fs/overlayfs/super.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *
   4  * Copyright (C) 2011 Novell Inc.
   5  */
   6
   7 #include <uapi/linux/magic.h>
   8 #include <linux/fs.h>
   9 #include <linux/namei.h>
  10 #include <linux/xattr.h>
  11 #include <linux/mount.h>
  12 #include <linux/parser.h>
  13 #include <linux/module.h>
  14 #include <linux/statfs.h>
  15 #include <linux/seq_file.h>
  16 #include <linux/posix_acl_xattr.h>
  17 #include <linux/exportfs.h>
  18 #include "overlayfs.h"
  19
  20 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  21 MODULE_DESCRIPTION("Overlay filesystem");
  22 MODULE_LICENSE("GPL");
  23
  24
  25 struct ovl_dir_cache;
  26
  27 #define OVL_MAX_STACK 500
  28
  29 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  30 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  31 MODULE_PARM_DESC(redirect_dir,
  32                  "Default to on or off for the redirect_dir feature");
  33
  34 static bool ovl_redirect_always_follow =
  35         IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
  36 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
  37                    bool, 0644);
  38 MODULE_PARM_DESC(redirect_always_follow,
  39                  "Follow redirects even if redirect_dir feature is turned off");
  40
  41 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
  42 module_param_named(index, ovl_index_def, bool, 0644);
  43 MODULE_PARM_DESC(index,
  44                  "Default to on or off for the inodes index feature");
  45
  46 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
  47 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
  48 MODULE_PARM_DESC(nfs_export,
  49                  "Default to on or off for the NFS export feature");
  50
  51 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
  52 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
  53 MODULE_PARM_DESC(xino_auto,
  54                  "Auto enable xino feature");
  55
  56 static void ovl_entry_stack_free(struct ovl_entry *oe)
  57 {
  58         unsigned int i;
  59
  60         for (i = 0; i < oe->numlower; i++)
  61                 dput(oe->lowerstack[i].dentry);
  62 }
  63
  64 static bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
  65 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
  66 MODULE_PARM_DESC(metacopy,
  67                  "Default to on or off for the metadata only copy up feature");
  68
  69 static void ovl_dentry_release(struct dentry *dentry)
  70 {
  71         struct ovl_entry *oe = dentry->d_fsdata;
  72
  73         if (oe) {
  74                 ovl_entry_stack_free(oe);
  75                 kfree_rcu(oe, rcu);
  76         }
  77 }
  78
  79 static struct dentry *ovl_d_real(struct dentry *dentry,
  80                                  const struct inode *inode)
  81 {
  82         struct dentry *real = NULL, *lower;
  83
  84         /* It's an overlay file */
  85         if (inode && d_inode(dentry) == inode)
  86                 return dentry;
  87
  88         if (!d_is_reg(dentry)) {
  89                 if (!inode || inode == d_inode(dentry))
  90                         return dentry;
  91                 goto bug;
  92         }
  93
  94         real = ovl_dentry_upper(dentry);
  95         if (real && (inode == d_inode(real)))
  96                 return real;
  97
  98         if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
  99                 return real;
 100
 101         lower = ovl_dentry_lowerdata(dentry);
 102         if (!lower)
 103                 goto bug;
 104         real = lower;
 105
 106         /* Handle recursion */
 107         real = d_real(real, inode);
 108
 109         if (!inode || inode == d_inode(real))
 110                 return real;
 111 bug:
 112         WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
 113              __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
 114              inode ? inode->i_ino : 0, real,
 115              real && d_inode(real) ? d_inode(real)->i_ino : 0);
 116         return dentry;
 117 }
 118
 119 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
 120 {
 121         int ret = 1;
 122
 123         if (weak) {
 124                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
 125                         ret =  d->d_op->d_weak_revalidate(d, flags);
 126         } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
 127                 ret = d->d_op->d_revalidate(d, flags);
 128                 if (!ret) {
 129                         if (!(flags & LOOKUP_RCU))
 130                                 d_invalidate(d);
 131                         ret = -ESTALE;
 132                 }
 133         }
 134         return ret;
 135 }
 136
 137 static int ovl_dentry_revalidate_common(struct dentry *dentry,
 138                                         unsigned int flags, bool weak)
 139 {
 140         struct ovl_entry *oe = dentry->d_fsdata;
 141         struct dentry *upper;
 142         unsigned int i;
 143         int ret = 1;
 144
 145         upper = ovl_dentry_upper(dentry);
 146         if (upper)
 147                 ret = ovl_revalidate_real(upper, flags, weak);
 148
 149         for (i = 0; ret > 0 && i < oe->numlower; i++) {
 150                 ret = ovl_revalidate_real(oe->lowerstack[i].dentry, flags,
 151                                           weak);
 152         }
 153         return ret;
 154 }
 155
 156 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 157 {
 158         return ovl_dentry_revalidate_common(dentry, flags, false);
 159 }
 160
 161 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 162 {
 163         return ovl_dentry_revalidate_common(dentry, flags, true);
 164 }
 165
 166 static const struct dentry_operations ovl_dentry_operations = {
 167         .d_release = ovl_dentry_release,
 168         .d_real = ovl_d_real,
 169         .d_revalidate = ovl_dentry_revalidate,
 170         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 171 };
 172
 173 static struct kmem_cache *ovl_inode_cachep;
 174
 175 static struct inode *ovl_alloc_inode(struct super_block *sb)
 176 {
 177         struct ovl_inode *oi = alloc_inode_sb(sb, ovl_inode_cachep, GFP_KERNEL);
 178
 179         if (!oi)
 180                 return NULL;
 181
 182         oi->cache = NULL;
 183         oi->redirect = NULL;
 184         oi->version = 0;
 185         oi->flags = 0;
 186         oi->__upperdentry = NULL;
 187         oi->lowerpath.dentry = NULL;
 188         oi->lowerpath.layer = NULL;
 189         oi->lowerdata = NULL;
 190         mutex_init(&oi->lock);
 191
 192         return &oi->vfs_inode;
 193 }
 194
 195 static void ovl_free_inode(struct inode *inode)
 196 {
 197         struct ovl_inode *oi = OVL_I(inode);
 198
 199         kfree(oi->redirect);
 200         mutex_destroy(&oi->lock);
 201         kmem_cache_free(ovl_inode_cachep, oi);
 202 }
 203
 204 static void ovl_destroy_inode(struct inode *inode)
 205 {
 206         struct ovl_inode *oi = OVL_I(inode);
 207
 208         dput(oi->__upperdentry);
 209         dput(oi->lowerpath.dentry);
 210         if (S_ISDIR(inode->i_mode))
 211                 ovl_dir_cache_free(inode);
 212         else
 213                 iput(oi->lowerdata);
 214 }
 215
 216 static void ovl_free_fs(struct ovl_fs *ofs)
 217 {
 218         struct vfsmount **mounts;
 219         unsigned i;
 220
 221         iput(ofs->workbasedir_trap);
 222         iput(ofs->indexdir_trap);
 223         iput(ofs->workdir_trap);
 224         dput(ofs->whiteout);
 225         dput(ofs->indexdir);
 226         dput(ofs->workdir);
 227         if (ofs->workdir_locked)
 228                 ovl_inuse_unlock(ofs->workbasedir);
 229         dput(ofs->workbasedir);
 230         if (ofs->upperdir_locked)
 231                 ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
 232
 233         /* Hack!  Reuse ofs->layers as a vfsmount array before freeing it */
 234         mounts = (struct vfsmount **) ofs->layers;
 235         for (i = 0; i < ofs->numlayer; i++) {
 236                 iput(ofs->layers[i].trap);
 237                 mounts[i] = ofs->layers[i].mnt;
 238         }
 239         kern_unmount_array(mounts, ofs->numlayer);
 240         kfree(ofs->layers);
 241         for (i = 0; i < ofs->numfs; i++)
 242                 free_anon_bdev(ofs->fs[i].pseudo_dev);
 243         kfree(ofs->fs);
 244
 245         kfree(ofs->config.lowerdir);
 246         kfree(ofs->config.upperdir);
 247         kfree(ofs->config.workdir);
 248         kfree(ofs->config.redirect_mode);
 249         if (ofs->creator_cred)
 250                 put_cred(ofs->creator_cred);
 251         kfree(ofs);
 252 }
 253
 254 static void ovl_put_super(struct super_block *sb)
 255 {
 256         struct ovl_fs *ofs = sb->s_fs_info;
 257
 258         ovl_free_fs(ofs);
 259 }
 260
 261 /* Sync real dirty inodes in upper filesystem (if it exists) */
 262 static int ovl_sync_fs(struct super_block *sb, int wait)
 263 {
 264         struct ovl_fs *ofs = sb->s_fs_info;
 265         struct super_block *upper_sb;
 266         int ret;
 267
 268         ret = ovl_sync_status(ofs);
 269         /*
 270          * We have to always set the err, because the return value isn't
 271          * checked in syncfs, and instead indirectly return an error via
 272          * the sb's writeback errseq, which VFS inspects after this call.
