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 <uapi/linux/magic.h>
  11 #include <linux/fs.h>
  12 #include <linux/namei.h>
  13 #include <linux/xattr.h>
  14 #include <linux/mount.h>
  15 #include <linux/parser.h>
  16 #include <linux/module.h>
  17 #include <linux/statfs.h>
  18 #include <linux/seq_file.h>
  19 #include <linux/posix_acl_xattr.h>
  20 #include "overlayfs.h"
  21 #include "ovl_entry.h"
  22
  23 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  24 MODULE_DESCRIPTION("Overlay filesystem");
  25 MODULE_LICENSE("GPL");
  26
  27
  28 struct ovl_dir_cache;
  29
  30 #define OVL_MAX_STACK 500
  31
  32 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  33 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  34 MODULE_PARM_DESC(ovl_redirect_dir_def,
  35                  "Default to on or off for the redirect_dir feature");
  36
  37 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
  38 module_param_named(index, ovl_index_def, bool, 0644);
  39 MODULE_PARM_DESC(ovl_index_def,
  40                  "Default to on or off for the inodes index feature");
  41
  42 static void ovl_dentry_release(struct dentry *dentry)
  43 {
  44         struct ovl_entry *oe = dentry->d_fsdata;
  45
  46         if (oe) {
  47                 unsigned int i;
  48
  49                 for (i = 0; i < oe->numlower; i++)
  50                         dput(oe->lowerstack[i].dentry);
  51                 kfree_rcu(oe, rcu);
  52         }
  53 }
  54
  55 static int ovl_check_append_only(struct inode *inode, int flag)
  56 {
  57         /*
  58          * This test was moot in vfs may_open() because overlay inode does
  59          * not have the S_APPEND flag, so re-check on real upper inode
  60          */
  61         if (IS_APPEND(inode)) {
  62                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
  63                         return -EPERM;
  64                 if (flag & O_TRUNC)
  65                         return -EPERM;
  66         }
  67
  68         return 0;
  69 }
  70
  71 static struct dentry *ovl_d_real(struct dentry *dentry,
  72                                  const struct inode *inode,
  73                                  unsigned int open_flags, unsigned int flags)
  74 {
  75         struct dentry *real;
  76         int err;
  77
  78         if (flags & D_REAL_UPPER)
  79                 return ovl_dentry_upper(dentry);
  80
  81         if (!d_is_reg(dentry)) {
  82                 if (!inode || inode == d_inode(dentry))
  83                         return dentry;
  84                 goto bug;
  85         }
  86
  87         if (open_flags) {
  88                 err = ovl_open_maybe_copy_up(dentry, open_flags);
  89                 if (err)
  90                         return ERR_PTR(err);
  91         }
  92
  93         real = ovl_dentry_upper(dentry);
  94         if (real && (!inode || inode == d_inode(real))) {
  95                 if (!inode) {
  96                         err = ovl_check_append_only(d_inode(real), open_flags);
  97                         if (err)
  98                                 return ERR_PTR(err);
  99                 }
 100                 return real;
 101         }
 102
 103         real = ovl_dentry_lower(dentry);
 104         if (!real)
 105                 goto bug;
 106
 107         /* Handle recursion */
 108         real = d_real(real, inode, open_flags, 0);
 109
 110         if (!inode || inode == d_inode(real))
 111                 return real;
 112 bug:
 113         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
 114              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 115         return dentry;
 116 }
 117
 118 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 119 {
 120         struct ovl_entry *oe = dentry->d_fsdata;
 121         unsigned int i;
 122         int ret = 1;
 123
 124         for (i = 0; i < oe->numlower; i++) {
 125                 struct dentry *d = oe->lowerstack[i].dentry;
 126
 127                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 128                         ret = d->d_op->d_revalidate(d, flags);
 129                         if (ret < 0)
 130                                 return ret;
 131                         if (!ret) {
 132                                 if (!(flags & LOOKUP_RCU))
 133                                         d_invalidate(d);
 134                                 return -ESTALE;
 135                         }
 136                 }
 137         }
 138         return 1;
 139 }
 140
 141 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 142 {
 143         struct ovl_entry *oe = dentry->d_fsdata;
 144         unsigned int i;
 145         int ret = 1;
 146
 147         for (i = 0; i < oe->numlower; i++) {
 148                 struct dentry *d = oe->lowerstack[i].dentry;
 149
