fs/overlayfs/super.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/fs.h>
  11 #include <linux/namei.h>
  12 #include <linux/xattr.h>
  13 #include <linux/security.h>
  14 #include <linux/mount.h>
  15 #include <linux/slab.h>
  16 #include <linux/parser.h>
  17 #include <linux/module.h>
  18 #include <linux/sched.h>
  19 #include <linux/statfs.h>
  20 #include <linux/seq_file.h>
  21 #include "overlayfs.h"
  22
  23 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  24 MODULE_DESCRIPTION("Overlay filesystem");
  25 MODULE_LICENSE("GPL");
  26
  27 #define OVERLAYFS_SUPER_MAGIC 0x794c7630
  28
  29 struct ovl_config {
  30         char *lowerdir;
  31         char *upperdir;
  32         char *workdir;
  33         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                 if (oe->numlower) {
  80                         if (S_ISDIR(dentry->d_inode->i_mode))
  81                                 type |= __OVL_PATH_MERGE;
  82                 } else if (!oe->opaque) {
  83                         type |= __OVL_PATH_PURE;
  84                 }
  85         } else {
  86                 if (oe->numlower > 1)
  87                         type |= __OVL_PATH_MERGE;
  88         }
  89         return type;
  90 }
  91
  92 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
  93 {
  94         return lockless_dereference(oe->__upperdentry);
  95 }
  96
  97 void ovl_path_upper(struct dentry *dentry, struct path *path)
  98 {
  99         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 100         struct ovl_entry *oe = dentry->d_fsdata;
 101
 102         path->mnt = ofs->upper_mnt;
 103         path->dentry = ovl_upperdentry_dereference(oe);
 104 }
 105
 106 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 107 {
 108         enum ovl_path_type type = ovl_path_type(dentry);
 109
 110         if (!OVL_TYPE_UPPER(type))
 111                 ovl_path_lower(dentry, path);
 112         else
 113                 ovl_path_upper(dentry, path);
 114
 115         return type;
 116 }
 117
 118 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 119 {
 120         struct ovl_entry *oe = dentry->d_fsdata;
 121
 122         return ovl_upperdentry_dereference(oe);
 123 }
 124
 125 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 126 {
 127         struct ovl_entry *oe = dentry->d_fsdata;
 128
 129         return __ovl_dentry_lower(oe);
 130 }
 131
 132 struct dentry *ovl_dentry_real(struct dentry *dentry)
 133 {
 134         struct ovl_entry *oe = dentry->d_fsdata;
 135         struct dentry *realdentry;
 136
 137         realdentry = ovl_upperdentry_dereference(oe);
 138         if (!realdentry)
 139                 realdentry = __ovl_dentry_lower(oe);
 140
 141         return realdentry;
 142 }
 143
 144 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
 145 {
 146         struct dentry *realdentry;
 147
 148         realdentry = ovl_upperdentry_dereference(oe);
 149         if (realdentry) {
 150                 *is_upper = true;
 151         } else {
 152                 realdentry = __ovl_dentry_lower(oe);
 153                 *is_upper = false;
 154         }
 155         return realdentry;
 156 }
 157
 158 struct vfsmount *ovl_entry_mnt_real(struct ovl_entry *oe, struct inode *inode,
 159                                     bool is_upper)
 160 {
 161         if (is_upper) {
 162                 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 163
 164                 return ofs->upper_mnt;
 165         } else {
 166                 return oe->numlower ? oe->lowerstack[0].mnt : NULL;
 167         }
 168 }
 169
 170 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 171 {
 172         struct ovl_entry *oe = dentry->d_fsdata;
 173
 174         return oe->cache;
 175 }
 176
 177 bool ovl_is_default_permissions(struct inode *inode)
 178 {
 179         struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 180
 181         return ofs->config.default_permissions;
 182 }
 183
 184 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 185 {
 186         struct ovl_entry *oe = dentry->d_fsdata;
 187
 188         oe->cache = cache;
 189 }
 190
 191 void ovl_path_lower(struct dentry *dentry, struct path *path)
 192 {
 193         struct ovl_entry *oe = dentry->d_fsdata;
 194
 195         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 196 }
 197
 198 int ovl_want_write(struct dentry *dentry)
 199 {
 200         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 201         return mnt_want_write(ofs->upper_mnt);
 202 }
 203
 204 void ovl_drop_write(struct dentry *dentry)
 205 {
 206         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 207         mnt_drop_write(ofs->upper_mnt);
 208 }
 209
 210 struct dentry *ovl_workdir(struct dentry *dentry)
 211 {
