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 bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
  68 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
  69 MODULE_PARM_DESC(ovl_metacopy_def,
  70                  "Default to on or off for the metadata only copy up feature");
  71
  72 static void ovl_dentry_release(struct dentry *dentry)
  73 {
  74         struct ovl_entry *oe = dentry->d_fsdata;
  75
  76         if (oe) {
  77                 ovl_entry_stack_free(oe);
  78                 kfree_rcu(oe, rcu);
  79         }
  80 }
  81
  82 static struct dentry *ovl_d_real(struct dentry *dentry,
  83                                  const struct inode *inode)
  84 {
  85         struct dentry *real;
  86
  87         /* It's an overlay file */
  88         if (inode && d_inode(dentry) == inode)
  89                 return dentry;
  90
  91         if (!d_is_reg(dentry)) {
  92                 if (!inode || inode == d_inode(dentry))
  93                         return dentry;
  94                 goto bug;
  95         }
  96
  97         real = ovl_dentry_upper(dentry);
  98         if (real && (!inode || inode == d_inode(real)))
  99                 return real;
 100
 101         real = ovl_dentry_lower(dentry);
 102         if (!real)
 103                 goto bug;
 104
 105         /* Handle recursion */
 106         real = d_real(real, inode);
 107
 108         if (!inode || inode == d_inode(real))
 109                 return real;
 110 bug:
 111         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
 112              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 113         return dentry;
 114 }
 115
 116 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 117 {
 118         struct ovl_entry *oe = dentry->d_fsdata;
 119         unsigned int i;
 120         int ret = 1;
 121
 122         for (i = 0; i < oe->numlower; i++) {
 123                 struct dentry *d = oe->lowerstack[i].dentry;
 124
 125                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 126                         ret = d->d_op->d_revalidate(d, flags);
 127                         if (ret < 0)
 128                                 return ret;
 129                         if (!ret) {
 130                                 if (!(flags & LOOKUP_RCU))
 131                                         d_invalidate(d);
 132                                 return -ESTALE;
 133                         }
 134                 }
 135         }
 136         return 1;
 137 }
 138
 139 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 140 {
 141         struct ovl_entry *oe = dentry->d_fsdata;
 142         unsigned int i;
 143         int ret = 1;
 144
 145         for (i = 0; i < oe->numlower; i++) {
 146                 struct dentry *d = oe->lowerstack[i].dentry;
 147
 148                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 149                         ret = d->d_op->d_weak_revalidate(d, flags);
 150                         if (ret <= 0)
 151                                 break;
 152                 }
 153         }
 154         return ret;
 155 }
 156
 157 static const struct dentry_operations ovl_dentry_operations = {
 158         .d_release = ovl_dentry_release,
 159         .d_real = ovl_d_real,
 160 };
 161
 162 static const struct dentry_operations ovl_reval_dentry_operations = {
 163         .d_release = ovl_dentry_release,
 164         .d_real = ovl_d_real,
 165         .d_revalidate = ovl_dentry_revalidate,
 166         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 167 };
 168
 169 static struct kmem_cache *ovl_inode_cachep;
 170
 171 static struct inode *ovl_alloc_inode(struct super_block *sb)
 172 {
 173         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 174
 175         if (!oi)
 176                 return NULL;
 177
 178         oi->cache = NULL;
 179         oi->redirect = NULL;
 180         oi->version = 0;
 181         oi->flags = 0;
 182         oi->__upperdentry = NULL;
 183         oi->lower = NULL;
 184         mutex_init(&oi->lock);
 185
 186         return &oi->vfs_inode;
 187 }
 188
 189 static void ovl_i_callback(struct rcu_head *head)
 190 {
 191         struct inode *inode = container_of(head, struct inode, i_rcu);
 192
 193         kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
 194 }
 195
 196 static void ovl_destroy_inode(struct inode *inode)
 197 {
 198         struct ovl_inode *oi = OVL_I(inode);
 199
 200         dput(oi->__upperdentry);
 201         iput(oi->lower);
 202         kfree(oi->redirect);
 203         ovl_dir_cache_free(inode);
 204         mutex_destroy(&oi->lock);
 205
 206         call_rcu(&inode->i_rcu, ovl_i_callback);
 207 }
 208
 209 static void ovl_free_fs(struct ovl_fs *ofs)
 210 {
 211         unsigned i;
 212
 213         dput(ofs->indexdir);
 214         dput(ofs->workdir);
 215         if (ofs->workdir_locked)
 216                 ovl_inuse_unlock(ofs->workbasedir);
 217         dput(ofs->workbasedir);
 218         if (ofs->upperdir_locked)
 219                 ovl_inuse_unlock(ofs->upper_mnt->mnt_root);
 220         mntput(ofs->upper_mnt);
 221         for (i = 0; i < ofs->numlower; i++)
 222                 mntput(ofs->lower_layers[i].mnt);
 223         for (i = 0; i < ofs->numlowerfs; i++)
 224                 free_anon_bdev(ofs->lower_fs[i].pseudo_dev);
 225         kfree(ofs->lower_layers);
 226         kfree(ofs->lower_fs);
 227
 228         kfree(ofs->config.lowerdir);
 229         kfree(ofs->config.upperdir);
 230         kfree(ofs->config.workdir);
 231         kfree(ofs->config.redirect_mode);
 232         if (ofs->creator_cred)
 233                 put_cred(ofs->creator_cred);
 234         kfree(ofs);
 235 }
 236
 237 static void ovl_put_super(struct super_block *sb)
 238 {
 239         struct ovl_fs *ofs = sb->s_fs_info;
 240
 241         ovl_free_fs(ofs);
 242 }
 243
 244 /* Sync real dirty inodes in upper filesystem (if it exists) */
 245 static int ovl_sync_fs(struct super_block *sb, int wait)
 246 {
 247         struct ovl_fs *ofs = sb->s_fs_info;
 248         struct super_block *upper_sb;
 249         int ret;
 250
 251         if (!ofs->upper_mnt)
 252                 return 0;
 253
 254         /*
 255          * If this is a sync(2) call or an emergency sync, all the super blocks
 256          * will be iterated, including upper_sb, so no need to do anything.
