kernel/fork.c: simplify copy_mm()
[platform/kernel/linux-starfive.git] / kernel / ptrace.c
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
36 #include <asm/syscall.h>        /* for syscall_get_* */
37
38 /*
39  * Access another process' address space via ptrace.
40  * Source/target buffer must be kernel space,
41  * Do not walk the page table directly, use get_user_pages
42  */
43 int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
44                      void *buf, int len, unsigned int gup_flags)
45 {
46         struct mm_struct *mm;
47         int ret;
48
49         mm = get_task_mm(tsk);
50         if (!mm)
51                 return 0;
52
53         if (!tsk->ptrace ||
54             (current != tsk->parent) ||
55             ((get_dumpable(mm) != SUID_DUMP_USER) &&
56              !ptracer_capable(tsk, mm->user_ns))) {
57                 mmput(mm);
58                 return 0;
59         }
60
61         ret = __access_remote_vm(mm, addr, buf, len, gup_flags);
62         mmput(mm);
63
64         return ret;
65 }
66
67
68 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent,
69                    const struct cred *ptracer_cred)
70 {
71         BUG_ON(!list_empty(&child->ptrace_entry));
72         list_add(&child->ptrace_entry, &new_parent->ptraced);
73         child->parent = new_parent;
74         child->ptracer_cred = get_cred(ptracer_cred);
75 }
76
77 /*
78  * ptrace a task: make the debugger its new parent and
79  * move it to the ptrace list.
80  *
81  * Must be called with the tasklist lock write-held.
82  */
83 static void ptrace_link(struct task_struct *child, struct task_struct *new_parent)
84 {
85         __ptrace_link(child, new_parent, current_cred());
86 }
87
88 /**
89  * __ptrace_unlink - unlink ptracee and restore its execution state
90  * @child: ptracee to be unlinked
91  *
92  * Remove @child from the ptrace list, move it back to the original parent,
93  * and restore the execution state so that it conforms to the group stop
94  * state.
95  *
96  * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
97  * exiting.  For PTRACE_DETACH, unless the ptracee has been killed between
98  * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
99  * If the ptracer is exiting, the ptracee can be in any state.
100  *
101  * After detach, the ptracee should be in a state which conforms to the
102  * group stop.  If the group is stopped or in the process of stopping, the
103  * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
104  * up from TASK_TRACED.
105  *
106  * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
107  * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
108  * to but in the opposite direction of what happens while attaching to a
109  * stopped task.  However, in this direction, the intermediate RUNNING
110  * state is not hidden even from the current ptracer and if it immediately
111  * re-attaches and performs a WNOHANG wait(2), it may fail.
112  *
113  * CONTEXT:
114  * write_lock_irq(tasklist_lock)
115  */
116 void __ptrace_unlink(struct task_struct *child)
117 {
118         const struct cred *old_cred;
119         BUG_ON(!child->ptrace);
120
121         clear_task_syscall_work(child, SYSCALL_TRACE);
122 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
123         clear_task_syscall_work(child, SYSCALL_EMU);
124 #endif
125
126         child->parent = child->real_parent;
127         list_del_init(&child->ptrace_entry);
128         old_cred = child->ptracer_cred;
129         child->ptracer_cred = NULL;
130         put_cred(old_cred);
131
132         spin_lock(&child->sighand->siglock);
133         child->ptrace = 0;
134         /*
135          * Clear all pending traps and TRAPPING.  TRAPPING should be
136          * cleared regardless of JOBCTL_STOP_PENDING.  Do it explicitly.
137          */
138         task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
139         task_clear_jobctl_trapping(child);
140
141         /*
142          * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
143          * @child isn't dead.
144          */
145         if (!(child->flags & PF_EXITING) &&
146             (child->signal->flags & SIGNAL_STOP_STOPPED ||
147              child->signal->group_stop_count)) {
148                 child->jobctl |= JOBCTL_STOP_PENDING;
149
150                 /*
151                  * This is only possible if this thread was cloned by the
152                  * traced task running in the stopped group, set the signal
153                  * for the future reports.
