other words,
3889 * grouping is a property of the flags argument not a property of the file
3890 * descriptor.
3891 *
3892 * Return: 0 on success, negative errno on failure
3893 */
3894SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3895 siginfo_t __user *, info, unsigned int, flags)
3896{
3897 int ret;
3898 struct fd f;
3899 struct pid *pid;
3900 kernel_siginfo_t kinfo;
3901
3902 /* Enforce flags be set to 0 until we add an extension. */
3903 if (flags)
3904 return -EINVAL;
3905
3906 f = fdget(fd: pidfd);
3907 if (!f.file)
3908 return -EBADF;
3909
3910 /* Is this a pidfd? */
3911 pid = pidfd_to_pid(file: f.file);
3912 if (IS_ERR(ptr: pid)) {
3913 ret = PTR_ERR(ptr: pid);
3914 goto err;
3915 }
3916
3917 ret = -EINVAL;
3918 if (!access_pidfd_pidns(pid))
3919 goto err;
3920
3921 if (info) {
3922 ret = copy_siginfo_from_user_any(kinfo: &kinfo, info);
3923 if (unlikely(ret))
3924 goto err;
3925
3926 ret = -EINVAL;
3927 if (unlikely(sig != kinfo.si_signo))
3928 goto err;
3929
3930 /* Only allow sending arbitrary signals to yourself. */
3931 ret = -EPERM;
3932 if ((task_pid(current) != pid) &&
3933 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3934 goto err;
3935 } else {
3936 prepare_kill_siginfo(sig, info: &kinfo);
3937 }
3938
3939 ret = kill_pid_info(sig, info: &kinfo, pid);
3940
3941err:
3942 fdput(fd: f);
3943 return ret;
3944}
3945
3946static int
3947do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3948{
3949 struct task_struct *p;
3950 int error = -ESRCH;
3951
3952 rcu_read_lock();
3953 p = find_task_by_vpid(nr: pid);
3954 if (p && (tgid <= 0 || task_tgid_vnr(tsk: p) == tgid)) {
3955 error = check_kill_permission(sig, info, t: p);
3956 /*
3957 * The null signal is a permissions and process existence
3958 * probe. No signal is actually delivered.
3959 */
3960 if (!error && sig) {
3961 error = do_send_sig_info(sig, info, p, type: PIDTYPE_PID);
3962 /*
3963 * If lock_task_sighand() failed we pretend the task
3964 * dies after receiving the signal. The window is tiny,
3965 * and the signal is private anyway.
3966 */
3967 if (unlikely(error == -ESRCH))
3968 error = 0;
3969 }
3970 }
3971 rcu_read_unlock();
3972
3973 return error;
3974}
3975
3976static int do_tkill(pid_t tgid, pid_t pid, int sig)
3977{
3978 struct kernel_siginfo info;
3979
3980 clear_siginfo(info: &info);
3981 info.si_signo = sig;
3982 info.si_errno = 0;
3983 info.si_code = SI_TKILL;
3984 info.si_pid = task_tgid_vnr(current);
3985 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3986
3987 return do_send_specific(tgid, pid, sig, info: &info);
3988}
3989
3990/**
3991 * sys_tgkill - send signal to one specific thread
3992 * @tgid: the thread group ID of the thread
3993 * @pid: the PID of the thread
3994 * @sig: signal to be sent
3995 *
3996 * This syscall also checks the @tgid and returns -ESRCH even if the PID
3997 * exists but it's not belonging to the target process anymore. This
3998 * method solves the problem of threads exiting and PIDs getting reused.
3999 */
4000SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
4001{
4002 /* This is only valid for single tasks */
4003 if (pid <= 0 || tgid <= 0)
4004 return -EINVAL;
4005
4006 return do_tkill(tgid, pid, sig);
4007}
4008
4009/**
4010 * sys_tkill - send signal to one specific task
4011 * @pid: the PID of the task
4012 * @sig: signal to be sent
4013 *
4014 * Send a signal to only one task, even if it's a CLONE_THREAD task.
