fs/overlayfs/super.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/fs.h>
  11 #include <linux/namei.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/xattr.h>
  14 #include <linux/security.h>
  15 #include <linux/mount.h>
  16 #include <linux/slab.h>
  17 #include <linux/parser.h>
  18 #include <linux/module.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/sched.h>
  21 #include <linux/statfs.h>
  22 #include <linux/seq_file.h>
  23 #include "overlayfs.h"
  24
  25 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  26 MODULE_DESCRIPTION("Overlay filesystem");
  27 MODULE_LICENSE("GPL");
  28
  29 struct ovl_config {
  30         char *lowerdir;
  31         char *upperdir;
  32         char *workdir;
  33         bool default_permissions;
  34 };
  35
  36 /* private information held for overlayfs's superblock */
  37 struct ovl_fs {
  38         struct vfsmount *upper_mnt;
  39         unsigned numlower;
  40         struct vfsmount **lower_mnt;
  41         struct dentry *workdir;
  42         long lower_namelen;
  43         /* pathnames of lower and upper dirs, for show_options */
  44         struct ovl_config config;
  45         /* creds of process who forced instantiation of super block */
  46         const struct cred *creator_cred;
  47 };
  48
  49 struct ovl_dir_cache;
  50
  51 /* private information held for every overlayfs dentry */
  52 struct ovl_entry {
  53         struct dentry *__upperdentry;
  54         struct ovl_dir_cache *cache;
  55         union {
  56                 struct {
  57                         u64 version;
  58                         bool opaque;
  59                 };
  60                 struct rcu_head rcu;
  61         };
  62         unsigned numlower;
  63         struct path lowerstack[];
  64 };
  65
  66 #define OVL_MAX_STACK 500
  67
  68 static struct dentry *__ovl_dentry_lower(struct ovl_entry *oe)
  69 {
  70         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
  71 }
  72
  73 enum ovl_path_type ovl_path_type(struct dentry *dentry)
  74 {
  75         struct ovl_entry *oe = dentry->d_fsdata;
  76         enum ovl_path_type type = 0;
  77
  78         if (oe->__upperdentry) {
  79                 type = __OVL_PATH_UPPER;
  80
  81                 /*
  82                  * Non-dir dentry can hold lower dentry from previous
  83                  * location. Its purity depends only on opaque flag.
  84                  */
  85                 if (oe->numlower && S_ISDIR(dentry->d_inode->i_mode))
  86                         type |= __OVL_PATH_MERGE;
  87                 else if (!oe->opaque)
  88                         type |= __OVL_PATH_PURE;
  89         } else {
  90                 if (oe->numlower > 1)
  91                         type |= __OVL_PATH_MERGE;
  92         }
  93         return type;
  94 }
  95
  96 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
  97 {
  98         return lockless_dereference(oe->__upperdentry);
  99 }
 100
 101 void ovl_path_upper(struct dentry *dentry, struct path *path)
 102 {
 103         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 104         struct ovl_entry *oe = dentry->d_fsdata;
 105
 106         path->mnt = ofs->upper_mnt;
 107         path->dentry = ovl_upperdentry_dereference(oe);
 108 }
 109
 110 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 111 {
 112         enum ovl_path_type type = ovl_path_type(dentry);
 113
 114         if (!OVL_TYPE_UPPER(type))
 115                 ovl_path_lower(dentry, path);
 116         else
 117                 ovl_path_upper(dentry, path);
 118
 119         return type;
 120 }
 121
 122 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 123 {
 124         struct ovl_entry *oe = dentry->d_fsdata;
 125
 126         return ovl_upperdentry_dereference(oe);
 127 }
 128
 129 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 130 {
 131         struct ovl_entry *oe = dentry->d_fsdata;
 132
 133         return __ovl_dentry_lower(oe);
 134 }
 135
 136 struct dentry *ovl_dentry_real(struct dentry *dentry)
 137 {
 138         struct ovl_entry *oe = dentry->d_fsdata;
 139         struct dentry *realdentry;
 140
 141         realdentry = ovl_upperdentry_dereference(oe);
 142         if (!realdentry)
 143                 realdentry = __ovl_dentry_lower(oe);
 144
 145         return realdentry;
 146 }
 147
 148 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
 149 {
 150         struct dentry *realdentry;
 151
 152         realdentry = ovl_upperdentry_dereference(oe);
 153         if (realdentry) {
 154                 *is_upper = true;
