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 <[email protected]>");
  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         if (ofs->config.userxattr)
 384                 seq_puts(m, ",userxattr");
 385         return 0;
 386 }
 387
 388 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 389 {
 390         struct ovl_fs *ofs = sb->s_fs_info;
 391         struct super_block *upper_sb;
 392         int ret = 0;
 393
 394         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 395                 return -EROFS;
 396
 397         if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
 398                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 399                 if (ovl_should_sync(ofs)) {
 400                         down_read(&upper_sb->s_umount);
 401                         ret = sync_filesystem(upper_sb);
 402                         up_read(&upper_sb->s_umount);
 403                 }
 404         }
 405
 406         return ret;
 407 }
 408
 409 static const struct super_operations ovl_super_operations = {
 410         .alloc_inode    = ovl_alloc_inode,
 411         .free_inode     = ovl_free_inode,
 412         .destroy_inode  = ovl_destroy_inode,
 413         .drop_inode     = generic_delete_inode,
 414         .put_super      = ovl_put_super,
 415         .sync_fs        = ovl_sync_fs,
 416         .statfs         = ovl_statfs,
 417         .show_options   = ovl_show_options,
 418         .remount_fs     = ovl_remount,
 419 };
 420
 421 enum {
 422         OPT_LOWERDIR,
 423         OPT_UPPERDIR,
 424         OPT_WORKDIR,
 425         OPT_DEFAULT_PERMISSIONS,
 426         OPT_REDIRECT_DIR,
 427         OPT_INDEX_ON,
 428         OPT_INDEX_OFF,
 429         OPT_UUID_ON,
 430         OPT_UUID_OFF,
 431         OPT_NFS_EXPORT_ON,
 432         OPT_USERXATTR,
 433         OPT_NFS_EXPORT_OFF,
 434         OPT_XINO_ON,
 435         OPT_XINO_OFF,
 436         OPT_XINO_AUTO,
 437         OPT_METACOPY_ON,
 438         OPT_METACOPY_OFF,
 439         OPT_VOLATILE,
 440         OPT_ERR,
 441 };
 442
 443 static const match_table_t ovl_tokens = {
 444         {OPT_LOWERDIR,                  "lowerdir=%s"},
 445         {OPT_UPPERDIR,                  "upperdir=%s"},
 446         {OPT_WORKDIR,                   "workdir=%s"},
 447         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 448         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 449         {OPT_INDEX_ON,                  "index=on"},
 450         {OPT_INDEX_OFF,                 "index=off"},
 451         {OPT_USERXATTR,                 "userxattr"},
 452         {OPT_UUID_ON,                   "uuid=on"},
 453         {OPT_UUID_OFF,                  "uuid=off"},
 454         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 455         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 456         {OPT_XINO_ON,                   "xino=on"},
 457         {OPT_XINO_OFF,                  "xino=off"},
 458         {OPT_XINO_AUTO,                 "xino=auto"},
 459         {OPT_METACOPY_ON,               "metacopy=on"},
 460         {OPT_METACOPY_OFF,              "metacopy=off"},
 461         {OPT_VOLATILE,                  "volatile"},
 462         {OPT_ERR,                       NULL}
 463 };
 464
 465 static char *ovl_next_opt(char **s)
 466 {
 467         char *sbegin = *s;
 468         char *p;
 469
 470         if (sbegin == NULL)
 471                 return NULL;
 472
 473         for (p = sbegin; *p; p++) {
 474                 if (*p == '\\') {
 475                         p++;
 476                         if (!*p)
 477                                 break;
 478                 } else if (*p == ',') {
 479                         *p = '\0';
 480                         *s = p + 1;
 481                         return sbegin;
 482                 }
 483         }
 484         *s = NULL;
 485         return sbegin;
 486 }
 487
 488 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 489 {
 490         if (strcmp(mode, "on") == 0) {
 491                 config->redirect_dir = true;
 492                 /*
 493                  * Does not make sense to have redirect creation without
 494                  * redirect following.
 495                  */
 496                 config->redirect_follow = true;
 497         } else if (strcmp(mode, "follow") == 0) {
 498                 config->redirect_follow = true;
 499         } else if (strcmp(mode, "off") == 0) {
 500                 if (ovl_redirect_always_follow)
 501                         config->redirect_follow = true;
 502         } else if (strcmp(mode, "nofollow") != 0) {
 503                 pr_err("bad mount option \"redirect_dir=%s\"\n",
 504                        mode);
 505                 return -EINVAL;
 506         }
 507
 508         return 0;
 509 }
 510
 511 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 512 {
 513         char *p;
 514         int err;
 515         bool metacopy_opt = false, redirect_opt = false;
 516         bool nfs_export_opt = false, index_opt = false;
 517
 518         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 519         if (!config->redirect_mode)
 520                 return -ENOMEM;
 521
 522         while ((p = ovl_next_opt(&opt)) != NULL) {
 523                 int token;
 524                 substring_t args[MAX_OPT_ARGS];
 525
 526                 if (!*p)
 527                         continue;
 528
 529                 token = match_token(p, ovl_tokens, args);
 530                 switch (token) {
 531                 case OPT_UPPERDIR:
 532                         kfree(config->upperdir);
 533                         config->upperdir = match_strdup(&args[0]);
 534                         if (!config->upperdir)
 535                                 return -ENOMEM;
 536                         break;
 537
 538                 case OPT_LOWERDIR:
 539                         kfree(config->lowerdir);
 540                         config->lowerdir = match_strdup(&args[0]);
 541                         if (!config->lowerdir)
 542                                 return -ENOMEM;
 543                         break;
 544
 545                 case OPT_WORKDIR:
 546                         kfree(config->workdir);
 547                         config->workdir = match_strdup(&args[0]);
 548                         if (!config->workdir)
 549                                 return -ENOMEM;
 550                         break;
 551
 552                 case OPT_DEFAULT_PERMISSIONS:
 553                         config->default_permissions = true;
 554                         break;
 555
 556                 case OPT_REDIRECT_DIR:
 557                         kfree(config->redirect_mode);
 558                         config->redirect_mode = match_strdup(&args[0]);
 559                         if (!config->redirect_mode)
 560                                 return -ENOMEM;
 561                         redirect_opt = true;
 562                         break;
 563
 564                 case OPT_INDEX_ON:
 565                         config->index = true;
 566                         index_opt = true;
 567                         break;
 568
 569                 case OPT_INDEX_OFF:
 570                         config->index = false;
 571                         index_opt = true;
 572                         break;
 573
 574                 case OPT_UUID_ON:
 575                         config->uuid = true;
 576                         break;
 577
 578                 case OPT_UUID_OFF:
 579                         config->uuid = false;
 580                         break;
 581
 582                 case OPT_NFS_EXPORT_ON:
 583                         config->nfs_export = true;
 584                         nfs_export_opt = true;
 585                         break;
 586
 587                 case OPT_NFS_EXPORT_OFF:
 588                         config->nfs_export = false;
 589                         nfs_export_opt = true;
 590                         break;
 591
 592                 case OPT_XINO_ON:
 593                         config->xino = OVL_XINO_ON;
 594                         break;
 595
 596                 case OPT_XINO_OFF:
 597                         config->xino = OVL_XINO_OFF;
 598                         break;
 599
 600                 case OPT_XINO_AUTO:
 601                         config->xino = OVL_XINO_AUTO;
 602                         break;
 603
 604                 case OPT_METACOPY_ON:
 605                         config->metacopy = true;
 606                         metacopy_opt = true;
 607                         break;
 608
 609                 case OPT_METACOPY_OFF:
 610                         config->metacopy = false;
 611                         metacopy_opt = true;
 612                         break;
 613
 614                 case OPT_VOLATILE:
 615                         config->ovl_volatile = true;
 616                         break;
 617
 618                 case OPT_USERXATTR:
 619                         config->userxattr = true;
 620                         break;
 621
 622                 default:
 623                         pr_err("unrecognized mount option \"%s\" or missing value\n",
 624                                         p);
 625                         return -EINVAL;
 626                 }
 627         }
 628
 629         /* Workdir/index are useless in non-upper mount */
 630         if (!config->upperdir) {
 631                 if (config->workdir) {
 632                         pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 633                                 config->workdir);
 634                         kfree(config->workdir);
 635                         config->workdir = NULL;
 636                 }
 637                 if (config->index && index_opt) {
 638                         pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
 639                         index_opt = false;
 640                 }
 641                 config->index = false;
 642         }
 643
 644         if (!config->upperdir && config->ovl_volatile) {
 645                 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
 646                 config->ovl_volatile = false;
 647         }
 648
 649         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 650         if (err)
 651                 return err;
 652
 653         /*
 654          * This is to make the logic below simpler.  It doesn't make any other
 655          * difference, since config->redirect_dir is only used for upper.
