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 <miklos@szeredi.hu>");
  24 MODULE_DESCRIPTION("Overlay filesystem");
  25 MODULE_LICENSE("GPL");
  26
  27
  28 struct ovl_dir_cache;
  29
  30 #define OVL_MAX_STACK 500
  31
  32 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  33 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  34 MODULE_PARM_DESC(ovl_redirect_dir_def,
  35                  "Default to on or off for the redirect_dir feature");
  36
  37 static 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 bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
  68 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
  69 MODULE_PARM_DESC(ovl_metacopy_def,
  70                  "Default to on or off for the metadata only copy up feature");
  71
  72 static void ovl_dentry_release(struct dentry *dentry)
  73 {
  74         struct ovl_entry *oe = dentry->d_fsdata;
  75
  76         if (oe) {
  77                 ovl_entry_stack_free(oe);
  78                 kfree_rcu(oe, rcu);
  79         }
  80 }
  81
  82 static struct dentry *ovl_d_real(struct dentry *dentry,
  83                                  const struct inode *inode)
  84 {
  85         struct dentry *real;
  86
  87         /* It's an overlay file */
  88         if (inode && d_inode(dentry) == inode)
  89                 return dentry;
  90
  91         if (!d_is_reg(dentry)) {
  92                 if (!inode || inode == d_inode(dentry))
  93                         return dentry;
  94                 goto bug;
  95         }
  96
  97         real = ovl_dentry_upper(dentry);
  98         if (real && (inode == d_inode(real)))
  99                 return real;
 100
 101         if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
 102                 return real;
 103
 104         real = ovl_dentry_lowerdata(dentry);
 105         if (!real)
 106                 goto bug;
 107
 108         /* Handle recursion */
 109         real = d_real(real, inode);
 110
 111         if (!inode || inode == d_inode(real))
 112                 return real;
 113 bug:
 114         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
 115              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 116         return dentry;
 117 }
 118
 119 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 120 {
 121         struct ovl_entry *oe = dentry->d_fsdata;
 122         unsigned int i;
 123         int ret = 1;
 124
 125         for (i = 0; i < oe->numlower; i++) {
 126                 struct dentry *d = oe->lowerstack[i].dentry;
 127
 128                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 129                         ret = d->d_op->d_revalidate(d, flags);
 130                         if (ret < 0)
 131                                 return ret;
 132                         if (!ret) {
 133                                 if (!(flags & LOOKUP_RCU))
 134                                         d_invalidate(d);
 135                                 return -ESTALE;
 136                         }
 137                 }
 138         }
 139         return 1;
 140 }
 141
 142 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 143 {
 144         struct ovl_entry *oe = dentry->d_fsdata;
 145         unsigned int i;
 146         int ret = 1;
 147
 148         for (i = 0; i < oe->numlower; i++) {
 149                 struct dentry *d = oe->lowerstack[i].dentry;
 150
 151                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 152                         ret = d->d_op->d_weak_revalidate(d, flags);
 153                         if (ret <= 0)
 154                                 break;
 155                 }
 156         }
 157         return ret;
 158 }
 159
 160 static const struct dentry_operations ovl_dentry_operations = {
 161         .d_release = ovl_dentry_release,
 162         .d_real = ovl_d_real,
 163 };
 164
 165 static const struct dentry_operations ovl_reval_dentry_operations = {
 166         .d_release = ovl_dentry_release,
 167         .d_real = ovl_d_real,
 168         .d_revalidate = ovl_dentry_revalidate,
 169         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 170 };
 171
 172 static struct kmem_cache *ovl_inode_cachep;
 173
 174 static struct inode *ovl_alloc_inode(struct super_block *sb)
 175 {
 176         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 177
 178         if (!oi)
 179                 return NULL;
 180
 181         oi->cache = NULL;
 182         oi->redirect = NULL;
 183         oi->version = 0;
 184         oi->flags = 0;
 185         oi->__upperdentry = NULL;
 186         oi->lower = NULL;
 187         oi->lowerdata = NULL;
 188         mutex_init(&oi->lock);
 189
 190         return &oi->vfs_inode;
 191 }
 192
 193 static void ovl_free_inode(struct inode *inode)
 194 {
 195         struct ovl_inode *oi = OVL_I(inode);
 196
 197         kfree(oi->redirect);
 198         mutex_destroy(&oi->lock);
 199         kmem_cache_free(ovl_inode_cachep, oi);
 200 }
 201
 202 static void ovl_destroy_inode(struct inode *inode)
 203 {
 204         struct ovl_inode *oi = OVL_I(inode);
 205
 206         dput(oi->__upperdentry);
 207         iput(oi->lower);
 208         if (S_ISDIR(inode->i_mode))
 209                 ovl_dir_cache_free(inode);
 210         else
 211                 iput(oi->lowerdata);
 212 }
 213
 214 static void ovl_free_fs(struct ovl_fs *ofs)
 215 {
 216         unsigned i;
 217
 218         dput(ofs->indexdir);
 219         dput(ofs->workdir);
 220         if (ofs->workdir_locked)
 221                 ovl_inuse_unlock(ofs->workbasedir);
 222         dput(ofs->workbasedir);
 223         if (ofs->upperdir_locked)
 224                 ovl_inuse_unlock(ofs->upper_mnt->mnt_root);
 225         mntput(ofs->upper_mnt);
 226         for (i = 0; i < ofs->numlower; i++)
 227                 mntput(ofs->lower_layers[i].mnt);
 228         for (i = 0; i < ofs->numlowerfs; i++)
 229                 free_anon_bdev(ofs->lower_fs[i].pseudo_dev);
 230         kfree(ofs->lower_layers);
 231         kfree(ofs->lower_fs);
 232
 233         kfree(ofs->config.lowerdir);
 234         kfree(ofs->config.upperdir);
 235         kfree(ofs->config.workdir);
 236         kfree(ofs->config.redirect_mode);
 237         if (ofs->creator_cred)
 238                 put_cred(ofs->creator_cred);
 239         kfree(ofs);
 240 }
 241
 242 static void ovl_put_super(struct super_block *sb)
 243 {
 244         struct ovl_fs *ofs = sb->s_fs_info;
 245
 246         ovl_free_fs(ofs);
 247 }
 248
 249 /* Sync real dirty inodes in upper filesystem (if it exists) */
 250 static int ovl_sync_fs(struct super_block *sb, int wait)
 251 {
 252         struct ovl_fs *ofs = sb->s_fs_info;
 253         struct super_block *upper_sb;
 254         int ret;
 255
 256         if (!ofs->upper_mnt)
 257                 return 0;
 258
 259         /*
 260          * If this is a sync(2) call or an emergency sync, all the super blocks
 261          * will be iterated, including upper_sb, so no need to do anything.
