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 <uapi/linux/magic.h>
  11 #include <linux/fs.h>
  12 #include <linux/namei.h>
  13 #include <linux/xattr.h>
  14 #include <linux/mount.h>
  15 #include <linux/parser.h>
  16 #include <linux/module.h>
  17 #include <linux/statfs.h>
  18 #include <linux/seq_file.h>
  19 #include <linux/posix_acl_xattr.h>
  20 #include <linux/exportfs.h>
  21 #include "overlayfs.h"
  22
  23 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  24 MODULE_DESCRIPTION("Overlay filesystem");
  25 MODULE_LICENSE("GPL");
  26
  27
  28 struct ovl_dir_cache;
  29
  30 #define OVL_MAX_STACK 500
  31
  32 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  33 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  34 MODULE_PARM_DESC(ovl_redirect_dir_def,
  35                  "Default to on or off for the redirect_dir feature");
  36
  37 static bool ovl_redirect_always_follow =
  38         IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
  39 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
  40                    bool, 0644);
  41 MODULE_PARM_DESC(ovl_redirect_always_follow,
  42                  "Follow redirects even if redirect_dir feature is turned off");
  43
  44 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
  45 module_param_named(index, ovl_index_def, bool, 0644);
  46 MODULE_PARM_DESC(ovl_index_def,
  47                  "Default to on or off for the inodes index feature");
  48
  49 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
  50 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
  51 MODULE_PARM_DESC(ovl_nfs_export_def,
  52                  "Default to on or off for the NFS export feature");
  53
  54 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
  55 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
  56 MODULE_PARM_DESC(ovl_xino_auto_def,
  57                  "Auto enable xino feature");
  58
  59 static void ovl_entry_stack_free(struct ovl_entry *oe)
  60 {
  61         unsigned int i;
  62
  63         for (i = 0; i < oe->numlower; i++)
  64                 dput(oe->lowerstack[i].dentry);
  65 }
  66
  67 static void ovl_dentry_release(struct dentry *dentry)
  68 {
  69         struct ovl_entry *oe = dentry->d_fsdata;
  70
  71         if (oe) {
  72                 ovl_entry_stack_free(oe);
  73                 kfree_rcu(oe, rcu);
  74         }
  75 }
  76
  77 static struct dentry *ovl_d_real(struct dentry *dentry,
  78                                  const struct inode *inode)
  79 {
  80         struct dentry *real;
  81
  82         /* It's an overlay file */
  83         if (inode && d_inode(dentry) == inode)
  84                 return dentry;
  85
  86         if (!d_is_reg(dentry)) {
  87                 if (!inode || inode == d_inode(dentry))
  88                         return dentry;
  89                 goto bug;
  90         }
  91
  92         real = ovl_dentry_upper(dentry);
  93         if (real && (!inode || inode == d_inode(real)))
  94                 return real;
  95
  96         real = ovl_dentry_lower(dentry);
  97         if (!real)
  98                 goto bug;
  99
 100         /* Handle recursion */
 101         real = d_real(real, inode);
 102
 103         if (!inode || inode == d_inode(real))
 104                 return real;
 105 bug:
 106         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
 107              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 108         return dentry;
 109 }
 110
 111 static int ovl_dentry_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_REVALIDATE) {
 121                         ret = d->d_op->d_revalidate(d, flags);
 122                         if (ret < 0)
 123                                 return ret;
 124                         if (!ret) {
 125                                 if (!(flags & LOOKUP_RCU))
 126                                         d_invalidate(d);
 127                                 return -ESTALE;
 128                         }
 129                 }
 130         }
 131         return 1;
 132 }
 133
 134 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 135 {
 136         struct ovl_entry *oe = dentry->d_fsdata;
 137         unsigned int i;
 138         int ret = 1;
 139
 140         for (i = 0; i < oe->numlower; i++) {
 141                 struct dentry *d = oe->lowerstack[i].dentry;
 142
 143                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 144                         ret = d->d_op->d_weak_revalidate(d, flags);
 145                         if (ret <= 0)
 146                                 break;
 147                 }
 148         }
 149         return ret;
 150 }
 151
 152 static const struct dentry_operations ovl_dentry_operations = {
 153         .d_release = ovl_dentry_release,
 154         .d_real = ovl_d_real,
 155 };
 156
 157 static const struct dentry_operations ovl_reval_dentry_operations = {
 158         .d_release = ovl_dentry_release,
 159         .d_real = ovl_d_real,
 160         .d_revalidate = ovl_dentry_revalidate,
 161         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 162 };
 163
 164 static struct kmem_cache *ovl_inode_cachep;
 165
 166 static struct inode *ovl_alloc_inode(struct super_block *sb)
 167 {
 168         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 169
 170         if (!oi)
 171                 return NULL;
 172
 173         oi->cache = NULL;
 174         oi->redirect = NULL;
 175         oi->version = 0;
 176         oi->flags = 0;
 177         oi->__upperdentry = NULL;
 178         oi->lower = NULL;
 179         mutex_init(&oi->lock);
 180
 181         return &oi->vfs_inode;
 182 }
 183
 184 static void ovl_i_callback(struct rcu_head *head)
 185 {
 186         struct inode *inode = container_of(head, struct inode, i_rcu);
 187
 188         kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
 189 }
 190
 191 static void ovl_destroy_inode(struct inode *inode)
 192 {
 193         struct ovl_inode *oi = OVL_I(inode);
 194
 195         dput(oi->__upperdentry);
 196         iput(oi->lower);
 197         kfree(oi->redirect);
 198         ovl_dir_cache_free(inode);
 199         mutex_destroy(&oi->lock);
 200
 201         call_rcu(&inode->i_rcu, ovl_i_callback);
 202 }
 203
 204 static void ovl_free_fs(struct ovl_fs *ofs)
 205 {
 206         unsigned i;
 207
 208         dput(ofs->indexdir);
 209         dput(ofs->workdir);
 210         if (ofs->workdir_locked)
 211                 ovl_inuse_unlock(ofs->workbasedir);
 212         dput(ofs->workbasedir);
 213         if (ofs->upperdir_locked)
 214                 ovl_inuse_unlock(ofs->upper_mnt->mnt_root);
 215         mntput(ofs->upper_mnt);
 216         for (i = 0; i < ofs->numlower; i++)
 217                 mntput(ofs->lower_layers[i].mnt);
 218         for (i = 0; i < ofs->numlowerfs; i++)
 219                 free_anon_bdev(ofs->lower_fs[i].pseudo_dev);
 220         kfree(ofs->lower_layers);
 221         kfree(ofs->lower_fs);
 222
 223         kfree(ofs->config.lowerdir);
 224         kfree(ofs->config.upperdir);
 225         kfree(ofs->config.workdir);
