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 <linux/exportfs.h>
  21 #include "overlayfs.h"
  22
  23 MODULE_AUTHOR("Miklos Szeredi <[email protected]>");
  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 bool ovl_redirect_always_follow =
  38         IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
  39 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
  40                    bool, 0644);
  41 MODULE_PARM_DESC(ovl_redirect_always_follow,
  42                  "Follow redirects even if redirect_dir feature is turned off");
  43
  44 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
  45 module_param_named(index, ovl_index_def, bool, 0644);
  46 MODULE_PARM_DESC(ovl_index_def,
  47                  "Default to on or off for the inodes index feature");
  48
  49 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
  50 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
  51 MODULE_PARM_DESC(ovl_nfs_export_def,
  52                  "Default to on or off for the NFS export feature");
  53
  54 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
  55 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
  56 MODULE_PARM_DESC(ovl_xino_auto_def,
  57                  "Auto enable xino feature");
  58
  59 static void ovl_entry_stack_free(struct ovl_entry *oe)
  60 {
  61         unsigned int i;
  62
  63         for (i = 0; i < oe->numlower; i++)
  64                 dput(oe->lowerstack[i].dentry);
  65 }
  66
  67 static void ovl_dentry_release(struct dentry *dentry)
  68 {
  69         struct ovl_entry *oe = dentry->d_fsdata;
  70
  71         if (oe) {
  72                 ovl_entry_stack_free(oe);
  73                 kfree_rcu(oe, rcu);
  74         }
  75 }
  76
  77 static int ovl_check_append_only(struct inode *inode, int flag)
  78 {
  79         /*
  80          * This test was moot in vfs may_open() because overlay inode does
  81          * not have the S_APPEND flag, so re-check on real upper inode
  82          */
  83         if (IS_APPEND(inode)) {
  84                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
  85                         return -EPERM;
  86                 if (flag & O_TRUNC)
  87                         return -EPERM;
  88         }
  89
  90         return 0;
  91 }
  92
  93 static struct dentry *ovl_d_real(struct dentry *dentry,
  94                                  const struct inode *inode,
  95                                  unsigned int open_flags, unsigned int flags)
  96 {
  97         struct dentry *real;
  98         int err;
  99
 100         if (flags & D_REAL_UPPER)
 101                 return ovl_dentry_upper(dentry);
 102
 103         if (!d_is_reg(dentry)) {
 104                 if (!inode || inode == d_inode(dentry))
 105                         return dentry;
 106                 goto bug;
 107         }
 108
 109         if (open_flags) {
 110                 err = ovl_open_maybe_copy_up(dentry, open_flags);
 111                 if (err)
 112                         return ERR_PTR(err);
 113         }
 114
 115         real = ovl_dentry_upper(dentry);
 116         if (real && (!inode || inode == d_inode(real))) {
 117                 if (!inode) {
 118                         err = ovl_check_append_only(d_inode(real), open_flags);
 119                         if (err)
 120                                 return ERR_PTR(err);
 121                 }
 122                 return real;
 123         }
 124
 125         real = ovl_dentry_lower(dentry);
 126         if (!real)
 127                 goto bug;
 128
 129         /* Handle recursion */
 130         real = d_real(real, inode, open_flags, 0);
 131
 132         if (!inode || inode == d_inode(real))
 133                 return real;
 134 bug:
 135         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
 136              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 137         return dentry;
 138 }
 139
 140 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 141 {
 142         struct ovl_entry *oe = dentry->d_fsdata;
 143         unsigned int i;
 144         int ret = 1;
 145
 146         for (i = 0; i < oe->numlower; i++) {
 147                 struct dentry *d = oe->lowerstack[i].dentry;
 148
 149                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 150                         ret = d->d_op->d_revalidate(d, flags);
 151                         if (ret < 0)
 152                                 return ret;
 153                         if (!ret) {
 154                                 if (!(flags & LOOKUP_RCU))
 155                                         d_invalidate(d);
 156                                 return -ESTALE;
 157                         }
 158                 }
 159         }
 160         return 1;
 161 }
 162
 163 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 164 {
 165         struct ovl_entry *oe = dentry->d_fsdata;
 166         unsigned int i;
 167         int ret = 1;
 168
 169         for (i = 0; i < oe->numlower; i++) {
 170                 struct dentry *d = oe->lowerstack[i].dentry;
 171
 172                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 173                         ret = d->d_op->d_weak_revalidate(d, flags);
 174                         if (ret <= 0)
 175                                 break;
 176                 }
 177         }
 178         return ret;
 179 }
 180
