fs/overlayfs/super.c

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