Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/ptrace.c
4 *
5 * (C) Copyright 1999 Linus Torvalds
6 *
7 * Common interfaces for "ptrace()" which we do not want
8 * to continually duplicate across every architecture.
9 */
10
11#include <linux/capability.h>
12#include <linux/export.h>
13#include <linux/sched.h>
14#include <linux/sched/mm.h>
15#include <linux/sched/coredump.h>
16#include <linux/sched/task.h>
17#include <linux/errno.h>
18#include <linux/mm.h>
19#include <linux/highmem.h>
20#include <linux/pagemap.h>
21#include <linux/ptrace.h>
22#include <linux/security.h>
23#include <linux/signal.h>
24#include <linux/uio.h>
25#include <linux/audit.h>
26#include <linux/pid_namespace.h>
27#include <linux/syscalls.h>
28#include <linux/uaccess.h>
29#include <linux/regset.h>
30#include <linux/hw_breakpoint.h>
31#include <linux/cn_proc.h>
32#include <linux/compat.h>
33#include <linux/sched/signal.h>
34
35#include <asm/syscall.h> /* for syscall_get_* */
36
37/*
38 * Access another process' address space via ptrace.
39 * Source/target buffer must be kernel space,
40 * Do not walk the page table directly, use get_user_pages
41 */
42int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
43 void *buf, int len, unsigned int gup_flags)
44{
45 struct mm_struct *mm;
46 int ret;
47
48 mm = get_task_mm(tsk);
49 if (!mm)
50 return 0;
51
52 if (!tsk->ptrace ||
53 (current != tsk->parent) ||
54 ((get_dumpable(mm) != SUID_DUMP_USER) &&
55 !ptracer_capable(tsk, mm->user_ns))) {
56 mmput(mm);
57 return 0;
58 }
59
60 ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
61 mmput(mm);
62
63 return ret;
64}
65
66
67void __ptrace_link(struct task_struct *child, struct task_struct *new_parent,
68 const struct cred *ptracer_cred)
69{
70 BUG_ON(!list_empty(&child->ptrace_entry));
71 list_add(&child->ptrace_entry, &new_parent->ptraced);
72 child->parent = new_parent;
73 child->ptracer_cred = get_cred(ptracer_cred);
74}
75
76/*
77 * ptrace a task: make the debugger its new parent and
78 * move it to the ptrace list.
79 *
80 * Must be called with the tasklist lock write-held.
81 */
82static void ptrace_link(struct task_struct *child, struct task_struct *new_parent)
83{
84 __ptrace_link(child, new_parent, current_cred());
85}
86
87/**
88 * __ptrace_unlink - unlink ptracee and restore its execution state
89 * @child: ptracee to be unlinked
90 *
91 * Remove @child from the ptrace list, move it back to the original parent,
92 * and restore the execution state so that it conforms to the group stop
93 * state.
94 *
95 * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
96 * exiting. For PTRACE_DETACH, unless the ptracee has been killed between
97 * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
98 * If the ptracer is exiting, the ptracee can be in any state.
99 *
100 * After detach, the ptracee should be in a state which conforms to the
101 * group stop. If the group is stopped or in the process of stopping, the
102 * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
103 * up from TASK_TRACED.
104 *
105 * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
106 * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
107 * to but in the opposite direction of what happens while attaching to a
108 * stopped task. However, in this direction, the intermediate RUNNING
109 * state is not hidden even from the current ptracer and if it immediately
110 * re-attaches and performs a WNOHANG wait(2), it may fail.
111 *
112 * CONTEXT:
113 * write_lock_irq(tasklist_lock)
114 */
115void __ptrace_unlink(struct task_struct *child)
116{
117 const struct cred *old_cred;
118 BUG_ON(!child->ptrace);
119
120 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
121#ifdef TIF_SYSCALL_EMU
122 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
123#endif
124
125 child->parent = child->real_parent;
126 list_del_init(&child->ptrace_entry);
127 old_cred = child->ptracer_cred;
128 child->ptracer_cred = NULL;
129 put_cred(old_cred);
130
131 spin_lock(&child->sighand->siglock);
132 child->ptrace = 0;
133 /*
134 * Clear all pending traps and TRAPPING. TRAPPING should be
135 * cleared regardless of JOBCTL_STOP_PENDING. Do it explicitly.
136 */
137 task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
138 task_clear_jobctl_trapping(child);
139
140 /*
141 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
142 * @child isn't dead.
143 */
144 if (!(child->flags & PF_EXITING) &&
145 (child->signal->flags & SIGNAL_STOP_STOPPED ||
146 child->signal->group_stop_count)) {
147 child->jobctl |= JOBCTL_STOP_PENDING;
148
149 /*
150 * This is only possible if this thread was cloned by the
151 * traced task running in the stopped group, set the signal
152 * for the future reports.
153 * FIXME: we should change ptrace_init_task() to handle this
154 * case.
155 */
156 if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
157 child->jobctl |= SIGSTOP;
158 }
159
160 /*
161 * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
162 * @child in the butt. Note that @resume should be used iff @child
163 * is in TASK_TRACED; otherwise, we might unduly disrupt
164 * TASK_KILLABLE sleeps.
165 */
166 if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
167 ptrace_signal_wake_up(child, true);
168
169 spin_unlock(&child->sighand->siglock);
170}
171
172/* Ensure that nothing can wake it up, even SIGKILL */
173static bool ptrace_freeze_traced(struct task_struct *task)
174{
175 bool ret = false;
176
177 /* Lockless, nobody but us can set this flag */
178 if (task->jobctl & JOBCTL_LISTENING)
179 return ret;
180
181 spin_lock_irq(&task->sighand->siglock);
182 if (task_is_traced(task) && !__fatal_signal_pending(task)) {
183 task->state = __TASK_TRACED;
184 ret = true;
185 }
186 spin_unlock_irq(&task->sighand->siglock);
187
188 return ret;
189}
190
191static void ptrace_unfreeze_traced(struct task_struct *task)
192{
193 if (task->state != __TASK_TRACED)
194 return;
195
196 WARN_ON(!task->ptrace || task->parent != current);
197
198 /*
199 * PTRACE_LISTEN can allow ptrace_trap_notify to wake us up remotely.
200 * Recheck state under the lock to close this race.
201 */
202 spin_lock_irq(&task->sighand->siglock);
203 if (task->state == __TASK_TRACED) {
204 if (__fatal_signal_pending(task))
205 wake_up_state(task, __TASK_TRACED);
206 else
207 task->state = TASK_TRACED;
208 }
209 spin_unlock_irq(&task->sighand->siglock);
210}
211
212/**
213 * ptrace_check_attach - check whether ptracee is ready for ptrace operation
214 * @child: ptracee to check for
215 * @ignore_state: don't check whether @child is currently %TASK_TRACED
216 *
217 * Check whether @child is being ptraced by %current and ready for further
218 * ptrace operations. If @ignore_state is %false, @child also should be in
219 * %TASK_TRACED state and on return the child is guaranteed to be traced
220 * and not executing. If @ignore_state is %true, @child can be in any
221 * state.
222 *
223 * CONTEXT:
224 * Grabs and releases tasklist_lock and @child->sighand->siglock.
225 *
226 * RETURNS:
227 * 0 on success, -ESRCH if %child is not ready.
228 */
229static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
230{
231 int ret = -ESRCH;
232
233 /*
234 * We take the read lock around doing both checks to close a
235 * possible race where someone else was tracing our child and
236 * detached between these two checks. After this locked check,
237 * we are sure that this is our traced child and that can only
238 * be changed by us so it's not changing right after this.
239 */
240 read_lock(&tasklist_lock);
241 if (child->ptrace && child->parent == current) {
242 WARN_ON(child->state == __TASK_TRACED);
243 /*
244 * child->sighand can't be NULL, release_task()
245 * does ptrace_unlink() before __exit_signal().
246 */
247 if (ignore_state || ptrace_freeze_traced(child))
248 ret = 0;
249 }
250 read_unlock(&tasklist_lock);
251
252 if (!ret && !ignore_state) {
253 if (!wait_task_inactive(child, __TASK_TRACED)) {
254 /*
255 * This can only happen if may_ptrace_stop() fails and
256 * ptrace_stop() changes ->state back to TASK_RUNNING,
257 * so we should not worry about leaking __TASK_TRACED.
258 */
259 WARN_ON(child->state == __TASK_TRACED);
260 ret = -ESRCH;
261 }
262 }
263
264 return ret;
265}
266
267static int ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
268{
269 if (mode & PTRACE_MODE_NOAUDIT)
270 return has_ns_capability_noaudit(current, ns, CAP_SYS_PTRACE);
271 else
272 return has_ns_capability(current, 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 ptrace_attach(struct task_struct *task, long request,
359 unsigned long addr,
360 unsigned long flags)
361{
362 bool seize = (request == PTRACE_SEIZE);
363 int retval;
364
365 retval = -EIO;
366 if (seize) {
367 if (addr != 0)
368 goto out;
369 if (flags & ~(unsigned long)PTRACE_O_MASK)
370 goto out;
371 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
372 } else {
373 flags = PT_PTRACED;
374 }
375
376 audit_ptrace(task);
377
378 retval = -EPERM;
379 if (unlikely(task->flags & PF_KTHREAD))
380 goto out;
381 if (same_thread_group(task, current))
382 goto out;
383
384 /*
385 * Protect exec's credential calculations against our interference;
386 * SUID, SGID and LSM creds get determined differently
387 * under ptrace.
