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 = NULL, *lower;
  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         lower = ovl_dentry_lowerdata(dentry);
 102         if (!lower)
 103                 goto bug;
 104         real = lower;
 105
 106         /* Handle recursion */
 107         real = d_real(real, inode);
 108
 109         if (!inode || inode == d_inode(real))
 110                 return real;
 111 bug:
 112         WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
 113              __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
 114              inode ? inode->i_ino : 0, real,
 115              real && d_inode(real) ? d_inode(real)->i_ino : 0);
 116         return dentry;
 117 }
 118
 119 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
 120 {
 121         int ret = 1;
 122
 123         if (weak) {
 124                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
 125                         ret =  d->d_op->d_weak_revalidate(d, flags);
 126         } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
 127                 ret = d->d_op->d_revalidate(d, flags);
 128                 if (!ret) {
 129                         if (!(flags & LOOKUP_RCU))
 130                                 d_invalidate(d);
 131                         ret = -ESTALE;
 132                 }
 133         }
 134         return ret;
 135 }
 136
 137 static int ovl_dentry_revalidate_common(struct dentry *dentry,
 138                                         unsigned int flags, bool weak)
 139 {
 140         struct ovl_entry *oe = dentry->d_fsdata;
 141         struct dentry *upper;
 142         unsigned int i;
 143         int ret = 1;
 144
 145         upper = ovl_dentry_upper(dentry);
 146         if (upper)
 147                 ret = ovl_revalidate_real(upper, flags, weak);
 148
 149         for (i = 0; ret > 0 && i < oe->numlower; i++) {
 150                 ret = ovl_revalidate_real(oe->lowerstack[i].dentry, flags,
 151                                           weak);
 152         }
 153         return ret;
 154 }
 155
 156 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 157 {
 158         return ovl_dentry_revalidate_common(dentry, flags, false);
 159 }
 160
 161 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 162 {
 163         return ovl_dentry_revalidate_common(dentry, flags, true);
 164 }
 165
 166 static const struct dentry_operations ovl_dentry_operations = {
 167         .d_release = ovl_dentry_release,
 168         .d_real = ovl_d_real,
 169         .d_revalidate = ovl_dentry_revalidate,
 170         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 171 };
 172
 173 static struct kmem_cache *ovl_inode_cachep;
 174
 175 static struct inode *ovl_alloc_inode(struct super_block *sb)
 176 {
 177         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 178
 179         if (!oi)
 180                 return NULL;
 181
 182         oi->cache = NULL;
 183         oi->redirect = NULL;
 184         oi->version = 0;
 185         oi->flags = 0;
 186         oi->__upperdentry = NULL;
 187         oi->lower = NULL;
 188         oi->lowerdata = NULL;
 189         mutex_init(&oi->lock);
 190
 191         return &oi->vfs_inode;
 192 }
 193
 194 static void ovl_free_inode(struct inode *inode)
 195 {
 196         struct ovl_inode *oi = OVL_I(inode);
 197
 198         kfree(oi->redirect);
 199         mutex_destroy(&oi->lock);
 200         kmem_cache_free(ovl_inode_cachep, oi);
 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 }
 214
 215 static void ovl_free_fs(struct ovl_fs *ofs)
 216 {
 217         struct vfsmount **mounts;
 218         unsigned i;
 219
 220         iput(ofs->workbasedir_trap);
 221         iput(ofs->indexdir_trap);
 222         iput(ofs->workdir_trap);
 223         dput(ofs->whiteout);
 224         dput(ofs->indexdir);
 225         dput(ofs->workdir);
 226         if (ofs->workdir_locked)
 227                 ovl_inuse_unlock(ofs->workbasedir);
 228         dput(ofs->workbasedir);
 229         if (ofs->upperdir_locked)
 230                 ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
 231
 232         /* Hack!  Reuse ofs->layers as a vfsmount array before freeing it */
 233         mounts = (struct vfsmount **) ofs->layers;
 234         for (i = 0; i < ofs->numlayer; i++) {
 235                 iput(ofs->layers[i].trap);
 236                 mounts[i] = ofs->layers[i].mnt;
 237         }
 238         kern_unmount_array(mounts, ofs->numlayer);
 239         kfree(ofs->layers);
 240         for (i = 0; i < ofs->numfs; i++)
 241                 free_anon_bdev(ofs->fs[i].pseudo_dev);
 242         kfree(ofs->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 (!ovl_upper_mnt(ofs))
 268                 return 0;
 269
 270         if (!ovl_should_sync(ofs))
 271                 return 0;
 272         /*
 273          * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
 274          * All the super blocks will be iterated, including upper_sb.
 275          *
 276          * If this is a syncfs(2) call, then we do need to call
 277          * sync_filesystem() on upper_sb, but enough if we do it when being
 278          * called with wait == 1.
 279          */
 280         if (!wait)
 281                 return 0;
 282
 283         upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 284
 285         down_read(&upper_sb->s_umount);
 286         ret = sync_filesystem(upper_sb);
 287         up_read(&upper_sb->s_umount);
 288
 289         return ret;
 290 }
 291
 292 /**
 293  * ovl_statfs
 294  * @sb: The overlayfs super block
 295  * @buf: The struct kstatfs to fill in with stats
 296  *
 297  * Get the filesystem statistics.  As writes always target the upper layer
 298  * filesystem pass the statfs to the upper filesystem (if it exists)
 299  */
 300 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 301 {
 302         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 303         struct dentry *root_dentry = dentry->d_sb->s_root;
 304         struct path path;
 305         int err;
 306
 307         ovl_path_real(root_dentry, &path);
 308
 309         err = vfs_statfs(&path, buf);
 310         if (!err) {
 311                 buf->f_namelen = ofs->namelen;
 312                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 313         }
 314
 315         return err;
 316 }
 317
 318 /* Will this overlay be forced to mount/remount ro? */
 319 static bool ovl_force_readonly(struct ovl_fs *ofs)
 320 {
 321         return (!ovl_upper_mnt(ofs) || !ofs->workdir);
 322 }
 323
 324 static const char *ovl_redirect_mode_def(void)
 325 {
 326         return ovl_redirect_dir_def ? "on" : "off";
 327 }
 328
 329 static const char * const ovl_xino_str[] = {
 330         "off",
 331         "auto",
 332         "on",
 333 };
 334
 335 static inline int ovl_xino_def(void)
 336 {
 337         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 338 }
 339
 340 /**
 341  * ovl_show_options
 342  *
 343  * Prints the mount options for a given superblock.
 344  * Returns zero; does not fail.
