1 /*
2 * linux/kernel/ptrace.c
3 *
4 * (C) Copyright 1999 Linus Torvalds
5 *
6 * Common interfaces for "ptrace()" which we do not want
7 * to continually duplicate across every architecture.
8 */
9
10 #include <linux/capability.h>
11 #include <linux/export.h>
12 #include <linux/sched.h>
13 #include <linux/errno.h>
14 #include <linux/mm.h>
15 #include <linux/highmem.h>
16 #include <linux/pagemap.h>
17 #include <linux/ptrace.h>
18 #include <linux/security.h>
19 #include <linux/signal.h>
20 #include <linux/uio.h>
21 #include <linux/audit.h>
22 #include <linux/pid_namespace.h>
23 #include <linux/syscalls.h>
24 #include <linux/uaccess.h>
25 #include <linux/regset.h>
26 #include <linux/hw_breakpoint.h>
27 #include <linux/cn_proc.h>
28 #include <linux/compat.h>
29
30
31 /*
32 * ptrace a task: make the debugger its new parent and
33 * move it to the ptrace list.
34 *
35 * Must be called with the tasklist lock write-held.
36 */
37 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent)
38 {
39 BUG_ON(!list_empty(&child->ptrace_entry));
40 list_add(&child->ptrace_entry, &new_parent->ptraced);
41 child->parent = new_parent;
42 rcu_read_lock();
43 child->ptracer_cred = get_cred(__task_cred(new_parent));
44 rcu_read_unlock();
45 }
46
47 /**
48 * __ptrace_unlink - unlink ptracee and restore its execution state
49 * @child: ptracee to be unlinked
50 *
51 * Remove @child from the ptrace list, move it back to the original parent,
52 * and restore the execution state so that it conforms to the group stop
53 * state.
54 *
55 * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
56 * exiting. For PTRACE_DETACH, unless the ptracee has been killed between
57 * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
58 * If the ptracer is exiting, the ptracee can be in any state.
59 *
60 * After detach, the ptracee should be in a state which conforms to the
61 * group stop. If the group is stopped or in the process of stopping, the
62 * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
63 * up from TASK_TRACED.
64 *
65 * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
66 * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
67 * to but in the opposite direction of what happens while attaching to a
68 * stopped task. However, in this direction, the intermediate RUNNING
69 * state is not hidden even from the current ptracer and if it immediately
70 * re-attaches and performs a WNOHANG wait(2), it may fail.
71 *
72 * CONTEXT:
73 * write_lock_irq(tasklist_lock)
74 */
75 void __ptrace_unlink(struct task_struct *child)
76 {
77 const struct cred *old_cred;
78 BUG_ON(!child->ptrace);
79
80 child->parent = child->real_parent;
81 list_del_init(&child->ptrace_entry);
82 old_cred = child->ptracer_cred;
83 child->ptracer_cred = NULL;
84 put_cred(old_cred);
85
86 spin_lock(&child->sighand->siglock);
87 child->ptrace = 0;
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 */
128 static 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
146 static 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 */
178 static 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
216 static 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. */
225 static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
226 {
227 const struct cred *cred = current_cred(), *tcred;
228 struct mm_struct *mm;
229 kuid_t caller_uid;
230 kgid_t caller_gid;
231
232 if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
233 WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
234 return -EPERM;
235 }
236
237 /* May we inspect the given task?
238 * This check is used both for attaching with ptrace
239 * and for allowing access to sensitive information in /proc.
240 *
241 * ptrace_attach denies several cases that /proc allows
242 * because setting up the necessary parent/child relationship
243 * or halting the specified task is impossible.
244 */
245
246 /* Don't let security modules deny introspection */
247 if (same_thread_group(task, current))
248 return 0;
249 rcu_read_lock();
250 if (mode & PTRACE_MODE_FSCREDS) {
251 caller_uid = cred->fsuid;
252 caller_gid = cred->fsgid;
253 } else {
254 /*
255 * Using the euid would make more sense here, but something
256 * in userland might rely on the old behavior, and this
257 * shouldn't be a security problem since
258 * PTRACE_MODE_REALCREDS implies that the caller explicitly
259 * used a syscall that requests access to another process
260 * (and not a filesystem syscall to procfs).
