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