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 static bool looks_like_a_spurious_pid(struct task_struct *task)
 174 {
 175         if (task->exit_code != ((PTRACE_EVENT_EXEC << 8) | SIGTRAP))
 176                 return false;
 177
 178         if (task_pid_vnr(task) == task->ptrace_message)
 179                 return false;
 180         /*
 181          * The tracee changed its pid but the PTRACE_EVENT_EXEC event
 182          * was not wait()'ed, most probably debugger targets the old
 183          * leader which was destroyed in de_thread().
 184          */
 185         return true;
 186 }
 187
 188 /* Ensure that nothing can wake it up, even SIGKILL */
 189 static bool ptrace_freeze_traced(struct task_struct *task)
 190 {
 191         bool ret = false;
 192
 193         /* Lockless, nobody but us can set this flag */
 194         if (task->jobctl & JOBCTL_LISTENING)
 195                 return ret;
 196
 197         spin_lock_irq(&task->sighand->siglock);
 198         if (task_is_traced(task) && !looks_like_a_spurious_pid(task) &&
 199             !__fatal_signal_pending(task)) {
 200                 WRITE_ONCE(task->__state, __TASK_TRACED);
 201                 ret = true;
 202         }
 203         spin_unlock_irq(&task->sighand->siglock);
 204
 205         return ret;
 206 }
 207
 208 static void ptrace_unfreeze_traced(struct task_struct *task)
 209 {
 210         if (READ_ONCE(task->__state) != __TASK_TRACED)
 211                 return;
 212
 213         WARN_ON(!task->ptrace || task->parent != current);
 214
 215         /*
 216          * PTRACE_LISTEN can allow ptrace_trap_notify to wake us up remotely.
 217          * Recheck state under the lock to close this race.
 218          */
 219         spin_lock_irq(&task->sighand->siglock);
 220         if (READ_ONCE(task->__state) == __TASK_TRACED) {
 221                 if (__fatal_signal_pending(task))
 222                         wake_up_state(task, __TASK_TRACED);
 223                 else
 224                         WRITE_ONCE(task->__state, TASK_TRACED);
 225         }
 226         spin_unlock_irq(&task->sighand->siglock);
 227 }
 228
 229 /**
 230  * ptrace_check_attach - check whether ptracee is ready for ptrace operation
 231  * @child: ptracee to check for
 232  * @ignore_state: don't check whether @child is currently %TASK_TRACED
 233  *
 234  * Check whether @child is being ptraced by %current and ready for further
 235  * ptrace operations.  If @ignore_state is %false, @child also should be in
 236  * %TASK_TRACED state and on return the child is guaranteed to be traced
 237  * and not executing.  If @ignore_state is %true, @child can be in any
 238  * state.
 239  *
 240  * CONTEXT:
 241  * Grabs and releases tasklist_lock and @child->sighand->siglock.
 242  *
 243  * RETURNS:
 244  * 0 on success, -ESRCH if %child is not ready.
 245  */
 246 static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
 247 {
 248         int ret = -ESRCH;
 249
 250         /*
 251          * We take the read lock around doing both checks to close a
 252          * possible race where someone else was tracing our child and
 253          * detached between these two checks.  After this locked check,
 254          * we are sure that this is our traced child and that can only
 255          * be changed by us so it's not changing right after this.
 256          */
 257         read_lock(&tasklist_lock);
 258         if (child->ptrace && child->parent == current) {
 259                 WARN_ON(READ_ONCE(child->__state) == __TASK_TRACED);
 260                 /*
 261                  * child->sighand can't be NULL, release_task()
 262                  * does ptrace_unlink() before __exit_signal().
 263                  */
 264                 if (ignore_state || ptrace_freeze_traced(child))
 265                         ret = 0;
 266         }
 267         read_unlock(&tasklist_lock);
 268
 269         if (!ret && !ignore_state) {
 270                 if (!wait_task_inactive(child, __TASK_TRACED)) {
 271                         /*
 272                          * This can only happen if may_ptrace_stop() fails and
 273                          * ptrace_stop() changes ->state back to TASK_RUNNING,
 274                          * so we should not worry about leaking __TASK_TRACED.
 275                          */
 276                         WARN_ON(READ_ONCE(child->__state) == __TASK_TRACED);
 277                         ret = -ESRCH;
 278                 }
 279         }
 280
 281         return ret;
 282 }
 283
 284 static bool ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
 285 {
 286         if (mode & PTRACE_MODE_NOAUDIT)
 287                 return ns_capable_noaudit(ns, CAP_SYS_PTRACE);
 288         return ns_capable(ns, CAP_SYS_PTRACE);
 289 }
 290
 291 /* Returns 0 on success, -errno on denial. */
 292 static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
 293 {
 294         const struct cred *cred = current_cred(), *tcred;
 295         struct mm_struct *mm;
 296         kuid_t caller_uid;
 297         kgid_t caller_gid;
 298
 299         if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
 300                 WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
 301                 return -EPERM;
 302         }
 303
 304         /* May we inspect the given task?
 305          * This check is used both for attaching with ptrace
 306          * and for allowing access to sensitive information in /proc.
 307          *
 308          * ptrace_attach denies several cases that /proc allows
 309          * because setting up the necessary parent/child relationship
 310          * or halting the specified task is impossible.
