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 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_i_callback(struct rcu_head *head)
 194 {
 195         struct inode *inode = container_of(head, struct inode, i_rcu);
 196
 197         kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
 198 }
 199
 200 static void ovl_destroy_inode(struct inode *inode)
 201 {
 202         struct ovl_inode *oi = OVL_I(inode);
 203
 204         dput(oi->__upperdentry);
 205         iput(oi->lower);
 206         if (S_ISDIR(inode->i_mode))
 207                 ovl_dir_cache_free(inode);
 208         else
 209                 iput(oi->lowerdata);
 210         kfree(oi->redirect);
 211         mutex_destroy(&oi->lock);
 212
 213         call_rcu(&inode->i_rcu, ovl_i_callback);
 214 }
 215
 216 static void ovl_free_fs(struct ovl_fs *ofs)
 217 {
 218         unsigned i;
 219
 220         dput(ofs->indexdir);
 221         dput(ofs->workdir);
 222         if (ofs->workdir_locked)
 223                 ovl_inuse_unlock(ofs->workbasedir);
 224         dput(ofs->workbasedir);
 225         if (ofs->upperdir_locked)
 226                 ovl_inuse_unlock(ofs->upper_mnt->mnt_root);
 227         mntput(ofs->upper_mnt);
 228         for (i = 0; i < ofs->numlower; i++)
 229                 mntput(ofs->lower_layers[i].mnt);
 230         for (i = 0; i < ofs->numlowerfs; i++)
 231                 free_anon_bdev(ofs->lower_fs[i].pseudo_dev);
 232         kfree(ofs->lower_layers);
 233         kfree(ofs->lower_fs);
 234
 235         kfree(ofs->config.lowerdir);
 236         kfree(ofs->config.upperdir);
 237         kfree(ofs->config.workdir);
 238         kfree(ofs->config.redirect_mode);
 239         if (ofs->creator_cred)
 240                 put_cred(ofs->creator_cred);
 241         kfree(ofs);
 242 }
 243
 244 static void ovl_put_super(struct super_block *sb)
 245 {
 246         struct ovl_fs *ofs = sb->s_fs_info;
 247
 248         ovl_free_fs(ofs);
 249 }
 250
 251 /* Sync real dirty inodes in upper filesystem (if it exists) */
 252 static int ovl_sync_fs(struct super_block *sb, int wait)
 253 {
 254         struct ovl_fs *ofs = sb->s_fs_info;
 255         struct super_block *upper_sb;
 256         int ret;
 257
 258         if (!ofs->upper_mnt)
 259                 return 0;
 260
 261         /*
 262          * If this is a sync(2) call or an emergency sync, all the super blocks
 263          * will be iterated, including upper_sb, so no need to do anything.
 264          *
 265          * If this is a syncfs(2) call, then we do need to call
 266          * sync_filesystem() on upper_sb, but enough if we do it when being
 267          * called with wait == 1.
 268          */
 269         if (!wait)
 270                 return 0;
 271
 272         upper_sb = ofs->upper_mnt->mnt_sb;
 273
 274         down_read(&upper_sb->s_umount);
 275         ret = sync_filesystem(upper_sb);
 276         up_read(&upper_sb->s_umount);
 277
 278         return ret;
 279 }
 280
 281 /**
 282  * ovl_statfs
 283  * @sb: The overlayfs super block
 284  * @buf: The struct kstatfs to fill in with stats
 285  *
 286  * Get the filesystem statistics.  As writes always target the upper layer
 287  * filesystem pass the statfs to the upper filesystem (if it exists)
 288  */
 289 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 290 {
 291         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 292         struct dentry *root_dentry = dentry->d_sb->s_root;
 293         struct path path;
 294         int err;
 295
 296         ovl_path_real(root_dentry, &path);
 297
 298         err = vfs_statfs(&path, buf);
 299         if (!err) {
 300                 buf->f_namelen = ofs->namelen;
 301                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 302         }
 303
 304         return err;
 305 }
 306
 307 /* Will this overlay be forced to mount/remount ro? */
 308 static bool ovl_force_readonly(struct ovl_fs *ofs)
 309 {
 310         return (!ofs->upper_mnt || !ofs->workdir);
 311 }
 312
 313 static const char *ovl_redirect_mode_def(void)
 314 {
 315         return ovl_redirect_dir_def ? "on" : "off";
 316 }
 317
 318 enum {
 319         OVL_XINO_OFF,
 320         OVL_XINO_AUTO,
 321         OVL_XINO_ON,
 322 };
 323
 324 static const char * const ovl_xino_str[] = {
 325         "off",
 326         "auto",
 327         "on",
 328 };
 329
 330 static inline int ovl_xino_def(void)
 331 {
 332         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 333 }
 334
 335 /**
 336  * ovl_show_options
 337  *
 338  * Prints the mount options for a given superblock.
 339  * Returns zero; does not fail.
