fs/overlayfs/super.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *
   4  * Copyright (C) 2011 Novell Inc.
   5  */
   6
   7 #include <uapi/linux/magic.h>
   8 #include <linux/fs.h>
   9 #include <linux/namei.h>
  10 #include <linux/xattr.h>
  11 #include <linux/mount.h>
  12 #include <linux/parser.h>
  13 #include <linux/module.h>
  14 #include <linux/statfs.h>
  15 #include <linux/seq_file.h>
  16 #include <linux/posix_acl_xattr.h>
  17 #include <linux/exportfs.h>
  18 #include "overlayfs.h"
  19
  20 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  21 MODULE_DESCRIPTION("Overlay filesystem");
  22 MODULE_LICENSE("GPL");
  23
  24
  25 struct ovl_dir_cache;
  26
  27 #define OVL_MAX_STACK 500
  28
  29 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  30 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  31 MODULE_PARM_DESC(redirect_dir,
  32                  "Default to on or off for the redirect_dir feature");
  33
  34 static bool ovl_redirect_always_follow =
  35         IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
  36 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
  37                    bool, 0644);
  38 MODULE_PARM_DESC(redirect_always_follow,
  39                  "Follow redirects even if redirect_dir feature is turned off");
  40
  41 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
  42 module_param_named(index, ovl_index_def, bool, 0644);
  43 MODULE_PARM_DESC(index,
  44                  "Default to on or off for the inodes index feature");
  45
  46 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
  47 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
  48 MODULE_PARM_DESC(nfs_export,
  49                  "Default to on or off for the NFS export feature");
  50
  51 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
  52 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
  53 MODULE_PARM_DESC(xino_auto,
  54                  "Auto enable xino feature");
  55
  56 static void ovl_entry_stack_free(struct ovl_entry *oe)
  57 {
  58         unsigned int i;
  59
  60         for (i = 0; i < oe->numlower; i++)
  61                 dput(oe->lowerstack[i].dentry);
  62 }
  63
  64 static bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
  65 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
  66 MODULE_PARM_DESC(metacopy,
  67                  "Default to on or off for the metadata only copy up feature");
  68
  69 static void ovl_dentry_release(struct dentry *dentry)
  70 {
  71         struct ovl_entry *oe = dentry->d_fsdata;
  72
  73         if (oe) {
  74                 ovl_entry_stack_free(oe);
  75                 kfree_rcu(oe, rcu);
  76         }
  77 }
  78
  79 static struct dentry *ovl_d_real(struct dentry *dentry,
  80                                  const struct inode *inode)
  81 {
  82         struct dentry *real = NULL, *lower;
  83
  84         /* It's an overlay file */
  85         if (inode && d_inode(dentry) == inode)
  86                 return dentry;
  87
  88         if (!d_is_reg(dentry)) {
  89                 if (!inode || inode == d_inode(dentry))
  90                         return dentry;
  91                 goto bug;
  92         }
  93
  94         real = ovl_dentry_upper(dentry);
  95         if (real && (inode == d_inode(real)))
  96                 return real;
  97
  98         if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
  99                 return real;
 100
 101         lower = ovl_dentry_lowerdata(dentry);
 102         if (!lower)
 103                 goto bug;
 104         real = lower;
 105
 106         /* Handle recursion */
 107         real = d_real(real, inode);
 108
 109         if (!inode || inode == d_inode(real))
 110                 return real;
 111 bug:
 112         WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
 113              __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
 114              inode ? inode->i_ino : 0, real,
 115              real && d_inode(real) ? d_inode(real)->i_ino : 0);
 116         return dentry;
 117 }
 118
 119 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
 120 {
 121         int ret = 1;
 122
 123         if (weak) {
 124                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
 125                         ret =  d->d_op->d_weak_revalidate(d, flags);
 126         } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
 127                 ret = d->d_op->d_revalidate(d, flags);
 128                 if (!ret) {
 129                         if (!(flags & LOOKUP_RCU))
 130                                 d_invalidate(d);
 131                         ret = -ESTALE;
 132                 }
 133         }
 134         return ret;
 135 }
 136
 137 static int ovl_dentry_revalidate_common(struct dentry *dentry,
 138                                         unsigned int flags, bool weak)
 139 {
 140         struct ovl_entry *oe = dentry->d_fsdata;
 141         struct dentry *upper;
 142         unsigned int i;
 143         int ret = 1;
 144
 145         upper = ovl_dentry_upper(dentry);
 146         if (upper)
 147                 ret = ovl_revalidate_real(upper, flags, weak);
 148
 149         for (i = 0; ret > 0 && i < oe->numlower; i++) {
 150                 ret = ovl_revalidate_real(oe->lowerstack[i].dentry, flags,
 151                                           weak);
 152         }
 153         return ret;
 154 }
 155
 156 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 157 {
 158         return ovl_dentry_revalidate_common(dentry, flags, false);
 159 }
 160
 161 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 162 {
 163         return ovl_dentry_revalidate_common(dentry, flags, true);
 164 }
 165
 166 static const struct dentry_operations ovl_dentry_operations = {
 167         .d_release = ovl_dentry_release,
 168         .d_real = ovl_d_real,
 169         .d_revalidate = ovl_dentry_revalidate,
 170         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 171 };
 172
 173 static struct kmem_cache *ovl_inode_cachep;
 174
 175 static struct inode *ovl_alloc_inode(struct super_block *sb)
 176 {
 177         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 178
 179         if (!oi)
 180                 return NULL;
 181
 182         oi->cache = NULL;
 183         oi->redirect = NULL;
 184         oi->version = 0;
 185         oi->flags = 0;
 186         oi->__upperdentry = NULL;
 187         oi->lower = NULL;
 188         oi->lowerdata = NULL;
 189         mutex_init(&oi->lock);
 190
 191         return &oi->vfs_inode;
 192 }
 193
 194 static void ovl_free_inode(struct inode *inode)
 195 {
 196         struct ovl_inode *oi = OVL_I(inode);
 197
 198         kfree(oi->redirect);
 199         mutex_destroy(&oi->lock);
 200         kmem_cache_free(ovl_inode_cachep, oi);
 201 }
 202
 203 static void ovl_destroy_inode(struct inode *inode)
 204 {
 205         struct ovl_inode *oi = OVL_I(inode);
 206
 207         dput(oi->__upperdentry);
 208         iput(oi->lower);
 209         if (S_ISDIR(inode->i_mode))
 210                 ovl_dir_cache_free(inode);
 211         else
 212                 iput(oi->lowerdata);
 213 }
 214
 215 static void ovl_free_fs(struct ovl_fs *ofs)
 216 {
 217         struct vfsmount **mounts;
 218         unsigned i;
 219
 220         iput(ofs->workbasedir_trap);
 221         iput(ofs->indexdir_trap);
 222         iput(ofs->workdir_trap);
 223         dput(ofs->whiteout);
 224         dput(ofs->indexdir);
 225         dput(ofs->workdir);
 226         if (ofs->workdir_locked)
 227                 ovl_inuse_unlock(ofs->workbasedir);
 228         dput(ofs->workbasedir);
 229         if (ofs->upperdir_locked)
 230                 ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
 231
 232         /* Hack!  Reuse ofs->layers as a vfsmount array before freeing it */
 233         mounts = (struct vfsmount **) ofs->layers;
 234         for (i = 0; i < ofs->numlayer; i++) {
 235                 iput(ofs->layers[i].trap);
 236                 mounts[i] = ofs->layers[i].mnt;
 237         }
 238         kern_unmount_array(mounts, ofs->numlayer);
 239         kfree(ofs->layers);
 240         for (i = 0; i < ofs->numfs; i++)
 241                 free_anon_bdev(ofs->fs[i].pseudo_dev);
 242         kfree(ofs->fs);
 243
 244         kfree(ofs->config.lowerdir);
 245         kfree(ofs->config.upperdir);
 246         kfree(ofs->config.workdir);
 247         kfree(ofs->config.redirect_mode);
 248         if (ofs->creator_cred)
 249                 put_cred(ofs->creator_cred);
 250         kfree(ofs);
 251 }
 252
 253 static void ovl_put_super(struct super_block *sb)
 254 {
 255         struct ovl_fs *ofs = sb->s_fs_info;
 256
 257         ovl_free_fs(ofs);
 258 }
 259
 260 /* Sync real dirty inodes in upper filesystem (if it exists) */
 261 static int ovl_sync_fs(struct super_block *sb, int wait)
 262 {
 263         struct ovl_fs *ofs = sb->s_fs_info;
 264         struct super_block *upper_sb;
 265         int ret;
 266
 267         ret = ovl_sync_status(ofs);
 268         /*
 269          * We have to always set the err, because the return value isn't
 270          * checked in syncfs, and instead indirectly return an error via
 271          * the sb's writeback errseq, which VFS inspects after this call.