 656          */
 657         if (!config->upperdir && config->redirect_follow)
 658                 config->redirect_dir = true;
 659
 660         /* Resolve metacopy -> redirect_dir dependency */
 661         if (config->metacopy && !config->redirect_dir) {
 662                 if (metacopy_opt && redirect_opt) {
 663                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
 664                                config->redirect_mode);
 665                         return -EINVAL;
 666                 }
 667                 if (redirect_opt) {
 668                         /*
 669                          * There was an explicit redirect_dir=... that resulted
 670                          * in this conflict.
 671                          */
 672                         pr_info("disabling metacopy due to redirect_dir=%s\n",
 673                                 config->redirect_mode);
 674                         config->metacopy = false;
 675                 } else {
 676                         /* Automatically enable redirect otherwise. */
 677                         config->redirect_follow = config->redirect_dir = true;
 678                 }
 679         }
 680
 681         /* Resolve nfs_export -> index dependency */
 682         if (config->nfs_export && !config->index) {
 683                 if (!config->upperdir && config->redirect_follow) {
 684                         pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
 685                         config->nfs_export = false;
 686                 } else if (nfs_export_opt && index_opt) {
 687                         pr_err("conflicting options: nfs_export=on,index=off\n");
 688                         return -EINVAL;
 689                 } else if (index_opt) {
 690                         /*
 691                          * There was an explicit index=off that resulted
 692                          * in this conflict.
 693                          */
 694                         pr_info("disabling nfs_export due to index=off\n");
 695                         config->nfs_export = false;
 696                 } else {
 697                         /* Automatically enable index otherwise. */
 698                         config->index = true;
 699                 }
 700         }
 701
 702         /* Resolve nfs_export -> !metacopy dependency */
 703         if (config->nfs_export && config->metacopy) {
 704                 if (nfs_export_opt && metacopy_opt) {
 705                         pr_err("conflicting options: nfs_export=on,metacopy=on\n");
 706                         return -EINVAL;
 707                 }
 708                 if (metacopy_opt) {
 709                         /*
 710                          * There was an explicit metacopy=on that resulted
 711                          * in this conflict.
 712                          */
 713                         pr_info("disabling nfs_export due to metacopy=on\n");
 714                         config->nfs_export = false;
 715                 } else {
 716                         /*
 717                          * There was an explicit nfs_export=on that resulted
 718                          * in this conflict.
 719                          */
 720                         pr_info("disabling metacopy due to nfs_export=on\n");
 721                         config->metacopy = false;
 722                 }
 723         }
 724
 725
 726         /* Resolve userxattr -> !redirect && !metacopy dependency */
 727         if (config->userxattr) {
 728                 if (config->redirect_follow && redirect_opt) {
 729                         pr_err("conflicting options: userxattr,redirect_dir=%s\n",
 730                                config->redirect_mode);
 731                         return -EINVAL;
 732                 }
 733                 if (config->metacopy && metacopy_opt) {
 734                         pr_err("conflicting options: userxattr,metacopy=on\n");
 735                         return -EINVAL;
 736                 }
 737                 /*
 738                  * Silently disable default setting of redirect and metacopy.
 739                  * This shall be the default in the future as well: these
 740                  * options must be explicitly enabled if used together with
 741                  * userxattr.
 742                  */
 743                 config->redirect_dir = config->redirect_follow = false;
 744                 config->metacopy = false;
 745         }
 746
 747         return 0;
 748 }
 749
 750 #define OVL_WORKDIR_NAME "work"
 751 #define OVL_INDEXDIR_NAME "index"
 752
 753 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 754                                          const char *name, bool persist)
 755 {
 756         struct inode *dir =  ofs->workbasedir->d_inode;
 757         struct vfsmount *mnt = ovl_upper_mnt(ofs);
 758         struct dentry *work;
 759         int err;
 760         bool retried = false;
 761
 762         inode_lock_nested(dir, I_MUTEX_PARENT);
 763 retry:
 764         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
 765
 766         if (!IS_ERR(work)) {
 767                 struct iattr attr = {
 768                         .ia_valid = ATTR_MODE,
 769                         .ia_mode = S_IFDIR | 0,
 770                 };
 771
 772                 if (work->d_inode) {
 773                         err = -EEXIST;
 774                         if (retried)
 775                                 goto out_dput;
 776
 777                         if (persist)
 778                                 goto out_unlock;
 779
 780                         retried = true;
 781                         err = ovl_workdir_cleanup(dir, mnt, work, 0);
 782                         dput(work);
 783                         if (err == -EINVAL) {
 784                                 work = ERR_PTR(err);
 785                                 goto out_unlock;
 786                         }
 787                         goto retry;
 788                 }
 789
 790                 err = ovl_mkdir_real(dir, &work, attr.ia_mode);
 791                 if (err)
 792                         goto out_dput;
 793
 794                 /* Weird filesystem returning with hashed negative (kernfs)? */
 795                 err = -EINVAL;
 796                 if (d_really_is_negative(work))
 797                         goto out_dput;
 798
 799                 /*
 800                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 801                  *
 802                  * a) success (there was a POSIX ACL xattr and was removed)
 803                  * b) -ENODATA (there was no POSIX ACL xattr)
 804                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 805                  *
 806                  * There are various other error values that could effectively
 807                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 808                  * if the xattr name is too long), but the set of filesystems
 809                  * allowed as upper are limited to "normal" ones, where checking
 810                  * for the above two errors is sufficient.
