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
  22 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  23 MODULE_DESCRIPTION("Overlay filesystem");
  24 MODULE_LICENSE("GPL");
  25
  26
  27 struct ovl_dir_cache;
  28
  29 #define OVL_MAX_STACK 500
  30
  31 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  32 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  33 MODULE_PARM_DESC(ovl_redirect_dir_def,
  34                  "Default to on or off for the redirect_dir feature");
  35
  36 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
  37 module_param_named(index, ovl_index_def, bool, 0644);
  38 MODULE_PARM_DESC(ovl_index_def,
  39                  "Default to on or off for the inodes index feature");
  40
  41 static void ovl_dentry_release(struct dentry *dentry)
  42 {
  43         struct ovl_entry *oe = dentry->d_fsdata;
  44
  45         if (oe) {
  46                 unsigned int i;
  47
  48                 for (i = 0; i < oe->numlower; i++)
  49                         dput(oe->lowerstack[i].dentry);
  50                 kfree_rcu(oe, rcu);
  51         }
  52 }
  53
  54 static int ovl_check_append_only(struct inode *inode, int flag)
  55 {
  56         /*
  57          * This test was moot in vfs may_open() because overlay inode does
  58          * not have the S_APPEND flag, so re-check on real upper inode
  59          */
  60         if (IS_APPEND(inode)) {
  61                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
  62                         return -EPERM;
  63                 if (flag & O_TRUNC)
  64                         return -EPERM;
  65         }
  66
  67         return 0;
  68 }
  69
  70 static struct dentry *ovl_d_real(struct dentry *dentry,
  71                                  const struct inode *inode,
  72                                  unsigned int open_flags, unsigned int flags)
  73 {
  74         struct dentry *real;
  75         int err;
  76
  77         if (flags & D_REAL_UPPER)
  78                 return ovl_dentry_upper(dentry);
  79
  80         if (!d_is_reg(dentry)) {
  81                 if (!inode || inode == d_inode(dentry))
  82                         return dentry;
  83                 goto bug;
  84         }
  85
  86         if (open_flags) {
  87                 err = ovl_open_maybe_copy_up(dentry, open_flags);
  88                 if (err)
  89                         return ERR_PTR(err);
  90         }
  91
  92         real = ovl_dentry_upper(dentry);
  93         if (real && (!inode || inode == d_inode(real))) {
  94                 if (!inode) {
  95                         err = ovl_check_append_only(d_inode(real), open_flags);
  96                         if (err)
  97                                 return ERR_PTR(err);
  98                 }
  99                 return real;
 100         }
 101
 102         real = ovl_dentry_lower(dentry);
 103         if (!real)
 104                 goto bug;
 105
 106         /* Handle recursion */
 107         real = d_real(real, inode, open_flags, 0);
 108
 109         if (!inode || inode == d_inode(real))
 110                 return real;
 111 bug:
 112         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
 113              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 114         return dentry;
 115 }
 116
 117 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 118 {
 119         struct ovl_entry *oe = dentry->d_fsdata;
 120         unsigned int i;
 121         int ret = 1;
 122
 123         for (i = 0; i < oe->numlower; i++) {
 124                 struct dentry *d = oe->lowerstack[i].dentry;
 125
 126                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 127                         ret = d->d_op->d_revalidate(d, flags);
 128                         if (ret < 0)
 129                                 return ret;
 130                         if (!ret) {
 131                                 if (!(flags & LOOKUP_RCU))
 132                                         d_invalidate(d);
 133                                 return -ESTALE;
 134                         }
 135                 }
 136         }
 137         return 1;
 138 }
 139
 140 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 141 {
 142         struct ovl_entry *oe = dentry->d_fsdata;
 143         unsigned int i;
 144         int ret = 1;
 145
 146         for (i = 0; i < oe->numlower; i++) {
 147                 struct dentry *d = oe->lowerstack[i].dentry;
 148
 149                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 150                         ret = d->d_op->d_weak_revalidate(d, flags);
