fs/overlayfs/super.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *
   4  * Copyright (C) 2011 Novell Inc.
   5  */
   6
   7 #include <uapi/linux/magic.h>
   8 #include <linux/fs.h>
   9 #include <linux/namei.h>
  10 #include <linux/xattr.h>
  11 #include <linux/mount.h>
  12 #include <linux/parser.h>
  13 #include <linux/module.h>
  14 #include <linux/statfs.h>
  15 #include <linux/seq_file.h>
  16 #include <linux/posix_acl_xattr.h>
  17 #include <linux/exportfs.h>
  18 #include "overlayfs.h"
  19
  20 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  21 MODULE_DESCRIPTION("Overlay filesystem");
  22 MODULE_LICENSE("GPL");
  23
  24
  25 struct ovl_dir_cache;
  26
  27 #define OVL_MAX_STACK 500
  28
  29 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  30 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  31 MODULE_PARM_DESC(redirect_dir,
  32                  "Default to on or off for the redirect_dir feature");
  33
  34 static bool ovl_redirect_always_follow =
  35         IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
  36 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
  37                    bool, 0644);
  38 MODULE_PARM_DESC(redirect_always_follow,
  39                  "Follow redirects even if redirect_dir feature is turned off");
  40
  41 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
  42 module_param_named(index, ovl_index_def, bool, 0644);
  43 MODULE_PARM_DESC(index,
  44                  "Default to on or off for the inodes index feature");
  45
  46 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
  47 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
  48 MODULE_PARM_DESC(nfs_export,
  49                  "Default to on or off for the NFS export feature");
  50
  51 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
  52 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
  53 MODULE_PARM_DESC(xino_auto,
  54                  "Auto enable xino feature");
  55
  56 static void ovl_entry_stack_free(struct ovl_entry *oe)
  57 {
  58         unsigned int i;
  59
  60         for (i = 0; i < oe->numlower; i++)
  61                 dput(oe->lowerstack[i].dentry);
  62 }
  63
  64 static bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
  65 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
  66 MODULE_PARM_DESC(metacopy,
  67                  "Default to on or off for the metadata only copy up feature");
  68
  69 static void ovl_dentry_release(struct dentry *dentry)
  70 {
  71         struct ovl_entry *oe = dentry->d_fsdata;
  72
  73         if (oe) {
  74                 ovl_entry_stack_free(oe);
  75                 kfree_rcu(oe, rcu);
  76         }
  77 }
  78
  79 static struct dentry *ovl_d_real(struct dentry *dentry,
  80                                  const struct inode *inode)
  81 {
  82         struct dentry *real;
  83
  84         /* It's an overlay file */
  85         if (inode && d_inode(dentry) == inode)
  86                 return dentry;
  87
  88         if (!d_is_reg(dentry)) {
  89                 if (!inode || inode == d_inode(dentry))
  90                         return dentry;
  91                 goto bug;
  92         }
  93
  94         real = ovl_dentry_upper(dentry);
  95         if (real && (inode == d_inode(real)))
  96                 return real;
  97
  98         if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
  99                 return real;
 100
 101         real = ovl_dentry_lowerdata(dentry);
 102         if (!real)
 103                 goto bug;
 104
 105         /* Handle recursion */
 106         real = d_real(real, inode);
 107
 108         if (!inode || inode == d_inode(real))
 109                 return real;
 110 bug:
 111         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
 112              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 113         return dentry;
 114 }
 115
 116 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 117 {
 118         struct ovl_entry *oe = dentry->d_fsdata;
 119         unsigned int i;
 120         int ret = 1;
 121
 122         for (i = 0; i < oe->numlower; i++) {
 123                 struct dentry *d = oe->lowerstack[i].dentry;
 124
 125                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 126                         ret = d->d_op->d_revalidate(d, flags);
 127                         if (ret < 0)
 128                                 return ret;
 129                         if (!ret) {
 130                                 if (!(flags & LOOKUP_RCU))
 131                                         d_invalidate(d);
 132                                 return -ESTALE;
 133                         }
 134                 }
 135         }
 136         return 1;
 137 }
 138
 139 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 140 {
 141         struct ovl_entry *oe = dentry->d_fsdata;
 142         unsigned int i;
 143         int ret = 1;
 144
 145         for (i = 0; i < oe->numlower; i++) {
 146                 struct dentry *d = oe->lowerstack[i].dentry;
 147
 148                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 149                         ret = d->d_op->d_weak_revalidate(d, flags);
 150                         if (ret <= 0)
 151                                 break;
 152                 }
 153         }
 154         return ret;
 155 }
 156
 157 static const struct dentry_operations ovl_dentry_operations = {
 158         .d_release = ovl_dentry_release,
 159         .d_real = ovl_d_real,
 160 };
 161
 162 static const struct dentry_operations ovl_reval_dentry_operations = {
 163         .d_release = ovl_dentry_release,
 164         .d_real = ovl_d_real,
 165         .d_revalidate = ovl_dentry_revalidate,
 166         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 167 };
 168
 169 static struct kmem_cache *ovl_inode_cachep;
 170
 171 static struct inode *ovl_alloc_inode(struct super_block *sb)
 172 {
 173         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
 174
 175         if (!oi)
 176                 return NULL;
 177
 178         oi->cache = NULL;
 179         oi->redirect = NULL;
 180         oi->version = 0;
 181         oi->flags = 0;
 182         oi->__upperdentry = NULL;
 183         oi->lower = NULL;
 184         oi->lowerdata = NULL;
 185         mutex_init(&oi->lock);
 186
 187         return &oi->vfs_inode;
 188 }
 189
 190 static void ovl_free_inode(struct inode *inode)
 191 {
 192         struct ovl_inode *oi = OVL_I(inode);
 193
 194         kfree(oi->redirect);
 195         mutex_destroy(&oi->lock);
 196         kmem_cache_free(ovl_inode_cachep, oi);
 197 }
 198
 199 static void ovl_destroy_inode(struct inode *inode)
 200 {
 201         struct ovl_inode *oi = OVL_I(inode);
 202
 203         dput(oi->__upperdentry);
 204         iput(oi->lower);
 205         if (S_ISDIR(inode->i_mode))
 206                 ovl_dir_cache_free(inode);
 207         else
 208                 iput(oi->lowerdata);
 209 }
 210
 211 static void ovl_free_fs(struct ovl_fs *ofs)
 212 {
 213         unsigned i;
 214
 215         iput(ofs->workbasedir_trap);
 216         iput(ofs->indexdir_trap);
 217         iput(ofs->workdir_trap);
 218         iput(ofs->upperdir_trap);
 219         dput(ofs->indexdir);
 220         dput(ofs->workdir);
 221         if (ofs->workdir_locked)
 222                 ovl_inuse_unlock(ofs->workbasedir);
 223         dput(ofs->workbasedir);
 224         if (ofs->upperdir_locked)
 225                 ovl_inuse_unlock(ofs->upper_mnt->mnt_root);
 226         mntput(ofs->upper_mnt);
 227         for (i = 1; i < ofs->numlayer; i++) {
 228                 iput(ofs->layers[i].trap);
 229                 mntput(ofs->layers[i].mnt);
 230         }
 231         kfree(ofs->layers);
 232         for (i = 0; i < ofs->numfs; i++)
 233                 free_anon_bdev(ofs->fs[i].pseudo_dev);
 234         kfree(ofs->fs);
 235
 236         kfree(ofs->config.lowerdir);
 237         kfree(ofs->config.upperdir);
 238         kfree(ofs->config.workdir);
 239         kfree(ofs->config.redirect_mode);
 240         if (ofs->creator_cred)
 241                 put_cred(ofs->creator_cred);
 242         kfree(ofs);
 243 }
 244
 245 static void ovl_put_super(struct super_block *sb)
 246 {
 247         struct ovl_fs *ofs = sb->s_fs_info;
 248
 249         ovl_free_fs(ofs);
 250 }
 251
 252 /* Sync real dirty inodes in upper filesystem (if it exists) */
 253 static int ovl_sync_fs(struct super_block *sb, int wait)
 254 {
 255         struct ovl_fs *ofs = sb->s_fs_info;
 256         struct super_block *upper_sb;
 257         int ret;
 258
 259         if (!ofs->upper_mnt)
 260                 return 0;
 261
 262         /*
 263          * If this is a sync(2) call or an emergency sync, all the super blocks
 264          * will be iterated, including upper_sb, so no need to do anything.
