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         kfree(oi->redirect);
 204         ovl_dir_cache_free(inode);
 205         mutex_destroy(&oi->lock);
 206
 207         call_rcu(&inode->i_rcu, ovl_i_callback);
 208 }
 209
 210 static void ovl_put_super(struct super_block *sb)
 211 {
 212         struct ovl_fs *ufs = sb->s_fs_info;
 213         unsigned i;
 214
 215         dput(ufs->indexdir);
 216         dput(ufs->workdir);
 217         if (ufs->workdir_locked)
 218                 ovl_inuse_unlock(ufs->workbasedir);
 219         dput(ufs->workbasedir);
 220         if (ufs->upper_mnt && ufs->upperdir_locked)
 221                 ovl_inuse_unlock(ufs->upper_mnt->mnt_root);
 222         mntput(ufs->upper_mnt);
 223         for (i = 0; i < ufs->numlower; i++)
 224                 mntput(ufs->lower_mnt[i]);
 225         kfree(ufs->lower_mnt);
 226
 227         kfree(ufs->config.lowerdir);
 228         kfree(ufs->config.upperdir);
 229         kfree(ufs->config.workdir);
 230         put_cred(ufs->creator_cred);
 231         kfree(ufs);
 232 }
 233
 234 static int ovl_sync_fs(struct super_block *sb, int wait)
 235 {
 236         struct ovl_fs *ufs = sb->s_fs_info;
 237         struct super_block *upper_sb;
 238         int ret;
 239
 240         if (!ufs->upper_mnt)
 241                 return 0;
 242         upper_sb = ufs->upper_mnt->mnt_sb;
 243         if (!upper_sb->s_op->sync_fs)
 244                 return 0;
 245
 246         /* real inodes have already been synced by sync_filesystem(ovl_sb) */
 247         down_read(&upper_sb->s_umount);
 248         ret = upper_sb->s_op->sync_fs(upper_sb, wait);
 249         up_read(&upper_sb->s_umount);
 250         return ret;
 251 }
 252
 253 /**
 254  * ovl_statfs
 255  * @sb: The overlayfs super block
 256  * @buf: The struct kstatfs to fill in with stats
 257  *
 258  * Get the filesystem statistics.  As writes always target the upper layer
 259  * filesystem pass the statfs to the upper filesystem (if it exists)
 260  */
 261 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 262 {
 263         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 264         struct dentry *root_dentry = dentry->d_sb->s_root;
 265         struct path path;
 266         int err;
 267
 268         ovl_path_real(root_dentry, &path);
 269
 270         err = vfs_statfs(&path, buf);
 271         if (!err) {
 272                 buf->f_namelen = ofs->namelen;
 273                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 274         }
 275
 276         return err;
 277 }
 278
 279 /* Will this overlay be forced to mount/remount ro? */
 280 static bool ovl_force_readonly(struct ovl_fs *ufs)
 281 {
 282         return (!ufs->upper_mnt || !ufs->workdir);
 283 }
 284
 285 /**
 286  * ovl_show_options
 287  *
 288  * Prints the mount options for a given superblock.
 289  * Returns zero; does not fail.
