fs/overlayfs/super.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/fs.h>
  11 #include <linux/namei.h>
  12 #include <linux/xattr.h>
  13 #include <linux/mount.h>
  14 #include <linux/parser.h>
  15 #include <linux/module.h>
  16 #include <linux/statfs.h>
  17 #include <linux/seq_file.h>
  18 #include <linux/posix_acl_xattr.h>
  19 #include "overlayfs.h"
  20 #include "ovl_entry.h"
  21
  22 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  23 MODULE_DESCRIPTION("Overlay filesystem");
  24 MODULE_LICENSE("GPL");
  25
  26
  27 struct ovl_dir_cache;
  28
  29 #define OVL_MAX_STACK 500
  30
  31
  32 static void ovl_dentry_release(struct dentry *dentry)
  33 {
  34         struct ovl_entry *oe = dentry->d_fsdata;
  35
  36         if (oe) {
  37                 unsigned int i;
  38
  39                 dput(oe->__upperdentry);
  40                 kfree(oe->redirect);
  41                 for (i = 0; i < oe->numlower; i++)
  42                         dput(oe->lowerstack[i].dentry);
  43                 kfree_rcu(oe, rcu);
  44         }
  45 }
  46
  47 static struct dentry *ovl_d_real(struct dentry *dentry,
  48                                  const struct inode *inode,
  49                                  unsigned int open_flags)
  50 {
  51         struct dentry *real;
  52
  53         if (!d_is_reg(dentry)) {
  54                 if (!inode || inode == d_inode(dentry))
  55                         return dentry;
  56                 goto bug;
  57         }
  58
  59         if (d_is_negative(dentry))
  60                 return dentry;
  61
  62         if (open_flags) {
  63                 int err = ovl_open_maybe_copy_up(dentry, open_flags);
  64
  65                 if (err)
  66                         return ERR_PTR(err);
  67         }
  68
  69         real = ovl_dentry_upper(dentry);
  70         if (real && (!inode || inode == d_inode(real)))
  71                 return real;
  72
  73         real = ovl_dentry_lower(dentry);
  74         if (!real)
  75                 goto bug;
  76
  77         /* Handle recursion */
  78         real = d_real(real, inode, open_flags);
  79
  80         if (!inode || inode == d_inode(real))
  81                 return real;
  82 bug:
  83         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
  84              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
  85         return dentry;
  86 }
  87
  88 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
  89 {
  90         struct ovl_entry *oe = dentry->d_fsdata;
  91         unsigned int i;
  92         int ret = 1;
  93
  94         for (i = 0; i < oe->numlower; i++) {
  95                 struct dentry *d = oe->lowerstack[i].dentry;
  96
  97                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
  98                         ret = d->d_op->d_revalidate(d, flags);
  99                         if (ret < 0)
 100                                 return ret;
 101                         if (!ret) {
 102                                 if (!(flags & LOOKUP_RCU))
 103                                         d_invalidate(d);
 104                                 return -ESTALE;
 105                         }
 106                 }
 107         }
 108         return 1;
 109 }
 110
 111 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 112 {
 113         struct ovl_entry *oe = dentry->d_fsdata;
 114         unsigned int i;
 115         int ret = 1;
 116
 117         for (i = 0; i < oe->numlower; i++) {
 118                 struct dentry *d = oe->lowerstack[i].dentry;
 119
 120                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 121                         ret = d->d_op->d_weak_revalidate(d, flags);
 122                         if (ret <= 0)
 123                                 break;
 124                 }
 125         }
 126         return ret;
 127 }
 128
 129 static const struct dentry_operations ovl_dentry_operations = {
 130         .d_release = ovl_dentry_release,
 131         .d_real = ovl_d_real,
 132 };
 133
 134 static const struct dentry_operations ovl_reval_dentry_operations = {
 135         .d_release = ovl_dentry_release,
 136         .d_real = ovl_d_real,
 137         .d_revalidate = ovl_dentry_revalidate,
 138         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 139 };
 140
