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 <linux/fs.h>
  11 #include <linux/namei.h>
  12 #include <linux/xattr.h>
  13 #include <linux/mount.h>
  14 #include <linux/parser.h>
  15 #include <linux/module.h>
  16 #include <linux/statfs.h>
  17 #include <linux/seq_file.h>
  18 #include <linux/posix_acl_xattr.h>
  19 #include "overlayfs.h"
  20 #include "ovl_entry.h"
  21
  22 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  23 MODULE_DESCRIPTION("Overlay filesystem");
  24 MODULE_LICENSE("GPL");
  25
  26
  27 struct ovl_dir_cache;
  28
  29 #define OVL_MAX_STACK 500
  30
  31 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  32 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  33 MODULE_PARM_DESC(ovl_redirect_dir_def,
  34                  "Default to on or off for the redirect_dir feature");
  35
  36 static void ovl_dentry_release(struct dentry *dentry)
  37 {
  38         struct ovl_entry *oe = dentry->d_fsdata;
  39
  40         if (oe) {
  41                 unsigned int i;
  42
  43                 dput(oe->__upperdentry);
  44                 kfree(oe->redirect);
  45                 for (i = 0; i < oe->numlower; i++)
  46                         dput(oe->lowerstack[i].dentry);
  47                 kfree_rcu(oe, rcu);
  48         }
  49 }
  50
  51 static struct dentry *ovl_d_real(struct dentry *dentry,
  52                                  const struct inode *inode,
  53                                  unsigned int open_flags)
  54 {
  55         struct dentry *real;
  56
  57         if (!d_is_reg(dentry)) {
  58                 if (!inode || inode == d_inode(dentry))
  59                         return dentry;
  60                 goto bug;
  61         }
  62
  63         if (d_is_negative(dentry))
  64                 return dentry;
  65
  66         if (open_flags) {
  67                 int err = ovl_open_maybe_copy_up(dentry, open_flags);
  68
  69                 if (err)
  70                         return ERR_PTR(err);
  71         }
  72
  73         real = ovl_dentry_upper(dentry);
  74         if (real && (!inode || inode == d_inode(real)))
  75                 return real;
  76
  77         real = ovl_dentry_lower(dentry);
  78         if (!real)
  79                 goto bug;
  80
  81         /* Handle recursion */
  82         real = d_real(real, inode, open_flags);
  83
  84         if (!inode || inode == d_inode(real))
  85                 return real;
  86 bug:
  87         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
  88              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
  89         return dentry;
  90 }
  91
  92 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
  93 {
  94         struct ovl_entry *oe = dentry->d_fsdata;
  95         unsigned int i;
  96         int ret = 1;
  97
  98         for (i = 0; i < oe->numlower; i++) {
  99                 struct dentry *d = oe->lowerstack[i].dentry;
 100
 101                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 102                         ret = d->d_op->d_revalidate(d, flags);
 103                         if (ret < 0)
 104                                 return ret;
 105                         if (!ret) {
 106                                 if (!(flags & LOOKUP_RCU))
 107                                         d_invalidate(d);
 108                                 return -ESTALE;
 109                         }
 110                 }
 111         }
 112         return 1;
 113 }
 114
 115 static int ovl_dentry_weak_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_WEAK_REVALIDATE) {
 125                         ret = d->d_op->d_weak_revalidate(d, flags);
 126                         if (ret <= 0)
 127                                 break;
 128                 }
 129         }
 130         return ret;
 131 }
 132
 133 static const struct dentry_operations ovl_dentry_operations = {
 134         .d_release = ovl_dentry_release,
 135         .d_real = ovl_d_real,
 136 };
 137
 138 static const struct dentry_operations ovl_reval_dentry_operations = {
 139         .d_release = ovl_dentry_release,
 140         .d_real = ovl_d_real,
 141         .d_revalidate = ovl_dentry_revalidate,
 142         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 143 };
 144
 145 static void ovl_put_super(struct super_block *sb)
 146 {
 147         struct ovl_fs *ufs = sb->s_fs_info;
 148         unsigned i;
 149
 150         dput(ufs->workdir);
 151         mntput(ufs->upper_mnt);
 152         for (i = 0; i < ufs->numlower; i++)
 153                 mntput(ufs->lower_mnt[i]);
 154         kfree(ufs->lower_mnt);
