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/*
31 * Access another process' address space via ptrace.
32 * Source/target buffer must be kernel space,
33 * Do not walk the page table directly, use get_user_pages
34 */
35int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
36 void *buf, int len, unsigned int gup_flags)
37{
38 struct mm_struct *mm;
39 int ret;
40
41 mm = get_task_mm(tsk);
42 if (!mm)
43 return 0;
44
45 if (!tsk->ptrace ||
46 (current != tsk->parent) ||
47 ((get_dumpable(mm) != SUID_DUMP_USER) &&
48 !ptracer_capable(tsk, mm->user_ns))) {
49 mmput(mm);
50 return 0;
51 }
52
53 ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
54 mmput(mm);
55
56 return ret;
57}
58
59
60/*
61 * ptrace a task: make the debugger its new parent and
62 * move it to the ptrace list.
63 *
64 * Must be called with the tasklist lock write-held.
65 */
66void __ptrace_link(struct task_struct *child, struct task_struct *new_parent)
67{
68 BUG_ON(!list_empty(&child->ptrace_entry));
69 list_add(&child->ptrace_entry, &new_parent->ptraced);
70 child->parent = new_parent;
71 rcu_read_lock();
72 child->ptracer_cred = get_cred(__task_cred(new_parent));
73 rcu_read_unlock();
74}
75
76/**
77 * __ptrace_unlink - unlink ptracee and restore its execution state
78 * @child: ptracee to be unlinked
79 *
80 * Remove @child from the ptrace list, move it back to the original parent,
81 * and restore the execution state so that it conforms to the group stop
82 * state.
83 *
84 * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
85 * exiting. For PTRACE_DETACH, unless the ptracee has been killed between
86 * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
87 * If the ptracer is exiting, the ptracee can be in any state.
88 *
89 * After detach, the ptracee should be in a state which conforms to the
90 * group stop. If the group is stopped or in the process of stopping, the
91 * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
92 * up from TASK_TRACED.
93 *
94 * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
95 * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
96 * to but in the opposite direction of what happens while attaching to a
97 * stopped task. However, in this direction, the intermediate RUNNING
98 * state is not hidden even from the current ptracer and if it immediately
99 * re-attaches and performs a WNOHANG wait(2), it may fail.
100 *
101 * CONTEXT:
102 * write_lock_irq(tasklist_lock)
103 */
104void __ptrace_unlink(struct task_struct *child)
105{
106 const struct cred *old_cred;
107 BUG_ON(!child->ptrace);
108
109 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
110
111 child->parent = child->real_parent;
112 list_del_init(&child->ptrace_entry);
113 old_cred = child->ptracer_cred;
114 child->ptracer_cred = NULL;
115 put_cred(old_cred);
116
117 spin_lock(&child->sighand->siglock);
118 child->ptrace = 0;
119 /*
120 * Clear all pending traps and TRAPPING. TRAPPING should be
121 * cleared regardless of JOBCTL_STOP_PENDING. Do it explicitly.
122 */
123 task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
124 task_clear_jobctl_trapping(child);
125
126 /*
127 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
128 * @child isn't dead.
129 */
130 if (!(child->flags & PF_EXITING) &&
131 (child->signal->flags & SIGNAL_STOP_STOPPED ||
132 child->signal->group_stop_count)) {
133 child->jobctl |= JOBCTL_STOP_PENDING;
134
135 /*
136 * This is only possible if this thread was cloned by the
137 * traced task running in the stopped group, set the signal
138 * for the future reports.
139 * FIXME: we should change ptrace_init_task() to handle this
140 * case.
141 */
142 if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
143 child->jobctl |= SIGSTOP;
144 }
145
146 /*
147 * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
148 * @child in the butt. Note that @resume should be used iff @child
149 * is in TASK_TRACED; otherwise, we might unduly disrupt
150 * TASK_KILLABLE sleeps.
151 */
152 if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
153 ptrace_signal_wake_up(child, true);
154
155 spin_unlock(&child->sighand->siglock);
156}
157
158/* Ensure that nothing can wake it up, even SIGKILL */
159static bool ptrace_freeze_traced(struct task_struct *task)
160{
161 bool ret = false;
162
163 /* Lockless, nobody but us can set this flag */
164 if (task->jobctl & JOBCTL_LISTENING)
165 return ret;
166
167 spin_lock_irq(&task->sighand->siglock);
168 if (task_is_traced(task) && !__fatal_signal_pending(task)) {
169 task->state = __TASK_TRACED;
170 ret = true;
171 }
172 spin_unlock_irq(&task->sighand->siglock);
173
174 return ret;
175}
176
177static void ptrace_unfreeze_traced(struct task_struct *task)
178{
179 if (task->state != __TASK_TRACED)
180 return;
181
182 WARN_ON(!task->ptrace || task->parent != current);
183
184 /*
185 * PTRACE_LISTEN can allow ptrace_trap_notify to wake us up remotely.
186 * Recheck state under the lock to close this race.
187 */
188 spin_lock_irq(&task->sighand->siglock);
189 if (task->state == __TASK_TRACED) {
190 if (__fatal_signal_pending(task))
191 wake_up_state(task, __TASK_TRACED);
192 else
193 task->state = TASK_TRACED;
194 }
195 spin_unlock_irq(&task->sighand->siglock);
196}
197
198/**
199 * ptrace_check_attach - check whether ptracee is ready for ptrace operation
200 * @child: ptracee to check for
201 * @ignore_state: don't check whether @child is currently %TASK_TRACED
202 *
203 * Check whether @child is being ptraced by %current and ready for further
204 * ptrace operations. If @ignore_state is %false, @child also should be in
205 * %TASK_TRACED state and on return the child is guaranteed to be traced
206 * and not executing. If @ignore_state is %true, @child can be in any
207 * state.
208 *
209 * CONTEXT:
210 * Grabs and releases tasklist_lock and @child->sighand->siglock.
211 *
212 * RETURNS:
213 * 0 on success, -ESRCH if %child is not ready.
214 */
215static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
216{
217 int ret = -ESRCH;
218
219 /*
220 * We take the read lock around doing both checks to close a
221 * possible race where someone else was tracing our child and
222 * detached between these two checks. After this locked check,
223 * we are sure that this is our traced child and that can only
224 * be changed by us so it's not changing right after this.
225 */
226 read_lock(&tasklist_lock);
227 if (child->ptrace && child->parent == current) {
228 WARN_ON(child->state == __TASK_TRACED);
229 /*
230 * child->sighand can't be NULL, release_task()
231 * does ptrace_unlink() before __exit_signal().
232 */
233 if (ignore_state || ptrace_freeze_traced(child))
234 ret = 0;
235 }
236 read_unlock(&tasklist_lock);
237
238 if (!ret && !ignore_state) {
239 if (!wait_task_inactive(child, __TASK_TRACED)) {
240 /*
241 * This can only happen if may_ptrace_stop() fails and
242 * ptrace_stop() changes ->state back to TASK_RUNNING,
243 * so we should not worry about leaking __TASK_TRACED.
244 */
245 WARN_ON(child->state == __TASK_TRACED);
246 ret = -ESRCH;
247 }
248 }
249
250 return ret;
251}
252
253static int ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
254{
255 if (mode & PTRACE_MODE_NOAUDIT)
256 return has_ns_capability_noaudit(current, ns, CAP_SYS_PTRACE);
257 else
258 return has_ns_capability(current, ns, CAP_SYS_PTRACE);
259}
260
261/* Returns 0 on success, -errno on denial. */
262static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
263{
264 const struct cred *cred = current_cred(), *tcred;
265 struct mm_struct *mm;
266 kuid_t caller_uid;
267 kgid_t caller_gid;
268
269 if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
270 WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
271 return -EPERM;
272 }
273
274 /* May we inspect the given task?
275 * This check is used both for attaching with ptrace
276 * and for allowing access to sensitive information in /proc.
277 *
278 * ptrace_attach denies several cases that /proc allows
279 * because setting up the necessary parent/child relationship
280 * or halting the specified task is impossible.
281 */
282
283 /* Don't let security modules deny introspection */
284 if (same_thread_group(task, current))
285 return 0;
286 rcu_read_lock();
287 if (mode & PTRACE_MODE_FSCREDS) {
288 caller_uid = cred->fsuid;
289 caller_gid = cred->fsgid;
290 } else {
291 /*
292 * Using the euid would make more sense here, but something
293 * in userland might rely on the old behavior, and this
294 * shouldn't be a security problem since
295 * PTRACE_MODE_REALCREDS implies that the caller explicitly
296 * used a syscall that requests access to another process
297 * (and not a filesystem syscall to procfs).
298 */
299 caller_uid = cred->uid;
300 caller_gid = cred->gid;
301 }
302 tcred = __task_cred(task);
303 if (uid_eq(caller_uid, tcred->euid) &&
304 uid_eq(caller_uid, tcred->suid) &&
305 uid_eq(caller_uid, tcred->uid) &&
306 gid_eq(caller_gid, tcred->egid) &&
307 gid_eq(caller_gid, tcred->sgid) &&
308 gid_eq(caller_gid, tcred->gid))
309 goto ok;
310 if (ptrace_has_cap(tcred->user_ns, mode))
311 goto ok;
312 rcu_read_unlock();
313 return -EPERM;
314ok:
315 rcu_read_unlock();
316 mm = task->mm;
317 if (mm &&
318 ((get_dumpable(mm) != SUID_DUMP_USER) &&
319 !ptrace_has_cap(mm->user_ns, mode)))
320 return -EPERM;
321
322 return security_ptrace_access_check(task, mode);
323}
324
325bool ptrace_may_access(struct task_struct *task, unsigned int mode)
326{
327 int err;
328 task_lock(task);
329 err = __ptrace_may_access(task, mode);
330 task_unlock(task);
331 return !err;
332}
333
334static int ptrace_attach(struct task_struct *task, long request,
335 unsigned long addr,
336 unsigned long flags)
337{
338 bool seize = (request == PTRACE_SEIZE);
339 int retval;
340
341 retval = -EIO;
342 if (seize) {
343 if (addr != 0)
344 goto out;
345 if (flags & ~(unsigned long)PTRACE_O_MASK)
346 goto out;
347 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
348 } else {
349 flags = PT_PTRACED;
350 }
351
352 audit_ptrace(task);
353
354 retval = -EPERM;
355 if (unlikely(task->flags & PF_KTHREAD))
356 goto out;
357 if (same_thread_group(task, current))
358 goto out;
359
360 /*
361 * Protect exec's credential calculations against our interference;
362 * SUID, SGID and LSM creds get determined differently
363 * under ptrace.
