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/pagemap.h>
  13 #include <linux/xattr.h>
  14 #include <linux/security.h>
  15 #include <linux/mount.h>
  16 #include <linux/slab.h>
  17 #include <linux/parser.h>
  18 #include <linux/module.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/sched.h>
  21 #include <linux/statfs.h>
  22 #include <linux/seq_file.h>
  23 #include "overlayfs.h"
  24
  25 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  26 MODULE_DESCRIPTION("Overlay filesystem");
  27 MODULE_LICENSE("GPL");
  28
  29 struct ovl_config {
  30         char *lowerdir;
  31         char *upperdir;
  32         char *workdir;
  33         bool default_permissions;
  34 };
  35
  36 /* private information held for overlayfs's superblock */
  37 struct ovl_fs {
  38         struct vfsmount *upper_mnt;
  39         unsigned numlower;
  40         struct vfsmount **lower_mnt;
  41         struct dentry *workdir;
  42         long lower_namelen;
  43         /* pathnames of lower and upper dirs, for show_options */
  44         struct ovl_config config;
  45 };
  46
  47 struct ovl_dir_cache;
  48
  49 /* private information held for every overlayfs dentry */
  50 struct ovl_entry {
  51         struct dentry *__upperdentry;
  52         struct ovl_dir_cache *cache;
  53         union {
  54                 struct {
  55                         u64 version;
  56                         bool opaque;
  57                 };
  58                 struct rcu_head rcu;
  59         };
  60         unsigned numlower;
  61         struct path lowerstack[];
  62 };
  63
  64 #define OVL_MAX_STACK 500
  65
  66 static struct dentry *__ovl_dentry_lower(struct ovl_entry *oe)
  67 {
  68         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
  69 }
  70
  71 enum ovl_path_type ovl_path_type(struct dentry *dentry)
  72 {
  73         struct ovl_entry *oe = dentry->d_fsdata;
  74         enum ovl_path_type type = 0;
  75
  76         if (oe->__upperdentry) {
  77                 type = __OVL_PATH_UPPER;
  78
  79                 /*
  80                  * Non-dir dentry can hold lower dentry from previous
  81                  * location. Its purity depends only on opaque flag.
  82                  */
  83                 if (oe->numlower && S_ISDIR(dentry->d_inode->i_mode))
  84                         type |= __OVL_PATH_MERGE;
  85                 else if (!oe->opaque)
  86                         type |= __OVL_PATH_PURE;
  87         } else {
  88                 if (oe->numlower > 1)
  89                         type |= __OVL_PATH_MERGE;
  90         }
  91         return type;
  92 }
  93
  94 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
  95 {
  96         return lockless_dereference(oe->__upperdentry);
  97 }
  98
  99 void ovl_path_upper(struct dentry *dentry, struct path *path)
 100 {
 101         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 102         struct ovl_entry *oe = dentry->d_fsdata;
 103
 104         path->mnt = ofs->upper_mnt;
 105         path->dentry = ovl_upperdentry_dereference(oe);
 106 }
 107
 108 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 109 {
 110         enum ovl_path_type type = ovl_path_type(dentry);
 111
 112         if (!OVL_TYPE_UPPER(type))
 113                 ovl_path_lower(dentry, path);
 114         else
 115                 ovl_path_upper(dentry, path);
 116
 117         return type;
 118 }
 119
 120 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 121 {
 122         struct ovl_entry *oe = dentry->d_fsdata;
 123
 124         return ovl_upperdentry_dereference(oe);
 125 }
 126
 127 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 128 {
 129         struct ovl_entry *oe = dentry->d_fsdata;
 130
 131         return __ovl_dentry_lower(oe);
 132 }
 133
 134 struct dentry *ovl_dentry_real(struct dentry *dentry)
 135 {
 136         struct ovl_entry *oe = dentry->d_fsdata;
 137         struct dentry *realdentry;
 138
 139         realdentry = ovl_upperdentry_dereference(oe);
 140         if (!realdentry)
 141                 realdentry = __ovl_dentry_lower(oe);
 142
