kernel/fs/coredump.c

   1 #include <linux/slab.h>
   2 #include <linux/file.h>
   3 #include <linux/fdtable.h>
   4 #include <linux/freezer.h>
   5 #include <linux/mm.h>
   6 #include <linux/stat.h>
   7 #include <linux/fcntl.h>
   8 #include <linux/swap.h>
   9 #include <linux/string.h>
  10 #include <linux/init.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/perf_event.h>
  13 #include <linux/highmem.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/key.h>
  16 #include <linux/personality.h>
  17 #include <linux/binfmts.h>
  18 #include <linux/coredump.h>
  19 #include <linux/utsname.h>
  20 #include <linux/pid_namespace.h>
  21 #include <linux/module.h>
  22 #include <linux/namei.h>
  23 #include <linux/mount.h>
  24 #include <linux/security.h>
  25 #include <linux/syscalls.h>
  26 #include <linux/tsacct_kern.h>
  27 #include <linux/cn_proc.h>
  28 #include <linux/audit.h>
  29 #include <linux/tracehook.h>
  30 #include <linux/kmod.h>
  31 #include <linux/fsnotify.h>
  32 #include <linux/fs_struct.h>
  33 #include <linux/pipe_fs_i.h>
  34 #include <linux/oom.h>
  35 #include <linux/compat.h>
  36 #include <linux/sched.h>
  37 #include <linux/fs.h>
  38 #include <linux/path.h>
  39
  40 #include <asm/uaccess.h>
  41 #include <asm/mmu_context.h>
  42 #include <asm/tlb.h>
  43 #include <asm/exec.h>
  44
  45 #include <trace/events/task.h>
  46 #include "internal.h"
  47
  48 #include <trace/events/sched.h>
  49
  50 int core_uses_pid;
  51 unsigned int core_pipe_limit;
  52 char core_pattern[CORENAME_MAX_SIZE] = "core";
  53 static int core_name_size = CORENAME_MAX_SIZE;
  54
  55 struct core_name {
  56         char *corename;
  57         int used, size;
  58 };
  59
  60 /* The maximal length of core_pattern is also specified in sysctl.c */
  61
  62 static int expand_corename(struct core_name *cn, int size)
  63 {
  64         char *corename = krealloc(cn->corename, size, GFP_KERNEL);
  65
  66         if (!corename)
  67                 return -ENOMEM;
  68
  69         if (size > core_name_size) /* racy but harmless */
  70                 core_name_size = size;
  71
  72         cn->size = ksize(corename);
  73         cn->corename = corename;
  74         return 0;
  75 }
  76
  77 static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
  78                                      va_list arg)
  79 {
  80         int free, need;
  81         va_list arg_copy;
  82
  83 again:
  84         free = cn->size - cn->used;
  85
  86         va_copy(arg_copy, arg);
  87         need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
  88         va_end(arg_copy);
  89
  90         if (need < free) {
  91                 cn->used += need;
  92                 return 0;
  93         }
  94
  95         if (!expand_corename(cn, cn->size + need - free + 1))
  96                 goto again;
  97
  98         return -ENOMEM;
  99 }
 100
 101 static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
 102 {
 103         va_list arg;
 104         int ret;
 105
 106         va_start(arg, fmt);
 107         ret = cn_vprintf(cn, fmt, arg);
 108         va_end(arg);
 109
 110         return ret;
 111 }
 112
 113 static __printf(2, 3)
 114 int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
 115 {
 116         int cur = cn->used;
 117         va_list arg;
 118         int ret;
 119
 120         va_start(arg, fmt);
 121         ret = cn_vprintf(cn, fmt, arg);
 122         va_end(arg);
 123
 124         for (; cur < cn->used; ++cur) {
 125                 if (cn->corename[cur] == '/')
 126                         cn->corename[cur] = '!';
 127         }
 128         return ret;
 129 }
 130
 131 static int cn_print_exe_file(struct core_name *cn)
 132 {
 133         struct file *exe_file;
 134         char *pathbuf, *path;
 135         int ret;
 136
 137         exe_file = get_mm_exe_file(current->mm);
 138         if (!exe_file)
 139                 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
 140
 141         pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
 142         if (!pathbuf) {
 143                 ret = -ENOMEM;
 144                 goto put_exe_file;
 145         }
 146
 147         path = file_path(exe_file, pathbuf, PATH_MAX);
 148         if (IS_ERR(path)) {
 149                 ret = PTR_ERR(path);
 150                 goto free_buf;
 151         }
 152
 153         ret = cn_esc_printf(cn, "%s", path);
 154
 155 free_buf:
 156         kfree(pathbuf);
 157 put_exe_file:
 158         fput(exe_file);
 159         return ret;
 160 }
 161
