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 <linux/file.h>
  19 #include "overlayfs.h"
  20
  21 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  22 MODULE_DESCRIPTION("Overlay filesystem");
  23 MODULE_LICENSE("GPL");
  24
  25
  26 struct ovl_dir_cache;
  27
  28 #define OVL_MAX_STACK 500
  29
  30 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  31 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  32 MODULE_PARM_DESC(redirect_dir,
  33                  "Default to on or off for the redirect_dir feature");
  34
  35 static bool ovl_redirect_always_follow =
  36         IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
  37 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
  38                    bool, 0644);
  39 MODULE_PARM_DESC(redirect_always_follow,
  40                  "Follow redirects even if redirect_dir feature is turned off");
  41
  42 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
  43 module_param_named(index, ovl_index_def, bool, 0644);
  44 MODULE_PARM_DESC(index,
  45                  "Default to on or off for the inodes index feature");
  46
  47 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
  48 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
  49 MODULE_PARM_DESC(nfs_export,
  50                  "Default to on or off for the NFS export feature");
  51
  52 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
  53 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
  54 MODULE_PARM_DESC(xino_auto,
  55                  "Auto enable xino feature");
  56
  57 static void ovl_entry_stack_free(struct ovl_entry *oe)
  58 {
  59         unsigned int i;
  60
  61         for (i = 0; i < oe->numlower; i++)
  62                 dput(oe->lowerstack[i].dentry);
  63 }
  64
  65 static bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
  66 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
  67 MODULE_PARM_DESC(metacopy,
  68                  "Default to on or off for the metadata only copy up feature");
  69
  70 static void ovl_dentry_release(struct dentry *dentry)
  71 {
  72         struct ovl_entry *oe = dentry->d_fsdata;
  73
  74         if (oe) {
  75                 ovl_entry_stack_free(oe);
  76                 kfree_rcu(oe, rcu);
  77         }
  78 }
  79
  80 static struct dentry *ovl_d_real(struct dentry *dentry,
  81                                  const struct inode *inode)
  82 {
  83         struct dentry *real = NULL, *lower;
  84
  85         /* It's an overlay file */
  86         if (inode && d_inode(dentry) == inode)
  87                 return dentry;
  88
  89         if (!d_is_reg(dentry)) {
  90                 if (!inode || inode == d_inode(dentry))
  91                         return dentry;
  92                 goto bug;
  93         }
  94
  95         real = ovl_dentry_upper(dentry);
  96         if (real && (inode == d_inode(real)))
  97                 return real;
  98
  99         if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
 100                 return real;
 101
 102         lower = ovl_dentry_lowerdata(dentry);
 103         if (!lower)
 104                 goto bug;
 105         real = lower;
 106
 107         /* Handle recursion */
 108         real = d_real(real, inode);
 109
 110         if (!inode || inode == d_inode(real))
 111                 return real;
 112 bug:
 113         WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
 114              __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
 115              inode ? inode->i_ino : 0, real,
 116              real && d_inode(real) ? d_inode(real)->i_ino : 0);
 117         return dentry;
 118 }
 119
 120 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
 121 {
 122         int ret = 1;
 123
 124         if (weak) {
 125                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
 126                         ret =  d->d_op->d_weak_revalidate(d, flags);
 127         } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
 128                 ret = d->d_op->d_revalidate(d, flags);
 129                 if (!ret) {
 130                         if (!(flags & LOOKUP_RCU))
 131                                 d_invalidate(d);
 132                         ret = -ESTALE;
 133                 }
 134         }
 135         return ret;
 136 }
 137
 138 static int ovl_dentry_revalidate_common(struct dentry *dentry,
 139                                         unsigned int flags, bool weak)
 140 {
 141         struct ovl_entry *oe = dentry->d_fsdata;
 142         struct inode *inode = d_inode_rcu(dentry);
 143         struct dentry *upper;
 144         unsigned int i;
 145         int ret = 1;
 146
 147         /* Careful in RCU mode */
 148         if (!inode)
 149                 return -ECHILD;
 150
 151         upper = ovl_i_dentry_upper(inode);
 152         if (upper)
 153                 ret = ovl_revalidate_real(upper, flags, weak);
 154
 155         for (i = 0; ret > 0 && i < oe->numlower; i++) {
 156                 ret = ovl_revalidate_real(oe->lowerstack[i].dentry, flags,
 157                                           weak);
 158         }
 159         return ret;
 160 }
 161
 162 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 163 {
 164         return ovl_dentry_revalidate_common(dentry, flags, false);
 165 }
 166
 167 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 168 {
 169         return ovl_dentry_revalidate_common(dentry, flags, true);
 170 }
 171
 172 static const struct dentry_operations ovl_dentry_operations = {
 173         .d_release = ovl_dentry_release,
 174         .d_real = ovl_d_real,
 175         .d_revalidate = ovl_dentry_revalidate,
 176         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 177 };
 178
 179 static struct kmem_cache *ovl_inode_cachep;
 180
 181 static struct inode *ovl_alloc_inode(struct super_block *sb)
 182 {
 183         struct ovl_inode *oi = alloc_inode_sb(sb, ovl_inode_cachep, GFP_KERNEL);
 184
 185         if (!oi)
 186                 return NULL;
 187
 188         oi->cache = NULL;
 189         oi->redirect = NULL;
 190         oi->version = 0;
 191         oi->flags = 0;
 192         oi->__upperdentry = NULL;
 193         oi->lowerpath.dentry = NULL;
 194         oi->lowerpath.layer = NULL;
 195         oi->lowerdata = NULL;
 196         mutex_init(&oi->lock);
 197
 198         return &oi->vfs_inode;
 199 }
 200
 201 static void ovl_free_inode(struct inode *inode)
 202 {
 203         struct ovl_inode *oi = OVL_I(inode);
 204
 205         kfree(oi->redirect);
 206         mutex_destroy(&oi->lock);
 207         kmem_cache_free(ovl_inode_cachep, oi);
 208 }
 209
 210 static void ovl_destroy_inode(struct inode *inode)
 211 {
 212         struct ovl_inode *oi = OVL_I(inode);
 213
 214         dput(oi->__upperdentry);
 215         dput(oi->lowerpath.dentry);
 216         if (S_ISDIR(inode->i_mode))
 217                 ovl_dir_cache_free(inode);
 218         else
 219                 iput(oi->lowerdata);
 220 }
 221
 222 static void ovl_free_fs(struct ovl_fs *ofs)
 223 {
 224         struct vfsmount **mounts;
 225         unsigned i;
 226
 227         iput(ofs->workbasedir_trap);
 228         iput(ofs->indexdir_trap);
 229         iput(ofs->workdir_trap);
 230         dput(ofs->whiteout);
 231         dput(ofs->indexdir);
 232         dput(ofs->workdir);
 233         if (ofs->workdir_locked)
 234                 ovl_inuse_unlock(ofs->workbasedir);
 235         dput(ofs->workbasedir);
 236         if (ofs->upperdir_locked)
 237                 ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
 238
 239         /* Hack!  Reuse ofs->layers as a vfsmount array before freeing it */
 240         mounts = (struct vfsmount **) ofs->layers;
 241         for (i = 0; i < ofs->numlayer; i++) {
 242                 iput(ofs->layers[i].trap);
 243                 mounts[i] = ofs->layers[i].mnt;
 244         }
 245         kern_unmount_array(mounts, ofs->numlayer);
 246         kfree(ofs->layers);
 247         for (i = 0; i < ofs->numfs; i++)
 248                 free_anon_bdev(ofs->fs[i].pseudo_dev);
 249         kfree(ofs->fs);
 250
 251         kfree(ofs->config.lowerdir);
 252         kfree(ofs->config.upperdir);
 253         kfree(ofs->config.workdir);
 254         kfree(ofs->config.redirect_mode);
 255         if (ofs->creator_cred)
 256                 put_cred(ofs->creator_cred);
 257         kfree(ofs);
 258 }
 259
 260 static void ovl_put_super(struct super_block *sb)
 261 {
 262         struct ovl_fs *ofs = sb->s_fs_info;
 263
 264         ovl_free_fs(ofs);
 265 }
 266
 267 /* Sync real dirty inodes in upper filesystem (if it exists) */
 268 static int ovl_sync_fs(struct super_block *sb, int wait)
 269 {
 270         struct ovl_fs *ofs = sb->s_fs_info;
 271         struct super_block *upper_sb;
 272         int ret;
 273
 274         ret = ovl_sync_status(ofs);
 275         /*
 276          * We have to always set the err, because the return value isn't
 277          * checked in syncfs, and instead indirectly return an error via
 278          * the sb's writeback errseq, which VFS inspects after this call.
