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