linux/kernel/ptrace.c
<<
>>
Prefs
   1/*
   2 * linux/kernel/ptrace.c
   3 *
   4 * (C) Copyright 1999 Linus Torvalds
   5 *
   6 * Common interfaces for "ptrace()" which we do not want
   7 * to continually duplicate across every architecture.
   8 */
   9
  10#include <linux/capability.h>
  11#include <linux/export.h>
  12#include <linux/sched.h>
  13#include <linux/errno.h>
  14#include <linux/mm.h>
  15#include <linux/highmem.h>
  16#include <linux/pagemap.h>
  17#include <linux/ptrace.h>
  18#include <linux/security.h>
  19#include <linux/signal.h>
  20#include <linux/uio.h>
  21#include <linux/audit.h>
  22#include <linux/pid_namespace.h>
  23#include <linux/syscalls.h>
  24#include <linux/uaccess.h>
  25#include <linux/regset.h>
  26#include <linux/hw_breakpoint.h>
  27#include <linux/cn_proc.h>
  28#include <linux/compat.h>
  29
  30
  31/*
  32 * ptrace a task: make the debugger its new parent and
  33 * move it to the ptrace list.
  34 *
  35 * Must be called with the tasklist lock write-held.
  36 */
  37void __ptrace_link(struct task_struct *child, struct task_struct *new_parent)
  38{
  39        BUG_ON(!list_empty(&child->ptrace_entry));
  40        list_add(&child->ptrace_entry, &new_parent->ptraced);
  41        child->parent = new_parent;
  42}
  43
  44/**
  45 * __ptrace_unlink - unlink ptracee and restore its execution state
  46 * @child: ptracee to be unlinked
  47 *
  48 * Remove @child from the ptrace list, move it back to the original parent,
  49 * and restore the execution state so that it conforms to the group stop
  50 * state.
  51 *
  52 * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
  53 * exiting.  For PTRACE_DETACH, unless the ptracee has been killed between
  54 * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
  55 * If the ptracer is exiting, the ptracee can be in any state.
  56 *
  57 * After detach, the ptracee should be in a state which conforms to the
  58 * group stop.  If the group is stopped or in the process of stopping, the
  59 * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
  60 * up from TASK_TRACED.
  61 *
  62 * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
  63 * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
  64 * to but in the opposite direction of what happens while attaching to a
  65 * stopped task.  However, in this direction, the intermediate RUNNING
  66 * state is not hidden even from the current ptracer and if it immediately
  67 * re-attaches and performs a WNOHANG wait(2), it may fail.
  68 *
  69 * CONTEXT:
  70 * write_lock_irq(tasklist_lock)
  71 */
  72void __ptrace_unlink(struct task_struct *child)
  73{
  74        BUG_ON(!child->ptrace);
  75
  76        clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
  77
  78        child->parent = child->real_parent;
  79        list_del_init(&child->ptrace_entry);
  80
  81        spin_lock(&child->sighand->siglock);
  82        child->ptrace = 0;
  83        /*
  84         * Clear all pending traps and TRAPPING.  TRAPPING should be
  85         * cleared regardless of JOBCTL_STOP_PENDING.  Do it explicitly.
  86         */
  87        task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
  88        task_clear_jobctl_trapping(child);
  89
  90        /*
  91         * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
  92         * @child isn't dead.
  93         */
  94        if (!(child->flags & PF_EXITING) &&
  95            (child->signal->flags & SIGNAL_STOP_STOPPED ||
  96             child->signal->group_stop_count)) {
  97                child->jobctl |= JOBCTL_STOP_PENDING;
  98
  99                /*
 100                 * This is only possible if this thread was cloned by the
 101                 * traced task running in the stopped group, set the signal
 102                 * for the future reports.
 103                 * FIXME: we should change ptrace_init_task() to handle this
 104                 * case.
 105                 */
 106                if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
 107                        child->jobctl |= SIGSTOP;
 108        }
 109
 110        /*
 111         * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
 112         * @child in the butt.  Note that @resume should be used iff @child
 113         * is in TASK_TRACED; otherwise, we might unduly disrupt
 114         * TASK_KILLABLE sleeps.
 115         */
 116        if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
 117                ptrace_signal_wake_up(child, true);
 118
 119        spin_unlock(&child->sighand->siglock);
 120}
 121
 122/* Ensure that nothing can wake it up, even SIGKILL */
 123static bool ptrace_freeze_traced(struct task_struct *task)
 124{
 125        bool ret = false;
 126
 127        /* Lockless, nobody but us can set this flag */
 128        if (task->jobctl & JOBCTL_LISTENING)
 129                return ret;
 130
 131        spin_lock_irq(&task->sighand->siglock);
 132        if (task_is_traced(task) && !__fatal_signal_pending(task)) {
 133                task->state = __TASK_TRACED;
 134                ret = true;
 135        }
 136        spin_unlock_irq(&task->sighand->siglock);
 137
 138        return ret;
 139}
 140
 141static void ptrace_unfreeze_traced(struct task_struct *task)
 142{
 143        if (task->state != __TASK_TRACED)
 144                return;
 145
 146        WARN_ON(!task->ptrace || task->parent != current);
 147
 148        spin_lock_irq(&task->sighand->siglock);
 149        if (__fatal_signal_pending(task))
 150                wake_up_state(task, __TASK_TRACED);
 151        else
 152                task->state = TASK_TRACED;
 153        spin_unlock_irq(&task->sighand->siglock);
 154}
 155
 156/**
 157 * ptrace_check_attach - check whether ptracee is ready for ptrace operation
 158 * @child: ptracee to check for
 159 * @ignore_state: don't check whether @child is currently %TASK_TRACED
 160 *
 161 * Check whether @child is being ptraced by %current and ready for further
 162 * ptrace operations.  If @ignore_state is %false, @child also should be in
 163 * %TASK_TRACED state and on return the child is guaranteed to be traced
 164 * and not executing.  If @ignore_state is %true, @child can be in any
 165 * state.