 262          *
 263          * If this is a syncfs(2) call, then we do need to call
 264          * sync_filesystem() on upper_sb, but enough if we do it when being
 265          * called with wait == 1.
 266          */
 267         if (!wait)
 268                 return 0;
 269
 270         upper_sb = ofs->upper_mnt->mnt_sb;
 271
 272         down_read(&upper_sb->s_umount);
 273         ret = sync_filesystem(upper_sb);
 274         up_read(&upper_sb->s_umount);
 275
 276         return ret;
 277 }
 278
 279 /**
 280  * ovl_statfs
 281  * @sb: The overlayfs super block
 282  * @buf: The struct kstatfs to fill in with stats
 283  *
 284  * Get the filesystem statistics.  As writes always target the upper layer
 285  * filesystem pass the statfs to the upper filesystem (if it exists)
 286  */
 287 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 288 {
 289         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 290         struct dentry *root_dentry = dentry->d_sb->s_root;
 291         struct path path;
 292         int err;
 293
 294         ovl_path_real(root_dentry, &path);
 295
 296         err = vfs_statfs(&path, buf);
 297         if (!err) {
 298                 buf->f_namelen = ofs->namelen;
 299                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 300         }
 301
 302         return err;
 303 }
 304
 305 /* Will this overlay be forced to mount/remount ro? */
 306 static bool ovl_force_readonly(struct ovl_fs *ofs)
 307 {
 308         return (!ofs->upper_mnt || !ofs->workdir);
 309 }
 310
 311 static const char *ovl_redirect_mode_def(void)
 312 {
 313         return ovl_redirect_dir_def ? "on" : "off";
 314 }
 315
 316 enum {
 317         OVL_XINO_OFF,
 318         OVL_XINO_AUTO,
 319         OVL_XINO_ON,
 320 };
 321
 322 static const char * const ovl_xino_str[] = {
 323         "off",
 324         "auto",
 325         "on",
 326 };
 327
 328 static inline int ovl_xino_def(void)
 329 {
 330         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 331 }
 332
 333 /**
 334  * ovl_show_options
 335  *
 336  * Prints the mount options for a given superblock.
 337  * Returns zero; does not fail.
 338  */
 339 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 340 {
 341         struct super_block *sb = dentry->d_sb;
 342         struct ovl_fs *ofs = sb->s_fs_info;
 343
 344         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 345         if (ofs->config.upperdir) {
 346                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 347                 seq_show_option(m, "workdir", ofs->config.workdir);
 348         }
 349         if (ofs->config.default_permissions)
 350                 seq_puts(m, ",default_permissions");
 351         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 352                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 353         if (ofs->config.index != ovl_index_def)
 354                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 355         if (ofs->config.nfs_export != ovl_nfs_export_def)
 356                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 357                                                 "on" : "off");
 358         if (ofs->config.xino != ovl_xino_def())
 359                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 360         if (ofs->config.metacopy != ovl_metacopy_def)
 361                 seq_printf(m, ",metacopy=%s",
 362                            ofs->config.metacopy ? "on" : "off");
 363         return 0;
 364 }
 365
 366 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 367 {
 368         struct ovl_fs *ofs = sb->s_fs_info;
 369
 370         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 371                 return -EROFS;
 372
 373         return 0;
 374 }
 375
 376 static const struct super_operations ovl_super_operations = {
 377         .alloc_inode    = ovl_alloc_inode,
 378         .free_inode     = ovl_free_inode,
 379         .destroy_inode  = ovl_destroy_inode,
 380         .drop_inode     = generic_delete_inode,
 381         .put_super      = ovl_put_super,
 382         .sync_fs        = ovl_sync_fs,
 383         .statfs         = ovl_statfs,
 384         .show_options   = ovl_show_options,
 385         .remount_fs     = ovl_remount,
 386 };
 387
 388 enum {
 389         OPT_LOWERDIR,
 390         OPT_UPPERDIR,
 391         OPT_WORKDIR,
 392         OPT_DEFAULT_PERMISSIONS,
 393         OPT_REDIRECT_DIR,
 394         OPT_INDEX_ON,
 395         OPT_INDEX_OFF,
 396         OPT_NFS_EXPORT_ON,
 397         OPT_NFS_EXPORT_OFF,
 398         OPT_XINO_ON,
 399         OPT_XINO_OFF,
 400         OPT_XINO_AUTO,
 401         OPT_METACOPY_ON,
 402         OPT_METACOPY_OFF,
 403         OPT_ERR,
 404 };
 405
 406 static const match_table_t ovl_tokens = {
 407         {OPT_LOWERDIR,                  "lowerdir=%s"},
 408         {OPT_UPPERDIR,                  "upperdir=%s"},
 409         {OPT_WORKDIR,                   "workdir=%s"},
 410         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 411         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 412         {OPT_INDEX_ON,                  "index=on"},
 413         {OPT_INDEX_OFF,                 "index=off"},
 414         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 415         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 416         {OPT_XINO_ON,                   "xino=on"},
 417         {OPT_XINO_OFF,                  "xino=off"},
 418         {OPT_XINO_AUTO,                 "xino=auto"},
 419         {OPT_METACOPY_ON,               "metacopy=on"},
 420         {OPT_METACOPY_OFF,              "metacopy=off"},
 421         {OPT_ERR,                       NULL}
 422 };
 423
 424 static char *ovl_next_opt(char **s)
 425 {
 426         char *sbegin = *s;
 427         char *p;
 428
 429         if (sbegin == NULL)
 430                 return NULL;
 431
 432         for (p = sbegin; *p; p++) {
 433                 if (*p == '\\') {
 434                         p++;
 435                         if (!*p)
 436                                 break;
 437                 } else if (*p == ',') {
 438                         *p = '\0';
 439                         *s = p + 1;
 440                         return sbegin;
 441                 }
 442         }
 443         *s = NULL;
 444         return sbegin;
 445 }
 446
 447 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 448 {
 449         if (strcmp(mode, "on") == 0) {
 450                 config->redirect_dir = true;
 451                 /*
 452                  * Does not make sense to have redirect creation without
 453                  * redirect following.
