fs/overlayfs/super.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/fs.h>
  11 #include <linux/namei.h>
  12 #include <linux/xattr.h>
  13 #include <linux/security.h>
  14 #include <linux/mount.h>
  15 #include <linux/slab.h>
  16 #include <linux/parser.h>
  17 #include <linux/module.h>
  18 #include <linux/sched.h>
  19 #include <linux/statfs.h>
  20 #include <linux/seq_file.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 #define OVERLAYFS_SUPER_MAGIC 0x794c7630
  28
  29 struct ovl_config {
  30         char *lowerdir;
  31         char *upperdir;
  32         char *workdir;
  33 };
  34
  35 /* private information held for overlayfs's superblock */
  36 struct ovl_fs {
  37         struct vfsmount *upper_mnt;
  38         unsigned numlower;
  39         struct vfsmount **lower_mnt;
  40         struct dentry *workdir;
  41         long lower_namelen;
  42         /* pathnames of lower and upper dirs, for show_options */
  43         struct ovl_config config;
  44 };
  45
  46 struct ovl_dir_cache;
  47
  48 /* private information held for every overlayfs dentry */
  49 struct ovl_entry {
  50         struct dentry *__upperdentry;
  51         struct ovl_dir_cache *cache;
  52         union {
  53                 struct {
  54                         u64 version;
  55                         bool opaque;
  56                 };
  57                 struct rcu_head rcu;
  58         };
  59         unsigned numlower;
  60         struct path lowerstack[];
  61 };
  62
  63 #define OVL_MAX_STACK 500
  64
  65 static struct dentry *__ovl_dentry_lower(struct ovl_entry *oe)
  66 {
  67         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
  68 }
  69
  70 enum ovl_path_type ovl_path_type(struct dentry *dentry)
  71 {
  72         struct ovl_entry *oe = dentry->d_fsdata;
  73         enum ovl_path_type type = 0;
  74
  75         if (oe->__upperdentry) {
  76                 type = __OVL_PATH_UPPER;
  77
  78                 if (oe->numlower) {
  79                         if (S_ISDIR(dentry->d_inode->i_mode))
  80                                 type |= __OVL_PATH_MERGE;
  81                 } else if (!oe->opaque) {
  82                         type |= __OVL_PATH_PURE;
  83                 }
  84         } else {
  85                 if (oe->numlower > 1)
  86                         type |= __OVL_PATH_MERGE;
  87         }
  88         return type;
  89 }
  90
  91 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
  92 {
  93         return lockless_dereference(oe->__upperdentry);
  94 }
  95
  96 void ovl_path_upper(struct dentry *dentry, struct path *path)
  97 {
  98         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
  99         struct ovl_entry *oe = dentry->d_fsdata;
 100
 101         path->mnt = ofs->upper_mnt;
 102         path->dentry = ovl_upperdentry_dereference(oe);
 103 }
 104
 105 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 106 {
 107         enum ovl_path_type type = ovl_path_type(dentry);
 108
 109         if (!OVL_TYPE_UPPER(type))
 110                 ovl_path_lower(dentry, path);
 111         else
 112                 ovl_path_upper(dentry, path);
 113
 114         return type;
 115 }
 116
 117 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 118 {
 119         struct ovl_entry *oe = dentry->d_fsdata;
 120
 121         return ovl_upperdentry_dereference(oe);
 122 }
 123
 124 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 125 {
 126         struct ovl_entry *oe = dentry->d_fsdata;
 127
 128         return __ovl_dentry_lower(oe);
 129 }
 130
 131 struct dentry *ovl_dentry_real(struct dentry *dentry)
 132 {
 133         struct ovl_entry *oe = dentry->d_fsdata;
 134         struct dentry *realdentry;
 135
 136         realdentry = ovl_upperdentry_dereference(oe);
 137         if (!realdentry)
 138                 realdentry = __ovl_dentry_lower(oe);
 139
 140         return realdentry;
 141 }
 142
 143 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
 144 {
 145         struct dentry *realdentry;
 146
 147         realdentry = ovl_upperdentry_dereference(oe);
 148         if (realdentry) {
 149                 *is_upper = true;
 150         } else {
 151                 realdentry = __ovl_dentry_lower(oe);
 152                 *is_upper = false;
 153         }
 154         return realdentry;
 155 }
 156
 157 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 158 {
 159         struct ovl_entry *oe = dentry->d_fsdata;
 160
