xref: /linux/kernel/ptrace.c (revision 17afab1de42236ee2f6235f4383cc6f3f13f8a10)
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/audit.h>
21 #include <linux/pid_namespace.h>
22 #include <linux/syscalls.h>
23 #include <linux/uaccess.h>
24 #include <linux/regset.h>
25 #include <linux/hw_breakpoint.h>
26 #include <linux/cn_proc.h>
27 #include <linux/compat.h>
28 
29 
30 static int ptrace_trapping_sleep_fn(void *flags)
31 {
32 	schedule();
33 	return 0;
34 }
35 
36 /*
37  * ptrace a task: make the debugger its new parent and
38  * move it to the ptrace list.
39  *
40  * Must be called with the tasklist lock write-held.
41  */
42 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent)
43 {
44 	BUG_ON(!list_empty(&child->ptrace_entry));
45 	list_add(&child->ptrace_entry, &new_parent->ptraced);
46 	child->parent = new_parent;
47 }
48 
49 /**
50  * __ptrace_unlink - unlink ptracee and restore its execution state
51  * @child: ptracee to be unlinked
52  *
53  * Remove @child from the ptrace list, move it back to the original parent,
54  * and restore the execution state so that it conforms to the group stop
55  * state.
56  *
57  * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
58  * exiting.  For PTRACE_DETACH, unless the ptracee has been killed between
59  * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
60  * If the ptracer is exiting, the ptracee can be in any state.
61  *
62  * After detach, the ptracee should be in a state which conforms to the
63  * group stop.  If the group is stopped or in the process of stopping, the
64  * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
65  * up from TASK_TRACED.
66  *
67  * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
68  * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
69  * to but in the opposite direction of what happens while attaching to a
70  * stopped task.  However, in this direction, the intermediate RUNNING
71  * state is not hidden even from the current ptracer and if it immediately
72  * re-attaches and performs a WNOHANG wait(2), it may fail.
73  *
74  * CONTEXT:
75  * write_lock_irq(tasklist_lock)
76  */
77 void __ptrace_unlink(struct task_struct *child)
78 {
79 	BUG_ON(!child->ptrace);
80 
81 	child->ptrace = 0;
82 	child->parent = child->real_parent;
83 	list_del_init(&child->ptrace_entry);
84 
85 	spin_lock(&child->sighand->siglock);
86 
87 	/*
88 	 * Clear all pending traps and TRAPPING.  TRAPPING should be
89 	 * cleared regardless of JOBCTL_STOP_PENDING.  Do it explicitly.
90 	 */
91 	task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
92 	task_clear_jobctl_trapping(child);
93 
94 	/*
95 	 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
96 	 * @child isn't dead.
97 	 */
98 	if (!(child->flags & PF_EXITING) &&
99 	    (child->signal->flags & SIGNAL_STOP_STOPPED ||
100 	     child->signal->group_stop_count)) {
101 		child->jobctl |= JOBCTL_STOP_PENDING;
102 
103 		/*
104 		 * This is only possible if this thread was cloned by the
105 		 * traced task running in the stopped group, set the signal
106 		 * for the future reports.
107 		 * FIXME: we should change ptrace_init_task() to handle this
108 		 * case.
109 		 */
110 		if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
111 			child->jobctl |= SIGSTOP;
112 	}
113 
114 	/*
115 	 * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
116 	 * @child in the butt.  Note that @resume should be used iff @child
117 	 * is in TASK_TRACED; otherwise, we might unduly disrupt
118 	 * TASK_KILLABLE sleeps.
119 	 */
120 	if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
121 		ptrace_signal_wake_up(child, true);
122 
123 	spin_unlock(&child->sighand->siglock);
124 }
125 
126 /* Ensure that nothing can wake it up, even SIGKILL */
127 static bool ptrace_freeze_traced(struct task_struct *task)
128 {
129 	bool ret = false;
130 
131 	/* Lockless, nobody but us can set this flag */
132 	if (task->jobctl & JOBCTL_LISTENING)
133 		return ret;
134 
135 	spin_lock_irq(&task->sighand->siglock);
136 	if (task_is_traced(task) && !__fatal_signal_pending(task)) {
137 		task->state = __TASK_TRACED;
138 		ret = true;
139 	}
140 	spin_unlock_irq(&task->sighand->siglock);
141 
142 	return ret;
143 }
144 
145 static void ptrace_unfreeze_traced(struct task_struct *task)
146 {
147 	if (task->state != __TASK_TRACED)
148 		return;
149 
150 	WARN_ON(!task->ptrace || task->parent != current);
151 
152 	spin_lock_irq(&task->sighand->siglock);
153 	if (__fatal_signal_pending(task))
154 		wake_up_state(task, __TASK_TRACED);
155 	else
156 		task->state = TASK_TRACED;
157 	spin_unlock_irq(&task->sighand->siglock);
158 }
159 
160 /**
161  * ptrace_check_attach - check whether ptracee is ready for ptrace operation
162  * @child: ptracee to check for
163  * @ignore_state: don't check whether @child is currently %TASK_TRACED
164  *
165  * Check whether @child is being ptraced by %current and ready for further
166  * ptrace operations.  If @ignore_state is %false, @child also should be in
167  * %TASK_TRACED state and on return the child is guaranteed to be traced
168  * and not executing.  If @ignore_state is %true, @child can be in any
169  * state.