 155
 156         kfree(ufs->config.lowerdir);
 157         kfree(ufs->config.upperdir);
 158         kfree(ufs->config.workdir);
 159         put_cred(ufs->creator_cred);
 160         kfree(ufs);
 161 }
 162
 163 static int ovl_sync_fs(struct super_block *sb, int wait)
 164 {
 165         struct ovl_fs *ufs = sb->s_fs_info;
 166         struct super_block *upper_sb;
 167         int ret;
 168
 169         if (!ufs->upper_mnt)
 170                 return 0;
 171         upper_sb = ufs->upper_mnt->mnt_sb;
 172         if (!upper_sb->s_op->sync_fs)
 173                 return 0;
 174
 175         /* real inodes have already been synced by sync_filesystem(ovl_sb) */
 176         down_read(&upper_sb->s_umount);
 177         ret = upper_sb->s_op->sync_fs(upper_sb, wait);
 178         up_read(&upper_sb->s_umount);
 179         return ret;
 180 }
 181
 182 /**
 183  * ovl_statfs
 184  * @sb: The overlayfs super block
 185  * @buf: The struct kstatfs to fill in with stats
 186  *
 187  * Get the filesystem statistics.  As writes always target the upper layer
 188  * filesystem pass the statfs to the upper filesystem (if it exists)
 189  */
 190 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 191 {
 192         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 193         struct dentry *root_dentry = dentry->d_sb->s_root;
 194         struct path path;
 195         int err;
 196
 197         ovl_path_real(root_dentry, &path);
 198
 199         err = vfs_statfs(&path, buf);
 200         if (!err) {
 201                 buf->f_namelen = ofs->namelen;
 202                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 203         }
 204
 205         return err;
 206 }
 207
 208 /**
 209  * ovl_show_options
 210  *
 211  * Prints the mount options for a given superblock.
 212  * Returns zero; does not fail.
 213  */
 214 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 215 {
 216         struct super_block *sb = dentry->d_sb;
 217         struct ovl_fs *ufs = sb->s_fs_info;
 218
 219         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 220         if (ufs->config.upperdir) {
 221                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 222                 seq_show_option(m, "workdir", ufs->config.workdir);
 223         }
 224         if (ufs->config.default_permissions)
 225                 seq_puts(m, ",default_permissions");
 226         if (ufs->config.redirect_dir != ovl_redirect_dir_def)
 227                 seq_printf(m, ",redirect_dir=%s",
 228                            ufs->config.redirect_dir ? "on" : "off");
 229         return 0;
 230 }
 231
 232 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 233 {
 234         struct ovl_fs *ufs = sb->s_fs_info;
 235
 236         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 237                 return -EROFS;
 238
 239         return 0;
 240 }
 241
 242 static const struct super_operations ovl_super_operations = {
 243         .put_super      = ovl_put_super,
 244         .sync_fs        = ovl_sync_fs,
 245         .statfs         = ovl_statfs,
 246         .show_options   = ovl_show_options,
 247         .remount_fs     = ovl_remount,
 248         .drop_inode     = generic_delete_inode,
 249 };
 250
 251 enum {
 252         OPT_LOWERDIR,
 253         OPT_UPPERDIR,
 254         OPT_WORKDIR,
 255         OPT_DEFAULT_PERMISSIONS,
 256         OPT_REDIRECT_DIR_ON,
 257         OPT_REDIRECT_DIR_OFF,
 258         OPT_ERR,
 259 };
 260
 261 static const match_table_t ovl_tokens = {
 262         {OPT_LOWERDIR,                  "lowerdir=%s"},
 263         {OPT_UPPERDIR,                  "upperdir=%s"},
 264         {OPT_WORKDIR,                   "workdir=%s"},
 265         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 266         {OPT_REDIRECT_DIR_ON,           "redirect_dir=on"},
 267         {OPT_REDIRECT_DIR_OFF,          "redirect_dir=off"},
 268         {OPT_ERR,                       NULL}
 269 };
 270
 271 static char *ovl_next_opt(char **s)
 272 {
 273         char *sbegin = *s;
 274         char *p;
 275
 276         if (sbegin == NULL)
 277                 return NULL;
 278
 279         for (p = sbegin; *p; p++) {
 280                 if (*p == '\\') {
 281                         p++;
 282                         if (!*p)
 283                                 break;
 284                 } else if (*p == ',') {
 285                         *p = '\0';
 286                         *s = p + 1;
 287                         return sbegin;
 288                 }
 289         }
 290         *s = NULL;
 291         return sbegin;
 292 }
 293
