xref: /linux/kernel/ptrace.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 	if (!valid_signal(data))
460 		return -EIO;
461 
462 	/* Architecture-specific hardware disable .. */
463 	ptrace_disable(child);
464 	clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
465 
466 	write_lock_irq(&tasklist_lock);
467 	/*
468 	 * We rely on ptrace_freeze_traced(). It can't be killed and
469 	 * untraced by another thread, it can't be a zombie.
470 	 */
471 	WARN_ON(!child->ptrace || child->exit_state);
472 	/*
473 	 * tasklist_lock avoids the race with wait_task_stopped(), see
474 	 * the comment in ptrace_resume().
475 	 */
476 	child->exit_code = data;
477 	__ptrace_detach(current, child);
478 	write_unlock_irq(&tasklist_lock);
479 
480 	proc_ptrace_connector(child, PTRACE_DETACH);
481 
482 	return 0;
483 }
484 
485 /*
486  * Detach all tasks we were using ptrace on. Called with tasklist held
487  * for writing.
488  */
489 void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
490 {
491 	struct task_struct *p, *n;
492 
493 	list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
494 		if (unlikely(p->ptrace & PT_EXITKILL))
495 			send_sig_info(SIGKILL, SEND_SIG_FORCED, p);
496 
497 		if (__ptrace_detach(tracer, p))
498 			list_add(&p->ptrace_entry, dead);
499 	}
500 }
501 
502 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
503 {
504 	int copied = 0;
505 
506 	while (len > 0) {
507 		char buf[128];
508 		int this_len, retval;
509 
510 		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
511 		retval = access_process_vm(tsk, src, buf, this_len, 0);
512 		if (!retval) {
513 			if (copied)
514 				break;
515 			return -EIO;
516 		}
517 		if (copy_to_user(dst, buf, retval))
518 			return -EFAULT;
519 		copied += retval;
520 		src += retval;
521 		dst += retval;
522 		len -= retval;
523 	}
524 	return copied;
525 }
526 
527 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
528 {
529 	int copied = 0;
530 
531 	while (len > 0) {
532 		char buf[128];
533 		int this_len, retval;
534 
535 		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
536 		if (copy_from_user(buf, src, this_len))
537 			return -EFAULT;
538 		retval = access_process_vm(tsk, dst, buf, this_len, 1);
539 		if (!retval) {
540 			if (copied)
541 				break;
542 			return -EIO;
543 		}
544 		copied += retval;
545 		src += retval;
546 		dst += retval;
547 		len -= retval;
548 	}
549 	return copied;
550 }
551 
552 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
553 {
554 	unsigned flags;
555 
556 	if (data & ~(unsigned long)PTRACE_O_MASK)
557 		return -EINVAL;
558 
559 	if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
560 		if (!config_enabled(CONFIG_CHECKPOINT_RESTORE) ||
561 		    !config_enabled(CONFIG_SECCOMP))
562 			return -EINVAL;
563 
564 		if (!capable(CAP_SYS_ADMIN))
565 			return -EPERM;
566 
567 		if (seccomp_mode(&current->seccomp) != SECCOMP_MODE_DISABLED ||
568 		    current->ptrace & PT_SUSPEND_SECCOMP)
569 			return -EPERM;
570 	}
571 
572 	/* Avoid intermediate state when all opts are cleared */
573 	flags = child->ptrace;
574 	flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
575 	flags |= (data << PT_OPT_FLAG_SHIFT);
576 	child->ptrace = flags;
577 
578 	return 0;
579 }
580 
581 static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
582 {
583 	unsigned long flags;
584 	int error = -ESRCH;
585 
586 	if (lock_task_sighand(child, &flags)) {
587 		error = -EINVAL;
588 		if (likely(child->last_siginfo != NULL)) {
589 			*info = *child->last_siginfo;
590 			error = 0;
591 		}
592 		unlock_task_sighand(child, &flags);
593 	}
594 	return error;
595 }
596 
597 static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
598 {
599 	unsigned long flags;
600 	int error = -ESRCH;
601 