 279          */
 280         if (ret < 0) {
 281                 errseq_set(&sb->s_wb_err, -EIO);
 282                 return -EIO;
 283         }
 284
 285         if (!ret)
 286                 return ret;
 287
 288         /*
 289          * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
 290          * All the super blocks will be iterated, including upper_sb.
 291          *
 292          * If this is a syncfs(2) call, then we do need to call
 293          * sync_filesystem() on upper_sb, but enough if we do it when being
 294          * called with wait == 1.
 295          */
 296         if (!wait)
 297                 return 0;
 298
 299         upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 300
 301         down_read(&upper_sb->s_umount);
 302         ret = sync_filesystem(upper_sb);
 303         up_read(&upper_sb->s_umount);
 304
 305         return ret;
 306 }
 307
 308 /**
 309  * ovl_statfs
 310  * @dentry: The dentry to query
 311  * @buf: The struct kstatfs to fill in with stats
 312  *
 313  * Get the filesystem statistics.  As writes always target the upper layer
 314  * filesystem pass the statfs to the upper filesystem (if it exists)
 315  */
 316 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 317 {
 318         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 319         struct dentry *root_dentry = dentry->d_sb->s_root;
 320         struct path path;
 321         int err;
 322
 323         ovl_path_real(root_dentry, &path);
 324
 325         err = vfs_statfs(&path, buf);
 326         if (!err) {
 327                 buf->f_namelen = ofs->namelen;
 328                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 329         }
 330
 331         return err;
 332 }
 333
 334 /* Will this overlay be forced to mount/remount ro? */
 335 static bool ovl_force_readonly(struct ovl_fs *ofs)
 336 {
 337         return (!ovl_upper_mnt(ofs) || !ofs->workdir);
 338 }
 339
 340 static const char *ovl_redirect_mode_def(void)
 341 {
 342         return ovl_redirect_dir_def ? "on" : "off";
 343 }
 344
 345 static const char * const ovl_xino_str[] = {
 346         "off",
 347         "auto",
 348         "on",
 349 };
 350
 351 static inline int ovl_xino_def(void)
 352 {
 353         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 354 }
 355
 356 /**
 357  * ovl_show_options
 358  * @m: the seq_file handle
 359  * @dentry: The dentry to query
 360  *
 361  * Prints the mount options for a given superblock.
 362  * Returns zero; does not fail.
 363  */
 364 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 365 {
 366         struct super_block *sb = dentry->d_sb;
 367         struct ovl_fs *ofs = sb->s_fs_info;
 368
 369         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 370         if (ofs->config.upperdir) {
 371                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 372                 seq_show_option(m, "workdir", ofs->config.workdir);
 373         }
 374         if (ofs->config.default_permissions)
 375                 seq_puts(m, ",default_permissions");
 376         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 377                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 378         if (ofs->config.index != ovl_index_def)
 379                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 380         if (!ofs->config.uuid)
 381                 seq_puts(m, ",uuid=off");
 382         if (ofs->config.nfs_export != ovl_nfs_export_def)
 383                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 384                                                 "on" : "off");
 385         if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
 386                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 387         if (ofs->config.metacopy != ovl_metacopy_def)
 388                 seq_printf(m, ",metacopy=%s",
 389                            ofs->config.metacopy ? "on" : "off");
 390         if (ofs->config.ovl_volatile)
 391                 seq_puts(m, ",volatile");
 392         if (ofs->config.userxattr)
 393                 seq_puts(m, ",userxattr");
 394         return 0;
 395 }
 396
 397 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 398 {
 399         struct ovl_fs *ofs = sb->s_fs_info;
 400         struct super_block *upper_sb;
 401         int ret = 0;
 402
 403         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 404                 return -EROFS;
 405
 406         if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
 407                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 408                 if (ovl_should_sync(ofs)) {
 409                         down_read(&upper_sb->s_umount);
 410                         ret = sync_filesystem(upper_sb);
 411                         up_read(&upper_sb->s_umount);
 412                 }
 413         }
 414
 415         return ret;
 416 }
 417
 418 static const struct super_operations ovl_super_operations = {
 419         .alloc_inode    = ovl_alloc_inode,
 420         .free_inode     = ovl_free_inode,
 421         .destroy_inode  = ovl_destroy_inode,
 422         .drop_inode     = generic_delete_inode,
 423         .put_super      = ovl_put_super,
 424         .sync_fs        = ovl_sync_fs,
 425         .statfs         = ovl_statfs,
 426         .show_options   = ovl_show_options,
 427         .remount_fs     = ovl_remount,
 428 };
 429
 430 enum {
 431         OPT_LOWERDIR,
 432         OPT_UPPERDIR,
 433         OPT_WORKDIR,
 434         OPT_DEFAULT_PERMISSIONS,
 435         OPT_REDIRECT_DIR,
 436         OPT_INDEX_ON,
 437         OPT_INDEX_OFF,
 438         OPT_UUID_ON,
 439         OPT_UUID_OFF,
 440         OPT_NFS_EXPORT_ON,
 441         OPT_USERXATTR,
 442         OPT_NFS_EXPORT_OFF,
 443         OPT_XINO_ON,
 444         OPT_XINO_OFF,
 445         OPT_XINO_AUTO,
 446         OPT_METACOPY_ON,
 447         OPT_METACOPY_OFF,
 448         OPT_VOLATILE,
 449         OPT_ERR,
 450 };
 451
 452 static const match_table_t ovl_tokens = {
 453         {OPT_LOWERDIR,                  "lowerdir=%s"},
 454         {OPT_UPPERDIR,                  "upperdir=%s"},
 455         {OPT_WORKDIR,                   "workdir=%s"},
 456         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 457         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 458         {OPT_INDEX_ON,                  "index=on"},
 459         {OPT_INDEX_OFF,                 "index=off"},
 460         {OPT_USERXATTR,                 "userxattr"},
 461         {OPT_UUID_ON,                   "uuid=on"},
 462         {OPT_UUID_OFF,                  "uuid=off"},
 463         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 464         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 465         {OPT_XINO_ON,                   "xino=on"},
 466         {OPT_XINO_OFF,                  "xino=off"},
 467         {OPT_XINO_AUTO,                 "xino=auto"},
 468         {OPT_METACOPY_ON,               "metacopy=on"},
 469         {OPT_METACOPY_OFF,              "metacopy=off"},
 470         {OPT_VOLATILE,                  "volatile"},
 471         {OPT_ERR,                       NULL}
 472 };
 473
 474 static char *ovl_next_opt(char **s)
 475 {
 476         char *sbegin = *s;
 477         char *p;
 478
 479         if (sbegin == NULL)
 480                 return NULL;
 481
 482         for (p = sbegin; *p; p++) {
 483                 if (*p == '\\') {
 484                         p++;
 485                         if (!*p)
 486                                 break;
 487                 } else if (*p == ',') {
 488                         *p = '\0';
 489                         *s = p + 1;
 490                         return sbegin;
 491                 }
 492         }
 493         *s = NULL;
 494         return sbegin;
 495 }
 496
 497 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 498 {
 499         if (strcmp(mode, "on") == 0) {
 500                 config->redirect_dir = true;
 501                 /*
 502                  * Does not make sense to have redirect creation without
 503                  * redirect following.
