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 #include <uapi/linux/pidfd.h>
51
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/signal.h>
54
55 #include <asm/param.h>
56 #include <linux/uaccess.h>
57 #include <asm/unistd.h>
58 #include <asm/siginfo.h>
59 #include <asm/cacheflush.h>
60 #include <asm/syscall.h> /* for syscall_get_* */
61
62 /*
63 * SLAB caches for signal bits.
64 */
65
66 static struct kmem_cache *sigqueue_cachep;
67
68 int print_fatal_signals __read_mostly;
69
sig_handler(struct task_struct * t,int sig)70 static void __user *sig_handler(struct task_struct *t, int sig)
71 {
72 return t->sighand->action[sig - 1].sa.sa_handler;
73 }
74
sig_handler_ignored(void __user * handler,int sig)75 static inline bool sig_handler_ignored(void __user *handler, int sig)
76 {
77 /* Is it explicitly or implicitly ignored? */
78 return handler == SIG_IGN ||
79 (handler == SIG_DFL && sig_kernel_ignore(sig));
80 }
81
sig_task_ignored(struct task_struct * t,int sig,bool force)82 static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
83 {
84 void __user *handler;
85
86 handler = sig_handler(t, sig);
87
88 /* SIGKILL and SIGSTOP may not be sent to the global init */
89 if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
90 return true;
91
92 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
93 handler == SIG_DFL && !(force && sig_kernel_only(sig)))
94 return true;
95
96 /* Only allow kernel generated signals to this kthread */
97 if (unlikely((t->flags & PF_KTHREAD) &&
98 (handler == SIG_KTHREAD_KERNEL) && !force))
99 return true;
100
101 return sig_handler_ignored(handler, sig);
102 }
103
sig_ignored(struct task_struct * t,int sig,bool force)104 static bool sig_ignored(struct task_struct *t, int sig, bool force)
105 {
106 /*
107 * Blocked signals are never ignored, since the
108 * signal handler may change by the time it is
109 * unblocked.
110 */
111 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
112 return false;
113
114 /*
115 * Tracers may want to know about even ignored signal unless it
116 * is SIGKILL which can't be reported anyway but can be ignored
117 * by SIGNAL_UNKILLABLE task.
118 */
119 if (t->ptrace && sig != SIGKILL)
120 return false;
121
122 return sig_task_ignored(t, sig, force);
123 }
124
125 /*
126 * Re-calculate pending state from the set of locally pending
127 * signals, globally pending signals, and blocked signals.
128 */
has_pending_signals(sigset_t * signal,sigset_t * blocked)129 static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
130 {
131 unsigned long ready;
132 long i;
133
134 switch (_NSIG_WORDS) {
135 default:
136 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
137 ready |= signal->sig[i] &~ blocked->sig[i];
138 break;
139
140 case 4: ready = signal->sig[3] &~ blocked->sig[3];
141 ready |= signal->sig[2] &~ blocked->sig[2];
142 ready |= signal->sig[1] &~ blocked->sig[1];
143 ready |= signal->sig[0] &~ blocked->sig[0];
144 break;
145
146 case 2: ready = signal->sig[1] &~ blocked->sig[1];
147 ready |= signal->sig[0] &~ blocked->sig[0];
148 break;
149
150 case 1: ready = signal->sig[0] &~ blocked->sig[0];
151 }
152 return ready != 0;
153 }
154
155 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
156
recalc_sigpending_tsk(struct task_struct * t)157 static bool recalc_sigpending_tsk(struct task_struct *t)
158 {
159 if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) ||
160 PENDING(&t->pending, &t->blocked) ||
161 PENDING(&t->signal->shared_pending, &t->blocked) ||
162 cgroup_task_frozen(t)) {
163 set_tsk_thread_flag(t, TIF_SIGPENDING);
164 return true;
165 }
166
167 /*
168 * We must never clear the flag in another thread, or in current
169 * when it's possible the current syscall is returning -ERESTART*.
170 * So we don't clear it here, and only callers who know they should do.
171 */
172 return false;
173 }
174
recalc_sigpending(void)175 void recalc_sigpending(void)
176 {
177 if (!recalc_sigpending_tsk(current) && !freezing(current))
178 clear_thread_flag(TIF_SIGPENDING);
179
180 }
181 EXPORT_SYMBOL(recalc_sigpending);
182
calculate_sigpending(void)183 void calculate_sigpending(void)
184 {
185 /* Have any signals or users of TIF_SIGPENDING been delayed
186 * until after fork?
187 */
188 spin_lock_irq(¤t->sighand->siglock);
189 set_tsk_thread_flag(current, TIF_SIGPENDING);
190 recalc_sigpending();
191 spin_unlock_irq(¤t->sighand->siglock);
192 }
193
194 /* Given the mask, find the first available signal that should be serviced. */
195
196 #define SYNCHRONOUS_MASK \
197 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
198 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
199
next_signal(struct sigpending * pending,sigset_t * mask)200 int next_signal(struct sigpending *pending, sigset_t *mask)
201 {
202 unsigned long i, *s, *m, x;
203 int sig = 0;
204
205 s = pending->signal.sig;
206 m = mask->sig;
207
208 /*
209 * Handle the first word specially: it contains the
210 * synchronous signals that need to be dequeued first.
211 */
212 x = *s &~ *m;
213 if (x) {
214 if (x & SYNCHRONOUS_MASK)
215 x &= SYNCHRONOUS_MASK;
216 sig = ffz(~x) + 1;
217 return sig;
218 }
219
220 switch (_NSIG_WORDS) {
221 default:
222 for (i = 1; i < _NSIG_WORDS; ++i) {
223 x = *++s &~ *++m;
224 if (!x)
225 continue;
226 sig = ffz(~x) + i*_NSIG_BPW + 1;
227 break;
228 }
229 break;
230
231 case 2:
232 x = s[1] &~ m[1];
233 if (!x)
234 break;
235 sig = ffz(~x) + _NSIG_BPW + 1;
236 break;
237
238 case 1:
239 /* Nothing to do */
240 break;
241 }
242
243 return sig;
244 }
245
print_dropped_signal(int sig)246 static inline void print_dropped_signal(int sig)
247 {
248 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
249
250 if (!print_fatal_signals)
251 return;
252
253 if (!__ratelimit(&ratelimit_state))
254 return;
255
256 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
257 current->comm, current->pid, sig);
258 }
259
260 /**
261 * task_set_jobctl_pending - set jobctl pending bits
262 * @task: target task
263 * @mask: pending bits to set
264 *
265 * Clear @mask from @task->jobctl. @mask must be subset of
266 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
267 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
268 * cleared. If @task is already being killed or exiting, this function
269 * becomes noop.
270 *
271 * CONTEXT:
272 * Must be called with @task->sighand->siglock held.
273 *
274 * RETURNS:
275 * %true if @mask is set, %false if made noop because @task was dying.
276 */
task_set_jobctl_pending(struct task_struct * task,unsigned long mask)277 bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
278 {
279 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
280 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
281 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
282
283 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
284 return false;
285
286 if (mask & JOBCTL_STOP_SIGMASK)
287 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
288
289 task->jobctl |= mask;
290 return true;
291 }
292
293 /**
294 * task_clear_jobctl_trapping - clear jobctl trapping bit
295 * @task: target task
296 *
297 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
298 * Clear it and wake up the ptracer. Note that we don't need any further
299 * locking. @task->siglock guarantees that @task->parent points to the
300 * ptracer.
301 *
302 * CONTEXT:
303 * Must be called with @task->sighand->siglock held.
304 */
task_clear_jobctl_trapping(struct task_struct * task)305 void task_clear_jobctl_trapping(struct task_struct *task)
306 {
307 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
308 task->jobctl &= ~JOBCTL_TRAPPING;
309 smp_mb(); /* advised by wake_up_bit() */
310 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
311 }
312 }
313
314 /**
315 * task_clear_jobctl_pending - clear jobctl pending bits
316 * @task: target task
317 * @mask: pending bits to clear
318 *
319 * Clear @mask from @task->jobctl. @mask must be subset of
320 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
321 * STOP bits are cleared together.
322 *
323 * If clearing of @mask leaves no stop or trap pending, this function calls
324 * task_clear_jobctl_trapping().
325 *
326 * CONTEXT:
327 * Must be called with @task->sighand->siglock held.
328 */
task_clear_jobctl_pending(struct task_struct * task,unsigned long mask)329 void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
330 {
331 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
332
333 if (mask & JOBCTL_STOP_PENDING)
334 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
335
336 task->jobctl &= ~mask;
337
338 if (!(task->jobctl & JOBCTL_PENDING_MASK))
339 task_clear_jobctl_trapping(task);
340 }
341
342 /**
343 * task_participate_group_stop - participate in a group stop
344 * @task: task participating in a group stop
345 *
346 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
347 * Group stop states are cleared and the group stop count is consumed if
348 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
349 * stop, the appropriate `SIGNAL_*` flags are set.
350 *
351 * CONTEXT:
352 * Must be called with @task->sighand->siglock held.
353 *
354 * RETURNS:
355 * %true if group stop completion should be notified to the parent, %false
356 * otherwise.
357 */
task_participate_group_stop(struct task_struct * task)358 static bool task_participate_group_stop(struct task_struct *task)
359 {
360 struct signal_struct *sig = task->signal;
361 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
362
363 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
364
365 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
366
367 if (!consume)
368 return false;
369
370 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
371 sig->group_stop_count--;
372
373 /*
374 * Tell the caller to notify completion iff we are entering into a
375 * fresh group stop. Read comment in do_signal_stop() for details.
376 */
377 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
378 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
379 return true;
380 }
381 return false;
382 }
383
task_join_group_stop(struct task_struct * task)384 void task_join_group_stop(struct task_struct *task)
385 {
386 unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK;
387 struct signal_struct *sig = current->signal;
388
389 if (sig->group_stop_count) {
390 sig->group_stop_count++;
391 mask |= JOBCTL_STOP_CONSUME;
392 } else if (!(sig->flags & SIGNAL_STOP_STOPPED))
393 return;
394
395 /* Have the new thread join an on-going signal group stop */
396 task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING);
397 }
398
399 /*
400 * allocate a new signal queue record
401 * - this may be called without locks if and only if t == current, otherwise an
402 * appropriate lock must be held to stop the target task from exiting
403 */
404 static struct sigqueue *
__sigqueue_alloc(int sig,struct task_struct * t,gfp_t gfp_flags,int override_rlimit,const unsigned int sigqueue_flags)405 __sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
406 int override_rlimit, const unsigned int sigqueue_flags)
407 {
408 struct sigqueue *q = NULL;
409 struct ucounts *ucounts;
410 long sigpending;
411
412 /*
413 * Protect access to @t credentials. This can go away when all
414 * callers hold rcu read lock.
415 *
416 * NOTE! A pending signal will hold on to the user refcount,
417 * and we get/put the refcount only when the sigpending count
418 * changes from/to zero.
419 */
420 rcu_read_lock();
421 ucounts = task_ucounts(t);
422 sigpending = inc_rlimit_get_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
423 rcu_read_unlock();
424 if (!sigpending)
425 return NULL;
426
427 if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
428 q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
429 } else {
430 print_dropped_signal(sig);
431 }
432
433 if (unlikely(q == NULL)) {
434 dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
435 } else {
436 INIT_LIST_HEAD(&q->list);
437 q->flags = sigqueue_flags;
438 q->ucounts = ucounts;
439 }
440 return q;
441 }
442
__sigqueue_free(struct sigqueue * q)443 static void __sigqueue_free(struct sigqueue *q)
444 {
445 if (q->flags & SIGQUEUE_PREALLOC)
446 return;
447 if (q->ucounts) {
448 dec_rlimit_put_ucounts(q->ucounts, UCOUNT_RLIMIT_SIGPENDING);
449 q->ucounts = NULL;
450 }
451 kmem_cache_free(sigqueue_cachep, q);
452 }
453
flush_sigqueue(struct sigpending * queue)454 void flush_sigqueue(struct sigpending *queue)
455 {
456 struct sigqueue *q;
457
458 sigemptyset(&queue->signal);
459 while (!list_empty(&queue->list)) {
460 q = list_entry(queue->list.next, struct sigqueue , list);
461 list_del_init(&q->list);
462 __sigqueue_free(q);
463 }
464 }
465
466 /*
467 * Flush all pending signals for this kthread.
468 */
flush_signals(struct task_struct * t)469 void flush_signals(struct task_struct *t)
470 {
471 unsigned long flags;
472
473 spin_lock_irqsave(&t->sighand->siglock, flags);
474 clear_tsk_thread_flag(t, TIF_SIGPENDING);
475 flush_sigqueue(&t->pending);
476 flush_sigqueue(&t->signal->shared_pending);
477 spin_unlock_irqrestore(&t->sighand->siglock, flags);
478 }
479 EXPORT_SYMBOL(flush_signals);
480
481 #ifdef CONFIG_POSIX_TIMERS
__flush_itimer_signals(struct sigpending * pending)482 static void __flush_itimer_signals(struct sigpending *pending)
483 {
484 sigset_t signal, retain;
485 struct sigqueue *q, *n;
486
487 signal = pending->signal;
488 sigemptyset(&retain);
489
490 list_for_each_entry_safe(q, n, &pending->list, list) {
491 int sig = q->info.si_signo;
492
493 if (likely(q->info.si_code != SI_TIMER)) {
494 sigaddset(&retain, sig);
495 } else {
496 sigdelset(&signal, sig);
497 list_del_init(&q->list);
498 __sigqueue_free(q);
499 }
500 }
501
502 sigorsets(&pending->signal, &signal, &retain);
503 }
504
flush_itimer_signals(void)505 void flush_itimer_signals(void)
506 {
507 struct task_struct *tsk = current;
508 unsigned long flags;
509
510 spin_lock_irqsave(&tsk->sighand->siglock, flags);
511 __flush_itimer_signals(&tsk->pending);
512 __flush_itimer_signals(&tsk->signal->shared_pending);
513 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
514 }
515 #endif
516
ignore_signals(struct task_struct * t)517 void ignore_signals(struct task_struct *t)
518 {
519 int i;
520
521 for (i = 0; i < _NSIG; ++i)
522 t->sighand->action[i].sa.sa_handler = SIG_IGN;
523
524 flush_signals(t);
525 }
526
527 /*
528 * Flush all handlers for a task.
529 */
530
531 void
flush_signal_handlers(struct task_struct * t,int force_default)532 flush_signal_handlers(struct task_struct *t, int force_default)
533 {
534 int i;
535 struct k_sigaction *ka = &t->sighand->action[0];
536 for (i = _NSIG ; i != 0 ; i--) {
537 if (force_default || ka->sa.sa_handler != SIG_IGN)
538 ka->sa.sa_handler = SIG_DFL;
539 ka->sa.sa_flags = 0;
540 #ifdef __ARCH_HAS_SA_RESTORER
541 ka->sa.sa_restorer = NULL;
542 #endif
543 sigemptyset(&ka->sa.sa_mask);
544 ka++;
545 }
546 }
547
unhandled_signal(struct task_struct * tsk,int sig)548 bool unhandled_signal(struct task_struct *tsk, int sig)
549 {
550 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
551 if (is_global_init(tsk))
552 return true;
553
554 if (handler != SIG_IGN && handler != SIG_DFL)
555 return false;
556
557 /* If dying, we handle all new signals by ignoring them */
558 if (fatal_signal_pending(tsk))
559 return false;
560
561 /* if ptraced, let the tracer determine */
562 return !tsk->ptrace;
563 }
564
collect_signal(int sig,struct sigpending * list,kernel_siginfo_t * info,bool * resched_timer)565 static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
566 bool *resched_timer)
567 {
568 struct sigqueue *q, *first = NULL;
569
570 /*
571 * Collect the siginfo appropriate to this signal. Check if
572 * there is another siginfo for the same signal.
573 */
574 list_for_each_entry(q, &list->list, list) {
575 if (q->info.si_signo == sig) {
576 if (first)
577 goto still_pending;
578 first = q;
579 }
580 }
581
582 sigdelset(&list->signal, sig);
583
584 if (first) {
585 still_pending:
586 list_del_init(&first->list);
587 copy_siginfo(info, &first->info);
588
589 *resched_timer =
590 (first->flags & SIGQUEUE_PREALLOC) &&
591 (info->si_code == SI_TIMER) &&
592 (info->si_sys_private);
593
594 __sigqueue_free(first);
595 } else {
596 /*
597 * Ok, it wasn't in the queue. This must be
598 * a fast-pathed signal or we must have been
599 * out of queue space. So zero out the info.
600 */
601 clear_siginfo(info);
602 info->si_signo = sig;
603 info->si_errno = 0;
604 info->si_code = SI_USER;
605 info->si_pid = 0;
606 info->si_uid = 0;
607 }
608 }
609
__dequeue_signal(struct sigpending * pending,sigset_t * mask,kernel_siginfo_t * info,bool * resched_timer)610 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
611 kernel_siginfo_t *info, bool *resched_timer)
612 {
613 int sig = next_signal(pending, mask);
614
615 if (sig)
616 collect_signal(sig, pending, info, resched_timer);
617 return sig;
618 }
619
620 /*
621 * Dequeue a signal and return the element to the caller, which is
622 * expected to free it.
623 *
624 * All callers have to hold the siglock.
625 */
dequeue_signal(struct task_struct * tsk,sigset_t * mask,kernel_siginfo_t * info,enum pid_type * type)626 int dequeue_signal(struct task_struct *tsk, sigset_t *mask,
627 kernel_siginfo_t *info, enum pid_type *type)
628 {
629 bool resched_timer = false;
630 int signr;
631
632 /* We only dequeue private signals from ourselves, we don't let
633 * signalfd steal them
634 */
635 *type = PIDTYPE_PID;
636 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
637 if (!signr) {
638 *type = PIDTYPE_TGID;
639 signr = __dequeue_signal(&tsk->signal->shared_pending,
640 mask, info, &resched_timer);
641 #ifdef CONFIG_POSIX_TIMERS
642 /*
643 * itimer signal ?
644 *
645 * itimers are process shared and we restart periodic
646 * itimers in the signal delivery path to prevent DoS
647 * attacks in the high resolution timer case. This is
648 * compliant with the old way of self-restarting
649 * itimers, as the SIGALRM is a legacy signal and only
650 * queued once. Changing the restart behaviour to
651 * restart the timer in the signal dequeue path is
652 * reducing the timer noise on heavy loaded !highres
653 * systems too.
654 */
655 if (unlikely(signr == SIGALRM)) {
656 struct hrtimer *tmr = &tsk->signal->real_timer;
657
658 if (!hrtimer_is_queued(tmr) &&
659 tsk->signal->it_real_incr != 0) {
660 hrtimer_forward(tmr, tmr->base->get_time(),
661 tsk->signal->it_real_incr);
662 hrtimer_restart(tmr);
663 }
664 }
665 #endif
666 }
667
668 recalc_sigpending();
669 if (!signr)
670 return 0;
671
672 if (unlikely(sig_kernel_stop(signr))) {
673 /*
674 * Set a marker that we have dequeued a stop signal. Our
675 * caller might release the siglock and then the pending
676 * stop signal it is about to process is no longer in the
677 * pending bitmasks, but must still be cleared by a SIGCONT
678 * (and overruled by a SIGKILL). So those cases clear this
679 * shared flag after we've set it. Note that this flag may
680 * remain set after the signal we return is ignored or
681 * handled. That doesn't matter because its only purpose
682 * is to alert stop-signal processing code when another
683 * processor has come along and cleared the flag.
684 */
685 current->jobctl |= JOBCTL_STOP_DEQUEUED;
686 }
687 #ifdef CONFIG_POSIX_TIMERS
688 if (resched_timer) {
689 /*
690 * Release the siglock to ensure proper locking order
691 * of timer locks outside of siglocks. Note, we leave
692 * irqs disabled here, since the posix-timers code is
693 * about to disable them again anyway.
694 */
695 spin_unlock(&tsk->sighand->siglock);
696 posixtimer_rearm(info);
697 spin_lock(&tsk->sighand->siglock);
698
699 /* Don't expose the si_sys_private value to userspace */
700 info->si_sys_private = 0;
701 }
702 #endif
703 return signr;
704 }
705 EXPORT_SYMBOL_GPL(dequeue_signal);
706
dequeue_synchronous_signal(kernel_siginfo_t * info)707 static int dequeue_synchronous_signal(kernel_siginfo_t *info)
708 {
709 struct task_struct *tsk = current;
710 struct sigpending *pending = &tsk->pending;
711 struct sigqueue *q, *sync = NULL;
712
713 /*
714 * Might a synchronous signal be in the queue?
