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         const struct cred *cred = current_cred(), *tcred;
 119
 120         /* May we inspect the given task?
 121          * This check is used both for attaching with ptrace
 122          * and for allowing access to sensitive information in /proc.
 123          *
 124          * ptrace_attach denies several cases that /proc allows
 125          * because setting up the necessary parent/child relationship
 126          * or halting the specified task is impossible.
 127          */
 128         int dumpable = 0;
 129         /* Don't let security modules deny introspection */
 130         if (task == current)
 131                 return 0;
 132         rcu_read_lock();
 133         tcred = __task_cred(task);
 134         if ((cred->uid != tcred->euid ||
 135              cred->uid != tcred->suid ||
 136              cred->uid != tcred->uid  ||
 137              cred->gid != tcred->egid ||
 138              cred->gid != tcred->sgid ||
 139              cred->gid != tcred->gid) &&
 140             !capable(CAP_SYS_PTRACE)) {
 141                 rcu_read_unlock();
 142                 return -EPERM;
 143         }
 144         rcu_read_unlock();
 145         smp_rmb();
 146         if (task->mm)
 147                 dumpable = get_dumpable(task->mm);
 148         if (!dumpable && !capable(CAP_SYS_PTRACE))
 149                 return -EPERM;
 150
 151         return security_ptrace_may_access(task, mode);
 152 }
 153
 154 bool ptrace_may_access(struct task_struct *task, unsigned int mode)
 155 {
 156         int err;
 157         task_lock(task);
 158         err = __ptrace_may_access(task, mode);
 159         task_unlock(task);
 160         return (!err ? true : false);
 161 }
 162
 163 int ptrace_attach(struct task_struct *task)
 164 {
 165         int retval;
 166         unsigned long flags;
 167
 168         audit_ptrace(task);
 169
 170         retval = -EPERM;
 171         if (same_thread_group(task, current))
 172                 goto out;
 173
 174 repeat:
 175         /*
 176          * Nasty, nasty.
 177          *
 178          * We want to hold both the task-lock and the
 179          * tasklist_lock for writing at the same time.
 180          * But that's against the rules (tasklist_lock
 181          * is taken for reading by interrupts on other
 182          * cpu's that may have task_lock).
 183          */
 184         task_lock(task);
 185         if (!write_trylock_irqsave(&tasklist_lock, flags)) {
 186                 task_unlock(task);
 187                 do {
 188                         cpu_relax();
 189                 } while (!write_can_lock(&tasklist_lock));
 190                 goto repeat;
 191         }
 192
 193         if (!task->mm)
 194                 goto bad;
 195         /* the same process cannot be attached many times */
 196         if (task->ptrace & PT_PTRACED)
 197                 goto bad;
 198         retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH);
 199         if (retval)
 200                 goto bad;
 201
 202         /* Go */
 203         task->ptrace |= PT_PTRACED;
 204         if (capable(CAP_SYS_PTRACE))
 205                 task->ptrace |= PT_PTRACE_CAP;
 206
 207         __ptrace_link(task, current);
 208
 209         send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
 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 *info)
 324 {
 325         int error = -ESRCH;
 326
 327         read_lock(&tasklist_lock);
 328         if (likely(child->sighand != NULL)) {
 329                 error = -EINVAL;
 330                 spin_lock_irq(&child->sighand->siglock);
 331                 if (likely(child->last_siginfo != NULL)) {
 332                         *info = *child->last_siginfo;
 333                         error = 0;
 334                 }
 335                 spin_unlock_irq(&child->sighand->siglock);
 336         }
 337         read_unlock(&tasklist_lock);
 338         return error;
 339 }
 340
 341 static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
 342 {
 343         int error = -ESRCH;
 344
 345         read_lock(&tasklist_lock);
 346         if (likely(child->sighand != NULL)) {
 347                 error = -EINVAL;
 348                 spin_lock_irq(&child->sighand->siglock);
 349                 if (likely(child->last_siginfo != NULL)) {
 350                         *child->last_siginfo = *info;
 351                         error = 0;
 352                 }
