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  * @dentry: The dentry to query
 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  * @m: the seq_file handle
 353  * @dentry: The dentry to query
 354  *
 355  * Prints the mount options for a given superblock.
 356  * Returns zero; does not fail.
 357  */
 358 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 359 {
 360         struct super_block *sb = dentry->d_sb;
 361         struct ovl_fs *ofs = sb->s_fs_info;
 362
 363         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 364         if (ofs->config.upperdir) {
 365                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 366                 seq_show_option(m, "workdir", ofs->config.workdir);
 367         }
 368         if (ofs->config.default_permissions)
 369                 seq_puts(m, ",default_permissions");
 370         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 371                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 372         if (ofs->config.index != ovl_index_def)
 373                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 374         if (!ofs->config.uuid)
 375                 seq_puts(m, ",uuid=off");
 376         if (ofs->config.nfs_export != ovl_nfs_export_def)
 377                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 378                                                 "on" : "off");
 379         if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
 380                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 381         if (ofs->config.metacopy != ovl_metacopy_def)
 382                 seq_printf(m, ",metacopy=%s",
 383                            ofs->config.metacopy ? "on" : "off");
 384         if (ofs->config.ovl_volatile)
 385                 seq_puts(m, ",volatile");
 386         if (ofs->config.userxattr)
 387                 seq_puts(m, ",userxattr");
 388         return 0;
 389 }
 390
 391 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 392 {
 393         struct ovl_fs *ofs = sb->s_fs_info;
 394         struct super_block *upper_sb;
 395         int ret = 0;
 396
 397         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 398                 return -EROFS;
 399
 400         if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
 401                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 402                 if (ovl_should_sync(ofs)) {
 403                         down_read(&upper_sb->s_umount);
 404                         ret = sync_filesystem(upper_sb);
 405                         up_read(&upper_sb->s_umount);
 406                 }
 407         }
 408
 409         return ret;
 410 }
 411
 412 static const struct super_operations ovl_super_operations = {
 413         .alloc_inode    = ovl_alloc_inode,
 414         .free_inode     = ovl_free_inode,
 415         .destroy_inode  = ovl_destroy_inode,
 416         .drop_inode     = generic_delete_inode,
 417         .put_super      = ovl_put_super,
 418         .sync_fs        = ovl_sync_fs,
 419         .statfs         = ovl_statfs,
 420         .show_options   = ovl_show_options,
 421         .remount_fs     = ovl_remount,
 422 };
 423
 424 enum {
 425         OPT_LOWERDIR,
 426         OPT_UPPERDIR,
 427         OPT_WORKDIR,
 428         OPT_DEFAULT_PERMISSIONS,
 429         OPT_REDIRECT_DIR,
 430         OPT_INDEX_ON,
 431         OPT_INDEX_OFF,
 432         OPT_UUID_ON,
 433         OPT_UUID_OFF,
 434         OPT_NFS_EXPORT_ON,
 435         OPT_USERXATTR,
 436         OPT_NFS_EXPORT_OFF,
 437         OPT_XINO_ON,
 438         OPT_XINO_OFF,
 439         OPT_XINO_AUTO,
 440         OPT_METACOPY_ON,
 441         OPT_METACOPY_OFF,
 442         OPT_VOLATILE,
 443         OPT_ERR,
 444 };
 445
 446 static const match_table_t ovl_tokens = {
 447         {OPT_LOWERDIR,                  "lowerdir=%s"},
 448         {OPT_UPPERDIR,                  "upperdir=%s"},
 449         {OPT_WORKDIR,                   "workdir=%s"},
 450         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 451         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 452         {OPT_INDEX_ON,                  "index=on"},
 453         {OPT_INDEX_OFF,                 "index=off"},
 454         {OPT_USERXATTR,                 "userxattr"},
 455         {OPT_UUID_ON,                   "uuid=on"},
 456         {OPT_UUID_OFF,                  "uuid=off"},
 457         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 458         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 459         {OPT_XINO_ON,                   "xino=on"},
 460         {OPT_XINO_OFF,                  "xino=off"},
 461         {OPT_XINO_AUTO,                 "xino=auto"},
 462         {OPT_METACOPY_ON,               "metacopy=on"},
 463         {OPT_METACOPY_OFF,              "metacopy=off"},
 464         {OPT_VOLATILE,                  "volatile"},
 465         {OPT_ERR,                       NULL}
 466 };
 467
 468 static char *ovl_next_opt(char **s)
 469 {
 470         char *sbegin = *s;
 471         char *p;
 472
 473         if (sbegin == NULL)
 474                 return NULL;
 475
 476         for (p = sbegin; *p; p++) {
 477                 if (*p == '\\') {
 478                         p++;
 479                         if (!*p)
 480                                 break;
 481                 } else if (*p == ',') {
 482                         *p = '\0';
 483                         *s = p + 1;
 484                         return sbegin;
 485                 }
 486         }
 487         *s = NULL;
 488         return sbegin;
 489 }
 490
 491 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 492 {
 493         if (strcmp(mode, "on") == 0) {
 494                 config->redirect_dir = true;
 495                 /*
 496                  * Does not make sense to have redirect creation without
 497                  * redirect following.
