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 void ovl_put_super(struct super_block *sb)
 169 {
 170         struct ovl_fs *ufs = sb->s_fs_info;
 171         unsigned i;
 172
 173         dput(ufs->workdir);
 174         mntput(ufs->upper_mnt);
 175         for (i = 0; i < ufs->numlower; i++)
 176                 mntput(ufs->lower_mnt[i]);
 177         kfree(ufs->lower_mnt);
 178
 179         kfree(ufs->config.lowerdir);
 180         kfree(ufs->config.upperdir);
 181         kfree(ufs->config.workdir);
 182         put_cred(ufs->creator_cred);
 183         kfree(ufs);
 184 }
 185
 186 static int ovl_sync_fs(struct super_block *sb, int wait)
 187 {
 188         struct ovl_fs *ufs = sb->s_fs_info;
 189         struct super_block *upper_sb;
 190         int ret;
 191
 192         if (!ufs->upper_mnt)
 193                 return 0;
 194         upper_sb = ufs->upper_mnt->mnt_sb;
 195         if (!upper_sb->s_op->sync_fs)
 196                 return 0;
 197
 198         /* real inodes have already been synced by sync_filesystem(ovl_sb) */
 199         down_read(&upper_sb->s_umount);
 200         ret = upper_sb->s_op->sync_fs(upper_sb, wait);
 201         up_read(&upper_sb->s_umount);
 202         return ret;
 203 }
 204
 205 /**
 206  * ovl_statfs
 207  * @sb: The overlayfs super block
 208  * @buf: The struct kstatfs to fill in with stats
 209  *
 210  * Get the filesystem statistics.  As writes always target the upper layer
 211  * filesystem pass the statfs to the upper filesystem (if it exists)
 212  */
 213 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 214 {
 215         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 216         struct dentry *root_dentry = dentry->d_sb->s_root;
 217         struct path path;
 218         int err;
 219
 220         ovl_path_real(root_dentry, &path);
 221
 222         err = vfs_statfs(&path, buf);
 223         if (!err) {
 224                 buf->f_namelen = ofs->namelen;
 225                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 226         }
 227
 228         return err;
 229 }
 230
 231 /**
 232  * ovl_show_options
 233  *
 234  * Prints the mount options for a given superblock.
 235  * Returns zero; does not fail.
 236  */
 237 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 238 {
 239         struct super_block *sb = dentry->d_sb;
 240         struct ovl_fs *ufs = sb->s_fs_info;
 241
 242         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 243         if (ufs->config.upperdir) {
 244                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 245                 seq_show_option(m, "workdir", ufs->config.workdir);
 246         }
 247         if (ufs->config.default_permissions)
 248                 seq_puts(m, ",default_permissions");
 249         if (ufs->config.redirect_dir != ovl_redirect_dir_def)
 250                 seq_printf(m, ",redirect_dir=%s",
 251                            ufs->config.redirect_dir ? "on" : "off");
 252         return 0;
 253 }
 254
 255 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 256 {
 257         struct ovl_fs *ufs = sb->s_fs_info;
 258
 259         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 260                 return -EROFS;
 261
 262         return 0;
 263 }
 264
 265 static const struct super_operations ovl_super_operations = {
 266         .put_super      = ovl_put_super,
 267         .sync_fs        = ovl_sync_fs,
 268         .statfs         = ovl_statfs,
 269         .show_options   = ovl_show_options,
 270         .remount_fs     = ovl_remount,
 271         .drop_inode     = generic_delete_inode,
 272 };
 273
 274 enum {
 275         OPT_LOWERDIR,
 276         OPT_UPPERDIR,
 277         OPT_WORKDIR,
 278         OPT_DEFAULT_PERMISSIONS,
 279         OPT_REDIRECT_DIR_ON,
 280         OPT_REDIRECT_DIR_OFF,
 281         OPT_ERR,
 282 };
 283
 284 static const match_table_t ovl_tokens = {
 285         {OPT_LOWERDIR,                  "lowerdir=%s"},
 286         {OPT_UPPERDIR,                  "upperdir=%s"},
 287         {OPT_WORKDIR,                   "workdir=%s"},
 288         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 289         {OPT_REDIRECT_DIR_ON,           "redirect_dir=on"},
 290         {OPT_REDIRECT_DIR_OFF,          "redirect_dir=off"},
 291         {OPT_ERR,                       NULL}
 292 };
 293
 294 static char *ovl_next_opt(char **s)
 295 {
 296         char *sbegin = *s;
 297         char *p;
 298
 299         if (sbegin == NULL)
 300                 return NULL;
 301
 302         for (p = sbegin; *p; p++) {
 303                 if (*p == '\\') {
 304                         p++;
 305                         if (!*p)
 306                                 break;
 307                 } else if (*p == ',') {
 308                         *p = '\0';
