kernel/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/sched.h>
  20 #include <linux/statfs.h>
  21 #include <linux/seq_file.h>
  22 #include "overlayfs.h"
  23
  24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  25 MODULE_DESCRIPTION("Overlay filesystem");
  26 MODULE_LICENSE("GPL");
  27
  28 #define OVERLAYFS_SUPER_MAGIC 0x794c7630
  29
  30 struct ovl_config {
  31         char *lowerdir;
  32         char *upperdir;
  33         char *workdir;
  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 ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 161 {
 162         struct ovl_entry *oe = dentry->d_fsdata;
 163
 164         return oe->cache;
 165 }
 166
 167 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 168 {
 169         struct ovl_entry *oe = dentry->d_fsdata;
 170
 171         oe->cache = cache;
 172 }
 173
 174 void ovl_path_lower(struct dentry *dentry, struct path *path)
 175 {
 176         struct ovl_entry *oe = dentry->d_fsdata;
 177
 178         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 179 }
 180
 181 int ovl_want_write(struct dentry *dentry)
 182 {
 183         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 184         return mnt_want_write(ofs->upper_mnt);
 185 }
 186
 187 void ovl_drop_write(struct dentry *dentry)
 188 {
 189         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 190         mnt_drop_write(ofs->upper_mnt);
 191 }
 192
 193 struct dentry *ovl_workdir(struct dentry *dentry)
 194 {
 195         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 196         return ofs->workdir;
 197 }
 198
 199 bool ovl_dentry_is_opaque(struct dentry *dentry)
 200 {
 201         struct ovl_entry *oe = dentry->d_fsdata;
 202         return oe->opaque;
 203 }
 204
 205 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 206 {
 207         struct ovl_entry *oe = dentry->d_fsdata;
 208         oe->opaque = opaque;
 209 }
 210
 211 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 212 {
 213         struct ovl_entry *oe = dentry->d_fsdata;
 214
 215         WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
 216         WARN_ON(oe->__upperdentry);
 217         BUG_ON(!upperdentry->d_inode);
 218         /*
 219          * Make sure upperdentry is consistent before making it visible to
 220          * ovl_upperdentry_dereference().
 221          */
 222         smp_wmb();
 223         oe->__upperdentry = upperdentry;
 224 }
 225
 226 void ovl_dentry_version_inc(struct dentry *dentry)
 227 {
 228         struct ovl_entry *oe = dentry->d_fsdata;
 229
 230         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 231         oe->version++;
 232 }
 233
 234 u64 ovl_dentry_version_get(struct dentry *dentry)
 235 {
 236         struct ovl_entry *oe = dentry->d_fsdata;
 237
 238         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 239         return oe->version;
 240 }
 241
 242 bool ovl_is_whiteout(struct dentry *dentry)
 243 {
 244         struct inode *inode = dentry->d_inode;
 245
 246         return inode && IS_WHITEOUT(inode);
 247 }
 248
 249 static bool ovl_is_opaquedir(struct dentry *dentry)
 250 {
 251         int res;
 252         char val;
 253         struct inode *inode = dentry->d_inode;
 254
 255         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
 256                 return false;
 257
 258         res = inode->i_op->getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
 259         if (res == 1 && val == 'y')
 260                 return true;
 261
 262         return false;
 263 }
 264
 265 static void ovl_dentry_release(struct dentry *dentry)
 266 {
 267         struct ovl_entry *oe = dentry->d_fsdata;
 268
 269         if (oe) {
 270                 unsigned int i;
 271
 272                 dput(oe->__upperdentry);
 273                 for (i = 0; i < oe->numlower; i++)