 498                  */
 499                 config->redirect_follow = true;
 500         } else if (strcmp(mode, "follow") == 0) {
 501                 config->redirect_follow = true;
 502         } else if (strcmp(mode, "off") == 0) {
 503                 if (ovl_redirect_always_follow)
 504                         config->redirect_follow = true;
 505         } else if (strcmp(mode, "nofollow") != 0) {
 506                 pr_err("bad mount option \"redirect_dir=%s\"\n",
 507                        mode);
 508                 return -EINVAL;
 509         }
 510
 511         return 0;
 512 }
 513
 514 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 515 {
 516         char *p;
 517         int err;
 518         bool metacopy_opt = false, redirect_opt = false;
 519         bool nfs_export_opt = false, index_opt = false;
 520
 521         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 522         if (!config->redirect_mode)
 523                 return -ENOMEM;
 524
 525         while ((p = ovl_next_opt(&opt)) != NULL) {
 526                 int token;
 527                 substring_t args[MAX_OPT_ARGS];
 528
 529                 if (!*p)
 530                         continue;
 531
 532                 token = match_token(p, ovl_tokens, args);
 533                 switch (token) {
 534                 case OPT_UPPERDIR:
 535                         kfree(config->upperdir);
 536                         config->upperdir = match_strdup(&args[0]);
 537                         if (!config->upperdir)
 538                                 return -ENOMEM;
 539                         break;
 540
 541                 case OPT_LOWERDIR:
 542                         kfree(config->lowerdir);
 543                         config->lowerdir = match_strdup(&args[0]);
 544                         if (!config->lowerdir)
 545                                 return -ENOMEM;
 546                         break;
 547
 548                 case OPT_WORKDIR:
 549                         kfree(config->workdir);
 550                         config->workdir = match_strdup(&args[0]);
 551                         if (!config->workdir)
 552                                 return -ENOMEM;
 553                         break;
 554
 555                 case OPT_DEFAULT_PERMISSIONS:
 556                         config->default_permissions = true;
 557                         break;
 558
 559                 case OPT_REDIRECT_DIR:
 560                         kfree(config->redirect_mode);
 561                         config->redirect_mode = match_strdup(&args[0]);
 562                         if (!config->redirect_mode)
 563                                 return -ENOMEM;
 564                         redirect_opt = true;
 565                         break;
 566
 567                 case OPT_INDEX_ON:
 568                         config->index = true;
 569                         index_opt = true;
 570                         break;
 571
 572                 case OPT_INDEX_OFF:
 573                         config->index = false;
 574                         index_opt = true;
 575                         break;
 576
 577                 case OPT_UUID_ON:
 578                         config->uuid = true;
 579                         break;
 580
 581                 case OPT_UUID_OFF:
 582                         config->uuid = false;
 583                         break;
 584
 585                 case OPT_NFS_EXPORT_ON:
 586                         config->nfs_export = true;
 587                         nfs_export_opt = true;
 588                         break;
 589
 590                 case OPT_NFS_EXPORT_OFF:
 591                         config->nfs_export = false;
 592                         nfs_export_opt = true;
 593                         break;
 594
 595                 case OPT_XINO_ON:
 596                         config->xino = OVL_XINO_ON;
 597                         break;
 598
 599                 case OPT_XINO_OFF:
 600                         config->xino = OVL_XINO_OFF;
 601                         break;
 602
 603                 case OPT_XINO_AUTO:
 604                         config->xino = OVL_XINO_AUTO;
 605                         break;
 606
 607                 case OPT_METACOPY_ON:
 608                         config->metacopy = true;
 609                         metacopy_opt = true;
 610                         break;
 611
 612                 case OPT_METACOPY_OFF:
 613                         config->metacopy = false;
 614                         metacopy_opt = true;
 615                         break;
 616
 617                 case OPT_VOLATILE:
 618                         config->ovl_volatile = true;
 619                         break;
 620
 621                 case OPT_USERXATTR:
 622                         config->userxattr = true;
 623                         break;
 624
 625                 default:
 626                         pr_err("unrecognized mount option \"%s\" or missing value\n",
 627                                         p);
 628                         return -EINVAL;
 629                 }
 630         }
 631
 632         /* Workdir/index are useless in non-upper mount */
 633         if (!config->upperdir) {
 634                 if (config->workdir) {
 635                         pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 636                                 config->workdir);
 637                         kfree(config->workdir);
 638                         config->workdir = NULL;
 639                 }
 640                 if (config->index && index_opt) {
 641                         pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
 642                         index_opt = false;
 643                 }
 644                 config->index = false;
 645         }
 646
 647         if (!config->upperdir && config->ovl_volatile) {
 648                 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
 649                 config->ovl_volatile = false;
 650         }
 651
 652         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 653         if (err)
 654                 return err;
 655
 656         /*
 657          * This is to make the logic below simpler.  It doesn't make any other
 658          * difference, since config->redirect_dir is only used for upper.
 659          */
 660         if (!config->upperdir && config->redirect_follow)
 661                 config->redirect_dir = true;
 662
 663         /* Resolve metacopy -> redirect_dir dependency */
 664         if (config->metacopy && !config->redirect_dir) {
 665                 if (metacopy_opt && redirect_opt) {
 666                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
 667                                config->redirect_mode);
 668                         return -EINVAL;
 669                 }
 670                 if (redirect_opt) {
 671                         /*
 672                          * There was an explicit redirect_dir=... that resulted
 673                          * in this conflict.
 674                          */
 675                         pr_info("disabling metacopy due to redirect_dir=%s\n",
 676                                 config->redirect_mode);
 677                         config->metacopy = false;
 678                 } else {
 679                         /* Automatically enable redirect otherwise. */
 680                         config->redirect_follow = config->redirect_dir = true;
 681                 }
 682         }
 683
 684         /* Resolve nfs_export -> index dependency */
 685         if (config->nfs_export && !config->index) {
 686                 if (!config->upperdir && config->redirect_follow) {
 687                         pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
 688                         config->nfs_export = false;
 689                 } else if (nfs_export_opt && index_opt) {
 690                         pr_err("conflicting options: nfs_export=on,index=off\n");
 691                         return -EINVAL;
 692                 } else if (index_opt) {
 693                         /*
 694                          * There was an explicit index=off that resulted
 695                          * in this conflict.
 696                          */
 697                         pr_info("disabling nfs_export due to index=off\n");
 698                         config->nfs_export = false;
 699                 } else {
 700                         /* Automatically enable index otherwise. */
 701                         config->index = true;
 702                 }
 703         }
 704
 705         /* Resolve nfs_export -> !metacopy dependency */
 706         if (config->nfs_export && config->metacopy) {
 707                 if (nfs_export_opt && metacopy_opt) {
 708                         pr_err("conflicting options: nfs_export=on,metacopy=on\n");
 709                         return -EINVAL;
 710                 }
 711                 if (metacopy_opt) {
 712                         /*
 713                          * There was an explicit metacopy=on that resulted
 714                          * in this conflict.
 715                          */
 716                         pr_info("disabling nfs_export due to metacopy=on\n");
 717                         config->nfs_export = false;
 718                 } else {
 719                         /*
 720                          * There was an explicit nfs_export=on that resulted
 721                          * in this conflict.
 722                          */
 723                         pr_info("disabling metacopy due to nfs_export=on\n");
 724                         config->metacopy = false;
 725                 }
 726         }
 727
 728
 729         /* Resolve userxattr -> !redirect && !metacopy dependency */
 730         if (config->userxattr) {
 731                 if (config->redirect_follow && redirect_opt) {
 732                         pr_err("conflicting options: userxattr,redirect_dir=%s\n",
 733                                config->redirect_mode);
 734                         return -EINVAL;
 735                 }
 736                 if (config->metacopy && metacopy_opt) {
 737                         pr_err("conflicting options: userxattr,metacopy=on\n");
 738                         return -EINVAL;
 739                 }
 740                 /*
 741                  * Silently disable default setting of redirect and metacopy.
 742                  * This shall be the default in the future as well: these
 743                  * options must be explicitly enabled if used together with
 744                  * userxattr.
 745                  */
 746                 config->redirect_dir = config->redirect_follow = false;
 747                 config->metacopy = false;
 748         }
 749
 750         return 0;
 751 }
 752
 753 #define OVL_WORKDIR_NAME "work"
 754 #define OVL_INDEXDIR_NAME "index"
 755
 756 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 757                                          const char *name, bool persist)
 758 {
 759         struct inode *dir =  ofs->workbasedir->d_inode;
 760         struct vfsmount *mnt = ovl_upper_mnt(ofs);
 761         struct dentry *work;
 762         int err;
 763         bool retried = false;
 764
 765         inode_lock_nested(dir, I_MUTEX_PARENT);
 766 retry:
 767         work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));
 768
 769         if (!IS_ERR(work)) {
 770                 struct iattr attr = {
 771                         .ia_valid = ATTR_MODE,
 772                         .ia_mode = S_IFDIR | 0,
 773                 };
 774
 775                 if (work->d_inode) {
 776                         err = -EEXIST;
 777                         if (retried)
 778                                 goto out_dput;
 779
 780                         if (persist)
 781                                 goto out_unlock;
 782
 783                         retried = true;
 784                         err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);
 785                         dput(work);
 786                         if (err == -EINVAL) {
 787                                 work = ERR_PTR(err);
 788                                 goto out_unlock;
 789                         }
 790                         goto retry;
 791                 }
 792
 793                 err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);
 794                 if (err)
 795                         goto out_dput;
 796
 797                 /* Weird filesystem returning with hashed negative (kernfs)? */
 798                 err = -EINVAL;
 799                 if (d_really_is_negative(work))
 800                         goto out_dput;
 801
 802                 /*
 803                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 804                  *
 805                  * a) success (there was a POSIX ACL xattr and was removed)
 806                  * b) -ENODATA (there was no POSIX ACL xattr)
 807                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 808                  *
 809                  * There are various other error values that could effectively
 810                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 811                  * if the xattr name is too long), but the set of filesystems
 812                  * allowed as upper are limited to "normal" ones, where checking
 813                  * for the above two errors is sufficient.