 273          */
 274         if (ret < 0) {
 275                 errseq_set(&sb->s_wb_err, -EIO);
 276                 return -EIO;
 277         }
 278
 279         if (!ret)
 280                 return ret;
 281
 282         /*
 283          * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
 284          * All the super blocks will be iterated, including upper_sb.
 285          *
 286          * If this is a syncfs(2) call, then we do need to call
 287          * sync_filesystem() on upper_sb, but enough if we do it when being
 288          * called with wait == 1.
 289          */
 290         if (!wait)
 291                 return 0;
 292
 293         upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 294
 295         down_read(&upper_sb->s_umount);
 296         ret = sync_filesystem(upper_sb);
 297         up_read(&upper_sb->s_umount);
 298
 299         return ret;
 300 }
 301
 302 /**
 303  * ovl_statfs
 304  * @sb: The overlayfs super block
 305  * @buf: The struct kstatfs to fill in with stats
 306  *
 307  * Get the filesystem statistics.  As writes always target the upper layer
 308  * filesystem pass the statfs to the upper filesystem (if it exists)
 309  */
 310 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 311 {
 312         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 313         struct dentry *root_dentry = dentry->d_sb->s_root;
 314         struct path path;
 315         int err;
 316
 317         ovl_path_real(root_dentry, &path);
 318
 319         err = vfs_statfs(&path, buf);
 320         if (!err) {
 321                 buf->f_namelen = ofs->namelen;
 322                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 323         }
 324
 325         return err;
 326 }
 327
 328 /* Will this overlay be forced to mount/remount ro? */
 329 static bool ovl_force_readonly(struct ovl_fs *ofs)
 330 {
 331         return (!ovl_upper_mnt(ofs) || !ofs->workdir);
 332 }
 333
 334 static const char *ovl_redirect_mode_def(void)
 335 {
 336         return ovl_redirect_dir_def ? "on" : "off";
 337 }
 338
 339 static const char * const ovl_xino_str[] = {
 340         "off",
 341         "auto",
 342         "on",
 343 };
 344
 345 static inline int ovl_xino_def(void)
 346 {
 347         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 348 }
 349
 350 /**
 351  * ovl_show_options
 352  *
 353  * Prints the mount options for a given superblock.
 354  * Returns zero; does not fail.
 355  */
 356 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 357 {
 358         struct super_block *sb = dentry->d_sb;
 359         struct ovl_fs *ofs = sb->s_fs_info;
 360
 361         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 362         if (ofs->config.upperdir) {
 363                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 364                 seq_show_option(m, "workdir", ofs->config.workdir);
 365         }
 366         if (ofs->config.default_permissions)
 367                 seq_puts(m, ",default_permissions");
 368         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 369                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 370         if (ofs->config.index != ovl_index_def)
 371                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 372         if (!ofs->config.uuid)
 373                 seq_puts(m, ",uuid=off");
 374         if (ofs->config.nfs_export != ovl_nfs_export_def)
 375                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 376                                                 "on" : "off");
 377         if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
 378                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 379         if (ofs->config.metacopy != ovl_metacopy_def)
 380                 seq_printf(m, ",metacopy=%s",
 381                            ofs->config.metacopy ? "on" : "off");
 382         if (ofs->config.ovl_volatile)
 383                 seq_puts(m, ",volatile");
 384         if (ofs->config.userxattr)
 385                 seq_puts(m, ",userxattr");
 386         return 0;
 387 }
 388
 389 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 390 {
 391         struct ovl_fs *ofs = sb->s_fs_info;
 392         struct super_block *upper_sb;
 393         int ret = 0;
 394
 395         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 396                 return -EROFS;
 397
 398         if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
 399                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 400                 if (ovl_should_sync(ofs)) {
 401                         down_read(&upper_sb->s_umount);
 402                         ret = sync_filesystem(upper_sb);
 403                         up_read(&upper_sb->s_umount);
 404                 }
 405         }
 406
 407         return ret;
 408 }
 409
 410 static const struct super_operations ovl_super_operations = {
 411         .alloc_inode    = ovl_alloc_inode,
 412         .free_inode     = ovl_free_inode,
 413         .destroy_inode  = ovl_destroy_inode,
 414         .drop_inode     = generic_delete_inode,
 415         .put_super      = ovl_put_super,
 416         .sync_fs        = ovl_sync_fs,
 417         .statfs         = ovl_statfs,
 418         .show_options   = ovl_show_options,
 419         .remount_fs     = ovl_remount,
 420 };
 421
 422 enum {
 423         OPT_LOWERDIR,
 424         OPT_UPPERDIR,
 425         OPT_WORKDIR,
 426         OPT_DEFAULT_PERMISSIONS,
 427         OPT_REDIRECT_DIR,
 428         OPT_INDEX_ON,
 429         OPT_INDEX_OFF,
 430         OPT_UUID_ON,
 431         OPT_UUID_OFF,
 432         OPT_NFS_EXPORT_ON,
 433         OPT_USERXATTR,
 434         OPT_NFS_EXPORT_OFF,
 435         OPT_XINO_ON,
 436         OPT_XINO_OFF,
 437         OPT_XINO_AUTO,
 438         OPT_METACOPY_ON,
 439         OPT_METACOPY_OFF,
 440         OPT_VOLATILE,
 441         OPT_ERR,
 442 };
 443
 444 static const match_table_t ovl_tokens = {
 445         {OPT_LOWERDIR,                  "lowerdir=%s"},
 446         {OPT_UPPERDIR,                  "upperdir=%s"},
 447         {OPT_WORKDIR,                   "workdir=%s"},
 448         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 449         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 450         {OPT_INDEX_ON,                  "index=on"},
 451         {OPT_INDEX_OFF,                 "index=off"},
 452         {OPT_USERXATTR,                 "userxattr"},
 453         {OPT_UUID_ON,                   "uuid=on"},
 454         {OPT_UUID_OFF,                  "uuid=off"},
 455         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 456         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 457         {OPT_XINO_ON,                   "xino=on"},
 458         {OPT_XINO_OFF,                  "xino=off"},
 459         {OPT_XINO_AUTO,                 "xino=auto"},
 460         {OPT_METACOPY_ON,               "metacopy=on"},
 461         {OPT_METACOPY_OFF,              "metacopy=off"},
 462         {OPT_VOLATILE,                  "volatile"},
 463         {OPT_ERR,                       NULL}
 464 };
 465
 466 static char *ovl_next_opt(char **s)
 467 {
 468         char *sbegin = *s;
 469         char *p;
 470
 471         if (sbegin == NULL)
 472                 return NULL;
 473
 474         for (p = sbegin; *p; p++) {
 475                 if (*p == '\\') {
 476                         p++;
 477                         if (!*p)
 478                                 break;
 479                 } else if (*p == ',') {
 480                         *p = '\0';
 481                         *s = p + 1;
 482                         return sbegin;
 483                 }
 484         }
 485         *s = NULL;
 486         return sbegin;
 487 }
 488
 489 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 490 {
 491         if (strcmp(mode, "on") == 0) {
 492                 config->redirect_dir = true;
 493                 /*
 494                  * Does not make sense to have redirect creation without
 495                  * redirect following.