 290  */
 291 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 292 {
 293         struct super_block *sb = dentry->d_sb;
 294         struct ovl_fs *ufs = sb->s_fs_info;
 295
 296         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 297         if (ufs->config.upperdir) {
 298                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 299                 seq_show_option(m, "workdir", ufs->config.workdir);
 300         }
 301         if (ufs->config.default_permissions)
 302                 seq_puts(m, ",default_permissions");
 303         if (ufs->config.redirect_dir != ovl_redirect_dir_def)
 304                 seq_printf(m, ",redirect_dir=%s",
 305                            ufs->config.redirect_dir ? "on" : "off");
 306         if (ufs->config.index != ovl_index_def)
 307                 seq_printf(m, ",index=%s",
 308                            ufs->config.index ? "on" : "off");
 309         return 0;
 310 }
 311
 312 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 313 {
 314         struct ovl_fs *ufs = sb->s_fs_info;
 315
 316         if (!(*flags & MS_RDONLY) && ovl_force_readonly(ufs))
 317                 return -EROFS;
 318
 319         return 0;
 320 }
 321
 322 static const struct super_operations ovl_super_operations = {
 323         .alloc_inode    = ovl_alloc_inode,
 324         .destroy_inode  = ovl_destroy_inode,
 325         .drop_inode     = generic_delete_inode,
 326         .put_super      = ovl_put_super,
 327         .sync_fs        = ovl_sync_fs,
 328         .statfs         = ovl_statfs,
 329         .show_options   = ovl_show_options,
 330         .remount_fs     = ovl_remount,
 331 };
 332
 333 enum {
 334         OPT_LOWERDIR,
 335         OPT_UPPERDIR,
 336         OPT_WORKDIR,
 337         OPT_DEFAULT_PERMISSIONS,
 338         OPT_REDIRECT_DIR_ON,
 339         OPT_REDIRECT_DIR_OFF,
 340         OPT_INDEX_ON,
 341         OPT_INDEX_OFF,
 342         OPT_ERR,
 343 };
 344
 345 static const match_table_t ovl_tokens = {
 346         {OPT_LOWERDIR,                  "lowerdir=%s"},
 347         {OPT_UPPERDIR,                  "upperdir=%s"},
 348         {OPT_WORKDIR,                   "workdir=%s"},
 349         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 350         {OPT_REDIRECT_DIR_ON,           "redirect_dir=on"},
 351         {OPT_REDIRECT_DIR_OFF,          "redirect_dir=off"},
 352         {OPT_INDEX_ON,                  "index=on"},
 353         {OPT_INDEX_OFF,                 "index=off"},
 354         {OPT_ERR,                       NULL}
 355 };
 356
 357 static char *ovl_next_opt(char **s)
 358 {
 359         char *sbegin = *s;
 360         char *p;
 361
 362         if (sbegin == NULL)
 363                 return NULL;
 364
 365         for (p = sbegin; *p; p++) {
 366                 if (*p == '\\') {
 367                         p++;
 368                         if (!*p)
 369                                 break;
 370                 } else if (*p == ',') {
 371                         *p = '\0';
 372                         *s = p + 1;
 373                         return sbegin;
 374                 }
 375         }
 376         *s = NULL;
 377         return sbegin;
 378 }
 379
 380 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 381 {
 382         char *p;
 383
 384         while ((p = ovl_next_opt(&opt)) != NULL) {
 385                 int token;
 386                 substring_t args[MAX_OPT_ARGS];
 387
 388                 if (!*p)
 389                         continue;
 390
 391                 token = match_token(p, ovl_tokens, args);
 392                 switch (token) {
 393                 case OPT_UPPERDIR:
 394                         kfree(config->upperdir);
 395                         config->upperdir = match_strdup(&args[0]);
 396                         if (!config->upperdir)
 397                                 return -ENOMEM;
 398                         break;
 399
 400                 case OPT_LOWERDIR:
 401                         kfree(config->lowerdir);
 402                         config->lowerdir = match_strdup(&args[0]);
 403                         if (!config->lowerdir)
 404                                 return -ENOMEM;
 405                         break;
 406
 407                 case OPT_WORKDIR:
 408                         kfree(config->workdir);
 409                         config->workdir = match_strdup(&args[0]);
 410                         if (!config->workdir)
 411                                 return -ENOMEM;
 412                         break;
 413
 414                 case OPT_DEFAULT_PERMISSIONS:
 415                         config->default_permissions = true;
 416                         break;
 417
 418                 case OPT_REDIRECT_DIR_ON:
 419                         config->redirect_dir = true;
 420                         break;
 421
 422                 case OPT_REDIRECT_DIR_OFF:
 423                         config->redirect_dir = false;
 424                         break;
 425
 426                 case OPT_INDEX_ON:
 427                         config->index = true;
 428                         break;
 429
 430                 case OPT_INDEX_OFF:
 431                         config->index = false;
 432                         break;
 433
 434                 default:
 435                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 436                         return -EINVAL;
 437                 }
 438         }
 439
 440         /* Workdir is useless in non-upper mount */
 441         if (!config->upperdir && config->workdir) {
 442                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 443                         config->workdir);
 444                 kfree(config->workdir);
 445                 config->workdir = NULL;
 446         }
 447
 448         return 0;
 449 }
 450
 451 #define OVL_WORKDIR_NAME "work"
 452 #define OVL_INDEXDIR_NAME "index"
 453
 454 static struct dentry *ovl_workdir_create(struct super_block *sb,
 455                                          struct ovl_fs *ufs,
 456                                          struct dentry *dentry,
 457                                          const char *name, bool persist)
 458 {
 459         struct inode *dir = dentry->d_inode;
 460         struct vfsmount *mnt = ufs->upper_mnt;
 461         struct dentry *work;
 462         int err;
 463         bool retried = false;