 212         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 213         return ofs->workdir;
 214 }
 215
 216 bool ovl_dentry_is_opaque(struct dentry *dentry)
 217 {
 218         struct ovl_entry *oe = dentry->d_fsdata;
 219         return oe->opaque;
 220 }
 221
 222 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 223 {
 224         struct ovl_entry *oe = dentry->d_fsdata;
 225         oe->opaque = opaque;
 226 }
 227
 228 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 229 {
 230         struct ovl_entry *oe = dentry->d_fsdata;
 231
 232         WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
 233         WARN_ON(oe->__upperdentry);
 234         BUG_ON(!upperdentry->d_inode);
 235         /*
 236          * Make sure upperdentry is consistent before making it visible to
 237          * ovl_upperdentry_dereference().
 238          */
 239         smp_wmb();
 240         oe->__upperdentry = upperdentry;
 241 }
 242
 243 void ovl_dentry_version_inc(struct dentry *dentry)
 244 {
 245         struct ovl_entry *oe = dentry->d_fsdata;
 246
 247         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 248         oe->version++;
 249 }
 250
 251 u64 ovl_dentry_version_get(struct dentry *dentry)
 252 {
 253         struct ovl_entry *oe = dentry->d_fsdata;
 254
 255         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 256         return oe->version;
 257 }
 258
 259 bool ovl_is_whiteout(struct dentry *dentry)
 260 {
 261         struct inode *inode = dentry->d_inode;
 262
 263         return inode && IS_WHITEOUT(inode);
 264 }
 265
 266 static bool ovl_is_opaquedir(struct dentry *dentry)
 267 {
 268         int res;
 269         char val;
 270         struct inode *inode = dentry->d_inode;
 271
 272         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
 273                 return false;
 274
 275         res = inode->i_op->getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
 276         if (res == 1 && val == 'y')
 277                 return true;
 278
 279         return false;
 280 }
 281
 282 static void ovl_dentry_release(struct dentry *dentry)
 283 {
 284         struct ovl_entry *oe = dentry->d_fsdata;
 285
 286         if (oe) {
 287                 unsigned int i;
 288
 289                 dput(oe->__upperdentry);
 290                 for (i = 0; i < oe->numlower; i++)
 291                         dput(oe->lowerstack[i].dentry);
 292                 kfree_rcu(oe, rcu);
 293         }
 294 }
 295
 296 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 297 {
 298         struct ovl_entry *oe = dentry->d_fsdata;
 299         unsigned int i;
 300         int ret = 1;
 301
 302         for (i = 0; i < oe->numlower; i++) {
 303                 struct dentry *d = oe->lowerstack[i].dentry;
 304
 305                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 306                         ret = d->d_op->d_revalidate(d, flags);
 307                         if (ret < 0)
 308                                 return ret;
 309                         if (!ret) {
 310                                 if (!(flags & LOOKUP_RCU))
 311                                         d_invalidate(d);
 312                                 return -ESTALE;
 313                         }
 314                 }
 315         }
 316         return 1;
 317 }
 318
 319 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 320 {
 321         struct ovl_entry *oe = dentry->d_fsdata;
 322         unsigned int i;
 323         int ret = 1;
 324
 325         for (i = 0; i < oe->numlower; i++) {
 326                 struct dentry *d = oe->lowerstack[i].dentry;
 327
 328                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 329                         ret = d->d_op->d_weak_revalidate(d, flags);
 330                         if (ret <= 0)
 331                                 break;
 332                 }
 333         }
 334         return ret;
 335 }
 336
 337 static const struct dentry_operations ovl_dentry_operations = {
 338         .d_release = ovl_dentry_release,
 339         .d_select_inode = ovl_d_select_inode,
 340 };
 341
 342 static const struct dentry_operations ovl_reval_dentry_operations = {
 343         .d_release = ovl_dentry_release,
 344         .d_revalidate = ovl_dentry_revalidate,
 345         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 346 };
 347
 348 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 349 {
 350         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 351         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 352
 353         if (oe)
 354                 oe->numlower = numlower;
 355
 356         return oe;
 357 }
 358
 359 static bool ovl_dentry_remote(struct dentry *dentry)
 360 {
 361         return dentry->d_flags &