715 */
716 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
717 return 0;
718
719 /*
720 * Return the first synchronous signal in the queue.
721 */
722 list_for_each_entry(q, &pending->list, list) {
723 /* Synchronous signals have a positive si_code */
724 if ((q->info.si_code > SI_USER) &&
725 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
726 sync = q;
727 goto next;
728 }
729 }
730 return 0;
731 next:
732 /*
733 * Check if there is another siginfo for the same signal.
734 */
735 list_for_each_entry_continue(q, &pending->list, list) {
736 if (q->info.si_signo == sync->info.si_signo)
737 goto still_pending;
738 }
739
740 sigdelset(&pending->signal, sync->info.si_signo);
741 recalc_sigpending();
742 still_pending:
743 list_del_init(&sync->list);
744 copy_siginfo(info, &sync->info);
745 __sigqueue_free(sync);
746 return info->si_signo;
747 }
748
749 /*
750 * Tell a process that it has a new active signal..
751 *
752 * NOTE! we rely on the previous spin_lock to
753 * lock interrupts for us! We can only be called with
754 * "siglock" held, and the local interrupt must
755 * have been disabled when that got acquired!
756 *
757 * No need to set need_resched since signal event passing
758 * goes through ->blocked
759 */
signal_wake_up_state(struct task_struct * t,unsigned int state)760 void signal_wake_up_state(struct task_struct *t, unsigned int state)
761 {
762 lockdep_assert_held(&t->sighand->siglock);
763
764 set_tsk_thread_flag(t, TIF_SIGPENDING);
765
766 /*
767 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
768 * case. We don't check t->state here because there is a race with it
769 * executing another processor and just now entering stopped state.
770 * By using wake_up_state, we ensure the process will wake up and
771 * handle its death signal.
772 */
773 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
774 kick_process(t);
775 }
776
777 /*
778 * Remove signals in mask from the pending set and queue.
779 * Returns 1 if any signals were found.
780 *
781 * All callers must be holding the siglock.
782 */
flush_sigqueue_mask(sigset_t * mask,struct sigpending * s)783 static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
784 {
785 struct sigqueue *q, *n;
786 sigset_t m;
787
788 sigandsets(&m, mask, &s->signal);
789 if (sigisemptyset(&m))
790 return;
791
792 sigandnsets(&s->signal, &s->signal, mask);
793 list_for_each_entry_safe(q, n, &s->list, list) {
794 if (sigismember(mask, q->info.si_signo)) {
795 list_del_init(&q->list);
796 __sigqueue_free(q);
797 }
798 }
799 }
800
is_si_special(const struct kernel_siginfo * info)801 static inline int is_si_special(const struct kernel_siginfo *info)
802 {
803 return info <= SEND_SIG_PRIV;
804 }
805
si_fromuser(const struct kernel_siginfo * info)806 static inline bool si_fromuser(const struct kernel_siginfo *info)
807 {
808 return info == SEND_SIG_NOINFO ||
809 (!is_si_special(info) && SI_FROMUSER(info));
810 }
811
812 /*
813 * called with RCU read lock from check_kill_permission()
814 */
kill_ok_by_cred(struct task_struct * t)815 static bool kill_ok_by_cred(struct task_struct *t)
816 {
817 const struct cred *cred = current_cred();
818 const struct cred *tcred = __task_cred(t);
819
820 return uid_eq(cred->euid, tcred->suid) ||
821 uid_eq(cred->euid, tcred->uid) ||
822 uid_eq(cred->uid, tcred->suid) ||
823 uid_eq(cred->uid, tcred->uid) ||
824 ns_capable(tcred->user_ns, CAP_KILL);
825 }
826
827 /*
828 * Bad permissions for sending the signal
829 * - the caller must hold the RCU read lock
830 */
check_kill_permission(int sig,struct kernel_siginfo * info,struct task_struct * t)831 static int check_kill_permission(int sig, struct kernel_siginfo *info,
832 struct task_struct *t)
833 {
834 struct pid *sid;
835 int error;
836
837 if (!valid_signal(sig))
838 return -EINVAL;
839
840 if (!si_fromuser(info))
841 return 0;
842
843 error = audit_signal_info(sig, t); /* Let audit system see the signal */
844 if (error)
845 return error;
846
847 if (!same_thread_group(current, t) &&
848 !kill_ok_by_cred(t)) {
849 switch (sig) {
850 case SIGCONT:
851 sid = task_session(t);
852 /*
853 * We don't return the error if sid == NULL. The
854 * task was unhashed, the caller must notice this.
855 */
856 if (!sid || sid == task_session(current))
857 break;
858 fallthrough;
859 default:
860 return -EPERM;
861 }
862 }
863
864 return security_task_kill(t, info, sig, NULL);
865 }
866
867 /**
868 * ptrace_trap_notify - schedule trap to notify ptracer
869 * @t: tracee wanting to notify tracer
870 *
871 * This function schedules sticky ptrace trap which is cleared on the next
872 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
873 * ptracer.
874 *
875 * If @t is running, STOP trap will be taken. If trapped for STOP and
876 * ptracer is listening for events, tracee is woken up so that it can
877 * re-trap for the new event. If trapped otherwise, STOP trap will be
878 * eventually taken without returning to userland after the existing traps
879 * are finished by PTRACE_CONT.
880 *
881 * CONTEXT:
882 * Must be called with @task->sighand->siglock held.
883 */
ptrace_trap_notify(struct task_struct * t)884 static void ptrace_trap_notify(struct task_struct *t)
885 {
886 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
887 lockdep_assert_held(&t->sighand->siglock);
888
889 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
890 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
891 }
892
893 /*
894 * Handle magic process-wide effects of stop/continue signals. Unlike
895 * the signal actions, these happen immediately at signal-generation
896 * time regardless of blocking, ignoring, or handling. This does the
897 * actual continuing for SIGCONT, but not the actual stopping for stop
898 * signals. The process stop is done as a signal action for SIG_DFL.
899 *
900 * Returns true if the signal should be actually delivered, otherwise
901 * it should be dropped.
902 */
prepare_signal(int sig,struct task_struct * p,bool force)903 static bool prepare_signal(int sig, struct task_struct *p, bool force)
904 {
905 struct signal_struct *signal = p->signal;
906 struct task_struct *t;
907 sigset_t flush;
908
909 if (signal->flags & SIGNAL_GROUP_EXIT) {
910 if (signal->core_state)
911 return sig == SIGKILL;
912 /*
913 * The process is in the middle of dying, drop the signal.
914 */
915 return false;
916 } else if (sig_kernel_stop(sig)) {
917 /*
918 * This is a stop signal. Remove SIGCONT from all queues.
919 */
920 siginitset(&flush, sigmask(SIGCONT));
921 flush_sigqueue_mask(&flush, &signal->shared_pending);
922 for_each_thread(p, t)
923 flush_sigqueue_mask(&flush, &t->pending);
924 } else if (sig == SIGCONT) {
925 unsigned int why;
926 /*
927 * Remove all stop signals from all queues, wake all threads.
928 */
929 siginitset(&flush, SIG_KERNEL_STOP_MASK);
930 flush_sigqueue_mask(&flush, &signal->shared_pending);
931 for_each_thread(p, t) {
932 flush_sigqueue_mask(&flush, &t->pending);
933 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
934 if (likely(!(t->ptrace & PT_SEIZED))) {
935 t->jobctl &= ~JOBCTL_STOPPED;
936 wake_up_state(t, __TASK_STOPPED);
937 } else
938 ptrace_trap_notify(t);
939 }
940
941 /*
942 * Notify the parent with CLD_CONTINUED if we were stopped.
943 *
944 * If we were in the middle of a group stop, we pretend it
945 * was already finished, and then continued. Since SIGCHLD
946 * doesn't queue we report only CLD_STOPPED, as if the next
947 * CLD_CONTINUED was dropped.
948 */
949 why = 0;
950 if (signal->flags & SIGNAL_STOP_STOPPED)
951 why |= SIGNAL_CLD_CONTINUED;
952 else if (signal->group_stop_count)
953 why |= SIGNAL_CLD_STOPPED;
954
955 if (why) {
956 /*
957 * The first thread which returns from do_signal_stop()
958 * will take ->siglock, notice SIGNAL_CLD_MASK, and
959 * notify its parent. See get_signal().
960 */
961 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
962 signal->group_stop_count = 0;
963 signal->group_exit_code = 0;
964 }
965 }
966
967 return !sig_ignored(p, sig, force);
968 }
969
970 /*
971 * Test if P wants to take SIG. After we've checked all threads with this,
972 * it's equivalent to finding no threads not blocking SIG. Any threads not
973 * blocking SIG were ruled out because they are not running and already
974 * have pending signals. Such threads will dequeue from the shared queue
975 * as soon as they're available, so putting the signal on the shared queue
976 * will be equivalent to sending it to one such thread.
977 */
wants_signal(int sig,struct task_struct * p)978 static inline bool wants_signal(int sig, struct task_struct *p)
979 {
980 if (sigismember(&p->blocked, sig))
981 return false;
982
983 if (p->flags & PF_EXITING)
984 return false;
985
986 if (sig == SIGKILL)
987 return true;
988
989 if (task_is_stopped_or_traced(p))
990 return false;
991
992 return task_curr(p) || !task_sigpending(p);
993 }
994
complete_signal(int sig,struct task_struct * p,enum pid_type type)995 static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
996 {
997 struct signal_struct *signal = p->signal;
998 struct task_struct *t;
999
1000 /*
1001 * Now find a thread we can wake up to take the signal off the queue.
1002 *
1003 * Try the suggested task first (may or may not be the main thread).
1004 */
1005 if (wants_signal(sig, p))
1006 t = p;
1007 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1008 /*
1009 * There is just one thread and it does not need to be woken.
1010 * It will dequeue unblocked signals before it runs again.
1011 */
1012 return;
1013 else {
1014 /*
1015 * Otherwise try to find a suitable thread.
1016 */
1017 t = signal->curr_target;
1018 while (!wants_signal(sig, t)) {
1019 t = next_thread(t);
1020 if (t == signal->curr_target)
1021 /*
1022 * No thread needs to be woken.
1023 * Any eligible threads will see
1024 * the signal in the queue soon.
1025 */
1026 return;
1027 }
1028 signal->curr_target = t;
1029 }
1030
1031 /*
1032 * Found a killable thread. If the signal will be fatal,
1033 * then start taking the whole group down immediately.
1034 */
1035 if (sig_fatal(p, sig) &&
1036 (signal->core_state || !(signal->flags & SIGNAL_GROUP_EXIT)) &&
1037 !sigismember(&t->real_blocked, sig) &&
1038 (sig == SIGKILL || !p->ptrace)) {
1039 /*
1040 * This signal will be fatal to the whole group.
1041 */
1042 if (!sig_kernel_coredump(sig)) {
1043 /*
1044 * Start a group exit and wake everybody up.
1045 * This way we don't have other threads
1046 * running and doing things after a slower
1047 * thread has the fatal signal pending.
1048 */
1049 signal->flags = SIGNAL_GROUP_EXIT;
1050 signal->group_exit_code = sig;
1051 signal->group_stop_count = 0;
1052 __for_each_thread(signal, t) {
1053 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1054 sigaddset(&t->pending.signal, SIGKILL);
1055 signal_wake_up(t, 1);
1056 }
1057 return;
1058 }
1059 }
1060
1061 /*
1062 * The signal is already in the shared-pending queue.
1063 * Tell the chosen thread to wake up and dequeue it.
1064 */
1065 signal_wake_up(t, sig == SIGKILL);
1066 return;
1067 }
1068
legacy_queue(struct sigpending * signals,int sig)1069 static inline bool legacy_queue(struct sigpending *signals, int sig)
1070 {
1071 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1072 }
1073
__send_signal_locked(int sig,struct kernel_siginfo * info,struct task_struct * t,enum pid_type type,bool force)1074 static int __send_signal_locked(int sig, struct kernel_siginfo *info,
1075 struct task_struct *t, enum pid_type type, bool force)
1076 {
1077 struct sigpending *pending;
1078 struct sigqueue *q;
1079 int override_rlimit;
1080 int ret = 0, result;
1081
1082 lockdep_assert_held(&t->sighand->siglock);
1083
1084 result = TRACE_SIGNAL_IGNORED;
1085 if (!prepare_signal(sig, t, force))
1086 goto ret;
1087
1088 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1089 /*
1090 * Short-circuit ignored signals and support queuing
1091 * exactly one non-rt signal, so that we can get more
1092 * detailed information about the cause of the signal.
1093 */
1094 result = TRACE_SIGNAL_ALREADY_PENDING;
1095 if (legacy_queue(pending, sig))
1096 goto ret;
1097
1098 result = TRACE_SIGNAL_DELIVERED;
1099 /*
1100 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1101 */
1102 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1103 goto out_set;
1104
1105 /*
1106 * Real-time signals must be queued if sent by sigqueue, or
1107 * some other real-time mechanism. It is implementation
1108 * defined whether kill() does so. We attempt to do so, on
1109 * the principle of least surprise, but since kill is not
1110 * allowed to fail with EAGAIN when low on memory we just
1111 * make sure at least one signal gets delivered and don't
1112 * pass on the info struct.
1113 */
1114 if (sig < SIGRTMIN)
1115 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1116 else
1117 override_rlimit = 0;
1118
1119 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit, 0);
1120
1121 if (q) {
1122 list_add_tail(&q->list, &pending->list);
1123 switch ((unsigned long) info) {
1124 case (unsigned long) SEND_SIG_NOINFO:
1125 clear_siginfo(&q->info);
1126 q->info.si_signo = sig;
1127 q->info.si_errno = 0;
1128 q->info.si_code = SI_USER;
1129 q->info.si_pid = task_tgid_nr_ns(current,
1130 task_active_pid_ns(t));
1131 rcu_read_lock();
1132 q->info.si_uid =
1133 from_kuid_munged(task_cred_xxx(t, user_ns),
1134 current_uid());
1135 rcu_read_unlock();
1136 break;
1137 case (unsigned long) SEND_SIG_PRIV:
1138 clear_siginfo(&q->info);
1139 q->info.si_signo = sig;
1140 q->info.si_errno = 0;
1141 q->info.si_code = SI_KERNEL;
1142 q->info.si_pid = 0;
1143 q->info.si_uid = 0;
1144 break;
1145 default:
1146 copy_siginfo(&q->info, info);
1147 break;
1148 }
1149 } else if (!is_si_special(info) &&
1150 sig >= SIGRTMIN && info->si_code != SI_USER) {
1151 /*
1152 * Queue overflow, abort. We may abort if the
1153 * signal was rt and sent by user using something
1154 * other than kill().
1155 */
1156 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1157 ret = -EAGAIN;
1158 goto ret;
1159 } else {
1160 /*
1161 * This is a silent loss of information. We still
1162 * send the signal, but the *info bits are lost.
1163 */
1164 result = TRACE_SIGNAL_LOSE_INFO;
1165 }
1166
1167 out_set:
1168 signalfd_notify(t, sig);
1169 sigaddset(&pending->signal, sig);
1170
1171 /* Let multiprocess signals appear after on-going forks */
1172 if (type > PIDTYPE_TGID) {
1173 struct multiprocess_signals *delayed;
1174 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1175 sigset_t *signal = &delayed->signal;
1176 /* Can't queue both a stop and a continue signal */
1177 if (sig == SIGCONT)
1178 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1179 else if (sig_kernel_stop(sig))
1180 sigdelset(signal, SIGCONT);
1181 sigaddset(signal, sig);
1182 }
1183 }
1184
1185 complete_signal(sig, t, type);
1186 ret:
1187 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1188 return ret;
1189 }
1190
has_si_pid_and_uid(struct kernel_siginfo * info)1191 static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1192 {
1193 bool ret = false;
1194 switch (siginfo_layout(info->si_signo, info->si_code)) {
1195 case SIL_KILL:
1196 case SIL_CHLD:
1197 case SIL_RT:
1198 ret = true;
1199 break;
1200 case SIL_TIMER:
1201 case SIL_POLL:
1202 case SIL_FAULT:
1203 case SIL_FAULT_TRAPNO:
1204 case SIL_FAULT_MCEERR:
1205 case SIL_FAULT_BNDERR:
1206 case SIL_FAULT_PKUERR:
1207 case SIL_FAULT_PERF_EVENT:
1208 case SIL_SYS:
1209 ret = false;
1210 break;
1211 }
1212 return ret;
1213 }
1214
send_signal_locked(int sig,struct kernel_siginfo * info,struct task_struct * t,enum pid_type type)1215 int send_signal_locked(int sig, struct kernel_siginfo *info,
1216 struct task_struct *t, enum pid_type type)
1217 {
1218 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1219 bool force = false;
1220
1221 if (info == SEND_SIG_NOINFO) {
1222 /* Force if sent from an ancestor pid namespace */
1223 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1224 } else if (info == SEND_SIG_PRIV) {
1225 /* Don't ignore kernel generated signals */
1226 force = true;
1227 } else if (has_si_pid_and_uid(info)) {
1228 /* SIGKILL and SIGSTOP is special or has ids */
1229 struct user_namespace *t_user_ns;
1230
1231 rcu_read_lock();
1232 t_user_ns = task_cred_xxx(t, user_ns);
1233 if (current_user_ns() != t_user_ns) {
1234 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1235 info->si_uid = from_kuid_munged(t_user_ns, uid);
1236 }
1237 rcu_read_unlock();
1238
1239 /* A kernel generated signal? */
1240 force = (info->si_code == SI_KERNEL);
1241
1242 /* From an ancestor pid namespace? */
1243 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1244 info->si_pid = 0;
1245 force = true;
1246 }
1247 }
1248 return __send_signal_locked(sig, info, t, type, force);
1249 }
1250
print_fatal_signal(int signr)1251 static void print_fatal_signal(int signr)
1252 {
1253 struct pt_regs *regs = task_pt_regs(current);
1254 struct file *exe_file;
1255
1256 exe_file = get_task_exe_file(current);
1257 if (exe_file) {
1258 pr_info("%pD: %s: potentially unexpected fatal signal %d.\n",
1259 exe_file, current->comm, signr);
1260 fput(exe_file);
1261 } else {
1262 pr_info("%s: potentially unexpected fatal signal %d.\n",
1263 current->comm, signr);
1264 }
1265
1266 #if defined(__i386__) && !defined(__arch_um__)
1267 pr_info("code at %08lx: ", regs->ip);
1268 {
1269 int i;
1270 for (i = 0; i < 16; i++) {
1271 unsigned char insn;
1272
1273 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1274 break;
1275 pr_cont("%02x ", insn);
1276 }
1277 }
1278 pr_cont("\n");
1279 #endif
1280 preempt_disable();
1281 show_regs(regs);
1282 preempt_enable();
1283 }
1284
setup_print_fatal_signals(char * str)1285 static int __init setup_print_fatal_signals(char *str)
1286 {
1287 get_option (&str, &print_fatal_signals);
1288
1289 return 1;
1290 }
1291
1292 __setup("print-fatal-signals=", setup_print_fatal_signals);
1293
do_send_sig_info(int sig,struct kernel_siginfo * info,struct task_struct * p,enum pid_type type)1294 int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1295 enum pid_type type)
1296 {
1297 unsigned long flags;
1298 int ret = -ESRCH;
1299
1300 if (lock_task_sighand(p, &flags)) {
1301 ret = send_signal_locked(sig, info, p, type);
1302 unlock_task_sighand(p, &flags);
1303 }
1304
1305 return ret;
1306 }
1307
1308 enum sig_handler {
1309 HANDLER_CURRENT, /* If reachable use the current handler */
1310 HANDLER_SIG_DFL, /* Always use SIG_DFL handler semantics */
1311 HANDLER_EXIT, /* Only visible as the process exit code */
1312 };
1313
1314 /*
1315 * Force a signal that the process can't ignore: if necessary
1316 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1317 *
1318 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1319 * since we do not want to have a signal handler that was blocked
1320 * be invoked when user space had explicitly blocked it.
1321 *
1322 * We don't want to have recursive SIGSEGV's etc, for example,
1323 * that is why we also clear SIGNAL_UNKILLABLE.
