xref: /linux/kernel/ptrace.c (revision 52990390f91c1c39ca742fc8f390b29891d95127)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * linux/kernel/ptrace.c
4  *
5  * (C) Copyright 1999 Linus Torvalds
6  *
7  * Common interfaces for "ptrace()" which we do not want
8  * to continually duplicate across every architecture.
9  */
10 
11 #include <linux/capability.h>
12 #include <linux/export.h>
13 #include <linux/sched.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/coredump.h>
16 #include <linux/sched/task.h>
17 #include <linux/errno.h>
18 #include <linux/mm.h>
19 #include <linux/highmem.h>
20 #include <linux/pagemap.h>
21 #include <linux/ptrace.h>
22 #include <linux/security.h>
23 #include <linux/signal.h>
24 #include <linux/uio.h>
25 #include <linux/audit.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/syscalls.h>
28 #include <linux/uaccess.h>
29 #include <linux/regset.h>
30 #include <linux/hw_breakpoint.h>
31 #include <linux/cn_proc.h>
32 #include <linux/compat.h>
33 #include <linux/sched/signal.h>
34 #include <linux/minmax.h>
35 #include <linux/syscall_user_dispatch.h>
36 
37 #include <asm/syscall.h>	/* for syscall_get_* */
38 
39 /*
40  * Access another process' address space via ptrace.
41  * Source/target buffer must be kernel space,
42  * Do not walk the page table directly, use get_user_pages
43  */
44 int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
45 		     void *buf, int len, unsigned int gup_flags)
46 {
47 	struct mm_struct *mm;
48 	int ret;
49 
50 	mm = get_task_mm(tsk);
51 	if (!mm)
52 		return 0;
53 
54 	if (!tsk->ptrace ||
55 	    (current != tsk->parent) ||
56 	    ((get_dumpable(mm) != SUID_DUMP_USER) &&
57 	     !ptracer_capable(tsk, mm->user_ns))) {
58 		mmput(mm);
59 		return 0;
60 	}
61 
62 	ret = __access_remote_vm(mm, addr, buf, len, gup_flags);
63 	mmput(mm);
64 
65 	return ret;
66 }
67 
68 
69 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent,
70 		   const struct cred *ptracer_cred)
71 {
72 	BUG_ON(!list_empty(&child->ptrace_entry));
73 	list_add(&child->ptrace_entry, &new_parent->ptraced);
74 	child->parent = new_parent;
75 	child->ptracer_cred = get_cred(ptracer_cred);
76 }
77 
78 /*
79  * ptrace a task: make the debugger its new parent and
80  * move it to the ptrace list.
81  *
82  * Must be called with the tasklist lock write-held.
83  */
84 static void ptrace_link(struct task_struct *child, struct task_struct *new_parent)
85 {
86 	__ptrace_link(child, new_parent, current_cred());
87 }
88 
89 /**
90  * __ptrace_unlink - unlink ptracee and restore its execution state
91  * @child: ptracee to be unlinked
92  *
93  * Remove @child from the ptrace list, move it back to the original parent,
94  * and restore the execution state so that it conforms to the group stop
95  * state.
96  *
97  * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
98  * exiting.  For PTRACE_DETACH, unless the ptracee has been killed between
99  * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
100  * If the ptracer is exiting, the ptracee can be in any state.
101  *
102  * After detach, the ptracee should be in a state which conforms to the
103  * group stop.  If the group is stopped or in the process of stopping, the
104  * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
105  * up from TASK_TRACED.
106  *
107  * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
108  * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
109  * to but in the opposite direction of what happens while attaching to a
110  * stopped task.  However, in this direction, the intermediate RUNNING
111  * state is not hidden even from the current ptracer and if it immediately
112  * re-attaches and performs a WNOHANG wait(2), it may fail.
113  *
114  * CONTEXT:
115  * write_lock_irq(tasklist_lock)
116  */
117 void __ptrace_unlink(struct task_struct *child)
118 {
119 	const struct cred *old_cred;
120 	BUG_ON(!child->ptrace);
121 
122 	clear_task_syscall_work(child, SYSCALL_TRACE);
123 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
124 	clear_task_syscall_work(child, SYSCALL_EMU);
125 #endif
126 
127 	child->parent = child->real_parent;
128 	list_del_init(&child->ptrace_entry);
129 	old_cred = child->ptracer_cred;
130 	child->ptracer_cred = NULL;
131 	put_cred(old_cred);
132 
133 	spin_lock(&child->sighand->siglock);
134 	child->ptrace = 0;
135 	/*
136 	 * Clear all pending traps and TRAPPING.  TRAPPING should be
137 	 * cleared regardless of JOBCTL_STOP_PENDING.  Do it explicitly.
138 	 */
139 	task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
140 	task_clear_jobctl_trapping(child);
141 
142 	/*
143 	 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
144 	 * @child isn't dead.
145 	 */
146 	if (!(child->flags & PF_EXITING) &&
147 	    (child->signal->flags & SIGNAL_STOP_STOPPED ||
148 	     child->signal->group_stop_count)) {
149 		child->jobctl |= JOBCTL_STOP_PENDING;
150 
151 		/*
152 		 * This is only possible if this thread was cloned by the
153 		 * traced task running in the stopped group, set the signal
154 		 * for the future reports.
155 		 * FIXME: we should change ptrace_init_task() to handle this
156 		 * case.
157 		 */
158 		if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
159 			child->jobctl |= SIGSTOP;
160 	}
161 
162 	/*
163 	 * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
164 	 * @child in the butt.  Note that @resume should be used iff @child
165 	 * is in TASK_TRACED; otherwise, we might unduly disrupt
166 	 * TASK_KILLABLE sleeps.
167 	 */
168 	if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
169 		ptrace_signal_wake_up(child, true);
170 
171 	spin_unlock(&child->sighand->siglock);
172 }
173 
174 static bool looks_like_a_spurious_pid(struct task_struct *task)
175 {
176 	if (task->exit_code != ((PTRACE_EVENT_EXEC << 8) | SIGTRAP))
177 		return false;
178 
179 	if (task_pid_vnr(task) == task->ptrace_message)
180 		return false;
181 	/*
182 	 * The tracee changed its pid but the PTRACE_EVENT_EXEC event
183 	 * was not wait()'ed, most probably debugger targets the old
184 	 * leader which was destroyed in de_thread().