 811                  */
 812                 err = vfs_removexattr(&init_user_ns, work,
 813                                       XATTR_NAME_POSIX_ACL_DEFAULT);
 814                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 815                         goto out_dput;
 816
 817                 err = vfs_removexattr(&init_user_ns, work,
 818                                       XATTR_NAME_POSIX_ACL_ACCESS);
 819                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 820                         goto out_dput;
 821
 822                 /* Clear any inherited mode bits */
 823                 inode_lock(work->d_inode);
 824                 err = notify_change(&init_user_ns, work, &attr, NULL);
 825                 inode_unlock(work->d_inode);
 826                 if (err)
 827                         goto out_dput;
 828         } else {
 829                 err = PTR_ERR(work);
 830                 goto out_err;
 831         }
 832 out_unlock:
 833         inode_unlock(dir);
 834         return work;
 835
 836 out_dput:
 837         dput(work);
 838 out_err:
 839         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
 840                 ofs->config.workdir, name, -err);
 841         work = NULL;
 842         goto out_unlock;
 843 }
 844
 845 static void ovl_unescape(char *s)
 846 {
 847         char *d = s;
 848
 849         for (;; s++, d++) {
 850                 if (*s == '\\')
 851                         s++;
 852                 *d = *s;
 853                 if (!*s)
 854                         break;
 855         }
 856 }
 857
 858 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 859 {
 860         int err = -EINVAL;
 861
 862         if (!*name) {
 863                 pr_err("empty lowerdir\n");
 864                 goto out;
 865         }
 866         err = kern_path(name, LOOKUP_FOLLOW, path);
 867         if (err) {
 868                 pr_err("failed to resolve '%s': %i\n", name, err);
 869                 goto out;
 870         }
 871         err = -EINVAL;
 872         if (ovl_dentry_weird(path->dentry)) {
 873                 pr_err("filesystem on '%s' not supported\n", name);
 874                 goto out_put;
 875         }
 876         if (mnt_user_ns(path->mnt) != &init_user_ns) {
 877                 pr_err("idmapped layers are currently not supported\n");
 878                 goto out_put;
 879         }
 880         if (!d_is_dir(path->dentry)) {
 881                 pr_err("'%s' not a directory\n", name);
 882                 goto out_put;
 883         }
 884         return 0;
 885
 886 out_put:
 887         path_put_init(path);
 888 out:
 889         return err;
 890 }
 891
 892 static int ovl_mount_dir(const char *name, struct path *path)
 893 {
 894         int err = -ENOMEM;
 895         char *tmp = kstrdup(name, GFP_KERNEL);
 896
 897         if (tmp) {
 898                 ovl_unescape(tmp);
 899                 err = ovl_mount_dir_noesc(tmp, path);
 900
 901                 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
 902                         pr_err("filesystem on '%s' not supported as upperdir\n",
 903                                tmp);
 904                         path_put_init(path);
 905                         err = -EINVAL;
 906                 }
 907                 kfree(tmp);
 908         }
 909         return err;
 910 }
 911
 912 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 913                              const char *name)
 914 {
 915         struct kstatfs statfs;
 916         int err = vfs_statfs(path, &statfs);
 917
 918         if (err)
 919                 pr_err("statfs failed on '%s'\n", name);
 920         else
 921                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 922
 923         return err;
 924 }
 925
 926 static int ovl_lower_dir(const char *name, struct path *path,
 927                          struct ovl_fs *ofs, int *stack_depth)
 928 {
 929         int fh_type;
 930         int err;
 931
 932         err = ovl_mount_dir_noesc(name, path);
 933         if (err)
 934                 return err;
 935
 936         err = ovl_check_namelen(path, ofs, name);
 937         if (err)
 938                 return err;
 939
 940         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 941
 942         /*
 943          * The inodes index feature and NFS export need to encode and decode
 944          * file handles, so they require that all layers support them.
 945          */
 946         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 947         if ((ofs->config.nfs_export ||
 948              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 949                 ofs->config.index = false;
 950                 ofs->config.nfs_export = false;
 951                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 952                         name);
 953         }
 954         /*
 955          * Decoding origin file handle is required for persistent st_ino.
 956          * Without persistent st_ino, xino=auto falls back to xino=off.
 957          */
 958         if (ofs->config.xino == OVL_XINO_AUTO &&
 959             ofs->config.upperdir && !fh_type) {
 960                 ofs->config.xino = OVL_XINO_OFF;
 961                 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
 962                         name);
 963         }
 964
 965         /* Check if lower fs has 32bit inode numbers */
 966         if (fh_type != FILEID_INO32_GEN)
 967                 ofs->xino_mode = -1;
 968
 969         return 0;
 970 }
 971
 972 /* Workdir should not be subdir of upperdir and vice versa */
 973 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 974 {
 975         bool ok = false;
 976
 977         if (workdir != upperdir) {
 978                 ok = (lock_rename(workdir, upperdir) == NULL);
 979                 unlock_rename(workdir, upperdir);
 980         }
 981         return ok;
 982 }
 983
 984 static unsigned int ovl_split_lowerdirs(char *str)
 985 {
 986         unsigned int ctr = 1;
 987         char *s, *d;
 988
 989         for (s = d = str;; s++, d++) {
 990                 if (*s == '\\') {
 991                         s++;
 992                 } else if (*s == ':') {
 993                         *d = '\0';
 994                         ctr++;
 995                         continue;
 996                 }
 997                 *d = *s;
 998                 if (!*s)
 999                         break;
1000         }
1001         return ctr;
1002 }
1003
1004 static int __maybe_unused
1005 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
1006                         struct dentry *dentry, struct inode *inode,
1007                         const char *name, void *buffer, size_t size)
1008 {
1009         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
1010 }
1011
1012 static int __maybe_unused
1013 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
1014                         struct user_namespace *mnt_userns,
1015                         struct dentry *dentry, struct inode *inode,
1016                         const char *name, const void *value,
1017                         size_t size, int flags)
1018 {
1019         struct dentry *workdir = ovl_workdir(dentry);
1020         struct inode *realinode = ovl_inode_real(inode);
1021         struct posix_acl *acl = NULL;
1022         int err;
1023
1024         /* Check that everything is OK before copy-up */
1025         if (value) {
1026                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
1027                 if (IS_ERR(acl))
1028                         return PTR_ERR(acl);
1029         }
1030         err = -EOPNOTSUPP;
1031         if (!IS_POSIXACL(d_inode(workdir)))
1032                 goto out_acl_release;
1033         if (!realinode->i_op->set_acl)
1034                 goto out_acl_release;
1035         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1036                 err = acl ? -EACCES : 0;
1037                 goto out_acl_release;
1038         }
1039         err = -EPERM;
1040         if (!inode_owner_or_capable(&init_user_ns, inode))
1041                 goto out_acl_release;
1042
1043         posix_acl_release(acl);
1044
1045         /*
1046          * Check if sgid bit needs to be cleared (actual setacl operation will
1047          * be done with mounter's capabilities and so that won't do it for us).
