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/sched.h>
  20 #include <linux/statfs.h>
  21 #include <linux/seq_file.h>
  22 #include "overlayfs.h"
  23
  24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  25 MODULE_DESCRIPTION("Overlay filesystem");
  26 MODULE_LICENSE("GPL");
  27
  28 #define OVERLAYFS_SUPER_MAGIC 0x794c7630
  29
  30 struct ovl_config {
  31         char *lowerdir;
  32         char *upperdir;
  33         char *workdir;
  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                 if (oe->numlower) {
  80                         if (S_ISDIR(dentry->d_inode->i_mode))
  81                                 type |= __OVL_PATH_MERGE;
  82                 } else if (!oe->opaque) {
  83                         type |= __OVL_PATH_PURE;
  84                 }
  85         } else {
  86                 if (oe->numlower > 1)
  87                         type |= __OVL_PATH_MERGE;
  88         }
  89         return type;
  90 }
  91
  92 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
  93 {
  94         return lockless_dereference(oe->__upperdentry);
  95 }
  96
  97 void ovl_path_upper(struct dentry *dentry, struct path *path)
  98 {
  99         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 100         struct ovl_entry *oe = dentry->d_fsdata;
 101
 102         path->mnt = ofs->upper_mnt;
 103         path->dentry = ovl_upperdentry_dereference(oe);
 104 }
 105
 106 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 107 {
 108         enum ovl_path_type type = ovl_path_type(dentry);
 109
 110         if (!OVL_TYPE_UPPER(type))
 111                 ovl_path_lower(dentry, path);
 112         else
 113                 ovl_path_upper(dentry, path);
 114
 115         return type;
 116 }
 117
 118 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 119 {
 120         struct ovl_entry *oe = dentry->d_fsdata;
 121
 122         return ovl_upperdentry_dereference(oe);
 123 }
 124
 125 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 126 {
 127         struct ovl_entry *oe = dentry->d_fsdata;
 128
 129         return __ovl_dentry_lower(oe);
 130 }
 131
 132 struct dentry *ovl_dentry_real(struct dentry *dentry)
 133 {
 134         struct ovl_entry *oe = dentry->d_fsdata;
 135         struct dentry *realdentry;
 136
 137         realdentry = ovl_upperdentry_dereference(oe);
 138         if (!realdentry)
 139                 realdentry = __ovl_dentry_lower(oe);
 140
 141         return realdentry;
 142 }
 143
 144 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
 145 {
 146         struct dentry *realdentry;
 147
 148         realdentry = ovl_upperdentry_dereference(oe);
 149         if (realdentry) {
 150                 *is_upper = true;
 151         } else {
 152                 realdentry = __ovl_dentry_lower(oe);
 153                 *is_upper = false;
 154         }
 155         return realdentry;
 156 }
 157
 158 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 159 {
 160         struct ovl_entry *oe = dentry->d_fsdata;
 161
 162         return oe->cache;
 163 }
 164
 165 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 166 {
 167         struct ovl_entry *oe = dentry->d_fsdata;
 168
 169         oe->cache = cache;
 170 }
 171
 172 void ovl_path_lower(struct dentry *dentry, struct path *path)
 173 {
 174         struct ovl_entry *oe = dentry->d_fsdata;
 175
 176         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 177 }
 178
 179 int ovl_want_write(struct dentry *dentry)
 180 {
 181         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 182         return mnt_want_write(ofs->upper_mnt);
 183 }
 184
 185 void ovl_drop_write(struct dentry *dentry)
 186 {
 187         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 188         mnt_drop_write(ofs->upper_mnt);
 189 }
 190
 191 struct dentry *ovl_workdir(struct dentry *dentry)
 192 {
 193         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 194         return ofs->workdir;
 195 }
 196
 197 bool ovl_dentry_is_opaque(struct dentry *dentry)
 198 {
 199         struct ovl_entry *oe = dentry->d_fsdata;
 200         return oe->opaque;
 201 }
 202
 203 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 204 {
 205         struct ovl_entry *oe = dentry->d_fsdata;
 206         oe->opaque = opaque;
 207 }
 208
 209 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 210 {
 211         struct ovl_entry *oe = dentry->d_fsdata;
 212
 213         WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
 214         WARN_ON(oe->__upperdentry);
 215         BUG_ON(!upperdentry->d_inode);
 216         /*
 217          * Make sure upperdentry is consistent before making it visible to
 218          * ovl_upperdentry_dereference().
