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                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 376                 err = PTR_ERR(this);
 377                 if (IS_ERR(this)) {
 378                         /*
 379                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 380                          */
 381                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 382                                 continue;
 383                         goto out_put;
 384                 }
 385                 if (!this)
 386                         continue;
 387                 if (ovl_is_whiteout(this)) {
 388                         dput(this);
 389                         break;
 390                 }
 391                 /*
 392                  * Only makes sense to check opaque dir if this is not the
 393                  * lowermost layer.
 394                  */
 395                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 396                         opaque = true;
 397
 398                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 399                              !S_ISDIR(this->d_inode->i_mode))) {
 400                         /*
 401                          * FIXME: check for upper-opaqueness maybe better done
 402                          * in remove code.
 403                          */
 404                         if (prev == upperdentry)
 405                                 upperopaque = true;
 406                         dput(this);
 407                         break;
 408                 }
 409                 /*
 410                  * If this is a non-directory then stop here.
 411                  */
 412                 if (!S_ISDIR(this->d_inode->i_mode))
 413                         opaque = true;
 414
 415                 stack[ctr].dentry = this;
 416                 stack[ctr].mnt = lowerpath.mnt;
 417                 ctr++;
 418                 prev = this;
 419                 if (opaque)
 420                         break;
 421         }
 422
 423         oe = ovl_alloc_entry(ctr);
 424         err = -ENOMEM;
 425         if (!oe)
 426                 goto out_put;
 427
 428         if (upperdentry || ctr) {
 429                 struct dentry *realdentry;
 430
 431                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 432
 433                 err = -ENOMEM;
 434                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 435                                       oe);
 436                 if (!inode)
 437                         goto out_free_oe;
 438                 ovl_copyattr(realdentry->d_inode, inode);
 439         }
 440
 441         oe->opaque = upperopaque;
 442         oe->__upperdentry = upperdentry;
 443         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 444         kfree(stack);
 445         dentry->d_fsdata = oe;
 446         d_add(dentry, inode);
 447
 448         return NULL;
 449
 450 out_free_oe:
 451         kfree(oe);
 452 out_put:
 453         for (i = 0; i < ctr; i++)
 454                 dput(stack[i].dentry);
 455         kfree(stack);
 456 out_put_upper:
 457         dput(upperdentry);
 458 out:
 459         return ERR_PTR(err);
 460 }
 461
 462 struct file *ovl_path_open(struct path *path, int flags)
 463 {
 464         return dentry_open(path, flags, current_cred());
 465 }
 466
 467 static void ovl_put_super(struct super_block *sb)
 468 {
 469         struct ovl_fs *ufs = sb->s_fs_info;
 470         unsigned i;
 471
 472         dput(ufs->workdir);
 473         mntput(ufs->upper_mnt);
 474         for (i = 0; i < ufs->numlower; i++)
 475                 mntput(ufs->lower_mnt[i]);
 476
 477         kfree(ufs->config.lowerdir);
 478         kfree(ufs->config.upperdir);
 479         kfree(ufs->config.workdir);
 480         kfree(ufs);
 481 }
 482
 483 /**
 484  * ovl_statfs
 485  * @sb: The overlayfs super block
 486  * @buf: The struct kstatfs to fill in with stats
 487  *
 488  * Get the filesystem statistics.  As writes always target the upper layer
 489  * filesystem pass the statfs to the upper filesystem (if it exists)
 490  */
 491 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 492 {
 493         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 494         struct dentry *root_dentry = dentry->d_sb->s_root;
 495         struct path path;
 496         int err;
 497
 498         ovl_path_real(root_dentry, &path);
 499
 500         err = vfs_statfs(&path, buf);
 501         if (!err) {
 502                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 503                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 504         }
 505
 506         return err;
 507 }
 508
 509 /**
 510  * ovl_show_options
 511  *
 512  * Prints the mount options for a given superblock.