 340  */
 341 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 342 {
 343         struct super_block *sb = dentry->d_sb;
 344         struct ovl_fs *ofs = sb->s_fs_info;
 345
 346         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 347         if (ofs->config.upperdir) {
 348                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 349                 seq_show_option(m, "workdir", ofs->config.workdir);
 350         }
 351         if (ofs->config.default_permissions)
 352                 seq_puts(m, ",default_permissions");
 353         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 354                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 355         if (ofs->config.index != ovl_index_def)
 356                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 357         if (ofs->config.nfs_export != ovl_nfs_export_def)
 358                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 359                                                 "on" : "off");
 360         if (ofs->config.xino != ovl_xino_def())
 361                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 362         if (ofs->config.metacopy != ovl_metacopy_def)
 363                 seq_printf(m, ",metacopy=%s",
 364                            ofs->config.metacopy ? "on" : "off");
 365         return 0;
 366 }
 367
 368 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 369 {
 370         struct ovl_fs *ofs = sb->s_fs_info;
 371
 372         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 373                 return -EROFS;
 374
 375         return 0;
 376 }
 377
 378 static const struct super_operations ovl_super_operations = {
 379         .alloc_inode    = ovl_alloc_inode,
 380         .destroy_inode  = ovl_destroy_inode,
 381         .drop_inode     = generic_delete_inode,
 382         .put_super      = ovl_put_super,
 383         .sync_fs        = ovl_sync_fs,
 384         .statfs         = ovl_statfs,
 385         .show_options   = ovl_show_options,
 386         .remount_fs     = ovl_remount,
 387 };
 388
 389 enum {
 390         OPT_LOWERDIR,
 391         OPT_UPPERDIR,
 392         OPT_WORKDIR,
 393         OPT_DEFAULT_PERMISSIONS,
 394         OPT_REDIRECT_DIR,
 395         OPT_INDEX_ON,
 396         OPT_INDEX_OFF,
 397         OPT_NFS_EXPORT_ON,
 398         OPT_NFS_EXPORT_OFF,
 399         OPT_XINO_ON,
 400         OPT_XINO_OFF,
 401         OPT_XINO_AUTO,
 402         OPT_METACOPY_ON,
 403         OPT_METACOPY_OFF,
 404         OPT_ERR,
 405 };
 406
 407 static const match_table_t ovl_tokens = {
 408         {OPT_LOWERDIR,                  "lowerdir=%s"},
 409         {OPT_UPPERDIR,                  "upperdir=%s"},
 410         {OPT_WORKDIR,                   "workdir=%s"},
 411         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 412         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 413         {OPT_INDEX_ON,                  "index=on"},
 414         {OPT_INDEX_OFF,                 "index=off"},
 415         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 416         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 417         {OPT_XINO_ON,                   "xino=on"},
 418         {OPT_XINO_OFF,                  "xino=off"},
 419         {OPT_XINO_AUTO,                 "xino=auto"},
 420         {OPT_METACOPY_ON,               "metacopy=on"},
 421         {OPT_METACOPY_OFF,              "metacopy=off"},
 422         {OPT_ERR,                       NULL}
 423 };
 424
 425 static char *ovl_next_opt(char **s)
 426 {
 427         char *sbegin = *s;
 428         char *p;
 429
 430         if (sbegin == NULL)
 431                 return NULL;
 432
 433         for (p = sbegin; *p; p++) {
 434                 if (*p == '\\') {
 435                         p++;
 436                         if (!*p)
 437                                 break;
 438                 } else if (*p == ',') {
 439                         *p = '\0';
 440                         *s = p + 1;
 441                         return sbegin;
 442                 }
 443         }
 444         *s = NULL;
 445         return sbegin;
 446 }
 447
 448 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 449 {
 450         if (strcmp(mode, "on") == 0) {
 451                 config->redirect_dir = true;
 452                 /*
 453                  * Does not make sense to have redirect creation without
 454                  * redirect following.
 455                  */
 456                 config->redirect_follow = true;
 457         } else if (strcmp(mode, "follow") == 0) {
 458                 config->redirect_follow = true;
 459         } else if (strcmp(mode, "off") == 0) {
 460                 if (ovl_redirect_always_follow)
 461                         config->redirect_follow = true;
 462         } else if (strcmp(mode, "nofollow") != 0) {
 463                 pr_err("overlayfs: bad mount option \"redirect_dir=%s\"\n",
 464                        mode);
 465                 return -EINVAL;
 466         }
 467
 468         return 0;
 469 }
 470
 471 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 472 {
 473         char *p;
 474         int err;
 475         bool metacopy_opt = false, redirect_opt = false;
 476
 477         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 478         if (!config->redirect_mode)
 479                 return -ENOMEM;
 480
 481         while ((p = ovl_next_opt(&opt)) != NULL) {
 482                 int token;
 483                 substring_t args[MAX_OPT_ARGS];
 484
 485                 if (!*p)
 486                         continue;
 487
 488                 token = match_token(p, ovl_tokens, args);
 489                 switch (token) {
 490                 case OPT_UPPERDIR:
 491                         kfree(config->upperdir);
 492                         config->upperdir = match_strdup(&args[0]);
 493                         if (!config->upperdir)
 494                                 return -ENOMEM;
 495                         break;
 496
 497                 case OPT_LOWERDIR:
 498                         kfree(config->lowerdir);
 499                         config->lowerdir = match_strdup(&args[0]);
 500                         if (!config->lowerdir)
 501                                 return -ENOMEM;
 502                         break;
 503
 504                 case OPT_WORKDIR:
 505                         kfree(config->workdir);
 506                         config->workdir = match_strdup(&args[0]);
 507                         if (!config->workdir)
 508                                 return -ENOMEM;
 509                         break;
 510
 511                 case OPT_DEFAULT_PERMISSIONS:
 512                         config->default_permissions = true;
 513                         break;
 514
 515                 case OPT_REDIRECT_DIR:
 516                         kfree(config->redirect_mode);
 517                         config->redirect_mode = match_strdup(&args[0]);
 518                         if (!config->redirect_mode)
 519                                 return -ENOMEM;
 520                         redirect_opt = true;
 521                         break;
 522
 523                 case OPT_INDEX_ON:
 524                         config->index = true;
 525                         break;
 526
 527                 case OPT_INDEX_OFF:
 528                         config->index = false;
 529                         break;
 530
 531                 case OPT_NFS_EXPORT_ON:
 532                         config->nfs_export = true;
 533                         break;
 534
 535                 case OPT_NFS_EXPORT_OFF:
 536                         config->nfs_export = false;
 537                         break;
 538
 539                 case OPT_XINO_ON:
 540                         config->xino = OVL_XINO_ON;
 541                         break;
 542
 543                 case OPT_XINO_OFF:
 544                         config->xino = OVL_XINO_OFF;
 545                         break;
 546
 547                 case OPT_XINO_AUTO:
 548                         config->xino = OVL_XINO_AUTO;
 549                         break;
 550
 551                 case OPT_METACOPY_ON:
 552                         config->metacopy = true;
 553                         metacopy_opt = true;
 554                         break;
 555
 556                 case OPT_METACOPY_OFF:
 557                         config->metacopy = false;
 558                         break;
 559
 560                 default:
 561                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 562                         return -EINVAL;
 563                 }
 564         }
 565
 566         /* Workdir is useless in non-upper mount */
 567         if (!config->upperdir && config->workdir) {
 568                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 569                         config->workdir);
 570                 kfree(config->workdir);
 571                 config->workdir = NULL;
 572         }
 573
 574         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 575         if (err)
 576                 return err;
 577
 578         /*
 579          * This is to make the logic below simpler.  It doesn't make any other
 580          * difference, since config->redirect_dir is only used for upper.
