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 */
37void __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 */
72void __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 */
122static 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
140static 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 */
172static 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
210static 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. */
219static 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;
248ok:
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
264bool 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
273static 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;
361unlock_tasklist:
362	write_unlock_irq(&tasklist_lock);
363unlock_creds:
364	mutex_unlock(&task->signal->cred_guard_mutex);
365out:
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 */
381static 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 */
407static 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 */
432static 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
457static int ptrace_detach(struct task_struct *child, unsigned int data)
458{
459	bool dead = false;
460
461	if (!valid_signal(data))
462		return -EIO;
463
464	/* Architecture-specific hardware disable .. */
465	ptrace_disable(child);
466	clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
467
468	write_lock_irq(&tasklist_lock);
469	/*
470	 * This child can be already killed. Make sure de_thread() or
471	 * our sub-thread doing do_wait() didn't do release_task() yet.
472	 */
473	if (child->ptrace) {
474		child->exit_code = data;
475		dead = __ptrace_detach(current, child);
476	}
477	write_unlock_irq(&tasklist_lock);
478
479	proc_ptrace_connector(child, PTRACE_DETACH);
480	if (unlikely(dead))
481		release_task(child);
482
483	return 0;
484}
485
486/*
487 * Detach all tasks we were using ptrace on. Called with tasklist held
488 * for writing, and returns with it held too. But note it can release
489 * and reacquire the lock.
490 */
491void exit_ptrace(struct task_struct *tracer)
492	__releases(&tasklist_lock)
493	__acquires(&tasklist_lock)
494{
495	struct task_struct *p, *n;
496	LIST_HEAD(ptrace_dead);
497
498	if (likely(list_empty(&tracer->ptraced)))
499		return;
500
501	list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
502		if (unlikely(p->ptrace & PT_EXITKILL))
503			send_sig_info(SIGKILL, SEND_SIG_FORCED, p);
504
505		if (__ptrace_detach(tracer, p))
506			list_add(&p->ptrace_entry, &ptrace_dead);
507	}
508
509	write_unlock_irq(&tasklist_lock);
510	BUG_ON(!list_empty(&tracer->ptraced));
511
512	list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_entry) {
513		list_del_init(&p->ptrace_entry);
514		release_task(p);
515	}
516
517	write_lock_irq(&tasklist_lock);
518}
519
520int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
521{
522	int copied = 0;
523
524	while (len > 0) {
525		char buf[128];
526		int this_len, retval;
527
528		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
529		retval = access_process_vm(tsk, src, buf, this_len, 0);
530		if (!retval) {
531			if (copied)
532				break;
533			return -EIO;
534		}
535		if (copy_to_user(dst, buf, retval))
536			return -EFAULT;
537		copied += retval;
538		src += retval;
539		dst += retval;
540		len -= retval;
541	}
542	return copied;
543}
544
545int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
546{
547	int copied = 0;
548
549	while (len > 0) {
550		char buf[128];
551		int this_len, retval;
552
553		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
554		if (copy_from_user(buf, src, this_len))
555			return -EFAULT;
556		retval = access_process_vm(tsk, dst, buf, this_len, 1);
557		if (!retval) {
558			if (copied)
559				break;
560			return -EIO;
561		}
562		copied += retval;
563		src += retval;
564		dst += retval;
565		len -= retval;
566	}
567	return copied;
568}
569
570static int ptrace_setoptions(struct task_struct *child, unsigned long data)
571{
572	unsigned flags;
573
574	if (data & ~(unsigned long)PTRACE_O_MASK)
575		return -EINVAL;
576
577	/* Avoid intermediate state when all opts are cleared */
578	flags = child->ptrace;
579	flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
580	flags |= (data << PT_OPT_FLAG_SHIFT);
581	child->ptrace = flags;
582
583	return 0;
584}
585
586static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
587{
588	unsigned long flags;
589	int error = -ESRCH;
590
591	if (lock_task_sighand(child, &flags)) {
592		error = -EINVAL;
593		if (likely(child->last_siginfo != NULL)) {
594			*info = *child->last_siginfo;
595			error = 0;
596		}
597		unlock_task_sighand(child, &flags);
598	}
599	return error;
600}
601
602static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
603{
604	unsigned long flags;
605	int error = -ESRCH;
606
