fs/coredump.c

   1 #include <linux/slab.h>
   2 #include <linux/file.h>
   3 #include <linux/fdtable.h>
   4 #include <linux/freezer.h>
   5 #include <linux/mm.h>
   6 #include <linux/stat.h>
   7 #include <linux/fcntl.h>
   8 #include <linux/swap.h>
   9 #include <linux/string.h>
  10 #include <linux/init.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/perf_event.h>
  13 #include <linux/highmem.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/key.h>
  16 #include <linux/personality.h>
  17 #include <linux/binfmts.h>
  18 #include <linux/coredump.h>
  19 #include <linux/sched/coredump.h>
  20 #include <linux/sched/signal.h>
  21 #include <linux/sched/task_stack.h>
  22 #include <linux/utsname.h>
  23 #include <linux/pid_namespace.h>
  24 #include <linux/module.h>
  25 #include <linux/namei.h>
  26 #include <linux/mount.h>
  27 #include <linux/security.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/tsacct_kern.h>
  30 #include <linux/cn_proc.h>
  31 #include <linux/audit.h>
  32 #include <linux/tracehook.h>
  33 #include <linux/kmod.h>
  34 #include <linux/fsnotify.h>
  35 #include <linux/fs_struct.h>
  36 #include <linux/pipe_fs_i.h>
  37 #include <linux/oom.h>
  38 #include <linux/compat.h>
  39 #include <linux/fs.h>
  40 #include <linux/path.h>
  41 #include <linux/timekeeping.h>
  42
  43 #include <linux/uaccess.h>
  44 #include <asm/mmu_context.h>
  45 #include <asm/tlb.h>
  46 #include <asm/exec.h>
  47
  48 #include <trace/events/task.h>
  49 #include "internal.h"
  50
  51 #include <trace/events/sched.h>
  52
  53 int core_uses_pid;
  54 unsigned int core_pipe_limit;
  55 char core_pattern[CORENAME_MAX_SIZE] = "core";
  56 static int core_name_size = CORENAME_MAX_SIZE;
  57
  58 struct core_name {
  59         char *corename;
  60         int used, size;
  61 };
  62
  63 /* The maximal length of core_pattern is also specified in sysctl.c */
  64
  65 static int expand_corename(struct core_name *cn, int size)
  66 {
  67         char *corename = krealloc(cn->corename, size, GFP_KERNEL);
  68
  69         if (!corename)
  70                 return -ENOMEM;
  71
  72         if (size > core_name_size) /* racy but harmless */
  73                 core_name_size = size;
  74
  75         cn->size = ksize(corename);
  76         cn->corename = corename;
  77         return 0;
  78 }
  79
  80 static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
  81                                      va_list arg)
  82 {
  83         int free, need;
  84         va_list arg_copy;
  85
  86 again:
  87         free = cn->size - cn->used;
  88
  89         va_copy(arg_copy, arg);
  90         need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
  91         va_end(arg_copy);
  92
  93         if (need < free) {
  94                 cn->used += need;
  95                 return 0;
  96         }
  97
  98         if (!expand_corename(cn, cn->size + need - free + 1))
  99                 goto again;
 100
 101         return -ENOMEM;
 102 }
 103
 104 static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
 105 {
 106         va_list arg;
 107         int ret;
 108
 109         va_start(arg, fmt);
 110         ret = cn_vprintf(cn, fmt, arg);
 111         va_end(arg);
 112
 113         return ret;
 114 }
 115
 116 static __printf(2, 3)
 117 int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
 118 {
 119         int cur = cn->used;
 120         va_list arg;
 121         int ret;
 122
 123         va_start(arg, fmt);
 124         ret = cn_vprintf(cn, fmt, arg);
 125         va_end(arg);
 126
 127         if (ret == 0) {
 128                 /*
 129                  * Ensure that this coredump name component can't cause the
 130                  * resulting corefile path to consist of a ".." or ".".
 131                  */
 132                 if ((cn->used - cur == 1 && cn->corename[cur] == '.') ||
 133                                 (cn->used - cur == 2 && cn->corename[cur] == '.'
 134                                 && cn->corename[cur+1] == '.'))