 143         return realdentry;
 144 }
 145
 146 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
 147 {
 148         struct dentry *realdentry;
 149
 150         realdentry = ovl_upperdentry_dereference(oe);
 151         if (realdentry) {
 152                 *is_upper = true;
 153         } else {
 154                 realdentry = __ovl_dentry_lower(oe);
 155                 *is_upper = false;
 156         }
 157         return realdentry;
 158 }
 159
 160 struct vfsmount *ovl_entry_mnt_real(struct ovl_entry *oe, struct inode *inode,
 161                                     bool is_upper)
 162 {
 163         if (is_upper) {
 164                 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 165
 166                 return ofs->upper_mnt;
 167         } else {
 168                 return oe->numlower ? oe->lowerstack[0].mnt : NULL;
 169         }
 170 }
 171
 172 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 173 {
 174         struct ovl_entry *oe = dentry->d_fsdata;
 175
 176         return oe->cache;
 177 }
 178
 179 bool ovl_is_default_permissions(struct inode *inode)
 180 {
 181         struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 182
 183         return ofs->config.default_permissions;
 184 }
 185
 186 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 187 {
 188         struct ovl_entry *oe = dentry->d_fsdata;
 189
 190         oe->cache = cache;
 191 }
 192
 193 void ovl_path_lower(struct dentry *dentry, struct path *path)
 194 {
 195         struct ovl_entry *oe = dentry->d_fsdata;
 196
 197         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 198 }
 199
 200 int ovl_want_write(struct dentry *dentry)
 201 {
 202         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 203         return mnt_want_write(ofs->upper_mnt);
 204 }
 205
 206 void ovl_drop_write(struct dentry *dentry)
 207 {
 208         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 209         mnt_drop_write(ofs->upper_mnt);
 210 }
 211
 212 struct dentry *ovl_workdir(struct dentry *dentry)
 213 {
 214         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 215         return ofs->workdir;
 216 }
 217
 218 bool ovl_dentry_is_opaque(struct dentry *dentry)
 219 {
 220         struct ovl_entry *oe = dentry->d_fsdata;
 221         return oe->opaque;
 222 }
 223
 224 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 225 {
 226         struct ovl_entry *oe = dentry->d_fsdata;
 227         oe->opaque = opaque;
 228 }
 229
 230 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 231 {
 232         struct ovl_entry *oe = dentry->d_fsdata;
 233
 234         WARN_ON(!inode_is_locked(upperdentry->d_parent->d_inode));
 235         WARN_ON(oe->__upperdentry);
 236         BUG_ON(!upperdentry->d_inode);
 237         /*
 238          * Make sure upperdentry is consistent before making it visible to
 239          * ovl_upperdentry_dereference().
 240          */
 241         smp_wmb();
 242         oe->__upperdentry = upperdentry;
 243 }
 244
 245 void ovl_dentry_version_inc(struct dentry *dentry)
 246 {
 247         struct ovl_entry *oe = dentry->d_fsdata;
 248
 249         WARN_ON(!inode_is_locked(dentry->d_inode));
 250         oe->version++;
 251 }
 252
 253 u64 ovl_dentry_version_get(struct dentry *dentry)
 254 {
 255         struct ovl_entry *oe = dentry->d_fsdata;
 256
 257         WARN_ON(!inode_is_locked(dentry->d_inode));
 258         return oe->version;
 259 }
 260
 261 bool ovl_is_whiteout(struct dentry *dentry)
 262 {
 263         struct inode *inode = dentry->d_inode;
 264
 265         return inode && IS_WHITEOUT(inode);
 266 }
 267
 268 static bool ovl_is_opaquedir(struct dentry *dentry)
 269 {
 270         int res;
 271         char val;
 272         struct inode *inode = dentry->d_inode;
 273
 274         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
 275                 return false;
 276
 277         res = inode->i_op->getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
 278         if (res == 1 && val == 'y')
 279                 return true;
 280
 281         return false;
 282 }
 283
