xref: /linux/kernel/ptrace.c (revision e0bf6c5ca2d3281f231c5f0c9bf145e9513644de)
1 /*
2  * linux/kernel/ptrace.c
3  *
4  * (C) Copyright 1999 Linus Torvalds
5  *
6  * Common interfaces for "ptrace()" which we do not want
7  * to continually duplicate across every architecture.
8  */
9 
10 #include <linux/capability.h>
11 #include <linux/export.h>
12 #include <linux/sched.h>
13 #include <linux/errno.h>
14 #include <linux/mm.h>
15 #include <linux/highmem.h>
16 #include <linux/pagemap.h>
17 #include <linux/ptrace.h>
18 #include <linux/security.h>
19 #include <linux/signal.h>
20 #include <linux/uio.h>
21 #include <linux/audit.h>
22 #include <linux/pid_namespace.h>
23 #include <linux/syscalls.h>
24 #include <linux/uaccess.h>
25 #include <linux/regset.h>
26 #include <linux/hw_breakpoint.h>
27 #include <linux/cn_proc.h>
28 #include <linux/compat.h>
29 
30 
31 /*
32  * ptrace a task: make the debugger its new parent and
33  * move it to the ptrace list.
34  *
35  * Must be called with the tasklist lock write-held.
36  */
37 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent)
38 {
39 	BUG_ON(!list_empty(&child->ptrace_entry));
40 	list_add(&child->ptrace_entry, &new_parent->ptraced);
41 	child->parent = new_parent;
42 }
43 
44 /**
45  * __ptrace_unlink - unlink ptracee and restore its execution state
46  * @child: ptracee to be unlinked
47  *
48  * Remove @child from the ptrace list, move it back to the original parent,
49  * and restore the execution state so that it conforms to the group stop
50  * state.
51  *
52  * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
53  * exiting.  For PTRACE_DETACH, unless the ptracee has been killed between
54  * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
55  * If the ptracer is exiting, the ptracee can be in any state.
56  *
57  * After detach, the ptracee should be in a state which conforms to the
58  * group stop.  If the group is stopped or in the process of stopping, the
59  * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
60  * up from TASK_TRACED.
61  *
62  * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
63  * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
64  * to but in the opposite direction of what happens while attaching to a
65  * stopped task.  However, in this direction, the intermediate RUNNING
66  * state is not hidden even from the current ptracer and if it immediately
67  * re-attaches and performs a WNOHANG wait(2), it may fail.
68  *
69  * CONTEXT:
70  * write_lock_irq(tasklist_lock)
71  */
72 void __ptrace_unlink(struct task_struct *child)
73 {
74 	BUG_ON(!child->ptrace);
75 
76 	child->ptrace = 0;
77 	child->parent = child->real_parent;
78 	list_del_init(&child->ptrace_entry);
79 
80 	spin_lock(&child->sighand->siglock);
81 
82 	/*
83 	 * Clear all pending traps and TRAPPING.  TRAPPING should be
84 	 * cleared regardless of JOBCTL_STOP_PENDING.  Do it explicitly.
85 	 */
86 	task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
87 	task_clear_jobctl_trapping(child);
88 
89 	/*
90 	 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
91 	 * @child isn't dead.
92 	 */
93 	if (!(child->flags & PF_EXITING) &&
94 	    (child->signal->flags & SIGNAL_STOP_STOPPED ||
95 	     child->signal->group_stop_count)) {
96 		child->jobctl |= JOBCTL_STOP_PENDING;
97 
98 		/*
99 		 * This is only possible if this thread was cloned by the
100 		 * traced task running in the stopped group, set the signal
101 		 * for the future reports.
102 		 * FIXME: we should change ptrace_init_task() to handle this
103 		 * case.
104 		 */
105 		if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
106 			child->jobctl |= SIGSTOP;
107 	}
108 
109 	/*
110 	 * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
111 	 * @child in the butt.  Note that @resume should be used iff @child
112 	 * is in TASK_TRACED; otherwise, we might unduly disrupt
113 	 * TASK_KILLABLE sleeps.
