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         ret = ovl_sync_status(ofs);
 268         /*
 269          * We have to always set the err, because the return value isn't
 270          * checked in syncfs, and instead indirectly return an error via
 271          * the sb's writeback errseq, which VFS inspects after this call.
 272          */
 273         if (ret < 0) {
 274                 errseq_set(&sb->s_wb_err, -EIO);
 275                 return -EIO;
 276         }
 277
 278         if (!ret)
 279                 return ret;
 280
 281         /*
 282          * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
 283          * All the super blocks will be iterated, including upper_sb.
 284          *
 285          * If this is a syncfs(2) call, then we do need to call
 286          * sync_filesystem() on upper_sb, but enough if we do it when being
 287          * called with wait == 1.
 288          */
 289         if (!wait)
 290                 return 0;
 291
 292         upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 293
 294         down_read(&upper_sb->s_umount);
 295         ret = sync_filesystem(upper_sb);
 296         up_read(&upper_sb->s_umount);
 297
 298         return ret;
 299 }
 300
 301 /**
 302  * ovl_statfs
 303  * @sb: The overlayfs super block
 304  * @buf: The struct kstatfs to fill in with stats
 305  *
 306  * Get the filesystem statistics.  As writes always target the upper layer
 307  * filesystem pass the statfs to the upper filesystem (if it exists)
 308  */
 309 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 310 {
 311         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 312         struct dentry *root_dentry = dentry->d_sb->s_root;
 313         struct path path;
 314         int err;
 315
 316         ovl_path_real(root_dentry, &path);
 317
 318         err = vfs_statfs(&path, buf);
 319         if (!err) {
 320                 buf->f_namelen = ofs->namelen;
 321                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 322         }
 323
 324         return err;
 325 }
 326
 327 /* Will this overlay be forced to mount/remount ro? */
 328 static bool ovl_force_readonly(struct ovl_fs *ofs)
 329 {
 330         return (!ovl_upper_mnt(ofs) || !ofs->workdir);
 331 }
 332
 333 static const char *ovl_redirect_mode_def(void)
 334 {
 335         return ovl_redirect_dir_def ? "on" : "off";
 336 }
 337
 338 static const char * const ovl_xino_str[] = {
 339         "off",
 340         "auto",
 341         "on",
 342 };
 343
 344 static inline int ovl_xino_def(void)
 345 {
 346         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 347 }
 348
 349 /**
 350  * ovl_show_options
 351  *
 352  * Prints the mount options for a given superblock.
 353  * Returns zero; does not fail.
 354  */
 355 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 356 {
 357         struct super_block *sb = dentry->d_sb;
 358         struct ovl_fs *ofs = sb->s_fs_info;
 359
 360         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 361         if (ofs->config.upperdir) {
 362                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 363                 seq_show_option(m, "workdir", ofs->config.workdir);
 364         }
 365         if (ofs->config.default_permissions)
 366                 seq_puts(m, ",default_permissions");
 367         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 368                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 369         if (ofs->config.index != ovl_index_def)
 370                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 371         if (!ofs->config.uuid)
 372                 seq_puts(m, ",uuid=off");
 373         if (ofs->config.nfs_export != ovl_nfs_export_def)
 374                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 375                                                 "on" : "off");
 376         if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
 377                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 378         if (ofs->config.metacopy != ovl_metacopy_def)
 379                 seq_printf(m, ",metacopy=%s",
 380                            ofs->config.metacopy ? "on" : "off");
 381         if (ofs->config.ovl_volatile)
 382                 seq_puts(m, ",volatile");
 383         return 0;
 384 }
 385
 386 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 387 {
 388         struct ovl_fs *ofs = sb->s_fs_info;
 389         struct super_block *upper_sb;
 390         int ret = 0;
 391
 392         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 393                 return -EROFS;
 394
 395         if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
 396                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 397                 if (ovl_should_sync(ofs)) {
 398                         down_read(&upper_sb->s_umount);
 399                         ret = sync_filesystem(upper_sb);
 400                         up_read(&upper_sb->s_umount);
 401                 }
 402         }
 403
 404         return ret;
 405 }
 406
 407 static const struct super_operations ovl_super_operations = {
 408         .alloc_inode    = ovl_alloc_inode,
 409         .free_inode     = ovl_free_inode,
 410         .destroy_inode  = ovl_destroy_inode,
 411         .drop_inode     = generic_delete_inode,
 412         .put_super      = ovl_put_super,
 413         .sync_fs        = ovl_sync_fs,
 414         .statfs         = ovl_statfs,
 415         .show_options   = ovl_show_options,
 416         .remount_fs     = ovl_remount,
 417 };
 418
 419 enum {
 420         OPT_LOWERDIR,
 421         OPT_UPPERDIR,
 422         OPT_WORKDIR,
 423         OPT_DEFAULT_PERMISSIONS,
 424         OPT_REDIRECT_DIR,
 425         OPT_INDEX_ON,
 426         OPT_INDEX_OFF,
 427         OPT_UUID_ON,
 428         OPT_UUID_OFF,
 429         OPT_NFS_EXPORT_ON,
 430         OPT_USERXATTR,
 431         OPT_NFS_EXPORT_OFF,
 432         OPT_XINO_ON,
 433         OPT_XINO_OFF,
 434         OPT_XINO_AUTO,
 435         OPT_METACOPY_ON,
 436         OPT_METACOPY_OFF,
 437         OPT_VOLATILE,
 438         OPT_ERR,
 439 };
 440
 441 static const match_table_t ovl_tokens = {
 442         {OPT_LOWERDIR,                  "lowerdir=%s"},
 443         {OPT_UPPERDIR,                  "upperdir=%s"},
 444         {OPT_WORKDIR,                   "workdir=%s"},
 445         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 446         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 447         {OPT_INDEX_ON,                  "index=on"},
 448         {OPT_INDEX_OFF,                 "index=off"},
 449         {OPT_USERXATTR,                 "userxattr"},
 450         {OPT_UUID_ON,                   "uuid=on"},
 451         {OPT_UUID_OFF,                  "uuid=off"},
 452         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 453         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 454         {OPT_XINO_ON,                   "xino=on"},
 455         {OPT_XINO_OFF,                  "xino=off"},
 456         {OPT_XINO_AUTO,                 "xino=auto"},
 457         {OPT_METACOPY_ON,               "metacopy=on"},
 458         {OPT_METACOPY_OFF,              "metacopy=off"},
 459         {OPT_VOLATILE,                  "volatile"},
 460         {OPT_ERR,                       NULL}
 461 };
 462
 463 static char *ovl_next_opt(char **s)
 464 {
 465         char *sbegin = *s;
 466         char *p;
 467
 468         if (sbegin == NULL)
 469                 return NULL;
 470
 471         for (p = sbegin; *p; p++) {
 472                 if (*p == '\\') {
 473                         p++;
 474                         if (!*p)
 475                                 break;
 476                 } else if (*p == ',') {
 477                         *p = '\0';
 478                         *s = p + 1;
 479                         return sbegin;
 480                 }
 481         }
 482         *s = NULL;
 483         return sbegin;
 484 }
 485
 486 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 487 {
 488         if (strcmp(mode, "on") == 0) {
 489                 config->redirect_dir = true;
 490                 /*
 491                  * Does not make sense to have redirect creation without
 492                  * redirect following.
