fs/overlayfs/super.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <uapi/linux/magic.h>
  11 #include <linux/fs.h>
  12 #include <linux/namei.h>
  13 #include <linux/xattr.h>
  14 #include <linux/mount.h>
  15 #include <linux/parser.h>
  16 #include <linux/module.h>
  17 #include <linux/statfs.h>
  18 #include <linux/seq_file.h>
  19 #include <linux/posix_acl_xattr.h>
  20 #include "overlayfs.h"
  21 #include "ovl_entry.h"
  22
  23 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  24 MODULE_DESCRIPTION("Overlay filesystem");
  25 MODULE_LICENSE("GPL");
  26
  27
  28 struct ovl_dir_cache;
  29
  30 #define OVL_MAX_STACK 500
  31
  32 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  33 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  34 MODULE_PARM_DESC(ovl_redirect_dir_def,
  35                  "Default to on or off for the redirect_dir feature");
  36
  37 static void ovl_dentry_release(struct dentry *dentry)
  38 {
  39         struct ovl_entry *oe = dentry->d_fsdata;
  40
  41         if (oe) {
  42                 unsigned int i;
  43
  44                 dput(oe->__upperdentry);
  45                 kfree(oe->redirect);
  46                 for (i = 0; i < oe->numlower; i++)
  47                         dput(oe->lowerstack[i].dentry);
  48                 kfree_rcu(oe, rcu);
  49         }
  50 }
  51
  52 static int ovl_check_append_only(struct inode *inode, int flag)
  53 {
  54         /*
  55          * This test was moot in vfs may_open() because overlay inode does
  56          * not have the S_APPEND flag, so re-check on real upper inode
  57          */
  58         if (IS_APPEND(inode)) {
  59                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
  60                         return -EPERM;
  61                 if (flag & O_TRUNC)
  62                         return -EPERM;
  63         }
  64
  65         return 0;
  66 }
  67
  68 static struct dentry *ovl_d_real(struct dentry *dentry,
  69                                  const struct inode *inode,
  70                                  unsigned int open_flags)
  71 {
  72         struct dentry *real;
  73         int err;
  74
  75         if (!d_is_reg(dentry)) {
  76                 if (!inode || inode == d_inode(dentry))
  77                         return dentry;
  78                 goto bug;
  79         }
  80
  81         if (d_is_negative(dentry))
  82                 return dentry;
  83
  84         if (open_flags) {
  85                 err = ovl_open_maybe_copy_up(dentry, open_flags);
  86                 if (err)
  87                         return ERR_PTR(err);
  88         }
  89
  90         real = ovl_dentry_upper(dentry);
  91         if (real && (!inode || inode == d_inode(real))) {
  92                 if (!inode) {
  93                         err = ovl_check_append_only(d_inode(real), open_flags);
  94                         if (err)
  95                                 return ERR_PTR(err);
  96                 }
  97                 return real;
  98         }
  99
 100         real = ovl_dentry_lower(dentry);
 101         if (!real)
 102                 goto bug;
 103
 104         /* Handle recursion */
 105         real = d_real(real, inode, open_flags);
 106
 107         if (!inode || inode == d_inode(real))
 108                 return real;
 109 bug:
 110         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
 111              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 112         return dentry;
 113 }
 114
 115 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 116 {
 117         struct ovl_entry *oe = dentry->d_fsdata;
 118         unsigned int i;
 119         int ret = 1;
 120
 121         for (i = 0; i < oe->numlower; i++) {
 122                 struct dentry *d = oe->lowerstack[i].dentry;
 123
 124                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 125                         ret = d->d_op->d_revalidate(d, flags);
 126                         if (ret < 0)
 127                                 return ret;
 128                         if (!ret) {