 581          */
 582         if (!config->upperdir && config->redirect_follow)
 583                 config->redirect_dir = true;
 584
 585         /* Resolve metacopy -> redirect_dir dependency */
 586         if (config->metacopy && !config->redirect_dir) {
 587                 if (metacopy_opt && redirect_opt) {
 588                         pr_err("overlayfs: conflicting options: metacopy=on,redirect_dir=%s\n",
 589                                config->redirect_mode);
 590                         return -EINVAL;
 591                 }
 592                 if (redirect_opt) {
 593                         /*
 594                          * There was an explicit redirect_dir=... that resulted
 595                          * in this conflict.
 596                          */
 597                         pr_info("overlayfs: disabling metacopy due to redirect_dir=%s\n",
 598                                 config->redirect_mode);
 599                         config->metacopy = false;
 600                 } else {
 601                         /* Automatically enable redirect otherwise. */
 602                         config->redirect_follow = config->redirect_dir = true;
 603                 }
 604         }
 605
 606         return 0;
 607 }
 608
 609 #define OVL_WORKDIR_NAME "work"
 610 #define OVL_INDEXDIR_NAME "index"
 611
 612 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 613                                          const char *name, bool persist)
 614 {
 615         struct inode *dir =  ofs->workbasedir->d_inode;
 616         struct vfsmount *mnt = ofs->upper_mnt;
 617         struct dentry *work;
 618         int err;
 619         bool retried = false;
 620         bool locked = false;
 621
 622         inode_lock_nested(dir, I_MUTEX_PARENT);
 623         locked = true;
 624
 625 retry:
 626         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
 627
 628         if (!IS_ERR(work)) {
 629                 struct iattr attr = {
 630                         .ia_valid = ATTR_MODE,
 631                         .ia_mode = S_IFDIR | 0,
 632                 };
 633
 634                 if (work->d_inode) {
 635                         err = -EEXIST;
 636                         if (retried)
 637                                 goto out_dput;
 638
 639                         if (persist)
 640                                 goto out_unlock;
 641
 642                         retried = true;
 643                         ovl_workdir_cleanup(dir, mnt, work, 0);
 644                         dput(work);
 645                         goto retry;
 646                 }
 647
 648                 work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
 649                 err = PTR_ERR(work);
 650                 if (IS_ERR(work))
 651                         goto out_err;
 652
 653                 /*
 654                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 655                  *
 656                  * a) success (there was a POSIX ACL xattr and was removed)
 657                  * b) -ENODATA (there was no POSIX ACL xattr)
 658                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 659                  *
 660                  * There are various other error values that could effectively
 661                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 662                  * if the xattr name is too long), but the set of filesystems
 663                  * allowed as upper are limited to "normal" ones, where checking
 664                  * for the above two errors is sufficient.
