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         /* creds of process who forced instantiation of super block */
  46         const struct cred *creator_cred;
  47 };
  48
  49 struct ovl_dir_cache;
  50
  51 /* private information held for every overlayfs dentry */
  52 struct ovl_entry {
  53         struct dentry *__upperdentry;
  54         struct ovl_dir_cache *cache;
  55         union {
  56                 struct {
  57                         u64 version;
  58                         bool opaque;
  59                 };
  60                 struct rcu_head rcu;
  61         };
  62         unsigned numlower;
  63         struct path lowerstack[];
  64 };
  65
  66 #define OVL_MAX_STACK 500
  67
  68 static struct dentry *__ovl_dentry_lower(struct ovl_entry *oe)
  69 {
  70         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
  71 }
  72
  73 enum ovl_path_type ovl_path_type(struct dentry *dentry)
  74 {
  75         struct ovl_entry *oe = dentry->d_fsdata;
  76         enum ovl_path_type type = 0;
  77
  78         if (oe->__upperdentry) {
  79                 type = __OVL_PATH_UPPER;
  80
  81                 /*
  82                  * Non-dir dentry can hold lower dentry from previous
  83                  * location. Its purity depends only on opaque flag.
  84                  */
  85                 if (oe->numlower && S_ISDIR(dentry->d_inode->i_mode))
  86                         type |= __OVL_PATH_MERGE;
  87                 else if (!oe->opaque)
  88                         type |= __OVL_PATH_PURE;
  89         } else {
  90                 if (oe->numlower > 1)
  91                         type |= __OVL_PATH_MERGE;
  92         }
  93         return type;
  94 }
  95
  96 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
  97 {
  98         return lockless_dereference(oe->__upperdentry);
  99 }
 100
 101 void ovl_path_upper(struct dentry *dentry, struct path *path)
 102 {
 103         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 104         struct ovl_entry *oe = dentry->d_fsdata;
 105
 106         path->mnt = ofs->upper_mnt;
 107         path->dentry = ovl_upperdentry_dereference(oe);
 108 }
 109
 110 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 111 {
 112         enum ovl_path_type type = ovl_path_type(dentry);
 113
 114         if (!OVL_TYPE_UPPER(type))
 115                 ovl_path_lower(dentry, path);
 116         else
 117                 ovl_path_upper(dentry, path);
 118
 119         return type;
 120 }
 121
 122 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 123 {
 124         struct ovl_entry *oe = dentry->d_fsdata;
 125
 126         return ovl_upperdentry_dereference(oe);
 127 }
 128
 129 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 130 {
 131         struct ovl_entry *oe = dentry->d_fsdata;
 132
 133         return __ovl_dentry_lower(oe);
 134 }
 135
 136 struct dentry *ovl_dentry_real(struct dentry *dentry)
 137 {
 138         struct ovl_entry *oe = dentry->d_fsdata;
 139         struct dentry *realdentry;
 140
 141         realdentry = ovl_upperdentry_dereference(oe);
 142         if (!realdentry)
 143                 realdentry = __ovl_dentry_lower(oe);
 144
 145         return realdentry;
 146 }
 147
 148 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
 149 {
 150         struct dentry *realdentry;
 151
 152         realdentry = ovl_upperdentry_dereference(oe);
 153         if (realdentry) {
 154                 *is_upper = true;
 155         } else {
 156                 realdentry = __ovl_dentry_lower(oe);
 157                 *is_upper = false;
 158         }
 159         return realdentry;
 160 }
 161
 162 struct vfsmount *ovl_entry_mnt_real(struct ovl_entry *oe, struct inode *inode,
 163                                     bool is_upper)
 164 {
 165         if (is_upper) {
 166                 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 167
 168                 return ofs->upper_mnt;
 169         } else {
 170                 return oe->numlower ? oe->lowerstack[0].mnt : NULL;
 171         }
 172 }
 173
 174 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 175 {
 176         struct ovl_entry *oe = dentry->d_fsdata;
 177
 178         return oe->cache;
 179 }
 180
 181 bool ovl_is_default_permissions(struct inode *inode)