 353                 spin_unlock_irq(&child->sighand->siglock);
 354         }
 355         read_unlock(&tasklist_lock);
 356         return error;
 357 }
 358
 359
 360 #ifdef PTRACE_SINGLESTEP
 361 #define is_singlestep(request)          ((request) == PTRACE_SINGLESTEP)
 362 #else
 363 #define is_singlestep(request)          0
 364 #endif
 365
 366 #ifdef PTRACE_SINGLEBLOCK
 367 #define is_singleblock(request)         ((request) == PTRACE_SINGLEBLOCK)
 368 #else
 369 #define is_singleblock(request)         0
 370 #endif
 371
 372 #ifdef PTRACE_SYSEMU
 373 #define is_sysemu_singlestep(request)   ((request) == PTRACE_SYSEMU_SINGLESTEP)
 374 #else
 375 #define is_sysemu_singlestep(request)   0
 376 #endif
 377
 378 static int ptrace_resume(struct task_struct *child, long request, long data)
 379 {
 380         if (!valid_signal(data))
 381                 return -EIO;
 382
 383         if (request == PTRACE_SYSCALL)
 384                 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 385         else
 386                 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 387
 388 #ifdef TIF_SYSCALL_EMU
 389         if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
 390                 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 391         else
 392                 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 393 #endif
 394
 395         if (is_singleblock(request)) {
 396                 if (unlikely(!arch_has_block_step()))
 397                         return -EIO;
 398                 user_enable_block_step(child);
 399         } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
 400                 if (unlikely(!arch_has_single_step()))
 401                         return -EIO;
 402                 user_enable_single_step(child);
 403         }
 404         else
 405                 user_disable_single_step(child);
 406
 407         child->exit_code = data;
 408         wake_up_process(child);
 409
 410         return 0;
 411 }
 412
 413 int ptrace_request(struct task_struct *child, long request,
 414                    long addr, long data)
 415 {
 416         int ret = -EIO;
 417         siginfo_t siginfo;
 418
 419         switch (request) {
 420         case PTRACE_PEEKTEXT:
 421         case PTRACE_PEEKDATA:
 422                 return generic_ptrace_peekdata(child, addr, data);
 423         case PTRACE_POKETEXT:
 424         case PTRACE_POKEDATA:
 425                 return generic_ptrace_pokedata(child, addr, data);
 426
 427 #ifdef PTRACE_OLDSETOPTIONS
 428         case PTRACE_OLDSETOPTIONS:
 429 #endif
 430         case PTRACE_SETOPTIONS:
 431                 ret = ptrace_setoptions(child, data);
 432                 break;
 433         case PTRACE_GETEVENTMSG:
 434                 ret = put_user(child->ptrace_message, (unsigned long __user *) data);
 435                 break;
 436
 437         case PTRACE_GETSIGINFO:
 438                 ret = ptrace_getsiginfo(child, &siginfo);
 439                 if (!ret)
 440                         ret = copy_siginfo_to_user((siginfo_t __user *) data,
 441                                                    &siginfo);
 442                 break;
 443
 444         case PTRACE_SETSIGINFO:
 445                 if (copy_from_user(&siginfo, (siginfo_t __user *) data,
 446                                    sizeof siginfo))
 447                         ret = -EFAULT;
 448                 else
 449                         ret = ptrace_setsiginfo(child, &siginfo);
 450                 break;
 451
 452         case PTRACE_DETACH:      /* detach a process that was attached. */
 453                 ret = ptrace_detach(child, data);
 454                 break;
 455
 456 #ifdef PTRACE_SINGLESTEP
 457         case PTRACE_SINGLESTEP:
 458 #endif
 459 #ifdef PTRACE_SINGLEBLOCK
 460         case PTRACE_SINGLEBLOCK:
 461 #endif
 462 #ifdef PTRACE_SYSEMU
 463         case PTRACE_SYSEMU:
 464         case PTRACE_SYSEMU_SINGLESTEP:
 465 #endif
 466         case PTRACE_SYSCALL:
 467         case PTRACE_CONT:
 468                 return ptrace_resume(child, request, data);
 469
 470         case PTRACE_KILL:
 471                 if (child->exit_state)  /* already dead */
 472                         return 0;
 473                 return ptrace_resume(child, request, SIGKILL);
 474
 475         default:
 476                 break;
 477         }
 478
 479         return ret;
 480 }
 481
 482 /**
 483  * ptrace_traceme  --  helper for PTRACE_TRACEME
 484  *
 485  * Performs checks and sets PT_PTRACED.
