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 #ifdef CONFIG_PREEMPT_RT
 201                 unsigned long flags;
 202
 203                 raw_spin_lock_irqsave(&task->pi_lock, flags);
 204                 if (READ_ONCE(task->__state) & __TASK_TRACED)
 205                         WRITE_ONCE(task->__state, __TASK_TRACED);
 206                 else
 207                         task->saved_state = __TASK_TRACED;
 208                 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
 209 #else
 210                 WRITE_ONCE(task->__state, __TASK_TRACED);
 211 #endif
 212                 ret = true;
 213         }
 214         spin_unlock_irq(&task->sighand->siglock);
 215
 216         return ret;
 217 }
 218
 219 static void ptrace_unfreeze_traced(struct task_struct *task)
 220 {
 221         unsigned long flags;
 222         bool frozen = true;
 223
 224         if (!IS_ENABLED(CONFIG_PREEMPT_RT) &&
 225             READ_ONCE(task->__state) != __TASK_TRACED)
 226                 return;
 227
 228         WARN_ON(!task->ptrace || task->parent != current);
 229
 230         /*
 231          * PTRACE_LISTEN can allow ptrace_trap_notify to wake us up remotely.
 232          * Recheck state under the lock to close this race.
 233          */
 234         spin_lock_irq(&task->sighand->siglock);
 235         raw_spin_lock_irqsave(&task->pi_lock, flags);
 236         if (READ_ONCE(task->__state) == __TASK_TRACED)
 237                 WRITE_ONCE(task->__state, TASK_TRACED);
 238
 239 #ifdef CONFIG_PREEMPT_RT
 240         else if (task->saved_state == __TASK_TRACED)
 241                 task->saved_state = TASK_TRACED;
 242 #endif
 243         else
 244                 frozen = false;
 245         raw_spin_unlock_irqrestore(&task->pi_lock, flags);
 246
 247         if (frozen && __fatal_signal_pending(task))
 248                 wake_up_state(task, __TASK_TRACED);
 249
 250         spin_unlock_irq(&task->sighand->siglock);
 251 }
 252
 253 /**
 254  * ptrace_check_attach - check whether ptracee is ready for ptrace operation
 255  * @child: ptracee to check for
 256  * @ignore_state: don't check whether @child is currently %TASK_TRACED
 257  *
 258  * Check whether @child is being ptraced by %current and ready for further
 259  * ptrace operations.  If @ignore_state is %false, @child also should be in
 260  * %TASK_TRACED state and on return the child is guaranteed to be traced
 261  * and not executing.  If @ignore_state is %true, @child can be in any
 262  * state.
 263  *
 264  * CONTEXT:
 265  * Grabs and releases tasklist_lock and @child->sighand->siglock.
 266  *
 267  * RETURNS:
 268  * 0 on success, -ESRCH if %child is not ready.
 269  */
 270 static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
 271 {
 272         int ret = -ESRCH;
 273
 274         /*
 275          * We take the read lock around doing both checks to close a
 276          * possible race where someone else was tracing our child and
 277          * detached between these two checks.  After this locked check,
 278          * we are sure that this is our traced child and that can only
 279          * be changed by us so it's not changing right after this.
 280          */
 281         read_lock(&tasklist_lock);
 282         if (child->ptrace && child->parent == current) {
 283                 WARN_ON(READ_ONCE(child->__state) == __TASK_TRACED);
 284                 /*
 285                  * child->sighand can't be NULL, release_task()
 286                  * does ptrace_unlink() before __exit_signal().
 287                  */
 288                 if (ignore_state || ptrace_freeze_traced(child))
 289                         ret = 0;
 290         }
 291         read_unlock(&tasklist_lock);
 292
 293         if (!ret && !ignore_state) {
 294                 if (!wait_task_inactive(child, __TASK_TRACED)) {
 295                         /*
 296                          * This can only happen if may_ptrace_stop() fails and
 297                          * ptrace_stop() changes ->state back to TASK_RUNNING,
 298                          * so we should not worry about leaking __TASK_TRACED.
 299                          */
 300                         WARN_ON(READ_ONCE(child->__state) == __TASK_TRACED);
 301                         ret = -ESRCH;
 302                 }
 303         }
 304
 305         return ret;
 306 }
 307
 308 static bool ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
 309 {
 310         if (mode & PTRACE_MODE_NOAUDIT)
 311                 return ns_capable_noaudit(ns, CAP_SYS_PTRACE);
 312         return ns_capable(ns, CAP_SYS_PTRACE);
 313 }
 314
 315 /* Returns 0 on success, -errno on denial. */
 316 static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
 317 {
 318         const struct cred *cred = current_cred(), *tcred;
 319         struct mm_struct *mm;
 320         kuid_t caller_uid;
 321         kgid_t caller_gid;
 322
 323         if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
 324                 WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
 325                 return -EPERM;
 326         }
 327
 328         /* May we inspect the given task?
 329          * This check is used both for attaching with ptrace
 330          * and for allowing access to sensitive information in /proc.
 331          *
 332          * ptrace_attach denies several cases that /proc allows
 333          * because setting up the necessary parent/child relationship
 334          * or halting the specified task is impossible.
 335          */
 336
 337         /* Don't let security modules deny introspection */
 338         if (same_thread_group(task, current))
 339                 return 0;
 340         rcu_read_lock();
 341         if (mode & PTRACE_MODE_FSCREDS) {
 342                 caller_uid = cred->fsuid;
 343                 caller_gid = cred->fsgid;
 344         } else {
 345                 /*
 346                  * Using the euid would make more sense here, but something
 347                  * in userland might rely on the old behavior, and this
 348                  * shouldn't be a security problem since
 349                  * PTRACE_MODE_REALCREDS implies that the caller explicitly
 350                  * used a syscall that requests access to another process
 351                  * (and not a filesystem syscall to procfs).