 362                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 363 }
 364
 365 static bool ovl_dentry_weird(struct dentry *dentry)
 366 {
 367         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 368                                   DCACHE_MANAGE_TRANSIT |
 369                                   DCACHE_OP_HASH |
 370                                   DCACHE_OP_COMPARE);
 371 }
 372
 373 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 374                                              struct qstr *name)
 375 {
 376         struct dentry *dentry;
 377
 378         mutex_lock(&dir->d_inode->i_mutex);
 379         dentry = lookup_one_len(name->name, dir, name->len);
 380         mutex_unlock(&dir->d_inode->i_mutex);
 381
 382         if (IS_ERR(dentry)) {
 383                 if (PTR_ERR(dentry) == -ENOENT)
 384                         dentry = NULL;
 385         } else if (!dentry->d_inode) {
 386                 dput(dentry);
 387                 dentry = NULL;
 388         } else if (ovl_dentry_weird(dentry)) {
 389                 dput(dentry);
 390                 /* Don't support traversing automounts and other weirdness */
 391                 dentry = ERR_PTR(-EREMOTE);
 392         }
 393         return dentry;
 394 }
 395
 396 /*
 397  * Returns next layer in stack starting from top.
 398  * Returns -1 if this is the last layer.
 399  */
 400 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 401 {
 402         struct ovl_entry *oe = dentry->d_fsdata;
 403
 404         BUG_ON(idx < 0);
 405         if (idx == 0) {
 406                 ovl_path_upper(dentry, path);
 407                 if (path->dentry)
 408                         return oe->numlower ? 1 : -1;
 409                 idx++;
 410         }
 411         BUG_ON(idx > oe->numlower);
 412         *path = oe->lowerstack[idx - 1];
 413
 414         return (idx < oe->numlower) ? idx + 1 : -1;
 415 }
 416
 417 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 418                           unsigned int flags)
 419 {
 420         struct ovl_entry *oe;
 421         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 422         struct path *stack = NULL;
 423         struct dentry *upperdir, *upperdentry = NULL;
 424         unsigned int ctr = 0;
 425         struct inode *inode = NULL;
 426         bool upperopaque = false;
 427         struct dentry *this, *prev = NULL;
 428         unsigned int i;
 429         int err;
 430
 431         upperdir = ovl_upperdentry_dereference(poe);
 432         if (upperdir) {
 433                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 434                 err = PTR_ERR(this);
 435                 if (IS_ERR(this))
 436                         goto out;
 437
 438                 if (this) {
 439                         if (unlikely(ovl_dentry_remote(this))) {
 440                                 dput(this);
 441                                 err = -EREMOTE;
 442                                 goto out;
 443                         }
 444                         if (ovl_is_whiteout(this)) {
 445                                 dput(this);
 446                                 this = NULL;
 447                                 upperopaque = true;
 448                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 449                                 upperopaque = true;
 450                         }
 451                 }
 452                 upperdentry = prev = this;
 453         }
 454
 455         if (!upperopaque && poe->numlower) {
 456                 err = -ENOMEM;
 457                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 458                 if (!stack)
 459                         goto out_put_upper;
 460         }
 461
 462         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 463                 bool opaque = false;
 464                 struct path lowerpath = poe->lowerstack[i];
 465
 466                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 467                 err = PTR_ERR(this);
 468                 if (IS_ERR(this)) {
 469                         /*
 470                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 471                          */
 472                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 473                                 continue;
 474                         goto out_put;
 475                 }
 476                 if (!this)
 477                         continue;
 478                 if (ovl_is_whiteout(this)) {
 479                         dput(this);
 480                         break;
 481                 }
 482                 /*
 483                  * Only makes sense to check opaque dir if this is not the
 484                  * lowermost layer.
