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