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 int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 277 {
 278         struct ovl_entry *oe = dentry->d_fsdata;
 279         unsigned int i;
 280         int ret = 1;
 281
 282         for (i = 0; i < oe->numlower; i++) {
 283                 struct dentry *d = oe->lowerstack[i].dentry;
 284
 285                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 286                         ret = d->d_op->d_revalidate(d, flags);
 287                         if (ret < 0)
 288                                 return ret;
 289                         if (!ret) {
 290                                 if (!(flags & LOOKUP_RCU))
 291                                         d_invalidate(d);
 292                                 return -ESTALE;
 293                         }
 294                 }
 295         }
 296         return 1;
 297 }
 298
 299 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 300 {
 301         struct ovl_entry *oe = dentry->d_fsdata;
 302         unsigned int i;
 303         int ret = 1;
 304
 305         for (i = 0; i < oe->numlower; i++) {
 306                 struct dentry *d = oe->lowerstack[i].dentry;
 307
 308                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 309                         ret = d->d_op->d_weak_revalidate(d, flags);
 310                         if (ret <= 0)
 311                                 break;
 312                 }
 313         }
 314         return ret;
 315 }
 316
 317 static const struct dentry_operations ovl_dentry_operations = {
 318         .d_release = ovl_dentry_release,
 319         .d_select_inode = ovl_d_select_inode,
 320 };
 321
 322 static const struct dentry_operations ovl_reval_dentry_operations = {
 323         .d_release = ovl_dentry_release,
 324         .d_revalidate = ovl_dentry_revalidate,
 325         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 326 };
 327
 328 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 329 {
 330         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 331         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 332
 333         if (oe)
 334                 oe->numlower = numlower;
 335
 336         return oe;
 337 }
 338
 339 static bool ovl_dentry_remote(struct dentry *dentry)
 340 {
 341         return dentry->d_flags &
 342                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 343 }
 344
 345 static bool ovl_dentry_weird(struct dentry *dentry)
 346 {
 347         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 348                                   DCACHE_MANAGE_TRANSIT |
 349                                   DCACHE_OP_HASH |
 350                                   DCACHE_OP_COMPARE);
 351 }
 352
 353 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 354                                              struct qstr *name)
 355 {
 356         struct dentry *dentry;
 357
 358         mutex_lock(&dir->d_inode->i_mutex);
 359         dentry = lookup_one_len(name->name, dir, name->len);
 360         mutex_unlock(&dir->d_inode->i_mutex);
 361
 362         if (IS_ERR(dentry)) {
 363                 if (PTR_ERR(dentry) == -ENOENT)
 364                         dentry = NULL;
 365         } else if (!dentry->d_inode) {
 366                 dput(dentry);
 367                 dentry = NULL;
 368         } else if (ovl_dentry_weird(dentry)) {
 369                 dput(dentry);
 370                 /* Don't support traversing automounts and other weirdness */
 371                 dentry = ERR_PTR(-EREMOTE);
 372         }
 373         return dentry;
 374 }
 375
 376 /*
 377  * Returns next layer in stack starting from top.
 378  * Returns -1 if this is the last layer.