114 	 */
115 	if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
116 		ptrace_signal_wake_up(child, true);
117 
118 	spin_unlock(&child->sighand->siglock);
119 }
120 
121 /* Ensure that nothing can wake it up, even SIGKILL */
122 static bool ptrace_freeze_traced(struct task_struct *task)
123 {
124 	bool ret = false;
125 
126 	/* Lockless, nobody but us can set this flag */
127 	if (task->jobctl & JOBCTL_LISTENING)
128 		return ret;
129 
130 	spin_lock_irq(&task->sighand->siglock);
131 	if (task_is_traced(task) && !__fatal_signal_pending(task)) {
132 		task->state = __TASK_TRACED;
133 		ret = true;
134 	}
135 	spin_unlock_irq(&task->sighand->siglock);
136 
137 	return ret;
138 }
139 
140 static void ptrace_unfreeze_traced(struct task_struct *task)
141 {
142 	if (task->state != __TASK_TRACED)
143 		return;
144 
145 	WARN_ON(!task->ptrace || task->parent != current);
146 
147 	spin_lock_irq(&task->sighand->siglock);
148 	if (__fatal_signal_pending(task))
149 		wake_up_state(task, __TASK_TRACED);
150 	else
151 		task->state = TASK_TRACED;
152 	spin_unlock_irq(&task->sighand->siglock);
153 }
154 
155 /**
156  * ptrace_check_attach - check whether ptracee is ready for ptrace operation
157  * @child: ptracee to check for
158  * @ignore_state: don't check whether @child is currently %TASK_TRACED
159  *
160  * Check whether @child is being ptraced by %current and ready for further
161  * ptrace operations.  If @ignore_state is %false, @child also should be in
162  * %TASK_TRACED state and on return the child is guaranteed to be traced
163  * and not executing.  If @ignore_state is %true, @child can be in any
164  * state.
165  *
166  * CONTEXT:
167  * Grabs and releases tasklist_lock and @child->sighand->siglock.
168  *
169  * RETURNS:
170  * 0 on success, -ESRCH if %child is not ready.
171  */
172 static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
173 {
174 	int ret = -ESRCH;
175 
176 	/*
177 	 * We take the read lock around doing both checks to close a
178 	 * possible race where someone else was tracing our child and
179 	 * detached between these two checks.  After this locked check,
180 	 * we are sure that this is our traced child and that can only
181 	 * be changed by us so it's not changing right after this.
182 	 */
183 	read_lock(&tasklist_lock);
184 	if (child->ptrace && child->parent == current) {
185 		WARN_ON(child->state == __TASK_TRACED);
186 		/*
187 		 * child->sighand can't be NULL, release_task()
188 		 * does ptrace_unlink() before __exit_signal().
189 		 */
190 		if (ignore_state || ptrace_freeze_traced(child))
191 			ret = 0;
192 	}
193 	read_unlock(&tasklist_lock);
194 
195 	if (!ret && !ignore_state) {
196 		if (!wait_task_inactive(child, __TASK_TRACED)) {
197 			/*
198 			 * This can only happen if may_ptrace_stop() fails and
199 			 * ptrace_stop() changes ->state back to TASK_RUNNING,
200 			 * so we should not worry about leaking __TASK_TRACED.
201 			 */
202 			WARN_ON(child->state == __TASK_TRACED);
203 			ret = -ESRCH;
204 		}
205 	}
206 
207 	return ret;
208 }
209 
210 static int ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
211 {
212 	if (mode & PTRACE_MODE_NOAUDIT)
213 		return has_ns_capability_noaudit(current, ns, CAP_SYS_PTRACE);
214 	else
215 		return has_ns_capability(current, ns, CAP_SYS_PTRACE);
216 }
217 
218 /* Returns 0 on success, -errno on denial. */
219 static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
220 {
221 	const struct cred *cred = current_cred(), *tcred;
222 
223 	/* May we inspect the given task?
224 	 * This check is used both for attaching with ptrace
225 	 * and for allowing access to sensitive information in /proc.
226 	 *
227 	 * ptrace_attach denies several cases that /proc allows
228 	 * because setting up the necessary parent/child relationship
229 	 * or halting the specified task is impossible.
230 	 */
231 	int dumpable = 0;
232 	/* Don't let security modules deny introspection */
233 	if (same_thread_group(task, current))
234 		return 0;
235 	rcu_read_lock();
236 	tcred = __task_cred(task);
237 	if (uid_eq(cred->uid, tcred->euid) &&
238 	    uid_eq(cred->uid, tcred->suid) &&
239 	    uid_eq(cred->uid, tcred->uid)  &&
240 	    gid_eq(cred->gid, tcred->egid) &&
241 	    gid_eq(cred->gid, tcred->sgid) &&
242 	    gid_eq(cred->gid, tcred->gid))
243 		goto ok;
244 	if (ptrace_has_cap(tcred->user_ns, mode))
245 		goto ok;
246 	rcu_read_unlock();
247 	return -EPERM;
248 ok:
249 	rcu_read_unlock();
250 	smp_rmb();
251 	if (task->mm)
252 		dumpable = get_dumpable(task->mm);
253 	rcu_read_lock();
254 	if (dumpable != SUID_DUMP_USER &&
255 	    !ptrace_has_cap(__task_cred(task)->user_ns, mode)) {
256 		rcu_read_unlock();
257 		return -EPERM;
258 	}
259 	rcu_read_unlock();
260 
261 	return security_ptrace_access_check(task, mode);
262 }
263 
264 bool ptrace_may_access(struct task_struct *task, unsigned int mode)
265 {
266 	int err;
267 	task_lock(task);
268 	err = __ptrace_may_access(task, mode);
269 	task_unlock(task);
270 	return !err;
271 }
272 
273 static int ptrace_attach(struct task_struct *task, long request,
274 			 unsigned long addr,
275 			 unsigned long flags)
276 {
277 	bool seize = (request == PTRACE_SEIZE);
278 	int retval;
279 
280 	retval = -EIO;
281 	if (seize) {
282 		if (addr != 0)
283 			goto out;
284 		if (flags & ~(unsigned long)PTRACE_O_MASK)
285 			goto out;
286 		flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
287 	} else {
288 		flags = PT_PTRACED;
289 	}
290 
291 	audit_ptrace(task);
292 
293 	retval = -EPERM;
294 	if (unlikely(task->flags & PF_KTHREAD))
295 		goto out;
296 	if (same_thread_group(task, current))
297 		goto out;
298 
299 	/*
300 	 * Protect exec's credential calculations against our interference;
301 	 * SUID, SGID and LSM creds get determined differently
302 	 * under ptrace.