 493                  */
 494                 config->redirect_follow = true;
 495         } else if (strcmp(mode, "follow") == 0) {
 496                 config->redirect_follow = true;
 497         } else if (strcmp(mode, "off") == 0) {
 498                 if (ovl_redirect_always_follow)
 499                         config->redirect_follow = true;
 500         } else if (strcmp(mode, "nofollow") != 0) {
 501                 pr_err("bad mount option \"redirect_dir=%s\"\n",
 502                        mode);
 503                 return -EINVAL;
 504         }
 505
 506         return 0;
 507 }
 508
 509 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 510 {
 511         char *p;
 512         int err;
 513         bool metacopy_opt = false, redirect_opt = false;
 514         bool nfs_export_opt = false, index_opt = false;
 515
 516         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 517         if (!config->redirect_mode)
 518                 return -ENOMEM;
 519
 520         while ((p = ovl_next_opt(&opt)) != NULL) {
 521                 int token;
 522                 substring_t args[MAX_OPT_ARGS];
 523
 524                 if (!*p)
 525                         continue;
 526
 527                 token = match_token(p, ovl_tokens, args);
 528                 switch (token) {
 529                 case OPT_UPPERDIR:
 530                         kfree(config->upperdir);
 531                         config->upperdir = match_strdup(&args[0]);
 532                         if (!config->upperdir)
 533                                 return -ENOMEM;
 534                         break;
 535
 536                 case OPT_LOWERDIR:
 537                         kfree(config->lowerdir);
 538                         config->lowerdir = match_strdup(&args[0]);
 539                         if (!config->lowerdir)
 540                                 return -ENOMEM;
 541                         break;
 542
 543                 case OPT_WORKDIR:
 544                         kfree(config->workdir);
 545                         config->workdir = match_strdup(&args[0]);
 546                         if (!config->workdir)
 547                                 return -ENOMEM;
 548                         break;
 549
 550                 case OPT_DEFAULT_PERMISSIONS:
 551                         config->default_permissions = true;
 552                         break;
 553
 554                 case OPT_REDIRECT_DIR:
 555                         kfree(config->redirect_mode);
 556                         config->redirect_mode = match_strdup(&args[0]);
 557                         if (!config->redirect_mode)
 558                                 return -ENOMEM;
 559                         redirect_opt = true;
 560                         break;
 561
 562                 case OPT_INDEX_ON:
 563                         config->index = true;
 564                         index_opt = true;
 565                         break;
 566
 567                 case OPT_INDEX_OFF:
 568                         config->index = false;
 569                         index_opt = true;
 570                         break;
 571
 572                 case OPT_UUID_ON:
 573                         config->uuid = true;
 574                         break;
 575
 576                 case OPT_UUID_OFF:
 577                         config->uuid = false;
 578                         break;
 579
 580                 case OPT_NFS_EXPORT_ON:
 581                         config->nfs_export = true;
 582                         nfs_export_opt = true;
 583                         break;
 584
 585                 case OPT_NFS_EXPORT_OFF:
 586                         config->nfs_export = false;
 587                         nfs_export_opt = true;
 588                         break;
 589
 590                 case OPT_XINO_ON:
 591                         config->xino = OVL_XINO_ON;
 592                         break;
 593
 594                 case OPT_XINO_OFF:
 595                         config->xino = OVL_XINO_OFF;
 596                         break;
 597
 598                 case OPT_XINO_AUTO:
 599                         config->xino = OVL_XINO_AUTO;
 600                         break;
 601
 602                 case OPT_METACOPY_ON:
 603                         config->metacopy = true;
 604                         metacopy_opt = true;
 605                         break;
 606
 607                 case OPT_METACOPY_OFF:
 608                         config->metacopy = false;
 609                         metacopy_opt = true;
 610                         break;
 611
 612                 case OPT_VOLATILE:
 613                         config->ovl_volatile = true;
 614                         break;
 615
 616                 case OPT_USERXATTR:
 617                         config->userxattr = true;
 618                         break;
 619
 620                 default:
 621                         pr_err("unrecognized mount option \"%s\" or missing value\n",
 622                                         p);
 623                         return -EINVAL;
 624                 }
 625         }
 626
 627         /* Workdir/index are useless in non-upper mount */
 628         if (!config->upperdir) {
 629                 if (config->workdir) {
 630                         pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 631                                 config->workdir);
 632                         kfree(config->workdir);
 633                         config->workdir = NULL;
 634                 }
 635                 if (config->index && index_opt) {
 636                         pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
 637                         index_opt = false;
 638                 }
 639                 config->index = false;
 640         }
 641
 642         if (!config->upperdir && config->ovl_volatile) {
 643                 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
 644                 config->ovl_volatile = false;
 645         }
 646
 647         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 648         if (err)
 649                 return err;
 650
 651         /*
 652          * This is to make the logic below simpler.  It doesn't make any other
 653          * difference, since config->redirect_dir is only used for upper.
 654          */
 655         if (!config->upperdir && config->redirect_follow)
 656                 config->redirect_dir = true;
 657
 658         /* Resolve metacopy -> redirect_dir dependency */
 659         if (config->metacopy && !config->redirect_dir) {
 660                 if (metacopy_opt && redirect_opt) {
 661                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
 662                                config->redirect_mode);
 663                         return -EINVAL;
 664                 }
 665                 if (redirect_opt) {
 666                         /*
 667                          * There was an explicit redirect_dir=... that resulted
 668                          * in this conflict.
 669                          */
 670                         pr_info("disabling metacopy due to redirect_dir=%s\n",
 671                                 config->redirect_mode);
 672                         config->metacopy = false;
 673                 } else {
 674                         /* Automatically enable redirect otherwise. */
 675                         config->redirect_follow = config->redirect_dir = true;
 676                 }
 677         }
 678
 679         /* Resolve nfs_export -> index dependency */
 680         if (config->nfs_export && !config->index) {
 681                 if (!config->upperdir && config->redirect_follow) {
 682                         pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
 683                         config->nfs_export = false;
 684                 } else if (nfs_export_opt && index_opt) {
 685                         pr_err("conflicting options: nfs_export=on,index=off\n");
 686                         return -EINVAL;
 687                 } else if (index_opt) {
 688                         /*
 689                          * There was an explicit index=off that resulted
 690                          * in this conflict.
 691                          */
 692                         pr_info("disabling nfs_export due to index=off\n");
 693                         config->nfs_export = false;
 694                 } else {
 695                         /* Automatically enable index otherwise. */
 696                         config->index = true;
 697                 }
 698         }
 699
 700         /* Resolve nfs_export -> !metacopy dependency */
 701         if (config->nfs_export && config->metacopy) {
 702                 if (nfs_export_opt && metacopy_opt) {
 703                         pr_err("conflicting options: nfs_export=on,metacopy=on\n");
 704                         return -EINVAL;
 705                 }
 706                 if (metacopy_opt) {
 707                         /*
 708                          * There was an explicit metacopy=on that resulted
 709                          * in this conflict.
 710                          */
 711                         pr_info("disabling nfs_export due to metacopy=on\n");
 712                         config->nfs_export = false;
 713                 } else {
 714                         /*
 715                          * There was an explicit nfs_export=on that resulted
 716                          * in this conflict.
