fs/overlayfs/super.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/fs.h>
  11 #include <linux/namei.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/xattr.h>
  14 #include <linux/security.h>
  15 #include <linux/mount.h>
  16 #include <linux/slab.h>
  17 #include <linux/parser.h>
  18 #include <linux/module.h>
  19 #include <linux/sched.h>
  20 #include <linux/statfs.h>
  21 #include <linux/seq_file.h>
  22 #include <linux/posix_acl_xattr.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 static void ovl_inode_init(struct inode *inode, struct inode *realinode,
 149                            bool is_upper)
 150 {
 151         WRITE_ONCE(inode->i_private, (unsigned long) realinode |
 152                    (is_upper ? OVL_ISUPPER_MASK : 0));
 153 }
 154
 155 struct vfsmount *ovl_entry_mnt_real(struct ovl_entry *oe, struct inode *inode,
 156                                     bool is_upper)
 157 {
 158         if (is_upper) {
 159                 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 160
 161                 return ofs->upper_mnt;
 162         } else {
 163                 return oe->numlower ? oe->lowerstack[0].mnt : NULL;
 164         }
 165 }
 166
 167 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 168 {
 169         struct ovl_entry *oe = dentry->d_fsdata;
 170
 171         return oe->cache;
 172 }
 173
 174 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 175 {
 176         struct ovl_entry *oe = dentry->d_fsdata;
 177
 178         oe->cache = cache;
 179 }
 180
 181 void ovl_path_lower(struct dentry *dentry, struct path *path)
 182 {
 183         struct ovl_entry *oe = dentry->d_fsdata;
 184
 185         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 186 }
 187
 188 int ovl_want_write(struct dentry *dentry)
 189 {
 190         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 191         return mnt_want_write(ofs->upper_mnt);
 192 }
 193
 194 void ovl_drop_write(struct dentry *dentry)
 195 {
 196         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 197         mnt_drop_write(ofs->upper_mnt);
 198 }
 199
 200 struct dentry *ovl_workdir(struct dentry *dentry)
 201 {
 202         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 203         return ofs->workdir;
 204 }
 205
 206 bool ovl_dentry_is_opaque(struct dentry *dentry)
 207 {
 208         struct ovl_entry *oe = dentry->d_fsdata;
 209         return oe->opaque;
 210 }
 211
 212 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 213 {
 214         struct ovl_entry *oe = dentry->d_fsdata;
 215         oe->opaque = opaque;
 216 }
 217
 218 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 219 {
 220         struct ovl_entry *oe = dentry->d_fsdata;
 221
 222         WARN_ON(!inode_is_locked(upperdentry->d_parent->d_inode));
 223         WARN_ON(oe->__upperdentry);
 224         /*
 225          * Make sure upperdentry is consistent before making it visible to
 226          * ovl_upperdentry_dereference().
 227          */
 228         smp_wmb();
 229         oe->__upperdentry = upperdentry;
 230 }
 231
 232 void ovl_inode_update(struct inode *inode, struct inode *upperinode)
 233 {
 234         WARN_ON(!upperinode);
 235         WARN_ON(!inode_unhashed(inode));
 236         WRITE_ONCE(inode->i_private,
 237                    (unsigned long) upperinode | OVL_ISUPPER_MASK);
 238         if (!S_ISDIR(upperinode->i_mode))
 239                 __insert_inode_hash(inode, (unsigned long) upperinode);
 240 }
 241
 242 void ovl_dentry_version_inc(struct dentry *dentry)
 243 {
 244         struct ovl_entry *oe = dentry->d_fsdata;
 245
 246         WARN_ON(!inode_is_locked(dentry->d_inode));
 247         oe->version++;
 248 }
 249
 250 u64 ovl_dentry_version_get(struct dentry *dentry)
 251 {
 252         struct ovl_entry *oe = dentry->d_fsdata;
 253
 254         WARN_ON(!inode_is_locked(dentry->d_inode));
 255         return oe->version;
 256 }
 257
 258 bool ovl_is_whiteout(struct dentry *dentry)
 259 {
 260         struct inode *inode = dentry->d_inode;
 261
 262         return inode && IS_WHITEOUT(inode);
 263 }
 264
 265 const struct cred *ovl_override_creds(struct super_block *sb)
 266 {
 267         struct ovl_fs *ofs = sb->s_fs_info;
 268
 269         return override_creds(ofs->creator_cred);
 270 }
 271
 272 static bool ovl_is_opaquedir(struct dentry *dentry)
 273 {
 274         int res;
 275         char val;
 276
 277         if (!d_is_dir(dentry))
 278                 return false;
 279
 280         res = vfs_getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
 281         if (res == 1 && val == 'y')
 282                 return true;
 283
 284         return false;
 285 }
 286
 287 static void ovl_dentry_release(struct dentry *dentry)
 288 {
 289         struct ovl_entry *oe = dentry->d_fsdata;
 290
 291         if (oe) {
 292                 unsigned int i;
 293
 294                 dput(oe->__upperdentry);
 295                 for (i = 0; i < oe->numlower; i++)
 296                         dput(oe->lowerstack[i].dentry);
