fs/overlayfs/super.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *
   4  * Copyright (C) 2011 Novell Inc.
   5  */
   6
   7 #include <uapi/linux/magic.h>
   8 #include <linux/fs.h>
   9 #include <linux/namei.h>
  10 #include <linux/xattr.h>
  11 #include <linux/mount.h>
  12 #include <linux/parser.h>
  13 #include <linux/module.h>
  14 #include <linux/statfs.h>
  15 #include <linux/seq_file.h>
  16 #include <linux/posix_acl_xattr.h>
  17 #include <linux/exportfs.h>
  18 #include "overlayfs.h"
  19
  20 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  21 MODULE_DESCRIPTION("Overlay filesystem");
  22 MODULE_LICENSE("GPL");
  23
  24
  25 struct ovl_dir_cache;
  26
  27 #define OVL_MAX_STACK 500
  28
  29 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
  30 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
  31 MODULE_PARM_DESC(redirect_dir,
  32                  "Default to on or off for the redirect_dir feature");
  33
  34 static bool ovl_redirect_always_follow =
  35         IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
  36 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
  37                    bool, 0644);
  38 MODULE_PARM_DESC(redirect_always_follow,
  39                  "Follow redirects even if redirect_dir feature is turned off");
  40
  41 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
  42 module_param_named(index, ovl_index_def, bool, 0644);
  43 MODULE_PARM_DESC(index,
  44                  "Default to on or off for the inodes index feature");
  45
  46 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
  47 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
  48 MODULE_PARM_DESC(nfs_export,
  49                  "Default to on or off for the NFS export feature");
  50
  51 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
  52 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
  53 MODULE_PARM_DESC(xino_auto,
  54                  "Auto enable xino feature");
  55
  56 static void ovl_entry_stack_free(struct ovl_entry *oe)
  57 {
  58         unsigned int i;
  59
  60         for (i = 0; i < oe->numlower; i++)
  61                 dput(oe->lowerstack[i].dentry);
  62 }
  63
  64 static bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
  65 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
  66 MODULE_PARM_DESC(metacopy,
  67                  "Default to on or off for the metadata only copy up feature");
  68
  69 static void ovl_dentry_release(struct dentry *dentry)
  70 {
  71         struct ovl_entry *oe = dentry->d_fsdata;
  72
  73         if (oe) {
  74                 ovl_entry_stack_free(oe);
  75                 kfree_rcu(oe, rcu);
  76         }
  77 }
  78
  79 static struct dentry *ovl_d_real(struct dentry *dentry,
  80                                  const struct inode *inode)
  81 {
  82         struct dentry *real = NULL, *lower;
  83
  84         /* It's an overlay file */
  85         if (inode && d_inode(dentry) == inode)
  86                 return dentry;
  87
  88         if (!d_is_reg(dentry)) {
  89                 if (!inode || inode == d_inode(dentry))
  90                         return dentry;
  91                 goto bug;
  92         }
  93
  94         real = ovl_dentry_upper(dentry);
  95         if (real && (inode == d_inode(real)))
  96                 return real;
  97
  98         if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
  99                 return real;
 100
 101         lower = ovl_dentry_lowerdata(dentry);
 102         if (!lower)
 103                 goto bug;
 104         real = lower;
 105
 106         /* Handle recursion */
 107         real = d_real(real, inode);
 108
 109         if (!inode || inode == d_inode(real))
 110                 return real;
 111 bug:
 112         WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
 113              __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
 114              inode ? inode->i_ino : 0, real,
 115              real && d_inode(real) ? d_inode(real)->i_ino : 0);
 116         return dentry;
 117 }
 118
 119 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
 120 {
 121         int ret = 1;
 122
 123         if (weak) {
 124                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
 125                         ret =  d->d_op->d_weak_revalidate(d, flags);
 126         } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
 127                 ret = d->d_op->d_revalidate(d, flags);
 128                 if (!ret) {
 129                         if (!(flags & LOOKUP_RCU))
 130                                 d_invalidate(d);
 131                         ret = -ESTALE;
 132                 }
 133         }
 134         return ret;
 135 }
 136
 137 static int ovl_dentry_revalidate_common(struct dentry *dentry,
 138                                         unsigned int flags, bool weak)
 139 {
 140         struct ovl_entry *oe = dentry->d_fsdata;
 141         struct dentry *upper;
 142         unsigned int i;
 143         int ret = 1;
 144
 145         upper = ovl_dentry_upper(dentry);
 146         if (upper)
 147                 ret = ovl_revalidate_real(upper, flags, weak);
 148
 149         for (i = 0; ret > 0 && i < oe->numlower; i++) {
 150                 ret = ovl_revalidate_real(oe->lowerstack[i].dentry, flags,
 151                                           weak);
 152         }
 153         return ret;
 154 }
 155
 156 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 157 {
 158         return ovl_dentry_revalidate_common(dentry, flags, false);
 159 }
 160
 161 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 162 {
 163         return ovl_dentry_revalidate_common(dentry, flags, true);
 164 }
 165
 166 static const struct dentry_operations ovl_dentry_operations = {
 167         .d_release = ovl_dentry_release,
 168         .d_real = ovl_d_real,
 169         .d_revalidate = ovl_dentry_revalidate,
