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 inode *ovl_inode_real(struct inode *inode)
 163 {
 164         bool tmp;
 165
 166         return d_inode(ovl_entry_real(inode->i_private, &tmp));
 167 }
 168
 169 struct vfsmount *ovl_entry_mnt_real(struct ovl_entry *oe, struct inode *inode,
 170                                     bool is_upper)
 171 {
 172         if (is_upper) {
 173                 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 174
 175                 return ofs->upper_mnt;
 176         } else {
 177                 return oe->numlower ? oe->lowerstack[0].mnt : NULL;
 178         }
 179 }
 180
 181 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 182 {
 183         struct ovl_entry *oe = dentry->d_fsdata;
 184
 185         return oe->cache;
 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,
 308                                  const struct inode *inode,
 309                                  unsigned int open_flags)
 310 {
 311         struct dentry *real;
 312
 313         if (d_is_dir(dentry)) {
 314                 if (!inode || inode == d_inode(dentry))
 315                         return dentry;
 316                 goto bug;
 317         }
 318
 319         if (d_is_negative(dentry))
 320                 return dentry;
 321
 322         if (open_flags) {
 323                 int err = ovl_open_maybe_copy_up(dentry, open_flags);
 324
 325                 if (err)
 326                         return ERR_PTR(err);
 327         }
 328
 329         real = ovl_dentry_upper(dentry);
 330         if (real && (!inode || inode == d_inode(real)))
 331                 return real;
 332
 333         real = ovl_dentry_lower(dentry);
 334         if (!real)
 335                 goto bug;
 336
 337         if (!inode || inode == d_inode(real))
 338                 return real;
 339
 340         /* Handle recursion */
 341         return d_real(real, inode, open_flags);
 342 bug:
 343         WARN(1, "ovl_d_real(%pd4, %s:%lu\n): real dentry not found\n", dentry,
 344              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 345         return dentry;
 346 }
 347
 348 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 349 {
 350         struct ovl_entry *oe = dentry->d_fsdata;
 351         unsigned int i;
 352         int ret = 1;
 353
 354         for (i = 0; i < oe->numlower; i++) {
 355                 struct dentry *d = oe->lowerstack[i].dentry;
 356
 357                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 358                         ret = d->d_op->d_revalidate(d, flags);
 359                         if (ret < 0)
 360                                 return ret;
 361                         if (!ret) {
 362                                 if (!(flags & LOOKUP_RCU))
 363                                         d_invalidate(d);
 364                                 return -ESTALE;
 365                         }
 366                 }
 367         }
 368         return 1;
 369 }
 370
 371 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 372 {
 373         struct ovl_entry *oe = dentry->d_fsdata;
 374         unsigned int i;
 375         int ret = 1;
 376
 377         for (i = 0; i < oe->numlower; i++) {
 378                 struct dentry *d = oe->lowerstack[i].dentry;
 379
 380                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 381                         ret = d->d_op->d_weak_revalidate(d, flags);
 382                         if (ret <= 0)
 383                                 break;
 384                 }
 385         }
 386         return ret;
 387 }
 388
 389 static const struct dentry_operations ovl_dentry_operations = {
 390         .d_release = ovl_dentry_release,
 391         .d_real = ovl_d_real,
 392 };
 393
 394 static const struct dentry_operations ovl_reval_dentry_operations = {
 395         .d_release = ovl_dentry_release,
 396         .d_real = ovl_d_real,
 397         .d_revalidate = ovl_dentry_revalidate,
 398         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 399 };
 400
 401 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 402 {
 403         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 404         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 405
 406         if (oe)
 407                 oe->numlower = numlower;
 408
 409         return oe;
 410 }
 411
 412 static bool ovl_dentry_remote(struct dentry *dentry)
 413 {
 414         return dentry->d_flags &
 415                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 416 }
 417
 418 static bool ovl_dentry_weird(struct dentry *dentry)
 419 {
 420         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 421                                   DCACHE_MANAGE_TRANSIT |
 422                                   DCACHE_OP_HASH |
 423                                   DCACHE_OP_COMPARE);
 424 }
 425
 426 static inline struct dentry *ovl_lookup_real(struct super_block *ovl_sb,
 427                                              struct dentry *dir,
 428                                              struct qstr *name)
 429 {
 430         const struct cred *old_cred;
 431         struct dentry *dentry;
 432
 433         old_cred = ovl_override_creds(ovl_sb);
 434         dentry = lookup_one_len_unlocked(name->name, dir, name->len);
 435         revert_creds(old_cred);
 436
 437         if (IS_ERR(dentry)) {
 438                 if (PTR_ERR(dentry) == -ENOENT)
 439                         dentry = NULL;
 440         } else if (!dentry->d_inode) {
 441                 dput(dentry);
 442                 dentry = NULL;
 443         } else if (ovl_dentry_weird(dentry)) {
 444                 dput(dentry);
 445                 /* Don't support traversing automounts and other weirdness */
 446                 dentry = ERR_PTR(-EREMOTE);
 447         }
 448         return dentry;
 449 }
 450
 451 /*
 452  * Returns next layer in stack starting from top.