1324 */
1325 static int
force_sig_info_to_task(struct kernel_siginfo * info,struct task_struct * t,enum sig_handler handler)1326 force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t,
1327 enum sig_handler handler)
1328 {
1329 unsigned long int flags;
1330 int ret, blocked, ignored;
1331 struct k_sigaction *action;
1332 int sig = info->si_signo;
1333
1334 spin_lock_irqsave(&t->sighand->siglock, flags);
1335 action = &t->sighand->action[sig-1];
1336 ignored = action->sa.sa_handler == SIG_IGN;
1337 blocked = sigismember(&t->blocked, sig);
1338 if (blocked || ignored || (handler != HANDLER_CURRENT)) {
1339 action->sa.sa_handler = SIG_DFL;
1340 if (handler == HANDLER_EXIT)
1341 action->sa.sa_flags |= SA_IMMUTABLE;
1342 if (blocked)
1343 sigdelset(&t->blocked, sig);
1344 }
1345 /*
1346 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1347 * debugging to leave init killable. But HANDLER_EXIT is always fatal.
1348 */
1349 if (action->sa.sa_handler == SIG_DFL &&
1350 (!t->ptrace || (handler == HANDLER_EXIT)))
1351 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1352 ret = send_signal_locked(sig, info, t, PIDTYPE_PID);
1353 /* This can happen if the signal was already pending and blocked */
1354 if (!task_sigpending(t))
1355 signal_wake_up(t, 0);
1356 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1357
1358 return ret;
1359 }
1360
force_sig_info(struct kernel_siginfo * info)1361 int force_sig_info(struct kernel_siginfo *info)
1362 {
1363 return force_sig_info_to_task(info, current, HANDLER_CURRENT);
1364 }
1365
1366 /*
1367 * Nuke all other threads in the group.
1368 */
zap_other_threads(struct task_struct * p)1369 int zap_other_threads(struct task_struct *p)
1370 {
1371 struct task_struct *t;
1372 int count = 0;
1373
1374 p->signal->group_stop_count = 0;
1375
1376 for_other_threads(p, t) {
1377 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1378 count++;
1379
1380 /* Don't bother with already dead threads */
1381 if (t->exit_state)
1382 continue;
1383 sigaddset(&t->pending.signal, SIGKILL);
1384 signal_wake_up(t, 1);
1385 }
1386
1387 return count;
1388 }
1389
__lock_task_sighand(struct task_struct * tsk,unsigned long * flags)1390 struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1391 unsigned long *flags)
1392 {
1393 struct sighand_struct *sighand;
1394
1395 rcu_read_lock();
1396 for (;;) {
1397 sighand = rcu_dereference(tsk->sighand);
1398 if (unlikely(sighand == NULL))
1399 break;
1400
1401 /*
1402 * This sighand can be already freed and even reused, but
1403 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1404 * initializes ->siglock: this slab can't go away, it has
1405 * the same object type, ->siglock can't be reinitialized.
1406 *
1407 * We need to ensure that tsk->sighand is still the same
1408 * after we take the lock, we can race with de_thread() or
1409 * __exit_signal(). In the latter case the next iteration
1410 * must see ->sighand == NULL.
1411 */
1412 spin_lock_irqsave(&sighand->siglock, *flags);
1413 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1414 break;
1415 spin_unlock_irqrestore(&sighand->siglock, *flags);
1416 }
1417 rcu_read_unlock();
1418
1419 return sighand;
1420 }
1421
1422 #ifdef CONFIG_LOCKDEP
lockdep_assert_task_sighand_held(struct task_struct * task)1423 void lockdep_assert_task_sighand_held(struct task_struct *task)
1424 {
1425 struct sighand_struct *sighand;
1426
1427 rcu_read_lock();
1428 sighand = rcu_dereference(task->sighand);
1429 if (sighand)
1430 lockdep_assert_held(&sighand->siglock);
1431 else
1432 WARN_ON_ONCE(1);
1433 rcu_read_unlock();
1434 }
1435 #endif
1436
1437 /*
1438 * send signal info to all the members of a thread group or to the
1439 * individual thread if type == PIDTYPE_PID.
1440 */
group_send_sig_info(int sig,struct kernel_siginfo * info,struct task_struct * p,enum pid_type type)1441 int group_send_sig_info(int sig, struct kernel_siginfo *info,
1442 struct task_struct *p, enum pid_type type)
1443 {
1444 int ret;
1445
1446 rcu_read_lock();
1447 ret = check_kill_permission(sig, info, p);
1448 rcu_read_unlock();
1449
1450 if (!ret && sig)
1451 ret = do_send_sig_info(sig, info, p, type);
1452
1453 return ret;
1454 }
1455
1456 /*
1457 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1458 * control characters do (^C, ^Z etc)
1459 * - the caller must hold at least a readlock on tasklist_lock
1460 */
__kill_pgrp_info(int sig,struct kernel_siginfo * info,struct pid * pgrp)1461 int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1462 {
1463 struct task_struct *p = NULL;
1464 int ret = -ESRCH;
1465
1466 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1467 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1468 /*
1469 * If group_send_sig_info() succeeds at least once ret
1470 * becomes 0 and after that the code below has no effect.
1471 * Otherwise we return the last err or -ESRCH if this
1472 * process group is empty.
1473 */
1474 if (ret)
1475 ret = err;
1476 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1477
1478 return ret;
1479 }
1480
kill_pid_info_type(int sig,struct kernel_siginfo * info,struct pid * pid,enum pid_type type)1481 static int kill_pid_info_type(int sig, struct kernel_siginfo *info,
1482 struct pid *pid, enum pid_type type)
1483 {
1484 int error = -ESRCH;
1485 struct task_struct *p;
1486
1487 for (;;) {
1488 rcu_read_lock();
1489 p = pid_task(pid, PIDTYPE_PID);
1490 if (p)
1491 error = group_send_sig_info(sig, info, p, type);
1492 rcu_read_unlock();
1493 if (likely(!p || error != -ESRCH))
1494 return error;
1495 /*
1496 * The task was unhashed in between, try again. If it
1497 * is dead, pid_task() will return NULL, if we race with
1498 * de_thread() it will find the new leader.
1499 */
1500 }
1501 }
1502
kill_pid_info(int sig,struct kernel_siginfo * info,struct pid * pid)1503 int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1504 {
1505 return kill_pid_info_type(sig, info, pid, PIDTYPE_TGID);
1506 }
1507
kill_proc_info(int sig,struct kernel_siginfo * info,pid_t pid)1508 static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1509 {
1510 int error;
1511 rcu_read_lock();
1512 error = kill_pid_info(sig, info, find_vpid(pid));
1513 rcu_read_unlock();
1514 return error;
1515 }
1516
kill_as_cred_perm(const struct cred * cred,struct task_struct * target)1517 static inline bool kill_as_cred_perm(const struct cred *cred,
1518 struct task_struct *target)
1519 {
1520 const struct cred *pcred = __task_cred(target);
1521
1522 return uid_eq(cred->euid, pcred->suid) ||
1523 uid_eq(cred->euid, pcred->uid) ||
1524 uid_eq(cred->uid, pcred->suid) ||
1525 uid_eq(cred->uid, pcred->uid);
1526 }
1527
1528 /*
1529 * The usb asyncio usage of siginfo is wrong. The glibc support
1530 * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1531 * AKA after the generic fields:
1532 * kernel_pid_t si_pid;
1533 * kernel_uid32_t si_uid;
1534 * sigval_t si_value;
1535 *
1536 * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1537 * after the generic fields is:
1538 * void __user *si_addr;
1539 *
1540 * This is a practical problem when there is a 64bit big endian kernel
1541 * and a 32bit userspace. As the 32bit address will encoded in the low
1542 * 32bits of the pointer. Those low 32bits will be stored at higher
1543 * address than appear in a 32 bit pointer. So userspace will not
1544 * see the address it was expecting for it's completions.
1545 *
1546 * There is nothing in the encoding that can allow
1547 * copy_siginfo_to_user32 to detect this confusion of formats, so
1548 * handle this by requiring the caller of kill_pid_usb_asyncio to
1549 * notice when this situration takes place and to store the 32bit
1550 * pointer in sival_int, instead of sival_addr of the sigval_t addr
1551 * parameter.
1552 */
kill_pid_usb_asyncio(int sig,int errno,sigval_t addr,struct pid * pid,const struct cred * cred)1553 int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1554 struct pid *pid, const struct cred *cred)
1555 {
1556 struct kernel_siginfo info;
1557 struct task_struct *p;
1558 unsigned long flags;
1559 int ret = -EINVAL;
1560
1561 if (!valid_signal(sig))
1562 return ret;
1563
1564 clear_siginfo(&info);
1565 info.si_signo = sig;
1566 info.si_errno = errno;
1567 info.si_code = SI_ASYNCIO;
1568 *((sigval_t *)&info.si_pid) = addr;
1569
1570 rcu_read_lock();
1571 p = pid_task(pid, PIDTYPE_PID);
1572 if (!p) {
1573 ret = -ESRCH;
1574 goto out_unlock;
1575 }
1576 if (!kill_as_cred_perm(cred, p)) {
1577 ret = -EPERM;
1578 goto out_unlock;
1579 }
1580 ret = security_task_kill(p, &info, sig, cred);
1581 if (ret)
1582 goto out_unlock;
1583
1584 if (sig) {
1585 if (lock_task_sighand(p, &flags)) {
1586 ret = __send_signal_locked(sig, &info, p, PIDTYPE_TGID, false);
1587 unlock_task_sighand(p, &flags);
1588 } else
1589 ret = -ESRCH;
1590 }
1591 out_unlock:
1592 rcu_read_unlock();
1593 return ret;
1594 }
1595 EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1596
1597 /*
1598 * kill_something_info() interprets pid in interesting ways just like kill(2).
1599 *
1600 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1601 * is probably wrong. Should make it like BSD or SYSV.
1602 */
1603
kill_something_info(int sig,struct kernel_siginfo * info,pid_t pid)1604 static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1605 {
1606 int ret;
1607
1608 if (pid > 0)
1609 return kill_proc_info(sig, info, pid);
1610
1611 /* -INT_MIN is undefined. Exclude this case to avoid a UBSAN warning */
1612 if (pid == INT_MIN)
1613 return -ESRCH;
1614
1615 read_lock(&tasklist_lock);
1616 if (pid != -1) {
1617 ret = __kill_pgrp_info(sig, info,
1618 pid ? find_vpid(-pid) : task_pgrp(current));
1619 } else {
1620 int retval = 0, count = 0;
1621 struct task_struct * p;
1622
1623 for_each_process(p) {
1624 if (task_pid_vnr(p) > 1 &&
1625 !same_thread_group(p, current)) {
1626 int err = group_send_sig_info(sig, info, p,
1627 PIDTYPE_MAX);
1628 ++count;
1629 if (err != -EPERM)
1630 retval = err;
1631 }
1632 }
1633 ret = count ? retval : -ESRCH;
1634 }
1635 read_unlock(&tasklist_lock);
1636
1637 return ret;
1638 }
1639
1640 /*
1641 * These are for backward compatibility with the rest of the kernel source.
1642 */
1643
send_sig_info(int sig,struct kernel_siginfo * info,struct task_struct * p)1644 int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1645 {
1646 /*
1647 * Make sure legacy kernel users don't send in bad values
1648 * (normal paths check this in check_kill_permission).
1649 */
1650 if (!valid_signal(sig))
1651 return -EINVAL;
1652
1653 return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1654 }
1655 EXPORT_SYMBOL(send_sig_info);
1656
1657 #define __si_special(priv) \
1658 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1659
1660 int
send_sig(int sig,struct task_struct * p,int priv)1661 send_sig(int sig, struct task_struct *p, int priv)
1662 {
1663 return send_sig_info(sig, __si_special(priv), p);
1664 }
1665 EXPORT_SYMBOL(send_sig);
1666
force_sig(int sig)1667 void force_sig(int sig)
1668 {
1669 struct kernel_siginfo info;
1670
1671 clear_siginfo(&info);
1672 info.si_signo = sig;
1673 info.si_errno = 0;
1674 info.si_code = SI_KERNEL;
1675 info.si_pid = 0;
1676 info.si_uid = 0;
1677 force_sig_info(&info);
1678 }
1679 EXPORT_SYMBOL(force_sig);
1680
force_fatal_sig(int sig)1681 void force_fatal_sig(int sig)
1682 {
1683 struct kernel_siginfo info;
1684
1685 clear_siginfo(&info);
1686 info.si_signo = sig;
1687 info.si_errno = 0;
1688 info.si_code = SI_KERNEL;
1689 info.si_pid = 0;
1690 info.si_uid = 0;
1691 force_sig_info_to_task(&info, current, HANDLER_SIG_DFL);
1692 }
1693
force_exit_sig(int sig)1694 void force_exit_sig(int sig)
1695 {
1696 struct kernel_siginfo info;
1697
1698 clear_siginfo(&info);
1699 info.si_signo = sig;
1700 info.si_errno = 0;
1701 info.si_code = SI_KERNEL;
1702 info.si_pid = 0;
1703 info.si_uid = 0;
1704 force_sig_info_to_task(&info, current, HANDLER_EXIT);
1705 }
1706
1707 /*
1708 * When things go south during signal handling, we
1709 * will force a SIGSEGV. And if the signal that caused
1710 * the problem was already a SIGSEGV, we'll want to
1711 * make sure we don't even try to deliver the signal..
1712 */
force_sigsegv(int sig)1713 void force_sigsegv(int sig)
1714 {
1715 if (sig == SIGSEGV)
1716 force_fatal_sig(SIGSEGV);
1717 else
1718 force_sig(SIGSEGV);
1719 }
1720
force_sig_fault_to_task(int sig,int code,void __user * addr,struct task_struct * t)1721 int force_sig_fault_to_task(int sig, int code, void __user *addr,
1722 struct task_struct *t)
1723 {
1724 struct kernel_siginfo info;
1725
1726 clear_siginfo(&info);
1727 info.si_signo = sig;
1728 info.si_errno = 0;
1729 info.si_code = code;
1730 info.si_addr = addr;
1731 return force_sig_info_to_task(&info, t, HANDLER_CURRENT);
1732 }
1733
force_sig_fault(int sig,int code,void __user * addr)1734 int force_sig_fault(int sig, int code, void __user *addr)
1735 {
1736 return force_sig_fault_to_task(sig, code, addr, current);
1737 }
1738
send_sig_fault(int sig,int code,void __user * addr,struct task_struct * t)1739 int send_sig_fault(int sig, int code, void __user *addr, struct task_struct *t)
1740 {
1741 struct kernel_siginfo info;
1742
1743 clear_siginfo(&info);
1744 info.si_signo = sig;
1745 info.si_errno = 0;
1746 info.si_code = code;
1747 info.si_addr = addr;
1748 return send_sig_info(info.si_signo, &info, t);
1749 }
1750
force_sig_mceerr(int code,void __user * addr,short lsb)1751 int force_sig_mceerr(int code, void __user *addr, short lsb)
1752 {
1753 struct kernel_siginfo info;
1754
1755 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1756 clear_siginfo(&info);
1757 info.si_signo = SIGBUS;
1758 info.si_errno = 0;
1759 info.si_code = code;
1760 info.si_addr = addr;
1761 info.si_addr_lsb = lsb;
1762 return force_sig_info(&info);
1763 }
1764
send_sig_mceerr(int code,void __user * addr,short lsb,struct task_struct * t)1765 int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1766 {
1767 struct kernel_siginfo info;
1768
1769 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1770 clear_siginfo(&info);
1771 info.si_signo = SIGBUS;
1772 info.si_errno = 0;
1773 info.si_code = code;
1774 info.si_addr = addr;
1775 info.si_addr_lsb = lsb;
1776 return send_sig_info(info.si_signo, &info, t);
1777 }
1778 EXPORT_SYMBOL(send_sig_mceerr);
1779
force_sig_bnderr(void __user * addr,void __user * lower,void __user * upper)1780 int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1781 {
1782 struct kernel_siginfo info;
1783
1784 clear_siginfo(&info);
1785 info.si_signo = SIGSEGV;
1786 info.si_errno = 0;
1787 info.si_code = SEGV_BNDERR;
1788 info.si_addr = addr;
1789 info.si_lower = lower;
1790 info.si_upper = upper;
1791 return force_sig_info(&info);
1792 }
1793
1794 #ifdef SEGV_PKUERR
force_sig_pkuerr(void __user * addr,u32 pkey)1795 int force_sig_pkuerr(void __user *addr, u32 pkey)
1796 {
1797 struct kernel_siginfo info;
1798
1799 clear_siginfo(&info);
1800 info.si_signo = SIGSEGV;
1801 info.si_errno = 0;
1802 info.si_code = SEGV_PKUERR;
1803 info.si_addr = addr;
1804 info.si_pkey = pkey;
1805 return force_sig_info(&info);
1806 }
1807 #endif
1808
send_sig_perf(void __user * addr,u32 type,u64 sig_data)1809 int send_sig_perf(void __user *addr, u32 type, u64 sig_data)
1810 {
1811 struct kernel_siginfo info;
1812
1813 clear_siginfo(&info);
1814 info.si_signo = SIGTRAP;
1815 info.si_errno = 0;
1816 info.si_code = TRAP_PERF;
1817 info.si_addr = addr;
1818 info.si_perf_data = sig_data;
1819 info.si_perf_type = type;
1820
1821 /*
1822 * Signals generated by perf events should not terminate the whole
1823 * process if SIGTRAP is blocked, however, delivering the signal
1824 * asynchronously is better than not delivering at all. But tell user
1825 * space if the signal was asynchronous, so it can clearly be
1826 * distinguished from normal synchronous ones.
1827 */
1828 info.si_perf_flags = sigismember(¤t->blocked, info.si_signo) ?
1829 TRAP_PERF_FLAG_ASYNC :
1830 0;
1831
1832 return send_sig_info(info.si_signo, &info, current);
1833 }
1834
1835 /**
1836 * force_sig_seccomp - signals the task to allow in-process syscall emulation
1837 * @syscall: syscall number to send to userland
1838 * @reason: filter-supplied reason code to send to userland (via si_errno)
1839 * @force_coredump: true to trigger a coredump
1840 *
1841 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
1842 */
force_sig_seccomp(int syscall,int reason,bool force_coredump)1843 int force_sig_seccomp(int syscall, int reason, bool force_coredump)
1844 {
1845 struct kernel_siginfo info;
1846
1847 clear_siginfo(&info);
1848 info.si_signo = SIGSYS;
1849 info.si_code = SYS_SECCOMP;
1850 info.si_call_addr = (void __user *)KSTK_EIP(current);
1851 info.si_errno = reason;
1852 info.si_arch = syscall_get_arch(current);
1853 info.si_syscall = syscall;
1854 return force_sig_info_to_task(&info, current,
1855 force_coredump ? HANDLER_EXIT : HANDLER_CURRENT);
1856 }
1857
1858 /* For the crazy architectures that include trap information in
1859 * the errno field, instead of an actual errno value.
1860 */
force_sig_ptrace_errno_trap(int errno,void __user * addr)1861 int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1862 {
1863 struct kernel_siginfo info;
1864
1865 clear_siginfo(&info);
1866 info.si_signo = SIGTRAP;
1867 info.si_errno = errno;
1868 info.si_code = TRAP_HWBKPT;
1869 info.si_addr = addr;
1870 return force_sig_info(&info);
1871 }
1872
1873 /* For the rare architectures that include trap information using
1874 * si_trapno.
1875 */
force_sig_fault_trapno(int sig,int code,void __user * addr,int trapno)1876 int force_sig_fault_trapno(int sig, int code, void __user *addr, int trapno)
1877 {
1878 struct kernel_siginfo info;
1879
1880 clear_siginfo(&info);
1881 info.si_signo = sig;
1882 info.si_errno = 0;
1883 info.si_code = code;
1884 info.si_addr = addr;
1885 info.si_trapno = trapno;
1886 return force_sig_info(&info);
1887 }
1888
1889 /* For the rare architectures that include trap information using
1890 * si_trapno.