 151                         if (ret <= 0)
 152                                 break;
 153                 }
 154         }
 155         return ret;
 156 }
 157
 158 static const struct dentry_operations ovl_dentry_operations = {
 159         .d_release = ovl_dentry_release,
 160         .d_real = ovl_d_real,
 161 };
 162
 163 static const struct dentry_operations ovl_reval_dentry_operations = {
 164         .d_release = ovl_dentry_release,
 165         .d_real = ovl_d_real,
 166         .d_revalidate = ovl_dentry_revalidate,
 167         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 168 };
 169
 170 static struct kmem_cache *ovl_inode_cachep;
 171
 172 static struct inode *ovl_alloc_inode(struct super_block *sb)
 173 {
 174         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 175
 176         if (!oi)
 177                 return NULL;
 178
 179         oi->cache = NULL;
 180         oi->redirect = NULL;
 181         oi->version = 0;
 182         oi->flags = 0;
 183         oi->__upperdentry = NULL;
 184         oi->lower = NULL;
 185         mutex_init(&oi->lock);
 186
 187         return &oi->vfs_inode;
 188 }
 189
 190 static void ovl_i_callback(struct rcu_head *head)
 191 {
 192         struct inode *inode = container_of(head, struct inode, i_rcu);
 193
 194         kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
 195 }
 196
 197 static void ovl_destroy_inode(struct inode *inode)
 198 {
 199         struct ovl_inode *oi = OVL_I(inode);
 200
 201         dput(oi->__upperdentry);
 202         kfree(oi->redirect);
 203         ovl_dir_cache_free(inode);
 204         mutex_destroy(&oi->lock);
 205
 206         call_rcu(&inode->i_rcu, ovl_i_callback);
 207 }
 208
 209 static void ovl_free_fs(struct ovl_fs *ufs)
 210 {
 211         unsigned i;
 212
 213         dput(ufs->indexdir);
 214         dput(ufs->workdir);
 215         if (ufs->workdir_locked)
 216                 ovl_inuse_unlock(ufs->workbasedir);
 217         dput(ufs->workbasedir);
 218         if (ufs->upperdir_locked)
 219                 ovl_inuse_unlock(ufs->upper_mnt->mnt_root);
 220         mntput(ufs->upper_mnt);
 221         for (i = 0; i < ufs->numlower; i++) {
 222                 mntput(ufs->lower_layers[i].mnt);
 223                 free_anon_bdev(ufs->lower_layers[i].pseudo_dev);
 224         }
 225         kfree(ufs->lower_layers);
 226
 227         kfree(ufs->config.lowerdir);
 228         kfree(ufs->config.upperdir);
 229         kfree(ufs->config.workdir);
 230         if (ufs->creator_cred)
 231                 put_cred(ufs->creator_cred);
 232         kfree(ufs);
 233 }
 234
 235 static void ovl_put_super(struct super_block *sb)
 236 {
 237         struct ovl_fs *ofs = sb->s_fs_info;
 238
 239         ovl_free_fs(ofs);
 240 }
 241
 242 static int ovl_sync_fs(struct super_block *sb, int wait)
 243 {
 244         struct ovl_fs *ufs = sb->s_fs_info;
 245         struct super_block *upper_sb;
 246         int ret;
 247
 248         if (!ufs->upper_mnt)
 249                 return 0;
 250         upper_sb = ufs->upper_mnt->mnt_sb;
 251         if (!upper_sb->s_op->sync_fs)
 252                 return 0;
 253
 254         /* real inodes have already been synced by sync_filesystem(ovl_sb) */
 255         down_read(&upper_sb->s_umount);
 256         ret = upper_sb->s_op->sync_fs(upper_sb, wait);
 257         up_read(&upper_sb->s_umount);
 258         return ret;
 259 }
 260
 261 /**
 262  * ovl_statfs
 263  * @sb: The overlayfs super block
 264  * @buf: The struct kstatfs to fill in with stats
 265  *
 266  * Get the filesystem statistics.  As writes always target the upper layer
 267  * filesystem pass the statfs to the upper filesystem (if it exists)
 268  */
 269 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 270 {
 271         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 272         struct dentry *root_dentry = dentry->d_sb->s_root;
 273         struct path path;
 274         int err;
 275
 276         ovl_path_real(root_dentry, &path);
 277
 278         err = vfs_statfs(&path, buf);
 279         if (!err) {
 280                 buf->f_namelen = ofs->namelen;
 281                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 282         }
 283
 284         return err;
 285 }
 286
 287 /* Will this overlay be forced to mount/remount ro? */
 288 static bool ovl_force_readonly(struct ovl_fs *ufs)
 289 {
 290         return (!ufs->upper_mnt || !ufs->workdir);
 291 }
 292
 293 /**
 294  * ovl_show_options
 295  *
 296  * Prints the mount options for a given superblock.
 297  * Returns zero; does not fail.
