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(const 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                 /* The above comment can be understood in two ways:
1026                  *
1027                  * 1. We just want to check whether the basic POSIX ACL format
1028                  *    is ok. For example, if the header is correct and the size
1029                  *    is sane.
1030                  * 2. We want to know whether the ACL_{GROUP,USER} entries can
1031                  *    be mapped according to the underlying filesystem.
1032                  *
1033                  * Currently, we only check 1. If we wanted to check 2. we
1034                  * would need to pass the mnt_userns and the fs_userns of the
1035                  * underlying filesystem. But frankly, I think checking 1. is
1036                  * enough to start the copy-up.
1037                  */
1038                 acl = vfs_set_acl_prepare(&init_user_ns, &init_user_ns, value, size);
1039                 if (IS_ERR(acl))
1040                         return PTR_ERR(acl);
1041         }
1042         err = -EOPNOTSUPP;
1043         if (!IS_POSIXACL(d_inode(workdir)))
1044                 goto out_acl_release;
1045         if (!realinode->i_op->set_acl)
1046                 goto out_acl_release;
1047         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1048                 err = acl ? -EACCES : 0;
1049                 goto out_acl_release;
1050         }
1051         err = -EPERM;
1052         if (!inode_owner_or_capable(&init_user_ns, inode))
1053                 goto out_acl_release;
1054
1055         posix_acl_release(acl);
1056
1057         /*
1058          * Check if sgid bit needs to be cleared (actual setacl operation will
1059          * be done with mounter's capabilities and so that won't do it for us).
1060          */
1061         if (unlikely(inode->i_mode & S_ISGID) &&
1062             handler->flags == ACL_TYPE_ACCESS &&
1063             !in_group_p(inode->i_gid) &&
1064             !capable_wrt_inode_uidgid(&init_user_ns, inode, CAP_FSETID)) {
1065                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
1066
1067                 err = ovl_setattr(&init_user_ns, dentry, &iattr);
1068                 if (err)
1069                         return err;
1070         }
1071
1072         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
1073         return err;
1074
1075 out_acl_release:
1076         posix_acl_release(acl);
1077         return err;
1078 }
1079
1080 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1081                              struct dentry *dentry, struct inode *inode,
1082                              const char *name, void *buffer, size_t size)
1083 {
1084         return -EOPNOTSUPP;
1085 }
1086
1087 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1088                              struct user_namespace *mnt_userns,
1089                              struct dentry *dentry, struct inode *inode,
1090                              const char *name, const void *value,
1091                              size_t size, int flags)
1092 {
1093         return -EOPNOTSUPP;
1094 }
1095
1096 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1097                                struct dentry *dentry, struct inode *inode,
1098                                const char *name, void *buffer, size_t size)
1099 {
1100         return ovl_xattr_get(dentry, inode, name, buffer, size);
1101 }
1102
1103 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1104                                struct user_namespace *mnt_userns,
1105                                struct dentry *dentry, struct inode *inode,
1106                                const char *name, const void *value,
1107                                size_t size, int flags)
1108 {
1109         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1110 }
1111
1112 static const struct xattr_handler __maybe_unused
1113 ovl_posix_acl_access_xattr_handler = {
1114         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1115         .flags = ACL_TYPE_ACCESS,
1116         .get = ovl_posix_acl_xattr_get,
1117         .set = ovl_posix_acl_xattr_set,
1118 };
1119
1120 static const struct xattr_handler __maybe_unused
1121 ovl_posix_acl_default_xattr_handler = {
1122         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1123         .flags = ACL_TYPE_DEFAULT,
1124         .get = ovl_posix_acl_xattr_get,
1125         .set = ovl_posix_acl_xattr_set,
1126 };
1127
1128 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1129         .prefix = OVL_XATTR_TRUSTED_PREFIX,
1130         .get = ovl_own_xattr_get,
1131         .set = ovl_own_xattr_set,
1132 };
1133
1134 static const struct xattr_handler ovl_own_user_xattr_handler = {
1135         .prefix = OVL_XATTR_USER_PREFIX,
1136         .get = ovl_own_xattr_get,
1137         .set = ovl_own_xattr_set,
1138 };
1139
1140 static const struct xattr_handler ovl_other_xattr_handler = {
1141         .prefix = "", /* catch all */
1142         .get = ovl_other_xattr_get,
1143         .set = ovl_other_xattr_set,
1144 };
1145
1146 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1147 #ifdef CONFIG_FS_POSIX_ACL
1148         &ovl_posix_acl_access_xattr_handler,
1149         &ovl_posix_acl_default_xattr_handler,
1150 #endif
1151         &ovl_own_trusted_xattr_handler,
1152         &ovl_other_xattr_handler,
1153         NULL
1154 };
1155
1156 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1157 #ifdef CONFIG_FS_POSIX_ACL
1158         &ovl_posix_acl_access_xattr_handler,
1159         &ovl_posix_acl_default_xattr_handler,
1160 #endif
1161         &ovl_own_user_xattr_handler,
1162         &ovl_other_xattr_handler,
1163         NULL
1164 };
1165
1166 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1167                           struct inode **ptrap, const char *name)
1168 {
1169         struct inode *trap;
1170         int err;
1171
1172         trap = ovl_get_trap_inode(sb, dir);
1173         err = PTR_ERR_OR_ZERO(trap);
1174         if (err) {
1175                 if (err == -ELOOP)
1176                         pr_err("conflicting %s path\n", name);
1177                 return err;
1178         }
1179
1180         *ptrap = trap;
1181         return 0;
1182 }
1183
1184 /*
1185  * Determine how we treat concurrent use of upperdir/workdir based on the
1186  * index feature. This is papering over mount leaks of container runtimes,
1187  * for example, an old overlay mount is leaked and now its upperdir is
1188  * attempted to be used as a lower layer in a new overlay mount.
