summaryrefslogtreecommitdiff
path: root/kernel/ptrace.c (plain)
blob: efba851ee0183f2e8b8c35c729271b009a3e5382
1/*
2 * linux/kernel/ptrace.c
3 *
4 * (C) Copyright 1999 Linus Torvalds
5 *
6 * Common interfaces for "ptrace()" which we do not want
7 * to continually duplicate across every architecture.
8 */
9
10#include <linux/capability.h>
11#include <linux/export.h>
12#include <linux/sched.h>
13#include <linux/errno.h>
14#include <linux/mm.h>
15#include <linux/highmem.h>
16#include <linux/pagemap.h>
17#include <linux/ptrace.h>
18#include <linux/security.h>
19#include <linux/signal.h>
20#include <linux/uio.h>
21#include <linux/audit.h>
22#include <linux/pid_namespace.h>
23#include <linux/syscalls.h>
24#include <linux/uaccess.h>
25#include <linux/regset.h>
26#include <linux/hw_breakpoint.h>
27#include <linux/cn_proc.h>
28#include <linux/compat.h>
29
30/*
31 * Access another process' address space via ptrace.
32 * Source/target buffer must be kernel space,
33 * Do not walk the page table directly, use get_user_pages
34 */
35int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
36 void *buf, int len, unsigned int gup_flags)
37{
38 struct mm_struct *mm;
39 int ret;
40
41 mm = get_task_mm(tsk);
42 if (!mm)
43 return 0;
44
45 if (!tsk->ptrace ||
46 (current != tsk->parent) ||
47 ((get_dumpable(mm) != SUID_DUMP_USER) &&
48 !ptracer_capable(tsk, mm->user_ns))) {
49 mmput(mm);
50 return 0;
51 }
52
53 ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
54 mmput(mm);
55
56 return ret;
57}
58
59
60void __ptrace_link(struct task_struct *child, struct task_struct *new_parent,
61 const struct cred *ptracer_cred)
62{
63 BUG_ON(!list_empty(&child->ptrace_entry));
64 list_add(&child->ptrace_entry, &new_parent->ptraced);
65 child->parent = new_parent;
66 child->ptracer_cred = get_cred(ptracer_cred);
67}
68
69/*
70 * ptrace a task: make the debugger its new parent and
71 * move it to the ptrace list.
72 *
73 * Must be called with the tasklist lock write-held.
74 */
75static void ptrace_link(struct task_struct *child, struct task_struct *new_parent)
76{
77 rcu_read_lock();
78 __ptrace_link(child, new_parent, __task_cred(new_parent));
79 rcu_read_unlock();
80}
81
82/**
83 * __ptrace_unlink - unlink ptracee and restore its execution state
84 * @child: ptracee to be unlinked
85 *
86 * Remove @child from the ptrace list, move it back to the original parent,
87 * and restore the execution state so that it conforms to the group stop
88 * state.
89 *
90 * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
91 * exiting. For PTRACE_DETACH, unless the ptracee has been killed between
92 * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
93 * If the ptracer is exiting, the ptracee can be in any state.
94 *
95 * After detach, the ptracee should be in a state which conforms to the
96 * group stop. If the group is stopped or in the process of stopping, the
97 * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
98 * up from TASK_TRACED.
99 *
100 * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
101 * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
102 * to but in the opposite direction of what happens while attaching to a
103 * stopped task. However, in this direction, the intermediate RUNNING
104 * state is not hidden even from the current ptracer and if it immediately
105 * re-attaches and performs a WNOHANG wait(2), it may fail.
106 *
107 * CONTEXT:
108 * write_lock_irq(tasklist_lock)
109 */
110void __ptrace_unlink(struct task_struct *child)
111{
112 const struct cred *old_cred;
113 BUG_ON(!child->ptrace);
114
115 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
116
117 child->parent = child->real_parent;
118 list_del_init(&child->ptrace_entry);
119 old_cred = child->ptracer_cred;
120 child->ptracer_cred = NULL;
121 put_cred(old_cred);
122
123 spin_lock(&child->sighand->siglock);
124 child->ptrace = 0;
125 /*
126 * Clear all pending traps and TRAPPING. TRAPPING should be
127 * cleared regardless of JOBCTL_STOP_PENDING. Do it explicitly.
128 */
129 task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
130 task_clear_jobctl_trapping(child);
131
132 /*
133 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
134 * @child isn't dead.
135 */
136 if (!(child->flags & PF_EXITING) &&
137 (child->signal->flags & SIGNAL_STOP_STOPPED ||
138 child->signal->group_stop_count)) {
139 child->jobctl |= JOBCTL_STOP_PENDING;
140
141 /*
142 * This is only possible if this thread was cloned by the
143 * traced task running in the stopped group, set the signal
144 * for the future reports.
