fs/overlayfs/super.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *
   4  * Copyright (C) 2011 Novell Inc.
   5  */
   6
   7 #include <uapi/linux/magic.h>
   8 #include <linux/fs.h>
   9 #include <linux/namei.h>
  10 #include <linux/xattr.h>
  11 #include <linux/mount.h>
  12 #include <linux/parser.h>
  13 #include <linux/module.h>
  14 #include <linux/statfs.h>
  15 #include <linux/seq_file.h>
  16 #include <linux/posix_acl_xattr.h>
  17 #include <linux/exportfs.h>
  18 #include "overlayfs.h"
  19
  20 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  21 MODULE_DESCRIPTION("Overlay filesystem");
  22 MODULE_LICENSE("GPL");
  23
  24
  25 struct ovl_dir_cache;
  26
  27 #define OVL_MAX_STACK 500
  28
  29 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  30 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  31 MODULE_PARM_DESC(redirect_dir,
  32                  "Default to on or off for the redirect_dir feature");
  33
  34 static bool ovl_redirect_always_follow =
  35         IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
  36 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
  37                    bool, 0644);
  38 MODULE_PARM_DESC(redirect_always_follow,
  39                  "Follow redirects even if redirect_dir feature is turned off");
  40
  41 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
  42 module_param_named(index, ovl_index_def, bool, 0644);
  43 MODULE_PARM_DESC(index,
  44                  "Default to on or off for the inodes index feature");
  45
  46 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
  47 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
  48 MODULE_PARM_DESC(nfs_export,
  49                  "Default to on or off for the NFS export feature");
  50
  51 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
  52 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
  53 MODULE_PARM_DESC(xino_auto,
  54                  "Auto enable xino feature");
  55
  56 static void ovl_entry_stack_free(struct ovl_entry *oe)
  57 {
  58         unsigned int i;
  59
  60         for (i = 0; i < oe->numlower; i++)
  61                 dput(oe->lowerstack[i].dentry);
  62 }
  63
  64 static bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
  65 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
  66 MODULE_PARM_DESC(metacopy,
  67                  "Default to on or off for the metadata only copy up feature");
  68
  69 static void ovl_dentry_release(struct dentry *dentry)
  70 {
  71         struct ovl_entry *oe = dentry->d_fsdata;
  72
  73         if (oe) {
  74                 ovl_entry_stack_free(oe);
  75                 kfree_rcu(oe, rcu);
  76         }
  77 }
  78
  79 static struct dentry *ovl_d_real(struct dentry *dentry,
  80                                  const struct inode *inode)
  81 {
  82         struct dentry *real = NULL, *lower;
  83
  84         /* It's an overlay file */
  85         if (inode && d_inode(dentry) == inode)
  86                 return dentry;
  87
  88         if (!d_is_reg(dentry)) {
  89                 if (!inode || inode == d_inode(dentry))
  90                         return dentry;
  91                 goto bug;
  92         }
  93
  94         real = ovl_dentry_upper(dentry);
  95         if (real && (inode == d_inode(real)))
  96                 return real;
  97
  98         if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
  99                 return real;
 100
 101         lower = ovl_dentry_lowerdata(dentry);
 102         if (!lower)
 103                 goto bug;
 104         real = lower;
 105
 106         /* Handle recursion */
 107         real = d_real(real, inode);
 108
 109         if (!inode || inode == d_inode(real))
 110                 return real;
 111 bug:
 112         WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
 113              __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
 114              inode ? inode->i_ino : 0, real,
 115              real && d_inode(real) ? d_inode(real)->i_ino : 0);
 116         return dentry;
 117 }
 118
 119 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
 120 {
 121         int ret = 1;
 122
 123         if (weak) {
 124                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
 125                         ret =  d->d_op->d_weak_revalidate(d, flags);
 126         } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
 127                 ret = d->d_op->d_revalidate(d, flags);
 128                 if (!ret) {
 129                         if (!(flags & LOOKUP_RCU))
 130                                 d_invalidate(d);
 131                         ret = -ESTALE;
 132                 }
 133         }
 134         return ret;
 135 }
 136
 137 static int ovl_dentry_revalidate_common(struct dentry *dentry,
 138                                         unsigned int flags, bool weak)
 139 {
 140         struct ovl_entry *oe = dentry->d_fsdata;
 141         struct dentry *upper;
 142         unsigned int i;
 143         int ret = 1;
 144
 145         upper = ovl_dentry_upper(dentry);
 146         if (upper)
 147                 ret = ovl_revalidate_real(upper, flags, weak);
 148
 149         for (i = 0; ret > 0 && i < oe->numlower; i++) {
 150                 ret = ovl_revalidate_real(oe->lowerstack[i].dentry, flags,
 151                                           weak);
 152         }
 153         return ret;
 154 }
 155
 156 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 157 {
 158         return ovl_dentry_revalidate_common(dentry, flags, false);
 159 }
 160
 161 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 162 {
 163         return ovl_dentry_revalidate_common(dentry, flags, true);
 164 }
 165
 166 static const struct dentry_operations ovl_dentry_operations = {
 167         .d_release = ovl_dentry_release,
 168         .d_real = ovl_d_real,
 169         .d_revalidate = ovl_dentry_revalidate,
 170         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 171 };
 172
 173 static struct kmem_cache *ovl_inode_cachep;
 174
 175 static struct inode *ovl_alloc_inode(struct super_block *sb)
 176 {
 177         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 178
 179         if (!oi)
 180                 return NULL;
 181
 182         oi->cache = NULL;
 183         oi->redirect = NULL;
 184         oi->version = 0;
 185         oi->flags = 0;
 186         oi->__upperdentry = NULL;
 187         oi->lower = NULL;
 188         oi->lowerdata = NULL;
 189         mutex_init(&oi->lock);
 190
 191         return &oi->vfs_inode;
 192 }
 193
 194 static void ovl_free_inode(struct inode *inode)
 195 {
 196         struct ovl_inode *oi = OVL_I(inode);
 197
 198         kfree(oi->redirect);
 199         mutex_destroy(&oi->lock);
 200         kmem_cache_free(ovl_inode_cachep, oi);
 201 }
 202
 203 static void ovl_destroy_inode(struct inode *inode)
 204 {
 205         struct ovl_inode *oi = OVL_I(inode);
 206
 207         dput(oi->__upperdentry);
 208         iput(oi->lower);
 209         if (S_ISDIR(inode->i_mode))
 210                 ovl_dir_cache_free(inode);
 211         else
 212                 iput(oi->lowerdata);
 213 }
 214
 215 static void ovl_free_fs(struct ovl_fs *ofs)
 216 {
 217         struct vfsmount **mounts;
 218         unsigned i;
 219
 220         iput(ofs->workbasedir_trap);
 221         iput(ofs->indexdir_trap);
 222         iput(ofs->workdir_trap);
 223         dput(ofs->whiteout);
 224         dput(ofs->indexdir);
 225         dput(ofs->workdir);
 226         if (ofs->workdir_locked)
 227                 ovl_inuse_unlock(ofs->workbasedir);
 228         dput(ofs->workbasedir);
 229         if (ofs->upperdir_locked)
 230                 ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
 231
 232         /* Hack!  Reuse ofs->layers as a vfsmount array before freeing it */
 233         mounts = (struct vfsmount **) ofs->layers;
 234         for (i = 0; i < ofs->numlayer; i++) {
 235                 iput(ofs->layers[i].trap);
 236                 mounts[i] = ofs->layers[i].mnt;
 237         }
 238         kern_unmount_array(mounts, ofs->numlayer);
 239         kfree(ofs->layers);
 240         for (i = 0; i < ofs->numfs; i++)
 241                 free_anon_bdev(ofs->fs[i].pseudo_dev);
 242         kfree(ofs->fs);
 243
 244         kfree(ofs->config.lowerdir);
 245         kfree(ofs->config.upperdir);
 246         kfree(ofs->config.workdir);
 247         kfree(ofs->config.redirect_mode);
 248         if (ofs->creator_cred)
 249                 put_cred(ofs->creator_cred);
 250         kfree(ofs);
 251 }
 252
 253 static void ovl_put_super(struct super_block *sb)
 254 {
 255         struct ovl_fs *ofs = sb->s_fs_info;
 256
 257         ovl_free_fs(ofs);
 258 }
 259
 260 /* Sync real dirty inodes in upper filesystem (if it exists) */
 261 static int ovl_sync_fs(struct super_block *sb, int wait)
 262 {
 263         struct ovl_fs *ofs = sb->s_fs_info;
 264         struct super_block *upper_sb;
 265         int ret;
 266
 267         if (!ovl_upper_mnt(ofs))
 268                 return 0;
 269
 270         if (!ovl_should_sync(ofs))
 271                 return 0;
 272         /*
 273          * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
 274          * All the super blocks will be iterated, including upper_sb.
