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 check_ptrace_options(unsigned long data)
 375 {
 376         if (data & ~(unsigned long)PTRACE_O_MASK)
 377                 return -EINVAL;
 378
 379         if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
 380                 if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
 381                     !IS_ENABLED(CONFIG_SECCOMP))
 382                         return -EINVAL;
 383
 384                 if (!capable(CAP_SYS_ADMIN))
 385                         return -EPERM;
 386
 387                 if (seccomp_mode(&current->seccomp) != SECCOMP_MODE_DISABLED ||
 388                     current->ptrace & PT_SUSPEND_SECCOMP)
 389                         return -EPERM;
 390         }
 391         return 0;
 392 }
 393
 394 static int ptrace_attach(struct task_struct *task, long request,
 395                          unsigned long addr,
 396                          unsigned long flags)
 397 {
 398         bool seize = (request == PTRACE_SEIZE);
 399         int retval;
 400
 401         retval = -EIO;
 402         if (seize) {
 403                 if (addr != 0)
 404                         goto out;
 405                 /*
 406                  * This duplicates the check in check_ptrace_options() because
 407                  * ptrace_attach() and ptrace_setoptions() have historically
 408                  * used different error codes for unknown ptrace options.
 409                  */
 410                 if (flags & ~(unsigned long)PTRACE_O_MASK)
 411                         goto out;
 412                 retval = check_ptrace_options(flags);
 413                 if (retval)
 414                         return retval;
 415                 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
 416         } else {
 417                 flags = PT_PTRACED;
 418         }
 419
 420         audit_ptrace(task);
 421
 422         retval = -EPERM;
 423         if (unlikely(task->flags & PF_KTHREAD))
 424                 goto out;
 425         if (same_thread_group(task, current))
 426                 goto out;
 427
 428         /*
 429          * Protect exec's credential calculations against our interference;
 430          * SUID, SGID and LSM creds get determined differently
 431          * under ptrace.
 432          */
 433         retval = -ERESTARTNOINTR;
 434         if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
 435                 goto out;
 436
 437         task_lock(task);
 438         retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
 439         task_unlock(task);
 440         if (retval)
 441                 goto unlock_creds;
 442
 443         write_lock_irq(&tasklist_lock);
 444         retval = -EPERM;
 445         if (unlikely(task->exit_state))
 446                 goto unlock_tasklist;
 447         if (task->ptrace)
 448                 goto unlock_tasklist;
 449
 450         if (seize)
 451                 flags |= PT_SEIZED;
 452         task->ptrace = flags;
 453
 454         ptrace_link(task, current);
 455
 456         /* SEIZE doesn't trap tracee on attach */
 457         if (!seize)
 458                 send_sig_info(SIGSTOP, SEND_SIG_PRIV, task);
 459
 460         spin_lock(&task->sighand->siglock);
 461
 462         /*
 463          * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
 464          * TRAPPING, and kick it so that it transits to TRACED.  TRAPPING
 465          * will be cleared if the child completes the transition or any
 466          * event which clears the group stop states happens.  We'll wait
 467          * for the transition to complete before returning from this
 468          * function.
 469          *
 470          * This hides STOPPED -> RUNNING -> TRACED transition from the
 471          * attaching thread but a different thread in the same group can
 472          * still observe the transient RUNNING state.  IOW, if another
 473          * thread's WNOHANG wait(2) on the stopped tracee races against
 474          * ATTACH, the wait(2) may fail due to the transient RUNNING.
 475          *
 476          * The following task_is_stopped() test is safe as both transitions
 477          * in and out of STOPPED are protected by siglock.
 478          */
 479         if (task_is_stopped(task) &&
 480             task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
 481                 signal_wake_up_state(task, __TASK_STOPPED);
 482
 483         spin_unlock(&task->sighand->siglock);
 484
 485         retval = 0;
 486 unlock_tasklist:
 487         write_unlock_irq(&tasklist_lock);
 488 unlock_creds:
 489         mutex_unlock(&task->signal->cred_guard_mutex);
 490 out:
 491         if (!retval) {
 492                 /*
 493                  * We do not bother to change retval or clear JOBCTL_TRAPPING
 494                  * if wait_on_bit() was interrupted by SIGKILL. The tracer will
 495                  * not return to user-mode, it will exit and clear this bit in
 496                  * __ptrace_unlink() if it wasn't already cleared by the tracee;
 497                  * and until then nobody can ptrace this task.
 498                  */
 499                 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
 500                 proc_ptrace_connector(task, PTRACE_ATTACH);
 501         }
 502
 503         return retval;
 504 }
 505
 506 /**
 507  * ptrace_traceme  --  helper for PTRACE_TRACEME
 508  *
 509  * Performs checks and sets PT_PTRACED.