 309                         *s = p + 1;
 310                         return sbegin;
 311                 }
 312         }
 313         *s = NULL;
 314         return sbegin;
 315 }
 316
 317 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 318 {
 319         char *p;
 320
 321         while ((p = ovl_next_opt(&opt)) != NULL) {
 322                 int token;
 323                 substring_t args[MAX_OPT_ARGS];
 324
 325                 if (!*p)
 326                         continue;
 327
 328                 token = match_token(p, ovl_tokens, args);
 329                 switch (token) {
 330                 case OPT_UPPERDIR:
 331                         kfree(config->upperdir);
 332                         config->upperdir = match_strdup(&args[0]);
 333                         if (!config->upperdir)
 334                                 return -ENOMEM;
 335                         break;
 336
 337                 case OPT_LOWERDIR:
 338                         kfree(config->lowerdir);
 339                         config->lowerdir = match_strdup(&args[0]);
 340                         if (!config->lowerdir)
 341                                 return -ENOMEM;
 342                         break;
 343
 344                 case OPT_WORKDIR:
 345                         kfree(config->workdir);
 346                         config->workdir = match_strdup(&args[0]);
 347                         if (!config->workdir)
 348                                 return -ENOMEM;
 349                         break;
 350
 351                 case OPT_DEFAULT_PERMISSIONS:
 352                         config->default_permissions = true;
 353                         break;
 354
 355                 case OPT_REDIRECT_DIR_ON:
 356                         config->redirect_dir = true;
 357                         break;
 358
 359                 case OPT_REDIRECT_DIR_OFF:
 360                         config->redirect_dir = false;
 361                         break;
 362
 363                 default:
 364                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 365                         return -EINVAL;
 366                 }
 367         }
 368
 369         /* Workdir is useless in non-upper mount */
 370         if (!config->upperdir && config->workdir) {
 371                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 372                         config->workdir);
 373                 kfree(config->workdir);
 374                 config->workdir = NULL;
 375         }
 376
 377         return 0;
 378 }
 379
 380 #define OVL_WORKDIR_NAME "work"
 381
 382 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 383                                          struct dentry *dentry)
 384 {
 385         struct inode *dir = dentry->d_inode;
 386         struct dentry *work;
 387         int err;
 388         bool retried = false;
 389
 390         err = mnt_want_write(mnt);
 391         if (err)
 392                 return ERR_PTR(err);
 393
 394         inode_lock_nested(dir, I_MUTEX_PARENT);
 395 retry:
 396         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 397                               strlen(OVL_WORKDIR_NAME));
 398
 399         if (!IS_ERR(work)) {
 400                 struct iattr attr = {
 401                         .ia_valid = ATTR_MODE,
 402                         .ia_mode = S_IFDIR | 0,
 403                 };
 404
 405                 if (work->d_inode) {
 406                         err = -EEXIST;
 407                         if (retried)
 408                                 goto out_dput;
 409
 410                         retried = true;
 411                         ovl_workdir_cleanup(dir, mnt, work, 0);
 412                         dput(work);
 413                         goto retry;
 414                 }
 415
 416                 err = ovl_create_real(dir, work,
 417                                       &(struct cattr){.mode = S_IFDIR | 0},
 418                                       NULL, true);
 419                 if (err)
 420                         goto out_dput;
 421
 422                 /*
 423                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 424                  *
 425                  * a) success (there was a POSIX ACL xattr and was removed)
 426                  * b) -ENODATA (there was no POSIX ACL xattr)
 427                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 428                  *
 429                  * There are various other error values that could effectively
 430                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 431                  * if the xattr name is too long), but the set of filesystems
 432                  * allowed as upper are limited to "normal" ones, where checking
 433                  * for the above two errors is sufficient.
