fs/overlayfs/super.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/fs.h>
  11 #include <linux/namei.h>
  12 #include <linux/xattr.h>
  13 #include <linux/security.h>
  14 #include <linux/mount.h>
  15 #include <linux/slab.h>
  16 #include <linux/parser.h>
  17 #include <linux/module.h>
  18 #include <linux/sched.h>
  19 #include <linux/statfs.h>
  20 #include <linux/seq_file.h>
  21 #include "overlayfs.h"
  22
  23 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  24 MODULE_DESCRIPTION("Overlay filesystem");
  25 MODULE_LICENSE("GPL");
  26
  27 #define OVERLAYFS_SUPER_MAGIC 0x794c7630
  28
  29 struct ovl_config {
  30         char *lowerdir;
  31         char *upperdir;
  32         char *workdir;
  33 };
  34
  35 /* private information held for overlayfs's superblock */
  36 struct ovl_fs {
  37         struct vfsmount *upper_mnt;
  38         unsigned numlower;
  39         struct vfsmount **lower_mnt;
  40         struct dentry *workdir;
  41         long lower_namelen;
  42         /* pathnames of lower and upper dirs, for show_options */
  43         struct ovl_config config;
  44 };
  45
  46 struct ovl_dir_cache;
  47
  48 /* private information held for every overlayfs dentry */
  49 struct ovl_entry {
  50         struct dentry *__upperdentry;
  51         struct ovl_dir_cache *cache;
  52         union {
  53                 struct {
  54                         u64 version;
  55                         bool opaque;
  56                 };
  57                 struct rcu_head rcu;
  58         };
  59         unsigned numlower;
  60         struct path lowerstack[];
  61 };
  62
  63 #define OVL_MAX_STACK 500
  64
  65 static struct dentry *__ovl_dentry_lower(struct ovl_entry *oe)
  66 {
  67         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
  68 }
  69
  70 enum ovl_path_type ovl_path_type(struct dentry *dentry)
  71 {
  72         struct ovl_entry *oe = dentry->d_fsdata;
  73         enum ovl_path_type type = 0;
  74
  75         if (oe->__upperdentry) {
  76                 type = __OVL_PATH_UPPER;
  77
  78                 if (oe->numlower) {
  79                         if (S_ISDIR(dentry->d_inode->i_mode))
  80                                 type |= __OVL_PATH_MERGE;
  81                 } else if (!oe->opaque) {
  82                         type |= __OVL_PATH_PURE;
  83                 }
  84         } else {
  85                 if (oe->numlower > 1)
  86                         type |= __OVL_PATH_MERGE;
  87         }
  88         return type;
  89 }
  90
  91 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
  92 {
  93         return lockless_dereference(oe->__upperdentry);
  94 }
  95
  96 void ovl_path_upper(struct dentry *dentry, struct path *path)
  97 {
  98         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
  99         struct ovl_entry *oe = dentry->d_fsdata;
 100
 101         path->mnt = ofs->upper_mnt;
 102         path->dentry = ovl_upperdentry_dereference(oe);
 103 }
 104
 105 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 106 {
 107         enum ovl_path_type type = ovl_path_type(dentry);
 108
 109         if (!OVL_TYPE_UPPER(type))
 110                 ovl_path_lower(dentry, path);
 111         else
 112                 ovl_path_upper(dentry, path);
 113
 114         return type;
 115 }
 116
 117 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 118 {
 119         struct ovl_entry *oe = dentry->d_fsdata;
 120
 121         return ovl_upperdentry_dereference(oe);
 122 }
 123
 124 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 125 {
 126         struct ovl_entry *oe = dentry->d_fsdata;
 127
 128         return __ovl_dentry_lower(oe);
 129 }
 130
 131 struct dentry *ovl_dentry_real(struct dentry *dentry)
 132 {
 133         struct ovl_entry *oe = dentry->d_fsdata;
 134         struct dentry *realdentry;
 135
 136         realdentry = ovl_upperdentry_dereference(oe);
 137         if (!realdentry)
 138                 realdentry = __ovl_dentry_lower(oe);
 139
 140         return realdentry;
 141 }
 142
 143 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
 144 {
 145         struct dentry *realdentry;
 146
 147         realdentry = ovl_upperdentry_dereference(oe);
 148         if (realdentry) {
 149                 *is_upper = true;
 150         } else {
 151                 realdentry = __ovl_dentry_lower(oe);
 152                 *is_upper = false;
 153         }
 154         return realdentry;
 155 }
 156
 157 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 158 {
 159         struct ovl_entry *oe = dentry->d_fsdata;
 160
 161         return oe->cache;
 162 }
 163
 164 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 165 {
 166         struct ovl_entry *oe = dentry->d_fsdata;