 311          */
 312
 313         /* Don't let security modules deny introspection */
 314         if (same_thread_group(task, current))
 315                 return 0;
 316         rcu_read_lock();
 317         if (mode & PTRACE_MODE_FSCREDS) {
 318                 caller_uid = cred->fsuid;
 319                 caller_gid = cred->fsgid;
 320         } else {
 321                 /*
 322                  * Using the euid would make more sense here, but something
 323                  * in userland might rely on the old behavior, and this
 324                  * shouldn't be a security problem since
 325                  * PTRACE_MODE_REALCREDS implies that the caller explicitly
 326                  * used a syscall that requests access to another process
 327                  * (and not a filesystem syscall to procfs).
 328                  */
 329                 caller_uid = cred->uid;
 330                 caller_gid = cred->gid;
 331         }
 332         tcred = __task_cred(task);
 333         if (uid_eq(caller_uid, tcred->euid) &&
 334             uid_eq(caller_uid, tcred->suid) &&
 335             uid_eq(caller_uid, tcred->uid)  &&
 336             gid_eq(caller_gid, tcred->egid) &&
 337             gid_eq(caller_gid, tcred->sgid) &&
 338             gid_eq(caller_gid, tcred->gid))
 339                 goto ok;
 340         if (ptrace_has_cap(tcred->user_ns, mode))
 341                 goto ok;
 342         rcu_read_unlock();
 343         return -EPERM;
 344 ok:
 345         rcu_read_unlock();
 346         /*
 347          * If a task drops privileges and becomes nondumpable (through a syscall
 348          * like setresuid()) while we are trying to access it, we must ensure
 349          * that the dumpability is read after the credentials; otherwise,
 350          * we may be able to attach to a task that we shouldn't be able to
 351          * attach to (as if the task had dropped privileges without becoming
 352          * nondumpable).
 353          * Pairs with a write barrier in commit_creds().
 354          */
 355         smp_rmb();
 356         mm = task->mm;
 357         if (mm &&
 358             ((get_dumpable(mm) != SUID_DUMP_USER) &&
 359              !ptrace_has_cap(mm->user_ns, mode)))
 360             return -EPERM;
 361
 362         return security_ptrace_access_check(task, mode);
 363 }
 364
 365 bool ptrace_may_access(struct task_struct *task, unsigned int mode)
 366 {
 367         int err;
 368         task_lock(task);
 369         err = __ptrace_may_access(task, mode);
 370         task_unlock(task);
 371         return !err;
 372 }
 373
 374 static int ptrace_attach(struct task_struct *task, long request,
 375                          unsigned long addr,
 376                          unsigned long flags)
 377 {
 378         bool seize = (request == PTRACE_SEIZE);
 379         int retval;
 380
 381         retval = -EIO;
 382         if (seize) {
 383                 if (addr != 0)
 384                         goto out;
 385                 if (flags & ~(unsigned long)PTRACE_O_MASK)
 386                         goto out;
 387                 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
 388         } else {
 389                 flags = PT_PTRACED;
 390         }
 391
 392         audit_ptrace(task);
 393
 394         retval = -EPERM;
 395         if (unlikely(task->flags & PF_KTHREAD))
 396                 goto out;
 397         if (same_thread_group(task, current))
 398                 goto out;
 399
 400         /*
 401          * Protect exec's credential calculations against our interference;
 402          * SUID, SGID and LSM creds get determined differently
 403          * under ptrace.
 404          */
 405         retval = -ERESTARTNOINTR;
 406         if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
 407                 goto out;
 408
 409         task_lock(task);
 410         retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
 411         task_unlock(task);
 412         if (retval)
 413                 goto unlock_creds;
 414
 415         write_lock_irq(&tasklist_lock);
 416         retval = -EPERM;
 417         if (unlikely(task->exit_state))
 418                 goto unlock_tasklist;
 419         if (task->ptrace)
 420                 goto unlock_tasklist;
 421
 422         task->ptrace = flags;
 423
 424         ptrace_link(task, current);
 425
 426         /* SEIZE doesn't trap tracee on attach */
 427         if (!seize)
 428                 send_sig_info(SIGSTOP, SEND_SIG_PRIV, task);
 429
 430         spin_lock(&task->sighand->siglock);
 431
 432         /*
 433          * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
 434          * TRAPPING, and kick it so that it transits to TRACED.  TRAPPING
 435          * will be cleared if the child completes the transition or any
 436          * event which clears the group stop states happens.  We'll wait
 437          * for the transition to complete before returning from this
 438          * function.
 439          *
 440          * This hides STOPPED -> RUNNING -> TRACED transition from the
 441          * attaching thread but a different thread in the same group can
 442          * still observe the transient RUNNING state.  IOW, if another
 443          * thread's WNOHANG wait(2) on the stopped tracee races against
 444          * ATTACH, the wait(2) may fail due to the transient RUNNING.
 445          *
 446          * The following task_is_stopped() test is safe as both transitions
 447          * in and out of STOPPED are protected by siglock.
 448          */
 449         if (task_is_stopped(task) &&
 450             task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
 451                 signal_wake_up_state(task, __TASK_STOPPED);
 452
 453         spin_unlock(&task->sighand->siglock);
 454
 455         retval = 0;
 456 unlock_tasklist:
 457         write_unlock_irq(&tasklist_lock);
 458 unlock_creds:
 459         mutex_unlock(&task->signal->cred_guard_mutex);
 460 out:
 461         if (!retval) {
 462                 /*
 463                  * We do not bother to change retval or clear JOBCTL_TRAPPING
 464                  * if wait_on_bit() was interrupted by SIGKILL. The tracer will
 465                  * not return to user-mode, it will exit and clear this bit in
 466                  * __ptrace_unlink() if it wasn't already cleared by the tracee;
 467                  * and until then nobody can ptrace this task.