 464         bool locked = false;
 465
 466         err = mnt_want_write(mnt);
 467         if (err)
 468                 goto out_err;
 469
 470         inode_lock_nested(dir, I_MUTEX_PARENT);
 471         locked = true;
 472
 473 retry:
 474         work = lookup_one_len(name, dentry, strlen(name));
 475
 476         if (!IS_ERR(work)) {
 477                 struct iattr attr = {
 478                         .ia_valid = ATTR_MODE,
 479                         .ia_mode = S_IFDIR | 0,
 480                 };
 481
 482                 if (work->d_inode) {
 483                         err = -EEXIST;
 484                         if (retried)
 485                                 goto out_dput;
 486
 487                         if (persist)
 488                                 goto out_unlock;
 489
 490                         retried = true;
 491                         ovl_workdir_cleanup(dir, mnt, work, 0);
 492                         dput(work);
 493                         goto retry;
 494                 }
 495
 496                 err = ovl_create_real(dir, work,
 497                                       &(struct cattr){.mode = S_IFDIR | 0},
 498                                       NULL, true);
 499                 if (err)
 500                         goto out_dput;
 501
 502                 /*
 503                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 504                  *
 505                  * a) success (there was a POSIX ACL xattr and was removed)
 506                  * b) -ENODATA (there was no POSIX ACL xattr)
 507                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 508                  *
 509                  * There are various other error values that could effectively
 510                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 511                  * if the xattr name is too long), but the set of filesystems
 512                  * allowed as upper are limited to "normal" ones, where checking
 513                  * for the above two errors is sufficient.
 514                  */
 515                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 516                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 517                         goto out_dput;
 518
 519                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 520                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 521                         goto out_dput;
 522
 523                 /* Clear any inherited mode bits */
 524                 inode_lock(work->d_inode);
 525                 err = notify_change(work, &attr, NULL);
 526                 inode_unlock(work->d_inode);
 527                 if (err)
 528                         goto out_dput;
 529         } else {
 530                 err = PTR_ERR(work);
 531                 goto out_err;
 532         }
 533 out_unlock:
 534         mnt_drop_write(mnt);
 535         if (locked)
 536                 inode_unlock(dir);
 537
 538         return work;
 539
 540 out_dput:
 541         dput(work);
 542 out_err:
 543         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 544                 ufs->config.workdir, name, -err);
 545         sb->s_flags |= MS_RDONLY;
 546         work = NULL;
 547         goto out_unlock;
 548 }
 549
 550 static void ovl_unescape(char *s)
 551 {
 552         char *d = s;
 553
 554         for (;; s++, d++) {
 555                 if (*s == '\\')
 556                         s++;
 557                 *d = *s;
 558                 if (!*s)
 559                         break;
 560         }
 561 }
 562
 563 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 564 {
 565         int err = -EINVAL;
 566
 567         if (!*name) {
 568                 pr_err("overlayfs: empty lowerdir\n");
 569                 goto out;
 570         }
 571         err = kern_path(name, LOOKUP_FOLLOW, path);
 572         if (err) {
 573                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 574                 goto out;
 575         }
 576         err = -EINVAL;
 577         if (ovl_dentry_weird(path->dentry)) {
 578                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 579                 goto out_put;
 580         }
 581         if (!d_is_dir(path->dentry)) {
 582                 pr_err("overlayfs: '%s' not a directory\n", name);
 583                 goto out_put;
 584         }
 585         return 0;
 586
 587 out_put:
 588         path_put(path);
 589 out:
 590         return err;
 591 }
 592
 593 static int ovl_mount_dir(const char *name, struct path *path)
 594 {
 595         int err = -ENOMEM;
 596         char *tmp = kstrdup(name, GFP_KERNEL);
 597
 598         if (tmp) {
 599                 ovl_unescape(tmp);
 600                 err = ovl_mount_dir_noesc(tmp, path);
 601
 602                 if (!err)
 603                         if (ovl_dentry_remote(path->dentry)) {
 604                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 605                                        tmp);
 606                                 path_put(path);
 607                                 err = -EINVAL;
 608                         }
 609                 kfree(tmp);
 610         }
 611         return err;
 612 }
 613
 614 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 615                              const char *name)
 616 {
 617         struct kstatfs statfs;
 618         int err = vfs_statfs(path, &statfs);
 619
 620         if (err)
 621                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 622         else
 623                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 624
 625         return err;
 626 }
 627
 628 static int ovl_lower_dir(const char *name, struct path *path,
 629                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 630 {
 631         int err;
 632
 633         err = ovl_mount_dir_noesc(name, path);
 634         if (err)
 635                 goto out;
 636
 637         err = ovl_check_namelen(path, ofs, name);
 638         if (err)
 639                 goto out_put;
 640
 641         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 642
 643         if (ovl_dentry_remote(path->dentry))
 644                 *remote = true;
 645
 646         /*
 647          * The inodes index feature needs to encode and decode file
 648          * handles, so it requires that all layers support them.