 167
 168         oe->cache = cache;
 169 }
 170
 171 void ovl_path_lower(struct dentry *dentry, struct path *path)
 172 {
 173         struct ovl_entry *oe = dentry->d_fsdata;
 174
 175         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 176 }
 177
 178 int ovl_want_write(struct dentry *dentry)
 179 {
 180         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 181         return mnt_want_write(ofs->upper_mnt);
 182 }
 183
 184 void ovl_drop_write(struct dentry *dentry)
 185 {
 186         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 187         mnt_drop_write(ofs->upper_mnt);
 188 }
 189
 190 struct dentry *ovl_workdir(struct dentry *dentry)
 191 {
 192         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 193         return ofs->workdir;
 194 }
 195
 196 bool ovl_dentry_is_opaque(struct dentry *dentry)
 197 {
 198         struct ovl_entry *oe = dentry->d_fsdata;
 199         return oe->opaque;
 200 }
 201
 202 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 203 {
 204         struct ovl_entry *oe = dentry->d_fsdata;
 205         oe->opaque = opaque;
 206 }
 207
 208 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 209 {
 210         struct ovl_entry *oe = dentry->d_fsdata;
 211
 212         WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
 213         WARN_ON(oe->__upperdentry);
 214         BUG_ON(!upperdentry->d_inode);
 215         /*
 216          * Make sure upperdentry is consistent before making it visible to
 217          * ovl_upperdentry_dereference().
 218          */
 219         smp_wmb();
 220         oe->__upperdentry = upperdentry;
 221 }
 222
 223 void ovl_dentry_version_inc(struct dentry *dentry)
 224 {
 225         struct ovl_entry *oe = dentry->d_fsdata;
 226
 227         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 228         oe->version++;
 229 }
 230
 231 u64 ovl_dentry_version_get(struct dentry *dentry)
 232 {
 233         struct ovl_entry *oe = dentry->d_fsdata;
 234
 235         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 236         return oe->version;
 237 }
 238
 239 bool ovl_is_whiteout(struct dentry *dentry)
 240 {
 241         struct inode *inode = dentry->d_inode;
 242
 243         return inode && IS_WHITEOUT(inode);
 244 }
 245
 246 static bool ovl_is_opaquedir(struct dentry *dentry)
 247 {
 248         int res;
 249         char val;
 250         struct inode *inode = dentry->d_inode;
 251
 252         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
 253                 return false;
 254
 255         res = inode->i_op->getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
 256         if (res == 1 && val == 'y')
 257                 return true;
 258
 259         return false;
 260 }
 261
 262 static void ovl_dentry_release(struct dentry *dentry)
 263 {
 264         struct ovl_entry *oe = dentry->d_fsdata;
 265
 266         if (oe) {
 267                 unsigned int i;
 268
 269                 dput(oe->__upperdentry);
 270                 for (i = 0; i < oe->numlower; i++)
 271                         dput(oe->lowerstack[i].dentry);
 272                 kfree_rcu(oe, rcu);
 273         }
 274 }
 275
 276 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 277 {
 278         struct ovl_entry *oe = dentry->d_fsdata;
 279         unsigned int i;
 280         int ret = 1;
 281
 282         for (i = 0; i < oe->numlower; i++) {
 283                 struct dentry *d = oe->lowerstack[i].dentry;
 284
 285                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 286                         ret = d->d_op->d_revalidate(d, flags);
 287                         if (ret < 0)
 288                                 return ret;
 289                         if (!ret) {
 290                                 if (!(flags & LOOKUP_RCU))
 291                                         d_invalidate(d);
 292                                 return -ESTALE;
 293                         }
 294                 }
 295         }
 296         return 1;
 297 }
 298
 299 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 300 {
 301         struct ovl_entry *oe = dentry->d_fsdata;
 302         unsigned int i;
 303         int ret = 1;
 304
 305         for (i = 0; i < oe->numlower; i++) {
 306                 struct dentry *d = oe->lowerstack[i].dentry;
 307
 308                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 309                         ret = d->d_op->d_weak_revalidate(d, flags);
 310                         if (ret <= 0)
 311                                 break;
 312                 }
 313         }
 314         return ret;
 315 }
 316
 317 static const struct dentry_operations ovl_dentry_operations = {
