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