 257          *
 258          * If this is a syncfs(2) call, then we do need to call
 259          * sync_filesystem() on upper_sb, but enough if we do it when being
 260          * called with wait == 1.
 261          */
 262         if (!wait)
 263                 return 0;
 264
 265         upper_sb = ofs->upper_mnt->mnt_sb;
 266
 267         down_read(&upper_sb->s_umount);
 268         ret = sync_filesystem(upper_sb);
 269         up_read(&upper_sb->s_umount);
 270
 271         return ret;
 272 }
 273
 274 /**
 275  * ovl_statfs
 276  * @sb: The overlayfs super block
 277  * @buf: The struct kstatfs to fill in with stats
 278  *
 279  * Get the filesystem statistics.  As writes always target the upper layer
 280  * filesystem pass the statfs to the upper filesystem (if it exists)
 281  */
 282 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 283 {
 284         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 285         struct dentry *root_dentry = dentry->d_sb->s_root;
 286         struct path path;
 287         int err;
 288
 289         ovl_path_real(root_dentry, &path);
 290
 291         err = vfs_statfs(&path, buf);
 292         if (!err) {
 293                 buf->f_namelen = ofs->namelen;
 294                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 295         }
 296
 297         return err;
 298 }
 299
 300 /* Will this overlay be forced to mount/remount ro? */
 301 static bool ovl_force_readonly(struct ovl_fs *ofs)
 302 {
 303         return (!ofs->upper_mnt || !ofs->workdir);
 304 }
 305
 306 static const char *ovl_redirect_mode_def(void)
 307 {
 308         return ovl_redirect_dir_def ? "on" : "off";
 309 }
 310
 311 enum {
 312         OVL_XINO_OFF,
 313         OVL_XINO_AUTO,
 314         OVL_XINO_ON,
 315 };
 316
 317 static const char * const ovl_xino_str[] = {
 318         "off",
 319         "auto",
 320         "on",
 321 };
 322
 323 static inline int ovl_xino_def(void)
 324 {
 325         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 326 }
 327
 328 /**
 329  * ovl_show_options
 330  *
 331  * Prints the mount options for a given superblock.
 332  * Returns zero; does not fail.
 333  */
 334 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 335 {
 336         struct super_block *sb = dentry->d_sb;
 337         struct ovl_fs *ofs = sb->s_fs_info;
 338
 339         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 340         if (ofs->config.upperdir) {
 341                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 342                 seq_show_option(m, "workdir", ofs->config.workdir);
 343         }
 344         if (ofs->config.default_permissions)
 345                 seq_puts(m, ",default_permissions");
 346         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 347                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 348         if (ofs->config.index != ovl_index_def)
 349                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 350         if (ofs->config.nfs_export != ovl_nfs_export_def)
 351                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 352                                                 "on" : "off");
 353         if (ofs->config.xino != ovl_xino_def())
 354                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 355         if (ofs->config.metacopy != ovl_metacopy_def)
 356                 seq_printf(m, ",metacopy=%s",
 357                            ofs->config.metacopy ? "on" : "off");
 358         return 0;
 359 }
 360
 361 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 362 {
 363         struct ovl_fs *ofs = sb->s_fs_info;
 364
 365         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 366                 return -EROFS;
 367
 368         return 0;
 369 }
 370
 371 static const struct super_operations ovl_super_operations = {
 372         .alloc_inode    = ovl_alloc_inode,
 373         .destroy_inode  = ovl_destroy_inode,
 374         .drop_inode     = generic_delete_inode,
 375         .put_super      = ovl_put_super,
 376         .sync_fs        = ovl_sync_fs,
 377         .statfs         = ovl_statfs,
 378         .show_options   = ovl_show_options,
 379         .remount_fs     = ovl_remount,
 380 };
 381
 382 enum {
 383         OPT_LOWERDIR,
 384         OPT_UPPERDIR,
 385         OPT_WORKDIR,
 386         OPT_DEFAULT_PERMISSIONS,
 387         OPT_REDIRECT_DIR,
 388         OPT_INDEX_ON,
 389         OPT_INDEX_OFF,
 390         OPT_NFS_EXPORT_ON,
 391         OPT_NFS_EXPORT_OFF,
 392         OPT_XINO_ON,
 393         OPT_XINO_OFF,
 394         OPT_XINO_AUTO,
 395         OPT_METACOPY_ON,
 396         OPT_METACOPY_OFF,
 397         OPT_ERR,
 398 };
 399
 400 static const match_table_t ovl_tokens = {
 401         {OPT_LOWERDIR,                  "lowerdir=%s"},
 402         {OPT_UPPERDIR,                  "upperdir=%s"},
 403         {OPT_WORKDIR,                   "workdir=%s"},
 404         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 405         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 406         {OPT_INDEX_ON,                  "index=on"},
 407         {OPT_INDEX_OFF,                 "index=off"},
 408         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 409         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 410         {OPT_XINO_ON,                   "xino=on"},
 411         {OPT_XINO_OFF,                  "xino=off"},
 412         {OPT_XINO_AUTO,                 "xino=auto"},
 413         {OPT_METACOPY_ON,               "metacopy=on"},
 414         {OPT_METACOPY_OFF,              "metacopy=off"},
 415         {OPT_ERR,                       NULL}
 416 };
 417
 418 static char *ovl_next_opt(char **s)
 419 {
 420         char *sbegin = *s;
 421         char *p;
 422
 423         if (sbegin == NULL)
 424                 return NULL;
 425
 426         for (p = sbegin; *p; p++) {
 427                 if (*p == '\\') {
 428                         p++;
 429                         if (!*p)
 430                                 break;
 431                 } else if (*p == ',') {
 432                         *p = '\0';
 433                         *s = p + 1;
 434                         return sbegin;
 435                 }
 436         }
 437         *s = NULL;
 438         return sbegin;
 439 }
 440
 441 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 442 {
 443         if (strcmp(mode, "on") == 0) {
 444                 config->redirect_dir = true;
 445                 /*
 446                  * Does not make sense to have redirect creation without
 447                  * redirect following.