4015 */
4016SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
4017{
4018 /* This is only valid for single tasks */
4019 if (pid <= 0)
4020 return -EINVAL;
4021
4022 return do_tkill(tgid: 0, pid, sig);
4023}
4024
4025static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
4026{
4027 /* Not even root can pretend to send signals from the kernel.
4028 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4029 */
4030 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4031 (task_pid_vnr(current) != pid))
4032 return -EPERM;
4033
4034 /* POSIX.1b doesn't mention process groups. */
4035 return kill_proc_info(sig, info, pid);
4036}
4037
4038/**
4039 * sys_rt_sigqueueinfo - send signal information to a signal
4040 * @pid: the PID of the thread
4041 * @sig: signal to be sent
4042 * @uinfo: signal info to be sent
4043 */
4044SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
4045 siginfo_t __user *, uinfo)
4046{
4047 kernel_siginfo_t info;
4048 int ret = __copy_siginfo_from_user(signo: sig, to: &info, from: uinfo);
4049 if (unlikely(ret))
4050 return ret;
4051 return do_rt_sigqueueinfo(pid, sig, info: &info);
4052}
4053
4054#ifdef CONFIG_COMPAT
4055COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
4056 compat_pid_t, pid,
4057 int, sig,
4058 struct compat_siginfo __user *, uinfo)
4059{
4060 kernel_siginfo_t info;
4061 int ret = __copy_siginfo_from_user32(signo: sig, to: &info, ufrom: uinfo);
4062 if (unlikely(ret))
4063 return ret;
4064 return do_rt_sigqueueinfo(pid, sig, info: &info);
4065}
4066#endif
4067
4068static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
4069{
4070 /* This is only valid for single tasks */
4071 if (pid <= 0 || tgid <= 0)
4072 return -EINVAL;
4073
4074 /* Not even root can pretend to send signals from the kernel.
4075 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4076 */
4077 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4078 (task_pid_vnr(current) != pid))
4079 return -EPERM;
4080
4081 return do_send_specific(tgid, pid, sig, info);
4082}
4083
4084SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
4085 siginfo_t __user *, uinfo)
4086{
4087 kernel_siginfo_t info;
4088 int ret = __copy_siginfo_from_user(signo: sig, to: &info, from: uinfo);
4089 if (unlikely(ret))
4090 return ret;
4091 return do_rt_tgsigqueueinfo(tgid, pid, sig, info: &info);
4092}
4093
4094#ifdef CONFIG_COMPAT
4095COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
4096 compat_pid_t, tgid,
4097 compat_pid_t, pid,
4098 int, sig,
4099 struct compat_siginfo __user *, uinfo)
4100{
4101 kernel_siginfo_t info;
4102 int ret = __copy_siginfo_from_user32(signo: sig, to: &info, ufrom: uinfo);
4103 if (unlikely(ret))
4104 return ret;
4105 return do_rt_tgsigqueueinfo(tgid, pid, sig, info: &info);
4106}
4107#endif
4108
4109/*
4110 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
4111 */
4112void kernel_sigaction(int sig, __sighandler_t action)
4113{
4114 spin_lock_irq(lock: &current->sighand->siglock);
4115 current->sighand->action[sig - 1].sa.sa_handler = action;
4116 if (action == SIG_IGN) {
4117 sigset_t mask;
4118
4119 sigemptyset(set: &mask);
4120 sigaddset(set: &mask, sig: sig);
4121
4122 flush_sigqueue_mask(mask: &mask, s: &current->signal->shared_pending);
4123 flush_sigqueue_mask(mask: &mask, s: &current->pending);
4124 recalc_sigpending();
4125 }
4126 spin_unlock_irq(lock: &current->sighand->siglock);
4127}
4128EXPORT_SYMBOL(kernel_sigaction);
4129
4130void __weak sigaction_compat_abi(struct k_sigaction *act,
4131 struct k_sigaction *oact)
4132{
4133}
4134
4135int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4136{
4137 struct task_struct *p = current, *t;
4138 struct k_sigaction *k;
4139 sigset_t mask;
4140
4141 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4142 return -EINVAL;
4143
4144 k = &p->sighand->action[sig-1];
4145
4146 spin_lock_irq(lock: &p->sighand->siglock);
4147 if (k->sa.sa_flags & SA_IMMUTABLE) {
4148 spin_unlock_irq(lock: &p->sighand->siglock);
4149 return -EINVAL;
4150 }
4151 if (oact)
4152 *oact = *k;
4153
4154 /*
4155 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4156 * e.g. by having an architecture use the bit in their uapi.