602 	if (lock_task_sighand(child, &flags)) {
603 		error = -EINVAL;
604 		if (likely(child->last_siginfo != NULL)) {
605 			*child->last_siginfo = *info;
606 			error = 0;
607 		}
608 		unlock_task_sighand(child, &flags);
609 	}
610 	return error;
611 }
612 
613 static int ptrace_peek_siginfo(struct task_struct *child,
614 				unsigned long addr,
615 				unsigned long data)
616 {
617 	struct ptrace_peeksiginfo_args arg;
618 	struct sigpending *pending;
619 	struct sigqueue *q;
620 	int ret, i;
621 
622 	ret = copy_from_user(&arg, (void __user *) addr,
623 				sizeof(struct ptrace_peeksiginfo_args));
624 	if (ret)
625 		return -EFAULT;
626 
627 	if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
628 		return -EINVAL; /* unknown flags */
629 
630 	if (arg.nr < 0)
631 		return -EINVAL;
632 
633 	if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
634 		pending = &child->signal->shared_pending;
635 	else
636 		pending = &child->pending;
637 
638 	for (i = 0; i < arg.nr; ) {
639 		siginfo_t info;
640 		s32 off = arg.off + i;
641 
642 		spin_lock_irq(&child->sighand->siglock);
643 		list_for_each_entry(q, &pending->list, list) {
644 			if (!off--) {
645 				copy_siginfo(&info, &q->info);
646 				break;
647 			}
648 		}
649 		spin_unlock_irq(&child->sighand->siglock);
650 
651 		if (off >= 0) /* beyond the end of the list */
652 			break;
653 
654 #ifdef CONFIG_COMPAT
655 		if (unlikely(is_compat_task())) {
656 			compat_siginfo_t __user *uinfo = compat_ptr(data);
657 
658 			if (copy_siginfo_to_user32(uinfo, &info) ||
659 			    __put_user(info.si_code, &uinfo->si_code)) {
660 				ret = -EFAULT;
661 				break;
662 			}
663 
664 		} else
665 #endif
666 		{
667 			siginfo_t __user *uinfo = (siginfo_t __user *) data;
668 
669 			if (copy_siginfo_to_user(uinfo, &info) ||
670 			    __put_user(info.si_code, &uinfo->si_code)) {
671 				ret = -EFAULT;
672 				break;
673 			}
674 		}
675 
676 		data += sizeof(siginfo_t);
677 		i++;
678 
679 		if (signal_pending(current))
680 			break;
681 
682 		cond_resched();
683 	}
684 
685 	if (i > 0)
686 		return i;
687 
688 	return ret;
689 }
690 
691 #ifdef PTRACE_SINGLESTEP
692 #define is_singlestep(request)		((request) == PTRACE_SINGLESTEP)
693 #else
694 #define is_singlestep(request)		0
695 #endif
696 
697 #ifdef PTRACE_SINGLEBLOCK
698 #define is_singleblock(request)		((request) == PTRACE_SINGLEBLOCK)
699 #else
700 #define is_singleblock(request)		0
701 #endif
702 
703 #ifdef PTRACE_SYSEMU
704 #define is_sysemu_singlestep(request)	((request) == PTRACE_SYSEMU_SINGLESTEP)
705 #else
706 #define is_sysemu_singlestep(request)	0
707 #endif
708 
709 static int ptrace_resume(struct task_struct *child, long request,
710 			 unsigned long data)
711 {
712 	bool need_siglock;
713 
714 	if (!valid_signal(data))
715 		return -EIO;
716 
717 	if (request == PTRACE_SYSCALL)
718 		set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
719 	else
720 		clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
721 
722 #ifdef TIF_SYSCALL_EMU
723 	if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
724 		set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
725 	else
726 		clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
727 #endif
728 
729 	if (is_singleblock(request)) {
730 		if (unlikely(!arch_has_block_step()))
731 			return -EIO;
732 		user_enable_block_step(child);
733 	} else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
734 		if (unlikely(!arch_has_single_step()))
735 			return -EIO;
736 		user_enable_single_step(child);
737 	} else {
738 		user_disable_single_step(child);
739 	}
740 
741 	/*
742 	 * Change ->exit_code and ->state under siglock to avoid the race
743 	 * with wait_task_stopped() in between; a non-zero ->exit_code will
744 	 * wrongly look like another report from tracee.