261 */
262 caller_uid = cred->uid;
263 caller_gid = cred->gid;
264 }
265 tcred = __task_cred(task);
266 if (uid_eq(caller_uid, tcred->euid) &&
267 uid_eq(caller_uid, tcred->suid) &&
268 uid_eq(caller_uid, tcred->uid) &&
269 gid_eq(caller_gid, tcred->egid) &&
270 gid_eq(caller_gid, tcred->sgid) &&
271 gid_eq(caller_gid, tcred->gid))
272 goto ok;
273 if (ptrace_has_cap(tcred->user_ns, mode))
274 goto ok;
275 rcu_read_unlock();
276 return -EPERM;
277 ok:
278 rcu_read_unlock();
279 mm = task->mm;
280 if (mm &&
281 ((get_dumpable(mm) != SUID_DUMP_USER) &&
282 !ptrace_has_cap(mm->user_ns, mode)))
283 return -EPERM;
284
285 return security_ptrace_access_check(task, mode);
286 }
287
288 bool ptrace_may_access(struct task_struct *task, unsigned int mode)
289 {
290 int err;
291 task_lock(task);
292 err = __ptrace_may_access(task, mode);
293 task_unlock(task);
294 return !err;
295 }
296
297 static int ptrace_attach(struct task_struct *task, long request,
298 unsigned long addr,
299 unsigned long flags)
300 {
301 bool seize = (request == PTRACE_SEIZE);
302 int retval;
303
304 retval = -EIO;
305 if (seize) {
306 if (addr != 0)
307 goto out;
308 if (flags & ~(unsigned long)PTRACE_O_MASK)
309 goto out;
310 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
311 } else {
312 flags = PT_PTRACED;
313 }
314
315 audit_ptrace(task);
316
317 retval = -EPERM;
318 if (unlikely(task->flags & PF_KTHREAD))
319 goto out;
320 if (same_thread_group(task, current))
321 goto out;
322
323 /*
324 * Protect exec's credential calculations against our interference;
325 * SUID, SGID and LSM creds get determined differently
326 * under ptrace.
327 */
328 retval = -ERESTARTNOINTR;
329 if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
330 goto out;
331
332 task_lock(task);
333 retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
334 task_unlock(task);
335 if (retval)
336 goto unlock_creds;
337
338 write_lock_irq(&tasklist_lock);
339 retval = -EPERM;
340 if (unlikely(task->exit_state))
341 goto unlock_tasklist;
342 if (task->ptrace)
343 goto unlock_tasklist;
344
345 if (seize)
346 flags |= PT_SEIZED;
347 task->ptrace = flags;
348
349 __ptrace_link(task, current);
350
351 /* SEIZE doesn't trap tracee on attach */
352 if (!seize)
353 send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
354
355 spin_lock(&task->sighand->siglock);
356
357 /*
358 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
359 * TRAPPING, and kick it so that it transits to TRACED. TRAPPING
360 * will be cleared if the child completes the transition or any
361 * event which clears the group stop states happens. We'll wait
362 * for the transition to complete before returning from this
363 * function.
364 *
365 * This hides STOPPED -> RUNNING -> TRACED transition from the
366 * attaching thread but a different thread in the same group can
367 * still observe the transient RUNNING state. IOW, if another
368 * thread's WNOHANG wait(2) on the stopped tracee races against
369 * ATTACH, the wait(2) may fail due to the transient RUNNING.
370 *
371 * The following task_is_stopped() test is safe as both transitions
372 * in and out of STOPPED are protected by siglock.
373 */
374 if (task_is_stopped(task) &&
375 task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
376 signal_wake_up_state(task, __TASK_STOPPED);
377
378 spin_unlock(&task->sighand->siglock);
379
380 retval = 0;
381 unlock_tasklist:
382 write_unlock_irq(&tasklist_lock);
383 unlock_creds:
384 mutex_unlock(&task->signal->cred_guard_mutex);
385 out:
386 if (!retval) {
387 /*
388 * We do not bother to change retval or clear JOBCTL_TRAPPING
389 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
390 * not return to user-mode, it will exit and clear this bit in
391 * __ptrace_unlink() if it wasn't already cleared by the tracee;
392 * and until then nobody can ptrace this task.