 150                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 151                         ret = d->d_op->d_weak_revalidate(d, flags);
 152                         if (ret <= 0)
 153                                 break;
 154                 }
 155         }
 156         return ret;
 157 }
 158
 159 static const struct dentry_operations ovl_dentry_operations = {
 160         .d_release = ovl_dentry_release,
 161         .d_real = ovl_d_real,
 162 };
 163
 164 static const struct dentry_operations ovl_reval_dentry_operations = {
 165         .d_release = ovl_dentry_release,
 166         .d_real = ovl_d_real,
 167         .d_revalidate = ovl_dentry_revalidate,
 168         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 169 };
 170
 171 static struct kmem_cache *ovl_inode_cachep;
 172
 173 static struct inode *ovl_alloc_inode(struct super_block *sb)
 174 {
 175         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 176
 177         if (!oi)
 178                 return NULL;
 179
 180         oi->cache = NULL;
 181         oi->redirect = NULL;
 182         oi->version = 0;
 183         oi->flags = 0;
 184         oi->__upperdentry = NULL;
 185         oi->lower = NULL;
 186         mutex_init(&oi->lock);
 187
 188         return &oi->vfs_inode;
 189 }
 190
 191 static void ovl_i_callback(struct rcu_head *head)
 192 {
 193         struct inode *inode = container_of(head, struct inode, i_rcu);
 194
 195         kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
 196 }
 197
 198 static void ovl_destroy_inode(struct inode *inode)
 199 {
 200         struct ovl_inode *oi = OVL_I(inode);
 201
 202         dput(oi->__upperdentry);
 203         iput(oi->lower);
 204         kfree(oi->redirect);
 205         ovl_dir_cache_free(inode);
 206         mutex_destroy(&oi->lock);
 207
 208         call_rcu(&inode->i_rcu, ovl_i_callback);
 209 }
 210
 211 static void ovl_put_super(struct super_block *sb)
 212 {
 213         struct ovl_fs *ufs = sb->s_fs_info;
 214         unsigned i;
 215
 216         dput(ufs->indexdir);
 217         dput(ufs->workdir);
 218         if (ufs->workdir_locked)
 219                 ovl_inuse_unlock(ufs->workbasedir);
 220         dput(ufs->workbasedir);
 221         if (ufs->upper_mnt && ufs->upperdir_locked)
 222                 ovl_inuse_unlock(ufs->upper_mnt->mnt_root);
 223         mntput(ufs->upper_mnt);
 224         for (i = 0; i < ufs->numlower; i++)
 225                 mntput(ufs->lower_mnt[i]);
 226         kfree(ufs->lower_mnt);
 227
 228         kfree(ufs->config.lowerdir);
 229         kfree(ufs->config.upperdir);
 230         kfree(ufs->config.workdir);
 231         put_cred(ufs->creator_cred);
 232         kfree(ufs);
 233 }
 234
 235 /* Sync real dirty inodes in upper filesystem (if it exists) */
 236 static int ovl_sync_fs(struct super_block *sb, int wait)
 237 {
 238         struct ovl_fs *ufs = sb->s_fs_info;
 239         struct super_block *upper_sb;
 240         int ret;
 241
 242         if (!ufs->upper_mnt)
 243                 return 0;
 244
 245         /*
 246          * If this is a sync(2) call or an emergency sync, all the super blocks
 247          * will be iterated, including upper_sb, so no need to do anything.
 248          *
 249          * If this is a syncfs(2) call, then we do need to call
 250          * sync_filesystem() on upper_sb, but enough if we do it when being
 251          * called with wait == 1.
 252          */
 253         if (!wait)
 254                 return 0;
 255
 256         upper_sb = ufs->upper_mnt->mnt_sb;
 257
 258         down_read(&upper_sb->s_umount);
 259         ret = sync_filesystem(upper_sb);
 260         up_read(&upper_sb->s_umount);
 261
 262         return ret;
 263 }
 264
 265 /**
 266  * ovl_statfs
 267  * @sb: The overlayfs super block
 268  * @buf: The struct kstatfs to fill in with stats
 269  *
 270  * Get the filesystem statistics.  As writes always target the upper layer
 271  * filesystem pass the statfs to the upper filesystem (if it exists)
 272  */
 273 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 274 {
 275         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 276         struct dentry *root_dentry = dentry->d_sb->s_root;
 277         struct path path;
 278         int err;
 279
 280         ovl_path_real(root_dentry, &path);
 281
 282         err = vfs_statfs(&path, buf);
 283         if (!err) {
 284                 buf->f_namelen = ofs->namelen;
 285                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 286         }
 287
 288         return err;
 289 }
 290
 291 /* Will this overlay be forced to mount/remount ro? */
 292 static bool ovl_force_readonly(struct ovl_fs *ufs)
 293 {
 294         return (!ufs->upper_mnt || !ufs->workdir);
 295 }
 296
 297 /**
 298  * ovl_show_options
 299  *
 300  * Prints the mount options for a given superblock.
 301  * Returns zero; does not fail.