4157 */
4158 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4159
4160 /*
4161 * Clear unknown flag bits in order to allow userspace to detect missing
4162 * support for flag bits and to allow the kernel to use non-uapi bits
4163 * internally.
4164 */
4165 if (act)
4166 act->sa.sa_flags &= UAPI_SA_FLAGS;
4167 if (oact)
4168 oact->sa.sa_flags &= UAPI_SA_FLAGS;
4169
4170 sigaction_compat_abi(act, oact);
4171
4172 if (act) {
4173 sigdelsetmask(set: &act->sa.sa_mask,
4174 sigmask(SIGKILL) | sigmask(SIGSTOP));
4175 *k = *act;
4176 /*
4177 * POSIX 3.3.1.3:
4178 * "Setting a signal action to SIG_IGN for a signal that is
4179 * pending shall cause the pending signal to be discarded,
4180 * whether or not it is blocked."
4181 *
4182 * "Setting a signal action to SIG_DFL for a signal that is
4183 * pending and whose default action is to ignore the signal
4184 * (for example, SIGCHLD), shall cause the pending signal to
4185 * be discarded, whether or not it is blocked"
4186 */
4187 if (sig_handler_ignored(handler: sig_handler(t: p, sig), sig)) {
4188 sigemptyset(set: &mask);
4189 sigaddset(set: &mask, sig: sig);
4190 flush_sigqueue_mask(mask: &mask, s: &p->signal->shared_pending);
4191 for_each_thread(p, t)
4192 flush_sigqueue_mask(mask: &mask, s: &t->pending);
4193 }
4194 }
4195
4196 spin_unlock_irq(lock: &p->sighand->siglock);
4197 return 0;
4198}
4199
4200#ifdef CONFIG_DYNAMIC_SIGFRAME
4201static inline void sigaltstack_lock(void)
4202 __acquires(&current->sighand->siglock)
4203{
4204 spin_lock_irq(lock: &current->sighand->siglock);
4205}
4206
4207static inline void sigaltstack_unlock(void)
4208 __releases(&current->sighand->siglock)
4209{
4210 spin_unlock_irq(lock: &current->sighand->siglock);
4211}
4212#else
4213static inline void sigaltstack_lock(void) { }
4214static inline void sigaltstack_unlock(void) { }
4215#endif
4216
4217static int
4218do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4219 size_t min_ss_size)
4220{
4221 struct task_struct *t = current;
4222 int ret = 0;
4223
4224 if (oss) {
4225 memset(oss, 0, sizeof(stack_t));
4226 oss->ss_sp = (void __user *) t->sas_ss_sp;
4227 oss->ss_size = t->sas_ss_size;
4228 oss->ss_flags = sas_ss_flags(sp) |
4229 (current->sas_ss_flags & SS_FLAG_BITS);
4230 }
4231
4232 if (ss) {
4233 void __user *ss_sp = ss->ss_sp;
4234 size_t ss_size = ss->ss_size;
4235 unsigned ss_flags = ss->ss_flags;
4236 int ss_mode;
4237
4238 if (unlikely(on_sig_stack(sp)))
4239 return -EPERM;
4240
4241 ss_mode = ss_flags & ~SS_FLAG_BITS;
4242 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4243 ss_mode != 0))
4244 return -EINVAL;
4245
4246 /*
4247 * Return before taking any locks if no actual
4248 * sigaltstack changes were requested.