 510  * Should be used by all ptrace implementations for PTRACE_TRACEME.
 511  */
 512 static int ptrace_traceme(void)
 513 {
 514         int ret = -EPERM;
 515
 516         write_lock_irq(&tasklist_lock);
 517         /* Are we already being traced? */
 518         if (!current->ptrace) {
 519                 ret = security_ptrace_traceme(current->parent);
 520                 /*
 521                  * Check PF_EXITING to ensure ->real_parent has not passed
 522                  * exit_ptrace(). Otherwise we don't report the error but
 523                  * pretend ->real_parent untraces us right after return.
 524                  */
 525                 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
 526                         current->ptrace = PT_PTRACED;
 527                         ptrace_link(current, current->real_parent);
 528                 }
 529         }
 530         write_unlock_irq(&tasklist_lock);
 531
 532         return ret;
 533 }
 534
 535 /*
 536  * Called with irqs disabled, returns true if childs should reap themselves.
 537  */
 538 static int ignoring_children(struct sighand_struct *sigh)
 539 {
 540         int ret;
 541         spin_lock(&sigh->siglock);
 542         ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
 543               (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
 544         spin_unlock(&sigh->siglock);
 545         return ret;
 546 }
 547
 548 /*
 549  * Called with tasklist_lock held for writing.
 550  * Unlink a traced task, and clean it up if it was a traced zombie.
 551  * Return true if it needs to be reaped with release_task().
 552  * (We can't call release_task() here because we already hold tasklist_lock.)
 553  *
 554  * If it's a zombie, our attachedness prevented normal parent notification
 555  * or self-reaping.  Do notification now if it would have happened earlier.
 556  * If it should reap itself, return true.
 557  *
 558  * If it's our own child, there is no notification to do. But if our normal
 559  * children self-reap, then this child was prevented by ptrace and we must
 560  * reap it now, in that case we must also wake up sub-threads sleeping in
 561  * do_wait().
 562  */
 563 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
 564 {
 565         bool dead;
 566
 567         __ptrace_unlink(p);
 568
 569         if (p->exit_state != EXIT_ZOMBIE)
 570                 return false;
 571
 572         dead = !thread_group_leader(p);
 573
 574         if (!dead && thread_group_empty(p)) {
 575                 if (!same_thread_group(p->real_parent, tracer))
 576                         dead = do_notify_parent(p, p->exit_signal);
 577                 else if (ignoring_children(tracer->sighand)) {
 578                         __wake_up_parent(p, tracer);
 579                         dead = true;
 580                 }
 581         }
 582         /* Mark it as in the process of being reaped. */
 583         if (dead)
 584                 p->exit_state = EXIT_DEAD;
 585         return dead;
 586 }
 587
 588 static int ptrace_detach(struct task_struct *child, unsigned int data)
 589 {
 590         if (!valid_signal(data))
 591                 return -EIO;
 592
 593         /* Architecture-specific hardware disable .. */
 594         ptrace_disable(child);
 595
 596         write_lock_irq(&tasklist_lock);
 597         /*
 598          * We rely on ptrace_freeze_traced(). It can't be killed and
 599          * untraced by another thread, it can't be a zombie.
 600          */
 601         WARN_ON(!child->ptrace || child->exit_state);
 602         /*
 603          * tasklist_lock avoids the race with wait_task_stopped(), see
 604          * the comment in ptrace_resume().
 605          */
 606         child->exit_code = data;
 607         __ptrace_detach(current, child);
 608         write_unlock_irq(&tasklist_lock);
 609
 610         proc_ptrace_connector(child, PTRACE_DETACH);
 611
 612         return 0;
 613 }
 614
 615 /*
 616  * Detach all tasks we were using ptrace on. Called with tasklist held
 617  * for writing.
