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