 272          */
 273         if (ret < 0) {
 274                 errseq_set(&sb->s_wb_err, -EIO);
 275                 return -EIO;
 276         }
 277
 278         if (!ret)
 279                 return ret;
 280
 281         /*
 282          * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
 283          * All the super blocks will be iterated, including upper_sb.
 284          *
 285          * If this is a syncfs(2) call, then we do need to call
 286          * sync_filesystem() on upper_sb, but enough if we do it when being
 287          * called with wait == 1.
 288          */
 289         if (!wait)
 290                 return 0;
 291
 292         upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 293
 294         down_read(&upper_sb->s_umount);
 295         ret = sync_filesystem(upper_sb);
 296         up_read(&upper_sb->s_umount);
 297
 298         return ret;
 299 }
 300
 301 /**
 302  * ovl_statfs
 303  * @sb: The overlayfs super block
 304  * @buf: The struct kstatfs to fill in with stats
 305  *
 306  * Get the filesystem statistics.  As writes always target the upper layer
 307  * filesystem pass the statfs to the upper filesystem (if it exists)
 308  */
 309 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 310 {
 311         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 312         struct dentry *root_dentry = dentry->d_sb->s_root;
 313         struct path path;
 314         int err;
 315
 316         ovl_path_real(root_dentry, &path);
 317
 318         err = vfs_statfs(&path, buf);
 319         if (!err) {
 320                 buf->f_namelen = ofs->namelen;
 321                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 322         }
 323
 324         return err;
 325 }
 326
 327 /* Will this overlay be forced to mount/remount ro? */
 328 static bool ovl_force_readonly(struct ovl_fs *ofs)
 329 {
 330         return (!ovl_upper_mnt(ofs) || !ofs->workdir);
 331 }
 332
 333 static const char *ovl_redirect_mode_def(void)
 334 {
 335         return ovl_redirect_dir_def ? "on" : "off";
 336 }
 337
 338 static const char * const ovl_xino_str[] = {
 339         "off",
 340         "auto",
 341         "on",
 342 };
 343
 344 static inline int ovl_xino_def(void)
 345 {
 346         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 347 }
 348
 349 /**
 350  * ovl_show_options
 351  *
 352  * Prints the mount options for a given superblock.
 353  * Returns zero; does not fail.
 354  */
 355 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 356 {
 357         struct super_block *sb = dentry->d_sb;
 358         struct ovl_fs *ofs = sb->s_fs_info;
 359
 360         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 361         if (ofs->config.upperdir) {
 362                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 363                 seq_show_option(m, "workdir", ofs->config.workdir);
 364         }
 365         if (ofs->config.default_permissions)
 366                 seq_puts(m, ",default_permissions");
 367         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 368                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 369         if (ofs->config.index != ovl_index_def)
 370                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 371         if (!ofs->config.uuid)
 372                 seq_puts(m, ",uuid=off");
 373         if (ofs->config.nfs_export != ovl_nfs_export_def)
 374                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 375                                                 "on" : "off");
 376         if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
 377                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 378         if (ofs->config.metacopy != ovl_metacopy_def)
 379                 seq_printf(m, ",metacopy=%s",
 380                            ofs->config.metacopy ? "on" : "off");
 381         if (ofs->config.ovl_volatile)
 382                 seq_puts(m, ",volatile");
 383         return 0;
 384 }
 385
 386 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 387 {
 388         struct ovl_fs *ofs = sb->s_fs_info;
 389         struct super_block *upper_sb;
 390         int ret = 0;
 391
 392         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 393                 return -EROFS;
 394
 395         if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
 396                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 397                 if (ovl_should_sync(ofs)) {
 398                         down_read(&upper_sb->s_umount);
 399                         ret = sync_filesystem(upper_sb);
 400                         up_read(&upper_sb->s_umount);
 401                 }
 402         }
 403
 404         return ret;
 405 }
 406
 407 static const struct super_operations ovl_super_operations = {
 408         .alloc_inode    = ovl_alloc_inode,
 409         .free_inode     = ovl_free_inode,
 410         .destroy_inode  = ovl_destroy_inode,
 411         .drop_inode     = generic_delete_inode,
 412         .put_super      = ovl_put_super,
 413         .sync_fs        = ovl_sync_fs,
 414         .statfs         = ovl_statfs,
 415         .show_options   = ovl_show_options,
 416         .remount_fs     = ovl_remount,
 417 };
 418
 419 enum {
 420         OPT_LOWERDIR,
 421         OPT_UPPERDIR,
 422         OPT_WORKDIR,
 423         OPT_DEFAULT_PERMISSIONS,
 424         OPT_REDIRECT_DIR,
 425         OPT_INDEX_ON,
 426         OPT_INDEX_OFF,
 427         OPT_UUID_ON,
 428         OPT_UUID_OFF,
 429         OPT_NFS_EXPORT_ON,
 430         OPT_USERXATTR,
 431         OPT_NFS_EXPORT_OFF,
 432         OPT_XINO_ON,
 433         OPT_XINO_OFF,
 434         OPT_XINO_AUTO,
 435         OPT_METACOPY_ON,
 436         OPT_METACOPY_OFF,
 437         OPT_VOLATILE,
 438         OPT_ERR,
 439 };
 440
 441 static const match_table_t ovl_tokens = {
 442         {OPT_LOWERDIR,                  "lowerdir=%s"},
 443         {OPT_UPPERDIR,                  "upperdir=%s"},
 444         {OPT_WORKDIR,                   "workdir=%s"},
 445         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 446         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 447         {OPT_INDEX_ON,                  "index=on"},
 448         {OPT_INDEX_OFF,                 "index=off"},
 449         {OPT_USERXATTR,                 "userxattr"},
 450         {OPT_UUID_ON,                   "uuid=on"},
 451         {OPT_UUID_OFF,                  "uuid=off"},
 452         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 453         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 454         {OPT_XINO_ON,                   "xino=on"},
 455         {OPT_XINO_OFF,                  "xino=off"},
 456         {OPT_XINO_AUTO,                 "xino=auto"},
 457         {OPT_METACOPY_ON,               "metacopy=on"},
 458         {OPT_METACOPY_OFF,              "metacopy=off"},
 459         {OPT_VOLATILE,                  "volatile"},
 460         {OPT_ERR,                       NULL}
 461 };
 462
 463 static char *ovl_next_opt(char **s)
 464 {
 465         char *sbegin = *s;
 466         char *p;
 467
 468         if (sbegin == NULL)
 469                 return NULL;
 470
 471         for (p = sbegin; *p; p++) {
 472                 if (*p == '\\') {
 473                         p++;
 474                         if (!*p)
 475                                 break;
 476                 } else if (*p == ',') {
 477                         *p = '\0';
 478                         *s = p + 1;
 479                         return sbegin;
 480                 }
 481         }
 482         *s = NULL;
 483         return sbegin;
 484 }
 485
 486 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 487 {
 488         if (strcmp(mode, "on") == 0) {
 489                 config->redirect_dir = true;
 490                 /*
 491                  * Does not make sense to have redirect creation without
 492                  * redirect following.