 352                  */
 353                 caller_uid = cred->uid;
 354                 caller_gid = cred->gid;
 355         }
 356         tcred = __task_cred(task);
 357         if (uid_eq(caller_uid, tcred->euid) &&
 358             uid_eq(caller_uid, tcred->suid) &&
 359             uid_eq(caller_uid, tcred->uid)  &&
 360             gid_eq(caller_gid, tcred->egid) &&
 361             gid_eq(caller_gid, tcred->sgid) &&
 362             gid_eq(caller_gid, tcred->gid))
 363                 goto ok;
 364         if (ptrace_has_cap(tcred->user_ns, mode))
 365                 goto ok;
 366         rcu_read_unlock();
 367         return -EPERM;
 368 ok:
 369         rcu_read_unlock();
 370         /*
 371          * If a task drops privileges and becomes nondumpable (through a syscall
 372          * like setresuid()) while we are trying to access it, we must ensure
 373          * that the dumpability is read after the credentials; otherwise,
 374          * we may be able to attach to a task that we shouldn't be able to
 375          * attach to (as if the task had dropped privileges without becoming
 376          * nondumpable).
 377          * Pairs with a write barrier in commit_creds().
 378          */
 379         smp_rmb();
 380         mm = task->mm;
 381         if (mm &&
 382             ((get_dumpable(mm) != SUID_DUMP_USER) &&
 383              !ptrace_has_cap(mm->user_ns, mode)))
 384             return -EPERM;
 385
 386         return security_ptrace_access_check(task, mode);
 387 }
 388
 389 bool ptrace_may_access(struct task_struct *task, unsigned int mode)
 390 {
 391         int err;
 392         task_lock(task);
 393         err = __ptrace_may_access(task, mode);
 394         task_unlock(task);
 395         return !err;
 396 }
 397
 398 static int check_ptrace_options(unsigned long data)
 399 {
 400         if (data & ~(unsigned long)PTRACE_O_MASK)
 401                 return -EINVAL;
 402
 403         if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
 404                 if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
 405                     !IS_ENABLED(CONFIG_SECCOMP))
 406                         return -EINVAL;
 407
 408                 if (!capable(CAP_SYS_ADMIN))
 409                         return -EPERM;
 410
 411                 if (seccomp_mode(&current->seccomp) != SECCOMP_MODE_DISABLED ||
 412                     current->ptrace & PT_SUSPEND_SECCOMP)
 413                         return -EPERM;
 414         }
 415         return 0;
 416 }
 417
 418 static int ptrace_attach(struct task_struct *task, long request,
 419                          unsigned long addr,
 420                          unsigned long flags)
 421 {
 422         bool seize = (request == PTRACE_SEIZE);
 423         int retval;
 424
 425         retval = -EIO;
 426         if (seize) {
 427                 if (addr != 0)
 428                         goto out;
 429                 /*
 430                  * This duplicates the check in check_ptrace_options() because
 431                  * ptrace_attach() and ptrace_setoptions() have historically
 432                  * used different error codes for unknown ptrace options.
 433                  */
 434                 if (flags & ~(unsigned long)PTRACE_O_MASK)
 435                         goto out;
 436                 retval = check_ptrace_options(flags);
 437                 if (retval)
 438                         return retval;
 439                 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
 440         } else {
 441                 flags = PT_PTRACED;
 442         }
 443
 444         audit_ptrace(task);
 445
 446         retval = -EPERM;
 447         if (unlikely(task->flags & PF_KTHREAD))
 448                 goto out;
 449         if (same_thread_group(task, current))
 450                 goto out;
 451
 452         /*
 453          * Protect exec's credential calculations against our interference;
 454          * SUID, SGID and LSM creds get determined differently
 455          * under ptrace.
 456          */
 457         retval = -ERESTARTNOINTR;
 458         if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
 459                 goto out;
 460
 461         task_lock(task);
 462         retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
 463         task_unlock(task);
 464         if (retval)
 465                 goto unlock_creds;
 466
 467         write_lock_irq(&tasklist_lock);
 468         retval = -EPERM;
 469         if (unlikely(task->exit_state))
 470                 goto unlock_tasklist;
 471         if (task->ptrace)
 472                 goto unlock_tasklist;
 473
 474         if (seize)
 475                 flags |= PT_SEIZED;
 476         task->ptrace = flags;
 477
 478         ptrace_link(task, current);
 479
 480         /* SEIZE doesn't trap tracee on attach */
 481         if (!seize)
 482                 send_sig_info(SIGSTOP, SEND_SIG_PRIV, task);
 483
 484         spin_lock(&task->sighand->siglock);
 485
 486         /*
 487          * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
 488          * TRAPPING, and kick it so that it transits to TRACED.  TRAPPING
 489          * will be cleared if the child completes the transition or any
 490          * event which clears the group stop states happens.  We'll wait
 491          * for the transition to complete before returning from this
 492          * function.
 493          *
 494          * This hides STOPPED -> RUNNING -> TRACED transition from the
 495          * attaching thread but a different thread in the same group can
 496          * still observe the transient RUNNING state.  IOW, if another
 497          * thread's WNOHANG wait(2) on the stopped tracee races against
 498          * ATTACH, the wait(2) may fail due to the transient RUNNING.
 499          *
 500          * The following task_is_stopped() test is safe as both transitions
 501          * in and out of STOPPED are protected by siglock.
