kernel/ptrace.c

   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/module.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/smp_lock.h>
  18 #include <linux/ptrace.h>
  19 #include <linux/security.h>
  20 #include <linux/signal.h>
  21 #include <linux/audit.h>
  22 #include <linux/pid_namespace.h>
  23 #include <linux/syscalls.h>
  24
  25 #include <asm/pgtable.h>
  26 #include <asm/uaccess.h>
  27
  28 /*
  29  * ptrace a task: make the debugger its new parent and
  30  * move it to the ptrace list.
  31  *
  32  * Must be called with the tasklist lock write-held.
  33  */
  34 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent)
  35 {
  36         BUG_ON(!list_empty(&child->ptrace_list));
  37         if (child->parent == new_parent)
  38                 return;
  39         list_add(&child->ptrace_list, &child->parent->ptrace_children);
  40         remove_parent(child);
  41         child->parent = new_parent;
  42         add_parent(child);
  43 }
  44
  45 /*
  46  * Turn a tracing stop into a normal stop now, since with no tracer there
  47  * would be no way to wake it up with SIGCONT or SIGKILL.  If there was a
  48  * signal sent that would resume the child, but didn't because it was in
  49  * TASK_TRACED, resume it now.
  50  * Requires that irqs be disabled.
  51  */
  52 void ptrace_untrace(struct task_struct *child)
  53 {
  54         spin_lock(&child->sighand->siglock);
  55         if (task_is_traced(child)) {
  56                 if (child->signal->flags & SIGNAL_STOP_STOPPED) {
  57                         __set_task_state(child, TASK_STOPPED);
  58                 } else {
  59                         signal_wake_up(child, 1);
  60                 }
  61         }
  62         spin_unlock(&child->sighand->siglock);
  63 }
  64
  65 /*
  66  * unptrace a task: move it back to its original parent and
  67  * remove it from the ptrace list.
  68  *
  69  * Must be called with the tasklist lock write-held.
  70  */
  71 void __ptrace_unlink(struct task_struct *child)
  72 {
  73         BUG_ON(!child->ptrace);
  74
  75         child->ptrace = 0;
  76         if (!list_empty(&child->ptrace_list)) {
  77                 list_del_init(&child->ptrace_list);
  78                 remove_parent(child);
  79                 child->parent = child->real_parent;
  80                 add_parent(child);
  81         }
  82
  83         if (task_is_traced(child))
  84                 ptrace_untrace(child);
  85 }
  86
  87 /*
  88  * Check that we have indeed attached to the thing..
  89  */
  90 int ptrace_check_attach(struct task_struct *child, int kill)
  91 {
  92         int ret = -ESRCH;
  93
  94         /*
  95          * We take the read lock around doing both checks to close a
  96          * possible race where someone else was tracing our child and
  97          * detached between these two checks.  After this locked check,
  98          * we are sure that this is our traced child and that can only
  99          * be changed by us so it's not changing right after this.
 100          */
 101         read_lock(&tasklist_lock);
 102         if ((child->ptrace & PT_PTRACED) && child->parent == current) {
 103                 ret = 0;
 104                 /*
 105                  * child->sighand can't be NULL, release_task()
 106                  * does ptrace_unlink() before __exit_signal().
 107                  */
 108                 spin_lock_irq(&child->sighand->siglock);
 109                 if (task_is_stopped(child))
 110                         child->state = TASK_TRACED;
 111                 else if (!task_is_traced(child) && !kill)
 112                         ret = -ESRCH;
 113                 spin_unlock_irq(&child->sighand->siglock);
 114         }
 115         read_unlock(&tasklist_lock);
 116
 117         if (!ret && !kill)
 118                 wait_task_inactive(child);
 119
 120         /* All systems go.. */
 121         return ret;
 122 }
 123
 124 int __ptrace_may_attach(struct task_struct *task)
 125 {
 126         /* May we inspect the given task?
