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 = alloc_inode_sb(sb, 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->lowerpath.dentry = NULL;
 188         oi->lowerpath.layer = NULL;
 189         oi->lowerdata = NULL;
 190         mutex_init(&oi->lock);
 191
 192         return &oi->vfs_inode;
 193 }
 194
 195 static void ovl_free_inode(struct inode *inode)
 196 {
 197         struct ovl_inode *oi = OVL_I(inode);
 198
 199         kfree(oi->redirect);
 200         mutex_destroy(&oi->lock);
 201         kmem_cache_free(ovl_inode_cachep, oi);
 202 }
 203
 204 static void ovl_destroy_inode(struct inode *inode)
 205 {
 206         struct ovl_inode *oi = OVL_I(inode);
 207
 208         dput(oi->__upperdentry);
 209         dput(oi->lowerpath.dentry);
 210         if (S_ISDIR(inode->i_mode))
 211                 ovl_dir_cache_free(inode);
 212         else
 213                 iput(oi->lowerdata);
 214 }
 215
 216 static void ovl_free_fs(struct ovl_fs *ofs)
 217 {
 218         struct vfsmount **mounts;
 219         unsigned i;
 220
 221         iput(ofs->workbasedir_trap);
 222         iput(ofs->indexdir_trap);
 223         iput(ofs->workdir_trap);
 224         dput(ofs->whiteout);
 225         dput(ofs->indexdir);
 226         dput(ofs->workdir);
 227         if (ofs->workdir_locked)
 228                 ovl_inuse_unlock(ofs->workbasedir);
 229         dput(ofs->workbasedir);
 230         if (ofs->upperdir_locked)
 231                 ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
 232
 233         /* Hack!  Reuse ofs->layers as a vfsmount array before freeing it */
 234         mounts = (struct vfsmount **) ofs->layers;
 235         for (i = 0; i < ofs->numlayer; i++) {
 236                 iput(ofs->layers[i].trap);
 237                 mounts[i] = ofs->layers[i].mnt;
 238         }
 239         kern_unmount_array(mounts, ofs->numlayer);
 240         kfree(ofs->layers);
 241         for (i = 0; i < ofs->numfs; i++)
 242                 free_anon_bdev(ofs->fs[i].pseudo_dev);
 243         kfree(ofs->fs);
 244
 245         kfree(ofs->config.lowerdir);
 246         kfree(ofs->config.upperdir);
 247         kfree(ofs->config.workdir);
 248         kfree(ofs->config.redirect_mode);
 249         if (ofs->creator_cred)
 250                 put_cred(ofs->creator_cred);
 251         kfree(ofs);
 252 }
 253
 254 static void ovl_put_super(struct super_block *sb)
 255 {
 256         struct ovl_fs *ofs = sb->s_fs_info;
 257
 258         ovl_free_fs(ofs);
 259 }
 260
 261 /* Sync real dirty inodes in upper filesystem (if it exists) */
 262 static int ovl_sync_fs(struct super_block *sb, int wait)
 263 {
 264         struct ovl_fs *ofs = sb->s_fs_info;
 265         struct super_block *upper_sb;
 266         int ret;
 267
 268         ret = ovl_sync_status(ofs);
 269         /*
 270          * We have to always set the err, because the return value isn't
 271          * checked in syncfs, and instead indirectly return an error via
 272          * the sb's writeback errseq, which VFS inspects after this call.
 273          */
 274         if (ret < 0) {
 275                 errseq_set(&sb->s_wb_err, -EIO);
 276                 return -EIO;
 277         }
 278
 279         if (!ret)
 280                 return ret;
 281
 282         /*
 283          * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
 284          * All the super blocks will be iterated, including upper_sb.
 285          *
 286          * If this is a syncfs(2) call, then we do need to call
 287          * sync_filesystem() on upper_sb, but enough if we do it when being
 288          * called with wait == 1.
 289          */
 290         if (!wait)
 291                 return 0;
 292
 293         upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 294
 295         down_read(&upper_sb->s_umount);
 296         ret = sync_filesystem(upper_sb);
 297         up_read(&upper_sb->s_umount);
 298
 299         return ret;
 300 }
 301
 302 /**
 303  * ovl_statfs
 304  * @sb: The overlayfs super block
 305  * @buf: The struct kstatfs to fill in with stats
 306  *
 307  * Get the filesystem statistics.  As writes always target the upper layer
 308  * filesystem pass the statfs to the upper filesystem (if it exists)
 309  */
 310 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 311 {
 312         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 313         struct dentry *root_dentry = dentry->d_sb->s_root;
 314         struct path path;
 315         int err;
 316
 317         ovl_path_real(root_dentry, &path);
 318
 319         err = vfs_statfs(&path, buf);
 320         if (!err) {
 321                 buf->f_namelen = ofs->namelen;
 322                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 323         }
 324
 325         return err;
 326 }
 327
 328 /* Will this overlay be forced to mount/remount ro? */
 329 static bool ovl_force_readonly(struct ovl_fs *ofs)
 330 {
 331         return (!ovl_upper_mnt(ofs) || !ofs->workdir);
 332 }
 333
 334 static const char *ovl_redirect_mode_def(void)
 335 {
 336         return ovl_redirect_dir_def ? "on" : "off";
 337 }
 338
 339 static const char * const ovl_xino_str[] = {
 340         "off",
 341         "auto",
 342         "on",
 343 };
 344
 345 static inline int ovl_xino_def(void)
 346 {
 347         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 348 }
 349
 350 /**
 351  * ovl_show_options
 352  *
 353  * Prints the mount options for a given superblock.
 354  * Returns zero; does not fail.