745 	 *
746 	 * Note that we need siglock even if ->exit_code == data and/or this
747 	 * status was not reported yet, the new status must not be cleared by
748 	 * wait_task_stopped() after resume.
749 	 *
750 	 * If data == 0 we do not care if wait_task_stopped() reports the old
751 	 * status and clears the code too; this can't race with the tracee, it
752 	 * takes siglock after resume.
753 	 */
754 	need_siglock = data && !thread_group_empty(current);
755 	if (need_siglock)
756 		spin_lock_irq(&child->sighand->siglock);
757 	child->exit_code = data;
758 	wake_up_state(child, __TASK_TRACED);
759 	if (need_siglock)
760 		spin_unlock_irq(&child->sighand->siglock);
761 
762 	return 0;
763 }
764 
765 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
766 
767 static const struct user_regset *
768 find_regset(const struct user_regset_view *view, unsigned int type)
769 {
770 	const struct user_regset *regset;
771 	int n;
772 
773 	for (n = 0; n < view->n; ++n) {
774 		regset = view->regsets + n;
775 		if (regset->core_note_type == type)
776 			return regset;
777 	}
778 
779 	return NULL;
780 }
781 
782 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
783 			 struct iovec *kiov)
784 {
785 	const struct user_regset_view *view = task_user_regset_view(task);
786 	const struct user_regset *regset = find_regset(view, type);
787 	int regset_no;
788 
789 	if (!regset || (kiov->iov_len % regset->size) != 0)
790 		return -EINVAL;
791 
792 	regset_no = regset - view->regsets;
793 	kiov->iov_len = min(kiov->iov_len,
794 			    (__kernel_size_t) (regset->n * regset->size));
795 
796 	if (req == PTRACE_GETREGSET)
797 		return copy_regset_to_user(task, view, regset_no, 0,
798 					   kiov->iov_len, kiov->iov_base);
799 	else
800 		return copy_regset_from_user(task, view, regset_no, 0,
801 					     kiov->iov_len, kiov->iov_base);
802 }
803 
804 /*
805  * This is declared in linux/regset.h and defined in machine-dependent
806  * code.  We put the export here, near the primary machine-neutral use,
807  * to ensure no machine forgets it.
808  */
809 EXPORT_SYMBOL_GPL(task_user_regset_view);
810 #endif
811 
812 int ptrace_request(struct task_struct *child, long request,
813 		   unsigned long addr, unsigned long data)
814 {
815 	bool seized = child->ptrace & PT_SEIZED;
816 	int ret = -EIO;
817 	siginfo_t siginfo, *si;
818 	void __user *datavp = (void __user *) data;
819 	unsigned long __user *datalp = datavp;
820 	unsigned long flags;
821 
822 	switch (request) {
823 	case PTRACE_PEEKTEXT:
824 	case PTRACE_PEEKDATA:
825 		return generic_ptrace_peekdata(child, addr, data);
826 	case PTRACE_POKETEXT:
827 	case PTRACE_POKEDATA:
828 		return generic_ptrace_pokedata(child, addr, data);
829 
830 #ifdef PTRACE_OLDSETOPTIONS
831 	case PTRACE_OLDSETOPTIONS:
832 #endif
833 	case PTRACE_SETOPTIONS:
834 		ret = ptrace_setoptions(child, data);
835 		break;
836 	case PTRACE_GETEVENTMSG:
837 		ret = put_user(child->ptrace_message, datalp);
838 		break;
839 
840 	case PTRACE_PEEKSIGINFO:
841 		ret = ptrace_peek_siginfo(child, addr, data);
842 		break;
843 
844 	case PTRACE_GETSIGINFO:
845 		ret = ptrace_getsiginfo(child, &siginfo);
846 		if (!ret)
847 			ret = copy_siginfo_to_user(datavp, &siginfo);
848 		break;
849 
850 	case PTRACE_SETSIGINFO:
851 		if (copy_from_user(&siginfo, datavp, sizeof siginfo))
852 			ret = -EFAULT;
853 		else
854 			ret = ptrace_setsiginfo(child, &siginfo);
855 		break;
856 
857 	case PTRACE_GETSIGMASK:
858 		if (addr != sizeof(sigset_t)) {
859 			ret = -EINVAL;
860 			break;
861 		}
862 
863 		if (copy_to_user(datavp, &child->blocked, sizeof(sigset_t)))
864 			ret = -EFAULT;
865 		else
866 			ret = 0;
867 
868 		break;
869 
870 	case PTRACE_SETSIGMASK: {
871 		sigset_t new_set;
872 
873 		if (addr != sizeof(sigset_t)) {
874 			ret = -EINVAL;
875 			break;
876 		}
877 
878 		if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
879 			ret = -EFAULT;
880 			break;
881 		}
882 
883 		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
884 
885 		/*
886 		 * Every thread does recalc_sigpending() after resume, so
887 		 * retarget_shared_pending() and recalc_sigpending() are not
888 		 * called here.