1048          */
1049         if (unlikely(inode->i_mode & S_ISGID) &&
1050             handler->flags == ACL_TYPE_ACCESS &&
1051             !in_group_p(inode->i_gid) &&
1052             !capable_wrt_inode_uidgid(&init_user_ns, inode, CAP_FSETID)) {
1053                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
1054
1055                 err = ovl_setattr(&init_user_ns, dentry, &iattr);
1056                 if (err)
1057                         return err;
1058         }
1059
1060         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
1061         return err;
1062
1063 out_acl_release:
1064         posix_acl_release(acl);
1065         return err;
1066 }
1067
1068 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1069                              struct dentry *dentry, struct inode *inode,
1070                              const char *name, void *buffer, size_t size)
1071 {
1072         return -EOPNOTSUPP;
1073 }
1074
1075 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1076                              struct user_namespace *mnt_userns,
1077                              struct dentry *dentry, struct inode *inode,
1078                              const char *name, const void *value,
1079                              size_t size, int flags)
1080 {
1081         return -EOPNOTSUPP;
1082 }
1083
1084 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1085                                struct dentry *dentry, struct inode *inode,
1086                                const char *name, void *buffer, size_t size)
1087 {
1088         return ovl_xattr_get(dentry, inode, name, buffer, size);
1089 }
1090
1091 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1092                                struct user_namespace *mnt_userns,
1093                                struct dentry *dentry, struct inode *inode,
1094                                const char *name, const void *value,
1095                                size_t size, int flags)
1096 {
1097         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1098 }
1099
1100 static const struct xattr_handler __maybe_unused
1101 ovl_posix_acl_access_xattr_handler = {
1102         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1103         .flags = ACL_TYPE_ACCESS,
1104         .get = ovl_posix_acl_xattr_get,
1105         .set = ovl_posix_acl_xattr_set,
1106 };
1107
1108 static const struct xattr_handler __maybe_unused
1109 ovl_posix_acl_default_xattr_handler = {
1110         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1111         .flags = ACL_TYPE_DEFAULT,
1112         .get = ovl_posix_acl_xattr_get,
1113         .set = ovl_posix_acl_xattr_set,
1114 };
1115
1116 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1117         .prefix = OVL_XATTR_TRUSTED_PREFIX,
1118         .get = ovl_own_xattr_get,
1119         .set = ovl_own_xattr_set,
1120 };
1121
1122 static const struct xattr_handler ovl_own_user_xattr_handler = {
1123         .prefix = OVL_XATTR_USER_PREFIX,
1124         .get = ovl_own_xattr_get,
1125         .set = ovl_own_xattr_set,
1126 };
1127
1128 static const struct xattr_handler ovl_other_xattr_handler = {
1129         .prefix = "", /* catch all */
1130         .get = ovl_other_xattr_get,
1131         .set = ovl_other_xattr_set,
1132 };
1133
1134 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1135 #ifdef CONFIG_FS_POSIX_ACL
1136         &ovl_posix_acl_access_xattr_handler,
1137         &ovl_posix_acl_default_xattr_handler,
1138 #endif
1139         &ovl_own_trusted_xattr_handler,
1140         &ovl_other_xattr_handler,
1141         NULL
1142 };
1143
1144 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1145 #ifdef CONFIG_FS_POSIX_ACL
1146         &ovl_posix_acl_access_xattr_handler,
1147         &ovl_posix_acl_default_xattr_handler,
1148 #endif
1149         &ovl_own_user_xattr_handler,
1150         &ovl_other_xattr_handler,
1151         NULL
1152 };
1153
1154 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1155                           struct inode **ptrap, const char *name)
1156 {
1157         struct inode *trap;
1158         int err;
1159
1160         trap = ovl_get_trap_inode(sb, dir);
1161         err = PTR_ERR_OR_ZERO(trap);
1162         if (err) {
1163                 if (err == -ELOOP)
1164                         pr_err("conflicting %s path\n", name);
1165                 return err;
1166         }
1167
1168         *ptrap = trap;
1169         return 0;
1170 }
1171
1172 /*
1173  * Determine how we treat concurrent use of upperdir/workdir based on the
1174  * index feature. This is papering over mount leaks of container runtimes,
1175  * for example, an old overlay mount is leaked and now its upperdir is
1176  * attempted to be used as a lower layer in a new overlay mount.
1177  */
1178 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1179 {
1180         if (ofs->config.index) {
1181                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1182                        name);
1183                 return -EBUSY;
1184         } else {
1185                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1186                         name);
1187                 return 0;
1188         }
1189 }
1190
1191 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1192                          struct ovl_layer *upper_layer, struct path *upperpath)
1193 {
1194         struct vfsmount *upper_mnt;
1195         int err;
1196
1197         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1198         if (err)
1199                 goto out;
1200
1201         /* Upperdir path should not be r/o */
1202         if (__mnt_is_readonly(upperpath->mnt)) {
1203                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1204                 err = -EINVAL;
1205                 goto out;
1206         }
1207
1208         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1209         if (err)
1210                 goto out;
1211
1212         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1213                              "upperdir");
1214         if (err)
1215                 goto out;
1216
1217         upper_mnt = clone_private_mount(upperpath);
1218         err = PTR_ERR(upper_mnt);
1219         if (IS_ERR(upper_mnt)) {
1220                 pr_err("failed to clone upperpath\n");
1221                 goto out;
1222         }
1223
1224         /* Don't inherit atime flags */
1225         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1226         upper_layer->mnt = upper_mnt;
1227         upper_layer->idx = 0;
1228         upper_layer->fsid = 0;
1229
1230         /*
1231          * Inherit SB_NOSEC flag from upperdir.
1232          *
1233          * This optimization changes behavior when a security related attribute
1234          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1235          * okay because we don't yet have guarantees in that case, but it will
1236          * need careful treatment once we want to honour changes to underlying
1237          * filesystems.
1238          */
1239         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1240                 sb->s_flags |= SB_NOSEC;
1241
1242         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1243                 ofs->upperdir_locked = true;
1244         } else {
1245                 err = ovl_report_in_use(ofs, "upperdir");
1246                 if (err)
1247                         goto out;
1248         }
1249
1250         err = 0;
1251 out:
1252         return err;
1253 }
1254
1255 /*
1256  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1257  * negative values if error is encountered.
