Loading...
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
31static int ptrace_trapping_sleep_fn(void *flags)
32{
33 schedule();
34 return 0;
35}
36
37/*
38 * ptrace a task: make the debugger its new parent and
39 * move it to the ptrace list.
40 *
41 * Must be called with the tasklist lock write-held.
42 */
43void __ptrace_link(struct task_struct *child, struct task_struct *new_parent)
44{
45 BUG_ON(!list_empty(&child->ptrace_entry));
46 list_add(&child->ptrace_entry, &new_parent->ptraced);
47 child->parent = new_parent;
48}
49
50/**
51 * __ptrace_unlink - unlink ptracee and restore its execution state
52 * @child: ptracee to be unlinked
53 *
54 * Remove @child from the ptrace list, move it back to the original parent,
55 * and restore the execution state so that it conforms to the group stop
56 * state.
57 *
58 * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
59 * exiting. For PTRACE_DETACH, unless the ptracee has been killed between
60 * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
61 * If the ptracer is exiting, the ptracee can be in any state.
62 *
63 * After detach, the ptracee should be in a state which conforms to the
64 * group stop. If the group is stopped or in the process of stopping, the
65 * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
66 * up from TASK_TRACED.
67 *
68 * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
69 * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
70 * to but in the opposite direction of what happens while attaching to a
71 * stopped task. However, in this direction, the intermediate RUNNING
72 * state is not hidden even from the current ptracer and if it immediately
73 * re-attaches and performs a WNOHANG wait(2), it may fail.
74 *
75 * CONTEXT:
76 * write_lock_irq(tasklist_lock)
77 */
78void __ptrace_unlink(struct task_struct *child)
79{
80 BUG_ON(!child->ptrace);
81
82 child->ptrace = 0;
83 child->parent = child->real_parent;
84 list_del_init(&child->ptrace_entry);
85
86 spin_lock(&child->sighand->siglock);
87
88 /*
89 * Clear all pending traps and TRAPPING. TRAPPING should be
90 * cleared regardless of JOBCTL_STOP_PENDING. Do it explicitly.
91 */
92 task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
93 task_clear_jobctl_trapping(child);
94
95 /*
96 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
97 * @child isn't dead.
98 */
99 if (!(child->flags & PF_EXITING) &&
100 (child->signal->flags & SIGNAL_STOP_STOPPED ||
101 child->signal->group_stop_count)) {
102 child->jobctl |= JOBCTL_STOP_PENDING;
103
104 /*
105 * This is only possible if this thread was cloned by the
106 * traced task running in the stopped group, set the signal
107 * for the future reports.
108 * FIXME: we should change ptrace_init_task() to handle this
109 * case.
110 */
111 if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
112 child->jobctl |= SIGSTOP;
113 }
114
115 /*
116 * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
117 * @child in the butt. Note that @resume should be used iff @child
118 * is in TASK_TRACED; otherwise, we might unduly disrupt
119 * TASK_KILLABLE sleeps.
120 */
121 if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
122 ptrace_signal_wake_up(child, true);
123
124 spin_unlock(&child->sighand->siglock);
125}
126
127/* Ensure that nothing can wake it up, even SIGKILL */
128static bool ptrace_freeze_traced(struct task_struct *task)
129{
130 bool ret = false;
131
132 /* Lockless, nobody but us can set this flag */
133 if (task->jobctl & JOBCTL_LISTENING)
134 return ret;
135
136 spin_lock_irq(&task->sighand->siglock);
137 if (task_is_traced(task) && !__fatal_signal_pending(task)) {
138 task->state = __TASK_TRACED;
139 ret = true;
140 }
141 spin_unlock_irq(&task->sighand->siglock);
142
143 return ret;
144}
145
146static void ptrace_unfreeze_traced(struct task_struct *task)
147{
148 if (task->state != __TASK_TRACED)
149 return;
150
151 WARN_ON(!task->ptrace || task->parent != current);
152
153 spin_lock_irq(&task->sighand->siglock);
154 if (__fatal_signal_pending(task))
155 wake_up_state(task, __TASK_TRACED);
156 else
157 task->state = TASK_TRACED;
158 spin_unlock_irq(&task->sighand->siglock);
159}
160
161/**
162 * ptrace_check_attach - check whether ptracee is ready for ptrace operation
163 * @child: ptracee to check for
164 * @ignore_state: don't check whether @child is currently %TASK_TRACED
165 *
166 * Check whether @child is being ptraced by %current and ready for further
167 * ptrace operations. If @ignore_state is %false, @child also should be in
168 * %TASK_TRACED state and on return the child is guaranteed to be traced
169 * and not executing. If @ignore_state is %true, @child can be in any
170 * state.
171 *
172 * CONTEXT:
173 * Grabs and releases tasklist_lock and @child->sighand->siglock.
174 *
175 * RETURNS:
176 * 0 on success, -ESRCH if %child is not ready.
177 */
178static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
179{
180 int ret = -ESRCH;
181
182 /*
183 * We take the read lock around doing both checks to close a
184 * possible race where someone else was tracing our child and
185 * detached between these two checks. After this locked check,
186 * we are sure that this is our traced child and that can only
187 * be changed by us so it's not changing right after this.
188 */
189 read_lock(&tasklist_lock);
190 if (child->ptrace && child->parent == current) {
191 WARN_ON(child->state == __TASK_TRACED);
192 /*
193 * child->sighand can't be NULL, release_task()
194 * does ptrace_unlink() before __exit_signal().
195 */
196 if (ignore_state || ptrace_freeze_traced(child))
197 ret = 0;
198 }
199 read_unlock(&tasklist_lock);
200
201 if (!ret && !ignore_state) {
202 if (!wait_task_inactive(child, __TASK_TRACED)) {
203 /*
204 * This can only happen if may_ptrace_stop() fails and
205 * ptrace_stop() changes ->state back to TASK_RUNNING,
206 * so we should not worry about leaking __TASK_TRACED.
207 */
208 WARN_ON(child->state == __TASK_TRACED);
209 ret = -ESRCH;
210 }
211 }
212
213 return ret;
214}
215
216static int ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
217{
218 if (mode & PTRACE_MODE_NOAUDIT)
219 return has_ns_capability_noaudit(current, ns, CAP_SYS_PTRACE);
220 else
221 return has_ns_capability(current, ns, CAP_SYS_PTRACE);
222}
223
224/* Returns 0 on success, -errno on denial. */
225static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
226{
227 const struct cred *cred = current_cred(), *tcred;
228
229 /* May we inspect the given task?
230 * This check is used both for attaching with ptrace
231 * and for allowing access to sensitive information in /proc.
232 *
233 * ptrace_attach denies several cases that /proc allows
234 * because setting up the necessary parent/child relationship
235 * or halting the specified task is impossible.
236 */
237 int dumpable = 0;
238 /* Don't let security modules deny introspection */
239 if (same_thread_group(task, current))
240 return 0;
241 rcu_read_lock();
242 tcred = __task_cred(task);
243 if (uid_eq(cred->uid, tcred->euid) &&
244 uid_eq(cred->uid, tcred->suid) &&
245 uid_eq(cred->uid, tcred->uid) &&
246 gid_eq(cred->gid, tcred->egid) &&
247 gid_eq(cred->gid, tcred->sgid) &&
248 gid_eq(cred->gid, tcred->gid))
249 goto ok;
250 if (ptrace_has_cap(tcred->user_ns, mode))
251 goto ok;
252 rcu_read_unlock();
253 return -EPERM;
254ok:
255 rcu_read_unlock();
256 smp_rmb();
257 if (task->mm)
258 dumpable = get_dumpable(task->mm);
259 rcu_read_lock();
260 if (dumpable != SUID_DUMP_USER &&
261 !ptrace_has_cap(__task_cred(task)->user_ns, mode)) {
262 rcu_read_unlock();
263 return -EPERM;
264 }
265 rcu_read_unlock();
266
267 return security_ptrace_access_check(task, mode);
268}
269
270bool ptrace_may_access(struct task_struct *task, unsigned int mode)
271{
272 int err;
273 task_lock(task);
274 err = __ptrace_may_access(task, mode);
275 task_unlock(task);
276 return !err;
277}
278
279static int ptrace_attach(struct task_struct *task, long request,
280 unsigned long addr,
281 unsigned long flags)
282{
283 bool seize = (request == PTRACE_SEIZE);
284 int retval;
285
286 retval = -EIO;
287 if (seize) {
288 if (addr != 0)
289 goto out;
290 if (flags & ~(unsigned long)PTRACE_O_MASK)
291 goto out;
292 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
293 } else {
294 flags = PT_PTRACED;
295 }
296
297 audit_ptrace(task);
298
299 retval = -EPERM;
300 if (unlikely(task->flags & PF_KTHREAD))
301 goto out;
302 if (same_thread_group(task, current))
303 goto out;
304
305 /*
306 * Protect exec's credential calculations against our interference;
307 * SUID, SGID and LSM creds get determined differently
308 * under ptrace.