 155         } else {
 156                 realdentry = __ovl_dentry_lower(oe);
 157                 *is_upper = false;
 158         }
 159         return realdentry;
 160 }
 161
 162 struct vfsmount *ovl_entry_mnt_real(struct ovl_entry *oe, struct inode *inode,
 163                                     bool is_upper)
 164 {
 165         if (is_upper) {
 166                 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 167
 168                 return ofs->upper_mnt;
 169         } else {
 170                 return oe->numlower ? oe->lowerstack[0].mnt : NULL;
 171         }
 172 }
 173
 174 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 175 {
 176         struct ovl_entry *oe = dentry->d_fsdata;
 177
 178         return oe->cache;
 179 }
 180
 181 bool ovl_is_default_permissions(struct inode *inode)
 182 {
 183         struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 184
 185         return ofs->config.default_permissions;
 186 }
 187
 188 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 189 {
 190         struct ovl_entry *oe = dentry->d_fsdata;
 191
 192         oe->cache = cache;
 193 }
 194
 195 void ovl_path_lower(struct dentry *dentry, struct path *path)
 196 {
 197         struct ovl_entry *oe = dentry->d_fsdata;
 198
 199         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 200 }
 201
 202 int ovl_want_write(struct dentry *dentry)
 203 {
 204         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 205         return mnt_want_write(ofs->upper_mnt);
 206 }
 207
 208 void ovl_drop_write(struct dentry *dentry)
 209 {
 210         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 211         mnt_drop_write(ofs->upper_mnt);
 212 }
 213
 214 struct dentry *ovl_workdir(struct dentry *dentry)
 215 {
 216         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 217         return ofs->workdir;
 218 }
 219
 220 bool ovl_dentry_is_opaque(struct dentry *dentry)
 221 {
 222         struct ovl_entry *oe = dentry->d_fsdata;
 223         return oe->opaque;
 224 }
 225
 226 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 227 {
 228         struct ovl_entry *oe = dentry->d_fsdata;
 229         oe->opaque = opaque;
 230 }
 231
 232 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 233 {
 234         struct ovl_entry *oe = dentry->d_fsdata;
 235
 236         WARN_ON(!inode_is_locked(upperdentry->d_parent->d_inode));
 237         WARN_ON(oe->__upperdentry);
 238         BUG_ON(!upperdentry->d_inode);
 239         /*
 240          * Make sure upperdentry is consistent before making it visible to
 241          * ovl_upperdentry_dereference().
 242          */
 243         smp_wmb();
 244         oe->__upperdentry = upperdentry;
 245 }
 246
 247 void ovl_dentry_version_inc(struct dentry *dentry)
 248 {
 249         struct ovl_entry *oe = dentry->d_fsdata;
 250
 251         WARN_ON(!inode_is_locked(dentry->d_inode));
 252         oe->version++;
 253 }
 254
 255 u64 ovl_dentry_version_get(struct dentry *dentry)
 256 {
 257         struct ovl_entry *oe = dentry->d_fsdata;
 258
 259         WARN_ON(!inode_is_locked(dentry->d_inode));
 260         return oe->version;
 261 }
 262
 263 bool ovl_is_whiteout(struct dentry *dentry)
 264 {
 265         struct inode *inode = dentry->d_inode;
 266
 267         return inode && IS_WHITEOUT(inode);
 268 }
 269
 270 const struct cred *ovl_override_creds(struct super_block *sb)
 271 {
 272         struct ovl_fs *ofs = sb->s_fs_info;
 273
 274         return override_creds(ofs->creator_cred);
 275 }
 276
 277 static bool ovl_is_opaquedir(struct dentry *dentry)
 278 {
 279         int res;
 280         char val;
 281         struct inode *inode = dentry->d_inode;
 282
 283         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
 284                 return false;
 285
 286         res = inode->i_op->getxattr(dentry, inode, OVL_XATTR_OPAQUE, &val, 1);
 287         if (res == 1 && val == 'y')
 288                 return true;
 289
 290         return false;
 291 }
 292
 293 static void ovl_dentry_release(struct dentry *dentry)
 294 {
 295         struct ovl_entry *oe = dentry->d_fsdata;
 296
 297         if (oe) {
 298                 unsigned int i;
 299
 300                 dput(oe->__upperdentry);
 301                 for (i = 0; i < oe->numlower; i++)
 302                         dput(oe->lowerstack[i].dentry);
 303                 kfree_rcu(oe, rcu);
 304         }
 305 }
 306
 307 static struct dentry *ovl_d_real(struct dentry *dentry, struct inode *inode)
 308 {
 309         struct dentry *real;