 496                  */
 497                 config->redirect_follow = true;
 498         } else if (strcmp(mode, "follow") == 0) {
 499                 config->redirect_follow = true;
 500         } else if (strcmp(mode, "off") == 0) {
 501                 if (ovl_redirect_always_follow)
 502                         config->redirect_follow = true;
 503         } else if (strcmp(mode, "nofollow") != 0) {
 504                 pr_err("bad mount option \"redirect_dir=%s\"\n",
 505                        mode);
 506                 return -EINVAL;
 507         }
 508
 509         return 0;
 510 }
 511
 512 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 513 {
 514         char *p;
 515         int err;
 516         bool metacopy_opt = false, redirect_opt = false;
 517         bool nfs_export_opt = false, index_opt = false;
 518
 519         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 520         if (!config->redirect_mode)
 521                 return -ENOMEM;
 522
 523         while ((p = ovl_next_opt(&opt)) != NULL) {
 524                 int token;
 525                 substring_t args[MAX_OPT_ARGS];
 526
 527                 if (!*p)
 528                         continue;
 529
 530                 token = match_token(p, ovl_tokens, args);
 531                 switch (token) {
 532                 case OPT_UPPERDIR:
 533                         kfree(config->upperdir);
 534                         config->upperdir = match_strdup(&args[0]);
 535                         if (!config->upperdir)
 536                                 return -ENOMEM;
 537                         break;
 538
 539                 case OPT_LOWERDIR:
 540                         kfree(config->lowerdir);
 541                         config->lowerdir = match_strdup(&args[0]);
 542                         if (!config->lowerdir)
 543                                 return -ENOMEM;
 544                         break;
 545
 546                 case OPT_WORKDIR:
 547                         kfree(config->workdir);
 548                         config->workdir = match_strdup(&args[0]);
 549                         if (!config->workdir)
 550                                 return -ENOMEM;
 551                         break;
 552
 553                 case OPT_DEFAULT_PERMISSIONS:
 554                         config->default_permissions = true;
 555                         break;
 556
 557                 case OPT_REDIRECT_DIR:
 558                         kfree(config->redirect_mode);
 559                         config->redirect_mode = match_strdup(&args[0]);
 560                         if (!config->redirect_mode)
 561                                 return -ENOMEM;
 562                         redirect_opt = true;
 563                         break;
 564
 565                 case OPT_INDEX_ON:
 566                         config->index = true;
 567                         index_opt = true;
 568                         break;
 569
 570                 case OPT_INDEX_OFF:
 571                         config->index = false;
 572                         index_opt = true;
 573                         break;
 574
 575                 case OPT_UUID_ON:
 576                         config->uuid = true;
 577                         break;
 578
 579                 case OPT_UUID_OFF:
 580                         config->uuid = false;
 581                         break;
 582
 583                 case OPT_NFS_EXPORT_ON:
 584                         config->nfs_export = true;
 585                         nfs_export_opt = true;
 586                         break;
 587
 588                 case OPT_NFS_EXPORT_OFF:
 589                         config->nfs_export = false;
 590                         nfs_export_opt = true;
 591                         break;
 592
 593                 case OPT_XINO_ON:
 594                         config->xino = OVL_XINO_ON;
 595                         break;
 596
 597                 case OPT_XINO_OFF:
 598                         config->xino = OVL_XINO_OFF;
 599                         break;
 600
 601                 case OPT_XINO_AUTO:
 602                         config->xino = OVL_XINO_AUTO;
 603                         break;
 604
 605                 case OPT_METACOPY_ON:
 606                         config->metacopy = true;
 607                         metacopy_opt = true;
 608                         break;
 609
 610                 case OPT_METACOPY_OFF:
 611                         config->metacopy = false;
 612                         metacopy_opt = true;
 613                         break;
 614
 615                 case OPT_VOLATILE:
 616                         config->ovl_volatile = true;
 617                         break;
 618
 619                 case OPT_USERXATTR:
 620                         config->userxattr = true;
 621                         break;
 622
 623                 default:
 624                         pr_err("unrecognized mount option \"%s\" or missing value\n",
 625                                         p);
 626                         return -EINVAL;
 627                 }
 628         }
 629
 630         /* Workdir/index are useless in non-upper mount */
 631         if (!config->upperdir) {
 632                 if (config->workdir) {
 633                         pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 634                                 config->workdir);
 635                         kfree(config->workdir);
 636                         config->workdir = NULL;
 637                 }
 638                 if (config->index && index_opt) {
 639                         pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
 640                         index_opt = false;
 641                 }
 642                 config->index = false;
 643         }
 644
 645         if (!config->upperdir && config->ovl_volatile) {
 646                 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
 647                 config->ovl_volatile = false;
 648         }
 649
 650         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 651         if (err)
 652                 return err;
 653
 654         /*
 655          * This is to make the logic below simpler.  It doesn't make any other
 656          * difference, since config->redirect_dir is only used for upper.
 657          */
 658         if (!config->upperdir && config->redirect_follow)
 659                 config->redirect_dir = true;
 660
 661         /* Resolve metacopy -> redirect_dir dependency */
 662         if (config->metacopy && !config->redirect_dir) {
 663                 if (metacopy_opt && redirect_opt) {
 664                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
 665                                config->redirect_mode);
 666                         return -EINVAL;
 667                 }
 668                 if (redirect_opt) {
 669                         /*
 670                          * There was an explicit redirect_dir=... that resulted
 671                          * in this conflict.
 672                          */
 673                         pr_info("disabling metacopy due to redirect_dir=%s\n",
 674                                 config->redirect_mode);
 675                         config->metacopy = false;
 676                 } else {
 677                         /* Automatically enable redirect otherwise. */
 678                         config->redirect_follow = config->redirect_dir = true;
 679                 }
 680         }
 681
 682         /* Resolve nfs_export -> index dependency */
 683         if (config->nfs_export && !config->index) {
 684                 if (!config->upperdir && config->redirect_follow) {
 685                         pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
 686                         config->nfs_export = false;
 687                 } else if (nfs_export_opt && index_opt) {
 688                         pr_err("conflicting options: nfs_export=on,index=off\n");
 689                         return -EINVAL;
 690                 } else if (index_opt) {
 691                         /*
 692                          * There was an explicit index=off that resulted
 693                          * in this conflict.
 694                          */
 695                         pr_info("disabling nfs_export due to index=off\n");
 696                         config->nfs_export = false;
 697                 } else {
 698                         /* Automatically enable index otherwise. */
 699                         config->index = true;
 700                 }
 701         }
 702
 703         /* Resolve nfs_export -> !metacopy dependency */
 704         if (config->nfs_export && config->metacopy) {
 705                 if (nfs_export_opt && metacopy_opt) {
 706                         pr_err("conflicting options: nfs_export=on,metacopy=on\n");
 707                         return -EINVAL;
 708                 }
 709                 if (metacopy_opt) {
 710                         /*
 711                          * There was an explicit metacopy=on that resulted
 712                          * in this conflict.
 713                          */
 714                         pr_info("disabling nfs_export due to metacopy=on\n");
 715                         config->nfs_export = false;
 716                 } else {
 717                         /*
 718                          * There was an explicit nfs_export=on that resulted
 719                          * in this conflict.
 720                          */
 721                         pr_info("disabling metacopy due to nfs_export=on\n");
 722                         config->metacopy = false;
 723                 }
 724         }
 725
 726
 727         /* Resolve userxattr -> !redirect && !metacopy dependency */
 728         if (config->userxattr) {
 729                 if (config->redirect_follow && redirect_opt) {
 730                         pr_err("conflicting options: userxattr,redirect_dir=%s\n",
 731                                config->redirect_mode);
 732                         return -EINVAL;
 733                 }
 734                 if (config->metacopy && metacopy_opt) {
 735                         pr_err("conflicting options: userxattr,metacopy=on\n");
 736                         return -EINVAL;
 737                 }
 738                 /*
 739                  * Silently disable default setting of redirect and metacopy.
 740                  * This shall be the default in the future as well: these
 741                  * options must be explicitly enabled if used together with
 742                  * userxattr.
 743                  */
 744                 config->redirect_dir = config->redirect_follow = false;
 745                 config->metacopy = false;
 746         }
 747
 748         return 0;
 749 }
 750
 751 #define OVL_WORKDIR_NAME "work"
 752 #define OVL_INDEXDIR_NAME "index"
 753
 754 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 755                                          const char *name, bool persist)
 756 {
 757         struct inode *dir =  ofs->workbasedir->d_inode;
 758         struct vfsmount *mnt = ovl_upper_mnt(ofs);
 759         struct dentry *work;
 760         int err;
 761         bool retried = false;
 762
 763         inode_lock_nested(dir, I_MUTEX_PARENT);
 764 retry:
 765         work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));
 766
 767         if (!IS_ERR(work)) {
 768                 struct iattr attr = {
 769                         .ia_valid = ATTR_MODE,
 770                         .ia_mode = S_IFDIR | 0,
 771                 };
 772
 773                 if (work->d_inode) {
 774                         err = -EEXIST;
 775                         if (retried)
 776                                 goto out_dput;
 777
 778                         if (persist)
 779                                 goto out_unlock;
 780
 781                         retried = true;
 782                         err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);
 783                         dput(work);
 784                         if (err == -EINVAL) {
 785                                 work = ERR_PTR(err);
 786                                 goto out_unlock;
 787                         }
 788                         goto retry;
 789                 }
 790
 791                 err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);
 792                 if (err)
 793                         goto out_dput;
 794
 795                 /* Weird filesystem returning with hashed negative (kernfs)? */
 796                 err = -EINVAL;
 797                 if (d_really_is_negative(work))
 798                         goto out_dput;
 799
 800                 /*
 801                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 802                  *
 803                  * a) success (there was a POSIX ACL xattr and was removed)
 804                  * b) -ENODATA (there was no POSIX ACL xattr)
 805                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 806                  *
 807                  * There are various other error values that could effectively
 808                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 809                  * if the xattr name is too long), but the set of filesystems
 810                  * allowed as upper are limited to "normal" ones, where checking
 811                  * for the above two errors is sufficient.