 504                  */
 505                 config->redirect_follow = true;
 506         } else if (strcmp(mode, "follow") == 0) {
 507                 config->redirect_follow = true;
 508         } else if (strcmp(mode, "off") == 0) {
 509                 if (ovl_redirect_always_follow)
 510                         config->redirect_follow = true;
 511         } else if (strcmp(mode, "nofollow") != 0) {
 512                 pr_err("bad mount option \"redirect_dir=%s\"\n",
 513                        mode);
 514                 return -EINVAL;
 515         }
 516
 517         return 0;
 518 }
 519
 520 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 521 {
 522         char *p;
 523         int err;
 524         bool metacopy_opt = false, redirect_opt = false;
 525         bool nfs_export_opt = false, index_opt = false;
 526
 527         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 528         if (!config->redirect_mode)
 529                 return -ENOMEM;
 530
 531         while ((p = ovl_next_opt(&opt)) != NULL) {
 532                 int token;
 533                 substring_t args[MAX_OPT_ARGS];
 534
 535                 if (!*p)
 536                         continue;
 537
 538                 token = match_token(p, ovl_tokens, args);
 539                 switch (token) {
 540                 case OPT_UPPERDIR:
 541                         kfree(config->upperdir);
 542                         config->upperdir = match_strdup(&args[0]);
 543                         if (!config->upperdir)
 544                                 return -ENOMEM;
 545                         break;
 546
 547                 case OPT_LOWERDIR:
 548                         kfree(config->lowerdir);
 549                         config->lowerdir = match_strdup(&args[0]);
 550                         if (!config->lowerdir)
 551                                 return -ENOMEM;
 552                         break;
 553
 554                 case OPT_WORKDIR:
 555                         kfree(config->workdir);
 556                         config->workdir = match_strdup(&args[0]);
 557                         if (!config->workdir)
 558                                 return -ENOMEM;
 559                         break;
 560
 561                 case OPT_DEFAULT_PERMISSIONS:
 562                         config->default_permissions = true;
 563                         break;
 564
 565                 case OPT_REDIRECT_DIR:
 566                         kfree(config->redirect_mode);
 567                         config->redirect_mode = match_strdup(&args[0]);
 568                         if (!config->redirect_mode)
 569                                 return -ENOMEM;
 570                         redirect_opt = true;
 571                         break;
 572
 573                 case OPT_INDEX_ON:
 574                         config->index = true;
 575                         index_opt = true;
 576                         break;
 577
 578                 case OPT_INDEX_OFF:
 579                         config->index = false;
 580                         index_opt = true;
 581                         break;
 582
 583                 case OPT_UUID_ON:
 584                         config->uuid = true;
 585                         break;
 586
 587                 case OPT_UUID_OFF:
 588                         config->uuid = false;
 589                         break;
 590
 591                 case OPT_NFS_EXPORT_ON:
 592                         config->nfs_export = true;
 593                         nfs_export_opt = true;
 594                         break;
 595
 596                 case OPT_NFS_EXPORT_OFF:
 597                         config->nfs_export = false;
 598                         nfs_export_opt = true;
 599                         break;
 600
 601                 case OPT_XINO_ON:
 602                         config->xino = OVL_XINO_ON;
 603                         break;
 604
 605                 case OPT_XINO_OFF:
 606                         config->xino = OVL_XINO_OFF;
 607                         break;
 608
 609                 case OPT_XINO_AUTO:
 610                         config->xino = OVL_XINO_AUTO;
 611                         break;
 612
 613                 case OPT_METACOPY_ON:
 614                         config->metacopy = true;
 615                         metacopy_opt = true;
 616                         break;
 617
 618                 case OPT_METACOPY_OFF:
 619                         config->metacopy = false;
 620                         metacopy_opt = true;
 621                         break;
 622
 623                 case OPT_VOLATILE:
 624                         config->ovl_volatile = true;
 625                         break;
 626
 627                 case OPT_USERXATTR:
 628                         config->userxattr = true;
 629                         break;
 630
 631                 default:
 632                         pr_err("unrecognized mount option \"%s\" or missing value\n",
 633                                         p);
 634                         return -EINVAL;
 635                 }
 636         }
 637
 638         /* Workdir/index are useless in non-upper mount */
 639         if (!config->upperdir) {
 640                 if (config->workdir) {
 641                         pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 642                                 config->workdir);
 643                         kfree(config->workdir);
 644                         config->workdir = NULL;
 645                 }
 646                 if (config->index && index_opt) {
 647                         pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
 648                         index_opt = false;
 649                 }
 650                 config->index = false;
 651         }
 652
 653         if (!config->upperdir && config->ovl_volatile) {
 654                 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
 655                 config->ovl_volatile = false;
 656         }
 657
 658         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 659         if (err)
 660                 return err;
 661
 662         /*
 663          * This is to make the logic below simpler.  It doesn't make any other
 664          * difference, since config->redirect_dir is only used for upper.
 665          */
 666         if (!config->upperdir && config->redirect_follow)
 667                 config->redirect_dir = true;
 668
 669         /* Resolve metacopy -> redirect_dir dependency */
 670         if (config->metacopy && !config->redirect_dir) {
 671                 if (metacopy_opt && redirect_opt) {
 672                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
 673                                config->redirect_mode);
 674                         return -EINVAL;
 675                 }
 676                 if (redirect_opt) {
 677                         /*
 678                          * There was an explicit redirect_dir=... that resulted
 679                          * in this conflict.
 680                          */
 681                         pr_info("disabling metacopy due to redirect_dir=%s\n",
 682                                 config->redirect_mode);
 683                         config->metacopy = false;
 684                 } else {
 685                         /* Automatically enable redirect otherwise. */
 686                         config->redirect_follow = config->redirect_dir = true;
 687                 }
 688         }
 689
 690         /* Resolve nfs_export -> index dependency */
 691         if (config->nfs_export && !config->index) {
 692                 if (!config->upperdir && config->redirect_follow) {
 693                         pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
 694                         config->nfs_export = false;
 695                 } else if (nfs_export_opt && index_opt) {
 696                         pr_err("conflicting options: nfs_export=on,index=off\n");
 697                         return -EINVAL;
 698                 } else if (index_opt) {
 699                         /*
 700                          * There was an explicit index=off that resulted
 701                          * in this conflict.
 702                          */
 703                         pr_info("disabling nfs_export due to index=off\n");
 704                         config->nfs_export = false;
 705                 } else {
 706                         /* Automatically enable index otherwise. */
 707                         config->index = true;
 708                 }
 709         }
 710
 711         /* Resolve nfs_export -> !metacopy dependency */
 712         if (config->nfs_export && config->metacopy) {
 713                 if (nfs_export_opt && metacopy_opt) {
 714                         pr_err("conflicting options: nfs_export=on,metacopy=on\n");
 715                         return -EINVAL;
 716                 }
 717                 if (metacopy_opt) {
 718                         /*
 719                          * There was an explicit metacopy=on that resulted
 720                          * in this conflict.
 721                          */
 722                         pr_info("disabling nfs_export due to metacopy=on\n");
 723                         config->nfs_export = false;
 724                 } else {
 725                         /*
 726                          * There was an explicit nfs_export=on that resulted
 727                          * in this conflict.
 728                          */
 729                         pr_info("disabling metacopy due to nfs_export=on\n");
 730                         config->metacopy = false;
 731                 }
 732         }
 733
 734
 735         /* Resolve userxattr -> !redirect && !metacopy dependency */
 736         if (config->userxattr) {
 737                 if (config->redirect_follow && redirect_opt) {
 738                         pr_err("conflicting options: userxattr,redirect_dir=%s\n",
 739                                config->redirect_mode);
 740                         return -EINVAL;
 741                 }
 742                 if (config->metacopy && metacopy_opt) {
 743                         pr_err("conflicting options: userxattr,metacopy=on\n");
 744                         return -EINVAL;
 745                 }
 746                 /*
 747                  * Silently disable default setting of redirect and metacopy.
 748                  * This shall be the default in the future as well: these
 749                  * options must be explicitly enabled if used together with
 750                  * userxattr.