 448                  */
 449                 config->redirect_follow = true;
 450         } else if (strcmp(mode, "follow") == 0) {
 451                 config->redirect_follow = true;
 452         } else if (strcmp(mode, "off") == 0) {
 453                 if (ovl_redirect_always_follow)
 454                         config->redirect_follow = true;
 455         } else if (strcmp(mode, "nofollow") != 0) {
 456                 pr_err("overlayfs: bad mount option \"redirect_dir=%s\"\n",
 457                        mode);
 458                 return -EINVAL;
 459         }
 460
 461         return 0;
 462 }
 463
 464 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 465 {
 466         char *p;
 467         int err;
 468
 469         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 470         if (!config->redirect_mode)
 471                 return -ENOMEM;
 472
 473         while ((p = ovl_next_opt(&opt)) != NULL) {
 474                 int token;
 475                 substring_t args[MAX_OPT_ARGS];
 476
 477                 if (!*p)
 478                         continue;
 479
 480                 token = match_token(p, ovl_tokens, args);
 481                 switch (token) {
 482                 case OPT_UPPERDIR:
 483                         kfree(config->upperdir);
 484                         config->upperdir = match_strdup(&args[0]);
 485                         if (!config->upperdir)
 486                                 return -ENOMEM;
 487                         break;
 488
 489                 case OPT_LOWERDIR:
 490                         kfree(config->lowerdir);
 491                         config->lowerdir = match_strdup(&args[0]);
 492                         if (!config->lowerdir)
 493                                 return -ENOMEM;
 494                         break;
 495
 496                 case OPT_WORKDIR:
 497                         kfree(config->workdir);
 498                         config->workdir = match_strdup(&args[0]);
 499                         if (!config->workdir)
 500                                 return -ENOMEM;
 501                         break;
 502
 503                 case OPT_DEFAULT_PERMISSIONS:
 504                         config->default_permissions = true;
 505                         break;
 506
 507                 case OPT_REDIRECT_DIR:
 508                         kfree(config->redirect_mode);
 509                         config->redirect_mode = match_strdup(&args[0]);
 510                         if (!config->redirect_mode)
 511                                 return -ENOMEM;
 512                         break;
 513
 514                 case OPT_INDEX_ON:
 515                         config->index = true;
 516                         break;
 517
 518                 case OPT_INDEX_OFF:
 519                         config->index = false;
 520                         break;
 521
 522                 case OPT_NFS_EXPORT_ON:
 523                         config->nfs_export = true;
 524                         break;
 525
 526                 case OPT_NFS_EXPORT_OFF:
 527                         config->nfs_export = false;
 528                         break;
 529
 530                 case OPT_XINO_ON:
 531                         config->xino = OVL_XINO_ON;
 532                         break;
 533
 534                 case OPT_XINO_OFF:
 535                         config->xino = OVL_XINO_OFF;
 536                         break;
 537
 538                 case OPT_XINO_AUTO:
 539                         config->xino = OVL_XINO_AUTO;
 540                         break;
 541
 542                 case OPT_METACOPY_ON:
 543                         config->metacopy = true;
 544                         break;
 545
 546                 case OPT_METACOPY_OFF:
 547                         config->metacopy = false;
 548                         break;
 549
 550                 default:
 551                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 552                         return -EINVAL;
 553                 }
 554         }
 555
 556         /* Workdir is useless in non-upper mount */
 557         if (!config->upperdir && config->workdir) {
 558                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 559                         config->workdir);
 560                 kfree(config->workdir);
 561                 config->workdir = NULL;
 562         }
 563
 564         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 565         if (err)
 566                 return err;
 567
 568         /* metacopy feature with upper requires redirect_dir=on */
 569         if (config->upperdir && config->metacopy && !config->redirect_dir) {
 570                 pr_warn("overlayfs: metadata only copy up requires \"redirect_dir=on\", falling back to metacopy=off.\n");
 571                 config->metacopy = false;
 572         } else if (config->metacopy && !config->redirect_follow) {
 573                 pr_warn("overlayfs: metadata only copy up requires \"redirect_dir=follow\" on non-upper mount, falling back to metacopy=off.\n");
 574                 config->metacopy = false;
 575         }
 576
 577         return 0;
 578 }
 579
 580 #define OVL_WORKDIR_NAME "work"
 581 #define OVL_INDEXDIR_NAME "index"
 582
 583 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 584                                          const char *name, bool persist)
 585 {
 586         struct inode *dir =  ofs->workbasedir->d_inode;
 587         struct vfsmount *mnt = ofs->upper_mnt;
 588         struct dentry *work;
 589         int err;
 590         bool retried = false;
 591         bool locked = false;
 592
 593         inode_lock_nested(dir, I_MUTEX_PARENT);
 594         locked = true;
 595
 596 retry:
 597         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
 598
 599         if (!IS_ERR(work)) {
 600                 struct iattr attr = {
 601                         .ia_valid = ATTR_MODE,
 602                         .ia_mode = S_IFDIR | 0,
 603                 };
 604
 605                 if (work->d_inode) {
 606                         err = -EEXIST;
 607                         if (retried)
 608                                 goto out_dput;
 609
 610                         if (persist)
 611                                 goto out_unlock;
 612
 613                         retried = true;
 614                         ovl_workdir_cleanup(dir, mnt, work, 0);
 615                         dput(work);
 616                         goto retry;
 617                 }
 618
 619                 work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
 620                 err = PTR_ERR(work);
 621                 if (IS_ERR(work))
 622                         goto out_err;
 623
 624                 /*
 625                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 626                  *
 627                  * a) success (there was a POSIX ACL xattr and was removed)
 628                  * b) -ENODATA (there was no POSIX ACL xattr)
 629                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 630                  *
 631                  * There are various other error values that could effectively
 632                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 633                  * if the xattr name is too long), but the set of filesystems
 634                  * allowed as upper are limited to "normal" ones, where checking
 635                  * for the above two errors is sufficient.
