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 __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
  78                                      va_list arg)
  79 {
  80         int free, need;
  81         va_list arg_copy;
  82
  83 again:
  84         free = cn->size - cn->used;
  85
  86         va_copy(arg_copy, arg);
  87         need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
  88         va_end(arg_copy);
  89
  90         if (need < free) {
  91                 cn->used += need;
  92                 return 0;
  93         }
  94
  95         if (!expand_corename(cn, cn->size + need - free + 1))
  96                 goto again;
  97
  98         return -ENOMEM;
  99 }
 100
 101 static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
 102 {
 103         va_list arg;
 104         int ret;
 105
 106         va_start(arg, fmt);
 107         ret = cn_vprintf(cn, fmt, arg);
 108         va_end(arg);
 109
 110         return ret;
 111 }
 112
 113 static __printf(2, 3)
 114 int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
 115 {
 116         int cur = cn->used;
 117         va_list arg;
 118         int ret;
 119
 120         va_start(arg, fmt);
 121         ret = cn_vprintf(cn, fmt, arg);
 122         va_end(arg);
 123
 124         if (ret == 0) {
 125                 /*
 126                  * Ensure that this coredump name component can't cause the
 127                  * resulting corefile path to consist of a ".." or ".".
 128                  */
 129                 if ((cn->used - cur == 1 && cn->corename[cur] == '.') ||
 130                                 (cn->used - cur == 2 && cn->corename[cur] == '.'
 131                                 && cn->corename[cur+1] == '.'))
 132                         cn->corename[cur] = '!';
 133
 134                 /*
 135                  * Empty names are fishy and could be used to create a "//" in a
 136                  * corefile name, causing the coredump to happen one directory
 137                  * level too high. Enforce that all components of the core
 138                  * pattern are at least one character long.
 139                  */
 140                 if (cn->used == cur)
 141                         ret = cn_printf(cn, "!");
 142         }
 143
 144         for (; cur < cn->used; ++cur) {
 145                 if (cn->corename[cur] == '/')
 146                         cn->corename[cur] = '!';
 147         }
 148         return ret;
 149 }
 150
 151 static int cn_print_exe_file(struct core_name *cn)
 152 {
 153         struct file *exe_file;
 154         char *pathbuf, *path;
 155         int ret;
 156
 157         exe_file = get_mm_exe_file(current->mm);
 158         if (!exe_file)
 159                 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
 160
 161         pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
 162         if (!pathbuf) {
 163                 ret = -ENOMEM;
 164                 goto put_exe_file;
 165         }
 166
 167         path = file_path(exe_file, pathbuf, PATH_MAX);
 168         if (IS_ERR(path)) {
 169                 ret = PTR_ERR(path);
 170                 goto free_buf;
 171         }
 172
 173         ret = cn_esc_printf(cn, "%s", path);
 174
 175 free_buf:
 176         kfree(pathbuf);
 177 put_exe_file:
 178         fput(exe_file);
 179         return ret;
 180 }
 181
 182 /* format_corename will inspect the pattern parameter, and output a
 183  * name into corename, which must have space for at least
 184  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
 185  */
 186 static int format_corename(struct core_name *cn, struct coredump_params *cprm)
 187 {
 188         const struct cred *cred = current_cred();
 189         const char *pat_ptr = core_pattern;
 190         int ispipe = (*pat_ptr == '|');
 191         int pid_in_pattern = 0;
 192         int err = 0;
 193
 194         cn->used = 0;
 195         cn->corename = NULL;
 196         if (expand_corename(cn, core_name_size))
 197                 return -ENOMEM;
 198         cn->corename[0] = '\0';
 199
 200         if (ispipe)
 201                 ++pat_ptr;
 202
 203         /* Repeat as long as we have more pattern to process and more output
 204            space */
 205         while (*pat_ptr) {
 206                 if (*pat_ptr != '%') {
 207                         err = cn_printf(cn, "%c", *pat_ptr++);
 208                 } else {
 209                         switch (*++pat_ptr) {
 210                         /* single % at the end, drop that */
 211                         case 0:
 212                                 goto out;
 213                         /* Double percent, output one percent */
 214                         case '%':
 215                                 err = cn_printf(cn, "%c", '%');
 216                                 break;
 217                         /* pid */
 218                         case 'p':
 219                                 pid_in_pattern = 1;
 220                                 err = cn_printf(cn, "%d",
 221                                               task_tgid_vnr(current));
 222                                 break;
 223                         /* global pid */
 224                         case 'P':
 225                                 err = cn_printf(cn, "%d",
 226                                               task_tgid_nr(current));
 227                                 break;
 228                         case 'i':
 229                                 err = cn_printf(cn, "%d",
 230                                               task_pid_vnr(current));
 231                                 break;
 232                         case 'I':
 233                                 err = cn_printf(cn, "%d",
 234                                               task_pid_nr(current));