 182 {
 183         struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 184
 185         return ofs->config.default_permissions;
 186 }
 187
 188 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 189 {
 190         struct ovl_entry *oe = dentry->d_fsdata;
 191
 192         oe->cache = cache;
 193 }
 194
 195 void ovl_path_lower(struct dentry *dentry, struct path *path)
 196 {
 197         struct ovl_entry *oe = dentry->d_fsdata;
 198
 199         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 200 }
 201
 202 int ovl_want_write(struct dentry *dentry)
 203 {
 204         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 205         return mnt_want_write(ofs->upper_mnt);
 206 }
 207
 208 void ovl_drop_write(struct dentry *dentry)
 209 {
 210         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 211         mnt_drop_write(ofs->upper_mnt);
 212 }
 213
 214 struct dentry *ovl_workdir(struct dentry *dentry)
 215 {
 216         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 217         return ofs->workdir;
 218 }
 219
 220 bool ovl_dentry_is_opaque(struct dentry *dentry)
 221 {
 222         struct ovl_entry *oe = dentry->d_fsdata;
 223         return oe->opaque;
 224 }
 225
 226 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 227 {
 228         struct ovl_entry *oe = dentry->d_fsdata;
 229         oe->opaque = opaque;
 230 }
 231
 232 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 233 {
 234         struct ovl_entry *oe = dentry->d_fsdata;
 235
 236         WARN_ON(!inode_is_locked(upperdentry->d_parent->d_inode));
 237         WARN_ON(oe->__upperdentry);
 238         BUG_ON(!upperdentry->d_inode);
 239         /*
 240          * Make sure upperdentry is consistent before making it visible to
 241          * ovl_upperdentry_dereference().
 242          */
 243         smp_wmb();
 244         oe->__upperdentry = upperdentry;
 245 }
 246
 247 void ovl_dentry_version_inc(struct dentry *dentry)
 248 {
 249         struct ovl_entry *oe = dentry->d_fsdata;
 250
 251         WARN_ON(!inode_is_locked(dentry->d_inode));
 252         oe->version++;
 253 }
 254
 255 u64 ovl_dentry_version_get(struct dentry *dentry)
 256 {
 257         struct ovl_entry *oe = dentry->d_fsdata;
 258
 259         WARN_ON(!inode_is_locked(dentry->d_inode));
 260         return oe->version;
 261 }
 262
 263 bool ovl_is_whiteout(struct dentry *dentry)
 264 {
 265         struct inode *inode = dentry->d_inode;
 266
 267         return inode && IS_WHITEOUT(inode);
 268 }
 269
 270 const struct cred *ovl_override_creds(struct super_block *sb)
 271 {
 272         struct ovl_fs *ofs = sb->s_fs_info;
 273
 274         return override_creds(ofs->creator_cred);
 275 }
 276
 277 static bool ovl_is_opaquedir(struct dentry *dentry)
 278 {
 279         int res;
 280         char val;
 281         struct inode *inode = dentry->d_inode;
 282
 283         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
 284                 return false;
 285
 286         res = inode->i_op->getxattr(dentry, inode, OVL_XATTR_OPAQUE, &val, 1);
 287         if (res == 1 && val == 'y')
 288                 return true;
 289
 290         return false;
 291 }
 292
 293 static void ovl_dentry_release(struct dentry *dentry)
 294 {
 295         struct ovl_entry *oe = dentry->d_fsdata;
 296
 297         if (oe) {
 298                 unsigned int i;
 299
 300                 dput(oe->__upperdentry);
 301                 for (i = 0; i < oe->numlower; i++)
 302                         dput(oe->lowerstack[i].dentry);
 303                 kfree_rcu(oe, rcu);
 304         }
 305 }
 306
 307 static struct dentry *ovl_d_real(struct dentry *dentry,
 308                                  const struct inode *inode,
 309                                  unsigned int open_flags)
 310 {
 311         struct dentry *real;
 312
 313         if (d_is_dir(dentry)) {
 314                 if (!inode || inode == d_inode(dentry))
 315                         return dentry;
 316                 goto bug;
 317         }
 318
 319         if (d_is_negative(dentry))
 320                 return dentry;
 321
 322         if (open_flags) {
 323                 int err = ovl_open_maybe_copy_up(dentry, open_flags);
 324
 325                 if (err)