 129                                 if (!(flags & LOOKUP_RCU))
 130                                         d_invalidate(d);
 131                                 return -ESTALE;
 132                         }
 133                 }
 134         }
 135         return 1;
 136 }
 137
 138 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 139 {
 140         struct ovl_entry *oe = dentry->d_fsdata;
 141         unsigned int i;
 142         int ret = 1;
 143
 144         for (i = 0; i < oe->numlower; i++) {
 145                 struct dentry *d = oe->lowerstack[i].dentry;
 146
 147                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 148                         ret = d->d_op->d_weak_revalidate(d, flags);
 149                         if (ret <= 0)
 150                                 break;
 151                 }
 152         }
 153         return ret;
 154 }
 155
 156 static const struct dentry_operations ovl_dentry_operations = {
 157         .d_release = ovl_dentry_release,
 158         .d_real = ovl_d_real,
 159 };
 160
 161 static const struct dentry_operations ovl_reval_dentry_operations = {
 162         .d_release = ovl_dentry_release,
 163         .d_real = ovl_d_real,
 164         .d_revalidate = ovl_dentry_revalidate,
 165         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 166 };
 167
 168 static struct kmem_cache *ovl_inode_cachep;
 169
 170 static struct inode *ovl_alloc_inode(struct super_block *sb)
 171 {
 172         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 173
 174         return &oi->vfs_inode;
 175 }
 176
 177 static void ovl_i_callback(struct rcu_head *head)
 178 {
 179         struct inode *inode = container_of(head, struct inode, i_rcu);
 180
 181         kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
 182 }
 183
 184 static void ovl_destroy_inode(struct inode *inode)
 185 {
 186         call_rcu(&inode->i_rcu, ovl_i_callback);
 187 }
 188
 189 static void ovl_put_super(struct super_block *sb)
 190 {
 191         struct ovl_fs *ufs = sb->s_fs_info;
 192         unsigned i;
 193
 194         dput(ufs->workdir);
 195         mntput(ufs->upper_mnt);
 196         for (i = 0; i < ufs->numlower; i++)
 197                 mntput(ufs->lower_mnt[i]);
 198         kfree(ufs->lower_mnt);
 199
 200         kfree(ufs->config.lowerdir);
 201         kfree(ufs->config.upperdir);
 202         kfree(ufs->config.workdir);
 203         put_cred(ufs->creator_cred);
 204         kfree(ufs);
 205 }
 206
 207 static int ovl_sync_fs(struct super_block *sb, int wait)
 208 {
 209         struct ovl_fs *ufs = sb->s_fs_info;
 210         struct super_block *upper_sb;
 211         int ret;
 212
 213         if (!ufs->upper_mnt)
 214                 return 0;
 215         upper_sb = ufs->upper_mnt->mnt_sb;
 216         if (!upper_sb->s_op->sync_fs)
 217                 return 0;
 218
 219         /* real inodes have already been synced by sync_filesystem(ovl_sb) */
 220         down_read(&upper_sb->s_umount);
 221         ret = upper_sb->s_op->sync_fs(upper_sb, wait);
 222         up_read(&upper_sb->s_umount);
 223         return ret;
 224 }
 225
 226 /**
 227  * ovl_statfs
 228  * @sb: The overlayfs super block
 229  * @buf: The struct kstatfs to fill in with stats
 230  *
 231  * Get the filesystem statistics.  As writes always target the upper layer
 232  * filesystem pass the statfs to the upper filesystem (if it exists)
 233  */
 234 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 235 {
 236         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 237         struct dentry *root_dentry = dentry->d_sb->s_root;
 238         struct path path;
 239         int err;
 240
 241         ovl_path_real(root_dentry, &path);
 242
 243         err = vfs_statfs(&path, buf);
 244         if (!err) {
 245                 buf->f_namelen = ofs->namelen;
 246                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 247         }
 248
 249         return err;
 250 }
 251
 252 /**
 253  * ovl_show_options
 254  *
 255  * Prints the mount options for a given superblock.
 256  * Returns zero; does not fail.