 618  */
 619 void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
 620 {
 621         struct task_struct *p, *n;
 622
 623         list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
 624                 if (unlikely(p->ptrace & PT_EXITKILL))
 625                         send_sig_info(SIGKILL, SEND_SIG_PRIV, p);
 626
 627                 if (__ptrace_detach(tracer, p))
 628                         list_add(&p->ptrace_entry, dead);
 629         }
 630 }
 631
 632 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
 633 {
 634         int copied = 0;
 635
 636         while (len > 0) {
 637                 char buf[128];
 638                 int this_len, retval;
 639
 640                 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 641                 retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
 642
 643                 if (!retval) {
 644                         if (copied)
 645                                 break;
 646                         return -EIO;
 647                 }
 648                 if (copy_to_user(dst, buf, retval))
 649                         return -EFAULT;
 650                 copied += retval;
 651                 src += retval;
 652                 dst += retval;
 653                 len -= retval;
 654         }
 655         return copied;
 656 }
 657
 658 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
 659 {
 660         int copied = 0;
 661
 662         while (len > 0) {
 663                 char buf[128];
 664                 int this_len, retval;
 665
 666                 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 667                 if (copy_from_user(buf, src, this_len))
 668                         return -EFAULT;
 669                 retval = ptrace_access_vm(tsk, dst, buf, this_len,
 670                                 FOLL_FORCE | FOLL_WRITE);
 671                 if (!retval) {
 672                         if (copied)
 673                                 break;
 674                         return -EIO;
 675                 }
 676                 copied += retval;
 677                 src += retval;
 678                 dst += retval;
 679                 len -= retval;
 680         }
 681         return copied;
 682 }
 683
 684 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
 685 {
 686         unsigned flags;
 687         int ret;
 688
 689         ret = check_ptrace_options(data);
 690         if (ret)
 691                 return ret;
 692
 693         /* Avoid intermediate state when all opts are cleared */
 694         flags = child->ptrace;
 695         flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
 696         flags |= (data << PT_OPT_FLAG_SHIFT);
 697         child->ptrace = flags;
 698
 699         return 0;
 700 }
 701
 702 static int ptrace_getsiginfo(struct task_struct *child, kernel_siginfo_t *info)
 703 {
 704         unsigned long flags;
 705         int error = -ESRCH;
 706
 707         if (lock_task_sighand(child, &flags)) {
 708                 error = -EINVAL;
 709                 if (likely(child->last_siginfo != NULL)) {
 710                         copy_siginfo(info, child->last_siginfo);
 711                         error = 0;
 712                 }
 713                 unlock_task_sighand(child, &flags);
 714         }
 715         return error;
 716 }
 717
 718 static int ptrace_setsiginfo(struct task_struct *child, const kernel_siginfo_t *info)
 719 {
 720         unsigned long flags;
 721         int error = -ESRCH;
 722
 723         if (lock_task_sighand(child, &flags)) {
 724                 error = -EINVAL;
 725                 if (likely(child->last_siginfo != NULL)) {
 726                         copy_siginfo(child->last_siginfo, info);
 727                         error = 0;
 728                 }
 729                 unlock_task_sighand(child, &flags);
 730         }
 731         return error;
 732 }
 733
 734 static int ptrace_peek_siginfo(struct task_struct *child,
 735                                 unsigned long addr,
 736                                 unsigned long data)
 737 {
 738         struct ptrace_peeksiginfo_args arg;
 739         struct sigpending *pending;
 740         struct sigqueue *q;
 741         int ret, i;
 742
 743         ret = copy_from_user(&arg, (void __user *) addr,
 744                                 sizeof(struct ptrace_peeksiginfo_args));
 745         if (ret)
 746                 return -EFAULT;
 747
 748         if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
 749                 return -EINVAL; /* unknown flags */
 750
 751         if (arg.nr < 0)
 752                 return -EINVAL;
 753
 754         /* Ensure arg.off fits in an unsigned long */
 755         if (arg.off > ULONG_MAX)
 756                 return 0;
 757
 758         if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
 759                 pending = &child->signal->shared_pending;
 760         else
 761                 pending = &child->pending;
 762
 763         for (i = 0; i < arg.nr; ) {
 764                 kernel_siginfo_t info;
 765                 unsigned long off = arg.off + i;
 766                 bool found = false;
 767
 768                 spin_lock_irq(&child->sighand->siglock);
 769                 list_for_each_entry(q, &pending->list, list) {
 770                         if (!off--) {
 771                                 found = true;
 772                                 copy_siginfo(&info, &q->info);
 773                                 break;
 774                         }
 775                 }
 776                 spin_unlock_irq(&child->sighand->siglock);
 777
 778                 if (!found) /* beyond the end of the list */
 779                         break;
 780
 781 #ifdef CONFIG_COMPAT
 782                 if (unlikely(in_compat_syscall())) {
 783                         compat_siginfo_t __user *uinfo = compat_ptr(data);
 784
 785                         if (copy_siginfo_to_user32(uinfo, &info)) {
 786                                 ret = -EFAULT;