1891 */
send_sig_fault_trapno(int sig,int code,void __user * addr,int trapno,struct task_struct * t)1892 int send_sig_fault_trapno(int sig, int code, void __user *addr, int trapno,
1893 struct task_struct *t)
1894 {
1895 struct kernel_siginfo info;
1896
1897 clear_siginfo(&info);
1898 info.si_signo = sig;
1899 info.si_errno = 0;
1900 info.si_code = code;
1901 info.si_addr = addr;
1902 info.si_trapno = trapno;
1903 return send_sig_info(info.si_signo, &info, t);
1904 }
1905
kill_pgrp_info(int sig,struct kernel_siginfo * info,struct pid * pgrp)1906 static int kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1907 {
1908 int ret;
1909 read_lock(&tasklist_lock);
1910 ret = __kill_pgrp_info(sig, info, pgrp);
1911 read_unlock(&tasklist_lock);
1912 return ret;
1913 }
1914
kill_pgrp(struct pid * pid,int sig,int priv)1915 int kill_pgrp(struct pid *pid, int sig, int priv)
1916 {
1917 return kill_pgrp_info(sig, __si_special(priv), pid);
1918 }
1919 EXPORT_SYMBOL(kill_pgrp);
1920
kill_pid(struct pid * pid,int sig,int priv)1921 int kill_pid(struct pid *pid, int sig, int priv)
1922 {
1923 return kill_pid_info(sig, __si_special(priv), pid);
1924 }
1925 EXPORT_SYMBOL(kill_pid);
1926
1927 /*
1928 * These functions support sending signals using preallocated sigqueue
1929 * structures. This is needed "because realtime applications cannot
1930 * afford to lose notifications of asynchronous events, like timer
1931 * expirations or I/O completions". In the case of POSIX Timers
1932 * we allocate the sigqueue structure from the timer_create. If this
1933 * allocation fails we are able to report the failure to the application
1934 * with an EAGAIN error.
1935 */
sigqueue_alloc(void)1936 struct sigqueue *sigqueue_alloc(void)
1937 {
1938 return __sigqueue_alloc(-1, current, GFP_KERNEL, 0, SIGQUEUE_PREALLOC);
1939 }
1940
sigqueue_free(struct sigqueue * q)1941 void sigqueue_free(struct sigqueue *q)
1942 {
1943 unsigned long flags;
1944 spinlock_t *lock = ¤t->sighand->siglock;
1945
1946 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1947 /*
1948 * We must hold ->siglock while testing q->list
1949 * to serialize with collect_signal() or with
1950 * __exit_signal()->flush_sigqueue().
1951 */
1952 spin_lock_irqsave(lock, flags);
1953 q->flags &= ~SIGQUEUE_PREALLOC;
1954 /*
1955 * If it is queued it will be freed when dequeued,
1956 * like the "regular" sigqueue.
1957 */
1958 if (!list_empty(&q->list))
1959 q = NULL;
1960 spin_unlock_irqrestore(lock, flags);
1961
1962 if (q)
1963 __sigqueue_free(q);
1964 }
1965
send_sigqueue(struct sigqueue * q,struct pid * pid,enum pid_type type)1966 int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1967 {
1968 int sig = q->info.si_signo;
1969 struct sigpending *pending;
1970 struct task_struct *t;
1971 unsigned long flags;
1972 int ret, result;
1973
1974 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1975
1976 ret = -1;
1977 rcu_read_lock();
1978
1979 /*
1980 * This function is used by POSIX timers to deliver a timer signal.
1981 * Where type is PIDTYPE_PID (such as for timers with SIGEV_THREAD_ID
1982 * set), the signal must be delivered to the specific thread (queues
1983 * into t->pending).
1984 *
1985 * Where type is not PIDTYPE_PID, signals must be delivered to the
1986 * process. In this case, prefer to deliver to current if it is in
1987 * the same thread group as the target process, which avoids
1988 * unnecessarily waking up a potentially idle task.
1989 */
1990 t = pid_task(pid, type);
1991 if (!t)
1992 goto ret;
1993 if (type != PIDTYPE_PID && same_thread_group(t, current))
1994 t = current;
1995 if (!likely(lock_task_sighand(t, &flags)))
1996 goto ret;
1997
1998 ret = 1; /* the signal is ignored */
1999 result = TRACE_SIGNAL_IGNORED;
2000 if (!prepare_signal(sig, t, false))
2001 goto out;
2002
2003 ret = 0;
2004 if (unlikely(!list_empty(&q->list))) {
2005 /*
2006 * If an SI_TIMER entry is already queue just increment
2007 * the overrun count.
2008 */
2009 BUG_ON(q->info.si_code != SI_TIMER);
2010 q->info.si_overrun++;
2011 result = TRACE_SIGNAL_ALREADY_PENDING;
2012 goto out;
2013 }
2014 q->info.si_overrun = 0;
2015
2016 signalfd_notify(t, sig);
2017 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
2018 list_add_tail(&q->list, &pending->list);
2019 sigaddset(&pending->signal, sig);
2020 complete_signal(sig, t, type);
2021 result = TRACE_SIGNAL_DELIVERED;
2022 out:
2023 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
2024 unlock_task_sighand(t, &flags);
2025 ret:
2026 rcu_read_unlock();
2027 return ret;
2028 }
2029
do_notify_pidfd(struct task_struct * task)2030 void do_notify_pidfd(struct task_struct *task)
2031 {
2032 struct pid *pid = task_pid(task);
2033
2034 WARN_ON(task->exit_state == 0);
2035
2036 __wake_up(&pid->wait_pidfd, TASK_NORMAL, 0,
2037 poll_to_key(EPOLLIN | EPOLLRDNORM));
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 */
do_notify_parent(struct task_struct * tsk,int sig)2047 bool 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 * tsk is a group leader and has no threads, wake up the
2064 * non-PIDFD_THREAD waiters.
2065 */
2066 if (thread_group_empty(tsk))
2067 do_notify_pidfd(tsk);
2068
2069 if (sig != SIGCHLD) {
2070 /*
2071 * This is only possible if parent == real_parent.
2072 * Check if it has changed security domain.
2073 */
2074 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
2075 sig = SIGCHLD;
2076 }
2077
2078 clear_siginfo(&info);
2079 info.si_signo = sig;
2080 info.si_errno = 0;
2081 /*
2082 * We are under tasklist_lock here so our parent is tied to
2083 * us and cannot change.
2084 *
2085 * task_active_pid_ns will always return the same pid namespace
2086 * until a task passes through release_task.
2087 *
2088 * write_lock() currently calls preempt_disable() which is the
2089 * same as rcu_read_lock(), but according to Oleg, this is not
2090 * correct to rely on this
2091 */
2092 rcu_read_lock();
2093 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
2094 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
2095 task_uid(tsk));
2096 rcu_read_unlock();
2097
2098 task_cputime(tsk, &utime, &stime);
2099 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
2100 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
2101
2102 info.si_status = tsk->exit_code & 0x7f;
2103 if (tsk->exit_code & 0x80)
2104 info.si_code = CLD_DUMPED;
2105 else if (tsk->exit_code & 0x7f)
2106 info.si_code = CLD_KILLED;
2107 else {
2108 info.si_code = CLD_EXITED;
2109 info.si_status = tsk->exit_code >> 8;
2110 }
2111
2112 psig = tsk->parent->sighand;
2113 spin_lock_irqsave(&psig->siglock, flags);
2114 if (!tsk->ptrace && sig == SIGCHLD &&
2115 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
2116 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
2117 /*
2118 * We are exiting and our parent doesn't care. POSIX.1
2119 * defines special semantics for setting SIGCHLD to SIG_IGN
2120 * or setting the SA_NOCLDWAIT flag: we should be reaped
2121 * automatically and not left for our parent's wait4 call.
2122 * Rather than having the parent do it as a magic kind of
2123 * signal handler, we just set this to tell do_exit that we
2124 * can be cleaned up without becoming a zombie. Note that
2125 * we still call __wake_up_parent in this case, because a
2126 * blocked sys_wait4 might now return -ECHILD.
2127 *
2128 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
2129 * is implementation-defined: we do (if you don't want
2130 * it, just use SIG_IGN instead).
2131 */
2132 autoreap = true;
2133 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2134 sig = 0;
2135 }
2136 /*
2137 * Send with __send_signal as si_pid and si_uid are in the
2138 * parent's namespaces.
2139 */
2140 if (valid_signal(sig) && sig)
2141 __send_signal_locked(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2142 __wake_up_parent(tsk, tsk->parent);
2143 spin_unlock_irqrestore(&psig->siglock, flags);
2144
2145 return autoreap;
2146 }
2147
2148 /**
2149 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2150 * @tsk: task reporting the state change
2151 * @for_ptracer: the notification is for ptracer
2152 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2153 *
2154 * Notify @tsk's parent that the stopped/continued state has changed. If
2155 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2156 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2157 *
2158 * CONTEXT:
2159 * Must be called with tasklist_lock at least read locked.
2160 */
do_notify_parent_cldstop(struct task_struct * tsk,bool for_ptracer,int why)2161 static void do_notify_parent_cldstop(struct task_struct *tsk,
2162 bool for_ptracer, int why)
2163 {
2164 struct kernel_siginfo info;
2165 unsigned long flags;
2166 struct task_struct *parent;
2167 struct sighand_struct *sighand;
2168 u64 utime, stime;
2169
2170 if (for_ptracer) {
2171 parent = tsk->parent;
2172 } else {
2173 tsk = tsk->group_leader;
2174 parent = tsk->real_parent;
2175 }
2176
2177 clear_siginfo(&info);
2178 info.si_signo = SIGCHLD;
2179 info.si_errno = 0;
2180 /*
2181 * see comment in do_notify_parent() about the following 4 lines
2182 */
2183 rcu_read_lock();
2184 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2185 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2186 rcu_read_unlock();
2187
2188 task_cputime(tsk, &utime, &stime);
2189 info.si_utime = nsec_to_clock_t(utime);
2190 info.si_stime = nsec_to_clock_t(stime);
2191
2192 info.si_code = why;
2193 switch (why) {
2194 case CLD_CONTINUED:
2195 info.si_status = SIGCONT;
2196 break;
2197 case CLD_STOPPED:
2198 info.si_status = tsk->signal->group_exit_code & 0x7f;
2199 break;
2200 case CLD_TRAPPED:
2201 info.si_status = tsk->exit_code & 0x7f;
2202 break;
2203 default:
2204 BUG();
2205 }
2206
2207 sighand = parent->sighand;
2208 spin_lock_irqsave(&sighand->siglock, flags);
2209 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2210 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2211 send_signal_locked(SIGCHLD, &info, parent, PIDTYPE_TGID);
2212 /*
2213 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2214 */
2215 __wake_up_parent(tsk, parent);
2216 spin_unlock_irqrestore(&sighand->siglock, flags);
2217 }
2218
2219 /*
2220 * This must be called with current->sighand->siglock held.
2221 *
2222 * This should be the path for all ptrace stops.
2223 * We always set current->last_siginfo while stopped here.
2224 * That makes it a way to test a stopped process for
2225 * being ptrace-stopped vs being job-control-stopped.
2226 *
2227 * Returns the signal the ptracer requested the code resume
2228 * with. If the code did not stop because the tracer is gone,
2229 * the stop signal remains unchanged unless clear_code.
2230 */
ptrace_stop(int exit_code,int why,unsigned long message,kernel_siginfo_t * info)2231 static int ptrace_stop(int exit_code, int why, unsigned long message,
2232 kernel_siginfo_t *info)
2233 __releases(¤t->sighand->siglock)
2234 __acquires(¤t->sighand->siglock)
2235 {
2236 bool gstop_done = false;
2237
2238 if (arch_ptrace_stop_needed()) {
2239 /*
2240 * The arch code has something special to do before a
2241 * ptrace stop. This is allowed to block, e.g. for faults
2242 * on user stack pages. We can't keep the siglock while
2243 * calling arch_ptrace_stop, so we must release it now.
2244 * To preserve proper semantics, we must do this before
2245 * any signal bookkeeping like checking group_stop_count.
2246 */
2247 spin_unlock_irq(¤t->sighand->siglock);
2248 arch_ptrace_stop();
2249 spin_lock_irq(¤t->sighand->siglock);
2250 }
2251
2252 /*
2253 * After this point ptrace_signal_wake_up or signal_wake_up
2254 * will clear TASK_TRACED if ptrace_unlink happens or a fatal
2255 * signal comes in. Handle previous ptrace_unlinks and fatal
2256 * signals here to prevent ptrace_stop sleeping in schedule.
2257 */
2258 if (!current->ptrace || __fatal_signal_pending(current))
2259 return exit_code;
2260
2261 set_special_state(TASK_TRACED);
2262 current->jobctl |= JOBCTL_TRACED;
2263
2264 /*
2265 * We're committing to trapping. TRACED should be visible before
2266 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2267 * Also, transition to TRACED and updates to ->jobctl should be
2268 * atomic with respect to siglock and should be done after the arch
2269 * hook as siglock is released and regrabbed across it.
2270 *
2271 * TRACER TRACEE
2272 *
2273 * ptrace_attach()
2274 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2275 * do_wait()
2276 * set_current_state() smp_wmb();
2277 * ptrace_do_wait()
2278 * wait_task_stopped()
2279 * task_stopped_code()
2280 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2281 */
2282 smp_wmb();
2283
2284 current->ptrace_message = message;
2285 current->last_siginfo = info;
2286 current->exit_code = exit_code;
2287
2288 /*
2289 * If @why is CLD_STOPPED, we're trapping to participate in a group
2290 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2291 * across siglock relocks since INTERRUPT was scheduled, PENDING
2292 * could be clear now. We act as if SIGCONT is received after
2293 * TASK_TRACED is entered - ignore it.
2294 */
2295 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2296 gstop_done = task_participate_group_stop(current);
2297
2298 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2299 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2300 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2301 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2302
2303 /* entering a trap, clear TRAPPING */
2304 task_clear_jobctl_trapping(current);
2305
2306 spin_unlock_irq(¤t->sighand->siglock);
2307 read_lock(&tasklist_lock);
2308 /*
2309 * Notify parents of the stop.
2310 *
2311 * While ptraced, there are two parents - the ptracer and
2312 * the real_parent of the group_leader. The ptracer should
2313 * know about every stop while the real parent is only
2314 * interested in the completion of group stop. The states
2315 * for the two don't interact with each other. Notify
2316 * separately unless they're gonna be duplicates.
2317 */
2318 if (current->ptrace)
2319 do_notify_parent_cldstop(current, true, why);
2320 if (gstop_done && (!current->ptrace || ptrace_reparented(current)))
2321 do_notify_parent_cldstop(current, false, why);
2322
2323 /*
2324 * The previous do_notify_parent_cldstop() invocation woke ptracer.
2325 * One a PREEMPTION kernel this can result in preemption requirement
2326 * which will be fulfilled after read_unlock() and the ptracer will be
2327 * put on the CPU.
2328 * The ptracer is in wait_task_inactive(, __TASK_TRACED) waiting for
2329 * this task wait in schedule(). If this task gets preempted then it
2330 * remains enqueued on the runqueue. The ptracer will observe this and
2331 * then sleep for a delay of one HZ tick. In the meantime this task
2332 * gets scheduled, enters schedule() and will wait for the ptracer.
2333 *
2334 * This preemption point is not bad from a correctness point of
2335 * view but extends the runtime by one HZ tick time due to the
2336 * ptracer's sleep. The preempt-disable section ensures that there
2337 * will be no preemption between unlock and schedule() and so
2338 * improving the performance since the ptracer will observe that
2339 * the tracee is scheduled out once it gets on the CPU.
2340 *
2341 * On PREEMPT_RT locking tasklist_lock does not disable preemption.
2342 * Therefore the task can be preempted after do_notify_parent_cldstop()
2343 * before unlocking tasklist_lock so there is no benefit in doing this.
2344 *
2345 * In fact disabling preemption is harmful on PREEMPT_RT because
2346 * the spinlock_t in cgroup_enter_frozen() must not be acquired
2347 * with preemption disabled due to the 'sleeping' spinlock
2348 * substitution of RT.
2349 */
2350 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2351 preempt_disable();
2352 read_unlock(&tasklist_lock);
2353 cgroup_enter_frozen();
2354 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2355 preempt_enable_no_resched();
2356 schedule();
2357 cgroup_leave_frozen(true);
2358
2359 /*
2360 * We are back. Now reacquire the siglock before touching
2361 * last_siginfo, so that we are sure to have synchronized with
2362 * any signal-sending on another CPU that wants to examine it.
2363 */
2364 spin_lock_irq(¤t->sighand->siglock);
2365 exit_code = current->exit_code;
2366 current->last_siginfo = NULL;
2367 current->ptrace_message = 0;
2368 current->exit_code = 0;
2369
2370 /* LISTENING can be set only during STOP traps, clear it */
2371 current->jobctl &= ~(JOBCTL_LISTENING | JOBCTL_PTRACE_FROZEN);
2372
2373 /*
2374 * Queued signals ignored us while we were stopped for tracing.
2375 * So check for any that we should take before resuming user mode.
2376 * This sets TIF_SIGPENDING, but never clears it.
2377 */
2378 recalc_sigpending_tsk(current);
2379 return exit_code;
2380 }
2381
ptrace_do_notify(int signr,int exit_code,int why,unsigned long message)2382 static int ptrace_do_notify(int signr, int exit_code, int why, unsigned long message)
2383 {
2384 kernel_siginfo_t info;
2385
2386 clear_siginfo(&info);
2387 info.si_signo = signr;
2388 info.si_code = exit_code;
2389 info.si_pid = task_pid_vnr(current);
2390 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2391
2392 /* Let the debugger run. */
2393 return ptrace_stop(exit_code, why, message, &info);
2394 }
2395
ptrace_notify(int exit_code,unsigned long message)2396 int ptrace_notify(int exit_code, unsigned long message)
2397 {
2398 int signr;
2399
2400 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2401 if (unlikely(task_work_pending(current)))
2402 task_work_run();
2403
2404 spin_lock_irq(¤t->sighand->siglock);
2405 signr = ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED, message);
2406 spin_unlock_irq(¤t->sighand->siglock);
2407 return signr;
2408 }
2409
2410 /**
2411 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2412 * @signr: signr causing group stop if initiating
2413 *
2414 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2415 * and participate in it. If already set, participate in the existing
2416 * group stop. If participated in a group stop (and thus slept), %true is
2417 * returned with siglock released.
2418 *
2419 * If ptraced, this function doesn't handle stop itself. Instead,
2420 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2421 * untouched. The caller must ensure that INTERRUPT trap handling takes
2422 * places afterwards.
2423 *
2424 * CONTEXT:
2425 * Must be called with @current->sighand->siglock held, which is released
2426 * on %true return.
2427 *
2428 * RETURNS:
2429 * %false if group stop is already cancelled or ptrace trap is scheduled.
2430 * %true if participated in group stop.
2431 */
do_signal_stop(int signr)2432 static bool do_signal_stop(int signr)
2433 __releases(¤t->sighand->siglock)
2434 {
2435 struct signal_struct *sig = current->signal;
2436
2437 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2438 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2439 struct task_struct *t;
2440
2441 /* signr will be recorded in task->jobctl for retries */
2442 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2443
2444 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2445 unlikely(sig->flags & SIGNAL_GROUP_EXIT) ||
2446 unlikely(sig->group_exec_task))
2447 return false;
2448 /*
2449 * There is no group stop already in progress. We must
2450 * initiate one now.
2451 *
2452 * While ptraced, a task may be resumed while group stop is
2453 * still in effect and then receive a stop signal and
2454 * initiate another group stop. This deviates from the
2455 * usual behavior as two consecutive stop signals can't
2456 * cause two group stops when !ptraced. That is why we
2457 * also check !task_is_stopped(t) below.
2458 *
2459 * The condition can be distinguished by testing whether
2460 * SIGNAL_STOP_STOPPED is already set. Don't generate
2461 * group_exit_code in such case.
2462 *
2463 * This is not necessary for SIGNAL_STOP_CONTINUED because
2464 * an intervening stop signal is required to cause two
2465 * continued events regardless of ptrace.
2466 */
2467 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2468 sig->group_exit_code = signr;
2469
2470 sig->group_stop_count = 0;
2471 if (task_set_jobctl_pending(current, signr | gstop))
2472 sig->group_stop_count++;
2473
2474 for_other_threads(current, t) {
2475 /*
2476 * Setting state to TASK_STOPPED for a group
2477 * stop is always done with the siglock held,
2478 * so this check has no races.