 717                          */
 718                         pr_info("disabling metacopy due to nfs_export=on\n");
 719                         config->metacopy = false;
 720                 }
 721         }
 722
 723
 724         /* Resolve userxattr -> !redirect && !metacopy dependency */
 725         if (config->userxattr) {
 726                 if (config->redirect_follow && redirect_opt) {
 727                         pr_err("conflicting options: userxattr,redirect_dir=%s\n",
 728                                config->redirect_mode);
 729                         return -EINVAL;
 730                 }
 731                 if (config->metacopy && metacopy_opt) {
 732                         pr_err("conflicting options: userxattr,metacopy=on\n");
 733                         return -EINVAL;
 734                 }
 735                 /*
 736                  * Silently disable default setting of redirect and metacopy.
 737                  * This shall be the default in the future as well: these
 738                  * options must be explicitly enabled if used together with
 739                  * userxattr.
 740                  */
 741                 config->redirect_dir = config->redirect_follow = false;
 742                 config->metacopy = false;
 743         }
 744
 745         return 0;
 746 }
 747
 748 #define OVL_WORKDIR_NAME "work"
 749 #define OVL_INDEXDIR_NAME "index"
 750
 751 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 752                                          const char *name, bool persist)
 753 {
 754         struct inode *dir =  ofs->workbasedir->d_inode;
 755         struct vfsmount *mnt = ovl_upper_mnt(ofs);
 756         struct dentry *work;
 757         int err;
 758         bool retried = false;
 759
 760         inode_lock_nested(dir, I_MUTEX_PARENT);
 761 retry:
 762         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
 763
 764         if (!IS_ERR(work)) {
 765                 struct iattr attr = {
 766                         .ia_valid = ATTR_MODE,
 767                         .ia_mode = S_IFDIR | 0,
 768                 };
 769
 770                 if (work->d_inode) {
 771                         err = -EEXIST;
 772                         if (retried)
 773                                 goto out_dput;
 774
 775                         if (persist)
 776                                 goto out_unlock;
 777
 778                         retried = true;
 779                         err = ovl_workdir_cleanup(dir, mnt, work, 0);
 780                         dput(work);
 781                         if (err == -EINVAL) {
 782                                 work = ERR_PTR(err);
 783                                 goto out_unlock;
 784                         }
 785                         goto retry;
 786                 }
 787
 788                 work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
 789                 err = PTR_ERR(work);
 790                 if (IS_ERR(work))
 791                         goto out_err;
 792
 793                 /*
 794                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 795                  *
 796                  * a) success (there was a POSIX ACL xattr and was removed)
 797                  * b) -ENODATA (there was no POSIX ACL xattr)
 798                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 799                  *
 800                  * There are various other error values that could effectively
 801                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 802                  * if the xattr name is too long), but the set of filesystems
 803                  * allowed as upper are limited to "normal" ones, where checking
 804                  * for the above two errors is sufficient.
 805                  */
 806                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 807                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 808                         goto out_dput;
 809
 810                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 811                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 812                         goto out_dput;
 813
 814                 /* Clear any inherited mode bits */
 815                 inode_lock(work->d_inode);
 816                 err = notify_change(work, &attr, NULL);
 817                 inode_unlock(work->d_inode);
 818                 if (err)
 819                         goto out_dput;
 820         } else {
 821                 err = PTR_ERR(work);
 822                 goto out_err;
 823         }
 824 out_unlock:
 825         inode_unlock(dir);
 826         return work;
 827
 828 out_dput:
 829         dput(work);
 830 out_err:
 831         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
 832                 ofs->config.workdir, name, -err);
 833         work = NULL;
 834         goto out_unlock;
 835 }
 836
 837 static void ovl_unescape(char *s)
 838 {
 839         char *d = s;
 840
 841         for (;; s++, d++) {
 842                 if (*s == '\\')
 843                         s++;
 844                 *d = *s;
 845                 if (!*s)
 846                         break;
 847         }
 848 }
 849
 850 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 851 {
 852         int err = -EINVAL;
 853
 854         if (!*name) {
 855                 pr_err("empty lowerdir\n");
 856                 goto out;
 857         }
 858         err = kern_path(name, LOOKUP_FOLLOW, path);
 859         if (err) {
 860                 pr_err("failed to resolve '%s': %i\n", name, err);
 861                 goto out;
 862         }
 863         err = -EINVAL;
 864         if (ovl_dentry_weird(path->dentry)) {
 865                 pr_err("filesystem on '%s' not supported\n", name);
 866                 goto out_put;
 867         }
 868         if (!d_is_dir(path->dentry)) {
 869                 pr_err("'%s' not a directory\n", name);
 870                 goto out_put;
 871         }
 872         return 0;
 873
 874 out_put:
 875         path_put_init(path);
 876 out:
 877         return err;
 878 }
 879
 880 static int ovl_mount_dir(const char *name, struct path *path)
 881 {
 882         int err = -ENOMEM;
 883         char *tmp = kstrdup(name, GFP_KERNEL);
 884
 885         if (tmp) {
 886                 ovl_unescape(tmp);
 887                 err = ovl_mount_dir_noesc(tmp, path);
 888
 889                 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
 890                         pr_err("filesystem on '%s' not supported as upperdir\n",
 891                                tmp);
 892                         path_put_init(path);
 893                         err = -EINVAL;
 894                 }
 895                 kfree(tmp);
 896         }
 897         return err;
 898 }
 899
 900 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 901                              const char *name)
 902 {
 903         struct kstatfs statfs;
 904         int err = vfs_statfs(path, &statfs);
 905
 906         if (err)
 907                 pr_err("statfs failed on '%s'\n", name);
 908         else
 909                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 910
 911         return err;
 912 }
 913
 914 static int ovl_lower_dir(const char *name, struct path *path,
 915                          struct ovl_fs *ofs, int *stack_depth)
 916 {
 917         int fh_type;
 918         int err;
 919
 920         err = ovl_mount_dir_noesc(name, path);
 921         if (err)
 922                 return err;
 923
 924         err = ovl_check_namelen(path, ofs, name);
 925         if (err)
 926                 return err;
 927
 928         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 929
 930         /*
 931          * The inodes index feature and NFS export need to encode and decode
 932          * file handles, so they require that all layers support them.