 345  */
 346 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 347 {
 348         struct super_block *sb = dentry->d_sb;
 349         struct ovl_fs *ofs = sb->s_fs_info;
 350
 351         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 352         if (ofs->config.upperdir) {
 353                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 354                 seq_show_option(m, "workdir", ofs->config.workdir);
 355         }
 356         if (ofs->config.default_permissions)
 357                 seq_puts(m, ",default_permissions");
 358         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 359                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 360         if (ofs->config.index != ovl_index_def)
 361                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 362         if (!ofs->config.uuid)
 363                 seq_puts(m, ",uuid=off");
 364         if (ofs->config.nfs_export != ovl_nfs_export_def)
 365                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 366                                                 "on" : "off");
 367         if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
 368                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 369         if (ofs->config.metacopy != ovl_metacopy_def)
 370                 seq_printf(m, ",metacopy=%s",
 371                            ofs->config.metacopy ? "on" : "off");
 372         if (ofs->config.ovl_volatile)
 373                 seq_puts(m, ",volatile");
 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         struct super_block *upper_sb;
 381         int ret = 0;
 382
 383         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 384                 return -EROFS;
 385
 386         if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
 387                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 388                 if (ovl_should_sync(ofs)) {
 389                         down_read(&upper_sb->s_umount);
 390                         ret = sync_filesystem(upper_sb);
 391                         up_read(&upper_sb->s_umount);
 392                 }
 393         }
 394
 395         return ret;
 396 }
 397
 398 static const struct super_operations ovl_super_operations = {
 399         .alloc_inode    = ovl_alloc_inode,
 400         .free_inode     = ovl_free_inode,
 401         .destroy_inode  = ovl_destroy_inode,
 402         .drop_inode     = generic_delete_inode,
 403         .put_super      = ovl_put_super,
 404         .sync_fs        = ovl_sync_fs,
 405         .statfs         = ovl_statfs,
 406         .show_options   = ovl_show_options,
 407         .remount_fs     = ovl_remount,
 408 };
 409
 410 enum {
 411         OPT_LOWERDIR,
 412         OPT_UPPERDIR,
 413         OPT_WORKDIR,
 414         OPT_DEFAULT_PERMISSIONS,
 415         OPT_REDIRECT_DIR,
 416         OPT_INDEX_ON,
 417         OPT_INDEX_OFF,
 418         OPT_UUID_ON,
 419         OPT_UUID_OFF,
 420         OPT_NFS_EXPORT_ON,
 421         OPT_USERXATTR,
 422         OPT_NFS_EXPORT_OFF,
 423         OPT_XINO_ON,
 424         OPT_XINO_OFF,
 425         OPT_XINO_AUTO,
 426         OPT_METACOPY_ON,
 427         OPT_METACOPY_OFF,
 428         OPT_VOLATILE,
 429         OPT_ERR,
 430 };
 431
 432 static const match_table_t ovl_tokens = {
 433         {OPT_LOWERDIR,                  "lowerdir=%s"},
 434         {OPT_UPPERDIR,                  "upperdir=%s"},
 435         {OPT_WORKDIR,                   "workdir=%s"},
 436         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 437         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 438         {OPT_INDEX_ON,                  "index=on"},
 439         {OPT_INDEX_OFF,                 "index=off"},
 440         {OPT_USERXATTR,                 "userxattr"},
 441         {OPT_UUID_ON,                   "uuid=on"},
 442         {OPT_UUID_OFF,                  "uuid=off"},
 443         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 444         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 445         {OPT_XINO_ON,                   "xino=on"},
 446         {OPT_XINO_OFF,                  "xino=off"},
 447         {OPT_XINO_AUTO,                 "xino=auto"},
 448         {OPT_METACOPY_ON,               "metacopy=on"},
 449         {OPT_METACOPY_OFF,              "metacopy=off"},
 450         {OPT_VOLATILE,                  "volatile"},
 451         {OPT_ERR,                       NULL}
 452 };
 453
 454 static char *ovl_next_opt(char **s)
 455 {
 456         char *sbegin = *s;
 457         char *p;
 458
 459         if (sbegin == NULL)
 460                 return NULL;
 461
 462         for (p = sbegin; *p; p++) {
 463                 if (*p == '\\') {
 464                         p++;
 465                         if (!*p)
 466                                 break;
 467                 } else if (*p == ',') {
 468                         *p = '\0';
 469                         *s = p + 1;
 470                         return sbegin;
 471                 }
 472         }
 473         *s = NULL;
 474         return sbegin;
 475 }
 476
 477 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 478 {
 479         if (strcmp(mode, "on") == 0) {
 480                 config->redirect_dir = true;
 481                 /*
 482                  * Does not make sense to have redirect creation without
 483                  * redirect following.
 484                  */
 485                 config->redirect_follow = true;
 486         } else if (strcmp(mode, "follow") == 0) {
 487                 config->redirect_follow = true;
 488         } else if (strcmp(mode, "off") == 0) {
 489                 if (ovl_redirect_always_follow)
 490                         config->redirect_follow = true;
 491         } else if (strcmp(mode, "nofollow") != 0) {
 492                 pr_err("bad mount option \"redirect_dir=%s\"\n",
 493                        mode);
 494                 return -EINVAL;
 495         }
 496
 497         return 0;
 498 }
 499
 500 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 501 {
 502         char *p;
 503         int err;
 504         bool metacopy_opt = false, redirect_opt = false;
 505         bool nfs_export_opt = false, index_opt = false;
 506
 507         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 508         if (!config->redirect_mode)
 509                 return -ENOMEM;
 510
 511         while ((p = ovl_next_opt(&opt)) != NULL) {
 512                 int token;
 513                 substring_t args[MAX_OPT_ARGS];
 514
 515                 if (!*p)
 516                         continue;
 517
 518                 token = match_token(p, ovl_tokens, args);
 519                 switch (token) {
 520                 case OPT_UPPERDIR:
 521                         kfree(config->upperdir);
 522                         config->upperdir = match_strdup(&args[0]);
 523                         if (!config->upperdir)
 524                                 return -ENOMEM;
 525                         break;
 526
 527                 case OPT_LOWERDIR:
 528                         kfree(config->lowerdir);
 529                         config->lowerdir = match_strdup(&args[0]);
 530                         if (!config->lowerdir)
 531                                 return -ENOMEM;
 532                         break;
 533
 534                 case OPT_WORKDIR:
 535                         kfree(config->workdir);
 536                         config->workdir = match_strdup(&args[0]);
 537                         if (!config->workdir)
 538                                 return -ENOMEM;
 539                         break;
 540
 541                 case OPT_DEFAULT_PERMISSIONS:
 542                         config->default_permissions = true;
 543                         break;
 544
 545                 case OPT_REDIRECT_DIR:
 546                         kfree(config->redirect_mode);
 547                         config->redirect_mode = match_strdup(&args[0]);
 548                         if (!config->redirect_mode)
 549                                 return -ENOMEM;
 550                         redirect_opt = true;
 551                         break;
 552
 553                 case OPT_INDEX_ON:
 554                         config->index = true;
 555                         index_opt = true;
 556                         break;
 557
 558                 case OPT_INDEX_OFF:
 559                         config->index = false;
 560                         index_opt = true;
 561                         break;
 562
 563                 case OPT_UUID_ON:
 564                         config->uuid = true;
 565                         break;
 566
 567                 case OPT_UUID_OFF:
 568                         config->uuid = false;
 569                         break;
 570
 571                 case OPT_NFS_EXPORT_ON:
 572                         config->nfs_export = true;
 573                         nfs_export_opt = true;
 574                         break;
 575
 576                 case OPT_NFS_EXPORT_OFF:
 577                         config->nfs_export = false;
 578                         nfs_export_opt = true;
 579                         break;
 580
 581                 case OPT_XINO_ON:
 582                         config->xino = OVL_XINO_ON;
 583                         break;
 584
 585                 case OPT_XINO_OFF:
 586                         config->xino = OVL_XINO_OFF;
 587                         break;
 588
 589                 case OPT_XINO_AUTO:
 590                         config->xino = OVL_XINO_AUTO;
 591                         break;
 592
 593                 case OPT_METACOPY_ON:
 594                         config->metacopy = true;
 595                         metacopy_opt = true;
 596                         break;
 597
 598                 case OPT_METACOPY_OFF:
 599                         config->metacopy = false;
 600                         metacopy_opt = true;
 601                         break;
 602
 603                 case OPT_VOLATILE:
 604                         config->ovl_volatile = true;
 605                         break;
 606
 607                 case OPT_USERXATTR:
 608                         config->userxattr = true;
 609                         break;
 610
 611                 default:
 612                         pr_err("unrecognized mount option \"%s\" or missing value\n",
 613                                         p);
 614                         return -EINVAL;
 615                 }
 616         }
 617
 618         /* Workdir/index are useless in non-upper mount */
 619         if (!config->upperdir) {
 620                 if (config->workdir) {
 621                         pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 622                                 config->workdir);
 623                         kfree(config->workdir);
 624                         config->workdir = NULL;
 625                 }
 626                 if (config->index && index_opt) {
 627                         pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
 628                         index_opt = false;
 629                 }
 630                 config->index = false;
 631         }
 632
 633         if (!config->upperdir && config->ovl_volatile) {
 634                 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
 635                 config->ovl_volatile = false;
 636         }
 637
 638         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 639         if (err)
 640                 return err;
 641
 642         /*
 643          * This is to make the logic below simpler.  It doesn't make any other
 644          * difference, since config->redirect_dir is only used for upper.