 127          * This check is used both for attaching with ptrace
 128          * and for allowing access to sensitive information in /proc.
 129          *
 130          * ptrace_attach denies several cases that /proc allows
 131          * because setting up the necessary parent/child relationship
 132          * or halting the specified task is impossible.
 133          */
 134         int dumpable = 0;
 135         /* Don't let security modules deny introspection */
 136         if (task == current)
 137                 return 0;
 138         if (((current->uid != task->euid) ||
 139              (current->uid != task->suid) ||
 140              (current->uid != task->uid) ||
 141              (current->gid != task->egid) ||
 142              (current->gid != task->sgid) ||
 143              (current->gid != task->gid)) && !capable(CAP_SYS_PTRACE))
 144                 return -EPERM;
 145         smp_rmb();
 146         if (task->mm)
 147                 dumpable = get_dumpable(task->mm);
 148         if (!dumpable && !capable(CAP_SYS_PTRACE))
 149                 return -EPERM;
 150
 151         return security_ptrace(current, task);
 152 }
 153
 154 int ptrace_may_attach(struct task_struct *task)
 155 {
 156         int err;
 157         task_lock(task);
 158         err = __ptrace_may_attach(task);
 159         task_unlock(task);
 160         return !err;
 161 }
 162
 163 int ptrace_attach(struct task_struct *task)
 164 {
 165         int retval;
 166         unsigned long flags;
 167
 168         audit_ptrace(task);
 169
 170         retval = -EPERM;
 171         if (task->pid <= 1)
 172                 goto out;
 173         if (same_thread_group(task, current))
 174                 goto out;
 175
 176 repeat:
 177         /*
 178          * Nasty, nasty.
 179          *
 180          * We want to hold both the task-lock and the
 181          * tasklist_lock for writing at the same time.
 182          * But that's against the rules (tasklist_lock
 183          * is taken for reading by interrupts on other
 184          * cpu's that may have task_lock).
 185          */
 186         task_lock(task);
 187         if (!write_trylock_irqsave(&tasklist_lock, flags)) {
 188                 task_unlock(task);
 189                 do {
 190                         cpu_relax();
 191                 } while (!write_can_lock(&tasklist_lock));
 192                 goto repeat;
 193         }
 194
 195         if (!task->mm)
 196                 goto bad;
 197         /* the same process cannot be attached many times */
 198         if (task->ptrace & PT_PTRACED)
 199                 goto bad;
 200         retval = __ptrace_may_attach(task);
 201         if (retval)
 202                 goto bad;
 203
 204         /* Go */
 205         task->ptrace |= PT_PTRACED;
 206         if (capable(CAP_SYS_PTRACE))
 207                 task->ptrace |= PT_PTRACE_CAP;
 208
 209         __ptrace_link(task, current);
 210
 211         force_sig_specific(SIGSTOP, task);
 212
 213 bad:
 214         write_unlock_irqrestore(&tasklist_lock, flags);
 215         task_unlock(task);
 216 out:
 217         return retval;
 218 }
 219
 220 static inline void __ptrace_detach(struct task_struct *child, unsigned int data)
 221 {
 222         child->exit_code = data;
 223         /* .. re-parent .. */
 224         __ptrace_unlink(child);
 225         /* .. and wake it up. */
 226         if (child->exit_state != EXIT_ZOMBIE)
 227                 wake_up_process(child);
 228 }
 229
 230 int ptrace_detach(struct task_struct *child, unsigned int data)
 231 {
 232         if (!valid_signal(data))
 233                 return -EIO;
 234
 235         /* Architecture-specific hardware disable .. */
 236         ptrace_disable(child);
 237         clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 238
 239         write_lock_irq(&tasklist_lock);
 240         /* protect against de_thread()->release_task() */
 241         if (child->ptrace)
 242                 __ptrace_detach(child, data);
 243         write_unlock_irq(&tasklist_lock);
 244
 245         return 0;
 246 }
 247
 248 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