 468                  */
 469                 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
 470                 proc_ptrace_connector(task, PTRACE_ATTACH);
 471         }
 472
 473         return retval;
 474 }
 475
 476 /**
 477  * ptrace_traceme  --  helper for PTRACE_TRACEME
 478  *
 479  * Performs checks and sets PT_PTRACED.
 480  * Should be used by all ptrace implementations for PTRACE_TRACEME.
 481  */
 482 static int ptrace_traceme(void)
 483 {
 484         int ret = -EPERM;
 485
 486         write_lock_irq(&tasklist_lock);
 487         /* Are we already being traced? */
 488         if (!current->ptrace) {
 489                 ret = security_ptrace_traceme(current->parent);
 490                 /*
 491                  * Check PF_EXITING to ensure ->real_parent has not passed
 492                  * exit_ptrace(). Otherwise we don't report the error but
 493                  * pretend ->real_parent untraces us right after return.
 494                  */
 495                 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
 496                         current->ptrace = PT_PTRACED;
 497                         ptrace_link(current, current->real_parent);
 498                 }
 499         }
 500         write_unlock_irq(&tasklist_lock);
 501
 502         return ret;
 503 }
 504
 505 /*
 506  * Called with irqs disabled, returns true if childs should reap themselves.
 507  */
 508 static int ignoring_children(struct sighand_struct *sigh)
 509 {
 510         int ret;
 511         spin_lock(&sigh->siglock);
 512         ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
 513               (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
 514         spin_unlock(&sigh->siglock);
 515         return ret;
 516 }
 517
 518 /*
 519  * Called with tasklist_lock held for writing.
 520  * Unlink a traced task, and clean it up if it was a traced zombie.
 521  * Return true if it needs to be reaped with release_task().
 522  * (We can't call release_task() here because we already hold tasklist_lock.)
 523  *
 524  * If it's a zombie, our attachedness prevented normal parent notification
 525  * or self-reaping.  Do notification now if it would have happened earlier.
 526  * If it should reap itself, return true.
 527  *
 528  * If it's our own child, there is no notification to do. But if our normal
 529  * children self-reap, then this child was prevented by ptrace and we must
 530  * reap it now, in that case we must also wake up sub-threads sleeping in
 531  * do_wait().
 532  */
 533 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
 534 {
 535         bool dead;
 536
 537         __ptrace_unlink(p);
 538
 539         if (p->exit_state != EXIT_ZOMBIE)
 540                 return false;
 541
 542         dead = !thread_group_leader(p);
 543
 544         if (!dead && thread_group_empty(p)) {
 545                 if (!same_thread_group(p->real_parent, tracer))
 546                         dead = do_notify_parent(p, p->exit_signal);
 547                 else if (ignoring_children(tracer->sighand)) {
 548                         __wake_up_parent(p, tracer);
 549                         dead = true;
 550                 }
 551         }
 552         /* Mark it as in the process of being reaped. */
 553         if (dead)
 554                 p->exit_state = EXIT_DEAD;
 555         return dead;
 556 }
 557
 558 static int ptrace_detach(struct task_struct *child, unsigned int data)
 559 {
 560         if (!valid_signal(data))
 561                 return -EIO;
 562
 563         /* Architecture-specific hardware disable .. */
 564         ptrace_disable(child);
 565
 566         write_lock_irq(&tasklist_lock);
 567         /*
 568          * We rely on ptrace_freeze_traced(). It can't be killed and
 569          * untraced by another thread, it can't be a zombie.
 570          */
 571         WARN_ON(!child->ptrace || child->exit_state);
 572         /*
 573          * tasklist_lock avoids the race with wait_task_stopped(), see
 574          * the comment in ptrace_resume().
 575          */
 576         child->exit_code = data;
 577         __ptrace_detach(current, child);
 578         write_unlock_irq(&tasklist_lock);
 579
 580         proc_ptrace_connector(child, PTRACE_DETACH);
 581
 582         return 0;
 583 }
 584
 585 /*
 586  * Detach all tasks we were using ptrace on. Called with tasklist held
 587  * for writing.