 486  * Should be used by all ptrace implementations for PTRACE_TRACEME.
 487  */
 488 int ptrace_traceme(void)
 489 {
 490         int ret = -EPERM;
 491
 492         /*
 493          * Are we already being traced?
 494          */
 495 repeat:
 496         task_lock(current);
 497         if (!(current->ptrace & PT_PTRACED)) {
 498                 /*
 499                  * See ptrace_attach() comments about the locking here.
 500                  */
 501                 unsigned long flags;
 502                 if (!write_trylock_irqsave(&tasklist_lock, flags)) {
 503                         task_unlock(current);
 504                         do {
 505                                 cpu_relax();
 506                         } while (!write_can_lock(&tasklist_lock));
 507                         goto repeat;
 508                 }
 509
 510                 ret = security_ptrace_traceme(current->parent);
 511
 512                 /*
 513                  * Set the ptrace bit in the process ptrace flags.
 514                  * Then link us on our parent's ptraced list.
 515                  */
 516                 if (!ret) {
 517                         current->ptrace |= PT_PTRACED;
 518                         __ptrace_link(current, current->real_parent);
 519                 }
 520
 521                 write_unlock_irqrestore(&tasklist_lock, flags);
 522         }
 523         task_unlock(current);
 524         return ret;
 525 }
 526
 527 /**
 528  * ptrace_get_task_struct  --  grab a task struct reference for ptrace
 529  * @pid:       process id to grab a task_struct reference of
 530  *
 531  * This function is a helper for ptrace implementations.  It checks
 532  * permissions and then grabs a task struct for use of the actual
 533  * ptrace implementation.
 534  *
 535  * Returns the task_struct for @pid or an ERR_PTR() on failure.
 536  */
 537 struct task_struct *ptrace_get_task_struct(pid_t pid)
 538 {
 539         struct task_struct *child;
 540
 541         read_lock(&tasklist_lock);
 542         child = find_task_by_vpid(pid);
 543         if (child)
 544                 get_task_struct(child);
 545
 546         read_unlock(&tasklist_lock);
 547         if (!child)
 548                 return ERR_PTR(-ESRCH);
 549         return child;
 550 }
 551
 552 #ifndef arch_ptrace_attach
 553 #define arch_ptrace_attach(child)       do { } while (0)
 554 #endif
 555
 556 asmlinkage long sys_ptrace(long request, long pid, long addr, long data)
 557 {
 558         struct task_struct *child;
 559         long ret;
 560
 561         /*
 562          * This lock_kernel fixes a subtle race with suid exec
 563          */
 564         lock_kernel();
 565         if (request == PTRACE_TRACEME) {
 566                 ret = ptrace_traceme();
 567                 if (!ret)
 568                         arch_ptrace_attach(current);
 569                 goto out;
 570         }
 571
 572         child = ptrace_get_task_struct(pid);
 573         if (IS_ERR(child)) {
 574                 ret = PTR_ERR(child);
 575                 goto out;
 576         }
 577
 578         if (request == PTRACE_ATTACH) {
 579                 ret = ptrace_attach(child);
 580                 /*
 581                  * Some architectures need to do book-keeping after
 582                  * a ptrace attach.