 645          */
 646         if (!config->upperdir && config->redirect_follow)
 647                 config->redirect_dir = true;
 648
 649         /* Resolve metacopy -> redirect_dir dependency */
 650         if (config->metacopy && !config->redirect_dir) {
 651                 if (metacopy_opt && redirect_opt) {
 652                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
 653                                config->redirect_mode);
 654                         return -EINVAL;
 655                 }
 656                 if (redirect_opt) {
 657                         /*
 658                          * There was an explicit redirect_dir=... that resulted
 659                          * in this conflict.
 660                          */
 661                         pr_info("disabling metacopy due to redirect_dir=%s\n",
 662                                 config->redirect_mode);
 663                         config->metacopy = false;
 664                 } else {
 665                         /* Automatically enable redirect otherwise. */
 666                         config->redirect_follow = config->redirect_dir = true;
 667                 }
 668         }
 669
 670         /* Resolve nfs_export -> index dependency */
 671         if (config->nfs_export && !config->index) {
 672                 if (!config->upperdir && config->redirect_follow) {
 673                         pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
 674                         config->nfs_export = false;
 675                 } else if (nfs_export_opt && index_opt) {
 676                         pr_err("conflicting options: nfs_export=on,index=off\n");
 677                         return -EINVAL;
 678                 } else if (index_opt) {
 679                         /*
 680                          * There was an explicit index=off that resulted
 681                          * in this conflict.
 682                          */
 683                         pr_info("disabling nfs_export due to index=off\n");
 684                         config->nfs_export = false;
 685                 } else {
 686                         /* Automatically enable index otherwise. */
 687                         config->index = true;
 688                 }
 689         }
 690
 691         /* Resolve nfs_export -> !metacopy dependency */
 692         if (config->nfs_export && config->metacopy) {
 693                 if (nfs_export_opt && metacopy_opt) {
 694                         pr_err("conflicting options: nfs_export=on,metacopy=on\n");
 695                         return -EINVAL;
 696                 }
 697                 if (metacopy_opt) {
 698                         /*
 699                          * There was an explicit metacopy=on that resulted
 700                          * in this conflict.
 701                          */
 702                         pr_info("disabling nfs_export due to metacopy=on\n");
 703                         config->nfs_export = false;
 704                 } else {
 705                         /*
 706                          * There was an explicit nfs_export=on that resulted
 707                          * in this conflict.
 708                          */
 709                         pr_info("disabling metacopy due to nfs_export=on\n");
 710                         config->metacopy = false;
 711                 }
 712         }
 713
 714
 715         /* Resolve userxattr -> !redirect && !metacopy dependency */
 716         if (config->userxattr) {
 717                 if (config->redirect_follow && redirect_opt) {
 718                         pr_err("conflicting options: userxattr,redirect_dir=%s\n",
 719                                config->redirect_mode);
 720                         return -EINVAL;
 721                 }
 722                 if (config->metacopy && metacopy_opt) {
 723                         pr_err("conflicting options: userxattr,metacopy=on\n");
 724                         return -EINVAL;
 725                 }
 726                 /*
 727                  * Silently disable default setting of redirect and metacopy.
 728                  * This shall be the default in the future as well: these
 729                  * options must be explicitly enabled if used together with
 730                  * userxattr.
 731                  */
 732                 config->redirect_dir = config->redirect_follow = false;
 733                 config->metacopy = false;
 734         }
 735
 736         return 0;
 737 }
 738
 739 #define OVL_WORKDIR_NAME "work"
 740 #define OVL_INDEXDIR_NAME "index"
 741
 742 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 743                                          const char *name, bool persist)
 744 {
 745         struct inode *dir =  ofs->workbasedir->d_inode;
 746         struct vfsmount *mnt = ovl_upper_mnt(ofs);
 747         struct dentry *work;
 748         int err;
 749         bool retried = false;
 750
 751         inode_lock_nested(dir, I_MUTEX_PARENT);
 752 retry:
 753         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
 754
 755         if (!IS_ERR(work)) {
 756                 struct iattr attr = {
 757                         .ia_valid = ATTR_MODE,
 758                         .ia_mode = S_IFDIR | 0,
 759                 };
 760
 761                 if (work->d_inode) {
 762                         err = -EEXIST;
 763                         if (retried)
 764                                 goto out_dput;
 765
 766                         if (persist)
 767                                 goto out_unlock;
 768
 769                         retried = true;
 770                         err = ovl_workdir_cleanup(dir, mnt, work, 0);
 771                         dput(work);
 772                         if (err == -EINVAL) {
 773                                 work = ERR_PTR(err);
 774                                 goto out_unlock;
 775                         }
 776                         goto retry;
 777                 }
 778
 779                 work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
 780                 err = PTR_ERR(work);
 781                 if (IS_ERR(work))
 782                         goto out_err;
 783
 784                 /*
 785                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 786                  *
 787                  * a) success (there was a POSIX ACL xattr and was removed)
 788                  * b) -ENODATA (there was no POSIX ACL xattr)
 789                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 790                  *
 791                  * There are various other error values that could effectively
 792                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 793                  * if the xattr name is too long), but the set of filesystems
 794                  * allowed as upper are limited to "normal" ones, where checking
 795                  * for the above two errors is sufficient.
 796                  */
 797                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 798                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 799                         goto out_dput;
 800
 801                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 802                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 803                         goto out_dput;
 804
 805                 /* Clear any inherited mode bits */
 806                 inode_lock(work->d_inode);
 807                 err = notify_change(work, &attr, NULL);
 808                 inode_unlock(work->d_inode);
 809                 if (err)
 810                         goto out_dput;
 811         } else {
 812                 err = PTR_ERR(work);
 813                 goto out_err;
 814         }
 815 out_unlock:
 816         inode_unlock(dir);
 817         return work;
 818
 819 out_dput:
 820         dput(work);
 821 out_err:
 822         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
 823                 ofs->config.workdir, name, -err);
 824         work = NULL;
 825         goto out_unlock;
 826 }
 827
 828 static void ovl_unescape(char *s)
 829 {
 830         char *d = s;
 831
 832         for (;; s++, d++) {
 833                 if (*s == '\\')
 834                         s++;
 835                 *d = *s;
 836                 if (!*s)
 837                         break;
 838         }
 839 }
 840
 841 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 842 {
 843         int err = -EINVAL;
 844
 845         if (!*name) {
 846                 pr_err("empty lowerdir\n");
 847                 goto out;
 848         }
 849         err = kern_path(name, LOOKUP_FOLLOW, path);
 850         if (err) {
 851                 pr_err("failed to resolve '%s': %i\n", name, err);
 852                 goto out;
 853         }
 854         err = -EINVAL;
 855         if (ovl_dentry_weird(path->dentry)) {
 856                 pr_err("filesystem on '%s' not supported\n", name);
 857                 goto out_put;
 858         }
 859         if (!d_is_dir(path->dentry)) {
 860                 pr_err("'%s' not a directory\n", name);
 861                 goto out_put;
 862         }
 863         return 0;
 864
 865 out_put:
 866         path_put_init(path);
 867 out:
 868         return err;
 869 }
 870
 871 static int ovl_mount_dir(const char *name, struct path *path)
 872 {
 873         int err = -ENOMEM;
 874         char *tmp = kstrdup(name, GFP_KERNEL);
 875
 876         if (tmp) {
 877                 ovl_unescape(tmp);
 878                 err = ovl_mount_dir_noesc(tmp, path);
 879
 880                 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
 881                         pr_err("filesystem on '%s' not supported as upperdir\n",
 882                                tmp);
 883                         path_put_init(path);
 884                         err = -EINVAL;
 885                 }
 886                 kfree(tmp);
 887         }
 888         return err;
 889 }
 890
 891 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 892                              const char *name)
 893 {
 894         struct kstatfs statfs;
 895         int err = vfs_statfs(path, &statfs);
 896
 897         if (err)
 898                 pr_err("statfs failed on '%s'\n", name);
 899         else
 900                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 901
 902         return err;
 903 }
 904
 905 static int ovl_lower_dir(const char *name, struct path *path,
 906                          struct ovl_fs *ofs, int *stack_depth)
 907 {
 908         int fh_type;
 909         int err;
 910
 911         err = ovl_mount_dir_noesc(name, path);
 912         if (err)
 913                 return err;
 914
 915         err = ovl_check_namelen(path, ofs, name);
 916         if (err)
 917                 return err;
 918
 919         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 920
 921         /*
 922          * The inodes index feature and NFS export need to encode and decode
 923          * file handles, so they require that all layers support them.
