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 <[email protected]>");
  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         oi->cache = NULL;
 178         oi->redirect = NULL;
 179         oi->version = 0;
 180         oi->flags = 0;
 181         oi->__upperdentry = NULL;
 182         oi->lower = NULL;
 183         mutex_init(&oi->lock);
 184
 185         return &oi->vfs_inode;
 186 }
 187
 188 static void ovl_i_callback(struct rcu_head *head)
 189 {
 190         struct inode *inode = container_of(head, struct inode, i_rcu);
 191
 192         kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
 193 }
 194
 195 static void ovl_destroy_inode(struct inode *inode)
 196 {
 197         struct ovl_inode *oi = OVL_I(inode);
 198
 199         dput(oi->__upperdentry);
 200         kfree(oi->redirect);
 201         ovl_dir_cache_free(inode);
 202         mutex_destroy(&oi->lock);
 203
 204         call_rcu(&inode->i_rcu, ovl_i_callback);
 205 }
 206
 207 static void ovl_put_super(struct super_block *sb)
 208 {
 209         struct ovl_fs *ufs = sb->s_fs_info;
 210         unsigned i;
 211
 212         dput(ufs->indexdir);
 213         dput(ufs->workdir);
 214         ovl_inuse_unlock(ufs->workbasedir);
 215         dput(ufs->workbasedir);
 216         if (ufs->upper_mnt)
 217                 ovl_inuse_unlock(ufs->upper_mnt->mnt_root);
 218         mntput(ufs->upper_mnt);
 219         for (i = 0; i < ufs->numlower; i++)
 220                 mntput(ufs->lower_mnt[i]);
 221         kfree(ufs->lower_mnt);
 222
 223         kfree(ufs->config.lowerdir);
 224         kfree(ufs->config.upperdir);
 225         kfree(ufs->config.workdir);
 226         put_cred(ufs->creator_cred);
 227         kfree(ufs);
 228 }
 229
 230 static int ovl_sync_fs(struct super_block *sb, int wait)
 231 {
 232         struct ovl_fs *ufs = sb->s_fs_info;
 233         struct super_block *upper_sb;
 234         int ret;
 235
 236         if (!ufs->upper_mnt)
 237                 return 0;
 238         upper_sb = ufs->upper_mnt->mnt_sb;
 239         if (!upper_sb->s_op->sync_fs)
 240                 return 0;
 241
 242         /* real inodes have already been synced by sync_filesystem(ovl_sb) */
 243         down_read(&upper_sb->s_umount);
 244         ret = upper_sb->s_op->sync_fs(upper_sb, wait);
 245         up_read(&upper_sb->s_umount);
 246         return ret;
 247 }
 248
 249 /**
 250  * ovl_statfs
 251  * @sb: The overlayfs super block
 252  * @buf: The struct kstatfs to fill in with stats
 253  *
 254  * Get the filesystem statistics.  As writes always target the upper layer
 255  * filesystem pass the statfs to the upper filesystem (if it exists)
 256  */
 257 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 258 {
 259         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 260         struct dentry *root_dentry = dentry->d_sb->s_root;
 261         struct path path;
 262         int err;
 263
 264         ovl_path_real(root_dentry, &path);
 265
 266         err = vfs_statfs(&path, buf);
 267         if (!err) {
 268                 buf->f_namelen = ofs->namelen;
 269                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 270         }
 271
 272         return err;
 273 }
 274
 275 /* Will this overlay be forced to mount/remount ro? */
 276 static bool ovl_force_readonly(struct ovl_fs *ufs)
 277 {
 278         return (!ufs->upper_mnt || !ufs->workdir);
 279 }
 280
 281 /**
 282  * ovl_show_options
 283  *
 284  * Prints the mount options for a given superblock.
 285  * Returns zero; does not fail.