 502          */
 503         if (task_is_stopped(task) &&
 504             task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
 505                 signal_wake_up_state(task, __TASK_STOPPED);
 506
 507         spin_unlock(&task->sighand->siglock);
 508
 509         retval = 0;
 510 unlock_tasklist:
 511         write_unlock_irq(&tasklist_lock);
 512 unlock_creds:
 513         mutex_unlock(&task->signal->cred_guard_mutex);
 514 out:
 515         if (!retval) {
 516                 /*
 517                  * We do not bother to change retval or clear JOBCTL_TRAPPING
 518                  * if wait_on_bit() was interrupted by SIGKILL. The tracer will
 519                  * not return to user-mode, it will exit and clear this bit in
 520                  * __ptrace_unlink() if it wasn't already cleared by the tracee;
 521                  * and until then nobody can ptrace this task.
 522                  */
 523                 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
 524                 proc_ptrace_connector(task, PTRACE_ATTACH);
 525         }
 526
 527         return retval;
 528 }
 529
 530 /**
 531  * ptrace_traceme  --  helper for PTRACE_TRACEME
 532  *
 533  * Performs checks and sets PT_PTRACED.
 534  * Should be used by all ptrace implementations for PTRACE_TRACEME.
 535  */
 536 static int ptrace_traceme(void)
 537 {
 538         int ret = -EPERM;
 539
 540         write_lock_irq(&tasklist_lock);
 541         /* Are we already being traced? */
 542         if (!current->ptrace) {
 543                 ret = security_ptrace_traceme(current->parent);
 544                 /*
 545                  * Check PF_EXITING to ensure ->real_parent has not passed
 546                  * exit_ptrace(). Otherwise we don't report the error but
 547                  * pretend ->real_parent untraces us right after return.
 548                  */
 549                 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
 550                         current->ptrace = PT_PTRACED;
 551                         ptrace_link(current, current->real_parent);
 552                 }
 553         }
 554         write_unlock_irq(&tasklist_lock);
 555
 556         return ret;
 557 }
 558
 559 /*
 560  * Called with irqs disabled, returns true if childs should reap themselves.
 561  */
 562 static int ignoring_children(struct sighand_struct *sigh)
 563 {
 564         int ret;
 565         spin_lock(&sigh->siglock);
 566         ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
 567               (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
 568         spin_unlock(&sigh->siglock);
 569         return ret;
 570 }
 571
 572 /*
 573  * Called with tasklist_lock held for writing.
 574  * Unlink a traced task, and clean it up if it was a traced zombie.
 575  * Return true if it needs to be reaped with release_task().
 576  * (We can't call release_task() here because we already hold tasklist_lock.)
 577  *
 578  * If it's a zombie, our attachedness prevented normal parent notification
 579  * or self-reaping.  Do notification now if it would have happened earlier.
 580  * If it should reap itself, return true.
 581  *
 582  * If it's our own child, there is no notification to do. But if our normal
 583  * children self-reap, then this child was prevented by ptrace and we must
 584  * reap it now, in that case we must also wake up sub-threads sleeping in
 585  * do_wait().
 586  */
 587 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
 588 {
 589         bool dead;
 590
 591         __ptrace_unlink(p);
 592
 593         if (p->exit_state != EXIT_ZOMBIE)
 594                 return false;
 595
 596         dead = !thread_group_leader(p);
 597
 598         if (!dead && thread_group_empty(p)) {
 599                 if (!same_thread_group(p->real_parent, tracer))
 600                         dead = do_notify_parent(p, p->exit_signal);
 601                 else if (ignoring_children(tracer->sighand)) {
 602                         __wake_up_parent(p, tracer);
 603                         dead = true;
 604                 }
 605         }
 606         /* Mark it as in the process of being reaped. */
 607         if (dead)
 608                 p->exit_state = EXIT_DEAD;
 609         return dead;
 610 }
 611
 612 static int ptrace_detach(struct task_struct *child, unsigned int data)
 613 {
 614         if (!valid_signal(data))
 615                 return -EIO;
 616
 617         /* Architecture-specific hardware disable .. */
 618         ptrace_disable(child);
 619
 620         write_lock_irq(&tasklist_lock);
 621         /*
 622          * We rely on ptrace_freeze_traced(). It can't be killed and
 623          * untraced by another thread, it can't be a zombie.
 624          */
 625         WARN_ON(!child->ptrace || child->exit_state);
 626         /*
 627          * tasklist_lock avoids the race with wait_task_stopped(), see
 628          * the comment in ptrace_resume().
 629          */
 630         child->exit_code = data;
 631         __ptrace_detach(current, child);
 632         write_unlock_irq(&tasklist_lock);
 633
 634         proc_ptrace_connector(child, PTRACE_DETACH);
 635
 636         return 0;
 637 }
 638
 639 /*
 640  * Detach all tasks we were using ptrace on. Called with tasklist held
 641  * for writing.