145 * FIXME: we should change ptrace_init_task() to handle this
146 * case.
147 */
148 if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
149 child->jobctl |= SIGSTOP;
150 }
151
152 /*
153 * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
154 * @child in the butt. Note that @resume should be used iff @child
155 * is in TASK_TRACED; otherwise, we might unduly disrupt
156 * TASK_KILLABLE sleeps.
157 */
158 if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
159 ptrace_signal_wake_up(child, true);
160
161 spin_unlock(&child->sighand->siglock);
162}
163
164/* Ensure that nothing can wake it up, even SIGKILL */
165static bool ptrace_freeze_traced(struct task_struct *task)
166{
167 bool ret = false;
168
169 /* Lockless, nobody but us can set this flag */
170 if (task->jobctl & JOBCTL_LISTENING)
171 return ret;
172
173 spin_lock_irq(&task->sighand->siglock);
174 if (task_is_traced(task) && !__fatal_signal_pending(task)) {
175 task->state = __TASK_TRACED;
176 ret = true;
177 }
178 spin_unlock_irq(&task->sighand->siglock);
179
180 return ret;
181}
182
183static void ptrace_unfreeze_traced(struct task_struct *task)
184{
185 if (task->state != __TASK_TRACED)
186 return;
187
188 WARN_ON(!task->ptrace || task->parent != current);
189
190 /*
191 * PTRACE_LISTEN can allow ptrace_trap_notify to wake us up remotely.
192 * Recheck state under the lock to close this race.
193 */
194 spin_lock_irq(&task->sighand->siglock);
195 if (task->state == __TASK_TRACED) {
196 if (__fatal_signal_pending(task))
197 wake_up_state(task, __TASK_TRACED);
198 else
199 task->state = TASK_TRACED;
200 }
201 spin_unlock_irq(&task->sighand->siglock);
202}
203
204/**
205 * ptrace_check_attach - check whether ptracee is ready for ptrace operation
206 * @child: ptracee to check for
207 * @ignore_state: don't check whether @child is currently %TASK_TRACED
208 *
209 * Check whether @child is being ptraced by %current and ready for further
210 * ptrace operations. If @ignore_state is %false, @child also should be in
211 * %TASK_TRACED state and on return the child is guaranteed to be traced
212 * and not executing. If @ignore_state is %true, @child can be in any
213 * state.
214 *
215 * CONTEXT:
216 * Grabs and releases tasklist_lock and @child->sighand->siglock.
217 *
218 * RETURNS:
219 * 0 on success, -ESRCH if %child is not ready.
220 */
221static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
222{
223 int ret = -ESRCH;
224
225 /*
226 * We take the read lock around doing both checks to close a
227 * possible race where someone else was tracing our child and
228 * detached between these two checks. After this locked check,
229 * we are sure that this is our traced child and that can only
230 * be changed by us so it's not changing right after this.
231 */
232 read_lock(&tasklist_lock);
233 if (child->ptrace && child->parent == current) {
234 WARN_ON(child->state == __TASK_TRACED);
235 /*
236 * child->sighand can't be NULL, release_task()
237 * does ptrace_unlink() before __exit_signal().
238 */
239 if (ignore_state || ptrace_freeze_traced(child))
240 ret = 0;
241 }
242 read_unlock(&tasklist_lock);
243
244 if (!ret && !ignore_state) {
245 if (!wait_task_inactive(child, __TASK_TRACED)) {
246 /*
247 * This can only happen if may_ptrace_stop() fails and
248 * ptrace_stop() changes ->state back to TASK_RUNNING,
249 * so we should not worry about leaking __TASK_TRACED.
250 */
251 WARN_ON(child->state == __TASK_TRACED);
252 ret = -ESRCH;
253 }
254 }
255
256 return ret;
257}
258
259static int ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
260{
261 if (mode & PTRACE_MODE_SCHED)
262 return false;
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 if (mode & PTRACE_MODE_SCHED)
332 return 0;
333 return security_ptrace_access_check(task, mode);
334}
335
336bool ptrace_may_access_sched(struct task_struct *task, unsigned int mode)
337{
338 return __ptrace_may_access(task, mode | PTRACE_MODE_SCHED);
339}
340
341bool ptrace_may_access(struct task_struct *task, unsigned int mode)
342{
343 int err;
344 task_lock(task);
345 err = __ptrace_may_access(task, mode);
346 task_unlock(task);
347 return !err;
348}
349
350static int ptrace_attach(struct task_struct *task, long request,
351 unsigned long addr,
352 unsigned long flags)
353{
354 bool seize = (request == PTRACE_SEIZE);
355 int retval;
356
357 retval = -EIO;
358 if (seize) {
359 if (addr != 0)
360 goto out;
361 if (flags & ~(unsigned long)PTRACE_O_MASK)
362 goto out;
363 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
364 } else {
365 flags = PT_PTRACED;
366 }
367
368 audit_ptrace(task);
369
370 retval = -EPERM;
371 if (unlikely(task->flags & PF_KTHREAD))
372 goto out;
373 if (same_thread_group(task, current))
374 goto out;
375
376 /*
377 * Protect exec's credential calculations against our interference;
378 * SUID, SGID and LSM creds get determined differently
379 * under ptrace.