 379  */
 380 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 381 {
 382         struct ovl_entry *oe = dentry->d_fsdata;
 383
 384         BUG_ON(idx < 0);
 385         if (idx == 0) {
 386                 ovl_path_upper(dentry, path);
 387                 if (path->dentry)
 388                         return oe->numlower ? 1 : -1;
 389                 idx++;
 390         }
 391         BUG_ON(idx > oe->numlower);
 392         *path = oe->lowerstack[idx - 1];
 393
 394         return (idx < oe->numlower) ? idx + 1 : -1;
 395 }
 396
 397 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 398                           unsigned int flags)
 399 {
 400         struct ovl_entry *oe;
 401         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 402         struct path *stack = NULL;
 403         struct dentry *upperdir, *upperdentry = NULL;
 404         unsigned int ctr = 0;
 405         struct inode *inode = NULL;
 406         bool upperopaque = false;
 407         struct dentry *this, *prev = NULL;
 408         unsigned int i;
 409         int err;
 410
 411         upperdir = ovl_upperdentry_dereference(poe);
 412         if (upperdir) {
 413                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 414                 err = PTR_ERR(this);
 415                 if (IS_ERR(this))
 416                         goto out;
 417
 418                 if (this) {
 419                         if (unlikely(ovl_dentry_remote(this))) {
 420                                 dput(this);
 421                                 err = -EREMOTE;
 422                                 goto out;
 423                         }
 424                         if (ovl_is_whiteout(this)) {
 425                                 dput(this);
 426                                 this = NULL;
 427                                 upperopaque = true;
 428                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 429                                 upperopaque = true;
 430                         }
 431                 }
 432                 upperdentry = prev = this;
 433         }
 434
 435         if (!upperopaque && poe->numlower) {
 436                 err = -ENOMEM;
 437                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 438                 if (!stack)
 439                         goto out_put_upper;
 440         }
 441
 442         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 443                 bool opaque = false;
 444                 struct path lowerpath = poe->lowerstack[i];
 445
 446                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 447                 err = PTR_ERR(this);
 448                 if (IS_ERR(this)) {
 449                         /*
 450                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 451                          */
 452                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 453                                 continue;
 454                         goto out_put;
 455                 }
 456                 if (!this)
 457                         continue;
 458                 if (ovl_is_whiteout(this)) {
 459                         dput(this);
 460                         break;
 461                 }
 462                 /*
 463                  * Only makes sense to check opaque dir if this is not the
 464                  * lowermost layer.
 465                  */
 466                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 467                         opaque = true;
 468
 469                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 470                              !S_ISDIR(this->d_inode->i_mode))) {
 471                         /*
 472                          * FIXME: check for upper-opaqueness maybe better done
 473                          * in remove code.
 474                          */
 475                         if (prev == upperdentry)
 476                                 upperopaque = true;
 477                         dput(this);
 478                         break;
 479                 }
 480                 /*
 481                  * If this is a non-directory then stop here.
 482                  */
 483                 if (!S_ISDIR(this->d_inode->i_mode))
 484                         opaque = true;
 485
 486                 stack[ctr].dentry = this;
 487                 stack[ctr].mnt = lowerpath.mnt;
 488                 ctr++;
 489                 prev = this;
 490                 if (opaque)
 491                         break;
 492         }
 493
 494         oe = ovl_alloc_entry(ctr);
 495         err = -ENOMEM;
 496         if (!oe)
 497                 goto out_put;
 498
 499         if (upperdentry || ctr) {
 500                 struct dentry *realdentry;
 501
 502                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 503
 504                 err = -ENOMEM;
 505                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 506                                       oe);
 507                 if (!inode)
 508                         goto out_free_oe;
 509                 ovl_copyattr(realdentry->d_inode, inode);
 510         }
 511
 512         oe->opaque = upperopaque;
 513         oe->__upperdentry = upperdentry;
 514         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 515         kfree(stack);
 516         dentry->d_fsdata = oe;
 517         d_add(dentry, inode);
 518
 519         return NULL;
 520
 521 out_free_oe:
 522         kfree(oe);
 523 out_put:
 524         for (i = 0; i < ctr; i++)
 525                 dput(stack[i].dentry);
 526         kfree(stack);
 527 out_put_upper:
 528         dput(upperdentry);
 529 out:
 530         return ERR_PTR(err);
 531 }
 532
 533 struct file *ovl_path_open(struct path *path, int flags)
 534 {
 535         return dentry_open(path, flags, current_cred());
 536 }
 537
 538 static void ovl_put_super(struct super_block *sb)
 539 {
 540         struct ovl_fs *ufs = sb->s_fs_info;
 541         unsigned i;
 542
 543         dput(ufs->workdir);
 544         mntput(ufs->upper_mnt);
 545         for (i = 0; i < ufs->numlower; i++)
 546                 mntput(ufs->lower_mnt[i]);
 547         kfree(ufs->lower_mnt);
 548
 549         kfree(ufs->config.lowerdir);
 550         kfree(ufs->config.upperdir);
 551         kfree(ufs->config.workdir);
 552         kfree(ufs);
 553 }
 554
 555 /**
 556  * ovl_statfs
 557  * @sb: The overlayfs super block
 558  * @buf: The struct kstatfs to fill in with stats
 559  *
 560  * Get the filesystem statistics.  As writes always target the upper layer
 561  * filesystem pass the statfs to the upper filesystem (if it exists)
 562  */
 563 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 564 {
 565         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 566         struct dentry *root_dentry = dentry->d_sb->s_root;
 567         struct path path;
 568         int err;
 569
 570         ovl_path_real(root_dentry, &path);
 571
 572         err = vfs_statfs(&path, buf);
 573         if (!err) {
 574                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 575                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 576         }
 577
 578         return err;
 579 }
 580
 581 /**
 582  * ovl_show_options
 583  *
 584  * Prints the mount options for a given superblock.