 434                  */
 435                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 436                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 437                         goto out_dput;
 438
 439                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 440                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 441                         goto out_dput;
 442
 443                 /* Clear any inherited mode bits */
 444                 inode_lock(work->d_inode);
 445                 err = notify_change(work, &attr, NULL);
 446                 inode_unlock(work->d_inode);
 447                 if (err)
 448                         goto out_dput;
 449         }
 450 out_unlock:
 451         inode_unlock(dir);
 452         mnt_drop_write(mnt);
 453
 454         return work;
 455
 456 out_dput:
 457         dput(work);
 458         work = ERR_PTR(err);
 459         goto out_unlock;
 460 }
 461
 462 static void ovl_unescape(char *s)
 463 {
 464         char *d = s;
 465
 466         for (;; s++, d++) {
 467                 if (*s == '\\')
 468                         s++;
 469                 *d = *s;
 470                 if (!*s)
 471                         break;
 472         }
 473 }
 474
 475 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 476 {
 477         int err = -EINVAL;
 478
 479         if (!*name) {
 480                 pr_err("overlayfs: empty lowerdir\n");
 481                 goto out;
 482         }
 483         err = kern_path(name, LOOKUP_FOLLOW, path);
 484         if (err) {
 485                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 486                 goto out;
 487         }
 488         err = -EINVAL;
 489         if (ovl_dentry_weird(path->dentry)) {
 490                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 491                 goto out_put;
 492         }
 493         if (!d_is_dir(path->dentry)) {
 494                 pr_err("overlayfs: '%s' not a directory\n", name);
 495                 goto out_put;
 496         }
 497         return 0;
 498
 499 out_put:
 500         path_put(path);
 501 out:
 502         return err;
 503 }
 504
 505 static int ovl_mount_dir(const char *name, struct path *path)
 506 {
 507         int err = -ENOMEM;
 508         char *tmp = kstrdup(name, GFP_KERNEL);
 509
 510         if (tmp) {
 511                 ovl_unescape(tmp);
 512                 err = ovl_mount_dir_noesc(tmp, path);
 513
 514                 if (!err)
 515                         if (ovl_dentry_remote(path->dentry)) {
 516                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 517                                        tmp);
 518                                 path_put(path);
 519                                 err = -EINVAL;
 520                         }
 521                 kfree(tmp);
 522         }
 523         return err;
 524 }
 525
 526 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 527                              const char *name)
 528 {
 529         struct kstatfs statfs;
 530         int err = vfs_statfs(path, &statfs);
 531
 532         if (err)
 533                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 534         else
 535                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 536
 537         return err;
 538 }
 539
 540 static int ovl_lower_dir(const char *name, struct path *path,
 541                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 542 {
 543         int err;
 544
 545         err = ovl_mount_dir_noesc(name, path);
 546         if (err)
 547                 goto out;
 548
 549         err = ovl_check_namelen(path, ofs, name);
 550         if (err)
 551                 goto out_put;
 552
 553         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 554
 555         if (ovl_dentry_remote(path->dentry))
 556                 *remote = true;
 557
 558         return 0;
 559
 560 out_put:
 561         path_put(path);
 562 out:
 563         return err;
 564 }
 565
 566 /* Workdir should not be subdir of upperdir and vice versa */
 567 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 568 {
 569         bool ok = false;
 570
 571         if (workdir != upperdir) {
 572                 ok = (lock_rename(workdir, upperdir) == NULL);
 573                 unlock_rename(workdir, upperdir);
 574         }
 575         return ok;
 576 }
 577
 578 static unsigned int ovl_split_lowerdirs(char *str)
 579 {
 580         unsigned int ctr = 1;
 581         char *s, *d;
 582
 583         for (s = d = str;; s++, d++) {
 584                 if (*s == '\\') {