380 */
381 retval = -ERESTARTNOINTR;
382 if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
383 goto out;
384
385 task_lock(task);
386 retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
387 task_unlock(task);
388 if (retval)
389 goto unlock_creds;
390
391 write_lock_irq(&tasklist_lock);
392 retval = -EPERM;
393 if (unlikely(task->exit_state))
394 goto unlock_tasklist;
395 if (task->ptrace)
396 goto unlock_tasklist;
397
398 if (seize)
399 flags |= PT_SEIZED;
400 task->ptrace = flags;
401
402 ptrace_link(task, current);
403
404 /* SEIZE doesn't trap tracee on attach */
405 if (!seize)
406 send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
407
408 spin_lock(&task->sighand->siglock);
409
410 /*
411 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
412 * TRAPPING, and kick it so that it transits to TRACED. TRAPPING
413 * will be cleared if the child completes the transition or any
414 * event which clears the group stop states happens. We'll wait
415 * for the transition to complete before returning from this
416 * function.
417 *
418 * This hides STOPPED -> RUNNING -> TRACED transition from the
419 * attaching thread but a different thread in the same group can
420 * still observe the transient RUNNING state. IOW, if another
421 * thread's WNOHANG wait(2) on the stopped tracee races against
422 * ATTACH, the wait(2) may fail due to the transient RUNNING.
423 *
424 * The following task_is_stopped() test is safe as both transitions
425 * in and out of STOPPED are protected by siglock.
426 */
427 if (task_is_stopped(task) &&
428 task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
429 signal_wake_up_state(task, __TASK_STOPPED);
430
431 spin_unlock(&task->sighand->siglock);
432
433 retval = 0;
434unlock_tasklist:
435 write_unlock_irq(&tasklist_lock);
436unlock_creds:
437 mutex_unlock(&task->signal->cred_guard_mutex);
438out:
439 if (!retval) {
440 /*
441 * We do not bother to change retval or clear JOBCTL_TRAPPING
442 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
443 * not return to user-mode, it will exit and clear this bit in
444 * __ptrace_unlink() if it wasn't already cleared by the tracee;
445 * and until then nobody can ptrace this task.
446 */
447 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
448 proc_ptrace_connector(task, PTRACE_ATTACH);
449 }
450
451 return retval;
452}
453
454/**
455 * ptrace_traceme -- helper for PTRACE_TRACEME
456 *
457 * Performs checks and sets PT_PTRACED.
458 * Should be used by all ptrace implementations for PTRACE_TRACEME.
459 */
460static int ptrace_traceme(void)
461{
462 int ret = -EPERM;
463
464 write_lock_irq(&tasklist_lock);
465 /* Are we already being traced? */
466 if (!current->ptrace) {
467 ret = security_ptrace_traceme(current->parent);
468 /*
469 * Check PF_EXITING to ensure ->real_parent has not passed
470 * exit_ptrace(). Otherwise we don't report the error but
471 * pretend ->real_parent untraces us right after return.
472 */
473 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
474 current->ptrace = PT_PTRACED;
475 ptrace_link(current, current->real_parent);
476 }
477 }
478 write_unlock_irq(&tasklist_lock);
479
480 return ret;
481}
482
483/*
484 * Called with irqs disabled, returns true if childs should reap themselves.
485 */
486static int ignoring_children(struct sighand_struct *sigh)
487{
488 int ret;
489 spin_lock(&sigh->siglock);
490 ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
491 (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
492 spin_unlock(&sigh->siglock);
493 return ret;
494}
495
496/*
497 * Called with tasklist_lock held for writing.
498 * Unlink a traced task, and clean it up if it was a traced zombie.
499 * Return true if it needs to be reaped with release_task().
500 * (We can't call release_task() here because we already hold tasklist_lock.)
501 *
502 * If it's a zombie, our attachedness prevented normal parent notification
503 * or self-reaping. Do notification now if it would have happened earlier.
504 * If it should reap itself, return true.
505 *
506 * If it's our own child, there is no notification to do. But if our normal
507 * children self-reap, then this child was prevented by ptrace and we must
508 * reap it now, in that case we must also wake up sub-threads sleeping in
509 * do_wait().