 161         return oe->cache;
 162 }
 163
 164 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 165 {
 166         struct ovl_entry *oe = dentry->d_fsdata;
 167
 168         oe->cache = cache;
 169 }
 170
 171 void ovl_path_lower(struct dentry *dentry, struct path *path)
 172 {
 173         struct ovl_entry *oe = dentry->d_fsdata;
 174
 175         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 176 }
 177
 178 int ovl_want_write(struct dentry *dentry)
 179 {
 180         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 181         return mnt_want_write(ofs->upper_mnt);
 182 }
 183
 184 void ovl_drop_write(struct dentry *dentry)
 185 {
 186         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 187         mnt_drop_write(ofs->upper_mnt);
 188 }
 189
 190 struct dentry *ovl_workdir(struct dentry *dentry)
 191 {
 192         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 193         return ofs->workdir;
 194 }
 195
 196 bool ovl_dentry_is_opaque(struct dentry *dentry)
 197 {
 198         struct ovl_entry *oe = dentry->d_fsdata;
 199         return oe->opaque;
 200 }
 201
 202 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 203 {
 204         struct ovl_entry *oe = dentry->d_fsdata;
 205         oe->opaque = opaque;
 206 }
 207
 208 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 209 {
 210         struct ovl_entry *oe = dentry->d_fsdata;
 211
 212         WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
 213         WARN_ON(oe->__upperdentry);
 214         BUG_ON(!upperdentry->d_inode);
 215         /*
 216          * Make sure upperdentry is consistent before making it visible to
 217          * ovl_upperdentry_dereference().
 218          */
 219         smp_wmb();
 220         oe->__upperdentry = upperdentry;
 221 }
 222
 223 void ovl_dentry_version_inc(struct dentry *dentry)
 224 {
 225         struct ovl_entry *oe = dentry->d_fsdata;
 226
 227         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 228         oe->version++;
 229 }
 230
 231 u64 ovl_dentry_version_get(struct dentry *dentry)
 232 {
 233         struct ovl_entry *oe = dentry->d_fsdata;
 234
 235         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 236         return oe->version;
 237 }
 238
 239 bool ovl_is_whiteout(struct dentry *dentry)
 240 {
 241         struct inode *inode = dentry->d_inode;
 242
 243         return inode && IS_WHITEOUT(inode);
 244 }
 245
 246 static bool ovl_is_opaquedir(struct dentry *dentry)
 247 {
 248         int res;
 249         char val;
 250         struct inode *inode = dentry->d_inode;
 251
 252         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
 253                 return false;
 254
 255         res = inode->i_op->getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
 256         if (res == 1 && val == 'y')
 257                 return true;
 258
 259         return false;
 260 }
 261
 262 static void ovl_dentry_release(struct dentry *dentry)
 263 {
 264         struct ovl_entry *oe = dentry->d_fsdata;
 265
 266         if (oe) {
 267                 unsigned int i;
 268
 269                 dput(oe->__upperdentry);
 270                 for (i = 0; i < oe->numlower; i++)
 271                         dput(oe->lowerstack[i].dentry);
 272                 kfree_rcu(oe, rcu);
 273         }
 274 }
 275
 276 static const struct dentry_operations ovl_dentry_operations = {
 277         .d_release = ovl_dentry_release,
 278 };
 279
 280 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 281 {
 282         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 283         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 284
 285         if (oe)
 286                 oe->numlower = numlower;
 287
 288         return oe;
 289 }
 290
 291 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 292                                              struct qstr *name)
 293 {
 294         struct dentry *dentry;
 295
 296         mutex_lock(&dir->d_inode->i_mutex);
 297         dentry = lookup_one_len(name->name, dir, name->len);
 298         mutex_unlock(&dir->d_inode->i_mutex);
 299
 300         if (IS_ERR(dentry)) {
 301                 if (PTR_ERR(dentry) == -ENOENT)
 302                         dentry = NULL;
 303         } else if (!dentry->d_inode) {
 304                 dput(dentry);
 305                 dentry = NULL;
 306         }
 307         return dentry;
 308 }
 309
 310 /*
 311  * Returns next layer in stack starting from top.