303 	 */
304 	retval = -ERESTARTNOINTR;
305 	if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
306 		goto out;
307 
308 	task_lock(task);
309 	retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH);
310 	task_unlock(task);
311 	if (retval)
312 		goto unlock_creds;
313 
314 	write_lock_irq(&tasklist_lock);
315 	retval = -EPERM;
316 	if (unlikely(task->exit_state))
317 		goto unlock_tasklist;
318 	if (task->ptrace)
319 		goto unlock_tasklist;
320 
321 	if (seize)
322 		flags |= PT_SEIZED;
323 	rcu_read_lock();
324 	if (ns_capable(__task_cred(task)->user_ns, CAP_SYS_PTRACE))
325 		flags |= PT_PTRACE_CAP;
326 	rcu_read_unlock();
327 	task->ptrace = flags;
328 
329 	__ptrace_link(task, current);
330 
331 	/* SEIZE doesn't trap tracee on attach */
332 	if (!seize)
333 		send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
334 
335 	spin_lock(&task->sighand->siglock);
336 
337 	/*
338 	 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
339 	 * TRAPPING, and kick it so that it transits to TRACED.  TRAPPING
340 	 * will be cleared if the child completes the transition or any
341 	 * event which clears the group stop states happens.  We'll wait
342 	 * for the transition to complete before returning from this
343 	 * function.
344 	 *
345 	 * This hides STOPPED -> RUNNING -> TRACED transition from the
346 	 * attaching thread but a different thread in the same group can
347 	 * still observe the transient RUNNING state.  IOW, if another
348 	 * thread's WNOHANG wait(2) on the stopped tracee races against
349 	 * ATTACH, the wait(2) may fail due to the transient RUNNING.
350 	 *
351 	 * The following task_is_stopped() test is safe as both transitions
352 	 * in and out of STOPPED are protected by siglock.
353 	 */
354 	if (task_is_stopped(task) &&
355 	    task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
356 		signal_wake_up_state(task, __TASK_STOPPED);
357 
358 	spin_unlock(&task->sighand->siglock);
359 
360 	retval = 0;
361 unlock_tasklist:
362 	write_unlock_irq(&tasklist_lock);
363 unlock_creds:
364 	mutex_unlock(&task->signal->cred_guard_mutex);
365 out:
366 	if (!retval) {
367 		wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT,
368 			    TASK_UNINTERRUPTIBLE);
369 		proc_ptrace_connector(task, PTRACE_ATTACH);
370 	}
371 
372 	return retval;
373 }
374 
375 /**
376  * ptrace_traceme  --  helper for PTRACE_TRACEME
377  *
378  * Performs checks and sets PT_PTRACED.
379  * Should be used by all ptrace implementations for PTRACE_TRACEME.
380  */
381 static int ptrace_traceme(void)
382 {
383 	int ret = -EPERM;
384 
385 	write_lock_irq(&tasklist_lock);
386 	/* Are we already being traced? */
387 	if (!current->ptrace) {
388 		ret = security_ptrace_traceme(current->parent);
389 		/*
390 		 * Check PF_EXITING to ensure ->real_parent has not passed
391 		 * exit_ptrace(). Otherwise we don't report the error but
392 		 * pretend ->real_parent untraces us right after return.
393 		 */
394 		if (!ret && !(current->real_parent->flags & PF_EXITING)) {
395 			current->ptrace = PT_PTRACED;
396 			__ptrace_link(current, current->real_parent);
397 		}
398 	}
399 	write_unlock_irq(&tasklist_lock);
400 
401 	return ret;
402 }
403 
404 /*
405  * Called with irqs disabled, returns true if childs should reap themselves.
406  */
407 static int ignoring_children(struct sighand_struct *sigh)
408 {
409 	int ret;
410 	spin_lock(&sigh->siglock);
411 	ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
412 	      (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
413 	spin_unlock(&sigh->siglock);
414 	return ret;
415 }
416 
417 /*
418  * Called with tasklist_lock held for writing.