154                  * FIXME: we should change ptrace_init_task() to handle this
155                  * case.
156                  */
157                 if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
158                         child->jobctl |= SIGSTOP;
159         }
160
161         /*
162          * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
163          * @child in the butt.  Note that @resume should be used iff @child
164          * is in TASK_TRACED; otherwise, we might unduly disrupt
165          * TASK_KILLABLE sleeps.
166          */
167         if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
168                 ptrace_signal_wake_up(child, true);
169
170         spin_unlock(&child->sighand->siglock);
171 }
172
173 /* Ensure that nothing can wake it up, even SIGKILL */
174 static bool ptrace_freeze_traced(struct task_struct *task)
175 {
176         bool ret = false;
177
178         /* Lockless, nobody but us can set this flag */
179         if (task->jobctl & JOBCTL_LISTENING)
180                 return ret;
181
182         spin_lock_irq(&task->sighand->siglock);
183         if (task_is_traced(task) && !__fatal_signal_pending(task)) {
184                 task->state = __TASK_TRACED;
185                 ret = true;
186         }
187         spin_unlock_irq(&task->sighand->siglock);
188
189         return ret;
190 }
191
192 static void ptrace_unfreeze_traced(struct task_struct *task)
193 {
194         if (task->state != __TASK_TRACED)
195                 return;
196
197         WARN_ON(!task->ptrace || task->parent != current);
198
199         /*
200          * PTRACE_LISTEN can allow ptrace_trap_notify to wake us up remotely.
201          * Recheck state under the lock to close this race.
202          */
203         spin_lock_irq(&task->sighand->siglock);
204         if (task->state == __TASK_TRACED) {
205                 if (__fatal_signal_pending(task))
206                         wake_up_state(task, __TASK_TRACED);
207                 else
208                         task->state = TASK_TRACED;
209         }
210         spin_unlock_irq(&task->sighand->siglock);
211 }
212
213 /**
214  * ptrace_check_attach - check whether ptracee is ready for ptrace operation
215  * @child: ptracee to check for
216  * @ignore_state: don't check whether @child is currently %TASK_TRACED
217  *
218  * Check whether @child is being ptraced by %current and ready for further
219  * ptrace operations.  If @ignore_state is %false, @child also should be in
220  * %TASK_TRACED state and on return the child is guaranteed to be traced
221  * and not executing.  If @ignore_state is %true, @child can be in any
222  * state.
223  *
224  * CONTEXT:
225  * Grabs and releases tasklist_lock and @child->sighand->siglock.
226  *
227  * RETURNS:
228  * 0 on success, -ESRCH if %child is not ready.
229  */
230 static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
231 {
232         int ret = -ESRCH;
233
234         /*
235          * We take the read lock around doing both checks to close a
236          * possible race where someone else was tracing our child and
237          * detached between these two checks.  After this locked check,
238          * we are sure that this is our traced child and that can only
239          * be changed by us so it's not changing right after this.
240          */
241         read_lock(&tasklist_lock);
242         if (child->ptrace && child->parent == current) {
243                 WARN_ON(child->state == __TASK_TRACED);
244                 /*
245                  * child->sighand can't be NULL, release_task()
246                  * does ptrace_unlink() before __exit_signal().
247                  */
248                 if (ignore_state || ptrace_freeze_traced(child))
249                         ret = 0;
250         }
251         read_unlock(&tasklist_lock);
252
253         if (!ret && !ignore_state) {
254                 if (!wait_task_inactive(child, __TASK_TRACED)) {
255                         /*
256                          * This can only happen if may_ptrace_stop() fails and
257                          * ptrace_stop() changes ->state back to TASK_RUNNING,
258                          * so we should not worry about leaking __TASK_TRACED.
