fs/overlayfs/super.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <uapi/linux/magic.h>
  11 #include <linux/fs.h>
  12 #include <linux/namei.h>
  13 #include <linux/xattr.h>
  14 #include <linux/mount.h>
  15 #include <linux/parser.h>
  16 #include <linux/module.h>
  17 #include <linux/statfs.h>
  18 #include <linux/seq_file.h>
  19 #include <linux/posix_acl_xattr.h>
  20 #include "overlayfs.h"
  21
  22 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  23 MODULE_DESCRIPTION("Overlay filesystem");
  24 MODULE_LICENSE("GPL");
  25
  26
  27 struct ovl_dir_cache;
  28
  29 #define OVL_MAX_STACK 500
  30
  31 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  32 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  33 MODULE_PARM_DESC(ovl_redirect_dir_def,
  34                  "Default to on or off for the redirect_dir feature");
  35
  36 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
  37 module_param_named(index, ovl_index_def, bool, 0644);
  38 MODULE_PARM_DESC(ovl_index_def,
  39                  "Default to on or off for the inodes index feature");
  40
  41 static void ovl_dentry_release(struct dentry *dentry)
  42 {
  43         struct ovl_entry *oe = dentry->d_fsdata;
  44
  45         if (oe) {
  46                 unsigned int i;
  47
  48                 for (i = 0; i < oe->numlower; i++)
  49                         dput(oe->lowerstack[i].dentry);
  50                 kfree_rcu(oe, rcu);
  51         }
  52 }
  53
  54 static int ovl_check_append_only(struct inode *inode, int flag)
  55 {
  56         /*
  57          * This test was moot in vfs may_open() because overlay inode does
  58          * not have the S_APPEND flag, so re-check on real upper inode
  59          */
  60         if (IS_APPEND(inode)) {
  61                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
  62                         return -EPERM;
  63                 if (flag & O_TRUNC)
  64                         return -EPERM;
  65         }
  66
  67         return 0;
  68 }
  69
  70 static struct dentry *ovl_d_real(struct dentry *dentry,
  71                                  const struct inode *inode,
  72                                  unsigned int open_flags, unsigned int flags)
  73 {
  74         struct dentry *real;
  75         int err;
  76
  77         if (flags & D_REAL_UPPER)
  78                 return ovl_dentry_upper(dentry);
  79
  80         if (!d_is_reg(dentry)) {
  81                 if (!inode || inode == d_inode(dentry))
  82                         return dentry;
  83                 goto bug;
  84         }
  85
  86         if (open_flags) {
  87                 err = ovl_open_maybe_copy_up(dentry, open_flags);
  88                 if (err)
  89                         return ERR_PTR(err);
  90         }
  91
  92         real = ovl_dentry_upper(dentry);
  93         if (real && (!inode || inode == d_inode(real))) {
  94                 if (!inode) {
  95                         err = ovl_check_append_only(d_inode(real), open_flags);
  96                         if (err)
  97                                 return ERR_PTR(err);
  98                 }
  99                 return real;
 100         }
 101
 102         real = ovl_dentry_lower(dentry);
 103         if (!real)
 104                 goto bug;
 105
 106         /* Handle recursion */
 107         real = d_real(real, inode, open_flags, 0);
 108
 109         if (!inode || inode == d_inode(real))
 110                 return real;
 111 bug:
 112         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
 113              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 114         return dentry;
 115 }
 116
 117 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 118 {
 119         struct ovl_entry *oe = dentry->d_fsdata;
 120         unsigned int i;
 121         int ret = 1;
 122
 123         for (i = 0; i < oe->numlower; i++) {
 124                 struct dentry *d = oe->lowerstack[i].dentry;
 125
 126                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 127                         ret = d->d_op->d_revalidate(d, flags);
 128                         if (ret < 0)
 129                                 return ret;
 130                         if (!ret) {
 131                                 if (!(flags & LOOKUP_RCU))
 132                                         d_invalidate(d);
 133                                 return -ESTALE;
 134                         }
 135                 }
 136         }
 137         return 1;
 138 }
 139
 140 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 141 {
 142         struct ovl_entry *oe = dentry->d_fsdata;
 143         unsigned int i;
 144         int ret = 1;
 145
 146         for (i = 0; i < oe->numlower; i++) {