 284 static void ovl_dentry_release(struct dentry *dentry)
 285 {
 286         struct ovl_entry *oe = dentry->d_fsdata;
 287
 288         if (oe) {
 289                 unsigned int i;
 290
 291                 dput(oe->__upperdentry);
 292                 for (i = 0; i < oe->numlower; i++)
 293                         dput(oe->lowerstack[i].dentry);
 294                 kfree_rcu(oe, rcu);
 295         }
 296 }
 297
 298 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 299 {
 300         struct ovl_entry *oe = dentry->d_fsdata;
 301         unsigned int i;
 302         int ret = 1;
 303
 304         for (i = 0; i < oe->numlower; i++) {
 305                 struct dentry *d = oe->lowerstack[i].dentry;
 306
 307                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 308                         ret = d->d_op->d_revalidate(d, flags);
 309                         if (ret < 0)
 310                                 return ret;
 311                         if (!ret) {
 312                                 if (!(flags & LOOKUP_RCU))
 313                                         d_invalidate(d);
 314                                 return -ESTALE;
 315                         }
 316                 }
 317         }
 318         return 1;
 319 }
 320
 321 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 322 {
 323         struct ovl_entry *oe = dentry->d_fsdata;
 324         unsigned int i;
 325         int ret = 1;
 326
 327         for (i = 0; i < oe->numlower; i++) {
 328                 struct dentry *d = oe->lowerstack[i].dentry;
 329
 330                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 331                         ret = d->d_op->d_weak_revalidate(d, flags);
 332                         if (ret <= 0)
 333                                 break;
 334                 }
 335         }
 336         return ret;
 337 }
 338
 339 static const struct dentry_operations ovl_dentry_operations = {
 340         .d_release = ovl_dentry_release,
 341         .d_select_inode = ovl_d_select_inode,
 342 };
 343
 344 static const struct dentry_operations ovl_reval_dentry_operations = {
 345         .d_release = ovl_dentry_release,
 346         .d_select_inode = ovl_d_select_inode,
 347         .d_revalidate = ovl_dentry_revalidate,
 348         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 349 };
 350
 351 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 352 {
 353         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 354         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 355
 356         if (oe)
 357                 oe->numlower = numlower;
 358
 359         return oe;
 360 }
 361
 362 static bool ovl_dentry_remote(struct dentry *dentry)
 363 {
 364         return dentry->d_flags &
 365                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 366 }
 367
 368 static bool ovl_dentry_weird(struct dentry *dentry)
 369 {
 370         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 371                                   DCACHE_MANAGE_TRANSIT |
 372                                   DCACHE_OP_HASH |
 373                                   DCACHE_OP_COMPARE);
 374 }
 375
 376 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 377                                              struct qstr *name)
 378 {
 379         struct dentry *dentry;
 380
 381         inode_lock(dir->d_inode);
 382         dentry = lookup_one_len(name->name, dir, name->len);
 383         inode_unlock(dir->d_inode);
 384
 385         if (IS_ERR(dentry)) {
 386                 if (PTR_ERR(dentry) == -ENOENT)
 387                         dentry = NULL;
 388         } else if (!dentry->d_inode) {
 389                 dput(dentry);
 390                 dentry = NULL;
 391         } else if (ovl_dentry_weird(dentry)) {
 392                 dput(dentry);
 393                 /* Don't support traversing automounts and other weirdness */
 394                 dentry = ERR_PTR(-EREMOTE);
 395         }
 396         return dentry;
 397 }
 398
 399 /*
 400  * Returns next layer in stack starting from top.
 401  * Returns -1 if this is the last layer.