 454                  */
 455                 config->redirect_follow = true;
 456         } else if (strcmp(mode, "follow") == 0) {
 457                 config->redirect_follow = true;
 458         } else if (strcmp(mode, "off") == 0) {
 459                 if (ovl_redirect_always_follow)
 460                         config->redirect_follow = true;
 461         } else if (strcmp(mode, "nofollow") != 0) {
 462                 pr_err("overlayfs: bad mount option \"redirect_dir=%s\"\n",
 463                        mode);
 464                 return -EINVAL;
 465         }
 466
 467         return 0;
 468 }
 469
 470 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 471 {
 472         char *p;
 473         int err;
 474         bool metacopy_opt = false, redirect_opt = false;
 475
 476         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 477         if (!config->redirect_mode)
 478                 return -ENOMEM;
 479
 480         while ((p = ovl_next_opt(&opt)) != NULL) {
 481                 int token;
 482                 substring_t args[MAX_OPT_ARGS];
 483
 484                 if (!*p)
 485                         continue;
 486
 487                 token = match_token(p, ovl_tokens, args);
 488                 switch (token) {
 489                 case OPT_UPPERDIR:
 490                         kfree(config->upperdir);
 491                         config->upperdir = match_strdup(&args[0]);
 492                         if (!config->upperdir)
 493                                 return -ENOMEM;
 494                         break;
 495
 496                 case OPT_LOWERDIR:
 497                         kfree(config->lowerdir);
 498                         config->lowerdir = match_strdup(&args[0]);
 499                         if (!config->lowerdir)
 500                                 return -ENOMEM;
 501                         break;
 502
 503                 case OPT_WORKDIR:
 504                         kfree(config->workdir);
 505                         config->workdir = match_strdup(&args[0]);
 506                         if (!config->workdir)
 507                                 return -ENOMEM;
 508                         break;
 509
 510                 case OPT_DEFAULT_PERMISSIONS:
 511                         config->default_permissions = true;
 512                         break;
 513
 514                 case OPT_REDIRECT_DIR:
 515                         kfree(config->redirect_mode);
 516                         config->redirect_mode = match_strdup(&args[0]);
 517                         if (!config->redirect_mode)
 518                                 return -ENOMEM;
 519                         redirect_opt = true;
 520                         break;
 521
 522                 case OPT_INDEX_ON:
 523                         config->index = true;
 524                         break;
 525
 526                 case OPT_INDEX_OFF:
 527                         config->index = false;
 528                         break;
 529
 530                 case OPT_NFS_EXPORT_ON:
 531                         config->nfs_export = true;
 532                         break;
 533
 534                 case OPT_NFS_EXPORT_OFF:
 535                         config->nfs_export = false;
 536                         break;
 537
 538                 case OPT_XINO_ON:
 539                         config->xino = OVL_XINO_ON;
 540                         break;
 541
 542                 case OPT_XINO_OFF:
 543                         config->xino = OVL_XINO_OFF;
 544                         break;
 545
 546                 case OPT_XINO_AUTO:
 547                         config->xino = OVL_XINO_AUTO;
 548                         break;
 549
 550                 case OPT_METACOPY_ON:
 551                         config->metacopy = true;
 552                         metacopy_opt = true;
 553                         break;
 554
 555                 case OPT_METACOPY_OFF:
 556                         config->metacopy = false;
 557                         break;
 558
 559                 default:
 560                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 561                         return -EINVAL;
 562                 }
 563         }
 564
 565         /* Workdir is useless in non-upper mount */
 566         if (!config->upperdir && config->workdir) {
 567                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 568                         config->workdir);
 569                 kfree(config->workdir);
 570                 config->workdir = NULL;
 571         }
 572
 573         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 574         if (err)
 575                 return err;
 576
 577         /*
 578          * This is to make the logic below simpler.  It doesn't make any other
 579          * difference, since config->redirect_dir is only used for upper.
 580          */
 581         if (!config->upperdir && config->redirect_follow)
 582                 config->redirect_dir = true;
 583
 584         /* Resolve metacopy -> redirect_dir dependency */
 585         if (config->metacopy && !config->redirect_dir) {
 586                 if (metacopy_opt && redirect_opt) {
 587                         pr_err("overlayfs: conflicting options: metacopy=on,redirect_dir=%s\n",
 588                                config->redirect_mode);
 589                         return -EINVAL;
 590                 }
 591                 if (redirect_opt) {
 592                         /*
 593                          * There was an explicit redirect_dir=... that resulted
 594                          * in this conflict.
 595                          */
 596                         pr_info("overlayfs: disabling metacopy due to redirect_dir=%s\n",
 597                                 config->redirect_mode);
 598                         config->metacopy = false;
 599                 } else {
 600                         /* Automatically enable redirect otherwise. */
 601                         config->redirect_follow = config->redirect_dir = true;
 602                 }
 603         }
 604
 605         return 0;
 606 }
 607
 608 #define OVL_WORKDIR_NAME "work"
 609 #define OVL_INDEXDIR_NAME "index"
 610
 611 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 612                                          const char *name, bool persist)
 613 {
 614         struct inode *dir =  ofs->workbasedir->d_inode;
 615         struct vfsmount *mnt = ofs->upper_mnt;
 616         struct dentry *work;
 617         int err;
 618         bool retried = false;
 619         bool locked = false;
 620
 621         inode_lock_nested(dir, I_MUTEX_PARENT);
 622         locked = true;
 623
 624 retry:
 625         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
 626
 627         if (!IS_ERR(work)) {
 628                 struct iattr attr = {
 629                         .ia_valid = ATTR_MODE,
 630                         .ia_mode = S_IFDIR | 0,
 631                 };
 632
 633                 if (work->d_inode) {
 634                         err = -EEXIST;
 635                         if (retried)
 636                                 goto out_dput;
 637
 638                         if (persist)
 639                                 goto out_unlock;
 640
 641                         retried = true;
 642                         ovl_workdir_cleanup(dir, mnt, work, 0);
 643                         dput(work);
 644                         goto retry;
 645                 }
 646
 647                 work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
 648                 err = PTR_ERR(work);
 649                 if (IS_ERR(work))
 650                         goto out_err;
 651
 652                 /*
 653                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 654                  *
 655                  * a) success (there was a POSIX ACL xattr and was removed)
 656                  * b) -ENODATA (there was no POSIX ACL xattr)
 657                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 658                  *
 659                  * There are various other error values that could effectively
 660                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 661                  * if the xattr name is too long), but the set of filesystems
 662                  * allowed as upper are limited to "normal" ones, where checking
 663                  * for the above two errors is sufficient.