 286  */
 287 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 288 {
 289         struct super_block *sb = dentry->d_sb;
 290         struct ovl_fs *ufs = sb->s_fs_info;
 291
 292         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 293         if (ufs->config.upperdir) {
 294                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 295                 seq_show_option(m, "workdir", ufs->config.workdir);
 296         }
 297         if (ufs->config.default_permissions)
 298                 seq_puts(m, ",default_permissions");
 299         if (ufs->config.redirect_dir != ovl_redirect_dir_def)
 300                 seq_printf(m, ",redirect_dir=%s",
 301                            ufs->config.redirect_dir ? "on" : "off");
 302         if (ufs->config.index != ovl_index_def)
 303                 seq_printf(m, ",index=%s",
 304                            ufs->config.index ? "on" : "off");
 305         return 0;
 306 }
 307
 308 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 309 {
 310         struct ovl_fs *ufs = sb->s_fs_info;
 311
 312         if (!(*flags & MS_RDONLY) && ovl_force_readonly(ufs))
 313                 return -EROFS;
 314
 315         return 0;
 316 }
 317
 318 static const struct super_operations ovl_super_operations = {
 319         .alloc_inode    = ovl_alloc_inode,
 320         .destroy_inode  = ovl_destroy_inode,
 321         .drop_inode     = generic_delete_inode,
 322         .put_super      = ovl_put_super,
 323         .sync_fs        = ovl_sync_fs,
 324         .statfs         = ovl_statfs,
 325         .show_options   = ovl_show_options,
 326         .remount_fs     = ovl_remount,
 327 };
 328
 329 enum {
 330         OPT_LOWERDIR,
 331         OPT_UPPERDIR,
 332         OPT_WORKDIR,
 333         OPT_DEFAULT_PERMISSIONS,
 334         OPT_REDIRECT_DIR_ON,
 335         OPT_REDIRECT_DIR_OFF,
 336         OPT_INDEX_ON,
 337         OPT_INDEX_OFF,
 338         OPT_ERR,
 339 };
 340
 341 static const match_table_t ovl_tokens = {
 342         {OPT_LOWERDIR,                  "lowerdir=%s"},
 343         {OPT_UPPERDIR,                  "upperdir=%s"},
 344         {OPT_WORKDIR,                   "workdir=%s"},
 345         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 346         {OPT_REDIRECT_DIR_ON,           "redirect_dir=on"},
 347         {OPT_REDIRECT_DIR_OFF,          "redirect_dir=off"},
 348         {OPT_INDEX_ON,                  "index=on"},
 349         {OPT_INDEX_OFF,                 "index=off"},
 350         {OPT_ERR,                       NULL}
 351 };
 352
 353 static char *ovl_next_opt(char **s)
 354 {
 355         char *sbegin = *s;
 356         char *p;
 357
 358         if (sbegin == NULL)
 359                 return NULL;
 360
 361         for (p = sbegin; *p; p++) {
 362                 if (*p == '\\') {
 363                         p++;
 364                         if (!*p)
 365                                 break;
 366                 } else if (*p == ',') {
 367                         *p = '\0';
 368                         *s = p + 1;
 369                         return sbegin;
 370                 }
 371         }
 372         *s = NULL;
 373         return sbegin;
 374 }
 375
 376 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 377 {
 378         char *p;
 379
 380         while ((p = ovl_next_opt(&opt)) != NULL) {
 381                 int token;
 382                 substring_t args[MAX_OPT_ARGS];
 383
 384                 if (!*p)
 385                         continue;
 386
 387                 token = match_token(p, ovl_tokens, args);
 388                 switch (token) {
 389                 case OPT_UPPERDIR:
 390                         kfree(config->upperdir);
 391                         config->upperdir = match_strdup(&args[0]);
 392                         if (!config->upperdir)
 393                                 return -ENOMEM;
 394                         break;
 395
 396                 case OPT_LOWERDIR:
 397                         kfree(config->lowerdir);
 398                         config->lowerdir = match_strdup(&args[0]);
 399                         if (!config->lowerdir)
 400                                 return -ENOMEM;
 401                         break;
 402
 403                 case OPT_WORKDIR:
 404                         kfree(config->workdir);
 405                         config->workdir = match_strdup(&args[0]);
 406                         if (!config->workdir)
 407                                 return -ENOMEM;
 408                         break;
 409
 410                 case OPT_DEFAULT_PERMISSIONS:
 411                         config->default_permissions = true;
 412                         break;
 413
 414                 case OPT_REDIRECT_DIR_ON:
 415                         config->redirect_dir = true;
 416                         break;
 417
 418                 case OPT_REDIRECT_DIR_OFF:
 419                         config->redirect_dir = false;
 420                         break;
 421
 422                 case OPT_INDEX_ON:
 423                         config->index = true;
 424                         break;
 425
 426                 case OPT_INDEX_OFF:
 427                         config->index = false;
 428                         break;
 429
 430                 default:
 431                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 432                         return -EINVAL;
 433                 }
 434         }
 435
 436         /* Workdir is useless in non-upper mount */
 437         if (!config->upperdir && config->workdir) {
 438                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 439                         config->workdir);
 440                 kfree(config->workdir);
 441                 config->workdir = NULL;
 442         }
 443
 444         return 0;
 445 }
 446
 447 #define OVL_WORKDIR_NAME "work"
 448 #define OVL_INDEXDIR_NAME "index"
 449
 450 static struct dentry *ovl_workdir_create(struct super_block *sb,
 451                                          struct ovl_fs *ufs,
 452                                          struct dentry *dentry,
 453                                          const char *name, bool persist)
 454 {
 455         struct inode *dir = dentry->d_inode;
 456         struct vfsmount *mnt = ufs->upper_mnt;
 457         struct dentry *work;
 458         int err;
 459         bool retried = false;
 460         bool locked = false;
 461
 462         err = mnt_want_write(mnt);
 463         if (err)
 464                 goto out_err;
 465
 466         inode_lock_nested(dir, I_MUTEX_PARENT);
 467         locked = true;
 468
 469 retry:
 470         work = lookup_one_len(name, dentry, strlen(name));
 471
 472         if (!IS_ERR(work)) {
 473                 struct iattr attr = {
 474                         .ia_valid = ATTR_MODE,
 475                         .ia_mode = S_IFDIR | 0,
 476                 };
 477
 478                 if (work->d_inode) {
 479                         err = -EEXIST;
 480                         if (retried)
 481                                 goto out_dput;
 482
 483                         if (persist)
 484                                 goto out_unlock;
 485
 486                         retried = true;
 487                         ovl_workdir_cleanup(dir, mnt, work, 0);
 488                         dput(work);
 489                         goto retry;
 490                 }
 491
 492                 err = ovl_create_real(dir, work,
 493                                       &(struct cattr){.mode = S_IFDIR | 0},
 494                                       NULL, true);
 495                 if (err)
 496                         goto out_dput;
 497
 498                 /*
 499                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 500                  *
 501                  * a) success (there was a POSIX ACL xattr and was removed)
 502                  * b) -ENODATA (there was no POSIX ACL xattr)
 503                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 504                  *
 505                  * There are various other error values that could effectively
 506                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 507                  * if the xattr name is too long), but the set of filesystems
 508                  * allowed as upper are limited to "normal" ones, where checking
 509                  * for the above two errors is sufficient.
 510                  */
 511                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 512                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 513                         goto out_dput;
 514
 515                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 516                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 517                         goto out_dput;
 518
 519                 /* Clear any inherited mode bits */
 520                 inode_lock(work->d_inode);
 521                 err = notify_change(work, &attr, NULL);
 522                 inode_unlock(work->d_inode);
 523                 if (err)
 524                         goto out_dput;
 525         } else {
 526                 err = PTR_ERR(work);
 527                 goto out_err;
 528         }
 529 out_unlock:
 530         mnt_drop_write(mnt);
 531         if (locked)
 532                 inode_unlock(dir);
 533
 534         return work;
 535
 536 out_dput:
 537         dput(work);
 538 out_err:
 539         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 540                 ufs->config.workdir, name, -err);
 541         sb->s_flags |= MS_RDONLY;
 542         work = NULL;
 543         goto out_unlock;
 544 }
 545
 546 static void ovl_unescape(char *s)
 547 {
 548         char *d = s;
 549
 550         for (;; s++, d++) {
 551                 if (*s == '\\')
 552                         s++;
 553                 *d = *s;
 554                 if (!*s)
 555                         break;
 556         }
 557 }
 558
 559 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 560 {
 561         int err = -EINVAL;
 562
 563         if (!*name) {
 564                 pr_err("overlayfs: empty lowerdir\n");
 565                 goto out;
 566         }
 567         err = kern_path(name, LOOKUP_FOLLOW, path);
 568         if (err) {
 569                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 570                 goto out;
 571         }
 572         err = -EINVAL;
 573         if (ovl_dentry_weird(path->dentry)) {
 574                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 575                 goto out_put;