 226         kfree(ofs->config.redirect_mode);
 227         if (ofs->creator_cred)
 228                 put_cred(ofs->creator_cred);
 229         kfree(ofs);
 230 }
 231
 232 static void ovl_put_super(struct super_block *sb)
 233 {
 234         struct ovl_fs *ofs = sb->s_fs_info;
 235
 236         ovl_free_fs(ofs);
 237 }
 238
 239 /* Sync real dirty inodes in upper filesystem (if it exists) */
 240 static int ovl_sync_fs(struct super_block *sb, int wait)
 241 {
 242         struct ovl_fs *ofs = sb->s_fs_info;
 243         struct super_block *upper_sb;
 244         int ret;
 245
 246         if (!ofs->upper_mnt)
 247                 return 0;
 248
 249         /*
 250          * If this is a sync(2) call or an emergency sync, all the super blocks
 251          * will be iterated, including upper_sb, so no need to do anything.
 252          *
 253          * If this is a syncfs(2) call, then we do need to call
 254          * sync_filesystem() on upper_sb, but enough if we do it when being
 255          * called with wait == 1.
 256          */
 257         if (!wait)
 258                 return 0;
 259
 260         upper_sb = ofs->upper_mnt->mnt_sb;
 261
 262         down_read(&upper_sb->s_umount);
 263         ret = sync_filesystem(upper_sb);
 264         up_read(&upper_sb->s_umount);
 265
 266         return ret;
 267 }
 268
 269 /**
 270  * ovl_statfs
 271  * @sb: The overlayfs super block
 272  * @buf: The struct kstatfs to fill in with stats
 273  *
 274  * Get the filesystem statistics.  As writes always target the upper layer
 275  * filesystem pass the statfs to the upper filesystem (if it exists)
 276  */
 277 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 278 {
 279         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 280         struct dentry *root_dentry = dentry->d_sb->s_root;
 281         struct path path;
 282         int err;
 283
 284         ovl_path_real(root_dentry, &path);
 285
 286         err = vfs_statfs(&path, buf);
 287         if (!err) {
 288                 buf->f_namelen = ofs->namelen;
 289                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 290         }
 291
 292         return err;
 293 }
 294
 295 /* Will this overlay be forced to mount/remount ro? */
 296 static bool ovl_force_readonly(struct ovl_fs *ofs)
 297 {
 298         return (!ofs->upper_mnt || !ofs->workdir);
 299 }
 300
 301 static const char *ovl_redirect_mode_def(void)
 302 {
 303         return ovl_redirect_dir_def ? "on" : "off";
 304 }
 305
 306 enum {
 307         OVL_XINO_OFF,
 308         OVL_XINO_AUTO,
 309         OVL_XINO_ON,
 310 };
 311
 312 static const char * const ovl_xino_str[] = {
 313         "off",
 314         "auto",
 315         "on",
 316 };
 317
 318 static inline int ovl_xino_def(void)
 319 {
 320         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 321 }
 322
 323 /**
 324  * ovl_show_options
 325  *
 326  * Prints the mount options for a given superblock.
 327  * Returns zero; does not fail.
 328  */
 329 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 330 {
 331         struct super_block *sb = dentry->d_sb;
 332         struct ovl_fs *ofs = sb->s_fs_info;
 333
 334         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 335         if (ofs->config.upperdir) {
 336                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 337                 seq_show_option(m, "workdir", ofs->config.workdir);
 338         }
 339         if (ofs->config.default_permissions)
 340                 seq_puts(m, ",default_permissions");
 341         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 342                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 343         if (ofs->config.index != ovl_index_def)
 344                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 345         if (ofs->config.nfs_export != ovl_nfs_export_def)
 346                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 347                                                 "on" : "off");
 348         if (ofs->config.xino != ovl_xino_def())
 349                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 350         return 0;
 351 }
 352
 353 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 354 {
 355         struct ovl_fs *ofs = sb->s_fs_info;
 356
 357         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 358                 return -EROFS;
 359
 360         return 0;
 361 }
 362
 363 static const struct super_operations ovl_super_operations = {
 364         .alloc_inode    = ovl_alloc_inode,
 365         .destroy_inode  = ovl_destroy_inode,
 366         .drop_inode     = generic_delete_inode,
 367         .put_super      = ovl_put_super,
 368         .sync_fs        = ovl_sync_fs,
 369         .statfs         = ovl_statfs,
 370         .show_options   = ovl_show_options,
 371         .remount_fs     = ovl_remount,
 372 };
 373
 374 enum {
 375         OPT_LOWERDIR,
 376         OPT_UPPERDIR,
 377         OPT_WORKDIR,
 378         OPT_DEFAULT_PERMISSIONS,
 379         OPT_REDIRECT_DIR,
 380         OPT_INDEX_ON,
 381         OPT_INDEX_OFF,
 382         OPT_NFS_EXPORT_ON,
 383         OPT_NFS_EXPORT_OFF,
 384         OPT_XINO_ON,
 385         OPT_XINO_OFF,
 386         OPT_XINO_AUTO,
 387         OPT_ERR,
 388 };
 389
 390 static const match_table_t ovl_tokens = {
 391         {OPT_LOWERDIR,                  "lowerdir=%s"},
 392         {OPT_UPPERDIR,                  "upperdir=%s"},
 393         {OPT_WORKDIR,                   "workdir=%s"},
 394         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 395         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 396         {OPT_INDEX_ON,                  "index=on"},
 397         {OPT_INDEX_OFF,                 "index=off"},
 398         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 399         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 400         {OPT_XINO_ON,                   "xino=on"},
 401         {OPT_XINO_OFF,                  "xino=off"},
 402         {OPT_XINO_AUTO,                 "xino=auto"},
 403         {OPT_ERR,                       NULL}
 404 };
 405
 406 static char *ovl_next_opt(char **s)
 407 {
 408         char *sbegin = *s;
 409         char *p;
 410
 411         if (sbegin == NULL)
 412                 return NULL;
 413
 414         for (p = sbegin; *p; p++) {
 415                 if (*p == '\\') {
 416                         p++;
 417                         if (!*p)
 418                                 break;
 419                 } else if (*p == ',') {
 420                         *p = '\0';
 421                         *s = p + 1;
 422                         return sbegin;
 423                 }
 424         }
 425         *s = NULL;
 426         return sbegin;
 427 }
 428
 429 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 430 {
 431         if (strcmp(mode, "on") == 0) {
 432                 config->redirect_dir = true;
 433                 /*
 434                  * Does not make sense to have redirect creation without
 435                  * redirect following.