 787                                 break;
 788                         }
 789
 790                 } else
 791 #endif
 792                 {
 793                         siginfo_t __user *uinfo = (siginfo_t __user *) data;
 794
 795                         if (copy_siginfo_to_user(uinfo, &info)) {
 796                                 ret = -EFAULT;
 797                                 break;
 798                         }
 799                 }
 800
 801                 data += sizeof(siginfo_t);
 802                 i++;
 803
 804                 if (signal_pending(current))
 805                         break;
 806
 807                 cond_resched();
 808         }
 809
 810         if (i > 0)
 811                 return i;
 812
 813         return ret;
 814 }
 815
 816 #ifdef CONFIG_RSEQ
 817 static long ptrace_get_rseq_configuration(struct task_struct *task,
 818                                           unsigned long size, void __user *data)
 819 {
 820         struct ptrace_rseq_configuration conf = {
 821                 .rseq_abi_pointer = (u64)(uintptr_t)task->rseq,
 822                 .rseq_abi_size = sizeof(*task->rseq),
 823                 .signature = task->rseq_sig,
 824                 .flags = 0,
 825         };
 826
 827         size = min_t(unsigned long, size, sizeof(conf));
 828         if (copy_to_user(data, &conf, size))
 829                 return -EFAULT;
 830         return sizeof(conf);
 831 }
 832 #endif
 833
 834 #ifdef PTRACE_SINGLESTEP
 835 #define is_singlestep(request)          ((request) == PTRACE_SINGLESTEP)
 836 #else
 837 #define is_singlestep(request)          0
 838 #endif
 839
 840 #ifdef PTRACE_SINGLEBLOCK
 841 #define is_singleblock(request)         ((request) == PTRACE_SINGLEBLOCK)
 842 #else
 843 #define is_singleblock(request)         0
 844 #endif
 845
 846 #ifdef PTRACE_SYSEMU
 847 #define is_sysemu_singlestep(request)   ((request) == PTRACE_SYSEMU_SINGLESTEP)
 848 #else
 849 #define is_sysemu_singlestep(request)   0
 850 #endif
 851
 852 static int ptrace_resume(struct task_struct *child, long request,
 853                          unsigned long data)
 854 {
 855         bool need_siglock;
 856
 857         if (!valid_signal(data))
 858                 return -EIO;
 859
 860         if (request == PTRACE_SYSCALL)
 861                 set_task_syscall_work(child, SYSCALL_TRACE);
 862         else
 863                 clear_task_syscall_work(child, SYSCALL_TRACE);
 864
 865 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
 866         if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
 867                 set_task_syscall_work(child, SYSCALL_EMU);
 868         else
 869                 clear_task_syscall_work(child, SYSCALL_EMU);
 870 #endif
 871
 872         if (is_singleblock(request)) {
 873                 if (unlikely(!arch_has_block_step()))
 874                         return -EIO;
 875                 user_enable_block_step(child);
 876         } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
 877                 if (unlikely(!arch_has_single_step()))
 878                         return -EIO;
 879                 user_enable_single_step(child);
 880         } else {
 881                 user_disable_single_step(child);
 882         }
 883
 884         /*
 885          * Change ->exit_code and ->state under siglock to avoid the race
 886          * with wait_task_stopped() in between; a non-zero ->exit_code will
 887          * wrongly look like another report from tracee.
 888          *
 889          * Note that we need siglock even if ->exit_code == data and/or this
 890          * status was not reported yet, the new status must not be cleared by
 891          * wait_task_stopped() after resume.
 892          *
 893          * If data == 0 we do not care if wait_task_stopped() reports the old
 894          * status and clears the code too; this can't race with the tracee, it
 895          * takes siglock after resume.
 896          */
 897         need_siglock = data && !thread_group_empty(current);
 898         if (need_siglock)
 899                 spin_lock_irq(&child->sighand->siglock);
 900         child->exit_code = data;
 901         wake_up_state(child, __TASK_TRACED);
 902         if (need_siglock)
 903                 spin_unlock_irq(&child->sighand->siglock);
 904
 905         return 0;
 906 }
 907
 908 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 909
 910 static const struct user_regset *
 911 find_regset(const struct user_regset_view *view, unsigned int type)
 912 {
 913         const struct user_regset *regset;
 914         int n;
 915
 916         for (n = 0; n < view->n; ++n) {
 917                 regset = view->regsets + n;
 918                 if (regset->core_note_type == type)
 919                         return regset;
 920         }
 921
 922         return NULL;
 923 }
 924
 925 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
 926                          struct iovec *kiov)
 927 {
 928         const struct user_regset_view *view = task_user_regset_view(task);
 929         const struct user_regset *regset = find_regset(view, type);
 930         int regset_no;
 931
 932         if (!regset || (kiov->iov_len % regset->size) != 0)
 933                 return -EINVAL;
 934
 935         regset_no = regset - view->regsets;
 936         kiov->iov_len = min(kiov->iov_len,
 937                             (__kernel_size_t) (regset->n * regset->size));
 938
 939         if (req == PTRACE_GETREGSET)
 940                 return copy_regset_to_user(task, view, regset_no, 0,
 941                                            kiov->iov_len, kiov->iov_base);
 942         else
 943                 return copy_regset_from_user(task, view, regset_no, 0,
 944                                              kiov->iov_len, kiov->iov_base);
 945 }
 946
 947 /*
 948  * This is declared in linux/regset.h and defined in machine-dependent
 949  * code.  We put the export here, near the primary machine-neutral use,
 950  * to ensure no machine forgets it.