393 */
394 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
395 proc_ptrace_connector(task, PTRACE_ATTACH);
396 }
397
398 return retval;
399 }
400
401 /**
402 * ptrace_traceme -- helper for PTRACE_TRACEME
403 *
404 * Performs checks and sets PT_PTRACED.
405 * Should be used by all ptrace implementations for PTRACE_TRACEME.
406 */
407 static int ptrace_traceme(void)
408 {
409 int ret = -EPERM;
410
411 write_lock_irq(&tasklist_lock);
412 /* Are we already being traced? */
413 if (!current->ptrace) {
414 ret = security_ptrace_traceme(current->parent);
415 /*
416 * Check PF_EXITING to ensure ->real_parent has not passed
417 * exit_ptrace(). Otherwise we don't report the error but
418 * pretend ->real_parent untraces us right after return.
419 */
420 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
421 current->ptrace = PT_PTRACED;
422 __ptrace_link(current, current->real_parent);
423 }
424 }
425 write_unlock_irq(&tasklist_lock);
426
427 return ret;
428 }
429
430 /*
431 * Called with irqs disabled, returns true if childs should reap themselves.
432 */
433 static int ignoring_children(struct sighand_struct *sigh)
434 {
435 int ret;
436 spin_lock(&sigh->siglock);
437 ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
438 (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
439 spin_unlock(&sigh->siglock);
440 return ret;
441 }
442
443 /*
444 * Called with tasklist_lock held for writing.
445 * Unlink a traced task, and clean it up if it was a traced zombie.
446 * Return true if it needs to be reaped with release_task().
447 * (We can't call release_task() here because we already hold tasklist_lock.)
448 *
449 * If it's a zombie, our attachedness prevented normal parent notification
450 * or self-reaping. Do notification now if it would have happened earlier.
451 * If it should reap itself, return true.
452 *
453 * If it's our own child, there is no notification to do. But if our normal
454 * children self-reap, then this child was prevented by ptrace and we must
455 * reap it now, in that case we must also wake up sub-threads sleeping in
456 * do_wait().
457 */
458 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
459 {
460 bool dead;
461
462 __ptrace_unlink(p);
463
464 if (p->exit_state != EXIT_ZOMBIE)
465 return false;
466
467 dead = !thread_group_leader(p);
468
469 if (!dead && thread_group_empty(p)) {
470 if (!same_thread_group(p->real_parent, tracer))
471 dead = do_notify_parent(p, p->exit_signal);
472 else if (ignoring_children(tracer->sighand)) {
473 __wake_up_parent(p, tracer);
474 dead = true;
475 }
476 }
477 /* Mark it as in the process of being reaped. */
478 if (dead)
479 p->exit_state = EXIT_DEAD;
480 return dead;
481 }
482
483 static int ptrace_detach(struct task_struct *child, unsigned int data)
484 {
485 if (!valid_signal(data))
486 return -EIO;
487
488 /* Architecture-specific hardware disable .. */
489 ptrace_disable(child);
490 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
491
492 write_lock_irq(&tasklist_lock);
493 /*
494 * We rely on ptrace_freeze_traced(). It can't be killed and
495 * untraced by another thread, it can't be a zombie.
496 */
497 WARN_ON(!child->ptrace || child->exit_state);
498 /*
499 * tasklist_lock avoids the race with wait_task_stopped(), see
500 * the comment in ptrace_resume().
501 */
502 child->exit_code = data;
503 __ptrace_detach(current, child);
504 write_unlock_irq(&tasklist_lock);
505
506 proc_ptrace_connector(child, PTRACE_DETACH);
507
508 return 0;
509 }
510
511 /*
512 * Detach all tasks we were using ptrace on. Called with tasklist held
513 * for writing.