1258  */
1259 static int ovl_check_rename_whiteout(struct dentry *workdir)
1260 {
1261         struct inode *dir = d_inode(workdir);
1262         struct dentry *temp;
1263         struct dentry *dest;
1264         struct dentry *whiteout;
1265         struct name_snapshot name;
1266         int err;
1267
1268         inode_lock_nested(dir, I_MUTEX_PARENT);
1269
1270         temp = ovl_create_temp(workdir, OVL_CATTR(S_IFREG | 0));
1271         err = PTR_ERR(temp);
1272         if (IS_ERR(temp))
1273                 goto out_unlock;
1274
1275         dest = ovl_lookup_temp(workdir);
1276         err = PTR_ERR(dest);
1277         if (IS_ERR(dest)) {
1278                 dput(temp);
1279                 goto out_unlock;
1280         }
1281
1282         /* Name is inline and stable - using snapshot as a copy helper */
1283         take_dentry_name_snapshot(&name, temp);
1284         err = ovl_do_rename(dir, temp, dir, dest, RENAME_WHITEOUT);
1285         if (err) {
1286                 if (err == -EINVAL)
1287                         err = 0;
1288                 goto cleanup_temp;
1289         }
1290
1291         whiteout = lookup_one_len(name.name.name, workdir, name.name.len);
1292         err = PTR_ERR(whiteout);
1293         if (IS_ERR(whiteout))
1294                 goto cleanup_temp;
1295
1296         err = ovl_is_whiteout(whiteout);
1297
1298         /* Best effort cleanup of whiteout and temp file */
1299         if (err)
1300                 ovl_cleanup(dir, whiteout);
1301         dput(whiteout);
1302
1303 cleanup_temp:
1304         ovl_cleanup(dir, temp);
1305         release_dentry_name_snapshot(&name);
1306         dput(temp);
1307         dput(dest);
1308
1309 out_unlock:
1310         inode_unlock(dir);
1311
1312         return err;
1313 }
1314
1315 static struct dentry *ovl_lookup_or_create(struct dentry *parent,
1316                                            const char *name, umode_t mode)
1317 {
1318         size_t len = strlen(name);
1319         struct dentry *child;
1320
1321         inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1322         child = lookup_one_len(name, parent, len);
1323         if (!IS_ERR(child) && !child->d_inode)
1324                 child = ovl_create_real(parent->d_inode, child,
1325                                         OVL_CATTR(mode));
1326         inode_unlock(parent->d_inode);
1327         dput(parent);
1328
1329         return child;
1330 }
1331
1332 /*
1333  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1334  * present.
1335  */
1336 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1337 {
1338         unsigned int ctr;
1339         struct dentry *d = dget(ofs->workbasedir);
1340         static const char *const volatile_path[] = {
1341                 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1342         };
1343         const char *const *name = volatile_path;
1344
1345         for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1346                 d = ovl_lookup_or_create(d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1347                 if (IS_ERR(d))
1348                         return PTR_ERR(d);
1349         }
1350         dput(d);
1351         return 0;
1352 }
1353
1354 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1355                             struct path *workpath)
1356 {
1357         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1358         struct dentry *temp, *workdir;
1359         bool rename_whiteout;
1360         bool d_type;
1361         int fh_type;
1362         int err;
1363
1364         err = mnt_want_write(mnt);
1365         if (err)
1366                 return err;
1367
1368         workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1369         err = PTR_ERR(workdir);
1370         if (IS_ERR_OR_NULL(workdir))
1371                 goto out;
1372
1373         ofs->workdir = workdir;
1374
1375         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1376         if (err)
1377                 goto out;
1378
1379         /*
1380          * Upper should support d_type, else whiteouts are visible.  Given
1381          * workdir and upper are on same fs, we can do iterate_dir() on
1382          * workdir. This check requires successful creation of workdir in
1383          * previous step.
1384          */
1385         err = ovl_check_d_type_supported(workpath);
1386         if (err < 0)
1387                 goto out;
1388
1389         d_type = err;
1390         if (!d_type)
1391                 pr_warn("upper fs needs to support d_type.\n");
1392
1393         /* Check if upper/work fs supports O_TMPFILE */
1394         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1395         ofs->tmpfile = !IS_ERR(temp);
1396         if (ofs->tmpfile)
1397                 dput(temp);
1398         else
1399                 pr_warn("upper fs does not support tmpfile.\n");
1400
1401
1402         /* Check if upper/work fs supports RENAME_WHITEOUT */
1403         err = ovl_check_rename_whiteout(ofs->workdir);
1404         if (err < 0)
1405                 goto out;
1406
1407         rename_whiteout = err;
1408         if (!rename_whiteout)
1409                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1410
1411         /*
1412          * Check if upper/work fs supports (trusted|user).overlay.* xattr
1413          */
1414         err = ovl_do_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1415         if (err) {
1416                 ofs->noxattr = true;
1417                 if (ofs->config.index || ofs->config.metacopy) {
1418                         ofs->config.index = false;
1419                         ofs->config.metacopy = false;
1420                         pr_warn("upper fs does not support xattr, falling back to index=off,metacopy=off.\n");
1421                 }
1422                 /*
1423                  * xattr support is required for persistent st_ino.
1424                  * Without persistent st_ino, xino=auto falls back to xino=off.
1425                  */
1426                 if (ofs->config.xino == OVL_XINO_AUTO) {
1427                         ofs->config.xino = OVL_XINO_OFF;
1428                         pr_warn("upper fs does not support xattr, falling back to xino=off.\n");
1429                 }
1430                 err = 0;
1431         } else {
1432                 ovl_do_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1433         }
1434
1435         /*
1436          * We allowed sub-optimal upper fs configuration and don't want to break
1437          * users over kernel upgrade, but we never allowed remote upper fs, so
1438          * we can enforce strict requirements for remote upper fs.
1439          */
1440         if (ovl_dentry_remote(ofs->workdir) &&
1441             (!d_type || !rename_whiteout || ofs->noxattr)) {
1442                 pr_err("upper fs missing required features.\n");
1443                 err = -EINVAL;
1444                 goto out;
1445         }
1446
1447         /*
1448          * For volatile mount, create a incompat/volatile/dirty file to keep
1449          * track of it.
1450          */
1451         if (ofs->config.ovl_volatile) {
1452                 err = ovl_create_volatile_dirty(ofs);
1453                 if (err < 0) {
1454                         pr_err("Failed to create volatile/dirty file.\n");
1455                         goto out;
1456                 }
1457         }
1458
1459         /* Check if upper/work fs supports file handles */
1460         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1461         if (ofs->config.index && !fh_type) {
1462                 ofs->config.index = false;
1463                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1464         }
1465
1466         /* Check if upper fs has 32bit inode numbers */
1467         if (fh_type != FILEID_INO32_GEN)
1468                 ofs->xino_mode = -1;
1469
1470         /* NFS export of r/w mount depends on index */
1471         if (ofs->config.nfs_export && !ofs->config.index) {
1472                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1473                 ofs->config.nfs_export = false;
1474         }
1475 out:
1476         mnt_drop_write(mnt);
1477         return err;
1478 }
1479
1480 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1481                            struct path *upperpath)
1482 {
1483         int err;
1484         struct path workpath = { };
1485
1486         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1487         if (err)
1488                 goto out;
1489
1490         err = -EINVAL;
1491         if (upperpath->mnt != workpath.mnt) {
1492                 pr_err("workdir and upperdir must reside under the same mount\n");
1493                 goto out;
1494         }
1495         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1496                 pr_err("workdir and upperdir must be separate subtrees\n");
1497                 goto out;
1498         }
1499
1500         ofs->workbasedir = dget(workpath.dentry);
1501
1502         if (ovl_inuse_trylock(ofs->workbasedir)) {
1503                 ofs->workdir_locked = true;
1504         } else {
1505                 err = ovl_report_in_use(ofs, "workdir");
1506                 if (err)
1507                         goto out;
1508         }
1509
1510         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1511                              "workdir");
1512         if (err)
1513                 goto out;
1514
1515         err = ovl_make_workdir(sb, ofs, &workpath);
1516
1517 out:
1518         path_put(&workpath);
1519
1520         return err;
1521 }
1522
1523 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1524                             struct ovl_entry *oe, struct path *upperpath)
1525 {
1526         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1527         struct dentry *indexdir;
1528         int err;
1529
1530         err = mnt_want_write(mnt);
1531         if (err)
1532                 return err;
1533
1534         /* Verify lower root is upper root origin */
1535         err = ovl_verify_origin(ofs, upperpath->dentry,
1536                                 oe->lowerstack[0].dentry, true);
1537         if (err) {
1538                 pr_err("failed to verify upper root origin\n");
1539                 goto out;
1540         }
1541
1542         /* index dir will act also as workdir */
1543         iput(ofs->workdir_trap);
1544         ofs->workdir_trap = NULL;
1545         dput(ofs->workdir);
1546         ofs->workdir = NULL;
1547         indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1548         if (IS_ERR(indexdir)) {
1549                 err = PTR_ERR(indexdir);
1550         } else if (indexdir) {
1551                 ofs->indexdir = indexdir;
1552                 ofs->workdir = dget(indexdir);
1553
1554                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1555                                      "indexdir");
1556                 if (err)
1557                         goto out;
1558
1559                 /*
1560                  * Verify upper root is exclusively associated with index dir.