 181 static const struct dentry_operations ovl_dentry_operations = {
 182         .d_release = ovl_dentry_release,
 183         .d_real = ovl_d_real,
 184 };
 185
 186 static const struct dentry_operations ovl_reval_dentry_operations = {
 187         .d_release = ovl_dentry_release,
 188         .d_real = ovl_d_real,
 189         .d_revalidate = ovl_dentry_revalidate,
 190         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 191 };
 192
 193 static struct kmem_cache *ovl_inode_cachep;
 194
 195 static struct inode *ovl_alloc_inode(struct super_block *sb)
 196 {
 197         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 198
 199         if (!oi)
 200                 return NULL;
 201
 202         oi->cache = NULL;
 203         oi->redirect = NULL;
 204         oi->version = 0;
 205         oi->flags = 0;
 206         oi->__upperdentry = NULL;
 207         oi->lower = NULL;
 208         mutex_init(&oi->lock);
 209
 210         return &oi->vfs_inode;
 211 }
 212
 213 static void ovl_i_callback(struct rcu_head *head)
 214 {
 215         struct inode *inode = container_of(head, struct inode, i_rcu);
 216
 217         kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
 218 }
 219
 220 static void ovl_destroy_inode(struct inode *inode)
 221 {
 222         struct ovl_inode *oi = OVL_I(inode);
 223
 224         dput(oi->__upperdentry);
 225         iput(oi->lower);
 226         kfree(oi->redirect);
 227         ovl_dir_cache_free(inode);
 228         mutex_destroy(&oi->lock);
 229
 230         call_rcu(&inode->i_rcu, ovl_i_callback);
 231 }
 232
 233 static void ovl_free_fs(struct ovl_fs *ofs)
 234 {
 235         unsigned i;
 236
 237         dput(ofs->indexdir);
 238         dput(ofs->workdir);
 239         if (ofs->workdir_locked)
 240                 ovl_inuse_unlock(ofs->workbasedir);
 241         dput(ofs->workbasedir);
 242         if (ofs->upperdir_locked)
 243                 ovl_inuse_unlock(ofs->upper_mnt->mnt_root);
 244         mntput(ofs->upper_mnt);
 245         for (i = 0; i < ofs->numlower; i++)
 246                 mntput(ofs->lower_layers[i].mnt);
 247         for (i = 0; i < ofs->numlowerfs; i++)
 248                 free_anon_bdev(ofs->lower_fs[i].pseudo_dev);
 249         kfree(ofs->lower_layers);
 250         kfree(ofs->lower_fs);
 251
 252         kfree(ofs->config.lowerdir);
 253         kfree(ofs->config.upperdir);
 254         kfree(ofs->config.workdir);
 255         kfree(ofs->config.redirect_mode);
 256         if (ofs->creator_cred)
 257                 put_cred(ofs->creator_cred);
 258         kfree(ofs);
 259 }
 260
 261 static void ovl_put_super(struct super_block *sb)
 262 {
 263         struct ovl_fs *ofs = sb->s_fs_info;
 264
 265         ovl_free_fs(ofs);
 266 }
 267
 268 /* Sync real dirty inodes in upper filesystem (if it exists) */
 269 static int ovl_sync_fs(struct super_block *sb, int wait)
 270 {
 271         struct ovl_fs *ofs = sb->s_fs_info;
 272         struct super_block *upper_sb;
 273         int ret;
 274
 275         if (!ofs->upper_mnt)
 276                 return 0;
 277
 278         /*
 279          * If this is a sync(2) call or an emergency sync, all the super blocks
 280          * will be iterated, including upper_sb, so no need to do anything.
 281          *
 282          * If this is a syncfs(2) call, then we do need to call
 283          * sync_filesystem() on upper_sb, but enough if we do it when being
 284          * called with wait == 1.
 285          */
 286         if (!wait)
 287                 return 0;
 288
 289         upper_sb = ofs->upper_mnt->mnt_sb;
 290
 291         down_read(&upper_sb->s_umount);
 292         ret = sync_filesystem(upper_sb);
 293         up_read(&upper_sb->s_umount);
 294
 295         return ret;
 296 }
 297
 298 /**
 299  * ovl_statfs
 300  * @sb: The overlayfs super block
 301  * @buf: The struct kstatfs to fill in with stats
 302  *
 303  * Get the filesystem statistics.  As writes always target the upper layer
 304  * filesystem pass the statfs to the upper filesystem (if it exists)
 305  */
 306 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 307 {
 308         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 309         struct dentry *root_dentry = dentry->d_sb->s_root;
 310         struct path path;
 311         int err;
 312
 313         ovl_path_real(root_dentry, &path);
 314
 315         err = vfs_statfs(&path, buf);
 316         if (!err) {
 317                 buf->f_namelen = ofs->namelen;
 318                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 319         }
 320
 321         return err;
 322 }
 323
 324 /* Will this overlay be forced to mount/remount ro? */
 325 static bool ovl_force_readonly(struct ovl_fs *ofs)
 326 {
 327         return (!ofs->upper_mnt || !ofs->workdir);
 328 }
 329
 330 static const char *ovl_redirect_mode_def(void)
 331 {
 332         return ovl_redirect_dir_def ? "on" : "off";
 333 }
 334
 335 enum {
 336         OVL_XINO_OFF,
 337         OVL_XINO_AUTO,
 338         OVL_XINO_ON,
 339 };
 340
 341 static const char * const ovl_xino_str[] = {
 342         "off",
 343         "auto",
 344         "on",
 345 };
 346
 347 static inline int ovl_xino_def(void)
 348 {
 349         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 350 }
 351
 352 /**
 353  * ovl_show_options
 354  *
 355  * Prints the mount options for a given superblock.
 356  * Returns zero; does not fail.