 642  */
 643 void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
 644 {
 645         struct task_struct *p, *n;
 646
 647         list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
 648                 if (unlikely(p->ptrace & PT_EXITKILL))
 649                         send_sig_info(SIGKILL, SEND_SIG_PRIV, p);
 650
 651                 if (__ptrace_detach(tracer, p))
 652                         list_add(&p->ptrace_entry, dead);
 653         }
 654 }
 655
 656 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *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                 retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
 666
 667                 if (!retval) {
 668                         if (copied)
 669                                 break;
 670                         return -EIO;
 671                 }
 672                 if (copy_to_user(dst, buf, retval))
 673                         return -EFAULT;
 674                 copied += retval;
 675                 src += retval;
 676                 dst += retval;
 677                 len -= retval;
 678         }
 679         return copied;
 680 }
 681
 682 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
 683 {
 684         int copied = 0;
 685
 686         while (len > 0) {
 687                 char buf[128];
 688                 int this_len, retval;
 689
 690                 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 691                 if (copy_from_user(buf, src, this_len))
 692                         return -EFAULT;
 693                 retval = ptrace_access_vm(tsk, dst, buf, this_len,
 694                                 FOLL_FORCE | FOLL_WRITE);
 695                 if (!retval) {
 696                         if (copied)
 697                                 break;
 698                         return -EIO;
 699                 }
 700                 copied += retval;
 701                 src += retval;
 702                 dst += retval;
 703                 len -= retval;
 704         }
 705         return copied;
 706 }
 707
 708 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
 709 {
 710         unsigned flags;
 711         int ret;
 712
 713         ret = check_ptrace_options(data);
 714         if (ret)
 715                 return ret;
 716
 717         /* Avoid intermediate state when all opts are cleared */
 718         flags = child->ptrace;
 719         flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
 720         flags |= (data << PT_OPT_FLAG_SHIFT);
 721         child->ptrace = flags;
 722
 723         return 0;
 724 }
 725
 726 static int ptrace_getsiginfo(struct task_struct *child, kernel_siginfo_t *info)
 727 {
 728         unsigned long flags;
 729         int error = -ESRCH;
 730
 731         if (lock_task_sighand(child, &flags)) {
 732                 error = -EINVAL;
 733                 if (likely(child->last_siginfo != NULL)) {
 734                         copy_siginfo(info, child->last_siginfo);
 735                         error = 0;
 736                 }
 737                 unlock_task_sighand(child, &flags);
 738         }
 739         return error;
 740 }
 741
 742 static int ptrace_setsiginfo(struct task_struct *child, const kernel_siginfo_t *info)
 743 {
 744         unsigned long flags;
 745         int error = -ESRCH;
 746
 747         if (lock_task_sighand(child, &flags)) {
 748                 error = -EINVAL;
 749                 if (likely(child->last_siginfo != NULL)) {
 750                         copy_siginfo(child->last_siginfo, info);
 751                         error = 0;
 752                 }
 753                 unlock_task_sighand(child, &flags);
 754         }
 755         return error;
 756 }
 757
 758 static int ptrace_peek_siginfo(struct task_struct *child,
 759                                 unsigned long addr,
 760                                 unsigned long data)
 761 {
 762         struct ptrace_peeksiginfo_args arg;
 763         struct sigpending *pending;
 764         struct sigqueue *q;
 765         int ret, i;
 766
 767         ret = copy_from_user(&arg, (void __user *) addr,
 768                                 sizeof(struct ptrace_peeksiginfo_args));
 769         if (ret)
 770                 return -EFAULT;
 771
 772         if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
 773                 return -EINVAL; /* unknown flags */
 774
 775         if (arg.nr < 0)
 776                 return -EINVAL;
 777
 778         /* Ensure arg.off fits in an unsigned long */
 779         if (arg.off > ULONG_MAX)
 780                 return 0;
 781
 782         if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
 783                 pending = &child->signal->shared_pending;
 784         else
 785                 pending = &child->pending;
 786
 787         for (i = 0; i < arg.nr; ) {
 788                 kernel_siginfo_t info;
 789                 unsigned long off = arg.off + i;
 790                 bool found = false;
 791
 792                 spin_lock_irq(&child->sighand->siglock);
 793                 list_for_each_entry(q, &pending->list, list) {
 794                         if (!off--) {
 795                                 found = true;
 796                                 copy_siginfo(&info, &q->info);
 797                                 break;
 798                         }
 799                 }
 800                 spin_unlock_irq(&child->sighand->siglock);
 801
 802                 if (!found) /* beyond the end of the list */
 803                         break;
 804
 805 #ifdef CONFIG_COMPAT
 806                 if (unlikely(in_compat_syscall())) {
 807                         compat_siginfo_t __user *uinfo = compat_ptr(data);
 808
 809                         if (copy_siginfo_to_user32(uinfo, &info)) {
 810                                 ret = -EFAULT;
 811                                 break;
 812                         }
 813
 814                 } else
 815 #endif
 816                 {
 817                         siginfo_t __user *uinfo = (siginfo_t __user *) data;
 818
 819                         if (copy_siginfo_to_user(uinfo, &info)) {
 820                                 ret = -EFAULT;
 821                                 break;
 822                         }
 823                 }
 824
 825                 data += sizeof(siginfo_t);
 826                 i++;
 827
 828                 if (signal_pending(current))
 829                         break;
 830
 831                 cond_resched();
 832         }
 833
 834         if (i > 0)
 835                 return i;
 836
 837         return ret;