 665                  */
 666                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 667                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 668                         goto out_dput;
 669
 670                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 671                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 672                         goto out_dput;
 673
 674                 /* Clear any inherited mode bits */
 675                 inode_lock(work->d_inode);
 676                 err = notify_change(work, &attr, NULL);
 677                 inode_unlock(work->d_inode);
 678                 if (err)
 679                         goto out_dput;
 680         } else {
 681                 err = PTR_ERR(work);
 682                 goto out_err;
 683         }
 684 out_unlock:
 685         if (locked)
 686                 inode_unlock(dir);
 687
 688         return work;
 689
 690 out_dput:
 691         dput(work);
 692 out_err:
 693         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 694                 ofs->config.workdir, name, -err);
 695         work = NULL;
 696         goto out_unlock;
 697 }
 698
 699 static void ovl_unescape(char *s)
 700 {
 701         char *d = s;
 702
 703         for (;; s++, d++) {
 704                 if (*s == '\\')
 705                         s++;
 706                 *d = *s;
 707                 if (!*s)
 708                         break;
 709         }
 710 }
 711
 712 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 713 {
 714         int err = -EINVAL;
 715
 716         if (!*name) {
 717                 pr_err("overlayfs: empty lowerdir\n");
 718                 goto out;
 719         }
 720         err = kern_path(name, LOOKUP_FOLLOW, path);
 721         if (err) {
 722                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 723                 goto out;
 724         }
 725         err = -EINVAL;
 726         if (ovl_dentry_weird(path->dentry)) {
 727                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 728                 goto out_put;
 729         }
 730         if (!d_is_dir(path->dentry)) {
 731                 pr_err("overlayfs: '%s' not a directory\n", name);
 732                 goto out_put;
 733         }
 734         return 0;
 735
 736 out_put:
 737         path_put_init(path);
 738 out:
 739         return err;
 740 }
 741
 742 static int ovl_mount_dir(const char *name, struct path *path)
 743 {
 744         int err = -ENOMEM;
 745         char *tmp = kstrdup(name, GFP_KERNEL);
 746
 747         if (tmp) {
 748                 ovl_unescape(tmp);
 749                 err = ovl_mount_dir_noesc(tmp, path);
 750
 751                 if (!err)
 752                         if (ovl_dentry_remote(path->dentry)) {
 753                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 754                                        tmp);
 755                                 path_put_init(path);
 756                                 err = -EINVAL;
 757                         }
 758                 kfree(tmp);
 759         }
 760         return err;
 761 }
 762
 763 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 764                              const char *name)
 765 {
 766         struct kstatfs statfs;
 767         int err = vfs_statfs(path, &statfs);
 768
 769         if (err)
 770                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 771         else
 772                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 773
 774         return err;
 775 }
 776
 777 static int ovl_lower_dir(const char *name, struct path *path,
 778                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 779 {
 780         int fh_type;
 781         int err;
 782
 783         err = ovl_mount_dir_noesc(name, path);
 784         if (err)
 785                 goto out;
 786
 787         err = ovl_check_namelen(path, ofs, name);
 788         if (err)
 789                 goto out_put;
 790
 791         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 792
 793         if (ovl_dentry_remote(path->dentry))
 794                 *remote = true;
 795
 796         /*
 797          * The inodes index feature and NFS export need to encode and decode
 798          * file handles, so they require that all layers support them.
 799          */
 800         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 801         if ((ofs->config.nfs_export ||
 802              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 803                 ofs->config.index = false;
 804                 ofs->config.nfs_export = false;
 805                 pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 806                         name);
 807         }
 808
 809         /* Check if lower fs has 32bit inode numbers */
 810         if (fh_type != FILEID_INO32_GEN)
 811                 ofs->xino_bits = 0;
 812
 813         return 0;
 814
 815 out_put:
 816         path_put_init(path);
 817 out:
 818         return err;
 819 }
 820
 821 /* Workdir should not be subdir of upperdir and vice versa */
 822 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 823 {
 824         bool ok = false;
 825
 826         if (workdir != upperdir) {
 827                 ok = (lock_rename(workdir, upperdir) == NULL);
 828                 unlock_rename(workdir, upperdir);
 829         }
 830         return ok;
 831 }
 832
 833 static unsigned int ovl_split_lowerdirs(char *str)
 834 {
 835         unsigned int ctr = 1;
 836         char *s, *d;
 837
 838         for (s = d = str;; s++, d++) {
 839                 if (*s == '\\') {
 840                         s++;
 841                 } else if (*s == ':') {
 842                         *d = '\0';
 843                         ctr++;
 844                         continue;
 845                 }
 846                 *d = *s;
 847                 if (!*s)
 848                         break;
 849         }
 850         return ctr;
 851 }
 852
 853 static int __maybe_unused
 854 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 855                         struct dentry *dentry, struct inode *inode,
 856                         const char *name, void *buffer, size_t size)
 857 {
 858         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 859 }
 860
 861 static int __maybe_unused
 862 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 863                         struct dentry *dentry, struct inode *inode,
 864                         const char *name, const void *value,
 865                         size_t size, int flags)
 866 {
 867         struct dentry *workdir = ovl_workdir(dentry);
 868         struct inode *realinode = ovl_inode_real(inode);
 869         struct posix_acl *acl = NULL;
 870         int err;
 871
 872         /* Check that everything is OK before copy-up */
 873         if (value) {
 874                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 875                 if (IS_ERR(acl))
 876                         return PTR_ERR(acl);
 877         }
 878         err = -EOPNOTSUPP;
 879         if (!IS_POSIXACL(d_inode(workdir)))
 880                 goto out_acl_release;
 881         if (!realinode->i_op->set_acl)
 882                 goto out_acl_release;
 883         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 884                 err = acl ? -EACCES : 0;
 885                 goto out_acl_release;
 886         }
 887         err = -EPERM;
 888         if (!inode_owner_or_capable(inode))
 889                 goto out_acl_release;
 890
 891         posix_acl_release(acl);
 892
 893         /*
 894          * Check if sgid bit needs to be cleared (actual setacl operation will
 895          * be done with mounter's capabilities and so that won't do it for us).