364 */
365 retval = -ERESTARTNOINTR;
366 if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
367 goto out;
368
369 task_lock(task);
370 retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
371 task_unlock(task);
372 if (retval)
373 goto unlock_creds;
374
375 write_lock_irq(&tasklist_lock);
376 retval = -EPERM;
377 if (unlikely(task->exit_state))
378 goto unlock_tasklist;
379 if (task->ptrace)
380 goto unlock_tasklist;
381
382 if (seize)
383 flags |= PT_SEIZED;
384 task->ptrace = flags;
385
386 __ptrace_link(task, current);
387
388 /* SEIZE doesn't trap tracee on attach */
389 if (!seize)
390 send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
391
392 spin_lock(&task->sighand->siglock);
393
394 /*
395 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
396 * TRAPPING, and kick it so that it transits to TRACED. TRAPPING
397 * will be cleared if the child completes the transition or any
398 * event which clears the group stop states happens. We'll wait
399 * for the transition to complete before returning from this
400 * function.
401 *
402 * This hides STOPPED -> RUNNING -> TRACED transition from the
403 * attaching thread but a different thread in the same group can
404 * still observe the transient RUNNING state. IOW, if another
405 * thread's WNOHANG wait(2) on the stopped tracee races against
406 * ATTACH, the wait(2) may fail due to the transient RUNNING.
407 *
408 * The following task_is_stopped() test is safe as both transitions
409 * in and out of STOPPED are protected by siglock.
410 */
411 if (task_is_stopped(task) &&
412 task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
413 signal_wake_up_state(task, __TASK_STOPPED);
414
415 spin_unlock(&task->sighand->siglock);
416
417 retval = 0;
418unlock_tasklist:
419 write_unlock_irq(&tasklist_lock);
420unlock_creds:
421 mutex_unlock(&task->signal->cred_guard_mutex);
422out:
423 if (!retval) {
424 /*
425 * We do not bother to change retval or clear JOBCTL_TRAPPING
426 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
427 * not return to user-mode, it will exit and clear this bit in
428 * __ptrace_unlink() if it wasn't already cleared by the tracee;
429 * and until then nobody can ptrace this task.
430 */
431 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
432 proc_ptrace_connector(task, PTRACE_ATTACH);
433 }
434
435 return retval;
436}
437
438/**
439 * ptrace_traceme -- helper for PTRACE_TRACEME
440 *
441 * Performs checks and sets PT_PTRACED.
442 * Should be used by all ptrace implementations for PTRACE_TRACEME.
443 */
444static int ptrace_traceme(void)
445{
446 int ret = -EPERM;
447
448 write_lock_irq(&tasklist_lock);
449 /* Are we already being traced? */
450 if (!current->ptrace) {
451 ret = security_ptrace_traceme(current->parent);
452 /*
453 * Check PF_EXITING to ensure ->real_parent has not passed
454 * exit_ptrace(). Otherwise we don't report the error but
455 * pretend ->real_parent untraces us right after return.
456 */
457 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
458 current->ptrace = PT_PTRACED;
459 __ptrace_link(current, current->real_parent);
460 }
461 }
462 write_unlock_irq(&tasklist_lock);
463
464 return ret;
465}
466
467/*
468 * Called with irqs disabled, returns true if childs should reap themselves.
469 */
470static int ignoring_children(struct sighand_struct *sigh)
471{
472 int ret;
473 spin_lock(&sigh->siglock);
474 ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
475 (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
476 spin_unlock(&sigh->siglock);
477 return ret;
478}
479
480/*
481 * Called with tasklist_lock held for writing.
482 * Unlink a traced task, and clean it up if it was a traced zombie.
483 * Return true if it needs to be reaped with release_task().
484 * (We can't call release_task() here because we already hold tasklist_lock.)
485 *
486 * If it's a zombie, our attachedness prevented normal parent notification
487 * or self-reaping. Do notification now if it would have happened earlier.
488 * If it should reap itself, return true.
489 *
490 * If it's our own child, there is no notification to do. But if our normal
491 * children self-reap, then this child was prevented by ptrace and we must
492 * reap it now, in that case we must also wake up sub-threads sleeping in
493 * do_wait().
494 */
495static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
496{
497 bool dead;
498
499 __ptrace_unlink(p);
500
501 if (p->exit_state != EXIT_ZOMBIE)
502 return false;
503
504 dead = !thread_group_leader(p);
505
506 if (!dead && thread_group_empty(p)) {
507 if (!same_thread_group(p->real_parent, tracer))
508 dead = do_notify_parent(p, p->exit_signal);
509 else if (ignoring_children(tracer->sighand)) {
510 __wake_up_parent(p, tracer);
511 dead = true;
512 }
513 }
514 /* Mark it as in the process of being reaped. */
515 if (dead)
516 p->exit_state = EXIT_DEAD;
517 return dead;
518}
519
520static int ptrace_detach(struct task_struct *child, unsigned int data)
521{
522 if (!valid_signal(data))
523 return -EIO;
524
525 /* Architecture-specific hardware disable .. */
526 ptrace_disable(child);
527
528 write_lock_irq(&tasklist_lock);
529 /*
530 * We rely on ptrace_freeze_traced(). It can't be killed and
531 * untraced by another thread, it can't be a zombie.
532 */
533 WARN_ON(!child->ptrace || child->exit_state);
534 /*
535 * tasklist_lock avoids the race with wait_task_stopped(), see
536 * the comment in ptrace_resume().
537 */
538 child->exit_code = data;
539 __ptrace_detach(current, child);
540 write_unlock_irq(&tasklist_lock);
541
542 proc_ptrace_connector(child, PTRACE_DETACH);
543
544 return 0;
545}
546
547/*
548 * Detach all tasks we were using ptrace on. Called with tasklist held
549 * for writing.
550 */
551void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
552{
553 struct task_struct *p, *n;
554
555 list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
556 if (unlikely(p->ptrace & PT_EXITKILL))
557 send_sig_info(SIGKILL, SEND_SIG_FORCED, p);
558
559 if (__ptrace_detach(tracer, p))
560 list_add(&p->ptrace_entry, dead);
561 }
562}
563
564int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
565{
566 int copied = 0;
567
568 while (len > 0) {
569 char buf[128];
570 int this_len, retval;
571
572 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
573 retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
574
575 if (!retval) {
576 if (copied)
577 break;
578 return -EIO;
579 }
580 if (copy_to_user(dst, buf, retval))
581 return -EFAULT;
582 copied += retval;
583 src += retval;
584 dst += retval;
585 len -= retval;
586 }
587 return copied;
588}
589
590int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
591{
592 int copied = 0;
593
594 while (len > 0) {
595 char buf[128];
596 int this_len, retval;
597
598 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
599 if (copy_from_user(buf, src, this_len))
600 return -EFAULT;
601 retval = ptrace_access_vm(tsk, dst, buf, this_len,
602 FOLL_FORCE | FOLL_WRITE);
603 if (!retval) {
604 if (copied)
605 break;
606 return -EIO;
607 }
608 copied += retval;
609 src += retval;
610 dst += retval;
611 len -= retval;
612 }
613 return copied;
614}
615
616static int ptrace_setoptions(struct task_struct *child, unsigned long data)
617{
618 unsigned flags;
619
620 if (data & ~(unsigned long)PTRACE_O_MASK)
621 return -EINVAL;
622
623 if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
624 if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
625 !IS_ENABLED(CONFIG_SECCOMP))
626 return -EINVAL;
627
628 if (!capable(CAP_SYS_ADMIN))
629 return -EPERM;
630
631 if (seccomp_mode(¤t->seccomp) != SECCOMP_MODE_DISABLED ||
632 current->ptrace & PT_SUSPEND_SECCOMP)
633 return -EPERM;
634 }
635
636 /* Avoid intermediate state when all opts are cleared */
637 flags = child->ptrace;
638 flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
639 flags |= (data << PT_OPT_FLAG_SHIFT);
640 child->ptrace = flags;
641
642 return 0;
643}
644
645static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
646{
647 unsigned long flags;
648 int error = -ESRCH;
649
650 if (lock_task_sighand(child, &flags)) {
651 error = -EINVAL;
652 if (likely(child->last_siginfo != NULL)) {
653 *info = *child->last_siginfo;
654 error = 0;
655 }
656 unlock_task_sighand(child, &flags);
657 }
658 return error;
659}
660
661static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
662{
663 unsigned long flags;
664 int error = -ESRCH;
665
666 if (lock_task_sighand(child, &flags)) {
667 error = -EINVAL;
668 if (likely(child->last_siginfo != NULL)) {
669 *child->last_siginfo = *info;
670 error = 0;
671 }
672 unlock_task_sighand(child, &flags);
673 }
674 return error;
675}
676
677static int ptrace_peek_siginfo(struct task_struct *child,
678 unsigned long addr,
679 unsigned long data)
680{
681 struct ptrace_peeksiginfo_args arg;
682 struct sigpending *pending;
683 struct sigqueue *q;
684 int ret, i;
685
686 ret = copy_from_user(&arg, (void __user *) addr,
687 sizeof(struct ptrace_peeksiginfo_args));
688 if (ret)
689 return -EFAULT;
690
691 if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
692 return -EINVAL; /* unknown flags */
693
694 if (arg.nr < 0)
695 return -EINVAL;
696
697 if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
698 pending = &child->signal->shared_pending;
699 else
700 pending = &child->pending;
701
702 for (i = 0; i < arg.nr; ) {
703 siginfo_t info;
704 s32 off = arg.off + i;
705
706 spin_lock_irq(&child->sighand->siglock);
707 list_for_each_entry(q, &pending->list, list) {
708 if (!off--) {
709 copy_siginfo(&info, &q->info);
710 break;
711 }
712 }
713 spin_unlock_irq(&child->sighand->siglock);
714
715 if (off >= 0) /* beyond the end of the list */
716 break;
717
718#ifdef CONFIG_COMPAT
719 if (unlikely(in_compat_syscall())) {
720 compat_siginfo_t __user *uinfo = compat_ptr(data);
721
722 if (copy_siginfo_to_user32(uinfo, &info) ||
723 __put_user(info.si_code, &uinfo->si_code)) {
724 ret = -EFAULT;
725 break;
726 }
727
728 } else
729#endif
730 {
731 siginfo_t __user *uinfo = (siginfo_t __user *) data;
732
733 if (copy_siginfo_to_user(uinfo, &info) ||
734 __put_user(info.si_code, &uinfo->si_code)) {
735 ret = -EFAULT;
736 break;
737 }
738 }
739
740 data += sizeof(siginfo_t);
741 i++;
742
743 if (signal_pending(current))
744 break;
745
746 cond_resched();
747 }
748
749 if (i > 0)
750 return i;
751
752 return ret;
753}
754
755#ifdef PTRACE_SINGLESTEP
756#define is_singlestep(request) ((request) == PTRACE_SINGLESTEP)
757#else
758#define is_singlestep(request) 0
759#endif
760
761#ifdef PTRACE_SINGLEBLOCK
762#define is_singleblock(request) ((request) == PTRACE_SINGLEBLOCK)
763#else
764#define is_singleblock(request) 0
765#endif
766
767#ifdef PTRACE_SYSEMU
768#define is_sysemu_singlestep(request) ((request) == PTRACE_SYSEMU_SINGLESTEP)
769#else
770#define is_sysemu_singlestep(request) 0
771#endif
772
773static int ptrace_resume(struct task_struct *child, long request,
774 unsigned long data)
775{
776 bool need_siglock;
777
778 if (!valid_signal(data))
779 return -EIO;
780
781 if (request == PTRACE_SYSCALL)
782 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
783 else
784 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
785
786#ifdef TIF_SYSCALL_EMU
787 if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
788 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
789 else
790 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
791#endif
792
793 if (is_singleblock(request)) {
794 if (unlikely(!arch_has_block_step()))
795 return -EIO;
796 user_enable_block_step(child);
797 } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
798 if (unlikely(!arch_has_single_step()))
799 return -EIO;
800 user_enable_single_step(child);
801 } else {
802 user_disable_single_step(child);
803 }
804
805 /*
806 * Change ->exit_code and ->state under siglock to avoid the race
807 * with wait_task_stopped() in between; a non-zero ->exit_code will
808 * wrongly look like another report from tracee.