 436                  */
 437                 config->redirect_follow = true;
 438         } else if (strcmp(mode, "follow") == 0) {
 439                 config->redirect_follow = true;
 440         } else if (strcmp(mode, "off") == 0) {
 441                 if (ovl_redirect_always_follow)
 442                         config->redirect_follow = true;
 443         } else if (strcmp(mode, "nofollow") != 0) {
 444                 pr_err("overlayfs: bad mount option \"redirect_dir=%s\"\n",
 445                        mode);
 446                 return -EINVAL;
 447         }
 448
 449         return 0;
 450 }
 451
 452 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 453 {
 454         char *p;
 455
 456         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 457         if (!config->redirect_mode)
 458                 return -ENOMEM;
 459
 460         while ((p = ovl_next_opt(&opt)) != NULL) {
 461                 int token;
 462                 substring_t args[MAX_OPT_ARGS];
 463
 464                 if (!*p)
 465                         continue;
 466
 467                 token = match_token(p, ovl_tokens, args);
 468                 switch (token) {
 469                 case OPT_UPPERDIR:
 470                         kfree(config->upperdir);
 471                         config->upperdir = match_strdup(&args[0]);
 472                         if (!config->upperdir)
 473                                 return -ENOMEM;
 474                         break;
 475
 476                 case OPT_LOWERDIR:
 477                         kfree(config->lowerdir);
 478                         config->lowerdir = match_strdup(&args[0]);
 479                         if (!config->lowerdir)
 480                                 return -ENOMEM;
 481                         break;
 482
 483                 case OPT_WORKDIR:
 484                         kfree(config->workdir);
 485                         config->workdir = match_strdup(&args[0]);
 486                         if (!config->workdir)
 487                                 return -ENOMEM;
 488                         break;
 489
 490                 case OPT_DEFAULT_PERMISSIONS:
 491                         config->default_permissions = true;
 492                         break;
 493
 494                 case OPT_REDIRECT_DIR:
 495                         kfree(config->redirect_mode);
 496                         config->redirect_mode = match_strdup(&args[0]);
 497                         if (!config->redirect_mode)
 498                                 return -ENOMEM;
 499                         break;
 500
 501                 case OPT_INDEX_ON:
 502                         config->index = true;
 503                         break;
 504
 505                 case OPT_INDEX_OFF:
 506                         config->index = false;
 507                         break;
 508
 509                 case OPT_NFS_EXPORT_ON:
 510                         config->nfs_export = true;
 511                         break;
 512
 513                 case OPT_NFS_EXPORT_OFF:
 514                         config->nfs_export = false;
 515                         break;
 516
 517                 case OPT_XINO_ON:
 518                         config->xino = OVL_XINO_ON;
 519                         break;
 520
 521                 case OPT_XINO_OFF:
 522                         config->xino = OVL_XINO_OFF;
 523                         break;
 524
 525                 case OPT_XINO_AUTO:
 526                         config->xino = OVL_XINO_AUTO;
 527                         break;
 528
 529                 default:
 530                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 531                         return -EINVAL;
 532                 }
 533         }
 534
 535         /* Workdir is useless in non-upper mount */
 536         if (!config->upperdir && config->workdir) {
 537                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 538                         config->workdir);
 539                 kfree(config->workdir);
 540                 config->workdir = NULL;
 541         }
 542
 543         return ovl_parse_redirect_mode(config, config->redirect_mode);
 544 }
 545
 546 #define OVL_WORKDIR_NAME "work"
 547 #define OVL_INDEXDIR_NAME "index"
 548
 549 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 550                                          const char *name, bool persist)
 551 {
 552         struct inode *dir =  ofs->workbasedir->d_inode;
 553         struct vfsmount *mnt = ofs->upper_mnt;
 554         struct dentry *work;
 555         int err;
 556         bool retried = false;
 557         bool locked = false;
 558
 559         inode_lock_nested(dir, I_MUTEX_PARENT);
 560         locked = true;
 561
 562 retry:
 563         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
 564
 565         if (!IS_ERR(work)) {
 566                 struct iattr attr = {
 567                         .ia_valid = ATTR_MODE,
 568                         .ia_mode = S_IFDIR | 0,
 569                 };
 570
 571                 if (work->d_inode) {
 572                         err = -EEXIST;
 573                         if (retried)
 574                                 goto out_dput;
 575
 576                         if (persist)
 577                                 goto out_unlock;
 578
 579                         retried = true;
 580                         ovl_workdir_cleanup(dir, mnt, work, 0);
 581                         dput(work);
 582                         goto retry;
 583                 }
 584
 585                 work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
 586                 err = PTR_ERR(work);
 587                 if (IS_ERR(work))
 588                         goto out_err;
 589
 590                 /*
 591                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 592                  *
 593                  * a) success (there was a POSIX ACL xattr and was removed)
 594                  * b) -ENODATA (there was no POSIX ACL xattr)
 595                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 596                  *
 597                  * There are various other error values that could effectively
 598                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 599                  * if the xattr name is too long), but the set of filesystems
 600                  * allowed as upper are limited to "normal" ones, where checking
 601                  * for the above two errors is sufficient.