 493                  */
 494                 config->redirect_follow = true;
 495         } else if (strcmp(mode, "follow") == 0) {
 496                 config->redirect_follow = true;
 497         } else if (strcmp(mode, "off") == 0) {
 498                 if (ovl_redirect_always_follow)
 499                         config->redirect_follow = true;
 500         } else if (strcmp(mode, "nofollow") != 0) {
 501                 pr_err("bad mount option \"redirect_dir=%s\"\n",
 502                        mode);
 503                 return -EINVAL;
 504         }
 505
 506         return 0;
 507 }
 508
 509 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 510 {
 511         char *p;
 512         int err;
 513         bool metacopy_opt = false, redirect_opt = false;
 514         bool nfs_export_opt = false, index_opt = false;
 515
 516         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 517         if (!config->redirect_mode)
 518                 return -ENOMEM;
 519
 520         while ((p = ovl_next_opt(&opt)) != NULL) {
 521                 int token;
 522                 substring_t args[MAX_OPT_ARGS];
 523
 524                 if (!*p)
 525                         continue;
 526
 527                 token = match_token(p, ovl_tokens, args);
 528                 switch (token) {
 529                 case OPT_UPPERDIR:
 530                         kfree(config->upperdir);
 531                         config->upperdir = match_strdup(&args[0]);
 532                         if (!config->upperdir)
 533                                 return -ENOMEM;
 534                         break;
 535
 536                 case OPT_LOWERDIR:
 537                         kfree(config->lowerdir);
 538                         config->lowerdir = match_strdup(&args[0]);
 539                         if (!config->lowerdir)
 540                                 return -ENOMEM;
 541                         break;
 542
 543                 case OPT_WORKDIR:
 544                         kfree(config->workdir);
 545                         config->workdir = match_strdup(&args[0]);
 546                         if (!config->workdir)
 547                                 return -ENOMEM;
 548                         break;
 549
 550                 case OPT_DEFAULT_PERMISSIONS:
 551                         config->default_permissions = true;
 552                         break;
 553
 554                 case OPT_REDIRECT_DIR:
 555                         kfree(config->redirect_mode);
 556                         config->redirect_mode = match_strdup(&args[0]);
 557                         if (!config->redirect_mode)
 558                                 return -ENOMEM;
 559                         redirect_opt = true;
 560                         break;
 561
 562                 case OPT_INDEX_ON:
 563                         config->index = true;
 564                         index_opt = true;
 565                         break;
 566
 567                 case OPT_INDEX_OFF:
 568                         config->index = false;
 569                         index_opt = true;
 570                         break;
 571
 572                 case OPT_UUID_ON:
 573                         config->uuid = true;
 574                         break;
 575
 576                 case OPT_UUID_OFF:
 577                         config->uuid = false;
 578                         break;
 579
 580                 case OPT_NFS_EXPORT_ON:
 581                         config->nfs_export = true;
 582                         nfs_export_opt = true;
 583                         break;
 584
 585                 case OPT_NFS_EXPORT_OFF:
 586                         config->nfs_export = false;
 587                         nfs_export_opt = true;
 588                         break;
 589
 590                 case OPT_XINO_ON:
 591                         config->xino = OVL_XINO_ON;
 592                         break;
 593
 594                 case OPT_XINO_OFF:
 595                         config->xino = OVL_XINO_OFF;
 596                         break;
 597
 598                 case OPT_XINO_AUTO:
 599                         config->xino = OVL_XINO_AUTO;
 600                         break;
 601
 602                 case OPT_METACOPY_ON:
 603                         config->metacopy = true;
 604                         metacopy_opt = true;
 605                         break;
 606
 607                 case OPT_METACOPY_OFF:
 608                         config->metacopy = false;
 609                         metacopy_opt = true;
 610                         break;
 611
 612                 case OPT_VOLATILE:
 613                         config->ovl_volatile = true;
 614                         break;
 615
 616                 case OPT_USERXATTR:
 617                         config->userxattr = true;
 618                         break;
 619
 620                 default:
 621                         pr_err("unrecognized mount option \"%s\" or missing value\n",
 622                                         p);
 623                         return -EINVAL;
 624                 }
 625         }
 626
 627         /* Workdir/index are useless in non-upper mount */
 628         if (!config->upperdir) {
 629                 if (config->workdir) {
 630                         pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 631                                 config->workdir);
 632                         kfree(config->workdir);
 633                         config->workdir = NULL;
 634                 }
 635                 if (config->index && index_opt) {
 636                         pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
 637                         index_opt = false;
 638                 }
 639                 config->index = false;
 640         }
 641
 642         if (!config->upperdir && config->ovl_volatile) {
 643                 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
 644                 config->ovl_volatile = false;
 645         }
 646
 647         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 648         if (err)
 649                 return err;
 650
 651         /*
 652          * This is to make the logic below simpler.  It doesn't make any other
 653          * difference, since config->redirect_dir is only used for upper.
 654          */
 655         if (!config->upperdir && config->redirect_follow)
 656                 config->redirect_dir = true;
 657
 658         /* Resolve metacopy -> redirect_dir dependency */
 659         if (config->metacopy && !config->redirect_dir) {
 660                 if (metacopy_opt && redirect_opt) {
 661                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
 662                                config->redirect_mode);
 663                         return -EINVAL;
 664                 }
 665                 if (redirect_opt) {
 666                         /*
 667                          * There was an explicit redirect_dir=... that resulted
 668                          * in this conflict.
 669                          */
 670                         pr_info("disabling metacopy due to redirect_dir=%s\n",
 671                                 config->redirect_mode);
 672                         config->metacopy = false;
 673                 } else {
 674                         /* Automatically enable redirect otherwise. */
 675                         config->redirect_follow = config->redirect_dir = true;
 676                 }
 677         }
 678
 679         /* Resolve nfs_export -> index dependency */
 680         if (config->nfs_export && !config->index) {
 681                 if (!config->upperdir && config->redirect_follow) {
 682                         pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
 683                         config->nfs_export = false;
 684                 } else if (nfs_export_opt && index_opt) {
 685                         pr_err("conflicting options: nfs_export=on,index=off\n");
 686                         return -EINVAL;
 687                 } else if (index_opt) {
 688                         /*
 689                          * There was an explicit index=off that resulted
 690                          * in this conflict.
 691                          */
 692                         pr_info("disabling nfs_export due to index=off\n");
 693                         config->nfs_export = false;
 694                 } else {
 695                         /* Automatically enable index otherwise. */
 696                         config->index = true;
 697                 }
 698         }
 699
 700         /* Resolve nfs_export -> !metacopy dependency */
 701         if (config->nfs_export && config->metacopy) {
 702                 if (nfs_export_opt && metacopy_opt) {
 703                         pr_err("conflicting options: nfs_export=on,metacopy=on\n");
 704                         return -EINVAL;
 705                 }
 706                 if (metacopy_opt) {
 707                         /*
 708                          * There was an explicit metacopy=on that resulted
 709                          * in this conflict.
 710                          */
 711                         pr_info("disabling nfs_export due to metacopy=on\n");
 712                         config->nfs_export = false;
 713                 } else {
 714                         /*
 715                          * There was an explicit nfs_export=on that resulted
 716                          * in this conflict.
 717                          */
 718                         pr_info("disabling metacopy due to nfs_export=on\n");
 719                         config->metacopy = false;
 720                 }
 721         }
 722
 723
 724         /* Resolve userxattr -> !redirect && !metacopy dependency */
 725         if (config->userxattr) {
 726                 if (config->redirect_follow && redirect_opt) {
 727                         pr_err("conflicting options: userxattr,redirect_dir=%s\n",
 728                                config->redirect_mode);
 729                         return -EINVAL;
 730                 }
 731                 if (config->metacopy && metacopy_opt) {
 732                         pr_err("conflicting options: userxattr,metacopy=on\n");
 733                         return -EINVAL;
 734                 }
 735                 /*
 736                  * Silently disable default setting of redirect and metacopy.
 737                  * This shall be the default in the future as well: these
 738                  * options must be explicitly enabled if used together with
 739                  * userxattr.
 740                  */
 741                 config->redirect_dir = config->redirect_follow = false;
 742                 config->metacopy = false;
 743         }
 744
 745         return 0;
 746 }
 747
 748 #define OVL_WORKDIR_NAME "work"
 749 #define OVL_INDEXDIR_NAME "index"
 750
 751 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 752                                          const char *name, bool persist)
 753 {
 754         struct inode *dir =  ofs->workbasedir->d_inode;
 755         struct vfsmount *mnt = ovl_upper_mnt(ofs);
 756         struct dentry *work;
 757         int err;
 758         bool retried = false;
 759
 760         inode_lock_nested(dir, I_MUTEX_PARENT);
 761 retry:
 762         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
 763
 764         if (!IS_ERR(work)) {
 765                 struct iattr attr = {
 766                         .ia_valid = ATTR_MODE,
 767                         .ia_mode = S_IFDIR | 0,
 768                 };
 769
 770                 if (work->d_inode) {
 771                         err = -EEXIST;
 772                         if (retried)
 773                                 goto out_dput;
 774
 775                         if (persist)
 776                                 goto out_unlock;
 777
 778                         retried = true;
 779                         err = ovl_workdir_cleanup(dir, mnt, work, 0);
 780                         dput(work);
 781                         if (err == -EINVAL) {
 782                                 work = ERR_PTR(err);
 783                                 goto out_unlock;
 784                         }
 785                         goto retry;
 786                 }
 787
 788                 work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
 789                 err = PTR_ERR(work);
 790                 if (IS_ERR(work))
 791                         goto out_err;
 792
 793                 /*
 794                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 795                  *
 796                  * a) success (there was a POSIX ACL xattr and was removed)
 797                  * b) -ENODATA (there was no POSIX ACL xattr)
 798                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 799                  *
 800                  * There are various other error values that could effectively
 801                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 802                  * if the xattr name is too long), but the set of filesystems
 803                  * allowed as upper are limited to "normal" ones, where checking
 804                  * for the above two errors is sufficient.