 275          *
 276          * If this is a syncfs(2) call, then we do need to call
 277          * sync_filesystem() on upper_sb, but enough if we do it when being
 278          * called with wait == 1.
 279          */
 280         if (!wait)
 281                 return 0;
 282
 283         upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 284
 285         down_read(&upper_sb->s_umount);
 286         ret = sync_filesystem(upper_sb);
 287         up_read(&upper_sb->s_umount);
 288
 289         return ret;
 290 }
 291
 292 /**
 293  * ovl_statfs
 294  * @sb: The overlayfs super block
 295  * @buf: The struct kstatfs to fill in with stats
 296  *
 297  * Get the filesystem statistics.  As writes always target the upper layer
 298  * filesystem pass the statfs to the upper filesystem (if it exists)
 299  */
 300 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 301 {
 302         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 303         struct dentry *root_dentry = dentry->d_sb->s_root;
 304         struct path path;
 305         int err;
 306
 307         ovl_path_real(root_dentry, &path);
 308
 309         err = vfs_statfs(&path, buf);
 310         if (!err) {
 311                 buf->f_namelen = ofs->namelen;
 312                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 313         }
 314
 315         return err;
 316 }
 317
 318 /* Will this overlay be forced to mount/remount ro? */
 319 static bool ovl_force_readonly(struct ovl_fs *ofs)
 320 {
 321         return (!ovl_upper_mnt(ofs) || !ofs->workdir);
 322 }
 323
 324 static const char *ovl_redirect_mode_def(void)
 325 {
 326         return ovl_redirect_dir_def ? "on" : "off";
 327 }
 328
 329 static const char * const ovl_xino_str[] = {
 330         "off",
 331         "auto",
 332         "on",
 333 };
 334
 335 static inline int ovl_xino_def(void)
 336 {
 337         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 338 }
 339
 340 /**
 341  * ovl_show_options
 342  *
 343  * Prints the mount options for a given superblock.
 344  * Returns zero; does not fail.
 345  */
 346 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 347 {
 348         struct super_block *sb = dentry->d_sb;
 349         struct ovl_fs *ofs = sb->s_fs_info;
 350
 351         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 352         if (ofs->config.upperdir) {
 353                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 354                 seq_show_option(m, "workdir", ofs->config.workdir);
 355         }
 356         if (ofs->config.default_permissions)
 357                 seq_puts(m, ",default_permissions");
 358         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 359                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 360         if (ofs->config.index != ovl_index_def)
 361                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 362         if (!ofs->config.uuid)
 363                 seq_puts(m, ",uuid=off");
 364         if (ofs->config.nfs_export != ovl_nfs_export_def)
 365                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 366                                                 "on" : "off");
 367         if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
 368                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 369         if (ofs->config.metacopy != ovl_metacopy_def)
 370                 seq_printf(m, ",metacopy=%s",
 371                            ofs->config.metacopy ? "on" : "off");
 372         if (ofs->config.ovl_volatile)
 373                 seq_puts(m, ",volatile");
 374         return 0;
 375 }
 376
 377 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 378 {
 379         struct ovl_fs *ofs = sb->s_fs_info;
 380         struct super_block *upper_sb;
 381         int ret = 0;
 382
 383         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 384                 return -EROFS;
 385
 386         if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
 387                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 388                 if (ovl_should_sync(ofs)) {
 389                         down_read(&upper_sb->s_umount);
 390                         ret = sync_filesystem(upper_sb);
 391                         up_read(&upper_sb->s_umount);
 392                 }
 393         }
 394
 395         return ret;
 396 }
 397
 398 static const struct super_operations ovl_super_operations = {
 399         .alloc_inode    = ovl_alloc_inode,
 400         .free_inode     = ovl_free_inode,
 401         .destroy_inode  = ovl_destroy_inode,
 402         .drop_inode     = generic_delete_inode,
 403         .put_super      = ovl_put_super,
 404         .sync_fs        = ovl_sync_fs,
 405         .statfs         = ovl_statfs,
 406         .show_options   = ovl_show_options,
 407         .remount_fs     = ovl_remount,
 408 };
 409
 410 enum {
 411         OPT_LOWERDIR,
 412         OPT_UPPERDIR,
 413         OPT_WORKDIR,
 414         OPT_DEFAULT_PERMISSIONS,
 415         OPT_REDIRECT_DIR,
 416         OPT_INDEX_ON,
 417         OPT_INDEX_OFF,
 418         OPT_UUID_ON,
 419         OPT_UUID_OFF,
 420         OPT_NFS_EXPORT_ON,
 421         OPT_USERXATTR,
 422         OPT_NFS_EXPORT_OFF,
 423         OPT_XINO_ON,
 424         OPT_XINO_OFF,
 425         OPT_XINO_AUTO,
 426         OPT_METACOPY_ON,
 427         OPT_METACOPY_OFF,
 428         OPT_VOLATILE,
 429         OPT_ERR,
 430 };
 431
 432 static const match_table_t ovl_tokens = {
 433         {OPT_LOWERDIR,                  "lowerdir=%s"},
 434         {OPT_UPPERDIR,                  "upperdir=%s"},
 435         {OPT_WORKDIR,                   "workdir=%s"},
 436         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 437         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 438         {OPT_INDEX_ON,                  "index=on"},
 439         {OPT_INDEX_OFF,                 "index=off"},
 440         {OPT_USERXATTR,                 "userxattr"},
 441         {OPT_UUID_ON,                   "uuid=on"},
 442         {OPT_UUID_OFF,                  "uuid=off"},
 443         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 444         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 445         {OPT_XINO_ON,                   "xino=on"},
 446         {OPT_XINO_OFF,                  "xino=off"},
 447         {OPT_XINO_AUTO,                 "xino=auto"},
 448         {OPT_METACOPY_ON,               "metacopy=on"},
 449         {OPT_METACOPY_OFF,              "metacopy=off"},
 450         {OPT_VOLATILE,                  "volatile"},
 451         {OPT_ERR,                       NULL}
 452 };
 453
 454 static char *ovl_next_opt(char **s)
 455 {
 456         char *sbegin = *s;
 457         char *p;
 458
 459         if (sbegin == NULL)
 460                 return NULL;
 461
 462         for (p = sbegin; *p; p++) {
 463                 if (*p == '\\') {
 464                         p++;
 465                         if (!*p)
 466                                 break;
 467                 } else if (*p == ',') {
 468                         *p = '\0';
 469                         *s = p + 1;
 470                         return sbegin;
 471                 }
 472         }
 473         *s = NULL;
 474         return sbegin;
 475 }
 476
 477 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 478 {
 479         if (strcmp(mode, "on") == 0) {
 480                 config->redirect_dir = true;
 481                 /*
 482                  * Does not make sense to have redirect creation without
 483                  * redirect following.