 135                         cn->corename[cur] = '!';
 136
 137                 /*
 138                  * Empty names are fishy and could be used to create a "//" in a
 139                  * corefile name, causing the coredump to happen one directory
 140                  * level too high. Enforce that all components of the core
 141                  * pattern are at least one character long.
 142                  */
 143                 if (cn->used == cur)
 144                         ret = cn_printf(cn, "!");
 145         }
 146
 147         for (; cur < cn->used; ++cur) {
 148                 if (cn->corename[cur] == '/')
 149                         cn->corename[cur] = '!';
 150         }
 151         return ret;
 152 }
 153
 154 static int cn_print_exe_file(struct core_name *cn)
 155 {
 156         struct file *exe_file;
 157         char *pathbuf, *path;
 158         int ret;
 159
 160         exe_file = get_mm_exe_file(current->mm);
 161         if (!exe_file)
 162                 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
 163
 164         pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
 165         if (!pathbuf) {
 166                 ret = -ENOMEM;
 167                 goto put_exe_file;
 168         }
 169
 170         path = file_path(exe_file, pathbuf, PATH_MAX);
 171         if (IS_ERR(path)) {
 172                 ret = PTR_ERR(path);
 173                 goto free_buf;
 174         }
 175
 176         ret = cn_esc_printf(cn, "%s", path);
 177
 178 free_buf:
 179         kfree(pathbuf);
 180 put_exe_file:
 181         fput(exe_file);
 182         return ret;
 183 }
 184
 185 /* format_corename will inspect the pattern parameter, and output a
 186  * name into corename, which must have space for at least
 187  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
 188  */
 189 static int format_corename(struct core_name *cn, struct coredump_params *cprm)
 190 {
 191         const struct cred *cred = current_cred();
 192         const char *pat_ptr = core_pattern;
 193         int ispipe = (*pat_ptr == '|');
 194         int pid_in_pattern = 0;
 195         int err = 0;
 196
 197         cn->used = 0;
 198         cn->corename = NULL;
 199         if (expand_corename(cn, core_name_size))
 200                 return -ENOMEM;
 201         cn->corename[0] = '\0';
 202
 203         if (ispipe)
 204                 ++pat_ptr;
 205
 206         /* Repeat as long as we have more pattern to process and more output
 207            space */
 208         while (*pat_ptr) {
 209                 if (*pat_ptr != '%') {
 210                         err = cn_printf(cn, "%c", *pat_ptr++);
 211                 } else {
 212                         switch (*++pat_ptr) {
 213                         /* single % at the end, drop that */
 214                         case 0:
 215                                 goto out;
 216                         /* Double percent, output one percent */
 217                         case '%':
 218                                 err = cn_printf(cn, "%c", '%');
 219                                 break;
 220                         /* pid */
 221                         case 'p':
 222                                 pid_in_pattern = 1;
 223                                 err = cn_printf(cn, "%d",
 224                                               task_tgid_vnr(current));
 225                                 break;
 226                         /* global pid */
 227                         case 'P':
 228                                 err = cn_printf(cn, "%d",
 229                                               task_tgid_nr(current));
 230                                 break;
 231                         case 'i':
 232                                 err = cn_printf(cn, "%d",
 233                                               task_pid_vnr(current));
 234                                 break;
 235                         case 'I':
 236                                 err = cn_printf(cn, "%d",
 237                                               task_pid_nr(current));
 238                                 break;
 239                         /* uid */
 240                         case 'u':
 241                                 err = cn_printf(cn, "%u",
 242                                                 from_kuid(&init_user_ns,
 243                                                           cred->uid));
 244                                 break;
 245                         /* gid */
 246                         case 'g':
 247                                 err = cn_printf(cn, "%u",
 248                                                 from_kgid(&init_user_ns,
 249                                                           cred->gid));
 250                                 break;
 251                         case 'd':
 252                                 err = cn_printf(cn, "%d",
 253                                         __get_dumpable(cprm->mm_flags));
 254                                 break;
 255                         /* signal that caused the coredump */
 256                         case 's':
 257                                 err = cn_printf(cn, "%d",
 258                                                 cprm->siginfo->si_signo);
 259                                 break;
 260                         /* UNIX time of coredump */
 261                         case 't': {
