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 <linux/file.h>
  19 #include "overlayfs.h"
  20
  21 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  22 MODULE_DESCRIPTION("Overlay filesystem");
  23 MODULE_LICENSE("GPL");
  24
  25
  26 struct ovl_dir_cache;
  27
  28 #define OVL_MAX_STACK 500
  29
  30 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  31 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  32 MODULE_PARM_DESC(redirect_dir,
  33                  "Default to on or off for the redirect_dir feature");
  34
  35 static bool ovl_redirect_always_follow =
  36         IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
  37 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
  38                    bool, 0644);
  39 MODULE_PARM_DESC(redirect_always_follow,
  40                  "Follow redirects even if redirect_dir feature is turned off");
  41
  42 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
  43 module_param_named(index, ovl_index_def, bool, 0644);
  44 MODULE_PARM_DESC(index,
  45                  "Default to on or off for the inodes index feature");
  46
  47 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
  48 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
  49 MODULE_PARM_DESC(nfs_export,
  50                  "Default to on or off for the NFS export feature");
  51
  52 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
  53 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
  54 MODULE_PARM_DESC(xino_auto,
  55                  "Auto enable xino feature");
  56
  57 static void ovl_entry_stack_free(struct ovl_entry *oe)
  58 {
  59         unsigned int i;
  60
  61         for (i = 0; i < oe->numlower; i++)
  62                 dput(oe->lowerstack[i].dentry);
  63 }
  64
  65 static bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
  66 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
  67 MODULE_PARM_DESC(metacopy,
  68                  "Default to on or off for the metadata only copy up feature");
  69
  70 static void ovl_dentry_release(struct dentry *dentry)
  71 {
  72         struct ovl_entry *oe = dentry->d_fsdata;
  73
  74         if (oe) {
  75                 ovl_entry_stack_free(oe);
  76                 kfree_rcu(oe, rcu);
  77         }
  78 }
  79
  80 static struct dentry *ovl_d_real(struct dentry *dentry,
  81                                  const struct inode *inode)
  82 {
  83         struct dentry *real = NULL, *lower;
  84
  85         /* It's an overlay file */
  86         if (inode && d_inode(dentry) == inode)
  87                 return dentry;
  88
  89         if (!d_is_reg(dentry)) {
  90                 if (!inode || inode == d_inode(dentry))
  91                         return dentry;
  92                 goto bug;
  93         }
  94
  95         real = ovl_dentry_upper(dentry);
  96         if (real && (inode == d_inode(real)))
  97                 return real;
  98
  99         if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
 100                 return real;
 101
 102         lower = ovl_dentry_lowerdata(dentry);
 103         if (!lower)
 104                 goto bug;
 105         real = lower;
 106
 107         /* Handle recursion */
 108         real = d_real(real, inode);
 109
 110         if (!inode || inode == d_inode(real))
 111                 return real;
 112 bug:
 113         WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
 114              __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
 115              inode ? inode->i_ino : 0, real,
 116              real && d_inode(real) ? d_inode(real)->i_ino : 0);
 117         return dentry;
 118 }
 119
 120 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
 121 {
 122         int ret = 1;
 123
 124         if (weak) {
 125                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
 126                         ret =  d->d_op->d_weak_revalidate(d, flags);
 127         } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
 128                 ret = d->d_op->d_revalidate(d, flags);
 129                 if (!ret) {
 130                         if (!(flags & LOOKUP_RCU))
 131                                 d_invalidate(d);
 132                         ret = -ESTALE;
 133                 }
 134         }
 135         return ret;
 136 }
 137
 138 static int ovl_dentry_revalidate_common(struct dentry *dentry,
 139                                         unsigned int flags, bool weak)
 140 {
 141         struct ovl_entry *oe = dentry->d_fsdata;
 142         struct inode *inode = d_inode_rcu(dentry);
 143         struct dentry *upper;
 144         unsigned int i;
 145         int ret = 1;
 146
 147         /* Careful in RCU mode */
 148         if (!inode)
 149                 return -ECHILD;
 150
 151         upper = ovl_i_dentry_upper(inode);
 152         if (upper)
 153                 ret = ovl_revalidate_real(upper, flags, weak);
 154
 155         for (i = 0; ret > 0 && i < oe->numlower; i++) {
 156                 ret = ovl_revalidate_real(oe->lowerstack[i].dentry, flags,
 157                                           weak);
 158         }
 159         return ret;
 160 }
 161
 162 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 163 {
 164         return ovl_dentry_revalidate_common(dentry, flags, false);
 165 }
 166
 167 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 168 {
 169         return ovl_dentry_revalidate_common(dentry, flags, true);
 170 }
 171
 172 static const struct dentry_operations ovl_dentry_operations = {
 173         .d_release = ovl_dentry_release,
 174         .d_real = ovl_d_real,
 175         .d_revalidate = ovl_dentry_revalidate,
 176         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 177 };
 178
 179 static struct kmem_cache *ovl_inode_cachep;
 180
 181 static struct inode *ovl_alloc_inode(struct super_block *sb)
 182 {
 183         struct ovl_inode *oi = alloc_inode_sb(sb, ovl_inode_cachep, GFP_KERNEL);
 184
 185         if (!oi)
 186                 return NULL;
 187
 188         oi->cache = NULL;
 189         oi->redirect = NULL;
 190         oi->version = 0;
 191         oi->flags = 0;
 192         oi->__upperdentry = NULL;
 193         oi->lowerpath.dentry = NULL;
 194         oi->lowerpath.layer = NULL;
 195         oi->lowerdata = NULL;
 196         mutex_init(&oi->lock);
 197
 198         return &oi->vfs_inode;
 199 }
 200
 201 static void ovl_free_inode(struct inode *inode)
 202 {
 203         struct ovl_inode *oi = OVL_I(inode);
 204
 205         kfree(oi->redirect);
 206         mutex_destroy(&oi->lock);
 207         kmem_cache_free(ovl_inode_cachep, oi);
 208 }
 209
 210 static void ovl_destroy_inode(struct inode *inode)
 211 {
 212         struct ovl_inode *oi = OVL_I(inode);
 213
 214         dput(oi->__upperdentry);
 215         dput(oi->lowerpath.dentry);
 216         if (S_ISDIR(inode->i_mode))
 217                 ovl_dir_cache_free(inode);
 218         else
 219                 iput(oi->lowerdata);
 220 }
 221
 222 static void ovl_free_fs(struct ovl_fs *ofs)
 223 {
 224         struct vfsmount **mounts;
 225         unsigned i;
 226
 227         iput(ofs->workbasedir_trap);
 228         iput(ofs->indexdir_trap);
 229         iput(ofs->workdir_trap);
 230         dput(ofs->whiteout);
 231         dput(ofs->indexdir);
 232         dput(ofs->workdir);
 233         if (ofs->workdir_locked)
 234                 ovl_inuse_unlock(ofs->workbasedir);
 235         dput(ofs->workbasedir);
 236         if (ofs->upperdir_locked)
 237                 ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
 238
 239         /* Hack!  Reuse ofs->layers as a vfsmount array before freeing it */
 240         mounts = (struct vfsmount **) ofs->layers;
 241         for (i = 0; i < ofs->numlayer; i++) {
 242                 iput(ofs->layers[i].trap);
 243                 mounts[i] = ofs->layers[i].mnt;
 244         }
 245         kern_unmount_array(mounts, ofs->numlayer);
 246         kfree(ofs->layers);
 247         for (i = 0; i < ofs->numfs; i++)
 248                 free_anon_bdev(ofs->fs[i].pseudo_dev);
 249         kfree(ofs->fs);
 250
 251         kfree(ofs->config.lowerdir);
 252         kfree(ofs->config.upperdir);
 253         kfree(ofs->config.workdir);
 254         kfree(ofs->config.redirect_mode);
 255         if (ofs->creator_cred)
 256                 put_cred(ofs->creator_cred);
 257         kfree(ofs);
 258 }
 259
 260 static void ovl_put_super(struct super_block *sb)
 261 {
 262         struct ovl_fs *ofs = sb->s_fs_info;
 263
 264         ovl_free_fs(ofs);
 265 }
 266
 267 /* Sync real dirty inodes in upper filesystem (if it exists) */
 268 static int ovl_sync_fs(struct super_block *sb, int wait)
 269 {
 270         struct ovl_fs *ofs = sb->s_fs_info;
 271         struct super_block *upper_sb;
 272         int ret;
 273
 274         ret = ovl_sync_status(ofs);
 275         /*
 276          * We have to always set the err, because the return value isn't
 277          * checked in syncfs, and instead indirectly return an error via
 278          * the sb's writeback errseq, which VFS inspects after this call.