 355  */
 356 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 357 {
 358         struct super_block *sb = dentry->d_sb;
 359         struct ovl_fs *ofs = sb->s_fs_info;
 360
 361         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 362         if (ofs->config.upperdir) {
 363                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 364                 seq_show_option(m, "workdir", ofs->config.workdir);
 365         }
 366         if (ofs->config.default_permissions)
 367                 seq_puts(m, ",default_permissions");
 368         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 369                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 370         if (ofs->config.index != ovl_index_def)
 371                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 372         if (!ofs->config.uuid)
 373                 seq_puts(m, ",uuid=off");
 374         if (ofs->config.nfs_export != ovl_nfs_export_def)
 375                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 376                                                 "on" : "off");
 377         if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
 378                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 379         if (ofs->config.metacopy != ovl_metacopy_def)
 380                 seq_printf(m, ",metacopy=%s",
 381                            ofs->config.metacopy ? "on" : "off");
 382         if (ofs->config.ovl_volatile)
 383                 seq_puts(m, ",volatile");
 384         if (ofs->config.userxattr)
 385                 seq_puts(m, ",userxattr");
 386         return 0;
 387 }
 388
 389 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 390 {
 391         struct ovl_fs *ofs = sb->s_fs_info;
 392         struct super_block *upper_sb;
 393         int ret = 0;
 394
 395         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 396                 return -EROFS;
 397
 398         if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
 399                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 400                 if (ovl_should_sync(ofs)) {
 401                         down_read(&upper_sb->s_umount);
 402                         ret = sync_filesystem(upper_sb);
 403                         up_read(&upper_sb->s_umount);
 404                 }
 405         }
 406
 407         return ret;
 408 }
 409
 410 static const struct super_operations ovl_super_operations = {
 411         .alloc_inode    = ovl_alloc_inode,
 412         .free_inode     = ovl_free_inode,
 413         .destroy_inode  = ovl_destroy_inode,
 414         .drop_inode     = generic_delete_inode,
 415         .put_super      = ovl_put_super,
 416         .sync_fs        = ovl_sync_fs,
 417         .statfs         = ovl_statfs,
 418         .show_options   = ovl_show_options,
 419         .remount_fs     = ovl_remount,
 420 };
 421
 422 enum {
 423         OPT_LOWERDIR,
 424         OPT_UPPERDIR,
 425         OPT_WORKDIR,
 426         OPT_DEFAULT_PERMISSIONS,
 427         OPT_REDIRECT_DIR,
 428         OPT_INDEX_ON,
 429         OPT_INDEX_OFF,
 430         OPT_UUID_ON,
 431         OPT_UUID_OFF,
 432         OPT_NFS_EXPORT_ON,
 433         OPT_USERXATTR,
 434         OPT_NFS_EXPORT_OFF,
 435         OPT_XINO_ON,
 436         OPT_XINO_OFF,
 437         OPT_XINO_AUTO,
 438         OPT_METACOPY_ON,
 439         OPT_METACOPY_OFF,
 440         OPT_VOLATILE,
 441         OPT_ERR,
 442 };
 443
 444 static const match_table_t ovl_tokens = {
 445         {OPT_LOWERDIR,                  "lowerdir=%s"},
 446         {OPT_UPPERDIR,                  "upperdir=%s"},
 447         {OPT_WORKDIR,                   "workdir=%s"},
 448         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 449         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 450         {OPT_INDEX_ON,                  "index=on"},
 451         {OPT_INDEX_OFF,                 "index=off"},
 452         {OPT_USERXATTR,                 "userxattr"},
 453         {OPT_UUID_ON,                   "uuid=on"},
 454         {OPT_UUID_OFF,                  "uuid=off"},
 455         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 456         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 457         {OPT_XINO_ON,                   "xino=on"},
 458         {OPT_XINO_OFF,                  "xino=off"},
 459         {OPT_XINO_AUTO,                 "xino=auto"},
 460         {OPT_METACOPY_ON,               "metacopy=on"},
 461         {OPT_METACOPY_OFF,              "metacopy=off"},
 462         {OPT_VOLATILE,                  "volatile"},
 463         {OPT_ERR,                       NULL}
 464 };
 465
 466 static char *ovl_next_opt(char **s)
 467 {
 468         char *sbegin = *s;
 469         char *p;
 470
 471         if (sbegin == NULL)
 472                 return NULL;
 473
 474         for (p = sbegin; *p; p++) {
 475                 if (*p == '\\') {
 476                         p++;
 477                         if (!*p)
 478                                 break;
 479                 } else if (*p == ',') {
 480                         *p = '\0';
 481                         *s = p + 1;
 482                         return sbegin;
 483                 }
 484         }
 485         *s = NULL;
 486         return sbegin;
 487 }
 488
 489 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 490 {
 491         if (strcmp(mode, "on") == 0) {
 492                 config->redirect_dir = true;
 493                 /*
 494                  * Does not make sense to have redirect creation without
 495                  * redirect following.
 496                  */
 497                 config->redirect_follow = true;
 498         } else if (strcmp(mode, "follow") == 0) {
 499                 config->redirect_follow = true;
 500         } else if (strcmp(mode, "off") == 0) {
 501                 if (ovl_redirect_always_follow)
 502                         config->redirect_follow = true;
 503         } else if (strcmp(mode, "nofollow") != 0) {
 504                 pr_err("bad mount option \"redirect_dir=%s\"\n",
 505                        mode);
 506                 return -EINVAL;
 507         }
 508
 509         return 0;
 510 }
 511
 512 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 513 {
 514         char *p;
 515         int err;
 516         bool metacopy_opt = false, redirect_opt = false;
 517         bool nfs_export_opt = false, index_opt = false;
 518
 519         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 520         if (!config->redirect_mode)
 521                 return -ENOMEM;
 522
 523         while ((p = ovl_next_opt(&opt)) != NULL) {
 524                 int token;
 525                 substring_t args[MAX_OPT_ARGS];
 526
 527                 if (!*p)
 528                         continue;
 529
 530                 token = match_token(p, ovl_tokens, args);
 531                 switch (token) {
 532                 case OPT_UPPERDIR:
 533                         kfree(config->upperdir);
 534                         config->upperdir = match_strdup(&args[0]);
 535                         if (!config->upperdir)
 536                                 return -ENOMEM;
 537                         break;
 538
 539                 case OPT_LOWERDIR:
 540                         kfree(config->lowerdir);
 541                         config->lowerdir = match_strdup(&args[0]);
 542                         if (!config->lowerdir)
 543                                 return -ENOMEM;
 544                         break;
 545
 546                 case OPT_WORKDIR:
 547                         kfree(config->workdir);
 548                         config->workdir = match_strdup(&args[0]);
 549                         if (!config->workdir)
 550                                 return -ENOMEM;
 551                         break;
 552
 553                 case OPT_DEFAULT_PERMISSIONS:
 554                         config->default_permissions = true;
 555                         break;
 556
 557                 case OPT_REDIRECT_DIR:
 558                         kfree(config->redirect_mode);
 559                         config->redirect_mode = match_strdup(&args[0]);
 560                         if (!config->redirect_mode)
 561                                 return -ENOMEM;
 562                         redirect_opt = true;
 563                         break;
 564
 565                 case OPT_INDEX_ON:
 566                         config->index = true;
 567                         index_opt = true;
 568                         break;
 569
 570                 case OPT_INDEX_OFF:
 571                         config->index = false;
 572                         index_opt = true;
 573                         break;
 574
 575                 case OPT_UUID_ON:
 576                         config->uuid = true;
 577                         break;
 578
 579                 case OPT_UUID_OFF:
 580                         config->uuid = false;
 581                         break;
 582
 583                 case OPT_NFS_EXPORT_ON:
 584                         config->nfs_export = true;
 585                         nfs_export_opt = true;
 586                         break;
 587
 588                 case OPT_NFS_EXPORT_OFF:
 589                         config->nfs_export = false;
 590                         nfs_export_opt = true;
 591                         break;
 592
 593                 case OPT_XINO_ON:
 594                         config->xino = OVL_XINO_ON;
 595                         break;
 596
 597                 case OPT_XINO_OFF:
 598                         config->xino = OVL_XINO_OFF;
 599                         break;
 600
 601                 case OPT_XINO_AUTO:
 602                         config->xino = OVL_XINO_AUTO;
 603                         break;
 604
 605                 case OPT_METACOPY_ON:
 606                         config->metacopy = true;
 607                         metacopy_opt = true;
 608                         break;
 609
 610                 case OPT_METACOPY_OFF:
 611                         config->metacopy = false;
 612                         metacopy_opt = true;
 613                         break;
 614
 615                 case OPT_VOLATILE:
 616                         config->ovl_volatile = true;
 617                         break;
 618
 619                 case OPT_USERXATTR:
 620                         config->userxattr = true;
 621                         break;
 622
 623                 default:
 624                         pr_err("unrecognized mount option \"%s\" or missing value\n",
 625                                         p);
 626                         return -EINVAL;
 627                 }
 628         }
 629
 630         /* Workdir/index are useless in non-upper mount */
 631         if (!config->upperdir) {
 632                 if (config->workdir) {
 633                         pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 634                                 config->workdir);
 635                         kfree(config->workdir);
 636                         config->workdir = NULL;
 637                 }
 638                 if (config->index && index_opt) {
 639                         pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
 640                         index_opt = false;
 641                 }
 642                 config->index = false;
 643         }
 644
 645         if (!config->upperdir && config->ovl_volatile) {
 646                 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
 647                 config->ovl_volatile = false;
 648         }
 649
 650         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 651         if (err)
 652                 return err;
 653
 654         /*
 655          * This is to make the logic below simpler.  It doesn't make any other
 656          * difference, since config->redirect_dir is only used for upper.
 657          */
 658         if (!config->upperdir && config->redirect_follow)
 659                 config->redirect_dir = true;
 660
 661         /* Resolve metacopy -> redirect_dir dependency */
 662         if (config->metacopy && !config->redirect_dir) {
 663                 if (metacopy_opt && redirect_opt) {
 664                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
 665                                config->redirect_mode);
 666                         return -EINVAL;
 667                 }
 668                 if (redirect_opt) {
 669                         /*
 670                          * There was an explicit redirect_dir=... that resulted
 671                          * in this conflict.