 166 *
 167 * CONTEXT:
 168 * Grabs and releases tasklist_lock and @child->sighand->siglock.
 169 *
 170 * RETURNS:
 171 * 0 on success, -ESRCH if %child is not ready.
 172 */
 173static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
 174{
 175        int ret = -ESRCH;
 176
 177        /*
 178         * We take the read lock around doing both checks to close a
 179         * possible race where someone else was tracing our child and
 180         * detached between these two checks.  After this locked check,
 181         * we are sure that this is our traced child and that can only
 182         * be changed by us so it's not changing right after this.
 183         */
 184        read_lock(&tasklist_lock);
 185        if (child->ptrace && child->parent == current) {
 186                WARN_ON(child->state == __TASK_TRACED);
 187                /*
 188                 * child->sighand can't be NULL, release_task()
 189                 * does ptrace_unlink() before __exit_signal().
 190                 */
 191                if (ignore_state || ptrace_freeze_traced(child))
 192                        ret = 0;
 193        }
 194        read_unlock(&tasklist_lock);
 195
 196        if (!ret && !ignore_state) {
 197                if (!wait_task_inactive(child, __TASK_TRACED)) {
 198                        /*
 199                         * This can only happen if may_ptrace_stop() fails and
 200                         * ptrace_stop() changes ->state back to TASK_RUNNING,
 201                         * so we should not worry about leaking __TASK_TRACED.
 202                         */
 203                        WARN_ON(child->state == __TASK_TRACED);
 204                        ret = -ESRCH;
 205                }
 206        }
 207
 208        return ret;
 209}
 210
 211static int ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
 212{
 213        if (mode & PTRACE_MODE_NOAUDIT)
 214                return has_ns_capability_noaudit(current, ns, CAP_SYS_PTRACE);
 215        else
 216                return has_ns_capability(current, ns, CAP_SYS_PTRACE);
 217}
 218
 219/* Returns 0 on success, -errno on denial. */
 220static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
 221{
 222        const struct cred *cred = current_cred(), *tcred;
 223        int dumpable = 0;
 224        kuid_t caller_uid;
 225        kgid_t caller_gid;
 226
 227        if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
 228                WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
 229                return -EPERM;
 230        }
 231
 232        /* May we inspect the given task?
 233         * This check is used both for attaching with ptrace
 234         * and for allowing access to sensitive information in /proc.
 235         *
 236         * ptrace_attach denies several cases that /proc allows
 237         * because setting up the necessary parent/child relationship
 238         * or halting the specified task is impossible.
 239         */
 240
 241        /* Don't let security modules deny introspection */
 242        if (same_thread_group(task, current))
 243                return 0;
 244        rcu_read_lock();
 245        if (mode & PTRACE_MODE_FSCREDS) {
 246                caller_uid = cred->fsuid;
 247                caller_gid = cred->fsgid;
 248        } else {
 249                /*
 250                 * Using the euid would make more sense here, but something
 251                 * in userland might rely on the old behavior, and this
 252                 * shouldn't be a security problem since
 253                 * PTRACE_MODE_REALCREDS implies that the caller explicitly
 254                 * used a syscall that requests access to another process
 255                 * (and not a filesystem syscall to procfs).
 256                 */
 257                caller_uid = cred->uid;
 258                caller_gid = cred->gid;
 259        }
 260        tcred = __task_cred(task);
 261        if (uid_eq(caller_uid, tcred->euid) &&
 262            uid_eq(caller_uid, tcred->suid) &&
 263            uid_eq(caller_uid, tcred->uid)  &&
 264            gid_eq(caller_gid, tcred->egid) &&
 265            gid_eq(caller_gid, tcred->sgid) &&
 266            gid_eq(caller_gid, tcred->gid))
 267                goto ok;
 268        if (ptrace_has_cap(tcred->user_ns, mode))
 269                goto ok;
 270        rcu_read_unlock();
 271        return -EPERM;
 272ok:
 273        rcu_read_unlock();
 274        smp_rmb();
 275        if (task->mm)
 276                dumpable = get_dumpable(task->mm);
 277        rcu_read_lock();
 278        if (dumpable != SUID_DUMP_USER &&
 279            !ptrace_has_cap(__task_cred(task)->user_ns, mode)) {
 280                rcu_read_unlock();
 281                return -EPERM;
 282        }
 283        rcu_read_unlock();
 284
 285        return security_ptrace_access_check(task, mode);
 286}
 287
 288bool ptrace_may_access(struct task_struct *task, unsigned int mode)
 289{
 290        int err;
 291        task_lock(task);
 292        err = __ptrace_may_access(task, mode);
 293        task_unlock(task);
 294        return !err;
 295}
 296
 297static int ptrace_attach(struct task_struct *task, long request,
 298                         unsigned long addr,
 299                         unsigned long flags)
 300{
 301        bool seize = (request == PTRACE_SEIZE);
 302        int retval;
 303
 304        retval = -EIO;
 305        if (seize) {
 306                if (addr != 0)
 307                        goto out;
 308                if (flags & ~(unsigned long)PTRACE_O_MASK)
 309                        goto out;
 310                flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
 311        } else {
 312                flags = PT_PTRACED;
 313        }
 314
 315        audit_ptrace(task);
 316
 317        retval = -EPERM;
 318        if (unlikely(task->flags & PF_KTHREAD))
 319                goto out;
 320        if (same_thread_group(task, current))
 321                goto out;
 322
 323        /*
 324         * Protect exec's credential calculations against our interference;
 325         * SUID, SGID and LSM creds get determined differently
 326         * under ptrace.
