fs/coredump.c

   1 #include <linux/slab.h>
   2 #include <linux/file.h>
   3 #include <linux/fdtable.h>
   4 #include <linux/mm.h>
   5 #include <linux/stat.h>
   6 #include <linux/fcntl.h>
   7 #include <linux/swap.h>
   8 #include <linux/string.h>
   9 #include <linux/init.h>
  10 #include <linux/pagemap.h>
  11 #include <linux/perf_event.h>
  12 #include <linux/highmem.h>
  13 #include <linux/spinlock.h>
  14 #include <linux/key.h>
  15 #include <linux/personality.h>
  16 #include <linux/binfmts.h>
  17 #include <linux/coredump.h>
  18 #include <linux/utsname.h>
  19 #include <linux/pid_namespace.h>
  20 #include <linux/module.h>
  21 #include <linux/namei.h>
  22 #include <linux/mount.h>
  23 #include <linux/security.h>
  24 #include <linux/syscalls.h>
  25 #include <linux/tsacct_kern.h>
  26 #include <linux/cn_proc.h>
  27 #include <linux/audit.h>
  28 #include <linux/tracehook.h>
  29 #include <linux/kmod.h>
  30 #include <linux/fsnotify.h>
  31 #include <linux/fs_struct.h>
  32 #include <linux/pipe_fs_i.h>
  33 #include <linux/oom.h>
  34 #include <linux/compat.h>
  35 #include <linux/timekeeping.h>
  36
  37 #include <asm/uaccess.h>
  38 #include <asm/mmu_context.h>
  39 #include <asm/tlb.h>
  40 #include <asm/exec.h>
  41
  42 #include <trace/events/task.h>
  43 #include "internal.h"
  44
  45 #include <trace/events/sched.h>
  46
  47 int core_uses_pid;
  48 unsigned int core_pipe_limit;
  49 char core_pattern[CORENAME_MAX_SIZE] = "core";
  50 static int core_name_size = CORENAME_MAX_SIZE;
  51
  52 struct core_name {
  53         char *corename;
  54         int used, size;
  55 };
  56
  57 /* The maximal length of core_pattern is also specified in sysctl.c */
  58
  59 static int expand_corename(struct core_name *cn, int size)
  60 {
  61         char *corename = krealloc(cn->corename, size, GFP_KERNEL);
  62
  63         if (!corename)
  64                 return -ENOMEM;
  65
  66         if (size > core_name_size) /* racy but harmless */
  67                 core_name_size = size;
  68
  69         cn->size = ksize(corename);
  70         cn->corename = corename;
  71         return 0;
  72 }
  73
  74 static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
  75                                      va_list arg)
  76 {
  77         int free, need;
  78         va_list arg_copy;
  79
  80 again:
  81         free = cn->size - cn->used;
  82
  83         va_copy(arg_copy, arg);
  84         need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
  85         va_end(arg_copy);
  86
  87         if (need < free) {
  88                 cn->used += need;
  89                 return 0;
  90         }
  91
  92         if (!expand_corename(cn, cn->size + need - free + 1))
  93                 goto again;
  94
  95         return -ENOMEM;
  96 }
  97
  98 static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
  99 {
 100         va_list arg;
 101         int ret;
 102
 103         va_start(arg, fmt);
 104         ret = cn_vprintf(cn, fmt, arg);
 105         va_end(arg);
 106
 107         return ret;
 108 }
 109
 110 static __printf(2, 3)
 111 int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
 112 {
 113         int cur = cn->used;
 114         va_list arg;
 115         int ret;
 116
 117         va_start(arg, fmt);
 118         ret = cn_vprintf(cn, fmt, arg);
 119         va_end(arg);
 120
 121         if (ret == 0) {
 122                 /*
 123                  * Ensure that this coredump name component can't cause the
 124                  * resulting corefile path to consist of a ".." or ".".
 125                  */
 126                 if ((cn->used - cur == 1 && cn->corename[cur] == '.') ||
 127                                 (cn->used - cur == 2 && cn->corename[cur] == '.'