 402  */
 403 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 404 {
 405         struct ovl_entry *oe = dentry->d_fsdata;
 406
 407         BUG_ON(idx < 0);
 408         if (idx == 0) {
 409                 ovl_path_upper(dentry, path);
 410                 if (path->dentry)
 411                         return oe->numlower ? 1 : -1;
 412                 idx++;
 413         }
 414         BUG_ON(idx > oe->numlower);
 415         *path = oe->lowerstack[idx - 1];
 416
 417         return (idx < oe->numlower) ? idx + 1 : -1;
 418 }
 419
 420 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 421                           unsigned int flags)
 422 {
 423         struct ovl_entry *oe;
 424         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 425         struct path *stack = NULL;
 426         struct dentry *upperdir, *upperdentry = NULL;
 427         unsigned int ctr = 0;
 428         struct inode *inode = NULL;
 429         bool upperopaque = false;
 430         struct dentry *this, *prev = NULL;
 431         unsigned int i;
 432         int err;
 433
 434         upperdir = ovl_upperdentry_dereference(poe);
 435         if (upperdir) {
 436                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 437                 err = PTR_ERR(this);
 438                 if (IS_ERR(this))
 439                         goto out;
 440
 441                 if (this) {
 442                         if (unlikely(ovl_dentry_remote(this))) {
 443                                 dput(this);
 444                                 err = -EREMOTE;
 445                                 goto out;
 446                         }
 447                         if (ovl_is_whiteout(this)) {
 448                                 dput(this);
 449                                 this = NULL;
 450                                 upperopaque = true;
 451                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 452                                 upperopaque = true;
 453                         }
 454                 }
 455                 upperdentry = prev = this;
 456         }
 457
 458         if (!upperopaque && poe->numlower) {
 459                 err = -ENOMEM;
 460                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 461                 if (!stack)
 462                         goto out_put_upper;
 463         }
 464
 465         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 466                 bool opaque = false;
 467                 struct path lowerpath = poe->lowerstack[i];
 468
 469                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 470                 err = PTR_ERR(this);
 471                 if (IS_ERR(this)) {
 472                         /*
 473                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 474                          */
 475                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 476                                 continue;
 477                         goto out_put;
 478                 }
 479                 if (!this)
 480                         continue;
 481                 if (ovl_is_whiteout(this)) {
 482                         dput(this);
 483                         break;
 484                 }
 485                 /*
 486                  * Only makes sense to check opaque dir if this is not the
 487                  * lowermost layer.
 488                  */
 489                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 490                         opaque = true;
 491
 492                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 493                              !S_ISDIR(this->d_inode->i_mode))) {
 494                         /*
 495                          * FIXME: check for upper-opaqueness maybe better done
 496                          * in remove code.
 497                          */
 498                         if (prev == upperdentry)
 499                                 upperopaque = true;
 500                         dput(this);
 501                         break;
 502                 }
 503                 /*
 504                  * If this is a non-directory then stop here.
 505                  */
 506                 if (!S_ISDIR(this->d_inode->i_mode))
 507                         opaque = true;
 508
 509                 stack[ctr].dentry = this;
 510                 stack[ctr].mnt = lowerpath.mnt;
 511                 ctr++;
 512                 prev = this;
 513                 if (opaque)
 514                         break;
 515         }
 516
 517         oe = ovl_alloc_entry(ctr);
 518         err = -ENOMEM;
 519         if (!oe)
 520                 goto out_put;
 521
 522         if (upperdentry || ctr) {
 523                 struct dentry *realdentry;
 524
 525                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 526
 527                 err = -ENOMEM;
 528                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 529                                       oe);
 530                 if (!inode)
 531                         goto out_free_oe;
 532                 ovl_copyattr(realdentry->d_inode, inode);
 533         }
 534
 535         oe->opaque = upperopaque;
 536         oe->__upperdentry = upperdentry;
 537         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 538         kfree(stack);
 539         dentry->d_fsdata = oe;
 540         d_add(dentry, inode);
 541
 542         return NULL;
 543
 544 out_free_oe:
 545         kfree(oe);
 546 out_put:
 547         for (i = 0; i < ctr; i++)
 548                 dput(stack[i].dentry);
 549         kfree(stack);
 550 out_put_upper:
 551         dput(upperdentry);
 552 out:
 553         return ERR_PTR(err);
 554 }
 555
 556 struct file *ovl_path_open(struct path *path, int flags)
 557 {
 558         return dentry_open(path, flags, current_cred());
 559 }
 560
 561 static void ovl_put_super(struct super_block *sb)
 562 {
 563         struct ovl_fs *ufs = sb->s_fs_info;
 564         unsigned i;
 565
 566         dput(ufs->workdir);
 567         mntput(ufs->upper_mnt);
 568         for (i = 0; i < ufs->numlower; i++)
 569                 mntput(ufs->lower_mnt[i]);
 570         kfree(ufs->lower_mnt);
 571
 572         kfree(ufs->config.lowerdir);
 573         kfree(ufs->config.upperdir);
 574         kfree(ufs->config.workdir);
 575         kfree(ufs);
 576 }
 577
 578 /**
 579  * ovl_statfs
 580  * @sb: The overlayfs super block
 581  * @buf: The struct kstatfs to fill in with stats
 582  *
 583  * Get the filesystem statistics.  As writes always target the upper layer
 584  * filesystem pass the statfs to the upper filesystem (if it exists)
 585  */
 586 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 587 {
 588         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 589         struct dentry *root_dentry = dentry->d_sb->s_root;
 590         struct path path;
 591         int err;
 592
 593         ovl_path_real(root_dentry, &path);
 594
 595         err = vfs_statfs(&path, buf);
 596         if (!err) {
 597                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 598                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 599         }
 600
 601         return err;
 602 }
 603
 604 /**
 605  * ovl_show_options
 606  *
 607  * Prints the mount options for a given superblock.
