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