 274                         dput(oe->lowerstack[i].dentry);
 275                 kfree_rcu(oe, rcu);
 276         }
 277 }
 278
 279 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 280 {
 281         struct ovl_entry *oe = dentry->d_fsdata;
 282         unsigned int i;
 283         int ret = 1;
 284
 285         for (i = 0; i < oe->numlower; i++) {
 286                 struct dentry *d = oe->lowerstack[i].dentry;
 287
 288                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 289                         ret = d->d_op->d_revalidate(d, flags);
 290                         if (ret < 0)
 291                                 return ret;
 292                         if (!ret) {
 293                                 if (!(flags & LOOKUP_RCU))
 294                                         d_invalidate(d);
 295                                 return -ESTALE;
 296                         }
 297                 }
 298         }
 299         return 1;
 300 }
 301
 302 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 303 {
 304         struct ovl_entry *oe = dentry->d_fsdata;
 305         unsigned int i;
 306         int ret = 1;
 307
 308         for (i = 0; i < oe->numlower; i++) {
 309                 struct dentry *d = oe->lowerstack[i].dentry;
 310
 311                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 312                         ret = d->d_op->d_weak_revalidate(d, flags);
 313                         if (ret <= 0)
 314                                 break;
 315                 }
 316         }
 317         return ret;
 318 }
 319
 320 static const struct dentry_operations ovl_dentry_operations = {
 321         .d_release = ovl_dentry_release,
 322         .d_select_inode = ovl_d_select_inode,
 323 };
 324
 325 static const struct dentry_operations ovl_reval_dentry_operations = {
 326         .d_release = ovl_dentry_release,
 327         .d_select_inode = ovl_d_select_inode,
 328         .d_revalidate = ovl_dentry_revalidate,
 329         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 330 };
 331
 332 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 333 {
 334         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 335         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 336
 337         if (oe)
 338                 oe->numlower = numlower;
 339
 340         return oe;
 341 }
 342
 343 static bool ovl_dentry_remote(struct dentry *dentry)
 344 {
 345         return dentry->d_flags &
 346                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 347 }
 348
 349 static bool ovl_dentry_weird(struct dentry *dentry)
 350 {
 351         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 352                                   DCACHE_MANAGE_TRANSIT |
 353                                   DCACHE_OP_HASH |
 354                                   DCACHE_OP_COMPARE);
 355 }
 356
 357 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 358                                              struct qstr *name)
 359 {
 360         struct dentry *dentry;
 361
 362         mutex_lock(&dir->d_inode->i_mutex);
 363         dentry = lookup_one_len(name->name, dir, name->len);
 364         mutex_unlock(&dir->d_inode->i_mutex);
 365
 366         if (IS_ERR(dentry)) {
 367                 if (PTR_ERR(dentry) == -ENOENT)
 368                         dentry = NULL;
 369         } else if (!dentry->d_inode) {
 370                 dput(dentry);
 371                 dentry = NULL;
 372         } else if (ovl_dentry_weird(dentry)) {
 373                 dput(dentry);
 374                 /* Don't support traversing automounts and other weirdness */
 375                 dentry = ERR_PTR(-EREMOTE);
 376         }
 377         return dentry;
 378 }
 379
 380 /*
 381  * Returns next layer in stack starting from top.
 382  * Returns -1 if this is the last layer.