 128                                 && cn->corename[cur+1] == '.'))
 129                         cn->corename[cur] = '!';
 130
 131                 /*
 132                  * Empty names are fishy and could be used to create a "//" in a
 133                  * corefile name, causing the coredump to happen one directory
 134                  * level too high. Enforce that all components of the core
 135                  * pattern are at least one character long.
 136                  */
 137                 if (cn->used == cur)
 138                         ret = cn_printf(cn, "!");
 139         }
 140
 141         for (; cur < cn->used; ++cur) {
 142                 if (cn->corename[cur] == '/')
 143                         cn->corename[cur] = '!';
 144         }
 145         return ret;
 146 }
 147
 148 static int cn_print_exe_file(struct core_name *cn)
 149 {
 150         struct file *exe_file;
 151         char *pathbuf, *path;
 152         int ret;
 153
 154         exe_file = get_mm_exe_file(current->mm);
 155         if (!exe_file)
 156                 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
 157
 158         pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
 159         if (!pathbuf) {
 160                 ret = -ENOMEM;
 161                 goto put_exe_file;
 162         }
 163
 164         path = file_path(exe_file, pathbuf, PATH_MAX);
 165         if (IS_ERR(path)) {
 166                 ret = PTR_ERR(path);
 167                 goto free_buf;
 168         }
 169
 170         ret = cn_esc_printf(cn, "%s", path);
 171
 172 free_buf:
 173         kfree(pathbuf);
 174 put_exe_file:
 175         fput(exe_file);
 176         return ret;
 177 }
 178
 179 /* format_corename will inspect the pattern parameter, and output a
 180  * name into corename, which must have space for at least
 181  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
 182  */
 183 static int format_corename(struct core_name *cn, struct coredump_params *cprm)
 184 {
 185         const struct cred *cred = current_cred();
 186         const char *pat_ptr = core_pattern;
 187         int ispipe = (*pat_ptr == '|');
 188         int pid_in_pattern = 0;
 189         int err = 0;
 190
 191         cn->used = 0;
 192         cn->corename = NULL;
 193         if (expand_corename(cn, core_name_size))
 194                 return -ENOMEM;
 195         cn->corename[0] = '\0';
 196
 197         if (ispipe)
 198                 ++pat_ptr;
 199
 200         /* Repeat as long as we have more pattern to process and more output
 201            space */
 202         while (*pat_ptr) {
 203                 if (*pat_ptr != '%') {
 204                         err = cn_printf(cn, "%c", *pat_ptr++);
 205                 } else {
 206                         switch (*++pat_ptr) {
 207                         /* single % at the end, drop that */
 208                         case 0:
 209                                 goto out;
 210                         /* Double percent, output one percent */
 211                         case '%':
 212                                 err = cn_printf(cn, "%c", '%');
 213                                 break;
 214                         /* pid */
 215                         case 'p':
 216                                 pid_in_pattern = 1;
 217                                 err = cn_printf(cn, "%d",
 218                                               task_tgid_vnr(current));
 219                                 break;
 220                         /* global pid */
 221                         case 'P':
 222                                 err = cn_printf(cn, "%d",
 223                                               task_tgid_nr(current));
 224                                 break;
 225                         case 'i':
 226                                 err = cn_printf(cn, "%d",
 227                                               task_pid_vnr(current));
 228                                 break;
 229                         case 'I':
 230                                 err = cn_printf(cn, "%d",
 231                                               task_pid_nr(current));
 232                                 break;
 233                         /* uid */
 234                         case 'u':
 235                                 err = cn_printf(cn, "%u",
 236                                                 from_kuid(&init_user_ns,
 237                                                           cred->uid));
 238                                 break;
 239                         /* gid */
 240                         case 'g':
 241                                 err = cn_printf(cn, "%u",
 242                                                 from_kgid(&init_user_ns,
 243                                                           cred->gid));
 244                                 break;
 245                         case 'd':
 246                                 err = cn_printf(cn, "%d",
 247                                         __get_dumpable(cprm->mm_flags));
 248                                 break;