4249 */
4250 if (t->sas_ss_sp == (unsigned long)ss_sp &&
4251 t->sas_ss_size == ss_size &&
4252 t->sas_ss_flags == ss_flags)
4253 return 0;
4254
4255 sigaltstack_lock();
4256 if (ss_mode == SS_DISABLE) {
4257 ss_size = 0;
4258 ss_sp = NULL;
4259 } else {
4260 if (unlikely(ss_size < min_ss_size))
4261 ret = -ENOMEM;
4262 if (!sigaltstack_size_valid(ss_size))
4263 ret = -ENOMEM;
4264 }
4265 if (!ret) {
4266 t->sas_ss_sp = (unsigned long) ss_sp;
4267 t->sas_ss_size = ss_size;
4268 t->sas_ss_flags = ss_flags;
4269 }
4270 sigaltstack_unlock();
4271 }
4272 return ret;
4273}
4274
4275SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4276{
4277 stack_t new, old;
4278 int err;
4279 if (uss && copy_from_user(to: &new, from: uss, n: sizeof(stack_t)))
4280 return -EFAULT;
4281 err = do_sigaltstack(ss: uss ? &new : NULL, oss: uoss ? &old : NULL,
4282 current_user_stack_pointer(),
4283 MINSIGSTKSZ);
4284 if (!err && uoss && copy_to_user(to: uoss, from: &old, n: sizeof(stack_t)))
4285 err = -EFAULT;
4286 return err;
4287}
4288
4289int restore_altstack(const stack_t __user *uss)
4290{
4291 stack_t new;
4292 if (copy_from_user(to: &new, from: uss, n: sizeof(stack_t)))
4293 return -EFAULT;
4294 (void)do_sigaltstack(ss: &new, NULL, current_user_stack_pointer(),
4295 MINSIGSTKSZ);
4296 /* squash all but EFAULT for now */
4297 return 0;
4298}
4299
4300int __save_altstack(stack_t __user *uss, unsigned long sp)
4301{
4302 struct task_struct *t = current;
4303 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4304 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4305 __put_user(t->sas_ss_size, &uss->ss_size);
4306 return err;
4307}
4308
4309#ifdef CONFIG_COMPAT
4310static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4311 compat_stack_t __user *uoss_ptr)
4312{
4313 stack_t uss, uoss;
4314 int ret;
4315
4316 if (uss_ptr) {
4317 compat_stack_t uss32;
4318 if (copy_from_user(to: &uss32, from: uss_ptr, n: sizeof(compat_stack_t)))
4319 return -EFAULT;
4320 uss.ss_sp = compat_ptr(uptr: uss32.ss_sp);
4321 uss.ss_flags = uss32.ss_flags;
4322 uss.ss_size = uss32.ss_size;
4323 }
4324 ret = do_sigaltstack(ss: uss_ptr ? &uss : NULL, oss: &uoss,
4325 compat_user_stack_pointer(),
4326 COMPAT_MINSIGSTKSZ);
4327 if (ret >= 0 && uoss_ptr) {
4328 compat_stack_t old;
4329 memset(&old, 0, sizeof(old));
4330 old.ss_sp = ptr_to_compat(uptr: uoss.ss_sp);
4331 old.ss_flags = uoss.ss_flags;
4332 old.ss_size = uoss.ss_size;
4333 if (copy_to_user(to: uoss_ptr, from: &old, n: sizeof(compat_stack_t)))
4334 ret = -EFAULT;
4335 }
4336 return ret;
4337}
4338
4339COMPAT_SYSCALL_DEFINE2(sigaltstack,
4340 const compat_stack_t __user *, uss_ptr,
4341 compat_stack_t __user *, uoss_ptr)
4342{
4343 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4344}
4345
4346int compat_restore_altstack(const compat_stack_t __user *uss)
4347{
4348 int err = do_compat_sigaltstack(uss_ptr: uss, NULL);
4349 /* squash all but -EFAULT for now */
4350 return err == -EFAULT ? err : 0;
4351}
4352
4353int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4354{
4355 int err;
4356 struct task_struct *t = current;
4357 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4358 &uss->ss_sp) |
4359 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4360 __put_user(t->sas_ss_size, &uss->ss_size);
4361 return err;
4362}
4363#endif
4364
4365#ifdef __ARCH_WANT_SYS_SIGPENDING
4366
4367/**
4368 * sys_sigpending - examine pending signals
4369 * @uset: where mask of pending signal is returned
4370 */
4371SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4372{
4373 sigset_t set;
4374
4375 if (sizeof(old_sigset_t) > sizeof(*uset))
4376 return -EINVAL;
4377
4378 do_sigpending(set: &set);
4379
4380 if (copy_to_user(to: uset, from: &set, n: sizeof(old_sigset_t)))
4381 return -EFAULT;
4382
4383 return 0;
4384}
4385
4386#ifdef CONFIG_COMPAT
4387COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4388{
4389 sigset_t set;
4390
4391 do_sigpending(set: &set);
4392
4393 return put_user(set.sig[0], set32);
4394}
4395#endif
4396
4397#endif
4398
4399#ifdef __ARCH_WANT_SYS_SIGPROCMASK
4400/**
4401 * sys_sigprocmask - examine and change blocked signals
4402 * @how: whether to add, remove, or set signals
4403 * @nset: signals to add or remove (if non-null)
4404 * @oset: previous value of signal mask if non-null
4405 *
4406 * Some platforms have their own version with special arguments;
4407 * others support only sys_rt_sigprocmask.