 219          */
 220         smp_wmb();
 221         oe->__upperdentry = upperdentry;
 222 }
 223
 224 void ovl_dentry_version_inc(struct dentry *dentry)
 225 {
 226         struct ovl_entry *oe = dentry->d_fsdata;
 227
 228         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 229         oe->version++;
 230 }
 231
 232 u64 ovl_dentry_version_get(struct dentry *dentry)
 233 {
 234         struct ovl_entry *oe = dentry->d_fsdata;
 235
 236         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 237         return oe->version;
 238 }
 239
 240 bool ovl_is_whiteout(struct dentry *dentry)
 241 {
 242         struct inode *inode = dentry->d_inode;
 243
 244         return inode && IS_WHITEOUT(inode);
 245 }
 246
 247 static bool ovl_is_opaquedir(struct dentry *dentry)
 248 {
 249         int res;
 250         char val;
 251         struct inode *inode = dentry->d_inode;
 252
 253         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
 254                 return false;
 255
 256         res = inode->i_op->getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
 257         if (res == 1 && val == 'y')
 258                 return true;
 259
 260         return false;
 261 }
 262
 263 static void ovl_dentry_release(struct dentry *dentry)
 264 {
 265         struct ovl_entry *oe = dentry->d_fsdata;
 266
 267         if (oe) {
 268                 unsigned int i;
 269
 270                 dput(oe->__upperdentry);
 271                 for (i = 0; i < oe->numlower; i++)
 272                         dput(oe->lowerstack[i].dentry);
 273                 kfree_rcu(oe, rcu);
 274         }
 275 }
 276
 277 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 278 {
 279         struct ovl_entry *oe = dentry->d_fsdata;
 280         unsigned int i;
 281         int ret = 1;
 282
 283         for (i = 0; i < oe->numlower; i++) {
 284                 struct dentry *d = oe->lowerstack[i].dentry;
 285
 286                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 287                         ret = d->d_op->d_revalidate(d, flags);
 288                         if (ret < 0)
 289                                 return ret;
 290                         if (!ret) {
 291                                 if (!(flags & LOOKUP_RCU))
 292                                         d_invalidate(d);
 293                                 return -ESTALE;
 294                         }
 295                 }
 296         }
 297         return 1;
 298 }
 299
 300 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 301 {
 302         struct ovl_entry *oe = dentry->d_fsdata;
 303         unsigned int i;
 304         int ret = 1;
 305
 306         for (i = 0; i < oe->numlower; i++) {
 307                 struct dentry *d = oe->lowerstack[i].dentry;
 308
 309                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 310                         ret = d->d_op->d_weak_revalidate(d, flags);
 311                         if (ret <= 0)
 312                                 break;
 313                 }
 314         }
 315         return ret;
 316 }
 317
 318 static const struct dentry_operations ovl_dentry_operations = {
 319         .d_release = ovl_dentry_release,
 320         .d_select_inode = ovl_d_select_inode,
 321 };
 322
 323 static const struct dentry_operations ovl_reval_dentry_operations = {
 324         .d_release = ovl_dentry_release,
 325         .d_revalidate = ovl_dentry_revalidate,
 326         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 327 };
 328
 329 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 330 {
 331         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 332         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 333
 334         if (oe)
 335                 oe->numlower = numlower;
 336
 337         return oe;
 338 }
 339
 340 static bool ovl_dentry_remote(struct dentry *dentry)
 341 {
 342         return dentry->d_flags &
 343                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 344 }
 345
 346 static bool ovl_dentry_weird(struct dentry *dentry)
 347 {
 348         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 349                                   DCACHE_MANAGE_TRANSIT |
 350                                   DCACHE_OP_HASH |
 351                                   DCACHE_OP_COMPARE);
 352 }
 353
 354 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 355                                              struct qstr *name)
 356 {
 357         struct dentry *dentry;
 358
 359         mutex_lock(&dir->d_inode->i_mutex);
 360         dentry = lookup_one_len(name->name, dir, name->len);
 361         mutex_unlock(&dir->d_inode->i_mutex);
 362
 363         if (IS_ERR(dentry)) {
 364                 if (PTR_ERR(dentry) == -ENOENT)
 365                         dentry = NULL;
 366         } else if (!dentry->d_inode) {
 367                 dput(dentry);
 368                 dentry = NULL;
 369         } else if (ovl_dentry_weird(dentry)) {
 370                 dput(dentry);
 371                 /* Don't support traversing automounts and other weirdness */
 372                 dentry = ERR_PTR(-EREMOTE);
 373         }
 374         return dentry;
 375 }
 376
 377 /*
 378  * Returns next layer in stack starting from top.
