fs/overlayfs/super.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *
   4  * Copyright (C) 2011 Novell Inc.
   5  */
   6
   7 #include <uapi/linux/magic.h>
   8 #include <linux/fs.h>
   9 #include <linux/namei.h>
  10 #include <linux/xattr.h>
  11 #include <linux/mount.h>
  12 #include <linux/parser.h>
  13 #include <linux/module.h>
  14 #include <linux/statfs.h>
  15 #include <linux/seq_file.h>
  16 #include <linux/posix_acl_xattr.h>
  17 #include <linux/exportfs.h>
  18 #include "overlayfs.h"
  19
  20 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  21 MODULE_DESCRIPTION("Overlay filesystem");
  22 MODULE_LICENSE("GPL");
  23
  24
  25 struct ovl_dir_cache;
  26
  27 #define OVL_MAX_STACK 500
  28
  29 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  30 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  31 MODULE_PARM_DESC(redirect_dir,
  32                  "Default to on or off for the redirect_dir feature");
  33
  34 static bool ovl_redirect_always_follow =
  35         IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
  36 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
  37                    bool, 0644);
  38 MODULE_PARM_DESC(redirect_always_follow,
  39                  "Follow redirects even if redirect_dir feature is turned off");
  40
  41 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
  42 module_param_named(index, ovl_index_def, bool, 0644);
  43 MODULE_PARM_DESC(index,
  44                  "Default to on or off for the inodes index feature");
  45
  46 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
  47 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
  48 MODULE_PARM_DESC(nfs_export,
  49                  "Default to on or off for the NFS export feature");
  50
  51 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
  52 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
  53 MODULE_PARM_DESC(xino_auto,
  54                  "Auto enable xino feature");
  55
  56 static void ovl_entry_stack_free(struct ovl_entry *oe)
  57 {
  58         unsigned int i;
  59
  60         for (i = 0; i < oe->numlower; i++)
  61                 dput(oe->lowerstack[i].dentry);
  62 }
  63
  64 static bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
  65 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
  66 MODULE_PARM_DESC(metacopy,
  67                  "Default to on or off for the metadata only copy up feature");
  68
  69 static void ovl_dentry_release(struct dentry *dentry)
  70 {
  71         struct ovl_entry *oe = dentry->d_fsdata;
  72
  73         if (oe) {
  74                 ovl_entry_stack_free(oe);
  75                 kfree_rcu(oe, rcu);
  76         }
  77 }
  78
  79 static struct dentry *ovl_d_real(struct dentry *dentry,
  80                                  const struct inode *inode)
  81 {
  82         struct dentry *real;
  83
  84         /* It's an overlay file */
  85         if (inode && d_inode(dentry) == inode)
  86                 return dentry;
  87
  88         if (!d_is_reg(dentry)) {
  89                 if (!inode || inode == d_inode(dentry))
  90                         return dentry;
  91                 goto bug;
  92         }
  93
  94         real = ovl_dentry_upper(dentry);
  95         if (real && (inode == d_inode(real)))
  96                 return real;
  97
  98         if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
  99                 return real;
 100
 101         real = ovl_dentry_lowerdata(dentry);
 102         if (!real)
 103                 goto bug;
 104
 105         /* Handle recursion */
 106         real = d_real(real, inode);
 107
 108         if (!inode || inode == d_inode(real))
 109                 return real;
 110 bug:
 111         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
 112              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 113         return dentry;
 114 }
 115
 116 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
 117 {
 118         int ret = 1;
 119
 120         if (weak) {
 121                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
 122                         ret =  d->d_op->d_weak_revalidate(d, flags);
 123         } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
 124                 ret = d->d_op->d_revalidate(d, flags);
 125                 if (!ret) {
 126                         if (!(flags & LOOKUP_RCU))
 127                                 d_invalidate(d);
 128                         ret = -ESTALE;
 129                 }
 130         }
 131         return ret;
 132 }
 133
 134 static int ovl_dentry_revalidate_common(struct dentry *dentry,
 135                                         unsigned int flags, bool weak)
 136 {
 137         struct ovl_entry *oe = dentry->d_fsdata;
 138         struct dentry *upper;
 139         unsigned int i;
 140         int ret = 1;
 141
 142         upper = ovl_dentry_upper(dentry);
 143         if (upper)
 144                 ret = ovl_revalidate_real(upper, flags, weak);
 145
 146         for (i = 0; ret > 0 && i < oe->numlower; i++) {
 147                 ret = ovl_revalidate_real(oe->lowerstack[i].dentry, flags,
 148                                           weak);
 149         }
 150         return ret;
 151 }
 152
 153 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 154 {
 155         return ovl_dentry_revalidate_common(dentry, flags, false);
 156 }
 157
 158 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 159 {
 160         return ovl_dentry_revalidate_common(dentry, flags, true);
 161 }
 162
 163 static const struct dentry_operations ovl_dentry_operations = {
 164         .d_release = ovl_dentry_release,
 165         .d_real = ovl_d_real,
 166         .d_revalidate = ovl_dentry_revalidate,
 167         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 168 };
 169
 170 static struct kmem_cache *ovl_inode_cachep;
 171
 172 static struct inode *ovl_alloc_inode(struct super_block *sb)
 173 {
 174         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 175
 176         if (!oi)
 177                 return NULL;
 178
 179         oi->cache = NULL;
 180         oi->redirect = NULL;
 181         oi->version = 0;
 182         oi->flags = 0;
 183         oi->__upperdentry = NULL;
 184         oi->lower = NULL;
 185         oi->lowerdata = NULL;
 186         mutex_init(&oi->lock);
 187
 188         return &oi->vfs_inode;
 189 }
 190
 191 static void ovl_free_inode(struct inode *inode)
 192 {
 193         struct ovl_inode *oi = OVL_I(inode);
 194
 195         kfree(oi->redirect);
 196         mutex_destroy(&oi->lock);
 197         kmem_cache_free(ovl_inode_cachep, oi);
 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 }
 211
 212 static void ovl_free_fs(struct ovl_fs *ofs)
 213 {
 214         struct vfsmount **mounts;
 215         unsigned i;
 216
 217         iput(ofs->workbasedir_trap);
 218         iput(ofs->indexdir_trap);
 219         iput(ofs->workdir_trap);
 220         dput(ofs->whiteout);
 221         dput(ofs->indexdir);
 222         dput(ofs->workdir);
 223         if (ofs->workdir_locked)
 224                 ovl_inuse_unlock(ofs->workbasedir);
 225         dput(ofs->workbasedir);
 226         if (ofs->upperdir_locked)
 227                 ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
 228
 229         /* Hack!  Reuse ofs->layers as a vfsmount array before freeing it */
 230         mounts = (struct vfsmount **) ofs->layers;
 231         for (i = 0; i < ofs->numlayer; i++) {
 232                 iput(ofs->layers[i].trap);
 233                 mounts[i] = ofs->layers[i].mnt;
 234         }
 235         kern_unmount_array(mounts, ofs->numlayer);
 236         kfree(ofs->layers);
 237         for (i = 0; i < ofs->numfs; i++)
 238                 free_anon_bdev(ofs->fs[i].pseudo_dev);
 239         kfree(ofs->fs);
 240
 241         kfree(ofs->config.lowerdir);
 242         kfree(ofs->config.upperdir);
 243         kfree(ofs->config.workdir);
 244         kfree(ofs->config.redirect_mode);
 245         if (ofs->creator_cred)
 246                 put_cred(ofs->creator_cred);
 247         kfree(ofs);
 248 }
 249
 250 static void ovl_put_super(struct super_block *sb)
 251 {
 252         struct ovl_fs *ofs = sb->s_fs_info;
 253
 254         ovl_free_fs(ofs);
 255 }
 256
 257 /* Sync real dirty inodes in upper filesystem (if it exists) */
 258 static int ovl_sync_fs(struct super_block *sb, int wait)
 259 {
 260         struct ovl_fs *ofs = sb->s_fs_info;
 261         struct super_block *upper_sb;
 262         int ret;
 263
 264         if (!ovl_upper_mnt(ofs))
 265                 return 0;
 266
 267         if (!ovl_should_sync(ofs))
 268                 return 0;
 269         /*
 270          * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
 271          * All the super blocks will be iterated, including upper_sb.
