1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/kernel/signal.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * 1997-11-02 Modified for POSIX.1b signals by Richard Henderson
8 *
9 * 2003-06-02 Jim Houston - Concurrent Computer Corp.
10 * Changes to use preallocated sigqueue structures
11 * to allow signals to be sent reliably.
12 */
13
14 #include <linux/slab.h>
15 #include <linux/export.h>
16 #include <linux/init.h>
17 #include <linux/sched/mm.h>
18 #include <linux/sched/user.h>
19 #include <linux/sched/debug.h>
20 #include <linux/sched/task.h>
21 #include <linux/sched/task_stack.h>
22 #include <linux/sched/cputime.h>
23 #include <linux/file.h>
24 #include <linux/fs.h>
25 #include <linux/mm.h>
26 #include <linux/proc_fs.h>
27 #include <linux/tty.h>
28 #include <linux/binfmts.h>
29 #include <linux/coredump.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
32 #include <linux/ptrace.h>
33 #include <linux/signal.h>
34 #include <linux/signalfd.h>
35 #include <linux/ratelimit.h>
36 #include <linux/task_work.h>
37 #include <linux/capability.h>
38 #include <linux/freezer.h>
39 #include <linux/pid_namespace.h>
40 #include <linux/nsproxy.h>
41 #include <linux/user_namespace.h>
42 #include <linux/uprobes.h>
43 #include <linux/compat.h>
44 #include <linux/cn_proc.h>
45 #include <linux/compiler.h>
46 #include <linux/posix-timers.h>
47 #include <linux/cgroup.h>
48 #include <linux/audit.h>
49 #include <linux/sysctl.h>
50
51 #define CREATE_TRACE_POINTS
52 #include <trace/events/signal.h>
53
54 #include <asm/param.h>
55 #include <linux/uaccess.h>
56 #include <asm/unistd.h>
57 #include <asm/siginfo.h>
58 #include <asm/cacheflush.h>
59 #include <asm/syscall.h> /* for syscall_get_* */
60
61 /*
62 * SLAB caches for signal bits.
63 */
64
65 static struct kmem_cache *sigqueue_cachep;
66
67 int print_fatal_signals __read_mostly;
68
sig_handler(struct task_struct * t,int sig)69 static void __user *sig_handler(struct task_struct *t, int sig)
70 {
71 return t->sighand->action[sig - 1].sa.sa_handler;
72 }
73
sig_handler_ignored(void __user * handler,int sig)74 static inline bool sig_handler_ignored(void __user *handler, int sig)
75 {
76 /* Is it explicitly or implicitly ignored? */
77 return handler == SIG_IGN ||
78 (handler == SIG_DFL && sig_kernel_ignore(sig));
79 }
80
sig_task_ignored(struct task_struct * t,int sig,bool force)81 static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
82 {
83 void __user *handler;
84
85 handler = sig_handler(t, sig);
86
87 /* SIGKILL and SIGSTOP may not be sent to the global init */
88 if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
89 return true;
90
91 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
92 handler == SIG_DFL && !(force && sig_kernel_only(sig)))
93 return true;
94
95 /* Only allow kernel generated signals to this kthread */
96 if (unlikely((t->flags & PF_KTHREAD) &&
97 (handler == SIG_KTHREAD_KERNEL) && !force))
98 return true;
99
100 return sig_handler_ignored(handler, sig);
101 }
102
sig_ignored(struct task_struct * t,int sig,bool force)103 static bool sig_ignored(struct task_struct *t, int sig, bool force)
104 {
105 /*
106 * Blocked signals are never ignored, since the
107 * signal handler may change by the time it is
108 * unblocked.
109 */
110 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
111 return false;
112
113 /*
114 * Tracers may want to know about even ignored signal unless it
115 * is SIGKILL which can't be reported anyway but can be ignored
116 * by SIGNAL_UNKILLABLE task.
117 */
118 if (t->ptrace && sig != SIGKILL)
119 return false;
120
121 return sig_task_ignored(t, sig, force);
122 }
123
124 /*
125 * Re-calculate pending state from the set of locally pending
126 * signals, globally pending signals, and blocked signals.
127 */
has_pending_signals(sigset_t * signal,sigset_t * blocked)128 static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
129 {
130 unsigned long ready;
131 long i;
132
133 switch (_NSIG_WORDS) {
134 default:
135 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
136 ready |= signal->sig[i] &~ blocked->sig[i];
137 break;
138
139 case 4: ready = signal->sig[3] &~ blocked->sig[3];
140 ready |= signal->sig[2] &~ blocked->sig[2];
141 ready |= signal->sig[1] &~ blocked->sig[1];
142 ready |= signal->sig[0] &~ blocked->sig[0];
143 break;
144
145 case 2: ready = signal->sig[1] &~ blocked->sig[1];
146 ready |= signal->sig[0] &~ blocked->sig[0];
147 break;
148
149 case 1: ready = signal->sig[0] &~ blocked->sig[0];
150 }
151 return ready != 0;
152 }
153
154 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
155
recalc_sigpending_tsk(struct task_struct * t)156 static bool recalc_sigpending_tsk(struct task_struct *t)
157 {
158 if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) ||
159 PENDING(&t->pending, &t->blocked) ||
160 PENDING(&t->signal->shared_pending, &t->blocked) ||
161 cgroup_task_frozen(t)) {
162 set_tsk_thread_flag(t, TIF_SIGPENDING);
163 return true;
164 }
165
166 /*
167 * We must never clear the flag in another thread, or in current
168 * when it's possible the current syscall is returning -ERESTART*.
169 * So we don't clear it here, and only callers who know they should do.
170 */
171 return false;
172 }
173
174 /*
175 * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
176 * This is superfluous when called on current, the wakeup is a harmless no-op.
177 */
recalc_sigpending_and_wake(struct task_struct * t)178 void recalc_sigpending_and_wake(struct task_struct *t)
179 {
180 if (recalc_sigpending_tsk(t))
181 signal_wake_up(t, 0);
182 }
183
recalc_sigpending(void)184 void recalc_sigpending(void)
185 {
186 if (!recalc_sigpending_tsk(current) && !freezing(current))
187 clear_thread_flag(TIF_SIGPENDING);
188
189 }
190 EXPORT_SYMBOL(recalc_sigpending);
191
calculate_sigpending(void)192 void calculate_sigpending(void)
193 {
194 /* Have any signals or users of TIF_SIGPENDING been delayed
195 * until after fork?
196 */
197 spin_lock_irq(¤t->sighand->siglock);
198 set_tsk_thread_flag(current, TIF_SIGPENDING);
199 recalc_sigpending();
200 spin_unlock_irq(¤t->sighand->siglock);
201 }
202
203 /* Given the mask, find the first available signal that should be serviced. */
204
205 #define SYNCHRONOUS_MASK \
206 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
207 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
208
next_signal(struct sigpending * pending,sigset_t * mask)209 int next_signal(struct sigpending *pending, sigset_t *mask)
210 {
211 unsigned long i, *s, *m, x;
212 int sig = 0;
213
214 s = pending->signal.sig;
215 m = mask->sig;
216
217 /*
218 * Handle the first word specially: it contains the
219 * synchronous signals that need to be dequeued first.
220 */
221 x = *s &~ *m;
222 if (x) {
223 if (x & SYNCHRONOUS_MASK)
224 x &= SYNCHRONOUS_MASK;
225 sig = ffz(~x) + 1;
226 return sig;
227 }
228
229 switch (_NSIG_WORDS) {
230 default:
231 for (i = 1; i < _NSIG_WORDS; ++i) {
232 x = *++s &~ *++m;
233 if (!x)
234 continue;
235 sig = ffz(~x) + i*_NSIG_BPW + 1;
236 break;
237 }
238 break;
239
240 case 2:
241 x = s[1] &~ m[1];
242 if (!x)
243 break;
244 sig = ffz(~x) + _NSIG_BPW + 1;
245 break;
246
247 case 1:
248 /* Nothing to do */
249 break;
250 }
251
252 return sig;
253 }
254
print_dropped_signal(int sig)255 static inline void print_dropped_signal(int sig)
256 {
257 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
258
259 if (!print_fatal_signals)
260 return;
261
262 if (!__ratelimit(&ratelimit_state))
263 return;
264
265 pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
266 current->comm, current->pid, sig);
267 }
268
269 /**
270 * task_set_jobctl_pending - set jobctl pending bits
271 * @task: target task
272 * @mask: pending bits to set
273 *
274 * Clear @mask from @task->jobctl. @mask must be subset of
275 * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
276 * %JOBCTL_TRAPPING. If stop signo is being set, the existing signo is
277 * cleared. If @task is already being killed or exiting, this function
278 * becomes noop.
279 *
280 * CONTEXT:
281 * Must be called with @task->sighand->siglock held.
282 *
283 * RETURNS:
284 * %true if @mask is set, %false if made noop because @task was dying.
285 */
task_set_jobctl_pending(struct task_struct * task,unsigned long mask)286 bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
287 {
288 BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
289 JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
290 BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
291
292 if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
293 return false;
294
295 if (mask & JOBCTL_STOP_SIGMASK)
296 task->jobctl &= ~JOBCTL_STOP_SIGMASK;
297
298 task->jobctl |= mask;
299 return true;
300 }
301
302 /**
303 * task_clear_jobctl_trapping - clear jobctl trapping bit
304 * @task: target task
305 *
306 * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
307 * Clear it and wake up the ptracer. Note that we don't need any further
308 * locking. @task->siglock guarantees that @task->parent points to the
309 * ptracer.
310 *
311 * CONTEXT:
312 * Must be called with @task->sighand->siglock held.
313 */
task_clear_jobctl_trapping(struct task_struct * task)314 void task_clear_jobctl_trapping(struct task_struct *task)
315 {
316 if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
317 task->jobctl &= ~JOBCTL_TRAPPING;
318 smp_mb(); /* advised by wake_up_bit() */
319 wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
320 }
321 }
322
323 /**
324 * task_clear_jobctl_pending - clear jobctl pending bits
325 * @task: target task
326 * @mask: pending bits to clear
327 *
328 * Clear @mask from @task->jobctl. @mask must be subset of
329 * %JOBCTL_PENDING_MASK. If %JOBCTL_STOP_PENDING is being cleared, other
330 * STOP bits are cleared together.
331 *
332 * If clearing of @mask leaves no stop or trap pending, this function calls
333 * task_clear_jobctl_trapping().
334 *
335 * CONTEXT:
336 * Must be called with @task->sighand->siglock held.
337 */
task_clear_jobctl_pending(struct task_struct * task,unsigned long mask)338 void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
339 {
340 BUG_ON(mask & ~JOBCTL_PENDING_MASK);
341
342 if (mask & JOBCTL_STOP_PENDING)
343 mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
344
345 task->jobctl &= ~mask;
346
347 if (!(task->jobctl & JOBCTL_PENDING_MASK))
348 task_clear_jobctl_trapping(task);
349 }
350
351 /**
352 * task_participate_group_stop - participate in a group stop
353 * @task: task participating in a group stop
354 *
355 * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
356 * Group stop states are cleared and the group stop count is consumed if
357 * %JOBCTL_STOP_CONSUME was set. If the consumption completes the group
358 * stop, the appropriate `SIGNAL_*` flags are set.
359 *
360 * CONTEXT:
361 * Must be called with @task->sighand->siglock held.
362 *
363 * RETURNS:
364 * %true if group stop completion should be notified to the parent, %false
365 * otherwise.
366 */
task_participate_group_stop(struct task_struct * task)367 static bool task_participate_group_stop(struct task_struct *task)
368 {
369 struct signal_struct *sig = task->signal;
370 bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
371
372 WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
373
374 task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
375
376 if (!consume)
377 return false;
378
379 if (!WARN_ON_ONCE(sig->group_stop_count == 0))
380 sig->group_stop_count--;
381
382 /*
383 * Tell the caller to notify completion iff we are entering into a
384 * fresh group stop. Read comment in do_signal_stop() for details.
385 */
386 if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
387 signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
388 return true;
389 }
390 return false;
391 }
392
task_join_group_stop(struct task_struct * task)393 void task_join_group_stop(struct task_struct *task)
394 {
395 unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK;
396 struct signal_struct *sig = current->signal;
397
398 if (sig->group_stop_count) {
399 sig->group_stop_count++;
400 mask |= JOBCTL_STOP_CONSUME;
401 } else if (!(sig->flags & SIGNAL_STOP_STOPPED))
402 return;
403
404 /* Have the new thread join an on-going signal group stop */
405 task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING);
406 }
407
408 /*
409 * allocate a new signal queue record
410 * - this may be called without locks if and only if t == current, otherwise an
411 * appropriate lock must be held to stop the target task from exiting
412 */
413 static struct sigqueue *
__sigqueue_alloc(int sig,struct task_struct * t,gfp_t gfp_flags,int override_rlimit,const unsigned int sigqueue_flags)414 __sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
415 int override_rlimit, const unsigned int sigqueue_flags)
416 {
417 struct sigqueue *q = NULL;
418 struct ucounts *ucounts = NULL;
419 long sigpending;
420
421 /*
422 * Protect access to @t credentials. This can go away when all
423 * callers hold rcu read lock.
424 *
425 * NOTE! A pending signal will hold on to the user refcount,
426 * and we get/put the refcount only when the sigpending count
427 * changes from/to zero.
428 */
429 rcu_read_lock();
430 ucounts = task_ucounts(t);
431 sigpending = inc_rlimit_get_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
432 rcu_read_unlock();
433 if (!sigpending)
434 return NULL;
435
436 if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
437 q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
438 } else {
439 print_dropped_signal(sig);
440 }
441
442 if (unlikely(q == NULL)) {
443 dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
444 } else {
445 INIT_LIST_HEAD(&q->list);
446 q->flags = sigqueue_flags;
447 q->ucounts = ucounts;
448 }
449 return q;
450 }
451
__sigqueue_free(struct sigqueue * q)452 static void __sigqueue_free(struct sigqueue *q)
453 {
454 if (q->flags & SIGQUEUE_PREALLOC)
455 return;
456 if (q->ucounts) {
457 dec_rlimit_put_ucounts(q->ucounts, UCOUNT_RLIMIT_SIGPENDING);
458 q->ucounts = NULL;
459 }
460 kmem_cache_free(sigqueue_cachep, q);
461 }
462
flush_sigqueue(struct sigpending * queue)463 void flush_sigqueue(struct sigpending *queue)
464 {
465 struct sigqueue *q;
466
467 sigemptyset(&queue->signal);
468 while (!list_empty(&queue->list)) {
469 q = list_entry(queue->list.next, struct sigqueue , list);
470 list_del_init(&q->list);
471 __sigqueue_free(q);
472 }
473 }
474
475 /*
476 * Flush all pending signals for this kthread.
477 */
flush_signals(struct task_struct * t)478 void flush_signals(struct task_struct *t)
479 {
480 unsigned long flags;
481
482 spin_lock_irqsave(&t->sighand->siglock, flags);
483 clear_tsk_thread_flag(t, TIF_SIGPENDING);
484 flush_sigqueue(&t->pending);
485 flush_sigqueue(&t->signal->shared_pending);
486 spin_unlock_irqrestore(&t->sighand->siglock, flags);
487 }
488 EXPORT_SYMBOL(flush_signals);
489
490 #ifdef CONFIG_POSIX_TIMERS
__flush_itimer_signals(struct sigpending * pending)491 static void __flush_itimer_signals(struct sigpending *pending)
492 {
493 sigset_t signal, retain;
494 struct sigqueue *q, *n;
495
496 signal = pending->signal;
497 sigemptyset(&retain);
498
499 list_for_each_entry_safe(q, n, &pending->list, list) {
500 int sig = q->info.si_signo;
501
502 if (likely(q->info.si_code != SI_TIMER)) {
503 sigaddset(&retain, sig);
504 } else {
505 sigdelset(&signal, sig);
506 list_del_init(&q->list);
507 __sigqueue_free(q);
508 }
509 }
510
511 sigorsets(&pending->signal, &signal, &retain);
512 }
513
flush_itimer_signals(void)514 void flush_itimer_signals(void)
515 {
516 struct task_struct *tsk = current;
517 unsigned long flags;
518
519 spin_lock_irqsave(&tsk->sighand->siglock, flags);
520 __flush_itimer_signals(&tsk->pending);
521 __flush_itimer_signals(&tsk->signal->shared_pending);
522 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
523 }
524 #endif
525
ignore_signals(struct task_struct * t)526 void ignore_signals(struct task_struct *t)
527 {
528 int i;
529
530 for (i = 0; i < _NSIG; ++i)
531 t->sighand->action[i].sa.sa_handler = SIG_IGN;
532
533 flush_signals(t);
534 }
535
536 /*
537 * Flush all handlers for a task.
538 */
539
540 void
flush_signal_handlers(struct task_struct * t,int force_default)541 flush_signal_handlers(struct task_struct *t, int force_default)
542 {
543 int i;
544 struct k_sigaction *ka = &t->sighand->action[0];
545 for (i = _NSIG ; i != 0 ; i--) {
546 if (force_default || ka->sa.sa_handler != SIG_IGN)
547 ka->sa.sa_handler = SIG_DFL;
548 ka->sa.sa_flags = 0;
549 #ifdef __ARCH_HAS_SA_RESTORER
550 ka->sa.sa_restorer = NULL;
551 #endif
552 sigemptyset(&ka->sa.sa_mask);
553 ka++;
554 }
555 }
556
unhandled_signal(struct task_struct * tsk,int sig)557 bool unhandled_signal(struct task_struct *tsk, int sig)
558 {
559 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
560 if (is_global_init(tsk))
561 return true;
562
563 if (handler != SIG_IGN && handler != SIG_DFL)
564 return false;
565
566 /* If dying, we handle all new signals by ignoring them */
567 if (fatal_signal_pending(tsk))
568 return false;
569
570 /* if ptraced, let the tracer determine */
571 return !tsk->ptrace;
572 }
573
collect_signal(int sig,struct sigpending * list,kernel_siginfo_t * info,bool * resched_timer)574 static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
575 bool *resched_timer)
576 {
577 struct sigqueue *q, *first = NULL;
578
579 /*
580 * Collect the siginfo appropriate to this signal. Check if
581 * there is another siginfo for the same signal.
582 */
583 list_for_each_entry(q, &list->list, list) {
584 if (q->info.si_signo == sig) {
585 if (first)
586 goto still_pending;
587 first = q;
588 }
589 }
590
591 sigdelset(&list->signal, sig);
592
593 if (first) {
594 still_pending:
595 list_del_init(&first->list);
596 copy_siginfo(info, &first->info);
597
598 *resched_timer =
599 (first->flags & SIGQUEUE_PREALLOC) &&
600 (info->si_code == SI_TIMER) &&
601 (info->si_sys_private);
602
603 __sigqueue_free(first);
604 } else {
605 /*
606 * Ok, it wasn't in the queue. This must be
607 * a fast-pathed signal or we must have been
608 * out of queue space. So zero out the info.
609 */
610 clear_siginfo(info);
611 info->si_signo = sig;
612 info->si_errno = 0;
613 info->si_code = SI_USER;
614 info->si_pid = 0;
615 info->si_uid = 0;
616 }
617 }
618
__dequeue_signal(struct sigpending * pending,sigset_t * mask,kernel_siginfo_t * info,bool * resched_timer)619 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
620 kernel_siginfo_t *info, bool *resched_timer)
621 {
622 int sig = next_signal(pending, mask);
623
624 if (sig)
625 collect_signal(sig, pending, info, resched_timer);
626 return sig;
627 }
628
629 /*
630 * Dequeue a signal and return the element to the caller, which is
631 * expected to free it.
632 *
633 * All callers have to hold the siglock.