 636                  */
 637                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 638                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 639                         goto out_dput;
 640
 641                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 642                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 643                         goto out_dput;
 644
 645                 /* Clear any inherited mode bits */
 646                 inode_lock(work->d_inode);
 647                 err = notify_change(work, &attr, NULL);
 648                 inode_unlock(work->d_inode);
 649                 if (err)
 650                         goto out_dput;
 651         } else {
 652                 err = PTR_ERR(work);
 653                 goto out_err;
 654         }
 655 out_unlock:
 656         if (locked)
 657                 inode_unlock(dir);
 658
 659         return work;
 660
 661 out_dput:
 662         dput(work);
 663 out_err:
 664         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 665                 ofs->config.workdir, name, -err);
 666         work = NULL;
 667         goto out_unlock;
 668 }
 669
 670 static void ovl_unescape(char *s)
 671 {
 672         char *d = s;
 673
 674         for (;; s++, d++) {
 675                 if (*s == '\\')
 676                         s++;
 677                 *d = *s;
 678                 if (!*s)
 679                         break;
 680         }
 681 }
 682
 683 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 684 {
 685         int err = -EINVAL;
 686
 687         if (!*name) {
 688                 pr_err("overlayfs: empty lowerdir\n");
 689                 goto out;
 690         }
 691         err = kern_path(name, LOOKUP_FOLLOW, path);
 692         if (err) {
 693                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 694                 goto out;
 695         }
 696         err = -EINVAL;
 697         if (ovl_dentry_weird(path->dentry)) {
 698                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 699                 goto out_put;
 700         }
 701         if (!d_is_dir(path->dentry)) {
 702                 pr_err("overlayfs: '%s' not a directory\n", name);
 703                 goto out_put;
 704         }
 705         return 0;
 706
 707 out_put:
 708         path_put_init(path);
 709 out:
 710         return err;
 711 }
 712
 713 static int ovl_mount_dir(const char *name, struct path *path)
 714 {
 715         int err = -ENOMEM;
 716         char *tmp = kstrdup(name, GFP_KERNEL);
 717
 718         if (tmp) {
 719                 ovl_unescape(tmp);
 720                 err = ovl_mount_dir_noesc(tmp, path);
 721
 722                 if (!err)
 723                         if (ovl_dentry_remote(path->dentry)) {
 724                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 725                                        tmp);
 726                                 path_put_init(path);
 727                                 err = -EINVAL;
 728                         }
 729                 kfree(tmp);
 730         }
 731         return err;
 732 }
 733
 734 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 735                              const char *name)
 736 {
 737         struct kstatfs statfs;
 738         int err = vfs_statfs(path, &statfs);
 739
 740         if (err)
 741                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 742         else
 743                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 744
 745         return err;
 746 }
 747
 748 static int ovl_lower_dir(const char *name, struct path *path,
 749                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 750 {
 751         int fh_type;
 752         int err;
 753
 754         err = ovl_mount_dir_noesc(name, path);
 755         if (err)
 756                 goto out;
 757
 758         err = ovl_check_namelen(path, ofs, name);
 759         if (err)
 760                 goto out_put;
 761
 762         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 763
 764         if (ovl_dentry_remote(path->dentry))
 765                 *remote = true;
 766
 767         /*
 768          * The inodes index feature and NFS export need to encode and decode
 769          * file handles, so they require that all layers support them.
 770          */
 771         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 772         if ((ofs->config.nfs_export ||
 773              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 774                 ofs->config.index = false;
 775                 ofs->config.nfs_export = false;
 776                 pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 777                         name);
 778         }
 779
 780         /* Check if lower fs has 32bit inode numbers */
 781         if (fh_type != FILEID_INO32_GEN)
 782                 ofs->xino_bits = 0;
 783
 784         return 0;
 785
 786 out_put:
 787         path_put_init(path);
 788 out:
 789         return err;
 790 }
 791
 792 /* Workdir should not be subdir of upperdir and vice versa */
 793 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 794 {
 795         bool ok = false;
 796
 797         if (workdir != upperdir) {
 798                 ok = (lock_rename(workdir, upperdir) == NULL);
 799                 unlock_rename(workdir, upperdir);
 800         }
 801         return ok;
 802 }
 803
 804 static unsigned int ovl_split_lowerdirs(char *str)
 805 {
 806         unsigned int ctr = 1;
 807         char *s, *d;
 808
 809         for (s = d = str;; s++, d++) {
 810                 if (*s == '\\') {
 811                         s++;
 812                 } else if (*s == ':') {
 813                         *d = '\0';
 814                         ctr++;
 815                         continue;
 816                 }
 817                 *d = *s;
 818                 if (!*s)
 819                         break;
 820         }
 821         return ctr;
 822 }
 823
 824 static int __maybe_unused
 825 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 826                         struct dentry *dentry, struct inode *inode,
 827                         const char *name, void *buffer, size_t size)
 828 {
 829         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 830 }
 831
 832 static int __maybe_unused
 833 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 834                         struct dentry *dentry, struct inode *inode,
 835                         const char *name, const void *value,
 836                         size_t size, int flags)
 837 {
 838         struct dentry *workdir = ovl_workdir(dentry);
 839         struct inode *realinode = ovl_inode_real(inode);
 840         struct posix_acl *acl = NULL;
 841         int err;
 842
 843         /* Check that everything is OK before copy-up */
 844         if (value) {
 845                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 846                 if (IS_ERR(acl))
 847                         return PTR_ERR(acl);
 848         }
 849         err = -EOPNOTSUPP;
 850         if (!IS_POSIXACL(d_inode(workdir)))
 851                 goto out_acl_release;
 852         if (!realinode->i_op->set_acl)
 853                 goto out_acl_release;
 854         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 855                 err = acl ? -EACCES : 0;
 856                 goto out_acl_release;
 857         }
 858         err = -EPERM;
 859         if (!inode_owner_or_capable(inode))
 860                 goto out_acl_release;
 861
 862         posix_acl_release(acl);
 863
 864         /*
 865          * Check if sgid bit needs to be cleared (actual setacl operation will
 866          * be done with mounter's capabilities and so that won't do it for us).