259                          */
260                         WARN_ON(child->state == __TASK_TRACED);
261                         ret = -ESRCH;
262                 }
263         }
264
265         return ret;
266 }
267
268 static 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. */
276 static 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;
328 ok:
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
349 bool 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
358 static 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;
442 unlock_tasklist:
443         write_unlock_irq(&tasklist_lock);
444 unlock_creds:
445         mutex_unlock(&task->signal->cred_guard_mutex);
446 out:
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  */
468 static 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  */
494 static 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  */
519 static 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
544 static 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  */
575 void 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
588 int 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
614 int 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
640 static 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(&current->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
669 static 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
685 static 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
701 static 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 CONFIG_RSEQ
784 static long ptrace_get_rseq_configuration(struct task_struct *task,
785                                           unsigned long size, void __user *data)
786 {
787         struct ptrace_rseq_configuration conf = {
788                 .rseq_abi_pointer = (u64)(uintptr_t)task->rseq,
789                 .rseq_abi_size = sizeof(*task->rseq),
790                 .signature = task->rseq_sig,
791                 .flags = 0,
792         };
793
794         size = min_t(unsigned long, size, sizeof(conf));
795         if (copy_to_user(data, &conf, size))
796                 return -EFAULT;
797         return sizeof(conf);
798 }
799 #endif
800
801 #ifdef PTRACE_SINGLESTEP
802 #define is_singlestep(request)          ((request) == PTRACE_SINGLESTEP)
803 #else
804 #define is_singlestep(request)          0
805 #endif
806
807 #ifdef PTRACE_SINGLEBLOCK
808 #define is_singleblock(request)         ((request) == PTRACE_SINGLEBLOCK)
809 #else
810 #define is_singleblock(request)         0
811 #endif
812
813 #ifdef PTRACE_SYSEMU
814 #define is_sysemu_singlestep(request)   ((request) == PTRACE_SYSEMU_SINGLESTEP)
815 #else
816 #define is_sysemu_singlestep(request)   0
817 #endif
818
819 static int ptrace_resume(struct task_struct *child, long request,
820                          unsigned long data)
821 {
822         bool need_siglock;
823
824         if (!valid_signal(data))
825                 return -EIO;
826
827         if (request == PTRACE_SYSCALL)
828                 set_task_syscall_work(child, SYSCALL_TRACE);
829         else
830                 clear_task_syscall_work(child, SYSCALL_TRACE);
831
832 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
833         if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
834                 set_task_syscall_work(child, SYSCALL_EMU);
835         else
836                 clear_task_syscall_work(child, SYSCALL_EMU);
837 #endif
838
839         if (is_singleblock(request)) {
840                 if (unlikely(!arch_has_block_step()))
841                         return -EIO;
842                 user_enable_block_step(child);
843         } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
844                 if (unlikely(!arch_has_single_step()))
845                         return -EIO;
846                 user_enable_single_step(child);
847         } else {
848                 user_disable_single_step(child);
849         }
850
851         /*
852          * Change ->exit_code and ->state under siglock to avoid the race
853          * with wait_task_stopped() in between; a non-zero ->exit_code will
854          * wrongly look like another report from tracee.
855          *
856          * Note that we need siglock even if ->exit_code == data and/or this
857          * status was not reported yet, the new status must not be cleared by
858          * wait_task_stopped() after resume.
859          *
860          * If data == 0 we do not care if wait_task_stopped() reports the old
861          * status and clears the code too; this can't race with the tracee, it
862          * takes siglock after resume.