 262                                 time64_t time;
 263
 264                                 time = ktime_get_real_seconds();
 265                                 err = cn_printf(cn, "%lld", time);
 266                                 break;
 267                         }
 268                         /* hostname */
 269                         case 'h':
 270                                 down_read(&uts_sem);
 271                                 err = cn_esc_printf(cn, "%s",
 272                                               utsname()->nodename);
 273                                 up_read(&uts_sem);
 274                                 break;
 275                         /* executable */
 276                         case 'e':
 277                                 err = cn_esc_printf(cn, "%s", current->comm);
 278                                 break;
 279                         case 'E':
 280                                 err = cn_print_exe_file(cn);
 281                                 break;
 282                         /* core limit size */
 283                         case 'c':
 284                                 err = cn_printf(cn, "%lu",
 285                                               rlimit(RLIMIT_CORE));
 286                                 break;
 287                         default:
 288                                 break;
 289                         }
 290                         ++pat_ptr;
 291                 }
 292
 293                 if (err)
 294                         return err;
 295         }
 296
 297 out:
 298         /* Backward compatibility with core_uses_pid:
 299          *
 300          * If core_pattern does not include a %p (as is the default)
 301          * and core_uses_pid is set, then .%pid will be appended to
 302          * the filename. Do not do this for piped commands. */
 303         if (!ispipe && !pid_in_pattern && core_uses_pid) {
 304                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
 305                 if (err)
 306                         return err;
 307         }
 308         return ispipe;
 309 }
 310
 311 static int zap_process(struct task_struct *start, int exit_code, int flags)
 312 {
 313         struct task_struct *t;
 314         int nr = 0;
 315
 316         /* ignore all signals except SIGKILL, see prepare_signal() */
 317         start->signal->flags = SIGNAL_GROUP_COREDUMP | flags;
 318         start->signal->group_exit_code = exit_code;
 319         start->signal->group_stop_count = 0;
 320
 321         for_each_thread(start, t) {
 322                 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
 323                 if (t != current && t->mm) {
 324                         sigaddset(&t->pending.signal, SIGKILL);
 325                         signal_wake_up(t, 1);
 326                         nr++;
 327                 }
 328         }
 329
 330         return nr;
 331 }
 332
 333 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
 334                         struct core_state *core_state, int exit_code)
 335 {
 336         struct task_struct *g, *p;
 337         unsigned long flags;
 338         int nr = -EAGAIN;
 339
 340         spin_lock_irq(&tsk->sighand->siglock);
 341         if (!signal_group_exit(tsk->signal)) {
 342                 mm->core_state = core_state;
 343                 tsk->signal->group_exit_task = tsk;
 344                 nr = zap_process(tsk, exit_code, 0);
 345                 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
 346         }
 347         spin_unlock_irq(&tsk->sighand->siglock);
 348         if (unlikely(nr < 0))
 349                 return nr;
 350
 351         tsk->flags |= PF_DUMPCORE;
 352         if (atomic_read(&mm->mm_users) == nr + 1)
 353                 goto done;
 354         /*
 355          * We should find and kill all tasks which use this mm, and we should
 356          * count them correctly into ->nr_threads. We don't take tasklist
 357          * lock, but this is safe wrt:
 358          *
 359          * fork:
 360          *      None of sub-threads can fork after zap_process(leader). All
 361          *      processes which were created before this point should be
 362          *      visible to zap_threads() because copy_process() adds the new
 363          *      process to the tail of init_task.tasks list, and lock/unlock
 364          *      of ->siglock provides a memory barrier.
 365          *
 366          * do_exit:
 367          *      The caller holds mm->mmap_sem. This means that the task which
 368          *      uses this mm can't pass exit_mm(), so it can't exit or clear
 369          *      its ->mm.
 370          *
 371          * de_thread:
 372          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
 373          *      we must see either old or new leader, this does not matter.
 374          *      However, it can change p->sighand, so lock_task_sighand(p)
 375          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
 376          *      it can't fail.
 377          *
 378          *      Note also that "g" can be the old leader with ->mm == NULL
 379          *      and already unhashed and thus removed from ->thread_group.
 380          *      This is OK, __unhash_process()->list_del_rcu() does not
 381          *      clear the ->next pointer, we will find the new leader via
 382          *      next_thread().