 357  */
 358 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 359 {
 360         struct super_block *sb = dentry->d_sb;
 361         struct ovl_fs *ofs = sb->s_fs_info;
 362
 363         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 364         if (ofs->config.upperdir) {
 365                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 366                 seq_show_option(m, "workdir", ofs->config.workdir);
 367         }
 368         if (ofs->config.default_permissions)
 369                 seq_puts(m, ",default_permissions");
 370         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 371                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 372         if (ofs->config.index != ovl_index_def)
 373                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 374         if (ofs->config.nfs_export != ovl_nfs_export_def)
 375                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 376                                                 "on" : "off");
 377         if (ofs->config.xino != ovl_xino_def())
 378                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 379         return 0;
 380 }
 381
 382 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 383 {
 384         struct ovl_fs *ofs = sb->s_fs_info;
 385
 386         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 387                 return -EROFS;
 388
 389         return 0;
 390 }
 391
 392 static const struct super_operations ovl_super_operations = {
 393         .alloc_inode    = ovl_alloc_inode,
 394         .destroy_inode  = ovl_destroy_inode,
 395         .drop_inode     = generic_delete_inode,
 396         .put_super      = ovl_put_super,
 397         .sync_fs        = ovl_sync_fs,
 398         .statfs         = ovl_statfs,
 399         .show_options   = ovl_show_options,
 400         .remount_fs     = ovl_remount,
 401 };
 402
 403 enum {
 404         OPT_LOWERDIR,
 405         OPT_UPPERDIR,
 406         OPT_WORKDIR,
 407         OPT_DEFAULT_PERMISSIONS,
 408         OPT_REDIRECT_DIR,
 409         OPT_INDEX_ON,
 410         OPT_INDEX_OFF,
 411         OPT_NFS_EXPORT_ON,
 412         OPT_NFS_EXPORT_OFF,
 413         OPT_XINO_ON,
 414         OPT_XINO_OFF,
 415         OPT_XINO_AUTO,
 416         OPT_ERR,
 417 };
 418
 419 static const match_table_t ovl_tokens = {
 420         {OPT_LOWERDIR,                  "lowerdir=%s"},
 421         {OPT_UPPERDIR,                  "upperdir=%s"},
 422         {OPT_WORKDIR,                   "workdir=%s"},
 423         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 424         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 425         {OPT_INDEX_ON,                  "index=on"},
 426         {OPT_INDEX_OFF,                 "index=off"},
 427         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 428         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 429         {OPT_XINO_ON,                   "xino=on"},
 430         {OPT_XINO_OFF,                  "xino=off"},
 431         {OPT_XINO_AUTO,                 "xino=auto"},
 432         {OPT_ERR,                       NULL}
 433 };
 434
 435 static char *ovl_next_opt(char **s)
 436 {
 437         char *sbegin = *s;
 438         char *p;
 439
 440         if (sbegin == NULL)
 441                 return NULL;
 442
 443         for (p = sbegin; *p; p++) {
 444                 if (*p == '\\') {
 445                         p++;
 446                         if (!*p)
 447                                 break;
 448                 } else if (*p == ',') {
 449                         *p = '\0';
 450                         *s = p + 1;
 451                         return sbegin;
 452                 }
 453         }
 454         *s = NULL;
 455         return sbegin;
 456 }
 457
 458 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 459 {
 460         if (strcmp(mode, "on") == 0) {
 461                 config->redirect_dir = true;
 462                 /*
 463                  * Does not make sense to have redirect creation without
 464                  * redirect following.
 465                  */
 466                 config->redirect_follow = true;
 467         } else if (strcmp(mode, "follow") == 0) {
 468                 config->redirect_follow = true;
 469         } else if (strcmp(mode, "off") == 0) {
 470                 if (ovl_redirect_always_follow)
 471                         config->redirect_follow = true;
 472         } else if (strcmp(mode, "nofollow") != 0) {
 473                 pr_err("overlayfs: bad mount option \"redirect_dir=%s\"\n",
 474                        mode);
 475                 return -EINVAL;
 476         }
 477
 478         return 0;
 479 }
 480
 481 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 482 {
 483         char *p;
 484
 485         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 486         if (!config->redirect_mode)
 487                 return -ENOMEM;
 488
 489         while ((p = ovl_next_opt(&opt)) != NULL) {
 490                 int token;
 491                 substring_t args[MAX_OPT_ARGS];
 492
 493                 if (!*p)
 494                         continue;
 495
 496                 token = match_token(p, ovl_tokens, args);
 497                 switch (token) {
 498                 case OPT_UPPERDIR:
 499                         kfree(config->upperdir);
 500                         config->upperdir = match_strdup(&args[0]);
 501                         if (!config->upperdir)
 502                                 return -ENOMEM;
 503                         break;
 504
 505                 case OPT_LOWERDIR:
 506                         kfree(config->lowerdir);
 507                         config->lowerdir = match_strdup(&args[0]);
 508                         if (!config->lowerdir)
 509                                 return -ENOMEM;
 510                         break;
 511
 512                 case OPT_WORKDIR:
 513                         kfree(config->workdir);
 514                         config->workdir = match_strdup(&args[0]);
 515                         if (!config->workdir)
 516                                 return -ENOMEM;
 517                         break;
 518
 519                 case OPT_DEFAULT_PERMISSIONS:
 520                         config->default_permissions = true;
 521                         break;
 522
 523                 case OPT_REDIRECT_DIR:
 524                         kfree(config->redirect_mode);
 525                         config->redirect_mode = match_strdup(&args[0]);
 526                         if (!config->redirect_mode)
 527                                 return -ENOMEM;
 528                         break;
 529
 530                 case OPT_INDEX_ON:
 531                         config->index = true;
 532                         break;
 533
 534                 case OPT_INDEX_OFF:
 535                         config->index = false;
 536                         break;
 537
 538                 case OPT_NFS_EXPORT_ON:
 539                         config->nfs_export = true;
 540                         break;
 541
 542                 case OPT_NFS_EXPORT_OFF:
 543                         config->nfs_export = false;
 544                         break;
 545
 546                 case OPT_XINO_ON:
 547                         config->xino = OVL_XINO_ON;
 548                         break;
 549
 550                 case OPT_XINO_OFF:
 551                         config->xino = OVL_XINO_OFF;
 552                         break;
 553
 554                 case OPT_XINO_AUTO:
 555                         config->xino = OVL_XINO_AUTO;
 556                         break;
 557
 558                 default:
 559                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 560                         return -EINVAL;
 561                 }
 562         }
 563
 564         /* Workdir is useless in non-upper mount */
 565         if (!config->upperdir && config->workdir) {
 566                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 567                         config->workdir);
 568                 kfree(config->workdir);
 569                 config->workdir = NULL;
 570         }
 571
 572         return ovl_parse_redirect_mode(config, config->redirect_mode);
 573 }
 574
 575 #define OVL_WORKDIR_NAME "work"
 576 #define OVL_INDEXDIR_NAME "index"
 577
 578 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 579                                          const char *name, bool persist)
 580 {
 581         struct inode *dir =  ofs->workbasedir->d_inode;
 582         struct vfsmount *mnt = ofs->upper_mnt;
 583         struct dentry *work;
 584         int err;
 585         bool retried = false;
 586         bool locked = false;
 587
 588         inode_lock_nested(dir, I_MUTEX_PARENT);
 589         locked = true;
 590
 591 retry:
 592         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
 593
 594         if (!IS_ERR(work)) {
 595                 struct iattr attr = {
 596                         .ia_valid = ATTR_MODE,
 597                         .ia_mode = S_IFDIR | 0,
 598                 };
 599
 600                 if (work->d_inode) {
 601                         err = -EEXIST;
 602                         if (retried)
 603                                 goto out_dput;
 604
 605                         if (persist)
 606                                 goto out_unlock;
 607
 608                         retried = true;
 609                         ovl_workdir_cleanup(dir, mnt, work, 0);
 610                         dput(work);
 611                         goto retry;
 612                 }
 613
 614                 err = ovl_create_real(dir, work,
 615                                       &(struct cattr){.mode = S_IFDIR | 0},
 616                                       NULL, true);
 617                 if (err)
 618                         goto out_dput;
 619
 620                 /*
 621                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 622                  *
 623                  * a) success (there was a POSIX ACL xattr and was removed)
 624                  * b) -ENODATA (there was no POSIX ACL xattr)
 625                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 626                  *
 627                  * There are various other error values that could effectively
 628                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 629                  * if the xattr name is too long), but the set of filesystems
 630                  * allowed as upper are limited to "normal" ones, where checking
 631                  * for the above two errors is sufficient.