 951  */
 952 EXPORT_SYMBOL_GPL(task_user_regset_view);
 953
 954 static unsigned long
 955 ptrace_get_syscall_info_entry(struct task_struct *child, struct pt_regs *regs,
 956                               struct ptrace_syscall_info *info)
 957 {
 958         unsigned long args[ARRAY_SIZE(info->entry.args)];
 959         int i;
 960
 961         info->op = PTRACE_SYSCALL_INFO_ENTRY;
 962         info->entry.nr = syscall_get_nr(child, regs);
 963         syscall_get_arguments(child, regs, args);
 964         for (i = 0; i < ARRAY_SIZE(args); i++)
 965                 info->entry.args[i] = args[i];
 966
 967         /* args is the last field in struct ptrace_syscall_info.entry */
 968         return offsetofend(struct ptrace_syscall_info, entry.args);
 969 }
 970
 971 static unsigned long
 972 ptrace_get_syscall_info_seccomp(struct task_struct *child, struct pt_regs *regs,
 973                                 struct ptrace_syscall_info *info)
 974 {
 975         /*
 976          * As struct ptrace_syscall_info.entry is currently a subset
 977          * of struct ptrace_syscall_info.seccomp, it makes sense to
 978          * initialize that subset using ptrace_get_syscall_info_entry().
 979          * This can be reconsidered in the future if these structures
 980          * diverge significantly enough.
 981          */
 982         ptrace_get_syscall_info_entry(child, regs, info);
 983         info->op = PTRACE_SYSCALL_INFO_SECCOMP;
 984         info->seccomp.ret_data = child->ptrace_message;
 985
 986         /* ret_data is the last field in struct ptrace_syscall_info.seccomp */
 987         return offsetofend(struct ptrace_syscall_info, seccomp.ret_data);
 988 }
 989
 990 static unsigned long
 991 ptrace_get_syscall_info_exit(struct task_struct *child, struct pt_regs *regs,
 992                              struct ptrace_syscall_info *info)
 993 {
 994         info->op = PTRACE_SYSCALL_INFO_EXIT;
 995         info->exit.rval = syscall_get_error(child, regs);
 996         info->exit.is_error = !!info->exit.rval;
 997         if (!info->exit.is_error)
 998                 info->exit.rval = syscall_get_return_value(child, regs);
 999
1000         /* is_error is the last field in struct ptrace_syscall_info.exit */
1001         return offsetofend(struct ptrace_syscall_info, exit.is_error);
1002 }
1003
1004 static int
1005 ptrace_get_syscall_info(struct task_struct *child, unsigned long user_size,
1006                         void __user *datavp)
1007 {
1008         struct pt_regs *regs = task_pt_regs(child);
1009         struct ptrace_syscall_info info = {
1010                 .op = PTRACE_SYSCALL_INFO_NONE,
1011                 .arch = syscall_get_arch(child),
1012                 .instruction_pointer = instruction_pointer(regs),
1013                 .stack_pointer = user_stack_pointer(regs),
1014         };
1015         unsigned long actual_size = offsetof(struct ptrace_syscall_info, entry);
1016         unsigned long write_size;
1017
1018         /*
1019          * This does not need lock_task_sighand() to access
1020          * child->last_siginfo because ptrace_freeze_traced()
1021          * called earlier by ptrace_check_attach() ensures that
1022          * the tracee cannot go away and clear its last_siginfo.