185 	 */
186 	return true;
187 }
188 
189 /*
190  * Ensure that nothing can wake it up, even SIGKILL
191  *
192  * A task is switched to this state while a ptrace operation is in progress;
193  * such that the ptrace operation is uninterruptible.
194  */
195 static bool ptrace_freeze_traced(struct task_struct *task)
196 {
197 	bool ret = false;
198 
199 	/* Lockless, nobody but us can set this flag */
200 	if (task->jobctl & JOBCTL_LISTENING)
201 		return ret;
202 
203 	spin_lock_irq(&task->sighand->siglock);
204 	if (task_is_traced(task) && !looks_like_a_spurious_pid(task) &&
205 	    !__fatal_signal_pending(task)) {
206 		task->jobctl |= JOBCTL_PTRACE_FROZEN;
207 		ret = true;
208 	}
209 	spin_unlock_irq(&task->sighand->siglock);
210 
211 	return ret;
212 }
213 
214 static void ptrace_unfreeze_traced(struct task_struct *task)
215 {
216 	unsigned long flags;
217 
218 	/*
219 	 * The child may be awake and may have cleared
220 	 * JOBCTL_PTRACE_FROZEN (see ptrace_resume).  The child will
221 	 * not set JOBCTL_PTRACE_FROZEN or enter __TASK_TRACED anew.
222 	 */
223 	if (lock_task_sighand(task, &flags)) {
224 		task->jobctl &= ~JOBCTL_PTRACE_FROZEN;
225 		if (__fatal_signal_pending(task)) {
226 			task->jobctl &= ~JOBCTL_TRACED;
227 			wake_up_state(task, __TASK_TRACED);
228 		}
229 		unlock_task_sighand(task, &flags);
230 	}
231 }
232 
233 /**
234  * ptrace_check_attach - check whether ptracee is ready for ptrace operation
235  * @child: ptracee to check for
236  * @ignore_state: don't check whether @child is currently %TASK_TRACED
237  *
238  * Check whether @child is being ptraced by %current and ready for further
239  * ptrace operations.  If @ignore_state is %false, @child also should be in
240  * %TASK_TRACED state and on return the child is guaranteed to be traced
241  * and not executing.  If @ignore_state is %true, @child can be in any
242  * state.
243  *
244  * CONTEXT:
245  * Grabs and releases tasklist_lock and @child->sighand->siglock.
246  *
247  * RETURNS:
248  * 0 on success, -ESRCH if %child is not ready.
249  */
250 static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
251 {
252 	int ret = -ESRCH;
253 
254 	/*
255 	 * We take the read lock around doing both checks to close a
256 	 * possible race where someone else was tracing our child and
257 	 * detached between these two checks.  After this locked check,
258 	 * we are sure that this is our traced child and that can only
259 	 * be changed by us so it's not changing right after this.
260 	 */
261 	read_lock(&tasklist_lock);
262 	if (child->ptrace && child->parent == current) {
263 		/*
264 		 * child->sighand can't be NULL, release_task()
265 		 * does ptrace_unlink() before __exit_signal().
266 		 */
267 		if (ignore_state || ptrace_freeze_traced(child))
268 			ret = 0;
269 	}
270 	read_unlock(&tasklist_lock);
271 
272 	if (!ret && !ignore_state &&
273 	    WARN_ON_ONCE(!wait_task_inactive(child, __TASK_TRACED|TASK_FROZEN)))
274 		ret = -ESRCH;
275 
276 	return ret;
277 }
278 
279 static bool ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
280 {
281 	if (mode & PTRACE_MODE_NOAUDIT)
282 		return ns_capable_noaudit(ns, CAP_SYS_PTRACE);
283 	return ns_capable(ns, CAP_SYS_PTRACE);
284 }
285 
286 /* Returns 0 on success, -errno on denial. */
287 static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
288 {
289 	const struct cred *cred = current_cred(), *tcred;
290 	struct mm_struct *mm;
291 	kuid_t caller_uid;
292 	kgid_t caller_gid;
293 
294 	if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
295 		WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
296 		return -EPERM;
297 	}
298 
299 	/* May we inspect the given task?
300 	 * This check is used both for attaching with ptrace
301 	 * and for allowing access to sensitive information in /proc.
302 	 *
303 	 * ptrace_attach denies several cases that /proc allows
304 	 * because setting up the necessary parent/child relationship
305 	 * or halting the specified task is impossible.
306 	 */
307 
308 	/* Don't let security modules deny introspection */
309 	if (same_thread_group(task, current))
310 		return 0;
311 	rcu_read_lock();
312 	if (mode & PTRACE_MODE_FSCREDS) {
313 		caller_uid = cred->fsuid;
314 		caller_gid = cred->fsgid;
315 	} else {
316 		/*
317 		 * Using the euid would make more sense here, but something
318 		 * in userland might rely on the old behavior, and this
319 		 * shouldn't be a security problem since
320 		 * PTRACE_MODE_REALCREDS implies that the caller explicitly
321 		 * used a syscall that requests access to another process
322 		 * (and not a filesystem syscall to procfs).
323 		 */
324 		caller_uid = cred->uid;
325 		caller_gid = cred->gid;
326 	}
327 	tcred = __task_cred(task);
328 	if (uid_eq(caller_uid, tcred->euid) &&
329 	    uid_eq(caller_uid, tcred->suid) &&
330 	    uid_eq(caller_uid, tcred->uid)  &&
331 	    gid_eq(caller_gid, tcred->egid) &&
332 	    gid_eq(caller_gid, tcred->sgid) &&
333 	    gid_eq(caller_gid, tcred->gid))
334 		goto ok;
335 	if (ptrace_has_cap(tcred->user_ns, mode))
336 		goto ok;
337 	rcu_read_unlock();
338 	return -EPERM;
339 ok:
340 	rcu_read_unlock();
341 	/*
342 	 * If a task drops privileges and becomes nondumpable (through a syscall
343 	 * like setresuid()) while we are trying to access it, we must ensure
344 	 * that the dumpability is read after the credentials; otherwise,
345 	 * we may be able to attach to a task that we shouldn't be able to
346 	 * attach to (as if the task had dropped privileges without becoming
347 	 * nondumpable).