 235                                 break;
 236                         /* uid */
 237                         case 'u':
 238                                 err = cn_printf(cn, "%u",
 239                                                 from_kuid(&init_user_ns,
 240                                                           cred->uid));
 241                                 break;
 242                         /* gid */
 243                         case 'g':
 244                                 err = cn_printf(cn, "%u",
 245                                                 from_kgid(&init_user_ns,
 246                                                           cred->gid));
 247                                 break;
 248                         case 'd':
 249                                 err = cn_printf(cn, "%d",
 250                                         __get_dumpable(cprm->mm_flags));
 251                                 break;
 252                         /* signal that caused the coredump */
 253                         case 's':
 254                                 err = cn_printf(cn, "%d",
 255                                                 cprm->siginfo->si_signo);
 256                                 break;
 257                         /* UNIX time of coredump */
 258                         case 't': {
 259                                 time64_t time;
 260
 261                                 time = ktime_get_real_seconds();
 262                                 err = cn_printf(cn, "%lld", time);
 263                                 break;
 264                         }
 265                         /* hostname */
 266                         case 'h':
 267                                 down_read(&uts_sem);
 268                                 err = cn_esc_printf(cn, "%s",
 269                                               utsname()->nodename);
 270                                 up_read(&uts_sem);
 271                                 break;
 272                         /* executable */
 273                         case 'e':
 274                                 err = cn_esc_printf(cn, "%s", current->comm);
 275                                 break;
 276                         case 'E':
 277                                 err = cn_print_exe_file(cn);
 278                                 break;
 279                         /* core limit size */
 280                         case 'c':
 281                                 err = cn_printf(cn, "%lu",
 282                                               rlimit(RLIMIT_CORE));
 283                                 break;
 284                         default:
 285                                 break;
 286                         }
 287                         ++pat_ptr;
 288                 }
 289
 290                 if (err)
 291                         return err;
 292         }
 293
 294 out:
 295         /* Backward compatibility with core_uses_pid:
 296          *
 297          * If core_pattern does not include a %p (as is the default)
 298          * and core_uses_pid is set, then .%pid will be appended to
 299          * the filename. Do not do this for piped commands. */
 300         if (!ispipe && !pid_in_pattern && core_uses_pid) {
 301                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
 302                 if (err)
 303                         return err;
 304         }
 305         return ispipe;
 306 }
 307
 308 static int zap_process(struct task_struct *start, int exit_code, int flags)
 309 {
 310         struct task_struct *t;
 311         int nr = 0;
 312
 313         /* ignore all signals except SIGKILL, see prepare_signal() */
 314         start->signal->flags = SIGNAL_GROUP_COREDUMP | flags;
 315         start->signal->group_exit_code = exit_code;
 316         start->signal->group_stop_count = 0;
 317
 318         for_each_thread(start, t) {
 319                 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
 320                 if (t != current && t->mm) {
 321                         sigaddset(&t->pending.signal, SIGKILL);
 322                         signal_wake_up(t, 1);
 323                         nr++;
 324                 }
 325         }
 326
 327         return nr;
 328 }
 329
 330 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
 331                         struct core_state *core_state, int exit_code)
 332 {
 333         struct task_struct *g, *p;
 334         unsigned long flags;
 335         int nr = -EAGAIN;
 336
 337         spin_lock_irq(&tsk->sighand->siglock);
 338         if (!signal_group_exit(tsk->signal)) {
 339                 mm->core_state = core_state;
 340                 tsk->signal->group_exit_task = tsk;
 341                 nr = zap_process(tsk, exit_code, 0);
 342                 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
 343         }
 344         spin_unlock_irq(&tsk->sighand->siglock);
 345         if (unlikely(nr < 0))
 346                 return nr;
 347
 348         tsk->flags |= PF_DUMPCORE;
 349         if (atomic_read(&mm->mm_users) == nr + 1)
 350                 goto done;
 351         /*
 352          * We should find and kill all tasks which use this mm, and we should
 353          * count them correctly into ->nr_threads. We don't take tasklist
 354          * lock, but this is safe wrt:
 355          *
 356          * fork:
 357          *      None of sub-threads can fork after zap_process(leader). All
 358          *      processes which were created before this point should be
 359          *      visible to zap_threads() because copy_process() adds the new
 360          *      process to the tail of init_task.tasks list, and lock/unlock
 361          *      of ->siglock provides a memory barrier.