514 */
515 void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
516 {
517 struct task_struct *p, *n;
518
519 list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
520 if (unlikely(p->ptrace & PT_EXITKILL))
521 send_sig_info(SIGKILL, SEND_SIG_FORCED, p);
522
523 if (__ptrace_detach(tracer, p))
524 list_add(&p->ptrace_entry, dead);
525 }
526 }
527
528 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
529 {
530 int copied = 0;
531
532 while (len > 0) {
533 char buf[128];
534 int this_len, retval;
535
536 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
537 retval = access_process_vm(tsk, src, buf, this_len, 0);
538 if (!retval) {
539 if (copied)
540 break;
541 return -EIO;
542 }
543 if (copy_to_user(dst, buf, retval))
544 return -EFAULT;
545 copied += retval;
546 src += retval;
547 dst += retval;
548 len -= retval;
549 }
550 return copied;
551 }
552
553 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
554 {
555 int copied = 0;
556
557 while (len > 0) {
558 char buf[128];
559 int this_len, retval;
560
561 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
562 if (copy_from_user(buf, src, this_len))
563 return -EFAULT;
564 retval = access_process_vm(tsk, dst, buf, this_len, 1);
565 if (!retval) {
566 if (copied)
567 break;
568 return -EIO;
569 }
570 copied += retval;
571 src += retval;
572 dst += retval;
573 len -= retval;
574 }
575 return copied;
576 }
577
578 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
579 {
580 unsigned flags;
581
582 if (data & ~(unsigned long)PTRACE_O_MASK)
583 return -EINVAL;
584
585 if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
586 if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
587 !IS_ENABLED(CONFIG_SECCOMP))
588 return -EINVAL;
589
590 if (!capable(CAP_SYS_ADMIN))
591 return -EPERM;
592
593 if (seccomp_mode(¤t->seccomp) != SECCOMP_MODE_DISABLED ||
594 current->ptrace & PT_SUSPEND_SECCOMP)
595 return -EPERM;
596 }
597
598 /* Avoid intermediate state when all opts are cleared */
599 flags = child->ptrace;
600 flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
601 flags |= (data << PT_OPT_FLAG_SHIFT);
602 child->ptrace = flags;
603
604 return 0;
605 }
606
607 static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
608 {
609 unsigned long flags;
610 int error = -ESRCH;
611
612 if (lock_task_sighand(child, &flags)) {
613 error = -EINVAL;
614 if (likely(child->last_siginfo != NULL)) {
615 *info = *child->last_siginfo;
616 error = 0;
617 }
618 unlock_task_sighand(child, &flags);
619 }
620 return error;
621 }
622
623 static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
624 {
625 unsigned long flags;
626 int error = -ESRCH;
627
628 if (lock_task_sighand(child, &flags)) {
629 error = -EINVAL;
630 if (likely(child->last_siginfo != NULL)) {
631 *child->last_siginfo = *info;
632 error = 0;
633 }
634 unlock_task_sighand(child, &flags);
635 }
636 return error;
637 }
638
639 static int ptrace_peek_siginfo(struct task_struct *child,
640 unsigned long addr,
641 unsigned long data)
642 {
643 struct ptrace_peeksiginfo_args arg;
644 struct sigpending *pending;
645 struct sigqueue *q;
646 int ret, i;
647
648 ret = copy_from_user(&arg, (void __user *) addr,
649 sizeof(struct ptrace_peeksiginfo_args));
650 if (ret)
651 return -EFAULT;
652
653 if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
654 return -EINVAL; /* unknown flags */
655
656 if (arg.nr < 0)
657 return -EINVAL;
658
659 if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
660 pending = &child->signal->shared_pending;
661 else
662 pending = &child->pending;
663
664 for (i = 0; i < arg.nr; ) {
665 siginfo_t info;
666 s32 off = arg.off + i;
667
668 spin_lock_irq(&child->sighand->siglock);
669 list_for_each_entry(q, &pending->list, list) {
670 if (!off--) {
671 copy_siginfo(&info, &q->info);
672 break;
673 }
674 }
675 spin_unlock_irq(&child->sighand->siglock);
676
677 if (off >= 0) /* beyond the end of the list */
678 break;
679
680 #ifdef CONFIG_COMPAT
681 if (unlikely(in_compat_syscall())) {
682 compat_siginfo_t __user *uinfo = compat_ptr(data);
683
684 if (copy_siginfo_to_user32(uinfo, &info) ||