 453  * Returns -1 if this is the last layer.
 454  */
 455 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 456 {
 457         struct ovl_entry *oe = dentry->d_fsdata;
 458
 459         BUG_ON(idx < 0);
 460         if (idx == 0) {
 461                 ovl_path_upper(dentry, path);
 462                 if (path->dentry)
 463                         return oe->numlower ? 1 : -1;
 464                 idx++;
 465         }
 466         BUG_ON(idx > oe->numlower);
 467         *path = oe->lowerstack[idx - 1];
 468
 469         return (idx < oe->numlower) ? idx + 1 : -1;
 470 }
 471
 472 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 473                           unsigned int flags)
 474 {
 475         struct ovl_entry *oe;
 476         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 477         struct path *stack = NULL;
 478         struct dentry *upperdir, *upperdentry = NULL;
 479         unsigned int ctr = 0;
 480         struct inode *inode = NULL;
 481         bool upperopaque = false;
 482         struct dentry *this, *prev = NULL;
 483         unsigned int i;
 484         int err;
 485
 486         upperdir = ovl_upperdentry_dereference(poe);
 487         if (upperdir) {
 488                 this = ovl_lookup_real(dentry->d_sb, upperdir, &dentry->d_name);
 489                 err = PTR_ERR(this);
 490                 if (IS_ERR(this))
 491                         goto out;
 492
 493                 if (this) {
 494                         if (unlikely(ovl_dentry_remote(this))) {
 495                                 dput(this);
 496                                 err = -EREMOTE;
 497                                 goto out;
 498                         }
 499                         if (ovl_is_whiteout(this)) {
 500                                 dput(this);
 501                                 this = NULL;
 502                                 upperopaque = true;
 503                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 504                                 upperopaque = true;
 505                         }
 506                 }
 507                 upperdentry = prev = this;
 508         }
 509
 510         if (!upperopaque && poe->numlower) {
 511                 err = -ENOMEM;
 512                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 513                 if (!stack)
 514                         goto out_put_upper;
 515         }
 516
 517         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 518                 bool opaque = false;
 519                 struct path lowerpath = poe->lowerstack[i];
 520
 521                 this = ovl_lookup_real(dentry->d_sb,
 522                                        lowerpath.dentry, &dentry->d_name);
 523                 err = PTR_ERR(this);
 524                 if (IS_ERR(this)) {
 525                         /*
 526                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 527                          */
 528                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 529                                 continue;
 530                         goto out_put;
 531                 }
 532                 if (!this)
 533                         continue;
 534                 if (ovl_is_whiteout(this)) {
 535                         dput(this);
 536                         break;
 537                 }
 538                 /*
 539                  * Only makes sense to check opaque dir if this is not the
 540                  * lowermost layer.
 541                  */
 542                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 543                         opaque = true;
 544
 545                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 546                              !S_ISDIR(this->d_inode->i_mode))) {
 547                         /*
 548                          * FIXME: check for upper-opaqueness maybe better done
 549                          * in remove code.