634 */
dequeue_signal(struct task_struct * tsk,sigset_t * mask,kernel_siginfo_t * info,enum pid_type * type)635 int dequeue_signal(struct task_struct *tsk, sigset_t *mask,
636 kernel_siginfo_t *info, enum pid_type *type)
637 {
638 bool resched_timer = false;
639 int signr;
640
641 /* We only dequeue private signals from ourselves, we don't let
642 * signalfd steal them
643 */
644 *type = PIDTYPE_PID;
645 signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
646 if (!signr) {
647 *type = PIDTYPE_TGID;
648 signr = __dequeue_signal(&tsk->signal->shared_pending,
649 mask, info, &resched_timer);
650 #ifdef CONFIG_POSIX_TIMERS
651 /*
652 * itimer signal ?
653 *
654 * itimers are process shared and we restart periodic
655 * itimers in the signal delivery path to prevent DoS
656 * attacks in the high resolution timer case. This is
657 * compliant with the old way of self-restarting
658 * itimers, as the SIGALRM is a legacy signal and only
659 * queued once. Changing the restart behaviour to
660 * restart the timer in the signal dequeue path is
661 * reducing the timer noise on heavy loaded !highres
662 * systems too.
663 */
664 if (unlikely(signr == SIGALRM)) {
665 struct hrtimer *tmr = &tsk->signal->real_timer;
666
667 if (!hrtimer_is_queued(tmr) &&
668 tsk->signal->it_real_incr != 0) {
669 hrtimer_forward(tmr, tmr->base->get_time(),
670 tsk->signal->it_real_incr);
671 hrtimer_restart(tmr);
672 }
673 }
674 #endif
675 }
676
677 recalc_sigpending();
678 if (!signr)
679 return 0;
680
681 if (unlikely(sig_kernel_stop(signr))) {
682 /*
683 * Set a marker that we have dequeued a stop signal. Our
684 * caller might release the siglock and then the pending
685 * stop signal it is about to process is no longer in the
686 * pending bitmasks, but must still be cleared by a SIGCONT
687 * (and overruled by a SIGKILL). So those cases clear this
688 * shared flag after we've set it. Note that this flag may
689 * remain set after the signal we return is ignored or
690 * handled. That doesn't matter because its only purpose
691 * is to alert stop-signal processing code when another
692 * processor has come along and cleared the flag.
693 */
694 current->jobctl |= JOBCTL_STOP_DEQUEUED;
695 }
696 #ifdef CONFIG_POSIX_TIMERS
697 if (resched_timer) {
698 /*
699 * Release the siglock to ensure proper locking order
700 * of timer locks outside of siglocks. Note, we leave
701 * irqs disabled here, since the posix-timers code is
702 * about to disable them again anyway.
703 */
704 spin_unlock(&tsk->sighand->siglock);
705 posixtimer_rearm(info);
706 spin_lock(&tsk->sighand->siglock);
707
708 /* Don't expose the si_sys_private value to userspace */
709 info->si_sys_private = 0;
710 }
711 #endif
712 return signr;
713 }
714 EXPORT_SYMBOL_GPL(dequeue_signal);
715
dequeue_synchronous_signal(kernel_siginfo_t * info)716 static int dequeue_synchronous_signal(kernel_siginfo_t *info)
717 {
718 struct task_struct *tsk = current;
719 struct sigpending *pending = &tsk->pending;
720 struct sigqueue *q, *sync = NULL;
721
722 /*
723 * Might a synchronous signal be in the queue?
724 */
725 if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
726 return 0;
727
728 /*
729 * Return the first synchronous signal in the queue.
730 */
731 list_for_each_entry(q, &pending->list, list) {
732 /* Synchronous signals have a positive si_code */
733 if ((q->info.si_code > SI_USER) &&
734 (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
735 sync = q;
736 goto next;
737 }
738 }
739 return 0;
740 next:
741 /*
742 * Check if there is another siginfo for the same signal.
743 */
744 list_for_each_entry_continue(q, &pending->list, list) {
745 if (q->info.si_signo == sync->info.si_signo)
746 goto still_pending;
747 }
748
749 sigdelset(&pending->signal, sync->info.si_signo);
750 recalc_sigpending();
751 still_pending:
752 list_del_init(&sync->list);
753 copy_siginfo(info, &sync->info);
754 __sigqueue_free(sync);
755 return info->si_signo;
756 }
757
758 /*
759 * Tell a process that it has a new active signal..
760 *
761 * NOTE! we rely on the previous spin_lock to
762 * lock interrupts for us! We can only be called with
763 * "siglock" held, and the local interrupt must
764 * have been disabled when that got acquired!
765 *
766 * No need to set need_resched since signal event passing
767 * goes through ->blocked
768 */
signal_wake_up_state(struct task_struct * t,unsigned int state)769 void signal_wake_up_state(struct task_struct *t, unsigned int state)
770 {
771 lockdep_assert_held(&t->sighand->siglock);
772
773 set_tsk_thread_flag(t, TIF_SIGPENDING);
774
775 /*
776 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
777 * case. We don't check t->state here because there is a race with it
778 * executing another processor and just now entering stopped state.
779 * By using wake_up_state, we ensure the process will wake up and
780 * handle its death signal.
781 */
782 if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
783 kick_process(t);
784 }
785
786 /*
787 * Remove signals in mask from the pending set and queue.
788 * Returns 1 if any signals were found.
789 *
790 * All callers must be holding the siglock.
791 */
flush_sigqueue_mask(sigset_t * mask,struct sigpending * s)792 static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
793 {
794 struct sigqueue *q, *n;
795 sigset_t m;
796
797 sigandsets(&m, mask, &s->signal);
798 if (sigisemptyset(&m))
799 return;
800
801 sigandnsets(&s->signal, &s->signal, mask);
802 list_for_each_entry_safe(q, n, &s->list, list) {
803 if (sigismember(mask, q->info.si_signo)) {
804 list_del_init(&q->list);
805 __sigqueue_free(q);
806 }
807 }
808 }
809
is_si_special(const struct kernel_siginfo * info)810 static inline int is_si_special(const struct kernel_siginfo *info)
811 {
812 return info <= SEND_SIG_PRIV;
813 }
814
si_fromuser(const struct kernel_siginfo * info)815 static inline bool si_fromuser(const struct kernel_siginfo *info)
816 {
817 return info == SEND_SIG_NOINFO ||
818 (!is_si_special(info) && SI_FROMUSER(info));
819 }
820
821 /*
822 * called with RCU read lock from check_kill_permission()
823 */
kill_ok_by_cred(struct task_struct * t)824 static bool kill_ok_by_cred(struct task_struct *t)
825 {
826 const struct cred *cred = current_cred();
827 const struct cred *tcred = __task_cred(t);
828
829 return uid_eq(cred->euid, tcred->suid) ||
830 uid_eq(cred->euid, tcred->uid) ||
831 uid_eq(cred->uid, tcred->suid) ||
832 uid_eq(cred->uid, tcred->uid) ||
833 ns_capable(tcred->user_ns, CAP_KILL);
834 }
835
836 /*
837 * Bad permissions for sending the signal
838 * - the caller must hold the RCU read lock
839 */
check_kill_permission(int sig,struct kernel_siginfo * info,struct task_struct * t)840 static int check_kill_permission(int sig, struct kernel_siginfo *info,
841 struct task_struct *t)
842 {
843 struct pid *sid;
844 int error;
845
846 if (!valid_signal(sig))
847 return -EINVAL;
848
849 if (!si_fromuser(info))
850 return 0;
851
852 error = audit_signal_info(sig, t); /* Let audit system see the signal */
853 if (error)
854 return error;
855
856 if (!same_thread_group(current, t) &&
857 !kill_ok_by_cred(t)) {
858 switch (sig) {
859 case SIGCONT:
860 sid = task_session(t);
861 /*
862 * We don't return the error if sid == NULL. The
863 * task was unhashed, the caller must notice this.
864 */
865 if (!sid || sid == task_session(current))
866 break;
867 fallthrough;
868 default:
869 return -EPERM;
870 }
871 }
872
873 return security_task_kill(t, info, sig, NULL);
874 }
875
876 /**
877 * ptrace_trap_notify - schedule trap to notify ptracer
878 * @t: tracee wanting to notify tracer
879 *
880 * This function schedules sticky ptrace trap which is cleared on the next
881 * TRAP_STOP to notify ptracer of an event. @t must have been seized by
882 * ptracer.
883 *
884 * If @t is running, STOP trap will be taken. If trapped for STOP and
885 * ptracer is listening for events, tracee is woken up so that it can
886 * re-trap for the new event. If trapped otherwise, STOP trap will be
887 * eventually taken without returning to userland after the existing traps
888 * are finished by PTRACE_CONT.
889 *
890 * CONTEXT:
891 * Must be called with @task->sighand->siglock held.
892 */
ptrace_trap_notify(struct task_struct * t)893 static void ptrace_trap_notify(struct task_struct *t)
894 {
895 WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
896 lockdep_assert_held(&t->sighand->siglock);
897
898 task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
899 ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
900 }
901
902 /*
903 * Handle magic process-wide effects of stop/continue signals. Unlike
904 * the signal actions, these happen immediately at signal-generation
905 * time regardless of blocking, ignoring, or handling. This does the
906 * actual continuing for SIGCONT, but not the actual stopping for stop
907 * signals. The process stop is done as a signal action for SIG_DFL.
908 *
909 * Returns true if the signal should be actually delivered, otherwise
910 * it should be dropped.
911 */
prepare_signal(int sig,struct task_struct * p,bool force)912 static bool prepare_signal(int sig, struct task_struct *p, bool force)
913 {
914 struct signal_struct *signal = p->signal;
915 struct task_struct *t;
916 sigset_t flush;
917
918 if (signal->flags & SIGNAL_GROUP_EXIT) {
919 if (signal->core_state)
920 return sig == SIGKILL;
921 /*
922 * The process is in the middle of dying, drop the signal.
923 */
924 return false;
925 } else if (sig_kernel_stop(sig)) {
926 /*
927 * This is a stop signal. Remove SIGCONT from all queues.
928 */
929 siginitset(&flush, sigmask(SIGCONT));
930 flush_sigqueue_mask(&flush, &signal->shared_pending);
931 for_each_thread(p, t)
932 flush_sigqueue_mask(&flush, &t->pending);
933 } else if (sig == SIGCONT) {
934 unsigned int why;
935 /*
936 * Remove all stop signals from all queues, wake all threads.
937 */
938 siginitset(&flush, SIG_KERNEL_STOP_MASK);
939 flush_sigqueue_mask(&flush, &signal->shared_pending);
940 for_each_thread(p, t) {
941 flush_sigqueue_mask(&flush, &t->pending);
942 task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
943 if (likely(!(t->ptrace & PT_SEIZED))) {
944 t->jobctl &= ~JOBCTL_STOPPED;
945 wake_up_state(t, __TASK_STOPPED);
946 } else
947 ptrace_trap_notify(t);
948 }
949
950 /*
951 * Notify the parent with CLD_CONTINUED if we were stopped.
952 *
953 * If we were in the middle of a group stop, we pretend it
954 * was already finished, and then continued. Since SIGCHLD
955 * doesn't queue we report only CLD_STOPPED, as if the next
956 * CLD_CONTINUED was dropped.
957 */
958 why = 0;
959 if (signal->flags & SIGNAL_STOP_STOPPED)
960 why |= SIGNAL_CLD_CONTINUED;
961 else if (signal->group_stop_count)
962 why |= SIGNAL_CLD_STOPPED;
963
964 if (why) {
965 /*
966 * The first thread which returns from do_signal_stop()
967 * will take ->siglock, notice SIGNAL_CLD_MASK, and
968 * notify its parent. See get_signal().
969 */
970 signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
971 signal->group_stop_count = 0;
972 signal->group_exit_code = 0;
973 }
974 }
975
976 return !sig_ignored(p, sig, force);
977 }
978
979 /*
980 * Test if P wants to take SIG. After we've checked all threads with this,
981 * it's equivalent to finding no threads not blocking SIG. Any threads not
982 * blocking SIG were ruled out because they are not running and already
983 * have pending signals. Such threads will dequeue from the shared queue
984 * as soon as they're available, so putting the signal on the shared queue
985 * will be equivalent to sending it to one such thread.
986 */
wants_signal(int sig,struct task_struct * p)987 static inline bool wants_signal(int sig, struct task_struct *p)
988 {
989 if (sigismember(&p->blocked, sig))
990 return false;
991
992 if (p->flags & PF_EXITING)
993 return false;
994
995 if (sig == SIGKILL)
996 return true;
997
998 if (task_is_stopped_or_traced(p))
999 return false;
1000
1001 return task_curr(p) || !task_sigpending(p);
1002 }
1003
complete_signal(int sig,struct task_struct * p,enum pid_type type)1004 static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
1005 {
1006 struct signal_struct *signal = p->signal;
1007 struct task_struct *t;
1008
1009 /*
1010 * Now find a thread we can wake up to take the signal off the queue.
1011 *
1012 * Try the suggested task first (may or may not be the main thread).
1013 */
1014 if (wants_signal(sig, p))
1015 t = p;
1016 else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1017 /*
1018 * There is just one thread and it does not need to be woken.
1019 * It will dequeue unblocked signals before it runs again.
1020 */
1021 return;
1022 else {
1023 /*
1024 * Otherwise try to find a suitable thread.
1025 */
1026 t = signal->curr_target;
1027 while (!wants_signal(sig, t)) {
1028 t = next_thread(t);
1029 if (t == signal->curr_target)
1030 /*
1031 * No thread needs to be woken.
1032 * Any eligible threads will see
1033 * the signal in the queue soon.
1034 */
1035 return;
1036 }
1037 signal->curr_target = t;
1038 }
1039
1040 /*
1041 * Found a killable thread. If the signal will be fatal,
1042 * then start taking the whole group down immediately.
1043 */
1044 if (sig_fatal(p, sig) &&
1045 (signal->core_state || !(signal->flags & SIGNAL_GROUP_EXIT)) &&
1046 !sigismember(&t->real_blocked, sig) &&
1047 (sig == SIGKILL || !p->ptrace)) {
1048 /*
1049 * This signal will be fatal to the whole group.
1050 */
1051 if (!sig_kernel_coredump(sig)) {
1052 /*
1053 * Start a group exit and wake everybody up.
1054 * This way we don't have other threads
1055 * running and doing things after a slower
1056 * thread has the fatal signal pending.
1057 */
1058 signal->flags = SIGNAL_GROUP_EXIT;
1059 signal->group_exit_code = sig;
1060 signal->group_stop_count = 0;
1061 t = p;
1062 do {
1063 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1064 sigaddset(&t->pending.signal, SIGKILL);
1065 signal_wake_up(t, 1);
1066 } while_each_thread(p, t);
1067 return;
1068 }
1069 }
1070
1071 /*
1072 * The signal is already in the shared-pending queue.
1073 * Tell the chosen thread to wake up and dequeue it.
1074 */
1075 signal_wake_up(t, sig == SIGKILL);
1076 return;
1077 }
1078
legacy_queue(struct sigpending * signals,int sig)1079 static inline bool legacy_queue(struct sigpending *signals, int sig)
1080 {
1081 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1082 }
1083
__send_signal_locked(int sig,struct kernel_siginfo * info,struct task_struct * t,enum pid_type type,bool force)1084 static int __send_signal_locked(int sig, struct kernel_siginfo *info,
1085 struct task_struct *t, enum pid_type type, bool force)
1086 {
1087 struct sigpending *pending;
1088 struct sigqueue *q;
1089 int override_rlimit;
1090 int ret = 0, result;
1091
1092 lockdep_assert_held(&t->sighand->siglock);
1093
1094 result = TRACE_SIGNAL_IGNORED;
1095 if (!prepare_signal(sig, t, force))
1096 goto ret;
1097
1098 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1099 /*
1100 * Short-circuit ignored signals and support queuing
1101 * exactly one non-rt signal, so that we can get more
1102 * detailed information about the cause of the signal.
1103 */
1104 result = TRACE_SIGNAL_ALREADY_PENDING;
1105 if (legacy_queue(pending, sig))
1106 goto ret;
1107
1108 result = TRACE_SIGNAL_DELIVERED;
1109 /*
1110 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1111 */
1112 if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1113 goto out_set;
1114
1115 /*
1116 * Real-time signals must be queued if sent by sigqueue, or
1117 * some other real-time mechanism. It is implementation
1118 * defined whether kill() does so. We attempt to do so, on
1119 * the principle of least surprise, but since kill is not
1120 * allowed to fail with EAGAIN when low on memory we just
1121 * make sure at least one signal gets delivered and don't
1122 * pass on the info struct.
1123 */
1124 if (sig < SIGRTMIN)
1125 override_rlimit = (is_si_special(info) || info->si_code >= 0);
1126 else
1127 override_rlimit = 0;
1128
1129 q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit, 0);
1130
1131 if (q) {
1132 list_add_tail(&q->list, &pending->list);
1133 switch ((unsigned long) info) {
1134 case (unsigned long) SEND_SIG_NOINFO:
1135 clear_siginfo(&q->info);
1136 q->info.si_signo = sig;
1137 q->info.si_errno = 0;
1138 q->info.si_code = SI_USER;
1139 q->info.si_pid = task_tgid_nr_ns(current,
1140 task_active_pid_ns(t));
1141 rcu_read_lock();
1142 q->info.si_uid =
1143 from_kuid_munged(task_cred_xxx(t, user_ns),
1144 current_uid());
1145 rcu_read_unlock();
1146 break;
1147 case (unsigned long) SEND_SIG_PRIV:
1148 clear_siginfo(&q->info);
1149 q->info.si_signo = sig;
1150 q->info.si_errno = 0;
1151 q->info.si_code = SI_KERNEL;
1152 q->info.si_pid = 0;
1153 q->info.si_uid = 0;
1154 break;
1155 default:
1156 copy_siginfo(&q->info, info);
1157 break;
1158 }
1159 } else if (!is_si_special(info) &&
1160 sig >= SIGRTMIN && info->si_code != SI_USER) {
1161 /*
1162 * Queue overflow, abort. We may abort if the
1163 * signal was rt and sent by user using something
1164 * other than kill().
1165 */
1166 result = TRACE_SIGNAL_OVERFLOW_FAIL;
1167 ret = -EAGAIN;
1168 goto ret;
1169 } else {
1170 /*
1171 * This is a silent loss of information. We still
1172 * send the signal, but the *info bits are lost.