863          */
864         need_siglock = data && !thread_group_empty(current);
865         if (need_siglock)
866                 spin_lock_irq(&child->sighand->siglock);
867         child->exit_code = data;
868         wake_up_state(child, __TASK_TRACED);
869         if (need_siglock)
870                 spin_unlock_irq(&child->sighand->siglock);
871
872         return 0;
873 }
874
875 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
876
877 static const struct user_regset *
878 find_regset(const struct user_regset_view *view, unsigned int type)
879 {
880         const struct user_regset *regset;
881         int n;
882
883         for (n = 0; n < view->n; ++n) {
884                 regset = view->regsets + n;
885                 if (regset->core_note_type == type)
886                         return regset;
887         }
888
889         return NULL;
890 }
891
892 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
893                          struct iovec *kiov)
894 {
895         const struct user_regset_view *view = task_user_regset_view(task);
896         const struct user_regset *regset = find_regset(view, type);
897         int regset_no;
898
899         if (!regset || (kiov->iov_len % regset->size) != 0)
900                 return -EINVAL;
901
902         regset_no = regset - view->regsets;
903         kiov->iov_len = min(kiov->iov_len,
904                             (__kernel_size_t) (regset->n * regset->size));
905
906         if (req == PTRACE_GETREGSET)
907                 return copy_regset_to_user(task, view, regset_no, 0,
908                                            kiov->iov_len, kiov->iov_base);
909         else
910                 return copy_regset_from_user(task, view, regset_no, 0,
911                                              kiov->iov_len, kiov->iov_base);
912 }
913
914 /*
915  * This is declared in linux/regset.h and defined in machine-dependent
916  * code.  We put the export here, near the primary machine-neutral use,
917  * to ensure no machine forgets it.
918  */
919 EXPORT_SYMBOL_GPL(task_user_regset_view);
920
921 static unsigned long
922 ptrace_get_syscall_info_entry(struct task_struct *child, struct pt_regs *regs,
923                               struct ptrace_syscall_info *info)
924 {
925         unsigned long args[ARRAY_SIZE(info->entry.args)];
926         int i;
927
928         info->op = PTRACE_SYSCALL_INFO_ENTRY;
929         info->entry.nr = syscall_get_nr(child, regs);
930         syscall_get_arguments(child, regs, args);
931         for (i = 0; i < ARRAY_SIZE(args); i++)
932                 info->entry.args[i] = args[i];
933
934         /* args is the last field in struct ptrace_syscall_info.entry */
935         return offsetofend(struct ptrace_syscall_info, entry.args);
936 }
937
938 static unsigned long
939 ptrace_get_syscall_info_seccomp(struct task_struct *child, struct pt_regs *regs,
940                                 struct ptrace_syscall_info *info)
941 {
942         /*
943          * As struct ptrace_syscall_info.entry is currently a subset
944          * of struct ptrace_syscall_info.seccomp, it makes sense to
945          * initialize that subset using ptrace_get_syscall_info_entry().
946          * This can be reconsidered in the future if these structures
947          * diverge significantly enough.
948          */
949         ptrace_get_syscall_info_entry(child, regs, info);
950         info->op = PTRACE_SYSCALL_INFO_SECCOMP;
951         info->seccomp.ret_data = child->ptrace_message;
952
953         /* ret_data is the last field in struct ptrace_syscall_info.seccomp */
954         return offsetofend(struct ptrace_syscall_info, seccomp.ret_data);
955 }
956
957 static unsigned long
958 ptrace_get_syscall_info_exit(struct task_struct *child, struct pt_regs *regs,
959                              struct ptrace_syscall_info *info)
960 {
961         info->op = PTRACE_SYSCALL_INFO_EXIT;
962         info->exit.rval = syscall_get_error(child, regs);
963         info->exit.is_error = !!info->exit.rval;
964         if (!info->exit.is_error)
965                 info->exit.rval = syscall_get_return_value(child, regs);
966
967         /* is_error is the last field in struct ptrace_syscall_info.exit */
968         return offsetofend(struct ptrace_syscall_info, exit.is_error);
969 }
970
971 static int
972 ptrace_get_syscall_info(struct task_struct *child, unsigned long user_size,
973                         void __user *datavp)
974 {
975         struct pt_regs *regs = task_pt_regs(child);
976         struct ptrace_syscall_info info = {
977                 .op = PTRACE_SYSCALL_INFO_NONE,
978                 .arch = syscall_get_arch(child),
979                 .instruction_pointer = instruction_pointer(regs),
980                 .stack_pointer = user_stack_pointer(regs),
981         };
982         unsigned long actual_size = offsetof(struct ptrace_syscall_info, entry);
983         unsigned long write_size;
984
985         /*
986          * This does not need lock_task_sighand() to access
987          * child->last_siginfo because ptrace_freeze_traced()
988          * called earlier by ptrace_check_attach() ensures that
989          * the tracee cannot go away and clear its last_siginfo.