 805                  */
 806                 err = vfs_removexattr(&init_user_ns, work,
 807                                       XATTR_NAME_POSIX_ACL_DEFAULT);
 808                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 809                         goto out_dput;
 810
 811                 err = vfs_removexattr(&init_user_ns, work,
 812                                       XATTR_NAME_POSIX_ACL_ACCESS);
 813                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 814                         goto out_dput;
 815
 816                 /* Clear any inherited mode bits */
 817                 inode_lock(work->d_inode);
 818                 err = notify_change(&init_user_ns, work, &attr, NULL);
 819                 inode_unlock(work->d_inode);
 820                 if (err)
 821                         goto out_dput;
 822         } else {
 823                 err = PTR_ERR(work);
 824                 goto out_err;
 825         }
 826 out_unlock:
 827         inode_unlock(dir);
 828         return work;
 829
 830 out_dput:
 831         dput(work);
 832 out_err:
 833         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
 834                 ofs->config.workdir, name, -err);
 835         work = NULL;
 836         goto out_unlock;
 837 }
 838
 839 static void ovl_unescape(char *s)
 840 {
 841         char *d = s;
 842
 843         for (;; s++, d++) {
 844                 if (*s == '\\')
 845                         s++;
 846                 *d = *s;
 847                 if (!*s)
 848                         break;
 849         }
 850 }
 851
 852 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 853 {
 854         int err = -EINVAL;
 855
 856         if (!*name) {
 857                 pr_err("empty lowerdir\n");
 858                 goto out;
 859         }
 860         err = kern_path(name, LOOKUP_FOLLOW, path);
 861         if (err) {
 862                 pr_err("failed to resolve '%s': %i\n", name, err);
 863                 goto out;
 864         }
 865         err = -EINVAL;
 866         if (ovl_dentry_weird(path->dentry)) {
 867                 pr_err("filesystem on '%s' not supported\n", name);
 868                 goto out_put;
 869         }
 870         if (mnt_user_ns(path->mnt) != &init_user_ns) {
 871                 pr_err("idmapped layers are currently not supported\n");
 872                 goto out_put;
 873         }
 874         if (!d_is_dir(path->dentry)) {
 875                 pr_err("'%s' not a directory\n", name);
 876                 goto out_put;
 877         }
 878         return 0;
 879
 880 out_put:
 881         path_put_init(path);
 882 out:
 883         return err;
 884 }
 885
 886 static int ovl_mount_dir(const char *name, struct path *path)
 887 {
 888         int err = -ENOMEM;
 889         char *tmp = kstrdup(name, GFP_KERNEL);
 890
 891         if (tmp) {
 892                 ovl_unescape(tmp);
 893                 err = ovl_mount_dir_noesc(tmp, path);
 894
 895                 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
 896                         pr_err("filesystem on '%s' not supported as upperdir\n",
 897                                tmp);
 898                         path_put_init(path);
 899                         err = -EINVAL;
 900                 }
 901                 kfree(tmp);
 902         }
 903         return err;
 904 }
 905
 906 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 907                              const char *name)
 908 {
 909         struct kstatfs statfs;
 910         int err = vfs_statfs(path, &statfs);
 911
 912         if (err)
 913                 pr_err("statfs failed on '%s'\n", name);
 914         else
 915                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 916
 917         return err;
 918 }
 919
 920 static int ovl_lower_dir(const char *name, struct path *path,
 921                          struct ovl_fs *ofs, int *stack_depth)
 922 {
 923         int fh_type;
 924         int err;
 925
 926         err = ovl_mount_dir_noesc(name, path);
 927         if (err)
 928                 return err;
 929
 930         err = ovl_check_namelen(path, ofs, name);
 931         if (err)
 932                 return err;
 933
 934         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 935
 936         /*
 937          * The inodes index feature and NFS export need to encode and decode
 938          * file handles, so they require that all layers support them.
 939          */
 940         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 941         if ((ofs->config.nfs_export ||
 942              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 943                 ofs->config.index = false;
 944                 ofs->config.nfs_export = false;
 945                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 946                         name);
 947         }
 948
 949         /* Check if lower fs has 32bit inode numbers */
 950         if (fh_type != FILEID_INO32_GEN)
 951                 ofs->xino_mode = -1;
 952
 953         return 0;
 954 }
 955
 956 /* Workdir should not be subdir of upperdir and vice versa */
 957 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 958 {
 959         bool ok = false;
 960
 961         if (workdir != upperdir) {
 962                 ok = (lock_rename(workdir, upperdir) == NULL);
 963                 unlock_rename(workdir, upperdir);
 964         }
 965         return ok;
 966 }
 967
 968 static unsigned int ovl_split_lowerdirs(char *str)
 969 {
 970         unsigned int ctr = 1;
 971         char *s, *d;
 972
 973         for (s = d = str;; s++, d++) {
 974                 if (*s == '\\') {
 975                         s++;
 976                 } else if (*s == ':') {
 977                         *d = '\0';
 978                         ctr++;
 979                         continue;
 980                 }
 981                 *d = *s;
 982                 if (!*s)
 983                         break;
 984         }
 985         return ctr;
 986 }
 987
 988 static int __maybe_unused
 989 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 990                         struct dentry *dentry, struct inode *inode,
 991                         const char *name, void *buffer, size_t size)
 992 {
 993         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 994 }
 995
 996 static int __maybe_unused
 997 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 998                         struct user_namespace *mnt_userns,
 999                         struct dentry *dentry, struct inode *inode,
1000                         const char *name, const void *value,
1001                         size_t size, int flags)
1002 {
1003         struct dentry *workdir = ovl_workdir(dentry);
1004         struct inode *realinode = ovl_inode_real(inode);
1005         struct posix_acl *acl = NULL;
1006         int err;
1007
1008         /* Check that everything is OK before copy-up */
1009         if (value) {
1010                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
1011                 if (IS_ERR(acl))
1012                         return PTR_ERR(acl);
1013         }
1014         err = -EOPNOTSUPP;
1015         if (!IS_POSIXACL(d_inode(workdir)))
1016                 goto out_acl_release;
1017         if (!realinode->i_op->set_acl)
1018                 goto out_acl_release;
1019         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1020                 err = acl ? -EACCES : 0;
1021                 goto out_acl_release;
1022         }
1023         err = -EPERM;
1024         if (!inode_owner_or_capable(&init_user_ns, inode))
1025                 goto out_acl_release;
1026
1027         posix_acl_release(acl);
1028
1029         /*
1030          * Check if sgid bit needs to be cleared (actual setacl operation will
1031          * be done with mounter's capabilities and so that won't do it for us).