 814                  */
 815                 err = ovl_do_removexattr(ofs, work,
 816                                          XATTR_NAME_POSIX_ACL_DEFAULT);
 817                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 818                         goto out_dput;
 819
 820                 err = ovl_do_removexattr(ofs, work,
 821                                          XATTR_NAME_POSIX_ACL_ACCESS);
 822                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 823                         goto out_dput;
 824
 825                 /* Clear any inherited mode bits */
 826                 inode_lock(work->d_inode);
 827                 err = ovl_do_notify_change(ofs, work, &attr);
 828                 inode_unlock(work->d_inode);
 829                 if (err)
 830                         goto out_dput;
 831         } else {
 832                 err = PTR_ERR(work);
 833                 goto out_err;
 834         }
 835 out_unlock:
 836         inode_unlock(dir);
 837         return work;
 838
 839 out_dput:
 840         dput(work);
 841 out_err:
 842         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
 843                 ofs->config.workdir, name, -err);
 844         work = NULL;
 845         goto out_unlock;
 846 }
 847
 848 static void ovl_unescape(char *s)
 849 {
 850         char *d = s;
 851
 852         for (;; s++, d++) {
 853                 if (*s == '\\')
 854                         s++;
 855                 *d = *s;
 856                 if (!*s)
 857                         break;
 858         }
 859 }
 860
 861 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 862 {
 863         int err = -EINVAL;
 864
 865         if (!*name) {
 866                 pr_err("empty lowerdir\n");
 867                 goto out;
 868         }
 869         err = kern_path(name, LOOKUP_FOLLOW, path);
 870         if (err) {
 871                 pr_err("failed to resolve '%s': %i\n", name, err);
 872                 goto out;
 873         }
 874         err = -EINVAL;
 875         if (ovl_dentry_weird(path->dentry)) {
 876                 pr_err("filesystem on '%s' not supported\n", name);
 877                 goto out_put;
 878         }
 879         if (!d_is_dir(path->dentry)) {
 880                 pr_err("'%s' not a directory\n", name);
 881                 goto out_put;
 882         }
 883         return 0;
 884
 885 out_put:
 886         path_put_init(path);
 887 out:
 888         return err;
 889 }
 890
 891 static int ovl_mount_dir(const char *name, struct path *path)
 892 {
 893         int err = -ENOMEM;
 894         char *tmp = kstrdup(name, GFP_KERNEL);
 895
 896         if (tmp) {
 897                 ovl_unescape(tmp);
 898                 err = ovl_mount_dir_noesc(tmp, path);
 899
 900                 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
 901                         pr_err("filesystem on '%s' not supported as upperdir\n",
 902                                tmp);
 903                         path_put_init(path);
 904                         err = -EINVAL;
 905                 }
 906                 kfree(tmp);
 907         }
 908         return err;
 909 }
 910
 911 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 912                              const char *name)
 913 {
 914         struct kstatfs statfs;
 915         int err = vfs_statfs(path, &statfs);
 916
 917         if (err)
 918                 pr_err("statfs failed on '%s'\n", name);
 919         else
 920                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 921
 922         return err;
 923 }
 924
 925 static int ovl_lower_dir(const char *name, struct path *path,
 926                          struct ovl_fs *ofs, int *stack_depth)
 927 {
 928         int fh_type;
 929         int err;
 930
 931         err = ovl_mount_dir_noesc(name, path);
 932         if (err)
 933                 return err;
 934
 935         err = ovl_check_namelen(path, ofs, name);
 936         if (err)
 937                 return err;
 938
 939         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 940
 941         /*
 942          * The inodes index feature and NFS export need to encode and decode
 943          * file handles, so they require that all layers support them.
 944          */
 945         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 946         if ((ofs->config.nfs_export ||
 947              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 948                 ofs->config.index = false;
 949                 ofs->config.nfs_export = false;
 950                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 951                         name);
 952         }
 953         /*
 954          * Decoding origin file handle is required for persistent st_ino.
 955          * Without persistent st_ino, xino=auto falls back to xino=off.
 956          */
 957         if (ofs->config.xino == OVL_XINO_AUTO &&
 958             ofs->config.upperdir && !fh_type) {
 959                 ofs->config.xino = OVL_XINO_OFF;
 960                 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
 961                         name);
 962         }
 963
 964         /* Check if lower fs has 32bit inode numbers */
 965         if (fh_type != FILEID_INO32_GEN)
 966                 ofs->xino_mode = -1;
 967
 968         return 0;
 969 }
 970
 971 /* Workdir should not be subdir of upperdir and vice versa */
 972 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 973 {
 974         bool ok = false;
 975
 976         if (workdir != upperdir) {
 977                 ok = (lock_rename(workdir, upperdir) == NULL);
 978                 unlock_rename(workdir, upperdir);
 979         }
 980         return ok;
 981 }
 982
 983 static unsigned int ovl_split_lowerdirs(char *str)
 984 {
 985         unsigned int ctr = 1;
 986         char *s, *d;
 987
 988         for (s = d = str;; s++, d++) {
 989                 if (*s == '\\') {
 990                         s++;
 991                 } else if (*s == ':') {
 992                         *d = '\0';
 993                         ctr++;
 994                         continue;
 995                 }
 996                 *d = *s;
 997                 if (!*s)
 998                         break;
 999         }
1000         return ctr;
1001 }
1002
1003 static int __maybe_unused
1004 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
1005                         struct dentry *dentry, struct inode *inode,
1006                         const char *name, void *buffer, size_t size)
1007 {
1008         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
1009 }
1010
1011 static int __maybe_unused
1012 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
1013                         struct user_namespace *mnt_userns,
1014                         struct dentry *dentry, struct inode *inode,
1015                         const char *name, const void *value,
1016                         size_t size, int flags)
1017 {
1018         struct dentry *workdir = ovl_workdir(dentry);
1019         struct inode *realinode = ovl_inode_real(inode);
1020         struct posix_acl *acl = NULL;
1021         int err;
1022
1023         /* Check that everything is OK before copy-up */
1024         if (value) {
1025                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
1026                 if (IS_ERR(acl))
1027                         return PTR_ERR(acl);
1028         }
1029         err = -EOPNOTSUPP;
1030         if (!IS_POSIXACL(d_inode(workdir)))
1031                 goto out_acl_release;
1032         if (!realinode->i_op->set_acl)
1033                 goto out_acl_release;
1034         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1035                 err = acl ? -EACCES : 0;
1036                 goto out_acl_release;
1037         }
1038         err = -EPERM;
1039         if (!inode_owner_or_capable(&init_user_ns, inode))
1040                 goto out_acl_release;
1041
1042         posix_acl_release(acl);
1043
1044         /*
1045          * Check if sgid bit needs to be cleared (actual setacl operation will
1046          * be done with mounter's capabilities and so that won't do it for us).