 279          */
 280         if (ret < 0) {
 281                 errseq_set(&sb->s_wb_err, -EIO);
 282                 return -EIO;
 283         }
 284
 285         if (!ret)
 286                 return ret;
 287
 288         /*
 289          * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
 290          * All the super blocks will be iterated, including upper_sb.
 291          *
 292          * If this is a syncfs(2) call, then we do need to call
 293          * sync_filesystem() on upper_sb, but enough if we do it when being
 294          * called with wait == 1.
 295          */
 296         if (!wait)
 297                 return 0;
 298
 299         upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 300
 301         down_read(&upper_sb->s_umount);
 302         ret = sync_filesystem(upper_sb);
 303         up_read(&upper_sb->s_umount);
 304
 305         return ret;
 306 }
 307
 308 /**
 309  * ovl_statfs
 310  * @dentry: The dentry to query
 311  * @buf: The struct kstatfs to fill in with stats
 312  *
 313  * Get the filesystem statistics.  As writes always target the upper layer
 314  * filesystem pass the statfs to the upper filesystem (if it exists)
 315  */
 316 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 317 {
 318         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 319         struct dentry *root_dentry = dentry->d_sb->s_root;
 320         struct path path;
 321         int err;
 322
 323         ovl_path_real(root_dentry, &path);
 324
 325         err = vfs_statfs(&path, buf);
 326         if (!err) {
 327                 buf->f_namelen = ofs->namelen;
 328                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 329         }
 330
 331         return err;
 332 }
 333
 334 /* Will this overlay be forced to mount/remount ro? */
 335 static bool ovl_force_readonly(struct ovl_fs *ofs)
 336 {
 337         return (!ovl_upper_mnt(ofs) || !ofs->workdir);
 338 }
 339
 340 static const char *ovl_redirect_mode_def(void)
 341 {
 342         return ovl_redirect_dir_def ? "on" : "off";
 343 }
 344
 345 static const char * const ovl_xino_str[] = {
 346         "off",
 347         "auto",
 348         "on",
 349 };
 350
 351 static inline int ovl_xino_def(void)
 352 {
 353         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 354 }
 355
 356 /**
 357  * ovl_show_options
 358  * @m: the seq_file handle
 359  * @dentry: The dentry to query
 360  *
 361  * Prints the mount options for a given superblock.
 362  * Returns zero; does not fail.
 363  */
 364 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 365 {
 366         struct super_block *sb = dentry->d_sb;
 367         struct ovl_fs *ofs = sb->s_fs_info;
 368
 369         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 370         if (ofs->config.upperdir) {
 371                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 372                 seq_show_option(m, "workdir", ofs->config.workdir);
 373         }
 374         if (ofs->config.default_permissions)
 375                 seq_puts(m, ",default_permissions");
 376         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 377                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 378         if (ofs->config.index != ovl_index_def)
 379                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 380         if (!ofs->config.uuid)
 381                 seq_puts(m, ",uuid=off");
 382         if (ofs->config.nfs_export != ovl_nfs_export_def)
 383                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 384                                                 "on" : "off");
 385         if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
 386                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 387         if (ofs->config.metacopy != ovl_metacopy_def)
 388                 seq_printf(m, ",metacopy=%s",
 389                            ofs->config.metacopy ? "on" : "off");
 390         if (ofs->config.ovl_volatile)
 391                 seq_puts(m, ",volatile");
 392         if (ofs->config.userxattr)
 393                 seq_puts(m, ",userxattr");
 394         return 0;
 395 }
 396
 397 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 398 {
 399         struct ovl_fs *ofs = sb->s_fs_info;
 400         struct super_block *upper_sb;
 401         int ret = 0;
 402
 403         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 404                 return -EROFS;
 405
 406         if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
 407                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 408                 if (ovl_should_sync(ofs)) {
 409                         down_read(&upper_sb->s_umount);
 410                         ret = sync_filesystem(upper_sb);
 411                         up_read(&upper_sb->s_umount);
 412                 }
 413         }
 414
 415         return ret;
 416 }
 417
 418 static const struct super_operations ovl_super_operations = {
 419         .alloc_inode    = ovl_alloc_inode,
 420         .free_inode     = ovl_free_inode,
 421         .destroy_inode  = ovl_destroy_inode,
 422         .drop_inode     = generic_delete_inode,
 423         .put_super      = ovl_put_super,
 424         .sync_fs        = ovl_sync_fs,
 425         .statfs         = ovl_statfs,
 426         .show_options   = ovl_show_options,
 427         .remount_fs     = ovl_remount,
 428 };
 429
 430 enum {
 431         OPT_LOWERDIR,
 432         OPT_UPPERDIR,
 433         OPT_WORKDIR,
 434         OPT_DEFAULT_PERMISSIONS,
 435         OPT_REDIRECT_DIR,
 436         OPT_INDEX_ON,
 437         OPT_INDEX_OFF,
 438         OPT_UUID_ON,
 439         OPT_UUID_OFF,
 440         OPT_NFS_EXPORT_ON,
 441         OPT_USERXATTR,
 442         OPT_NFS_EXPORT_OFF,
 443         OPT_XINO_ON,
 444         OPT_XINO_OFF,
 445         OPT_XINO_AUTO,
 446         OPT_METACOPY_ON,
 447         OPT_METACOPY_OFF,
 448         OPT_VOLATILE,
 449         OPT_ERR,
 450 };
 451
 452 static const match_table_t ovl_tokens = {
 453         {OPT_LOWERDIR,                  "lowerdir=%s"},
 454         {OPT_UPPERDIR,                  "upperdir=%s"},
 455         {OPT_WORKDIR,                   "workdir=%s"},
 456         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 457         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 458         {OPT_INDEX_ON,                  "index=on"},
 459         {OPT_INDEX_OFF,                 "index=off"},
 460         {OPT_USERXATTR,                 "userxattr"},
 461         {OPT_UUID_ON,                   "uuid=on"},
 462         {OPT_UUID_OFF,                  "uuid=off"},
 463         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 464         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 465         {OPT_XINO_ON,                   "xino=on"},
 466         {OPT_XINO_OFF,                  "xino=off"},
 467         {OPT_XINO_AUTO,                 "xino=auto"},
 468         {OPT_METACOPY_ON,               "metacopy=on"},
 469         {OPT_METACOPY_OFF,              "metacopy=off"},
 470         {OPT_VOLATILE,                  "volatile"},
 471         {OPT_ERR,                       NULL}
 472 };
 473
 474 static char *ovl_next_opt(char **s)
 475 {
 476         char *sbegin = *s;
 477         char *p;
 478
 479         if (sbegin == NULL)
 480                 return NULL;
 481
 482         for (p = sbegin; *p; p++) {
 483                 if (*p == '\\') {
 484                         p++;
 485                         if (!*p)
 486                                 break;
 487                 } else if (*p == ',') {
 488                         *p = '\0';
 489                         *s = p + 1;
 490                         return sbegin;
 491                 }
 492         }
 493         *s = NULL;
 494         return sbegin;
 495 }
 496
 497 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 498 {
 499         if (strcmp(mode, "on") == 0) {
 500                 config->redirect_dir = true;
 501                 /*
 502                  * Does not make sense to have redirect creation without
 503                  * redirect following.