 312  * Returns -1 if this is the last layer.
 313  */
 314 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 315 {
 316         struct ovl_entry *oe = dentry->d_fsdata;
 317
 318         BUG_ON(idx < 0);
 319         if (idx == 0) {
 320                 ovl_path_upper(dentry, path);
 321                 if (path->dentry)
 322                         return oe->numlower ? 1 : -1;
 323                 idx++;
 324         }
 325         BUG_ON(idx > oe->numlower);
 326         *path = oe->lowerstack[idx - 1];
 327
 328         return (idx < oe->numlower) ? idx + 1 : -1;
 329 }
 330
 331 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 332                           unsigned int flags)
 333 {
 334         struct ovl_entry *oe;
 335         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 336         struct path *stack = NULL;
 337         struct dentry *upperdir, *upperdentry = NULL;
 338         unsigned int ctr = 0;
 339         struct inode *inode = NULL;
 340         bool upperopaque = false;
 341         struct dentry *this, *prev = NULL;
 342         unsigned int i;
 343         int err;
 344
 345         upperdir = ovl_upperdentry_dereference(poe);
 346         if (upperdir) {
 347                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 348                 err = PTR_ERR(this);
 349                 if (IS_ERR(this))
 350                         goto out;
 351
 352                 if (this) {
 353                         if (ovl_is_whiteout(this)) {
 354                                 dput(this);
 355                                 this = NULL;
 356                                 upperopaque = true;
 357                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 358                                 upperopaque = true;
 359                         }
 360                 }
 361                 upperdentry = prev = this;
 362         }
 363
 364         if (!upperopaque && poe->numlower) {
 365                 err = -ENOMEM;
 366                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 367                 if (!stack)
 368                         goto out_put_upper;
 369         }
 370
 371         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 372                 bool opaque = false;
 373                 struct path lowerpath = poe->lowerstack[i];
 374
 375                 opaque = false;
 376                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 377                 err = PTR_ERR(this);
 378                 if (IS_ERR(this)) {
 379                         /*
 380                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 381                          */
 382                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 383                                 continue;
 384                         goto out_put;
 385                 }
 386                 if (!this)
 387                         continue;
 388                 if (ovl_is_whiteout(this)) {
 389                         dput(this);
 390                         break;
 391                 }
 392                 /*
 393                  * Only makes sense to check opaque dir if this is not the
 394                  * lowermost layer.
 395                  */
 396                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 397                         opaque = true;
 398                 /*
 399                  * If this is a non-directory then stop here.
 400                  *
 401                  * FIXME: check for opaqueness maybe better done in remove code.
 402                  */
 403                 if (!S_ISDIR(this->d_inode->i_mode)) {
 404                         opaque = true;
 405                 } else if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 406                                     !S_ISDIR(this->d_inode->i_mode))) {
 407                         if (prev == upperdentry)
 408                                 upperopaque = true;
 409                         dput(this);
 410                         break;
 411                 }
 412                 stack[ctr].dentry = this;
 413                 stack[ctr].mnt = lowerpath.mnt;
 414                 ctr++;
 415                 prev = this;
 416                 if (opaque)
 417                         break;
 418         }
 419
 420         oe = ovl_alloc_entry(ctr);
 421         err = -ENOMEM;
 422         if (!oe)
 423                 goto out_put;
 424
 425         if (upperdentry || ctr) {
 426                 struct dentry *realdentry;
 427
 428                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 429
 430                 err = -ENOMEM;
 431                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 432                                       oe);
 433                 if (!inode)
 434                         goto out_free_oe;
 435                 ovl_copyattr(realdentry->d_inode, inode);
 436         }
 437
 438         oe->opaque = upperopaque;
 439         oe->__upperdentry = upperdentry;
 440         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 441         kfree(stack);
 442         dentry->d_fsdata = oe;
 443         d_add(dentry, inode);
 444