 632                  */
 633                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 634                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 635                         goto out_dput;
 636
 637                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 638                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 639                         goto out_dput;
 640
 641                 /* Clear any inherited mode bits */
 642                 inode_lock(work->d_inode);
 643                 err = notify_change(work, &attr, NULL);
 644                 inode_unlock(work->d_inode);
 645                 if (err)
 646                         goto out_dput;
 647         } else {
 648                 err = PTR_ERR(work);
 649                 goto out_err;
 650         }
 651 out_unlock:
 652         if (locked)
 653                 inode_unlock(dir);
 654
 655         return work;
 656
 657 out_dput:
 658         dput(work);
 659 out_err:
 660         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 661                 ofs->config.workdir, name, -err);
 662         work = NULL;
 663         goto out_unlock;
 664 }
 665
 666 static void ovl_unescape(char *s)
 667 {
 668         char *d = s;
 669
 670         for (;; s++, d++) {
 671                 if (*s == '\\')
 672                         s++;
 673                 *d = *s;
 674                 if (!*s)
 675                         break;
 676         }
 677 }
 678
 679 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 680 {
 681         int err = -EINVAL;
 682
 683         if (!*name) {
 684                 pr_err("overlayfs: empty lowerdir\n");
 685                 goto out;
 686         }
 687         err = kern_path(name, LOOKUP_FOLLOW, path);
 688         if (err) {
 689                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 690                 goto out;
 691         }
 692         err = -EINVAL;
 693         if (ovl_dentry_weird(path->dentry)) {
 694                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 695                 goto out_put;
 696         }
 697         if (!d_is_dir(path->dentry)) {
 698                 pr_err("overlayfs: '%s' not a directory\n", name);
 699                 goto out_put;
 700         }
 701         return 0;
 702
 703 out_put:
 704         path_put_init(path);
 705 out:
 706         return err;
 707 }
 708
 709 static int ovl_mount_dir(const char *name, struct path *path)
 710 {
 711         int err = -ENOMEM;
 712         char *tmp = kstrdup(name, GFP_KERNEL);
 713
 714         if (tmp) {
 715                 ovl_unescape(tmp);
 716                 err = ovl_mount_dir_noesc(tmp, path);
 717
 718                 if (!err)
 719                         if (ovl_dentry_remote(path->dentry)) {
 720                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 721                                        tmp);
 722                                 path_put_init(path);
 723                                 err = -EINVAL;
 724                         }
 725                 kfree(tmp);
 726         }
 727         return err;
 728 }
 729
 730 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 731                              const char *name)
 732 {
 733         struct kstatfs statfs;
 734         int err = vfs_statfs(path, &statfs);
 735
 736         if (err)
 737                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 738         else
 739                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 740
 741         return err;
 742 }
 743
 744 static int ovl_lower_dir(const char *name, struct path *path,
 745                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 746 {
 747         int fh_type;
 748         int err;
 749
 750         err = ovl_mount_dir_noesc(name, path);
 751         if (err)
 752                 goto out;
 753
 754         err = ovl_check_namelen(path, ofs, name);
 755         if (err)
 756                 goto out_put;
 757
 758         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 759
 760         if (ovl_dentry_remote(path->dentry))
 761                 *remote = true;
 762
 763         /*
 764          * The inodes index feature and NFS export need to encode and decode
 765          * file handles, so they require that all layers support them.
 766          */
 767         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 768         if ((ofs->config.nfs_export ||
 769              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 770                 ofs->config.index = false;
 771                 ofs->config.nfs_export = false;
 772                 pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 773                         name);
 774         }
 775
 776         /* Check if lower fs has 32bit inode numbers */
 777         if (fh_type != FILEID_INO32_GEN)
 778                 ofs->xino_bits = 0;
 779
 780         return 0;
 781
 782 out_put:
 783         path_put_init(path);
 784 out:
 785         return err;
 786 }
 787
 788 /* Workdir should not be subdir of upperdir and vice versa */
 789 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 790 {
 791         bool ok = false;
 792
 793         if (workdir != upperdir) {
 794                 ok = (lock_rename(workdir, upperdir) == NULL);
 795                 unlock_rename(workdir, upperdir);
 796         }
 797         return ok;
 798 }
 799
 800 static unsigned int ovl_split_lowerdirs(char *str)
 801 {
 802         unsigned int ctr = 1;
 803         char *s, *d;
 804
 805         for (s = d = str;; s++, d++) {
 806                 if (*s == '\\') {
 807                         s++;
 808                 } else if (*s == ':') {
 809                         *d = '\0';
 810                         ctr++;
 811                         continue;
 812                 }
 813                 *d = *s;
 814                 if (!*s)
 815                         break;
 816         }
 817         return ctr;
 818 }
 819
 820 static int __maybe_unused
 821 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 822                         struct dentry *dentry, struct inode *inode,
 823                         const char *name, void *buffer, size_t size)
 824 {
 825         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 826 }
 827
 828 static int __maybe_unused
 829 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 830                         struct dentry *dentry, struct inode *inode,
 831                         const char *name, const void *value,
 832                         size_t size, int flags)
 833 {
 834         struct dentry *workdir = ovl_workdir(dentry);
 835         struct inode *realinode = ovl_inode_real(inode);
 836         struct posix_acl *acl = NULL;
 837         int err;
 838
 839         /* Check that everything is OK before copy-up */
 840         if (value) {
 841                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 842                 if (IS_ERR(acl))
 843                         return PTR_ERR(acl);
 844         }
 845         err = -EOPNOTSUPP;
 846         if (!IS_POSIXACL(d_inode(workdir)))
 847                 goto out_acl_release;
 848         if (!realinode->i_op->set_acl)
 849                 goto out_acl_release;
 850         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 851                 err = acl ? -EACCES : 0;
 852                 goto out_acl_release;
 853         }
 854         err = -EPERM;
 855         if (!inode_owner_or_capable(inode))
 856                 goto out_acl_release;
 857
 858         posix_acl_release(acl);
 859
 860         /*
 861          * Check if sgid bit needs to be cleared (actual setacl operation will
 862          * be done with mounter's capabilities and so that won't do it for us).