348 	 * Pairs with a write barrier in commit_creds().
349 	 */
350 	smp_rmb();
351 	mm = task->mm;
352 	if (mm &&
353 	    ((get_dumpable(mm) != SUID_DUMP_USER) &&
354 	     !ptrace_has_cap(mm->user_ns, mode)))
355 	    return -EPERM;
356 
357 	return security_ptrace_access_check(task, mode);
358 }
359 
360 bool ptrace_may_access(struct task_struct *task, unsigned int mode)
361 {
362 	int err;
363 	task_lock(task);
364 	err = __ptrace_may_access(task, mode);
365 	task_unlock(task);
366 	return !err;
367 }
368 
369 static int check_ptrace_options(unsigned long data)
370 {
371 	if (data & ~(unsigned long)PTRACE_O_MASK)
372 		return -EINVAL;
373 
374 	if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
375 		if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
376 		    !IS_ENABLED(CONFIG_SECCOMP))
377 			return -EINVAL;
378 
379 		if (!capable(CAP_SYS_ADMIN))
380 			return -EPERM;
381 
382 		if (seccomp_mode(&current->seccomp) != SECCOMP_MODE_DISABLED ||
383 		    current->ptrace & PT_SUSPEND_SECCOMP)
384 			return -EPERM;
385 	}
386 	return 0;
387 }
388 
389 static int ptrace_attach(struct task_struct *task, long request,
390 			 unsigned long addr,
391 			 unsigned long flags)
392 {
393 	bool seize = (request == PTRACE_SEIZE);
394 	int retval;
395 
396 	retval = -EIO;
397 	if (seize) {
398 		if (addr != 0)
399 			goto out;
400 		/*
401 		 * This duplicates the check in check_ptrace_options() because
402 		 * ptrace_attach() and ptrace_setoptions() have historically
403 		 * used different error codes for unknown ptrace options.
404 		 */
405 		if (flags & ~(unsigned long)PTRACE_O_MASK)
406 			goto out;
407 		retval = check_ptrace_options(flags);
408 		if (retval)
409 			return retval;
410 		flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
411 	} else {
412 		flags = PT_PTRACED;
413 	}
414 
415 	audit_ptrace(task);
416 
417 	retval = -EPERM;
418 	if (unlikely(task->flags & PF_KTHREAD))
419 		goto out;
420 	if (same_thread_group(task, current))
421 		goto out;
422 
423 	/*
424 	 * Protect exec's credential calculations against our interference;
425 	 * SUID, SGID and LSM creds get determined differently
426 	 * under ptrace.
427 	 */
428 	retval = -ERESTARTNOINTR;
429 	if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
430 		goto out;
431 
432 	task_lock(task);
433 	retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
434 	task_unlock(task);
435 	if (retval)
436 		goto unlock_creds;
437 
438 	write_lock_irq(&tasklist_lock);
439 	retval = -EPERM;
440 	if (unlikely(task->exit_state))
441 		goto unlock_tasklist;
442 	if (task->ptrace)
443 		goto unlock_tasklist;
444 
445 	task->ptrace = flags;
446 
447 	ptrace_link(task, current);
448 
449 	/* SEIZE doesn't trap tracee on attach */
450 	if (!seize)
451 		send_sig_info(SIGSTOP, SEND_SIG_PRIV, task);
452 
453 	spin_lock(&task->sighand->siglock);
454 
455 	/*
456 	 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
457 	 * TRAPPING, and kick it so that it transits to TRACED.  TRAPPING
458 	 * will be cleared if the child completes the transition or any
459 	 * event which clears the group stop states happens.  We'll wait
460 	 * for the transition to complete before returning from this
461 	 * function.
462 	 *
463 	 * This hides STOPPED -> RUNNING -> TRACED transition from the
464 	 * attaching thread but a different thread in the same group can
465 	 * still observe the transient RUNNING state.  IOW, if another
466 	 * thread's WNOHANG wait(2) on the stopped tracee races against
467 	 * ATTACH, the wait(2) may fail due to the transient RUNNING.
468 	 *
469 	 * The following task_is_stopped() test is safe as both transitions
470 	 * in and out of STOPPED are protected by siglock.
471 	 */
472 	if (task_is_stopped(task) &&
473 	    task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING)) {
474 		task->jobctl &= ~JOBCTL_STOPPED;
475 		signal_wake_up_state(task, __TASK_STOPPED);
476 	}
477 
478 	spin_unlock(&task->sighand->siglock);
479 
480 	retval = 0;
481 unlock_tasklist:
482 	write_unlock_irq(&tasklist_lock);
483 unlock_creds:
484 	mutex_unlock(&task->signal->cred_guard_mutex);
485 out:
486 	if (!retval) {
487 		/*
488 		 * We do not bother to change retval or clear JOBCTL_TRAPPING
489 		 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
490 		 * not return to user-mode, it will exit and clear this bit in
491 		 * __ptrace_unlink() if it wasn't already cleared by the tracee;
492 		 * and until then nobody can ptrace this task.
493 		 */
494 		wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
495 		proc_ptrace_connector(task, PTRACE_ATTACH);
496 	}
497 
498 	return retval;
499 }
500 
501 /**
502  * ptrace_traceme  --  helper for PTRACE_TRACEME
503  *
504  * Performs checks and sets PT_PTRACED.
505  * Should be used by all ptrace implementations for PTRACE_TRACEME.
506  */
507 static int ptrace_traceme(void)
508 {
509 	int ret = -EPERM;
510 
511 	write_lock_irq(&tasklist_lock);
512 	/* Are we already being traced? */
513 	if (!current->ptrace) {
514 		ret = security_ptrace_traceme(current->parent);
515 		/*
516 		 * Check PF_EXITING to ensure ->real_parent has not passed
517 		 * exit_ptrace(). Otherwise we don't report the error but
518 		 * pretend ->real_parent untraces us right after return.
519 		 */
520 		if (!ret && !(current->real_parent->flags & PF_EXITING)) {
521 			current->ptrace = PT_PTRACED;
522 			ptrace_link(current, current->real_parent);
523 		}
524 	}
525 	write_unlock_irq(&tasklist_lock);
526 
527 	return ret;
528 }
529 
530 /*
531  * Called with irqs disabled, returns true if childs should reap themselves.
