fs/overlayfs/super.c

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