 298  */
 299 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 300 {
 301         struct super_block *sb = dentry->d_sb;
 302         struct ovl_fs *ufs = sb->s_fs_info;
 303
 304         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 305         if (ufs->config.upperdir) {
 306                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 307                 seq_show_option(m, "workdir", ufs->config.workdir);
 308         }
 309         if (ufs->config.default_permissions)
 310                 seq_puts(m, ",default_permissions");
 311         if (ufs->config.redirect_dir != ovl_redirect_dir_def)
 312                 seq_printf(m, ",redirect_dir=%s",
 313                            ufs->config.redirect_dir ? "on" : "off");
 314         if (ufs->config.index != ovl_index_def)
 315                 seq_printf(m, ",index=%s",
 316                            ufs->config.index ? "on" : "off");
 317         return 0;
 318 }
 319
 320 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 321 {
 322         struct ovl_fs *ufs = sb->s_fs_info;
 323
 324         if (!(*flags & MS_RDONLY) && ovl_force_readonly(ufs))
 325                 return -EROFS;
 326
 327         return 0;
 328 }
 329
 330 static const struct super_operations ovl_super_operations = {
 331         .alloc_inode    = ovl_alloc_inode,
 332         .destroy_inode  = ovl_destroy_inode,
 333         .drop_inode     = generic_delete_inode,
 334         .put_super      = ovl_put_super,
 335         .sync_fs        = ovl_sync_fs,
 336         .statfs         = ovl_statfs,
 337         .show_options   = ovl_show_options,
 338         .remount_fs     = ovl_remount,
 339 };
 340
 341 enum {
 342         OPT_LOWERDIR,
 343         OPT_UPPERDIR,
 344         OPT_WORKDIR,
 345         OPT_DEFAULT_PERMISSIONS,
 346         OPT_REDIRECT_DIR_ON,
 347         OPT_REDIRECT_DIR_OFF,
 348         OPT_INDEX_ON,
 349         OPT_INDEX_OFF,
 350         OPT_ERR,
 351 };
 352
 353 static const match_table_t ovl_tokens = {
 354         {OPT_LOWERDIR,                  "lowerdir=%s"},
 355         {OPT_UPPERDIR,                  "upperdir=%s"},
 356         {OPT_WORKDIR,                   "workdir=%s"},
 357         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 358         {OPT_REDIRECT_DIR_ON,           "redirect_dir=on"},
 359         {OPT_REDIRECT_DIR_OFF,          "redirect_dir=off"},
 360         {OPT_INDEX_ON,                  "index=on"},
 361         {OPT_INDEX_OFF,                 "index=off"},
 362         {OPT_ERR,                       NULL}
 363 };
 364
 365 static char *ovl_next_opt(char **s)
 366 {
 367         char *sbegin = *s;
 368         char *p;
 369
 370         if (sbegin == NULL)
 371                 return NULL;
 372
 373         for (p = sbegin; *p; p++) {
 374                 if (*p == '\\') {
 375                         p++;
 376                         if (!*p)
 377                                 break;
 378                 } else if (*p == ',') {
 379                         *p = '\0';
 380                         *s = p + 1;
 381                         return sbegin;
 382                 }
 383         }
 384         *s = NULL;
 385         return sbegin;
 386 }
 387
 388 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 389 {
 390         char *p;
 391
 392         while ((p = ovl_next_opt(&opt)) != NULL) {
 393                 int token;
 394                 substring_t args[MAX_OPT_ARGS];
 395
 396                 if (!*p)
 397                         continue;
 398
 399                 token = match_token(p, ovl_tokens, args);
 400                 switch (token) {
 401                 case OPT_UPPERDIR:
 402                         kfree(config->upperdir);
 403                         config->upperdir = match_strdup(&args[0]);
 404                         if (!config->upperdir)
 405                                 return -ENOMEM;
 406                         break;
 407
 408                 case OPT_LOWERDIR:
 409                         kfree(config->lowerdir);
 410                         config->lowerdir = match_strdup(&args[0]);
 411                         if (!config->lowerdir)
 412                                 return -ENOMEM;
 413                         break;
 414
 415                 case OPT_WORKDIR:
 416                         kfree(config->workdir);
 417                         config->workdir = match_strdup(&args[0]);
 418                         if (!config->workdir)
 419                                 return -ENOMEM;
 420                         break;
 421
 422                 case OPT_DEFAULT_PERMISSIONS:
 423                         config->default_permissions = true;
 424                         break;
 425
 426                 case OPT_REDIRECT_DIR_ON:
 427                         config->redirect_dir = true;
 428                         break;
 429
 430                 case OPT_REDIRECT_DIR_OFF:
 431                         config->redirect_dir = false;
 432                         break;
 433
 434                 case OPT_INDEX_ON:
 435                         config->index = true;
 436                         break;
 437
 438                 case OPT_INDEX_OFF:
 439                         config->index = false;
 440                         break;
 441
 442                 default:
 443                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 444                         return -EINVAL;
 445                 }
 446         }
 447
 448         /* Workdir is useless in non-upper mount */
 449         if (!config->upperdir && config->workdir) {
 450                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 451                         config->workdir);
 452                 kfree(config->workdir);
 453                 config->workdir = NULL;
 454         }
 455
 456         return 0;
 457 }
 458
 459 #define OVL_WORKDIR_NAME "work"
 460 #define OVL_INDEXDIR_NAME "index"
 461
 462 static struct dentry *ovl_workdir_create(struct ovl_fs *ufs,
 463                                          const char *name, bool persist)
 464 {
 465         struct inode *dir =  ufs->workbasedir->d_inode;
 466         struct vfsmount *mnt = ufs->upper_mnt;
 467         struct dentry *work;
 468         int err;
 469         bool retried = false;
 470         bool locked = false;
 471
 472         err = mnt_want_write(mnt);
 473         if (err)
 474                 goto out_err;
 475
 476         inode_lock_nested(dir, I_MUTEX_PARENT);
 477         locked = true;
 478
 479 retry:
 480         work = lookup_one_len(name, ufs->workbasedir, strlen(name));
 481
 482         if (!IS_ERR(work)) {
 483                 struct iattr attr = {
 484                         .ia_valid = ATTR_MODE,
 485                         .ia_mode = S_IFDIR | 0,
 486                 };
 487
 488                 if (work->d_inode) {
 489                         err = -EEXIST;
 490                         if (retried)
 491                                 goto out_dput;
 492
 493                         if (persist)
 494                                 goto out_unlock;
 495
 496                         retried = true;
 497                         ovl_workdir_cleanup(dir, mnt, work, 0);
 498                         dput(work);
 499                         goto retry;
 500                 }
 501
 502                 err = ovl_create_real(dir, work,
 503                                       &(struct cattr){.mode = S_IFDIR | 0},
 504                                       NULL, true);
 505                 if (err)
 506                         goto out_dput;
 507
 508                 /*
 509                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 510                  *
 511                  * a) success (there was a POSIX ACL xattr and was removed)
 512                  * b) -ENODATA (there was no POSIX ACL xattr)
 513                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 514                  *
 515                  * There are various other error values that could effectively
 516                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 517                  * if the xattr name is too long), but the set of filesystems
 518                  * allowed as upper are limited to "normal" ones, where checking
 519                  * for the above two errors is sufficient.