 383  */
 384 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 385 {
 386         struct ovl_entry *oe = dentry->d_fsdata;
 387
 388         BUG_ON(idx < 0);
 389         if (idx == 0) {
 390                 ovl_path_upper(dentry, path);
 391                 if (path->dentry)
 392                         return oe->numlower ? 1 : -1;
 393                 idx++;
 394         }
 395         BUG_ON(idx > oe->numlower);
 396         *path = oe->lowerstack[idx - 1];
 397
 398         return (idx < oe->numlower) ? idx + 1 : -1;
 399 }
 400
 401 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 402                           unsigned int flags)
 403 {
 404         struct ovl_entry *oe;
 405         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 406         struct path *stack = NULL;
 407         struct dentry *upperdir, *upperdentry = NULL;
 408         unsigned int ctr = 0;
 409         struct inode *inode = NULL;
 410         bool upperopaque = false;
 411         struct dentry *this, *prev = NULL;
 412         unsigned int i;
 413         int err;
 414
 415         upperdir = ovl_upperdentry_dereference(poe);
 416         if (upperdir) {
 417                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 418                 err = PTR_ERR(this);
 419                 if (IS_ERR(this))
 420                         goto out;
 421
 422                 if (this) {
 423                         if (unlikely(ovl_dentry_remote(this))) {
 424                                 dput(this);
 425                                 err = -EREMOTE;
 426                                 goto out;
 427                         }
 428                         if (ovl_is_whiteout(this)) {
 429                                 dput(this);
 430                                 this = NULL;
 431                                 upperopaque = true;
 432                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 433                                 upperopaque = true;
 434                         }
 435                 }
 436                 upperdentry = prev = this;
 437         }
 438
 439         if (!upperopaque && poe->numlower) {
 440                 err = -ENOMEM;
 441                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 442                 if (!stack)
 443                         goto out_put_upper;
 444         }
 445
 446         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 447                 bool opaque = false;
 448                 struct path lowerpath = poe->lowerstack[i];
 449
 450                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 451                 err = PTR_ERR(this);
 452                 if (IS_ERR(this)) {
 453                         /*
 454                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 455                          */
 456                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 457                                 continue;
 458                         goto out_put;
 459                 }
 460                 if (!this)
 461                         continue;
 462                 if (ovl_is_whiteout(this)) {
 463                         dput(this);
 464                         break;
 465                 }
 466                 /*
 467                  * Only makes sense to check opaque dir if this is not the
 468                  * lowermost layer.
 469                  */
 470                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 471                         opaque = true;
 472
 473                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 474                              !S_ISDIR(this->d_inode->i_mode))) {
 475                         /*
 476                          * FIXME: check for upper-opaqueness maybe better done
 477                          * in remove code.
 478                          */
 479                         if (prev == upperdentry)
 480                                 upperopaque = true;
 481                         dput(this);
 482                         break;
 483                 }
 484                 /*
 485                  * If this is a non-directory then stop here.
 486                  */
 487                 if (!S_ISDIR(this->d_inode->i_mode))
 488                         opaque = true;
 489
 490                 stack[ctr].dentry = this;
 491                 stack[ctr].mnt = lowerpath.mnt;
 492                 ctr++;
 493                 prev = this;
 494                 if (opaque)
 495                         break;
 496         }
 497
 498         oe = ovl_alloc_entry(ctr);
 499         err = -ENOMEM;
 500         if (!oe)
 501                 goto out_put;
 502
 503         if (upperdentry || ctr) {
 504                 struct dentry *realdentry;
 505
 506                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 507
 508                 err = -ENOMEM;
 509                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 510                                       oe);
 511                 if (!inode)
 512                         goto out_free_oe;
 513                 ovl_copyattr(realdentry->d_inode, inode);
 514         }
 515
 516         oe->opaque = upperopaque;
 517         oe->__upperdentry = upperdentry;
 518         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 519         kfree(stack);
 520         dentry->d_fsdata = oe;
 521         d_add(dentry, inode);
 522
 523         return NULL;
 524
 525 out_free_oe:
 526         kfree(oe);
 527 out_put:
 528         for (i = 0; i < ctr; i++)
 529                 dput(stack[i].dentry);
 530         kfree(stack);
 531 out_put_upper:
 532         dput(upperdentry);
 533 out:
 534         return ERR_PTR(err);
 535 }
 536
 537 struct file *ovl_path_open(struct path *path, int flags)
 538 {
 539         return dentry_open(path, flags, current_cred());
 540 }
 541
 542 static void ovl_put_super(struct super_block *sb)
 543 {
 544         struct ovl_fs *ufs = sb->s_fs_info;
 545         unsigned i;
 546
 547         dput(ufs->workdir);
 548         mntput(ufs->upper_mnt);
 549         for (i = 0; i < ufs->numlower; i++)
 550                 mntput(ufs->lower_mnt[i]);
 551         kfree(ufs->lower_mnt);
 552
 553         kfree(ufs->config.lowerdir);
 554         kfree(ufs->config.upperdir);
 555         kfree(ufs->config.workdir);
 556         kfree(ufs);
 557 }
 558
 559 /**
 560  * ovl_statfs
 561  * @sb: The overlayfs super block
 562  * @buf: The struct kstatfs to fill in with stats
 563  *
 564  * Get the filesystem statistics.  As writes always target the upper layer
 565  * filesystem pass the statfs to the upper filesystem (if it exists)
 566  */
 567 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 568 {
 569         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 570         struct dentry *root_dentry = dentry->d_sb->s_root;
 571         struct path path;
 572         int err;
 573
 574         ovl_path_real(root_dentry, &path);
 575
 576         err = vfs_statfs(&path, buf);
 577         if (!err) {
 578                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 579                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 580         }
 581
 582         return err;
 583 }
 584
 585 /**
 586  * ovl_show_options
 587  *
 588  * Prints the mount options for a given superblock.