510 */
511static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
512{
513 bool dead;
514
515 __ptrace_unlink(p);
516
517 if (p->exit_state != EXIT_ZOMBIE)
518 return false;
519
520 dead = !thread_group_leader(p);
521
522 if (!dead && thread_group_empty(p)) {
523 if (!same_thread_group(p->real_parent, tracer))
524 dead = do_notify_parent(p, p->exit_signal);
525 else if (ignoring_children(tracer->sighand)) {
526 __wake_up_parent(p, tracer);
527 dead = true;
528 }
529 }
530 /* Mark it as in the process of being reaped. */
531 if (dead)
532 p->exit_state = EXIT_DEAD;
533 return dead;
534}
535
536static int ptrace_detach(struct task_struct *child, unsigned int data)
537{
538 if (!valid_signal(data))
539 return -EIO;
540
541 /* Architecture-specific hardware disable .. */
542 ptrace_disable(child);
543
544 write_lock_irq(&tasklist_lock);
545 /*
546 * We rely on ptrace_freeze_traced(). It can't be killed and
547 * untraced by another thread, it can't be a zombie.
548 */
549 WARN_ON(!child->ptrace || child->exit_state);
550 /*
551 * tasklist_lock avoids the race with wait_task_stopped(), see
552 * the comment in ptrace_resume().
553 */
554 child->exit_code = data;
555 __ptrace_detach(current, child);
556 write_unlock_irq(&tasklist_lock);
557
558 proc_ptrace_connector(child, PTRACE_DETACH);
559
560 return 0;
561}
562
563/*
564 * Detach all tasks we were using ptrace on. Called with tasklist held
565 * for writing.
566 */
567void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
568{
569 struct task_struct *p, *n;
570
571 list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
572 if (unlikely(p->ptrace & PT_EXITKILL))
573 send_sig_info(SIGKILL, SEND_SIG_FORCED, p);
574
575 if (__ptrace_detach(tracer, p))
576 list_add(&p->ptrace_entry, dead);
577 }
578}
579
580int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
581{
582 int copied = 0;
583
584 while (len > 0) {
585 char buf[128];
586 int this_len, retval;
587
588 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
589 retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
590
591 if (!retval) {
592 if (copied)
593 break;
594 return -EIO;
595 }
596 if (copy_to_user(dst, buf, retval))
597 return -EFAULT;
598 copied += retval;
599 src += retval;
600 dst += retval;
601 len -= retval;
602 }
603 return copied;
604}
605
606int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
607{
608 int copied = 0;
609
610 while (len > 0) {
611 char buf[128];
612 int this_len, retval;
613
614 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
615 if (copy_from_user(buf, src, this_len))
616 return -EFAULT;
617 retval = ptrace_access_vm(tsk, dst, buf, this_len,
618 FOLL_FORCE | FOLL_WRITE);
619 if (!retval) {
620 if (copied)
621 break;
622 return -EIO;
623 }
624 copied += retval;
625 src += retval;
626 dst += retval;
627 len -= retval;
628 }
629 return copied;
630}
631
632static int ptrace_setoptions(struct task_struct *child, unsigned long data)
633{
634 unsigned flags;
635
636 if (data & ~(unsigned long)PTRACE_O_MASK)
637 return -EINVAL;
638
639 if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
640 if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
641 !IS_ENABLED(CONFIG_SECCOMP))
642 return -EINVAL;
643
644 if (!capable(CAP_SYS_ADMIN))
645 return -EPERM;
646
647 if (seccomp_mode(&current->seccomp) != SECCOMP_MODE_DISABLED ||
648 current->ptrace & PT_SUSPEND_SECCOMP)
649 return -EPERM;
650 }
651
652 /* Avoid intermediate state when all opts are cleared */
653 flags = child->ptrace;
654 flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
655 flags |= (data << PT_OPT_FLAG_SHIFT);
656 child->ptrace = flags;
657
658 return 0;
659}
660
661static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
662{
663 unsigned long flags;
664 int error = -ESRCH;
665
666 if (lock_task_sighand(child, &flags)) {
667 error = -EINVAL;
668 if (likely(child->last_siginfo != NULL)) {
669 *info = *child->last_siginfo;
670 error = 0;
671 }
672 unlock_task_sighand(child, &flags);
673 }
674 return error;
675}
676