607	if (lock_task_sighand(child, &flags)) {
608		error = -EINVAL;
609		if (likely(child->last_siginfo != NULL)) {
610			*child->last_siginfo = *info;
611			error = 0;
612		}
613		unlock_task_sighand(child, &flags);
614	}
615	return error;
616}
617
618static int ptrace_peek_siginfo(struct task_struct *child,
619				unsigned long addr,
620				unsigned long data)
621{
622	struct ptrace_peeksiginfo_args arg;
623	struct sigpending *pending;
624	struct sigqueue *q;
625	int ret, i;
626
627	ret = copy_from_user(&arg, (void __user *) addr,
628				sizeof(struct ptrace_peeksiginfo_args));
629	if (ret)
630		return -EFAULT;
631
632	if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
633		return -EINVAL; /* unknown flags */
634
635	if (arg.nr < 0)
636		return -EINVAL;
637
638	if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
639		pending = &child->signal->shared_pending;
640	else
641		pending = &child->pending;
642
643	for (i = 0; i < arg.nr; ) {
644		siginfo_t info;
645		s32 off = arg.off + i;
646
647		spin_lock_irq(&child->sighand->siglock);
648		list_for_each_entry(q, &pending->list, list) {
649			if (!off--) {
650				copy_siginfo(&info, &q->info);
651				break;
652			}
653		}
654		spin_unlock_irq(&child->sighand->siglock);
655
656		if (off >= 0) /* beyond the end of the list */
657			break;
658
659#ifdef CONFIG_COMPAT
660		if (unlikely(is_compat_task())) {
661			compat_siginfo_t __user *uinfo = compat_ptr(data);
662
663			if (copy_siginfo_to_user32(uinfo, &info) ||
664			    __put_user(info.si_code, &uinfo->si_code)) {
665				ret = -EFAULT;
666				break;
667			}
668
669		} else
670#endif
671		{
672			siginfo_t __user *uinfo = (siginfo_t __user *) data;
673
674			if (copy_siginfo_to_user(uinfo, &info) ||
675			    __put_user(info.si_code, &uinfo->si_code)) {
676				ret = -EFAULT;
677				break;
678			}
679		}
680
681		data += sizeof(siginfo_t);
682		i++;
683
684		if (signal_pending(current))
685			break;
686
687		cond_resched();
688	}
689
690	if (i > 0)
691		return i;
692
693	return ret;
694}
695
696#ifdef PTRACE_SINGLESTEP
697#define is_singlestep(request)		((request) == PTRACE_SINGLESTEP)
698#else
699#define is_singlestep(request)		0
700#endif
701
702#ifdef PTRACE_SINGLEBLOCK
703#define is_singleblock(request)		((request) == PTRACE_SINGLEBLOCK)
704#else
705#define is_singleblock(request)		0
706#endif
707
708#ifdef PTRACE_SYSEMU
709#define is_sysemu_singlestep(request)	((request) == PTRACE_SYSEMU_SINGLESTEP)
710#else
711#define is_sysemu_singlestep(request)	0
712#endif
713
714static int ptrace_resume(struct task_struct *child, long request,
715			 unsigned long data)
716{
717	if (!valid_signal(data))
718		return -EIO;
719
720	if (request == PTRACE_SYSCALL)
721		set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
722	else
723		clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
724
725#ifdef TIF_SYSCALL_EMU
726	if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
727		set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
728	else
729		clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
730#endif
731
732	if (is_singleblock(request)) {
733		if (unlikely(!arch_has_block_step()))
734			return -EIO;
735		user_enable_block_step(child);
736	} else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
737		if (unlikely(!arch_has_single_step()))
738			return -EIO;
739		user_enable_single_step(child);
740	} else {
741		user_disable_single_step(child);
742	}
743
744	child->exit_code = data;
745	wake_up_state(child, __TASK_TRACED);
746
747	return 0;
748}
749
750#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
751
752static const struct user_regset *
753find_regset(const struct user_regset_view *view, unsigned int type)
754{
755	const struct user_regset *regset;
756	int n;
757
758	for (n = 0; n < view->n; ++n) {
759		regset = view->regsets + n;
760		if (regset->core_note_type == type)
761			return regset;
762	}
763
764	return NULL;
765}
766
767static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
768			 struct iovec *kiov)
769{
770	const struct user_regset_view *view = task_user_regset_view(task);
771	const struct user_regset *regset = find_regset(view, type);
772	int regset_no;
773
774	if (!regset || (kiov->iov_len % regset->size) != 0)
775		return -EINVAL;
776
777	regset_no = regset - view->regsets;
778	kiov->iov_len = min(kiov->iov_len,
779			    (__kernel_size_t) (regset->n * regset->size));
780
781	if (req == PTRACE_GETREGSET)
782		return copy_regset_to_user(task, view, regset_no, 0,
783					   kiov->iov_len, kiov->iov_base);
784	else
785		return copy_regset_from_user(task, view, regset_no, 0,
786					     kiov->iov_len, kiov->iov_base);
787}
788
789/*
790 * This is declared in linux/regset.h and defined in machine-dependent
791 * code.  We put the export here, near the primary machine-neutral use,
792 * to ensure no machine forgets it.