4408 */
4409
4410SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4411 old_sigset_t __user *, oset)
4412{
4413 old_sigset_t old_set, new_set;
4414 sigset_t new_blocked;
4415
4416 old_set = current->blocked.sig[0];
4417
4418 if (nset) {
4419 if (copy_from_user(to: &new_set, from: nset, n: sizeof(*nset)))
4420 return -EFAULT;
4421
4422 new_blocked = current->blocked;
4423
4424 switch (how) {
4425 case SIG_BLOCK:
4426 sigaddsetmask(set: &new_blocked, mask: new_set);
4427 break;
4428 case SIG_UNBLOCK:
4429 sigdelsetmask(set: &new_blocked, mask: new_set);
4430 break;
4431 case SIG_SETMASK:
4432 new_blocked.sig[0] = new_set;
4433 break;
4434 default:
4435 return -EINVAL;
4436 }
4437
4438 set_current_blocked(&new_blocked);
4439 }
4440
4441 if (oset) {
4442 if (copy_to_user(to: oset, from: &old_set, n: sizeof(*oset)))
4443 return -EFAULT;
4444 }
4445
4446 return 0;
4447}
4448#endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4449
4450#ifndef CONFIG_ODD_RT_SIGACTION
4451/**
4452 * sys_rt_sigaction - alter an action taken by a process
4453 * @sig: signal to be sent
4454 * @act: new sigaction
4455 * @oact: used to save the previous sigaction
4456 * @sigsetsize: size of sigset_t type
4457 */
4458SYSCALL_DEFINE4(rt_sigaction, int, sig,
4459 const struct sigaction __user *, act,
4460 struct sigaction __user *, oact,
4461 size_t, sigsetsize)
4462{
4463 struct k_sigaction new_sa, old_sa;
4464 int ret;
4465
4466 /* XXX: Don't preclude handling different sized sigset_t's. */
4467 if (sigsetsize != sizeof(sigset_t))
4468 return -EINVAL;
4469
4470 if (act && copy_from_user(to: &new_sa.sa, from: act, n: sizeof(new_sa.sa)))
4471 return -EFAULT;
4472
4473 ret = do_sigaction(sig, act: act ? &new_sa : NULL, oact: oact ? &old_sa : NULL);
4474 if (ret)
4475 return ret;
4476
4477 if (oact && copy_to_user(to: oact, from: &old_sa.sa, n: sizeof(old_sa.sa)))
4478 return -EFAULT;
4479
4480 return 0;
4481}
4482#ifdef CONFIG_COMPAT
4483COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4484 const struct compat_sigaction __user *, act,
4485 struct compat_sigaction __user *, oact,
4486 compat_size_t, sigsetsize)
4487{
4488 struct k_sigaction new_ka, old_ka;
4489#ifdef __ARCH_HAS_SA_RESTORER
4490 compat_uptr_t restorer;
4491#endif
4492 int ret;
4493
4494 /* XXX: Don't preclude handling different sized sigset_t's. */
4495 if (sigsetsize != sizeof(compat_sigset_t))
4496 return -EINVAL;
4497
4498 if (act) {
4499 compat_uptr_t handler;
4500 ret = get_user(handler, &act->sa_handler);
4501 new_ka.sa.sa_handler = compat_ptr(uptr: handler);
4502#ifdef __ARCH_HAS_SA_RESTORER
4503 ret |= get_user(restorer, &act->sa_restorer);
4504 new_ka.sa.sa_restorer = compat_ptr(uptr: restorer);
4505#endif
4506 ret |= get_compat_sigset(set: &new_ka.sa.sa_mask, compat: &act->sa_mask);
4507 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4508 if (ret)
4509 return -EFAULT;
4510 }
4511
4512 ret = do_sigaction(sig, act: act ? &new_ka : NULL, oact: oact ? &old_ka : NULL);
4513 if (!ret && oact) {
4514 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4515 &oact->sa_handler);
4516 ret |= put_compat_sigset(compat: &oact->sa_mask, set: &old_ka.sa.sa_mask,
4517 size: sizeof(oact->sa_mask));
4518 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4519#ifdef __ARCH_HAS_SA_RESTORER