 383          */
 384         rcu_read_lock();
 385         for_each_process(g) {
 386                 if (g == tsk->group_leader)
 387                         continue;
 388                 if (g->flags & PF_KTHREAD)
 389                         continue;
 390
 391                 for_each_thread(g, p) {
 392                         if (unlikely(!p->mm))
 393                                 continue;
 394                         if (unlikely(p->mm == mm)) {
 395                                 lock_task_sighand(p, &flags);
 396                                 nr += zap_process(p, exit_code,
 397                                                         SIGNAL_GROUP_EXIT);
 398                                 unlock_task_sighand(p, &flags);
 399                         }
 400                         break;
 401                 }
 402         }
 403         rcu_read_unlock();
 404 done:
 405         atomic_set(&core_state->nr_threads, nr);
 406         return nr;
 407 }
 408
 409 static int coredump_wait(int exit_code, struct core_state *core_state)
 410 {
 411         struct task_struct *tsk = current;
 412         struct mm_struct *mm = tsk->mm;
 413         int core_waiters = -EBUSY;
 414
 415         init_completion(&core_state->startup);
 416         core_state->dumper.task = tsk;
 417         core_state->dumper.next = NULL;
 418
 419         if (down_write_killable(&mm->mmap_sem))
 420                 return -EINTR;
 421
 422         if (!mm->core_state)
 423                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
 424         up_write(&mm->mmap_sem);
 425
 426         if (core_waiters > 0) {
 427                 struct core_thread *ptr;
 428
 429                 freezer_do_not_count();
 430                 wait_for_completion(&core_state->startup);
 431                 freezer_count();
 432                 /*
 433                  * Wait for all the threads to become inactive, so that
 434                  * all the thread context (extended register state, like
 435                  * fpu etc) gets copied to the memory.
 436                  */
 437                 ptr = core_state->dumper.next;
 438                 while (ptr != NULL) {
 439                         wait_task_inactive(ptr->task, 0);
 440                         ptr = ptr->next;
 441                 }
 442         }
 443
 444         return core_waiters;
 445 }
 446
 447 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
 448 {
 449         struct core_thread *curr, *next;
 450         struct task_struct *task;
 451
 452         spin_lock_irq(&current->sighand->siglock);
 453         if (core_dumped && !__fatal_signal_pending(current))
 454                 current->signal->group_exit_code |= 0x80;
 455         current->signal->group_exit_task = NULL;
 456         current->signal->flags = SIGNAL_GROUP_EXIT;
 457         spin_unlock_irq(&current->sighand->siglock);
 458
 459         next = mm->core_state->dumper.next;
 460         while ((curr = next) != NULL) {
 461                 next = curr->next;
 462                 task = curr->task;
 463                 /*
 464                  * see exit_mm(), curr->task must not see
 465                  * ->task == NULL before we read ->next.
 466                  */
 467                 smp_mb();
 468                 curr->task = NULL;
 469                 wake_up_process(task);
 470         }
 471
 472         mm->core_state = NULL;
 473 }
 474
 475 static bool dump_interrupted(void)
 476 {
 477         /*
 478          * SIGKILL or freezing() interrupt the coredumping. Perhaps we
 479          * can do try_to_freeze() and check __fatal_signal_pending(),
 480          * but then we need to teach dump_write() to restart and clear
 481          * TIF_SIGPENDING.
 482          */
 483         return signal_pending(current);
 484 }
 485
 486 static void wait_for_dump_helpers(struct file *file)
 487 {
 488         struct pipe_inode_info *pipe = file->private_data;
 489
 490         pipe_lock(pipe);
 491         pipe->readers++;
 492         pipe->writers--;
 493         wake_up_interruptible_sync(&pipe->wait);
 494         kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
 495         pipe_unlock(pipe);
 496
 497         /*
 498          * We actually want wait_event_freezable() but then we need
 499          * to clear TIF_SIGPENDING and improve dump_interrupted().
 500          */
 501         wait_event_interruptible(pipe->wait, pipe->readers == 1);
 502
 503         pipe_lock(pipe);
 504         pipe->readers--;
 505         pipe->writers++;
 506         pipe_unlock(pipe);
 507 }
 508
 509 /*
 510  * umh_pipe_setup
 511  * helper function to customize the process used
 512  * to collect the core in userspace.  Specifically
 513  * it sets up a pipe and installs it as fd 0 (stdin)
 514  * for the process.  Returns 0 on success, or
 515  * PTR_ERR on failure.