 520                  */
 521                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 522                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 523                         goto out_dput;
 524
 525                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 526                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 527                         goto out_dput;
 528
 529                 /* Clear any inherited mode bits */
 530                 inode_lock(work->d_inode);
 531                 err = notify_change(work, &attr, NULL);
 532                 inode_unlock(work->d_inode);
 533                 if (err)
 534                         goto out_dput;
 535         } else {
 536                 err = PTR_ERR(work);
 537                 goto out_err;
 538         }
 539 out_unlock:
 540         mnt_drop_write(mnt);
 541         if (locked)
 542                 inode_unlock(dir);
 543
 544         return work;
 545
 546 out_dput:
 547         dput(work);
 548 out_err:
 549         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 550                 ufs->config.workdir, name, -err);
 551         work = NULL;
 552         goto out_unlock;
 553 }
 554
 555 static void ovl_unescape(char *s)
 556 {
 557         char *d = s;
 558
 559         for (;; s++, d++) {
 560                 if (*s == '\\')
 561                         s++;
 562                 *d = *s;
 563                 if (!*s)
 564                         break;
 565         }
 566 }
 567
 568 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 569 {
 570         int err = -EINVAL;
 571
 572         if (!*name) {
 573                 pr_err("overlayfs: empty lowerdir\n");
 574                 goto out;
 575         }
 576         err = kern_path(name, LOOKUP_FOLLOW, path);
 577         if (err) {
 578                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 579                 goto out;
 580         }
 581         err = -EINVAL;
 582         if (ovl_dentry_weird(path->dentry)) {
 583                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 584                 goto out_put;
 585         }
 586         if (!d_is_dir(path->dentry)) {
 587                 pr_err("overlayfs: '%s' not a directory\n", name);
 588                 goto out_put;
 589         }
 590         return 0;
 591
 592 out_put:
 593         path_put_init(path);
 594 out:
 595         return err;
 596 }
 597
 598 static int ovl_mount_dir(const char *name, struct path *path)
 599 {
 600         int err = -ENOMEM;
 601         char *tmp = kstrdup(name, GFP_KERNEL);
 602
 603         if (tmp) {
 604                 ovl_unescape(tmp);
 605                 err = ovl_mount_dir_noesc(tmp, path);
 606
 607                 if (!err)
 608                         if (ovl_dentry_remote(path->dentry)) {
 609                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 610                                        tmp);
 611                                 path_put_init(path);
 612                                 err = -EINVAL;
 613                         }
 614                 kfree(tmp);
 615         }
 616         return err;
 617 }
 618
 619 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 620                              const char *name)
 621 {
 622         struct kstatfs statfs;
 623         int err = vfs_statfs(path, &statfs);
 624
 625         if (err)
 626                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 627         else
 628                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 629
 630         return err;
 631 }
 632
 633 static int ovl_lower_dir(const char *name, struct path *path,
 634                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 635 {
 636         int err;
 637
 638         err = ovl_mount_dir_noesc(name, path);
 639         if (err)
 640                 goto out;
 641
 642         err = ovl_check_namelen(path, ofs, name);
 643         if (err)
 644                 goto out_put;
 645
 646         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 647
 648         if (ovl_dentry_remote(path->dentry))
 649                 *remote = true;
 650
 651         /*
 652          * The inodes index feature needs to encode and decode file
 653          * handles, so it requires that all layers support them.