 147                 struct dentry *d = oe->lowerstack[i].dentry;
 148
 149                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 150                         ret = d->d_op->d_weak_revalidate(d, flags);
 151                         if (ret <= 0)
 152                                 break;
 153                 }
 154         }
 155         return ret;
 156 }
 157
 158 static const struct dentry_operations ovl_dentry_operations = {
 159         .d_release = ovl_dentry_release,
 160         .d_real = ovl_d_real,
 161 };
 162
 163 static const struct dentry_operations ovl_reval_dentry_operations = {
 164         .d_release = ovl_dentry_release,
 165         .d_real = ovl_d_real,
 166         .d_revalidate = ovl_dentry_revalidate,
 167         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 168 };
 169
 170 static struct kmem_cache *ovl_inode_cachep;
 171
 172 static struct inode *ovl_alloc_inode(struct super_block *sb)
 173 {
 174         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 175
 176         if (!oi)
 177                 return NULL;
 178
 179         oi->cache = NULL;
 180         oi->redirect = NULL;
 181         oi->version = 0;
 182         oi->flags = 0;
 183         oi->__upperdentry = NULL;
 184         oi->lower = NULL;
 185         mutex_init(&oi->lock);
 186
 187         return &oi->vfs_inode;
 188 }
 189
 190 static void ovl_i_callback(struct rcu_head *head)
 191 {
 192         struct inode *inode = container_of(head, struct inode, i_rcu);
 193
 194         kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
 195 }
 196
 197 static void ovl_destroy_inode(struct inode *inode)
 198 {
 199         struct ovl_inode *oi = OVL_I(inode);
 200
 201         dput(oi->__upperdentry);
 202         kfree(oi->redirect);
 203         ovl_dir_cache_free(inode);
 204         mutex_destroy(&oi->lock);
 205
 206         call_rcu(&inode->i_rcu, ovl_i_callback);
 207 }
 208
 209 static void ovl_free_fs(struct ovl_fs *ufs)
 210 {
 211         unsigned i;
 212
 213         dput(ufs->indexdir);
 214         dput(ufs->workdir);
 215         if (ufs->workdir_locked)
 216                 ovl_inuse_unlock(ufs->workbasedir);
 217         dput(ufs->workbasedir);
 218         if (ufs->upperdir_locked)
 219                 ovl_inuse_unlock(ufs->upper_mnt->mnt_root);
 220         mntput(ufs->upper_mnt);
 221         for (i = 0; i < ufs->numlower; i++) {
 222                 mntput(ufs->lower_layers[i].mnt);
 223                 free_anon_bdev(ufs->lower_layers[i].pseudo_dev);
 224         }
 225         kfree(ufs->lower_layers);
 226
 227         kfree(ufs->config.lowerdir);
 228         kfree(ufs->config.upperdir);
 229         kfree(ufs->config.workdir);
 230         if (ufs->creator_cred)
 231                 put_cred(ufs->creator_cred);
 232         kfree(ufs);
 233 }
 234
 235 static void ovl_put_super(struct super_block *sb)
 236 {
 237         struct ovl_fs *ofs = sb->s_fs_info;
 238
 239         ovl_free_fs(ofs);
 240 }
 241
 242 static int ovl_sync_fs(struct super_block *sb, int wait)
 243 {
 244         struct ovl_fs *ufs = sb->s_fs_info;
 245         struct super_block *upper_sb;
 246         int ret;
 247
 248         if (!ufs->upper_mnt)
 249                 return 0;
 250         upper_sb = ufs->upper_mnt->mnt_sb;
 251         if (!upper_sb->s_op->sync_fs)
 252                 return 0;
 253
 254         /* real inodes have already been synced by sync_filesystem(ovl_sb) */
 255         down_read(&upper_sb->s_umount);
 256         ret = upper_sb->s_op->sync_fs(upper_sb, wait);
 257         up_read(&upper_sb->s_umount);
 258         return ret;
 259 }
 260
 261 /**
 262  * ovl_statfs
 263  * @sb: The overlayfs super block
 264  * @buf: The struct kstatfs to fill in with stats
 265  *
 266  * Get the filesystem statistics.  As writes always target the upper layer
 267  * filesystem pass the statfs to the upper filesystem (if it exists)
 268  */
 269 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 270 {
 271         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 272         struct dentry *root_dentry = dentry->d_sb->s_root;
 273         struct path path;
 274         int err;
 275
 276         ovl_path_real(root_dentry, &path);
 277
 278         err = vfs_statfs(&path, buf);
 279         if (!err) {
 280                 buf->f_namelen = ofs->namelen;
 281                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 282         }
 283
 284         return err;
 285 }
 286
 287 /* Will this overlay be forced to mount/remount ro? */
 288 static bool ovl_force_readonly(struct ovl_fs *ufs)
 289 {
 290         return (!ufs->upper_mnt || !ufs->workdir);
 291 }
 292
 293 /**
 294  * ovl_show_options
 295  *
 296  * Prints the mount options for a given superblock.
 297  * Returns zero; does not fail.