1189  */
1190 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1191 {
1192         if (ofs->config.index) {
1193                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1194                        name);
1195                 return -EBUSY;
1196         } else {
1197                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1198                         name);
1199                 return 0;
1200         }
1201 }
1202
1203 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1204                          struct ovl_layer *upper_layer, struct path *upperpath)
1205 {
1206         struct vfsmount *upper_mnt;
1207         int err;
1208
1209         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1210         if (err)
1211                 goto out;
1212
1213         /* Upperdir path should not be r/o */
1214         if (__mnt_is_readonly(upperpath->mnt)) {
1215                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1216                 err = -EINVAL;
1217                 goto out;
1218         }
1219
1220         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1221         if (err)
1222                 goto out;
1223
1224         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1225                              "upperdir");
1226         if (err)
1227                 goto out;
1228
1229         upper_mnt = clone_private_mount(upperpath);
1230         err = PTR_ERR(upper_mnt);
1231         if (IS_ERR(upper_mnt)) {
1232                 pr_err("failed to clone upperpath\n");
1233                 goto out;
1234         }
1235
1236         /* Don't inherit atime flags */
1237         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1238         upper_layer->mnt = upper_mnt;
1239         upper_layer->idx = 0;
1240         upper_layer->fsid = 0;
1241
1242         /*
1243          * Inherit SB_NOSEC flag from upperdir.
1244          *
1245          * This optimization changes behavior when a security related attribute
1246          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1247          * okay because we don't yet have guarantees in that case, but it will
1248          * need careful treatment once we want to honour changes to underlying
1249          * filesystems.
1250          */
1251         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1252                 sb->s_flags |= SB_NOSEC;
1253
1254         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1255                 ofs->upperdir_locked = true;
1256         } else {
1257                 err = ovl_report_in_use(ofs, "upperdir");
1258                 if (err)
1259                         goto out;
1260         }
1261
1262         err = 0;
1263 out:
1264         return err;
1265 }
1266
1267 /*
1268  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1269  * negative values if error is encountered.
1270  */
1271 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
1272 {
1273         struct dentry *workdir = ofs->workdir;
1274         struct inode *dir = d_inode(workdir);
1275         struct dentry *temp;
1276         struct dentry *dest;
1277         struct dentry *whiteout;
1278         struct name_snapshot name;
1279         int err;
1280
1281         inode_lock_nested(dir, I_MUTEX_PARENT);
1282
1283         temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
1284         err = PTR_ERR(temp);
1285         if (IS_ERR(temp))
1286                 goto out_unlock;
1287
1288         dest = ovl_lookup_temp(ofs, workdir);
1289         err = PTR_ERR(dest);
1290         if (IS_ERR(dest)) {
1291                 dput(temp);
1292                 goto out_unlock;
1293         }
1294
1295         /* Name is inline and stable - using snapshot as a copy helper */
1296         take_dentry_name_snapshot(&name, temp);
1297         err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
1298         if (err) {
1299                 if (err == -EINVAL)
1300                         err = 0;
1301                 goto cleanup_temp;
1302         }
1303
1304         whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
1305         err = PTR_ERR(whiteout);
1306         if (IS_ERR(whiteout))
1307                 goto cleanup_temp;
1308
1309         err = ovl_is_whiteout(whiteout);
1310
1311         /* Best effort cleanup of whiteout and temp file */
1312         if (err)
1313                 ovl_cleanup(ofs, dir, whiteout);
1314         dput(whiteout);
1315
1316 cleanup_temp:
1317         ovl_cleanup(ofs, dir, temp);
1318         release_dentry_name_snapshot(&name);
1319         dput(temp);
1320         dput(dest);
1321
1322 out_unlock:
1323         inode_unlock(dir);
1324
1325         return err;
1326 }
1327
1328 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
1329                                            struct dentry *parent,
1330                                            const char *name, umode_t mode)
1331 {
1332         size_t len = strlen(name);
1333         struct dentry *child;
1334
1335         inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1336         child = ovl_lookup_upper(ofs, name, parent, len);
1337         if (!IS_ERR(child) && !child->d_inode)
1338                 child = ovl_create_real(ofs, parent->d_inode, child,
1339                                         OVL_CATTR(mode));
1340         inode_unlock(parent->d_inode);
1341         dput(parent);
1342
1343         return child;
1344 }
1345
1346 /*
1347  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1348  * present.