 265          *
 266          * If this is a syncfs(2) call, then we do need to call
 267          * sync_filesystem() on upper_sb, but enough if we do it when being
 268          * called with wait == 1.
 269          */
 270         if (!wait)
 271                 return 0;
 272
 273         upper_sb = ofs->upper_mnt->mnt_sb;
 274
 275         down_read(&upper_sb->s_umount);
 276         ret = sync_filesystem(upper_sb);
 277         up_read(&upper_sb->s_umount);
 278
 279         return ret;
 280 }
 281
 282 /**
 283  * ovl_statfs
 284  * @sb: The overlayfs super block
 285  * @buf: The struct kstatfs to fill in with stats
 286  *
 287  * Get the filesystem statistics.  As writes always target the upper layer
 288  * filesystem pass the statfs to the upper filesystem (if it exists)
 289  */
 290 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 291 {
 292         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 293         struct dentry *root_dentry = dentry->d_sb->s_root;
 294         struct path path;
 295         int err;
 296
 297         ovl_path_real(root_dentry, &path);
 298
 299         err = vfs_statfs(&path, buf);
 300         if (!err) {
 301                 buf->f_namelen = ofs->namelen;
 302                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 303         }
 304
 305         return err;
 306 }
 307
 308 /* Will this overlay be forced to mount/remount ro? */
 309 static bool ovl_force_readonly(struct ovl_fs *ofs)
 310 {
 311         return (!ofs->upper_mnt || !ofs->workdir);
 312 }
 313
 314 static const char *ovl_redirect_mode_def(void)
 315 {
 316         return ovl_redirect_dir_def ? "on" : "off";
 317 }
 318
 319 enum {
 320         OVL_XINO_OFF,
 321         OVL_XINO_AUTO,
 322         OVL_XINO_ON,
 323 };
 324
 325 static const char * const ovl_xino_str[] = {
 326         "off",
 327         "auto",
 328         "on",
 329 };
 330
 331 static inline int ovl_xino_def(void)
 332 {
 333         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 334 }
 335
 336 /**
 337  * ovl_show_options
 338  *
 339  * Prints the mount options for a given superblock.
 340  * Returns zero; does not fail.
 341  */
 342 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 343 {
 344         struct super_block *sb = dentry->d_sb;
 345         struct ovl_fs *ofs = sb->s_fs_info;
 346
 347         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 348         if (ofs->config.upperdir) {
 349                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 350                 seq_show_option(m, "workdir", ofs->config.workdir);
 351         }
 352         if (ofs->config.default_permissions)
 353                 seq_puts(m, ",default_permissions");
 354         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 355                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 356         if (ofs->config.index != ovl_index_def)
 357                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 358         if (ofs->config.nfs_export != ovl_nfs_export_def)
 359                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 360                                                 "on" : "off");
 361         if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
 362                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 363         if (ofs->config.metacopy != ovl_metacopy_def)
 364                 seq_printf(m, ",metacopy=%s",
 365                            ofs->config.metacopy ? "on" : "off");
 366         return 0;
 367 }
 368
 369 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 370 {
 371         struct ovl_fs *ofs = sb->s_fs_info;
 372
 373         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 374                 return -EROFS;
 375
 376         return 0;
 377 }
 378
 379 static const struct super_operations ovl_super_operations = {
 380         .alloc_inode    = ovl_alloc_inode,
 381         .free_inode     = ovl_free_inode,
 382         .destroy_inode  = ovl_destroy_inode,
 383         .drop_inode     = generic_delete_inode,
 384         .put_super      = ovl_put_super,
 385         .sync_fs        = ovl_sync_fs,
 386         .statfs         = ovl_statfs,
 387         .show_options   = ovl_show_options,
 388         .remount_fs     = ovl_remount,
 389 };
 390
 391 enum {
 392         OPT_LOWERDIR,
 393         OPT_UPPERDIR,
 394         OPT_WORKDIR,
 395         OPT_DEFAULT_PERMISSIONS,
 396         OPT_REDIRECT_DIR,
 397         OPT_INDEX_ON,
 398         OPT_INDEX_OFF,
 399         OPT_NFS_EXPORT_ON,
 400         OPT_NFS_EXPORT_OFF,
 401         OPT_XINO_ON,
 402         OPT_XINO_OFF,
 403         OPT_XINO_AUTO,
 404         OPT_METACOPY_ON,
 405         OPT_METACOPY_OFF,
 406         OPT_ERR,
 407 };
 408
 409 static const match_table_t ovl_tokens = {
 410         {OPT_LOWERDIR,                  "lowerdir=%s"},
 411         {OPT_UPPERDIR,                  "upperdir=%s"},
 412         {OPT_WORKDIR,                   "workdir=%s"},
 413         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 414         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 415         {OPT_INDEX_ON,                  "index=on"},
 416         {OPT_INDEX_OFF,                 "index=off"},
 417         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 418         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 419         {OPT_XINO_ON,                   "xino=on"},
 420         {OPT_XINO_OFF,                  "xino=off"},
 421         {OPT_XINO_AUTO,                 "xino=auto"},
 422         {OPT_METACOPY_ON,               "metacopy=on"},
 423         {OPT_METACOPY_OFF,              "metacopy=off"},
 424         {OPT_ERR,                       NULL}
 425 };
 426
 427 static char *ovl_next_opt(char **s)
 428 {
 429         char *sbegin = *s;
 430         char *p;
 431
 432         if (sbegin == NULL)
 433                 return NULL;
 434
 435         for (p = sbegin; *p; p++) {
 436                 if (*p == '\\') {
 437                         p++;
 438                         if (!*p)
 439                                 break;
 440                 } else if (*p == ',') {
 441                         *p = '\0';
 442                         *s = p + 1;
 443                         return sbegin;
 444                 }
 445         }
 446         *s = NULL;
 447         return sbegin;
 448 }
 449
 450 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 451 {
 452         if (strcmp(mode, "on") == 0) {
 453                 config->redirect_dir = true;
 454                 /*
 455                  * Does not make sense to have redirect creation without
 456                  * redirect following.