 664                  */
 665                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 666                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 667                         goto out_dput;
 668
 669                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 670                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 671                         goto out_dput;
 672
 673                 /* Clear any inherited mode bits */
 674                 inode_lock(work->d_inode);
 675                 err = notify_change(work, &attr, NULL);
 676                 inode_unlock(work->d_inode);
 677                 if (err)
 678                         goto out_dput;
 679         } else {
 680                 err = PTR_ERR(work);
 681                 goto out_err;
 682         }
 683 out_unlock:
 684         if (locked)
 685                 inode_unlock(dir);
 686
 687         return work;
 688
 689 out_dput:
 690         dput(work);
 691 out_err:
 692         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 693                 ofs->config.workdir, name, -err);
 694         work = NULL;
 695         goto out_unlock;
 696 }
 697
 698 static void ovl_unescape(char *s)
 699 {
 700         char *d = s;
 701
 702         for (;; s++, d++) {
 703                 if (*s == '\\')
 704                         s++;
 705                 *d = *s;
 706                 if (!*s)
 707                         break;
 708         }
 709 }
 710
 711 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 712 {
 713         int err = -EINVAL;
 714
 715         if (!*name) {
 716                 pr_err("overlayfs: empty lowerdir\n");
 717                 goto out;
 718         }
 719         err = kern_path(name, LOOKUP_FOLLOW, path);
 720         if (err) {
 721                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 722                 goto out;
 723         }
 724         err = -EINVAL;
 725         if (ovl_dentry_weird(path->dentry)) {
 726                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 727                 goto out_put;
 728         }
 729         if (!d_is_dir(path->dentry)) {
 730                 pr_err("overlayfs: '%s' not a directory\n", name);
 731                 goto out_put;
 732         }
 733         return 0;
 734
 735 out_put:
 736         path_put_init(path);
 737 out:
 738         return err;
 739 }
 740
 741 static int ovl_mount_dir(const char *name, struct path *path)
 742 {
 743         int err = -ENOMEM;
 744         char *tmp = kstrdup(name, GFP_KERNEL);
 745
 746         if (tmp) {
 747                 ovl_unescape(tmp);
 748                 err = ovl_mount_dir_noesc(tmp, path);
 749
 750                 if (!err)
 751                         if (ovl_dentry_remote(path->dentry)) {
 752                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 753                                        tmp);
 754                                 path_put_init(path);
 755                                 err = -EINVAL;
 756                         }
 757                 kfree(tmp);
 758         }
 759         return err;
 760 }
 761
 762 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 763                              const char *name)
 764 {
 765         struct kstatfs statfs;
 766         int err = vfs_statfs(path, &statfs);
 767
 768         if (err)
 769                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 770         else
 771                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 772
 773         return err;
 774 }
 775
 776 static int ovl_lower_dir(const char *name, struct path *path,
 777                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 778 {
 779         int fh_type;
 780         int err;
 781
 782         err = ovl_mount_dir_noesc(name, path);
 783         if (err)
 784                 goto out;
 785
 786         err = ovl_check_namelen(path, ofs, name);
 787         if (err)
 788                 goto out_put;
 789
 790         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 791
 792         if (ovl_dentry_remote(path->dentry))
 793                 *remote = true;
 794
 795         /*
 796          * The inodes index feature and NFS export need to encode and decode
 797          * file handles, so they require that all layers support them.
 798          */
 799         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 800         if ((ofs->config.nfs_export ||
 801              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 802                 ofs->config.index = false;
 803                 ofs->config.nfs_export = false;
 804                 pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 805                         name);
 806         }
 807
 808         /* Check if lower fs has 32bit inode numbers */
 809         if (fh_type != FILEID_INO32_GEN)
 810                 ofs->xino_bits = 0;
 811
 812         return 0;
 813
 814 out_put:
 815         path_put_init(path);
 816 out:
 817         return err;
 818 }
 819
 820 /* Workdir should not be subdir of upperdir and vice versa */
 821 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 822 {
 823         bool ok = false;
 824
 825         if (workdir != upperdir) {
 826                 ok = (lock_rename(workdir, upperdir) == NULL);
 827                 unlock_rename(workdir, upperdir);
 828         }
 829         return ok;
 830 }
 831
 832 static unsigned int ovl_split_lowerdirs(char *str)
 833 {
 834         unsigned int ctr = 1;
 835         char *s, *d;
 836
 837         for (s = d = str;; s++, d++) {
 838                 if (*s == '\\') {
 839                         s++;
 840                 } else if (*s == ':') {
 841                         *d = '\0';
 842                         ctr++;
 843                         continue;
 844                 }
 845                 *d = *s;
 846                 if (!*s)
 847                         break;
 848         }
 849         return ctr;
 850 }
 851
 852 static int __maybe_unused
 853 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 854                         struct dentry *dentry, struct inode *inode,
 855                         const char *name, void *buffer, size_t size)
 856 {
 857         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 858 }
 859
 860 static int __maybe_unused
 861 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 862                         struct dentry *dentry, struct inode *inode,
 863                         const char *name, const void *value,
 864                         size_t size, int flags)
 865 {
 866         struct dentry *workdir = ovl_workdir(dentry);
 867         struct inode *realinode = ovl_inode_real(inode);
 868         struct posix_acl *acl = NULL;
 869         int err;
 870
 871         /* Check that everything is OK before copy-up */
 872         if (value) {
 873                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 874                 if (IS_ERR(acl))
 875                         return PTR_ERR(acl);
 876         }
 877         err = -EOPNOTSUPP;
 878         if (!IS_POSIXACL(d_inode(workdir)))
 879                 goto out_acl_release;
 880         if (!realinode->i_op->set_acl)
 881                 goto out_acl_release;
 882         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 883                 err = acl ? -EACCES : 0;
 884                 goto out_acl_release;
 885         }
 886         err = -EPERM;
 887         if (!inode_owner_or_capable(inode))
 888                 goto out_acl_release;
 889
 890         posix_acl_release(acl);
 891
 892         /*
 893          * Check if sgid bit needs to be cleared (actual setacl operation will
 894          * be done with mounter's capabilities and so that won't do it for us).