 602                  */
 603                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 604                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 605                         goto out_dput;
 606
 607                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 608                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 609                         goto out_dput;
 610
 611                 /* Clear any inherited mode bits */
 612                 inode_lock(work->d_inode);
 613                 err = notify_change(work, &attr, NULL);
 614                 inode_unlock(work->d_inode);
 615                 if (err)
 616                         goto out_dput;
 617         } else {
 618                 err = PTR_ERR(work);
 619                 goto out_err;
 620         }
 621 out_unlock:
 622         if (locked)
 623                 inode_unlock(dir);
 624
 625         return work;
 626
 627 out_dput:
 628         dput(work);
 629 out_err:
 630         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 631                 ofs->config.workdir, name, -err);
 632         work = NULL;
 633         goto out_unlock;
 634 }
 635
 636 static void ovl_unescape(char *s)
 637 {
 638         char *d = s;
 639
 640         for (;; s++, d++) {
 641                 if (*s == '\\')
 642                         s++;
 643                 *d = *s;
 644                 if (!*s)
 645                         break;
 646         }
 647 }
 648
 649 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 650 {
 651         int err = -EINVAL;
 652
 653         if (!*name) {
 654                 pr_err("overlayfs: empty lowerdir\n");
 655                 goto out;
 656         }
 657         err = kern_path(name, LOOKUP_FOLLOW, path);
 658         if (err) {
 659                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 660                 goto out;
 661         }
 662         err = -EINVAL;
 663         if (ovl_dentry_weird(path->dentry)) {
 664                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 665                 goto out_put;
 666         }
 667         if (!d_is_dir(path->dentry)) {
 668                 pr_err("overlayfs: '%s' not a directory\n", name);
 669                 goto out_put;
 670         }
 671         return 0;
 672
 673 out_put:
 674         path_put_init(path);
 675 out:
 676         return err;
 677 }
 678
 679 static int ovl_mount_dir(const char *name, struct path *path)
 680 {
 681         int err = -ENOMEM;
 682         char *tmp = kstrdup(name, GFP_KERNEL);
 683
 684         if (tmp) {
 685                 ovl_unescape(tmp);
 686                 err = ovl_mount_dir_noesc(tmp, path);
 687
 688                 if (!err)
 689                         if (ovl_dentry_remote(path->dentry)) {
 690                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 691                                        tmp);
 692                                 path_put_init(path);
 693                                 err = -EINVAL;
 694                         }
 695                 kfree(tmp);
 696         }
 697         return err;
 698 }
 699
 700 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 701                              const char *name)
 702 {
 703         struct kstatfs statfs;
 704         int err = vfs_statfs(path, &statfs);
 705
 706         if (err)
 707                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 708         else
 709                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 710
 711         return err;
 712 }
 713
 714 static int ovl_lower_dir(const char *name, struct path *path,
 715                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 716 {
 717         int fh_type;
 718         int err;
 719
 720         err = ovl_mount_dir_noesc(name, path);
 721         if (err)
 722                 goto out;
 723
 724         err = ovl_check_namelen(path, ofs, name);
 725         if (err)
 726                 goto out_put;
 727
 728         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 729
 730         if (ovl_dentry_remote(path->dentry))
 731                 *remote = true;
 732
 733         /*
 734          * The inodes index feature and NFS export need to encode and decode
 735          * file handles, so they require that all layers support them.
 736          */
 737         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 738         if ((ofs->config.nfs_export ||
 739              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 740                 ofs->config.index = false;
 741                 ofs->config.nfs_export = false;
 742                 pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 743                         name);
 744         }
 745
 746         /* Check if lower fs has 32bit inode numbers */
 747         if (fh_type != FILEID_INO32_GEN)
 748                 ofs->xino_bits = 0;
 749
 750         return 0;
 751
 752 out_put:
 753         path_put_init(path);
 754 out:
 755         return err;
 756 }
 757
 758 /* Workdir should not be subdir of upperdir and vice versa */
 759 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 760 {
 761         bool ok = false;
 762
 763         if (workdir != upperdir) {
 764                 ok = (lock_rename(workdir, upperdir) == NULL);
 765                 unlock_rename(workdir, upperdir);
 766         }
 767         return ok;
 768 }
 769
 770 static unsigned int ovl_split_lowerdirs(char *str)
 771 {
 772         unsigned int ctr = 1;
 773         char *s, *d;
 774
 775         for (s = d = str;; s++, d++) {
 776                 if (*s == '\\') {
 777                         s++;
 778                 } else if (*s == ':') {
 779                         *d = '\0';
 780                         ctr++;
 781                         continue;
 782                 }
 783                 *d = *s;
 784                 if (!*s)
 785                         break;
 786         }
 787         return ctr;
 788 }
 789
 790 static int __maybe_unused
 791 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 792                         struct dentry *dentry, struct inode *inode,
 793                         const char *name, void *buffer, size_t size)
 794 {
 795         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 796 }
 797
 798 static int __maybe_unused
 799 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 800                         struct dentry *dentry, struct inode *inode,
 801                         const char *name, const void *value,
 802                         size_t size, int flags)
 803 {
 804         struct dentry *workdir = ovl_workdir(dentry);
 805         struct inode *realinode = ovl_inode_real(inode);
 806         struct posix_acl *acl = NULL;
 807         int err;
 808
 809         /* Check that everything is OK before copy-up */
 810         if (value) {
 811                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 812                 if (IS_ERR(acl))
 813                         return PTR_ERR(acl);
 814         }
 815         err = -EOPNOTSUPP;
 816         if (!IS_POSIXACL(d_inode(workdir)))
 817                 goto out_acl_release;
 818         if (!realinode->i_op->set_acl)
 819                 goto out_acl_release;
 820         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 821                 err = acl ? -EACCES : 0;
 822                 goto out_acl_release;
 823         }
 824         err = -EPERM;
 825         if (!inode_owner_or_capable(inode))
 826                 goto out_acl_release;
 827
 828         posix_acl_release(acl);
 829
 830         /*
 831          * Check if sgid bit needs to be cleared (actual setacl operation will
 832          * be done with mounter's capabilities and so that won't do it for us).