 484                  */
 485                 config->redirect_follow = true;
 486         } else if (strcmp(mode, "follow") == 0) {
 487                 config->redirect_follow = true;
 488         } else if (strcmp(mode, "off") == 0) {
 489                 if (ovl_redirect_always_follow)
 490                         config->redirect_follow = true;
 491         } else if (strcmp(mode, "nofollow") != 0) {
 492                 pr_err("bad mount option \"redirect_dir=%s\"\n",
 493                        mode);
 494                 return -EINVAL;
 495         }
 496
 497         return 0;
 498 }
 499
 500 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 501 {
 502         char *p;
 503         int err;
 504         bool metacopy_opt = false, redirect_opt = false;
 505         bool nfs_export_opt = false, index_opt = false;
 506
 507         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 508         if (!config->redirect_mode)
 509                 return -ENOMEM;
 510
 511         while ((p = ovl_next_opt(&opt)) != NULL) {
 512                 int token;
 513                 substring_t args[MAX_OPT_ARGS];
 514
 515                 if (!*p)
 516                         continue;
 517
 518                 token = match_token(p, ovl_tokens, args);
 519                 switch (token) {
 520                 case OPT_UPPERDIR:
 521                         kfree(config->upperdir);
 522                         config->upperdir = match_strdup(&args[0]);
 523                         if (!config->upperdir)
 524                                 return -ENOMEM;
 525                         break;
 526
 527                 case OPT_LOWERDIR:
 528                         kfree(config->lowerdir);
 529                         config->lowerdir = match_strdup(&args[0]);
 530                         if (!config->lowerdir)
 531                                 return -ENOMEM;
 532                         break;
 533
 534                 case OPT_WORKDIR:
 535                         kfree(config->workdir);
 536                         config->workdir = match_strdup(&args[0]);
 537                         if (!config->workdir)
 538                                 return -ENOMEM;
 539                         break;
 540
 541                 case OPT_DEFAULT_PERMISSIONS:
 542                         config->default_permissions = true;
 543                         break;
 544
 545                 case OPT_REDIRECT_DIR:
 546                         kfree(config->redirect_mode);
 547                         config->redirect_mode = match_strdup(&args[0]);
 548                         if (!config->redirect_mode)
 549                                 return -ENOMEM;
 550                         redirect_opt = true;
 551                         break;
 552
 553                 case OPT_INDEX_ON:
 554                         config->index = true;
 555                         index_opt = true;
 556                         break;
 557
 558                 case OPT_INDEX_OFF:
 559                         config->index = false;
 560                         index_opt = true;
 561                         break;
 562
 563                 case OPT_UUID_ON:
 564                         config->uuid = true;
 565                         break;
 566
 567                 case OPT_UUID_OFF:
 568                         config->uuid = false;
 569                         break;
 570
 571                 case OPT_NFS_EXPORT_ON:
 572                         config->nfs_export = true;
 573                         nfs_export_opt = true;
 574                         break;
 575
 576                 case OPT_NFS_EXPORT_OFF:
 577                         config->nfs_export = false;
 578                         nfs_export_opt = true;
 579                         break;
 580
 581                 case OPT_XINO_ON:
 582                         config->xino = OVL_XINO_ON;
 583                         break;
 584
 585                 case OPT_XINO_OFF:
 586                         config->xino = OVL_XINO_OFF;
 587                         break;
 588
 589                 case OPT_XINO_AUTO:
 590                         config->xino = OVL_XINO_AUTO;
 591                         break;
 592
 593                 case OPT_METACOPY_ON:
 594                         config->metacopy = true;
 595                         metacopy_opt = true;
 596                         break;
 597
 598                 case OPT_METACOPY_OFF:
 599                         config->metacopy = false;
 600                         metacopy_opt = true;
 601                         break;
 602
 603                 case OPT_VOLATILE:
 604                         config->ovl_volatile = true;
 605                         break;
 606
 607                 case OPT_USERXATTR:
 608                         config->userxattr = true;
 609                         break;
 610
 611                 default:
 612                         pr_err("unrecognized mount option \"%s\" or missing value\n",
 613                                         p);
 614                         return -EINVAL;
 615                 }
 616         }
 617
 618         /* Workdir/index are useless in non-upper mount */
 619         if (!config->upperdir) {
 620                 if (config->workdir) {
 621                         pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 622                                 config->workdir);
 623                         kfree(config->workdir);
 624                         config->workdir = NULL;
 625                 }
 626                 if (config->index && index_opt) {
 627                         pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
 628                         index_opt = false;
 629                 }
 630                 config->index = false;
 631         }
 632
 633         if (!config->upperdir && config->ovl_volatile) {
 634                 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
 635                 config->ovl_volatile = false;
 636         }
 637
 638         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 639         if (err)
 640                 return err;
 641
 642         /*
 643          * This is to make the logic below simpler.  It doesn't make any other
 644          * difference, since config->redirect_dir is only used for upper.
 645          */
 646         if (!config->upperdir && config->redirect_follow)
 647                 config->redirect_dir = true;
 648
 649         /* Resolve metacopy -> redirect_dir dependency */
 650         if (config->metacopy && !config->redirect_dir) {
 651                 if (metacopy_opt && redirect_opt) {
 652                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
 653                                config->redirect_mode);
 654                         return -EINVAL;
 655                 }
 656                 if (redirect_opt) {
 657                         /*
 658                          * There was an explicit redirect_dir=... that resulted
 659                          * in this conflict.
 660                          */
 661                         pr_info("disabling metacopy due to redirect_dir=%s\n",
 662                                 config->redirect_mode);
 663                         config->metacopy = false;
 664                 } else {
 665                         /* Automatically enable redirect otherwise. */
 666                         config->redirect_follow = config->redirect_dir = true;
 667                 }
 668         }
 669
 670         /* Resolve nfs_export -> index dependency */
 671         if (config->nfs_export && !config->index) {
 672                 if (!config->upperdir && config->redirect_follow) {
 673                         pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
 674                         config->nfs_export = false;
 675                 } else if (nfs_export_opt && index_opt) {
 676                         pr_err("conflicting options: nfs_export=on,index=off\n");
 677                         return -EINVAL;
 678                 } else if (index_opt) {
 679                         /*
 680                          * There was an explicit index=off that resulted
 681                          * in this conflict.
 682                          */
 683                         pr_info("disabling nfs_export due to index=off\n");
 684                         config->nfs_export = false;
 685                 } else {
 686                         /* Automatically enable index otherwise. */
 687                         config->index = true;
 688                 }
 689         }
 690
 691         /* Resolve nfs_export -> !metacopy dependency */
 692         if (config->nfs_export && config->metacopy) {
 693                 if (nfs_export_opt && metacopy_opt) {
 694                         pr_err("conflicting options: nfs_export=on,metacopy=on\n");
 695                         return -EINVAL;
 696                 }
 697                 if (metacopy_opt) {
 698                         /*
 699                          * There was an explicit metacopy=on that resulted
 700                          * in this conflict.
 701                          */
 702                         pr_info("disabling nfs_export due to metacopy=on\n");
 703                         config->nfs_export = false;
 704                 } else {
 705                         /*
 706                          * There was an explicit nfs_export=on that resulted
 707                          * in this conflict.
 708                          */
 709                         pr_info("disabling metacopy due to nfs_export=on\n");
 710                         config->metacopy = false;
 711                 }
 712         }
 713
 714
 715         /* Resolve userxattr -> !redirect && !metacopy dependency */
 716         if (config->userxattr) {
 717                 if (config->redirect_follow && redirect_opt) {
 718                         pr_err("conflicting options: userxattr,redirect_dir=%s\n",
 719                                config->redirect_mode);
 720                         return -EINVAL;
 721                 }
 722                 if (config->metacopy && metacopy_opt) {
 723                         pr_err("conflicting options: userxattr,metacopy=on\n");
 724                         return -EINVAL;
 725                 }
 726                 /*
 727                  * Silently disable default setting of redirect and metacopy.
 728                  * This shall be the default in the future as well: these
 729                  * options must be explicitly enabled if used together with
 730                  * userxattr.
 731                  */
 732                 config->redirect_dir = config->redirect_follow = false;
 733                 config->metacopy = false;
 734         }
 735
 736         return 0;
 737 }
 738
 739 #define OVL_WORKDIR_NAME "work"
 740 #define OVL_INDEXDIR_NAME "index"
 741
 742 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 743                                          const char *name, bool persist)
 744 {
 745         struct inode *dir =  ofs->workbasedir->d_inode;
 746         struct vfsmount *mnt = ovl_upper_mnt(ofs);
 747         struct dentry *work;
 748         int err;
 749         bool retried = false;
 750
 751         inode_lock_nested(dir, I_MUTEX_PARENT);
 752 retry:
 753         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
 754
 755         if (!IS_ERR(work)) {
 756                 struct iattr attr = {
 757                         .ia_valid = ATTR_MODE,
 758                         .ia_mode = S_IFDIR | 0,
 759                 };
 760
 761                 if (work->d_inode) {
 762                         err = -EEXIST;
 763                         if (retried)
 764                                 goto out_dput;
 765
 766                         if (persist)
 767                                 goto out_unlock;
 768
 769                         retried = true;
 770                         err = ovl_workdir_cleanup(dir, mnt, work, 0);
 771                         dput(work);
 772                         if (err == -EINVAL) {
 773                                 work = ERR_PTR(err);
 774                                 goto out_unlock;
 775                         }
 776                         goto retry;
 777                 }
 778
 779                 work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
 780                 err = PTR_ERR(work);
 781                 if (IS_ERR(work))
 782                         goto out_err;
 783
 784                 /*
 785                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 786                  *
 787                  * a) success (there was a POSIX ACL xattr and was removed)
 788                  * b) -ENODATA (there was no POSIX ACL xattr)
 789                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 790                  *
 791                  * There are various other error values that could effectively
 792                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 793                  * if the xattr name is too long), but the set of filesystems
 794                  * allowed as upper are limited to "normal" ones, where checking
 795                  * for the above two errors is sufficient.