 608  * Returns zero; does not fail.
 609  */
 610 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 611 {
 612         struct super_block *sb = dentry->d_sb;
 613         struct ovl_fs *ufs = sb->s_fs_info;
 614
 615         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 616         if (ufs->config.upperdir) {
 617                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 618                 seq_show_option(m, "workdir", ufs->config.workdir);
 619         }
 620         if (ufs->config.default_permissions)
 621                 seq_puts(m, ",default_permissions");
 622         return 0;
 623 }
 624
 625 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 626 {
 627         struct ovl_fs *ufs = sb->s_fs_info;
 628
 629         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 630                 return -EROFS;
 631
 632         return 0;
 633 }
 634
 635 static const struct super_operations ovl_super_operations = {
 636         .put_super      = ovl_put_super,
 637         .statfs         = ovl_statfs,
 638         .show_options   = ovl_show_options,
 639         .remount_fs     = ovl_remount,
 640 };
 641
 642 enum {
 643         OPT_LOWERDIR,
 644         OPT_UPPERDIR,
 645         OPT_WORKDIR,
 646         OPT_DEFAULT_PERMISSIONS,
 647         OPT_ERR,
 648 };
 649
 650 static const match_table_t ovl_tokens = {
 651         {OPT_LOWERDIR,                  "lowerdir=%s"},
 652         {OPT_UPPERDIR,                  "upperdir=%s"},
 653         {OPT_WORKDIR,                   "workdir=%s"},
 654         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 655         {OPT_ERR,                       NULL}
 656 };
 657
 658 static char *ovl_next_opt(char **s)
 659 {
 660         char *sbegin = *s;
 661         char *p;
 662
 663         if (sbegin == NULL)
 664                 return NULL;
 665
 666         for (p = sbegin; *p; p++) {
 667                 if (*p == '\\') {
 668                         p++;
 669                         if (!*p)
 670                                 break;
 671                 } else if (*p == ',') {
 672                         *p = '\0';
 673                         *s = p + 1;
 674                         return sbegin;
 675                 }
 676         }
 677         *s = NULL;
 678         return sbegin;
 679 }
 680
 681 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 682 {
 683         char *p;
 684
 685         while ((p = ovl_next_opt(&opt)) != NULL) {
 686                 int token;
 687                 substring_t args[MAX_OPT_ARGS];
 688
 689                 if (!*p)
 690                         continue;
 691
 692                 token = match_token(p, ovl_tokens, args);
 693                 switch (token) {
 694                 case OPT_UPPERDIR:
 695                         kfree(config->upperdir);
 696                         config->upperdir = match_strdup(&args[0]);
 697                         if (!config->upperdir)
 698                                 return -ENOMEM;
 699                         break;
 700
 701                 case OPT_LOWERDIR:
 702                         kfree(config->lowerdir);
 703                         config->lowerdir = match_strdup(&args[0]);
 704                         if (!config->lowerdir)
 705                                 return -ENOMEM;
 706                         break;
 707
 708                 case OPT_WORKDIR:
 709                         kfree(config->workdir);
 710                         config->workdir = match_strdup(&args[0]);
 711                         if (!config->workdir)
 712                                 return -ENOMEM;
 713                         break;
 714
 715                 case OPT_DEFAULT_PERMISSIONS:
 716                         config->default_permissions = true;
 717                         break;
 718
 719                 default:
 720                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 721                         return -EINVAL;
 722                 }
 723         }
 724
 725         /* Workdir is useless in non-upper mount */
 726         if (!config->upperdir && config->workdir) {
 727                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 728                         config->workdir);
 729                 kfree(config->workdir);
 730                 config->workdir = NULL;
 731         }
 732
 733         return 0;
 734 }
 735
 736 #define OVL_WORKDIR_NAME "work"
 737
 738 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 739                                          struct dentry *dentry)
 740 {
 741         struct inode *dir = dentry->d_inode;
 742         struct dentry *work;
 743         int err;
 744         bool retried = false;
 745
 746         err = mnt_want_write(mnt);
 747         if (err)
 748                 return ERR_PTR(err);
 749
 750         inode_lock_nested(dir, I_MUTEX_PARENT);
 751 retry:
 752         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 753                               strlen(OVL_WORKDIR_NAME));
 754
 755         if (!IS_ERR(work)) {
 756                 struct kstat stat = {