793 */
794EXPORT_SYMBOL_GPL(task_user_regset_view);
795#endif
796
797int ptrace_request(struct task_struct *child, long request,
798		   unsigned long addr, unsigned long data)
799{
800	bool seized = child->ptrace & PT_SEIZED;
801	int ret = -EIO;
802	siginfo_t siginfo, *si;
803	void __user *datavp = (void __user *) data;
804	unsigned long __user *datalp = datavp;
805	unsigned long flags;
806
807	switch (request) {
808	case PTRACE_PEEKTEXT:
809	case PTRACE_PEEKDATA:
810		return generic_ptrace_peekdata(child, addr, data);
811	case PTRACE_POKETEXT:
812	case PTRACE_POKEDATA:
813		return generic_ptrace_pokedata(child, addr, data);
814
815#ifdef PTRACE_OLDSETOPTIONS
816	case PTRACE_OLDSETOPTIONS:
817#endif
818	case PTRACE_SETOPTIONS:
819		ret = ptrace_setoptions(child, data);
820		break;
821	case PTRACE_GETEVENTMSG:
822		ret = put_user(child->ptrace_message, datalp);
823		break;
824
825	case PTRACE_PEEKSIGINFO:
826		ret = ptrace_peek_siginfo(child, addr, data);
827		break;
828
829	case PTRACE_GETSIGINFO:
830		ret = ptrace_getsiginfo(child, &siginfo);
831		if (!ret)
832			ret = copy_siginfo_to_user(datavp, &siginfo);
833		break;
834
835	case PTRACE_SETSIGINFO:
836		if (copy_from_user(&siginfo, datavp, sizeof siginfo))
837			ret = -EFAULT;
838		else
839			ret = ptrace_setsiginfo(child, &siginfo);
840		break;
841
842	case PTRACE_GETSIGMASK:
843		if (addr != sizeof(sigset_t)) {
844			ret = -EINVAL;
845			break;
846		}
847
848		if (copy_to_user(datavp, &child->blocked, sizeof(sigset_t)))
849			ret = -EFAULT;
850		else
851			ret = 0;
852
853		break;
854
855	case PTRACE_SETSIGMASK: {
856		sigset_t new_set;
857
858		if (addr != sizeof(sigset_t)) {
859			ret = -EINVAL;
860			break;
861		}
862
863		if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
864			ret = -EFAULT;
865			break;
866		}
867
868		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
869
870		/*
871		 * Every thread does recalc_sigpending() after resume, so
872		 * retarget_shared_pending() and recalc_sigpending() are not
873		 * called here.
874		 */
875		spin_lock_irq(&child->sighand->siglock);
876		child->blocked = new_set;
877		spin_unlock_irq(&child->sighand->siglock);
878
879		ret = 0;
880		break;
881	}
882
883	case PTRACE_INTERRUPT:
884		/*
885		 * Stop tracee without any side-effect on signal or job
886		 * control.  At least one trap is guaranteed to happen
887		 * after this request.  If @child is already trapped, the
888		 * current trap is not disturbed and another trap will
889		 * happen after the current trap is ended with PTRACE_CONT.
890		 *
891		 * The actual trap might not be PTRACE_EVENT_STOP trap but
892		 * the pending condition is cleared regardless.
893		 */
894		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
895			break;
896
897		/*
898		 * INTERRUPT doesn't disturb existing trap sans one
899		 * exception.  If ptracer issued LISTEN for the current
900		 * STOP, this INTERRUPT should clear LISTEN and re-trap
901		 * tracee into STOP.
902		 */
903		if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
904			ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
905
906		unlock_task_sighand(child, &flags);
907		ret = 0;
908		break;
909
910	case PTRACE_LISTEN:
911		/*
912		 * Listen for events.  Tracee must be in STOP.  It's not
913		 * resumed per-se but is not considered to be in TRACED by
914		 * wait(2) or ptrace(2).  If an async event (e.g. group
915		 * stop state change) happens, tracee will enter STOP trap
916		 * again.  Alternatively, ptracer can issue INTERRUPT to
917		 * finish listening and re-trap tracee into STOP.