 796                  */
 797                 err = vfs_removexattr(&init_user_ns, work,
 798                                       XATTR_NAME_POSIX_ACL_DEFAULT);
 799                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 800                         goto out_dput;
 801
 802                 err = vfs_removexattr(&init_user_ns, work,
 803                                       XATTR_NAME_POSIX_ACL_ACCESS);
 804                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 805                         goto out_dput;
 806
 807                 /* Clear any inherited mode bits */
 808                 inode_lock(work->d_inode);
 809                 err = notify_change(&init_user_ns, work, &attr, NULL);
 810                 inode_unlock(work->d_inode);
 811                 if (err)
 812                         goto out_dput;
 813         } else {
 814                 err = PTR_ERR(work);
 815                 goto out_err;
 816         }
 817 out_unlock:
 818         inode_unlock(dir);
 819         return work;
 820
 821 out_dput:
 822         dput(work);
 823 out_err:
 824         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
 825                 ofs->config.workdir, name, -err);
 826         work = NULL;
 827         goto out_unlock;
 828 }
 829
 830 static void ovl_unescape(char *s)
 831 {
 832         char *d = s;
 833
 834         for (;; s++, d++) {
 835                 if (*s == '\\')
 836                         s++;
 837                 *d = *s;
 838                 if (!*s)
 839                         break;
 840         }
 841 }
 842
 843 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 844 {
 845         int err = -EINVAL;
 846
 847         if (!*name) {
 848                 pr_err("empty lowerdir\n");
 849                 goto out;
 850         }
 851         err = kern_path(name, LOOKUP_FOLLOW, path);
 852         if (err) {
 853                 pr_err("failed to resolve '%s': %i\n", name, err);
 854                 goto out;
 855         }
 856         err = -EINVAL;
 857         if (ovl_dentry_weird(path->dentry)) {
 858                 pr_err("filesystem on '%s' not supported\n", name);
 859                 goto out_put;
 860         }
 861         if (!d_is_dir(path->dentry)) {
 862                 pr_err("'%s' not a directory\n", name);
 863                 goto out_put;
 864         }
 865         return 0;
 866
 867 out_put:
 868         path_put_init(path);
 869 out:
 870         return err;
 871 }
 872
 873 static int ovl_mount_dir(const char *name, struct path *path)
 874 {
 875         int err = -ENOMEM;
 876         char *tmp = kstrdup(name, GFP_KERNEL);
 877
 878         if (tmp) {
 879                 ovl_unescape(tmp);
 880                 err = ovl_mount_dir_noesc(tmp, path);
 881
 882                 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
 883                         pr_err("filesystem on '%s' not supported as upperdir\n",
 884                                tmp);
 885                         path_put_init(path);
 886                         err = -EINVAL;
 887                 }
 888                 kfree(tmp);
 889         }
 890         return err;
 891 }
 892
 893 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 894                              const char *name)
 895 {
 896         struct kstatfs statfs;
 897         int err = vfs_statfs(path, &statfs);
 898
 899         if (err)
 900                 pr_err("statfs failed on '%s'\n", name);
 901         else
 902                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 903
 904         return err;
 905 }
 906
 907 static int ovl_lower_dir(const char *name, struct path *path,
 908                          struct ovl_fs *ofs, int *stack_depth)
 909 {
 910         int fh_type;
 911         int err;
 912
 913         err = ovl_mount_dir_noesc(name, path);
 914         if (err)
 915                 return err;
 916
 917         err = ovl_check_namelen(path, ofs, name);
 918         if (err)
 919                 return err;
 920
 921         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 922
 923         /*
 924          * The inodes index feature and NFS export need to encode and decode
 925          * file handles, so they require that all layers support them.
 926          */
 927         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 928         if ((ofs->config.nfs_export ||
 929              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 930                 ofs->config.index = false;
 931                 ofs->config.nfs_export = false;
 932                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 933                         name);
 934         }
 935
 936         /* Check if lower fs has 32bit inode numbers */
 937         if (fh_type != FILEID_INO32_GEN)
 938                 ofs->xino_mode = -1;
 939
 940         return 0;
 941 }
 942
 943 /* Workdir should not be subdir of upperdir and vice versa */
 944 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 945 {
 946         bool ok = false;
 947
 948         if (workdir != upperdir) {
 949                 ok = (lock_rename(workdir, upperdir) == NULL);
 950                 unlock_rename(workdir, upperdir);
 951         }
 952         return ok;
 953 }
 954
 955 static unsigned int ovl_split_lowerdirs(char *str)
 956 {
 957         unsigned int ctr = 1;
 958         char *s, *d;
 959
 960         for (s = d = str;; s++, d++) {
 961                 if (*s == '\\') {
 962                         s++;
 963                 } else if (*s == ':') {
 964                         *d = '\0';
 965                         ctr++;
 966                         continue;
 967                 }
 968                 *d = *s;
 969                 if (!*s)
 970                         break;
 971         }
 972         return ctr;
 973 }
 974
 975 static int __maybe_unused
 976 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 977                         struct dentry *dentry, struct inode *inode,
 978                         const char *name, void *buffer, size_t size)
 979 {
 980         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 981 }
 982
 983 static int __maybe_unused
 984 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 985                         struct user_namespace *mnt_userns,
 986                         struct dentry *dentry, struct inode *inode,
 987                         const char *name, const void *value,
 988                         size_t size, int flags)
 989 {
 990         struct dentry *workdir = ovl_workdir(dentry);
 991         struct inode *realinode = ovl_inode_real(inode);
 992         struct posix_acl *acl = NULL;
 993         int err;
 994
 995         /* Check that everything is OK before copy-up */
 996         if (value) {
 997                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 998                 if (IS_ERR(acl))
 999                         return PTR_ERR(acl);
1000         }
1001         err = -EOPNOTSUPP;
1002         if (!IS_POSIXACL(d_inode(workdir)))
1003                 goto out_acl_release;
1004         if (!realinode->i_op->set_acl)
1005                 goto out_acl_release;
1006         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1007                 err = acl ? -EACCES : 0;
1008                 goto out_acl_release;
1009         }
1010         err = -EPERM;
1011         if (!inode_owner_or_capable(&init_user_ns, inode))
1012                 goto out_acl_release;
1013
1014         posix_acl_release(acl);
1015
1016         /*
1017          * Check if sgid bit needs to be cleared (actual setacl operation will
1018          * be done with mounter's capabilities and so that won't do it for us).