 838 }
 839
 840 #ifdef CONFIG_RSEQ
 841 static long ptrace_get_rseq_configuration(struct task_struct *task,
 842                                           unsigned long size, void __user *data)
 843 {
 844         struct ptrace_rseq_configuration conf = {
 845                 .rseq_abi_pointer = (u64)(uintptr_t)task->rseq,
 846                 .rseq_abi_size = sizeof(*task->rseq),
 847                 .signature = task->rseq_sig,
 848                 .flags = 0,
 849         };
 850
 851         size = min_t(unsigned long, size, sizeof(conf));
 852         if (copy_to_user(data, &conf, size))
 853                 return -EFAULT;
 854         return sizeof(conf);
 855 }
 856 #endif
 857
 858 #ifdef PTRACE_SINGLESTEP
 859 #define is_singlestep(request)          ((request) == PTRACE_SINGLESTEP)
 860 #else
 861 #define is_singlestep(request)          0
 862 #endif
 863
 864 #ifdef PTRACE_SINGLEBLOCK
 865 #define is_singleblock(request)         ((request) == PTRACE_SINGLEBLOCK)
 866 #else
 867 #define is_singleblock(request)         0
 868 #endif
 869
 870 #ifdef PTRACE_SYSEMU
 871 #define is_sysemu_singlestep(request)   ((request) == PTRACE_SYSEMU_SINGLESTEP)
 872 #else
 873 #define is_sysemu_singlestep(request)   0
 874 #endif
 875
 876 static int ptrace_resume(struct task_struct *child, long request,
 877                          unsigned long data)
 878 {
 879         bool need_siglock;
 880
 881         if (!valid_signal(data))
 882                 return -EIO;
 883
 884         if (request == PTRACE_SYSCALL)
 885                 set_task_syscall_work(child, SYSCALL_TRACE);
 886         else
 887                 clear_task_syscall_work(child, SYSCALL_TRACE);
 888
 889 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
 890         if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
 891                 set_task_syscall_work(child, SYSCALL_EMU);
 892         else
 893                 clear_task_syscall_work(child, SYSCALL_EMU);
 894 #endif
 895
 896         if (is_singleblock(request)) {
 897                 if (unlikely(!arch_has_block_step()))
 898                         return -EIO;
 899                 user_enable_block_step(child);
 900         } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
 901                 if (unlikely(!arch_has_single_step()))
 902                         return -EIO;
 903                 user_enable_single_step(child);
 904         } else {
 905                 user_disable_single_step(child);
 906         }
 907
 908         /*
 909          * Change ->exit_code and ->state under siglock to avoid the race
 910          * with wait_task_stopped() in between; a non-zero ->exit_code will
 911          * wrongly look like another report from tracee.
 912          *
 913          * Note that we need siglock even if ->exit_code == data and/or this
 914          * status was not reported yet, the new status must not be cleared by
 915          * wait_task_stopped() after resume.
 916          *
 917          * If data == 0 we do not care if wait_task_stopped() reports the old
 918          * status and clears the code too; this can't race with the tracee, it
 919          * takes siglock after resume.
 920          */
 921         need_siglock = data && !thread_group_empty(current);
 922         if (need_siglock)
 923                 spin_lock_irq(&child->sighand->siglock);
 924         child->exit_code = data;
 925         wake_up_state(child, __TASK_TRACED);
 926         if (need_siglock)
 927                 spin_unlock_irq(&child->sighand->siglock);
 928
 929         return 0;
 930 }
 931
 932 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 933
 934 static const struct user_regset *
 935 find_regset(const struct user_regset_view *view, unsigned int type)
 936 {
 937         const struct user_regset *regset;
 938         int n;
 939
 940         for (n = 0; n < view->n; ++n) {
 941                 regset = view->regsets + n;
 942                 if (regset->core_note_type == type)
 943                         return regset;
 944         }
 945
 946         return NULL;
 947 }
 948
 949 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
 950                          struct iovec *kiov)
 951 {
 952         const struct user_regset_view *view = task_user_regset_view(task);
 953         const struct user_regset *regset = find_regset(view, type);
 954         int regset_no;
 955
 956         if (!regset || (kiov->iov_len % regset->size) != 0)
 957                 return -EINVAL;
 958
 959         regset_no = regset - view->regsets;
 960         kiov->iov_len = min(kiov->iov_len,
 961                             (__kernel_size_t) (regset->n * regset->size));
 962
 963         if (req == PTRACE_GETREGSET)
 964                 return copy_regset_to_user(task, view, regset_no, 0,
 965                                            kiov->iov_len, kiov->iov_base);
 966         else
 967                 return copy_regset_from_user(task, view, regset_no, 0,
 968                                              kiov->iov_len, kiov->iov_base);
 969 }
 970
 971 /*
 972  * This is declared in linux/regset.h and defined in machine-dependent
 973  * code.  We put the export here, near the primary machine-neutral use,
 974  * to ensure no machine forgets it.
 975  */
 976 EXPORT_SYMBOL_GPL(task_user_regset_view);
 977
 978 static unsigned long
 979 ptrace_get_syscall_info_entry(struct task_struct *child, struct pt_regs *regs,
 980                               struct ptrace_syscall_info *info)
 981 {
 982         unsigned long args[ARRAY_SIZE(info->entry.args)];
 983         int i;
 984
 985         info->op = PTRACE_SYSCALL_INFO_ENTRY;
 986         info->entry.nr = syscall_get_nr(child, regs);
 987         syscall_get_arguments(child, regs, args);
 988         for (i = 0; i < ARRAY_SIZE(args); i++)
 989                 info->entry.args[i] = args[i];
 990
 991         /* args is the last field in struct ptrace_syscall_info.entry */
 992         return offsetofend(struct ptrace_syscall_info, entry.args);
 993 }
 994
 995 static unsigned long
 996 ptrace_get_syscall_info_seccomp(struct task_struct *child, struct pt_regs *regs,
 997                                 struct ptrace_syscall_info *info)
 998 {
 999         /*
1000          * As struct ptrace_syscall_info.entry is currently a subset
1001          * of struct ptrace_syscall_info.seccomp, it makes sense to
1002          * initialize that subset using ptrace_get_syscall_info_entry().
1003          * This can be reconsidered in the future if these structures
1004          * diverge significantly enough.