 457                  */
 458                 config->redirect_follow = true;
 459         } else if (strcmp(mode, "follow") == 0) {
 460                 config->redirect_follow = true;
 461         } else if (strcmp(mode, "off") == 0) {
 462                 if (ovl_redirect_always_follow)
 463                         config->redirect_follow = true;
 464         } else if (strcmp(mode, "nofollow") != 0) {
 465                 pr_err("bad mount option \"redirect_dir=%s\"\n",
 466                        mode);
 467                 return -EINVAL;
 468         }
 469
 470         return 0;
 471 }
 472
 473 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 474 {
 475         char *p;
 476         int err;
 477         bool metacopy_opt = false, redirect_opt = false;
 478
 479         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 480         if (!config->redirect_mode)
 481                 return -ENOMEM;
 482
 483         while ((p = ovl_next_opt(&opt)) != NULL) {
 484                 int token;
 485                 substring_t args[MAX_OPT_ARGS];
 486
 487                 if (!*p)
 488                         continue;
 489
 490                 token = match_token(p, ovl_tokens, args);
 491                 switch (token) {
 492                 case OPT_UPPERDIR:
 493                         kfree(config->upperdir);
 494                         config->upperdir = match_strdup(&args[0]);
 495                         if (!config->upperdir)
 496                                 return -ENOMEM;
 497                         break;
 498
 499                 case OPT_LOWERDIR:
 500                         kfree(config->lowerdir);
 501                         config->lowerdir = match_strdup(&args[0]);
 502                         if (!config->lowerdir)
 503                                 return -ENOMEM;
 504                         break;
 505
 506                 case OPT_WORKDIR:
 507                         kfree(config->workdir);
 508                         config->workdir = match_strdup(&args[0]);
 509                         if (!config->workdir)
 510                                 return -ENOMEM;
 511                         break;
 512
 513                 case OPT_DEFAULT_PERMISSIONS:
 514                         config->default_permissions = true;
 515                         break;
 516
 517                 case OPT_REDIRECT_DIR:
 518                         kfree(config->redirect_mode);
 519                         config->redirect_mode = match_strdup(&args[0]);
 520                         if (!config->redirect_mode)
 521                                 return -ENOMEM;
 522                         redirect_opt = true;
 523                         break;
 524
 525                 case OPT_INDEX_ON:
 526                         config->index = true;
 527                         break;
 528
 529                 case OPT_INDEX_OFF:
 530                         config->index = false;
 531                         break;
 532
 533                 case OPT_NFS_EXPORT_ON:
 534                         config->nfs_export = true;
 535                         break;
 536
 537                 case OPT_NFS_EXPORT_OFF:
 538                         config->nfs_export = false;
 539                         break;
 540
 541                 case OPT_XINO_ON:
 542                         config->xino = OVL_XINO_ON;
 543                         break;
 544
 545                 case OPT_XINO_OFF:
 546                         config->xino = OVL_XINO_OFF;
 547                         break;
 548
 549                 case OPT_XINO_AUTO:
 550                         config->xino = OVL_XINO_AUTO;
 551                         break;
 552
 553                 case OPT_METACOPY_ON:
 554                         config->metacopy = true;
 555                         metacopy_opt = true;
 556                         break;
 557
 558                 case OPT_METACOPY_OFF:
 559                         config->metacopy = false;
 560                         break;
 561
 562                 default:
 563                         pr_err("unrecognized mount option \"%s\" or missing value\n",
 564                                         p);
 565                         return -EINVAL;
 566                 }
 567         }
 568
 569         /* Workdir is useless in non-upper mount */
 570         if (!config->upperdir && config->workdir) {
 571                 pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 572                         config->workdir);
 573                 kfree(config->workdir);
 574                 config->workdir = NULL;
 575         }
 576
 577         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 578         if (err)
 579                 return err;
 580
 581         /*
 582          * This is to make the logic below simpler.  It doesn't make any other
 583          * difference, since config->redirect_dir is only used for upper.
 584          */
 585         if (!config->upperdir && config->redirect_follow)
 586                 config->redirect_dir = true;
 587
 588         /* Resolve metacopy -> redirect_dir dependency */
 589         if (config->metacopy && !config->redirect_dir) {
 590                 if (metacopy_opt && redirect_opt) {
 591                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
 592                                config->redirect_mode);
 593                         return -EINVAL;
 594                 }
 595                 if (redirect_opt) {
 596                         /*
 597                          * There was an explicit redirect_dir=... that resulted
 598                          * in this conflict.
 599                          */
 600                         pr_info("disabling metacopy due to redirect_dir=%s\n",
 601                                 config->redirect_mode);
 602                         config->metacopy = false;
 603                 } else {
 604                         /* Automatically enable redirect otherwise. */
 605                         config->redirect_follow = config->redirect_dir = true;
 606                 }
 607         }
 608
 609         return 0;
 610 }
 611
 612 #define OVL_WORKDIR_NAME "work"
 613 #define OVL_INDEXDIR_NAME "index"
 614
 615 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 616                                          const char *name, bool persist)
 617 {
 618         struct inode *dir =  ofs->workbasedir->d_inode;
 619         struct vfsmount *mnt = ofs->upper_mnt;
 620         struct dentry *work;
 621         int err;
 622         bool retried = false;
 623         bool locked = false;
 624
 625         inode_lock_nested(dir, I_MUTEX_PARENT);
 626         locked = true;
 627
 628 retry:
 629         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
 630
 631         if (!IS_ERR(work)) {
 632                 struct iattr attr = {
 633                         .ia_valid = ATTR_MODE,
 634                         .ia_mode = S_IFDIR | 0,
 635                 };
 636
 637                 if (work->d_inode) {
 638                         err = -EEXIST;
 639                         if (retried)
 640                                 goto out_dput;
 641
 642                         if (persist)
 643                                 goto out_unlock;
 644
 645                         retried = true;
 646                         ovl_workdir_cleanup(dir, mnt, work, 0);
 647                         dput(work);
 648                         goto retry;
 649                 }
 650
 651                 work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
 652                 err = PTR_ERR(work);
 653                 if (IS_ERR(work))
 654                         goto out_err;
 655
 656                 /*
 657                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 658                  *
 659                  * a) success (there was a POSIX ACL xattr and was removed)
 660                  * b) -ENODATA (there was no POSIX ACL xattr)
 661                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 662                  *
 663                  * There are various other error values that could effectively
 664                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 665                  * if the xattr name is too long), but the set of filesystems
 666                  * allowed as upper are limited to "normal" ones, where checking
 667                  * for the above two errors is sufficient.