 585                         s++;
 586                 } else if (*s == ':') {
 587                         *d = '\0';
 588                         ctr++;
 589                         continue;
 590                 }
 591                 *d = *s;
 592                 if (!*s)
 593                         break;
 594         }
 595         return ctr;
 596 }
 597
 598 static int __maybe_unused
 599 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 600                         struct dentry *dentry, struct inode *inode,
 601                         const char *name, void *buffer, size_t size)
 602 {
 603         return ovl_xattr_get(dentry, handler->name, buffer, size);
 604 }
 605
 606 static int __maybe_unused
 607 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 608                         struct dentry *dentry, struct inode *inode,
 609                         const char *name, const void *value,
 610                         size_t size, int flags)
 611 {
 612         struct dentry *workdir = ovl_workdir(dentry);
 613         struct inode *realinode = ovl_inode_real(inode, NULL);
 614         struct posix_acl *acl = NULL;
 615         int err;
 616
 617         /* Check that everything is OK before copy-up */
 618         if (value) {
 619                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 620                 if (IS_ERR(acl))
 621                         return PTR_ERR(acl);
 622         }
 623         err = -EOPNOTSUPP;
 624         if (!IS_POSIXACL(d_inode(workdir)))
 625                 goto out_acl_release;
 626         if (!realinode->i_op->set_acl)
 627                 goto out_acl_release;
 628         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 629                 err = acl ? -EACCES : 0;
 630                 goto out_acl_release;
 631         }
 632         err = -EPERM;
 633         if (!inode_owner_or_capable(inode))
 634                 goto out_acl_release;
 635
 636         posix_acl_release(acl);
 637
 638         /*
 639          * Check if sgid bit needs to be cleared (actual setacl operation will
 640          * be done with mounter's capabilities and so that won't do it for us).
 641          */
 642         if (unlikely(inode->i_mode & S_ISGID) &&
 643             handler->flags == ACL_TYPE_ACCESS &&
 644             !in_group_p(inode->i_gid) &&
 645             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 646                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 647
 648                 err = ovl_setattr(dentry, &iattr);
 649                 if (err)
 650                         return err;
 651         }
 652
 653         err = ovl_xattr_set(dentry, handler->name, value, size, flags);
 654         if (!err)
 655                 ovl_copyattr(ovl_inode_real(inode, NULL), inode);
 656
 657         return err;
 658
 659 out_acl_release:
 660         posix_acl_release(acl);
 661         return err;
 662 }
 663
 664 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 665                              struct dentry *dentry, struct inode *inode,
 666                              const char *name, void *buffer, size_t size)
 667 {
 668         return -EOPNOTSUPP;
 669 }
 670
 671 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 672                              struct dentry *dentry, struct inode *inode,
 673                              const char *name, const void *value,
 674                              size_t size, int flags)
 675 {
 676         return -EOPNOTSUPP;
 677 }
 678
 679 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 680                                struct dentry *dentry, struct inode *inode,
 681                                const char *name, void *buffer, size_t size)
 682 {
 683         return ovl_xattr_get(dentry, name, buffer, size);
 684 }
 685
 686 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 687                                struct dentry *dentry, struct inode *inode,
 688                                const char *name, const void *value,
 689                                size_t size, int flags)
 690 {
 691         return ovl_xattr_set(dentry, name, value, size, flags);
 692 }
 693
 694 static const struct xattr_handler __maybe_unused
 695 ovl_posix_acl_access_xattr_handler = {
 696         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 697         .flags = ACL_TYPE_ACCESS,
 698         .get = ovl_posix_acl_xattr_get,
 699         .set = ovl_posix_acl_xattr_set,
 700 };
 701
 702 static const struct xattr_handler __maybe_unused
 703 ovl_posix_acl_default_xattr_handler = {
 704         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 705         .flags = ACL_TYPE_DEFAULT,
