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