 272          *
 273          * If this is a syncfs(2) call, then we do need to call
 274          * sync_filesystem() on upper_sb, but enough if we do it when being
 275          * called with wait == 1.
 276          */
 277         if (!wait)
 278                 return 0;
 279
 280         upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 281
 282         down_read(&upper_sb->s_umount);
 283         ret = sync_filesystem(upper_sb);
 284         up_read(&upper_sb->s_umount);
 285
 286         return ret;
 287 }
 288
 289 /**
 290  * ovl_statfs
 291  * @sb: The overlayfs super block
 292  * @buf: The struct kstatfs to fill in with stats
 293  *
 294  * Get the filesystem statistics.  As writes always target the upper layer
 295  * filesystem pass the statfs to the upper filesystem (if it exists)
 296  */
 297 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 298 {
 299         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 300         struct dentry *root_dentry = dentry->d_sb->s_root;
 301         struct path path;
 302         int err;
 303
 304         ovl_path_real(root_dentry, &path);
 305
 306         err = vfs_statfs(&path, buf);
 307         if (!err) {
 308                 buf->f_namelen = ofs->namelen;
 309                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 310         }
 311
 312         return err;
 313 }
 314
 315 /* Will this overlay be forced to mount/remount ro? */
 316 static bool ovl_force_readonly(struct ovl_fs *ofs)
 317 {
 318         return (!ovl_upper_mnt(ofs) || !ofs->workdir);
 319 }
 320
 321 static const char *ovl_redirect_mode_def(void)
 322 {
 323         return ovl_redirect_dir_def ? "on" : "off";
 324 }
 325
 326 static const char * const ovl_xino_str[] = {
 327         "off",
 328         "auto",
 329         "on",
 330 };
 331
 332 static inline int ovl_xino_def(void)
 333 {
 334         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 335 }
 336
 337 /**
 338  * ovl_show_options
 339  *
 340  * Prints the mount options for a given superblock.
 341  * Returns zero; does not fail.
 342  */
 343 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 344 {
 345         struct super_block *sb = dentry->d_sb;
 346         struct ovl_fs *ofs = sb->s_fs_info;
 347
 348         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 349         if (ofs->config.upperdir) {
 350                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 351                 seq_show_option(m, "workdir", ofs->config.workdir);
 352         }
 353         if (ofs->config.default_permissions)
 354                 seq_puts(m, ",default_permissions");
 355         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 356                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 357         if (ofs->config.index != ovl_index_def)
 358                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 359         if (ofs->config.nfs_export != ovl_nfs_export_def)
 360                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 361                                                 "on" : "off");
 362         if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
 363                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 364         if (ofs->config.metacopy != ovl_metacopy_def)
 365                 seq_printf(m, ",metacopy=%s",
 366                            ofs->config.metacopy ? "on" : "off");
 367         if (ofs->config.ovl_volatile)
 368                 seq_puts(m, ",volatile");
 369         return 0;
 370 }
 371
 372 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 373 {
 374         struct ovl_fs *ofs = sb->s_fs_info;
 375         struct super_block *upper_sb;
 376         int ret = 0;
 377
 378         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 379                 return -EROFS;
 380
 381         if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
 382                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 383                 if (ovl_should_sync(ofs)) {
 384                         down_read(&upper_sb->s_umount);
 385                         ret = sync_filesystem(upper_sb);
 386                         up_read(&upper_sb->s_umount);
 387                 }
 388         }
 389
 390         return ret;
 391 }
 392
 393 static const struct super_operations ovl_super_operations = {
 394         .alloc_inode    = ovl_alloc_inode,
 395         .free_inode     = ovl_free_inode,
 396         .destroy_inode  = ovl_destroy_inode,
 397         .drop_inode     = generic_delete_inode,
 398         .put_super      = ovl_put_super,
 399         .sync_fs        = ovl_sync_fs,
 400         .statfs         = ovl_statfs,
 401         .show_options   = ovl_show_options,
 402         .remount_fs     = ovl_remount,
 403 };
 404
 405 enum {
 406         OPT_LOWERDIR,
 407         OPT_UPPERDIR,
 408         OPT_WORKDIR,
 409         OPT_DEFAULT_PERMISSIONS,
 410         OPT_REDIRECT_DIR,
 411         OPT_INDEX_ON,
 412         OPT_INDEX_OFF,
 413         OPT_NFS_EXPORT_ON,
 414         OPT_NFS_EXPORT_OFF,
 415         OPT_XINO_ON,
 416         OPT_XINO_OFF,
 417         OPT_XINO_AUTO,
 418         OPT_METACOPY_ON,
 419         OPT_METACOPY_OFF,
 420         OPT_VOLATILE,
 421         OPT_ERR,
 422 };
 423
 424 static const match_table_t ovl_tokens = {
 425         {OPT_LOWERDIR,                  "lowerdir=%s"},
 426         {OPT_UPPERDIR,                  "upperdir=%s"},
 427         {OPT_WORKDIR,                   "workdir=%s"},
 428         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 429         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 430         {OPT_INDEX_ON,                  "index=on"},
 431         {OPT_INDEX_OFF,                 "index=off"},
 432         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 433         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 434         {OPT_XINO_ON,                   "xino=on"},
 435         {OPT_XINO_OFF,                  "xino=off"},
 436         {OPT_XINO_AUTO,                 "xino=auto"},
 437         {OPT_METACOPY_ON,               "metacopy=on"},
 438         {OPT_METACOPY_OFF,              "metacopy=off"},
 439         {OPT_VOLATILE,                  "volatile"},
 440         {OPT_ERR,                       NULL}
 441 };
 442
 443 static char *ovl_next_opt(char **s)
 444 {
 445         char *sbegin = *s;
 446         char *p;
 447
 448         if (sbegin == NULL)
 449                 return NULL;
 450
 451         for (p = sbegin; *p; p++) {
 452                 if (*p == '\\') {
 453                         p++;
 454                         if (!*p)
 455                                 break;
 456                 } else if (*p == ',') {
 457                         *p = '\0';
 458                         *s = p + 1;
 459                         return sbegin;
 460                 }
 461         }
 462         *s = NULL;
 463         return sbegin;
 464 }
 465
 466 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 467 {
 468         if (strcmp(mode, "on") == 0) {
 469                 config->redirect_dir = true;
 470                 /*
 471                  * Does not make sense to have redirect creation without
 472                  * redirect following.