309 */
310 retval = -ERESTARTNOINTR;
311 if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
312 goto out;
313
314 task_lock(task);
315 retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH);
316 task_unlock(task);
317 if (retval)
318 goto unlock_creds;
319
320 write_lock_irq(&tasklist_lock);
321 retval = -EPERM;
322 if (unlikely(task->exit_state))
323 goto unlock_tasklist;
324 if (task->ptrace)
325 goto unlock_tasklist;
326
327 if (seize)
328 flags |= PT_SEIZED;
329 rcu_read_lock();
330 if (ns_capable(__task_cred(task)->user_ns, CAP_SYS_PTRACE))
331 flags |= PT_PTRACE_CAP;
332 rcu_read_unlock();
333 task->ptrace = flags;
334
335 __ptrace_link(task, current);
336
337 /* SEIZE doesn't trap tracee on attach */
338 if (!seize)
339 send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
340
341 spin_lock(&task->sighand->siglock);
342
343 /*
344 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
345 * TRAPPING, and kick it so that it transits to TRACED. TRAPPING
346 * will be cleared if the child completes the transition or any
347 * event which clears the group stop states happens. We'll wait
348 * for the transition to complete before returning from this
349 * function.
350 *
351 * This hides STOPPED -> RUNNING -> TRACED transition from the
352 * attaching thread but a different thread in the same group can
353 * still observe the transient RUNNING state. IOW, if another
354 * thread's WNOHANG wait(2) on the stopped tracee races against
355 * ATTACH, the wait(2) may fail due to the transient RUNNING.
356 *
357 * The following task_is_stopped() test is safe as both transitions
358 * in and out of STOPPED are protected by siglock.
359 */
360 if (task_is_stopped(task) &&
361 task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
362 signal_wake_up_state(task, __TASK_STOPPED);
363
364 spin_unlock(&task->sighand->siglock);
365
366 retval = 0;
367unlock_tasklist:
368 write_unlock_irq(&tasklist_lock);
369unlock_creds:
370 mutex_unlock(&task->signal->cred_guard_mutex);
371out:
372 if (!retval) {
373 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT,
374 ptrace_trapping_sleep_fn, TASK_UNINTERRUPTIBLE);
375 proc_ptrace_connector(task, PTRACE_ATTACH);
376 }
377
378 return retval;
379}
380
381/**
382 * ptrace_traceme -- helper for PTRACE_TRACEME
383 *
384 * Performs checks and sets PT_PTRACED.
385 * Should be used by all ptrace implementations for PTRACE_TRACEME.
386 */
387static int ptrace_traceme(void)
388{
389 int ret = -EPERM;
390
391 write_lock_irq(&tasklist_lock);
392 /* Are we already being traced? */
393 if (!current->ptrace) {
394 ret = security_ptrace_traceme(current->parent);
395 /*
396 * Check PF_EXITING to ensure ->real_parent has not passed
397 * exit_ptrace(). Otherwise we don't report the error but
398 * pretend ->real_parent untraces us right after return.
399 */
400 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
401 current->ptrace = PT_PTRACED;
402 __ptrace_link(current, current->real_parent);
403 }
404 }
405 write_unlock_irq(&tasklist_lock);
406
407 return ret;
408}
409
410/*
411 * Called with irqs disabled, returns true if childs should reap themselves.
412 */
413static int ignoring_children(struct sighand_struct *sigh)
414{
415 int ret;
416 spin_lock(&sigh->siglock);
417 ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
418 (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
419 spin_unlock(&sigh->siglock);
420 return ret;
421}
422
423/*
424 * Called with tasklist_lock held for writing.
425 * Unlink a traced task, and clean it up if it was a traced zombie.
426 * Return true if it needs to be reaped with release_task().
427 * (We can't call release_task() here because we already hold tasklist_lock.)
428 *
429 * If it's a zombie, our attachedness prevented normal parent notification
430 * or self-reaping. Do notification now if it would have happened earlier.
431 * If it should reap itself, return true.
432 *
433 * If it's our own child, there is no notification to do. But if our normal
434 * children self-reap, then this child was prevented by ptrace and we must
435 * reap it now, in that case we must also wake up sub-threads sleeping in
436 * do_wait().
437 */
438static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
439{
440 bool dead;
441
442 __ptrace_unlink(p);
443
444 if (p->exit_state != EXIT_ZOMBIE)
445 return false;
446
447 dead = !thread_group_leader(p);
448
449 if (!dead && thread_group_empty(p)) {
450 if (!same_thread_group(p->real_parent, tracer))
451 dead = do_notify_parent(p, p->exit_signal);
452 else if (ignoring_children(tracer->sighand)) {
453 __wake_up_parent(p, tracer);
454 dead = true;
455 }
456 }
457 /* Mark it as in the process of being reaped. */
458 if (dead)
459 p->exit_state = EXIT_DEAD;
460 return dead;
461}
462
463static int ptrace_detach(struct task_struct *child, unsigned int data)
464{
465 bool dead = false;
466
467 if (!valid_signal(data))
468 return -EIO;
469
470 /* Architecture-specific hardware disable .. */
471 ptrace_disable(child);
472 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
473
474 write_lock_irq(&tasklist_lock);
475 /*
476 * This child can be already killed. Make sure de_thread() or
477 * our sub-thread doing do_wait() didn't do release_task() yet.
478 */
479 if (child->ptrace) {
480 child->exit_code = data;
481 dead = __ptrace_detach(current, child);
482 }
483 write_unlock_irq(&tasklist_lock);
484
485 proc_ptrace_connector(child, PTRACE_DETACH);
486 if (unlikely(dead))
487 release_task(child);
488
489 return 0;
490}
491
492/*
493 * Detach all tasks we were using ptrace on. Called with tasklist held
494 * for writing, and returns with it held too. But note it can release
495 * and reacquire the lock.
496 */
497void exit_ptrace(struct task_struct *tracer)
498 __releases(&tasklist_lock)
499 __acquires(&tasklist_lock)
500{
501 struct task_struct *p, *n;
502 LIST_HEAD(ptrace_dead);
503
504 if (likely(list_empty(&tracer->ptraced)))
505 return;
506
507 list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
508 if (unlikely(p->ptrace & PT_EXITKILL))
509 send_sig_info(SIGKILL, SEND_SIG_FORCED, p);
510
511 if (__ptrace_detach(tracer, p))
512 list_add(&p->ptrace_entry, &ptrace_dead);
513 }
514
515 write_unlock_irq(&tasklist_lock);
516 BUG_ON(!list_empty(&tracer->ptraced));
517
518 list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_entry) {
519 list_del_init(&p->ptrace_entry);
520 release_task(p);
521 }
522
523 write_lock_irq(&tasklist_lock);
524}
525
526int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
527{
528 int copied = 0;
529
530 while (len > 0) {
531 char buf[128];
532 int this_len, retval;
533
534 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
535 retval = access_process_vm(tsk, src, buf, this_len, 0);
536 if (!retval) {
537 if (copied)
538 break;
539 return -EIO;
540 }
541 if (copy_to_user(dst, buf, retval))
542 return -EFAULT;
543 copied += retval;
544 src += retval;
545 dst += retval;
546 len -= retval;
547 }
548 return copied;
549}
550
551int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
552{
553 int copied = 0;
554
555 while (len > 0) {
556 char buf[128];
557 int this_len, retval;
558
559 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
560 if (copy_from_user(buf, src, this_len))
561 return -EFAULT;
562 retval = access_process_vm(tsk, dst, buf, this_len, 1);
563 if (!retval) {
564 if (copied)
565 break;
566 return -EIO;
567 }
568 copied += retval;
569 src += retval;
570 dst += retval;
571 len -= retval;
572 }
573 return copied;
574}
575
576static int ptrace_setoptions(struct task_struct *child, unsigned long data)
577{
578 unsigned flags;
579
580 if (data & ~(unsigned long)PTRACE_O_MASK)
581 return -EINVAL;
582
583 /* Avoid intermediate state when all opts are cleared */
584 flags = child->ptrace;
585 flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
586 flags |= (data << PT_OPT_FLAG_SHIFT);
587 child->ptrace = flags;
588
589 return 0;
590}
591
592static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