 654          */
 655         if (ofs->config.index && !ovl_can_decode_fh(path->dentry->d_sb)) {
 656                 ofs->config.index = false;
 657                 pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off.\n", name);
 658         }
 659
 660         return 0;
 661
 662 out_put:
 663         path_put_init(path);
 664 out:
 665         return err;
 666 }
 667
 668 /* Workdir should not be subdir of upperdir and vice versa */
 669 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 670 {
 671         bool ok = false;
 672
 673         if (workdir != upperdir) {
 674                 ok = (lock_rename(workdir, upperdir) == NULL);
 675                 unlock_rename(workdir, upperdir);
 676         }
 677         return ok;
 678 }
 679
 680 static unsigned int ovl_split_lowerdirs(char *str)
 681 {
 682         unsigned int ctr = 1;
 683         char *s, *d;
 684
 685         for (s = d = str;; s++, d++) {
 686                 if (*s == '\\') {
 687                         s++;
 688                 } else if (*s == ':') {
 689                         *d = '\0';
 690                         ctr++;
 691                         continue;
 692                 }
 693                 *d = *s;
 694                 if (!*s)
 695                         break;
 696         }
 697         return ctr;
 698 }
 699
 700 static int __maybe_unused
 701 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 702                         struct dentry *dentry, struct inode *inode,
 703                         const char *name, void *buffer, size_t size)
 704 {
 705         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 706 }
 707
 708 static int __maybe_unused
 709 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 710                         struct dentry *dentry, struct inode *inode,
 711                         const char *name, const void *value,
 712                         size_t size, int flags)
 713 {
 714         struct dentry *workdir = ovl_workdir(dentry);
 715         struct inode *realinode = ovl_inode_real(inode);
 716         struct posix_acl *acl = NULL;
 717         int err;
 718
 719         /* Check that everything is OK before copy-up */
 720         if (value) {
 721                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 722                 if (IS_ERR(acl))
 723                         return PTR_ERR(acl);
 724         }
 725         err = -EOPNOTSUPP;
 726         if (!IS_POSIXACL(d_inode(workdir)))
 727                 goto out_acl_release;
 728         if (!realinode->i_op->set_acl)
 729                 goto out_acl_release;
 730         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 731                 err = acl ? -EACCES : 0;
 732                 goto out_acl_release;
 733         }
 734         err = -EPERM;
 735         if (!inode_owner_or_capable(inode))
 736                 goto out_acl_release;
 737
 738         posix_acl_release(acl);
 739
 740         /*
 741          * Check if sgid bit needs to be cleared (actual setacl operation will
 742          * be done with mounter's capabilities and so that won't do it for us).
 743          */
 744         if (unlikely(inode->i_mode & S_ISGID) &&
 745             handler->flags == ACL_TYPE_ACCESS &&
 746             !in_group_p(inode->i_gid) &&
 747             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 748                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 749
 750                 err = ovl_setattr(dentry, &iattr);
 751                 if (err)
 752                         return err;
 753         }
 754
 755         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
 756         if (!err)
 757                 ovl_copyattr(ovl_inode_real(inode), inode);
 758
 759         return err;
 760
 761 out_acl_release:
 762         posix_acl_release(acl);
 763         return err;
 764 }
 765
 766 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 767                              struct dentry *dentry, struct inode *inode,
 768                              const char *name, void *buffer, size_t size)
 769 {
 770         return -EOPNOTSUPP;
 771 }
 772
 773 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 774                              struct dentry *dentry, struct inode *inode,
 775                              const char *name, const void *value,
 776                              size_t size, int flags)
 777 {
 778         return -EOPNOTSUPP;
 779 }
 780
 781 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 782                                struct dentry *dentry, struct inode *inode,
 783                                const char *name, void *buffer, size_t size)
 784 {
 785         return ovl_xattr_get(dentry, inode, name, buffer, size);
 786 }
 787
 788 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 789                                struct dentry *dentry, struct inode *inode,
 790                                const char *name, const void *value,
 791                                size_t size, int flags)
 792 {
 793         return ovl_xattr_set(dentry, inode, name, value, size, flags);
 794 }
 795
 796 static const struct xattr_handler __maybe_unused
 797 ovl_posix_acl_access_xattr_handler = {
 798         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 799         .flags = ACL_TYPE_ACCESS,
 800         .get = ovl_posix_acl_xattr_get,
 801         .set = ovl_posix_acl_xattr_set,
 802 };
 803
 804 static const struct xattr_handler __maybe_unused
 805 ovl_posix_acl_default_xattr_handler = {
 806         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 807         .flags = ACL_TYPE_DEFAULT,
 808         .get = ovl_posix_acl_xattr_get,
 809         .set = ovl_posix_acl_xattr_set,
 810 };
 811
 812 static const struct xattr_handler ovl_own_xattr_handler = {
 813         .prefix = OVL_XATTR_PREFIX,
 814         .get = ovl_own_xattr_get,
 815         .set = ovl_own_xattr_set,
 816 };
 817
 818 static const struct xattr_handler ovl_other_xattr_handler = {
 819         .prefix = "", /* catch all */
 820         .get = ovl_other_xattr_get,