 379  * Returns -1 if this is the last layer.
 380  */
 381 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 382 {
 383         struct ovl_entry *oe = dentry->d_fsdata;
 384
 385         BUG_ON(idx < 0);
 386         if (idx == 0) {
 387                 ovl_path_upper(dentry, path);
 388                 if (path->dentry)
 389                         return oe->numlower ? 1 : -1;
 390                 idx++;
 391         }
 392         BUG_ON(idx > oe->numlower);
 393         *path = oe->lowerstack[idx - 1];
 394
 395         return (idx < oe->numlower) ? idx + 1 : -1;
 396 }
 397
 398 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 399                           unsigned int flags)
 400 {
 401         struct ovl_entry *oe;
 402         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 403         struct path *stack = NULL;
 404         struct dentry *upperdir, *upperdentry = NULL;
 405         unsigned int ctr = 0;
 406         struct inode *inode = NULL;
 407         bool upperopaque = false;
 408         struct dentry *this, *prev = NULL;
 409         unsigned int i;
 410         int err;
 411
 412         upperdir = ovl_upperdentry_dereference(poe);
 413         if (upperdir) {
 414                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 415                 err = PTR_ERR(this);
 416                 if (IS_ERR(this))
 417                         goto out;
 418
 419                 if (this) {
 420                         if (unlikely(ovl_dentry_remote(this))) {
 421                                 dput(this);
 422                                 err = -EREMOTE;
 423                                 goto out;
 424                         }
 425                         if (ovl_is_whiteout(this)) {
 426                                 dput(this);
 427                                 this = NULL;
 428                                 upperopaque = true;
 429                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 430                                 upperopaque = true;
 431                         }
 432                 }
 433                 upperdentry = prev = this;
 434         }
 435
 436         if (!upperopaque && poe->numlower) {
 437                 err = -ENOMEM;
 438                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 439                 if (!stack)
 440                         goto out_put_upper;
 441         }
 442
 443         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 444                 bool opaque = false;
 445                 struct path lowerpath = poe->lowerstack[i];
 446
 447                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 448                 err = PTR_ERR(this);
 449                 if (IS_ERR(this)) {
 450                         /*
 451                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 452                          */
 453                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 454                                 continue;
 455                         goto out_put;
 456                 }
 457                 if (!this)
 458                         continue;
 459                 if (ovl_is_whiteout(this)) {
 460                         dput(this);
 461                         break;
 462                 }
 463                 /*
 464                  * Only makes sense to check opaque dir if this is not the
 465                  * lowermost layer.
 466                  */
 467                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 468                         opaque = true;
 469
 470                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 471                              !S_ISDIR(this->d_inode->i_mode))) {
 472                         /*
 473                          * FIXME: check for upper-opaqueness maybe better done
 474                          * in remove code.
 475                          */
 476                         if (prev == upperdentry)
 477                                 upperopaque = true;
 478                         dput(this);
 479                         break;
 480                 }
 481                 /*
 482                  * If this is a non-directory then stop here.
