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