 589  * Returns zero; does not fail.
 590  */
 591 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 592 {
 593         struct super_block *sb = dentry->d_sb;
 594         struct ovl_fs *ufs = sb->s_fs_info;
 595
 596         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 597         if (ufs->config.upperdir) {
 598                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 599                 seq_show_option(m, "workdir", ufs->config.workdir);
 600         }
 601         return 0;
 602 }
 603
 604 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 605 {
 606         struct ovl_fs *ufs = sb->s_fs_info;
 607
 608         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 609                 return -EROFS;
 610
 611         return 0;
 612 }
 613
 614 static const struct super_operations ovl_super_operations = {
 615         .put_super      = ovl_put_super,
 616         .statfs         = ovl_statfs,
 617         .show_options   = ovl_show_options,
 618         .remount_fs     = ovl_remount,
 619 };
 620
 621 enum {
 622         OPT_LOWERDIR,
 623         OPT_UPPERDIR,
 624         OPT_WORKDIR,
 625         OPT_ERR,
 626 };
 627
 628 static const match_table_t ovl_tokens = {
 629         {OPT_LOWERDIR,                  "lowerdir=%s"},
 630         {OPT_UPPERDIR,                  "upperdir=%s"},
 631         {OPT_WORKDIR,                   "workdir=%s"},
 632         {OPT_ERR,                       NULL}
 633 };
 634
 635 static char *ovl_next_opt(char **s)
 636 {
 637         char *sbegin = *s;
 638         char *p;
 639
 640         if (sbegin == NULL)
 641                 return NULL;
 642
 643         for (p = sbegin; *p; p++) {
 644                 if (*p == '\\') {
 645                         p++;
 646                         if (!*p)
 647                                 break;
 648                 } else if (*p == ',') {
 649                         *p = '\0';
 650                         *s = p + 1;
 651                         return sbegin;
 652                 }
 653         }
 654         *s = NULL;
 655         return sbegin;
 656 }
 657
 658 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 659 {
 660         char *p;
 661
 662         while ((p = ovl_next_opt(&opt)) != NULL) {
 663                 int token;
 664                 substring_t args[MAX_OPT_ARGS];
 665
 666                 if (!*p)
 667                         continue;
 668
 669                 token = match_token(p, ovl_tokens, args);
 670                 switch (token) {
 671                 case OPT_UPPERDIR:
 672                         kfree(config->upperdir);
 673                         config->upperdir = match_strdup(&args[0]);
 674                         if (!config->upperdir)
 675                                 return -ENOMEM;
 676                         break;
 677
 678                 case OPT_LOWERDIR:
 679                         kfree(config->lowerdir);
 680                         config->lowerdir = match_strdup(&args[0]);
 681                         if (!config->lowerdir)
 682                                 return -ENOMEM;
 683                         break;
 684
 685                 case OPT_WORKDIR:
 686                         kfree(config->workdir);
 687                         config->workdir = match_strdup(&args[0]);
 688                         if (!config->workdir)
 689                                 return -ENOMEM;
 690                         break;
 691
 692                 default:
 693                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 694                         return -EINVAL;
 695                 }
 696         }
 697
 698         /* Workdir is useless in non-upper mount */
 699         if (!config->upperdir && config->workdir) {
 700                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 701                         config->workdir);
 702                 kfree(config->workdir);
 703                 config->workdir = NULL;
 704         }
 705
 706         return 0;
 707 }
 708
 709 #define OVL_WORKDIR_NAME "work"
 710
 711 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 712                                          struct dentry *dentry)
 713 {
 714         struct inode *dir = dentry->d_inode;
 715         struct dentry *work;
 716         int err;
 717         bool retried = false;
 718
 719         err = mnt_want_write(mnt);
 720         if (err)
 721                 return ERR_PTR(err);
 722
 723         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 724 retry:
 725         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 726                               strlen(OVL_WORKDIR_NAME));
 727
 728         if (!IS_ERR(work)) {
 729                 struct kstat stat = {
 730                         .mode = S_IFDIR | 0,
 731                 };
 732
 733                 if (work->d_inode) {
 734                         err = -EEXIST;
 735                         if (retried)
 736                                 goto out_dput;
 737