2479 */
2480 if (!task_is_stopped(t) &&
2481 task_set_jobctl_pending(t, signr | gstop)) {
2482 sig->group_stop_count++;
2483 if (likely(!(t->ptrace & PT_SEIZED)))
2484 signal_wake_up(t, 0);
2485 else
2486 ptrace_trap_notify(t);
2487 }
2488 }
2489 }
2490
2491 if (likely(!current->ptrace)) {
2492 int notify = 0;
2493
2494 /*
2495 * If there are no other threads in the group, or if there
2496 * is a group stop in progress and we are the last to stop,
2497 * report to the parent.
2498 */
2499 if (task_participate_group_stop(current))
2500 notify = CLD_STOPPED;
2501
2502 current->jobctl |= JOBCTL_STOPPED;
2503 set_special_state(TASK_STOPPED);
2504 spin_unlock_irq(¤t->sighand->siglock);
2505
2506 /*
2507 * Notify the parent of the group stop completion. Because
2508 * we're not holding either the siglock or tasklist_lock
2509 * here, ptracer may attach inbetween; however, this is for
2510 * group stop and should always be delivered to the real
2511 * parent of the group leader. The new ptracer will get
2512 * its notification when this task transitions into
2513 * TASK_TRACED.
2514 */
2515 if (notify) {
2516 read_lock(&tasklist_lock);
2517 do_notify_parent_cldstop(current, false, notify);
2518 read_unlock(&tasklist_lock);
2519 }
2520
2521 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2522 cgroup_enter_frozen();
2523 schedule();
2524 return true;
2525 } else {
2526 /*
2527 * While ptraced, group stop is handled by STOP trap.
2528 * Schedule it and let the caller deal with it.
2529 */
2530 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2531 return false;
2532 }
2533 }
2534
2535 /**
2536 * do_jobctl_trap - take care of ptrace jobctl traps
2537 *
2538 * When PT_SEIZED, it's used for both group stop and explicit
2539 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2540 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2541 * the stop signal; otherwise, %SIGTRAP.
2542 *
2543 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2544 * number as exit_code and no siginfo.
2545 *
2546 * CONTEXT:
2547 * Must be called with @current->sighand->siglock held, which may be
2548 * released and re-acquired before returning with intervening sleep.
2549 */
do_jobctl_trap(void)2550 static void do_jobctl_trap(void)
2551 {
2552 struct signal_struct *signal = current->signal;
2553 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2554
2555 if (current->ptrace & PT_SEIZED) {
2556 if (!signal->group_stop_count &&
2557 !(signal->flags & SIGNAL_STOP_STOPPED))
2558 signr = SIGTRAP;
2559 WARN_ON_ONCE(!signr);
2560 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2561 CLD_STOPPED, 0);
2562 } else {
2563 WARN_ON_ONCE(!signr);
2564 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2565 }
2566 }
2567
2568 /**
2569 * do_freezer_trap - handle the freezer jobctl trap
2570 *
2571 * Puts the task into frozen state, if only the task is not about to quit.
2572 * In this case it drops JOBCTL_TRAP_FREEZE.
2573 *
2574 * CONTEXT:
2575 * Must be called with @current->sighand->siglock held,
2576 * which is always released before returning.
2577 */
do_freezer_trap(void)2578 static void do_freezer_trap(void)
2579 __releases(¤t->sighand->siglock)
2580 {
2581 /*
2582 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2583 * let's make another loop to give it a chance to be handled.
2584 * In any case, we'll return back.
2585 */
2586 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2587 JOBCTL_TRAP_FREEZE) {
2588 spin_unlock_irq(¤t->sighand->siglock);
2589 return;
2590 }
2591
2592 /*
2593 * Now we're sure that there is no pending fatal signal and no
2594 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2595 * immediately (if there is a non-fatal signal pending), and
2596 * put the task into sleep.
2597 */
2598 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2599 clear_thread_flag(TIF_SIGPENDING);
2600 spin_unlock_irq(¤t->sighand->siglock);
2601 cgroup_enter_frozen();
2602 schedule();
2603 }
2604
ptrace_signal(int signr,kernel_siginfo_t * info,enum pid_type type)2605 static int ptrace_signal(int signr, kernel_siginfo_t *info, enum pid_type type)
2606 {
2607 /*
2608 * We do not check sig_kernel_stop(signr) but set this marker
2609 * unconditionally because we do not know whether debugger will
2610 * change signr. This flag has no meaning unless we are going
2611 * to stop after return from ptrace_stop(). In this case it will
2612 * be checked in do_signal_stop(), we should only stop if it was
2613 * not cleared by SIGCONT while we were sleeping. See also the
2614 * comment in dequeue_signal().
2615 */
2616 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2617 signr = ptrace_stop(signr, CLD_TRAPPED, 0, info);
2618
2619 /* We're back. Did the debugger cancel the sig? */
2620 if (signr == 0)
2621 return signr;
2622
2623 /*
2624 * Update the siginfo structure if the signal has
2625 * changed. If the debugger wanted something
2626 * specific in the siginfo structure then it should
2627 * have updated *info via PTRACE_SETSIGINFO.
2628 */
2629 if (signr != info->si_signo) {
2630 clear_siginfo(info);
2631 info->si_signo = signr;
2632 info->si_errno = 0;
2633 info->si_code = SI_USER;
2634 rcu_read_lock();
2635 info->si_pid = task_pid_vnr(current->parent);
2636 info->si_uid = from_kuid_munged(current_user_ns(),
2637 task_uid(current->parent));
2638 rcu_read_unlock();
2639 }
2640
2641 /* If the (new) signal is now blocked, requeue it. */
2642 if (sigismember(¤t->blocked, signr) ||
2643 fatal_signal_pending(current)) {
2644 send_signal_locked(signr, info, current, type);
2645 signr = 0;
2646 }
2647
2648 return signr;
2649 }
2650
hide_si_addr_tag_bits(struct ksignal * ksig)2651 static void hide_si_addr_tag_bits(struct ksignal *ksig)
2652 {
2653 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2654 case SIL_FAULT:
2655 case SIL_FAULT_TRAPNO:
2656 case SIL_FAULT_MCEERR:
2657 case SIL_FAULT_BNDERR:
2658 case SIL_FAULT_PKUERR:
2659 case SIL_FAULT_PERF_EVENT:
2660 ksig->info.si_addr = arch_untagged_si_addr(
2661 ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2662 break;
2663 case SIL_KILL:
2664 case SIL_TIMER:
2665 case SIL_POLL:
2666 case SIL_CHLD:
2667 case SIL_RT:
2668 case SIL_SYS:
2669 break;
2670 }
2671 }
2672
get_signal(struct ksignal * ksig)2673 bool get_signal(struct ksignal *ksig)
2674 {
2675 struct sighand_struct *sighand = current->sighand;
2676 struct signal_struct *signal = current->signal;
2677 int signr;
2678
2679 clear_notify_signal();
2680 if (unlikely(task_work_pending(current)))
2681 task_work_run();
2682
2683 if (!task_sigpending(current))
2684 return false;
2685
2686 if (unlikely(uprobe_deny_signal()))
2687 return false;
2688
2689 /*
2690 * Do this once, we can't return to user-mode if freezing() == T.
2691 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2692 * thus do not need another check after return.
2693 */
2694 try_to_freeze();
2695
2696 relock:
2697 spin_lock_irq(&sighand->siglock);
2698
2699 /*
2700 * Every stopped thread goes here after wakeup. Check to see if
2701 * we should notify the parent, prepare_signal(SIGCONT) encodes
2702 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2703 */
2704 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2705 int why;
2706
2707 if (signal->flags & SIGNAL_CLD_CONTINUED)
2708 why = CLD_CONTINUED;
2709 else
2710 why = CLD_STOPPED;
2711
2712 signal->flags &= ~SIGNAL_CLD_MASK;
2713
2714 spin_unlock_irq(&sighand->siglock);
2715
2716 /*
2717 * Notify the parent that we're continuing. This event is
2718 * always per-process and doesn't make whole lot of sense
2719 * for ptracers, who shouldn't consume the state via
2720 * wait(2) either, but, for backward compatibility, notify
2721 * the ptracer of the group leader too unless it's gonna be
2722 * a duplicate.
2723 */
2724 read_lock(&tasklist_lock);
2725 do_notify_parent_cldstop(current, false, why);
2726
2727 if (ptrace_reparented(current->group_leader))
2728 do_notify_parent_cldstop(current->group_leader,
2729 true, why);
2730 read_unlock(&tasklist_lock);
2731
2732 goto relock;
2733 }
2734
2735 for (;;) {
2736 struct k_sigaction *ka;
2737 enum pid_type type;
2738
2739 /* Has this task already been marked for death? */
2740 if ((signal->flags & SIGNAL_GROUP_EXIT) ||
2741 signal->group_exec_task) {
2742 signr = SIGKILL;
2743 sigdelset(¤t->pending.signal, SIGKILL);
2744 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2745 &sighand->action[SIGKILL-1]);
2746 recalc_sigpending();
2747 /*
2748 * implies do_group_exit() or return to PF_USER_WORKER,
2749 * no need to initialize ksig->info/etc.
2750 */
2751 goto fatal;
2752 }
2753
2754 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2755 do_signal_stop(0))
2756 goto relock;
2757
2758 if (unlikely(current->jobctl &
2759 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2760 if (current->jobctl & JOBCTL_TRAP_MASK) {
2761 do_jobctl_trap();
2762 spin_unlock_irq(&sighand->siglock);
2763 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2764 do_freezer_trap();
2765
2766 goto relock;
2767 }
2768
2769 /*
2770 * If the task is leaving the frozen state, let's update
2771 * cgroup counters and reset the frozen bit.
2772 */
2773 if (unlikely(cgroup_task_frozen(current))) {
2774 spin_unlock_irq(&sighand->siglock);
2775 cgroup_leave_frozen(false);
2776 goto relock;
2777 }
2778
2779 /*
2780 * Signals generated by the execution of an instruction
2781 * need to be delivered before any other pending signals
2782 * so that the instruction pointer in the signal stack
2783 * frame points to the faulting instruction.
2784 */
2785 type = PIDTYPE_PID;
2786 signr = dequeue_synchronous_signal(&ksig->info);
2787 if (!signr)
2788 signr = dequeue_signal(current, ¤t->blocked,
2789 &ksig->info, &type);
2790
2791 if (!signr)
2792 break; /* will return 0 */
2793
2794 if (unlikely(current->ptrace) && (signr != SIGKILL) &&
2795 !(sighand->action[signr -1].sa.sa_flags & SA_IMMUTABLE)) {
2796 signr = ptrace_signal(signr, &ksig->info, type);
2797 if (!signr)
2798 continue;
2799 }
2800
2801 ka = &sighand->action[signr-1];
2802
2803 /* Trace actually delivered signals. */
2804 trace_signal_deliver(signr, &ksig->info, ka);
2805
2806 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2807 continue;
2808 if (ka->sa.sa_handler != SIG_DFL) {
2809 /* Run the handler. */
2810 ksig->ka = *ka;
2811
2812 if (ka->sa.sa_flags & SA_ONESHOT)
2813 ka->sa.sa_handler = SIG_DFL;
2814
2815 break; /* will return non-zero "signr" value */
2816 }
2817
2818 /*
2819 * Now we are doing the default action for this signal.
2820 */
2821 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2822 continue;
2823
2824 /*
2825 * Global init gets no signals it doesn't want.
2826 * Container-init gets no signals it doesn't want from same
2827 * container.
2828 *
2829 * Note that if global/container-init sees a sig_kernel_only()
2830 * signal here, the signal must have been generated internally
2831 * or must have come from an ancestor namespace. In either
2832 * case, the signal cannot be dropped.
2833 */
2834 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2835 !sig_kernel_only(signr))
2836 continue;
2837
2838 if (sig_kernel_stop(signr)) {
2839 /*
2840 * The default action is to stop all threads in
2841 * the thread group. The job control signals
2842 * do nothing in an orphaned pgrp, but SIGSTOP
2843 * always works. Note that siglock needs to be
2844 * dropped during the call to is_orphaned_pgrp()
2845 * because of lock ordering with tasklist_lock.
2846 * This allows an intervening SIGCONT to be posted.
2847 * We need to check for that and bail out if necessary.
2848 */
2849 if (signr != SIGSTOP) {
2850 spin_unlock_irq(&sighand->siglock);
2851
2852 /* signals can be posted during this window */
2853
2854 if (is_current_pgrp_orphaned())
2855 goto relock;
2856
2857 spin_lock_irq(&sighand->siglock);
2858 }
2859
2860 if (likely(do_signal_stop(signr))) {
2861 /* It released the siglock. */
2862 goto relock;
2863 }
2864
2865 /*
2866 * We didn't actually stop, due to a race
2867 * with SIGCONT or something like that.
2868 */
2869 continue;
2870 }
2871
2872 fatal:
2873 spin_unlock_irq(&sighand->siglock);
2874 if (unlikely(cgroup_task_frozen(current)))
2875 cgroup_leave_frozen(true);
2876
2877 /*
2878 * Anything else is fatal, maybe with a core dump.
2879 */
2880 current->flags |= PF_SIGNALED;
2881
2882 if (sig_kernel_coredump(signr)) {
2883 if (print_fatal_signals)
2884 print_fatal_signal(signr);
2885 proc_coredump_connector(current);
2886 /*
2887 * If it was able to dump core, this kills all
2888 * other threads in the group and synchronizes with
2889 * their demise. If we lost the race with another
2890 * thread getting here, it set group_exit_code
2891 * first and our do_group_exit call below will use
2892 * that value and ignore the one we pass it.
2893 */
2894 do_coredump(&ksig->info);
2895 }
2896
2897 /*
2898 * PF_USER_WORKER threads will catch and exit on fatal signals
2899 * themselves. They have cleanup that must be performed, so we
2900 * cannot call do_exit() on their behalf. Note that ksig won't
2901 * be properly initialized, PF_USER_WORKER's shouldn't use it.
2902 */
2903 if (current->flags & PF_USER_WORKER)
2904 goto out;
2905
2906 /*
2907 * Death signals, no core dump.
2908 */
2909 do_group_exit(signr);
2910 /* NOTREACHED */
2911 }
2912 spin_unlock_irq(&sighand->siglock);
2913
2914 ksig->sig = signr;
2915
2916 if (signr && !(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
2917 hide_si_addr_tag_bits(ksig);
2918 out:
2919 return signr > 0;
2920 }
2921
2922 /**
2923 * signal_delivered - called after signal delivery to update blocked signals
2924 * @ksig: kernel signal struct
2925 * @stepping: nonzero if debugger single-step or block-step in use
2926 *
2927 * This function should be called when a signal has successfully been
2928 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2929 * is always blocked), and the signal itself is blocked unless %SA_NODEFER
2930 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2931 */
signal_delivered(struct ksignal * ksig,int stepping)2932 static void signal_delivered(struct ksignal *ksig, int stepping)
2933 {
2934 sigset_t blocked;
2935
2936 /* A signal was successfully delivered, and the
2937 saved sigmask was stored on the signal frame,
2938 and will be restored by sigreturn. So we can
2939 simply clear the restore sigmask flag. */
2940 clear_restore_sigmask();
2941
2942 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask);
2943 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2944 sigaddset(&blocked, ksig->sig);
2945 set_current_blocked(&blocked);
2946 if (current->sas_ss_flags & SS_AUTODISARM)
2947 sas_ss_reset(current);
2948 if (stepping)
2949 ptrace_notify(SIGTRAP, 0);
2950 }
2951
signal_setup_done(int failed,struct ksignal * ksig,int stepping)2952 void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2953 {
2954 if (failed)
2955 force_sigsegv(ksig->sig);
2956 else
2957 signal_delivered(ksig, stepping);
2958 }
2959
2960 /*
2961 * It could be that complete_signal() picked us to notify about the
2962 * group-wide signal. Other threads should be notified now to take
2963 * the shared signals in @which since we will not.
2964 */
retarget_shared_pending(struct task_struct * tsk,sigset_t * which)2965 static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2966 {
2967 sigset_t retarget;
2968 struct task_struct *t;
2969
2970 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2971 if (sigisemptyset(&retarget))
2972 return;
2973
2974 for_other_threads(tsk, t) {
2975 if (t->flags & PF_EXITING)
2976 continue;
2977
2978 if (!has_pending_signals(&retarget, &t->blocked))
2979 continue;
2980 /* Remove the signals this thread can handle. */
2981 sigandsets(&retarget, &retarget, &t->blocked);
2982
2983 if (!task_sigpending(t))
2984 signal_wake_up(t, 0);
2985
2986 if (sigisemptyset(&retarget))
2987 break;
2988 }
2989 }
2990
exit_signals(struct task_struct * tsk)2991 void exit_signals(struct task_struct *tsk)
2992 {
2993 int group_stop = 0;
2994 sigset_t unblocked;
2995
2996 /*
2997 * @tsk is about to have PF_EXITING set - lock out users which
2998 * expect stable threadgroup.
2999 */
3000 cgroup_threadgroup_change_begin(tsk);
3001
3002 if (thread_group_empty(tsk) || (tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
3003 sched_mm_cid_exit_signals(tsk);
3004 tsk->flags |= PF_EXITING;
3005 cgroup_threadgroup_change_end(tsk);
3006 return;
3007 }
3008
3009 spin_lock_irq(&tsk->sighand->siglock);
3010 /*
3011 * From now this task is not visible for group-wide signals,
3012 * see wants_signal(), do_signal_stop().
3013 */
3014 sched_mm_cid_exit_signals(tsk);
3015 tsk->flags |= PF_EXITING;
3016
3017 cgroup_threadgroup_change_end(tsk);
3018
3019 if (!task_sigpending(tsk))
3020 goto out;
3021
3022 unblocked = tsk->blocked;
3023 signotset(&unblocked);
3024 retarget_shared_pending(tsk, &unblocked);
3025
3026 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
3027 task_participate_group_stop(tsk))
3028 group_stop = CLD_STOPPED;
3029 out:
3030 spin_unlock_irq(&tsk->sighand->siglock);
3031
3032 /*
3033 * If group stop has completed, deliver the notification. This
3034 * should always go to the real parent of the group leader.
3035 */
3036 if (unlikely(group_stop)) {
3037 read_lock(&tasklist_lock);
3038 do_notify_parent_cldstop(tsk, false, group_stop);
3039 read_unlock(&tasklist_lock);
3040 }
3041 }
3042
3043 /*
3044 * System call entry points.
3045 */
3046
3047 /**
3048 * sys_restart_syscall - restart a system call
3049 */
SYSCALL_DEFINE0(restart_syscall)3050 SYSCALL_DEFINE0(restart_syscall)
3051 {
3052 struct restart_block *restart = ¤t->restart_block;
3053 return restart->fn(restart);
3054 }
3055
do_no_restart_syscall(struct restart_block * param)3056 long do_no_restart_syscall(struct restart_block *param)
3057 {
3058 return -EINTR;
3059 }
3060
__set_task_blocked(struct task_struct * tsk,const sigset_t * newset)3061 static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
3062 {
3063 if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
3064 sigset_t newblocked;
3065 /* A set of now blocked but previously unblocked signals. */
3066 sigandnsets(&newblocked, newset, ¤t->blocked);
3067 retarget_shared_pending(tsk, &newblocked);
3068 }
3069 tsk->blocked = *newset;
3070 recalc_sigpending();
3071 }
3072
3073 /**
3074 * set_current_blocked - change current->blocked mask
3075 * @newset: new mask
3076 *
3077 * It is wrong to change ->blocked directly, this helper should be used
3078 * to ensure the process can't miss a shared signal we are going to block.
3079 */
set_current_blocked(sigset_t * newset)3080 void set_current_blocked(sigset_t *newset)
3081 {
3082 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
3083 __set_current_blocked(newset);
3084 }
3085
__set_current_blocked(const sigset_t * newset)3086 void __set_current_blocked(const sigset_t *newset)
3087 {
3088 struct task_struct *tsk = current;
3089
3090 /*
3091 * In case the signal mask hasn't changed, there is nothing we need
3092 * to do. The current->blocked shouldn't be modified by other task.
3093 */
3094 if (sigequalsets(&tsk->blocked, newset))
3095 return;
3096
3097 spin_lock_irq(&tsk->sighand->siglock);
3098 __set_task_blocked(tsk, newset);
3099 spin_unlock_irq(&tsk->sighand->siglock);
3100 }
3101
3102 /*
3103 * This is also useful for kernel threads that want to temporarily
3104 * (or permanently) block certain signals.
3105 *
3106 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
3107 * interface happily blocks "unblockable" signals like SIGKILL
3108 * and friends.