170  *
171  * CONTEXT:
172  * Grabs and releases tasklist_lock and @child->sighand->siglock.
173  *
174  * RETURNS:
175  * 0 on success, -ESRCH if %child is not ready.
176  */
177 static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
178 {
179 	int ret = -ESRCH;
180 
181 	/*
182 	 * We take the read lock around doing both checks to close a
183 	 * possible race where someone else was tracing our child and
184 	 * detached between these two checks.  After this locked check,
185 	 * we are sure that this is our traced child and that can only
186 	 * be changed by us so it's not changing right after this.
187 	 */
188 	read_lock(&tasklist_lock);
189 	if (child->ptrace && child->parent == current) {
190 		WARN_ON(child->state == __TASK_TRACED);
191 		/*
192 		 * child->sighand can't be NULL, release_task()
193 		 * does ptrace_unlink() before __exit_signal().
194 		 */
195 		if (ignore_state || ptrace_freeze_traced(child))
196 			ret = 0;
197 	}
198 	read_unlock(&tasklist_lock);
199 
200 	if (!ret && !ignore_state) {
201 		if (!wait_task_inactive(child, __TASK_TRACED)) {
202 			/*
203 			 * This can only happen if may_ptrace_stop() fails and
204 			 * ptrace_stop() changes ->state back to TASK_RUNNING,
205 			 * so we should not worry about leaking __TASK_TRACED.
206 			 */
207 			WARN_ON(child->state == __TASK_TRACED);
208 			ret = -ESRCH;
209 		}
210 	}
211 
212 	return ret;
213 }
214 
215 static int ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
216 {
217 	if (mode & PTRACE_MODE_NOAUDIT)
218 		return has_ns_capability_noaudit(current, ns, CAP_SYS_PTRACE);
219 	else
220 		return has_ns_capability(current, ns, CAP_SYS_PTRACE);
221 }
222 
223 /* Returns 0 on success, -errno on denial. */
224 static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
225 {
226 	const struct cred *cred = current_cred(), *tcred;
227 
228 	/* May we inspect the given task?
229 	 * This check is used both for attaching with ptrace
230 	 * and for allowing access to sensitive information in /proc.
231 	 *
232 	 * ptrace_attach denies several cases that /proc allows
233 	 * because setting up the necessary parent/child relationship
234 	 * or halting the specified task is impossible.
235 	 */
236 	int dumpable = 0;
237 	/* Don't let security modules deny introspection */
238 	if (task == current)
239 		return 0;
240 	rcu_read_lock();
241 	tcred = __task_cred(task);
242 	if (uid_eq(cred->uid, tcred->euid) &&
243 	    uid_eq(cred->uid, tcred->suid) &&
244 	    uid_eq(cred->uid, tcred->uid)  &&
245 	    gid_eq(cred->gid, tcred->egid) &&
246 	    gid_eq(cred->gid, tcred->sgid) &&
247 	    gid_eq(cred->gid, tcred->gid))
248 		goto ok;
249 	if (ptrace_has_cap(tcred->user_ns, mode))
250 		goto ok;
251 	rcu_read_unlock();
252 	return -EPERM;
253 ok:
254 	rcu_read_unlock();
255 	smp_rmb();
256 	if (task->mm)
257 		dumpable = get_dumpable(task->mm);
258 	rcu_read_lock();
259 	if (!dumpable && !ptrace_has_cap(__task_cred(task)->user_ns, mode)) {
260 		rcu_read_unlock();
261 		return -EPERM;
262 	}
263 	rcu_read_unlock();
264 
265 	return security_ptrace_access_check(task, mode);
266 }
267 
268 bool ptrace_may_access(struct task_struct *task, unsigned int mode)
269 {
270 	int err;
271 	task_lock(task);
272 	err = __ptrace_may_access(task, mode);
273 	task_unlock(task);
274 	return !err;
275 }
276 
277 static int ptrace_attach(struct task_struct *task, long request,
278 			 unsigned long addr,
279 			 unsigned long flags)
280 {
281 	bool seize = (request == PTRACE_SEIZE);
282 	int retval;
283 
284 	retval = -EIO;
285 	if (seize) {
286 		if (addr != 0)
287 			goto out;
288 		if (flags & ~(unsigned long)PTRACE_O_MASK)
289 			goto out;
290 		flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
291 	} else {
292 		flags = PT_PTRACED;
293 	}
294 
295 	audit_ptrace(task);
296 
297 	retval = -EPERM;
298 	if (unlikely(task->flags & PF_KTHREAD))
299 		goto out;
300 	if (same_thread_group(task, current))
301 		goto out;
302 
303 	/*
304 	 * Protect exec's credential calculations against our interference;
305 	 * SUID, SGID and LSM creds get determined differently
306 	 * under ptrace.