 318         .d_release = ovl_dentry_release,
 319         .d_select_inode = ovl_d_select_inode,
 320 };
 321
 322 static const struct dentry_operations ovl_reval_dentry_operations = {
 323         .d_release = ovl_dentry_release,
 324         .d_revalidate = ovl_dentry_revalidate,
 325         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 326 };
 327
 328 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 329 {
 330         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 331         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 332
 333         if (oe)
 334                 oe->numlower = numlower;
 335
 336         return oe;
 337 }
 338
 339 static bool ovl_dentry_remote(struct dentry *dentry)
 340 {
 341         return dentry->d_flags &
 342                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 343 }
 344
 345 static bool ovl_dentry_weird(struct dentry *dentry)
 346 {
 347         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 348                                   DCACHE_MANAGE_TRANSIT |
 349                                   DCACHE_OP_HASH |
 350                                   DCACHE_OP_COMPARE);
 351 }
 352
 353 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 354                                              struct qstr *name)
 355 {
 356         struct dentry *dentry;
 357
 358         mutex_lock(&dir->d_inode->i_mutex);
 359         dentry = lookup_one_len(name->name, dir, name->len);
 360         mutex_unlock(&dir->d_inode->i_mutex);
 361
 362         if (IS_ERR(dentry)) {
 363                 if (PTR_ERR(dentry) == -ENOENT)
 364                         dentry = NULL;
 365         } else if (!dentry->d_inode) {
 366                 dput(dentry);
 367                 dentry = NULL;
 368         } else if (ovl_dentry_weird(dentry)) {
 369                 dput(dentry);
 370                 /* Don't support traversing automounts and other weirdness */
 371                 dentry = ERR_PTR(-EREMOTE);
 372         }
 373         return dentry;
 374 }
 375
 376 /*
 377  * Returns next layer in stack starting from top.
 378  * Returns -1 if this is the last layer.
 379  */
 380 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 381 {
 382         struct ovl_entry *oe = dentry->d_fsdata;
 383
 384         BUG_ON(idx < 0);
 385         if (idx == 0) {
 386                 ovl_path_upper(dentry, path);
 387                 if (path->dentry)
 388                         return oe->numlower ? 1 : -1;
 389                 idx++;
 390         }
 391         BUG_ON(idx > oe->numlower);
 392         *path = oe->lowerstack[idx - 1];
 393
 394         return (idx < oe->numlower) ? idx + 1 : -1;
 395 }
 396
 397 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 398                           unsigned int flags)
 399 {
 400         struct ovl_entry *oe;
 401         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 402         struct path *stack = NULL;
 403         struct dentry *upperdir, *upperdentry = NULL;
 404         unsigned int ctr = 0;
 405         struct inode *inode = NULL;
 406         bool upperopaque = false;
 407         struct dentry *this, *prev = NULL;
 408         unsigned int i;
 409         int err;
 410
 411         upperdir = ovl_upperdentry_dereference(poe);
 412         if (upperdir) {
 413                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 414                 err = PTR_ERR(this);
 415                 if (IS_ERR(this))
 416                         goto out;
 417
 418                 if (this) {
 419                         if (unlikely(ovl_dentry_remote(this))) {
 420                                 dput(this);
 421                                 err = -EREMOTE;
 422                                 goto out;
 423                         }
 424                         if (ovl_is_whiteout(this)) {
 425                                 dput(this);
 426                                 this = NULL;
 427                                 upperopaque = true;
 428                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 429                                 upperopaque = true;
 430                         }
 431                 }
 432                 upperdentry = prev = this;
 433         }
 434
 435         if (!upperopaque && poe->numlower) {
 436                 err = -ENOMEM;
 437                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 438                 if (!stack)
 439                         goto out_put_upper;
 440         }
 441
 442         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 443                 bool opaque = false;
 444                 struct path lowerpath = poe->lowerstack[i];
 445