419  * Unlink a traced task, and clean it up if it was a traced zombie.
420  * Return true if it needs to be reaped with release_task().
421  * (We can't call release_task() here because we already hold tasklist_lock.)
422  *
423  * If it's a zombie, our attachedness prevented normal parent notification
424  * or self-reaping.  Do notification now if it would have happened earlier.
425  * If it should reap itself, return true.
426  *
427  * If it's our own child, there is no notification to do. But if our normal
428  * children self-reap, then this child was prevented by ptrace and we must
429  * reap it now, in that case we must also wake up sub-threads sleeping in
430  * do_wait().
431  */
432 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
433 {
434 	bool dead;
435 
436 	__ptrace_unlink(p);
437 
438 	if (p->exit_state != EXIT_ZOMBIE)
439 		return false;
440 
441 	dead = !thread_group_leader(p);
442 
443 	if (!dead && thread_group_empty(p)) {
444 		if (!same_thread_group(p->real_parent, tracer))
445 			dead = do_notify_parent(p, p->exit_signal);
446 		else if (ignoring_children(tracer->sighand)) {
447 			__wake_up_parent(p, tracer);
448 			dead = true;
449 		}
450 	}
451 	/* Mark it as in the process of being reaped. */
452 	if (dead)
453 		p->exit_state = EXIT_DEAD;
454 	return dead;
455 }
456 
457 static int ptrace_detach(struct task_struct *child, unsigned int data)
458 {
459 	bool dead = false;
460 
461 	if (!valid_signal(data))
462 		return -EIO;
463 
464 	/* Architecture-specific hardware disable .. */
465 	ptrace_disable(child);
466 	clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
467 
468 	write_lock_irq(&tasklist_lock);
469 	/*
470 	 * This child can be already killed. Make sure de_thread() or
471 	 * our sub-thread doing do_wait() didn't do release_task() yet.
472 	 */
473 	if (child->ptrace) {
474 		child->exit_code = data;
475 		dead = __ptrace_detach(current, child);
476 	}
477 	write_unlock_irq(&tasklist_lock);
478 
479 	proc_ptrace_connector(child, PTRACE_DETACH);
480 	if (unlikely(dead))
481 		release_task(child);
482 
483 	return 0;
484 }
485 
486 /*
487  * Detach all tasks we were using ptrace on. Called with tasklist held
488  * for writing.
489  */
490 void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
491 {
492 	struct task_struct *p, *n;
493 
494 	list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
495 		if (unlikely(p->ptrace & PT_EXITKILL))
496 			send_sig_info(SIGKILL, SEND_SIG_FORCED, p);
497 
498 		if (__ptrace_detach(tracer, p))
499 			list_add(&p->ptrace_entry, dead);
500 	}
501 }
502 
503 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
504 {
505 	int copied = 0;
506 
507 	while (len > 0) {
508 		char buf[128];
509 		int this_len, retval;
510 
511 		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
512 		retval = access_process_vm(tsk, src, buf, this_len, 0);
513 		if (!retval) {
514 			if (copied)
515 				break;
516 			return -EIO;
517 		}
518 		if (copy_to_user(dst, buf, retval))
519 			return -EFAULT;
520 		copied += retval;
521 		src += retval;
522 		dst += retval;
523 		len -= retval;
524 	}
525 	return copied;
526 }
527 
528 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
529 {
530 	int copied = 0;
531 
532 	while (len > 0) {
533 		char buf[128];
534 		int this_len, retval;
535 
536 		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
537 		if (copy_from_user(buf, src, this_len))
538 			return -EFAULT;
539 		retval = access_process_vm(tsk, dst, buf, this_len, 1);
540 		if (!retval) {
541 			if (copied)
542 				break;
543 			return -EIO;
544 		}
545 		copied += retval;
546 		src += retval;
547 		dst += retval;
548 		len -= retval;
549 	}
550 	return copied;
551 }
552 
553 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
554 {
555 	unsigned flags;
556 
557 	if (data & ~(unsigned long)PTRACE_O_MASK)
558 		return -EINVAL;
559 
560 	/* Avoid intermediate state when all opts are cleared */
561 	flags = child->ptrace;
562 	flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
563 	flags |= (data << PT_OPT_FLAG_SHIFT);
564 	child->ptrace = flags;
565 
566 	return 0;
567 }
568 
569 static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
570 {
571 	unsigned long flags;
572 	int error = -ESRCH;
573 
574 	if (lock_task_sighand(child, &flags)) {
575 		error = -EINVAL;
576 		if (likely(child->last_siginfo != NULL)) {
577 			*info = *child->last_siginfo;
578 			error = 0;
579 		}
580 		unlock_task_sighand(child, &flags);
581 	}
582 	return error;
583 }
584 
585 static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
586 {
587 	unsigned long flags;
588 	int error = -ESRCH;
589 
590 	if (lock_task_sighand(child, &flags)) {
591 		error = -EINVAL;
592 		if (likely(child->last_siginfo != NULL)) {
593 			*child->last_siginfo = *info;
594 			error = 0;
595 		}