4520 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4521 &oact->sa_restorer);
4522#endif
4523 }
4524 return ret;
4525}
4526#endif
4527#endif /* !CONFIG_ODD_RT_SIGACTION */
4528
4529#ifdef CONFIG_OLD_SIGACTION
4530SYSCALL_DEFINE3(sigaction, int, sig,
4531 const struct old_sigaction __user *, act,
4532 struct old_sigaction __user *, oact)
4533{
4534 struct k_sigaction new_ka, old_ka;
4535 int ret;
4536
4537 if (act) {
4538 old_sigset_t mask;
4539 if (!access_ok(act, sizeof(*act)) ||
4540 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4541 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4542 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4543 __get_user(mask, &act->sa_mask))
4544 return -EFAULT;
4545#ifdef __ARCH_HAS_KA_RESTORER
4546 new_ka.ka_restorer = NULL;
4547#endif
4548 siginitset(&new_ka.sa.sa_mask, mask);
4549 }
4550
4551 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4552
4553 if (!ret && oact) {
4554 if (!access_ok(oact, sizeof(*oact)) ||
4555 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4556 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4557 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4558 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4559 return -EFAULT;
4560 }
4561
4562 return ret;
4563}
4564#endif
4565#ifdef CONFIG_COMPAT_OLD_SIGACTION
4566COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4567 const struct compat_old_sigaction __user *, act,
4568 struct compat_old_sigaction __user *, oact)
4569{
4570 struct k_sigaction new_ka, old_ka;
4571 int ret;
4572 compat_old_sigset_t mask;
4573 compat_uptr_t handler, restorer;
4574
4575 if (act) {
4576 if (!access_ok(act, sizeof(*act)) ||
4577 __get_user(handler, &act->sa_handler) ||
4578 __get_user(restorer, &act->sa_restorer) ||
4579 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4580 __get_user(mask, &act->sa_mask))
4581 return -EFAULT;
4582
4583#ifdef __ARCH_HAS_KA_RESTORER
4584 new_ka.ka_restorer = NULL;
4585#endif
4586 new_ka.sa.sa_handler = compat_ptr(uptr: handler);
4587 new_ka.sa.sa_restorer = compat_ptr(uptr: restorer);
4588 siginitset(set: &new_ka.sa.sa_mask, mask);
4589 }
4590
4591 ret = do_sigaction(sig, act: act ? &new_ka : NULL, oact: oact ? &old_ka : NULL);
4592
4593 if (!ret && oact) {
4594 if (!access_ok(oact, sizeof(*oact)) ||
4595 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4596 &oact->sa_handler) ||
4597 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4598 &oact->sa_restorer) ||
4599 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4600 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4601 return -EFAULT;
4602 }
4603 return ret;
4604}
4605#endif
4606
4607#ifdef CONFIG_SGETMASK_SYSCALL
4608
4609/*
4610 * For backwards compatibility. Functionality superseded by sigprocmask.
4611 */
4612SYSCALL_DEFINE0(sgetmask)
4613{
4614 /* SMP safe */
4615 return current->blocked.sig[0];
4616}
4617
4618SYSCALL_DEFINE1(ssetmask, int, newmask)
4619{
4620 int old = current->blocked.sig[0];
4621 sigset_t newset;
4622
4623 siginitset(set: &newset, mask: newmask);
4624 set_current_blocked(&newset);
4625
4626 return old;
4627}
4628#endif /* CONFIG_SGETMASK_SYSCALL */
4629
4630#ifdef __ARCH_WANT_SYS_SIGNAL
4631/*
4632 * For backwards compatibility. Functionality superseded by sigaction.