 588  */
 589 void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
 590 {
 591         struct task_struct *p, *n;
 592
 593         list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
 594                 if (unlikely(p->ptrace & PT_EXITKILL))
 595                         send_sig_info(SIGKILL, SEND_SIG_PRIV, p);
 596
 597                 if (__ptrace_detach(tracer, p))
 598                         list_add(&p->ptrace_entry, dead);
 599         }
 600 }
 601
 602 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
 603 {
 604         int copied = 0;
 605
 606         while (len > 0) {
 607                 char buf[128];
 608                 int this_len, retval;
 609
 610                 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 611                 retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
 612
 613                 if (!retval) {
 614                         if (copied)
 615                                 break;
 616                         return -EIO;
 617                 }
 618                 if (copy_to_user(dst, buf, retval))
 619                         return -EFAULT;
 620                 copied += retval;
 621                 src += retval;
 622                 dst += retval;
 623                 len -= retval;
 624         }
 625         return copied;
 626 }
 627
 628 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
 629 {
 630         int copied = 0;
 631
 632         while (len > 0) {
 633                 char buf[128];
 634                 int this_len, retval;
 635
 636                 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 637                 if (copy_from_user(buf, src, this_len))
 638                         return -EFAULT;
 639                 retval = ptrace_access_vm(tsk, dst, buf, this_len,
 640                                 FOLL_FORCE | FOLL_WRITE);
 641                 if (!retval) {
 642                         if (copied)
 643                                 break;
 644                         return -EIO;
 645                 }
 646                 copied += retval;
 647                 src += retval;
 648                 dst += retval;
 649                 len -= retval;
 650         }
 651         return copied;
 652 }
 653
 654 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
 655 {
 656         unsigned flags;
 657
 658         if (data & ~(unsigned long)PTRACE_O_MASK)
 659                 return -EINVAL;
 660
 661         if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
 662                 if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
 663                     !IS_ENABLED(CONFIG_SECCOMP))
 664                         return -EINVAL;
 665
 666                 if (!capable(CAP_SYS_ADMIN))
 667                         return -EPERM;
 668
 669                 if (seccomp_mode(&current->seccomp) != SECCOMP_MODE_DISABLED ||
 670                     current->ptrace & PT_SUSPEND_SECCOMP)
 671                         return -EPERM;
 672         }
 673
 674         /* Avoid intermediate state when all opts are cleared */
 675         flags = child->ptrace;
 676         flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
 677         flags |= (data << PT_OPT_FLAG_SHIFT);
 678         child->ptrace = flags;
 679
 680         return 0;
 681 }
 682
 683 static int ptrace_getsiginfo(struct task_struct *child, kernel_siginfo_t *info)
 684 {
 685         unsigned long flags;
 686         int error = -ESRCH;
 687
 688         if (lock_task_sighand(child, &flags)) {
 689                 error = -EINVAL;
 690                 if (likely(child->last_siginfo != NULL)) {
 691                         copy_siginfo(info, child->last_siginfo);
 692                         error = 0;
 693                 }
 694                 unlock_task_sighand(child, &flags);
 695         }
 696         return error;
 697 }
 698
 699 static int ptrace_setsiginfo(struct task_struct *child, const kernel_siginfo_t *info)
 700 {
 701         unsigned long flags;
 702         int error = -ESRCH;
 703
 704         if (lock_task_sighand(child, &flags)) {
 705                 error = -EINVAL;
 706                 if (likely(child->last_siginfo != NULL)) {
 707                         copy_siginfo(child->last_siginfo, info);
 708                         error = 0;
 709                 }
 710                 unlock_task_sighand(child, &flags);
 711         }
 712         return error;
 713 }
 714
 715 static int ptrace_peek_siginfo(struct task_struct *child,
 716                                 unsigned long addr,
 717                                 unsigned long data)
 718 {
 719         struct ptrace_peeksiginfo_args arg;
 720         struct sigpending *pending;
 721         struct sigqueue *q;
 722         int ret, i;
 723
 724         ret = copy_from_user(&arg, (void __user *) addr,
 725                                 sizeof(struct ptrace_peeksiginfo_args));
 726         if (ret)
 727                 return -EFAULT;
 728
 729         if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
 730                 return -EINVAL; /* unknown flags */
 731
 732         if (arg.nr < 0)
 733                 return -EINVAL;
 734
 735         /* Ensure arg.off fits in an unsigned long */
 736         if (arg.off > ULONG_MAX)
 737                 return 0;
 738
 739         if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
 740                 pending = &child->signal->shared_pending;
 741         else
 742                 pending = &child->pending;
 743
 744         for (i = 0; i < arg.nr; ) {
 745                 kernel_siginfo_t info;
 746                 unsigned long off = arg.off + i;
 747                 bool found = false;
 748
 749                 spin_lock_irq(&child->sighand->siglock);
 750                 list_for_each_entry(q, &pending->list, list) {
 751                         if (!off--) {
 752                                 found = true;
 753                                 copy_siginfo(&info, &q->info);
 754                                 break;
 755                         }
 756                 }
 757                 spin_unlock_irq(&child->sighand->siglock);
 758
 759                 if (!found) /* beyond the end of the list */
 760                         break;
 761
 762 #ifdef CONFIG_COMPAT
 763                 if (unlikely(in_compat_syscall())) {
 764                         compat_siginfo_t __user *uinfo = compat_ptr(data);
 765
 766                         if (copy_siginfo_to_user32(uinfo, &info)) {
 767                                 ret = -EFAULT;
 768                                 break;
 769                         }
 770
 771                 } else
 772 #endif
 773                 {
 774                         siginfo_t __user *uinfo = (siginfo_t __user *) data;
 775
 776                         if (copy_siginfo_to_user(uinfo, &info)) {
 777                                 ret = -EFAULT;
 778                                 break;
 779                         }
 780                 }
 781
 782                 data += sizeof(siginfo_t);
 783                 i++;
 784
 785                 if (signal_pending(current))
 786                         break;
 787
 788                 cond_resched();
 789         }
 790
 791         if (i > 0)
 792                 return i;
 793
 794         return ret;
 795 }
 796
 797 #ifdef CONFIG_RSEQ
 798 static long ptrace_get_rseq_configuration(struct task_struct *task,
 799                                           unsigned long size, void __user *data)
 800 {
 801         struct ptrace_rseq_configuration conf = {
 802                 .rseq_abi_pointer = (u64)(uintptr_t)task->rseq,
 803                 .rseq_abi_size = sizeof(*task->rseq),
 804                 .signature = task->rseq_sig,
 805                 .flags = 0,
 806         };
 807
 808         size = min_t(unsigned long, size, sizeof(conf));
 809         if (copy_to_user(data, &conf, size))
 810                 return -EFAULT;
 811         return sizeof(conf);
 812 }
 813 #endif
 814
 815 #ifdef PTRACE_SINGLESTEP
 816 #define is_singlestep(request)          ((request) == PTRACE_SINGLESTEP)
 817 #else
 818 #define is_singlestep(request)          0
 819 #endif
 820
 821 #ifdef PTRACE_SINGLEBLOCK
 822 #define is_singleblock(request)         ((request) == PTRACE_SINGLEBLOCK)
 823 #else
 824 #define is_singleblock(request)         0
 825 #endif
 826
 827 #ifdef PTRACE_SYSEMU
 828 #define is_sysemu_singlestep(request)   ((request) == PTRACE_SYSEMU_SINGLESTEP)
 829 #else
 830 #define is_sysemu_singlestep(request)   0
 831 #endif
 832
 833 static int ptrace_resume(struct task_struct *child, long request,
 834                          unsigned long data)
 835 {
 836         bool need_siglock;
 837
 838         if (!valid_signal(data))
 839                 return -EIO;
 840
 841         if (request == PTRACE_SYSCALL)
 842                 set_task_syscall_work(child, SYSCALL_TRACE);
 843         else
 844                 clear_task_syscall_work(child, SYSCALL_TRACE);
 845
 846 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
 847         if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
 848                 set_task_syscall_work(child, SYSCALL_EMU);
 849         else
 850                 clear_task_syscall_work(child, SYSCALL_EMU);
 851 #endif
 852
 853         if (is_singleblock(request)) {
 854                 if (unlikely(!arch_has_block_step()))
 855                         return -EIO;
 856                 user_enable_block_step(child);
 857         } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
 858                 if (unlikely(!arch_has_single_step()))
 859                         return -EIO;
 860                 user_enable_single_step(child);
 861         } else {
 862                 user_disable_single_step(child);
 863         }
 864
 865         /*
 866          * Change ->exit_code and ->state under siglock to avoid the race
 867          * with wait_task_stopped() in between; a non-zero ->exit_code will
 868          * wrongly look like another report from tracee.