 513  * Returns zero; does not fail.
 514  */
 515 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 516 {
 517         struct super_block *sb = dentry->d_sb;
 518         struct ovl_fs *ufs = sb->s_fs_info;
 519
 520         seq_printf(m, ",lowerdir=%s", ufs->config.lowerdir);
 521         if (ufs->config.upperdir) {
 522                 seq_printf(m, ",upperdir=%s", ufs->config.upperdir);
 523                 seq_printf(m, ",workdir=%s", ufs->config.workdir);
 524         }
 525         return 0;
 526 }
 527
 528 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 529 {
 530         struct ovl_fs *ufs = sb->s_fs_info;
 531
 532         if (!(*flags & MS_RDONLY) &&
 533             (!ufs->upper_mnt || (ufs->upper_mnt->mnt_sb->s_flags & MS_RDONLY)))
 534                 return -EROFS;
 535
 536         return 0;
 537 }
 538
 539 static const struct super_operations ovl_super_operations = {
 540         .put_super      = ovl_put_super,
 541         .statfs         = ovl_statfs,
 542         .show_options   = ovl_show_options,
 543         .remount_fs     = ovl_remount,
 544 };
 545
 546 enum {
 547         OPT_LOWERDIR,
 548         OPT_UPPERDIR,
 549         OPT_WORKDIR,
 550         OPT_ERR,
 551 };
 552
 553 static const match_table_t ovl_tokens = {
 554         {OPT_LOWERDIR,                  "lowerdir=%s"},
 555         {OPT_UPPERDIR,                  "upperdir=%s"},
 556         {OPT_WORKDIR,                   "workdir=%s"},
 557         {OPT_ERR,                       NULL}
 558 };
 559
 560 static char *ovl_next_opt(char **s)
 561 {
 562         char *sbegin = *s;
 563         char *p;
 564
 565         if (sbegin == NULL)
 566                 return NULL;
 567
 568         for (p = sbegin; *p; p++) {
 569                 if (*p == '\\') {
 570                         p++;
 571                         if (!*p)
 572                                 break;
 573                 } else if (*p == ',') {
 574                         *p = '\0';
 575                         *s = p + 1;
 576                         return sbegin;
 577                 }
 578         }
 579         *s = NULL;
 580         return sbegin;
 581 }
 582
 583 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 584 {
 585         char *p;
 586
 587         while ((p = ovl_next_opt(&opt)) != NULL) {
 588                 int token;
 589                 substring_t args[MAX_OPT_ARGS];
 590
 591                 if (!*p)
 592                         continue;
 593
 594                 token = match_token(p, ovl_tokens, args);
 595                 switch (token) {
 596                 case OPT_UPPERDIR:
 597                         kfree(config->upperdir);
 598                         config->upperdir = match_strdup(&args[0]);
 599                         if (!config->upperdir)
 600                                 return -ENOMEM;
 601                         break;
 602
 603                 case OPT_LOWERDIR:
 604                         kfree(config->lowerdir);
 605                         config->lowerdir = match_strdup(&args[0]);
 606                         if (!config->lowerdir)
 607                                 return -ENOMEM;
 608                         break;
 609
 610                 case OPT_WORKDIR:
 611                         kfree(config->workdir);
 612                         config->workdir = match_strdup(&args[0]);
 613                         if (!config->workdir)
 614                                 return -ENOMEM;
 615                         break;
 616
 617                 default:
 618                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 619                         return -EINVAL;
 620                 }
 621         }
 622         return 0;
 623 }
 624
 625 #define OVL_WORKDIR_NAME "work"
 626
 627 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 628                                          struct dentry *dentry)
 629 {
 630         struct inode *dir = dentry->d_inode;
 631         struct dentry *work;
 632         int err;
 633         bool retried = false;
 634
 635         err = mnt_want_write(mnt);
 636         if (err)
 637                 return ERR_PTR(err);
 638
 639         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 640 retry:
 641         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 642                               strlen(OVL_WORKDIR_NAME));
 643
 644         if (!IS_ERR(work)) {
 645                 struct kstat stat = {
 646                         .mode = S_IFDIR | 0,
 647                 };
 648
 649                 if (work->d_inode) {
 650                         err = -EEXIST;
 651                         if (retried)
 652                                 goto out_dput;
 653