1032          */
1033         if (unlikely(inode->i_mode & S_ISGID) &&
1034             handler->flags == ACL_TYPE_ACCESS &&
1035             !in_group_p(inode->i_gid) &&
1036             !capable_wrt_inode_uidgid(&init_user_ns, inode, CAP_FSETID)) {
1037                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
1038
1039                 err = ovl_setattr(&init_user_ns, dentry, &iattr);
1040                 if (err)
1041                         return err;
1042         }
1043
1044         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
1045         if (!err)
1046                 ovl_copyattr(ovl_inode_real(inode), inode);
1047
1048         return err;
1049
1050 out_acl_release:
1051         posix_acl_release(acl);
1052         return err;
1053 }
1054
1055 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1056                              struct dentry *dentry, struct inode *inode,
1057                              const char *name, void *buffer, size_t size)
1058 {
1059         return -EOPNOTSUPP;
1060 }
1061
1062 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1063                              struct user_namespace *mnt_userns,
1064                              struct dentry *dentry, struct inode *inode,
1065                              const char *name, const void *value,
1066                              size_t size, int flags)
1067 {
1068         return -EOPNOTSUPP;
1069 }
1070
1071 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1072                                struct dentry *dentry, struct inode *inode,
1073                                const char *name, void *buffer, size_t size)
1074 {
1075         return ovl_xattr_get(dentry, inode, name, buffer, size);
1076 }
1077
1078 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1079                                struct user_namespace *mnt_userns,
1080                                struct dentry *dentry, struct inode *inode,
1081                                const char *name, const void *value,
1082                                size_t size, int flags)
1083 {
1084         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1085 }
1086
1087 static const struct xattr_handler __maybe_unused
1088 ovl_posix_acl_access_xattr_handler = {
1089         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1090         .flags = ACL_TYPE_ACCESS,
1091         .get = ovl_posix_acl_xattr_get,
1092         .set = ovl_posix_acl_xattr_set,
1093 };
1094
1095 static const struct xattr_handler __maybe_unused
1096 ovl_posix_acl_default_xattr_handler = {
1097         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1098         .flags = ACL_TYPE_DEFAULT,
1099         .get = ovl_posix_acl_xattr_get,
1100         .set = ovl_posix_acl_xattr_set,
1101 };
1102
1103 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1104         .prefix = OVL_XATTR_TRUSTED_PREFIX,
1105         .get = ovl_own_xattr_get,
1106         .set = ovl_own_xattr_set,
1107 };
1108
1109 static const struct xattr_handler ovl_own_user_xattr_handler = {
1110         .prefix = OVL_XATTR_USER_PREFIX,
1111         .get = ovl_own_xattr_get,
1112         .set = ovl_own_xattr_set,
1113 };
1114
1115 static const struct xattr_handler ovl_other_xattr_handler = {
1116         .prefix = "", /* catch all */
1117         .get = ovl_other_xattr_get,
1118         .set = ovl_other_xattr_set,
1119 };
1120
1121 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1122 #ifdef CONFIG_FS_POSIX_ACL
1123         &ovl_posix_acl_access_xattr_handler,
1124         &ovl_posix_acl_default_xattr_handler,
1125 #endif
1126         &ovl_own_trusted_xattr_handler,
1127         &ovl_other_xattr_handler,
1128         NULL
1129 };
1130
1131 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1132 #ifdef CONFIG_FS_POSIX_ACL
1133         &ovl_posix_acl_access_xattr_handler,
1134         &ovl_posix_acl_default_xattr_handler,
1135 #endif
1136         &ovl_own_user_xattr_handler,
1137         &ovl_other_xattr_handler,
1138         NULL
1139 };
1140
1141 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1142                           struct inode **ptrap, const char *name)
1143 {
1144         struct inode *trap;
1145         int err;
1146
1147         trap = ovl_get_trap_inode(sb, dir);
1148         err = PTR_ERR_OR_ZERO(trap);
1149         if (err) {
1150                 if (err == -ELOOP)
1151                         pr_err("conflicting %s path\n", name);
1152                 return err;
1153         }
1154
1155         *ptrap = trap;
1156         return 0;
1157 }
1158
1159 /*
1160  * Determine how we treat concurrent use of upperdir/workdir based on the
1161  * index feature. This is papering over mount leaks of container runtimes,
1162  * for example, an old overlay mount is leaked and now its upperdir is
1163  * attempted to be used as a lower layer in a new overlay mount.
1164  */
1165 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1166 {
1167         if (ofs->config.index) {
1168                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1169                        name);
1170                 return -EBUSY;
1171         } else {
1172                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1173                         name);
1174                 return 0;
1175         }
1176 }
1177
1178 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1179                          struct ovl_layer *upper_layer, struct path *upperpath)
1180 {
1181         struct vfsmount *upper_mnt;
1182         int err;
1183
1184         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1185         if (err)
1186                 goto out;
1187
1188         /* Upper fs should not be r/o */
1189         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
1190                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1191                 err = -EINVAL;
1192                 goto out;
1193         }
1194
1195         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1196         if (err)
1197                 goto out;
1198
1199         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1200                              "upperdir");
1201         if (err)
1202                 goto out;
1203
1204         upper_mnt = clone_private_mount(upperpath);
1205         err = PTR_ERR(upper_mnt);
1206         if (IS_ERR(upper_mnt)) {
1207                 pr_err("failed to clone upperpath\n");
1208                 goto out;
1209         }
1210
1211         /* Don't inherit atime flags */
1212         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1213         upper_layer->mnt = upper_mnt;
1214         upper_layer->idx = 0;
1215         upper_layer->fsid = 0;
1216
1217         /*
1218          * Inherit SB_NOSEC flag from upperdir.
1219          *
1220          * This optimization changes behavior when a security related attribute
1221          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1222          * okay because we don't yet have guarantees in that case, but it will
1223          * need careful treatment once we want to honour changes to underlying
1224          * filesystems.
1225          */
1226         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1227                 sb->s_flags |= SB_NOSEC;
1228
1229         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1230                 ofs->upperdir_locked = true;
1231         } else {
1232                 err = ovl_report_in_use(ofs, "upperdir");
1233                 if (err)
1234                         goto out;
1235         }
1236
1237         err = 0;
1238 out:
1239         return err;
1240 }
1241
1242 /*
1243  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1244  * negative values if error is encountered.
1245  */
1246 static int ovl_check_rename_whiteout(struct dentry *workdir)
1247 {
1248         struct inode *dir = d_inode(workdir);
1249         struct dentry *temp;
1250         struct dentry *dest;
1251         struct dentry *whiteout;
1252         struct name_snapshot name;
1253         int err;
1254
1255         inode_lock_nested(dir, I_MUTEX_PARENT);
1256
1257         temp = ovl_create_temp(workdir, OVL_CATTR(S_IFREG | 0));
1258         err = PTR_ERR(temp);
1259         if (IS_ERR(temp))
1260                 goto out_unlock;
1261
1262         dest = ovl_lookup_temp(workdir);
1263         err = PTR_ERR(dest);
1264         if (IS_ERR(dest)) {
1265                 dput(temp);
1266                 goto out_unlock;
1267         }
1268
1269         /* Name is inline and stable - using snapshot as a copy helper */
1270         take_dentry_name_snapshot(&name, temp);
1271         err = ovl_do_rename(dir, temp, dir, dest, RENAME_WHITEOUT);
1272         if (err) {
1273                 if (err == -EINVAL)
1274                         err = 0;
1275                 goto cleanup_temp;
1276         }
1277
1278         whiteout = lookup_one_len(name.name.name, workdir, name.name.len);
1279         err = PTR_ERR(whiteout);
1280         if (IS_ERR(whiteout))
1281                 goto cleanup_temp;
1282
1283         err = ovl_is_whiteout(whiteout);
1284
1285         /* Best effort cleanup of whiteout and temp file */
1286         if (err)
1287                 ovl_cleanup(dir, whiteout);
1288         dput(whiteout);
1289
1290 cleanup_temp:
1291         ovl_cleanup(dir, temp);
1292         release_dentry_name_snapshot(&name);
1293         dput(temp);
1294         dput(dest);
1295
1296 out_unlock:
1297         inode_unlock(dir);
1298
1299         return err;
1300 }
1301
1302 static struct dentry *ovl_lookup_or_create(struct dentry *parent,
1303                                            const char *name, umode_t mode)
1304 {
1305         size_t len = strlen(name);
1306         struct dentry *child;
1307
1308         inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1309         child = lookup_one_len(name, parent, len);
1310         if (!IS_ERR(child) && !child->d_inode)
1311                 child = ovl_create_real(parent->d_inode, child,
1312                                         OVL_CATTR(mode));
1313         inode_unlock(parent->d_inode);
1314         dput(parent);
1315
1316         return child;
1317 }
1318
1319 /*
1320  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1321  * present.
1322  */
1323 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1324 {
1325         unsigned int ctr;
1326         struct dentry *d = dget(ofs->workbasedir);
1327         static const char *const volatile_path[] = {
1328                 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1329         };
1330         const char *const *name = volatile_path;
1331
1332         for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1333                 d = ovl_lookup_or_create(d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1334                 if (IS_ERR(d))
1335                         return PTR_ERR(d);
1336         }
1337         dput(d);
1338         return 0;
1339 }
1340
1341 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1342                             struct path *workpath)
1343 {
1344         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1345         struct dentry *temp, *workdir;
1346         bool rename_whiteout;
1347         bool d_type;
1348         int fh_type;
1349         int err;
1350
1351         err = mnt_want_write(mnt);
1352         if (err)
1353                 return err;
1354
1355         workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1356         err = PTR_ERR(workdir);
1357         if (IS_ERR_OR_NULL(workdir))
1358                 goto out;
1359
1360         ofs->workdir = workdir;
1361
1362         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1363         if (err)
1364                 goto out;
1365
1366         /*
1367          * Upper should support d_type, else whiteouts are visible.  Given
1368          * workdir and upper are on same fs, we can do iterate_dir() on
1369          * workdir. This check requires successful creation of workdir in
1370          * previous step.
