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 struct dentry *ovl_d_real(struct dentry *dentry, struct inode *inode)
 280 {
 281         struct dentry *real;
 282
 283         if (d_is_dir(dentry)) {
 284                 if (!inode || inode == d_inode(dentry))
 285                         return dentry;
 286                 goto bug;
 287         }
 288
 289         real = ovl_dentry_upper(dentry);
 290         if (real && (!inode || inode == d_inode(real)))
 291                 return real;
 292
 293         real = ovl_dentry_lower(dentry);
 294         if (!real)
 295                 goto bug;
 296
 297         if (!inode || inode == d_inode(real))
 298                 return real;
 299
 300         /* Handle recursion */
 301         if (real->d_flags & DCACHE_OP_REAL)
 302                 return real->d_op->d_real(real, inode);
 303
 304 bug:
 305         WARN(1, "ovl_d_real(%pd4, %s:%lu\n): real dentry not found\n", dentry,
 306              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 307         return dentry;
 308 }
 309
 310 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 311 {
 312         struct ovl_entry *oe = dentry->d_fsdata;
 313         unsigned int i;
 314         int ret = 1;
 315
 316         for (i = 0; i < oe->numlower; i++) {
 317                 struct dentry *d = oe->lowerstack[i].dentry;
 318
 319                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 320                         ret = d->d_op->d_revalidate(d, flags);
 321                         if (ret < 0)
 322                                 return ret;
 323                         if (!ret) {
 324                                 if (!(flags & LOOKUP_RCU))
 325                                         d_invalidate(d);
 326                                 return -ESTALE;
 327                         }
 328                 }
 329         }
 330         return 1;
 331 }
 332
 333 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 334 {
 335         struct ovl_entry *oe = dentry->d_fsdata;
 336         unsigned int i;
 337         int ret = 1;
 338
 339         for (i = 0; i < oe->numlower; i++) {
 340                 struct dentry *d = oe->lowerstack[i].dentry;
 341
 342                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 343                         ret = d->d_op->d_weak_revalidate(d, flags);
 344                         if (ret <= 0)
 345                                 break;
 346                 }
 347         }
 348         return ret;
 349 }
 350
 351 static const struct dentry_operations ovl_dentry_operations = {
 352         .d_release = ovl_dentry_release,
 353         .d_select_inode = ovl_d_select_inode,
 354         .d_real = ovl_d_real,
 355 };
 356
 357 static const struct dentry_operations ovl_reval_dentry_operations = {
 358         .d_release = ovl_dentry_release,
 359         .d_select_inode = ovl_d_select_inode,
 360         .d_real = ovl_d_real,
 361         .d_revalidate = ovl_dentry_revalidate,
 362         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 363 };
 364
 365 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 366 {
 367         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 368         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 369
 370         if (oe)
 371                 oe->numlower = numlower;
 372
 373         return oe;
 374 }
 375
 376 static bool ovl_dentry_remote(struct dentry *dentry)
 377 {
 378         return dentry->d_flags &
 379                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE |
 380                  DCACHE_OP_REAL);
 381 }
 382
 383 static bool ovl_dentry_weird(struct dentry *dentry)
 384 {
 385         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 386                                   DCACHE_MANAGE_TRANSIT |
 387                                   DCACHE_OP_HASH |
 388                                   DCACHE_OP_COMPARE);
 389 }
 390
 391 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 392                                              struct qstr *name)
 393 {
 394         struct dentry *dentry;
 395
 396         mutex_lock(&dir->d_inode->i_mutex);
 397         dentry = lookup_one_len(name->name, dir, name->len);
 398         mutex_unlock(&dir->d_inode->i_mutex);
 399
 400         if (IS_ERR(dentry)) {
 401                 if (PTR_ERR(dentry) == -ENOENT)
 402                         dentry = NULL;
 403         } else if (!dentry->d_inode) {
 404                 dput(dentry);
 405                 dentry = NULL;
 406         } else if (ovl_dentry_weird(dentry)) {
 407                 dput(dentry);
 408                 /* Don't support traversing automounts and other weirdness */
 409                 dentry = ERR_PTR(-EREMOTE);
 410         }
 411         return dentry;
 412 }
 413
 414 /*
 415  * Returns next layer in stack starting from top.
 416  * Returns -1 if this is the last layer.