918		 */
919		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
920			break;
921
922		si = child->last_siginfo;
923		if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
924			child->jobctl |= JOBCTL_LISTENING;
925			/*
926			 * If NOTIFY is set, it means event happened between
927			 * start of this trap and now.  Trigger re-trap.
928			 */
929			if (child->jobctl & JOBCTL_TRAP_NOTIFY)
930				ptrace_signal_wake_up(child, true);
931			ret = 0;
932		}
933		unlock_task_sighand(child, &flags);
934		break;
935
936	case PTRACE_DETACH:	 /* detach a process that was attached. */
937		ret = ptrace_detach(child, data);
938		break;
939
940#ifdef CONFIG_BINFMT_ELF_FDPIC
941	case PTRACE_GETFDPIC: {
942		struct mm_struct *mm = get_task_mm(child);
943		unsigned long tmp = 0;
944
945		ret = -ESRCH;
946		if (!mm)
947			break;
948
949		switch (addr) {
950		case PTRACE_GETFDPIC_EXEC:
951			tmp = mm->context.exec_fdpic_loadmap;
952			break;
953		case PTRACE_GETFDPIC_INTERP:
954			tmp = mm->context.interp_fdpic_loadmap;
955			break;
956		default:
957			break;
958		}
959		mmput(mm);
960
961		ret = put_user(tmp, datalp);
962		break;
963	}
964#endif
965
966#ifdef PTRACE_SINGLESTEP
967	case PTRACE_SINGLESTEP:
968#endif
969#ifdef PTRACE_SINGLEBLOCK
970	case PTRACE_SINGLEBLOCK:
971#endif
972#ifdef PTRACE_SYSEMU
973	case PTRACE_SYSEMU:
974	case PTRACE_SYSEMU_SINGLESTEP:
975#endif
976	case PTRACE_SYSCALL:
977	case PTRACE_CONT:
978		return ptrace_resume(child, request, data);
979
980	case PTRACE_KILL:
981		if (child->exit_state)	/* already dead */
982			return 0;
983		return ptrace_resume(child, request, SIGKILL);
984
985#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
986	case PTRACE_GETREGSET:
987	case PTRACE_SETREGSET: {
988		struct iovec kiov;
989		struct iovec __user *uiov = datavp;
990
991		if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
992			return -EFAULT;
993
994		if (__get_user(kiov.iov_base, &uiov->iov_base) ||
995		    __get_user(kiov.iov_len, &uiov->iov_len))
996			return -EFAULT;
997
998		ret = ptrace_regset(child, request, addr, &kiov);
999		if (!ret)
1000			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1001		break;
1002	}
1003#endif
1004	default:
1005		break;
1006	}
1007
1008	return ret;
1009}
1010
1011static struct task_struct *ptrace_get_task_struct(pid_t pid)
1012{
1013	struct task_struct *child;
1014
1015	rcu_read_lock();
1016	child = find_task_by_vpid(pid);
1017	if (child)
1018		get_task_struct(child);
1019	rcu_read_unlock();
1020
1021	if (!child)
1022		return ERR_PTR(-ESRCH);
1023	return child;
1024}
1025
1026#ifndef arch_ptrace_attach
1027#define arch_ptrace_attach(child)	do { } while (0)
1028#endif
1029
1030SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1031		unsigned long, data)
1032{
1033	struct task_struct *child;
1034	long ret;
1035
1036	if (request == PTRACE_TRACEME) {
1037		ret = ptrace_traceme();
1038		if (!ret)
1039			arch_ptrace_attach(current);
1040		goto out;
1041	}
1042
1043	child = ptrace_get_task_struct(pid);
1044	if (IS_ERR(child)) {
1045		ret = PTR_ERR(child);
1046		goto out;
1047	}
1048
1049	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1050		ret = ptrace_attach(child, request, addr, data);
1051		/*
1052		 * Some architectures need to do book-keeping after
1053		 * a ptrace attach.