 297                 kfree_rcu(oe, rcu);
 298         }
 299 }
 300
 301 static struct dentry *ovl_d_real(struct dentry *dentry,
 302                                  const struct inode *inode,
 303                                  unsigned int open_flags)
 304 {
 305         struct dentry *real;
 306
 307         if (d_is_dir(dentry)) {
 308                 if (!inode || inode == d_inode(dentry))
 309                         return dentry;
 310                 goto bug;
 311         }
 312
 313         if (d_is_negative(dentry))
 314                 return dentry;
 315
 316         if (open_flags) {
 317                 int err = ovl_open_maybe_copy_up(dentry, open_flags);
 318
 319                 if (err)
 320                         return ERR_PTR(err);
 321         }
 322
 323         real = ovl_dentry_upper(dentry);
 324         if (real && (!inode || inode == d_inode(real)))
 325                 return real;
 326
 327         real = ovl_dentry_lower(dentry);
 328         if (!real)
 329                 goto bug;
 330
 331         if (!inode || inode == d_inode(real))
 332                 return real;
 333
 334         /* Handle recursion */
 335         return d_real(real, inode, open_flags);
 336 bug:
 337         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
 338              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 339         return dentry;
 340 }
 341
 342 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 343 {
 344         struct ovl_entry *oe = dentry->d_fsdata;
 345         unsigned int i;
 346         int ret = 1;
 347
 348         for (i = 0; i < oe->numlower; i++) {
 349                 struct dentry *d = oe->lowerstack[i].dentry;
 350
 351                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 352                         ret = d->d_op->d_revalidate(d, flags);
 353                         if (ret < 0)
 354                                 return ret;
 355                         if (!ret) {
 356                                 if (!(flags & LOOKUP_RCU))
 357                                         d_invalidate(d);
 358                                 return -ESTALE;
 359                         }
 360                 }
 361         }
 362         return 1;
 363 }
 364
 365 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 366 {
 367         struct ovl_entry *oe = dentry->d_fsdata;
 368         unsigned int i;
 369         int ret = 1;
 370
 371         for (i = 0; i < oe->numlower; i++) {
 372                 struct dentry *d = oe->lowerstack[i].dentry;
 373
 374                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 375                         ret = d->d_op->d_weak_revalidate(d, flags);
 376                         if (ret <= 0)
 377                                 break;
 378                 }
 379         }
 380         return ret;
 381 }
 382
 383 static const struct dentry_operations ovl_dentry_operations = {
 384         .d_release = ovl_dentry_release,
 385         .d_real = ovl_d_real,
 386 };
 387
 388 static const struct dentry_operations ovl_reval_dentry_operations = {
 389         .d_release = ovl_dentry_release,
 390         .d_real = ovl_d_real,
 391         .d_revalidate = ovl_dentry_revalidate,
 392         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 393 };
 394
 395 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 396 {
 397         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 398         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 399
 400         if (oe)
 401                 oe->numlower = numlower;
 402
 403         return oe;
 404 }
 405
 406 static bool ovl_dentry_remote(struct dentry *dentry)
 407 {
 408         return dentry->d_flags &
 409                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE |
 410                  DCACHE_OP_REAL);
 411 }
 412
 413 static bool ovl_dentry_weird(struct dentry *dentry)
 414 {
 415         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 416                                   DCACHE_MANAGE_TRANSIT |
 417                                   DCACHE_OP_HASH |
 418                                   DCACHE_OP_COMPARE);
 419 }
 420
 421 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 422                                              const struct qstr *name)
 423 {
 424         struct dentry *dentry;
 425
 426         dentry = lookup_one_len_unlocked(name->name, dir, name->len);
 427
 428         if (IS_ERR(dentry)) {
 429                 if (PTR_ERR(dentry) == -ENOENT)
 430                         dentry = NULL;
 431         } else if (!dentry->d_inode) {
 432                 dput(dentry);
 433                 dentry = NULL;
 434         } else if (ovl_dentry_weird(dentry)) {
 435                 dput(dentry);
 436                 /* Don't support traversing automounts and other weirdness */
 437                 dentry = ERR_PTR(-EREMOTE);
 438         }
 439         return dentry;
 440 }
 441
 442 /*
 443  * Returns next layer in stack starting from top.