1173 */
1174 result = TRACE_SIGNAL_LOSE_INFO;
1175 }
1176
1177 out_set:
1178 signalfd_notify(t, sig);
1179 sigaddset(&pending->signal, sig);
1180
1181 /* Let multiprocess signals appear after on-going forks */
1182 if (type > PIDTYPE_TGID) {
1183 struct multiprocess_signals *delayed;
1184 hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1185 sigset_t *signal = &delayed->signal;
1186 /* Can't queue both a stop and a continue signal */
1187 if (sig == SIGCONT)
1188 sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1189 else if (sig_kernel_stop(sig))
1190 sigdelset(signal, SIGCONT);
1191 sigaddset(signal, sig);
1192 }
1193 }
1194
1195 complete_signal(sig, t, type);
1196 ret:
1197 trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1198 return ret;
1199 }
1200
has_si_pid_and_uid(struct kernel_siginfo * info)1201 static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1202 {
1203 bool ret = false;
1204 switch (siginfo_layout(info->si_signo, info->si_code)) {
1205 case SIL_KILL:
1206 case SIL_CHLD:
1207 case SIL_RT:
1208 ret = true;
1209 break;
1210 case SIL_TIMER:
1211 case SIL_POLL:
1212 case SIL_FAULT:
1213 case SIL_FAULT_TRAPNO:
1214 case SIL_FAULT_MCEERR:
1215 case SIL_FAULT_BNDERR:
1216 case SIL_FAULT_PKUERR:
1217 case SIL_FAULT_PERF_EVENT:
1218 case SIL_SYS:
1219 ret = false;
1220 break;
1221 }
1222 return ret;
1223 }
1224
send_signal_locked(int sig,struct kernel_siginfo * info,struct task_struct * t,enum pid_type type)1225 int send_signal_locked(int sig, struct kernel_siginfo *info,
1226 struct task_struct *t, enum pid_type type)
1227 {
1228 /* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1229 bool force = false;
1230
1231 if (info == SEND_SIG_NOINFO) {
1232 /* Force if sent from an ancestor pid namespace */
1233 force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1234 } else if (info == SEND_SIG_PRIV) {
1235 /* Don't ignore kernel generated signals */
1236 force = true;
1237 } else if (has_si_pid_and_uid(info)) {
1238 /* SIGKILL and SIGSTOP is special or has ids */
1239 struct user_namespace *t_user_ns;
1240
1241 rcu_read_lock();
1242 t_user_ns = task_cred_xxx(t, user_ns);
1243 if (current_user_ns() != t_user_ns) {
1244 kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1245 info->si_uid = from_kuid_munged(t_user_ns, uid);
1246 }
1247 rcu_read_unlock();
1248
1249 /* A kernel generated signal? */
1250 force = (info->si_code == SI_KERNEL);
1251
1252 /* From an ancestor pid namespace? */
1253 if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1254 info->si_pid = 0;
1255 force = true;
1256 }
1257 }
1258 return __send_signal_locked(sig, info, t, type, force);
1259 }
1260
print_fatal_signal(int signr)1261 static void print_fatal_signal(int signr)
1262 {
1263 struct pt_regs *regs = task_pt_regs(current);
1264 struct file *exe_file;
1265
1266 exe_file = get_task_exe_file(current);
1267 if (exe_file) {
1268 pr_info("%pD: %s: potentially unexpected fatal signal %d.\n",
1269 exe_file, current->comm, signr);
1270 fput(exe_file);
1271 } else {
1272 pr_info("%s: potentially unexpected fatal signal %d.\n",
1273 current->comm, signr);
1274 }
1275
1276 #if defined(__i386__) && !defined(__arch_um__)
1277 pr_info("code at %08lx: ", regs->ip);
1278 {
1279 int i;
1280 for (i = 0; i < 16; i++) {
1281 unsigned char insn;
1282
1283 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1284 break;
1285 pr_cont("%02x ", insn);
1286 }
1287 }
1288 pr_cont("\n");
1289 #endif
1290 preempt_disable();
1291 show_regs(regs);
1292 preempt_enable();
1293 }
1294
setup_print_fatal_signals(char * str)1295 static int __init setup_print_fatal_signals(char *str)
1296 {
1297 get_option (&str, &print_fatal_signals);
1298
1299 return 1;
1300 }
1301
1302 __setup("print-fatal-signals=", setup_print_fatal_signals);
1303
do_send_sig_info(int sig,struct kernel_siginfo * info,struct task_struct * p,enum pid_type type)1304 int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1305 enum pid_type type)
1306 {
1307 unsigned long flags;
1308 int ret = -ESRCH;
1309
1310 if (lock_task_sighand(p, &flags)) {
1311 ret = send_signal_locked(sig, info, p, type);
1312 unlock_task_sighand(p, &flags);
1313 }
1314
1315 return ret;
1316 }
1317
1318 enum sig_handler {
1319 HANDLER_CURRENT, /* If reachable use the current handler */
1320 HANDLER_SIG_DFL, /* Always use SIG_DFL handler semantics */
1321 HANDLER_EXIT, /* Only visible as the process exit code */
1322 };
1323
1324 /*
1325 * Force a signal that the process can't ignore: if necessary
1326 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1327 *
1328 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1329 * since we do not want to have a signal handler that was blocked
1330 * be invoked when user space had explicitly blocked it.
1331 *
1332 * We don't want to have recursive SIGSEGV's etc, for example,
1333 * that is why we also clear SIGNAL_UNKILLABLE.
1334 */
1335 static int
force_sig_info_to_task(struct kernel_siginfo * info,struct task_struct * t,enum sig_handler handler)1336 force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t,
1337 enum sig_handler handler)
1338 {
1339 unsigned long int flags;
1340 int ret, blocked, ignored;
1341 struct k_sigaction *action;
1342 int sig = info->si_signo;
1343
1344 spin_lock_irqsave(&t->sighand->siglock, flags);
1345 action = &t->sighand->action[sig-1];
1346 ignored = action->sa.sa_handler == SIG_IGN;
1347 blocked = sigismember(&t->blocked, sig);
1348 if (blocked || ignored || (handler != HANDLER_CURRENT)) {
1349 action->sa.sa_handler = SIG_DFL;
1350 if (handler == HANDLER_EXIT)
1351 action->sa.sa_flags |= SA_IMMUTABLE;
1352 if (blocked) {
1353 sigdelset(&t->blocked, sig);
1354 recalc_sigpending_and_wake(t);
1355 }
1356 }
1357 /*
1358 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1359 * debugging to leave init killable. But HANDLER_EXIT is always fatal.
1360 */
1361 if (action->sa.sa_handler == SIG_DFL &&
1362 (!t->ptrace || (handler == HANDLER_EXIT)))
1363 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1364 ret = send_signal_locked(sig, info, t, PIDTYPE_PID);
1365 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1366
1367 return ret;
1368 }
1369
force_sig_info(struct kernel_siginfo * info)1370 int force_sig_info(struct kernel_siginfo *info)
1371 {
1372 return force_sig_info_to_task(info, current, HANDLER_CURRENT);
1373 }
1374
1375 /*
1376 * Nuke all other threads in the group.
1377 */
zap_other_threads(struct task_struct * p)1378 int zap_other_threads(struct task_struct *p)
1379 {
1380 struct task_struct *t = p;
1381 int count = 0;
1382
1383 p->signal->group_stop_count = 0;
1384
1385 while_each_thread(p, t) {
1386 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1387 /* Don't require de_thread to wait for the vhost_worker */
1388 if ((t->flags & (PF_IO_WORKER | PF_USER_WORKER)) != PF_USER_WORKER)
1389 count++;
1390
1391 /* Don't bother with already dead threads */
1392 if (t->exit_state)
1393 continue;
1394 sigaddset(&t->pending.signal, SIGKILL);
1395 signal_wake_up(t, 1);
1396 }
1397
1398 return count;
1399 }
1400
__lock_task_sighand(struct task_struct * tsk,unsigned long * flags)1401 struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1402 unsigned long *flags)
1403 {
1404 struct sighand_struct *sighand;
1405
1406 rcu_read_lock();
1407 for (;;) {
1408 sighand = rcu_dereference(tsk->sighand);
1409 if (unlikely(sighand == NULL))
1410 break;
1411
1412 /*
1413 * This sighand can be already freed and even reused, but
1414 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1415 * initializes ->siglock: this slab can't go away, it has
1416 * the same object type, ->siglock can't be reinitialized.
1417 *
1418 * We need to ensure that tsk->sighand is still the same
1419 * after we take the lock, we can race with de_thread() or
1420 * __exit_signal(). In the latter case the next iteration
1421 * must see ->sighand == NULL.
1422 */
1423 spin_lock_irqsave(&sighand->siglock, *flags);
1424 if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1425 break;
1426 spin_unlock_irqrestore(&sighand->siglock, *flags);
1427 }
1428 rcu_read_unlock();
1429
1430 return sighand;
1431 }
1432
1433 #ifdef CONFIG_LOCKDEP
lockdep_assert_task_sighand_held(struct task_struct * task)1434 void lockdep_assert_task_sighand_held(struct task_struct *task)
1435 {
1436 struct sighand_struct *sighand;
1437
1438 rcu_read_lock();
1439 sighand = rcu_dereference(task->sighand);
1440 if (sighand)
1441 lockdep_assert_held(&sighand->siglock);
1442 else
1443 WARN_ON_ONCE(1);
1444 rcu_read_unlock();
1445 }
1446 #endif
1447
1448 /*
1449 * send signal info to all the members of a group
1450 */
group_send_sig_info(int sig,struct kernel_siginfo * info,struct task_struct * p,enum pid_type type)1451 int group_send_sig_info(int sig, struct kernel_siginfo *info,
1452 struct task_struct *p, enum pid_type type)
1453 {
1454 int ret;
1455
1456 rcu_read_lock();
1457 ret = check_kill_permission(sig, info, p);
1458 rcu_read_unlock();
1459
1460 if (!ret && sig)
1461 ret = do_send_sig_info(sig, info, p, type);
1462
1463 return ret;
1464 }
1465
1466 /*
1467 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1468 * control characters do (^C, ^Z etc)
1469 * - the caller must hold at least a readlock on tasklist_lock
1470 */
__kill_pgrp_info(int sig,struct kernel_siginfo * info,struct pid * pgrp)1471 int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1472 {
1473 struct task_struct *p = NULL;
1474 int retval, success;
1475
1476 success = 0;
1477 retval = -ESRCH;
1478 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1479 int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1480 success |= !err;
1481 retval = err;
1482 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1483 return success ? 0 : retval;
1484 }
1485
kill_pid_info(int sig,struct kernel_siginfo * info,struct pid * pid)1486 int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1487 {
1488 int error = -ESRCH;
1489 struct task_struct *p;
1490
1491 for (;;) {
1492 rcu_read_lock();
1493 p = pid_task(pid, PIDTYPE_PID);
1494 if (p)
1495 error = group_send_sig_info(sig, info, p, PIDTYPE_TGID);
1496 rcu_read_unlock();
1497 if (likely(!p || error != -ESRCH))
1498 return error;
1499
1500 /*
1501 * The task was unhashed in between, try again. If it
1502 * is dead, pid_task() will return NULL, if we race with
1503 * de_thread() it will find the new leader.
1504 */
1505 }
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 ___ARCH_SI_IA64 (int imm,unsigned int flags,unsigned long isr),struct task_struct * t)1721 int force_sig_fault_to_task(int sig, int code, void __user *addr
1722 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1723 , struct task_struct *t)
1724 {
1725 struct kernel_siginfo info;
1726
1727 clear_siginfo(&info);
1728 info.si_signo = sig;
1729 info.si_errno = 0;
1730 info.si_code = code;
1731 info.si_addr = addr;
1732 #ifdef __ia64__
1733 info.si_imm = imm;
1734 info.si_flags = flags;
1735 info.si_isr = isr;
1736 #endif
1737 return force_sig_info_to_task(&info, t, HANDLER_CURRENT);
1738 }
1739
force_sig_fault(int sig,int code,void __user * addr ___ARCH_SI_IA64 (int imm,unsigned int flags,unsigned long isr))1740 int force_sig_fault(int sig, int code, void __user *addr
1741 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr))
1742 {
1743 return force_sig_fault_to_task(sig, code, addr
1744 ___ARCH_SI_IA64(imm, flags, isr), current);
1745 }
1746
send_sig_fault(int sig,int code,void __user * addr ___ARCH_SI_IA64 (int imm,unsigned int flags,unsigned long isr),struct task_struct * t)1747 int send_sig_fault(int sig, int code, void __user *addr
1748 ___ARCH_SI_IA64(int imm, unsigned int flags, unsigned long isr)
1749 , struct task_struct *t)
1750 {
1751 struct kernel_siginfo info;
1752
1753 clear_siginfo(&info);
1754 info.si_signo = sig;
1755 info.si_errno = 0;
1756 info.si_code = code;
1757 info.si_addr = addr;
1758 #ifdef __ia64__
1759 info.si_imm = imm;
1760 info.si_flags = flags;
1761 info.si_isr = isr;
1762 #endif
1763 return send_sig_info(info.si_signo, &info, t);
1764 }
1765
force_sig_mceerr(int code,void __user * addr,short lsb)1766 int force_sig_mceerr(int code, void __user *addr, short lsb)
1767 {
1768 struct kernel_siginfo info;
1769
1770 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1771 clear_siginfo(&info);
1772 info.si_signo = SIGBUS;
1773 info.si_errno = 0;
1774 info.si_code = code;
1775 info.si_addr = addr;
1776 info.si_addr_lsb = lsb;
1777 return force_sig_info(&info);
1778 }
1779
send_sig_mceerr(int code,void __user * addr,short lsb,struct task_struct * t)1780 int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1781 {
1782 struct kernel_siginfo info;
1783
1784 WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1785 clear_siginfo(&info);
1786 info.si_signo = SIGBUS;
1787 info.si_errno = 0;
1788 info.si_code = code;
1789 info.si_addr = addr;
1790 info.si_addr_lsb = lsb;
1791 return send_sig_info(info.si_signo, &info, t);
1792 }
1793 EXPORT_SYMBOL(send_sig_mceerr);
1794
force_sig_bnderr(void __user * addr,void __user * lower,void __user * upper)1795 int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
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_BNDERR;
1803 info.si_addr = addr;
1804 info.si_lower = lower;
1805 info.si_upper = upper;
1806 return force_sig_info(&info);
1807 }
1808
1809 #ifdef SEGV_PKUERR
force_sig_pkuerr(void __user * addr,u32 pkey)1810 int force_sig_pkuerr(void __user *addr, u32 pkey)
1811 {
1812 struct kernel_siginfo info;
1813
1814 clear_siginfo(&info);
1815 info.si_signo = SIGSEGV;
1816 info.si_errno = 0;
1817 info.si_code = SEGV_PKUERR;
1818 info.si_addr = addr;
1819 info.si_pkey = pkey;
1820 return force_sig_info(&info);
1821 }
1822 #endif
1823
send_sig_perf(void __user * addr,u32 type,u64 sig_data)1824 int send_sig_perf(void __user *addr, u32 type, u64 sig_data)
1825 {
1826 struct kernel_siginfo info;
1827
1828 clear_siginfo(&info);
1829 info.si_signo = SIGTRAP;
1830 info.si_errno = 0;
1831 info.si_code = TRAP_PERF;
1832 info.si_addr = addr;
1833 info.si_perf_data = sig_data;
1834 info.si_perf_type = type;
1835
1836 /*
1837 * Signals generated by perf events should not terminate the whole
1838 * process if SIGTRAP is blocked, however, delivering the signal
1839 * asynchronously is better than not delivering at all. But tell user
1840 * space if the signal was asynchronous, so it can clearly be
1841 * distinguished from normal synchronous ones.
1842 */
1843 info.si_perf_flags = sigismember(¤t->blocked, info.si_signo) ?
1844 TRAP_PERF_FLAG_ASYNC :
1845 0;
1846
1847 return send_sig_info(info.si_signo, &info, current);
1848 }
1849
1850 /**
1851 * force_sig_seccomp - signals the task to allow in-process syscall emulation
1852 * @syscall: syscall number to send to userland
1853 * @reason: filter-supplied reason code to send to userland (via si_errno)
1854 * @force_coredump: true to trigger a coredump
1855 *
1856 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
1857 */
force_sig_seccomp(int syscall,int reason,bool force_coredump)1858 int force_sig_seccomp(int syscall, int reason, bool force_coredump)
1859 {
1860 struct kernel_siginfo info;
1861
1862 clear_siginfo(&info);
1863 info.si_signo = SIGSYS;
1864 info.si_code = SYS_SECCOMP;
1865 info.si_call_addr = (void __user *)KSTK_EIP(current);
1866 info.si_errno = reason;
1867 info.si_arch = syscall_get_arch(current);
1868 info.si_syscall = syscall;
1869 return force_sig_info_to_task(&info, current,
1870 force_coredump ? HANDLER_EXIT : HANDLER_CURRENT);
1871 }
1872
1873 /* For the crazy architectures that include trap information in
1874 * the errno field, instead of an actual errno value.
1875 */
force_sig_ptrace_errno_trap(int errno,void __user * addr)1876 int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1877 {
1878 struct kernel_siginfo info;
1879
1880 clear_siginfo(&info);
1881 info.si_signo = SIGTRAP;
1882 info.si_errno = errno;
1883 info.si_code = TRAP_HWBKPT;
1884 info.si_addr = addr;
1885 return force_sig_info(&info);
1886 }
1887
1888 /* For the rare architectures that include trap information using
1889 * si_trapno.
1890 */
force_sig_fault_trapno(int sig,int code,void __user * addr,int trapno)1891 int force_sig_fault_trapno(int sig, int code, void __user *addr, int trapno)
1892 {
1893 struct kernel_siginfo info;
1894
1895 clear_siginfo(&info);
1896 info.si_signo = sig;
1897 info.si_errno = 0;
1898 info.si_code = code;
1899 info.si_addr = addr;
1900 info.si_trapno = trapno;
1901 return force_sig_info(&info);
1902 }
1903
1904 /* For the rare architectures that include trap information using
1905 * si_trapno.
1906 */
send_sig_fault_trapno(int sig,int code,void __user * addr,int trapno,struct task_struct * t)1907 int send_sig_fault_trapno(int sig, int code, void __user *addr, int trapno,
1908 struct task_struct *t)
1909 {
1910 struct kernel_siginfo info;
1911
1912 clear_siginfo(&info);
1913 info.si_signo = sig;
1914 info.si_errno = 0;
1915 info.si_code = code;
1916 info.si_addr = addr;
1917 info.si_trapno = trapno;
1918 return send_sig_info(info.si_signo, &info, t);
1919 }
1920
kill_pgrp(struct pid * pid,int sig,int priv)1921 int kill_pgrp(struct pid *pid, int sig, int priv)
1922 {
1923 int ret;
1924
1925 read_lock(&tasklist_lock);
1926 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1927 read_unlock(&tasklist_lock);
1928
1929 return ret;
1930 }
1931 EXPORT_SYMBOL(kill_pgrp);
1932
kill_pid(struct pid * pid,int sig,int priv)1933 int kill_pid(struct pid *pid, int sig, int priv)
1934 {
1935 return kill_pid_info(sig, __si_special(priv), pid);
1936 }
1937 EXPORT_SYMBOL(kill_pid);
1938
1939 /*
1940 * These functions support sending signals using preallocated sigqueue
1941 * structures. This is needed "because realtime applications cannot
1942 * afford to lose notifications of asynchronous events, like timer
1943 * expirations or I/O completions". In the case of POSIX Timers
1944 * we allocate the sigqueue structure from the timer_create. If this
1945 * allocation fails we are able to report the failure to the application
1946 * with an EAGAIN error.
1947 */
sigqueue_alloc(void)1948 struct sigqueue *sigqueue_alloc(void)
1949 {
1950 return __sigqueue_alloc(-1, current, GFP_KERNEL, 0, SIGQUEUE_PREALLOC);
1951 }
1952
sigqueue_free(struct sigqueue * q)1953 void sigqueue_free(struct sigqueue *q)
1954 {
1955 unsigned long flags;
1956 spinlock_t *lock = ¤t->sighand->siglock;
1957
1958 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1959 /*
1960 * We must hold ->siglock while testing q->list
1961 * to serialize with collect_signal() or with
1962 * __exit_signal()->flush_sigqueue().
1963 */
1964 spin_lock_irqsave(lock, flags);
1965 q->flags &= ~SIGQUEUE_PREALLOC;
1966 /*
1967 * If it is queued it will be freed when dequeued,
1968 * like the "regular" sigqueue.
1969 */
1970 if (!list_empty(&q->list))
1971 q = NULL;
1972 spin_unlock_irqrestore(lock, flags);
1973
1974 if (q)
1975 __sigqueue_free(q);
1976 }
1977
send_sigqueue(struct sigqueue * q,struct pid * pid,enum pid_type type)1978 int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1979 {
1980 int sig = q->info.si_signo;
1981 struct sigpending *pending;
1982 struct task_struct *t;
1983 unsigned long flags;
1984 int ret, result;
1985
1986 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1987
1988 ret = -1;
1989 rcu_read_lock();
1990
1991 /*
1992 * This function is used by POSIX timers to deliver a timer signal.
