xref: /linux/kernel/ptrace.c (revision 215a0d305c5651928eb67c96bcedd0a6c297dfce)
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 
35 /*
36  * Access another process' address space via ptrace.
37  * Source/target buffer must be kernel space,
38  * Do not walk the page table directly, use get_user_pages
39  */
40 int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
41 		     void *buf, int len, unsigned int gup_flags)
42 {
43 	struct mm_struct *mm;
44 	int ret;
45 
46 	mm = get_task_mm(tsk);
47 	if (!mm)
48 		return 0;
49 
50 	if (!tsk->ptrace ||
51 	    (current != tsk->parent) ||
52 	    ((get_dumpable(mm) != SUID_DUMP_USER) &&
53 	     !ptracer_capable(tsk, mm->user_ns))) {
54 		mmput(mm);
55 		return 0;
56 	}
57 
58 	ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
59 	mmput(mm);
60 
61 	return ret;
62 }
63 
64 
65 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent,
66 		   const struct cred *ptracer_cred)
67 {
68 	BUG_ON(!list_empty(&child->ptrace_entry));
69 	list_add(&child->ptrace_entry, &new_parent->ptraced);
70 	child->parent = new_parent;
71 	child->ptracer_cred = get_cred(ptracer_cred);
72 }
73 
74 /*
75  * ptrace a task: make the debugger its new parent and
76  * move it to the ptrace list.
77  *
78  * Must be called with the tasklist lock write-held.
79  */
80 static void ptrace_link(struct task_struct *child, struct task_struct *new_parent)
81 {
82 	__ptrace_link(child, new_parent, current_cred());
83 }
84 
85 /**
86  * __ptrace_unlink - unlink ptracee and restore its execution state
87  * @child: ptracee to be unlinked
88  *
89  * Remove @child from the ptrace list, move it back to the original parent,
90  * and restore the execution state so that it conforms to the group stop
91  * state.
92  *
93  * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
94  * exiting.  For PTRACE_DETACH, unless the ptracee has been killed between
95  * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
96  * If the ptracer is exiting, the ptracee can be in any state.
97  *
98  * After detach, the ptracee should be in a state which conforms to the
99  * group stop.  If the group is stopped or in the process of stopping, the
100  * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
101  * up from TASK_TRACED.
102  *
103  * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
104  * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
105  * to but in the opposite direction of what happens while attaching to a
106  * stopped task.  However, in this direction, the intermediate RUNNING
107  * state is not hidden even from the current ptracer and if it immediately
108  * re-attaches and performs a WNOHANG wait(2), it may fail.
109  *
110  * CONTEXT:
111  * write_lock_irq(tasklist_lock)
112  */
113 void __ptrace_unlink(struct task_struct *child)
114 {
115 	const struct cred *old_cred;
116 	BUG_ON(!child->ptrace);
117 
118 	clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
119 
120 	child->parent = child->real_parent;
121 	list_del_init(&child->ptrace_entry);
122 	old_cred = child->ptracer_cred;
123 	child->ptracer_cred = NULL;
124 	put_cred(old_cred);
125 
126 	spin_lock(&child->sighand->siglock);
127 	child->ptrace = 0;
128 	/*
129 	 * Clear all pending traps and TRAPPING.  TRAPPING should be
130 	 * cleared regardless of JOBCTL_STOP_PENDING.  Do it explicitly.
131 	 */
132 	task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
133 	task_clear_jobctl_trapping(child);
134 
135 	/*
136 	 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
137 	 * @child isn't dead.
138 	 */
139 	if (!(child->flags & PF_EXITING) &&
140 	    (child->signal->flags & SIGNAL_STOP_STOPPED ||
141 	     child->signal->group_stop_count)) {
142 		child->jobctl |= JOBCTL_STOP_PENDING;
143 
144 		/*
145 		 * This is only possible if this thread was cloned by the
146 		 * traced task running in the stopped group, set the signal
147 		 * for the future reports.
148 		 * FIXME: we should change ptrace_init_task() to handle this
149 		 * case.
150 		 */
151 		if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
152 			child->jobctl |= SIGSTOP;
153 	}
154 
155 	/*
156 	 * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
157 	 * @child in the butt.  Note that @resume should be used iff @child
158 	 * is in TASK_TRACED; otherwise, we might unduly disrupt
159 	 * TASK_KILLABLE sleeps.
160 	 */
161 	if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
162 		ptrace_signal_wake_up(child, true);
163 
164 	spin_unlock(&child->sighand->siglock);
165 }
166 
167 /* Ensure that nothing can wake it up, even SIGKILL */
168 static bool ptrace_freeze_traced(struct task_struct *task)
169 {
170 	bool ret = false;
171 
172 	/* Lockless, nobody but us can set this flag */
173 	if (task->jobctl & JOBCTL_LISTENING)
174 		return ret;
175 
176 	spin_lock_irq(&task->sighand->siglock);
177 	if (task_is_traced(task) && !__fatal_signal_pending(task)) {
178 		task->state = __TASK_TRACED;
179 		ret = true;
180 	}
181 	spin_unlock_irq(&task->sighand->siglock);
182 
183 	return ret;
184 }
185 
186 static void ptrace_unfreeze_traced(struct task_struct *task)
187 {
188 	if (task->state != __TASK_TRACED)
189 		return;
190 
191 	WARN_ON(!task->ptrace || task->parent != current);
192 
193 	/*
194 	 * PTRACE_LISTEN can allow ptrace_trap_notify to wake us up remotely.