 362          *
 363          * do_exit:
 364          *      The caller holds mm->mmap_sem. This means that the task which
 365          *      uses this mm can't pass exit_mm(), so it can't exit or clear
 366          *      its ->mm.
 367          *
 368          * de_thread:
 369          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
 370          *      we must see either old or new leader, this does not matter.
 371          *      However, it can change p->sighand, so lock_task_sighand(p)
 372          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
 373          *      it can't fail.
 374          *
 375          *      Note also that "g" can be the old leader with ->mm == NULL
 376          *      and already unhashed and thus removed from ->thread_group.
 377          *      This is OK, __unhash_process()->list_del_rcu() does not
 378          *      clear the ->next pointer, we will find the new leader via
 379          *      next_thread().
 380          */
 381         rcu_read_lock();
 382         for_each_process(g) {
 383                 if (g == tsk->group_leader)
 384                         continue;
 385                 if (g->flags & PF_KTHREAD)
 386                         continue;
 387
 388                 for_each_thread(g, p) {
 389                         if (unlikely(!p->mm))
 390                                 continue;
 391                         if (unlikely(p->mm == mm)) {
 392                                 lock_task_sighand(p, &flags);
 393                                 nr += zap_process(p, exit_code,
 394                                                         SIGNAL_GROUP_EXIT);
 395                                 unlock_task_sighand(p, &flags);
 396                         }
 397                         break;
 398                 }
 399         }
 400         rcu_read_unlock();
 401 done:
 402         atomic_set(&core_state->nr_threads, nr);
 403         return nr;
 404 }
 405
 406 static int coredump_wait(int exit_code, struct core_state *core_state)
 407 {
 408         struct task_struct *tsk = current;
 409         struct mm_struct *mm = tsk->mm;
 410         int core_waiters = -EBUSY;
 411
 412         init_completion(&core_state->startup);
 413         core_state->dumper.task = tsk;
 414         core_state->dumper.next = NULL;
 415
 416         down_write(&mm->mmap_sem);
 417         if (!mm->core_state)
 418                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
 419         up_write(&mm->mmap_sem);
 420
 421         if (core_waiters > 0) {
 422                 struct core_thread *ptr;
 423
 424                 wait_for_completion(&core_state->startup);
 425                 /*
 426                  * Wait for all the threads to become inactive, so that
 427                  * all the thread context (extended register state, like
 428                  * fpu etc) gets copied to the memory.
 429                  */
 430                 ptr = core_state->dumper.next;
 431                 while (ptr != NULL) {
 432                         wait_task_inactive(ptr->task, 0);
 433                         ptr = ptr->next;
 434                 }
 435         }
 436
 437         return core_waiters;
 438 }
 439
 440 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
 441 {
 442         struct core_thread *curr, *next;
 443         struct task_struct *task;
 444
 445         spin_lock_irq(&current->sighand->siglock);
 446         if (core_dumped && !__fatal_signal_pending(current))
 447                 current->signal->group_exit_code |= 0x80;
 448         current->signal->group_exit_task = NULL;
 449         current->signal->flags = SIGNAL_GROUP_EXIT;
 450         spin_unlock_irq(&current->sighand->siglock);
 451
 452         next = mm->core_state->dumper.next;
 453         while ((curr = next) != NULL) {
 454                 next = curr->next;
 455                 task = curr->task;
 456                 /*
 457                  * see exit_mm(), curr->task must not see
 458                  * ->task == NULL before we read ->next.
 459                  */
 460                 smp_mb();
 461                 curr->task = NULL;
 462                 wake_up_process(task);
 463         }
 464
 465         mm->core_state = NULL;
 466 }
 467
 468 static bool dump_interrupted(void)
 469 {
 470         /*
 471          * SIGKILL or freezing() interrupt the coredumping. Perhaps we
 472          * can do try_to_freeze() and check __fatal_signal_pending(),
 473          * but then we need to teach dump_write() to restart and clear
 474          * TIF_SIGPENDING.