 833          */
 834         if (unlikely(inode->i_mode & S_ISGID) &&
 835             handler->flags == ACL_TYPE_ACCESS &&
 836             !in_group_p(inode->i_gid) &&
 837             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 838                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 839
 840                 err = ovl_setattr(dentry, &iattr);
 841                 if (err)
 842                         return err;
 843         }
 844
 845         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
 846         if (!err)
 847                 ovl_copyattr(ovl_inode_real(inode), inode);
 848
 849         return err;
 850
 851 out_acl_release:
 852         posix_acl_release(acl);
 853         return err;
 854 }
 855
 856 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 857                              struct dentry *dentry, struct inode *inode,
 858                              const char *name, void *buffer, size_t size)
 859 {
 860         return -EOPNOTSUPP;
 861 }
 862
 863 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 864                              struct dentry *dentry, struct inode *inode,
 865                              const char *name, const void *value,
 866                              size_t size, int flags)
 867 {
 868         return -EOPNOTSUPP;
 869 }
 870
 871 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 872                                struct dentry *dentry, struct inode *inode,
 873                                const char *name, void *buffer, size_t size)
 874 {
 875         return ovl_xattr_get(dentry, inode, name, buffer, size);
 876 }
 877
 878 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 879                                struct dentry *dentry, struct inode *inode,
 880                                const char *name, const void *value,
 881                                size_t size, int flags)
 882 {
 883         return ovl_xattr_set(dentry, inode, name, value, size, flags);
 884 }
 885
 886 static const struct xattr_handler __maybe_unused
 887 ovl_posix_acl_access_xattr_handler = {
 888         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 889         .flags = ACL_TYPE_ACCESS,
 890         .get = ovl_posix_acl_xattr_get,
 891         .set = ovl_posix_acl_xattr_set,
 892 };
 893
 894 static const struct xattr_handler __maybe_unused
 895 ovl_posix_acl_default_xattr_handler = {
 896         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 897         .flags = ACL_TYPE_DEFAULT,
 898         .get = ovl_posix_acl_xattr_get,
 899         .set = ovl_posix_acl_xattr_set,
 900 };
 901
 902 static const struct xattr_handler ovl_own_xattr_handler = {
 903         .prefix = OVL_XATTR_PREFIX,
 904         .get = ovl_own_xattr_get,
 905         .set = ovl_own_xattr_set,
 906 };
 907
 908 static const struct xattr_handler ovl_other_xattr_handler = {
 909         .prefix = "", /* catch all */
 910         .get = ovl_other_xattr_get,
 911         .set = ovl_other_xattr_set,
 912 };
 913
 914 static const struct xattr_handler *ovl_xattr_handlers[] = {
 915 #ifdef CONFIG_FS_POSIX_ACL
 916         &ovl_posix_acl_access_xattr_handler,
 917         &ovl_posix_acl_default_xattr_handler,
 918 #endif
 919         &ovl_own_xattr_handler,
 920         &ovl_other_xattr_handler,
 921         NULL
 922 };
 923
 924 static int ovl_get_upper(struct ovl_fs *ofs, struct path *upperpath)
 925 {
 926         struct vfsmount *upper_mnt;
 927         int err;
 928
 929         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
 930         if (err)
 931                 goto out;
 932
 933         /* Upper fs should not be r/o */
 934         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
 935                 pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 936                 err = -EINVAL;
 937                 goto out;
 938         }
 939
 940         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
 941         if (err)
 942                 goto out;
 943
 944         err = -EBUSY;
 945         if (ovl_inuse_trylock(upperpath->dentry)) {
 946                 ofs->upperdir_locked = true;
 947         } else if (ofs->config.index) {
 948                 pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n");
 949                 goto out;
 950         } else {
 951                 pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
 952         }
 953
 954         upper_mnt = clone_private_mount(upperpath);
 955         err = PTR_ERR(upper_mnt);
 956         if (IS_ERR(upper_mnt)) {
 957                 pr_err("overlayfs: failed to clone upperpath\n");
 958                 goto out;
 959         }
 960
 961         /* Don't inherit atime flags */
 962         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
 963         ofs->upper_mnt = upper_mnt;
 964         err = 0;
 965 out:
 966         return err;
 967 }
 968
 969 static int ovl_make_workdir(struct ovl_fs *ofs, struct path *workpath)
 970 {
 971         struct vfsmount *mnt = ofs->upper_mnt;
 972         struct dentry *temp;
 973         int fh_type;
 974         int err;
 975
 976         err = mnt_want_write(mnt);
 977         if (err)
 978                 return err;
 979
 980         ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
 981         if (!ofs->workdir)
 982                 goto out;
 983
 984         /*
 985          * Upper should support d_type, else whiteouts are visible.  Given
 986          * workdir and upper are on same fs, we can do iterate_dir() on
 987          * workdir. This check requires successful creation of workdir in
 988          * previous step.
