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