 757                         .mode = S_IFDIR | 0,
 758                 };
 759
 760                 if (work->d_inode) {
 761                         err = -EEXIST;
 762                         if (retried)
 763                                 goto out_dput;
 764
 765                         retried = true;
 766                         ovl_cleanup(dir, work);
 767                         dput(work);
 768                         goto retry;
 769                 }
 770
 771                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 772                 if (err)
 773                         goto out_dput;
 774         }
 775 out_unlock:
 776         inode_unlock(dir);
 777         mnt_drop_write(mnt);
 778
 779         return work;
 780
 781 out_dput:
 782         dput(work);
 783         work = ERR_PTR(err);
 784         goto out_unlock;
 785 }
 786
 787 static void ovl_unescape(char *s)
 788 {
 789         char *d = s;
 790
 791         for (;; s++, d++) {
 792                 if (*s == '\\')
 793                         s++;
 794                 *d = *s;
 795                 if (!*s)
 796                         break;
 797         }
 798 }
 799
 800 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 801 {
 802         int err = -EINVAL;
 803
 804         if (!*name) {
 805                 pr_err("overlayfs: empty lowerdir\n");
 806                 goto out;
 807         }
 808         err = kern_path(name, LOOKUP_FOLLOW, path);
 809         if (err) {
 810                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 811                 goto out;
 812         }
 813         err = -EINVAL;
 814         if (ovl_dentry_weird(path->dentry)) {
 815                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 816                 goto out_put;
 817         }
 818         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 819                 pr_err("overlayfs: '%s' not a directory\n", name);
 820                 goto out_put;
 821         }
 822         return 0;
 823
 824 out_put:
 825         path_put(path);
 826 out:
 827         return err;
 828 }
 829
 830 static int ovl_mount_dir(const char *name, struct path *path)
 831 {
 832         int err = -ENOMEM;
 833         char *tmp = kstrdup(name, GFP_KERNEL);
 834
 835         if (tmp) {
 836                 ovl_unescape(tmp);
 837                 err = ovl_mount_dir_noesc(tmp, path);
 838
 839                 if (!err)
 840                         if (ovl_dentry_remote(path->dentry)) {
 841                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 842                                        tmp);
 843                                 path_put(path);
 844                                 err = -EINVAL;
 845                         }
 846                 kfree(tmp);
 847         }
 848         return err;
 849 }
 850
 851 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 852                          int *stack_depth, bool *remote)
 853 {
 854         int err;
 855         struct kstatfs statfs;
 856
 857         err = ovl_mount_dir_noesc(name, path);
 858         if (err)
 859                 goto out;
 860
 861         err = vfs_statfs(path, &statfs);
 862         if (err) {
 863                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 864                 goto out_put;
 865         }
 866         *namelen = max(*namelen, statfs.f_namelen);
 867         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 868
 869         if (ovl_dentry_remote(path->dentry))
 870                 *remote = true;
 871
 872         return 0;
 873
 874 out_put:
 875         path_put(path);
 876 out:
 877         return err;
 878 }
 879
 880 /* Workdir should not be subdir of upperdir and vice versa */
 881 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 882 {
 883         bool ok = false;
 884
 885         if (workdir != upperdir) {
 886                 ok = (lock_rename(workdir, upperdir) == NULL);
 887                 unlock_rename(workdir, upperdir);
 888         }
 889         return ok;
 890 }
 891
 892 static unsigned int ovl_split_lowerdirs(char *str)
 893 {
 894         unsigned int ctr = 1;
 895         char *s, *d;
 896
 897         for (s = d = str;; s++, d++) {
 898                 if (*s == '\\') {
 899                         s++;
 900                 } else if (*s == ':') {
 901                         *d = '\0';
 902                         ctr++;
 903                         continue;
 904                 }
 905                 *d = *s;
 906                 if (!*s)
 907                         break;
 908         }
 909         return ctr;
 910 }
 911
 912 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 913 {
 914         struct path upperpath = { NULL, NULL };
 915         struct path workpath = { NULL, NULL };
 916         struct dentry *root_dentry;
 917         struct ovl_entry *oe;
 918         struct ovl_fs *ufs;
 919         struct path *stack = NULL;
 920         char *lowertmp;
 921         char *lower;
 922         unsigned int numlower;
 923         unsigned int stacklen = 0;
 924         unsigned int i;
 925         bool remote = false;