593{
594 unsigned long flags;
595 int error = -ESRCH;
596
597 if (lock_task_sighand(child, &flags)) {
598 error = -EINVAL;
599 if (likely(child->last_siginfo != NULL)) {
600 *info = *child->last_siginfo;
601 error = 0;
602 }
603 unlock_task_sighand(child, &flags);
604 }
605 return error;
606}
607
608static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
609{
610 unsigned long flags;
611 int error = -ESRCH;
612
613 if (lock_task_sighand(child, &flags)) {
614 error = -EINVAL;
615 if (likely(child->last_siginfo != NULL)) {
616 *child->last_siginfo = *info;
617 error = 0;
618 }
619 unlock_task_sighand(child, &flags);
620 }
621 return error;
622}
623
624static int ptrace_peek_siginfo(struct task_struct *child,
625 unsigned long addr,
626 unsigned long data)
627{
628 struct ptrace_peeksiginfo_args arg;
629 struct sigpending *pending;
630 struct sigqueue *q;
631 int ret, i;
632
633 ret = copy_from_user(&arg, (void __user *) addr,
634 sizeof(struct ptrace_peeksiginfo_args));
635 if (ret)
636 return -EFAULT;
637
638 if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
639 return -EINVAL; /* unknown flags */
640
641 if (arg.nr < 0)
642 return -EINVAL;
643
644 if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
645 pending = &child->signal->shared_pending;
646 else
647 pending = &child->pending;
648
649 for (i = 0; i < arg.nr; ) {
650 siginfo_t info;
651 s32 off = arg.off + i;
652
653 spin_lock_irq(&child->sighand->siglock);
654 list_for_each_entry(q, &pending->list, list) {
655 if (!off--) {
656 copy_siginfo(&info, &q->info);
657 break;
658 }
659 }
660 spin_unlock_irq(&child->sighand->siglock);
661
662 if (off >= 0) /* beyond the end of the list */
663 break;
664
665#ifdef CONFIG_COMPAT
666 if (unlikely(is_compat_task())) {
667 compat_siginfo_t __user *uinfo = compat_ptr(data);
668
669 if (copy_siginfo_to_user32(uinfo, &info) ||
670 __put_user(info.si_code, &uinfo->si_code)) {
671 ret = -EFAULT;
672 break;
673 }
674
675 } else
676#endif
677 {
678 siginfo_t __user *uinfo = (siginfo_t __user *) data;
679
680 if (copy_siginfo_to_user(uinfo, &info) ||
681 __put_user(info.si_code, &uinfo->si_code)) {
682 ret = -EFAULT;
683 break;
684 }
685 }
686
687 data += sizeof(siginfo_t);
688 i++;
689
690 if (signal_pending(current))
691 break;
692
693 cond_resched();
694 }
695
696 if (i > 0)
697 return i;
698
699 return ret;
700}
701
702#ifdef PTRACE_SINGLESTEP
703#define is_singlestep(request) ((request) == PTRACE_SINGLESTEP)
704#else
705#define is_singlestep(request) 0
706#endif
707
708#ifdef PTRACE_SINGLEBLOCK
709#define is_singleblock(request) ((request) == PTRACE_SINGLEBLOCK)
710#else
711#define is_singleblock(request) 0
712#endif
713
714#ifdef PTRACE_SYSEMU
715#define is_sysemu_singlestep(request) ((request) == PTRACE_SYSEMU_SINGLESTEP)
716#else
717#define is_sysemu_singlestep(request) 0
718#endif
719
720static int ptrace_resume(struct task_struct *child, long request,
721 unsigned long data)
722{
723 if (!valid_signal(data))
724 return -EIO;
725
726 if (request == PTRACE_SYSCALL)
727 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
728 else
729 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
730
731#ifdef TIF_SYSCALL_EMU
732 if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
733 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
734 else
735 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
736#endif
737
738 if (is_singleblock(request)) {
739 if (unlikely(!arch_has_block_step()))
740 return -EIO;
741 user_enable_block_step(child);
742 } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
743 if (unlikely(!arch_has_single_step()))
744 return -EIO;
745 user_enable_single_step(child);
746 } else {
747 user_disable_single_step(child);
748 }
749
750 child->exit_code = data;
751 wake_up_state(child, __TASK_TRACED);
752
753 return 0;
754}
755
756#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
757
758static const struct user_regset *
759find_regset(const struct user_regset_view *view, unsigned int type)
760{
761 const struct user_regset *regset;
762 int n;
763
764 for (n = 0; n < view->n; ++n) {
765 regset = view->regsets + n;
766 if (regset->core_note_type == type)
767 return regset;
768 }
769
770 return NULL;
771}
772
773static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
774 struct iovec *kiov)
775{
776 const struct user_regset_view *view = task_user_regset_view(task);
777 const struct user_regset *regset = find_regset(view, type);
778 int regset_no;
779
780 if (!regset || (kiov->iov_len % regset->size) != 0)
781 return -EINVAL;
782
783 regset_no = regset - view->regsets;
784 kiov->iov_len = min(kiov->iov_len,
785 (__kernel_size_t) (regset->n * regset->size));
786
787 if (req == PTRACE_GETREGSET)
788 return copy_regset_to_user(task, view, regset_no, 0,
789 kiov->iov_len, kiov->iov_base);
790 else
791 return copy_regset_from_user(task, view, regset_no, 0,
792 kiov->iov_len, kiov->iov_base);
793}
794
795/*
796 * This is declared in linux/regset.h and defined in machine-dependent
797 * code. We put the export here, near the primary machine-neutral use,
798 * to ensure no machine forgets it.
799 */
800EXPORT_SYMBOL_GPL(task_user_regset_view);
801#endif
802
803int ptrace_request(struct task_struct *child, long request,
804 unsigned long addr, unsigned long data)
805{
806 bool seized = child->ptrace & PT_SEIZED;
807 int ret = -EIO;
808 siginfo_t siginfo, *si;
809 void __user *datavp = (void __user *) data;
810 unsigned long __user *datalp = datavp;
811 unsigned long flags;
812
813 switch (request) {
814 case PTRACE_PEEKTEXT:
815 case PTRACE_PEEKDATA:
816 return generic_ptrace_peekdata(child, addr, data);
817 case PTRACE_POKETEXT:
818 case PTRACE_POKEDATA:
819 return generic_ptrace_pokedata(child, addr, data);
820
821#ifdef PTRACE_OLDSETOPTIONS
822 case PTRACE_OLDSETOPTIONS:
823#endif
824 case PTRACE_SETOPTIONS:
825 ret = ptrace_setoptions(child, data);
826 break;
827 case PTRACE_GETEVENTMSG:
828 ret = put_user(child->ptrace_message, datalp);
829 break;
830
831 case PTRACE_PEEKSIGINFO:
832 ret = ptrace_peek_siginfo(child, addr, data);
833 break;
834
835 case PTRACE_GETSIGINFO:
836 ret = ptrace_getsiginfo(child, &siginfo);
837 if (!ret)
838 ret = copy_siginfo_to_user(datavp, &siginfo);
839 break;
840
841 case PTRACE_SETSIGINFO:
842 if (copy_from_user(&siginfo, datavp, sizeof siginfo))
843 ret = -EFAULT;
844 else
845 ret = ptrace_setsiginfo(child, &siginfo);
846 break;
847
848 case PTRACE_GETSIGMASK:
849 if (addr != sizeof(sigset_t)) {
850 ret = -EINVAL;
851 break;
852 }
853
854 if (copy_to_user(datavp, &child->blocked, sizeof(sigset_t)))
855 ret = -EFAULT;
856 else
857 ret = 0;
858
859 break;
860
861 case PTRACE_SETSIGMASK: {
862 sigset_t new_set;
863
864 if (addr != sizeof(sigset_t)) {
865 ret = -EINVAL;
866 break;
867 }
868
869 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
870 ret = -EFAULT;
871 break;
872 }
873
874 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
875
876 /*
877 * Every thread does recalc_sigpending() after resume, so
878 * retarget_shared_pending() and recalc_sigpending() are not
879 * called here.
880 */
881 spin_lock_irq(&child->sighand->siglock);
882 child->blocked = new_set;
883 spin_unlock_irq(&child->sighand->siglock);
884
885 ret = 0;
886 break;
887 }
888
889 case PTRACE_INTERRUPT:
890 /*
891 * Stop tracee without any side-effect on signal or job
892 * control. At least one trap is guaranteed to happen
893 * after this request. If @child is already trapped, the
894 * current trap is not disturbed and another trap will
895 * happen after the current trap is ended with PTRACE_CONT.
896 *
897 * The actual trap might not be PTRACE_EVENT_STOP trap but
898 * the pending condition is cleared regardless.
899 */
900 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
901 break;
902
903 /*
904 * INTERRUPT doesn't disturb existing trap sans one
905 * exception. If ptracer issued LISTEN for the current
906 * STOP, this INTERRUPT should clear LISTEN and re-trap
907 * tracee into STOP.
