fs/overlayfs/super.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/fs.h>
  11 #include <linux/namei.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/xattr.h>
  14 #include <linux/security.h>
  15 #include <linux/mount.h>
  16 #include <linux/slab.h>
  17 #include <linux/parser.h>
  18 #include <linux/module.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/sched.h>
  21 #include <linux/statfs.h>
  22 #include <linux/seq_file.h>
  23 #include "overlayfs.h"
  24
  25 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  26 MODULE_DESCRIPTION("Overlay filesystem");
  27 MODULE_LICENSE("GPL");
  28
  29 struct ovl_config {
  30         char *lowerdir;
  31         char *upperdir;
  32         char *workdir;
  33         bool default_permissions;
  34 };
  35
  36 /* private information held for overlayfs's superblock */
  37 struct ovl_fs {
  38         struct vfsmount *upper_mnt;
  39         unsigned numlower;
  40         struct vfsmount **lower_mnt;
  41         struct dentry *workdir;
  42         long lower_namelen;
  43         /* pathnames of lower and upper dirs, for show_options */
  44         struct ovl_config config;
  45 };
  46
  47 struct ovl_dir_cache;
  48
  49 /* private information held for every overlayfs dentry */
  50 struct ovl_entry {
  51         struct dentry *__upperdentry;
  52         struct ovl_dir_cache *cache;
  53         union {
  54                 struct {
  55                         u64 version;
  56                         bool opaque;
  57                 };
  58                 struct rcu_head rcu;
  59         };
  60         unsigned numlower;
  61         struct path lowerstack[];
  62 };
  63
  64 #define OVL_MAX_STACK 500
  65
  66 static struct dentry *__ovl_dentry_lower(struct ovl_entry *oe)
  67 {
  68         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
  69 }
  70
  71 enum ovl_path_type ovl_path_type(struct dentry *dentry)
  72 {
  73         struct ovl_entry *oe = dentry->d_fsdata;
  74         enum ovl_path_type type = 0;
  75
  76         if (oe->__upperdentry) {
  77                 type = __OVL_PATH_UPPER;
  78
  79                 /*
  80                  * Non-dir dentry can hold lower dentry from previous
  81                  * location. Its purity depends only on opaque flag.
  82                  */
  83                 if (oe->numlower && S_ISDIR(dentry->d_inode->i_mode))
  84                         type |= __OVL_PATH_MERGE;
  85                 else if (!oe->opaque)
  86                         type |= __OVL_PATH_PURE;
  87         } else {
  88                 if (oe->numlower > 1)
  89                         type |= __OVL_PATH_MERGE;
  90         }
  91         return type;
  92 }
  93
  94 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
  95 {
  96         return lockless_dereference(oe->__upperdentry);
  97 }
  98
  99 void ovl_path_upper(struct dentry *dentry, struct path *path)
 100 {
 101         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 102         struct ovl_entry *oe = dentry->d_fsdata;
 103
 104         path->mnt = ofs->upper_mnt;
 105         path->dentry = ovl_upperdentry_dereference(oe);
 106 }
 107
 108 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 109 {
 110         enum ovl_path_type type = ovl_path_type(dentry);
 111
 112         if (!OVL_TYPE_UPPER(type))
 113                 ovl_path_lower(dentry, path);
 114         else
 115                 ovl_path_upper(dentry, path);
 116
 117         return type;
 118 }
 119
 120 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 121 {
 122         struct ovl_entry *oe = dentry->d_fsdata;
 123
 124         return ovl_upperdentry_dereference(oe);
 125 }
 126
 127 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 128 {
 129         struct ovl_entry *oe = dentry->d_fsdata;
 130
 131         return __ovl_dentry_lower(oe);
 132 }
 133
 134 struct dentry *ovl_dentry_real(struct dentry *dentry)
 135 {
 136         struct ovl_entry *oe = dentry->d_fsdata;
 137         struct dentry *realdentry;
 138
 139         realdentry = ovl_upperdentry_dereference(oe);
 140         if (!realdentry)
 141                 realdentry = __ovl_dentry_lower(oe);
 142
 143         return realdentry;
 144 }
 145
 146 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
 147 {
 148         struct dentry *realdentry;
 149
 150         realdentry = ovl_upperdentry_dereference(oe);
 151         if (realdentry) {
 152                 *is_upper = true;
 153         } else {