 583                  */
 584                 if (!ret)
 585                         arch_ptrace_attach(child);
 586                 goto out_put_task_struct;
 587         }
 588
 589         ret = ptrace_check_attach(child, request == PTRACE_KILL);
 590         if (ret < 0)
 591                 goto out_put_task_struct;
 592
 593         ret = arch_ptrace(child, request, addr, data);
 594         if (ret < 0)
 595                 goto out_put_task_struct;
 596
 597  out_put_task_struct:
 598         put_task_struct(child);
 599  out:
 600         unlock_kernel();
 601         return ret;
 602 }
 603
 604 int generic_ptrace_peekdata(struct task_struct *tsk, long addr, long data)
 605 {
 606         unsigned long tmp;
 607         int copied;
 608
 609         copied = access_process_vm(tsk, addr, &tmp, sizeof(tmp), 0);
 610         if (copied != sizeof(tmp))
 611                 return -EIO;
 612         return put_user(tmp, (unsigned long __user *)data);
 613 }
 614
 615 int generic_ptrace_pokedata(struct task_struct *tsk, long addr, long data)
 616 {
 617         int copied;
 618
 619         copied = access_process_vm(tsk, addr, &data, sizeof(data), 1);
 620         return (copied == sizeof(data)) ? 0 : -EIO;
 621 }
 622
 623 #if defined CONFIG_COMPAT && defined __ARCH_WANT_COMPAT_SYS_PTRACE
 624 #include <linux/compat.h>
 625
 626 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
 627                           compat_ulong_t addr, compat_ulong_t data)
 628 {
 629         compat_ulong_t __user *datap = compat_ptr(data);
 630         compat_ulong_t word;
 631         siginfo_t siginfo;
 632         int ret;
 633
 634         switch (request) {
 635         case PTRACE_PEEKTEXT:
 636         case PTRACE_PEEKDATA:
 637                 ret = access_process_vm(child, addr, &word, sizeof(word), 0);
 638                 if (ret != sizeof(word))
 639                         ret = -EIO;
 640                 else
 641                         ret = put_user(word, datap);
 642                 break;
 643
 644         case PTRACE_POKETEXT:
 645         case PTRACE_POKEDATA:
 646                 ret = access_process_vm(child, addr, &data, sizeof(data), 1);
 647                 ret = (ret != sizeof(data) ? -EIO : 0);
 648                 break;
 649
 650         case PTRACE_GETEVENTMSG:
 651                 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
 652                 break;
 653
 654         case PTRACE_GETSIGINFO:
 655                 ret = ptrace_getsiginfo(child, &siginfo);
 656                 if (!ret)
 657                         ret = copy_siginfo_to_user32(
 658                                 (struct compat_siginfo __user *) datap,
 659                                 &siginfo);
 660                 break;
 661
 662         case PTRACE_SETSIGINFO:
 663                 memset(&siginfo, 0, sizeof siginfo);
 664                 if (copy_siginfo_from_user32(
 665                             &siginfo, (struct compat_siginfo __user *) datap))
 666                         ret = -EFAULT;
 667                 else
 668                         ret = ptrace_setsiginfo(child, &siginfo);
 669                 break;
 670
 671         default:
 672                 ret = ptrace_request(child, request, addr, data);
 673         }
 674
 675         return ret;
 676 }
 677
 678 asmlinkage long compat_sys_ptrace(compat_long_t request, compat_long_t pid,
 679                                   compat_long_t addr, compat_long_t data)
 680 {
 681         struct task_struct *child;
 682         long ret;
 683
 684         /*
 685          * This lock_kernel fixes a subtle race with suid exec
 686          */
 687         lock_kernel();
 688         if (request == PTRACE_TRACEME) {
 689                 ret = ptrace_traceme();
 690                 goto out;
 691         }
 692
 693         child = ptrace_get_task_struct(pid);
 694         if (IS_ERR(child)) {
 695                 ret = PTR_ERR(child);
 696                 goto out;
 697         }
 698
 699         if (request == PTRACE_ATTACH) {
 700                 ret = ptrace_attach(child);
 701                 /*
 702                  * Some architectures need to do book-keeping after
 703                  * a ptrace attach.
 704                  */
 705                 if (!ret)
 706                         arch_ptrace_attach(child);
 707                 goto out_put_task_struct;
 708         }
 709
 710         ret = ptrace_check_attach(child, request == PTRACE_KILL);
 711         if (!ret)
 712                 ret = compat_arch_ptrace(child, request, addr, data);
 713
 714  out_put_task_struct:
 715         put_task_struct(child);
 716  out:
 717         unlock_kernel();
 718         return ret;
 719 }
 720 #endif  /* CONFIG_COMPAT && __ARCH_WANT_COMPAT_SYS_PTRACE */