1349  */
1350 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1351 {
1352         unsigned int ctr;
1353         struct dentry *d = dget(ofs->workbasedir);
1354         static const char *const volatile_path[] = {
1355                 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1356         };
1357         const char *const *name = volatile_path;
1358
1359         for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1360                 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1361                 if (IS_ERR(d))
1362                         return PTR_ERR(d);
1363         }
1364         dput(d);
1365         return 0;
1366 }
1367
1368 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1369                             const struct path *workpath)
1370 {
1371         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1372         struct dentry *temp, *workdir;
1373         bool rename_whiteout;
1374         bool d_type;
1375         int fh_type;
1376         int err;
1377
1378         err = mnt_want_write(mnt);
1379         if (err)
1380                 return err;
1381
1382         workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1383         err = PTR_ERR(workdir);
1384         if (IS_ERR_OR_NULL(workdir))
1385                 goto out;
1386
1387         ofs->workdir = workdir;
1388
1389         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1390         if (err)
1391                 goto out;
1392
1393         /*
1394          * Upper should support d_type, else whiteouts are visible.  Given
1395          * workdir and upper are on same fs, we can do iterate_dir() on
1396          * workdir. This check requires successful creation of workdir in
1397          * previous step.
1398          */
1399         err = ovl_check_d_type_supported(workpath);
1400         if (err < 0)
1401                 goto out;
1402
1403         d_type = err;
1404         if (!d_type)
1405                 pr_warn("upper fs needs to support d_type.\n");
1406
1407         /* Check if upper/work fs supports O_TMPFILE */
1408         temp = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
1409         ofs->tmpfile = !IS_ERR(temp);
1410         if (ofs->tmpfile)
1411                 dput(temp);
1412         else
1413                 pr_warn("upper fs does not support tmpfile.\n");
1414
1415
1416         /* Check if upper/work fs supports RENAME_WHITEOUT */
1417         err = ovl_check_rename_whiteout(ofs);
1418         if (err < 0)
1419                 goto out;
1420
1421         rename_whiteout = err;
1422         if (!rename_whiteout)
1423                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1424
1425         /*
1426          * Check if upper/work fs supports (trusted|user).overlay.* xattr
1427          */
1428         err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1429         if (err) {
1430                 pr_warn("failed to set xattr on upper\n");
1431                 ofs->noxattr = true;
1432                 if (ofs->config.index || ofs->config.metacopy) {
1433                         ofs->config.index = false;
1434                         ofs->config.metacopy = false;
1435                         pr_warn("...falling back to index=off,metacopy=off.\n");
1436                 }
1437                 /*
1438                  * xattr support is required for persistent st_ino.
1439                  * Without persistent st_ino, xino=auto falls back to xino=off.
1440                  */
1441                 if (ofs->config.xino == OVL_XINO_AUTO) {
1442                         ofs->config.xino = OVL_XINO_OFF;
1443                         pr_warn("...falling back to xino=off.\n");
1444                 }
1445                 if (err == -EPERM && !ofs->config.userxattr)
1446                         pr_info("try mounting with 'userxattr' option\n");
1447                 err = 0;
1448         } else {
1449                 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1450         }
1451
1452         /*
1453          * We allowed sub-optimal upper fs configuration and don't want to break
1454          * users over kernel upgrade, but we never allowed remote upper fs, so
1455          * we can enforce strict requirements for remote upper fs.
1456          */
1457         if (ovl_dentry_remote(ofs->workdir) &&
1458             (!d_type || !rename_whiteout || ofs->noxattr)) {
1459                 pr_err("upper fs missing required features.\n");
1460                 err = -EINVAL;
1461                 goto out;
1462         }
1463
1464         /*
1465          * For volatile mount, create a incompat/volatile/dirty file to keep
1466          * track of it.