 154                 realdentry = __ovl_dentry_lower(oe);
 155                 *is_upper = false;
 156         }
 157         return realdentry;
 158 }
 159
 160 struct vfsmount *ovl_entry_mnt_real(struct ovl_entry *oe, struct inode *inode,
 161                                     bool is_upper)
 162 {
 163         if (is_upper) {
 164                 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 165
 166                 return ofs->upper_mnt;
 167         } else {
 168                 return oe->numlower ? oe->lowerstack[0].mnt : NULL;
 169         }
 170 }
 171
 172 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 173 {
 174         struct ovl_entry *oe = dentry->d_fsdata;
 175
 176         return oe->cache;
 177 }
 178
 179 bool ovl_is_default_permissions(struct inode *inode)
 180 {
 181         struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 182
 183         return ofs->config.default_permissions;
 184 }
 185
 186 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 187 {
 188         struct ovl_entry *oe = dentry->d_fsdata;
 189
 190         oe->cache = cache;
 191 }
 192
 193 void ovl_path_lower(struct dentry *dentry, struct path *path)
 194 {
 195         struct ovl_entry *oe = dentry->d_fsdata;
 196
 197         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 198 }
 199
 200 int ovl_want_write(struct dentry *dentry)
 201 {
 202         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 203         return mnt_want_write(ofs->upper_mnt);
 204 }
 205
 206 void ovl_drop_write(struct dentry *dentry)
 207 {
 208         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 209         mnt_drop_write(ofs->upper_mnt);
 210 }
 211
 212 struct dentry *ovl_workdir(struct dentry *dentry)
 213 {
 214         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 215         return ofs->workdir;
 216 }
 217
 218 bool ovl_dentry_is_opaque(struct dentry *dentry)
 219 {
 220         struct ovl_entry *oe = dentry->d_fsdata;
 221         return oe->opaque;
 222 }
 223
 224 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 225 {
 226         struct ovl_entry *oe = dentry->d_fsdata;
 227         oe->opaque = opaque;
 228 }
 229
 230 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 231 {
 232         struct ovl_entry *oe = dentry->d_fsdata;
 233
 234         WARN_ON(!inode_is_locked(upperdentry->d_parent->d_inode));
 235         WARN_ON(oe->__upperdentry);
 236         BUG_ON(!upperdentry->d_inode);
 237         /*
 238          * Make sure upperdentry is consistent before making it visible to
 239          * ovl_upperdentry_dereference().
 240          */
 241         smp_wmb();
 242         oe->__upperdentry = upperdentry;
 243 }
 244
 245 void ovl_dentry_version_inc(struct dentry *dentry)
 246 {
 247         struct ovl_entry *oe = dentry->d_fsdata;
 248
 249         WARN_ON(!inode_is_locked(dentry->d_inode));
 250         oe->version++;
 251 }
 252
 253 u64 ovl_dentry_version_get(struct dentry *dentry)
 254 {
 255         struct ovl_entry *oe = dentry->d_fsdata;
 256
 257         WARN_ON(!inode_is_locked(dentry->d_inode));
 258         return oe->version;
 259 }
 260
 261 bool ovl_is_whiteout(struct dentry *dentry)
 262 {
 263         struct inode *inode = dentry->d_inode;
 264
 265         return inode && IS_WHITEOUT(inode);
 266 }
 267
 268 static bool ovl_is_opaquedir(struct dentry *dentry)
 269 {
 270         int res;
 271         char val;
 272         struct inode *inode = dentry->d_inode;
 273
 274         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
 275                 return false;
 276
 277         res = inode->i_op->getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
 278         if (res == 1 && val == 'y')
 279                 return true;
 280
 281         return false;
 282 }
 283
 284 static void ovl_dentry_release(struct dentry *dentry)
 285 {
 286         struct ovl_entry *oe = dentry->d_fsdata;
 287
 288         if (oe) {
 289                 unsigned int i;
 290
 291                 dput(oe->__upperdentry);
 292                 for (i = 0; i < oe->numlower; i++)
 293                         dput(oe->lowerstack[i].dentry);
 294                 kfree_rcu(oe, rcu);
 295         }
 296 }
 297
 298 static struct dentry *ovl_d_real(struct dentry *dentry, struct inode *inode)
 299 {
 300         struct dentry *real;
 301
 302         if (d_is_dir(dentry)) {
 303                 if (!inode || inode == d_inode(dentry))
 304                         return dentry;
 305                 goto bug;