 926         int err;
 927
 928         err = -ENOMEM;
 929         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 930         if (!ufs)
 931                 goto out;
 932
 933         err = ovl_parse_opt((char *) data, &ufs->config);
 934         if (err)
 935                 goto out_free_config;
 936
 937         err = -EINVAL;
 938         if (!ufs->config.lowerdir) {
 939                 pr_err("overlayfs: missing 'lowerdir'\n");
 940                 goto out_free_config;
 941         }
 942
 943         sb->s_stack_depth = 0;
 944         sb->s_maxbytes = MAX_LFS_FILESIZE;
 945         if (ufs->config.upperdir) {
 946                 if (!ufs->config.workdir) {
 947                         pr_err("overlayfs: missing 'workdir'\n");
 948                         goto out_free_config;
 949                 }
 950
 951                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 952                 if (err)
 953                         goto out_free_config;
 954
 955                 /* Upper fs should not be r/o */
 956                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 957                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 958                         err = -EINVAL;
 959                         goto out_put_upperpath;
 960                 }
 961
 962                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 963                 if (err)
 964                         goto out_put_upperpath;
 965
 966                 err = -EINVAL;
 967                 if (upperpath.mnt != workpath.mnt) {
 968                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 969                         goto out_put_workpath;
 970                 }
 971                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 972                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 973                         goto out_put_workpath;
 974                 }
 975                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 976         }
 977         err = -ENOMEM;
 978         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 979         if (!lowertmp)
 980                 goto out_put_workpath;
 981
 982         err = -EINVAL;
 983         stacklen = ovl_split_lowerdirs(lowertmp);
 984         if (stacklen > OVL_MAX_STACK) {
 985                 pr_err("overlayfs: too many lower directries, limit is %d\n",
 986                        OVL_MAX_STACK);
 987                 goto out_free_lowertmp;
 988         } else if (!ufs->config.upperdir && stacklen == 1) {
 989                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 990                 goto out_free_lowertmp;
 991         }
 992
 993         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 994         if (!stack)
 995                 goto out_free_lowertmp;
 996
 997         lower = lowertmp;
 998         for (numlower = 0; numlower < stacklen; numlower++) {
 999                 err = ovl_lower_dir(lower, &stack[numlower],
1000                                     &ufs->lower_namelen, &sb->s_stack_depth,
1001                                     &remote);
1002                 if (err)
1003                         goto out_put_lowerpath;
1004
1005                 lower = strchr(lower, '\0') + 1;
1006         }
1007
1008         err = -EINVAL;
1009         sb->s_stack_depth++;
1010         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1011                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1012                 goto out_put_lowerpath;
1013         }
1014
1015         if (ufs->config.upperdir) {
1016                 ufs->upper_mnt = clone_private_mount(&upperpath);
1017                 err = PTR_ERR(ufs->upper_mnt);
1018                 if (IS_ERR(ufs->upper_mnt)) {
1019                         pr_err("overlayfs: failed to clone upperpath\n");
1020                         goto out_put_lowerpath;
1021                 }
1022
1023                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
1024                 err = PTR_ERR(ufs->workdir);
1025                 if (IS_ERR(ufs->workdir)) {
1026                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1027                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1028                         sb->s_flags |= MS_RDONLY;
1029                         ufs->workdir = NULL;
1030                 }
1031         }
1032
1033         err = -ENOMEM;
1034         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1035         if (ufs->lower_mnt == NULL)
1036                 goto out_put_workdir;
1037         for (i = 0; i < numlower; i++) {
1038                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1039
1040                 err = PTR_ERR(mnt);
1041                 if (IS_ERR(mnt)) {
1042                         pr_err("overlayfs: failed to clone lowerpath\n");
1043                         goto out_put_lower_mnt;
1044                 }
1045                 /*
1046                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1047                  * will fail instead of modifying lower fs.