1467          */
1468         if (ofs->config.ovl_volatile) {
1469                 err = ovl_create_volatile_dirty(ofs);
1470                 if (err < 0) {
1471                         pr_err("Failed to create volatile/dirty file.\n");
1472                         goto out;
1473                 }
1474         }
1475
1476         /* Check if upper/work fs supports file handles */
1477         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1478         if (ofs->config.index && !fh_type) {
1479                 ofs->config.index = false;
1480                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1481         }
1482
1483         /* Check if upper fs has 32bit inode numbers */
1484         if (fh_type != FILEID_INO32_GEN)
1485                 ofs->xino_mode = -1;
1486
1487         /* NFS export of r/w mount depends on index */
1488         if (ofs->config.nfs_export && !ofs->config.index) {
1489                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1490                 ofs->config.nfs_export = false;
1491         }
1492 out:
1493         mnt_drop_write(mnt);
1494         return err;
1495 }
1496
1497 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1498                            const struct path *upperpath)
1499 {
1500         int err;
1501         struct path workpath = { };
1502
1503         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1504         if (err)
1505                 goto out;
1506
1507         err = -EINVAL;
1508         if (upperpath->mnt != workpath.mnt) {
1509                 pr_err("workdir and upperdir must reside under the same mount\n");
1510                 goto out;
1511         }
1512         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1513                 pr_err("workdir and upperdir must be separate subtrees\n");
1514                 goto out;
1515         }
1516
1517         ofs->workbasedir = dget(workpath.dentry);
1518
1519         if (ovl_inuse_trylock(ofs->workbasedir)) {
1520                 ofs->workdir_locked = true;
1521         } else {
1522                 err = ovl_report_in_use(ofs, "workdir");
1523                 if (err)
1524                         goto out;
1525         }
1526
1527         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1528                              "workdir");
1529         if (err)
1530                 goto out;
1531
1532         err = ovl_make_workdir(sb, ofs, &workpath);
1533
1534 out:
1535         path_put(&workpath);
1536
1537         return err;
1538 }
1539
1540 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1541                             struct ovl_entry *oe, const struct path *upperpath)
1542 {
1543         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1544         struct dentry *indexdir;
1545         int err;
1546
1547         err = mnt_want_write(mnt);
1548         if (err)
1549                 return err;
1550
1551         /* Verify lower root is upper root origin */
1552         err = ovl_verify_origin(ofs, upperpath->dentry,
1553                                 oe->lowerstack[0].dentry, true);
1554         if (err) {
1555                 pr_err("failed to verify upper root origin\n");
1556                 goto out;
1557         }
1558
1559         /* index dir will act also as workdir */
1560         iput(ofs->workdir_trap);
1561         ofs->workdir_trap = NULL;
1562         dput(ofs->workdir);
1563         ofs->workdir = NULL;
1564         indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1565         if (IS_ERR(indexdir)) {
1566                 err = PTR_ERR(indexdir);
1567         } else if (indexdir) {
1568                 ofs->indexdir = indexdir;
1569                 ofs->workdir = dget(indexdir);
1570
1571                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1572                                      "indexdir");
1573                 if (err)
1574                         goto out;
1575
1576                 /*
1577                  * Verify upper root is exclusively associated with index dir.
1578                  * Older kernels stored upper fh in ".overlay.origin"
1579                  * xattr. If that xattr exists, verify that it is a match to
1580                  * upper dir file handle. In any case, verify or set xattr
1581                  * ".overlay.upper" to indicate that index may have
1582                  * directory entries.
1583                  */
1584                 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1585                         err = ovl_verify_set_fh(ofs, ofs->indexdir,
1586                                                 OVL_XATTR_ORIGIN,
1587                                                 upperpath->dentry, true, false);
1588                         if (err)
1589                                 pr_err("failed to verify index dir 'origin' xattr\n");
1590                 }
1591                 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1592                                        true);
1593                 if (err)
1594                         pr_err("failed to verify index dir 'upper' xattr\n");
1595
1596                 /* Cleanup bad/stale/orphan index entries */
1597                 if (!err)
1598                         err = ovl_indexdir_cleanup(ofs);
1599         }
1600         if (err || !ofs->indexdir)
1601                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1602
1603 out:
1604         mnt_drop_write(mnt);
1605         return err;
1606 }
1607
1608 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1609 {
1610         unsigned int i;
1611
1612         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1613                 return true;
1614
1615         /*
1616          * We allow using single lower with null uuid for index and nfs_export
1617          * for example to support those features with single lower squashfs.
1618          * To avoid regressions in setups of overlay with re-formatted lower
1619          * squashfs, do not allow decoding origin with lower null uuid unless
1620          * user opted-in to one of the new features that require following the
1621          * lower inode of non-dir upper.