 485                  */
 486                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 487                         opaque = true;
 488
 489                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 490                              !S_ISDIR(this->d_inode->i_mode))) {
 491                         /*
 492                          * FIXME: check for upper-opaqueness maybe better done
 493                          * in remove code.
 494                          */
 495                         if (prev == upperdentry)
 496                                 upperopaque = true;
 497                         dput(this);
 498                         break;
 499                 }
 500                 /*
 501                  * If this is a non-directory then stop here.
 502                  */
 503                 if (!S_ISDIR(this->d_inode->i_mode))
 504                         opaque = true;
 505
 506                 stack[ctr].dentry = this;
 507                 stack[ctr].mnt = lowerpath.mnt;
 508                 ctr++;
 509                 prev = this;
 510                 if (opaque)
 511                         break;
 512         }
 513
 514         oe = ovl_alloc_entry(ctr);
 515         err = -ENOMEM;
 516         if (!oe)
 517                 goto out_put;
 518
 519         if (upperdentry || ctr) {
 520                 struct dentry *realdentry;
 521
 522                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 523
 524                 err = -ENOMEM;
 525                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 526                                       oe);
 527                 if (!inode)
 528                         goto out_free_oe;
 529                 ovl_copyattr(realdentry->d_inode, inode);
 530         }
 531
 532         oe->opaque = upperopaque;
 533         oe->__upperdentry = upperdentry;
 534         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 535         kfree(stack);
 536         dentry->d_fsdata = oe;
 537         d_add(dentry, inode);
 538
 539         return NULL;
 540
 541 out_free_oe:
 542         kfree(oe);
 543 out_put:
 544         for (i = 0; i < ctr; i++)
 545                 dput(stack[i].dentry);
 546         kfree(stack);
 547 out_put_upper:
 548         dput(upperdentry);
 549 out:
 550         return ERR_PTR(err);
 551 }
 552
 553 struct file *ovl_path_open(struct path *path, int flags)
 554 {
 555         return dentry_open(path, flags, current_cred());
 556 }
 557
 558 static void ovl_put_super(struct super_block *sb)
 559 {
 560         struct ovl_fs *ufs = sb->s_fs_info;
 561         unsigned i;
 562
 563         dput(ufs->workdir);
 564         mntput(ufs->upper_mnt);
 565         for (i = 0; i < ufs->numlower; i++)
 566                 mntput(ufs->lower_mnt[i]);
 567         kfree(ufs->lower_mnt);
 568
 569         kfree(ufs->config.lowerdir);
 570         kfree(ufs->config.upperdir);
 571         kfree(ufs->config.workdir);
 572         kfree(ufs);
 573 }
 574
 575 /**
 576  * ovl_statfs
 577  * @sb: The overlayfs super block
 578  * @buf: The struct kstatfs to fill in with stats
 579  *
 580  * Get the filesystem statistics.  As writes always target the upper layer
 581  * filesystem pass the statfs to the upper filesystem (if it exists)
 582  */
 583 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 584 {
 585         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 586         struct dentry *root_dentry = dentry->d_sb->s_root;
 587         struct path path;
 588         int err;
 589
 590         ovl_path_real(root_dentry, &path);
 591
 592         err = vfs_statfs(&path, buf);
 593         if (!err) {
 594                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 595                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 596         }
 597
 598         return err;
 599 }
 600
 601 /**
 602  * ovl_show_options
 603  *
 604  * Prints the mount options for a given superblock.