 924          */
 925         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 926         if ((ofs->config.nfs_export ||
 927              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 928                 ofs->config.index = false;
 929                 ofs->config.nfs_export = false;
 930                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 931                         name);
 932         }
 933
 934         /* Check if lower fs has 32bit inode numbers */
 935         if (fh_type != FILEID_INO32_GEN)
 936                 ofs->xino_mode = -1;
 937
 938         return 0;
 939 }
 940
 941 /* Workdir should not be subdir of upperdir and vice versa */
 942 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 943 {
 944         bool ok = false;
 945
 946         if (workdir != upperdir) {
 947                 ok = (lock_rename(workdir, upperdir) == NULL);
 948                 unlock_rename(workdir, upperdir);
 949         }
 950         return ok;
 951 }
 952
 953 static unsigned int ovl_split_lowerdirs(char *str)
 954 {
 955         unsigned int ctr = 1;
 956         char *s, *d;
 957
 958         for (s = d = str;; s++, d++) {
 959                 if (*s == '\\') {
 960                         s++;
 961                 } else if (*s == ':') {
 962                         *d = '\0';
 963                         ctr++;
 964                         continue;
 965                 }
 966                 *d = *s;
 967                 if (!*s)
 968                         break;
 969         }
 970         return ctr;
 971 }
 972
 973 static int __maybe_unused
 974 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 975                         struct dentry *dentry, struct inode *inode,
 976                         const char *name, void *buffer, size_t size)
 977 {
 978         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 979 }
 980
 981 static int __maybe_unused
 982 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 983                         struct dentry *dentry, struct inode *inode,
 984                         const char *name, const void *value,
 985                         size_t size, int flags)
 986 {
 987         struct dentry *workdir = ovl_workdir(dentry);
 988         struct inode *realinode = ovl_inode_real(inode);
 989         struct posix_acl *acl = NULL;
 990         int err;
 991
 992         /* Check that everything is OK before copy-up */
 993         if (value) {
 994                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 995                 if (IS_ERR(acl))
 996                         return PTR_ERR(acl);
 997         }
 998         err = -EOPNOTSUPP;
 999         if (!IS_POSIXACL(d_inode(workdir)))
1000                 goto out_acl_release;
1001         if (!realinode->i_op->set_acl)
1002                 goto out_acl_release;
1003         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1004                 err = acl ? -EACCES : 0;
1005                 goto out_acl_release;
1006         }
1007         err = -EPERM;
1008         if (!inode_owner_or_capable(inode))
1009                 goto out_acl_release;
1010
1011         posix_acl_release(acl);
1012
1013         /*
1014          * Check if sgid bit needs to be cleared (actual setacl operation will
1015          * be done with mounter's capabilities and so that won't do it for us).
1016          */
1017         if (unlikely(inode->i_mode & S_ISGID) &&
1018             handler->flags == ACL_TYPE_ACCESS &&
1019             !in_group_p(inode->i_gid) &&
1020             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
1021                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
1022
1023                 err = ovl_setattr(dentry, &iattr);
1024                 if (err)
1025                         return err;
1026         }
1027
1028         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
1029         if (!err)
1030                 ovl_copyattr(ovl_inode_real(inode), inode);
1031
1032         return err;
1033
1034 out_acl_release:
1035         posix_acl_release(acl);
1036         return err;
1037 }
1038
1039 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1040                              struct dentry *dentry, struct inode *inode,
1041                              const char *name, void *buffer, size_t size)
1042 {
1043         return -EOPNOTSUPP;
1044 }
1045
1046 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1047                              struct dentry *dentry, struct inode *inode,
1048                              const char *name, const void *value,
1049                              size_t size, int flags)
1050 {
1051         return -EOPNOTSUPP;
1052 }
1053
1054 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1055                                struct dentry *dentry, struct inode *inode,
1056                                const char *name, void *buffer, size_t size)
1057 {
1058         return ovl_xattr_get(dentry, inode, name, buffer, size);
1059 }
1060
1061 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1062                                struct dentry *dentry, struct inode *inode,
1063                                const char *name, const void *value,
1064                                size_t size, int flags)
1065 {
1066         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1067 }
1068
1069 static const struct xattr_handler __maybe_unused
1070 ovl_posix_acl_access_xattr_handler = {
1071         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1072         .flags = ACL_TYPE_ACCESS,
1073         .get = ovl_posix_acl_xattr_get,
1074         .set = ovl_posix_acl_xattr_set,
1075 };
1076
1077 static const struct xattr_handler __maybe_unused
1078 ovl_posix_acl_default_xattr_handler = {
1079         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1080         .flags = ACL_TYPE_DEFAULT,
1081         .get = ovl_posix_acl_xattr_get,
1082         .set = ovl_posix_acl_xattr_set,
1083 };
1084
1085 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1086         .prefix = OVL_XATTR_TRUSTED_PREFIX,
1087         .get = ovl_own_xattr_get,
1088         .set = ovl_own_xattr_set,
1089 };
1090
1091 static const struct xattr_handler ovl_own_user_xattr_handler = {
1092         .prefix = OVL_XATTR_USER_PREFIX,
1093         .get = ovl_own_xattr_get,
1094         .set = ovl_own_xattr_set,
1095 };
1096
1097 static const struct xattr_handler ovl_other_xattr_handler = {
1098         .prefix = "", /* catch all */
1099         .get = ovl_other_xattr_get,
1100         .set = ovl_other_xattr_set,
1101 };
1102
1103 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1104 #ifdef CONFIG_FS_POSIX_ACL
1105         &ovl_posix_acl_access_xattr_handler,
1106         &ovl_posix_acl_default_xattr_handler,
1107 #endif
1108         &ovl_own_trusted_xattr_handler,
1109         &ovl_other_xattr_handler,
1110         NULL
1111 };
1112
1113 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1114 #ifdef CONFIG_FS_POSIX_ACL
1115         &ovl_posix_acl_access_xattr_handler,
1116         &ovl_posix_acl_default_xattr_handler,
1117 #endif
1118         &ovl_own_user_xattr_handler,
1119         &ovl_other_xattr_handler,
1120         NULL
1121 };
1122
1123 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1124                           struct inode **ptrap, const char *name)
1125 {
1126         struct inode *trap;
1127         int err;
1128
1129         trap = ovl_get_trap_inode(sb, dir);
1130         err = PTR_ERR_OR_ZERO(trap);
1131         if (err) {
1132                 if (err == -ELOOP)
1133                         pr_err("conflicting %s path\n", name);
1134                 return err;
1135         }
1136
1137         *ptrap = trap;
1138         return 0;
1139 }
1140
1141 /*
1142  * Determine how we treat concurrent use of upperdir/workdir based on the
1143  * index feature. This is papering over mount leaks of container runtimes,
1144  * for example, an old overlay mount is leaked and now its upperdir is
1145  * attempted to be used as a lower layer in a new overlay mount.