1019          */
1020         if (unlikely(inode->i_mode & S_ISGID) &&
1021             handler->flags == ACL_TYPE_ACCESS &&
1022             !in_group_p(inode->i_gid) &&
1023             !capable_wrt_inode_uidgid(&init_user_ns, inode, CAP_FSETID)) {
1024                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
1025
1026                 err = ovl_setattr(dentry, &iattr);
1027                 if (err)
1028                         return err;
1029         }
1030
1031         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
1032         if (!err)
1033                 ovl_copyattr(ovl_inode_real(inode), inode);
1034
1035         return err;
1036
1037 out_acl_release:
1038         posix_acl_release(acl);
1039         return err;
1040 }
1041
1042 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1043                              struct dentry *dentry, struct inode *inode,
1044                              const char *name, void *buffer, size_t size)
1045 {
1046         return -EOPNOTSUPP;
1047 }
1048
1049 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1050                              struct user_namespace *mnt_userns,
1051                              struct dentry *dentry, struct inode *inode,
1052                              const char *name, const void *value,
1053                              size_t size, int flags)
1054 {
1055         return -EOPNOTSUPP;
1056 }
1057
1058 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1059                                struct dentry *dentry, struct inode *inode,
1060                                const char *name, void *buffer, size_t size)
1061 {
1062         return ovl_xattr_get(dentry, inode, name, buffer, size);
1063 }
1064
1065 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1066                                struct user_namespace *mnt_userns,
1067                                struct dentry *dentry, struct inode *inode,
1068                                const char *name, const void *value,
1069                                size_t size, int flags)
1070 {
1071         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1072 }
1073
1074 static const struct xattr_handler __maybe_unused
1075 ovl_posix_acl_access_xattr_handler = {
1076         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1077         .flags = ACL_TYPE_ACCESS,
1078         .get = ovl_posix_acl_xattr_get,
1079         .set = ovl_posix_acl_xattr_set,
1080 };
1081
1082 static const struct xattr_handler __maybe_unused
1083 ovl_posix_acl_default_xattr_handler = {
1084         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1085         .flags = ACL_TYPE_DEFAULT,
1086         .get = ovl_posix_acl_xattr_get,
1087         .set = ovl_posix_acl_xattr_set,
1088 };
1089
1090 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1091         .prefix = OVL_XATTR_TRUSTED_PREFIX,
1092         .get = ovl_own_xattr_get,
1093         .set = ovl_own_xattr_set,
1094 };
1095
1096 static const struct xattr_handler ovl_own_user_xattr_handler = {
1097         .prefix = OVL_XATTR_USER_PREFIX,
1098         .get = ovl_own_xattr_get,
1099         .set = ovl_own_xattr_set,
1100 };
1101
1102 static const struct xattr_handler ovl_other_xattr_handler = {
1103         .prefix = "", /* catch all */
1104         .get = ovl_other_xattr_get,
1105         .set = ovl_other_xattr_set,
1106 };
1107
1108 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1109 #ifdef CONFIG_FS_POSIX_ACL
1110         &ovl_posix_acl_access_xattr_handler,
1111         &ovl_posix_acl_default_xattr_handler,
1112 #endif
1113         &ovl_own_trusted_xattr_handler,
1114         &ovl_other_xattr_handler,
1115         NULL
1116 };
1117
1118 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1119 #ifdef CONFIG_FS_POSIX_ACL
1120         &ovl_posix_acl_access_xattr_handler,
1121         &ovl_posix_acl_default_xattr_handler,
1122 #endif
1123         &ovl_own_user_xattr_handler,
1124         &ovl_other_xattr_handler,
1125         NULL
1126 };
1127
1128 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1129                           struct inode **ptrap, const char *name)
1130 {
1131         struct inode *trap;
1132         int err;
1133
1134         trap = ovl_get_trap_inode(sb, dir);
1135         err = PTR_ERR_OR_ZERO(trap);
1136         if (err) {
1137                 if (err == -ELOOP)
1138                         pr_err("conflicting %s path\n", name);
1139                 return err;
1140         }
1141
1142         *ptrap = trap;
1143         return 0;
1144 }
1145
1146 /*
1147  * Determine how we treat concurrent use of upperdir/workdir based on the
1148  * index feature. This is papering over mount leaks of container runtimes,
1149  * for example, an old overlay mount is leaked and now its upperdir is
1150  * attempted to be used as a lower layer in a new overlay mount.
1151  */
1152 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1153 {
1154         if (ofs->config.index) {
1155                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1156                        name);
1157                 return -EBUSY;
1158         } else {
1159                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1160                         name);
1161                 return 0;
1162         }
1163 }
1164
1165 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1166                          struct ovl_layer *upper_layer, struct path *upperpath)
1167 {
1168         struct vfsmount *upper_mnt;
1169         int err;
1170
1171         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1172         if (err)
1173                 goto out;
1174
1175         /* Upper fs should not be r/o */
1176         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
1177                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1178                 err = -EINVAL;
1179                 goto out;
1180         }
1181
1182         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1183         if (err)
1184                 goto out;
1185
1186         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1187                              "upperdir");
1188         if (err)
1189                 goto out;
1190
1191         upper_mnt = clone_private_mount(upperpath);
1192         err = PTR_ERR(upper_mnt);
1193         if (IS_ERR(upper_mnt)) {
1194                 pr_err("failed to clone upperpath\n");
1195                 goto out;
1196         }
1197
1198         /* Don't inherit atime flags */
1199         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1200         upper_layer->mnt = upper_mnt;
1201         upper_layer->idx = 0;
1202         upper_layer->fsid = 0;
1203
1204         /*
1205          * Inherit SB_NOSEC flag from upperdir.
1206          *
1207          * This optimization changes behavior when a security related attribute
1208          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1209          * okay because we don't yet have guarantees in that case, but it will
1210          * need careful treatment once we want to honour changes to underlying
1211          * filesystems.
1212          */
1213         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1214                 sb->s_flags |= SB_NOSEC;
1215
1216         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1217                 ofs->upperdir_locked = true;
1218         } else {
1219                 err = ovl_report_in_use(ofs, "upperdir");
1220                 if (err)
1221                         goto out;
1222         }
1223
1224         err = 0;
1225 out:
1226         return err;
1227 }
1228
1229 /*
1230  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1231  * negative values if error is encountered.
1232  */
1233 static int ovl_check_rename_whiteout(struct dentry *workdir)
1234 {
1235         struct inode *dir = d_inode(workdir);
1236         struct dentry *temp;
1237         struct dentry *dest;
1238         struct dentry *whiteout;
1239         struct name_snapshot name;
1240         int err;
1241
1242         inode_lock_nested(dir, I_MUTEX_PARENT);
1243
1244         temp = ovl_create_temp(workdir, OVL_CATTR(S_IFREG | 0));
1245         err = PTR_ERR(temp);
1246         if (IS_ERR(temp))
1247                 goto out_unlock;
1248
1249         dest = ovl_lookup_temp(workdir);
1250         err = PTR_ERR(dest);
1251         if (IS_ERR(dest)) {
1252                 dput(temp);
1253                 goto out_unlock;
1254         }
1255
1256         /* Name is inline and stable - using snapshot as a copy helper */
1257         take_dentry_name_snapshot(&name, temp);
1258         err = ovl_do_rename(dir, temp, dir, dest, RENAME_WHITEOUT);
1259         if (err) {
1260                 if (err == -EINVAL)
1261                         err = 0;
1262                 goto cleanup_temp;
1263         }
1264
1265         whiteout = lookup_one_len(name.name.name, workdir, name.name.len);
1266         err = PTR_ERR(whiteout);
1267         if (IS_ERR(whiteout))
1268                 goto cleanup_temp;
1269
1270         err = ovl_is_whiteout(whiteout);
1271
1272         /* Best effort cleanup of whiteout and temp file */
1273         if (err)
1274                 ovl_cleanup(dir, whiteout);
1275         dput(whiteout);
1276
1277 cleanup_temp:
1278         ovl_cleanup(dir, temp);
1279         release_dentry_name_snapshot(&name);
1280         dput(temp);
1281         dput(dest);
1282
1283 out_unlock:
1284         inode_unlock(dir);
1285
1286         return err;
1287 }
1288
1289 static struct dentry *ovl_lookup_or_create(struct dentry *parent,
1290                                            const char *name, umode_t mode)
1291 {
1292         size_t len = strlen(name);
1293         struct dentry *child;
1294
1295         inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1296         child = lookup_one_len(name, parent, len);
1297         if (!IS_ERR(child) && !child->d_inode)
1298                 child = ovl_create_real(parent->d_inode, child,
1299                                         OVL_CATTR(mode));
1300         inode_unlock(parent->d_inode);
1301         dput(parent);
1302
1303         return child;
1304 }
1305
1306 /*
1307  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1308  * present.
1309  */
1310 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1311 {
1312         unsigned int ctr;
1313         struct dentry *d = dget(ofs->workbasedir);
1314         static const char *const volatile_path[] = {
1315                 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1316         };
1317         const char *const *name = volatile_path;
1318
1319         for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1320                 d = ovl_lookup_or_create(d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1321                 if (IS_ERR(d))
1322                         return PTR_ERR(d);
1323         }
1324         dput(d);
1325         return 0;
1326 }
1327
1328 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1329                             struct path *workpath)
1330 {
1331         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1332         struct dentry *temp, *workdir;
1333         bool rename_whiteout;
1334         bool d_type;
1335         int fh_type;
1336         int err;
1337
1338         err = mnt_want_write(mnt);
1339         if (err)
1340                 return err;
1341
1342         workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1343         err = PTR_ERR(workdir);
1344         if (IS_ERR_OR_NULL(workdir))
1345                 goto out;
1346
1347         ofs->workdir = workdir;
1348
1349         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1350         if (err)
1351                 goto out;
1352
1353         /*
1354          * Upper should support d_type, else whiteouts are visible.  Given
1355          * workdir and upper are on same fs, we can do iterate_dir() on
1356          * workdir. This check requires successful creation of workdir in
1357          * previous step.
