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