 821         .set = ovl_other_xattr_set,
 822 };
 823
 824 static const struct xattr_handler *ovl_xattr_handlers[] = {
 825 #ifdef CONFIG_FS_POSIX_ACL
 826         &ovl_posix_acl_access_xattr_handler,
 827         &ovl_posix_acl_default_xattr_handler,
 828 #endif
 829         &ovl_own_xattr_handler,
 830         &ovl_other_xattr_handler,
 831         NULL
 832 };
 833
 834 static int ovl_get_upperpath(struct ovl_fs *ufs, struct path *upperpath)
 835 {
 836         int err;
 837
 838         err = ovl_mount_dir(ufs->config.upperdir, upperpath);
 839         if (err)
 840                 goto out;
 841
 842         /* Upper fs should not be r/o */
 843         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
 844                 pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 845                 err = -EINVAL;
 846                 goto out;
 847         }
 848
 849         err = ovl_check_namelen(upperpath, ufs, ufs->config.upperdir);
 850         if (err)
 851                 goto out;
 852
 853         err = -EBUSY;
 854         if (ovl_inuse_trylock(upperpath->dentry)) {
 855                 ufs->upperdir_locked = true;
 856         } else if (ufs->config.index) {
 857                 pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n");
 858                 goto out;
 859         } else {
 860                 pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
 861         }
 862         err = 0;
 863 out:
 864         return err;
 865 }
 866
 867 static int ovl_get_workpath(struct ovl_fs *ufs, struct path *upperpath,
 868                             struct path *workpath)
 869 {
 870         int err;
 871
 872         err = ovl_mount_dir(ufs->config.workdir, workpath);
 873         if (err)
 874                 goto out;
 875
 876         err = -EINVAL;
 877         if (upperpath->mnt != workpath->mnt) {
 878                 pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 879                 goto out;
 880         }
 881         if (!ovl_workdir_ok(workpath->dentry, upperpath->dentry)) {
 882                 pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 883                 goto out;
 884         }
 885
 886         err = -EBUSY;
 887         if (ovl_inuse_trylock(workpath->dentry)) {
 888                 ufs->workdir_locked = true;
 889         } else if (ufs->config.index) {
 890                 pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n");
 891                 goto out;
 892         } else {
 893                 pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
 894         }
 895
 896         ufs->workbasedir = dget(workpath->dentry);
 897         err = 0;
 898 out:
 899         return err;
 900 }
 901
 902 static int ovl_get_upper(struct ovl_fs *ufs, struct path *upperpath)
 903 {
 904         struct vfsmount *upper_mnt;
 905
 906         upper_mnt = clone_private_mount(upperpath);
 907         if (IS_ERR(upper_mnt)) {
 908                 pr_err("overlayfs: failed to clone upperpath\n");
 909                 return PTR_ERR(upper_mnt);
 910         }
 911
 912         /* Don't inherit atime flags */
 913         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
 914         ufs->upper_mnt = upper_mnt;
 915
 916         return 0;
 917 }
 918
 919 static int ovl_get_workdir(struct ovl_fs *ufs, struct path *workpath)
 920 {
 921         struct dentry *temp;
 922         int err;
 923
 924         ufs->workdir = ovl_workdir_create(ufs, OVL_WORKDIR_NAME, false);
 925         if (!ufs->workdir)
 926                 return 0;
 927
 928         /*
 929          * Upper should support d_type, else whiteouts are visible.  Given
 930          * workdir and upper are on same fs, we can do iterate_dir() on
 931          * workdir. This check requires successful creation of workdir in
 932          * previous step.
 933          */
 934         err = ovl_check_d_type_supported(workpath);
 935         if (err < 0)
 936                 return err;
 937
 938         /*
 939          * We allowed this configuration and don't want to break users over
 940          * kernel upgrade. So warn instead of erroring out.
 941          */
 942         if (!err)
 943                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
 944
 945         /* Check if upper/work fs supports O_TMPFILE */
 946         temp = ovl_do_tmpfile(ufs->workdir, S_IFREG | 0);
 947         ufs->tmpfile = !IS_ERR(temp);
 948         if (ufs->tmpfile)
 949                 dput(temp);
 950         else
 951                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
 952
 953         /*
 954          * Check if upper/work fs supports trusted.overlay.* xattr
 955          */
 956         err = ovl_do_setxattr(ufs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
 957         if (err) {
 958                 ufs->noxattr = true;
 959                 pr_warn("overlayfs: upper fs does not support xattr.\n");
 960         } else {
 961                 vfs_removexattr(ufs->workdir, OVL_XATTR_OPAQUE);
 962         }
 963
 964         /* Check if upper/work fs supports file handles */
 965         if (ufs->config.index &&
 966             !ovl_can_decode_fh(ufs->workdir->d_sb)) {
 967                 ufs->config.index = false;
 968                 pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
 969         }
 970
 971         return 0;
 972 }
 973
 974 static int ovl_get_indexdir(struct ovl_fs *ufs, struct ovl_entry *oe,
 975                             struct path *upperpath)
 976 {
 977         int err;
 978
 979         /* Verify lower root is upper root origin */
 980         err = ovl_verify_origin(upperpath->dentry,
 981                                 oe->lowerstack[0].layer->mnt,
 982                                 oe->lowerstack[0].dentry,
 983                                 false, true);
 984         if (err) {
 985                 pr_err("overlayfs: failed to verify upper root origin\n");
 986                 goto out;
 987         }
 988
 989         ufs->indexdir = ovl_workdir_create(ufs, OVL_INDEXDIR_NAME, true);
 990         if (ufs->indexdir) {