532  */
533 static int ignoring_children(struct sighand_struct *sigh)
534 {
535 	int ret;
536 	spin_lock(&sigh->siglock);
537 	ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
538 	      (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
539 	spin_unlock(&sigh->siglock);
540 	return ret;
541 }
542 
543 /*
544  * Called with tasklist_lock held for writing.
545  * Unlink a traced task, and clean it up if it was a traced zombie.
546  * Return true if it needs to be reaped with release_task().
547  * (We can't call release_task() here because we already hold tasklist_lock.)
548  *
549  * If it's a zombie, our attachedness prevented normal parent notification
550  * or self-reaping.  Do notification now if it would have happened earlier.
551  * If it should reap itself, return true.
552  *
553  * If it's our own child, there is no notification to do. But if our normal
554  * children self-reap, then this child was prevented by ptrace and we must
555  * reap it now, in that case we must also wake up sub-threads sleeping in
556  * do_wait().
557  */
558 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
559 {
560 	bool dead;
561 
562 	__ptrace_unlink(p);
563 
564 	if (p->exit_state != EXIT_ZOMBIE)
565 		return false;
566 
567 	dead = !thread_group_leader(p);
568 
569 	if (!dead && thread_group_empty(p)) {
570 		if (!same_thread_group(p->real_parent, tracer))
571 			dead = do_notify_parent(p, p->exit_signal);
572 		else if (ignoring_children(tracer->sighand)) {
573 			__wake_up_parent(p, tracer);
574 			dead = true;
575 		}
576 	}
577 	/* Mark it as in the process of being reaped. */
578 	if (dead)
579 		p->exit_state = EXIT_DEAD;
580 	return dead;
581 }
582 
583 static int ptrace_detach(struct task_struct *child, unsigned int data)
584 {
585 	if (!valid_signal(data))
586 		return -EIO;
587 
588 	/* Architecture-specific hardware disable .. */
589 	ptrace_disable(child);
590 
591 	write_lock_irq(&tasklist_lock);
592 	/*
593 	 * We rely on ptrace_freeze_traced(). It can't be killed and
594 	 * untraced by another thread, it can't be a zombie.
595 	 */
596 	WARN_ON(!child->ptrace || child->exit_state);
597 	/*
598 	 * tasklist_lock avoids the race with wait_task_stopped(), see
599 	 * the comment in ptrace_resume().
600 	 */
601 	child->exit_code = data;
602 	__ptrace_detach(current, child);
603 	write_unlock_irq(&tasklist_lock);
604 
605 	proc_ptrace_connector(child, PTRACE_DETACH);
606 
607 	return 0;
608 }
609 
610 /*
611  * Detach all tasks we were using ptrace on. Called with tasklist held
612  * for writing.
613  */
614 void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
615 {
616 	struct task_struct *p, *n;
617 
618 	list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
619 		if (unlikely(p->ptrace & PT_EXITKILL))
620 			send_sig_info(SIGKILL, SEND_SIG_PRIV, p);
621 
622 		if (__ptrace_detach(tracer, p))
623 			list_add(&p->ptrace_entry, dead);
624 	}
625 }
626 
627 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
628 {
629 	int copied = 0;
630 
631 	while (len > 0) {
632 		char buf[128];
633 		int this_len, retval;
634 
635 		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
636 		retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
637 
638 		if (!retval) {
639 			if (copied)
640 				break;
641 			return -EIO;
642 		}
643 		if (copy_to_user(dst, buf, retval))
644 			return -EFAULT;
645 		copied += retval;
646 		src += retval;
647 		dst += retval;
648 		len -= retval;
649 	}
650 	return copied;
651 }
652 
653 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
654 {
655 	int copied = 0;
656 
657 	while (len > 0) {
658 		char buf[128];
659 		int this_len, retval;
660 
661 		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
662 		if (copy_from_user(buf, src, this_len))
663 			return -EFAULT;
664 		retval = ptrace_access_vm(tsk, dst, buf, this_len,
665 				FOLL_FORCE | FOLL_WRITE);
666 		if (!retval) {
667 			if (copied)
668 				break;
669 			return -EIO;
670 		}
671 		copied += retval;
672 		src += retval;
673 		dst += retval;
674 		len -= retval;
675 	}
676 	return copied;
677 }
678 
679 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
680 {
681 	unsigned flags;
682 	int ret;
683 
684 	ret = check_ptrace_options(data);
685 	if (ret)
686 		return ret;
687 
688 	/* Avoid intermediate state when all opts are cleared */
689 	flags = child->ptrace;
690 	flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
691 	flags |= (data << PT_OPT_FLAG_SHIFT);
692 	child->ptrace = flags;
693 
694 	return 0;
695 }
696 
697 static int ptrace_getsiginfo(struct task_struct *child, kernel_siginfo_t *info)
698 {
699 	unsigned long flags;
700 	int error = -ESRCH;
701 
702 	if (lock_task_sighand(child, &flags)) {
703 		error = -EINVAL;
704 		if (likely(child->last_siginfo != NULL)) {
705 			copy_siginfo(info, child->last_siginfo);
706 			error = 0;
707 		}
708 		unlock_task_sighand(child, &flags);
709 	}
710 	return error;
711 }
712 
713 static int ptrace_setsiginfo(struct task_struct *child, const kernel_siginfo_t *info)
714 {
715 	unsigned long flags;
716 	int error = -ESRCH;
717 
718 	if (lock_task_sighand(child, &flags)) {
719 		error = -EINVAL;
720 		if (likely(child->last_siginfo != NULL)) {
721 			copy_siginfo(child->last_siginfo, info);
722 			error = 0;
723 		}
724 		unlock_task_sighand(child, &flags);
725 	}
726 	return error;
727 }
728 
729 static int ptrace_peek_siginfo(struct task_struct *child,
730 				unsigned long addr,
731 				unsigned long data)
732 {
733 	struct ptrace_peeksiginfo_args arg;
734 	struct sigpending *pending;