 298  */
 299 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 300 {
 301         struct super_block *sb = dentry->d_sb;
 302         struct ovl_fs *ufs = sb->s_fs_info;
 303
 304         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 305         if (ufs->config.upperdir) {
 306                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 307                 seq_show_option(m, "workdir", ufs->config.workdir);
 308         }
 309         if (ufs->config.default_permissions)
 310                 seq_puts(m, ",default_permissions");
 311         if (ufs->config.redirect_dir != ovl_redirect_dir_def)
 312                 seq_printf(m, ",redirect_dir=%s",
 313                            ufs->config.redirect_dir ? "on" : "off");
 314         if (ufs->config.index != ovl_index_def)
 315                 seq_printf(m, ",index=%s",
 316                            ufs->config.index ? "on" : "off");
 317         return 0;
 318 }
 319
 320 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 321 {
 322         struct ovl_fs *ufs = sb->s_fs_info;
 323
 324         if (!(*flags & MS_RDONLY) && ovl_force_readonly(ufs))
 325                 return -EROFS;
 326
 327         return 0;
 328 }
 329
 330 static const struct super_operations ovl_super_operations = {
 331         .alloc_inode    = ovl_alloc_inode,
 332         .destroy_inode  = ovl_destroy_inode,
 333         .drop_inode     = generic_delete_inode,
 334         .put_super      = ovl_put_super,
 335         .sync_fs        = ovl_sync_fs,
 336         .statfs         = ovl_statfs,
 337         .show_options   = ovl_show_options,
 338         .remount_fs     = ovl_remount,
 339 };
 340
 341 enum {
 342         OPT_LOWERDIR,
 343         OPT_UPPERDIR,
 344         OPT_WORKDIR,
 345         OPT_DEFAULT_PERMISSIONS,
 346         OPT_REDIRECT_DIR_ON,
 347         OPT_REDIRECT_DIR_OFF,
 348         OPT_INDEX_ON,
 349         OPT_INDEX_OFF,
 350         OPT_ERR,
 351 };
 352
 353 static const match_table_t ovl_tokens = {
 354         {OPT_LOWERDIR,                  "lowerdir=%s"},
 355         {OPT_UPPERDIR,                  "upperdir=%s"},
 356         {OPT_WORKDIR,                   "workdir=%s"},
 357         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 358         {OPT_REDIRECT_DIR_ON,           "redirect_dir=on"},
 359         {OPT_REDIRECT_DIR_OFF,          "redirect_dir=off"},
 360         {OPT_INDEX_ON,                  "index=on"},
 361         {OPT_INDEX_OFF,                 "index=off"},
 362         {OPT_ERR,                       NULL}
 363 };
 364
 365 static char *ovl_next_opt(char **s)
 366 {
 367         char *sbegin = *s;
 368         char *p;
 369
 370         if (sbegin == NULL)
 371                 return NULL;
 372
 373         for (p = sbegin; *p; p++) {
 374                 if (*p == '\\') {
 375                         p++;
 376                         if (!*p)
 377                                 break;
 378                 } else if (*p == ',') {
 379                         *p = '\0';
 380                         *s = p + 1;
 381                         return sbegin;
 382                 }
 383         }
 384         *s = NULL;
 385         return sbegin;
 386 }
 387
 388 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 389 {
 390         char *p;
 391
 392         while ((p = ovl_next_opt(&opt)) != NULL) {
 393                 int token;
 394                 substring_t args[MAX_OPT_ARGS];
 395
 396                 if (!*p)
 397                         continue;
 398
 399                 token = match_token(p, ovl_tokens, args);
 400                 switch (token) {
 401                 case OPT_UPPERDIR:
 402                         kfree(config->upperdir);
 403                         config->upperdir = match_strdup(&args[0]);
 404                         if (!config->upperdir)
 405                                 return -ENOMEM;
 406                         break;
 407
 408                 case OPT_LOWERDIR:
 409                         kfree(config->lowerdir);
 410                         config->lowerdir = match_strdup(&args[0]);
 411                         if (!config->lowerdir)
 412                                 return -ENOMEM;
 413                         break;
 414
 415                 case OPT_WORKDIR:
 416                         kfree(config->workdir);
 417                         config->workdir = match_strdup(&args[0]);
 418                         if (!config->workdir)
 419                                 return -ENOMEM;
 420                         break;
 421
 422                 case OPT_DEFAULT_PERMISSIONS:
 423                         config->default_permissions = true;
 424                         break;
 425
 426                 case OPT_REDIRECT_DIR_ON:
 427                         config->redirect_dir = true;
 428                         break;
 429
 430                 case OPT_REDIRECT_DIR_OFF:
 431                         config->redirect_dir = false;
 432                         break;
 433
 434                 case OPT_INDEX_ON:
 435                         config->index = true;
 436                         break;
 437
 438                 case OPT_INDEX_OFF:
 439                         config->index = false;
 440                         break;
 441
 442                 default:
 443                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 444                         return -EINVAL;
 445                 }
 446         }
 447
 448         /* Workdir is useless in non-upper mount */
 449         if (!config->upperdir && config->workdir) {
 450                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 451                         config->workdir);
 452                 kfree(config->workdir);
 453                 config->workdir = NULL;
 454         }
 455
 456         return 0;
 457 }
 458
 459 #define OVL_WORKDIR_NAME "work"
 460 #define OVL_INDEXDIR_NAME "index"
 461
 462 static struct dentry *ovl_workdir_create(struct super_block *sb,
 463                                          struct ovl_fs *ufs,
 464                                          struct dentry *dentry,
 465                                          const char *name, bool persist)
 466 {
 467         struct inode *dir = dentry->d_inode;
 468         struct vfsmount *mnt = ufs->upper_mnt;
 469         struct dentry *work;
 470         int err;
 471         bool retried = false;
 472         bool locked = false;
 473
 474         err = mnt_want_write(mnt);
 475         if (err)
 476                 goto out_err;
 477
 478         inode_lock_nested(dir, I_MUTEX_PARENT);
 479         locked = true;
 480
 481 retry:
 482         work = lookup_one_len(name, dentry, strlen(name));
 483
 484         if (!IS_ERR(work)) {
 485                 struct iattr attr = {
 486                         .ia_valid = ATTR_MODE,
 487                         .ia_mode = S_IFDIR | 0,
 488                 };
 489
 490                 if (work->d_inode) {
 491                         err = -EEXIST;
 492                         if (retried)
 493                                 goto out_dput;
 494
 495                         if (persist)
 496                                 goto out_unlock;
 497
 498                         retried = true;
 499                         ovl_workdir_cleanup(dir, mnt, work, 0);
 500                         dput(work);
 501                         goto retry;
 502                 }
 503
 504                 err = ovl_create_real(dir, work,
 505                                       &(struct cattr){.mode = S_IFDIR | 0},
 506                                       NULL, true);
 507                 if (err)
 508                         goto out_dput;
 509
 510                 /*
 511                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 512                  *
 513                  * a) success (there was a POSIX ACL xattr and was removed)
 514                  * b) -ENODATA (there was no POSIX ACL xattr)
 515                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 516                  *
 517                  * There are various other error values that could effectively
 518                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 519                  * if the xattr name is too long), but the set of filesystems
 520                  * allowed as upper are limited to "normal" ones, where checking
 521                  * for the above two errors is sufficient.