 706         .get = ovl_posix_acl_xattr_get,
 707         .set = ovl_posix_acl_xattr_set,
 708 };
 709
 710 static const struct xattr_handler ovl_own_xattr_handler = {
 711         .prefix = OVL_XATTR_PREFIX,
 712         .get = ovl_own_xattr_get,
 713         .set = ovl_own_xattr_set,
 714 };
 715
 716 static const struct xattr_handler ovl_other_xattr_handler = {
 717         .prefix = "", /* catch all */
 718         .get = ovl_other_xattr_get,
 719         .set = ovl_other_xattr_set,
 720 };
 721
 722 static const struct xattr_handler *ovl_xattr_handlers[] = {
 723 #ifdef CONFIG_FS_POSIX_ACL
 724         &ovl_posix_acl_access_xattr_handler,
 725         &ovl_posix_acl_default_xattr_handler,
 726 #endif
 727         &ovl_own_xattr_handler,
 728         &ovl_other_xattr_handler,
 729         NULL
 730 };
 731
 732 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 733 {
 734         struct path upperpath = { };
 735         struct path workpath = { };
 736         struct dentry *root_dentry;
 737         struct inode *realinode;
 738         struct ovl_entry *oe;
 739         struct ovl_fs *ufs;
 740         struct path *stack = NULL;
 741         char *lowertmp;
 742         char *lower;
 743         unsigned int numlower;
 744         unsigned int stacklen = 0;
 745         unsigned int i;
 746         bool remote = false;
 747         struct cred *cred;
 748         int err;
 749
 750         err = -ENOMEM;
 751         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 752         if (!ufs)
 753                 goto out;
 754
 755         init_waitqueue_head(&ufs->copyup_wq);
 756         ufs->config.redirect_dir = ovl_redirect_dir_def;
 757         err = ovl_parse_opt((char *) data, &ufs->config);
 758         if (err)
 759                 goto out_free_config;
 760
 761         err = -EINVAL;
 762         if (!ufs->config.lowerdir) {
 763                 if (!silent)
 764                         pr_err("overlayfs: missing 'lowerdir'\n");
 765                 goto out_free_config;
 766         }
 767
 768         sb->s_stack_depth = 0;
 769         sb->s_maxbytes = MAX_LFS_FILESIZE;
 770         if (ufs->config.upperdir) {
 771                 if (!ufs->config.workdir) {
 772                         pr_err("overlayfs: missing 'workdir'\n");
 773                         goto out_free_config;
 774                 }
 775
 776                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 777                 if (err)
 778                         goto out_free_config;
 779
 780                 /* Upper fs should not be r/o */
 781                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 782                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 783                         err = -EINVAL;
 784                         goto out_put_upperpath;
 785                 }
 786
 787                 err = ovl_check_namelen(&upperpath, ufs, ufs->config.upperdir);
 788                 if (err)
 789                         goto out_put_upperpath;
 790
 791                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 792                 if (err)
 793                         goto out_put_upperpath;
 794
 795                 err = -EINVAL;
 796                 if (upperpath.mnt != workpath.mnt) {
 797                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 798                         goto out_put_workpath;
 799                 }
 800                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 801                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 802                         goto out_put_workpath;
 803                 }
 804                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 805         }
 806         err = -ENOMEM;
 807         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 808         if (!lowertmp)
 809                 goto out_put_workpath;
 810
 811         err = -EINVAL;
 812         stacklen = ovl_split_lowerdirs(lowertmp);
 813         if (stacklen > OVL_MAX_STACK) {
 814                 pr_err("overlayfs: too many lower directories, limit is %d\n",
 815                        OVL_MAX_STACK);
 816                 goto out_free_lowertmp;
 817         } else if (!ufs->config.upperdir && stacklen == 1) {
 818                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 819                 goto out_free_lowertmp;
 820         }
 821
 822         err = -ENOMEM;
 823         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 824         if (!stack)
 825                 goto out_free_lowertmp;
 826