 473                  */
 474                 config->redirect_follow = true;
 475         } else if (strcmp(mode, "follow") == 0) {
 476                 config->redirect_follow = true;
 477         } else if (strcmp(mode, "off") == 0) {
 478                 if (ovl_redirect_always_follow)
 479                         config->redirect_follow = true;
 480         } else if (strcmp(mode, "nofollow") != 0) {
 481                 pr_err("bad mount option \"redirect_dir=%s\"\n",
 482                        mode);
 483                 return -EINVAL;
 484         }
 485
 486         return 0;
 487 }
 488
 489 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 490 {
 491         char *p;
 492         int err;
 493         bool metacopy_opt = false, redirect_opt = false;
 494         bool nfs_export_opt = false, index_opt = false;
 495
 496         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 497         if (!config->redirect_mode)
 498                 return -ENOMEM;
 499
 500         while ((p = ovl_next_opt(&opt)) != NULL) {
 501                 int token;
 502                 substring_t args[MAX_OPT_ARGS];
 503
 504                 if (!*p)
 505                         continue;
 506
 507                 token = match_token(p, ovl_tokens, args);
 508                 switch (token) {
 509                 case OPT_UPPERDIR:
 510                         kfree(config->upperdir);
 511                         config->upperdir = match_strdup(&args[0]);
 512                         if (!config->upperdir)
 513                                 return -ENOMEM;
 514                         break;
 515
 516                 case OPT_LOWERDIR:
 517                         kfree(config->lowerdir);
 518                         config->lowerdir = match_strdup(&args[0]);
 519                         if (!config->lowerdir)
 520                                 return -ENOMEM;
 521                         break;
 522
 523                 case OPT_WORKDIR:
 524                         kfree(config->workdir);
 525                         config->workdir = match_strdup(&args[0]);
 526                         if (!config->workdir)
 527                                 return -ENOMEM;
 528                         break;
 529
 530                 case OPT_DEFAULT_PERMISSIONS:
 531                         config->default_permissions = true;
 532                         break;
 533
 534                 case OPT_REDIRECT_DIR:
 535                         kfree(config->redirect_mode);
 536                         config->redirect_mode = match_strdup(&args[0]);
 537                         if (!config->redirect_mode)
 538                                 return -ENOMEM;
 539                         redirect_opt = true;
 540                         break;
 541
 542                 case OPT_INDEX_ON:
 543                         config->index = true;
 544                         index_opt = true;
 545                         break;
 546
 547                 case OPT_INDEX_OFF:
 548                         config->index = false;
 549                         index_opt = true;
 550                         break;
 551
 552                 case OPT_NFS_EXPORT_ON:
 553                         config->nfs_export = true;
 554                         nfs_export_opt = true;
 555                         break;
 556
 557                 case OPT_NFS_EXPORT_OFF:
 558                         config->nfs_export = false;
 559                         nfs_export_opt = true;
 560                         break;
 561
 562                 case OPT_XINO_ON:
 563                         config->xino = OVL_XINO_ON;
 564                         break;
 565
 566                 case OPT_XINO_OFF:
 567                         config->xino = OVL_XINO_OFF;
 568                         break;
 569
 570                 case OPT_XINO_AUTO:
 571                         config->xino = OVL_XINO_AUTO;
 572                         break;
 573
 574                 case OPT_METACOPY_ON:
 575                         config->metacopy = true;
 576                         metacopy_opt = true;
 577                         break;
 578
 579                 case OPT_METACOPY_OFF:
 580                         config->metacopy = false;
 581                         metacopy_opt = true;
 582                         break;
 583
 584                 case OPT_VOLATILE:
 585                         config->ovl_volatile = true;
 586                         break;
 587
 588                 default:
 589                         pr_err("unrecognized mount option \"%s\" or missing value\n",
 590                                         p);
 591                         return -EINVAL;
 592                 }
 593         }
 594
 595         /* Workdir/index are useless in non-upper mount */
 596         if (!config->upperdir) {
 597                 if (config->workdir) {
 598                         pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 599                                 config->workdir);
 600                         kfree(config->workdir);
 601                         config->workdir = NULL;
 602                 }
 603                 if (config->index && index_opt) {
 604                         pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
 605                         index_opt = false;
 606                 }
 607                 config->index = false;
 608         }
 609
 610         if (!config->upperdir && config->ovl_volatile) {
 611                 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
 612                 config->ovl_volatile = false;
 613         }
 614
 615         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 616         if (err)
 617                 return err;
 618
 619         /*
 620          * This is to make the logic below simpler.  It doesn't make any other
 621          * difference, since config->redirect_dir is only used for upper.
 622          */
 623         if (!config->upperdir && config->redirect_follow)
 624                 config->redirect_dir = true;
 625
 626         /* Resolve metacopy -> redirect_dir dependency */
 627         if (config->metacopy && !config->redirect_dir) {
 628                 if (metacopy_opt && redirect_opt) {
 629                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
 630                                config->redirect_mode);
 631                         return -EINVAL;
 632                 }
 633                 if (redirect_opt) {
 634                         /*
 635                          * There was an explicit redirect_dir=... that resulted
 636                          * in this conflict.
 637                          */
 638                         pr_info("disabling metacopy due to redirect_dir=%s\n",
 639                                 config->redirect_mode);
 640                         config->metacopy = false;
 641                 } else {
 642                         /* Automatically enable redirect otherwise. */
 643                         config->redirect_follow = config->redirect_dir = true;
 644                 }
 645         }
 646
 647         /* Resolve nfs_export -> index dependency */
 648         if (config->nfs_export && !config->index) {
 649                 if (!config->upperdir && config->redirect_follow) {
 650                         pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
 651                         config->nfs_export = false;
 652                 } else if (nfs_export_opt && index_opt) {
 653                         pr_err("conflicting options: nfs_export=on,index=off\n");
 654                         return -EINVAL;
 655                 } else if (index_opt) {
 656                         /*
 657                          * There was an explicit index=off that resulted
 658                          * in this conflict.
 659                          */
 660                         pr_info("disabling nfs_export due to index=off\n");
 661                         config->nfs_export = false;
 662                 } else {
 663                         /* Automatically enable index otherwise. */
 664                         config->index = true;
 665                 }
 666         }
 667
 668         /* Resolve nfs_export -> !metacopy dependency */
 669         if (config->nfs_export && config->metacopy) {
 670                 if (nfs_export_opt && metacopy_opt) {
 671                         pr_err("conflicting options: nfs_export=on,metacopy=on\n");
 672                         return -EINVAL;
 673                 }
 674                 if (metacopy_opt) {
 675                         /*
 676                          * There was an explicit metacopy=on that resulted
 677                          * in this conflict.
 678                          */
 679                         pr_info("disabling nfs_export due to metacopy=on\n");
 680                         config->nfs_export = false;
 681                 } else {
 682                         /*
 683                          * There was an explicit nfs_export=on that resulted
 684                          * in this conflict.
 685                          */
 686                         pr_info("disabling metacopy due to nfs_export=on\n");
 687                         config->metacopy = false;
 688                 }
 689         }
 690
 691         return 0;
 692 }
 693
 694 #define OVL_WORKDIR_NAME "work"
 695 #define OVL_INDEXDIR_NAME "index"
 696
 697 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 698                                          const char *name, bool persist)
 699 {
 700         struct inode *dir =  ofs->workbasedir->d_inode;
 701         struct vfsmount *mnt = ovl_upper_mnt(ofs);
 702         struct dentry *work;
 703         int err;
 704         bool retried = false;
 705
 706         inode_lock_nested(dir, I_MUTEX_PARENT);
 707 retry:
 708         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
 709
 710         if (!IS_ERR(work)) {
 711                 struct iattr attr = {
 712                         .ia_valid = ATTR_MODE,
 713                         .ia_mode = S_IFDIR | 0,
 714                 };
 715
 716                 if (work->d_inode) {
 717                         err = -EEXIST;
 718                         if (retried)
 719                                 goto out_dput;
 720
 721                         if (persist)
 722                                 goto out_unlock;
 723
 724                         retried = true;
 725                         err = ovl_workdir_cleanup(dir, mnt, work, 0);
 726                         dput(work);
 727                         if (err == -EINVAL) {
 728                                 work = ERR_PTR(err);
 729                                 goto out_unlock;
 730                         }
 731                         goto retry;
 732                 }
 733
 734                 work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
 735                 err = PTR_ERR(work);
 736                 if (IS_ERR(work))
 737                         goto out_err;
 738
 739                 /*
 740                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 741                  *
 742                  * a) success (there was a POSIX ACL xattr and was removed)
 743                  * b) -ENODATA (there was no POSIX ACL xattr)
 744                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 745                  *
 746                  * There are various other error values that could effectively
 747                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 748                  * if the xattr name is too long), but the set of filesystems
 749                  * allowed as upper are limited to "normal" ones, where checking
 750                  * for the above two errors is sufficient.