 417  */
 418 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 419 {
 420         struct ovl_entry *oe = dentry->d_fsdata;
 421
 422         BUG_ON(idx < 0);
 423         if (idx == 0) {
 424                 ovl_path_upper(dentry, path);
 425                 if (path->dentry)
 426                         return oe->numlower ? 1 : -1;
 427                 idx++;
 428         }
 429         BUG_ON(idx > oe->numlower);
 430         *path = oe->lowerstack[idx - 1];
 431
 432         return (idx < oe->numlower) ? idx + 1 : -1;
 433 }
 434
 435 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 436                           unsigned int flags)
 437 {
 438         struct ovl_entry *oe;
 439         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 440         struct path *stack = NULL;
 441         struct dentry *upperdir, *upperdentry = NULL;
 442         unsigned int ctr = 0;
 443         struct inode *inode = NULL;
 444         bool upperopaque = false;
 445         struct dentry *this, *prev = NULL;
 446         unsigned int i;
 447         int err;
 448
 449         upperdir = ovl_upperdentry_dereference(poe);
 450         if (upperdir) {
 451                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 452                 err = PTR_ERR(this);
 453                 if (IS_ERR(this))
 454                         goto out;
 455
 456                 if (this) {
 457                         if (unlikely(ovl_dentry_remote(this))) {
 458                                 dput(this);
 459                                 err = -EREMOTE;
 460                                 goto out;
 461                         }
 462                         if (ovl_is_whiteout(this)) {
 463                                 dput(this);
 464                                 this = NULL;
 465                                 upperopaque = true;
 466                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 467                                 upperopaque = true;
 468                         }
 469                 }
 470                 upperdentry = prev = this;
 471         }
 472
 473         if (!upperopaque && poe->numlower) {
 474                 err = -ENOMEM;
 475                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 476                 if (!stack)
 477                         goto out_put_upper;
 478         }
 479
 480         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 481                 bool opaque = false;
 482                 struct path lowerpath = poe->lowerstack[i];
 483
 484                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 485                 err = PTR_ERR(this);
 486                 if (IS_ERR(this)) {
 487                         /*
 488                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 489                          */
 490                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 491                                 continue;
 492                         goto out_put;
 493                 }
 494                 if (!this)
 495                         continue;
 496                 if (ovl_is_whiteout(this)) {
 497                         dput(this);
 498                         break;
 499                 }
 500                 /*
 501                  * Only makes sense to check opaque dir if this is not the
 502                  * lowermost layer.
 503                  */
 504                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 505                         opaque = true;
 506
 507                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 508                              !S_ISDIR(this->d_inode->i_mode))) {
 509                         /*
 510                          * FIXME: check for upper-opaqueness maybe better done
 511                          * in remove code.
 512                          */
 513                         if (prev == upperdentry)
 514                                 upperopaque = true;
 515                         dput(this);
 516                         break;
 517                 }
 518                 /*
 519                  * If this is a non-directory then stop here.
 520                  */
 521                 if (!S_ISDIR(this->d_inode->i_mode))
 522                         opaque = true;
 523
 524                 stack[ctr].dentry = this;
 525                 stack[ctr].mnt = lowerpath.mnt;
 526                 ctr++;
 527                 prev = this;
 528                 if (opaque)
 529                         break;
 530         }
 531
 532         oe = ovl_alloc_entry(ctr);
 533         err = -ENOMEM;
 534         if (!oe)
 535                 goto out_put;
 536
 537         if (upperdentry || ctr) {
 538                 struct dentry *realdentry;
 539
 540                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 541
 542                 err = -ENOMEM;
 543                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 544                                       oe);
 545                 if (!inode)
 546                         goto out_free_oe;
 547                 ovl_copyattr(realdentry->d_inode, inode);
 548         }
 549
 550         oe->opaque = upperopaque;
 551         oe->__upperdentry = upperdentry;
 552         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 553         kfree(stack);
 554         dentry->d_fsdata = oe;
 555         d_add(dentry, inode);
 556
 557         return NULL;
 558
 559 out_free_oe:
 560         kfree(oe);
 561 out_put:
 562         for (i = 0; i < ctr; i++)
 563                 dput(stack[i].dentry);
 564         kfree(stack);
 565 out_put_upper:
 566         dput(upperdentry);
 567 out:
 568         return ERR_PTR(err);
 569 }
 570
 571 struct file *ovl_path_open(struct path *path, int flags)