1023          */
1024         switch (child->last_siginfo ? child->last_siginfo->si_code : 0) {
1025         case SIGTRAP | 0x80:
1026                 switch (child->ptrace_message) {
1027                 case PTRACE_EVENTMSG_SYSCALL_ENTRY:
1028                         actual_size = ptrace_get_syscall_info_entry(child, regs,
1029                                                                     &info);
1030                         break;
1031                 case PTRACE_EVENTMSG_SYSCALL_EXIT:
1032                         actual_size = ptrace_get_syscall_info_exit(child, regs,
1033                                                                    &info);
1034                         break;
1035                 }
1036                 break;
1037         case SIGTRAP | (PTRACE_EVENT_SECCOMP << 8):
1038                 actual_size = ptrace_get_syscall_info_seccomp(child, regs,
1039                                                               &info);
1040                 break;
1041         }
1042
1043         write_size = min(actual_size, user_size);
1044         return copy_to_user(datavp, &info, write_size) ? -EFAULT : actual_size;
1045 }
1046 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
1047
1048 int ptrace_request(struct task_struct *child, long request,
1049                    unsigned long addr, unsigned long data)
1050 {
1051         bool seized = child->ptrace & PT_SEIZED;
1052         int ret = -EIO;
1053         kernel_siginfo_t siginfo, *si;
1054         void __user *datavp = (void __user *) data;
1055         unsigned long __user *datalp = datavp;
1056         unsigned long flags;
1057
1058         switch (request) {
1059         case PTRACE_PEEKTEXT:
1060         case PTRACE_PEEKDATA:
1061                 return generic_ptrace_peekdata(child, addr, data);
1062         case PTRACE_POKETEXT:
1063         case PTRACE_POKEDATA:
1064                 return generic_ptrace_pokedata(child, addr, data);
1065
1066 #ifdef PTRACE_OLDSETOPTIONS
1067         case PTRACE_OLDSETOPTIONS:
1068 #endif
1069         case PTRACE_SETOPTIONS:
1070                 ret = ptrace_setoptions(child, data);
1071                 break;
1072         case PTRACE_GETEVENTMSG:
1073                 ret = put_user(child->ptrace_message, datalp);
1074                 break;
1075
1076         case PTRACE_PEEKSIGINFO:
1077                 ret = ptrace_peek_siginfo(child, addr, data);
1078                 break;
1079
1080         case PTRACE_GETSIGINFO:
1081                 ret = ptrace_getsiginfo(child, &siginfo);
1082                 if (!ret)
1083                         ret = copy_siginfo_to_user(datavp, &siginfo);
1084                 break;
1085
1086         case PTRACE_SETSIGINFO:
1087                 ret = copy_siginfo_from_user(&siginfo, datavp);
1088                 if (!ret)
1089                         ret = ptrace_setsiginfo(child, &siginfo);
1090                 break;
1091
1092         case PTRACE_GETSIGMASK: {
1093                 sigset_t *mask;
1094
1095                 if (addr != sizeof(sigset_t)) {
1096                         ret = -EINVAL;
1097                         break;
1098                 }
1099
1100                 if (test_tsk_restore_sigmask(child))
1101                         mask = &child->saved_sigmask;
1102                 else
1103                         mask = &child->blocked;
1104
1105                 if (copy_to_user(datavp, mask, sizeof(sigset_t)))
1106                         ret = -EFAULT;
1107                 else
1108                         ret = 0;
1109
1110                 break;
1111         }
1112
1113         case PTRACE_SETSIGMASK: {
1114                 sigset_t new_set;
1115
1116                 if (addr != sizeof(sigset_t)) {
1117                         ret = -EINVAL;
1118                         break;
1119                 }
1120
1121                 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
1122                         ret = -EFAULT;
1123                         break;
1124                 }
1125
1126                 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
1127
1128                 /*
1129                  * Every thread does recalc_sigpending() after resume, so
1130                  * retarget_shared_pending() and recalc_sigpending() are not
1131                  * called here.
1132                  */
1133                 spin_lock_irq(&child->sighand->siglock);
1134                 child->blocked = new_set;
1135                 spin_unlock_irq(&child->sighand->siglock);
1136
1137                 clear_tsk_restore_sigmask(child);
1138
1139                 ret = 0;
1140                 break;
1141         }
1142
1143         case PTRACE_INTERRUPT:
1144                 /*
1145                  * Stop tracee without any side-effect on signal or job
1146                  * control.  At least one trap is guaranteed to happen
1147                  * after this request.  If @child is already trapped, the
1148                  * current trap is not disturbed and another trap will
1149                  * happen after the current trap is ended with PTRACE_CONT.
1150                  *
1151                  * The actual trap might not be PTRACE_EVENT_STOP trap but
1152                  * the pending condition is cleared regardless.
1153                  */
1154                 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1155                         break;
1156
1157                 /*
1158                  * INTERRUPT doesn't disturb existing trap sans one
1159                  * exception.  If ptracer issued LISTEN for the current
1160                  * STOP, this INTERRUPT should clear LISTEN and re-trap
1161                  * tracee into STOP.
1162                  */
1163                 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
1164                         ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1165
1166                 unlock_task_sighand(child, &flags);
1167                 ret = 0;
1168                 break;
1169
1170         case PTRACE_LISTEN:
1171                 /*
1172                  * Listen for events.  Tracee must be in STOP.  It's not
1173                  * resumed per-se but is not considered to be in TRACED by
1174                  * wait(2) or ptrace(2).  If an async event (e.g. group
1175                  * stop state change) happens, tracee will enter STOP trap
1176                  * again.  Alternatively, ptracer can issue INTERRUPT to
1177                  * finish listening and re-trap tracee into STOP.