 445         return NULL;
 446
 447 out_free_oe:
 448         kfree(oe);
 449 out_put:
 450         for (i = 0; i < ctr; i++)
 451                 dput(stack[i].dentry);
 452         kfree(stack);
 453 out_put_upper:
 454         dput(upperdentry);
 455 out:
 456         return ERR_PTR(err);
 457 }
 458
 459 struct file *ovl_path_open(struct path *path, int flags)
 460 {
 461         return dentry_open(path, flags, current_cred());
 462 }
 463
 464 static void ovl_put_super(struct super_block *sb)
 465 {
 466         struct ovl_fs *ufs = sb->s_fs_info;
 467         unsigned i;
 468
 469         dput(ufs->workdir);
 470         mntput(ufs->upper_mnt);
 471         for (i = 0; i < ufs->numlower; i++)
 472                 mntput(ufs->lower_mnt[i]);
 473
 474         kfree(ufs->config.lowerdir);
 475         kfree(ufs->config.upperdir);
 476         kfree(ufs->config.workdir);
 477         kfree(ufs);
 478 }
 479
 480 /**
 481  * ovl_statfs
 482  * @sb: The overlayfs super block
 483  * @buf: The struct kstatfs to fill in with stats
 484  *
 485  * Get the filesystem statistics.  As writes always target the upper layer
 486  * filesystem pass the statfs to the upper filesystem (if it exists)
 487  */
 488 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 489 {
 490         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 491         struct dentry *root_dentry = dentry->d_sb->s_root;
 492         struct path path;
 493         int err;
 494
 495         ovl_path_real(root_dentry, &path);
 496
 497         err = vfs_statfs(&path, buf);
 498         if (!err) {
 499                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 500                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 501         }
 502
 503         return err;
 504 }
 505
 506 /**
 507  * ovl_show_options
 508  *
 509  * Prints the mount options for a given superblock.
 510  * Returns zero; does not fail.
 511  */
 512 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 513 {
 514         struct super_block *sb = dentry->d_sb;
 515         struct ovl_fs *ufs = sb->s_fs_info;
 516
 517         seq_printf(m, ",lowerdir=%s", ufs->config.lowerdir);
 518         if (ufs->config.upperdir) {
 519                 seq_printf(m, ",upperdir=%s", ufs->config.upperdir);
 520                 seq_printf(m, ",workdir=%s", ufs->config.workdir);
 521         }
 522         return 0;
 523 }
 524
 525 static const struct super_operations ovl_super_operations = {
 526         .put_super      = ovl_put_super,
 527         .statfs         = ovl_statfs,
 528         .show_options   = ovl_show_options,
 529 };
 530
 531 enum {
 532         OPT_LOWERDIR,
 533         OPT_UPPERDIR,
 534         OPT_WORKDIR,
 535         OPT_ERR,
 536 };
 537
 538 static const match_table_t ovl_tokens = {
 539         {OPT_LOWERDIR,                  "lowerdir=%s"},
 540         {OPT_UPPERDIR,                  "upperdir=%s"},
 541         {OPT_WORKDIR,                   "workdir=%s"},
 542         {OPT_ERR,                       NULL}
 543 };
 544
 545 static char *ovl_next_opt(char **s)
 546 {
 547         char *sbegin = *s;
 548         char *p;
 549
 550         if (sbegin == NULL)
 551                 return NULL;
 552
 553         for (p = sbegin; *p; p++) {
 554                 if (*p == '\\') {
 555                         p++;
 556                         if (!*p)
 557                                 break;
 558                 } else if (*p == ',') {
 559                         *p = '\0';
 560                         *s = p + 1;
 561                         return sbegin;
 562                 }
 563         }
 564         *s = NULL;
 565         return sbegin;
 566 }
 567
 568 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 569 {
 570         char *p;
 571
 572         while ((p = ovl_next_opt(&opt)) != NULL) {
 573                 int token;
 574                 substring_t args[MAX_OPT_ARGS];
 575
 576                 if (!*p)
 577                         continue;
 578
 579                 token = match_token(p, ovl_tokens, args);
 580                 switch (token) {
 581                 case OPT_UPPERDIR:
 582                         kfree(config->upperdir);
 583                         config->upperdir = match_strdup(&args[0]);
 584                         if (!config->upperdir)
 585                                 return -ENOMEM;
 586                         break;
 587
 588                 case OPT_LOWERDIR:
 589                         kfree(config->lowerdir);
 590                         config->lowerdir = match_strdup(&args[0]);
 591                         if (!config->lowerdir)
 592                                 return -ENOMEM;
 593                         break;
 594
 595                 case OPT_WORKDIR:
 596                         kfree(config->workdir);
 597                         config->workdir = match_strdup(&args[0]);
 598                         if (!config->workdir)
 599                                 return -ENOMEM;
 600                         break;
 601
 602                 default:
 603                         return -EINVAL;
 604                 }
 605         }
 606         return 0;
 607 }
 608
 609 #define OVL_WORKDIR_NAME "work"