1561                  * Older kernels stored upper fh in ".overlay.origin"
1562                  * xattr. If that xattr exists, verify that it is a match to
1563                  * upper dir file handle. In any case, verify or set xattr
1564                  * ".overlay.upper" to indicate that index may have
1565                  * directory entries.
1566                  */
1567                 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1568                         err = ovl_verify_set_fh(ofs, ofs->indexdir,
1569                                                 OVL_XATTR_ORIGIN,
1570                                                 upperpath->dentry, true, false);
1571                         if (err)
1572                                 pr_err("failed to verify index dir 'origin' xattr\n");
1573                 }
1574                 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1575                                        true);
1576                 if (err)
1577                         pr_err("failed to verify index dir 'upper' xattr\n");
1578
1579                 /* Cleanup bad/stale/orphan index entries */
1580                 if (!err)
1581                         err = ovl_indexdir_cleanup(ofs);
1582         }
1583         if (err || !ofs->indexdir)
1584                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1585
1586 out:
1587         mnt_drop_write(mnt);
1588         return err;
1589 }
1590
1591 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1592 {
1593         unsigned int i;
1594
1595         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1596                 return true;
1597
1598         /*
1599          * We allow using single lower with null uuid for index and nfs_export
1600          * for example to support those features with single lower squashfs.
1601          * To avoid regressions in setups of overlay with re-formatted lower
1602          * squashfs, do not allow decoding origin with lower null uuid unless
1603          * user opted-in to one of the new features that require following the
1604          * lower inode of non-dir upper.
1605          */
1606         if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
1607                 return false;
1608
1609         for (i = 0; i < ofs->numfs; i++) {
1610                 /*
1611                  * We use uuid to associate an overlay lower file handle with a
1612                  * lower layer, so we can accept lower fs with null uuid as long
1613                  * as all lower layers with null uuid are on the same fs.
1614                  * if we detect multiple lower fs with the same uuid, we
1615                  * disable lower file handle decoding on all of them.
1616                  */
1617                 if (ofs->fs[i].is_lower &&
1618                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1619                         ofs->fs[i].bad_uuid = true;
1620                         return false;
1621                 }
1622         }
1623         return true;
1624 }
1625
1626 /* Get a unique fsid for the layer */
1627 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1628 {
1629         struct super_block *sb = path->mnt->mnt_sb;
1630         unsigned int i;
1631         dev_t dev;
1632         int err;
1633         bool bad_uuid = false;
1634         bool warn = false;
1635
1636         for (i = 0; i < ofs->numfs; i++) {
1637                 if (ofs->fs[i].sb == sb)
1638                         return i;
1639         }
1640
1641         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1642                 bad_uuid = true;
1643                 if (ofs->config.xino == OVL_XINO_AUTO) {
1644                         ofs->config.xino = OVL_XINO_OFF;
1645                         warn = true;
1646                 }
1647                 if (ofs->config.index || ofs->config.nfs_export) {
1648                         ofs->config.index = false;
1649                         ofs->config.nfs_export = false;
1650                         warn = true;
1651                 }
1652                 if (warn) {
1653                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1654                                 uuid_is_null(&sb->s_uuid) ? "null" :
1655                                                             "conflicting",
1656                                 path->dentry, ovl_xino_str[ofs->config.xino]);
1657                 }
1658         }
1659
1660         err = get_anon_bdev(&dev);
1661         if (err) {
1662                 pr_err("failed to get anonymous bdev for lowerpath\n");
1663                 return err;
1664         }
1665
1666         ofs->fs[ofs->numfs].sb = sb;
1667         ofs->fs[ofs->numfs].pseudo_dev = dev;
1668         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1669
1670         return ofs->numfs++;
1671 }
1672
1673 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1674                           struct path *stack, unsigned int numlower,
1675                           struct ovl_layer *layers)
1676 {
1677         int err;
1678         unsigned int i;
1679
1680         err = -ENOMEM;
1681         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1682         if (ofs->fs == NULL)
1683                 goto out;
1684
1685         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1686         ofs->numfs++;
1687
1688         /*
1689          * All lower layers that share the same fs as upper layer, use the same
1690          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1691          * only overlay to simplify ovl_fs_free().
1692          * is_lower will be set if upper fs is shared with a lower layer.
1693          */
1694         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1695         if (err) {
1696                 pr_err("failed to get anonymous bdev for upper fs\n");
1697                 goto out;
1698         }
1699
1700         if (ovl_upper_mnt(ofs)) {
1701                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1702                 ofs->fs[0].is_lower = false;
1703         }
1704
1705         for (i = 0; i < numlower; i++) {
1706                 struct vfsmount *mnt;
1707                 struct inode *trap;
1708                 int fsid;
1709
1710                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1711                 if (err < 0)
1712                         goto out;
1713
1714                 /*
1715                  * Check if lower root conflicts with this overlay layers before
1716                  * checking if it is in-use as upperdir/workdir of "another"
1717                  * mount, because we do not bother to check in ovl_is_inuse() if
1718                  * the upperdir/workdir is in fact in-use by our
1719                  * upperdir/workdir.
1720                  */
1721                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1722                 if (err)
1723                         goto out;
1724
1725                 if (ovl_is_inuse(stack[i].dentry)) {
1726                         err = ovl_report_in_use(ofs, "lowerdir");
1727                         if (err) {
1728                                 iput(trap);
1729                                 goto out;
1730                         }
1731                 }
1732
1733                 mnt = clone_private_mount(&stack[i]);
1734                 err = PTR_ERR(mnt);
1735                 if (IS_ERR(mnt)) {
1736                         pr_err("failed to clone lowerpath\n");
1737                         iput(trap);
1738                         goto out;
1739                 }
1740
1741                 /*
1742                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1743                  * will fail instead of modifying lower fs.