 302  */
 303 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 304 {
 305         struct super_block *sb = dentry->d_sb;
 306         struct ovl_fs *ufs = sb->s_fs_info;
 307
 308         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 309         if (ufs->config.upperdir) {
 310                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 311                 seq_show_option(m, "workdir", ufs->config.workdir);
 312         }
 313         if (ufs->config.default_permissions)
 314                 seq_puts(m, ",default_permissions");
 315         if (ufs->config.redirect_dir != ovl_redirect_dir_def)
 316                 seq_printf(m, ",redirect_dir=%s",
 317                            ufs->config.redirect_dir ? "on" : "off");
 318         if (ufs->config.index != ovl_index_def)
 319                 seq_printf(m, ",index=%s",
 320                            ufs->config.index ? "on" : "off");
 321         return 0;
 322 }
 323
 324 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 325 {
 326         struct ovl_fs *ufs = sb->s_fs_info;
 327
 328         if (!(*flags & MS_RDONLY) && ovl_force_readonly(ufs))
 329                 return -EROFS;
 330
 331         return 0;
 332 }
 333
 334 static const struct super_operations ovl_super_operations = {
 335         .alloc_inode    = ovl_alloc_inode,
 336         .destroy_inode  = ovl_destroy_inode,
 337         .drop_inode     = generic_delete_inode,
 338         .put_super      = ovl_put_super,
 339         .sync_fs        = ovl_sync_fs,
 340         .statfs         = ovl_statfs,
 341         .show_options   = ovl_show_options,
 342         .remount_fs     = ovl_remount,
 343 };
 344
 345 enum {
 346         OPT_LOWERDIR,
 347         OPT_UPPERDIR,
 348         OPT_WORKDIR,
 349         OPT_DEFAULT_PERMISSIONS,
 350         OPT_REDIRECT_DIR_ON,
 351         OPT_REDIRECT_DIR_OFF,
 352         OPT_INDEX_ON,
 353         OPT_INDEX_OFF,
 354         OPT_ERR,
 355 };
 356
 357 static const match_table_t ovl_tokens = {
 358         {OPT_LOWERDIR,                  "lowerdir=%s"},
 359         {OPT_UPPERDIR,                  "upperdir=%s"},
 360         {OPT_WORKDIR,                   "workdir=%s"},
 361         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 362         {OPT_REDIRECT_DIR_ON,           "redirect_dir=on"},
 363         {OPT_REDIRECT_DIR_OFF,          "redirect_dir=off"},
 364         {OPT_INDEX_ON,                  "index=on"},
 365         {OPT_INDEX_OFF,                 "index=off"},
 366         {OPT_ERR,                       NULL}
 367 };
 368
 369 static char *ovl_next_opt(char **s)
 370 {
 371         char *sbegin = *s;
 372         char *p;
 373
 374         if (sbegin == NULL)
 375                 return NULL;
 376
 377         for (p = sbegin; *p; p++) {
 378                 if (*p == '\\') {
 379                         p++;
 380                         if (!*p)
 381                                 break;
 382                 } else if (*p == ',') {
 383                         *p = '\0';
 384                         *s = p + 1;
 385                         return sbegin;
 386                 }
 387         }
 388         *s = NULL;
 389         return sbegin;
 390 }
 391
 392 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 393 {
 394         char *p;
 395
 396         while ((p = ovl_next_opt(&opt)) != NULL) {
 397                 int token;
 398                 substring_t args[MAX_OPT_ARGS];
 399
 400                 if (!*p)
 401                         continue;
 402
 403                 token = match_token(p, ovl_tokens, args);
 404                 switch (token) {
 405                 case OPT_UPPERDIR:
 406                         kfree(config->upperdir);
 407                         config->upperdir = match_strdup(&args[0]);
 408                         if (!config->upperdir)
 409                                 return -ENOMEM;
 410                         break;
 411
 412                 case OPT_LOWERDIR:
 413                         kfree(config->lowerdir);
 414                         config->lowerdir = match_strdup(&args[0]);
 415                         if (!config->lowerdir)
 416                                 return -ENOMEM;
 417                         break;
 418
 419                 case OPT_WORKDIR:
 420                         kfree(config->workdir);
 421                         config->workdir = match_strdup(&args[0]);
 422                         if (!config->workdir)
 423                                 return -ENOMEM;
 424                         break;
 425
 426                 case OPT_DEFAULT_PERMISSIONS:
 427                         config->default_permissions = true;
 428                         break;
 429
 430                 case OPT_REDIRECT_DIR_ON:
 431                         config->redirect_dir = true;
 432                         break;
 433
 434                 case OPT_REDIRECT_DIR_OFF:
 435                         config->redirect_dir = false;
 436                         break;
 437
 438                 case OPT_INDEX_ON:
 439                         config->index = true;
 440                         break;
 441
 442                 case OPT_INDEX_OFF:
 443                         config->index = false;
 444                         break;
 445
 446                 default:
 447                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 448                         return -EINVAL;
 449                 }
 450         }
 451
 452         /* Workdir is useless in non-upper mount */
 453         if (!config->upperdir && config->workdir) {
 454                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 455                         config->workdir);
 456                 kfree(config->workdir);
 457                 config->workdir = NULL;
 458         }
 459
 460         return 0;
 461 }
 462
 463 #define OVL_WORKDIR_NAME "work"
 464 #define OVL_INDEXDIR_NAME "index"
 465
 466 static struct dentry *ovl_workdir_create(struct super_block *sb,
 467                                          struct ovl_fs *ufs,
 468                                          struct dentry *dentry,
 469                                          const char *name, bool persist)
 470 {
 471         struct inode *dir = dentry->d_inode;
 472         struct vfsmount *mnt = ufs->upper_mnt;
 473         struct dentry *work;
 474         int err;
 475         bool retried = false;