 812                  */
 813                 err = ovl_do_removexattr(ofs, work,
 814                                          XATTR_NAME_POSIX_ACL_DEFAULT);
 815                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 816                         goto out_dput;
 817
 818                 err = ovl_do_removexattr(ofs, work,
 819                                          XATTR_NAME_POSIX_ACL_ACCESS);
 820                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 821                         goto out_dput;
 822
 823                 /* Clear any inherited mode bits */
 824                 inode_lock(work->d_inode);
 825                 err = ovl_do_notify_change(ofs, work, &attr);
 826                 inode_unlock(work->d_inode);
 827                 if (err)
 828                         goto out_dput;
 829         } else {
 830                 err = PTR_ERR(work);
 831                 goto out_err;
 832         }
 833 out_unlock:
 834         inode_unlock(dir);
 835         return work;
 836
 837 out_dput:
 838         dput(work);
 839 out_err:
 840         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
 841                 ofs->config.workdir, name, -err);
 842         work = NULL;
 843         goto out_unlock;
 844 }
 845
 846 static void ovl_unescape(char *s)
 847 {
 848         char *d = s;
 849
 850         for (;; s++, d++) {
 851                 if (*s == '\\')
 852                         s++;
 853                 *d = *s;
 854                 if (!*s)
 855                         break;
 856         }
 857 }
 858
 859 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 860 {
 861         int err = -EINVAL;
 862
 863         if (!*name) {
 864                 pr_err("empty lowerdir\n");
 865                 goto out;
 866         }
 867         err = kern_path(name, LOOKUP_FOLLOW, path);
 868         if (err) {
 869                 pr_err("failed to resolve '%s': %i\n", name, err);
 870                 goto out;
 871         }
 872         err = -EINVAL;
 873         if (ovl_dentry_weird(path->dentry)) {
 874                 pr_err("filesystem on '%s' not supported\n", name);
 875                 goto out_put;
 876         }
 877         if (is_idmapped_mnt(path->mnt)) {
 878                 pr_err("idmapped layers are currently not supported\n");
 879                 goto out_put;
 880         }
 881         if (!d_is_dir(path->dentry)) {
 882                 pr_err("'%s' not a directory\n", name);
 883                 goto out_put;
 884         }
 885         return 0;
 886
 887 out_put:
 888         path_put_init(path);
 889 out:
 890         return err;
 891 }
 892
 893 static int ovl_mount_dir(const char *name, struct path *path)
 894 {
 895         int err = -ENOMEM;
 896         char *tmp = kstrdup(name, GFP_KERNEL);
 897
 898         if (tmp) {
 899                 ovl_unescape(tmp);
 900                 err = ovl_mount_dir_noesc(tmp, path);
 901
 902                 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
 903                         pr_err("filesystem on '%s' not supported as upperdir\n",
 904                                tmp);
 905                         path_put_init(path);
 906                         err = -EINVAL;
 907                 }
 908                 kfree(tmp);
 909         }
 910         return err;
 911 }
 912
 913 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 914                              const char *name)
 915 {
 916         struct kstatfs statfs;
 917         int err = vfs_statfs(path, &statfs);
 918
 919         if (err)
 920                 pr_err("statfs failed on '%s'\n", name);
 921         else
 922                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 923
 924         return err;
 925 }
 926
 927 static int ovl_lower_dir(const char *name, struct path *path,
 928                          struct ovl_fs *ofs, int *stack_depth)
 929 {
 930         int fh_type;
 931         int err;
 932
 933         err = ovl_mount_dir_noesc(name, path);
 934         if (err)
 935                 return err;
 936
 937         err = ovl_check_namelen(path, ofs, name);
 938         if (err)
 939                 return err;
 940
 941         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 942
 943         /*
 944          * The inodes index feature and NFS export need to encode and decode
 945          * file handles, so they require that all layers support them.
 946          */
 947         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 948         if ((ofs->config.nfs_export ||
 949              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 950                 ofs->config.index = false;
 951                 ofs->config.nfs_export = false;
 952                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 953                         name);
 954         }
 955         /*
 956          * Decoding origin file handle is required for persistent st_ino.
 957          * Without persistent st_ino, xino=auto falls back to xino=off.
 958          */
 959         if (ofs->config.xino == OVL_XINO_AUTO &&
 960             ofs->config.upperdir && !fh_type) {
 961                 ofs->config.xino = OVL_XINO_OFF;
 962                 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
 963                         name);
 964         }
 965
 966         /* Check if lower fs has 32bit inode numbers */
 967         if (fh_type != FILEID_INO32_GEN)
 968                 ofs->xino_mode = -1;
 969
 970         return 0;
 971 }
 972
 973 /* Workdir should not be subdir of upperdir and vice versa */
 974 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 975 {
 976         bool ok = false;
 977
 978         if (workdir != upperdir) {
 979                 ok = (lock_rename(workdir, upperdir) == NULL);
 980                 unlock_rename(workdir, upperdir);
 981         }
 982         return ok;
 983 }
 984
 985 static unsigned int ovl_split_lowerdirs(char *str)
 986 {
 987         unsigned int ctr = 1;
 988         char *s, *d;
 989
 990         for (s = d = str;; s++, d++) {
 991                 if (*s == '\\') {
 992                         s++;
 993                 } else if (*s == ':') {
 994                         *d = '\0';
 995                         ctr++;
 996                         continue;
 997                 }
 998                 *d = *s;
 999                 if (!*s)
1000                         break;
1001         }
1002         return ctr;
1003 }
1004
1005 static int __maybe_unused
1006 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
1007                         struct dentry *dentry, struct inode *inode,
1008                         const char *name, void *buffer, size_t size)
1009 {
1010         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
1011 }
1012
1013 static int __maybe_unused
1014 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
1015                         struct user_namespace *mnt_userns,
1016                         struct dentry *dentry, struct inode *inode,
1017                         const char *name, const void *value,
1018                         size_t size, int flags)
1019 {
1020         struct dentry *workdir = ovl_workdir(dentry);
1021         struct inode *realinode = ovl_inode_real(inode);
1022         struct posix_acl *acl = NULL;
1023         int err;
1024
1025         /* Check that everything is OK before copy-up */
1026         if (value) {
1027                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
1028                 if (IS_ERR(acl))
1029                         return PTR_ERR(acl);
1030         }
1031         err = -EOPNOTSUPP;
1032         if (!IS_POSIXACL(d_inode(workdir)))
1033                 goto out_acl_release;
1034         if (!realinode->i_op->set_acl)
1035                 goto out_acl_release;
1036         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1037                 err = acl ? -EACCES : 0;
1038                 goto out_acl_release;
1039         }
1040         err = -EPERM;
1041         if (!inode_owner_or_capable(&init_user_ns, inode))
1042                 goto out_acl_release;
1043
1044         posix_acl_release(acl);
1045
1046         /*
1047          * Check if sgid bit needs to be cleared (actual setacl operation will
1048          * be done with mounter's capabilities and so that won't do it for us).