809 *
810 * Note that we need siglock even if ->exit_code == data and/or this
811 * status was not reported yet, the new status must not be cleared by
812 * wait_task_stopped() after resume.
813 *
814 * If data == 0 we do not care if wait_task_stopped() reports the old
815 * status and clears the code too; this can't race with the tracee, it
816 * takes siglock after resume.
817 */
818 need_siglock = data && !thread_group_empty(current);
819 if (need_siglock)
820 spin_lock_irq(&child->sighand->siglock);
821 child->exit_code = data;
822 wake_up_state(child, __TASK_TRACED);
823 if (need_siglock)
824 spin_unlock_irq(&child->sighand->siglock);
825
826 return 0;
827}
828
829#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
830
831static const struct user_regset *
832find_regset(const struct user_regset_view *view, unsigned int type)
833{
834 const struct user_regset *regset;
835 int n;
836
837 for (n = 0; n < view->n; ++n) {
838 regset = view->regsets + n;
839 if (regset->core_note_type == type)
840 return regset;
841 }
842
843 return NULL;
844}
845
846static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
847 struct iovec *kiov)
848{
849 const struct user_regset_view *view = task_user_regset_view(task);
850 const struct user_regset *regset = find_regset(view, type);
851 int regset_no;
852
853 if (!regset || (kiov->iov_len % regset->size) != 0)
854 return -EINVAL;
855
856 regset_no = regset - view->regsets;
857 kiov->iov_len = min(kiov->iov_len,
858 (__kernel_size_t) (regset->n * regset->size));
859
860 if (req == PTRACE_GETREGSET)
861 return copy_regset_to_user(task, view, regset_no, 0,
862 kiov->iov_len, kiov->iov_base);
863 else
864 return copy_regset_from_user(task, view, regset_no, 0,
865 kiov->iov_len, kiov->iov_base);
866}
867
868/*
869 * This is declared in linux/regset.h and defined in machine-dependent
870 * code. We put the export here, near the primary machine-neutral use,
871 * to ensure no machine forgets it.
872 */
873EXPORT_SYMBOL_GPL(task_user_regset_view);
874#endif
875
876int ptrace_request(struct task_struct *child, long request,
877 unsigned long addr, unsigned long data)
878{
879 bool seized = child->ptrace & PT_SEIZED;
880 int ret = -EIO;
881 siginfo_t siginfo, *si;
882 void __user *datavp = (void __user *) data;
883 unsigned long __user *datalp = datavp;
884 unsigned long flags;
885
886 switch (request) {
887 case PTRACE_PEEKTEXT:
888 case PTRACE_PEEKDATA:
889 return generic_ptrace_peekdata(child, addr, data);
890 case PTRACE_POKETEXT:
891 case PTRACE_POKEDATA:
892 return generic_ptrace_pokedata(child, addr, data);
893
894#ifdef PTRACE_OLDSETOPTIONS
895 case PTRACE_OLDSETOPTIONS:
896#endif
897 case PTRACE_SETOPTIONS:
898 ret = ptrace_setoptions(child, data);
899 break;
900 case PTRACE_GETEVENTMSG:
901 ret = put_user(child->ptrace_message, datalp);
902 break;
903
904 case PTRACE_PEEKSIGINFO:
905 ret = ptrace_peek_siginfo(child, addr, data);
906 break;
907
908 case PTRACE_GETSIGINFO:
909 ret = ptrace_getsiginfo(child, &siginfo);
910 if (!ret)
911 ret = copy_siginfo_to_user(datavp, &siginfo);
912 break;
913
914 case PTRACE_SETSIGINFO:
915 if (copy_from_user(&siginfo, datavp, sizeof siginfo))
916 ret = -EFAULT;
917 else
918 ret = ptrace_setsiginfo(child, &siginfo);
919 break;
920
921 case PTRACE_GETSIGMASK:
922 if (addr != sizeof(sigset_t)) {
923 ret = -EINVAL;
924 break;
925 }
926
927 if (copy_to_user(datavp, &child->blocked, sizeof(sigset_t)))
928 ret = -EFAULT;
929 else
930 ret = 0;
931
932 break;
933
934 case PTRACE_SETSIGMASK: {
935 sigset_t new_set;
936
937 if (addr != sizeof(sigset_t)) {
938 ret = -EINVAL;
939 break;
940 }
941
942 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
943 ret = -EFAULT;
944 break;
945 }
946
947 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
948
949 /*
950 * Every thread does recalc_sigpending() after resume, so
951 * retarget_shared_pending() and recalc_sigpending() are not
952 * called here.
953 */
954 spin_lock_irq(&child->sighand->siglock);
955 child->blocked = new_set;
956 spin_unlock_irq(&child->sighand->siglock);
957
958 ret = 0;
959 break;
960 }
961
962 case PTRACE_INTERRUPT:
963 /*
964 * Stop tracee without any side-effect on signal or job
965 * control. At least one trap is guaranteed to happen
966 * after this request. If @child is already trapped, the
967 * current trap is not disturbed and another trap will
968 * happen after the current trap is ended with PTRACE_CONT.
969 *
970 * The actual trap might not be PTRACE_EVENT_STOP trap but
971 * the pending condition is cleared regardless.
972 */
973 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
974 break;
975
976 /*
977 * INTERRUPT doesn't disturb existing trap sans one
978 * exception. If ptracer issued LISTEN for the current
979 * STOP, this INTERRUPT should clear LISTEN and re-trap
980 * tracee into STOP.
981 */
982 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
983 ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
984
985 unlock_task_sighand(child, &flags);
986 ret = 0;
987 break;
988
989 case PTRACE_LISTEN:
990 /*
991 * Listen for events. Tracee must be in STOP. It's not
992 * resumed per-se but is not considered to be in TRACED by
993 * wait(2) or ptrace(2). If an async event (e.g. group
994 * stop state change) happens, tracee will enter STOP trap
995 * again. Alternatively, ptracer can issue INTERRUPT to
996 * finish listening and re-trap tracee into STOP.
997 */
998 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
999 break;
1000
1001 si = child->last_siginfo;
1002 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1003 child->jobctl |= JOBCTL_LISTENING;
1004 /*
1005 * If NOTIFY is set, it means event happened between
1006 * start of this trap and now. Trigger re-trap.