 306         }
 307
 308         real = ovl_dentry_upper(dentry);
 309         if (real && (!inode || inode == d_inode(real)))
 310                 return real;
 311
 312         real = ovl_dentry_lower(dentry);
 313         if (!real)
 314                 goto bug;
 315
 316         if (!inode || inode == d_inode(real))
 317                 return real;
 318
 319         /* Handle recursion */
 320         if (real->d_flags & DCACHE_OP_REAL)
 321                 return real->d_op->d_real(real, inode);
 322
 323 bug:
 324         WARN(1, "ovl_d_real(%pd4, %s:%lu\n): real dentry not found\n", dentry,
 325              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 326         return dentry;
 327 }
 328
 329 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 330 {
 331         struct ovl_entry *oe = dentry->d_fsdata;
 332         unsigned int i;
 333         int ret = 1;
 334
 335         for (i = 0; i < oe->numlower; i++) {
 336                 struct dentry *d = oe->lowerstack[i].dentry;
 337
 338                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 339                         ret = d->d_op->d_revalidate(d, flags);
 340                         if (ret < 0)
 341                                 return ret;
 342                         if (!ret) {
 343                                 if (!(flags & LOOKUP_RCU))
 344                                         d_invalidate(d);
 345                                 return -ESTALE;
 346                         }
 347                 }
 348         }
 349         return 1;
 350 }
 351
 352 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 353 {
 354         struct ovl_entry *oe = dentry->d_fsdata;
 355         unsigned int i;
 356         int ret = 1;
 357
 358         for (i = 0; i < oe->numlower; i++) {
 359                 struct dentry *d = oe->lowerstack[i].dentry;
 360
 361                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 362                         ret = d->d_op->d_weak_revalidate(d, flags);
 363                         if (ret <= 0)
 364                                 break;
 365                 }
 366         }
 367         return ret;
 368 }
 369
 370 static const struct dentry_operations ovl_dentry_operations = {
 371         .d_release = ovl_dentry_release,
 372         .d_select_inode = ovl_d_select_inode,
 373         .d_real = ovl_d_real,
 374 };
 375
 376 static const struct dentry_operations ovl_reval_dentry_operations = {
 377         .d_release = ovl_dentry_release,
 378         .d_select_inode = ovl_d_select_inode,
 379         .d_real = ovl_d_real,
 380         .d_revalidate = ovl_dentry_revalidate,
 381         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 382 };
 383
 384 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 385 {
 386         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 387         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 388
 389         if (oe)
 390                 oe->numlower = numlower;
 391
 392         return oe;
 393 }
 394
 395 static bool ovl_dentry_remote(struct dentry *dentry)
 396 {
 397         return dentry->d_flags &
 398                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 399 }
 400
 401 static bool ovl_dentry_weird(struct dentry *dentry)
 402 {
 403         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 404                                   DCACHE_MANAGE_TRANSIT |
 405                                   DCACHE_OP_HASH |
 406                                   DCACHE_OP_COMPARE);
 407 }
 408
 409 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 410                                              struct qstr *name)
 411 {
 412         struct dentry *dentry;
 413
 414         inode_lock(dir->d_inode);
 415         dentry = lookup_one_len(name->name, dir, name->len);
 416         inode_unlock(dir->d_inode);
 417
 418         if (IS_ERR(dentry)) {
 419                 if (PTR_ERR(dentry) == -ENOENT)
 420                         dentry = NULL;
 421         } else if (!dentry->d_inode) {
 422                 dput(dentry);
 423                 dentry = NULL;
 424         } else if (ovl_dentry_weird(dentry)) {
 425                 dput(dentry);
 426                 /* Don't support traversing automounts and other weirdness */
 427                 dentry = ERR_PTR(-EREMOTE);
 428         }
 429         return dentry;
 430 }
 431
 432 /*
 433  * Returns next layer in stack starting from top.
 434  * Returns -1 if this is the last layer.