 989          */
 990         err = ovl_check_d_type_supported(workpath);
 991         if (err < 0)
 992                 goto out;
 993
 994         /*
 995          * We allowed this configuration and don't want to break users over
 996          * kernel upgrade. So warn instead of erroring out.
 997          */
 998         if (!err)
 999                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
1000
1001         /* Check if upper/work fs supports O_TMPFILE */
1002         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1003         ofs->tmpfile = !IS_ERR(temp);
1004         if (ofs->tmpfile)
1005                 dput(temp);
1006         else
1007                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
1008
1009         /*
1010          * Check if upper/work fs supports trusted.overlay.* xattr
1011          */
1012         err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
1013         if (err) {
1014                 ofs->noxattr = true;
1015                 ofs->config.index = false;
1016                 pr_warn("overlayfs: upper fs does not support xattr, falling back to index=off.\n");
1017                 err = 0;
1018         } else {
1019                 vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
1020         }
1021
1022         /* Check if upper/work fs supports file handles */
1023         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1024         if (ofs->config.index && !fh_type) {
1025                 ofs->config.index = false;
1026                 pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1027         }
1028
1029         /* Check if upper fs has 32bit inode numbers */
1030         if (fh_type != FILEID_INO32_GEN)
1031                 ofs->xino_bits = 0;
1032
1033         /* NFS export of r/w mount depends on index */
1034         if (ofs->config.nfs_export && !ofs->config.index) {
1035                 pr_warn("overlayfs: NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1036                 ofs->config.nfs_export = false;
1037         }
1038
1039 out:
1040         mnt_drop_write(mnt);
1041         return err;
1042 }
1043
1044 static int ovl_get_workdir(struct ovl_fs *ofs, struct path *upperpath)
1045 {
1046         int err;
1047         struct path workpath = { };
1048
1049         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1050         if (err)
1051                 goto out;
1052
1053         err = -EINVAL;
1054         if (upperpath->mnt != workpath.mnt) {
1055                 pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1056                 goto out;
1057         }
1058         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1059                 pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1060                 goto out;
1061         }
1062
1063         err = -EBUSY;
1064         if (ovl_inuse_trylock(workpath.dentry)) {
1065                 ofs->workdir_locked = true;
1066         } else if (ofs->config.index) {
1067                 pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n");
1068                 goto out;
1069         } else {
1070                 pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
1071         }
1072
1073         ofs->workbasedir = dget(workpath.dentry);
1074         err = ovl_make_workdir(ofs, &workpath);
1075         if (err)
1076                 goto out;
1077
1078         err = 0;
1079 out:
1080         path_put(&workpath);
1081
1082         return err;
1083 }
1084
1085 static int ovl_get_indexdir(struct ovl_fs *ofs, struct ovl_entry *oe,
1086                             struct path *upperpath)
1087 {
1088         struct vfsmount *mnt = ofs->upper_mnt;
1089         int err;
1090
1091         err = mnt_want_write(mnt);
1092         if (err)
1093                 return err;
1094
1095         /* Verify lower root is upper root origin */
1096         err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
1097                                 true);
1098         if (err) {
1099                 pr_err("overlayfs: failed to verify upper root origin\n");
1100                 goto out;
1101         }
1102
1103         ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1104         if (ofs->indexdir) {
1105                 /*
1106                  * Verify upper root is exclusively associated with index dir.
1107                  * Older kernels stored upper fh in "trusted.overlay.origin"
1108                  * xattr. If that xattr exists, verify that it is a match to
1109                  * upper dir file handle. In any case, verify or set xattr
1110                  * "trusted.overlay.upper" to indicate that index may have
1111                  * directory entries.
1112                  */
1113                 if (ovl_check_origin_xattr(ofs->indexdir)) {
1114                         err = ovl_verify_set_fh(ofs->indexdir, OVL_XATTR_ORIGIN,
1115                                                 upperpath->dentry, true, false);
1116                         if (err)
1117                                 pr_err("overlayfs: failed to verify index dir 'origin' xattr\n");
1118                 }
1119                 err = ovl_verify_upper(ofs->indexdir, upperpath->dentry, true);
1120                 if (err)
1121                         pr_err("overlayfs: failed to verify index dir 'upper' xattr\n");
1122
1123                 /* Cleanup bad/stale/orphan index entries */
1124                 if (!err)
1125                         err = ovl_indexdir_cleanup(ofs);
1126         }
1127         if (err || !ofs->indexdir)
1128                 pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1129
1130 out:
1131         mnt_drop_write(mnt);
1132         return err;
1133 }
1134
1135 /* Get a unique fsid for the layer */
1136 static int ovl_get_fsid(struct ovl_fs *ofs, struct super_block *sb)
1137 {
1138         unsigned int i;
1139         dev_t dev;
1140         int err;
1141
1142         /* fsid 0 is reserved for upper fs even with non upper overlay */
1143         if (ofs->upper_mnt && ofs->upper_mnt->mnt_sb == sb)
1144                 return 0;
1145
1146         for (i = 0; i < ofs->numlowerfs; i++) {
1147                 if (ofs->lower_fs[i].sb == sb)
1148                         return i + 1;
1149         }
1150
1151         err = get_anon_bdev(&dev);
1152         if (err) {
1153                 pr_err("overlayfs: failed to get anonymous bdev for lowerpath\n");
1154                 return err;
1155         }
1156
1157         ofs->lower_fs[ofs->numlowerfs].sb = sb;
1158         ofs->lower_fs[ofs->numlowerfs].pseudo_dev = dev;
1159         ofs->numlowerfs++;
1160
1161         return ofs->numlowerfs;
1162 }
1163
1164 static int ovl_get_lower_layers(struct ovl_fs *ofs, struct path *stack,
1165                                 unsigned int numlower)
1166 {
1167         int err;
1168         unsigned int i;
1169
1170         err = -ENOMEM;
1171         ofs->lower_layers = kcalloc(numlower, sizeof(struct ovl_layer),
1172                                     GFP_KERNEL);
1173         if (ofs->lower_layers == NULL)
1174                 goto out;
1175
1176         ofs->lower_fs = kcalloc(numlower, sizeof(struct ovl_sb),
1177                                 GFP_KERNEL);
1178         if (ofs->lower_fs == NULL)
1179                 goto out;
1180
1181         for (i = 0; i < numlower; i++) {
1182                 struct vfsmount *mnt;
1183                 int fsid;
1184
1185                 err = fsid = ovl_get_fsid(ofs, stack[i].mnt->mnt_sb);
1186                 if (err < 0)
1187                         goto out;
1188
1189                 mnt = clone_private_mount(&stack[i]);
1190                 err = PTR_ERR(mnt);
1191                 if (IS_ERR(mnt)) {
1192                         pr_err("overlayfs: failed to clone lowerpath\n");
1193                         goto out;
1194                 }
1195
1196                 /*
1197                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1198                  * will fail instead of modifying lower fs.