 444  * Returns -1 if this is the last layer.
 445  */
 446 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 447 {
 448         struct ovl_entry *oe = dentry->d_fsdata;
 449
 450         BUG_ON(idx < 0);
 451         if (idx == 0) {
 452                 ovl_path_upper(dentry, path);
 453                 if (path->dentry)
 454                         return oe->numlower ? 1 : -1;
 455                 idx++;
 456         }
 457         BUG_ON(idx > oe->numlower);
 458         *path = oe->lowerstack[idx - 1];
 459
 460         return (idx < oe->numlower) ? idx + 1 : -1;
 461 }
 462
 463 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 464                           unsigned int flags)
 465 {
 466         struct ovl_entry *oe;
 467         const struct cred *old_cred;
 468         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 469         struct path *stack = NULL;
 470         struct dentry *upperdir, *upperdentry = NULL;
 471         unsigned int ctr = 0;
 472         struct inode *inode = NULL;
 473         bool upperopaque = false;
 474         struct dentry *this, *prev = NULL;
 475         unsigned int i;
 476         int err;
 477
 478         old_cred = ovl_override_creds(dentry->d_sb);
 479         upperdir = ovl_upperdentry_dereference(poe);
 480         if (upperdir) {
 481                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 482                 err = PTR_ERR(this);
 483                 if (IS_ERR(this))
 484                         goto out;
 485
 486                 if (this) {
 487                         if (unlikely(ovl_dentry_remote(this))) {
 488                                 dput(this);
 489                                 err = -EREMOTE;
 490                                 goto out;
 491                         }
 492                         if (ovl_is_whiteout(this)) {
 493                                 dput(this);
 494                                 this = NULL;
 495                                 upperopaque = true;
 496                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 497                                 upperopaque = true;
 498                         }
 499                 }
 500                 upperdentry = prev = this;
 501         }
 502
 503         if (!upperopaque && poe->numlower) {
 504                 err = -ENOMEM;
 505                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 506                 if (!stack)
 507                         goto out_put_upper;
 508         }
 509
 510         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 511                 bool opaque = false;
 512                 struct path lowerpath = poe->lowerstack[i];
 513
 514                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 515                 err = PTR_ERR(this);
 516                 if (IS_ERR(this)) {
 517                         /*
 518                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 519                          */
 520                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 521                                 continue;
 522                         goto out_put;
 523                 }
 524                 if (!this)
 525                         continue;
 526                 if (ovl_is_whiteout(this)) {
 527                         dput(this);
 528                         break;
 529                 }
 530                 /*
 531                  * Only makes sense to check opaque dir if this is not the
 532                  * lowermost layer.
 533                  */
 534                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 535                         opaque = true;
 536
 537                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 538                              !S_ISDIR(this->d_inode->i_mode))) {
 539                         /*
 540                          * FIXME: check for upper-opaqueness maybe better done
 541                          * in remove code.
 542                          */
 543                         if (prev == upperdentry)
 544                                 upperopaque = true;
 545                         dput(this);
 546                         break;
 547                 }
 548                 /*
 549                  * If this is a non-directory then stop here.
 550                  */
 551                 if (!S_ISDIR(this->d_inode->i_mode))
 552                         opaque = true;
 553
 554                 stack[ctr].dentry = this;
 555                 stack[ctr].mnt = lowerpath.mnt;
 556                 ctr++;
 557                 prev = this;
 558                 if (opaque)
 559                         break;
 560         }
 561
 562         oe = ovl_alloc_entry(ctr);
 563         err = -ENOMEM;
 564         if (!oe)
 565                 goto out_put;
 566
 567         if (upperdentry || ctr) {
 568                 struct dentry *realdentry;
 569                 struct inode *realinode;
 570
 571                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 572                 realinode = d_inode(realdentry);
 573
 574                 err = -ENOMEM;
 575                 if (upperdentry && !d_is_dir(upperdentry)) {
 576                         inode = ovl_get_inode(dentry->d_sb, realinode);
 577                 } else {
 578                         inode = ovl_new_inode(dentry->d_sb, realinode->i_mode);
 579                         if (inode)
 580                                 ovl_inode_init(inode, realinode, !!upperdentry);
 581                 }
 582                 if (!inode)
 583                         goto out_free_oe;