 896          */
 897         if (unlikely(inode->i_mode & S_ISGID) &&
 898             handler->flags == ACL_TYPE_ACCESS &&
 899             !in_group_p(inode->i_gid) &&
 900             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 901                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 902
 903                 err = ovl_setattr(dentry, &iattr);
 904                 if (err)
 905                         return err;
 906         }
 907
 908         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
 909         if (!err)
 910                 ovl_copyattr(ovl_inode_real(inode), inode);
 911
 912         return err;
 913
 914 out_acl_release:
 915         posix_acl_release(acl);
 916         return err;
 917 }
 918
 919 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 920                              struct dentry *dentry, struct inode *inode,
 921                              const char *name, void *buffer, size_t size)
 922 {
 923         return -EOPNOTSUPP;
 924 }
 925
 926 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 927                              struct dentry *dentry, struct inode *inode,
 928                              const char *name, const void *value,
 929                              size_t size, int flags)
 930 {
 931         return -EOPNOTSUPP;
 932 }
 933
 934 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 935                                struct dentry *dentry, struct inode *inode,
 936                                const char *name, void *buffer, size_t size)
 937 {
 938         return ovl_xattr_get(dentry, inode, name, buffer, size);
 939 }
 940
 941 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 942                                struct dentry *dentry, struct inode *inode,
 943                                const char *name, const void *value,
 944                                size_t size, int flags)
 945 {
 946         return ovl_xattr_set(dentry, inode, name, value, size, flags);
 947 }
 948
 949 static const struct xattr_handler __maybe_unused
 950 ovl_posix_acl_access_xattr_handler = {
 951         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 952         .flags = ACL_TYPE_ACCESS,
 953         .get = ovl_posix_acl_xattr_get,
 954         .set = ovl_posix_acl_xattr_set,
 955 };
 956
 957 static const struct xattr_handler __maybe_unused
 958 ovl_posix_acl_default_xattr_handler = {
 959         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 960         .flags = ACL_TYPE_DEFAULT,
 961         .get = ovl_posix_acl_xattr_get,
 962         .set = ovl_posix_acl_xattr_set,
 963 };
 964
 965 static const struct xattr_handler ovl_own_xattr_handler = {
 966         .prefix = OVL_XATTR_PREFIX,
 967         .get = ovl_own_xattr_get,
 968         .set = ovl_own_xattr_set,
 969 };
 970
 971 static const struct xattr_handler ovl_other_xattr_handler = {
 972         .prefix = "", /* catch all */
 973         .get = ovl_other_xattr_get,
 974         .set = ovl_other_xattr_set,
 975 };
 976
 977 static const struct xattr_handler *ovl_xattr_handlers[] = {
 978 #ifdef CONFIG_FS_POSIX_ACL
 979         &ovl_posix_acl_access_xattr_handler,
 980         &ovl_posix_acl_default_xattr_handler,
 981 #endif
 982         &ovl_own_xattr_handler,
 983         &ovl_other_xattr_handler,
 984         NULL
 985 };
 986
 987 static int ovl_get_upper(struct ovl_fs *ofs, struct path *upperpath)
 988 {
 989         struct vfsmount *upper_mnt;
 990         int err;
 991
 992         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
 993         if (err)
 994                 goto out;
 995
 996         /* Upper fs should not be r/o */
 997         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
 998                 pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 999                 err = -EINVAL;
1000                 goto out;
1001         }
1002
1003         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1004         if (err)
1005                 goto out;
1006
1007         upper_mnt = clone_private_mount(upperpath);
1008         err = PTR_ERR(upper_mnt);
1009         if (IS_ERR(upper_mnt)) {
1010                 pr_err("overlayfs: failed to clone upperpath\n");
1011                 goto out;
1012         }
1013
1014         /* Don't inherit atime flags */
1015         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1016         ofs->upper_mnt = upper_mnt;
1017
1018         err = -EBUSY;
1019         if (ovl_inuse_trylock(ofs->upper_mnt->mnt_root)) {
1020                 ofs->upperdir_locked = true;
1021         } else if (ofs->config.index) {
1022                 pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n");
1023                 goto out;
1024         } else {
1025                 pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
1026         }
1027
1028         err = 0;
1029 out:
1030         return err;
1031 }
1032
1033 static int ovl_make_workdir(struct ovl_fs *ofs, struct path *workpath)
1034 {
1035         struct vfsmount *mnt = ofs->upper_mnt;
1036         struct dentry *temp;
1037         int fh_type;
1038         int err;
1039
1040         err = mnt_want_write(mnt);
1041         if (err)
1042                 return err;
1043
1044         ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1045         if (!ofs->workdir)
1046                 goto out;
1047
1048         /*
1049          * Upper should support d_type, else whiteouts are visible.  Given
1050          * workdir and upper are on same fs, we can do iterate_dir() on
1051          * workdir. This check requires successful creation of workdir in
1052          * previous step.
1053          */
1054         err = ovl_check_d_type_supported(workpath);
1055         if (err < 0)
1056                 goto out;
1057
1058         /*
1059          * We allowed this configuration and don't want to break users over
1060          * kernel upgrade. So warn instead of erroring out.
1061          */
1062         if (!err)
1063                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
1064
1065         /* Check if upper/work fs supports O_TMPFILE */
1066         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1067         ofs->tmpfile = !IS_ERR(temp);
1068         if (ofs->tmpfile)
1069                 dput(temp);
1070         else
1071                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
1072
1073         /*
1074          * Check if upper/work fs supports trusted.overlay.* xattr
1075          */
1076         err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
1077         if (err) {
1078                 ofs->noxattr = true;
1079                 ofs->config.index = false;
1080                 ofs->config.metacopy = false;
1081                 pr_warn("overlayfs: upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1082                 err = 0;
1083         } else {
1084                 vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
1085         }
1086
1087         /* Check if upper/work fs supports file handles */
1088         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1089         if (ofs->config.index && !fh_type) {
1090                 ofs->config.index = false;
1091                 pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1092         }
1093
1094         /* Check if upper fs has 32bit inode numbers */
1095         if (fh_type != FILEID_INO32_GEN)
1096                 ofs->xino_bits = 0;
1097
1098         /* NFS export of r/w mount depends on index */
1099         if (ofs->config.nfs_export && !ofs->config.index) {
1100                 pr_warn("overlayfs: NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1101                 ofs->config.nfs_export = false;
1102         }
1103 out:
1104         mnt_drop_write(mnt);
1105         return err;
1106 }
1107
1108 static int ovl_get_workdir(struct ovl_fs *ofs, struct path *upperpath)
1109 {
1110         int err;
1111         struct path workpath = { };
1112
1113         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1114         if (err)
1115                 goto out;
1116
1117         err = -EINVAL;
1118         if (upperpath->mnt != workpath.mnt) {
1119                 pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1120                 goto out;
1121         }
1122         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1123                 pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1124                 goto out;
1125         }
1126
1127         ofs->workbasedir = dget(workpath.dentry);
1128
1129         err = -EBUSY;
1130         if (ovl_inuse_trylock(ofs->workbasedir)) {
1131                 ofs->workdir_locked = true;
1132         } else if (ofs->config.index) {
1133                 pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n");
1134                 goto out;
1135         } else {
1136                 pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
1137         }
1138
1139         err = ovl_make_workdir(ofs, &workpath);
1140         if (err)
1141                 goto out;
1142
1143         err = 0;
1144 out:
1145         path_put(&workpath);
1146
1147         return err;
1148 }
1149
1150 static int ovl_get_indexdir(struct ovl_fs *ofs, struct ovl_entry *oe,
1151                             struct path *upperpath)
1152 {
1153         struct vfsmount *mnt = ofs->upper_mnt;
1154         int err;
1155
1156         err = mnt_want_write(mnt);
1157         if (err)
1158                 return err;
1159
1160         /* Verify lower root is upper root origin */
1161         err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
1162                                 true);
1163         if (err) {
1164                 pr_err("overlayfs: failed to verify upper root origin\n");
1165                 goto out;
1166         }
1167
1168         ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1169         if (ofs->indexdir) {
1170                 /*
1171                  * Verify upper root is exclusively associated with index dir.