 751                  */
 752                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 753                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 754                         goto out_dput;
 755
 756                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 757                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 758                         goto out_dput;
 759
 760                 /* Clear any inherited mode bits */
 761                 inode_lock(work->d_inode);
 762                 err = notify_change(work, &attr, NULL);
 763                 inode_unlock(work->d_inode);
 764                 if (err)
 765                         goto out_dput;
 766         } else {
 767                 err = PTR_ERR(work);
 768                 goto out_err;
 769         }
 770 out_unlock:
 771         inode_unlock(dir);
 772         return work;
 773
 774 out_dput:
 775         dput(work);
 776 out_err:
 777         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
 778                 ofs->config.workdir, name, -err);
 779         work = NULL;
 780         goto out_unlock;
 781 }
 782
 783 static void ovl_unescape(char *s)
 784 {
 785         char *d = s;
 786
 787         for (;; s++, d++) {
 788                 if (*s == '\\')
 789                         s++;
 790                 *d = *s;
 791                 if (!*s)
 792                         break;
 793         }
 794 }
 795
 796 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 797 {
 798         int err = -EINVAL;
 799
 800         if (!*name) {
 801                 pr_err("empty lowerdir\n");
 802                 goto out;
 803         }
 804         err = kern_path(name, LOOKUP_FOLLOW, path);
 805         if (err) {
 806                 pr_err("failed to resolve '%s': %i\n", name, err);
 807                 goto out;
 808         }
 809         err = -EINVAL;
 810         if (ovl_dentry_weird(path->dentry)) {
 811                 pr_err("filesystem on '%s' not supported\n", name);
 812                 goto out_put;
 813         }
 814         if (!d_is_dir(path->dentry)) {
 815                 pr_err("'%s' not a directory\n", name);
 816                 goto out_put;
 817         }
 818         return 0;
 819
 820 out_put:
 821         path_put_init(path);
 822 out:
 823         return err;
 824 }
 825
 826 static int ovl_mount_dir(const char *name, struct path *path)
 827 {
 828         int err = -ENOMEM;
 829         char *tmp = kstrdup(name, GFP_KERNEL);
 830
 831         if (tmp) {
 832                 ovl_unescape(tmp);
 833                 err = ovl_mount_dir_noesc(tmp, path);
 834
 835                 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
 836                         pr_err("filesystem on '%s' not supported as upperdir\n",
 837                                tmp);
 838                         path_put_init(path);
 839                         err = -EINVAL;
 840                 }
 841                 kfree(tmp);
 842         }
 843         return err;
 844 }
 845
 846 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 847                              const char *name)
 848 {
 849         struct kstatfs statfs;
 850         int err = vfs_statfs(path, &statfs);
 851
 852         if (err)
 853                 pr_err("statfs failed on '%s'\n", name);
 854         else
 855                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 856
 857         return err;
 858 }
 859
 860 static int ovl_lower_dir(const char *name, struct path *path,
 861                          struct ovl_fs *ofs, int *stack_depth)
 862 {
 863         int fh_type;
 864         int err;
 865
 866         err = ovl_mount_dir_noesc(name, path);
 867         if (err)
 868                 return err;
 869
 870         err = ovl_check_namelen(path, ofs, name);
 871         if (err)
 872                 return err;
 873
 874         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 875
 876         /*
 877          * The inodes index feature and NFS export need to encode and decode
 878          * file handles, so they require that all layers support them.
 879          */
 880         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 881         if ((ofs->config.nfs_export ||
 882              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 883                 ofs->config.index = false;
 884                 ofs->config.nfs_export = false;
 885                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 886                         name);
 887         }
 888
 889         /* Check if lower fs has 32bit inode numbers */
 890         if (fh_type != FILEID_INO32_GEN)
 891                 ofs->xino_mode = -1;
 892
 893         return 0;
 894 }
 895
 896 /* Workdir should not be subdir of upperdir and vice versa */
 897 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 898 {
 899         bool ok = false;
 900
 901         if (workdir != upperdir) {
 902                 ok = (lock_rename(workdir, upperdir) == NULL);
 903                 unlock_rename(workdir, upperdir);
 904         }
 905         return ok;
 906 }
 907
 908 static unsigned int ovl_split_lowerdirs(char *str)
 909 {
 910         unsigned int ctr = 1;
 911         char *s, *d;
 912
 913         for (s = d = str;; s++, d++) {
 914                 if (*s == '\\') {
 915                         s++;
 916                 } else if (*s == ':') {
 917                         *d = '\0';
 918                         ctr++;
 919                         continue;
 920                 }
 921                 *d = *s;
 922                 if (!*s)
 923                         break;
 924         }
 925         return ctr;
 926 }
 927
 928 static int __maybe_unused
 929 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 930                         struct dentry *dentry, struct inode *inode,
 931                         const char *name, void *buffer, size_t size)
 932 {
 933         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 934 }
 935
 936 static int __maybe_unused
 937 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 938                         struct dentry *dentry, struct inode *inode,
 939                         const char *name, const void *value,
 940                         size_t size, int flags)
 941 {
 942         struct dentry *workdir = ovl_workdir(dentry);
 943         struct inode *realinode = ovl_inode_real(inode);
 944         struct posix_acl *acl = NULL;
 945         int err;
 946
 947         /* Check that everything is OK before copy-up */
 948         if (value) {
 949                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 950                 if (IS_ERR(acl))
 951                         return PTR_ERR(acl);
 952         }
 953         err = -EOPNOTSUPP;
 954         if (!IS_POSIXACL(d_inode(workdir)))
 955                 goto out_acl_release;
 956         if (!realinode->i_op->set_acl)
 957                 goto out_acl_release;
 958         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 959                 err = acl ? -EACCES : 0;
 960                 goto out_acl_release;
 961         }
 962         err = -EPERM;
 963         if (!inode_owner_or_capable(inode))
 964                 goto out_acl_release;
 965
 966         posix_acl_release(acl);
 967
 968         /*
 969          * Check if sgid bit needs to be cleared (actual setacl operation will
 970          * be done with mounter's capabilities and so that won't do it for us).
 971          */
 972         if (unlikely(inode->i_mode & S_ISGID) &&
 973             handler->flags == ACL_TYPE_ACCESS &&
 974             !in_group_p(inode->i_gid) &&
 975             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 976                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 977
 978                 err = ovl_setattr(dentry, &iattr);
 979                 if (err)
 980                         return err;
 981         }
 982
 983         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
 984         if (!err)
 985                 ovl_copyattr(ovl_inode_real(inode), inode);
 986
 987         return err;
 988
 989 out_acl_release:
 990         posix_acl_release(acl);
 991         return err;
 992 }
 993
 994 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 995                              struct dentry *dentry, struct inode *inode,
 996                              const char *name, void *buffer, size_t size)
 997 {
 998         return -EOPNOTSUPP;
 999 }
1000
1001 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1002                              struct dentry *dentry, struct inode *inode,
1003                              const char *name, const void *value,
1004                              size_t size, int flags)
1005 {
1006         return -EOPNOTSUPP;
1007 }
1008
1009 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1010                                struct dentry *dentry, struct inode *inode,
1011                                const char *name, void *buffer, size_t size)
1012 {
1013         return ovl_xattr_get(dentry, inode, name, buffer, size);
1014 }
1015
1016 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1017                                struct dentry *dentry, struct inode *inode,
1018                                const char *name, const void *value,
1019                                size_t size, int flags)
1020 {
1021         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1022 }
1023
1024 static const struct xattr_handler __maybe_unused
1025 ovl_posix_acl_access_xattr_handler = {
1026         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1027         .flags = ACL_TYPE_ACCESS,
1028         .get = ovl_posix_acl_xattr_get,
1029         .set = ovl_posix_acl_xattr_set,
1030 };
1031
1032 static const struct xattr_handler __maybe_unused
1033 ovl_posix_acl_default_xattr_handler = {
1034         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1035         .flags = ACL_TYPE_DEFAULT,
1036         .get = ovl_posix_acl_xattr_get,
1037         .set = ovl_posix_acl_xattr_set,
1038 };
1039
1040 static const struct xattr_handler ovl_own_xattr_handler = {
1041         .prefix = OVL_XATTR_PREFIX,
1042         .get = ovl_own_xattr_get,
1043         .set = ovl_own_xattr_set,
1044 };
1045
1046 static const struct xattr_handler ovl_other_xattr_handler = {
1047         .prefix = "", /* catch all */
1048         .get = ovl_other_xattr_get,
1049         .set = ovl_other_xattr_set,
1050 };
1051
1052 static const struct xattr_handler *ovl_xattr_handlers[] = {
1053 #ifdef CONFIG_FS_POSIX_ACL
1054         &ovl_posix_acl_access_xattr_handler,
1055         &ovl_posix_acl_default_xattr_handler,
1056 #endif
1057         &ovl_own_xattr_handler,
1058         &ovl_other_xattr_handler,
1059         NULL
1060 };
1061
1062 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1063                           struct inode **ptrap, const char *name)
1064 {
1065         struct inode *trap;
1066         int err;
1067
1068         trap = ovl_get_trap_inode(sb, dir);
1069         err = PTR_ERR_OR_ZERO(trap);
1070         if (err) {
1071                 if (err == -ELOOP)
1072                         pr_err("conflicting %s path\n", name);
1073                 return err;
1074         }
1075
1076         *ptrap = trap;
1077         return 0;
1078 }
1079
1080 /*
1081  * Determine how we treat concurrent use of upperdir/workdir based on the
1082  * index feature. This is papering over mount leaks of container runtimes,
1083  * for example, an old overlay mount is leaked and now its upperdir is
1084  * attempted to be used as a lower layer in a new overlay mount.