 672                          */
 673                         pr_info("disabling metacopy due to redirect_dir=%s\n",
 674                                 config->redirect_mode);
 675                         config->metacopy = false;
 676                 } else {
 677                         /* Automatically enable redirect otherwise. */
 678                         config->redirect_follow = config->redirect_dir = true;
 679                 }
 680         }
 681
 682         /* Resolve nfs_export -> index dependency */
 683         if (config->nfs_export && !config->index) {
 684                 if (!config->upperdir && config->redirect_follow) {
 685                         pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
 686                         config->nfs_export = false;
 687                 } else if (nfs_export_opt && index_opt) {
 688                         pr_err("conflicting options: nfs_export=on,index=off\n");
 689                         return -EINVAL;
 690                 } else if (index_opt) {
 691                         /*
 692                          * There was an explicit index=off that resulted
 693                          * in this conflict.
 694                          */
 695                         pr_info("disabling nfs_export due to index=off\n");
 696                         config->nfs_export = false;
 697                 } else {
 698                         /* Automatically enable index otherwise. */
 699                         config->index = true;
 700                 }
 701         }
 702
 703         /* Resolve nfs_export -> !metacopy dependency */
 704         if (config->nfs_export && config->metacopy) {
 705                 if (nfs_export_opt && metacopy_opt) {
 706                         pr_err("conflicting options: nfs_export=on,metacopy=on\n");
 707                         return -EINVAL;
 708                 }
 709                 if (metacopy_opt) {
 710                         /*
 711                          * There was an explicit metacopy=on that resulted
 712                          * in this conflict.
 713                          */
 714                         pr_info("disabling nfs_export due to metacopy=on\n");
 715                         config->nfs_export = false;
 716                 } else {
 717                         /*
 718                          * There was an explicit nfs_export=on that resulted
 719                          * in this conflict.
 720                          */
 721                         pr_info("disabling metacopy due to nfs_export=on\n");
 722                         config->metacopy = false;
 723                 }
 724         }
 725
 726
 727         /* Resolve userxattr -> !redirect && !metacopy dependency */
 728         if (config->userxattr) {
 729                 if (config->redirect_follow && redirect_opt) {
 730                         pr_err("conflicting options: userxattr,redirect_dir=%s\n",
 731                                config->redirect_mode);
 732                         return -EINVAL;
 733                 }
 734                 if (config->metacopy && metacopy_opt) {
 735                         pr_err("conflicting options: userxattr,metacopy=on\n");
 736                         return -EINVAL;
 737                 }
 738                 /*
 739                  * Silently disable default setting of redirect and metacopy.
 740                  * This shall be the default in the future as well: these
 741                  * options must be explicitly enabled if used together with
 742                  * userxattr.
 743                  */
 744                 config->redirect_dir = config->redirect_follow = false;
 745                 config->metacopy = false;
 746         }
 747
 748         return 0;
 749 }
 750
 751 #define OVL_WORKDIR_NAME "work"
 752 #define OVL_INDEXDIR_NAME "index"
 753
 754 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 755                                          const char *name, bool persist)
 756 {
 757         struct inode *dir =  ofs->workbasedir->d_inode;
 758         struct vfsmount *mnt = ovl_upper_mnt(ofs);
 759         struct dentry *work;
 760         int err;
 761         bool retried = false;
 762
 763         inode_lock_nested(dir, I_MUTEX_PARENT);
 764 retry:
 765         work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));
 766
 767         if (!IS_ERR(work)) {
 768                 struct iattr attr = {
 769                         .ia_valid = ATTR_MODE,
 770                         .ia_mode = S_IFDIR | 0,
 771                 };
 772
 773                 if (work->d_inode) {
 774                         err = -EEXIST;
 775                         if (retried)
 776                                 goto out_dput;
 777
 778                         if (persist)
 779                                 goto out_unlock;
 780
 781                         retried = true;
 782                         err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);
 783                         dput(work);
 784                         if (err == -EINVAL) {
 785                                 work = ERR_PTR(err);
 786                                 goto out_unlock;
 787                         }
 788                         goto retry;
 789                 }
 790
 791                 err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);
 792                 if (err)
 793                         goto out_dput;
 794
 795                 /* Weird filesystem returning with hashed negative (kernfs)? */
 796                 err = -EINVAL;
 797                 if (d_really_is_negative(work))
 798                         goto out_dput;
 799
 800                 /*
 801                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 802                  *
 803                  * a) success (there was a POSIX ACL xattr and was removed)
 804                  * b) -ENODATA (there was no POSIX ACL xattr)
 805                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 806                  *
 807                  * There are various other error values that could effectively
 808                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 809                  * if the xattr name is too long), but the set of filesystems
 810                  * allowed as upper are limited to "normal" ones, where checking
 811                  * for the above two errors is sufficient.
 812                  */
 813                 err = ovl_do_removexattr(ofs, work,
 814                                          XATTR_NAME_POSIX_ACL_DEFAULT);
 815                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 816                         goto out_dput;
 817
 818                 err = ovl_do_removexattr(ofs, work,
 819                                          XATTR_NAME_POSIX_ACL_ACCESS);
 820                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 821                         goto out_dput;
 822
 823                 /* Clear any inherited mode bits */
 824                 inode_lock(work->d_inode);
 825                 err = ovl_do_notify_change(ofs, work, &attr);
 826                 inode_unlock(work->d_inode);
 827                 if (err)
 828                         goto out_dput;
 829         } else {
 830                 err = PTR_ERR(work);
 831                 goto out_err;
 832         }
 833 out_unlock:
 834         inode_unlock(dir);
 835         return work;
 836
 837 out_dput:
 838         dput(work);
 839 out_err:
 840         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
 841                 ofs->config.workdir, name, -err);
 842         work = NULL;
 843         goto out_unlock;
 844 }
 845
 846 static void ovl_unescape(char *s)
 847 {
 848         char *d = s;
 849
 850         for (;; s++, d++) {
 851                 if (*s == '\\')
 852                         s++;
 853                 *d = *s;
 854                 if (!*s)
 855                         break;
 856         }
 857 }
 858
 859 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 860 {
 861         int err = -EINVAL;
 862
 863         if (!*name) {
 864                 pr_err("empty lowerdir\n");
 865                 goto out;
 866         }
 867         err = kern_path(name, LOOKUP_FOLLOW, path);
 868         if (err) {
 869                 pr_err("failed to resolve '%s': %i\n", name, err);
 870                 goto out;
 871         }
 872         err = -EINVAL;
 873         if (ovl_dentry_weird(path->dentry)) {
 874                 pr_err("filesystem on '%s' not supported\n", name);
 875                 goto out_put;
 876         }
 877         if (!d_is_dir(path->dentry)) {
 878                 pr_err("'%s' not a directory\n", name);
 879                 goto out_put;
 880         }
 881         return 0;
 882
 883 out_put:
 884         path_put_init(path);
 885 out:
 886         return err;
 887 }
 888
 889 static int ovl_mount_dir(const char *name, struct path *path)
 890 {
 891         int err = -ENOMEM;
 892         char *tmp = kstrdup(name, GFP_KERNEL);
 893
 894         if (tmp) {
 895                 ovl_unescape(tmp);
 896                 err = ovl_mount_dir_noesc(tmp, path);
 897
 898                 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
 899                         pr_err("filesystem on '%s' not supported as upperdir\n",
 900                                tmp);
 901                         path_put_init(path);
 902                         err = -EINVAL;
 903                 }
 904                 kfree(tmp);
 905         }
 906         return err;
 907 }
 908
 909 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 910                              const char *name)
 911 {
 912         struct kstatfs statfs;
 913         int err = vfs_statfs(path, &statfs);
 914
 915         if (err)
 916                 pr_err("statfs failed on '%s'\n", name);
 917         else
 918                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 919
 920         return err;
 921 }
 922
 923 static int ovl_lower_dir(const char *name, struct path *path,
 924                          struct ovl_fs *ofs, int *stack_depth)
 925 {
 926         int fh_type;
 927         int err;
 928
 929         err = ovl_mount_dir_noesc(name, path);
 930         if (err)
 931                 return err;
 932
 933         err = ovl_check_namelen(path, ofs, name);
 934         if (err)
 935                 return err;
 936
 937         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 938
 939         /*
 940          * The inodes index feature and NFS export need to encode and decode
 941          * file handles, so they require that all layers support them.