1049          */
1050         if (unlikely(inode->i_mode & S_ISGID) &&
1051             handler->flags == ACL_TYPE_ACCESS &&
1052             !in_group_p(inode->i_gid) &&
1053             !capable_wrt_inode_uidgid(&init_user_ns, inode, CAP_FSETID)) {
1054                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
1055
1056                 err = ovl_setattr(&init_user_ns, dentry, &iattr);
1057                 if (err)
1058                         return err;
1059         }
1060
1061         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
1062         return err;
1063
1064 out_acl_release:
1065         posix_acl_release(acl);
1066         return err;
1067 }
1068
1069 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1070                              struct dentry *dentry, struct inode *inode,
1071                              const char *name, void *buffer, size_t size)
1072 {
1073         return -EOPNOTSUPP;
1074 }
1075
1076 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1077                              struct user_namespace *mnt_userns,
1078                              struct dentry *dentry, struct inode *inode,
1079                              const char *name, const void *value,
1080                              size_t size, int flags)
1081 {
1082         return -EOPNOTSUPP;
1083 }
1084
1085 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1086                                struct dentry *dentry, struct inode *inode,
1087                                const char *name, void *buffer, size_t size)
1088 {
1089         return ovl_xattr_get(dentry, inode, name, buffer, size);
1090 }
1091
1092 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1093                                struct user_namespace *mnt_userns,
1094                                struct dentry *dentry, struct inode *inode,
1095                                const char *name, const void *value,
1096                                size_t size, int flags)
1097 {
1098         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1099 }
1100
1101 static const struct xattr_handler __maybe_unused
1102 ovl_posix_acl_access_xattr_handler = {
1103         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1104         .flags = ACL_TYPE_ACCESS,
1105         .get = ovl_posix_acl_xattr_get,
1106         .set = ovl_posix_acl_xattr_set,
1107 };
1108
1109 static const struct xattr_handler __maybe_unused
1110 ovl_posix_acl_default_xattr_handler = {
1111         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1112         .flags = ACL_TYPE_DEFAULT,
1113         .get = ovl_posix_acl_xattr_get,
1114         .set = ovl_posix_acl_xattr_set,
1115 };
1116
1117 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1118         .prefix = OVL_XATTR_TRUSTED_PREFIX,
1119         .get = ovl_own_xattr_get,
1120         .set = ovl_own_xattr_set,
1121 };
1122
1123 static const struct xattr_handler ovl_own_user_xattr_handler = {
1124         .prefix = OVL_XATTR_USER_PREFIX,
1125         .get = ovl_own_xattr_get,
1126         .set = ovl_own_xattr_set,
1127 };
1128
1129 static const struct xattr_handler ovl_other_xattr_handler = {
1130         .prefix = "", /* catch all */
1131         .get = ovl_other_xattr_get,
1132         .set = ovl_other_xattr_set,
1133 };
1134
1135 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1136 #ifdef CONFIG_FS_POSIX_ACL
1137         &ovl_posix_acl_access_xattr_handler,
1138         &ovl_posix_acl_default_xattr_handler,
1139 #endif
1140         &ovl_own_trusted_xattr_handler,
1141         &ovl_other_xattr_handler,
1142         NULL
1143 };
1144
1145 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1146 #ifdef CONFIG_FS_POSIX_ACL
1147         &ovl_posix_acl_access_xattr_handler,
1148         &ovl_posix_acl_default_xattr_handler,
1149 #endif
1150         &ovl_own_user_xattr_handler,
1151         &ovl_other_xattr_handler,
1152         NULL
1153 };
1154
1155 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1156                           struct inode **ptrap, const char *name)
1157 {
1158         struct inode *trap;
1159         int err;
1160
1161         trap = ovl_get_trap_inode(sb, dir);
1162         err = PTR_ERR_OR_ZERO(trap);
1163         if (err) {
1164                 if (err == -ELOOP)
1165                         pr_err("conflicting %s path\n", name);
1166                 return err;
1167         }
1168
1169         *ptrap = trap;
1170         return 0;
1171 }
1172
1173 /*
1174  * Determine how we treat concurrent use of upperdir/workdir based on the
1175  * index feature. This is papering over mount leaks of container runtimes,
1176  * for example, an old overlay mount is leaked and now its upperdir is
1177  * attempted to be used as a lower layer in a new overlay mount.
1178  */
1179 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1180 {
1181         if (ofs->config.index) {
1182                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1183                        name);
1184                 return -EBUSY;
1185         } else {
1186                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1187                         name);
1188                 return 0;
1189         }
1190 }
1191
1192 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1193                          struct ovl_layer *upper_layer, struct path *upperpath)
1194 {
1195         struct vfsmount *upper_mnt;
1196         int err;
1197
1198         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1199         if (err)
1200                 goto out;
1201
1202         /* Upperdir path should not be r/o */
1203         if (__mnt_is_readonly(upperpath->mnt)) {
1204                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1205                 err = -EINVAL;
1206                 goto out;
1207         }
1208
1209         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1210         if (err)
1211                 goto out;
1212
1213         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1214                              "upperdir");
1215         if (err)
1216                 goto out;
1217
1218         upper_mnt = clone_private_mount(upperpath);
1219         err = PTR_ERR(upper_mnt);
1220         if (IS_ERR(upper_mnt)) {
1221                 pr_err("failed to clone upperpath\n");
1222                 goto out;
1223         }
1224
1225         /* Don't inherit atime flags */
1226         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1227         upper_layer->mnt = upper_mnt;
1228         upper_layer->idx = 0;
1229         upper_layer->fsid = 0;
1230
1231         /*
1232          * Inherit SB_NOSEC flag from upperdir.
1233          *
1234          * This optimization changes behavior when a security related attribute
1235          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1236          * okay because we don't yet have guarantees in that case, but it will
1237          * need careful treatment once we want to honour changes to underlying
1238          * filesystems.
1239          */
1240         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1241                 sb->s_flags |= SB_NOSEC;
1242
1243         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1244                 ofs->upperdir_locked = true;
1245         } else {
1246                 err = ovl_report_in_use(ofs, "upperdir");
1247                 if (err)
1248                         goto out;
1249         }
1250
1251         err = 0;
1252 out:
1253         return err;
1254 }
1255
1256 /*
1257  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1258  * negative values if error is encountered.
1259  */
1260 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
1261 {
1262         struct dentry *workdir = ofs->workdir;
1263         struct inode *dir = d_inode(workdir);
1264         struct dentry *temp;
1265         struct dentry *dest;
1266         struct dentry *whiteout;
1267         struct name_snapshot name;
1268         int err;
1269
1270         inode_lock_nested(dir, I_MUTEX_PARENT);
1271
1272         temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
1273         err = PTR_ERR(temp);
1274         if (IS_ERR(temp))
1275                 goto out_unlock;
1276
1277         dest = ovl_lookup_temp(ofs, workdir);
1278         err = PTR_ERR(dest);
1279         if (IS_ERR(dest)) {
1280                 dput(temp);
1281                 goto out_unlock;
1282         }
1283
1284         /* Name is inline and stable - using snapshot as a copy helper */
1285         take_dentry_name_snapshot(&name, temp);
1286         err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
1287         if (err) {
1288                 if (err == -EINVAL)
1289                         err = 0;
1290                 goto cleanup_temp;
1291         }
1292
1293         whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
1294         err = PTR_ERR(whiteout);
1295         if (IS_ERR(whiteout))
1296                 goto cleanup_temp;
1297
1298         err = ovl_is_whiteout(whiteout);
1299
1300         /* Best effort cleanup of whiteout and temp file */
1301         if (err)
1302                 ovl_cleanup(ofs, dir, whiteout);
1303         dput(whiteout);
1304
1305 cleanup_temp:
1306         ovl_cleanup(ofs, dir, temp);
1307         release_dentry_name_snapshot(&name);
1308         dput(temp);
1309         dput(dest);
1310
1311 out_unlock:
1312         inode_unlock(dir);
1313
1314         return err;
1315 }
1316
1317 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
1318                                            struct dentry *parent,
1319                                            const char *name, umode_t mode)
1320 {
1321         size_t len = strlen(name);
1322         struct dentry *child;
1323
1324         inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1325         child = ovl_lookup_upper(ofs, name, parent, len);
1326         if (!IS_ERR(child) && !child->d_inode)
1327                 child = ovl_create_real(ofs, parent->d_inode, child,
1328                                         OVL_CATTR(mode));
1329         inode_unlock(parent->d_inode);
1330         dput(parent);
1331
1332         return child;
1333 }
1334
1335 /*
1336  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1337  * present.
1338  */
1339 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1340 {
1341         unsigned int ctr;
1342         struct dentry *d = dget(ofs->workbasedir);
1343         static const char *const volatile_path[] = {
1344                 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1345         };
1346         const char *const *name = volatile_path;
1347
1348         for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1349                 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1350                 if (IS_ERR(d))
1351                         return PTR_ERR(d);
1352         }
1353         dput(d);
1354         return 0;
1355 }
1356
1357 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1358                             struct path *workpath)
1359 {
1360         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1361         struct dentry *temp, *workdir;
1362         bool rename_whiteout;
1363         bool d_type;
1364         int fh_type;
1365         int err;
1366
1367         err = mnt_want_write(mnt);
1368         if (err)
1369                 return err;
1370
1371         workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1372         err = PTR_ERR(workdir);
1373         if (IS_ERR_OR_NULL(workdir))
1374                 goto out;
1375
1376         ofs->workdir = workdir;
1377
1378         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1379         if (err)
1380                 goto out;
1381
1382         /*
1383          * Upper should support d_type, else whiteouts are visible.  Given
1384          * workdir and upper are on same fs, we can do iterate_dir() on
1385          * workdir. This check requires successful creation of workdir in
1386          * previous step.