 162 /* format_corename will inspect the pattern parameter, and output a
 163  * name into corename, which must have space for at least
 164  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
 165  */
 166 static int format_corename(struct core_name *cn, struct coredump_params *cprm)
 167 {
 168         const struct cred *cred = current_cred();
 169         const char *pat_ptr = core_pattern;
 170         int ispipe = (*pat_ptr == '|');
 171         int pid_in_pattern = 0;
 172         int err = 0;
 173
 174         cn->used = 0;
 175         cn->corename = NULL;
 176         if (expand_corename(cn, core_name_size))
 177                 return -ENOMEM;
 178         cn->corename[0] = '\0';
 179
 180         if (ispipe)
 181                 ++pat_ptr;
 182
 183         /* Repeat as long as we have more pattern to process and more output
 184            space */
 185         while (*pat_ptr) {
 186                 if (*pat_ptr != '%') {
 187                         err = cn_printf(cn, "%c", *pat_ptr++);
 188                 } else {
 189                         switch (*++pat_ptr) {
 190                         /* single % at the end, drop that */
 191                         case 0:
 192                                 goto out;
 193                         /* Double percent, output one percent */
 194                         case '%':
 195                                 err = cn_printf(cn, "%c", '%');
 196                                 break;
 197                         /* pid */
 198                         case 'p':
 199                                 pid_in_pattern = 1;
 200                                 err = cn_printf(cn, "%d",
 201                                               task_tgid_vnr(current));
 202                                 break;
 203                         /* global pid */
 204                         case 'P':
 205                                 err = cn_printf(cn, "%d",
 206                                               task_tgid_nr(current));
 207                                 break;
 208                         case 'i':
 209                                 err = cn_printf(cn, "%d",
 210                                               task_pid_vnr(current));
 211                                 break;
 212                         case 'I':
 213                                 err = cn_printf(cn, "%d",
 214                                               task_pid_nr(current));
 215                                 break;
 216                         /* uid */
 217                         case 'u':
 218                                 err = cn_printf(cn, "%u",
 219                                                 from_kuid(&init_user_ns,
 220                                                           cred->uid));
 221                                 break;
 222                         /* gid */
 223                         case 'g':
 224                                 err = cn_printf(cn, "%u",
 225                                                 from_kgid(&init_user_ns,
 226                                                           cred->gid));
 227                                 break;
 228                         case 'd':
 229                                 err = cn_printf(cn, "%d",
 230                                         __get_dumpable(cprm->mm_flags));
 231                                 break;
 232                         /* signal that caused the coredump */
 233                         case 's':
 234                                 err = cn_printf(cn, "%d",
 235                                                 cprm->siginfo->si_signo);
 236                                 break;
 237                         /* UNIX time of coredump */
 238                         case 't': {
 239                                 struct timeval tv;
 240                                 do_gettimeofday(&tv);
 241                                 err = cn_printf(cn, "%lu", tv.tv_sec);
 242                                 break;
 243                         }
 244                         /* hostname */
 245                         case 'h':
 246                                 down_read(&uts_sem);
 247                                 err = cn_esc_printf(cn, "%s",
 248                                               utsname()->nodename);
 249                                 up_read(&uts_sem);
 250                                 break;
 251                         /* executable */
 252                         case 'e':
 253                                 err = cn_esc_printf(cn, "%s", current->comm);