 867          */
 868         if (unlikely(inode->i_mode & S_ISGID) &&
 869             handler->flags == ACL_TYPE_ACCESS &&
 870             !in_group_p(inode->i_gid) &&
 871             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 872                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 873
 874                 err = ovl_setattr(dentry, &iattr);
 875                 if (err)
 876                         return err;
 877         }
 878
 879         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
 880         if (!err)
 881                 ovl_copyattr(ovl_inode_real(inode), inode);
 882
 883         return err;
 884
 885 out_acl_release:
 886         posix_acl_release(acl);
 887         return err;
 888 }
 889
 890 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 891                              struct dentry *dentry, struct inode *inode,
 892                              const char *name, void *buffer, size_t size)
 893 {
 894         return -EOPNOTSUPP;
 895 }
 896
 897 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 898                              struct dentry *dentry, struct inode *inode,
 899                              const char *name, const void *value,
 900                              size_t size, int flags)
 901 {
 902         return -EOPNOTSUPP;
 903 }
 904
 905 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 906                                struct dentry *dentry, struct inode *inode,
 907                                const char *name, void *buffer, size_t size)
 908 {
 909         return ovl_xattr_get(dentry, inode, name, buffer, size);
 910 }
 911
 912 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 913                                struct dentry *dentry, struct inode *inode,
 914                                const char *name, const void *value,
 915                                size_t size, int flags)
 916 {
 917         return ovl_xattr_set(dentry, inode, name, value, size, flags);
 918 }
 919
 920 static const struct xattr_handler __maybe_unused
 921 ovl_posix_acl_access_xattr_handler = {
 922         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 923         .flags = ACL_TYPE_ACCESS,
 924         .get = ovl_posix_acl_xattr_get,
 925         .set = ovl_posix_acl_xattr_set,
 926 };
 927
 928 static const struct xattr_handler __maybe_unused
 929 ovl_posix_acl_default_xattr_handler = {
 930         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 931         .flags = ACL_TYPE_DEFAULT,
 932         .get = ovl_posix_acl_xattr_get,
 933         .set = ovl_posix_acl_xattr_set,
 934 };
 935
 936 static const struct xattr_handler ovl_own_xattr_handler = {
 937         .prefix = OVL_XATTR_PREFIX,
 938         .get = ovl_own_xattr_get,
 939         .set = ovl_own_xattr_set,
 940 };
 941
 942 static const struct xattr_handler ovl_other_xattr_handler = {
 943         .prefix = "", /* catch all */
 944         .get = ovl_other_xattr_get,
 945         .set = ovl_other_xattr_set,
 946 };
 947
 948 static const struct xattr_handler *ovl_xattr_handlers[] = {
 949 #ifdef CONFIG_FS_POSIX_ACL
 950         &ovl_posix_acl_access_xattr_handler,
 951         &ovl_posix_acl_default_xattr_handler,
 952 #endif
 953         &ovl_own_xattr_handler,
 954         &ovl_other_xattr_handler,
 955         NULL
 956 };
 957
 958 static int ovl_get_upper(struct ovl_fs *ofs, struct path *upperpath)
 959 {
 960         struct vfsmount *upper_mnt;
 961         int err;
 962
 963         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
 964         if (err)
 965                 goto out;
 966
 967         /* Upper fs should not be r/o */
 968         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
 969                 pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 970                 err = -EINVAL;
 971                 goto out;
 972         }
 973
 974         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
 975         if (err)
 976                 goto out;
 977
 978         err = -EBUSY;
 979         if (ovl_inuse_trylock(upperpath->dentry)) {
 980                 ofs->upperdir_locked = true;
 981         } else if (ofs->config.index) {
 982                 pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n");
 983                 goto out;
 984         } else {
 985                 pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
 986         }
 987
 988         upper_mnt = clone_private_mount(upperpath);
 989         err = PTR_ERR(upper_mnt);
 990         if (IS_ERR(upper_mnt)) {
 991                 pr_err("overlayfs: failed to clone upperpath\n");
 992                 goto out;
 993         }
 994
 995         /* Don't inherit atime flags */
 996         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
 997         ofs->upper_mnt = upper_mnt;
 998         err = 0;
 999 out:
1000         return err;
1001 }
1002
1003 static int ovl_make_workdir(struct ovl_fs *ofs, struct path *workpath)
1004 {
1005         struct vfsmount *mnt = ofs->upper_mnt;
1006         struct dentry *temp;
1007         int fh_type;
1008         int err;
1009
1010         err = mnt_want_write(mnt);
1011         if (err)
1012                 return err;
1013
1014         ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1015         if (!ofs->workdir)
1016                 goto out;
1017
1018         /*
1019          * Upper should support d_type, else whiteouts are visible.  Given
1020          * workdir and upper are on same fs, we can do iterate_dir() on
1021          * workdir. This check requires successful creation of workdir in
1022          * previous step.
1023          */
1024         err = ovl_check_d_type_supported(workpath);
1025         if (err < 0)
1026                 goto out;
1027
1028         /*
1029          * We allowed this configuration and don't want to break users over
1030          * kernel upgrade. So warn instead of erroring out.