 310
 311         if (d_is_dir(dentry)) {
 312                 if (!inode || inode == d_inode(dentry))
 313                         return dentry;
 314                 goto bug;
 315         }
 316
 317         real = ovl_dentry_upper(dentry);
 318         if (real && (!inode || inode == d_inode(real)))
 319                 return real;
 320
 321         real = ovl_dentry_lower(dentry);
 322         if (!real)
 323                 goto bug;
 324
 325         if (!inode || inode == d_inode(real))
 326                 return real;
 327
 328         /* Handle recursion */
 329         if (real->d_flags & DCACHE_OP_REAL)
 330                 return real->d_op->d_real(real, inode);
 331
 332 bug:
 333         WARN(1, "ovl_d_real(%pd4, %s:%lu\n): real dentry not found\n", dentry,
 334              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 335         return dentry;
 336 }
 337
 338 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 339 {
 340         struct ovl_entry *oe = dentry->d_fsdata;
 341         unsigned int i;
 342         int ret = 1;
 343
 344         for (i = 0; i < oe->numlower; i++) {
 345                 struct dentry *d = oe->lowerstack[i].dentry;
 346
 347                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 348                         ret = d->d_op->d_revalidate(d, flags);
 349                         if (ret < 0)
 350                                 return ret;
 351                         if (!ret) {
 352                                 if (!(flags & LOOKUP_RCU))
 353                                         d_invalidate(d);
 354                                 return -ESTALE;
 355                         }
 356                 }
 357         }
 358         return 1;
 359 }
 360
 361 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 362 {
 363         struct ovl_entry *oe = dentry->d_fsdata;
 364         unsigned int i;
 365         int ret = 1;
 366
 367         for (i = 0; i < oe->numlower; i++) {
 368                 struct dentry *d = oe->lowerstack[i].dentry;
 369
 370                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 371                         ret = d->d_op->d_weak_revalidate(d, flags);
 372                         if (ret <= 0)
 373                                 break;
 374                 }
 375         }
 376         return ret;
 377 }
 378
 379 static const struct dentry_operations ovl_dentry_operations = {
 380         .d_release = ovl_dentry_release,
 381         .d_select_inode = ovl_d_select_inode,
 382         .d_real = ovl_d_real,
 383 };
 384
 385 static const struct dentry_operations ovl_reval_dentry_operations = {
 386         .d_release = ovl_dentry_release,
 387         .d_select_inode = ovl_d_select_inode,
 388         .d_real = ovl_d_real,
 389         .d_revalidate = ovl_dentry_revalidate,
 390         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 391 };
 392
 393 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 394 {
 395         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 396         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 397
 398         if (oe)
 399                 oe->numlower = numlower;
 400
 401         return oe;
 402 }
 403
 404 static bool ovl_dentry_remote(struct dentry *dentry)
 405 {
 406         return dentry->d_flags &
 407                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 408 }
 409
 410 static bool ovl_dentry_weird(struct dentry *dentry)
 411 {
 412         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 413                                   DCACHE_MANAGE_TRANSIT |
 414                                   DCACHE_OP_HASH |
 415                                   DCACHE_OP_COMPARE);
 416 }
 417
 418 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 419                                              struct qstr *name)
 420 {
 421         struct dentry *dentry;
 422
 423         dentry = lookup_hash(name, dir);
 424
 425         if (IS_ERR(dentry)) {
 426                 if (PTR_ERR(dentry) == -ENOENT)
 427                         dentry = NULL;
 428         } else if (!dentry->d_inode) {
 429                 dput(dentry);
 430                 dentry = NULL;
 431         } else if (ovl_dentry_weird(dentry)) {
 432                 dput(dentry);
 433                 /* Don't support traversing automounts and other weirdness */
 434                 dentry = ERR_PTR(-EREMOTE);
 435         }
 436         return dentry;
 437 }
 438
 439 /*
 440  * Returns next layer in stack starting from top.
 441  * Returns -1 if this is the last layer.