 326                         return ERR_PTR(err);
 327         }
 328
 329         real = ovl_dentry_upper(dentry);
 330         if (real && (!inode || inode == d_inode(real)))
 331                 return real;
 332
 333         real = ovl_dentry_lower(dentry);
 334         if (!real)
 335                 goto bug;
 336
 337         if (!inode || inode == d_inode(real))
 338                 return real;
 339
 340         /* Handle recursion */
 341         return d_real(real, inode, open_flags);
 342 bug:
 343         WARN(1, "ovl_d_real(%pd4, %s:%lu\n): real dentry not found\n", dentry,
 344              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 345         return dentry;
 346 }
 347
 348 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 349 {
 350         struct ovl_entry *oe = dentry->d_fsdata;
 351         unsigned int i;
 352         int ret = 1;
 353
 354         for (i = 0; i < oe->numlower; i++) {
 355                 struct dentry *d = oe->lowerstack[i].dentry;
 356
 357                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 358                         ret = d->d_op->d_revalidate(d, flags);
 359                         if (ret < 0)
 360                                 return ret;
 361                         if (!ret) {
 362                                 if (!(flags & LOOKUP_RCU))
 363                                         d_invalidate(d);
 364                                 return -ESTALE;
 365                         }
 366                 }
 367         }
 368         return 1;
 369 }
 370
 371 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 372 {
 373         struct ovl_entry *oe = dentry->d_fsdata;
 374         unsigned int i;
 375         int ret = 1;
 376
 377         for (i = 0; i < oe->numlower; i++) {
 378                 struct dentry *d = oe->lowerstack[i].dentry;
 379
 380                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 381                         ret = d->d_op->d_weak_revalidate(d, flags);
 382                         if (ret <= 0)
 383                                 break;
 384                 }
 385         }
 386         return ret;
 387 }
 388
 389 static const struct dentry_operations ovl_dentry_operations = {
 390         .d_release = ovl_dentry_release,
 391         .d_real = ovl_d_real,
 392 };
 393
 394 static const struct dentry_operations ovl_reval_dentry_operations = {
 395         .d_release = ovl_dentry_release,
 396         .d_real = ovl_d_real,
 397         .d_revalidate = ovl_dentry_revalidate,
 398         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 399 };
 400
 401 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 402 {
 403         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 404         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 405
 406         if (oe)
 407                 oe->numlower = numlower;
 408
 409         return oe;
 410 }
 411
 412 static bool ovl_dentry_remote(struct dentry *dentry)
 413 {
 414         return dentry->d_flags &
 415                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 416 }
 417
 418 static bool ovl_dentry_weird(struct dentry *dentry)
 419 {
 420         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 421                                   DCACHE_MANAGE_TRANSIT |
 422                                   DCACHE_OP_HASH |
 423                                   DCACHE_OP_COMPARE);
 424 }
 425
 426 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 427                                              struct qstr *name)
 428 {
 429         struct dentry *dentry;
 430
 431         dentry = lookup_hash(name, dir);
 432
 433         if (IS_ERR(dentry)) {
 434                 if (PTR_ERR(dentry) == -ENOENT)
 435                         dentry = NULL;
 436         } else if (!dentry->d_inode) {
 437                 dput(dentry);
 438                 dentry = NULL;
 439         } else if (ovl_dentry_weird(dentry)) {
 440                 dput(dentry);
 441                 /* Don't support traversing automounts and other weirdness */
 442                 dentry = ERR_PTR(-EREMOTE);
 443         }
 444         return dentry;
 445 }
 446
 447 /*
 448  * Returns next layer in stack starting from top.
 449  * Returns -1 if this is the last layer.