 170         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 171 };
 172
 173 static struct kmem_cache *ovl_inode_cachep;
 174
 175 static struct inode *ovl_alloc_inode(struct super_block *sb)
 176 {
 177         struct ovl_inode *oi = alloc_inode_sb(sb, ovl_inode_cachep, GFP_KERNEL);
 178
 179         if (!oi)
 180                 return NULL;
 181
 182         oi->cache = NULL;
 183         oi->redirect = NULL;
 184         oi->version = 0;
 185         oi->flags = 0;
 186         oi->__upperdentry = NULL;
 187         oi->lowerpath.dentry = NULL;
 188         oi->lowerpath.layer = NULL;
 189         oi->lowerdata = NULL;
 190         mutex_init(&oi->lock);
 191
 192         return &oi->vfs_inode;
 193 }
 194
 195 static void ovl_free_inode(struct inode *inode)
 196 {
 197         struct ovl_inode *oi = OVL_I(inode);
 198
 199         kfree(oi->redirect);
 200         mutex_destroy(&oi->lock);
 201         kmem_cache_free(ovl_inode_cachep, oi);
 202 }
 203
 204 static void ovl_destroy_inode(struct inode *inode)
 205 {
 206         struct ovl_inode *oi = OVL_I(inode);
 207
 208         dput(oi->__upperdentry);
 209         dput(oi->lowerpath.dentry);
 210         if (S_ISDIR(inode->i_mode))
 211                 ovl_dir_cache_free(inode);
 212         else
 213                 iput(oi->lowerdata);
 214 }
 215
 216 static void ovl_free_fs(struct ovl_fs *ofs)
 217 {
 218         struct vfsmount **mounts;
 219         unsigned i;
 220
 221         iput(ofs->workbasedir_trap);
 222         iput(ofs->indexdir_trap);
 223         iput(ofs->workdir_trap);
 224         dput(ofs->whiteout);
 225         dput(ofs->indexdir);
 226         dput(ofs->workdir);
 227         if (ofs->workdir_locked)
 228                 ovl_inuse_unlock(ofs->workbasedir);
 229         dput(ofs->workbasedir);
 230         if (ofs->upperdir_locked)
 231                 ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
 232
 233         /* Hack!  Reuse ofs->layers as a vfsmount array before freeing it */
 234         mounts = (struct vfsmount **) ofs->layers;
 235         for (i = 0; i < ofs->numlayer; i++) {
 236                 iput(ofs->layers[i].trap);
 237                 mounts[i] = ofs->layers[i].mnt;
 238         }
 239         kern_unmount_array(mounts, ofs->numlayer);
 240         kfree(ofs->layers);
 241         for (i = 0; i < ofs->numfs; i++)
 242                 free_anon_bdev(ofs->fs[i].pseudo_dev);
 243         kfree(ofs->fs);
 244
 245         kfree(ofs->config.lowerdir);
 246         kfree(ofs->config.upperdir);
 247         kfree(ofs->config.workdir);
 248         kfree(ofs->config.redirect_mode);
 249         if (ofs->creator_cred)
 250                 put_cred(ofs->creator_cred);
 251         kfree(ofs);
 252 }
 253
 254 static void ovl_put_super(struct super_block *sb)
 255 {
 256         struct ovl_fs *ofs = sb->s_fs_info;
 257
 258         ovl_free_fs(ofs);
 259 }
 260
 261 /* Sync real dirty inodes in upper filesystem (if it exists) */
 262 static int ovl_sync_fs(struct super_block *sb, int wait)
 263 {
 264         struct ovl_fs *ofs = sb->s_fs_info;
 265         struct super_block *upper_sb;
 266         int ret;
 267
 268         ret = ovl_sync_status(ofs);
 269         /*
 270          * We have to always set the err, because the return value isn't
 271          * checked in syncfs, and instead indirectly return an error via
 272          * the sb's writeback errseq, which VFS inspects after this call.
 273          */
 274         if (ret < 0) {
 275                 errseq_set(&sb->s_wb_err, -EIO);
 276                 return -EIO;
 277         }
 278
 279         if (!ret)
 280                 return ret;
 281
 282         /*
 283          * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
 284          * All the super blocks will be iterated, including upper_sb.
 285          *
 286          * If this is a syncfs(2) call, then we do need to call
 287          * sync_filesystem() on upper_sb, but enough if we do it when being
 288          * called with wait == 1.
 289          */
 290         if (!wait)
 291                 return 0;
 292
 293         upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 294
 295         down_read(&upper_sb->s_umount);
 296         ret = sync_filesystem(upper_sb);
 297         up_read(&upper_sb->s_umount);
 298
 299         return ret;
 300 }
 301
 302 /**
 303  * ovl_statfs
 304  * @sb: The overlayfs super block
 305  * @buf: The struct kstatfs to fill in with stats
 306  *
 307  * Get the filesystem statistics.  As writes always target the upper layer
 308  * filesystem pass the statfs to the upper filesystem (if it exists)
 309  */
 310 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 311 {
 312         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 313         struct dentry *root_dentry = dentry->d_sb->s_root;
 314         struct path path;
 315         int err;
 316
 317         ovl_path_real(root_dentry, &path);
 318
 319         err = vfs_statfs(&path, buf);
 320         if (!err) {
 321                 buf->f_namelen = ofs->namelen;
 322                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 323         }
 324
 325         return err;
 326 }
 327
 328 /* Will this overlay be forced to mount/remount ro? */
 329 static bool ovl_force_readonly(struct ovl_fs *ofs)
 330 {
 331         return (!ovl_upper_mnt(ofs) || !ofs->workdir);
 332 }
 333
 334 static const char *ovl_redirect_mode_def(void)
 335 {
 336         return ovl_redirect_dir_def ? "on" : "off";
 337 }
 338
 339 static const char * const ovl_xino_str[] = {
 340         "off",
 341         "auto",
 342         "on",
 343 };
 344
 345 static inline int ovl_xino_def(void)
 346 {
 347         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
 348 }
 349
 350 /**
 351  * ovl_show_options
 352  *
 353  * Prints the mount options for a given superblock.