 863          */
 864         if (unlikely(inode->i_mode & S_ISGID) &&
 865             handler->flags == ACL_TYPE_ACCESS &&
 866             !in_group_p(inode->i_gid) &&
 867             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 868                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 869
 870                 err = ovl_setattr(dentry, &iattr);
 871                 if (err)
 872                         return err;
 873         }
 874
 875         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
 876         if (!err)
 877                 ovl_copyattr(ovl_inode_real(inode), inode);
 878
 879         return err;
 880
 881 out_acl_release:
 882         posix_acl_release(acl);
 883         return err;
 884 }
 885
 886 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 887                              struct dentry *dentry, struct inode *inode,
 888                              const char *name, void *buffer, size_t size)
 889 {
 890         return -EOPNOTSUPP;
 891 }
 892
 893 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 894                              struct dentry *dentry, struct inode *inode,
 895                              const char *name, const void *value,
 896                              size_t size, int flags)
 897 {
 898         return -EOPNOTSUPP;
 899 }
 900
 901 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 902                                struct dentry *dentry, struct inode *inode,
 903                                const char *name, void *buffer, size_t size)
 904 {
 905         return ovl_xattr_get(dentry, inode, name, buffer, size);
 906 }
 907
 908 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 909                                struct dentry *dentry, struct inode *inode,
 910                                const char *name, const void *value,
 911                                size_t size, int flags)
 912 {
 913         return ovl_xattr_set(dentry, inode, name, value, size, flags);
 914 }
 915
 916 static const struct xattr_handler __maybe_unused
 917 ovl_posix_acl_access_xattr_handler = {
 918         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 919         .flags = ACL_TYPE_ACCESS,
 920         .get = ovl_posix_acl_xattr_get,
 921         .set = ovl_posix_acl_xattr_set,
 922 };
 923
 924 static const struct xattr_handler __maybe_unused
 925 ovl_posix_acl_default_xattr_handler = {
 926         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 927         .flags = ACL_TYPE_DEFAULT,
 928         .get = ovl_posix_acl_xattr_get,
 929         .set = ovl_posix_acl_xattr_set,
 930 };
 931
 932 static const struct xattr_handler ovl_own_xattr_handler = {
 933         .prefix = OVL_XATTR_PREFIX,
 934         .get = ovl_own_xattr_get,
 935         .set = ovl_own_xattr_set,
 936 };
 937
 938 static const struct xattr_handler ovl_other_xattr_handler = {
 939         .prefix = "", /* catch all */
 940         .get = ovl_other_xattr_get,
 941         .set = ovl_other_xattr_set,
 942 };
 943
 944 static const struct xattr_handler *ovl_xattr_handlers[] = {
 945 #ifdef CONFIG_FS_POSIX_ACL
 946         &ovl_posix_acl_access_xattr_handler,
 947         &ovl_posix_acl_default_xattr_handler,
 948 #endif
 949         &ovl_own_xattr_handler,
 950         &ovl_other_xattr_handler,
 951         NULL
 952 };
 953
 954 static int ovl_get_upper(struct ovl_fs *ofs, struct path *upperpath)
 955 {
 956         struct vfsmount *upper_mnt;
 957         int err;
 958
 959         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
 960         if (err)
 961                 goto out;
 962
 963         /* Upper fs should not be r/o */
 964         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
 965                 pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 966                 err = -EINVAL;
 967                 goto out;
 968         }
 969
 970         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
 971         if (err)
 972                 goto out;
 973
 974         err = -EBUSY;
 975         if (ovl_inuse_trylock(upperpath->dentry)) {
 976                 ofs->upperdir_locked = true;
 977         } else if (ofs->config.index) {
 978                 pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n");
 979                 goto out;
 980         } else {
 981                 pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
 982         }
 983
 984         upper_mnt = clone_private_mount(upperpath);
 985         err = PTR_ERR(upper_mnt);
 986         if (IS_ERR(upper_mnt)) {
 987                 pr_err("overlayfs: failed to clone upperpath\n");
 988                 goto out;
 989         }
 990
 991         /* Don't inherit atime flags */
 992         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
 993         ofs->upper_mnt = upper_mnt;
 994         err = 0;
 995 out:
 996         return err;
 997 }
 998
 999 static int ovl_make_workdir(struct ovl_fs *ofs, struct path *workpath)
1000 {
1001         struct vfsmount *mnt = ofs->upper_mnt;
1002         struct dentry *temp;
1003         int fh_type;
1004         int err;
1005
1006         err = mnt_want_write(mnt);
1007         if (err)
1008                 return err;
1009
1010         ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1011         if (!ofs->workdir)
1012                 goto out;
1013
1014         /*
1015          * Upper should support d_type, else whiteouts are visible.  Given
1016          * workdir and upper are on same fs, we can do iterate_dir() on
1017          * workdir. This check requires successful creation of workdir in
1018          * previous step.