 327         */
 328        retval = -ERESTARTNOINTR;
 329        if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
 330                goto out;
 331
 332        task_lock(task);
 333        retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
 334        task_unlock(task);
 335        if (retval)
 336                goto unlock_creds;
 337
 338        write_lock_irq(&tasklist_lock);
 339        retval = -EPERM;
 340        if (unlikely(task->exit_state))
 341                goto unlock_tasklist;
 342        if (task->ptrace)
 343                goto unlock_tasklist;
 344
 345        if (seize)
 346                flags |= PT_SEIZED;
 347        rcu_read_lock();
 348        if (ns_capable(__task_cred(task)->user_ns, CAP_SYS_PTRACE))
 349                flags |= PT_PTRACE_CAP;
 350        rcu_read_unlock();
 351        task->ptrace = flags;
 352
 353        __ptrace_link(task, current);
 354
 355        /* SEIZE doesn't trap tracee on attach */
 356        if (!seize)
 357                send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
 358
 359        spin_lock(&task->sighand->siglock);
 360
 361        /*
 362         * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
 363         * TRAPPING, and kick it so that it transits to TRACED.  TRAPPING
 364         * will be cleared if the child completes the transition or any
 365         * event which clears the group stop states happens.  We'll wait
 366         * for the transition to complete before returning from this
 367         * function.
 368         *
 369         * This hides STOPPED -> RUNNING -> TRACED transition from the
 370         * attaching thread but a different thread in the same group can
 371         * still observe the transient RUNNING state.  IOW, if another
 372         * thread's WNOHANG wait(2) on the stopped tracee races against
 373         * ATTACH, the wait(2) may fail due to the transient RUNNING.
 374         *
 375         * The following task_is_stopped() test is safe as both transitions
 376         * in and out of STOPPED are protected by siglock.
 377         */
 378        if (task_is_stopped(task) &&
 379            task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
 380                signal_wake_up_state(task, __TASK_STOPPED);
 381
 382        spin_unlock(&task->sighand->siglock);
 383
 384        retval = 0;
 385unlock_tasklist:
 386        write_unlock_irq(&tasklist_lock);
 387unlock_creds:
 388        mutex_unlock(&task->signal->cred_guard_mutex);
 389out:
 390        if (!retval) {
 391                /*
 392                 * We do not bother to change retval or clear JOBCTL_TRAPPING
 393                 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
 394                 * not return to user-mode, it will exit and clear this bit in
 395                 * __ptrace_unlink() if it wasn't already cleared by the tracee;
 396                 * and until then nobody can ptrace this task.
 397                 */
 398                wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
 399                proc_ptrace_connector(task, PTRACE_ATTACH);
 400        }
 401
 402        return retval;
 403}
 404
 405/**
 406 * ptrace_traceme  --  helper for PTRACE_TRACEME
 407 *
 408 * Performs checks and sets PT_PTRACED.
 409 * Should be used by all ptrace implementations for PTRACE_TRACEME.
 410 */
 411static int ptrace_traceme(void)
 412{
 413        int ret = -EPERM;
 414
 415        write_lock_irq(&tasklist_lock);
 416        /* Are we already being traced? */
 417        if (!current->ptrace) {
 418                ret = security_ptrace_traceme(current->parent);
 419                /*
 420                 * Check PF_EXITING to ensure ->real_parent has not passed
 421                 * exit_ptrace(). Otherwise we don't report the error but
 422                 * pretend ->real_parent untraces us right after return.
 423                 */
 424                if (!ret && !(current->real_parent->flags & PF_EXITING)) {
 425                        current->ptrace = PT_PTRACED;
 426                        __ptrace_link(current, current->real_parent);
 427                }
 428        }
 429        write_unlock_irq(&tasklist_lock);
 430
 431        return ret;
 432}
 433
 434/*
 435 * Called with irqs disabled, returns true if childs should reap themselves.
 436 */
 437static int ignoring_children(struct sighand_struct *sigh)
 438{
 439        int ret;
 440        spin_lock(&sigh->siglock);
 441        ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
 442              (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
 443        spin_unlock(&sigh->siglock);
 444        return ret;
 445}
 446
 447/*
 448 * Called with tasklist_lock held for writing.