 751                  */
 752                 config->redirect_dir = config->redirect_follow = false;
 753                 config->metacopy = false;
 754         }
 755
 756         return 0;
 757 }
 758
 759 #define OVL_WORKDIR_NAME "work"
 760 #define OVL_INDEXDIR_NAME "index"
 761
 762 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 763                                          const char *name, bool persist)
 764 {
 765         struct inode *dir =  ofs->workbasedir->d_inode;
 766         struct vfsmount *mnt = ovl_upper_mnt(ofs);
 767         struct dentry *work;
 768         int err;
 769         bool retried = false;
 770
 771         inode_lock_nested(dir, I_MUTEX_PARENT);
 772 retry:
 773         work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));
 774
 775         if (!IS_ERR(work)) {
 776                 struct iattr attr = {
 777                         .ia_valid = ATTR_MODE,
 778                         .ia_mode = S_IFDIR | 0,
 779                 };
 780
 781                 if (work->d_inode) {
 782                         err = -EEXIST;
 783                         if (retried)
 784                                 goto out_dput;
 785
 786                         if (persist)
 787                                 goto out_unlock;
 788
 789                         retried = true;
 790                         err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);
 791                         dput(work);
 792                         if (err == -EINVAL) {
 793                                 work = ERR_PTR(err);
 794                                 goto out_unlock;
 795                         }
 796                         goto retry;
 797                 }
 798
 799                 err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);
 800                 if (err)
 801                         goto out_dput;
 802
 803                 /* Weird filesystem returning with hashed negative (kernfs)? */
 804                 err = -EINVAL;
 805                 if (d_really_is_negative(work))
 806                         goto out_dput;
 807
 808                 /*
 809                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 810                  *
 811                  * a) success (there was a POSIX ACL xattr and was removed)
 812                  * b) -ENODATA (there was no POSIX ACL xattr)
 813                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 814                  *
 815                  * There are various other error values that could effectively
 816                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 817                  * if the xattr name is too long), but the set of filesystems
 818                  * allowed as upper are limited to "normal" ones, where checking
 819                  * for the above two errors is sufficient.
 820                  */
 821                 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_DEFAULT);
 822                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 823                         goto out_dput;
 824
 825                 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_ACCESS);
 826                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 827                         goto out_dput;
 828
 829                 /* Clear any inherited mode bits */
 830                 inode_lock(work->d_inode);
 831                 err = ovl_do_notify_change(ofs, work, &attr);
 832                 inode_unlock(work->d_inode);
 833                 if (err)
 834                         goto out_dput;
 835         } else {
 836                 err = PTR_ERR(work);
 837                 goto out_err;
 838         }
 839 out_unlock:
 840         inode_unlock(dir);
 841         return work;
 842
 843 out_dput:
 844         dput(work);
 845 out_err:
 846         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
 847                 ofs->config.workdir, name, -err);
 848         work = NULL;
 849         goto out_unlock;
 850 }
 851
 852 static void ovl_unescape(char *s)
 853 {
 854         char *d = s;
 855
 856         for (;; s++, d++) {
 857                 if (*s == '\\')
 858                         s++;
 859                 *d = *s;
 860                 if (!*s)
 861                         break;
 862         }
 863 }
 864
 865 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 866 {
 867         int err = -EINVAL;
 868
 869         if (!*name) {
 870                 pr_err("empty lowerdir\n");
 871                 goto out;
 872         }
 873         err = kern_path(name, LOOKUP_FOLLOW, path);
 874         if (err) {
 875                 pr_err("failed to resolve '%s': %i\n", name, err);
 876                 goto out;
 877         }
 878         err = -EINVAL;
 879         if (ovl_dentry_weird(path->dentry)) {
 880                 pr_err("filesystem on '%s' not supported\n", name);
 881                 goto out_put;
 882         }
 883         if (!d_is_dir(path->dentry)) {
 884                 pr_err("'%s' not a directory\n", name);
 885                 goto out_put;
 886         }
 887         return 0;
 888
 889 out_put:
 890         path_put_init(path);
 891 out:
 892         return err;
 893 }
 894
 895 static int ovl_mount_dir(const char *name, struct path *path)
 896 {
 897         int err = -ENOMEM;
 898         char *tmp = kstrdup(name, GFP_KERNEL);
 899
 900         if (tmp) {
 901                 ovl_unescape(tmp);
 902                 err = ovl_mount_dir_noesc(tmp, path);
 903
 904                 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
 905                         pr_err("filesystem on '%s' not supported as upperdir\n",
 906                                tmp);
 907                         path_put_init(path);
 908                         err = -EINVAL;
 909                 }
 910                 kfree(tmp);
 911         }
 912         return err;
 913 }
 914
 915 static int ovl_check_namelen(const struct path *path, struct ovl_fs *ofs,
 916                              const char *name)
 917 {
 918         struct kstatfs statfs;
 919         int err = vfs_statfs(path, &statfs);
 920
 921         if (err)
 922                 pr_err("statfs failed on '%s'\n", name);
 923         else
 924                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 925
 926         return err;
 927 }
 928
 929 static int ovl_lower_dir(const char *name, struct path *path,
 930                          struct ovl_fs *ofs, int *stack_depth)
 931 {
 932         int fh_type;
 933         int err;
 934
 935         err = ovl_mount_dir_noesc(name, path);
 936         if (err)
 937                 return err;
 938
 939         err = ovl_check_namelen(path, ofs, name);
 940         if (err)
 941                 return err;
 942
 943         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 944
 945         /*
 946          * The inodes index feature and NFS export need to encode and decode
 947          * file handles, so they require that all layers support them.
 948          */
 949         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 950         if ((ofs->config.nfs_export ||
 951              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 952                 ofs->config.index = false;
 953                 ofs->config.nfs_export = false;
 954                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 955                         name);
 956         }
 957         /*
 958          * Decoding origin file handle is required for persistent st_ino.
 959          * Without persistent st_ino, xino=auto falls back to xino=off.
 960          */
 961         if (ofs->config.xino == OVL_XINO_AUTO &&
 962             ofs->config.upperdir && !fh_type) {
 963                 ofs->config.xino = OVL_XINO_OFF;
 964                 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
 965                         name);
 966         }
 967
 968         /* Check if lower fs has 32bit inode numbers */
 969         if (fh_type != FILEID_INO32_GEN)
 970                 ofs->xino_mode = -1;
 971
 972         return 0;
 973 }
 974
 975 /* Workdir should not be subdir of upperdir and vice versa */
 976 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 977 {
 978         bool ok = false;
 979
 980         if (workdir != upperdir) {
 981                 ok = (lock_rename(workdir, upperdir) == NULL);
 982                 unlock_rename(workdir, upperdir);
 983         }
 984         return ok;
 985 }
 986
 987 static unsigned int ovl_split_lowerdirs(char *str)
 988 {
 989         unsigned int ctr = 1;
 990         char *s, *d;
 991
 992         for (s = d = str;; s++, d++) {
 993                 if (*s == '\\') {
 994                         s++;
 995                 } else if (*s == ':') {
 996                         *d = '\0';
 997                         ctr++;
 998                         continue;
 999                 }
1000                 *d = *s;
1001                 if (!*s)
1002                         break;
1003         }
1004         return ctr;
1005 }
1006
1007 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1008                              struct dentry *dentry, struct inode *inode,
1009                              const char *name, void *buffer, size_t size)
1010 {
1011         return -EOPNOTSUPP;
1012 }
1013
1014 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1015                              struct mnt_idmap *idmap,
1016                              struct dentry *dentry, struct inode *inode,
1017                              const char *name, const void *value,
1018                              size_t size, int flags)
1019 {
1020         return -EOPNOTSUPP;
1021 }
1022
1023 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1024                                struct dentry *dentry, struct inode *inode,
1025                                const char *name, void *buffer, size_t size)
1026 {
1027         return ovl_xattr_get(dentry, inode, name, buffer, size);
1028 }
1029
1030 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1031                                struct mnt_idmap *idmap,
1032                                struct dentry *dentry, struct inode *inode,
1033                                const char *name, const void *value,
1034                                size_t size, int flags)
1035 {
1036         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1037 }
1038
1039 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1040         .prefix = OVL_XATTR_TRUSTED_PREFIX,
1041         .get = ovl_own_xattr_get,
1042         .set = ovl_own_xattr_set,
1043 };
1044
1045 static const struct xattr_handler ovl_own_user_xattr_handler = {
1046         .prefix = OVL_XATTR_USER_PREFIX,
1047         .get = ovl_own_xattr_get,
1048         .set = ovl_own_xattr_set,
1049 };
1050
1051 static const struct xattr_handler ovl_other_xattr_handler = {
1052         .prefix = "", /* catch all */
1053         .get = ovl_other_xattr_get,
1054         .set = ovl_other_xattr_set,
1055 };
1056
1057 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1058         &ovl_own_trusted_xattr_handler,
1059         &ovl_other_xattr_handler,
1060         NULL
1061 };
1062
1063 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1064         &ovl_own_user_xattr_handler,
1065         &ovl_other_xattr_handler,
1066         NULL
1067 };
1068
1069 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1070                           struct inode **ptrap, const char *name)
1071 {
1072         struct inode *trap;
1073         int err;
1074
1075         trap = ovl_get_trap_inode(sb, dir);
1076         err = PTR_ERR_OR_ZERO(trap);
1077         if (err) {
1078                 if (err == -ELOOP)
1079                         pr_err("conflicting %s path\n", name);
1080                 return err;
1081         }
1082
1083         *ptrap = trap;
1084         return 0;
1085 }
1086
1087 /*
1088  * Determine how we treat concurrent use of upperdir/workdir based on the
1089  * index feature. This is papering over mount leaks of container runtimes,
1090  * for example, an old overlay mount is leaked and now its upperdir is
1091  * attempted to be used as a lower layer in a new overlay mount.