 605  * Returns zero; does not fail.
 606  */
 607 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 608 {
 609         struct super_block *sb = dentry->d_sb;
 610         struct ovl_fs *ufs = sb->s_fs_info;
 611
 612         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 613         if (ufs->config.upperdir) {
 614                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 615                 seq_show_option(m, "workdir", ufs->config.workdir);
 616         }
 617         if (ufs->config.default_permissions)
 618                 seq_puts(m, ",default_permissions");
 619         return 0;
 620 }
 621
 622 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 623 {
 624         struct ovl_fs *ufs = sb->s_fs_info;
 625
 626         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 627                 return -EROFS;
 628
 629         return 0;
 630 }
 631
 632 static const struct super_operations ovl_super_operations = {
 633         .put_super      = ovl_put_super,
 634         .statfs         = ovl_statfs,
 635         .show_options   = ovl_show_options,
 636         .remount_fs     = ovl_remount,
 637 };
 638
 639 enum {
 640         OPT_LOWERDIR,
 641         OPT_UPPERDIR,
 642         OPT_WORKDIR,
 643         OPT_DEFAULT_PERMISSIONS,
 644         OPT_ERR,
 645 };
 646
 647 static const match_table_t ovl_tokens = {
 648         {OPT_LOWERDIR,                  "lowerdir=%s"},
 649         {OPT_UPPERDIR,                  "upperdir=%s"},
 650         {OPT_WORKDIR,                   "workdir=%s"},
 651         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 652         {OPT_ERR,                       NULL}
 653 };
 654
 655 static char *ovl_next_opt(char **s)
 656 {
 657         char *sbegin = *s;
 658         char *p;
 659
 660         if (sbegin == NULL)
 661                 return NULL;
 662
 663         for (p = sbegin; *p; p++) {
 664                 if (*p == '\\') {
 665                         p++;
 666                         if (!*p)
 667                                 break;
 668                 } else if (*p == ',') {
 669                         *p = '\0';
 670                         *s = p + 1;
 671                         return sbegin;
 672                 }
 673         }
 674         *s = NULL;
 675         return sbegin;
 676 }
 677
 678 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 679 {
 680         char *p;
 681
 682         while ((p = ovl_next_opt(&opt)) != NULL) {
 683                 int token;
 684                 substring_t args[MAX_OPT_ARGS];
 685
 686                 if (!*p)
 687                         continue;
 688
 689                 token = match_token(p, ovl_tokens, args);
 690                 switch (token) {
 691                 case OPT_UPPERDIR:
 692                         kfree(config->upperdir);
 693                         config->upperdir = match_strdup(&args[0]);
 694                         if (!config->upperdir)
 695                                 return -ENOMEM;
 696                         break;
 697
 698                 case OPT_LOWERDIR:
 699                         kfree(config->lowerdir);
 700                         config->lowerdir = match_strdup(&args[0]);
 701                         if (!config->lowerdir)
 702                                 return -ENOMEM;
 703                         break;
 704
 705                 case OPT_WORKDIR:
 706                         kfree(config->workdir);
 707                         config->workdir = match_strdup(&args[0]);
 708                         if (!config->workdir)
 709                                 return -ENOMEM;
 710                         break;
 711
 712                 case OPT_DEFAULT_PERMISSIONS:
 713                         config->default_permissions = true;
 714                         break;
 715
 716                 default:
 717                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 718                         return -EINVAL;
 719                 }
 720         }
 721
 722         /* Workdir is useless in non-upper mount */
 723         if (!config->upperdir && config->workdir) {
 724                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 725                         config->workdir);
 726                 kfree(config->workdir);
 727                 config->workdir = NULL;
 728         }
 729
 730         return 0;
 731 }
 732
 733 #define OVL_WORKDIR_NAME "work"
 734
 735 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 736                                          struct dentry *dentry)
 737 {
 738         struct inode *dir = dentry->d_inode;
 739         struct dentry *work;
 740         int err;
 741         bool retried = false;
 742
 743         err = mnt_want_write(mnt);
 744         if (err)
 745                 return ERR_PTR(err);
 746
 747         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 748 retry:
 749         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 750                               strlen(OVL_WORKDIR_NAME));
 751
 752         if (!IS_ERR(work)) {
 753                 struct kstat stat = {