 483                  */
 484                 if (!S_ISDIR(this->d_inode->i_mode))
 485                         opaque = true;
 486
 487                 stack[ctr].dentry = this;
 488                 stack[ctr].mnt = lowerpath.mnt;
 489                 ctr++;
 490                 prev = this;
 491                 if (opaque)
 492                         break;
 493         }
 494
 495         oe = ovl_alloc_entry(ctr);
 496         err = -ENOMEM;
 497         if (!oe)
 498                 goto out_put;
 499
 500         if (upperdentry || ctr) {
 501                 struct dentry *realdentry;
 502
 503                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 504
 505                 err = -ENOMEM;
 506                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 507                                       oe);
 508                 if (!inode)
 509                         goto out_free_oe;
 510                 ovl_copyattr(realdentry->d_inode, inode);
 511         }
 512
 513         oe->opaque = upperopaque;
 514         oe->__upperdentry = upperdentry;
 515         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 516         kfree(stack);
 517         dentry->d_fsdata = oe;
 518         d_add(dentry, inode);
 519
 520         return NULL;
 521
 522 out_free_oe:
 523         kfree(oe);
 524 out_put:
 525         for (i = 0; i < ctr; i++)
 526                 dput(stack[i].dentry);
 527         kfree(stack);
 528 out_put_upper:
 529         dput(upperdentry);
 530 out:
 531         return ERR_PTR(err);
 532 }
 533
 534 struct file *ovl_path_open(struct path *path, int flags)
 535 {
 536         return dentry_open(path, flags, current_cred());
 537 }
 538
 539 static void ovl_put_super(struct super_block *sb)
 540 {
 541         struct ovl_fs *ufs = sb->s_fs_info;
 542         unsigned i;
 543
 544         dput(ufs->workdir);
 545         mntput(ufs->upper_mnt);
 546         for (i = 0; i < ufs->numlower; i++)
 547                 mntput(ufs->lower_mnt[i]);
 548         kfree(ufs->lower_mnt);
 549
 550         kfree(ufs->config.lowerdir);
 551         kfree(ufs->config.upperdir);
 552         kfree(ufs->config.workdir);
 553         kfree(ufs);
 554 }
 555
 556 /**
 557  * ovl_statfs
 558  * @sb: The overlayfs super block
 559  * @buf: The struct kstatfs to fill in with stats
 560  *
 561  * Get the filesystem statistics.  As writes always target the upper layer
 562  * filesystem pass the statfs to the upper filesystem (if it exists)
 563  */
 564 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 565 {
 566         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 567         struct dentry *root_dentry = dentry->d_sb->s_root;
 568         struct path path;
 569         int err;
 570
 571         ovl_path_real(root_dentry, &path);
 572
 573         err = vfs_statfs(&path, buf);
 574         if (!err) {
 575                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 576                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 577         }
 578
 579         return err;
 580 }
 581
 582 /**
 583  * ovl_show_options
 584  *
 585  * Prints the mount options for a given superblock.
 586  * Returns zero; does not fail.
 587  */
 588 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 589 {
 590         struct super_block *sb = dentry->d_sb;
 591         struct ovl_fs *ufs = sb->s_fs_info;
 592
 593         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 594         if (ufs->config.upperdir) {
 595                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 596                 seq_show_option(m, "workdir", ufs->config.workdir);
 597         }
 598         return 0;
 599 }
 600
 601 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 602 {
 603         struct ovl_fs *ufs = sb->s_fs_info;
 604
 605         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 606                 return -EROFS;
 607
 608         return 0;
 609 }
 610
 611 static const struct super_operations ovl_super_operations = {
 612         .put_super      = ovl_put_super,
 613         .statfs         = ovl_statfs,
 614         .show_options   = ovl_show_options,
 615         .remount_fs     = ovl_remount,
 616 };
 617
 618 enum {
 619         OPT_LOWERDIR,
 620         OPT_UPPERDIR,
 621         OPT_WORKDIR,
 622         OPT_ERR,
 623 };
 624
 625 static const match_table_t ovl_tokens = {
 626         {OPT_LOWERDIR,                  "lowerdir=%s"},
 627         {OPT_UPPERDIR,                  "upperdir=%s"},
 628         {OPT_WORKDIR,                   "workdir=%s"},
 629         {OPT_ERR,                       NULL}
 630 };
 631
 632 static char *ovl_next_opt(char **s)
 633 {
 634         char *sbegin = *s;
 635         char *p;
 636
 637         if (sbegin == NULL)
 638                 return NULL;
 639
 640         for (p = sbegin; *p; p++) {
 641                 if (*p == '\\') {
 642                         p++;
 643                         if (!*p)
 644                                 break;
 645                 } else if (*p == ',') {
 646                         *p = '\0';
 647                         *s = p + 1;
 648                         return sbegin;
 649                 }
 650         }
 651         *s = NULL;
 652         return sbegin;
 653 }
 654
 655 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 656 {
 657         char *p;
 658
 659         while ((p = ovl_next_opt(&opt)) != NULL) {
 660                 int token;
 661                 substring_t args[MAX_OPT_ARGS];
 662
 663                 if (!*p)
 664                         continue;
 665
 666                 token = match_token(p, ovl_tokens, args);
 667                 switch (token) {
 668                 case OPT_UPPERDIR:
 669                         kfree(config->upperdir);
 670                         config->upperdir = match_strdup(&args[0]);
 671                         if (!config->upperdir)
 672                                 return -ENOMEM;
 673                         break;
 674
 675                 case OPT_LOWERDIR:
 676                         kfree(config->lowerdir);
 677                         config->lowerdir = match_strdup(&args[0]);
 678                         if (!config->lowerdir)
 679                                 return -ENOMEM;
 680                         break;
 681
 682                 case OPT_WORKDIR:
 683                         kfree(config->workdir);
 684                         config->workdir = match_strdup(&args[0]);
 685                         if (!config->workdir)
 686                                 return -ENOMEM;
 687                         break;
 688
 689                 default:
 690                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 691                         return -EINVAL;
 692                 }
 693         }
 694
 695         /* Workdir is useless in non-upper mount */
 696         if (!config->upperdir && config->workdir) {
 697                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 698                         config->workdir);
 699                 kfree(config->workdir);
 700                 config->workdir = NULL;
 701         }
 702
 703         return 0;
 704 }
 705
 706 #define OVL_WORKDIR_NAME "work"
 707
 708 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 709                                          struct dentry *dentry)
 710 {
 711         struct inode *dir = dentry->d_inode;
 712         struct dentry *work;
 713         int err;
 714         bool retried = false;
 715
 716         err = mnt_want_write(mnt);
 717         if (err)
 718                 return ERR_PTR(err);
 719
 720         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 721 retry:
 722         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 723                               strlen(OVL_WORKDIR_NAME));
 724
 725         if (!IS_ERR(work)) {
 726                 struct kstat stat = {
 727                         .mode = S_IFDIR | 0,
 728                 };
 729
 730                 if (work->d_inode) {
 731                         err = -EEXIST;
 732                         if (retried)
 733                                 goto out_dput;
 734
 735                         retried = true;
 736                         ovl_cleanup(dir, work);
 737                         dput(work);
 738                         goto retry;
 739                 }
 740
 741                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 742                 if (err)
 743                         goto out_dput;
 744         }
 745 out_unlock:
 746         mutex_unlock(&dir->i_mutex);
 747         mnt_drop_write(mnt);
 748
 749         return work;
 750
 751 out_dput:
 752         dput(work);
 753         work = ERR_PTR(err);
 754         goto out_unlock;
 755 }
 756
 757 static void ovl_unescape(char *s)
 758 {
 759         char *d = s;
 760
 761         for (;; s++, d++) {
 762                 if (*s == '\\')
 763                         s++;
 764                 *d = *s;
 765                 if (!*s)
 766                         break;
 767         }
 768 }
 769
 770 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 771 {
 772         int err = -EINVAL;
 773
 774         if (!*name) {
 775                 pr_err("overlayfs: empty lowerdir\n");
 776                 goto out;
 777         }
 778         err = kern_path(name, LOOKUP_FOLLOW, path);
 779         if (err) {
 780                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 781                 goto out;
 782         }
 783         err = -EINVAL;
 784         if (ovl_dentry_weird(path->dentry)) {
 785                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 786                 goto out_put;
 787         }
 788         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 789                 pr_err("overlayfs: '%s' not a directory\n", name);
 790                 goto out_put;
 791         }
 792         return 0;
 793
 794 out_put:
 795         path_put(path);
 796 out:
 797         return err;
 798 }
 799
 800 static int ovl_mount_dir(const char *name, struct path *path)
 801 {
 802         int err = -ENOMEM;
 803         char *tmp = kstrdup(name, GFP_KERNEL);
 804
 805         if (tmp) {
 806                 ovl_unescape(tmp);
 807                 err = ovl_mount_dir_noesc(tmp, path);
 808
 809                 if (!err)
 810                         if (ovl_dentry_remote(path->dentry)) {
 811                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 812                                        tmp);
 813                                 path_put(path);
 814                                 err = -EINVAL;
 815                         }
 816                 kfree(tmp);
 817         }
 818         return err;
 819 }
 820
 821 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 822                          int *stack_depth, bool *remote)
 823 {
 824         int err;
 825         struct kstatfs statfs;
 826
 827         err = ovl_mount_dir_noesc(name, path);
 828         if (err)
 829                 goto out;
 830
 831         err = vfs_statfs(path, &statfs);
 832         if (err) {
 833                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 834                 goto out_put;
 835         }
 836         *namelen = max(*namelen, statfs.f_namelen);
 837         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 838
 839         if (ovl_dentry_remote(path->dentry))
 840                 *remote = true;
 841
 842         return 0;
 843
 844 out_put:
 845         path_put(path);
 846 out:
 847         return err;
 848 }
 849
 850 /* Workdir should not be subdir of upperdir and vice versa */
 851 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 852 {
 853         bool ok = false;
 854
 855         if (workdir != upperdir) {
 856                 ok = (lock_rename(workdir, upperdir) == NULL);
 857                 unlock_rename(workdir, upperdir);
 858         }
 859         return ok;
 860 }
 861
 862 static unsigned int ovl_split_lowerdirs(char *str)
 863 {
 864         unsigned int ctr = 1;
 865         char *s, *d;
 866