 522                  */
 523                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 524                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 525                         goto out_dput;
 526
 527                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 528                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 529                         goto out_dput;
 530
 531                 /* Clear any inherited mode bits */
 532                 inode_lock(work->d_inode);
 533                 err = notify_change(work, &attr, NULL);
 534                 inode_unlock(work->d_inode);
 535                 if (err)
 536                         goto out_dput;
 537         } else {
 538                 err = PTR_ERR(work);
 539                 goto out_err;
 540         }
 541 out_unlock:
 542         mnt_drop_write(mnt);
 543         if (locked)
 544                 inode_unlock(dir);
 545
 546         return work;
 547
 548 out_dput:
 549         dput(work);
 550 out_err:
 551         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
 552                 ufs->config.workdir, name, -err);
 553         sb->s_flags |= MS_RDONLY;
 554         work = NULL;
 555         goto out_unlock;
 556 }
 557
 558 static void ovl_unescape(char *s)
 559 {
 560         char *d = s;
 561
 562         for (;; s++, d++) {
 563                 if (*s == '\\')
 564                         s++;
 565                 *d = *s;
 566                 if (!*s)
 567                         break;
 568         }
 569 }
 570
 571 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 572 {
 573         int err = -EINVAL;
 574
 575         if (!*name) {
 576                 pr_err("overlayfs: empty lowerdir\n");
 577                 goto out;
 578         }
 579         err = kern_path(name, LOOKUP_FOLLOW, path);
 580         if (err) {
 581                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 582                 goto out;
 583         }
 584         err = -EINVAL;
 585         if (ovl_dentry_weird(path->dentry)) {
 586                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 587                 goto out_put;
 588         }
 589         if (!d_is_dir(path->dentry)) {
 590                 pr_err("overlayfs: '%s' not a directory\n", name);
 591                 goto out_put;
 592         }
 593         return 0;
 594
 595 out_put:
 596         path_put_init(path);
 597 out:
 598         return err;
 599 }
 600
 601 static int ovl_mount_dir(const char *name, struct path *path)
 602 {
 603         int err = -ENOMEM;
 604         char *tmp = kstrdup(name, GFP_KERNEL);
 605
 606         if (tmp) {
 607                 ovl_unescape(tmp);
 608                 err = ovl_mount_dir_noesc(tmp, path);
 609
 610                 if (!err)
 611                         if (ovl_dentry_remote(path->dentry)) {
 612                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 613                                        tmp);
 614                                 path_put_init(path);
 615                                 err = -EINVAL;
 616                         }
 617                 kfree(tmp);
 618         }
 619         return err;
 620 }
 621
 622 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 623                              const char *name)
 624 {
 625         struct kstatfs statfs;
 626         int err = vfs_statfs(path, &statfs);
 627
 628         if (err)
 629                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 630         else
 631                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 632
 633         return err;
 634 }
 635
 636 static int ovl_lower_dir(const char *name, struct path *path,
 637                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 638 {
 639         int err;
 640
 641         err = ovl_mount_dir_noesc(name, path);
 642         if (err)
 643                 goto out;
 644
 645         err = ovl_check_namelen(path, ofs, name);
 646         if (err)
 647                 goto out_put;
 648
 649         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 650
 651         if (ovl_dentry_remote(path->dentry))
 652                 *remote = true;
 653
 654         /*
 655          * The inodes index feature needs to encode and decode file
 656          * handles, so it requires that all layers support them.