 141 static void ovl_put_super(struct super_block *sb)
 142 {
 143         struct ovl_fs *ufs = sb->s_fs_info;
 144         unsigned i;
 145
 146         dput(ufs->workdir);
 147         mntput(ufs->upper_mnt);
 148         for (i = 0; i < ufs->numlower; i++)
 149                 mntput(ufs->lower_mnt[i]);
 150         kfree(ufs->lower_mnt);
 151
 152         kfree(ufs->config.lowerdir);
 153         kfree(ufs->config.upperdir);
 154         kfree(ufs->config.workdir);
 155         put_cred(ufs->creator_cred);
 156         kfree(ufs);
 157 }
 158
 159 /**
 160  * ovl_statfs
 161  * @sb: The overlayfs super block
 162  * @buf: The struct kstatfs to fill in with stats
 163  *
 164  * Get the filesystem statistics.  As writes always target the upper layer
 165  * filesystem pass the statfs to the upper filesystem (if it exists)
 166  */
 167 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 168 {
 169         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 170         struct dentry *root_dentry = dentry->d_sb->s_root;
 171         struct path path;
 172         int err;
 173
 174         ovl_path_real(root_dentry, &path);
 175
 176         err = vfs_statfs(&path, buf);
 177         if (!err) {
 178                 buf->f_namelen = ofs->namelen;
 179                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 180         }
 181
 182         return err;
 183 }
 184
 185 /**
 186  * ovl_show_options
 187  *
 188  * Prints the mount options for a given superblock.
 189  * Returns zero; does not fail.
 190  */
 191 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 192 {
 193         struct super_block *sb = dentry->d_sb;
 194         struct ovl_fs *ufs = sb->s_fs_info;
 195
 196         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 197         if (ufs->config.upperdir) {
 198                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 199                 seq_show_option(m, "workdir", ufs->config.workdir);
 200         }
 201         if (ufs->config.default_permissions)
 202                 seq_puts(m, ",default_permissions");
 203         return 0;
 204 }
 205
 206 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 207 {
 208         struct ovl_fs *ufs = sb->s_fs_info;
 209
 210         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 211                 return -EROFS;
 212
 213         return 0;
 214 }
 215
 216 static const struct super_operations ovl_super_operations = {
 217         .put_super      = ovl_put_super,
 218         .statfs         = ovl_statfs,
 219         .show_options   = ovl_show_options,
 220         .remount_fs     = ovl_remount,
 221         .drop_inode     = generic_delete_inode,
 222 };
 223
 224 enum {
 225         OPT_LOWERDIR,
 226         OPT_UPPERDIR,
 227         OPT_WORKDIR,
 228         OPT_DEFAULT_PERMISSIONS,
 229         OPT_REDIRECT_DIR_ON,
 230         OPT_REDIRECT_DIR_OFF,
 231         OPT_ERR,
 232 };
 233
 234 static const match_table_t ovl_tokens = {
 235         {OPT_LOWERDIR,                  "lowerdir=%s"},
 236         {OPT_UPPERDIR,                  "upperdir=%s"},
 237         {OPT_WORKDIR,                   "workdir=%s"},
 238         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 239         {OPT_REDIRECT_DIR_ON,           "redirect_dir=on"},
 240         {OPT_REDIRECT_DIR_OFF,          "redirect_dir=off"},
 241         {OPT_ERR,                       NULL}
 242 };
 243
 244 static char *ovl_next_opt(char **s)
 245 {
 246         char *sbegin = *s;
 247         char *p;
 248
 249         if (sbegin == NULL)
 250                 return NULL;
 251
 252         for (p = sbegin; *p; p++) {
 253                 if (*p == '\\') {
 254                         p++;
 255                         if (!*p)
 256                                 break;
 257                 } else if (*p == ',') {
 258                         *p = '\0';
 259                         *s = p + 1;
 260                         return sbegin;
 261                 }
 262         }
 263         *s = NULL;
 264         return sbegin;
 265 }
 266
 267 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 268 {
 269         char *p;
 270
 271         while ((p = ovl_next_opt(&opt)) != NULL) {
 272                 int token;
 273                 substring_t args[MAX_OPT_ARGS];
 274
 275                 if (!*p)
 276                         continue;