1371          */
1372         err = ovl_check_d_type_supported(workpath);
1373         if (err < 0)
1374                 goto out;
1375
1376         d_type = err;
1377         if (!d_type)
1378                 pr_warn("upper fs needs to support d_type.\n");
1379
1380         /* Check if upper/work fs supports O_TMPFILE */
1381         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1382         ofs->tmpfile = !IS_ERR(temp);
1383         if (ofs->tmpfile)
1384                 dput(temp);
1385         else
1386                 pr_warn("upper fs does not support tmpfile.\n");
1387
1388
1389         /* Check if upper/work fs supports RENAME_WHITEOUT */
1390         err = ovl_check_rename_whiteout(ofs->workdir);
1391         if (err < 0)
1392                 goto out;
1393
1394         rename_whiteout = err;
1395         if (!rename_whiteout)
1396                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1397
1398         /*
1399          * Check if upper/work fs supports (trusted|user).overlay.* xattr
1400          */
1401         err = ovl_do_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1402         if (err) {
1403                 ofs->noxattr = true;
1404                 ofs->config.index = false;
1405                 ofs->config.metacopy = false;
1406                 pr_warn("upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1407                 err = 0;
1408         } else {
1409                 ovl_do_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1410         }
1411
1412         /*
1413          * We allowed sub-optimal upper fs configuration and don't want to break
1414          * users over kernel upgrade, but we never allowed remote upper fs, so
1415          * we can enforce strict requirements for remote upper fs.
1416          */
1417         if (ovl_dentry_remote(ofs->workdir) &&
1418             (!d_type || !rename_whiteout || ofs->noxattr)) {
1419                 pr_err("upper fs missing required features.\n");
1420                 err = -EINVAL;
1421                 goto out;
1422         }
1423
1424         /*
1425          * For volatile mount, create a incompat/volatile/dirty file to keep
1426          * track of it.
1427          */
1428         if (ofs->config.ovl_volatile) {
1429                 err = ovl_create_volatile_dirty(ofs);
1430                 if (err < 0) {
1431                         pr_err("Failed to create volatile/dirty file.\n");
1432                         goto out;
1433                 }
1434         }
1435
1436         /* Check if upper/work fs supports file handles */
1437         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1438         if (ofs->config.index && !fh_type) {
1439                 ofs->config.index = false;
1440                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1441         }
1442
1443         /* Check if upper fs has 32bit inode numbers */
1444         if (fh_type != FILEID_INO32_GEN)
1445                 ofs->xino_mode = -1;
1446
1447         /* NFS export of r/w mount depends on index */
1448         if (ofs->config.nfs_export && !ofs->config.index) {
1449                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1450                 ofs->config.nfs_export = false;
1451         }
1452 out:
1453         mnt_drop_write(mnt);
1454         return err;
1455 }
1456
1457 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1458                            struct path *upperpath)
1459 {
1460         int err;
1461         struct path workpath = { };
1462
1463         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1464         if (err)
1465                 goto out;
1466
1467         err = -EINVAL;
1468         if (upperpath->mnt != workpath.mnt) {
1469                 pr_err("workdir and upperdir must reside under the same mount\n");
1470                 goto out;
1471         }
1472         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1473                 pr_err("workdir and upperdir must be separate subtrees\n");
1474                 goto out;
1475         }
1476
1477         ofs->workbasedir = dget(workpath.dentry);
1478
1479         if (ovl_inuse_trylock(ofs->workbasedir)) {
1480                 ofs->workdir_locked = true;
1481         } else {
1482                 err = ovl_report_in_use(ofs, "workdir");
1483                 if (err)
1484                         goto out;
1485         }
1486
1487         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1488                              "workdir");
1489         if (err)
1490                 goto out;
1491
1492         err = ovl_make_workdir(sb, ofs, &workpath);
1493
1494 out:
1495         path_put(&workpath);
1496
1497         return err;
1498 }
1499
1500 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1501                             struct ovl_entry *oe, struct path *upperpath)
1502 {
1503         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1504         struct dentry *indexdir;
1505         int err;
1506
1507         err = mnt_want_write(mnt);
1508         if (err)
1509                 return err;
1510
1511         /* Verify lower root is upper root origin */
1512         err = ovl_verify_origin(ofs, upperpath->dentry,
1513                                 oe->lowerstack[0].dentry, true);
1514         if (err) {
1515                 pr_err("failed to verify upper root origin\n");
1516                 goto out;
1517         }
1518
1519         /* index dir will act also as workdir */
1520         iput(ofs->workdir_trap);
1521         ofs->workdir_trap = NULL;
1522         dput(ofs->workdir);
1523         ofs->workdir = NULL;
1524         indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1525         if (IS_ERR(indexdir)) {
1526                 err = PTR_ERR(indexdir);
1527         } else if (indexdir) {
1528                 ofs->indexdir = indexdir;
1529                 ofs->workdir = dget(indexdir);
1530
1531                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1532                                      "indexdir");
1533                 if (err)
1534                         goto out;
1535
1536                 /*
1537                  * Verify upper root is exclusively associated with index dir.
1538                  * Older kernels stored upper fh in ".overlay.origin"
1539                  * xattr. If that xattr exists, verify that it is a match to
1540                  * upper dir file handle. In any case, verify or set xattr
1541                  * ".overlay.upper" to indicate that index may have
1542                  * directory entries.
1543                  */
1544                 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1545                         err = ovl_verify_set_fh(ofs, ofs->indexdir,
1546                                                 OVL_XATTR_ORIGIN,
1547                                                 upperpath->dentry, true, false);
1548                         if (err)
1549                                 pr_err("failed to verify index dir 'origin' xattr\n");
1550                 }
1551                 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1552                                        true);
1553                 if (err)
1554                         pr_err("failed to verify index dir 'upper' xattr\n");
1555
1556                 /* Cleanup bad/stale/orphan index entries */
1557                 if (!err)
1558                         err = ovl_indexdir_cleanup(ofs);
1559         }
1560         if (err || !ofs->indexdir)
1561                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1562
1563 out:
1564         mnt_drop_write(mnt);
1565         return err;
1566 }
1567
1568 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1569 {
1570         unsigned int i;
1571
1572         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1573                 return true;
1574
1575         /*
1576          * We allow using single lower with null uuid for index and nfs_export
1577          * for example to support those features with single lower squashfs.
1578          * To avoid regressions in setups of overlay with re-formatted lower
1579          * squashfs, do not allow decoding origin with lower null uuid unless
1580          * user opted-in to one of the new features that require following the
1581          * lower inode of non-dir upper.
1582          */
1583         if (!ofs->config.index && !ofs->config.metacopy && !ofs->config.xino &&
1584             uuid_is_null(uuid))
1585                 return false;
1586
1587         for (i = 0; i < ofs->numfs; i++) {
1588                 /*
1589                  * We use uuid to associate an overlay lower file handle with a
1590                  * lower layer, so we can accept lower fs with null uuid as long
1591                  * as all lower layers with null uuid are on the same fs.
1592                  * if we detect multiple lower fs with the same uuid, we
1593                  * disable lower file handle decoding on all of them.