 657          */
 658         if (ofs->config.index && !ovl_can_decode_fh(path->dentry->d_sb)) {
 659                 ofs->config.index = false;
 660                 pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off.\n", name);
 661         }
 662
 663         return 0;
 664
 665 out_put:
 666         path_put_init(path);
 667 out:
 668         return err;
 669 }
 670
 671 /* Workdir should not be subdir of upperdir and vice versa */
 672 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 673 {
 674         bool ok = false;
 675
 676         if (workdir != upperdir) {
 677                 ok = (lock_rename(workdir, upperdir) == NULL);
 678                 unlock_rename(workdir, upperdir);
 679         }
 680         return ok;
 681 }
 682
 683 static unsigned int ovl_split_lowerdirs(char *str)
 684 {
 685         unsigned int ctr = 1;
 686         char *s, *d;
 687
 688         for (s = d = str;; s++, d++) {
 689                 if (*s == '\\') {
 690                         s++;
 691                 } else if (*s == ':') {
 692                         *d = '\0';
 693                         ctr++;
 694                         continue;
 695                 }
 696                 *d = *s;
 697                 if (!*s)
 698                         break;
 699         }
 700         return ctr;
 701 }
 702
 703 static int __maybe_unused
 704 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 705                         struct dentry *dentry, struct inode *inode,
 706                         const char *name, void *buffer, size_t size)
 707 {
 708         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 709 }
 710
 711 static int __maybe_unused
 712 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 713                         struct dentry *dentry, struct inode *inode,
 714                         const char *name, const void *value,
 715                         size_t size, int flags)
 716 {
 717         struct dentry *workdir = ovl_workdir(dentry);
 718         struct inode *realinode = ovl_inode_real(inode);
 719         struct posix_acl *acl = NULL;
 720         int err;
 721
 722         /* Check that everything is OK before copy-up */
 723         if (value) {
 724                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 725                 if (IS_ERR(acl))
 726                         return PTR_ERR(acl);
 727         }
 728         err = -EOPNOTSUPP;
 729         if (!IS_POSIXACL(d_inode(workdir)))
 730                 goto out_acl_release;
 731         if (!realinode->i_op->set_acl)
 732                 goto out_acl_release;
 733         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 734                 err = acl ? -EACCES : 0;
 735                 goto out_acl_release;
 736         }
 737         err = -EPERM;
 738         if (!inode_owner_or_capable(inode))
 739                 goto out_acl_release;
 740
 741         posix_acl_release(acl);
 742
 743         /*
 744          * Check if sgid bit needs to be cleared (actual setacl operation will
 745          * be done with mounter's capabilities and so that won't do it for us).
 746          */
 747         if (unlikely(inode->i_mode & S_ISGID) &&
 748             handler->flags == ACL_TYPE_ACCESS &&
 749             !in_group_p(inode->i_gid) &&
 750             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 751                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 752
 753                 err = ovl_setattr(dentry, &iattr);
 754                 if (err)
 755                         return err;
 756         }
 757
 758         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
 759         if (!err)
 760                 ovl_copyattr(ovl_inode_real(inode), inode);
 761
 762         return err;
 763
 764 out_acl_release:
 765         posix_acl_release(acl);
 766         return err;
 767 }
 768
 769 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 770                              struct dentry *dentry, struct inode *inode,
 771                              const char *name, void *buffer, size_t size)
 772 {
 773         return -EOPNOTSUPP;
 774 }
 775
 776 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 777                              struct dentry *dentry, struct inode *inode,
 778                              const char *name, const void *value,
 779                              size_t size, int flags)
 780 {
 781         return -EOPNOTSUPP;
 782 }
 783
 784 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 785                                struct dentry *dentry, struct inode *inode,
 786                                const char *name, void *buffer, size_t size)
 787 {
 788         return ovl_xattr_get(dentry, inode, name, buffer, size);
 789 }
 790
 791 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 792                                struct dentry *dentry, struct inode *inode,
 793                                const char *name, const void *value,
 794                                size_t size, int flags)
 795 {
 796         return ovl_xattr_set(dentry, inode, name, value, size, flags);
 797 }
 798
 799 static const struct xattr_handler __maybe_unused
 800 ovl_posix_acl_access_xattr_handler = {
 801         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 802         .flags = ACL_TYPE_ACCESS,
 803         .get = ovl_posix_acl_xattr_get,
 804         .set = ovl_posix_acl_xattr_set,
 805 };
 806
 807 static const struct xattr_handler __maybe_unused
 808 ovl_posix_acl_default_xattr_handler = {
 809         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 810         .flags = ACL_TYPE_DEFAULT,
 811         .get = ovl_posix_acl_xattr_get,
 812         .set = ovl_posix_acl_xattr_set,
 813 };
 814
 815 static const struct xattr_handler ovl_own_xattr_handler = {
 816         .prefix = OVL_XATTR_PREFIX,
 817         .get = ovl_own_xattr_get,
 818         .set = ovl_own_xattr_set,
 819 };
 820
 821 static const struct xattr_handler ovl_other_xattr_handler = {
 822         .prefix = "", /* catch all */
 823         .get = ovl_other_xattr_get,
 824         .set = ovl_other_xattr_set,