 827         err = -EINVAL;
 828         lower = lowertmp;
 829         for (numlower = 0; numlower < stacklen; numlower++) {
 830                 err = ovl_lower_dir(lower, &stack[numlower], ufs,
 831                                     &sb->s_stack_depth, &remote);
 832                 if (err)
 833                         goto out_put_lowerpath;
 834
 835                 lower = strchr(lower, '\0') + 1;
 836         }
 837
 838         err = -EINVAL;
 839         sb->s_stack_depth++;
 840         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
 841                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
 842                 goto out_put_lowerpath;
 843         }
 844
 845         if (ufs->config.upperdir) {
 846                 ufs->upper_mnt = clone_private_mount(&upperpath);
 847                 err = PTR_ERR(ufs->upper_mnt);
 848                 if (IS_ERR(ufs->upper_mnt)) {
 849                         pr_err("overlayfs: failed to clone upperpath\n");
 850                         goto out_put_lowerpath;
 851                 }
 852                 /* Don't inherit atime flags */
 853                 ufs->upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
 854
 855                 sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
 856
 857                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
 858                 err = PTR_ERR(ufs->workdir);
 859                 if (IS_ERR(ufs->workdir)) {
 860                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 861                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
 862                         sb->s_flags |= MS_RDONLY;
 863                         ufs->workdir = NULL;
 864                 }
 865
 866                 /*
 867                  * Upper should support d_type, else whiteouts are visible.
 868                  * Given workdir and upper are on same fs, we can do
 869                  * iterate_dir() on workdir. This check requires successful
 870                  * creation of workdir in previous step.
 871                  */
 872                 if (ufs->workdir) {
 873                         struct dentry *temp;
 874
 875                         err = ovl_check_d_type_supported(&workpath);
 876                         if (err < 0)
 877                                 goto out_put_workdir;
 878
 879                         /*
 880                          * We allowed this configuration and don't want to
 881                          * break users over kernel upgrade. So warn instead
 882                          * of erroring out.
 883                          */
 884                         if (!err)
 885                                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
 886
 887                         /* Check if upper/work fs supports O_TMPFILE */
 888                         temp = ovl_do_tmpfile(ufs->workdir, S_IFREG | 0);
 889                         ufs->tmpfile = !IS_ERR(temp);
 890                         if (ufs->tmpfile)
 891                                 dput(temp);
 892                         else
 893                                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
 894
 895                         /*
 896                          * Check if upper/work fs supports trusted.overlay.*
 897                          * xattr
 898                          */
 899                         err = ovl_do_setxattr(ufs->workdir, OVL_XATTR_OPAQUE,
 900                                               "0", 1, 0);
 901                         if (err) {
 902                                 ufs->noxattr = true;
 903                                 pr_warn("overlayfs: upper fs does not support xattr.\n");
 904                         } else {
 905                                 vfs_removexattr(ufs->workdir, OVL_XATTR_OPAQUE);
 906                         }
 907                 }
 908         }
 909
 910         err = -ENOMEM;
 911         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
 912         if (ufs->lower_mnt == NULL)
 913                 goto out_put_workdir;
 914         for (i = 0; i < numlower; i++) {
 915                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
 916
 917                 err = PTR_ERR(mnt);
 918                 if (IS_ERR(mnt)) {
 919                         pr_err("overlayfs: failed to clone lowerpath\n");
 920                         goto out_put_lower_mnt;
 921                 }
 922                 /*
 923                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
 924                  * will fail instead of modifying lower fs.