388 */
389 retval = -ERESTARTNOINTR;
390 if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
391 goto out;
392
393 task_lock(task);
394 retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
395 task_unlock(task);
396 if (retval)
397 goto unlock_creds;
398
399 write_lock_irq(&tasklist_lock);
400 retval = -EPERM;
401 if (unlikely(task->exit_state))
402 goto unlock_tasklist;
403 if (task->ptrace)
404 goto unlock_tasklist;
405
406 if (seize)
407 flags |= PT_SEIZED;
408 task->ptrace = flags;
409
410 ptrace_link(task, current);
411
412 /* SEIZE doesn't trap tracee on attach */
413 if (!seize)
414 send_sig_info(SIGSTOP, SEND_SIG_PRIV, task);
415
416 spin_lock(&task->sighand->siglock);
417
418 /*
419 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
420 * TRAPPING, and kick it so that it transits to TRACED. TRAPPING
421 * will be cleared if the child completes the transition or any
422 * event which clears the group stop states happens. We'll wait
423 * for the transition to complete before returning from this
424 * function.
425 *
426 * This hides STOPPED -> RUNNING -> TRACED transition from the
427 * attaching thread but a different thread in the same group can
428 * still observe the transient RUNNING state. IOW, if another
429 * thread's WNOHANG wait(2) on the stopped tracee races against
430 * ATTACH, the wait(2) may fail due to the transient RUNNING.
431 *
432 * The following task_is_stopped() test is safe as both transitions
433 * in and out of STOPPED are protected by siglock.
434 */
435 if (task_is_stopped(task) &&
436 task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
437 signal_wake_up_state(task, __TASK_STOPPED);
438
439 spin_unlock(&task->sighand->siglock);
440
441 retval = 0;
442unlock_tasklist:
443 write_unlock_irq(&tasklist_lock);
444unlock_creds:
445 mutex_unlock(&task->signal->cred_guard_mutex);
446out:
447 if (!retval) {
448 /*
449 * We do not bother to change retval or clear JOBCTL_TRAPPING
450 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
451 * not return to user-mode, it will exit and clear this bit in
452 * __ptrace_unlink() if it wasn't already cleared by the tracee;
453 * and until then nobody can ptrace this task.
454 */
455 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
456 proc_ptrace_connector(task, PTRACE_ATTACH);
457 }
458
459 return retval;
460}
461
462/**
463 * ptrace_traceme -- helper for PTRACE_TRACEME
464 *
465 * Performs checks and sets PT_PTRACED.
466 * Should be used by all ptrace implementations for PTRACE_TRACEME.
467 */
468static int ptrace_traceme(void)
469{
470 int ret = -EPERM;
471
472 write_lock_irq(&tasklist_lock);
473 /* Are we already being traced? */
474 if (!current->ptrace) {
475 ret = security_ptrace_traceme(current->parent);
476 /*
477 * Check PF_EXITING to ensure ->real_parent has not passed
478 * exit_ptrace(). Otherwise we don't report the error but
479 * pretend ->real_parent untraces us right after return.
480 */
481 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
482 current->ptrace = PT_PTRACED;
483 ptrace_link(current, current->real_parent);
484 }
485 }
486 write_unlock_irq(&tasklist_lock);
487
488 return ret;
489}
490
491/*
492 * Called with irqs disabled, returns true if childs should reap themselves.
493 */
494static int ignoring_children(struct sighand_struct *sigh)
495{
496 int ret;
497 spin_lock(&sigh->siglock);
498 ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
499 (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
500 spin_unlock(&sigh->siglock);
501 return ret;
502}
503
504/*
505 * Called with tasklist_lock held for writing.
506 * Unlink a traced task, and clean it up if it was a traced zombie.
507 * Return true if it needs to be reaped with release_task().
508 * (We can't call release_task() here because we already hold tasklist_lock.)
509 *
510 * If it's a zombie, our attachedness prevented normal parent notification
511 * or self-reaping. Do notification now if it would have happened earlier.
512 * If it should reap itself, return true.
513 *
514 * If it's our own child, there is no notification to do. But if our normal
515 * children self-reap, then this child was prevented by ptrace and we must
516 * reap it now, in that case we must also wake up sub-threads sleeping in
517 * do_wait().
518 */
519static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
520{
521 bool dead;
522
523 __ptrace_unlink(p);
524
525 if (p->exit_state != EXIT_ZOMBIE)
526 return false;
527
528 dead = !thread_group_leader(p);
529
530 if (!dead && thread_group_empty(p)) {
531 if (!same_thread_group(p->real_parent, tracer))
532 dead = do_notify_parent(p, p->exit_signal);
533 else if (ignoring_children(tracer->sighand)) {
534 __wake_up_parent(p, tracer);
535 dead = true;
536 }
537 }
538 /* Mark it as in the process of being reaped. */
539 if (dead)
540 p->exit_state = EXIT_DEAD;
541 return dead;
542}
543
544static int ptrace_detach(struct task_struct *child, unsigned int data)
545{
546 if (!valid_signal(data))
547 return -EIO;
548
549 /* Architecture-specific hardware disable .. */
550 ptrace_disable(child);
551
552 write_lock_irq(&tasklist_lock);
553 /*
554 * We rely on ptrace_freeze_traced(). It can't be killed and
555 * untraced by another thread, it can't be a zombie.
556 */
557 WARN_ON(!child->ptrace || child->exit_state);
558 /*
559 * tasklist_lock avoids the race with wait_task_stopped(), see
560 * the comment in ptrace_resume().
561 */
562 child->exit_code = data;
563 __ptrace_detach(current, child);
564 write_unlock_irq(&tasklist_lock);
565
566 proc_ptrace_connector(child, PTRACE_DETACH);
567
568 return 0;
569}
570
571/*
572 * Detach all tasks we were using ptrace on. Called with tasklist held
573 * for writing.
574 */
575void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
576{
577 struct task_struct *p, *n;
578
579 list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
580 if (unlikely(p->ptrace & PT_EXITKILL))
581 send_sig_info(SIGKILL, SEND_SIG_PRIV, p);
582
583 if (__ptrace_detach(tracer, p))
584 list_add(&p->ptrace_entry, dead);
585 }
586}
587
588int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
589{
590 int copied = 0;
591
592 while (len > 0) {
593 char buf[128];
594 int this_len, retval;
595
596 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
597 retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
598
599 if (!retval) {
600 if (copied)
601 break;
602 return -EIO;
603 }
604 if (copy_to_user(dst, buf, retval))
605 return -EFAULT;
606 copied += retval;
607 src += retval;
608 dst += retval;
609 len -= retval;
610 }
611 return copied;
612}
613
614int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
615{
616 int copied = 0;
617
618 while (len > 0) {
619 char buf[128];
620 int this_len, retval;
621
622 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
623 if (copy_from_user(buf, src, this_len))
624 return -EFAULT;
625 retval = ptrace_access_vm(tsk, dst, buf, this_len,
626 FOLL_FORCE | FOLL_WRITE);
627 if (!retval) {
628 if (copied)
629 break;
630 return -EIO;
631 }
632 copied += retval;
633 src += retval;
634 dst += retval;
635 len -= retval;
636 }
637 return copied;
638}
639
640static int ptrace_setoptions(struct task_struct *child, unsigned long data)
641{
642 unsigned flags;
643
644 if (data & ~(unsigned long)PTRACE_O_MASK)
645 return -EINVAL;
646
647 if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
648 if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
649 !IS_ENABLED(CONFIG_SECCOMP))
650 return -EINVAL;
651
652 if (!capable(CAP_SYS_ADMIN))
653 return -EPERM;
654
655 if (seccomp_mode(¤t->seccomp) != SECCOMP_MODE_DISABLED ||
656 current->ptrace & PT_SUSPEND_SECCOMP)
657 return -EPERM;
658 }
659
660 /* Avoid intermediate state when all opts are cleared */
661 flags = child->ptrace;
662 flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
663 flags |= (data << PT_OPT_FLAG_SHIFT);
664 child->ptrace = flags;
665
666 return 0;
667}
668
669static int ptrace_getsiginfo(struct task_struct *child, kernel_siginfo_t *info)
670{
671 unsigned long flags;
672 int error = -ESRCH;
673
674 if (lock_task_sighand(child, &flags)) {
675 error = -EINVAL;
676 if (likely(child->last_siginfo != NULL)) {
677 copy_siginfo(info, child->last_siginfo);
678 error = 0;
679 }
680 unlock_task_sighand(child, &flags);
681 }
682 return error;
683}
684
685static int ptrace_setsiginfo(struct task_struct *child, const kernel_siginfo_t *info)
686{
687 unsigned long flags;
688 int error = -ESRCH;
689
690 if (lock_task_sighand(child, &flags)) {
691 error = -EINVAL;
692 if (likely(child->last_siginfo != NULL)) {
693 copy_siginfo(child->last_siginfo, info);
694 error = 0;
695 }
696 unlock_task_sighand(child, &flags);
697 }
698 return error;
699}
700
701static int ptrace_peek_siginfo(struct task_struct *child,
702 unsigned long addr,
703 unsigned long data)
704{
705 struct ptrace_peeksiginfo_args arg;
706 struct sigpending *pending;
707 struct sigqueue *q;
708 int ret, i;
709
710 ret = copy_from_user(&arg, (void __user *) addr,
711 sizeof(struct ptrace_peeksiginfo_args));
712 if (ret)
713 return -EFAULT;
714
715 if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
716 return -EINVAL; /* unknown flags */
717
718 if (arg.nr < 0)
719 return -EINVAL;
720
721 /* Ensure arg.off fits in an unsigned long */
722 if (arg.off > ULONG_MAX)
723 return 0;
724
725 if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
726 pending = &child->signal->shared_pending;
727 else
728 pending = &child->pending;
729
730 for (i = 0; i < arg.nr; ) {
731 kernel_siginfo_t info;
732 unsigned long off = arg.off + i;
733 bool found = false;
734
735 spin_lock_irq(&child->sighand->siglock);
736 list_for_each_entry(q, &pending->list, list) {
737 if (!off--) {
738 found = true;
739 copy_siginfo(&info, &q->info);
740 break;
741 }
742 }
743 spin_unlock_irq(&child->sighand->siglock);
744
745 if (!found) /* beyond the end of the list */
746 break;
747
748#ifdef CONFIG_COMPAT
749 if (unlikely(in_compat_syscall())) {
750 compat_siginfo_t __user *uinfo = compat_ptr(data);
751
752 if (copy_siginfo_to_user32(uinfo, &info)) {
753 ret = -EFAULT;
754 break;
755 }
756
757 } else
758#endif
759 {
760 siginfo_t __user *uinfo = (siginfo_t __user *) data;
761
762 if (copy_siginfo_to_user(uinfo, &info)) {
763 ret = -EFAULT;
764 break;
765 }
766 }
767
768 data += sizeof(siginfo_t);
769 i++;
770
771 if (signal_pending(current))
772 break;
773
774 cond_resched();
775 }
776
777 if (i > 0)
778 return i;
779
780 return ret;
781}
782
783#ifdef PTRACE_SINGLESTEP
784#define is_singlestep(request) ((request) == PTRACE_SINGLESTEP)
785#else
786#define is_singlestep(request) 0
787#endif
788
789#ifdef PTRACE_SINGLEBLOCK
790#define is_singleblock(request) ((request) == PTRACE_SINGLEBLOCK)
791#else
792#define is_singleblock(request) 0
793#endif
794
795#ifdef PTRACE_SYSEMU
796#define is_sysemu_singlestep(request) ((request) == PTRACE_SYSEMU_SINGLESTEP)
797#else
798#define is_sysemu_singlestep(request) 0
799#endif
800
801static int ptrace_resume(struct task_struct *child, long request,
802 unsigned long data)
803{
804 bool need_siglock;
805
806 if (!valid_signal(data))
807 return -EIO;
808
809 if (request == PTRACE_SYSCALL)
810 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
811 else
812 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
813
814#ifdef TIF_SYSCALL_EMU
815 if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
816 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
817 else
818 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
819#endif
820
821 if (is_singleblock(request)) {
822 if (unlikely(!arch_has_block_step()))
823 return -EIO;
824 user_enable_block_step(child);
825 } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
826 if (unlikely(!arch_has_single_step()))
827 return -EIO;
828 user_enable_single_step(child);
829 } else {
830 user_disable_single_step(child);
831 }
832
833 /*
834 * Change ->exit_code and ->state under siglock to avoid the race
835 * with wait_task_stopped() in between; a non-zero ->exit_code will
836 * wrongly look like another report from tracee.