596 		unlock_task_sighand(child, &flags);
597 	}
598 	return error;
599 }
600 
601 static int ptrace_peek_siginfo(struct task_struct *child,
602 				unsigned long addr,
603 				unsigned long data)
604 {
605 	struct ptrace_peeksiginfo_args arg;
606 	struct sigpending *pending;
607 	struct sigqueue *q;
608 	int ret, i;
609 
610 	ret = copy_from_user(&arg, (void __user *) addr,
611 				sizeof(struct ptrace_peeksiginfo_args));
612 	if (ret)
613 		return -EFAULT;
614 
615 	if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
616 		return -EINVAL; /* unknown flags */
617 
618 	if (arg.nr < 0)
619 		return -EINVAL;
620 
621 	if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
622 		pending = &child->signal->shared_pending;
623 	else
624 		pending = &child->pending;
625 
626 	for (i = 0; i < arg.nr; ) {
627 		siginfo_t info;
628 		s32 off = arg.off + i;
629 
630 		spin_lock_irq(&child->sighand->siglock);
631 		list_for_each_entry(q, &pending->list, list) {
632 			if (!off--) {
633 				copy_siginfo(&info, &q->info);
634 				break;
635 			}
636 		}
637 		spin_unlock_irq(&child->sighand->siglock);
638 
639 		if (off >= 0) /* beyond the end of the list */
640 			break;
641 
642 #ifdef CONFIG_COMPAT
643 		if (unlikely(is_compat_task())) {
644 			compat_siginfo_t __user *uinfo = compat_ptr(data);
645 
646 			if (copy_siginfo_to_user32(uinfo, &info) ||
647 			    __put_user(info.si_code, &uinfo->si_code)) {
648 				ret = -EFAULT;
649 				break;
650 			}
651 
652 		} else
653 #endif
654 		{
655 			siginfo_t __user *uinfo = (siginfo_t __user *) data;
656 
657 			if (copy_siginfo_to_user(uinfo, &info) ||
658 			    __put_user(info.si_code, &uinfo->si_code)) {
659 				ret = -EFAULT;
660 				break;
661 			}
662 		}
663 
664 		data += sizeof(siginfo_t);
665 		i++;
666 
667 		if (signal_pending(current))
668 			break;
669 
670 		cond_resched();
671 	}
672 
673 	if (i > 0)
674 		return i;
675 
676 	return ret;
677 }
678 
679 #ifdef PTRACE_SINGLESTEP
680 #define is_singlestep(request)		((request) == PTRACE_SINGLESTEP)
681 #else
682 #define is_singlestep(request)		0
683 #endif
684 
685 #ifdef PTRACE_SINGLEBLOCK
686 #define is_singleblock(request)		((request) == PTRACE_SINGLEBLOCK)
687 #else
688 #define is_singleblock(request)		0
689 #endif
690 
691 #ifdef PTRACE_SYSEMU
692 #define is_sysemu_singlestep(request)	((request) == PTRACE_SYSEMU_SINGLESTEP)
693 #else
694 #define is_sysemu_singlestep(request)	0
695 #endif
696 
697 static int ptrace_resume(struct task_struct *child, long request,
698 			 unsigned long data)
699 {
700 	if (!valid_signal(data))
701 		return -EIO;
702 
703 	if (request == PTRACE_SYSCALL)
704 		set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
705 	else
706 		clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
707 
708 #ifdef TIF_SYSCALL_EMU
709 	if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
710 		set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
711 	else
712 		clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
713 #endif
714 
715 	if (is_singleblock(request)) {
716 		if (unlikely(!arch_has_block_step()))
717 			return -EIO;
718 		user_enable_block_step(child);
719 	} else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
720 		if (unlikely(!arch_has_single_step()))
721 			return -EIO;
722 		user_enable_single_step(child);
723 	} else {
724 		user_disable_single_step(child);
725 	}
726 
727 	child->exit_code = data;
728 	wake_up_state(child, __TASK_TRACED);
729 
730 	return 0;
731 }
732 
733 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
734 
735 static const struct user_regset *
736 find_regset(const struct user_regset_view *view, unsigned int type)
737 {
738 	const struct user_regset *regset;
739 	int n;
740 
741 	for (n = 0; n < view->n; ++n) {
742 		regset = view->regsets + n;
743 		if (regset->core_note_type == type)
744 			return regset;
745 	}
746 
747 	return NULL;
748 }
749 
750 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
751 			 struct iovec *kiov)
752 {
753 	const struct user_regset_view *view = task_user_regset_view(task);
754 	const struct user_regset *regset = find_regset(view, type);
755 	int regset_no;
756 
757 	if (!regset || (kiov->iov_len % regset->size) != 0)
758 		return -EINVAL;
759 
760 	regset_no = regset - view->regsets;
761 	kiov->iov_len = min(kiov->iov_len,
762 			    (__kernel_size_t) (regset->n * regset->size));
763 
764 	if (req == PTRACE_GETREGSET)
765 		return copy_regset_to_user(task, view, regset_no, 0,
766 					   kiov->iov_len, kiov->iov_base);
767 	else
768 		return copy_regset_from_user(task, view, regset_no, 0,
769 					     kiov->iov_len, kiov->iov_base);
770 }
771 
772 /*
773  * This is declared in linux/regset.h and defined in machine-dependent
774  * code.  We put the export here, near the primary machine-neutral use,
775  * to ensure no machine forgets it.