1005          */
1006         ptrace_get_syscall_info_entry(child, regs, info);
1007         info->op = PTRACE_SYSCALL_INFO_SECCOMP;
1008         info->seccomp.ret_data = child->ptrace_message;
1009
1010         /* ret_data is the last field in struct ptrace_syscall_info.seccomp */
1011         return offsetofend(struct ptrace_syscall_info, seccomp.ret_data);
1012 }
1013
1014 static unsigned long
1015 ptrace_get_syscall_info_exit(struct task_struct *child, struct pt_regs *regs,
1016                              struct ptrace_syscall_info *info)
1017 {
1018         info->op = PTRACE_SYSCALL_INFO_EXIT;
1019         info->exit.rval = syscall_get_error(child, regs);
1020         info->exit.is_error = !!info->exit.rval;
1021         if (!info->exit.is_error)
1022                 info->exit.rval = syscall_get_return_value(child, regs);
1023
1024         /* is_error is the last field in struct ptrace_syscall_info.exit */
1025         return offsetofend(struct ptrace_syscall_info, exit.is_error);
1026 }
1027
1028 static int
1029 ptrace_get_syscall_info(struct task_struct *child, unsigned long user_size,
1030                         void __user *datavp)
1031 {
1032         struct pt_regs *regs = task_pt_regs(child);
1033         struct ptrace_syscall_info info = {
1034                 .op = PTRACE_SYSCALL_INFO_NONE,
1035                 .arch = syscall_get_arch(child),
1036                 .instruction_pointer = instruction_pointer(regs),
1037                 .stack_pointer = user_stack_pointer(regs),
1038         };
1039         unsigned long actual_size = offsetof(struct ptrace_syscall_info, entry);
1040         unsigned long write_size;
1041
1042         /*
1043          * This does not need lock_task_sighand() to access
1044          * child->last_siginfo because ptrace_freeze_traced()
1045          * called earlier by ptrace_check_attach() ensures that
1046          * the tracee cannot go away and clear its last_siginfo.
1047          */
1048         switch (child->last_siginfo ? child->last_siginfo->si_code : 0) {
1049         case SIGTRAP | 0x80:
1050                 switch (child->ptrace_message) {
1051                 case PTRACE_EVENTMSG_SYSCALL_ENTRY:
1052                         actual_size = ptrace_get_syscall_info_entry(child, regs,
1053                                                                     &info);
1054                         break;
1055                 case PTRACE_EVENTMSG_SYSCALL_EXIT:
1056                         actual_size = ptrace_get_syscall_info_exit(child, regs,
1057                                                                    &info);
1058                         break;
1059                 }
1060                 break;
1061         case SIGTRAP | (PTRACE_EVENT_SECCOMP << 8):
1062                 actual_size = ptrace_get_syscall_info_seccomp(child, regs,
1063                                                               &info);
1064                 break;
1065         }
1066
1067         write_size = min(actual_size, user_size);
1068         return copy_to_user(datavp, &info, write_size) ? -EFAULT : actual_size;
1069 }
1070 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
1071
1072 int ptrace_request(struct task_struct *child, long request,
1073                    unsigned long addr, unsigned long data)
1074 {
1075         bool seized = child->ptrace & PT_SEIZED;
1076         int ret = -EIO;
1077         kernel_siginfo_t siginfo, *si;
1078         void __user *datavp = (void __user *) data;
1079         unsigned long __user *datalp = datavp;
1080         unsigned long flags;
1081
1082         switch (request) {
1083         case PTRACE_PEEKTEXT:
1084         case PTRACE_PEEKDATA:
1085                 return generic_ptrace_peekdata(child, addr, data);
1086         case PTRACE_POKETEXT:
1087         case PTRACE_POKEDATA:
1088                 return generic_ptrace_pokedata(child, addr, data);
1089
1090 #ifdef PTRACE_OLDSETOPTIONS
1091         case PTRACE_OLDSETOPTIONS:
1092 #endif
1093         case PTRACE_SETOPTIONS:
1094                 ret = ptrace_setoptions(child, data);
1095                 break;
1096         case PTRACE_GETEVENTMSG:
1097                 ret = put_user(child->ptrace_message, datalp);
1098                 break;
1099
1100         case PTRACE_PEEKSIGINFO:
1101                 ret = ptrace_peek_siginfo(child, addr, data);
1102                 break;
1103
1104         case PTRACE_GETSIGINFO:
1105                 ret = ptrace_getsiginfo(child, &siginfo);
1106                 if (!ret)
1107                         ret = copy_siginfo_to_user(datavp, &siginfo);
1108                 break;
1109
1110         case PTRACE_SETSIGINFO:
1111                 ret = copy_siginfo_from_user(&siginfo, datavp);
1112                 if (!ret)
1113                         ret = ptrace_setsiginfo(child, &siginfo);
1114                 break;
1115
1116         case PTRACE_GETSIGMASK: {
1117                 sigset_t *mask;
1118
1119                 if (addr != sizeof(sigset_t)) {
1120                         ret = -EINVAL;
1121                         break;
1122                 }
1123
1124                 if (test_tsk_restore_sigmask(child))
1125                         mask = &child->saved_sigmask;
1126                 else
1127                         mask = &child->blocked;
1128
1129                 if (copy_to_user(datavp, mask, sizeof(sigset_t)))
1130                         ret = -EFAULT;
1131                 else
1132                         ret = 0;
1133
1134                 break;
1135         }
1136
1137         case PTRACE_SETSIGMASK: {
1138                 sigset_t new_set;
1139
1140                 if (addr != sizeof(sigset_t)) {
1141                         ret = -EINVAL;
1142                         break;
1143                 }
1144
1145                 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
1146                         ret = -EFAULT;
1147                         break;
1148                 }
1149
1150                 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
1151
1152                 /*
1153                  * Every thread does recalc_sigpending() after resume, so
1154                  * retarget_shared_pending() and recalc_sigpending() are not
1155                  * called here.