 435  */
 436 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 437 {
 438         struct ovl_entry *oe = dentry->d_fsdata;
 439
 440         BUG_ON(idx < 0);
 441         if (idx == 0) {
 442                 ovl_path_upper(dentry, path);
 443                 if (path->dentry)
 444                         return oe->numlower ? 1 : -1;
 445                 idx++;
 446         }
 447         BUG_ON(idx > oe->numlower);
 448         *path = oe->lowerstack[idx - 1];
 449
 450         return (idx < oe->numlower) ? idx + 1 : -1;
 451 }
 452
 453 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 454                           unsigned int flags)
 455 {
 456         struct ovl_entry *oe;
 457         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 458         struct path *stack = NULL;
 459         struct dentry *upperdir, *upperdentry = NULL;
 460         unsigned int ctr = 0;
 461         struct inode *inode = NULL;
 462         bool upperopaque = false;
 463         struct dentry *this, *prev = NULL;
 464         unsigned int i;
 465         int err;
 466
 467         upperdir = ovl_upperdentry_dereference(poe);
 468         if (upperdir) {
 469                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 470                 err = PTR_ERR(this);
 471                 if (IS_ERR(this))
 472                         goto out;
 473
 474                 if (this) {
 475                         if (unlikely(ovl_dentry_remote(this))) {
 476                                 dput(this);
 477                                 err = -EREMOTE;
 478                                 goto out;
 479                         }
 480                         if (ovl_is_whiteout(this)) {
 481                                 dput(this);
 482                                 this = NULL;
 483                                 upperopaque = true;
 484                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 485                                 upperopaque = true;
 486                         }
 487                 }
 488                 upperdentry = prev = this;
 489         }
 490
 491         if (!upperopaque && poe->numlower) {
 492                 err = -ENOMEM;
 493                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 494                 if (!stack)
 495                         goto out_put_upper;
 496         }
 497
 498         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 499                 bool opaque = false;
 500                 struct path lowerpath = poe->lowerstack[i];
 501
 502                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 503                 err = PTR_ERR(this);
 504                 if (IS_ERR(this)) {
 505                         /*
 506                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 507                          */
 508                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 509                                 continue;
 510                         goto out_put;
 511                 }
 512                 if (!this)
 513                         continue;
 514                 if (ovl_is_whiteout(this)) {
 515                         dput(this);
 516                         break;
 517                 }
 518                 /*
 519                  * Only makes sense to check opaque dir if this is not the
 520                  * lowermost layer.
 521                  */
 522                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 523                         opaque = true;
 524
 525                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 526                              !S_ISDIR(this->d_inode->i_mode))) {
 527                         /*
 528                          * FIXME: check for upper-opaqueness maybe better done
 529                          * in remove code.
 530                          */
 531                         if (prev == upperdentry)
 532                                 upperopaque = true;
 533                         dput(this);
 534                         break;
 535                 }
 536                 /*
 537                  * If this is a non-directory then stop here.
 538                  */
 539                 if (!S_ISDIR(this->d_inode->i_mode))
 540                         opaque = true;
 541
 542                 stack[ctr].dentry = this;
 543                 stack[ctr].mnt = lowerpath.mnt;
 544                 ctr++;
 545                 prev = this;
 546                 if (opaque)
 547                         break;
 548         }
 549
 550         oe = ovl_alloc_entry(ctr);
 551         err = -ENOMEM;
 552         if (!oe)
 553                 goto out_put;
 554
 555         if (upperdentry || ctr) {
 556                 struct dentry *realdentry;
 557
 558                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 559
 560                 err = -ENOMEM;
 561                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 562                                       oe);
 563                 if (!inode)
 564                         goto out_free_oe;
 565                 ovl_copyattr(realdentry->d_inode, inode);
 566         }
 567
 568         oe->opaque = upperopaque;
 569         oe->__upperdentry = upperdentry;
 570         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 571         kfree(stack);
 572         dentry->d_fsdata = oe;
 573         d_add(dentry, inode);
 574
 575         return NULL;
 576
 577 out_free_oe:
 578         kfree(oe);
 579 out_put:
 580         for (i = 0; i < ctr; i++)
 581                 dput(stack[i].dentry);
 582         kfree(stack);
 583 out_put_upper:
 584         dput(upperdentry);
 585 out:
 586         return ERR_PTR(err);
 587 }
 588
 589 struct file *ovl_path_open(struct path *path, int flags)
 590 {
 591         return dentry_open(path, flags, current_cred());
 592 }
 593
 594 static void ovl_put_super(struct super_block *sb)
 595 {
 596         struct ovl_fs *ufs = sb->s_fs_info;
 597         unsigned i;
 598
 599         dput(ufs->workdir);
 600         mntput(ufs->upper_mnt);
 601         for (i = 0; i < ufs->numlower; i++)
 602                 mntput(ufs->lower_mnt[i]);
 603         kfree(ufs->lower_mnt);
 604
 605         kfree(ufs->config.lowerdir);
 606         kfree(ufs->config.upperdir);
 607         kfree(ufs->config.workdir);
 608         kfree(ufs);
 609 }
 610
 611 /**
 612  * ovl_statfs
 613  * @sb: The overlayfs super block
 614  * @buf: The struct kstatfs to fill in with stats
 615  *
 616  * Get the filesystem statistics.  As writes always target the upper layer
 617  * filesystem pass the statfs to the upper filesystem (if it exists)
 618  */
 619 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 620 {
 621         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 622         struct dentry *root_dentry = dentry->d_sb->s_root;
 623         struct path path;
 624         int err;
 625
 626         ovl_path_real(root_dentry, &path);
 627
 628         err = vfs_statfs(&path, buf);
 629         if (!err) {
 630                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 631                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 632         }
 633
 634         return err;
 635 }
 636
 637 /**
 638  * ovl_show_options
 639  *
 640  * Prints the mount options for a given superblock.