1092  */
1093 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1094 {
1095         if (ofs->config.index) {
1096                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1097                        name);
1098                 return -EBUSY;
1099         } else {
1100                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1101                         name);
1102                 return 0;
1103         }
1104 }
1105
1106 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1107                          struct ovl_layer *upper_layer, struct path *upperpath)
1108 {
1109         struct vfsmount *upper_mnt;
1110         int err;
1111
1112         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1113         if (err)
1114                 goto out;
1115
1116         /* Upperdir path should not be r/o */
1117         if (__mnt_is_readonly(upperpath->mnt)) {
1118                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1119                 err = -EINVAL;
1120                 goto out;
1121         }
1122
1123         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1124         if (err)
1125                 goto out;
1126
1127         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1128                              "upperdir");
1129         if (err)
1130                 goto out;
1131
1132         upper_mnt = clone_private_mount(upperpath);
1133         err = PTR_ERR(upper_mnt);
1134         if (IS_ERR(upper_mnt)) {
1135                 pr_err("failed to clone upperpath\n");
1136                 goto out;
1137         }
1138
1139         /* Don't inherit atime flags */
1140         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1141         upper_layer->mnt = upper_mnt;
1142         upper_layer->idx = 0;
1143         upper_layer->fsid = 0;
1144
1145         /*
1146          * Inherit SB_NOSEC flag from upperdir.
1147          *
1148          * This optimization changes behavior when a security related attribute
1149          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1150          * okay because we don't yet have guarantees in that case, but it will
1151          * need careful treatment once we want to honour changes to underlying
1152          * filesystems.
1153          */
1154         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1155                 sb->s_flags |= SB_NOSEC;
1156
1157         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1158                 ofs->upperdir_locked = true;
1159         } else {
1160                 err = ovl_report_in_use(ofs, "upperdir");
1161                 if (err)
1162                         goto out;
1163         }
1164
1165         err = 0;
1166 out:
1167         return err;
1168 }
1169
1170 /*
1171  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1172  * negative values if error is encountered.
1173  */
1174 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
1175 {
1176         struct dentry *workdir = ofs->workdir;
1177         struct inode *dir = d_inode(workdir);
1178         struct dentry *temp;
1179         struct dentry *dest;
1180         struct dentry *whiteout;
1181         struct name_snapshot name;
1182         int err;
1183
1184         inode_lock_nested(dir, I_MUTEX_PARENT);
1185
1186         temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
1187         err = PTR_ERR(temp);
1188         if (IS_ERR(temp))
1189                 goto out_unlock;
1190
1191         dest = ovl_lookup_temp(ofs, workdir);
1192         err = PTR_ERR(dest);
1193         if (IS_ERR(dest)) {
1194                 dput(temp);
1195                 goto out_unlock;
1196         }
1197
1198         /* Name is inline and stable - using snapshot as a copy helper */
1199         take_dentry_name_snapshot(&name, temp);
1200         err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
1201         if (err) {
1202                 if (err == -EINVAL)
1203                         err = 0;
1204                 goto cleanup_temp;
1205         }
1206
1207         whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
1208         err = PTR_ERR(whiteout);
1209         if (IS_ERR(whiteout))
1210                 goto cleanup_temp;
1211
1212         err = ovl_is_whiteout(whiteout);
1213
1214         /* Best effort cleanup of whiteout and temp file */
1215         if (err)
1216                 ovl_cleanup(ofs, dir, whiteout);
1217         dput(whiteout);
1218
1219 cleanup_temp:
1220         ovl_cleanup(ofs, dir, temp);
1221         release_dentry_name_snapshot(&name);
1222         dput(temp);
1223         dput(dest);
1224
1225 out_unlock:
1226         inode_unlock(dir);
1227
1228         return err;
1229 }
1230
1231 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
1232                                            struct dentry *parent,
1233                                            const char *name, umode_t mode)
1234 {
1235         size_t len = strlen(name);
1236         struct dentry *child;
1237
1238         inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1239         child = ovl_lookup_upper(ofs, name, parent, len);
1240         if (!IS_ERR(child) && !child->d_inode)
1241                 child = ovl_create_real(ofs, parent->d_inode, child,
1242                                         OVL_CATTR(mode));
1243         inode_unlock(parent->d_inode);
1244         dput(parent);
1245
1246         return child;
1247 }
1248
1249 /*
1250  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1251  * present.
1252  */
1253 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1254 {
1255         unsigned int ctr;
1256         struct dentry *d = dget(ofs->workbasedir);
1257         static const char *const volatile_path[] = {
1258                 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1259         };
1260         const char *const *name = volatile_path;
1261
1262         for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1263                 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1264                 if (IS_ERR(d))
1265                         return PTR_ERR(d);
1266         }
1267         dput(d);
1268         return 0;
1269 }
1270
1271 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1272                             const struct path *workpath)
1273 {
1274         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1275         struct dentry *workdir;
1276         struct file *tmpfile;
1277         bool rename_whiteout;
1278         bool d_type;
1279         int fh_type;
1280         int err;
1281
1282         err = mnt_want_write(mnt);
1283         if (err)
1284                 return err;
1285
1286         workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1287         err = PTR_ERR(workdir);
1288         if (IS_ERR_OR_NULL(workdir))
1289                 goto out;
1290
1291         ofs->workdir = workdir;
1292
1293         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1294         if (err)
1295                 goto out;
1296
1297         /*
1298          * Upper should support d_type, else whiteouts are visible.  Given
1299          * workdir and upper are on same fs, we can do iterate_dir() on
1300          * workdir. This check requires successful creation of workdir in
1301          * previous step.
1302          */
1303         err = ovl_check_d_type_supported(workpath);
1304         if (err < 0)
1305                 goto out;
1306
1307         d_type = err;
1308         if (!d_type)
1309                 pr_warn("upper fs needs to support d_type.\n");
1310
1311         /* Check if upper/work fs supports O_TMPFILE */
1312         tmpfile = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
1313         ofs->tmpfile = !IS_ERR(tmpfile);
1314         if (ofs->tmpfile)
1315                 fput(tmpfile);
1316         else
1317                 pr_warn("upper fs does not support tmpfile.\n");
1318
1319
1320         /* Check if upper/work fs supports RENAME_WHITEOUT */
1321         err = ovl_check_rename_whiteout(ofs);
1322         if (err < 0)
1323                 goto out;
1324
1325         rename_whiteout = err;
1326         if (!rename_whiteout)
1327                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1328
1329         /*
1330          * Check if upper/work fs supports (trusted|user).overlay.* xattr
1331          */
1332         err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1333         if (err) {
1334                 pr_warn("failed to set xattr on upper\n");
1335                 ofs->noxattr = true;
1336                 if (ofs->config.index || ofs->config.metacopy) {
1337                         ofs->config.index = false;
1338                         ofs->config.metacopy = false;
1339                         pr_warn("...falling back to index=off,metacopy=off.\n");
1340                 }
1341                 /*
1342                  * xattr support is required for persistent st_ino.