 354  * Returns zero; does not fail.
 355  */
 356 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 357 {
 358         struct super_block *sb = dentry->d_sb;
 359         struct ovl_fs *ofs = sb->s_fs_info;
 360
 361         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
 362         if (ofs->config.upperdir) {
 363                 seq_show_option(m, "upperdir", ofs->config.upperdir);
 364                 seq_show_option(m, "workdir", ofs->config.workdir);
 365         }
 366         if (ofs->config.default_permissions)
 367                 seq_puts(m, ",default_permissions");
 368         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
 369                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
 370         if (ofs->config.index != ovl_index_def)
 371                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
 372         if (!ofs->config.uuid)
 373                 seq_puts(m, ",uuid=off");
 374         if (ofs->config.nfs_export != ovl_nfs_export_def)
 375                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
 376                                                 "on" : "off");
 377         if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
 378                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
 379         if (ofs->config.metacopy != ovl_metacopy_def)
 380                 seq_printf(m, ",metacopy=%s",
 381                            ofs->config.metacopy ? "on" : "off");
 382         if (ofs->config.ovl_volatile)
 383                 seq_puts(m, ",volatile");
 384         if (ofs->config.userxattr)
 385                 seq_puts(m, ",userxattr");
 386         return 0;
 387 }
 388
 389 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 390 {
 391         struct ovl_fs *ofs = sb->s_fs_info;
 392         struct super_block *upper_sb;
 393         int ret = 0;
 394
 395         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
 396                 return -EROFS;
 397
 398         if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
 399                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
 400                 if (ovl_should_sync(ofs)) {
 401                         down_read(&upper_sb->s_umount);
 402                         ret = sync_filesystem(upper_sb);
 403                         up_read(&upper_sb->s_umount);
 404                 }
 405         }
 406
 407         return ret;
 408 }
 409
 410 static const struct super_operations ovl_super_operations = {
 411         .alloc_inode    = ovl_alloc_inode,
 412         .free_inode     = ovl_free_inode,
 413         .destroy_inode  = ovl_destroy_inode,
 414         .drop_inode     = generic_delete_inode,
 415         .put_super      = ovl_put_super,
 416         .sync_fs        = ovl_sync_fs,
 417         .statfs         = ovl_statfs,
 418         .show_options   = ovl_show_options,
 419         .remount_fs     = ovl_remount,
 420 };
 421
 422 enum {
 423         OPT_LOWERDIR,
 424         OPT_UPPERDIR,
 425         OPT_WORKDIR,
 426         OPT_DEFAULT_PERMISSIONS,
 427         OPT_REDIRECT_DIR,
 428         OPT_INDEX_ON,
 429         OPT_INDEX_OFF,
 430         OPT_UUID_ON,
 431         OPT_UUID_OFF,
 432         OPT_NFS_EXPORT_ON,
 433         OPT_USERXATTR,
 434         OPT_NFS_EXPORT_OFF,
 435         OPT_XINO_ON,
 436         OPT_XINO_OFF,
 437         OPT_XINO_AUTO,
 438         OPT_METACOPY_ON,
 439         OPT_METACOPY_OFF,
 440         OPT_VOLATILE,
 441         OPT_ERR,
 442 };
 443
 444 static const match_table_t ovl_tokens = {
 445         {OPT_LOWERDIR,                  "lowerdir=%s"},
 446         {OPT_UPPERDIR,                  "upperdir=%s"},
 447         {OPT_WORKDIR,                   "workdir=%s"},
 448         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 449         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
 450         {OPT_INDEX_ON,                  "index=on"},
 451         {OPT_INDEX_OFF,                 "index=off"},
 452         {OPT_USERXATTR,                 "userxattr"},
 453         {OPT_UUID_ON,                   "uuid=on"},
 454         {OPT_UUID_OFF,                  "uuid=off"},
 455         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
 456         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
 457         {OPT_XINO_ON,                   "xino=on"},
 458         {OPT_XINO_OFF,                  "xino=off"},
 459         {OPT_XINO_AUTO,                 "xino=auto"},
 460         {OPT_METACOPY_ON,               "metacopy=on"},
 461         {OPT_METACOPY_OFF,              "metacopy=off"},
 462         {OPT_VOLATILE,                  "volatile"},
 463         {OPT_ERR,                       NULL}
 464 };
 465
 466 static char *ovl_next_opt(char **s)
 467 {
 468         char *sbegin = *s;
 469         char *p;
 470
 471         if (sbegin == NULL)
 472                 return NULL;
 473
 474         for (p = sbegin; *p; p++) {
 475                 if (*p == '\\') {
 476                         p++;
 477                         if (!*p)
 478                                 break;
 479                 } else if (*p == ',') {
 480                         *p = '\0';
 481                         *s = p + 1;
 482                         return sbegin;
 483                 }
 484         }
 485         *s = NULL;
 486         return sbegin;
 487 }
 488
 489 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
 490 {
 491         if (strcmp(mode, "on") == 0) {
 492                 config->redirect_dir = true;
 493                 /*
 494                  * Does not make sense to have redirect creation without
 495                  * redirect following.