195 	 * Recheck state under the lock to close this race.
196 	 */
197 	spin_lock_irq(&task->sighand->siglock);
198 	if (task->state == __TASK_TRACED) {
199 		if (__fatal_signal_pending(task))
200 			wake_up_state(task, __TASK_TRACED);
201 		else
202 			task->state = TASK_TRACED;
203 	}
204 	spin_unlock_irq(&task->sighand->siglock);
205 }
206 
207 /**
208  * ptrace_check_attach - check whether ptracee is ready for ptrace operation
209  * @child: ptracee to check for
210  * @ignore_state: don't check whether @child is currently %TASK_TRACED
211  *
212  * Check whether @child is being ptraced by %current and ready for further
213  * ptrace operations.  If @ignore_state is %false, @child also should be in
214  * %TASK_TRACED state and on return the child is guaranteed to be traced
215  * and not executing.  If @ignore_state is %true, @child can be in any
216  * state.
217  *
218  * CONTEXT:
219  * Grabs and releases tasklist_lock and @child->sighand->siglock.
220  *
221  * RETURNS:
222  * 0 on success, -ESRCH if %child is not ready.
223  */
224 static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
225 {
226 	int ret = -ESRCH;
227 
228 	/*
229 	 * We take the read lock around doing both checks to close a
230 	 * possible race where someone else was tracing our child and
231 	 * detached between these two checks.  After this locked check,
232 	 * we are sure that this is our traced child and that can only
233 	 * be changed by us so it's not changing right after this.
234 	 */
235 	read_lock(&tasklist_lock);
236 	if (child->ptrace && child->parent == current) {
237 		WARN_ON(child->state == __TASK_TRACED);
238 		/*
239 		 * child->sighand can't be NULL, release_task()
240 		 * does ptrace_unlink() before __exit_signal().
241 		 */
242 		if (ignore_state || ptrace_freeze_traced(child))
243 			ret = 0;
244 	}
245 	read_unlock(&tasklist_lock);
246 
247 	if (!ret && !ignore_state) {
248 		if (!wait_task_inactive(child, __TASK_TRACED)) {
249 			/*
250 			 * This can only happen if may_ptrace_stop() fails and
251 			 * ptrace_stop() changes ->state back to TASK_RUNNING,
252 			 * so we should not worry about leaking __TASK_TRACED.
253 			 */
254 			WARN_ON(child->state == __TASK_TRACED);
255 			ret = -ESRCH;
256 		}
257 	}
258 
259 	return ret;
260 }
261 
262 static int ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
263 {
264 	if (mode & PTRACE_MODE_NOAUDIT)
265 		return has_ns_capability_noaudit(current, ns, CAP_SYS_PTRACE);
266 	else
267 		return has_ns_capability(current, ns, CAP_SYS_PTRACE);
268 }
269 
270 /* Returns 0 on success, -errno on denial. */
271 static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
272 {
273 	const struct cred *cred = current_cred(), *tcred;
274 	struct mm_struct *mm;
275 	kuid_t caller_uid;
276 	kgid_t caller_gid;
277 
278 	if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
279 		WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
280 		return -EPERM;
281 	}
282 
283 	/* May we inspect the given task?
284 	 * This check is used both for attaching with ptrace
285 	 * and for allowing access to sensitive information in /proc.
286 	 *
287 	 * ptrace_attach denies several cases that /proc allows
288 	 * because setting up the necessary parent/child relationship
289 	 * or halting the specified task is impossible.
290 	 */
291 
292 	/* Don't let security modules deny introspection */
293 	if (same_thread_group(task, current))
294 		return 0;
295 	rcu_read_lock();
296 	if (mode & PTRACE_MODE_FSCREDS) {
297 		caller_uid = cred->fsuid;
298 		caller_gid = cred->fsgid;
299 	} else {
300 		/*
301 		 * Using the euid would make more sense here, but something
302 		 * in userland might rely on the old behavior, and this
303 		 * shouldn't be a security problem since
304 		 * PTRACE_MODE_REALCREDS implies that the caller explicitly
305 		 * used a syscall that requests access to another process
306 		 * (and not a filesystem syscall to procfs).
307 		 */
308 		caller_uid = cred->uid;
309 		caller_gid = cred->gid;
310 	}
311 	tcred = __task_cred(task);
312 	if (uid_eq(caller_uid, tcred->euid) &&
313 	    uid_eq(caller_uid, tcred->suid) &&
314 	    uid_eq(caller_uid, tcred->uid)  &&
315 	    gid_eq(caller_gid, tcred->egid) &&
316 	    gid_eq(caller_gid, tcred->sgid) &&
317 	    gid_eq(caller_gid, tcred->gid))
318 		goto ok;
319 	if (ptrace_has_cap(tcred->user_ns, mode))
320 		goto ok;
321 	rcu_read_unlock();
322 	return -EPERM;
323 ok:
324 	rcu_read_unlock();
325 	/*
326 	 * If a task drops privileges and becomes nondumpable (through a syscall
327 	 * like setresuid()) while we are trying to access it, we must ensure
328 	 * that the dumpability is read after the credentials; otherwise,
329 	 * we may be able to attach to a task that we shouldn't be able to
330 	 * attach to (as if the task had dropped privileges without becoming
331 	 * nondumpable).