1146  */
1147 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1148 {
1149         if (ofs->config.index) {
1150                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1151                        name);
1152                 return -EBUSY;
1153         } else {
1154                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1155                         name);
1156                 return 0;
1157         }
1158 }
1159
1160 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1161                          struct ovl_layer *upper_layer, struct path *upperpath)
1162 {
1163         struct vfsmount *upper_mnt;
1164         int err;
1165
1166         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1167         if (err)
1168                 goto out;
1169
1170         /* Upper fs should not be r/o */
1171         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
1172                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1173                 err = -EINVAL;
1174                 goto out;
1175         }
1176
1177         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1178         if (err)
1179                 goto out;
1180
1181         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1182                              "upperdir");
1183         if (err)
1184                 goto out;
1185
1186         upper_mnt = clone_private_mount(upperpath);
1187         err = PTR_ERR(upper_mnt);
1188         if (IS_ERR(upper_mnt)) {
1189                 pr_err("failed to clone upperpath\n");
1190                 goto out;
1191         }
1192
1193         /* Don't inherit atime flags */
1194         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1195         upper_layer->mnt = upper_mnt;
1196         upper_layer->idx = 0;
1197         upper_layer->fsid = 0;
1198
1199         /*
1200          * Inherit SB_NOSEC flag from upperdir.
1201          *
1202          * This optimization changes behavior when a security related attribute
1203          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1204          * okay because we don't yet have guarantees in that case, but it will
1205          * need careful treatment once we want to honour changes to underlying
1206          * filesystems.
1207          */
1208         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1209                 sb->s_flags |= SB_NOSEC;
1210
1211         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1212                 ofs->upperdir_locked = true;
1213         } else {
1214                 err = ovl_report_in_use(ofs, "upperdir");
1215                 if (err)
1216                         goto out;
1217         }
1218
1219         err = 0;
1220 out:
1221         return err;
1222 }
1223
1224 /*
1225  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1226  * negative values if error is encountered.
1227  */
1228 static int ovl_check_rename_whiteout(struct dentry *workdir)
1229 {
1230         struct inode *dir = d_inode(workdir);
1231         struct dentry *temp;
1232         struct dentry *dest;
1233         struct dentry *whiteout;
1234         struct name_snapshot name;
1235         int err;
1236
1237         inode_lock_nested(dir, I_MUTEX_PARENT);
1238
1239         temp = ovl_create_temp(workdir, OVL_CATTR(S_IFREG | 0));
1240         err = PTR_ERR(temp);
1241         if (IS_ERR(temp))
1242                 goto out_unlock;
1243
1244         dest = ovl_lookup_temp(workdir);
1245         err = PTR_ERR(dest);
1246         if (IS_ERR(dest)) {
1247                 dput(temp);
1248                 goto out_unlock;
1249         }
1250
1251         /* Name is inline and stable - using snapshot as a copy helper */
1252         take_dentry_name_snapshot(&name, temp);
1253         err = ovl_do_rename(dir, temp, dir, dest, RENAME_WHITEOUT);
1254         if (err) {
1255                 if (err == -EINVAL)
1256                         err = 0;
1257                 goto cleanup_temp;
1258         }
1259
1260         whiteout = lookup_one_len(name.name.name, workdir, name.name.len);
1261         err = PTR_ERR(whiteout);
1262         if (IS_ERR(whiteout))
1263                 goto cleanup_temp;
1264
1265         err = ovl_is_whiteout(whiteout);
1266
1267         /* Best effort cleanup of whiteout and temp file */
1268         if (err)
1269                 ovl_cleanup(dir, whiteout);
1270         dput(whiteout);
1271
1272 cleanup_temp:
1273         ovl_cleanup(dir, temp);
1274         release_dentry_name_snapshot(&name);
1275         dput(temp);
1276         dput(dest);
1277
1278 out_unlock:
1279         inode_unlock(dir);
1280
1281         return err;
1282 }
1283
1284 static struct dentry *ovl_lookup_or_create(struct dentry *parent,
1285                                            const char *name, umode_t mode)
1286 {
1287         size_t len = strlen(name);
1288         struct dentry *child;
1289
1290         inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1291         child = lookup_one_len(name, parent, len);
1292         if (!IS_ERR(child) && !child->d_inode)
1293                 child = ovl_create_real(parent->d_inode, child,
1294                                         OVL_CATTR(mode));
1295         inode_unlock(parent->d_inode);
1296         dput(parent);
1297
1298         return child;
1299 }
1300
1301 /*
1302  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1303  * present.
1304  */
1305 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1306 {
1307         unsigned int ctr;
1308         struct dentry *d = dget(ofs->workbasedir);
1309         static const char *const volatile_path[] = {
1310                 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1311         };
1312         const char *const *name = volatile_path;
1313
1314         for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1315                 d = ovl_lookup_or_create(d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1316                 if (IS_ERR(d))
1317                         return PTR_ERR(d);
1318         }
1319         dput(d);
1320         return 0;
1321 }
1322
1323 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1324                             struct path *workpath)
1325 {
1326         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1327         struct dentry *temp, *workdir;
1328         bool rename_whiteout;
1329         bool d_type;
1330         int fh_type;
1331         int err;
1332
1333         err = mnt_want_write(mnt);
1334         if (err)
1335                 return err;
1336
1337         workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1338         err = PTR_ERR(workdir);
1339         if (IS_ERR_OR_NULL(workdir))
1340                 goto out;
1341
1342         ofs->workdir = workdir;
1343
1344         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1345         if (err)
1346                 goto out;
1347
1348         /*
1349          * Upper should support d_type, else whiteouts are visible.  Given
1350          * workdir and upper are on same fs, we can do iterate_dir() on
1351          * workdir. This check requires successful creation of workdir in
1352          * previous step.
1353          */
1354         err = ovl_check_d_type_supported(workpath);
1355         if (err < 0)
1356                 goto out;
1357
1358         d_type = err;
1359         if (!d_type)
1360                 pr_warn("upper fs needs to support d_type.\n");
1361
1362         /* Check if upper/work fs supports O_TMPFILE */
1363         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1364         ofs->tmpfile = !IS_ERR(temp);
1365         if (ofs->tmpfile)
1366                 dput(temp);
1367         else
1368                 pr_warn("upper fs does not support tmpfile.\n");
1369
1370
1371         /* Check if upper/work fs supports RENAME_WHITEOUT */
1372         err = ovl_check_rename_whiteout(ofs->workdir);
1373         if (err < 0)
1374                 goto out;
1375
1376         rename_whiteout = err;
1377         if (!rename_whiteout)
1378                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1379
1380         /*
1381          * Check if upper/work fs supports (trusted|user).overlay.* xattr
1382          */
1383         err = ovl_do_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1384         if (err) {
1385                 ofs->noxattr = true;
1386                 ofs->config.index = false;
1387                 ofs->config.metacopy = false;
1388                 pr_warn("upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1389                 err = 0;
1390         } else {
1391                 ovl_do_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1392         }
1393
1394         /*
1395          * We allowed sub-optimal upper fs configuration and don't want to break
1396          * users over kernel upgrade, but we never allowed remote upper fs, so
1397          * we can enforce strict requirements for remote upper fs.