 668                  */
 669                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
 670                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 671                         goto out_dput;
 672
 673                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
 674                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 675                         goto out_dput;
 676
 677                 /* Clear any inherited mode bits */
 678                 inode_lock(work->d_inode);
 679                 err = notify_change(work, &attr, NULL);
 680                 inode_unlock(work->d_inode);
 681                 if (err)
 682                         goto out_dput;
 683         } else {
 684                 err = PTR_ERR(work);
 685                 goto out_err;
 686         }
 687 out_unlock:
 688         if (locked)
 689                 inode_unlock(dir);
 690
 691         return work;
 692
 693 out_dput:
 694         dput(work);
 695 out_err:
 696         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
 697                 ofs->config.workdir, name, -err);
 698         work = NULL;
 699         goto out_unlock;
 700 }
 701
 702 static void ovl_unescape(char *s)
 703 {
 704         char *d = s;
 705
 706         for (;; s++, d++) {
 707                 if (*s == '\\')
 708                         s++;
 709                 *d = *s;
 710                 if (!*s)
 711                         break;
 712         }
 713 }
 714
 715 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 716 {
 717         int err = -EINVAL;
 718
 719         if (!*name) {
 720                 pr_err("empty lowerdir\n");
 721                 goto out;
 722         }
 723         err = kern_path(name, LOOKUP_FOLLOW, path);
 724         if (err) {
 725                 pr_err("failed to resolve '%s': %i\n", name, err);
 726                 goto out;
 727         }
 728         err = -EINVAL;
 729         if (ovl_dentry_weird(path->dentry)) {
 730                 pr_err("filesystem on '%s' not supported\n", name);
 731                 goto out_put;
 732         }
 733         if (!d_is_dir(path->dentry)) {
 734                 pr_err("'%s' not a directory\n", name);
 735                 goto out_put;
 736         }
 737         return 0;
 738
 739 out_put:
 740         path_put_init(path);
 741 out:
 742         return err;
 743 }
 744
 745 static int ovl_mount_dir(const char *name, struct path *path)
 746 {
 747         int err = -ENOMEM;
 748         char *tmp = kstrdup(name, GFP_KERNEL);
 749
 750         if (tmp) {
 751                 ovl_unescape(tmp);
 752                 err = ovl_mount_dir_noesc(tmp, path);
 753
 754                 if (!err)
 755                         if (ovl_dentry_remote(path->dentry)) {
 756                                 pr_err("filesystem on '%s' not supported as upperdir\n",
 757                                        tmp);
 758                                 path_put_init(path);
 759                                 err = -EINVAL;
 760                         }
 761                 kfree(tmp);
 762         }
 763         return err;
 764 }
 765
 766 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 767                              const char *name)
 768 {
 769         struct kstatfs statfs;
 770         int err = vfs_statfs(path, &statfs);
 771
 772         if (err)
 773                 pr_err("statfs failed on '%s'\n", name);
 774         else
 775                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 776
 777         return err;
 778 }
 779
 780 static int ovl_lower_dir(const char *name, struct path *path,
 781                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
 782 {
 783         int fh_type;
 784         int err;
 785
 786         err = ovl_mount_dir_noesc(name, path);
 787         if (err)
 788                 goto out;
 789
 790         err = ovl_check_namelen(path, ofs, name);
 791         if (err)
 792                 goto out_put;
 793
 794         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 795
 796         if (ovl_dentry_remote(path->dentry))
 797                 *remote = true;
 798
 799         /*
 800          * The inodes index feature and NFS export need to encode and decode
 801          * file handles, so they require that all layers support them.
 802          */
 803         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 804         if ((ofs->config.nfs_export ||
 805              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 806                 ofs->config.index = false;
 807                 ofs->config.nfs_export = false;
 808                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 809                         name);
 810         }
 811
 812         /* Check if lower fs has 32bit inode numbers */
 813         if (fh_type != FILEID_INO32_GEN)
 814                 ofs->xino_mode = -1;
 815
 816         return 0;
 817
 818 out_put:
 819         path_put_init(path);
 820 out:
 821         return err;
 822 }
 823
 824 /* Workdir should not be subdir of upperdir and vice versa */
 825 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 826 {
 827         bool ok = false;
 828
 829         if (workdir != upperdir) {
 830                 ok = (lock_rename(workdir, upperdir) == NULL);
 831                 unlock_rename(workdir, upperdir);
 832         }
 833         return ok;
 834 }
 835
 836 static unsigned int ovl_split_lowerdirs(char *str)
 837 {
 838         unsigned int ctr = 1;
 839         char *s, *d;
 840
 841         for (s = d = str;; s++, d++) {
 842                 if (*s == '\\') {
 843                         s++;
 844                 } else if (*s == ':') {
 845                         *d = '\0';
 846                         ctr++;
 847                         continue;
 848                 }
 849                 *d = *s;
 850                 if (!*s)
 851                         break;
 852         }
 853         return ctr;
 854 }
 855
 856 static int __maybe_unused
 857 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
 858                         struct dentry *dentry, struct inode *inode,
 859                         const char *name, void *buffer, size_t size)
 860 {
 861         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
 862 }
 863
 864 static int __maybe_unused
 865 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 866                         struct dentry *dentry, struct inode *inode,
 867                         const char *name, const void *value,
 868                         size_t size, int flags)
 869 {
 870         struct dentry *workdir = ovl_workdir(dentry);
 871         struct inode *realinode = ovl_inode_real(inode);
 872         struct posix_acl *acl = NULL;
 873         int err;
 874
 875         /* Check that everything is OK before copy-up */
 876         if (value) {
 877                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 878                 if (IS_ERR(acl))
 879                         return PTR_ERR(acl);
 880         }
 881         err = -EOPNOTSUPP;
 882         if (!IS_POSIXACL(d_inode(workdir)))
 883                 goto out_acl_release;
 884         if (!realinode->i_op->set_acl)
 885                 goto out_acl_release;
 886         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 887                 err = acl ? -EACCES : 0;
 888                 goto out_acl_release;
 889         }
 890         err = -EPERM;
 891         if (!inode_owner_or_capable(inode))
 892                 goto out_acl_release;
 893
 894         posix_acl_release(acl);
 895
 896         /*
 897          * Check if sgid bit needs to be cleared (actual setacl operation will
 898          * be done with mounter's capabilities and so that won't do it for us).