 584                 ovl_copyattr(realdentry->d_inode, inode);
 585         }
 586
 587         revert_creds(old_cred);
 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         revert_creds(old_cred);
 607         return ERR_PTR(err);
 608 }
 609
 610 struct file *ovl_path_open(struct path *path, int flags)
 611 {
 612         return dentry_open(path, flags | O_NOATIME, current_cred());
 613 }
 614
 615 static void ovl_put_super(struct super_block *sb)
 616 {
 617         struct ovl_fs *ufs = sb->s_fs_info;
 618         unsigned i;
 619
 620         dput(ufs->workdir);
 621         mntput(ufs->upper_mnt);
 622         for (i = 0; i < ufs->numlower; i++)
 623                 mntput(ufs->lower_mnt[i]);
 624         kfree(ufs->lower_mnt);
 625
 626         kfree(ufs->config.lowerdir);
 627         kfree(ufs->config.upperdir);
 628         kfree(ufs->config.workdir);
 629         put_cred(ufs->creator_cred);
 630         kfree(ufs);
 631 }
 632
 633 /**
 634  * ovl_statfs
 635  * @sb: The overlayfs super block
 636  * @buf: The struct kstatfs to fill in with stats
 637  *
 638  * Get the filesystem statistics.  As writes always target the upper layer
 639  * filesystem pass the statfs to the upper filesystem (if it exists)
 640  */
 641 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 642 {
 643         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 644         struct dentry *root_dentry = dentry->d_sb->s_root;
 645         struct path path;
 646         int err;
 647
 648         ovl_path_real(root_dentry, &path);
 649
 650         err = vfs_statfs(&path, buf);
 651         if (!err) {
 652                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 653                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 654         }
 655
 656         return err;
 657 }
 658
 659 /**
 660  * ovl_show_options
 661  *
 662  * Prints the mount options for a given superblock.
 663  * Returns zero; does not fail.
 664  */
 665 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 666 {
 667         struct super_block *sb = dentry->d_sb;
 668         struct ovl_fs *ufs = sb->s_fs_info;
 669
 670         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 671         if (ufs->config.upperdir) {
 672                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 673                 seq_show_option(m, "workdir", ufs->config.workdir);
 674         }
 675         if (ufs->config.default_permissions)
 676                 seq_puts(m, ",default_permissions");
 677         return 0;
 678 }
 679
 680 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 681 {
 682         struct ovl_fs *ufs = sb->s_fs_info;
 683
 684         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 685                 return -EROFS;
 686
 687         return 0;
 688 }
 689
 690 static const struct super_operations ovl_super_operations = {
 691         .put_super      = ovl_put_super,
 692         .statfs         = ovl_statfs,
 693         .show_options   = ovl_show_options,
 694         .remount_fs     = ovl_remount,
 695         .drop_inode     = generic_delete_inode,
 696 };
 697
 698 enum {
 699         OPT_LOWERDIR,
 700         OPT_UPPERDIR,
 701         OPT_WORKDIR,
 702         OPT_DEFAULT_PERMISSIONS,
 703         OPT_ERR,
 704 };
 705
 706 static const match_table_t ovl_tokens = {
 707         {OPT_LOWERDIR,                  "lowerdir=%s"},
 708         {OPT_UPPERDIR,                  "upperdir=%s"},
 709         {OPT_WORKDIR,                   "workdir=%s"},
 710         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 711         {OPT_ERR,                       NULL}
 712 };
 713
 714 static char *ovl_next_opt(char **s)
 715 {
 716         char *sbegin = *s;
 717         char *p;
 718
 719         if (sbegin == NULL)
 720                 return NULL;
 721
 722         for (p = sbegin; *p; p++) {
 723                 if (*p == '\\') {
 724                         p++;
 725                         if (!*p)
 726                                 break;
 727                 } else if (*p == ',') {
 728                         *p = '\0';
 729                         *s = p + 1;
 730                         return sbegin;
 731                 }
 732         }
 733         *s = NULL;
 734         return sbegin;
 735 }
 736
 737 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 738 {
 739         char *p;
 740
 741         while ((p = ovl_next_opt(&opt)) != NULL) {
 742                 int token;
 743                 substring_t args[MAX_OPT_ARGS];
 744
 745                 if (!*p)
 746                         continue;
 747
 748                 token = match_token(p, ovl_tokens, args);
 749                 switch (token) {
 750                 case OPT_UPPERDIR:
 751                         kfree(config->upperdir);
 752                         config->upperdir = match_strdup(&args[0]);
 753                         if (!config->upperdir)
 754                                 return -ENOMEM;
 755                         break;
 756
 757                 case OPT_LOWERDIR:
 758                         kfree(config->lowerdir);
 759                         config->lowerdir = match_strdup(&args[0]);
 760                         if (!config->lowerdir)
 761                                 return -ENOMEM;
 762                         break;
 763
 764                 case OPT_WORKDIR:
 765                         kfree(config->workdir);
 766                         config->workdir = match_strdup(&args[0]);
 767                         if (!config->workdir)
 768                                 return -ENOMEM;
 769                         break;
 770