 476         bool locked = false;
 477
 478         err = mnt_want_write(mnt);
 479         if (err)
 480                 goto out_err;
 481
 482         inode_lock_nested(dir, I_MUTEX_PARENT);
 483         locked = true;
 484
 485 retry:
 486         work = lookup_one_len(name, dentry, strlen(name));
 487
 488         if (!IS_ERR(work)) {
 489                 struct iattr attr = {
 490                         .ia_valid = ATTR_MODE,
 491                         .ia_mode = S_IFDIR | 0,
 492                 };
 493
 494                 if (work->d_inode) {
 495                         err = -EEXIST;
 496                         if (retried)
 497                                 goto out_dput;
 498
 499                         if (persist)
 500                                 goto out_unlock;
 501
 502                         retried = true;
 503                         ovl_workdir_cleanup(dir, mnt, work, 0);
 504                         dput(work);
 505                         goto retry;
 506                 }
 507
 508                 err = ovl_create_real(dir, work,
 509                                       &(struct cattr){.mode = S_IFDIR | 0},
 510                                       NULL, true);
 511                 if (err)
 512                         goto out_dput;
 513
 514                 /*
 515                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 516                  *
 517                  * a) success (there was a POSIX ACL xattr and was removed)
 518                  * b) -ENODATA (there was no POSIX ACL xattr)
 519                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 520                  *
 521                  * There are various other error values that could effectively
 522                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 523                  * if the xattr name is too long), but the set of filesystems
 524                  * allowed as upper are limited to "normal" ones, where checking
 525                  * for the above two errors is sufficient.
 526                  */
 527                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 528                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 529                         goto out_dput;
 530
 531                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 532                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 533                         goto out_dput;
 534
 535                 /* Clear any inherited mode bits */
 536                 inode_lock(work->d_inode);
 537                 err = notify_change(work, &attr, NULL);
 538                 inode_unlock(work->d_inode);
 539                 if (err)
 540                         goto out_dput;
 541         } else {
 542                 err = PTR_ERR(work);
 543                 goto out_err;
 544         }
 545 out_unlock:
 546         mnt_drop_write(mnt);
 547         if (locked)
 548                 inode_unlock(dir);
 549
 550         return work;
 551
 552 out_dput:
 553         dput(work);
 554 out_err:
 555         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 556                 ufs->config.workdir, name, -err);
 557         sb->s_flags |= MS_RDONLY;
 558         work = NULL;
 559         goto out_unlock;
 560 }
 561
 562 static void ovl_unescape(char *s)
 563 {
 564         char *d = s;
 565
 566         for (;; s++, d++) {
 567                 if (*s == '\\')
 568                         s++;
 569                 *d = *s;
 570                 if (!*s)
 571                         break;
 572         }
 573 }
 574
 575 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 576 {
 577         int err = -EINVAL;
 578
 579         if (!*name) {
 580                 pr_err("overlayfs: empty lowerdir\n");
 581                 goto out;
 582         }
 583         err = kern_path(name, LOOKUP_FOLLOW, path);
 584         if (err) {
 585                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 586                 goto out;
 587         }
 588         err = -EINVAL;
 589         if (ovl_dentry_weird(path->dentry)) {
 590                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 591                 goto out_put;
 592         }
 593         if (!d_is_dir(path->dentry)) {
 594                 pr_err("overlayfs: '%s' not a directory\n", name);
 595                 goto out_put;
 596         }
 597         return 0;
 598
 599 out_put:
 600         path_put(path);
 601 out:
 602         return err;
 603 }
 604
 605 static int ovl_mount_dir(const char *name, struct path *path)
 606 {
 607         int err = -ENOMEM;
 608         char *tmp = kstrdup(name, GFP_KERNEL);
 609
 610         if (tmp) {
 611                 ovl_unescape(tmp);
 612                 err = ovl_mount_dir_noesc(tmp, path);
 613
 614                 if (!err)
 615                         if (ovl_dentry_remote(path->dentry)) {
 616                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 617                                        tmp);
 618                                 path_put(path);
 619                                 err = -EINVAL;
 620                         }
 621                 kfree(tmp);
 622         }
 623         return err;
 624 }
 625
 626 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 627                              const char *name)
 628 {
 629         struct kstatfs statfs;
 630         int err = vfs_statfs(path, &statfs);
 631
 632         if (err)
 633                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 634         else
 635                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 636
 637         return err;
 638 }
 639
 640 static int ovl_lower_dir(const char *name, struct path *path,
 641                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 642 {
 643         int err;
 644
 645         err = ovl_mount_dir_noesc(name, path);
 646         if (err)
 647                 goto out;
 648
 649         err = ovl_check_namelen(path, ofs, name);
 650         if (err)
 651                 goto out_put;
 652
 653         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 654
 655         if (ovl_dentry_remote(path->dentry))
 656                 *remote = true;
 657
 658         /*
 659          * The inodes index feature needs to encode and decode file
 660          * handles, so it requires that all layers support them.