1387          */
1388         err = ovl_check_d_type_supported(workpath);
1389         if (err < 0)
1390                 goto out;
1391
1392         d_type = err;
1393         if (!d_type)
1394                 pr_warn("upper fs needs to support d_type.\n");
1395
1396         /* Check if upper/work fs supports O_TMPFILE */
1397         temp = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
1398         ofs->tmpfile = !IS_ERR(temp);
1399         if (ofs->tmpfile)
1400                 dput(temp);
1401         else
1402                 pr_warn("upper fs does not support tmpfile.\n");
1403
1404
1405         /* Check if upper/work fs supports RENAME_WHITEOUT */
1406         err = ovl_check_rename_whiteout(ofs);
1407         if (err < 0)
1408                 goto out;
1409
1410         rename_whiteout = err;
1411         if (!rename_whiteout)
1412                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1413
1414         /*
1415          * Check if upper/work fs supports (trusted|user).overlay.* xattr
1416          */
1417         err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1418         if (err) {
1419                 ofs->noxattr = true;
1420                 if (ofs->config.index || ofs->config.metacopy) {
1421                         ofs->config.index = false;
1422                         ofs->config.metacopy = false;
1423                         pr_warn("upper fs does not support xattr, falling back to index=off,metacopy=off.\n");
1424                 }
1425                 /*
1426                  * xattr support is required for persistent st_ino.
1427                  * Without persistent st_ino, xino=auto falls back to xino=off.
1428                  */
1429                 if (ofs->config.xino == OVL_XINO_AUTO) {
1430                         ofs->config.xino = OVL_XINO_OFF;
1431                         pr_warn("upper fs does not support xattr, falling back to xino=off.\n");
1432                 }
1433                 err = 0;
1434         } else {
1435                 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1436         }
1437
1438         /*
1439          * We allowed sub-optimal upper fs configuration and don't want to break
1440          * users over kernel upgrade, but we never allowed remote upper fs, so
1441          * we can enforce strict requirements for remote upper fs.
1442          */
1443         if (ovl_dentry_remote(ofs->workdir) &&
1444             (!d_type || !rename_whiteout || ofs->noxattr)) {
1445                 pr_err("upper fs missing required features.\n");
1446                 err = -EINVAL;
1447                 goto out;
1448         }
1449
1450         /*
1451          * For volatile mount, create a incompat/volatile/dirty file to keep
1452          * track of it.
1453          */
1454         if (ofs->config.ovl_volatile) {
1455                 err = ovl_create_volatile_dirty(ofs);
1456                 if (err < 0) {
1457                         pr_err("Failed to create volatile/dirty file.\n");
1458                         goto out;
1459                 }
1460         }
1461
1462         /* Check if upper/work fs supports file handles */
1463         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1464         if (ofs->config.index && !fh_type) {
1465                 ofs->config.index = false;
1466                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1467         }
1468
1469         /* Check if upper fs has 32bit inode numbers */
1470         if (fh_type != FILEID_INO32_GEN)
1471                 ofs->xino_mode = -1;
1472
1473         /* NFS export of r/w mount depends on index */
1474         if (ofs->config.nfs_export && !ofs->config.index) {
1475                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1476                 ofs->config.nfs_export = false;
1477         }
1478 out:
1479         mnt_drop_write(mnt);
1480         return err;
1481 }
1482
1483 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1484                            struct path *upperpath)
1485 {
1486         int err;
1487         struct path workpath = { };
1488
1489         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1490         if (err)
1491                 goto out;
1492
1493         err = -EINVAL;
1494         if (upperpath->mnt != workpath.mnt) {
1495                 pr_err("workdir and upperdir must reside under the same mount\n");
1496                 goto out;
1497         }
1498         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1499                 pr_err("workdir and upperdir must be separate subtrees\n");
1500                 goto out;
1501         }
1502
1503         ofs->workbasedir = dget(workpath.dentry);
1504
1505         if (ovl_inuse_trylock(ofs->workbasedir)) {
1506                 ofs->workdir_locked = true;
1507         } else {
1508                 err = ovl_report_in_use(ofs, "workdir");
1509                 if (err)
1510                         goto out;
1511         }
1512
1513         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1514                              "workdir");
1515         if (err)
1516                 goto out;
1517
1518         err = ovl_make_workdir(sb, ofs, &workpath);
1519
1520 out:
1521         path_put(&workpath);
1522
1523         return err;
1524 }
1525
1526 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1527                             struct ovl_entry *oe, struct path *upperpath)
1528 {
1529         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1530         struct dentry *indexdir;
1531         int err;
1532
1533         err = mnt_want_write(mnt);
1534         if (err)
1535                 return err;
1536
1537         /* Verify lower root is upper root origin */
1538         err = ovl_verify_origin(ofs, upperpath->dentry,
1539                                 oe->lowerstack[0].dentry, true);
1540         if (err) {
1541                 pr_err("failed to verify upper root origin\n");
1542                 goto out;
1543         }
1544
1545         /* index dir will act also as workdir */
1546         iput(ofs->workdir_trap);
1547         ofs->workdir_trap = NULL;
1548         dput(ofs->workdir);
1549         ofs->workdir = NULL;
1550         indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1551         if (IS_ERR(indexdir)) {
1552                 err = PTR_ERR(indexdir);
1553         } else if (indexdir) {
1554                 ofs->indexdir = indexdir;
1555                 ofs->workdir = dget(indexdir);
1556
1557                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1558                                      "indexdir");
1559                 if (err)
1560                         goto out;
1561
1562                 /*
1563                  * Verify upper root is exclusively associated with index dir.
1564                  * Older kernels stored upper fh in ".overlay.origin"
1565                  * xattr. If that xattr exists, verify that it is a match to
1566                  * upper dir file handle. In any case, verify or set xattr
1567                  * ".overlay.upper" to indicate that index may have
1568                  * directory entries.
1569                  */
1570                 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1571                         err = ovl_verify_set_fh(ofs, ofs->indexdir,
1572                                                 OVL_XATTR_ORIGIN,
1573                                                 upperpath->dentry, true, false);
1574                         if (err)
1575                                 pr_err("failed to verify index dir 'origin' xattr\n");
1576                 }
1577                 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1578                                        true);
1579                 if (err)
1580                         pr_err("failed to verify index dir 'upper' xattr\n");
1581
1582                 /* Cleanup bad/stale/orphan index entries */
1583                 if (!err)
1584                         err = ovl_indexdir_cleanup(ofs);
1585         }
1586         if (err || !ofs->indexdir)
1587                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1588
1589 out:
1590         mnt_drop_write(mnt);
1591         return err;
1592 }
1593
1594 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1595 {
1596         unsigned int i;
1597
1598         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1599                 return true;
1600
1601         /*
1602          * We allow using single lower with null uuid for index and nfs_export
1603          * for example to support those features with single lower squashfs.
1604          * To avoid regressions in setups of overlay with re-formatted lower
1605          * squashfs, do not allow decoding origin with lower null uuid unless
1606          * user opted-in to one of the new features that require following the
1607          * lower inode of non-dir upper.
1608          */
1609         if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
1610                 return false;
1611
1612         for (i = 0; i < ofs->numfs; i++) {
1613                 /*
1614                  * We use uuid to associate an overlay lower file handle with a
1615                  * lower layer, so we can accept lower fs with null uuid as long
1616                  * as all lower layers with null uuid are on the same fs.
1617                  * if we detect multiple lower fs with the same uuid, we
1618                  * disable lower file handle decoding on all of them.