 450  */
 451 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 452 {
 453         struct ovl_entry *oe = dentry->d_fsdata;
 454
 455         BUG_ON(idx < 0);
 456         if (idx == 0) {
 457                 ovl_path_upper(dentry, path);
 458                 if (path->dentry)
 459                         return oe->numlower ? 1 : -1;
 460                 idx++;
 461         }
 462         BUG_ON(idx > oe->numlower);
 463         *path = oe->lowerstack[idx - 1];
 464
 465         return (idx < oe->numlower) ? idx + 1 : -1;
 466 }
 467
 468 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 469                           unsigned int flags)
 470 {
 471         struct ovl_entry *oe;
 472         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 473         struct path *stack = NULL;
 474         struct dentry *upperdir, *upperdentry = NULL;
 475         unsigned int ctr = 0;
 476         struct inode *inode = NULL;
 477         bool upperopaque = false;
 478         struct dentry *this, *prev = NULL;
 479         unsigned int i;
 480         int err;
 481
 482         upperdir = ovl_upperdentry_dereference(poe);
 483         if (upperdir) {
 484                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 485                 err = PTR_ERR(this);
 486                 if (IS_ERR(this))
 487                         goto out;
 488
 489                 if (this) {
 490                         if (unlikely(ovl_dentry_remote(this))) {
 491                                 dput(this);
 492                                 err = -EREMOTE;
 493                                 goto out;
 494                         }
 495                         if (ovl_is_whiteout(this)) {
 496                                 dput(this);
 497                                 this = NULL;
 498                                 upperopaque = true;
 499                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 500                                 upperopaque = true;
 501                         }
 502                 }
 503                 upperdentry = prev = this;
 504         }
 505
 506         if (!upperopaque && poe->numlower) {
 507                 err = -ENOMEM;
 508                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 509                 if (!stack)
 510                         goto out_put_upper;
 511         }
 512
 513         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 514                 bool opaque = false;
 515                 struct path lowerpath = poe->lowerstack[i];
 516
 517                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 518                 err = PTR_ERR(this);
 519                 if (IS_ERR(this)) {
 520                         /*
 521                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 522                          */
 523                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 524                                 continue;
 525                         goto out_put;
 526                 }
 527                 if (!this)
 528                         continue;
 529                 if (ovl_is_whiteout(this)) {
 530                         dput(this);
 531                         break;
 532                 }
 533                 /*
 534                  * Only makes sense to check opaque dir if this is not the
 535                  * lowermost layer.
 536                  */
 537                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 538                         opaque = true;
 539
 540                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 541                              !S_ISDIR(this->d_inode->i_mode))) {
 542                         /*
 543                          * FIXME: check for upper-opaqueness maybe better done
 544                          * in remove code.
 545                          */
 546                         if (prev == upperdentry)
 547                                 upperopaque = true;
 548                         dput(this);
 549                         break;
 550                 }
 551                 /*
 552                  * If this is a non-directory then stop here.
 553                  */
 554                 if (!S_ISDIR(this->d_inode->i_mode))
 555                         opaque = true;
 556
 557                 stack[ctr].dentry = this;
 558                 stack[ctr].mnt = lowerpath.mnt;
 559                 ctr++;
 560                 prev = this;
 561                 if (opaque)
 562                         break;
 563         }
 564
 565         oe = ovl_alloc_entry(ctr);
 566         err = -ENOMEM;
 567         if (!oe)
 568                 goto out_put;
 569
 570         if (upperdentry || ctr) {
 571                 struct dentry *realdentry;
 572
 573                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 574
 575                 err = -ENOMEM;
 576                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 577                                       oe);
 578                 if (!inode)
 579                         goto out_free_oe;
 580                 ovl_copyattr(realdentry->d_inode, inode);
 581         }
 582
 583         oe->opaque = upperopaque;
 584         oe->__upperdentry = upperdentry;
 585         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 586         kfree(stack);
 587         dentry->d_fsdata = oe;
 588         d_add(dentry, inode);
 589
 590         return NULL;
 591
 592 out_free_oe:
 593         kfree(oe);
 594 out_put:
 595         for (i = 0; i < ctr; i++)
 596                 dput(stack[i].dentry);
 597         kfree(stack);
 598 out_put_upper:
 599         dput(upperdentry);
 600 out:
 601         return ERR_PTR(err);
 602 }
 603
 604 struct file *ovl_path_open(struct path *path, int flags)
 605 {
 606         return dentry_open(path, flags, current_cred());
 607 }
 608
 609 static void ovl_put_super(struct super_block *sb)
 610 {
 611         struct ovl_fs *ufs = sb->s_fs_info;
 612         unsigned i;
 613
 614         dput(ufs->workdir);
 615         mntput(ufs->upper_mnt);
 616         for (i = 0; i < ufs->numlower; i++)
 617                 mntput(ufs->lower_mnt[i]);
 618         kfree(ufs->lower_mnt);
 619
 620         kfree(ufs->config.lowerdir);
 621         kfree(ufs->config.upperdir);
 622         kfree(ufs->config.workdir);
 623         put_cred(ufs->creator_cred);
 624         kfree(ufs);
 625 }
 626
 627 /**
 628  * ovl_statfs
 629  * @sb: The overlayfs super block
 630  * @buf: The struct kstatfs to fill in with stats
 631  *
 632  * Get the filesystem statistics.  As writes always target the upper layer
 633  * filesystem pass the statfs to the upper filesystem (if it exists)
 634  */
 635 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 636 {
 637         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 638         struct dentry *root_dentry = dentry->d_sb->s_root;
 639         struct path path;
 640         int err;
 641
 642         ovl_path_real(root_dentry, &path);
 643
 644         err = vfs_statfs(&path, buf);
 645         if (!err) {
 646                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 647                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 648         }
 649
 650         return err;
 651 }
 652
 653 /**
 654  * ovl_show_options
 655  *
 656  * Prints the mount options for a given superblock.