332 	 * Pairs with a write barrier in commit_creds().
333 	 */
334 	smp_rmb();
335 	mm = task->mm;
336 	if (mm &&
337 	    ((get_dumpable(mm) != SUID_DUMP_USER) &&
338 	     !ptrace_has_cap(mm->user_ns, mode)))
339 	    return -EPERM;
340 
341 	return security_ptrace_access_check(task, mode);
342 }
343 
344 bool ptrace_may_access(struct task_struct *task, unsigned int mode)
345 {
346 	int err;
347 	task_lock(task);
348 	err = __ptrace_may_access(task, mode);
349 	task_unlock(task);
350 	return !err;
351 }
352 
353 static int ptrace_attach(struct task_struct *task, long request,
354 			 unsigned long addr,
355 			 unsigned long flags)
356 {
357 	bool seize = (request == PTRACE_SEIZE);
358 	int retval;
359 
360 	retval = -EIO;
361 	if (seize) {
362 		if (addr != 0)
363 			goto out;
364 		if (flags & ~(unsigned long)PTRACE_O_MASK)
365 			goto out;
366 		flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
367 	} else {
368 		flags = PT_PTRACED;
369 	}
370 
371 	audit_ptrace(task);
372 
373 	retval = -EPERM;
374 	if (unlikely(task->flags & PF_KTHREAD))
375 		goto out;
376 	if (same_thread_group(task, current))
377 		goto out;
378 
379 	/*
380 	 * Protect exec's credential calculations against our interference;
381 	 * SUID, SGID and LSM creds get determined differently
382 	 * under ptrace.
383 	 */
384 	retval = -ERESTARTNOINTR;
385 	if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
386 		goto out;
387 
388 	task_lock(task);
389 	retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
390 	task_unlock(task);
391 	if (retval)
392 		goto unlock_creds;
393 
394 	write_lock_irq(&tasklist_lock);
395 	retval = -EPERM;
396 	if (unlikely(task->exit_state))
397 		goto unlock_tasklist;
398 	if (task->ptrace)
399 		goto unlock_tasklist;
400 
401 	if (seize)
402 		flags |= PT_SEIZED;
403 	task->ptrace = flags;
404 
405 	ptrace_link(task, current);
406 
407 	/* SEIZE doesn't trap tracee on attach */
408 	if (!seize)
409 		send_sig_info(SIGSTOP, SEND_SIG_PRIV, task);
410 
411 	spin_lock(&task->sighand->siglock);
412 
413 	/*
414 	 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
415 	 * TRAPPING, and kick it so that it transits to TRACED.  TRAPPING
416 	 * will be cleared if the child completes the transition or any
417 	 * event which clears the group stop states happens.  We'll wait
418 	 * for the transition to complete before returning from this
419 	 * function.
420 	 *
421 	 * This hides STOPPED -> RUNNING -> TRACED transition from the
422 	 * attaching thread but a different thread in the same group can
423 	 * still observe the transient RUNNING state.  IOW, if another
424 	 * thread's WNOHANG wait(2) on the stopped tracee races against
425 	 * ATTACH, the wait(2) may fail due to the transient RUNNING.
426 	 *
427 	 * The following task_is_stopped() test is safe as both transitions
428 	 * in and out of STOPPED are protected by siglock.
429 	 */
430 	if (task_is_stopped(task) &&
431 	    task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
432 		signal_wake_up_state(task, __TASK_STOPPED);
433 
434 	spin_unlock(&task->sighand->siglock);
435 
436 	retval = 0;
437 unlock_tasklist:
438 	write_unlock_irq(&tasklist_lock);
439 unlock_creds:
440 	mutex_unlock(&task->signal->cred_guard_mutex);
441 out:
442 	if (!retval) {
443 		/*
444 		 * We do not bother to change retval or clear JOBCTL_TRAPPING
445 		 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
446 		 * not return to user-mode, it will exit and clear this bit in
447 		 * __ptrace_unlink() if it wasn't already cleared by the tracee;
448 		 * and until then nobody can ptrace this task.
449 		 */
450 		wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
451 		proc_ptrace_connector(task, PTRACE_ATTACH);
452 	}
453 
454 	return retval;
455 }
456 
457 /**
458  * ptrace_traceme  --  helper for PTRACE_TRACEME
459  *
460  * Performs checks and sets PT_PTRACED.
461  * Should be used by all ptrace implementations for PTRACE_TRACEME.
462  */
463 static int ptrace_traceme(void)
464 {
465 	int ret = -EPERM;
466 
467 	write_lock_irq(&tasklist_lock);
468 	/* Are we already being traced? */
469 	if (!current->ptrace) {
470 		ret = security_ptrace_traceme(current->parent);
471 		/*
472 		 * Check PF_EXITING to ensure ->real_parent has not passed
473 		 * exit_ptrace(). Otherwise we don't report the error but
474 		 * pretend ->real_parent untraces us right after return.
475 		 */
476 		if (!ret && !(current->real_parent->flags & PF_EXITING)) {
477 			current->ptrace = PT_PTRACED;
478 			ptrace_link(current, current->real_parent);
479 		}
480 	}
481 	write_unlock_irq(&tasklist_lock);
482 
483 	return ret;
484 }
485 
486 /*
487  * Called with irqs disabled, returns true if childs should reap themselves.