1993 * Where type is PIDTYPE_PID (such as for timers with SIGEV_THREAD_ID
1994 * set), the signal must be delivered to the specific thread (queues
1995 * into t->pending).
1996 *
1997 * Where type is not PIDTYPE_PID, signals must be delivered to the
1998 * process. In this case, prefer to deliver to current if it is in
1999 * the same thread group as the target process, which avoids
2000 * unnecessarily waking up a potentially idle task.
2001 */
2002 t = pid_task(pid, type);
2003 if (!t)
2004 goto ret;
2005 if (type != PIDTYPE_PID && same_thread_group(t, current))
2006 t = current;
2007 if (!likely(lock_task_sighand(t, &flags)))
2008 goto ret;
2009
2010 ret = 1; /* the signal is ignored */
2011 result = TRACE_SIGNAL_IGNORED;
2012 if (!prepare_signal(sig, t, false))
2013 goto out;
2014
2015 ret = 0;
2016 if (unlikely(!list_empty(&q->list))) {
2017 /*
2018 * If an SI_TIMER entry is already queue just increment
2019 * the overrun count.
2020 */
2021 BUG_ON(q->info.si_code != SI_TIMER);
2022 q->info.si_overrun++;
2023 result = TRACE_SIGNAL_ALREADY_PENDING;
2024 goto out;
2025 }
2026 q->info.si_overrun = 0;
2027
2028 signalfd_notify(t, sig);
2029 pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
2030 list_add_tail(&q->list, &pending->list);
2031 sigaddset(&pending->signal, sig);
2032 complete_signal(sig, t, type);
2033 result = TRACE_SIGNAL_DELIVERED;
2034 out:
2035 trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
2036 unlock_task_sighand(t, &flags);
2037 ret:
2038 rcu_read_unlock();
2039 return ret;
2040 }
2041
do_notify_pidfd(struct task_struct * task)2042 static void do_notify_pidfd(struct task_struct *task)
2043 {
2044 struct pid *pid;
2045
2046 WARN_ON(task->exit_state == 0);
2047 pid = task_pid(task);
2048 wake_up_all(&pid->wait_pidfd);
2049 }
2050
2051 /*
2052 * Let a parent know about the death of a child.
2053 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
2054 *
2055 * Returns true if our parent ignored us and so we've switched to
2056 * self-reaping.
2057 */
do_notify_parent(struct task_struct * tsk,int sig)2058 bool do_notify_parent(struct task_struct *tsk, int sig)
2059 {
2060 struct kernel_siginfo info;
2061 unsigned long flags;
2062 struct sighand_struct *psig;
2063 bool autoreap = false;
2064 u64 utime, stime;
2065
2066 WARN_ON_ONCE(sig == -1);
2067
2068 /* do_notify_parent_cldstop should have been called instead. */
2069 WARN_ON_ONCE(task_is_stopped_or_traced(tsk));
2070
2071 WARN_ON_ONCE(!tsk->ptrace &&
2072 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
2073
2074 /* Wake up all pidfd waiters */
2075 do_notify_pidfd(tsk);
2076
2077 if (sig != SIGCHLD) {
2078 /*
2079 * This is only possible if parent == real_parent.
2080 * Check if it has changed security domain.
2081 */
2082 if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
2083 sig = SIGCHLD;
2084 }
2085
2086 clear_siginfo(&info);
2087 info.si_signo = sig;
2088 info.si_errno = 0;
2089 /*
2090 * We are under tasklist_lock here so our parent is tied to
2091 * us and cannot change.
2092 *
2093 * task_active_pid_ns will always return the same pid namespace
2094 * until a task passes through release_task.
2095 *
2096 * write_lock() currently calls preempt_disable() which is the
2097 * same as rcu_read_lock(), but according to Oleg, this is not
2098 * correct to rely on this
2099 */
2100 rcu_read_lock();
2101 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
2102 info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
2103 task_uid(tsk));
2104 rcu_read_unlock();
2105
2106 task_cputime(tsk, &utime, &stime);
2107 info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
2108 info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
2109
2110 info.si_status = tsk->exit_code & 0x7f;
2111 if (tsk->exit_code & 0x80)
2112 info.si_code = CLD_DUMPED;
2113 else if (tsk->exit_code & 0x7f)
2114 info.si_code = CLD_KILLED;
2115 else {
2116 info.si_code = CLD_EXITED;
2117 info.si_status = tsk->exit_code >> 8;
2118 }
2119
2120 psig = tsk->parent->sighand;
2121 spin_lock_irqsave(&psig->siglock, flags);
2122 if (!tsk->ptrace && sig == SIGCHLD &&
2123 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
2124 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
2125 /*
2126 * We are exiting and our parent doesn't care. POSIX.1
2127 * defines special semantics for setting SIGCHLD to SIG_IGN
2128 * or setting the SA_NOCLDWAIT flag: we should be reaped
2129 * automatically and not left for our parent's wait4 call.
2130 * Rather than having the parent do it as a magic kind of
2131 * signal handler, we just set this to tell do_exit that we
2132 * can be cleaned up without becoming a zombie. Note that
2133 * we still call __wake_up_parent in this case, because a
2134 * blocked sys_wait4 might now return -ECHILD.
2135 *
2136 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
2137 * is implementation-defined: we do (if you don't want
2138 * it, just use SIG_IGN instead).
2139 */
2140 autoreap = true;
2141 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2142 sig = 0;
2143 }
2144 /*
2145 * Send with __send_signal as si_pid and si_uid are in the
2146 * parent's namespaces.
2147 */
2148 if (valid_signal(sig) && sig)
2149 __send_signal_locked(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2150 __wake_up_parent(tsk, tsk->parent);
2151 spin_unlock_irqrestore(&psig->siglock, flags);
2152
2153 return autoreap;
2154 }
2155
2156 /**
2157 * do_notify_parent_cldstop - notify parent of stopped/continued state change
2158 * @tsk: task reporting the state change
2159 * @for_ptracer: the notification is for ptracer
2160 * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2161 *
2162 * Notify @tsk's parent that the stopped/continued state has changed. If
2163 * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2164 * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2165 *
2166 * CONTEXT:
2167 * Must be called with tasklist_lock at least read locked.
2168 */
do_notify_parent_cldstop(struct task_struct * tsk,bool for_ptracer,int why)2169 static void do_notify_parent_cldstop(struct task_struct *tsk,
2170 bool for_ptracer, int why)
2171 {
2172 struct kernel_siginfo info;
2173 unsigned long flags;
2174 struct task_struct *parent;
2175 struct sighand_struct *sighand;
2176 u64 utime, stime;
2177
2178 if (for_ptracer) {
2179 parent = tsk->parent;
2180 } else {
2181 tsk = tsk->group_leader;
2182 parent = tsk->real_parent;
2183 }
2184
2185 clear_siginfo(&info);
2186 info.si_signo = SIGCHLD;
2187 info.si_errno = 0;
2188 /*
2189 * see comment in do_notify_parent() about the following 4 lines
2190 */
2191 rcu_read_lock();
2192 info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2193 info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2194 rcu_read_unlock();
2195
2196 task_cputime(tsk, &utime, &stime);
2197 info.si_utime = nsec_to_clock_t(utime);
2198 info.si_stime = nsec_to_clock_t(stime);
2199
2200 info.si_code = why;
2201 switch (why) {
2202 case CLD_CONTINUED:
2203 info.si_status = SIGCONT;
2204 break;
2205 case CLD_STOPPED:
2206 info.si_status = tsk->signal->group_exit_code & 0x7f;
2207 break;
2208 case CLD_TRAPPED:
2209 info.si_status = tsk->exit_code & 0x7f;
2210 break;
2211 default:
2212 BUG();
2213 }
2214
2215 sighand = parent->sighand;
2216 spin_lock_irqsave(&sighand->siglock, flags);
2217 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2218 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2219 send_signal_locked(SIGCHLD, &info, parent, PIDTYPE_TGID);
2220 /*
2221 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2222 */
2223 __wake_up_parent(tsk, parent);
2224 spin_unlock_irqrestore(&sighand->siglock, flags);
2225 }
2226
2227 /*
2228 * This must be called with current->sighand->siglock held.
2229 *
2230 * This should be the path for all ptrace stops.
2231 * We always set current->last_siginfo while stopped here.
2232 * That makes it a way to test a stopped process for
2233 * being ptrace-stopped vs being job-control-stopped.
2234 *
2235 * Returns the signal the ptracer requested the code resume
2236 * with. If the code did not stop because the tracer is gone,
2237 * the stop signal remains unchanged unless clear_code.
2238 */
ptrace_stop(int exit_code,int why,unsigned long message,kernel_siginfo_t * info)2239 static int ptrace_stop(int exit_code, int why, unsigned long message,
2240 kernel_siginfo_t *info)
2241 __releases(¤t->sighand->siglock)
2242 __acquires(¤t->sighand->siglock)
2243 {
2244 bool gstop_done = false;
2245
2246 if (arch_ptrace_stop_needed()) {
2247 /*
2248 * The arch code has something special to do before a
2249 * ptrace stop. This is allowed to block, e.g. for faults
2250 * on user stack pages. We can't keep the siglock while
2251 * calling arch_ptrace_stop, so we must release it now.
2252 * To preserve proper semantics, we must do this before
2253 * any signal bookkeeping like checking group_stop_count.
2254 */
2255 spin_unlock_irq(¤t->sighand->siglock);
2256 arch_ptrace_stop();
2257 spin_lock_irq(¤t->sighand->siglock);
2258 }
2259
2260 /*
2261 * After this point ptrace_signal_wake_up or signal_wake_up
2262 * will clear TASK_TRACED if ptrace_unlink happens or a fatal
2263 * signal comes in. Handle previous ptrace_unlinks and fatal
2264 * signals here to prevent ptrace_stop sleeping in schedule.
2265 */
2266 if (!current->ptrace || __fatal_signal_pending(current))
2267 return exit_code;
2268
2269 set_special_state(TASK_TRACED);
2270 current->jobctl |= JOBCTL_TRACED;
2271
2272 /*
2273 * We're committing to trapping. TRACED should be visible before
2274 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2275 * Also, transition to TRACED and updates to ->jobctl should be
2276 * atomic with respect to siglock and should be done after the arch
2277 * hook as siglock is released and regrabbed across it.
2278 *
2279 * TRACER TRACEE
2280 *
2281 * ptrace_attach()
2282 * [L] wait_on_bit(JOBCTL_TRAPPING) [S] set_special_state(TRACED)
2283 * do_wait()
2284 * set_current_state() smp_wmb();
2285 * ptrace_do_wait()
2286 * wait_task_stopped()
2287 * task_stopped_code()
2288 * [L] task_is_traced() [S] task_clear_jobctl_trapping();
2289 */
2290 smp_wmb();
2291
2292 current->ptrace_message = message;
2293 current->last_siginfo = info;
2294 current->exit_code = exit_code;
2295
2296 /*
2297 * If @why is CLD_STOPPED, we're trapping to participate in a group
2298 * stop. Do the bookkeeping. Note that if SIGCONT was delievered
2299 * across siglock relocks since INTERRUPT was scheduled, PENDING
2300 * could be clear now. We act as if SIGCONT is received after
2301 * TASK_TRACED is entered - ignore it.
2302 */
2303 if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2304 gstop_done = task_participate_group_stop(current);
2305
2306 /* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2307 task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2308 if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2309 task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2310
2311 /* entering a trap, clear TRAPPING */
2312 task_clear_jobctl_trapping(current);
2313
2314 spin_unlock_irq(¤t->sighand->siglock);
2315 read_lock(&tasklist_lock);
2316 /*
2317 * Notify parents of the stop.
2318 *
2319 * While ptraced, there are two parents - the ptracer and
2320 * the real_parent of the group_leader. The ptracer should
2321 * know about every stop while the real parent is only
2322 * interested in the completion of group stop. The states
2323 * for the two don't interact with each other. Notify
2324 * separately unless they're gonna be duplicates.
2325 */
2326 if (current->ptrace)
2327 do_notify_parent_cldstop(current, true, why);
2328 if (gstop_done && (!current->ptrace || ptrace_reparented(current)))
2329 do_notify_parent_cldstop(current, false, why);
2330
2331 /*
2332 * Don't want to allow preemption here, because
2333 * sys_ptrace() needs this task to be inactive.
2334 *
2335 * XXX: implement read_unlock_no_resched().
2336 */
2337 preempt_disable();
2338 read_unlock(&tasklist_lock);
2339 cgroup_enter_frozen();
2340 preempt_enable_no_resched();
2341 schedule();
2342 cgroup_leave_frozen(true);
2343
2344 /*
2345 * We are back. Now reacquire the siglock before touching
2346 * last_siginfo, so that we are sure to have synchronized with
2347 * any signal-sending on another CPU that wants to examine it.
2348 */
2349 spin_lock_irq(¤t->sighand->siglock);
2350 exit_code = current->exit_code;
2351 current->last_siginfo = NULL;
2352 current->ptrace_message = 0;
2353 current->exit_code = 0;
2354
2355 /* LISTENING can be set only during STOP traps, clear it */
2356 current->jobctl &= ~(JOBCTL_LISTENING | JOBCTL_PTRACE_FROZEN);
2357
2358 /*
2359 * Queued signals ignored us while we were stopped for tracing.
2360 * So check for any that we should take before resuming user mode.
2361 * This sets TIF_SIGPENDING, but never clears it.
2362 */
2363 recalc_sigpending_tsk(current);
2364 return exit_code;
2365 }
2366
ptrace_do_notify(int signr,int exit_code,int why,unsigned long message)2367 static int ptrace_do_notify(int signr, int exit_code, int why, unsigned long message)
2368 {
2369 kernel_siginfo_t info;
2370
2371 clear_siginfo(&info);
2372 info.si_signo = signr;
2373 info.si_code = exit_code;
2374 info.si_pid = task_pid_vnr(current);
2375 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2376
2377 /* Let the debugger run. */
2378 return ptrace_stop(exit_code, why, message, &info);
2379 }
2380
ptrace_notify(int exit_code,unsigned long message)2381 int ptrace_notify(int exit_code, unsigned long message)
2382 {
2383 int signr;
2384
2385 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2386 if (unlikely(task_work_pending(current)))
2387 task_work_run();
2388
2389 spin_lock_irq(¤t->sighand->siglock);
2390 signr = ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED, message);
2391 spin_unlock_irq(¤t->sighand->siglock);
2392 return signr;
2393 }
2394
2395 /**
2396 * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2397 * @signr: signr causing group stop if initiating
2398 *
2399 * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2400 * and participate in it. If already set, participate in the existing
2401 * group stop. If participated in a group stop (and thus slept), %true is
2402 * returned with siglock released.
2403 *
2404 * If ptraced, this function doesn't handle stop itself. Instead,
2405 * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2406 * untouched. The caller must ensure that INTERRUPT trap handling takes
2407 * places afterwards.
2408 *
2409 * CONTEXT:
2410 * Must be called with @current->sighand->siglock held, which is released
2411 * on %true return.
2412 *
2413 * RETURNS:
2414 * %false if group stop is already cancelled or ptrace trap is scheduled.
2415 * %true if participated in group stop.
2416 */
do_signal_stop(int signr)2417 static bool do_signal_stop(int signr)
2418 __releases(¤t->sighand->siglock)
2419 {
2420 struct signal_struct *sig = current->signal;
2421
2422 if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2423 unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2424 struct task_struct *t;
2425
2426 /* signr will be recorded in task->jobctl for retries */
2427 WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2428
2429 if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2430 unlikely(sig->flags & SIGNAL_GROUP_EXIT) ||
2431 unlikely(sig->group_exec_task))
2432 return false;
2433 /*
2434 * There is no group stop already in progress. We must
2435 * initiate one now.
2436 *
2437 * While ptraced, a task may be resumed while group stop is
2438 * still in effect and then receive a stop signal and
2439 * initiate another group stop. This deviates from the
2440 * usual behavior as two consecutive stop signals can't
2441 * cause two group stops when !ptraced. That is why we
2442 * also check !task_is_stopped(t) below.
2443 *
2444 * The condition can be distinguished by testing whether
2445 * SIGNAL_STOP_STOPPED is already set. Don't generate
2446 * group_exit_code in such case.
2447 *
2448 * This is not necessary for SIGNAL_STOP_CONTINUED because
2449 * an intervening stop signal is required to cause two
2450 * continued events regardless of ptrace.
2451 */
2452 if (!(sig->flags & SIGNAL_STOP_STOPPED))
2453 sig->group_exit_code = signr;
2454
2455 sig->group_stop_count = 0;
2456
2457 if (task_set_jobctl_pending(current, signr | gstop))
2458 sig->group_stop_count++;
2459
2460 t = current;
2461 while_each_thread(current, t) {
2462 /*
2463 * Setting state to TASK_STOPPED for a group
2464 * stop is always done with the siglock held,
2465 * so this check has no races.
2466 */
2467 if (!task_is_stopped(t) &&
2468 task_set_jobctl_pending(t, signr | gstop)) {
2469 sig->group_stop_count++;
2470 if (likely(!(t->ptrace & PT_SEIZED)))
2471 signal_wake_up(t, 0);
2472 else
2473 ptrace_trap_notify(t);
2474 }
2475 }
2476 }
2477
2478 if (likely(!current->ptrace)) {
2479 int notify = 0;
2480
2481 /*
2482 * If there are no other threads in the group, or if there
2483 * is a group stop in progress and we are the last to stop,
2484 * report to the parent.
2485 */
2486 if (task_participate_group_stop(current))
2487 notify = CLD_STOPPED;
2488
2489 current->jobctl |= JOBCTL_STOPPED;
2490 set_special_state(TASK_STOPPED);
2491 spin_unlock_irq(¤t->sighand->siglock);
2492
2493 /*
2494 * Notify the parent of the group stop completion. Because
2495 * we're not holding either the siglock or tasklist_lock
2496 * here, ptracer may attach inbetween; however, this is for
2497 * group stop and should always be delivered to the real
2498 * parent of the group leader. The new ptracer will get
2499 * its notification when this task transitions into
2500 * TASK_TRACED.
2501 */
2502 if (notify) {
2503 read_lock(&tasklist_lock);
2504 do_notify_parent_cldstop(current, false, notify);
2505 read_unlock(&tasklist_lock);
2506 }
2507
2508 /* Now we don't run again until woken by SIGCONT or SIGKILL */
2509 cgroup_enter_frozen();
2510 schedule();
2511 return true;
2512 } else {
2513 /*
2514 * While ptraced, group stop is handled by STOP trap.
2515 * Schedule it and let the caller deal with it.
2516 */
2517 task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2518 return false;
2519 }
2520 }
2521
2522 /**
2523 * do_jobctl_trap - take care of ptrace jobctl traps
2524 *
2525 * When PT_SEIZED, it's used for both group stop and explicit
2526 * SEIZE/INTERRUPT traps. Both generate PTRACE_EVENT_STOP trap with
2527 * accompanying siginfo. If stopped, lower eight bits of exit_code contain
2528 * the stop signal; otherwise, %SIGTRAP.
2529 *
2530 * When !PT_SEIZED, it's used only for group stop trap with stop signal
2531 * number as exit_code and no siginfo.
2532 *
2533 * CONTEXT:
2534 * Must be called with @current->sighand->siglock held, which may be
2535 * released and re-acquired before returning with intervening sleep.
