fs/overlayfs/super.c

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