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