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