 572 {
 573         return dentry_open(path, flags, current_cred());
 574 }
 575
 576 static void ovl_put_super(struct super_block *sb)
 577 {
 578         struct ovl_fs *ufs = sb->s_fs_info;
 579         unsigned i;
 580
 581         dput(ufs->workdir);
 582         mntput(ufs->upper_mnt);
 583         for (i = 0; i < ufs->numlower; i++)
 584                 mntput(ufs->lower_mnt[i]);
 585         kfree(ufs->lower_mnt);
 586
 587         kfree(ufs->config.lowerdir);
 588         kfree(ufs->config.upperdir);
 589         kfree(ufs->config.workdir);
 590         kfree(ufs);
 591 }
 592
 593 /**
 594  * ovl_statfs
 595  * @sb: The overlayfs super block
 596  * @buf: The struct kstatfs to fill in with stats
 597  *
 598  * Get the filesystem statistics.  As writes always target the upper layer
 599  * filesystem pass the statfs to the upper filesystem (if it exists)
 600  */
 601 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 602 {
 603         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 604         struct dentry *root_dentry = dentry->d_sb->s_root;
 605         struct path path;
 606         int err;
 607
 608         ovl_path_real(root_dentry, &path);
 609
 610         err = vfs_statfs(&path, buf);
 611         if (!err) {
 612                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 613                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 614         }
 615
 616         return err;
 617 }
 618
 619 /**
 620  * ovl_show_options
 621  *
 622  * Prints the mount options for a given superblock.
 623  * Returns zero; does not fail.
 624  */
 625 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 626 {
 627         struct super_block *sb = dentry->d_sb;
 628         struct ovl_fs *ufs = sb->s_fs_info;
 629
 630         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 631         if (ufs->config.upperdir) {
 632                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 633                 seq_show_option(m, "workdir", ufs->config.workdir);
 634         }
 635         return 0;
 636 }
 637
 638 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 639 {
 640         struct ovl_fs *ufs = sb->s_fs_info;
 641
 642         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 643                 return -EROFS;
 644
 645         return 0;
 646 }
 647
 648 static const struct super_operations ovl_super_operations = {
 649         .put_super      = ovl_put_super,
 650         .statfs         = ovl_statfs,
 651         .show_options   = ovl_show_options,
 652         .remount_fs     = ovl_remount,
 653 };
 654
 655 enum {
 656         OPT_LOWERDIR,
 657         OPT_UPPERDIR,
 658         OPT_WORKDIR,
 659         OPT_ERR,
 660 };
 661
 662 static const match_table_t ovl_tokens = {
 663         {OPT_LOWERDIR,                  "lowerdir=%s"},
 664         {OPT_UPPERDIR,                  "upperdir=%s"},
 665         {OPT_WORKDIR,                   "workdir=%s"},
 666         {OPT_ERR,                       NULL}
 667 };
 668
 669 static char *ovl_next_opt(char **s)
 670 {
 671         char *sbegin = *s;
 672         char *p;
 673
 674         if (sbegin == NULL)
 675                 return NULL;
 676
 677         for (p = sbegin; *p; p++) {
 678                 if (*p == '\\') {
 679                         p++;
 680                         if (!*p)
 681                                 break;
 682                 } else if (*p == ',') {
 683                         *p = '\0';
 684                         *s = p + 1;
 685                         return sbegin;
 686                 }
 687         }
 688         *s = NULL;
 689         return sbegin;
 690 }
 691
 692 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 693 {
 694         char *p;
 695
 696         while ((p = ovl_next_opt(&opt)) != NULL) {
 697                 int token;
 698                 substring_t args[MAX_OPT_ARGS];
 699
 700                 if (!*p)
 701                         continue;
 702
 703                 token = match_token(p, ovl_tokens, args);
 704                 switch (token) {
 705                 case OPT_UPPERDIR:
 706                         kfree(config->upperdir);
 707                         config->upperdir = match_strdup(&args[0]);
 708                         if (!config->upperdir)
 709                                 return -ENOMEM;
 710                         break;
 711
 712                 case OPT_LOWERDIR:
 713                         kfree(config->lowerdir);
 714                         config->lowerdir = match_strdup(&args[0]);
 715                         if (!config->lowerdir)
 716                                 return -ENOMEM;
 717                         break;
 718
 719                 case OPT_WORKDIR:
 720                         kfree(config->workdir);
 721                         config->workdir = match_strdup(&args[0]);
 722                         if (!config->workdir)
 723                                 return -ENOMEM;
 724                         break;
 725
 726                 default:
 727                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 728                         return -EINVAL;
 729                 }
 730         }
 731
 732         /* Workdir is useless in non-upper mount */
 733         if (!config->upperdir && config->workdir) {
 734                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 735                         config->workdir);
 736                 kfree(config->workdir);
 737                 config->workdir = NULL;
 738         }
 739
 740         return 0;
 741 }
 742
 743 #define OVL_WORKDIR_NAME "work"
 744
 745 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 746                                          struct dentry *dentry)
 747 {
 748         struct inode *dir = dentry->d_inode;