 899          */
 900         if (unlikely(inode->i_mode & S_ISGID) &&
 901             handler->flags == ACL_TYPE_ACCESS &&
 902             !in_group_p(inode->i_gid) &&
 903             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
 904                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 905
 906                 err = ovl_setattr(dentry, &iattr);
 907                 if (err)
 908                         return err;
 909         }
 910
 911         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
 912         if (!err)
 913                 ovl_copyattr(ovl_inode_real(inode), inode);
 914
 915         return err;
 916
 917 out_acl_release:
 918         posix_acl_release(acl);
 919         return err;
 920 }
 921
 922 static int ovl_own_xattr_get(const struct xattr_handler *handler,
 923                              struct dentry *dentry, struct inode *inode,
 924                              const char *name, void *buffer, size_t size)
 925 {
 926         return -EOPNOTSUPP;
 927 }
 928
 929 static int ovl_own_xattr_set(const struct xattr_handler *handler,
 930                              struct dentry *dentry, struct inode *inode,
 931                              const char *name, const void *value,
 932                              size_t size, int flags)
 933 {
 934         return -EOPNOTSUPP;
 935 }
 936
 937 static int ovl_other_xattr_get(const struct xattr_handler *handler,
 938                                struct dentry *dentry, struct inode *inode,
 939                                const char *name, void *buffer, size_t size)
 940 {
 941         return ovl_xattr_get(dentry, inode, name, buffer, size);
 942 }
 943
 944 static int ovl_other_xattr_set(const struct xattr_handler *handler,
 945                                struct dentry *dentry, struct inode *inode,
 946                                const char *name, const void *value,
 947                                size_t size, int flags)
 948 {
 949         return ovl_xattr_set(dentry, inode, name, value, size, flags);
 950 }
 951
 952 static const struct xattr_handler __maybe_unused
 953 ovl_posix_acl_access_xattr_handler = {
 954         .name = XATTR_NAME_POSIX_ACL_ACCESS,
 955         .flags = ACL_TYPE_ACCESS,
 956         .get = ovl_posix_acl_xattr_get,
 957         .set = ovl_posix_acl_xattr_set,
 958 };
 959
 960 static const struct xattr_handler __maybe_unused
 961 ovl_posix_acl_default_xattr_handler = {
 962         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
 963         .flags = ACL_TYPE_DEFAULT,
 964         .get = ovl_posix_acl_xattr_get,
 965         .set = ovl_posix_acl_xattr_set,
 966 };
 967
 968 static const struct xattr_handler ovl_own_xattr_handler = {
 969         .prefix = OVL_XATTR_PREFIX,
 970         .get = ovl_own_xattr_get,
 971         .set = ovl_own_xattr_set,
 972 };
 973
 974 static const struct xattr_handler ovl_other_xattr_handler = {
 975         .prefix = "", /* catch all */
 976         .get = ovl_other_xattr_get,
 977         .set = ovl_other_xattr_set,
 978 };
 979
 980 static const struct xattr_handler *ovl_xattr_handlers[] = {
 981 #ifdef CONFIG_FS_POSIX_ACL
 982         &ovl_posix_acl_access_xattr_handler,
 983         &ovl_posix_acl_default_xattr_handler,
 984 #endif
 985         &ovl_own_xattr_handler,
 986         &ovl_other_xattr_handler,
 987         NULL
 988 };
 989
 990 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
 991                           struct inode **ptrap, const char *name)
 992 {
 993         struct inode *trap;
 994         int err;
 995
 996         trap = ovl_get_trap_inode(sb, dir);
 997         err = PTR_ERR_OR_ZERO(trap);
 998         if (err) {
 999                 if (err == -ELOOP)
1000                         pr_err("conflicting %s path\n", name);
1001                 return err;
1002         }
1003
1004         *ptrap = trap;
1005         return 0;
1006 }
1007
1008 /*
1009  * Determine how we treat concurrent use of upperdir/workdir based on the
1010  * index feature. This is papering over mount leaks of container runtimes,
1011  * for example, an old overlay mount is leaked and now its upperdir is
1012  * attempted to be used as a lower layer in a new overlay mount.
1013  */
1014 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1015 {
1016         if (ofs->config.index) {
1017                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1018                        name);
1019                 return -EBUSY;
1020         } else {
1021                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1022                         name);
1023                 return 0;
1024         }
1025 }
1026
1027 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1028                          struct path *upperpath)
1029 {
1030         struct vfsmount *upper_mnt;
1031         int err;
1032
1033         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1034         if (err)
1035                 goto out;
1036
1037         /* Upper fs should not be r/o */
1038         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
1039                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1040                 err = -EINVAL;
1041                 goto out;
1042         }
1043
1044         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1045         if (err)
1046                 goto out;
1047
1048         err = ovl_setup_trap(sb, upperpath->dentry, &ofs->upperdir_trap,
1049                              "upperdir");
1050         if (err)
1051                 goto out;
1052
1053         upper_mnt = clone_private_mount(upperpath);
1054         err = PTR_ERR(upper_mnt);
1055         if (IS_ERR(upper_mnt)) {
1056                 pr_err("failed to clone upperpath\n");
1057                 goto out;
1058         }
1059
1060         /* Don't inherit atime flags */
1061         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1062         ofs->upper_mnt = upper_mnt;
1063
1064         if (ovl_inuse_trylock(ofs->upper_mnt->mnt_root)) {
1065                 ofs->upperdir_locked = true;
1066         } else {
1067                 err = ovl_report_in_use(ofs, "upperdir");
1068                 if (err)
1069                         goto out;
1070         }
1071
1072         err = 0;
1073 out:
1074         return err;
1075 }
1076
1077 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1078                             struct path *workpath)
1079 {
1080         struct vfsmount *mnt = ofs->upper_mnt;
1081         struct dentry *temp;
1082         int fh_type;
1083         int err;
1084
1085         err = mnt_want_write(mnt);
1086         if (err)
1087                 return err;
1088
1089         ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1090         if (!ofs->workdir)
1091                 goto out;
1092
1093         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1094         if (err)
1095                 goto out;
1096
1097         /*
1098          * Upper should support d_type, else whiteouts are visible.  Given
1099          * workdir and upper are on same fs, we can do iterate_dir() on
1100          * workdir. This check requires successful creation of workdir in
1101          * previous step.
1102          */
1103         err = ovl_check_d_type_supported(workpath);
1104         if (err < 0)
1105                 goto out;
1106
1107         /*
1108          * We allowed this configuration and don't want to break users over
1109          * kernel upgrade. So warn instead of erroring out.