 294 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 295 {
 296         char *p;
 297
 298         while ((p = ovl_next_opt(&opt)) != NULL) {
 299                 int token;
 300                 substring_t args[MAX_OPT_ARGS];
 301
 302                 if (!*p)
 303                         continue;
 304
 305                 token = match_token(p, ovl_tokens, args);
 306                 switch (token) {
 307                 case OPT_UPPERDIR:
 308                         kfree(config->upperdir);
 309                         config->upperdir = match_strdup(&args[0]);
 310                         if (!config->upperdir)
 311                                 return -ENOMEM;
 312                         break;
 313
 314                 case OPT_LOWERDIR:
 315                         kfree(config->lowerdir);
 316                         config->lowerdir = match_strdup(&args[0]);
 317                         if (!config->lowerdir)
 318                                 return -ENOMEM;
 319                         break;
 320
 321                 case OPT_WORKDIR:
 322                         kfree(config->workdir);
 323                         config->workdir = match_strdup(&args[0]);
 324                         if (!config->workdir)
 325                                 return -ENOMEM;
 326                         break;
 327
 328                 case OPT_DEFAULT_PERMISSIONS:
 329                         config->default_permissions = true;
 330                         break;
 331
 332                 case OPT_REDIRECT_DIR_ON:
 333                         config->redirect_dir = true;
 334                         break;
 335
 336                 case OPT_REDIRECT_DIR_OFF:
 337                         config->redirect_dir = false;
 338                         break;
 339
 340                 default:
 341                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 342                         return -EINVAL;
 343                 }
 344         }
 345
 346         /* Workdir is useless in non-upper mount */
 347         if (!config->upperdir && config->workdir) {
 348                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 349                         config->workdir);
 350                 kfree(config->workdir);
 351                 config->workdir = NULL;
 352         }
 353
 354         return 0;
 355 }
 356
 357 #define OVL_WORKDIR_NAME "work"
 358
 359 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 360                                          struct dentry *dentry)
 361 {
 362         struct inode *dir = dentry->d_inode;
 363         struct dentry *work;
 364         int err;
 365         bool retried = false;
 366
 367         err = mnt_want_write(mnt);
 368         if (err)
 369                 return ERR_PTR(err);
 370
 371         inode_lock_nested(dir, I_MUTEX_PARENT);
 372 retry:
 373         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 374                               strlen(OVL_WORKDIR_NAME));
 375
 376         if (!IS_ERR(work)) {
 377                 struct iattr attr = {
 378                         .ia_valid = ATTR_MODE,
 379                         .ia_mode = S_IFDIR | 0,
 380                 };
 381
 382                 if (work->d_inode) {
 383                         err = -EEXIST;
 384                         if (retried)
 385                                 goto out_dput;
 386
 387                         retried = true;
 388                         ovl_workdir_cleanup(dir, mnt, work, 0);
 389                         dput(work);
 390                         goto retry;
 391                 }
 392
 393                 err = ovl_create_real(dir, work,
 394                                       &(struct cattr){.mode = S_IFDIR | 0},
 395                                       NULL, true);
 396                 if (err)
 397                         goto out_dput;
 398
 399                 /*
 400                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 401                  *
 402                  * a) success (there was a POSIX ACL xattr and was removed)
 403                  * b) -ENODATA (there was no POSIX ACL xattr)
 404                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 405                  *
 406                  * There are various other error values that could effectively
 407                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 408                  * if the xattr name is too long), but the set of filesystems
 409                  * allowed as upper are limited to "normal" ones, where checking
 410                  * for the above two errors is sufficient.