1744                  */
1745                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1746
1747                 layers[ofs->numlayer].trap = trap;
1748                 layers[ofs->numlayer].mnt = mnt;
1749                 layers[ofs->numlayer].idx = ofs->numlayer;
1750                 layers[ofs->numlayer].fsid = fsid;
1751                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1752                 ofs->numlayer++;
1753                 ofs->fs[fsid].is_lower = true;
1754         }
1755
1756         /*
1757          * When all layers on same fs, overlay can use real inode numbers.
1758          * With mount option "xino=<on|auto>", mounter declares that there are
1759          * enough free high bits in underlying fs to hold the unique fsid.
1760          * If overlayfs does encounter underlying inodes using the high xino
1761          * bits reserved for fsid, it emits a warning and uses the original
1762          * inode number or a non persistent inode number allocated from a
1763          * dedicated range.
1764          */
1765         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1766                 if (ofs->config.xino == OVL_XINO_ON)
1767                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1768                 ofs->xino_mode = 0;
1769         } else if (ofs->config.xino == OVL_XINO_OFF) {
1770                 ofs->xino_mode = -1;
1771         } else if (ofs->xino_mode < 0) {
1772                 /*
1773                  * This is a roundup of number of bits needed for encoding
1774                  * fsid, where fsid 0 is reserved for upper fs (even with
1775                  * lower only overlay) +1 extra bit is reserved for the non
1776                  * persistent inode number range that is used for resolving
1777                  * xino lower bits overflow.
1778                  */
1779                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1780                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1781         }
1782
1783         if (ofs->xino_mode > 0) {
1784                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1785                         ofs->xino_mode);
1786         }
1787
1788         err = 0;
1789 out:
1790         return err;
1791 }
1792
1793 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1794                                 const char *lower, unsigned int numlower,
1795                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1796 {
1797         int err;
1798         struct path *stack = NULL;
1799         unsigned int i;
1800         struct ovl_entry *oe;
1801
1802         if (!ofs->config.upperdir && numlower == 1) {
1803                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1804                 return ERR_PTR(-EINVAL);
1805         }
1806
1807         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1808         if (!stack)
1809                 return ERR_PTR(-ENOMEM);
1810
1811         err = -EINVAL;
1812         for (i = 0; i < numlower; i++) {
1813                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1814                 if (err)
1815                         goto out_err;
1816
1817                 lower = strchr(lower, '\0') + 1;
1818         }
1819
1820         err = -EINVAL;
1821         sb->s_stack_depth++;
1822         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1823                 pr_err("maximum fs stacking depth exceeded\n");
1824                 goto out_err;
1825         }
1826
1827         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1828         if (err)
1829                 goto out_err;
1830
1831         err = -ENOMEM;
1832         oe = ovl_alloc_entry(numlower);
1833         if (!oe)
1834                 goto out_err;
1835
1836         for (i = 0; i < numlower; i++) {
1837                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1838                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1839         }
1840
1841 out:
1842         for (i = 0; i < numlower; i++)
1843                 path_put(&stack[i]);
1844         kfree(stack);
1845
1846         return oe;
1847
1848 out_err:
1849         oe = ERR_PTR(err);
1850         goto out;
1851 }
1852
1853 /*
1854  * Check if this layer root is a descendant of:
1855  * - another layer of this overlayfs instance
1856  * - upper/work dir of any overlayfs instance
1857  */
1858 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1859                            struct dentry *dentry, const char *name,
1860                            bool is_lower)
1861 {
1862         struct dentry *next = dentry, *parent;
1863         int err = 0;
1864
1865         if (!dentry)
1866                 return 0;
1867
1868         parent = dget_parent(next);
1869
1870         /* Walk back ancestors to root (inclusive) looking for traps */
1871         while (!err && parent != next) {
1872                 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1873                         err = -ELOOP;
1874                         pr_err("overlapping %s path\n", name);
1875                 } else if (ovl_is_inuse(parent)) {
1876                         err = ovl_report_in_use(ofs, name);
1877                 }
1878                 next = parent;
1879                 parent = dget_parent(next);
1880                 dput(next);
1881         }
1882
1883         dput(parent);
1884
1885         return err;
1886 }
1887
1888 /*
1889  * Check if any of the layers or work dirs overlap.
1890  */
1891 static int ovl_check_overlapping_layers(struct super_block *sb,
1892                                         struct ovl_fs *ofs)
1893 {
1894         int i, err;
1895
1896         if (ovl_upper_mnt(ofs)) {
1897                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1898                                       "upperdir", false);
1899                 if (err)
1900                         return err;
1901
1902                 /*
1903                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1904                  * this instance and covers overlapping work and index dirs,
1905                  * unless work or index dir have been moved since created inside
1906                  * workbasedir.  In that case, we already have their traps in
1907                  * inode cache and we will catch that case on lookup.
1908                  */
1909                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1910                                       false);
1911                 if (err)
1912                         return err;
1913         }
1914
1915         for (i = 1; i < ofs->numlayer; i++) {
1916                 err = ovl_check_layer(sb, ofs,
1917                                       ofs->layers[i].mnt->mnt_root,
1918                                       "lowerdir", true);
1919                 if (err)
1920                         return err;
1921         }
1922
1923         return 0;
1924 }
1925
1926 static struct dentry *ovl_get_root(struct super_block *sb,
1927                                    struct dentry *upperdentry,
1928                                    struct ovl_entry *oe)
1929 {
1930         struct dentry *root;
1931         struct ovl_path *lowerpath = &oe->lowerstack[0];
1932         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1933         int fsid = lowerpath->layer->fsid;
1934         struct ovl_inode_params oip = {
1935                 .upperdentry = upperdentry,
1936                 .lowerpath = lowerpath,
1937         };
1938
1939         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1940         if (!root)
1941                 return NULL;
1942
1943         root->d_fsdata = oe;
1944
1945         if (upperdentry) {
1946                 /* Root inode uses upper st_ino/i_ino */
1947                 ino = d_inode(upperdentry)->i_ino;
1948                 fsid = 0;
1949                 ovl_dentry_set_upper_alias(root);
1950                 if (ovl_is_impuredir(sb, upperdentry))
1951                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1952         }
1953
1954         /* Root is always merge -> can have whiteouts */
1955         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1956         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1957         ovl_set_upperdata(d_inode(root));
1958         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1959         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1960
1961         return root;
1962 }
1963
1964 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1965 {
1966         struct path upperpath = { };
1967         struct dentry *root_dentry;
1968         struct ovl_entry *oe;
1969         struct ovl_fs *ofs;
1970         struct ovl_layer *layers;
1971         struct cred *cred;
1972         char *splitlower = NULL;
1973         unsigned int numlower;