677static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
678{
679 unsigned long flags;
680 int error = -ESRCH;
681
682 if (lock_task_sighand(child, &flags)) {
683 error = -EINVAL;
684 if (likely(child->last_siginfo != NULL)) {
685 *child->last_siginfo = *info;
686 error = 0;
687 }
688 unlock_task_sighand(child, &flags);
689 }
690 return error;
691}
692
693static int ptrace_peek_siginfo(struct task_struct *child,
694 unsigned long addr,
695 unsigned long data)
696{
697 struct ptrace_peeksiginfo_args arg;
698 struct sigpending *pending;
699 struct sigqueue *q;
700 int ret, i;
701
702 ret = copy_from_user(&arg, (void __user *) addr,
703 sizeof(struct ptrace_peeksiginfo_args));
704 if (ret)
705 return -EFAULT;
706
707 if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
708 return -EINVAL; /* unknown flags */
709
710 if (arg.nr < 0)
711 return -EINVAL;
712
713 if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
714 pending = &child->signal->shared_pending;
715 else
716 pending = &child->pending;
717
718 for (i = 0; i < arg.nr; ) {
719 siginfo_t info;
720 s32 off = arg.off + i;
721
722 spin_lock_irq(&child->sighand->siglock);
723 list_for_each_entry(q, &pending->list, list) {
724 if (!off--) {
725 copy_siginfo(&info, &q->info);
726 break;
727 }
728 }
729 spin_unlock_irq(&child->sighand->siglock);
730
731 if (off >= 0) /* beyond the end of the list */
732 break;
733
734#ifdef CONFIG_COMPAT
735 if (unlikely(in_compat_syscall())) {
736 compat_siginfo_t __user *uinfo = compat_ptr(data);
737
738 if (copy_siginfo_to_user32(uinfo, &info) ||
739 __put_user(info.si_code, &uinfo->si_code)) {
740 ret = -EFAULT;
741 break;
742 }
743
744 } else
745#endif
746 {
747 siginfo_t __user *uinfo = (siginfo_t __user *) data;
748
749 if (copy_siginfo_to_user(uinfo, &info) ||
750 __put_user(info.si_code, &uinfo->si_code)) {
751 ret = -EFAULT;
752 break;
753 }
754 }
755
756 data += sizeof(siginfo_t);
757 i++;
758
759 if (signal_pending(current))
760 break;
761
762 cond_resched();
763 }
764
765 if (i > 0)
766 return i;
767
768 return ret;
769}
770
771#ifdef PTRACE_SINGLESTEP
772#define is_singlestep(request) ((request) == PTRACE_SINGLESTEP)
773#else
774#define is_singlestep(request) 0
775#endif
776
777#ifdef PTRACE_SINGLEBLOCK
778#define is_singleblock(request) ((request) == PTRACE_SINGLEBLOCK)
779#else
780#define is_singleblock(request) 0
781#endif
782
783#ifdef PTRACE_SYSEMU
784#define is_sysemu_singlestep(request) ((request) == PTRACE_SYSEMU_SINGLESTEP)
785#else
786#define is_sysemu_singlestep(request) 0
787#endif
788
789static int ptrace_resume(struct task_struct *child, long request,
790 unsigned long data)
791{
792 bool need_siglock;
793
794 if (!valid_signal(data))
795 return -EIO;
796
797 if (request == PTRACE_SYSCALL)
798 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
799 else
800 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
801
802#ifdef TIF_SYSCALL_EMU
803 if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
804 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
805 else
806 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
807#endif
808
809 if (is_singleblock(request)) {
810 if (unlikely(!arch_has_block_step()))
811 return -EIO;
812 user_enable_block_step(child);
813 } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
814 if (unlikely(!arch_has_single_step()))
815 return -EIO;
816 user_enable_single_step(child);
817 } else {
818 user_disable_single_step(child);
819 }
820
821 /*
822 * Change ->exit_code and ->state under siglock to avoid the race
823 * with wait_task_stopped() in between; a non-zero ->exit_code will
824 * wrongly look like another report from tracee.
825 *
826 * Note that we need siglock even if ->exit_code == data and/or this
827 * status was not reported yet, the new status must not be cleared by
828 * wait_task_stopped() after resume.
829 *
830 * If data == 0 we do not care if wait_task_stopped() reports the old
831 * status and clears the code too; this can't race with the tracee, it
832 * takes siglock after resume.