488  */
489 static int ignoring_children(struct sighand_struct *sigh)
490 {
491 	int ret;
492 	spin_lock(&sigh->siglock);
493 	ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
494 	      (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
495 	spin_unlock(&sigh->siglock);
496 	return ret;
497 }
498 
499 /*
500  * Called with tasklist_lock held for writing.
501  * Unlink a traced task, and clean it up if it was a traced zombie.
502  * Return true if it needs to be reaped with release_task().
503  * (We can't call release_task() here because we already hold tasklist_lock.)
504  *
505  * If it's a zombie, our attachedness prevented normal parent notification
506  * or self-reaping.  Do notification now if it would have happened earlier.
507  * If it should reap itself, return true.
508  *
509  * If it's our own child, there is no notification to do. But if our normal
510  * children self-reap, then this child was prevented by ptrace and we must
511  * reap it now, in that case we must also wake up sub-threads sleeping in
512  * do_wait().
513  */
514 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
515 {
516 	bool dead;
517 
518 	__ptrace_unlink(p);
519 
520 	if (p->exit_state != EXIT_ZOMBIE)
521 		return false;
522 
523 	dead = !thread_group_leader(p);
524 
525 	if (!dead && thread_group_empty(p)) {
526 		if (!same_thread_group(p->real_parent, tracer))
527 			dead = do_notify_parent(p, p->exit_signal);
528 		else if (ignoring_children(tracer->sighand)) {
529 			__wake_up_parent(p, tracer);
530 			dead = true;
531 		}
532 	}
533 	/* Mark it as in the process of being reaped. */
534 	if (dead)
535 		p->exit_state = EXIT_DEAD;
536 	return dead;
537 }
538 
539 static int ptrace_detach(struct task_struct *child, unsigned int data)
540 {
541 	if (!valid_signal(data))
542 		return -EIO;
543 
544 	/* Architecture-specific hardware disable .. */
545 	ptrace_disable(child);
546 
547 	write_lock_irq(&tasklist_lock);
548 	/*
549 	 * We rely on ptrace_freeze_traced(). It can't be killed and
550 	 * untraced by another thread, it can't be a zombie.
551 	 */
552 	WARN_ON(!child->ptrace || child->exit_state);
553 	/*
554 	 * tasklist_lock avoids the race with wait_task_stopped(), see
555 	 * the comment in ptrace_resume().
556 	 */
557 	child->exit_code = data;
558 	__ptrace_detach(current, child);
559 	write_unlock_irq(&tasklist_lock);
560 
561 	proc_ptrace_connector(child, PTRACE_DETACH);
562 
563 	return 0;
564 }
565 
566 /*
567  * Detach all tasks we were using ptrace on. Called with tasklist held
568  * for writing.
569  */
570 void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
571 {
572 	struct task_struct *p, *n;
573 
574 	list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
575 		if (unlikely(p->ptrace & PT_EXITKILL))
576 			send_sig_info(SIGKILL, SEND_SIG_PRIV, p);
577 
578 		if (__ptrace_detach(tracer, p))
579 			list_add(&p->ptrace_entry, dead);
580 	}
581 }
582 
583 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
584 {
585 	int copied = 0;
586 
587 	while (len > 0) {
588 		char buf[128];
589 		int this_len, retval;
590 
591 		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
592 		retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
593 
594 		if (!retval) {
595 			if (copied)
596 				break;
597 			return -EIO;
598 		}
599 		if (copy_to_user(dst, buf, retval))
600 			return -EFAULT;
601 		copied += retval;
602 		src += retval;
603 		dst += retval;
604 		len -= retval;
605 	}
606 	return copied;
607 }
608 
609 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
610 {
611 	int copied = 0;
612 
613 	while (len > 0) {
614 		char buf[128];
615 		int this_len, retval;
616 
617 		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
618 		if (copy_from_user(buf, src, this_len))
619 			return -EFAULT;
620 		retval = ptrace_access_vm(tsk, dst, buf, this_len,
621 				FOLL_FORCE | FOLL_WRITE);
622 		if (!retval) {
623 			if (copied)
624 				break;
625 			return -EIO;
626 		}
627 		copied += retval;
628 		src += retval;
629 		dst += retval;
630 		len -= retval;
631 	}
632 	return copied;
633 }
634 
635 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
636 {
637 	unsigned flags;
638 
639 	if (data & ~(unsigned long)PTRACE_O_MASK)
640 		return -EINVAL;
641 
642 	if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
643 		if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
644 		    !IS_ENABLED(CONFIG_SECCOMP))
645 			return -EINVAL;
646 
647 		if (!capable(CAP_SYS_ADMIN))
648 			return -EPERM;
649 
650 		if (seccomp_mode(&current->seccomp) != SECCOMP_MODE_DISABLED ||
651 		    current->ptrace & PT_SUSPEND_SECCOMP)
652 			return -EPERM;
653 	}
654 
655 	/* Avoid intermediate state when all opts are cleared */
656 	flags = child->ptrace;
657 	flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
658 	flags |= (data << PT_OPT_FLAG_SHIFT);
659 	child->ptrace = flags;