 504                  */
 505                 config->redirect_follow = true;
 506         } else if (strcmp(mode, "follow") == 0) {
 507                 config->redirect_follow = true;
 508         } else if (strcmp(mode, "off") == 0) {
 509                 if (ovl_redirect_always_follow)
 510                         config->redirect_follow = true;
 511         } else if (strcmp(mode, "nofollow") != 0) {
 512                 pr_err("bad mount option \"redirect_dir=%s\"\n",
 513                        mode);
 514                 return -EINVAL;
 515         }
 516
 517         return 0;
 518 }
 519
 520 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 521 {
 522         char *p;
 523         int err;
 524         bool metacopy_opt = false, redirect_opt = false;
 525         bool nfs_export_opt = false, index_opt = false;
 526
 527         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 528         if (!config->redirect_mode)
 529                 return -ENOMEM;
 530
 531         while ((p = ovl_next_opt(&opt)) != NULL) {
 532                 int token;
 533                 substring_t args[MAX_OPT_ARGS];
 534
 535                 if (!*p)
 536                         continue;
 537
 538                 token = match_token(p, ovl_tokens, args);
 539                 switch (token) {
 540                 case OPT_UPPERDIR:
 541                         kfree(config->upperdir);
 542                         config->upperdir = match_strdup(&args[0]);
 543                         if (!config->upperdir)
 544                                 return -ENOMEM;
 545                         break;
 546
 547                 case OPT_LOWERDIR:
 548                         kfree(config->lowerdir);
 549                         config->lowerdir = match_strdup(&args[0]);
 550                         if (!config->lowerdir)
 551                                 return -ENOMEM;
 552                         break;
 553
 554                 case OPT_WORKDIR:
 555                         kfree(config->workdir);
 556                         config->workdir = match_strdup(&args[0]);
 557                         if (!config->workdir)
 558                                 return -ENOMEM;
 559                         break;
 560
 561                 case OPT_DEFAULT_PERMISSIONS:
 562                         config->default_permissions = true;
 563                         break;
 564
 565                 case OPT_REDIRECT_DIR:
 566                         kfree(config->redirect_mode);
 567                         config->redirect_mode = match_strdup(&args[0]);
 568                         if (!config->redirect_mode)
 569                                 return -ENOMEM;
 570                         redirect_opt = true;
 571                         break;
 572
 573                 case OPT_INDEX_ON:
 574                         config->index = true;
 575                         index_opt = true;
 576                         break;
 577
 578                 case OPT_INDEX_OFF:
 579                         config->index = false;
 580                         index_opt = true;
 581                         break;
 582
 583                 case OPT_UUID_ON:
 584                         config->uuid = true;
 585                         break;
 586
 587                 case OPT_UUID_OFF:
 588                         config->uuid = false;
 589                         break;
 590
 591                 case OPT_NFS_EXPORT_ON:
 592                         config->nfs_export = true;
 593                         nfs_export_opt = true;
 594                         break;
 595
 596                 case OPT_NFS_EXPORT_OFF:
 597                         config->nfs_export = false;
 598                         nfs_export_opt = true;
 599                         break;
 600
 601                 case OPT_XINO_ON:
 602                         config->xino = OVL_XINO_ON;
 603                         break;
 604
 605                 case OPT_XINO_OFF:
 606                         config->xino = OVL_XINO_OFF;
 607                         break;
 608
 609                 case OPT_XINO_AUTO:
 610                         config->xino = OVL_XINO_AUTO;
 611                         break;
 612
 613                 case OPT_METACOPY_ON:
 614                         config->metacopy = true;
 615                         metacopy_opt = true;
 616                         break;
 617
 618                 case OPT_METACOPY_OFF:
 619                         config->metacopy = false;
 620                         metacopy_opt = true;
 621                         break;
 622
 623                 case OPT_VOLATILE:
 624                         config->ovl_volatile = true;
 625                         break;
 626
 627                 case OPT_USERXATTR:
 628                         config->userxattr = true;
 629                         break;
 630
 631                 default:
 632                         pr_err("unrecognized mount option \"%s\" or missing value\n",
 633                                         p);
 634                         return -EINVAL;
 635                 }
 636         }
 637
 638         /* Workdir/index are useless in non-upper mount */
 639         if (!config->upperdir) {
 640                 if (config->workdir) {
 641                         pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 642                                 config->workdir);
 643                         kfree(config->workdir);
 644                         config->workdir = NULL;
 645                 }
 646                 if (config->index && index_opt) {
 647                         pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
 648                         index_opt = false;
 649                 }
 650                 config->index = false;
 651         }
 652
 653         if (!config->upperdir && config->ovl_volatile) {
 654                 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
 655                 config->ovl_volatile = false;
 656         }
 657
 658         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 659         if (err)
 660                 return err;
 661
 662         /*
 663          * This is to make the logic below simpler.  It doesn't make any other
 664          * difference, since config->redirect_dir is only used for upper.
 665          */
 666         if (!config->upperdir && config->redirect_follow)
 667                 config->redirect_dir = true;
 668
 669         /* Resolve metacopy -> redirect_dir dependency */
 670         if (config->metacopy && !config->redirect_dir) {
 671                 if (metacopy_opt && redirect_opt) {
 672                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
 673                                config->redirect_mode);
 674                         return -EINVAL;
 675                 }
 676                 if (redirect_opt) {
 677                         /*
 678                          * There was an explicit redirect_dir=... that resulted
 679                          * in this conflict.
 680                          */
 681                         pr_info("disabling metacopy due to redirect_dir=%s\n",
 682                                 config->redirect_mode);
 683                         config->metacopy = false;
 684                 } else {
 685                         /* Automatically enable redirect otherwise. */
 686                         config->redirect_follow = config->redirect_dir = true;
 687                 }
 688         }
 689
 690         /* Resolve nfs_export -> index dependency */
 691         if (config->nfs_export && !config->index) {
 692                 if (!config->upperdir && config->redirect_follow) {
 693                         pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
 694                         config->nfs_export = false;
 695                 } else if (nfs_export_opt && index_opt) {
 696                         pr_err("conflicting options: nfs_export=on,index=off\n");
 697                         return -EINVAL;
 698                 } else if (index_opt) {
 699                         /*
 700                          * There was an explicit index=off that resulted
 701                          * in this conflict.
 702                          */
 703                         pr_info("disabling nfs_export due to index=off\n");
 704                         config->nfs_export = false;
 705                 } else {
 706                         /* Automatically enable index otherwise. */
 707                         config->index = true;
 708                 }
 709         }
 710
 711         /* Resolve nfs_export -> !metacopy dependency */
 712         if (config->nfs_export && config->metacopy) {
 713                 if (nfs_export_opt && metacopy_opt) {
 714                         pr_err("conflicting options: nfs_export=on,metacopy=on\n");
 715                         return -EINVAL;
 716                 }
 717                 if (metacopy_opt) {
 718                         /*
 719                          * There was an explicit metacopy=on that resulted
 720                          * in this conflict.
 721                          */
 722                         pr_info("disabling nfs_export due to metacopy=on\n");
 723                         config->nfs_export = false;
 724                 } else {
 725                         /*
 726                          * There was an explicit nfs_export=on that resulted
 727                          * in this conflict.
 728                          */
 729                         pr_info("disabling metacopy due to nfs_export=on\n");
 730                         config->metacopy = false;
 731                 }
 732         }
 733
 734
 735         /* Resolve userxattr -> !redirect && !metacopy dependency */
 736         if (config->userxattr) {
 737                 if (config->redirect_follow && redirect_opt) {
 738                         pr_err("conflicting options: userxattr,redirect_dir=%s\n",
 739                                config->redirect_mode);
 740                         return -EINVAL;
 741                 }
 742                 if (config->metacopy && metacopy_opt) {
 743                         pr_err("conflicting options: userxattr,metacopy=on\n");
 744                         return -EINVAL;
 745                 }
 746                 /*
 747                  * Silently disable default setting of redirect and metacopy.
 748                  * This shall be the default in the future as well: these
 749                  * options must be explicitly enabled if used together with
 750                  * userxattr.
 751                  */
 752                 config->redirect_dir = config->redirect_follow = false;
 753                 config->metacopy = false;
 754         }
 755
 756         return 0;
 757 }
 758
 759 #define OVL_WORKDIR_NAME "work"
 760 #define OVL_INDEXDIR_NAME "index"
 761
 762 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 763                                          const char *name, bool persist)
 764 {
 765         struct inode *dir =  ofs->workbasedir->d_inode;
 766         struct vfsmount *mnt = ovl_upper_mnt(ofs);
 767         struct dentry *work;
 768         int err;
 769         bool retried = false;
 770
 771         inode_lock_nested(dir, I_MUTEX_PARENT);
 772 retry:
 773         work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));
 774
 775         if (!IS_ERR(work)) {
 776                 struct iattr attr = {
 777                         .ia_valid = ATTR_MODE,
 778                         .ia_mode = S_IFDIR | 0,
 779                 };
 780
 781                 if (work->d_inode) {
 782                         err = -EEXIST;
 783                         if (retried)
 784                                 goto out_dput;
 785
 786                         if (persist)
 787                                 goto out_unlock;
 788
 789                         retried = true;
 790                         err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);
 791                         dput(work);
 792                         if (err == -EINVAL) {
 793                                 work = ERR_PTR(err);
 794                                 goto out_unlock;
 795                         }
 796                         goto retry;
 797                 }
 798
 799                 err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);
 800                 if (err)
 801                         goto out_dput;
 802
 803                 /* Weird filesystem returning with hashed negative (kernfs)? */
 804                 err = -EINVAL;
 805                 if (d_really_is_negative(work))
 806                         goto out_dput;
 807
 808                 /*
 809                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 810                  *
 811                  * a) success (there was a POSIX ACL xattr and was removed)
 812                  * b) -ENODATA (there was no POSIX ACL xattr)
 813                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 814                  *
 815                  * There are various other error values that could effectively
 816                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 817                  * if the xattr name is too long), but the set of filesystems
 818                  * allowed as upper are limited to "normal" ones, where checking
 819                  * for the above two errors is sufficient.