908 */
909 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
910 ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
911
912 unlock_task_sighand(child, &flags);
913 ret = 0;
914 break;
915
916 case PTRACE_LISTEN:
917 /*
918 * Listen for events. Tracee must be in STOP. It's not
919 * resumed per-se but is not considered to be in TRACED by
920 * wait(2) or ptrace(2). If an async event (e.g. group
921 * stop state change) happens, tracee will enter STOP trap
922 * again. Alternatively, ptracer can issue INTERRUPT to
923 * finish listening and re-trap tracee into STOP.
924 */
925 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
926 break;
927
928 si = child->last_siginfo;
929 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
930 child->jobctl |= JOBCTL_LISTENING;
931 /*
932 * If NOTIFY is set, it means event happened between
933 * start of this trap and now. Trigger re-trap.
934 */
935 if (child->jobctl & JOBCTL_TRAP_NOTIFY)
936 ptrace_signal_wake_up(child, true);
937 ret = 0;
938 }
939 unlock_task_sighand(child, &flags);
940 break;
941
942 case PTRACE_DETACH: /* detach a process that was attached. */
943 ret = ptrace_detach(child, data);
944 break;
945
946#ifdef CONFIG_BINFMT_ELF_FDPIC
947 case PTRACE_GETFDPIC: {
948 struct mm_struct *mm = get_task_mm(child);
949 unsigned long tmp = 0;
950
951 ret = -ESRCH;
952 if (!mm)
953 break;
954
955 switch (addr) {
956 case PTRACE_GETFDPIC_EXEC:
957 tmp = mm->context.exec_fdpic_loadmap;
958 break;
959 case PTRACE_GETFDPIC_INTERP:
960 tmp = mm->context.interp_fdpic_loadmap;
961 break;
962 default:
963 break;
964 }
965 mmput(mm);
966
967 ret = put_user(tmp, datalp);
968 break;
969 }
970#endif
971
972#ifdef PTRACE_SINGLESTEP
973 case PTRACE_SINGLESTEP:
974#endif
975#ifdef PTRACE_SINGLEBLOCK
976 case PTRACE_SINGLEBLOCK:
977#endif
978#ifdef PTRACE_SYSEMU
979 case PTRACE_SYSEMU:
980 case PTRACE_SYSEMU_SINGLESTEP:
981#endif
982 case PTRACE_SYSCALL:
983 case PTRACE_CONT:
984 return ptrace_resume(child, request, data);
985
986 case PTRACE_KILL:
987 if (child->exit_state) /* already dead */
988 return 0;
989 return ptrace_resume(child, request, SIGKILL);
990
991#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
992 case PTRACE_GETREGSET:
993 case PTRACE_SETREGSET: {
994 struct iovec kiov;
995 struct iovec __user *uiov = datavp;
996
997 if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
998 return -EFAULT;
999
1000 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1001 __get_user(kiov.iov_len, &uiov->iov_len))
1002 return -EFAULT;
1003
1004 ret = ptrace_regset(child, request, addr, &kiov);
1005 if (!ret)
1006 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1007 break;
1008 }
1009#endif
1010 default:
1011 break;
1012 }
1013
1014 return ret;
1015}
1016
1017static struct task_struct *ptrace_get_task_struct(pid_t pid)
1018{
1019 struct task_struct *child;
1020
1021 rcu_read_lock();
1022 child = find_task_by_vpid(pid);
1023 if (child)
1024 get_task_struct(child);
1025 rcu_read_unlock();
1026
1027 if (!child)
1028 return ERR_PTR(-ESRCH);
1029 return child;
1030}
1031
1032#ifndef arch_ptrace_attach
1033#define arch_ptrace_attach(child) do { } while (0)
1034#endif
1035
1036SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1037 unsigned long, data)
1038{
1039 struct task_struct *child;
1040 long ret;
1041
1042 if (request == PTRACE_TRACEME) {
1043 ret = ptrace_traceme();
1044 if (!ret)
1045 arch_ptrace_attach(current);
1046 goto out;
1047 }
1048
1049 child = ptrace_get_task_struct(pid);
1050 if (IS_ERR(child)) {
1051 ret = PTR_ERR(child);
1052 goto out;
1053 }
1054
1055 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1056 ret = ptrace_attach(child, request, addr, data);
1057 /*
1058 * Some architectures need to do book-keeping after
1059 * a ptrace attach.
1060 */
1061 if (!ret)
1062 arch_ptrace_attach(child);
1063 goto out_put_task_struct;
1064 }
1065
1066 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1067 request == PTRACE_INTERRUPT);
1068 if (ret < 0)
1069 goto out_put_task_struct;
1070
1071 ret = arch_ptrace(child, request, addr, data);
1072 if (ret || request != PTRACE_DETACH)
1073 ptrace_unfreeze_traced(child);
1074
1075 out_put_task_struct:
1076 put_task_struct(child);
1077 out:
1078 return ret;
1079}
1080
1081int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1082 unsigned long data)
1083{
1084 unsigned long tmp;
1085 int copied;
1086
1087 copied = access_process_vm(tsk, addr, &tmp, sizeof(tmp), 0);
1088 if (copied != sizeof(tmp))
1089 return -EIO;
1090 return put_user(tmp, (unsigned long __user *)data);
1091}
1092
1093int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1094 unsigned long data)
1095{
1096 int copied;
1097
1098 copied = access_process_vm(tsk, addr, &data, sizeof(data), 1);
1099 return (copied == sizeof(data)) ? 0 : -EIO;
1100}
1101
1102#if defined CONFIG_COMPAT
1103#include <linux/compat.h>
1104
1105int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1106 compat_ulong_t addr, compat_ulong_t data)
1107{
1108 compat_ulong_t __user *datap = compat_ptr(data);
1109 compat_ulong_t word;
1110 siginfo_t siginfo;
1111 int ret;
1112
1113 switch (request) {
1114 case PTRACE_PEEKTEXT:
1115 case PTRACE_PEEKDATA:
1116 ret = access_process_vm(child, addr, &word, sizeof(word), 0);
1117 if (ret != sizeof(word))
1118 ret = -EIO;
1119 else
1120 ret = put_user(word, datap);
1121 break;
1122
1123 case PTRACE_POKETEXT:
1124 case PTRACE_POKEDATA:
1125 ret = access_process_vm(child, addr, &data, sizeof(data), 1);
1126 ret = (ret != sizeof(data) ? -EIO : 0);
1127 break;
1128
1129 case PTRACE_GETEVENTMSG:
1130 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1131 break;
1132
1133 case PTRACE_GETSIGINFO:
1134 ret = ptrace_getsiginfo(child, &siginfo);
1135 if (!ret)
1136 ret = copy_siginfo_to_user32(
1137 (struct compat_siginfo __user *) datap,
1138 &siginfo);
1139 break;
1140
1141 case PTRACE_SETSIGINFO:
1142 memset(&siginfo, 0, sizeof siginfo);
1143 if (copy_siginfo_from_user32(
1144 &siginfo, (struct compat_siginfo __user *) datap))
1145 ret = -EFAULT;
1146 else
1147 ret = ptrace_setsiginfo(child, &siginfo);
1148 break;
1149#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1150 case PTRACE_GETREGSET:
1151 case PTRACE_SETREGSET:
1152 {
1153 struct iovec kiov;
1154 struct compat_iovec __user *uiov =
1155 (struct compat_iovec __user *) datap;
1156 compat_uptr_t ptr;
1157 compat_size_t len;
1158
1159 if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1160 return -EFAULT;
1161
1162 if (__get_user(ptr, &uiov->iov_base) ||
1163 __get_user(len, &uiov->iov_len))
1164 return -EFAULT;
1165
1166 kiov.iov_base = compat_ptr(ptr);
1167 kiov.iov_len = len;
1168
1169 ret = ptrace_regset(child, request, addr, &kiov);
1170 if (!ret)
1171 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1172 break;
1173 }
1174#endif
1175
1176 default:
1177 ret = ptrace_request(child, request, addr, data);
1178 }
1179
1180 return ret;
1181}
1182
1183COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1184 compat_long_t, addr, compat_long_t, data)
1185{
1186 struct task_struct *child;
1187 long ret;
1188
1189 if (request == PTRACE_TRACEME) {
1190 ret = ptrace_traceme();
1191 goto out;
1192 }
1193
1194 child = ptrace_get_task_struct(pid);
1195 if (IS_ERR(child)) {
1196 ret = PTR_ERR(child);
1197 goto out;
1198 }
1199
1200 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1201 ret = ptrace_attach(child, request, addr, data);
1202 /*
1203 * Some architectures need to do book-keeping after
1204 * a ptrace attach.