307 	 */
308 	retval = -ERESTARTNOINTR;
309 	if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
310 		goto out;
311 
312 	task_lock(task);
313 	retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH);
314 	task_unlock(task);
315 	if (retval)
316 		goto unlock_creds;
317 
318 	write_lock_irq(&tasklist_lock);
319 	retval = -EPERM;
320 	if (unlikely(task->exit_state))
321 		goto unlock_tasklist;
322 	if (task->ptrace)
323 		goto unlock_tasklist;
324 
325 	if (seize)
326 		flags |= PT_SEIZED;
327 	rcu_read_lock();
328 	if (ns_capable(__task_cred(task)->user_ns, CAP_SYS_PTRACE))
329 		flags |= PT_PTRACE_CAP;
330 	rcu_read_unlock();
331 	task->ptrace = flags;
332 
333 	__ptrace_link(task, current);
334 
335 	/* SEIZE doesn't trap tracee on attach */
336 	if (!seize)
337 		send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
338 
339 	spin_lock(&task->sighand->siglock);
340 
341 	/*
342 	 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
343 	 * TRAPPING, and kick it so that it transits to TRACED.  TRAPPING
344 	 * will be cleared if the child completes the transition or any
345 	 * event which clears the group stop states happens.  We'll wait
346 	 * for the transition to complete before returning from this
347 	 * function.
348 	 *
349 	 * This hides STOPPED -> RUNNING -> TRACED transition from the
350 	 * attaching thread but a different thread in the same group can
351 	 * still observe the transient RUNNING state.  IOW, if another
352 	 * thread's WNOHANG wait(2) on the stopped tracee races against
353 	 * ATTACH, the wait(2) may fail due to the transient RUNNING.
354 	 *
355 	 * The following task_is_stopped() test is safe as both transitions
356 	 * in and out of STOPPED are protected by siglock.
357 	 */
358 	if (task_is_stopped(task) &&
359 	    task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
360 		signal_wake_up_state(task, __TASK_STOPPED);
361 
362 	spin_unlock(&task->sighand->siglock);
363 
364 	retval = 0;
365 unlock_tasklist:
366 	write_unlock_irq(&tasklist_lock);
367 unlock_creds:
368 	mutex_unlock(&task->signal->cred_guard_mutex);
369 out:
370 	if (!retval) {
371 		wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT,
372 			    ptrace_trapping_sleep_fn, TASK_UNINTERRUPTIBLE);
373 		proc_ptrace_connector(task, PTRACE_ATTACH);
374 	}
375 
376 	return retval;
377 }
378 
379 /**
380  * ptrace_traceme  --  helper for PTRACE_TRACEME
381  *
382  * Performs checks and sets PT_PTRACED.
383  * Should be used by all ptrace implementations for PTRACE_TRACEME.
384  */
385 static int ptrace_traceme(void)
386 {
387 	int ret = -EPERM;
388 
389 	write_lock_irq(&tasklist_lock);
390 	/* Are we already being traced? */
391 	if (!current->ptrace) {
392 		ret = security_ptrace_traceme(current->parent);
393 		/*
394 		 * Check PF_EXITING to ensure ->real_parent has not passed
395 		 * exit_ptrace(). Otherwise we don't report the error but
396 		 * pretend ->real_parent untraces us right after return.
397 		 */
398 		if (!ret && !(current->real_parent->flags & PF_EXITING)) {
399 			current->ptrace = PT_PTRACED;
400 			__ptrace_link(current, current->real_parent);
401 		}
402 	}
403 	write_unlock_irq(&tasklist_lock);
404 
405 	return ret;
406 }
407 
408 /*
409  * Called with irqs disabled, returns true if childs should reap themselves.
410  */
411 static int ignoring_children(struct sighand_struct *sigh)
412 {
413 	int ret;
414 	spin_lock(&sigh->siglock);
415 	ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
416 	      (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
417 	spin_unlock(&sigh->siglock);
418 	return ret;
419 }
420 
421 /*
422  * Called with tasklist_lock held for writing.
423  * Unlink a traced task, and clean it up if it was a traced zombie.
424  * Return true if it needs to be reaped with release_task().
425  * (We can't call release_task() here because we already hold tasklist_lock.)
426  *
427  * If it's a zombie, our attachedness prevented normal parent notification
428  * or self-reaping.  Do notification now if it would have happened earlier.
429  * If it should reap itself, return true.