1085  */
1086 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1087 {
1088         if (ofs->config.index) {
1089                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1090                        name);
1091                 return -EBUSY;
1092         } else {
1093                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1094                         name);
1095                 return 0;
1096         }
1097 }
1098
1099 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1100                          struct ovl_layer *upper_layer, struct path *upperpath)
1101 {
1102         struct vfsmount *upper_mnt;
1103         int err;
1104
1105         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1106         if (err)
1107                 goto out;
1108
1109         /* Upper fs should not be r/o */
1110         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
1111                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1112                 err = -EINVAL;
1113                 goto out;
1114         }
1115
1116         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1117         if (err)
1118                 goto out;
1119
1120         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1121                              "upperdir");
1122         if (err)
1123                 goto out;
1124
1125         upper_mnt = clone_private_mount(upperpath);
1126         err = PTR_ERR(upper_mnt);
1127         if (IS_ERR(upper_mnt)) {
1128                 pr_err("failed to clone upperpath\n");
1129                 goto out;
1130         }
1131
1132         /* Don't inherit atime flags */
1133         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1134         upper_layer->mnt = upper_mnt;
1135         upper_layer->idx = 0;
1136         upper_layer->fsid = 0;
1137
1138         /*
1139          * Inherit SB_NOSEC flag from upperdir.
1140          *
1141          * This optimization changes behavior when a security related attribute
1142          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1143          * okay because we don't yet have guarantees in that case, but it will
1144          * need careful treatment once we want to honour changes to underlying
1145          * filesystems.
1146          */
1147         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1148                 sb->s_flags |= SB_NOSEC;
1149
1150         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1151                 ofs->upperdir_locked = true;
1152         } else {
1153                 err = ovl_report_in_use(ofs, "upperdir");
1154                 if (err)
1155                         goto out;
1156         }
1157
1158         err = 0;
1159 out:
1160         return err;
1161 }
1162
1163 /*
1164  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1165  * negative values if error is encountered.
1166  */
1167 static int ovl_check_rename_whiteout(struct dentry *workdir)
1168 {
1169         struct inode *dir = d_inode(workdir);
1170         struct dentry *temp;
1171         struct dentry *dest;
1172         struct dentry *whiteout;
1173         struct name_snapshot name;
1174         int err;
1175
1176         inode_lock_nested(dir, I_MUTEX_PARENT);
1177
1178         temp = ovl_create_temp(workdir, OVL_CATTR(S_IFREG | 0));
1179         err = PTR_ERR(temp);
1180         if (IS_ERR(temp))
1181                 goto out_unlock;
1182
1183         dest = ovl_lookup_temp(workdir);
1184         err = PTR_ERR(dest);
1185         if (IS_ERR(dest)) {
1186                 dput(temp);
1187                 goto out_unlock;
1188         }
1189
1190         /* Name is inline and stable - using snapshot as a copy helper */
1191         take_dentry_name_snapshot(&name, temp);
1192         err = ovl_do_rename(dir, temp, dir, dest, RENAME_WHITEOUT);
1193         if (err) {
1194                 if (err == -EINVAL)
1195                         err = 0;
1196                 goto cleanup_temp;
1197         }
1198
1199         whiteout = lookup_one_len(name.name.name, workdir, name.name.len);
1200         err = PTR_ERR(whiteout);
1201         if (IS_ERR(whiteout))
1202                 goto cleanup_temp;
1203
1204         err = ovl_is_whiteout(whiteout);
1205
1206         /* Best effort cleanup of whiteout and temp file */
1207         if (err)
1208                 ovl_cleanup(dir, whiteout);
1209         dput(whiteout);
1210
1211 cleanup_temp:
1212         ovl_cleanup(dir, temp);
1213         release_dentry_name_snapshot(&name);
1214         dput(temp);
1215         dput(dest);
1216
1217 out_unlock:
1218         inode_unlock(dir);
1219
1220         return err;
1221 }
1222
1223 static struct dentry *ovl_lookup_or_create(struct dentry *parent,
1224                                            const char *name, umode_t mode)
1225 {
1226         size_t len = strlen(name);
1227         struct dentry *child;
1228
1229         inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1230         child = lookup_one_len(name, parent, len);
1231         if (!IS_ERR(child) && !child->d_inode)
1232                 child = ovl_create_real(parent->d_inode, child,
1233                                         OVL_CATTR(mode));
1234         inode_unlock(parent->d_inode);
1235         dput(parent);
1236
1237         return child;
1238 }
1239
1240 /*
1241  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1242  * present.
1243  */
1244 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1245 {
1246         unsigned int ctr;
1247         struct dentry *d = dget(ofs->workbasedir);
1248         static const char *const volatile_path[] = {
1249                 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1250         };
1251         const char *const *name = volatile_path;
1252
1253         for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1254                 d = ovl_lookup_or_create(d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1255                 if (IS_ERR(d))
1256                         return PTR_ERR(d);
1257         }
1258         dput(d);
1259         return 0;
1260 }
1261
1262 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1263                             struct path *workpath)
1264 {
1265         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1266         struct dentry *temp, *workdir;
1267         bool rename_whiteout;
1268         bool d_type;
1269         int fh_type;
1270         int err;
1271
1272         err = mnt_want_write(mnt);
1273         if (err)
1274                 return err;
1275
1276         workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1277         err = PTR_ERR(workdir);
1278         if (IS_ERR_OR_NULL(workdir))
1279                 goto out;
1280
1281         ofs->workdir = workdir;
1282
1283         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1284         if (err)
1285                 goto out;
1286
1287         /*
1288          * Upper should support d_type, else whiteouts are visible.  Given
1289          * workdir and upper are on same fs, we can do iterate_dir() on
1290          * workdir. This check requires successful creation of workdir in
1291          * previous step.