1358          */
1359         err = ovl_check_d_type_supported(workpath);
1360         if (err < 0)
1361                 goto out;
1362
1363         d_type = err;
1364         if (!d_type)
1365                 pr_warn("upper fs needs to support d_type.\n");
1366
1367         /* Check if upper/work fs supports O_TMPFILE */
1368         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1369         ofs->tmpfile = !IS_ERR(temp);
1370         if (ofs->tmpfile)
1371                 dput(temp);
1372         else
1373                 pr_warn("upper fs does not support tmpfile.\n");
1374
1375
1376         /* Check if upper/work fs supports RENAME_WHITEOUT */
1377         err = ovl_check_rename_whiteout(ofs->workdir);
1378         if (err < 0)
1379                 goto out;
1380
1381         rename_whiteout = err;
1382         if (!rename_whiteout)
1383                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1384
1385         /*
1386          * Check if upper/work fs supports (trusted|user).overlay.* xattr
1387          */
1388         err = ovl_do_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1389         if (err) {
1390                 ofs->noxattr = true;
1391                 ofs->config.index = false;
1392                 ofs->config.metacopy = false;
1393                 pr_warn("upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1394                 err = 0;
1395         } else {
1396                 ovl_do_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1397         }
1398
1399         /*
1400          * We allowed sub-optimal upper fs configuration and don't want to break
1401          * users over kernel upgrade, but we never allowed remote upper fs, so
1402          * we can enforce strict requirements for remote upper fs.
1403          */
1404         if (ovl_dentry_remote(ofs->workdir) &&
1405             (!d_type || !rename_whiteout || ofs->noxattr)) {
1406                 pr_err("upper fs missing required features.\n");
1407                 err = -EINVAL;
1408                 goto out;
1409         }
1410
1411         /*
1412          * For volatile mount, create a incompat/volatile/dirty file to keep
1413          * track of it.
1414          */
1415         if (ofs->config.ovl_volatile) {
1416                 err = ovl_create_volatile_dirty(ofs);
1417                 if (err < 0) {
1418                         pr_err("Failed to create volatile/dirty file.\n");
1419                         goto out;
1420                 }
1421         }
1422
1423         /* Check if upper/work fs supports file handles */
1424         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1425         if (ofs->config.index && !fh_type) {
1426                 ofs->config.index = false;
1427                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1428         }
1429
1430         /* Check if upper fs has 32bit inode numbers */
1431         if (fh_type != FILEID_INO32_GEN)
1432                 ofs->xino_mode = -1;
1433
1434         /* NFS export of r/w mount depends on index */
1435         if (ofs->config.nfs_export && !ofs->config.index) {
1436                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1437                 ofs->config.nfs_export = false;
1438         }
1439 out:
1440         mnt_drop_write(mnt);
1441         return err;
1442 }
1443
1444 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1445                            struct path *upperpath)
1446 {
1447         int err;
1448         struct path workpath = { };
1449
1450         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1451         if (err)
1452                 goto out;
1453
1454         err = -EINVAL;
1455         if (upperpath->mnt != workpath.mnt) {
1456                 pr_err("workdir and upperdir must reside under the same mount\n");
1457                 goto out;
1458         }
1459         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1460                 pr_err("workdir and upperdir must be separate subtrees\n");
1461                 goto out;
1462         }
1463
1464         ofs->workbasedir = dget(workpath.dentry);
1465
1466         if (ovl_inuse_trylock(ofs->workbasedir)) {
1467                 ofs->workdir_locked = true;
1468         } else {
1469                 err = ovl_report_in_use(ofs, "workdir");
1470                 if (err)
1471                         goto out;
1472         }
1473
1474         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1475                              "workdir");
1476         if (err)
1477                 goto out;
1478
1479         err = ovl_make_workdir(sb, ofs, &workpath);
1480
1481 out:
1482         path_put(&workpath);
1483
1484         return err;
1485 }
1486
1487 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1488                             struct ovl_entry *oe, struct path *upperpath)
1489 {
1490         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1491         struct dentry *indexdir;
1492         int err;
1493
1494         err = mnt_want_write(mnt);
1495         if (err)
1496                 return err;
1497
1498         /* Verify lower root is upper root origin */
1499         err = ovl_verify_origin(ofs, upperpath->dentry,
1500                                 oe->lowerstack[0].dentry, true);
1501         if (err) {
1502                 pr_err("failed to verify upper root origin\n");
1503                 goto out;
1504         }
1505
1506         /* index dir will act also as workdir */
1507         iput(ofs->workdir_trap);
1508         ofs->workdir_trap = NULL;
1509         dput(ofs->workdir);
1510         ofs->workdir = NULL;
1511         indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1512         if (IS_ERR(indexdir)) {
1513                 err = PTR_ERR(indexdir);
1514         } else if (indexdir) {
1515                 ofs->indexdir = indexdir;
1516                 ofs->workdir = dget(indexdir);
1517
1518                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1519                                      "indexdir");
1520                 if (err)
1521                         goto out;
1522
1523                 /*
1524                  * Verify upper root is exclusively associated with index dir.
1525                  * Older kernels stored upper fh in ".overlay.origin"
1526                  * xattr. If that xattr exists, verify that it is a match to
1527                  * upper dir file handle. In any case, verify or set xattr
1528                  * ".overlay.upper" to indicate that index may have
1529                  * directory entries.
1530                  */
1531                 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1532                         err = ovl_verify_set_fh(ofs, ofs->indexdir,
1533                                                 OVL_XATTR_ORIGIN,
1534                                                 upperpath->dentry, true, false);
1535                         if (err)
1536                                 pr_err("failed to verify index dir 'origin' xattr\n");
1537                 }
1538                 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1539                                        true);
1540                 if (err)
1541                         pr_err("failed to verify index dir 'upper' xattr\n");
1542
1543                 /* Cleanup bad/stale/orphan index entries */
1544                 if (!err)
1545                         err = ovl_indexdir_cleanup(ofs);
1546         }
1547         if (err || !ofs->indexdir)
1548                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1549
1550 out:
1551         mnt_drop_write(mnt);
1552         return err;
1553 }
1554
1555 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1556 {
1557         unsigned int i;
1558
1559         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1560                 return true;
1561
1562         /*
1563          * We allow using single lower with null uuid for index and nfs_export
1564          * for example to support those features with single lower squashfs.
1565          * To avoid regressions in setups of overlay with re-formatted lower
1566          * squashfs, do not allow decoding origin with lower null uuid unless
1567          * user opted-in to one of the new features that require following the
1568          * lower inode of non-dir upper.
1569          */
1570         if (!ofs->config.index && !ofs->config.metacopy && !ofs->config.xino &&
1571             uuid_is_null(uuid))
1572                 return false;
1573
1574         for (i = 0; i < ofs->numfs; i++) {
1575                 /*
1576                  * We use uuid to associate an overlay lower file handle with a
1577                  * lower layer, so we can accept lower fs with null uuid as long
1578                  * as all lower layers with null uuid are on the same fs.
1579                  * if we detect multiple lower fs with the same uuid, we
1580                  * disable lower file handle decoding on all of them.