889 		 */
890 		spin_lock_irq(&child->sighand->siglock);
891 		child->blocked = new_set;
892 		spin_unlock_irq(&child->sighand->siglock);
893 
894 		ret = 0;
895 		break;
896 	}
897 
898 	case PTRACE_INTERRUPT:
899 		/*
900 		 * Stop tracee without any side-effect on signal or job
901 		 * control.  At least one trap is guaranteed to happen
902 		 * after this request.  If @child is already trapped, the
903 		 * current trap is not disturbed and another trap will
904 		 * happen after the current trap is ended with PTRACE_CONT.
905 		 *
906 		 * The actual trap might not be PTRACE_EVENT_STOP trap but
907 		 * the pending condition is cleared regardless.
908 		 */
909 		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
910 			break;
911 
912 		/*
913 		 * INTERRUPT doesn't disturb existing trap sans one
914 		 * exception.  If ptracer issued LISTEN for the current
915 		 * STOP, this INTERRUPT should clear LISTEN and re-trap
916 		 * tracee into STOP.
917 		 */
918 		if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
919 			ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
920 
921 		unlock_task_sighand(child, &flags);
922 		ret = 0;
923 		break;
924 
925 	case PTRACE_LISTEN:
926 		/*
927 		 * Listen for events.  Tracee must be in STOP.  It's not
928 		 * resumed per-se but is not considered to be in TRACED by
929 		 * wait(2) or ptrace(2).  If an async event (e.g. group
930 		 * stop state change) happens, tracee will enter STOP trap
931 		 * again.  Alternatively, ptracer can issue INTERRUPT to
932 		 * finish listening and re-trap tracee into STOP.
933 		 */
934 		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
935 			break;
936 
937 		si = child->last_siginfo;
938 		if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
939 			child->jobctl |= JOBCTL_LISTENING;
940 			/*
941 			 * If NOTIFY is set, it means event happened between
942 			 * start of this trap and now.  Trigger re-trap.
