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