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