 550                          */
 551                         if (prev == upperdentry)
 552                                 upperopaque = true;
 553                         dput(this);
 554                         break;
 555                 }
 556                 /*
 557                  * If this is a non-directory then stop here.
 558                  */
 559                 if (!S_ISDIR(this->d_inode->i_mode))
 560                         opaque = true;
 561
 562                 stack[ctr].dentry = this;
 563                 stack[ctr].mnt = lowerpath.mnt;
 564                 ctr++;
 565                 prev = this;
 566                 if (opaque)
 567                         break;
 568         }
 569
 570         oe = ovl_alloc_entry(ctr);
 571         err = -ENOMEM;
 572         if (!oe)
 573                 goto out_put;
 574
 575         if (upperdentry || ctr) {
 576                 struct dentry *realdentry;
 577
 578                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 579
 580                 err = -ENOMEM;
 581                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 582                                       oe);
 583                 if (!inode)
 584                         goto out_free_oe;
 585                 ovl_copyattr(realdentry->d_inode, inode);
 586         }
 587
 588         oe->opaque = upperopaque;
 589         oe->__upperdentry = upperdentry;
 590         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 591         kfree(stack);
 592         dentry->d_fsdata = oe;
 593         d_add(dentry, inode);
 594
 595         return NULL;
 596
 597 out_free_oe:
 598         kfree(oe);
 599 out_put:
 600         for (i = 0; i < ctr; i++)
 601                 dput(stack[i].dentry);
 602         kfree(stack);
 603 out_put_upper:
 604         dput(upperdentry);
 605 out:
 606         return ERR_PTR(err);
 607 }
 608
 609 struct file *ovl_path_open(struct path *path, int flags)
 610 {
 611         return dentry_open(path, flags | O_NOATIME, current_cred());
 612 }
 613
 614 static void ovl_put_super(struct super_block *sb)
 615 {
 616         struct ovl_fs *ufs = sb->s_fs_info;
 617         unsigned i;
 618
 619         dput(ufs->workdir);
 620         mntput(ufs->upper_mnt);
 621         for (i = 0; i < ufs->numlower; i++)
 622                 mntput(ufs->lower_mnt[i]);
 623         kfree(ufs->lower_mnt);
 624
 625         kfree(ufs->config.lowerdir);
 626         kfree(ufs->config.upperdir);
 627         kfree(ufs->config.workdir);
 628         put_cred(ufs->creator_cred);
 629         kfree(ufs);
 630 }
 631
 632 /**
 633  * ovl_statfs
 634  * @sb: The overlayfs super block
 635  * @buf: The struct kstatfs to fill in with stats
 636  *
 637  * Get the filesystem statistics.  As writes always target the upper layer
 638  * filesystem pass the statfs to the upper filesystem (if it exists)
 639  */
 640 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 641 {
 642         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 643         struct dentry *root_dentry = dentry->d_sb->s_root;
 644         struct path path;
 645         int err;
 646
 647         ovl_path_real(root_dentry, &path);
 648
 649         err = vfs_statfs(&path, buf);
 650         if (!err) {
 651                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 652                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 653         }
 654
 655         return err;
 656 }
 657
 658 /**
 659  * ovl_show_options
 660  *
 661  * Prints the mount options for a given superblock.
 662  * Returns zero; does not fail.