 754                         .mode = S_IFDIR | 0,
 755                 };
 756
 757                 if (work->d_inode) {
 758                         err = -EEXIST;
 759                         if (retried)
 760                                 goto out_dput;
 761
 762                         retried = true;
 763                         ovl_cleanup(dir, work);
 764                         dput(work);
 765                         goto retry;
 766                 }
 767
 768                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 769                 if (err)
 770                         goto out_dput;
 771         }
 772 out_unlock:
 773         mutex_unlock(&dir->i_mutex);
 774         mnt_drop_write(mnt);
 775
 776         return work;
 777
 778 out_dput:
 779         dput(work);
 780         work = ERR_PTR(err);
 781         goto out_unlock;
 782 }
 783
 784 static void ovl_unescape(char *s)
 785 {
 786         char *d = s;
 787
 788         for (;; s++, d++) {
 789                 if (*s == '\\')
 790                         s++;
 791                 *d = *s;
 792                 if (!*s)
 793                         break;
 794         }
 795 }
 796
 797 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 798 {
 799         int err = -EINVAL;
 800
 801         if (!*name) {
 802                 pr_err("overlayfs: empty lowerdir\n");
 803                 goto out;
 804         }
 805         err = kern_path(name, LOOKUP_FOLLOW, path);
 806         if (err) {
 807                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 808                 goto out;
 809         }
 810         err = -EINVAL;
 811         if (ovl_dentry_weird(path->dentry)) {
 812                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 813                 goto out_put;
 814         }
 815         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 816                 pr_err("overlayfs: '%s' not a directory\n", name);
 817                 goto out_put;
 818         }
 819         return 0;
 820
 821 out_put:
 822         path_put(path);
 823 out:
 824         return err;
 825 }
 826
 827 static int ovl_mount_dir(const char *name, struct path *path)
 828 {
 829         int err = -ENOMEM;
 830         char *tmp = kstrdup(name, GFP_KERNEL);
 831
 832         if (tmp) {
 833                 ovl_unescape(tmp);
 834                 err = ovl_mount_dir_noesc(tmp, path);
 835
 836                 if (!err)
 837                         if (ovl_dentry_remote(path->dentry)) {
 838                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 839                                        tmp);
 840                                 path_put(path);
 841                                 err = -EINVAL;
 842                         }
 843                 kfree(tmp);
 844         }
 845         return err;
 846 }
 847
 848 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 849                          int *stack_depth, bool *remote)
 850 {
 851         int err;
 852         struct kstatfs statfs;
 853
 854         err = ovl_mount_dir_noesc(name, path);
 855         if (err)
 856                 goto out;
 857
 858         err = vfs_statfs(path, &statfs);
 859         if (err) {
 860                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 861                 goto out_put;
 862         }
 863         *namelen = max(*namelen, statfs.f_namelen);
 864         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 865
 866         if (ovl_dentry_remote(path->dentry))
 867                 *remote = true;
 868
 869         return 0;
 870
 871 out_put:
 872         path_put(path);
 873 out:
 874         return err;
 875 }
 876
 877 /* Workdir should not be subdir of upperdir and vice versa */
 878 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 879 {
 880         bool ok = false;
 881
 882         if (workdir != upperdir) {
 883                 ok = (lock_rename(workdir, upperdir) == NULL);
 884                 unlock_rename(workdir, upperdir);
 885         }
 886         return ok;
 887 }
 888
 889 static unsigned int ovl_split_lowerdirs(char *str)
 890 {
 891         unsigned int ctr = 1;
 892         char *s, *d;
 893
 894         for (s = d = str;; s++, d++) {
 895                 if (*s == '\\') {
 896                         s++;
 897                 } else if (*s == ':') {
 898                         *d = '\0';
 899                         ctr++;
 900                         continue;
 901                 }
 902                 *d = *s;
 903                 if (!*s)
 904                         break;
 905         }
 906         return ctr;
 907 }
 908
 909 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 910 {
 911         struct path upperpath = { NULL, NULL };
 912         struct path workpath = { NULL, NULL };
 913         struct dentry *root_dentry;
 914         struct ovl_entry *oe;
 915         struct ovl_fs *ufs;
 916         struct path *stack = NULL;
 917         char *lowertmp;
 918         char *lower;
 919         unsigned int numlower;
 920         unsigned int stacklen = 0;
 921         unsigned int i;