 641  * Returns zero; does not fail.
 642  */
 643 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 644 {
 645         struct super_block *sb = dentry->d_sb;
 646         struct ovl_fs *ufs = sb->s_fs_info;
 647
 648         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 649         if (ufs->config.upperdir) {
 650                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 651                 seq_show_option(m, "workdir", ufs->config.workdir);
 652         }
 653         if (ufs->config.default_permissions)
 654                 seq_puts(m, ",default_permissions");
 655         return 0;
 656 }
 657
 658 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 659 {
 660         struct ovl_fs *ufs = sb->s_fs_info;
 661
 662         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 663                 return -EROFS;
 664
 665         return 0;
 666 }
 667
 668 static const struct super_operations ovl_super_operations = {
 669         .put_super      = ovl_put_super,
 670         .statfs         = ovl_statfs,
 671         .show_options   = ovl_show_options,
 672         .remount_fs     = ovl_remount,
 673 };
 674
 675 enum {
 676         OPT_LOWERDIR,
 677         OPT_UPPERDIR,
 678         OPT_WORKDIR,
 679         OPT_DEFAULT_PERMISSIONS,
 680         OPT_ERR,
 681 };
 682
 683 static const match_table_t ovl_tokens = {
 684         {OPT_LOWERDIR,                  "lowerdir=%s"},
 685         {OPT_UPPERDIR,                  "upperdir=%s"},
 686         {OPT_WORKDIR,                   "workdir=%s"},
 687         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 688         {OPT_ERR,                       NULL}
 689 };
 690
 691 static char *ovl_next_opt(char **s)
 692 {
 693         char *sbegin = *s;
 694         char *p;
 695
 696         if (sbegin == NULL)
 697                 return NULL;
 698
 699         for (p = sbegin; *p; p++) {
 700                 if (*p == '\\') {
 701                         p++;
 702                         if (!*p)
 703                                 break;
 704                 } else if (*p == ',') {
 705                         *p = '\0';
 706                         *s = p + 1;
 707                         return sbegin;
 708                 }
 709         }
 710         *s = NULL;
 711         return sbegin;
 712 }
 713
 714 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 715 {
 716         char *p;
 717
 718         while ((p = ovl_next_opt(&opt)) != NULL) {
 719                 int token;
 720                 substring_t args[MAX_OPT_ARGS];
 721
 722                 if (!*p)
 723                         continue;
 724
 725                 token = match_token(p, ovl_tokens, args);
 726                 switch (token) {
 727                 case OPT_UPPERDIR:
 728                         kfree(config->upperdir);
 729                         config->upperdir = match_strdup(&args[0]);
 730                         if (!config->upperdir)
 731                                 return -ENOMEM;
 732                         break;
 733
 734                 case OPT_LOWERDIR:
 735                         kfree(config->lowerdir);
 736                         config->lowerdir = match_strdup(&args[0]);
 737                         if (!config->lowerdir)
 738                                 return -ENOMEM;
 739                         break;
 740
 741                 case OPT_WORKDIR:
 742                         kfree(config->workdir);
 743                         config->workdir = match_strdup(&args[0]);
 744                         if (!config->workdir)
 745                                 return -ENOMEM;
 746                         break;
 747
 748                 case OPT_DEFAULT_PERMISSIONS:
 749                         config->default_permissions = true;
 750                         break;
 751
 752                 default:
 753                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 754                         return -EINVAL;
 755                 }
 756         }
 757
 758         /* Workdir is useless in non-upper mount */
 759         if (!config->upperdir && config->workdir) {
 760                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 761                         config->workdir);
 762                 kfree(config->workdir);
 763                 config->workdir = NULL;
 764         }
 765
 766         return 0;
 767 }
 768
 769 #define OVL_WORKDIR_NAME "work"
 770
 771 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 772                                          struct dentry *dentry)
 773 {
 774         struct inode *dir = dentry->d_inode;
 775         struct dentry *work;
 776         int err;
 777         bool retried = false;
 778
 779         err = mnt_want_write(mnt);
 780         if (err)
 781                 return ERR_PTR(err);
 782
 783         inode_lock_nested(dir, I_MUTEX_PARENT);
 784 retry:
 785         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 786                               strlen(OVL_WORKDIR_NAME));
 787
 788         if (!IS_ERR(work)) {
 789                 struct kstat stat = {
 790                         .mode = S_IFDIR | 0,
 791                 };
 792
 793                 if (work->d_inode) {
 794                         err = -EEXIST;
 795                         if (retried)
 796                                 goto out_dput;
 797
 798                         retried = true;
 799                         ovl_cleanup(dir, work);
 800                         dput(work);
 801                         goto retry;
 802                 }
 803
 804                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 805                 if (err)
 806                         goto out_dput;
 807         }
 808 out_unlock:
 809         inode_unlock(dir);
 810         mnt_drop_write(mnt);
 811
 812         return work;
 813
 814 out_dput:
 815         dput(work);
 816         work = ERR_PTR(err);
 817         goto out_unlock;
 818 }
 819
 820 static void ovl_unescape(char *s)
 821 {
 822         char *d = s;
 823
 824         for (;; s++, d++) {
 825                 if (*s == '\\')
 826                         s++;
 827                 *d = *s;
 828                 if (!*s)
 829                         break;
 830         }
 831 }
 832
 833 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 834 {
 835         int err = -EINVAL;
 836
 837         if (!*name) {
 838                 pr_err("overlayfs: empty lowerdir\n");
 839                 goto out;
 840         }
 841         err = kern_path(name, LOOKUP_FOLLOW, path);
 842         if (err) {
 843                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 844                 goto out;
 845         }
 846         err = -EINVAL;
 847         if (ovl_dentry_weird(path->dentry)) {
 848                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 849                 goto out_put;
 850         }
 851         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 852                 pr_err("overlayfs: '%s' not a directory\n", name);
 853                 goto out_put;
 854         }
 855         return 0;
 856
 857 out_put:
 858         path_put(path);
 859 out:
 860         return err;
 861 }
 862
 863 static int ovl_mount_dir(const char *name, struct path *path)
 864 {
 865         int err = -ENOMEM;
 866         char *tmp = kstrdup(name, GFP_KERNEL);
 867
 868         if (tmp) {
 869                 ovl_unescape(tmp);
 870                 err = ovl_mount_dir_noesc(tmp, path);
 871
 872                 if (!err)
 873                         if (ovl_dentry_remote(path->dentry)) {
 874                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 875                                        tmp);
 876                                 path_put(path);
 877                                 err = -EINVAL;
 878                         }
 879                 kfree(tmp);
 880         }
 881         return err;
 882 }
 883
 884 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 885                          int *stack_depth, bool *remote)
 886 {
 887         int err;
 888         struct kstatfs statfs;
 889
 890         err = ovl_mount_dir_noesc(name, path);
 891         if (err)
 892                 goto out;
 893
 894         err = vfs_statfs(path, &statfs);
 895         if (err) {
 896                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 897                 goto out_put;
 898         }
 899         *namelen = max(*namelen, statfs.f_namelen);
 900         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 901
 902         if (ovl_dentry_remote(path->dentry))
 903                 *remote = true;
 904
 905         return 0;
 906
 907 out_put:
 908         path_put(path);
 909 out:
 910         return err;
 911 }
 912
 913 /* Workdir should not be subdir of upperdir and vice versa */
 914 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 915 {
 916         bool ok = false;
 917
 918         if (workdir != upperdir) {
 919                 ok = (lock_rename(workdir, upperdir) == NULL);
 920                 unlock_rename(workdir, upperdir);
 921         }
 922         return ok;
 923 }
 924
 925 static unsigned int ovl_split_lowerdirs(char *str)
 926 {
 927         unsigned int ctr = 1;
 928         char *s, *d;
 929
 930         for (s = d = str;; s++, d++) {
 931                 if (*s == '\\') {
 932                         s++;
 933                 } else if (*s == ':') {
 934                         *d = '\0';
 935                         ctr++;
 936                         continue;
 937                 }
 938                 *d = *s;
 939                 if (!*s)
 940                         break;
 941         }
 942         return ctr;
 943 }
 944
 945 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 946 {
 947         struct path upperpath = { NULL, NULL };