 411                  */
 412                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 413                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 414                         goto out_dput;
 415
 416                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 417                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 418                         goto out_dput;
 419
 420                 /* Clear any inherited mode bits */
 421                 inode_lock(work->d_inode);
 422                 err = notify_change(work, &attr, NULL);
 423                 inode_unlock(work->d_inode);
 424                 if (err)
 425                         goto out_dput;
 426         }
 427 out_unlock:
 428         inode_unlock(dir);
 429         mnt_drop_write(mnt);
 430
 431         return work;
 432
 433 out_dput:
 434         dput(work);
 435         work = ERR_PTR(err);
 436         goto out_unlock;
 437 }
 438
 439 static void ovl_unescape(char *s)
 440 {
 441         char *d = s;
 442
 443         for (;; s++, d++) {
 444                 if (*s == '\\')
 445                         s++;
 446                 *d = *s;
 447                 if (!*s)
 448                         break;
 449         }
 450 }
 451
 452 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 453 {
 454         int err = -EINVAL;
 455
 456         if (!*name) {
 457                 pr_err("overlayfs: empty lowerdir\n");
 458                 goto out;
 459         }
 460         err = kern_path(name, LOOKUP_FOLLOW, path);
 461         if (err) {
 462                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 463                 goto out;
 464         }
 465         err = -EINVAL;
 466         if (ovl_dentry_weird(path->dentry)) {
 467                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 468                 goto out_put;
 469         }
 470         if (!d_is_dir(path->dentry)) {
 471                 pr_err("overlayfs: '%s' not a directory\n", name);
 472                 goto out_put;
 473         }
 474         return 0;
 475
 476 out_put:
 477         path_put(path);
 478 out:
 479         return err;
 480 }
 481
 482 static int ovl_mount_dir(const char *name, struct path *path)
 483 {
 484         int err = -ENOMEM;
 485         char *tmp = kstrdup(name, GFP_KERNEL);
 486
 487         if (tmp) {
 488                 ovl_unescape(tmp);
 489                 err = ovl_mount_dir_noesc(tmp, path);
 490
 491                 if (!err)
 492                         if (ovl_dentry_remote(path->dentry)) {
 493                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 494                                        tmp);
 495                                 path_put(path);
 496                                 err = -EINVAL;
 497                         }
 498                 kfree(tmp);
 499         }
 500         return err;
 501 }
 502
 503 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 504                              const char *name)
 505 {
 506         struct kstatfs statfs;
 507         int err = vfs_statfs(path, &statfs);
 508
 509         if (err)
 510                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 511         else
 512                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 513
 514         return err;
 515 }
 516
 517 static int ovl_lower_dir(const char *name, struct path *path,
 518                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 519 {
 520         int err;
 521
 522         err = ovl_mount_dir_noesc(name, path);
 523         if (err)
 524                 goto out;
 525
 526         err = ovl_check_namelen(path, ofs, name);
 527         if (err)
 528                 goto out_put;
 529
 530         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 531
 532         if (ovl_dentry_remote(path->dentry))
 533                 *remote = true;
 534
 535         return 0;
 536
 537 out_put:
 538         path_put(path);
 539 out:
 540         return err;
 541 }
 542
 543 /* Workdir should not be subdir of upperdir and vice versa */
 544 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 545 {
 546         bool ok = false;
 547
 548         if (workdir != upperdir) {
 549                 ok = (lock_rename(workdir, upperdir) == NULL);
 550                 unlock_rename(workdir, upperdir);
 551         }
 552         return ok;
 553 }
 554
 555 static unsigned int ovl_split_lowerdirs(char *str)
 556 {
 557         unsigned int ctr = 1;
 558         char *s, *d;
 559
 560         for (s = d = str;; s++, d++) {
 561                 if (*s == '\\') {
 562                         s++;
 563                 } else if (*s == ':') {
 564                         *d = '\0';
 565                         ctr++;
 566                         continue;
 567                 }
 568                 *d = *s;
 569                 if (!*s)
 570                         break;
 571         }
 572         return ctr;
 573 }
 574
 575 static int __maybe_unused
 576 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 577                         struct dentry *dentry, struct inode *inode,
 578                         const char *name, void *buffer, size_t size)
 579 {
 580         return ovl_xattr_get(dentry, handler->name, buffer, size);
 581 }
 582
 583 static int __maybe_unused
 584 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 585                         struct dentry *dentry, struct inode *inode,
 586                         const char *name, const void *value,
 587                         size_t size, int flags)
 588 {
 589         struct dentry *workdir = ovl_workdir(dentry);
 590         struct inode *realinode = ovl_inode_real(inode, NULL);
 591         struct posix_acl *acl = NULL;
 592         int err;
 593
 594         /* Check that everything is OK before copy-up */
 595         if (value) {
 596                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 597                 if (IS_ERR(acl))
 598                         return PTR_ERR(acl);
 599         }
 600         err = -EOPNOTSUPP;
 601         if (!IS_POSIXACL(d_inode(workdir)))
 602                 goto out_acl_release;
 603         if (!realinode->i_op->set_acl)
 604                 goto out_acl_release;
 605         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 606                 err = acl ? -EACCES : 0;
 607                 goto out_acl_release;
 608         }
 609         err = -EPERM;
 610         if (!inode_owner_or_capable(inode))
 611                 goto out_acl_release;
 612
 613         posix_acl_release(acl);
 614
 615         /*
 616          * Check if sgid bit needs to be cleared (actual setacl operation will
 617          * be done with mounter's capabilities and so that won't do it for us).