685 __put_user(info.si_code, &uinfo->si_code)) {
686 ret = -EFAULT;
687 break;
688 }
689
690 } else
691 #endif
692 {
693 siginfo_t __user *uinfo = (siginfo_t __user *) data;
694
695 if (copy_siginfo_to_user(uinfo, &info) ||
696 __put_user(info.si_code, &uinfo->si_code)) {
697 ret = -EFAULT;
698 break;
699 }
700 }
701
702 data += sizeof(siginfo_t);
703 i++;
704
705 if (signal_pending(current))
706 break;
707
708 cond_resched();
709 }
710
711 if (i > 0)
712 return i;
713
714 return ret;
715 }
716
717 #ifdef PTRACE_SINGLESTEP
718 #define is_singlestep(request) ((request) == PTRACE_SINGLESTEP)
719 #else
720 #define is_singlestep(request) 0
721 #endif
722
723 #ifdef PTRACE_SINGLEBLOCK
724 #define is_singleblock(request) ((request) == PTRACE_SINGLEBLOCK)
725 #else
726 #define is_singleblock(request) 0
727 #endif
728
729 #ifdef PTRACE_SYSEMU
730 #define is_sysemu_singlestep(request) ((request) == PTRACE_SYSEMU_SINGLESTEP)
731 #else
732 #define is_sysemu_singlestep(request) 0
733 #endif
734
735 static int ptrace_resume(struct task_struct *child, long request,
736 unsigned long data)
737 {
738 bool need_siglock;
739
740 if (!valid_signal(data))
741 return -EIO;
742
743 if (request == PTRACE_SYSCALL)
744 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
745 else
746 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
747
748 #ifdef TIF_SYSCALL_EMU
749 if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
750 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
751 else
752 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
753 #endif
754
755 if (is_singleblock(request)) {
756 if (unlikely(!arch_has_block_step()))
757 return -EIO;
758 user_enable_block_step(child);
759 } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
760 if (unlikely(!arch_has_single_step()))
761 return -EIO;
762 user_enable_single_step(child);
763 } else {
764 user_disable_single_step(child);
765 }
766
767 /*
768 * Change ->exit_code and ->state under siglock to avoid the race
769 * with wait_task_stopped() in between; a non-zero ->exit_code will
770 * wrongly look like another report from tracee.
771 *
772 * Note that we need siglock even if ->exit_code == data and/or this
773 * status was not reported yet, the new status must not be cleared by
774 * wait_task_stopped() after resume.
775 *
776 * If data == 0 we do not care if wait_task_stopped() reports the old
777 * status and clears the code too; this can't race with the tracee, it
778 * takes siglock after resume.
779 */
780 need_siglock = data && !thread_group_empty(current);
781 if (need_siglock)
782 spin_lock_irq(&child->sighand->siglock);
783 child->exit_code = data;
784 wake_up_state(child, __TASK_TRACED);
785 if (need_siglock)
786 spin_unlock_irq(&child->sighand->siglock);
787
788 return 0;
789 }
790
791 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
792
793 static const struct user_regset *
794 find_regset(const struct user_regset_view *view, unsigned int type)
795 {
796 const struct user_regset *regset;
797 int n;
798
799 for (n = 0; n < view->n; ++n) {
800 regset = view->regsets + n;
801 if (regset->core_note_type == type)
802 return regset;
803 }
804
805 return NULL;
806 }
807
808 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
809 struct iovec *kiov)
810 {
811 const struct user_regset_view *view = task_user_regset_view(task);
812 const struct user_regset *regset = find_regset(view, type);
813 int regset_no;
814
815 if (!regset || (kiov->iov_len % regset->size) != 0)
816 return -EINVAL;
817
818 regset_no = regset - view->regsets;
819 kiov->iov_len = min(kiov->iov_len,
820 (__kernel_size_t) (regset->n * regset->size));
821
822 if (req == PTRACE_GETREGSET)
823 return copy_regset_to_user(task, view, regset_no, 0,
824 kiov->iov_len, kiov->iov_base);
825 else
826 return copy_regset_from_user(task, view, regset_no, 0,
827 kiov->iov_len, kiov->iov_base);
828 }
829
830 /*
831 * This is declared in linux/regset.h and defined in machine-dependent
832 * code. We put the export here, near the primary machine-neutral use,
833 * to ensure no machine forgets it.