 446                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 447                 err = PTR_ERR(this);
 448                 if (IS_ERR(this)) {
 449                         /*
 450                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 451                          */
 452                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 453                                 continue;
 454                         goto out_put;
 455                 }
 456                 if (!this)
 457                         continue;
 458                 if (ovl_is_whiteout(this)) {
 459                         dput(this);
 460                         break;
 461                 }
 462                 /*
 463                  * Only makes sense to check opaque dir if this is not the
 464                  * lowermost layer.
 465                  */
 466                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 467                         opaque = true;
 468
 469                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 470                              !S_ISDIR(this->d_inode->i_mode))) {
 471                         /*
 472                          * FIXME: check for upper-opaqueness maybe better done
 473                          * in remove code.
 474                          */
 475                         if (prev == upperdentry)
 476                                 upperopaque = true;
 477                         dput(this);
 478                         break;
 479                 }
 480                 /*
 481                  * If this is a non-directory then stop here.
 482                  */
 483                 if (!S_ISDIR(this->d_inode->i_mode))
 484                         opaque = true;
 485
 486                 stack[ctr].dentry = this;
 487                 stack[ctr].mnt = lowerpath.mnt;
 488                 ctr++;
 489                 prev = this;
 490                 if (opaque)
 491                         break;
 492         }
 493
 494         oe = ovl_alloc_entry(ctr);
 495         err = -ENOMEM;
 496         if (!oe)
 497                 goto out_put;
 498
 499         if (upperdentry || ctr) {
 500                 struct dentry *realdentry;
 501
 502                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 503
 504                 err = -ENOMEM;
 505                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 506                                       oe);
 507                 if (!inode)
 508                         goto out_free_oe;
 509                 ovl_copyattr(realdentry->d_inode, inode);
 510         }
 511
 512         oe->opaque = upperopaque;
 513         oe->__upperdentry = upperdentry;
 514         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 515         kfree(stack);
 516         dentry->d_fsdata = oe;
 517         d_add(dentry, inode);
 518
 519         return NULL;
 520
 521 out_free_oe:
 522         kfree(oe);
 523 out_put:
 524         for (i = 0; i < ctr; i++)
 525                 dput(stack[i].dentry);
 526         kfree(stack);
 527 out_put_upper:
 528         dput(upperdentry);
 529 out:
 530         return ERR_PTR(err);
 531 }
 532
 533 struct file *ovl_path_open(struct path *path, int flags)
 534 {
 535         return dentry_open(path, flags, current_cred());
 536 }
 537
 538 static void ovl_put_super(struct super_block *sb)
 539 {
 540         struct ovl_fs *ufs = sb->s_fs_info;
 541         unsigned i;
 542
 543         dput(ufs->workdir);
 544         mntput(ufs->upper_mnt);
 545         for (i = 0; i < ufs->numlower; i++)
 546                 mntput(ufs->lower_mnt[i]);
 547
 548         kfree(ufs->config.lowerdir);
 549         kfree(ufs->config.upperdir);
 550         kfree(ufs->config.workdir);
 551         kfree(ufs);
 552 }
 553
 554 /**
 555  * ovl_statfs
 556  * @sb: The overlayfs super block
 557  * @buf: The struct kstatfs to fill in with stats
 558  *
 559  * Get the filesystem statistics.  As writes always target the upper layer
 560  * filesystem pass the statfs to the upper filesystem (if it exists)
 561  */
 562 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 563 {
 564         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 565         struct dentry *root_dentry = dentry->d_sb->s_root;
 566         struct path path;
 567         int err;
 568
 569         ovl_path_real(root_dentry, &path);
 570
 571         err = vfs_statfs(&path, buf);
 572         if (!err) {
 573                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 574                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 575         }
 576
 577         return err;
 578 }
 579
 580 /**
 581  * ovl_show_options
 582  *
 583  * Prints the mount options for a given superblock.