1047          */
1048         if (unlikely(inode->i_mode & S_ISGID) &&
1049             handler->flags == ACL_TYPE_ACCESS &&
1050             !in_group_p(inode->i_gid) &&
1051             !capable_wrt_inode_uidgid(&init_user_ns, inode, CAP_FSETID)) {
1052                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
1053
1054                 err = ovl_setattr(&init_user_ns, dentry, &iattr);
1055                 if (err)
1056                         return err;
1057         }
1058
1059         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
1060         return err;
1061
1062 out_acl_release:
1063         posix_acl_release(acl);
1064         return err;
1065 }
1066
1067 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1068                              struct dentry *dentry, struct inode *inode,
1069                              const char *name, void *buffer, size_t size)
1070 {
1071         return -EOPNOTSUPP;
1072 }
1073
1074 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1075                              struct user_namespace *mnt_userns,
1076                              struct dentry *dentry, struct inode *inode,
1077                              const char *name, const void *value,
1078                              size_t size, int flags)
1079 {
1080         return -EOPNOTSUPP;
1081 }
1082
1083 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1084                                struct dentry *dentry, struct inode *inode,
1085                                const char *name, void *buffer, size_t size)
1086 {
1087         return ovl_xattr_get(dentry, inode, name, buffer, size);
1088 }
1089
1090 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1091                                struct user_namespace *mnt_userns,
1092                                struct dentry *dentry, struct inode *inode,
1093                                const char *name, const void *value,
1094                                size_t size, int flags)
1095 {
1096         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1097 }
1098
1099 static const struct xattr_handler __maybe_unused
1100 ovl_posix_acl_access_xattr_handler = {
1101         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1102         .flags = ACL_TYPE_ACCESS,
1103         .get = ovl_posix_acl_xattr_get,
1104         .set = ovl_posix_acl_xattr_set,
1105 };
1106
1107 static const struct xattr_handler __maybe_unused
1108 ovl_posix_acl_default_xattr_handler = {
1109         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1110         .flags = ACL_TYPE_DEFAULT,
1111         .get = ovl_posix_acl_xattr_get,
1112         .set = ovl_posix_acl_xattr_set,
1113 };
1114
1115 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1116         .prefix = OVL_XATTR_TRUSTED_PREFIX,
1117         .get = ovl_own_xattr_get,
1118         .set = ovl_own_xattr_set,
1119 };
1120
1121 static const struct xattr_handler ovl_own_user_xattr_handler = {
1122         .prefix = OVL_XATTR_USER_PREFIX,
1123         .get = ovl_own_xattr_get,
1124         .set = ovl_own_xattr_set,
1125 };
1126
1127 static const struct xattr_handler ovl_other_xattr_handler = {
1128         .prefix = "", /* catch all */
1129         .get = ovl_other_xattr_get,
1130         .set = ovl_other_xattr_set,
1131 };
1132
1133 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1134 #ifdef CONFIG_FS_POSIX_ACL
1135         &ovl_posix_acl_access_xattr_handler,
1136         &ovl_posix_acl_default_xattr_handler,
1137 #endif
1138         &ovl_own_trusted_xattr_handler,
1139         &ovl_other_xattr_handler,
1140         NULL
1141 };
1142
1143 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1144 #ifdef CONFIG_FS_POSIX_ACL
1145         &ovl_posix_acl_access_xattr_handler,
1146         &ovl_posix_acl_default_xattr_handler,
1147 #endif
1148         &ovl_own_user_xattr_handler,
1149         &ovl_other_xattr_handler,
1150         NULL
1151 };
1152
1153 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1154                           struct inode **ptrap, const char *name)
1155 {
1156         struct inode *trap;
1157         int err;
1158
1159         trap = ovl_get_trap_inode(sb, dir);
1160         err = PTR_ERR_OR_ZERO(trap);
1161         if (err) {
1162                 if (err == -ELOOP)
1163                         pr_err("conflicting %s path\n", name);
1164                 return err;
1165         }
1166
1167         *ptrap = trap;
1168         return 0;
1169 }
1170
1171 /*
1172  * Determine how we treat concurrent use of upperdir/workdir based on the
1173  * index feature. This is papering over mount leaks of container runtimes,
1174  * for example, an old overlay mount is leaked and now its upperdir is
1175  * attempted to be used as a lower layer in a new overlay mount.
1176  */
1177 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1178 {
1179         if (ofs->config.index) {
1180                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1181                        name);
1182                 return -EBUSY;
1183         } else {
1184                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1185                         name);
1186                 return 0;
1187         }
1188 }
1189
1190 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1191                          struct ovl_layer *upper_layer, struct path *upperpath)
1192 {
1193         struct vfsmount *upper_mnt;
1194         int err;
1195
1196         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1197         if (err)
1198                 goto out;
1199
1200         /* Upperdir path should not be r/o */
1201         if (__mnt_is_readonly(upperpath->mnt)) {
1202                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1203                 err = -EINVAL;
1204                 goto out;
1205         }
1206
1207         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1208         if (err)
1209                 goto out;
1210
1211         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1212                              "upperdir");
1213         if (err)
1214                 goto out;
1215
1216         upper_mnt = clone_private_mount(upperpath);
1217         err = PTR_ERR(upper_mnt);
1218         if (IS_ERR(upper_mnt)) {
1219                 pr_err("failed to clone upperpath\n");
1220                 goto out;
1221         }
1222
1223         /* Don't inherit atime flags */
1224         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1225         upper_layer->mnt = upper_mnt;
1226         upper_layer->idx = 0;
1227         upper_layer->fsid = 0;
1228
1229         /*
1230          * Inherit SB_NOSEC flag from upperdir.
1231          *
1232          * This optimization changes behavior when a security related attribute
1233          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1234          * okay because we don't yet have guarantees in that case, but it will
1235          * need careful treatment once we want to honour changes to underlying
1236          * filesystems.
1237          */
1238         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1239                 sb->s_flags |= SB_NOSEC;
1240
1241         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1242                 ofs->upperdir_locked = true;
1243         } else {
1244                 err = ovl_report_in_use(ofs, "upperdir");
1245                 if (err)
1246                         goto out;
1247         }
1248
1249         err = 0;
1250 out:
1251         return err;
1252 }
1253
1254 /*
1255  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1256  * negative values if error is encountered.