 738                         retried = true;
 739                         ovl_cleanup(dir, work);
 740                         dput(work);
 741                         goto retry;
 742                 }
 743
 744                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 745                 if (err)
 746                         goto out_dput;
 747         }
 748 out_unlock:
 749         mutex_unlock(&dir->i_mutex);
 750         mnt_drop_write(mnt);
 751
 752         return work;
 753
 754 out_dput:
 755         dput(work);
 756         work = ERR_PTR(err);
 757         goto out_unlock;
 758 }
 759
 760 static void ovl_unescape(char *s)
 761 {
 762         char *d = s;
 763
 764         for (;; s++, d++) {
 765                 if (*s == '\\')
 766                         s++;
 767                 *d = *s;
 768                 if (!*s)
 769                         break;
 770         }
 771 }
 772
 773 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 774 {
 775         int err = -EINVAL;
 776
 777         if (!*name) {
 778                 pr_err("overlayfs: empty lowerdir\n");
 779                 goto out;
 780         }
 781         err = kern_path(name, LOOKUP_FOLLOW, path);
 782         if (err) {
 783                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 784                 goto out;
 785         }
 786         err = -EINVAL;
 787         if (ovl_dentry_weird(path->dentry)) {
 788                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 789                 goto out_put;
 790         }
 791         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 792                 pr_err("overlayfs: '%s' not a directory\n", name);
 793                 goto out_put;
 794         }
 795         return 0;
 796
 797 out_put:
 798         path_put(path);
 799 out:
 800         return err;
 801 }
 802
 803 static int ovl_mount_dir(const char *name, struct path *path)
 804 {
 805         int err = -ENOMEM;
 806         char *tmp = kstrdup(name, GFP_KERNEL);
 807
 808         if (tmp) {
 809                 ovl_unescape(tmp);
 810                 err = ovl_mount_dir_noesc(tmp, path);
 811
 812                 if (!err)
 813                         if (ovl_dentry_remote(path->dentry)) {
 814                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 815                                        tmp);
 816                                 path_put(path);
 817                                 err = -EINVAL;
 818                         }
 819                 kfree(tmp);
 820         }
 821         return err;
 822 }
 823
 824 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 825                          int *stack_depth, bool *remote)
 826 {
 827         int err;
 828         struct kstatfs statfs;
 829
 830         err = ovl_mount_dir_noesc(name, path);
 831         if (err)
 832                 goto out;
 833
 834         err = vfs_statfs(path, &statfs);
 835         if (err) {
 836                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 837                 goto out_put;
 838         }
 839         *namelen = max(*namelen, statfs.f_namelen);
 840         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 841
 842         if (ovl_dentry_remote(path->dentry))
 843                 *remote = true;
 844
 845         return 0;
 846
 847 out_put:
 848         path_put(path);
 849 out:
 850         return err;
 851 }
 852
 853 /* Workdir should not be subdir of upperdir and vice versa */
 854 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 855 {
 856         bool ok = false;
 857
 858         if (workdir != upperdir) {
 859                 ok = (lock_rename(workdir, upperdir) == NULL);
 860                 unlock_rename(workdir, upperdir);
 861         }
 862         return ok;
 863 }
 864
 865 static unsigned int ovl_split_lowerdirs(char *str)
 866 {
 867         unsigned int ctr = 1;
 868         char *s, *d;
 869
 870         for (s = d = str;; s++, d++) {
 871                 if (*s == '\\') {
 872                         s++;
 873                 } else if (*s == ':') {
 874                         *d = '\0';
 875                         ctr++;
 876                         continue;
 877                 }
 878                 *d = *s;
 879                 if (!*s)
 880                         break;
 881         }
 882         return ctr;
 883 }
 884
 885 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 886 {
 887         struct path upperpath = { NULL, NULL };
 888         struct path workpath = { NULL, NULL };
 889         struct dentry *root_dentry;
 890         struct ovl_entry *oe;
 891         struct ovl_fs *ufs;
 892         struct path *stack = NULL;
 893         char *lowertmp;
 894         char *lower;
 895         unsigned int numlower;
 896         unsigned int stacklen = 0;
 897         unsigned int i;
 898         bool remote = false;
 899         int err;
 900
 901         err = -ENOMEM;
 902         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 903         if (!ufs)