1205 */
1206 if (!ret)
1207 arch_ptrace_attach(child);
1208 goto out_put_task_struct;
1209 }
1210
1211 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1212 request == PTRACE_INTERRUPT);
1213 if (!ret) {
1214 ret = compat_arch_ptrace(child, request, addr, data);
1215 if (ret || request != PTRACE_DETACH)
1216 ptrace_unfreeze_traced(child);
1217 }
1218
1219 out_put_task_struct:
1220 put_task_struct(child);
1221 out:
1222 return ret;
1223}
1224#endif /* CONFIG_COMPAT */
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
36#include <asm/syscall.h> /* for syscall_get_* */
37
38/*
39 * Access another process' address space via ptrace.
40 * Source/target buffer must be kernel space,
41 * Do not walk the page table directly, use get_user_pages
42 */
43int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
44 void *buf, int len, unsigned int gup_flags)
45{
46 struct mm_struct *mm;
47 int ret;
48
49 mm = get_task_mm(tsk);
50 if (!mm)
51 return 0;
52
53 if (!tsk->ptrace ||
54 (current != tsk->parent) ||
55 ((get_dumpable(mm) != SUID_DUMP_USER) &&
56 !ptracer_capable(tsk, mm->user_ns))) {
57 mmput(mm);
58 return 0;
59 }
60
61 ret = __access_remote_vm(mm, addr, buf, len, gup_flags);
62 mmput(mm);
63
64 return ret;
65}
66
67
68void __ptrace_link(struct task_struct *child, struct task_struct *new_parent,
69 const struct cred *ptracer_cred)
70{
71 BUG_ON(!list_empty(&child->ptrace_entry));
72 list_add(&child->ptrace_entry, &new_parent->ptraced);
73 child->parent = new_parent;
74 child->ptracer_cred = get_cred(ptracer_cred);
75}
76
77/*
78 * ptrace a task: make the debugger its new parent and
79 * move it to the ptrace list.
80 *
81 * Must be called with the tasklist lock write-held.
82 */
83static void ptrace_link(struct task_struct *child, struct task_struct *new_parent)
84{
85 __ptrace_link(child, new_parent, current_cred());
86}
87
88/**
89 * __ptrace_unlink - unlink ptracee and restore its execution state
90 * @child: ptracee to be unlinked
91 *
92 * Remove @child from the ptrace list, move it back to the original parent,
93 * and restore the execution state so that it conforms to the group stop
94 * state.
95 *
96 * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
97 * exiting. For PTRACE_DETACH, unless the ptracee has been killed between
98 * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
99 * If the ptracer is exiting, the ptracee can be in any state.
100 *
101 * After detach, the ptracee should be in a state which conforms to the
102 * group stop. If the group is stopped or in the process of stopping, the
103 * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
104 * up from TASK_TRACED.
105 *
106 * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
107 * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
108 * to but in the opposite direction of what happens while attaching to a
109 * stopped task. However, in this direction, the intermediate RUNNING
110 * state is not hidden even from the current ptracer and if it immediately
111 * re-attaches and performs a WNOHANG wait(2), it may fail.
112 *
113 * CONTEXT:
114 * write_lock_irq(tasklist_lock)
115 */
116void __ptrace_unlink(struct task_struct *child)
117{
118 const struct cred *old_cred;
119 BUG_ON(!child->ptrace);
120
121 clear_task_syscall_work(child, SYSCALL_TRACE);
122#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
123 clear_task_syscall_work(child, SYSCALL_EMU);
124#endif
125
126 child->parent = child->real_parent;
127 list_del_init(&child->ptrace_entry);
128 old_cred = child->ptracer_cred;
129 child->ptracer_cred = NULL;
130 put_cred(old_cred);
131
132 spin_lock(&child->sighand->siglock);
133 child->ptrace = 0;
134 /*
135 * Clear all pending traps and TRAPPING. TRAPPING should be
136 * cleared regardless of JOBCTL_STOP_PENDING. Do it explicitly.
137 */
138 task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
139 task_clear_jobctl_trapping(child);
140
141 /*
142 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
143 * @child isn't dead.
144 */
145 if (!(child->flags & PF_EXITING) &&
146 (child->signal->flags & SIGNAL_STOP_STOPPED ||
147 child->signal->group_stop_count)) {
148 child->jobctl |= JOBCTL_STOP_PENDING;
149
150 /*
151 * This is only possible if this thread was cloned by the
152 * traced task running in the stopped group, set the signal
153 * for the future reports.
154 * FIXME: we should change ptrace_init_task() to handle this
155 * case.
156 */
157 if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
158 child->jobctl |= SIGSTOP;
159 }
160
161 /*
162 * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
163 * @child in the butt. Note that @resume should be used iff @child
164 * is in TASK_TRACED; otherwise, we might unduly disrupt
165 * TASK_KILLABLE sleeps.
166 */
167 if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
168 ptrace_signal_wake_up(child, true);
169
170 spin_unlock(&child->sighand->siglock);
171}
172
173static bool looks_like_a_spurious_pid(struct task_struct *task)
174{
175 if (task->exit_code != ((PTRACE_EVENT_EXEC << 8) | SIGTRAP))
176 return false;
177
178 if (task_pid_vnr(task) == task->ptrace_message)
179 return false;
180 /*
181 * The tracee changed its pid but the PTRACE_EVENT_EXEC event
182 * was not wait()'ed, most probably debugger targets the old
183 * leader which was destroyed in de_thread().
184 */
185 return true;
186}
187
188/* Ensure that nothing can wake it up, even SIGKILL */
189static bool ptrace_freeze_traced(struct task_struct *task)
190{
191 bool ret = false;
192
193 /* Lockless, nobody but us can set this flag */
194 if (task->jobctl & JOBCTL_LISTENING)
195 return ret;
196
197 spin_lock_irq(&task->sighand->siglock);
198 if (task_is_traced(task) && !looks_like_a_spurious_pid(task) &&
199 !__fatal_signal_pending(task)) {
200 WRITE_ONCE(task->__state, __TASK_TRACED);
201 ret = true;
202 }
203 spin_unlock_irq(&task->sighand->siglock);
204
205 return ret;
206}
207
208static void ptrace_unfreeze_traced(struct task_struct *task)
209{
210 if (READ_ONCE(task->__state) != __TASK_TRACED)
211 return;
212
213 WARN_ON(!task->ptrace || task->parent != current);
214
215 /*
216 * PTRACE_LISTEN can allow ptrace_trap_notify to wake us up remotely.
217 * Recheck state under the lock to close this race.
218 */
219 spin_lock_irq(&task->sighand->siglock);
220 if (READ_ONCE(task->__state) == __TASK_TRACED) {
221 if (__fatal_signal_pending(task))
222 wake_up_state(task, __TASK_TRACED);
223 else
224 WRITE_ONCE(task->__state, TASK_TRACED);
225 }
226 spin_unlock_irq(&task->sighand->siglock);
227}
228
229/**
230 * ptrace_check_attach - check whether ptracee is ready for ptrace operation
231 * @child: ptracee to check for
232 * @ignore_state: don't check whether @child is currently %TASK_TRACED
233 *
234 * Check whether @child is being ptraced by %current and ready for further
235 * ptrace operations. If @ignore_state is %false, @child also should be in
236 * %TASK_TRACED state and on return the child is guaranteed to be traced
237 * and not executing. If @ignore_state is %true, @child can be in any
238 * state.
239 *
240 * CONTEXT:
241 * Grabs and releases tasklist_lock and @child->sighand->siglock.
242 *
243 * RETURNS:
244 * 0 on success, -ESRCH if %child is not ready.
245 */
246static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
247{
248 int ret = -ESRCH;
249
250 /*
251 * We take the read lock around doing both checks to close a
252 * possible race where someone else was tracing our child and
253 * detached between these two checks. After this locked check,
254 * we are sure that this is our traced child and that can only
255 * be changed by us so it's not changing right after this.
256 */
257 read_lock(&tasklist_lock);
258 if (child->ptrace && child->parent == current) {
259 WARN_ON(READ_ONCE(child->__state) == __TASK_TRACED);
260 /*
261 * child->sighand can't be NULL, release_task()
262 * does ptrace_unlink() before __exit_signal().
263 */
264 if (ignore_state || ptrace_freeze_traced(child))
265 ret = 0;
266 }
267 read_unlock(&tasklist_lock);
268
269 if (!ret && !ignore_state) {
270 if (!wait_task_inactive(child, __TASK_TRACED)) {
271 /*
272 * This can only happen if may_ptrace_stop() fails and
273 * ptrace_stop() changes ->state back to TASK_RUNNING,
274 * so we should not worry about leaking __TASK_TRACED.