1019          */
1020         err = ovl_check_d_type_supported(workpath);
1021         if (err < 0)
1022                 goto out;
1023
1024         /*
1025          * We allowed this configuration and don't want to break users over
1026          * kernel upgrade. So warn instead of erroring out.
1027          */
1028         if (!err)
1029                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
1030
1031         /* Check if upper/work fs supports O_TMPFILE */
1032         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1033         ofs->tmpfile = !IS_ERR(temp);
1034         if (ofs->tmpfile)
1035                 dput(temp);
1036         else
1037                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
1038
1039         /*
1040          * Check if upper/work fs supports trusted.overlay.* xattr
1041          */
1042         err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
1043         if (err) {
1044                 ofs->noxattr = true;
1045                 ofs->config.index = false;
1046                 pr_warn("overlayfs: upper fs does not support xattr, falling back to index=off.\n");
1047                 err = 0;
1048         } else {
1049                 vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
1050         }
1051
1052         /* Check if upper/work fs supports file handles */
1053         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1054         if (ofs->config.index && !fh_type) {
1055                 ofs->config.index = false;
1056                 pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1057         }
1058
1059         /* Check if upper fs has 32bit inode numbers */
1060         if (fh_type != FILEID_INO32_GEN)
1061                 ofs->xino_bits = 0;
1062
1063         /* NFS export of r/w mount depends on index */
1064         if (ofs->config.nfs_export && !ofs->config.index) {
1065                 pr_warn("overlayfs: NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1066                 ofs->config.nfs_export = false;
1067         }
1068
1069 out:
1070         mnt_drop_write(mnt);
1071         return err;
1072 }
1073
1074 static int ovl_get_workdir(struct ovl_fs *ofs, struct path *upperpath)
1075 {
1076         int err;
1077         struct path workpath = { };
1078
1079         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1080         if (err)
1081                 goto out;
1082
1083         err = -EINVAL;
1084         if (upperpath->mnt != workpath.mnt) {
1085                 pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1086                 goto out;
1087         }
1088         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1089                 pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1090                 goto out;
1091         }
1092
1093         err = -EBUSY;
1094         if (ovl_inuse_trylock(workpath.dentry)) {
1095                 ofs->workdir_locked = true;
1096         } else if (ofs->config.index) {
1097                 pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n");
1098                 goto out;
1099         } else {
1100                 pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
1101         }
1102
1103         ofs->workbasedir = dget(workpath.dentry);
1104         err = ovl_make_workdir(ofs, &workpath);
1105         if (err)
1106                 goto out;
1107
1108         err = 0;
1109 out:
1110         path_put(&workpath);
1111
1112         return err;
1113 }
1114
1115 static int ovl_get_indexdir(struct ovl_fs *ofs, struct ovl_entry *oe,
1116                             struct path *upperpath)
1117 {
1118         struct vfsmount *mnt = ofs->upper_mnt;
1119         int err;
1120
1121         err = mnt_want_write(mnt);
1122         if (err)
1123                 return err;
1124
1125         /* Verify lower root is upper root origin */
1126         err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
1127                                 true);
1128         if (err) {
1129                 pr_err("overlayfs: failed to verify upper root origin\n");
1130                 goto out;
1131         }
1132
1133         ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1134         if (ofs->indexdir) {
1135                 /*
1136                  * Verify upper root is exclusively associated with index dir.
1137                  * Older kernels stored upper fh in "trusted.overlay.origin"
1138                  * xattr. If that xattr exists, verify that it is a match to
1139                  * upper dir file handle. In any case, verify or set xattr
1140                  * "trusted.overlay.upper" to indicate that index may have
1141                  * directory entries.
1142                  */
1143                 if (ovl_check_origin_xattr(ofs->indexdir)) {
1144                         err = ovl_verify_set_fh(ofs->indexdir, OVL_XATTR_ORIGIN,
1145                                                 upperpath->dentry, true, false);
1146                         if (err)
1147                                 pr_err("overlayfs: failed to verify index dir 'origin' xattr\n");
1148                 }
1149                 err = ovl_verify_upper(ofs->indexdir, upperpath->dentry, true);
1150                 if (err)
1151                         pr_err("overlayfs: failed to verify index dir 'upper' xattr\n");
1152
1153                 /* Cleanup bad/stale/orphan index entries */
1154                 if (!err)
1155                         err = ovl_indexdir_cleanup(ofs);
1156         }
1157         if (err || !ofs->indexdir)
1158                 pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1159
1160 out:
1161         mnt_drop_write(mnt);
1162         return err;
1163 }
1164
1165 /* Get a unique fsid for the layer */
1166 static int ovl_get_fsid(struct ovl_fs *ofs, struct super_block *sb)
1167 {
1168         unsigned int i;
1169         dev_t dev;
1170         int err;
1171
1172         /* fsid 0 is reserved for upper fs even with non upper overlay */
1173         if (ofs->upper_mnt && ofs->upper_mnt->mnt_sb == sb)
1174                 return 0;
1175
1176         for (i = 0; i < ofs->numlowerfs; i++) {
1177                 if (ofs->lower_fs[i].sb == sb)
1178                         return i + 1;
1179         }
1180
1181         err = get_anon_bdev(&dev);
1182         if (err) {
1183                 pr_err("overlayfs: failed to get anonymous bdev for lowerpath\n");
1184                 return err;
1185         }
1186
1187         ofs->lower_fs[ofs->numlowerfs].sb = sb;
1188         ofs->lower_fs[ofs->numlowerfs].pseudo_dev = dev;
1189         ofs->numlowerfs++;
1190
1191         return ofs->numlowerfs;
1192 }
1193
1194 static int ovl_get_lower_layers(struct ovl_fs *ofs, struct path *stack,
1195                                 unsigned int numlower)
1196 {
1197         int err;
1198         unsigned int i;
1199
1200         err = -ENOMEM;
1201         ofs->lower_layers = kcalloc(numlower, sizeof(struct ovl_layer),
1202                                     GFP_KERNEL);
1203         if (ofs->lower_layers == NULL)
1204                 goto out;
1205
1206         ofs->lower_fs = kcalloc(numlower, sizeof(struct ovl_sb),
1207                                 GFP_KERNEL);
1208         if (ofs->lower_fs == NULL)
1209                 goto out;
1210
1211         for (i = 0; i < numlower; i++) {
1212                 struct vfsmount *mnt;
1213                 int fsid;
1214
1215                 err = fsid = ovl_get_fsid(ofs, stack[i].mnt->mnt_sb);
1216                 if (err < 0)
1217                         goto out;
1218
1219                 mnt = clone_private_mount(&stack[i]);
1220                 err = PTR_ERR(mnt);
1221                 if (IS_ERR(mnt)) {
1222                         pr_err("overlayfs: failed to clone lowerpath\n");
1223                         goto out;
1224                 }
1225
1226                 /*
1227                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1228                  * will fail instead of modifying lower fs.