1257  */
1258 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
1259 {
1260         struct dentry *workdir = ofs->workdir;
1261         struct inode *dir = d_inode(workdir);
1262         struct dentry *temp;
1263         struct dentry *dest;
1264         struct dentry *whiteout;
1265         struct name_snapshot name;
1266         int err;
1267
1268         inode_lock_nested(dir, I_MUTEX_PARENT);
1269
1270         temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
1271         err = PTR_ERR(temp);
1272         if (IS_ERR(temp))
1273                 goto out_unlock;
1274
1275         dest = ovl_lookup_temp(ofs, workdir);
1276         err = PTR_ERR(dest);
1277         if (IS_ERR(dest)) {
1278                 dput(temp);
1279                 goto out_unlock;
1280         }
1281
1282         /* Name is inline and stable - using snapshot as a copy helper */
1283         take_dentry_name_snapshot(&name, temp);
1284         err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
1285         if (err) {
1286                 if (err == -EINVAL)
1287                         err = 0;
1288                 goto cleanup_temp;
1289         }
1290
1291         whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
1292         err = PTR_ERR(whiteout);
1293         if (IS_ERR(whiteout))
1294                 goto cleanup_temp;
1295
1296         err = ovl_is_whiteout(whiteout);
1297
1298         /* Best effort cleanup of whiteout and temp file */
1299         if (err)
1300                 ovl_cleanup(ofs, dir, whiteout);
1301         dput(whiteout);
1302
1303 cleanup_temp:
1304         ovl_cleanup(ofs, dir, temp);
1305         release_dentry_name_snapshot(&name);
1306         dput(temp);
1307         dput(dest);
1308
1309 out_unlock:
1310         inode_unlock(dir);
1311
1312         return err;
1313 }
1314
1315 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
1316                                            struct dentry *parent,
1317                                            const char *name, umode_t mode)
1318 {
1319         size_t len = strlen(name);
1320         struct dentry *child;
1321
1322         inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1323         child = ovl_lookup_upper(ofs, name, parent, len);
1324         if (!IS_ERR(child) && !child->d_inode)
1325                 child = ovl_create_real(ofs, parent->d_inode, child,
1326                                         OVL_CATTR(mode));
1327         inode_unlock(parent->d_inode);
1328         dput(parent);
1329
1330         return child;
1331 }
1332
1333 /*
1334  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1335  * present.
1336  */
1337 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1338 {
1339         unsigned int ctr;
1340         struct dentry *d = dget(ofs->workbasedir);
1341         static const char *const volatile_path[] = {
1342                 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1343         };
1344         const char *const *name = volatile_path;
1345
1346         for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1347                 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1348                 if (IS_ERR(d))
1349                         return PTR_ERR(d);
1350         }
1351         dput(d);
1352         return 0;
1353 }
1354
1355 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1356                             struct path *workpath)
1357 {
1358         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1359         struct dentry *temp, *workdir;
1360         bool rename_whiteout;
1361         bool d_type;
1362         int fh_type;
1363         int err;
1364
1365         err = mnt_want_write(mnt);
1366         if (err)
1367                 return err;
1368
1369         workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1370         err = PTR_ERR(workdir);
1371         if (IS_ERR_OR_NULL(workdir))
1372                 goto out;
1373
1374         ofs->workdir = workdir;
1375
1376         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1377         if (err)
1378                 goto out;
1379
1380         /*
1381          * Upper should support d_type, else whiteouts are visible.  Given
1382          * workdir and upper are on same fs, we can do iterate_dir() on
1383          * workdir. This check requires successful creation of workdir in
1384          * previous step.
1385          */
1386         err = ovl_check_d_type_supported(workpath);
1387         if (err < 0)
1388                 goto out;
1389
1390         d_type = err;
1391         if (!d_type)
1392                 pr_warn("upper fs needs to support d_type.\n");
1393
1394         /* Check if upper/work fs supports O_TMPFILE */
1395         temp = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
1396         ofs->tmpfile = !IS_ERR(temp);
1397         if (ofs->tmpfile)
1398                 dput(temp);
1399         else
1400                 pr_warn("upper fs does not support tmpfile.\n");
1401
1402
1403         /* Check if upper/work fs supports RENAME_WHITEOUT */
1404         err = ovl_check_rename_whiteout(ofs);
1405         if (err < 0)
1406                 goto out;
1407
1408         rename_whiteout = err;
1409         if (!rename_whiteout)
1410                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1411
1412         /*
1413          * Check if upper/work fs supports (trusted|user).overlay.* xattr
1414          */
1415         err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1416         if (err) {
1417                 pr_warn("failed to set xattr on upper\n");
1418                 ofs->noxattr = true;
1419                 if (ofs->config.index || ofs->config.metacopy) {
1420                         ofs->config.index = false;
1421                         ofs->config.metacopy = false;
1422                         pr_warn("...falling back to index=off,metacopy=off.\n");
1423                 }
1424                 /*
1425                  * xattr support is required for persistent st_ino.
1426                  * Without persistent st_ino, xino=auto falls back to xino=off.
1427                  */
1428                 if (ofs->config.xino == OVL_XINO_AUTO) {
1429                         ofs->config.xino = OVL_XINO_OFF;
1430                         pr_warn("...falling back to xino=off.\n");
1431                 }
1432                 if (err == -EPERM && !ofs->config.userxattr)
1433                         pr_info("try mounting with 'userxattr' option\n");
1434                 err = 0;
1435         } else {
1436                 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1437         }
1438
1439         /*
1440          * We allowed sub-optimal upper fs configuration and don't want to break
1441          * users over kernel upgrade, but we never allowed remote upper fs, so
1442          * we can enforce strict requirements for remote upper fs.
1443          */
1444         if (ovl_dentry_remote(ofs->workdir) &&
1445             (!d_type || !rename_whiteout || ofs->noxattr)) {
1446                 pr_err("upper fs missing required features.\n");
1447                 err = -EINVAL;
1448                 goto out;
1449         }
1450
1451         /*
1452          * For volatile mount, create a incompat/volatile/dirty file to keep
1453          * track of it.
1454          */
1455         if (ofs->config.ovl_volatile) {
1456                 err = ovl_create_volatile_dirty(ofs);
1457                 if (err < 0) {
1458                         pr_err("Failed to create volatile/dirty file.\n");
1459                         goto out;
1460                 }
1461         }
1462
1463         /* Check if upper/work fs supports file handles */
1464         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1465         if (ofs->config.index && !fh_type) {
1466                 ofs->config.index = false;
1467                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1468         }
1469
1470         /* Check if upper fs has 32bit inode numbers */
1471         if (fh_type != FILEID_INO32_GEN)
1472                 ofs->xino_mode = -1;
1473
1474         /* NFS export of r/w mount depends on index */
1475         if (ofs->config.nfs_export && !ofs->config.index) {
1476                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1477                 ofs->config.nfs_export = false;
1478         }
1479 out:
1480         mnt_drop_write(mnt);
1481         return err;
1482 }
1483
1484 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1485                            struct path *upperpath)
1486 {
1487         int err;
1488         struct path workpath = { };
1489
1490         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1491         if (err)
1492                 goto out;
1493
1494         err = -EINVAL;
1495         if (upperpath->mnt != workpath.mnt) {
1496                 pr_err("workdir and upperdir must reside under the same mount\n");
1497                 goto out;
1498         }
1499         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1500                 pr_err("workdir and upperdir must be separate subtrees\n");
1501                 goto out;
1502         }
1503
1504         ofs->workbasedir = dget(workpath.dentry);
1505
1506         if (ovl_inuse_trylock(ofs->workbasedir)) {
1507                 ofs->workdir_locked = true;
1508         } else {
1509                 err = ovl_report_in_use(ofs, "workdir");
1510                 if (err)
1511                         goto out;
1512         }
1513
1514         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1515                              "workdir");
1516         if (err)
1517                 goto out;
1518
1519         err = ovl_make_workdir(sb, ofs, &workpath);
1520
1521 out:
1522         path_put(&workpath);
1523
1524         return err;
1525 }
1526
1527 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1528                             struct ovl_entry *oe, struct path *upperpath)
1529 {
1530         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1531         struct dentry *indexdir;
1532         int err;
1533
1534         err = mnt_want_write(mnt);
1535         if (err)
1536                 return err;
1537
1538         /* Verify lower root is upper root origin */
1539         err = ovl_verify_origin(ofs, upperpath->dentry,
1540                                 oe->lowerstack[0].dentry, true);
1541         if (err) {
1542                 pr_err("failed to verify upper root origin\n");
1543                 goto out;
1544         }
1545
1546         /* index dir will act also as workdir */
1547         iput(ofs->workdir_trap);
1548         ofs->workdir_trap = NULL;
1549         dput(ofs->workdir);
1550         ofs->workdir = NULL;
1551         indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1552         if (IS_ERR(indexdir)) {
1553                 err = PTR_ERR(indexdir);
1554         } else if (indexdir) {
1555                 ofs->indexdir = indexdir;
1556                 ofs->workdir = dget(indexdir);
1557
1558                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1559                                      "indexdir");
1560                 if (err)
1561                         goto out;
1562
1563                 /*
1564                  * Verify upper root is exclusively associated with index dir.