 249 {
 250         int copied = 0;
 251
 252         while (len > 0) {
 253                 char buf[128];
 254                 int this_len, retval;
 255
 256                 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 257                 retval = access_process_vm(tsk, src, buf, this_len, 0);
 258                 if (!retval) {
 259                         if (copied)
 260                                 break;
 261                         return -EIO;
 262                 }
 263                 if (copy_to_user(dst, buf, retval))
 264                         return -EFAULT;
 265                 copied += retval;
 266                 src += retval;
 267                 dst += retval;
 268                 len -= retval;
 269         }
 270         return copied;
 271 }
 272
 273 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
 274 {
 275         int copied = 0;
 276
 277         while (len > 0) {
 278                 char buf[128];
 279                 int this_len, retval;
 280
 281                 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 282                 if (copy_from_user(buf, src, this_len))
 283                         return -EFAULT;
 284                 retval = access_process_vm(tsk, dst, buf, this_len, 1);
 285                 if (!retval) {
 286                         if (copied)
 287                                 break;
 288                         return -EIO;
 289                 }
 290                 copied += retval;
 291                 src += retval;
 292                 dst += retval;
 293                 len -= retval;
 294         }
 295         return copied;
 296 }
 297
 298 static int ptrace_setoptions(struct task_struct *child, long data)
 299 {
 300         child->ptrace &= ~PT_TRACE_MASK;
 301
 302         if (data & PTRACE_O_TRACESYSGOOD)
 303                 child->ptrace |= PT_TRACESYSGOOD;
 304
 305         if (data & PTRACE_O_TRACEFORK)
 306                 child->ptrace |= PT_TRACE_FORK;
 307
 308         if (data & PTRACE_O_TRACEVFORK)
 309                 child->ptrace |= PT_TRACE_VFORK;
 310
 311         if (data & PTRACE_O_TRACECLONE)
 312                 child->ptrace |= PT_TRACE_CLONE;
 313
 314         if (data & PTRACE_O_TRACEEXEC)
 315                 child->ptrace |= PT_TRACE_EXEC;
 316
 317         if (data & PTRACE_O_TRACEVFORKDONE)
 318                 child->ptrace |= PT_TRACE_VFORK_DONE;
 319
 320         if (data & PTRACE_O_TRACEEXIT)
 321                 child->ptrace |= PT_TRACE_EXIT;
 322
 323         return (data & ~PTRACE_O_MASK) ? -EINVAL : 0;
 324 }
 325
 326 static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
 327 {
 328         int error = -ESRCH;
 329
 330         read_lock(&tasklist_lock);
 331         if (likely(child->sighand != NULL)) {
 332                 error = -EINVAL;
 333                 spin_lock_irq(&child->sighand->siglock);
 334                 if (likely(child->last_siginfo != NULL)) {
 335                         *info = *child->last_siginfo;
 336                         error = 0;
 337                 }
 338                 spin_unlock_irq(&child->sighand->siglock);
 339         }
 340         read_unlock(&tasklist_lock);
 341         return error;
 342 }
 343
 344 static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
 345 {
 346         int error = -ESRCH;
 347
 348         read_lock(&tasklist_lock);
 349         if (likely(child->sighand != NULL)) {
 350                 error = -EINVAL;
 351                 spin_lock_irq(&child->sighand->siglock);
 352                 if (likely(child->last_siginfo != NULL)) {
 353                         *child->last_siginfo = *info;
 354                         error = 0;
 355                 }
 356                 spin_unlock_irq(&child->sighand->siglock);
 357         }
 358         read_unlock(&tasklist_lock);
 359         return error;
 360 }
 361
 362
 363 #ifdef PTRACE_SINGLESTEP
 364 #define is_singlestep(request)          ((request) == PTRACE_SINGLESTEP)
 365 #else
 366 #define is_singlestep(request)          0
 367 #endif
 368
 369 #ifdef PTRACE_SINGLEBLOCK