 584  * Returns zero; does not fail.
 585  */
 586 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 587 {
 588         struct super_block *sb = dentry->d_sb;
 589         struct ovl_fs *ufs = sb->s_fs_info;
 590
 591         seq_printf(m, ",lowerdir=%s", ufs->config.lowerdir);
 592         if (ufs->config.upperdir) {
 593                 seq_printf(m, ",upperdir=%s", ufs->config.upperdir);
 594                 seq_printf(m, ",workdir=%s", ufs->config.workdir);
 595         }
 596         return 0;
 597 }
 598
 599 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 600 {
 601         struct ovl_fs *ufs = sb->s_fs_info;
 602
 603         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 604                 return -EROFS;
 605
 606         return 0;
 607 }
 608
 609 static const struct super_operations ovl_super_operations = {
 610         .put_super      = ovl_put_super,
 611         .statfs         = ovl_statfs,
 612         .show_options   = ovl_show_options,
 613         .remount_fs     = ovl_remount,
 614 };
 615
 616 enum {
 617         OPT_LOWERDIR,
 618         OPT_UPPERDIR,
 619         OPT_WORKDIR,
 620         OPT_ERR,
 621 };
 622
 623 static const match_table_t ovl_tokens = {
 624         {OPT_LOWERDIR,                  "lowerdir=%s"},
 625         {OPT_UPPERDIR,                  "upperdir=%s"},
 626         {OPT_WORKDIR,                   "workdir=%s"},
 627         {OPT_ERR,                       NULL}
 628 };
 629
 630 static char *ovl_next_opt(char **s)
 631 {
 632         char *sbegin = *s;
 633         char *p;
 634
 635         if (sbegin == NULL)
 636                 return NULL;
 637
 638         for (p = sbegin; *p; p++) {
 639                 if (*p == '\\') {
 640                         p++;
 641                         if (!*p)
 642                                 break;
 643                 } else if (*p == ',') {
 644                         *p = '\0';
 645                         *s = p + 1;
 646                         return sbegin;
 647                 }
 648         }
 649         *s = NULL;
 650         return sbegin;
 651 }
 652
 653 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 654 {
 655         char *p;
 656
 657         while ((p = ovl_next_opt(&opt)) != NULL) {
 658                 int token;
 659                 substring_t args[MAX_OPT_ARGS];
 660
 661                 if (!*p)
 662                         continue;
 663
 664                 token = match_token(p, ovl_tokens, args);
 665                 switch (token) {
 666                 case OPT_UPPERDIR:
 667                         kfree(config->upperdir);
 668                         config->upperdir = match_strdup(&args[0]);
 669                         if (!config->upperdir)
 670                                 return -ENOMEM;
 671                         break;
 672
 673                 case OPT_LOWERDIR:
 674                         kfree(config->lowerdir);
 675                         config->lowerdir = match_strdup(&args[0]);
 676                         if (!config->lowerdir)
 677                                 return -ENOMEM;
 678                         break;
 679
 680                 case OPT_WORKDIR:
 681                         kfree(config->workdir);
 682                         config->workdir = match_strdup(&args[0]);
 683                         if (!config->workdir)
 684                                 return -ENOMEM;
 685                         break;
 686
 687                 default:
 688                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 689                         return -EINVAL;
 690                 }
 691         }
 692
 693         /* Workdir is useless in non-upper mount */
 694         if (!config->upperdir && config->workdir) {
 695                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 696                         config->workdir);
 697                 kfree(config->workdir);
 698                 config->workdir = NULL;
 699         }
 700
 701         return 0;
 702 }
 703
 704 #define OVL_WORKDIR_NAME "work"
 705
 706 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 707                                          struct dentry *dentry)
 708 {
 709         struct inode *dir = dentry->d_inode;
 710         struct dentry *work;
 711         int err;
 712         bool retried = false;
 713
 714         err = mnt_want_write(mnt);
 715         if (err)
 716                 return ERR_PTR(err);
 717
 718         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 719 retry:
 720         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 721                               strlen(OVL_WORKDIR_NAME));
 722
 723         if (!IS_ERR(work)) {
 724                 struct kstat stat = {
 725                         .mode = S_IFDIR | 0,
 726                 };
 727
 728                 if (work->d_inode) {
 729                         err = -EEXIST;
 730                         if (retried)
 731                                 goto out_dput;