 749         struct dentry *work;
 750         int err;
 751         bool retried = false;
 752
 753         err = mnt_want_write(mnt);
 754         if (err)
 755                 return ERR_PTR(err);
 756
 757         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 758 retry:
 759         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 760                               strlen(OVL_WORKDIR_NAME));
 761
 762         if (!IS_ERR(work)) {
 763                 struct kstat stat = {
 764                         .mode = S_IFDIR | 0,
 765                 };
 766                 struct iattr attr = {
 767                         .ia_valid = ATTR_MODE,
 768                         .ia_mode = stat.mode,
 769                 };
 770
 771                 if (work->d_inode) {
 772                         err = -EEXIST;
 773                         if (retried)
 774                                 goto out_dput;
 775
 776                         retried = true;
 777                         ovl_cleanup(dir, work);
 778                         dput(work);
 779                         goto retry;
 780                 }
 781
 782                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 783                 if (err)
 784                         goto out_dput;
 785
 786                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 787                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 788                         goto out_dput;
 789
 790                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 791                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 792                         goto out_dput;
 793
 794                 /* Clear any inherited mode bits */
 795                 inode_lock(work->d_inode);
 796                 err = notify_change(work, &attr, NULL);
 797                 inode_unlock(work->d_inode);
 798                 if (err)
 799                         goto out_dput;
 800         }
 801 out_unlock:
 802         mutex_unlock(&dir->i_mutex);
 803         mnt_drop_write(mnt);
 804
 805         return work;
 806
 807 out_dput:
 808         dput(work);
 809         work = ERR_PTR(err);
 810         goto out_unlock;
 811 }
 812
 813 static void ovl_unescape(char *s)
 814 {
 815         char *d = s;
 816
 817         for (;; s++, d++) {
 818                 if (*s == '\\')
 819                         s++;
 820                 *d = *s;
 821                 if (!*s)
 822                         break;
 823         }
 824 }
 825
 826 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 827 {
 828         int err = -EINVAL;
 829
 830         if (!*name) {
 831                 pr_err("overlayfs: empty lowerdir\n");
 832                 goto out;
 833         }
 834         err = kern_path(name, LOOKUP_FOLLOW, path);
 835         if (err) {
 836                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 837                 goto out;
 838         }
 839         err = -EINVAL;
 840         if (ovl_dentry_weird(path->dentry)) {
 841                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 842                 goto out_put;
 843         }
 844         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 845                 pr_err("overlayfs: '%s' not a directory\n", name);
 846                 goto out_put;
 847         }
 848         return 0;
 849
 850 out_put:
 851         path_put(path);
 852 out:
 853         return err;
 854 }
 855
 856 static int ovl_mount_dir(const char *name, struct path *path)
 857 {
 858         int err = -ENOMEM;
 859         char *tmp = kstrdup(name, GFP_KERNEL);
 860
 861         if (tmp) {
 862                 ovl_unescape(tmp);
 863                 err = ovl_mount_dir_noesc(tmp, path);
 864
 865                 if (!err)
 866                         if (ovl_dentry_remote(path->dentry)) {
 867                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 868                                        tmp);
 869                                 path_put(path);
 870                                 err = -EINVAL;
 871                         }
 872                 kfree(tmp);
 873         }
 874         return err;
 875 }
 876
 877 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 878                          int *stack_depth, bool *remote)
 879 {
 880         int err;
 881         struct kstatfs statfs;
 882
 883         err = ovl_mount_dir_noesc(name, path);
 884         if (err)
 885                 goto out;
 886
 887         err = vfs_statfs(path, &statfs);
 888         if (err) {
 889                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 890                 goto out_put;
 891         }
 892         *namelen = max(*namelen, statfs.f_namelen);
 893         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 894
 895         if (ovl_dentry_remote(path->dentry))
 896                 *remote = true;
 897
 898         return 0;
 899
 900 out_put:
 901         path_put(path);
 902 out:
 903         return err;
 904 }
 905
 906 /* Workdir should not be subdir of upperdir and vice versa */
 907 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 908 {
 909         bool ok = false;
 910
 911         if (workdir != upperdir) {
 912                 ok = (lock_rename(workdir, upperdir) == NULL);
 913                 unlock_rename(workdir, upperdir);
 914         }
 915         return ok;
 916 }
 917
 918 static unsigned int ovl_split_lowerdirs(char *str)
 919 {
 920         unsigned int ctr = 1;
 921         char *s, *d;
 922
 923         for (s = d = str;; s++, d++) {
 924                 if (*s == '\\') {
 925                         s++;
 926                 } else if (*s == ':') {
 927                         *d = '\0';
 928                         ctr++;
 929                         continue;