1565                  * Older kernels stored upper fh in ".overlay.origin"
1566                  * xattr. If that xattr exists, verify that it is a match to
1567                  * upper dir file handle. In any case, verify or set xattr
1568                  * ".overlay.upper" to indicate that index may have
1569                  * directory entries.
1570                  */
1571                 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1572                         err = ovl_verify_set_fh(ofs, ofs->indexdir,
1573                                                 OVL_XATTR_ORIGIN,
1574                                                 upperpath->dentry, true, false);
1575                         if (err)
1576                                 pr_err("failed to verify index dir 'origin' xattr\n");
1577                 }
1578                 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1579                                        true);
1580                 if (err)
1581                         pr_err("failed to verify index dir 'upper' xattr\n");
1582
1583                 /* Cleanup bad/stale/orphan index entries */
1584                 if (!err)
1585                         err = ovl_indexdir_cleanup(ofs);
1586         }
1587         if (err || !ofs->indexdir)
1588                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1589
1590 out:
1591         mnt_drop_write(mnt);
1592         return err;
1593 }
1594
1595 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1596 {
1597         unsigned int i;
1598
1599         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1600                 return true;
1601
1602         /*
1603          * We allow using single lower with null uuid for index and nfs_export
1604          * for example to support those features with single lower squashfs.
1605          * To avoid regressions in setups of overlay with re-formatted lower
1606          * squashfs, do not allow decoding origin with lower null uuid unless
1607          * user opted-in to one of the new features that require following the
1608          * lower inode of non-dir upper.
1609          */
1610         if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
1611                 return false;
1612
1613         for (i = 0; i < ofs->numfs; i++) {
1614                 /*
1615                  * We use uuid to associate an overlay lower file handle with a
1616                  * lower layer, so we can accept lower fs with null uuid as long
1617                  * as all lower layers with null uuid are on the same fs.
1618                  * if we detect multiple lower fs with the same uuid, we
1619                  * disable lower file handle decoding on all of them.
1620                  */
1621                 if (ofs->fs[i].is_lower &&
1622                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1623                         ofs->fs[i].bad_uuid = true;
1624                         return false;
1625                 }
1626         }
1627         return true;
1628 }
1629
1630 /* Get a unique fsid for the layer */
1631 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1632 {
1633         struct super_block *sb = path->mnt->mnt_sb;
1634         unsigned int i;
1635         dev_t dev;
1636         int err;
1637         bool bad_uuid = false;
1638         bool warn = false;
1639
1640         for (i = 0; i < ofs->numfs; i++) {
1641                 if (ofs->fs[i].sb == sb)
1642                         return i;
1643         }
1644
1645         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1646                 bad_uuid = true;
1647                 if (ofs->config.xino == OVL_XINO_AUTO) {
1648                         ofs->config.xino = OVL_XINO_OFF;
1649                         warn = true;
1650                 }
1651                 if (ofs->config.index || ofs->config.nfs_export) {
1652                         ofs->config.index = false;
1653                         ofs->config.nfs_export = false;
1654                         warn = true;
1655                 }
1656                 if (warn) {
1657                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1658                                 uuid_is_null(&sb->s_uuid) ? "null" :
1659                                                             "conflicting",
1660                                 path->dentry, ovl_xino_str[ofs->config.xino]);
1661                 }
1662         }
1663
1664         err = get_anon_bdev(&dev);
1665         if (err) {
1666                 pr_err("failed to get anonymous bdev for lowerpath\n");
1667                 return err;
1668         }
1669
1670         ofs->fs[ofs->numfs].sb = sb;
1671         ofs->fs[ofs->numfs].pseudo_dev = dev;
1672         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1673
1674         return ofs->numfs++;
1675 }
1676
1677 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1678                           struct path *stack, unsigned int numlower,
1679                           struct ovl_layer *layers)
1680 {
1681         int err;
1682         unsigned int i;
1683
1684         err = -ENOMEM;
1685         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1686         if (ofs->fs == NULL)
1687                 goto out;
1688
1689         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1690         ofs->numfs++;
1691
1692         /*
1693          * All lower layers that share the same fs as upper layer, use the same
1694          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1695          * only overlay to simplify ovl_fs_free().
1696          * is_lower will be set if upper fs is shared with a lower layer.
1697          */
1698         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1699         if (err) {
1700                 pr_err("failed to get anonymous bdev for upper fs\n");
1701                 goto out;
1702         }
1703
1704         if (ovl_upper_mnt(ofs)) {
1705                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1706                 ofs->fs[0].is_lower = false;
1707         }
1708
1709         for (i = 0; i < numlower; i++) {
1710                 struct vfsmount *mnt;
1711                 struct inode *trap;
1712                 int fsid;
1713
1714                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1715                 if (err < 0)
1716                         goto out;
1717
1718                 /*
1719                  * Check if lower root conflicts with this overlay layers before
1720                  * checking if it is in-use as upperdir/workdir of "another"
1721                  * mount, because we do not bother to check in ovl_is_inuse() if
1722                  * the upperdir/workdir is in fact in-use by our
1723                  * upperdir/workdir.
1724                  */
1725                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1726                 if (err)
1727                         goto out;
1728
1729                 if (ovl_is_inuse(stack[i].dentry)) {
1730                         err = ovl_report_in_use(ofs, "lowerdir");
1731                         if (err) {
1732                                 iput(trap);
1733                                 goto out;
1734                         }
1735                 }
1736
1737                 mnt = clone_private_mount(&stack[i]);
1738                 err = PTR_ERR(mnt);
1739                 if (IS_ERR(mnt)) {
1740                         pr_err("failed to clone lowerpath\n");
1741                         iput(trap);
1742                         goto out;
1743                 }
1744
1745                 /*
1746                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1747                  * will fail instead of modifying lower fs.