1581                  */
1582                 if (ofs->fs[i].is_lower &&
1583                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1584                         ofs->fs[i].bad_uuid = true;
1585                         return false;
1586                 }
1587         }
1588         return true;
1589 }
1590
1591 /* Get a unique fsid for the layer */
1592 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1593 {
1594         struct super_block *sb = path->mnt->mnt_sb;
1595         unsigned int i;
1596         dev_t dev;
1597         int err;
1598         bool bad_uuid = false;
1599
1600         for (i = 0; i < ofs->numfs; i++) {
1601                 if (ofs->fs[i].sb == sb)
1602                         return i;
1603         }
1604
1605         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1606                 bad_uuid = true;
1607                 if (ofs->config.index || ofs->config.nfs_export) {
1608                         ofs->config.index = false;
1609                         ofs->config.nfs_export = false;
1610                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to index=off,nfs_export=off.\n",
1611                                 uuid_is_null(&sb->s_uuid) ? "null" :
1612                                                             "conflicting",
1613                                 path->dentry);
1614                 }
1615         }
1616
1617         err = get_anon_bdev(&dev);
1618         if (err) {
1619                 pr_err("failed to get anonymous bdev for lowerpath\n");
1620                 return err;
1621         }
1622
1623         ofs->fs[ofs->numfs].sb = sb;
1624         ofs->fs[ofs->numfs].pseudo_dev = dev;
1625         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1626
1627         return ofs->numfs++;
1628 }
1629
1630 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1631                           struct path *stack, unsigned int numlower,
1632                           struct ovl_layer *layers)
1633 {
1634         int err;
1635         unsigned int i;
1636
1637         err = -ENOMEM;
1638         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1639         if (ofs->fs == NULL)
1640                 goto out;
1641
1642         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1643         ofs->numfs++;
1644
1645         /*
1646          * All lower layers that share the same fs as upper layer, use the same
1647          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1648          * only overlay to simplify ovl_fs_free().
1649          * is_lower will be set if upper fs is shared with a lower layer.
1650          */
1651         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1652         if (err) {
1653                 pr_err("failed to get anonymous bdev for upper fs\n");
1654                 goto out;
1655         }
1656
1657         if (ovl_upper_mnt(ofs)) {
1658                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1659                 ofs->fs[0].is_lower = false;
1660         }
1661
1662         for (i = 0; i < numlower; i++) {
1663                 struct vfsmount *mnt;
1664                 struct inode *trap;
1665                 int fsid;
1666
1667                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1668                 if (err < 0)
1669                         goto out;
1670
1671                 /*
1672                  * Check if lower root conflicts with this overlay layers before
1673                  * checking if it is in-use as upperdir/workdir of "another"
1674                  * mount, because we do not bother to check in ovl_is_inuse() if
1675                  * the upperdir/workdir is in fact in-use by our
1676                  * upperdir/workdir.
1677                  */
1678                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1679                 if (err)
1680                         goto out;
1681
1682                 if (ovl_is_inuse(stack[i].dentry)) {
1683                         err = ovl_report_in_use(ofs, "lowerdir");
1684                         if (err) {
1685                                 iput(trap);
1686                                 goto out;
1687                         }
1688                 }
1689
1690                 mnt = clone_private_mount(&stack[i]);
1691                 err = PTR_ERR(mnt);
1692                 if (IS_ERR(mnt)) {
1693                         pr_err("failed to clone lowerpath\n");
1694                         iput(trap);
1695                         goto out;
1696                 }
1697
1698                 /*
1699                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1700                  * will fail instead of modifying lower fs.
1701                  */
1702                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1703
1704                 layers[ofs->numlayer].trap = trap;
1705                 layers[ofs->numlayer].mnt = mnt;
1706                 layers[ofs->numlayer].idx = ofs->numlayer;
1707                 layers[ofs->numlayer].fsid = fsid;
1708                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1709                 ofs->numlayer++;
1710                 ofs->fs[fsid].is_lower = true;
1711         }
1712
1713         /*
1714          * When all layers on same fs, overlay can use real inode numbers.
1715          * With mount option "xino=<on|auto>", mounter declares that there are
1716          * enough free high bits in underlying fs to hold the unique fsid.
1717          * If overlayfs does encounter underlying inodes using the high xino
1718          * bits reserved for fsid, it emits a warning and uses the original
1719          * inode number or a non persistent inode number allocated from a
1720          * dedicated range.
1721          */
1722         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1723                 if (ofs->config.xino == OVL_XINO_ON)
1724                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1725                 ofs->xino_mode = 0;
1726         } else if (ofs->config.xino == OVL_XINO_OFF) {
1727                 ofs->xino_mode = -1;
1728         } else if (ofs->xino_mode < 0) {
1729                 /*
1730                  * This is a roundup of number of bits needed for encoding
1731                  * fsid, where fsid 0 is reserved for upper fs (even with
1732                  * lower only overlay) +1 extra bit is reserved for the non
1733                  * persistent inode number range that is used for resolving
1734                  * xino lower bits overflow.
1735                  */
1736                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1737                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1738         }
1739
1740         if (ofs->xino_mode > 0) {
1741                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1742                         ofs->xino_mode);
1743         }
1744
1745         err = 0;
1746 out:
1747         return err;
1748 }
1749
1750 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1751                                 const char *lower, unsigned int numlower,
1752                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1753 {
1754         int err;
1755         struct path *stack = NULL;
1756         unsigned int i;
1757         struct ovl_entry *oe;
1758
1759         if (!ofs->config.upperdir && numlower == 1) {
1760                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1761                 return ERR_PTR(-EINVAL);
1762         }
1763
1764         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1765         if (!stack)
1766                 return ERR_PTR(-ENOMEM);
1767
1768         err = -EINVAL;
1769         for (i = 0; i < numlower; i++) {
1770                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1771                 if (err)
1772                         goto out_err;
1773
1774                 lower = strchr(lower, '\0') + 1;
1775         }
1776
1777         err = -EINVAL;
1778         sb->s_stack_depth++;
1779         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1780                 pr_err("maximum fs stacking depth exceeded\n");
1781                 goto out_err;
1782         }
1783
1784         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1785         if (err)
1786                 goto out_err;
1787
1788         err = -ENOMEM;
1789         oe = ovl_alloc_entry(numlower);
1790         if (!oe)
1791                 goto out_err;
1792
1793         for (i = 0; i < numlower; i++) {
1794                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1795                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1796         }
1797
1798 out:
1799         for (i = 0; i < numlower; i++)
1800                 path_put(&stack[i]);
1801         kfree(stack);
1802
1803         return oe;
1804
1805 out_err:
1806         oe = ERR_PTR(err);
1807         goto out;
1808 }
1809
1810 /*
1811  * Check if this layer root is a descendant of:
1812  * - another layer of this overlayfs instance
1813  * - upper/work dir of any overlayfs instance
1814  */
1815 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1816                            struct dentry *dentry, const char *name)
1817 {
1818         struct dentry *next = dentry, *parent;
1819         int err = 0;
1820
1821         if (!dentry)
1822                 return 0;
1823
1824         parent = dget_parent(next);
1825
1826         /* Walk back ancestors to root (inclusive) looking for traps */
1827         while (!err && parent != next) {
1828                 if (ovl_lookup_trap_inode(sb, parent)) {
1829                         err = -ELOOP;
1830                         pr_err("overlapping %s path\n", name);
1831                 } else if (ovl_is_inuse(parent)) {
1832                         err = ovl_report_in_use(ofs, name);
1833                 }
1834                 next = parent;
1835                 parent = dget_parent(next);
1836                 dput(next);
1837         }
1838
1839         dput(parent);
1840
1841         return err;
1842 }
1843
1844 /*
1845  * Check if any of the layers or work dirs overlap.
1846  */
1847 static int ovl_check_overlapping_layers(struct super_block *sb,
1848                                         struct ovl_fs *ofs)
1849 {
1850         int i, err;
1851
1852         if (ovl_upper_mnt(ofs)) {
1853                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1854                                       "upperdir");
1855                 if (err)
1856                         return err;
1857
1858                 /*
1859                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1860                  * this instance and covers overlapping work and index dirs,
1861                  * unless work or index dir have been moved since created inside
1862                  * workbasedir.  In that case, we already have their traps in
1863                  * inode cache and we will catch that case on lookup.