 585  * Returns zero; does not fail.
 586  */
 587 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 588 {
 589         struct super_block *sb = dentry->d_sb;
 590         struct ovl_fs *ufs = sb->s_fs_info;
 591
 592         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 593         if (ufs->config.upperdir) {
 594                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 595                 seq_show_option(m, "workdir", ufs->config.workdir);
 596         }
 597         return 0;
 598 }
 599
 600 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 601 {
 602         struct ovl_fs *ufs = sb->s_fs_info;
 603
 604         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 605                 return -EROFS;
 606
 607         return 0;
 608 }
 609
 610 static const struct super_operations ovl_super_operations = {
 611         .put_super      = ovl_put_super,
 612         .statfs         = ovl_statfs,
 613         .show_options   = ovl_show_options,
 614         .remount_fs     = ovl_remount,
 615 };
 616
 617 enum {
 618         OPT_LOWERDIR,
 619         OPT_UPPERDIR,
 620         OPT_WORKDIR,
 621         OPT_ERR,
 622 };
 623
 624 static const match_table_t ovl_tokens = {
 625         {OPT_LOWERDIR,                  "lowerdir=%s"},
 626         {OPT_UPPERDIR,                  "upperdir=%s"},
 627         {OPT_WORKDIR,                   "workdir=%s"},
 628         {OPT_ERR,                       NULL}
 629 };
 630
 631 static char *ovl_next_opt(char **s)
 632 {
 633         char *sbegin = *s;
 634         char *p;
 635
 636         if (sbegin == NULL)
 637                 return NULL;
 638
 639         for (p = sbegin; *p; p++) {
 640                 if (*p == '\\') {
 641                         p++;
 642                         if (!*p)
 643                                 break;
 644                 } else if (*p == ',') {
 645                         *p = '\0';
 646                         *s = p + 1;
 647                         return sbegin;
 648                 }
 649         }
 650         *s = NULL;
 651         return sbegin;
 652 }
 653
 654 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 655 {
 656         char *p;
 657
 658         while ((p = ovl_next_opt(&opt)) != NULL) {
 659                 int token;
 660                 substring_t args[MAX_OPT_ARGS];
 661
 662                 if (!*p)
 663                         continue;
 664
 665                 token = match_token(p, ovl_tokens, args);
 666                 switch (token) {
 667                 case OPT_UPPERDIR:
 668                         kfree(config->upperdir);
 669                         config->upperdir = match_strdup(&args[0]);
 670                         if (!config->upperdir)
 671                                 return -ENOMEM;
 672                         break;
 673
 674                 case OPT_LOWERDIR:
 675                         kfree(config->lowerdir);
 676                         config->lowerdir = match_strdup(&args[0]);
 677                         if (!config->lowerdir)
 678                                 return -ENOMEM;
 679                         break;
 680
 681                 case OPT_WORKDIR:
 682                         kfree(config->workdir);
 683                         config->workdir = match_strdup(&args[0]);
 684                         if (!config->workdir)
 685                                 return -ENOMEM;
 686                         break;
 687
 688                 default:
 689                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 690                         return -EINVAL;
 691                 }
 692         }
 693
 694         /* Workdir is useless in non-upper mount */
 695         if (!config->upperdir && config->workdir) {
 696                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 697                         config->workdir);
 698                 kfree(config->workdir);
 699                 config->workdir = NULL;
 700         }
 701
 702         return 0;
 703 }
 704
 705 #define OVL_WORKDIR_NAME "work"
 706
 707 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 708                                          struct dentry *dentry)
 709 {
 710         struct inode *dir = dentry->d_inode;
 711         struct dentry *work;
 712         int err;
 713         bool retried = false;
 714
 715         err = mnt_want_write(mnt);
 716         if (err)
 717                 return ERR_PTR(err);
 718
 719         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 720 retry:
 721         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 722                               strlen(OVL_WORKDIR_NAME));
 723
 724         if (!IS_ERR(work)) {
 725                 struct kstat stat = {
 726                         .mode = S_IFDIR | 0,
 727                 };
 728
 729                 if (work->d_inode) {
 730                         err = -EEXIST;
 731                         if (retried)
 732                                 goto out_dput;