 895          */
 896         if (unlikely(inode->i_mode & S_ISGID) &&
 897             handler->flags == ACL_TYPE_ACCESS &&
 898             !in_group_p(inode->i_gid) &&
 899             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 900                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 901
 902                 err = ovl_setattr(dentry, &iattr);
 903                 if (err)
 904                         return err;
 905         }
 906
 907         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
 908         if (!err)
 909                 ovl_copyattr(ovl_inode_real(inode), inode);
 910
 911         return err;
 912
 913 out_acl_release:
 914         posix_acl_release(acl);
 915         return err;
 916 }
 917
 918 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 919                              struct dentry *dentry, struct inode *inode,
 920                              const char *name, void *buffer, size_t size)
 921 {
 922         return -EOPNOTSUPP;
 923 }
 924
 925 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 926                              struct dentry *dentry, struct inode *inode,
 927                              const char *name, const void *value,
 928                              size_t size, int flags)
 929 {
 930         return -EOPNOTSUPP;
 931 }
 932
 933 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 934                                struct dentry *dentry, struct inode *inode,
 935                                const char *name, void *buffer, size_t size)
 936 {
 937         return ovl_xattr_get(dentry, inode, name, buffer, size);
 938 }
 939
 940 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 941                                struct dentry *dentry, struct inode *inode,
 942                                const char *name, const void *value,
 943                                size_t size, int flags)
 944 {
 945         return ovl_xattr_set(dentry, inode, name, value, size, flags);
 946 }
 947
 948 static const struct xattr_handler __maybe_unused
 949 ovl_posix_acl_access_xattr_handler = {
 950         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 951         .flags = ACL_TYPE_ACCESS,
 952         .get = ovl_posix_acl_xattr_get,
 953         .set = ovl_posix_acl_xattr_set,
 954 };
 955
 956 static const struct xattr_handler __maybe_unused
 957 ovl_posix_acl_default_xattr_handler = {
 958         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 959         .flags = ACL_TYPE_DEFAULT,
 960         .get = ovl_posix_acl_xattr_get,
 961         .set = ovl_posix_acl_xattr_set,
 962 };
 963
 964 static const struct xattr_handler ovl_own_xattr_handler = {
 965         .prefix = OVL_XATTR_PREFIX,
 966         .get = ovl_own_xattr_get,
 967         .set = ovl_own_xattr_set,
 968 };
 969
 970 static const struct xattr_handler ovl_other_xattr_handler = {
 971         .prefix = "", /* catch all */
 972         .get = ovl_other_xattr_get,
 973         .set = ovl_other_xattr_set,
 974 };
 975
 976 static const struct xattr_handler *ovl_xattr_handlers[] = {
 977 #ifdef CONFIG_FS_POSIX_ACL
 978         &ovl_posix_acl_access_xattr_handler,
 979         &ovl_posix_acl_default_xattr_handler,
 980 #endif
 981         &ovl_own_xattr_handler,
 982         &ovl_other_xattr_handler,
 983         NULL
 984 };
 985
 986 static int ovl_get_upper(struct ovl_fs *ofs, struct path *upperpath)
 987 {
 988         struct vfsmount *upper_mnt;
 989         int err;
 990
 991         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
 992         if (err)
 993                 goto out;
 994
 995         /* Upper fs should not be r/o */
 996         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
 997                 pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 998                 err = -EINVAL;
 999                 goto out;
1000         }
1001
1002         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1003         if (err)
1004                 goto out;
1005
1006         upper_mnt = clone_private_mount(upperpath);
1007         err = PTR_ERR(upper_mnt);
1008         if (IS_ERR(upper_mnt)) {
1009                 pr_err("overlayfs: failed to clone upperpath\n");
1010                 goto out;
1011         }
1012
1013         /* Don't inherit atime flags */
1014         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1015         ofs->upper_mnt = upper_mnt;
1016
1017         err = -EBUSY;
1018         if (ovl_inuse_trylock(ofs->upper_mnt->mnt_root)) {
1019                 ofs->upperdir_locked = true;
1020         } else if (ofs->config.index) {
1021                 pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n");
1022                 goto out;
1023         } else {
1024                 pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
1025         }
1026
1027         err = 0;
1028 out:
1029         return err;
1030 }
1031
1032 static int ovl_make_workdir(struct ovl_fs *ofs, struct path *workpath)
1033 {
1034         struct vfsmount *mnt = ofs->upper_mnt;
1035         struct dentry *temp;
1036         int fh_type;
1037         int err;
1038
1039         err = mnt_want_write(mnt);
1040         if (err)
1041                 return err;
1042
1043         ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1044         if (!ofs->workdir)
1045                 goto out;
1046
1047         /*
1048          * Upper should support d_type, else whiteouts are visible.  Given
1049          * workdir and upper are on same fs, we can do iterate_dir() on
1050          * workdir. This check requires successful creation of workdir in
1051          * previous step.
1052          */
1053         err = ovl_check_d_type_supported(workpath);
1054         if (err < 0)
1055                 goto out;
1056
1057         /*
1058          * We allowed this configuration and don't want to break users over
1059          * kernel upgrade. So warn instead of erroring out.