 663  */
 664 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 665 {
 666         struct super_block *sb = dentry->d_sb;
 667         struct ovl_fs *ufs = sb->s_fs_info;
 668
 669         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 670         if (ufs->config.upperdir) {
 671                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 672                 seq_show_option(m, "workdir", ufs->config.workdir);
 673         }
 674         if (ufs->config.default_permissions)
 675                 seq_puts(m, ",default_permissions");
 676         return 0;
 677 }
 678
 679 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 680 {
 681         struct ovl_fs *ufs = sb->s_fs_info;
 682
 683         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 684                 return -EROFS;
 685
 686         return 0;
 687 }
 688
 689 static const struct super_operations ovl_super_operations = {
 690         .put_super      = ovl_put_super,
 691         .statfs         = ovl_statfs,
 692         .show_options   = ovl_show_options,
 693         .remount_fs     = ovl_remount,
 694         .drop_inode     = generic_delete_inode,
 695 };
 696
 697 enum {
 698         OPT_LOWERDIR,
 699         OPT_UPPERDIR,
 700         OPT_WORKDIR,
 701         OPT_DEFAULT_PERMISSIONS,
 702         OPT_ERR,
 703 };
 704
 705 static const match_table_t ovl_tokens = {
 706         {OPT_LOWERDIR,                  "lowerdir=%s"},
 707         {OPT_UPPERDIR,                  "upperdir=%s"},
 708         {OPT_WORKDIR,                   "workdir=%s"},
 709         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 710         {OPT_ERR,                       NULL}
 711 };
 712
 713 static char *ovl_next_opt(char **s)
 714 {
 715         char *sbegin = *s;
 716         char *p;
 717
 718         if (sbegin == NULL)
 719                 return NULL;
 720
 721         for (p = sbegin; *p; p++) {
 722                 if (*p == '\\') {
 723                         p++;
 724                         if (!*p)
 725                                 break;
 726                 } else if (*p == ',') {
 727                         *p = '\0';
 728                         *s = p + 1;
 729                         return sbegin;
 730                 }
 731         }
 732         *s = NULL;
 733         return sbegin;
 734 }
 735
 736 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 737 {
 738         char *p;
 739
 740         while ((p = ovl_next_opt(&opt)) != NULL) {
 741                 int token;
 742                 substring_t args[MAX_OPT_ARGS];
 743
 744                 if (!*p)
 745                         continue;
 746
 747                 token = match_token(p, ovl_tokens, args);
 748                 switch (token) {
 749                 case OPT_UPPERDIR:
 750                         kfree(config->upperdir);
 751                         config->upperdir = match_strdup(&args[0]);
 752                         if (!config->upperdir)
 753                                 return -ENOMEM;
 754                         break;
 755
 756                 case OPT_LOWERDIR:
 757                         kfree(config->lowerdir);
 758                         config->lowerdir = match_strdup(&args[0]);
 759                         if (!config->lowerdir)
 760                                 return -ENOMEM;
 761                         break;
 762
 763                 case OPT_WORKDIR:
 764                         kfree(config->workdir);
 765                         config->workdir = match_strdup(&args[0]);
 766                         if (!config->workdir)
 767                                 return -ENOMEM;
 768                         break;
 769
 770                 case OPT_DEFAULT_PERMISSIONS:
 771                         config->default_permissions = true;
 772                         break;
 773
 774                 default:
 775                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 776                         return -EINVAL;
 777                 }
 778         }
 779
 780         /* Workdir is useless in non-upper mount */
 781         if (!config->upperdir && config->workdir) {
 782                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 783                         config->workdir);
 784                 kfree(config->workdir);
 785                 config->workdir = NULL;
 786         }
 787
 788         return 0;
 789 }
 790
 791 #define OVL_WORKDIR_NAME "work"
 792
 793 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 794                                          struct dentry *dentry)
 795 {
 796         struct inode *dir = dentry->d_inode;
 797         struct dentry *work;
 798         int err;
 799         bool retried = false;
 800
 801         err = mnt_want_write(mnt);
 802         if (err)
 803                 return ERR_PTR(err);
 804
 805         inode_lock_nested(dir, I_MUTEX_PARENT);
 806 retry:
 807         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 808                               strlen(OVL_WORKDIR_NAME));
 809
 810         if (!IS_ERR(work)) {
 811                 struct kstat stat = {
 812                         .mode = S_IFDIR | 0,
 813                 };
 814
 815                 if (work->d_inode) {