 516  * Note that it also sets the core limit to 1.  This
 517  * is a special value that we use to trap recursive
 518  * core dumps
 519  */
 520 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
 521 {
 522         struct file *files[2];
 523         struct coredump_params *cp = (struct coredump_params *)info->data;
 524         int err = create_pipe_files(files, 0);
 525         if (err)
 526                 return err;
 527
 528         cp->file = files[1];
 529
 530         err = replace_fd(0, files[0], 0);
 531         fput(files[0]);
 532         /* and disallow core files too */
 533         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
 534
 535         return err;
 536 }
 537
 538 void do_coredump(const siginfo_t *siginfo)
 539 {
 540         struct core_state core_state;
 541         struct core_name cn;
 542         struct mm_struct *mm = current->mm;
 543         struct linux_binfmt * binfmt;
 544         const struct cred *old_cred;
 545         struct cred *cred;
 546         int retval = 0;
 547         int ispipe;
 548         struct files_struct *displaced;
 549         /* require nonrelative corefile path and be extra careful */
 550         bool need_suid_safe = false;
 551         bool core_dumped = false;
 552         static atomic_t core_dump_count = ATOMIC_INIT(0);
 553         struct coredump_params cprm = {
 554                 .siginfo = siginfo,
 555                 .regs = signal_pt_regs(),
 556                 .limit = rlimit(RLIMIT_CORE),
 557                 /*
 558                  * We must use the same mm->flags while dumping core to avoid
 559                  * inconsistency of bit flags, since this flag is not protected
 560                  * by any locks.
 561                  */
 562                 .mm_flags = mm->flags,
 563         };
 564
 565         audit_core_dumps(siginfo->si_signo);
 566
 567         binfmt = mm->binfmt;
 568         if (!binfmt || !binfmt->core_dump)
 569                 goto fail;
 570         if (!__get_dumpable(cprm.mm_flags))
 571                 goto fail;
 572
 573         cred = prepare_creds();
 574         if (!cred)
 575                 goto fail;
 576         /*
 577          * We cannot trust fsuid as being the "true" uid of the process
 578          * nor do we know its entire history. We only know it was tainted
 579          * so we dump it as root in mode 2, and only into a controlled
 580          * environment (pipe handler or fully qualified path).
 581          */
 582         if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
 583                 /* Setuid core dump mode */
 584                 cred->fsuid = GLOBAL_ROOT_UID;  /* Dump root private */
 585                 need_suid_safe = true;
 586         }
 587
 588         retval = coredump_wait(siginfo->si_signo, &core_state);
 589         if (retval < 0)
 590                 goto fail_creds;
 591
 592         old_cred = override_creds(cred);
 593
 594         ispipe = format_corename(&cn, &cprm);
 595
 596         if (ispipe) {
 597                 int dump_count;
 598                 char **helper_argv;
 599                 struct subprocess_info *sub_info;
 600
 601                 if (ispipe < 0) {
 602                         printk(KERN_WARNING "format_corename failed\n");
 603                         printk(KERN_WARNING "Aborting core\n");
 604                         goto fail_unlock;
 605                 }
 606
 607                 if (cprm.limit == 1) {
 608                         /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
 609                          *
 610                          * Normally core limits are irrelevant to pipes, since
 611                          * we're not writing to the file system, but we use
 612                          * cprm.limit of 1 here as a special value, this is a
 613                          * consistent way to catch recursive crashes.
 614                          * We can still crash if the core_pattern binary sets
 615                          * RLIM_CORE = !1, but it runs as root, and can do
 616                          * lots of stupid things.
 617                          *
 618                          * Note that we use task_tgid_vnr here to grab the pid
 619                          * of the process group leader.  That way we get the
 620                          * right pid if a thread in a multi-threaded
 621                          * core_pattern process dies.