 942          */
 943         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 944         if ((ofs->config.nfs_export ||
 945              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 946                 ofs->config.index = false;
 947                 ofs->config.nfs_export = false;
 948                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 949                         name);
 950         }
 951         /*
 952          * Decoding origin file handle is required for persistent st_ino.
 953          * Without persistent st_ino, xino=auto falls back to xino=off.
 954          */
 955         if (ofs->config.xino == OVL_XINO_AUTO &&
 956             ofs->config.upperdir && !fh_type) {
 957                 ofs->config.xino = OVL_XINO_OFF;
 958                 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
 959                         name);
 960         }
 961
 962         /* Check if lower fs has 32bit inode numbers */
 963         if (fh_type != FILEID_INO32_GEN)
 964                 ofs->xino_mode = -1;
 965
 966         return 0;
 967 }
 968
 969 /* Workdir should not be subdir of upperdir and vice versa */
 970 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 971 {
 972         bool ok = false;
 973
 974         if (workdir != upperdir) {
 975                 ok = (lock_rename(workdir, upperdir) == NULL);
 976                 unlock_rename(workdir, upperdir);
 977         }
 978         return ok;
 979 }
 980
 981 static unsigned int ovl_split_lowerdirs(char *str)
 982 {
 983         unsigned int ctr = 1;
 984         char *s, *d;
 985
 986         for (s = d = str;; s++, d++) {
 987                 if (*s == '\\') {
 988                         s++;
 989                 } else if (*s == ':') {
 990                         *d = '\0';
 991                         ctr++;
 992                         continue;
 993                 }
 994                 *d = *s;
 995                 if (!*s)
 996                         break;
 997         }
 998         return ctr;
 999 }
1000
1001 static int __maybe_unused
1002 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
1003                         struct dentry *dentry, struct inode *inode,
1004                         const char *name, void *buffer, size_t size)
1005 {
1006         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
1007 }
1008
1009 static int __maybe_unused
1010 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
1011                         struct user_namespace *mnt_userns,
1012                         struct dentry *dentry, struct inode *inode,
1013                         const char *name, const void *value,
1014                         size_t size, int flags)
1015 {
1016         struct dentry *workdir = ovl_workdir(dentry);
1017         struct inode *realinode = ovl_inode_real(inode);
1018         struct posix_acl *acl = NULL;
1019         int err;
1020
1021         /* Check that everything is OK before copy-up */
1022         if (value) {
1023                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
1024                 if (IS_ERR(acl))
1025                         return PTR_ERR(acl);
1026         }
1027         err = -EOPNOTSUPP;
1028         if (!IS_POSIXACL(d_inode(workdir)))
1029                 goto out_acl_release;
1030         if (!realinode->i_op->set_acl)
1031                 goto out_acl_release;
1032         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1033                 err = acl ? -EACCES : 0;
1034                 goto out_acl_release;
1035         }
1036         err = -EPERM;
1037         if (!inode_owner_or_capable(&init_user_ns, inode))
1038                 goto out_acl_release;
1039
1040         posix_acl_release(acl);
1041
1042         /*
1043          * Check if sgid bit needs to be cleared (actual setacl operation will
1044          * be done with mounter's capabilities and so that won't do it for us).
1045          */
1046         if (unlikely(inode->i_mode & S_ISGID) &&
1047             handler->flags == ACL_TYPE_ACCESS &&
1048             !in_group_p(inode->i_gid) &&
1049             !capable_wrt_inode_uidgid(&init_user_ns, inode, CAP_FSETID)) {
1050                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
1051
1052                 err = ovl_setattr(&init_user_ns, dentry, &iattr);
1053                 if (err)
1054                         return err;
1055         }
1056
1057         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
1058         return err;
1059
1060 out_acl_release:
1061         posix_acl_release(acl);
1062         return err;
1063 }
1064
1065 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1066                              struct dentry *dentry, struct inode *inode,
1067                              const char *name, void *buffer, size_t size)
1068 {
1069         return -EOPNOTSUPP;
1070 }
1071
1072 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1073                              struct user_namespace *mnt_userns,
1074                              struct dentry *dentry, struct inode *inode,
1075                              const char *name, const void *value,
1076                              size_t size, int flags)
1077 {
1078         return -EOPNOTSUPP;
1079 }
1080
1081 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1082                                struct dentry *dentry, struct inode *inode,
1083                                const char *name, void *buffer, size_t size)
1084 {
1085         return ovl_xattr_get(dentry, inode, name, buffer, size);
1086 }
1087
1088 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1089                                struct user_namespace *mnt_userns,
1090                                struct dentry *dentry, struct inode *inode,
1091                                const char *name, const void *value,
1092                                size_t size, int flags)
1093 {
1094         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1095 }
1096
1097 static const struct xattr_handler __maybe_unused
1098 ovl_posix_acl_access_xattr_handler = {
1099         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1100         .flags = ACL_TYPE_ACCESS,
1101         .get = ovl_posix_acl_xattr_get,
1102         .set = ovl_posix_acl_xattr_set,
1103 };
1104
1105 static const struct xattr_handler __maybe_unused
1106 ovl_posix_acl_default_xattr_handler = {
1107         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1108         .flags = ACL_TYPE_DEFAULT,
1109         .get = ovl_posix_acl_xattr_get,
1110         .set = ovl_posix_acl_xattr_set,
1111 };
1112
1113 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1114         .prefix = OVL_XATTR_TRUSTED_PREFIX,
1115         .get = ovl_own_xattr_get,
1116         .set = ovl_own_xattr_set,
1117 };
1118
1119 static const struct xattr_handler ovl_own_user_xattr_handler = {
1120         .prefix = OVL_XATTR_USER_PREFIX,
1121         .get = ovl_own_xattr_get,
1122         .set = ovl_own_xattr_set,
1123 };
1124
1125 static const struct xattr_handler ovl_other_xattr_handler = {
1126         .prefix = "", /* catch all */
1127         .get = ovl_other_xattr_get,
1128         .set = ovl_other_xattr_set,
1129 };
1130
1131 static const struct xattr_handler *ovl_trusted_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_trusted_xattr_handler,
1137         &ovl_other_xattr_handler,
1138         NULL
1139 };
1140
1141 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1142 #ifdef CONFIG_FS_POSIX_ACL
1143         &ovl_posix_acl_access_xattr_handler,
1144         &ovl_posix_acl_default_xattr_handler,
1145 #endif
1146         &ovl_own_user_xattr_handler,
1147         &ovl_other_xattr_handler,
1148         NULL
1149 };
1150
1151 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1152                           struct inode **ptrap, const char *name)
1153 {
1154         struct inode *trap;
1155         int err;
1156
1157         trap = ovl_get_trap_inode(sb, dir);
1158         err = PTR_ERR_OR_ZERO(trap);
1159         if (err) {
1160                 if (err == -ELOOP)
1161                         pr_err("conflicting %s path\n", name);
1162                 return err;
1163         }
1164
1165         *ptrap = trap;
1166         return 0;
1167 }
1168
1169 /*
1170  * Determine how we treat concurrent use of upperdir/workdir based on the
1171  * index feature. This is papering over mount leaks of container runtimes,
1172  * for example, an old overlay mount is leaked and now its upperdir is
1173  * attempted to be used as a lower layer in a new overlay mount.