4633 */
4634SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4635{
4636 struct k_sigaction new_sa, old_sa;
4637 int ret;
4638
4639 new_sa.sa.sa_handler = handler;
4640 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4641 sigemptyset(set: &new_sa.sa.sa_mask);
4642
4643 ret = do_sigaction(sig, act: &new_sa, oact: &old_sa);
4644
4645 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4646}
4647#endif /* __ARCH_WANT_SYS_SIGNAL */
4648
4649#ifdef __ARCH_WANT_SYS_PAUSE
4650
4651SYSCALL_DEFINE0(pause)
4652{
4653 while (!signal_pending(current)) {
4654 __set_current_state(TASK_INTERRUPTIBLE);
4655 schedule();
4656 }
4657 return -ERESTARTNOHAND;
4658}
4659
4660#endif
4661
4662static int sigsuspend(sigset_t *set)
4663{
4664 current->saved_sigmask = current->blocked;
4665 set_current_blocked(set);
4666
4667 while (!signal_pending(current)) {
4668 __set_current_state(TASK_INTERRUPTIBLE);
4669 schedule();
4670 }
4671 set_restore_sigmask();
4672 return -ERESTARTNOHAND;
4673}
4674
4675/**
4676 * sys_rt_sigsuspend - replace the signal mask for a value with the
4677 * @unewset value until a signal is received
4678 * @unewset: new signal mask value
4679 * @sigsetsize: size of sigset_t type
4680 */
4681SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4682{
4683 sigset_t newset;
4684
4685 /* XXX: Don't preclude handling different sized sigset_t's. */
4686 if (sigsetsize != sizeof(sigset_t))
4687 return -EINVAL;
4688
4689 if (copy_from_user(to: &newset, from: unewset, n: sizeof(newset)))
4690 return -EFAULT;
4691 return sigsuspend(set: &newset);
4692}
4693
4694#ifdef CONFIG_COMPAT
4695COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4696{
4697 sigset_t newset;
4698
4699 /* XXX: Don't preclude handling different sized sigset_t's. */
4700 if (sigsetsize != sizeof(sigset_t))
4701 return -EINVAL;
4702
4703 if (get_compat_sigset(set: &newset, compat: unewset))
4704 return -EFAULT;
4705 return sigsuspend(set: &newset);
4706}
4707#endif
4708
4709#ifdef CONFIG_OLD_SIGSUSPEND
4710SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4711{
4712 sigset_t blocked;
4713 siginitset(&blocked, mask);
4714 return sigsuspend(&blocked);
4715}
4716#endif
4717#ifdef CONFIG_OLD_SIGSUSPEND3
4718SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4719{
4720 sigset_t blocked;
4721 siginitset(set: &blocked, mask);
4722 return sigsuspend(set: &blocked);
4723}
4724#endif
4725
4726__weak const char *arch_vma_name(struct vm_area_struct *vma)
4727{
4728 return NULL;
4729}
4730
4731static inline void siginfo_buildtime_checks(void)
4732{
4733 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4734
4735 /* Verify the offsets in the two siginfos match */
4736#define CHECK_OFFSET(field) \
4737 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4738
4739 /* kill */
4740 CHECK_OFFSET(si_pid);
4741 CHECK_OFFSET(si_uid);
4742
4743 /* timer */
4744 CHECK_OFFSET(si_tid);
4745 CHECK_OFFSET(si_overrun);
4746 CHECK_OFFSET(si_value);
4747
4748 /* rt */
4749 CHECK_OFFSET(si_pid);
4750 CHECK_OFFSET(si_uid);
4751 CHECK_OFFSET(si_value);
4752
4753 /* sigchld */
4754 CHECK_OFFSET(si_pid);
4755 CHECK_OFFSET(si_uid);
4756 CHECK_OFFSET(si_status);
4757 CHECK_OFFSET(si_utime);
4758 CHECK_OFFSET(si_stime);
4759
4760 /* sigfault */
4761 CHECK_OFFSET(si_addr);
4762 CHECK_OFFSET(si_trapno);
4763 CHECK_OFFSET(si_addr_lsb);