1199                  */
1200                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1201
1202                 ofs->lower_layers[ofs->numlower].mnt = mnt;
1203                 ofs->lower_layers[ofs->numlower].idx = i + 1;
1204                 ofs->lower_layers[ofs->numlower].fsid = fsid;
1205                 if (fsid) {
1206                         ofs->lower_layers[ofs->numlower].fs =
1207                                 &ofs->lower_fs[fsid - 1];
1208                 }
1209                 ofs->numlower++;
1210         }
1211
1212         /*
1213          * When all layers on same fs, overlay can use real inode numbers.
1214          * With mount option "xino=on", mounter declares that there are enough
1215          * free high bits in underlying fs to hold the unique fsid.
1216          * If overlayfs does encounter underlying inodes using the high xino
1217          * bits reserved for fsid, it emits a warning and uses the original
1218          * inode number.
1219          */
1220         if (!ofs->numlowerfs || (ofs->numlowerfs == 1 && !ofs->upper_mnt)) {
1221                 ofs->xino_bits = 0;
1222                 ofs->config.xino = OVL_XINO_OFF;
1223         } else if (ofs->config.xino == OVL_XINO_ON && !ofs->xino_bits) {
1224                 /*
1225                  * This is a roundup of number of bits needed for numlowerfs+1
1226                  * (i.e. ilog2(numlowerfs+1 - 1) + 1). fsid 0 is reserved for
1227                  * upper fs even with non upper overlay.
1228                  */
1229                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 31);
1230                 ofs->xino_bits = ilog2(ofs->numlowerfs) + 1;
1231         }
1232
1233         if (ofs->xino_bits) {
1234                 pr_info("overlayfs: \"xino\" feature enabled using %d upper inode bits.\n",
1235                         ofs->xino_bits);
1236         }
1237
1238         err = 0;
1239 out:
1240         return err;
1241 }
1242
1243 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1244                                             struct ovl_fs *ofs)
1245 {
1246         int err;
1247         char *lowertmp, *lower;
1248         struct path *stack = NULL;
1249         unsigned int stacklen, numlower = 0, i;
1250         bool remote = false;
1251         struct ovl_entry *oe;
1252
1253         err = -ENOMEM;
1254         lowertmp = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1255         if (!lowertmp)
1256                 goto out_err;
1257
1258         err = -EINVAL;
1259         stacklen = ovl_split_lowerdirs(lowertmp);
1260         if (stacklen > OVL_MAX_STACK) {
1261                 pr_err("overlayfs: too many lower directories, limit is %d\n",
1262                        OVL_MAX_STACK);
1263                 goto out_err;
1264         } else if (!ofs->config.upperdir && stacklen == 1) {
1265                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1266                 goto out_err;
1267         } else if (!ofs->config.upperdir && ofs->config.nfs_export &&
1268                    ofs->config.redirect_follow) {
1269                 pr_warn("overlayfs: NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
1270                 ofs->config.nfs_export = false;
1271         }
1272
1273         err = -ENOMEM;
1274         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1275         if (!stack)
1276                 goto out_err;
1277
1278         err = -EINVAL;
1279         lower = lowertmp;
1280         for (numlower = 0; numlower < stacklen; numlower++) {
1281                 err = ovl_lower_dir(lower, &stack[numlower], ofs,
1282                                     &sb->s_stack_depth, &remote);
1283                 if (err)
1284                         goto out_err;
1285
1286                 lower = strchr(lower, '\0') + 1;
1287         }
1288
1289         err = -EINVAL;
1290         sb->s_stack_depth++;
1291         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1292                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1293                 goto out_err;
1294         }
1295
1296         err = ovl_get_lower_layers(ofs, stack, numlower);
1297         if (err)
1298                 goto out_err;
1299
1300         err = -ENOMEM;
1301         oe = ovl_alloc_entry(numlower);
1302         if (!oe)
1303                 goto out_err;
1304
1305         for (i = 0; i < numlower; i++) {
1306                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1307                 oe->lowerstack[i].layer = &ofs->lower_layers[i];
1308         }
1309
1310         if (remote)
1311                 sb->s_d_op = &ovl_reval_dentry_operations;
1312         else
1313                 sb->s_d_op = &ovl_dentry_operations;
1314
1315 out:
1316         for (i = 0; i < numlower; i++)
1317                 path_put(&stack[i]);
1318         kfree(stack);
1319         kfree(lowertmp);
1320
1321         return oe;
1322
1323 out_err:
1324         oe = ERR_PTR(err);
1325         goto out;
1326 }
1327
1328 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1329 {
1330         struct path upperpath = { };
1331         struct dentry *root_dentry;
1332         struct ovl_entry *oe;
1333         struct ovl_fs *ofs;
1334         struct cred *cred;
1335         int err;
1336
1337         err = -ENOMEM;
1338         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1339         if (!ofs)
1340                 goto out;
1341
1342         ofs->creator_cred = cred = prepare_creds();
1343         if (!cred)
1344                 goto out_err;
1345
1346         ofs->config.index = ovl_index_def;
1347         ofs->config.nfs_export = ovl_nfs_export_def;
1348         ofs->config.xino = ovl_xino_def();
1349         err = ovl_parse_opt((char *) data, &ofs->config);
1350         if (err)
1351                 goto out_err;
1352
1353         err = -EINVAL;