 449 * Unlink a traced task, and clean it up if it was a traced zombie.
 450 * Return true if it needs to be reaped with release_task().
 451 * (We can't call release_task() here because we already hold tasklist_lock.)
 452 *
 453 * If it's a zombie, our attachedness prevented normal parent notification
 454 * or self-reaping.  Do notification now if it would have happened earlier.
 455 * If it should reap itself, return true.
 456 *
 457 * If it's our own child, there is no notification to do. But if our normal
 458 * children self-reap, then this child was prevented by ptrace and we must
 459 * reap it now, in that case we must also wake up sub-threads sleeping in
 460 * do_wait().
 461 */
 462static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
 463{
 464        bool dead;
 465
 466        __ptrace_unlink(p);
 467
 468        if (p->exit_state != EXIT_ZOMBIE)
 469                return false;
 470
 471        dead = !thread_group_leader(p);
 472
 473        if (!dead && thread_group_empty(p)) {
 474                if (!same_thread_group(p->real_parent, tracer))
 475                        dead = do_notify_parent(p, p->exit_signal);
 476                else if (ignoring_children(tracer->sighand)) {
 477                        __wake_up_parent(p, tracer);
 478                        dead = true;
 479                }
 480        }
 481        /* Mark it as in the process of being reaped. */
 482        if (dead)
 483                p->exit_state = EXIT_DEAD;
 484        return dead;
 485}
 486
 487static int ptrace_detach(struct task_struct *child, unsigned int data)
 488{
 489        if (!valid_signal(data))
 490                return -EIO;
 491
 492        /* Architecture-specific hardware disable .. */
 493        ptrace_disable(child);
 494
 495        write_lock_irq(&tasklist_lock);
 496        /*
 497         * We rely on ptrace_freeze_traced(). It can't be killed and
 498         * untraced by another thread, it can't be a zombie.
 499         */
 500        WARN_ON(!child->ptrace || child->exit_state);
 501        /*
 502         * tasklist_lock avoids the race with wait_task_stopped(), see
 503         * the comment in ptrace_resume().
 504         */
 505        child->exit_code = data;
 506        __ptrace_detach(current, child);
 507        write_unlock_irq(&tasklist_lock);
 508
 509        proc_ptrace_connector(child, PTRACE_DETACH);
 510
 511        return 0;
 512}
 513
 514/*
 515 * Detach all tasks we were using ptrace on. Called with tasklist held
 516 * for writing.
 517 */
 518void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
 519{
 520        struct task_struct *p, *n;
 521
 522        list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
 523                if (unlikely(p->ptrace & PT_EXITKILL))
 524                        send_sig_info(SIGKILL, SEND_SIG_FORCED, p);
 525
 526                if (__ptrace_detach(tracer, p))
 527                        list_add(&p->ptrace_entry, dead);
 528        }
 529}
 530
 531int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
 532{
 533        int copied = 0;
 534
 535        while (len > 0) {
 536                char buf[128];
 537                int this_len, retval;
 538
 539                this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 540                retval = access_process_vm(tsk, src, buf, this_len, FOLL_FORCE);
 541                if (!retval) {
 542                        if (copied)
 543                                break;
 544                        return -EIO;
 545                }
 546                if (copy_to_user(dst, buf, retval))
 547                        return -EFAULT;
 548                copied += retval;
 549                src += retval;
 550                dst += retval;
 551                len -= retval;
 552        }
 553        return copied;
 554}
 555
 556int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
 557{
 558        int copied = 0;
 559
 560        while (len > 0) {
 561                char buf[128];
 562                int this_len, retval;
 563
 564                this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 565                if (copy_from_user(buf, src, this_len))
 566                        return -EFAULT;
 567                retval = access_process_vm(tsk, dst, buf, this_len,
 568                                FOLL_FORCE | FOLL_WRITE);
 569                if (!retval) {
 570                        if (copied)
 571                                break;
 572                        return -EIO;
 573                }
 574                copied += retval;
 575                src += retval;
 576                dst += retval;
 577                len -= retval;
 578        }
 579        return copied;
 580}
 581
 582static int ptrace_setoptions(struct task_struct *child, unsigned long data)
 583{
 584        unsigned flags;
 585
 586        if (data & ~(unsigned long)PTRACE_O_MASK)
 587                return -EINVAL;
 588
 589        if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
 590                if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
 591                    !IS_ENABLED(CONFIG_SECCOMP))
 592                        return -EINVAL;
 593
 594                if (!capable(CAP_SYS_ADMIN))
 595                        return -EPERM;
 596
 597                if (seccomp_mode(&current->seccomp) != SECCOMP_MODE_DISABLED ||
 598                    current->ptrace & PT_SUSPEND_SECCOMP)
 599                        return -EPERM;
 600        }
 601
 602        /* Avoid intermediate state when all opts are cleared */
 603        flags = child->ptrace;
 604        flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
 605        flags |= (data << PT_OPT_FLAG_SHIFT);
 606        child->ptrace = flags;
 607
 608        return 0;
 609}
 610
 611static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
 612{
 613        unsigned long flags;
 614        int error = -ESRCH;
 615
 616        if (lock_task_sighand(child, &flags)) {
 617                error = -EINVAL;