660 
661 	return 0;
662 }
663 
664 static int ptrace_getsiginfo(struct task_struct *child, kernel_siginfo_t *info)
665 {
666 	unsigned long flags;
667 	int error = -ESRCH;
668 
669 	if (lock_task_sighand(child, &flags)) {
670 		error = -EINVAL;
671 		if (likely(child->last_siginfo != NULL)) {
672 			copy_siginfo(info, child->last_siginfo);
673 			error = 0;
674 		}
675 		unlock_task_sighand(child, &flags);
676 	}
677 	return error;
678 }
679 
680 static int ptrace_setsiginfo(struct task_struct *child, const kernel_siginfo_t *info)
681 {
682 	unsigned long flags;
683 	int error = -ESRCH;
684 
685 	if (lock_task_sighand(child, &flags)) {
686 		error = -EINVAL;
687 		if (likely(child->last_siginfo != NULL)) {
688 			copy_siginfo(child->last_siginfo, info);
689 			error = 0;
690 		}
691 		unlock_task_sighand(child, &flags);
692 	}
693 	return error;
694 }
695 
696 static int ptrace_peek_siginfo(struct task_struct *child,
697 				unsigned long addr,
698 				unsigned long data)
699 {
700 	struct ptrace_peeksiginfo_args arg;
701 	struct sigpending *pending;
702 	struct sigqueue *q;
703 	int ret, i;
704 
705 	ret = copy_from_user(&arg, (void __user *) addr,
706 				sizeof(struct ptrace_peeksiginfo_args));
707 	if (ret)
708 		return -EFAULT;
709 
710 	if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
711 		return -EINVAL; /* unknown flags */
712 
713 	if (arg.nr < 0)
714 		return -EINVAL;
715 
716 	/* Ensure arg.off fits in an unsigned long */
717 	if (arg.off > ULONG_MAX)
718 		return 0;
719 
720 	if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
721 		pending = &child->signal->shared_pending;
722 	else
723 		pending = &child->pending;
724 
725 	for (i = 0; i < arg.nr; ) {
726 		kernel_siginfo_t info;
727 		unsigned long off = arg.off + i;
728 		bool found = false;
729 
730 		spin_lock_irq(&child->sighand->siglock);
731 		list_for_each_entry(q, &pending->list, list) {
732 			if (!off--) {
733 				found = true;
734 				copy_siginfo(&info, &q->info);
735 				break;
736 			}
737 		}
738 		spin_unlock_irq(&child->sighand->siglock);
739 
740 		if (!found) /* beyond the end of the list */
741 			break;
742 
743 #ifdef CONFIG_COMPAT
744 		if (unlikely(in_compat_syscall())) {
745 			compat_siginfo_t __user *uinfo = compat_ptr(data);
746 
747 			if (copy_siginfo_to_user32(uinfo, &info)) {
748 				ret = -EFAULT;
749 				break;
750 			}
751 
752 		} else
753 #endif
754 		{
755 			siginfo_t __user *uinfo = (siginfo_t __user *) data;
756 
757 			if (copy_siginfo_to_user(uinfo, &info)) {
758 				ret = -EFAULT;
759 				break;
760 			}
761 		}
762 
763 		data += sizeof(siginfo_t);
764 		i++;
765 
766 		if (signal_pending(current))
767 			break;
768 
769 		cond_resched();
770 	}
771 
772 	if (i > 0)
773 		return i;
774 
775 	return ret;
776 }
777 
778 #ifdef PTRACE_SINGLESTEP
779 #define is_singlestep(request)		((request) == PTRACE_SINGLESTEP)
780 #else
781 #define is_singlestep(request)		0
782 #endif
783 
784 #ifdef PTRACE_SINGLEBLOCK
785 #define is_singleblock(request)		((request) == PTRACE_SINGLEBLOCK)
786 #else
787 #define is_singleblock(request)		0
788 #endif
789 
790 #ifdef PTRACE_SYSEMU
791 #define is_sysemu_singlestep(request)	((request) == PTRACE_SYSEMU_SINGLESTEP)
792 #else
793 #define is_sysemu_singlestep(request)	0
794 #endif
795 
796 static int ptrace_resume(struct task_struct *child, long request,
797 			 unsigned long data)
798 {
799 	bool need_siglock;
800 
801 	if (!valid_signal(data))
802 		return -EIO;
803 
804 	if (request == PTRACE_SYSCALL)
805 		set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
806 	else
807 		clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
808 
809 #ifdef TIF_SYSCALL_EMU
810 	if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
811 		set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
812 	else
813 		clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
814 #endif
815 
816 	if (is_singleblock(request)) {
817 		if (unlikely(!arch_has_block_step()))
818 			return -EIO;
819 		user_enable_block_step(child);
820 	} else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
821 		if (unlikely(!arch_has_single_step()))
822 			return -EIO;
823 		user_enable_single_step(child);
824 	} else {
825 		user_disable_single_step(child);
826 	}
827 
828 	/*
829 	 * Change ->exit_code and ->state under siglock to avoid the race
830 	 * with wait_task_stopped() in between; a non-zero ->exit_code will
831 	 * wrongly look like another report from tracee.
832 	 *
833 	 * Note that we need siglock even if ->exit_code == data and/or this
834 	 * status was not reported yet, the new status must not be cleared by
835 	 * wait_task_stopped() after resume.
836 	 *
837 	 * If data == 0 we do not care if wait_task_stopped() reports the old
838 	 * status and clears the code too; this can't race with the tracee, it
839 	 * takes siglock after resume.