 475          */
 476         return signal_pending(current);
 477 }
 478
 479 static void wait_for_dump_helpers(struct file *file)
 480 {
 481         struct pipe_inode_info *pipe = file->private_data;
 482
 483         pipe_lock(pipe);
 484         pipe->readers++;
 485         pipe->writers--;
 486         wake_up_interruptible_sync(&pipe->wait);
 487         kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 488         pipe_unlock(pipe);
 489
 490         /*
 491          * We actually want wait_event_freezable() but then we need
 492          * to clear TIF_SIGPENDING and improve dump_interrupted().
 493          */
 494         wait_event_interruptible(pipe->wait, pipe->readers == 1);
 495
 496         pipe_lock(pipe);
 497         pipe->readers--;
 498         pipe->writers++;
 499         pipe_unlock(pipe);
 500 }
 501
 502 /*
 503  * umh_pipe_setup
 504  * helper function to customize the process used
 505  * to collect the core in userspace.  Specifically
 506  * it sets up a pipe and installs it as fd 0 (stdin)
 507  * for the process.  Returns 0 on success, or
 508  * PTR_ERR on failure.
 509  * Note that it also sets the core limit to 1.  This
 510  * is a special value that we use to trap recursive
 511  * core dumps
 512  */
 513 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
 514 {
 515         struct file *files[2];
 516         struct coredump_params *cp = (struct coredump_params *)info->data;
 517         int err = create_pipe_files(files, 0);
 518         if (err)
 519                 return err;
 520
 521         cp->file = files[1];
 522
 523         err = replace_fd(0, files[0], 0);
 524         fput(files[0]);
 525         /* and disallow core files too */
 526         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
 527
 528         return err;
 529 }
 530
 531 void do_coredump(const siginfo_t *siginfo)
 532 {
 533         struct core_state core_state;
 534         struct core_name cn;
 535         struct mm_struct *mm = current->mm;
 536         struct linux_binfmt * binfmt;
 537         const struct cred *old_cred;
 538         struct cred *cred;
 539         int retval = 0;
 540         int ispipe;
 541         struct files_struct *displaced;
 542         /* require nonrelative corefile path and be extra careful */
 543         bool need_suid_safe = false;
 544         bool core_dumped = false;
 545         static atomic_t core_dump_count = ATOMIC_INIT(0);
 546         struct coredump_params cprm = {
 547                 .siginfo = siginfo,
 548                 .regs = signal_pt_regs(),
 549                 .limit = rlimit(RLIMIT_CORE),
 550                 /*
 551                  * We must use the same mm->flags while dumping core to avoid
 552                  * inconsistency of bit flags, since this flag is not protected
 553                  * by any locks.
 554                  */
 555                 .mm_flags = mm->flags,
 556         };
 557
 558         audit_core_dumps(siginfo->si_signo);
 559
 560         binfmt = mm->binfmt;
 561         if (!binfmt || !binfmt->core_dump)
 562                 goto fail;
 563         if (!__get_dumpable(cprm.mm_flags))
 564                 goto fail;
 565
 566         cred = prepare_creds();
 567         if (!cred)
 568                 goto fail;
 569         /*
 570          * We cannot trust fsuid as being the "true" uid of the process
 571          * nor do we know its entire history. We only know it was tainted
 572          * so we dump it as root in mode 2, and only into a controlled
 573          * environment (pipe handler or fully qualified path).
 574          */
 575         if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
 576                 /* Setuid core dump mode */
 577                 cred->fsuid = GLOBAL_ROOT_UID;  /* Dump root private */
 578                 need_suid_safe = true;
 579         }
 580
 581         retval = coredump_wait(siginfo->si_signo, &core_state);
 582         if (retval < 0)
 583                 goto fail_creds;
 584
 585         old_cred = override_creds(cred);
 586
 587         ispipe = format_corename(&cn, &cprm);
 588
 589         if (ispipe) {
 590                 int dump_count;
 591                 char **helper_argv;
 592                 struct subprocess_info *sub_info;
 593
 594                 if (ispipe < 0) {
 595                         printk(KERN_WARNING "format_corename failed\n");
 596                         printk(KERN_WARNING "Aborting core\n");
 597                         goto fail_unlock;
 598                 }
 599
 600                 if (cprm.limit == 1) {
 601                         /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
 602                          *
 603                          * Normally core limits are irrelevant to pipes, since
 604                          * we're not writing to the file system, but we use
 605                          * cprm.limit of 1 here as a special value, this is a
 606                          * consistent way to catch recursive crashes.