 249                         /* signal that caused the coredump */
 250                         case 's':
 251                                 err = cn_printf(cn, "%d",
 252                                                 cprm->siginfo->si_signo);
 253                                 break;
 254                         /* UNIX time of coredump */
 255                         case 't': {
 256                                 time64_t time;
 257
 258                                 time = ktime_get_real_seconds();
 259                                 err = cn_printf(cn, "%lld", time);
 260                                 break;
 261                         }
 262                         /* hostname */
 263                         case 'h':
 264                                 down_read(&uts_sem);
 265                                 err = cn_esc_printf(cn, "%s",
 266                                               utsname()->nodename);
 267                                 up_read(&uts_sem);
 268                                 break;
 269                         /* executable */
 270                         case 'e':
 271                                 err = cn_esc_printf(cn, "%s", current->comm);
 272                                 break;
 273                         case 'E':
 274                                 err = cn_print_exe_file(cn);
 275                                 break;
 276                         /* core limit size */
 277                         case 'c':
 278                                 err = cn_printf(cn, "%lu",
 279                                               rlimit(RLIMIT_CORE));
 280                                 break;
 281                         default:
 282                                 break;
 283                         }
 284                         ++pat_ptr;
 285                 }
 286
 287                 if (err)
 288                         return err;
 289         }
 290
 291 out:
 292         /* Backward compatibility with core_uses_pid:
 293          *
 294          * If core_pattern does not include a %p (as is the default)
 295          * and core_uses_pid is set, then .%pid will be appended to
 296          * the filename. Do not do this for piped commands. */
 297         if (!ispipe && !pid_in_pattern && core_uses_pid) {
 298                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
 299                 if (err)
 300                         return err;
 301         }
 302         return ispipe;
 303 }
 304
 305 static int zap_process(struct task_struct *start, int exit_code, int flags)
 306 {
 307         struct task_struct *t;
 308         int nr = 0;
 309
 310         /* ignore all signals except SIGKILL, see prepare_signal() */
 311         start->signal->flags = SIGNAL_GROUP_COREDUMP | flags;
 312         start->signal->group_exit_code = exit_code;
 313         start->signal->group_stop_count = 0;
 314
 315         for_each_thread(start, t) {
 316                 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
 317                 if (t != current && t->mm) {
 318                         sigaddset(&t->pending.signal, SIGKILL);
 319                         signal_wake_up(t, 1);
 320                         nr++;
 321                 }
 322         }
 323
 324         return nr;
 325 }
 326
 327 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
 328                         struct core_state *core_state, int exit_code)
 329 {
 330         struct task_struct *g, *p;
 331         unsigned long flags;
 332         int nr = -EAGAIN;
 333
 334         spin_lock_irq(&tsk->sighand->siglock);
 335         if (!signal_group_exit(tsk->signal)) {
 336                 mm->core_state = core_state;
 337                 tsk->signal->group_exit_task = tsk;
 338                 nr = zap_process(tsk, exit_code, 0);
 339                 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
 340         }
 341         spin_unlock_irq(&tsk->sighand->siglock);
 342         if (unlikely(nr < 0))
 343                 return nr;
 344
 345         tsk->flags |= PF_DUMPCORE;
 346         if (atomic_read(&mm->mm_users) == nr + 1)
 347                 goto done;
 348         /*
 349          * We should find and kill all tasks which use this mm, and we should
 350          * count them correctly into ->nr_threads. We don't take tasklist
 351          * lock, but this is safe wrt:
 352          *
 353          * fork:
 354          *      None of sub-threads can fork after zap_process(leader). All
 355          *      processes which were created before this point should be
 356          *      visible to zap_threads() because copy_process() adds the new
 357          *      process to the tail of init_task.tasks list, and lock/unlock
 358          *      of ->siglock provides a memory barrier.
 359          *
 360          * do_exit:
 361          *      The caller holds mm->mmap_sem. This means that the task which
 362          *      uses this mm can't pass exit_mm(), so it can't exit or clear
 363          *      its ->mm.