990          */
991         switch (child->last_siginfo ? child->last_siginfo->si_code : 0) {
992         case SIGTRAP | 0x80:
993                 switch (child->ptrace_message) {
994                 case PTRACE_EVENTMSG_SYSCALL_ENTRY:
995                         actual_size = ptrace_get_syscall_info_entry(child, regs,
996                                                                     &info);
997                         break;
998                 case PTRACE_EVENTMSG_SYSCALL_EXIT:
999                         actual_size = ptrace_get_syscall_info_exit(child, regs,
1000                                                                    &info);
1001                         break;
1002                 }
1003                 break;
1004         case SIGTRAP | (PTRACE_EVENT_SECCOMP << 8):
1005                 actual_size = ptrace_get_syscall_info_seccomp(child, regs,
1006                                                               &info);
1007                 break;
1008         }
1009
1010         write_size = min(actual_size, user_size);
1011         return copy_to_user(datavp, &info, write_size) ? -EFAULT : actual_size;
1012 }
1013 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
1014
1015 int ptrace_request(struct task_struct *child, long request,
1016                    unsigned long addr, unsigned long data)
1017 {
1018         bool seized = child->ptrace & PT_SEIZED;
1019         int ret = -EIO;
1020         kernel_siginfo_t siginfo, *si;
1021         void __user *datavp = (void __user *) data;
1022         unsigned long __user *datalp = datavp;
1023         unsigned long flags;
1024
1025         switch (request) {
1026         case PTRACE_PEEKTEXT:
1027         case PTRACE_PEEKDATA:
1028                 return generic_ptrace_peekdata(child, addr, data);
1029         case PTRACE_POKETEXT:
1030         case PTRACE_POKEDATA:
1031                 return generic_ptrace_pokedata(child, addr, data);
1032
1033 #ifdef PTRACE_OLDSETOPTIONS
1034         case PTRACE_OLDSETOPTIONS:
1035 #endif
1036         case PTRACE_SETOPTIONS:
1037                 ret = ptrace_setoptions(child, data);
1038                 break;
1039         case PTRACE_GETEVENTMSG:
1040                 ret = put_user(child->ptrace_message, datalp);
1041                 break;
1042
1043         case PTRACE_PEEKSIGINFO:
1044                 ret = ptrace_peek_siginfo(child, addr, data);
1045                 break;
1046
1047         case PTRACE_GETSIGINFO:
1048                 ret = ptrace_getsiginfo(child, &siginfo);
1049                 if (!ret)
1050                         ret = copy_siginfo_to_user(datavp, &siginfo);
1051                 break;
1052
1053         case PTRACE_SETSIGINFO:
1054                 ret = copy_siginfo_from_user(&siginfo, datavp);
1055                 if (!ret)
1056                         ret = ptrace_setsiginfo(child, &siginfo);
1057                 break;
1058
1059         case PTRACE_GETSIGMASK: {
1060                 sigset_t *mask;
1061
1062                 if (addr != sizeof(sigset_t)) {
1063                         ret = -EINVAL;
1064                         break;
1065                 }
1066
1067                 if (test_tsk_restore_sigmask(child))
1068                         mask = &child->saved_sigmask;
1069                 else
1070                         mask = &child->blocked;
1071
1072                 if (copy_to_user(datavp, mask, sizeof(sigset_t)))
1073                         ret = -EFAULT;
1074                 else
1075                         ret = 0;
1076
1077                 break;
1078         }
1079
1080         case PTRACE_SETSIGMASK: {
1081                 sigset_t new_set;
1082
1083                 if (addr != sizeof(sigset_t)) {
1084                         ret = -EINVAL;
1085                         break;
1086                 }
1087
1088                 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
1089                         ret = -EFAULT;
1090                         break;
1091                 }
1092
1093                 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
1094
1095                 /*
1096                  * Every thread does recalc_sigpending() after resume, so
1097                  * retarget_shared_pending() and recalc_sigpending() are not
1098                  * called here.