1343                  * Without persistent st_ino, xino=auto falls back to xino=off.
1344                  */
1345                 if (ofs->config.xino == OVL_XINO_AUTO) {
1346                         ofs->config.xino = OVL_XINO_OFF;
1347                         pr_warn("...falling back to xino=off.\n");
1348                 }
1349                 if (err == -EPERM && !ofs->config.userxattr)
1350                         pr_info("try mounting with 'userxattr' option\n");
1351                 err = 0;
1352         } else {
1353                 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1354         }
1355
1356         /*
1357          * We allowed sub-optimal upper fs configuration and don't want to break
1358          * users over kernel upgrade, but we never allowed remote upper fs, so
1359          * we can enforce strict requirements for remote upper fs.
1360          */
1361         if (ovl_dentry_remote(ofs->workdir) &&
1362             (!d_type || !rename_whiteout || ofs->noxattr)) {
1363                 pr_err("upper fs missing required features.\n");
1364                 err = -EINVAL;
1365                 goto out;
1366         }
1367
1368         /*
1369          * For volatile mount, create a incompat/volatile/dirty file to keep
1370          * track of it.
1371          */
1372         if (ofs->config.ovl_volatile) {
1373                 err = ovl_create_volatile_dirty(ofs);
1374                 if (err < 0) {
1375                         pr_err("Failed to create volatile/dirty file.\n");
1376                         goto out;
1377                 }
1378         }
1379
1380         /* Check if upper/work fs supports file handles */
1381         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1382         if (ofs->config.index && !fh_type) {
1383                 ofs->config.index = false;
1384                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1385         }
1386
1387         /* Check if upper fs has 32bit inode numbers */
1388         if (fh_type != FILEID_INO32_GEN)
1389                 ofs->xino_mode = -1;
1390
1391         /* NFS export of r/w mount depends on index */
1392         if (ofs->config.nfs_export && !ofs->config.index) {
1393                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1394                 ofs->config.nfs_export = false;
1395         }
1396 out:
1397         mnt_drop_write(mnt);
1398         return err;
1399 }
1400
1401 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1402                            const struct path *upperpath)
1403 {
1404         int err;
1405         struct path workpath = { };
1406
1407         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1408         if (err)
1409                 goto out;
1410
1411         err = -EINVAL;
1412         if (upperpath->mnt != workpath.mnt) {
1413                 pr_err("workdir and upperdir must reside under the same mount\n");
1414                 goto out;
1415         }
1416         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1417                 pr_err("workdir and upperdir must be separate subtrees\n");
1418                 goto out;
1419         }
1420
1421         ofs->workbasedir = dget(workpath.dentry);
1422
1423         if (ovl_inuse_trylock(ofs->workbasedir)) {
1424                 ofs->workdir_locked = true;
1425         } else {
1426                 err = ovl_report_in_use(ofs, "workdir");
1427                 if (err)
1428                         goto out;
1429         }
1430
1431         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1432                              "workdir");
1433         if (err)
1434                 goto out;
1435
1436         err = ovl_make_workdir(sb, ofs, &workpath);
1437
1438 out:
1439         path_put(&workpath);
1440
1441         return err;
1442 }
1443
1444 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1445                             struct ovl_entry *oe, const struct path *upperpath)
1446 {
1447         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1448         struct dentry *indexdir;
1449         int err;
1450
1451         err = mnt_want_write(mnt);
1452         if (err)
1453                 return err;
1454
1455         /* Verify lower root is upper root origin */
1456         err = ovl_verify_origin(ofs, upperpath->dentry,
1457                                 oe->lowerstack[0].dentry, true);
1458         if (err) {
1459                 pr_err("failed to verify upper root origin\n");
1460                 goto out;
1461         }
1462
1463         /* index dir will act also as workdir */
1464         iput(ofs->workdir_trap);
1465         ofs->workdir_trap = NULL;
1466         dput(ofs->workdir);
1467         ofs->workdir = NULL;
1468         indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1469         if (IS_ERR(indexdir)) {
1470                 err = PTR_ERR(indexdir);
1471         } else if (indexdir) {
1472                 ofs->indexdir = indexdir;
1473                 ofs->workdir = dget(indexdir);
1474
1475                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1476                                      "indexdir");
1477                 if (err)
1478                         goto out;
1479
1480                 /*
1481                  * Verify upper root is exclusively associated with index dir.
1482                  * Older kernels stored upper fh in ".overlay.origin"
1483                  * xattr. If that xattr exists, verify that it is a match to
1484                  * upper dir file handle. In any case, verify or set xattr
1485                  * ".overlay.upper" to indicate that index may have
1486                  * directory entries.
1487                  */
1488                 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1489                         err = ovl_verify_set_fh(ofs, ofs->indexdir,
1490                                                 OVL_XATTR_ORIGIN,
1491                                                 upperpath->dentry, true, false);
1492                         if (err)
1493                                 pr_err("failed to verify index dir 'origin' xattr\n");
1494                 }
1495                 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1496                                        true);
1497                 if (err)
1498                         pr_err("failed to verify index dir 'upper' xattr\n");
1499
1500                 /* Cleanup bad/stale/orphan index entries */
1501                 if (!err)
1502                         err = ovl_indexdir_cleanup(ofs);
1503         }
1504         if (err || !ofs->indexdir)
1505                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1506
1507 out:
1508         mnt_drop_write(mnt);
1509         return err;
1510 }
1511
1512 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1513 {
1514         unsigned int i;
1515
1516         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1517                 return true;
1518
1519         /*
1520          * We allow using single lower with null uuid for index and nfs_export
1521          * for example to support those features with single lower squashfs.
1522          * To avoid regressions in setups of overlay with re-formatted lower
1523          * squashfs, do not allow decoding origin with lower null uuid unless
1524          * user opted-in to one of the new features that require following the
1525          * lower inode of non-dir upper.
1526          */
1527         if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
1528                 return false;
1529
1530         for (i = 0; i < ofs->numfs; i++) {
1531                 /*
1532                  * We use uuid to associate an overlay lower file handle with a
1533                  * lower layer, so we can accept lower fs with null uuid as long
1534                  * as all lower layers with null uuid are on the same fs.
1535                  * if we detect multiple lower fs with the same uuid, we
1536                  * disable lower file handle decoding on all of them.
1537                  */
1538                 if (ofs->fs[i].is_lower &&
1539                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1540                         ofs->fs[i].bad_uuid = true;
1541                         return false;
1542                 }
1543         }
1544         return true;
1545 }
1546
1547 /* Get a unique fsid for the layer */
1548 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1549 {
1550         struct super_block *sb = path->mnt->mnt_sb;
1551         unsigned int i;
1552         dev_t dev;
1553         int err;
1554         bool bad_uuid = false;
1555         bool warn = false;
1556
1557         for (i = 0; i < ofs->numfs; i++) {
1558                 if (ofs->fs[i].sb == sb)
1559                         return i;
1560         }
1561
1562         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1563                 bad_uuid = true;
1564                 if (ofs->config.xino == OVL_XINO_AUTO) {
1565                         ofs->config.xino = OVL_XINO_OFF;
1566                         warn = true;
1567                 }
1568                 if (ofs->config.index || ofs->config.nfs_export) {
1569                         ofs->config.index = false;
1570                         ofs->config.nfs_export = false;
1571                         warn = true;
1572                 }
1573                 if (warn) {
1574                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1575                                 uuid_is_null(&sb->s_uuid) ? "null" :
1576                                                             "conflicting",
1577                                 path->dentry, ovl_xino_str[ofs->config.xino]);
1578                 }
1579         }
1580
1581         err = get_anon_bdev(&dev);
1582         if (err) {
1583                 pr_err("failed to get anonymous bdev for lowerpath\n");
1584                 return err;
1585         }
1586
1587         ofs->fs[ofs->numfs].sb = sb;
1588         ofs->fs[ofs->numfs].pseudo_dev = dev;
1589         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1590
1591         return ofs->numfs++;
1592 }
1593
1594 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1595                           struct path *stack, unsigned int numlower,
1596                           struct ovl_layer *layers)
1597 {
1598         int err;
1599         unsigned int i;
1600
1601         err = -ENOMEM;
1602         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1603         if (ofs->fs == NULL)
1604                 goto out;
1605
1606         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1607         ofs->numfs++;
1608
1609         /*
1610          * All lower layers that share the same fs as upper layer, use the same
1611          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1612          * only overlay to simplify ovl_fs_free().