430  *
431  * If it's our own child, there is no notification to do. But if our normal
432  * children self-reap, then this child was prevented by ptrace and we must
433  * reap it now, in that case we must also wake up sub-threads sleeping in
434  * do_wait().
435  */
436 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
437 {
438 	bool dead;
439 
440 	__ptrace_unlink(p);
441 
442 	if (p->exit_state != EXIT_ZOMBIE)
443 		return false;
444 
445 	dead = !thread_group_leader(p);
446 
447 	if (!dead && thread_group_empty(p)) {
448 		if (!same_thread_group(p->real_parent, tracer))
449 			dead = do_notify_parent(p, p->exit_signal);
450 		else if (ignoring_children(tracer->sighand)) {
451 			__wake_up_parent(p, tracer);
452 			dead = true;
453 		}
454 	}
455 	/* Mark it as in the process of being reaped. */
456 	if (dead)
457 		p->exit_state = EXIT_DEAD;
458 	return dead;
459 }
460 
461 static int ptrace_detach(struct task_struct *child, unsigned int data)
462 {
463 	bool dead = false;
464 
465 	if (!valid_signal(data))
466 		return -EIO;
467 
468 	/* Architecture-specific hardware disable .. */
469 	ptrace_disable(child);
470 	clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
471 
472 	write_lock_irq(&tasklist_lock);
473 	/*
474 	 * This child can be already killed. Make sure de_thread() or
475 	 * our sub-thread doing do_wait() didn't do release_task() yet.
476 	 */
477 	if (child->ptrace) {
478 		child->exit_code = data;
479 		dead = __ptrace_detach(current, child);
480 	}
481 	write_unlock_irq(&tasklist_lock);
482 
483 	proc_ptrace_connector(child, PTRACE_DETACH);
484 	if (unlikely(dead))
485 		release_task(child);
486 
487 	return 0;
488 }
489 
490 /*
491  * Detach all tasks we were using ptrace on. Called with tasklist held
492  * for writing, and returns with it held too. But note it can release
493  * and reacquire the lock.
494  */
495 void exit_ptrace(struct task_struct *tracer)
496 	__releases(&tasklist_lock)
497 	__acquires(&tasklist_lock)
498 {
499 	struct task_struct *p, *n;
500 	LIST_HEAD(ptrace_dead);
501 
502 	if (likely(list_empty(&tracer->ptraced)))
503 		return;
504 
505 	list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
506 		if (unlikely(p->ptrace & PT_EXITKILL))
507 			send_sig_info(SIGKILL, SEND_SIG_FORCED, p);
508 
509 		if (__ptrace_detach(tracer, p))
510 			list_add(&p->ptrace_entry, &ptrace_dead);
511 	}
512 
513 	write_unlock_irq(&tasklist_lock);
514 	BUG_ON(!list_empty(&tracer->ptraced));
515 
516 	list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_entry) {
517 		list_del_init(&p->ptrace_entry);
518 		release_task(p);
519 	}
520 
521 	write_lock_irq(&tasklist_lock);
522 }
523 
524 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
525 {
526 	int copied = 0;
527 
528 	while (len > 0) {
529 		char buf[128];
530 		int this_len, retval;
531 
532 		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
533 		retval = access_process_vm(tsk, src, buf, this_len, 0);
534 		if (!retval) {
535 			if (copied)
536 				break;
537 			return -EIO;
538 		}
539 		if (copy_to_user(dst, buf, retval))
540 			return -EFAULT;
541 		copied += retval;
542 		src += retval;
543 		dst += retval;
544 		len -= retval;
545 	}
546 	return copied;
547 }
548 
549 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
550 {
551 	int copied = 0;
552 
553 	while (len > 0) {
554 		char buf[128];
555 		int this_len, retval;
556 
557 		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
558 		if (copy_from_user(buf, src, this_len))
559 			return -EFAULT;
560 		retval = access_process_vm(tsk, dst, buf, this_len, 1);
561 		if (!retval) {
562 			if (copied)
563 				break;
564 			return -EIO;
565 		}
566 		copied += retval;
567 		src += retval;
568 		dst += retval;
569 		len -= retval;
570 	}
571 	return copied;
572 }
573 
574 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
575 {
576 	unsigned flags;
577 
578 	if (data & ~(unsigned long)PTRACE_O_MASK)
579 		return -EINVAL;
580 
581 	/* Avoid intermediate state when all opts are cleared */
582 	flags = child->ptrace;
583 	flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
584 	flags |= (data << PT_OPT_FLAG_SHIFT);
585 	child->ptrace = flags;
586 
587 	return 0;
588 }
589 
590 static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
591 {
592 	unsigned long flags;
593 	int error = -ESRCH;
594 
595 	if (lock_task_sighand(child, &flags)) {
596 		error = -EINVAL;
597 		if (likely(child->last_siginfo != NULL)) {
598 			*info = *child->last_siginfo;
599 			error = 0;
600 		}
601 		unlock_task_sighand(child, &flags);
602 	}
603 	return error;
604 }
605 
606 static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
607 {
608 	unsigned long flags;
609 	int error = -ESRCH;
610 
611 	if (lock_task_sighand(child, &flags)) {
612 		error = -EINVAL;
613 		if (likely(child->last_siginfo != NULL)) {