 254                                 break;
 255                         case 'E':
 256                                 err = cn_print_exe_file(cn);
 257                                 break;
 258                         /* core limit size */
 259                         case 'c':
 260                                 err = cn_printf(cn, "%lu",
 261                                               rlimit(RLIMIT_CORE));
 262                                 break;
 263                         default:
 264                                 break;
 265                         }
 266                         ++pat_ptr;
 267                 }
 268
 269                 if (err)
 270                         return err;
 271         }
 272
 273 out:
 274         /* Backward compatibility with core_uses_pid:
 275          *
 276          * If core_pattern does not include a %p (as is the default)
 277          * and core_uses_pid is set, then .%pid will be appended to
 278          * the filename. Do not do this for piped commands. */
 279         if (!ispipe && !pid_in_pattern && core_uses_pid) {
 280                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
 281                 if (err)
 282                         return err;
 283         }
 284         return ispipe;
 285 }
 286
 287 static int zap_process(struct task_struct *start, int exit_code, int flags)
 288 {
 289         struct task_struct *t;
 290         int nr = 0;
 291
 292         /* ignore all signals except SIGKILL, see prepare_signal() */
 293         start->signal->flags = SIGNAL_GROUP_COREDUMP | flags;
 294         start->signal->group_exit_code = exit_code;
 295         start->signal->group_stop_count = 0;
 296
 297         for_each_thread(start, t) {
 298                 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
 299                 if (t != current && t->mm) {
 300                         sigaddset(&t->pending.signal, SIGKILL);
 301                         signal_wake_up(t, 1);
 302                         nr++;
 303                 }
 304         }
 305
 306         return nr;
 307 }
 308
 309 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
 310                         struct core_state *core_state, int exit_code)
 311 {
 312         struct task_struct *g, *p;
 313         unsigned long flags;
 314         int nr = -EAGAIN;
 315
 316         spin_lock_irq(&tsk->sighand->siglock);
 317         if (!signal_group_exit(tsk->signal)) {
 318                 mm->core_state = core_state;
 319                 tsk->signal->group_exit_task = tsk;
 320                 nr = zap_process(tsk, exit_code, 0);
 321                 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
 322         }
 323         spin_unlock_irq(&tsk->sighand->siglock);
 324         if (unlikely(nr < 0))
 325                 return nr;
 326
 327         tsk->flags |= PF_DUMPCORE;
 328         if (atomic_read(&mm->mm_users) == nr + 1)
 329                 goto done;
 330         /*
 331          * We should find and kill all tasks which use this mm, and we should
 332          * count them correctly into ->nr_threads. We don't take tasklist
 333          * lock, but this is safe wrt:
 334          *
 335          * fork:
 336          *      None of sub-threads can fork after zap_process(leader). All
 337          *      processes which were created before this point should be
 338          *      visible to zap_threads() because copy_process() adds the new
 339          *      process to the tail of init_task.tasks list, and lock/unlock
 340          *      of ->siglock provides a memory barrier.
 341          *
 342          * do_exit:
 343          *      The caller holds mm->mmap_sem. This means that the task which
 344          *      uses this mm can't pass exit_mm(), so it can't exit or clear
 345          *      its ->mm.
 346          *
 347          * de_thread:
 348          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
 349          *      we must see either old or new leader, this does not matter.
 350          *      However, it can change p->sighand, so lock_task_sighand(p)
 351          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
 352          *      it can't fail.
 353          *
 354          *      Note also that "g" can be the old leader with ->mm == NULL
 355          *      and already unhashed and thus removed from ->thread_group.
 356          *      This is OK, __unhash_process()->list_del_rcu() does not
 357          *      clear the ->next pointer, we will find the new leader via
 358          *      next_thread().