1054		 */
1055		if (!ret)
1056			arch_ptrace_attach(child);
1057		goto out_put_task_struct;
1058	}
1059
1060	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1061				  request == PTRACE_INTERRUPT);
1062	if (ret < 0)
1063		goto out_put_task_struct;
1064
1065	ret = arch_ptrace(child, request, addr, data);
1066	if (ret || request != PTRACE_DETACH)
1067		ptrace_unfreeze_traced(child);
1068
1069 out_put_task_struct:
1070	put_task_struct(child);
1071 out:
1072	return ret;
1073}
1074
1075int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1076			    unsigned long data)
1077{
1078	unsigned long tmp;
1079	int copied;
1080
1081	copied = access_process_vm(tsk, addr, &tmp, sizeof(tmp), 0);
1082	if (copied != sizeof(tmp))
1083		return -EIO;
1084	return put_user(tmp, (unsigned long __user *)data);
1085}
1086
1087int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1088			    unsigned long data)
1089{
1090	int copied;
1091
1092	copied = access_process_vm(tsk, addr, &data, sizeof(data), 1);
1093	return (copied == sizeof(data)) ? 0 : -EIO;
1094}
1095
1096#if defined CONFIG_COMPAT
1097#include <linux/compat.h>
1098
1099int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1100			  compat_ulong_t addr, compat_ulong_t data)
1101{
1102	compat_ulong_t __user *datap = compat_ptr(data);
1103	compat_ulong_t word;
1104	siginfo_t siginfo;
1105	int ret;
1106
1107	switch (request) {
1108	case PTRACE_PEEKTEXT:
1109	case PTRACE_PEEKDATA:
1110		ret = access_process_vm(child, addr, &word, sizeof(word), 0);
1111		if (ret != sizeof(word))
1112			ret = -EIO;
1113		else
1114			ret = put_user(word, datap);
1115		break;
1116
1117	case PTRACE_POKETEXT:
1118	case PTRACE_POKEDATA:
1119		ret = access_process_vm(child, addr, &data, sizeof(data), 1);
1120		ret = (ret != sizeof(data) ? -EIO : 0);
1121		break;
1122
1123	case PTRACE_GETEVENTMSG:
1124		ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1125		break;
1126
1127	case PTRACE_GETSIGINFO:
1128		ret = ptrace_getsiginfo(child, &siginfo);
1129		if (!ret)
1130			ret = copy_siginfo_to_user32(
1131				(struct compat_siginfo __user *) datap,
1132				&siginfo);
1133		break;
1134
1135	case PTRACE_SETSIGINFO:
1136		memset(&siginfo, 0, sizeof siginfo);
1137		if (copy_siginfo_from_user32(
1138			    &siginfo, (struct compat_siginfo __user *) datap))
1139			ret = -EFAULT;
1140		else
1141			ret = ptrace_setsiginfo(child, &siginfo);
1142		break;
1143#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1144	case PTRACE_GETREGSET:
1145	case PTRACE_SETREGSET:
1146	{
1147		struct iovec kiov;
1148		struct compat_iovec __user *uiov =
1149			(struct compat_iovec __user *) datap;
1150		compat_uptr_t ptr;
1151		compat_size_t len;
1152
1153		if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1154			return -EFAULT;
1155
1156		if (__get_user(ptr, &uiov->iov_base) ||
1157		    __get_user(len, &uiov->iov_len))
1158			return -EFAULT;
1159
1160		kiov.iov_base = compat_ptr(ptr);
1161		kiov.iov_len = len;
1162
1163		ret = ptrace_regset(child, request, addr, &kiov);
1164		if (!ret)
1165			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1166		break;
1167	}
1168#endif
1169
1170	default:
1171		ret = ptrace_request(child, request, addr, data);
1172	}
1173
1174	return ret;
1175}
1176
1177COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1178		       compat_long_t, addr, compat_long_t, data)
1179{
1180	struct task_struct *child;
1181	long ret;
1182
1183	if (request == PTRACE_TRACEME) {
1184		ret = ptrace_traceme();
1185		goto out;
1186	}
1187
1188	child = ptrace_get_task_struct(pid);
1189	if (IS_ERR(child)) {
1190		ret = PTR_ERR(child);
1191		goto out;
1192	}
1193
1194	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1195		ret = ptrace_attach(child, request, addr, data);
1196		/*
1197		 * Some architectures need to do book-keeping after
1198		 * a ptrace attach.
1199		 */
1200		if (!ret)
1201			arch_ptrace_attach(child);
1202		goto out_put_task_struct;
1203	}
1204
1205	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1206				  request == PTRACE_INTERRUPT);
1207	if (!ret) {
1208		ret = compat_arch_ptrace(child, request, addr, data);
1209		if (ret || request != PTRACE_DETACH)
1210			ptrace_unfreeze_traced(child);
1211	}
1212
1213 out_put_task_struct:
1214	put_task_struct(child);
1215 out:
1216	return ret;
1217}
1218#endif	/* CONFIG_COMPAT */
1219