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 = NULL, *lower;
  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         lower = ovl_dentry_lowerdata(dentry);
 105         if (!lower)
 106                 goto bug;
 107         real = lower;
 108
 109         /* Handle recursion */
 110         real = d_real(real, inode);
 111
 112         if (!inode || inode == d_inode(real))
 113                 return real;
 114 bug:
 115         WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
 116              __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
 117              inode ? inode->i_ino : 0, real,
 118              real && d_inode(real) ? d_inode(real)->i_ino : 0);
 119         return dentry;
 120 }
 121
 122 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 123 {
 124         struct ovl_entry *oe = dentry->d_fsdata;
 125         unsigned int i;
 126         int ret = 1;
 127
 128         for (i = 0; i < oe->numlower; i++) {
 129                 struct dentry *d = oe->lowerstack[i].dentry;
 130
 131                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 132                         ret = d->d_op->d_revalidate(d, flags);
 133                         if (ret < 0)
 134                                 return ret;
 135                         if (!ret) {
 136                                 if (!(flags & LOOKUP_RCU))
 137                                         d_invalidate(d);
 138                                 return -ESTALE;
 139                         }
 140                 }
 141         }
 142         return 1;
 143 }
 144
 145 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 146 {
 147         struct ovl_entry *oe = dentry->d_fsdata;
 148         unsigned int i;
 149         int ret = 1;
 150
 151         for (i = 0; i < oe->numlower; i++) {
 152                 struct dentry *d = oe->lowerstack[i].dentry;
 153
 154                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 155                         ret = d->d_op->d_weak_revalidate(d, flags);
 156                         if (ret <= 0)
 157                                 break;
 158                 }
 159         }
 160         return ret;
 161 }
 162
 163 static const struct dentry_operations ovl_dentry_operations = {
 164         .d_release = ovl_dentry_release,
 165         .d_real = ovl_d_real,
 166 };
 167
 168 static const struct dentry_operations ovl_reval_dentry_operations = {
 169         .d_release = ovl_dentry_release,
 170         .d_real = ovl_d_real,
 171         .d_revalidate = ovl_dentry_revalidate,
 172         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 173 };
 174
 175 static struct kmem_cache *ovl_inode_cachep;
 176
 177 static struct inode *ovl_alloc_inode(struct super_block *sb)
 178 {
 179         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 180
 181         if (!oi)
 182                 return NULL;
 183
 184         oi->cache = NULL;
 185         oi->redirect = NULL;
 186         oi->version = 0;
 187         oi->flags = 0;
 188         oi->__upperdentry = NULL;
 189         oi->lower = NULL;
 190         oi->lowerdata = NULL;
 191         mutex_init(&oi->lock);
 192
 193         return &oi->vfs_inode;
 194 }
 195
 196 static void ovl_i_callback(struct rcu_head *head)
 197 {
 198         struct inode *inode = container_of(head, struct inode, i_rcu);
 199
 200         kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
 201 }
 202
 203 static void ovl_destroy_inode(struct inode *inode)
 204 {
 205         struct ovl_inode *oi = OVL_I(inode);
 206
 207         dput(oi->__upperdentry);
 208         iput(oi->lower);
 209         if (S_ISDIR(inode->i_mode))
 210                 ovl_dir_cache_free(inode);
 211         else
 212                 iput(oi->lowerdata);
 213         kfree(oi->redirect);
 214         mutex_destroy(&oi->lock);
 215
 216         call_rcu(&inode->i_rcu, ovl_i_callback);
 217 }
 218
 219 static void ovl_free_fs(struct ovl_fs *ofs)
 220 {
 221         unsigned i;
 222
 223         iput(ofs->workbasedir_trap);
 224         iput(ofs->indexdir_trap);
 225         iput(ofs->workdir_trap);
 226         iput(ofs->upperdir_trap);
 227         dput(ofs->indexdir);
 228         dput(ofs->workdir);
 229         if (ofs->workdir_locked)
 230                 ovl_inuse_unlock(ofs->workbasedir);
 231         dput(ofs->workbasedir);
 232         if (ofs->upperdir_locked)
 233                 ovl_inuse_unlock(ofs->upper_mnt->mnt_root);
 234         mntput(ofs->upper_mnt);
 235         for (i = 0; i < ofs->numlower; i++) {
 236                 iput(ofs->lower_layers[i].trap);
 237                 mntput(ofs->lower_layers[i].mnt);
 238         }
 239         for (i = 0; i < ofs->numlowerfs; i++)
 240                 free_anon_bdev(ofs->lower_fs[i].pseudo_dev);
 241         kfree(ofs->lower_layers);
 242         kfree(ofs->lower_fs);
 243
 244         kfree(ofs->config.lowerdir);
 245         kfree(ofs->config.upperdir);
 246         kfree(ofs->config.workdir);
 247         kfree(ofs->config.redirect_mode);
 248         if (ofs->creator_cred)
 249                 put_cred(ofs->creator_cred);
 250         kfree(ofs);
 251 }
 252
 253 static void ovl_put_super(struct super_block *sb)
 254 {
 255         struct ovl_fs *ofs = sb->s_fs_info;
 256
 257         ovl_free_fs(ofs);
 258 }
 259
 260 /* Sync real dirty inodes in upper filesystem (if it exists) */
 261 static int ovl_sync_fs(struct super_block *sb, int wait)
 262 {
 263         struct ovl_fs *ofs = sb->s_fs_info;
 264         struct super_block *upper_sb;
 265         int ret;
 266
 267         if (!ofs->upper_mnt)
 268                 return 0;
 269
 270         /*
 271          * If this is a sync(2) call or an emergency sync, all the super blocks
 272          * will be iterated, including upper_sb, so no need to do anything.
 273          *
 274          * If this is a syncfs(2) call, then we do need to call
 275          * sync_filesystem() on upper_sb, but enough if we do it when being
 276          * called with wait == 1.