1594                  */
1595                 if (ofs->fs[i].is_lower &&
1596                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1597                         ofs->fs[i].bad_uuid = true;
1598                         return false;
1599                 }
1600         }
1601         return true;
1602 }
1603
1604 /* Get a unique fsid for the layer */
1605 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1606 {
1607         struct super_block *sb = path->mnt->mnt_sb;
1608         unsigned int i;
1609         dev_t dev;
1610         int err;
1611         bool bad_uuid = false;
1612
1613         for (i = 0; i < ofs->numfs; i++) {
1614                 if (ofs->fs[i].sb == sb)
1615                         return i;
1616         }
1617
1618         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1619                 bad_uuid = true;
1620                 if (ofs->config.index || ofs->config.nfs_export) {
1621                         ofs->config.index = false;
1622                         ofs->config.nfs_export = false;
1623                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to index=off,nfs_export=off.\n",
1624                                 uuid_is_null(&sb->s_uuid) ? "null" :
1625                                                             "conflicting",
1626                                 path->dentry);
1627                 }
1628         }
1629
1630         err = get_anon_bdev(&dev);
1631         if (err) {
1632                 pr_err("failed to get anonymous bdev for lowerpath\n");
1633                 return err;
1634         }
1635
1636         ofs->fs[ofs->numfs].sb = sb;
1637         ofs->fs[ofs->numfs].pseudo_dev = dev;
1638         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1639
1640         return ofs->numfs++;
1641 }
1642
1643 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1644                           struct path *stack, unsigned int numlower,
1645                           struct ovl_layer *layers)
1646 {
1647         int err;
1648         unsigned int i;
1649
1650         err = -ENOMEM;
1651         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1652         if (ofs->fs == NULL)
1653                 goto out;
1654
1655         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1656         ofs->numfs++;
1657
1658         /*
1659          * All lower layers that share the same fs as upper layer, use the same
1660          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1661          * only overlay to simplify ovl_fs_free().
1662          * is_lower will be set if upper fs is shared with a lower layer.
1663          */
1664         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1665         if (err) {
1666                 pr_err("failed to get anonymous bdev for upper fs\n");
1667                 goto out;
1668         }
1669
1670         if (ovl_upper_mnt(ofs)) {
1671                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1672                 ofs->fs[0].is_lower = false;
1673         }
1674
1675         for (i = 0; i < numlower; i++) {
1676                 struct vfsmount *mnt;
1677                 struct inode *trap;
1678                 int fsid;
1679
1680                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1681                 if (err < 0)
1682                         goto out;
1683
1684                 /*
1685                  * Check if lower root conflicts with this overlay layers before
1686                  * checking if it is in-use as upperdir/workdir of "another"
1687                  * mount, because we do not bother to check in ovl_is_inuse() if
1688                  * the upperdir/workdir is in fact in-use by our
1689                  * upperdir/workdir.
1690                  */
1691                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1692                 if (err)
1693                         goto out;
1694
1695                 if (ovl_is_inuse(stack[i].dentry)) {
1696                         err = ovl_report_in_use(ofs, "lowerdir");
1697                         if (err) {
1698                                 iput(trap);
1699                                 goto out;
1700                         }
1701                 }
1702
1703                 mnt = clone_private_mount(&stack[i]);
1704                 err = PTR_ERR(mnt);
1705                 if (IS_ERR(mnt)) {
1706                         pr_err("failed to clone lowerpath\n");
1707                         iput(trap);
1708                         goto out;
1709                 }
1710
1711                 /*
1712                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1713                  * will fail instead of modifying lower fs.
1714                  */
1715                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1716
1717                 layers[ofs->numlayer].trap = trap;
1718                 layers[ofs->numlayer].mnt = mnt;
1719                 layers[ofs->numlayer].idx = ofs->numlayer;
1720                 layers[ofs->numlayer].fsid = fsid;
1721                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1722                 ofs->numlayer++;
1723                 ofs->fs[fsid].is_lower = true;
1724         }
1725
1726         /*
1727          * When all layers on same fs, overlay can use real inode numbers.
1728          * With mount option "xino=<on|auto>", mounter declares that there are
1729          * enough free high bits in underlying fs to hold the unique fsid.
1730          * If overlayfs does encounter underlying inodes using the high xino
1731          * bits reserved for fsid, it emits a warning and uses the original
1732          * inode number or a non persistent inode number allocated from a
1733          * dedicated range.
1734          */
1735         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1736                 if (ofs->config.xino == OVL_XINO_ON)
1737                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1738                 ofs->xino_mode = 0;
1739         } else if (ofs->config.xino == OVL_XINO_OFF) {
1740                 ofs->xino_mode = -1;
1741         } else if (ofs->xino_mode < 0) {
1742                 /*
1743                  * This is a roundup of number of bits needed for encoding
1744                  * fsid, where fsid 0 is reserved for upper fs (even with
1745                  * lower only overlay) +1 extra bit is reserved for the non
1746                  * persistent inode number range that is used for resolving
1747                  * xino lower bits overflow.
1748                  */
1749                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1750                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1751         }
1752
1753         if (ofs->xino_mode > 0) {
1754                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1755                         ofs->xino_mode);
1756         }
1757
1758         err = 0;
1759 out:
1760         return err;
1761 }
1762
1763 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1764                                 const char *lower, unsigned int numlower,
1765                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1766 {
1767         int err;
1768         struct path *stack = NULL;
1769         unsigned int i;
1770         struct ovl_entry *oe;
1771
1772         if (!ofs->config.upperdir && numlower == 1) {
1773                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1774                 return ERR_PTR(-EINVAL);
1775         }
1776
1777         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1778         if (!stack)
1779                 return ERR_PTR(-ENOMEM);
1780
1781         err = -EINVAL;
1782         for (i = 0; i < numlower; i++) {
1783                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1784                 if (err)
1785                         goto out_err;
1786
1787                 lower = strchr(lower, '\0') + 1;
1788         }
1789
1790         err = -EINVAL;
1791         sb->s_stack_depth++;
1792         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1793                 pr_err("maximum fs stacking depth exceeded\n");
1794                 goto out_err;
1795         }
1796
1797         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1798         if (err)
1799                 goto out_err;
1800
1801         err = -ENOMEM;
1802         oe = ovl_alloc_entry(numlower);
1803         if (!oe)
1804                 goto out_err;
1805
1806         for (i = 0; i < numlower; i++) {
1807                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1808                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1809         }
1810
1811 out:
1812         for (i = 0; i < numlower; i++)
1813                 path_put(&stack[i]);
1814         kfree(stack);
1815
1816         return oe;
1817
1818 out_err:
1819         oe = ERR_PTR(err);
1820         goto out;
1821 }
1822
1823 /*
1824  * Check if this layer root is a descendant of:
1825  * - another layer of this overlayfs instance
1826  * - upper/work dir of any overlayfs instance
1827  */
1828 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1829                            struct dentry *dentry, const char *name)
1830 {
1831         struct dentry *next = dentry, *parent;
1832         int err = 0;
1833
1834         if (!dentry)
1835                 return 0;
1836
1837         parent = dget_parent(next);
1838
1839         /* Walk back ancestors to root (inclusive) looking for traps */
1840         while (!err && parent != next) {
1841                 if (ovl_lookup_trap_inode(sb, parent)) {
1842                         err = -ELOOP;
1843                         pr_err("overlapping %s path\n", name);
1844                 } else if (ovl_is_inuse(parent)) {
1845                         err = ovl_report_in_use(ofs, name);
1846                 }
1847                 next = parent;
1848                 parent = dget_parent(next);
1849                 dput(next);
1850         }
1851
1852         dput(parent);
1853
1854         return err;
1855 }
1856
1857 /*
1858  * Check if any of the layers or work dirs overlap.
1859  */
1860 static int ovl_check_overlapping_layers(struct super_block *sb,
1861                                         struct ovl_fs *ofs)
1862 {
1863         int i, err;
1864
1865         if (ovl_upper_mnt(ofs)) {
1866                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1867                                       "upperdir");
1868                 if (err)
1869                         return err;
1870
1871                 /*
1872                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1873                  * this instance and covers overlapping work and index dirs,
1874                  * unless work or index dir have been moved since created inside
1875                  * workbasedir.  In that case, we already have their traps in
1876                  * inode cache and we will catch that case on lookup.