840 	 */
841 	need_siglock = data && !thread_group_empty(current);
842 	if (need_siglock)
843 		spin_lock_irq(&child->sighand->siglock);
844 	child->exit_code = data;
845 	wake_up_state(child, __TASK_TRACED);
846 	if (need_siglock)
847 		spin_unlock_irq(&child->sighand->siglock);
848 
849 	return 0;
850 }
851 
852 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
853 
854 static const struct user_regset *
855 find_regset(const struct user_regset_view *view, unsigned int type)
856 {
857 	const struct user_regset *regset;
858 	int n;
859 
860 	for (n = 0; n < view->n; ++n) {
861 		regset = view->regsets + n;
862 		if (regset->core_note_type == type)
863 			return regset;
864 	}
865 
866 	return NULL;
867 }
868 
869 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
870 			 struct iovec *kiov)
871 {
872 	const struct user_regset_view *view = task_user_regset_view(task);
873 	const struct user_regset *regset = find_regset(view, type);
874 	int regset_no;
875 
876 	if (!regset || (kiov->iov_len % regset->size) != 0)
877 		return -EINVAL;
878 
879 	regset_no = regset - view->regsets;
880 	kiov->iov_len = min(kiov->iov_len,
881 			    (__kernel_size_t) (regset->n * regset->size));
882 
883 	if (req == PTRACE_GETREGSET)
884 		return copy_regset_to_user(task, view, regset_no, 0,
885 					   kiov->iov_len, kiov->iov_base);
886 	else
887 		return copy_regset_from_user(task, view, regset_no, 0,
888 					     kiov->iov_len, kiov->iov_base);
889 }
890 
891 /*
892  * This is declared in linux/regset.h and defined in machine-dependent
893  * code.  We put the export here, near the primary machine-neutral use,
894  * to ensure no machine forgets it.
895  */
896 EXPORT_SYMBOL_GPL(task_user_regset_view);
897 #endif
898 
899 int ptrace_request(struct task_struct *child, long request,
900 		   unsigned long addr, unsigned long data)
901 {
902 	bool seized = child->ptrace & PT_SEIZED;
903 	int ret = -EIO;
904 	kernel_siginfo_t siginfo, *si;
905 	void __user *datavp = (void __user *) data;
906 	unsigned long __user *datalp = datavp;
907 	unsigned long flags;
908 
909 	switch (request) {
910 	case PTRACE_PEEKTEXT:
911 	case PTRACE_PEEKDATA:
912 		return generic_ptrace_peekdata(child, addr, data);
913 	case PTRACE_POKETEXT:
914 	case PTRACE_POKEDATA:
915 		return generic_ptrace_pokedata(child, addr, data);
916 
917 #ifdef PTRACE_OLDSETOPTIONS
918 	case PTRACE_OLDSETOPTIONS:
919 #endif
920 	case PTRACE_SETOPTIONS:
921 		ret = ptrace_setoptions(child, data);
922 		break;
923 	case PTRACE_GETEVENTMSG:
924 		ret = put_user(child->ptrace_message, datalp);
925 		break;
926 
927 	case PTRACE_PEEKSIGINFO:
928 		ret = ptrace_peek_siginfo(child, addr, data);
929 		break;
930 
931 	case PTRACE_GETSIGINFO:
932 		ret = ptrace_getsiginfo(child, &siginfo);
933 		if (!ret)
934 			ret = copy_siginfo_to_user(datavp, &siginfo);
935 		break;
936 
937 	case PTRACE_SETSIGINFO:
938 		ret = copy_siginfo_from_user(&siginfo, datavp);
939 		if (!ret)
940 			ret = ptrace_setsiginfo(child, &siginfo);
941 		break;
942 
943 	case PTRACE_GETSIGMASK: {
944 		sigset_t *mask;
945 
946 		if (addr != sizeof(sigset_t)) {
947 			ret = -EINVAL;
948 			break;
949 		}
950 
951 		if (test_tsk_restore_sigmask(child))
952 			mask = &child->saved_sigmask;
953 		else
954 			mask = &child->blocked;
955 
956 		if (copy_to_user(datavp, mask, sizeof(sigset_t)))
957 			ret = -EFAULT;
958 		else
959 			ret = 0;
960 
961 		break;
962 	}
963 
964 	case PTRACE_SETSIGMASK: {
965 		sigset_t new_set;
966 
967 		if (addr != sizeof(sigset_t)) {
968 			ret = -EINVAL;
969 			break;
970 		}
971 
972 		if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
973 			ret = -EFAULT;
974 			break;
975 		}
976 
977 		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
978 
979 		/*
980 		 * Every thread does recalc_sigpending() after resume, so
981 		 * retarget_shared_pending() and recalc_sigpending() are not
982 		 * called here.
983 		 */
984 		spin_lock_irq(&child->sighand->siglock);
985 		child->blocked = new_set;
986 		spin_unlock_irq(&child->sighand->siglock);
987 
988 		clear_tsk_restore_sigmask(child);
989 
990 		ret = 0;
991 		break;
992 	}
993 
994 	case PTRACE_INTERRUPT:
995 		/*
996 		 * Stop tracee without any side-effect on signal or job
997 		 * control.  At least one trap is guaranteed to happen
998 		 * after this request.  If @child is already trapped, the
999 		 * current trap is not disturbed and another trap will
1000 		 * happen after the current trap is ended with PTRACE_CONT.