1031          */
1032         if (!err)
1033                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
1034
1035         /* Check if upper/work fs supports O_TMPFILE */
1036         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1037         ofs->tmpfile = !IS_ERR(temp);
1038         if (ofs->tmpfile)
1039                 dput(temp);
1040         else
1041                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
1042
1043         /*
1044          * Check if upper/work fs supports trusted.overlay.* xattr
1045          */
1046         err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
1047         if (err) {
1048                 ofs->noxattr = true;
1049                 ofs->config.index = false;
1050                 ofs->config.metacopy = false;
1051                 pr_warn("overlayfs: upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1052                 err = 0;
1053         } else {
1054                 vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
1055         }
1056
1057         /* Check if upper/work fs supports file handles */
1058         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1059         if (ofs->config.index && !fh_type) {
1060                 ofs->config.index = false;
1061                 pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1062         }
1063
1064         /* Check if upper fs has 32bit inode numbers */
1065         if (fh_type != FILEID_INO32_GEN)
1066                 ofs->xino_bits = 0;
1067
1068         /* NFS export of r/w mount depends on index */
1069         if (ofs->config.nfs_export && !ofs->config.index) {
1070                 pr_warn("overlayfs: NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1071                 ofs->config.nfs_export = false;
1072         }
1073 out:
1074         mnt_drop_write(mnt);
1075         return err;
1076 }
1077
1078 static int ovl_get_workdir(struct ovl_fs *ofs, struct path *upperpath)
1079 {
1080         int err;
1081         struct path workpath = { };
1082
1083         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1084         if (err)
1085                 goto out;
1086
1087         err = -EINVAL;
1088         if (upperpath->mnt != workpath.mnt) {
1089                 pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1090                 goto out;
1091         }
1092         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1093                 pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1094                 goto out;
1095         }
1096
1097         err = -EBUSY;
1098         if (ovl_inuse_trylock(workpath.dentry)) {
1099                 ofs->workdir_locked = true;
1100         } else if (ofs->config.index) {
1101                 pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n");
1102                 goto out;
1103         } else {
1104                 pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
1105         }
1106
1107         ofs->workbasedir = dget(workpath.dentry);
1108         err = ovl_make_workdir(ofs, &workpath);
1109         if (err)
1110                 goto out;
1111
1112         err = 0;
1113 out:
1114         path_put(&workpath);
1115
1116         return err;
1117 }
1118
1119 static int ovl_get_indexdir(struct ovl_fs *ofs, struct ovl_entry *oe,
1120                             struct path *upperpath)
1121 {
1122         struct vfsmount *mnt = ofs->upper_mnt;
1123         int err;
1124
1125         err = mnt_want_write(mnt);
1126         if (err)
1127                 return err;
1128
1129         /* Verify lower root is upper root origin */
1130         err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
1131                                 true);
1132         if (err) {
1133                 pr_err("overlayfs: failed to verify upper root origin\n");
1134                 goto out;
1135         }
1136
1137         ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1138         if (ofs->indexdir) {
1139                 /*
1140                  * Verify upper root is exclusively associated with index dir.
1141                  * Older kernels stored upper fh in "trusted.overlay.origin"
1142                  * xattr. If that xattr exists, verify that it is a match to
1143                  * upper dir file handle. In any case, verify or set xattr
1144                  * "trusted.overlay.upper" to indicate that index may have
1145                  * directory entries.
1146                  */
1147                 if (ovl_check_origin_xattr(ofs->indexdir)) {
1148                         err = ovl_verify_set_fh(ofs->indexdir, OVL_XATTR_ORIGIN,
1149                                                 upperpath->dentry, true, false);
1150                         if (err)
1151                                 pr_err("overlayfs: failed to verify index dir 'origin' xattr\n");
1152                 }
1153                 err = ovl_verify_upper(ofs->indexdir, upperpath->dentry, true);
1154                 if (err)
1155                         pr_err("overlayfs: failed to verify index dir 'upper' xattr\n");
1156
1157                 /* Cleanup bad/stale/orphan index entries */
1158                 if (!err)
1159                         err = ovl_indexdir_cleanup(ofs);
1160         }
1161         if (err || !ofs->indexdir)
1162                 pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1163
1164 out:
1165         mnt_drop_write(mnt);
1166         return err;
1167 }
1168
1169 /* Get a unique fsid for the layer */
1170 static int ovl_get_fsid(struct ovl_fs *ofs, struct super_block *sb)
1171 {
1172         unsigned int i;
1173         dev_t dev;
1174         int err;
1175
1176         /* fsid 0 is reserved for upper fs even with non upper overlay */
1177         if (ofs->upper_mnt && ofs->upper_mnt->mnt_sb == sb)
1178                 return 0;
1179
1180         for (i = 0; i < ofs->numlowerfs; i++) {
1181                 if (ofs->lower_fs[i].sb == sb)
1182                         return i + 1;
1183         }
1184
1185         err = get_anon_bdev(&dev);
1186         if (err) {
1187                 pr_err("overlayfs: failed to get anonymous bdev for lowerpath\n");
1188                 return err;
1189         }
1190
1191         ofs->lower_fs[ofs->numlowerfs].sb = sb;
1192         ofs->lower_fs[ofs->numlowerfs].pseudo_dev = dev;
1193         ofs->numlowerfs++;
1194
1195         return ofs->numlowerfs;
1196 }
1197
1198 static int ovl_get_lower_layers(struct ovl_fs *ofs, struct path *stack,
1199                                 unsigned int numlower)
1200 {
1201         int err;
1202         unsigned int i;
1203
1204         err = -ENOMEM;
1205         ofs->lower_layers = kcalloc(numlower, sizeof(struct ovl_layer),
1206                                     GFP_KERNEL);
1207         if (ofs->lower_layers == NULL)
1208                 goto out;
1209
1210         ofs->lower_fs = kcalloc(numlower, sizeof(struct ovl_sb),
1211                                 GFP_KERNEL);
1212         if (ofs->lower_fs == NULL)
1213                 goto out;
1214
1215         for (i = 0; i < numlower; i++) {
1216                 struct vfsmount *mnt;
1217                 int fsid;
1218
1219                 err = fsid = ovl_get_fsid(ofs, stack[i].mnt->mnt_sb);
1220                 if (err < 0)
1221                         goto out;
1222
1223                 mnt = clone_private_mount(&stack[i]);
1224                 err = PTR_ERR(mnt);
1225                 if (IS_ERR(mnt)) {
1226                         pr_err("overlayfs: failed to clone lowerpath\n");
1227                         goto out;
1228                 }
1229
1230                 /*
1231                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1232                  * will fail instead of modifying lower fs.