1229                  */
1230                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1231
1232                 ofs->lower_layers[ofs->numlower].mnt = mnt;
1233                 ofs->lower_layers[ofs->numlower].idx = i + 1;
1234                 ofs->lower_layers[ofs->numlower].fsid = fsid;
1235                 if (fsid) {
1236                         ofs->lower_layers[ofs->numlower].fs =
1237                                 &ofs->lower_fs[fsid - 1];
1238                 }
1239                 ofs->numlower++;
1240         }
1241
1242         /*
1243          * When all layers on same fs, overlay can use real inode numbers.
1244          * With mount option "xino=on", mounter declares that there are enough
1245          * free high bits in underlying fs to hold the unique fsid.
1246          * If overlayfs does encounter underlying inodes using the high xino
1247          * bits reserved for fsid, it emits a warning and uses the original
1248          * inode number.
1249          */
1250         if (!ofs->numlowerfs || (ofs->numlowerfs == 1 && !ofs->upper_mnt)) {
1251                 ofs->xino_bits = 0;
1252                 ofs->config.xino = OVL_XINO_OFF;
1253         } else if (ofs->config.xino == OVL_XINO_ON && !ofs->xino_bits) {
1254                 /*
1255                  * This is a roundup of number of bits needed for numlowerfs+1
1256                  * (i.e. ilog2(numlowerfs+1 - 1) + 1). fsid 0 is reserved for
1257                  * upper fs even with non upper overlay.
1258                  */
1259                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 31);
1260                 ofs->xino_bits = ilog2(ofs->numlowerfs) + 1;
1261         }
1262
1263         if (ofs->xino_bits) {
1264                 pr_info("overlayfs: \"xino\" feature enabled using %d upper inode bits.\n",
1265                         ofs->xino_bits);
1266         }
1267
1268         err = 0;
1269 out:
1270         return err;
1271 }
1272
1273 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1274                                             struct ovl_fs *ofs)
1275 {
1276         int err;
1277         char *lowertmp, *lower;
1278         struct path *stack = NULL;
1279         unsigned int stacklen, numlower = 0, i;
1280         bool remote = false;
1281         struct ovl_entry *oe;
1282
1283         err = -ENOMEM;
1284         lowertmp = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1285         if (!lowertmp)
1286                 goto out_err;
1287
1288         err = -EINVAL;
1289         stacklen = ovl_split_lowerdirs(lowertmp);
1290         if (stacklen > OVL_MAX_STACK) {
1291                 pr_err("overlayfs: too many lower directories, limit is %d\n",
1292                        OVL_MAX_STACK);
1293                 goto out_err;
1294         } else if (!ofs->config.upperdir && stacklen == 1) {
1295                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1296                 goto out_err;
1297         } else if (!ofs->config.upperdir && ofs->config.nfs_export &&
1298                    ofs->config.redirect_follow) {
1299                 pr_warn("overlayfs: NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
1300                 ofs->config.nfs_export = false;
1301         }
1302
1303         err = -ENOMEM;
1304         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1305         if (!stack)
1306                 goto out_err;
1307
1308         err = -EINVAL;
1309         lower = lowertmp;
1310         for (numlower = 0; numlower < stacklen; numlower++) {
1311                 err = ovl_lower_dir(lower, &stack[numlower], ofs,
1312                                     &sb->s_stack_depth, &remote);
1313                 if (err)
1314                         goto out_err;
1315
1316                 lower = strchr(lower, '\0') + 1;
1317         }
1318
1319         err = -EINVAL;
1320         sb->s_stack_depth++;
1321         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1322                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1323                 goto out_err;
1324         }
1325
1326         err = ovl_get_lower_layers(ofs, stack, numlower);
1327         if (err)
1328                 goto out_err;
1329
1330         err = -ENOMEM;
1331         oe = ovl_alloc_entry(numlower);
1332         if (!oe)
1333                 goto out_err;
1334
1335         for (i = 0; i < numlower; i++) {
1336                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1337                 oe->lowerstack[i].layer = &ofs->lower_layers[i];
1338         }
1339
1340         if (remote)
1341                 sb->s_d_op = &ovl_reval_dentry_operations;
1342         else
1343                 sb->s_d_op = &ovl_dentry_operations;
1344
1345 out:
1346         for (i = 0; i < numlower; i++)
1347                 path_put(&stack[i]);
1348         kfree(stack);
1349         kfree(lowertmp);
1350
1351         return oe;
1352
1353 out_err:
1354         oe = ERR_PTR(err);
1355         goto out;
1356 }
1357
1358 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1359 {
1360         struct path upperpath = { };
1361         struct dentry *root_dentry;
1362         struct ovl_entry *oe;
1363         struct ovl_fs *ofs;
1364         struct cred *cred;
1365         int err;
1366
1367         err = -ENOMEM;
1368         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1369         if (!ofs)
1370                 goto out;
1371
1372         ofs->creator_cred = cred = prepare_creds();
1373         if (!cred)
1374                 goto out_err;
1375
1376         ofs->config.index = ovl_index_def;
1377         ofs->config.nfs_export = ovl_nfs_export_def;
1378         ofs->config.xino = ovl_xino_def();
1379         err = ovl_parse_opt((char *) data, &ofs->config);
1380         if (err)
1381                 goto out_err;
1382
1383         err = -EINVAL;