2536 */
do_jobctl_trap(void)2537 static void do_jobctl_trap(void)
2538 {
2539 struct signal_struct *signal = current->signal;
2540 int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2541
2542 if (current->ptrace & PT_SEIZED) {
2543 if (!signal->group_stop_count &&
2544 !(signal->flags & SIGNAL_STOP_STOPPED))
2545 signr = SIGTRAP;
2546 WARN_ON_ONCE(!signr);
2547 ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2548 CLD_STOPPED, 0);
2549 } else {
2550 WARN_ON_ONCE(!signr);
2551 ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2552 }
2553 }
2554
2555 /**
2556 * do_freezer_trap - handle the freezer jobctl trap
2557 *
2558 * Puts the task into frozen state, if only the task is not about to quit.
2559 * In this case it drops JOBCTL_TRAP_FREEZE.
2560 *
2561 * CONTEXT:
2562 * Must be called with @current->sighand->siglock held,
2563 * which is always released before returning.
2564 */
do_freezer_trap(void)2565 static void do_freezer_trap(void)
2566 __releases(¤t->sighand->siglock)
2567 {
2568 /*
2569 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2570 * let's make another loop to give it a chance to be handled.
2571 * In any case, we'll return back.
2572 */
2573 if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2574 JOBCTL_TRAP_FREEZE) {
2575 spin_unlock_irq(¤t->sighand->siglock);
2576 return;
2577 }
2578
2579 /*
2580 * Now we're sure that there is no pending fatal signal and no
2581 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2582 * immediately (if there is a non-fatal signal pending), and
2583 * put the task into sleep.
2584 */
2585 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2586 clear_thread_flag(TIF_SIGPENDING);
2587 spin_unlock_irq(¤t->sighand->siglock);
2588 cgroup_enter_frozen();
2589 schedule();
2590 }
2591
ptrace_signal(int signr,kernel_siginfo_t * info,enum pid_type type)2592 static int ptrace_signal(int signr, kernel_siginfo_t *info, enum pid_type type)
2593 {
2594 /*
2595 * We do not check sig_kernel_stop(signr) but set this marker
2596 * unconditionally because we do not know whether debugger will
2597 * change signr. This flag has no meaning unless we are going
2598 * to stop after return from ptrace_stop(). In this case it will
2599 * be checked in do_signal_stop(), we should only stop if it was
2600 * not cleared by SIGCONT while we were sleeping. See also the
2601 * comment in dequeue_signal().
2602 */
2603 current->jobctl |= JOBCTL_STOP_DEQUEUED;
2604 signr = ptrace_stop(signr, CLD_TRAPPED, 0, info);
2605
2606 /* We're back. Did the debugger cancel the sig? */
2607 if (signr == 0)
2608 return signr;
2609
2610 /*
2611 * Update the siginfo structure if the signal has
2612 * changed. If the debugger wanted something
2613 * specific in the siginfo structure then it should
2614 * have updated *info via PTRACE_SETSIGINFO.
2615 */
2616 if (signr != info->si_signo) {
2617 clear_siginfo(info);
2618 info->si_signo = signr;
2619 info->si_errno = 0;
2620 info->si_code = SI_USER;
2621 rcu_read_lock();
2622 info->si_pid = task_pid_vnr(current->parent);
2623 info->si_uid = from_kuid_munged(current_user_ns(),
2624 task_uid(current->parent));
2625 rcu_read_unlock();
2626 }
2627
2628 /* If the (new) signal is now blocked, requeue it. */
2629 if (sigismember(¤t->blocked, signr) ||
2630 fatal_signal_pending(current)) {
2631 send_signal_locked(signr, info, current, type);
2632 signr = 0;
2633 }
2634
2635 return signr;
2636 }
2637
hide_si_addr_tag_bits(struct ksignal * ksig)2638 static void hide_si_addr_tag_bits(struct ksignal *ksig)
2639 {
2640 switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2641 case SIL_FAULT:
2642 case SIL_FAULT_TRAPNO:
2643 case SIL_FAULT_MCEERR:
2644 case SIL_FAULT_BNDERR:
2645 case SIL_FAULT_PKUERR:
2646 case SIL_FAULT_PERF_EVENT:
2647 ksig->info.si_addr = arch_untagged_si_addr(
2648 ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2649 break;
2650 case SIL_KILL:
2651 case SIL_TIMER:
2652 case SIL_POLL:
2653 case SIL_CHLD:
2654 case SIL_RT:
2655 case SIL_SYS:
2656 break;
2657 }
2658 }
2659
get_signal(struct ksignal * ksig)2660 bool get_signal(struct ksignal *ksig)
2661 {
2662 struct sighand_struct *sighand = current->sighand;
2663 struct signal_struct *signal = current->signal;
2664 int signr;
2665
2666 clear_notify_signal();
2667 if (unlikely(task_work_pending(current)))
2668 task_work_run();
2669
2670 if (!task_sigpending(current))
2671 return false;
2672
2673 if (unlikely(uprobe_deny_signal()))
2674 return false;
2675
2676 /*
2677 * Do this once, we can't return to user-mode if freezing() == T.
2678 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2679 * thus do not need another check after return.
2680 */
2681 try_to_freeze();
2682
2683 relock:
2684 spin_lock_irq(&sighand->siglock);
2685
2686 /*
2687 * Every stopped thread goes here after wakeup. Check to see if
2688 * we should notify the parent, prepare_signal(SIGCONT) encodes
2689 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2690 */
2691 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2692 int why;
2693
2694 if (signal->flags & SIGNAL_CLD_CONTINUED)
2695 why = CLD_CONTINUED;
2696 else
2697 why = CLD_STOPPED;
2698
2699 signal->flags &= ~SIGNAL_CLD_MASK;
2700
2701 spin_unlock_irq(&sighand->siglock);
2702
2703 /*
2704 * Notify the parent that we're continuing. This event is
2705 * always per-process and doesn't make whole lot of sense
2706 * for ptracers, who shouldn't consume the state via
2707 * wait(2) either, but, for backward compatibility, notify
2708 * the ptracer of the group leader too unless it's gonna be
2709 * a duplicate.
2710 */
2711 read_lock(&tasklist_lock);
2712 do_notify_parent_cldstop(current, false, why);
2713
2714 if (ptrace_reparented(current->group_leader))
2715 do_notify_parent_cldstop(current->group_leader,
2716 true, why);
2717 read_unlock(&tasklist_lock);
2718
2719 goto relock;
2720 }
2721
2722 for (;;) {
2723 struct k_sigaction *ka;
2724 enum pid_type type;
2725
2726 /* Has this task already been marked for death? */
2727 if ((signal->flags & SIGNAL_GROUP_EXIT) ||
2728 signal->group_exec_task) {
2729 clear_siginfo(&ksig->info);
2730 ksig->info.si_signo = signr = SIGKILL;
2731 sigdelset(¤t->pending.signal, SIGKILL);
2732 trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2733 &sighand->action[SIGKILL - 1]);
2734 recalc_sigpending();
2735 goto fatal;
2736 }
2737
2738 if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2739 do_signal_stop(0))
2740 goto relock;
2741
2742 if (unlikely(current->jobctl &
2743 (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2744 if (current->jobctl & JOBCTL_TRAP_MASK) {
2745 do_jobctl_trap();
2746 spin_unlock_irq(&sighand->siglock);
2747 } else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2748 do_freezer_trap();
2749
2750 goto relock;
2751 }
2752
2753 /*
2754 * If the task is leaving the frozen state, let's update
2755 * cgroup counters and reset the frozen bit.
2756 */
2757 if (unlikely(cgroup_task_frozen(current))) {
2758 spin_unlock_irq(&sighand->siglock);
2759 cgroup_leave_frozen(false);
2760 goto relock;
2761 }
2762
2763 /*
2764 * Signals generated by the execution of an instruction
2765 * need to be delivered before any other pending signals
2766 * so that the instruction pointer in the signal stack
2767 * frame points to the faulting instruction.
2768 */
2769 type = PIDTYPE_PID;
2770 signr = dequeue_synchronous_signal(&ksig->info);
2771 if (!signr)
2772 signr = dequeue_signal(current, ¤t->blocked,
2773 &ksig->info, &type);
2774
2775 if (!signr)
2776 break; /* will return 0 */
2777
2778 if (unlikely(current->ptrace) && (signr != SIGKILL) &&
2779 !(sighand->action[signr -1].sa.sa_flags & SA_IMMUTABLE)) {
2780 signr = ptrace_signal(signr, &ksig->info, type);
2781 if (!signr)
2782 continue;
2783 }
2784
2785 ka = &sighand->action[signr-1];
2786
2787 /* Trace actually delivered signals. */
2788 trace_signal_deliver(signr, &ksig->info, ka);
2789
2790 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
2791 continue;
2792 if (ka->sa.sa_handler != SIG_DFL) {
2793 /* Run the handler. */
2794 ksig->ka = *ka;
2795
2796 if (ka->sa.sa_flags & SA_ONESHOT)
2797 ka->sa.sa_handler = SIG_DFL;
2798
2799 break; /* will return non-zero "signr" value */
2800 }
2801
2802 /*
2803 * Now we are doing the default action for this signal.
2804 */
2805 if (sig_kernel_ignore(signr)) /* Default is nothing. */
2806 continue;
2807
2808 /*
2809 * Global init gets no signals it doesn't want.
2810 * Container-init gets no signals it doesn't want from same
2811 * container.
2812 *
2813 * Note that if global/container-init sees a sig_kernel_only()
2814 * signal here, the signal must have been generated internally
2815 * or must have come from an ancestor namespace. In either
2816 * case, the signal cannot be dropped.
2817 */
2818 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2819 !sig_kernel_only(signr))
2820 continue;
2821
2822 if (sig_kernel_stop(signr)) {
2823 /*
2824 * The default action is to stop all threads in
2825 * the thread group. The job control signals
2826 * do nothing in an orphaned pgrp, but SIGSTOP
2827 * always works. Note that siglock needs to be
2828 * dropped during the call to is_orphaned_pgrp()
2829 * because of lock ordering with tasklist_lock.
2830 * This allows an intervening SIGCONT to be posted.
2831 * We need to check for that and bail out if necessary.
2832 */
2833 if (signr != SIGSTOP) {
2834 spin_unlock_irq(&sighand->siglock);
2835
2836 /* signals can be posted during this window */
2837
2838 if (is_current_pgrp_orphaned())
2839 goto relock;
2840
2841 spin_lock_irq(&sighand->siglock);
2842 }
2843
2844 if (likely(do_signal_stop(ksig->info.si_signo))) {
2845 /* It released the siglock. */
2846 goto relock;
2847 }
2848
2849 /*
2850 * We didn't actually stop, due to a race
2851 * with SIGCONT or something like that.
2852 */
2853 continue;
2854 }
2855
2856 fatal:
2857 spin_unlock_irq(&sighand->siglock);
2858 if (unlikely(cgroup_task_frozen(current)))
2859 cgroup_leave_frozen(true);
2860
2861 /*
2862 * Anything else is fatal, maybe with a core dump.
2863 */
2864 current->flags |= PF_SIGNALED;
2865
2866 if (sig_kernel_coredump(signr)) {
2867 if (print_fatal_signals)
2868 print_fatal_signal(ksig->info.si_signo);
2869 proc_coredump_connector(current);
2870 /*
2871 * If it was able to dump core, this kills all
2872 * other threads in the group and synchronizes with
2873 * their demise. If we lost the race with another
2874 * thread getting here, it set group_exit_code
2875 * first and our do_group_exit call below will use
2876 * that value and ignore the one we pass it.
2877 */
2878 do_coredump(&ksig->info);
2879 }
2880
2881 /*
2882 * PF_USER_WORKER threads will catch and exit on fatal signals
2883 * themselves. They have cleanup that must be performed, so
2884 * we cannot call do_exit() on their behalf.
2885 */
2886 if (current->flags & PF_USER_WORKER)
2887 goto out;
2888
2889 /*
2890 * Death signals, no core dump.
2891 */
2892 do_group_exit(ksig->info.si_signo);
2893 /* NOTREACHED */
2894 }
2895 spin_unlock_irq(&sighand->siglock);
2896 out:
2897 ksig->sig = signr;
2898
2899 if (!(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
2900 hide_si_addr_tag_bits(ksig);
2901
2902 return ksig->sig > 0;
2903 }
2904
2905 /**
2906 * signal_delivered - called after signal delivery to update blocked signals
2907 * @ksig: kernel signal struct
2908 * @stepping: nonzero if debugger single-step or block-step in use
2909 *
2910 * This function should be called when a signal has successfully been
2911 * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2912 * is always blocked), and the signal itself is blocked unless %SA_NODEFER
2913 * is set in @ksig->ka.sa.sa_flags. Tracing is notified.
2914 */
signal_delivered(struct ksignal * ksig,int stepping)2915 static void signal_delivered(struct ksignal *ksig, int stepping)
2916 {
2917 sigset_t blocked;
2918
2919 /* A signal was successfully delivered, and the
2920 saved sigmask was stored on the signal frame,
2921 and will be restored by sigreturn. So we can
2922 simply clear the restore sigmask flag. */
2923 clear_restore_sigmask();
2924
2925 sigorsets(&blocked, ¤t->blocked, &ksig->ka.sa.sa_mask);
2926 if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2927 sigaddset(&blocked, ksig->sig);
2928 set_current_blocked(&blocked);
2929 if (current->sas_ss_flags & SS_AUTODISARM)
2930 sas_ss_reset(current);
2931 if (stepping)
2932 ptrace_notify(SIGTRAP, 0);
2933 }
2934
signal_setup_done(int failed,struct ksignal * ksig,int stepping)2935 void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2936 {
2937 if (failed)
2938 force_sigsegv(ksig->sig);
2939 else
2940 signal_delivered(ksig, stepping);
2941 }
2942
2943 /*
2944 * It could be that complete_signal() picked us to notify about the
2945 * group-wide signal. Other threads should be notified now to take
2946 * the shared signals in @which since we will not.
2947 */
retarget_shared_pending(struct task_struct * tsk,sigset_t * which)2948 static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2949 {
2950 sigset_t retarget;
2951 struct task_struct *t;
2952
2953 sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2954 if (sigisemptyset(&retarget))
2955 return;
2956
2957 t = tsk;
2958 while_each_thread(tsk, t) {
2959 if (t->flags & PF_EXITING)
2960 continue;
2961
2962 if (!has_pending_signals(&retarget, &t->blocked))
2963 continue;
2964 /* Remove the signals this thread can handle. */
2965 sigandsets(&retarget, &retarget, &t->blocked);
2966
2967 if (!task_sigpending(t))
2968 signal_wake_up(t, 0);
2969
2970 if (sigisemptyset(&retarget))
2971 break;
2972 }
2973 }
2974
exit_signals(struct task_struct * tsk)2975 void exit_signals(struct task_struct *tsk)
2976 {
2977 int group_stop = 0;
2978 sigset_t unblocked;
2979
2980 /*
2981 * @tsk is about to have PF_EXITING set - lock out users which
2982 * expect stable threadgroup.
2983 */
2984 cgroup_threadgroup_change_begin(tsk);
2985
2986 if (thread_group_empty(tsk) || (tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
2987 sched_mm_cid_exit_signals(tsk);
2988 tsk->flags |= PF_EXITING;
2989 cgroup_threadgroup_change_end(tsk);
2990 return;
2991 }
2992
2993 spin_lock_irq(&tsk->sighand->siglock);
2994 /*
2995 * From now this task is not visible for group-wide signals,
2996 * see wants_signal(), do_signal_stop().
2997 */
2998 sched_mm_cid_exit_signals(tsk);
2999 tsk->flags |= PF_EXITING;
3000
3001 cgroup_threadgroup_change_end(tsk);
3002
3003 if (!task_sigpending(tsk))
3004 goto out;
3005
3006 unblocked = tsk->blocked;
3007 signotset(&unblocked);
3008 retarget_shared_pending(tsk, &unblocked);
3009
3010 if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
3011 task_participate_group_stop(tsk))
3012 group_stop = CLD_STOPPED;
3013 out:
3014 spin_unlock_irq(&tsk->sighand->siglock);
3015
3016 /*
3017 * If group stop has completed, deliver the notification. This
3018 * should always go to the real parent of the group leader.
3019 */
3020 if (unlikely(group_stop)) {
3021 read_lock(&tasklist_lock);
3022 do_notify_parent_cldstop(tsk, false, group_stop);
3023 read_unlock(&tasklist_lock);
3024 }
3025 }
3026
3027 /*
3028 * System call entry points.
3029 */
3030
3031 /**
3032 * sys_restart_syscall - restart a system call
3033 */
SYSCALL_DEFINE0(restart_syscall)3034 SYSCALL_DEFINE0(restart_syscall)
3035 {
3036 struct restart_block *restart = ¤t->restart_block;
3037 return restart->fn(restart);
3038 }
3039
do_no_restart_syscall(struct restart_block * param)3040 long do_no_restart_syscall(struct restart_block *param)
3041 {
3042 return -EINTR;
3043 }
3044
__set_task_blocked(struct task_struct * tsk,const sigset_t * newset)3045 static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
3046 {
3047 if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
3048 sigset_t newblocked;
3049 /* A set of now blocked but previously unblocked signals. */
3050 sigandnsets(&newblocked, newset, ¤t->blocked);
3051 retarget_shared_pending(tsk, &newblocked);
3052 }
3053 tsk->blocked = *newset;
3054 recalc_sigpending();
3055 }
3056
3057 /**
3058 * set_current_blocked - change current->blocked mask
3059 * @newset: new mask
3060 *
3061 * It is wrong to change ->blocked directly, this helper should be used
3062 * to ensure the process can't miss a shared signal we are going to block.
3063 */
set_current_blocked(sigset_t * newset)3064 void set_current_blocked(sigset_t *newset)
3065 {
3066 sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
3067 __set_current_blocked(newset);
3068 }
3069
__set_current_blocked(const sigset_t * newset)3070 void __set_current_blocked(const sigset_t *newset)
3071 {
3072 struct task_struct *tsk = current;
3073
3074 /*
3075 * In case the signal mask hasn't changed, there is nothing we need
3076 * to do. The current->blocked shouldn't be modified by other task.
3077 */
3078 if (sigequalsets(&tsk->blocked, newset))
3079 return;
3080
3081 spin_lock_irq(&tsk->sighand->siglock);
3082 __set_task_blocked(tsk, newset);
3083 spin_unlock_irq(&tsk->sighand->siglock);
3084 }
3085
3086 /*
3087 * This is also useful for kernel threads that want to temporarily
3088 * (or permanently) block certain signals.
3089 *
3090 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
3091 * interface happily blocks "unblockable" signals like SIGKILL
3092 * and friends.
3093 */
sigprocmask(int how,sigset_t * set,sigset_t * oldset)3094 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
3095 {
3096 struct task_struct *tsk = current;
3097 sigset_t newset;
3098
3099 /* Lockless, only current can change ->blocked, never from irq */
3100 if (oldset)
3101 *oldset = tsk->blocked;
3102
3103 switch (how) {
3104 case SIG_BLOCK:
3105 sigorsets(&newset, &tsk->blocked, set);
3106 break;
3107 case SIG_UNBLOCK:
3108 sigandnsets(&newset, &tsk->blocked, set);
3109 break;
3110 case SIG_SETMASK:
3111 newset = *set;
3112 break;
3113 default:
3114 return -EINVAL;
3115 }
3116
3117 __set_current_blocked(&newset);
3118 return 0;
3119 }
3120 EXPORT_SYMBOL(sigprocmask);
3121
3122 /*
3123 * The api helps set app-provided sigmasks.
3124 *
3125 * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3126 * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3127 *
3128 * Note that it does set_restore_sigmask() in advance, so it must be always
3129 * paired with restore_saved_sigmask_unless() before return from syscall.