 867         for (s = d = str;; s++, d++) {
 868                 if (*s == '\\') {
 869                         s++;
 870                 } else if (*s == ':') {
 871                         *d = '\0';
 872                         ctr++;
 873                         continue;
 874                 }
 875                 *d = *s;
 876                 if (!*s)
 877                         break;
 878         }
 879         return ctr;
 880 }
 881
 882 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 883 {
 884         struct path upperpath = { NULL, NULL };
 885         struct path workpath = { NULL, NULL };
 886         struct dentry *root_dentry;
 887         struct ovl_entry *oe;
 888         struct ovl_fs *ufs;
 889         struct path *stack = NULL;
 890         char *lowertmp;
 891         char *lower;
 892         unsigned int numlower;
 893         unsigned int stacklen = 0;
 894         unsigned int i;
 895         bool remote = false;
 896         int err;
 897
 898         err = -ENOMEM;
 899         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 900         if (!ufs)
 901                 goto out;
 902
 903         err = ovl_parse_opt((char *) data, &ufs->config);
 904         if (err)
 905                 goto out_free_config;
 906
 907         err = -EINVAL;
 908         if (!ufs->config.lowerdir) {
 909                 pr_err("overlayfs: missing 'lowerdir'\n");
 910                 goto out_free_config;
 911         }
 912
 913         sb->s_stack_depth = 0;
 914         sb->s_maxbytes = MAX_LFS_FILESIZE;
 915         if (ufs->config.upperdir) {
 916                 if (!ufs->config.workdir) {
 917                         pr_err("overlayfs: missing 'workdir'\n");
 918                         goto out_free_config;
 919                 }
 920
 921                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 922                 if (err)
 923                         goto out_free_config;
 924
 925                 /* Upper fs should not be r/o */
 926                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 927                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 928                         err = -EINVAL;
 929                         goto out_put_upperpath;
 930                 }
 931
 932                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 933                 if (err)
 934                         goto out_put_upperpath;
 935
 936                 err = -EINVAL;
 937                 if (upperpath.mnt != workpath.mnt) {
 938                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 939                         goto out_put_workpath;
 940                 }
 941                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 942                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 943                         goto out_put_workpath;
 944                 }
 945                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 946         }
 947         err = -ENOMEM;
 948         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 949         if (!lowertmp)
 950                 goto out_put_workpath;
 951
 952         err = -EINVAL;
 953         stacklen = ovl_split_lowerdirs(lowertmp);
 954         if (stacklen > OVL_MAX_STACK) {
 955                 pr_err("overlayfs: too many lower directries, limit is %d\n",
 956                        OVL_MAX_STACK);
 957                 goto out_free_lowertmp;
 958         } else if (!ufs->config.upperdir && stacklen == 1) {
 959                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 960                 goto out_free_lowertmp;
 961         }
 962
 963         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 964         if (!stack)
 965                 goto out_free_lowertmp;
 966
 967         lower = lowertmp;
 968         for (numlower = 0; numlower < stacklen; numlower++) {
 969                 err = ovl_lower_dir(lower, &stack[numlower],
 970                                     &ufs->lower_namelen, &sb->s_stack_depth,
 971                                     &remote);
 972                 if (err)
 973                         goto out_put_lowerpath;
 974
 975                 lower = strchr(lower, '\0') + 1;
 976         }
 977
 978         err = -EINVAL;
 979         sb->s_stack_depth++;
 980         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
 981                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
 982                 goto out_put_lowerpath;
 983         }
 984
 985         if (ufs->config.upperdir) {
 986                 ufs->upper_mnt = clone_private_mount(&upperpath);
 987                 err = PTR_ERR(ufs->upper_mnt);
 988                 if (IS_ERR(ufs->upper_mnt)) {
 989                         pr_err("overlayfs: failed to clone upperpath\n");
 990                         goto out_put_lowerpath;
 991                 }
 992
 993                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
 994                 err = PTR_ERR(ufs->workdir);
 995                 if (IS_ERR(ufs->workdir)) {
 996                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 997                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
 998                         sb->s_flags |= MS_RDONLY;
 999                         ufs->workdir = NULL;
1000                 }
1001         }
1002
1003         err = -ENOMEM;
1004         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1005         if (ufs->lower_mnt == NULL)
1006                 goto out_put_workdir;
1007         for (i = 0; i < numlower; i++) {
1008                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1009
1010                 err = PTR_ERR(mnt);
1011                 if (IS_ERR(mnt)) {
1012                         pr_err("overlayfs: failed to clone lowerpath\n");
1013                         goto out_put_lower_mnt;
1014                 }
1015                 /*
1016                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1017                  * will fail instead of modifying lower fs.