 257  */
 258 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 259 {
 260         struct super_block *sb = dentry->d_sb;
 261         struct ovl_fs *ufs = sb->s_fs_info;
 262
 263         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 264         if (ufs->config.upperdir) {
 265                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 266                 seq_show_option(m, "workdir", ufs->config.workdir);
 267         }
 268         if (ufs->config.default_permissions)
 269                 seq_puts(m, ",default_permissions");
 270         if (ufs->config.redirect_dir != ovl_redirect_dir_def)
 271                 seq_printf(m, ",redirect_dir=%s",
 272                            ufs->config.redirect_dir ? "on" : "off");
 273         return 0;
 274 }
 275
 276 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 277 {
 278         struct ovl_fs *ufs = sb->s_fs_info;
 279
 280         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 281                 return -EROFS;
 282
 283         return 0;
 284 }
 285
 286 static const struct super_operations ovl_super_operations = {
 287         .alloc_inode    = ovl_alloc_inode,
 288         .destroy_inode  = ovl_destroy_inode,
 289         .drop_inode     = generic_delete_inode,
 290         .put_super      = ovl_put_super,
 291         .sync_fs        = ovl_sync_fs,
 292         .statfs         = ovl_statfs,
 293         .show_options   = ovl_show_options,
 294         .remount_fs     = ovl_remount,
 295 };
 296
 297 enum {
 298         OPT_LOWERDIR,
 299         OPT_UPPERDIR,
 300         OPT_WORKDIR,
 301         OPT_DEFAULT_PERMISSIONS,
 302         OPT_REDIRECT_DIR_ON,
 303         OPT_REDIRECT_DIR_OFF,
 304         OPT_ERR,
 305 };
 306
 307 static const match_table_t ovl_tokens = {
 308         {OPT_LOWERDIR,                  "lowerdir=%s"},
 309         {OPT_UPPERDIR,                  "upperdir=%s"},
 310         {OPT_WORKDIR,                   "workdir=%s"},
 311         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 312         {OPT_REDIRECT_DIR_ON,           "redirect_dir=on"},
 313         {OPT_REDIRECT_DIR_OFF,          "redirect_dir=off"},
 314         {OPT_ERR,                       NULL}
 315 };
 316
 317 static char *ovl_next_opt(char **s)
 318 {
 319         char *sbegin = *s;
 320         char *p;
 321
 322         if (sbegin == NULL)
 323                 return NULL;
 324
 325         for (p = sbegin; *p; p++) {
 326                 if (*p == '\\') {
 327                         p++;
 328                         if (!*p)
 329                                 break;
 330                 } else if (*p == ',') {
 331                         *p = '\0';
 332                         *s = p + 1;
 333                         return sbegin;
 334                 }
 335         }
 336         *s = NULL;
 337         return sbegin;
 338 }
 339
 340 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 341 {
 342         char *p;
 343
 344         while ((p = ovl_next_opt(&opt)) != NULL) {
 345                 int token;
 346                 substring_t args[MAX_OPT_ARGS];
 347
 348                 if (!*p)
 349                         continue;
 350
 351                 token = match_token(p, ovl_tokens, args);
 352                 switch (token) {
 353                 case OPT_UPPERDIR:
 354                         kfree(config->upperdir);
 355                         config->upperdir = match_strdup(&args[0]);
 356                         if (!config->upperdir)
 357                                 return -ENOMEM;
 358                         break;
 359
 360                 case OPT_LOWERDIR:
 361                         kfree(config->lowerdir);
 362                         config->lowerdir = match_strdup(&args[0]);
 363                         if (!config->lowerdir)
 364                                 return -ENOMEM;
 365                         break;
 366
 367                 case OPT_WORKDIR:
 368                         kfree(config->workdir);
 369                         config->workdir = match_strdup(&args[0]);
 370                         if (!config->workdir)
 371                                 return -ENOMEM;
 372                         break;
 373
 374                 case OPT_DEFAULT_PERMISSIONS:
 375                         config->default_permissions = true;
 376                         break;
 377
 378                 case OPT_REDIRECT_DIR_ON:
 379                         config->redirect_dir = true;
 380                         break;
 381
 382                 case OPT_REDIRECT_DIR_OFF:
 383                         config->redirect_dir = false;
 384                         break;
 385
 386                 default:
 387                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 388                         return -EINVAL;
 389                 }
 390         }
 391
 392         /* Workdir is useless in non-upper mount */
 393         if (!config->upperdir && config->workdir) {
 394                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 395                         config->workdir);
 396                 kfree(config->workdir);
 397                 config->workdir = NULL;
 398         }
 399
 400         return 0;
 401 }
 402
 403 #define OVL_WORKDIR_NAME "work"
 404
 405 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 406                                          struct dentry *dentry)
 407 {
 408         struct inode *dir = dentry->d_inode;
 409         struct dentry *work;
 410         int err;
 411         bool retried = false;
 412
 413         err = mnt_want_write(mnt);
 414         if (err)
 415                 return ERR_PTR(err);
 416
 417         inode_lock_nested(dir, I_MUTEX_PARENT);
 418 retry:
 419         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 420                               strlen(OVL_WORKDIR_NAME));
 421
 422         if (!IS_ERR(work)) {
 423                 struct iattr attr = {
 424                         .ia_valid = ATTR_MODE,
 425                         .ia_mode = S_IFDIR | 0,
 426                 };
 427
 428                 if (work->d_inode) {
 429                         err = -EEXIST;
 430                         if (retried)
 431                                 goto out_dput;
 432
 433                         retried = true;
 434                         ovl_workdir_cleanup(dir, mnt, work, 0);
 435                         dput(work);
 436                         goto retry;
 437                 }
 438
 439                 err = ovl_create_real(dir, work,
 440                                       &(struct cattr){.mode = S_IFDIR | 0},
 441                                       NULL, true);
 442                 if (err)
 443                         goto out_dput;
 444
 445                 /*
 446                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 447                  *
 448                  * a) success (there was a POSIX ACL xattr and was removed)
 449                  * b) -ENODATA (there was no POSIX ACL xattr)
 450                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 451                  *
 452                  * There are various other error values that could effectively
 453                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 454                  * if the xattr name is too long), but the set of filesystems
 455                  * allowed as upper are limited to "normal" ones, where checking
 456                  * for the above two errors is sufficient.