 869          *
 870          * Note that we need siglock even if ->exit_code == data and/or this
 871          * status was not reported yet, the new status must not be cleared by
 872          * wait_task_stopped() after resume.
 873          *
 874          * If data == 0 we do not care if wait_task_stopped() reports the old
 875          * status and clears the code too; this can't race with the tracee, it
 876          * takes siglock after resume.
 877          */
 878         need_siglock = data && !thread_group_empty(current);
 879         if (need_siglock)
 880                 spin_lock_irq(&child->sighand->siglock);
 881         child->exit_code = data;
 882         wake_up_state(child, __TASK_TRACED);
 883         if (need_siglock)
 884                 spin_unlock_irq(&child->sighand->siglock);
 885
 886         return 0;
 887 }
 888
 889 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 890
 891 static const struct user_regset *
 892 find_regset(const struct user_regset_view *view, unsigned int type)
 893 {
 894         const struct user_regset *regset;
 895         int n;
 896
 897         for (n = 0; n < view->n; ++n) {
 898                 regset = view->regsets + n;
 899                 if (regset->core_note_type == type)
 900                         return regset;
 901         }
 902
 903         return NULL;
 904 }
 905
 906 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
 907                          struct iovec *kiov)
 908 {
 909         const struct user_regset_view *view = task_user_regset_view(task);
 910         const struct user_regset *regset = find_regset(view, type);
 911         int regset_no;
 912
 913         if (!regset || (kiov->iov_len % regset->size) != 0)
 914                 return -EINVAL;
 915
 916         regset_no = regset - view->regsets;
 917         kiov->iov_len = min(kiov->iov_len,
 918                             (__kernel_size_t) (regset->n * regset->size));
 919
 920         if (req == PTRACE_GETREGSET)
 921                 return copy_regset_to_user(task, view, regset_no, 0,
 922                                            kiov->iov_len, kiov->iov_base);
 923         else
 924                 return copy_regset_from_user(task, view, regset_no, 0,
 925                                              kiov->iov_len, kiov->iov_base);
 926 }
 927
 928 /*
 929  * This is declared in linux/regset.h and defined in machine-dependent
 930  * code.  We put the export here, near the primary machine-neutral use,
 931  * to ensure no machine forgets it.
 932  */
 933 EXPORT_SYMBOL_GPL(task_user_regset_view);
 934
 935 static unsigned long
 936 ptrace_get_syscall_info_entry(struct task_struct *child, struct pt_regs *regs,
 937                               struct ptrace_syscall_info *info)
 938 {
 939         unsigned long args[ARRAY_SIZE(info->entry.args)];
 940         int i;
 941
 942         info->op = PTRACE_SYSCALL_INFO_ENTRY;
 943         info->entry.nr = syscall_get_nr(child, regs);
 944         syscall_get_arguments(child, regs, args);
 945         for (i = 0; i < ARRAY_SIZE(args); i++)
 946                 info->entry.args[i] = args[i];
 947
 948         /* args is the last field in struct ptrace_syscall_info.entry */
 949         return offsetofend(struct ptrace_syscall_info, entry.args);
 950 }
 951
 952 static unsigned long
 953 ptrace_get_syscall_info_seccomp(struct task_struct *child, struct pt_regs *regs,
 954                                 struct ptrace_syscall_info *info)
 955 {
 956         /*
 957          * As struct ptrace_syscall_info.entry is currently a subset
 958          * of struct ptrace_syscall_info.seccomp, it makes sense to
 959          * initialize that subset using ptrace_get_syscall_info_entry().
 960          * This can be reconsidered in the future if these structures
 961          * diverge significantly enough.