 649          */
 650         if (ofs->config.index && !ovl_can_decode_fh(path->dentry->d_sb)) {
 651                 ofs->config.index = false;
 652                 pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off.\n", name);
 653         }
 654
 655         return 0;
 656
 657 out_put:
 658         path_put(path);
 659 out:
 660         return err;
 661 }
 662
 663 /* Workdir should not be subdir of upperdir and vice versa */
 664 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 665 {
 666         bool ok = false;
 667
 668         if (workdir != upperdir) {
 669                 ok = (lock_rename(workdir, upperdir) == NULL);
 670                 unlock_rename(workdir, upperdir);
 671         }
 672         return ok;
 673 }
 674
 675 static unsigned int ovl_split_lowerdirs(char *str)
 676 {
 677         unsigned int ctr = 1;
 678         char *s, *d;
 679
 680         for (s = d = str;; s++, d++) {
 681                 if (*s == '\\') {
 682                         s++;
 683                 } else if (*s == ':') {
 684                         *d = '\0';
 685                         ctr++;
 686                         continue;
 687                 }
 688                 *d = *s;
 689                 if (!*s)
 690                         break;
 691         }
 692         return ctr;
 693 }
 694
 695 static int __maybe_unused
 696 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 697                         struct dentry *dentry, struct inode *inode,
 698                         const char *name, void *buffer, size_t size)
 699 {
 700         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 701 }
 702
 703 static int __maybe_unused
 704 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 705                         struct dentry *dentry, struct inode *inode,
 706                         const char *name, const void *value,
 707                         size_t size, int flags)
 708 {
 709         struct dentry *workdir = ovl_workdir(dentry);
 710         struct inode *realinode = ovl_inode_real(inode);
 711         struct posix_acl *acl = NULL;
 712         int err;
 713
 714         /* Check that everything is OK before copy-up */
 715         if (value) {
 716                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 717                 if (IS_ERR(acl))
 718                         return PTR_ERR(acl);
 719         }
 720         err = -EOPNOTSUPP;
 721         if (!IS_POSIXACL(d_inode(workdir)))
 722                 goto out_acl_release;
 723         if (!realinode->i_op->set_acl)
 724                 goto out_acl_release;
 725         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 726                 err = acl ? -EACCES : 0;
 727                 goto out_acl_release;
 728         }
 729         err = -EPERM;
 730         if (!inode_owner_or_capable(inode))
 731                 goto out_acl_release;
 732
 733         posix_acl_release(acl);
 734
 735         /*
 736          * Check if sgid bit needs to be cleared (actual setacl operation will
 737          * be done with mounter's capabilities and so that won't do it for us).