 457                  */
 458                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 459                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 460                         goto out_dput;
 461
 462                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 463                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 464                         goto out_dput;
 465
 466                 /* Clear any inherited mode bits */
 467                 inode_lock(work->d_inode);
 468                 err = notify_change(work, &attr, NULL);
 469                 inode_unlock(work->d_inode);
 470                 if (err)
 471                         goto out_dput;
 472         }
 473 out_unlock:
 474         inode_unlock(dir);
 475         mnt_drop_write(mnt);
 476
 477         return work;
 478
 479 out_dput:
 480         dput(work);
 481         work = ERR_PTR(err);
 482         goto out_unlock;
 483 }
 484
 485 static void ovl_unescape(char *s)
 486 {
 487         char *d = s;
 488
 489         for (;; s++, d++) {
 490                 if (*s == '\\')
 491                         s++;
 492                 *d = *s;
 493                 if (!*s)
 494                         break;
 495         }
 496 }
 497
 498 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 499 {
 500         int err = -EINVAL;
 501
 502         if (!*name) {
 503                 pr_err("overlayfs: empty lowerdir\n");
 504                 goto out;
 505         }
 506         err = kern_path(name, LOOKUP_FOLLOW, path);
 507         if (err) {
 508                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 509                 goto out;
 510         }
 511         err = -EINVAL;
 512         if (ovl_dentry_weird(path->dentry)) {
 513                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 514                 goto out_put;
 515         }
 516         if (!d_is_dir(path->dentry)) {
 517                 pr_err("overlayfs: '%s' not a directory\n", name);
 518                 goto out_put;
 519         }
 520         return 0;
 521
 522 out_put:
 523         path_put(path);
 524 out:
 525         return err;
 526 }
 527
 528 static int ovl_mount_dir(const char *name, struct path *path)
 529 {
 530         int err = -ENOMEM;
 531         char *tmp = kstrdup(name, GFP_KERNEL);
 532
 533         if (tmp) {
 534                 ovl_unescape(tmp);
 535                 err = ovl_mount_dir_noesc(tmp, path);
 536
 537                 if (!err)
 538                         if (ovl_dentry_remote(path->dentry)) {
 539                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 540                                        tmp);
 541                                 path_put(path);
 542                                 err = -EINVAL;
 543                         }
 544                 kfree(tmp);
 545         }
 546         return err;
 547 }
 548
 549 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 550                              const char *name)
 551 {
 552         struct kstatfs statfs;
 553         int err = vfs_statfs(path, &statfs);
 554
 555         if (err)
 556                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 557         else
 558                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 559
 560         return err;
 561 }
 562
 563 static int ovl_lower_dir(const char *name, struct path *path,
 564                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 565 {
 566         int err;
 567
 568         err = ovl_mount_dir_noesc(name, path);
 569         if (err)
 570                 goto out;
 571
 572         err = ovl_check_namelen(path, ofs, name);
 573         if (err)
 574                 goto out_put;
 575
 576         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 577
 578         if (ovl_dentry_remote(path->dentry))
 579                 *remote = true;
 580
 581         return 0;
 582
 583 out_put:
 584         path_put(path);
 585 out:
 586         return err;
 587 }
 588
 589 /* Workdir should not be subdir of upperdir and vice versa */
 590 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 591 {
 592         bool ok = false;
 593
 594         if (workdir != upperdir) {
 595                 ok = (lock_rename(workdir, upperdir) == NULL);
 596                 unlock_rename(workdir, upperdir);
 597         }
 598         return ok;
 599 }
 600
 601 static unsigned int ovl_split_lowerdirs(char *str)
 602 {
 603         unsigned int ctr = 1;
 604         char *s, *d;
 605
 606         for (s = d = str;; s++, d++) {
 607                 if (*s == '\\') {
 608                         s++;
 609                 } else if (*s == ':') {
 610                         *d = '\0';
 611                         ctr++;
 612                         continue;
 613                 }
 614                 *d = *s;
 615                 if (!*s)
 616                         break;
 617         }
 618         return ctr;
 619 }
 620
 621 static int __maybe_unused
 622 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 623                         struct dentry *dentry, struct inode *inode,