 732
 733                         retried = true;
 734                         ovl_cleanup(dir, work);
 735                         dput(work);
 736                         goto retry;
 737                 }
 738
 739                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 740                 if (err)
 741                         goto out_dput;
 742         }
 743 out_unlock:
 744         mutex_unlock(&dir->i_mutex);
 745         mnt_drop_write(mnt);
 746
 747         return work;
 748
 749 out_dput:
 750         dput(work);
 751         work = ERR_PTR(err);
 752         goto out_unlock;
 753 }
 754
 755 static void ovl_unescape(char *s)
 756 {
 757         char *d = s;
 758
 759         for (;; s++, d++) {
 760                 if (*s == '\\')
 761                         s++;
 762                 *d = *s;
 763                 if (!*s)
 764                         break;
 765         }
 766 }
 767
 768 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 769 {
 770         int err = -EINVAL;
 771
 772         if (!*name) {
 773                 pr_err("overlayfs: empty lowerdir\n");
 774                 goto out;
 775         }
 776         err = kern_path(name, LOOKUP_FOLLOW, path);
 777         if (err) {
 778                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 779                 goto out;
 780         }
 781         err = -EINVAL;
 782         if (ovl_dentry_weird(path->dentry)) {
 783                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 784                 goto out_put;
 785         }
 786         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 787                 pr_err("overlayfs: '%s' not a directory\n", name);
 788                 goto out_put;
 789         }
 790         return 0;
 791
 792 out_put:
 793         path_put(path);
 794 out:
 795         return err;
 796 }
 797
 798 static int ovl_mount_dir(const char *name, struct path *path)
 799 {
 800         int err = -ENOMEM;
 801         char *tmp = kstrdup(name, GFP_KERNEL);
 802
 803         if (tmp) {
 804                 ovl_unescape(tmp);
 805                 err = ovl_mount_dir_noesc(tmp, path);
 806
 807                 if (!err)
 808                         if (ovl_dentry_remote(path->dentry)) {
 809                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 810                                        tmp);
 811                                 path_put(path);
 812                                 err = -EINVAL;
 813                         }
 814                 kfree(tmp);
 815         }
 816         return err;
 817 }
 818
 819 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 820                          int *stack_depth, bool *remote)
 821 {
 822         int err;
 823         struct kstatfs statfs;
 824
 825         err = ovl_mount_dir_noesc(name, path);
 826         if (err)
 827                 goto out;
 828
 829         err = vfs_statfs(path, &statfs);
 830         if (err) {
 831                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 832                 goto out_put;
 833         }
 834         *namelen = max(*namelen, statfs.f_namelen);
 835         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 836
 837         if (ovl_dentry_remote(path->dentry))
 838                 *remote = true;
 839
 840         return 0;
 841
 842 out_put:
 843         path_put(path);
 844 out:
 845         return err;
 846 }
 847
 848 /* Workdir should not be subdir of upperdir and vice versa */
 849 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 850 {
 851         bool ok = false;
 852
 853         if (workdir != upperdir) {
 854                 ok = (lock_rename(workdir, upperdir) == NULL);
 855                 unlock_rename(workdir, upperdir);
 856         }
 857         return ok;
 858 }
 859
 860 static unsigned int ovl_split_lowerdirs(char *str)
 861 {
 862         unsigned int ctr = 1;
 863         char *s, *d;
 864
 865         for (s = d = str;; s++, d++) {
 866                 if (*s == '\\') {
 867                         s++;
 868                 } else if (*s == ':') {
 869                         *d = '\0';
 870                         ctr++;
 871                         continue;
 872                 }
 873                 *d = *s;
 874                 if (!*s)
 875                         break;
 876         }
 877         return ctr;
 878 }
 879
 880 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 881 {
 882         struct path upperpath = { NULL, NULL };
 883         struct path workpath = { NULL, NULL };
 884         struct dentry *root_dentry;
 885         struct ovl_entry *oe;
 886         struct ovl_fs *ufs;
 887         struct path *stack = NULL;
 888         char *lowertmp;
 889         char *lower;
 890         unsigned int numlower;
 891         unsigned int stacklen = 0;
 892         unsigned int i;
 893         bool remote = false;
 894         int err;
 895
 896         err = -ENOMEM;