1007 */
1008 if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1009 ptrace_signal_wake_up(child, true);
1010 ret = 0;
1011 }
1012 unlock_task_sighand(child, &flags);
1013 break;
1014
1015 case PTRACE_DETACH: /* detach a process that was attached. */
1016 ret = ptrace_detach(child, data);
1017 break;
1018
1019#ifdef CONFIG_BINFMT_ELF_FDPIC
1020 case PTRACE_GETFDPIC: {
1021 struct mm_struct *mm = get_task_mm(child);
1022 unsigned long tmp = 0;
1023
1024 ret = -ESRCH;
1025 if (!mm)
1026 break;
1027
1028 switch (addr) {
1029 case PTRACE_GETFDPIC_EXEC:
1030 tmp = mm->context.exec_fdpic_loadmap;
1031 break;
1032 case PTRACE_GETFDPIC_INTERP:
1033 tmp = mm->context.interp_fdpic_loadmap;
1034 break;
1035 default:
1036 break;
1037 }
1038 mmput(mm);
1039
1040 ret = put_user(tmp, datalp);
1041 break;
1042 }
1043#endif
1044
1045#ifdef PTRACE_SINGLESTEP
1046 case PTRACE_SINGLESTEP:
1047#endif
1048#ifdef PTRACE_SINGLEBLOCK
1049 case PTRACE_SINGLEBLOCK:
1050#endif
1051#ifdef PTRACE_SYSEMU
1052 case PTRACE_SYSEMU:
1053 case PTRACE_SYSEMU_SINGLESTEP:
1054#endif
1055 case PTRACE_SYSCALL:
1056 case PTRACE_CONT:
1057 return ptrace_resume(child, request, data);
1058
1059 case PTRACE_KILL:
1060 if (child->exit_state) /* already dead */
1061 return 0;
1062 return ptrace_resume(child, request, SIGKILL);
1063
1064#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1065 case PTRACE_GETREGSET:
1066 case PTRACE_SETREGSET: {
1067 struct iovec kiov;
1068 struct iovec __user *uiov = datavp;
1069
1070 if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1071 return -EFAULT;
1072
1073 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1074 __get_user(kiov.iov_len, &uiov->iov_len))
1075 return -EFAULT;
1076
1077 ret = ptrace_regset(child, request, addr, &kiov);
1078 if (!ret)
1079 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1080 break;
1081 }
1082#endif
1083
1084 case PTRACE_SECCOMP_GET_FILTER:
1085 ret = seccomp_get_filter(child, addr, datavp);
1086 break;
1087
1088 default:
1089 break;
1090 }
1091
1092 return ret;
1093}
1094
1095static struct task_struct *ptrace_get_task_struct(pid_t pid)
1096{
1097 struct task_struct *child;
1098
1099 rcu_read_lock();
1100 child = find_task_by_vpid(pid);
1101 if (child)
1102 get_task_struct(child);
1103 rcu_read_unlock();
1104
1105 if (!child)
1106 return ERR_PTR(-ESRCH);
1107 return child;
1108}
1109
1110#ifndef arch_ptrace_attach
1111#define arch_ptrace_attach(child) do { } while (0)
1112#endif
1113
1114SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1115 unsigned long, data)
1116{
1117 struct task_struct *child;
1118 long ret;
1119
1120 if (request == PTRACE_TRACEME) {
1121 ret = ptrace_traceme();
1122 if (!ret)
1123 arch_ptrace_attach(current);
1124 goto out;
1125 }
1126
1127 child = ptrace_get_task_struct(pid);
1128 if (IS_ERR(child)) {
1129 ret = PTR_ERR(child);
1130 goto out;
1131 }
1132
1133 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1134 ret = ptrace_attach(child, request, addr, data);
1135 /*
1136 * Some architectures need to do book-keeping after
1137 * a ptrace attach.
1138 */
1139 if (!ret)
1140 arch_ptrace_attach(child);
1141 goto out_put_task_struct;
1142 }
1143
1144 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1145 request == PTRACE_INTERRUPT);
1146 if (ret < 0)
1147 goto out_put_task_struct;
1148
1149 ret = arch_ptrace(child, request, addr, data);
1150 if (ret || request != PTRACE_DETACH)
1151 ptrace_unfreeze_traced(child);
1152
1153 out_put_task_struct:
1154 put_task_struct(child);
1155 out:
1156 return ret;
1157}
1158
1159int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1160 unsigned long data)
1161{
1162 unsigned long tmp;
1163 int copied;
1164
1165 copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1166 if (copied != sizeof(tmp))
1167 return -EIO;
1168 return put_user(tmp, (unsigned long __user *)data);
1169}
1170
1171int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1172 unsigned long data)
1173{
1174 int copied;
1175
1176 copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1177 FOLL_FORCE | FOLL_WRITE);
1178 return (copied == sizeof(data)) ? 0 : -EIO;
1179}
1180
1181#if defined CONFIG_COMPAT
1182
1183int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1184 compat_ulong_t addr, compat_ulong_t data)
1185{
1186 compat_ulong_t __user *datap = compat_ptr(data);
1187 compat_ulong_t word;
1188 siginfo_t siginfo;
1189 int ret;
1190
1191 switch (request) {
1192 case PTRACE_PEEKTEXT:
1193 case PTRACE_PEEKDATA:
1194 ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1195 FOLL_FORCE);
1196 if (ret != sizeof(word))
1197 ret = -EIO;
1198 else
1199 ret = put_user(word, datap);
1200 break;
1201
1202 case PTRACE_POKETEXT:
1203 case PTRACE_POKEDATA:
1204 ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1205 FOLL_FORCE | FOLL_WRITE);
1206 ret = (ret != sizeof(data) ? -EIO : 0);
1207 break;
1208
1209 case PTRACE_GETEVENTMSG:
1210 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1211 break;
1212
1213 case PTRACE_GETSIGINFO:
1214 ret = ptrace_getsiginfo(child, &siginfo);
1215 if (!ret)
1216 ret = copy_siginfo_to_user32(
1217 (struct compat_siginfo __user *) datap,
1218 &siginfo);
1219 break;
1220
1221 case PTRACE_SETSIGINFO:
1222 memset(&siginfo, 0, sizeof siginfo);
1223 if (copy_siginfo_from_user32(
1224 &siginfo, (struct compat_siginfo __user *) datap))
1225 ret = -EFAULT;
1226 else
1227 ret = ptrace_setsiginfo(child, &siginfo);
1228 break;
1229#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1230 case PTRACE_GETREGSET:
1231 case PTRACE_SETREGSET:
1232 {
1233 struct iovec kiov;
1234 struct compat_iovec __user *uiov =
1235 (struct compat_iovec __user *) datap;
1236 compat_uptr_t ptr;
1237 compat_size_t len;
1238
1239 if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1240 return -EFAULT;
1241
1242 if (__get_user(ptr, &uiov->iov_base) ||
1243 __get_user(len, &uiov->iov_len))
1244 return -EFAULT;
1245
1246 kiov.iov_base = compat_ptr(ptr);
1247 kiov.iov_len = len;
1248
1249 ret = ptrace_regset(child, request, addr, &kiov);
1250 if (!ret)
1251 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1252 break;
1253 }
1254#endif
1255
1256 default:
1257 ret = ptrace_request(child, request, addr, data);
1258 }
1259
1260 return ret;
1261}
1262
1263COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1264 compat_long_t, addr, compat_long_t, data)
1265{
1266 struct task_struct *child;
1267 long ret;
1268
1269 if (request == PTRACE_TRACEME) {
1270 ret = ptrace_traceme();
1271 goto out;
1272 }
1273
1274 child = ptrace_get_task_struct(pid);
1275 if (IS_ERR(child)) {
1276 ret = PTR_ERR(child);
1277 goto out;
1278 }
1279
1280 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1281 ret = ptrace_attach(child, request, addr, data);
1282 /*
1283 * Some architectures need to do book-keeping after
1284 * a ptrace attach.
1285 */
1286 if (!ret)
1287 arch_ptrace_attach(child);
1288 goto out_put_task_struct;
1289 }
1290
1291 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1292 request == PTRACE_INTERRUPT);
1293 if (!ret) {
1294 ret = compat_arch_ptrace(child, request, addr, data);
1295 if (ret || request != PTRACE_DETACH)
1296 ptrace_unfreeze_traced(child);
1297 }
1298
1299 out_put_task_struct:
1300 put_task_struct(child);
1301 out:
1302 return ret;
1303}
1304#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#include <linux/syscall_user_dispatch.h>
36
37#include <asm/syscall.h> /* for syscall_get_* */
38
39/*
40 * Access another process' address space via ptrace.
41 * Source/target buffer must be kernel space,
42 * Do not walk the page table directly, use get_user_pages
43 */
44int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
45 void *buf, int len, unsigned int gup_flags)
46{
47 struct mm_struct *mm;
48 int ret;
49
50 mm = get_task_mm(tsk);
51 if (!mm)
52 return 0;
53
54 if (!tsk->ptrace ||
55 (current != tsk->parent) ||
56 ((get_dumpable(mm) != SUID_DUMP_USER) &&
57 !ptracer_capable(tsk, mm->user_ns))) {
58 mmput(mm);
59 return 0;
60 }
61
62 ret = access_remote_vm(mm, addr, buf, len, gup_flags);
63 mmput(mm);
64
65 return ret;
66}
67
68
69void __ptrace_link(struct task_struct *child, struct task_struct *new_parent,
70 const struct cred *ptracer_cred)
71{
72 BUG_ON(!list_empty(&child->ptrace_entry));
73 list_add(&child->ptrace_entry, &new_parent->ptraced);
74 child->parent = new_parent;
75 child->ptracer_cred = get_cred(ptracer_cred);
76}
77
78/*
79 * ptrace a task: make the debugger its new parent and
80 * move it to the ptrace list.
81 *
82 * Must be called with the tasklist lock write-held.
83 */
84static void ptrace_link(struct task_struct *child, struct task_struct *new_parent)
85{
86 __ptrace_link(child, new_parent, current_cred());
87}
88
89/**
90 * __ptrace_unlink - unlink ptracee and restore its execution state
91 * @child: ptracee to be unlinked
92 *
93 * Remove @child from the ptrace list, move it back to the original parent,
94 * and restore the execution state so that it conforms to the group stop
95 * state.
96 *
97 * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
98 * exiting. For PTRACE_DETACH, unless the ptracee has been killed between
99 * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
100 * If the ptracer is exiting, the ptracee can be in any state.
101 *
102 * After detach, the ptracee should be in a state which conforms to the
103 * group stop. If the group is stopped or in the process of stopping, the
104 * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
105 * up from TASK_TRACED.
106 *
107 * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
108 * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
109 * to but in the opposite direction of what happens while attaching to a
110 * stopped task. However, in this direction, the intermediate RUNNING
111 * state is not hidden even from the current ptracer and if it immediately
112 * re-attaches and performs a WNOHANG wait(2), it may fail.
113 *
114 * CONTEXT:
115 * write_lock_irq(tasklist_lock)
116 */
117void __ptrace_unlink(struct task_struct *child)
118{
119 const struct cred *old_cred;
120 BUG_ON(!child->ptrace);
121
122 clear_task_syscall_work(child, SYSCALL_TRACE);
123#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
124 clear_task_syscall_work(child, SYSCALL_EMU);
125#endif
126
127 child->parent = child->real_parent;
128 list_del_init(&child->ptrace_entry);
129 old_cred = child->ptracer_cred;
130 child->ptracer_cred = NULL;
131 put_cred(old_cred);
132
133 spin_lock(&child->sighand->siglock);
134 child->ptrace = 0;
135 /*
136 * Clear all pending traps and TRAPPING. TRAPPING should be
137 * cleared regardless of JOBCTL_STOP_PENDING. Do it explicitly.