 277          */
 278         if (!wait)
 279                 return 0;
 280
 281         upper_sb = ofs->upper_mnt->mnt_sb;
 282
 283         down_read(&upper_sb->s_umount);
 284         ret = sync_filesystem(upper_sb);
 285         up_read(&upper_sb->s_umount);
 286
 287         return ret;
 288 }
 289
 290 /**
 291  * ovl_statfs
 292  * @sb: The overlayfs super block
 293  * @buf: The struct kstatfs to fill in with stats
 294  *
 295  * Get the filesystem statistics.  As writes always target the upper layer
 296  * filesystem pass the statfs to the upper filesystem (if it exists)
 297  */
 298 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 299 {
 300         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 301         struct dentry *root_dentry = dentry->d_sb->s_root;
 302         struct path path;
 303         int err;
 304
 305         ovl_path_real(root_dentry, &path);
 306
 307         err = vfs_statfs(&path, buf);
 308         if (!err) {
 309                 buf->f_namelen = ofs->namelen;
 310                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 311         }
 312
 313         return err;
 314 }
 315
 316 /* Will this overlay be forced to mount/remount ro? */
 317 static bool ovl_force_readonly(struct ovl_fs *ofs)
 318 {
 319         return (!ofs->upper_mnt || !ofs->workdir);
 320 }
 321
 322 static const char *ovl_redirect_mode_def(void)
 323 {
 324         return ovl_redirect_dir_def ? "on" : "off";
 325 }
 326
 327 enum {
 328         OVL_XINO_OFF,
 329         OVL_XINO_AUTO,
 330         OVL_XINO_ON,
 331 };
 332
 333 static const char * const ovl_xino_str[] = {
 334         "off",
 335         "auto",
 336         "on",
 337 };
 338
 339 static inline int ovl_xino_def(void)
 340 {
 341         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 342 }
 343
 344 /**
 345  * ovl_show_options
 346  *
 347  * Prints the mount options for a given superblock.
 348  * Returns zero; does not fail.
 349  */
 350 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 351 {
 352         struct super_block *sb = dentry->d_sb;
 353         struct ovl_fs *ofs = sb->s_fs_info;
 354
 355         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 356         if (ofs->config.upperdir) {
 357                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 358                 seq_show_option(m, "workdir", ofs->config.workdir);
 359         }
 360         if (ofs->config.default_permissions)
 361                 seq_puts(m, ",default_permissions");
 362         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 363                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 364         if (ofs->config.index != ovl_index_def)
 365                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 366         if (ofs->config.nfs_export != ovl_nfs_export_def)
 367                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 368                                                 "on" : "off");
 369         if (ofs->config.xino != ovl_xino_def())
 370                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 371         if (ofs->config.metacopy != ovl_metacopy_def)
 372                 seq_printf(m, ",metacopy=%s",
 373                            ofs->config.metacopy ? "on" : "off");
 374         return 0;
 375 }
 376
 377 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 378 {
 379         struct ovl_fs *ofs = sb->s_fs_info;
 380
 381         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 382                 return -EROFS;
 383
 384         return 0;
 385 }
 386
 387 static const struct super_operations ovl_super_operations = {
 388         .alloc_inode    = ovl_alloc_inode,
 389         .destroy_inode  = ovl_destroy_inode,
 390         .drop_inode     = generic_delete_inode,
 391         .put_super      = ovl_put_super,
 392         .sync_fs        = ovl_sync_fs,
 393         .statfs         = ovl_statfs,
 394         .show_options   = ovl_show_options,
 395         .remount_fs     = ovl_remount,
 396 };
 397
 398 enum {
 399         OPT_LOWERDIR,
 400         OPT_UPPERDIR,
 401         OPT_WORKDIR,
 402         OPT_DEFAULT_PERMISSIONS,
 403         OPT_REDIRECT_DIR,
 404         OPT_INDEX_ON,
 405         OPT_INDEX_OFF,
 406         OPT_NFS_EXPORT_ON,
 407         OPT_NFS_EXPORT_OFF,
 408         OPT_XINO_ON,
 409         OPT_XINO_OFF,
 410         OPT_XINO_AUTO,
 411         OPT_METACOPY_ON,
 412         OPT_METACOPY_OFF,
 413         OPT_ERR,
 414 };
 415
 416 static const match_table_t ovl_tokens = {
 417         {OPT_LOWERDIR,                  "lowerdir=%s"},
 418         {OPT_UPPERDIR,                  "upperdir=%s"},
 419         {OPT_WORKDIR,                   "workdir=%s"},
 420         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 421         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 422         {OPT_INDEX_ON,                  "index=on"},
 423         {OPT_INDEX_OFF,                 "index=off"},
 424         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 425         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 426         {OPT_XINO_ON,                   "xino=on"},
 427         {OPT_XINO_OFF,                  "xino=off"},
 428         {OPT_XINO_AUTO,                 "xino=auto"},
 429         {OPT_METACOPY_ON,               "metacopy=on"},
 430         {OPT_METACOPY_OFF,              "metacopy=off"},
 431         {OPT_ERR,                       NULL}
 432 };
 433
 434 static char *ovl_next_opt(char **s)
 435 {
 436         char *sbegin = *s;
 437         char *p;
 438
 439         if (sbegin == NULL)
 440                 return NULL;
 441
 442         for (p = sbegin; *p; p++) {
 443                 if (*p == '\\') {
 444                         p++;
 445                         if (!*p)
 446                                 break;
 447                 } else if (*p == ',') {
 448                         *p = '\0';
 449                         *s = p + 1;
 450                         return sbegin;
 451                 }
 452         }
 453         *s = NULL;
 454         return sbegin;
 455 }
 456
 457 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 458 {
 459         if (strcmp(mode, "on") == 0) {
 460                 config->redirect_dir = true;
 461                 /*
 462                  * Does not make sense to have redirect creation without
 463                  * redirect following.