 897         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 898         if (!ufs)
 899                 goto out;
 900
 901         err = ovl_parse_opt((char *) data, &ufs->config);
 902         if (err)
 903                 goto out_free_config;
 904
 905         err = -EINVAL;
 906         if (!ufs->config.lowerdir) {
 907                 pr_err("overlayfs: missing 'lowerdir'\n");
 908                 goto out_free_config;
 909         }
 910
 911         sb->s_stack_depth = 0;
 912         if (ufs->config.upperdir) {
 913                 if (!ufs->config.workdir) {
 914                         pr_err("overlayfs: missing 'workdir'\n");
 915                         goto out_free_config;
 916                 }
 917
 918                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 919                 if (err)
 920                         goto out_free_config;
 921
 922                 /* Upper fs should not be r/o */
 923                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 924                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 925                         err = -EINVAL;
 926                         goto out_put_upperpath;
 927                 }
 928
 929                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 930                 if (err)
 931                         goto out_put_upperpath;
 932
 933                 err = -EINVAL;
 934                 if (upperpath.mnt != workpath.mnt) {
 935                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 936                         goto out_put_workpath;
 937                 }
 938                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 939                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 940                         goto out_put_workpath;
 941                 }
 942                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 943         }
 944         err = -ENOMEM;
 945         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 946         if (!lowertmp)
 947                 goto out_put_workpath;
 948
 949         err = -EINVAL;
 950         stacklen = ovl_split_lowerdirs(lowertmp);
 951         if (stacklen > OVL_MAX_STACK) {
 952                 pr_err("overlayfs: too many lower directries, limit is %d\n",
 953                        OVL_MAX_STACK);
 954                 goto out_free_lowertmp;
 955         } else if (!ufs->config.upperdir && stacklen == 1) {
 956                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 957                 goto out_free_lowertmp;
 958         }
 959
 960         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 961         if (!stack)
 962                 goto out_free_lowertmp;
 963
 964         lower = lowertmp;
 965         for (numlower = 0; numlower < stacklen; numlower++) {
 966                 err = ovl_lower_dir(lower, &stack[numlower],
 967                                     &ufs->lower_namelen, &sb->s_stack_depth,
 968                                     &remote);
 969                 if (err)
 970                         goto out_put_lowerpath;
 971
 972                 lower = strchr(lower, '\0') + 1;
 973         }
 974
 975         err = -EINVAL;
 976         sb->s_stack_depth++;
 977         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
 978                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
 979                 goto out_put_lowerpath;
 980         }
 981
 982         if (ufs->config.upperdir) {
 983                 ufs->upper_mnt = clone_private_mount(&upperpath);
 984                 err = PTR_ERR(ufs->upper_mnt);
 985                 if (IS_ERR(ufs->upper_mnt)) {
 986                         pr_err("overlayfs: failed to clone upperpath\n");
 987                         goto out_put_lowerpath;
 988                 }
 989
 990                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
 991                 err = PTR_ERR(ufs->workdir);
 992                 if (IS_ERR(ufs->workdir)) {
 993                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 994                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
 995                         sb->s_flags |= MS_RDONLY;
 996                         ufs->workdir = NULL;
 997                 }
 998         }
 999
1000         err = -ENOMEM;
1001         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1002         if (ufs->lower_mnt == NULL)
1003                 goto out_put_workdir;
1004         for (i = 0; i < numlower; i++) {
1005                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1006
1007                 err = PTR_ERR(mnt);
1008                 if (IS_ERR(mnt)) {
1009                         pr_err("overlayfs: failed to clone lowerpath\n");
1010                         goto out_put_lower_mnt;
1011                 }
1012                 /*
1013                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1014                  * will fail instead of modifying lower fs.