 933          */
 934         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 935         if ((ofs->config.nfs_export ||
 936              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 937                 ofs->config.index = false;
 938                 ofs->config.nfs_export = false;
 939                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 940                         name);
 941         }
 942
 943         /* Check if lower fs has 32bit inode numbers */
 944         if (fh_type != FILEID_INO32_GEN)
 945                 ofs->xino_mode = -1;
 946
 947         return 0;
 948 }
 949
 950 /* Workdir should not be subdir of upperdir and vice versa */
 951 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 952 {
 953         bool ok = false;
 954
 955         if (workdir != upperdir) {
 956                 ok = (lock_rename(workdir, upperdir) == NULL);
 957                 unlock_rename(workdir, upperdir);
 958         }
 959         return ok;
 960 }
 961
 962 static unsigned int ovl_split_lowerdirs(char *str)
 963 {
 964         unsigned int ctr = 1;
 965         char *s, *d;
 966
 967         for (s = d = str;; s++, d++) {
 968                 if (*s == '\\') {
 969                         s++;
 970                 } else if (*s == ':') {
 971                         *d = '\0';
 972                         ctr++;
 973                         continue;
 974                 }
 975                 *d = *s;
 976                 if (!*s)
 977                         break;
 978         }
 979         return ctr;
 980 }
 981
 982 static int __maybe_unused
 983 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 984                         struct dentry *dentry, struct inode *inode,
 985                         const char *name, void *buffer, size_t size)
 986 {
 987         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 988 }
 989
 990 static int __maybe_unused
 991 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 992                         struct dentry *dentry, struct inode *inode,
 993                         const char *name, const void *value,
 994                         size_t size, int flags)
 995 {
 996         struct dentry *workdir = ovl_workdir(dentry);
 997         struct inode *realinode = ovl_inode_real(inode);
 998         struct posix_acl *acl = NULL;
 999         int err;
1000
1001         /* Check that everything is OK before copy-up */
1002         if (value) {
1003                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
1004                 if (IS_ERR(acl))
1005                         return PTR_ERR(acl);
1006         }
1007         err = -EOPNOTSUPP;
1008         if (!IS_POSIXACL(d_inode(workdir)))
1009                 goto out_acl_release;
1010         if (!realinode->i_op->set_acl)
1011                 goto out_acl_release;
1012         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1013                 err = acl ? -EACCES : 0;
1014                 goto out_acl_release;
1015         }
1016         err = -EPERM;
1017         if (!inode_owner_or_capable(inode))
1018                 goto out_acl_release;
1019
1020         posix_acl_release(acl);
1021
1022         /*
1023          * Check if sgid bit needs to be cleared (actual setacl operation will
1024          * be done with mounter's capabilities and so that won't do it for us).
1025          */
1026         if (unlikely(inode->i_mode & S_ISGID) &&
1027             handler->flags == ACL_TYPE_ACCESS &&
1028             !in_group_p(inode->i_gid) &&
1029             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
1030                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
1031
1032                 err = ovl_setattr(dentry, &iattr);
1033                 if (err)
1034                         return err;
1035         }
1036
1037         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
1038         if (!err)
1039                 ovl_copyattr(ovl_inode_real(inode), inode);
1040
1041         return err;
1042
1043 out_acl_release:
1044         posix_acl_release(acl);
1045         return err;
1046 }
1047
1048 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1049                              struct dentry *dentry, struct inode *inode,
1050                              const char *name, void *buffer, size_t size)
1051 {
1052         return -EOPNOTSUPP;
1053 }
1054
1055 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1056                              struct dentry *dentry, struct inode *inode,
1057                              const char *name, const void *value,
1058                              size_t size, int flags)
1059 {
1060         return -EOPNOTSUPP;
1061 }
1062
1063 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1064                                struct dentry *dentry, struct inode *inode,
1065                                const char *name, void *buffer, size_t size)
1066 {
1067         return ovl_xattr_get(dentry, inode, name, buffer, size);
1068 }
1069
1070 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1071                                struct dentry *dentry, struct inode *inode,
1072                                const char *name, const void *value,
1073                                size_t size, int flags)
1074 {
1075         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1076 }
1077
1078 static const struct xattr_handler __maybe_unused
1079 ovl_posix_acl_access_xattr_handler = {
1080         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1081         .flags = ACL_TYPE_ACCESS,
1082         .get = ovl_posix_acl_xattr_get,
1083         .set = ovl_posix_acl_xattr_set,
1084 };
1085
1086 static const struct xattr_handler __maybe_unused
1087 ovl_posix_acl_default_xattr_handler = {
1088         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1089         .flags = ACL_TYPE_DEFAULT,
1090         .get = ovl_posix_acl_xattr_get,
1091         .set = ovl_posix_acl_xattr_set,
1092 };
1093
1094 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1095         .prefix = OVL_XATTR_TRUSTED_PREFIX,
1096         .get = ovl_own_xattr_get,
1097         .set = ovl_own_xattr_set,
1098 };
1099
1100 static const struct xattr_handler ovl_own_user_xattr_handler = {
1101         .prefix = OVL_XATTR_USER_PREFIX,
1102         .get = ovl_own_xattr_get,
1103         .set = ovl_own_xattr_set,
1104 };
1105
1106 static const struct xattr_handler ovl_other_xattr_handler = {
1107         .prefix = "", /* catch all */
1108         .get = ovl_other_xattr_get,
1109         .set = ovl_other_xattr_set,
1110 };
1111
1112 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1113 #ifdef CONFIG_FS_POSIX_ACL
1114         &ovl_posix_acl_access_xattr_handler,
1115         &ovl_posix_acl_default_xattr_handler,
1116 #endif
1117         &ovl_own_trusted_xattr_handler,
1118         &ovl_other_xattr_handler,
1119         NULL
1120 };
1121
1122 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1123 #ifdef CONFIG_FS_POSIX_ACL
1124         &ovl_posix_acl_access_xattr_handler,
1125         &ovl_posix_acl_default_xattr_handler,
1126 #endif
1127         &ovl_own_user_xattr_handler,
1128         &ovl_other_xattr_handler,
1129         NULL
1130 };
1131
1132 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1133                           struct inode **ptrap, const char *name)
1134 {
1135         struct inode *trap;
1136         int err;
1137
1138         trap = ovl_get_trap_inode(sb, dir);
1139         err = PTR_ERR_OR_ZERO(trap);
1140         if (err) {
1141                 if (err == -ELOOP)
1142                         pr_err("conflicting %s path\n", name);
1143                 return err;
1144         }
1145
1146         *ptrap = trap;
1147         return 0;
1148 }
1149
1150 /*
1151  * Determine how we treat concurrent use of upperdir/workdir based on the
1152  * index feature. This is papering over mount leaks of container runtimes,
1153  * for example, an old overlay mount is leaked and now its upperdir is
1154  * attempted to be used as a lower layer in a new overlay mount.
1155  */
1156 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1157 {
1158         if (ofs->config.index) {
1159                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1160                        name);
1161                 return -EBUSY;
1162         } else {
1163                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1164                         name);
1165                 return 0;
1166         }
1167 }
1168
1169 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1170                          struct ovl_layer *upper_layer, struct path *upperpath)
1171 {
1172         struct vfsmount *upper_mnt;
1173         int err;
1174
1175         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1176         if (err)
1177                 goto out;
1178
1179         /* Upper fs should not be r/o */
1180         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
1181                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1182                 err = -EINVAL;
1183                 goto out;
1184         }
1185
1186         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1187         if (err)
1188                 goto out;
1189
1190         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1191                              "upperdir");
1192         if (err)
1193                 goto out;
1194
1195         upper_mnt = clone_private_mount(upperpath);
1196         err = PTR_ERR(upper_mnt);
1197         if (IS_ERR(upper_mnt)) {
1198                 pr_err("failed to clone upperpath\n");
1199                 goto out;
1200         }
1201
1202         /* Don't inherit atime flags */
1203         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1204         upper_layer->mnt = upper_mnt;
1205         upper_layer->idx = 0;
1206         upper_layer->fsid = 0;
1207
1208         /*
1209          * Inherit SB_NOSEC flag from upperdir.
1210          *
1211          * This optimization changes behavior when a security related attribute
1212          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1213          * okay because we don't yet have guarantees in that case, but it will
1214          * need careful treatment once we want to honour changes to underlying
1215          * filesystems.