 464                  */
 465                 config->redirect_follow = true;
 466         } else if (strcmp(mode, "follow") == 0) {
 467                 config->redirect_follow = true;
 468         } else if (strcmp(mode, "off") == 0) {
 469                 if (ovl_redirect_always_follow)
 470                         config->redirect_follow = true;
 471         } else if (strcmp(mode, "nofollow") != 0) {
 472                 pr_err("overlayfs: bad mount option \"redirect_dir=%s\"\n",
 473                        mode);
 474                 return -EINVAL;
 475         }
 476
 477         return 0;
 478 }
 479
 480 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 481 {
 482         char *p;
 483         int err;
 484         bool metacopy_opt = false, redirect_opt = false;
 485
 486         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 487         if (!config->redirect_mode)
 488                 return -ENOMEM;
 489
 490         while ((p = ovl_next_opt(&opt)) != NULL) {
 491                 int token;
 492                 substring_t args[MAX_OPT_ARGS];
 493
 494                 if (!*p)
 495                         continue;
 496
 497                 token = match_token(p, ovl_tokens, args);
 498                 switch (token) {
 499                 case OPT_UPPERDIR:
 500                         kfree(config->upperdir);
 501                         config->upperdir = match_strdup(&args[0]);
 502                         if (!config->upperdir)
 503                                 return -ENOMEM;
 504                         break;
 505
 506                 case OPT_LOWERDIR:
 507                         kfree(config->lowerdir);
 508                         config->lowerdir = match_strdup(&args[0]);
 509                         if (!config->lowerdir)
 510                                 return -ENOMEM;
 511                         break;
 512
 513                 case OPT_WORKDIR:
 514                         kfree(config->workdir);
 515                         config->workdir = match_strdup(&args[0]);
 516                         if (!config->workdir)
 517                                 return -ENOMEM;
 518                         break;
 519
 520                 case OPT_DEFAULT_PERMISSIONS:
 521                         config->default_permissions = true;
 522                         break;
 523
 524                 case OPT_REDIRECT_DIR:
 525                         kfree(config->redirect_mode);
 526                         config->redirect_mode = match_strdup(&args[0]);
 527                         if (!config->redirect_mode)
 528                                 return -ENOMEM;
 529                         redirect_opt = true;
 530                         break;
 531
 532                 case OPT_INDEX_ON:
 533                         config->index = true;
 534                         break;
 535
 536                 case OPT_INDEX_OFF:
 537                         config->index = false;
 538                         break;
 539
 540                 case OPT_NFS_EXPORT_ON:
 541                         config->nfs_export = true;
 542                         break;
 543
 544                 case OPT_NFS_EXPORT_OFF:
 545                         config->nfs_export = false;
 546                         break;
 547
 548                 case OPT_XINO_ON:
 549                         config->xino = OVL_XINO_ON;
 550                         break;
 551
 552                 case OPT_XINO_OFF:
 553                         config->xino = OVL_XINO_OFF;
 554                         break;
 555
 556                 case OPT_XINO_AUTO:
 557                         config->xino = OVL_XINO_AUTO;
 558                         break;
 559
 560                 case OPT_METACOPY_ON:
 561                         config->metacopy = true;
 562                         metacopy_opt = true;
 563                         break;
 564
 565                 case OPT_METACOPY_OFF:
 566                         config->metacopy = false;
 567                         break;
 568
 569                 default:
 570                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 571                         return -EINVAL;
 572                 }
 573         }
 574
 575         /* Workdir is useless in non-upper mount */
 576         if (!config->upperdir && config->workdir) {
 577                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 578                         config->workdir);
 579                 kfree(config->workdir);
 580                 config->workdir = NULL;
 581         }
 582
 583         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 584         if (err)
 585                 return err;
 586
 587         /*
 588          * This is to make the logic below simpler.  It doesn't make any other
 589          * difference, since config->redirect_dir is only used for upper.
 590          */
 591         if (!config->upperdir && config->redirect_follow)
 592                 config->redirect_dir = true;
 593
 594         /* Resolve metacopy -> redirect_dir dependency */
 595         if (config->metacopy && !config->redirect_dir) {
 596                 if (metacopy_opt && redirect_opt) {
 597                         pr_err("overlayfs: conflicting options: metacopy=on,redirect_dir=%s\n",
 598                                config->redirect_mode);
 599                         return -EINVAL;
 600                 }
 601                 if (redirect_opt) {
 602                         /*
 603                          * There was an explicit redirect_dir=... that resulted
 604                          * in this conflict.
 605                          */
 606                         pr_info("overlayfs: disabling metacopy due to redirect_dir=%s\n",
 607                                 config->redirect_mode);
 608                         config->metacopy = false;
 609                 } else {
 610                         /* Automatically enable redirect otherwise. */
 611                         config->redirect_follow = config->redirect_dir = true;
 612                 }
 613         }
 614
 615         return 0;
 616 }
 617
 618 #define OVL_WORKDIR_NAME "work"
 619 #define OVL_INDEXDIR_NAME "index"
 620
 621 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 622                                          const char *name, bool persist)
 623 {
 624         struct inode *dir =  ofs->workbasedir->d_inode;
 625         struct vfsmount *mnt = ofs->upper_mnt;
 626         struct dentry *work;
 627         int err;
 628         bool retried = false;
 629         bool locked = false;
 630
 631         inode_lock_nested(dir, I_MUTEX_PARENT);
 632         locked = true;
 633
 634 retry:
 635         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
 636
 637         if (!IS_ERR(work)) {
 638                 struct iattr attr = {
 639                         .ia_valid = ATTR_MODE,
 640                         .ia_mode = S_IFDIR | 0,
 641                 };
 642
 643                 if (work->d_inode) {
 644                         err = -EEXIST;
 645                         if (retried)
 646                                 goto out_dput;
 647
 648                         if (persist)
 649                                 goto out_unlock;
 650
 651                         retried = true;
 652                         ovl_workdir_cleanup(dir, mnt, work, 0);
 653                         dput(work);
 654                         goto retry;
 655                 }
 656
 657                 err = ovl_mkdir_real(dir, &work, attr.ia_mode);
 658                 if (err)
 659                         goto out_dput;
 660
 661                 /* Weird filesystem returning with hashed negative (kernfs)? */
 662                 err = -EINVAL;
 663                 if (d_really_is_negative(work))
 664                         goto out_dput;
 665
 666                 /*
 667                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 668                  *
 669                  * a) success (there was a POSIX ACL xattr and was removed)
 670                  * b) -ENODATA (there was no POSIX ACL xattr)
 671                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 672                  *
 673                  * There are various other error values that could effectively
 674                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 675                  * if the xattr name is too long), but the set of filesystems
 676                  * allowed as upper are limited to "normal" ones, where checking
 677                  * for the above two errors is sufficient.