 607                          * We can still crash if the core_pattern binary sets
 608                          * RLIM_CORE = !1, but it runs as root, and can do
 609                          * lots of stupid things.
 610                          *
 611                          * Note that we use task_tgid_vnr here to grab the pid
 612                          * of the process group leader.  That way we get the
 613                          * right pid if a thread in a multi-threaded
 614                          * core_pattern process dies.
 615                          */
 616                         printk(KERN_WARNING
 617                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
 618                                 task_tgid_vnr(current), current->comm);
 619                         printk(KERN_WARNING "Aborting core\n");
 620                         goto fail_unlock;
 621                 }
 622                 cprm.limit = RLIM_INFINITY;
 623
 624                 dump_count = atomic_inc_return(&core_dump_count);
 625                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
 626                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
 627                                task_tgid_vnr(current), current->comm);
 628                         printk(KERN_WARNING "Skipping core dump\n");
 629                         goto fail_dropcount;
 630                 }
 631
 632                 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
 633                 if (!helper_argv) {
 634                         printk(KERN_WARNING "%s failed to allocate memory\n",
 635                                __func__);
 636                         goto fail_dropcount;
 637                 }
 638
 639                 retval = -ENOMEM;
 640                 sub_info = call_usermodehelper_setup(helper_argv[0],
 641                                                 helper_argv, NULL, GFP_KERNEL,
 642                                                 umh_pipe_setup, NULL, &cprm);
 643                 if (sub_info)
 644                         retval = call_usermodehelper_exec(sub_info,
 645                                                           UMH_WAIT_EXEC);
 646
 647                 argv_free(helper_argv);
 648                 if (retval) {
 649                         printk(KERN_INFO "Core dump to |%s pipe failed\n",
 650                                cn.corename);
 651                         goto close_fail;
 652                 }
 653         } else {
 654                 struct inode *inode;
 655                 int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
 656                                  O_LARGEFILE | O_EXCL;
 657
 658                 if (cprm.limit < binfmt->min_coredump)
 659                         goto fail_unlock;
 660
 661                 if (need_suid_safe && cn.corename[0] != '/') {
 662                         printk(KERN_WARNING "Pid %d(%s) can only dump core "\
 663                                 "to fully qualified path!\n",
 664                                 task_tgid_vnr(current), current->comm);
 665                         printk(KERN_WARNING "Skipping core dump\n");
 666                         goto fail_unlock;
 667                 }
 668
 669                 /*
 670                  * Unlink the file if it exists unless this is a SUID
 671                  * binary - in that case, we're running around with root
 672                  * privs and don't want to unlink another user's coredump.
 673                  */
 674                 if (!need_suid_safe) {
 675                         mm_segment_t old_fs;
 676
 677                         old_fs = get_fs();
 678                         set_fs(KERNEL_DS);
 679                         /*
 680                          * If it doesn't exist, that's fine. If there's some
 681                          * other problem, we'll catch it at the filp_open().
 682                          */
 683                         (void) sys_unlink((const char __user *)cn.corename);
 684                         set_fs(old_fs);
 685                 }
 686
 687                 /*
 688                  * There is a race between unlinking and creating the
 689                  * file, but if that causes an EEXIST here, that's
 690                  * fine - another process raced with us while creating
 691                  * the corefile, and the other process won. To userspace,
 692                  * what matters is that at least one of the two processes
 693                  * writes its coredump successfully, not which one.
 694                  */
 695                 if (need_suid_safe) {
 696                         /*
 697                          * Using user namespaces, normal user tasks can change
 698                          * their current->fs->root to point to arbitrary
 699                          * directories. Since the intention of the "only dump
 700                          * with a fully qualified path" rule is to control where
 701                          * coredumps may be placed using root privileges,
 702                          * current->fs->root must not be used. Instead, use the
 703                          * root directory of init_task.