1622          */
1623         if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
1624                 return false;
1625
1626         for (i = 0; i < ofs->numfs; i++) {
1627                 /*
1628                  * We use uuid to associate an overlay lower file handle with a
1629                  * lower layer, so we can accept lower fs with null uuid as long
1630                  * as all lower layers with null uuid are on the same fs.
1631                  * if we detect multiple lower fs with the same uuid, we
1632                  * disable lower file handle decoding on all of them.
1633                  */
1634                 if (ofs->fs[i].is_lower &&
1635                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1636                         ofs->fs[i].bad_uuid = true;
1637                         return false;
1638                 }
1639         }
1640         return true;
1641 }
1642
1643 /* Get a unique fsid for the layer */
1644 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1645 {
1646         struct super_block *sb = path->mnt->mnt_sb;
1647         unsigned int i;
1648         dev_t dev;
1649         int err;
1650         bool bad_uuid = false;
1651         bool warn = false;
1652
1653         for (i = 0; i < ofs->numfs; i++) {
1654                 if (ofs->fs[i].sb == sb)
1655                         return i;
1656         }
1657
1658         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1659                 bad_uuid = true;
1660                 if (ofs->config.xino == OVL_XINO_AUTO) {
1661                         ofs->config.xino = OVL_XINO_OFF;
1662                         warn = true;
1663                 }
1664                 if (ofs->config.index || ofs->config.nfs_export) {
1665                         ofs->config.index = false;
1666                         ofs->config.nfs_export = false;
1667                         warn = true;
1668                 }
1669                 if (warn) {
1670                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1671                                 uuid_is_null(&sb->s_uuid) ? "null" :
1672                                                             "conflicting",
1673                                 path->dentry, ovl_xino_str[ofs->config.xino]);
1674                 }
1675         }
1676
1677         err = get_anon_bdev(&dev);
1678         if (err) {
1679                 pr_err("failed to get anonymous bdev for lowerpath\n");
1680                 return err;
1681         }
1682
1683         ofs->fs[ofs->numfs].sb = sb;
1684         ofs->fs[ofs->numfs].pseudo_dev = dev;
1685         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1686
1687         return ofs->numfs++;
1688 }
1689
1690 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1691                           struct path *stack, unsigned int numlower,
1692                           struct ovl_layer *layers)
1693 {
1694         int err;
1695         unsigned int i;
1696
1697         err = -ENOMEM;
1698         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1699         if (ofs->fs == NULL)
1700                 goto out;
1701
1702         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1703         ofs->numfs++;
1704
1705         /*
1706          * All lower layers that share the same fs as upper layer, use the same
1707          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1708          * only overlay to simplify ovl_fs_free().
1709          * is_lower will be set if upper fs is shared with a lower layer.
1710          */
1711         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1712         if (err) {
1713                 pr_err("failed to get anonymous bdev for upper fs\n");
1714                 goto out;
1715         }
1716
1717         if (ovl_upper_mnt(ofs)) {
1718                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1719                 ofs->fs[0].is_lower = false;
1720         }
1721
1722         for (i = 0; i < numlower; i++) {
1723                 struct vfsmount *mnt;
1724                 struct inode *trap;
1725                 int fsid;
1726
1727                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1728                 if (err < 0)
1729                         goto out;
1730
1731                 /*
1732                  * Check if lower root conflicts with this overlay layers before
1733                  * checking if it is in-use as upperdir/workdir of "another"
1734                  * mount, because we do not bother to check in ovl_is_inuse() if
1735                  * the upperdir/workdir is in fact in-use by our
1736                  * upperdir/workdir.
1737                  */
1738                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1739                 if (err)
1740                         goto out;
1741
1742                 if (ovl_is_inuse(stack[i].dentry)) {
1743                         err = ovl_report_in_use(ofs, "lowerdir");
1744                         if (err) {
1745                                 iput(trap);
1746                                 goto out;
1747                         }
1748                 }
1749
1750                 mnt = clone_private_mount(&stack[i]);
1751                 err = PTR_ERR(mnt);
1752                 if (IS_ERR(mnt)) {
1753                         pr_err("failed to clone lowerpath\n");
1754                         iput(trap);
1755                         goto out;
1756                 }
1757
1758                 /*
1759                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1760                  * will fail instead of modifying lower fs.
1761                  */
1762                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1763
1764                 layers[ofs->numlayer].trap = trap;
1765                 layers[ofs->numlayer].mnt = mnt;
1766                 layers[ofs->numlayer].idx = ofs->numlayer;
1767                 layers[ofs->numlayer].fsid = fsid;
1768                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1769                 ofs->numlayer++;
1770                 ofs->fs[fsid].is_lower = true;
1771         }
1772
1773         /*
1774          * When all layers on same fs, overlay can use real inode numbers.