 576         }
 577         if (!d_is_dir(path->dentry)) {
 578                 pr_err("overlayfs: '%s' not a directory\n", name);
 579                 goto out_put;
 580         }
 581         return 0;
 582
 583 out_put:
 584         path_put(path);
 585 out:
 586         return err;
 587 }
 588
 589 static int ovl_mount_dir(const char *name, struct path *path)
 590 {
 591         int err = -ENOMEM;
 592         char *tmp = kstrdup(name, GFP_KERNEL);
 593
 594         if (tmp) {
 595                 ovl_unescape(tmp);
 596                 err = ovl_mount_dir_noesc(tmp, path);
 597
 598                 if (!err)
 599                         if (ovl_dentry_remote(path->dentry)) {
 600                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 601                                        tmp);
 602                                 path_put(path);
 603                                 err = -EINVAL;
 604                         }
 605                 kfree(tmp);
 606         }
 607         return err;
 608 }
 609
 610 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 611                              const char *name)
 612 {
 613         struct kstatfs statfs;
 614         int err = vfs_statfs(path, &statfs);
 615
 616         if (err)
 617                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 618         else
 619                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 620
 621         return err;
 622 }
 623
 624 static int ovl_lower_dir(const char *name, struct path *path,
 625                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 626 {
 627         int err;
 628
 629         err = ovl_mount_dir_noesc(name, path);
 630         if (err)
 631                 goto out;
 632
 633         err = ovl_check_namelen(path, ofs, name);
 634         if (err)
 635                 goto out_put;
 636
 637         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 638
 639         if (ovl_dentry_remote(path->dentry))
 640                 *remote = true;
 641
 642         /*
 643          * The inodes index feature needs to encode and decode file
 644          * handles, so it requires that all layers support them.
 645          */
 646         if (ofs->config.index && !ovl_can_decode_fh(path->dentry->d_sb)) {
 647                 ofs->config.index = false;
 648                 pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off.\n", name);
 649         }
 650
 651         return 0;
 652
 653 out_put:
 654         path_put(path);
 655 out:
 656         return err;
 657 }
 658
 659 /* Workdir should not be subdir of upperdir and vice versa */
 660 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 661 {
 662         bool ok = false;
 663
 664         if (workdir != upperdir) {
 665                 ok = (lock_rename(workdir, upperdir) == NULL);
 666                 unlock_rename(workdir, upperdir);
 667         }
 668         return ok;
 669 }
 670
 671 static unsigned int ovl_split_lowerdirs(char *str)
 672 {
 673         unsigned int ctr = 1;
 674         char *s, *d;
 675
 676         for (s = d = str;; s++, d++) {
 677                 if (*s == '\\') {
 678                         s++;
 679                 } else if (*s == ':') {
 680                         *d = '\0';
 681                         ctr++;
 682                         continue;
 683                 }
 684                 *d = *s;
 685                 if (!*s)
 686                         break;
 687         }
 688         return ctr;
 689 }
 690
 691 static int __maybe_unused
 692 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 693                         struct dentry *dentry, struct inode *inode,
 694                         const char *name, void *buffer, size_t size)
 695 {
 696         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 697 }
 698
 699 static int __maybe_unused
 700 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 701                         struct dentry *dentry, struct inode *inode,
 702                         const char *name, const void *value,
 703                         size_t size, int flags)
 704 {
 705         struct dentry *workdir = ovl_workdir(dentry);
 706         struct inode *realinode = ovl_inode_real(inode);
 707         struct posix_acl *acl = NULL;
 708         int err;
 709
 710         /* Check that everything is OK before copy-up */
 711         if (value) {
 712                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 713                 if (IS_ERR(acl))
 714                         return PTR_ERR(acl);
 715         }
 716         err = -EOPNOTSUPP;
 717         if (!IS_POSIXACL(d_inode(workdir)))
 718                 goto out_acl_release;
 719         if (!realinode->i_op->set_acl)
 720                 goto out_acl_release;
 721         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 722                 err = acl ? -EACCES : 0;
 723                 goto out_acl_release;
 724         }
 725         err = -EPERM;
 726         if (!inode_owner_or_capable(inode))
 727                 goto out_acl_release;
 728
 729         posix_acl_release(acl);
 730
 731         /*
 732          * Check if sgid bit needs to be cleared (actual setacl operation will
 733          * be done with mounter's capabilities and so that won't do it for us).