 370 #define is_singleblock(request)         ((request) == PTRACE_SINGLEBLOCK)
 371 #else
 372 #define is_singleblock(request)         0
 373 #endif
 374
 375 #ifdef PTRACE_SYSEMU
 376 #define is_sysemu_singlestep(request)   ((request) == PTRACE_SYSEMU_SINGLESTEP)
 377 #else
 378 #define is_sysemu_singlestep(request)   0
 379 #endif
 380
 381 static int ptrace_resume(struct task_struct *child, long request, long data)
 382 {
 383         if (!valid_signal(data))
 384                 return -EIO;
 385
 386         if (request == PTRACE_SYSCALL)
 387                 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 388         else
 389                 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 390
 391 #ifdef TIF_SYSCALL_EMU
 392         if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
 393                 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 394         else
 395                 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 396 #endif
 397
 398         if (is_singleblock(request)) {
 399                 if (unlikely(!arch_has_block_step()))
 400                         return -EIO;
 401                 user_enable_block_step(child);
 402         } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
 403                 if (unlikely(!arch_has_single_step()))
 404                         return -EIO;
 405                 user_enable_single_step(child);
 406         }
 407         else
 408                 user_disable_single_step(child);
 409
 410         child->exit_code = data;
 411         wake_up_process(child);
 412
 413         return 0;
 414 }
 415
 416 int ptrace_request(struct task_struct *child, long request,
 417                    long addr, long data)
 418 {
 419         int ret = -EIO;
 420         siginfo_t siginfo;
 421
 422         switch (request) {
 423         case PTRACE_PEEKTEXT:
 424         case PTRACE_PEEKDATA:
 425                 return generic_ptrace_peekdata(child, addr, data);
 426         case PTRACE_POKETEXT:
 427         case PTRACE_POKEDATA:
 428                 return generic_ptrace_pokedata(child, addr, data);
 429
 430 #ifdef PTRACE_OLDSETOPTIONS
 431         case PTRACE_OLDSETOPTIONS:
 432 #endif
 433         case PTRACE_SETOPTIONS:
 434                 ret = ptrace_setoptions(child, data);
 435                 break;
 436         case PTRACE_GETEVENTMSG:
 437                 ret = put_user(child->ptrace_message, (unsigned long __user *) data);
 438                 break;
 439
 440         case PTRACE_GETSIGINFO:
 441                 ret = ptrace_getsiginfo(child, &siginfo);
 442                 if (!ret)
 443                         ret = copy_siginfo_to_user((siginfo_t __user *) data,
 444                                                    &siginfo);
 445                 break;
 446
 447         case PTRACE_SETSIGINFO:
 448                 if (copy_from_user(&siginfo, (siginfo_t __user *) data,
 449                                    sizeof siginfo))
 450                         ret = -EFAULT;
 451                 else
 452                         ret = ptrace_setsiginfo(child, &siginfo);
 453                 break;
 454
 455         case PTRACE_DETACH:      /* detach a process that was attached. */
 456                 ret = ptrace_detach(child, data);
 457                 break;
 458
 459 #ifdef PTRACE_SINGLESTEP
 460         case PTRACE_SINGLESTEP:
 461 #endif
 462 #ifdef PTRACE_SINGLEBLOCK
 463         case PTRACE_SINGLEBLOCK:
 464 #endif
 465 #ifdef PTRACE_SYSEMU
 466         case PTRACE_SYSEMU:
 467         case PTRACE_SYSEMU_SINGLESTEP:
 468 #endif
 469         case PTRACE_SYSCALL:
 470         case PTRACE_CONT:
 471                 return ptrace_resume(child, request, data);
 472
 473         case PTRACE_KILL:
 474                 if (child->exit_state)  /* already dead */
 475                         return 0;
 476                 return ptrace_resume(child, request, SIGKILL);
 477
 478         default:
 479                 break;
 480         }
 481
 482         return ret;
 483 }
 484
 485 /**
 486  * ptrace_traceme  --  helper for PTRACE_TRACEME
 487  *
 488  * Performs checks and sets PT_PTRACED.