275 */
276 WARN_ON(READ_ONCE(child->__state) == __TASK_TRACED);
277 ret = -ESRCH;
278 }
279 }
280
281 return ret;
282}
283
284static bool ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
285{
286 if (mode & PTRACE_MODE_NOAUDIT)
287 return ns_capable_noaudit(ns, CAP_SYS_PTRACE);
288 return ns_capable(ns, CAP_SYS_PTRACE);
289}
290
291/* Returns 0 on success, -errno on denial. */
292static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
293{
294 const struct cred *cred = current_cred(), *tcred;
295 struct mm_struct *mm;
296 kuid_t caller_uid;
297 kgid_t caller_gid;
298
299 if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
300 WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
301 return -EPERM;
302 }
303
304 /* May we inspect the given task?
305 * This check is used both for attaching with ptrace
306 * and for allowing access to sensitive information in /proc.
307 *
308 * ptrace_attach denies several cases that /proc allows
309 * because setting up the necessary parent/child relationship
310 * or halting the specified task is impossible.
311 */
312
313 /* Don't let security modules deny introspection */
314 if (same_thread_group(task, current))
315 return 0;
316 rcu_read_lock();
317 if (mode & PTRACE_MODE_FSCREDS) {
318 caller_uid = cred->fsuid;
319 caller_gid = cred->fsgid;
320 } else {
321 /*
322 * Using the euid would make more sense here, but something
323 * in userland might rely on the old behavior, and this
324 * shouldn't be a security problem since
325 * PTRACE_MODE_REALCREDS implies that the caller explicitly
326 * used a syscall that requests access to another process
327 * (and not a filesystem syscall to procfs).
328 */
329 caller_uid = cred->uid;
330 caller_gid = cred->gid;
331 }
332 tcred = __task_cred(task);
333 if (uid_eq(caller_uid, tcred->euid) &&
334 uid_eq(caller_uid, tcred->suid) &&
335 uid_eq(caller_uid, tcred->uid) &&
336 gid_eq(caller_gid, tcred->egid) &&
337 gid_eq(caller_gid, tcred->sgid) &&
338 gid_eq(caller_gid, tcred->gid))
339 goto ok;
340 if (ptrace_has_cap(tcred->user_ns, mode))
341 goto ok;
342 rcu_read_unlock();
343 return -EPERM;
344ok:
345 rcu_read_unlock();
346 /*
347 * If a task drops privileges and becomes nondumpable (through a syscall
348 * like setresuid()) while we are trying to access it, we must ensure
349 * that the dumpability is read after the credentials; otherwise,
350 * we may be able to attach to a task that we shouldn't be able to
351 * attach to (as if the task had dropped privileges without becoming
352 * nondumpable).
353 * Pairs with a write barrier in commit_creds().
354 */
355 smp_rmb();
356 mm = task->mm;
357 if (mm &&
358 ((get_dumpable(mm) != SUID_DUMP_USER) &&
359 !ptrace_has_cap(mm->user_ns, mode)))
360 return -EPERM;
361
362 return security_ptrace_access_check(task, mode);
363}
364
365bool ptrace_may_access(struct task_struct *task, unsigned int mode)
366{
367 int err;
368 task_lock(task);
369 err = __ptrace_may_access(task, mode);
370 task_unlock(task);
371 return !err;
372}
373
374static int ptrace_attach(struct task_struct *task, long request,
375 unsigned long addr,
376 unsigned long flags)
377{
378 bool seize = (request == PTRACE_SEIZE);
379 int retval;
380
381 retval = -EIO;
382 if (seize) {
383 if (addr != 0)
384 goto out;
385 if (flags & ~(unsigned long)PTRACE_O_MASK)
386 goto out;
387 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
388 } else {
389 flags = PT_PTRACED;
390 }
391
392 audit_ptrace(task);
393
394 retval = -EPERM;
395 if (unlikely(task->flags & PF_KTHREAD))
396 goto out;
397 if (same_thread_group(task, current))
398 goto out;
399
400 /*
401 * Protect exec's credential calculations against our interference;
402 * SUID, SGID and LSM creds get determined differently
403 * under ptrace.
404 */
405 retval = -ERESTARTNOINTR;
406 if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
407 goto out;
408
409 task_lock(task);
410 retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
411 task_unlock(task);
412 if (retval)
413 goto unlock_creds;
414
415 write_lock_irq(&tasklist_lock);
416 retval = -EPERM;
417 if (unlikely(task->exit_state))
418 goto unlock_tasklist;
419 if (task->ptrace)
420 goto unlock_tasklist;
421
422 if (seize)
423 flags |= PT_SEIZED;
424 task->ptrace = flags;
425
426 ptrace_link(task, current);
427
428 /* SEIZE doesn't trap tracee on attach */
429 if (!seize)
430 send_sig_info(SIGSTOP, SEND_SIG_PRIV, task);
431
432 spin_lock(&task->sighand->siglock);
433
434 /*
435 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
436 * TRAPPING, and kick it so that it transits to TRACED. TRAPPING
437 * will be cleared if the child completes the transition or any
438 * event which clears the group stop states happens. We'll wait
439 * for the transition to complete before returning from this
440 * function.
441 *
442 * This hides STOPPED -> RUNNING -> TRACED transition from the
443 * attaching thread but a different thread in the same group can
444 * still observe the transient RUNNING state. IOW, if another
445 * thread's WNOHANG wait(2) on the stopped tracee races against
446 * ATTACH, the wait(2) may fail due to the transient RUNNING.
447 *
448 * The following task_is_stopped() test is safe as both transitions
449 * in and out of STOPPED are protected by siglock.
450 */
451 if (task_is_stopped(task) &&
452 task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
453 signal_wake_up_state(task, __TASK_STOPPED);
454
455 spin_unlock(&task->sighand->siglock);
456
457 retval = 0;
458unlock_tasklist:
459 write_unlock_irq(&tasklist_lock);
460unlock_creds:
461 mutex_unlock(&task->signal->cred_guard_mutex);
462out:
463 if (!retval) {
464 /*
465 * We do not bother to change retval or clear JOBCTL_TRAPPING
466 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
467 * not return to user-mode, it will exit and clear this bit in
468 * __ptrace_unlink() if it wasn't already cleared by the tracee;
469 * and until then nobody can ptrace this task.
470 */
471 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
472 proc_ptrace_connector(task, PTRACE_ATTACH);
473 }
474
475 return retval;
476}
477
478/**
479 * ptrace_traceme -- helper for PTRACE_TRACEME
480 *
481 * Performs checks and sets PT_PTRACED.
482 * Should be used by all ptrace implementations for PTRACE_TRACEME.
483 */
484static int ptrace_traceme(void)
485{
486 int ret = -EPERM;
487
488 write_lock_irq(&tasklist_lock);
489 /* Are we already being traced? */
490 if (!current->ptrace) {
491 ret = security_ptrace_traceme(current->parent);
492 /*
493 * Check PF_EXITING to ensure ->real_parent has not passed
494 * exit_ptrace(). Otherwise we don't report the error but
495 * pretend ->real_parent untraces us right after return.
496 */
497 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
498 current->ptrace = PT_PTRACED;
499 ptrace_link(current, current->real_parent);
500 }
501 }
502 write_unlock_irq(&tasklist_lock);
503
504 return ret;
505}
506
507/*
508 * Called with irqs disabled, returns true if childs should reap themselves.
509 */
510static int ignoring_children(struct sighand_struct *sigh)
511{
512 int ret;
513 spin_lock(&sigh->siglock);
514 ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
515 (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
516 spin_unlock(&sigh->siglock);
517 return ret;
518}
519
520/*
521 * Called with tasklist_lock held for writing.
522 * Unlink a traced task, and clean it up if it was a traced zombie.
523 * Return true if it needs to be reaped with release_task().
524 * (We can't call release_task() here because we already hold tasklist_lock.)
525 *
526 * If it's a zombie, our attachedness prevented normal parent notification
527 * or self-reaping. Do notification now if it would have happened earlier.
528 * If it should reap itself, return true.
529 *
530 * If it's our own child, there is no notification to do. But if our normal
531 * children self-reap, then this child was prevented by ptrace and we must
532 * reap it now, in that case we must also wake up sub-threads sleeping in
533 * do_wait().