1619                  */
1620                 if (ofs->fs[i].is_lower &&
1621                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1622                         ofs->fs[i].bad_uuid = true;
1623                         return false;
1624                 }
1625         }
1626         return true;
1627 }
1628
1629 /* Get a unique fsid for the layer */
1630 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1631 {
1632         struct super_block *sb = path->mnt->mnt_sb;
1633         unsigned int i;
1634         dev_t dev;
1635         int err;
1636         bool bad_uuid = false;
1637         bool warn = false;
1638
1639         for (i = 0; i < ofs->numfs; i++) {
1640                 if (ofs->fs[i].sb == sb)
1641                         return i;
1642         }
1643
1644         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1645                 bad_uuid = true;
1646                 if (ofs->config.xino == OVL_XINO_AUTO) {
1647                         ofs->config.xino = OVL_XINO_OFF;
1648                         warn = true;
1649                 }
1650                 if (ofs->config.index || ofs->config.nfs_export) {
1651                         ofs->config.index = false;
1652                         ofs->config.nfs_export = false;
1653                         warn = true;
1654                 }
1655                 if (warn) {
1656                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1657                                 uuid_is_null(&sb->s_uuid) ? "null" :
1658                                                             "conflicting",
1659                                 path->dentry, ovl_xino_str[ofs->config.xino]);
1660                 }
1661         }
1662
1663         err = get_anon_bdev(&dev);
1664         if (err) {
1665                 pr_err("failed to get anonymous bdev for lowerpath\n");
1666                 return err;
1667         }
1668
1669         ofs->fs[ofs->numfs].sb = sb;
1670         ofs->fs[ofs->numfs].pseudo_dev = dev;
1671         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1672
1673         return ofs->numfs++;
1674 }
1675
1676 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1677                           struct path *stack, unsigned int numlower,
1678                           struct ovl_layer *layers)
1679 {
1680         int err;
1681         unsigned int i;
1682
1683         err = -ENOMEM;
1684         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1685         if (ofs->fs == NULL)
1686                 goto out;
1687
1688         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1689         ofs->numfs++;
1690
1691         /*
1692          * All lower layers that share the same fs as upper layer, use the same
1693          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1694          * only overlay to simplify ovl_fs_free().
1695          * is_lower will be set if upper fs is shared with a lower layer.
1696          */
1697         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1698         if (err) {
1699                 pr_err("failed to get anonymous bdev for upper fs\n");
1700                 goto out;
1701         }
1702
1703         if (ovl_upper_mnt(ofs)) {
1704                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1705                 ofs->fs[0].is_lower = false;
1706         }
1707
1708         for (i = 0; i < numlower; i++) {
1709                 struct vfsmount *mnt;
1710                 struct inode *trap;
1711                 int fsid;
1712
1713                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1714                 if (err < 0)
1715                         goto out;
1716
1717                 /*
1718                  * Check if lower root conflicts with this overlay layers before
1719                  * checking if it is in-use as upperdir/workdir of "another"
1720                  * mount, because we do not bother to check in ovl_is_inuse() if
1721                  * the upperdir/workdir is in fact in-use by our
1722                  * upperdir/workdir.
1723                  */
1724                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1725                 if (err)
1726                         goto out;
1727
1728                 if (ovl_is_inuse(stack[i].dentry)) {
1729                         err = ovl_report_in_use(ofs, "lowerdir");
1730                         if (err) {
1731                                 iput(trap);
1732                                 goto out;
1733                         }
1734                 }
1735
1736                 mnt = clone_private_mount(&stack[i]);
1737                 err = PTR_ERR(mnt);
1738                 if (IS_ERR(mnt)) {
1739                         pr_err("failed to clone lowerpath\n");
1740                         iput(trap);
1741                         goto out;
1742                 }
1743
1744                 /*
1745                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1746                  * will fail instead of modifying lower fs.
1747                  */
1748                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1749
1750                 layers[ofs->numlayer].trap = trap;
1751                 layers[ofs->numlayer].mnt = mnt;
1752                 layers[ofs->numlayer].idx = ofs->numlayer;
1753                 layers[ofs->numlayer].fsid = fsid;
1754                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1755                 ofs->numlayer++;
1756                 ofs->fs[fsid].is_lower = true;
1757         }
1758
1759         /*
1760          * When all layers on same fs, overlay can use real inode numbers.
1761          * With mount option "xino=<on|auto>", mounter declares that there are
1762          * enough free high bits in underlying fs to hold the unique fsid.
1763          * If overlayfs does encounter underlying inodes using the high xino
1764          * bits reserved for fsid, it emits a warning and uses the original
1765          * inode number or a non persistent inode number allocated from a
1766          * dedicated range.
1767          */
1768         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1769                 if (ofs->config.xino == OVL_XINO_ON)
1770                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1771                 ofs->xino_mode = 0;
1772         } else if (ofs->config.xino == OVL_XINO_OFF) {
1773                 ofs->xino_mode = -1;
1774         } else if (ofs->xino_mode < 0) {
1775                 /*
1776                  * This is a roundup of number of bits needed for encoding
1777                  * fsid, where fsid 0 is reserved for upper fs (even with
1778                  * lower only overlay) +1 extra bit is reserved for the non
1779                  * persistent inode number range that is used for resolving
1780                  * xino lower bits overflow.
1781                  */
1782                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1783                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1784         }
1785
1786         if (ofs->xino_mode > 0) {
1787                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1788                         ofs->xino_mode);
1789         }
1790
1791         err = 0;
1792 out:
1793         return err;
1794 }
1795
1796 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1797                                 const char *lower, unsigned int numlower,
1798                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1799 {
1800         int err;
1801         struct path *stack = NULL;
1802         unsigned int i;
1803         struct ovl_entry *oe;
1804
1805         if (!ofs->config.upperdir && numlower == 1) {
1806                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1807                 return ERR_PTR(-EINVAL);
1808         }
1809
1810         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1811         if (!stack)
1812                 return ERR_PTR(-ENOMEM);
1813
1814         err = -EINVAL;
1815         for (i = 0; i < numlower; i++) {
1816                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1817                 if (err)
1818                         goto out_err;
1819
1820                 lower = strchr(lower, '\0') + 1;
1821         }
1822
1823         err = -EINVAL;
1824         sb->s_stack_depth++;
1825         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1826                 pr_err("maximum fs stacking depth exceeded\n");
1827                 goto out_err;
1828         }
1829
1830         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1831         if (err)
1832                 goto out_err;
1833
1834         err = -ENOMEM;
1835         oe = ovl_alloc_entry(numlower);
1836         if (!oe)
1837                 goto out_err;
1838
1839         for (i = 0; i < numlower; i++) {
1840                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1841                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1842         }
1843
1844 out:
1845         for (i = 0; i < numlower; i++)
1846                 path_put(&stack[i]);
1847         kfree(stack);
1848
1849         return oe;
1850
1851 out_err:
1852         oe = ERR_PTR(err);
1853         goto out;
1854 }
1855
1856 /*
1857  * Check if this layer root is a descendant of:
1858  * - another layer of this overlayfs instance
1859  * - upper/work dir of any overlayfs instance
1860  */
1861 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1862                            struct dentry *dentry, const char *name,
1863                            bool is_lower)
1864 {
1865         struct dentry *next = dentry, *parent;
1866         int err = 0;
1867
1868         if (!dentry)
1869                 return 0;
1870
1871         parent = dget_parent(next);
1872
1873         /* Walk back ancestors to root (inclusive) looking for traps */
1874         while (!err && parent != next) {
1875                 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1876                         err = -ELOOP;
1877                         pr_err("overlapping %s path\n", name);
1878                 } else if (ovl_is_inuse(parent)) {
1879                         err = ovl_report_in_use(ofs, name);
1880                 }
1881                 next = parent;
1882                 parent = dget_parent(next);
1883                 dput(next);
1884         }
1885
1886         dput(parent);
1887
1888         return err;
1889 }
1890
1891 /*
1892  * Check if any of the layers or work dirs overlap.
1893  */
1894 static int ovl_check_overlapping_layers(struct super_block *sb,
1895                                         struct ovl_fs *ofs)
1896 {
1897         int i, err;
1898
1899         if (ovl_upper_mnt(ofs)) {
1900                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1901                                       "upperdir", false);
1902                 if (err)
1903                         return err;
1904
1905                 /*
1906                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1907                  * this instance and covers overlapping work and index dirs,
1908                  * unless work or index dir have been moved since created inside
1909                  * workbasedir.  In that case, we already have their traps in
1910                  * inode cache and we will catch that case on lookup.