 962          */
 963         ptrace_get_syscall_info_entry(child, regs, info);
 964         info->op = PTRACE_SYSCALL_INFO_SECCOMP;
 965         info->seccomp.ret_data = child->ptrace_message;
 966
 967         /* ret_data is the last field in struct ptrace_syscall_info.seccomp */
 968         return offsetofend(struct ptrace_syscall_info, seccomp.ret_data);
 969 }
 970
 971 static unsigned long
 972 ptrace_get_syscall_info_exit(struct task_struct *child, struct pt_regs *regs,
 973                              struct ptrace_syscall_info *info)
 974 {
 975         info->op = PTRACE_SYSCALL_INFO_EXIT;
 976         info->exit.rval = syscall_get_error(child, regs);
 977         info->exit.is_error = !!info->exit.rval;
 978         if (!info->exit.is_error)
 979                 info->exit.rval = syscall_get_return_value(child, regs);
 980
 981         /* is_error is the last field in struct ptrace_syscall_info.exit */
 982         return offsetofend(struct ptrace_syscall_info, exit.is_error);
 983 }
 984
 985 static int
 986 ptrace_get_syscall_info(struct task_struct *child, unsigned long user_size,
 987                         void __user *datavp)
 988 {
 989         struct pt_regs *regs = task_pt_regs(child);
 990         struct ptrace_syscall_info info = {
 991                 .op = PTRACE_SYSCALL_INFO_NONE,
 992                 .arch = syscall_get_arch(child),
 993                 .instruction_pointer = instruction_pointer(regs),
 994                 .stack_pointer = user_stack_pointer(regs),
 995         };
 996         unsigned long actual_size = offsetof(struct ptrace_syscall_info, entry);
 997         unsigned long write_size;
 998
 999         /*
1000          * This does not need lock_task_sighand() to access
1001          * child->last_siginfo because ptrace_freeze_traced()
1002          * called earlier by ptrace_check_attach() ensures that
1003          * the tracee cannot go away and clear its last_siginfo.
1004          */
1005         switch (child->last_siginfo ? child->last_siginfo->si_code : 0) {
1006         case SIGTRAP | 0x80:
1007                 switch (child->ptrace_message) {
1008                 case PTRACE_EVENTMSG_SYSCALL_ENTRY:
1009                         actual_size = ptrace_get_syscall_info_entry(child, regs,
1010                                                                     &info);
1011                         break;
1012                 case PTRACE_EVENTMSG_SYSCALL_EXIT:
1013                         actual_size = ptrace_get_syscall_info_exit(child, regs,
1014                                                                    &info);
1015                         break;
1016                 }
1017                 break;
1018         case SIGTRAP | (PTRACE_EVENT_SECCOMP << 8):
1019                 actual_size = ptrace_get_syscall_info_seccomp(child, regs,
1020                                                               &info);
1021                 break;
1022         }
1023
1024         write_size = min(actual_size, user_size);
1025         return copy_to_user(datavp, &info, write_size) ? -EFAULT : actual_size;
1026 }
1027 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
1028
1029 int ptrace_request(struct task_struct *child, long request,
1030                    unsigned long addr, unsigned long data)
1031 {
1032         bool seized = child->ptrace & PT_SEIZED;
1033         int ret = -EIO;
1034         kernel_siginfo_t siginfo, *si;
1035         void __user *datavp = (void __user *) data;
1036         unsigned long __user *datalp = datavp;
1037         unsigned long flags;
1038
1039         switch (request) {
1040         case PTRACE_PEEKTEXT:
1041         case PTRACE_PEEKDATA:
1042                 return generic_ptrace_peekdata(child, addr, data);
1043         case PTRACE_POKETEXT:
1044         case PTRACE_POKEDATA:
1045                 return generic_ptrace_pokedata(child, addr, data);
1046
1047 #ifdef PTRACE_OLDSETOPTIONS
1048         case PTRACE_OLDSETOPTIONS:
1049 #endif
1050         case PTRACE_SETOPTIONS:
1051                 ret = ptrace_setoptions(child, data);
1052                 break;
1053         case PTRACE_GETEVENTMSG:
1054                 ret = put_user(child->ptrace_message, datalp);
1055                 break;
1056
1057         case PTRACE_PEEKSIGINFO:
1058                 ret = ptrace_peek_siginfo(child, addr, data);
1059                 break;
1060
1061         case PTRACE_GETSIGINFO:
1062                 ret = ptrace_getsiginfo(child, &siginfo);
1063                 if (!ret)
1064                         ret = copy_siginfo_to_user(datavp, &siginfo);
1065                 break;
1066
1067         case PTRACE_SETSIGINFO:
1068                 ret = copy_siginfo_from_user(&siginfo, datavp);
1069                 if (!ret)
1070                         ret = ptrace_setsiginfo(child, &siginfo);
1071                 break;
1072
1073         case PTRACE_GETSIGMASK: {
1074                 sigset_t *mask;
1075
1076                 if (addr != sizeof(sigset_t)) {
1077                         ret = -EINVAL;
1078                         break;
1079                 }
1080
1081                 if (test_tsk_restore_sigmask(child))
1082                         mask = &child->saved_sigmask;
1083                 else
1084                         mask = &child->blocked;
1085
1086                 if (copy_to_user(datavp, mask, sizeof(sigset_t)))
1087                         ret = -EFAULT;
1088                 else
1089                         ret = 0;
1090
1091                 break;
1092         }
1093
1094         case PTRACE_SETSIGMASK: {
1095                 sigset_t new_set;
1096
1097                 if (addr != sizeof(sigset_t)) {
1098                         ret = -EINVAL;
1099                         break;
1100                 }
1101
1102                 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
1103                         ret = -EFAULT;
1104                         break;
1105                 }
1106
1107                 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
1108
1109                 /*
1110                  * Every thread does recalc_sigpending() after resume, so
1111                  * retarget_shared_pending() and recalc_sigpending() are not
1112                  * called here.