833 */
834 need_siglock = data && !thread_group_empty(current);
835 if (need_siglock)
836 spin_lock_irq(&child->sighand->siglock);
837 child->exit_code = data;
838 wake_up_state(child, __TASK_TRACED);
839 if (need_siglock)
840 spin_unlock_irq(&child->sighand->siglock);
841
842 return 0;
843}
844
845#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
846
847static const struct user_regset *
848find_regset(const struct user_regset_view *view, unsigned int type)
849{
850 const struct user_regset *regset;
851 int n;
852
853 for (n = 0; n < view->n; ++n) {
854 regset = view->regsets + n;
855 if (regset->core_note_type == type)
856 return regset;
857 }
858
859 return NULL;
860}
861
862static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
863 struct iovec *kiov)
864{
865 const struct user_regset_view *view = task_user_regset_view(task);
866 const struct user_regset *regset = find_regset(view, type);
867 int regset_no;
868
869 if (!regset || (kiov->iov_len % regset->size) != 0)
870 return -EINVAL;
871
872 regset_no = regset - view->regsets;
873 kiov->iov_len = min(kiov->iov_len,
874 (__kernel_size_t) (regset->n * regset->size));
875
876 if (req == PTRACE_GETREGSET)
877 return copy_regset_to_user(task, view, regset_no, 0,
878 kiov->iov_len, kiov->iov_base);
879 else
880 return copy_regset_from_user(task, view, regset_no, 0,
881 kiov->iov_len, kiov->iov_base);
882}
883
884/*
885 * This is declared in linux/regset.h and defined in machine-dependent
886 * code. We put the export here, near the primary machine-neutral use,
887 * to ensure no machine forgets it.
888 */
889EXPORT_SYMBOL_GPL(task_user_regset_view);
890#endif
891
892int ptrace_request(struct task_struct *child, long request,
893 unsigned long addr, unsigned long data)
894{
895 bool seized = child->ptrace & PT_SEIZED;
896 int ret = -EIO;
897 siginfo_t siginfo, *si;
898 void __user *datavp = (void __user *) data;
899 unsigned long __user *datalp = datavp;
900 unsigned long flags;
901
902 switch (request) {
903 case PTRACE_PEEKTEXT:
904 case PTRACE_PEEKDATA:
905 return generic_ptrace_peekdata(child, addr, data);
906 case PTRACE_POKETEXT:
907 case PTRACE_POKEDATA:
908 return generic_ptrace_pokedata(child, addr, data);
909
910#ifdef PTRACE_OLDSETOPTIONS
911 case PTRACE_OLDSETOPTIONS:
912#endif
913 case PTRACE_SETOPTIONS:
914 ret = ptrace_setoptions(child, data);
915 break;
916 case PTRACE_GETEVENTMSG:
917 ret = put_user(child->ptrace_message, datalp);
918 break;
919
920 case PTRACE_PEEKSIGINFO:
921 ret = ptrace_peek_siginfo(child, addr, data);
922 break;
923
924 case PTRACE_GETSIGINFO:
925 ret = ptrace_getsiginfo(child, &siginfo);
926 if (!ret)
927 ret = copy_siginfo_to_user(datavp, &siginfo);
928 break;
929
930 case PTRACE_SETSIGINFO:
931 if (copy_from_user(&siginfo, datavp, sizeof siginfo))
932 ret = -EFAULT;
933 else
934 ret = ptrace_setsiginfo(child, &siginfo);
935 break;
936
937 case PTRACE_GETSIGMASK:
938 if (addr != sizeof(sigset_t)) {
939 ret = -EINVAL;
940 break;
941 }
942
943 if (copy_to_user(datavp, &child->blocked, sizeof(sigset_t)))
944 ret = -EFAULT;
945 else
946 ret = 0;
947
948 break;
949
950 case PTRACE_SETSIGMASK: {
951 sigset_t new_set;
952
953 if (addr != sizeof(sigset_t)) {
954 ret = -EINVAL;
955 break;
956 }
957
958 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
959 ret = -EFAULT;
960 break;
961 }
962
963 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
964
965 /*
966 * Every thread does recalc_sigpending() after resume, so
967 * retarget_shared_pending() and recalc_sigpending() are not
968 * called here.
969 */
970 spin_lock_irq(&child->sighand->siglock);
971 child->blocked = new_set;
972 spin_unlock_irq(&child->sighand->siglock);
973
974 ret = 0;
975 break;
976 }
977
978 case PTRACE_INTERRUPT:
979 /*
980 * Stop tracee without any side-effect on signal or job
981 * control. At least one trap is guaranteed to happen
982 * after this request. If @child is already trapped, the
983 * current trap is not disturbed and another trap will
984 * happen after the current trap is ended with PTRACE_CONT.
985 *
986 * The actual trap might not be PTRACE_EVENT_STOP trap but
987 * the pending condition is cleared regardless.
988 */
989 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
990 break;
991
992 /*
993 * INTERRUPT doesn't disturb existing trap sans one
994 * exception. If ptracer issued LISTEN for the current
995 * STOP, this INTERRUPT should clear LISTEN and re-trap
996 * tracee into STOP.