 904                 goto out;
 905
 906         err = ovl_parse_opt((char *) data, &ufs->config);
 907         if (err)
 908                 goto out_free_config;
 909
 910         err = -EINVAL;
 911         if (!ufs->config.lowerdir) {
 912                 pr_err("overlayfs: missing 'lowerdir'\n");
 913                 goto out_free_config;
 914         }
 915
 916         sb->s_stack_depth = 0;
 917         sb->s_maxbytes = MAX_LFS_FILESIZE;
 918         if (ufs->config.upperdir) {
 919                 if (!ufs->config.workdir) {
 920                         pr_err("overlayfs: missing 'workdir'\n");
 921                         goto out_free_config;
 922                 }
 923
 924                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 925                 if (err)
 926                         goto out_free_config;
 927
 928                 /* Upper fs should not be r/o */
 929                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 930                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 931                         err = -EINVAL;
 932                         goto out_put_upperpath;
 933                 }
 934
 935                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 936                 if (err)
 937                         goto out_put_upperpath;
 938
 939                 err = -EINVAL;
 940                 if (upperpath.mnt != workpath.mnt) {
 941                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 942                         goto out_put_workpath;
 943                 }
 944                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 945                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 946                         goto out_put_workpath;
 947                 }
 948                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 949         }
 950         err = -ENOMEM;
 951         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 952         if (!lowertmp)
 953                 goto out_put_workpath;
 954
 955         err = -EINVAL;
 956         stacklen = ovl_split_lowerdirs(lowertmp);
 957         if (stacklen > OVL_MAX_STACK) {
 958                 pr_err("overlayfs: too many lower directries, limit is %d\n",
 959                        OVL_MAX_STACK);
 960                 goto out_free_lowertmp;
 961         } else if (!ufs->config.upperdir && stacklen == 1) {
 962                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 963                 goto out_free_lowertmp;
 964         }
 965
 966         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 967         if (!stack)
 968                 goto out_free_lowertmp;
 969
 970         lower = lowertmp;
 971         for (numlower = 0; numlower < stacklen; numlower++) {
 972                 err = ovl_lower_dir(lower, &stack[numlower],
 973                                     &ufs->lower_namelen, &sb->s_stack_depth,
 974                                     &remote);
 975                 if (err)
 976                         goto out_put_lowerpath;
 977
 978                 lower = strchr(lower, '\0') + 1;
 979         }
 980
 981         err = -EINVAL;
 982         sb->s_stack_depth++;
 983         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
 984                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
 985                 goto out_put_lowerpath;
 986         }
 987
 988         if (ufs->config.upperdir) {
 989                 ufs->upper_mnt = clone_private_mount(&upperpath);
 990                 err = PTR_ERR(ufs->upper_mnt);
 991                 if (IS_ERR(ufs->upper_mnt)) {
 992                         pr_err("overlayfs: failed to clone upperpath\n");
 993                         goto out_put_lowerpath;
 994                 }
 995
 996                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
 997                 err = PTR_ERR(ufs->workdir);
 998                 if (IS_ERR(ufs->workdir)) {
 999                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1000                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1001                         sb->s_flags |= MS_RDONLY;
1002                         ufs->workdir = NULL;
1003                 }
1004         }
1005
1006         err = -ENOMEM;
1007         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1008         if (ufs->lower_mnt == NULL)
1009                 goto out_put_workdir;
1010         for (i = 0; i < numlower; i++) {
1011                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1012
1013                 err = PTR_ERR(mnt);
1014                 if (IS_ERR(mnt)) {
1015                         pr_err("overlayfs: failed to clone lowerpath\n");
1016                         goto out_put_lower_mnt;
1017                 }
1018                 /*
1019                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1020                  * will fail instead of modifying lower fs.