1775          * With mount option "xino=<on|auto>", mounter declares that there are
1776          * enough free high bits in underlying fs to hold the unique fsid.
1777          * If overlayfs does encounter underlying inodes using the high xino
1778          * bits reserved for fsid, it emits a warning and uses the original
1779          * inode number or a non persistent inode number allocated from a
1780          * dedicated range.
1781          */
1782         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1783                 if (ofs->config.xino == OVL_XINO_ON)
1784                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1785                 ofs->xino_mode = 0;
1786         } else if (ofs->config.xino == OVL_XINO_OFF) {
1787                 ofs->xino_mode = -1;
1788         } else if (ofs->xino_mode < 0) {
1789                 /*
1790                  * This is a roundup of number of bits needed for encoding
1791                  * fsid, where fsid 0 is reserved for upper fs (even with
1792                  * lower only overlay) +1 extra bit is reserved for the non
1793                  * persistent inode number range that is used for resolving
1794                  * xino lower bits overflow.
1795                  */
1796                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1797                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1798         }
1799
1800         if (ofs->xino_mode > 0) {
1801                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1802                         ofs->xino_mode);
1803         }
1804
1805         err = 0;
1806 out:
1807         return err;
1808 }
1809
1810 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1811                                 const char *lower, unsigned int numlower,
1812                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1813 {
1814         int err;
1815         struct path *stack = NULL;
1816         unsigned int i;
1817         struct ovl_entry *oe;
1818
1819         if (!ofs->config.upperdir && numlower == 1) {
1820                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1821                 return ERR_PTR(-EINVAL);
1822         }
1823
1824         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1825         if (!stack)
1826                 return ERR_PTR(-ENOMEM);
1827
1828         err = -EINVAL;
1829         for (i = 0; i < numlower; i++) {
1830                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1831                 if (err)
1832                         goto out_err;
1833
1834                 lower = strchr(lower, '\0') + 1;
1835         }
1836
1837         err = -EINVAL;
1838         sb->s_stack_depth++;
1839         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1840                 pr_err("maximum fs stacking depth exceeded\n");
1841                 goto out_err;
1842         }
1843
1844         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1845         if (err)
1846                 goto out_err;
1847
1848         err = -ENOMEM;
1849         oe = ovl_alloc_entry(numlower);
1850         if (!oe)
1851                 goto out_err;
1852
1853         for (i = 0; i < numlower; i++) {
1854                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1855                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1856         }
1857
1858 out:
1859         for (i = 0; i < numlower; i++)
1860                 path_put(&stack[i]);
1861         kfree(stack);
1862
1863         return oe;
1864
1865 out_err:
1866         oe = ERR_PTR(err);
1867         goto out;
1868 }
1869
1870 /*
1871  * Check if this layer root is a descendant of:
1872  * - another layer of this overlayfs instance
1873  * - upper/work dir of any overlayfs instance
1874  */
1875 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1876                            struct dentry *dentry, const char *name,
1877                            bool is_lower)
1878 {
1879         struct dentry *next = dentry, *parent;
1880         int err = 0;
1881
1882         if (!dentry)
1883                 return 0;
1884
1885         parent = dget_parent(next);
1886
1887         /* Walk back ancestors to root (inclusive) looking for traps */
1888         while (!err && parent != next) {
1889                 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1890                         err = -ELOOP;
1891                         pr_err("overlapping %s path\n", name);
1892                 } else if (ovl_is_inuse(parent)) {
1893                         err = ovl_report_in_use(ofs, name);
1894                 }
1895                 next = parent;
1896                 parent = dget_parent(next);
1897                 dput(next);
1898         }
1899
1900         dput(parent);
1901
1902         return err;
1903 }
1904
1905 /*
1906  * Check if any of the layers or work dirs overlap.
1907  */
1908 static int ovl_check_overlapping_layers(struct super_block *sb,
1909                                         struct ovl_fs *ofs)
1910 {
1911         int i, err;
1912
1913         if (ovl_upper_mnt(ofs)) {
1914                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1915                                       "upperdir", false);
1916                 if (err)
1917                         return err;
1918
1919                 /*
1920                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1921                  * this instance and covers overlapping work and index dirs,
1922                  * unless work or index dir have been moved since created inside
1923                  * workbasedir.  In that case, we already have their traps in
1924                  * inode cache and we will catch that case on lookup.