 948         struct path workpath = { NULL, NULL };
 949         struct dentry *root_dentry;
 950         struct ovl_entry *oe;
 951         struct ovl_fs *ufs;
 952         struct path *stack = NULL;
 953         char *lowertmp;
 954         char *lower;
 955         unsigned int numlower;
 956         unsigned int stacklen = 0;
 957         unsigned int i;
 958         bool remote = false;
 959         int err;
 960
 961         err = -ENOMEM;
 962         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 963         if (!ufs)
 964                 goto out;
 965
 966         err = ovl_parse_opt((char *) data, &ufs->config);
 967         if (err)
 968                 goto out_free_config;
 969
 970         err = -EINVAL;
 971         if (!ufs->config.lowerdir) {
 972                 if (!silent)
 973                         pr_err("overlayfs: missing 'lowerdir'\n");
 974                 goto out_free_config;
 975         }
 976
 977         sb->s_stack_depth = 0;
 978         sb->s_maxbytes = MAX_LFS_FILESIZE;
 979         if (ufs->config.upperdir) {
 980                 if (!ufs->config.workdir) {
 981                         pr_err("overlayfs: missing 'workdir'\n");
 982                         goto out_free_config;
 983                 }
 984
 985                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 986                 if (err)
 987                         goto out_free_config;
 988
 989                 /* Upper fs should not be r/o */
 990                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 991                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 992                         err = -EINVAL;
 993                         goto out_put_upperpath;
 994                 }
 995
 996                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 997                 if (err)
 998                         goto out_put_upperpath;
 999
1000                 err = -EINVAL;
1001                 if (upperpath.mnt != workpath.mnt) {
1002                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1003                         goto out_put_workpath;
1004                 }
1005                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
1006                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1007                         goto out_put_workpath;
1008                 }
1009                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
1010         }
1011         err = -ENOMEM;
1012         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
1013         if (!lowertmp)
1014                 goto out_put_workpath;
1015
1016         err = -EINVAL;
1017         stacklen = ovl_split_lowerdirs(lowertmp);
1018         if (stacklen > OVL_MAX_STACK) {
1019                 pr_err("overlayfs: too many lower directries, limit is %d\n",
1020                        OVL_MAX_STACK);
1021                 goto out_free_lowertmp;
1022         } else if (!ufs->config.upperdir && stacklen == 1) {
1023                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1024                 goto out_free_lowertmp;
1025         }
1026
1027         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1028         if (!stack)
1029                 goto out_free_lowertmp;
1030
1031         lower = lowertmp;
1032         for (numlower = 0; numlower < stacklen; numlower++) {
1033                 err = ovl_lower_dir(lower, &stack[numlower],
1034                                     &ufs->lower_namelen, &sb->s_stack_depth,
1035                                     &remote);
1036                 if (err)
1037                         goto out_put_lowerpath;
1038
1039                 lower = strchr(lower, '\0') + 1;
1040         }
1041
1042         err = -EINVAL;
1043         sb->s_stack_depth++;
1044         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1045                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1046                 goto out_put_lowerpath;
1047         }
1048
1049         if (ufs->config.upperdir) {
1050                 ufs->upper_mnt = clone_private_mount(&upperpath);
1051                 err = PTR_ERR(ufs->upper_mnt);
1052                 if (IS_ERR(ufs->upper_mnt)) {
1053                         pr_err("overlayfs: failed to clone upperpath\n");
1054                         goto out_put_lowerpath;
1055                 }
1056
1057                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
1058                 err = PTR_ERR(ufs->workdir);
1059                 if (IS_ERR(ufs->workdir)) {
1060                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1061                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1062                         sb->s_flags |= MS_RDONLY;
1063                         ufs->workdir = NULL;
1064                 }
1065
1066                 /*
1067                  * Upper should support d_type, else whiteouts are visible.