 678                  */
 679                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 680                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 681                         goto out_dput;
 682
 683                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 684                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 685                         goto out_dput;
 686
 687                 /* Clear any inherited mode bits */
 688                 inode_lock(work->d_inode);
 689                 err = notify_change(work, &attr, NULL);
 690                 inode_unlock(work->d_inode);
 691                 if (err)
 692                         goto out_dput;
 693         } else {
 694                 err = PTR_ERR(work);
 695                 goto out_err;
 696         }
 697 out_unlock:
 698         if (locked)
 699                 inode_unlock(dir);
 700
 701         return work;
 702
 703 out_dput:
 704         dput(work);
 705 out_err:
 706         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 707                 ofs->config.workdir, name, -err);
 708         work = NULL;
 709         goto out_unlock;
 710 }
 711
 712 static void ovl_unescape(char *s)
 713 {
 714         char *d = s;
 715
 716         for (;; s++, d++) {
 717                 if (*s == '\\')
 718                         s++;
 719                 *d = *s;
 720                 if (!*s)
 721                         break;
 722         }
 723 }
 724
 725 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 726 {
 727         int err = -EINVAL;
 728
 729         if (!*name) {
 730                 pr_err("overlayfs: empty lowerdir\n");
 731                 goto out;
 732         }
 733         err = kern_path(name, LOOKUP_FOLLOW, path);
 734         if (err) {
 735                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 736                 goto out;
 737         }
 738         err = -EINVAL;
 739         if (ovl_dentry_weird(path->dentry)) {
 740                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 741                 goto out_put;
 742         }
 743         if (!d_is_dir(path->dentry)) {
 744                 pr_err("overlayfs: '%s' not a directory\n", name);
 745                 goto out_put;
 746         }
 747         return 0;
 748
 749 out_put:
 750         path_put_init(path);
 751 out:
 752         return err;
 753 }
 754
 755 static int ovl_mount_dir(const char *name, struct path *path)
 756 {
 757         int err = -ENOMEM;
 758         char *tmp = kstrdup(name, GFP_KERNEL);
 759
 760         if (tmp) {
 761                 ovl_unescape(tmp);
 762                 err = ovl_mount_dir_noesc(tmp, path);
 763
 764                 if (!err)
 765                         if (ovl_dentry_remote(path->dentry)) {
 766                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 767                                        tmp);
 768                                 path_put_init(path);
 769                                 err = -EINVAL;
 770                         }
 771                 kfree(tmp);
 772         }
 773         return err;
 774 }
 775
 776 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 777                              const char *name)
 778 {
 779         struct kstatfs statfs;
 780         int err = vfs_statfs(path, &statfs);
 781
 782         if (err)
 783                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 784         else
 785                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 786
 787         return err;
 788 }
 789
 790 static int ovl_lower_dir(const char *name, struct path *path,
 791                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 792 {
 793         int fh_type;
 794         int err;
 795
 796         err = ovl_mount_dir_noesc(name, path);
 797         if (err)
 798                 goto out;
 799
 800         err = ovl_check_namelen(path, ofs, name);
 801         if (err)
 802                 goto out_put;
 803
 804         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 805
 806         if (ovl_dentry_remote(path->dentry))
 807                 *remote = true;
 808
 809         /*
 810          * The inodes index feature and NFS export need to encode and decode
 811          * file handles, so they require that all layers support them.
 812          */
 813         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 814         if ((ofs->config.nfs_export ||
 815              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 816                 ofs->config.index = false;
 817                 ofs->config.nfs_export = false;
 818                 pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 819                         name);
 820         }
 821
 822         /* Check if lower fs has 32bit inode numbers */
 823         if (fh_type != FILEID_INO32_GEN)
 824                 ofs->xino_bits = 0;
 825
 826         return 0;
 827
 828 out_put:
 829         path_put_init(path);
 830 out:
 831         return err;
 832 }
 833
 834 /* Workdir should not be subdir of upperdir and vice versa */
 835 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 836 {
 837         bool ok = false;
 838
 839         if (workdir != upperdir) {
 840                 ok = (lock_rename(workdir, upperdir) == NULL);
 841                 unlock_rename(workdir, upperdir);
 842         }
 843         return ok;
 844 }
 845
 846 static unsigned int ovl_split_lowerdirs(char *str)
 847 {
 848         unsigned int ctr = 1;
 849         char *s, *d;
 850
 851         for (s = d = str;; s++, d++) {
 852                 if (*s == '\\') {
 853                         s++;
 854                 } else if (*s == ':') {
 855                         *d = '\0';
 856                         ctr++;
 857                         continue;
 858                 }
 859                 *d = *s;
 860                 if (!*s)
 861                         break;
 862         }
 863         return ctr;
 864 }
 865
 866 static int __maybe_unused
 867 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 868                         struct dentry *dentry, struct inode *inode,
 869                         const char *name, void *buffer, size_t size)
 870 {
 871         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 872 }
 873
 874 static int __maybe_unused
 875 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 876                         struct dentry *dentry, struct inode *inode,
 877                         const char *name, const void *value,
 878                         size_t size, int flags)
 879 {
 880         struct dentry *workdir = ovl_workdir(dentry);
 881         struct inode *realinode = ovl_inode_real(inode);
 882         struct posix_acl *acl = NULL;
 883         int err;
 884
 885         /* Check that everything is OK before copy-up */
 886         if (value) {
 887                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 888                 if (IS_ERR(acl))
 889                         return PTR_ERR(acl);
 890         }
 891         err = -EOPNOTSUPP;
 892         if (!IS_POSIXACL(d_inode(workdir)))
 893                 goto out_acl_release;
 894         if (!realinode->i_op->set_acl)
 895                 goto out_acl_release;
 896         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 897                 err = acl ? -EACCES : 0;
 898                 goto out_acl_release;
 899         }
 900         err = -EPERM;
 901         if (!inode_owner_or_capable(inode))
 902                 goto out_acl_release;
 903
 904         posix_acl_release(acl);
 905
 906         /*
 907          * Check if sgid bit needs to be cleared (actual setacl operation will
 908          * be done with mounter's capabilities and so that won't do it for us).