 733
 734                         retried = true;
 735                         ovl_cleanup(dir, work);
 736                         dput(work);
 737                         goto retry;
 738                 }
 739
 740                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 741                 if (err)
 742                         goto out_dput;
 743         }
 744 out_unlock:
 745         mutex_unlock(&dir->i_mutex);
 746         mnt_drop_write(mnt);
 747
 748         return work;
 749
 750 out_dput:
 751         dput(work);
 752         work = ERR_PTR(err);
 753         goto out_unlock;
 754 }
 755
 756 static void ovl_unescape(char *s)
 757 {
 758         char *d = s;
 759
 760         for (;; s++, d++) {
 761                 if (*s == '\\')
 762                         s++;
 763                 *d = *s;
 764                 if (!*s)
 765                         break;
 766         }
 767 }
 768
 769 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 770 {
 771         int err = -EINVAL;
 772
 773         if (!*name) {
 774                 pr_err("overlayfs: empty lowerdir\n");
 775                 goto out;
 776         }
 777         err = kern_path(name, LOOKUP_FOLLOW, path);
 778         if (err) {
 779                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 780                 goto out;
 781         }
 782         err = -EINVAL;
 783         if (ovl_dentry_weird(path->dentry)) {
 784                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 785                 goto out_put;
 786         }
 787         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 788                 pr_err("overlayfs: '%s' not a directory\n", name);
 789                 goto out_put;
 790         }
 791         return 0;
 792
 793 out_put:
 794         path_put(path);
 795 out:
 796         return err;
 797 }
 798
 799 static int ovl_mount_dir(const char *name, struct path *path)
 800 {
 801         int err = -ENOMEM;
 802         char *tmp = kstrdup(name, GFP_KERNEL);
 803
 804         if (tmp) {
 805                 ovl_unescape(tmp);
 806                 err = ovl_mount_dir_noesc(tmp, path);
 807
 808                 if (!err)
 809                         if (ovl_dentry_remote(path->dentry)) {
 810                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 811                                        tmp);
 812                                 path_put(path);
 813                                 err = -EINVAL;
 814                         }
 815                 kfree(tmp);
 816         }
 817         return err;
 818 }
 819
 820 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 821                          int *stack_depth, bool *remote)
 822 {
 823         int err;
 824         struct kstatfs statfs;
 825
 826         err = ovl_mount_dir_noesc(name, path);
 827         if (err)
 828                 goto out;
 829
 830         err = vfs_statfs(path, &statfs);
 831         if (err) {
 832                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 833                 goto out_put;
 834         }
 835         *namelen = max(*namelen, statfs.f_namelen);
 836         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 837
 838         if (ovl_dentry_remote(path->dentry))
 839                 *remote = true;
 840
 841         return 0;
 842
 843 out_put:
 844         path_put(path);
 845 out:
 846         return err;
 847 }
 848
 849 /* Workdir should not be subdir of upperdir and vice versa */
 850 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 851 {
 852         bool ok = false;
 853
 854         if (workdir != upperdir) {
 855                 ok = (lock_rename(workdir, upperdir) == NULL);
 856                 unlock_rename(workdir, upperdir);
 857         }
 858         return ok;
 859 }
 860
 861 static unsigned int ovl_split_lowerdirs(char *str)
 862 {
 863         unsigned int ctr = 1;
 864         char *s, *d;
 865
 866         for (s = d = str;; s++, d++) {
 867                 if (*s == '\\') {
 868                         s++;
 869                 } else if (*s == ':') {
 870                         *d = '\0';
 871                         ctr++;
 872                         continue;
 873                 }
 874                 *d = *s;
 875                 if (!*s)
 876                         break;
 877         }
 878         return ctr;
 879 }
 880
 881 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 882 {
 883         struct path upperpath = { NULL, NULL };
 884         struct path workpath = { NULL, NULL };
 885         struct dentry *root_dentry;
 886         struct ovl_entry *oe;
 887         struct ovl_fs *ufs;
 888         struct path *stack = NULL;
 889         char *lowertmp;
 890         char *lower;
 891         unsigned int numlower;
 892         unsigned int stacklen = 0;
 893         unsigned int i;
 894         bool remote = false;
 895         int err;
 896
 897         err = -ENOMEM;