534 */
535static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
536{
537 bool dead;
538
539 __ptrace_unlink(p);
540
541 if (p->exit_state != EXIT_ZOMBIE)
542 return false;
543
544 dead = !thread_group_leader(p);
545
546 if (!dead && thread_group_empty(p)) {
547 if (!same_thread_group(p->real_parent, tracer))
548 dead = do_notify_parent(p, p->exit_signal);
549 else if (ignoring_children(tracer->sighand)) {
550 __wake_up_parent(p, tracer);
551 dead = true;
552 }
553 }
554 /* Mark it as in the process of being reaped. */
555 if (dead)
556 p->exit_state = EXIT_DEAD;
557 return dead;
558}
559
560static int ptrace_detach(struct task_struct *child, unsigned int data)
561{
562 if (!valid_signal(data))
563 return -EIO;
564
565 /* Architecture-specific hardware disable .. */
566 ptrace_disable(child);
567
568 write_lock_irq(&tasklist_lock);
569 /*
570 * We rely on ptrace_freeze_traced(). It can't be killed and
571 * untraced by another thread, it can't be a zombie.
572 */
573 WARN_ON(!child->ptrace || child->exit_state);
574 /*
575 * tasklist_lock avoids the race with wait_task_stopped(), see
576 * the comment in ptrace_resume().
577 */
578 child->exit_code = data;
579 __ptrace_detach(current, child);
580 write_unlock_irq(&tasklist_lock);
581
582 proc_ptrace_connector(child, PTRACE_DETACH);
583
584 return 0;
585}
586
587/*
588 * Detach all tasks we were using ptrace on. Called with tasklist held
589 * for writing.
590 */
591void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
592{
593 struct task_struct *p, *n;
594
595 list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
596 if (unlikely(p->ptrace & PT_EXITKILL))
597 send_sig_info(SIGKILL, SEND_SIG_PRIV, p);
598
599 if (__ptrace_detach(tracer, p))
600 list_add(&p->ptrace_entry, dead);
601 }
602}
603
604int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
605{
606 int copied = 0;
607
608 while (len > 0) {
609 char buf[128];
610 int this_len, retval;
611
612 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
613 retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
614
615 if (!retval) {
616 if (copied)
617 break;
618 return -EIO;
619 }
620 if (copy_to_user(dst, buf, retval))
621 return -EFAULT;
622 copied += retval;
623 src += retval;
624 dst += retval;
625 len -= retval;
626 }
627 return copied;
628}
629
630int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
631{
632 int copied = 0;
633
634 while (len > 0) {
635 char buf[128];
636 int this_len, retval;
637
638 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
639 if (copy_from_user(buf, src, this_len))
640 return -EFAULT;
641 retval = ptrace_access_vm(tsk, dst, buf, this_len,
642 FOLL_FORCE | FOLL_WRITE);
643 if (!retval) {
644 if (copied)
645 break;
646 return -EIO;
647 }
648 copied += retval;
649 src += retval;
650 dst += retval;
651 len -= retval;
652 }
653 return copied;
654}
655
656static int ptrace_setoptions(struct task_struct *child, unsigned long data)
657{
658 unsigned flags;
659
660 if (data & ~(unsigned long)PTRACE_O_MASK)
661 return -EINVAL;
662
663 if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
664 if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
665 !IS_ENABLED(CONFIG_SECCOMP))
666 return -EINVAL;
667
668 if (!capable(CAP_SYS_ADMIN))
669 return -EPERM;
670
671 if (seccomp_mode(¤t->seccomp) != SECCOMP_MODE_DISABLED ||
672 current->ptrace & PT_SUSPEND_SECCOMP)
673 return -EPERM;
674 }
675
676 /* Avoid intermediate state when all opts are cleared */
677 flags = child->ptrace;
678 flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
679 flags |= (data << PT_OPT_FLAG_SHIFT);
680 child->ptrace = flags;
681
682 return 0;
683}
684
685static int ptrace_getsiginfo(struct task_struct *child, kernel_siginfo_t *info)
686{
687 unsigned long flags;
688 int error = -ESRCH;
689
690 if (lock_task_sighand(child, &flags)) {
691 error = -EINVAL;
692 if (likely(child->last_siginfo != NULL)) {
693 copy_siginfo(info, child->last_siginfo);
694 error = 0;
695 }
696 unlock_task_sighand(child, &flags);
697 }
698 return error;
699}
700
701static int ptrace_setsiginfo(struct task_struct *child, const kernel_siginfo_t *info)
702{
703 unsigned long flags;
704 int error = -ESRCH;
705
706 if (lock_task_sighand(child, &flags)) {
707 error = -EINVAL;
708 if (likely(child->last_siginfo != NULL)) {
709 copy_siginfo(child->last_siginfo, info);
710 error = 0;
711 }
712 unlock_task_sighand(child, &flags);
713 }
714 return error;
715}
716
717static int ptrace_peek_siginfo(struct task_struct *child,
718 unsigned long addr,
719 unsigned long data)
720{
721 struct ptrace_peeksiginfo_args arg;
722 struct sigpending *pending;
723 struct sigqueue *q;
724 int ret, i;
725
726 ret = copy_from_user(&arg, (void __user *) addr,
727 sizeof(struct ptrace_peeksiginfo_args));
728 if (ret)
729 return -EFAULT;
730
731 if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
732 return -EINVAL; /* unknown flags */
733
734 if (arg.nr < 0)
735 return -EINVAL;
736
737 /* Ensure arg.off fits in an unsigned long */
738 if (arg.off > ULONG_MAX)
739 return 0;
740
741 if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
742 pending = &child->signal->shared_pending;
743 else
744 pending = &child->pending;
745
746 for (i = 0; i < arg.nr; ) {
747 kernel_siginfo_t info;
748 unsigned long off = arg.off + i;
749 bool found = false;
750
751 spin_lock_irq(&child->sighand->siglock);
752 list_for_each_entry(q, &pending->list, list) {
753 if (!off--) {
754 found = true;
755 copy_siginfo(&info, &q->info);
756 break;
757 }
758 }
759 spin_unlock_irq(&child->sighand->siglock);
760
761 if (!found) /* beyond the end of the list */
762 break;
763
764#ifdef CONFIG_COMPAT
765 if (unlikely(in_compat_syscall())) {
766 compat_siginfo_t __user *uinfo = compat_ptr(data);
767
768 if (copy_siginfo_to_user32(uinfo, &info)) {
769 ret = -EFAULT;
770 break;
771 }
772
773 } else
774#endif
775 {
776 siginfo_t __user *uinfo = (siginfo_t __user *) data;
777
778 if (copy_siginfo_to_user(uinfo, &info)) {
779 ret = -EFAULT;
780 break;
781 }
782 }
783
784 data += sizeof(siginfo_t);
785 i++;
786
787 if (signal_pending(current))
788 break;
789
790 cond_resched();
791 }
792
793 if (i > 0)
794 return i;
795
796 return ret;
797}
798
799#ifdef CONFIG_RSEQ
800static long ptrace_get_rseq_configuration(struct task_struct *task,
801 unsigned long size, void __user *data)
802{
803 struct ptrace_rseq_configuration conf = {
804 .rseq_abi_pointer = (u64)(uintptr_t)task->rseq,
805 .rseq_abi_size = sizeof(*task->rseq),
806 .signature = task->rseq_sig,
807 .flags = 0,
808 };
809
810 size = min_t(unsigned long, size, sizeof(conf));
811 if (copy_to_user(data, &conf, size))
812 return -EFAULT;
813 return sizeof(conf);
814}
815#endif
816
817#ifdef PTRACE_SINGLESTEP
818#define is_singlestep(request) ((request) == PTRACE_SINGLESTEP)
819#else
820#define is_singlestep(request) 0
821#endif
822
823#ifdef PTRACE_SINGLEBLOCK
824#define is_singleblock(request) ((request) == PTRACE_SINGLEBLOCK)
825#else
826#define is_singleblock(request) 0
827#endif
828
829#ifdef PTRACE_SYSEMU
830#define is_sysemu_singlestep(request) ((request) == PTRACE_SYSEMU_SINGLESTEP)
831#else
832#define is_sysemu_singlestep(request) 0
833#endif
834
835static int ptrace_resume(struct task_struct *child, long request,
836 unsigned long data)
837{
838 bool need_siglock;
839
840 if (!valid_signal(data))
841 return -EIO;
842
843 if (request == PTRACE_SYSCALL)
844 set_task_syscall_work(child, SYSCALL_TRACE);
845 else
846 clear_task_syscall_work(child, SYSCALL_TRACE);
847
848#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
849 if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
850 set_task_syscall_work(child, SYSCALL_EMU);
851 else
852 clear_task_syscall_work(child, SYSCALL_EMU);
853#endif
854
855 if (is_singleblock(request)) {
856 if (unlikely(!arch_has_block_step()))
857 return -EIO;
858 user_enable_block_step(child);
859 } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
860 if (unlikely(!arch_has_single_step()))
861 return -EIO;
862 user_enable_single_step(child);
863 } else {
864 user_disable_single_step(child);
865 }
866
867 /*
868 * Change ->exit_code and ->state under siglock to avoid the race
869 * with wait_task_stopped() in between; a non-zero ->exit_code will
870 * wrongly look like another report from tracee.