 618                if (likely(child->last_siginfo != NULL)) {
 619                        *info = *child->last_siginfo;
 620                        error = 0;
 621                }
 622                unlock_task_sighand(child, &flags);
 623        }
 624        return error;
 625}
 626
 627static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
 628{
 629        unsigned long flags;
 630        int error = -ESRCH;
 631
 632        if (lock_task_sighand(child, &flags)) {
 633                error = -EINVAL;
 634                if (likely(child->last_siginfo != NULL)) {
 635                        *child->last_siginfo = *info;
 636                        error = 0;
 637                }
 638                unlock_task_sighand(child, &flags);
 639        }
 640        return error;
 641}
 642
 643static int ptrace_peek_siginfo(struct task_struct *child,
 644                                unsigned long addr,
 645                                unsigned long data)
 646{
 647        struct ptrace_peeksiginfo_args arg;
 648        struct sigpending *pending;
 649        struct sigqueue *q;
 650        int ret, i;
 651
 652        ret = copy_from_user(&arg, (void __user *) addr,
 653                                sizeof(struct ptrace_peeksiginfo_args));
 654        if (ret)
 655                return -EFAULT;
 656
 657        if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
 658                return -EINVAL; /* unknown flags */
 659
 660        if (arg.nr < 0)
 661                return -EINVAL;
 662
 663        if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
 664                pending = &child->signal->shared_pending;
 665        else
 666                pending = &child->pending;
 667
 668        for (i = 0; i < arg.nr; ) {
 669                siginfo_t info;
 670                s32 off = arg.off + i;
 671
 672                spin_lock_irq(&child->sighand->siglock);
 673                list_for_each_entry(q, &pending->list, list) {
 674                        if (!off--) {
 675                                copy_siginfo(&info, &q->info);
 676                                break;
 677                        }
 678                }
 679                spin_unlock_irq(&child->sighand->siglock);
 680
 681                if (off >= 0) /* beyond the end of the list */
 682                        break;
 683
 684#ifdef CONFIG_COMPAT
 685                if (unlikely(in_compat_syscall())) {
 686                        compat_siginfo_t __user *uinfo = compat_ptr(data);
 687
 688                        if (copy_siginfo_to_user32(uinfo, &info) ||
 689                            __put_user(info.si_code, &uinfo->si_code)) {
 690                                ret = -EFAULT;
 691                                break;
 692                        }
 693
 694                } else
 695#endif
 696                {
 697                        siginfo_t __user *uinfo = (siginfo_t __user *) data;
 698
 699                        if (copy_siginfo_to_user(uinfo, &info) ||
 700                            __put_user(info.si_code, &uinfo->si_code)) {
 701                                ret = -EFAULT;
 702                                break;
 703                        }
 704                }
 705
 706                data += sizeof(siginfo_t);
 707                i++;
 708
 709                if (signal_pending(current))
 710                        break;
 711
 712                cond_resched();
 713        }
 714
 715        if (i > 0)
 716                return i;
 717
 718        return ret;
 719}
 720
 721#ifdef PTRACE_SINGLESTEP
 722#define is_singlestep(request)          ((request) == PTRACE_SINGLESTEP)
 723#else
 724#define is_singlestep(request)          0
 725#endif
 726
 727#ifdef PTRACE_SINGLEBLOCK
 728#define is_singleblock(request)         ((request) == PTRACE_SINGLEBLOCK)
 729#else
 730#define is_singleblock(request)         0
 731#endif
 732
 733#ifdef PTRACE_SYSEMU
 734#define is_sysemu_singlestep(request)   ((request) == PTRACE_SYSEMU_SINGLESTEP)
 735#else
 736#define is_sysemu_singlestep(request)   0
 737#endif
 738
 739static int ptrace_resume(struct task_struct *child, long request,
 740                         unsigned long data)
 741{
 742        bool need_siglock;
 743
 744        if (!valid_signal(data))
 745                return -EIO;
 746
 747        if (request == PTRACE_SYSCALL)
 748                set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 749        else
 750                clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 751
 752#ifdef TIF_SYSCALL_EMU
 753        if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
 754                set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 755        else
 756                clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 757#endif
 758
 759        if (is_singleblock(request)) {
 760                if (unlikely(!arch_has_block_step()))
 761                        return -EIO;
 762                user_enable_block_step(child);
 763        } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
 764                if (unlikely(!arch_has_single_step()))
 765                        return -EIO;
 766                user_enable_single_step(child);
 767        } else {
 768                user_disable_single_step(child);
 769        }
 770
 771        /*
 772         * Change ->exit_code and ->state under siglock to avoid the race
 773         * with wait_task_stopped() in between; a non-zero ->exit_code will
 774         * wrongly look like another report from tracee.
 775         *
 776         * Note that we need siglock even if ->exit_code == data and/or this
 777         * status was not reported yet, the new status must not be cleared by
 778         * wait_task_stopped() after resume.
 779         *
 780         * If data == 0 we do not care if wait_task_stopped() reports the old
 781         * status and clears the code too; this can't race with the tracee, it
 782         * takes siglock after resume.