 364          *
 365          * de_thread:
 366          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
 367          *      we must see either old or new leader, this does not matter.
 368          *      However, it can change p->sighand, so lock_task_sighand(p)
 369          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
 370          *      it can't fail.
 371          *
 372          *      Note also that "g" can be the old leader with ->mm == NULL
 373          *      and already unhashed and thus removed from ->thread_group.
 374          *      This is OK, __unhash_process()->list_del_rcu() does not
 375          *      clear the ->next pointer, we will find the new leader via
 376          *      next_thread().
 377          */
 378         rcu_read_lock();
 379         for_each_process(g) {
 380                 if (g == tsk->group_leader)
 381                         continue;
 382                 if (g->flags & PF_KTHREAD)
 383                         continue;
 384
 385                 for_each_thread(g, p) {
 386                         if (unlikely(!p->mm))
 387                                 continue;
 388                         if (unlikely(p->mm == mm)) {
 389                                 lock_task_sighand(p, &flags);
 390                                 nr += zap_process(p, exit_code,
 391                                                         SIGNAL_GROUP_EXIT);
 392                                 unlock_task_sighand(p, &flags);
 393                         }
 394                         break;
 395                 }
 396         }
 397         rcu_read_unlock();
 398 done:
 399         atomic_set(&core_state->nr_threads, nr);
 400         return nr;
 401 }
 402
 403 static int coredump_wait(int exit_code, struct core_state *core_state)
 404 {
 405         struct task_struct *tsk = current;
 406         struct mm_struct *mm = tsk->mm;
 407         int core_waiters = -EBUSY;
 408
 409         init_completion(&core_state->startup);
 410         core_state->dumper.task = tsk;
 411         core_state->dumper.next = NULL;
 412
 413         down_write(&mm->mmap_sem);
 414         if (!mm->core_state)
 415                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
 416         up_write(&mm->mmap_sem);
 417
 418         if (core_waiters > 0) {
 419                 struct core_thread *ptr;
 420
 421                 wait_for_completion(&core_state->startup);
 422                 /*
 423                  * Wait for all the threads to become inactive, so that
 424                  * all the thread context (extended register state, like
 425                  * fpu etc) gets copied to the memory.
 426                  */
 427                 ptr = core_state->dumper.next;
 428                 while (ptr != NULL) {
 429                         wait_task_inactive(ptr->task, 0);
 430                         ptr = ptr->next;
 431                 }
 432         }
 433
 434         return core_waiters;
 435 }
 436
 437 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
 438 {
 439         struct core_thread *curr, *next;
 440         struct task_struct *task;
 441
 442         spin_lock_irq(&current->sighand->siglock);
 443         if (core_dumped && !__fatal_signal_pending(current))
 444                 current->signal->group_exit_code |= 0x80;
 445         current->signal->group_exit_task = NULL;
 446         current->signal->flags = SIGNAL_GROUP_EXIT;
 447         spin_unlock_irq(&current->sighand->siglock);
 448
 449         next = mm->core_state->dumper.next;
 450         while ((curr = next) != NULL) {
 451                 next = curr->next;
 452                 task = curr->task;
 453                 /*
 454                  * see exit_mm(), curr->task must not see
 455                  * ->task == NULL before we read ->next.
 456                  */
 457                 smp_mb();
 458                 curr->task = NULL;
 459                 wake_up_process(task);
 460         }
 461
 462         mm->core_state = NULL;
 463 }
 464
 465 static bool dump_interrupted(void)
 466 {
 467         /*
 468          * SIGKILL or freezing() interrupt the coredumping. Perhaps we
 469          * can do try_to_freeze() and check __fatal_signal_pending(),
 470          * but then we need to teach dump_write() to restart and clear
 471          * TIF_SIGPENDING.