735 	struct sigqueue *q;
736 	int ret, i;
737 
738 	ret = copy_from_user(&arg, (void __user *) addr,
739 				sizeof(struct ptrace_peeksiginfo_args));
740 	if (ret)
741 		return -EFAULT;
742 
743 	if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
744 		return -EINVAL; /* unknown flags */
745 
746 	if (arg.nr < 0)
747 		return -EINVAL;
748 
749 	/* Ensure arg.off fits in an unsigned long */
750 	if (arg.off > ULONG_MAX)
751 		return 0;
752 
753 	if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
754 		pending = &child->signal->shared_pending;
755 	else
756 		pending = &child->pending;
757 
758 	for (i = 0; i < arg.nr; ) {
759 		kernel_siginfo_t info;
760 		unsigned long off = arg.off + i;
761 		bool found = false;
762 
763 		spin_lock_irq(&child->sighand->siglock);
764 		list_for_each_entry(q, &pending->list, list) {
765 			if (!off--) {
766 				found = true;
767 				copy_siginfo(&info, &q->info);
768 				break;
769 			}
770 		}
771 		spin_unlock_irq(&child->sighand->siglock);
772 
773 		if (!found) /* beyond the end of the list */
774 			break;
775 
776 #ifdef CONFIG_COMPAT
777 		if (unlikely(in_compat_syscall())) {
778 			compat_siginfo_t __user *uinfo = compat_ptr(data);
779 
780 			if (copy_siginfo_to_user32(uinfo, &info)) {
781 				ret = -EFAULT;
782 				break;
783 			}
784 
785 		} else
786 #endif
787 		{
788 			siginfo_t __user *uinfo = (siginfo_t __user *) data;
789 
790 			if (copy_siginfo_to_user(uinfo, &info)) {
791 				ret = -EFAULT;
792 				break;
793 			}
794 		}
795 
796 		data += sizeof(siginfo_t);
797 		i++;
798 
799 		if (signal_pending(current))
800 			break;
801 
802 		cond_resched();
803 	}
804 
805 	if (i > 0)
806 		return i;
807 
808 	return ret;
809 }
810 
811 #ifdef CONFIG_RSEQ
812 static long ptrace_get_rseq_configuration(struct task_struct *task,
813 					  unsigned long size, void __user *data)
814 {
815 	struct ptrace_rseq_configuration conf = {
816 		.rseq_abi_pointer = (u64)(uintptr_t)task->rseq,
817 		.rseq_abi_size = task->rseq_len,
818 		.signature = task->rseq_sig,
819 		.flags = 0,
820 	};
821 
822 	size = min_t(unsigned long, size, sizeof(conf));
823 	if (copy_to_user(data, &conf, size))
824 		return -EFAULT;
825 	return sizeof(conf);
826 }
827 #endif
828 
829 #define is_singlestep(request)		((request) == PTRACE_SINGLESTEP)
830 
831 #ifdef PTRACE_SINGLEBLOCK
832 #define is_singleblock(request)		((request) == PTRACE_SINGLEBLOCK)
833 #else
834 #define is_singleblock(request)		0
835 #endif
836 
837 #ifdef PTRACE_SYSEMU
838 #define is_sysemu_singlestep(request)	((request) == PTRACE_SYSEMU_SINGLESTEP)
839 #else
840 #define is_sysemu_singlestep(request)	0
841 #endif
842 
843 static int ptrace_resume(struct task_struct *child, long request,
844 			 unsigned long data)
845 {
846 	if (!valid_signal(data))
847 		return -EIO;
848 
849 	if (request == PTRACE_SYSCALL)
850 		set_task_syscall_work(child, SYSCALL_TRACE);
851 	else
852 		clear_task_syscall_work(child, SYSCALL_TRACE);
853 
854 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
855 	if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
856 		set_task_syscall_work(child, SYSCALL_EMU);
857 	else
858 		clear_task_syscall_work(child, SYSCALL_EMU);
859 #endif
860 
861 	if (is_singleblock(request)) {
862 		if (unlikely(!arch_has_block_step()))
863 			return -EIO;
864 		user_enable_block_step(child);
865 	} else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
866 		if (unlikely(!arch_has_single_step()))
867 			return -EIO;
868 		user_enable_single_step(child);
869 	} else {
870 		user_disable_single_step(child);
871 	}
872 
873 	/*
874 	 * Change ->exit_code and ->state under siglock to avoid the race
875 	 * with wait_task_stopped() in between; a non-zero ->exit_code will
876 	 * wrongly look like another report from tracee.
877 	 *
878 	 * Note that we need siglock even if ->exit_code == data and/or this
879 	 * status was not reported yet, the new status must not be cleared by
880 	 * wait_task_stopped() after resume.
881 	 */
882 	spin_lock_irq(&child->sighand->siglock);
883 	child->exit_code = data;
884 	child->jobctl &= ~JOBCTL_TRACED;
885 	wake_up_state(child, __TASK_TRACED);
886 	spin_unlock_irq(&child->sighand->siglock);
887 
888 	return 0;
889 }
890 
891 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
892 
893 static const struct user_regset *
894 find_regset(const struct user_regset_view *view, unsigned int type)
895 {
896 	const struct user_regset *regset;
897 	int n;
898 
899 	for (n = 0; n < view->n; ++n) {
900 		regset = view->regsets + n;
901 		if (regset->core_note_type == type)
902 			return regset;
903 	}
904 
905 	return NULL;
906 }
907 
908 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
909 			 struct iovec *kiov)
910 {
911 	const struct user_regset_view *view = task_user_regset_view(task);
912 	const struct user_regset *regset = find_regset(view, type);
913 	int regset_no;
914 
915 	if (!regset || (kiov->iov_len % regset->size) != 0)
916 		return -EINVAL;
917 
918 	regset_no = regset - view->regsets;
919 	kiov->iov_len = min(kiov->iov_len,
920 			    (__kernel_size_t) (regset->n * regset->size));
921 
922 	if (req == PTRACE_GETREGSET)
923 		return copy_regset_to_user(task, view, regset_no, 0,
924 					   kiov->iov_len, kiov->iov_base);
925 	else
926 		return copy_regset_from_user(task, view, regset_no, 0,
927 					     kiov->iov_len, kiov->iov_base);
928 }
929 
930 /*
931  * This is declared in linux/regset.h and defined in machine-dependent
932  * code.  We put the export here, near the primary machine-neutral use,
933  * to ensure no machine forgets it.