 820                  */
 821                 err = ovl_do_removexattr(ofs, work,
 822                                          XATTR_NAME_POSIX_ACL_DEFAULT);
 823                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 824                         goto out_dput;
 825
 826                 err = ovl_do_removexattr(ofs, work,
 827                                          XATTR_NAME_POSIX_ACL_ACCESS);
 828                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 829                         goto out_dput;
 830
 831                 /* Clear any inherited mode bits */
 832                 inode_lock(work->d_inode);
 833                 err = ovl_do_notify_change(ofs, work, &attr);
 834                 inode_unlock(work->d_inode);
 835                 if (err)
 836                         goto out_dput;
 837         } else {
 838                 err = PTR_ERR(work);
 839                 goto out_err;
 840         }
 841 out_unlock:
 842         inode_unlock(dir);
 843         return work;
 844
 845 out_dput:
 846         dput(work);
 847 out_err:
 848         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
 849                 ofs->config.workdir, name, -err);
 850         work = NULL;
 851         goto out_unlock;
 852 }
 853
 854 static void ovl_unescape(char *s)
 855 {
 856         char *d = s;
 857
 858         for (;; s++, d++) {
 859                 if (*s == '\\')
 860                         s++;
 861                 *d = *s;
 862                 if (!*s)
 863                         break;
 864         }
 865 }
 866
 867 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 868 {
 869         int err = -EINVAL;
 870
 871         if (!*name) {
 872                 pr_err("empty lowerdir\n");
 873                 goto out;
 874         }
 875         err = kern_path(name, LOOKUP_FOLLOW, path);
 876         if (err) {
 877                 pr_err("failed to resolve '%s': %i\n", name, err);
 878                 goto out;
 879         }
 880         err = -EINVAL;
 881         if (ovl_dentry_weird(path->dentry)) {
 882                 pr_err("filesystem on '%s' not supported\n", name);
 883                 goto out_put;
 884         }
 885         if (!d_is_dir(path->dentry)) {
 886                 pr_err("'%s' not a directory\n", name);
 887                 goto out_put;
 888         }
 889         return 0;
 890
 891 out_put:
 892         path_put_init(path);
 893 out:
 894         return err;
 895 }
 896
 897 static int ovl_mount_dir(const char *name, struct path *path)
 898 {
 899         int err = -ENOMEM;
 900         char *tmp = kstrdup(name, GFP_KERNEL);
 901
 902         if (tmp) {
 903                 ovl_unescape(tmp);
 904                 err = ovl_mount_dir_noesc(tmp, path);
 905
 906                 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
 907                         pr_err("filesystem on '%s' not supported as upperdir\n",
 908                                tmp);
 909                         path_put_init(path);
 910                         err = -EINVAL;
 911                 }
 912                 kfree(tmp);
 913         }
 914         return err;
 915 }
 916
 917 static int ovl_check_namelen(const struct path *path, struct ovl_fs *ofs,
 918                              const char *name)
 919 {
 920         struct kstatfs statfs;
 921         int err = vfs_statfs(path, &statfs);
 922
 923         if (err)
 924                 pr_err("statfs failed on '%s'\n", name);
 925         else
 926                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 927
 928         return err;
 929 }
 930
 931 static int ovl_lower_dir(const char *name, struct path *path,
 932                          struct ovl_fs *ofs, int *stack_depth)
 933 {
 934         int fh_type;
 935         int err;
 936
 937         err = ovl_mount_dir_noesc(name, path);
 938         if (err)
 939                 return err;
 940
 941         err = ovl_check_namelen(path, ofs, name);
 942         if (err)
 943                 return err;
 944
 945         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 946
 947         /*
 948          * The inodes index feature and NFS export need to encode and decode
 949          * file handles, so they require that all layers support them.
 950          */
 951         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 952         if ((ofs->config.nfs_export ||
 953              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 954                 ofs->config.index = false;
 955                 ofs->config.nfs_export = false;
 956                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 957                         name);
 958         }
 959         /*
 960          * Decoding origin file handle is required for persistent st_ino.
 961          * Without persistent st_ino, xino=auto falls back to xino=off.
 962          */
 963         if (ofs->config.xino == OVL_XINO_AUTO &&
 964             ofs->config.upperdir && !fh_type) {
 965                 ofs->config.xino = OVL_XINO_OFF;
 966                 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
 967                         name);
 968         }
 969
 970         /* Check if lower fs has 32bit inode numbers */
 971         if (fh_type != FILEID_INO32_GEN)
 972                 ofs->xino_mode = -1;
 973
 974         return 0;
 975 }
 976
 977 /* Workdir should not be subdir of upperdir and vice versa */
 978 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 979 {
 980         bool ok = false;
 981
 982         if (workdir != upperdir) {
 983                 ok = (lock_rename(workdir, upperdir) == NULL);
 984                 unlock_rename(workdir, upperdir);
 985         }
 986         return ok;
 987 }
 988
 989 static unsigned int ovl_split_lowerdirs(char *str)
 990 {
 991         unsigned int ctr = 1;
 992         char *s, *d;
 993
 994         for (s = d = str;; s++, d++) {
 995                 if (*s == '\\') {
 996                         s++;
 997                 } else if (*s == ':') {
 998                         *d = '\0';
 999                         ctr++;
1000                         continue;
1001                 }
1002                 *d = *s;
1003                 if (!*s)
1004                         break;
1005         }
1006         return ctr;
1007 }
1008
1009 static int __maybe_unused
1010 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
1011                         struct dentry *dentry, struct inode *inode,
1012                         const char *name, void *buffer, size_t size)
1013 {
1014         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
1015 }
1016
1017 static int __maybe_unused
1018 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
1019                         struct user_namespace *mnt_userns,
1020                         struct dentry *dentry, struct inode *inode,
1021                         const char *name, const void *value,
1022                         size_t size, int flags)
1023 {
1024         struct dentry *workdir = ovl_workdir(dentry);
1025         struct inode *realinode = ovl_inode_real(inode);
1026         struct posix_acl *acl = NULL;
1027         int err;
1028
1029         /* Check that everything is OK before copy-up */
1030         if (value) {
1031                 /* The above comment can be understood in two ways:
1032                  *
1033                  * 1. We just want to check whether the basic POSIX ACL format
1034                  *    is ok. For example, if the header is correct and the size
1035                  *    is sane.
1036                  * 2. We want to know whether the ACL_{GROUP,USER} entries can
1037                  *    be mapped according to the underlying filesystem.
1038                  *
1039                  * Currently, we only check 1. If we wanted to check 2. we
1040                  * would need to pass the mnt_userns and the fs_userns of the
1041                  * underlying filesystem. But frankly, I think checking 1. is
1042                  * enough to start the copy-up.
1043                  */
1044                 acl = vfs_set_acl_prepare(&init_user_ns, &init_user_ns, value, size);
1045                 if (IS_ERR(acl))
1046                         return PTR_ERR(acl);
1047         }
1048         err = -EOPNOTSUPP;
1049         if (!IS_POSIXACL(d_inode(workdir)))
1050                 goto out_acl_release;
1051         if (!realinode->i_op->set_acl)
1052                 goto out_acl_release;
1053         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1054                 err = acl ? -EACCES : 0;
1055                 goto out_acl_release;
1056         }
1057         err = -EPERM;
1058         if (!inode_owner_or_capable(&init_user_ns, inode))
1059                 goto out_acl_release;
1060
1061         posix_acl_release(acl);
1062
1063         /*
1064          * Check if sgid bit needs to be cleared (actual setacl operation will
1065          * be done with mounter's capabilities and so that won't do it for us).