1060          */
1061         if (!err)
1062                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
1063
1064         /* Check if upper/work fs supports O_TMPFILE */
1065         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1066         ofs->tmpfile = !IS_ERR(temp);
1067         if (ofs->tmpfile)
1068                 dput(temp);
1069         else
1070                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
1071
1072         /*
1073          * Check if upper/work fs supports trusted.overlay.* xattr
1074          */
1075         err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
1076         if (err) {
1077                 ofs->noxattr = true;
1078                 ofs->config.index = false;
1079                 ofs->config.metacopy = false;
1080                 pr_warn("overlayfs: upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1081                 err = 0;
1082         } else {
1083                 vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
1084         }
1085
1086         /* Check if upper/work fs supports file handles */
1087         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1088         if (ofs->config.index && !fh_type) {
1089                 ofs->config.index = false;
1090                 pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1091         }
1092
1093         /* Check if upper fs has 32bit inode numbers */
1094         if (fh_type != FILEID_INO32_GEN)
1095                 ofs->xino_bits = 0;
1096
1097         /* NFS export of r/w mount depends on index */
1098         if (ofs->config.nfs_export && !ofs->config.index) {
1099                 pr_warn("overlayfs: NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1100                 ofs->config.nfs_export = false;
1101         }
1102 out:
1103         mnt_drop_write(mnt);
1104         return err;
1105 }
1106
1107 static int ovl_get_workdir(struct ovl_fs *ofs, struct path *upperpath)
1108 {
1109         int err;
1110         struct path workpath = { };
1111
1112         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1113         if (err)
1114                 goto out;
1115
1116         err = -EINVAL;
1117         if (upperpath->mnt != workpath.mnt) {
1118                 pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1119                 goto out;
1120         }
1121         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1122                 pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1123                 goto out;
1124         }
1125
1126         ofs->workbasedir = dget(workpath.dentry);
1127
1128         err = -EBUSY;
1129         if (ovl_inuse_trylock(ofs->workbasedir)) {
1130                 ofs->workdir_locked = true;
1131         } else if (ofs->config.index) {
1132                 pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n");
1133                 goto out;
1134         } else {
1135                 pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
1136         }
1137
1138         err = ovl_make_workdir(ofs, &workpath);
1139         if (err)
1140                 goto out;
1141
1142         err = 0;
1143 out:
1144         path_put(&workpath);
1145
1146         return err;
1147 }
1148
1149 static int ovl_get_indexdir(struct ovl_fs *ofs, struct ovl_entry *oe,
1150                             struct path *upperpath)
1151 {
1152         struct vfsmount *mnt = ofs->upper_mnt;
1153         int err;
1154
1155         err = mnt_want_write(mnt);
1156         if (err)
1157                 return err;
1158
1159         /* Verify lower root is upper root origin */
1160         err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
1161                                 true);
1162         if (err) {
1163                 pr_err("overlayfs: failed to verify upper root origin\n");
1164                 goto out;
1165         }
1166
1167         ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1168         if (ofs->indexdir) {
1169                 /*
1170                  * Verify upper root is exclusively associated with index dir.
1171                  * Older kernels stored upper fh in "trusted.overlay.origin"
1172                  * xattr. If that xattr exists, verify that it is a match to
1173                  * upper dir file handle. In any case, verify or set xattr
1174                  * "trusted.overlay.upper" to indicate that index may have
1175                  * directory entries.
1176                  */
1177                 if (ovl_check_origin_xattr(ofs->indexdir)) {
1178                         err = ovl_verify_set_fh(ofs->indexdir, OVL_XATTR_ORIGIN,
1179                                                 upperpath->dentry, true, false);
1180                         if (err)
1181                                 pr_err("overlayfs: failed to verify index dir 'origin' xattr\n");
1182                 }
1183                 err = ovl_verify_upper(ofs->indexdir, upperpath->dentry, true);
1184                 if (err)
1185                         pr_err("overlayfs: failed to verify index dir 'upper' xattr\n");
1186
1187                 /* Cleanup bad/stale/orphan index entries */
1188                 if (!err)
1189                         err = ovl_indexdir_cleanup(ofs);
1190         }
1191         if (err || !ofs->indexdir)
1192                 pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1193
1194 out:
1195         mnt_drop_write(mnt);
1196         return err;
1197 }
1198
1199 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1200 {
1201         unsigned int i;
1202
1203         if (!ofs->config.nfs_export && !(ofs->config.index && ofs->upper_mnt))
1204                 return true;
1205
1206         for (i = 0; i < ofs->numlowerfs; i++) {
1207                 /*
1208                  * We use uuid to associate an overlay lower file handle with a
1209                  * lower layer, so we can accept lower fs with null uuid as long
1210                  * as all lower layers with null uuid are on the same fs.
1211                  */
1212                 if (uuid_equal(&ofs->lower_fs[i].sb->s_uuid, uuid))
1213                         return false;
1214         }
1215         return true;
1216 }
1217
1218 /* Get a unique fsid for the layer */
1219 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1220 {
1221         struct super_block *sb = path->mnt->mnt_sb;
1222         unsigned int i;
1223         dev_t dev;
1224         int err;
1225
1226         /* fsid 0 is reserved for upper fs even with non upper overlay */
1227         if (ofs->upper_mnt && ofs->upper_mnt->mnt_sb == sb)
1228                 return 0;
1229
1230         for (i = 0; i < ofs->numlowerfs; i++) {
1231                 if (ofs->lower_fs[i].sb == sb)
1232                         return i + 1;
1233         }
1234
1235         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1236                 ofs->config.index = false;
1237                 ofs->config.nfs_export = false;
1238                 pr_warn("overlayfs: %s uuid detected in lower fs '%pd2', falling back to index=off,nfs_export=off.\n",
1239                         uuid_is_null(&sb->s_uuid) ? "null" : "conflicting",
1240                         path->dentry);
1241         }
1242
1243         err = get_anon_bdev(&dev);
1244         if (err) {
1245                 pr_err("overlayfs: failed to get anonymous bdev for lowerpath\n");
1246                 return err;
1247         }
1248
1249         ofs->lower_fs[ofs->numlowerfs].sb = sb;
1250         ofs->lower_fs[ofs->numlowerfs].pseudo_dev = dev;
1251         ofs->numlowerfs++;
1252
1253         return ofs->numlowerfs;
1254 }
1255
1256 static int ovl_get_lower_layers(struct ovl_fs *ofs, struct path *stack,
1257                                 unsigned int numlower)
1258 {
1259         int err;
1260         unsigned int i;
1261
1262         err = -ENOMEM;
1263         ofs->lower_layers = kcalloc(numlower, sizeof(struct ovl_layer),
1264                                     GFP_KERNEL);
1265         if (ofs->lower_layers == NULL)
1266                 goto out;
1267
1268         ofs->lower_fs = kcalloc(numlower, sizeof(struct ovl_sb),
1269                                 GFP_KERNEL);
1270         if (ofs->lower_fs == NULL)
1271                 goto out;
1272
1273         for (i = 0; i < numlower; i++) {
1274                 struct vfsmount *mnt;
1275                 int fsid;
1276
1277                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1278                 if (err < 0)
1279                         goto out;
1280
1281                 mnt = clone_private_mount(&stack[i]);
1282                 err = PTR_ERR(mnt);
1283                 if (IS_ERR(mnt)) {
1284                         pr_err("overlayfs: failed to clone lowerpath\n");
1285                         goto out;
1286                 }
1287
1288                 /*
1289                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1290                  * will fail instead of modifying lower fs.