 472          */
 473         return signal_pending(current);
 474 }
 475
 476 static void wait_for_dump_helpers(struct file *file)
 477 {
 478         struct pipe_inode_info *pipe = file->private_data;
 479
 480         pipe_lock(pipe);
 481         pipe->readers++;
 482         pipe->writers--;
 483         wake_up_interruptible_sync(&pipe->wait);
 484         kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 485         pipe_unlock(pipe);
 486
 487         /*
 488          * We actually want wait_event_freezable() but then we need
 489          * to clear TIF_SIGPENDING and improve dump_interrupted().
 490          */
 491         wait_event_interruptible(pipe->wait, pipe->readers == 1);
 492
 493         pipe_lock(pipe);
 494         pipe->readers--;
 495         pipe->writers++;
 496         pipe_unlock(pipe);
 497 }
 498
 499 /*
 500  * umh_pipe_setup
 501  * helper function to customize the process used
 502  * to collect the core in userspace.  Specifically
 503  * it sets up a pipe and installs it as fd 0 (stdin)
 504  * for the process.  Returns 0 on success, or
 505  * PTR_ERR on failure.
 506  * Note that it also sets the core limit to 1.  This
 507  * is a special value that we use to trap recursive
 508  * core dumps
 509  */
 510 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
 511 {
 512         struct file *files[2];
 513         struct coredump_params *cp = (struct coredump_params *)info->data;
 514         int err = create_pipe_files(files, 0);
 515         if (err)
 516                 return err;
 517
 518         cp->file = files[1];
 519
 520         err = replace_fd(0, files[0], 0);
 521         fput(files[0]);
 522         /* and disallow core files too */
 523         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
 524
 525         return err;
 526 }
 527
 528 void do_coredump(const siginfo_t *siginfo)
 529 {
 530         struct core_state core_state;
 531         struct core_name cn;
 532         struct mm_struct *mm = current->mm;
 533         struct linux_binfmt * binfmt;
 534         const struct cred *old_cred;
 535         struct cred *cred;
 536         int retval = 0;
 537         int ispipe;
 538         struct files_struct *displaced;
 539         /* require nonrelative corefile path and be extra careful */
 540         bool need_suid_safe = false;
 541         bool core_dumped = false;
 542         static atomic_t core_dump_count = ATOMIC_INIT(0);
 543         struct coredump_params cprm = {
 544                 .siginfo = siginfo,
 545                 .regs = signal_pt_regs(),
 546                 .limit = rlimit(RLIMIT_CORE),
 547                 /*
 548                  * We must use the same mm->flags while dumping core to avoid
 549                  * inconsistency of bit flags, since this flag is not protected
 550                  * by any locks.
 551                  */
 552                 .mm_flags = mm->flags,
 553         };
 554
 555         audit_core_dumps(siginfo->si_signo);
 556
 557         binfmt = mm->binfmt;
 558         if (!binfmt || !binfmt->core_dump)
 559                 goto fail;
 560         if (!__get_dumpable(cprm.mm_flags))
 561                 goto fail;
 562
 563         cred = prepare_creds();
 564         if (!cred)
 565                 goto fail;
 566         /*
 567          * We cannot trust fsuid as being the "true" uid of the process
 568          * nor do we know its entire history. We only know it was tainted
 569          * so we dump it as root in mode 2, and only into a controlled
 570          * environment (pipe handler or fully qualified path).
 571          */
 572         if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
 573                 /* Setuid core dump mode */
 574                 cred->fsuid = GLOBAL_ROOT_UID;  /* Dump root private */
 575                 need_suid_safe = true;
 576         }
 577
 578         retval = coredump_wait(siginfo->si_signo, &core_state);
 579         if (retval < 0)
 580                 goto fail_creds;
 581
 582         old_cred = override_creds(cred);
 583
 584         ispipe = format_corename(&cn, &cprm);
 585
 586         if (ispipe) {
 587                 int dump_count;
 588                 char **helper_argv;
 589                 struct subprocess_info *sub_info;
 590
 591                 if (ispipe < 0) {
 592                         printk(KERN_WARNING "format_corename failed\n");
 593                         printk(KERN_WARNING "Aborting core\n");
 594                         goto fail_unlock;
 595                 }
 596
 597                 if (cprm.limit == 1) {
 598                         /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
 599                          *
 600                          * Normally core limits are irrelevant to pipes, since
 601                          * we're not writing to the file system, but we use
 602                          * cprm.limit of 1 here as a special value, this is a
 603                          * consistent way to catch recursive crashes.