1015                  */
1016                 mnt->mnt_flags |= MNT_READONLY;
1017
1018                 ufs->lower_mnt[ufs->numlower] = mnt;
1019                 ufs->numlower++;
1020         }
1021
1022         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1023         if (!ufs->upper_mnt)
1024                 sb->s_flags |= MS_RDONLY;
1025
1026         if (remote)
1027                 sb->s_d_op = &ovl_reval_dentry_operations;
1028         else
1029                 sb->s_d_op = &ovl_dentry_operations;
1030
1031         err = -ENOMEM;
1032         oe = ovl_alloc_entry(numlower);
1033         if (!oe)
1034                 goto out_put_lower_mnt;
1035
1036         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1037         if (!root_dentry)
1038                 goto out_free_oe;
1039
1040         mntput(upperpath.mnt);
1041         for (i = 0; i < numlower; i++)
1042                 mntput(stack[i].mnt);
1043         path_put(&workpath);
1044         kfree(lowertmp);
1045
1046         oe->__upperdentry = upperpath.dentry;
1047         for (i = 0; i < numlower; i++) {
1048                 oe->lowerstack[i].dentry = stack[i].dentry;
1049                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1050         }
1051
1052         root_dentry->d_fsdata = oe;
1053
1054         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1055         sb->s_op = &ovl_super_operations;
1056         sb->s_root = root_dentry;
1057         sb->s_fs_info = ufs;
1058
1059         return 0;
1060
1061 out_free_oe:
1062         kfree(oe);
1063 out_put_lower_mnt:
1064         for (i = 0; i < ufs->numlower; i++)
1065                 mntput(ufs->lower_mnt[i]);
1066         kfree(ufs->lower_mnt);
1067 out_put_workdir:
1068         dput(ufs->workdir);
1069         mntput(ufs->upper_mnt);
1070 out_put_lowerpath:
1071         for (i = 0; i < numlower; i++)
1072                 path_put(&stack[i]);
1073         kfree(stack);
1074 out_free_lowertmp:
1075         kfree(lowertmp);
1076 out_put_workpath:
1077         path_put(&workpath);
1078 out_put_upperpath:
1079         path_put(&upperpath);
1080 out_free_config:
1081         kfree(ufs->config.lowerdir);
1082         kfree(ufs->config.upperdir);
1083         kfree(ufs->config.workdir);
1084         kfree(ufs);
1085 out:
1086         return err;
1087 }
1088
1089 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1090                                 const char *dev_name, void *raw_data)
1091 {
1092         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1093 }
1094
1095 static struct file_system_type ovl_fs_type = {
1096         .owner          = THIS_MODULE,
1097         .name           = "overlay",
1098         .mount          = ovl_mount,
1099         .kill_sb        = kill_anon_super,
1100 };
1101 MODULE_ALIAS_FS("overlay");
1102
1103 static int __init ovl_init(void)
1104 {
1105         return register_filesystem(&ovl_fs_type);
1106 }
1107
1108 static void __exit ovl_exit(void)
1109 {
1110         unregister_filesystem(&ovl_fs_type);
1111 }
1112
1113 module_init(ovl_init);
1114 module_exit(ovl_exit);