3130 */
set_user_sigmask(const sigset_t __user * umask,size_t sigsetsize)3131 int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3132 {
3133 sigset_t kmask;
3134
3135 if (!umask)
3136 return 0;
3137 if (sigsetsize != sizeof(sigset_t))
3138 return -EINVAL;
3139 if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3140 return -EFAULT;
3141
3142 set_restore_sigmask();
3143 current->saved_sigmask = current->blocked;
3144 set_current_blocked(&kmask);
3145
3146 return 0;
3147 }
3148
3149 #ifdef CONFIG_COMPAT
set_compat_user_sigmask(const compat_sigset_t __user * umask,size_t sigsetsize)3150 int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3151 size_t sigsetsize)
3152 {
3153 sigset_t kmask;
3154
3155 if (!umask)
3156 return 0;
3157 if (sigsetsize != sizeof(compat_sigset_t))
3158 return -EINVAL;
3159 if (get_compat_sigset(&kmask, umask))
3160 return -EFAULT;
3161
3162 set_restore_sigmask();
3163 current->saved_sigmask = current->blocked;
3164 set_current_blocked(&kmask);
3165
3166 return 0;
3167 }
3168 #endif
3169
3170 /**
3171 * sys_rt_sigprocmask - change the list of currently blocked signals
3172 * @how: whether to add, remove, or set signals
3173 * @nset: stores pending signals
3174 * @oset: previous value of signal mask if non-null
3175 * @sigsetsize: size of sigset_t type
3176 */
SYSCALL_DEFINE4(rt_sigprocmask,int,how,sigset_t __user *,nset,sigset_t __user *,oset,size_t,sigsetsize)3177 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3178 sigset_t __user *, oset, size_t, sigsetsize)
3179 {
3180 sigset_t old_set, new_set;
3181 int error;
3182
3183 /* XXX: Don't preclude handling different sized sigset_t's. */
3184 if (sigsetsize != sizeof(sigset_t))
3185 return -EINVAL;
3186
3187 old_set = current->blocked;
3188
3189 if (nset) {
3190 if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3191 return -EFAULT;
3192 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3193
3194 error = sigprocmask(how, &new_set, NULL);
3195 if (error)
3196 return error;
3197 }
3198
3199 if (oset) {
3200 if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3201 return -EFAULT;
3202 }
3203
3204 return 0;
3205 }
3206
3207 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(rt_sigprocmask,int,how,compat_sigset_t __user *,nset,compat_sigset_t __user *,oset,compat_size_t,sigsetsize)3208 COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3209 compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3210 {
3211 sigset_t old_set = current->blocked;
3212
3213 /* XXX: Don't preclude handling different sized sigset_t's. */
3214 if (sigsetsize != sizeof(sigset_t))
3215 return -EINVAL;
3216
3217 if (nset) {
3218 sigset_t new_set;
3219 int error;
3220 if (get_compat_sigset(&new_set, nset))
3221 return -EFAULT;
3222 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3223
3224 error = sigprocmask(how, &new_set, NULL);
3225 if (error)
3226 return error;
3227 }
3228 return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3229 }
3230 #endif
3231
do_sigpending(sigset_t * set)3232 static void do_sigpending(sigset_t *set)
3233 {
3234 spin_lock_irq(¤t->sighand->siglock);
3235 sigorsets(set, ¤t->pending.signal,
3236 ¤t->signal->shared_pending.signal);
3237 spin_unlock_irq(¤t->sighand->siglock);
3238
3239 /* Outside the lock because only this thread touches it. */
3240 sigandsets(set, ¤t->blocked, set);
3241 }
3242
3243 /**
3244 * sys_rt_sigpending - examine a pending signal that has been raised
3245 * while blocked
3246 * @uset: stores pending signals
3247 * @sigsetsize: size of sigset_t type or larger
3248 */
SYSCALL_DEFINE2(rt_sigpending,sigset_t __user *,uset,size_t,sigsetsize)3249 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3250 {
3251 sigset_t set;
3252
3253 if (sigsetsize > sizeof(*uset))
3254 return -EINVAL;
3255
3256 do_sigpending(&set);
3257
3258 if (copy_to_user(uset, &set, sigsetsize))
3259 return -EFAULT;
3260
3261 return 0;
3262 }
3263
3264 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(rt_sigpending,compat_sigset_t __user *,uset,compat_size_t,sigsetsize)3265 COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3266 compat_size_t, sigsetsize)
3267 {
3268 sigset_t set;
3269
3270 if (sigsetsize > sizeof(*uset))
3271 return -EINVAL;
3272
3273 do_sigpending(&set);
3274
3275 return put_compat_sigset(uset, &set, sigsetsize);
3276 }
3277 #endif
3278
3279 static const struct {
3280 unsigned char limit, layout;
3281 } sig_sicodes[] = {
3282 [SIGILL] = { NSIGILL, SIL_FAULT },
3283 [SIGFPE] = { NSIGFPE, SIL_FAULT },
3284 [SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3285 [SIGBUS] = { NSIGBUS, SIL_FAULT },
3286 [SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3287 #if defined(SIGEMT)
3288 [SIGEMT] = { NSIGEMT, SIL_FAULT },
3289 #endif
3290 [SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3291 [SIGPOLL] = { NSIGPOLL, SIL_POLL },
3292 [SIGSYS] = { NSIGSYS, SIL_SYS },
3293 };
3294
known_siginfo_layout(unsigned sig,int si_code)3295 static bool known_siginfo_layout(unsigned sig, int si_code)
3296 {
3297 if (si_code == SI_KERNEL)
3298 return true;
3299 else if ((si_code > SI_USER)) {
3300 if (sig_specific_sicodes(sig)) {
3301 if (si_code <= sig_sicodes[sig].limit)
3302 return true;
3303 }
3304 else if (si_code <= NSIGPOLL)
3305 return true;
3306 }
3307 else if (si_code >= SI_DETHREAD)
3308 return true;
3309 else if (si_code == SI_ASYNCNL)
3310 return true;
3311 return false;
3312 }
3313
siginfo_layout(unsigned sig,int si_code)3314 enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3315 {
3316 enum siginfo_layout layout = SIL_KILL;
3317 if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3318 if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3319 (si_code <= sig_sicodes[sig].limit)) {
3320 layout = sig_sicodes[sig].layout;
3321 /* Handle the exceptions */
3322 if ((sig == SIGBUS) &&
3323 (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3324 layout = SIL_FAULT_MCEERR;
3325 else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3326 layout = SIL_FAULT_BNDERR;
3327 #ifdef SEGV_PKUERR
3328 else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3329 layout = SIL_FAULT_PKUERR;
3330 #endif
3331 else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3332 layout = SIL_FAULT_PERF_EVENT;
3333 else if (IS_ENABLED(CONFIG_SPARC) &&
3334 (sig == SIGILL) && (si_code == ILL_ILLTRP))
3335 layout = SIL_FAULT_TRAPNO;
3336 else if (IS_ENABLED(CONFIG_ALPHA) &&
3337 ((sig == SIGFPE) ||
3338 ((sig == SIGTRAP) && (si_code == TRAP_UNK))))
3339 layout = SIL_FAULT_TRAPNO;
3340 }
3341 else if (si_code <= NSIGPOLL)
3342 layout = SIL_POLL;
3343 } else {
3344 if (si_code == SI_TIMER)
3345 layout = SIL_TIMER;
3346 else if (si_code == SI_SIGIO)
3347 layout = SIL_POLL;
3348 else if (si_code < 0)
3349 layout = SIL_RT;
3350 }
3351 return layout;
3352 }
3353
si_expansion(const siginfo_t __user * info)3354 static inline char __user *si_expansion(const siginfo_t __user *info)
3355 {
3356 return ((char __user *)info) + sizeof(struct kernel_siginfo);
3357 }
3358
copy_siginfo_to_user(siginfo_t __user * to,const kernel_siginfo_t * from)3359 int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3360 {
3361 char __user *expansion = si_expansion(to);
3362 if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3363 return -EFAULT;
3364 if (clear_user(expansion, SI_EXPANSION_SIZE))
3365 return -EFAULT;
3366 return 0;
3367 }
3368
post_copy_siginfo_from_user(kernel_siginfo_t * info,const siginfo_t __user * from)3369 static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3370 const siginfo_t __user *from)
3371 {
3372 if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3373 char __user *expansion = si_expansion(from);
3374 char buf[SI_EXPANSION_SIZE];
3375 int i;
3376 /*
3377 * An unknown si_code might need more than
3378 * sizeof(struct kernel_siginfo) bytes. Verify all of the
3379 * extra bytes are 0. This guarantees copy_siginfo_to_user
3380 * will return this data to userspace exactly.
3381 */
3382 if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3383 return -EFAULT;
3384 for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3385 if (buf[i] != 0)
3386 return -E2BIG;
3387 }
3388 }
3389 return 0;
3390 }
3391
__copy_siginfo_from_user(int signo,kernel_siginfo_t * to,const siginfo_t __user * from)3392 static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3393 const siginfo_t __user *from)
3394 {
3395 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3396 return -EFAULT;
3397 to->si_signo = signo;
3398 return post_copy_siginfo_from_user(to, from);
3399 }
3400
copy_siginfo_from_user(kernel_siginfo_t * to,const siginfo_t __user * from)3401 int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3402 {
3403 if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3404 return -EFAULT;
3405 return post_copy_siginfo_from_user(to, from);
3406 }
3407
3408 #ifdef CONFIG_COMPAT
3409 /**
3410 * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3411 * @to: compat siginfo destination
3412 * @from: kernel siginfo source
3413 *
3414 * Note: This function does not work properly for the SIGCHLD on x32, but
3415 * fortunately it doesn't have to. The only valid callers for this function are
3416 * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3417 * The latter does not care because SIGCHLD will never cause a coredump.
3418 */
copy_siginfo_to_external32(struct compat_siginfo * to,const struct kernel_siginfo * from)3419 void copy_siginfo_to_external32(struct compat_siginfo *to,
3420 const struct kernel_siginfo *from)
3421 {
3422 memset(to, 0, sizeof(*to));
3423
3424 to->si_signo = from->si_signo;
3425 to->si_errno = from->si_errno;
3426 to->si_code = from->si_code;
3427 switch(siginfo_layout(from->si_signo, from->si_code)) {
3428 case SIL_KILL:
3429 to->si_pid = from->si_pid;
3430 to->si_uid = from->si_uid;
3431 break;
3432 case SIL_TIMER:
3433 to->si_tid = from->si_tid;
3434 to->si_overrun = from->si_overrun;
3435 to->si_int = from->si_int;
3436 break;
3437 case SIL_POLL:
3438 to->si_band = from->si_band;
3439 to->si_fd = from->si_fd;
3440 break;
3441 case SIL_FAULT:
3442 to->si_addr = ptr_to_compat(from->si_addr);
3443 break;
3444 case SIL_FAULT_TRAPNO:
3445 to->si_addr = ptr_to_compat(from->si_addr);
3446 to->si_trapno = from->si_trapno;
3447 break;
3448 case SIL_FAULT_MCEERR:
3449 to->si_addr = ptr_to_compat(from->si_addr);
3450 to->si_addr_lsb = from->si_addr_lsb;
3451 break;
3452 case SIL_FAULT_BNDERR:
3453 to->si_addr = ptr_to_compat(from->si_addr);
3454 to->si_lower = ptr_to_compat(from->si_lower);
3455 to->si_upper = ptr_to_compat(from->si_upper);
3456 break;
3457 case SIL_FAULT_PKUERR:
3458 to->si_addr = ptr_to_compat(from->si_addr);
3459 to->si_pkey = from->si_pkey;
3460 break;
3461 case SIL_FAULT_PERF_EVENT:
3462 to->si_addr = ptr_to_compat(from->si_addr);
3463 to->si_perf_data = from->si_perf_data;
3464 to->si_perf_type = from->si_perf_type;
3465 to->si_perf_flags = from->si_perf_flags;
3466 break;
3467 case SIL_CHLD:
3468 to->si_pid = from->si_pid;
3469 to->si_uid = from->si_uid;
3470 to->si_status = from->si_status;
3471 to->si_utime = from->si_utime;
3472 to->si_stime = from->si_stime;
3473 break;
3474 case SIL_RT:
3475 to->si_pid = from->si_pid;
3476 to->si_uid = from->si_uid;
3477 to->si_int = from->si_int;
3478 break;
3479 case SIL_SYS:
3480 to->si_call_addr = ptr_to_compat(from->si_call_addr);
3481 to->si_syscall = from->si_syscall;
3482 to->si_arch = from->si_arch;
3483 break;
3484 }
3485 }
3486
__copy_siginfo_to_user32(struct compat_siginfo __user * to,const struct kernel_siginfo * from)3487 int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3488 const struct kernel_siginfo *from)
3489 {
3490 struct compat_siginfo new;
3491
3492 copy_siginfo_to_external32(&new, from);
3493 if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3494 return -EFAULT;
3495 return 0;
3496 }
3497
post_copy_siginfo_from_user32(kernel_siginfo_t * to,const struct compat_siginfo * from)3498 static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3499 const struct compat_siginfo *from)
3500 {
3501 clear_siginfo(to);
3502 to->si_signo = from->si_signo;
3503 to->si_errno = from->si_errno;
3504 to->si_code = from->si_code;
3505 switch(siginfo_layout(from->si_signo, from->si_code)) {
3506 case SIL_KILL:
3507 to->si_pid = from->si_pid;
3508 to->si_uid = from->si_uid;
3509 break;
3510 case SIL_TIMER:
3511 to->si_tid = from->si_tid;
3512 to->si_overrun = from->si_overrun;
3513 to->si_int = from->si_int;
3514 break;
3515 case SIL_POLL:
3516 to->si_band = from->si_band;
3517 to->si_fd = from->si_fd;
3518 break;
3519 case SIL_FAULT:
3520 to->si_addr = compat_ptr(from->si_addr);
3521 break;
3522 case SIL_FAULT_TRAPNO:
3523 to->si_addr = compat_ptr(from->si_addr);
3524 to->si_trapno = from->si_trapno;
3525 break;
3526 case SIL_FAULT_MCEERR:
3527 to->si_addr = compat_ptr(from->si_addr);
3528 to->si_addr_lsb = from->si_addr_lsb;
3529 break;
3530 case SIL_FAULT_BNDERR:
3531 to->si_addr = compat_ptr(from->si_addr);
3532 to->si_lower = compat_ptr(from->si_lower);
3533 to->si_upper = compat_ptr(from->si_upper);
3534 break;
3535 case SIL_FAULT_PKUERR:
3536 to->si_addr = compat_ptr(from->si_addr);
3537 to->si_pkey = from->si_pkey;
3538 break;
3539 case SIL_FAULT_PERF_EVENT:
3540 to->si_addr = compat_ptr(from->si_addr);
3541 to->si_perf_data = from->si_perf_data;
3542 to->si_perf_type = from->si_perf_type;
3543 to->si_perf_flags = from->si_perf_flags;
3544 break;
3545 case SIL_CHLD:
3546 to->si_pid = from->si_pid;
3547 to->si_uid = from->si_uid;
3548 to->si_status = from->si_status;
3549 #ifdef CONFIG_X86_X32_ABI
3550 if (in_x32_syscall()) {
3551 to->si_utime = from->_sifields._sigchld_x32._utime;
3552 to->si_stime = from->_sifields._sigchld_x32._stime;
3553 } else
3554 #endif
3555 {
3556 to->si_utime = from->si_utime;
3557 to->si_stime = from->si_stime;
3558 }
3559 break;
3560 case SIL_RT:
3561 to->si_pid = from->si_pid;
3562 to->si_uid = from->si_uid;
3563 to->si_int = from->si_int;
3564 break;
3565 case SIL_SYS:
3566 to->si_call_addr = compat_ptr(from->si_call_addr);
3567 to->si_syscall = from->si_syscall;
3568 to->si_arch = from->si_arch;
3569 break;
3570 }
3571 return 0;
3572 }
3573
__copy_siginfo_from_user32(int signo,struct kernel_siginfo * to,const struct compat_siginfo __user * ufrom)3574 static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3575 const struct compat_siginfo __user *ufrom)
3576 {
3577 struct compat_siginfo from;
3578
3579 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3580 return -EFAULT;
3581
3582 from.si_signo = signo;
3583 return post_copy_siginfo_from_user32(to, &from);
3584 }
3585
copy_siginfo_from_user32(struct kernel_siginfo * to,const struct compat_siginfo __user * ufrom)3586 int copy_siginfo_from_user32(struct kernel_siginfo *to,
3587 const struct compat_siginfo __user *ufrom)
3588 {
3589 struct compat_siginfo from;
3590
3591 if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3592 return -EFAULT;
3593
3594 return post_copy_siginfo_from_user32(to, &from);
3595 }
3596 #endif /* CONFIG_COMPAT */
3597
3598 /**
3599 * do_sigtimedwait - wait for queued signals specified in @which
3600 * @which: queued signals to wait for
3601 * @info: if non-null, the signal's siginfo is returned here
3602 * @ts: upper bound on process time suspension
3603 */
do_sigtimedwait(const sigset_t * which,kernel_siginfo_t * info,const struct timespec64 * ts)3604 static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3605 const struct timespec64 *ts)
3606 {
3607 ktime_t *to = NULL, timeout = KTIME_MAX;
3608 struct task_struct *tsk = current;
3609 sigset_t mask = *which;
3610 enum pid_type type;
3611 int sig, ret = 0;
3612
3613 if (ts) {
3614 if (!timespec64_valid(ts))
3615 return -EINVAL;
3616 timeout = timespec64_to_ktime(*ts);
3617 to = &timeout;
3618 }
3619
3620 /*
3621 * Invert the set of allowed signals to get those we want to block.
3622 */
3623 sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3624 signotset(&mask);
3625
3626 spin_lock_irq(&tsk->sighand->siglock);
3627 sig = dequeue_signal(tsk, &mask, info, &type);
3628 if (!sig && timeout) {
3629 /*
3630 * None ready, temporarily unblock those we're interested
3631 * while we are sleeping in so that we'll be awakened when
3632 * they arrive. Unblocking is always fine, we can avoid
3633 * set_current_blocked().