1174  */
1175 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1176 {
1177         if (ofs->config.index) {
1178                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1179                        name);
1180                 return -EBUSY;
1181         } else {
1182                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1183                         name);
1184                 return 0;
1185         }
1186 }
1187
1188 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1189                          struct ovl_layer *upper_layer, struct path *upperpath)
1190 {
1191         struct vfsmount *upper_mnt;
1192         int err;
1193
1194         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1195         if (err)
1196                 goto out;
1197
1198         /* Upperdir path should not be r/o */
1199         if (__mnt_is_readonly(upperpath->mnt)) {
1200                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1201                 err = -EINVAL;
1202                 goto out;
1203         }
1204
1205         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1206         if (err)
1207                 goto out;
1208
1209         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1210                              "upperdir");
1211         if (err)
1212                 goto out;
1213
1214         upper_mnt = clone_private_mount(upperpath);
1215         err = PTR_ERR(upper_mnt);
1216         if (IS_ERR(upper_mnt)) {
1217                 pr_err("failed to clone upperpath\n");
1218                 goto out;
1219         }
1220
1221         /* Don't inherit atime flags */
1222         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1223         upper_layer->mnt = upper_mnt;
1224         upper_layer->idx = 0;
1225         upper_layer->fsid = 0;
1226
1227         /*
1228          * Inherit SB_NOSEC flag from upperdir.
1229          *
1230          * This optimization changes behavior when a security related attribute
1231          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1232          * okay because we don't yet have guarantees in that case, but it will
1233          * need careful treatment once we want to honour changes to underlying
1234          * filesystems.
1235          */
1236         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1237                 sb->s_flags |= SB_NOSEC;
1238
1239         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1240                 ofs->upperdir_locked = true;
1241         } else {
1242                 err = ovl_report_in_use(ofs, "upperdir");
1243                 if (err)
1244                         goto out;
1245         }
1246
1247         err = 0;
1248 out:
1249         return err;
1250 }
1251
1252 /*
1253  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1254  * negative values if error is encountered.
1255  */
1256 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
1257 {
1258         struct dentry *workdir = ofs->workdir;
1259         struct inode *dir = d_inode(workdir);
1260         struct dentry *temp;
1261         struct dentry *dest;
1262         struct dentry *whiteout;
1263         struct name_snapshot name;
1264         int err;
1265
1266         inode_lock_nested(dir, I_MUTEX_PARENT);
1267
1268         temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
1269         err = PTR_ERR(temp);
1270         if (IS_ERR(temp))
1271                 goto out_unlock;
1272
1273         dest = ovl_lookup_temp(ofs, workdir);
1274         err = PTR_ERR(dest);
1275         if (IS_ERR(dest)) {
1276                 dput(temp);
1277                 goto out_unlock;
1278         }
1279
1280         /* Name is inline and stable - using snapshot as a copy helper */
1281         take_dentry_name_snapshot(&name, temp);
1282         err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
1283         if (err) {
1284                 if (err == -EINVAL)
1285                         err = 0;
1286                 goto cleanup_temp;
1287         }
1288
1289         whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
1290         err = PTR_ERR(whiteout);
1291         if (IS_ERR(whiteout))
1292                 goto cleanup_temp;
1293
1294         err = ovl_is_whiteout(whiteout);
1295
1296         /* Best effort cleanup of whiteout and temp file */
1297         if (err)
1298                 ovl_cleanup(ofs, dir, whiteout);
1299         dput(whiteout);
1300
1301 cleanup_temp:
1302         ovl_cleanup(ofs, dir, temp);
1303         release_dentry_name_snapshot(&name);
1304         dput(temp);
1305         dput(dest);
1306
1307 out_unlock:
1308         inode_unlock(dir);
1309
1310         return err;
1311 }
1312
1313 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
1314                                            struct dentry *parent,
1315                                            const char *name, umode_t mode)
1316 {
1317         size_t len = strlen(name);
1318         struct dentry *child;
1319
1320         inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1321         child = ovl_lookup_upper(ofs, name, parent, len);
1322         if (!IS_ERR(child) && !child->d_inode)
1323                 child = ovl_create_real(ofs, parent->d_inode, child,
1324                                         OVL_CATTR(mode));
1325         inode_unlock(parent->d_inode);
1326         dput(parent);
1327
1328         return child;
1329 }
1330
1331 /*
1332  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1333  * present.
1334  */
1335 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1336 {
1337         unsigned int ctr;
1338         struct dentry *d = dget(ofs->workbasedir);
1339         static const char *const volatile_path[] = {
1340                 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1341         };
1342         const char *const *name = volatile_path;
1343
1344         for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1345                 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1346                 if (IS_ERR(d))
1347                         return PTR_ERR(d);
1348         }
1349         dput(d);
1350         return 0;
1351 }
1352
1353 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1354                             struct path *workpath)
1355 {
1356         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1357         struct dentry *temp, *workdir;
1358         bool rename_whiteout;
1359         bool d_type;
1360         int fh_type;
1361         int err;
1362
1363         err = mnt_want_write(mnt);
1364         if (err)
1365                 return err;
1366
1367         workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1368         err = PTR_ERR(workdir);
1369         if (IS_ERR_OR_NULL(workdir))
1370                 goto out;
1371
1372         ofs->workdir = workdir;
1373
1374         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1375         if (err)
1376                 goto out;
1377
1378         /*
1379          * Upper should support d_type, else whiteouts are visible.  Given
1380          * workdir and upper are on same fs, we can do iterate_dir() on
1381          * workdir. This check requires successful creation of workdir in
1382          * previous step.
1383          */
1384         err = ovl_check_d_type_supported(workpath);
1385         if (err < 0)
1386                 goto out;
1387
1388         d_type = err;
1389         if (!d_type)
1390                 pr_warn("upper fs needs to support d_type.\n");
1391
1392         /* Check if upper/work fs supports O_TMPFILE */
1393         temp = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
1394         ofs->tmpfile = !IS_ERR(temp);
1395         if (ofs->tmpfile)
1396                 dput(temp);
1397         else
1398                 pr_warn("upper fs does not support tmpfile.\n");
1399
1400
1401         /* Check if upper/work fs supports RENAME_WHITEOUT */
1402         err = ovl_check_rename_whiteout(ofs);
1403         if (err < 0)
1404                 goto out;
1405
1406         rename_whiteout = err;
1407         if (!rename_whiteout)
1408                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1409
1410         /*
1411          * Check if upper/work fs supports (trusted|user).overlay.* xattr
1412          */
1413         err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1414         if (err) {
1415                 ofs->noxattr = true;
1416                 if (ofs->config.index || ofs->config.metacopy) {
1417                         ofs->config.index = false;
1418                         ofs->config.metacopy = false;
1419                         pr_warn("upper fs does not support xattr, falling back to index=off,metacopy=off.\n");
1420                 }
1421                 /*
1422                  * xattr support is required for persistent st_ino.
1423                  * Without persistent st_ino, xino=auto falls back to xino=off.
1424                  */
1425                 if (ofs->config.xino == OVL_XINO_AUTO) {
1426                         ofs->config.xino = OVL_XINO_OFF;
1427                         pr_warn("upper fs does not support xattr, falling back to xino=off.\n");
1428                 }
1429                 err = 0;
1430         } else {
1431                 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1432         }
1433
1434         /*
1435          * We allowed sub-optimal upper fs configuration and don't want to break
1436          * users over kernel upgrade, but we never allowed remote upper fs, so
1437          * we can enforce strict requirements for remote upper fs.