 734          */
 735         if (unlikely(inode->i_mode & S_ISGID) &&
 736             handler->flags == ACL_TYPE_ACCESS &&
 737             !in_group_p(inode->i_gid) &&
 738             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 739                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 740
 741                 err = ovl_setattr(dentry, &iattr);
 742                 if (err)
 743                         return err;
 744         }
 745
 746         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
 747         if (!err)
 748                 ovl_copyattr(ovl_inode_real(inode), inode);
 749
 750         return err;
 751
 752 out_acl_release:
 753         posix_acl_release(acl);
 754         return err;
 755 }
 756
 757 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 758                              struct dentry *dentry, struct inode *inode,
 759                              const char *name, void *buffer, size_t size)
 760 {
 761         return -EOPNOTSUPP;
 762 }
 763
 764 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 765                              struct dentry *dentry, struct inode *inode,
 766                              const char *name, const void *value,
 767                              size_t size, int flags)
 768 {
 769         return -EOPNOTSUPP;
 770 }
 771
 772 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 773                                struct dentry *dentry, struct inode *inode,
 774                                const char *name, void *buffer, size_t size)
 775 {
 776         return ovl_xattr_get(dentry, inode, name, buffer, size);
 777 }
 778
 779 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 780                                struct dentry *dentry, struct inode *inode,
 781                                const char *name, const void *value,
 782                                size_t size, int flags)
 783 {
 784         return ovl_xattr_set(dentry, inode, name, value, size, flags);
 785 }
 786
 787 static const struct xattr_handler __maybe_unused
 788 ovl_posix_acl_access_xattr_handler = {
 789         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 790         .flags = ACL_TYPE_ACCESS,
 791         .get = ovl_posix_acl_xattr_get,
 792         .set = ovl_posix_acl_xattr_set,
 793 };
 794
 795 static const struct xattr_handler __maybe_unused
 796 ovl_posix_acl_default_xattr_handler = {
 797         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 798         .flags = ACL_TYPE_DEFAULT,
 799         .get = ovl_posix_acl_xattr_get,
 800         .set = ovl_posix_acl_xattr_set,
 801 };
 802
 803 static const struct xattr_handler ovl_own_xattr_handler = {
 804         .prefix = OVL_XATTR_PREFIX,
 805         .get = ovl_own_xattr_get,
 806         .set = ovl_own_xattr_set,
 807 };
 808
 809 static const struct xattr_handler ovl_other_xattr_handler = {
 810         .prefix = "", /* catch all */
 811         .get = ovl_other_xattr_get,
 812         .set = ovl_other_xattr_set,
 813 };
 814
 815 static const struct xattr_handler *ovl_xattr_handlers[] = {
 816 #ifdef CONFIG_FS_POSIX_ACL
 817         &ovl_posix_acl_access_xattr_handler,
 818         &ovl_posix_acl_default_xattr_handler,
 819 #endif
 820         &ovl_own_xattr_handler,
 821         &ovl_other_xattr_handler,
 822         NULL
 823 };
 824
 825 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 826 {
 827         struct path upperpath = { };
 828         struct path workpath = { };
 829         struct dentry *root_dentry;
 830         struct ovl_entry *oe;
 831         struct ovl_fs *ufs;
 832         struct path *stack = NULL;
 833         char *lowertmp;
 834         char *lower;
 835         unsigned int numlower;
 836         unsigned int stacklen = 0;
 837         unsigned int i;
 838         bool remote = false;
 839         struct cred *cred;
 840         int err;
 841
 842         err = -ENOMEM;
 843         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 844         if (!ufs)
 845                 goto out;
 846
 847         ufs->config.redirect_dir = ovl_redirect_dir_def;
 848         ufs->config.index = ovl_index_def;
 849         err = ovl_parse_opt((char *) data, &ufs->config);
 850         if (err)
 851                 goto out_free_config;
 852
 853         err = -EINVAL;
 854         if (!ufs->config.lowerdir) {
 855                 if (!silent)
 856                         pr_err("overlayfs: missing 'lowerdir'\n");
 857                 goto out_free_config;
 858         }
 859
 860         sb->s_stack_depth = 0;
 861         sb->s_maxbytes = MAX_LFS_FILESIZE;
 862         if (ufs->config.upperdir) {
 863                 if (!ufs->config.workdir) {
 864                         pr_err("overlayfs: missing 'workdir'\n");
 865                         goto out_free_config;
 866                 }
 867
 868                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 869                 if (err)
 870                         goto out_free_config;