614 			*child->last_siginfo = *info;
615 			error = 0;
616 		}
617 		unlock_task_sighand(child, &flags);
618 	}
619 	return error;
620 }
621 
622 static int ptrace_peek_siginfo(struct task_struct *child,
623 				unsigned long addr,
624 				unsigned long data)
625 {
626 	struct ptrace_peeksiginfo_args arg;
627 	struct sigpending *pending;
628 	struct sigqueue *q;
629 	int ret, i;
630 
631 	ret = copy_from_user(&arg, (void __user *) addr,
632 				sizeof(struct ptrace_peeksiginfo_args));
633 	if (ret)
634 		return -EFAULT;
635 
636 	if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
637 		return -EINVAL; /* unknown flags */
638 
639 	if (arg.nr < 0)
640 		return -EINVAL;
641 
642 	if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
643 		pending = &child->signal->shared_pending;
644 	else
645 		pending = &child->pending;
646 
647 	for (i = 0; i < arg.nr; ) {
648 		siginfo_t info;
649 		s32 off = arg.off + i;
650 
651 		spin_lock_irq(&child->sighand->siglock);
652 		list_for_each_entry(q, &pending->list, list) {
653 			if (!off--) {
654 				copy_siginfo(&info, &q->info);
655 				break;
656 			}
657 		}
658 		spin_unlock_irq(&child->sighand->siglock);
659 
660 		if (off >= 0) /* beyond the end of the list */
661 			break;
662 
663 #ifdef CONFIG_COMPAT
664 		if (unlikely(is_compat_task())) {
665 			compat_siginfo_t __user *uinfo = compat_ptr(data);
666 
667 			ret = copy_siginfo_to_user32(uinfo, &info);
668 			ret |= __put_user(info.si_code, &uinfo->si_code);
669 		} else
670 #endif
671 		{
672 			siginfo_t __user *uinfo = (siginfo_t __user *) data;
673 
674 			ret = copy_siginfo_to_user(uinfo, &info);
675 			ret |= __put_user(info.si_code, &uinfo->si_code);
676 		}
677 
678 		if (ret) {
679 			ret = -EFAULT;
680 			break;
681 		}
682 
683 		data += sizeof(siginfo_t);
684 		i++;
685 
686 		if (signal_pending(current))
687 			break;
688 
689 		cond_resched();
690 	}
691 
692 	if (i > 0)
693 		return i;
694 
695 	return ret;
696 }
697 
698 #ifdef PTRACE_SINGLESTEP
699 #define is_singlestep(request)		((request) == PTRACE_SINGLESTEP)
700 #else
701 #define is_singlestep(request)		0
702 #endif
703 
704 #ifdef PTRACE_SINGLEBLOCK
705 #define is_singleblock(request)		((request) == PTRACE_SINGLEBLOCK)
706 #else
707 #define is_singleblock(request)		0
708 #endif
709 
710 #ifdef PTRACE_SYSEMU
711 #define is_sysemu_singlestep(request)	((request) == PTRACE_SYSEMU_SINGLESTEP)
712 #else
713 #define is_sysemu_singlestep(request)	0
714 #endif
715 
716 static int ptrace_resume(struct task_struct *child, long request,
717 			 unsigned long data)
718 {
719 	if (!valid_signal(data))
720 		return -EIO;
721 
722 	if (request == PTRACE_SYSCALL)
723 		set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
724 	else
725 		clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
726 
727 #ifdef TIF_SYSCALL_EMU
728 	if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
729 		set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
730 	else
731 		clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
732 #endif
733 
734 	if (is_singleblock(request)) {
735 		if (unlikely(!arch_has_block_step()))
736 			return -EIO;
737 		user_enable_block_step(child);
738 	} else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
739 		if (unlikely(!arch_has_single_step()))
740 			return -EIO;
741 		user_enable_single_step(child);
742 	} else {
743 		user_disable_single_step(child);
744 	}
745 
746 	child->exit_code = data;
747 	wake_up_state(child, __TASK_TRACED);
748 
749 	return 0;
750 }
751 
752 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
753 
754 static const struct user_regset *
755 find_regset(const struct user_regset_view *view, unsigned int type)
756 {
757 	const struct user_regset *regset;
758 	int n;
759 
760 	for (n = 0; n < view->n; ++n) {
761 		regset = view->regsets + n;
762 		if (regset->core_note_type == type)
763 			return regset;
764 	}
765 
766 	return NULL;
767 }
768 
769 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
770 			 struct iovec *kiov)
771 {
772 	const struct user_regset_view *view = task_user_regset_view(task);
773 	const struct user_regset *regset = find_regset(view, type);
774 	int regset_no;
775 
776 	if (!regset || (kiov->iov_len % regset->size) != 0)
777 		return -EINVAL;
778 
779 	regset_no = regset - view->regsets;
780 	kiov->iov_len = min(kiov->iov_len,
781 			    (__kernel_size_t) (regset->n * regset->size));
782 
783 	if (req == PTRACE_GETREGSET)
784 		return copy_regset_to_user(task, view, regset_no, 0,
785 					   kiov->iov_len, kiov->iov_base);
786 	else
787 		return copy_regset_from_user(task, view, regset_no, 0,
788 					     kiov->iov_len, kiov->iov_base);
789 }
790 
791 /*
792  * This is declared in linux/regset.h and defined in machine-dependent
793  * code.  We put the export here, near the primary machine-neutral use,
794  * to ensure no machine forgets it.