 489  * Should be used by all ptrace implementations for PTRACE_TRACEME.
 490  */
 491 int ptrace_traceme(void)
 492 {
 493         int ret = -EPERM;
 494
 495         /*
 496          * Are we already being traced?
 497          */
 498         task_lock(current);
 499         if (!(current->ptrace & PT_PTRACED)) {
 500                 ret = security_ptrace(current->parent, current);
 501                 /*
 502                  * Set the ptrace bit in the process ptrace flags.
 503                  */
 504                 if (!ret)
 505                         current->ptrace |= PT_PTRACED;
 506         }
 507         task_unlock(current);
 508         return ret;
 509 }
 510
 511 /**
 512  * ptrace_get_task_struct  --  grab a task struct reference for ptrace
 513  * @pid:       process id to grab a task_struct reference of
 514  *
 515  * This function is a helper for ptrace implementations.  It checks
 516  * permissions and then grabs a task struct for use of the actual
 517  * ptrace implementation.
 518  *
 519  * Returns the task_struct for @pid or an ERR_PTR() on failure.
 520  */
 521 struct task_struct *ptrace_get_task_struct(pid_t pid)
 522 {
 523         struct task_struct *child;
 524
 525         /*
 526          * Tracing init is not allowed.
 527          */
 528         if (pid == 1)
 529                 return ERR_PTR(-EPERM);
 530
 531         read_lock(&tasklist_lock);
 532         child = find_task_by_vpid(pid);
 533         if (child)
 534                 get_task_struct(child);
 535
 536         read_unlock(&tasklist_lock);
 537         if (!child)
 538                 return ERR_PTR(-ESRCH);
 539         return child;
 540 }
 541
 542 #ifndef arch_ptrace_attach
 543 #define arch_ptrace_attach(child)       do { } while (0)
 544 #endif
 545
 546 #ifndef __ARCH_SYS_PTRACE
 547 asmlinkage long sys_ptrace(long request, long pid, long addr, long data)
 548 {
 549         struct task_struct *child;
 550         long ret;
 551
 552         /*
 553          * This lock_kernel fixes a subtle race with suid exec
 554          */
 555         lock_kernel();
 556         if (request == PTRACE_TRACEME) {
 557                 ret = ptrace_traceme();
 558                 if (!ret)
 559                         arch_ptrace_attach(current);
 560                 goto out;
 561         }
 562
 563         child = ptrace_get_task_struct(pid);
 564         if (IS_ERR(child)) {
 565                 ret = PTR_ERR(child);
 566                 goto out;
 567         }
 568
 569         if (request == PTRACE_ATTACH) {
 570                 ret = ptrace_attach(child);
 571                 /*
 572                  * Some architectures need to do book-keeping after
 573                  * a ptrace attach.