 871
 872                 /* Upper fs should not be r/o */
 873                 if (sb_rdonly(upperpath.mnt->mnt_sb)) {
 874                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 875                         err = -EINVAL;
 876                         goto out_put_upperpath;
 877                 }
 878
 879                 err = ovl_check_namelen(&upperpath, ufs, ufs->config.upperdir);
 880                 if (err)
 881                         goto out_put_upperpath;
 882
 883                 err = -EBUSY;
 884                 if (!ovl_inuse_trylock(upperpath.dentry)) {
 885                         pr_err("overlayfs: upperdir is in-use by another mount\n");
 886                         goto out_put_upperpath;
 887                 }
 888
 889                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 890                 if (err)
 891                         goto out_unlock_upperdentry;
 892
 893                 err = -EINVAL;
 894                 if (upperpath.mnt != workpath.mnt) {
 895                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 896                         goto out_put_workpath;
 897                 }
 898                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 899                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 900                         goto out_put_workpath;
 901                 }
 902
 903                 err = -EBUSY;
 904                 if (!ovl_inuse_trylock(workpath.dentry)) {
 905                         pr_err("overlayfs: workdir is in-use by another mount\n");
 906                         goto out_put_workpath;
 907                 }
 908
 909                 ufs->workbasedir = workpath.dentry;
 910                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 911         }
 912         err = -ENOMEM;
 913         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 914         if (!lowertmp)
 915                 goto out_unlock_workdentry;
 916
 917         err = -EINVAL;
 918         stacklen = ovl_split_lowerdirs(lowertmp);
 919         if (stacklen > OVL_MAX_STACK) {
 920                 pr_err("overlayfs: too many lower directories, limit is %d\n",
 921                        OVL_MAX_STACK);
 922                 goto out_free_lowertmp;
 923         } else if (!ufs->config.upperdir && stacklen == 1) {
 924                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 925                 goto out_free_lowertmp;
 926         }
 927
 928         err = -ENOMEM;
 929         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 930         if (!stack)
 931                 goto out_free_lowertmp;
 932
 933         err = -EINVAL;
 934         lower = lowertmp;
 935         for (numlower = 0; numlower < stacklen; numlower++) {
 936                 err = ovl_lower_dir(lower, &stack[numlower], ufs,
 937                                     &sb->s_stack_depth, &remote);
 938                 if (err)
 939                         goto out_put_lowerpath;
 940
 941                 lower = strchr(lower, '\0') + 1;
 942         }
 943
 944         err = -EINVAL;
 945         sb->s_stack_depth++;
 946         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
 947                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
 948                 goto out_put_lowerpath;
 949         }
 950
 951         if (ufs->config.upperdir) {
 952                 ufs->upper_mnt = clone_private_mount(&upperpath);
 953                 err = PTR_ERR(ufs->upper_mnt);
 954                 if (IS_ERR(ufs->upper_mnt)) {
 955                         pr_err("overlayfs: failed to clone upperpath\n");
 956                         goto out_put_lowerpath;
 957                 }
 958
 959                 /* Don't inherit atime flags */
 960                 ufs->upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
 961
 962                 sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
 963
 964                 ufs->workdir = ovl_workdir_create(sb, ufs, workpath.dentry,
 965                                                   OVL_WORKDIR_NAME, false);
 966                 /*
 967                  * Upper should support d_type, else whiteouts are visible.
 968                  * Given workdir and upper are on same fs, we can do
 969                  * iterate_dir() on workdir. This check requires successful
 970                  * creation of workdir in previous step.
 971                  */
 972                 if (ufs->workdir) {
 973                         struct dentry *temp;
 974
 975                         err = ovl_check_d_type_supported(&workpath);
 976                         if (err < 0)
 977                                 goto out_put_workdir;
 978
 979                         /*
 980                          * We allowed this configuration and don't want to
 981                          * break users over kernel upgrade. So warn instead
 982                          * of erroring out.