997 */
998 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
999 ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1000
1001 unlock_task_sighand(child, &flags);
1002 ret = 0;
1003 break;
1004
1005 case PTRACE_LISTEN:
1006 /*
1007 * Listen for events. Tracee must be in STOP. It's not
1008 * resumed per-se but is not considered to be in TRACED by
1009 * wait(2) or ptrace(2). If an async event (e.g. group
1010 * stop state change) happens, tracee will enter STOP trap
1011 * again. Alternatively, ptracer can issue INTERRUPT to
1012 * finish listening and re-trap tracee into STOP.
1013 */
1014 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1015 break;
1016
1017 si = child->last_siginfo;
1018 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1019 child->jobctl |= JOBCTL_LISTENING;
1020 /*
1021 * If NOTIFY is set, it means event happened between
1022 * start of this trap and now. Trigger re-trap.
1023 */
1024 if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1025 ptrace_signal_wake_up(child, true);
1026 ret = 0;
1027 }
1028 unlock_task_sighand(child, &flags);
1029 break;
1030
1031 case PTRACE_DETACH: /* detach a process that was attached. */
1032 ret = ptrace_detach(child, data);
1033 break;
1034
1035#ifdef CONFIG_BINFMT_ELF_FDPIC
1036 case PTRACE_GETFDPIC: {
1037 struct mm_struct *mm = get_task_mm(child);
1038 unsigned long tmp = 0;
1039
1040 ret = -ESRCH;
1041 if (!mm)
1042 break;
1043
1044 switch (addr) {
1045 case PTRACE_GETFDPIC_EXEC:
1046 tmp = mm->context.exec_fdpic_loadmap;
1047 break;
1048 case PTRACE_GETFDPIC_INTERP:
1049 tmp = mm->context.interp_fdpic_loadmap;
1050 break;
1051 default:
1052 break;
1053 }
1054 mmput(mm);
1055
1056 ret = put_user(tmp, datalp);
1057 break;
1058 }
1059#endif
1060
1061#ifdef PTRACE_SINGLESTEP
1062 case PTRACE_SINGLESTEP:
1063#endif
1064#ifdef PTRACE_SINGLEBLOCK
1065 case PTRACE_SINGLEBLOCK:
1066#endif
1067#ifdef PTRACE_SYSEMU
1068 case PTRACE_SYSEMU:
1069 case PTRACE_SYSEMU_SINGLESTEP:
1070#endif
1071 case PTRACE_SYSCALL:
1072 case PTRACE_CONT:
1073 return ptrace_resume(child, request, data);
1074
1075 case PTRACE_KILL:
1076 if (child->exit_state) /* already dead */
1077 return 0;
1078 return ptrace_resume(child, request, SIGKILL);
1079
1080#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1081 case PTRACE_GETREGSET:
1082 case PTRACE_SETREGSET: {
1083 struct iovec kiov;
1084 struct iovec __user *uiov = datavp;
1085
1086 if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1087 return -EFAULT;
1088
1089 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1090 __get_user(kiov.iov_len, &uiov->iov_len))
1091 return -EFAULT;
1092
1093 ret = ptrace_regset(child, request, addr, &kiov);
1094 if (!ret)
1095 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1096 break;
1097 }
1098#endif
1099
1100 case PTRACE_SECCOMP_GET_FILTER:
1101 ret = seccomp_get_filter(child, addr, datavp);
1102 break;
1103
1104 default:
1105 break;
1106 }
1107
1108 return ret;
1109}
1110
1111static struct task_struct *ptrace_get_task_struct(pid_t pid)
1112{
1113 struct task_struct *child;
1114
1115 rcu_read_lock();
1116 child = find_task_by_vpid(pid);
1117 if (child)
1118 get_task_struct(child);
1119 rcu_read_unlock();
1120
1121 if (!child)
1122 return ERR_PTR(-ESRCH);
1123 return child;
1124}
1125
1126#ifndef arch_ptrace_attach
1127#define arch_ptrace_attach(child) do { } while (0)
1128#endif
1129
1130SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1131 unsigned long, data)
1132{
1133 struct task_struct *child;
1134 long ret;
1135
1136 if (request == PTRACE_TRACEME) {
1137 ret = ptrace_traceme();
1138 if (!ret)
1139 arch_ptrace_attach(current);
1140 goto out;
1141 }
1142
1143 child = ptrace_get_task_struct(pid);
1144 if (IS_ERR(child)) {
1145 ret = PTR_ERR(child);
1146 goto out;
1147 }
1148
1149 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1150 ret = ptrace_attach(child, request, addr, data);
1151 /*
1152 * Some architectures need to do book-keeping after
1153 * a ptrace attach.