 624                         const char *name, void *buffer, size_t size)
 625 {
 626         return ovl_xattr_get(dentry, handler->name, buffer, size);
 627 }
 628
 629 static int __maybe_unused
 630 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 631                         struct dentry *dentry, struct inode *inode,
 632                         const char *name, const void *value,
 633                         size_t size, int flags)
 634 {
 635         struct dentry *workdir = ovl_workdir(dentry);
 636         struct inode *realinode = ovl_inode_real(inode, NULL);
 637         struct posix_acl *acl = NULL;
 638         int err;
 639
 640         /* Check that everything is OK before copy-up */
 641         if (value) {
 642                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 643                 if (IS_ERR(acl))
 644                         return PTR_ERR(acl);
 645         }
 646         err = -EOPNOTSUPP;
 647         if (!IS_POSIXACL(d_inode(workdir)))
 648                 goto out_acl_release;
 649         if (!realinode->i_op->set_acl)
 650                 goto out_acl_release;
 651         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 652                 err = acl ? -EACCES : 0;
 653                 goto out_acl_release;
 654         }
 655         err = -EPERM;
 656         if (!inode_owner_or_capable(inode))
 657                 goto out_acl_release;
 658
 659         posix_acl_release(acl);
 660
 661         /*
 662          * Check if sgid bit needs to be cleared (actual setacl operation will
 663          * be done with mounter's capabilities and so that won't do it for us).
 664          */
 665         if (unlikely(inode->i_mode & S_ISGID) &&
 666             handler->flags == ACL_TYPE_ACCESS &&
 667             !in_group_p(inode->i_gid) &&
 668             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 669                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 670
 671                 err = ovl_setattr(dentry, &iattr);
 672                 if (err)
 673                         return err;
 674         }
 675
 676         err = ovl_xattr_set(dentry, handler->name, value, size, flags);
 677         if (!err)
 678                 ovl_copyattr(ovl_inode_real(inode, NULL), inode);
 679
 680         return err;
 681
 682 out_acl_release:
 683         posix_acl_release(acl);
 684         return err;
 685 }
 686
 687 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 688                              struct dentry *dentry, struct inode *inode,
 689                              const char *name, void *buffer, size_t size)
 690 {
 691         return -EOPNOTSUPP;
 692 }
 693
 694 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 695                              struct dentry *dentry, struct inode *inode,
 696                              const char *name, const void *value,
 697                              size_t size, int flags)
 698 {
 699         return -EOPNOTSUPP;
 700 }
 701
 702 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 703                                struct dentry *dentry, struct inode *inode,
 704                                const char *name, void *buffer, size_t size)
 705 {
 706         return ovl_xattr_get(dentry, name, buffer, size);
 707 }
 708
 709 static int ovl_other_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         return ovl_xattr_set(dentry, name, value, size, flags);
 715 }
 716
 717 static const struct xattr_handler __maybe_unused
 718 ovl_posix_acl_access_xattr_handler = {
 719         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 720         .flags = ACL_TYPE_ACCESS,
 721         .get = ovl_posix_acl_xattr_get,
 722         .set = ovl_posix_acl_xattr_set,
 723 };
 724
 725 static const struct xattr_handler __maybe_unused
 726 ovl_posix_acl_default_xattr_handler = {
 727         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 728         .flags = ACL_TYPE_DEFAULT,
 729         .get = ovl_posix_acl_xattr_get,
 730         .set = ovl_posix_acl_xattr_set,
 731 };
 732
 733 static const struct xattr_handler ovl_own_xattr_handler = {
 734         .prefix = OVL_XATTR_PREFIX,
 735         .get = ovl_own_xattr_get,
 736         .set = ovl_own_xattr_set,
 737 };
 738
 739 static const struct xattr_handler ovl_other_xattr_handler = {
 740         .prefix = "", /* catch all */
 741         .get = ovl_other_xattr_get,
 742         .set = ovl_other_xattr_set,
 743 };
 744
 745 static const struct xattr_handler *ovl_xattr_handlers[] = {
 746 #ifdef CONFIG_FS_POSIX_ACL
 747         &ovl_posix_acl_access_xattr_handler,
 748         &ovl_posix_acl_default_xattr_handler,
 749 #endif
 750         &ovl_own_xattr_handler,
 751         &ovl_other_xattr_handler,
 752         NULL
 753 };
 754
 755 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 756 {
 757         struct path upperpath = { };
 758         struct path workpath = { };
 759         struct dentry *root_dentry;
 760         struct inode *realinode;
 761         struct ovl_entry *oe;
 762         struct ovl_fs *ufs;
 763         struct path *stack = NULL;
 764         char *lowertmp;
 765         char *lower;
 766         unsigned int numlower;