 991                 /* Verify upper root is index dir origin */
 992                 err = ovl_verify_origin(ufs->indexdir, ufs->upper_mnt,
 993                                         upperpath->dentry, true, true);
 994                 if (err)
 995                         pr_err("overlayfs: failed to verify index dir origin\n");
 996
 997                 /* Cleanup bad/stale/orphan index entries */
 998                 if (!err)
 999                         err = ovl_indexdir_cleanup(ufs->indexdir,
1000                                                    ufs->upper_mnt,
1001                                                    oe->lowerstack,
1002                                                    oe->numlower);
1003         }
1004         if (err || !ufs->indexdir)
1005                 pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1006
1007 out:
1008         return err;
1009 }
1010
1011 static int ovl_get_lowerstack(struct super_block *sb, struct ovl_fs *ufs,
1012                               struct path **stackp, unsigned int *stacklenp)
1013 {
1014         int err;
1015         char *lowertmp, *lower;
1016         struct path *stack;
1017         unsigned int stacklen, numlower, i;
1018         bool remote = false;
1019
1020         err = -ENOMEM;
1021         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
1022         if (!lowertmp)
1023                 goto out;
1024
1025         err = -EINVAL;
1026         stacklen = ovl_split_lowerdirs(lowertmp);
1027         if (stacklen > OVL_MAX_STACK) {
1028                 pr_err("overlayfs: too many lower directories, limit is %d\n",
1029                        OVL_MAX_STACK);
1030                 goto out;
1031         } else if (!ufs->config.upperdir && stacklen == 1) {
1032                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1033                 goto out;
1034         }
1035
1036         err = -ENOMEM;
1037         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1038         if (!stack)
1039                 goto out;
1040
1041         err = -EINVAL;
1042         lower = lowertmp;
1043         for (numlower = 0; numlower < stacklen; numlower++) {
1044                 err = ovl_lower_dir(lower, &stack[numlower], ufs,
1045                                     &sb->s_stack_depth, &remote);
1046                 if (err)
1047                         goto out_free_stack;
1048
1049                 lower = strchr(lower, '\0') + 1;
1050         }
1051
1052         err = -EINVAL;
1053         sb->s_stack_depth++;
1054         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1055                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1056                 goto out_free_stack;
1057         }
1058
1059         *stackp = stack;
1060         *stacklenp = numlower;
1061
1062         if (remote)
1063                 sb->s_d_op = &ovl_reval_dentry_operations;
1064         else
1065                 sb->s_d_op = &ovl_dentry_operations;
1066
1067         err = 0;
1068
1069 out:
1070         kfree(lowertmp);
1071         return err;
1072
1073 out_free_stack:
1074         for (i = 0; i < numlower; i++)
1075                 path_put(&stack[i]);
1076         kfree(stack);
1077         goto out;
1078 }
1079
1080 static int ovl_get_lower_layers(struct ovl_fs *ufs, struct path *stack,
1081                                 unsigned int numlower)
1082 {
1083         int err;
1084         unsigned int i;
1085
1086         err = -ENOMEM;
1087         ufs->lower_layers = kcalloc(numlower, sizeof(struct ovl_layer),
1088                                     GFP_KERNEL);
1089         if (ufs->lower_layers == NULL)
1090                 goto out;
1091         for (i = 0; i < numlower; i++) {
1092                 struct vfsmount *mnt;
1093                 dev_t dev;
1094
1095                 err = get_anon_bdev(&dev);
1096                 if (err) {
1097                         pr_err("overlayfs: failed to get anonymous bdev for lowerpath\n");
1098                         goto out;
1099                 }
1100
1101                 mnt = clone_private_mount(&stack[i]);
1102                 err = PTR_ERR(mnt);
1103                 if (IS_ERR(mnt)) {
1104                         pr_err("overlayfs: failed to clone lowerpath\n");
1105                         free_anon_bdev(dev);
1106                         goto out;
1107                 }
1108                 /*
1109                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1110                  * will fail instead of modifying lower fs.
1111                  */
1112                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1113
1114                 ufs->lower_layers[ufs->numlower].mnt = mnt;
1115                 ufs->lower_layers[ufs->numlower].pseudo_dev = dev;
1116                 ufs->numlower++;
1117
1118                 /* Check if all lower layers are on same sb */
1119                 if (i == 0)
1120                         ufs->same_sb = mnt->mnt_sb;
1121                 else if (ufs->same_sb != mnt->mnt_sb)
1122                         ufs->same_sb = NULL;
1123         }
1124         err = 0;
1125 out:
1126         return err;
1127 }
1128
1129 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1130 {
1131         struct path upperpath = { };
1132         struct path workpath = { };
1133         struct dentry *root_dentry;
1134         struct ovl_entry *oe = NULL;
1135         struct ovl_fs *ufs;
1136         struct path *stack = NULL;
1137         unsigned int numlower = 0;
1138         unsigned int i;
1139         struct cred *cred;
1140         int err;
1141
1142         err = -ENOMEM;
1143         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1144         if (!ufs)
1145                 goto out;
1146
1147         ufs->creator_cred = cred = prepare_creds();
1148         if (!cred)
1149                 goto out_err;
1150
1151         ufs->config.redirect_dir = ovl_redirect_dir_def;
1152         ufs->config.index = ovl_index_def;
1153         err = ovl_parse_opt((char *) data, &ufs->config);
1154         if (err)
1155                 goto out_err;
1156
1157         err = -EINVAL;