 925                  */
 926                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
 927
 928                 ufs->lower_mnt[ufs->numlower] = mnt;
 929                 ufs->numlower++;
 930
 931                 /* Check if all lower layers are on same sb */
 932                 if (i == 0)
 933                         ufs->same_sb = mnt->mnt_sb;
 934                 else if (ufs->same_sb != mnt->mnt_sb)
 935                         ufs->same_sb = NULL;
 936         }
 937
 938         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
 939         if (!ufs->upper_mnt)
 940                 sb->s_flags |= MS_RDONLY;
 941         else if (ufs->upper_mnt->mnt_sb != ufs->same_sb)
 942                 ufs->same_sb = NULL;
 943
 944         if (remote)
 945                 sb->s_d_op = &ovl_reval_dentry_operations;
 946         else
 947                 sb->s_d_op = &ovl_dentry_operations;
 948
 949         ufs->creator_cred = cred = prepare_creds();
 950         if (!cred)
 951                 goto out_put_lower_mnt;
 952
 953         /* Never override disk quota limits or use reserved space */
 954         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
 955
 956         err = -ENOMEM;
 957         oe = ovl_alloc_entry(numlower);
 958         if (!oe)
 959                 goto out_put_cred;
 960
 961         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
 962         sb->s_op = &ovl_super_operations;
 963         sb->s_xattr = ovl_xattr_handlers;
 964         sb->s_fs_info = ufs;
 965         sb->s_flags |= MS_POSIXACL | MS_NOREMOTELOCK;
 966
 967         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
 968         if (!root_dentry)
 969                 goto out_free_oe;
 970
 971         mntput(upperpath.mnt);
 972         for (i = 0; i < numlower; i++)
 973                 mntput(stack[i].mnt);
 974         path_put(&workpath);
 975         kfree(lowertmp);
 976
 977         if (upperpath.dentry) {
 978                 oe->__upperdentry = upperpath.dentry;
 979                 oe->impure = ovl_is_impuredir(upperpath.dentry);
 980         }
 981         for (i = 0; i < numlower; i++) {
 982                 oe->lowerstack[i].dentry = stack[i].dentry;
 983                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
 984         }
 985         kfree(stack);
 986
 987         root_dentry->d_fsdata = oe;
 988
 989         realinode = d_inode(ovl_dentry_real(root_dentry));
 990         ovl_inode_init(d_inode(root_dentry), realinode, !!upperpath.dentry);
 991         ovl_copyattr(realinode, d_inode(root_dentry));
 992
 993         sb->s_root = root_dentry;
 994
 995         return 0;
 996
 997 out_free_oe:
 998         kfree(oe);
 999 out_put_cred:
1000         put_cred(ufs->creator_cred);
1001 out_put_lower_mnt:
1002         for (i = 0; i < ufs->numlower; i++)
1003                 mntput(ufs->lower_mnt[i]);
1004         kfree(ufs->lower_mnt);
1005 out_put_workdir:
1006         dput(ufs->workdir);
1007         mntput(ufs->upper_mnt);
1008 out_put_lowerpath:
1009         for (i = 0; i < numlower; i++)
1010                 path_put(&stack[i]);
1011         kfree(stack);
1012 out_free_lowertmp:
1013         kfree(lowertmp);
1014 out_put_workpath:
1015         path_put(&workpath);
1016 out_put_upperpath:
1017         path_put(&upperpath);
1018 out_free_config:
1019         kfree(ufs->config.lowerdir);
1020         kfree(ufs->config.upperdir);
1021         kfree(ufs->config.workdir);
1022         kfree(ufs);
1023 out:
1024         return err;
1025 }
1026
1027 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1028                                 const char *dev_name, void *raw_data)
1029 {
1030         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1031 }
1032
1033 static struct file_system_type ovl_fs_type = {
1034         .owner          = THIS_MODULE,
1035         .name           = "overlay",
1036         .mount          = ovl_mount,
1037         .kill_sb        = kill_anon_super,
1038 };
1039 MODULE_ALIAS_FS("overlay");
1040
1041 static int __init ovl_init(void)
1042 {
1043         return register_filesystem(&ovl_fs_type);
1044 }
1045
1046 static void __exit ovl_exit(void)
1047 {
1048         unregister_filesystem(&ovl_fs_type);
1049 }
1050
1051 module_init(ovl_init);
1052 module_exit(ovl_exit);