1911                  */
1912                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1913                                       false);
1914                 if (err)
1915                         return err;
1916         }
1917
1918         for (i = 1; i < ofs->numlayer; i++) {
1919                 err = ovl_check_layer(sb, ofs,
1920                                       ofs->layers[i].mnt->mnt_root,
1921                                       "lowerdir", true);
1922                 if (err)
1923                         return err;
1924         }
1925
1926         return 0;
1927 }
1928
1929 static struct dentry *ovl_get_root(struct super_block *sb,
1930                                    struct dentry *upperdentry,
1931                                    struct ovl_entry *oe)
1932 {
1933         struct dentry *root;
1934         struct ovl_path *lowerpath = &oe->lowerstack[0];
1935         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1936         int fsid = lowerpath->layer->fsid;
1937         struct ovl_inode_params oip = {
1938                 .upperdentry = upperdentry,
1939                 .lowerpath = lowerpath,
1940         };
1941
1942         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1943         if (!root)
1944                 return NULL;
1945
1946         root->d_fsdata = oe;
1947
1948         if (upperdentry) {
1949                 /* Root inode uses upper st_ino/i_ino */
1950                 ino = d_inode(upperdentry)->i_ino;
1951                 fsid = 0;
1952                 ovl_dentry_set_upper_alias(root);
1953                 if (ovl_is_impuredir(sb, upperdentry))
1954                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1955         }
1956
1957         /* Root is always merge -> can have whiteouts */
1958         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1959         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1960         ovl_set_upperdata(d_inode(root));
1961         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1962         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1963
1964         return root;
1965 }
1966
1967 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1968 {
1969         struct path upperpath = { };
1970         struct dentry *root_dentry;
1971         struct ovl_entry *oe;
1972         struct ovl_fs *ofs;
1973         struct ovl_layer *layers;
1974         struct cred *cred;
1975         char *splitlower = NULL;
1976         unsigned int numlower;
1977         int err;
1978
1979         err = -EIO;
1980         if (WARN_ON(sb->s_user_ns != current_user_ns()))
1981                 goto out;
1982
1983         sb->s_d_op = &ovl_dentry_operations;
1984
1985         err = -ENOMEM;
1986         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1987         if (!ofs)
1988                 goto out;
1989
1990         err = -ENOMEM;
1991         ofs->creator_cred = cred = prepare_creds();
1992         if (!cred)
1993                 goto out_err;
1994
1995         /* Is there a reason anyone would want not to share whiteouts? */
1996         ofs->share_whiteout = true;
1997
1998         ofs->config.index = ovl_index_def;
1999         ofs->config.uuid = true;
2000         ofs->config.nfs_export = ovl_nfs_export_def;
2001         ofs->config.xino = ovl_xino_def();
2002         ofs->config.metacopy = ovl_metacopy_def;
2003         err = ovl_parse_opt((char *) data, &ofs->config);
2004         if (err)
2005                 goto out_err;
2006
2007         err = -EINVAL;
2008         if (!ofs->config.lowerdir) {
2009                 if (!silent)
2010                         pr_err("missing 'lowerdir'\n");
2011                 goto out_err;
2012         }
2013
2014         err = -ENOMEM;
2015         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
2016         if (!splitlower)
2017                 goto out_err;
2018
2019         err = -EINVAL;
2020         numlower = ovl_split_lowerdirs(splitlower);
2021         if (numlower > OVL_MAX_STACK) {
2022                 pr_err("too many lower directories, limit is %d\n",
2023                        OVL_MAX_STACK);
2024                 goto out_err;
2025         }
2026
2027         err = -ENOMEM;
2028         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
2029         if (!layers)
2030                 goto out_err;
2031
2032         ofs->layers = layers;
2033         /* Layer 0 is reserved for upper even if there's no upper */
2034         ofs->numlayer = 1;
2035
2036         sb->s_stack_depth = 0;
2037         sb->s_maxbytes = MAX_LFS_FILESIZE;
2038         atomic_long_set(&ofs->last_ino, 1);
2039         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
2040         if (ofs->config.xino != OVL_XINO_OFF) {
2041                 ofs->xino_mode = BITS_PER_LONG - 32;
2042                 if (!ofs->xino_mode) {
2043                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
2044                         ofs->config.xino = OVL_XINO_OFF;
2045                 }
2046         }
2047
2048         /* alloc/destroy_inode needed for setting up traps in inode cache */
2049         sb->s_op = &ovl_super_operations;
2050
2051         if (ofs->config.upperdir) {
2052                 struct super_block *upper_sb;
2053
2054                 err = -EINVAL;
2055                 if (!ofs->config.workdir) {
2056                         pr_err("missing 'workdir'\n");
2057                         goto out_err;
2058                 }
2059
2060                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
2061                 if (err)
2062                         goto out_err;
2063
2064                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
2065                 if (!ovl_should_sync(ofs)) {
2066                         ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
2067                         if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
2068                                 err = -EIO;
2069                                 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
2070                                 goto out_err;
2071                         }
2072                 }
2073
2074                 err = ovl_get_workdir(sb, ofs, &upperpath);
2075                 if (err)
2076                         goto out_err;
2077
2078                 if (!ofs->workdir)
2079                         sb->s_flags |= SB_RDONLY;
2080
2081                 sb->s_stack_depth = upper_sb->s_stack_depth;
2082                 sb->s_time_gran = upper_sb->s_time_gran;
2083         }
2084         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2085         err = PTR_ERR(oe);
2086         if (IS_ERR(oe))
2087                 goto out_err;
2088
2089         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2090         if (!ovl_upper_mnt(ofs))
2091                 sb->s_flags |= SB_RDONLY;
2092
2093         if (!ofs->config.uuid && ofs->numfs > 1) {
2094                 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2095                 ofs->config.uuid = true;
2096         }
2097
2098         if (!ovl_force_readonly(ofs) && ofs->config.index) {
2099                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2100                 if (err)
2101                         goto out_free_oe;
2102
2103                 /* Force r/o mount with no index dir */
2104                 if (!ofs->indexdir)
2105                         sb->s_flags |= SB_RDONLY;
2106         }
2107
2108         err = ovl_check_overlapping_layers(sb, ofs);
2109         if (err)
2110                 goto out_free_oe;
2111
2112         /* Show index=off in /proc/mounts for forced r/o mount */
2113         if (!ofs->indexdir) {
2114                 ofs->config.index = false;
2115                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2116                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2117                         ofs->config.nfs_export = false;
2118                 }
2119         }
2120
2121         if (ofs->config.metacopy && ofs->config.nfs_export) {
2122                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2123                 ofs->config.nfs_export = false;
2124         }
2125
2126         if (ofs->config.nfs_export)
2127                 sb->s_export_op = &ovl_export_operations;
2128
2129         /* Never override disk quota limits or use reserved space */
2130         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2131
2132         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2133         sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2134                 ovl_trusted_xattr_handlers;
2135         sb->s_fs_info = ofs;
2136         sb->s_flags |= SB_POSIXACL;
2137         sb->s_iflags |= SB_I_SKIP_SYNC;
2138
2139         err = -ENOMEM;
2140         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2141         if (!root_dentry)
2142                 goto out_free_oe;
2143
2144         mntput(upperpath.mnt);
2145         kfree(splitlower);
2146
2147         sb->s_root = root_dentry;
2148
2149         return 0;
2150
2151 out_free_oe:
2152         ovl_entry_stack_free(oe);
2153         kfree(oe);
2154 out_err:
2155         kfree(splitlower);
2156         path_put(&upperpath);
2157         ovl_free_fs(ofs);
2158 out:
2159         return err;
2160 }
2161
2162 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2163                                 const char *dev_name, void *raw_data)
2164 {
2165         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2166 }
2167
2168 static struct file_system_type ovl_fs_type = {
2169         .owner          = THIS_MODULE,
2170         .name           = "overlay",
2171         .fs_flags       = FS_USERNS_MOUNT,
2172         .mount          = ovl_mount,
2173         .kill_sb        = kill_anon_super,
2174 };
2175 MODULE_ALIAS_FS("overlay");
2176
2177 static void ovl_inode_init_once(void *foo)
2178 {
2179         struct ovl_inode *oi = foo;
2180
2181         inode_init_once(&oi->vfs_inode);
2182 }
2183
2184 static int __init ovl_init(void)
2185 {
2186         int err;
2187
2188         ovl_inode_cachep = kmem_cache_create("ovl_inode",
2189                                              sizeof(struct ovl_inode), 0,
2190                                              (SLAB_RECLAIM_ACCOUNT|
2191                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2192                                              ovl_inode_init_once);
2193         if (ovl_inode_cachep == NULL)
2194                 return -ENOMEM;
2195
2196         err = ovl_aio_request_cache_init();
2197         if (!err) {
2198                 err = register_filesystem(&ovl_fs_type);
2199                 if (!err)
2200                         return 0;
2201
2202                 ovl_aio_request_cache_destroy();
2203         }
2204         kmem_cache_destroy(ovl_inode_cachep);
2205
2206         return err;
2207 }
2208
2209 static void __exit ovl_exit(void)
2210 {
2211         unregister_filesystem(&ovl_fs_type);
2212
2213         /*
2214          * Make sure all delayed rcu free inodes are flushed before we
2215          * destroy cache.
2216          */
2217         rcu_barrier();
2218         kmem_cache_destroy(ovl_inode_cachep);
2219         ovl_aio_request_cache_destroy();
2220 }
2221
2222 module_init(ovl_init);
2223 module_exit(ovl_exit);