1099                  */
1100                 spin_lock_irq(&child->sighand->siglock);
1101                 child->blocked = new_set;
1102                 spin_unlock_irq(&child->sighand->siglock);
1103
1104                 clear_tsk_restore_sigmask(child);
1105
1106                 ret = 0;
1107                 break;
1108         }
1109
1110         case PTRACE_INTERRUPT:
1111                 /*
1112                  * Stop tracee without any side-effect on signal or job
1113                  * control.  At least one trap is guaranteed to happen
1114                  * after this request.  If @child is already trapped, the
1115                  * current trap is not disturbed and another trap will
1116                  * happen after the current trap is ended with PTRACE_CONT.
1117                  *
1118                  * The actual trap might not be PTRACE_EVENT_STOP trap but
1119                  * the pending condition is cleared regardless.
1120                  */
1121                 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1122                         break;
1123
1124                 /*
1125                  * INTERRUPT doesn't disturb existing trap sans one
1126                  * exception.  If ptracer issued LISTEN for the current
1127                  * STOP, this INTERRUPT should clear LISTEN and re-trap
1128                  * tracee into STOP.
1129                  */
1130                 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
1131                         ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1132
1133                 unlock_task_sighand(child, &flags);
1134                 ret = 0;
1135                 break;
1136
1137         case PTRACE_LISTEN:
1138                 /*
1139                  * Listen for events.  Tracee must be in STOP.  It's not
1140                  * resumed per-se but is not considered to be in TRACED by
1141                  * wait(2) or ptrace(2).  If an async event (e.g. group
1142                  * stop state change) happens, tracee will enter STOP trap
1143                  * again.  Alternatively, ptracer can issue INTERRUPT to
1144                  * finish listening and re-trap tracee into STOP.
1145                  */
1146                 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1147                         break;
1148
1149                 si = child->last_siginfo;
1150                 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1151                         child->jobctl |= JOBCTL_LISTENING;
1152                         /*
1153                          * If NOTIFY is set, it means event happened between
1154                          * start of this trap and now.  Trigger re-trap.
1155                          */
1156                         if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1157                                 ptrace_signal_wake_up(child, true);
1158                         ret = 0;
1159                 }
1160                 unlock_task_sighand(child, &flags);
1161                 break;
1162
1163         case PTRACE_DETACH:      /* detach a process that was attached. */
1164                 ret = ptrace_detach(child, data);
1165                 break;
1166
1167 #ifdef CONFIG_BINFMT_ELF_FDPIC
1168         case PTRACE_GETFDPIC: {
1169                 struct mm_struct *mm = get_task_mm(child);
1170                 unsigned long tmp = 0;
1171
1172                 ret = -ESRCH;
1173                 if (!mm)