1925                  */
1926                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1927                                       false);
1928                 if (err)
1929                         return err;
1930         }
1931
1932         for (i = 1; i < ofs->numlayer; i++) {
1933                 err = ovl_check_layer(sb, ofs,
1934                                       ofs->layers[i].mnt->mnt_root,
1935                                       "lowerdir", true);
1936                 if (err)
1937                         return err;
1938         }
1939
1940         return 0;
1941 }
1942
1943 static struct dentry *ovl_get_root(struct super_block *sb,
1944                                    struct dentry *upperdentry,
1945                                    struct ovl_entry *oe)
1946 {
1947         struct dentry *root;
1948         struct ovl_path *lowerpath = &oe->lowerstack[0];
1949         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1950         int fsid = lowerpath->layer->fsid;
1951         struct ovl_inode_params oip = {
1952                 .upperdentry = upperdentry,
1953                 .lowerpath = lowerpath,
1954         };
1955
1956         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1957         if (!root)
1958                 return NULL;
1959
1960         root->d_fsdata = oe;
1961
1962         if (upperdentry) {
1963                 /* Root inode uses upper st_ino/i_ino */
1964                 ino = d_inode(upperdentry)->i_ino;
1965                 fsid = 0;
1966                 ovl_dentry_set_upper_alias(root);
1967                 if (ovl_is_impuredir(sb, upperdentry))
1968                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1969         }
1970
1971         /* Root is always merge -> can have whiteouts */
1972         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1973         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1974         ovl_set_upperdata(d_inode(root));
1975         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1976         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1977
1978         return root;
1979 }
1980
1981 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1982 {
1983         struct path upperpath = { };
1984         struct dentry *root_dentry;
1985         struct ovl_entry *oe;
1986         struct ovl_fs *ofs;
1987         struct ovl_layer *layers;
1988         struct cred *cred;
1989         char *splitlower = NULL;
1990         unsigned int numlower;
1991         int err;
1992
1993         err = -EIO;
1994         if (WARN_ON(sb->s_user_ns != current_user_ns()))
1995                 goto out;
1996
1997         sb->s_d_op = &ovl_dentry_operations;
1998
1999         err = -ENOMEM;
2000         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
2001         if (!ofs)
2002                 goto out;
2003
2004         err = -ENOMEM;
2005         ofs->creator_cred = cred = prepare_creds();
2006         if (!cred)
2007                 goto out_err;
2008
2009         /* Is there a reason anyone would want not to share whiteouts? */
2010         ofs->share_whiteout = true;
2011
2012         ofs->config.index = ovl_index_def;
2013         ofs->config.uuid = true;
2014         ofs->config.nfs_export = ovl_nfs_export_def;
2015         ofs->config.xino = ovl_xino_def();
2016         ofs->config.metacopy = ovl_metacopy_def;
2017         err = ovl_parse_opt((char *) data, &ofs->config);
2018         if (err)
2019                 goto out_err;
2020
2021         err = -EINVAL;
2022         if (!ofs->config.lowerdir) {
2023                 if (!silent)
2024                         pr_err("missing 'lowerdir'\n");
2025                 goto out_err;
2026         }
2027
2028         err = -ENOMEM;
2029         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
2030         if (!splitlower)
2031                 goto out_err;
2032
2033         err = -EINVAL;
2034         numlower = ovl_split_lowerdirs(splitlower);
2035         if (numlower > OVL_MAX_STACK) {
2036                 pr_err("too many lower directories, limit is %d\n",
2037                        OVL_MAX_STACK);
2038                 goto out_err;
2039         }
2040
2041         err = -ENOMEM;
2042         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
2043         if (!layers)
2044                 goto out_err;
2045
2046         ofs->layers = layers;
2047         /* Layer 0 is reserved for upper even if there's no upper */
2048         ofs->numlayer = 1;
2049
2050         sb->s_stack_depth = 0;
2051         sb->s_maxbytes = MAX_LFS_FILESIZE;
2052         atomic_long_set(&ofs->last_ino, 1);
2053         /* Assume underlying fs uses 32bit inodes unless proven otherwise */
2054         if (ofs->config.xino != OVL_XINO_OFF) {
2055                 ofs->xino_mode = BITS_PER_LONG - 32;
2056                 if (!ofs->xino_mode) {
2057                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
2058                         ofs->config.xino = OVL_XINO_OFF;
2059                 }
2060         }
2061
2062         /* alloc/destroy_inode needed for setting up traps in inode cache */
2063         sb->s_op = &ovl_super_operations;
2064
2065         if (ofs->config.upperdir) {
2066                 struct super_block *upper_sb;
2067
2068                 err = -EINVAL;
2069                 if (!ofs->config.workdir) {
2070                         pr_err("missing 'workdir'\n");
2071                         goto out_err;
2072                 }
2073
2074                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
2075                 if (err)
2076                         goto out_err;
2077