1233                  */
1234                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1235
1236                 ofs->lower_layers[ofs->numlower].mnt = mnt;
1237                 ofs->lower_layers[ofs->numlower].idx = i + 1;
1238                 ofs->lower_layers[ofs->numlower].fsid = fsid;
1239                 if (fsid) {
1240                         ofs->lower_layers[ofs->numlower].fs =
1241                                 &ofs->lower_fs[fsid - 1];
1242                 }
1243                 ofs->numlower++;
1244         }
1245
1246         /*
1247          * When all layers on same fs, overlay can use real inode numbers.
1248          * With mount option "xino=on", mounter declares that there are enough
1249          * free high bits in underlying fs to hold the unique fsid.
1250          * If overlayfs does encounter underlying inodes using the high xino
1251          * bits reserved for fsid, it emits a warning and uses the original
1252          * inode number.
1253          */
1254         if (!ofs->numlowerfs || (ofs->numlowerfs == 1 && !ofs->upper_mnt)) {
1255                 ofs->xino_bits = 0;
1256                 ofs->config.xino = OVL_XINO_OFF;
1257         } else if (ofs->config.xino == OVL_XINO_ON && !ofs->xino_bits) {
1258                 /*
1259                  * This is a roundup of number of bits needed for numlowerfs+1
1260                  * (i.e. ilog2(numlowerfs+1 - 1) + 1). fsid 0 is reserved for
1261                  * upper fs even with non upper overlay.
1262                  */
1263                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 31);
1264                 ofs->xino_bits = ilog2(ofs->numlowerfs) + 1;
1265         }
1266
1267         if (ofs->xino_bits) {
1268                 pr_info("overlayfs: \"xino\" feature enabled using %d upper inode bits.\n",
1269                         ofs->xino_bits);
1270         }
1271
1272         err = 0;
1273 out:
1274         return err;
1275 }
1276
1277 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1278                                             struct ovl_fs *ofs)
1279 {
1280         int err;
1281         char *lowertmp, *lower;
1282         struct path *stack = NULL;
1283         unsigned int stacklen, numlower = 0, i;
1284         bool remote = false;
1285         struct ovl_entry *oe;
1286
1287         err = -ENOMEM;
1288         lowertmp = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1289         if (!lowertmp)
1290                 goto out_err;
1291
1292         err = -EINVAL;
1293         stacklen = ovl_split_lowerdirs(lowertmp);
1294         if (stacklen > OVL_MAX_STACK) {
1295                 pr_err("overlayfs: too many lower directories, limit is %d\n",
1296                        OVL_MAX_STACK);
1297                 goto out_err;
1298         } else if (!ofs->config.upperdir && stacklen == 1) {
1299                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1300                 goto out_err;
1301         } else if (!ofs->config.upperdir && ofs->config.nfs_export &&
1302                    ofs->config.redirect_follow) {
1303                 pr_warn("overlayfs: NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
1304                 ofs->config.nfs_export = false;
1305         }
1306
1307         err = -ENOMEM;
1308         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1309         if (!stack)
1310                 goto out_err;
1311
1312         err = -EINVAL;
1313         lower = lowertmp;
1314         for (numlower = 0; numlower < stacklen; numlower++) {
1315                 err = ovl_lower_dir(lower, &stack[numlower], ofs,
1316                                     &sb->s_stack_depth, &remote);
1317                 if (err)
1318                         goto out_err;
1319
1320                 lower = strchr(lower, '\0') + 1;
1321         }
1322
1323         err = -EINVAL;
1324         sb->s_stack_depth++;
1325         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1326                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1327                 goto out_err;
1328         }
1329
1330         err = ovl_get_lower_layers(ofs, stack, numlower);
1331         if (err)
1332                 goto out_err;
1333
1334         err = -ENOMEM;
1335         oe = ovl_alloc_entry(numlower);
1336         if (!oe)
1337                 goto out_err;
1338
1339         for (i = 0; i < numlower; i++) {
1340                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1341                 oe->lowerstack[i].layer = &ofs->lower_layers[i];
1342         }
1343
1344         if (remote)
1345                 sb->s_d_op = &ovl_reval_dentry_operations;
1346         else
1347                 sb->s_d_op = &ovl_dentry_operations;
1348
1349 out:
1350         for (i = 0; i < numlower; i++)
1351                 path_put(&stack[i]);
1352         kfree(stack);
1353         kfree(lowertmp);
1354
1355         return oe;
1356
1357 out_err:
1358         oe = ERR_PTR(err);
1359         goto out;
1360 }
1361
1362 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1363 {
1364         struct path upperpath = { };
1365         struct dentry *root_dentry;
1366         struct ovl_entry *oe;
1367         struct ovl_fs *ofs;
1368         struct cred *cred;
1369         int err;
1370
1371         err = -ENOMEM;
1372         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1373         if (!ofs)
1374                 goto out;
1375
1376         ofs->creator_cred = cred = prepare_creds();
1377         if (!cred)
1378                 goto out_err;
1379
1380         ofs->config.index = ovl_index_def;
1381         ofs->config.nfs_export = ovl_nfs_export_def;
1382         ofs->config.xino = ovl_xino_def();
1383         ofs->config.metacopy = ovl_metacopy_def;
1384         err = ovl_parse_opt((char *) data, &ofs->config);
1385         if (err)
1386                 goto out_err;
1387
1388         err = -EINVAL;
1389         if (!ofs->config.lowerdir) {
1390                 if (!silent)
1391                         pr_err("overlayfs: missing 'lowerdir'\n");