 618          */
 619         if (unlikely(inode->i_mode & S_ISGID) &&
 620             handler->flags == ACL_TYPE_ACCESS &&
 621             !in_group_p(inode->i_gid) &&
 622             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 623                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 624
 625                 err = ovl_setattr(dentry, &iattr);
 626                 if (err)
 627                         return err;
 628         }
 629
 630         err = ovl_xattr_set(dentry, handler->name, value, size, flags);
 631         if (!err)
 632                 ovl_copyattr(ovl_inode_real(inode, NULL), inode);
 633
 634         return err;
 635
 636 out_acl_release:
 637         posix_acl_release(acl);
 638         return err;
 639 }
 640
 641 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 642                              struct dentry *dentry, struct inode *inode,
 643                              const char *name, void *buffer, size_t size)
 644 {
 645         return -EOPNOTSUPP;
 646 }
 647
 648 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 649                              struct dentry *dentry, struct inode *inode,
 650                              const char *name, const void *value,
 651                              size_t size, int flags)
 652 {
 653         return -EOPNOTSUPP;
 654 }
 655
 656 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 657                                struct dentry *dentry, struct inode *inode,
 658                                const char *name, void *buffer, size_t size)
 659 {
 660         return ovl_xattr_get(dentry, name, buffer, size);
 661 }
 662
 663 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 664                                struct dentry *dentry, struct inode *inode,
 665                                const char *name, const void *value,
 666                                size_t size, int flags)
 667 {
 668         return ovl_xattr_set(dentry, name, value, size, flags);
 669 }
 670
 671 static const struct xattr_handler __maybe_unused
 672 ovl_posix_acl_access_xattr_handler = {
 673         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 674         .flags = ACL_TYPE_ACCESS,
 675         .get = ovl_posix_acl_xattr_get,
 676         .set = ovl_posix_acl_xattr_set,
 677 };
 678
 679 static const struct xattr_handler __maybe_unused
 680 ovl_posix_acl_default_xattr_handler = {
 681         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 682         .flags = ACL_TYPE_DEFAULT,
 683         .get = ovl_posix_acl_xattr_get,
 684         .set = ovl_posix_acl_xattr_set,
 685 };
 686
 687 static const struct xattr_handler ovl_own_xattr_handler = {
 688         .prefix = OVL_XATTR_PREFIX,
 689         .get = ovl_own_xattr_get,
 690         .set = ovl_own_xattr_set,
 691 };
 692
 693 static const struct xattr_handler ovl_other_xattr_handler = {
 694         .prefix = "", /* catch all */
 695         .get = ovl_other_xattr_get,
 696         .set = ovl_other_xattr_set,
 697 };
 698
 699 static const struct xattr_handler *ovl_xattr_handlers[] = {
 700 #ifdef CONFIG_FS_POSIX_ACL
 701         &ovl_posix_acl_access_xattr_handler,
 702         &ovl_posix_acl_default_xattr_handler,
 703 #endif
 704         &ovl_own_xattr_handler,
 705         &ovl_other_xattr_handler,
 706         NULL
 707 };
 708
 709 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 710 {
 711         struct path upperpath = { NULL, NULL };
 712         struct path workpath = { NULL, NULL };
 713         struct dentry *root_dentry;
 714         struct inode *realinode;
 715         struct ovl_entry *oe;
 716         struct ovl_fs *ufs;
 717         struct path *stack = NULL;
 718         char *lowertmp;
 719         char *lower;
 720         unsigned int numlower;
 721         unsigned int stacklen = 0;
 722         unsigned int i;
 723         bool remote = false;
 724         int err;
 725
 726         err = -ENOMEM;
 727         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 728         if (!ufs)
 729                 goto out;
 730
 731         init_waitqueue_head(&ufs->copyup_wq);
 732         ufs->config.redirect_dir = ovl_redirect_dir_def;
 733         err = ovl_parse_opt((char *) data, &ufs->config);
 734         if (err)
 735                 goto out_free_config;
 736
 737         err = -EINVAL;
 738         if (!ufs->config.lowerdir) {
 739                 if (!silent)
 740                         pr_err("overlayfs: missing 'lowerdir'\n");
 741                 goto out_free_config;
 742         }
 743
 744         sb->s_stack_depth = 0;
 745         sb->s_maxbytes = MAX_LFS_FILESIZE;
 746         if (ufs->config.upperdir) {
 747                 if (!ufs->config.workdir) {
 748                         pr_err("overlayfs: missing 'workdir'\n");
 749                         goto out_free_config;