 661          */
 662         if (ofs->config.index && !ovl_can_decode_fh(path->dentry->d_sb)) {
 663                 ofs->config.index = false;
 664                 pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off.\n", name);
 665         }
 666
 667         return 0;
 668
 669 out_put:
 670         path_put(path);
 671 out:
 672         return err;
 673 }
 674
 675 /* Workdir should not be subdir of upperdir and vice versa */
 676 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 677 {
 678         bool ok = false;
 679
 680         if (workdir != upperdir) {
 681                 ok = (lock_rename(workdir, upperdir) == NULL);
 682                 unlock_rename(workdir, upperdir);
 683         }
 684         return ok;
 685 }
 686
 687 static unsigned int ovl_split_lowerdirs(char *str)
 688 {
 689         unsigned int ctr = 1;
 690         char *s, *d;
 691
 692         for (s = d = str;; s++, d++) {
 693                 if (*s == '\\') {
 694                         s++;
 695                 } else if (*s == ':') {
 696                         *d = '\0';
 697                         ctr++;
 698                         continue;
 699                 }
 700                 *d = *s;
 701                 if (!*s)
 702                         break;
 703         }
 704         return ctr;
 705 }
 706
 707 static int __maybe_unused
 708 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 709                         struct dentry *dentry, struct inode *inode,
 710                         const char *name, void *buffer, size_t size)
 711 {
 712         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 713 }
 714
 715 static int __maybe_unused
 716 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 717                         struct dentry *dentry, struct inode *inode,
 718                         const char *name, const void *value,
 719                         size_t size, int flags)
 720 {
 721         struct dentry *workdir = ovl_workdir(dentry);
 722         struct inode *realinode = ovl_inode_real(inode);
 723         struct posix_acl *acl = NULL;
 724         int err;
 725
 726         /* Check that everything is OK before copy-up */
 727         if (value) {
 728                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 729                 if (IS_ERR(acl))
 730                         return PTR_ERR(acl);
 731         }
 732         err = -EOPNOTSUPP;
 733         if (!IS_POSIXACL(d_inode(workdir)))
 734                 goto out_acl_release;
 735         if (!realinode->i_op->set_acl)
 736                 goto out_acl_release;
 737         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 738                 err = acl ? -EACCES : 0;
 739                 goto out_acl_release;
 740         }
 741         err = -EPERM;
 742         if (!inode_owner_or_capable(inode))
 743                 goto out_acl_release;
 744
 745         posix_acl_release(acl);
 746
 747         /*
 748          * Check if sgid bit needs to be cleared (actual setacl operation will
 749          * be done with mounter's capabilities and so that won't do it for us).
 750          */
 751         if (unlikely(inode->i_mode & S_ISGID) &&
 752             handler->flags == ACL_TYPE_ACCESS &&
 753             !in_group_p(inode->i_gid) &&
 754             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 755                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 756
 757                 err = ovl_setattr(dentry, &iattr);
 758                 if (err)
 759                         return err;
 760         }
 761
 762         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
 763         if (!err)
 764                 ovl_copyattr(ovl_inode_real(inode), inode);
 765
 766         return err;
 767
 768 out_acl_release:
 769         posix_acl_release(acl);
 770         return err;
 771 }
 772
 773 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 774                              struct dentry *dentry, struct inode *inode,
 775                              const char *name, void *buffer, size_t size)
 776 {
 777         return -EOPNOTSUPP;
 778 }
 779
 780 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 781                              struct dentry *dentry, struct inode *inode,
 782                              const char *name, const void *value,
 783                              size_t size, int flags)
 784 {
 785         return -EOPNOTSUPP;
 786 }
 787
 788 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 789                                struct dentry *dentry, struct inode *inode,
 790                                const char *name, void *buffer, size_t size)
 791 {
 792         return ovl_xattr_get(dentry, inode, name, buffer, size);
 793 }
 794
 795 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 796                                struct dentry *dentry, struct inode *inode,
 797                                const char *name, const void *value,
 798                                size_t size, int flags)
 799 {
 800         return ovl_xattr_set(dentry, inode, name, value, size, flags);
 801 }
 802
 803 static const struct xattr_handler __maybe_unused
 804 ovl_posix_acl_access_xattr_handler = {
 805         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 806         .flags = ACL_TYPE_ACCESS,
 807         .get = ovl_posix_acl_xattr_get,
 808         .set = ovl_posix_acl_xattr_set,
 809 };
 810
 811 static const struct xattr_handler __maybe_unused
 812 ovl_posix_acl_default_xattr_handler = {
 813         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 814         .flags = ACL_TYPE_DEFAULT,
 815         .get = ovl_posix_acl_xattr_get,
 816         .set = ovl_posix_acl_xattr_set,
 817 };
 818
 819 static const struct xattr_handler ovl_own_xattr_handler = {
 820         .prefix = OVL_XATTR_PREFIX,
 821         .get = ovl_own_xattr_get,
 822         .set = ovl_own_xattr_set,
 823 };
 824
 825 static const struct xattr_handler ovl_other_xattr_handler = {
 826         .prefix = "", /* catch all */
 827         .get = ovl_other_xattr_get,
 828         .set = ovl_other_xattr_set,
 829 };
 830
 831 static const struct xattr_handler *ovl_xattr_handlers[] = {
 832 #ifdef CONFIG_FS_POSIX_ACL
 833         &ovl_posix_acl_access_xattr_handler,
 834         &ovl_posix_acl_default_xattr_handler,
 835 #endif
 836         &ovl_own_xattr_handler,
