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