1110          */
1111         if (!err)
1112                 pr_warn("upper fs needs to support d_type.\n");
1113
1114         /* Check if upper/work fs supports O_TMPFILE */
1115         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1116         ofs->tmpfile = !IS_ERR(temp);
1117         if (ofs->tmpfile)
1118                 dput(temp);
1119         else
1120                 pr_warn("upper fs does not support tmpfile.\n");
1121
1122         /*
1123          * Check if upper/work fs supports trusted.overlay.* xattr
1124          */
1125         err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
1126         if (err) {
1127                 ofs->noxattr = true;
1128                 ofs->config.index = false;
1129                 ofs->config.metacopy = false;
1130                 pr_warn("upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1131                 err = 0;
1132         } else {
1133                 vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
1134         }
1135
1136         /* Check if upper/work fs supports file handles */
1137         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1138         if (ofs->config.index && !fh_type) {
1139                 ofs->config.index = false;
1140                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1141         }
1142
1143         /* Check if upper fs has 32bit inode numbers */
1144         if (fh_type != FILEID_INO32_GEN)
1145                 ofs->xino_mode = -1;
1146
1147         /* NFS export of r/w mount depends on index */
1148         if (ofs->config.nfs_export && !ofs->config.index) {
1149                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1150                 ofs->config.nfs_export = false;
1151         }
1152 out:
1153         mnt_drop_write(mnt);
1154         return err;
1155 }
1156
1157 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1158                            struct path *upperpath)
1159 {
1160         int err;
1161         struct path workpath = { };
1162
1163         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1164         if (err)
1165                 goto out;
1166
1167         err = -EINVAL;
1168         if (upperpath->mnt != workpath.mnt) {
1169                 pr_err("workdir and upperdir must reside under the same mount\n");
1170                 goto out;
1171         }
1172         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1173                 pr_err("workdir and upperdir must be separate subtrees\n");
1174                 goto out;
1175         }
1176
1177         ofs->workbasedir = dget(workpath.dentry);
1178
1179         if (ovl_inuse_trylock(ofs->workbasedir)) {
1180                 ofs->workdir_locked = true;
1181         } else {
1182                 err = ovl_report_in_use(ofs, "workdir");
1183                 if (err)
1184                         goto out;
1185         }
1186
1187         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1188                              "workdir");
1189         if (err)
1190                 goto out;
1191
1192         err = ovl_make_workdir(sb, ofs, &workpath);
1193
1194 out:
1195         path_put(&workpath);
1196
1197         return err;
1198 }
1199
1200 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1201                             struct ovl_entry *oe, struct path *upperpath)
1202 {
1203         struct vfsmount *mnt = ofs->upper_mnt;
1204         int err;
1205
1206         err = mnt_want_write(mnt);
1207         if (err)
1208                 return err;
1209
1210         /* Verify lower root is upper root origin */
1211         err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
1212                                 true);
1213         if (err) {
1214                 pr_err("failed to verify upper root origin\n");
1215                 goto out;
1216         }
1217
1218         ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1219         if (ofs->indexdir) {
1220                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1221                                      "indexdir");
1222                 if (err)
1223                         goto out;
1224
1225                 /*
1226                  * Verify upper root is exclusively associated with index dir.
1227                  * Older kernels stored upper fh in "trusted.overlay.origin"
1228                  * xattr. If that xattr exists, verify that it is a match to
1229                  * upper dir file handle. In any case, verify or set xattr
1230                  * "trusted.overlay.upper" to indicate that index may have
1231                  * directory entries.
1232                  */
1233                 if (ovl_check_origin_xattr(ofs->indexdir)) {
1234                         err = ovl_verify_set_fh(ofs->indexdir, OVL_XATTR_ORIGIN,
1235                                                 upperpath->dentry, true, false);
1236                         if (err)
1237                                 pr_err("failed to verify index dir 'origin' xattr\n");
1238                 }
1239                 err = ovl_verify_upper(ofs->indexdir, upperpath->dentry, true);
1240                 if (err)
1241                         pr_err("failed to verify index dir 'upper' xattr\n");
1242
1243                 /* Cleanup bad/stale/orphan index entries */
1244                 if (!err)
1245                         err = ovl_indexdir_cleanup(ofs);
1246         }
1247         if (err || !ofs->indexdir)
1248                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1249
1250 out:
1251         mnt_drop_write(mnt);
1252         return err;
1253 }
1254
1255 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1256 {
1257         unsigned int i;
1258
1259         if (!ofs->config.nfs_export && !ofs->upper_mnt)
1260                 return true;
1261
1262         for (i = 0; i < ofs->numfs; i++) {
1263                 /*
1264                  * We use uuid to associate an overlay lower file handle with a
1265                  * lower layer, so we can accept lower fs with null uuid as long
1266                  * as all lower layers with null uuid are on the same fs.
1267                  * if we detect multiple lower fs with the same uuid, we
1268                  * disable lower file handle decoding on all of them.
1269                  */
1270                 if (ofs->fs[i].is_lower &&
1271                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1272                         ofs->fs[i].bad_uuid = true;
1273                         return false;
1274                 }
1275         }
1276         return true;
1277 }
1278
1279 /* Get a unique fsid for the layer */
1280 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1281 {
1282         struct super_block *sb = path->mnt->mnt_sb;
1283         unsigned int i;
1284         dev_t dev;
1285         int err;
1286         bool bad_uuid = false;
1287
1288         for (i = 0; i < ofs->numfs; i++) {
1289                 if (ofs->fs[i].sb == sb)
1290                         return i;
1291         }
1292
1293         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1294                 bad_uuid = true;
1295                 if (ofs->config.index || ofs->config.nfs_export) {
1296                         ofs->config.index = false;
1297                         ofs->config.nfs_export = false;
1298                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to index=off,nfs_export=off.\n",
1299                                 uuid_is_null(&sb->s_uuid) ? "null" :
1300                                                             "conflicting",
1301                                 path->dentry);
1302                 }
1303         }
1304
1305         err = get_anon_bdev(&dev);
1306         if (err) {
1307                 pr_err("failed to get anonymous bdev for lowerpath\n");
1308                 return err;
1309         }
1310
1311         ofs->fs[ofs->numfs].sb = sb;
1312         ofs->fs[ofs->numfs].pseudo_dev = dev;
1313         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1314
1315         return ofs->numfs++;
1316 }
1317
1318 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1319                           struct path *stack, unsigned int numlower)
1320 {
1321         int err;
1322         unsigned int i;
1323         struct ovl_layer *layers;
1324
1325         err = -ENOMEM;
1326         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1327         if (!layers)
1328                 goto out;
1329         ofs->layers = layers;
1330
1331         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1332         if (ofs->fs == NULL)
1333                 goto out;
1334
1335         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1336         ofs->numfs++;
1337
1338         layers[0].mnt = ofs->upper_mnt;
1339         layers[0].idx = 0;
1340         layers[0].fsid = 0;
1341         ofs->numlayer = 1;
1342
1343         /*
1344          * All lower layers that share the same fs as upper layer, use the same
1345          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1346          * only overlay to simplify ovl_fs_free().
1347          * is_lower will be set if upper fs is shared with a lower layer.
1348          */
1349         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1350         if (err) {
1351                 pr_err("failed to get anonymous bdev for upper fs\n");
1352                 goto out;
1353         }
1354
1355         if (ofs->upper_mnt) {
1356                 ofs->fs[0].sb = ofs->upper_mnt->mnt_sb;
1357                 ofs->fs[0].is_lower = false;
1358         }
1359
1360         for (i = 0; i < numlower; i++) {
1361                 struct vfsmount *mnt;
1362                 struct inode *trap;
1363                 int fsid;
1364
1365                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1366                 if (err < 0)
1367                         goto out;
1368
1369                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1370                 if (err)
1371                         goto out;
1372
1373                 if (ovl_is_inuse(stack[i].dentry)) {
1374                         err = ovl_report_in_use(ofs, "lowerdir");
1375                         if (err)
1376                                 goto out;
1377                 }
1378
1379                 mnt = clone_private_mount(&stack[i]);
1380                 err = PTR_ERR(mnt);
1381                 if (IS_ERR(mnt)) {
1382                         pr_err("failed to clone lowerpath\n");
1383                         iput(trap);
1384                         goto out;
1385                 }
1386
1387                 /*
1388                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1389                  * will fail instead of modifying lower fs.