1174                         break;
1175
1176                 switch (addr) {
1177                 case PTRACE_GETFDPIC_EXEC:
1178                         tmp = mm->context.exec_fdpic_loadmap;
1179                         break;
1180                 case PTRACE_GETFDPIC_INTERP:
1181                         tmp = mm->context.interp_fdpic_loadmap;
1182                         break;
1183                 default:
1184                         break;
1185                 }
1186                 mmput(mm);
1187
1188                 ret = put_user(tmp, datalp);
1189                 break;
1190         }
1191 #endif
1192
1193 #ifdef PTRACE_SINGLESTEP
1194         case PTRACE_SINGLESTEP:
1195 #endif
1196 #ifdef PTRACE_SINGLEBLOCK
1197         case PTRACE_SINGLEBLOCK:
1198 #endif
1199 #ifdef PTRACE_SYSEMU
1200         case PTRACE_SYSEMU:
1201         case PTRACE_SYSEMU_SINGLESTEP:
1202 #endif
1203         case PTRACE_SYSCALL:
1204         case PTRACE_CONT:
1205                 return ptrace_resume(child, request, data);
1206
1207         case PTRACE_KILL:
1208                 if (child->exit_state)  /* already dead */
1209                         return 0;
1210                 return ptrace_resume(child, request, SIGKILL);
1211
1212 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1213         case PTRACE_GETREGSET:
1214         case PTRACE_SETREGSET: {
1215                 struct iovec kiov;
1216                 struct iovec __user *uiov = datavp;
1217
1218                 if (!access_ok(uiov, sizeof(*uiov)))
1219                         return -EFAULT;
1220
1221                 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1222                     __get_user(kiov.iov_len, &uiov->iov_len))
1223                         return -EFAULT;
1224
1225                 ret = ptrace_regset(child, request, addr, &kiov);
1226                 if (!ret)
1227                         ret = __put_user(kiov.iov_len, &uiov->iov_len);
1228                 break;
1229         }
1230
1231         case PTRACE_GET_SYSCALL_INFO:
1232                 ret = ptrace_get_syscall_info(child, addr, datavp);
1233                 break;
1234 #endif
1235
1236         case PTRACE_SECCOMP_GET_FILTER:
1237                 ret = seccomp_get_filter(child, addr, datavp);
1238                 break;
1239
1240         case PTRACE_SECCOMP_GET_METADATA:
1241                 ret = seccomp_get_metadata(child, addr, datavp);
1242                 break;
1243
1244 #ifdef CONFIG_RSEQ
1245         case PTRACE_GET_RSEQ_CONFIGURATION:
1246                 ret = ptrace_get_rseq_configuration(child, addr, datavp);
1247                 break;
1248 #endif
1249
1250         default:
1251                 break;
1252         }
1253
1254         return ret;
1255 }
1256
1257 #ifndef arch_ptrace_attach
1258 #define arch_ptrace_attach(child)       do { } while (0)
1259 #endif
1260
1261 SYSCALL_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                 if (!ret)
1270                         arch_ptrace_attach(current);
1271                 goto out;
1272         }
1273
1274         child = find_get_task_by_vpid(pid);
1275         if (!child) {
1276                 ret = -ESRCH;
1277                 goto out;
1278         }
1279
1280         if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1281                 ret = ptrace_attach(child, request, addr, data);
1282                 /*
1283                  * Some architectures need to do book-keeping after
1284                  * a ptrace attach.