1156                  */
1157                 spin_lock_irq(&child->sighand->siglock);
1158                 child->blocked = new_set;
1159                 spin_unlock_irq(&child->sighand->siglock);
1160
1161                 clear_tsk_restore_sigmask(child);
1162
1163                 ret = 0;
1164                 break;
1165         }
1166
1167         case PTRACE_INTERRUPT:
1168                 /*
1169                  * Stop tracee without any side-effect on signal or job
1170                  * control.  At least one trap is guaranteed to happen
1171                  * after this request.  If @child is already trapped, the
1172                  * current trap is not disturbed and another trap will
1173                  * happen after the current trap is ended with PTRACE_CONT.
1174                  *
1175                  * The actual trap might not be PTRACE_EVENT_STOP trap but
1176                  * the pending condition is cleared regardless.
1177                  */
1178                 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1179                         break;
1180
1181                 /*
1182                  * INTERRUPT doesn't disturb existing trap sans one
1183                  * exception.  If ptracer issued LISTEN for the current
1184                  * STOP, this INTERRUPT should clear LISTEN and re-trap
1185                  * tracee into STOP.
1186                  */
1187                 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
1188                         ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1189
1190                 unlock_task_sighand(child, &flags);
1191                 ret = 0;
1192                 break;
1193
1194         case PTRACE_LISTEN:
1195                 /*
1196                  * Listen for events.  Tracee must be in STOP.  It's not
1197                  * resumed per-se but is not considered to be in TRACED by
1198                  * wait(2) or ptrace(2).  If an async event (e.g. group
1199                  * stop state change) happens, tracee will enter STOP trap
1200                  * again.  Alternatively, ptracer can issue INTERRUPT to
1201                  * finish listening and re-trap tracee into STOP.
1202                  */
1203                 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1204                         break;
1205
1206                 si = child->last_siginfo;
1207                 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1208                         child->jobctl |= JOBCTL_LISTENING;
1209                         /*
1210                          * If NOTIFY is set, it means event happened between
1211                          * start of this trap and now.  Trigger re-trap.
1212                          */
1213                         if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1214                                 ptrace_signal_wake_up(child, true);
1215                         ret = 0;
1216                 }
1217                 unlock_task_sighand(child, &flags);
1218                 break;
1219
1220         case PTRACE_DETACH:      /* detach a process that was attached. */
1221                 ret = ptrace_detach(child, data);
1222                 break;
1223
1224 #ifdef CONFIG_BINFMT_ELF_FDPIC
1225         case PTRACE_GETFDPIC: {
1226                 struct mm_struct *mm = get_task_mm(child);
1227                 unsigned long tmp = 0;
1228
1229                 ret = -ESRCH;
1230                 if (!mm)
1231                         break;
1232
1233                 switch (addr) {
1234                 case PTRACE_GETFDPIC_EXEC:
1235                         tmp = mm->context.exec_fdpic_loadmap;
1236                         break;
1237                 case PTRACE_GETFDPIC_INTERP:
1238                         tmp = mm->context.interp_fdpic_loadmap;
1239                         break;
1240                 default:
1241                         break;
1242                 }
1243                 mmput(mm);
1244
1245                 ret = put_user(tmp, datalp);
1246                 break;
1247         }
1248 #endif
1249
1250 #ifdef PTRACE_SINGLESTEP
1251         case PTRACE_SINGLESTEP:
1252 #endif
1253 #ifdef PTRACE_SINGLEBLOCK
1254         case PTRACE_SINGLEBLOCK:
1255 #endif
1256 #ifdef PTRACE_SYSEMU
1257         case PTRACE_SYSEMU:
1258         case PTRACE_SYSEMU_SINGLESTEP:
1259 #endif
1260         case PTRACE_SYSCALL:
1261         case PTRACE_CONT:
1262                 return ptrace_resume(child, request, data);
1263
1264         case PTRACE_KILL:
1265                 send_sig_info(SIGKILL, SEND_SIG_NOINFO, child);
1266                 return 0;
1267
1268 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1269         case PTRACE_GETREGSET:
1270         case PTRACE_SETREGSET: {
1271                 struct iovec kiov;
1272                 struct iovec __user *uiov = datavp;
1273
1274                 if (!access_ok(uiov, sizeof(*uiov)))
1275                         return -EFAULT;
1276
1277                 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1278                     __get_user(kiov.iov_len, &uiov->iov_len))
1279                         return -EFAULT;
1280
1281                 ret = ptrace_regset(child, request, addr, &kiov);
1282                 if (!ret)
1283                         ret = __put_user(kiov.iov_len, &uiov->iov_len);
1284                 break;
1285         }
1286
1287         case PTRACE_GET_SYSCALL_INFO:
1288                 ret = ptrace_get_syscall_info(child, addr, datavp);
1289                 break;
1290 #endif
1291
1292         case PTRACE_SECCOMP_GET_FILTER:
1293                 ret = seccomp_get_filter(child, addr, datavp);
1294                 break;
1295
1296         case PTRACE_SECCOMP_GET_METADATA:
1297                 ret = seccomp_get_metadata(child, addr, datavp);
1298                 break;
1299
1300 #ifdef CONFIG_RSEQ
1301         case PTRACE_GET_RSEQ_CONFIGURATION:
1302                 ret = ptrace_get_rseq_configuration(child, addr, datavp);
1303                 break;
1304 #endif
1305
1306         default:
1307                 break;
1308         }
1309
1310         return ret;
1311 }
1312
1313 #ifndef arch_ptrace_attach
1314 #define arch_ptrace_attach(child)       do { } while (0)
1315 #endif
1316
1317 SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1318                 unsigned long, data)
1319 {
1320         struct task_struct *child;
1321         long ret;
1322
1323         if (request == PTRACE_TRACEME) {
1324                 ret = ptrace_traceme();
1325                 if (!ret)
1326                         arch_ptrace_attach(current);
1327                 goto out;
1328         }
1329
1330         child = find_get_task_by_vpid(pid);
1331         if (!child) {
1332                 ret = -ESRCH;
1333                 goto out;
1334         }
1335
1336         if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1337                 ret = ptrace_attach(child, request, addr, data);
1338                 /*
1339                  * Some architectures need to do book-keeping after
1340                  * a ptrace attach.