 359          */
 360         rcu_read_lock();
 361         for_each_process(g) {
 362                 if (g == tsk->group_leader)
 363                         continue;
 364                 if (g->flags & PF_KTHREAD)
 365                         continue;
 366
 367                 for_each_thread(g, p) {
 368                         if (unlikely(!p->mm))
 369                                 continue;
 370                         if (unlikely(p->mm == mm)) {
 371                                 lock_task_sighand(p, &flags);
 372                                 nr += zap_process(p, exit_code,
 373                                                         SIGNAL_GROUP_EXIT);
 374                                 unlock_task_sighand(p, &flags);
 375                         }
 376                         break;
 377                 }
 378         }
 379         rcu_read_unlock();
 380 done:
 381         atomic_set(&core_state->nr_threads, nr);
 382         return nr;
 383 }
 384
 385 static int coredump_wait(int exit_code, struct core_state *core_state)
 386 {
 387         struct task_struct *tsk = current;
 388         struct mm_struct *mm = tsk->mm;
 389         int core_waiters = -EBUSY;
 390
 391         init_completion(&core_state->startup);
 392         core_state->dumper.task = tsk;
 393         core_state->dumper.next = NULL;
 394
 395         down_write(&mm->mmap_sem);
 396         if (!mm->core_state)
 397                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
 398         up_write(&mm->mmap_sem);
 399
 400         if (core_waiters > 0) {
 401                 struct core_thread *ptr;
 402
 403                 freezer_do_not_count();
 404                 wait_for_completion(&core_state->startup);
 405                 freezer_count();
 406                 /*
 407                  * Wait for all the threads to become inactive, so that
 408                  * all the thread context (extended register state, like
 409                  * fpu etc) gets copied to the memory.
 410                  */
 411                 ptr = core_state->dumper.next;
 412                 while (ptr != NULL) {
 413                         wait_task_inactive(ptr->task, 0);
 414                         ptr = ptr->next;
 415                 }
 416         }
 417
 418         return core_waiters;
 419 }
 420
 421 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
 422 {
 423         struct core_thread *curr, *next;
 424         struct task_struct *task;
 425
 426         spin_lock_irq(&current->sighand->siglock);
 427         if (core_dumped && !__fatal_signal_pending(current))
 428                 current->signal->group_exit_code |= 0x80;
 429         current->signal->group_exit_task = NULL;
 430         current->signal->flags = SIGNAL_GROUP_EXIT;
 431         spin_unlock_irq(&current->sighand->siglock);
 432
 433         next = mm->core_state->dumper.next;
 434         while ((curr = next) != NULL) {
 435                 next = curr->next;
 436                 task = curr->task;
 437                 /*
 438                  * see exit_mm(), curr->task must not see
 439                  * ->task == NULL before we read ->next.
 440                  */
 441                 smp_mb();
 442                 curr->task = NULL;
 443                 wake_up_process(task);
 444         }
 445
 446         mm->core_state = NULL;
 447 }
 448
 449 static bool dump_interrupted(void)
 450 {
 451         /*
 452          * SIGKILL or freezing() interrupt the coredumping. Perhaps we
 453          * can do try_to_freeze() and check __fatal_signal_pending(),
 454          * but then we need to teach dump_write() to restart and clear
 455          * TIF_SIGPENDING.
 456          */
 457         return signal_pending(current);
 458 }
 459
 460 static void wait_for_dump_helpers(struct file *file)
 461 {
 462         struct pipe_inode_info *pipe = file->private_data;
 463
 464         pipe_lock(pipe);
 465         pipe->readers++;
 466         pipe->writers--;
 467         wake_up_interruptible_sync(&pipe->wait);
 468         kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 469         pipe_unlock(pipe);
 470
 471         /*
 472          * We actually want wait_event_freezable() but then we need
 473          * to clear TIF_SIGPENDING and improve dump_interrupted().
 474          */
 475         wait_event_interruptible(pipe->wait, pipe->readers == 1);
 476
 477         pipe_lock(pipe);
 478         pipe->readers--;
 479         pipe->writers++;
 480         pipe_unlock(pipe);
 481 }
 482
 483 /*
 484  * umh_pipe_setup
 485  * helper function to customize the process used
 486  * to collect the core in userspace.  Specifically
 487  * it sets up a pipe and installs it as fd 0 (stdin)
 488  * for the process.  Returns 0 on success, or
 489  * PTR_ERR on failure.