 771                 case OPT_DEFAULT_PERMISSIONS:
 772                         config->default_permissions = true;
 773                         break;
 774
 775                 default:
 776                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 777                         return -EINVAL;
 778                 }
 779         }
 780
 781         /* Workdir is useless in non-upper mount */
 782         if (!config->upperdir && config->workdir) {
 783                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 784                         config->workdir);
 785                 kfree(config->workdir);
 786                 config->workdir = NULL;
 787         }
 788
 789         return 0;
 790 }
 791
 792 #define OVL_WORKDIR_NAME "work"
 793
 794 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 795                                          struct dentry *dentry)
 796 {
 797         struct inode *dir = dentry->d_inode;
 798         struct dentry *work;
 799         int err;
 800         bool retried = false;
 801
 802         err = mnt_want_write(mnt);
 803         if (err)
 804                 return ERR_PTR(err);
 805
 806         inode_lock_nested(dir, I_MUTEX_PARENT);
 807 retry:
 808         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 809                               strlen(OVL_WORKDIR_NAME));
 810
 811         if (!IS_ERR(work)) {
 812                 struct kstat stat = {
 813                         .mode = S_IFDIR | 0,
 814                 };
 815                 struct iattr attr = {
 816                         .ia_valid = ATTR_MODE,
 817                         .ia_mode = stat.mode,
 818                 };
 819
 820                 if (work->d_inode) {
 821                         err = -EEXIST;
 822                         if (retried)
 823                                 goto out_dput;
 824
 825                         retried = true;
 826                         ovl_workdir_cleanup(dir, mnt, work, 0);
 827                         dput(work);
 828                         goto retry;
 829                 }
 830
 831                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 832                 if (err)
 833                         goto out_dput;
 834
 835                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 836                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 837                         goto out_dput;
 838
 839                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 840                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 841                         goto out_dput;
 842
 843                 /* Clear any inherited mode bits */
 844                 inode_lock(work->d_inode);
 845                 err = notify_change(work, &attr, NULL);
 846                 inode_unlock(work->d_inode);
 847                 if (err)
 848                         goto out_dput;
 849         }
 850 out_unlock:
 851         inode_unlock(dir);
 852         mnt_drop_write(mnt);
 853
 854         return work;
 855
 856 out_dput:
 857         dput(work);
 858         work = ERR_PTR(err);
 859         goto out_unlock;
 860 }
 861
 862 static void ovl_unescape(char *s)
 863 {
 864         char *d = s;
 865
 866         for (;; s++, d++) {
 867                 if (*s == '\\')
 868                         s++;
 869                 *d = *s;
 870                 if (!*s)
 871                         break;
 872         }
 873 }
 874
 875 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 876 {
 877         int err = -EINVAL;
 878
 879         if (!*name) {
 880                 pr_err("overlayfs: empty lowerdir\n");
 881                 goto out;
 882         }
 883         err = kern_path(name, LOOKUP_FOLLOW, path);
 884         if (err) {
 885                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 886                 goto out;
 887         }
 888         err = -EINVAL;
 889         if (ovl_dentry_weird(path->dentry)) {
 890                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 891                 goto out_put;
 892         }
 893         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 894                 pr_err("overlayfs: '%s' not a directory\n", name);
 895                 goto out_put;
 896         }
 897         return 0;
 898
 899 out_put:
 900         path_put(path);
 901 out:
 902         return err;
 903 }
 904
 905 static int ovl_mount_dir(const char *name, struct path *path)
 906 {
 907         int err = -ENOMEM;
 908         char *tmp = kstrdup(name, GFP_KERNEL);
 909
 910         if (tmp) {
 911                 ovl_unescape(tmp);
 912                 err = ovl_mount_dir_noesc(tmp, path);
 913
 914                 if (!err)
 915                         if (ovl_dentry_remote(path->dentry)) {
 916                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 917                                        tmp);
 918                                 path_put(path);
 919                                 err = -EINVAL;
 920                         }
 921                 kfree(tmp);
 922         }
 923         return err;
 924 }
 925
 926 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 927                          int *stack_depth, bool *remote)
 928 {
 929         int err;
 930         struct kstatfs statfs;
 931
 932         err = ovl_mount_dir_noesc(name, path);
 933         if (err)
 934                 goto out;
 935
 936         err = vfs_statfs(path, &statfs);
 937         if (err) {
 938                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 939                 goto out_put;
 940         }