 922         bool remote = false;
 923         int err;
 924
 925         err = -ENOMEM;
 926         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 927         if (!ufs)
 928                 goto out;
 929
 930         err = ovl_parse_opt((char *) data, &ufs->config);
 931         if (err)
 932                 goto out_free_config;
 933
 934         err = -EINVAL;
 935         if (!ufs->config.lowerdir) {
 936                 pr_err("overlayfs: missing 'lowerdir'\n");
 937                 goto out_free_config;
 938         }
 939
 940         sb->s_stack_depth = 0;
 941         if (ufs->config.upperdir) {
 942                 if (!ufs->config.workdir) {
 943                         pr_err("overlayfs: missing 'workdir'\n");
 944                         goto out_free_config;
 945                 }
 946
 947                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 948                 if (err)
 949                         goto out_free_config;
 950
 951                 /* Upper fs should not be r/o */
 952                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 953                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 954                         err = -EINVAL;
 955                         goto out_put_upperpath;
 956                 }
 957
 958                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 959                 if (err)
 960                         goto out_put_upperpath;
 961
 962                 err = -EINVAL;
 963                 if (upperpath.mnt != workpath.mnt) {
 964                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 965                         goto out_put_workpath;
 966                 }
 967                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 968                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 969                         goto out_put_workpath;
 970                 }
 971                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 972         }
 973         err = -ENOMEM;
 974         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 975         if (!lowertmp)
 976                 goto out_put_workpath;
 977
 978         err = -EINVAL;
 979         stacklen = ovl_split_lowerdirs(lowertmp);
 980         if (stacklen > OVL_MAX_STACK) {
 981                 pr_err("overlayfs: too many lower directries, limit is %d\n",
 982                        OVL_MAX_STACK);
 983                 goto out_free_lowertmp;
 984         } else if (!ufs->config.upperdir && stacklen == 1) {
 985                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 986                 goto out_free_lowertmp;
 987         }
 988
 989         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 990         if (!stack)
 991                 goto out_free_lowertmp;
 992
 993         lower = lowertmp;
 994         for (numlower = 0; numlower < stacklen; numlower++) {
 995                 err = ovl_lower_dir(lower, &stack[numlower],
 996                                     &ufs->lower_namelen, &sb->s_stack_depth,
 997                                     &remote);
 998                 if (err)
 999                         goto out_put_lowerpath;
1000
1001                 lower = strchr(lower, '\0') + 1;
1002         }
1003
1004         err = -EINVAL;
1005         sb->s_stack_depth++;
1006         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1007                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1008                 goto out_put_lowerpath;
1009         }
1010
1011         if (ufs->config.upperdir) {
1012                 ufs->upper_mnt = clone_private_mount(&upperpath);
1013                 err = PTR_ERR(ufs->upper_mnt);
1014                 if (IS_ERR(ufs->upper_mnt)) {
1015                         pr_err("overlayfs: failed to clone upperpath\n");
1016                         goto out_put_lowerpath;
1017                 }
1018
1019                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
1020                 err = PTR_ERR(ufs->workdir);
1021                 if (IS_ERR(ufs->workdir)) {
1022                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1023                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1024                         sb->s_flags |= MS_RDONLY;
1025                         ufs->workdir = NULL;
1026                 }
1027         }
1028
1029         err = -ENOMEM;
1030         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1031         if (ufs->lower_mnt == NULL)
1032                 goto out_put_workdir;
1033         for (i = 0; i < numlower; i++) {
1034                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1035
1036                 err = PTR_ERR(mnt);
1037                 if (IS_ERR(mnt)) {
1038                         pr_err("overlayfs: failed to clone lowerpath\n");
1039                         goto out_put_lower_mnt;
1040                 }
1041                 /*
1042                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1043                  * will fail instead of modifying lower fs.