 604                          * We can still crash if the core_pattern binary sets
 605                          * RLIM_CORE = !1, but it runs as root, and can do
 606                          * lots of stupid things.
 607                          *
 608                          * Note that we use task_tgid_vnr here to grab the pid
 609                          * of the process group leader.  That way we get the
 610                          * right pid if a thread in a multi-threaded
 611                          * core_pattern process dies.
 612                          */
 613                         printk(KERN_WARNING
 614                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
 615                                 task_tgid_vnr(current), current->comm);
 616                         printk(KERN_WARNING "Aborting core\n");
 617                         goto fail_unlock;
 618                 }
 619                 cprm.limit = RLIM_INFINITY;
 620
 621                 dump_count = atomic_inc_return(&core_dump_count);
 622                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
 623                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
 624                                task_tgid_vnr(current), current->comm);
 625                         printk(KERN_WARNING "Skipping core dump\n");
 626                         goto fail_dropcount;
 627                 }
 628
 629                 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
 630                 if (!helper_argv) {
 631                         printk(KERN_WARNING "%s failed to allocate memory\n",
 632                                __func__);
 633                         goto fail_dropcount;
 634                 }
 635
 636                 retval = -ENOMEM;
 637                 sub_info = call_usermodehelper_setup(helper_argv[0],
 638                                                 helper_argv, NULL, GFP_KERNEL,
 639                                                 umh_pipe_setup, NULL, &cprm);
 640                 if (sub_info)
 641                         retval = call_usermodehelper_exec(sub_info,
 642                                                           UMH_WAIT_EXEC);
 643
 644                 argv_free(helper_argv);
 645                 if (retval) {
 646                         printk(KERN_INFO "Core dump to |%s pipe failed\n",
 647                                cn.corename);
 648                         goto close_fail;
 649                 }
 650         } else {
 651                 struct inode *inode;
 652
 653                 if (cprm.limit < binfmt->min_coredump)
 654                         goto fail_unlock;
 655
 656                 if (need_suid_safe && cn.corename[0] != '/') {
 657                         printk(KERN_WARNING "Pid %d(%s) can only dump core "\
 658                                 "to fully qualified path!\n",
 659                                 task_tgid_vnr(current), current->comm);
 660                         printk(KERN_WARNING "Skipping core dump\n");
 661                         goto fail_unlock;
 662                 }
 663
 664                 /*
 665                  * Unlink the file if it exists unless this is a SUID
 666                  * binary - in that case, we're running around with root
 667                  * privs and don't want to unlink another user's coredump.
 668                  */
 669                 if (!need_suid_safe) {
 670                         mm_segment_t old_fs;
 671
 672                         old_fs = get_fs();
 673                         set_fs(KERNEL_DS);
 674                         /*
 675                          * If it doesn't exist, that's fine. If there's some
 676                          * other problem, we'll catch it at the filp_open().
 677                          */
 678                         (void) sys_unlink((const char __user *)cn.corename);
 679                         set_fs(old_fs);
 680                 }
 681
 682                 /*
 683                  * There is a race between unlinking and creating the
 684                  * file, but if that causes an EEXIST here, that's
 685                  * fine - another process raced with us while creating
 686                  * the corefile, and the other process won. To userspace,
 687                  * what matters is that at least one of the two processes
 688                  * writes its coredump successfully, not which one.