 738          */
 739         if (unlikely(inode->i_mode & S_ISGID) &&
 740             handler->flags == ACL_TYPE_ACCESS &&
 741             !in_group_p(inode->i_gid) &&
 742             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 743                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 744
 745                 err = ovl_setattr(dentry, &iattr);
 746                 if (err)
 747                         return err;
 748         }
 749
 750         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
 751         if (!err)
 752                 ovl_copyattr(ovl_inode_real(inode), inode);
 753
 754         return err;
 755
 756 out_acl_release:
 757         posix_acl_release(acl);
 758         return err;
 759 }
 760
 761 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 762                              struct dentry *dentry, struct inode *inode,
 763                              const char *name, void *buffer, size_t size)
 764 {
 765         return -EOPNOTSUPP;
 766 }
 767
 768 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 769                              struct dentry *dentry, struct inode *inode,
 770                              const char *name, const void *value,
 771                              size_t size, int flags)
 772 {
 773         return -EOPNOTSUPP;
 774 }
 775
 776 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 777                                struct dentry *dentry, struct inode *inode,
 778                                const char *name, void *buffer, size_t size)
 779 {
 780         return ovl_xattr_get(dentry, inode, name, buffer, size);
 781 }
 782
 783 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 784                                struct dentry *dentry, struct inode *inode,
 785                                const char *name, const void *value,
 786                                size_t size, int flags)
 787 {
 788         return ovl_xattr_set(dentry, inode, name, value, size, flags);
 789 }
 790
 791 static const struct xattr_handler __maybe_unused
 792 ovl_posix_acl_access_xattr_handler = {
 793         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 794         .flags = ACL_TYPE_ACCESS,
 795         .get = ovl_posix_acl_xattr_get,
 796         .set = ovl_posix_acl_xattr_set,
 797 };
 798
 799 static const struct xattr_handler __maybe_unused
 800 ovl_posix_acl_default_xattr_handler = {
 801         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 802         .flags = ACL_TYPE_DEFAULT,
 803         .get = ovl_posix_acl_xattr_get,
 804         .set = ovl_posix_acl_xattr_set,
 805 };
 806
 807 static const struct xattr_handler ovl_own_xattr_handler = {
 808         .prefix = OVL_XATTR_PREFIX,
 809         .get = ovl_own_xattr_get,
 810         .set = ovl_own_xattr_set,
 811 };
 812
 813 static const struct xattr_handler ovl_other_xattr_handler = {
 814         .prefix = "", /* catch all */
 815         .get = ovl_other_xattr_get,
 816         .set = ovl_other_xattr_set,
 817 };
 818
 819 static const struct xattr_handler *ovl_xattr_handlers[] = {
 820 #ifdef CONFIG_FS_POSIX_ACL
 821         &ovl_posix_acl_access_xattr_handler,
 822         &ovl_posix_acl_default_xattr_handler,
 823 #endif
 824         &ovl_own_xattr_handler,
 825         &ovl_other_xattr_handler,
 826         NULL
 827 };
 828
 829 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 830 {
 831         struct path upperpath = { };
 832         struct path workpath = { };
 833         struct dentry *root_dentry;
 834         struct ovl_entry *oe;
 835         struct ovl_fs *ufs;
 836         struct path *stack = NULL;
 837         char *lowertmp;
 838         char *lower;
 839         unsigned int numlower;
 840         unsigned int stacklen = 0;
 841         unsigned int i;
 842         bool remote = false;
 843         struct cred *cred;
 844         int err;
 845
 846         err = -ENOMEM;
 847         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 848         if (!ufs)
 849                 goto out;
 850
 851         ufs->config.redirect_dir = ovl_redirect_dir_def;
 852         ufs->config.index = ovl_index_def;
 853         err = ovl_parse_opt((char *) data, &ufs->config);
 854         if (err)
 855                 goto out_free_config;
 856
 857         err = -EINVAL;
 858         if (!ufs->config.lowerdir) {
 859                 if (!silent)
 860                         pr_err("overlayfs: missing 'lowerdir'\n");
 861                 goto out_free_config;
 862         }
 863
 864         sb->s_stack_depth = 0;
 865         sb->s_maxbytes = MAX_LFS_FILESIZE;
 866         if (ufs->config.upperdir) {
 867                 if (!ufs->config.workdir) {
 868                         pr_err("overlayfs: missing 'workdir'\n");
 869                         goto out_free_config;
 870                 }
 871
 872                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 873                 if (err)
 874                         goto out_free_config;
 875
 876                 /* Upper fs should not be r/o */
 877                 if (sb_rdonly(upperpath.mnt->mnt_sb)) {
 878                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 879                         err = -EINVAL;