1044                  */
1045                 mnt->mnt_flags |= MNT_READONLY;
1046
1047                 ufs->lower_mnt[ufs->numlower] = mnt;
1048                 ufs->numlower++;
1049         }
1050
1051         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1052         if (!ufs->upper_mnt)
1053                 sb->s_flags |= MS_RDONLY;
1054
1055         if (remote)
1056                 sb->s_d_op = &ovl_reval_dentry_operations;
1057         else
1058                 sb->s_d_op = &ovl_dentry_operations;
1059
1060         err = -ENOMEM;
1061         oe = ovl_alloc_entry(numlower);
1062         if (!oe)
1063                 goto out_put_lower_mnt;
1064
1065         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1066         if (!root_dentry)
1067                 goto out_free_oe;
1068
1069         mntput(upperpath.mnt);
1070         for (i = 0; i < numlower; i++)
1071                 mntput(stack[i].mnt);
1072         path_put(&workpath);
1073         kfree(lowertmp);
1074
1075         oe->__upperdentry = upperpath.dentry;
1076         for (i = 0; i < numlower; i++) {
1077                 oe->lowerstack[i].dentry = stack[i].dentry;
1078                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1079         }
1080         kfree(stack);
1081
1082         root_dentry->d_fsdata = oe;
1083
1084         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1085         sb->s_op = &ovl_super_operations;
1086         sb->s_root = root_dentry;
1087         sb->s_fs_info = ufs;
1088
1089         return 0;
1090
1091 out_free_oe:
1092         kfree(oe);
1093 out_put_lower_mnt:
1094         for (i = 0; i < ufs->numlower; i++)
1095                 mntput(ufs->lower_mnt[i]);
1096         kfree(ufs->lower_mnt);
1097 out_put_workdir:
1098         dput(ufs->workdir);
1099         mntput(ufs->upper_mnt);
1100 out_put_lowerpath:
1101         for (i = 0; i < numlower; i++)
1102                 path_put(&stack[i]);
1103         kfree(stack);
1104 out_free_lowertmp:
1105         kfree(lowertmp);
1106 out_put_workpath:
1107         path_put(&workpath);
1108 out_put_upperpath:
1109         path_put(&upperpath);
1110 out_free_config:
1111         kfree(ufs->config.lowerdir);
1112         kfree(ufs->config.upperdir);
1113         kfree(ufs->config.workdir);
1114         kfree(ufs);
1115 out:
1116         return err;
1117 }
1118
1119 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1120                                 const char *dev_name, void *raw_data)
1121 {
1122         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1123 }
1124
1125 static struct file_system_type ovl_fs_type = {
1126         .owner          = THIS_MODULE,
1127         .name           = "overlay",
1128         .mount          = ovl_mount,
1129         .kill_sb        = kill_anon_super,
1130 };
1131 MODULE_ALIAS_FS("overlay");
1132
1133 static int __init ovl_init(void)
1134 {
1135         return register_filesystem(&ovl_fs_type);
1136 }
1137
1138 static void __exit ovl_exit(void)
1139 {
1140         unregister_filesystem(&ovl_fs_type);
1141 }
1142
1143 module_init(ovl_init);
1144 module_exit(ovl_exit);