1384         if (!ofs->config.lowerdir) {
1385                 if (!silent)
1386                         pr_err("overlayfs: missing 'lowerdir'\n");
1387                 goto out_err;
1388         }
1389
1390         sb->s_stack_depth = 0;
1391         sb->s_maxbytes = MAX_LFS_FILESIZE;
1392         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1393         if (ofs->config.xino != OVL_XINO_OFF)
1394                 ofs->xino_bits = BITS_PER_LONG - 32;
1395
1396         if (ofs->config.upperdir) {
1397                 if (!ofs->config.workdir) {
1398                         pr_err("overlayfs: missing 'workdir'\n");
1399                         goto out_err;
1400                 }
1401
1402                 err = ovl_get_upper(ofs, &upperpath);
1403                 if (err)
1404                         goto out_err;
1405
1406                 err = ovl_get_workdir(ofs, &upperpath);
1407                 if (err)
1408                         goto out_err;
1409
1410                 if (!ofs->workdir)
1411                         sb->s_flags |= SB_RDONLY;
1412
1413                 sb->s_stack_depth = ofs->upper_mnt->mnt_sb->s_stack_depth;
1414                 sb->s_time_gran = ofs->upper_mnt->mnt_sb->s_time_gran;
1415
1416         }
1417         oe = ovl_get_lowerstack(sb, ofs);
1418         err = PTR_ERR(oe);
1419         if (IS_ERR(oe))
1420                 goto out_err;
1421
1422         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1423         if (!ofs->upper_mnt)
1424                 sb->s_flags |= SB_RDONLY;
1425
1426         if (!(ovl_force_readonly(ofs)) && ofs->config.index) {
1427                 err = ovl_get_indexdir(ofs, oe, &upperpath);
1428                 if (err)
1429                         goto out_free_oe;
1430
1431                 /* Force r/o mount with no index dir */
1432                 if (!ofs->indexdir) {
1433                         dput(ofs->workdir);
1434                         ofs->workdir = NULL;
1435                         sb->s_flags |= SB_RDONLY;
1436                 }
1437
1438         }
1439
1440         /* Show index=off in /proc/mounts for forced r/o mount */
1441         if (!ofs->indexdir) {
1442                 ofs->config.index = false;
1443                 if (ofs->upper_mnt && ofs->config.nfs_export) {
1444                         pr_warn("overlayfs: NFS export requires an index dir, falling back to nfs_export=off.\n");
1445                         ofs->config.nfs_export = false;
1446                 }
1447         }
1448
1449         if (ofs->config.nfs_export)
1450                 sb->s_export_op = &ovl_export_operations;
1451
1452         /* Never override disk quota limits or use reserved space */
1453         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1454
1455         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1456         sb->s_op = &ovl_super_operations;
1457         sb->s_xattr = ovl_xattr_handlers;
1458         sb->s_fs_info = ofs;
1459         sb->s_flags |= SB_POSIXACL | SB_NOREMOTELOCK;
1460
1461         err = -ENOMEM;
1462         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1463         if (!root_dentry)
1464                 goto out_free_oe;
1465
1466         root_dentry->d_fsdata = oe;
1467
1468         mntput(upperpath.mnt);
1469         if (upperpath.dentry) {
1470                 ovl_dentry_set_upper_alias(root_dentry);
1471                 if (ovl_is_impuredir(upperpath.dentry))
1472                         ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1473         }
1474
1475         /* Root is always merge -> can have whiteouts */
1476         ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1477         ovl_dentry_set_flag(OVL_E_CONNECTED, root_dentry);
1478         ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1479                        ovl_dentry_lower(root_dentry));
1480
1481         sb->s_root = root_dentry;
1482
1483         return 0;
1484
1485 out_free_oe:
1486         ovl_entry_stack_free(oe);
1487         kfree(oe);
1488 out_err:
1489         path_put(&upperpath);
1490         ovl_free_fs(ofs);
1491 out:
1492         return err;
1493 }
1494
1495 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1496                                 const char *dev_name, void *raw_data)
1497 {
1498         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1499 }
1500
1501 static struct file_system_type ovl_fs_type = {
1502         .owner          = THIS_MODULE,
1503         .name           = "overlay",
1504         .mount          = ovl_mount,
1505         .kill_sb        = kill_anon_super,
1506 };
1507 MODULE_ALIAS_FS("overlay");
1508
1509 static void ovl_inode_init_once(void *foo)
1510 {
1511         struct ovl_inode *oi = foo;
1512
1513         inode_init_once(&oi->vfs_inode);
1514 }
1515
1516 static int __init ovl_init(void)
1517 {
1518         int err;
1519
1520         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1521                                              sizeof(struct ovl_inode), 0,
1522                                              (SLAB_RECLAIM_ACCOUNT|
1523                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1524                                              ovl_inode_init_once);
1525         if (ovl_inode_cachep == NULL)
1526                 return -ENOMEM;
1527
1528         err = register_filesystem(&ovl_fs_type);
1529         if (err)
1530                 kmem_cache_destroy(ovl_inode_cachep);
1531
1532         return err;
1533 }
1534
1535 static void __exit ovl_exit(void)
1536 {
1537         unregister_filesystem(&ovl_fs_type);
1538
1539         /*
1540          * Make sure all delayed rcu free inodes are flushed before we
1541          * destroy cache.
1542          */
1543         rcu_barrier();
1544         kmem_cache_destroy(ovl_inode_cachep);
1545
1546 }
1547
1548 module_init(ovl_init);
1549 module_exit(ovl_exit);