 654                         retried = true;
 655                         ovl_cleanup(dir, work);
 656                         dput(work);
 657                         goto retry;
 658                 }
 659
 660                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 661                 if (err)
 662                         goto out_dput;
 663         }
 664 out_unlock:
 665         mutex_unlock(&dir->i_mutex);
 666         mnt_drop_write(mnt);
 667
 668         return work;
 669
 670 out_dput:
 671         dput(work);
 672         work = ERR_PTR(err);
 673         goto out_unlock;
 674 }
 675
 676 static void ovl_unescape(char *s)
 677 {
 678         char *d = s;
 679
 680         for (;; s++, d++) {
 681                 if (*s == '\\')
 682                         s++;
 683                 *d = *s;
 684                 if (!*s)
 685                         break;
 686         }
 687 }
 688
 689 static bool ovl_is_allowed_fs_type(struct dentry *root)
 690 {
 691         const struct dentry_operations *dop = root->d_op;
 692
 693         /*
 694          * We don't support:
 695          *  - automount filesystems
 696          *  - filesystems with revalidate (FIXME for lower layer)
 697          *  - filesystems with case insensitive names
 698          */
 699         if (dop &&
 700             (dop->d_manage || dop->d_automount ||
 701              dop->d_revalidate || dop->d_weak_revalidate ||
 702              dop->d_compare || dop->d_hash)) {
 703                 return false;
 704         }
 705         return true;
 706 }
 707
 708 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 709 {
 710         int err = -EINVAL;
 711
 712         if (!*name) {
 713                 pr_err("overlayfs: empty lowerdir\n");
 714                 goto out;
 715         }
 716         err = kern_path(name, LOOKUP_FOLLOW, path);
 717         if (err) {
 718                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 719                 goto out;
 720         }
 721         err = -EINVAL;
 722         if (!ovl_is_allowed_fs_type(path->dentry)) {
 723                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 724                 goto out_put;
 725         }
 726         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 727                 pr_err("overlayfs: '%s' not a directory\n", name);
 728                 goto out_put;
 729         }
 730         return 0;
 731
 732 out_put:
 733         path_put(path);
 734 out:
 735         return err;
 736 }
 737
 738 static int ovl_mount_dir(const char *name, struct path *path)
 739 {
 740         int err = -ENOMEM;
 741         char *tmp = kstrdup(name, GFP_KERNEL);
 742
 743         if (tmp) {
 744                 ovl_unescape(tmp);
 745                 err = ovl_mount_dir_noesc(tmp, path);
 746                 kfree(tmp);
 747         }
 748         return err;
 749 }
 750
 751 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 752                          int *stack_depth)
 753 {
 754         int err;
 755         struct kstatfs statfs;
 756
 757         err = ovl_mount_dir_noesc(name, path);
 758         if (err)
 759                 goto out;
 760
 761         err = vfs_statfs(path, &statfs);
 762         if (err) {
 763                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 764                 goto out_put;
 765         }
 766         *namelen = max(*namelen, statfs.f_namelen);
 767         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 768
 769         return 0;
 770
 771 out_put:
 772         path_put(path);
 773 out:
 774         return err;
 775 }
 776
 777 /* Workdir should not be subdir of upperdir and vice versa */
 778 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 779 {
 780         bool ok = false;
 781
 782         if (workdir != upperdir) {
 783                 ok = (lock_rename(workdir, upperdir) == NULL);
 784                 unlock_rename(workdir, upperdir);
 785         }
 786         return ok;
 787 }
 788
 789 static unsigned int ovl_split_lowerdirs(char *str)
 790 {
 791         unsigned int ctr = 1;
 792         char *s, *d;
 793
 794         for (s = d = str;; s++, d++) {
 795                 if (*s == '\\') {
 796                         s++;
 797                 } else if (*s == ':') {
 798                         *d = '\0';
 799                         ctr++;
 800                         continue;
 801                 }
 802                 *d = *s;
 803                 if (!*s)
 804                         break;
 805         }
 806         return ctr;
 807 }
 808
 809 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 810 {
 811         struct path upperpath = { NULL, NULL };
 812         struct path workpath = { NULL, NULL };