795  */
796 EXPORT_SYMBOL_GPL(task_user_regset_view);
797 #endif
798 
799 int ptrace_request(struct task_struct *child, long request,
800 		   unsigned long addr, unsigned long data)
801 {
802 	bool seized = child->ptrace & PT_SEIZED;
803 	int ret = -EIO;
804 	siginfo_t siginfo, *si;
805 	void __user *datavp = (void __user *) data;
806 	unsigned long __user *datalp = datavp;
807 	unsigned long flags;
808 
809 	switch (request) {
810 	case PTRACE_PEEKTEXT:
811 	case PTRACE_PEEKDATA:
812 		return generic_ptrace_peekdata(child, addr, data);
813 	case PTRACE_POKETEXT:
814 	case PTRACE_POKEDATA:
815 		return generic_ptrace_pokedata(child, addr, data);
816 
817 #ifdef PTRACE_OLDSETOPTIONS
818 	case PTRACE_OLDSETOPTIONS:
819 #endif
820 	case PTRACE_SETOPTIONS:
821 		ret = ptrace_setoptions(child, data);
822 		break;
823 	case PTRACE_GETEVENTMSG:
824 		ret = put_user(child->ptrace_message, datalp);
825 		break;
826 
827 	case PTRACE_PEEKSIGINFO:
828 		ret = ptrace_peek_siginfo(child, addr, data);
829 		break;
830 
831 	case PTRACE_GETSIGINFO:
832 		ret = ptrace_getsiginfo(child, &siginfo);
833 		if (!ret)
834 			ret = copy_siginfo_to_user(datavp, &siginfo);
835 		break;
836 
837 	case PTRACE_SETSIGINFO:
838 		if (copy_from_user(&siginfo, datavp, sizeof siginfo))
839 			ret = -EFAULT;
840 		else
841 			ret = ptrace_setsiginfo(child, &siginfo);
842 		break;
843 
844 	case PTRACE_INTERRUPT:
845 		/*
846 		 * Stop tracee without any side-effect on signal or job
847 		 * control.  At least one trap is guaranteed to happen
848 		 * after this request.  If @child is already trapped, the
849 		 * current trap is not disturbed and another trap will
850 		 * happen after the current trap is ended with PTRACE_CONT.
851 		 *
852 		 * The actual trap might not be PTRACE_EVENT_STOP trap but
853 		 * the pending condition is cleared regardless.
854 		 */
855 		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
856 			break;
857 
858 		/*
859 		 * INTERRUPT doesn't disturb existing trap sans one
860 		 * exception.  If ptracer issued LISTEN for the current
861 		 * STOP, this INTERRUPT should clear LISTEN and re-trap
862 		 * tracee into STOP.
863 		 */
864 		if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
865 			ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
866 
867 		unlock_task_sighand(child, &flags);
868 		ret = 0;
869 		break;
870 
871 	case PTRACE_LISTEN:
872 		/*
873 		 * Listen for events.  Tracee must be in STOP.  It's not
874 		 * resumed per-se but is not considered to be in TRACED by
875 		 * wait(2) or ptrace(2).  If an async event (e.g. group
876 		 * stop state change) happens, tracee will enter STOP trap
877 		 * again.  Alternatively, ptracer can issue INTERRUPT to
878 		 * finish listening and re-trap tracee into STOP.
879 		 */
880 		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
881 			break;
882 
883 		si = child->last_siginfo;
884 		if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
885 			child->jobctl |= JOBCTL_LISTENING;
886 			/*
887 			 * If NOTIFY is set, it means event happened between
888 			 * start of this trap and now.  Trigger re-trap.