 825 };
 826
 827 static const struct xattr_handler *ovl_xattr_handlers[] = {
 828 #ifdef CONFIG_FS_POSIX_ACL
 829         &ovl_posix_acl_access_xattr_handler,
 830         &ovl_posix_acl_default_xattr_handler,
 831 #endif
 832         &ovl_own_xattr_handler,
 833         &ovl_other_xattr_handler,
 834         NULL
 835 };
 836
 837 static int ovl_get_upperpath(struct ovl_fs *ufs, struct path *upperpath)
 838 {
 839         int err;
 840
 841         err = ovl_mount_dir(ufs->config.upperdir, upperpath);
 842         if (err)
 843                 goto out;
 844
 845         /* Upper fs should not be r/o */
 846         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
 847                 pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 848                 err = -EINVAL;
 849                 goto out;
 850         }
 851
 852         err = ovl_check_namelen(upperpath, ufs, ufs->config.upperdir);
 853         if (err)
 854                 goto out;
 855
 856         err = -EBUSY;
 857         if (ovl_inuse_trylock(upperpath->dentry)) {
 858                 ufs->upperdir_locked = true;
 859         } else if (ufs->config.index) {
 860                 pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n");
 861                 goto out;
 862         } else {
 863                 pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
 864         }
 865         err = 0;
 866 out:
 867         return err;
 868 }
 869
 870 static int ovl_get_workpath(struct ovl_fs *ufs, struct path *upperpath,
 871                             struct path *workpath)
 872 {
 873         int err;
 874
 875         err = ovl_mount_dir(ufs->config.workdir, workpath);
 876         if (err)
 877                 goto out;
 878
 879         err = -EINVAL;
 880         if (upperpath->mnt != workpath->mnt) {
 881                 pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 882                 goto out;
 883         }
 884         if (!ovl_workdir_ok(workpath->dentry, upperpath->dentry)) {
 885                 pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 886                 goto out;
 887         }
 888
 889         err = -EBUSY;
 890         if (ovl_inuse_trylock(workpath->dentry)) {
 891                 ufs->workdir_locked = true;
 892         } else if (ufs->config.index) {
 893                 pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n");
 894                 goto out;
 895         } else {
 896                 pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
 897         }
 898
 899         ufs->workbasedir = dget(workpath->dentry);
 900         err = 0;
 901 out:
 902         return err;
 903 }
 904
 905 static int ovl_get_upper(struct ovl_fs *ufs, struct path *upperpath)
 906 {
 907         struct vfsmount *upper_mnt;
 908
 909         upper_mnt = clone_private_mount(upperpath);
 910         if (IS_ERR(upper_mnt)) {
 911                 pr_err("overlayfs: failed to clone upperpath\n");
 912                 return PTR_ERR(upper_mnt);
 913         }
 914
 915         /* Don't inherit atime flags */
 916         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
 917         ufs->upper_mnt = upper_mnt;
 918
 919         return 0;
 920 }
 921
 922 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ufs,
 923                            struct path *workpath)
 924 {
 925         struct dentry *temp;
 926         int err;
 927
 928         ufs->workdir = ovl_workdir_create(sb, ufs, ufs->workbasedir,
 929                                           OVL_WORKDIR_NAME, false);
 930         if (!ufs->workdir)
 931                 return 0;
 932
 933         /*
 934          * Upper should support d_type, else whiteouts are visible.  Given
 935          * workdir and upper are on same fs, we can do iterate_dir() on
 936          * workdir. This check requires successful creation of workdir in
 937          * previous step.
 938          */
 939         err = ovl_check_d_type_supported(workpath);
 940         if (err < 0)
 941                 return err;
 942
 943         /*
 944          * We allowed this configuration and don't want to break users over
 945          * kernel upgrade. So warn instead of erroring out.
 946          */
 947         if (!err)
 948                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
 949
 950         /* Check if upper/work fs supports O_TMPFILE */
 951         temp = ovl_do_tmpfile(ufs->workdir, S_IFREG | 0);
 952         ufs->tmpfile = !IS_ERR(temp);
 953         if (ufs->tmpfile)
 954                 dput(temp);
 955         else
 956                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
 957
 958         /*
 959          * Check if upper/work fs supports trusted.overlay.* xattr
 960          */
 961         err = ovl_do_setxattr(ufs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
 962         if (err) {
 963                 ufs->noxattr = true;
 964                 pr_warn("overlayfs: upper fs does not support xattr.\n");
 965         } else {
 966                 vfs_removexattr(ufs->workdir, OVL_XATTR_OPAQUE);
 967         }
 968
 969         /* Check if upper/work fs supports file handles */
 970         if (ufs->config.index &&
 971             !ovl_can_decode_fh(ufs->workdir->d_sb)) {
 972                 ufs->config.index = false;
 973                 pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
 974         }
 975
 976         return 0;
 977 }
 978
 979 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ufs,
 980                             struct ovl_entry *oe,
 981                             struct path *upperpath)
 982 {
 983         int err;
 984
 985         /* Verify lower root is upper root origin */
 986         err = ovl_verify_origin(upperpath->dentry,
 987                                 oe->lowerstack[0].layer->mnt,
 988                                 oe->lowerstack[0].dentry,
 989                                 false, true);
 990         if (err) {
 991                 pr_err("overlayfs: failed to verify upper root origin\n");
 992                 goto out;
 993         }
 994