 816                         err = -EEXIST;
 817                         if (retried)
 818                                 goto out_dput;
 819
 820                         retried = true;
 821                         ovl_cleanup(dir, work);
 822                         dput(work);
 823                         goto retry;
 824                 }
 825
 826                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 827                 if (err)
 828                         goto out_dput;
 829         }
 830 out_unlock:
 831         inode_unlock(dir);
 832         mnt_drop_write(mnt);
 833
 834         return work;
 835
 836 out_dput:
 837         dput(work);
 838         work = ERR_PTR(err);
 839         goto out_unlock;
 840 }
 841
 842 static void ovl_unescape(char *s)
 843 {
 844         char *d = s;
 845
 846         for (;; s++, d++) {
 847                 if (*s == '\\')
 848                         s++;
 849                 *d = *s;
 850                 if (!*s)
 851                         break;
 852         }
 853 }
 854
 855 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 856 {
 857         int err = -EINVAL;
 858
 859         if (!*name) {
 860                 pr_err("overlayfs: empty lowerdir\n");
 861                 goto out;
 862         }
 863         err = kern_path(name, LOOKUP_FOLLOW, path);
 864         if (err) {
 865                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 866                 goto out;
 867         }
 868         err = -EINVAL;
 869         if (ovl_dentry_weird(path->dentry)) {
 870                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 871                 goto out_put;
 872         }
 873         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 874                 pr_err("overlayfs: '%s' not a directory\n", name);
 875                 goto out_put;
 876         }
 877         return 0;
 878
 879 out_put:
 880         path_put(path);
 881 out:
 882         return err;
 883 }
 884
 885 static int ovl_mount_dir(const char *name, struct path *path)
 886 {
 887         int err = -ENOMEM;
 888         char *tmp = kstrdup(name, GFP_KERNEL);
 889
 890         if (tmp) {
 891                 ovl_unescape(tmp);
 892                 err = ovl_mount_dir_noesc(tmp, path);
 893
 894                 if (!err)
 895                         if (ovl_dentry_remote(path->dentry)) {
 896                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 897                                        tmp);
 898                                 path_put(path);
 899                                 err = -EINVAL;
 900                         }
 901                 kfree(tmp);
 902         }
 903         return err;
 904 }
 905
 906 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 907                          int *stack_depth, bool *remote)
 908 {
 909         int err;
 910         struct kstatfs statfs;
 911
 912         err = ovl_mount_dir_noesc(name, path);
 913         if (err)
 914                 goto out;
 915
 916         err = vfs_statfs(path, &statfs);
 917         if (err) {
 918                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 919                 goto out_put;
 920         }
 921         *namelen = max(*namelen, statfs.f_namelen);
 922         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 923
 924         if (ovl_dentry_remote(path->dentry))
 925                 *remote = true;
 926
 927         return 0;
 928
 929 out_put:
 930         path_put(path);
 931 out:
 932         return err;
 933 }
 934
 935 /* Workdir should not be subdir of upperdir and vice versa */
 936 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 937 {
 938         bool ok = false;
 939
 940         if (workdir != upperdir) {
 941                 ok = (lock_rename(workdir, upperdir) == NULL);
 942                 unlock_rename(workdir, upperdir);
 943         }
 944         return ok;
 945 }
 946
 947 static unsigned int ovl_split_lowerdirs(char *str)
 948 {
 949         unsigned int ctr = 1;
 950         char *s, *d;
 951
 952         for (s = d = str;; s++, d++) {
 953                 if (*s == '\\') {
 954                         s++;
 955                 } else if (*s == ':') {
 956                         *d = '\0';
 957                         ctr++;
 958                         continue;
 959                 }
 960                 *d = *s;
 961                 if (!*s)
 962                         break;
 963         }
 964         return ctr;
 965 }
 966
 967 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 968 {
 969         struct path upperpath = { NULL, NULL };
 970         struct path workpath = { NULL, NULL };
 971         struct dentry *root_dentry;
 972         struct ovl_entry *oe;
 973         struct ovl_fs *ufs;
 974         struct path *stack = NULL;
 975         char *lowertmp;
 976         char *lower;
 977         unsigned int numlower;
 978         unsigned int stacklen = 0;
 979         unsigned int i;
 980         bool remote = false;
 981         int err;
 982
 983         err = -ENOMEM;
 984         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 985         if (!ufs)
 986                 goto out;
 987
 988         err = ovl_parse_opt((char *) data, &ufs->config);