1066          */
1067         if (unlikely(inode->i_mode & S_ISGID) &&
1068             handler->flags == ACL_TYPE_ACCESS &&
1069             !in_group_p(inode->i_gid) &&
1070             !capable_wrt_inode_uidgid(&init_user_ns, inode, CAP_FSETID)) {
1071                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
1072
1073                 err = ovl_setattr(&init_user_ns, dentry, &iattr);
1074                 if (err)
1075                         return err;
1076         }
1077
1078         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
1079         return err;
1080
1081 out_acl_release:
1082         posix_acl_release(acl);
1083         return err;
1084 }
1085
1086 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1087                              struct dentry *dentry, struct inode *inode,
1088                              const char *name, void *buffer, size_t size)
1089 {
1090         return -EOPNOTSUPP;
1091 }
1092
1093 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1094                              struct user_namespace *mnt_userns,
1095                              struct dentry *dentry, struct inode *inode,
1096                              const char *name, const void *value,
1097                              size_t size, int flags)
1098 {
1099         return -EOPNOTSUPP;
1100 }
1101
1102 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1103                                struct dentry *dentry, struct inode *inode,
1104                                const char *name, void *buffer, size_t size)
1105 {
1106         return ovl_xattr_get(dentry, inode, name, buffer, size);
1107 }
1108
1109 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1110                                struct user_namespace *mnt_userns,
1111                                struct dentry *dentry, struct inode *inode,
1112                                const char *name, const void *value,
1113                                size_t size, int flags)
1114 {
1115         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1116 }
1117
1118 static const struct xattr_handler __maybe_unused
1119 ovl_posix_acl_access_xattr_handler = {
1120         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1121         .flags = ACL_TYPE_ACCESS,
1122         .get = ovl_posix_acl_xattr_get,
1123         .set = ovl_posix_acl_xattr_set,
1124 };
1125
1126 static const struct xattr_handler __maybe_unused
1127 ovl_posix_acl_default_xattr_handler = {
1128         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1129         .flags = ACL_TYPE_DEFAULT,
1130         .get = ovl_posix_acl_xattr_get,
1131         .set = ovl_posix_acl_xattr_set,
1132 };
1133
1134 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1135         .prefix = OVL_XATTR_TRUSTED_PREFIX,
1136         .get = ovl_own_xattr_get,
1137         .set = ovl_own_xattr_set,
1138 };
1139
1140 static const struct xattr_handler ovl_own_user_xattr_handler = {
1141         .prefix = OVL_XATTR_USER_PREFIX,
1142         .get = ovl_own_xattr_get,
1143         .set = ovl_own_xattr_set,
1144 };
1145
1146 static const struct xattr_handler ovl_other_xattr_handler = {
1147         .prefix = "", /* catch all */
1148         .get = ovl_other_xattr_get,
1149         .set = ovl_other_xattr_set,
1150 };
1151
1152 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1153 #ifdef CONFIG_FS_POSIX_ACL
1154         &ovl_posix_acl_access_xattr_handler,
1155         &ovl_posix_acl_default_xattr_handler,
1156 #endif
1157         &ovl_own_trusted_xattr_handler,
1158         &ovl_other_xattr_handler,
1159         NULL
1160 };
1161
1162 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1163 #ifdef CONFIG_FS_POSIX_ACL
1164         &ovl_posix_acl_access_xattr_handler,
1165         &ovl_posix_acl_default_xattr_handler,
1166 #endif
1167         &ovl_own_user_xattr_handler,
1168         &ovl_other_xattr_handler,
1169         NULL
1170 };
1171
1172 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1173                           struct inode **ptrap, const char *name)
1174 {
1175         struct inode *trap;
1176         int err;
1177
1178         trap = ovl_get_trap_inode(sb, dir);
1179         err = PTR_ERR_OR_ZERO(trap);
1180         if (err) {
1181                 if (err == -ELOOP)
1182                         pr_err("conflicting %s path\n", name);
1183                 return err;
1184         }
1185
1186         *ptrap = trap;
1187         return 0;
1188 }
1189
1190 /*
1191  * Determine how we treat concurrent use of upperdir/workdir based on the
1192  * index feature. This is papering over mount leaks of container runtimes,
1193  * for example, an old overlay mount is leaked and now its upperdir is
1194  * attempted to be used as a lower layer in a new overlay mount.
1195  */
1196 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1197 {
1198         if (ofs->config.index) {
1199                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1200                        name);
1201                 return -EBUSY;
1202         } else {
1203                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1204                         name);
1205                 return 0;
1206         }
1207 }
1208
1209 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1210                          struct ovl_layer *upper_layer, struct path *upperpath)
1211 {
1212         struct vfsmount *upper_mnt;
1213         int err;
1214
1215         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1216         if (err)
1217                 goto out;
1218
1219         /* Upperdir path should not be r/o */
1220         if (__mnt_is_readonly(upperpath->mnt)) {
1221                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1222                 err = -EINVAL;
1223                 goto out;
1224         }
1225
1226         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1227         if (err)
1228                 goto out;
1229
1230         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1231                              "upperdir");
1232         if (err)
1233                 goto out;
1234
1235         upper_mnt = clone_private_mount(upperpath);
1236         err = PTR_ERR(upper_mnt);
1237         if (IS_ERR(upper_mnt)) {
1238                 pr_err("failed to clone upperpath\n");
1239                 goto out;
1240         }
1241
1242         /* Don't inherit atime flags */
1243         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1244         upper_layer->mnt = upper_mnt;
1245         upper_layer->idx = 0;
1246         upper_layer->fsid = 0;
1247
1248         /*
1249          * Inherit SB_NOSEC flag from upperdir.
1250          *
1251          * This optimization changes behavior when a security related attribute
1252          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1253          * okay because we don't yet have guarantees in that case, but it will
1254          * need careful treatment once we want to honour changes to underlying
1255          * filesystems.
1256          */
1257         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1258                 sb->s_flags |= SB_NOSEC;
1259
1260         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1261                 ofs->upperdir_locked = true;
1262         } else {
1263                 err = ovl_report_in_use(ofs, "upperdir");
1264                 if (err)
1265                         goto out;
1266         }
1267
1268         err = 0;
1269 out:
1270         return err;
1271 }
1272
1273 /*
1274  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1275  * negative values if error is encountered.
1276  */
1277 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
1278 {
1279         struct dentry *workdir = ofs->workdir;
1280         struct inode *dir = d_inode(workdir);
1281         struct dentry *temp;
1282         struct dentry *dest;
1283         struct dentry *whiteout;
1284         struct name_snapshot name;
1285         int err;
1286
1287         inode_lock_nested(dir, I_MUTEX_PARENT);
1288
1289         temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
1290         err = PTR_ERR(temp);
1291         if (IS_ERR(temp))
1292                 goto out_unlock;
1293
1294         dest = ovl_lookup_temp(ofs, workdir);
1295         err = PTR_ERR(dest);
1296         if (IS_ERR(dest)) {
1297                 dput(temp);
1298                 goto out_unlock;
1299         }
1300
1301         /* Name is inline and stable - using snapshot as a copy helper */
1302         take_dentry_name_snapshot(&name, temp);
1303         err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
1304         if (err) {
1305                 if (err == -EINVAL)
1306                         err = 0;
1307                 goto cleanup_temp;
1308         }
1309
1310         whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
1311         err = PTR_ERR(whiteout);
1312         if (IS_ERR(whiteout))
1313                 goto cleanup_temp;
1314
1315         err = ovl_is_whiteout(whiteout);
1316
1317         /* Best effort cleanup of whiteout and temp file */
1318         if (err)
1319                 ovl_cleanup(ofs, dir, whiteout);
1320         dput(whiteout);
1321
1322 cleanup_temp:
1323         ovl_cleanup(ofs, dir, temp);
1324         release_dentry_name_snapshot(&name);
1325         dput(temp);
1326         dput(dest);
1327
1328 out_unlock:
1329         inode_unlock(dir);
1330
1331         return err;
1332 }
1333
1334 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
1335                                            struct dentry *parent,
1336                                            const char *name, umode_t mode)
1337 {
1338         size_t len = strlen(name);
1339         struct dentry *child;
1340
1341         inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1342         child = ovl_lookup_upper(ofs, name, parent, len);
1343         if (!IS_ERR(child) && !child->d_inode)
1344                 child = ovl_create_real(ofs, parent->d_inode, child,
1345                                         OVL_CATTR(mode));
1346         inode_unlock(parent->d_inode);
1347         dput(parent);
1348
1349         return child;
1350 }
1351
1352 /*
1353  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1354  * present.
1355  */
1356 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1357 {
1358         unsigned int ctr;
1359         struct dentry *d = dget(ofs->workbasedir);
1360         static const char *const volatile_path[] = {
1361                 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1362         };
1363         const char *const *name = volatile_path;
1364
1365         for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1366                 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1367                 if (IS_ERR(d))
1368                         return PTR_ERR(d);
1369         }
1370         dput(d);
1371         return 0;
1372 }
1373
1374 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1375                             const struct path *workpath)
1376 {
1377         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1378         struct dentry *workdir;
1379         struct file *tmpfile;
1380         bool rename_whiteout;
1381         bool d_type;
1382         int fh_type;
1383         int err;
1384
1385         err = mnt_want_write(mnt);
1386         if (err)
1387                 return err;
1388
1389         workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1390         err = PTR_ERR(workdir);
1391         if (IS_ERR_OR_NULL(workdir))
1392                 goto out;
1393
1394         ofs->workdir = workdir;
1395
1396         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1397         if (err)
1398                 goto out;
1399
1400         /*
1401          * Upper should support d_type, else whiteouts are visible.  Given
1402          * workdir and upper are on same fs, we can do iterate_dir() on
1403          * workdir. This check requires successful creation of workdir in
1404          * previous step.