1748                  */
1749                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1750
1751                 layers[ofs->numlayer].trap = trap;
1752                 layers[ofs->numlayer].mnt = mnt;
1753                 layers[ofs->numlayer].idx = ofs->numlayer;
1754                 layers[ofs->numlayer].fsid = fsid;
1755                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1756                 ofs->numlayer++;
1757                 ofs->fs[fsid].is_lower = true;
1758         }
1759
1760         /*
1761          * When all layers on same fs, overlay can use real inode numbers.
1762          * With mount option "xino=<on|auto>", mounter declares that there are
1763          * enough free high bits in underlying fs to hold the unique fsid.
1764          * If overlayfs does encounter underlying inodes using the high xino
1765          * bits reserved for fsid, it emits a warning and uses the original
1766          * inode number or a non persistent inode number allocated from a
1767          * dedicated range.
1768          */
1769         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1770                 if (ofs->config.xino == OVL_XINO_ON)
1771                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1772                 ofs->xino_mode = 0;
1773         } else if (ofs->config.xino == OVL_XINO_OFF) {
1774                 ofs->xino_mode = -1;
1775         } else if (ofs->xino_mode < 0) {
1776                 /*
1777                  * This is a roundup of number of bits needed for encoding
1778                  * fsid, where fsid 0 is reserved for upper fs (even with
1779                  * lower only overlay) +1 extra bit is reserved for the non
1780                  * persistent inode number range that is used for resolving
1781                  * xino lower bits overflow.
1782                  */
1783                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1784                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1785         }
1786
1787         if (ofs->xino_mode > 0) {
1788                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1789                         ofs->xino_mode);
1790         }
1791
1792         err = 0;
1793 out:
1794         return err;
1795 }
1796
1797 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1798                                 const char *lower, unsigned int numlower,
1799                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1800 {
1801         int err;
1802         struct path *stack = NULL;
1803         unsigned int i;
1804         struct ovl_entry *oe;
1805
1806         if (!ofs->config.upperdir && numlower == 1) {
1807                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1808                 return ERR_PTR(-EINVAL);
1809         }
1810
1811         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1812         if (!stack)
1813                 return ERR_PTR(-ENOMEM);
1814
1815         err = -EINVAL;
1816         for (i = 0; i < numlower; i++) {
1817                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1818                 if (err)
1819                         goto out_err;
1820
1821                 lower = strchr(lower, '\0') + 1;
1822         }
1823
1824         err = -EINVAL;
1825         sb->s_stack_depth++;
1826         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1827                 pr_err("maximum fs stacking depth exceeded\n");
1828                 goto out_err;
1829         }
1830
1831         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1832         if (err)
1833                 goto out_err;
1834
1835         err = -ENOMEM;
1836         oe = ovl_alloc_entry(numlower);
1837         if (!oe)
1838                 goto out_err;
1839
1840         for (i = 0; i < numlower; i++) {
1841                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1842                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1843         }
1844
1845 out:
1846         for (i = 0; i < numlower; i++)
1847                 path_put(&stack[i]);
1848         kfree(stack);
1849
1850         return oe;
1851
1852 out_err:
1853         oe = ERR_PTR(err);
1854         goto out;
1855 }
1856
1857 /*
1858  * Check if this layer root is a descendant of:
1859  * - another layer of this overlayfs instance
1860  * - upper/work dir of any overlayfs instance
1861  */
1862 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1863                            struct dentry *dentry, const char *name,
1864                            bool is_lower)
1865 {
1866         struct dentry *next = dentry, *parent;
1867         int err = 0;
1868
1869         if (!dentry)
1870                 return 0;
1871
1872         parent = dget_parent(next);
1873
1874         /* Walk back ancestors to root (inclusive) looking for traps */
1875         while (!err && parent != next) {
1876                 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1877                         err = -ELOOP;
1878                         pr_err("overlapping %s path\n", name);
1879                 } else if (ovl_is_inuse(parent)) {
1880                         err = ovl_report_in_use(ofs, name);
1881                 }
1882                 next = parent;
1883                 parent = dget_parent(next);
1884                 dput(next);
1885         }
1886
1887         dput(parent);
1888
1889         return err;
1890 }
1891
1892 /*
1893  * Check if any of the layers or work dirs overlap.
1894  */
1895 static int ovl_check_overlapping_layers(struct super_block *sb,
1896                                         struct ovl_fs *ofs)
1897 {
1898         int i, err;
1899
1900         if (ovl_upper_mnt(ofs)) {
1901                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1902                                       "upperdir", false);
1903                 if (err)
1904                         return err;
1905
1906                 /*
1907                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1908                  * this instance and covers overlapping work and index dirs,
1909                  * unless work or index dir have been moved since created inside
1910                  * workbasedir.  In that case, we already have their traps in
1911                  * inode cache and we will catch that case on lookup.
1912                  */
1913                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1914                                       false);
1915                 if (err)
1916                         return err;
1917         }
1918
1919         for (i = 1; i < ofs->numlayer; i++) {
1920                 err = ovl_check_layer(sb, ofs,
1921                                       ofs->layers[i].mnt->mnt_root,
1922                                       "lowerdir", true);
1923                 if (err)
1924                         return err;
1925         }
1926
1927         return 0;
1928 }
1929
1930 static struct dentry *ovl_get_root(struct super_block *sb,
1931                                    struct dentry *upperdentry,
1932                                    struct ovl_entry *oe)
1933 {
1934         struct dentry *root;
1935         struct ovl_path *lowerpath = &oe->lowerstack[0];
1936         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1937         int fsid = lowerpath->layer->fsid;
1938         struct ovl_inode_params oip = {
1939                 .upperdentry = upperdentry,
1940                 .lowerpath = lowerpath,
1941         };
1942
1943         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1944         if (!root)
1945                 return NULL;
1946
1947         root->d_fsdata = oe;
1948
1949         if (upperdentry) {
1950                 /* Root inode uses upper st_ino/i_ino */
1951                 ino = d_inode(upperdentry)->i_ino;
1952                 fsid = 0;
1953                 ovl_dentry_set_upper_alias(root);
1954                 if (ovl_is_impuredir(sb, upperdentry))
1955                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1956         }
1957
1958         /* Root is always merge -> can have whiteouts */
1959         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1960         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1961         ovl_set_upperdata(d_inode(root));
1962         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1963         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1964
1965         return root;
1966 }
1967
1968 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1969 {
1970         struct path upperpath = { };
1971         struct dentry *root_dentry;
1972         struct ovl_entry *oe;
1973         struct ovl_fs *ofs;
1974         struct ovl_layer *layers;
1975         struct cred *cred;
1976         char *splitlower = NULL;
1977         unsigned int numlower;