 989         if (err)
 990                 goto out_free_config;
 991
 992         err = -EINVAL;
 993         if (!ufs->config.lowerdir) {
 994                 if (!silent)
 995                         pr_err("overlayfs: missing 'lowerdir'\n");
 996                 goto out_free_config;
 997         }
 998
 999         sb->s_stack_depth = 0;
1000         sb->s_maxbytes = MAX_LFS_FILESIZE;
1001         if (ufs->config.upperdir) {
1002                 if (!ufs->config.workdir) {
1003                         pr_err("overlayfs: missing 'workdir'\n");
1004                         goto out_free_config;
1005                 }
1006
1007                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
1008                 if (err)
1009                         goto out_free_config;
1010
1011                 /* Upper fs should not be r/o */
1012                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
1013                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
1014                         err = -EINVAL;
1015                         goto out_put_upperpath;
1016                 }
1017
1018                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
1019                 if (err)
1020                         goto out_put_upperpath;
1021
1022                 err = -EINVAL;
1023                 if (upperpath.mnt != workpath.mnt) {
1024                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1025                         goto out_put_workpath;
1026                 }
1027                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
1028                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1029                         goto out_put_workpath;
1030                 }
1031                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
1032         }
1033         err = -ENOMEM;
1034         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
1035         if (!lowertmp)
1036                 goto out_put_workpath;
1037
1038         err = -EINVAL;
1039         stacklen = ovl_split_lowerdirs(lowertmp);
1040         if (stacklen > OVL_MAX_STACK) {
1041                 pr_err("overlayfs: too many lower directries, limit is %d\n",
1042                        OVL_MAX_STACK);
1043                 goto out_free_lowertmp;
1044         } else if (!ufs->config.upperdir && stacklen == 1) {
1045                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1046                 goto out_free_lowertmp;
1047         }
1048
1049         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1050         if (!stack)
1051                 goto out_free_lowertmp;
1052
1053         lower = lowertmp;
1054         for (numlower = 0; numlower < stacklen; numlower++) {
1055                 err = ovl_lower_dir(lower, &stack[numlower],
1056                                     &ufs->lower_namelen, &sb->s_stack_depth,
1057                                     &remote);
1058                 if (err)
1059                         goto out_put_lowerpath;
1060
1061                 lower = strchr(lower, '\0') + 1;
1062         }
1063
1064         err = -EINVAL;
1065         sb->s_stack_depth++;
1066         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1067                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1068                 goto out_put_lowerpath;
1069         }
1070
1071         if (ufs->config.upperdir) {
1072                 ufs->upper_mnt = clone_private_mount(&upperpath);
1073                 err = PTR_ERR(ufs->upper_mnt);
1074                 if (IS_ERR(ufs->upper_mnt)) {
1075                         pr_err("overlayfs: failed to clone upperpath\n");
1076                         goto out_put_lowerpath;
1077                 }
1078                 /* Don't inherit atime flags */
1079                 ufs->upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1080
1081                 sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
1082
1083                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
1084                 err = PTR_ERR(ufs->workdir);
1085                 if (IS_ERR(ufs->workdir)) {
1086                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1087                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1088                         sb->s_flags |= MS_RDONLY;
1089                         ufs->workdir = NULL;
1090                 }
1091
1092                 /*
1093                  * Upper should support d_type, else whiteouts are visible.
1094                  * Given workdir and upper are on same fs, we can do
1095                  * iterate_dir() on workdir. This check requires successful
1096                  * creation of workdir in previous step.
1097                  */
1098                 if (ufs->workdir) {
1099                         err = ovl_check_d_type_supported(&workpath);
1100                         if (err < 0)
1101                                 goto out_put_workdir;
1102
1103                         /*
1104                          * We allowed this configuration and don't want to
1105                          * break users over kernel upgrade. So warn instead
1106                          * of erroring out.