 909          */
 910         if (unlikely(inode->i_mode & S_ISGID) &&
 911             handler->flags == ACL_TYPE_ACCESS &&
 912             !in_group_p(inode->i_gid) &&
 913             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 914                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 915
 916                 err = ovl_setattr(dentry, &iattr);
 917                 if (err)
 918                         return err;
 919         }
 920
 921         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
 922         if (!err)
 923                 ovl_copyattr(ovl_inode_real(inode), inode);
 924
 925         return err;
 926
 927 out_acl_release:
 928         posix_acl_release(acl);
 929         return err;
 930 }
 931
 932 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 933                              struct dentry *dentry, struct inode *inode,
 934                              const char *name, void *buffer, size_t size)
 935 {
 936         return -EOPNOTSUPP;
 937 }
 938
 939 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 940                              struct dentry *dentry, struct inode *inode,
 941                              const char *name, const void *value,
 942                              size_t size, int flags)
 943 {
 944         return -EOPNOTSUPP;
 945 }
 946
 947 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 948                                struct dentry *dentry, struct inode *inode,
 949                                const char *name, void *buffer, size_t size)
 950 {
 951         return ovl_xattr_get(dentry, inode, name, buffer, size);
 952 }
 953
 954 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 955                                struct dentry *dentry, struct inode *inode,
 956                                const char *name, const void *value,
 957                                size_t size, int flags)
 958 {
 959         return ovl_xattr_set(dentry, inode, name, value, size, flags);
 960 }
 961
 962 static const struct xattr_handler __maybe_unused
 963 ovl_posix_acl_access_xattr_handler = {
 964         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 965         .flags = ACL_TYPE_ACCESS,
 966         .get = ovl_posix_acl_xattr_get,
 967         .set = ovl_posix_acl_xattr_set,
 968 };
 969
 970 static const struct xattr_handler __maybe_unused
 971 ovl_posix_acl_default_xattr_handler = {
 972         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 973         .flags = ACL_TYPE_DEFAULT,
 974         .get = ovl_posix_acl_xattr_get,
 975         .set = ovl_posix_acl_xattr_set,
 976 };
 977
 978 static const struct xattr_handler ovl_own_xattr_handler = {
 979         .prefix = OVL_XATTR_PREFIX,
 980         .get = ovl_own_xattr_get,
 981         .set = ovl_own_xattr_set,
 982 };
 983
 984 static const struct xattr_handler ovl_other_xattr_handler = {
 985         .prefix = "", /* catch all */
 986         .get = ovl_other_xattr_get,
 987         .set = ovl_other_xattr_set,
 988 };
 989
 990 static const struct xattr_handler *ovl_xattr_handlers[] = {
 991 #ifdef CONFIG_FS_POSIX_ACL
 992         &ovl_posix_acl_access_xattr_handler,
 993         &ovl_posix_acl_default_xattr_handler,
 994 #endif
 995         &ovl_own_xattr_handler,
 996         &ovl_other_xattr_handler,
 997         NULL
 998 };
 999
1000 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1001                           struct inode **ptrap, const char *name)
1002 {
1003         struct inode *trap;
1004         int err;
1005
1006         trap = ovl_get_trap_inode(sb, dir);
1007         err = PTR_ERR_OR_ZERO(trap);
1008         if (err) {
1009                 if (err == -ELOOP)
1010                         pr_err("overlayfs: conflicting %s path\n", name);
1011                 return err;
1012         }
1013
1014         *ptrap = trap;
1015         return 0;
1016 }
1017
1018 /*
1019  * Determine how we treat concurrent use of upperdir/workdir based on the
1020  * index feature. This is papering over mount leaks of container runtimes,
1021  * for example, an old overlay mount is leaked and now its upperdir is
1022  * attempted to be used as a lower layer in a new overlay mount.
1023  */
1024 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1025 {
1026         if (ofs->config.index) {
1027                 pr_err("overlayfs: %s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1028                        name);
1029                 return -EBUSY;
1030         } else {
1031                 pr_warn("overlayfs: %s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1032                         name);
1033                 return 0;
1034         }
1035 }
1036
1037 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1038                          struct path *upperpath)
1039 {
1040         struct vfsmount *upper_mnt;
1041         int err;
1042
1043         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1044         if (err)
1045                 goto out;
1046
1047         /* Upper fs should not be r/o */
1048         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
1049                 pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
1050                 err = -EINVAL;
1051                 goto out;
1052         }
1053
1054         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1055         if (err)
1056                 goto out;
1057
1058         err = ovl_setup_trap(sb, upperpath->dentry, &ofs->upperdir_trap,
1059                              "upperdir");
1060         if (err)
1061                 goto out;
1062
1063         upper_mnt = clone_private_mount(upperpath);
1064         err = PTR_ERR(upper_mnt);
1065         if (IS_ERR(upper_mnt)) {
1066                 pr_err("overlayfs: failed to clone upperpath\n");
1067                 goto out;
1068         }
1069
1070         /* Don't inherit atime flags */
1071         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1072         ofs->upper_mnt = upper_mnt;
1073
1074         if (ovl_inuse_trylock(ofs->upper_mnt->mnt_root)) {
1075                 ofs->upperdir_locked = true;
1076         } else {
1077                 err = ovl_report_in_use(ofs, "upperdir");
1078                 if (err)
1079                         goto out;
1080         }
1081
1082         err = 0;
1083 out:
1084         return err;
1085 }
1086
1087 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1088                             struct path *workpath)
1089 {
1090         struct vfsmount *mnt = ofs->upper_mnt;
1091         struct dentry *temp;
1092         int fh_type;
1093         int err;
1094
1095         err = mnt_want_write(mnt);
1096         if (err)
1097                 return err;
1098
1099         ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1100         if (!ofs->workdir)
1101                 goto out;
1102
1103         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1104         if (err)
1105                 goto out;
1106
1107         /*
1108          * Upper should support d_type, else whiteouts are visible.  Given
1109          * workdir and upper are on same fs, we can do iterate_dir() on
1110          * workdir. This check requires successful creation of workdir in
1111          * previous step.
1112          */
1113         err = ovl_check_d_type_supported(workpath);
1114         if (err < 0)
1115                 goto out;
1116
1117         /*
1118          * We allowed this configuration and don't want to break users over
1119          * kernel upgrade. So warn instead of erroring out.