871 *
872 * Note that we need siglock even if ->exit_code == data and/or this
873 * status was not reported yet, the new status must not be cleared by
874 * wait_task_stopped() after resume.
875 *
876 * If data == 0 we do not care if wait_task_stopped() reports the old
877 * status and clears the code too; this can't race with the tracee, it
878 * takes siglock after resume.
879 */
880 need_siglock = data && !thread_group_empty(current);
881 if (need_siglock)
882 spin_lock_irq(&child->sighand->siglock);
883 child->exit_code = data;
884 wake_up_state(child, __TASK_TRACED);
885 if (need_siglock)
886 spin_unlock_irq(&child->sighand->siglock);
887
888 return 0;
889}
890
891#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
892
893static const struct user_regset *
894find_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
908static 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 */
935EXPORT_SYMBOL_GPL(task_user_regset_view);
936
937static unsigned long
938ptrace_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
954static unsigned long
955ptrace_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
973static unsigned long
974ptrace_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
987static int
988ptrace_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
1031int 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#ifdef PTRACE_SINGLESTEP
1210 case PTRACE_SINGLESTEP:
1211#endif
1212#ifdef PTRACE_SINGLEBLOCK
1213 case PTRACE_SINGLEBLOCK:
1214#endif
1215#ifdef PTRACE_SYSEMU
1216 case PTRACE_SYSEMU:
1217 case PTRACE_SYSEMU_SINGLESTEP:
1218#endif
1219 case PTRACE_SYSCALL:
1220 case PTRACE_CONT:
1221 return ptrace_resume(child, request, data);
1222
1223 case PTRACE_KILL:
1224 if (child->exit_state) /* already dead */
1225 return 0;
1226 return ptrace_resume(child, request, SIGKILL);
1227
1228#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1229 case PTRACE_GETREGSET:
1230 case PTRACE_SETREGSET: {
1231 struct iovec kiov;
1232 struct iovec __user *uiov = datavp;
1233
1234 if (!access_ok(uiov, sizeof(*uiov)))
1235 return -EFAULT;
1236
1237 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1238 __get_user(kiov.iov_len, &uiov->iov_len))
1239 return -EFAULT;
1240
1241 ret = ptrace_regset(child, request, addr, &kiov);
1242 if (!ret)
1243 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1244 break;
1245 }
1246
1247 case PTRACE_GET_SYSCALL_INFO:
1248 ret = ptrace_get_syscall_info(child, addr, datavp);
1249 break;
1250#endif
1251
1252 case PTRACE_SECCOMP_GET_FILTER:
1253 ret = seccomp_get_filter(child, addr, datavp);
1254 break;
1255
1256 case PTRACE_SECCOMP_GET_METADATA:
1257 ret = seccomp_get_metadata(child, addr, datavp);
1258 break;
1259
1260#ifdef CONFIG_RSEQ
1261 case PTRACE_GET_RSEQ_CONFIGURATION:
1262 ret = ptrace_get_rseq_configuration(child, addr, datavp);
1263 break;
1264#endif
1265
1266 default:
1267 break;
1268 }
1269
1270 return ret;
1271}
1272
1273#ifndef arch_ptrace_attach
1274#define arch_ptrace_attach(child) do { } while (0)
1275#endif
1276
1277SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1278 unsigned long, data)
1279{
1280 struct task_struct *child;
1281 long ret;
1282
1283 if (request == PTRACE_TRACEME) {
1284 ret = ptrace_traceme();
1285 if (!ret)
1286 arch_ptrace_attach(current);
1287 goto out;
1288 }
1289
1290 child = find_get_task_by_vpid(pid);
1291 if (!child) {
1292 ret = -ESRCH;
1293 goto out;
1294 }
1295
1296 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1297 ret = ptrace_attach(child, request, addr, data);
1298 /*
1299 * Some architectures need to do book-keeping after
1300 * a ptrace attach.
1301 */
1302 if (!ret)
1303 arch_ptrace_attach(child);
1304 goto out_put_task_struct;
1305 }
1306
1307 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1308 request == PTRACE_INTERRUPT);
1309 if (ret < 0)
1310 goto out_put_task_struct;
1311
1312 ret = arch_ptrace(child, request, addr, data);
1313 if (ret || request != PTRACE_DETACH)
1314 ptrace_unfreeze_traced(child);
1315
1316 out_put_task_struct:
1317 put_task_struct(child);
1318 out:
1319 return ret;
1320}
1321
1322int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1323 unsigned long data)
1324{
1325 unsigned long tmp;
1326 int copied;
1327
1328 copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1329 if (copied != sizeof(tmp))
1330 return -EIO;
1331 return put_user(tmp, (unsigned long __user *)data);
1332}
1333
1334int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1335 unsigned long data)
1336{
1337 int copied;
1338
1339 copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1340 FOLL_FORCE | FOLL_WRITE);
1341 return (copied == sizeof(data)) ? 0 : -EIO;
1342}
1343
1344#if defined CONFIG_COMPAT
1345
1346int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1347 compat_ulong_t addr, compat_ulong_t data)
1348{
1349 compat_ulong_t __user *datap = compat_ptr(data);
1350 compat_ulong_t word;
1351 kernel_siginfo_t siginfo;
1352 int ret;
1353
1354 switch (request) {
1355 case PTRACE_PEEKTEXT:
1356 case PTRACE_PEEKDATA:
1357 ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1358 FOLL_FORCE);
1359 if (ret != sizeof(word))
1360 ret = -EIO;
1361 else
1362 ret = put_user(word, datap);
1363 break;
1364
1365 case PTRACE_POKETEXT:
1366 case PTRACE_POKEDATA:
1367 ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1368 FOLL_FORCE | FOLL_WRITE);
1369 ret = (ret != sizeof(data) ? -EIO : 0);
1370 break;
1371
1372 case PTRACE_GETEVENTMSG:
1373 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1374 break;
1375
1376 case PTRACE_GETSIGINFO:
1377 ret = ptrace_getsiginfo(child, &siginfo);
1378 if (!ret)
1379 ret = copy_siginfo_to_user32(
1380 (struct compat_siginfo __user *) datap,
1381 &siginfo);
1382 break;
1383
1384 case PTRACE_SETSIGINFO:
1385 ret = copy_siginfo_from_user32(
1386 &siginfo, (struct compat_siginfo __user *) datap);
1387 if (!ret)
1388 ret = ptrace_setsiginfo(child, &siginfo);
1389 break;
1390#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1391 case PTRACE_GETREGSET:
1392 case PTRACE_SETREGSET:
1393 {
1394 struct iovec kiov;
1395 struct compat_iovec __user *uiov =
1396 (struct compat_iovec __user *) datap;
1397 compat_uptr_t ptr;
1398 compat_size_t len;
1399
1400 if (!access_ok(uiov, sizeof(*uiov)))
1401 return -EFAULT;
1402
1403 if (__get_user(ptr, &uiov->iov_base) ||
1404 __get_user(len, &uiov->iov_len))
1405 return -EFAULT;
1406
1407 kiov.iov_base = compat_ptr(ptr);
1408 kiov.iov_len = len;
1409
1410 ret = ptrace_regset(child, request, addr, &kiov);
1411 if (!ret)
1412 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1413 break;
1414 }
1415#endif
1416
1417 default:
1418 ret = ptrace_request(child, request, addr, data);
1419 }
1420
1421 return ret;
1422}
1423
1424COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1425 compat_long_t, addr, compat_long_t, data)
1426{
1427 struct task_struct *child;
1428 long ret;
1429
1430 if (request == PTRACE_TRACEME) {
1431 ret = ptrace_traceme();
1432 goto out;
1433 }
1434
1435 child = find_get_task_by_vpid(pid);
1436 if (!child) {
1437 ret = -ESRCH;
1438 goto out;
1439 }
1440
1441 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1442 ret = ptrace_attach(child, request, addr, data);
1443 /*
1444 * Some architectures need to do book-keeping after
1445 * a ptrace attach.
1446 */
1447 if (!ret)
1448 arch_ptrace_attach(child);
1449 goto out_put_task_struct;
1450 }
1451
1452 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1453 request == PTRACE_INTERRUPT);
1454 if (!ret) {
1455 ret = compat_arch_ptrace(child, request, addr, data);
1456 if (ret || request != PTRACE_DETACH)
1457 ptrace_unfreeze_traced(child);
1458 }
1459
1460 out_put_task_struct:
1461 put_task_struct(child);
1462 out:
1463 return ret;
1464}
1465#endif /* CONFIG_COMPAT */