889 			 */
890 			if (child->jobctl & JOBCTL_TRAP_NOTIFY)
891 				ptrace_signal_wake_up(child, true);
892 			ret = 0;
893 		}
894 		unlock_task_sighand(child, &flags);
895 		break;
896 
897 	case PTRACE_DETACH:	 /* detach a process that was attached. */
898 		ret = ptrace_detach(child, data);
899 		break;
900 
901 #ifdef CONFIG_BINFMT_ELF_FDPIC
902 	case PTRACE_GETFDPIC: {
903 		struct mm_struct *mm = get_task_mm(child);
904 		unsigned long tmp = 0;
905 
906 		ret = -ESRCH;
907 		if (!mm)
908 			break;
909 
910 		switch (addr) {
911 		case PTRACE_GETFDPIC_EXEC:
912 			tmp = mm->context.exec_fdpic_loadmap;
913 			break;
914 		case PTRACE_GETFDPIC_INTERP:
915 			tmp = mm->context.interp_fdpic_loadmap;
916 			break;
917 		default:
918 			break;
919 		}
920 		mmput(mm);
921 
922 		ret = put_user(tmp, datalp);
923 		break;
924 	}
925 #endif
926 
927 #ifdef PTRACE_SINGLESTEP
928 	case PTRACE_SINGLESTEP:
929 #endif
930 #ifdef PTRACE_SINGLEBLOCK
931 	case PTRACE_SINGLEBLOCK:
932 #endif
933 #ifdef PTRACE_SYSEMU
934 	case PTRACE_SYSEMU:
935 	case PTRACE_SYSEMU_SINGLESTEP:
936 #endif
937 	case PTRACE_SYSCALL:
938 	case PTRACE_CONT:
939 		return ptrace_resume(child, request, data);
940 
941 	case PTRACE_KILL:
942 		if (child->exit_state)	/* already dead */
943 			return 0;
944 		return ptrace_resume(child, request, SIGKILL);
945 
946 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
947 	case PTRACE_GETREGSET:
948 	case PTRACE_SETREGSET:
949 	{
950 		struct iovec kiov;
951 		struct iovec __user *uiov = datavp;
952 
953 		if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
954 			return -EFAULT;
955 
956 		if (__get_user(kiov.iov_base, &uiov->iov_base) ||
957 		    __get_user(kiov.iov_len, &uiov->iov_len))
958 			return -EFAULT;
959 
960 		ret = ptrace_regset(child, request, addr, &kiov);
961 		if (!ret)
962 			ret = __put_user(kiov.iov_len, &uiov->iov_len);
963 		break;
964 	}
965 #endif
966 	default:
967 		break;
968 	}
969 
970 	return ret;
971 }
972 
973 static struct task_struct *ptrace_get_task_struct(pid_t pid)
974 {
975 	struct task_struct *child;
976 
977 	rcu_read_lock();
978 	child = find_task_by_vpid(pid);
979 	if (child)
980 		get_task_struct(child);
981 	rcu_read_unlock();
982 
983 	if (!child)
984 		return ERR_PTR(-ESRCH);
985 	return child;
986 }
987 
988 #ifndef arch_ptrace_attach
989 #define arch_ptrace_attach(child)	do { } while (0)
990 #endif
991 
992 SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
993 		unsigned long, data)
994 {
995 	struct task_struct *child;
996 	long ret;
997 
998 	if (request == PTRACE_TRACEME) {
999 		ret = ptrace_traceme();
1000 		if (!ret)
1001 			arch_ptrace_attach(current);
1002 		goto out;
1003 	}
1004 
1005 	child = ptrace_get_task_struct(pid);
1006 	if (IS_ERR(child)) {
1007 		ret = PTR_ERR(child);
1008 		goto out;
1009 	}
1010 
1011 	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1012 		ret = ptrace_attach(child, request, addr, data);
1013 		/*
1014 		 * Some architectures need to do book-keeping after
1015 		 * a ptrace attach.