1438          */
1439         if (ovl_dentry_remote(ofs->workdir) &&
1440             (!d_type || !rename_whiteout || ofs->noxattr)) {
1441                 pr_err("upper fs missing required features.\n");
1442                 err = -EINVAL;
1443                 goto out;
1444         }
1445
1446         /*
1447          * For volatile mount, create a incompat/volatile/dirty file to keep
1448          * track of it.
1449          */
1450         if (ofs->config.ovl_volatile) {
1451                 err = ovl_create_volatile_dirty(ofs);
1452                 if (err < 0) {
1453                         pr_err("Failed to create volatile/dirty file.\n");
1454                         goto out;
1455                 }
1456         }
1457
1458         /* Check if upper/work fs supports file handles */
1459         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1460         if (ofs->config.index && !fh_type) {
1461                 ofs->config.index = false;
1462                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1463         }
1464
1465         /* Check if upper fs has 32bit inode numbers */
1466         if (fh_type != FILEID_INO32_GEN)
1467                 ofs->xino_mode = -1;
1468
1469         /* NFS export of r/w mount depends on index */
1470         if (ofs->config.nfs_export && !ofs->config.index) {
1471                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1472                 ofs->config.nfs_export = false;
1473         }
1474 out:
1475         mnt_drop_write(mnt);
1476         return err;
1477 }
1478
1479 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1480                            struct path *upperpath)
1481 {
1482         int err;
1483         struct path workpath = { };
1484
1485         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1486         if (err)
1487                 goto out;
1488
1489         err = -EINVAL;
1490         if (upperpath->mnt != workpath.mnt) {
1491                 pr_err("workdir and upperdir must reside under the same mount\n");
1492                 goto out;
1493         }
1494         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1495                 pr_err("workdir and upperdir must be separate subtrees\n");
1496                 goto out;
1497         }
1498
1499         ofs->workbasedir = dget(workpath.dentry);
1500
1501         if (ovl_inuse_trylock(ofs->workbasedir)) {
1502                 ofs->workdir_locked = true;
1503         } else {
1504                 err = ovl_report_in_use(ofs, "workdir");
1505                 if (err)
1506                         goto out;
1507         }
1508
1509         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1510                              "workdir");
1511         if (err)
1512                 goto out;
1513
1514         err = ovl_make_workdir(sb, ofs, &workpath);
1515
1516 out:
1517         path_put(&workpath);
1518
1519         return err;
1520 }
1521
1522 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1523                             struct ovl_entry *oe, struct path *upperpath)
1524 {
1525         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1526         struct dentry *indexdir;
1527         int err;
1528
1529         err = mnt_want_write(mnt);
1530         if (err)
1531                 return err;
1532
1533         /* Verify lower root is upper root origin */
1534         err = ovl_verify_origin(ofs, upperpath->dentry,
1535                                 oe->lowerstack[0].dentry, true);
1536         if (err) {
1537                 pr_err("failed to verify upper root origin\n");
1538                 goto out;
1539         }
1540
1541         /* index dir will act also as workdir */
1542         iput(ofs->workdir_trap);
1543         ofs->workdir_trap = NULL;
1544         dput(ofs->workdir);
1545         ofs->workdir = NULL;
1546         indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1547         if (IS_ERR(indexdir)) {
1548                 err = PTR_ERR(indexdir);
1549         } else if (indexdir) {
1550                 ofs->indexdir = indexdir;
1551                 ofs->workdir = dget(indexdir);
1552
1553                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1554                                      "indexdir");
1555                 if (err)
1556                         goto out;
1557
1558                 /*
1559                  * Verify upper root is exclusively associated with index dir.
1560                  * Older kernels stored upper fh in ".overlay.origin"
1561                  * xattr. If that xattr exists, verify that it is a match to
1562                  * upper dir file handle. In any case, verify or set xattr
1563                  * ".overlay.upper" to indicate that index may have
1564                  * directory entries.
1565                  */
1566                 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1567                         err = ovl_verify_set_fh(ofs, ofs->indexdir,
1568                                                 OVL_XATTR_ORIGIN,
1569                                                 upperpath->dentry, true, false);
1570                         if (err)
1571                                 pr_err("failed to verify index dir 'origin' xattr\n");
1572                 }
1573                 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1574                                        true);
1575                 if (err)
1576                         pr_err("failed to verify index dir 'upper' xattr\n");
1577
1578                 /* Cleanup bad/stale/orphan index entries */
1579                 if (!err)
1580                         err = ovl_indexdir_cleanup(ofs);
1581         }
1582         if (err || !ofs->indexdir)
1583                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1584
1585 out:
1586         mnt_drop_write(mnt);
1587         return err;
1588 }
1589
1590 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1591 {
1592         unsigned int i;
1593
1594         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1595                 return true;
1596
1597         /*
1598          * We allow using single lower with null uuid for index and nfs_export
1599          * for example to support those features with single lower squashfs.
1600          * To avoid regressions in setups of overlay with re-formatted lower
1601          * squashfs, do not allow decoding origin with lower null uuid unless
1602          * user opted-in to one of the new features that require following the
1603          * lower inode of non-dir upper.
1604          */
1605         if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
1606                 return false;
1607
1608         for (i = 0; i < ofs->numfs; i++) {
1609                 /*
1610                  * We use uuid to associate an overlay lower file handle with a
1611                  * lower layer, so we can accept lower fs with null uuid as long
1612                  * as all lower layers with null uuid are on the same fs.
1613                  * if we detect multiple lower fs with the same uuid, we
1614                  * disable lower file handle decoding on all of them.
1615                  */
1616                 if (ofs->fs[i].is_lower &&
1617                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1618                         ofs->fs[i].bad_uuid = true;
1619                         return false;
1620                 }
1621         }
1622         return true;
1623 }
1624
1625 /* Get a unique fsid for the layer */
1626 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1627 {
1628         struct super_block *sb = path->mnt->mnt_sb;
1629         unsigned int i;
1630         dev_t dev;
1631         int err;
1632         bool bad_uuid = false;
1633         bool warn = false;
1634
1635         for (i = 0; i < ofs->numfs; i++) {
1636                 if (ofs->fs[i].sb == sb)
1637                         return i;
1638         }
1639
1640         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1641                 bad_uuid = true;
1642                 if (ofs->config.xino == OVL_XINO_AUTO) {
1643                         ofs->config.xino = OVL_XINO_OFF;
1644                         warn = true;
1645                 }
1646                 if (ofs->config.index || ofs->config.nfs_export) {
1647                         ofs->config.index = false;
1648                         ofs->config.nfs_export = false;
1649                         warn = true;
1650                 }
1651                 if (warn) {
1652                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1653                                 uuid_is_null(&sb->s_uuid) ? "null" :
1654                                                             "conflicting",
1655                                 path->dentry, ovl_xino_str[ofs->config.xino]);
1656                 }
1657         }
1658
1659         err = get_anon_bdev(&dev);
1660         if (err) {
1661                 pr_err("failed to get anonymous bdev for lowerpath\n");
1662                 return err;
1663         }
1664
1665         ofs->fs[ofs->numfs].sb = sb;
1666         ofs->fs[ofs->numfs].pseudo_dev = dev;
1667         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1668
1669         return ofs->numfs++;
1670 }
1671
1672 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1673                           struct path *stack, unsigned int numlower,
1674                           struct ovl_layer *layers)
1675 {
1676         int err;
1677         unsigned int i;
1678
1679         err = -ENOMEM;
1680         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1681         if (ofs->fs == NULL)
1682                 goto out;
1683
1684         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1685         ofs->numfs++;
1686
1687         /*
1688          * All lower layers that share the same fs as upper layer, use the same
1689          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1690          * only overlay to simplify ovl_fs_free().
1691          * is_lower will be set if upper fs is shared with a lower layer.