 657  * Returns zero; does not fail.
 658  */
 659 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 660 {
 661         struct super_block *sb = dentry->d_sb;
 662         struct ovl_fs *ufs = sb->s_fs_info;
 663
 664         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 665         if (ufs->config.upperdir) {
 666                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 667                 seq_show_option(m, "workdir", ufs->config.workdir);
 668         }
 669         if (ufs->config.default_permissions)
 670                 seq_puts(m, ",default_permissions");
 671         return 0;
 672 }
 673
 674 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 675 {
 676         struct ovl_fs *ufs = sb->s_fs_info;
 677
 678         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 679                 return -EROFS;
 680
 681         return 0;
 682 }
 683
 684 static const struct super_operations ovl_super_operations = {
 685         .put_super      = ovl_put_super,
 686         .statfs         = ovl_statfs,
 687         .show_options   = ovl_show_options,
 688         .remount_fs     = ovl_remount,
 689 };
 690
 691 enum {
 692         OPT_LOWERDIR,
 693         OPT_UPPERDIR,
 694         OPT_WORKDIR,
 695         OPT_DEFAULT_PERMISSIONS,
 696         OPT_ERR,
 697 };
 698
 699 static const match_table_t ovl_tokens = {
 700         {OPT_LOWERDIR,                  "lowerdir=%s"},
 701         {OPT_UPPERDIR,                  "upperdir=%s"},
 702         {OPT_WORKDIR,                   "workdir=%s"},
 703         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 704         {OPT_ERR,                       NULL}
 705 };
 706
 707 static char *ovl_next_opt(char **s)
 708 {
 709         char *sbegin = *s;
 710         char *p;
 711
 712         if (sbegin == NULL)
 713                 return NULL;
 714
 715         for (p = sbegin; *p; p++) {
 716                 if (*p == '\\') {
 717                         p++;
 718                         if (!*p)
 719                                 break;
 720                 } else if (*p == ',') {
 721                         *p = '\0';
 722                         *s = p + 1;
 723                         return sbegin;
 724                 }
 725         }
 726         *s = NULL;
 727         return sbegin;
 728 }
 729
 730 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 731 {
 732         char *p;
 733
 734         while ((p = ovl_next_opt(&opt)) != NULL) {
 735                 int token;
 736                 substring_t args[MAX_OPT_ARGS];
 737
 738                 if (!*p)
 739                         continue;
 740
 741                 token = match_token(p, ovl_tokens, args);
 742                 switch (token) {
 743                 case OPT_UPPERDIR:
 744                         kfree(config->upperdir);
 745                         config->upperdir = match_strdup(&args[0]);
 746                         if (!config->upperdir)
 747                                 return -ENOMEM;
 748                         break;
 749
 750                 case OPT_LOWERDIR:
 751                         kfree(config->lowerdir);
 752                         config->lowerdir = match_strdup(&args[0]);
 753                         if (!config->lowerdir)
 754                                 return -ENOMEM;
 755                         break;
 756
 757                 case OPT_WORKDIR:
 758                         kfree(config->workdir);
 759                         config->workdir = match_strdup(&args[0]);
 760                         if (!config->workdir)
 761                                 return -ENOMEM;
 762                         break;
 763
 764                 case OPT_DEFAULT_PERMISSIONS:
 765                         config->default_permissions = true;
 766                         break;
 767
 768                 default:
 769                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 770                         return -EINVAL;
 771                 }
 772         }
 773
 774         /* Workdir is useless in non-upper mount */
 775         if (!config->upperdir && config->workdir) {
 776                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 777                         config->workdir);
 778                 kfree(config->workdir);
 779                 config->workdir = NULL;
 780         }
 781
 782         return 0;
 783 }
 784
 785 #define OVL_WORKDIR_NAME "work"
 786
 787 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 788                                          struct dentry *dentry)
 789 {
 790         struct inode *dir = dentry->d_inode;
 791         struct dentry *work;
 792         int err;
 793         bool retried = false;
 794
 795         err = mnt_want_write(mnt);
 796         if (err)
 797                 return ERR_PTR(err);
 798
 799         inode_lock_nested(dir, I_MUTEX_PARENT);
 800 retry:
 801         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 802                               strlen(OVL_WORKDIR_NAME));