1001 		 *
1002 		 * The actual trap might not be PTRACE_EVENT_STOP trap but
1003 		 * the pending condition is cleared regardless.
1004 		 */
1005 		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1006 			break;
1007 
1008 		/*
1009 		 * INTERRUPT doesn't disturb existing trap sans one
1010 		 * exception.  If ptracer issued LISTEN for the current
1011 		 * STOP, this INTERRUPT should clear LISTEN and re-trap
1012 		 * tracee into STOP.
1013 		 */
1014 		if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
1015 			ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1016 
1017 		unlock_task_sighand(child, &flags);
1018 		ret = 0;
1019 		break;
1020 
1021 	case PTRACE_LISTEN:
1022 		/*
1023 		 * Listen for events.  Tracee must be in STOP.  It's not
1024 		 * resumed per-se but is not considered to be in TRACED by
1025 		 * wait(2) or ptrace(2).  If an async event (e.g. group
1026 		 * stop state change) happens, tracee will enter STOP trap
1027 		 * again.  Alternatively, ptracer can issue INTERRUPT to
1028 		 * finish listening and re-trap tracee into STOP.
1029 		 */
1030 		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1031 			break;
1032 
1033 		si = child->last_siginfo;
1034 		if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1035 			child->jobctl |= JOBCTL_LISTENING;
1036 			/*
1037 			 * If NOTIFY is set, it means event happened between
1038 			 * start of this trap and now.  Trigger re-trap.
1039 			 */
1040 			if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1041 				ptrace_signal_wake_up(child, true);
1042 			ret = 0;
1043 		}
1044 		unlock_task_sighand(child, &flags);
1045 		break;
1046 
1047 	case PTRACE_DETACH:	 /* detach a process that was attached. */
1048 		ret = ptrace_detach(child, data);
1049 		break;
1050 
1051 #ifdef CONFIG_BINFMT_ELF_FDPIC
1052 	case PTRACE_GETFDPIC: {
1053 		struct mm_struct *mm = get_task_mm(child);
1054 		unsigned long tmp = 0;
1055 
1056 		ret = -ESRCH;
1057 		if (!mm)
1058 			break;
1059 
1060 		switch (addr) {
1061 		case PTRACE_GETFDPIC_EXEC:
1062 			tmp = mm->context.exec_fdpic_loadmap;
1063 			break;
1064 		case PTRACE_GETFDPIC_INTERP:
1065 			tmp = mm->context.interp_fdpic_loadmap;
1066 			break;
1067 		default:
1068 			break;
1069 		}
1070 		mmput(mm);
1071 
1072 		ret = put_user(tmp, datalp);
1073 		break;
1074 	}
1075 #endif
1076 
1077 #ifdef PTRACE_SINGLESTEP
1078 	case PTRACE_SINGLESTEP:
1079 #endif
1080 #ifdef PTRACE_SINGLEBLOCK
1081 	case PTRACE_SINGLEBLOCK:
1082 #endif
1083 #ifdef PTRACE_SYSEMU
1084 	case PTRACE_SYSEMU:
1085 	case PTRACE_SYSEMU_SINGLESTEP:
1086 #endif
1087 	case PTRACE_SYSCALL:
1088 	case PTRACE_CONT:
1089 		return ptrace_resume(child, request, data);
1090 
1091 	case PTRACE_KILL:
1092 		if (child->exit_state)	/* already dead */
1093 			return 0;
1094 		return ptrace_resume(child, request, SIGKILL);
1095 
1096 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1097 	case PTRACE_GETREGSET:
1098 	case PTRACE_SETREGSET: {
1099 		struct iovec kiov;
1100 		struct iovec __user *uiov = datavp;
1101 
1102 		if (!access_ok(uiov, sizeof(*uiov)))
1103 			return -EFAULT;
1104 
1105 		if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1106 		    __get_user(kiov.iov_len, &uiov->iov_len))
1107 			return -EFAULT;
1108 
1109 		ret = ptrace_regset(child, request, addr, &kiov);
1110 		if (!ret)
1111 			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1112 		break;
1113 	}
1114 #endif
1115 
1116 	case PTRACE_SECCOMP_GET_FILTER:
1117 		ret = seccomp_get_filter(child, addr, datavp);
1118 		break;
1119 
1120 	case PTRACE_SECCOMP_GET_METADATA:
1121 		ret = seccomp_get_metadata(child, addr, datavp);
1122 		break;
1123 
1124 	default:
1125 		break;
1126 	}
1127 
1128 	return ret;
1129 }
1130 
1131 #ifndef arch_ptrace_attach
1132 #define arch_ptrace_attach(child)	do { } while (0)
1133 #endif
1134 
1135 SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1136 		unsigned long, data)
1137 {
1138 	struct task_struct *child;
1139 	long ret;
1140 
1141 	if (request == PTRACE_TRACEME) {
1142 		ret = ptrace_traceme();
1143 		if (!ret)
1144 			arch_ptrace_attach(current);
1145 		goto out;
1146 	}
1147 
1148 	child = find_get_task_by_vpid(pid);
1149 	if (!child) {
1150 		ret = -ESRCH;
1151 		goto out;
1152 	}
1153 
1154 	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1155 		ret = ptrace_attach(child, request, addr, data);
1156 		/*
1157 		 * Some architectures need to do book-keeping after
1158 		 * a ptrace attach.