1877                  */
1878                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir");
1879                 if (err)
1880                         return err;
1881         }
1882
1883         for (i = 1; i < ofs->numlayer; i++) {
1884                 err = ovl_check_layer(sb, ofs,
1885                                       ofs->layers[i].mnt->mnt_root,
1886                                       "lowerdir");
1887                 if (err)
1888                         return err;
1889         }
1890
1891         return 0;
1892 }
1893
1894 static struct dentry *ovl_get_root(struct super_block *sb,
1895                                    struct dentry *upperdentry,
1896                                    struct ovl_entry *oe)
1897 {
1898         struct dentry *root;
1899         struct ovl_path *lowerpath = &oe->lowerstack[0];
1900         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1901         int fsid = lowerpath->layer->fsid;
1902         struct ovl_inode_params oip = {
1903                 .upperdentry = upperdentry,
1904                 .lowerpath = lowerpath,
1905         };
1906
1907         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1908         if (!root)
1909                 return NULL;
1910
1911         root->d_fsdata = oe;
1912
1913         if (upperdentry) {
1914                 /* Root inode uses upper st_ino/i_ino */
1915                 ino = d_inode(upperdentry)->i_ino;
1916                 fsid = 0;
1917                 ovl_dentry_set_upper_alias(root);
1918                 if (ovl_is_impuredir(sb, upperdentry))
1919                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1920         }
1921
1922         /* Root is always merge -> can have whiteouts */
1923         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1924         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1925         ovl_set_upperdata(d_inode(root));
1926         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1927         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1928
1929         return root;
1930 }
1931
1932 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1933 {
1934         struct path upperpath = { };
1935         struct dentry *root_dentry;
1936         struct ovl_entry *oe;
1937         struct ovl_fs *ofs;
1938         struct ovl_layer *layers;
1939         struct cred *cred;
1940         char *splitlower = NULL;
1941         unsigned int numlower;
1942         int err;
1943
1944         err = -EIO;
1945         if (WARN_ON(sb->s_user_ns != current_user_ns()))
1946                 goto out;
1947
1948         sb->s_d_op = &ovl_dentry_operations;
1949
1950         err = -ENOMEM;
1951         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1952         if (!ofs)
1953                 goto out;
1954
1955         ofs->creator_cred = cred = prepare_creds();
1956         if (!cred)
1957                 goto out_err;
1958
1959         /* Is there a reason anyone would want not to share whiteouts? */
1960         ofs->share_whiteout = true;
1961
1962         ofs->config.index = ovl_index_def;
1963         ofs->config.uuid = true;
1964         ofs->config.nfs_export = ovl_nfs_export_def;
1965         ofs->config.xino = ovl_xino_def();
1966         ofs->config.metacopy = ovl_metacopy_def;
1967         err = ovl_parse_opt((char *) data, &ofs->config);
1968         if (err)
1969                 goto out_err;
1970
1971         err = -EINVAL;
1972         if (!ofs->config.lowerdir) {
1973                 if (!silent)
1974                         pr_err("missing 'lowerdir'\n");
1975                 goto out_err;
1976         }
1977
1978         err = -ENOMEM;
1979         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1980         if (!splitlower)
1981                 goto out_err;
1982
1983         numlower = ovl_split_lowerdirs(splitlower);
1984         if (numlower > OVL_MAX_STACK) {
1985                 pr_err("too many lower directories, limit is %d\n",
1986                        OVL_MAX_STACK);
1987                 goto out_err;
1988         }
1989
1990         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1991         if (!layers)
1992                 goto out_err;
1993
1994         ofs->layers = layers;
1995         /* Layer 0 is reserved for upper even if there's no upper */
1996         ofs->numlayer = 1;
1997
1998         sb->s_stack_depth = 0;
1999         sb->s_maxbytes = MAX_LFS_FILESIZE;
2000         atomic_long_set(&ofs->last_ino, 1);
2001         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
2002         if (ofs->config.xino != OVL_XINO_OFF) {
2003                 ofs->xino_mode = BITS_PER_LONG - 32;
2004                 if (!ofs->xino_mode) {
2005                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
2006                         ofs->config.xino = OVL_XINO_OFF;
2007                 }
2008         }
2009
2010         /* alloc/destroy_inode needed for setting up traps in inode cache */
2011         sb->s_op = &ovl_super_operations;
2012
2013         if (ofs->config.upperdir) {
2014                 struct super_block *upper_sb;
2015
2016                 if (!ofs->config.workdir) {
2017                         pr_err("missing 'workdir'\n");
2018                         goto out_err;
2019                 }
2020
2021                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
2022                 if (err)
2023                         goto out_err;
2024
2025                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
2026                 if (!ovl_should_sync(ofs)) {
2027                         ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
2028                         if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
2029                                 err = -EIO;
2030                                 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
2031                                 goto out_err;
2032                         }
2033                 }
2034
2035                 err = ovl_get_workdir(sb, ofs, &upperpath);
2036                 if (err)
2037                         goto out_err;
2038
2039                 if (!ofs->workdir)
2040                         sb->s_flags |= SB_RDONLY;
2041
2042                 sb->s_stack_depth = upper_sb->s_stack_depth;
2043                 sb->s_time_gran = upper_sb->s_time_gran;
2044         }
2045         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2046         err = PTR_ERR(oe);
2047         if (IS_ERR(oe))
2048                 goto out_err;
2049
2050         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2051         if (!ovl_upper_mnt(ofs))
2052                 sb->s_flags |= SB_RDONLY;
2053
2054         if (!ofs->config.uuid && ofs->numfs > 1) {
2055                 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2056                 ofs->config.uuid = true;
2057         }
2058
2059         if (!ovl_force_readonly(ofs) && ofs->config.index) {
2060                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2061                 if (err)
2062                         goto out_free_oe;
2063
2064                 /* Force r/o mount with no index dir */
2065                 if (!ofs->indexdir)
2066                         sb->s_flags |= SB_RDONLY;
2067         }
2068
2069         err = ovl_check_overlapping_layers(sb, ofs);
2070         if (err)
2071                 goto out_free_oe;
2072
2073         /* Show index=off in /proc/mounts for forced r/o mount */
2074         if (!ofs->indexdir) {
2075                 ofs->config.index = false;
2076                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2077                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2078                         ofs->config.nfs_export = false;
2079                 }
2080         }
2081
2082         if (ofs->config.metacopy && ofs->config.nfs_export) {
2083                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2084                 ofs->config.nfs_export = false;
2085         }
2086
2087         if (ofs->config.nfs_export)
2088                 sb->s_export_op = &ovl_export_operations;
2089
2090         /* Never override disk quota limits or use reserved space */
2091         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2092
2093         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2094         sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2095                 ovl_trusted_xattr_handlers;
2096         sb->s_fs_info = ofs;
2097         sb->s_flags |= SB_POSIXACL;
2098         sb->s_iflags |= SB_I_SKIP_SYNC;
2099
2100         err = -ENOMEM;
2101         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2102         if (!root_dentry)
2103                 goto out_free_oe;
2104
2105         mntput(upperpath.mnt);
2106         kfree(splitlower);
2107
2108         sb->s_root = root_dentry;
2109
2110         return 0;
2111
2112 out_free_oe:
2113         ovl_entry_stack_free(oe);
2114         kfree(oe);
2115 out_err:
2116         kfree(splitlower);
2117         path_put(&upperpath);
2118         ovl_free_fs(ofs);
2119 out:
2120         return err;
2121 }
2122
2123 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2124                                 const char *dev_name, void *raw_data)
2125 {
2126         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2127 }
2128
2129 static struct file_system_type ovl_fs_type = {
2130         .owner          = THIS_MODULE,
2131         .name           = "overlay",
2132         .fs_flags       = FS_USERNS_MOUNT,
2133         .mount          = ovl_mount,
2134         .kill_sb        = kill_anon_super,
2135 };
2136 MODULE_ALIAS_FS("overlay");
2137
2138 static void ovl_inode_init_once(void *foo)
2139 {
2140         struct ovl_inode *oi = foo;
2141
2142         inode_init_once(&oi->vfs_inode);
2143 }
2144
2145 static int __init ovl_init(void)
2146 {
2147         int err;
2148
2149         ovl_inode_cachep = kmem_cache_create("ovl_inode",
2150                                              sizeof(struct ovl_inode), 0,
2151                                              (SLAB_RECLAIM_ACCOUNT|
2152                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2153                                              ovl_inode_init_once);
2154         if (ovl_inode_cachep == NULL)
2155                 return -ENOMEM;
2156
2157         err = ovl_aio_request_cache_init();
2158         if (!err) {
2159                 err = register_filesystem(&ovl_fs_type);
2160                 if (!err)
2161                         return 0;
2162
2163                 ovl_aio_request_cache_destroy();
2164         }
2165         kmem_cache_destroy(ovl_inode_cachep);
2166
2167         return err;
2168 }
2169
2170 static void __exit ovl_exit(void)
2171 {
2172         unregister_filesystem(&ovl_fs_type);
2173
2174         /*
2175          * Make sure all delayed rcu free inodes are flushed before we
2176          * destroy cache.
2177          */
2178         rcu_barrier();
2179         kmem_cache_destroy(ovl_inode_cachep);
2180         ovl_aio_request_cache_destroy();
2181 }
2182
2183 module_init(ovl_init);
2184 module_exit(ovl_exit);