 767         unsigned int stacklen = 0;
 768         unsigned int i;
 769         bool remote = false;
 770         struct cred *cred;
 771         int err;
 772
 773         err = -ENOMEM;
 774         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 775         if (!ufs)
 776                 goto out;
 777
 778         init_waitqueue_head(&ufs->copyup_wq);
 779         ufs->config.redirect_dir = ovl_redirect_dir_def;
 780         err = ovl_parse_opt((char *) data, &ufs->config);
 781         if (err)
 782                 goto out_free_config;
 783
 784         err = -EINVAL;
 785         if (!ufs->config.lowerdir) {
 786                 if (!silent)
 787                         pr_err("overlayfs: missing 'lowerdir'\n");
 788                 goto out_free_config;
 789         }
 790
 791         sb->s_stack_depth = 0;
 792         sb->s_maxbytes = MAX_LFS_FILESIZE;
 793         if (ufs->config.upperdir) {
 794                 if (!ufs->config.workdir) {
 795                         pr_err("overlayfs: missing 'workdir'\n");
 796                         goto out_free_config;
 797                 }
 798
 799                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 800                 if (err)
 801                         goto out_free_config;
 802
 803                 /* Upper fs should not be r/o */
 804                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 805                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 806                         err = -EINVAL;
 807                         goto out_put_upperpath;
 808                 }
 809
 810                 err = ovl_check_namelen(&upperpath, ufs, ufs->config.upperdir);
 811                 if (err)
 812                         goto out_put_upperpath;
 813
 814                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 815                 if (err)
 816                         goto out_put_upperpath;
 817
 818                 err = -EINVAL;
 819                 if (upperpath.mnt != workpath.mnt) {
 820                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 821                         goto out_put_workpath;
 822                 }
 823                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 824                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 825                         goto out_put_workpath;
 826                 }
 827                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 828         }
 829         err = -ENOMEM;
 830         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 831         if (!lowertmp)
 832                 goto out_put_workpath;
 833
 834         err = -EINVAL;
 835         stacklen = ovl_split_lowerdirs(lowertmp);
 836         if (stacklen > OVL_MAX_STACK) {
 837                 pr_err("overlayfs: too many lower directories, limit is %d\n",
 838                        OVL_MAX_STACK);
 839                 goto out_free_lowertmp;
 840         } else if (!ufs->config.upperdir && stacklen == 1) {
 841                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 842                 goto out_free_lowertmp;
 843         }
 844
 845         err = -ENOMEM;
 846         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 847         if (!stack)
 848                 goto out_free_lowertmp;
 849
 850         err = -EINVAL;
 851         lower = lowertmp;
 852         for (numlower = 0; numlower < stacklen; numlower++) {
 853                 err = ovl_lower_dir(lower, &stack[numlower], ufs,
 854                                     &sb->s_stack_depth, &remote);
 855                 if (err)
 856                         goto out_put_lowerpath;
 857
 858                 lower = strchr(lower, '\0') + 1;
 859         }
 860
 861         err = -EINVAL;
 862         sb->s_stack_depth++;
 863         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
 864                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
 865                 goto out_put_lowerpath;
 866         }
 867
 868         if (ufs->config.upperdir) {
 869                 ufs->upper_mnt = clone_private_mount(&upperpath);
 870                 err = PTR_ERR(ufs->upper_mnt);
 871                 if (IS_ERR(ufs->upper_mnt)) {
 872                         pr_err("overlayfs: failed to clone upperpath\n");
 873                         goto out_put_lowerpath;
 874                 }
 875                 /* Don't inherit atime flags */
 876                 ufs->upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
 877
 878                 sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
 879
 880                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
 881                 err = PTR_ERR(ufs->workdir);
 882                 if (IS_ERR(ufs->workdir)) {
 883                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 884                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
 885                         sb->s_flags |= MS_RDONLY;
 886                         ufs->workdir = NULL;
 887                 }
 888
 889                 /*
 890                  * Upper should support d_type, else whiteouts are visible.