1068                  * Given workdir and upper are on same fs, we can do
1069                  * iterate_dir() on workdir.
1070                  */
1071                 err = ovl_check_d_type_supported(&workpath);
1072                 if (err < 0)
1073                         goto out_put_workdir;
1074
1075                 if (!err) {
1076                         pr_err("overlayfs: upper fs needs to support d_type.\n");
1077                         err = -EINVAL;
1078                         goto out_put_workdir;
1079                 }
1080         }
1081
1082         err = -ENOMEM;
1083         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1084         if (ufs->lower_mnt == NULL)
1085                 goto out_put_workdir;
1086         for (i = 0; i < numlower; i++) {
1087                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1088
1089                 err = PTR_ERR(mnt);
1090                 if (IS_ERR(mnt)) {
1091                         pr_err("overlayfs: failed to clone lowerpath\n");
1092                         goto out_put_lower_mnt;
1093                 }
1094                 /*
1095                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1096                  * will fail instead of modifying lower fs.
1097                  */
1098                 mnt->mnt_flags |= MNT_READONLY;
1099
1100                 ufs->lower_mnt[ufs->numlower] = mnt;
1101                 ufs->numlower++;
1102         }
1103
1104         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1105         if (!ufs->upper_mnt)
1106                 sb->s_flags |= MS_RDONLY;
1107
1108         if (remote)
1109                 sb->s_d_op = &ovl_reval_dentry_operations;
1110         else
1111                 sb->s_d_op = &ovl_dentry_operations;
1112
1113         err = -ENOMEM;
1114         oe = ovl_alloc_entry(numlower);
1115         if (!oe)
1116                 goto out_put_lower_mnt;
1117
1118         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1119         if (!root_dentry)
1120                 goto out_free_oe;
1121
1122         mntput(upperpath.mnt);
1123         for (i = 0; i < numlower; i++)
1124                 mntput(stack[i].mnt);
1125         path_put(&workpath);
1126         kfree(lowertmp);
1127
1128         oe->__upperdentry = upperpath.dentry;
1129         for (i = 0; i < numlower; i++) {
1130                 oe->lowerstack[i].dentry = stack[i].dentry;
1131                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1132         }
1133         kfree(stack);
1134
1135         root_dentry->d_fsdata = oe;
1136
1137         ovl_copyattr(ovl_dentry_real(root_dentry)->d_inode,
1138                      root_dentry->d_inode);
1139
1140         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1141         sb->s_op = &ovl_super_operations;
1142         sb->s_root = root_dentry;
1143         sb->s_fs_info = ufs;
1144
1145         return 0;
1146
1147 out_free_oe:
1148         kfree(oe);
1149 out_put_lower_mnt:
1150         for (i = 0; i < ufs->numlower; i++)
1151                 mntput(ufs->lower_mnt[i]);
1152         kfree(ufs->lower_mnt);
1153 out_put_workdir:
1154         dput(ufs->workdir);
1155         mntput(ufs->upper_mnt);
1156 out_put_lowerpath:
1157         for (i = 0; i < numlower; i++)
1158                 path_put(&stack[i]);
1159         kfree(stack);
1160 out_free_lowertmp:
1161         kfree(lowertmp);
1162 out_put_workpath:
1163         path_put(&workpath);
1164 out_put_upperpath:
1165         path_put(&upperpath);
1166 out_free_config:
1167         kfree(ufs->config.lowerdir);
1168         kfree(ufs->config.upperdir);
1169         kfree(ufs->config.workdir);
1170         kfree(ufs);
1171 out:
1172         return err;
1173 }
1174
1175 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1176                                 const char *dev_name, void *raw_data)
1177 {
1178         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1179 }
1180
1181 static struct file_system_type ovl_fs_type = {
1182         .owner          = THIS_MODULE,
1183         .name           = "overlay",
1184         .mount          = ovl_mount,
1185         .kill_sb        = kill_anon_super,
1186 };
1187 MODULE_ALIAS_FS("overlay");
1188
1189 static int __init ovl_init(void)
1190 {
1191         return register_filesystem(&ovl_fs_type);
1192 }
1193
1194 static void __exit ovl_exit(void)
1195 {
1196         unregister_filesystem(&ovl_fs_type);
1197 }
1198
1199 module_init(ovl_init);
1200 module_exit(ovl_exit);