1398          */
1399         if (ovl_dentry_remote(ofs->workdir) &&
1400             (!d_type || !rename_whiteout || ofs->noxattr)) {
1401                 pr_err("upper fs missing required features.\n");
1402                 err = -EINVAL;
1403                 goto out;
1404         }
1405
1406         /*
1407          * For volatile mount, create a incompat/volatile/dirty file to keep
1408          * track of it.
1409          */
1410         if (ofs->config.ovl_volatile) {
1411                 err = ovl_create_volatile_dirty(ofs);
1412                 if (err < 0) {
1413                         pr_err("Failed to create volatile/dirty file.\n");
1414                         goto out;
1415                 }
1416         }
1417
1418         /* Check if upper/work fs supports file handles */
1419         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1420         if (ofs->config.index && !fh_type) {
1421                 ofs->config.index = false;
1422                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1423         }
1424
1425         /* Check if upper fs has 32bit inode numbers */
1426         if (fh_type != FILEID_INO32_GEN)
1427                 ofs->xino_mode = -1;
1428
1429         /* NFS export of r/w mount depends on index */
1430         if (ofs->config.nfs_export && !ofs->config.index) {
1431                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1432                 ofs->config.nfs_export = false;
1433         }
1434 out:
1435         mnt_drop_write(mnt);
1436         return err;
1437 }
1438
1439 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1440                            struct path *upperpath)
1441 {
1442         int err;
1443         struct path workpath = { };
1444
1445         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1446         if (err)
1447                 goto out;
1448
1449         err = -EINVAL;
1450         if (upperpath->mnt != workpath.mnt) {
1451                 pr_err("workdir and upperdir must reside under the same mount\n");
1452                 goto out;
1453         }
1454         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1455                 pr_err("workdir and upperdir must be separate subtrees\n");
1456                 goto out;
1457         }
1458
1459         ofs->workbasedir = dget(workpath.dentry);
1460
1461         if (ovl_inuse_trylock(ofs->workbasedir)) {
1462                 ofs->workdir_locked = true;
1463         } else {
1464                 err = ovl_report_in_use(ofs, "workdir");
1465                 if (err)
1466                         goto out;
1467         }
1468
1469         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1470                              "workdir");
1471         if (err)
1472                 goto out;
1473
1474         err = ovl_make_workdir(sb, ofs, &workpath);
1475
1476 out:
1477         path_put(&workpath);
1478
1479         return err;
1480 }
1481
1482 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1483                             struct ovl_entry *oe, struct path *upperpath)
1484 {
1485         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1486         struct dentry *indexdir;
1487         int err;
1488
1489         err = mnt_want_write(mnt);
1490         if (err)
1491                 return err;
1492
1493         /* Verify lower root is upper root origin */
1494         err = ovl_verify_origin(ofs, upperpath->dentry,
1495                                 oe->lowerstack[0].dentry, true);
1496         if (err) {
1497                 pr_err("failed to verify upper root origin\n");
1498                 goto out;
1499         }
1500
1501         /* index dir will act also as workdir */
1502         iput(ofs->workdir_trap);
1503         ofs->workdir_trap = NULL;
1504         dput(ofs->workdir);
1505         ofs->workdir = NULL;
1506         indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1507         if (IS_ERR(indexdir)) {
1508                 err = PTR_ERR(indexdir);
1509         } else if (indexdir) {
1510                 ofs->indexdir = indexdir;
1511                 ofs->workdir = dget(indexdir);
1512
1513                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1514                                      "indexdir");
1515                 if (err)
1516                         goto out;
1517
1518                 /*
1519                  * Verify upper root is exclusively associated with index dir.
1520                  * Older kernels stored upper fh in ".overlay.origin"
1521                  * xattr. If that xattr exists, verify that it is a match to
1522                  * upper dir file handle. In any case, verify or set xattr
1523                  * ".overlay.upper" to indicate that index may have
1524                  * directory entries.
1525                  */
1526                 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1527                         err = ovl_verify_set_fh(ofs, ofs->indexdir,
1528                                                 OVL_XATTR_ORIGIN,
1529                                                 upperpath->dentry, true, false);
1530                         if (err)
1531                                 pr_err("failed to verify index dir 'origin' xattr\n");
1532                 }
1533                 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1534                                        true);
1535                 if (err)
1536                         pr_err("failed to verify index dir 'upper' xattr\n");
1537
1538                 /* Cleanup bad/stale/orphan index entries */
1539                 if (!err)
1540                         err = ovl_indexdir_cleanup(ofs);
1541         }
1542         if (err || !ofs->indexdir)
1543                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1544
1545 out:
1546         mnt_drop_write(mnt);
1547         return err;
1548 }
1549
1550 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1551 {
1552         unsigned int i;
1553
1554         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1555                 return true;
1556
1557         /*
1558          * We allow using single lower with null uuid for index and nfs_export
1559          * for example to support those features with single lower squashfs.
1560          * To avoid regressions in setups of overlay with re-formatted lower
1561          * squashfs, do not allow decoding origin with lower null uuid unless
1562          * user opted-in to one of the new features that require following the
1563          * lower inode of non-dir upper.
1564          */
1565         if (!ofs->config.index && !ofs->config.metacopy && !ofs->config.xino &&
1566             uuid_is_null(uuid))
1567                 return false;
1568
1569         for (i = 0; i < ofs->numfs; i++) {
1570                 /*
1571                  * We use uuid to associate an overlay lower file handle with a
1572                  * lower layer, so we can accept lower fs with null uuid as long
1573                  * as all lower layers with null uuid are on the same fs.
1574                  * if we detect multiple lower fs with the same uuid, we
1575                  * disable lower file handle decoding on all of them.
1576                  */
1577                 if (ofs->fs[i].is_lower &&
1578                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1579                         ofs->fs[i].bad_uuid = true;
1580                         return false;
1581                 }
1582         }
1583         return true;
1584 }
1585
1586 /* Get a unique fsid for the layer */
1587 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1588 {
1589         struct super_block *sb = path->mnt->mnt_sb;
1590         unsigned int i;
1591         dev_t dev;
1592         int err;
1593         bool bad_uuid = false;
1594
1595         for (i = 0; i < ofs->numfs; i++) {
1596                 if (ofs->fs[i].sb == sb)
1597                         return i;
1598         }
1599
1600         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1601                 bad_uuid = true;
1602                 if (ofs->config.index || ofs->config.nfs_export) {
1603                         ofs->config.index = false;
1604                         ofs->config.nfs_export = false;
1605                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to index=off,nfs_export=off.\n",
1606                                 uuid_is_null(&sb->s_uuid) ? "null" :
1607                                                             "conflicting",
1608                                 path->dentry);
1609                 }
1610         }
1611
1612         err = get_anon_bdev(&dev);
1613         if (err) {
1614                 pr_err("failed to get anonymous bdev for lowerpath\n");
1615                 return err;
1616         }
1617
1618         ofs->fs[ofs->numfs].sb = sb;
1619         ofs->fs[ofs->numfs].pseudo_dev = dev;
1620         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1621
1622         return ofs->numfs++;
1623 }
1624
1625 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1626                           struct path *stack, unsigned int numlower,
1627                           struct ovl_layer *layers)
1628 {
1629         int err;
1630         unsigned int i;
1631
1632         err = -ENOMEM;
1633         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1634         if (ofs->fs == NULL)
1635                 goto out;
1636
1637         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1638         ofs->numfs++;
1639
1640         /*
1641          * All lower layers that share the same fs as upper layer, use the same
1642          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1643          * only overlay to simplify ovl_fs_free().
1644          * is_lower will be set if upper fs is shared with a lower layer.