1216          */
1217         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1218                 sb->s_flags |= SB_NOSEC;
1219
1220         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1221                 ofs->upperdir_locked = true;
1222         } else {
1223                 err = ovl_report_in_use(ofs, "upperdir");
1224                 if (err)
1225                         goto out;
1226         }
1227
1228         err = 0;
1229 out:
1230         return err;
1231 }
1232
1233 /*
1234  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1235  * negative values if error is encountered.
1236  */
1237 static int ovl_check_rename_whiteout(struct dentry *workdir)
1238 {
1239         struct inode *dir = d_inode(workdir);
1240         struct dentry *temp;
1241         struct dentry *dest;
1242         struct dentry *whiteout;
1243         struct name_snapshot name;
1244         int err;
1245
1246         inode_lock_nested(dir, I_MUTEX_PARENT);
1247
1248         temp = ovl_create_temp(workdir, OVL_CATTR(S_IFREG | 0));
1249         err = PTR_ERR(temp);
1250         if (IS_ERR(temp))
1251                 goto out_unlock;
1252
1253         dest = ovl_lookup_temp(workdir);
1254         err = PTR_ERR(dest);
1255         if (IS_ERR(dest)) {
1256                 dput(temp);
1257                 goto out_unlock;
1258         }
1259
1260         /* Name is inline and stable - using snapshot as a copy helper */
1261         take_dentry_name_snapshot(&name, temp);
1262         err = ovl_do_rename(dir, temp, dir, dest, RENAME_WHITEOUT);
1263         if (err) {
1264                 if (err == -EINVAL)
1265                         err = 0;
1266                 goto cleanup_temp;
1267         }
1268
1269         whiteout = lookup_one_len(name.name.name, workdir, name.name.len);
1270         err = PTR_ERR(whiteout);
1271         if (IS_ERR(whiteout))
1272                 goto cleanup_temp;
1273
1274         err = ovl_is_whiteout(whiteout);
1275
1276         /* Best effort cleanup of whiteout and temp file */
1277         if (err)
1278                 ovl_cleanup(dir, whiteout);
1279         dput(whiteout);
1280
1281 cleanup_temp:
1282         ovl_cleanup(dir, temp);
1283         release_dentry_name_snapshot(&name);
1284         dput(temp);
1285         dput(dest);
1286
1287 out_unlock:
1288         inode_unlock(dir);
1289
1290         return err;
1291 }
1292
1293 static struct dentry *ovl_lookup_or_create(struct dentry *parent,
1294                                            const char *name, umode_t mode)
1295 {
1296         size_t len = strlen(name);
1297         struct dentry *child;
1298
1299         inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1300         child = lookup_one_len(name, parent, len);
1301         if (!IS_ERR(child) && !child->d_inode)
1302                 child = ovl_create_real(parent->d_inode, child,
1303                                         OVL_CATTR(mode));
1304         inode_unlock(parent->d_inode);
1305         dput(parent);
1306
1307         return child;
1308 }
1309
1310 /*
1311  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1312  * present.
1313  */
1314 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1315 {
1316         unsigned int ctr;
1317         struct dentry *d = dget(ofs->workbasedir);
1318         static const char *const volatile_path[] = {
1319                 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1320         };
1321         const char *const *name = volatile_path;
1322
1323         for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1324                 d = ovl_lookup_or_create(d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1325                 if (IS_ERR(d))
1326                         return PTR_ERR(d);
1327         }
1328         dput(d);
1329         return 0;
1330 }
1331
1332 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1333                             struct path *workpath)
1334 {
1335         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1336         struct dentry *temp, *workdir;
1337         bool rename_whiteout;
1338         bool d_type;
1339         int fh_type;
1340         int err;
1341
1342         err = mnt_want_write(mnt);
1343         if (err)
1344                 return err;
1345
1346         workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1347         err = PTR_ERR(workdir);
1348         if (IS_ERR_OR_NULL(workdir))
1349                 goto out;
1350
1351         ofs->workdir = workdir;
1352
1353         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1354         if (err)
1355                 goto out;
1356
1357         /*
1358          * Upper should support d_type, else whiteouts are visible.  Given
1359          * workdir and upper are on same fs, we can do iterate_dir() on
1360          * workdir. This check requires successful creation of workdir in
1361          * previous step.
1362          */
1363         err = ovl_check_d_type_supported(workpath);
1364         if (err < 0)
1365                 goto out;
1366
1367         d_type = err;
1368         if (!d_type)
1369                 pr_warn("upper fs needs to support d_type.\n");
1370
1371         /* Check if upper/work fs supports O_TMPFILE */
1372         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1373         ofs->tmpfile = !IS_ERR(temp);
1374         if (ofs->tmpfile)
1375                 dput(temp);
1376         else
1377                 pr_warn("upper fs does not support tmpfile.\n");
1378
1379
1380         /* Check if upper/work fs supports RENAME_WHITEOUT */
1381         err = ovl_check_rename_whiteout(ofs->workdir);
1382         if (err < 0)
1383                 goto out;
1384
1385         rename_whiteout = err;
1386         if (!rename_whiteout)
1387                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1388
1389         /*
1390          * Check if upper/work fs supports (trusted|user).overlay.* xattr
1391          */
1392         err = ovl_do_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1393         if (err) {
1394                 ofs->noxattr = true;
1395                 ofs->config.index = false;
1396                 ofs->config.metacopy = false;
1397                 pr_warn("upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1398                 err = 0;
1399         } else {
1400                 ovl_do_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1401         }
1402
1403         /*
1404          * We allowed sub-optimal upper fs configuration and don't want to break
1405          * users over kernel upgrade, but we never allowed remote upper fs, so
1406          * we can enforce strict requirements for remote upper fs.
1407          */
1408         if (ovl_dentry_remote(ofs->workdir) &&
1409             (!d_type || !rename_whiteout || ofs->noxattr)) {
1410                 pr_err("upper fs missing required features.\n");
1411                 err = -EINVAL;
1412                 goto out;
1413         }
1414
1415         /*
1416          * For volatile mount, create a incompat/volatile/dirty file to keep
1417          * track of it.