1048                  */
1049                 mnt->mnt_flags |= MNT_READONLY;
1050
1051                 ufs->lower_mnt[ufs->numlower] = mnt;
1052                 ufs->numlower++;
1053         }
1054
1055         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1056         if (!ufs->upper_mnt)
1057                 sb->s_flags |= MS_RDONLY;
1058
1059         if (remote)
1060                 sb->s_d_op = &ovl_reval_dentry_operations;
1061         else
1062                 sb->s_d_op = &ovl_dentry_operations;
1063
1064         err = -ENOMEM;
1065         oe = ovl_alloc_entry(numlower);
1066         if (!oe)
1067                 goto out_put_lower_mnt;
1068
1069         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1070         if (!root_dentry)
1071                 goto out_free_oe;
1072
1073         mntput(upperpath.mnt);
1074         for (i = 0; i < numlower; i++)
1075                 mntput(stack[i].mnt);
1076         path_put(&workpath);
1077         kfree(lowertmp);
1078
1079         oe->__upperdentry = upperpath.dentry;
1080         for (i = 0; i < numlower; i++) {
1081                 oe->lowerstack[i].dentry = stack[i].dentry;
1082                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1083         }
1084         kfree(stack);
1085
1086         root_dentry->d_fsdata = oe;
1087
1088         ovl_copyattr(ovl_dentry_real(root_dentry)->d_inode,
1089                      root_dentry->d_inode);
1090
1091         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1092         sb->s_op = &ovl_super_operations;
1093         sb->s_root = root_dentry;
1094         sb->s_fs_info = ufs;
1095
1096         return 0;
1097
1098 out_free_oe:
1099         kfree(oe);
1100 out_put_lower_mnt:
1101         for (i = 0; i < ufs->numlower; i++)
1102                 mntput(ufs->lower_mnt[i]);
1103         kfree(ufs->lower_mnt);
1104 out_put_workdir:
1105         dput(ufs->workdir);
1106         mntput(ufs->upper_mnt);
1107 out_put_lowerpath:
1108         for (i = 0; i < numlower; i++)
1109                 path_put(&stack[i]);
1110         kfree(stack);
1111 out_free_lowertmp:
1112         kfree(lowertmp);
1113 out_put_workpath:
1114         path_put(&workpath);
1115 out_put_upperpath:
1116         path_put(&upperpath);
1117 out_free_config:
1118         kfree(ufs->config.lowerdir);
1119         kfree(ufs->config.upperdir);
1120         kfree(ufs->config.workdir);
1121         kfree(ufs);
1122 out:
1123         return err;
1124 }
1125
1126 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1127                                 const char *dev_name, void *raw_data)
1128 {
1129         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1130 }
1131
1132 static struct file_system_type ovl_fs_type = {
1133         .owner          = THIS_MODULE,
1134         .name           = "overlay",
1135         .mount          = ovl_mount,
1136         .kill_sb        = kill_anon_super,
1137 };
1138 MODULE_ALIAS_FS("overlay");
1139
1140 static int __init ovl_init(void)
1141 {
1142         return register_filesystem(&ovl_fs_type);
1143 }
1144
1145 static void __exit ovl_exit(void)
1146 {
1147         unregister_filesystem(&ovl_fs_type);
1148 }
1149
1150 module_init(ovl_init);
1151 module_exit(ovl_exit);