 574                  */
 575                 if (!ret)
 576                         arch_ptrace_attach(child);
 577                 goto out_put_task_struct;
 578         }
 579
 580         ret = ptrace_check_attach(child, request == PTRACE_KILL);
 581         if (ret < 0)
 582                 goto out_put_task_struct;
 583
 584         ret = arch_ptrace(child, request, addr, data);
 585         if (ret < 0)
 586                 goto out_put_task_struct;
 587
 588  out_put_task_struct:
 589         put_task_struct(child);
 590  out:
 591         unlock_kernel();
 592         return ret;
 593 }
 594 #endif /* __ARCH_SYS_PTRACE */
 595
 596 int generic_ptrace_peekdata(struct task_struct *tsk, long addr, long data)
 597 {
 598         unsigned long tmp;
 599         int copied;
 600
 601         copied = access_process_vm(tsk, addr, &tmp, sizeof(tmp), 0);
 602         if (copied != sizeof(tmp))
 603                 return -EIO;
 604         return put_user(tmp, (unsigned long __user *)data);
 605 }
 606
 607 int generic_ptrace_pokedata(struct task_struct *tsk, long addr, long data)
 608 {
 609         int copied;
 610
 611         copied = access_process_vm(tsk, addr, &data, sizeof(data), 1);
 612         return (copied == sizeof(data)) ? 0 : -EIO;
 613 }
 614
 615 #if defined CONFIG_COMPAT && defined __ARCH_WANT_COMPAT_SYS_PTRACE
 616 #include <linux/compat.h>
 617
 618 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
 619                           compat_ulong_t addr, compat_ulong_t data)
 620 {
 621         compat_ulong_t __user *datap = compat_ptr(data);
 622         compat_ulong_t word;
 623         siginfo_t siginfo;
 624         int ret;
 625
 626         switch (request) {
 627         case PTRACE_PEEKTEXT:
 628         case PTRACE_PEEKDATA:
 629                 ret = access_process_vm(child, addr, &word, sizeof(word), 0);
 630                 if (ret != sizeof(word))
 631                         ret = -EIO;
 632                 else
 633                         ret = put_user(word, datap);
 634                 break;
 635
 636         case PTRACE_POKETEXT:
 637         case PTRACE_POKEDATA:
 638                 ret = access_process_vm(child, addr, &data, sizeof(data), 1);
 639                 ret = (ret != sizeof(data) ? -EIO : 0);
 640                 break;
 641
 642         case PTRACE_GETEVENTMSG:
 643                 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
 644                 break;
 645
 646         case PTRACE_GETSIGINFO:
 647                 ret = ptrace_getsiginfo(child, &siginfo);
 648                 if (!ret)
 649                         ret = copy_siginfo_to_user32(
 650                                 (struct compat_siginfo __user *) datap,
 651                                 &siginfo);
 652                 break;
 653
 654         case PTRACE_SETSIGINFO:
 655                 memset(&siginfo, 0, sizeof siginfo);
 656                 if (copy_siginfo_from_user32(
 657                             &siginfo, (struct compat_siginfo __user *) datap))
 658                         ret = -EFAULT;
 659                 else
 660                         ret = ptrace_setsiginfo(child, &siginfo);
 661                 break;
 662
 663         default:
 664                 ret = ptrace_request(child, request, addr, data);
 665         }
 666
 667         return ret;
 668 }
 669
 670 asmlinkage long compat_sys_ptrace(compat_long_t request, compat_long_t pid,
 671                                   compat_long_t addr, compat_long_t data)
 672 {
 673         struct task_struct *child;
 674         long ret;
 675
 676         /*
 677          * This lock_kernel fixes a subtle race with suid exec
 678          */
 679         lock_kernel();
 680         if (request == PTRACE_TRACEME) {
 681                 ret = ptrace_traceme();
 682                 goto out;
 683         }
 684
 685         child = ptrace_get_task_struct(pid);
 686         if (IS_ERR(child)) {
 687                 ret = PTR_ERR(child);
 688                 goto out;
 689         }
 690
 691         if (request == PTRACE_ATTACH) {
 692                 ret = ptrace_attach(child);
 693                 /*
 694                  * Some architectures need to do book-keeping after
 695                  * a ptrace attach.
 696                  */
 697                 if (!ret)
 698                         arch_ptrace_attach(child);
 699                 goto out_put_task_struct;
 700         }
 701
 702         ret = ptrace_check_attach(child, request == PTRACE_KILL);
 703         if (!ret)
 704                 ret = compat_arch_ptrace(child, request, addr, data);
 705
 706  out_put_task_struct:
 707         put_task_struct(child);
 708  out:
 709         unlock_kernel();
 710         return ret;
 711 }
 712 #endif  /* CONFIG_COMPAT && __ARCH_WANT_COMPAT_SYS_PTRACE */