 941         *namelen = max(*namelen, statfs.f_namelen);
 942         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 943
 944         if (ovl_dentry_remote(path->dentry))
 945                 *remote = true;
 946
 947         return 0;
 948
 949 out_put:
 950         path_put(path);
 951 out:
 952         return err;
 953 }
 954
 955 /* Workdir should not be subdir of upperdir and vice versa */
 956 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 957 {
 958         bool ok = false;
 959
 960         if (workdir != upperdir) {
 961                 ok = (lock_rename(workdir, upperdir) == NULL);
 962                 unlock_rename(workdir, upperdir);
 963         }
 964         return ok;
 965 }
 966
 967 static unsigned int ovl_split_lowerdirs(char *str)
 968 {
 969         unsigned int ctr = 1;
 970         char *s, *d;
 971
 972         for (s = d = str;; s++, d++) {
 973                 if (*s == '\\') {
 974                         s++;
 975                 } else if (*s == ':') {
 976                         *d = '\0';
 977                         ctr++;
 978                         continue;
 979                 }
 980                 *d = *s;
 981                 if (!*s)
 982                         break;
 983         }
 984         return ctr;
 985 }
 986
 987 static int __maybe_unused
 988 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 989                         struct dentry *dentry, struct inode *inode,
 990                         const char *name, void *buffer, size_t size)
 991 {
 992         return ovl_xattr_get(dentry, handler->name, buffer, size);
 993 }
 994
 995 static int __maybe_unused
 996 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 997                         struct dentry *dentry, struct inode *inode,
 998                         const char *name, const void *value,
 999                         size_t size, int flags)
1000 {
1001         struct dentry *workdir = ovl_workdir(dentry);
1002         struct inode *realinode = ovl_inode_real(inode, NULL);
1003         struct posix_acl *acl = NULL;
1004         int err;
1005
1006         /* Check that everything is OK before copy-up */
1007         if (value) {
1008                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
1009                 if (IS_ERR(acl))
1010                         return PTR_ERR(acl);
1011         }
1012         err = -EOPNOTSUPP;
1013         if (!IS_POSIXACL(d_inode(workdir)))
1014                 goto out_acl_release;
1015         if (!realinode->i_op->set_acl)
1016                 goto out_acl_release;
1017         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1018                 err = acl ? -EACCES : 0;
1019                 goto out_acl_release;
1020         }
1021         err = -EPERM;
1022         if (!inode_owner_or_capable(inode))
1023                 goto out_acl_release;
1024
1025         posix_acl_release(acl);
1026
1027         err = ovl_xattr_set(dentry, handler->name, value, size, flags);
1028         if (!err)
1029                 ovl_copyattr(ovl_inode_real(inode, NULL), inode);
1030
1031         return err;
1032
1033 out_acl_release:
1034         posix_acl_release(acl);
1035         return err;
1036 }
1037
1038 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1039                              struct dentry *dentry, struct inode *inode,
1040                              const char *name, void *buffer, size_t size)
1041 {
1042         return -EPERM;
1043 }
1044
1045 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1046                              struct dentry *dentry, struct inode *inode,
1047                              const char *name, const void *value,
1048                              size_t size, int flags)
1049 {
1050         return -EPERM;
1051 }
1052
1053 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1054                                struct dentry *dentry, struct inode *inode,
1055                                const char *name, void *buffer, size_t size)
1056 {
1057         return ovl_xattr_get(dentry, name, buffer, size);
1058 }
1059
1060 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1061                                struct dentry *dentry, struct inode *inode,
1062                                const char *name, const void *value,
1063                                size_t size, int flags)
1064 {
1065         return ovl_xattr_set(dentry, name, value, size, flags);
1066 }
1067
1068 static const struct xattr_handler __maybe_unused
1069 ovl_posix_acl_access_xattr_handler = {
1070         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1071         .flags = ACL_TYPE_ACCESS,
1072         .get = ovl_posix_acl_xattr_get,
1073         .set = ovl_posix_acl_xattr_set,
1074 };
1075
1076 static const struct xattr_handler __maybe_unused
1077 ovl_posix_acl_default_xattr_handler = {
1078         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1079         .flags = ACL_TYPE_DEFAULT,
1080         .get = ovl_posix_acl_xattr_get,
1081         .set = ovl_posix_acl_xattr_set,
1082 };
1083
1084 static const struct xattr_handler ovl_own_xattr_handler = {
1085         .prefix = OVL_XATTR_PREFIX,
1086         .get = ovl_own_xattr_get,
1087         .set = ovl_own_xattr_set,
1088 };
1089
1090 static const struct xattr_handler ovl_other_xattr_handler = {
1091         .prefix = "", /* catch all */
1092         .get = ovl_other_xattr_get,
1093         .set = ovl_other_xattr_set,
1094 };
1095
1096 static const struct xattr_handler *ovl_xattr_handlers[] = {
1097 #ifdef CONFIG_FS_POSIX_ACL
1098         &ovl_posix_acl_access_xattr_handler,
1099         &ovl_posix_acl_default_xattr_handler,