1418          */
1419         if (ofs->config.ovl_volatile) {
1420                 err = ovl_create_volatile_dirty(ofs);
1421                 if (err < 0) {
1422                         pr_err("Failed to create volatile/dirty file.\n");
1423                         goto out;
1424                 }
1425         }
1426
1427         /* Check if upper/work fs supports file handles */
1428         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1429         if (ofs->config.index && !fh_type) {
1430                 ofs->config.index = false;
1431                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1432         }
1433
1434         /* Check if upper fs has 32bit inode numbers */
1435         if (fh_type != FILEID_INO32_GEN)
1436                 ofs->xino_mode = -1;
1437
1438         /* NFS export of r/w mount depends on index */
1439         if (ofs->config.nfs_export && !ofs->config.index) {
1440                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1441                 ofs->config.nfs_export = false;
1442         }
1443 out:
1444         mnt_drop_write(mnt);
1445         return err;
1446 }
1447
1448 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1449                            struct path *upperpath)
1450 {
1451         int err;
1452         struct path workpath = { };
1453
1454         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1455         if (err)
1456                 goto out;
1457
1458         err = -EINVAL;
1459         if (upperpath->mnt != workpath.mnt) {
1460                 pr_err("workdir and upperdir must reside under the same mount\n");
1461                 goto out;
1462         }
1463         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1464                 pr_err("workdir and upperdir must be separate subtrees\n");
1465                 goto out;
1466         }
1467
1468         ofs->workbasedir = dget(workpath.dentry);
1469
1470         if (ovl_inuse_trylock(ofs->workbasedir)) {
1471                 ofs->workdir_locked = true;
1472         } else {
1473                 err = ovl_report_in_use(ofs, "workdir");
1474                 if (err)
1475                         goto out;
1476         }
1477
1478         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1479                              "workdir");
1480         if (err)
1481                 goto out;
1482
1483         err = ovl_make_workdir(sb, ofs, &workpath);
1484
1485 out:
1486         path_put(&workpath);
1487
1488         return err;
1489 }
1490
1491 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1492                             struct ovl_entry *oe, struct path *upperpath)
1493 {
1494         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1495         struct dentry *indexdir;
1496         int err;
1497
1498         err = mnt_want_write(mnt);
1499         if (err)
1500                 return err;
1501
1502         /* Verify lower root is upper root origin */
1503         err = ovl_verify_origin(ofs, upperpath->dentry,
1504                                 oe->lowerstack[0].dentry, true);
1505         if (err) {
1506                 pr_err("failed to verify upper root origin\n");
1507                 goto out;
1508         }
1509
1510         /* index dir will act also as workdir */
1511         iput(ofs->workdir_trap);
1512         ofs->workdir_trap = NULL;
1513         dput(ofs->workdir);
1514         ofs->workdir = NULL;
1515         indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1516         if (IS_ERR(indexdir)) {
1517                 err = PTR_ERR(indexdir);
1518         } else if (indexdir) {
1519                 ofs->indexdir = indexdir;
1520                 ofs->workdir = dget(indexdir);
1521
1522                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1523                                      "indexdir");
1524                 if (err)
1525                         goto out;
1526
1527                 /*
1528                  * Verify upper root is exclusively associated with index dir.
1529                  * Older kernels stored upper fh in ".overlay.origin"
1530                  * xattr. If that xattr exists, verify that it is a match to
1531                  * upper dir file handle. In any case, verify or set xattr
1532                  * ".overlay.upper" to indicate that index may have
1533                  * directory entries.
1534                  */
1535                 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1536                         err = ovl_verify_set_fh(ofs, ofs->indexdir,
1537                                                 OVL_XATTR_ORIGIN,
1538                                                 upperpath->dentry, true, false);
1539                         if (err)
1540                                 pr_err("failed to verify index dir 'origin' xattr\n");
1541                 }
1542                 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1543                                        true);
1544                 if (err)
1545                         pr_err("failed to verify index dir 'upper' xattr\n");
1546
1547                 /* Cleanup bad/stale/orphan index entries */
1548                 if (!err)
1549                         err = ovl_indexdir_cleanup(ofs);
1550         }
1551         if (err || !ofs->indexdir)
1552                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1553
1554 out:
1555         mnt_drop_write(mnt);
1556         return err;
1557 }
1558
1559 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1560 {
1561         unsigned int i;
1562
1563         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1564                 return true;
1565
1566         /*
1567          * We allow using single lower with null uuid for index and nfs_export
1568          * for example to support those features with single lower squashfs.
1569          * To avoid regressions in setups of overlay with re-formatted lower
1570          * squashfs, do not allow decoding origin with lower null uuid unless
1571          * user opted-in to one of the new features that require following the
1572          * lower inode of non-dir upper.
1573          */
1574         if (!ofs->config.index && !ofs->config.metacopy && !ofs->config.xino &&
1575             uuid_is_null(uuid))
1576                 return false;
1577
1578         for (i = 0; i < ofs->numfs; i++) {
1579                 /*
1580                  * We use uuid to associate an overlay lower file handle with a
1581                  * lower layer, so we can accept lower fs with null uuid as long
1582                  * as all lower layers with null uuid are on the same fs.
1583                  * if we detect multiple lower fs with the same uuid, we
1584                  * disable lower file handle decoding on all of them.
1585                  */
1586                 if (ofs->fs[i].is_lower &&
1587                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1588                         ofs->fs[i].bad_uuid = true;
1589                         return false;
1590                 }
1591         }
1592         return true;
1593 }
1594
1595 /* Get a unique fsid for the layer */
1596 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1597 {
1598         struct super_block *sb = path->mnt->mnt_sb;
1599         unsigned int i;
1600         dev_t dev;
1601         int err;
1602         bool bad_uuid = false;
1603
1604         for (i = 0; i < ofs->numfs; i++) {
1605                 if (ofs->fs[i].sb == sb)
1606                         return i;
1607         }
1608
1609         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1610                 bad_uuid = true;
1611                 if (ofs->config.index || ofs->config.nfs_export) {
1612                         ofs->config.index = false;
1613                         ofs->config.nfs_export = false;
1614                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to index=off,nfs_export=off.\n",
1615                                 uuid_is_null(&sb->s_uuid) ? "null" :
1616                                                             "conflicting",
1617                                 path->dentry);
1618                 }
1619         }
1620
1621         err = get_anon_bdev(&dev);
1622         if (err) {
1623                 pr_err("failed to get anonymous bdev for lowerpath\n");
1624                 return err;
1625         }
1626
1627         ofs->fs[ofs->numfs].sb = sb;
1628         ofs->fs[ofs->numfs].pseudo_dev = dev;
1629         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1630
1631         return ofs->numfs++;
1632 }
1633
1634 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1635                           struct path *stack, unsigned int numlower,
1636                           struct ovl_layer *layers)
1637 {
1638         int err;
1639         unsigned int i;
1640
1641         err = -ENOMEM;
1642         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1643         if (ofs->fs == NULL)
1644                 goto out;
1645
1646         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1647         ofs->numfs++;
1648
1649         /*
1650          * All lower layers that share the same fs as upper layer, use the same
1651          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1652          * only overlay to simplify ovl_fs_free().
1653          * is_lower will be set if upper fs is shared with a lower layer.
1654          */
1655         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1656         if (err) {
1657                 pr_err("failed to get anonymous bdev for upper fs\n");
1658                 goto out;
1659         }
1660
1661         if (ovl_upper_mnt(ofs)) {
1662                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1663                 ofs->fs[0].is_lower = false;
1664         }
1665
1666         for (i = 0; i < numlower; i++) {
1667                 struct vfsmount *mnt;
1668                 struct inode *trap;
1669                 int fsid;
1670
1671                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1672                 if (err < 0)
1673                         goto out;
1674
1675                 /*
1676                  * Check if lower root conflicts with this overlay layers before
1677                  * checking if it is in-use as upperdir/workdir of "another"
1678                  * mount, because we do not bother to check in ovl_is_inuse() if
1679                  * the upperdir/workdir is in fact in-use by our
1680                  * upperdir/workdir.