1178                  */
1179                 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1180                         break;
1181
1182                 si = child->last_siginfo;
1183                 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1184                         child->jobctl |= JOBCTL_LISTENING;
1185                         /*
1186                          * If NOTIFY is set, it means event happened between
1187                          * start of this trap and now.  Trigger re-trap.
1188                          */
1189                         if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1190                                 ptrace_signal_wake_up(child, true);
1191                         ret = 0;
1192                 }
1193                 unlock_task_sighand(child, &flags);
1194                 break;
1195
1196         case PTRACE_DETACH:      /* detach a process that was attached. */
1197                 ret = ptrace_detach(child, data);
1198                 break;
1199
1200 #ifdef CONFIG_BINFMT_ELF_FDPIC
1201         case PTRACE_GETFDPIC: {
1202                 struct mm_struct *mm = get_task_mm(child);
1203                 unsigned long tmp = 0;
1204
1205                 ret = -ESRCH;
1206                 if (!mm)
1207                         break;
1208
1209                 switch (addr) {
1210                 case PTRACE_GETFDPIC_EXEC:
1211                         tmp = mm->context.exec_fdpic_loadmap;
1212                         break;
1213                 case PTRACE_GETFDPIC_INTERP:
1214                         tmp = mm->context.interp_fdpic_loadmap;
1215                         break;
1216                 default:
1217                         break;
1218                 }
1219                 mmput(mm);
1220
1221                 ret = put_user(tmp, datalp);
1222                 break;
1223         }
1224 #endif
1225
1226 #ifdef PTRACE_SINGLESTEP
1227         case PTRACE_SINGLESTEP:
1228 #endif
1229 #ifdef PTRACE_SINGLEBLOCK
1230         case PTRACE_SINGLEBLOCK:
1231 #endif
1232 #ifdef PTRACE_SYSEMU
1233         case PTRACE_SYSEMU:
1234         case PTRACE_SYSEMU_SINGLESTEP:
1235 #endif
1236         case PTRACE_SYSCALL:
1237         case PTRACE_CONT:
1238                 return ptrace_resume(child, request, data);
1239
1240         case PTRACE_KILL:
1241                 send_sig_info(SIGKILL, SEND_SIG_NOINFO, child);
1242                 return 0;
1243
1244 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1245         case PTRACE_GETREGSET:
1246         case PTRACE_SETREGSET: {
1247                 struct iovec kiov;
1248                 struct iovec __user *uiov = datavp;
1249
1250                 if (!access_ok(uiov, sizeof(*uiov)))
1251                         return -EFAULT;
1252
1253                 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1254                     __get_user(kiov.iov_len, &uiov->iov_len))
1255                         return -EFAULT;
1256
1257                 ret = ptrace_regset(child, request, addr, &kiov);
1258                 if (!ret)
1259                         ret = __put_user(kiov.iov_len, &uiov->iov_len);
1260                 break;
1261         }
1262
1263         case PTRACE_GET_SYSCALL_INFO:
1264                 ret = ptrace_get_syscall_info(child, addr, datavp);
1265                 break;
1266 #endif
1267
1268         case PTRACE_SECCOMP_GET_FILTER:
1269                 ret = seccomp_get_filter(child, addr, datavp);
1270                 break;
1271
1272         case PTRACE_SECCOMP_GET_METADATA:
1273                 ret = seccomp_get_metadata(child, addr, datavp);
1274                 break;
1275
1276 #ifdef CONFIG_RSEQ
1277         case PTRACE_GET_RSEQ_CONFIGURATION:
1278                 ret = ptrace_get_rseq_configuration(child, addr, datavp);
1279                 break;
1280 #endif
1281
1282         default:
1283                 break;
1284         }
1285
1286         return ret;
1287 }
1288
1289 #ifndef arch_ptrace_attach
1290 #define arch_ptrace_attach(child)       do { } while (0)
1291 #endif
1292
1293 SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1294                 unsigned long, data)
1295 {
1296         struct task_struct *child;
1297         long ret;
1298
1299         if (request == PTRACE_TRACEME) {
1300                 ret = ptrace_traceme();
1301                 if (!ret)
1302                         arch_ptrace_attach(current);
1303                 goto out;
1304         }
1305
1306         child = find_get_task_by_vpid(pid);
1307         if (!child) {
1308                 ret = -ESRCH;
1309                 goto out;
1310         }
1311
1312         if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1313                 ret = ptrace_attach(child, request, addr, data);
1314                 /*
1315                  * Some architectures need to do book-keeping after
1316                  * a ptrace attach.