776  */
777 EXPORT_SYMBOL_GPL(task_user_regset_view);
778 #endif
779 
780 int ptrace_request(struct task_struct *child, long request,
781 		   unsigned long addr, unsigned long data)
782 {
783 	bool seized = child->ptrace & PT_SEIZED;
784 	int ret = -EIO;
785 	siginfo_t siginfo, *si;
786 	void __user *datavp = (void __user *) data;
787 	unsigned long __user *datalp = datavp;
788 	unsigned long flags;
789 
790 	switch (request) {
791 	case PTRACE_PEEKTEXT:
792 	case PTRACE_PEEKDATA:
793 		return generic_ptrace_peekdata(child, addr, data);
794 	case PTRACE_POKETEXT:
795 	case PTRACE_POKEDATA:
796 		return generic_ptrace_pokedata(child, addr, data);
797 
798 #ifdef PTRACE_OLDSETOPTIONS
799 	case PTRACE_OLDSETOPTIONS:
800 #endif
801 	case PTRACE_SETOPTIONS:
802 		ret = ptrace_setoptions(child, data);
803 		break;
804 	case PTRACE_GETEVENTMSG:
805 		ret = put_user(child->ptrace_message, datalp);
806 		break;
807 
808 	case PTRACE_PEEKSIGINFO:
809 		ret = ptrace_peek_siginfo(child, addr, data);
810 		break;
811 
812 	case PTRACE_GETSIGINFO:
813 		ret = ptrace_getsiginfo(child, &siginfo);
814 		if (!ret)
815 			ret = copy_siginfo_to_user(datavp, &siginfo);
816 		break;
817 
818 	case PTRACE_SETSIGINFO:
819 		if (copy_from_user(&siginfo, datavp, sizeof siginfo))
820 			ret = -EFAULT;
821 		else
822 			ret = ptrace_setsiginfo(child, &siginfo);
823 		break;
824 
825 	case PTRACE_GETSIGMASK:
826 		if (addr != sizeof(sigset_t)) {
827 			ret = -EINVAL;
828 			break;
829 		}
830 
831 		if (copy_to_user(datavp, &child->blocked, sizeof(sigset_t)))
832 			ret = -EFAULT;
833 		else
834 			ret = 0;
835 
836 		break;
837 
838 	case PTRACE_SETSIGMASK: {
839 		sigset_t new_set;
840 
841 		if (addr != sizeof(sigset_t)) {
842 			ret = -EINVAL;
843 			break;
844 		}
845 
846 		if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
847 			ret = -EFAULT;
848 			break;
849 		}
850 
851 		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
852 
853 		/*
854 		 * Every thread does recalc_sigpending() after resume, so
855 		 * retarget_shared_pending() and recalc_sigpending() are not
856 		 * called here.
857 		 */
858 		spin_lock_irq(&child->sighand->siglock);
859 		child->blocked = new_set;
860 		spin_unlock_irq(&child->sighand->siglock);
861 
862 		ret = 0;
863 		break;
864 	}
865 
866 	case PTRACE_INTERRUPT:
867 		/*
868 		 * Stop tracee without any side-effect on signal or job
869 		 * control.  At least one trap is guaranteed to happen
870 		 * after this request.  If @child is already trapped, the
871 		 * current trap is not disturbed and another trap will
872 		 * happen after the current trap is ended with PTRACE_CONT.
873 		 *
874 		 * The actual trap might not be PTRACE_EVENT_STOP trap but
875 		 * the pending condition is cleared regardless.
876 		 */
877 		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
878 			break;
879 
880 		/*
881 		 * INTERRUPT doesn't disturb existing trap sans one
882 		 * exception.  If ptracer issued LISTEN for the current
883 		 * STOP, this INTERRUPT should clear LISTEN and re-trap
884 		 * tracee into STOP.
885 		 */
886 		if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
887 			ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
888 
889 		unlock_task_sighand(child, &flags);
890 		ret = 0;
891 		break;
892 
893 	case PTRACE_LISTEN:
894 		/*
895 		 * Listen for events.  Tracee must be in STOP.  It's not
896 		 * resumed per-se but is not considered to be in TRACED by
897 		 * wait(2) or ptrace(2).  If an async event (e.g. group
898 		 * stop state change) happens, tracee will enter STOP trap
899 		 * again.  Alternatively, ptracer can issue INTERRUPT to
900 		 * finish listening and re-trap tracee into STOP.