 277
 278                 token = match_token(p, ovl_tokens, args);
 279                 switch (token) {
 280                 case OPT_UPPERDIR:
 281                         kfree(config->upperdir);
 282                         config->upperdir = match_strdup(&args[0]);
 283                         if (!config->upperdir)
 284                                 return -ENOMEM;
 285                         break;
 286
 287                 case OPT_LOWERDIR:
 288                         kfree(config->lowerdir);
 289                         config->lowerdir = match_strdup(&args[0]);
 290                         if (!config->lowerdir)
 291                                 return -ENOMEM;
 292                         break;
 293
 294                 case OPT_WORKDIR:
 295                         kfree(config->workdir);
 296                         config->workdir = match_strdup(&args[0]);
 297                         if (!config->workdir)
 298                                 return -ENOMEM;
 299                         break;
 300
 301                 case OPT_DEFAULT_PERMISSIONS:
 302                         config->default_permissions = true;
 303                         break;
 304
 305                 case OPT_REDIRECT_DIR_ON:
 306                         config->redirect_dir = true;
 307                         break;
 308
 309                 case OPT_REDIRECT_DIR_OFF:
 310                         config->redirect_dir = false;
 311                         break;
 312
 313                 default:
 314                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 315                         return -EINVAL;
 316                 }
 317         }
 318
 319         /* Workdir is useless in non-upper mount */
 320         if (!config->upperdir && config->workdir) {
 321                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 322                         config->workdir);
 323                 kfree(config->workdir);
 324                 config->workdir = NULL;
 325         }
 326
 327         return 0;
 328 }
 329
 330 #define OVL_WORKDIR_NAME "work"
 331
 332 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 333                                          struct dentry *dentry)
 334 {
 335         struct inode *dir = dentry->d_inode;
 336         struct dentry *work;
 337         int err;
 338         bool retried = false;
 339
 340         err = mnt_want_write(mnt);
 341         if (err)
 342                 return ERR_PTR(err);
 343
 344         inode_lock_nested(dir, I_MUTEX_PARENT);
 345 retry:
 346         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 347                               strlen(OVL_WORKDIR_NAME));
 348
 349         if (!IS_ERR(work)) {
 350                 struct kstat stat = {
 351                         .mode = S_IFDIR | 0,
 352                 };
 353                 struct iattr attr = {
 354                         .ia_valid = ATTR_MODE,
 355                         .ia_mode = stat.mode,
 356                 };
 357
 358                 if (work->d_inode) {
 359                         err = -EEXIST;
 360                         if (retried)
 361                                 goto out_dput;
 362
 363                         retried = true;
 364                         ovl_workdir_cleanup(dir, mnt, work, 0);
 365                         dput(work);
 366                         goto retry;
 367                 }
 368
 369                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 370                 if (err)
 371                         goto out_dput;
 372
 373                 /*
 374                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 375                  *
 376                  * a) success (there was a POSIX ACL xattr and was removed)
 377                  * b) -ENODATA (there was no POSIX ACL xattr)
 378                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 379                  *
 380                  * There are various other error values that could effectively
 381                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 382                  * if the xattr name is too long), but the set of filesystems
 383                  * allowed as upper are limited to "normal" ones, where checking
 384                  * for the above two errors is sufficient.