1405          */
1406         err = ovl_check_d_type_supported(workpath);
1407         if (err < 0)
1408                 goto out;
1409
1410         d_type = err;
1411         if (!d_type)
1412                 pr_warn("upper fs needs to support d_type.\n");
1413
1414         /* Check if upper/work fs supports O_TMPFILE */
1415         tmpfile = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
1416         ofs->tmpfile = !IS_ERR(tmpfile);
1417         if (ofs->tmpfile)
1418                 fput(tmpfile);
1419         else
1420                 pr_warn("upper fs does not support tmpfile.\n");
1421
1422
1423         /* Check if upper/work fs supports RENAME_WHITEOUT */
1424         err = ovl_check_rename_whiteout(ofs);
1425         if (err < 0)
1426                 goto out;
1427
1428         rename_whiteout = err;
1429         if (!rename_whiteout)
1430                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1431
1432         /*
1433          * Check if upper/work fs supports (trusted|user).overlay.* xattr
1434          */
1435         err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1436         if (err) {
1437                 pr_warn("failed to set xattr on upper\n");
1438                 ofs->noxattr = true;
1439                 if (ofs->config.index || ofs->config.metacopy) {
1440                         ofs->config.index = false;
1441                         ofs->config.metacopy = false;
1442                         pr_warn("...falling back to index=off,metacopy=off.\n");
1443                 }
1444                 /*
1445                  * xattr support is required for persistent st_ino.
1446                  * Without persistent st_ino, xino=auto falls back to xino=off.
1447                  */
1448                 if (ofs->config.xino == OVL_XINO_AUTO) {
1449                         ofs->config.xino = OVL_XINO_OFF;
1450                         pr_warn("...falling back to xino=off.\n");
1451                 }
1452                 if (err == -EPERM && !ofs->config.userxattr)
1453                         pr_info("try mounting with 'userxattr' option\n");
1454                 err = 0;
1455         } else {
1456                 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1457         }
1458
1459         /*
1460          * We allowed sub-optimal upper fs configuration and don't want to break
1461          * users over kernel upgrade, but we never allowed remote upper fs, so
1462          * we can enforce strict requirements for remote upper fs.
1463          */
1464         if (ovl_dentry_remote(ofs->workdir) &&
1465             (!d_type || !rename_whiteout || ofs->noxattr)) {
1466                 pr_err("upper fs missing required features.\n");
1467                 err = -EINVAL;
1468                 goto out;
1469         }
1470
1471         /*
1472          * For volatile mount, create a incompat/volatile/dirty file to keep
1473          * track of it.
1474          */
1475         if (ofs->config.ovl_volatile) {
1476                 err = ovl_create_volatile_dirty(ofs);
1477                 if (err < 0) {
1478                         pr_err("Failed to create volatile/dirty file.\n");
1479                         goto out;
1480                 }
1481         }
1482
1483         /* Check if upper/work fs supports file handles */
1484         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1485         if (ofs->config.index && !fh_type) {
1486                 ofs->config.index = false;
1487                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1488         }
1489
1490         /* Check if upper fs has 32bit inode numbers */
1491         if (fh_type != FILEID_INO32_GEN)
1492                 ofs->xino_mode = -1;
1493
1494         /* NFS export of r/w mount depends on index */
1495         if (ofs->config.nfs_export && !ofs->config.index) {
1496                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1497                 ofs->config.nfs_export = false;
1498         }
1499 out:
1500         mnt_drop_write(mnt);
1501         return err;
1502 }
1503
1504 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1505                            const struct path *upperpath)
1506 {
1507         int err;
1508         struct path workpath = { };
1509
1510         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1511         if (err)
1512                 goto out;
1513
1514         err = -EINVAL;
1515         if (upperpath->mnt != workpath.mnt) {
1516                 pr_err("workdir and upperdir must reside under the same mount\n");
1517                 goto out;
1518         }
1519         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1520                 pr_err("workdir and upperdir must be separate subtrees\n");
1521                 goto out;
1522         }
1523
1524         ofs->workbasedir = dget(workpath.dentry);
1525
1526         if (ovl_inuse_trylock(ofs->workbasedir)) {
1527                 ofs->workdir_locked = true;
1528         } else {
1529                 err = ovl_report_in_use(ofs, "workdir");
1530                 if (err)
1531                         goto out;
1532         }
1533
1534         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1535                              "workdir");
1536         if (err)
1537                 goto out;
1538
1539         err = ovl_make_workdir(sb, ofs, &workpath);
1540
1541 out:
1542         path_put(&workpath);
1543
1544         return err;
1545 }
1546
1547 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1548                             struct ovl_entry *oe, const struct path *upperpath)
1549 {
1550         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1551         struct dentry *indexdir;
1552         int err;
1553
1554         err = mnt_want_write(mnt);
1555         if (err)
1556                 return err;
1557
1558         /* Verify lower root is upper root origin */
1559         err = ovl_verify_origin(ofs, upperpath->dentry,
1560                                 oe->lowerstack[0].dentry, true);
1561         if (err) {
1562                 pr_err("failed to verify upper root origin\n");
1563                 goto out;
1564         }
1565
1566         /* index dir will act also as workdir */
1567         iput(ofs->workdir_trap);
1568         ofs->workdir_trap = NULL;
1569         dput(ofs->workdir);
1570         ofs->workdir = NULL;
1571         indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1572         if (IS_ERR(indexdir)) {
1573                 err = PTR_ERR(indexdir);
1574         } else if (indexdir) {
1575                 ofs->indexdir = indexdir;
1576                 ofs->workdir = dget(indexdir);
1577
1578                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1579                                      "indexdir");
1580                 if (err)
1581                         goto out;
1582
1583                 /*
1584                  * Verify upper root is exclusively associated with index dir.
1585                  * Older kernels stored upper fh in ".overlay.origin"
1586                  * xattr. If that xattr exists, verify that it is a match to
1587                  * upper dir file handle. In any case, verify or set xattr
1588                  * ".overlay.upper" to indicate that index may have
1589                  * directory entries.
1590                  */
1591                 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1592                         err = ovl_verify_set_fh(ofs, ofs->indexdir,
1593                                                 OVL_XATTR_ORIGIN,
1594                                                 upperpath->dentry, true, false);
1595                         if (err)
1596                                 pr_err("failed to verify index dir 'origin' xattr\n");
1597                 }
1598                 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1599                                        true);
1600                 if (err)
1601                         pr_err("failed to verify index dir 'upper' xattr\n");
1602
1603                 /* Cleanup bad/stale/orphan index entries */
1604                 if (!err)
1605                         err = ovl_indexdir_cleanup(ofs);
1606         }
1607         if (err || !ofs->indexdir)
1608                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1609
1610 out:
1611         mnt_drop_write(mnt);
1612         return err;
1613 }
1614
1615 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1616 {
1617         unsigned int i;
1618
1619         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1620                 return true;
1621
1622         /*
1623          * We allow using single lower with null uuid for index and nfs_export
1624          * for example to support those features with single lower squashfs.
1625          * To avoid regressions in setups of overlay with re-formatted lower
1626          * squashfs, do not allow decoding origin with lower null uuid unless
1627          * user opted-in to one of the new features that require following the
1628          * lower inode of non-dir upper.
1629          */
1630         if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
1631                 return false;
1632
1633         for (i = 0; i < ofs->numfs; i++) {
1634                 /*
1635                  * We use uuid to associate an overlay lower file handle with a
1636                  * lower layer, so we can accept lower fs with null uuid as long
1637                  * as all lower layers with null uuid are on the same fs.
1638                  * if we detect multiple lower fs with the same uuid, we
1639                  * disable lower file handle decoding on all of them.