 610
 611 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 612                                          struct dentry *dentry)
 613 {
 614         struct inode *dir = dentry->d_inode;
 615         struct dentry *work;
 616         int err;
 617         bool retried = false;
 618
 619         err = mnt_want_write(mnt);
 620         if (err)
 621                 return ERR_PTR(err);
 622
 623         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 624 retry:
 625         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 626                               strlen(OVL_WORKDIR_NAME));
 627
 628         if (!IS_ERR(work)) {
 629                 struct kstat stat = {
 630                         .mode = S_IFDIR | 0,
 631                 };
 632
 633                 if (work->d_inode) {
 634                         err = -EEXIST;
 635                         if (retried)
 636                                 goto out_dput;
 637
 638                         retried = true;
 639                         ovl_cleanup(dir, work);
 640                         dput(work);
 641                         goto retry;
 642                 }
 643
 644                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 645                 if (err)
 646                         goto out_dput;
 647         }
 648 out_unlock:
 649         mutex_unlock(&dir->i_mutex);
 650         mnt_drop_write(mnt);
 651
 652         return work;
 653
 654 out_dput:
 655         dput(work);
 656         work = ERR_PTR(err);
 657         goto out_unlock;
 658 }
 659
 660 static void ovl_unescape(char *s)
 661 {
 662         char *d = s;
 663
 664         for (;; s++, d++) {
 665                 if (*s == '\\')
 666                         s++;
 667                 *d = *s;
 668                 if (!*s)
 669                         break;
 670         }
 671 }
 672
 673 static bool ovl_is_allowed_fs_type(struct dentry *root)
 674 {
 675         const struct dentry_operations *dop = root->d_op;
 676
 677         /*
 678          * We don't support:
 679          *  - automount filesystems
 680          *  - filesystems with revalidate (FIXME for lower layer)
 681          *  - filesystems with case insensitive names
 682          */
 683         if (dop &&
 684             (dop->d_manage || dop->d_automount ||
 685              dop->d_revalidate || dop->d_weak_revalidate ||
 686              dop->d_compare || dop->d_hash)) {
 687                 return false;
 688         }
 689         return true;
 690 }
 691
 692 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 693 {
 694         int err = -EINVAL;
 695
 696         if (!*name) {
 697                 pr_err("overlayfs: empty lowerdir\n");
 698                 goto out;
 699         }
 700         err = kern_path(name, LOOKUP_FOLLOW, path);
 701         if (err) {
 702                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 703                 goto out;
 704         }
 705         err = -EINVAL;
 706         if (!ovl_is_allowed_fs_type(path->dentry)) {
 707                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 708                 goto out_put;
 709         }
 710         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 711                 pr_err("overlayfs: '%s' not a directory\n", name);
 712                 goto out_put;
 713         }
 714         return 0;
 715
 716 out_put:
 717         path_put(path);
 718 out:
 719         return err;
 720 }
 721
 722 static int ovl_mount_dir(const char *name, struct path *path)
 723 {
 724         int err = -ENOMEM;
 725         char *tmp = kstrdup(name, GFP_KERNEL);
 726
 727         if (tmp) {
 728                 ovl_unescape(tmp);
 729                 err = ovl_mount_dir_noesc(tmp, path);
 730                 kfree(tmp);
 731         }
 732         return err;
 733 }
 734
 735 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 736                          int *stack_depth)
 737 {
 738         int err;
 739         struct kstatfs statfs;
 740
 741         err = ovl_mount_dir_noesc(name, path);
 742         if (err)
 743                 goto out;
 744
 745         err = vfs_statfs(path, &statfs);
 746         if (err) {
 747                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 748                 goto out_put;
 749         }
 750         *namelen = max(*namelen, statfs.f_namelen);
 751         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 752
 753         return 0;
 754
 755 out_put:
 756         path_put(path);
 757 out:
 758         return err;
 759 }
 760
 761 /* Workdir should not be subdir of upperdir and vice versa */
 762 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 763 {
 764         bool ok = false;
 765
 766         if (workdir != upperdir) {
 767                 ok = (lock_rename(workdir, upperdir) == NULL);
 768                 unlock_rename(workdir, upperdir);
 769         }
 770         return ok;
 771 }
 772
 773 static unsigned int ovl_split_lowerdirs(char *str)
 774 {
 775         unsigned int ctr = 1;
 776         char *s, *d;
 777
 778         for (s = d = str;; s++, d++) {
 779                 if (*s == '\\') {
 780                         s++;