 880                         goto out_put_upperpath;
 881                 }
 882
 883                 err = ovl_check_namelen(&upperpath, ufs, ufs->config.upperdir);
 884                 if (err)
 885                         goto out_put_upperpath;
 886
 887                 err = -EBUSY;
 888                 if (ovl_inuse_trylock(upperpath.dentry)) {
 889                         ufs->upperdir_locked = true;
 890                 } else if (ufs->config.index) {
 891                         pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n");
 892                         goto out_put_upperpath;
 893                 } else {
 894                         pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
 895                 }
 896
 897                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 898                 if (err)
 899                         goto out_unlock_upperdentry;
 900
 901                 err = -EINVAL;
 902                 if (upperpath.mnt != workpath.mnt) {
 903                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 904                         goto out_put_workpath;
 905                 }
 906                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 907                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 908                         goto out_put_workpath;
 909                 }
 910
 911                 err = -EBUSY;
 912                 if (ovl_inuse_trylock(workpath.dentry)) {
 913                         ufs->workdir_locked = true;
 914                 } else if (ufs->config.index) {
 915                         pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n");
 916                         goto out_put_workpath;
 917                 } else {
 918                         pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
 919                 }
 920
 921                 ufs->workbasedir = workpath.dentry;
 922                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 923         }
 924         err = -ENOMEM;
 925         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 926         if (!lowertmp)
 927                 goto out_unlock_workdentry;
 928
 929         err = -EINVAL;
 930         stacklen = ovl_split_lowerdirs(lowertmp);
 931         if (stacklen > OVL_MAX_STACK) {
 932                 pr_err("overlayfs: too many lower directories, limit is %d\n",
 933                        OVL_MAX_STACK);
 934                 goto out_free_lowertmp;
 935         } else if (!ufs->config.upperdir && stacklen == 1) {
 936                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 937                 goto out_free_lowertmp;
 938         }
 939
 940         err = -ENOMEM;
 941         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 942         if (!stack)
 943                 goto out_free_lowertmp;
 944
 945         err = -EINVAL;
 946         lower = lowertmp;
 947         for (numlower = 0; numlower < stacklen; numlower++) {
 948                 err = ovl_lower_dir(lower, &stack[numlower], ufs,
 949                                     &sb->s_stack_depth, &remote);
 950                 if (err)
 951                         goto out_put_lowerpath;
 952
 953                 lower = strchr(lower, '\0') + 1;
 954         }
 955
 956         err = -EINVAL;
 957         sb->s_stack_depth++;
 958         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
 959                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
 960                 goto out_put_lowerpath;
 961         }
 962
 963         if (ufs->config.upperdir) {
 964                 ufs->upper_mnt = clone_private_mount(&upperpath);
 965                 err = PTR_ERR(ufs->upper_mnt);
 966                 if (IS_ERR(ufs->upper_mnt)) {
 967                         pr_err("overlayfs: failed to clone upperpath\n");
 968                         goto out_put_lowerpath;
 969                 }
 970
 971                 /* Don't inherit atime flags */
 972                 ufs->upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
 973
 974                 sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
 975
 976                 ufs->workdir = ovl_workdir_create(sb, ufs, workpath.dentry,
 977                                                   OVL_WORKDIR_NAME, false);
 978                 /*
 979                  * Upper should support d_type, else whiteouts are visible.
 980                  * Given workdir and upper are on same fs, we can do
 981                  * iterate_dir() on workdir. This check requires successful
 982                  * creation of workdir in previous step.
 983                  */
 984                 if (ufs->workdir) {
 985                         struct dentry *temp;
 986
 987                         err = ovl_check_d_type_supported(&workpath);
 988                         if (err < 0)
 989                                 goto out_put_workdir;
 990
 991                         /*
 992                          * We allowed this configuration and don't want to
 993                          * break users over kernel upgrade. So warn instead
 994                          * of erroring out.