 930                 }
 931                 *d = *s;
 932                 if (!*s)
 933                         break;
 934         }
 935         return ctr;
 936 }
 937
 938 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 939 {
 940         struct path upperpath = { NULL, NULL };
 941         struct path workpath = { NULL, NULL };
 942         struct dentry *root_dentry;
 943         struct ovl_entry *oe;
 944         struct ovl_fs *ufs;
 945         struct path *stack = NULL;
 946         char *lowertmp;
 947         char *lower;
 948         unsigned int numlower;
 949         unsigned int stacklen = 0;
 950         unsigned int i;
 951         bool remote = false;
 952         int err;
 953
 954         err = -ENOMEM;
 955         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 956         if (!ufs)
 957                 goto out;
 958
 959         err = ovl_parse_opt((char *) data, &ufs->config);
 960         if (err)
 961                 goto out_free_config;
 962
 963         err = -EINVAL;
 964         if (!ufs->config.lowerdir) {
 965                 pr_err("overlayfs: missing 'lowerdir'\n");
 966                 goto out_free_config;
 967         }
 968
 969         sb->s_stack_depth = 0;
 970         sb->s_maxbytes = MAX_LFS_FILESIZE;
 971         if (ufs->config.upperdir) {
 972                 if (!ufs->config.workdir) {
 973                         pr_err("overlayfs: missing 'workdir'\n");
 974                         goto out_free_config;
 975                 }
 976
 977                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 978                 if (err)
 979                         goto out_free_config;
 980
 981                 /* Upper fs should not be r/o */
 982                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 983                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 984                         err = -EINVAL;
 985                         goto out_put_upperpath;
 986                 }
 987
 988                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 989                 if (err)
 990                         goto out_put_upperpath;
 991
 992                 err = -EINVAL;
 993                 if (upperpath.mnt != workpath.mnt) {
 994                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 995                         goto out_put_workpath;
 996                 }
 997                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 998                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 999                         goto out_put_workpath;
1000                 }
1001                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
1002         }
1003         err = -ENOMEM;
1004         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
1005         if (!lowertmp)
1006                 goto out_put_workpath;
1007
1008         err = -EINVAL;
1009         stacklen = ovl_split_lowerdirs(lowertmp);
1010         if (stacklen > OVL_MAX_STACK) {
1011                 pr_err("overlayfs: too many lower directries, limit is %d\n",
1012                        OVL_MAX_STACK);
1013                 goto out_free_lowertmp;
1014         } else if (!ufs->config.upperdir && stacklen == 1) {
1015                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1016                 goto out_free_lowertmp;
1017         }
1018
1019         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1020         if (!stack)
1021                 goto out_free_lowertmp;
1022
1023         lower = lowertmp;
1024         for (numlower = 0; numlower < stacklen; numlower++) {
1025                 err = ovl_lower_dir(lower, &stack[numlower],
1026                                     &ufs->lower_namelen, &sb->s_stack_depth,
1027                                     &remote);
1028                 if (err)
1029                         goto out_put_lowerpath;
1030
1031                 lower = strchr(lower, '\0') + 1;
1032         }
1033
1034         err = -EINVAL;
1035         sb->s_stack_depth++;
1036         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1037                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1038                 goto out_put_lowerpath;
1039         }
1040
1041         if (ufs->config.upperdir) {
1042                 ufs->upper_mnt = clone_private_mount(&upperpath);
1043                 err = PTR_ERR(ufs->upper_mnt);
1044                 if (IS_ERR(ufs->upper_mnt)) {
1045                         pr_err("overlayfs: failed to clone upperpath\n");
1046                         goto out_put_lowerpath;
1047                 }
1048
1049                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
1050                 err = PTR_ERR(ufs->workdir);
1051                 if (IS_ERR(ufs->workdir)) {
1052                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1053                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1054                         sb->s_flags |= MS_RDONLY;
1055                         ufs->workdir = NULL;
1056                 }
1057         }
1058
1059         err = -ENOMEM;
1060         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1061         if (ufs->lower_mnt == NULL)
1062                 goto out_put_workdir;
1063         for (i = 0; i < numlower; i++) {
1064                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1065
1066                 err = PTR_ERR(mnt);
1067                 if (IS_ERR(mnt)) {
1068                         pr_err("overlayfs: failed to clone lowerpath\n");
1069                         goto out_put_lower_mnt;
1070                 }
1071                 /*
1072                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1073                  * will fail instead of modifying lower fs.