 442  */
 443 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 444 {
 445         struct ovl_entry *oe = dentry->d_fsdata;
 446
 447         BUG_ON(idx < 0);
 448         if (idx == 0) {
 449                 ovl_path_upper(dentry, path);
 450                 if (path->dentry)
 451                         return oe->numlower ? 1 : -1;
 452                 idx++;
 453         }
 454         BUG_ON(idx > oe->numlower);
 455         *path = oe->lowerstack[idx - 1];
 456
 457         return (idx < oe->numlower) ? idx + 1 : -1;
 458 }
 459
 460 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 461                           unsigned int flags)
 462 {
 463         struct ovl_entry *oe;
 464         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 465         struct path *stack = NULL;
 466         struct dentry *upperdir, *upperdentry = NULL;
 467         unsigned int ctr = 0;
 468         struct inode *inode = NULL;
 469         bool upperopaque = false;
 470         struct dentry *this, *prev = NULL;
 471         unsigned int i;
 472         int err;
 473
 474         upperdir = ovl_upperdentry_dereference(poe);
 475         if (upperdir) {
 476                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 477                 err = PTR_ERR(this);
 478                 if (IS_ERR(this))
 479                         goto out;
 480
 481                 if (this) {
 482                         if (unlikely(ovl_dentry_remote(this))) {
 483                                 dput(this);
 484                                 err = -EREMOTE;
 485                                 goto out;
 486                         }
 487                         if (ovl_is_whiteout(this)) {
 488                                 dput(this);
 489                                 this = NULL;
 490                                 upperopaque = true;
 491                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 492                                 upperopaque = true;
 493                         }
 494                 }
 495                 upperdentry = prev = this;
 496         }
 497
 498         if (!upperopaque && poe->numlower) {
 499                 err = -ENOMEM;
 500                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 501                 if (!stack)
 502                         goto out_put_upper;
 503         }
 504
 505         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 506                 bool opaque = false;
 507                 struct path lowerpath = poe->lowerstack[i];
 508
 509                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 510                 err = PTR_ERR(this);
 511                 if (IS_ERR(this)) {
 512                         /*
 513                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 514                          */
 515                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 516                                 continue;
 517                         goto out_put;
 518                 }
 519                 if (!this)
 520                         continue;
 521                 if (ovl_is_whiteout(this)) {
 522                         dput(this);
 523                         break;
 524                 }
 525                 /*
 526                  * Only makes sense to check opaque dir if this is not the
 527                  * lowermost layer.
 528                  */
 529                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 530                         opaque = true;
 531
 532                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 533                              !S_ISDIR(this->d_inode->i_mode))) {
 534                         /*
 535                          * FIXME: check for upper-opaqueness maybe better done
 536                          * in remove code.
 537                          */
 538                         if (prev == upperdentry)
 539                                 upperopaque = true;
 540                         dput(this);
 541                         break;
 542                 }
 543                 /*
 544                  * If this is a non-directory then stop here.
 545                  */
 546                 if (!S_ISDIR(this->d_inode->i_mode))
 547                         opaque = true;
 548
 549                 stack[ctr].dentry = this;
 550                 stack[ctr].mnt = lowerpath.mnt;
 551                 ctr++;
 552                 prev = this;
 553                 if (opaque)
 554                         break;
 555         }
 556
 557         oe = ovl_alloc_entry(ctr);
 558         err = -ENOMEM;
 559         if (!oe)
 560                 goto out_put;
 561
 562         if (upperdentry || ctr) {
 563                 struct dentry *realdentry;
 564
 565                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 566
 567                 err = -ENOMEM;
 568                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 569                                       oe);
 570                 if (!inode)
 571                         goto out_free_oe;
 572                 ovl_copyattr(realdentry->d_inode, inode);
 573         }
 574
 575         oe->opaque = upperopaque;
 576         oe->__upperdentry = upperdentry;
 577         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 578         kfree(stack);
 579         dentry->d_fsdata = oe;
 580         d_add(dentry, inode);
 581
 582         return NULL;
 583
 584 out_free_oe:
 585         kfree(oe);
 586 out_put:
 587         for (i = 0; i < ctr; i++)
 588                 dput(stack[i].dentry);
 589         kfree(stack);
 590 out_put_upper:
 591         dput(upperdentry);
 592 out:
 593         return ERR_PTR(err);
 594 }
 595
 596 struct file *ovl_path_open(struct path *path, int flags)
 597 {
 598         return dentry_open(path, flags, current_cred());
 599 }
 600
 601 static void ovl_put_super(struct super_block *sb)
 602 {
 603         struct ovl_fs *ufs = sb->s_fs_info;
 604         unsigned i;
 605
 606         dput(ufs->workdir);
 607         mntput(ufs->upper_mnt);
 608         for (i = 0; i < ufs->numlower; i++)
 609                 mntput(ufs->lower_mnt[i]);
 610         kfree(ufs->lower_mnt);
 611
 612         kfree(ufs->config.lowerdir);
 613         kfree(ufs->config.upperdir);
 614         kfree(ufs->config.workdir);
 615         put_cred(ufs->creator_cred);
 616         kfree(ufs);
 617 }
 618
 619 /**
 620  * ovl_statfs
 621  * @sb: The overlayfs super block
 622  * @buf: The struct kstatfs to fill in with stats
 623  *
 624  * Get the filesystem statistics.  As writes always target the upper layer
 625  * filesystem pass the statfs to the upper filesystem (if it exists)
 626  */
 627 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 628 {
 629         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 630         struct dentry *root_dentry = dentry->d_sb->s_root;
 631         struct path path;
 632         int err;
 633
 634         ovl_path_real(root_dentry, &path);
 635
 636         err = vfs_statfs(&path, buf);
 637         if (!err) {
 638                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 639                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 640         }
 641
 642         return err;
 643 }
 644
 645 /**
 646  * ovl_show_options
 647  *
 648  * Prints the mount options for a given superblock.