1018                  */
1019                 mnt->mnt_flags |= MNT_READONLY;
1020
1021                 ufs->lower_mnt[ufs->numlower] = mnt;
1022                 ufs->numlower++;
1023         }
1024
1025         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1026         if (!ufs->upper_mnt)
1027                 sb->s_flags |= MS_RDONLY;
1028
1029         if (remote)
1030                 sb->s_d_op = &ovl_reval_dentry_operations;
1031         else
1032                 sb->s_d_op = &ovl_dentry_operations;
1033
1034         err = -ENOMEM;
1035         oe = ovl_alloc_entry(numlower);
1036         if (!oe)
1037                 goto out_put_lower_mnt;
1038
1039         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1040         if (!root_dentry)
1041                 goto out_free_oe;
1042
1043         mntput(upperpath.mnt);
1044         for (i = 0; i < numlower; i++)
1045                 mntput(stack[i].mnt);
1046         path_put(&workpath);
1047         kfree(lowertmp);
1048
1049         oe->__upperdentry = upperpath.dentry;
1050         for (i = 0; i < numlower; i++) {
1051                 oe->lowerstack[i].dentry = stack[i].dentry;
1052                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1053         }
1054         kfree(stack);
1055
1056         root_dentry->d_fsdata = oe;
1057
1058         ovl_copyattr(ovl_dentry_real(root_dentry)->d_inode,
1059                      root_dentry->d_inode);
1060
1061         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1062         sb->s_op = &ovl_super_operations;
1063         sb->s_root = root_dentry;
1064         sb->s_fs_info = ufs;
1065
1066         return 0;
1067
1068 out_free_oe:
1069         kfree(oe);
1070 out_put_lower_mnt:
1071         for (i = 0; i < ufs->numlower; i++)
1072                 mntput(ufs->lower_mnt[i]);
1073         kfree(ufs->lower_mnt);
1074 out_put_workdir:
1075         dput(ufs->workdir);
1076         mntput(ufs->upper_mnt);
1077 out_put_lowerpath:
1078         for (i = 0; i < numlower; i++)
1079                 path_put(&stack[i]);
1080         kfree(stack);
1081 out_free_lowertmp:
1082         kfree(lowertmp);
1083 out_put_workpath:
1084         path_put(&workpath);
1085 out_put_upperpath:
1086         path_put(&upperpath);
1087 out_free_config:
1088         kfree(ufs->config.lowerdir);
1089         kfree(ufs->config.upperdir);
1090         kfree(ufs->config.workdir);
1091         kfree(ufs);
1092 out:
1093         return err;
1094 }
1095
1096 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1097                                 const char *dev_name, void *raw_data)
1098 {
1099         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1100 }
1101
1102 static struct file_system_type ovl_fs_type = {
1103         .owner          = THIS_MODULE,
1104         .name           = "overlay",
1105         .mount          = ovl_mount,
1106         .kill_sb        = kill_anon_super,
1107 };
1108 MODULE_ALIAS_FS("overlay");
1109
1110 static int __init ovl_init(void)
1111 {
1112         return register_filesystem(&ovl_fs_type);
1113 }
1114
1115 static void __exit ovl_exit(void)
1116 {
1117         unregister_filesystem(&ovl_fs_type);
1118 }
1119
1120 module_init(ovl_init);
1121 module_exit(ovl_exit);