943 			 */
944 			if (child->jobctl & JOBCTL_TRAP_NOTIFY)
945 				ptrace_signal_wake_up(child, true);
946 			ret = 0;
947 		}
948 		unlock_task_sighand(child, &flags);
949 		break;
950 
951 	case PTRACE_DETACH:	 /* detach a process that was attached. */
952 		ret = ptrace_detach(child, data);
953 		break;
954 
955 #ifdef CONFIG_BINFMT_ELF_FDPIC
956 	case PTRACE_GETFDPIC: {
957 		struct mm_struct *mm = get_task_mm(child);
958 		unsigned long tmp = 0;
959 
960 		ret = -ESRCH;
961 		if (!mm)
962 			break;
963 
964 		switch (addr) {
965 		case PTRACE_GETFDPIC_EXEC:
966 			tmp = mm->context.exec_fdpic_loadmap;
967 			break;
968 		case PTRACE_GETFDPIC_INTERP:
969 			tmp = mm->context.interp_fdpic_loadmap;
970 			break;
971 		default:
972 			break;
973 		}
974 		mmput(mm);
975 
976 		ret = put_user(tmp, datalp);
977 		break;
978 	}
979 #endif
980 
981 #ifdef PTRACE_SINGLESTEP
982 	case PTRACE_SINGLESTEP:
983 #endif
984 #ifdef PTRACE_SINGLEBLOCK
985 	case PTRACE_SINGLEBLOCK:
986 #endif
987 #ifdef PTRACE_SYSEMU
988 	case PTRACE_SYSEMU:
989 	case PTRACE_SYSEMU_SINGLESTEP:
990 #endif
991 	case PTRACE_SYSCALL:
992 	case PTRACE_CONT:
993 		return ptrace_resume(child, request, data);
994 
995 	case PTRACE_KILL:
996 		if (child->exit_state)	/* already dead */
997 			return 0;
998 		return ptrace_resume(child, request, SIGKILL);
999 
1000 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1001 	case PTRACE_GETREGSET:
1002 	case PTRACE_SETREGSET: {
1003 		struct iovec kiov;
1004 		struct iovec __user *uiov = datavp;
1005 
1006 		if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1007 			return -EFAULT;
1008 
1009 		if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1010 		    __get_user(kiov.iov_len, &uiov->iov_len))
1011 			return -EFAULT;
1012 
1013 		ret = ptrace_regset(child, request, addr, &kiov);
1014 		if (!ret)
1015 			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1016 		break;
1017 	}
1018 #endif
1019 	default:
1020 		break;
1021 	}
1022 
1023 	return ret;
1024 }
1025 
1026 static struct task_struct *ptrace_get_task_struct(pid_t pid)
1027 {
1028 	struct task_struct *child;
1029 
1030 	rcu_read_lock();
1031 	child = find_task_by_vpid(pid);
1032 	if (child)
1033 		get_task_struct(child);
1034 	rcu_read_unlock();
1035 
1036 	if (!child)
1037 		return ERR_PTR(-ESRCH);
1038 	return child;
1039 }
1040 
1041 #ifndef arch_ptrace_attach
1042 #define arch_ptrace_attach(child)	do { } while (0)
1043 #endif
1044 
1045 SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1046 		unsigned long, data)
1047 {
1048 	struct task_struct *child;
1049 	long ret;
1050 
1051 	if (request == PTRACE_TRACEME) {
1052 		ret = ptrace_traceme();
1053 		if (!ret)
1054 			arch_ptrace_attach(current);
1055 		goto out;
1056 	}
1057 
1058 	child = ptrace_get_task_struct(pid);
1059 	if (IS_ERR(child)) {
1060 		ret = PTR_ERR(child);
1061 		goto out;
1062 	}
1063 
1064 	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1065 		ret = ptrace_attach(child, request, addr, data);
1066 		/*
1067 		 * Some architectures need to do book-keeping after
1068 		 * a ptrace attach.
1069 		 */
1070 		if (!ret)
1071 			arch_ptrace_attach(child);
1072 		goto out_put_task_struct;
1073 	}
1074 
1075 	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1076 				  request == PTRACE_INTERRUPT);
1077 	if (ret < 0)
1078 		goto out_put_task_struct;
1079 
1080 	ret = arch_ptrace(child, request, addr, data);
1081 	if (ret || request != PTRACE_DETACH)
1082 		ptrace_unfreeze_traced(child);
1083 
1084  out_put_task_struct:
1085 	put_task_struct(child);
1086  out:
1087 	return ret;
1088 }
1089 
1090 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1091 			    unsigned long data)
1092 {
1093 	unsigned long tmp;
1094 	int copied;
1095 