1681                  */
1682                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1683                 if (err)
1684                         goto out;
1685
1686                 if (ovl_is_inuse(stack[i].dentry)) {
1687                         err = ovl_report_in_use(ofs, "lowerdir");
1688                         if (err) {
1689                                 iput(trap);
1690                                 goto out;
1691                         }
1692                 }
1693
1694                 mnt = clone_private_mount(&stack[i]);
1695                 err = PTR_ERR(mnt);
1696                 if (IS_ERR(mnt)) {
1697                         pr_err("failed to clone lowerpath\n");
1698                         iput(trap);
1699                         goto out;
1700                 }
1701
1702                 /*
1703                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1704                  * will fail instead of modifying lower fs.
1705                  */
1706                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1707
1708                 layers[ofs->numlayer].trap = trap;
1709                 layers[ofs->numlayer].mnt = mnt;
1710                 layers[ofs->numlayer].idx = ofs->numlayer;
1711                 layers[ofs->numlayer].fsid = fsid;
1712                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1713                 ofs->numlayer++;
1714                 ofs->fs[fsid].is_lower = true;
1715         }
1716
1717         /*
1718          * When all layers on same fs, overlay can use real inode numbers.
1719          * With mount option "xino=<on|auto>", mounter declares that there are
1720          * enough free high bits in underlying fs to hold the unique fsid.
1721          * If overlayfs does encounter underlying inodes using the high xino
1722          * bits reserved for fsid, it emits a warning and uses the original
1723          * inode number or a non persistent inode number allocated from a
1724          * dedicated range.
1725          */
1726         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1727                 if (ofs->config.xino == OVL_XINO_ON)
1728                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1729                 ofs->xino_mode = 0;
1730         } else if (ofs->config.xino == OVL_XINO_OFF) {
1731                 ofs->xino_mode = -1;
1732         } else if (ofs->xino_mode < 0) {
1733                 /*
1734                  * This is a roundup of number of bits needed for encoding
1735                  * fsid, where fsid 0 is reserved for upper fs (even with
1736                  * lower only overlay) +1 extra bit is reserved for the non
1737                  * persistent inode number range that is used for resolving
1738                  * xino lower bits overflow.
1739                  */
1740                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1741                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1742         }
1743
1744         if (ofs->xino_mode > 0) {
1745                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1746                         ofs->xino_mode);
1747         }
1748
1749         err = 0;
1750 out:
1751         return err;
1752 }
1753
1754 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1755                                 const char *lower, unsigned int numlower,
1756                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1757 {
1758         int err;
1759         struct path *stack = NULL;
1760         unsigned int i;
1761         struct ovl_entry *oe;
1762
1763         if (!ofs->config.upperdir && numlower == 1) {
1764                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1765                 return ERR_PTR(-EINVAL);
1766         }
1767
1768         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1769         if (!stack)
1770                 return ERR_PTR(-ENOMEM);
1771
1772         err = -EINVAL;
1773         for (i = 0; i < numlower; i++) {
1774                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1775                 if (err)
1776                         goto out_err;
1777
1778                 lower = strchr(lower, '\0') + 1;
1779         }
1780
1781         err = -EINVAL;
1782         sb->s_stack_depth++;
1783         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1784                 pr_err("maximum fs stacking depth exceeded\n");
1785                 goto out_err;
1786         }
1787
1788         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1789         if (err)
1790                 goto out_err;
1791
1792         err = -ENOMEM;
1793         oe = ovl_alloc_entry(numlower);
1794         if (!oe)
1795                 goto out_err;
1796
1797         for (i = 0; i < numlower; i++) {
1798                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1799                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1800         }
1801
1802 out:
1803         for (i = 0; i < numlower; i++)
1804                 path_put(&stack[i]);
1805         kfree(stack);
1806
1807         return oe;
1808
1809 out_err:
1810         oe = ERR_PTR(err);
1811         goto out;
1812 }
1813
1814 /*
1815  * Check if this layer root is a descendant of:
1816  * - another layer of this overlayfs instance
1817  * - upper/work dir of any overlayfs instance
1818  */
1819 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1820                            struct dentry *dentry, const char *name)
1821 {
1822         struct dentry *next = dentry, *parent;
1823         int err = 0;
1824
1825         if (!dentry)
1826                 return 0;
1827
1828         parent = dget_parent(next);
1829
1830         /* Walk back ancestors to root (inclusive) looking for traps */
1831         while (!err && parent != next) {
1832                 if (ovl_lookup_trap_inode(sb, parent)) {
1833                         err = -ELOOP;
1834                         pr_err("overlapping %s path\n", name);
1835                 } else if (ovl_is_inuse(parent)) {
1836                         err = ovl_report_in_use(ofs, name);
1837                 }
1838                 next = parent;
1839                 parent = dget_parent(next);
1840                 dput(next);
1841         }
1842
1843         dput(parent);
1844
1845         return err;
1846 }
1847
1848 /*
1849  * Check if any of the layers or work dirs overlap.
1850  */
1851 static int ovl_check_overlapping_layers(struct super_block *sb,
1852                                         struct ovl_fs *ofs)
1853 {
1854         int i, err;
1855
1856         if (ovl_upper_mnt(ofs)) {
1857                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1858                                       "upperdir");
1859                 if (err)
1860                         return err;
1861
1862                 /*
1863                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1864                  * this instance and covers overlapping work and index dirs,
1865                  * unless work or index dir have been moved since created inside
1866                  * workbasedir.  In that case, we already have their traps in
1867                  * inode cache and we will catch that case on lookup.