4764 CHECK_OFFSET(si_lower);
4765 CHECK_OFFSET(si_upper);
4766 CHECK_OFFSET(si_pkey);
4767 CHECK_OFFSET(si_perf_data);
4768 CHECK_OFFSET(si_perf_type);
4769 CHECK_OFFSET(si_perf_flags);
4770
4771 /* sigpoll */
4772 CHECK_OFFSET(si_band);
4773 CHECK_OFFSET(si_fd);
4774
4775 /* sigsys */
4776 CHECK_OFFSET(si_call_addr);
4777 CHECK_OFFSET(si_syscall);
4778 CHECK_OFFSET(si_arch);
4779#undef CHECK_OFFSET
4780
4781 /* usb asyncio */
4782 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4783 offsetof(struct siginfo, si_addr));
4784 if (sizeof(int) == sizeof(void __user *)) {
4785 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4786 sizeof(void __user *));
4787 } else {
4788 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4789 sizeof_field(struct siginfo, si_uid)) !=
4790 sizeof(void __user *));
4791 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4792 offsetof(struct siginfo, si_uid));
4793 }
4794#ifdef CONFIG_COMPAT
4795 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4796 offsetof(struct compat_siginfo, si_addr));
4797 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4798 sizeof(compat_uptr_t));
4799 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4800 sizeof_field(struct siginfo, si_pid));
4801#endif
4802}
4803
4804#if defined(CONFIG_SYSCTL)
4805static struct ctl_table signal_debug_table[] = {
4806#ifdef CONFIG_SYSCTL_EXCEPTION_TRACE
4807 {
4808 .procname = "exception-trace",
4809 .data = &show_unhandled_signals,
4810 .maxlen = sizeof(int),
4811 .mode = 0644,
4812 .proc_handler = proc_dointvec
4813 },
4814#endif
4815 { }
4816};
4817
4818static int __init init_signal_sysctls(void)
4819{
4820 register_sysctl_init("debug", signal_debug_table);
4821 return 0;
4822}
4823early_initcall(init_signal_sysctls);
4824#endif /* CONFIG_SYSCTL */
4825
4826void __init signals_init(void)
4827{
4828 siginfo_buildtime_checks();
4829
4830 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC | SLAB_ACCOUNT);
4831}
4832
4833#ifdef CONFIG_KGDB_KDB
4834#include <linux/kdb.h>
4835/*
4836 * kdb_send_sig - Allows kdb to send signals without exposing
4837 * signal internals. This function checks if the required locks are
4838 * available before calling the main signal code, to avoid kdb
4839 * deadlocks.
4840 */
4841void kdb_send_sig(struct task_struct *t, int sig)
4842{
4843 static struct task_struct *kdb_prev_t;
4844 int new_t, ret;
4845 if (!spin_trylock(lock: &t->sighand->siglock)) {
4846 kdb_printf("Can't do kill command now.\n"
4847 "The sigmask lock is held somewhere else in "
4848 "kernel, try again later\n");
4849 return;
4850 }
4851 new_t = kdb_prev_t != t;
4852 kdb_prev_t = t;
4853 if (!task_is_running(t) && new_t) {
4854 spin_unlock(lock: &t->sighand->siglock);
4855 kdb_printf("Process is not RUNNING, sending a signal from "
4856 "kdb risks deadlock\n"
4857 "on the run queue locks. "
4858 "The signal has _not_ been sent.\n"
4859 "Reissue the kill command if you want to risk "
4860 "the deadlock.\n");
4861 return;
4862 }
4863 ret = send_signal_locked(sig, SEND_SIG_PRIV, t, type: PIDTYPE_PID);
4864 spin_unlock(lock: &t->sighand->siglock);
4865 if (ret)
4866 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4867 sig, t->pid);
4868 else
4869 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4870}
4871#endif /* CONFIG_KGDB_KDB */
4872

source code of linux/kernel/signal.c