1392                 goto out_err;
1393         }
1394
1395         sb->s_stack_depth = 0;
1396         sb->s_maxbytes = MAX_LFS_FILESIZE;
1397         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1398         if (ofs->config.xino != OVL_XINO_OFF)
1399                 ofs->xino_bits = BITS_PER_LONG - 32;
1400
1401         if (ofs->config.upperdir) {
1402                 if (!ofs->config.workdir) {
1403                         pr_err("overlayfs: missing 'workdir'\n");
1404                         goto out_err;
1405                 }
1406
1407                 err = ovl_get_upper(ofs, &upperpath);
1408                 if (err)
1409                         goto out_err;
1410
1411                 err = ovl_get_workdir(ofs, &upperpath);
1412                 if (err)
1413                         goto out_err;
1414
1415                 if (!ofs->workdir)
1416                         sb->s_flags |= SB_RDONLY;
1417
1418                 sb->s_stack_depth = ofs->upper_mnt->mnt_sb->s_stack_depth;
1419                 sb->s_time_gran = ofs->upper_mnt->mnt_sb->s_time_gran;
1420
1421         }
1422         oe = ovl_get_lowerstack(sb, ofs);
1423         err = PTR_ERR(oe);
1424         if (IS_ERR(oe))
1425                 goto out_err;
1426
1427         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1428         if (!ofs->upper_mnt)
1429                 sb->s_flags |= SB_RDONLY;
1430
1431         if (!(ovl_force_readonly(ofs)) && ofs->config.index) {
1432                 err = ovl_get_indexdir(ofs, oe, &upperpath);
1433                 if (err)
1434                         goto out_free_oe;
1435
1436                 /* Force r/o mount with no index dir */
1437                 if (!ofs->indexdir) {
1438                         dput(ofs->workdir);
1439                         ofs->workdir = NULL;
1440                         sb->s_flags |= SB_RDONLY;
1441                 }
1442
1443         }
1444
1445         /* Show index=off in /proc/mounts for forced r/o mount */
1446         if (!ofs->indexdir) {
1447                 ofs->config.index = false;
1448                 if (ofs->upper_mnt && ofs->config.nfs_export) {
1449                         pr_warn("overlayfs: NFS export requires an index dir, falling back to nfs_export=off.\n");
1450                         ofs->config.nfs_export = false;
1451                 }
1452         }
1453
1454         if (ofs->config.metacopy && ofs->config.nfs_export) {
1455                 pr_warn("overlayfs: NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1456                 ofs->config.nfs_export = false;
1457         }
1458
1459         if (ofs->config.nfs_export)
1460                 sb->s_export_op = &ovl_export_operations;
1461
1462         /* Never override disk quota limits or use reserved space */
1463         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1464
1465         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1466         sb->s_op = &ovl_super_operations;
1467         sb->s_xattr = ovl_xattr_handlers;
1468         sb->s_fs_info = ofs;
1469         sb->s_flags |= SB_POSIXACL;
1470
1471         err = -ENOMEM;
1472         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1473         if (!root_dentry)
1474                 goto out_free_oe;
1475
1476         root_dentry->d_fsdata = oe;
1477
1478         mntput(upperpath.mnt);
1479         if (upperpath.dentry) {
1480                 ovl_dentry_set_upper_alias(root_dentry);
1481                 if (ovl_is_impuredir(upperpath.dentry))
1482                         ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1483         }
1484
1485         /* Root is always merge -> can have whiteouts */
1486         ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1487         ovl_dentry_set_flag(OVL_E_CONNECTED, root_dentry);
1488         ovl_set_upperdata(d_inode(root_dentry));
1489         ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1490                        ovl_dentry_lower(root_dentry));
1491
1492         sb->s_root = root_dentry;
1493
1494         return 0;
1495
1496 out_free_oe:
1497         ovl_entry_stack_free(oe);
1498         kfree(oe);
1499 out_err:
1500         path_put(&upperpath);
1501         ovl_free_fs(ofs);
1502 out:
1503         return err;
1504 }
1505
1506 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1507                                 const char *dev_name, void *raw_data)
1508 {
1509         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1510 }
1511
1512 static struct file_system_type ovl_fs_type = {
1513         .owner          = THIS_MODULE,
1514         .name           = "overlay",
1515         .mount          = ovl_mount,
1516         .kill_sb        = kill_anon_super,
1517 };
1518 MODULE_ALIAS_FS("overlay");
1519
1520 static void ovl_inode_init_once(void *foo)
1521 {
1522         struct ovl_inode *oi = foo;
1523
1524         inode_init_once(&oi->vfs_inode);
1525 }
1526
1527 static int __init ovl_init(void)
1528 {
1529         int err;
1530
1531         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1532                                              sizeof(struct ovl_inode), 0,
1533                                              (SLAB_RECLAIM_ACCOUNT|
1534                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1535                                              ovl_inode_init_once);
1536         if (ovl_inode_cachep == NULL)
1537                 return -ENOMEM;
1538
1539         err = register_filesystem(&ovl_fs_type);
1540         if (err)
1541                 kmem_cache_destroy(ovl_inode_cachep);
1542
1543         return err;
1544 }
1545
1546 static void __exit ovl_exit(void)
1547 {
1548         unregister_filesystem(&ovl_fs_type);
1549
1550         /*
1551          * Make sure all delayed rcu free inodes are flushed before we
1552          * destroy cache.
1553          */
1554         rcu_barrier();
1555         kmem_cache_destroy(ovl_inode_cachep);
1556
1557 }
1558
1559 module_init(ovl_init);
1560 module_exit(ovl_exit);