1292          */
1293         err = ovl_check_d_type_supported(workpath);
1294         if (err < 0)
1295                 goto out;
1296
1297         d_type = err;
1298         if (!d_type)
1299                 pr_warn("upper fs needs to support d_type.\n");
1300
1301         /* Check if upper/work fs supports O_TMPFILE */
1302         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1303         ofs->tmpfile = !IS_ERR(temp);
1304         if (ofs->tmpfile)
1305                 dput(temp);
1306         else
1307                 pr_warn("upper fs does not support tmpfile.\n");
1308
1309
1310         /* Check if upper/work fs supports RENAME_WHITEOUT */
1311         err = ovl_check_rename_whiteout(ofs->workdir);
1312         if (err < 0)
1313                 goto out;
1314
1315         rename_whiteout = err;
1316         if (!rename_whiteout)
1317                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1318
1319         /*
1320          * Check if upper/work fs supports trusted.overlay.* xattr
1321          */
1322         err = ovl_do_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1323         if (err) {
1324                 ofs->noxattr = true;
1325                 ofs->config.index = false;
1326                 ofs->config.metacopy = false;
1327                 pr_warn("upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1328                 err = 0;
1329         } else {
1330                 ovl_do_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1331         }
1332
1333         /*
1334          * We allowed sub-optimal upper fs configuration and don't want to break
1335          * users over kernel upgrade, but we never allowed remote upper fs, so
1336          * we can enforce strict requirements for remote upper fs.
1337          */
1338         if (ovl_dentry_remote(ofs->workdir) &&
1339             (!d_type || !rename_whiteout || ofs->noxattr)) {
1340                 pr_err("upper fs missing required features.\n");
1341                 err = -EINVAL;
1342                 goto out;
1343         }
1344
1345         /*
1346          * For volatile mount, create a incompat/volatile/dirty file to keep
1347          * track of it.
1348          */
1349         if (ofs->config.ovl_volatile) {
1350                 err = ovl_create_volatile_dirty(ofs);
1351                 if (err < 0) {
1352                         pr_err("Failed to create volatile/dirty file.\n");
1353                         goto out;
1354                 }
1355         }
1356
1357         /* Check if upper/work fs supports file handles */
1358         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1359         if (ofs->config.index && !fh_type) {
1360                 ofs->config.index = false;
1361                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1362         }
1363
1364         /* Check if upper fs has 32bit inode numbers */
1365         if (fh_type != FILEID_INO32_GEN)
1366                 ofs->xino_mode = -1;
1367
1368         /* NFS export of r/w mount depends on index */
1369         if (ofs->config.nfs_export && !ofs->config.index) {
1370                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1371                 ofs->config.nfs_export = false;
1372         }
1373 out:
1374         mnt_drop_write(mnt);
1375         return err;
1376 }
1377
1378 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1379                            struct path *upperpath)
1380 {
1381         int err;
1382         struct path workpath = { };
1383
1384         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1385         if (err)
1386                 goto out;
1387
1388         err = -EINVAL;
1389         if (upperpath->mnt != workpath.mnt) {
1390                 pr_err("workdir and upperdir must reside under the same mount\n");
1391                 goto out;
1392         }
1393         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1394                 pr_err("workdir and upperdir must be separate subtrees\n");
1395                 goto out;
1396         }
1397
1398         ofs->workbasedir = dget(workpath.dentry);
1399
1400         if (ovl_inuse_trylock(ofs->workbasedir)) {
1401                 ofs->workdir_locked = true;
1402         } else {
1403                 err = ovl_report_in_use(ofs, "workdir");
1404                 if (err)
1405                         goto out;
1406         }
1407
1408         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1409                              "workdir");
1410         if (err)
1411                 goto out;
1412
1413         err = ovl_make_workdir(sb, ofs, &workpath);
1414
1415 out:
1416         path_put(&workpath);
1417
1418         return err;
1419 }
1420
1421 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1422                             struct ovl_entry *oe, struct path *upperpath)
1423 {
1424         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1425         struct dentry *indexdir;
1426         int err;
1427
1428         err = mnt_want_write(mnt);
1429         if (err)
1430                 return err;
1431
1432         /* Verify lower root is upper root origin */
1433         err = ovl_verify_origin(ofs, upperpath->dentry,
1434                                 oe->lowerstack[0].dentry, true);
1435         if (err) {
1436                 pr_err("failed to verify upper root origin\n");
1437                 goto out;
1438         }
1439
1440         /* index dir will act also as workdir */
1441         iput(ofs->workdir_trap);
1442         ofs->workdir_trap = NULL;
1443         dput(ofs->workdir);
1444         ofs->workdir = NULL;
1445         indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1446         if (IS_ERR(indexdir)) {
1447                 err = PTR_ERR(indexdir);
1448         } else if (indexdir) {
1449                 ofs->indexdir = indexdir;
1450                 ofs->workdir = dget(indexdir);
1451
1452                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1453                                      "indexdir");
1454                 if (err)
1455                         goto out;
1456
1457                 /*
1458                  * Verify upper root is exclusively associated with index dir.
1459                  * Older kernels stored upper fh in "trusted.overlay.origin"
1460                  * xattr. If that xattr exists, verify that it is a match to
1461                  * upper dir file handle. In any case, verify or set xattr
1462                  * "trusted.overlay.upper" to indicate that index may have
1463                  * directory entries.
1464                  */
1465                 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1466                         err = ovl_verify_set_fh(ofs, ofs->indexdir,
1467                                                 OVL_XATTR_ORIGIN,
1468                                                 upperpath->dentry, true, false);
1469                         if (err)
1470                                 pr_err("failed to verify index dir 'origin' xattr\n");
1471                 }
1472                 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1473                                        true);
1474                 if (err)
1475                         pr_err("failed to verify index dir 'upper' xattr\n");
1476
1477                 /* Cleanup bad/stale/orphan index entries */
1478                 if (!err)
1479                         err = ovl_indexdir_cleanup(ofs);
1480         }
1481         if (err || !ofs->indexdir)
1482                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1483
1484 out:
1485         mnt_drop_write(mnt);
1486         return err;
1487 }
1488
1489 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1490 {
1491         unsigned int i;
1492
1493         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1494                 return true;
1495
1496         /*
1497          * We allow using single lower with null uuid for index and nfs_export
1498          * for example to support those features with single lower squashfs.
1499          * To avoid regressions in setups of overlay with re-formatted lower
1500          * squashfs, do not allow decoding origin with lower null uuid unless
1501          * user opted-in to one of the new features that require following the
1502          * lower inode of non-dir upper.
1503          */
1504         if (!ofs->config.index && !ofs->config.metacopy && !ofs->config.xino &&
1505             uuid_is_null(uuid))
1506                 return false;
1507
1508         for (i = 0; i < ofs->numfs; i++) {
1509                 /*
1510                  * We use uuid to associate an overlay lower file handle with a
1511                  * lower layer, so we can accept lower fs with null uuid as long
1512                  * as all lower layers with null uuid are on the same fs.
1513                  * if we detect multiple lower fs with the same uuid, we
1514                  * disable lower file handle decoding on all of them.
1515                  */
1516                 if (ofs->fs[i].is_lower &&
1517                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1518                         ofs->fs[i].bad_uuid = true;
1519                         return false;
1520                 }
1521         }
1522         return true;
1523 }
1524
1525 /* Get a unique fsid for the layer */
1526 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1527 {
1528         struct super_block *sb = path->mnt->mnt_sb;
1529         unsigned int i;
1530         dev_t dev;
1531         int err;
1532         bool bad_uuid = false;
1533
1534         for (i = 0; i < ofs->numfs; i++) {
1535                 if (ofs->fs[i].sb == sb)
1536                         return i;
1537         }
1538
1539         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1540                 bad_uuid = true;
1541                 if (ofs->config.index || ofs->config.nfs_export) {
1542                         ofs->config.index = false;
1543                         ofs->config.nfs_export = false;
1544                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to index=off,nfs_export=off.\n",
1545                                 uuid_is_null(&sb->s_uuid) ? "null" :
1546                                                             "conflicting",
1547                                 path->dentry);
1548                 }
1549         }
1550
1551         err = get_anon_bdev(&dev);
1552         if (err) {
1553                 pr_err("failed to get anonymous bdev for lowerpath\n");
1554                 return err;
1555         }
1556
1557         ofs->fs[ofs->numfs].sb = sb;
1558         ofs->fs[ofs->numfs].pseudo_dev = dev;
1559         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1560
1561         return ofs->numfs++;
1562 }
1563
1564 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1565                           struct path *stack, unsigned int numlower,
1566                           struct ovl_layer *layers)
1567 {
1568         int err;
1569         unsigned int i;
1570
1571         err = -ENOMEM;
1572         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1573         if (ofs->fs == NULL)
1574                 goto out;
1575
1576         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1577         ofs->numfs++;
1578
1579         /*
1580          * All lower layers that share the same fs as upper layer, use the same
1581          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1582          * only overlay to simplify ovl_fs_free().