834 */
835 EXPORT_SYMBOL_GPL(task_user_regset_view);
836 #endif
837
838 int ptrace_request(struct task_struct *child, long request,
839 unsigned long addr, unsigned long data)
840 {
841 bool seized = child->ptrace & PT_SEIZED;
842 int ret = -EIO;
843 siginfo_t siginfo, *si;
844 void __user *datavp = (void __user *) data;
845 unsigned long __user *datalp = datavp;
846 unsigned long flags;
847
848 switch (request) {
849 case PTRACE_PEEKTEXT:
850 case PTRACE_PEEKDATA:
851 return generic_ptrace_peekdata(child, addr, data);
852 case PTRACE_POKETEXT:
853 case PTRACE_POKEDATA:
854 return generic_ptrace_pokedata(child, addr, data);
855
856 #ifdef PTRACE_OLDSETOPTIONS
857 case PTRACE_OLDSETOPTIONS:
858 #endif
859 case PTRACE_SETOPTIONS:
860 ret = ptrace_setoptions(child, data);
861 break;
862 case PTRACE_GETEVENTMSG:
863 ret = put_user(child->ptrace_message, datalp);
864 break;
865
866 case PTRACE_PEEKSIGINFO:
867 ret = ptrace_peek_siginfo(child, addr, data);
868 break;
869
870 case PTRACE_GETSIGINFO:
871 ret = ptrace_getsiginfo(child, &siginfo);
872 if (!ret)
873 ret = copy_siginfo_to_user(datavp, &siginfo);
874 break;
875
876 case PTRACE_SETSIGINFO:
877 if (copy_from_user(&siginfo, datavp, sizeof siginfo))
878 ret = -EFAULT;
879 else
880 ret = ptrace_setsiginfo(child, &siginfo);
881 break;
882
883 case PTRACE_GETSIGMASK:
884 if (addr != sizeof(sigset_t)) {
885 ret = -EINVAL;
886 break;
887 }
888
889 if (copy_to_user(datavp, &child->blocked, sizeof(sigset_t)))
890 ret = -EFAULT;
891 else
892 ret = 0;
893
894 break;
895
896 case PTRACE_SETSIGMASK: {
897 sigset_t new_set;
898
899 if (addr != sizeof(sigset_t)) {
900 ret = -EINVAL;
901 break;
902 }
903
904 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
905 ret = -EFAULT;
906 break;
907 }
908
909 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
910
911 /*
912 * Every thread does recalc_sigpending() after resume, so
913 * retarget_shared_pending() and recalc_sigpending() are not
914 * called here.
915 */
916 spin_lock_irq(&child->sighand->siglock);
917 child->blocked = new_set;
918 spin_unlock_irq(&child->sighand->siglock);
919
920 ret = 0;
921 break;
922 }
923
924 case PTRACE_INTERRUPT:
925 /*
926 * Stop tracee without any side-effect on signal or job
927 * control. At least one trap is guaranteed to happen
928 * after this request. If @child is already trapped, the
929 * current trap is not disturbed and another trap will
930 * happen after the current trap is ended with PTRACE_CONT.
931 *
932 * The actual trap might not be PTRACE_EVENT_STOP trap but
933 * the pending condition is cleared regardless.
934 */
935 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
936 break;
937
938 /*
939 * INTERRUPT doesn't disturb existing trap sans one
940 * exception. If ptracer issued LISTEN for the current
941 * STOP, this INTERRUPT should clear LISTEN and re-trap
942 * tracee into STOP.
943 */
944 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
945 ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
946
947 unlock_task_sighand(child, &flags);
948 ret = 0;
949 break;
950
951 case PTRACE_LISTEN:
952 /*
953 * Listen for events. Tracee must be in STOP. It's not
954 * resumed per-se but is not considered to be in TRACED by
955 * wait(2) or ptrace(2). If an async event (e.g. group
956 * stop state change) happens, tracee will enter STOP trap
957 * again. Alternatively, ptracer can issue INTERRUPT to
958 * finish listening and re-trap tracee into STOP.
959 */
960 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
961 break;
962
963 si = child->last_siginfo;
964 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
965 child->jobctl |= JOBCTL_LISTENING;
966 /*
967 * If NOTIFY is set, it means event happened between
968 * start of this trap and now. Trigger re-trap.