 813         struct dentry *root_dentry;
 814         struct ovl_entry *oe;
 815         struct ovl_fs *ufs;
 816         struct path *stack = NULL;
 817         char *lowertmp;
 818         char *lower;
 819         unsigned int numlower;
 820         unsigned int stacklen = 0;
 821         unsigned int i;
 822         int err;
 823
 824         err = -ENOMEM;
 825         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 826         if (!ufs)
 827                 goto out;
 828
 829         err = ovl_parse_opt((char *) data, &ufs->config);
 830         if (err)
 831                 goto out_free_config;
 832
 833         err = -EINVAL;
 834         if (!ufs->config.lowerdir) {
 835                 pr_err("overlayfs: missing 'lowerdir'\n");
 836                 goto out_free_config;
 837         }
 838
 839         sb->s_stack_depth = 0;
 840         if (ufs->config.upperdir) {
 841                 /* FIXME: workdir is not needed for a R/O mount */
 842                 if (!ufs->config.workdir) {
 843                         pr_err("overlayfs: missing 'workdir'\n");
 844                         goto out_free_config;
 845                 }
 846
 847                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 848                 if (err)
 849                         goto out_free_config;
 850
 851                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 852                 if (err)
 853                         goto out_put_upperpath;
 854
 855                 err = -EINVAL;
 856                 if (upperpath.mnt != workpath.mnt) {
 857                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 858                         goto out_put_workpath;
 859                 }
 860                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 861                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 862                         goto out_put_workpath;
 863                 }
 864                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 865         }
 866         err = -ENOMEM;
 867         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 868         if (!lowertmp)
 869                 goto out_put_workpath;
 870
 871         err = -EINVAL;
 872         stacklen = ovl_split_lowerdirs(lowertmp);
 873         if (stacklen > OVL_MAX_STACK)
 874                 goto out_free_lowertmp;
 875
 876         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 877         if (!stack)
 878                 goto out_free_lowertmp;
 879
 880         lower = lowertmp;
 881         for (numlower = 0; numlower < stacklen; numlower++) {
 882                 err = ovl_lower_dir(lower, &stack[numlower],
 883                                     &ufs->lower_namelen, &sb->s_stack_depth);
 884                 if (err)
 885                         goto out_put_lowerpath;
 886
 887                 lower = strchr(lower, '\0') + 1;
 888         }
 889
 890         err = -EINVAL;
 891         sb->s_stack_depth++;
 892         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
 893                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
 894                 goto out_put_lowerpath;
 895         }
 896
 897         if (ufs->config.upperdir) {
 898                 ufs->upper_mnt = clone_private_mount(&upperpath);
 899                 err = PTR_ERR(ufs->upper_mnt);
 900                 if (IS_ERR(ufs->upper_mnt)) {
 901                         pr_err("overlayfs: failed to clone upperpath\n");
 902                         goto out_put_lowerpath;
 903                 }
 904
 905                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
 906                 err = PTR_ERR(ufs->workdir);
 907                 if (IS_ERR(ufs->workdir)) {
 908                         pr_err("overlayfs: failed to create directory %s/%s\n",
 909                                ufs->config.workdir, OVL_WORKDIR_NAME);
 910                         goto out_put_upper_mnt;
 911                 }
 912         }
 913
 914         err = -ENOMEM;
 915         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
 916         if (ufs->lower_mnt == NULL)
 917                 goto out_put_workdir;
 918         for (i = 0; i < numlower; i++) {
 919                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
 920
 921                 err = PTR_ERR(mnt);
 922                 if (IS_ERR(mnt)) {
 923                         pr_err("overlayfs: failed to clone lowerpath\n");
 924                         goto out_put_lower_mnt;
 925                 }
 926                 /*
 927                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
 928                  * will fail instead of modifying lower fs.