1390                  */
1391                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1392
1393                 layers[ofs->numlayer].trap = trap;
1394                 layers[ofs->numlayer].mnt = mnt;
1395                 layers[ofs->numlayer].idx = ofs->numlayer;
1396                 layers[ofs->numlayer].fsid = fsid;
1397                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1398                 ofs->numlayer++;
1399                 ofs->fs[fsid].is_lower = true;
1400         }
1401
1402         /*
1403          * When all layers on same fs, overlay can use real inode numbers.
1404          * With mount option "xino=on", mounter declares that there are enough
1405          * free high bits in underlying fs to hold the unique fsid.
1406          * If overlayfs does encounter underlying inodes using the high xino
1407          * bits reserved for fsid, it emits a warning and uses the original
1408          * inode number.
1409          */
1410         if (ofs->numfs - !ofs->upper_mnt == 1) {
1411                 if (ofs->config.xino == OVL_XINO_ON)
1412                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1413                 ofs->xino_mode = 0;
1414         } else if (ofs->config.xino == OVL_XINO_OFF) {
1415                 ofs->xino_mode = -1;
1416         } else if (ofs->config.xino == OVL_XINO_ON && ofs->xino_mode < 0) {
1417                 /*
1418                  * This is a roundup of number of bits needed for encoding
1419                  * fsid, where fsid 0 is reserved for upper fs even with
1420                  * lower only overlay.
1421                  */
1422                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 31);
1423                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 1;
1424         }
1425
1426         if (ofs->xino_mode > 0) {
1427                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1428                         ofs->xino_mode);
1429         }
1430
1431         err = 0;
1432 out:
1433         return err;
1434 }
1435
1436 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1437                                             struct ovl_fs *ofs)
1438 {
1439         int err;
1440         char *lowertmp, *lower;
1441         struct path *stack = NULL;
1442         unsigned int stacklen, numlower = 0, i;
1443         bool remote = false;
1444         struct ovl_entry *oe;
1445
1446         err = -ENOMEM;
1447         lowertmp = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1448         if (!lowertmp)
1449                 goto out_err;
1450
1451         err = -EINVAL;
1452         stacklen = ovl_split_lowerdirs(lowertmp);
1453         if (stacklen > OVL_MAX_STACK) {
1454                 pr_err("too many lower directories, limit is %d\n",
1455                        OVL_MAX_STACK);
1456                 goto out_err;
1457         } else if (!ofs->config.upperdir && stacklen == 1) {
1458                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1459                 goto out_err;
1460         } else if (!ofs->config.upperdir && ofs->config.nfs_export &&
1461                    ofs->config.redirect_follow) {
1462                 pr_warn("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
1463                 ofs->config.nfs_export = false;
1464         }
1465
1466         err = -ENOMEM;
1467         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1468         if (!stack)
1469                 goto out_err;
1470
1471         err = -EINVAL;
1472         lower = lowertmp;
1473         for (numlower = 0; numlower < stacklen; numlower++) {
1474                 err = ovl_lower_dir(lower, &stack[numlower], ofs,
1475                                     &sb->s_stack_depth, &remote);
1476                 if (err)
1477                         goto out_err;
1478
1479                 lower = strchr(lower, '\0') + 1;
1480         }
1481
1482         err = -EINVAL;
1483         sb->s_stack_depth++;
1484         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1485                 pr_err("maximum fs stacking depth exceeded\n");
1486                 goto out_err;
1487         }
1488
1489         err = ovl_get_layers(sb, ofs, stack, numlower);
1490         if (err)
1491                 goto out_err;
1492
1493         err = -ENOMEM;
1494         oe = ovl_alloc_entry(numlower);
1495         if (!oe)
1496                 goto out_err;
1497
1498         for (i = 0; i < numlower; i++) {
1499                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1500                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1501         }
1502
1503         if (remote)
1504                 sb->s_d_op = &ovl_reval_dentry_operations;
1505         else
1506                 sb->s_d_op = &ovl_dentry_operations;
1507
1508 out:
1509         for (i = 0; i < numlower; i++)
1510                 path_put(&stack[i]);
1511         kfree(stack);
1512         kfree(lowertmp);
1513
1514         return oe;
1515
1516 out_err:
1517         oe = ERR_PTR(err);
1518         goto out;
1519 }
1520
1521 /*
1522  * Check if this layer root is a descendant of:
1523  * - another layer of this overlayfs instance
1524  * - upper/work dir of any overlayfs instance
1525  */
1526 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1527                            struct dentry *dentry, const char *name)
1528 {
1529         struct dentry *next = dentry, *parent;
1530         int err = 0;
1531
1532         if (!dentry)
1533                 return 0;
1534
1535         parent = dget_parent(next);
1536
1537         /* Walk back ancestors to root (inclusive) looking for traps */
1538         while (!err && parent != next) {
1539                 if (ovl_lookup_trap_inode(sb, parent)) {
1540                         err = -ELOOP;
1541                         pr_err("overlapping %s path\n", name);
1542                 } else if (ovl_is_inuse(parent)) {
1543                         err = ovl_report_in_use(ofs, name);
1544                 }
1545                 next = parent;
1546                 parent = dget_parent(next);
1547                 dput(next);
1548         }
1549
1550         dput(parent);
1551
1552         return err;
1553 }
1554
1555 /*
1556  * Check if any of the layers or work dirs overlap.
1557  */
1558 static int ovl_check_overlapping_layers(struct super_block *sb,
1559                                         struct ovl_fs *ofs)
1560 {
1561         int i, err;
1562
1563         if (ofs->upper_mnt) {
1564                 err = ovl_check_layer(sb, ofs, ofs->upper_mnt->mnt_root,
1565                                       "upperdir");
1566                 if (err)
1567                         return err;
1568
1569                 /*
1570                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1571                  * this instance and covers overlapping work and index dirs,
1572                  * unless work or index dir have been moved since created inside
1573                  * workbasedir.  In that case, we already have their traps in
1574                  * inode cache and we will catch that case on lookup.