 750                 }
 751
 752                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 753                 if (err)
 754                         goto out_free_config;
 755
 756                 /* Upper fs should not be r/o */
 757                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 758                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 759                         err = -EINVAL;
 760                         goto out_put_upperpath;
 761                 }
 762
 763                 err = ovl_check_namelen(&upperpath, ufs, ufs->config.upperdir);
 764                 if (err)
 765                         goto out_put_upperpath;
 766
 767                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 768                 if (err)
 769                         goto out_put_upperpath;
 770
 771                 err = -EINVAL;
 772                 if (upperpath.mnt != workpath.mnt) {
 773                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 774                         goto out_put_workpath;
 775                 }
 776                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 777                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 778                         goto out_put_workpath;
 779                 }
 780                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 781         }
 782         err = -ENOMEM;
 783         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 784         if (!lowertmp)
 785                 goto out_put_workpath;
 786
 787         err = -EINVAL;
 788         stacklen = ovl_split_lowerdirs(lowertmp);
 789         if (stacklen > OVL_MAX_STACK) {
 790                 pr_err("overlayfs: too many lower directories, limit is %d\n",
 791                        OVL_MAX_STACK);
 792                 goto out_free_lowertmp;
 793         } else if (!ufs->config.upperdir && stacklen == 1) {
 794                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 795                 goto out_free_lowertmp;
 796         }
 797
 798         err = -ENOMEM;
 799         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 800         if (!stack)
 801                 goto out_free_lowertmp;
 802
 803         err = -EINVAL;
 804         lower = lowertmp;
 805         for (numlower = 0; numlower < stacklen; numlower++) {
 806                 err = ovl_lower_dir(lower, &stack[numlower], ufs,
 807                                     &sb->s_stack_depth, &remote);
 808                 if (err)
 809                         goto out_put_lowerpath;
 810
 811                 lower = strchr(lower, '\0') + 1;
 812         }
 813
 814         err = -EINVAL;
 815         sb->s_stack_depth++;
 816         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
 817                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
 818                 goto out_put_lowerpath;
 819         }
 820
 821         if (ufs->config.upperdir) {
 822                 ufs->upper_mnt = clone_private_mount(&upperpath);
 823                 err = PTR_ERR(ufs->upper_mnt);
 824                 if (IS_ERR(ufs->upper_mnt)) {
 825                         pr_err("overlayfs: failed to clone upperpath\n");
 826                         goto out_put_lowerpath;
 827                 }
 828                 /* Don't inherit atime flags */
 829                 ufs->upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
 830
 831                 sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
 832
 833                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
 834                 err = PTR_ERR(ufs->workdir);
 835                 if (IS_ERR(ufs->workdir)) {
 836                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 837                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
 838                         sb->s_flags |= MS_RDONLY;
 839                         ufs->workdir = NULL;
 840                 }
 841
 842                 /*
 843                  * Upper should support d_type, else whiteouts are visible.
 844                  * Given workdir and upper are on same fs, we can do
 845                  * iterate_dir() on workdir. This check requires successful
 846                  * creation of workdir in previous step.
 847                  */
 848                 if (ufs->workdir) {
 849                         struct dentry *temp;
 850
 851                         err = ovl_check_d_type_supported(&workpath);
 852                         if (err < 0)
 853                                 goto out_put_workdir;
 854
 855                         /*
 856                          * We allowed this configuration and don't want to
 857                          * break users over kernel upgrade. So warn instead
 858                          * of erroring out.