 496                  */
 497                 config->redirect_follow = true;
 498         } else if (strcmp(mode, "follow") == 0) {
 499                 config->redirect_follow = true;
 500         } else if (strcmp(mode, "off") == 0) {
 501                 if (ovl_redirect_always_follow)
 502                         config->redirect_follow = true;
 503         } else if (strcmp(mode, "nofollow") != 0) {
 504                 pr_err("bad mount option \"redirect_dir=%s\"\n",
 505                        mode);
 506                 return -EINVAL;
 507         }
 508
 509         return 0;
 510 }
 511
 512 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 513 {
 514         char *p;
 515         int err;
 516         bool metacopy_opt = false, redirect_opt = false;
 517         bool nfs_export_opt = false, index_opt = false;
 518
 519         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
 520         if (!config->redirect_mode)
 521                 return -ENOMEM;
 522
 523         while ((p = ovl_next_opt(&opt)) != NULL) {
 524                 int token;
 525                 substring_t args[MAX_OPT_ARGS];
 526
 527                 if (!*p)
 528                         continue;
 529
 530                 token = match_token(p, ovl_tokens, args);
 531                 switch (token) {
 532                 case OPT_UPPERDIR:
 533                         kfree(config->upperdir);
 534                         config->upperdir = match_strdup(&args[0]);
 535                         if (!config->upperdir)
 536                                 return -ENOMEM;
 537                         break;
 538
 539                 case OPT_LOWERDIR:
 540                         kfree(config->lowerdir);
 541                         config->lowerdir = match_strdup(&args[0]);
 542                         if (!config->lowerdir)
 543                                 return -ENOMEM;
 544                         break;
 545
 546                 case OPT_WORKDIR:
 547                         kfree(config->workdir);
 548                         config->workdir = match_strdup(&args[0]);
 549                         if (!config->workdir)
 550                                 return -ENOMEM;
 551                         break;
 552
 553                 case OPT_DEFAULT_PERMISSIONS:
 554                         config->default_permissions = true;
 555                         break;
 556
 557                 case OPT_REDIRECT_DIR:
 558                         kfree(config->redirect_mode);
 559                         config->redirect_mode = match_strdup(&args[0]);
 560                         if (!config->redirect_mode)
 561                                 return -ENOMEM;
 562                         redirect_opt = true;
 563                         break;
 564
 565                 case OPT_INDEX_ON:
 566                         config->index = true;
 567                         index_opt = true;
 568                         break;
 569
 570                 case OPT_INDEX_OFF:
 571                         config->index = false;
 572                         index_opt = true;
 573                         break;
 574
 575                 case OPT_UUID_ON:
 576                         config->uuid = true;
 577                         break;
 578
 579                 case OPT_UUID_OFF:
 580                         config->uuid = false;
 581                         break;
 582
 583                 case OPT_NFS_EXPORT_ON:
 584                         config->nfs_export = true;
 585                         nfs_export_opt = true;
 586                         break;
 587
 588                 case OPT_NFS_EXPORT_OFF:
 589                         config->nfs_export = false;
 590                         nfs_export_opt = true;
 591                         break;
 592
 593                 case OPT_XINO_ON:
 594                         config->xino = OVL_XINO_ON;
 595                         break;
 596
 597                 case OPT_XINO_OFF:
 598                         config->xino = OVL_XINO_OFF;
 599                         break;
 600
 601                 case OPT_XINO_AUTO:
 602                         config->xino = OVL_XINO_AUTO;
 603                         break;
 604
 605                 case OPT_METACOPY_ON:
 606                         config->metacopy = true;
 607                         metacopy_opt = true;
 608                         break;
 609
 610                 case OPT_METACOPY_OFF:
 611                         config->metacopy = false;
 612                         metacopy_opt = true;
 613                         break;
 614
 615                 case OPT_VOLATILE:
 616                         config->ovl_volatile = true;
 617                         break;
 618
 619                 case OPT_USERXATTR:
 620                         config->userxattr = true;
 621                         break;
 622
 623                 default:
 624                         pr_err("unrecognized mount option \"%s\" or missing value\n",
 625                                         p);
 626                         return -EINVAL;
 627                 }
 628         }
 629
 630         /* Workdir/index are useless in non-upper mount */
 631         if (!config->upperdir) {
 632                 if (config->workdir) {
 633                         pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 634                                 config->workdir);
 635                         kfree(config->workdir);
 636                         config->workdir = NULL;
 637                 }
 638                 if (config->index && index_opt) {
 639                         pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
 640                         index_opt = false;
 641                 }
 642                 config->index = false;
 643         }
 644
 645         if (!config->upperdir && config->ovl_volatile) {
 646                 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
 647                 config->ovl_volatile = false;
 648         }
 649
 650         err = ovl_parse_redirect_mode(config, config->redirect_mode);
 651         if (err)
 652                 return err;
 653
 654         /*
 655          * This is to make the logic below simpler.  It doesn't make any other
 656          * difference, since config->redirect_dir is only used for upper.
 657          */
 658         if (!config->upperdir && config->redirect_follow)
 659                 config->redirect_dir = true;
 660
 661         /* Resolve metacopy -> redirect_dir dependency */
 662         if (config->metacopy && !config->redirect_dir) {
 663                 if (metacopy_opt && redirect_opt) {
 664                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
 665                                config->redirect_mode);
 666                         return -EINVAL;
 667                 }
 668                 if (redirect_opt) {
 669                         /*
 670                          * There was an explicit redirect_dir=... that resulted
 671                          * in this conflict.