 995         ufs->indexdir = ovl_workdir_create(sb, ufs, ufs->workbasedir,
 996                                            OVL_INDEXDIR_NAME, true);
 997         if (ufs->indexdir) {
 998                 /* Verify upper root is index dir origin */
 999                 err = ovl_verify_origin(ufs->indexdir, ufs->upper_mnt,
1000                                         upperpath->dentry, true, true);
1001                 if (err)
1002                         pr_err("overlayfs: failed to verify index dir origin\n");
1003
1004                 /* Cleanup bad/stale/orphan index entries */
1005                 if (!err)
1006                         err = ovl_indexdir_cleanup(ufs->indexdir,
1007                                                    ufs->upper_mnt,
1008                                                    oe->lowerstack,
1009                                                    oe->numlower);
1010         }
1011         if (err || !ufs->indexdir)
1012                 pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1013
1014 out:
1015         return err;
1016 }
1017
1018 static int ovl_get_lowerstack(struct super_block *sb, struct ovl_fs *ufs,
1019                               struct path **stackp, unsigned int *stacklenp)
1020 {
1021         int err;
1022         char *lowertmp, *lower;
1023         struct path *stack;
1024         unsigned int stacklen, numlower, i;
1025         bool remote = false;
1026
1027         err = -ENOMEM;
1028         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
1029         if (!lowertmp)
1030                 goto out;
1031
1032         err = -EINVAL;
1033         stacklen = ovl_split_lowerdirs(lowertmp);
1034         if (stacklen > OVL_MAX_STACK) {
1035                 pr_err("overlayfs: too many lower directories, limit is %d\n",
1036                        OVL_MAX_STACK);
1037                 goto out;
1038         } else if (!ufs->config.upperdir && stacklen == 1) {
1039                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1040                 goto out;
1041         }
1042
1043         err = -ENOMEM;
1044         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1045         if (!stack)
1046                 goto out;
1047
1048         err = -EINVAL;
1049         lower = lowertmp;
1050         for (numlower = 0; numlower < stacklen; numlower++) {
1051                 err = ovl_lower_dir(lower, &stack[numlower], ufs,
1052                                     &sb->s_stack_depth, &remote);
1053                 if (err)
1054                         goto out_free_stack;
1055
1056                 lower = strchr(lower, '\0') + 1;
1057         }
1058
1059         err = -EINVAL;
1060         sb->s_stack_depth++;
1061         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1062                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1063                 goto out_free_stack;
1064         }
1065
1066         *stackp = stack;
1067         *stacklenp = numlower;
1068
1069         if (remote)
1070                 sb->s_d_op = &ovl_reval_dentry_operations;
1071         else
1072                 sb->s_d_op = &ovl_dentry_operations;
1073
1074         err = 0;
1075
1076 out:
1077         kfree(lowertmp);
1078         return err;
1079
1080 out_free_stack:
1081         for (i = 0; i < numlower; i++)
1082                 path_put(&stack[i]);
1083         kfree(stack);
1084         goto out;
1085 }
1086
1087 static int ovl_get_lower_layers(struct ovl_fs *ufs, struct path *stack,
1088                                 unsigned int numlower)
1089 {
1090         int err;
1091         unsigned int i;
1092
1093         err = -ENOMEM;
1094         ufs->lower_layers = kcalloc(numlower, sizeof(struct ovl_layer),
1095                                     GFP_KERNEL);
1096         if (ufs->lower_layers == NULL)
1097                 goto out;
1098         for (i = 0; i < numlower; i++) {
1099                 struct vfsmount *mnt;
1100                 dev_t dev;
1101
1102                 err = get_anon_bdev(&dev);
1103                 if (err) {
1104                         pr_err("overlayfs: failed to get anonymous bdev for lowerpath\n");
1105                         goto out;
1106                 }
1107
1108                 mnt = clone_private_mount(&stack[i]);
1109                 err = PTR_ERR(mnt);
1110                 if (IS_ERR(mnt)) {
1111                         pr_err("overlayfs: failed to clone lowerpath\n");
1112                         free_anon_bdev(dev);
1113                         goto out;
1114                 }
1115                 /*
1116                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1117                  * will fail instead of modifying lower fs.
1118                  */
1119                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1120
1121                 ufs->lower_layers[ufs->numlower].mnt = mnt;
1122                 ufs->lower_layers[ufs->numlower].pseudo_dev = dev;
1123                 ufs->numlower++;
1124
1125                 /* Check if all lower layers are on same sb */
1126                 if (i == 0)
1127                         ufs->same_sb = mnt->mnt_sb;
1128                 else if (ufs->same_sb != mnt->mnt_sb)
1129                         ufs->same_sb = NULL;
1130         }
1131         err = 0;
1132 out:
1133         return err;
1134 }
1135
1136 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1137 {
1138         struct path upperpath = { };
1139         struct path workpath = { };
1140         struct dentry *root_dentry;
1141         struct ovl_entry *oe = NULL;
1142         struct ovl_fs *ufs;
1143         struct path *stack = NULL;
1144         unsigned int numlower = 0;
1145         unsigned int i;
1146         struct cred *cred;
1147         int err;
1148
1149         err = -ENOMEM;
1150         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1151         if (!ufs)
1152                 goto out;
1153
1154         ufs->creator_cred = cred = prepare_creds();
1155         if (!cred)
1156                 goto out_err;
1157
1158         ufs->config.redirect_dir = ovl_redirect_dir_def;