1864                  */
1865                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir");
1866                 if (err)
1867                         return err;
1868         }
1869
1870         for (i = 1; i < ofs->numlayer; i++) {
1871                 err = ovl_check_layer(sb, ofs,
1872                                       ofs->layers[i].mnt->mnt_root,
1873                                       "lowerdir");
1874                 if (err)
1875                         return err;
1876         }
1877
1878         return 0;
1879 }
1880
1881 static struct dentry *ovl_get_root(struct super_block *sb,
1882                                    struct dentry *upperdentry,
1883                                    struct ovl_entry *oe)
1884 {
1885         struct dentry *root;
1886         struct ovl_path *lowerpath = &oe->lowerstack[0];
1887         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1888         int fsid = lowerpath->layer->fsid;
1889         struct ovl_inode_params oip = {
1890                 .upperdentry = upperdentry,
1891                 .lowerpath = lowerpath,
1892         };
1893
1894         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1895         if (!root)
1896                 return NULL;
1897
1898         root->d_fsdata = oe;
1899
1900         if (upperdentry) {
1901                 /* Root inode uses upper st_ino/i_ino */
1902                 ino = d_inode(upperdentry)->i_ino;
1903                 fsid = 0;
1904                 ovl_dentry_set_upper_alias(root);
1905                 if (ovl_is_impuredir(sb, upperdentry))
1906                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1907         }
1908
1909         /* Root is always merge -> can have whiteouts */
1910         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1911         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1912         ovl_set_upperdata(d_inode(root));
1913         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1914         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1915
1916         return root;
1917 }
1918
1919 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1920 {
1921         struct path upperpath = { };
1922         struct dentry *root_dentry;
1923         struct ovl_entry *oe;
1924         struct ovl_fs *ofs;
1925         struct ovl_layer *layers;
1926         struct cred *cred;
1927         char *splitlower = NULL;
1928         unsigned int numlower;
1929         int err;
1930
1931         sb->s_d_op = &ovl_dentry_operations;
1932
1933         err = -ENOMEM;
1934         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1935         if (!ofs)
1936                 goto out;
1937
1938         ofs->creator_cred = cred = prepare_creds();
1939         if (!cred)
1940                 goto out_err;
1941
1942         /* Is there a reason anyone would want not to share whiteouts? */
1943         ofs->share_whiteout = true;
1944
1945         ofs->config.index = ovl_index_def;
1946         ofs->config.uuid = true;
1947         ofs->config.nfs_export = ovl_nfs_export_def;
1948         ofs->config.xino = ovl_xino_def();
1949         ofs->config.metacopy = ovl_metacopy_def;
1950         err = ovl_parse_opt((char *) data, &ofs->config);
1951         if (err)
1952                 goto out_err;
1953
1954         err = -EINVAL;
1955         if (!ofs->config.lowerdir) {
1956                 if (!silent)
1957                         pr_err("missing 'lowerdir'\n");
1958                 goto out_err;
1959         }
1960
1961         err = -ENOMEM;
1962         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1963         if (!splitlower)
1964                 goto out_err;
1965
1966         numlower = ovl_split_lowerdirs(splitlower);
1967         if (numlower > OVL_MAX_STACK) {
1968                 pr_err("too many lower directories, limit is %d\n",
1969                        OVL_MAX_STACK);
1970                 goto out_err;
1971         }
1972
1973         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1974         if (!layers)
1975                 goto out_err;
1976
1977         ofs->layers = layers;
1978         /* Layer 0 is reserved for upper even if there's no upper */
1979         ofs->numlayer = 1;
1980
1981         sb->s_stack_depth = 0;
1982         sb->s_maxbytes = MAX_LFS_FILESIZE;
1983         atomic_long_set(&ofs->last_ino, 1);
1984         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1985         if (ofs->config.xino != OVL_XINO_OFF) {
1986                 ofs->xino_mode = BITS_PER_LONG - 32;
1987                 if (!ofs->xino_mode) {
1988                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1989                         ofs->config.xino = OVL_XINO_OFF;
1990                 }
1991         }
1992
1993         /* alloc/destroy_inode needed for setting up traps in inode cache */
1994         sb->s_op = &ovl_super_operations;
1995
1996         if (ofs->config.upperdir) {
1997                 if (!ofs->config.workdir) {
1998                         pr_err("missing 'workdir'\n");
1999                         goto out_err;
2000                 }
2001
2002                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
2003                 if (err)
2004                         goto out_err;
2005
2006                 err = ovl_get_workdir(sb, ofs, &upperpath);
2007                 if (err)
2008                         goto out_err;
2009
2010                 if (!ofs->workdir)
2011                         sb->s_flags |= SB_RDONLY;
2012
2013                 sb->s_stack_depth = ovl_upper_mnt(ofs)->mnt_sb->s_stack_depth;
2014                 sb->s_time_gran = ovl_upper_mnt(ofs)->mnt_sb->s_time_gran;
2015
2016         }
2017         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2018         err = PTR_ERR(oe);
2019         if (IS_ERR(oe))
2020                 goto out_err;
2021
2022         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2023         if (!ovl_upper_mnt(ofs))
2024                 sb->s_flags |= SB_RDONLY;
2025
2026         if (!ofs->config.uuid && ofs->numfs > 1) {
2027                 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2028                 ofs->config.uuid = true;
2029         }
2030
2031         if (!ovl_force_readonly(ofs) && ofs->config.index) {
2032                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2033                 if (err)
2034                         goto out_free_oe;
2035
2036                 /* Force r/o mount with no index dir */
2037                 if (!ofs->indexdir)
2038                         sb->s_flags |= SB_RDONLY;
2039         }
2040
2041         err = ovl_check_overlapping_layers(sb, ofs);
2042         if (err)
2043                 goto out_free_oe;
2044
2045         /* Show index=off in /proc/mounts for forced r/o mount */
2046         if (!ofs->indexdir) {
2047                 ofs->config.index = false;
2048                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2049                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2050                         ofs->config.nfs_export = false;
2051                 }
2052         }
2053
2054         if (ofs->config.metacopy && ofs->config.nfs_export) {
2055                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2056                 ofs->config.nfs_export = false;
2057         }
2058
2059         if (ofs->config.nfs_export)
2060                 sb->s_export_op = &ovl_export_operations;
2061
2062         /* Never override disk quota limits or use reserved space */
2063         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2064
2065         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2066         sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2067                 ovl_trusted_xattr_handlers;
2068         sb->s_fs_info = ofs;
2069         sb->s_flags |= SB_POSIXACL;
2070         sb->s_iflags |= SB_I_SKIP_SYNC;
2071
2072         err = -ENOMEM;
2073         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2074         if (!root_dentry)
2075                 goto out_free_oe;
2076
2077         mntput(upperpath.mnt);
2078         kfree(splitlower);
2079
2080         sb->s_root = root_dentry;
2081
2082         return 0;
2083
2084 out_free_oe:
2085         ovl_entry_stack_free(oe);
2086         kfree(oe);
2087 out_err:
2088         kfree(splitlower);
2089         path_put(&upperpath);
2090         ovl_free_fs(ofs);
2091 out:
2092         return err;
2093 }
2094
2095 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2096                                 const char *dev_name, void *raw_data)
2097 {
2098         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2099 }
2100
2101 static struct file_system_type ovl_fs_type = {
2102         .owner          = THIS_MODULE,
2103         .name           = "overlay",
2104         .fs_flags       = FS_USERNS_MOUNT,
2105         .mount          = ovl_mount,
2106         .kill_sb        = kill_anon_super,
2107 };
2108 MODULE_ALIAS_FS("overlay");
2109
2110 static void ovl_inode_init_once(void *foo)
2111 {
2112         struct ovl_inode *oi = foo;
2113
2114         inode_init_once(&oi->vfs_inode);
2115 }
2116
2117 static int __init ovl_init(void)
2118 {
2119         int err;
2120
2121         ovl_inode_cachep = kmem_cache_create("ovl_inode",
2122                                              sizeof(struct ovl_inode), 0,
2123                                              (SLAB_RECLAIM_ACCOUNT|
2124                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2125                                              ovl_inode_init_once);
2126         if (ovl_inode_cachep == NULL)
2127                 return -ENOMEM;
2128
2129         err = ovl_aio_request_cache_init();
2130         if (!err) {
2131                 err = register_filesystem(&ovl_fs_type);
2132                 if (!err)
2133                         return 0;
2134
2135                 ovl_aio_request_cache_destroy();
2136         }
2137         kmem_cache_destroy(ovl_inode_cachep);
2138
2139         return err;
2140 }
2141
2142 static void __exit ovl_exit(void)
2143 {
2144         unregister_filesystem(&ovl_fs_type);
2145
2146         /*
2147          * Make sure all delayed rcu free inodes are flushed before we
2148          * destroy cache.
2149          */
2150         rcu_barrier();
2151         kmem_cache_destroy(ovl_inode_cachep);
2152         ovl_aio_request_cache_destroy();
2153 }
2154
2155 module_init(ovl_init);
2156 module_exit(ovl_exit);