 622                          */
 623                         printk(KERN_WARNING
 624                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
 625                                 task_tgid_vnr(current), current->comm);
 626                         printk(KERN_WARNING "Aborting core\n");
 627                         goto fail_unlock;
 628                 }
 629                 cprm.limit = RLIM_INFINITY;
 630
 631                 dump_count = atomic_inc_return(&core_dump_count);
 632                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
 633                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
 634                                task_tgid_vnr(current), current->comm);
 635                         printk(KERN_WARNING "Skipping core dump\n");
 636                         goto fail_dropcount;
 637                 }
 638
 639                 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
 640                 if (!helper_argv) {
 641                         printk(KERN_WARNING "%s failed to allocate memory\n",
 642                                __func__);
 643                         goto fail_dropcount;
 644                 }
 645
 646                 retval = -ENOMEM;
 647                 sub_info = call_usermodehelper_setup(helper_argv[0],
 648                                                 helper_argv, NULL, GFP_KERNEL,
 649                                                 umh_pipe_setup, NULL, &cprm);
 650                 if (sub_info)
 651                         retval = call_usermodehelper_exec(sub_info,
 652                                                           UMH_WAIT_EXEC);
 653
 654                 argv_free(helper_argv);
 655                 if (retval) {
 656                         printk(KERN_INFO "Core dump to |%s pipe failed\n",
 657                                cn.corename);
 658                         goto close_fail;
 659                 }
 660         } else {
 661                 struct inode *inode;
 662                 int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
 663                                  O_LARGEFILE | O_EXCL;
 664
 665                 if (cprm.limit < binfmt->min_coredump)
 666                         goto fail_unlock;
 667
 668                 if (need_suid_safe && cn.corename[0] != '/') {
 669                         printk(KERN_WARNING "Pid %d(%s) can only dump core "\
 670                                 "to fully qualified path!\n",
 671                                 task_tgid_vnr(current), current->comm);
 672                         printk(KERN_WARNING "Skipping core dump\n");
 673                         goto fail_unlock;
 674                 }
 675
 676                 /*
 677                  * Unlink the file if it exists unless this is a SUID
 678                  * binary - in that case, we're running around with root
 679                  * privs and don't want to unlink another user's coredump.
 680                  */
 681                 if (!need_suid_safe) {
 682                         mm_segment_t old_fs;
 683
 684                         old_fs = get_fs();
 685                         set_fs(KERNEL_DS);
 686                         /*
 687                          * If it doesn't exist, that's fine. If there's some
 688                          * other problem, we'll catch it at the filp_open().
 689                          */
 690                         (void) sys_unlink((const char __user *)cn.corename);
 691                         set_fs(old_fs);
 692                 }
 693
 694                 /*
 695                  * There is a race between unlinking and creating the
 696                  * file, but if that causes an EEXIST here, that's
 697                  * fine - another process raced with us while creating
 698                  * the corefile, and the other process won. To userspace,
 699                  * what matters is that at least one of the two processes
 700                  * writes its coredump successfully, not which one.
 701                  */
 702                 if (need_suid_safe) {
 703                         /*
 704                          * Using user namespaces, normal user tasks can change
 705                          * their current->fs->root to point to arbitrary
 706                          * directories. Since the intention of the "only dump
 707                          * with a fully qualified path" rule is to control where
 708                          * coredumps may be placed using root privileges,
 709                          * current->fs->root must not be used. Instead, use the
 710                          * root directory of init_task.
 711                          */
 712                         struct path root;
 713
 714                         task_lock(&init_task);
 715                         get_fs_root(init_task.fs, &root);
 716                         task_unlock(&init_task);
 717                         cprm.file = file_open_root(root.dentry, root.mnt,
 718                                 cn.corename, open_flags, 0600);
 719                         path_put(&root);
 720                 } else {
 721                         cprm.file = filp_open(cn.corename, open_flags, 0600);
 722                 }
 723                 if (IS_ERR(cprm.file))
 724                         goto fail_unlock;
 725
 726                 inode = file_inode(cprm.file);
 727                 if (inode->i_nlink > 1)
 728                         goto close_fail;
 729                 if (d_unhashed(cprm.file->f_path.dentry))
 730                         goto close_fail;
 731                 /*
 732                  * AK: actually i see no reason to not allow this for named
 733                  * pipes etc, but keep the previous behaviour for now.