1172                  * Older kernels stored upper fh in "trusted.overlay.origin"
1173                  * xattr. If that xattr exists, verify that it is a match to
1174                  * upper dir file handle. In any case, verify or set xattr
1175                  * "trusted.overlay.upper" to indicate that index may have
1176                  * directory entries.
1177                  */
1178                 if (ovl_check_origin_xattr(ofs->indexdir)) {
1179                         err = ovl_verify_set_fh(ofs->indexdir, OVL_XATTR_ORIGIN,
1180                                                 upperpath->dentry, true, false);
1181                         if (err)
1182                                 pr_err("overlayfs: failed to verify index dir 'origin' xattr\n");
1183                 }
1184                 err = ovl_verify_upper(ofs->indexdir, upperpath->dentry, true);
1185                 if (err)
1186                         pr_err("overlayfs: failed to verify index dir 'upper' xattr\n");
1187
1188                 /* Cleanup bad/stale/orphan index entries */
1189                 if (!err)
1190                         err = ovl_indexdir_cleanup(ofs);
1191         }
1192         if (err || !ofs->indexdir)
1193                 pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1194
1195 out:
1196         mnt_drop_write(mnt);
1197         return err;
1198 }
1199
1200 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1201 {
1202         unsigned int i;
1203
1204         if (!ofs->config.nfs_export && !(ofs->config.index && ofs->upper_mnt))
1205                 return true;
1206
1207         for (i = 0; i < ofs->numlowerfs; i++) {
1208                 /*
1209                  * We use uuid to associate an overlay lower file handle with a
1210                  * lower layer, so we can accept lower fs with null uuid as long
1211                  * as all lower layers with null uuid are on the same fs.
1212                  */
1213                 if (uuid_equal(&ofs->lower_fs[i].sb->s_uuid, uuid))
1214                         return false;
1215         }
1216         return true;
1217 }
1218
1219 /* Get a unique fsid for the layer */
1220 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1221 {
1222         struct super_block *sb = path->mnt->mnt_sb;
1223         unsigned int i;
1224         dev_t dev;
1225         int err;
1226
1227         /* fsid 0 is reserved for upper fs even with non upper overlay */
1228         if (ofs->upper_mnt && ofs->upper_mnt->mnt_sb == sb)
1229                 return 0;
1230
1231         for (i = 0; i < ofs->numlowerfs; i++) {
1232                 if (ofs->lower_fs[i].sb == sb)
1233                         return i + 1;
1234         }
1235
1236         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1237                 ofs->config.index = false;
1238                 ofs->config.nfs_export = false;
1239                 pr_warn("overlayfs: %s uuid detected in lower fs '%pd2', falling back to index=off,nfs_export=off.\n",
1240                         uuid_is_null(&sb->s_uuid) ? "null" : "conflicting",
1241                         path->dentry);
1242         }
1243
1244         err = get_anon_bdev(&dev);
1245         if (err) {
1246                 pr_err("overlayfs: failed to get anonymous bdev for lowerpath\n");
1247                 return err;
1248         }
1249
1250         ofs->lower_fs[ofs->numlowerfs].sb = sb;
1251         ofs->lower_fs[ofs->numlowerfs].pseudo_dev = dev;
1252         ofs->numlowerfs++;
1253
1254         return ofs->numlowerfs;
1255 }
1256
1257 static int ovl_get_lower_layers(struct ovl_fs *ofs, struct path *stack,
1258                                 unsigned int numlower)
1259 {
1260         int err;
1261         unsigned int i;
1262
1263         err = -ENOMEM;
1264         ofs->lower_layers = kcalloc(numlower, sizeof(struct ovl_layer),
1265                                     GFP_KERNEL);
1266         if (ofs->lower_layers == NULL)
1267                 goto out;
1268
1269         ofs->lower_fs = kcalloc(numlower, sizeof(struct ovl_sb),
1270                                 GFP_KERNEL);
1271         if (ofs->lower_fs == NULL)
1272                 goto out;
1273
1274         for (i = 0; i < numlower; i++) {
1275                 struct vfsmount *mnt;
1276                 int fsid;
1277
1278                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1279                 if (err < 0)
1280                         goto out;
1281
1282                 mnt = clone_private_mount(&stack[i]);
1283                 err = PTR_ERR(mnt);
1284                 if (IS_ERR(mnt)) {
1285                         pr_err("overlayfs: failed to clone lowerpath\n");
1286                         goto out;
1287                 }
1288
1289                 /*
1290                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1291                  * will fail instead of modifying lower fs.