 859                          */
 860                         if (!err)
 861                                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
 862
 863                         /* Check if upper/work fs supports O_TMPFILE */
 864                         temp = ovl_do_tmpfile(ufs->workdir, S_IFREG | 0);
 865                         ufs->tmpfile = !IS_ERR(temp);
 866                         if (ufs->tmpfile)
 867                                 dput(temp);
 868                         else
 869                                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
 870                 }
 871         }
 872
 873         err = -ENOMEM;
 874         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
 875         if (ufs->lower_mnt == NULL)
 876                 goto out_put_workdir;
 877         for (i = 0; i < numlower; i++) {
 878                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
 879
 880                 err = PTR_ERR(mnt);
 881                 if (IS_ERR(mnt)) {
 882                         pr_err("overlayfs: failed to clone lowerpath\n");
 883                         goto out_put_lower_mnt;
 884                 }
 885                 /*
 886                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
 887                  * will fail instead of modifying lower fs.
 888                  */
 889                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
 890
 891                 ufs->lower_mnt[ufs->numlower] = mnt;
 892                 ufs->numlower++;
 893         }
 894
 895         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
 896         if (!ufs->upper_mnt)
 897                 sb->s_flags |= MS_RDONLY;
 898
 899         if (remote)
 900                 sb->s_d_op = &ovl_reval_dentry_operations;
 901         else
 902                 sb->s_d_op = &ovl_dentry_operations;
 903
 904         ufs->creator_cred = prepare_creds();
 905         if (!ufs->creator_cred)
 906                 goto out_put_lower_mnt;
 907
 908         err = -ENOMEM;
 909         oe = ovl_alloc_entry(numlower);
 910         if (!oe)
 911                 goto out_put_cred;
 912
 913         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
 914         sb->s_op = &ovl_super_operations;
 915         sb->s_xattr = ovl_xattr_handlers;
 916         sb->s_fs_info = ufs;
 917         sb->s_flags |= MS_POSIXACL | MS_NOREMOTELOCK;
 918
 919         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
 920         if (!root_dentry)
 921                 goto out_free_oe;
 922
 923         mntput(upperpath.mnt);
 924         for (i = 0; i < numlower; i++)
 925                 mntput(stack[i].mnt);
 926         path_put(&workpath);
 927         kfree(lowertmp);
 928
 929         oe->__upperdentry = upperpath.dentry;
 930         for (i = 0; i < numlower; i++) {
 931                 oe->lowerstack[i].dentry = stack[i].dentry;
 932                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
 933         }
 934         kfree(stack);
 935
 936         root_dentry->d_fsdata = oe;
 937
 938         realinode = d_inode(ovl_dentry_real(root_dentry));
 939         ovl_inode_init(d_inode(root_dentry), realinode, !!upperpath.dentry);
 940         ovl_copyattr(realinode, d_inode(root_dentry));
 941
 942         sb->s_root = root_dentry;
 943
 944         return 0;
 945
 946 out_free_oe:
 947         kfree(oe);
 948 out_put_cred:
 949         put_cred(ufs->creator_cred);
 950 out_put_lower_mnt:
 951         for (i = 0; i < ufs->numlower; i++)
 952                 mntput(ufs->lower_mnt[i]);
 953         kfree(ufs->lower_mnt);
 954 out_put_workdir:
 955         dput(ufs->workdir);
 956         mntput(ufs->upper_mnt);
 957 out_put_lowerpath:
 958         for (i = 0; i < numlower; i++)
 959                 path_put(&stack[i]);
 960         kfree(stack);
 961 out_free_lowertmp:
 962         kfree(lowertmp);
 963 out_put_workpath:
 964         path_put(&workpath);
 965 out_put_upperpath:
 966         path_put(&upperpath);
 967 out_free_config:
 968         kfree(ufs->config.lowerdir);
 969         kfree(ufs->config.upperdir);
 970         kfree(ufs->config.workdir);
 971         kfree(ufs);
 972 out:
 973         return err;
 974 }
 975
 976 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
 977                                 const char *dev_name, void *raw_data)
 978 {
 979         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
 980 }
 981
 982 static struct file_system_type ovl_fs_type = {
 983         .owner          = THIS_MODULE,
 984         .name           = "overlay",
 985         .mount          = ovl_mount,
 986         .kill_sb        = kill_anon_super,
 987 };
 988 MODULE_ALIAS_FS("overlay");
 989
 990 static int __init ovl_init(void)
 991 {
 992         return register_filesystem(&ovl_fs_type);
 993 }
 994
 995 static void __exit ovl_exit(void)
 996 {
 997         unregister_filesystem(&ovl_fs_type);
 998 }
 999
1000 module_init(ovl_init);
1001 module_exit(ovl_exit);