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