3634 */
3635 tsk->real_blocked = tsk->blocked;
3636 sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3637 recalc_sigpending();
3638 spin_unlock_irq(&tsk->sighand->siglock);
3639
3640 __set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
3641 ret = schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3642 HRTIMER_MODE_REL);
3643 spin_lock_irq(&tsk->sighand->siglock);
3644 __set_task_blocked(tsk, &tsk->real_blocked);
3645 sigemptyset(&tsk->real_blocked);
3646 sig = dequeue_signal(tsk, &mask, info, &type);
3647 }
3648 spin_unlock_irq(&tsk->sighand->siglock);
3649
3650 if (sig)
3651 return sig;
3652 return ret ? -EINTR : -EAGAIN;
3653 }
3654
3655 /**
3656 * sys_rt_sigtimedwait - synchronously wait for queued signals specified
3657 * in @uthese
3658 * @uthese: queued signals to wait for
3659 * @uinfo: if non-null, the signal's siginfo is returned here
3660 * @uts: upper bound on process time suspension
3661 * @sigsetsize: size of sigset_t type
3662 */
SYSCALL_DEFINE4(rt_sigtimedwait,const sigset_t __user *,uthese,siginfo_t __user *,uinfo,const struct __kernel_timespec __user *,uts,size_t,sigsetsize)3663 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3664 siginfo_t __user *, uinfo,
3665 const struct __kernel_timespec __user *, uts,
3666 size_t, sigsetsize)
3667 {
3668 sigset_t these;
3669 struct timespec64 ts;
3670 kernel_siginfo_t info;
3671 int ret;
3672
3673 /* XXX: Don't preclude handling different sized sigset_t's. */
3674 if (sigsetsize != sizeof(sigset_t))
3675 return -EINVAL;
3676
3677 if (copy_from_user(&these, uthese, sizeof(these)))
3678 return -EFAULT;
3679
3680 if (uts) {
3681 if (get_timespec64(&ts, uts))
3682 return -EFAULT;
3683 }
3684
3685 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3686
3687 if (ret > 0 && uinfo) {
3688 if (copy_siginfo_to_user(uinfo, &info))
3689 ret = -EFAULT;
3690 }
3691
3692 return ret;
3693 }
3694
3695 #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)3696 SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3697 siginfo_t __user *, uinfo,
3698 const struct old_timespec32 __user *, uts,
3699 size_t, sigsetsize)
3700 {
3701 sigset_t these;
3702 struct timespec64 ts;
3703 kernel_siginfo_t info;
3704 int ret;
3705
3706 if (sigsetsize != sizeof(sigset_t))
3707 return -EINVAL;
3708
3709 if (copy_from_user(&these, uthese, sizeof(these)))
3710 return -EFAULT;
3711
3712 if (uts) {
3713 if (get_old_timespec32(&ts, uts))
3714 return -EFAULT;
3715 }
3716
3717 ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3718
3719 if (ret > 0 && uinfo) {
3720 if (copy_siginfo_to_user(uinfo, &info))
3721 ret = -EFAULT;
3722 }
3723
3724 return ret;
3725 }
3726 #endif
3727
3728 #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)3729 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3730 struct compat_siginfo __user *, uinfo,
3731 struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3732 {
3733 sigset_t s;
3734 struct timespec64 t;
3735 kernel_siginfo_t info;
3736 long ret;
3737
3738 if (sigsetsize != sizeof(sigset_t))
3739 return -EINVAL;
3740
3741 if (get_compat_sigset(&s, uthese))
3742 return -EFAULT;
3743
3744 if (uts) {
3745 if (get_timespec64(&t, uts))
3746 return -EFAULT;
3747 }
3748
3749 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3750
3751 if (ret > 0 && uinfo) {
3752 if (copy_siginfo_to_user32(uinfo, &info))
3753 ret = -EFAULT;
3754 }
3755
3756 return ret;
3757 }
3758
3759 #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)3760 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3761 struct compat_siginfo __user *, uinfo,
3762 struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3763 {
3764 sigset_t s;
3765 struct timespec64 t;
3766 kernel_siginfo_t info;
3767 long ret;
3768
3769 if (sigsetsize != sizeof(sigset_t))
3770 return -EINVAL;
3771
3772 if (get_compat_sigset(&s, uthese))
3773 return -EFAULT;
3774
3775 if (uts) {
3776 if (get_old_timespec32(&t, uts))
3777 return -EFAULT;
3778 }
3779
3780 ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3781
3782 if (ret > 0 && uinfo) {
3783 if (copy_siginfo_to_user32(uinfo, &info))
3784 ret = -EFAULT;
3785 }
3786
3787 return ret;
3788 }
3789 #endif
3790 #endif
3791
prepare_kill_siginfo(int sig,struct kernel_siginfo * info)3792 static inline void prepare_kill_siginfo(int sig, struct kernel_siginfo *info)
3793 {
3794 clear_siginfo(info);
3795 info->si_signo = sig;
3796 info->si_errno = 0;
3797 info->si_code = SI_USER;
3798 info->si_pid = task_tgid_vnr(current);
3799 info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3800 }
3801
3802 /**
3803 * sys_kill - send a signal to a process
3804 * @pid: the PID of the process
3805 * @sig: signal to be sent
3806 */
SYSCALL_DEFINE2(kill,pid_t,pid,int,sig)3807 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3808 {
3809 struct kernel_siginfo info;
3810
3811 prepare_kill_siginfo(sig, &info);
3812
3813 return kill_something_info(sig, &info, pid);
3814 }
3815
3816 /*
3817 * Verify that the signaler and signalee either are in the same pid namespace
3818 * or that the signaler's pid namespace is an ancestor of the signalee's pid
3819 * namespace.
3820 */
access_pidfd_pidns(struct pid * pid)3821 static bool access_pidfd_pidns(struct pid *pid)
3822 {
3823 struct pid_namespace *active = task_active_pid_ns(current);
3824 struct pid_namespace *p = ns_of_pid(pid);
3825
3826 for (;;) {
3827 if (!p)
3828 return false;
3829 if (p == active)
3830 break;
3831 p = p->parent;
3832 }
3833
3834 return true;
3835 }
3836
copy_siginfo_from_user_any(kernel_siginfo_t * kinfo,siginfo_t __user * info)3837 static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3838 siginfo_t __user *info)
3839 {
3840 #ifdef CONFIG_COMPAT
3841 /*
3842 * Avoid hooking up compat syscalls and instead handle necessary
3843 * conversions here. Note, this is a stop-gap measure and should not be
3844 * considered a generic solution.
3845 */
3846 if (in_compat_syscall())
3847 return copy_siginfo_from_user32(
3848 kinfo, (struct compat_siginfo __user *)info);
3849 #endif
3850 return copy_siginfo_from_user(kinfo, info);
3851 }
3852
pidfd_to_pid(const struct file * file)3853 static struct pid *pidfd_to_pid(const struct file *file)
3854 {
3855 struct pid *pid;
3856
3857 pid = pidfd_pid(file);
3858 if (!IS_ERR(pid))
3859 return pid;
3860
3861 return tgid_pidfd_to_pid(file);
3862 }
3863
3864 /**
3865 * sys_pidfd_send_signal - Signal a process through a pidfd
3866 * @pidfd: file descriptor of the process
3867 * @sig: signal to send
3868 * @info: signal info
3869 * @flags: future flags
3870 *
3871 * The syscall currently only signals via PIDTYPE_PID which covers
3872 * kill(<positive-pid>, <signal>. It does not signal threads or process
3873 * groups.
3874 * In order to extend the syscall to threads and process groups the @flags
3875 * argument should be used. In essence, the @flags argument will determine
3876 * what is signaled and not the file descriptor itself. Put in other words,
3877 * grouping is a property of the flags argument not a property of the file
3878 * descriptor.
3879 *
3880 * Return: 0 on success, negative errno on failure
3881 */
SYSCALL_DEFINE4(pidfd_send_signal,int,pidfd,int,sig,siginfo_t __user *,info,unsigned int,flags)3882 SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3883 siginfo_t __user *, info, unsigned int, flags)
3884 {
3885 int ret;
3886 struct fd f;
3887 struct pid *pid;
3888 kernel_siginfo_t kinfo;
3889
3890 /* Enforce flags be set to 0 until we add an extension. */
3891 if (flags)
3892 return -EINVAL;
3893
3894 f = fdget(pidfd);
3895 if (!f.file)
3896 return -EBADF;
3897
3898 /* Is this a pidfd? */
3899 pid = pidfd_to_pid(f.file);
3900 if (IS_ERR(pid)) {
3901 ret = PTR_ERR(pid);
3902 goto err;
3903 }
3904
3905 ret = -EINVAL;
3906 if (!access_pidfd_pidns(pid))
3907 goto err;
3908
3909 if (info) {
3910 ret = copy_siginfo_from_user_any(&kinfo, info);
3911 if (unlikely(ret))
3912 goto err;
3913
3914 ret = -EINVAL;
3915 if (unlikely(sig != kinfo.si_signo))
3916 goto err;
3917
3918 /* Only allow sending arbitrary signals to yourself. */
3919 ret = -EPERM;
3920 if ((task_pid(current) != pid) &&
3921 (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3922 goto err;
3923 } else {
3924 prepare_kill_siginfo(sig, &kinfo);
3925 }
3926
3927 ret = kill_pid_info(sig, &kinfo, pid);
3928
3929 err:
3930 fdput(f);
3931 return ret;
3932 }
3933
3934 static int
do_send_specific(pid_t tgid,pid_t pid,int sig,struct kernel_siginfo * info)3935 do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3936 {
3937 struct task_struct *p;
3938 int error = -ESRCH;
3939
3940 rcu_read_lock();
3941 p = find_task_by_vpid(pid);
3942 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3943 error = check_kill_permission(sig, info, p);
3944 /*
3945 * The null signal is a permissions and process existence
3946 * probe. No signal is actually delivered.
3947 */
3948 if (!error && sig) {
3949 error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
3950 /*
3951 * If lock_task_sighand() failed we pretend the task
3952 * dies after receiving the signal. The window is tiny,
3953 * and the signal is private anyway.
3954 */
3955 if (unlikely(error == -ESRCH))
3956 error = 0;
3957 }
3958 }
3959 rcu_read_unlock();
3960
3961 return error;
3962 }
3963
do_tkill(pid_t tgid,pid_t pid,int sig)3964 static int do_tkill(pid_t tgid, pid_t pid, int sig)
3965 {
3966 struct kernel_siginfo info;
3967
3968 clear_siginfo(&info);
3969 info.si_signo = sig;
3970 info.si_errno = 0;
3971 info.si_code = SI_TKILL;
3972 info.si_pid = task_tgid_vnr(current);
3973 info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
3974
3975 return do_send_specific(tgid, pid, sig, &info);
3976 }
3977
3978 /**
3979 * sys_tgkill - send signal to one specific thread
3980 * @tgid: the thread group ID of the thread
3981 * @pid: the PID of the thread
3982 * @sig: signal to be sent
3983 *
3984 * This syscall also checks the @tgid and returns -ESRCH even if the PID
3985 * exists but it's not belonging to the target process anymore. This
3986 * method solves the problem of threads exiting and PIDs getting reused.
3987 */
SYSCALL_DEFINE3(tgkill,pid_t,tgid,pid_t,pid,int,sig)3988 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
3989 {
3990 /* This is only valid for single tasks */
3991 if (pid <= 0 || tgid <= 0)
3992 return -EINVAL;
3993
3994 return do_tkill(tgid, pid, sig);
3995 }
3996
3997 /**
3998 * sys_tkill - send signal to one specific task
3999 * @pid: the PID of the task
4000 * @sig: signal to be sent
4001 *
4002 * Send a signal to only one task, even if it's a CLONE_THREAD task.
4003 */
SYSCALL_DEFINE2(tkill,pid_t,pid,int,sig)4004 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
4005 {
4006 /* This is only valid for single tasks */
4007 if (pid <= 0)
4008 return -EINVAL;
4009
4010 return do_tkill(0, pid, sig);
4011 }
4012
do_rt_sigqueueinfo(pid_t pid,int sig,kernel_siginfo_t * info)4013 static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
4014 {
4015 /* Not even root can pretend to send signals from the kernel.
4016 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4017 */
4018 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4019 (task_pid_vnr(current) != pid))
4020 return -EPERM;
4021
4022 /* POSIX.1b doesn't mention process groups. */
4023 return kill_proc_info(sig, info, pid);
4024 }
4025
4026 /**
4027 * sys_rt_sigqueueinfo - send signal information to a signal
4028 * @pid: the PID of the thread
4029 * @sig: signal to be sent
4030 * @uinfo: signal info to be sent
4031 */
SYSCALL_DEFINE3(rt_sigqueueinfo,pid_t,pid,int,sig,siginfo_t __user *,uinfo)4032 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
4033 siginfo_t __user *, uinfo)
4034 {
4035 kernel_siginfo_t info;
4036 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4037 if (unlikely(ret))
4038 return ret;
4039 return do_rt_sigqueueinfo(pid, sig, &info);
4040 }
4041
4042 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,compat_pid_t,pid,int,sig,struct compat_siginfo __user *,uinfo)4043 COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
4044 compat_pid_t, pid,
4045 int, sig,
4046 struct compat_siginfo __user *, uinfo)
4047 {
4048 kernel_siginfo_t info;
4049 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4050 if (unlikely(ret))
4051 return ret;
4052 return do_rt_sigqueueinfo(pid, sig, &info);
4053 }
4054 #endif
4055
do_rt_tgsigqueueinfo(pid_t tgid,pid_t pid,int sig,kernel_siginfo_t * info)4056 static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
4057 {
4058 /* This is only valid for single tasks */
4059 if (pid <= 0 || tgid <= 0)
4060 return -EINVAL;
4061
4062 /* Not even root can pretend to send signals from the kernel.
4063 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4064 */
4065 if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4066 (task_pid_vnr(current) != pid))
4067 return -EPERM;
4068
4069 return do_send_specific(tgid, pid, sig, info);
4070 }
4071
SYSCALL_DEFINE4(rt_tgsigqueueinfo,pid_t,tgid,pid_t,pid,int,sig,siginfo_t __user *,uinfo)4072 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
4073 siginfo_t __user *, uinfo)
4074 {
4075 kernel_siginfo_t info;
4076 int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4077 if (unlikely(ret))
4078 return ret;
4079 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4080 }
4081
4082 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,compat_pid_t,tgid,compat_pid_t,pid,int,sig,struct compat_siginfo __user *,uinfo)4083 COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
4084 compat_pid_t, tgid,
4085 compat_pid_t, pid,
4086 int, sig,
4087 struct compat_siginfo __user *, uinfo)
4088 {
4089 kernel_siginfo_t info;
4090 int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4091 if (unlikely(ret))
4092 return ret;
4093 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4094 }
4095 #endif
4096
4097 /*
4098 * For kthreads only, must not be used if cloned with CLONE_SIGHAND
4099 */
kernel_sigaction(int sig,__sighandler_t action)4100 void kernel_sigaction(int sig, __sighandler_t action)
4101 {
4102 spin_lock_irq(¤t->sighand->siglock);
4103 current->sighand->action[sig - 1].sa.sa_handler = action;
4104 if (action == SIG_IGN) {
4105 sigset_t mask;
4106
4107 sigemptyset(&mask);
4108 sigaddset(&mask, sig);
4109
4110 flush_sigqueue_mask(&mask, ¤t->signal->shared_pending);
4111 flush_sigqueue_mask(&mask, ¤t->pending);
4112 recalc_sigpending();
4113 }
4114 spin_unlock_irq(¤t->sighand->siglock);
4115 }
4116 EXPORT_SYMBOL(kernel_sigaction);
4117
sigaction_compat_abi(struct k_sigaction * act,struct k_sigaction * oact)4118 void __weak sigaction_compat_abi(struct k_sigaction *act,
4119 struct k_sigaction *oact)
4120 {
4121 }
4122
do_sigaction(int sig,struct k_sigaction * act,struct k_sigaction * oact)4123 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4124 {
4125 struct task_struct *p = current, *t;
4126 struct k_sigaction *k;
4127 sigset_t mask;
4128
4129 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4130 return -EINVAL;
4131
4132 k = &p->sighand->action[sig-1];
4133
4134 spin_lock_irq(&p->sighand->siglock);
4135 if (k->sa.sa_flags & SA_IMMUTABLE) {
4136 spin_unlock_irq(&p->sighand->siglock);
4137 return -EINVAL;
4138 }
4139 if (oact)
4140 *oact = *k;
4141
4142 /*
4143 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4144 * e.g. by having an architecture use the bit in their uapi.
4145 */
4146 BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4147
4148 /*
4149 * Clear unknown flag bits in order to allow userspace to detect missing
4150 * support for flag bits and to allow the kernel to use non-uapi bits
4151 * internally.
4152 */
4153 if (act)
4154 act->sa.sa_flags &= UAPI_SA_FLAGS;
4155 if (oact)
4156 oact->sa.sa_flags &= UAPI_SA_FLAGS;
4157
4158 sigaction_compat_abi(act, oact);
4159
4160 if (act) {
4161 sigdelsetmask(&act->sa.sa_mask,
4162 sigmask(SIGKILL) | sigmask(SIGSTOP));
4163 *k = *act;
4164 /*
4165 * POSIX 3.3.1.3:
4166 * "Setting a signal action to SIG_IGN for a signal that is
4167 * pending shall cause the pending signal to be discarded,
4168 * whether or not it is blocked."
4169 *
4170 * "Setting a signal action to SIG_DFL for a signal that is
4171 * pending and whose default action is to ignore the signal
4172 * (for example, SIGCHLD), shall cause the pending signal to
4173 * be discarded, whether or not it is blocked"
4174 */
4175 if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4176 sigemptyset(&mask);
4177 sigaddset(&mask, sig);
4178 flush_sigqueue_mask(&mask, &p->signal->shared_pending);
4179 for_each_thread(p, t)
4180 flush_sigqueue_mask(&mask, &t->pending);
4181 }
4182 }
4183
4184 spin_unlock_irq(&p->sighand->siglock);
4185 return 0;
4186 }
4187
4188 #ifdef CONFIG_DYNAMIC_SIGFRAME
sigaltstack_lock(void)4189 static inline void sigaltstack_lock(void)
4190 __acquires(¤t->sighand->siglock)
4191 {
4192 spin_lock_irq(¤t->sighand->siglock);
4193 }
4194
sigaltstack_unlock(void)4195 static inline void sigaltstack_unlock(void)
4196 __releases(¤t->sighand->siglock)
4197 {
4198 spin_unlock_irq(¤t->sighand->siglock);
4199 }
4200 #else
sigaltstack_lock(void)4201 static inline void sigaltstack_lock(void) { }
sigaltstack_unlock(void)4202 static inline void sigaltstack_unlock(void) { }
4203 #endif
4204
4205 static int
do_sigaltstack(const stack_t * ss,stack_t * oss,unsigned long sp,size_t min_ss_size)4206 do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4207 size_t min_ss_size)
4208 {
4209 struct task_struct *t = current;
4210 int ret = 0;
4211
4212 if (oss) {
4213 memset(oss, 0, sizeof(stack_t));
4214 oss->ss_sp = (void __user *) t->sas_ss_sp;
4215 oss->ss_size = t->sas_ss_size;
4216 oss->ss_flags = sas_ss_flags(sp) |
4217 (current->sas_ss_flags & SS_FLAG_BITS);
4218 }
4219
4220 if (ss) {
4221 void __user *ss_sp = ss->ss_sp;
4222 size_t ss_size = ss->ss_size;
4223 unsigned ss_flags = ss->ss_flags;
4224 int ss_mode;
4225
4226 if (unlikely(on_sig_stack(sp)))
4227 return -EPERM;
4228
4229 ss_mode = ss_flags & ~SS_FLAG_BITS;
4230 if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4231 ss_mode != 0))
4232 return -EINVAL;
4233
4234 /*
4235 * Return before taking any locks if no actual
4236 * sigaltstack changes were requested.