837 *
838 * Note that we need siglock even if ->exit_code == data and/or this
839 * status was not reported yet, the new status must not be cleared by
840 * wait_task_stopped() after resume.
841 *
842 * If data == 0 we do not care if wait_task_stopped() reports the old
843 * status and clears the code too; this can't race with the tracee, it
844 * takes siglock after resume.
845 */
846 need_siglock = data && !thread_group_empty(current);
847 if (need_siglock)
848 spin_lock_irq(&child->sighand->siglock);
849 child->exit_code = data;
850 wake_up_state(child, __TASK_TRACED);
851 if (need_siglock)
852 spin_unlock_irq(&child->sighand->siglock);
853
854 return 0;
855}
856
857#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
858
859static const struct user_regset *
860find_regset(const struct user_regset_view *view, unsigned int type)
861{
862 const struct user_regset *regset;
863 int n;
864
865 for (n = 0; n < view->n; ++n) {
866 regset = view->regsets + n;
867 if (regset->core_note_type == type)
868 return regset;
869 }
870
871 return NULL;
872}
873
874static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
875 struct iovec *kiov)
876{
877 const struct user_regset_view *view = task_user_regset_view(task);
878 const struct user_regset *regset = find_regset(view, type);
879 int regset_no;
880
881 if (!regset || (kiov->iov_len % regset->size) != 0)
882 return -EINVAL;
883
884 regset_no = regset - view->regsets;
885 kiov->iov_len = min(kiov->iov_len,
886 (__kernel_size_t) (regset->n * regset->size));
887
888 if (req == PTRACE_GETREGSET)
889 return copy_regset_to_user(task, view, regset_no, 0,
890 kiov->iov_len, kiov->iov_base);
891 else
892 return copy_regset_from_user(task, view, regset_no, 0,
893 kiov->iov_len, kiov->iov_base);
894}
895
896/*
897 * This is declared in linux/regset.h and defined in machine-dependent
898 * code. We put the export here, near the primary machine-neutral use,
899 * to ensure no machine forgets it.
900 */
901EXPORT_SYMBOL_GPL(task_user_regset_view);
902
903static unsigned long
904ptrace_get_syscall_info_entry(struct task_struct *child, struct pt_regs *regs,
905 struct ptrace_syscall_info *info)
906{
907 unsigned long args[ARRAY_SIZE(info->entry.args)];
908 int i;
909
910 info->op = PTRACE_SYSCALL_INFO_ENTRY;
911 info->entry.nr = syscall_get_nr(child, regs);
912 syscall_get_arguments(child, regs, args);
913 for (i = 0; i < ARRAY_SIZE(args); i++)
914 info->entry.args[i] = args[i];
915
916 /* args is the last field in struct ptrace_syscall_info.entry */
917 return offsetofend(struct ptrace_syscall_info, entry.args);
918}
919
920static unsigned long
921ptrace_get_syscall_info_seccomp(struct task_struct *child, struct pt_regs *regs,
922 struct ptrace_syscall_info *info)
923{
924 /*
925 * As struct ptrace_syscall_info.entry is currently a subset
926 * of struct ptrace_syscall_info.seccomp, it makes sense to
927 * initialize that subset using ptrace_get_syscall_info_entry().
928 * This can be reconsidered in the future if these structures
929 * diverge significantly enough.
930 */
931 ptrace_get_syscall_info_entry(child, regs, info);
932 info->op = PTRACE_SYSCALL_INFO_SECCOMP;
933 info->seccomp.ret_data = child->ptrace_message;
934
935 /* ret_data is the last field in struct ptrace_syscall_info.seccomp */
936 return offsetofend(struct ptrace_syscall_info, seccomp.ret_data);
937}
938
939static unsigned long
940ptrace_get_syscall_info_exit(struct task_struct *child, struct pt_regs *regs,
941 struct ptrace_syscall_info *info)
942{
943 info->op = PTRACE_SYSCALL_INFO_EXIT;
944 info->exit.rval = syscall_get_error(child, regs);
945 info->exit.is_error = !!info->exit.rval;
946 if (!info->exit.is_error)
947 info->exit.rval = syscall_get_return_value(child, regs);
948
949 /* is_error is the last field in struct ptrace_syscall_info.exit */
950 return offsetofend(struct ptrace_syscall_info, exit.is_error);
951}
952
953static int
954ptrace_get_syscall_info(struct task_struct *child, unsigned long user_size,
955 void __user *datavp)
956{
957 struct pt_regs *regs = task_pt_regs(child);
958 struct ptrace_syscall_info info = {
959 .op = PTRACE_SYSCALL_INFO_NONE,
960 .arch = syscall_get_arch(child),
961 .instruction_pointer = instruction_pointer(regs),
962 .stack_pointer = user_stack_pointer(regs),
963 };
964 unsigned long actual_size = offsetof(struct ptrace_syscall_info, entry);
965 unsigned long write_size;
966
967 /*
968 * This does not need lock_task_sighand() to access
969 * child->last_siginfo because ptrace_freeze_traced()
970 * called earlier by ptrace_check_attach() ensures that
971 * the tracee cannot go away and clear its last_siginfo.
972 */
973 switch (child->last_siginfo ? child->last_siginfo->si_code : 0) {
974 case SIGTRAP | 0x80:
975 switch (child->ptrace_message) {
976 case PTRACE_EVENTMSG_SYSCALL_ENTRY:
977 actual_size = ptrace_get_syscall_info_entry(child, regs,
978 &info);
979 break;
980 case PTRACE_EVENTMSG_SYSCALL_EXIT:
981 actual_size = ptrace_get_syscall_info_exit(child, regs,
982 &info);
983 break;
984 }
985 break;
986 case SIGTRAP | (PTRACE_EVENT_SECCOMP << 8):
987 actual_size = ptrace_get_syscall_info_seccomp(child, regs,
988 &info);
989 break;
990 }
991
992 write_size = min(actual_size, user_size);
993 return copy_to_user(datavp, &info, write_size) ? -EFAULT : actual_size;
994}
995#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
996
997int ptrace_request(struct task_struct *child, long request,
998 unsigned long addr, unsigned long data)
999{
1000 bool seized = child->ptrace & PT_SEIZED;
1001 int ret = -EIO;
1002 kernel_siginfo_t siginfo, *si;
1003 void __user *datavp = (void __user *) data;
1004 unsigned long __user *datalp = datavp;
1005 unsigned long flags;
1006
1007 switch (request) {
1008 case PTRACE_PEEKTEXT:
1009 case PTRACE_PEEKDATA:
1010 return generic_ptrace_peekdata(child, addr, data);
1011 case PTRACE_POKETEXT:
1012 case PTRACE_POKEDATA:
1013 return generic_ptrace_pokedata(child, addr, data);
1014
1015#ifdef PTRACE_OLDSETOPTIONS
1016 case PTRACE_OLDSETOPTIONS:
1017#endif
1018 case PTRACE_SETOPTIONS:
1019 ret = ptrace_setoptions(child, data);
1020 break;
1021 case PTRACE_GETEVENTMSG:
1022 ret = put_user(child->ptrace_message, datalp);
1023 break;
1024
1025 case PTRACE_PEEKSIGINFO:
1026 ret = ptrace_peek_siginfo(child, addr, data);
1027 break;
1028
1029 case PTRACE_GETSIGINFO:
1030 ret = ptrace_getsiginfo(child, &siginfo);
1031 if (!ret)
1032 ret = copy_siginfo_to_user(datavp, &siginfo);
1033 break;
1034
1035 case PTRACE_SETSIGINFO:
1036 ret = copy_siginfo_from_user(&siginfo, datavp);
1037 if (!ret)
1038 ret = ptrace_setsiginfo(child, &siginfo);
1039 break;
1040
1041 case PTRACE_GETSIGMASK: {
1042 sigset_t *mask;
1043
1044 if (addr != sizeof(sigset_t)) {
1045 ret = -EINVAL;
1046 break;
1047 }
1048
1049 if (test_tsk_restore_sigmask(child))
1050 mask = &child->saved_sigmask;
1051 else
1052 mask = &child->blocked;
1053
1054 if (copy_to_user(datavp, mask, sizeof(sigset_t)))
1055 ret = -EFAULT;
1056 else
1057 ret = 0;
1058
1059 break;
1060 }
1061
1062 case PTRACE_SETSIGMASK: {
1063 sigset_t new_set;
1064
1065 if (addr != sizeof(sigset_t)) {
1066 ret = -EINVAL;
1067 break;
1068 }
1069
1070 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
1071 ret = -EFAULT;
1072 break;
1073 }
1074
1075 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
1076
1077 /*
1078 * Every thread does recalc_sigpending() after resume, so
1079 * retarget_shared_pending() and recalc_sigpending() are not
1080 * called here.