 734                  */
 735                 if (!S_ISREG(inode->i_mode))
 736                         goto close_fail;
 737                 /*
 738                  * Don't dump core if the filesystem changed owner or mode
 739                  * of the file during file creation. This is an issue when
 740                  * a process dumps core while its cwd is e.g. on a vfat
 741                  * filesystem.
 742                  */
 743                 if (!uid_eq(inode->i_uid, current_fsuid()))
 744                         goto close_fail;
 745                 if ((inode->i_mode & 0677) != 0600)
 746                         goto close_fail;
 747                 if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
 748                         goto close_fail;
 749                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
 750                         goto close_fail;
 751         }
 752
 753         /* get us an unshared descriptor table; almost always a no-op */
 754         retval = unshare_files(&displaced);
 755         if (retval)
 756                 goto close_fail;
 757         if (displaced)
 758                 put_files_struct(displaced);
 759         if (!dump_interrupted()) {
 760                 file_start_write(cprm.file);
 761                 core_dumped = binfmt->core_dump(&cprm);
 762                 file_end_write(cprm.file);
 763         }
 764         if (ispipe && core_pipe_limit)
 765                 wait_for_dump_helpers(cprm.file);
 766 close_fail:
 767         if (cprm.file)
 768                 filp_close(cprm.file, NULL);
 769 fail_dropcount:
 770         if (ispipe)
 771                 atomic_dec(&core_dump_count);
 772 fail_unlock:
 773         kfree(cn.corename);
 774         coredump_finish(mm, core_dumped);
 775         revert_creds(old_cred);
 776 fail_creds:
 777         put_cred(cred);
 778 fail:
 779         return;
 780 }
 781
 782 /*
 783  * Core dumping helper functions.  These are the only things you should
 784  * do on a core-file: use only these functions to write out all the
 785  * necessary info.
 786  */
 787 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
 788 {
 789         struct file *file = cprm->file;
 790         loff_t pos = file->f_pos;
 791         ssize_t n;
 792         if (cprm->written + nr > cprm->limit)
 793                 return 0;
 794         while (nr) {
 795                 if (dump_interrupted())
 796                         return 0;
 797                 n = __kernel_write(file, addr, nr, &pos);
 798                 if (n <= 0)
 799                         return 0;
 800                 file->f_pos = pos;
 801                 cprm->written += n;
 802                 cprm->pos += n;
 803                 nr -= n;
 804         }
 805         return 1;
 806 }
 807 EXPORT_SYMBOL(dump_emit);
 808
 809 int dump_skip(struct coredump_params *cprm, size_t nr)
 810 {
 811         static char zeroes[PAGE_SIZE];
 812         struct file *file = cprm->file;
 813         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
 814                 if (dump_interrupted() ||
 815                     file->f_op->llseek(file, nr, SEEK_CUR) < 0)
 816                         return 0;
 817                 cprm->pos += nr;
 818                 return 1;
 819         } else {
 820                 while (nr > PAGE_SIZE) {
 821                         if (!dump_emit(cprm, zeroes, PAGE_SIZE))
 822                                 return 0;
 823                         nr -= PAGE_SIZE;
 824                 }
 825                 return dump_emit(cprm, zeroes, nr);
 826         }
 827 }
 828 EXPORT_SYMBOL(dump_skip);
 829
 830 int dump_align(struct coredump_params *cprm, int align)
 831 {
 832         unsigned mod = cprm->pos & (align - 1);
 833         if (align & (align - 1))
 834                 return 0;
 835         return mod ? dump_skip(cprm, align - mod) : 1;
 836 }
 837 EXPORT_SYMBOL(dump_align);
 838
 839 /*
 840  * Ensures that file size is big enough to contain the current file
 841  * postion. This prevents gdb from complaining about a truncated file
 842  * if the last "write" to the file was dump_skip.
 843  */
 844 void dump_truncate(struct coredump_params *cprm)
 845 {
 846         struct file *file = cprm->file;
 847         loff_t offset;
 848
 849         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
 850                 offset = file->f_op->llseek(file, 0, SEEK_CUR);
 851                 if (i_size_read(file->f_mapping->host) < offset)
 852                         do_truncate(file->f_path.dentry, offset, 0, file);
 853         }
 854 }
 855 EXPORT_SYMBOL(dump_truncate);