1575                  */
1576                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir");
1577                 if (err)
1578                         return err;
1579         }
1580
1581         for (i = 1; i < ofs->numlayer; i++) {
1582                 err = ovl_check_layer(sb, ofs,
1583                                       ofs->layers[i].mnt->mnt_root,
1584                                       "lowerdir");
1585                 if (err)
1586                         return err;
1587         }
1588
1589         return 0;
1590 }
1591
1592 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1593 {
1594         struct path upperpath = { };
1595         struct dentry *root_dentry;
1596         struct ovl_entry *oe;
1597         struct ovl_fs *ofs;
1598         struct cred *cred;
1599         int err;
1600
1601         err = -ENOMEM;
1602         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1603         if (!ofs)
1604                 goto out;
1605
1606         ofs->creator_cred = cred = prepare_creds();
1607         if (!cred)
1608                 goto out_err;
1609
1610         ofs->config.index = ovl_index_def;
1611         ofs->config.nfs_export = ovl_nfs_export_def;
1612         ofs->config.xino = ovl_xino_def();
1613         ofs->config.metacopy = ovl_metacopy_def;
1614         err = ovl_parse_opt((char *) data, &ofs->config);
1615         if (err)
1616                 goto out_err;
1617
1618         err = -EINVAL;
1619         if (!ofs->config.lowerdir) {
1620                 if (!silent)
1621                         pr_err("missing 'lowerdir'\n");
1622                 goto out_err;
1623         }
1624
1625         sb->s_stack_depth = 0;
1626         sb->s_maxbytes = MAX_LFS_FILESIZE;
1627         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1628         if (ofs->config.xino != OVL_XINO_OFF) {
1629                 ofs->xino_mode = BITS_PER_LONG - 32;
1630                 if (!ofs->xino_mode) {
1631                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1632                         ofs->config.xino = OVL_XINO_OFF;
1633                 }
1634         }
1635
1636         /* alloc/destroy_inode needed for setting up traps in inode cache */
1637         sb->s_op = &ovl_super_operations;
1638
1639         if (ofs->config.upperdir) {
1640                 if (!ofs->config.workdir) {
1641                         pr_err("missing 'workdir'\n");
1642                         goto out_err;
1643                 }
1644
1645                 err = ovl_get_upper(sb, ofs, &upperpath);
1646                 if (err)
1647                         goto out_err;
1648
1649                 err = ovl_get_workdir(sb, ofs, &upperpath);
1650                 if (err)
1651                         goto out_err;
1652
1653                 if (!ofs->workdir)
1654                         sb->s_flags |= SB_RDONLY;
1655
1656                 sb->s_stack_depth = ofs->upper_mnt->mnt_sb->s_stack_depth;
1657                 sb->s_time_gran = ofs->upper_mnt->mnt_sb->s_time_gran;
1658
1659         }
1660         oe = ovl_get_lowerstack(sb, ofs);
1661         err = PTR_ERR(oe);
1662         if (IS_ERR(oe))
1663                 goto out_err;
1664
1665         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1666         if (!ofs->upper_mnt)
1667                 sb->s_flags |= SB_RDONLY;
1668
1669         if (!(ovl_force_readonly(ofs)) && ofs->config.index) {
1670                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
1671                 if (err)
1672                         goto out_free_oe;
1673
1674                 /* Force r/o mount with no index dir */
1675                 if (!ofs->indexdir) {
1676                         dput(ofs->workdir);
1677                         ofs->workdir = NULL;
1678                         sb->s_flags |= SB_RDONLY;
1679                 }
1680
1681         }
1682
1683         err = ovl_check_overlapping_layers(sb, ofs);
1684         if (err)
1685                 goto out_free_oe;
1686
1687         /* Show index=off in /proc/mounts for forced r/o mount */
1688         if (!ofs->indexdir) {
1689                 ofs->config.index = false;
1690                 if (ofs->upper_mnt && ofs->config.nfs_export) {
1691                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
1692                         ofs->config.nfs_export = false;
1693                 }
1694         }
1695
1696         if (ofs->config.metacopy && ofs->config.nfs_export) {
1697                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1698                 ofs->config.nfs_export = false;
1699         }
1700
1701         if (ofs->config.nfs_export)
1702                 sb->s_export_op = &ovl_export_operations;
1703
1704         /* Never override disk quota limits or use reserved space */
1705         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1706
1707         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1708         sb->s_xattr = ovl_xattr_handlers;
1709         sb->s_fs_info = ofs;
1710         sb->s_flags |= SB_POSIXACL;
1711
1712         err = -ENOMEM;
1713         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1714         if (!root_dentry)
1715                 goto out_free_oe;
1716
1717         root_dentry->d_fsdata = oe;
1718
1719         mntput(upperpath.mnt);
1720         if (upperpath.dentry) {
1721                 ovl_dentry_set_upper_alias(root_dentry);
1722                 if (ovl_is_impuredir(upperpath.dentry))
1723                         ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1724         }
1725
1726         /* Root is always merge -> can have whiteouts */
1727         ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1728         ovl_dentry_set_flag(OVL_E_CONNECTED, root_dentry);
1729         ovl_set_upperdata(d_inode(root_dentry));
1730         ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1731                        ovl_dentry_lower(root_dentry), NULL);
1732
1733         sb->s_root = root_dentry;
1734
1735         return 0;
1736
1737 out_free_oe:
1738         ovl_entry_stack_free(oe);
1739         kfree(oe);
1740 out_err:
1741         path_put(&upperpath);
1742         ovl_free_fs(ofs);
1743 out:
1744         return err;
1745 }
1746
1747 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1748                                 const char *dev_name, void *raw_data)
1749 {
1750         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1751 }
1752
1753 static struct file_system_type ovl_fs_type = {
1754         .owner          = THIS_MODULE,
1755         .name           = "overlay",
1756         .mount          = ovl_mount,
1757         .kill_sb        = kill_anon_super,
1758 };
1759 MODULE_ALIAS_FS("overlay");
1760
1761 static void ovl_inode_init_once(void *foo)
1762 {
1763         struct ovl_inode *oi = foo;
1764
1765         inode_init_once(&oi->vfs_inode);
1766 }
1767
1768 static int __init ovl_init(void)
1769 {
1770         int err;
1771
1772         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1773                                              sizeof(struct ovl_inode), 0,
1774                                              (SLAB_RECLAIM_ACCOUNT|
1775                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1776                                              ovl_inode_init_once);
1777         if (ovl_inode_cachep == NULL)
1778                 return -ENOMEM;
1779
1780         err = ovl_aio_request_cache_init();
1781         if (!err) {
1782                 err = register_filesystem(&ovl_fs_type);
1783                 if (!err)
1784                         return 0;
1785
1786                 ovl_aio_request_cache_destroy();
1787         }
1788         kmem_cache_destroy(ovl_inode_cachep);
1789
1790         return err;
1791 }
1792
1793 static void __exit ovl_exit(void)
1794 {
1795         unregister_filesystem(&ovl_fs_type);
1796
1797         /*
1798          * Make sure all delayed rcu free inodes are flushed before we
1799          * destroy cache.
1800          */
1801         rcu_barrier();
1802         kmem_cache_destroy(ovl_inode_cachep);
1803         ovl_aio_request_cache_destroy();
1804 }
1805
1806 module_init(ovl_init);
1807 module_exit(ovl_exit);