1081 */
1082 spin_lock_irq(&child->sighand->siglock);
1083 child->blocked = new_set;
1084 spin_unlock_irq(&child->sighand->siglock);
1085
1086 clear_tsk_restore_sigmask(child);
1087
1088 ret = 0;
1089 break;
1090 }
1091
1092 case PTRACE_INTERRUPT:
1093 /*
1094 * Stop tracee without any side-effect on signal or job
1095 * control. At least one trap is guaranteed to happen
1096 * after this request. If @child is already trapped, the
1097 * current trap is not disturbed and another trap will
1098 * happen after the current trap is ended with PTRACE_CONT.
1099 *
1100 * The actual trap might not be PTRACE_EVENT_STOP trap but
1101 * the pending condition is cleared regardless.
1102 */
1103 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1104 break;
1105
1106 /*
1107 * INTERRUPT doesn't disturb existing trap sans one
1108 * exception. If ptracer issued LISTEN for the current
1109 * STOP, this INTERRUPT should clear LISTEN and re-trap
1110 * tracee into STOP.
1111 */
1112 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
1113 ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1114
1115 unlock_task_sighand(child, &flags);
1116 ret = 0;
1117 break;
1118
1119 case PTRACE_LISTEN:
1120 /*
1121 * Listen for events. Tracee must be in STOP. It's not
1122 * resumed per-se but is not considered to be in TRACED by
1123 * wait(2) or ptrace(2). If an async event (e.g. group
1124 * stop state change) happens, tracee will enter STOP trap
1125 * again. Alternatively, ptracer can issue INTERRUPT to
1126 * finish listening and re-trap tracee into STOP.
1127 */
1128 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1129 break;
1130
1131 si = child->last_siginfo;
1132 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1133 child->jobctl |= JOBCTL_LISTENING;
1134 /*
1135 * If NOTIFY is set, it means event happened between
1136 * start of this trap and now. Trigger re-trap.
1137 */
1138 if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1139 ptrace_signal_wake_up(child, true);
1140 ret = 0;
1141 }
1142 unlock_task_sighand(child, &flags);
1143 break;
1144
1145 case PTRACE_DETACH: /* detach a process that was attached. */
1146 ret = ptrace_detach(child, data);
1147 break;
1148
1149#ifdef CONFIG_BINFMT_ELF_FDPIC
1150 case PTRACE_GETFDPIC: {
1151 struct mm_struct *mm = get_task_mm(child);
1152 unsigned long tmp = 0;
1153
1154 ret = -ESRCH;
1155 if (!mm)
1156 break;
1157
1158 switch (addr) {
1159 case PTRACE_GETFDPIC_EXEC:
1160 tmp = mm->context.exec_fdpic_loadmap;
1161 break;
1162 case PTRACE_GETFDPIC_INTERP:
1163 tmp = mm->context.interp_fdpic_loadmap;
1164 break;
1165 default:
1166 break;
1167 }
1168 mmput(mm);
1169
1170 ret = put_user(tmp, datalp);
1171 break;
1172 }
1173#endif
1174
1175#ifdef PTRACE_SINGLESTEP
1176 case PTRACE_SINGLESTEP:
1177#endif
1178#ifdef PTRACE_SINGLEBLOCK
1179 case PTRACE_SINGLEBLOCK:
1180#endif
1181#ifdef PTRACE_SYSEMU
1182 case PTRACE_SYSEMU:
1183 case PTRACE_SYSEMU_SINGLESTEP:
1184#endif
1185 case PTRACE_SYSCALL:
1186 case PTRACE_CONT:
1187 return ptrace_resume(child, request, data);
1188
1189 case PTRACE_KILL:
1190 if (child->exit_state) /* already dead */
1191 return 0;
1192 return ptrace_resume(child, request, SIGKILL);
1193
1194#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1195 case PTRACE_GETREGSET:
1196 case PTRACE_SETREGSET: {
1197 struct iovec kiov;
1198 struct iovec __user *uiov = datavp;
1199
1200 if (!access_ok(uiov, sizeof(*uiov)))
1201 return -EFAULT;
1202
1203 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1204 __get_user(kiov.iov_len, &uiov->iov_len))
1205 return -EFAULT;
1206
1207 ret = ptrace_regset(child, request, addr, &kiov);
1208 if (!ret)
1209 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1210 break;
1211 }
1212
1213 case PTRACE_GET_SYSCALL_INFO:
1214 ret = ptrace_get_syscall_info(child, addr, datavp);
1215 break;
1216#endif
1217
1218 case PTRACE_SECCOMP_GET_FILTER:
1219 ret = seccomp_get_filter(child, addr, datavp);
1220 break;
1221
1222 case PTRACE_SECCOMP_GET_METADATA:
1223 ret = seccomp_get_metadata(child, addr, datavp);
1224 break;
1225
1226 default:
1227 break;
1228 }
1229
1230 return ret;
1231}
1232
1233#ifndef arch_ptrace_attach
1234#define arch_ptrace_attach(child) do { } while (0)
1235#endif
1236
1237SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1238 unsigned long, data)
1239{
1240 struct task_struct *child;
1241 long ret;
1242
1243 if (request == PTRACE_TRACEME) {
1244 ret = ptrace_traceme();
1245 if (!ret)
1246 arch_ptrace_attach(current);
1247 goto out;
1248 }
1249
1250 child = find_get_task_by_vpid(pid);
1251 if (!child) {
1252 ret = -ESRCH;
1253 goto out;
1254 }
1255
1256 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1257 ret = ptrace_attach(child, request, addr, data);
1258 /*
1259 * Some architectures need to do book-keeping after
1260 * a ptrace attach.
1261 */
1262 if (!ret)
1263 arch_ptrace_attach(child);
1264 goto out_put_task_struct;
1265 }
1266
1267 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1268 request == PTRACE_INTERRUPT);
1269 if (ret < 0)
1270 goto out_put_task_struct;
1271
1272 ret = arch_ptrace(child, request, addr, data);
1273 if (ret || request != PTRACE_DETACH)
1274 ptrace_unfreeze_traced(child);
1275
1276 out_put_task_struct:
1277 put_task_struct(child);
1278 out:
1279 return ret;
1280}
1281
1282int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1283 unsigned long data)
1284{
1285 unsigned long tmp;
1286 int copied;
1287
1288 copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1289 if (copied != sizeof(tmp))
1290 return -EIO;
1291 return put_user(tmp, (unsigned long __user *)data);
1292}
1293
1294int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1295 unsigned long data)
1296{
1297 int copied;
1298
1299 copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1300 FOLL_FORCE | FOLL_WRITE);
1301 return (copied == sizeof(data)) ? 0 : -EIO;
1302}
1303
1304#if defined CONFIG_COMPAT
1305
1306int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1307 compat_ulong_t addr, compat_ulong_t data)
1308{
1309 compat_ulong_t __user *datap = compat_ptr(data);
1310 compat_ulong_t word;
1311 kernel_siginfo_t siginfo;
1312 int ret;
1313
1314 switch (request) {
1315 case PTRACE_PEEKTEXT:
1316 case PTRACE_PEEKDATA:
1317 ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1318 FOLL_FORCE);
1319 if (ret != sizeof(word))
1320 ret = -EIO;
1321 else
1322 ret = put_user(word, datap);
1323 break;
1324
1325 case PTRACE_POKETEXT:
1326 case PTRACE_POKEDATA:
1327 ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1328 FOLL_FORCE | FOLL_WRITE);
1329 ret = (ret != sizeof(data) ? -EIO : 0);
1330 break;
1331
1332 case PTRACE_GETEVENTMSG:
1333 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1334 break;
1335
1336 case PTRACE_GETSIGINFO:
1337 ret = ptrace_getsiginfo(child, &siginfo);
1338 if (!ret)
1339 ret = copy_siginfo_to_user32(
1340 (struct compat_siginfo __user *) datap,
1341 &siginfo);
1342 break;
1343
1344 case PTRACE_SETSIGINFO:
1345 ret = copy_siginfo_from_user32(
1346 &siginfo, (struct compat_siginfo __user *) datap);
1347 if (!ret)
1348 ret = ptrace_setsiginfo(child, &siginfo);
1349 break;
1350#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1351 case PTRACE_GETREGSET:
1352 case PTRACE_SETREGSET:
1353 {
1354 struct iovec kiov;
1355 struct compat_iovec __user *uiov =
1356 (struct compat_iovec __user *) datap;
1357 compat_uptr_t ptr;
1358 compat_size_t len;
1359
1360 if (!access_ok(uiov, sizeof(*uiov)))
1361 return -EFAULT;
1362
1363 if (__get_user(ptr, &uiov->iov_base) ||
1364 __get_user(len, &uiov->iov_len))
1365 return -EFAULT;
1366
1367 kiov.iov_base = compat_ptr(ptr);
1368 kiov.iov_len = len;
1369
1370 ret = ptrace_regset(child, request, addr, &kiov);
1371 if (!ret)
1372 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1373 break;
1374 }
1375#endif
1376
1377 default:
1378 ret = ptrace_request(child, request, addr, data);
1379 }
1380
1381 return ret;
1382}
1383
1384COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1385 compat_long_t, addr, compat_long_t, data)
1386{
1387 struct task_struct *child;
1388 long ret;
1389
1390 if (request == PTRACE_TRACEME) {
1391 ret = ptrace_traceme();
1392 goto out;
1393 }
1394
1395 child = find_get_task_by_vpid(pid);
1396 if (!child) {
1397 ret = -ESRCH;
1398 goto out;
1399 }
1400
1401 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1402 ret = ptrace_attach(child, request, addr, data);
1403 /*
1404 * Some architectures need to do book-keeping after
1405 * a ptrace attach.