1113                  */
1114                 spin_lock_irq(&child->sighand->siglock);
1115                 child->blocked = new_set;
1116                 spin_unlock_irq(&child->sighand->siglock);
1117
1118                 clear_tsk_restore_sigmask(child);
1119
1120                 ret = 0;
1121                 break;
1122         }
1123
1124         case PTRACE_INTERRUPT:
1125                 /*
1126                  * Stop tracee without any side-effect on signal or job
1127                  * control.  At least one trap is guaranteed to happen
1128                  * after this request.  If @child is already trapped, the
1129                  * current trap is not disturbed and another trap will
1130                  * happen after the current trap is ended with PTRACE_CONT.
1131                  *
1132                  * The actual trap might not be PTRACE_EVENT_STOP trap but
1133                  * the pending condition is cleared regardless.
1134                  */
1135                 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1136                         break;
1137
1138                 /*
1139                  * INTERRUPT doesn't disturb existing trap sans one
1140                  * exception.  If ptracer issued LISTEN for the current
1141                  * STOP, this INTERRUPT should clear LISTEN and re-trap
1142                  * tracee into STOP.
1143                  */
1144                 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
1145                         ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1146
1147                 unlock_task_sighand(child, &flags);
1148                 ret = 0;
1149                 break;
1150
1151         case PTRACE_LISTEN:
1152                 /*
1153                  * Listen for events.  Tracee must be in STOP.  It's not
1154                  * resumed per-se but is not considered to be in TRACED by
1155                  * wait(2) or ptrace(2).  If an async event (e.g. group
1156                  * stop state change) happens, tracee will enter STOP trap
1157                  * again.  Alternatively, ptracer can issue INTERRUPT to
1158                  * finish listening and re-trap tracee into STOP.
1159                  */
1160                 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1161                         break;
1162
1163                 si = child->last_siginfo;
1164                 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1165                         child->jobctl |= JOBCTL_LISTENING;
1166                         /*
1167                          * If NOTIFY is set, it means event happened between
1168                          * start of this trap and now.  Trigger re-trap.
1169                          */
1170                         if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1171                                 ptrace_signal_wake_up(child, true);
1172                         ret = 0;
1173                 }
1174                 unlock_task_sighand(child, &flags);
1175                 break;
1176
1177         case PTRACE_DETACH:      /* detach a process that was attached. */
1178                 ret = ptrace_detach(child, data);
1179                 break;
1180
1181 #ifdef CONFIG_BINFMT_ELF_FDPIC
1182         case PTRACE_GETFDPIC: {
1183                 struct mm_struct *mm = get_task_mm(child);
1184                 unsigned long tmp = 0;
1185
1186                 ret = -ESRCH;
1187                 if (!mm)
1188                         break;
1189
1190                 switch (addr) {
1191                 case PTRACE_GETFDPIC_EXEC:
1192                         tmp = mm->context.exec_fdpic_loadmap;
1193                         break;
1194                 case PTRACE_GETFDPIC_INTERP:
1195                         tmp = mm->context.interp_fdpic_loadmap;
1196                         break;
1197                 default:
1198                         break;
1199                 }
1200                 mmput(mm);
1201
1202                 ret = put_user(tmp, datalp);
1203                 break;
1204         }
1205 #endif
1206
1207 #ifdef PTRACE_SINGLESTEP
1208         case PTRACE_SINGLESTEP:
1209 #endif
1210 #ifdef PTRACE_SINGLEBLOCK
1211         case PTRACE_SINGLEBLOCK:
1212 #endif
1213 #ifdef PTRACE_SYSEMU
1214         case PTRACE_SYSEMU:
1215         case PTRACE_SYSEMU_SINGLESTEP:
1216 #endif
1217         case PTRACE_SYSCALL:
1218         case PTRACE_CONT:
1219                 return ptrace_resume(child, request, data);
1220
1221         case PTRACE_KILL:
1222                 if (child->exit_state)  /* already dead */
1223                         return 0;
1224                 return ptrace_resume(child, request, SIGKILL);
1225
1226 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1227         case PTRACE_GETREGSET:
1228         case PTRACE_SETREGSET: {
1229                 struct iovec kiov;
1230                 struct iovec __user *uiov = datavp;
1231
1232                 if (!access_ok(uiov, sizeof(*uiov)))
1233                         return -EFAULT;
1234
1235                 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1236                     __get_user(kiov.iov_len, &uiov->iov_len))
1237                         return -EFAULT;
1238
1239                 ret = ptrace_regset(child, request, addr, &kiov);
1240                 if (!ret)
1241                         ret = __put_user(kiov.iov_len, &uiov->iov_len);
1242                 break;
1243         }
1244
1245         case PTRACE_GET_SYSCALL_INFO:
1246                 ret = ptrace_get_syscall_info(child, addr, datavp);
1247                 break;
1248 #endif
1249
1250         case PTRACE_SECCOMP_GET_FILTER:
1251                 ret = seccomp_get_filter(child, addr, datavp);
1252                 break;
1253
1254         case PTRACE_SECCOMP_GET_METADATA:
1255                 ret = seccomp_get_metadata(child, addr, datavp);
1256                 break;
1257
1258 #ifdef CONFIG_RSEQ
1259         case PTRACE_GET_RSEQ_CONFIGURATION:
1260                 ret = ptrace_get_rseq_configuration(child, addr, datavp);
1261                 break;
1262 #endif
1263
1264         default:
1265                 break;
1266         }
1267
1268         return ret;
1269 }
1270
1271 #ifndef arch_ptrace_attach
1272 #define arch_ptrace_attach(child)       do { } while (0)
1273 #endif
1274
1275 SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1276                 unsigned long, data)
1277 {
1278         struct task_struct *child;
1279         long ret;
1280
1281         if (request == PTRACE_TRACEME) {
1282                 ret = ptrace_traceme();
1283                 if (!ret)
1284                         arch_ptrace_attach(current);
1285                 goto out;
1286         }
1287
1288         child = find_get_task_by_vpid(pid);
1289         if (!child) {
1290                 ret = -ESRCH;
1291                 goto out;
1292         }
1293
1294         if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1295                 ret = ptrace_attach(child, request, addr, data);
1296                 /*
1297                  * Some architectures need to do book-keeping after
1298                  * a ptrace attach.