1613          * is_lower will be set if upper fs is shared with a lower layer.
1614          */
1615         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1616         if (err) {
1617                 pr_err("failed to get anonymous bdev for upper fs\n");
1618                 goto out;
1619         }
1620
1621         if (ovl_upper_mnt(ofs)) {
1622                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1623                 ofs->fs[0].is_lower = false;
1624         }
1625
1626         for (i = 0; i < numlower; i++) {
1627                 struct vfsmount *mnt;
1628                 struct inode *trap;
1629                 int fsid;
1630
1631                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1632                 if (err < 0)
1633                         goto out;
1634
1635                 /*
1636                  * Check if lower root conflicts with this overlay layers before
1637                  * checking if it is in-use as upperdir/workdir of "another"
1638                  * mount, because we do not bother to check in ovl_is_inuse() if
1639                  * the upperdir/workdir is in fact in-use by our
1640                  * upperdir/workdir.
1641                  */
1642                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1643                 if (err)
1644                         goto out;
1645
1646                 if (ovl_is_inuse(stack[i].dentry)) {
1647                         err = ovl_report_in_use(ofs, "lowerdir");
1648                         if (err) {
1649                                 iput(trap);
1650                                 goto out;
1651                         }
1652                 }
1653
1654                 mnt = clone_private_mount(&stack[i]);
1655                 err = PTR_ERR(mnt);
1656                 if (IS_ERR(mnt)) {
1657                         pr_err("failed to clone lowerpath\n");
1658                         iput(trap);
1659                         goto out;
1660                 }
1661
1662                 /*
1663                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1664                  * will fail instead of modifying lower fs.
1665                  */
1666                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1667
1668                 layers[ofs->numlayer].trap = trap;
1669                 layers[ofs->numlayer].mnt = mnt;
1670                 layers[ofs->numlayer].idx = ofs->numlayer;
1671                 layers[ofs->numlayer].fsid = fsid;
1672                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1673                 ofs->numlayer++;
1674                 ofs->fs[fsid].is_lower = true;
1675         }
1676
1677         /*
1678          * When all layers on same fs, overlay can use real inode numbers.
1679          * With mount option "xino=<on|auto>", mounter declares that there are
1680          * enough free high bits in underlying fs to hold the unique fsid.
1681          * If overlayfs does encounter underlying inodes using the high xino
1682          * bits reserved for fsid, it emits a warning and uses the original
1683          * inode number or a non persistent inode number allocated from a
1684          * dedicated range.
1685          */
1686         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1687                 if (ofs->config.xino == OVL_XINO_ON)
1688                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1689                 ofs->xino_mode = 0;
1690         } else if (ofs->config.xino == OVL_XINO_OFF) {
1691                 ofs->xino_mode = -1;
1692         } else if (ofs->xino_mode < 0) {
1693                 /*
1694                  * This is a roundup of number of bits needed for encoding
1695                  * fsid, where fsid 0 is reserved for upper fs (even with
1696                  * lower only overlay) +1 extra bit is reserved for the non
1697                  * persistent inode number range that is used for resolving
1698                  * xino lower bits overflow.
1699                  */
1700                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1701                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1702         }
1703
1704         if (ofs->xino_mode > 0) {
1705                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1706                         ofs->xino_mode);
1707         }
1708
1709         err = 0;
1710 out:
1711         return err;
1712 }
1713
1714 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1715                                 const char *lower, unsigned int numlower,
1716                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1717 {
1718         int err;
1719         struct path *stack = NULL;
1720         unsigned int i;
1721         struct ovl_entry *oe;
1722
1723         if (!ofs->config.upperdir && numlower == 1) {
1724                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1725                 return ERR_PTR(-EINVAL);
1726         }
1727
1728         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1729         if (!stack)
1730                 return ERR_PTR(-ENOMEM);
1731
1732         err = -EINVAL;
1733         for (i = 0; i < numlower; i++) {
1734                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1735                 if (err)
1736                         goto out_err;
1737
1738                 lower = strchr(lower, '\0') + 1;
1739         }
1740
1741         err = -EINVAL;
1742         sb->s_stack_depth++;
1743         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1744                 pr_err("maximum fs stacking depth exceeded\n");
1745                 goto out_err;
1746         }
1747
1748         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1749         if (err)
1750                 goto out_err;
1751
1752         err = -ENOMEM;
1753         oe = ovl_alloc_entry(numlower);
1754         if (!oe)
1755                 goto out_err;
1756
1757         for (i = 0; i < numlower; i++) {
1758                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1759                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1760         }
1761
1762 out:
1763         for (i = 0; i < numlower; i++)
1764                 path_put(&stack[i]);
1765         kfree(stack);
1766
1767         return oe;
1768
1769 out_err:
1770         oe = ERR_PTR(err);
1771         goto out;
1772 }
1773
1774 /*
1775  * Check if this layer root is a descendant of:
1776  * - another layer of this overlayfs instance
1777  * - upper/work dir of any overlayfs instance
1778  */
1779 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1780                            struct dentry *dentry, const char *name,
1781                            bool is_lower)
1782 {
1783         struct dentry *next = dentry, *parent;
1784         int err = 0;
1785
1786         if (!dentry)
1787                 return 0;
1788
1789         parent = dget_parent(next);
1790
1791         /* Walk back ancestors to root (inclusive) looking for traps */
1792         while (!err && parent != next) {
1793                 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1794                         err = -ELOOP;
1795                         pr_err("overlapping %s path\n", name);
1796                 } else if (ovl_is_inuse(parent)) {
1797                         err = ovl_report_in_use(ofs, name);
1798                 }
1799                 next = parent;
1800                 parent = dget_parent(next);
1801                 dput(next);
1802         }
1803
1804         dput(parent);
1805
1806         return err;
1807 }
1808
1809 /*
1810  * Check if any of the layers or work dirs overlap.
1811  */
1812 static int ovl_check_overlapping_layers(struct super_block *sb,
1813                                         struct ovl_fs *ofs)
1814 {
1815         int i, err;
1816
1817         if (ovl_upper_mnt(ofs)) {
1818                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1819                                       "upperdir", false);
1820                 if (err)
1821                         return err;
1822
1823                 /*
1824                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1825                  * this instance and covers overlapping work and index dirs,
1826                  * unless work or index dir have been moved since created inside
1827                  * workbasedir.  In that case, we already have their traps in
1828                  * inode cache and we will catch that case on lookup.