 672                          */
 673                         pr_info("disabling metacopy due to redirect_dir=%s\n",
 674                                 config->redirect_mode);
 675                         config->metacopy = false;
 676                 } else {
 677                         /* Automatically enable redirect otherwise. */
 678                         config->redirect_follow = config->redirect_dir = true;
 679                 }
 680         }
 681
 682         /* Resolve nfs_export -> index dependency */
 683         if (config->nfs_export && !config->index) {
 684                 if (!config->upperdir && config->redirect_follow) {
 685                         pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
 686                         config->nfs_export = false;
 687                 } else if (nfs_export_opt && index_opt) {
 688                         pr_err("conflicting options: nfs_export=on,index=off\n");
 689                         return -EINVAL;
 690                 } else if (index_opt) {
 691                         /*
 692                          * There was an explicit index=off that resulted
 693                          * in this conflict.
 694                          */
 695                         pr_info("disabling nfs_export due to index=off\n");
 696                         config->nfs_export = false;
 697                 } else {
 698                         /* Automatically enable index otherwise. */
 699                         config->index = true;
 700                 }
 701         }
 702
 703         /* Resolve nfs_export -> !metacopy dependency */
 704         if (config->nfs_export && config->metacopy) {
 705                 if (nfs_export_opt && metacopy_opt) {
 706                         pr_err("conflicting options: nfs_export=on,metacopy=on\n");
 707                         return -EINVAL;
 708                 }
 709                 if (metacopy_opt) {
 710                         /*
 711                          * There was an explicit metacopy=on that resulted
 712                          * in this conflict.
 713                          */
 714                         pr_info("disabling nfs_export due to metacopy=on\n");
 715                         config->nfs_export = false;
 716                 } else {
 717                         /*
 718                          * There was an explicit nfs_export=on that resulted
 719                          * in this conflict.
 720                          */
 721                         pr_info("disabling metacopy due to nfs_export=on\n");
 722                         config->metacopy = false;
 723                 }
 724         }
 725
 726
 727         /* Resolve userxattr -> !redirect && !metacopy dependency */
 728         if (config->userxattr) {
 729                 if (config->redirect_follow && redirect_opt) {
 730                         pr_err("conflicting options: userxattr,redirect_dir=%s\n",
 731                                config->redirect_mode);
 732                         return -EINVAL;
 733                 }
 734                 if (config->metacopy && metacopy_opt) {
 735                         pr_err("conflicting options: userxattr,metacopy=on\n");
 736                         return -EINVAL;
 737                 }
 738                 /*
 739                  * Silently disable default setting of redirect and metacopy.
 740                  * This shall be the default in the future as well: these
 741                  * options must be explicitly enabled if used together with
 742                  * userxattr.
 743                  */
 744                 config->redirect_dir = config->redirect_follow = false;
 745                 config->metacopy = false;
 746         }
 747
 748         return 0;
 749 }
 750
 751 #define OVL_WORKDIR_NAME "work"
 752 #define OVL_INDEXDIR_NAME "index"
 753
 754 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
 755                                          const char *name, bool persist)
 756 {
 757         struct inode *dir =  ofs->workbasedir->d_inode;
 758         struct vfsmount *mnt = ovl_upper_mnt(ofs);
 759         struct dentry *work;
 760         int err;
 761         bool retried = false;
 762
 763         inode_lock_nested(dir, I_MUTEX_PARENT);
 764 retry:
 765         work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));
 766
 767         if (!IS_ERR(work)) {
 768                 struct iattr attr = {
 769                         .ia_valid = ATTR_MODE,
 770                         .ia_mode = S_IFDIR | 0,
 771                 };
 772
 773                 if (work->d_inode) {
 774                         err = -EEXIST;
 775                         if (retried)
 776                                 goto out_dput;
 777
 778                         if (persist)
 779                                 goto out_unlock;
 780
 781                         retried = true;
 782                         err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);
 783                         dput(work);
 784                         if (err == -EINVAL) {
 785                                 work = ERR_PTR(err);
 786                                 goto out_unlock;
 787                         }
 788                         goto retry;
 789                 }
 790
 791                 err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);
 792                 if (err)
 793                         goto out_dput;
 794
 795                 /* Weird filesystem returning with hashed negative (kernfs)? */
 796                 err = -EINVAL;
 797                 if (d_really_is_negative(work))
 798                         goto out_dput;
 799
 800                 /*
 801                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
 802                  *
 803                  * a) success (there was a POSIX ACL xattr and was removed)
 804                  * b) -ENODATA (there was no POSIX ACL xattr)
 805                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
 806                  *
 807                  * There are various other error values that could effectively
 808                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
 809                  * if the xattr name is too long), but the set of filesystems
 810                  * allowed as upper are limited to "normal" ones, where checking
 811                  * for the above two errors is sufficient.