934  */
935 EXPORT_SYMBOL_GPL(task_user_regset_view);
936 
937 static unsigned long
938 ptrace_get_syscall_info_entry(struct task_struct *child, struct pt_regs *regs,
939 			      struct ptrace_syscall_info *info)
940 {
941 	unsigned long args[ARRAY_SIZE(info->entry.args)];
942 	int i;
943 
944 	info->op = PTRACE_SYSCALL_INFO_ENTRY;
945 	info->entry.nr = syscall_get_nr(child, regs);
946 	syscall_get_arguments(child, regs, args);
947 	for (i = 0; i < ARRAY_SIZE(args); i++)
948 		info->entry.args[i] = args[i];
949 
950 	/* args is the last field in struct ptrace_syscall_info.entry */
951 	return offsetofend(struct ptrace_syscall_info, entry.args);
952 }
953 
954 static unsigned long
955 ptrace_get_syscall_info_seccomp(struct task_struct *child, struct pt_regs *regs,
956 				struct ptrace_syscall_info *info)
957 {
958 	/*
959 	 * As struct ptrace_syscall_info.entry is currently a subset
960 	 * of struct ptrace_syscall_info.seccomp, it makes sense to
961 	 * initialize that subset using ptrace_get_syscall_info_entry().
962 	 * This can be reconsidered in the future if these structures
963 	 * diverge significantly enough.
964 	 */
965 	ptrace_get_syscall_info_entry(child, regs, info);
966 	info->op = PTRACE_SYSCALL_INFO_SECCOMP;
967 	info->seccomp.ret_data = child->ptrace_message;
968 
969 	/* ret_data is the last field in struct ptrace_syscall_info.seccomp */
970 	return offsetofend(struct ptrace_syscall_info, seccomp.ret_data);
971 }
972 
973 static unsigned long
974 ptrace_get_syscall_info_exit(struct task_struct *child, struct pt_regs *regs,
975 			     struct ptrace_syscall_info *info)
976 {
977 	info->op = PTRACE_SYSCALL_INFO_EXIT;
978 	info->exit.rval = syscall_get_error(child, regs);
979 	info->exit.is_error = !!info->exit.rval;
980 	if (!info->exit.is_error)
981 		info->exit.rval = syscall_get_return_value(child, regs);
982 
983 	/* is_error is the last field in struct ptrace_syscall_info.exit */
984 	return offsetofend(struct ptrace_syscall_info, exit.is_error);
985 }
986 
987 static int
988 ptrace_get_syscall_info(struct task_struct *child, unsigned long user_size,
989 			void __user *datavp)
990 {
991 	struct pt_regs *regs = task_pt_regs(child);
992 	struct ptrace_syscall_info info = {
993 		.op = PTRACE_SYSCALL_INFO_NONE,
994 		.arch = syscall_get_arch(child),
995 		.instruction_pointer = instruction_pointer(regs),
996 		.stack_pointer = user_stack_pointer(regs),
997 	};
998 	unsigned long actual_size = offsetof(struct ptrace_syscall_info, entry);
999 	unsigned long write_size;
1000 
1001 	/*
1002 	 * This does not need lock_task_sighand() to access
1003 	 * child->last_siginfo because ptrace_freeze_traced()
1004 	 * called earlier by ptrace_check_attach() ensures that
1005 	 * the tracee cannot go away and clear its last_siginfo.
1006 	 */
1007 	switch (child->last_siginfo ? child->last_siginfo->si_code : 0) {
1008 	case SIGTRAP | 0x80:
1009 		switch (child->ptrace_message) {
1010 		case PTRACE_EVENTMSG_SYSCALL_ENTRY:
1011 			actual_size = ptrace_get_syscall_info_entry(child, regs,
1012 								    &info);
1013 			break;
1014 		case PTRACE_EVENTMSG_SYSCALL_EXIT:
1015 			actual_size = ptrace_get_syscall_info_exit(child, regs,
1016 								   &info);
1017 			break;
1018 		}
1019 		break;
1020 	case SIGTRAP | (PTRACE_EVENT_SECCOMP << 8):
1021 		actual_size = ptrace_get_syscall_info_seccomp(child, regs,
1022 							      &info);
1023 		break;
1024 	}
1025 
1026 	write_size = min(actual_size, user_size);
1027 	return copy_to_user(datavp, &info, write_size) ? -EFAULT : actual_size;
1028 }
1029 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
1030 
1031 int ptrace_request(struct task_struct *child, long request,
1032 		   unsigned long addr, unsigned long data)
1033 {
1034 	bool seized = child->ptrace & PT_SEIZED;
1035 	int ret = -EIO;
1036 	kernel_siginfo_t siginfo, *si;
1037 	void __user *datavp = (void __user *) data;
1038 	unsigned long __user *datalp = datavp;
1039 	unsigned long flags;
1040 
1041 	switch (request) {
1042 	case PTRACE_PEEKTEXT:
1043 	case PTRACE_PEEKDATA:
1044 		return generic_ptrace_peekdata(child, addr, data);
1045 	case PTRACE_POKETEXT:
1046 	case PTRACE_POKEDATA:
1047 		return generic_ptrace_pokedata(child, addr, data);
1048 
1049 #ifdef PTRACE_OLDSETOPTIONS
1050 	case PTRACE_OLDSETOPTIONS:
1051 #endif
1052 	case PTRACE_SETOPTIONS:
1053 		ret = ptrace_setoptions(child, data);
1054 		break;
1055 	case PTRACE_GETEVENTMSG:
1056 		ret = put_user(child->ptrace_message, datalp);
1057 		break;
1058 
1059 	case PTRACE_PEEKSIGINFO:
1060 		ret = ptrace_peek_siginfo(child, addr, data);
1061 		break;
1062 
1063 	case PTRACE_GETSIGINFO:
1064 		ret = ptrace_getsiginfo(child, &siginfo);
1065 		if (!ret)
1066 			ret = copy_siginfo_to_user(datavp, &siginfo);
1067 		break;
1068 
1069 	case PTRACE_SETSIGINFO:
1070 		ret = copy_siginfo_from_user(&siginfo, datavp);
1071 		if (!ret)
1072 			ret = ptrace_setsiginfo(child, &siginfo);
1073 		break;
1074 
1075 	case PTRACE_GETSIGMASK: {
1076 		sigset_t *mask;
1077 
1078 		if (addr != sizeof(sigset_t)) {
1079 			ret = -EINVAL;
1080 			break;
1081 		}
1082 
1083 		if (test_tsk_restore_sigmask(child))
1084 			mask = &child->saved_sigmask;
1085 		else
1086 			mask = &child->blocked;
1087 
1088 		if (copy_to_user(datavp, mask, sizeof(sigset_t)))
1089 			ret = -EFAULT;
1090 		else
1091 			ret = 0;
1092 
1093 		break;
1094 	}
1095 
1096 	case PTRACE_SETSIGMASK: {
1097 		sigset_t new_set;
1098 
1099 		if (addr != sizeof(sigset_t)) {
1100 			ret = -EINVAL;
1101 			break;
1102 		}
1103 
1104 		if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
1105 			ret = -EFAULT;
1106 			break;
1107 		}
1108 
1109 		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
1110 
1111 		/*
1112 		 * Every thread does recalc_sigpending() after resume, so
1113 		 * retarget_shared_pending() and recalc_sigpending() are not
1114 		 * called here.