1974         int err;
1975
1976         err = -EIO;
1977         if (WARN_ON(sb->s_user_ns != current_user_ns()))
1978                 goto out;
1979
1980         sb->s_d_op = &ovl_dentry_operations;
1981
1982         err = -ENOMEM;
1983         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1984         if (!ofs)
1985                 goto out;
1986
1987         err = -ENOMEM;
1988         ofs->creator_cred = cred = prepare_creds();
1989         if (!cred)
1990                 goto out_err;
1991
1992         /* Is there a reason anyone would want not to share whiteouts? */
1993         ofs->share_whiteout = true;
1994
1995         ofs->config.index = ovl_index_def;
1996         ofs->config.uuid = true;
1997         ofs->config.nfs_export = ovl_nfs_export_def;
1998         ofs->config.xino = ovl_xino_def();
1999         ofs->config.metacopy = ovl_metacopy_def;
2000         err = ovl_parse_opt((char *) data, &ofs->config);
2001         if (err)
2002                 goto out_err;
2003
2004         err = -EINVAL;
2005         if (!ofs->config.lowerdir) {
2006                 if (!silent)
2007                         pr_err("missing 'lowerdir'\n");
2008                 goto out_err;
2009         }
2010
2011         err = -ENOMEM;
2012         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
2013         if (!splitlower)
2014                 goto out_err;
2015
2016         err = -EINVAL;
2017         numlower = ovl_split_lowerdirs(splitlower);
2018         if (numlower > OVL_MAX_STACK) {
2019                 pr_err("too many lower directories, limit is %d\n",
2020                        OVL_MAX_STACK);
2021                 goto out_err;
2022         }
2023
2024         err = -ENOMEM;
2025         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
2026         if (!layers)
2027                 goto out_err;
2028
2029         ofs->layers = layers;
2030         /* Layer 0 is reserved for upper even if there's no upper */
2031         ofs->numlayer = 1;
2032
2033         sb->s_stack_depth = 0;
2034         sb->s_maxbytes = MAX_LFS_FILESIZE;
2035         atomic_long_set(&ofs->last_ino, 1);
2036         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
2037         if (ofs->config.xino != OVL_XINO_OFF) {
2038                 ofs->xino_mode = BITS_PER_LONG - 32;
2039                 if (!ofs->xino_mode) {
2040                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
2041                         ofs->config.xino = OVL_XINO_OFF;
2042                 }
2043         }
2044
2045         /* alloc/destroy_inode needed for setting up traps in inode cache */
2046         sb->s_op = &ovl_super_operations;
2047
2048         if (ofs->config.upperdir) {
2049                 struct super_block *upper_sb;
2050
2051                 err = -EINVAL;
2052                 if (!ofs->config.workdir) {
2053                         pr_err("missing 'workdir'\n");
2054                         goto out_err;
2055                 }
2056
2057                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
2058                 if (err)
2059                         goto out_err;
2060
2061                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
2062                 if (!ovl_should_sync(ofs)) {
2063                         ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
2064                         if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
2065                                 err = -EIO;
2066                                 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
2067                                 goto out_err;
2068                         }
2069                 }
2070
2071                 err = ovl_get_workdir(sb, ofs, &upperpath);
2072                 if (err)
2073                         goto out_err;
2074
2075                 if (!ofs->workdir)
2076                         sb->s_flags |= SB_RDONLY;
2077
2078                 sb->s_stack_depth = upper_sb->s_stack_depth;
2079                 sb->s_time_gran = upper_sb->s_time_gran;
2080         }
2081         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2082         err = PTR_ERR(oe);
2083         if (IS_ERR(oe))
2084                 goto out_err;
2085
2086         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2087         if (!ovl_upper_mnt(ofs))
2088                 sb->s_flags |= SB_RDONLY;
2089
2090         if (!ofs->config.uuid && ofs->numfs > 1) {
2091                 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2092                 ofs->config.uuid = true;
2093         }
2094
2095         if (!ovl_force_readonly(ofs) && ofs->config.index) {
2096                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2097                 if (err)
2098                         goto out_free_oe;
2099
2100                 /* Force r/o mount with no index dir */
2101                 if (!ofs->indexdir)
2102                         sb->s_flags |= SB_RDONLY;
2103         }
2104
2105         err = ovl_check_overlapping_layers(sb, ofs);
2106         if (err)
2107                 goto out_free_oe;
2108
2109         /* Show index=off in /proc/mounts for forced r/o mount */
2110         if (!ofs->indexdir) {
2111                 ofs->config.index = false;
2112                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2113                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2114                         ofs->config.nfs_export = false;
2115                 }
2116         }
2117
2118         if (ofs->config.metacopy && ofs->config.nfs_export) {
2119                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2120                 ofs->config.nfs_export = false;
2121         }
2122
2123         if (ofs->config.nfs_export)
2124                 sb->s_export_op = &ovl_export_operations;
2125
2126         /* Never override disk quota limits or use reserved space */
2127         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2128
2129         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2130         sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2131                 ovl_trusted_xattr_handlers;
2132         sb->s_fs_info = ofs;
2133         sb->s_flags |= SB_POSIXACL;
2134         sb->s_iflags |= SB_I_SKIP_SYNC;
2135
2136         err = -ENOMEM;
2137         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2138         if (!root_dentry)
2139                 goto out_free_oe;
2140
2141         mntput(upperpath.mnt);
2142         kfree(splitlower);
2143
2144         sb->s_root = root_dentry;
2145
2146         return 0;
2147
2148 out_free_oe:
2149         ovl_entry_stack_free(oe);
2150         kfree(oe);
2151 out_err:
2152         kfree(splitlower);
2153         path_put(&upperpath);
2154         ovl_free_fs(ofs);
2155 out:
2156         return err;
2157 }
2158
2159 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2160                                 const char *dev_name, void *raw_data)
2161 {
2162         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2163 }
2164
2165 static struct file_system_type ovl_fs_type = {
2166         .owner          = THIS_MODULE,
2167         .name           = "overlay",
2168         .fs_flags       = FS_USERNS_MOUNT,
2169         .mount          = ovl_mount,
2170         .kill_sb        = kill_anon_super,
2171 };
2172 MODULE_ALIAS_FS("overlay");
2173
2174 static void ovl_inode_init_once(void *foo)
2175 {
2176         struct ovl_inode *oi = foo;
2177
2178         inode_init_once(&oi->vfs_inode);
2179 }
2180
2181 static int __init ovl_init(void)
2182 {
2183         int err;
2184
2185         ovl_inode_cachep = kmem_cache_create("ovl_inode",
2186                                              sizeof(struct ovl_inode), 0,
2187                                              (SLAB_RECLAIM_ACCOUNT|
2188                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2189                                              ovl_inode_init_once);
2190         if (ovl_inode_cachep == NULL)
2191                 return -ENOMEM;
2192
2193         err = ovl_aio_request_cache_init();
2194         if (!err) {
2195                 err = register_filesystem(&ovl_fs_type);
2196                 if (!err)
2197                         return 0;
2198
2199                 ovl_aio_request_cache_destroy();
2200         }
2201         kmem_cache_destroy(ovl_inode_cachep);
2202
2203         return err;
2204 }
2205
2206 static void __exit ovl_exit(void)
2207 {
2208         unregister_filesystem(&ovl_fs_type);
2209
2210         /*
2211          * Make sure all delayed rcu free inodes are flushed before we
2212          * destroy cache.
2213          */
2214         rcu_barrier();
2215         kmem_cache_destroy(ovl_inode_cachep);
2216         ovl_aio_request_cache_destroy();
2217 }
2218
2219 module_init(ovl_init);
2220 module_exit(ovl_exit);