 803
 804         if (!IS_ERR(work)) {
 805                 struct kstat stat = {
 806                         .mode = S_IFDIR | 0,
 807                 };
 808
 809                 if (work->d_inode) {
 810                         err = -EEXIST;
 811                         if (retried)
 812                                 goto out_dput;
 813
 814                         retried = true;
 815                         ovl_cleanup(dir, work);
 816                         dput(work);
 817                         goto retry;
 818                 }
 819
 820                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 821                 if (err)
 822                         goto out_dput;
 823         }
 824 out_unlock:
 825         inode_unlock(dir);
 826         mnt_drop_write(mnt);
 827
 828         return work;
 829
 830 out_dput:
 831         dput(work);
 832         work = ERR_PTR(err);
 833         goto out_unlock;
 834 }
 835
 836 static void ovl_unescape(char *s)
 837 {
 838         char *d = s;
 839
 840         for (;; s++, d++) {
 841                 if (*s == '\\')
 842                         s++;
 843                 *d = *s;
 844                 if (!*s)
 845                         break;
 846         }
 847 }
 848
 849 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 850 {
 851         int err = -EINVAL;
 852
 853         if (!*name) {
 854                 pr_err("overlayfs: empty lowerdir\n");
 855                 goto out;
 856         }
 857         err = kern_path(name, LOOKUP_FOLLOW, path);
 858         if (err) {
 859                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 860                 goto out;
 861         }
 862         err = -EINVAL;
 863         if (ovl_dentry_weird(path->dentry)) {
 864                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 865                 goto out_put;
 866         }
 867         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 868                 pr_err("overlayfs: '%s' not a directory\n", name);
 869                 goto out_put;
 870         }
 871         return 0;
 872
 873 out_put:
 874         path_put(path);
 875 out:
 876         return err;
 877 }
 878
 879 static int ovl_mount_dir(const char *name, struct path *path)
 880 {
 881         int err = -ENOMEM;
 882         char *tmp = kstrdup(name, GFP_KERNEL);
 883
 884         if (tmp) {
 885                 ovl_unescape(tmp);
 886                 err = ovl_mount_dir_noesc(tmp, path);
 887
 888                 if (!err)
 889                         if (ovl_dentry_remote(path->dentry)) {
 890                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 891                                        tmp);
 892                                 path_put(path);
 893                                 err = -EINVAL;
 894                         }
 895                 kfree(tmp);
 896         }
 897         return err;
 898 }
 899
 900 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 901                          int *stack_depth, bool *remote)
 902 {
 903         int err;
 904         struct kstatfs statfs;
 905
 906         err = ovl_mount_dir_noesc(name, path);
 907         if (err)
 908                 goto out;
 909
 910         err = vfs_statfs(path, &statfs);
 911         if (err) {
 912                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 913                 goto out_put;
 914         }
 915         *namelen = max(*namelen, statfs.f_namelen);
 916         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 917
 918         if (ovl_dentry_remote(path->dentry))
 919                 *remote = true;
 920
 921         return 0;
 922
 923 out_put:
 924         path_put(path);
 925 out:
 926         return err;
 927 }
 928
 929 /* Workdir should not be subdir of upperdir and vice versa */
 930 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 931 {
 932         bool ok = false;
 933
 934         if (workdir != upperdir) {
 935                 ok = (lock_rename(workdir, upperdir) == NULL);
 936                 unlock_rename(workdir, upperdir);
 937         }
 938         return ok;
 939 }
 940
 941 static unsigned int ovl_split_lowerdirs(char *str)
 942 {
 943         unsigned int ctr = 1;
 944         char *s, *d;
 945
 946         for (s = d = str;; s++, d++) {
 947                 if (*s == '\\') {
 948                         s++;
 949                 } else if (*s == ':') {
 950                         *d = '\0';
 951                         ctr++;
 952                         continue;
 953                 }
 954                 *d = *s;
 955                 if (!*s)
 956                         break;
 957         }
 958         return ctr;
 959 }
 960
 961 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 962 {
 963         struct path upperpath = { NULL, NULL };
 964         struct path workpath = { NULL, NULL };
 965         struct dentry *root_dentry;
 966         struct ovl_entry *oe;
 967         struct ovl_fs *ufs;