 995                          */
 996                         if (!err)
 997                                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
 998
 999                         /* Check if upper/work fs supports O_TMPFILE */
1000                         temp = ovl_do_tmpfile(ufs->workdir, S_IFREG | 0);
1001                         ufs->tmpfile = !IS_ERR(temp);
1002                         if (ufs->tmpfile)
1003                                 dput(temp);
1004                         else
1005                                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
1006
1007                         /*
1008                          * Check if upper/work fs supports trusted.overlay.*
1009                          * xattr
1010                          */
1011                         err = ovl_do_setxattr(ufs->workdir, OVL_XATTR_OPAQUE,
1012                                               "0", 1, 0);
1013                         if (err) {
1014                                 ufs->noxattr = true;
1015                                 pr_warn("overlayfs: upper fs does not support xattr.\n");
1016                         } else {
1017                                 vfs_removexattr(ufs->workdir, OVL_XATTR_OPAQUE);
1018                         }
1019
1020                         /* Check if upper/work fs supports file handles */
1021                         if (ufs->config.index &&
1022                             !ovl_can_decode_fh(ufs->workdir->d_sb)) {
1023                                 ufs->config.index = false;
1024                                 pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1025                         }
1026                 }
1027         }
1028
1029         err = -ENOMEM;
1030         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1031         if (ufs->lower_mnt == NULL)
1032                 goto out_put_workdir;
1033         for (i = 0; i < numlower; i++) {
1034                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1035
1036                 err = PTR_ERR(mnt);
1037                 if (IS_ERR(mnt)) {
1038                         pr_err("overlayfs: failed to clone lowerpath\n");
1039                         goto out_put_lower_mnt;
1040                 }
1041                 /*
1042                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1043                  * will fail instead of modifying lower fs.
1044                  */
1045                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1046
1047                 ufs->lower_mnt[ufs->numlower] = mnt;
1048                 ufs->numlower++;
1049
1050                 /* Check if all lower layers are on same sb */
1051                 if (i == 0)
1052                         ufs->same_sb = mnt->mnt_sb;
1053                 else if (ufs->same_sb != mnt->mnt_sb)
1054                         ufs->same_sb = NULL;
1055         }
1056
1057         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1058         if (!ufs->upper_mnt)
1059                 sb->s_flags |= MS_RDONLY;
1060         else if (ufs->upper_mnt->mnt_sb != ufs->same_sb)
1061                 ufs->same_sb = NULL;
1062
1063         if (!(ovl_force_readonly(ufs)) && ufs->config.index) {
1064                 /* Verify lower root is upper root origin */
1065                 err = ovl_verify_origin(upperpath.dentry, ufs->lower_mnt[0],
1066                                         stack[0].dentry, false, true);
1067                 if (err) {
1068                         pr_err("overlayfs: failed to verify upper root origin\n");
1069                         goto out_put_lower_mnt;
1070                 }
1071
1072                 ufs->indexdir = ovl_workdir_create(sb, ufs, workpath.dentry,
1073                                                    OVL_INDEXDIR_NAME, true);
1074                 if (ufs->indexdir) {
1075                         /* Verify upper root is index dir origin */
1076                         err = ovl_verify_origin(ufs->indexdir, ufs->upper_mnt,
1077                                                 upperpath.dentry, true, true);
1078                         if (err)
1079                                 pr_err("overlayfs: failed to verify index dir origin\n");
1080
1081                         /* Cleanup bad/stale/orphan index entries */
1082                         if (!err)
1083                                 err = ovl_indexdir_cleanup(ufs->indexdir,
1084                                                            ufs->upper_mnt,
1085                                                            stack, numlower);
1086                 }
1087                 if (err || !ufs->indexdir)
1088                         pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1089                 if (err)
1090                         goto out_put_indexdir;
1091         }
1092
1093         /* Show index=off/on in /proc/mounts for any of the reasons above */