 783         */
 784        need_siglock = data && !thread_group_empty(current);
 785        if (need_siglock)
 786                spin_lock_irq(&child->sighand->siglock);
 787        child->exit_code = data;
 788        wake_up_state(child, __TASK_TRACED);
 789        if (need_siglock)
 790                spin_unlock_irq(&child->sighand->siglock);
 791
 792        return 0;
 793}
 794
 795#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 796
 797static const struct user_regset *
 798find_regset(const struct user_regset_view *view, unsigned int type)
 799{
 800        const struct user_regset *regset;
 801        int n;
 802
 803        for (n = 0; n < view->n; ++n) {
 804                regset = view->regsets + n;
 805                if (regset->core_note_type == type)
 806                        return regset;
 807        }
 808
 809        return NULL;
 810}
 811
 812static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
 813                         struct iovec *kiov)
 814{
 815        const struct user_regset_view *view = task_user_regset_view(task);
 816        const struct user_regset *regset = find_regset(view, type);
 817        int regset_no;
 818
 819        if (!regset || (kiov->iov_len % regset->size) != 0)
 820                return -EINVAL;
 821
 822        regset_no = regset - view->regsets;
 823        kiov->iov_len = min(kiov->iov_len,
 824                            (__kernel_size_t) (regset->n * regset->size));
 825
 826        if (req == PTRACE_GETREGSET)
 827                return copy_regset_to_user(task, view, regset_no, 0,
 828                                           kiov->iov_len, kiov->iov_base);
 829        else
 830                return copy_regset_from_user(task, view, regset_no, 0,
 831                                             kiov->iov_len, kiov->iov_base);
 832}
 833
 834/*
 835 * This is declared in linux/regset.h and defined in machine-dependent
 836 * code.  We put the export here, near the primary machine-neutral use,
 837 * to ensure no machine forgets it.
 838 */
 839EXPORT_SYMBOL_GPL(task_user_regset_view);
 840#endif
 841
 842int ptrace_request(struct task_struct *child, long request,
 843                   unsigned long addr, unsigned long data)
 844{
 845        bool seized = child->ptrace & PT_SEIZED;
 846        int ret = -EIO;
 847        siginfo_t siginfo, *si;
 848        void __user *datavp = (void __user *) data;
 849        unsigned long __user *datalp = datavp;
 850        unsigned long flags;
 851
 852        switch (request) {
 853        case PTRACE_PEEKTEXT:
 854        case PTRACE_PEEKDATA:
 855                return generic_ptrace_peekdata(child, addr, data);
 856        case PTRACE_POKETEXT:
 857        case PTRACE_POKEDATA:
 858                return generic_ptrace_pokedata(child, addr, data);
 859
 860#ifdef PTRACE_OLDSETOPTIONS
 861        case PTRACE_OLDSETOPTIONS:
 862#endif
 863        case PTRACE_SETOPTIONS:
 864                ret = ptrace_setoptions(child, data);
 865                break;
 866        case PTRACE_GETEVENTMSG:
 867                ret = put_user(child->ptrace_message, datalp);
 868                break;
 869
 870        case PTRACE_PEEKSIGINFO:
 871                ret = ptrace_peek_siginfo(child, addr, data);
 872                break;
 873
 874        case PTRACE_GETSIGINFO:
 875                ret = ptrace_getsiginfo(child, &siginfo);
 876                if (!ret)
 877                        ret = copy_siginfo_to_user(datavp, &siginfo);
 878                break;
 879
 880        case PTRACE_SETSIGINFO:
 881                if (copy_from_user(&siginfo, datavp, sizeof siginfo))
 882                        ret = -EFAULT;
 883                else
 884                        ret = ptrace_setsiginfo(child, &siginfo);
 885                break;
 886
 887        case PTRACE_GETSIGMASK:
 888                if (addr != sizeof(sigset_t)) {
 889                        ret = -EINVAL;
 890                        break;
 891                }
 892
 893                if (copy_to_user(datavp, &child->blocked, sizeof(sigset_t)))
 894                        ret = -EFAULT;
 895                else
 896                        ret = 0;
 897
 898                break;
 899
 900        case PTRACE_SETSIGMASK: {
 901                sigset_t new_set;
 902
 903                if (addr != sizeof(sigset_t)) {
 904                        ret = -EINVAL;
 905                        break;
 906                }
 907
 908                if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
 909                        ret = -EFAULT;
 910                        break;
 911                }
 912
 913                sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
 914
 915                /*
 916                 * Every thread does recalc_sigpending() after resume, so
 917                 * retarget_shared_pending() and recalc_sigpending() are not
 918                 * called here.
 919                 */
 920                spin_lock_irq(&child->sighand->siglock);
 921                child->blocked = new_set;
 922                spin_unlock_irq(&child->sighand->siglock);
 923
 924                ret = 0;
 925                break;
 926        }
 927
 928        case PTRACE_INTERRUPT:
 929                /*
 930                 * Stop tracee without any side-effect on signal or job
 931                 * control.  At least one trap is guaranteed to happen
 932                 * after this request.  If @child is already trapped, the
 933                 * current trap is not disturbed and another trap will
 934                 * happen after the current trap is ended with PTRACE_CONT.
 935                 *
 936                 * The actual trap might not be PTRACE_EVENT_STOP trap but
 937                 * the pending condition is cleared regardless.