 649  * Returns zero; does not fail.
 650  */
 651 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 652 {
 653         struct super_block *sb = dentry->d_sb;
 654         struct ovl_fs *ufs = sb->s_fs_info;
 655
 656         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 657         if (ufs->config.upperdir) {
 658                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 659                 seq_show_option(m, "workdir", ufs->config.workdir);
 660         }
 661         if (ufs->config.default_permissions)
 662                 seq_puts(m, ",default_permissions");
 663         return 0;
 664 }
 665
 666 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 667 {
 668         struct ovl_fs *ufs = sb->s_fs_info;
 669
 670         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 671                 return -EROFS;
 672
 673         return 0;
 674 }
 675
 676 static const struct super_operations ovl_super_operations = {
 677         .put_super      = ovl_put_super,
 678         .statfs         = ovl_statfs,
 679         .show_options   = ovl_show_options,
 680         .remount_fs     = ovl_remount,
 681 };
 682
 683 enum {
 684         OPT_LOWERDIR,
 685         OPT_UPPERDIR,
 686         OPT_WORKDIR,
 687         OPT_DEFAULT_PERMISSIONS,
 688         OPT_ERR,
 689 };
 690
 691 static const match_table_t ovl_tokens = {
 692         {OPT_LOWERDIR,                  "lowerdir=%s"},
 693         {OPT_UPPERDIR,                  "upperdir=%s"},
 694         {OPT_WORKDIR,                   "workdir=%s"},
 695         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 696         {OPT_ERR,                       NULL}
 697 };
 698
 699 static char *ovl_next_opt(char **s)
 700 {
 701         char *sbegin = *s;
 702         char *p;
 703
 704         if (sbegin == NULL)
 705                 return NULL;
 706
 707         for (p = sbegin; *p; p++) {
 708                 if (*p == '\\') {
 709                         p++;
 710                         if (!*p)
 711                                 break;
 712                 } else if (*p == ',') {
 713                         *p = '\0';
 714                         *s = p + 1;
 715                         return sbegin;
 716                 }
 717         }
 718         *s = NULL;
 719         return sbegin;
 720 }
 721
 722 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 723 {
 724         char *p;
 725
 726         while ((p = ovl_next_opt(&opt)) != NULL) {
 727                 int token;
 728                 substring_t args[MAX_OPT_ARGS];
 729
 730                 if (!*p)
 731                         continue;
 732
 733                 token = match_token(p, ovl_tokens, args);
 734                 switch (token) {
 735                 case OPT_UPPERDIR:
 736                         kfree(config->upperdir);
 737                         config->upperdir = match_strdup(&args[0]);
 738                         if (!config->upperdir)
 739                                 return -ENOMEM;
 740                         break;
 741
 742                 case OPT_LOWERDIR:
 743                         kfree(config->lowerdir);
 744                         config->lowerdir = match_strdup(&args[0]);
 745                         if (!config->lowerdir)
 746                                 return -ENOMEM;
 747                         break;
 748
 749                 case OPT_WORKDIR:
 750                         kfree(config->workdir);
 751                         config->workdir = match_strdup(&args[0]);
 752                         if (!config->workdir)
 753                                 return -ENOMEM;
 754                         break;
 755
 756                 case OPT_DEFAULT_PERMISSIONS:
 757                         config->default_permissions = true;
 758                         break;
 759
 760                 default:
 761                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 762                         return -EINVAL;
 763                 }
 764         }
 765
 766         /* Workdir is useless in non-upper mount */
 767         if (!config->upperdir && config->workdir) {
 768                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 769                         config->workdir);
 770                 kfree(config->workdir);
 771                 config->workdir = NULL;
 772         }
 773
 774         return 0;
 775 }
 776
 777 #define OVL_WORKDIR_NAME "work"
 778
 779 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 780                                          struct dentry *dentry)
 781 {
 782         struct inode *dir = dentry->d_inode;
 783         struct dentry *work;
 784         int err;
 785         bool retried = false;
 786
 787         err = mnt_want_write(mnt);
 788         if (err)
 789                 return ERR_PTR(err);
 790
 791         inode_lock_nested(dir, I_MUTEX_PARENT);
 792 retry:
 793         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 794                               strlen(OVL_WORKDIR_NAME));