1645          */
1646         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1647         if (err) {
1648                 pr_err("failed to get anonymous bdev for upper fs\n");
1649                 goto out;
1650         }
1651
1652         if (ovl_upper_mnt(ofs)) {
1653                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1654                 ofs->fs[0].is_lower = false;
1655         }
1656
1657         for (i = 0; i < numlower; i++) {
1658                 struct vfsmount *mnt;
1659                 struct inode *trap;
1660                 int fsid;
1661
1662                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1663                 if (err < 0)
1664                         goto out;
1665
1666                 /*
1667                  * Check if lower root conflicts with this overlay layers before
1668                  * checking if it is in-use as upperdir/workdir of "another"
1669                  * mount, because we do not bother to check in ovl_is_inuse() if
1670                  * the upperdir/workdir is in fact in-use by our
1671                  * upperdir/workdir.
1672                  */
1673                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1674                 if (err)
1675                         goto out;
1676
1677                 if (ovl_is_inuse(stack[i].dentry)) {
1678                         err = ovl_report_in_use(ofs, "lowerdir");
1679                         if (err) {
1680                                 iput(trap);
1681                                 goto out;
1682                         }
1683                 }
1684
1685                 mnt = clone_private_mount(&stack[i]);
1686                 err = PTR_ERR(mnt);
1687                 if (IS_ERR(mnt)) {
1688                         pr_err("failed to clone lowerpath\n");
1689                         iput(trap);
1690                         goto out;
1691                 }
1692
1693                 /*
1694                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1695                  * will fail instead of modifying lower fs.
1696                  */
1697                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1698
1699                 layers[ofs->numlayer].trap = trap;
1700                 layers[ofs->numlayer].mnt = mnt;
1701                 layers[ofs->numlayer].idx = ofs->numlayer;
1702                 layers[ofs->numlayer].fsid = fsid;
1703                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1704                 ofs->numlayer++;
1705                 ofs->fs[fsid].is_lower = true;
1706         }
1707
1708         /*
1709          * When all layers on same fs, overlay can use real inode numbers.
1710          * With mount option "xino=<on|auto>", mounter declares that there are
1711          * enough free high bits in underlying fs to hold the unique fsid.
1712          * If overlayfs does encounter underlying inodes using the high xino
1713          * bits reserved for fsid, it emits a warning and uses the original
1714          * inode number or a non persistent inode number allocated from a
1715          * dedicated range.
1716          */
1717         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1718                 if (ofs->config.xino == OVL_XINO_ON)
1719                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1720                 ofs->xino_mode = 0;
1721         } else if (ofs->config.xino == OVL_XINO_OFF) {
1722                 ofs->xino_mode = -1;
1723         } else if (ofs->xino_mode < 0) {
1724                 /*
1725                  * This is a roundup of number of bits needed for encoding
1726                  * fsid, where fsid 0 is reserved for upper fs (even with
1727                  * lower only overlay) +1 extra bit is reserved for the non
1728                  * persistent inode number range that is used for resolving
1729                  * xino lower bits overflow.
1730                  */
1731                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1732                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1733         }
1734
1735         if (ofs->xino_mode > 0) {
1736                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1737                         ofs->xino_mode);
1738         }
1739
1740         err = 0;
1741 out:
1742         return err;
1743 }
1744
1745 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1746                                 const char *lower, unsigned int numlower,
1747                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1748 {
1749         int err;
1750         struct path *stack = NULL;
1751         unsigned int i;
1752         struct ovl_entry *oe;
1753
1754         if (!ofs->config.upperdir && numlower == 1) {
1755                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1756                 return ERR_PTR(-EINVAL);
1757         }
1758
1759         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1760         if (!stack)
1761                 return ERR_PTR(-ENOMEM);
1762
1763         err = -EINVAL;
1764         for (i = 0; i < numlower; i++) {
1765                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1766                 if (err)
1767                         goto out_err;
1768
1769                 lower = strchr(lower, '\0') + 1;
1770         }
1771
1772         err = -EINVAL;
1773         sb->s_stack_depth++;
1774         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1775                 pr_err("maximum fs stacking depth exceeded\n");
1776                 goto out_err;
1777         }
1778
1779         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1780         if (err)
1781                 goto out_err;
1782
1783         err = -ENOMEM;
1784         oe = ovl_alloc_entry(numlower);
1785         if (!oe)
1786                 goto out_err;
1787
1788         for (i = 0; i < numlower; i++) {
1789                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1790                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1791         }
1792
1793 out:
1794         for (i = 0; i < numlower; i++)
1795                 path_put(&stack[i]);
1796         kfree(stack);
1797
1798         return oe;
1799
1800 out_err:
1801         oe = ERR_PTR(err);
1802         goto out;
1803 }
1804
1805 /*
1806  * Check if this layer root is a descendant of:
1807  * - another layer of this overlayfs instance
1808  * - upper/work dir of any overlayfs instance
1809  */
1810 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1811                            struct dentry *dentry, const char *name)
1812 {
1813         struct dentry *next = dentry, *parent;
1814         int err = 0;
1815
1816         if (!dentry)
1817                 return 0;
1818
1819         parent = dget_parent(next);
1820
1821         /* Walk back ancestors to root (inclusive) looking for traps */
1822         while (!err && parent != next) {
1823                 if (ovl_lookup_trap_inode(sb, parent)) {
1824                         err = -ELOOP;
1825                         pr_err("overlapping %s path\n", name);
1826                 } else if (ovl_is_inuse(parent)) {
1827                         err = ovl_report_in_use(ofs, name);
1828                 }
1829                 next = parent;
1830                 parent = dget_parent(next);
1831                 dput(next);
1832         }
1833
1834         dput(parent);
1835
1836         return err;
1837 }
1838
1839 /*
1840  * Check if any of the layers or work dirs overlap.
1841  */
1842 static int ovl_check_overlapping_layers(struct super_block *sb,
1843                                         struct ovl_fs *ofs)
1844 {
1845         int i, err;
1846
1847         if (ovl_upper_mnt(ofs)) {
1848                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1849                                       "upperdir");
1850                 if (err)
1851                         return err;
1852
1853                 /*
1854                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1855                  * this instance and covers overlapping work and index dirs,
1856                  * unless work or index dir have been moved since created inside
1857                  * workbasedir.  In that case, we already have their traps in
1858                  * inode cache and we will catch that case on lookup.
