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