 938                 */
 939                if (unlikely(!seized || !lock_task_sighand(child, &flags)))
 940                        break;
 941
 942                /*
 943                 * INTERRUPT doesn't disturb existing trap sans one
 944                 * exception.  If ptracer issued LISTEN for the current
 945                 * STOP, this INTERRUPT should clear LISTEN and re-trap
 946                 * tracee into STOP.
 947                 */
 948                if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
 949                        ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
 950
 951                unlock_task_sighand(child, &flags);
 952                ret = 0;
 953                break;
 954
 955        case PTRACE_LISTEN:
 956                /*
 957                 * Listen for events.  Tracee must be in STOP.  It's not
 958                 * resumed per-se but is not considered to be in TRACED by
 959                 * wait(2) or ptrace(2).  If an async event (e.g. group
 960                 * stop state change) happens, tracee will enter STOP trap
 961                 * again.  Alternatively, ptracer can issue INTERRUPT to
 962                 * finish listening and re-trap tracee into STOP.
 963                 */
 964                if (unlikely(!seized || !lock_task_sighand(child, &flags)))
 965                        break;
 966
 967                si = child->last_siginfo;
 968                if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
 969                        child->jobctl |= JOBCTL_LISTENING;
 970                        /*
 971                         * If NOTIFY is set, it means event happened between
 972                         * start of this trap and now.  Trigger re-trap.
 973                         */
 974                        if (child->jobctl & JOBCTL_TRAP_NOTIFY)
 975                                ptrace_signal_wake_up(child, true);
 976                        ret = 0;
 977                }
 978                unlock_task_sighand(child, &flags);
 979                break;
 980
 981        case PTRACE_DETACH:      /* detach a process that was attached. */
 982                ret = ptrace_detach(child, data);
 983                break;
 984
 985#ifdef CONFIG_BINFMT_ELF_FDPIC
 986        case PTRACE_GETFDPIC: {
 987                struct mm_struct *mm = get_task_mm(child);
 988                unsigned long tmp = 0;
 989
 990                ret = -ESRCH;
 991                if (!mm)
 992                        break;
 993
 994                switch (addr) {
 995                case PTRACE_GETFDPIC_EXEC:
 996                        tmp = mm->context.exec_fdpic_loadmap;
 997                        break;
 998                case PTRACE_GETFDPIC_INTERP:
 999                        tmp = mm->context.interp_fdpic_loadmap;
1000                        break;
1001                default:
1002                        break;
1003                }
1004                mmput(mm);
1005
1006                ret = put_user(tmp, datalp);
1007                break;
1008        }
1009#endif
1010
1011#ifdef PTRACE_SINGLESTEP
1012        case PTRACE_SINGLESTEP:
1013#endif
1014#ifdef PTRACE_SINGLEBLOCK
1015        case PTRACE_SINGLEBLOCK:
1016#endif
1017#ifdef PTRACE_SYSEMU
1018        case PTRACE_SYSEMU:
1019        case PTRACE_SYSEMU_SINGLESTEP:
1020#endif
1021        case PTRACE_SYSCALL:
1022        case PTRACE_CONT:
1023                return ptrace_resume(child, request, data);
1024
1025        case PTRACE_KILL:
1026                if (child->exit_state)  /* already dead */
1027                        return 0;
1028                return ptrace_resume(child, request, SIGKILL);
1029
1030#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1031        case PTRACE_GETREGSET:
1032        case PTRACE_SETREGSET: {
1033                struct iovec kiov;
1034                struct iovec __user *uiov = datavp;
1035
1036                if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1037                        return -EFAULT;
1038
1039                if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1040                    __get_user(kiov.iov_len, &uiov->iov_len))
1041                        return -EFAULT;
1042
1043                ret = ptrace_regset(child, request, addr, &kiov);
1044                if (!ret)
1045                        ret = __put_user(kiov.iov_len, &uiov->iov_len);
1046                break;
1047        }
1048#endif
1049
1050        case PTRACE_SECCOMP_GET_FILTER:
1051                ret = seccomp_get_filter(child, addr, datavp);
1052                break;
1053
1054        default:
1055                break;
1056        }
1057
1058        return ret;
1059}
1060
1061static struct task_struct *ptrace_get_task_struct(pid_t pid)
1062{
1063        struct task_struct *child;
1064
1065        rcu_read_lock();
1066        child = find_task_by_vpid(pid);
1067        if (child)
1068                get_task_struct(child);
1069        rcu_read_unlock();
1070
1071        if (!child)
1072                return ERR_PTR(-ESRCH);
1073        return child;
1074}
1075
1076#ifndef arch_ptrace_attach
1077#define arch_ptrace_attach(child)       do { } while (0)
1078#endif
1079
1080SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1081                unsigned long, data)
1082{
1083        struct task_struct *child;
1084        long ret;
1085
1086        if (request == PTRACE_TRACEME) {
1087                ret = ptrace_traceme();
1088                if (!ret)
1089                        arch_ptrace_attach(current);
1090                goto out;
1091        }
1092
1093        child = ptrace_get_task_struct(pid);
1094        if (IS_ERR(child)) {
1095                ret = PTR_ERR(child);
1096                goto out;
1097        }
1098
1099        if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1100                ret = ptrace_attach(child, request, addr, data);
1101                /*
1102                 * Some architectures need to do book-keeping after
1103                 * a ptrace attach.