 704                          */
 705                         struct path root;
 706
 707                         task_lock(&init_task);
 708                         get_fs_root(init_task.fs, &root);
 709                         task_unlock(&init_task);
 710                         cprm.file = file_open_root(root.dentry, root.mnt,
 711                                 cn.corename, open_flags, 0600);
 712                         path_put(&root);
 713                 } else {
 714                         cprm.file = filp_open(cn.corename, open_flags, 0600);
 715                 }
 716                 if (IS_ERR(cprm.file))
 717                         goto fail_unlock;
 718
 719                 inode = file_inode(cprm.file);
 720                 if (inode->i_nlink > 1)
 721                         goto close_fail;
 722                 if (d_unhashed(cprm.file->f_path.dentry))
 723                         goto close_fail;
 724                 /*
 725                  * AK: actually i see no reason to not allow this for named
 726                  * pipes etc, but keep the previous behaviour for now.
 727                  */
 728                 if (!S_ISREG(inode->i_mode))
 729                         goto close_fail;
 730                 /*
 731                  * Don't dump core if the filesystem changed owner or mode
 732                  * of the file during file creation. This is an issue when
 733                  * a process dumps core while its cwd is e.g. on a vfat
 734                  * filesystem.
 735                  */
 736                 if (!uid_eq(inode->i_uid, current_fsuid()))
 737                         goto close_fail;
 738                 if ((inode->i_mode & 0677) != 0600)
 739                         goto close_fail;
 740                 if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
 741                         goto close_fail;
 742                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
 743                         goto close_fail;
 744         }
 745
 746         /* get us an unshared descriptor table; almost always a no-op */
 747         retval = unshare_files(&displaced);
 748         if (retval)
 749                 goto close_fail;
 750         if (displaced)
 751                 put_files_struct(displaced);
 752         if (!dump_interrupted()) {
 753                 file_start_write(cprm.file);
 754                 core_dumped = binfmt->core_dump(&cprm);
 755                 file_end_write(cprm.file);
 756         }
 757         if (ispipe && core_pipe_limit)
 758                 wait_for_dump_helpers(cprm.file);
 759 close_fail:
 760         if (cprm.file)
 761                 filp_close(cprm.file, NULL);
 762 fail_dropcount:
 763         if (ispipe)
 764                 atomic_dec(&core_dump_count);
 765 fail_unlock:
 766         kfree(cn.corename);
 767         coredump_finish(mm, core_dumped);
 768         revert_creds(old_cred);
 769 fail_creds:
 770         put_cred(cred);
 771 fail:
 772         return;
 773 }
 774
 775 /*
 776  * Core dumping helper functions.  These are the only things you should
 777  * do on a core-file: use only these functions to write out all the
 778  * necessary info.
 779  */
 780 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
 781 {
 782         struct file *file = cprm->file;
 783         loff_t pos = file->f_pos;
 784         ssize_t n;
 785         if (cprm->written + nr > cprm->limit)
 786                 return 0;
 787         while (nr) {
 788                 if (dump_interrupted())
 789                         return 0;
 790                 n = __kernel_write(file, addr, nr, &pos);
 791                 if (n <= 0)
 792                         return 0;
 793                 file->f_pos = pos;
 794                 cprm->written += n;
 795                 nr -= n;
 796         }
 797         return 1;
 798 }
 799 EXPORT_SYMBOL(dump_emit);
 800
 801 int dump_skip(struct coredump_params *cprm, size_t nr)
 802 {
 803         static char zeroes[PAGE_SIZE];
 804         struct file *file = cprm->file;
 805         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
 806                 if (cprm->written + nr > cprm->limit)
 807                         return 0;
 808                 if (dump_interrupted() ||
 809                     file->f_op->llseek(file, nr, SEEK_CUR) < 0)
 810                         return 0;
 811                 cprm->written += nr;
 812                 return 1;
 813         } else {
 814                 while (nr > PAGE_SIZE) {
 815                         if (!dump_emit(cprm, zeroes, PAGE_SIZE))
 816                                 return 0;
 817                         nr -= PAGE_SIZE;
 818                 }
 819                 return dump_emit(cprm, zeroes, nr);
 820         }
 821 }
 822 EXPORT_SYMBOL(dump_skip);
 823
 824 int dump_align(struct coredump_params *cprm, int align)
 825 {
 826         unsigned mod = cprm->written & (align - 1);
 827         if (align & (align - 1))
 828                 return 0;
 829         return mod ? dump_skip(cprm, align - mod) : 1;
 830 }
 831 EXPORT_SYMBOL(dump_align);