138 */
139 task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
140 task_clear_jobctl_trapping(child);
141
142 /*
143 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
144 * @child isn't dead.
145 */
146 if (!(child->flags & PF_EXITING) &&
147 (child->signal->flags & SIGNAL_STOP_STOPPED ||
148 child->signal->group_stop_count))
149 child->jobctl |= JOBCTL_STOP_PENDING;
150
151 /*
152 * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
153 * @child in the butt. Note that @resume should be used iff @child
154 * is in TASK_TRACED; otherwise, we might unduly disrupt
155 * TASK_KILLABLE sleeps.
156 */
157 if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
158 ptrace_signal_wake_up(child, true);
159
160 spin_unlock(&child->sighand->siglock);
161}
162
163static bool looks_like_a_spurious_pid(struct task_struct *task)
164{
165 if (task->exit_code != ((PTRACE_EVENT_EXEC << 8) | SIGTRAP))
166 return false;
167
168 if (task_pid_vnr(task) == task->ptrace_message)
169 return false;
170 /*
171 * The tracee changed its pid but the PTRACE_EVENT_EXEC event
172 * was not wait()'ed, most probably debugger targets the old
173 * leader which was destroyed in de_thread().
174 */
175 return true;
176}
177
178/*
179 * Ensure that nothing can wake it up, even SIGKILL
180 *
181 * A task is switched to this state while a ptrace operation is in progress;
182 * such that the ptrace operation is uninterruptible.
183 */
184static bool ptrace_freeze_traced(struct task_struct *task)
185{
186 bool ret = false;
187
188 /* Lockless, nobody but us can set this flag */
189 if (task->jobctl & JOBCTL_LISTENING)
190 return ret;
191
192 spin_lock_irq(&task->sighand->siglock);
193 if (task_is_traced(task) && !looks_like_a_spurious_pid(task) &&
194 !__fatal_signal_pending(task)) {
195 task->jobctl |= JOBCTL_PTRACE_FROZEN;
196 ret = true;
197 }
198 spin_unlock_irq(&task->sighand->siglock);
199
200 return ret;
201}
202
203static void ptrace_unfreeze_traced(struct task_struct *task)
204{
205 unsigned long flags;
206
207 /*
208 * The child may be awake and may have cleared
209 * JOBCTL_PTRACE_FROZEN (see ptrace_resume). The child will
210 * not set JOBCTL_PTRACE_FROZEN or enter __TASK_TRACED anew.
211 */
212 if (lock_task_sighand(task, &flags)) {
213 task->jobctl &= ~JOBCTL_PTRACE_FROZEN;
214 if (__fatal_signal_pending(task)) {
215 task->jobctl &= ~JOBCTL_TRACED;
216 wake_up_state(task, __TASK_TRACED);
217 }
218 unlock_task_sighand(task, &flags);
219 }
220}
221
222/**
223 * ptrace_check_attach - check whether ptracee is ready for ptrace operation
224 * @child: ptracee to check for
225 * @ignore_state: don't check whether @child is currently %TASK_TRACED
226 *
227 * Check whether @child is being ptraced by %current and ready for further
228 * ptrace operations. If @ignore_state is %false, @child also should be in
229 * %TASK_TRACED state and on return the child is guaranteed to be traced
230 * and not executing. If @ignore_state is %true, @child can be in any
231 * state.
232 *
233 * CONTEXT:
234 * Grabs and releases tasklist_lock and @child->sighand->siglock.
235 *
236 * RETURNS:
237 * 0 on success, -ESRCH if %child is not ready.
238 */
239static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
240{
241 int ret = -ESRCH;
242
243 /*
244 * We take the read lock around doing both checks to close a
245 * possible race where someone else was tracing our child and
246 * detached between these two checks. After this locked check,
247 * we are sure that this is our traced child and that can only
248 * be changed by us so it's not changing right after this.
249 */
250 read_lock(&tasklist_lock);
251 if (child->ptrace && child->parent == current) {
252 /*
253 * child->sighand can't be NULL, release_task()
254 * does ptrace_unlink() before __exit_signal().
255 */
256 if (ignore_state || ptrace_freeze_traced(child))
257 ret = 0;
258 }
259 read_unlock(&tasklist_lock);
260
261 if (!ret && !ignore_state &&
262 WARN_ON_ONCE(!wait_task_inactive(child, __TASK_TRACED|TASK_FROZEN)))
263 ret = -ESRCH;
264
265 return ret;
266}
267
268static bool ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
269{
270 if (mode & PTRACE_MODE_NOAUDIT)
271 return ns_capable_noaudit(ns, CAP_SYS_PTRACE);
272 return ns_capable(ns, CAP_SYS_PTRACE);
273}
274
275/* Returns 0 on success, -errno on denial. */
276static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
277{
278 const struct cred *cred = current_cred(), *tcred;
279 struct mm_struct *mm;
280 kuid_t caller_uid;
281 kgid_t caller_gid;
282
283 if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
284 WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
285 return -EPERM;
286 }
287
288 /* May we inspect the given task?
289 * This check is used both for attaching with ptrace
290 * and for allowing access to sensitive information in /proc.
291 *
292 * ptrace_attach denies several cases that /proc allows
293 * because setting up the necessary parent/child relationship
294 * or halting the specified task is impossible.
295 */
296
297 /* Don't let security modules deny introspection */
298 if (same_thread_group(task, current))
299 return 0;
300 rcu_read_lock();
301 if (mode & PTRACE_MODE_FSCREDS) {
302 caller_uid = cred->fsuid;
303 caller_gid = cred->fsgid;
304 } else {
305 /*
306 * Using the euid would make more sense here, but something
307 * in userland might rely on the old behavior, and this
308 * shouldn't be a security problem since
309 * PTRACE_MODE_REALCREDS implies that the caller explicitly
310 * used a syscall that requests access to another process
311 * (and not a filesystem syscall to procfs).
312 */
313 caller_uid = cred->uid;
314 caller_gid = cred->gid;
315 }
316 tcred = __task_cred(task);
317 if (uid_eq(caller_uid, tcred->euid) &&
318 uid_eq(caller_uid, tcred->suid) &&
319 uid_eq(caller_uid, tcred->uid) &&
320 gid_eq(caller_gid, tcred->egid) &&
321 gid_eq(caller_gid, tcred->sgid) &&
322 gid_eq(caller_gid, tcred->gid))
323 goto ok;
324 if (ptrace_has_cap(tcred->user_ns, mode))
325 goto ok;
326 rcu_read_unlock();
327 return -EPERM;
328ok:
329 rcu_read_unlock();
330 /*
331 * If a task drops privileges and becomes nondumpable (through a syscall
332 * like setresuid()) while we are trying to access it, we must ensure
333 * that the dumpability is read after the credentials; otherwise,
334 * we may be able to attach to a task that we shouldn't be able to
335 * attach to (as if the task had dropped privileges without becoming
336 * nondumpable).
337 * Pairs with a write barrier in commit_creds().
338 */
339 smp_rmb();
340 mm = task->mm;
341 if (mm &&
342 ((get_dumpable(mm) != SUID_DUMP_USER) &&
343 !ptrace_has_cap(mm->user_ns, mode)))
344 return -EPERM;
345
346 return security_ptrace_access_check(task, mode);
347}
348
349bool ptrace_may_access(struct task_struct *task, unsigned int mode)
350{
351 int err;
352 task_lock(task);
353 err = __ptrace_may_access(task, mode);
354 task_unlock(task);
355 return !err;
356}
357
358static int check_ptrace_options(unsigned long data)
359{
360 if (data & ~(unsigned long)PTRACE_O_MASK)
361 return -EINVAL;
362
363 if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
364 if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
365 !IS_ENABLED(CONFIG_SECCOMP))
366 return -EINVAL;
367
368 if (!capable(CAP_SYS_ADMIN))
369 return -EPERM;
370
371 if (seccomp_mode(¤t->seccomp) != SECCOMP_MODE_DISABLED ||
372 current->ptrace & PT_SUSPEND_SECCOMP)
373 return -EPERM;
374 }
375 return 0;
376}
377
378static inline void ptrace_set_stopped(struct task_struct *task, bool seize)
379{
380 guard(spinlock)(&task->sighand->siglock);
381
382 /* SEIZE doesn't trap tracee on attach */
383 if (!seize)
384 send_signal_locked(SIGSTOP, SEND_SIG_PRIV, task, PIDTYPE_PID);
385 /*
386 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
387 * TRAPPING, and kick it so that it transits to TRACED. TRAPPING
388 * will be cleared if the child completes the transition or any
389 * event which clears the group stop states happens. We'll wait
390 * for the transition to complete before returning from this
391 * function.
392 *
393 * This hides STOPPED -> RUNNING -> TRACED transition from the
394 * attaching thread but a different thread in the same group can
395 * still observe the transient RUNNING state. IOW, if another
396 * thread's WNOHANG wait(2) on the stopped tracee races against
397 * ATTACH, the wait(2) may fail due to the transient RUNNING.
398 *
399 * The following task_is_stopped() test is safe as both transitions
400 * in and out of STOPPED are protected by siglock.
401 */
402 if (task_is_stopped(task) &&
403 task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING)) {
404 task->jobctl &= ~JOBCTL_STOPPED;
405 signal_wake_up_state(task, __TASK_STOPPED);
406 }
407}
408
409static int ptrace_attach(struct task_struct *task, long request,
410 unsigned long addr,
411 unsigned long flags)
412{
413 bool seize = (request == PTRACE_SEIZE);
414 int retval;
415
416 if (seize) {
417 if (addr != 0)
418 return -EIO;
419 /*
420 * This duplicates the check in check_ptrace_options() because
421 * ptrace_attach() and ptrace_setoptions() have historically
422 * used different error codes for unknown ptrace options.
423 */
424 if (flags & ~(unsigned long)PTRACE_O_MASK)
425 return -EIO;
426
427 retval = check_ptrace_options(flags);
428 if (retval)
429 return retval;
430 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
431 } else {
432 flags = PT_PTRACED;
433 }
434
435 audit_ptrace(task);
436
437 if (unlikely(task->flags & PF_KTHREAD))
438 return -EPERM;
439 if (same_thread_group(task, current))
440 return -EPERM;
441
442 /*
443 * Protect exec's credential calculations against our interference;
444 * SUID, SGID and LSM creds get determined differently
445 * under ptrace.