1859                  */
1860                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir");
1861                 if (err)
1862                         return err;
1863         }
1864
1865         for (i = 1; i < ofs->numlayer; i++) {
1866                 err = ovl_check_layer(sb, ofs,
1867                                       ofs->layers[i].mnt->mnt_root,
1868                                       "lowerdir");
1869                 if (err)
1870                         return err;
1871         }
1872
1873         return 0;
1874 }
1875
1876 static struct dentry *ovl_get_root(struct super_block *sb,
1877                                    struct dentry *upperdentry,
1878                                    struct ovl_entry *oe)
1879 {
1880         struct dentry *root;
1881         struct ovl_path *lowerpath = &oe->lowerstack[0];
1882         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1883         int fsid = lowerpath->layer->fsid;
1884         struct ovl_inode_params oip = {
1885                 .upperdentry = upperdentry,
1886                 .lowerpath = lowerpath,
1887         };
1888
1889         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1890         if (!root)
1891                 return NULL;
1892
1893         root->d_fsdata = oe;
1894
1895         if (upperdentry) {
1896                 /* Root inode uses upper st_ino/i_ino */
1897                 ino = d_inode(upperdentry)->i_ino;
1898                 fsid = 0;
1899                 ovl_dentry_set_upper_alias(root);
1900                 if (ovl_is_impuredir(sb, upperdentry))
1901                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1902         }
1903
1904         /* Root is always merge -> can have whiteouts */
1905         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1906         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1907         ovl_set_upperdata(d_inode(root));
1908         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1909         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1910
1911         return root;
1912 }
1913
1914 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1915 {
1916         struct path upperpath = { };
1917         struct dentry *root_dentry;
1918         struct ovl_entry *oe;
1919         struct ovl_fs *ofs;
1920         struct ovl_layer *layers;
1921         struct cred *cred;
1922         char *splitlower = NULL;
1923         unsigned int numlower;
1924         int err;
1925
1926         sb->s_d_op = &ovl_dentry_operations;
1927
1928         err = -ENOMEM;
1929         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1930         if (!ofs)
1931                 goto out;
1932
1933         ofs->creator_cred = cred = prepare_creds();
1934         if (!cred)
1935                 goto out_err;
1936
1937         /* Is there a reason anyone would want not to share whiteouts? */
1938         ofs->share_whiteout = true;
1939
1940         ofs->config.index = ovl_index_def;
1941         ofs->config.uuid = true;
1942         ofs->config.nfs_export = ovl_nfs_export_def;
1943         ofs->config.xino = ovl_xino_def();
1944         ofs->config.metacopy = ovl_metacopy_def;
1945         err = ovl_parse_opt((char *) data, &ofs->config);
1946         if (err)
1947                 goto out_err;
1948
1949         err = -EINVAL;
1950         if (!ofs->config.lowerdir) {
1951                 if (!silent)
1952                         pr_err("missing 'lowerdir'\n");
1953                 goto out_err;
1954         }
1955
1956         err = -ENOMEM;
1957         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1958         if (!splitlower)
1959                 goto out_err;
1960
1961         numlower = ovl_split_lowerdirs(splitlower);
1962         if (numlower > OVL_MAX_STACK) {
1963                 pr_err("too many lower directories, limit is %d\n",
1964                        OVL_MAX_STACK);
1965                 goto out_err;
1966         }
1967
1968         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1969         if (!layers)
1970                 goto out_err;
1971
1972         ofs->layers = layers;
1973         /* Layer 0 is reserved for upper even if there's no upper */
1974         ofs->numlayer = 1;
1975
1976         sb->s_stack_depth = 0;
1977         sb->s_maxbytes = MAX_LFS_FILESIZE;
1978         atomic_long_set(&ofs->last_ino, 1);
1979         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1980         if (ofs->config.xino != OVL_XINO_OFF) {
1981                 ofs->xino_mode = BITS_PER_LONG - 32;
1982                 if (!ofs->xino_mode) {
1983                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1984                         ofs->config.xino = OVL_XINO_OFF;
1985                 }
1986         }
1987
1988         /* alloc/destroy_inode needed for setting up traps in inode cache */
1989         sb->s_op = &ovl_super_operations;
1990
1991         if (ofs->config.upperdir) {
1992                 if (!ofs->config.workdir) {
1993                         pr_err("missing 'workdir'\n");
1994                         goto out_err;
1995                 }
1996
1997                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
1998                 if (err)
1999                         goto out_err;
2000
2001                 err = ovl_get_workdir(sb, ofs, &upperpath);
2002                 if (err)
2003                         goto out_err;
2004
2005                 if (!ofs->workdir)
2006                         sb->s_flags |= SB_RDONLY;
2007
2008                 sb->s_stack_depth = ovl_upper_mnt(ofs)->mnt_sb->s_stack_depth;
2009                 sb->s_time_gran = ovl_upper_mnt(ofs)->mnt_sb->s_time_gran;
2010
2011         }
2012         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2013         err = PTR_ERR(oe);
2014         if (IS_ERR(oe))
2015                 goto out_err;
2016
2017         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2018         if (!ovl_upper_mnt(ofs))
2019                 sb->s_flags |= SB_RDONLY;
2020
2021         if (!ofs->config.uuid && ofs->numfs > 1) {
2022                 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2023                 ofs->config.uuid = true;
2024         }
2025
2026         if (!ovl_force_readonly(ofs) && ofs->config.index) {
2027                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2028                 if (err)
2029                         goto out_free_oe;
2030
2031                 /* Force r/o mount with no index dir */
2032                 if (!ofs->indexdir)
2033                         sb->s_flags |= SB_RDONLY;
2034         }
2035
2036         err = ovl_check_overlapping_layers(sb, ofs);
2037         if (err)
2038                 goto out_free_oe;
2039
2040         /* Show index=off in /proc/mounts for forced r/o mount */
2041         if (!ofs->indexdir) {
2042                 ofs->config.index = false;
2043                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2044                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2045                         ofs->config.nfs_export = false;
2046                 }
2047         }
2048
2049         if (ofs->config.metacopy && ofs->config.nfs_export) {
2050                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2051                 ofs->config.nfs_export = false;
2052         }
2053
2054         if (ofs->config.nfs_export)
2055                 sb->s_export_op = &ovl_export_operations;
2056
2057         /* Never override disk quota limits or use reserved space */
2058         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2059
2060         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2061         sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2062                 ovl_trusted_xattr_handlers;
2063         sb->s_fs_info = ofs;
2064         sb->s_flags |= SB_POSIXACL;
2065         sb->s_iflags |= SB_I_SKIP_SYNC;
2066
2067         err = -ENOMEM;
2068         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2069         if (!root_dentry)
2070                 goto out_free_oe;
2071
2072         mntput(upperpath.mnt);
2073         kfree(splitlower);
2074
2075         sb->s_root = root_dentry;
2076
2077         return 0;
2078
2079 out_free_oe:
2080         ovl_entry_stack_free(oe);
2081         kfree(oe);
2082 out_err:
2083         kfree(splitlower);
2084         path_put(&upperpath);
2085         ovl_free_fs(ofs);
2086 out:
2087         return err;
2088 }
2089
2090 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2091                                 const char *dev_name, void *raw_data)
2092 {
2093         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2094 }
2095
2096 static struct file_system_type ovl_fs_type = {
2097         .owner          = THIS_MODULE,
2098         .name           = "overlay",
2099         .fs_flags       = FS_USERNS_MOUNT,
2100         .mount          = ovl_mount,
2101         .kill_sb        = kill_anon_super,
2102 };
2103 MODULE_ALIAS_FS("overlay");
2104
2105 static void ovl_inode_init_once(void *foo)
2106 {
2107         struct ovl_inode *oi = foo;
2108
2109         inode_init_once(&oi->vfs_inode);
2110 }
2111
2112 static int __init ovl_init(void)
2113 {
2114         int err;
2115
2116         ovl_inode_cachep = kmem_cache_create("ovl_inode",
2117                                              sizeof(struct ovl_inode), 0,
2118                                              (SLAB_RECLAIM_ACCOUNT|
2119                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2120                                              ovl_inode_init_once);
2121         if (ovl_inode_cachep == NULL)
2122                 return -ENOMEM;
2123
2124         err = ovl_aio_request_cache_init();
2125         if (!err) {
2126                 err = register_filesystem(&ovl_fs_type);
2127                 if (!err)
2128                         return 0;
2129
2130                 ovl_aio_request_cache_destroy();
2131         }
2132         kmem_cache_destroy(ovl_inode_cachep);
2133
2134         return err;
2135 }
2136
2137 static void __exit ovl_exit(void)
2138 {
2139         unregister_filesystem(&ovl_fs_type);
2140
2141         /*
2142          * Make sure all delayed rcu free inodes are flushed before we
2143          * destroy cache.
2144          */
2145         rcu_barrier();
2146         kmem_cache_destroy(ovl_inode_cachep);
2147         ovl_aio_request_cache_destroy();
2148 }
2149
2150 module_init(ovl_init);
2151 module_exit(ovl_exit);