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