 891                  * Given workdir and upper are on same fs, we can do
 892                  * iterate_dir() on workdir. This check requires successful
 893                  * creation of workdir in previous step.
 894                  */
 895                 if (ufs->workdir) {
 896                         struct dentry *temp;
 897
 898                         err = ovl_check_d_type_supported(&workpath);
 899                         if (err < 0)
 900                                 goto out_put_workdir;
 901
 902                         /*
 903                          * We allowed this configuration and don't want to
 904                          * break users over kernel upgrade. So warn instead
 905                          * of erroring out.
 906                          */
 907                         if (!err)
 908                                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
 909
 910                         /* Check if upper/work fs supports O_TMPFILE */
 911                         temp = ovl_do_tmpfile(ufs->workdir, S_IFREG | 0);
 912                         ufs->tmpfile = !IS_ERR(temp);
 913                         if (ufs->tmpfile)
 914                                 dput(temp);
 915                         else
 916                                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
 917
 918                         /*
 919                          * Check if upper/work fs supports trusted.overlay.*
 920                          * xattr
 921                          */
 922                         err = ovl_do_setxattr(ufs->workdir, OVL_XATTR_OPAQUE,
 923                                               "0", 1, 0);
 924                         if (err) {
 925                                 ufs->noxattr = true;
 926                                 pr_warn("overlayfs: upper fs does not support xattr.\n");
 927                         } else {
 928                                 vfs_removexattr(ufs->workdir, OVL_XATTR_OPAQUE);
 929                         }
 930                 }
 931         }
 932
 933         err = -ENOMEM;
 934         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
 935         if (ufs->lower_mnt == NULL)
 936                 goto out_put_workdir;
 937         for (i = 0; i < numlower; i++) {
 938                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
 939
 940                 err = PTR_ERR(mnt);
 941                 if (IS_ERR(mnt)) {
 942                         pr_err("overlayfs: failed to clone lowerpath\n");
 943                         goto out_put_lower_mnt;
 944                 }
 945                 /*
 946                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
 947                  * will fail instead of modifying lower fs.