1406 */
1407 if (!ret)
1408 arch_ptrace_attach(child);
1409 goto out_put_task_struct;
1410 }
1411
1412 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1413 request == PTRACE_INTERRUPT);
1414 if (!ret) {
1415 ret = compat_arch_ptrace(child, request, addr, data);
1416 if (ret || request != PTRACE_DETACH)
1417 ptrace_unfreeze_traced(child);
1418 }
1419
1420 out_put_task_struct:
1421 put_task_struct(child);
1422 out:
1423 return ret;
1424}
1425#endif /* CONFIG_COMPAT */
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/sched/mm.h>
14#include <linux/sched/coredump.h>
15#include <linux/sched/task.h>
16#include <linux/errno.h>
17#include <linux/mm.h>
18#include <linux/highmem.h>
19#include <linux/pagemap.h>
20#include <linux/ptrace.h>
21#include <linux/security.h>
22#include <linux/signal.h>
23#include <linux/uio.h>
24#include <linux/audit.h>
25#include <linux/pid_namespace.h>
26#include <linux/syscalls.h>
27#include <linux/uaccess.h>
28#include <linux/regset.h>
29#include <linux/hw_breakpoint.h>
30#include <linux/cn_proc.h>
31#include <linux/compat.h>
32
33/*
34 * Access another process' address space via ptrace.
35 * Source/target buffer must be kernel space,
36 * Do not walk the page table directly, use get_user_pages
37 */
38int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
39 void *buf, int len, unsigned int gup_flags)
40{
41 struct mm_struct *mm;
42 int ret;
43
44 mm = get_task_mm(tsk);
45 if (!mm)
46 return 0;
47
48 if (!tsk->ptrace ||
49 (current != tsk->parent) ||
50 ((get_dumpable(mm) != SUID_DUMP_USER) &&
51 !ptracer_capable(tsk, mm->user_ns))) {
52 mmput(mm);
53 return 0;
54 }
55
56 ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
57 mmput(mm);
58
59 return ret;
60}
61
62
63void __ptrace_link(struct task_struct *child, struct task_struct *new_parent,
64 const struct cred *ptracer_cred)
65{
66 BUG_ON(!list_empty(&child->ptrace_entry));
67 list_add(&child->ptrace_entry, &new_parent->ptraced);
68 child->parent = new_parent;
69 child->ptracer_cred = get_cred(ptracer_cred);
70}
71
72/*
73 * ptrace a task: make the debugger its new parent and
74 * move it to the ptrace list.
75 *
76 * Must be called with the tasklist lock write-held.
77 */
78static void ptrace_link(struct task_struct *child, struct task_struct *new_parent)
79{
80 rcu_read_lock();
81 __ptrace_link(child, new_parent, __task_cred(new_parent));
82 rcu_read_unlock();
83}
84
85/**
86 * __ptrace_unlink - unlink ptracee and restore its execution state
87 * @child: ptracee to be unlinked
88 *
89 * Remove @child from the ptrace list, move it back to the original parent,
90 * and restore the execution state so that it conforms to the group stop
91 * state.
92 *
93 * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
94 * exiting. For PTRACE_DETACH, unless the ptracee has been killed between
95 * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
96 * If the ptracer is exiting, the ptracee can be in any state.
97 *
98 * After detach, the ptracee should be in a state which conforms to the
99 * group stop. If the group is stopped or in the process of stopping, the
100 * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
101 * up from TASK_TRACED.
102 *
103 * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
104 * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
105 * to but in the opposite direction of what happens while attaching to a
106 * stopped task. However, in this direction, the intermediate RUNNING
107 * state is not hidden even from the current ptracer and if it immediately
108 * re-attaches and performs a WNOHANG wait(2), it may fail.
109 *
110 * CONTEXT:
111 * write_lock_irq(tasklist_lock)
112 */
113void __ptrace_unlink(struct task_struct *child)
114{
115 const struct cred *old_cred;
116 BUG_ON(!child->ptrace);
117
118 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
119
120 child->parent = child->real_parent;
121 list_del_init(&child->ptrace_entry);
122 old_cred = child->ptracer_cred;
123 child->ptracer_cred = NULL;
124 put_cred(old_cred);
125
126 spin_lock(&child->sighand->siglock);
127 child->ptrace = 0;
128 /*
129 * Clear all pending traps and TRAPPING. TRAPPING should be
130 * cleared regardless of JOBCTL_STOP_PENDING. Do it explicitly.
131 */
132 task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
133 task_clear_jobctl_trapping(child);
134
135 /*
136 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
137 * @child isn't dead.
138 */
139 if (!(child->flags & PF_EXITING) &&
140 (child->signal->flags & SIGNAL_STOP_STOPPED ||
141 child->signal->group_stop_count)) {
142 child->jobctl |= JOBCTL_STOP_PENDING;
143
144 /*
145 * This is only possible if this thread was cloned by the
146 * traced task running in the stopped group, set the signal
147 * for the future reports.
148 * FIXME: we should change ptrace_init_task() to handle this
149 * case.
150 */
151 if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
152 child->jobctl |= SIGSTOP;
153 }
154
155 /*
156 * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
157 * @child in the butt. Note that @resume should be used iff @child
158 * is in TASK_TRACED; otherwise, we might unduly disrupt
159 * TASK_KILLABLE sleeps.
160 */
161 if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
162 ptrace_signal_wake_up(child, true);
163
164 spin_unlock(&child->sighand->siglock);
165}
166
167/* Ensure that nothing can wake it up, even SIGKILL */
168static bool ptrace_freeze_traced(struct task_struct *task)
169{
170 bool ret = false;
171
172 /* Lockless, nobody but us can set this flag */
173 if (task->jobctl & JOBCTL_LISTENING)
174 return ret;
175
176 spin_lock_irq(&task->sighand->siglock);
177 if (task_is_traced(task) && !__fatal_signal_pending(task)) {
178 task->state = __TASK_TRACED;
179 ret = true;
180 }
181 spin_unlock_irq(&task->sighand->siglock);
182
183 return ret;
184}
185
186static void ptrace_unfreeze_traced(struct task_struct *task)
187{
188 if (task->state != __TASK_TRACED)
189 return;
190
191 WARN_ON(!task->ptrace || task->parent != current);
192
193 /*
194 * PTRACE_LISTEN can allow ptrace_trap_notify to wake us up remotely.
195 * Recheck state under the lock to close this race.
196 */
197 spin_lock_irq(&task->sighand->siglock);
198 if (task->state == __TASK_TRACED) {
199 if (__fatal_signal_pending(task))
200 wake_up_state(task, __TASK_TRACED);
201 else
202 task->state = TASK_TRACED;
203 }
204 spin_unlock_irq(&task->sighand->siglock);
205}
206
207/**
208 * ptrace_check_attach - check whether ptracee is ready for ptrace operation
209 * @child: ptracee to check for
210 * @ignore_state: don't check whether @child is currently %TASK_TRACED
211 *
212 * Check whether @child is being ptraced by %current and ready for further
213 * ptrace operations. If @ignore_state is %false, @child also should be in
214 * %TASK_TRACED state and on return the child is guaranteed to be traced
215 * and not executing. If @ignore_state is %true, @child can be in any
216 * state.
217 *
218 * CONTEXT:
219 * Grabs and releases tasklist_lock and @child->sighand->siglock.
220 *
221 * RETURNS:
222 * 0 on success, -ESRCH if %child is not ready.
223 */
224static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
225{
226 int ret = -ESRCH;
227
228 /*
229 * We take the read lock around doing both checks to close a
230 * possible race where someone else was tracing our child and
231 * detached between these two checks. After this locked check,
232 * we are sure that this is our traced child and that can only
233 * be changed by us so it's not changing right after this.
234 */
235 read_lock(&tasklist_lock);
236 if (child->ptrace && child->parent == current) {
237 WARN_ON(child->state == __TASK_TRACED);
238 /*
239 * child->sighand can't be NULL, release_task()
240 * does ptrace_unlink() before __exit_signal().
241 */
242 if (ignore_state || ptrace_freeze_traced(child))
243 ret = 0;
244 }
245 read_unlock(&tasklist_lock);
246
247 if (!ret && !ignore_state) {
248 if (!wait_task_inactive(child, __TASK_TRACED)) {
249 /*
250 * This can only happen if may_ptrace_stop() fails and
251 * ptrace_stop() changes ->state back to TASK_RUNNING,
252 * so we should not worry about leaking __TASK_TRACED.