1292                  */
1293                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1294
1295                 ofs->lower_layers[ofs->numlower].mnt = mnt;
1296                 ofs->lower_layers[ofs->numlower].idx = i + 1;
1297                 ofs->lower_layers[ofs->numlower].fsid = fsid;
1298                 if (fsid) {
1299                         ofs->lower_layers[ofs->numlower].fs =
1300                                 &ofs->lower_fs[fsid - 1];
1301                 }
1302                 ofs->numlower++;
1303         }
1304
1305         /*
1306          * When all layers on same fs, overlay can use real inode numbers.
1307          * With mount option "xino=on", mounter declares that there are enough
1308          * free high bits in underlying fs to hold the unique fsid.
1309          * If overlayfs does encounter underlying inodes using the high xino
1310          * bits reserved for fsid, it emits a warning and uses the original
1311          * inode number.
1312          */
1313         if (!ofs->numlowerfs || (ofs->numlowerfs == 1 && !ofs->upper_mnt)) {
1314                 ofs->xino_bits = 0;
1315                 ofs->config.xino = OVL_XINO_OFF;
1316         } else if (ofs->config.xino == OVL_XINO_ON && !ofs->xino_bits) {
1317                 /*
1318                  * This is a roundup of number of bits needed for numlowerfs+1
1319                  * (i.e. ilog2(numlowerfs+1 - 1) + 1). fsid 0 is reserved for
1320                  * upper fs even with non upper overlay.
1321                  */
1322                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 31);
1323                 ofs->xino_bits = ilog2(ofs->numlowerfs) + 1;
1324         }
1325
1326         if (ofs->xino_bits) {
1327                 pr_info("overlayfs: \"xino\" feature enabled using %d upper inode bits.\n",
1328                         ofs->xino_bits);
1329         }
1330
1331         err = 0;
1332 out:
1333         return err;
1334 }
1335
1336 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1337                                             struct ovl_fs *ofs)
1338 {
1339         int err;
1340         char *lowertmp, *lower;
1341         struct path *stack = NULL;
1342         unsigned int stacklen, numlower = 0, i;
1343         bool remote = false;
1344         struct ovl_entry *oe;
1345
1346         err = -ENOMEM;
1347         lowertmp = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1348         if (!lowertmp)
1349                 goto out_err;
1350
1351         err = -EINVAL;
1352         stacklen = ovl_split_lowerdirs(lowertmp);
1353         if (stacklen > OVL_MAX_STACK) {
1354                 pr_err("overlayfs: too many lower directories, limit is %d\n",
1355                        OVL_MAX_STACK);
1356                 goto out_err;
1357         } else if (!ofs->config.upperdir && stacklen == 1) {
1358                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1359                 goto out_err;
1360         } else if (!ofs->config.upperdir && ofs->config.nfs_export &&
1361                    ofs->config.redirect_follow) {
1362                 pr_warn("overlayfs: NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
1363                 ofs->config.nfs_export = false;
1364         }
1365
1366         err = -ENOMEM;
1367         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1368         if (!stack)
1369                 goto out_err;
1370
1371         err = -EINVAL;
1372         lower = lowertmp;
1373         for (numlower = 0; numlower < stacklen; numlower++) {
1374                 err = ovl_lower_dir(lower, &stack[numlower], ofs,
1375                                     &sb->s_stack_depth, &remote);
1376                 if (err)
1377                         goto out_err;
1378
1379                 lower = strchr(lower, '\0') + 1;
1380         }
1381
1382         err = -EINVAL;
1383         sb->s_stack_depth++;
1384         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1385                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1386                 goto out_err;
1387         }
1388
1389         err = ovl_get_lower_layers(ofs, stack, numlower);
1390         if (err)
1391                 goto out_err;
1392
1393         err = -ENOMEM;
1394         oe = ovl_alloc_entry(numlower);
1395         if (!oe)
1396                 goto out_err;
1397
1398         for (i = 0; i < numlower; i++) {
1399                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1400                 oe->lowerstack[i].layer = &ofs->lower_layers[i];
1401         }
1402
1403         if (remote)
1404                 sb->s_d_op = &ovl_reval_dentry_operations;
1405         else
1406                 sb->s_d_op = &ovl_dentry_operations;
1407
1408 out:
1409         for (i = 0; i < numlower; i++)
1410                 path_put(&stack[i]);
1411         kfree(stack);
1412         kfree(lowertmp);
1413
1414         return oe;
1415
1416 out_err:
1417         oe = ERR_PTR(err);
1418         goto out;
1419 }
1420
1421 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1422 {
1423         struct path upperpath = { };
1424         struct dentry *root_dentry;
1425         struct ovl_entry *oe;
1426         struct ovl_fs *ofs;
1427         struct cred *cred;
1428         int err;
1429
1430         err = -ENOMEM;
1431         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1432         if (!ofs)
1433                 goto out;
1434
1435         ofs->creator_cred = cred = prepare_creds();
1436         if (!cred)
1437                 goto out_err;
1438
1439         ofs->config.index = ovl_index_def;
1440         ofs->config.nfs_export = ovl_nfs_export_def;
1441         ofs->config.xino = ovl_xino_def();
1442         ofs->config.metacopy = ovl_metacopy_def;
1443         err = ovl_parse_opt((char *) data, &ofs->config);
1444         if (err)
1445                 goto out_err;
1446
1447         err = -EINVAL;
1448         if (!ofs->config.lowerdir) {
1449                 if (!silent)
1450                         pr_err("overlayfs: missing 'lowerdir'\n");
