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