446 */
447 scoped_cond_guard (mutex_intr, return -ERESTARTNOINTR,
448 &task->signal->cred_guard_mutex) {
449
450 scoped_guard (task_lock, task) {
451 retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
452 if (retval)
453 return retval;
454 }
455
456 scoped_guard (write_lock_irq, &tasklist_lock) {
457 if (unlikely(task->exit_state))
458 return -EPERM;
459 if (task->ptrace)
460 return -EPERM;
461
462 task->ptrace = flags;
463 ptrace_link(task, current);
464 ptrace_set_stopped(task, seize);
465 }
466 }
467
468 /*
469 * We do not bother to change retval or clear JOBCTL_TRAPPING
470 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
471 * not return to user-mode, it will exit and clear this bit in
472 * __ptrace_unlink() if it wasn't already cleared by the tracee;
473 * and until then nobody can ptrace this task.
474 */
475 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
476 proc_ptrace_connector(task, PTRACE_ATTACH);
477
478 return 0;
479}
480
481/**
482 * ptrace_traceme -- helper for PTRACE_TRACEME
483 *
484 * Performs checks and sets PT_PTRACED.
485 * Should be used by all ptrace implementations for PTRACE_TRACEME.
486 */
487static int ptrace_traceme(void)
488{
489 int ret = -EPERM;
490
491 write_lock_irq(&tasklist_lock);
492 /* Are we already being traced? */
493 if (!current->ptrace) {
494 ret = security_ptrace_traceme(current->parent);
495 /*
496 * Check PF_EXITING to ensure ->real_parent has not passed
497 * exit_ptrace(). Otherwise we don't report the error but
498 * pretend ->real_parent untraces us right after return.
499 */
500 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
501 current->ptrace = PT_PTRACED;
502 ptrace_link(current, current->real_parent);
503 }
504 }
505 write_unlock_irq(&tasklist_lock);
506
507 return ret;
508}
509
510/*
511 * Called with irqs disabled, returns true if childs should reap themselves.
512 */
513static int ignoring_children(struct sighand_struct *sigh)
514{
515 int ret;
516 spin_lock(&sigh->siglock);
517 ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
518 (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
519 spin_unlock(&sigh->siglock);
520 return ret;
521}
522
523/*
524 * Called with tasklist_lock held for writing.
525 * Unlink a traced task, and clean it up if it was a traced zombie.
526 * Return true if it needs to be reaped with release_task().
527 * (We can't call release_task() here because we already hold tasklist_lock.)
528 *
529 * If it's a zombie, our attachedness prevented normal parent notification
530 * or self-reaping. Do notification now if it would have happened earlier.
531 * If it should reap itself, return true.
532 *
533 * If it's our own child, there is no notification to do. But if our normal
534 * children self-reap, then this child was prevented by ptrace and we must
535 * reap it now, in that case we must also wake up sub-threads sleeping in
536 * do_wait().
537 */
538static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
539{
540 bool dead;
541
542 __ptrace_unlink(p);
543
544 if (p->exit_state != EXIT_ZOMBIE)
545 return false;
546
547 dead = !thread_group_leader(p);
548
549 if (!dead && thread_group_empty(p)) {
550 if (!same_thread_group(p->real_parent, tracer))
551 dead = do_notify_parent(p, p->exit_signal);
552 else if (ignoring_children(tracer->sighand)) {
553 __wake_up_parent(p, tracer);
554 dead = true;
555 }
556 }
557 /* Mark it as in the process of being reaped. */
558 if (dead)
559 p->exit_state = EXIT_DEAD;
560 return dead;
561}
562
563static int ptrace_detach(struct task_struct *child, unsigned int data)
564{
565 if (!valid_signal(data))
566 return -EIO;
567
568 /* Architecture-specific hardware disable .. */
569 ptrace_disable(child);
570
571 write_lock_irq(&tasklist_lock);
572 /*
573 * We rely on ptrace_freeze_traced(). It can't be killed and
574 * untraced by another thread, it can't be a zombie.
575 */
576 WARN_ON(!child->ptrace || child->exit_state);
577 /*
578 * tasklist_lock avoids the race with wait_task_stopped(), see
579 * the comment in ptrace_resume().
580 */
581 child->exit_code = data;
582 __ptrace_detach(current, child);
583 write_unlock_irq(&tasklist_lock);
584
585 proc_ptrace_connector(child, PTRACE_DETACH);
586
587 return 0;
588}
589
590/*
591 * Detach all tasks we were using ptrace on. Called with tasklist held
592 * for writing.
593 */
594void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
595{
596 struct task_struct *p, *n;
597
598 list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
599 if (unlikely(p->ptrace & PT_EXITKILL))
600 send_sig_info(SIGKILL, SEND_SIG_PRIV, p);
601
602 if (__ptrace_detach(tracer, p))
603 list_add(&p->ptrace_entry, dead);
604 }
605}
606
607int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
608{
609 int copied = 0;
610
611 while (len > 0) {
612 char buf[128];
613 int this_len, retval;
614
615 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
616 retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
617
618 if (!retval) {
619 if (copied)
620 break;
621 return -EIO;
622 }
623 if (copy_to_user(dst, buf, retval))
624 return -EFAULT;
625 copied += retval;
626 src += retval;
627 dst += retval;
628 len -= retval;
629 }
630 return copied;
631}
632
633int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
634{
635 int copied = 0;
636
637 while (len > 0) {
638 char buf[128];
639 int this_len, retval;
640
641 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
642 if (copy_from_user(buf, src, this_len))
643 return -EFAULT;
644 retval = ptrace_access_vm(tsk, dst, buf, this_len,
645 FOLL_FORCE | FOLL_WRITE);
646 if (!retval) {
647 if (copied)
648 break;
649 return -EIO;
650 }
651 copied += retval;
652 src += retval;
653 dst += retval;
654 len -= retval;
655 }
656 return copied;
657}
658
659static int ptrace_setoptions(struct task_struct *child, unsigned long data)
660{
661 unsigned flags;
662 int ret;
663
664 ret = check_ptrace_options(data);
665 if (ret)
666 return ret;
667
668 /* Avoid intermediate state when all opts are cleared */
669 flags = child->ptrace;
670 flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
671 flags |= (data << PT_OPT_FLAG_SHIFT);
672 child->ptrace = flags;
673
674 return 0;
675}
676
677static int ptrace_getsiginfo(struct task_struct *child, kernel_siginfo_t *info)
678{
679 unsigned long flags;
680 int error = -ESRCH;
681
682 if (lock_task_sighand(child, &flags)) {
683 error = -EINVAL;
684 if (likely(child->last_siginfo != NULL)) {
685 copy_siginfo(info, child->last_siginfo);
686 error = 0;
687 }
688 unlock_task_sighand(child, &flags);
689 }
690 return error;
691}
692
693static int ptrace_setsiginfo(struct task_struct *child, const kernel_siginfo_t *info)
694{
695 unsigned long flags;
696 int error = -ESRCH;
697
698 if (lock_task_sighand(child, &flags)) {
699 error = -EINVAL;
700 if (likely(child->last_siginfo != NULL)) {
701 copy_siginfo(child->last_siginfo, info);
702 error = 0;
703 }
704 unlock_task_sighand(child, &flags);
705 }
706 return error;
707}
708
709static int ptrace_peek_siginfo(struct task_struct *child,
710 unsigned long addr,
711 unsigned long data)
712{
713 struct ptrace_peeksiginfo_args arg;
714 struct sigpending *pending;
715 struct sigqueue *q;
716 int ret, i;
717
718 ret = copy_from_user(&arg, (void __user *) addr,
719 sizeof(struct ptrace_peeksiginfo_args));
720 if (ret)
721 return -EFAULT;
722
723 if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
724 return -EINVAL; /* unknown flags */
725
726 if (arg.nr < 0)
727 return -EINVAL;
728
729 /* Ensure arg.off fits in an unsigned long */
730 if (arg.off > ULONG_MAX)
731 return 0;
732
733 if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
734 pending = &child->signal->shared_pending;
735 else
736 pending = &child->pending;
737
738 for (i = 0; i < arg.nr; ) {
739 kernel_siginfo_t info;
740 unsigned long off = arg.off + i;
741 bool found = false;
742
743 spin_lock_irq(&child->sighand->siglock);
744 list_for_each_entry(q, &pending->list, list) {
745 if (!off--) {
746 found = true;
747 copy_siginfo(&info, &q->info);
748 break;
749 }
750 }
751 spin_unlock_irq(&child->sighand->siglock);
752
753 if (!found) /* beyond the end of the list */
754 break;
755
756#ifdef CONFIG_COMPAT
757 if (unlikely(in_compat_syscall())) {
758 compat_siginfo_t __user *uinfo = compat_ptr(data);
759
760 if (copy_siginfo_to_user32(uinfo, &info)) {
761 ret = -EFAULT;
762 break;
763 }
764
765 } else
766#endif
767 {
768 siginfo_t __user *uinfo = (siginfo_t __user *) data;
769
770 if (copy_siginfo_to_user(uinfo, &info)) {
771 ret = -EFAULT;
772 break;
773 }
774 }
775
776 data += sizeof(siginfo_t);
777 i++;
778
779 if (signal_pending(current))
780 break;
781
782 cond_resched();
783 }
784
785 if (i > 0)
786 return i;
787
788 return ret;
789}
790
791#ifdef CONFIG_RSEQ
792static long ptrace_get_rseq_configuration(struct task_struct *task,
793 unsigned long size, void __user *data)
794{
795 struct ptrace_rseq_configuration conf = {
796 .rseq_abi_pointer = (u64)(uintptr_t)task->rseq,
797 .rseq_abi_size = task->rseq_len,
798 .signature = task->rseq_sig,
799 .flags = 0,
800 };
801
802 size = min_t(unsigned long, size, sizeof(conf));
803 if (copy_to_user(data, &conf, size))
804 return -EFAULT;
805 return sizeof(conf);
806}
807#endif
808
809#define is_singlestep(request) ((request) == PTRACE_SINGLESTEP)
810
811#ifdef PTRACE_SINGLEBLOCK
812#define is_singleblock(request) ((request) == PTRACE_SINGLEBLOCK)
813#else
814#define is_singleblock(request) 0
815#endif
816
817#ifdef PTRACE_SYSEMU
818#define is_sysemu_singlestep(request) ((request) == PTRACE_SYSEMU_SINGLESTEP)
819#else
820#define is_sysemu_singlestep(request) 0
821#endif
822
823static int ptrace_resume(struct task_struct *child, long request,
824 unsigned long data)
825{
826 if (!valid_signal(data))
827 return -EIO;
828
829 if (request == PTRACE_SYSCALL)
830 set_task_syscall_work(child, SYSCALL_TRACE);
831 else
832 clear_task_syscall_work(child, SYSCALL_TRACE);
833
834#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
835 if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
836 set_task_syscall_work(child, SYSCALL_EMU);
837 else
838 clear_task_syscall_work(child, SYSCALL_EMU);
839#endif
840
841 if (is_singleblock(request)) {
842 if (unlikely(!arch_has_block_step()))
843 return -EIO;
844 user_enable_block_step(child);
845 } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
846 if (unlikely(!arch_has_single_step()))
847 return -EIO;
848 user_enable_single_step(child);
849 } else {
850 user_disable_single_step(child);
851 }
852
853 /*
854 * Change ->exit_code and ->state under siglock to avoid the race
855 * with wait_task_stopped() in between; a non-zero ->exit_code will
856 * wrongly look like another report from tracee.