 983                          */
 984                         if (!err)
 985                                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
 986
 987                         /* Check if upper/work fs supports O_TMPFILE */
 988                         temp = ovl_do_tmpfile(ufs->workdir, S_IFREG | 0);
 989                         ufs->tmpfile = !IS_ERR(temp);
 990                         if (ufs->tmpfile)
 991                                 dput(temp);
 992                         else
 993                                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
 994
 995                         /*
 996                          * Check if upper/work fs supports trusted.overlay.*
 997                          * xattr
 998                          */
 999                         err = ovl_do_setxattr(ufs->workdir, OVL_XATTR_OPAQUE,
1000                                               "0", 1, 0);
1001                         if (err) {
1002                                 ufs->noxattr = true;
1003                                 pr_warn("overlayfs: upper fs does not support xattr.\n");
1004                         } else {
1005                                 vfs_removexattr(ufs->workdir, OVL_XATTR_OPAQUE);
1006                         }
1007
1008                         /* Check if upper/work fs supports file handles */
1009                         if (ufs->config.index &&
1010                             !ovl_can_decode_fh(ufs->workdir->d_sb)) {
1011                                 ufs->config.index = false;
1012                                 pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1013                         }
1014                 }
1015         }
1016
1017         err = -ENOMEM;
1018         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1019         if (ufs->lower_mnt == NULL)
1020                 goto out_put_workdir;
1021         for (i = 0; i < numlower; i++) {
1022                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1023
1024                 err = PTR_ERR(mnt);
1025                 if (IS_ERR(mnt)) {
1026                         pr_err("overlayfs: failed to clone lowerpath\n");
1027                         goto out_put_lower_mnt;
1028                 }
1029                 /*
1030                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1031                  * will fail instead of modifying lower fs.
1032                  */
1033                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1034
1035                 ufs->lower_mnt[ufs->numlower] = mnt;
1036                 ufs->numlower++;
1037
1038                 /* Check if all lower layers are on same sb */
1039                 if (i == 0)
1040                         ufs->same_sb = mnt->mnt_sb;
1041                 else if (ufs->same_sb != mnt->mnt_sb)
1042                         ufs->same_sb = NULL;
1043         }
1044
1045         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1046         if (!ufs->upper_mnt)
1047                 sb->s_flags |= MS_RDONLY;
1048         else if (ufs->upper_mnt->mnt_sb != ufs->same_sb)
1049                 ufs->same_sb = NULL;
1050
1051         if (!(ovl_force_readonly(ufs)) && ufs->config.index) {
1052                 /* Verify lower root is upper root origin */
1053                 err = ovl_verify_origin(upperpath.dentry, ufs->lower_mnt[0],
1054                                         stack[0].dentry, false, true);
1055                 if (err) {
1056                         pr_err("overlayfs: failed to verify upper root origin\n");
1057                         goto out_put_lower_mnt;
1058                 }
1059
1060                 ufs->indexdir = ovl_workdir_create(sb, ufs, workpath.dentry,
1061                                                    OVL_INDEXDIR_NAME, true);
1062                 if (ufs->indexdir) {
1063                         /* Verify upper root is index dir origin */
1064                         err = ovl_verify_origin(ufs->indexdir, ufs->upper_mnt,
1065                                                 upperpath.dentry, true, true);
1066                         if (err)
1067                                 pr_err("overlayfs: failed to verify index dir origin\n");
1068
1069                         /* Cleanup bad/stale/orphan index entries */
1070                         if (!err)
1071                                 err = ovl_indexdir_cleanup(ufs->indexdir,
1072                                                            ufs->upper_mnt,
1073                                                            stack, numlower);
1074                 }
1075                 if (err || !ufs->indexdir)
1076                         pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1077                 if (err)
1078                         goto out_put_indexdir;
1079         }
1080