 781                 } else if (*s == ':') {
 782                         *d = '\0';
 783                         ctr++;
 784                         continue;
 785                 }
 786                 *d = *s;
 787                 if (!*s)
 788                         break;
 789         }
 790         return ctr;
 791 }
 792
 793 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 794 {
 795         struct path upperpath = { NULL, NULL };
 796         struct path workpath = { NULL, NULL };
 797         struct dentry *root_dentry;
 798         struct ovl_entry *oe;
 799         struct ovl_fs *ufs;
 800         struct path *stack = NULL;
 801         char *lowertmp;
 802         char *lower;
 803         unsigned int numlower;
 804         unsigned int stacklen = 0;
 805         unsigned int i;
 806         int err;
 807
 808         err = -ENOMEM;
 809         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 810         if (!ufs)
 811                 goto out;
 812
 813         err = ovl_parse_opt((char *) data, &ufs->config);
 814         if (err)
 815                 goto out_free_config;
 816
 817         err = -EINVAL;
 818         if (!ufs->config.lowerdir) {
 819                 pr_err("overlayfs: missing 'lowerdir'\n");
 820                 goto out_free_config;
 821         }
 822
 823         sb->s_stack_depth = 0;
 824         if (ufs->config.upperdir) {
 825                 /* FIXME: workdir is not needed for a R/O mount */
 826                 if (!ufs->config.workdir) {
 827                         pr_err("overlayfs: missing 'workdir'\n");
 828                         goto out_free_config;
 829                 }
 830
 831                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 832                 if (err)
 833                         goto out_free_config;
 834
 835                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 836                 if (err)
 837                         goto out_put_upperpath;
 838
 839                 err = -EINVAL;
 840                 if (upperpath.mnt != workpath.mnt) {
 841                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 842                         goto out_put_workpath;
 843                 }
 844                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 845                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 846                         goto out_put_workpath;
 847                 }
 848                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 849         }
 850         err = -ENOMEM;
 851         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 852         if (!lowertmp)
 853                 goto out_put_workpath;
 854
 855         err = -EINVAL;
 856         stacklen = ovl_split_lowerdirs(lowertmp);
 857         if (stacklen > OVL_MAX_STACK)
 858                 goto out_free_lowertmp;
 859
 860         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 861         if (!stack)
 862                 goto out_free_lowertmp;
 863
 864         lower = lowertmp;
 865         for (numlower = 0; numlower < stacklen; numlower++) {
 866                 err = ovl_lower_dir(lower, &stack[numlower],
 867                                     &ufs->lower_namelen, &sb->s_stack_depth);
 868                 if (err)
 869                         goto out_put_lowerpath;
 870
 871                 lower = strchr(lower, '\0') + 1;
 872         }
 873
 874         err = -EINVAL;
 875         sb->s_stack_depth++;
 876         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
 877                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
 878                 goto out_put_lowerpath;
 879         }
 880
 881         if (ufs->config.upperdir) {
 882                 ufs->upper_mnt = clone_private_mount(&upperpath);
 883                 err = PTR_ERR(ufs->upper_mnt);
 884                 if (IS_ERR(ufs->upper_mnt)) {
 885                         pr_err("overlayfs: failed to clone upperpath\n");
 886                         goto out_put_lowerpath;
 887                 }
 888
 889                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
 890                 err = PTR_ERR(ufs->workdir);
 891                 if (IS_ERR(ufs->workdir)) {
 892                         pr_err("overlayfs: failed to create directory %s/%s\n",
 893                                ufs->config.workdir, OVL_WORKDIR_NAME);
 894                         goto out_put_upper_mnt;
 895                 }
 896         }
 897
 898         err = -ENOMEM;
 899         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
 900         if (ufs->lower_mnt == NULL)
 901                 goto out_put_workdir;
 902         for (i = 0; i < numlower; i++) {
 903                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
 904
 905                 err = PTR_ERR(mnt);
 906                 if (IS_ERR(mnt)) {
 907                         pr_err("overlayfs: failed to clone lowerpath\n");
 908                         goto out_put_lower_mnt;
 909                 }
 910                 /*
 911                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
 912                  * will fail instead of modifying lower fs.