1074                  */
1075                 mnt->mnt_flags |= MNT_READONLY;
1076
1077                 ufs->lower_mnt[ufs->numlower] = mnt;
1078                 ufs->numlower++;
1079         }
1080
1081         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1082         if (!ufs->upper_mnt)
1083                 sb->s_flags |= MS_RDONLY;
1084
1085         if (remote)
1086                 sb->s_d_op = &ovl_reval_dentry_operations;
1087         else
1088                 sb->s_d_op = &ovl_dentry_operations;
1089
1090         err = -ENOMEM;
1091         oe = ovl_alloc_entry(numlower);
1092         if (!oe)
1093                 goto out_put_lower_mnt;
1094
1095         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1096         if (!root_dentry)
1097                 goto out_free_oe;
1098
1099         mntput(upperpath.mnt);
1100         for (i = 0; i < numlower; i++)
1101                 mntput(stack[i].mnt);
1102         path_put(&workpath);
1103         kfree(lowertmp);
1104
1105         oe->__upperdentry = upperpath.dentry;
1106         for (i = 0; i < numlower; i++) {
1107                 oe->lowerstack[i].dentry = stack[i].dentry;
1108                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1109         }
1110         kfree(stack);
1111
1112         root_dentry->d_fsdata = oe;
1113
1114         ovl_copyattr(ovl_dentry_real(root_dentry)->d_inode,
1115                      root_dentry->d_inode);
1116
1117         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1118         sb->s_op = &ovl_super_operations;
1119         sb->s_root = root_dentry;
1120         sb->s_fs_info = ufs;
1121
1122         return 0;
1123
1124 out_free_oe:
1125         kfree(oe);
1126 out_put_lower_mnt:
1127         for (i = 0; i < ufs->numlower; i++)
1128                 mntput(ufs->lower_mnt[i]);
1129         kfree(ufs->lower_mnt);
1130 out_put_workdir:
1131         dput(ufs->workdir);
1132         mntput(ufs->upper_mnt);
1133 out_put_lowerpath:
1134         for (i = 0; i < numlower; i++)
1135                 path_put(&stack[i]);
1136         kfree(stack);
1137 out_free_lowertmp:
1138         kfree(lowertmp);
1139 out_put_workpath:
1140         path_put(&workpath);
1141 out_put_upperpath:
1142         path_put(&upperpath);
1143 out_free_config:
1144         kfree(ufs->config.lowerdir);
1145         kfree(ufs->config.upperdir);
1146         kfree(ufs->config.workdir);
1147         kfree(ufs);
1148 out:
1149         return err;
1150 }
1151
1152 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1153                                 const char *dev_name, void *raw_data)
1154 {
1155         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1156 }
1157
1158 static struct file_system_type ovl_fs_type = {
1159         .owner          = THIS_MODULE,
1160         .name           = "overlay",
1161         .mount          = ovl_mount,
1162         .kill_sb        = kill_anon_super,
1163 };
1164 MODULE_ALIAS_FS("overlay");
1165
1166 static int __init ovl_init(void)
1167 {
1168         return register_filesystem(&ovl_fs_type);
1169 }
1170
1171 static void __exit ovl_exit(void)
1172 {
1173         unregister_filesystem(&ovl_fs_type);
1174 }
1175
1176 module_init(ovl_init);
1177 module_exit(ovl_exit);