 837         &ovl_other_xattr_handler,
 838         NULL
 839 };
 840
 841 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 842 {
 843         struct path upperpath = { };
 844         struct path workpath = { };
 845         struct dentry *root_dentry;
 846         struct ovl_entry *oe;
 847         struct ovl_fs *ufs;
 848         struct path *stack = NULL;
 849         char *lowertmp;
 850         char *lower;
 851         unsigned int numlower;
 852         unsigned int stacklen = 0;
 853         unsigned int i;
 854         bool remote = false;
 855         struct cred *cred;
 856         int err;
 857
 858         err = -ENOMEM;
 859         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 860         if (!ufs)
 861                 goto out;
 862
 863         ufs->config.redirect_dir = ovl_redirect_dir_def;
 864         ufs->config.index = ovl_index_def;
 865         err = ovl_parse_opt((char *) data, &ufs->config);
 866         if (err)
 867                 goto out_free_config;
 868
 869         err = -EINVAL;
 870         if (!ufs->config.lowerdir) {
 871                 if (!silent)
 872                         pr_err("overlayfs: missing 'lowerdir'\n");
 873                 goto out_free_config;
 874         }
 875
 876         sb->s_stack_depth = 0;
 877         sb->s_maxbytes = MAX_LFS_FILESIZE;
 878         if (ufs->config.upperdir) {
 879                 if (!ufs->config.workdir) {
 880                         pr_err("overlayfs: missing 'workdir'\n");
 881                         goto out_free_config;
 882                 }
 883
 884                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 885                 if (err)
 886                         goto out_free_config;
 887
 888                 /* Upper fs should not be r/o */
 889                 if (sb_rdonly(upperpath.mnt->mnt_sb)) {
 890                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 891                         err = -EINVAL;
 892                         goto out_put_upperpath;
 893                 }
 894
 895                 err = ovl_check_namelen(&upperpath, ufs, ufs->config.upperdir);
 896                 if (err)
 897                         goto out_put_upperpath;
 898
 899                 err = -EBUSY;
 900                 if (ovl_inuse_trylock(upperpath.dentry)) {
 901                         ufs->upperdir_locked = true;
 902                 } else if (ufs->config.index) {
 903                         pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n");
 904                         goto out_put_upperpath;
 905                 } else {
 906                         pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
 907                 }
 908
 909                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 910                 if (err)
 911                         goto out_unlock_upperdentry;
 912
 913                 err = -EINVAL;
 914                 if (upperpath.mnt != workpath.mnt) {
 915                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 916                         goto out_put_workpath;
 917                 }
 918                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 919                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 920                         goto out_put_workpath;
 921                 }
 922
 923                 err = -EBUSY;
 924                 if (ovl_inuse_trylock(workpath.dentry)) {
 925                         ufs->workdir_locked = true;
 926                 } else if (ufs->config.index) {
 927                         pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n");
 928                         goto out_put_workpath;
 929                 } else {
 930                         pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
 931                 }
 932
 933                 ufs->workbasedir = workpath.dentry;
 934                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 935         }
 936         err = -ENOMEM;
 937         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 938         if (!lowertmp)
 939                 goto out_unlock_workdentry;
 940
 941         err = -EINVAL;
 942         stacklen = ovl_split_lowerdirs(lowertmp);
 943         if (stacklen > OVL_MAX_STACK) {
 944                 pr_err("overlayfs: too many lower directories, limit is %d\n",
 945                        OVL_MAX_STACK);
 946                 goto out_free_lowertmp;
 947         } else if (!ufs->config.upperdir && stacklen == 1) {
 948                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 949                 goto out_free_lowertmp;
 950         }
 951
 952         err = -ENOMEM;
 953         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 954         if (!stack)
 955                 goto out_free_lowertmp;
 956
 957         err = -EINVAL;
 958         lower = lowertmp;
 959         for (numlower = 0; numlower < stacklen; numlower++) {
 960                 err = ovl_lower_dir(lower, &stack[numlower], ufs,
 961                                     &sb->s_stack_depth, &remote);
 962                 if (err)
 963                         goto out_put_lowerpath;
 964
 965                 lower = strchr(lower, '\0') + 1;
 966         }
 967
 968         err = -EINVAL;
 969         sb->s_stack_depth++;
 970         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
 971                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
 972                 goto out_put_lowerpath;
 973         }
 974
 975         if (ufs->config.upperdir) {
 976                 ufs->upper_mnt = clone_private_mount(&upperpath);
 977                 err = PTR_ERR(ufs->upper_mnt);
 978                 if (IS_ERR(ufs->upper_mnt)) {
 979                         pr_err("overlayfs: failed to clone upperpath\n");
 980                         goto out_put_lowerpath;
 981                 }
 982
 983                 /* Don't inherit atime flags */
 984                 ufs->upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
 985
 986                 sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
 987
 988                 ufs->workdir = ovl_workdir_create(sb, ufs, workpath.dentry,
 989                                                   OVL_WORKDIR_NAME, false);
 990                 /*
 991                  * Upper should support d_type, else whiteouts are visible.