1451                 goto out_err;
1452         }
1453
1454         sb->s_stack_depth = 0;
1455         sb->s_maxbytes = MAX_LFS_FILESIZE;
1456         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1457         if (ofs->config.xino != OVL_XINO_OFF)
1458                 ofs->xino_bits = BITS_PER_LONG - 32;
1459
1460         if (ofs->config.upperdir) {
1461                 if (!ofs->config.workdir) {
1462                         pr_err("overlayfs: missing 'workdir'\n");
1463                         goto out_err;
1464                 }
1465
1466                 err = ovl_get_upper(ofs, &upperpath);
1467                 if (err)
1468                         goto out_err;
1469
1470                 err = ovl_get_workdir(ofs, &upperpath);
1471                 if (err)
1472                         goto out_err;
1473
1474                 if (!ofs->workdir)
1475                         sb->s_flags |= SB_RDONLY;
1476
1477                 sb->s_stack_depth = ofs->upper_mnt->mnt_sb->s_stack_depth;
1478                 sb->s_time_gran = ofs->upper_mnt->mnt_sb->s_time_gran;
1479
1480         }
1481         oe = ovl_get_lowerstack(sb, ofs);
1482         err = PTR_ERR(oe);
1483         if (IS_ERR(oe))
1484                 goto out_err;
1485
1486         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1487         if (!ofs->upper_mnt)
1488                 sb->s_flags |= SB_RDONLY;
1489
1490         if (!(ovl_force_readonly(ofs)) && ofs->config.index) {
1491                 err = ovl_get_indexdir(ofs, oe, &upperpath);
1492                 if (err)
1493                         goto out_free_oe;
1494
1495                 /* Force r/o mount with no index dir */
1496                 if (!ofs->indexdir) {
1497                         dput(ofs->workdir);
1498                         ofs->workdir = NULL;
1499                         sb->s_flags |= SB_RDONLY;
1500                 }
1501
1502         }
1503
1504         /* Show index=off in /proc/mounts for forced r/o mount */
1505         if (!ofs->indexdir) {
1506                 ofs->config.index = false;
1507                 if (ofs->upper_mnt && ofs->config.nfs_export) {
1508                         pr_warn("overlayfs: NFS export requires an index dir, falling back to nfs_export=off.\n");
1509                         ofs->config.nfs_export = false;
1510                 }
1511         }
1512
1513         if (ofs->config.metacopy && ofs->config.nfs_export) {
1514                 pr_warn("overlayfs: NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1515                 ofs->config.nfs_export = false;
1516         }
1517
1518         if (ofs->config.nfs_export)
1519                 sb->s_export_op = &ovl_export_operations;
1520
1521         /* Never override disk quota limits or use reserved space */
1522         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1523
1524         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1525         sb->s_op = &ovl_super_operations;
1526         sb->s_xattr = ovl_xattr_handlers;
1527         sb->s_fs_info = ofs;
1528         sb->s_flags |= SB_POSIXACL;
1529
1530         err = -ENOMEM;
1531         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1532         if (!root_dentry)
1533                 goto out_free_oe;
1534
1535         root_dentry->d_fsdata = oe;
1536
1537         mntput(upperpath.mnt);
1538         if (upperpath.dentry) {
1539                 ovl_dentry_set_upper_alias(root_dentry);
1540                 if (ovl_is_impuredir(upperpath.dentry))
1541                         ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1542         }
1543
1544         /* Root is always merge -> can have whiteouts */
1545         ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1546         ovl_dentry_set_flag(OVL_E_CONNECTED, root_dentry);
1547         ovl_set_upperdata(d_inode(root_dentry));
1548         ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1549                        ovl_dentry_lower(root_dentry), NULL);
1550
1551         sb->s_root = root_dentry;
1552
1553         return 0;
1554
1555 out_free_oe:
1556         ovl_entry_stack_free(oe);
1557         kfree(oe);
1558 out_err:
1559         path_put(&upperpath);
1560         ovl_free_fs(ofs);
1561 out:
1562         return err;
1563 }
1564
1565 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1566                                 const char *dev_name, void *raw_data)
1567 {
1568         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1569 }
1570
1571 static struct file_system_type ovl_fs_type = {
1572         .owner          = THIS_MODULE,
1573         .name           = "overlay",
1574         .mount          = ovl_mount,
1575         .kill_sb        = kill_anon_super,
1576 };
1577 MODULE_ALIAS_FS("overlay");
1578
1579 static void ovl_inode_init_once(void *foo)
1580 {
1581         struct ovl_inode *oi = foo;
1582
1583         inode_init_once(&oi->vfs_inode);
1584 }
1585
1586 static int __init ovl_init(void)
1587 {
1588         int err;
1589
1590         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1591                                              sizeof(struct ovl_inode), 0,
1592                                              (SLAB_RECLAIM_ACCOUNT|
1593                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1594                                              ovl_inode_init_once);
1595         if (ovl_inode_cachep == NULL)
1596                 return -ENOMEM;
1597
1598         err = register_filesystem(&ovl_fs_type);
1599         if (err)
1600                 kmem_cache_destroy(ovl_inode_cachep);
1601
1602         return err;
1603 }
1604
1605 static void __exit ovl_exit(void)
1606 {
1607         unregister_filesystem(&ovl_fs_type);
1608
1609         /*
1610          * Make sure all delayed rcu free inodes are flushed before we
1611          * destroy cache.
1612          */
1613         rcu_barrier();
1614         kmem_cache_destroy(ovl_inode_cachep);
1615
1616 }
1617
1618 module_init(ovl_init);
1619 module_exit(ovl_exit);