1120          */
1121         if (!err)
1122                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
1123
1124         /* Check if upper/work fs supports O_TMPFILE */
1125         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1126         ofs->tmpfile = !IS_ERR(temp);
1127         if (ofs->tmpfile)
1128                 dput(temp);
1129         else
1130                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
1131
1132         /*
1133          * Check if upper/work fs supports trusted.overlay.* xattr
1134          */
1135         err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
1136         if (err) {
1137                 ofs->noxattr = true;
1138                 ofs->config.index = false;
1139                 ofs->config.metacopy = false;
1140                 pr_warn("overlayfs: upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1141                 err = 0;
1142         } else {
1143                 vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
1144         }
1145
1146         /* Check if upper/work fs supports file handles */
1147         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1148         if (ofs->config.index && !fh_type) {
1149                 ofs->config.index = false;
1150                 pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1151         }
1152
1153         /* Check if upper fs has 32bit inode numbers */
1154         if (fh_type != FILEID_INO32_GEN)
1155                 ofs->xino_bits = 0;
1156
1157         /* NFS export of r/w mount depends on index */
1158         if (ofs->config.nfs_export && !ofs->config.index) {
1159                 pr_warn("overlayfs: NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1160                 ofs->config.nfs_export = false;
1161         }
1162 out:
1163         mnt_drop_write(mnt);
1164         return err;
1165 }
1166
1167 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1168                            struct path *upperpath)
1169 {
1170         int err;
1171         struct path workpath = { };
1172
1173         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1174         if (err)
1175                 goto out;
1176
1177         err = -EINVAL;
1178         if (upperpath->mnt != workpath.mnt) {
1179                 pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1180                 goto out;
1181         }
1182         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1183                 pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1184                 goto out;
1185         }
1186
1187         ofs->workbasedir = dget(workpath.dentry);
1188
1189         if (ovl_inuse_trylock(ofs->workbasedir)) {
1190                 ofs->workdir_locked = true;
1191         } else {
1192                 err = ovl_report_in_use(ofs, "workdir");
1193                 if (err)
1194                         goto out;
1195         }
1196
1197         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1198                              "workdir");
1199         if (err)
1200                 goto out;
1201
1202         err = ovl_make_workdir(sb, ofs, &workpath);
1203
1204 out:
1205         path_put(&workpath);
1206
1207         return err;
1208 }
1209
1210 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1211                             struct ovl_entry *oe, struct path *upperpath)
1212 {
1213         struct vfsmount *mnt = ofs->upper_mnt;
1214         int err;
1215
1216         err = mnt_want_write(mnt);
1217         if (err)
1218                 return err;
1219
1220         /* Verify lower root is upper root origin */
1221         err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
1222                                 true);
1223         if (err) {
1224                 pr_err("overlayfs: failed to verify upper root origin\n");
1225                 goto out;
1226         }
1227
1228         ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1229         if (ofs->indexdir) {
1230                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1231                                      "indexdir");
1232                 if (err)
1233                         goto out;
1234
1235                 /*
1236                  * Verify upper root is exclusively associated with index dir.
1237                  * Older kernels stored upper fh in "trusted.overlay.origin"
1238                  * xattr. If that xattr exists, verify that it is a match to
1239                  * upper dir file handle. In any case, verify or set xattr
1240                  * "trusted.overlay.upper" to indicate that index may have
1241                  * directory entries.
1242                  */
1243                 if (ovl_check_origin_xattr(ofs->indexdir)) {
1244                         err = ovl_verify_set_fh(ofs->indexdir, OVL_XATTR_ORIGIN,
1245                                                 upperpath->dentry, true, false);
1246                         if (err)
1247                                 pr_err("overlayfs: failed to verify index dir 'origin' xattr\n");
1248                 }
1249                 err = ovl_verify_upper(ofs->indexdir, upperpath->dentry, true);
1250                 if (err)
1251                         pr_err("overlayfs: failed to verify index dir 'upper' xattr\n");
1252
1253                 /* Cleanup bad/stale/orphan index entries */
1254                 if (!err)
1255                         err = ovl_indexdir_cleanup(ofs);
1256         }
1257         if (err || !ofs->indexdir)
1258                 pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1259
1260 out:
1261         mnt_drop_write(mnt);
1262         return err;
1263 }
1264
1265 /* Get a unique fsid for the layer */
1266 static int ovl_get_fsid(struct ovl_fs *ofs, struct super_block *sb)
1267 {
1268         unsigned int i;
1269         dev_t dev;
1270         int err;
1271
1272         /* fsid 0 is reserved for upper fs even with non upper overlay */
1273         if (ofs->upper_mnt && ofs->upper_mnt->mnt_sb == sb)
1274                 return 0;
1275
1276         for (i = 0; i < ofs->numlowerfs; i++) {
1277                 if (ofs->lower_fs[i].sb == sb)
1278                         return i + 1;
1279         }
1280
1281         err = get_anon_bdev(&dev);
1282         if (err) {
1283                 pr_err("overlayfs: failed to get anonymous bdev for lowerpath\n");
1284                 return err;
1285         }
1286
1287         ofs->lower_fs[ofs->numlowerfs].sb = sb;
1288         ofs->lower_fs[ofs->numlowerfs].pseudo_dev = dev;
1289         ofs->numlowerfs++;
1290
1291         return ofs->numlowerfs;
1292 }
1293
1294 static int ovl_get_lower_layers(struct super_block *sb, struct ovl_fs *ofs,
1295                                 struct path *stack, unsigned int numlower)
1296 {
1297         int err;
1298         unsigned int i;
1299
1300         err = -ENOMEM;
1301         ofs->lower_layers = kcalloc(numlower, sizeof(struct ovl_layer),
1302                                     GFP_KERNEL);
1303         if (ofs->lower_layers == NULL)
1304                 goto out;
1305
1306         ofs->lower_fs = kcalloc(numlower, sizeof(struct ovl_sb),
1307                                 GFP_KERNEL);
1308         if (ofs->lower_fs == NULL)
1309                 goto out;
1310
1311         for (i = 0; i < numlower; i++) {
1312                 struct vfsmount *mnt;
1313                 struct inode *trap;
1314                 int fsid;
1315
1316                 err = fsid = ovl_get_fsid(ofs, stack[i].mnt->mnt_sb);
1317                 if (err < 0)
1318                         goto out;
1319
1320                 /*
1321                  * Check if lower root conflicts with this overlay layers before
1322                  * checking if it is in-use as upperdir/workdir of "another"
1323                  * mount, because we do not bother to check in ovl_is_inuse() if
1324                  * the upperdir/workdir is in fact in-use by our
1325                  * upperdir/workdir.
1326                  */
1327                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1328                 if (err)
1329                         goto out;
1330
1331                 if (ovl_is_inuse(stack[i].dentry)) {
1332                         err = ovl_report_in_use(ofs, "lowerdir");
1333                         if (err) {
1334                                 iput(trap);
1335                                 goto out;
1336                         }
1337                 }
1338
1339                 mnt = clone_private_mount(&stack[i]);
1340                 err = PTR_ERR(mnt);
1341                 if (IS_ERR(mnt)) {
1342                         pr_err("overlayfs: failed to clone lowerpath\n");
1343                         iput(trap);
1344                         goto out;
1345                 }
1346
1347                 /*
1348                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1349                  * will fail instead of modifying lower fs.
1350                  */
1351                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1352
1353                 ofs->lower_layers[ofs->numlower].trap = trap;
1354                 ofs->lower_layers[ofs->numlower].mnt = mnt;
1355                 ofs->lower_layers[ofs->numlower].idx = i + 1;
1356                 ofs->lower_layers[ofs->numlower].fsid = fsid;
1357                 if (fsid) {
1358                         ofs->lower_layers[ofs->numlower].fs =
1359                                 &ofs->lower_fs[fsid - 1];
1360                 }
1361                 ofs->numlower++;
1362         }
1363
1364         /*
1365          * When all layers on same fs, overlay can use real inode numbers.