1100 #endif
1101         &ovl_own_xattr_handler,
1102         &ovl_other_xattr_handler,
1103         NULL
1104 };
1105
1106 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1107 {
1108         struct path upperpath = { NULL, NULL };
1109         struct path workpath = { NULL, NULL };
1110         struct dentry *root_dentry;
1111         struct inode *realinode;
1112         struct ovl_entry *oe;
1113         struct ovl_fs *ufs;
1114         struct path *stack = NULL;
1115         char *lowertmp;
1116         char *lower;
1117         unsigned int numlower;
1118         unsigned int stacklen = 0;
1119         unsigned int i;
1120         bool remote = false;
1121         int err;
1122
1123         err = -ENOMEM;
1124         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1125         if (!ufs)
1126                 goto out;
1127
1128         err = ovl_parse_opt((char *) data, &ufs->config);
1129         if (err)
1130                 goto out_free_config;
1131
1132         err = -EINVAL;
1133         if (!ufs->config.lowerdir) {
1134                 if (!silent)
1135                         pr_err("overlayfs: missing 'lowerdir'\n");
1136                 goto out_free_config;
1137         }
1138
1139         sb->s_stack_depth = 0;
1140         sb->s_maxbytes = MAX_LFS_FILESIZE;
1141         if (ufs->config.upperdir) {
1142                 if (!ufs->config.workdir) {
1143                         pr_err("overlayfs: missing 'workdir'\n");
1144                         goto out_free_config;
1145                 }
1146
1147                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
1148                 if (err)
1149                         goto out_free_config;
1150
1151                 /* Upper fs should not be r/o */
1152                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
1153                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
1154                         err = -EINVAL;
1155                         goto out_put_upperpath;
1156                 }
1157
1158                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
1159                 if (err)
1160                         goto out_put_upperpath;
1161
1162                 err = -EINVAL;
1163                 if (upperpath.mnt != workpath.mnt) {
1164                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1165                         goto out_put_workpath;
1166                 }
1167                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
1168                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1169                         goto out_put_workpath;
1170                 }
1171                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
1172         }
1173         err = -ENOMEM;
1174         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
1175         if (!lowertmp)
1176                 goto out_put_workpath;
1177
1178         err = -EINVAL;
1179         stacklen = ovl_split_lowerdirs(lowertmp);
1180         if (stacklen > OVL_MAX_STACK) {
1181                 pr_err("overlayfs: too many lower directories, limit is %d\n",
1182                        OVL_MAX_STACK);
1183                 goto out_free_lowertmp;
1184         } else if (!ufs->config.upperdir && stacklen == 1) {
1185                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1186                 goto out_free_lowertmp;
1187         }
1188
1189         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1190         if (!stack)
1191                 goto out_free_lowertmp;
1192
1193         lower = lowertmp;
1194         for (numlower = 0; numlower < stacklen; numlower++) {
1195                 err = ovl_lower_dir(lower, &stack[numlower],
1196                                     &ufs->lower_namelen, &sb->s_stack_depth,
1197                                     &remote);
1198                 if (err)
1199                         goto out_put_lowerpath;
1200
1201                 lower = strchr(lower, '\0') + 1;
1202         }
1203
1204         err = -EINVAL;
1205         sb->s_stack_depth++;
1206         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1207                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1208                 goto out_put_lowerpath;
1209         }
1210
1211         if (ufs->config.upperdir) {
1212                 ufs->upper_mnt = clone_private_mount(&upperpath);
1213                 err = PTR_ERR(ufs->upper_mnt);
1214                 if (IS_ERR(ufs->upper_mnt)) {
1215                         pr_err("overlayfs: failed to clone upperpath\n");
1216                         goto out_put_lowerpath;
1217                 }
1218                 /* Don't inherit atime flags */
1219                 ufs->upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1220
1221                 sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
1222
1223                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
1224                 err = PTR_ERR(ufs->workdir);
1225                 if (IS_ERR(ufs->workdir)) {
1226                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1227                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1228                         sb->s_flags |= MS_RDONLY;
1229                         ufs->workdir = NULL;
1230                 }
1231
1232                 /*
1233                  * Upper should support d_type, else whiteouts are visible.
1234                  * Given workdir and upper are on same fs, we can do
1235                  * iterate_dir() on workdir. This check requires successful
1236                  * creation of workdir in previous step.