253 */
254 WARN_ON(child->state == __TASK_TRACED);
255 ret = -ESRCH;
256 }
257 }
258
259 return ret;
260}
261
262static int ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
263{
264 if (mode & PTRACE_MODE_NOAUDIT)
265 return has_ns_capability_noaudit(current, ns, CAP_SYS_PTRACE);
266 else
267 return has_ns_capability(current, ns, CAP_SYS_PTRACE);
268}
269
270/* Returns 0 on success, -errno on denial. */
271static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
272{
273 const struct cred *cred = current_cred(), *tcred;
274 struct mm_struct *mm;
275 kuid_t caller_uid;
276 kgid_t caller_gid;
277
278 if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
279 WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
280 return -EPERM;
281 }
282
283 /* May we inspect the given task?
284 * This check is used both for attaching with ptrace
285 * and for allowing access to sensitive information in /proc.
286 *
287 * ptrace_attach denies several cases that /proc allows
288 * because setting up the necessary parent/child relationship
289 * or halting the specified task is impossible.
290 */
291
292 /* Don't let security modules deny introspection */
293 if (same_thread_group(task, current))
294 return 0;
295 rcu_read_lock();
296 if (mode & PTRACE_MODE_FSCREDS) {
297 caller_uid = cred->fsuid;
298 caller_gid = cred->fsgid;
299 } else {
300 /*
301 * Using the euid would make more sense here, but something
302 * in userland might rely on the old behavior, and this
303 * shouldn't be a security problem since
304 * PTRACE_MODE_REALCREDS implies that the caller explicitly
305 * used a syscall that requests access to another process
306 * (and not a filesystem syscall to procfs).
307 */
308 caller_uid = cred->uid;
309 caller_gid = cred->gid;
310 }
311 tcred = __task_cred(task);
312 if (uid_eq(caller_uid, tcred->euid) &&
313 uid_eq(caller_uid, tcred->suid) &&
314 uid_eq(caller_uid, tcred->uid) &&
315 gid_eq(caller_gid, tcred->egid) &&
316 gid_eq(caller_gid, tcred->sgid) &&
317 gid_eq(caller_gid, tcred->gid))
318 goto ok;
319 if (ptrace_has_cap(tcred->user_ns, mode))
320 goto ok;
321 rcu_read_unlock();
322 return -EPERM;
323ok:
324 rcu_read_unlock();
325 mm = task->mm;
326 if (mm &&
327 ((get_dumpable(mm) != SUID_DUMP_USER) &&
328 !ptrace_has_cap(mm->user_ns, mode)))
329 return -EPERM;
330
331 return security_ptrace_access_check(task, mode);
332}
333
334bool ptrace_may_access(struct task_struct *task, unsigned int mode)
335{
336 int err;
337 task_lock(task);
338 err = __ptrace_may_access(task, mode);
339 task_unlock(task);
340 return !err;
341}
342
343static int ptrace_attach(struct task_struct *task, long request,
344 unsigned long addr,
345 unsigned long flags)
346{
347 bool seize = (request == PTRACE_SEIZE);
348 int retval;
349
350 retval = -EIO;
351 if (seize) {
352 if (addr != 0)
353 goto out;
354 if (flags & ~(unsigned long)PTRACE_O_MASK)
355 goto out;
356 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
357 } else {
358 flags = PT_PTRACED;
359 }
360
361 audit_ptrace(task);
362
363 retval = -EPERM;
364 if (unlikely(task->flags & PF_KTHREAD))
365 goto out;
366 if (same_thread_group(task, current))
367 goto out;
368
369 /*
370 * Protect exec's credential calculations against our interference;
371 * SUID, SGID and LSM creds get determined differently
372 * under ptrace.
373 */
374 retval = -ERESTARTNOINTR;
375 if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
376 goto out;
377
378 task_lock(task);
379 retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
380 task_unlock(task);
381 if (retval)
382 goto unlock_creds;
383
384 write_lock_irq(&tasklist_lock);
385 retval = -EPERM;
386 if (unlikely(task->exit_state))
387 goto unlock_tasklist;
388 if (task->ptrace)
389 goto unlock_tasklist;
390
391 if (seize)
392 flags |= PT_SEIZED;
393 task->ptrace = flags;
394
395 ptrace_link(task, current);
396
397 /* SEIZE doesn't trap tracee on attach */
398 if (!seize)
399 send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
400
401 spin_lock(&task->sighand->siglock);
402
403 /*
404 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
405 * TRAPPING, and kick it so that it transits to TRACED. TRAPPING
406 * will be cleared if the child completes the transition or any
407 * event which clears the group stop states happens. We'll wait
408 * for the transition to complete before returning from this
409 * function.
410 *
411 * This hides STOPPED -> RUNNING -> TRACED transition from the
412 * attaching thread but a different thread in the same group can
413 * still observe the transient RUNNING state. IOW, if another
414 * thread's WNOHANG wait(2) on the stopped tracee races against
415 * ATTACH, the wait(2) may fail due to the transient RUNNING.
416 *
417 * The following task_is_stopped() test is safe as both transitions
418 * in and out of STOPPED are protected by siglock.
419 */
420 if (task_is_stopped(task) &&
421 task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
422 signal_wake_up_state(task, __TASK_STOPPED);
423
424 spin_unlock(&task->sighand->siglock);
425
426 retval = 0;
427unlock_tasklist:
428 write_unlock_irq(&tasklist_lock);
429unlock_creds:
430 mutex_unlock(&task->signal->cred_guard_mutex);
431out:
432 if (!retval) {
433 /*
434 * We do not bother to change retval or clear JOBCTL_TRAPPING
435 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
436 * not return to user-mode, it will exit and clear this bit in
437 * __ptrace_unlink() if it wasn't already cleared by the tracee;
438 * and until then nobody can ptrace this task.
439 */
440 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
441 proc_ptrace_connector(task, PTRACE_ATTACH);
442 }
443
444 return retval;
445}
446
447/**
448 * ptrace_traceme -- helper for PTRACE_TRACEME
449 *
450 * Performs checks and sets PT_PTRACED.
451 * Should be used by all ptrace implementations for PTRACE_TRACEME.
452 */
453static int ptrace_traceme(void)
454{
455 int ret = -EPERM;
456
457 write_lock_irq(&tasklist_lock);
458 /* Are we already being traced? */
459 if (!current->ptrace) {
460 ret = security_ptrace_traceme(current->parent);
461 /*
462 * Check PF_EXITING to ensure ->real_parent has not passed
463 * exit_ptrace(). Otherwise we don't report the error but
464 * pretend ->real_parent untraces us right after return.
465 */
466 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
467 current->ptrace = PT_PTRACED;
468 ptrace_link(current, current->real_parent);
469 }
470 }
471 write_unlock_irq(&tasklist_lock);
472
473 return ret;
474}
475
476/*
477 * Called with irqs disabled, returns true if childs should reap themselves.
478 */
479static int ignoring_children(struct sighand_struct *sigh)
480{
481 int ret;
482 spin_lock(&sigh->siglock);
483 ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
484 (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
485 spin_unlock(&sigh->siglock);
486 return ret;
487}
488
489/*
490 * Called with tasklist_lock held for writing.
491 * Unlink a traced task, and clean it up if it was a traced zombie.
492 * Return true if it needs to be reaped with release_task().
493 * (We can't call release_task() here because we already hold tasklist_lock.)
494 *
495 * If it's a zombie, our attachedness prevented normal parent notification
496 * or self-reaping. Do notification now if it would have happened earlier.
497 * If it should reap itself, return true.
498 *
499 * If it's our own child, there is no notification to do. But if our normal
500 * children self-reap, then this child was prevented by ptrace and we must
501 * reap it now, in that case we must also wake up sub-threads sleeping in
502 * do_wait().
503 */
504static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
505{
506 bool dead;
507
508 __ptrace_unlink(p);
509
510 if (p->exit_state != EXIT_ZOMBIE)
511 return false;
512
513 dead = !thread_group_leader(p);
514
515 if (!dead && thread_group_empty(p)) {
516 if (!same_thread_group(p->real_parent, tracer))
517 dead = do_notify_parent(p, p->exit_signal);
518 else if (ignoring_children(tracer->sighand)) {
519 __wake_up_parent(p, tracer);
520 dead = true;
521 }
522 }
523 /* Mark it as in the process of being reaped. */
524 if (dead)
525 p->exit_state = EXIT_DEAD;
526 return dead;
527}
528
529static int ptrace_detach(struct task_struct *child, unsigned int data)
530{
531 if (!valid_signal(data))
532 return -EIO;
533
534 /* Architecture-specific hardware disable .. */
535 ptrace_disable(child);
536
537 write_lock_irq(&tasklist_lock);
538 /*
539 * We rely on ptrace_freeze_traced(). It can't be killed and
540 * untraced by another thread, it can't be a zombie.
541 */
542 WARN_ON(!child->ptrace || child->exit_state);
543 /*
544 * tasklist_lock avoids the race with wait_task_stopped(), see
545 * the comment in ptrace_resume().
546 */
547 child->exit_code = data;
548 __ptrace_detach(current, child);
549 write_unlock_irq(&tasklist_lock);
550
551 proc_ptrace_connector(child, PTRACE_DETACH);
552
553 return 0;
554}
555
556/*
557 * Detach all tasks we were using ptrace on. Called with tasklist held
558 * for writing.