1366          * With mount option "xino=on", mounter declares that there are enough
1367          * free high bits in underlying fs to hold the unique fsid.
1368          * If overlayfs does encounter underlying inodes using the high xino
1369          * bits reserved for fsid, it emits a warning and uses the original
1370          * inode number.
1371          */
1372         if (!ofs->numlowerfs || (ofs->numlowerfs == 1 && !ofs->upper_mnt)) {
1373                 ofs->xino_bits = 0;
1374                 ofs->config.xino = OVL_XINO_OFF;
1375         } else if (ofs->config.xino == OVL_XINO_ON && !ofs->xino_bits) {
1376                 /*
1377                  * This is a roundup of number of bits needed for numlowerfs+1
1378                  * (i.e. ilog2(numlowerfs+1 - 1) + 1). fsid 0 is reserved for
1379                  * upper fs even with non upper overlay.
1380                  */
1381                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 31);
1382                 ofs->xino_bits = ilog2(ofs->numlowerfs) + 1;
1383         }
1384
1385         if (ofs->xino_bits) {
1386                 pr_info("overlayfs: \"xino\" feature enabled using %d upper inode bits.\n",
1387                         ofs->xino_bits);
1388         }
1389
1390         err = 0;
1391 out:
1392         return err;
1393 }
1394
1395 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1396                                             struct ovl_fs *ofs)
1397 {
1398         int err;
1399         char *lowertmp, *lower;
1400         struct path *stack = NULL;
1401         unsigned int stacklen, numlower = 0, i;
1402         bool remote = false;
1403         struct ovl_entry *oe;
1404
1405         err = -ENOMEM;
1406         lowertmp = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1407         if (!lowertmp)
1408                 goto out_err;
1409
1410         err = -EINVAL;
1411         stacklen = ovl_split_lowerdirs(lowertmp);
1412         if (stacklen > OVL_MAX_STACK) {
1413                 pr_err("overlayfs: too many lower directories, limit is %d\n",
1414                        OVL_MAX_STACK);
1415                 goto out_err;
1416         } else if (!ofs->config.upperdir && stacklen == 1) {
1417                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1418                 goto out_err;
1419         } else if (!ofs->config.upperdir && ofs->config.nfs_export &&
1420                    ofs->config.redirect_follow) {
1421                 pr_warn("overlayfs: NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
1422                 ofs->config.nfs_export = false;
1423         }
1424
1425         err = -ENOMEM;
1426         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1427         if (!stack)
1428                 goto out_err;
1429
1430         err = -EINVAL;
1431         lower = lowertmp;
1432         for (numlower = 0; numlower < stacklen; numlower++) {
1433                 err = ovl_lower_dir(lower, &stack[numlower], ofs,
1434                                     &sb->s_stack_depth, &remote);
1435                 if (err)
1436                         goto out_err;
1437
1438                 lower = strchr(lower, '\0') + 1;
1439         }
1440
1441         err = -EINVAL;
1442         sb->s_stack_depth++;
1443         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1444                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1445                 goto out_err;
1446         }
1447
1448         err = ovl_get_lower_layers(sb, ofs, stack, numlower);
1449         if (err)
1450                 goto out_err;
1451
1452         err = -ENOMEM;
1453         oe = ovl_alloc_entry(numlower);
1454         if (!oe)
1455                 goto out_err;
1456
1457         for (i = 0; i < numlower; i++) {
1458                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1459                 oe->lowerstack[i].layer = &ofs->lower_layers[i];
1460         }
1461
1462         if (remote)
1463                 sb->s_d_op = &ovl_reval_dentry_operations;
1464         else
1465                 sb->s_d_op = &ovl_dentry_operations;
1466
1467 out:
1468         for (i = 0; i < numlower; i++)
1469                 path_put(&stack[i]);
1470         kfree(stack);
1471         kfree(lowertmp);
1472
1473         return oe;
1474
1475 out_err:
1476         oe = ERR_PTR(err);
1477         goto out;
1478 }
1479
1480 /*
1481  * Check if this layer root is a descendant of:
1482  * - another layer of this overlayfs instance
1483  * - upper/work dir of any overlayfs instance
1484  */
1485 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1486                            struct dentry *dentry, const char *name,
1487                            bool is_lower)
1488 {
1489         struct dentry *next = dentry, *parent;
1490         int err = 0;
1491
1492         if (!dentry)
1493                 return 0;
1494
1495         parent = dget_parent(next);
1496
1497         /* Walk back ancestors to root (inclusive) looking for traps */
1498         while (!err && parent != next) {
1499                 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1500                         err = -ELOOP;
1501                         pr_err("overlayfs: overlapping %s path\n", name);
1502                 } else if (ovl_is_inuse(parent)) {
1503                         err = ovl_report_in_use(ofs, name);
1504                 }
1505                 next = parent;
1506                 parent = dget_parent(next);
1507                 dput(next);
1508         }
1509
1510         dput(parent);
1511
1512         return err;
1513 }
1514
1515 /*
1516  * Check if any of the layers or work dirs overlap.
1517  */
1518 static int ovl_check_overlapping_layers(struct super_block *sb,
1519                                         struct ovl_fs *ofs)
1520 {
1521         int i, err;
1522
1523         if (ofs->upper_mnt) {
1524                 err = ovl_check_layer(sb, ofs, ofs->upper_mnt->mnt_root,
1525                                       "upperdir", false);
1526                 if (err)
1527                         return err;
1528
1529                 /*
1530                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1531                  * this instance and covers overlapping work and index dirs,
1532                  * unless work or index dir have been moved since created inside
1533                  * workbasedir.  In that case, we already have their traps in
1534                  * inode cache and we will catch that case on lookup.