1094         if (!ufs->indexdir)
1095                 ufs->config.index = false;
1096
1097         if (remote)
1098                 sb->s_d_op = &ovl_reval_dentry_operations;
1099         else
1100                 sb->s_d_op = &ovl_dentry_operations;
1101
1102         err = -ENOMEM;
1103         ufs->creator_cred = cred = prepare_creds();
1104         if (!cred)
1105                 goto out_put_indexdir;
1106
1107         /* Never override disk quota limits or use reserved space */
1108         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1109
1110         err = -ENOMEM;
1111         oe = ovl_alloc_entry(numlower);
1112         if (!oe)
1113                 goto out_put_cred;
1114
1115         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1116         sb->s_op = &ovl_super_operations;
1117         sb->s_xattr = ovl_xattr_handlers;
1118         sb->s_fs_info = ufs;
1119         sb->s_flags |= MS_POSIXACL | MS_NOREMOTELOCK;
1120
1121         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1122         if (!root_dentry)
1123                 goto out_free_oe;
1124
1125         mntput(upperpath.mnt);
1126         for (i = 0; i < numlower; i++)
1127                 mntput(stack[i].mnt);
1128         mntput(workpath.mnt);
1129         kfree(lowertmp);
1130
1131         if (upperpath.dentry) {
1132                 oe->has_upper = true;
1133                 if (ovl_is_impuredir(upperpath.dentry))
1134                         ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1135         }
1136         for (i = 0; i < numlower; i++) {
1137                 oe->lowerstack[i].dentry = stack[i].dentry;
1138                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1139         }
1140         kfree(stack);
1141
1142         root_dentry->d_fsdata = oe;
1143
1144         ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1145                        ovl_dentry_lower(root_dentry));
1146
1147         sb->s_root = root_dentry;
1148
1149         return 0;
1150
1151 out_free_oe:
1152         kfree(oe);
1153 out_put_cred:
1154         put_cred(ufs->creator_cred);
1155 out_put_indexdir:
1156         dput(ufs->indexdir);
1157 out_put_lower_mnt:
1158         for (i = 0; i < ufs->numlower; i++)
1159                 mntput(ufs->lower_mnt[i]);
1160         kfree(ufs->lower_mnt);
1161 out_put_workdir:
1162         dput(ufs->workdir);
1163         mntput(ufs->upper_mnt);
1164 out_put_lowerpath:
1165         for (i = 0; i < numlower; i++)
1166                 path_put(&stack[i]);
1167         kfree(stack);
1168 out_free_lowertmp:
1169         kfree(lowertmp);
1170 out_unlock_workdentry:
1171         if (ufs->workdir_locked)
1172                 ovl_inuse_unlock(workpath.dentry);
1173 out_put_workpath:
1174         path_put(&workpath);
1175 out_unlock_upperdentry:
1176         if (ufs->upperdir_locked)
1177                 ovl_inuse_unlock(upperpath.dentry);
1178 out_put_upperpath:
1179         path_put(&upperpath);
1180 out_free_config:
1181         kfree(ufs->config.lowerdir);
1182         kfree(ufs->config.upperdir);
1183         kfree(ufs->config.workdir);
1184         kfree(ufs);
1185 out:
1186         return err;
1187 }
1188
1189 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1190                                 const char *dev_name, void *raw_data)
1191 {
1192         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1193 }
1194
1195 static struct file_system_type ovl_fs_type = {
1196         .owner          = THIS_MODULE,
1197         .name           = "overlay",
1198         .mount          = ovl_mount,
1199         .kill_sb        = kill_anon_super,
1200 };
1201 MODULE_ALIAS_FS("overlay");
1202
1203 static void ovl_inode_init_once(void *foo)
1204 {
1205         struct ovl_inode *oi = foo;
1206
1207         inode_init_once(&oi->vfs_inode);
1208 }
1209
1210 static int __init ovl_init(void)
1211 {
1212         int err;
1213
1214         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1215                                              sizeof(struct ovl_inode), 0,
1216                                              (SLAB_RECLAIM_ACCOUNT|
1217                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1218                                              ovl_inode_init_once);
1219         if (ovl_inode_cachep == NULL)
1220                 return -ENOMEM;
1221
1222         err = register_filesystem(&ovl_fs_type);
1223         if (err)
1224                 kmem_cache_destroy(ovl_inode_cachep);
1225
1226         return err;
1227 }
1228
1229 static void __exit ovl_exit(void)
1230 {
1231         unregister_filesystem(&ovl_fs_type);
1232
1233         /*
1234          * Make sure all delayed rcu free inodes are flushed before we
1235          * destroy cache.
1236          */
1237         rcu_barrier();
1238         kmem_cache_destroy(ovl_inode_cachep);
1239
1240 }
1241
1242 module_init(ovl_init);
1243 module_exit(ovl_exit);