857 *
858 * Note that we need siglock even if ->exit_code == data and/or this
859 * status was not reported yet, the new status must not be cleared by
860 * wait_task_stopped() after resume.
861 */
862 spin_lock_irq(&child->sighand->siglock);
863 child->exit_code = data;
864 child->jobctl &= ~JOBCTL_TRACED;
865 wake_up_state(child, __TASK_TRACED);
866 spin_unlock_irq(&child->sighand->siglock);
867
868 return 0;
869}
870
871#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
872
873static const struct user_regset *
874find_regset(const struct user_regset_view *view, unsigned int type)
875{
876 const struct user_regset *regset;
877 int n;
878
879 for (n = 0; n < view->n; ++n) {
880 regset = view->regsets + n;
881 if (regset->core_note_type == type)
882 return regset;
883 }
884
885 return NULL;
886}
887
888static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
889 struct iovec *kiov)
890{
891 const struct user_regset_view *view = task_user_regset_view(task);
892 const struct user_regset *regset = find_regset(view, type);
893 int regset_no;
894
895 if (!regset || (kiov->iov_len % regset->size) != 0)
896 return -EINVAL;
897
898 regset_no = regset - view->regsets;
899 kiov->iov_len = min(kiov->iov_len,
900 (__kernel_size_t) (regset->n * regset->size));
901
902 if (req == PTRACE_GETREGSET)
903 return copy_regset_to_user(task, view, regset_no, 0,
904 kiov->iov_len, kiov->iov_base);
905 else
906 return copy_regset_from_user(task, view, regset_no, 0,
907 kiov->iov_len, kiov->iov_base);
908}
909
910/*
911 * This is declared in linux/regset.h and defined in machine-dependent
912 * code. We put the export here, near the primary machine-neutral use,
913 * to ensure no machine forgets it.
914 */
915EXPORT_SYMBOL_GPL(task_user_regset_view);
916
917static unsigned long
918ptrace_get_syscall_info_entry(struct task_struct *child, struct pt_regs *regs,
919 struct ptrace_syscall_info *info)
920{
921 unsigned long args[ARRAY_SIZE(info->entry.args)];
922 int i;
923
924 info->op = PTRACE_SYSCALL_INFO_ENTRY;
925 info->entry.nr = syscall_get_nr(child, regs);
926 syscall_get_arguments(child, regs, args);
927 for (i = 0; i < ARRAY_SIZE(args); i++)
928 info->entry.args[i] = args[i];
929
930 /* args is the last field in struct ptrace_syscall_info.entry */
931 return offsetofend(struct ptrace_syscall_info, entry.args);
932}
933
934static unsigned long
935ptrace_get_syscall_info_seccomp(struct task_struct *child, struct pt_regs *regs,
936 struct ptrace_syscall_info *info)
937{
938 /*
939 * As struct ptrace_syscall_info.entry is currently a subset
940 * of struct ptrace_syscall_info.seccomp, it makes sense to
941 * initialize that subset using ptrace_get_syscall_info_entry().
942 * This can be reconsidered in the future if these structures
943 * diverge significantly enough.
944 */
945 ptrace_get_syscall_info_entry(child, regs, info);
946 info->op = PTRACE_SYSCALL_INFO_SECCOMP;
947 info->seccomp.ret_data = child->ptrace_message;
948
949 /* ret_data is the last field in struct ptrace_syscall_info.seccomp */
950 return offsetofend(struct ptrace_syscall_info, seccomp.ret_data);
951}
952
953static unsigned long
954ptrace_get_syscall_info_exit(struct task_struct *child, struct pt_regs *regs,
955 struct ptrace_syscall_info *info)
956{
957 info->op = PTRACE_SYSCALL_INFO_EXIT;
958 info->exit.rval = syscall_get_error(child, regs);
959 info->exit.is_error = !!info->exit.rval;
960 if (!info->exit.is_error)
961 info->exit.rval = syscall_get_return_value(child, regs);
962
963 /* is_error is the last field in struct ptrace_syscall_info.exit */
964 return offsetofend(struct ptrace_syscall_info, exit.is_error);
965}
966
967static int
968ptrace_get_syscall_info(struct task_struct *child, unsigned long user_size,
969 void __user *datavp)
970{
971 struct pt_regs *regs = task_pt_regs(child);
972 struct ptrace_syscall_info info = {
973 .op = PTRACE_SYSCALL_INFO_NONE,
974 .arch = syscall_get_arch(child),
975 .instruction_pointer = instruction_pointer(regs),
976 .stack_pointer = user_stack_pointer(regs),
977 };
978 unsigned long actual_size = offsetof(struct ptrace_syscall_info, entry);
979 unsigned long write_size;
980
981 /*
982 * This does not need lock_task_sighand() to access
983 * child->last_siginfo because ptrace_freeze_traced()
984 * called earlier by ptrace_check_attach() ensures that
985 * the tracee cannot go away and clear its last_siginfo.
986 */
987 switch (child->last_siginfo ? child->last_siginfo->si_code : 0) {
988 case SIGTRAP | 0x80:
989 switch (child->ptrace_message) {
990 case PTRACE_EVENTMSG_SYSCALL_ENTRY:
991 actual_size = ptrace_get_syscall_info_entry(child, regs,
992 &info);
993 break;
994 case PTRACE_EVENTMSG_SYSCALL_EXIT:
995 actual_size = ptrace_get_syscall_info_exit(child, regs,
996 &info);
997 break;
998 }
999 break;
1000 case SIGTRAP | (PTRACE_EVENT_SECCOMP << 8):
1001 actual_size = ptrace_get_syscall_info_seccomp(child, regs,
1002 &info);
1003 break;
1004 }
1005
1006 write_size = min(actual_size, user_size);
1007 return copy_to_user(datavp, &info, write_size) ? -EFAULT : actual_size;
1008}
1009#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
1010
1011int ptrace_request(struct task_struct *child, long request,
1012 unsigned long addr, unsigned long data)
1013{
1014 bool seized = child->ptrace & PT_SEIZED;
1015 int ret = -EIO;
1016 kernel_siginfo_t siginfo, *si;
1017 void __user *datavp = (void __user *) data;
1018 unsigned long __user *datalp = datavp;
1019 unsigned long flags;
1020
1021 switch (request) {
1022 case PTRACE_PEEKTEXT:
1023 case PTRACE_PEEKDATA:
1024 return generic_ptrace_peekdata(child, addr, data);
1025 case PTRACE_POKETEXT:
1026 case PTRACE_POKEDATA:
1027 return generic_ptrace_pokedata(child, addr, data);
1028
1029#ifdef PTRACE_OLDSETOPTIONS
1030 case PTRACE_OLDSETOPTIONS:
1031#endif
1032 case PTRACE_SETOPTIONS:
1033 ret = ptrace_setoptions(child, data);
1034 break;
1035 case PTRACE_GETEVENTMSG:
1036 ret = put_user(child->ptrace_message, datalp);
1037 break;
1038
1039 case PTRACE_PEEKSIGINFO:
1040 ret = ptrace_peek_siginfo(child, addr, data);
1041 break;
1042
1043 case PTRACE_GETSIGINFO:
1044 ret = ptrace_getsiginfo(child, &siginfo);
1045 if (!ret)
1046 ret = copy_siginfo_to_user(datavp, &siginfo);
1047 break;
1048
1049 case PTRACE_SETSIGINFO:
1050 ret = copy_siginfo_from_user(&siginfo, datavp);
1051 if (!ret)
1052 ret = ptrace_setsiginfo(child, &siginfo);
1053 break;
1054
1055 case PTRACE_GETSIGMASK: {
1056 sigset_t *mask;
1057
1058 if (addr != sizeof(sigset_t)) {
1059 ret = -EINVAL;
1060 break;
1061 }
1062
1063 if (test_tsk_restore_sigmask(child))
1064 mask = &child->saved_sigmask;
1065 else
1066 mask = &child->blocked;
1067
1068 if (copy_to_user(datavp, mask, sizeof(sigset_t)))
1069 ret = -EFAULT;
1070 else
1071 ret = 0;
1072
1073 break;
1074 }
1075
1076 case PTRACE_SETSIGMASK: {
1077 sigset_t new_set;
1078
1079 if (addr != sizeof(sigset_t)) {
1080 ret = -EINVAL;
1081 break;
1082 }
1083
1084 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
1085 ret = -EFAULT;
1086 break;
1087 }
1088
1089 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
1090
1091 /*
1092 * Every thread does recalc_sigpending() after resume, so
1093 * retarget_shared_pending() and recalc_sigpending() are not
1094 * called here.