2078                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
2079                 if (!ovl_should_sync(ofs)) {
2080                         ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
2081                         if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
2082                                 err = -EIO;
2083                                 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
2084                                 goto out_err;
2085                         }
2086                 }
2087
2088                 err = ovl_get_workdir(sb, ofs, &upperpath);
2089                 if (err)
2090                         goto out_err;
2091
2092                 if (!ofs->workdir)
2093                         sb->s_flags |= SB_RDONLY;
2094
2095                 sb->s_stack_depth = upper_sb->s_stack_depth;
2096                 sb->s_time_gran = upper_sb->s_time_gran;
2097         }
2098         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2099         err = PTR_ERR(oe);
2100         if (IS_ERR(oe))
2101                 goto out_err;
2102
2103         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2104         if (!ovl_upper_mnt(ofs))
2105                 sb->s_flags |= SB_RDONLY;
2106
2107         if (!ofs->config.uuid && ofs->numfs > 1) {
2108                 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2109                 ofs->config.uuid = true;
2110         }
2111
2112         if (!ovl_force_readonly(ofs) && ofs->config.index) {
2113                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2114                 if (err)
2115                         goto out_free_oe;
2116
2117                 /* Force r/o mount with no index dir */
2118                 if (!ofs->indexdir)
2119                         sb->s_flags |= SB_RDONLY;
2120         }
2121
2122         err = ovl_check_overlapping_layers(sb, ofs);
2123         if (err)
2124                 goto out_free_oe;
2125
2126         /* Show index=off in /proc/mounts for forced r/o mount */
2127         if (!ofs->indexdir) {
2128                 ofs->config.index = false;
2129                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2130                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2131                         ofs->config.nfs_export = false;
2132                 }
2133         }
2134
2135         if (ofs->config.metacopy && ofs->config.nfs_export) {
2136                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2137                 ofs->config.nfs_export = false;
2138         }
2139
2140         if (ofs->config.nfs_export)
2141                 sb->s_export_op = &ovl_export_operations;
2142
2143         /* Never override disk quota limits or use reserved space */
2144         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2145
2146         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2147         sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2148                 ovl_trusted_xattr_handlers;
2149         sb->s_fs_info = ofs;
2150         sb->s_flags |= SB_POSIXACL;
2151         sb->s_iflags |= SB_I_SKIP_SYNC;
2152
2153         err = -ENOMEM;
2154         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2155         if (!root_dentry)
2156                 goto out_free_oe;
2157
2158         mntput(upperpath.mnt);
2159         kfree(splitlower);
2160
2161         sb->s_root = root_dentry;
2162
2163         return 0;
2164
2165 out_free_oe:
2166         ovl_entry_stack_free(oe);
2167         kfree(oe);
2168 out_err:
2169         kfree(splitlower);
2170         path_put(&upperpath);
2171         ovl_free_fs(ofs);
2172 out:
2173         return err;
2174 }
2175
2176 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2177                                 const char *dev_name, void *raw_data)
2178 {
2179         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2180 }
2181
2182 static struct file_system_type ovl_fs_type = {
2183         .owner          = THIS_MODULE,
2184         .name           = "overlay",
2185         .fs_flags       = FS_USERNS_MOUNT,
2186         .mount          = ovl_mount,
2187         .kill_sb        = kill_anon_super,
2188 };
2189 MODULE_ALIAS_FS("overlay");
2190
2191 static void ovl_inode_init_once(void *foo)
2192 {
2193         struct ovl_inode *oi = foo;
2194
2195         inode_init_once(&oi->vfs_inode);
2196 }
2197
2198 static int __init ovl_init(void)
2199 {
2200         int err;
2201
2202         ovl_inode_cachep = kmem_cache_create("ovl_inode",
2203                                              sizeof(struct ovl_inode), 0,
2204                                              (SLAB_RECLAIM_ACCOUNT|
2205                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2206                                              ovl_inode_init_once);
2207         if (ovl_inode_cachep == NULL)
2208                 return -ENOMEM;
2209
2210         err = ovl_aio_request_cache_init();
2211         if (!err) {
2212                 err = register_filesystem(&ovl_fs_type);
2213                 if (!err)
2214                         return 0;
2215
2216                 ovl_aio_request_cache_destroy();
2217         }
2218         kmem_cache_destroy(ovl_inode_cachep);
2219
2220         return err;
2221 }
2222
2223 static void __exit ovl_exit(void)
2224 {
2225         unregister_filesystem(&ovl_fs_type);
2226
2227         /*
2228          * Make sure all delayed rcu free inodes are flushed before we
2229          * destroy cache.
2230          */
2231         rcu_barrier();
2232         kmem_cache_destroy(ovl_inode_cachep);
2233         ovl_aio_request_cache_destroy();
2234 }
2235
2236 module_init(ovl_init);
2237 module_exit(ovl_exit);