 812                  */
 813                 err = ovl_do_removexattr(ofs, work,
 814                                          XATTR_NAME_POSIX_ACL_DEFAULT);
 815                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 816                         goto out_dput;
 817
 818                 err = ovl_do_removexattr(ofs, work,
 819                                          XATTR_NAME_POSIX_ACL_ACCESS);
 820                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
 821                         goto out_dput;
 822
 823                 /* Clear any inherited mode bits */
 824                 inode_lock(work->d_inode);
 825                 err = ovl_do_notify_change(ofs, work, &attr);
 826                 inode_unlock(work->d_inode);
 827                 if (err)
 828                         goto out_dput;
 829         } else {
 830                 err = PTR_ERR(work);
 831                 goto out_err;
 832         }
 833 out_unlock:
 834         inode_unlock(dir);
 835         return work;
 836
 837 out_dput:
 838         dput(work);
 839 out_err:
 840         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
 841                 ofs->config.workdir, name, -err);
 842         work = NULL;
 843         goto out_unlock;
 844 }
 845
 846 static void ovl_unescape(char *s)
 847 {
 848         char *d = s;
 849
 850         for (;; s++, d++) {
 851                 if (*s == '\\')
 852                         s++;
 853                 *d = *s;
 854                 if (!*s)
 855                         break;
 856         }
 857 }
 858
 859 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 860 {
 861         int err = -EINVAL;
 862
 863         if (!*name) {
 864                 pr_err("empty lowerdir\n");
 865                 goto out;
 866         }
 867         err = kern_path(name, LOOKUP_FOLLOW, path);
 868         if (err) {
 869                 pr_err("failed to resolve '%s': %i\n", name, err);
 870                 goto out;
 871         }
 872         err = -EINVAL;
 873         if (ovl_dentry_weird(path->dentry)) {
 874                 pr_err("filesystem on '%s' not supported\n", name);
 875                 goto out_put;
 876         }
 877         if (!d_is_dir(path->dentry)) {
 878                 pr_err("'%s' not a directory\n", name);
 879                 goto out_put;
 880         }
 881         return 0;
 882
 883 out_put:
 884         path_put_init(path);
 885 out:
 886         return err;
 887 }
 888
 889 static int ovl_mount_dir(const char *name, struct path *path)
 890 {
 891         int err = -ENOMEM;
 892         char *tmp = kstrdup(name, GFP_KERNEL);
 893
 894         if (tmp) {
 895                 ovl_unescape(tmp);
 896                 err = ovl_mount_dir_noesc(tmp, path);
 897
 898                 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
 899                         pr_err("filesystem on '%s' not supported as upperdir\n",
 900                                tmp);
 901                         path_put_init(path);
 902                         err = -EINVAL;
 903                 }
 904                 kfree(tmp);
 905         }
 906         return err;
 907 }
 908
 909 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
 910                              const char *name)
 911 {
 912         struct kstatfs statfs;
 913         int err = vfs_statfs(path, &statfs);
 914
 915         if (err)
 916                 pr_err("statfs failed on '%s'\n", name);
 917         else
 918                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
 919
 920         return err;
 921 }
 922
 923 static int ovl_lower_dir(const char *name, struct path *path,
 924                          struct ovl_fs *ofs, int *stack_depth)
 925 {
 926         int fh_type;
 927         int err;
 928
 929         err = ovl_mount_dir_noesc(name, path);
 930         if (err)
 931                 return err;
 932
 933         err = ovl_check_namelen(path, ofs, name);
 934         if (err)
 935                 return err;
 936
 937         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 938
 939         /*
 940          * The inodes index feature and NFS export need to encode and decode
 941          * file handles, so they require that all layers support them.
 942          */
 943         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
 944         if ((ofs->config.nfs_export ||
 945              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
 946                 ofs->config.index = false;
 947                 ofs->config.nfs_export = false;
 948                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
 949                         name);
 950         }
 951         /*
 952          * Decoding origin file handle is required for persistent st_ino.
 953          * Without persistent st_ino, xino=auto falls back to xino=off.
 954          */
 955         if (ofs->config.xino == OVL_XINO_AUTO &&
 956             ofs->config.upperdir && !fh_type) {
 957                 ofs->config.xino = OVL_XINO_OFF;
 958                 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
 959                         name);
 960         }
 961
 962         /* Check if lower fs has 32bit inode numbers */
 963         if (fh_type != FILEID_INO32_GEN)
 964                 ofs->xino_mode = -1;
 965
 966         return 0;
 967 }
 968
 969 /* Workdir should not be subdir of upperdir and vice versa */
 970 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 971 {
 972         bool ok = false;
 973
 974         if (workdir != upperdir) {
 975                 ok = (lock_rename(workdir, upperdir) == NULL);
 976                 unlock_rename(workdir, upperdir);
 977         }
 978         return ok;
 979 }
 980
 981 static unsigned int ovl_split_lowerdirs(char *str)
 982 {
 983         unsigned int ctr = 1;
 984         char *s, *d;
 985
 986         for (s = d = str;; s++, d++) {
 987                 if (*s == '\\') {
 988                         s++;
 989                 } else if (*s == ':') {
 990                         *d = '\0';
 991                         ctr++;
 992                         continue;
 993                 }
 994                 *d = *s;
 995                 if (!*s)
 996                         break;
 997         }
 998         return ctr;
 999 }
1000
1001 static int __maybe_unused
1002 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
1003                         struct dentry *dentry, struct inode *inode,
1004                         const char *name, void *buffer, size_t size)
1005 {
1006         if (!IS_POSIXACL(inode))
1007                 return -EOPNOTSUPP;
1008
1009         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
1010 }
1011
1012 static int __maybe_unused
1013 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
1014                         struct user_namespace *mnt_userns,
1015                         struct dentry *dentry, struct inode *inode,
1016                         const char *name, const void *value,
1017                         size_t size, int flags)
1018 {
1019         struct dentry *workdir = ovl_workdir(dentry);
1020         struct inode *realinode = ovl_inode_real(inode);
1021         struct posix_acl *acl = NULL;
1022         int err;
1023
1024         if (!IS_POSIXACL(inode))
1025                 return -EOPNOTSUPP;
1026
1027         /* Check that everything is OK before copy-up */
1028         if (value) {
1029                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
1030                 if (IS_ERR(acl))
1031                         return PTR_ERR(acl);
1032         }
1033         err = -EOPNOTSUPP;
1034         if (!IS_POSIXACL(d_inode(workdir)))
1035                 goto out_acl_release;
1036         if (!realinode->i_op->set_acl)
1037                 goto out_acl_release;
1038         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1039                 err = acl ? -EACCES : 0;
1040                 goto out_acl_release;
1041         }
1042         err = -EPERM;
1043         if (!inode_owner_or_capable(&init_user_ns, inode))
1044                 goto out_acl_release;
1045
1046         posix_acl_release(acl);
1047
1048         /*
1049          * Check if sgid bit needs to be cleared (actual setacl operation will
1050          * be done with mounter's capabilities and so that won't do it for us).