 385                  */
 386                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 387                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 388                         goto out_dput;
 389
 390                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 391                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 392                         goto out_dput;
 393
 394                 /* Clear any inherited mode bits */
 395                 inode_lock(work->d_inode);
 396                 err = notify_change(work, &attr, NULL);
 397                 inode_unlock(work->d_inode);
 398                 if (err)
 399                         goto out_dput;
 400         }
 401 out_unlock:
 402         inode_unlock(dir);
 403         mnt_drop_write(mnt);
 404
 405         return work;
 406
 407 out_dput:
 408         dput(work);
 409         work = ERR_PTR(err);
 410         goto out_unlock;
 411 }
 412
 413 static void ovl_unescape(char *s)
 414 {
 415         char *d = s;
 416
 417         for (;; s++, d++) {
 418                 if (*s == '\\')
 419                         s++;
 420                 *d = *s;
 421                 if (!*s)
 422                         break;
 423         }
 424 }
 425
 426 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 427 {
 428         int err = -EINVAL;
 429
 430         if (!*name) {
 431                 pr_err("overlayfs: empty lowerdir\n");
 432                 goto out;
 433         }
 434         err = kern_path(name, LOOKUP_FOLLOW, path);
 435         if (err) {
 436                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 437                 goto out;
 438         }
 439         err = -EINVAL;
 440         if (ovl_dentry_weird(path->dentry)) {
 441                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 442                 goto out_put;
 443         }
 444         if (!d_is_dir(path->dentry)) {
 445                 pr_err("overlayfs: '%s' not a directory\n", name);
 446                 goto out_put;
 447         }
 448         return 0;
 449
 450 out_put:
 451         path_put(path);
 452 out:
 453         return err;
 454 }
 455
 456 static int ovl_mount_dir(const char *name, struct path *path)
 457 {
 458         int err = -ENOMEM;
 459         char *tmp = kstrdup(name, GFP_KERNEL);
 460
 461         if (tmp) {
 462                 ovl_unescape(tmp);
 463                 err = ovl_mount_dir_noesc(tmp, path);
 464
 465                 if (!err)
 466                         if (ovl_dentry_remote(path->dentry)) {
 467                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 468                                        tmp);
 469                                 path_put(path);
 470                                 err = -EINVAL;
 471                         }
 472                 kfree(tmp);
 473         }
 474         return err;
 475 }
 476
 477 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 478                              const char *name)
 479 {
 480         struct kstatfs statfs;
 481         int err = vfs_statfs(path, &statfs);
 482
 483         if (err)
 484                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 485         else
 486                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 487
 488         return err;
 489 }
 490
 491 static int ovl_lower_dir(const char *name, struct path *path,
 492                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 493 {
 494         int err;
 495
 496         err = ovl_mount_dir_noesc(name, path);
 497         if (err)
 498                 goto out;
 499
 500         err = ovl_check_namelen(path, ofs, name);
 501         if (err)
 502                 goto out_put;
 503
 504         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 505
 506         if (ovl_dentry_remote(path->dentry))
 507                 *remote = true;
 508
 509         return 0;
 510
 511 out_put:
 512         path_put(path);
 513 out:
 514         return err;
 515 }
 516
 517 /* Workdir should not be subdir of upperdir and vice versa */
 518 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 519 {
 520         bool ok = false;
 521
 522         if (workdir != upperdir) {
 523                 ok = (lock_rename(workdir, upperdir) == NULL);
 524                 unlock_rename(workdir, upperdir);
 525         }
 526         return ok;
 527 }
 528
 529 static unsigned int ovl_split_lowerdirs(char *str)
 530 {
 531         unsigned int ctr = 1;
 532         char *s, *d;
 533
 534         for (s = d = str;; s++, d++) {
 535                 if (*s == '\\') {
 536                         s++;
 537                 } else if (*s == ':') {
 538                         *d = '\0';
 539                         ctr++;
 540                         continue;
 541                 }
 542                 *d = *s;
 543                 if (!*s)
 544                         break;
 545         }
 546         return ctr;
 547 }
 548
 549 static int __maybe_unused
 550 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 551                         struct dentry *dentry, struct inode *inode,
 552                         const char *name, void *buffer, size_t size)
 553 {
 554         return ovl_xattr_get(dentry, handler->name, buffer, size);
 555 }
 556
 557 static int __maybe_unused
 558 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 559                         struct dentry *dentry, struct inode *inode,
 560                         const char *name, const void *value,
 561                         size_t size, int flags)
 562 {
 563         struct dentry *workdir = ovl_workdir(dentry);
 564         struct inode *realinode = ovl_inode_real(inode, NULL);
 565         struct posix_acl *acl = NULL;
 566         int err;
 567
 568         /* Check that everything is OK before copy-up */
 569         if (value) {
 570                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 571                 if (IS_ERR(acl))
 572                         return PTR_ERR(acl);
 573         }
 574         err = -EOPNOTSUPP;
 575         if (!IS_POSIXACL(d_inode(workdir)))
 576                 goto out_acl_release;
 577         if (!realinode->i_op->set_acl)
 578                 goto out_acl_release;
 579         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 580                 err = acl ? -EACCES : 0;
 581                 goto out_acl_release;
 582         }
 583         err = -EPERM;
 584         if (!inode_owner_or_capable(inode))
 585                 goto out_acl_release;
 586
 587         posix_acl_release(acl);
 588
 589         /*
 590          * Check if sgid bit needs to be cleared (actual setacl operation will
 591          * be done with mounter's capabilities and so that won't do it for us).