 992                  * Given workdir and upper are on same fs, we can do
 993                  * iterate_dir() on workdir. This check requires successful
 994                  * creation of workdir in previous step.
 995                  */
 996                 if (ufs->workdir) {
 997                         struct dentry *temp;
 998
 999                         err = ovl_check_d_type_supported(&workpath);
1000                         if (err < 0)
1001                                 goto out_put_workdir;
1002
1003                         /*
1004                          * We allowed this configuration and don't want to
1005                          * break users over kernel upgrade. So warn instead
1006                          * of erroring out.
1007                          */
1008                         if (!err)
1009                                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
1010
1011                         /* Check if upper/work fs supports O_TMPFILE */
1012                         temp = ovl_do_tmpfile(ufs->workdir, S_IFREG | 0);
1013                         ufs->tmpfile = !IS_ERR(temp);
1014                         if (ufs->tmpfile)
1015                                 dput(temp);
1016                         else
1017                                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
1018
1019                         /*
1020                          * Check if upper/work fs supports trusted.overlay.*
1021                          * xattr
1022                          */
1023                         err = ovl_do_setxattr(ufs->workdir, OVL_XATTR_OPAQUE,
1024                                               "0", 1, 0);
1025                         if (err) {
1026                                 ufs->noxattr = true;
1027                                 pr_warn("overlayfs: upper fs does not support xattr.\n");
1028                         } else {
1029                                 vfs_removexattr(ufs->workdir, OVL_XATTR_OPAQUE);
1030                         }
1031
1032                         /* Check if upper/work fs supports file handles */
1033                         if (ufs->config.index &&
1034                             !ovl_can_decode_fh(ufs->workdir->d_sb)) {
1035                                 ufs->config.index = false;
1036                                 pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1037                         }
1038                 }
1039         }
1040
1041         err = -ENOMEM;
1042         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1043         if (ufs->lower_mnt == NULL)
1044                 goto out_put_workdir;
1045         for (i = 0; i < numlower; i++) {
1046                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1047
1048                 err = PTR_ERR(mnt);
1049                 if (IS_ERR(mnt)) {
1050                         pr_err("overlayfs: failed to clone lowerpath\n");
1051                         goto out_put_lower_mnt;
1052                 }
1053                 /*
1054                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1055                  * will fail instead of modifying lower fs.
1056                  */
1057                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1058
1059                 ufs->lower_mnt[ufs->numlower] = mnt;
1060                 ufs->numlower++;
1061
1062                 /* Check if all lower layers are on same sb */
1063                 if (i == 0)
1064                         ufs->same_sb = mnt->mnt_sb;
1065                 else if (ufs->same_sb != mnt->mnt_sb)
1066                         ufs->same_sb = NULL;
1067         }
1068
1069         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1070         if (!ufs->upper_mnt)
1071                 sb->s_flags |= MS_RDONLY;
1072         else if (ufs->upper_mnt->mnt_sb != ufs->same_sb)
1073                 ufs->same_sb = NULL;
1074
1075         if (!(ovl_force_readonly(ufs)) && ufs->config.index) {
1076                 /* Verify lower root is upper root origin */
1077                 err = ovl_verify_origin(upperpath.dentry, ufs->lower_mnt[0],
1078                                         stack[0].dentry, false, true);
1079                 if (err) {
1080                         pr_err("overlayfs: failed to verify upper root origin\n");
1081                         goto out_put_lower_mnt;
1082                 }
1083
1084                 ufs->indexdir = ovl_workdir_create(sb, ufs, workpath.dentry,
1085                                                    OVL_INDEXDIR_NAME, true);
1086                 if (ufs->indexdir) {
1087                         /* Verify upper root is index dir origin */
1088                         err = ovl_verify_origin(ufs->indexdir, ufs->upper_mnt,
1089                                                 upperpath.dentry, true, true);
1090                         if (err)
1091                                 pr_err("overlayfs: failed to verify index dir origin\n");
1092
1093                         /* Cleanup bad/stale/orphan index entries */
1094                         if (!err)
1095                                 err = ovl_indexdir_cleanup(ufs->indexdir,