1640                  */
1641                 if (ofs->fs[i].is_lower &&
1642                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1643                         ofs->fs[i].bad_uuid = true;
1644                         return false;
1645                 }
1646         }
1647         return true;
1648 }
1649
1650 /* Get a unique fsid for the layer */
1651 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1652 {
1653         struct super_block *sb = path->mnt->mnt_sb;
1654         unsigned int i;
1655         dev_t dev;
1656         int err;
1657         bool bad_uuid = false;
1658         bool warn = false;
1659
1660         for (i = 0; i < ofs->numfs; i++) {
1661                 if (ofs->fs[i].sb == sb)
1662                         return i;
1663         }
1664
1665         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1666                 bad_uuid = true;
1667                 if (ofs->config.xino == OVL_XINO_AUTO) {
1668                         ofs->config.xino = OVL_XINO_OFF;
1669                         warn = true;
1670                 }
1671                 if (ofs->config.index || ofs->config.nfs_export) {
1672                         ofs->config.index = false;
1673                         ofs->config.nfs_export = false;
1674                         warn = true;
1675                 }
1676                 if (warn) {
1677                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1678                                 uuid_is_null(&sb->s_uuid) ? "null" :
1679                                                             "conflicting",
1680                                 path->dentry, ovl_xino_str[ofs->config.xino]);
1681                 }
1682         }
1683
1684         err = get_anon_bdev(&dev);
1685         if (err) {
1686                 pr_err("failed to get anonymous bdev for lowerpath\n");
1687                 return err;
1688         }
1689
1690         ofs->fs[ofs->numfs].sb = sb;
1691         ofs->fs[ofs->numfs].pseudo_dev = dev;
1692         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1693
1694         return ofs->numfs++;
1695 }
1696
1697 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1698                           struct path *stack, unsigned int numlower,
1699                           struct ovl_layer *layers)
1700 {
1701         int err;
1702         unsigned int i;
1703
1704         err = -ENOMEM;
1705         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1706         if (ofs->fs == NULL)
1707                 goto out;
1708
1709         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1710         ofs->numfs++;
1711
1712         /*
1713          * All lower layers that share the same fs as upper layer, use the same
1714          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1715          * only overlay to simplify ovl_fs_free().
1716          * is_lower will be set if upper fs is shared with a lower layer.
1717          */
1718         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1719         if (err) {
1720                 pr_err("failed to get anonymous bdev for upper fs\n");
1721                 goto out;
1722         }
1723
1724         if (ovl_upper_mnt(ofs)) {
1725                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1726                 ofs->fs[0].is_lower = false;
1727         }
1728
1729         for (i = 0; i < numlower; i++) {
1730                 struct vfsmount *mnt;
1731                 struct inode *trap;
1732                 int fsid;
1733
1734                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1735                 if (err < 0)
1736                         goto out;
1737
1738                 /*
1739                  * Check if lower root conflicts with this overlay layers before
1740                  * checking if it is in-use as upperdir/workdir of "another"
1741                  * mount, because we do not bother to check in ovl_is_inuse() if
1742                  * the upperdir/workdir is in fact in-use by our
1743                  * upperdir/workdir.
1744                  */
1745                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1746                 if (err)
1747                         goto out;
1748
1749                 if (ovl_is_inuse(stack[i].dentry)) {
1750                         err = ovl_report_in_use(ofs, "lowerdir");
1751                         if (err) {
1752                                 iput(trap);
1753                                 goto out;
1754                         }
1755                 }
1756
1757                 mnt = clone_private_mount(&stack[i]);
1758                 err = PTR_ERR(mnt);
1759                 if (IS_ERR(mnt)) {
1760                         pr_err("failed to clone lowerpath\n");
1761                         iput(trap);
1762                         goto out;
1763                 }
1764
1765                 /*
1766                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1767                  * will fail instead of modifying lower fs.
1768                  */
1769                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1770
1771                 layers[ofs->numlayer].trap = trap;
1772                 layers[ofs->numlayer].mnt = mnt;
1773                 layers[ofs->numlayer].idx = ofs->numlayer;
1774                 layers[ofs->numlayer].fsid = fsid;
1775                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1776                 ofs->numlayer++;
1777                 ofs->fs[fsid].is_lower = true;
1778         }
1779
1780         /*
1781          * When all layers on same fs, overlay can use real inode numbers.
1782          * With mount option "xino=<on|auto>", mounter declares that there are
1783          * enough free high bits in underlying fs to hold the unique fsid.
1784          * If overlayfs does encounter underlying inodes using the high xino
1785          * bits reserved for fsid, it emits a warning and uses the original
1786          * inode number or a non persistent inode number allocated from a
1787          * dedicated range.
1788          */
1789         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1790                 if (ofs->config.xino == OVL_XINO_ON)
1791                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1792                 ofs->xino_mode = 0;
1793         } else if (ofs->config.xino == OVL_XINO_OFF) {
1794                 ofs->xino_mode = -1;
1795         } else if (ofs->xino_mode < 0) {
1796                 /*
1797                  * This is a roundup of number of bits needed for encoding
1798                  * fsid, where fsid 0 is reserved for upper fs (even with
1799                  * lower only overlay) +1 extra bit is reserved for the non
1800                  * persistent inode number range that is used for resolving
1801                  * xino lower bits overflow.
1802                  */
1803                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1804                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1805         }
1806
1807         if (ofs->xino_mode > 0) {
1808                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1809                         ofs->xino_mode);
1810         }
1811
1812         err = 0;
1813 out:
1814         return err;
1815 }
1816
1817 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1818                                 const char *lower, unsigned int numlower,
1819                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1820 {
1821         int err;
1822         struct path *stack = NULL;
1823         unsigned int i;
1824         struct ovl_entry *oe;
1825
1826         if (!ofs->config.upperdir && numlower == 1) {
1827                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1828                 return ERR_PTR(-EINVAL);
1829         }
1830
1831         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1832         if (!stack)
1833                 return ERR_PTR(-ENOMEM);
1834
1835         err = -EINVAL;
1836         for (i = 0; i < numlower; i++) {
1837                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1838                 if (err)
1839                         goto out_err;
1840
1841                 lower = strchr(lower, '\0') + 1;
1842         }
1843
1844         err = -EINVAL;
1845         sb->s_stack_depth++;
1846         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1847                 pr_err("maximum fs stacking depth exceeded\n");
1848                 goto out_err;
1849         }
1850
1851         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1852         if (err)
1853                 goto out_err;
1854
1855         err = -ENOMEM;
1856         oe = ovl_alloc_entry(numlower);
1857         if (!oe)
1858                 goto out_err;
1859
1860         for (i = 0; i < numlower; i++) {
1861                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1862                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1863         }
1864
1865 out:
1866         for (i = 0; i < numlower; i++)
1867                 path_put(&stack[i]);
1868         kfree(stack);
1869
1870         return oe;
1871
1872 out_err:
1873         oe = ERR_PTR(err);
1874         goto out;
1875 }
1876
1877 /*
1878  * Check if this layer root is a descendant of:
1879  * - another layer of this overlayfs instance
1880  * - upper/work dir of any overlayfs instance
1881  */
1882 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1883                            struct dentry *dentry, const char *name,
1884                            bool is_lower)
1885 {
1886         struct dentry *next = dentry, *parent;
1887         int err = 0;
1888
1889         if (!dentry)
1890                 return 0;
1891
1892         parent = dget_parent(next);
1893
1894         /* Walk back ancestors to root (inclusive) looking for traps */
1895         while (!err && parent != next) {
1896                 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1897                         err = -ELOOP;
1898                         pr_err("overlapping %s path\n", name);
1899                 } else if (ovl_is_inuse(parent)) {
1900                         err = ovl_report_in_use(ofs, name);
1901                 }
1902                 next = parent;
1903                 parent = dget_parent(next);
1904                 dput(next);
1905         }
1906
1907         dput(parent);
1908
1909         return err;
1910 }
1911
1912 /*
1913  * Check if any of the layers or work dirs overlap.
1914  */
1915 static int ovl_check_overlapping_layers(struct super_block *sb,
1916                                         struct ovl_fs *ofs)
1917 {
1918         int i, err;
1919
1920         if (ovl_upper_mnt(ofs)) {
1921                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1922                                       "upperdir", false);
1923                 if (err)
1924                         return err;
1925
1926                 /*
1927                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1928                  * this instance and covers overlapping work and index dirs,
1929                  * unless work or index dir have been moved since created inside
1930                  * workbasedir.  In that case, we already have their traps in
1931                  * inode cache and we will catch that case on lookup.