 592          */
 593         if (unlikely(inode->i_mode & S_ISGID) &&
 594             handler->flags == ACL_TYPE_ACCESS &&
 595             !in_group_p(inode->i_gid) &&
 596             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 597                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 598
 599                 err = ovl_setattr(dentry, &iattr);
 600                 if (err)
 601                         return err;
 602         }
 603
 604         err = ovl_xattr_set(dentry, handler->name, value, size, flags);
 605         if (!err)
 606                 ovl_copyattr(ovl_inode_real(inode, NULL), inode);
 607
 608         return err;
 609
 610 out_acl_release:
 611         posix_acl_release(acl);
 612         return err;
 613 }
 614
 615 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 616                              struct dentry *dentry, struct inode *inode,
 617                              const char *name, void *buffer, size_t size)
 618 {
 619         return -EOPNOTSUPP;
 620 }
 621
 622 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 623                              struct dentry *dentry, struct inode *inode,
 624                              const char *name, const void *value,
 625                              size_t size, int flags)
 626 {
 627         return -EOPNOTSUPP;
 628 }
 629
 630 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 631                                struct dentry *dentry, struct inode *inode,
 632                                const char *name, void *buffer, size_t size)
 633 {
 634         return ovl_xattr_get(dentry, name, buffer, size);
 635 }
 636
 637 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 638                                struct dentry *dentry, struct inode *inode,
 639                                const char *name, const void *value,
 640                                size_t size, int flags)
 641 {
 642         return ovl_xattr_set(dentry, name, value, size, flags);
 643 }
 644
 645 static const struct xattr_handler __maybe_unused
 646 ovl_posix_acl_access_xattr_handler = {
 647         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 648         .flags = ACL_TYPE_ACCESS,
 649         .get = ovl_posix_acl_xattr_get,
 650         .set = ovl_posix_acl_xattr_set,
 651 };
 652
 653 static const struct xattr_handler __maybe_unused
 654 ovl_posix_acl_default_xattr_handler = {
 655         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 656         .flags = ACL_TYPE_DEFAULT,
 657         .get = ovl_posix_acl_xattr_get,
 658         .set = ovl_posix_acl_xattr_set,
 659 };
 660
 661 static const struct xattr_handler ovl_own_xattr_handler = {
 662         .prefix = OVL_XATTR_PREFIX,
 663         .get = ovl_own_xattr_get,
 664         .set = ovl_own_xattr_set,
 665 };
 666
 667 static const struct xattr_handler ovl_other_xattr_handler = {
 668         .prefix = "", /* catch all */
 669         .get = ovl_other_xattr_get,
 670         .set = ovl_other_xattr_set,
 671 };
 672
 673 static const struct xattr_handler *ovl_xattr_handlers[] = {
 674 #ifdef CONFIG_FS_POSIX_ACL
 675         &ovl_posix_acl_access_xattr_handler,
 676         &ovl_posix_acl_default_xattr_handler,
 677 #endif
 678         &ovl_own_xattr_handler,
 679         &ovl_other_xattr_handler,
 680         NULL
 681 };
 682
 683 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 684 {
 685         struct path upperpath = { NULL, NULL };
 686         struct path workpath = { NULL, NULL };
 687         struct dentry *root_dentry;
 688         struct inode *realinode;
 689         struct ovl_entry *oe;
 690         struct ovl_fs *ufs;
 691         struct path *stack = NULL;
 692         char *lowertmp;
 693         char *lower;
 694         unsigned int numlower;
 695         unsigned int stacklen = 0;
 696         unsigned int i;
 697         bool remote = false;
 698         int err;
 699
 700         err = -ENOMEM;
 701         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 702         if (!ufs)
 703                 goto out;
 704
 705         err = ovl_parse_opt((char *) data, &ufs->config);
 706         if (err)
 707                 goto out_free_config;
 708
 709         err = -EINVAL;
 710         if (!ufs->config.lowerdir) {
 711                 if (!silent)
 712                         pr_err("overlayfs: missing 'lowerdir'\n");
 713                 goto out_free_config;
 714         }
 715
 716         sb->s_stack_depth = 0;
 717         sb->s_maxbytes = MAX_LFS_FILESIZE;
 718         if (ufs->config.upperdir) {
 719                 if (!ufs->config.workdir) {
 720                         pr_err("overlayfs: missing 'workdir'\n");
 721                         goto out_free_config;