 490  * Note that it also sets the core limit to 1.  This
 491  * is a special value that we use to trap recursive
 492  * core dumps
 493  */
 494 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
 495 {
 496         struct file *files[2];
 497         struct coredump_params *cp = (struct coredump_params *)info->data;
 498         int err = create_pipe_files(files, 0);
 499         if (err)
 500                 return err;
 501
 502         cp->file = files[1];
 503
 504         err = replace_fd(0, files[0], 0);
 505         fput(files[0]);
 506         /* and disallow core files too */
 507         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
 508
 509         return err;
 510 }
 511
 512 void do_coredump(const siginfo_t *siginfo)
 513 {
 514         struct core_state core_state;
 515         struct core_name cn;
 516         struct mm_struct *mm = current->mm;
 517         struct linux_binfmt * binfmt;
 518         const struct cred *old_cred;
 519         struct cred *cred;
 520         int retval = 0;
 521         int ispipe;
 522         struct files_struct *displaced;
 523         /* require nonrelative corefile path and be extra careful */
 524         bool need_suid_safe = false;
 525         bool core_dumped = false;
 526         static atomic_t core_dump_count = ATOMIC_INIT(0);
 527         struct coredump_params cprm = {
 528                 .siginfo = siginfo,
 529                 .regs = signal_pt_regs(),
 530                 .limit = rlimit(RLIMIT_CORE),
 531                 /*
 532                  * We must use the same mm->flags while dumping core to avoid
 533                  * inconsistency of bit flags, since this flag is not protected
 534                  * by any locks.
 535                  */
 536                 .mm_flags = mm->flags,
 537         };
 538
 539         audit_core_dumps(siginfo->si_signo);
 540
 541         binfmt = mm->binfmt;
 542         if (!binfmt || !binfmt->core_dump)
 543                 goto fail;
 544         if (!__get_dumpable(cprm.mm_flags))
 545                 goto fail;
 546
 547         cred = prepare_creds();
 548         if (!cred)
 549                 goto fail;
 550         /*
 551          * We cannot trust fsuid as being the "true" uid of the process
 552          * nor do we know its entire history. We only know it was tainted
 553          * so we dump it as root in mode 2, and only into a controlled
 554          * environment (pipe handler or fully qualified path).
 555          */
 556         if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
 557                 /* Setuid core dump mode */
 558                 cred->fsuid = GLOBAL_ROOT_UID;  /* Dump root private */
 559                 need_suid_safe = true;
 560         }
 561
 562         retval = coredump_wait(siginfo->si_signo, &core_state);
 563         if (retval < 0)
 564                 goto fail_creds;
 565
 566         old_cred = override_creds(cred);
 567
 568         ispipe = format_corename(&cn, &cprm);
 569
 570         if (ispipe) {
 571                 int dump_count;
 572                 char **helper_argv;
 573                 struct subprocess_info *sub_info;
 574
 575                 if (ispipe < 0) {
 576                         printk(KERN_WARNING "format_corename failed\n");
 577                         printk(KERN_WARNING "Aborting core\n");
 578                         goto fail_unlock;
 579                 }
 580
 581                 if (cprm.limit == 1) {
 582                         /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
 583                          *
 584                          * Normally core limits are irrelevant to pipes, since
 585                          * we're not writing to the file system, but we use
 586                          * cprm.limit of 1 here as a special value, this is a
 587                          * consistent way to catch recursive crashes.
 588                          * We can still crash if the core_pattern binary sets
 589                          * RLIM_CORE = !1, but it runs as root, and can do
 590                          * lots of stupid things.
 591                          *
 592                          * Note that we use task_tgid_vnr here to grab the pid
 593                          * of the process group leader.  That way we get the
 594                          * right pid if a thread in a multi-threaded
 595                          * core_pattern process dies.