901 		 */
902 		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
903 			break;
904 
905 		si = child->last_siginfo;
906 		if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
907 			child->jobctl |= JOBCTL_LISTENING;
908 			/*
909 			 * If NOTIFY is set, it means event happened between
910 			 * start of this trap and now.  Trigger re-trap.
911 			 */
912 			if (child->jobctl & JOBCTL_TRAP_NOTIFY)
913 				ptrace_signal_wake_up(child, true);
914 			ret = 0;
915 		}
916 		unlock_task_sighand(child, &flags);
917 		break;
918 
919 	case PTRACE_DETACH:	 /* detach a process that was attached. */
920 		ret = ptrace_detach(child, data);
921 		break;
922 
923 #ifdef CONFIG_BINFMT_ELF_FDPIC
924 	case PTRACE_GETFDPIC: {
925 		struct mm_struct *mm = get_task_mm(child);
926 		unsigned long tmp = 0;
927 
928 		ret = -ESRCH;
929 		if (!mm)
930 			break;
931 
932 		switch (addr) {
933 		case PTRACE_GETFDPIC_EXEC:
934 			tmp = mm->context.exec_fdpic_loadmap;
935 			break;
936 		case PTRACE_GETFDPIC_INTERP:
937 			tmp = mm->context.interp_fdpic_loadmap;
938 			break;
939 		default:
940 			break;
941 		}
942 		mmput(mm);
943 
944 		ret = put_user(tmp, datalp);
945 		break;
946 	}
947 #endif
948 
949 #ifdef PTRACE_SINGLESTEP
950 	case PTRACE_SINGLESTEP:
951 #endif
952 #ifdef PTRACE_SINGLEBLOCK
953 	case PTRACE_SINGLEBLOCK:
954 #endif
955 #ifdef PTRACE_SYSEMU
956 	case PTRACE_SYSEMU:
957 	case PTRACE_SYSEMU_SINGLESTEP:
958 #endif
959 	case PTRACE_SYSCALL:
960 	case PTRACE_CONT:
961 		return ptrace_resume(child, request, data);
962 
963 	case PTRACE_KILL:
964 		if (child->exit_state)	/* already dead */
965 			return 0;
966 		return ptrace_resume(child, request, SIGKILL);
967 
968 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
969 	case PTRACE_GETREGSET:
970 	case PTRACE_SETREGSET: {
971 		struct iovec kiov;
972 		struct iovec __user *uiov = datavp;
973 
974 		if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
975 			return -EFAULT;
976 
977 		if (__get_user(kiov.iov_base, &uiov->iov_base) ||
978 		    __get_user(kiov.iov_len, &uiov->iov_len))
979 			return -EFAULT;
980 
981 		ret = ptrace_regset(child, request, addr, &kiov);
982 		if (!ret)
983 			ret = __put_user(kiov.iov_len, &uiov->iov_len);
984 		break;
985 	}
986 #endif
987 	default:
988 		break;
989 	}
990 
991 	return ret;
992 }
993 
994 static struct task_struct *ptrace_get_task_struct(pid_t pid)
995 {
996 	struct task_struct *child;
997 
998 	rcu_read_lock();
999 	child = find_task_by_vpid(pid);
1000 	if (child)
1001 		get_task_struct(child);
1002 	rcu_read_unlock();
1003 
1004 	if (!child)
1005 		return ERR_PTR(-ESRCH);
1006 	return child;
1007 }
1008 
1009 #ifndef arch_ptrace_attach
1010 #define arch_ptrace_attach(child)	do { } while (0)
1011 #endif
1012 
1013 SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1014 		unsigned long, data)
1015 {
1016 	struct task_struct *child;
1017 	long ret;
1018 
1019 	if (request == PTRACE_TRACEME) {
1020 		ret = ptrace_traceme();
1021 		if (!ret)
1022 			arch_ptrace_attach(current);
1023 		goto out;
1024 	}
1025 
1026 	child = ptrace_get_task_struct(pid);
1027 	if (IS_ERR(child)) {
1028 		ret = PTR_ERR(child);
1029 		goto out;
1030 	}
1031 
1032 	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1033 		ret = ptrace_attach(child, request, addr, data);
1034 		/*
1035 		 * Some architectures need to do book-keeping after
1036 		 * a ptrace attach.