1299                  */
1300                 if (!ret)
1301                         arch_ptrace_attach(child);
1302                 goto out_put_task_struct;
1303         }
1304
1305         ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1306                                   request == PTRACE_INTERRUPT);
1307         if (ret < 0)
1308                 goto out_put_task_struct;
1309
1310         ret = arch_ptrace(child, request, addr, data);
1311         if (ret || request != PTRACE_DETACH)
1312                 ptrace_unfreeze_traced(child);
1313
1314  out_put_task_struct:
1315         put_task_struct(child);
1316  out:
1317         return ret;
1318 }
1319
1320 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1321                             unsigned long data)
1322 {
1323         unsigned long tmp;
1324         int copied;
1325
1326         copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1327         if (copied != sizeof(tmp))
1328                 return -EIO;
1329         return put_user(tmp, (unsigned long __user *)data);
1330 }
1331
1332 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1333                             unsigned long data)
1334 {
1335         int copied;
1336
1337         copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1338                         FOLL_FORCE | FOLL_WRITE);
1339         return (copied == sizeof(data)) ? 0 : -EIO;
1340 }
1341
1342 #if defined CONFIG_COMPAT
1343
1344 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1345                           compat_ulong_t addr, compat_ulong_t data)
1346 {
1347         compat_ulong_t __user *datap = compat_ptr(data);
1348         compat_ulong_t word;
1349         kernel_siginfo_t siginfo;
1350         int ret;
1351
1352         switch (request) {
1353         case PTRACE_PEEKTEXT:
1354         case PTRACE_PEEKDATA:
1355                 ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1356                                 FOLL_FORCE);
1357                 if (ret != sizeof(word))
1358                         ret = -EIO;
1359                 else
1360                         ret = put_user(word, datap);
1361                 break;
1362
1363         case PTRACE_POKETEXT:
1364         case PTRACE_POKEDATA:
1365                 ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1366                                 FOLL_FORCE | FOLL_WRITE);
1367                 ret = (ret != sizeof(data) ? -EIO : 0);
1368                 break;
1369
1370         case PTRACE_GETEVENTMSG:
1371                 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1372                 break;
1373
1374         case PTRACE_GETSIGINFO:
1375                 ret = ptrace_getsiginfo(child, &siginfo);
1376                 if (!ret)
1377                         ret = copy_siginfo_to_user32(
1378                                 (struct compat_siginfo __user *) datap,
1379                                 &siginfo);
1380                 break;
1381
1382         case PTRACE_SETSIGINFO:
1383                 ret = copy_siginfo_from_user32(
1384                         &siginfo, (struct compat_siginfo __user *) datap);
1385                 if (!ret)
1386                         ret = ptrace_setsiginfo(child, &siginfo);
1387                 break;
1388 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1389         case PTRACE_GETREGSET:
1390         case PTRACE_SETREGSET:
1391         {
1392                 struct iovec kiov;
1393                 struct compat_iovec __user *uiov =
1394                         (struct compat_iovec __user *) datap;
1395                 compat_uptr_t ptr;
1396                 compat_size_t len;
1397
1398                 if (!access_ok(uiov, sizeof(*uiov)))
1399                         return -EFAULT;
1400
1401                 if (__get_user(ptr, &uiov->iov_base) ||
1402                     __get_user(len, &uiov->iov_len))
1403                         return -EFAULT;
1404
1405                 kiov.iov_base = compat_ptr(ptr);
1406                 kiov.iov_len = len;
1407
1408                 ret = ptrace_regset(child, request, addr, &kiov);
1409                 if (!ret)
1410                         ret = __put_user(kiov.iov_len, &uiov->iov_len);
1411                 break;
1412         }
1413 #endif
1414
1415         default:
1416                 ret = ptrace_request(child, request, addr, data);
1417         }
1418
1419         return ret;
1420 }
1421
1422 COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1423                        compat_long_t, addr, compat_long_t, data)
1424 {
1425         struct task_struct *child;
1426         long ret;
1427
1428         if (request == PTRACE_TRACEME) {
1429                 ret = ptrace_traceme();
1430                 goto out;
1431         }
1432
1433         child = find_get_task_by_vpid(pid);
1434         if (!child) {
1435                 ret = -ESRCH;
1436                 goto out;
1437         }
1438
1439         if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1440                 ret = ptrace_attach(child, request, addr, data);
1441                 /*
1442                  * Some architectures need to do book-keeping after
1443                  * a ptrace attach.
1444                  */
1445                 if (!ret)
1446                         arch_ptrace_attach(child);
1447                 goto out_put_task_struct;
1448         }
1449
1450         ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1451                                   request == PTRACE_INTERRUPT);
1452         if (!ret) {
1453                 ret = compat_arch_ptrace(child, request, addr, data);
1454                 if (ret || request != PTRACE_DETACH)
1455                         ptrace_unfreeze_traced(child);
1456         }
1457
1458  out_put_task_struct:
1459         put_task_struct(child);
1460  out:
1461         return ret;
1462 }
1463 #endif  /* CONFIG_COMPAT */