 596                          */
 597                         printk(KERN_WARNING
 598                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
 599                                 task_tgid_vnr(current), current->comm);
 600                         printk(KERN_WARNING "Aborting core\n");
 601                         goto fail_unlock;
 602                 }
 603                 cprm.limit = RLIM_INFINITY;
 604
 605                 dump_count = atomic_inc_return(&core_dump_count);
 606                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
 607                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
 608                                task_tgid_vnr(current), current->comm);
 609                         printk(KERN_WARNING "Skipping core dump\n");
 610                         goto fail_dropcount;
 611                 }
 612
 613                 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
 614                 if (!helper_argv) {
 615                         printk(KERN_WARNING "%s failed to allocate memory\n",
 616                                __func__);
 617                         goto fail_dropcount;
 618                 }
 619
 620                 retval = -ENOMEM;
 621                 sub_info = call_usermodehelper_setup(helper_argv[0],
 622                                                 helper_argv, NULL, GFP_KERNEL,
 623                                                 umh_pipe_setup, NULL, &cprm);
 624                 if (sub_info)
 625                         retval = call_usermodehelper_exec(sub_info,
 626                                                           UMH_WAIT_EXEC);
 627
 628                 argv_free(helper_argv);
 629                 if (retval) {
 630                         printk(KERN_INFO "Core dump to |%s pipe failed\n",
 631                                cn.corename);
 632                         goto close_fail;
 633                 }
 634         } else {
 635                 struct inode *inode;
 636                 int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
 637                                  O_LARGEFILE | O_EXCL;
 638
 639                 if (cprm.limit < binfmt->min_coredump)
 640                         goto fail_unlock;
 641
 642                 if (need_suid_safe && cn.corename[0] != '/') {
 643                         printk(KERN_WARNING "Pid %d(%s) can only dump core "\
 644                                 "to fully qualified path!\n",
 645                                 task_tgid_vnr(current), current->comm);
 646                         printk(KERN_WARNING "Skipping core dump\n");
 647                         goto fail_unlock;
 648                 }
 649
 650                 /*
 651                  * Unlink the file if it exists unless this is a SUID
 652                  * binary - in that case, we're running around with root
 653                  * privs and don't want to unlink another user's coredump.
 654                  */
 655                 if (!need_suid_safe) {
 656                         mm_segment_t old_fs;
 657
 658                         old_fs = get_fs();
 659                         set_fs(KERNEL_DS);
 660                         /*
 661                          * If it doesn't exist, that's fine. If there's some
 662                          * other problem, we'll catch it at the filp_open().
 663                          */
 664                         (void) sys_unlink((const char __user *)cn.corename);
 665                         set_fs(old_fs);
 666                 }
 667
 668                 /*
 669                  * There is a race between unlinking and creating the
 670                  * file, but if that causes an EEXIST here, that's
 671                  * fine - another process raced with us while creating
 672                  * the corefile, and the other process won. To userspace,
 673                  * what matters is that at least one of the two processes
 674                  * writes its coredump successfully, not which one.
 675                  */
 676                 if (need_suid_safe) {
 677                         /*
 678                          * Using user namespaces, normal user tasks can change
 679                          * their current->fs->root to point to arbitrary
 680                          * directories. Since the intention of the "only dump
 681                          * with a fully qualified path" rule is to control where
 682                          * coredumps may be placed using root privileges,
 683                          * current->fs->root must not be used. Instead, use the
 684                          * root directory of init_task.
 685                          */
 686                         struct path root;
 687
 688                         task_lock(&init_task);
 689                         get_fs_root(init_task.fs, &root);
 690                         task_unlock(&init_task);
 691                         cprm.file = file_open_root(root.dentry, root.mnt,
 692                                 cn.corename, open_flags, 0600);
 693                         path_put(&root);
 694                 } else {
 695                         cprm.file = filp_open(cn.corename, open_flags, 0600);
 696                 }
 697                 if (IS_ERR(cprm.file))
 698                         goto fail_unlock;
 699
 700                 inode = file_inode(cprm.file);
 701                 if (inode->i_nlink > 1)
 702                         goto close_fail;
 703                 if (d_unhashed(cprm.file->f_path.dentry))
 704                         goto close_fail;
 705                 /*
 706                  * AK: actually i see no reason to not allow this for named
 707                  * pipes etc, but keep the previous behaviour for now.