 948                  */
 949                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
 950
 951                 ufs->lower_mnt[ufs->numlower] = mnt;
 952                 ufs->numlower++;
 953
 954                 /* Check if all lower layers are on same sb */
 955                 if (i == 0)
 956                         ufs->same_sb = mnt->mnt_sb;
 957                 else if (ufs->same_sb != mnt->mnt_sb)
 958                         ufs->same_sb = NULL;
 959         }
 960
 961         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
 962         if (!ufs->upper_mnt)
 963                 sb->s_flags |= MS_RDONLY;
 964         else if (ufs->upper_mnt->mnt_sb != ufs->same_sb)
 965                 ufs->same_sb = NULL;
 966
 967         if (remote)
 968                 sb->s_d_op = &ovl_reval_dentry_operations;
 969         else
 970                 sb->s_d_op = &ovl_dentry_operations;
 971
 972         ufs->creator_cred = cred = prepare_creds();
 973         if (!cred)
 974                 goto out_put_lower_mnt;
 975
 976         /* Never override disk quota limits or use reserved space */
 977         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
 978
 979         err = -ENOMEM;
 980         oe = ovl_alloc_entry(numlower);
 981         if (!oe)
 982                 goto out_put_cred;
 983
 984         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
 985         sb->s_op = &ovl_super_operations;
 986         sb->s_xattr = ovl_xattr_handlers;
 987         sb->s_fs_info = ufs;
 988         sb->s_flags |= MS_POSIXACL | MS_NOREMOTELOCK;
 989
 990         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
 991         if (!root_dentry)
 992                 goto out_free_oe;
 993
 994         mntput(upperpath.mnt);
 995         for (i = 0; i < numlower; i++)
 996                 mntput(stack[i].mnt);
 997         path_put(&workpath);
 998         kfree(lowertmp);
 999
1000         if (upperpath.dentry) {
1001                 oe->__upperdentry = upperpath.dentry;
1002                 oe->impure = ovl_is_impuredir(upperpath.dentry);
1003         }
1004         for (i = 0; i < numlower; i++) {
1005                 oe->lowerstack[i].dentry = stack[i].dentry;
1006                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1007         }
1008         kfree(stack);
1009
1010         root_dentry->d_fsdata = oe;
1011
1012         realinode = d_inode(ovl_dentry_real(root_dentry));
1013         ovl_inode_init(d_inode(root_dentry), realinode, !!upperpath.dentry);
1014         ovl_copyattr(realinode, d_inode(root_dentry));
1015
1016         sb->s_root = root_dentry;
1017
1018         return 0;
1019
1020 out_free_oe:
1021         kfree(oe);
1022 out_put_cred:
1023         put_cred(ufs->creator_cred);
1024 out_put_lower_mnt:
1025         for (i = 0; i < ufs->numlower; i++)
1026                 mntput(ufs->lower_mnt[i]);
1027         kfree(ufs->lower_mnt);
1028 out_put_workdir:
1029         dput(ufs->workdir);
1030         mntput(ufs->upper_mnt);
1031 out_put_lowerpath:
1032         for (i = 0; i < numlower; i++)
1033                 path_put(&stack[i]);
1034         kfree(stack);
1035 out_free_lowertmp:
1036         kfree(lowertmp);
1037 out_put_workpath:
1038         path_put(&workpath);
1039 out_put_upperpath:
1040         path_put(&upperpath);
1041 out_free_config:
1042         kfree(ufs->config.lowerdir);
1043         kfree(ufs->config.upperdir);
1044         kfree(ufs->config.workdir);
1045         kfree(ufs);
1046 out:
1047         return err;
1048 }
1049
1050 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1051                                 const char *dev_name, void *raw_data)
1052 {
1053         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1054 }
1055
1056 static struct file_system_type ovl_fs_type = {
1057         .owner          = THIS_MODULE,
1058         .name           = "overlay",
1059         .mount          = ovl_mount,
1060         .kill_sb        = kill_anon_super,
1061 };
1062 MODULE_ALIAS_FS("overlay");
1063
1064 static void ovl_inode_init_once(void *foo)
1065 {
1066         struct ovl_inode *oi = foo;
1067
1068         inode_init_once(&oi->vfs_inode);
1069 }
1070
1071 static int __init ovl_init(void)
1072 {
1073         int err;
1074
1075         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1076                                              sizeof(struct ovl_inode), 0,
1077                                              (SLAB_RECLAIM_ACCOUNT|
1078                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1079                                              ovl_inode_init_once);
1080         if (ovl_inode_cachep == NULL)
1081                 return -ENOMEM;
1082
1083         err = register_filesystem(&ovl_fs_type);
1084         if (err)
1085                 kmem_cache_destroy(ovl_inode_cachep);
1086
1087         return err;
1088 }
1089
1090 static void __exit ovl_exit(void)
1091 {
1092         unregister_filesystem(&ovl_fs_type);
1093
1094         /*
1095          * Make sure all delayed rcu free inodes are flushed before we
1096          * destroy cache.
1097          */
1098         rcu_barrier();
1099         kmem_cache_destroy(ovl_inode_cachep);
1100
1101 }
1102
1103 module_init(ovl_init);
1104 module_exit(ovl_exit);