 968         struct path *stack = NULL;
 969         char *lowertmp;
 970         char *lower;
 971         unsigned int numlower;
 972         unsigned int stacklen = 0;
 973         unsigned int i;
 974         bool remote = false;
 975         int err;
 976
 977         err = -ENOMEM;
 978         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 979         if (!ufs)
 980                 goto out;
 981
 982         err = ovl_parse_opt((char *) data, &ufs->config);
 983         if (err)
 984                 goto out_free_config;
 985
 986         err = -EINVAL;
 987         if (!ufs->config.lowerdir) {
 988                 if (!silent)
 989                         pr_err("overlayfs: missing 'lowerdir'\n");
 990                 goto out_free_config;
 991         }
 992
 993         sb->s_stack_depth = 0;
 994         sb->s_maxbytes = MAX_LFS_FILESIZE;
 995         if (ufs->config.upperdir) {
 996                 if (!ufs->config.workdir) {
 997                         pr_err("overlayfs: missing 'workdir'\n");
 998                         goto out_free_config;
 999                 }
1000
1001                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
1002                 if (err)
1003                         goto out_free_config;
1004
1005                 /* Upper fs should not be r/o */
1006                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
1007                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
1008                         err = -EINVAL;
1009                         goto out_put_upperpath;
1010                 }
1011
1012                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
1013                 if (err)
1014                         goto out_put_upperpath;
1015
1016                 err = -EINVAL;
1017                 if (upperpath.mnt != workpath.mnt) {
1018                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1019                         goto out_put_workpath;
1020                 }
1021                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
1022                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1023                         goto out_put_workpath;
1024                 }
1025                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
1026         }
1027         err = -ENOMEM;
1028         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
1029         if (!lowertmp)
1030                 goto out_put_workpath;
1031
1032         err = -EINVAL;
1033         stacklen = ovl_split_lowerdirs(lowertmp);
1034         if (stacklen > OVL_MAX_STACK) {
1035                 pr_err("overlayfs: too many lower directries, limit is %d\n",
1036                        OVL_MAX_STACK);
1037                 goto out_free_lowertmp;
1038         } else if (!ufs->config.upperdir && stacklen == 1) {
1039                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1040                 goto out_free_lowertmp;
1041         }
1042
1043         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1044         if (!stack)
1045                 goto out_free_lowertmp;
1046
1047         lower = lowertmp;
1048         for (numlower = 0; numlower < stacklen; numlower++) {
1049                 err = ovl_lower_dir(lower, &stack[numlower],
1050                                     &ufs->lower_namelen, &sb->s_stack_depth,
1051                                     &remote);
1052                 if (err)
1053                         goto out_put_lowerpath;
1054
1055                 lower = strchr(lower, '\0') + 1;
1056         }
1057
1058         err = -EINVAL;
1059         sb->s_stack_depth++;
1060         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1061                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1062                 goto out_put_lowerpath;
1063         }
1064
1065         if (ufs->config.upperdir) {
1066                 ufs->upper_mnt = clone_private_mount(&upperpath);
1067                 err = PTR_ERR(ufs->upper_mnt);
1068                 if (IS_ERR(ufs->upper_mnt)) {
1069                         pr_err("overlayfs: failed to clone upperpath\n");
1070                         goto out_put_lowerpath;
1071                 }
1072
1073                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
1074                 err = PTR_ERR(ufs->workdir);
1075                 if (IS_ERR(ufs->workdir)) {
1076                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1077                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1078                         sb->s_flags |= MS_RDONLY;
1079                         ufs->workdir = NULL;
1080                 }
1081
1082                 /*
1083                  * Upper should support d_type, else whiteouts are visible.
1084                  * Given workdir and upper are on same fs, we can do
1085                  * iterate_dir() on workdir. This check requires successful
1086                  * creation of workdir in previous step.