 795
 796         if (!IS_ERR(work)) {
 797                 struct kstat stat = {
 798                         .mode = S_IFDIR | 0,
 799                 };
 800
 801                 if (work->d_inode) {
 802                         err = -EEXIST;
 803                         if (retried)
 804                                 goto out_dput;
 805
 806                         retried = true;
 807                         ovl_cleanup(dir, work);
 808                         dput(work);
 809                         goto retry;
 810                 }
 811
 812                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 813                 if (err)
 814                         goto out_dput;
 815         }
 816 out_unlock:
 817         inode_unlock(dir);
 818         mnt_drop_write(mnt);
 819
 820         return work;
 821
 822 out_dput:
 823         dput(work);
 824         work = ERR_PTR(err);
 825         goto out_unlock;
 826 }
 827
 828 static void ovl_unescape(char *s)
 829 {
 830         char *d = s;
 831
 832         for (;; s++, d++) {
 833                 if (*s == '\\')
 834                         s++;
 835                 *d = *s;
 836                 if (!*s)
 837                         break;
 838         }
 839 }
 840
 841 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 842 {
 843         int err = -EINVAL;
 844
 845         if (!*name) {
 846                 pr_err("overlayfs: empty lowerdir\n");
 847                 goto out;
 848         }
 849         err = kern_path(name, LOOKUP_FOLLOW, path);
 850         if (err) {
 851                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 852                 goto out;
 853         }
 854         err = -EINVAL;
 855         if (ovl_dentry_weird(path->dentry)) {
 856                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 857                 goto out_put;
 858         }
 859         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 860                 pr_err("overlayfs: '%s' not a directory\n", name);
 861                 goto out_put;
 862         }
 863         return 0;
 864
 865 out_put:
 866         path_put(path);
 867 out:
 868         return err;
 869 }
 870
 871 static int ovl_mount_dir(const char *name, struct path *path)
 872 {
 873         int err = -ENOMEM;
 874         char *tmp = kstrdup(name, GFP_KERNEL);
 875
 876         if (tmp) {
 877                 ovl_unescape(tmp);
 878                 err = ovl_mount_dir_noesc(tmp, path);
 879
 880                 if (!err)
 881                         if (ovl_dentry_remote(path->dentry)) {
 882                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 883                                        tmp);
 884                                 path_put(path);
 885                                 err = -EINVAL;
 886                         }
 887                 kfree(tmp);
 888         }
 889         return err;
 890 }
 891
 892 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 893                          int *stack_depth, bool *remote)
 894 {
 895         int err;
 896         struct kstatfs statfs;
 897
 898         err = ovl_mount_dir_noesc(name, path);
 899         if (err)
 900                 goto out;
 901
 902         err = vfs_statfs(path, &statfs);
 903         if (err) {
 904                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 905                 goto out_put;
 906         }
 907         *namelen = max(*namelen, statfs.f_namelen);
 908         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 909
 910         if (ovl_dentry_remote(path->dentry))
 911                 *remote = true;
 912
 913         return 0;
 914
 915 out_put:
 916         path_put(path);
 917 out:
 918         return err;
 919 }
 920
 921 /* Workdir should not be subdir of upperdir and vice versa */
 922 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 923 {
 924         bool ok = false;
 925
 926         if (workdir != upperdir) {
 927                 ok = (lock_rename(workdir, upperdir) == NULL);
 928                 unlock_rename(workdir, upperdir);
 929         }
 930         return ok;
 931 }
 932
 933 static unsigned int ovl_split_lowerdirs(char *str)
 934 {
 935         unsigned int ctr = 1;
 936         char *s, *d;
 937
 938         for (s = d = str;; s++, d++) {
 939                 if (*s == '\\') {
 940                         s++;
 941                 } else if (*s == ':') {
 942                         *d = '\0';
 943                         ctr++;
 944                         continue;
 945                 }
 946                 *d = *s;
 947                 if (!*s)
 948                         break;
 949         }
 950         return ctr;
 951 }
 952
 953 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 954 {
 955         struct path upperpath = { NULL, NULL };
 956         struct path workpath = { NULL, NULL };
 957         struct dentry *root_dentry;
 958         struct ovl_entry *oe;
 959         struct ovl_fs *ufs;