1115 		 */
1116 		spin_lock_irq(&child->sighand->siglock);
1117 		child->blocked = new_set;
1118 		spin_unlock_irq(&child->sighand->siglock);
1119 
1120 		clear_tsk_restore_sigmask(child);
1121 
1122 		ret = 0;
1123 		break;
1124 	}
1125 
1126 	case PTRACE_INTERRUPT:
1127 		/*
1128 		 * Stop tracee without any side-effect on signal or job
1129 		 * control.  At least one trap is guaranteed to happen
1130 		 * after this request.  If @child is already trapped, the
1131 		 * current trap is not disturbed and another trap will
1132 		 * happen after the current trap is ended with PTRACE_CONT.
1133 		 *
1134 		 * The actual trap might not be PTRACE_EVENT_STOP trap but
1135 		 * the pending condition is cleared regardless.
1136 		 */
1137 		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1138 			break;
1139 
1140 		/*
1141 		 * INTERRUPT doesn't disturb existing trap sans one
1142 		 * exception.  If ptracer issued LISTEN for the current
1143 		 * STOP, this INTERRUPT should clear LISTEN and re-trap
1144 		 * tracee into STOP.
1145 		 */
1146 		if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
1147 			ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1148 
1149 		unlock_task_sighand(child, &flags);
1150 		ret = 0;
1151 		break;
1152 
1153 	case PTRACE_LISTEN:
1154 		/*
1155 		 * Listen for events.  Tracee must be in STOP.  It's not
1156 		 * resumed per-se but is not considered to be in TRACED by
1157 		 * wait(2) or ptrace(2).  If an async event (e.g. group
1158 		 * stop state change) happens, tracee will enter STOP trap
1159 		 * again.  Alternatively, ptracer can issue INTERRUPT to
1160 		 * finish listening and re-trap tracee into STOP.
1161 		 */
1162 		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1163 			break;
1164 
1165 		si = child->last_siginfo;
1166 		if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1167 			child->jobctl |= JOBCTL_LISTENING;
1168 			/*
1169 			 * If NOTIFY is set, it means event happened between
1170 			 * start of this trap and now.  Trigger re-trap.
1171 			 */
1172 			if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1173 				ptrace_signal_wake_up(child, true);
1174 			ret = 0;
1175 		}
1176 		unlock_task_sighand(child, &flags);
1177 		break;
1178 
1179 	case PTRACE_DETACH:	 /* detach a process that was attached. */
1180 		ret = ptrace_detach(child, data);
1181 		break;
1182 
1183 #ifdef CONFIG_BINFMT_ELF_FDPIC
1184 	case PTRACE_GETFDPIC: {
1185 		struct mm_struct *mm = get_task_mm(child);
1186 		unsigned long tmp = 0;
1187 
1188 		ret = -ESRCH;
1189 		if (!mm)
1190 			break;
1191 
1192 		switch (addr) {
1193 		case PTRACE_GETFDPIC_EXEC:
1194 			tmp = mm->context.exec_fdpic_loadmap;
1195 			break;
1196 		case PTRACE_GETFDPIC_INTERP:
1197 			tmp = mm->context.interp_fdpic_loadmap;
1198 			break;
1199 		default:
1200 			break;
1201 		}
1202 		mmput(mm);
1203 
1204 		ret = put_user(tmp, datalp);
1205 		break;
1206 	}
1207 #endif
1208 
1209 	case PTRACE_SINGLESTEP:
1210 #ifdef PTRACE_SINGLEBLOCK
1211 	case PTRACE_SINGLEBLOCK:
1212 #endif
1213 #ifdef PTRACE_SYSEMU
1214 	case PTRACE_SYSEMU:
1215 	case PTRACE_SYSEMU_SINGLESTEP:
1216 #endif
1217 	case PTRACE_SYSCALL:
1218 	case PTRACE_CONT:
1219 		return ptrace_resume(child, request, data);
1220 
1221 	case PTRACE_KILL:
1222 		send_sig_info(SIGKILL, SEND_SIG_NOINFO, child);
1223 		return 0;
1224 
1225 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1226 	case PTRACE_GETREGSET:
1227 	case PTRACE_SETREGSET: {
1228 		struct iovec kiov;
1229 		struct iovec __user *uiov = datavp;
1230 
1231 		if (!access_ok(uiov, sizeof(*uiov)))
1232 			return -EFAULT;
1233 
1234 		if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1235 		    __get_user(kiov.iov_len, &uiov->iov_len))
1236 			return -EFAULT;
1237 
1238 		ret = ptrace_regset(child, request, addr, &kiov);
1239 		if (!ret)
1240 			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1241 		break;
1242 	}
1243 
1244 	case PTRACE_GET_SYSCALL_INFO:
1245 		ret = ptrace_get_syscall_info(child, addr, datavp);
1246 		break;
1247 #endif
1248 
1249 	case PTRACE_SECCOMP_GET_FILTER:
1250 		ret = seccomp_get_filter(child, addr, datavp);
1251 		break;
1252 
1253 	case PTRACE_SECCOMP_GET_METADATA:
1254 		ret = seccomp_get_metadata(child, addr, datavp);
1255 		break;
1256 
1257 #ifdef CONFIG_RSEQ
1258 	case PTRACE_GET_RSEQ_CONFIGURATION:
1259 		ret = ptrace_get_rseq_configuration(child, addr, datavp);
1260 		break;
1261 #endif
1262 
1263 	case PTRACE_SET_SYSCALL_USER_DISPATCH_CONFIG:
1264 		ret = syscall_user_dispatch_set_config(child, addr, datavp);
1265 		break;
1266 
1267 	case PTRACE_GET_SYSCALL_USER_DISPATCH_CONFIG:
1268 		ret = syscall_user_dispatch_get_config(child, addr, datavp);
1269 		break;
1270 
1271 	default:
1272 		break;
1273 	}
1274 
1275 	return ret;
1276 }
1277 
1278 SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1279 		unsigned long, data)
1280 {
1281 	struct task_struct *child;
1282 	long ret;
1283 
1284 	if (request == PTRACE_TRACEME) {
1285 		ret = ptrace_traceme();
1286 		goto out;
1287 	}
1288 
1289 	child = find_get_task_by_vpid(pid);
1290 	if (!child) {
1291 		ret = -ESRCH;
1292 		goto out;
1293 	}
1294 
1295 	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1296 		ret = ptrace_attach(child, request, addr, data);
1297 		goto out_put_task_struct;
1298 	}
1299 
1300 	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1301 				  request == PTRACE_INTERRUPT);
1302 	if (ret < 0)
1303 		goto out_put_task_struct;
1304 
1305 	ret = arch_ptrace(child, request, addr, data);
1306 	if (ret || request != PTRACE_DETACH)
1307 		ptrace_unfreeze_traced(child);
1308 
1309  out_put_task_struct:
1310 	put_task_struct(child);
1311  out:
1312 	return ret;
1313 }
1314 
1315 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1316 			    unsigned long data)
1317 {
1318 	unsigned long tmp;
1319 	int copied;
1320 
1321 	copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1322 	if (copied != sizeof(tmp))
1323 		return -EIO;
1324 	return put_user(tmp, (unsigned long __user *)data);
1325 }
1326 
1327 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1328 			    unsigned long data)
1329 {
1330 	int copied;
1331 
1332 	copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1333 			FOLL_FORCE | FOLL_WRITE);
1334 	return (copied == sizeof(data)) ? 0 : -EIO;
1335 }
1336 
1337 #if defined CONFIG_COMPAT
1338 
1339 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1340 			  compat_ulong_t addr, compat_ulong_t data)
1341 {
1342 	compat_ulong_t __user *datap = compat_ptr(data);
1343 	compat_ulong_t word;
1344 	kernel_siginfo_t siginfo;
1345 	int ret;
1346 
1347 	switch (request) {
1348 	case PTRACE_PEEKTEXT:
1349 	case PTRACE_PEEKDATA:
1350 		ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1351 				FOLL_FORCE);
1352 		if (ret != sizeof(word))
1353 			ret = -EIO;
1354 		else
1355 			ret = put_user(word, datap);
1356 		break;
1357 
1358 	case PTRACE_POKETEXT:
1359 	case PTRACE_POKEDATA:
1360 		ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1361 				FOLL_FORCE | FOLL_WRITE);
1362 		ret = (ret != sizeof(data) ? -EIO : 0);
1363 		break;
1364 
1365 	case PTRACE_GETEVENTMSG:
1366 		ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1367 		break;
1368 
1369 	case PTRACE_GETSIGINFO:
1370 		ret = ptrace_getsiginfo(child, &siginfo);
1371 		if (!ret)
1372 			ret = copy_siginfo_to_user32(
1373 				(struct compat_siginfo __user *) datap,
1374 				&siginfo);
1375 		break;
1376 
1377 	case PTRACE_SETSIGINFO:
1378 		ret = copy_siginfo_from_user32(
1379 			&siginfo, (struct compat_siginfo __user *) datap);
1380 		if (!ret)
1381 			ret = ptrace_setsiginfo(child, &siginfo);
1382 		break;
1383 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1384 	case PTRACE_GETREGSET:
1385 	case PTRACE_SETREGSET:
1386 	{
1387 		struct iovec kiov;
1388 		struct compat_iovec __user *uiov =
1389 			(struct compat_iovec __user *) datap;
1390 		compat_uptr_t ptr;
1391 		compat_size_t len;
1392 
1393 		if (!access_ok(uiov, sizeof(*uiov)))
1394 			return -EFAULT;
1395 
1396 		if (__get_user(ptr, &uiov->iov_base) ||
1397 		    __get_user(len, &uiov->iov_len))
1398 			return -EFAULT;
1399 
1400 		kiov.iov_base = compat_ptr(ptr);
1401 		kiov.iov_len = len;
1402 
1403 		ret = ptrace_regset(child, request, addr, &kiov);
1404 		if (!ret)
1405 			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1406 		break;
1407 	}
1408 #endif
1409 
1410 	default:
1411 		ret = ptrace_request(child, request, addr, data);
1412 	}
1413 
1414 	return ret;
1415 }
1416 
1417 COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1418 		       compat_long_t, addr, compat_long_t, data)
1419 {
1420 	struct task_struct *child;
1421 	long ret;
1422 
1423 	if (request == PTRACE_TRACEME) {
1424 		ret = ptrace_traceme();
1425 		goto out;
1426 	}
1427 
1428 	child = find_get_task_by_vpid(pid);
1429 	if (!child) {
1430 		ret = -ESRCH;
1431 		goto out;
1432 	}
1433 
1434 	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1435 		ret = ptrace_attach(child, request, addr, data);
1436 		goto out_put_task_struct;
1437 	}
1438 
1439 	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1440 				  request == PTRACE_INTERRUPT);
1441 	if (!ret) {
1442 		ret = compat_arch_ptrace(child, request, addr, data);
1443 		if (ret || request != PTRACE_DETACH)
1444 			ptrace_unfreeze_traced(child);
1445 	}
1446 
1447  out_put_task_struct:
1448 	put_task_struct(child);
1449  out:
1450 	return ret;
1451 }
1452 #endif	/* CONFIG_COMPAT */
1453