1104                 */
1105                if (!ret)
1106                        arch_ptrace_attach(child);
1107                goto out_put_task_struct;
1108        }
1109
1110        ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1111                                  request == PTRACE_INTERRUPT);
1112        if (ret < 0)
1113                goto out_put_task_struct;
1114
1115        ret = arch_ptrace(child, request, addr, data);
1116        if (ret || request != PTRACE_DETACH)
1117                ptrace_unfreeze_traced(child);
1118
1119 out_put_task_struct:
1120        put_task_struct(child);
1121 out:
1122        return ret;
1123}
1124
1125int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1126                            unsigned long data)
1127{
1128        unsigned long tmp;
1129        int copied;
1130
1131        copied = access_process_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1132        if (copied != sizeof(tmp))
1133                return -EIO;
1134        return put_user(tmp, (unsigned long __user *)data);
1135}
1136
1137int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1138                            unsigned long data)
1139{
1140        int copied;
1141
1142        copied = access_process_vm(tsk, addr, &data, sizeof(data),
1143                        FOLL_FORCE | FOLL_WRITE);
1144        return (copied == sizeof(data)) ? 0 : -EIO;
1145}
1146
1147#if defined CONFIG_COMPAT
1148
1149int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1150                          compat_ulong_t addr, compat_ulong_t data)
1151{
1152        compat_ulong_t __user *datap = compat_ptr(data);
1153        compat_ulong_t word;
1154        siginfo_t siginfo;
1155        int ret;
1156
1157        switch (request) {
1158        case PTRACE_PEEKTEXT:
1159        case PTRACE_PEEKDATA:
1160                ret = access_process_vm(child, addr, &word, sizeof(word),
1161                                FOLL_FORCE);
1162                if (ret != sizeof(word))
1163                        ret = -EIO;
1164                else
1165                        ret = put_user(word, datap);
1166                break;
1167
1168        case PTRACE_POKETEXT:
1169        case PTRACE_POKEDATA:
1170                ret = access_process_vm(child, addr, &data, sizeof(data),
1171                                FOLL_FORCE | FOLL_WRITE);
1172                ret = (ret != sizeof(data) ? -EIO : 0);
1173                break;
1174
1175        case PTRACE_GETEVENTMSG:
1176                ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1177                break;
1178
1179        case PTRACE_GETSIGINFO:
1180                ret = ptrace_getsiginfo(child, &siginfo);
1181                if (!ret)
1182                        ret = copy_siginfo_to_user32(
1183                                (struct compat_siginfo __user *) datap,
1184                                &siginfo);
1185                break;
1186
1187        case PTRACE_SETSIGINFO:
1188                memset(&siginfo, 0, sizeof siginfo);
1189                if (copy_siginfo_from_user32(
1190                            &siginfo, (struct compat_siginfo __user *) datap))
1191                        ret = -EFAULT;
1192                else
1193                        ret = ptrace_setsiginfo(child, &siginfo);
1194                break;
1195#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1196        case PTRACE_GETREGSET:
1197        case PTRACE_SETREGSET:
1198        {
1199                struct iovec kiov;
1200                struct compat_iovec __user *uiov =
1201                        (struct compat_iovec __user *) datap;
1202                compat_uptr_t ptr;
1203                compat_size_t len;
1204
1205                if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1206                        return -EFAULT;
1207
1208                if (__get_user(ptr, &uiov->iov_base) ||
1209                    __get_user(len, &uiov->iov_len))
1210                        return -EFAULT;
1211
1212                kiov.iov_base = compat_ptr(ptr);
1213                kiov.iov_len = len;
1214
1215                ret = ptrace_regset(child, request, addr, &kiov);
1216                if (!ret)
1217                        ret = __put_user(kiov.iov_len, &uiov->iov_len);
1218                break;
1219        }
1220#endif
1221
1222        default:
1223                ret = ptrace_request(child, request, addr, data);
1224        }
1225
1226        return ret;
1227}
1228
1229COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1230                       compat_long_t, addr, compat_long_t, data)
1231{
1232        struct task_struct *child;
1233        long ret;
1234
1235        if (request == PTRACE_TRACEME) {
1236                ret = ptrace_traceme();
1237                goto out;
1238        }
1239
1240        child = ptrace_get_task_struct(pid);
1241        if (IS_ERR(child)) {
1242                ret = PTR_ERR(child);
1243                goto out;
1244        }
1245
1246        if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1247                ret = ptrace_attach(child, request, addr, data);
1248                /*
1249                 * Some architectures need to do book-keeping after
1250                 * a ptrace attach.
1251                 */
1252                if (!ret)
1253                        arch_ptrace_attach(child);
1254                goto out_put_task_struct;
1255        }
1256
1257        ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1258                                  request == PTRACE_INTERRUPT);
1259        if (!ret) {
1260                ret = compat_arch_ptrace(child, request, addr, data);
1261                if (ret || request != PTRACE_DETACH)
1262                        ptrace_unfreeze_traced(child);
1263        }
1264
1265 out_put_task_struct:
1266        put_task_struct(child);
1267 out:
1268        return ret;
1269}
1270#endif  /* CONFIG_COMPAT */
1271