559 */
560void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
561{
562 struct task_struct *p, *n;
563
564 list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
565 if (unlikely(p->ptrace & PT_EXITKILL))
566 send_sig_info(SIGKILL, SEND_SIG_FORCED, p);
567
568 if (__ptrace_detach(tracer, p))
569 list_add(&p->ptrace_entry, dead);
570 }
571}
572
573int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
574{
575 int copied = 0;
576
577 while (len > 0) {
578 char buf[128];
579 int this_len, retval;
580
581 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
582 retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
583
584 if (!retval) {
585 if (copied)
586 break;
587 return -EIO;
588 }
589 if (copy_to_user(dst, buf, retval))
590 return -EFAULT;
591 copied += retval;
592 src += retval;
593 dst += retval;
594 len -= retval;
595 }
596 return copied;
597}
598
599int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
600{
601 int copied = 0;
602
603 while (len > 0) {
604 char buf[128];
605 int this_len, retval;
606
607 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
608 if (copy_from_user(buf, src, this_len))
609 return -EFAULT;
610 retval = ptrace_access_vm(tsk, dst, buf, this_len,
611 FOLL_FORCE | FOLL_WRITE);
612 if (!retval) {
613 if (copied)
614 break;
615 return -EIO;
616 }
617 copied += retval;
618 src += retval;
619 dst += retval;
620 len -= retval;
621 }
622 return copied;
623}
624
625static int ptrace_setoptions(struct task_struct *child, unsigned long data)
626{
627 unsigned flags;
628
629 if (data & ~(unsigned long)PTRACE_O_MASK)
630 return -EINVAL;
631
632 if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
633 if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
634 !IS_ENABLED(CONFIG_SECCOMP))
635 return -EINVAL;
636
637 if (!capable(CAP_SYS_ADMIN))
638 return -EPERM;
639
640 if (seccomp_mode(¤t->seccomp) != SECCOMP_MODE_DISABLED ||
641 current->ptrace & PT_SUSPEND_SECCOMP)
642 return -EPERM;
643 }
644
645 /* Avoid intermediate state when all opts are cleared */
646 flags = child->ptrace;
647 flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
648 flags |= (data << PT_OPT_FLAG_SHIFT);
649 child->ptrace = flags;
650
651 return 0;
652}
653
654static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
655{
656 unsigned long flags;
657 int error = -ESRCH;
658
659 if (lock_task_sighand(child, &flags)) {
660 error = -EINVAL;
661 if (likely(child->last_siginfo != NULL)) {
662 copy_siginfo(info, child->last_siginfo);
663 error = 0;
664 }
665 unlock_task_sighand(child, &flags);
666 }
667 return error;
668}
669
670static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
671{
672 unsigned long flags;
673 int error = -ESRCH;
674
675 if (lock_task_sighand(child, &flags)) {
676 error = -EINVAL;
677 if (likely(child->last_siginfo != NULL)) {
678 copy_siginfo(child->last_siginfo, info);
679 error = 0;
680 }
681 unlock_task_sighand(child, &flags);
682 }
683 return error;
684}
685
686static int ptrace_peek_siginfo(struct task_struct *child,
687 unsigned long addr,
688 unsigned long data)
689{
690 struct ptrace_peeksiginfo_args arg;
691 struct sigpending *pending;
692 struct sigqueue *q;
693 int ret, i;
694
695 ret = copy_from_user(&arg, (void __user *) addr,
696 sizeof(struct ptrace_peeksiginfo_args));
697 if (ret)
698 return -EFAULT;
699
700 if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
701 return -EINVAL; /* unknown flags */
702
703 if (arg.nr < 0)
704 return -EINVAL;
705
706 if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
707 pending = &child->signal->shared_pending;
708 else
709 pending = &child->pending;
710
711 for (i = 0; i < arg.nr; ) {
712 siginfo_t info;
713 s32 off = arg.off + i;
714
715 spin_lock_irq(&child->sighand->siglock);
716 list_for_each_entry(q, &pending->list, list) {
717 if (!off--) {
718 copy_siginfo(&info, &q->info);
719 break;
720 }
721 }
722 spin_unlock_irq(&child->sighand->siglock);
723
724 if (off >= 0) /* beyond the end of the list */
725 break;
726
727#ifdef CONFIG_COMPAT
728 if (unlikely(in_compat_syscall())) {
729 compat_siginfo_t __user *uinfo = compat_ptr(data);
730
731 if (copy_siginfo_to_user32(uinfo, &info)) {
732 ret = -EFAULT;
733 break;
734 }
735
736 } else
737#endif
738 {
739 siginfo_t __user *uinfo = (siginfo_t __user *) data;
740
741 if (copy_siginfo_to_user(uinfo, &info)) {
742 ret = -EFAULT;
743 break;
744 }
745 }
746
747 data += sizeof(siginfo_t);
748 i++;
749
750 if (signal_pending(current))
751 break;
752
753 cond_resched();
754 }
755
756 if (i > 0)
757 return i;
758
759 return ret;
760}
761
762#ifdef PTRACE_SINGLESTEP
763#define is_singlestep(request) ((request) == PTRACE_SINGLESTEP)
764#else
765#define is_singlestep(request) 0
766#endif
767
768#ifdef PTRACE_SINGLEBLOCK
769#define is_singleblock(request) ((request) == PTRACE_SINGLEBLOCK)
770#else
771#define is_singleblock(request) 0
772#endif
773
774#ifdef PTRACE_SYSEMU
775#define is_sysemu_singlestep(request) ((request) == PTRACE_SYSEMU_SINGLESTEP)
776#else
777#define is_sysemu_singlestep(request) 0
778#endif
779
780static int ptrace_resume(struct task_struct *child, long request,
781 unsigned long data)
782{
783 bool need_siglock;
784
785 if (!valid_signal(data))
786 return -EIO;
787
788 if (request == PTRACE_SYSCALL)
789 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
790 else
791 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
792
793#ifdef TIF_SYSCALL_EMU
794 if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
795 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
796 else
797 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
798#endif
799
800 if (is_singleblock(request)) {
801 if (unlikely(!arch_has_block_step()))
802 return -EIO;
803 user_enable_block_step(child);
804 } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
805 if (unlikely(!arch_has_single_step()))
806 return -EIO;
807 user_enable_single_step(child);
808 } else {
809 user_disable_single_step(child);
810 }
811
812 /*
813 * Change ->exit_code and ->state under siglock to avoid the race
814 * with wait_task_stopped() in between; a non-zero ->exit_code will
815 * wrongly look like another report from tracee.
816 *
817 * Note that we need siglock even if ->exit_code == data and/or this
818 * status was not reported yet, the new status must not be cleared by
819 * wait_task_stopped() after resume.
820 *
821 * If data == 0 we do not care if wait_task_stopped() reports the old
822 * status and clears the code too; this can't race with the tracee, it
823 * takes siglock after resume.
824 */
825 need_siglock = data && !thread_group_empty(current);
826 if (need_siglock)
827 spin_lock_irq(&child->sighand->siglock);
828 child->exit_code = data;
829 wake_up_state(child, __TASK_TRACED);
830 if (need_siglock)
831 spin_unlock_irq(&child->sighand->siglock);
832
833 return 0;
834}
835
836#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
837
838static const struct user_regset *
839find_regset(const struct user_regset_view *view, unsigned int type)
840{
841 const struct user_regset *regset;
842 int n;
843
844 for (n = 0; n < view->n; ++n) {
845 regset = view->regsets + n;
846 if (regset->core_note_type == type)
847 return regset;
848 }
849
850 return NULL;
851}
852
853static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
854 struct iovec *kiov)
855{
856 const struct user_regset_view *view = task_user_regset_view(task);
857 const struct user_regset *regset = find_regset(view, type);
858 int regset_no;
859
860 if (!regset || (kiov->iov_len % regset->size) != 0)
861 return -EINVAL;
862
863 regset_no = regset - view->regsets;
864 kiov->iov_len = min(kiov->iov_len,
865 (__kernel_size_t) (regset->n * regset->size));
866
867 if (req == PTRACE_GETREGSET)
868 return copy_regset_to_user(task, view, regset_no, 0,
869 kiov->iov_len, kiov->iov_base);
870 else
871 return copy_regset_from_user(task, view, regset_no, 0,
872 kiov->iov_len, kiov->iov_base);
873}
874
875/*
876 * This is declared in linux/regset.h and defined in machine-dependent
877 * code. We put the export here, near the primary machine-neutral use,
878 * to ensure no machine forgets it.
879 */
880EXPORT_SYMBOL_GPL(task_user_regset_view);
881#endif
882
883int ptrace_request(struct task_struct *child, long request,
884 unsigned long addr, unsigned long data)
885{
886 bool seized = child->ptrace & PT_SEIZED;
887 int ret = -EIO;
888 siginfo_t siginfo, *si;
889 void __user *datavp = (void __user *) data;
890 unsigned long __user *datalp = datavp;
891 unsigned long flags;
892
893 switch (request) {
894 case PTRACE_PEEKTEXT:
895 case PTRACE_PEEKDATA:
896 return generic_ptrace_peekdata(child, addr, data);
897 case PTRACE_POKETEXT:
898 case PTRACE_POKEDATA:
899 return generic_ptrace_pokedata(child, addr, data);
900
901#ifdef PTRACE_OLDSETOPTIONS
902 case PTRACE_OLDSETOPTIONS:
903#endif
904 case PTRACE_SETOPTIONS:
905 ret = ptrace_setoptions(child, data);
906 break;
907 case PTRACE_GETEVENTMSG:
908 ret = put_user(child->ptrace_message, datalp);
909 break;
910
911 case PTRACE_PEEKSIGINFO:
912 ret = ptrace_peek_siginfo(child, addr, data);
913 break;
914
915 case PTRACE_GETSIGINFO:
916 ret = ptrace_getsiginfo(child, &siginfo);
917 if (!ret)
918 ret = copy_siginfo_to_user(datavp, &siginfo);
919 break;
920
921 case PTRACE_SETSIGINFO:
922 if (copy_from_user(&siginfo, datavp, sizeof siginfo))
923 ret = -EFAULT;
924 else
925 ret = ptrace_setsiginfo(child, &siginfo);
926 break;
927
928 case PTRACE_GETSIGMASK:
929 if (addr != sizeof(sigset_t)) {
930 ret = -EINVAL;
931 break;
932 }
933
934 if (copy_to_user(datavp, &child->blocked, sizeof(sigset_t)))
935 ret = -EFAULT;
936 else
937 ret = 0;
938
939 break;
940
941 case PTRACE_SETSIGMASK: {
942 sigset_t new_set;
943
944 if (addr != sizeof(sigset_t)) {
945 ret = -EINVAL;
946 break;
947 }
948
949 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
950 ret = -EFAULT;
951 break;
952 }
953
954 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
955
956 /*
957 * Every thread does recalc_sigpending() after resume, so
958 * retarget_shared_pending() and recalc_sigpending() are not
959 * called here.