1317                  */
1318                 if (!ret)
1319                         arch_ptrace_attach(child);
1320                 goto out_put_task_struct;
1321         }
1322
1323         ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1324                                   request == PTRACE_INTERRUPT);
1325         if (ret < 0)
1326                 goto out_put_task_struct;
1327
1328         ret = arch_ptrace(child, request, addr, data);
1329         if (ret || request != PTRACE_DETACH)
1330                 ptrace_unfreeze_traced(child);
1331
1332  out_put_task_struct:
1333         put_task_struct(child);
1334  out:
1335         return ret;
1336 }
1337
1338 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1339                             unsigned long data)
1340 {
1341         unsigned long tmp;
1342         int copied;
1343
1344         copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1345         if (copied != sizeof(tmp))
1346                 return -EIO;
1347         return put_user(tmp, (unsigned long __user *)data);
1348 }
1349
1350 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1351                             unsigned long data)
1352 {
1353         int copied;
1354
1355         copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1356                         FOLL_FORCE | FOLL_WRITE);
1357         return (copied == sizeof(data)) ? 0 : -EIO;
1358 }
1359
1360 #if defined CONFIG_COMPAT
1361
1362 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1363                           compat_ulong_t addr, compat_ulong_t data)
1364 {
1365         compat_ulong_t __user *datap = compat_ptr(data);
1366         compat_ulong_t word;
1367         kernel_siginfo_t siginfo;
1368         int ret;
1369
1370         switch (request) {
1371         case PTRACE_PEEKTEXT:
1372         case PTRACE_PEEKDATA:
1373                 ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1374                                 FOLL_FORCE);
1375                 if (ret != sizeof(word))
1376                         ret = -EIO;
1377                 else
1378                         ret = put_user(word, datap);
1379                 break;
1380
1381         case PTRACE_POKETEXT:
1382         case PTRACE_POKEDATA:
1383                 ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1384                                 FOLL_FORCE | FOLL_WRITE);
1385                 ret = (ret != sizeof(data) ? -EIO : 0);
1386                 break;
1387
1388         case PTRACE_GETEVENTMSG:
1389                 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1390                 break;
1391
1392         case PTRACE_GETSIGINFO:
1393                 ret = ptrace_getsiginfo(child, &siginfo);
1394                 if (!ret)
1395                         ret = copy_siginfo_to_user32(
1396                                 (struct compat_siginfo __user *) datap,
1397                                 &siginfo);
1398                 break;
1399
1400         case PTRACE_SETSIGINFO:
1401                 ret = copy_siginfo_from_user32(
1402                         &siginfo, (struct compat_siginfo __user *) datap);
1403                 if (!ret)
1404                         ret = ptrace_setsiginfo(child, &siginfo);
1405                 break;
1406 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1407         case PTRACE_GETREGSET:
1408         case PTRACE_SETREGSET:
1409         {
1410                 struct iovec kiov;
1411                 struct compat_iovec __user *uiov =
1412                         (struct compat_iovec __user *) datap;
1413                 compat_uptr_t ptr;
1414                 compat_size_t len;
1415
1416                 if (!access_ok(uiov, sizeof(*uiov)))
1417                         return -EFAULT;
1418
1419                 if (__get_user(ptr, &uiov->iov_base) ||
1420                     __get_user(len, &uiov->iov_len))
1421                         return -EFAULT;
1422
1423                 kiov.iov_base = compat_ptr(ptr);
1424                 kiov.iov_len = len;
1425
1426                 ret = ptrace_regset(child, request, addr, &kiov);
1427                 if (!ret)
1428                         ret = __put_user(kiov.iov_len, &uiov->iov_len);
1429                 break;
1430         }
1431 #endif
1432
1433         default:
1434                 ret = ptrace_request(child, request, addr, data);
1435         }
1436
1437         return ret;
1438 }
1439
1440 COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1441                        compat_long_t, addr, compat_long_t, data)
1442 {
1443         struct task_struct *child;
1444         long ret;
1445
1446         if (request == PTRACE_TRACEME) {
1447                 ret = ptrace_traceme();
1448                 goto out;
1449         }
1450
1451         child = find_get_task_by_vpid(pid);
1452         if (!child) {
1453                 ret = -ESRCH;
1454                 goto out;
1455         }
1456
1457         if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1458                 ret = ptrace_attach(child, request, addr, data);
1459                 /*
1460                  * Some architectures need to do book-keeping after
1461                  * a ptrace attach.
1462                  */
1463                 if (!ret)
1464                         arch_ptrace_attach(child);
1465                 goto out_put_task_struct;
1466         }
1467
1468         ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1469                                   request == PTRACE_INTERRUPT);
1470         if (!ret) {
1471                 ret = compat_arch_ptrace(child, request, addr, data);
1472                 if (ret || request != PTRACE_DETACH)
1473                         ptrace_unfreeze_traced(child);
1474         }
1475
1476  out_put_task_struct:
1477         put_task_struct(child);
1478  out:
1479         return ret;
1480 }
1481 #endif  /* CONFIG_COMPAT */