1285                  */
1286                 if (!ret)
1287                         arch_ptrace_attach(child);
1288                 goto out_put_task_struct;
1289         }
1290
1291         ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1292                                   request == PTRACE_INTERRUPT);
1293         if (ret < 0)
1294                 goto out_put_task_struct;
1295
1296         ret = arch_ptrace(child, request, addr, data);
1297         if (ret || request != PTRACE_DETACH)
1298                 ptrace_unfreeze_traced(child);
1299
1300  out_put_task_struct:
1301         put_task_struct(child);
1302  out:
1303         return ret;
1304 }
1305
1306 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1307                             unsigned long data)
1308 {
1309         unsigned long tmp;
1310         int copied;
1311
1312         copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1313         if (copied != sizeof(tmp))
1314                 return -EIO;
1315         return put_user(tmp, (unsigned long __user *)data);
1316 }
1317
1318 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1319                             unsigned long data)
1320 {
1321         int copied;
1322
1323         copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1324                         FOLL_FORCE | FOLL_WRITE);
1325         return (copied == sizeof(data)) ? 0 : -EIO;
1326 }
1327
1328 #if defined CONFIG_COMPAT
1329
1330 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1331                           compat_ulong_t addr, compat_ulong_t data)
1332 {
1333         compat_ulong_t __user *datap = compat_ptr(data);
1334         compat_ulong_t word;
1335         kernel_siginfo_t siginfo;
1336         int ret;
1337
1338         switch (request) {
1339         case PTRACE_PEEKTEXT:
1340         case PTRACE_PEEKDATA:
1341                 ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1342                                 FOLL_FORCE);
1343                 if (ret != sizeof(word))
1344                         ret = -EIO;
1345                 else
1346                         ret = put_user(word, datap);
1347                 break;
1348
1349         case PTRACE_POKETEXT:
1350         case PTRACE_POKEDATA:
1351                 ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1352                                 FOLL_FORCE | FOLL_WRITE);
1353                 ret = (ret != sizeof(data) ? -EIO : 0);
1354                 break;
1355
1356         case PTRACE_GETEVENTMSG:
1357                 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1358                 break;
1359
1360         case PTRACE_GETSIGINFO:
1361                 ret = ptrace_getsiginfo(child, &siginfo);
1362                 if (!ret)
1363                         ret = copy_siginfo_to_user32(
1364                                 (struct compat_siginfo __user *) datap,
1365                                 &siginfo);
1366                 break;
1367
1368         case PTRACE_SETSIGINFO:
1369                 ret = copy_siginfo_from_user32(
1370                         &siginfo, (struct compat_siginfo __user *) datap);
1371                 if (!ret)
1372                         ret = ptrace_setsiginfo(child, &siginfo);
1373                 break;
1374 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1375         case PTRACE_GETREGSET:
1376         case PTRACE_SETREGSET:
1377         {
1378                 struct iovec kiov;
1379                 struct compat_iovec __user *uiov =
1380                         (struct compat_iovec __user *) datap;
1381                 compat_uptr_t ptr;
1382                 compat_size_t len;
1383
1384                 if (!access_ok(uiov, sizeof(*uiov)))
1385                         return -EFAULT;
1386
1387                 if (__get_user(ptr, &uiov->iov_base) ||
1388                     __get_user(len, &uiov->iov_len))
1389                         return -EFAULT;
1390
1391                 kiov.iov_base = compat_ptr(ptr);
1392                 kiov.iov_len = len;
1393
1394                 ret = ptrace_regset(child, request, addr, &kiov);
1395                 if (!ret)
1396                         ret = __put_user(kiov.iov_len, &uiov->iov_len);
1397                 break;
1398         }
1399 #endif
1400
1401         default:
1402                 ret = ptrace_request(child, request, addr, data);
1403         }
1404
1405         return ret;
1406 }
1407
1408 COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1409                        compat_long_t, addr, compat_long_t, data)
1410 {
1411         struct task_struct *child;
1412         long ret;
1413
1414         if (request == PTRACE_TRACEME) {
1415                 ret = ptrace_traceme();
1416                 goto out;
1417         }
1418
1419         child = find_get_task_by_vpid(pid);
1420         if (!child) {
1421                 ret = -ESRCH;
1422                 goto out;
1423         }
1424
1425         if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1426                 ret = ptrace_attach(child, request, addr, data);
1427                 /*
1428                  * Some architectures need to do book-keeping after
1429                  * a ptrace attach.
1430                  */
1431                 if (!ret)
1432                         arch_ptrace_attach(child);
1433                 goto out_put_task_struct;
1434         }
1435
1436         ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1437                                   request == PTRACE_INTERRUPT);
1438         if (!ret) {
1439                 ret = compat_arch_ptrace(child, request, addr, data);
1440                 if (ret || request != PTRACE_DETACH)
1441                         ptrace_unfreeze_traced(child);
1442         }
1443
1444  out_put_task_struct:
1445         put_task_struct(child);
1446  out:
1447         return ret;
1448 }
1449 #endif  /* CONFIG_COMPAT */