1087                  */
1088                 if (ufs->workdir) {
1089                         err = ovl_check_d_type_supported(&workpath);
1090                         if (err < 0)
1091                                 goto out_put_workdir;
1092
1093                         if (!err) {
1094                                 pr_err("overlayfs: upper fs needs to support d_type.\n");
1095                                 err = -EINVAL;
1096                                 goto out_put_workdir;
1097                         }
1098                 }
1099         }
1100
1101         err = -ENOMEM;
1102         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1103         if (ufs->lower_mnt == NULL)
1104                 goto out_put_workdir;
1105         for (i = 0; i < numlower; i++) {
1106                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1107
1108                 err = PTR_ERR(mnt);
1109                 if (IS_ERR(mnt)) {
1110                         pr_err("overlayfs: failed to clone lowerpath\n");
1111                         goto out_put_lower_mnt;
1112                 }
1113                 /*
1114                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1115                  * will fail instead of modifying lower fs.
1116                  */
1117                 mnt->mnt_flags |= MNT_READONLY;
1118
1119                 ufs->lower_mnt[ufs->numlower] = mnt;
1120                 ufs->numlower++;
1121         }
1122
1123         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1124         if (!ufs->upper_mnt)
1125                 sb->s_flags |= MS_RDONLY;
1126
1127         if (remote)
1128                 sb->s_d_op = &ovl_reval_dentry_operations;
1129         else
1130                 sb->s_d_op = &ovl_dentry_operations;
1131
1132         ufs->creator_cred = prepare_creds();
1133         if (!ufs->creator_cred)
1134                 goto out_put_lower_mnt;
1135
1136         err = -ENOMEM;
1137         oe = ovl_alloc_entry(numlower);
1138         if (!oe)
1139                 goto out_put_cred;
1140
1141         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1142         if (!root_dentry)
1143                 goto out_free_oe;
1144
1145         mntput(upperpath.mnt);
1146         for (i = 0; i < numlower; i++)
1147                 mntput(stack[i].mnt);
1148         path_put(&workpath);
1149         kfree(lowertmp);
1150
1151         oe->__upperdentry = upperpath.dentry;
1152         for (i = 0; i < numlower; i++) {
1153                 oe->lowerstack[i].dentry = stack[i].dentry;
1154                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1155         }
1156         kfree(stack);
1157
1158         root_dentry->d_fsdata = oe;
1159
1160         ovl_copyattr(ovl_dentry_real(root_dentry)->d_inode,
1161                      root_dentry->d_inode);
1162
1163         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1164         sb->s_op = &ovl_super_operations;
1165         sb->s_root = root_dentry;
1166         sb->s_fs_info = ufs;
1167
1168         return 0;
1169
1170 out_free_oe:
1171         kfree(oe);
1172 out_put_cred:
1173         put_cred(ufs->creator_cred);
1174 out_put_lower_mnt:
1175         for (i = 0; i < ufs->numlower; i++)
1176                 mntput(ufs->lower_mnt[i]);
1177         kfree(ufs->lower_mnt);
1178 out_put_workdir:
1179         dput(ufs->workdir);
1180         mntput(ufs->upper_mnt);
1181 out_put_lowerpath:
1182         for (i = 0; i < numlower; i++)
1183                 path_put(&stack[i]);
1184         kfree(stack);
1185 out_free_lowertmp:
1186         kfree(lowertmp);
1187 out_put_workpath:
1188         path_put(&workpath);
1189 out_put_upperpath:
1190         path_put(&upperpath);
1191 out_free_config:
1192         kfree(ufs->config.lowerdir);
1193         kfree(ufs->config.upperdir);
1194         kfree(ufs->config.workdir);
1195         kfree(ufs);
1196 out:
1197         return err;
1198 }
1199
1200 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1201                                 const char *dev_name, void *raw_data)
1202 {
1203         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1204 }
1205
1206 static struct file_system_type ovl_fs_type = {
1207         .owner          = THIS_MODULE,
1208         .name           = "overlay",
1209         .mount          = ovl_mount,
1210         .kill_sb        = kill_anon_super,
1211 };
1212 MODULE_ALIAS_FS("overlay");
1213
1214 static int __init ovl_init(void)
1215 {
1216         return register_filesystem(&ovl_fs_type);
1217 }
1218
1219 static void __exit ovl_exit(void)
1220 {
1221         unregister_filesystem(&ovl_fs_type);
1222 }
1223
1224 module_init(ovl_init);
1225 module_exit(ovl_exit);