1692          */
1693         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1694         if (err) {
1695                 pr_err("failed to get anonymous bdev for upper fs\n");
1696                 goto out;
1697         }
1698
1699         if (ovl_upper_mnt(ofs)) {
1700                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1701                 ofs->fs[0].is_lower = false;
1702         }
1703
1704         for (i = 0; i < numlower; i++) {
1705                 struct vfsmount *mnt;
1706                 struct inode *trap;
1707                 int fsid;
1708
1709                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1710                 if (err < 0)
1711                         goto out;
1712
1713                 /*
1714                  * Check if lower root conflicts with this overlay layers before
1715                  * checking if it is in-use as upperdir/workdir of "another"
1716                  * mount, because we do not bother to check in ovl_is_inuse() if
1717                  * the upperdir/workdir is in fact in-use by our
1718                  * upperdir/workdir.
1719                  */
1720                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1721                 if (err)
1722                         goto out;
1723
1724                 if (ovl_is_inuse(stack[i].dentry)) {
1725                         err = ovl_report_in_use(ofs, "lowerdir");
1726                         if (err) {
1727                                 iput(trap);
1728                                 goto out;
1729                         }
1730                 }
1731
1732                 mnt = clone_private_mount(&stack[i]);
1733                 err = PTR_ERR(mnt);
1734                 if (IS_ERR(mnt)) {
1735                         pr_err("failed to clone lowerpath\n");
1736                         iput(trap);
1737                         goto out;
1738                 }
1739
1740                 /*
1741                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1742                  * will fail instead of modifying lower fs.
1743                  */
1744                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1745
1746                 layers[ofs->numlayer].trap = trap;
1747                 layers[ofs->numlayer].mnt = mnt;
1748                 layers[ofs->numlayer].idx = ofs->numlayer;
1749                 layers[ofs->numlayer].fsid = fsid;
1750                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1751                 ofs->numlayer++;
1752                 ofs->fs[fsid].is_lower = true;
1753         }
1754
1755         /*
1756          * When all layers on same fs, overlay can use real inode numbers.
1757          * With mount option "xino=<on|auto>", mounter declares that there are
1758          * enough free high bits in underlying fs to hold the unique fsid.
1759          * If overlayfs does encounter underlying inodes using the high xino
1760          * bits reserved for fsid, it emits a warning and uses the original
1761          * inode number or a non persistent inode number allocated from a
1762          * dedicated range.
1763          */
1764         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1765                 if (ofs->config.xino == OVL_XINO_ON)
1766                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1767                 ofs->xino_mode = 0;
1768         } else if (ofs->config.xino == OVL_XINO_OFF) {
1769                 ofs->xino_mode = -1;
1770         } else if (ofs->xino_mode < 0) {
1771                 /*
1772                  * This is a roundup of number of bits needed for encoding
1773                  * fsid, where fsid 0 is reserved for upper fs (even with
1774                  * lower only overlay) +1 extra bit is reserved for the non
1775                  * persistent inode number range that is used for resolving
1776                  * xino lower bits overflow.
1777                  */
1778                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1779                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1780         }
1781
1782         if (ofs->xino_mode > 0) {
1783                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1784                         ofs->xino_mode);
1785         }
1786
1787         err = 0;
1788 out:
1789         return err;
1790 }
1791
1792 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1793                                 const char *lower, unsigned int numlower,
1794                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1795 {
1796         int err;
1797         struct path *stack = NULL;
1798         unsigned int i;
1799         struct ovl_entry *oe;
1800
1801         if (!ofs->config.upperdir && numlower == 1) {
1802                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1803                 return ERR_PTR(-EINVAL);
1804         }
1805
1806         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1807         if (!stack)
1808                 return ERR_PTR(-ENOMEM);
1809
1810         err = -EINVAL;
1811         for (i = 0; i < numlower; i++) {
1812                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1813                 if (err)
1814                         goto out_err;
1815
1816                 lower = strchr(lower, '\0') + 1;
1817         }
1818
1819         err = -EINVAL;
1820         sb->s_stack_depth++;
1821         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1822                 pr_err("maximum fs stacking depth exceeded\n");
1823                 goto out_err;
1824         }
1825
1826         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1827         if (err)
1828                 goto out_err;
1829
1830         err = -ENOMEM;
1831         oe = ovl_alloc_entry(numlower);
1832         if (!oe)
1833                 goto out_err;
1834
1835         for (i = 0; i < numlower; i++) {
1836                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1837                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1838         }
1839
1840 out:
1841         for (i = 0; i < numlower; i++)
1842                 path_put(&stack[i]);
1843         kfree(stack);
1844
1845         return oe;
1846
1847 out_err:
1848         oe = ERR_PTR(err);
1849         goto out;
1850 }
1851
1852 /*
1853  * Check if this layer root is a descendant of:
1854  * - another layer of this overlayfs instance
1855  * - upper/work dir of any overlayfs instance
1856  */
1857 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1858                            struct dentry *dentry, const char *name,
1859                            bool is_lower)
1860 {
1861         struct dentry *next = dentry, *parent;
1862         int err = 0;
1863
1864         if (!dentry)
1865                 return 0;
1866
1867         parent = dget_parent(next);
1868
1869         /* Walk back ancestors to root (inclusive) looking for traps */
1870         while (!err && parent != next) {
1871                 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1872                         err = -ELOOP;
1873                         pr_err("overlapping %s path\n", name);
1874                 } else if (ovl_is_inuse(parent)) {
1875                         err = ovl_report_in_use(ofs, name);
1876                 }
1877                 next = parent;
1878                 parent = dget_parent(next);
1879                 dput(next);
1880         }
1881
1882         dput(parent);
1883
1884         return err;
1885 }
1886
1887 /*
1888  * Check if any of the layers or work dirs overlap.
1889  */
1890 static int ovl_check_overlapping_layers(struct super_block *sb,
1891                                         struct ovl_fs *ofs)
1892 {
1893         int i, err;
1894
1895         if (ovl_upper_mnt(ofs)) {
1896                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1897                                       "upperdir", false);
1898                 if (err)
1899                         return err;
1900
1901                 /*
1902                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1903                  * this instance and covers overlapping work and index dirs,
1904                  * unless work or index dir have been moved since created inside
1905                  * workbasedir.  In that case, we already have their traps in
1906                  * inode cache and we will catch that case on lookup.