1154 */
1155 if (!ret)
1156 arch_ptrace_attach(child);
1157 goto out_put_task_struct;
1158 }
1159
1160 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1161 request == PTRACE_INTERRUPT);
1162 if (ret < 0)
1163 goto out_put_task_struct;
1164
1165 ret = arch_ptrace(child, request, addr, data);
1166 if (ret || request != PTRACE_DETACH)
1167 ptrace_unfreeze_traced(child);
1168
1169 out_put_task_struct:
1170 put_task_struct(child);
1171 out:
1172 return ret;
1173}
1174
1175int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1176 unsigned long data)
1177{
1178 unsigned long tmp;
1179 int copied;
1180
1181 copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1182 if (copied != sizeof(tmp))
1183 return -EIO;
1184 return put_user(tmp, (unsigned long __user *)data);
1185}
1186
1187int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1188 unsigned long data)
1189{
1190 int copied;
1191
1192 copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1193 FOLL_FORCE | FOLL_WRITE);
1194 return (copied == sizeof(data)) ? 0 : -EIO;
1195}
1196
1197#if defined CONFIG_COMPAT
1198
1199int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1200 compat_ulong_t addr, compat_ulong_t data)
1201{
1202 compat_ulong_t __user *datap = compat_ptr(data);
1203 compat_ulong_t word;
1204 siginfo_t siginfo;
1205 int ret;
1206
1207 switch (request) {
1208 case PTRACE_PEEKTEXT:
1209 case PTRACE_PEEKDATA:
1210 ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1211 FOLL_FORCE);
1212 if (ret != sizeof(word))
1213 ret = -EIO;
1214 else
1215 ret = put_user(word, datap);
1216 break;
1217
1218 case PTRACE_POKETEXT:
1219 case PTRACE_POKEDATA:
1220 ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1221 FOLL_FORCE | FOLL_WRITE);
1222 ret = (ret != sizeof(data) ? -EIO : 0);
1223 break;
1224
1225 case PTRACE_GETEVENTMSG:
1226 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1227 break;
1228
1229 case PTRACE_GETSIGINFO:
1230 ret = ptrace_getsiginfo(child, &siginfo);
1231 if (!ret)
1232 ret = copy_siginfo_to_user32(
1233 (struct compat_siginfo __user *) datap,
1234 &siginfo);
1235 break;
1236
1237 case PTRACE_SETSIGINFO:
1238 memset(&siginfo, 0, sizeof siginfo);
1239 if (copy_siginfo_from_user32(
1240 &siginfo, (struct compat_siginfo __user *) datap))
1241 ret = -EFAULT;
1242 else
1243 ret = ptrace_setsiginfo(child, &siginfo);
1244 break;
1245#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1246 case PTRACE_GETREGSET:
1247 case PTRACE_SETREGSET:
1248 {
1249 struct iovec kiov;
1250 struct compat_iovec __user *uiov =
1251 (struct compat_iovec __user *) datap;
1252 compat_uptr_t ptr;
1253 compat_size_t len;
1254
1255 if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1256 return -EFAULT;
1257
1258 if (__get_user(ptr, &uiov->iov_base) ||
1259 __get_user(len, &uiov->iov_len))
1260 return -EFAULT;
1261
1262 kiov.iov_base = compat_ptr(ptr);
1263 kiov.iov_len = len;
1264
1265 ret = ptrace_regset(child, request, addr, &kiov);
1266 if (!ret)
1267 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1268 break;
1269 }
1270#endif
1271
1272 default:
1273 ret = ptrace_request(child, request, addr, data);
1274 }
1275
1276 return ret;
1277}
1278
1279COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1280 compat_long_t, addr, compat_long_t, data)
1281{
1282 struct task_struct *child;
1283 long ret;
1284
1285 if (request == PTRACE_TRACEME) {
1286 ret = ptrace_traceme();
1287 goto out;
1288 }
1289
1290 child = ptrace_get_task_struct(pid);
1291 if (IS_ERR(child)) {
1292 ret = PTR_ERR(child);
1293 goto out;
1294 }
1295
1296 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1297 ret = ptrace_attach(child, request, addr, data);
1298 /*
1299 * Some architectures need to do book-keeping after
1300 * a ptrace attach.
1301 */
1302 if (!ret)
1303 arch_ptrace_attach(child);
1304 goto out_put_task_struct;
1305 }
1306
1307 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1308 request == PTRACE_INTERRUPT);
1309 if (!ret) {
1310 ret = compat_arch_ptrace(child, request, addr, data);
1311 if (ret || request != PTRACE_DETACH)
1312 ptrace_unfreeze_traced(child);
1313 }
1314
1315 out_put_task_struct:
1316 put_task_struct(child);
1317 out:
1318 return ret;
1319}
1320#endif /* CONFIG_COMPAT */
1321
generated by cgit at 2024-05-18 13:27:27 (GMT)