 722                 }
 723
 724                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 725                 if (err)
 726                         goto out_free_config;
 727
 728                 /* Upper fs should not be r/o */
 729                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 730                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 731                         err = -EINVAL;
 732                         goto out_put_upperpath;
 733                 }
 734
 735                 err = ovl_check_namelen(&upperpath, ufs, ufs->config.upperdir);
 736                 if (err)
 737                         goto out_put_upperpath;
 738
 739                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 740                 if (err)
 741                         goto out_put_upperpath;
 742
 743                 err = -EINVAL;
 744                 if (upperpath.mnt != workpath.mnt) {
 745                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 746                         goto out_put_workpath;
 747                 }
 748                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 749                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 750                         goto out_put_workpath;
 751                 }
 752                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 753         }
 754         err = -ENOMEM;
 755         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 756         if (!lowertmp)
 757                 goto out_put_workpath;
 758
 759         err = -EINVAL;
 760         stacklen = ovl_split_lowerdirs(lowertmp);
 761         if (stacklen > OVL_MAX_STACK) {
 762                 pr_err("overlayfs: too many lower directories, limit is %d\n",
 763                        OVL_MAX_STACK);
 764                 goto out_free_lowertmp;
 765         } else if (!ufs->config.upperdir && stacklen == 1) {
 766                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 767                 goto out_free_lowertmp;
 768         }
 769
 770         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 771         if (!stack)
 772                 goto out_free_lowertmp;
 773
 774         lower = lowertmp;
 775         for (numlower = 0; numlower < stacklen; numlower++) {
 776                 err = ovl_lower_dir(lower, &stack[numlower], ufs,
 777                                     &sb->s_stack_depth, &remote);
 778                 if (err)
 779                         goto out_put_lowerpath;
 780
 781                 lower = strchr(lower, '\0') + 1;
 782         }
 783
 784         err = -EINVAL;
 785         sb->s_stack_depth++;
 786         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
 787                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
 788                 goto out_put_lowerpath;
 789         }
 790
 791         if (ufs->config.upperdir) {
 792                 ufs->upper_mnt = clone_private_mount(&upperpath);
 793                 err = PTR_ERR(ufs->upper_mnt);
 794                 if (IS_ERR(ufs->upper_mnt)) {
 795                         pr_err("overlayfs: failed to clone upperpath\n");
 796                         goto out_put_lowerpath;
 797                 }
 798                 /* Don't inherit atime flags */
 799                 ufs->upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
 800
 801                 sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
 802
 803                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
 804                 err = PTR_ERR(ufs->workdir);
 805                 if (IS_ERR(ufs->workdir)) {
 806                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 807                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
 808                         sb->s_flags |= MS_RDONLY;
 809                         ufs->workdir = NULL;
 810                 }
 811
 812                 /*
 813                  * Upper should support d_type, else whiteouts are visible.
 814                  * Given workdir and upper are on same fs, we can do
 815                  * iterate_dir() on workdir. This check requires successful
 816                  * creation of workdir in previous step.
 817                  */
 818                 if (ufs->workdir) {
 819                         err = ovl_check_d_type_supported(&workpath);
 820                         if (err < 0)
 821                                 goto out_put_workdir;
 822
 823                         /*
 824                          * We allowed this configuration and don't want to
 825                          * break users over kernel upgrade. So warn instead
 826                          * of erroring out.