 898         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 899         if (!ufs)
 900                 goto out;
 901
 902         err = ovl_parse_opt((char *) data, &ufs->config);
 903         if (err)
 904                 goto out_free_config;
 905
 906         err = -EINVAL;
 907         if (!ufs->config.lowerdir) {
 908                 pr_err("overlayfs: missing 'lowerdir'\n");
 909                 goto out_free_config;
 910         }
 911
 912         sb->s_stack_depth = 0;
 913         if (ufs->config.upperdir) {
 914                 if (!ufs->config.workdir) {
 915                         pr_err("overlayfs: missing 'workdir'\n");
 916                         goto out_free_config;
 917                 }
 918
 919                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 920                 if (err)
 921                         goto out_free_config;
 922
 923                 /* Upper fs should not be r/o */
 924                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 925                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 926                         err = -EINVAL;
 927                         goto out_put_upperpath;
 928                 }
 929
 930                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 931                 if (err)
 932                         goto out_put_upperpath;
 933
 934                 err = -EINVAL;
 935                 if (upperpath.mnt != workpath.mnt) {
 936                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 937                         goto out_put_workpath;
 938                 }
 939                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 940                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 941                         goto out_put_workpath;
 942                 }
 943                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 944         }
 945         err = -ENOMEM;
 946         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 947         if (!lowertmp)
 948                 goto out_put_workpath;
 949
 950         err = -EINVAL;
 951         stacklen = ovl_split_lowerdirs(lowertmp);
 952         if (stacklen > OVL_MAX_STACK) {
 953                 pr_err("overlayfs: too many lower directries, limit is %d\n",
 954                        OVL_MAX_STACK);
 955                 goto out_free_lowertmp;
 956         } else if (!ufs->config.upperdir && stacklen == 1) {
 957                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 958                 goto out_free_lowertmp;
 959         }
 960
 961         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 962         if (!stack)
 963                 goto out_free_lowertmp;
 964
 965         lower = lowertmp;
 966         for (numlower = 0; numlower < stacklen; numlower++) {
 967                 err = ovl_lower_dir(lower, &stack[numlower],
 968                                     &ufs->lower_namelen, &sb->s_stack_depth,
 969                                     &remote);
 970                 if (err)
 971                         goto out_put_lowerpath;
 972
 973                 lower = strchr(lower, '\0') + 1;
 974         }
 975
 976         err = -EINVAL;
 977         sb->s_stack_depth++;
 978         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
 979                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
 980                 goto out_put_lowerpath;
 981         }
 982
 983         if (ufs->config.upperdir) {
 984                 ufs->upper_mnt = clone_private_mount(&upperpath);
 985                 err = PTR_ERR(ufs->upper_mnt);
 986                 if (IS_ERR(ufs->upper_mnt)) {
 987                         pr_err("overlayfs: failed to clone upperpath\n");
 988                         goto out_put_lowerpath;
 989                 }
 990
 991                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
 992                 err = PTR_ERR(ufs->workdir);
 993                 if (IS_ERR(ufs->workdir)) {
 994                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 995                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
 996                         sb->s_flags |= MS_RDONLY;
 997                         ufs->workdir = NULL;
 998                 }
 999         }
1000
1001         err = -ENOMEM;
1002         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1003         if (ufs->lower_mnt == NULL)
1004                 goto out_put_workdir;
1005         for (i = 0; i < numlower; i++) {
1006                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1007
1008                 err = PTR_ERR(mnt);
1009                 if (IS_ERR(mnt)) {
1010                         pr_err("overlayfs: failed to clone lowerpath\n");
1011                         goto out_put_lower_mnt;
1012                 }
1013                 /*
1014                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1015                  * will fail instead of modifying lower fs.