1341                  */
1342                 if (!ret)
1343                         arch_ptrace_attach(child);
1344                 goto out_put_task_struct;
1345         }
1346
1347         ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1348                                   request == PTRACE_INTERRUPT);
1349         if (ret < 0)
1350                 goto out_put_task_struct;
1351
1352         ret = arch_ptrace(child, request, addr, data);
1353         if (ret || request != PTRACE_DETACH)
1354                 ptrace_unfreeze_traced(child);
1355
1356  out_put_task_struct:
1357         put_task_struct(child);
1358  out:
1359         return ret;
1360 }
1361
1362 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1363                             unsigned long data)
1364 {
1365         unsigned long tmp;
1366         int copied;
1367
1368         copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1369         if (copied != sizeof(tmp))
1370                 return -EIO;
1371         return put_user(tmp, (unsigned long __user *)data);
1372 }
1373
1374 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1375                             unsigned long data)
1376 {
1377         int copied;
1378
1379         copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1380                         FOLL_FORCE | FOLL_WRITE);
1381         return (copied == sizeof(data)) ? 0 : -EIO;
1382 }
1383
1384 #if defined CONFIG_COMPAT
1385
1386 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1387                           compat_ulong_t addr, compat_ulong_t data)
1388 {
1389         compat_ulong_t __user *datap = compat_ptr(data);
1390         compat_ulong_t word;
1391         kernel_siginfo_t siginfo;
1392         int ret;
1393
1394         switch (request) {
1395         case PTRACE_PEEKTEXT:
1396         case PTRACE_PEEKDATA:
1397                 ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1398                                 FOLL_FORCE);
1399                 if (ret != sizeof(word))
1400                         ret = -EIO;
1401                 else
1402                         ret = put_user(word, datap);
1403                 break;
1404
1405         case PTRACE_POKETEXT:
1406         case PTRACE_POKEDATA:
1407                 ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1408                                 FOLL_FORCE | FOLL_WRITE);
1409                 ret = (ret != sizeof(data) ? -EIO : 0);
1410                 break;
1411
1412         case PTRACE_GETEVENTMSG:
1413                 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1414                 break;
1415
1416         case PTRACE_GETSIGINFO:
1417                 ret = ptrace_getsiginfo(child, &siginfo);
1418                 if (!ret)
1419                         ret = copy_siginfo_to_user32(
1420                                 (struct compat_siginfo __user *) datap,
1421                                 &siginfo);
1422                 break;
1423
1424         case PTRACE_SETSIGINFO:
1425                 ret = copy_siginfo_from_user32(
1426                         &siginfo, (struct compat_siginfo __user *) datap);
1427                 if (!ret)
1428                         ret = ptrace_setsiginfo(child, &siginfo);
1429                 break;
1430 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1431         case PTRACE_GETREGSET:
1432         case PTRACE_SETREGSET:
1433         {
1434                 struct iovec kiov;
1435                 struct compat_iovec __user *uiov =
1436                         (struct compat_iovec __user *) datap;
1437                 compat_uptr_t ptr;
1438                 compat_size_t len;
1439
1440                 if (!access_ok(uiov, sizeof(*uiov)))
1441                         return -EFAULT;
1442
1443                 if (__get_user(ptr, &uiov->iov_base) ||
1444                     __get_user(len, &uiov->iov_len))
1445                         return -EFAULT;
1446
1447                 kiov.iov_base = compat_ptr(ptr);
1448                 kiov.iov_len = len;
1449
1450                 ret = ptrace_regset(child, request, addr, &kiov);
1451                 if (!ret)
1452                         ret = __put_user(kiov.iov_len, &uiov->iov_len);
1453                 break;
1454         }
1455 #endif
1456
1457         default:
1458                 ret = ptrace_request(child, request, addr, data);
1459         }
1460
1461         return ret;
1462 }
1463
1464 COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1465                        compat_long_t, addr, compat_long_t, data)
1466 {
1467         struct task_struct *child;
1468         long ret;
1469
1470         if (request == PTRACE_TRACEME) {
1471                 ret = ptrace_traceme();
1472                 goto out;
1473         }
1474
1475         child = find_get_task_by_vpid(pid);
1476         if (!child) {
1477                 ret = -ESRCH;
1478                 goto out;
1479         }
1480
1481         if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1482                 ret = ptrace_attach(child, request, addr, data);
1483                 /*
1484                  * Some architectures need to do book-keeping after
1485                  * a ptrace attach.
1486                  */
1487                 if (!ret)
1488                         arch_ptrace_attach(child);
1489                 goto out_put_task_struct;
1490         }
1491
1492         ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1493                                   request == PTRACE_INTERRUPT);
1494         if (!ret) {
1495                 ret = compat_arch_ptrace(child, request, addr, data);
1496                 if (ret || request != PTRACE_DETACH)
1497                         ptrace_unfreeze_traced(child);
1498         }
1499
1500  out_put_task_struct:
1501         put_task_struct(child);
1502  out:
1503         return ret;
1504 }
1505 #endif  /* CONFIG_COMPAT */