3109 */
sigprocmask(int how,sigset_t * set,sigset_t * oldset)3110 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
3111 {
3112 struct task_struct *tsk = current;
3113 sigset_t newset;
3114
3115 /* Lockless, only current can change ->blocked, never from irq */
3116 if (oldset)
3117 *oldset = tsk->blocked;
3118
3119 switch (how) {
3120 case SIG_BLOCK:
3121 sigorsets(&newset, &tsk->blocked, set);
3122 break;
3123 case SIG_UNBLOCK:
3124 sigandnsets(&newset, &tsk->blocked, set);
3125 break;
3126 case SIG_SETMASK:
3127 newset = *set;
3128 break;
3129 default:
3130 return -EINVAL;
3131 }
3132
3133 __set_current_blocked(&newset);
3134 return 0;
3135 }
3136 EXPORT_SYMBOL(sigprocmask);
3137
3138 /*
3139 * The api helps set app-provided sigmasks.
3140 *
3141 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3142 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3143 *
3144 * Note that it does set_restore_sigmask() in advance, so it must be always
3145 * paired with restore_saved_sigmask_unless() before return from syscall.
3146 */
set_user_sigmask(const sigset_t __user * umask,size_t sigsetsize)3147 int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3148 {
3149 sigset_t kmask;
3150
3151 if (!umask)
3152 return 0;
3153 if (sigsetsize != sizeof(sigset_t))
3154 return -EINVAL;
3155 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3156 return -EFAULT;
3157
3158 set_restore_sigmask();
3159 current->saved_sigmask = current->blocked;
3160 set_current_blocked(&kmask);
3161
3162 return 0;
3163 }
3164
3165 #ifdef CONFIG_COMPAT
set_compat_user_sigmask(const compat_sigset_t __user * umask,size_t sigsetsize)3166 int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3167 size_t sigsetsize)
3168 {
3169 sigset_t kmask;
3170
3171 if (!umask)
3172 return 0;
3173 if (sigsetsize != sizeof(compat_sigset_t))
3174 return -EINVAL;
3175 if (get_compat_sigset(&kmask, umask))
3176 return -EFAULT;
3177
3178 set_restore_sigmask();
3179 current->saved_sigmask = current->blocked;
3180 set_current_blocked(&kmask);
3181
3182 return 0;
3183 }
3184 #endif
3185
3186 /**
3187 * sys_rt_sigprocmask - change the list of currently blocked signals
3188 * @how: whether to add, remove, or set signals
3189 * @nset: stores pending signals
3190 * @oset: previous value of signal mask if non-null
3191 * @sigsetsize: size of sigset_t type
3192 */
SYSCALL_DEFINE4(rt_sigprocmask,int,how,sigset_t __user *,nset,sigset_t __user *,oset,size_t,sigsetsize)3193 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3194 sigset_t __user *, oset, size_t, sigsetsize)
3195 {
3196 sigset_t old_set, new_set;
3197 int error;
3198
3199 /* XXX: Don't preclude handling different sized sigset_t's. */
3200 if (sigsetsize != sizeof(sigset_t))
3201 return -EINVAL;
3202
3203 old_set = current->blocked;
3204
3205 if (nset) {
3206 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3207 return -EFAULT;
3208 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3209
3210 error = sigprocmask(how, &new_set, NULL);
3211 if (error)
3212 return error;
3213 }
3214
3215 if (oset) {
3216 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3217 return -EFAULT;
3218 }
3219
3220 return 0;
3221 }
3222
3223 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(rt_sigprocmask,int,how,compat_sigset_t __user *,nset,compat_sigset_t __user *,oset,compat_size_t,sigsetsize)3224 COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3225 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3226 {
3227 sigset_t old_set = current->blocked;
3228
3229 /* XXX: Don't preclude handling different sized sigset_t's. */
3230 if (sigsetsize != sizeof(sigset_t))
3231 return -EINVAL;
3232
3233 if (nset) {
3234 sigset_t new_set;
3235 int error;
3236 if (get_compat_sigset(&new_set, nset))
3237 return -EFAULT;
3238 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3239
3240 error = sigprocmask(how, &new_set, NULL);
3241 if (error)
3242 return error;
3243 }
3244 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3245 }
3246 #endif
3247
do_sigpending(sigset_t * set)3248 static void do_sigpending(sigset_t *set)
3249 {
3250 spin_lock_irq(¤t->sighand->siglock);
3251 sigorsets(set, ¤t->pending.signal,
3252 ¤t->signal->shared_pending.signal);
3253 spin_unlock_irq(¤t->sighand->siglock);
3254
3255 /* Outside the lock because only this thread touches it. */
3256 sigandsets(set, ¤t->blocked, set);
3257 }
3258
3259 /**
3260 * sys_rt_sigpending - examine a pending signal that has been raised
3261 * while blocked
3262 * @uset: stores pending signals
3263 * @sigsetsize: size of sigset_t type or larger
3264 */
SYSCALL_DEFINE2(rt_sigpending,sigset_t __user *,uset,size_t,sigsetsize)3265 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3266 {
3267 sigset_t set;
3268
3269 if (sigsetsize > sizeof(*uset))
3270 return -EINVAL;
3271
3272 do_sigpending(&set);
3273
3274 if (copy_to_user(uset, &set, sigsetsize))
3275 return -EFAULT;
3276
3277 return 0;
3278 }
3279
3280 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(rt_sigpending,compat_sigset_t __user *,uset,compat_size_t,sigsetsize)3281 COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3282 compat_size_t, sigsetsize)
3283 {
3284 sigset_t set;
3285
3286 if (sigsetsize > sizeof(*uset))
3287 return -EINVAL;
3288
3289 do_sigpending(&set);
3290
3291 return put_compat_sigset(uset, &set, sigsetsize);
3292 }
3293 #endif
3294
3295 static const struct {
3296 unsigned char limit, layout;
3297 } sig_sicodes[] = {
3298 [SIGILL] = { NSIGILL, SIL_FAULT },
3299 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3300 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3301 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3302 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3303 #if defined(SIGEMT)
3304 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3305 #endif
3306 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3307 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3308 [SIGSYS] = { NSIGSYS, SIL_SYS },
3309 };
3310
known_siginfo_layout(unsigned sig,int si_code)3311 static bool known_siginfo_layout(unsigned sig, int si_code)
3312 {
3313 if (si_code == SI_KERNEL)
3314 return true;
3315 else if ((si_code > SI_USER)) {
3316 if (sig_specific_sicodes(sig)) {
3317 if (si_code <= sig_sicodes[sig].limit)
3318 return true;
3319 }
3320 else if (si_code <= NSIGPOLL)
3321 return true;
3322 }
3323 else if (si_code >= SI_DETHREAD)
3324 return true;
3325 else if (si_code == SI_ASYNCNL)
3326 return true;
3327 return false;
3328 }
3329
siginfo_layout(unsigned sig,int si_code)3330 enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3331 {
3332 enum siginfo_layout layout = SIL_KILL;
3333 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3334 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3335 (si_code <= sig_sicodes[sig].limit)) {
3336 layout = sig_sicodes[sig].layout;
3337 /* Handle the exceptions */
3338 if ((sig == SIGBUS) &&
3339 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3340 layout = SIL_FAULT_MCEERR;
3341 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3342 layout = SIL_FAULT_BNDERR;
3343 #ifdef SEGV_PKUERR
3344 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3345 layout = SIL_FAULT_PKUERR;
3346 #endif
3347 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3348 layout = SIL_FAULT_PERF_EVENT;
3349 else if (IS_ENABLED(CONFIG_SPARC) &&
3350 (sig == SIGILL) && (si_code == ILL_ILLTRP))
3351 layout = SIL_FAULT_TRAPNO;
3352 else if (IS_ENABLED(CONFIG_ALPHA) &&
3353 ((sig == SIGFPE) ||
3354 ((sig == SIGTRAP) && (si_code == TRAP_UNK))))
3355 layout = SIL_FAULT_TRAPNO;
3356 }
3357 else if (si_code <= NSIGPOLL)
3358 layout = SIL_POLL;
3359 } else {
3360 if (si_code == SI_TIMER)
3361 layout = SIL_TIMER;
3362 else if (si_code == SI_SIGIO)
3363 layout = SIL_POLL;
3364 else if (si_code < 0)
3365 layout = SIL_RT;
3366 }
3367 return layout;
3368 }
3369
si_expansion(const siginfo_t __user * info)3370 static inline char __user *si_expansion(const siginfo_t __user *info)
3371 {
3372 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3373 }
3374
copy_siginfo_to_user(siginfo_t __user * to,const kernel_siginfo_t * from)3375 int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3376 {
3377 char __user *expansion = si_expansion(to);
3378 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3379 return -EFAULT;
3380 if (clear_user(expansion, SI_EXPANSION_SIZE))
3381 return -EFAULT;
3382 return 0;
3383 }
3384
post_copy_siginfo_from_user(kernel_siginfo_t * info,const siginfo_t __user * from)3385 static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3386 const siginfo_t __user *from)
3387 {
3388 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3389 char __user *expansion = si_expansion(from);
3390 char buf[SI_EXPANSION_SIZE];
3391 int i;
3392 /*
3393 * An unknown si_code might need more than
3394 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3395 * extra bytes are 0. This guarantees copy_siginfo_to_user
3396 * will return this data to userspace exactly.
3397 */
3398 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3399 return -EFAULT;
3400 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3401 if (buf[i] != 0)
3402 return -E2BIG;
3403 }
3404 }
3405 return 0;
3406 }
3407
__copy_siginfo_from_user(int signo,kernel_siginfo_t * to,const siginfo_t __user * from)3408 static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3409 const siginfo_t __user *from)
3410 {
3411 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3412 return -EFAULT;
3413 to->si_signo = signo;
3414 return post_copy_siginfo_from_user(to, from);
3415 }
3416
copy_siginfo_from_user(kernel_siginfo_t * to,const siginfo_t __user * from)3417 int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3418 {
3419 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3420 return -EFAULT;
3421 return post_copy_siginfo_from_user(to, from);
3422 }
3423
3424 #ifdef CONFIG_COMPAT
3425 /**
3426 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3427 * @to: compat siginfo destination
3428 * @from: kernel siginfo source
3429 *
3430 * Note: This function does not work properly for the SIGCHLD on x32, but
3431 * fortunately it doesn't have to. The only valid callers for this function are
3432 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3433 * The latter does not care because SIGCHLD will never cause a coredump.
3434 */
copy_siginfo_to_external32(struct compat_siginfo * to,const struct kernel_siginfo * from)3435 void copy_siginfo_to_external32(struct compat_siginfo *to,
3436 const struct kernel_siginfo *from)
3437 {
3438 memset(to, 0, sizeof(*to));
3439
3440 to->si_signo = from->si_signo;
3441 to->si_errno = from->si_errno;
3442 to->si_code = from->si_code;
3443 switch(siginfo_layout(from->si_signo, from->si_code)) {
3444 case SIL_KILL:
3445 to->si_pid = from->si_pid;
3446 to->si_uid = from->si_uid;
3447 break;
3448 case SIL_TIMER:
3449 to->si_tid = from->si_tid;
3450 to->si_overrun = from->si_overrun;
3451 to->si_int = from->si_int;
3452 break;
3453 case SIL_POLL:
3454 to->si_band = from->si_band;
3455 to->si_fd = from->si_fd;
3456 break;
3457 case SIL_FAULT:
3458 to->si_addr = ptr_to_compat(from->si_addr);
3459 break;
3460 case SIL_FAULT_TRAPNO:
3461 to->si_addr = ptr_to_compat(from->si_addr);
3462 to->si_trapno = from->si_trapno;
3463 break;
3464 case SIL_FAULT_MCEERR:
3465 to->si_addr = ptr_to_compat(from->si_addr);
3466 to->si_addr_lsb = from->si_addr_lsb;
3467 break;
3468 case SIL_FAULT_BNDERR:
3469 to->si_addr = ptr_to_compat(from->si_addr);
3470 to->si_lower = ptr_to_compat(from->si_lower);
3471 to->si_upper = ptr_to_compat(from->si_upper);
3472 break;
3473 case SIL_FAULT_PKUERR:
3474 to->si_addr = ptr_to_compat(from->si_addr);
3475 to->si_pkey = from->si_pkey;
3476 break;
3477 case SIL_FAULT_PERF_EVENT:
3478 to->si_addr = ptr_to_compat(from->si_addr);
3479 to->si_perf_data = from->si_perf_data;
3480 to->si_perf_type = from->si_perf_type;
3481 to->si_perf_flags = from->si_perf_flags;
3482 break;
3483 case SIL_CHLD:
3484 to->si_pid = from->si_pid;
3485 to->si_uid = from->si_uid;
3486 to->si_status = from->si_status;
3487 to->si_utime = from->si_utime;
3488 to->si_stime = from->si_stime;
3489 break;
3490 case SIL_RT:
3491 to->si_pid = from->si_pid;
3492 to->si_uid = from->si_uid;
3493 to->si_int = from->si_int;
3494 break;
3495 case SIL_SYS:
3496 to->si_call_addr = ptr_to_compat(from->si_call_addr);
3497 to->si_syscall = from->si_syscall;
3498 to->si_arch = from->si_arch;
3499 break;
3500 }
3501 }
3502
__copy_siginfo_to_user32(struct compat_siginfo __user * to,const struct kernel_siginfo * from)3503 int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3504 const struct kernel_siginfo *from)
3505 {
3506 struct compat_siginfo new;
3507
3508 copy_siginfo_to_external32(&new, from);
3509 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3510 return -EFAULT;
3511 return 0;
3512 }
3513
post_copy_siginfo_from_user32(kernel_siginfo_t * to,const struct compat_siginfo * from)3514 static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3515 const struct compat_siginfo *from)
3516 {
3517 clear_siginfo(to);
3518 to->si_signo = from->si_signo;
3519 to->si_errno = from->si_errno;
3520 to->si_code = from->si_code;
3521 switch(siginfo_layout(from->si_signo, from->si_code)) {
3522 case SIL_KILL:
3523 to->si_pid = from->si_pid;
3524 to->si_uid = from->si_uid;
3525 break;
3526 case SIL_TIMER:
3527 to->si_tid = from->si_tid;
3528 to->si_overrun = from->si_overrun;
3529 to->si_int = from->si_int;
3530 break;
3531 case SIL_POLL:
3532 to->si_band = from->si_band;
3533 to->si_fd = from->si_fd;
3534 break;
3535 case SIL_FAULT:
3536 to->si_addr = compat_ptr(from->si_addr);
3537 break;
3538 case SIL_FAULT_TRAPNO:
3539 to->si_addr = compat_ptr(from->si_addr);
3540 to->si_trapno = from->si_trapno;
3541 break;
3542 case SIL_FAULT_MCEERR:
3543 to->si_addr = compat_ptr(from->si_addr);
3544 to->si_addr_lsb = from->si_addr_lsb;
3545 break;
3546 case SIL_FAULT_BNDERR:
3547 to->si_addr = compat_ptr(from->si_addr);
3548 to->si_lower = compat_ptr(from->si_lower);
3549 to->si_upper = compat_ptr(from->si_upper);
3550 break;
3551 case SIL_FAULT_PKUERR:
3552 to->si_addr = compat_ptr(from->si_addr);
3553 to->si_pkey = from->si_pkey;
3554 break;
3555 case SIL_FAULT_PERF_EVENT:
3556 to->si_addr = compat_ptr(from->si_addr);
3557 to->si_perf_data = from->si_perf_data;
3558 to->si_perf_type = from->si_perf_type;
3559 to->si_perf_flags = from->si_perf_flags;
3560 break;
3561 case SIL_CHLD:
3562 to->si_pid = from->si_pid;
3563 to->si_uid = from->si_uid;
3564 to->si_status = from->si_status;
3565 #ifdef CONFIG_X86_X32_ABI
3566 if (in_x32_syscall()) {
3567 to->si_utime = from->_sifields._sigchld_x32._utime;
3568 to->si_stime = from->_sifields._sigchld_x32._stime;
3569 } else
3570 #endif
3571 {
3572 to->si_utime = from->si_utime;
3573 to->si_stime = from->si_stime;
3574 }
3575 break;
3576 case SIL_RT:
3577 to->si_pid = from->si_pid;
3578 to->si_uid = from->si_uid;
3579 to->si_int = from->si_int;
3580 break;
3581 case SIL_SYS:
3582 to->si_call_addr = compat_ptr(from->si_call_addr);
3583 to->si_syscall = from->si_syscall;
3584 to->si_arch = from->si_arch;
3585 break;
3586 }
3587 return 0;
3588 }
3589
__copy_siginfo_from_user32(int signo,struct kernel_siginfo * to,const struct compat_siginfo __user * ufrom)3590 static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3591 const struct compat_siginfo __user *ufrom)
3592 {
3593 struct compat_siginfo from;
3594
3595 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3596 return -EFAULT;
3597
3598 from.si_signo = signo;
3599 return post_copy_siginfo_from_user32(to, &from);
3600 }
3601
copy_siginfo_from_user32(struct kernel_siginfo * to,const struct compat_siginfo __user * ufrom)3602 int copy_siginfo_from_user32(struct kernel_siginfo *to,
3603 const struct compat_siginfo __user *ufrom)
3604 {
3605 struct compat_siginfo from;
3606
3607 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3608 return -EFAULT;
3609
3610 return post_copy_siginfo_from_user32(to, &from);
3611 }
3612 #endif /* CONFIG_COMPAT */
3613
3614 /**
3615 * do_sigtimedwait - wait for queued signals specified in @which
3616 * @which: queued signals to wait for
3617 * @info: if non-null, the signal's siginfo is returned here
3618 * @ts: upper bound on process time suspension
3619 */
do_sigtimedwait(const sigset_t * which,kernel_siginfo_t * info,const struct timespec64 * ts)3620 static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3621 const struct timespec64 *ts)
3622 {
3623 ktime_t *to = NULL, timeout = KTIME_MAX;
3624 struct task_struct *tsk = current;
3625 sigset_t mask = *which;
3626 enum pid_type type;
3627 int sig, ret = 0;
3628
3629 if (ts) {
3630 if (!timespec64_valid(ts))
3631 return -EINVAL;
3632 timeout = timespec64_to_ktime(*ts);
3633 to = &timeout;
3634 }
3635
3636 /*
3637 * Invert the set of allowed signals to get those we want to block.
3638 */
3639 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3640 signotset(&mask);
3641
3642 spin_lock_irq(&tsk->sighand->siglock);
3643 sig = dequeue_signal(tsk, &mask, info, &type);
3644 if (!sig && timeout) {
3645 /*
3646 * None ready, temporarily unblock those we're interested
3647 * while we are sleeping in so that we'll be awakened when
3648 * they arrive. Unblocking is always fine, we can avoid
3649 * set_current_blocked().