 960         struct path *stack = NULL;
 961         char *lowertmp;
 962         char *lower;
 963         unsigned int numlower;
 964         unsigned int stacklen = 0;
 965         unsigned int i;
 966         bool remote = false;
 967         int err;
 968
 969         err = -ENOMEM;
 970         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 971         if (!ufs)
 972                 goto out;
 973
 974         err = ovl_parse_opt((char *) data, &ufs->config);
 975         if (err)
 976                 goto out_free_config;
 977
 978         err = -EINVAL;
 979         if (!ufs->config.lowerdir) {
 980                 if (!silent)
 981                         pr_err("overlayfs: missing 'lowerdir'\n");
 982                 goto out_free_config;
 983         }
 984
 985         sb->s_stack_depth = 0;
 986         sb->s_maxbytes = MAX_LFS_FILESIZE;
 987         if (ufs->config.upperdir) {
 988                 if (!ufs->config.workdir) {
 989                         pr_err("overlayfs: missing 'workdir'\n");
 990                         goto out_free_config;
 991                 }
 992
 993                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 994                 if (err)
 995                         goto out_free_config;
 996
 997                 /* Upper fs should not be r/o */
 998                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 999                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
1000                         err = -EINVAL;
1001                         goto out_put_upperpath;
1002                 }
1003
1004                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
1005                 if (err)
1006                         goto out_put_upperpath;
1007
1008                 err = -EINVAL;
1009                 if (upperpath.mnt != workpath.mnt) {
1010                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1011                         goto out_put_workpath;
1012                 }
1013                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
1014                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1015                         goto out_put_workpath;
1016                 }
1017                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
1018         }
1019         err = -ENOMEM;
1020         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
1021         if (!lowertmp)
1022                 goto out_put_workpath;
1023
1024         err = -EINVAL;
1025         stacklen = ovl_split_lowerdirs(lowertmp);
1026         if (stacklen > OVL_MAX_STACK) {
1027                 pr_err("overlayfs: too many lower directries, limit is %d\n",
1028                        OVL_MAX_STACK);
1029                 goto out_free_lowertmp;
1030         } else if (!ufs->config.upperdir && stacklen == 1) {
1031                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1032                 goto out_free_lowertmp;
1033         }
1034
1035         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1036         if (!stack)
1037                 goto out_free_lowertmp;
1038
1039         lower = lowertmp;
1040         for (numlower = 0; numlower < stacklen; numlower++) {
1041                 err = ovl_lower_dir(lower, &stack[numlower],
1042                                     &ufs->lower_namelen, &sb->s_stack_depth,
1043                                     &remote);
1044                 if (err)
1045                         goto out_put_lowerpath;
1046
1047                 lower = strchr(lower, '\0') + 1;
1048         }
1049
1050         err = -EINVAL;
1051         sb->s_stack_depth++;
1052         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1053                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1054                 goto out_put_lowerpath;
1055         }
1056
1057         if (ufs->config.upperdir) {
1058                 ufs->upper_mnt = clone_private_mount(&upperpath);
1059                 err = PTR_ERR(ufs->upper_mnt);
1060                 if (IS_ERR(ufs->upper_mnt)) {
1061                         pr_err("overlayfs: failed to clone upperpath\n");
1062                         goto out_put_lowerpath;
1063                 }
1064
1065                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
1066                 err = PTR_ERR(ufs->workdir);
1067                 if (IS_ERR(ufs->workdir)) {
1068                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1069                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1070                         sb->s_flags |= MS_RDONLY;
1071                         ufs->workdir = NULL;
1072                 }
1073
1074                 /*
1075                  * Upper should support d_type, else whiteouts are visible.
1076                  * Given workdir and upper are on same fs, we can do
1077                  * iterate_dir() on workdir. This check requires successful
1078                  * creation of workdir in previous step.