1051          */
1052         if (unlikely(inode->i_mode & S_ISGID) &&
1053             handler->flags == ACL_TYPE_ACCESS &&
1054             !in_group_p(inode->i_gid) &&
1055             !capable_wrt_inode_uidgid(&init_user_ns, inode, CAP_FSETID)) {
1056                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
1057
1058                 err = ovl_setattr(&init_user_ns, dentry, &iattr);
1059                 if (err)
1060                         return err;
1061         }
1062
1063         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
1064         return err;
1065
1066 out_acl_release:
1067         posix_acl_release(acl);
1068         return err;
1069 }
1070
1071 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1072                              struct dentry *dentry, struct inode *inode,
1073                              const char *name, void *buffer, size_t size)
1074 {
1075         return -EOPNOTSUPP;
1076 }
1077
1078 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1079                              struct user_namespace *mnt_userns,
1080                              struct dentry *dentry, struct inode *inode,
1081                              const char *name, const void *value,
1082                              size_t size, int flags)
1083 {
1084         return -EOPNOTSUPP;
1085 }
1086
1087 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1088                                struct dentry *dentry, struct inode *inode,
1089                                const char *name, void *buffer, size_t size)
1090 {
1091         return ovl_xattr_get(dentry, inode, name, buffer, size);
1092 }
1093
1094 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1095                                struct user_namespace *mnt_userns,
1096                                struct dentry *dentry, struct inode *inode,
1097                                const char *name, const void *value,
1098                                size_t size, int flags)
1099 {
1100         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1101 }
1102
1103 static const struct xattr_handler __maybe_unused
1104 ovl_posix_acl_access_xattr_handler = {
1105         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1106         .flags = ACL_TYPE_ACCESS,
1107         .get = ovl_posix_acl_xattr_get,
1108         .set = ovl_posix_acl_xattr_set,
1109 };
1110
1111 static const struct xattr_handler __maybe_unused
1112 ovl_posix_acl_default_xattr_handler = {
1113         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1114         .flags = ACL_TYPE_DEFAULT,
1115         .get = ovl_posix_acl_xattr_get,
1116         .set = ovl_posix_acl_xattr_set,
1117 };
1118
1119 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1120         .prefix = OVL_XATTR_TRUSTED_PREFIX,
1121         .get = ovl_own_xattr_get,
1122         .set = ovl_own_xattr_set,
1123 };
1124
1125 static const struct xattr_handler ovl_own_user_xattr_handler = {
1126         .prefix = OVL_XATTR_USER_PREFIX,
1127         .get = ovl_own_xattr_get,
1128         .set = ovl_own_xattr_set,
1129 };
1130
1131 static const struct xattr_handler ovl_other_xattr_handler = {
1132         .prefix = "", /* catch all */
1133         .get = ovl_other_xattr_get,
1134         .set = ovl_other_xattr_set,
1135 };
1136
1137 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1138 #ifdef CONFIG_FS_POSIX_ACL
1139         &ovl_posix_acl_access_xattr_handler,
1140         &ovl_posix_acl_default_xattr_handler,
1141 #endif
1142         &ovl_own_trusted_xattr_handler,
1143         &ovl_other_xattr_handler,
1144         NULL
1145 };
1146
1147 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1148 #ifdef CONFIG_FS_POSIX_ACL
1149         &ovl_posix_acl_access_xattr_handler,
1150         &ovl_posix_acl_default_xattr_handler,
1151 #endif
1152         &ovl_own_user_xattr_handler,
1153         &ovl_other_xattr_handler,
1154         NULL
1155 };
1156
1157 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1158                           struct inode **ptrap, const char *name)
1159 {
1160         struct inode *trap;
1161         int err;
1162
1163         trap = ovl_get_trap_inode(sb, dir);
1164         err = PTR_ERR_OR_ZERO(trap);
1165         if (err) {
1166                 if (err == -ELOOP)
1167                         pr_err("conflicting %s path\n", name);
1168                 return err;
1169         }
1170
1171         *ptrap = trap;
1172         return 0;
1173 }
1174
1175 /*
1176  * Determine how we treat concurrent use of upperdir/workdir based on the
1177  * index feature. This is papering over mount leaks of container runtimes,
1178  * for example, an old overlay mount is leaked and now its upperdir is
1179  * attempted to be used as a lower layer in a new overlay mount.