1907                  */
1908                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1909                                       false);
1910                 if (err)
1911                         return err;
1912         }
1913
1914         for (i = 1; i < ofs->numlayer; i++) {
1915                 err = ovl_check_layer(sb, ofs,
1916                                       ofs->layers[i].mnt->mnt_root,
1917                                       "lowerdir", true);
1918                 if (err)
1919                         return err;
1920         }
1921
1922         return 0;
1923 }
1924
1925 static struct dentry *ovl_get_root(struct super_block *sb,
1926                                    struct dentry *upperdentry,
1927                                    struct ovl_entry *oe)
1928 {
1929         struct dentry *root;
1930         struct ovl_path *lowerpath = &oe->lowerstack[0];
1931         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1932         int fsid = lowerpath->layer->fsid;
1933         struct ovl_inode_params oip = {
1934                 .upperdentry = upperdentry,
1935                 .lowerpath = lowerpath,
1936         };
1937
1938         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1939         if (!root)
1940                 return NULL;
1941
1942         root->d_fsdata = oe;
1943
1944         if (upperdentry) {
1945                 /* Root inode uses upper st_ino/i_ino */
1946                 ino = d_inode(upperdentry)->i_ino;
1947                 fsid = 0;
1948                 ovl_dentry_set_upper_alias(root);
1949                 if (ovl_is_impuredir(sb, upperdentry))
1950                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1951         }
1952
1953         /* Root is always merge -> can have whiteouts */
1954         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1955         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1956         ovl_set_upperdata(d_inode(root));
1957         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1958         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1959
1960         return root;
1961 }
1962
1963 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1964 {
1965         struct path upperpath = { };
1966         struct dentry *root_dentry;
1967         struct ovl_entry *oe;
1968         struct ovl_fs *ofs;
1969         struct ovl_layer *layers;
1970         struct cred *cred;
1971         char *splitlower = NULL;
1972         unsigned int numlower;
1973         int err;
1974
1975         err = -EIO;
1976         if (WARN_ON(sb->s_user_ns != current_user_ns()))
1977                 goto out;
1978
1979         sb->s_d_op = &ovl_dentry_operations;
1980
1981         err = -ENOMEM;
1982         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1983         if (!ofs)
1984                 goto out;
1985
1986         err = -ENOMEM;
1987         ofs->creator_cred = cred = prepare_creds();
1988         if (!cred)
1989                 goto out_err;
1990
1991         /* Is there a reason anyone would want not to share whiteouts? */
1992         ofs->share_whiteout = true;
1993
1994         ofs->config.index = ovl_index_def;
1995         ofs->config.uuid = true;
1996         ofs->config.nfs_export = ovl_nfs_export_def;
1997         ofs->config.xino = ovl_xino_def();
1998         ofs->config.metacopy = ovl_metacopy_def;
1999         err = ovl_parse_opt((char *) data, &ofs->config);
2000         if (err)
2001                 goto out_err;
2002
2003         err = -EINVAL;
2004         if (!ofs->config.lowerdir) {
2005                 if (!silent)
2006                         pr_err("missing 'lowerdir'\n");
2007                 goto out_err;
2008         }
2009
2010         err = -ENOMEM;
2011         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
2012         if (!splitlower)
2013                 goto out_err;
2014
2015         err = -EINVAL;
2016         numlower = ovl_split_lowerdirs(splitlower);
2017         if (numlower > OVL_MAX_STACK) {
2018                 pr_err("too many lower directories, limit is %d\n",
2019                        OVL_MAX_STACK);
2020                 goto out_err;
2021         }
2022
2023         err = -ENOMEM;
2024         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
2025         if (!layers)
2026                 goto out_err;
2027
2028         ofs->layers = layers;
2029         /* Layer 0 is reserved for upper even if there's no upper */
2030         ofs->numlayer = 1;
2031
2032         sb->s_stack_depth = 0;
2033         sb->s_maxbytes = MAX_LFS_FILESIZE;
2034         atomic_long_set(&ofs->last_ino, 1);
2035         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
2036         if (ofs->config.xino != OVL_XINO_OFF) {
2037                 ofs->xino_mode = BITS_PER_LONG - 32;
2038                 if (!ofs->xino_mode) {
2039                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
2040                         ofs->config.xino = OVL_XINO_OFF;
2041                 }
2042         }
2043
2044         /* alloc/destroy_inode needed for setting up traps in inode cache */
2045         sb->s_op = &ovl_super_operations;
2046
2047         if (ofs->config.upperdir) {
2048                 struct super_block *upper_sb;
2049
2050                 err = -EINVAL;
2051                 if (!ofs->config.workdir) {
2052                         pr_err("missing 'workdir'\n");
2053                         goto out_err;
2054                 }
2055
2056                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
2057                 if (err)
2058                         goto out_err;
2059
2060                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
2061                 if (!ovl_should_sync(ofs)) {
2062                         ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
2063                         if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
2064                                 err = -EIO;
2065                                 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
2066                                 goto out_err;
2067                         }
2068                 }
2069
2070                 err = ovl_get_workdir(sb, ofs, &upperpath);
2071                 if (err)
2072                         goto out_err;
2073
2074                 if (!ofs->workdir)
2075                         sb->s_flags |= SB_RDONLY;
2076
2077                 sb->s_stack_depth = upper_sb->s_stack_depth;
2078                 sb->s_time_gran = upper_sb->s_time_gran;
2079         }
2080         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2081         err = PTR_ERR(oe);
2082         if (IS_ERR(oe))
2083                 goto out_err;
2084
2085         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2086         if (!ovl_upper_mnt(ofs))
2087                 sb->s_flags |= SB_RDONLY;
2088
2089         if (!ofs->config.uuid && ofs->numfs > 1) {
2090                 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2091                 ofs->config.uuid = true;
2092         }
2093
2094         if (!ovl_force_readonly(ofs) && ofs->config.index) {
2095                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2096                 if (err)
2097                         goto out_free_oe;
2098
2099                 /* Force r/o mount with no index dir */
2100                 if (!ofs->indexdir)
2101                         sb->s_flags |= SB_RDONLY;
2102         }
2103
2104         err = ovl_check_overlapping_layers(sb, ofs);
2105         if (err)
2106                 goto out_free_oe;
2107
2108         /* Show index=off in /proc/mounts for forced r/o mount */
2109         if (!ofs->indexdir) {
2110                 ofs->config.index = false;
2111                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2112                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2113                         ofs->config.nfs_export = false;
2114                 }
2115         }
2116
2117         if (ofs->config.metacopy && ofs->config.nfs_export) {
2118                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2119                 ofs->config.nfs_export = false;
2120         }
2121
2122         if (ofs->config.nfs_export)
2123                 sb->s_export_op = &ovl_export_operations;
2124
2125         /* Never override disk quota limits or use reserved space */
2126         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2127
2128         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2129         sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2130                 ovl_trusted_xattr_handlers;
2131         sb->s_fs_info = ofs;
2132         sb->s_flags |= SB_POSIXACL;
2133         sb->s_iflags |= SB_I_SKIP_SYNC;
2134
2135         err = -ENOMEM;
2136         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2137         if (!root_dentry)
2138                 goto out_free_oe;
2139
2140         mntput(upperpath.mnt);
2141         kfree(splitlower);
2142
2143         sb->s_root = root_dentry;
2144
2145         return 0;
2146
2147 out_free_oe:
2148         ovl_entry_stack_free(oe);
2149         kfree(oe);
2150 out_err:
2151         kfree(splitlower);
2152         path_put(&upperpath);
2153         ovl_free_fs(ofs);
2154 out:
2155         return err;
2156 }
2157
2158 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2159                                 const char *dev_name, void *raw_data)
2160 {
2161         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2162 }
2163
2164 static struct file_system_type ovl_fs_type = {
2165         .owner          = THIS_MODULE,
2166         .name           = "overlay",
2167         .fs_flags       = FS_USERNS_MOUNT,
2168         .mount          = ovl_mount,
2169         .kill_sb        = kill_anon_super,
2170 };
2171 MODULE_ALIAS_FS("overlay");
2172
2173 static void ovl_inode_init_once(void *foo)
2174 {
2175         struct ovl_inode *oi = foo;
2176
2177         inode_init_once(&oi->vfs_inode);
2178 }
2179
2180 static int __init ovl_init(void)
2181 {
2182         int err;
2183
2184         ovl_inode_cachep = kmem_cache_create("ovl_inode",
2185                                              sizeof(struct ovl_inode), 0,
2186                                              (SLAB_RECLAIM_ACCOUNT|
2187                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2188                                              ovl_inode_init_once);
2189         if (ovl_inode_cachep == NULL)
2190                 return -ENOMEM;
2191
2192         err = ovl_aio_request_cache_init();
2193         if (!err) {
2194                 err = register_filesystem(&ovl_fs_type);
2195                 if (!err)
2196                         return 0;
2197
2198                 ovl_aio_request_cache_destroy();
2199         }
2200         kmem_cache_destroy(ovl_inode_cachep);
2201
2202         return err;
2203 }
2204
2205 static void __exit ovl_exit(void)
2206 {
2207         unregister_filesystem(&ovl_fs_type);
2208
2209         /*
2210          * Make sure all delayed rcu free inodes are flushed before we
2211          * destroy cache.
2212          */
2213         rcu_barrier();
2214         kmem_cache_destroy(ovl_inode_cachep);
2215         ovl_aio_request_cache_destroy();
2216 }
2217
2218 module_init(ovl_init);
2219 module_exit(ovl_exit);