1107                          */
1108                         if (!err)
1109                                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
1110                 }
1111         }
1112
1113         err = -ENOMEM;
1114         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1115         if (ufs->lower_mnt == NULL)
1116                 goto out_put_workdir;
1117         for (i = 0; i < numlower; i++) {
1118                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1119
1120                 err = PTR_ERR(mnt);
1121                 if (IS_ERR(mnt)) {
1122                         pr_err("overlayfs: failed to clone lowerpath\n");
1123                         goto out_put_lower_mnt;
1124                 }
1125                 /*
1126                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1127                  * will fail instead of modifying lower fs.
1128                  */
1129                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1130
1131                 ufs->lower_mnt[ufs->numlower] = mnt;
1132                 ufs->numlower++;
1133         }
1134
1135         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1136         if (!ufs->upper_mnt)
1137                 sb->s_flags |= MS_RDONLY;
1138
1139         if (remote)
1140                 sb->s_d_op = &ovl_reval_dentry_operations;
1141         else
1142                 sb->s_d_op = &ovl_dentry_operations;
1143
1144         ufs->creator_cred = prepare_creds();
1145         if (!ufs->creator_cred)
1146                 goto out_put_lower_mnt;
1147
1148         err = -ENOMEM;
1149         oe = ovl_alloc_entry(numlower);
1150         if (!oe)
1151                 goto out_put_cred;
1152
1153         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1154         if (!root_dentry)
1155                 goto out_free_oe;
1156
1157         mntput(upperpath.mnt);
1158         for (i = 0; i < numlower; i++)
1159                 mntput(stack[i].mnt);
1160         path_put(&workpath);
1161         kfree(lowertmp);
1162
1163         oe->__upperdentry = upperpath.dentry;
1164         for (i = 0; i < numlower; i++) {
1165                 oe->lowerstack[i].dentry = stack[i].dentry;
1166                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1167         }
1168         kfree(stack);
1169
1170         root_dentry->d_fsdata = oe;
1171
1172         ovl_copyattr(ovl_dentry_real(root_dentry)->d_inode,
1173                      root_dentry->d_inode);
1174
1175         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1176         sb->s_op = &ovl_super_operations;
1177         sb->s_root = root_dentry;
1178         sb->s_fs_info = ufs;
1179         sb->s_flags |= MS_POSIXACL;
1180
1181         return 0;
1182
1183 out_free_oe:
1184         kfree(oe);
1185 out_put_cred:
1186         put_cred(ufs->creator_cred);
1187 out_put_lower_mnt:
1188         for (i = 0; i < ufs->numlower; i++)
1189                 mntput(ufs->lower_mnt[i]);
1190         kfree(ufs->lower_mnt);
1191 out_put_workdir:
1192         dput(ufs->workdir);
1193         mntput(ufs->upper_mnt);
1194 out_put_lowerpath:
1195         for (i = 0; i < numlower; i++)
1196                 path_put(&stack[i]);
1197         kfree(stack);
1198 out_free_lowertmp:
1199         kfree(lowertmp);
1200 out_put_workpath:
1201         path_put(&workpath);
1202 out_put_upperpath:
1203         path_put(&upperpath);
1204 out_free_config:
1205         kfree(ufs->config.lowerdir);
1206         kfree(ufs->config.upperdir);
1207         kfree(ufs->config.workdir);
1208         kfree(ufs);
1209 out:
1210         return err;
1211 }
1212
1213 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1214                                 const char *dev_name, void *raw_data)
1215 {
1216         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1217 }
1218
1219 static struct file_system_type ovl_fs_type = {
1220         .owner          = THIS_MODULE,
1221         .name           = "overlay",
1222         .mount          = ovl_mount,
1223         .kill_sb        = kill_anon_super,
1224 };
1225 MODULE_ALIAS_FS("overlay");
1226
1227 static int __init ovl_init(void)
1228 {
1229         return register_filesystem(&ovl_fs_type);
1230 }
1231
1232 static void __exit ovl_exit(void)
1233 {
1234         unregister_filesystem(&ovl_fs_type);
1235 }
1236
1237 module_init(ovl_init);
1238 module_exit(ovl_exit);