 929                  */
 930                 mnt->mnt_flags |= MNT_READONLY;
 931
 932                 ufs->lower_mnt[ufs->numlower] = mnt;
 933                 ufs->numlower++;
 934         }
 935
 936         /* If the upper fs is r/o or nonexistent, we mark overlayfs r/o too */
 937         if (!ufs->upper_mnt || (ufs->upper_mnt->mnt_sb->s_flags & MS_RDONLY))
 938                 sb->s_flags |= MS_RDONLY;
 939
 940         sb->s_d_op = &ovl_dentry_operations;
 941
 942         err = -ENOMEM;
 943         oe = ovl_alloc_entry(numlower);
 944         if (!oe)
 945                 goto out_put_lower_mnt;
 946
 947         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
 948         if (!root_dentry)
 949                 goto out_free_oe;
 950
 951         mntput(upperpath.mnt);
 952         for (i = 0; i < numlower; i++)
 953                 mntput(stack[i].mnt);
 954         path_put(&workpath);
 955         kfree(lowertmp);
 956
 957         oe->__upperdentry = upperpath.dentry;
 958         for (i = 0; i < numlower; i++) {
 959                 oe->lowerstack[i].dentry = stack[i].dentry;
 960                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
 961         }
 962
 963         root_dentry->d_fsdata = oe;
 964
 965         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
 966         sb->s_op = &ovl_super_operations;
 967         sb->s_root = root_dentry;
 968         sb->s_fs_info = ufs;
 969
 970         return 0;
 971
 972 out_free_oe:
 973         kfree(oe);
 974 out_put_lower_mnt:
 975         for (i = 0; i < ufs->numlower; i++)
 976                 mntput(ufs->lower_mnt[i]);
 977         kfree(ufs->lower_mnt);
 978 out_put_workdir:
 979         dput(ufs->workdir);
 980 out_put_upper_mnt:
 981         mntput(ufs->upper_mnt);
 982 out_put_lowerpath:
 983         for (i = 0; i < numlower; i++)
 984                 path_put(&stack[i]);
 985         kfree(stack);
 986 out_free_lowertmp:
 987         kfree(lowertmp);
 988 out_put_workpath:
 989         path_put(&workpath);
 990 out_put_upperpath:
 991         path_put(&upperpath);
 992 out_free_config:
 993         kfree(ufs->config.lowerdir);
 994         kfree(ufs->config.upperdir);
 995         kfree(ufs->config.workdir);
 996         kfree(ufs);
 997 out:
 998         return err;
 999 }
1000
1001 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1002                                 const char *dev_name, void *raw_data)
1003 {
1004         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1005 }
1006
1007 static struct file_system_type ovl_fs_type = {
1008         .owner          = THIS_MODULE,
1009         .name           = "overlay",
1010         .mount          = ovl_mount,
1011         .kill_sb        = kill_anon_super,
1012 };
1013 MODULE_ALIAS_FS("overlay");
1014
1015 static int __init ovl_init(void)
1016 {
1017         return register_filesystem(&ovl_fs_type);
1018 }
1019
1020 static void __exit ovl_exit(void)
1021 {
1022         unregister_filesystem(&ovl_fs_type);
1023 }
1024
1025 module_init(ovl_init);
1026 module_exit(ovl_exit);