960 */
961 spin_lock_irq(&child->sighand->siglock);
962 child->blocked = new_set;
963 spin_unlock_irq(&child->sighand->siglock);
964
965 ret = 0;
966 break;
967 }
968
969 case PTRACE_INTERRUPT:
970 /*
971 * Stop tracee without any side-effect on signal or job
972 * control. At least one trap is guaranteed to happen
973 * after this request. If @child is already trapped, the
974 * current trap is not disturbed and another trap will
975 * happen after the current trap is ended with PTRACE_CONT.
976 *
977 * The actual trap might not be PTRACE_EVENT_STOP trap but
978 * the pending condition is cleared regardless.
979 */
980 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
981 break;
982
983 /*
984 * INTERRUPT doesn't disturb existing trap sans one
985 * exception. If ptracer issued LISTEN for the current
986 * STOP, this INTERRUPT should clear LISTEN and re-trap
987 * tracee into STOP.
988 */
989 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
990 ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
991
992 unlock_task_sighand(child, &flags);
993 ret = 0;
994 break;
995
996 case PTRACE_LISTEN:
997 /*
998 * Listen for events. Tracee must be in STOP. It's not
999 * resumed per-se but is not considered to be in TRACED by
1000 * wait(2) or ptrace(2). If an async event (e.g. group
1001 * stop state change) happens, tracee will enter STOP trap
1002 * again. Alternatively, ptracer can issue INTERRUPT to
1003 * finish listening and re-trap tracee into STOP.
1004 */
1005 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1006 break;
1007
1008 si = child->last_siginfo;
1009 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1010 child->jobctl |= JOBCTL_LISTENING;
1011 /*
1012 * If NOTIFY is set, it means event happened between
1013 * start of this trap and now. Trigger re-trap.
1014 */
1015 if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1016 ptrace_signal_wake_up(child, true);
1017 ret = 0;
1018 }
1019 unlock_task_sighand(child, &flags);
1020 break;
1021
1022 case PTRACE_DETACH: /* detach a process that was attached. */
1023 ret = ptrace_detach(child, data);
1024 break;
1025
1026#ifdef CONFIG_BINFMT_ELF_FDPIC
1027 case PTRACE_GETFDPIC: {
1028 struct mm_struct *mm = get_task_mm(child);
1029 unsigned long tmp = 0;
1030
1031 ret = -ESRCH;
1032 if (!mm)
1033 break;
1034
1035 switch (addr) {
1036 case PTRACE_GETFDPIC_EXEC:
1037 tmp = mm->context.exec_fdpic_loadmap;
1038 break;
1039 case PTRACE_GETFDPIC_INTERP:
1040 tmp = mm->context.interp_fdpic_loadmap;
1041 break;
1042 default:
1043 break;
1044 }
1045 mmput(mm);
1046
1047 ret = put_user(tmp, datalp);
1048 break;
1049 }
1050#endif
1051
1052#ifdef PTRACE_SINGLESTEP
1053 case PTRACE_SINGLESTEP:
1054#endif
1055#ifdef PTRACE_SINGLEBLOCK
1056 case PTRACE_SINGLEBLOCK:
1057#endif
1058#ifdef PTRACE_SYSEMU
1059 case PTRACE_SYSEMU:
1060 case PTRACE_SYSEMU_SINGLESTEP:
1061#endif
1062 case PTRACE_SYSCALL:
1063 case PTRACE_CONT:
1064 return ptrace_resume(child, request, data);
1065
1066 case PTRACE_KILL:
1067 if (child->exit_state) /* already dead */
1068 return 0;
1069 return ptrace_resume(child, request, SIGKILL);
1070
1071#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1072 case PTRACE_GETREGSET:
1073 case PTRACE_SETREGSET: {
1074 struct iovec kiov;
1075 struct iovec __user *uiov = datavp;
1076
1077 if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1078 return -EFAULT;
1079
1080 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1081 __get_user(kiov.iov_len, &uiov->iov_len))
1082 return -EFAULT;
1083
1084 ret = ptrace_regset(child, request, addr, &kiov);
1085 if (!ret)
1086 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1087 break;
1088 }
1089#endif
1090
1091 case PTRACE_SECCOMP_GET_FILTER:
1092 ret = seccomp_get_filter(child, addr, datavp);
1093 break;
1094
1095 case PTRACE_SECCOMP_GET_METADATA:
1096 ret = seccomp_get_metadata(child, addr, datavp);
1097 break;
1098
1099 default:
1100 break;
1101 }
1102
1103 return ret;
1104}
1105
1106#ifndef arch_ptrace_attach
1107#define arch_ptrace_attach(child) do { } while (0)
1108#endif
1109
1110SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1111 unsigned long, data)
1112{
1113 struct task_struct *child;
1114 long ret;
1115
1116 if (request == PTRACE_TRACEME) {
1117 ret = ptrace_traceme();
1118 if (!ret)
1119 arch_ptrace_attach(current);
1120 goto out;
1121 }
1122
1123 child = find_get_task_by_vpid(pid);
1124 if (!child) {
1125 ret = -ESRCH;
1126 goto out;
1127 }
1128
1129 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1130 ret = ptrace_attach(child, request, addr, data);
1131 /*
1132 * Some architectures need to do book-keeping after
1133 * a ptrace attach.
1134 */
1135 if (!ret)
1136 arch_ptrace_attach(child);
1137 goto out_put_task_struct;
1138 }
1139
1140 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1141 request == PTRACE_INTERRUPT);
1142 if (ret < 0)
1143 goto out_put_task_struct;
1144
1145 ret = arch_ptrace(child, request, addr, data);
1146 if (ret || request != PTRACE_DETACH)
1147 ptrace_unfreeze_traced(child);
1148
1149 out_put_task_struct:
1150 put_task_struct(child);
1151 out:
1152 return ret;
1153}
1154
1155int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1156 unsigned long data)
1157{
1158 unsigned long tmp;
1159 int copied;
1160
1161 copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1162 if (copied != sizeof(tmp))
1163 return -EIO;
1164 return put_user(tmp, (unsigned long __user *)data);
1165}
1166
1167int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1168 unsigned long data)
1169{
1170 int copied;
1171
1172 copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1173 FOLL_FORCE | FOLL_WRITE);
1174 return (copied == sizeof(data)) ? 0 : -EIO;
1175}
1176
1177#if defined CONFIG_COMPAT
1178
1179int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1180 compat_ulong_t addr, compat_ulong_t data)
1181{
1182 compat_ulong_t __user *datap = compat_ptr(data);
1183 compat_ulong_t word;
1184 siginfo_t siginfo;
1185 int ret;
1186
1187 switch (request) {
1188 case PTRACE_PEEKTEXT:
1189 case PTRACE_PEEKDATA:
1190 ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1191 FOLL_FORCE);
1192 if (ret != sizeof(word))
1193 ret = -EIO;
1194 else
1195 ret = put_user(word, datap);
1196 break;
1197
1198 case PTRACE_POKETEXT:
1199 case PTRACE_POKEDATA:
1200 ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1201 FOLL_FORCE | FOLL_WRITE);
1202 ret = (ret != sizeof(data) ? -EIO : 0);
1203 break;
1204
1205 case PTRACE_GETEVENTMSG:
1206 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1207 break;
1208
1209 case PTRACE_GETSIGINFO:
1210 ret = ptrace_getsiginfo(child, &siginfo);
1211 if (!ret)
1212 ret = copy_siginfo_to_user32(
1213 (struct compat_siginfo __user *) datap,
1214 &siginfo);
1215 break;
1216
1217 case PTRACE_SETSIGINFO:
1218 if (copy_siginfo_from_user32(
1219 &siginfo, (struct compat_siginfo __user *) datap))
1220 ret = -EFAULT;
1221 else
1222 ret = ptrace_setsiginfo(child, &siginfo);
1223 break;
1224#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1225 case PTRACE_GETREGSET:
1226 case PTRACE_SETREGSET:
1227 {
1228 struct iovec kiov;
1229 struct compat_iovec __user *uiov =
1230 (struct compat_iovec __user *) datap;
1231 compat_uptr_t ptr;
1232 compat_size_t len;
1233
1234 if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1235 return -EFAULT;
1236
1237 if (__get_user(ptr, &uiov->iov_base) ||
1238 __get_user(len, &uiov->iov_len))
1239 return -EFAULT;
1240
1241 kiov.iov_base = compat_ptr(ptr);
1242 kiov.iov_len = len;
1243
1244 ret = ptrace_regset(child, request, addr, &kiov);
1245 if (!ret)
1246 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1247 break;
1248 }
1249#endif
1250
1251 default:
1252 ret = ptrace_request(child, request, addr, data);
1253 }
1254
1255 return ret;
1256}
1257
1258COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1259 compat_long_t, addr, compat_long_t, 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 /*
1278 * Some architectures need to do book-keeping after
1279 * a ptrace attach.
1280 */
1281 if (!ret)
1282 arch_ptrace_attach(child);
1283 goto out_put_task_struct;
1284 }
1285
1286 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1287 request == PTRACE_INTERRUPT);
1288 if (!ret) {
1289 ret = compat_arch_ptrace(child, request, addr, data);
1290 if (ret || request != PTRACE_DETACH)
1291 ptrace_unfreeze_traced(child);
1292 }
1293
1294 out_put_task_struct:
1295 put_task_struct(child);
1296 out:
1297 return ret;
1298}
1299#endif /* CONFIG_COMPAT */