1096 	copied = access_process_vm(tsk, addr, &tmp, sizeof(tmp), 0);
1097 	if (copied != sizeof(tmp))
1098 		return -EIO;
1099 	return put_user(tmp, (unsigned long __user *)data);
1100 }
1101 
1102 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1103 			    unsigned long data)
1104 {
1105 	int copied;
1106 
1107 	copied = access_process_vm(tsk, addr, &data, sizeof(data), 1);
1108 	return (copied == sizeof(data)) ? 0 : -EIO;
1109 }
1110 
1111 #if defined CONFIG_COMPAT
1112 
1113 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1114 			  compat_ulong_t addr, compat_ulong_t data)
1115 {
1116 	compat_ulong_t __user *datap = compat_ptr(data);
1117 	compat_ulong_t word;
1118 	siginfo_t siginfo;
1119 	int ret;
1120 
1121 	switch (request) {
1122 	case PTRACE_PEEKTEXT:
1123 	case PTRACE_PEEKDATA:
1124 		ret = access_process_vm(child, addr, &word, sizeof(word), 0);
1125 		if (ret != sizeof(word))
1126 			ret = -EIO;
1127 		else
1128 			ret = put_user(word, datap);
1129 		break;
1130 
1131 	case PTRACE_POKETEXT:
1132 	case PTRACE_POKEDATA:
1133 		ret = access_process_vm(child, addr, &data, sizeof(data), 1);
1134 		ret = (ret != sizeof(data) ? -EIO : 0);
1135 		break;
1136 
1137 	case PTRACE_GETEVENTMSG:
1138 		ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1139 		break;
1140 
1141 	case PTRACE_GETSIGINFO:
1142 		ret = ptrace_getsiginfo(child, &siginfo);
1143 		if (!ret)
1144 			ret = copy_siginfo_to_user32(
1145 				(struct compat_siginfo __user *) datap,
1146 				&siginfo);
1147 		break;
1148 
1149 	case PTRACE_SETSIGINFO:
1150 		memset(&siginfo, 0, sizeof siginfo);
1151 		if (copy_siginfo_from_user32(
1152 			    &siginfo, (struct compat_siginfo __user *) datap))
1153 			ret = -EFAULT;
1154 		else
1155 			ret = ptrace_setsiginfo(child, &siginfo);
1156 		break;
1157 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1158 	case PTRACE_GETREGSET:
1159 	case PTRACE_SETREGSET:
1160 	{
1161 		struct iovec kiov;
1162 		struct compat_iovec __user *uiov =
1163 			(struct compat_iovec __user *) datap;
1164 		compat_uptr_t ptr;
1165 		compat_size_t len;
1166 
1167 		if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1168 			return -EFAULT;
1169 
1170 		if (__get_user(ptr, &uiov->iov_base) ||
1171 		    __get_user(len, &uiov->iov_len))
1172 			return -EFAULT;
1173 
1174 		kiov.iov_base = compat_ptr(ptr);
1175 		kiov.iov_len = len;
1176 
1177 		ret = ptrace_regset(child, request, addr, &kiov);
1178 		if (!ret)
1179 			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1180 		break;
1181 	}
1182 #endif
1183 
1184 	default:
1185 		ret = ptrace_request(child, request, addr, data);
1186 	}
1187 
1188 	return ret;
1189 }
1190 
1191 COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1192 		       compat_long_t, addr, compat_long_t, data)
1193 {
1194 	struct task_struct *child;
1195 	long ret;
1196 
1197 	if (request == PTRACE_TRACEME) {
1198 		ret = ptrace_traceme();
1199 		goto out;
1200 	}
1201 
1202 	child = ptrace_get_task_struct(pid);
1203 	if (IS_ERR(child)) {
1204 		ret = PTR_ERR(child);
1205 		goto out;
1206 	}
1207 
1208 	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1209 		ret = ptrace_attach(child, request, addr, data);
1210 		/*
1211 		 * Some architectures need to do book-keeping after
1212 		 * a ptrace attach.
1213 		 */
1214 		if (!ret)
1215 			arch_ptrace_attach(child);
1216 		goto out_put_task_struct;
1217 	}
1218 
1219 	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1220 				  request == PTRACE_INTERRUPT);
1221 	if (!ret) {
1222 		ret = compat_arch_ptrace(child, request, addr, data);
1223 		if (ret || request != PTRACE_DETACH)
1224 			ptrace_unfreeze_traced(child);
1225 	}
1226 
1227  out_put_task_struct:
1228 	put_task_struct(child);
1229  out:
1230 	return ret;
1231 }
1232 #endif	/* CONFIG_COMPAT */
1233