1095 */
1096 spin_lock_irq(&child->sighand->siglock);
1097 child->blocked = new_set;
1098 spin_unlock_irq(&child->sighand->siglock);
1099
1100 clear_tsk_restore_sigmask(child);
1101
1102 ret = 0;
1103 break;
1104 }
1105
1106 case PTRACE_INTERRUPT:
1107 /*
1108 * Stop tracee without any side-effect on signal or job
1109 * control. At least one trap is guaranteed to happen
1110 * after this request. If @child is already trapped, the
1111 * current trap is not disturbed and another trap will
1112 * happen after the current trap is ended with PTRACE_CONT.
1113 *
1114 * The actual trap might not be PTRACE_EVENT_STOP trap but
1115 * the pending condition is cleared regardless.
1116 */
1117 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1118 break;
1119
1120 /*
1121 * INTERRUPT doesn't disturb existing trap sans one
1122 * exception. If ptracer issued LISTEN for the current
1123 * STOP, this INTERRUPT should clear LISTEN and re-trap
1124 * tracee into STOP.
1125 */
1126 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
1127 ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1128
1129 unlock_task_sighand(child, &flags);
1130 ret = 0;
1131 break;
1132
1133 case PTRACE_LISTEN:
1134 /*
1135 * Listen for events. Tracee must be in STOP. It's not
1136 * resumed per-se but is not considered to be in TRACED by
1137 * wait(2) or ptrace(2). If an async event (e.g. group
1138 * stop state change) happens, tracee will enter STOP trap
1139 * again. Alternatively, ptracer can issue INTERRUPT to
1140 * finish listening and re-trap tracee into STOP.
1141 */
1142 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1143 break;
1144
1145 si = child->last_siginfo;
1146 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1147 child->jobctl |= JOBCTL_LISTENING;
1148 /*
1149 * If NOTIFY is set, it means event happened between
1150 * start of this trap and now. Trigger re-trap.
1151 */
1152 if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1153 ptrace_signal_wake_up(child, true);
1154 ret = 0;
1155 }
1156 unlock_task_sighand(child, &flags);
1157 break;
1158
1159 case PTRACE_DETACH: /* detach a process that was attached. */
1160 ret = ptrace_detach(child, data);
1161 break;
1162
1163#ifdef CONFIG_BINFMT_ELF_FDPIC
1164 case PTRACE_GETFDPIC: {
1165 struct mm_struct *mm = get_task_mm(child);
1166 unsigned long tmp = 0;
1167
1168 ret = -ESRCH;
1169 if (!mm)
1170 break;
1171
1172 switch (addr) {
1173 case PTRACE_GETFDPIC_EXEC:
1174 tmp = mm->context.exec_fdpic_loadmap;
1175 break;
1176 case PTRACE_GETFDPIC_INTERP:
1177 tmp = mm->context.interp_fdpic_loadmap;
1178 break;
1179 default:
1180 break;
1181 }
1182 mmput(mm);
1183
1184 ret = put_user(tmp, datalp);
1185 break;
1186 }
1187#endif
1188
1189 case PTRACE_SINGLESTEP:
1190#ifdef PTRACE_SINGLEBLOCK
1191 case PTRACE_SINGLEBLOCK:
1192#endif
1193#ifdef PTRACE_SYSEMU
1194 case PTRACE_SYSEMU:
1195 case PTRACE_SYSEMU_SINGLESTEP:
1196#endif
1197 case PTRACE_SYSCALL:
1198 case PTRACE_CONT:
1199 return ptrace_resume(child, request, data);
1200
1201 case PTRACE_KILL:
1202 send_sig_info(SIGKILL, SEND_SIG_NOINFO, child);
1203 return 0;
1204
1205#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1206 case PTRACE_GETREGSET:
1207 case PTRACE_SETREGSET: {
1208 struct iovec kiov;
1209 struct iovec __user *uiov = datavp;
1210
1211 if (!access_ok(uiov, sizeof(*uiov)))
1212 return -EFAULT;
1213
1214 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1215 __get_user(kiov.iov_len, &uiov->iov_len))
1216 return -EFAULT;
1217
1218 ret = ptrace_regset(child, request, addr, &kiov);
1219 if (!ret)
1220 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1221 break;
1222 }
1223
1224 case PTRACE_GET_SYSCALL_INFO:
1225 ret = ptrace_get_syscall_info(child, addr, datavp);
1226 break;
1227#endif
1228
1229 case PTRACE_SECCOMP_GET_FILTER:
1230 ret = seccomp_get_filter(child, addr, datavp);
1231 break;
1232
1233 case PTRACE_SECCOMP_GET_METADATA:
1234 ret = seccomp_get_metadata(child, addr, datavp);
1235 break;
1236
1237#ifdef CONFIG_RSEQ
1238 case PTRACE_GET_RSEQ_CONFIGURATION:
1239 ret = ptrace_get_rseq_configuration(child, addr, datavp);
1240 break;
1241#endif
1242
1243 case PTRACE_SET_SYSCALL_USER_DISPATCH_CONFIG:
1244 ret = syscall_user_dispatch_set_config(child, addr, datavp);
1245 break;
1246
1247 case PTRACE_GET_SYSCALL_USER_DISPATCH_CONFIG:
1248 ret = syscall_user_dispatch_get_config(child, addr, datavp);
1249 break;
1250
1251 default:
1252 break;
1253 }
1254
1255 return ret;
1256}
1257
1258SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1259 unsigned long, data)
1260{
1261 struct task_struct *child;
1262 long ret;
1263
1264 if (request == PTRACE_TRACEME) {
1265 ret = ptrace_traceme();
1266 goto out;
1267 }
1268
1269 child = find_get_task_by_vpid(pid);
1270 if (!child) {
1271 ret = -ESRCH;
1272 goto out;
1273 }
1274
1275 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1276 ret = ptrace_attach(child, request, addr, data);
1277 goto out_put_task_struct;
1278 }
1279
1280 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1281 request == PTRACE_INTERRUPT);
1282 if (ret < 0)
1283 goto out_put_task_struct;
1284
1285 ret = arch_ptrace(child, request, addr, data);
1286 if (ret || request != PTRACE_DETACH)
1287 ptrace_unfreeze_traced(child);
1288
1289 out_put_task_struct:
1290 put_task_struct(child);
1291 out:
1292 return ret;
1293}
1294
1295int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1296 unsigned long data)
1297{
1298 unsigned long tmp;
1299 int copied;
1300
1301 copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1302 if (copied != sizeof(tmp))
1303 return -EIO;
1304 return put_user(tmp, (unsigned long __user *)data);
1305}
1306
1307int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1308 unsigned long data)
1309{
1310 int copied;
1311
1312 copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1313 FOLL_FORCE | FOLL_WRITE);
1314 return (copied == sizeof(data)) ? 0 : -EIO;
1315}
1316
1317#if defined CONFIG_COMPAT
1318
1319int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1320 compat_ulong_t addr, compat_ulong_t data)
1321{
1322 compat_ulong_t __user *datap = compat_ptr(data);
1323 compat_ulong_t word;
1324 kernel_siginfo_t siginfo;
1325 int ret;
1326
1327 switch (request) {
1328 case PTRACE_PEEKTEXT:
1329 case PTRACE_PEEKDATA:
1330 ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1331 FOLL_FORCE);
1332 if (ret != sizeof(word))
1333 ret = -EIO;
1334 else
1335 ret = put_user(word, datap);
1336 break;
1337
1338 case PTRACE_POKETEXT:
1339 case PTRACE_POKEDATA:
1340 ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1341 FOLL_FORCE | FOLL_WRITE);
1342 ret = (ret != sizeof(data) ? -EIO : 0);
1343 break;
1344
1345 case PTRACE_GETEVENTMSG:
1346 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1347 break;
1348
1349 case PTRACE_GETSIGINFO:
1350 ret = ptrace_getsiginfo(child, &siginfo);
1351 if (!ret)
1352 ret = copy_siginfo_to_user32(
1353 (struct compat_siginfo __user *) datap,
1354 &siginfo);
1355 break;
1356
1357 case PTRACE_SETSIGINFO:
1358 ret = copy_siginfo_from_user32(
1359 &siginfo, (struct compat_siginfo __user *) datap);
1360 if (!ret)
1361 ret = ptrace_setsiginfo(child, &siginfo);
1362 break;
1363#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1364 case PTRACE_GETREGSET:
1365 case PTRACE_SETREGSET:
1366 {
1367 struct iovec kiov;
1368 struct compat_iovec __user *uiov =
1369 (struct compat_iovec __user *) datap;
1370 compat_uptr_t ptr;
1371 compat_size_t len;
1372
1373 if (!access_ok(uiov, sizeof(*uiov)))
1374 return -EFAULT;
1375
1376 if (__get_user(ptr, &uiov->iov_base) ||
1377 __get_user(len, &uiov->iov_len))
1378 return -EFAULT;
1379
1380 kiov.iov_base = compat_ptr(ptr);
1381 kiov.iov_len = len;
1382
1383 ret = ptrace_regset(child, request, addr, &kiov);
1384 if (!ret)
1385 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1386 break;
1387 }
1388#endif
1389
1390 default:
1391 ret = ptrace_request(child, request, addr, data);
1392 }
1393
1394 return ret;
1395}
1396
1397COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1398 compat_long_t, addr, compat_long_t, data)
1399{
1400 struct task_struct *child;
1401 long ret;
1402
1403 if (request == PTRACE_TRACEME) {
1404 ret = ptrace_traceme();
1405 goto out;
1406 }
1407
1408 child = find_get_task_by_vpid(pid);
1409 if (!child) {
1410 ret = -ESRCH;
1411 goto out;
1412 }
1413
1414 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1415 ret = ptrace_attach(child, request, addr, data);
1416 goto out_put_task_struct;
1417 }
1418
1419 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1420 request == PTRACE_INTERRUPT);
1421 if (!ret) {
1422 ret = compat_arch_ptrace(child, request, addr, data);
1423 if (ret || request != PTRACE_DETACH)
1424 ptrace_unfreeze_traced(child);
1425 }
1426
1427 out_put_task_struct:
1428 put_task_struct(child);
1429 out:
1430 return ret;
1431}
1432#endif /* CONFIG_COMPAT */