1158         if (!ufs->config.lowerdir) {
1159                 if (!silent)
1160                         pr_err("overlayfs: missing 'lowerdir'\n");
1161                 goto out_err;
1162         }
1163
1164         sb->s_stack_depth = 0;
1165         sb->s_maxbytes = MAX_LFS_FILESIZE;
1166         if (ufs->config.upperdir) {
1167                 if (!ufs->config.workdir) {
1168                         pr_err("overlayfs: missing 'workdir'\n");
1169                         goto out_err;
1170                 }
1171
1172                 err = ovl_get_upperpath(ufs, &upperpath);
1173                 if (err)
1174                         goto out_err;
1175
1176                 err = ovl_get_upper(ufs, &upperpath);
1177                 if (err)
1178                         goto out_err;
1179
1180                 err = ovl_get_workpath(ufs, &upperpath, &workpath);
1181                 if (err)
1182                         goto out_err;
1183
1184                 err = ovl_get_workdir(ufs, &workpath);
1185                 if (err)
1186                         goto out_err;
1187
1188                 if (!ufs->workdir)
1189                         sb->s_flags |= MS_RDONLY;
1190
1191                 sb->s_stack_depth = ufs->upper_mnt->mnt_sb->s_stack_depth;
1192                 sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
1193
1194         }
1195         err = ovl_get_lowerstack(sb, ufs, &stack, &numlower);
1196         if (err)
1197                 goto out_err;
1198
1199         err = ovl_get_lower_layers(ufs, stack, numlower);
1200         if (err)
1201                 goto out_err;
1202
1203         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1204         if (!ufs->upper_mnt)
1205                 sb->s_flags |= MS_RDONLY;
1206         else if (ufs->upper_mnt->mnt_sb != ufs->same_sb)
1207                 ufs->same_sb = NULL;
1208
1209         err = -ENOMEM;
1210         oe = ovl_alloc_entry(numlower);
1211         if (!oe)
1212                 goto out_err;
1213
1214         for (i = 0; i < numlower; i++) {
1215                 oe->lowerstack[i].dentry = stack[i].dentry;
1216                 oe->lowerstack[i].layer = &(ufs->lower_layers[i]);
1217         }
1218
1219         if (!(ovl_force_readonly(ufs)) && ufs->config.index) {
1220                 err = ovl_get_indexdir(ufs, oe, &upperpath);
1221                 if (err)
1222                         goto out_err;
1223
1224                 if (!ufs->indexdir)
1225                         sb->s_flags |= MS_RDONLY;
1226         }
1227
1228         /* Show index=off/on in /proc/mounts for any of the reasons above */
1229         if (!ufs->indexdir)
1230                 ufs->config.index = false;
1231
1232         /* Never override disk quota limits or use reserved space */
1233         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1234
1235         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1236         sb->s_op = &ovl_super_operations;
1237         sb->s_xattr = ovl_xattr_handlers;
1238         sb->s_fs_info = ufs;
1239         sb->s_flags |= MS_POSIXACL | MS_NOREMOTELOCK;
1240
1241         err = -ENOMEM;
1242         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1243         if (!root_dentry)
1244                 goto out_err;
1245
1246         mntput(upperpath.mnt);
1247         for (i = 0; i < numlower; i++)
1248                 mntput(stack[i].mnt);
1249         kfree(stack);
1250         path_put(&workpath);
1251
1252         if (upperpath.dentry) {
1253                 oe->has_upper = true;
1254                 if (ovl_is_impuredir(upperpath.dentry))
1255                         ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1256         }
1257
1258         root_dentry->d_fsdata = oe;
1259
1260         /* Root is always merge -> can have whiteouts */
1261         ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1262         ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1263                        ovl_dentry_lower(root_dentry));
1264
1265         sb->s_root = root_dentry;
1266
1267         return 0;
1268
1269 out_err:
1270         kfree(oe);
1271         for (i = 0; i < numlower; i++)
1272                 path_put(&stack[i]);
1273         kfree(stack);
1274         path_put(&workpath);
1275         path_put(&upperpath);
1276         ovl_free_fs(ufs);
1277 out:
1278         return err;
1279 }
1280
1281 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1282                                 const char *dev_name, void *raw_data)
1283 {
1284         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1285 }
1286
1287 static struct file_system_type ovl_fs_type = {
1288         .owner          = THIS_MODULE,
1289         .name           = "overlay",
1290         .mount          = ovl_mount,
1291         .kill_sb        = kill_anon_super,
1292 };
1293 MODULE_ALIAS_FS("overlay");
1294
1295 static void ovl_inode_init_once(void *foo)
1296 {
1297         struct ovl_inode *oi = foo;
1298
1299         inode_init_once(&oi->vfs_inode);
1300 }
1301
1302 static int __init ovl_init(void)
1303 {
1304         int err;
1305
1306         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1307                                              sizeof(struct ovl_inode), 0,
1308                                              (SLAB_RECLAIM_ACCOUNT|
1309                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1310                                              ovl_inode_init_once);
1311         if (ovl_inode_cachep == NULL)
1312                 return -ENOMEM;
1313
1314         err = register_filesystem(&ovl_fs_type);
1315         if (err)
1316                 kmem_cache_destroy(ovl_inode_cachep);
1317
1318         return err;
1319 }
1320
1321 static void __exit ovl_exit(void)
1322 {
1323         unregister_filesystem(&ovl_fs_type);
1324
1325         /*
1326          * Make sure all delayed rcu free inodes are flushed before we
1327          * destroy cache.
1328          */
1329         rcu_barrier();
1330         kmem_cache_destroy(ovl_inode_cachep);
1331
1332 }
1333
1334 module_init(ovl_init);
1335 module_exit(ovl_exit);