1159         ufs->config.index = ovl_index_def;
1160         err = ovl_parse_opt((char *) data, &ufs->config);
1161         if (err)
1162                 goto out_err;
1163
1164         err = -EINVAL;
1165         if (!ufs->config.lowerdir) {
1166                 if (!silent)
1167                         pr_err("overlayfs: missing 'lowerdir'\n");
1168                 goto out_err;
1169         }
1170
1171         sb->s_stack_depth = 0;
1172         sb->s_maxbytes = MAX_LFS_FILESIZE;
1173         if (ufs->config.upperdir) {
1174                 if (!ufs->config.workdir) {
1175                         pr_err("overlayfs: missing 'workdir'\n");
1176                         goto out_err;
1177                 }
1178
1179                 err = ovl_get_upperpath(ufs, &upperpath);
1180                 if (err)
1181                         goto out_err;
1182
1183                 err = ovl_get_upper(ufs, &upperpath);
1184                 if (err)
1185                         goto out_err;
1186
1187                 err = ovl_get_workpath(ufs, &upperpath, &workpath);
1188                 if (err)
1189                         goto out_err;
1190
1191                 err = ovl_get_workdir(sb, ufs, &workpath);
1192                 if (err)
1193                         goto out_err;
1194
1195                 sb->s_stack_depth = ufs->upper_mnt->mnt_sb->s_stack_depth;
1196                 sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
1197
1198         }
1199         err = ovl_get_lowerstack(sb, ufs, &stack, &numlower);
1200         if (err)
1201                 goto out_err;
1202
1203         err = ovl_get_lower_layers(ufs, stack, numlower);
1204         if (err)
1205                 goto out_err;
1206
1207         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1208         if (!ufs->upper_mnt)
1209                 sb->s_flags |= MS_RDONLY;
1210         else if (ufs->upper_mnt->mnt_sb != ufs->same_sb)
1211                 ufs->same_sb = NULL;
1212
1213         err = -ENOMEM;
1214         oe = ovl_alloc_entry(numlower);
1215         if (!oe)
1216                 goto out_err;
1217
1218         for (i = 0; i < numlower; i++) {
1219                 oe->lowerstack[i].dentry = stack[i].dentry;
1220                 oe->lowerstack[i].layer = &(ufs->lower_layers[i]);
1221         }
1222
1223         if (!(ovl_force_readonly(ufs)) && ufs->config.index) {
1224                 err = ovl_get_indexdir(sb, ufs, oe, &upperpath);
1225                 if (err)
1226                         goto out_err;
1227         }
1228
1229         /* Show index=off/on in /proc/mounts for any of the reasons above */
1230         if (!ufs->indexdir)
1231                 ufs->config.index = false;
1232
1233         /* Never override disk quota limits or use reserved space */
1234         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1235
1236         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1237         sb->s_op = &ovl_super_operations;
1238         sb->s_xattr = ovl_xattr_handlers;
1239         sb->s_fs_info = ufs;
1240         sb->s_flags |= MS_POSIXACL | MS_NOREMOTELOCK;
1241
1242         err = -ENOMEM;
1243         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1244         if (!root_dentry)
1245                 goto out_err;
1246
1247         mntput(upperpath.mnt);
1248         for (i = 0; i < numlower; i++)
1249                 mntput(stack[i].mnt);
1250         kfree(stack);
1251         path_put(&workpath);
1252
1253         if (upperpath.dentry) {
1254                 oe->has_upper = true;
1255                 if (ovl_is_impuredir(upperpath.dentry))
1256                         ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1257         }
1258
1259         root_dentry->d_fsdata = oe;
1260
1261         /* Root is always merge -> can have whiteouts */
1262         ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1263         ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1264                        ovl_dentry_lower(root_dentry));
1265
1266         sb->s_root = root_dentry;
1267
1268         return 0;
1269
1270 out_err:
1271         kfree(oe);
1272         for (i = 0; i < numlower; i++)
1273                 path_put(&stack[i]);
1274         kfree(stack);
1275         path_put(&workpath);
1276         path_put(&upperpath);
1277         ovl_free_fs(ufs);
1278 out:
1279         return err;
1280 }
1281
1282 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1283                                 const char *dev_name, void *raw_data)
1284 {
1285         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1286 }
1287
1288 static struct file_system_type ovl_fs_type = {
1289         .owner          = THIS_MODULE,
1290         .name           = "overlay",
1291         .mount          = ovl_mount,
1292         .kill_sb        = kill_anon_super,
1293 };
1294 MODULE_ALIAS_FS("overlay");
1295
1296 static void ovl_inode_init_once(void *foo)
1297 {
1298         struct ovl_inode *oi = foo;
1299
1300         inode_init_once(&oi->vfs_inode);
1301 }
1302
1303 static int __init ovl_init(void)
1304 {
1305         int err;
1306
1307         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1308                                              sizeof(struct ovl_inode), 0,
1309                                              (SLAB_RECLAIM_ACCOUNT|
1310                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1311                                              ovl_inode_init_once);
1312         if (ovl_inode_cachep == NULL)
1313                 return -ENOMEM;
1314
1315         err = register_filesystem(&ovl_fs_type);
1316         if (err)
1317                 kmem_cache_destroy(ovl_inode_cachep);
1318
1319         return err;
1320 }
1321
1322 static void __exit ovl_exit(void)
1323 {
1324         unregister_filesystem(&ovl_fs_type);
1325
1326         /*
1327          * Make sure all delayed rcu free inodes are flushed before we
1328          * destroy cache.
1329          */
1330         rcu_barrier();
1331         kmem_cache_destroy(ovl_inode_cachep);
1332
1333 }
1334
1335 module_init(ovl_init);
1336 module_exit(ovl_exit);