1159 		 */
1160 		if (!ret)
1161 			arch_ptrace_attach(child);
1162 		goto out_put_task_struct;
1163 	}
1164 
1165 	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1166 				  request == PTRACE_INTERRUPT);
1167 	if (ret < 0)
1168 		goto out_put_task_struct;
1169 
1170 	ret = arch_ptrace(child, request, addr, data);
1171 	if (ret || request != PTRACE_DETACH)
1172 		ptrace_unfreeze_traced(child);
1173 
1174  out_put_task_struct:
1175 	put_task_struct(child);
1176  out:
1177 	return ret;
1178 }
1179 
1180 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1181 			    unsigned long data)
1182 {
1183 	unsigned long tmp;
1184 	int copied;
1185 
1186 	copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1187 	if (copied != sizeof(tmp))
1188 		return -EIO;
1189 	return put_user(tmp, (unsigned long __user *)data);
1190 }
1191 
1192 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1193 			    unsigned long data)
1194 {
1195 	int copied;
1196 
1197 	copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1198 			FOLL_FORCE | FOLL_WRITE);
1199 	return (copied == sizeof(data)) ? 0 : -EIO;
1200 }
1201 
1202 #if defined CONFIG_COMPAT
1203 
1204 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1205 			  compat_ulong_t addr, compat_ulong_t data)
1206 {
1207 	compat_ulong_t __user *datap = compat_ptr(data);
1208 	compat_ulong_t word;
1209 	kernel_siginfo_t siginfo;
1210 	int ret;
1211 
1212 	switch (request) {
1213 	case PTRACE_PEEKTEXT:
1214 	case PTRACE_PEEKDATA:
1215 		ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1216 				FOLL_FORCE);
1217 		if (ret != sizeof(word))
1218 			ret = -EIO;
1219 		else
1220 			ret = put_user(word, datap);
1221 		break;
1222 
1223 	case PTRACE_POKETEXT:
1224 	case PTRACE_POKEDATA:
1225 		ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1226 				FOLL_FORCE | FOLL_WRITE);
1227 		ret = (ret != sizeof(data) ? -EIO : 0);
1228 		break;
1229 
1230 	case PTRACE_GETEVENTMSG:
1231 		ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1232 		break;
1233 
1234 	case PTRACE_GETSIGINFO:
1235 		ret = ptrace_getsiginfo(child, &siginfo);
1236 		if (!ret)
1237 			ret = copy_siginfo_to_user32(
1238 				(struct compat_siginfo __user *) datap,
1239 				&siginfo);
1240 		break;
1241 
1242 	case PTRACE_SETSIGINFO:
1243 		ret = copy_siginfo_from_user32(
1244 			&siginfo, (struct compat_siginfo __user *) datap);
1245 		if (!ret)
1246 			ret = ptrace_setsiginfo(child, &siginfo);
1247 		break;
1248 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1249 	case PTRACE_GETREGSET:
1250 	case PTRACE_SETREGSET:
1251 	{
1252 		struct iovec kiov;
1253 		struct compat_iovec __user *uiov =
1254 			(struct compat_iovec __user *) datap;
1255 		compat_uptr_t ptr;
1256 		compat_size_t len;
1257 
1258 		if (!access_ok(uiov, sizeof(*uiov)))
1259 			return -EFAULT;
1260 
1261 		if (__get_user(ptr, &uiov->iov_base) ||
1262 		    __get_user(len, &uiov->iov_len))
1263 			return -EFAULT;
1264 
1265 		kiov.iov_base = compat_ptr(ptr);
1266 		kiov.iov_len = len;
1267 
1268 		ret = ptrace_regset(child, request, addr, &kiov);
1269 		if (!ret)
1270 			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1271 		break;
1272 	}
1273 #endif
1274 
1275 	default:
1276 		ret = ptrace_request(child, request, addr, data);
1277 	}
1278 
1279 	return ret;
1280 }
1281 
1282 COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1283 		       compat_long_t, addr, compat_long_t, data)
1284 {
1285 	struct task_struct *child;
1286 	long ret;
1287 
1288 	if (request == PTRACE_TRACEME) {
1289 		ret = ptrace_traceme();
1290 		goto out;
1291 	}
1292 
1293 	child = find_get_task_by_vpid(pid);
1294 	if (!child) {
1295 		ret = -ESRCH;
1296 		goto out;
1297 	}
1298 
1299 	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1300 		ret = ptrace_attach(child, request, addr, data);
1301 		/*
1302 		 * Some architectures need to do book-keeping after
1303 		 * a ptrace attach.
1304 		 */
1305 		if (!ret)
1306 			arch_ptrace_attach(child);
1307 		goto out_put_task_struct;
1308 	}
1309 
1310 	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1311 				  request == PTRACE_INTERRUPT);
1312 	if (!ret) {
1313 		ret = compat_arch_ptrace(child, request, addr, data);
1314 		if (ret || request != PTRACE_DETACH)
1315 			ptrace_unfreeze_traced(child);
1316 	}
1317 
1318  out_put_task_struct:
1319 	put_task_struct(child);
1320  out:
1321 	return ret;
1322 }
1323 #endif	/* CONFIG_COMPAT */
1324