1932                  */
1933                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1934                                       false);
1935                 if (err)
1936                         return err;
1937         }
1938
1939         for (i = 1; i < ofs->numlayer; i++) {
1940                 err = ovl_check_layer(sb, ofs,
1941                                       ofs->layers[i].mnt->mnt_root,
1942                                       "lowerdir", true);
1943                 if (err)
1944                         return err;
1945         }
1946
1947         return 0;
1948 }
1949
1950 static struct dentry *ovl_get_root(struct super_block *sb,
1951                                    struct dentry *upperdentry,
1952                                    struct ovl_entry *oe)
1953 {
1954         struct dentry *root;
1955         struct ovl_path *lowerpath = &oe->lowerstack[0];
1956         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1957         int fsid = lowerpath->layer->fsid;
1958         struct ovl_inode_params oip = {
1959                 .upperdentry = upperdentry,
1960                 .lowerpath = lowerpath,
1961         };
1962
1963         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1964         if (!root)
1965                 return NULL;
1966
1967         root->d_fsdata = oe;
1968
1969         if (upperdentry) {
1970                 /* Root inode uses upper st_ino/i_ino */
1971                 ino = d_inode(upperdentry)->i_ino;
1972                 fsid = 0;
1973                 ovl_dentry_set_upper_alias(root);
1974                 if (ovl_is_impuredir(sb, upperdentry))
1975                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1976         }
1977
1978         /* Root is always merge -> can have whiteouts */
1979         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1980         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1981         ovl_set_upperdata(d_inode(root));
1982         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1983         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1984
1985         return root;
1986 }
1987
1988 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1989 {
1990         struct path upperpath = { };
1991         struct dentry *root_dentry;
1992         struct ovl_entry *oe;
1993         struct ovl_fs *ofs;
1994         struct ovl_layer *layers;
1995         struct cred *cred;
1996         char *splitlower = NULL;
1997         unsigned int numlower;
1998         int err;
1999
2000         err = -EIO;
2001         if (WARN_ON(sb->s_user_ns != current_user_ns()))
2002                 goto out;
2003
2004         sb->s_d_op = &ovl_dentry_operations;
2005
2006         err = -ENOMEM;
2007         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
2008         if (!ofs)
2009                 goto out;
2010
2011         err = -ENOMEM;
2012         ofs->creator_cred = cred = prepare_creds();
2013         if (!cred)
2014                 goto out_err;
2015
2016         /* Is there a reason anyone would want not to share whiteouts? */
2017         ofs->share_whiteout = true;
2018
2019         ofs->config.index = ovl_index_def;
2020         ofs->config.uuid = true;
2021         ofs->config.nfs_export = ovl_nfs_export_def;
2022         ofs->config.xino = ovl_xino_def();
2023         ofs->config.metacopy = ovl_metacopy_def;
2024         err = ovl_parse_opt((char *) data, &ofs->config);
2025         if (err)
2026                 goto out_err;
2027
2028         err = -EINVAL;
2029         if (!ofs->config.lowerdir) {
2030                 if (!silent)
2031                         pr_err("missing 'lowerdir'\n");
2032                 goto out_err;
2033         }
2034
2035         err = -ENOMEM;
2036         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
2037         if (!splitlower)
2038                 goto out_err;
2039
2040         err = -EINVAL;
2041         numlower = ovl_split_lowerdirs(splitlower);
2042         if (numlower > OVL_MAX_STACK) {
2043                 pr_err("too many lower directories, limit is %d\n",
2044                        OVL_MAX_STACK);
2045                 goto out_err;
2046         }
2047
2048         err = -ENOMEM;
2049         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
2050         if (!layers)
2051                 goto out_err;
2052
2053         ofs->layers = layers;
2054         /* Layer 0 is reserved for upper even if there's no upper */
2055         ofs->numlayer = 1;
2056
2057         sb->s_stack_depth = 0;
2058         sb->s_maxbytes = MAX_LFS_FILESIZE;
2059         atomic_long_set(&ofs->last_ino, 1);
2060         /* Assume underlying fs uses 32bit inodes unless proven otherwise */
2061         if (ofs->config.xino != OVL_XINO_OFF) {
2062                 ofs->xino_mode = BITS_PER_LONG - 32;
2063                 if (!ofs->xino_mode) {
2064                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
2065                         ofs->config.xino = OVL_XINO_OFF;
2066                 }
2067         }
2068
2069         /* alloc/destroy_inode needed for setting up traps in inode cache */
2070         sb->s_op = &ovl_super_operations;
2071
2072         if (ofs->config.upperdir) {
2073                 struct super_block *upper_sb;
2074
2075                 err = -EINVAL;
2076                 if (!ofs->config.workdir) {
2077                         pr_err("missing 'workdir'\n");
2078                         goto out_err;
2079                 }
2080
2081                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
2082                 if (err)
2083                         goto out_err;
2084
2085                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
2086                 if (!ovl_should_sync(ofs)) {
2087                         ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
2088                         if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
2089                                 err = -EIO;
2090                                 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
2091                                 goto out_err;
2092                         }
2093                 }
2094
2095                 err = ovl_get_workdir(sb, ofs, &upperpath);
2096                 if (err)
2097                         goto out_err;
2098
2099                 if (!ofs->workdir)
2100                         sb->s_flags |= SB_RDONLY;
2101
2102                 sb->s_stack_depth = upper_sb->s_stack_depth;
2103                 sb->s_time_gran = upper_sb->s_time_gran;
2104         }
2105         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2106         err = PTR_ERR(oe);
2107         if (IS_ERR(oe))
2108                 goto out_err;
2109
2110         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2111         if (!ovl_upper_mnt(ofs))
2112                 sb->s_flags |= SB_RDONLY;
2113
2114         if (!ofs->config.uuid && ofs->numfs > 1) {
2115                 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2116                 ofs->config.uuid = true;
2117         }
2118
2119         if (!ovl_force_readonly(ofs) && ofs->config.index) {
2120                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2121                 if (err)
2122                         goto out_free_oe;
2123
2124                 /* Force r/o mount with no index dir */
2125                 if (!ofs->indexdir)
2126                         sb->s_flags |= SB_RDONLY;
2127         }
2128
2129         err = ovl_check_overlapping_layers(sb, ofs);
2130         if (err)
2131                 goto out_free_oe;
2132
2133         /* Show index=off in /proc/mounts for forced r/o mount */
2134         if (!ofs->indexdir) {
2135                 ofs->config.index = false;
2136                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2137                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2138                         ofs->config.nfs_export = false;
2139                 }
2140         }
2141
2142         if (ofs->config.metacopy && ofs->config.nfs_export) {
2143                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2144                 ofs->config.nfs_export = false;
2145         }
2146
2147         if (ofs->config.nfs_export)
2148                 sb->s_export_op = &ovl_export_operations;
2149
2150         /* Never override disk quota limits or use reserved space */
2151         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2152
2153         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2154         sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2155                 ovl_trusted_xattr_handlers;
2156         sb->s_fs_info = ofs;
2157         sb->s_flags |= SB_POSIXACL;
2158         sb->s_iflags |= SB_I_SKIP_SYNC;
2159
2160         err = -ENOMEM;
2161         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2162         if (!root_dentry)
2163                 goto out_free_oe;
2164
2165         mntput(upperpath.mnt);
2166         kfree(splitlower);
2167
2168         sb->s_root = root_dentry;
2169
2170         return 0;
2171
2172 out_free_oe:
2173         ovl_entry_stack_free(oe);
2174         kfree(oe);
2175 out_err:
2176         kfree(splitlower);
2177         path_put(&upperpath);
2178         ovl_free_fs(ofs);
2179 out:
2180         return err;
2181 }
2182
2183 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2184                                 const char *dev_name, void *raw_data)
2185 {
2186         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2187 }
2188
2189 static struct file_system_type ovl_fs_type = {
2190         .owner          = THIS_MODULE,
2191         .name           = "overlay",
2192         .fs_flags       = FS_USERNS_MOUNT,
2193         .mount          = ovl_mount,
2194         .kill_sb        = kill_anon_super,
2195 };
2196 MODULE_ALIAS_FS("overlay");
2197
2198 static void ovl_inode_init_once(void *foo)
2199 {
2200         struct ovl_inode *oi = foo;
2201
2202         inode_init_once(&oi->vfs_inode);
2203 }
2204
2205 static int __init ovl_init(void)
2206 {
2207         int err;
2208
2209         ovl_inode_cachep = kmem_cache_create("ovl_inode",
2210                                              sizeof(struct ovl_inode), 0,
2211                                              (SLAB_RECLAIM_ACCOUNT|
2212                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2213                                              ovl_inode_init_once);
2214         if (ovl_inode_cachep == NULL)
2215                 return -ENOMEM;
2216
2217         err = ovl_aio_request_cache_init();
2218         if (!err) {
2219                 err = register_filesystem(&ovl_fs_type);
2220                 if (!err)
2221                         return 0;
2222
2223                 ovl_aio_request_cache_destroy();
2224         }
2225         kmem_cache_destroy(ovl_inode_cachep);
2226
2227         return err;
2228 }
2229
2230 static void __exit ovl_exit(void)
2231 {
2232         unregister_filesystem(&ovl_fs_type);
2233
2234         /*
2235          * Make sure all delayed rcu free inodes are flushed before we
2236          * destroy cache.
2237          */
2238         rcu_barrier();
2239         kmem_cache_destroy(ovl_inode_cachep);
2240         ovl_aio_request_cache_destroy();
2241 }
2242
2243 module_init(ovl_init);
2244 module_exit(ovl_exit);