1081         /* Show index=off/on in /proc/mounts for any of the reasons above */
1082         if (!ufs->indexdir)
1083                 ufs->config.index = false;
1084
1085         if (remote)
1086                 sb->s_d_op = &ovl_reval_dentry_operations;
1087         else
1088                 sb->s_d_op = &ovl_dentry_operations;
1089
1090         err = -ENOMEM;
1091         ufs->creator_cred = cred = prepare_creds();
1092         if (!cred)
1093                 goto out_put_indexdir;
1094
1095         /* Never override disk quota limits or use reserved space */
1096         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1097
1098         err = -ENOMEM;
1099         oe = ovl_alloc_entry(numlower);
1100         if (!oe)
1101                 goto out_put_cred;
1102
1103         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1104         sb->s_op = &ovl_super_operations;
1105         sb->s_xattr = ovl_xattr_handlers;
1106         sb->s_fs_info = ufs;
1107         sb->s_flags |= MS_POSIXACL | MS_NOREMOTELOCK;
1108
1109         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1110         if (!root_dentry)
1111                 goto out_free_oe;
1112
1113         mntput(upperpath.mnt);
1114         for (i = 0; i < numlower; i++)
1115                 mntput(stack[i].mnt);
1116         mntput(workpath.mnt);
1117         kfree(lowertmp);
1118
1119         if (upperpath.dentry) {
1120                 oe->has_upper = true;
1121                 if (ovl_is_impuredir(upperpath.dentry))
1122                         ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1123         }
1124         for (i = 0; i < numlower; i++) {
1125                 oe->lowerstack[i].dentry = stack[i].dentry;
1126                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1127         }
1128         kfree(stack);
1129
1130         root_dentry->d_fsdata = oe;
1131
1132         ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1133                        ovl_dentry_lower(root_dentry));
1134
1135         sb->s_root = root_dentry;
1136
1137         return 0;
1138
1139 out_free_oe:
1140         kfree(oe);
1141 out_put_cred:
1142         put_cred(ufs->creator_cred);
1143 out_put_indexdir:
1144         dput(ufs->indexdir);
1145 out_put_lower_mnt:
1146         for (i = 0; i < ufs->numlower; i++)
1147                 mntput(ufs->lower_mnt[i]);
1148         kfree(ufs->lower_mnt);
1149 out_put_workdir:
1150         dput(ufs->workdir);
1151         mntput(ufs->upper_mnt);
1152 out_put_lowerpath:
1153         for (i = 0; i < numlower; i++)
1154                 path_put(&stack[i]);
1155         kfree(stack);
1156 out_free_lowertmp:
1157         kfree(lowertmp);
1158 out_unlock_workdentry:
1159         ovl_inuse_unlock(workpath.dentry);
1160 out_put_workpath:
1161         path_put(&workpath);
1162 out_unlock_upperdentry:
1163         ovl_inuse_unlock(upperpath.dentry);
1164 out_put_upperpath:
1165         path_put(&upperpath);
1166 out_free_config:
1167         kfree(ufs->config.lowerdir);
1168         kfree(ufs->config.upperdir);
1169         kfree(ufs->config.workdir);
1170         kfree(ufs);
1171 out:
1172         return err;
1173 }
1174
1175 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1176                                 const char *dev_name, void *raw_data)
1177 {
1178         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1179 }
1180
1181 static struct file_system_type ovl_fs_type = {
1182         .owner          = THIS_MODULE,
1183         .name           = "overlay",
1184         .mount          = ovl_mount,
1185         .kill_sb        = kill_anon_super,
1186 };
1187 MODULE_ALIAS_FS("overlay");
1188
1189 static void ovl_inode_init_once(void *foo)
1190 {
1191         struct ovl_inode *oi = foo;
1192
1193         inode_init_once(&oi->vfs_inode);
1194 }
1195
1196 static int __init ovl_init(void)
1197 {
1198         int err;
1199
1200         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1201                                              sizeof(struct ovl_inode), 0,
1202                                              (SLAB_RECLAIM_ACCOUNT|
1203                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1204                                              ovl_inode_init_once);
1205         if (ovl_inode_cachep == NULL)
1206                 return -ENOMEM;
1207
1208         err = register_filesystem(&ovl_fs_type);
1209         if (err)
1210                 kmem_cache_destroy(ovl_inode_cachep);
1211
1212         return err;
1213 }
1214
1215 static void __exit ovl_exit(void)
1216 {
1217         unregister_filesystem(&ovl_fs_type);
1218
1219         /*
1220          * Make sure all delayed rcu free inodes are flushed before we
1221          * destroy cache.
1222          */
1223         rcu_barrier();
1224         kmem_cache_destroy(ovl_inode_cachep);
1225
1226 }
1227
1228 module_init(ovl_init);
1229 module_exit(ovl_exit);