1291                  */
1292                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1293
1294                 ofs->lower_layers[ofs->numlower].mnt = mnt;
1295                 ofs->lower_layers[ofs->numlower].idx = i + 1;
1296                 ofs->lower_layers[ofs->numlower].fsid = fsid;
1297                 if (fsid) {
1298                         ofs->lower_layers[ofs->numlower].fs =
1299                                 &ofs->lower_fs[fsid - 1];
1300                 }
1301                 ofs->numlower++;
1302         }
1303
1304         /*
1305          * When all layers on same fs, overlay can use real inode numbers.
1306          * With mount option "xino=on", mounter declares that there are enough
1307          * free high bits in underlying fs to hold the unique fsid.
1308          * If overlayfs does encounter underlying inodes using the high xino
1309          * bits reserved for fsid, it emits a warning and uses the original
1310          * inode number.
1311          */
1312         if (!ofs->numlowerfs || (ofs->numlowerfs == 1 && !ofs->upper_mnt)) {
1313                 ofs->xino_bits = 0;
1314                 ofs->config.xino = OVL_XINO_OFF;
1315         } else if (ofs->config.xino == OVL_XINO_ON && !ofs->xino_bits) {
1316                 /*
1317                  * This is a roundup of number of bits needed for numlowerfs+1
1318                  * (i.e. ilog2(numlowerfs+1 - 1) + 1). fsid 0 is reserved for
1319                  * upper fs even with non upper overlay.
1320                  */
1321                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 31);
1322                 ofs->xino_bits = ilog2(ofs->numlowerfs) + 1;
1323         }
1324
1325         if (ofs->xino_bits) {
1326                 pr_info("overlayfs: \"xino\" feature enabled using %d upper inode bits.\n",
1327                         ofs->xino_bits);
1328         }
1329
1330         err = 0;
1331 out:
1332         return err;
1333 }
1334
1335 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1336                                             struct ovl_fs *ofs)
1337 {
1338         int err;
1339         char *lowertmp, *lower;
1340         struct path *stack = NULL;
1341         unsigned int stacklen, numlower = 0, i;
1342         bool remote = false;
1343         struct ovl_entry *oe;
1344
1345         err = -ENOMEM;
1346         lowertmp = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1347         if (!lowertmp)
1348                 goto out_err;
1349
1350         err = -EINVAL;
1351         stacklen = ovl_split_lowerdirs(lowertmp);
1352         if (stacklen > OVL_MAX_STACK) {
1353                 pr_err("overlayfs: too many lower directories, limit is %d\n",
1354                        OVL_MAX_STACK);
1355                 goto out_err;
1356         } else if (!ofs->config.upperdir && stacklen == 1) {
1357                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1358                 goto out_err;
1359         } else if (!ofs->config.upperdir && ofs->config.nfs_export &&
1360                    ofs->config.redirect_follow) {
1361                 pr_warn("overlayfs: NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
1362                 ofs->config.nfs_export = false;
1363         }
1364
1365         err = -ENOMEM;
1366         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1367         if (!stack)
1368                 goto out_err;
1369
1370         err = -EINVAL;
1371         lower = lowertmp;
1372         for (numlower = 0; numlower < stacklen; numlower++) {
1373                 err = ovl_lower_dir(lower, &stack[numlower], ofs,
1374                                     &sb->s_stack_depth, &remote);
1375                 if (err)
1376                         goto out_err;
1377
1378                 lower = strchr(lower, '\0') + 1;
1379         }
1380
1381         err = -EINVAL;
1382         sb->s_stack_depth++;
1383         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1384                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1385                 goto out_err;
1386         }
1387
1388         err = ovl_get_lower_layers(ofs, stack, numlower);
1389         if (err)
1390                 goto out_err;
1391
1392         err = -ENOMEM;
1393         oe = ovl_alloc_entry(numlower);
1394         if (!oe)
1395                 goto out_err;
1396
1397         for (i = 0; i < numlower; i++) {
1398                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1399                 oe->lowerstack[i].layer = &ofs->lower_layers[i];
1400         }
1401
1402         if (remote)
1403                 sb->s_d_op = &ovl_reval_dentry_operations;
1404         else
1405                 sb->s_d_op = &ovl_dentry_operations;
1406
1407 out:
1408         for (i = 0; i < numlower; i++)
1409                 path_put(&stack[i]);
1410         kfree(stack);
1411         kfree(lowertmp);
1412
1413         return oe;
1414
1415 out_err:
1416         oe = ERR_PTR(err);
1417         goto out;
1418 }
1419
1420 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1421 {
1422         struct path upperpath = { };
1423         struct dentry *root_dentry;
1424         struct ovl_entry *oe;
1425         struct ovl_fs *ofs;
1426         struct cred *cred;
1427         int err;
1428
1429         err = -ENOMEM;
1430         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1431         if (!ofs)
1432                 goto out;
1433
1434         ofs->creator_cred = cred = prepare_creds();
1435         if (!cred)
1436                 goto out_err;
1437
1438         ofs->config.index = ovl_index_def;
1439         ofs->config.nfs_export = ovl_nfs_export_def;
1440         ofs->config.xino = ovl_xino_def();
1441         ofs->config.metacopy = ovl_metacopy_def;
1442         err = ovl_parse_opt((char *) data, &ofs->config);
1443         if (err)
1444                 goto out_err;
1445
1446         err = -EINVAL;
1447         if (!ofs->config.lowerdir) {
1448                 if (!silent)
1449                         pr_err("overlayfs: missing 'lowerdir'\n");