1096                                                            ufs->upper_mnt,
1097                                                            stack, numlower);
1098                 }
1099                 if (err || !ufs->indexdir)
1100                         pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1101                 if (err)
1102                         goto out_put_indexdir;
1103         }
1104
1105         /* Show index=off/on in /proc/mounts for any of the reasons above */
1106         if (!ufs->indexdir)
1107                 ufs->config.index = false;
1108
1109         if (remote)
1110                 sb->s_d_op = &ovl_reval_dentry_operations;
1111         else
1112                 sb->s_d_op = &ovl_dentry_operations;
1113
1114         err = -ENOMEM;
1115         ufs->creator_cred = cred = prepare_creds();
1116         if (!cred)
1117                 goto out_put_indexdir;
1118
1119         /* Never override disk quota limits or use reserved space */
1120         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1121
1122         err = -ENOMEM;
1123         oe = ovl_alloc_entry(numlower);
1124         if (!oe)
1125                 goto out_put_cred;
1126
1127         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1128         sb->s_op = &ovl_super_operations;
1129         sb->s_xattr = ovl_xattr_handlers;
1130         sb->s_fs_info = ufs;
1131         sb->s_flags |= MS_POSIXACL | MS_NOREMOTELOCK;
1132
1133         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
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         mntput(workpath.mnt);
1141         kfree(lowertmp);
1142
1143         if (upperpath.dentry) {
1144                 oe->has_upper = true;
1145                 if (ovl_is_impuredir(upperpath.dentry))
1146                         ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1147         }
1148         for (i = 0; i < numlower; i++) {
1149                 oe->lowerstack[i].dentry = stack[i].dentry;
1150                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1151         }
1152         kfree(stack);
1153
1154         root_dentry->d_fsdata = oe;
1155
1156         ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1157                        ovl_dentry_lower(root_dentry));
1158
1159         sb->s_root = root_dentry;
1160
1161         return 0;
1162
1163 out_free_oe:
1164         kfree(oe);
1165 out_put_cred:
1166         put_cred(ufs->creator_cred);
1167 out_put_indexdir:
1168         dput(ufs->indexdir);
1169 out_put_lower_mnt:
1170         for (i = 0; i < ufs->numlower; i++)
1171                 mntput(ufs->lower_mnt[i]);
1172         kfree(ufs->lower_mnt);
1173 out_put_workdir:
1174         dput(ufs->workdir);
1175         mntput(ufs->upper_mnt);
1176 out_put_lowerpath:
1177         for (i = 0; i < numlower; i++)
1178                 path_put(&stack[i]);
1179         kfree(stack);
1180 out_free_lowertmp:
1181         kfree(lowertmp);
1182 out_unlock_workdentry:
1183         if (ufs->workdir_locked)
1184                 ovl_inuse_unlock(workpath.dentry);
1185 out_put_workpath:
1186         path_put(&workpath);
1187 out_unlock_upperdentry:
1188         if (ufs->upperdir_locked)
1189                 ovl_inuse_unlock(upperpath.dentry);
1190 out_put_upperpath:
1191         path_put(&upperpath);
1192 out_free_config:
1193         kfree(ufs->config.lowerdir);
1194         kfree(ufs->config.upperdir);
1195         kfree(ufs->config.workdir);
1196         kfree(ufs);
1197 out:
1198         return err;
1199 }
1200
1201 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1202                                 const char *dev_name, void *raw_data)
1203 {
1204         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1205 }
1206
1207 static struct file_system_type ovl_fs_type = {
1208         .owner          = THIS_MODULE,
1209         .name           = "overlay",
1210         .mount          = ovl_mount,
1211         .kill_sb        = kill_anon_super,
1212 };
1213 MODULE_ALIAS_FS("overlay");
1214
1215 static void ovl_inode_init_once(void *foo)
1216 {
1217         struct ovl_inode *oi = foo;
1218
1219         inode_init_once(&oi->vfs_inode);
1220 }
1221
1222 static int __init ovl_init(void)
1223 {
1224         int err;
1225
1226         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1227                                              sizeof(struct ovl_inode), 0,
1228                                              (SLAB_RECLAIM_ACCOUNT|
1229                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1230                                              ovl_inode_init_once);
1231         if (ovl_inode_cachep == NULL)
1232                 return -ENOMEM;
1233
1234         err = register_filesystem(&ovl_fs_type);
1235         if (err)
1236                 kmem_cache_destroy(ovl_inode_cachep);
1237
1238         return err;
1239 }
1240
1241 static void __exit ovl_exit(void)
1242 {
1243         unregister_filesystem(&ovl_fs_type);
1244
1245         /*
1246          * Make sure all delayed rcu free inodes are flushed before we
1247          * destroy cache.
1248          */
1249         rcu_barrier();
1250         kmem_cache_destroy(ovl_inode_cachep);
1251
1252 }
1253
1254 module_init(ovl_init);
1255 module_exit(ovl_exit);