969 */
970 if (child->jobctl & JOBCTL_TRAP_NOTIFY)
971 ptrace_signal_wake_up(child, true);
972 ret = 0;
973 }
974 unlock_task_sighand(child, &flags);
975 break;
976
977 case PTRACE_DETACH: /* detach a process that was attached. */
978 ret = ptrace_detach(child, data);
979 break;
980
981 #ifdef CONFIG_BINFMT_ELF_FDPIC
982 case PTRACE_GETFDPIC: {
983 struct mm_struct *mm = get_task_mm(child);
984 unsigned long tmp = 0;
985
986 ret = -ESRCH;
987 if (!mm)
988 break;
989
990 switch (addr) {
991 case PTRACE_GETFDPIC_EXEC:
992 tmp = mm->context.exec_fdpic_loadmap;
993 break;
994 case PTRACE_GETFDPIC_INTERP:
995 tmp = mm->context.interp_fdpic_loadmap;
996 break;
997 default:
998 break;
999 }
1000 mmput(mm);
1001
1002 ret = put_user(tmp, datalp);
1003 break;
1004 }
1005 #endif
1006
1007 #ifdef PTRACE_SINGLESTEP
1008 case PTRACE_SINGLESTEP:
1009 #endif
1010 #ifdef PTRACE_SINGLEBLOCK
1011 case PTRACE_SINGLEBLOCK:
1012 #endif
1013 #ifdef PTRACE_SYSEMU
1014 case PTRACE_SYSEMU:
1015 case PTRACE_SYSEMU_SINGLESTEP:
1016 #endif
1017 case PTRACE_SYSCALL:
1018 case PTRACE_CONT:
1019 return ptrace_resume(child, request, data);
1020
1021 case PTRACE_KILL:
1022 if (child->exit_state) /* already dead */
1023 return 0;
1024 return ptrace_resume(child, request, SIGKILL);
1025
1026 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1027 case PTRACE_GETREGSET:
1028 case PTRACE_SETREGSET: {
1029 struct iovec kiov;
1030 struct iovec __user *uiov = datavp;
1031
1032 if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1033 return -EFAULT;
1034
1035 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1036 __get_user(kiov.iov_len, &uiov->iov_len))
1037 return -EFAULT;
1038
1039 ret = ptrace_regset(child, request, addr, &kiov);
1040 if (!ret)
1041 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1042 break;
1043 }
1044 #endif
1045
1046 case PTRACE_SECCOMP_GET_FILTER:
1047 ret = seccomp_get_filter(child, addr, datavp);
1048 break;
1049
1050 default:
1051 break;
1052 }
1053
1054 return ret;
1055 }
1056
1057 static struct task_struct *ptrace_get_task_struct(pid_t pid)
1058 {
1059 struct task_struct *child;
1060
1061 rcu_read_lock();
1062 child = find_task_by_vpid(pid);
1063 if (child)
1064 get_task_struct(child);
1065 rcu_read_unlock();
1066
1067 if (!child)
1068 return ERR_PTR(-ESRCH);
1069 return child;
1070 }
1071
1072 #ifndef arch_ptrace_attach
1073 #define arch_ptrace_attach(child) do { } while (0)
1074 #endif
1075
1076 SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1077 unsigned long, data)
1078 {
1079 struct task_struct *child;
1080 long ret;
1081 {
1082 const int rc = ccs_ptrace_permission(request, pid);
1083 if (rc)
1084 return rc;
1085 }
1086
1087 if (request == PTRACE_TRACEME) {
1088 ret = ptrace_traceme();
1089 if (!ret)
1090 arch_ptrace_attach(current);
1091 goto out;
1092 }
1093
1094 child = ptrace_get_task_struct(pid);
1095 if (IS_ERR(child)) {
1096 ret = PTR_ERR(child);
1097 goto out;
1098 }
1099
1100 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1101 ret = ptrace_attach(child, request, addr, data);
1102 /*
1103 * Some architectures need to do book-keeping after
1104 * a ptrace attach.