1079                  */
1080                 if (ufs->workdir) {
1081                         err = ovl_check_d_type_supported(&workpath);
1082                         if (err < 0)
1083                                 goto out_put_workdir;
1084
1085                         if (!err) {
1086                                 pr_err("overlayfs: upper fs needs to support d_type.\n");
1087                                 err = -EINVAL;
1088                                 goto out_put_workdir;
1089                         }
1090                 }
1091         }
1092
1093         err = -ENOMEM;
1094         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1095         if (ufs->lower_mnt == NULL)
1096                 goto out_put_workdir;
1097         for (i = 0; i < numlower; i++) {
1098                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1099
1100                 err = PTR_ERR(mnt);
1101                 if (IS_ERR(mnt)) {
1102                         pr_err("overlayfs: failed to clone lowerpath\n");
1103                         goto out_put_lower_mnt;
1104                 }
1105                 /*
1106                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1107                  * will fail instead of modifying lower fs.
1108                  */
1109                 mnt->mnt_flags |= MNT_READONLY;
1110
1111                 ufs->lower_mnt[ufs->numlower] = mnt;
1112                 ufs->numlower++;
1113         }
1114
1115         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1116         if (!ufs->upper_mnt)
1117                 sb->s_flags |= MS_RDONLY;
1118
1119         if (remote)
1120                 sb->s_d_op = &ovl_reval_dentry_operations;
1121         else
1122                 sb->s_d_op = &ovl_dentry_operations;
1123
1124         ufs->creator_cred = prepare_creds();
1125         if (!ufs->creator_cred)
1126                 goto out_put_lower_mnt;
1127
1128         err = -ENOMEM;
1129         oe = ovl_alloc_entry(numlower);
1130         if (!oe)
1131                 goto out_put_cred;
1132
1133         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1134         if (!root_dentry)
1135                 goto out_free_oe;
1136
1137         mntput(upperpath.mnt);
1138         for (i = 0; i < numlower; i++)
1139                 mntput(stack[i].mnt);
1140         path_put(&workpath);
1141         kfree(lowertmp);
1142
1143         oe->__upperdentry = upperpath.dentry;
1144         for (i = 0; i < numlower; i++) {
1145                 oe->lowerstack[i].dentry = stack[i].dentry;
1146                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1147         }
1148         kfree(stack);
1149
1150         root_dentry->d_fsdata = oe;
1151
1152         ovl_copyattr(ovl_dentry_real(root_dentry)->d_inode,
1153                      root_dentry->d_inode);
1154
1155         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1156         sb->s_op = &ovl_super_operations;
1157         sb->s_root = root_dentry;
1158         sb->s_fs_info = ufs;
1159
1160         return 0;
1161
1162 out_free_oe:
1163         kfree(oe);
1164 out_put_cred:
1165         put_cred(ufs->creator_cred);
1166 out_put_lower_mnt:
1167         for (i = 0; i < ufs->numlower; i++)
1168                 mntput(ufs->lower_mnt[i]);
1169         kfree(ufs->lower_mnt);
1170 out_put_workdir:
1171         dput(ufs->workdir);
1172         mntput(ufs->upper_mnt);
1173 out_put_lowerpath:
1174         for (i = 0; i < numlower; i++)
1175                 path_put(&stack[i]);
1176         kfree(stack);
1177 out_free_lowertmp:
1178         kfree(lowertmp);
1179 out_put_workpath:
1180         path_put(&workpath);
1181 out_put_upperpath:
1182         path_put(&upperpath);
1183 out_free_config:
1184         kfree(ufs->config.lowerdir);
1185         kfree(ufs->config.upperdir);
1186         kfree(ufs->config.workdir);
1187         kfree(ufs);
1188 out:
1189         return err;
1190 }
1191
1192 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1193                                 const char *dev_name, void *raw_data)
1194 {
1195         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1196 }
1197
1198 static struct file_system_type ovl_fs_type = {
1199         .owner          = THIS_MODULE,
1200         .name           = "overlay",
1201         .mount          = ovl_mount,
1202         .kill_sb        = kill_anon_super,
1203 };
1204 MODULE_ALIAS_FS("overlay");
1205
1206 static int __init ovl_init(void)
1207 {
1208         return register_filesystem(&ovl_fs_type);
1209 }
1210
1211 static void __exit ovl_exit(void)
1212 {
1213         unregister_filesystem(&ovl_fs_type);
1214 }
1215
1216 module_init(ovl_init);
1217 module_exit(ovl_exit);