 913                  */
 914                 mnt->mnt_flags |= MNT_READONLY;
 915
 916                 ufs->lower_mnt[ufs->numlower] = mnt;
 917                 ufs->numlower++;
 918         }
 919
 920         /* If the upper fs is r/o or nonexistent, we mark overlayfs r/o too */
 921         if (!ufs->upper_mnt || (ufs->upper_mnt->mnt_sb->s_flags & MS_RDONLY))
 922                 sb->s_flags |= MS_RDONLY;
 923
 924         sb->s_d_op = &ovl_dentry_operations;
 925
 926         err = -ENOMEM;
 927         oe = ovl_alloc_entry(numlower);
 928         if (!oe)
 929                 goto out_put_lower_mnt;
 930
 931         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
 932         if (!root_dentry)
 933                 goto out_free_oe;
 934
 935         mntput(upperpath.mnt);
 936         for (i = 0; i < numlower; i++)
 937                 mntput(stack[i].mnt);
 938         path_put(&workpath);
 939         kfree(lowertmp);
 940
 941         oe->__upperdentry = upperpath.dentry;
 942         for (i = 0; i < numlower; i++) {
 943                 oe->lowerstack[i].dentry = stack[i].dentry;
 944                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
 945         }
 946
 947         root_dentry->d_fsdata = oe;
 948
 949         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
 950         sb->s_op = &ovl_super_operations;
 951         sb->s_root = root_dentry;
 952         sb->s_fs_info = ufs;
 953
 954         return 0;
 955
 956 out_free_oe:
 957         kfree(oe);
 958 out_put_lower_mnt:
 959         for (i = 0; i < ufs->numlower; i++)
 960                 mntput(ufs->lower_mnt[i]);
 961         kfree(ufs->lower_mnt);
 962 out_put_workdir:
 963         dput(ufs->workdir);
 964 out_put_upper_mnt:
 965         mntput(ufs->upper_mnt);
 966 out_put_lowerpath:
 967         for (i = 0; i < numlower; i++)
 968                 path_put(&stack[i]);
 969         kfree(stack);
 970 out_free_lowertmp:
 971         kfree(lowertmp);
 972 out_put_workpath:
 973         path_put(&workpath);
 974 out_put_upperpath:
 975         path_put(&upperpath);
 976 out_free_config:
 977         kfree(ufs->config.lowerdir);
 978         kfree(ufs->config.upperdir);
 979         kfree(ufs->config.workdir);
 980         kfree(ufs);
 981 out:
 982         return err;
 983 }
 984
 985 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
 986                                 const char *dev_name, void *raw_data)
 987 {
 988         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
 989 }
 990
 991 static struct file_system_type ovl_fs_type = {
 992         .owner          = THIS_MODULE,
 993         .name           = "overlay",
 994         .mount          = ovl_mount,
 995         .kill_sb        = kill_anon_super,
 996 };
 997 MODULE_ALIAS_FS("overlay");
 998
 999 static int __init ovl_init(void)
1000 {
1001         return register_filesystem(&ovl_fs_type);
1002 }
1003
1004 static void __exit ovl_exit(void)
1005 {
1006         unregister_filesystem(&ovl_fs_type);
1007 }
1008
1009 module_init(ovl_init);
1010 module_exit(ovl_exit);