3650 */
3651 tsk->real_blocked = tsk->blocked;
3652 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3653 recalc_sigpending();
3654 spin_unlock_irq(&tsk->sighand->siglock);
3655
3656 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
3657 ret = schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3658 HRTIMER_MODE_REL);
3659 spin_lock_irq(&tsk->sighand->siglock);
3660 __set_task_blocked(tsk, &tsk->real_blocked);
3661 sigemptyset(&tsk->real_blocked);
3662 sig = dequeue_signal(tsk, &mask, info, &type);
3663 }
3664 spin_unlock_irq(&tsk->sighand->siglock);
3665
3666 if (sig)
3667 return sig;
3668 return ret ? -EINTR : -EAGAIN;
3669 }
3670
3671 /**
3672 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3673 * in @uthese
3674 * @uthese: queued signals to wait for
3675 * @uinfo: if non-null, the signal's siginfo is returned here
3676 * @uts: upper bound on process time suspension
3677 * @sigsetsize: size of sigset_t type
3678 */
SYSCALL_DEFINE4(rt_sigtimedwait,const sigset_t __user *,uthese,siginfo_t __user *,uinfo,const struct __kernel_timespec __user *,uts,size_t,sigsetsize)3679 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3680 siginfo_t __user *, uinfo,
3681 const struct __kernel_timespec __user *, uts,
3682 size_t, sigsetsize)
3683 {
3684 sigset_t these;
3685 struct timespec64 ts;
3686 kernel_siginfo_t info;
3687 int ret;
3688
3689 /* XXX: Don't preclude handling different sized sigset_t's. */
3690 if (sigsetsize != sizeof(sigset_t))
3691 return -EINVAL;
3692
3693 if (copy_from_user(&these, uthese, sizeof(these)))
3694 return -EFAULT;
3695
3696 if (uts) {
3697 if (get_timespec64(&ts, uts))
3698 return -EFAULT;
3699 }
3700
3701 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3702
3703 if (ret > 0 && uinfo) {
3704 if (copy_siginfo_to_user(uinfo, &info))
3705 ret = -EFAULT;
3706 }
3707
3708 return ret;
3709 }
3710
3711 #ifdef CONFIG_COMPAT_32BIT_TIME
SYSCALL_DEFINE4(rt_sigtimedwait_time32,const sigset_t __user *,uthese,siginfo_t __user *,uinfo,const struct old_timespec32 __user *,uts,size_t,sigsetsize)3712 SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3713 siginfo_t __user *, uinfo,
3714 const struct old_timespec32 __user *, uts,
3715 size_t, sigsetsize)
3716 {
3717 sigset_t these;
3718 struct timespec64 ts;
3719 kernel_siginfo_t info;
3720 int ret;
3721
3722 if (sigsetsize != sizeof(sigset_t))
3723 return -EINVAL;
3724
3725 if (copy_from_user(&these, uthese, sizeof(these)))
3726 return -EFAULT;
3727
3728 if (uts) {
3729 if (get_old_timespec32(&ts, uts))
3730 return -EFAULT;
3731 }
3732
3733 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3734
3735 if (ret > 0 && uinfo) {
3736 if (copy_siginfo_to_user(uinfo, &info))
3737 ret = -EFAULT;
3738 }
3739
3740 return ret;
3741 }
3742 #endif
3743
3744 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64,compat_sigset_t __user *,uthese,struct compat_siginfo __user *,uinfo,struct __kernel_timespec __user *,uts,compat_size_t,sigsetsize)3745 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3746 struct compat_siginfo __user *, uinfo,
3747 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3748 {
3749 sigset_t s;
3750 struct timespec64 t;
3751 kernel_siginfo_t info;
3752 long ret;
3753
3754 if (sigsetsize != sizeof(sigset_t))
3755 return -EINVAL;
3756
3757 if (get_compat_sigset(&s, uthese))
3758 return -EFAULT;
3759
3760 if (uts) {
3761 if (get_timespec64(&t, uts))
3762 return -EFAULT;
3763 }
3764
3765 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3766
3767 if (ret > 0 && uinfo) {
3768 if (copy_siginfo_to_user32(uinfo, &info))
3769 ret = -EFAULT;
3770 }
3771
3772 return ret;
3773 }
3774
3775 #ifdef CONFIG_COMPAT_32BIT_TIME
COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32,compat_sigset_t __user *,uthese,struct compat_siginfo __user *,uinfo,struct old_timespec32 __user *,uts,compat_size_t,sigsetsize)3776 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3777 struct compat_siginfo __user *, uinfo,
3778 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3779 {
3780 sigset_t s;
3781 struct timespec64 t;
3782 kernel_siginfo_t info;
3783 long ret;
3784
3785 if (sigsetsize != sizeof(sigset_t))
3786 return -EINVAL;
3787
3788 if (get_compat_sigset(&s, uthese))
3789 return -EFAULT;
3790
3791 if (uts) {
3792 if (get_old_timespec32(&t, uts))
3793 return -EFAULT;
3794 }
3795
3796 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3797
3798 if (ret > 0 && uinfo) {
3799 if (copy_siginfo_to_user32(uinfo, &info))
3800 ret = -EFAULT;
3801 }
3802
3803 return ret;
3804 }
3805 #endif
3806 #endif
3807
prepare_kill_siginfo(int sig,struct kernel_siginfo * info,enum pid_type type)3808 static void prepare_kill_siginfo(int sig, struct kernel_siginfo *info,
3809 enum pid_type type)
3810 {
3811 clear_siginfo(info);
3812 info->si_signo = sig;
3813 info->si_errno = 0;
3814 info->si_code = (type == PIDTYPE_PID) ? SI_TKILL : SI_USER;
3815 info->si_pid = task_tgid_vnr(current);
3816 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3817 }
3818
3819 /**
3820 * sys_kill - send a signal to a process
3821 * @pid: the PID of the process
3822 * @sig: signal to be sent
3823 */
SYSCALL_DEFINE2(kill,pid_t,pid,int,sig)3824 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3825 {
3826 struct kernel_siginfo info;
3827
3828 prepare_kill_siginfo(sig, &info, PIDTYPE_TGID);
3829
3830 return kill_something_info(sig, &info, pid);
3831 }
3832
3833 /*
3834 * Verify that the signaler and signalee either are in the same pid namespace
3835 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3836 * namespace.
3837 */
access_pidfd_pidns(struct pid * pid)3838 static bool access_pidfd_pidns(struct pid *pid)
3839 {
3840 struct pid_namespace *active = task_active_pid_ns(current);
3841 struct pid_namespace *p = ns_of_pid(pid);
3842
3843 for (;;) {
3844 if (!p)
3845 return false;
3846 if (p == active)
3847 break;
3848 p = p->parent;
3849 }
3850
3851 return true;
3852 }
3853
copy_siginfo_from_user_any(kernel_siginfo_t * kinfo,siginfo_t __user * info)3854 static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3855 siginfo_t __user *info)
3856 {
3857 #ifdef CONFIG_COMPAT
3858 /*
3859 * Avoid hooking up compat syscalls and instead handle necessary
3860 * conversions here. Note, this is a stop-gap measure and should not be
3861 * considered a generic solution.
3862 */
3863 if (in_compat_syscall())
3864 return copy_siginfo_from_user32(
3865 kinfo, (struct compat_siginfo __user *)info);
3866 #endif
3867 return copy_siginfo_from_user(kinfo, info);
3868 }
3869
pidfd_to_pid(const struct file * file)3870 static struct pid *pidfd_to_pid(const struct file *file)
3871 {
3872 struct pid *pid;
3873
3874 pid = pidfd_pid(file);
3875 if (!IS_ERR(pid))
3876 return pid;
3877
3878 return tgid_pidfd_to_pid(file);
3879 }
3880
3881 #define PIDFD_SEND_SIGNAL_FLAGS \
3882 (PIDFD_SIGNAL_THREAD | PIDFD_SIGNAL_THREAD_GROUP | \
3883 PIDFD_SIGNAL_PROCESS_GROUP)
3884
3885 /**
3886 * sys_pidfd_send_signal - Signal a process through a pidfd
3887 * @pidfd: file descriptor of the process
3888 * @sig: signal to send
3889 * @info: signal info
3890 * @flags: future flags
3891 *
3892 * Send the signal to the thread group or to the individual thread depending
3893 * on PIDFD_THREAD.
3894 * In the future extension to @flags may be used to override the default scope
3895 * of @pidfd.
3896 *
3897 * Return: 0 on success, negative errno on failure
3898 */
SYSCALL_DEFINE4(pidfd_send_signal,int,pidfd,int,sig,siginfo_t __user *,info,unsigned int,flags)3899 SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3900 siginfo_t __user *, info, unsigned int, flags)
3901 {
3902 int ret;
3903 struct fd f;
3904 struct pid *pid;
3905 kernel_siginfo_t kinfo;
3906 enum pid_type type;
3907
3908 /* Enforce flags be set to 0 until we add an extension. */
3909 if (flags & ~PIDFD_SEND_SIGNAL_FLAGS)
3910 return -EINVAL;
3911
3912 /* Ensure that only a single signal scope determining flag is set. */
3913 if (hweight32(flags & PIDFD_SEND_SIGNAL_FLAGS) > 1)
3914 return -EINVAL;
3915
3916 f = fdget(pidfd);
3917 if (!f.file)
3918 return -EBADF;
3919
3920 /* Is this a pidfd? */
3921 pid = pidfd_to_pid(f.file);
3922 if (IS_ERR(pid)) {
3923 ret = PTR_ERR(pid);
3924 goto err;
3925 }
3926
3927 ret = -EINVAL;
3928 if (!access_pidfd_pidns(pid))
3929 goto err;
3930
3931 switch (flags) {
3932 case 0:
3933 /* Infer scope from the type of pidfd. */
3934 if (f.file->f_flags & PIDFD_THREAD)
3935 type = PIDTYPE_PID;
3936 else
3937 type = PIDTYPE_TGID;
3938 break;
3939 case PIDFD_SIGNAL_THREAD:
3940 type = PIDTYPE_PID;
3941 break;
3942 case PIDFD_SIGNAL_THREAD_GROUP:
3943 type = PIDTYPE_TGID;
3944 break;
3945 case PIDFD_SIGNAL_PROCESS_GROUP:
3946 type = PIDTYPE_PGID;
3947 break;
3948 }
3949
3950 if (info) {
3951 ret = copy_siginfo_from_user_any(&kinfo, info);
3952 if (unlikely(ret))
3953 goto err;
3954
3955 ret = -EINVAL;
3956 if (unlikely(sig != kinfo.si_signo))
3957 goto err;
3958
3959 /* Only allow sending arbitrary signals to yourself. */
3960 ret = -EPERM;
3961 if ((task_pid(current) != pid || type > PIDTYPE_TGID) &&
3962 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3963 goto err;
3964 } else {
3965 prepare_kill_siginfo(sig, &kinfo, type);
3966 }
3967
3968 if (type == PIDTYPE_PGID)
3969 ret = kill_pgrp_info(sig, &kinfo, pid);
3970 else
3971 ret = kill_pid_info_type(sig, &kinfo, pid, type);
3972 err:
3973 fdput(f);
3974 return ret;
3975 }
3976
3977 static int
do_send_specific(pid_t tgid,pid_t pid,int sig,struct kernel_siginfo * info)3978 do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3979 {
3980 struct task_struct *p;
3981 int error = -ESRCH;
3982
3983 rcu_read_lock();
3984 p = find_task_by_vpid(pid);
3985 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3986 error = check_kill_permission(sig, info, p);
3987 /*
3988 * The null signal is a permissions and process existence
3989 * probe. No signal is actually delivered.
3990 */
3991 if (!error && sig) {
3992 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3993 /*
3994 * If lock_task_sighand() failed we pretend the task
3995 * dies after receiving the signal. The window is tiny,
3996 * and the signal is private anyway.
3997 */
3998 if (unlikely(error == -ESRCH))
3999 error = 0;
4000 }
4001 }
4002 rcu_read_unlock();
4003
4004 return error;
4005 }
4006
do_tkill(pid_t tgid,pid_t pid,int sig)4007 static int do_tkill(pid_t tgid, pid_t pid, int sig)
4008 {
4009 struct kernel_siginfo info;
4010
4011 prepare_kill_siginfo(sig, &info, PIDTYPE_PID);
4012
4013 return do_send_specific(tgid, pid, sig, &info);
4014 }
4015
4016 /**
4017 * sys_tgkill - send signal to one specific thread
4018 * @tgid: the thread group ID of the thread
4019 * @pid: the PID of the thread
4020 * @sig: signal to be sent
4021 *
4022 * This syscall also checks the @tgid and returns -ESRCH even if the PID
4023 * exists but it's not belonging to the target process anymore. This
4024 * method solves the problem of threads exiting and PIDs getting reused.
4025 */
SYSCALL_DEFINE3(tgkill,pid_t,tgid,pid_t,pid,int,sig)4026 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
4027 {
4028 /* This is only valid for single tasks */
4029 if (pid <= 0 || tgid <= 0)
4030 return -EINVAL;
4031
4032 return do_tkill(tgid, pid, sig);
4033 }
4034
4035 /**
4036 * sys_tkill - send signal to one specific task
4037 * @pid: the PID of the task
4038 * @sig: signal to be sent
4039 *
4040 * Send a signal to only one task, even if it's a CLONE_THREAD task.
4041 */
SYSCALL_DEFINE2(tkill,pid_t,pid,int,sig)4042 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
4043 {
4044 /* This is only valid for single tasks */
4045 if (pid <= 0)
4046 return -EINVAL;
4047
4048 return do_tkill(0, pid, sig);
4049 }
4050
do_rt_sigqueueinfo(pid_t pid,int sig,kernel_siginfo_t * info)4051 static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
4052 {
4053 /* Not even root can pretend to send signals from the kernel.
4054 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4055 */
4056 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4057 (task_pid_vnr(current) != pid))
4058 return -EPERM;
4059
4060 /* POSIX.1b doesn't mention process groups. */
4061 return kill_proc_info(sig, info, pid);
4062 }
4063
4064 /**
4065 * sys_rt_sigqueueinfo - send signal information to a signal
4066 * @pid: the PID of the thread
4067 * @sig: signal to be sent
4068 * @uinfo: signal info to be sent
4069 */
SYSCALL_DEFINE3(rt_sigqueueinfo,pid_t,pid,int,sig,siginfo_t __user *,uinfo)4070 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
4071 siginfo_t __user *, uinfo)
4072 {
4073 kernel_siginfo_t info;
4074 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4075 if (unlikely(ret))
4076 return ret;
4077 return do_rt_sigqueueinfo(pid, sig, &info);
4078 }
4079
4080 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,compat_pid_t,pid,int,sig,struct compat_siginfo __user *,uinfo)4081 COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
4082 compat_pid_t, pid,
4083 int, sig,
4084 struct compat_siginfo __user *, uinfo)
4085 {
4086 kernel_siginfo_t info;
4087 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4088 if (unlikely(ret))
4089 return ret;
4090 return do_rt_sigqueueinfo(pid, sig, &info);
4091 }
4092 #endif
4093
do_rt_tgsigqueueinfo(pid_t tgid,pid_t pid,int sig,kernel_siginfo_t * info)4094 static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
4095 {
4096 /* This is only valid for single tasks */
4097 if (pid <= 0 || tgid <= 0)
4098 return -EINVAL;
4099
4100 /* Not even root can pretend to send signals from the kernel.
4101 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4102 */
4103 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4104 (task_pid_vnr(current) != pid))
4105 return -EPERM;
4106
4107 return do_send_specific(tgid, pid, sig, info);
4108 }
4109
SYSCALL_DEFINE4(rt_tgsigqueueinfo,pid_t,tgid,pid_t,pid,int,sig,siginfo_t __user *,uinfo)4110 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
4111 siginfo_t __user *, uinfo)
4112 {
4113 kernel_siginfo_t info;
4114 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4115 if (unlikely(ret))
4116 return ret;
4117 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4118 }
4119
4120 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,compat_pid_t,tgid,compat_pid_t,pid,int,sig,struct compat_siginfo __user *,uinfo)4121 COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
4122 compat_pid_t, tgid,
4123 compat_pid_t, pid,
4124 int, sig,
4125 struct compat_siginfo __user *, uinfo)
4126 {
4127 kernel_siginfo_t info;
4128 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4129 if (unlikely(ret))
4130 return ret;
4131 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4132 }
4133 #endif
4134
4135 /*
4136 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
4137 */
kernel_sigaction(int sig,__sighandler_t action)4138 void kernel_sigaction(int sig, __sighandler_t action)
4139 {
4140 spin_lock_irq(¤t->sighand->siglock);
4141 current->sighand->action[sig - 1].sa.sa_handler = action;
4142 if (action == SIG_IGN) {
4143 sigset_t mask;
4144
4145 sigemptyset(&mask);
4146 sigaddset(&mask, sig);
4147
4148 flush_sigqueue_mask(&mask, ¤t->signal->shared_pending);
4149 flush_sigqueue_mask(&mask, ¤t->pending);
4150 recalc_sigpending();
4151 }
4152 spin_unlock_irq(¤t->sighand->siglock);
4153 }
4154 EXPORT_SYMBOL(kernel_sigaction);
4155
sigaction_compat_abi(struct k_sigaction * act,struct k_sigaction * oact)4156 void __weak sigaction_compat_abi(struct k_sigaction *act,
4157 struct k_sigaction *oact)
4158 {
4159 }
4160
do_sigaction(int sig,struct k_sigaction * act,struct k_sigaction * oact)4161 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4162 {
4163 struct task_struct *p = current, *t;
4164 struct k_sigaction *k;
4165 sigset_t mask;
4166
4167 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4168 return -EINVAL;
4169
4170 k = &p->sighand->action[sig-1];
4171
4172 spin_lock_irq(&p->sighand->siglock);
4173 if (k->sa.sa_flags & SA_IMMUTABLE) {
4174 spin_unlock_irq(&p->sighand->siglock);
4175 return -EINVAL;
4176 }
4177 if (oact)
4178 *oact = *k;
4179
4180 /*
4181 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4182 * e.g. by having an architecture use the bit in their uapi.
4183 */
4184 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4185
4186 /*
4187 * Clear unknown flag bits in order to allow userspace to detect missing
4188 * support for flag bits and to allow the kernel to use non-uapi bits
4189 * internally.
4190 */
4191 if (act)
4192 act->sa.sa_flags &= UAPI_SA_FLAGS;
4193 if (oact)
4194 oact->sa.sa_flags &= UAPI_SA_FLAGS;
4195
4196 sigaction_compat_abi(act, oact);
4197
4198 if (act) {
4199 sigdelsetmask(&act->sa.sa_mask,
4200 sigmask(SIGKILL) | sigmask(SIGSTOP));
4201 *k = *act;
4202 /*
4203 * POSIX 3.3.1.3:
4204 * "Setting a signal action to SIG_IGN for a signal that is
4205 * pending shall cause the pending signal to be discarded,
4206 * whether or not it is blocked."
4207 *
4208 * "Setting a signal action to SIG_DFL for a signal that is
4209 * pending and whose default action is to ignore the signal
4210 * (for example, SIGCHLD), shall cause the pending signal to
4211 * be discarded, whether or not it is blocked"
4212 */
4213 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4214 sigemptyset(&mask);
4215 sigaddset(&mask, sig);
4216 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
4217 for_each_thread(p, t)
4218 flush_sigqueue_mask(&mask, &t->pending);
4219 }
4220 }
4221
4222 spin_unlock_irq(&p->sighand->siglock);
4223 return 0;
4224 }
4225
4226 #ifdef CONFIG_DYNAMIC_SIGFRAME
sigaltstack_lock(void)4227 static inline void sigaltstack_lock(void)
4228 __acquires(¤t->sighand->siglock)
4229 {
4230 spin_lock_irq(¤t->sighand->siglock);
4231 }
4232
sigaltstack_unlock(void)4233 static inline void sigaltstack_unlock(void)
4234 __releases(¤t->sighand->siglock)
4235 {
4236 spin_unlock_irq(¤t->sighand->siglock);
4237 }
4238 #else
sigaltstack_lock(void)4239 static inline void sigaltstack_lock(void) { }
sigaltstack_unlock(void)4240 static inline void sigaltstack_unlock(void) { }
4241 #endif
4242
4243 static int
do_sigaltstack(const stack_t * ss,stack_t * oss,unsigned long sp,size_t min_ss_size)4244 do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4245 size_t min_ss_size)
4246 {
4247 struct task_struct *t = current;
4248 int ret = 0;
4249
4250 if (oss) {
4251 memset(oss, 0, sizeof(stack_t));
4252 oss->ss_sp = (void __user *) t->sas_ss_sp;
4253 oss->ss_size = t->sas_ss_size;
4254 oss->ss_flags = sas_ss_flags(sp) |
4255 (current->sas_ss_flags & SS_FLAG_BITS);
4256 }
4257
4258 if (ss) {
4259 void __user *ss_sp = ss->ss_sp;
4260 size_t ss_size = ss->ss_size;
4261 unsigned ss_flags = ss->ss_flags;
4262 int ss_mode;
4263
4264 if (unlikely(on_sig_stack(sp)))
4265 return -EPERM;
4266
4267 ss_mode = ss_flags & ~SS_FLAG_BITS;
4268 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4269 ss_mode != 0))
4270 return -EINVAL;
4271
4272 /*
4273 * Return before taking any locks if no actual
4274 * sigaltstack changes were requested.