1583          * is_lower will be set if upper fs is shared with a lower layer.
1584          */
1585         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1586         if (err) {
1587                 pr_err("failed to get anonymous bdev for upper fs\n");
1588                 goto out;
1589         }
1590
1591         if (ovl_upper_mnt(ofs)) {
1592                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1593                 ofs->fs[0].is_lower = false;
1594         }
1595
1596         for (i = 0; i < numlower; i++) {
1597                 struct vfsmount *mnt;
1598                 struct inode *trap;
1599                 int fsid;
1600
1601                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1602                 if (err < 0)
1603                         goto out;
1604
1605                 /*
1606                  * Check if lower root conflicts with this overlay layers before
1607                  * checking if it is in-use as upperdir/workdir of "another"
1608                  * mount, because we do not bother to check in ovl_is_inuse() if
1609                  * the upperdir/workdir is in fact in-use by our
1610                  * upperdir/workdir.
1611                  */
1612                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1613                 if (err)
1614                         goto out;
1615
1616                 if (ovl_is_inuse(stack[i].dentry)) {
1617                         err = ovl_report_in_use(ofs, "lowerdir");
1618                         if (err) {
1619                                 iput(trap);
1620                                 goto out;
1621                         }
1622                 }
1623
1624                 mnt = clone_private_mount(&stack[i]);
1625                 err = PTR_ERR(mnt);
1626                 if (IS_ERR(mnt)) {
1627                         pr_err("failed to clone lowerpath\n");
1628                         iput(trap);
1629                         goto out;
1630                 }
1631
1632                 /*
1633                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1634                  * will fail instead of modifying lower fs.
1635                  */
1636                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1637
1638                 layers[ofs->numlayer].trap = trap;
1639                 layers[ofs->numlayer].mnt = mnt;
1640                 layers[ofs->numlayer].idx = ofs->numlayer;
1641                 layers[ofs->numlayer].fsid = fsid;
1642                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1643                 ofs->numlayer++;
1644                 ofs->fs[fsid].is_lower = true;
1645         }
1646
1647         /*
1648          * When all layers on same fs, overlay can use real inode numbers.
1649          * With mount option "xino=<on|auto>", mounter declares that there are
1650          * enough free high bits in underlying fs to hold the unique fsid.
1651          * If overlayfs does encounter underlying inodes using the high xino
1652          * bits reserved for fsid, it emits a warning and uses the original
1653          * inode number or a non persistent inode number allocated from a
1654          * dedicated range.
1655          */
1656         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1657                 if (ofs->config.xino == OVL_XINO_ON)
1658                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1659                 ofs->xino_mode = 0;
1660         } else if (ofs->config.xino == OVL_XINO_OFF) {
1661                 ofs->xino_mode = -1;
1662         } else if (ofs->xino_mode < 0) {
1663                 /*
1664                  * This is a roundup of number of bits needed for encoding
1665                  * fsid, where fsid 0 is reserved for upper fs (even with
1666                  * lower only overlay) +1 extra bit is reserved for the non
1667                  * persistent inode number range that is used for resolving
1668                  * xino lower bits overflow.
1669                  */
1670                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1671                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1672         }
1673
1674         if (ofs->xino_mode > 0) {
1675                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1676                         ofs->xino_mode);
1677         }
1678
1679         err = 0;
1680 out:
1681         return err;
1682 }
1683
1684 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1685                                 const char *lower, unsigned int numlower,
1686                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1687 {
1688         int err;
1689         struct path *stack = NULL;
1690         unsigned int i;
1691         struct ovl_entry *oe;
1692
1693         if (!ofs->config.upperdir && numlower == 1) {
1694                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1695                 return ERR_PTR(-EINVAL);
1696         }
1697
1698         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1699         if (!stack)
1700                 return ERR_PTR(-ENOMEM);
1701
1702         err = -EINVAL;
1703         for (i = 0; i < numlower; i++) {
1704                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1705                 if (err)
1706                         goto out_err;
1707
1708                 lower = strchr(lower, '\0') + 1;
1709         }
1710
1711         err = -EINVAL;
1712         sb->s_stack_depth++;
1713         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1714                 pr_err("maximum fs stacking depth exceeded\n");
1715                 goto out_err;
1716         }
1717
1718         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1719         if (err)
1720                 goto out_err;
1721
1722         err = -ENOMEM;
1723         oe = ovl_alloc_entry(numlower);
1724         if (!oe)
1725                 goto out_err;
1726
1727         for (i = 0; i < numlower; i++) {
1728                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1729                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1730         }
1731
1732 out:
1733         for (i = 0; i < numlower; i++)
1734                 path_put(&stack[i]);
1735         kfree(stack);
1736
1737         return oe;
1738
1739 out_err:
1740         oe = ERR_PTR(err);
1741         goto out;
1742 }
1743
1744 /*
1745  * Check if this layer root is a descendant of:
1746  * - another layer of this overlayfs instance
1747  * - upper/work dir of any overlayfs instance
1748  */
1749 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1750                            struct dentry *dentry, const char *name)
1751 {
1752         struct dentry *next = dentry, *parent;
1753         int err = 0;
1754
1755         if (!dentry)
1756                 return 0;
1757
1758         parent = dget_parent(next);
1759
1760         /* Walk back ancestors to root (inclusive) looking for traps */
1761         while (!err && parent != next) {
1762                 if (ovl_lookup_trap_inode(sb, parent)) {
1763                         err = -ELOOP;
1764                         pr_err("overlapping %s path\n", name);
1765                 } else if (ovl_is_inuse(parent)) {
1766                         err = ovl_report_in_use(ofs, name);
1767                 }
1768                 next = parent;
1769                 parent = dget_parent(next);
1770                 dput(next);
1771         }
1772
1773         dput(parent);
1774
1775         return err;
1776 }
1777
1778 /*
1779  * Check if any of the layers or work dirs overlap.
1780  */
1781 static int ovl_check_overlapping_layers(struct super_block *sb,
1782                                         struct ovl_fs *ofs)
1783 {
1784         int i, err;
1785
1786         if (ovl_upper_mnt(ofs)) {
1787                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1788                                       "upperdir");
1789                 if (err)
1790                         return err;
1791
1792                 /*
1793                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1794                  * this instance and covers overlapping work and index dirs,
1795                  * unless work or index dir have been moved since created inside
1796                  * workbasedir.  In that case, we already have their traps in
1797                  * inode cache and we will catch that case on lookup.