1535                  */
1536                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1537                                       false);
1538                 if (err)
1539                         return err;
1540         }
1541
1542         for (i = 0; i < ofs->numlower; i++) {
1543                 err = ovl_check_layer(sb, ofs,
1544                                       ofs->lower_layers[i].mnt->mnt_root,
1545                                       "lowerdir", true);
1546                 if (err)
1547                         return err;
1548         }
1549
1550         return 0;
1551 }
1552
1553 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1554 {
1555         struct path upperpath = { };
1556         struct dentry *root_dentry;
1557         struct ovl_entry *oe;
1558         struct ovl_fs *ofs;
1559         struct cred *cred;
1560         int err;
1561
1562         err = -ENOMEM;
1563         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1564         if (!ofs)
1565                 goto out;
1566
1567         ofs->creator_cred = cred = prepare_creds();
1568         if (!cred)
1569                 goto out_err;
1570
1571         ofs->config.index = ovl_index_def;
1572         ofs->config.nfs_export = ovl_nfs_export_def;
1573         ofs->config.xino = ovl_xino_def();
1574         ofs->config.metacopy = ovl_metacopy_def;
1575         err = ovl_parse_opt((char *) data, &ofs->config);
1576         if (err)
1577                 goto out_err;
1578
1579         err = -EINVAL;
1580         if (!ofs->config.lowerdir) {
1581                 if (!silent)
1582                         pr_err("overlayfs: missing 'lowerdir'\n");
1583                 goto out_err;
1584         }
1585
1586         sb->s_stack_depth = 0;
1587         sb->s_maxbytes = MAX_LFS_FILESIZE;
1588         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1589         if (ofs->config.xino != OVL_XINO_OFF)
1590                 ofs->xino_bits = BITS_PER_LONG - 32;
1591
1592         /* alloc/destroy_inode needed for setting up traps in inode cache */
1593         sb->s_op = &ovl_super_operations;
1594
1595         if (ofs->config.upperdir) {
1596                 if (!ofs->config.workdir) {
1597                         pr_err("overlayfs: missing 'workdir'\n");
1598                         goto out_err;
1599                 }
1600
1601                 err = ovl_get_upper(sb, ofs, &upperpath);
1602                 if (err)
1603                         goto out_err;
1604
1605                 err = ovl_get_workdir(sb, ofs, &upperpath);
1606                 if (err)
1607                         goto out_err;
1608
1609                 if (!ofs->workdir)
1610                         sb->s_flags |= SB_RDONLY;
1611
1612                 sb->s_stack_depth = ofs->upper_mnt->mnt_sb->s_stack_depth;
1613                 sb->s_time_gran = ofs->upper_mnt->mnt_sb->s_time_gran;
1614
1615         }
1616         oe = ovl_get_lowerstack(sb, ofs);
1617         err = PTR_ERR(oe);
1618         if (IS_ERR(oe))
1619                 goto out_err;
1620
1621         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1622         if (!ofs->upper_mnt)
1623                 sb->s_flags |= SB_RDONLY;
1624
1625         if (!(ovl_force_readonly(ofs)) && ofs->config.index) {
1626                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
1627                 if (err)
1628                         goto out_free_oe;
1629
1630                 /* Force r/o mount with no index dir */
1631                 if (!ofs->indexdir) {
1632                         dput(ofs->workdir);
1633                         ofs->workdir = NULL;
1634                         sb->s_flags |= SB_RDONLY;
1635                 }
1636
1637         }
1638
1639         err = ovl_check_overlapping_layers(sb, ofs);
1640         if (err)
1641                 goto out_free_oe;
1642
1643         /* Show index=off in /proc/mounts for forced r/o mount */
1644         if (!ofs->indexdir) {
1645                 ofs->config.index = false;
1646                 if (ofs->upper_mnt && ofs->config.nfs_export) {
1647                         pr_warn("overlayfs: NFS export requires an index dir, falling back to nfs_export=off.\n");
1648                         ofs->config.nfs_export = false;
1649                 }
1650         }
1651
1652         if (ofs->config.metacopy && ofs->config.nfs_export) {
1653                 pr_warn("overlayfs: NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1654                 ofs->config.nfs_export = false;
1655         }
1656
1657         if (ofs->config.nfs_export)
1658                 sb->s_export_op = &ovl_export_operations;
1659
1660         /* Never override disk quota limits or use reserved space */
1661         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1662
1663         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1664         sb->s_xattr = ovl_xattr_handlers;
1665         sb->s_fs_info = ofs;
1666         sb->s_flags |= SB_POSIXACL;
1667
1668         err = -ENOMEM;
1669         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1670         if (!root_dentry)
1671                 goto out_free_oe;
1672
1673         root_dentry->d_fsdata = oe;
1674
1675         mntput(upperpath.mnt);
1676         if (upperpath.dentry) {
1677                 ovl_dentry_set_upper_alias(root_dentry);
1678                 if (ovl_is_impuredir(upperpath.dentry))
1679                         ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1680         }
1681
1682         /* Root is always merge -> can have whiteouts */
1683         ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1684         ovl_dentry_set_flag(OVL_E_CONNECTED, root_dentry);
1685         ovl_set_upperdata(d_inode(root_dentry));
1686         ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1687                        ovl_dentry_lower(root_dentry), NULL);
1688
1689         sb->s_root = root_dentry;
1690
1691         return 0;
1692
1693 out_free_oe:
1694         ovl_entry_stack_free(oe);
1695         kfree(oe);
1696 out_err:
1697         path_put(&upperpath);
1698         ovl_free_fs(ofs);
1699 out:
1700         return err;
1701 }
1702
1703 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1704                                 const char *dev_name, void *raw_data)
1705 {
1706         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1707 }
1708
1709 static struct file_system_type ovl_fs_type = {
1710         .owner          = THIS_MODULE,
1711         .name           = "overlay",
1712         .mount          = ovl_mount,
1713         .kill_sb        = kill_anon_super,
1714 };
1715 MODULE_ALIAS_FS("overlay");
1716
1717 static void ovl_inode_init_once(void *foo)
1718 {
1719         struct ovl_inode *oi = foo;
1720
1721         inode_init_once(&oi->vfs_inode);
1722 }
1723
1724 static int __init ovl_init(void)
1725 {
1726         int err;
1727
1728         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1729                                              sizeof(struct ovl_inode), 0,
1730                                              (SLAB_RECLAIM_ACCOUNT|
1731                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1732                                              ovl_inode_init_once);
1733         if (ovl_inode_cachep == NULL)
1734                 return -ENOMEM;
1735
1736         err = register_filesystem(&ovl_fs_type);
1737         if (err)
1738                 kmem_cache_destroy(ovl_inode_cachep);
1739
1740         return err;
1741 }
1742
1743 static void __exit ovl_exit(void)
1744 {
1745         unregister_filesystem(&ovl_fs_type);
1746
1747         /*
1748          * Make sure all delayed rcu free inodes are flushed before we
1749          * destroy cache.
1750          */
1751         rcu_barrier();
1752         kmem_cache_destroy(ovl_inode_cachep);
1753
1754 }
1755
1756 module_init(ovl_init);
1757 module_exit(ovl_exit);