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