1237                  */
1238                 if (ufs->workdir) {
1239                         err = ovl_check_d_type_supported(&workpath);
1240                         if (err < 0)
1241                                 goto out_put_workdir;
1242
1243                         /*
1244                          * We allowed this configuration and don't want to
1245                          * break users over kernel upgrade. So warn instead
1246                          * of erroring out.
1247                          */
1248                         if (!err)
1249                                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
1250                 }
1251         }
1252
1253         err = -ENOMEM;
1254         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1255         if (ufs->lower_mnt == NULL)
1256                 goto out_put_workdir;
1257         for (i = 0; i < numlower; i++) {
1258                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1259
1260                 err = PTR_ERR(mnt);
1261                 if (IS_ERR(mnt)) {
1262                         pr_err("overlayfs: failed to clone lowerpath\n");
1263                         goto out_put_lower_mnt;
1264                 }
1265                 /*
1266                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1267                  * will fail instead of modifying lower fs.
1268                  */
1269                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1270
1271                 ufs->lower_mnt[ufs->numlower] = mnt;
1272                 ufs->numlower++;
1273         }
1274
1275         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1276         if (!ufs->upper_mnt)
1277                 sb->s_flags |= MS_RDONLY;
1278
1279         if (remote)
1280                 sb->s_d_op = &ovl_reval_dentry_operations;
1281         else
1282                 sb->s_d_op = &ovl_dentry_operations;
1283
1284         ufs->creator_cred = prepare_creds();
1285         if (!ufs->creator_cred)
1286                 goto out_put_lower_mnt;
1287
1288         err = -ENOMEM;
1289         oe = ovl_alloc_entry(numlower);
1290         if (!oe)
1291                 goto out_put_cred;
1292
1293         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR));
1294         if (!root_dentry)
1295                 goto out_free_oe;
1296
1297         mntput(upperpath.mnt);
1298         for (i = 0; i < numlower; i++)
1299                 mntput(stack[i].mnt);
1300         path_put(&workpath);
1301         kfree(lowertmp);
1302
1303         oe->__upperdentry = upperpath.dentry;
1304         for (i = 0; i < numlower; i++) {
1305                 oe->lowerstack[i].dentry = stack[i].dentry;
1306                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1307         }
1308         kfree(stack);
1309
1310         root_dentry->d_fsdata = oe;
1311
1312         realinode = d_inode(ovl_dentry_real(root_dentry));
1313         ovl_inode_init(d_inode(root_dentry), realinode, !!upperpath.dentry);
1314         ovl_copyattr(realinode, d_inode(root_dentry));
1315
1316         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1317         sb->s_op = &ovl_super_operations;
1318         sb->s_xattr = ovl_xattr_handlers;
1319         sb->s_root = root_dentry;
1320         sb->s_fs_info = ufs;
1321         sb->s_flags |= MS_POSIXACL;
1322
1323         return 0;
1324
1325 out_free_oe:
1326         kfree(oe);
1327 out_put_cred:
1328         put_cred(ufs->creator_cred);
1329 out_put_lower_mnt:
1330         for (i = 0; i < ufs->numlower; i++)
1331                 mntput(ufs->lower_mnt[i]);
1332         kfree(ufs->lower_mnt);
1333 out_put_workdir:
1334         dput(ufs->workdir);
1335         mntput(ufs->upper_mnt);
1336 out_put_lowerpath:
1337         for (i = 0; i < numlower; i++)
1338                 path_put(&stack[i]);
1339         kfree(stack);
1340 out_free_lowertmp:
1341         kfree(lowertmp);
1342 out_put_workpath:
1343         path_put(&workpath);
1344 out_put_upperpath:
1345         path_put(&upperpath);
1346 out_free_config:
1347         kfree(ufs->config.lowerdir);
1348         kfree(ufs->config.upperdir);
1349         kfree(ufs->config.workdir);
1350         kfree(ufs);
1351 out:
1352         return err;
1353 }
1354
1355 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1356                                 const char *dev_name, void *raw_data)
1357 {
1358         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1359 }
1360
1361 static struct file_system_type ovl_fs_type = {
1362         .owner          = THIS_MODULE,
1363         .name           = "overlay",
1364         .mount          = ovl_mount,
1365         .kill_sb        = kill_anon_super,
1366 };
1367 MODULE_ALIAS_FS("overlay");
1368
1369 static int __init ovl_init(void)
1370 {
1371         return register_filesystem(&ovl_fs_type);
1372 }
1373
1374 static void __exit ovl_exit(void)
1375 {
1376         unregister_filesystem(&ovl_fs_type);
1377 }
1378
1379 module_init(ovl_init);
1380 module_exit(ovl_exit);