1016                  */
1017                 mnt->mnt_flags |= MNT_READONLY;
1018
1019                 ufs->lower_mnt[ufs->numlower] = mnt;
1020                 ufs->numlower++;
1021         }
1022
1023         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1024         if (!ufs->upper_mnt)
1025                 sb->s_flags |= MS_RDONLY;
1026
1027         if (remote)
1028                 sb->s_d_op = &ovl_reval_dentry_operations;
1029         else
1030                 sb->s_d_op = &ovl_dentry_operations;
1031
1032         err = -ENOMEM;
1033         oe = ovl_alloc_entry(numlower);
1034         if (!oe)
1035                 goto out_put_lower_mnt;
1036
1037         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1038         if (!root_dentry)
1039                 goto out_free_oe;
1040
1041         mntput(upperpath.mnt);
1042         for (i = 0; i < numlower; i++)
1043                 mntput(stack[i].mnt);
1044         path_put(&workpath);
1045         kfree(lowertmp);
1046
1047         oe->__upperdentry = upperpath.dentry;
1048         for (i = 0; i < numlower; i++) {
1049                 oe->lowerstack[i].dentry = stack[i].dentry;
1050                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1051         }
1052         kfree(stack);
1053
1054         root_dentry->d_fsdata = oe;
1055
1056         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1057         sb->s_op = &ovl_super_operations;
1058         sb->s_root = root_dentry;
1059         sb->s_fs_info = ufs;
1060
1061         return 0;
1062
1063 out_free_oe:
1064         kfree(oe);
1065 out_put_lower_mnt:
1066         for (i = 0; i < ufs->numlower; i++)
1067                 mntput(ufs->lower_mnt[i]);
1068         kfree(ufs->lower_mnt);
1069 out_put_workdir:
1070         dput(ufs->workdir);
1071         mntput(ufs->upper_mnt);
1072 out_put_lowerpath:
1073         for (i = 0; i < numlower; i++)
1074                 path_put(&stack[i]);
1075         kfree(stack);
1076 out_free_lowertmp:
1077         kfree(lowertmp);
1078 out_put_workpath:
1079         path_put(&workpath);
1080 out_put_upperpath:
1081         path_put(&upperpath);
1082 out_free_config:
1083         kfree(ufs->config.lowerdir);
1084         kfree(ufs->config.upperdir);
1085         kfree(ufs->config.workdir);
1086         kfree(ufs);
1087 out:
1088         return err;
1089 }
1090
1091 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1092                                 const char *dev_name, void *raw_data)
1093 {
1094         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1095 }
1096
1097 static struct file_system_type ovl_fs_type = {
1098         .owner          = THIS_MODULE,
1099         .name           = "overlay",
1100         .mount          = ovl_mount,
1101         .kill_sb        = kill_anon_super,
1102 };
1103 MODULE_ALIAS_FS("overlay");
1104
1105 static int __init ovl_init(void)
1106 {
1107         return register_filesystem(&ovl_fs_type);
1108 }
1109
1110 static void __exit ovl_exit(void)
1111 {
1112         unregister_filesystem(&ovl_fs_type);
1113 }
1114
1115 module_init(ovl_init);
1116 module_exit(ovl_exit);