1021                  */
1022                 mnt->mnt_flags |= MNT_READONLY;
1023
1024                 ufs->lower_mnt[ufs->numlower] = mnt;
1025                 ufs->numlower++;
1026         }
1027
1028         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1029         if (!ufs->upper_mnt)
1030                 sb->s_flags |= MS_RDONLY;
1031
1032         if (remote)
1033                 sb->s_d_op = &ovl_reval_dentry_operations;
1034         else
1035                 sb->s_d_op = &ovl_dentry_operations;
1036
1037         err = -ENOMEM;
1038         oe = ovl_alloc_entry(numlower);
1039         if (!oe)
1040                 goto out_put_lower_mnt;
1041
1042         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1043         if (!root_dentry)
1044                 goto out_free_oe;
1045
1046         mntput(upperpath.mnt);
1047         for (i = 0; i < numlower; i++)
1048                 mntput(stack[i].mnt);
1049         path_put(&workpath);
1050         kfree(lowertmp);
1051
1052         oe->__upperdentry = upperpath.dentry;
1053         for (i = 0; i < numlower; i++) {
1054                 oe->lowerstack[i].dentry = stack[i].dentry;
1055                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1056         }
1057         kfree(stack);
1058
1059         root_dentry->d_fsdata = oe;
1060
1061         ovl_copyattr(ovl_dentry_real(root_dentry)->d_inode,
1062                      root_dentry->d_inode);
1063
1064         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1065         sb->s_op = &ovl_super_operations;
1066         sb->s_root = root_dentry;
1067         sb->s_fs_info = ufs;
1068
1069         return 0;
1070
1071 out_free_oe:
1072         kfree(oe);
1073 out_put_lower_mnt:
1074         for (i = 0; i < ufs->numlower; i++)
1075                 mntput(ufs->lower_mnt[i]);
1076         kfree(ufs->lower_mnt);
1077 out_put_workdir:
1078         dput(ufs->workdir);
1079         mntput(ufs->upper_mnt);
1080 out_put_lowerpath:
1081         for (i = 0; i < numlower; i++)
1082                 path_put(&stack[i]);
1083         kfree(stack);
1084 out_free_lowertmp:
1085         kfree(lowertmp);
1086 out_put_workpath:
1087         path_put(&workpath);
1088 out_put_upperpath:
1089         path_put(&upperpath);
1090 out_free_config:
1091         kfree(ufs->config.lowerdir);
1092         kfree(ufs->config.upperdir);
1093         kfree(ufs->config.workdir);
1094         kfree(ufs);
1095 out:
1096         return err;
1097 }
1098
1099 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1100                                 const char *dev_name, void *raw_data)
1101 {
1102         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1103 }
1104
1105 static struct file_system_type ovl_fs_type = {
1106         .owner          = THIS_MODULE,
1107         .name           = "overlay",
1108         .mount          = ovl_mount,
1109         .kill_sb        = kill_anon_super,
1110 };
1111 MODULE_ALIAS_FS("overlay");
1112
1113 static int __init ovl_init(void)
1114 {
1115         return register_filesystem(&ovl_fs_type);
1116 }
1117
1118 static void __exit ovl_exit(void)
1119 {
1120         unregister_filesystem(&ovl_fs_type);
1121 }
1122
1123 module_init(ovl_init);
1124 module_exit(ovl_exit);