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