 827                          */
 828                         if (!err)
 829                                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
 830                 }
 831         }
 832
 833         err = -ENOMEM;
 834         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
 835         if (ufs->lower_mnt == NULL)
 836                 goto out_put_workdir;
 837         for (i = 0; i < numlower; i++) {
 838                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
 839
 840                 err = PTR_ERR(mnt);
 841                 if (IS_ERR(mnt)) {
 842                         pr_err("overlayfs: failed to clone lowerpath\n");
 843                         goto out_put_lower_mnt;
 844                 }
 845                 /*
 846                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
 847                  * will fail instead of modifying lower fs.
 848                  */
 849                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
 850
 851                 ufs->lower_mnt[ufs->numlower] = mnt;
 852                 ufs->numlower++;
 853         }
 854
 855         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
 856         if (!ufs->upper_mnt)
 857                 sb->s_flags |= MS_RDONLY;
 858
 859         if (remote)
 860                 sb->s_d_op = &ovl_reval_dentry_operations;
 861         else
 862                 sb->s_d_op = &ovl_dentry_operations;
 863
 864         ufs->creator_cred = prepare_creds();
 865         if (!ufs->creator_cred)
 866                 goto out_put_lower_mnt;
 867
 868         err = -ENOMEM;
 869         oe = ovl_alloc_entry(numlower);
 870         if (!oe)
 871                 goto out_put_cred;
 872
 873         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
 874         sb->s_op = &ovl_super_operations;
 875         sb->s_xattr = ovl_xattr_handlers;
 876         sb->s_fs_info = ufs;
 877         sb->s_flags |= MS_POSIXACL | MS_NOREMOTELOCK;
 878
 879         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
 880         if (!root_dentry)
 881                 goto out_free_oe;
 882
 883         mntput(upperpath.mnt);
 884         for (i = 0; i < numlower; i++)
 885                 mntput(stack[i].mnt);
 886         path_put(&workpath);
 887         kfree(lowertmp);
 888
 889         oe->__upperdentry = upperpath.dentry;
 890         for (i = 0; i < numlower; i++) {
 891                 oe->lowerstack[i].dentry = stack[i].dentry;
 892                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
 893         }
 894         kfree(stack);
 895
 896         root_dentry->d_fsdata = oe;
 897
 898         realinode = d_inode(ovl_dentry_real(root_dentry));
 899         ovl_inode_init(d_inode(root_dentry), realinode, !!upperpath.dentry);
 900         ovl_copyattr(realinode, d_inode(root_dentry));
 901
 902         sb->s_root = root_dentry;
 903
 904         return 0;
 905
 906 out_free_oe:
 907         kfree(oe);
 908 out_put_cred:
 909         put_cred(ufs->creator_cred);
 910 out_put_lower_mnt:
 911         for (i = 0; i < ufs->numlower; i++)
 912                 mntput(ufs->lower_mnt[i]);
 913         kfree(ufs->lower_mnt);
 914 out_put_workdir:
 915         dput(ufs->workdir);
 916         mntput(ufs->upper_mnt);
 917 out_put_lowerpath:
 918         for (i = 0; i < numlower; i++)
 919                 path_put(&stack[i]);
 920         kfree(stack);
 921 out_free_lowertmp:
 922         kfree(lowertmp);
 923 out_put_workpath:
 924         path_put(&workpath);
 925 out_put_upperpath:
 926         path_put(&upperpath);
 927 out_free_config:
 928         kfree(ufs->config.lowerdir);
 929         kfree(ufs->config.upperdir);
 930         kfree(ufs->config.workdir);
 931         kfree(ufs);
 932 out:
 933         return err;
 934 }
 935
 936 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
 937                                 const char *dev_name, void *raw_data)
 938 {
 939         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
 940 }
 941
 942 static struct file_system_type ovl_fs_type = {
 943         .owner          = THIS_MODULE,
 944         .name           = "overlay",
 945         .mount          = ovl_mount,
 946         .kill_sb        = kill_anon_super,
 947 };
 948 MODULE_ALIAS_FS("overlay");
 949
 950 static int __init ovl_init(void)
 951 {
 952         return register_filesystem(&ovl_fs_type);
 953 }
 954
 955 static void __exit ovl_exit(void)
 956 {
 957         unregister_filesystem(&ovl_fs_type);
 958 }
 959
 960 module_init(ovl_init);
 961 module_exit(ovl_exit);