1978         int err;
1979
1980         err = -EIO;
1981         if (WARN_ON(sb->s_user_ns != current_user_ns()))
1982                 goto out;
1983
1984         sb->s_d_op = &ovl_dentry_operations;
1985
1986         err = -ENOMEM;
1987         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1988         if (!ofs)
1989                 goto out;
1990
1991         err = -ENOMEM;
1992         ofs->creator_cred = cred = prepare_creds();
1993         if (!cred)
1994                 goto out_err;
1995
1996         /* Is there a reason anyone would want not to share whiteouts? */
1997         ofs->share_whiteout = true;
1998
1999         ofs->config.index = ovl_index_def;
2000         ofs->config.uuid = true;
2001         ofs->config.nfs_export = ovl_nfs_export_def;
2002         ofs->config.xino = ovl_xino_def();
2003         ofs->config.metacopy = ovl_metacopy_def;
2004         err = ovl_parse_opt((char *) data, &ofs->config);
2005         if (err)
2006                 goto out_err;
2007
2008         err = -EINVAL;
2009         if (!ofs->config.lowerdir) {
2010                 if (!silent)
2011                         pr_err("missing 'lowerdir'\n");
2012                 goto out_err;
2013         }
2014
2015         err = -ENOMEM;
2016         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
2017         if (!splitlower)
2018                 goto out_err;
2019
2020         err = -EINVAL;
2021         numlower = ovl_split_lowerdirs(splitlower);
2022         if (numlower > OVL_MAX_STACK) {
2023                 pr_err("too many lower directories, limit is %d\n",
2024                        OVL_MAX_STACK);
2025                 goto out_err;
2026         }
2027
2028         err = -ENOMEM;
2029         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
2030         if (!layers)
2031                 goto out_err;
2032
2033         ofs->layers = layers;
2034         /* Layer 0 is reserved for upper even if there's no upper */
2035         ofs->numlayer = 1;
2036
2037         sb->s_stack_depth = 0;
2038         sb->s_maxbytes = MAX_LFS_FILESIZE;
2039         atomic_long_set(&ofs->last_ino, 1);
2040         /* Assume underlying fs uses 32bit inodes unless proven otherwise */
2041         if (ofs->config.xino != OVL_XINO_OFF) {
2042                 ofs->xino_mode = BITS_PER_LONG - 32;
2043                 if (!ofs->xino_mode) {
2044                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
2045                         ofs->config.xino = OVL_XINO_OFF;
2046                 }
2047         }
2048
2049         /* alloc/destroy_inode needed for setting up traps in inode cache */
2050         sb->s_op = &ovl_super_operations;
2051
2052         if (ofs->config.upperdir) {
2053                 struct super_block *upper_sb;
2054
2055                 err = -EINVAL;
2056                 if (!ofs->config.workdir) {
2057                         pr_err("missing 'workdir'\n");
2058                         goto out_err;
2059                 }
2060
2061                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
2062                 if (err)
2063                         goto out_err;
2064
2065                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
2066                 if (!ovl_should_sync(ofs)) {
2067                         ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
2068                         if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
2069                                 err = -EIO;
2070                                 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
2071                                 goto out_err;
2072                         }
2073                 }
2074
2075                 err = ovl_get_workdir(sb, ofs, &upperpath);
2076                 if (err)
2077                         goto out_err;
2078
2079                 if (!ofs->workdir)
2080                         sb->s_flags |= SB_RDONLY;
2081
2082                 sb->s_stack_depth = upper_sb->s_stack_depth;
2083                 sb->s_time_gran = upper_sb->s_time_gran;
2084         }
2085         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2086         err = PTR_ERR(oe);
2087         if (IS_ERR(oe))
2088                 goto out_err;
2089
2090         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2091         if (!ovl_upper_mnt(ofs))
2092                 sb->s_flags |= SB_RDONLY;
2093
2094         if (!ofs->config.uuid && ofs->numfs > 1) {
2095                 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2096                 ofs->config.uuid = true;
2097         }
2098
2099         if (!ovl_force_readonly(ofs) && ofs->config.index) {
2100                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2101                 if (err)
2102                         goto out_free_oe;
2103
2104                 /* Force r/o mount with no index dir */
2105                 if (!ofs->indexdir)
2106                         sb->s_flags |= SB_RDONLY;
2107         }
2108
2109         err = ovl_check_overlapping_layers(sb, ofs);
2110         if (err)
2111                 goto out_free_oe;
2112
2113         /* Show index=off in /proc/mounts for forced r/o mount */
2114         if (!ofs->indexdir) {
2115                 ofs->config.index = false;
2116                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2117                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2118                         ofs->config.nfs_export = false;
2119                 }
2120         }
2121
2122         if (ofs->config.metacopy && ofs->config.nfs_export) {
2123                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2124                 ofs->config.nfs_export = false;
2125         }
2126
2127         if (ofs->config.nfs_export)
2128                 sb->s_export_op = &ovl_export_operations;
2129
2130         /* Never override disk quota limits or use reserved space */
2131         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2132
2133         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2134         sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2135                 ovl_trusted_xattr_handlers;
2136         sb->s_fs_info = ofs;
2137         sb->s_flags |= SB_POSIXACL;
2138         sb->s_iflags |= SB_I_SKIP_SYNC;
2139
2140         err = -ENOMEM;
2141         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2142         if (!root_dentry)
2143                 goto out_free_oe;
2144
2145         mntput(upperpath.mnt);
2146         kfree(splitlower);
2147
2148         sb->s_root = root_dentry;
2149
2150         return 0;
2151
2152 out_free_oe:
2153         ovl_entry_stack_free(oe);
2154         kfree(oe);
2155 out_err:
2156         kfree(splitlower);
2157         path_put(&upperpath);
2158         ovl_free_fs(ofs);
2159 out:
2160         return err;
2161 }
2162
2163 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2164                                 const char *dev_name, void *raw_data)
2165 {
2166         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2167 }
2168
2169 static struct file_system_type ovl_fs_type = {
2170         .owner          = THIS_MODULE,
2171         .name           = "overlay",
2172         .fs_flags       = FS_USERNS_MOUNT,
2173         .mount          = ovl_mount,
2174         .kill_sb        = kill_anon_super,
2175 };
2176 MODULE_ALIAS_FS("overlay");
2177
2178 static void ovl_inode_init_once(void *foo)
2179 {
2180         struct ovl_inode *oi = foo;
2181
2182         inode_init_once(&oi->vfs_inode);
2183 }
2184
2185 static int __init ovl_init(void)
2186 {
2187         int err;
2188
2189         ovl_inode_cachep = kmem_cache_create("ovl_inode",
2190                                              sizeof(struct ovl_inode), 0,
2191                                              (SLAB_RECLAIM_ACCOUNT|
2192                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2193                                              ovl_inode_init_once);
2194         if (ovl_inode_cachep == NULL)
2195                 return -ENOMEM;
2196
2197         err = ovl_aio_request_cache_init();
2198         if (!err) {
2199                 err = register_filesystem(&ovl_fs_type);
2200                 if (!err)
2201                         return 0;
2202
2203                 ovl_aio_request_cache_destroy();
2204         }
2205         kmem_cache_destroy(ovl_inode_cachep);
2206
2207         return err;
2208 }
2209
2210 static void __exit ovl_exit(void)
2211 {
2212         unregister_filesystem(&ovl_fs_type);
2213
2214         /*
2215          * Make sure all delayed rcu free inodes are flushed before we
2216          * destroy cache.
2217          */
2218         rcu_barrier();
2219         kmem_cache_destroy(ovl_inode_cachep);
2220         ovl_aio_request_cache_destroy();
2221 }
2222
2223 module_init(ovl_init);
2224 module_exit(ovl_exit);