1105 */
1106 if (!ret)
1107 arch_ptrace_attach(child);
1108 goto out_put_task_struct;
1109 }
1110
1111 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1112 request == PTRACE_INTERRUPT);
1113 if (ret < 0)
1114 goto out_put_task_struct;
1115
1116 ret = arch_ptrace(child, request, addr, data);
1117 if (ret || request != PTRACE_DETACH)
1118 ptrace_unfreeze_traced(child);
1119
1120 out_put_task_struct:
1121 put_task_struct(child);
1122 out:
1123 return ret;
1124 }
1125
1126 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1127 unsigned long data)
1128 {
1129 unsigned long tmp;
1130 int copied;
1131
1132 copied = access_process_vm(tsk, addr, &tmp, sizeof(tmp), 0);
1133 if (copied != sizeof(tmp))
1134 return -EIO;
1135 return put_user(tmp, (unsigned long __user *)data);
1136 }
1137
1138 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1139 unsigned long data)
1140 {
1141 int copied;
1142
1143 copied = access_process_vm(tsk, addr, &data, sizeof(data), 1);
1144 return (copied == sizeof(data)) ? 0 : -EIO;
1145 }
1146
1147 #if defined CONFIG_COMPAT
1148
1149 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1150 compat_ulong_t addr, compat_ulong_t data)
1151 {
1152 compat_ulong_t __user *datap = compat_ptr(data);
1153 compat_ulong_t word;
1154 siginfo_t siginfo;
1155 int ret;
1156
1157 switch (request) {
1158 case PTRACE_PEEKTEXT:
1159 case PTRACE_PEEKDATA:
1160 ret = access_process_vm(child, addr, &word, sizeof(word), 0);
1161 if (ret != sizeof(word))
1162 ret = -EIO;
1163 else
1164 ret = put_user(word, datap);
1165 break;
1166
1167 case PTRACE_POKETEXT:
1168 case PTRACE_POKEDATA:
1169 ret = access_process_vm(child, addr, &data, sizeof(data), 1);
1170 ret = (ret != sizeof(data) ? -EIO : 0);
1171 break;
1172
1173 case PTRACE_GETEVENTMSG:
1174 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1175 break;
1176
1177 case PTRACE_GETSIGINFO:
1178 ret = ptrace_getsiginfo(child, &siginfo);
1179 if (!ret)
1180 ret = copy_siginfo_to_user32(
1181 (struct compat_siginfo __user *) datap,
1182 &siginfo);
1183 break;
1184
1185 case PTRACE_SETSIGINFO:
1186 memset(&siginfo, 0, sizeof siginfo);
1187 if (copy_siginfo_from_user32(
1188 &siginfo, (struct compat_siginfo __user *) datap))
1189 ret = -EFAULT;
1190 else
1191 ret = ptrace_setsiginfo(child, &siginfo);
1192 break;
1193 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1194 case PTRACE_GETREGSET:
1195 case PTRACE_SETREGSET:
1196 {
1197 struct iovec kiov;
1198 struct compat_iovec __user *uiov =
1199 (struct compat_iovec __user *) datap;
1200 compat_uptr_t ptr;
1201 compat_size_t len;
1202
1203 if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1204 return -EFAULT;
1205
1206 if (__get_user(ptr, &uiov->iov_base) ||
1207 __get_user(len, &uiov->iov_len))
1208 return -EFAULT;
1209
1210 kiov.iov_base = compat_ptr(ptr);
1211 kiov.iov_len = len;
1212
1213 ret = ptrace_regset(child, request, addr, &kiov);
1214 if (!ret)
1215 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1216 break;
1217 }
1218 #endif
1219
1220 default:
1221 ret = ptrace_request(child, request, addr, data);
1222 }
1223
1224 return ret;
1225 }
1226
1227 COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1228 compat_long_t, addr, compat_long_t, data)
1229 {
1230 struct task_struct *child;
1231 long ret;
1232 {
1233 const int rc = ccs_ptrace_permission(request, pid);
1234 if (rc)
1235 return rc;
1236 }
1237
1238 if (request == PTRACE_TRACEME) {
1239 ret = ptrace_traceme();
1240 goto out;
1241 }
1242
1243 child = ptrace_get_task_struct(pid);
1244 if (IS_ERR(child)) {
1245 ret = PTR_ERR(child);
1246 goto out;
1247 }
1248
1249 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1250 ret = ptrace_attach(child, request, addr, data);
1251 /*
1252 * Some architectures need to do book-keeping after
1253 * a ptrace attach.
1254 */
1255 if (!ret)
1256 arch_ptrace_attach(child);
1257 goto out_put_task_struct;
1258 }
1259
1260 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1261 request == PTRACE_INTERRUPT);
1262 if (!ret) {
1263 ret = compat_arch_ptrace(child, request, addr, data);
1264 if (ret || request != PTRACE_DETACH)
1265 ptrace_unfreeze_traced(child);
1266 }
1267
1268 out_put_task_struct:
1269 put_task_struct(child);
1270 out:
1271 return ret;
1272 }
1273 #endif /* CONFIG_COMPAT */
1274
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.