4275 */
4276 if (t->sas_ss_sp == (unsigned long)ss_sp &&
4277 t->sas_ss_size == ss_size &&
4278 t->sas_ss_flags == ss_flags)
4279 return 0;
4280
4281 sigaltstack_lock();
4282 if (ss_mode == SS_DISABLE) {
4283 ss_size = 0;
4284 ss_sp = NULL;
4285 } else {
4286 if (unlikely(ss_size < min_ss_size))
4287 ret = -ENOMEM;
4288 if (!sigaltstack_size_valid(ss_size))
4289 ret = -ENOMEM;
4290 }
4291 if (!ret) {
4292 t->sas_ss_sp = (unsigned long) ss_sp;
4293 t->sas_ss_size = ss_size;
4294 t->sas_ss_flags = ss_flags;
4295 }
4296 sigaltstack_unlock();
4297 }
4298 return ret;
4299 }
4300
SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss,stack_t __user *,uoss)4301 SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4302 {
4303 stack_t new, old;
4304 int err;
4305 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4306 return -EFAULT;
4307 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4308 current_user_stack_pointer(),
4309 MINSIGSTKSZ);
4310 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4311 err = -EFAULT;
4312 return err;
4313 }
4314
restore_altstack(const stack_t __user * uss)4315 int restore_altstack(const stack_t __user *uss)
4316 {
4317 stack_t new;
4318 if (copy_from_user(&new, uss, sizeof(stack_t)))
4319 return -EFAULT;
4320 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4321 MINSIGSTKSZ);
4322 /* squash all but EFAULT for now */
4323 return 0;
4324 }
4325
__save_altstack(stack_t __user * uss,unsigned long sp)4326 int __save_altstack(stack_t __user *uss, unsigned long sp)
4327 {
4328 struct task_struct *t = current;
4329 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4330 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4331 __put_user(t->sas_ss_size, &uss->ss_size);
4332 return err;
4333 }
4334
4335 #ifdef CONFIG_COMPAT
do_compat_sigaltstack(const compat_stack_t __user * uss_ptr,compat_stack_t __user * uoss_ptr)4336 static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4337 compat_stack_t __user *uoss_ptr)
4338 {
4339 stack_t uss, uoss;
4340 int ret;
4341
4342 if (uss_ptr) {
4343 compat_stack_t uss32;
4344 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4345 return -EFAULT;
4346 uss.ss_sp = compat_ptr(uss32.ss_sp);
4347 uss.ss_flags = uss32.ss_flags;
4348 uss.ss_size = uss32.ss_size;
4349 }
4350 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4351 compat_user_stack_pointer(),
4352 COMPAT_MINSIGSTKSZ);
4353 if (ret >= 0 && uoss_ptr) {
4354 compat_stack_t old;
4355 memset(&old, 0, sizeof(old));
4356 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4357 old.ss_flags = uoss.ss_flags;
4358 old.ss_size = uoss.ss_size;
4359 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4360 ret = -EFAULT;
4361 }
4362 return ret;
4363 }
4364
COMPAT_SYSCALL_DEFINE2(sigaltstack,const compat_stack_t __user *,uss_ptr,compat_stack_t __user *,uoss_ptr)4365 COMPAT_SYSCALL_DEFINE2(sigaltstack,
4366 const compat_stack_t __user *, uss_ptr,
4367 compat_stack_t __user *, uoss_ptr)
4368 {
4369 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4370 }
4371
compat_restore_altstack(const compat_stack_t __user * uss)4372 int compat_restore_altstack(const compat_stack_t __user *uss)
4373 {
4374 int err = do_compat_sigaltstack(uss, NULL);
4375 /* squash all but -EFAULT for now */
4376 return err == -EFAULT ? err : 0;
4377 }
4378
__compat_save_altstack(compat_stack_t __user * uss,unsigned long sp)4379 int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4380 {
4381 int err;
4382 struct task_struct *t = current;
4383 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4384 &uss->ss_sp) |
4385 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4386 __put_user(t->sas_ss_size, &uss->ss_size);
4387 return err;
4388 }
4389 #endif
4390
4391 #ifdef __ARCH_WANT_SYS_SIGPENDING
4392
4393 /**
4394 * sys_sigpending - examine pending signals
4395 * @uset: where mask of pending signal is returned
4396 */
SYSCALL_DEFINE1(sigpending,old_sigset_t __user *,uset)4397 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4398 {
4399 sigset_t set;
4400
4401 if (sizeof(old_sigset_t) > sizeof(*uset))
4402 return -EINVAL;
4403
4404 do_sigpending(&set);
4405
4406 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4407 return -EFAULT;
4408
4409 return 0;
4410 }
4411
4412 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE1(sigpending,compat_old_sigset_t __user *,set32)4413 COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4414 {
4415 sigset_t set;
4416
4417 do_sigpending(&set);
4418
4419 return put_user(set.sig[0], set32);
4420 }
4421 #endif
4422
4423 #endif
4424
4425 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
4426 /**
4427 * sys_sigprocmask - examine and change blocked signals
4428 * @how: whether to add, remove, or set signals
4429 * @nset: signals to add or remove (if non-null)
4430 * @oset: previous value of signal mask if non-null
4431 *
4432 * Some platforms have their own version with special arguments;
4433 * others support only sys_rt_sigprocmask.
4434 */
4435
SYSCALL_DEFINE3(sigprocmask,int,how,old_sigset_t __user *,nset,old_sigset_t __user *,oset)4436 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4437 old_sigset_t __user *, oset)
4438 {
4439 old_sigset_t old_set, new_set;
4440 sigset_t new_blocked;
4441
4442 old_set = current->blocked.sig[0];
4443
4444 if (nset) {
4445 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4446 return -EFAULT;
4447
4448 new_blocked = current->blocked;
4449
4450 switch (how) {
4451 case SIG_BLOCK:
4452 sigaddsetmask(&new_blocked, new_set);
4453 break;
4454 case SIG_UNBLOCK:
4455 sigdelsetmask(&new_blocked, new_set);
4456 break;
4457 case SIG_SETMASK:
4458 new_blocked.sig[0] = new_set;
4459 break;
4460 default:
4461 return -EINVAL;
4462 }
4463
4464 set_current_blocked(&new_blocked);
4465 }
4466
4467 if (oset) {
4468 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4469 return -EFAULT;
4470 }
4471
4472 return 0;
4473 }
4474 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4475
4476 #ifndef CONFIG_ODD_RT_SIGACTION
4477 /**
4478 * sys_rt_sigaction - alter an action taken by a process
4479 * @sig: signal to be sent
4480 * @act: new sigaction
4481 * @oact: used to save the previous sigaction
4482 * @sigsetsize: size of sigset_t type
4483 */
SYSCALL_DEFINE4(rt_sigaction,int,sig,const struct sigaction __user *,act,struct sigaction __user *,oact,size_t,sigsetsize)4484 SYSCALL_DEFINE4(rt_sigaction, int, sig,
4485 const struct sigaction __user *, act,
4486 struct sigaction __user *, oact,
4487 size_t, sigsetsize)
4488 {
4489 struct k_sigaction new_sa, old_sa;
4490 int ret;
4491
4492 /* XXX: Don't preclude handling different sized sigset_t's. */
4493 if (sigsetsize != sizeof(sigset_t))
4494 return -EINVAL;
4495
4496 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4497 return -EFAULT;
4498
4499 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4500 if (ret)
4501 return ret;
4502
4503 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4504 return -EFAULT;
4505
4506 return 0;
4507 }
4508 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(rt_sigaction,int,sig,const struct compat_sigaction __user *,act,struct compat_sigaction __user *,oact,compat_size_t,sigsetsize)4509 COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4510 const struct compat_sigaction __user *, act,
4511 struct compat_sigaction __user *, oact,
4512 compat_size_t, sigsetsize)
4513 {
4514 struct k_sigaction new_ka, old_ka;
4515 #ifdef __ARCH_HAS_SA_RESTORER
4516 compat_uptr_t restorer;
4517 #endif
4518 int ret;
4519
4520 /* XXX: Don't preclude handling different sized sigset_t's. */
4521 if (sigsetsize != sizeof(compat_sigset_t))
4522 return -EINVAL;
4523
4524 if (act) {
4525 compat_uptr_t handler;
4526 ret = get_user(handler, &act->sa_handler);
4527 new_ka.sa.sa_handler = compat_ptr(handler);
4528 #ifdef __ARCH_HAS_SA_RESTORER
4529 ret |= get_user(restorer, &act->sa_restorer);
4530 new_ka.sa.sa_restorer = compat_ptr(restorer);
4531 #endif
4532 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4533 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4534 if (ret)
4535 return -EFAULT;
4536 }
4537
4538 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4539 if (!ret && oact) {
4540 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4541 &oact->sa_handler);
4542 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4543 sizeof(oact->sa_mask));
4544 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4545 #ifdef __ARCH_HAS_SA_RESTORER
4546 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4547 &oact->sa_restorer);
4548 #endif
4549 }
4550 return ret;
4551 }
4552 #endif
4553 #endif /* !CONFIG_ODD_RT_SIGACTION */
4554
4555 #ifdef CONFIG_OLD_SIGACTION
SYSCALL_DEFINE3(sigaction,int,sig,const struct old_sigaction __user *,act,struct old_sigaction __user *,oact)4556 SYSCALL_DEFINE3(sigaction, int, sig,
4557 const struct old_sigaction __user *, act,
4558 struct old_sigaction __user *, oact)
4559 {
4560 struct k_sigaction new_ka, old_ka;
4561 int ret;
4562
4563 if (act) {
4564 old_sigset_t mask;
4565 if (!access_ok(act, sizeof(*act)) ||
4566 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4567 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4568 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4569 __get_user(mask, &act->sa_mask))
4570 return -EFAULT;
4571 #ifdef __ARCH_HAS_KA_RESTORER
4572 new_ka.ka_restorer = NULL;
4573 #endif
4574 siginitset(&new_ka.sa.sa_mask, mask);
4575 }
4576
4577 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4578
4579 if (!ret && oact) {
4580 if (!access_ok(oact, sizeof(*oact)) ||
4581 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4582 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4583 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4584 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4585 return -EFAULT;
4586 }
4587
4588 return ret;
4589 }
4590 #endif
4591 #ifdef CONFIG_COMPAT_OLD_SIGACTION
COMPAT_SYSCALL_DEFINE3(sigaction,int,sig,const struct compat_old_sigaction __user *,act,struct compat_old_sigaction __user *,oact)4592 COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4593 const struct compat_old_sigaction __user *, act,
4594 struct compat_old_sigaction __user *, oact)
4595 {
4596 struct k_sigaction new_ka, old_ka;
4597 int ret;
4598 compat_old_sigset_t mask;
4599 compat_uptr_t handler, restorer;
4600
4601 if (act) {
4602 if (!access_ok(act, sizeof(*act)) ||
4603 __get_user(handler, &act->sa_handler) ||
4604 __get_user(restorer, &act->sa_restorer) ||
4605 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4606 __get_user(mask, &act->sa_mask))
4607 return -EFAULT;
4608
4609 #ifdef __ARCH_HAS_KA_RESTORER
4610 new_ka.ka_restorer = NULL;
4611 #endif
4612 new_ka.sa.sa_handler = compat_ptr(handler);
4613 new_ka.sa.sa_restorer = compat_ptr(restorer);
4614 siginitset(&new_ka.sa.sa_mask, mask);
4615 }
4616
4617 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4618
4619 if (!ret && oact) {
4620 if (!access_ok(oact, sizeof(*oact)) ||
4621 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4622 &oact->sa_handler) ||
4623 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4624 &oact->sa_restorer) ||
4625 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4626 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4627 return -EFAULT;
4628 }
4629 return ret;
4630 }
4631 #endif
4632
4633 #ifdef CONFIG_SGETMASK_SYSCALL
4634
4635 /*
4636 * For backwards compatibility. Functionality superseded by sigprocmask.
4637 */
SYSCALL_DEFINE0(sgetmask)4638 SYSCALL_DEFINE0(sgetmask)
4639 {
4640 /* SMP safe */
4641 return current->blocked.sig[0];
4642 }
4643
SYSCALL_DEFINE1(ssetmask,int,newmask)4644 SYSCALL_DEFINE1(ssetmask, int, newmask)
4645 {
4646 int old = current->blocked.sig[0];
4647 sigset_t newset;
4648
4649 siginitset(&newset, newmask);
4650 set_current_blocked(&newset);
4651
4652 return old;
4653 }
4654 #endif /* CONFIG_SGETMASK_SYSCALL */
4655
4656 #ifdef __ARCH_WANT_SYS_SIGNAL
4657 /*
4658 * For backwards compatibility. Functionality superseded by sigaction.
4659 */
SYSCALL_DEFINE2(signal,int,sig,__sighandler_t,handler)4660 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4661 {
4662 struct k_sigaction new_sa, old_sa;
4663 int ret;
4664
4665 new_sa.sa.sa_handler = handler;
4666 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4667 sigemptyset(&new_sa.sa.sa_mask);
4668
4669 ret = do_sigaction(sig, &new_sa, &old_sa);
4670
4671 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4672 }
4673 #endif /* __ARCH_WANT_SYS_SIGNAL */
4674
4675 #ifdef __ARCH_WANT_SYS_PAUSE
4676
SYSCALL_DEFINE0(pause)4677 SYSCALL_DEFINE0(pause)
4678 {
4679 while (!signal_pending(current)) {
4680 __set_current_state(TASK_INTERRUPTIBLE);
4681 schedule();
4682 }
4683 return -ERESTARTNOHAND;
4684 }
4685
4686 #endif
4687
sigsuspend(sigset_t * set)4688 static int sigsuspend(sigset_t *set)
4689 {
4690 current->saved_sigmask = current->blocked;
4691 set_current_blocked(set);
4692
4693 while (!signal_pending(current)) {
4694 __set_current_state(TASK_INTERRUPTIBLE);
4695 schedule();
4696 }
4697 set_restore_sigmask();
4698 return -ERESTARTNOHAND;
4699 }
4700
4701 /**
4702 * sys_rt_sigsuspend - replace the signal mask for a value with the
4703 * @unewset value until a signal is received
4704 * @unewset: new signal mask value
4705 * @sigsetsize: size of sigset_t type
4706 */
SYSCALL_DEFINE2(rt_sigsuspend,sigset_t __user *,unewset,size_t,sigsetsize)4707 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4708 {
4709 sigset_t newset;
4710
4711 /* XXX: Don't preclude handling different sized sigset_t's. */
4712 if (sigsetsize != sizeof(sigset_t))
4713 return -EINVAL;
4714
4715 if (copy_from_user(&newset, unewset, sizeof(newset)))
4716 return -EFAULT;
4717 return sigsuspend(&newset);
4718 }
4719
4720 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(rt_sigsuspend,compat_sigset_t __user *,unewset,compat_size_t,sigsetsize)4721 COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4722 {
4723 sigset_t newset;
4724
4725 /* XXX: Don't preclude handling different sized sigset_t's. */
4726 if (sigsetsize != sizeof(sigset_t))
4727 return -EINVAL;
4728
4729 if (get_compat_sigset(&newset, unewset))
4730 return -EFAULT;
4731 return sigsuspend(&newset);
4732 }
4733 #endif
4734
4735 #ifdef CONFIG_OLD_SIGSUSPEND
SYSCALL_DEFINE1(sigsuspend,old_sigset_t,mask)4736 SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4737 {
4738 sigset_t blocked;
4739 siginitset(&blocked, mask);
4740 return sigsuspend(&blocked);
4741 }
4742 #endif
4743 #ifdef CONFIG_OLD_SIGSUSPEND3
SYSCALL_DEFINE3(sigsuspend,int,unused1,int,unused2,old_sigset_t,mask)4744 SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4745 {
4746 sigset_t blocked;
4747 siginitset(&blocked, mask);
4748 return sigsuspend(&blocked);
4749 }
4750 #endif
4751
arch_vma_name(struct vm_area_struct * vma)4752 __weak const char *arch_vma_name(struct vm_area_struct *vma)
4753 {
4754 return NULL;
4755 }
4756
siginfo_buildtime_checks(void)4757 static inline void siginfo_buildtime_checks(void)
4758 {
4759 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4760
4761 /* Verify the offsets in the two siginfos match */
4762 #define CHECK_OFFSET(field) \
4763 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4764
4765 /* kill */
4766 CHECK_OFFSET(si_pid);
4767 CHECK_OFFSET(si_uid);
4768
4769 /* timer */
4770 CHECK_OFFSET(si_tid);
4771 CHECK_OFFSET(si_overrun);
4772 CHECK_OFFSET(si_value);
4773
4774 /* rt */
4775 CHECK_OFFSET(si_pid);
4776 CHECK_OFFSET(si_uid);
4777 CHECK_OFFSET(si_value);
4778
4779 /* sigchld */
4780 CHECK_OFFSET(si_pid);
4781 CHECK_OFFSET(si_uid);
4782 CHECK_OFFSET(si_status);
4783 CHECK_OFFSET(si_utime);
4784 CHECK_OFFSET(si_stime);
4785
4786 /* sigfault */
4787 CHECK_OFFSET(si_addr);
4788 CHECK_OFFSET(si_trapno);
4789 CHECK_OFFSET(si_addr_lsb);
4790 CHECK_OFFSET(si_lower);
4791 CHECK_OFFSET(si_upper);
4792 CHECK_OFFSET(si_pkey);
4793 CHECK_OFFSET(si_perf_data);
4794 CHECK_OFFSET(si_perf_type);
4795 CHECK_OFFSET(si_perf_flags);
4796
4797 /* sigpoll */
4798 CHECK_OFFSET(si_band);
4799 CHECK_OFFSET(si_fd);
4800
4801 /* sigsys */
4802 CHECK_OFFSET(si_call_addr);
4803 CHECK_OFFSET(si_syscall);
4804 CHECK_OFFSET(si_arch);
4805 #undef CHECK_OFFSET
4806
4807 /* usb asyncio */
4808 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4809 offsetof(struct siginfo, si_addr));
4810 if (sizeof(int) == sizeof(void __user *)) {
4811 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4812 sizeof(void __user *));
4813 } else {
4814 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4815 sizeof_field(struct siginfo, si_uid)) !=
4816 sizeof(void __user *));
4817 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4818 offsetof(struct siginfo, si_uid));
4819 }
4820 #ifdef CONFIG_COMPAT
4821 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4822 offsetof(struct compat_siginfo, si_addr));
4823 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4824 sizeof(compat_uptr_t));
4825 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4826 sizeof_field(struct siginfo, si_pid));
4827 #endif
4828 }
4829
4830 #if defined(CONFIG_SYSCTL)
4831 static struct ctl_table signal_debug_table[] = {
4832 #ifdef CONFIG_SYSCTL_EXCEPTION_TRACE
4833 {
4834 .procname = "exception-trace",
4835 .data = &show_unhandled_signals,
4836 .maxlen = sizeof(int),
4837 .mode = 0644,
4838 .proc_handler = proc_dointvec
4839 },
4840 #endif
4841 };
4842
init_signal_sysctls(void)4843 static int __init init_signal_sysctls(void)
4844 {
4845 register_sysctl_init("debug", signal_debug_table);
4846 return 0;
4847 }
4848 early_initcall(init_signal_sysctls);
4849 #endif /* CONFIG_SYSCTL */
4850
signals_init(void)4851 void __init signals_init(void)
4852 {
4853 siginfo_buildtime_checks();
4854
4855 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC | SLAB_ACCOUNT);
4856 }
4857
4858 #ifdef CONFIG_KGDB_KDB
4859 #include <linux/kdb.h>
4860 /*
4861 * kdb_send_sig - Allows kdb to send signals without exposing
4862 * signal internals. This function checks if the required locks are
4863 * available before calling the main signal code, to avoid kdb
4864 * deadlocks.
4865 */
kdb_send_sig(struct task_struct * t,int sig)4866 void kdb_send_sig(struct task_struct *t, int sig)
4867 {
4868 static struct task_struct *kdb_prev_t;
4869 int new_t, ret;
4870 if (!spin_trylock(&t->sighand->siglock)) {
4871 kdb_printf("Can't do kill command now.\n"
4872 "The sigmask lock is held somewhere else in "
4873 "kernel, try again later\n");
4874 return;
4875 }
4876 new_t = kdb_prev_t != t;
4877 kdb_prev_t = t;
4878 if (!task_is_running(t) && new_t) {
4879 spin_unlock(&t->sighand->siglock);
4880 kdb_printf("Process is not RUNNING, sending a signal from "
4881 "kdb risks deadlock\n"
4882 "on the run queue locks. "
4883 "The signal has _not_ been sent.\n"
4884 "Reissue the kill command if you want to risk "
4885 "the deadlock.\n");
4886 return;
4887 }
4888 ret = send_signal_locked(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4889 spin_unlock(&t->sighand->siglock);
4890 if (ret)
4891 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4892 sig, t->pid);
4893 else
4894 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4895 }
4896 #endif /* CONFIG_KGDB_KDB */
4897