1037 		 */
1038 		if (!ret)
1039 			arch_ptrace_attach(child);
1040 		goto out_put_task_struct;
1041 	}
1042 
1043 	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1044 				  request == PTRACE_INTERRUPT);
1045 	if (ret < 0)
1046 		goto out_put_task_struct;
1047 
1048 	ret = arch_ptrace(child, request, addr, data);
1049 	if (ret || request != PTRACE_DETACH)
1050 		ptrace_unfreeze_traced(child);
1051 
1052  out_put_task_struct:
1053 	put_task_struct(child);
1054  out:
1055 	return ret;
1056 }
1057 
1058 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1059 			    unsigned long data)
1060 {
1061 	unsigned long tmp;
1062 	int copied;
1063 
1064 	copied = access_process_vm(tsk, addr, &tmp, sizeof(tmp), 0);
1065 	if (copied != sizeof(tmp))
1066 		return -EIO;
1067 	return put_user(tmp, (unsigned long __user *)data);
1068 }
1069 
1070 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1071 			    unsigned long data)
1072 {
1073 	int copied;
1074 
1075 	copied = access_process_vm(tsk, addr, &data, sizeof(data), 1);
1076 	return (copied == sizeof(data)) ? 0 : -EIO;
1077 }
1078 
1079 #if defined CONFIG_COMPAT
1080 
1081 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1082 			  compat_ulong_t addr, compat_ulong_t data)
1083 {
1084 	compat_ulong_t __user *datap = compat_ptr(data);
1085 	compat_ulong_t word;
1086 	siginfo_t siginfo;
1087 	int ret;
1088 
1089 	switch (request) {
1090 	case PTRACE_PEEKTEXT:
1091 	case PTRACE_PEEKDATA:
1092 		ret = access_process_vm(child, addr, &word, sizeof(word), 0);
1093 		if (ret != sizeof(word))
1094 			ret = -EIO;
1095 		else
1096 			ret = put_user(word, datap);
1097 		break;
1098 
1099 	case PTRACE_POKETEXT:
1100 	case PTRACE_POKEDATA:
1101 		ret = access_process_vm(child, addr, &data, sizeof(data), 1);
1102 		ret = (ret != sizeof(data) ? -EIO : 0);
1103 		break;
1104 
1105 	case PTRACE_GETEVENTMSG:
1106 		ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1107 		break;
1108 
1109 	case PTRACE_GETSIGINFO:
1110 		ret = ptrace_getsiginfo(child, &siginfo);
1111 		if (!ret)
1112 			ret = copy_siginfo_to_user32(
1113 				(struct compat_siginfo __user *) datap,
1114 				&siginfo);
1115 		break;
1116 
1117 	case PTRACE_SETSIGINFO:
1118 		memset(&siginfo, 0, sizeof siginfo);
1119 		if (copy_siginfo_from_user32(
1120 			    &siginfo, (struct compat_siginfo __user *) datap))
1121 			ret = -EFAULT;
1122 		else
1123 			ret = ptrace_setsiginfo(child, &siginfo);
1124 		break;
1125 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1126 	case PTRACE_GETREGSET:
1127 	case PTRACE_SETREGSET:
1128 	{
1129 		struct iovec kiov;
1130 		struct compat_iovec __user *uiov =
1131 			(struct compat_iovec __user *) datap;
1132 		compat_uptr_t ptr;
1133 		compat_size_t len;
1134 
1135 		if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1136 			return -EFAULT;
1137 
1138 		if (__get_user(ptr, &uiov->iov_base) ||
1139 		    __get_user(len, &uiov->iov_len))
1140 			return -EFAULT;
1141 
1142 		kiov.iov_base = compat_ptr(ptr);
1143 		kiov.iov_len = len;
1144 
1145 		ret = ptrace_regset(child, request, addr, &kiov);
1146 		if (!ret)
1147 			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1148 		break;
1149 	}
1150 #endif
1151 
1152 	default:
1153 		ret = ptrace_request(child, request, addr, data);
1154 	}
1155 
1156 	return ret;
1157 }
1158 
1159 COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1160 		       compat_long_t, addr, compat_long_t, data)
1161 {
1162 	struct task_struct *child;
1163 	long ret;
1164 
1165 	if (request == PTRACE_TRACEME) {
1166 		ret = ptrace_traceme();
1167 		goto out;
1168 	}
1169 
1170 	child = ptrace_get_task_struct(pid);
1171 	if (IS_ERR(child)) {
1172 		ret = PTR_ERR(child);
1173 		goto out;
1174 	}
1175 
1176 	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1177 		ret = ptrace_attach(child, request, addr, data);
1178 		/*
1179 		 * Some architectures need to do book-keeping after
1180 		 * a ptrace attach.
1181 		 */
1182 		if (!ret)
1183 			arch_ptrace_attach(child);
1184 		goto out_put_task_struct;
1185 	}
1186 
1187 	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1188 				  request == PTRACE_INTERRUPT);
1189 	if (!ret) {
1190 		ret = compat_arch_ptrace(child, request, addr, data);
1191 		if (ret || request != PTRACE_DETACH)
1192 			ptrace_unfreeze_traced(child);
1193 	}
1194 
1195  out_put_task_struct:
1196 	put_task_struct(child);
1197  out:
1198 	return ret;
1199 }
1200 #endif	/* CONFIG_COMPAT */
1201