1868                  */
1869                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir");
1870                 if (err)
1871                         return err;
1872         }
1873
1874         for (i = 1; i < ofs->numlayer; i++) {
1875                 err = ovl_check_layer(sb, ofs,
1876                                       ofs->layers[i].mnt->mnt_root,
1877                                       "lowerdir");
1878                 if (err)
1879                         return err;
1880         }
1881
1882         return 0;
1883 }
1884
1885 static struct dentry *ovl_get_root(struct super_block *sb,
1886                                    struct dentry *upperdentry,
1887                                    struct ovl_entry *oe)
1888 {
1889         struct dentry *root;
1890         struct ovl_path *lowerpath = &oe->lowerstack[0];
1891         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1892         int fsid = lowerpath->layer->fsid;
1893         struct ovl_inode_params oip = {
1894                 .upperdentry = upperdentry,
1895                 .lowerpath = lowerpath,
1896         };
1897
1898         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1899         if (!root)
1900                 return NULL;
1901
1902         root->d_fsdata = oe;
1903
1904         if (upperdentry) {
1905                 /* Root inode uses upper st_ino/i_ino */
1906                 ino = d_inode(upperdentry)->i_ino;
1907                 fsid = 0;
1908                 ovl_dentry_set_upper_alias(root);
1909                 if (ovl_is_impuredir(sb, upperdentry))
1910                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1911         }
1912
1913         /* Root is always merge -> can have whiteouts */
1914         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1915         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1916         ovl_set_upperdata(d_inode(root));
1917         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1918         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1919
1920         return root;
1921 }
1922
1923 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1924 {
1925         struct path upperpath = { };
1926         struct dentry *root_dentry;
1927         struct ovl_entry *oe;
1928         struct ovl_fs *ofs;
1929         struct ovl_layer *layers;
1930         struct cred *cred;
1931         char *splitlower = NULL;
1932         unsigned int numlower;
1933         int err;
1934
1935         err = -EIO;
1936         if (WARN_ON(sb->s_user_ns != current_user_ns()))
1937                 goto out;
1938
1939         sb->s_d_op = &ovl_dentry_operations;
1940
1941         err = -ENOMEM;
1942         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1943         if (!ofs)
1944                 goto out;
1945
1946         ofs->creator_cred = cred = prepare_creds();
1947         if (!cred)
1948                 goto out_err;
1949
1950         /* Is there a reason anyone would want not to share whiteouts? */
1951         ofs->share_whiteout = true;
1952
1953         ofs->config.index = ovl_index_def;
1954         ofs->config.uuid = true;
1955         ofs->config.nfs_export = ovl_nfs_export_def;
1956         ofs->config.xino = ovl_xino_def();
1957         ofs->config.metacopy = ovl_metacopy_def;
1958         err = ovl_parse_opt((char *) data, &ofs->config);
1959         if (err)
1960                 goto out_err;
1961
1962         err = -EINVAL;
1963         if (!ofs->config.lowerdir) {
1964                 if (!silent)
1965                         pr_err("missing 'lowerdir'\n");
1966                 goto out_err;
1967         }
1968
1969         err = -ENOMEM;
1970         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1971         if (!splitlower)
1972                 goto out_err;
1973
1974         numlower = ovl_split_lowerdirs(splitlower);
1975         if (numlower > OVL_MAX_STACK) {
1976                 pr_err("too many lower directories, limit is %d\n",
1977                        OVL_MAX_STACK);
1978                 goto out_err;
1979         }
1980
1981         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1982         if (!layers)
1983                 goto out_err;
1984
1985         ofs->layers = layers;
1986         /* Layer 0 is reserved for upper even if there's no upper */
1987         ofs->numlayer = 1;
1988
1989         sb->s_stack_depth = 0;
1990         sb->s_maxbytes = MAX_LFS_FILESIZE;
1991         atomic_long_set(&ofs->last_ino, 1);
1992         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1993         if (ofs->config.xino != OVL_XINO_OFF) {
1994                 ofs->xino_mode = BITS_PER_LONG - 32;
1995                 if (!ofs->xino_mode) {
1996                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1997                         ofs->config.xino = OVL_XINO_OFF;
1998                 }
1999         }
2000
2001         /* alloc/destroy_inode needed for setting up traps in inode cache */
2002         sb->s_op = &ovl_super_operations;
2003
2004         if (ofs->config.upperdir) {
2005                 struct super_block *upper_sb;
2006
2007                 if (!ofs->config.workdir) {
2008                         pr_err("missing 'workdir'\n");
2009                         goto out_err;
2010                 }
2011
2012                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
2013                 if (err)
2014                         goto out_err;
2015
2016                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
2017                 if (!ovl_should_sync(ofs)) {
2018                         ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
2019                         if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
2020                                 err = -EIO;
2021                                 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
2022                                 goto out_err;
2023                         }
2024                 }
2025
2026                 err = ovl_get_workdir(sb, ofs, &upperpath);
2027                 if (err)
2028                         goto out_err;
2029
2030                 if (!ofs->workdir)
2031                         sb->s_flags |= SB_RDONLY;
2032
2033                 sb->s_stack_depth = upper_sb->s_stack_depth;
2034                 sb->s_time_gran = upper_sb->s_time_gran;
2035         }
2036         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2037         err = PTR_ERR(oe);
2038         if (IS_ERR(oe))
2039                 goto out_err;
2040
2041         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2042         if (!ovl_upper_mnt(ofs))
2043                 sb->s_flags |= SB_RDONLY;
2044
2045         if (!ofs->config.uuid && ofs->numfs > 1) {
2046                 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2047                 ofs->config.uuid = true;
2048         }
2049
2050         if (!ovl_force_readonly(ofs) && ofs->config.index) {
2051                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2052                 if (err)
2053                         goto out_free_oe;
2054
2055                 /* Force r/o mount with no index dir */
2056                 if (!ofs->indexdir)
2057                         sb->s_flags |= SB_RDONLY;
2058         }
2059
2060         err = ovl_check_overlapping_layers(sb, ofs);
2061         if (err)
2062                 goto out_free_oe;
2063
2064         /* Show index=off in /proc/mounts for forced r/o mount */
2065         if (!ofs->indexdir) {
2066                 ofs->config.index = false;
2067                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2068                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2069                         ofs->config.nfs_export = false;
2070                 }
2071         }
2072
2073         if (ofs->config.metacopy && ofs->config.nfs_export) {
2074                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2075                 ofs->config.nfs_export = false;
2076         }
2077
2078         if (ofs->config.nfs_export)
2079                 sb->s_export_op = &ovl_export_operations;
2080
2081         /* Never override disk quota limits or use reserved space */
2082         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2083
2084         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2085         sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2086                 ovl_trusted_xattr_handlers;
2087         sb->s_fs_info = ofs;
2088         sb->s_flags |= SB_POSIXACL;
2089         sb->s_iflags |= SB_I_SKIP_SYNC;
2090
2091         err = -ENOMEM;
2092         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2093         if (!root_dentry)
2094                 goto out_free_oe;
2095
2096         mntput(upperpath.mnt);
2097         kfree(splitlower);
2098
2099         sb->s_root = root_dentry;
2100
2101         return 0;
2102
2103 out_free_oe:
2104         ovl_entry_stack_free(oe);
2105         kfree(oe);
2106 out_err:
2107         kfree(splitlower);
2108         path_put(&upperpath);
2109         ovl_free_fs(ofs);
2110 out:
2111         return err;
2112 }
2113
2114 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2115                                 const char *dev_name, void *raw_data)
2116 {
2117         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2118 }
2119
2120 static struct file_system_type ovl_fs_type = {
2121         .owner          = THIS_MODULE,
2122         .name           = "overlay",
2123         .fs_flags       = FS_USERNS_MOUNT,
2124         .mount          = ovl_mount,
2125         .kill_sb        = kill_anon_super,
2126 };
2127 MODULE_ALIAS_FS("overlay");
2128
2129 static void ovl_inode_init_once(void *foo)
2130 {
2131         struct ovl_inode *oi = foo;
2132
2133         inode_init_once(&oi->vfs_inode);
2134 }
2135
2136 static int __init ovl_init(void)
2137 {
2138         int err;
2139
2140         ovl_inode_cachep = kmem_cache_create("ovl_inode",
2141                                              sizeof(struct ovl_inode), 0,
2142                                              (SLAB_RECLAIM_ACCOUNT|
2143                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2144                                              ovl_inode_init_once);
2145         if (ovl_inode_cachep == NULL)
2146                 return -ENOMEM;
2147
2148         err = ovl_aio_request_cache_init();
2149         if (!err) {
2150                 err = register_filesystem(&ovl_fs_type);
2151                 if (!err)
2152                         return 0;
2153
2154                 ovl_aio_request_cache_destroy();
2155         }
2156         kmem_cache_destroy(ovl_inode_cachep);
2157
2158         return err;
2159 }
2160
2161 static void __exit ovl_exit(void)
2162 {
2163         unregister_filesystem(&ovl_fs_type);
2164
2165         /*
2166          * Make sure all delayed rcu free inodes are flushed before we
2167          * destroy cache.
2168          */
2169         rcu_barrier();
2170         kmem_cache_destroy(ovl_inode_cachep);
2171         ovl_aio_request_cache_destroy();
2172 }
2173
2174 module_init(ovl_init);
2175 module_exit(ovl_exit);