1354         if (!ofs->config.lowerdir) {
1355                 if (!silent)
1356                         pr_err("overlayfs: missing 'lowerdir'\n");
1357                 goto out_err;
1358         }
1359
1360         sb->s_stack_depth = 0;
1361         sb->s_maxbytes = MAX_LFS_FILESIZE;
1362         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1363         if (ofs->config.xino != OVL_XINO_OFF)
1364                 ofs->xino_bits = BITS_PER_LONG - 32;
1365
1366         if (ofs->config.upperdir) {
1367                 if (!ofs->config.workdir) {
1368                         pr_err("overlayfs: missing 'workdir'\n");
1369                         goto out_err;
1370                 }
1371
1372                 err = ovl_get_upper(ofs, &upperpath);
1373                 if (err)
1374                         goto out_err;
1375
1376                 err = ovl_get_workdir(ofs, &upperpath);
1377                 if (err)
1378                         goto out_err;
1379
1380                 if (!ofs->workdir)
1381                         sb->s_flags |= SB_RDONLY;
1382
1383                 sb->s_stack_depth = ofs->upper_mnt->mnt_sb->s_stack_depth;
1384                 sb->s_time_gran = ofs->upper_mnt->mnt_sb->s_time_gran;
1385
1386         }
1387         oe = ovl_get_lowerstack(sb, ofs);
1388         err = PTR_ERR(oe);
1389         if (IS_ERR(oe))
1390                 goto out_err;
1391
1392         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1393         if (!ofs->upper_mnt)
1394                 sb->s_flags |= SB_RDONLY;
1395
1396         if (!(ovl_force_readonly(ofs)) && ofs->config.index) {
1397                 err = ovl_get_indexdir(ofs, oe, &upperpath);
1398                 if (err)
1399                         goto out_free_oe;
1400
1401                 /* Force r/o mount with no index dir */
1402                 if (!ofs->indexdir) {
1403                         dput(ofs->workdir);
1404                         ofs->workdir = NULL;
1405                         sb->s_flags |= SB_RDONLY;
1406                 }
1407
1408         }
1409
1410         /* Show index=off in /proc/mounts for forced r/o mount */
1411         if (!ofs->indexdir) {
1412                 ofs->config.index = false;
1413                 if (ofs->upper_mnt && ofs->config.nfs_export) {
1414                         pr_warn("overlayfs: NFS export requires an index dir, falling back to nfs_export=off.\n");
1415                         ofs->config.nfs_export = false;
1416                 }
1417         }
1418
1419         if (ofs->config.nfs_export)
1420                 sb->s_export_op = &ovl_export_operations;
1421
1422         /* Never override disk quota limits or use reserved space */
1423         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1424
1425         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1426         sb->s_op = &ovl_super_operations;
1427         sb->s_xattr = ovl_xattr_handlers;
1428         sb->s_fs_info = ofs;
1429         sb->s_flags |= SB_POSIXACL;
1430
1431         err = -ENOMEM;
1432         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1433         if (!root_dentry)
1434                 goto out_free_oe;
1435
1436         root_dentry->d_fsdata = oe;
1437
1438         mntput(upperpath.mnt);
1439         if (upperpath.dentry) {
1440                 ovl_dentry_set_upper_alias(root_dentry);
1441                 if (ovl_is_impuredir(upperpath.dentry))
1442                         ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1443         }
1444
1445         /* Root is always merge -> can have whiteouts */
1446         ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1447         ovl_dentry_set_flag(OVL_E_CONNECTED, root_dentry);
1448         ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1449                        ovl_dentry_lower(root_dentry));
1450
1451         sb->s_root = root_dentry;
1452
1453         return 0;
1454
1455 out_free_oe:
1456         ovl_entry_stack_free(oe);
1457         kfree(oe);
1458 out_err:
1459         path_put(&upperpath);
1460         ovl_free_fs(ofs);
1461 out:
1462         return err;
1463 }
1464
1465 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1466                                 const char *dev_name, void *raw_data)
1467 {
1468         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1469 }
1470
1471 static struct file_system_type ovl_fs_type = {
1472         .owner          = THIS_MODULE,
1473         .name           = "overlay",
1474         .mount          = ovl_mount,
1475         .kill_sb        = kill_anon_super,
1476 };
1477 MODULE_ALIAS_FS("overlay");
1478
1479 static void ovl_inode_init_once(void *foo)
1480 {
1481         struct ovl_inode *oi = foo;
1482
1483         inode_init_once(&oi->vfs_inode);
1484 }
1485
1486 static int __init ovl_init(void)
1487 {
1488         int err;
1489
1490         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1491                                              sizeof(struct ovl_inode), 0,
1492                                              (SLAB_RECLAIM_ACCOUNT|
1493                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1494                                              ovl_inode_init_once);
1495         if (ovl_inode_cachep == NULL)
1496                 return -ENOMEM;
1497
1498         err = register_filesystem(&ovl_fs_type);
1499         if (err)
1500                 kmem_cache_destroy(ovl_inode_cachep);
1501
1502         return err;
1503 }
1504
1505 static void __exit ovl_exit(void)
1506 {
1507         unregister_filesystem(&ovl_fs_type);
1508
1509         /*
1510          * Make sure all delayed rcu free inodes are flushed before we
1511          * destroy cache.
1512          */
1513         rcu_barrier();
1514         kmem_cache_destroy(ovl_inode_cachep);
1515
1516 }
1517
1518 module_init(ovl_init);
1519 module_exit(ovl_exit);