1829                  */
1830                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1831                                       false);
1832                 if (err)
1833                         return err;
1834         }
1835
1836         for (i = 1; i < ofs->numlayer; i++) {
1837                 err = ovl_check_layer(sb, ofs,
1838                                       ofs->layers[i].mnt->mnt_root,
1839                                       "lowerdir", true);
1840                 if (err)
1841                         return err;
1842         }
1843
1844         return 0;
1845 }
1846
1847 static struct dentry *ovl_get_root(struct super_block *sb,
1848                                    struct dentry *upperdentry,
1849                                    struct ovl_entry *oe)
1850 {
1851         struct dentry *root;
1852         struct ovl_path *lowerpath = &oe->lowerstack[0];
1853         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1854         int fsid = lowerpath->layer->fsid;
1855         struct ovl_inode_params oip = {
1856                 .upperdentry = upperdentry,
1857                 .lowerpath = lowerpath,
1858         };
1859
1860         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1861         if (!root)
1862                 return NULL;
1863
1864         root->d_fsdata = oe;
1865
1866         if (upperdentry) {
1867                 /* Root inode uses upper st_ino/i_ino */
1868                 ino = d_inode(upperdentry)->i_ino;
1869                 fsid = 0;
1870                 ovl_dentry_set_upper_alias(root);
1871                 if (ovl_is_impuredir(sb, upperdentry))
1872                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1873         }
1874
1875         /* Root is always merge -> can have whiteouts */
1876         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1877         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1878         ovl_set_upperdata(d_inode(root));
1879         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1880         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1881
1882         return root;
1883 }
1884
1885 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1886 {
1887         struct path upperpath = { };
1888         struct dentry *root_dentry;
1889         struct ovl_entry *oe;
1890         struct ovl_fs *ofs;
1891         struct ovl_layer *layers;
1892         struct cred *cred;
1893         char *splitlower = NULL;
1894         unsigned int numlower;
1895         int err;
1896
1897         err = -EIO;
1898         if (WARN_ON(sb->s_user_ns != current_user_ns()))
1899                 goto out;
1900
1901         sb->s_d_op = &ovl_dentry_operations;
1902
1903         err = -ENOMEM;
1904         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1905         if (!ofs)
1906                 goto out;
1907
1908         err = -ENOMEM;
1909         ofs->creator_cred = cred = prepare_creds();
1910         if (!cred)
1911                 goto out_err;
1912
1913         /* Is there a reason anyone would want not to share whiteouts? */
1914         ofs->share_whiteout = true;
1915
1916         ofs->config.index = ovl_index_def;
1917         ofs->config.uuid = true;
1918         ofs->config.nfs_export = ovl_nfs_export_def;
1919         ofs->config.xino = ovl_xino_def();
1920         ofs->config.metacopy = ovl_metacopy_def;
1921         err = ovl_parse_opt((char *) data, &ofs->config);
1922         if (err)
1923                 goto out_err;
1924
1925         err = -EINVAL;
1926         if (!ofs->config.lowerdir) {
1927                 if (!silent)
1928                         pr_err("missing 'lowerdir'\n");
1929                 goto out_err;
1930         }
1931
1932         err = -ENOMEM;
1933         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1934         if (!splitlower)
1935                 goto out_err;
1936
1937         err = -EINVAL;
1938         numlower = ovl_split_lowerdirs(splitlower);
1939         if (numlower > OVL_MAX_STACK) {
1940                 pr_err("too many lower directories, limit is %d\n",
1941                        OVL_MAX_STACK);
1942                 goto out_err;
1943         }
1944
1945         err = -ENOMEM;
1946         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1947         if (!layers)
1948                 goto out_err;
1949
1950         ofs->layers = layers;
1951         /* Layer 0 is reserved for upper even if there's no upper */
1952         ofs->numlayer = 1;
1953
1954         sb->s_stack_depth = 0;
1955         sb->s_maxbytes = MAX_LFS_FILESIZE;
1956         atomic_long_set(&ofs->last_ino, 1);
1957         /* Assume underlying fs uses 32bit inodes unless proven otherwise */
1958         if (ofs->config.xino != OVL_XINO_OFF) {
1959                 ofs->xino_mode = BITS_PER_LONG - 32;
1960                 if (!ofs->xino_mode) {
1961                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1962                         ofs->config.xino = OVL_XINO_OFF;
1963                 }
1964         }
1965
1966         /* alloc/destroy_inode needed for setting up traps in inode cache */
1967         sb->s_op = &ovl_super_operations;
1968
1969         if (ofs->config.upperdir) {
1970                 struct super_block *upper_sb;
1971
1972                 err = -EINVAL;
1973                 if (!ofs->config.workdir) {
1974                         pr_err("missing 'workdir'\n");
1975                         goto out_err;
1976                 }
1977
1978                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
1979                 if (err)
1980                         goto out_err;
1981
1982                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
1983                 if (!ovl_should_sync(ofs)) {
1984                         ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
1985                         if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
1986                                 err = -EIO;
1987                                 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
1988                                 goto out_err;
1989                         }
1990                 }
1991
1992                 err = ovl_get_workdir(sb, ofs, &upperpath);
1993                 if (err)
1994                         goto out_err;
1995
1996                 if (!ofs->workdir)
1997                         sb->s_flags |= SB_RDONLY;
1998
1999                 sb->s_stack_depth = upper_sb->s_stack_depth;
2000                 sb->s_time_gran = upper_sb->s_time_gran;
2001         }
2002         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2003         err = PTR_ERR(oe);
2004         if (IS_ERR(oe))
2005                 goto out_err;
2006
2007         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2008         if (!ovl_upper_mnt(ofs))
2009                 sb->s_flags |= SB_RDONLY;
2010
2011         if (!ofs->config.uuid && ofs->numfs > 1) {
2012                 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2013                 ofs->config.uuid = true;
2014         }
2015
2016         if (!ovl_force_readonly(ofs) && ofs->config.index) {
2017                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2018                 if (err)
2019                         goto out_free_oe;
2020
2021                 /* Force r/o mount with no index dir */
2022                 if (!ofs->indexdir)
2023                         sb->s_flags |= SB_RDONLY;
2024         }
2025
2026         err = ovl_check_overlapping_layers(sb, ofs);
2027         if (err)
2028                 goto out_free_oe;
2029
2030         /* Show index=off in /proc/mounts for forced r/o mount */
2031         if (!ofs->indexdir) {
2032                 ofs->config.index = false;
2033                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2034                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2035                         ofs->config.nfs_export = false;
2036                 }
2037         }
2038
2039         if (ofs->config.metacopy && ofs->config.nfs_export) {
2040                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2041                 ofs->config.nfs_export = false;
2042         }
2043
2044         if (ofs->config.nfs_export)
2045                 sb->s_export_op = &ovl_export_operations;
2046
2047         /* Never override disk quota limits or use reserved space */
2048         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2049
2050         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2051         sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2052                 ovl_trusted_xattr_handlers;
2053         sb->s_fs_info = ofs;
2054         sb->s_flags |= SB_POSIXACL;
2055         sb->s_iflags |= SB_I_SKIP_SYNC;
2056
2057         err = -ENOMEM;
2058         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2059         if (!root_dentry)
2060                 goto out_free_oe;
2061
2062         mntput(upperpath.mnt);
2063         kfree(splitlower);
2064
2065         sb->s_root = root_dentry;
2066
2067         return 0;
2068
2069 out_free_oe:
2070         ovl_entry_stack_free(oe);
2071         kfree(oe);
2072 out_err:
2073         kfree(splitlower);
2074         path_put(&upperpath);
2075         ovl_free_fs(ofs);
2076 out:
2077         return err;
2078 }
2079
2080 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2081                                 const char *dev_name, void *raw_data)
2082 {
2083         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2084 }
2085
2086 static struct file_system_type ovl_fs_type = {
2087         .owner          = THIS_MODULE,
2088         .name           = "overlay",
2089         .fs_flags       = FS_USERNS_MOUNT,
2090         .mount          = ovl_mount,
2091         .kill_sb        = kill_anon_super,
2092 };
2093 MODULE_ALIAS_FS("overlay");
2094
2095 static void ovl_inode_init_once(void *foo)
2096 {
2097         struct ovl_inode *oi = foo;
2098
2099         inode_init_once(&oi->vfs_inode);
2100 }
2101
2102 static int __init ovl_init(void)
2103 {
2104         int err;
2105
2106         ovl_inode_cachep = kmem_cache_create("ovl_inode",
2107                                              sizeof(struct ovl_inode), 0,
2108                                              (SLAB_RECLAIM_ACCOUNT|
2109                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2110                                              ovl_inode_init_once);
2111         if (ovl_inode_cachep == NULL)
2112                 return -ENOMEM;
2113
2114         err = ovl_aio_request_cache_init();
2115         if (!err) {
2116                 err = register_filesystem(&ovl_fs_type);
2117                 if (!err)
2118                         return 0;
2119
2120                 ovl_aio_request_cache_destroy();
2121         }
2122         kmem_cache_destroy(ovl_inode_cachep);
2123
2124         return err;
2125 }
2126
2127 static void __exit ovl_exit(void)
2128 {
2129         unregister_filesystem(&ovl_fs_type);
2130
2131         /*
2132          * Make sure all delayed rcu free inodes are flushed before we
2133          * destroy cache.
2134          */
2135         rcu_barrier();
2136         kmem_cache_destroy(ovl_inode_cachep);
2137         ovl_aio_request_cache_destroy();
2138 }
2139
2140 module_init(ovl_init);
2141 module_exit(ovl_exit);