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