1798                  */
1799                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir");
1800                 if (err)
1801                         return err;
1802         }
1803
1804         for (i = 1; i < ofs->numlayer; i++) {
1805                 err = ovl_check_layer(sb, ofs,
1806                                       ofs->layers[i].mnt->mnt_root,
1807                                       "lowerdir");
1808                 if (err)
1809                         return err;
1810         }
1811
1812         return 0;
1813 }
1814
1815 static struct dentry *ovl_get_root(struct super_block *sb,
1816                                    struct dentry *upperdentry,
1817                                    struct ovl_entry *oe)
1818 {
1819         struct dentry *root;
1820         struct ovl_path *lowerpath = &oe->lowerstack[0];
1821         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1822         int fsid = lowerpath->layer->fsid;
1823         struct ovl_inode_params oip = {
1824                 .upperdentry = upperdentry,
1825                 .lowerpath = lowerpath,
1826         };
1827
1828         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1829         if (!root)
1830                 return NULL;
1831
1832         root->d_fsdata = oe;
1833
1834         if (upperdentry) {
1835                 /* Root inode uses upper st_ino/i_ino */
1836                 ino = d_inode(upperdentry)->i_ino;
1837                 fsid = 0;
1838                 ovl_dentry_set_upper_alias(root);
1839                 if (ovl_is_impuredir(sb, upperdentry))
1840                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1841         }
1842
1843         /* Root is always merge -> can have whiteouts */
1844         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1845         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1846         ovl_set_upperdata(d_inode(root));
1847         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1848         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1849
1850         return root;
1851 }
1852
1853 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1854 {
1855         struct path upperpath = { };
1856         struct dentry *root_dentry;
1857         struct ovl_entry *oe;
1858         struct ovl_fs *ofs;
1859         struct ovl_layer *layers;
1860         struct cred *cred;
1861         char *splitlower = NULL;
1862         unsigned int numlower;
1863         int err;
1864
1865         sb->s_d_op = &ovl_dentry_operations;
1866
1867         err = -ENOMEM;
1868         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1869         if (!ofs)
1870                 goto out;
1871
1872         ofs->creator_cred = cred = prepare_creds();
1873         if (!cred)
1874                 goto out_err;
1875
1876         /* Is there a reason anyone would want not to share whiteouts? */
1877         ofs->share_whiteout = true;
1878
1879         ofs->config.index = ovl_index_def;
1880         ofs->config.nfs_export = ovl_nfs_export_def;
1881         ofs->config.xino = ovl_xino_def();
1882         ofs->config.metacopy = ovl_metacopy_def;
1883         err = ovl_parse_opt((char *) data, &ofs->config);
1884         if (err)
1885                 goto out_err;
1886
1887         err = -EINVAL;
1888         if (!ofs->config.lowerdir) {
1889                 if (!silent)
1890                         pr_err("missing 'lowerdir'\n");
1891                 goto out_err;
1892         }
1893
1894         err = -ENOMEM;
1895         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1896         if (!splitlower)
1897                 goto out_err;
1898
1899         numlower = ovl_split_lowerdirs(splitlower);
1900         if (numlower > OVL_MAX_STACK) {
1901                 pr_err("too many lower directories, limit is %d\n",
1902                        OVL_MAX_STACK);
1903                 goto out_err;
1904         }
1905
1906         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1907         if (!layers)
1908                 goto out_err;
1909
1910         ofs->layers = layers;
1911         /* Layer 0 is reserved for upper even if there's no upper */
1912         ofs->numlayer = 1;
1913
1914         sb->s_stack_depth = 0;
1915         sb->s_maxbytes = MAX_LFS_FILESIZE;
1916         atomic_long_set(&ofs->last_ino, 1);
1917         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1918         if (ofs->config.xino != OVL_XINO_OFF) {
1919                 ofs->xino_mode = BITS_PER_LONG - 32;
1920                 if (!ofs->xino_mode) {
1921                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1922                         ofs->config.xino = OVL_XINO_OFF;
1923                 }
1924         }
1925
1926         /* alloc/destroy_inode needed for setting up traps in inode cache */
1927         sb->s_op = &ovl_super_operations;
1928
1929         if (ofs->config.upperdir) {
1930                 if (!ofs->config.workdir) {
1931                         pr_err("missing 'workdir'\n");
1932                         goto out_err;
1933                 }
1934
1935                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
1936                 if (err)
1937                         goto out_err;
1938
1939                 err = ovl_get_workdir(sb, ofs, &upperpath);
1940                 if (err)
1941                         goto out_err;
1942
1943                 if (!ofs->workdir)
1944                         sb->s_flags |= SB_RDONLY;
1945
1946                 sb->s_stack_depth = ovl_upper_mnt(ofs)->mnt_sb->s_stack_depth;
1947                 sb->s_time_gran = ovl_upper_mnt(ofs)->mnt_sb->s_time_gran;
1948
1949         }
1950         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
1951         err = PTR_ERR(oe);
1952         if (IS_ERR(oe))
1953                 goto out_err;
1954
1955         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1956         if (!ovl_upper_mnt(ofs))
1957                 sb->s_flags |= SB_RDONLY;
1958
1959         if (!ovl_force_readonly(ofs) && ofs->config.index) {
1960                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
1961                 if (err)
1962                         goto out_free_oe;
1963
1964                 /* Force r/o mount with no index dir */
1965                 if (!ofs->indexdir)
1966                         sb->s_flags |= SB_RDONLY;
1967         }
1968
1969         err = ovl_check_overlapping_layers(sb, ofs);
1970         if (err)
1971                 goto out_free_oe;
1972
1973         /* Show index=off in /proc/mounts for forced r/o mount */
1974         if (!ofs->indexdir) {
1975                 ofs->config.index = false;
1976                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
1977                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
1978                         ofs->config.nfs_export = false;
1979                 }
1980         }
1981
1982         if (ofs->config.metacopy && ofs->config.nfs_export) {
1983                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1984                 ofs->config.nfs_export = false;
1985         }
1986
1987         if (ofs->config.nfs_export)
1988                 sb->s_export_op = &ovl_export_operations;
1989
1990         /* Never override disk quota limits or use reserved space */
1991         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1992
1993         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1994         sb->s_xattr = ovl_xattr_handlers;
1995         sb->s_fs_info = ofs;
1996         sb->s_flags |= SB_POSIXACL;
1997         sb->s_iflags |= SB_I_SKIP_SYNC;
1998
1999         err = -ENOMEM;
2000         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2001         if (!root_dentry)
2002                 goto out_free_oe;
2003
2004         mntput(upperpath.mnt);
2005         kfree(splitlower);
2006
2007         sb->s_root = root_dentry;
2008
2009         return 0;
2010
2011 out_free_oe:
2012         ovl_entry_stack_free(oe);
2013         kfree(oe);
2014 out_err:
2015         kfree(splitlower);
2016         path_put(&upperpath);
2017         ovl_free_fs(ofs);
2018 out:
2019         return err;
2020 }
2021
2022 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2023                                 const char *dev_name, void *raw_data)
2024 {
2025         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2026 }
2027
2028 static struct file_system_type ovl_fs_type = {
2029         .owner          = THIS_MODULE,
2030         .name           = "overlay",
2031         .mount          = ovl_mount,
2032         .kill_sb        = kill_anon_super,
2033 };
2034 MODULE_ALIAS_FS("overlay");
2035
2036 static void ovl_inode_init_once(void *foo)
2037 {
2038         struct ovl_inode *oi = foo;
2039
2040         inode_init_once(&oi->vfs_inode);
2041 }
2042
2043 static int __init ovl_init(void)
2044 {
2045         int err;
2046
2047         ovl_inode_cachep = kmem_cache_create("ovl_inode",
2048                                              sizeof(struct ovl_inode), 0,
2049                                              (SLAB_RECLAIM_ACCOUNT|
2050                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2051                                              ovl_inode_init_once);
2052         if (ovl_inode_cachep == NULL)
2053                 return -ENOMEM;
2054
2055         err = ovl_aio_request_cache_init();
2056         if (!err) {
2057                 err = register_filesystem(&ovl_fs_type);
2058                 if (!err)
2059                         return 0;
2060
2061                 ovl_aio_request_cache_destroy();
2062         }
2063         kmem_cache_destroy(ovl_inode_cachep);
2064
2065         return err;
2066 }
2067
2068 static void __exit ovl_exit(void)
2069 {
2070         unregister_filesystem(&ovl_fs_type);
2071
2072         /*
2073          * Make sure all delayed rcu free inodes are flushed before we
2074          * destroy cache.
2075          */
2076         rcu_barrier();
2077         kmem_cache_destroy(ovl_inode_cachep);
2078         ovl_aio_request_cache_destroy();
2079 }
2080
2081 module_init(ovl_init);
2082 module_exit(ovl_exit);