 708                  */
 709                 if (!S_ISREG(inode->i_mode))
 710                         goto close_fail;
 711                 /*
 712                  * Don't dump core if the filesystem changed owner or mode
 713                  * of the file during file creation. This is an issue when
 714                  * a process dumps core while its cwd is e.g. on a vfat
 715                  * filesystem.
 716                  */
 717                 if (!uid_eq(inode->i_uid, current_fsuid()))
 718                         goto close_fail;
 719                 if ((inode->i_mode & 0677) != 0600)
 720                         goto close_fail;
 721                 if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
 722                         goto close_fail;
 723                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
 724                         goto close_fail;
 725         }
 726
 727         /* get us an unshared descriptor table; almost always a no-op */
 728         retval = unshare_files(&displaced);
 729         if (retval)
 730                 goto close_fail;
 731         if (displaced)
 732                 put_files_struct(displaced);
 733         if (!dump_interrupted()) {
 734                 file_start_write(cprm.file);
 735                 core_dumped = binfmt->core_dump(&cprm);
 736                 file_end_write(cprm.file);
 737         }
 738         if (ispipe && core_pipe_limit)
 739                 wait_for_dump_helpers(cprm.file);
 740 close_fail:
 741         if (cprm.file)
 742                 filp_close(cprm.file, NULL);
 743 fail_dropcount:
 744         if (ispipe)
 745                 atomic_dec(&core_dump_count);
 746 fail_unlock:
 747         kfree(cn.corename);
 748         coredump_finish(mm, core_dumped);
 749         revert_creds(old_cred);
 750 fail_creds:
 751         put_cred(cred);
 752 fail:
 753         return;
 754 }
 755
 756 /*
 757  * Core dumping helper functions.  These are the only things you should
 758  * do on a core-file: use only these functions to write out all the
 759  * necessary info.
 760  */
 761 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
 762 {
 763         struct file *file = cprm->file;
 764         loff_t pos = file->f_pos;
 765         ssize_t n;
 766         if (cprm->written + nr > cprm->limit)
 767                 return 0;
 768         while (nr) {
 769                 if (dump_interrupted())
 770                         return 0;
 771                 n = __kernel_write(file, addr, nr, &pos);
 772                 if (n <= 0)
 773                         return 0;
 774                 file->f_pos = pos;
 775                 cprm->written += n;
 776                 nr -= n;
 777         }
 778         return 1;
 779 }
 780 EXPORT_SYMBOL(dump_emit);
 781
 782 int dump_skip(struct coredump_params *cprm, size_t nr)
 783 {
 784         static char zeroes[PAGE_SIZE];
 785         struct file *file = cprm->file;
 786         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
 787                 if (cprm->written + nr > cprm->limit)
 788                         return 0;
 789                 if (dump_interrupted() ||
 790                     file->f_op->llseek(file, nr, SEEK_CUR) < 0)
 791                         return 0;
 792                 cprm->written += nr;
 793                 return 1;
 794         } else {
 795                 while (nr > PAGE_SIZE) {
 796                         if (!dump_emit(cprm, zeroes, PAGE_SIZE))
 797                                 return 0;
 798                         nr -= PAGE_SIZE;
 799                 }
 800                 return dump_emit(cprm, zeroes, nr);
 801         }
 802 }
 803 EXPORT_SYMBOL(dump_skip);
 804
 805 int dump_align(struct coredump_params *cprm, int align)
 806 {
 807         unsigned mod = cprm->written & (align - 1);
 808         if (align & (align - 1))
 809                 return 0;
 810         return mod ? dump_skip(cprm, align - mod) : 1;
 811 }
 812 EXPORT_SYMBOL(dump_align);