1450                 goto out_err;
1451         }
1452
1453         sb->s_stack_depth = 0;
1454         sb->s_maxbytes = MAX_LFS_FILESIZE;
1455         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1456         if (ofs->config.xino != OVL_XINO_OFF)
1457                 ofs->xino_bits = BITS_PER_LONG - 32;
1458
1459         if (ofs->config.upperdir) {
1460                 if (!ofs->config.workdir) {
1461                         pr_err("overlayfs: missing 'workdir'\n");
1462                         goto out_err;
1463                 }
1464
1465                 err = ovl_get_upper(ofs, &upperpath);
1466                 if (err)
1467                         goto out_err;
1468
1469                 err = ovl_get_workdir(ofs, &upperpath);
1470                 if (err)
1471                         goto out_err;
1472
1473                 if (!ofs->workdir)
1474                         sb->s_flags |= SB_RDONLY;
1475
1476                 sb->s_stack_depth = ofs->upper_mnt->mnt_sb->s_stack_depth;
1477                 sb->s_time_gran = ofs->upper_mnt->mnt_sb->s_time_gran;
1478
1479         }
1480         oe = ovl_get_lowerstack(sb, ofs);
1481         err = PTR_ERR(oe);
1482         if (IS_ERR(oe))
1483                 goto out_err;
1484
1485         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1486         if (!ofs->upper_mnt)
1487                 sb->s_flags |= SB_RDONLY;
1488
1489         if (!(ovl_force_readonly(ofs)) && ofs->config.index) {
1490                 err = ovl_get_indexdir(ofs, oe, &upperpath);
1491                 if (err)
1492                         goto out_free_oe;
1493
1494                 /* Force r/o mount with no index dir */
1495                 if (!ofs->indexdir) {
1496                         dput(ofs->workdir);
1497                         ofs->workdir = NULL;
1498                         sb->s_flags |= SB_RDONLY;
1499                 }
1500
1501         }
1502
1503         /* Show index=off in /proc/mounts for forced r/o mount */
1504         if (!ofs->indexdir) {
1505                 ofs->config.index = false;
1506                 if (ofs->upper_mnt && ofs->config.nfs_export) {
1507                         pr_warn("overlayfs: NFS export requires an index dir, falling back to nfs_export=off.\n");
1508                         ofs->config.nfs_export = false;
1509                 }
1510         }
1511
1512         if (ofs->config.metacopy && ofs->config.nfs_export) {
1513                 pr_warn("overlayfs: NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1514                 ofs->config.nfs_export = false;
1515         }
1516
1517         if (ofs->config.nfs_export)
1518                 sb->s_export_op = &ovl_export_operations;
1519
1520         /* Never override disk quota limits or use reserved space */
1521         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1522
1523         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1524         sb->s_op = &ovl_super_operations;
1525         sb->s_xattr = ovl_xattr_handlers;
1526         sb->s_fs_info = ofs;
1527         sb->s_flags |= SB_POSIXACL;
1528
1529         err = -ENOMEM;
1530         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1531         if (!root_dentry)
1532                 goto out_free_oe;
1533
1534         root_dentry->d_fsdata = oe;
1535
1536         mntput(upperpath.mnt);
1537         if (upperpath.dentry) {
1538                 ovl_dentry_set_upper_alias(root_dentry);
1539                 if (ovl_is_impuredir(upperpath.dentry))
1540                         ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1541         }
1542
1543         /* Root is always merge -> can have whiteouts */
1544         ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1545         ovl_dentry_set_flag(OVL_E_CONNECTED, root_dentry);
1546         ovl_set_upperdata(d_inode(root_dentry));
1547         ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1548                        ovl_dentry_lower(root_dentry), NULL);
1549
1550         sb->s_root = root_dentry;
1551
1552         return 0;
1553
1554 out_free_oe:
1555         ovl_entry_stack_free(oe);
1556         kfree(oe);
1557 out_err:
1558         path_put(&upperpath);
1559         ovl_free_fs(ofs);
1560 out:
1561         return err;
1562 }
1563
1564 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1565                                 const char *dev_name, void *raw_data)
1566 {
1567         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1568 }
1569
1570 static struct file_system_type ovl_fs_type = {
1571         .owner          = THIS_MODULE,
1572         .name           = "overlay",
1573         .mount          = ovl_mount,
1574         .kill_sb        = kill_anon_super,
1575 };
1576 MODULE_ALIAS_FS("overlay");
1577
1578 static void ovl_inode_init_once(void *foo)
1579 {
1580         struct ovl_inode *oi = foo;
1581
1582         inode_init_once(&oi->vfs_inode);
1583 }
1584
1585 static int __init ovl_init(void)
1586 {
1587         int err;
1588
1589         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1590                                              sizeof(struct ovl_inode), 0,
1591                                              (SLAB_RECLAIM_ACCOUNT|
1592                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1593                                              ovl_inode_init_once);
1594         if (ovl_inode_cachep == NULL)
1595                 return -ENOMEM;
1596
1597         err = register_filesystem(&ovl_fs_type);
1598         if (err)
1599                 kmem_cache_destroy(ovl_inode_cachep);
1600
1601         return err;
1602 }
1603
1604 static void __exit ovl_exit(void)
1605 {
1606         unregister_filesystem(&ovl_fs_type);
1607
1608         /*
1609          * Make sure all delayed rcu free inodes are flushed before we
1610          * destroy cache.
1611          */
1612         rcu_barrier();
1613         kmem_cache_destroy(ovl_inode_cachep);
1614
1615 }
1616
1617 module_init(ovl_init);
1618 module_exit(ovl_exit);