4237 */
4238 if (t->sas_ss_sp == (unsigned long)ss_sp &&
4239 t->sas_ss_size == ss_size &&
4240 t->sas_ss_flags == ss_flags)
4241 return 0;
4242
4243 sigaltstack_lock();
4244 if (ss_mode == SS_DISABLE) {
4245 ss_size = 0;
4246 ss_sp = NULL;
4247 } else {
4248 if (unlikely(ss_size < min_ss_size))
4249 ret = -ENOMEM;
4250 if (!sigaltstack_size_valid(ss_size))
4251 ret = -ENOMEM;
4252 }
4253 if (!ret) {
4254 t->sas_ss_sp = (unsigned long) ss_sp;
4255 t->sas_ss_size = ss_size;
4256 t->sas_ss_flags = ss_flags;
4257 }
4258 sigaltstack_unlock();
4259 }
4260 return ret;
4261 }
4262
SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss,stack_t __user *,uoss)4263 SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4264 {
4265 stack_t new, old;
4266 int err;
4267 if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4268 return -EFAULT;
4269 err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4270 current_user_stack_pointer(),
4271 MINSIGSTKSZ);
4272 if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4273 err = -EFAULT;
4274 return err;
4275 }
4276
restore_altstack(const stack_t __user * uss)4277 int restore_altstack(const stack_t __user *uss)
4278 {
4279 stack_t new;
4280 if (copy_from_user(&new, uss, sizeof(stack_t)))
4281 return -EFAULT;
4282 (void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4283 MINSIGSTKSZ);
4284 /* squash all but EFAULT for now */
4285 return 0;
4286 }
4287
__save_altstack(stack_t __user * uss,unsigned long sp)4288 int __save_altstack(stack_t __user *uss, unsigned long sp)
4289 {
4290 struct task_struct *t = current;
4291 int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4292 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4293 __put_user(t->sas_ss_size, &uss->ss_size);
4294 return err;
4295 }
4296
4297 #ifdef CONFIG_COMPAT
do_compat_sigaltstack(const compat_stack_t __user * uss_ptr,compat_stack_t __user * uoss_ptr)4298 static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4299 compat_stack_t __user *uoss_ptr)
4300 {
4301 stack_t uss, uoss;
4302 int ret;
4303
4304 if (uss_ptr) {
4305 compat_stack_t uss32;
4306 if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4307 return -EFAULT;
4308 uss.ss_sp = compat_ptr(uss32.ss_sp);
4309 uss.ss_flags = uss32.ss_flags;
4310 uss.ss_size = uss32.ss_size;
4311 }
4312 ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4313 compat_user_stack_pointer(),
4314 COMPAT_MINSIGSTKSZ);
4315 if (ret >= 0 && uoss_ptr) {
4316 compat_stack_t old;
4317 memset(&old, 0, sizeof(old));
4318 old.ss_sp = ptr_to_compat(uoss.ss_sp);
4319 old.ss_flags = uoss.ss_flags;
4320 old.ss_size = uoss.ss_size;
4321 if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4322 ret = -EFAULT;
4323 }
4324 return ret;
4325 }
4326
COMPAT_SYSCALL_DEFINE2(sigaltstack,const compat_stack_t __user *,uss_ptr,compat_stack_t __user *,uoss_ptr)4327 COMPAT_SYSCALL_DEFINE2(sigaltstack,
4328 const compat_stack_t __user *, uss_ptr,
4329 compat_stack_t __user *, uoss_ptr)
4330 {
4331 return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4332 }
4333
compat_restore_altstack(const compat_stack_t __user * uss)4334 int compat_restore_altstack(const compat_stack_t __user *uss)
4335 {
4336 int err = do_compat_sigaltstack(uss, NULL);
4337 /* squash all but -EFAULT for now */
4338 return err == -EFAULT ? err : 0;
4339 }
4340
__compat_save_altstack(compat_stack_t __user * uss,unsigned long sp)4341 int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4342 {
4343 int err;
4344 struct task_struct *t = current;
4345 err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4346 &uss->ss_sp) |
4347 __put_user(t->sas_ss_flags, &uss->ss_flags) |
4348 __put_user(t->sas_ss_size, &uss->ss_size);
4349 return err;
4350 }
4351 #endif
4352
4353 #ifdef __ARCH_WANT_SYS_SIGPENDING
4354
4355 /**
4356 * sys_sigpending - examine pending signals
4357 * @uset: where mask of pending signal is returned
4358 */
SYSCALL_DEFINE1(sigpending,old_sigset_t __user *,uset)4359 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4360 {
4361 sigset_t set;
4362
4363 if (sizeof(old_sigset_t) > sizeof(*uset))
4364 return -EINVAL;
4365
4366 do_sigpending(&set);
4367
4368 if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4369 return -EFAULT;
4370
4371 return 0;
4372 }
4373
4374 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE1(sigpending,compat_old_sigset_t __user *,set32)4375 COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4376 {
4377 sigset_t set;
4378
4379 do_sigpending(&set);
4380
4381 return put_user(set.sig[0], set32);
4382 }
4383 #endif
4384
4385 #endif
4386
4387 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
4388 /**
4389 * sys_sigprocmask - examine and change blocked signals
4390 * @how: whether to add, remove, or set signals
4391 * @nset: signals to add or remove (if non-null)
4392 * @oset: previous value of signal mask if non-null
4393 *
4394 * Some platforms have their own version with special arguments;
4395 * others support only sys_rt_sigprocmask.
4396 */
4397
SYSCALL_DEFINE3(sigprocmask,int,how,old_sigset_t __user *,nset,old_sigset_t __user *,oset)4398 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4399 old_sigset_t __user *, oset)
4400 {
4401 old_sigset_t old_set, new_set;
4402 sigset_t new_blocked;
4403
4404 old_set = current->blocked.sig[0];
4405
4406 if (nset) {
4407 if (copy_from_user(&new_set, nset, sizeof(*nset)))
4408 return -EFAULT;
4409
4410 new_blocked = current->blocked;
4411
4412 switch (how) {
4413 case SIG_BLOCK:
4414 sigaddsetmask(&new_blocked, new_set);
4415 break;
4416 case SIG_UNBLOCK:
4417 sigdelsetmask(&new_blocked, new_set);
4418 break;
4419 case SIG_SETMASK:
4420 new_blocked.sig[0] = new_set;
4421 break;
4422 default:
4423 return -EINVAL;
4424 }
4425
4426 set_current_blocked(&new_blocked);
4427 }
4428
4429 if (oset) {
4430 if (copy_to_user(oset, &old_set, sizeof(*oset)))
4431 return -EFAULT;
4432 }
4433
4434 return 0;
4435 }
4436 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4437
4438 #ifndef CONFIG_ODD_RT_SIGACTION
4439 /**
4440 * sys_rt_sigaction - alter an action taken by a process
4441 * @sig: signal to be sent
4442 * @act: new sigaction
4443 * @oact: used to save the previous sigaction
4444 * @sigsetsize: size of sigset_t type
4445 */
SYSCALL_DEFINE4(rt_sigaction,int,sig,const struct sigaction __user *,act,struct sigaction __user *,oact,size_t,sigsetsize)4446 SYSCALL_DEFINE4(rt_sigaction, int, sig,
4447 const struct sigaction __user *, act,
4448 struct sigaction __user *, oact,
4449 size_t, sigsetsize)
4450 {
4451 struct k_sigaction new_sa, old_sa;
4452 int ret;
4453
4454 /* XXX: Don't preclude handling different sized sigset_t's. */
4455 if (sigsetsize != sizeof(sigset_t))
4456 return -EINVAL;
4457
4458 if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4459 return -EFAULT;
4460
4461 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4462 if (ret)
4463 return ret;
4464
4465 if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4466 return -EFAULT;
4467
4468 return 0;
4469 }
4470 #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)4471 COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4472 const struct compat_sigaction __user *, act,
4473 struct compat_sigaction __user *, oact,
4474 compat_size_t, sigsetsize)
4475 {
4476 struct k_sigaction new_ka, old_ka;
4477 #ifdef __ARCH_HAS_SA_RESTORER
4478 compat_uptr_t restorer;
4479 #endif
4480 int ret;
4481
4482 /* XXX: Don't preclude handling different sized sigset_t's. */
4483 if (sigsetsize != sizeof(compat_sigset_t))
4484 return -EINVAL;
4485
4486 if (act) {
4487 compat_uptr_t handler;
4488 ret = get_user(handler, &act->sa_handler);
4489 new_ka.sa.sa_handler = compat_ptr(handler);
4490 #ifdef __ARCH_HAS_SA_RESTORER
4491 ret |= get_user(restorer, &act->sa_restorer);
4492 new_ka.sa.sa_restorer = compat_ptr(restorer);
4493 #endif
4494 ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4495 ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4496 if (ret)
4497 return -EFAULT;
4498 }
4499
4500 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4501 if (!ret && oact) {
4502 ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4503 &oact->sa_handler);
4504 ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4505 sizeof(oact->sa_mask));
4506 ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4507 #ifdef __ARCH_HAS_SA_RESTORER
4508 ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4509 &oact->sa_restorer);
4510 #endif
4511 }
4512 return ret;
4513 }
4514 #endif
4515 #endif /* !CONFIG_ODD_RT_SIGACTION */
4516
4517 #ifdef CONFIG_OLD_SIGACTION
SYSCALL_DEFINE3(sigaction,int,sig,const struct old_sigaction __user *,act,struct old_sigaction __user *,oact)4518 SYSCALL_DEFINE3(sigaction, int, sig,
4519 const struct old_sigaction __user *, act,
4520 struct old_sigaction __user *, oact)
4521 {
4522 struct k_sigaction new_ka, old_ka;
4523 int ret;
4524
4525 if (act) {
4526 old_sigset_t mask;
4527 if (!access_ok(act, sizeof(*act)) ||
4528 __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4529 __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4530 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4531 __get_user(mask, &act->sa_mask))
4532 return -EFAULT;
4533 #ifdef __ARCH_HAS_KA_RESTORER
4534 new_ka.ka_restorer = NULL;
4535 #endif
4536 siginitset(&new_ka.sa.sa_mask, mask);
4537 }
4538
4539 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4540
4541 if (!ret && oact) {
4542 if (!access_ok(oact, sizeof(*oact)) ||
4543 __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4544 __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4545 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4546 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4547 return -EFAULT;
4548 }
4549
4550 return ret;
4551 }
4552 #endif
4553 #ifdef CONFIG_COMPAT_OLD_SIGACTION
COMPAT_SYSCALL_DEFINE3(sigaction,int,sig,const struct compat_old_sigaction __user *,act,struct compat_old_sigaction __user *,oact)4554 COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4555 const struct compat_old_sigaction __user *, act,
4556 struct compat_old_sigaction __user *, oact)
4557 {
4558 struct k_sigaction new_ka, old_ka;
4559 int ret;
4560 compat_old_sigset_t mask;
4561 compat_uptr_t handler, restorer;
4562
4563 if (act) {
4564 if (!access_ok(act, sizeof(*act)) ||
4565 __get_user(handler, &act->sa_handler) ||
4566 __get_user(restorer, &act->sa_restorer) ||
4567 __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4568 __get_user(mask, &act->sa_mask))
4569 return -EFAULT;
4570
4571 #ifdef __ARCH_HAS_KA_RESTORER
4572 new_ka.ka_restorer = NULL;
4573 #endif
4574 new_ka.sa.sa_handler = compat_ptr(handler);
4575 new_ka.sa.sa_restorer = compat_ptr(restorer);
4576 siginitset(&new_ka.sa.sa_mask, mask);
4577 }
4578
4579 ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4580
4581 if (!ret && oact) {
4582 if (!access_ok(oact, sizeof(*oact)) ||
4583 __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4584 &oact->sa_handler) ||
4585 __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4586 &oact->sa_restorer) ||
4587 __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4588 __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4589 return -EFAULT;
4590 }
4591 return ret;
4592 }
4593 #endif
4594
4595 #ifdef CONFIG_SGETMASK_SYSCALL
4596
4597 /*
4598 * For backwards compatibility. Functionality superseded by sigprocmask.
4599 */
SYSCALL_DEFINE0(sgetmask)4600 SYSCALL_DEFINE0(sgetmask)
4601 {
4602 /* SMP safe */
4603 return current->blocked.sig[0];
4604 }
4605
SYSCALL_DEFINE1(ssetmask,int,newmask)4606 SYSCALL_DEFINE1(ssetmask, int, newmask)
4607 {
4608 int old = current->blocked.sig[0];
4609 sigset_t newset;
4610
4611 siginitset(&newset, newmask);
4612 set_current_blocked(&newset);
4613
4614 return old;
4615 }
4616 #endif /* CONFIG_SGETMASK_SYSCALL */
4617
4618 #ifdef __ARCH_WANT_SYS_SIGNAL
4619 /*
4620 * For backwards compatibility. Functionality superseded by sigaction.
4621 */
SYSCALL_DEFINE2(signal,int,sig,__sighandler_t,handler)4622 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4623 {
4624 struct k_sigaction new_sa, old_sa;
4625 int ret;
4626
4627 new_sa.sa.sa_handler = handler;
4628 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4629 sigemptyset(&new_sa.sa.sa_mask);
4630
4631 ret = do_sigaction(sig, &new_sa, &old_sa);
4632
4633 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4634 }
4635 #endif /* __ARCH_WANT_SYS_SIGNAL */
4636
4637 #ifdef __ARCH_WANT_SYS_PAUSE
4638
SYSCALL_DEFINE0(pause)4639 SYSCALL_DEFINE0(pause)
4640 {
4641 while (!signal_pending(current)) {
4642 __set_current_state(TASK_INTERRUPTIBLE);
4643 schedule();
4644 }
4645 return -ERESTARTNOHAND;
4646 }
4647
4648 #endif
4649
sigsuspend(sigset_t * set)4650 static int sigsuspend(sigset_t *set)
4651 {
4652 current->saved_sigmask = current->blocked;
4653 set_current_blocked(set);
4654
4655 while (!signal_pending(current)) {
4656 __set_current_state(TASK_INTERRUPTIBLE);
4657 schedule();
4658 }
4659 set_restore_sigmask();
4660 return -ERESTARTNOHAND;
4661 }
4662
4663 /**
4664 * sys_rt_sigsuspend - replace the signal mask for a value with the
4665 * @unewset value until a signal is received
4666 * @unewset: new signal mask value
4667 * @sigsetsize: size of sigset_t type
4668 */
SYSCALL_DEFINE2(rt_sigsuspend,sigset_t __user *,unewset,size_t,sigsetsize)4669 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4670 {
4671 sigset_t newset;
4672
4673 /* XXX: Don't preclude handling different sized sigset_t's. */
4674 if (sigsetsize != sizeof(sigset_t))
4675 return -EINVAL;
4676
4677 if (copy_from_user(&newset, unewset, sizeof(newset)))
4678 return -EFAULT;
4679 return sigsuspend(&newset);
4680 }
4681
4682 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(rt_sigsuspend,compat_sigset_t __user *,unewset,compat_size_t,sigsetsize)4683 COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4684 {
4685 sigset_t newset;
4686
4687 /* XXX: Don't preclude handling different sized sigset_t's. */
4688 if (sigsetsize != sizeof(sigset_t))
4689 return -EINVAL;
4690
4691 if (get_compat_sigset(&newset, unewset))
4692 return -EFAULT;
4693 return sigsuspend(&newset);
4694 }
4695 #endif
4696
4697 #ifdef CONFIG_OLD_SIGSUSPEND
SYSCALL_DEFINE1(sigsuspend,old_sigset_t,mask)4698 SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4699 {
4700 sigset_t blocked;
4701 siginitset(&blocked, mask);
4702 return sigsuspend(&blocked);
4703 }
4704 #endif
4705 #ifdef CONFIG_OLD_SIGSUSPEND3
SYSCALL_DEFINE3(sigsuspend,int,unused1,int,unused2,old_sigset_t,mask)4706 SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4707 {
4708 sigset_t blocked;
4709 siginitset(&blocked, mask);
4710 return sigsuspend(&blocked);
4711 }
4712 #endif
4713
arch_vma_name(struct vm_area_struct * vma)4714 __weak const char *arch_vma_name(struct vm_area_struct *vma)
4715 {
4716 return NULL;
4717 }
4718
siginfo_buildtime_checks(void)4719 static inline void siginfo_buildtime_checks(void)
4720 {
4721 BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4722
4723 /* Verify the offsets in the two siginfos match */
4724 #define CHECK_OFFSET(field) \
4725 BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4726
4727 /* kill */
4728 CHECK_OFFSET(si_pid);
4729 CHECK_OFFSET(si_uid);
4730
4731 /* timer */
4732 CHECK_OFFSET(si_tid);
4733 CHECK_OFFSET(si_overrun);
4734 CHECK_OFFSET(si_value);
4735
4736 /* rt */
4737 CHECK_OFFSET(si_pid);
4738 CHECK_OFFSET(si_uid);
4739 CHECK_OFFSET(si_value);
4740
4741 /* sigchld */
4742 CHECK_OFFSET(si_pid);
4743 CHECK_OFFSET(si_uid);
4744 CHECK_OFFSET(si_status);
4745 CHECK_OFFSET(si_utime);
4746 CHECK_OFFSET(si_stime);
4747
4748 /* sigfault */
4749 CHECK_OFFSET(si_addr);
4750 CHECK_OFFSET(si_trapno);
4751 CHECK_OFFSET(si_addr_lsb);
4752 CHECK_OFFSET(si_lower);
4753 CHECK_OFFSET(si_upper);
4754 CHECK_OFFSET(si_pkey);
4755 CHECK_OFFSET(si_perf_data);
4756 CHECK_OFFSET(si_perf_type);
4757 CHECK_OFFSET(si_perf_flags);
4758
4759 /* sigpoll */
4760 CHECK_OFFSET(si_band);
4761 CHECK_OFFSET(si_fd);
4762
4763 /* sigsys */
4764 CHECK_OFFSET(si_call_addr);
4765 CHECK_OFFSET(si_syscall);
4766 CHECK_OFFSET(si_arch);
4767 #undef CHECK_OFFSET
4768
4769 /* usb asyncio */
4770 BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4771 offsetof(struct siginfo, si_addr));
4772 if (sizeof(int) == sizeof(void __user *)) {
4773 BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4774 sizeof(void __user *));
4775 } else {
4776 BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4777 sizeof_field(struct siginfo, si_uid)) !=
4778 sizeof(void __user *));
4779 BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4780 offsetof(struct siginfo, si_uid));
4781 }
4782 #ifdef CONFIG_COMPAT
4783 BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4784 offsetof(struct compat_siginfo, si_addr));
4785 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4786 sizeof(compat_uptr_t));
4787 BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4788 sizeof_field(struct siginfo, si_pid));
4789 #endif
4790 }
4791
4792 #if defined(CONFIG_SYSCTL)
4793 static struct ctl_table signal_debug_table[] = {
4794 #ifdef CONFIG_SYSCTL_EXCEPTION_TRACE
4795 {
4796 .procname = "exception-trace",
4797 .data = &show_unhandled_signals,
4798 .maxlen = sizeof(int),
4799 .mode = 0644,
4800 .proc_handler = proc_dointvec
4801 },
4802 #endif
4803 { }
4804 };
4805
init_signal_sysctls(void)4806 static int __init init_signal_sysctls(void)
4807 {
4808 register_sysctl_init("debug", signal_debug_table);
4809 return 0;
4810 }
4811 early_initcall(init_signal_sysctls);
4812 #endif /* CONFIG_SYSCTL */
4813
signals_init(void)4814 void __init signals_init(void)
4815 {
4816 siginfo_buildtime_checks();
4817
4818 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC | SLAB_ACCOUNT);
4819 }
4820
4821 #ifdef CONFIG_KGDB_KDB
4822 #include <linux/kdb.h>
4823 /*
4824 * kdb_send_sig - Allows kdb to send signals without exposing
4825 * signal internals. This function checks if the required locks are
4826 * available before calling the main signal code, to avoid kdb
4827 * deadlocks.
4828 */
kdb_send_sig(struct task_struct * t,int sig)4829 void kdb_send_sig(struct task_struct *t, int sig)
4830 {
4831 static struct task_struct *kdb_prev_t;
4832 int new_t, ret;
4833 if (!spin_trylock(&t->sighand->siglock)) {
4834 kdb_printf("Can't do kill command now.\n"
4835 "The sigmask lock is held somewhere else in "
4836 "kernel, try again later\n");
4837 return;
4838 }
4839 new_t = kdb_prev_t != t;
4840 kdb_prev_t = t;
4841 if (!task_is_running(t) && new_t) {
4842 spin_unlock(&t->sighand->siglock);
4843 kdb_printf("Process is not RUNNING, sending a signal from "
4844 "kdb risks deadlock\n"
4845 "on the run queue locks. "
4846 "The signal has _not_ been sent.\n"
4847 "Reissue the kill command if you want to risk "
4848 "the deadlock.\n");
4849 return;
4850 }
4851 ret = send_signal_locked(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4852 spin_unlock(&t->sighand->siglock);
4853 if (ret)
4854 kdb_printf("Fail to deliver Signal %d to process %d.\n",
4855 sig, t->pid);
4856 else
4857 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4858 }
4859 #endif /* CONFIG_KGDB_KDB */
4860