1016 		 */
1017 		if (!ret)
1018 			arch_ptrace_attach(child);
1019 		goto out_put_task_struct;
1020 	}
1021 
1022 	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1023 				  request == PTRACE_INTERRUPT);
1024 	if (ret < 0)
1025 		goto out_put_task_struct;
1026 
1027 	ret = arch_ptrace(child, request, addr, data);
1028 	if (ret || request != PTRACE_DETACH)
1029 		ptrace_unfreeze_traced(child);
1030 
1031  out_put_task_struct:
1032 	put_task_struct(child);
1033  out:
1034 	return ret;
1035 }
1036 
1037 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1038 			    unsigned long data)
1039 {
1040 	unsigned long tmp;
1041 	int copied;
1042 
1043 	copied = access_process_vm(tsk, addr, &tmp, sizeof(tmp), 0);
1044 	if (copied != sizeof(tmp))
1045 		return -EIO;
1046 	return put_user(tmp, (unsigned long __user *)data);
1047 }
1048 
1049 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1050 			    unsigned long data)
1051 {
1052 	int copied;
1053 
1054 	copied = access_process_vm(tsk, addr, &data, sizeof(data), 1);
1055 	return (copied == sizeof(data)) ? 0 : -EIO;
1056 }
1057 
1058 #if defined CONFIG_COMPAT
1059 #include <linux/compat.h>
1060 
1061 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1062 			  compat_ulong_t addr, compat_ulong_t data)
1063 {
1064 	compat_ulong_t __user *datap = compat_ptr(data);
1065 	compat_ulong_t word;
1066 	siginfo_t siginfo;
1067 	int ret;
1068 
1069 	switch (request) {
1070 	case PTRACE_PEEKTEXT:
1071 	case PTRACE_PEEKDATA:
1072 		ret = access_process_vm(child, addr, &word, sizeof(word), 0);
1073 		if (ret != sizeof(word))
1074 			ret = -EIO;
1075 		else
1076 			ret = put_user(word, datap);
1077 		break;
1078 
1079 	case PTRACE_POKETEXT:
1080 	case PTRACE_POKEDATA:
1081 		ret = access_process_vm(child, addr, &data, sizeof(data), 1);
1082 		ret = (ret != sizeof(data) ? -EIO : 0);
1083 		break;
1084 
1085 	case PTRACE_GETEVENTMSG:
1086 		ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1087 		break;
1088 
1089 	case PTRACE_GETSIGINFO:
1090 		ret = ptrace_getsiginfo(child, &siginfo);
1091 		if (!ret)
1092 			ret = copy_siginfo_to_user32(
1093 				(struct compat_siginfo __user *) datap,
1094 				&siginfo);
1095 		break;
1096 
1097 	case PTRACE_SETSIGINFO:
1098 		memset(&siginfo, 0, sizeof siginfo);
1099 		if (copy_siginfo_from_user32(
1100 			    &siginfo, (struct compat_siginfo __user *) datap))
1101 			ret = -EFAULT;
1102 		else
1103 			ret = ptrace_setsiginfo(child, &siginfo);
1104 		break;
1105 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1106 	case PTRACE_GETREGSET:
1107 	case PTRACE_SETREGSET:
1108 	{
1109 		struct iovec kiov;
1110 		struct compat_iovec __user *uiov =
1111 			(struct compat_iovec __user *) datap;
1112 		compat_uptr_t ptr;
1113 		compat_size_t len;
1114 
1115 		if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1116 			return -EFAULT;
1117 
1118 		if (__get_user(ptr, &uiov->iov_base) ||
1119 		    __get_user(len, &uiov->iov_len))
1120 			return -EFAULT;
1121 
1122 		kiov.iov_base = compat_ptr(ptr);
1123 		kiov.iov_len = len;
1124 
1125 		ret = ptrace_regset(child, request, addr, &kiov);
1126 		if (!ret)
1127 			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1128 		break;
1129 	}
1130 #endif
1131 
1132 	default:
1133 		ret = ptrace_request(child, request, addr, data);
1134 	}
1135 
1136 	return ret;
1137 }
1138 
1139 asmlinkage long compat_sys_ptrace(compat_long_t request, compat_long_t pid,
1140 				  compat_long_t addr, compat_long_t data)
1141 {
1142 	struct task_struct *child;
1143 	long ret;
1144 
1145 	if (request == PTRACE_TRACEME) {
1146 		ret = ptrace_traceme();
1147 		goto out;
1148 	}
1149 
1150 	child = ptrace_get_task_struct(pid);
1151 	if (IS_ERR(child)) {
1152 		ret = PTR_ERR(child);
1153 		goto out;
1154 	}
1155 
1156 	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1157 		ret = ptrace_attach(child, request, addr, data);
1158 		/*
1159 		 * Some architectures need to do book-keeping after
1160 		 * a ptrace attach.
1161 		 */
1162 		if (!ret)
1163 			arch_ptrace_attach(child);
1164 		goto out_put_task_struct;
1165 	}
1166 
1167 	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1168 				  request == PTRACE_INTERRUPT);
1169 	if (!ret) {
1170 		ret = compat_arch_ptrace(child, request, addr, data);
1171 		if (ret || request != PTRACE_DETACH)
1172 			ptrace_unfreeze_traced(child);
1173 	}
1174 
1175  out_put_task_struct:
1176 	put_task_struct(child);
1177  out:
1178 	return ret;
1179 }
1180 #endif	/* CONFIG_COMPAT */
1181 
1182 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1183 int ptrace_get_breakpoints(struct task_struct *tsk)
1184 {
1185 	if (atomic_inc_not_zero(&tsk->ptrace_bp_refcnt))
1186 		return 0;
1187 
1188 	return -1;
1189 }
1190 
1191 void ptrace_put_breakpoints(struct task_struct *tsk)
1192 {
1193 	if (atomic_dec_and_test(&tsk->ptrace_bp_refcnt))
1194 		flush_ptrace_hw_breakpoint(tsk);
1195 }
1196 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1197