 689                  */
 690                 cprm.file = filp_open(cn.corename,
 691                                  O_CREAT | 2 | O_NOFOLLOW |
 692                                  O_LARGEFILE | O_EXCL,
 693                                  0600);
 694                 if (IS_ERR(cprm.file))
 695                         goto fail_unlock;
 696
 697                 inode = file_inode(cprm.file);
 698                 if (inode->i_nlink > 1)
 699                         goto close_fail;
 700                 if (d_unhashed(cprm.file->f_path.dentry))
 701                         goto close_fail;
 702                 /*
 703                  * AK: actually i see no reason to not allow this for named
 704                  * pipes etc, but keep the previous behaviour for now.
 705                  */
 706                 if (!S_ISREG(inode->i_mode))
 707                         goto close_fail;
 708                 /*
 709                  * Don't dump core if the filesystem changed owner or mode
 710                  * of the file during file creation. This is an issue when
 711                  * a process dumps core while its cwd is e.g. on a vfat
 712                  * filesystem.
 713                  */
 714                 if (!uid_eq(inode->i_uid, current_fsuid()))
 715                         goto close_fail;
 716                 if ((inode->i_mode & 0677) != 0600)
 717                         goto close_fail;
 718                 if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
 719                         goto close_fail;
 720                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
 721                         goto close_fail;
 722         }
 723
 724         /* get us an unshared descriptor table; almost always a no-op */
 725         retval = unshare_files(&displaced);
 726         if (retval)
 727                 goto close_fail;
 728         if (displaced)
 729                 put_files_struct(displaced);
 730         if (!dump_interrupted()) {
 731                 file_start_write(cprm.file);
 732                 core_dumped = binfmt->core_dump(&cprm);
 733                 file_end_write(cprm.file);
 734         }
 735         if (ispipe && core_pipe_limit)
 736                 wait_for_dump_helpers(cprm.file);
 737 close_fail:
 738         if (cprm.file)
 739                 filp_close(cprm.file, NULL);
 740 fail_dropcount:
 741         if (ispipe)
 742                 atomic_dec(&core_dump_count);
 743 fail_unlock:
 744         kfree(cn.corename);
 745         coredump_finish(mm, core_dumped);
 746         revert_creds(old_cred);
 747 fail_creds:
 748         put_cred(cred);
 749 fail:
 750         return;
 751 }
 752
 753 /*
 754  * Core dumping helper functions.  These are the only things you should
 755  * do on a core-file: use only these functions to write out all the
 756  * necessary info.
 757  */
 758 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
 759 {
 760         struct file *file = cprm->file;
 761         loff_t pos = file->f_pos;
 762         ssize_t n;
 763         if (cprm->written + nr > cprm->limit)
 764                 return 0;
 765         while (nr) {
 766                 if (dump_interrupted())
 767                         return 0;
 768                 n = __kernel_write(file, addr, nr, &pos);
 769                 if (n <= 0)
 770                         return 0;
 771                 file->f_pos = pos;
 772                 cprm->written += n;
 773                 nr -= n;
 774         }
 775         return 1;
 776 }
 777 EXPORT_SYMBOL(dump_emit);
 778
 779 int dump_skip(struct coredump_params *cprm, size_t nr)
 780 {
 781         static char zeroes[PAGE_SIZE];
 782         struct file *file = cprm->file;
 783         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
 784                 if (cprm->written + nr > cprm->limit)
 785                         return 0;
 786                 if (dump_interrupted() ||
 787                     file->f_op->llseek(file, nr, SEEK_CUR) < 0)
 788                         return 0;
 789                 cprm->written += nr;
 790                 return 1;
 791         } else {
 792                 while (nr > PAGE_SIZE) {
 793                         if (!dump_emit(cprm, zeroes, PAGE_SIZE))
 794                                 return 0;
 795                         nr -= PAGE_SIZE;
 796                 }
 797                 return dump_emit(cprm, zeroes, nr);
 798         }
 799 }
 800 EXPORT_SYMBOL(dump_skip);
 801
 802 int dump_align(struct coredump_params *cprm, int align)
 803 {
 804         unsigned mod = cprm->written & (align - 1);
 805         if (align & (align - 1))
 806                 return 0;
 807         return mod ? dump_skip(cprm, align - mod) : 1;
 808 }
 809 EXPORT_SYMBOL(dump_align);