1180  */
1181 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1182 {
1183         if (ofs->config.index) {
1184                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1185                        name);
1186                 return -EBUSY;
1187         } else {
1188                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1189                         name);
1190                 return 0;
1191         }
1192 }
1193
1194 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1195                          struct ovl_layer *upper_layer, struct path *upperpath)
1196 {
1197         struct vfsmount *upper_mnt;
1198         int err;
1199
1200         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1201         if (err)
1202                 goto out;
1203
1204         /* Upperdir path should not be r/o */
1205         if (__mnt_is_readonly(upperpath->mnt)) {
1206                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1207                 err = -EINVAL;
1208                 goto out;
1209         }
1210
1211         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1212         if (err)
1213                 goto out;
1214
1215         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1216                              "upperdir");
1217         if (err)
1218                 goto out;
1219
1220         upper_mnt = clone_private_mount(upperpath);
1221         err = PTR_ERR(upper_mnt);
1222         if (IS_ERR(upper_mnt)) {
1223                 pr_err("failed to clone upperpath\n");
1224                 goto out;
1225         }
1226
1227         /* Don't inherit atime flags */
1228         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1229         upper_layer->mnt = upper_mnt;
1230         upper_layer->idx = 0;
1231         upper_layer->fsid = 0;
1232
1233         /*
1234          * Inherit SB_NOSEC flag from upperdir.
1235          *
1236          * This optimization changes behavior when a security related attribute
1237          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1238          * okay because we don't yet have guarantees in that case, but it will
1239          * need careful treatment once we want to honour changes to underlying
1240          * filesystems.
1241          */
1242         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1243                 sb->s_flags |= SB_NOSEC;
1244
1245         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1246                 ofs->upperdir_locked = true;
1247         } else {
1248                 err = ovl_report_in_use(ofs, "upperdir");
1249                 if (err)
1250                         goto out;
1251         }
1252
1253         err = 0;
1254 out:
1255         return err;
1256 }
1257
1258 /*
1259  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1260  * negative values if error is encountered.
1261  */
1262 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
1263 {
1264         struct dentry *workdir = ofs->workdir;
1265         struct inode *dir = d_inode(workdir);
1266         struct dentry *temp;
1267         struct dentry *dest;
1268         struct dentry *whiteout;
1269         struct name_snapshot name;
1270         int err;
1271
1272         inode_lock_nested(dir, I_MUTEX_PARENT);
1273
1274         temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
1275         err = PTR_ERR(temp);
1276         if (IS_ERR(temp))
1277                 goto out_unlock;
1278
1279         dest = ovl_lookup_temp(ofs, workdir);
1280         err = PTR_ERR(dest);
1281         if (IS_ERR(dest)) {
1282                 dput(temp);
1283                 goto out_unlock;
1284         }
1285
1286         /* Name is inline and stable - using snapshot as a copy helper */
1287         take_dentry_name_snapshot(&name, temp);
1288         err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
1289         if (err) {
1290                 if (err == -EINVAL)
1291                         err = 0;
1292                 goto cleanup_temp;
1293         }
1294
1295         whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
1296         err = PTR_ERR(whiteout);
1297         if (IS_ERR(whiteout))
1298                 goto cleanup_temp;
1299
1300         err = ovl_is_whiteout(whiteout);
1301
1302         /* Best effort cleanup of whiteout and temp file */
1303         if (err)
1304                 ovl_cleanup(ofs, dir, whiteout);
1305         dput(whiteout);
1306
1307 cleanup_temp:
1308         ovl_cleanup(ofs, dir, temp);
1309         release_dentry_name_snapshot(&name);
1310         dput(temp);
1311         dput(dest);
1312
1313 out_unlock:
1314         inode_unlock(dir);
1315
1316         return err;
1317 }
1318
1319 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
1320                                            struct dentry *parent,
1321                                            const char *name, umode_t mode)
1322 {
1323         size_t len = strlen(name);
1324         struct dentry *child;
1325
1326         inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1327         child = ovl_lookup_upper(ofs, name, parent, len);
1328         if (!IS_ERR(child) && !child->d_inode)
1329                 child = ovl_create_real(ofs, parent->d_inode, child,
1330                                         OVL_CATTR(mode));
1331         inode_unlock(parent->d_inode);
1332         dput(parent);
1333
1334         return child;
1335 }
1336
1337 /*
1338  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1339  * present.
1340  */
1341 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1342 {
1343         unsigned int ctr;
1344         struct dentry *d = dget(ofs->workbasedir);
1345         static const char *const volatile_path[] = {
1346                 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1347         };
1348         const char *const *name = volatile_path;
1349
1350         for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1351                 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1352                 if (IS_ERR(d))
1353                         return PTR_ERR(d);
1354         }
1355         dput(d);
1356         return 0;
1357 }
1358
1359 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1360                             struct path *workpath)
1361 {
1362         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1363         struct dentry *temp, *workdir;
1364         bool rename_whiteout;
1365         bool d_type;
1366         int fh_type;
1367         int err;
1368
1369         err = mnt_want_write(mnt);
1370         if (err)
1371                 return err;
1372
1373         workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1374         err = PTR_ERR(workdir);
1375         if (IS_ERR_OR_NULL(workdir))
1376                 goto out;
1377
1378         ofs->workdir = workdir;
1379
1380         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1381         if (err)
1382                 goto out;
1383
1384         /*
1385          * Upper should support d_type, else whiteouts are visible.  Given
1386          * workdir and upper are on same fs, we can do iterate_dir() on
1387          * workdir. This check requires successful creation of workdir in
1388          * previous step.
1389          */
1390         err = ovl_check_d_type_supported(workpath);
1391         if (err < 0)
1392                 goto out;
1393
1394         d_type = err;
1395         if (!d_type)
1396                 pr_warn("upper fs needs to support d_type.\n");
1397
1398         /* Check if upper/work fs supports O_TMPFILE */
1399         temp = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
1400         ofs->tmpfile = !IS_ERR(temp);
1401         if (ofs->tmpfile)
1402                 dput(temp);
1403         else
1404                 pr_warn("upper fs does not support tmpfile.\n");
1405
1406
1407         /* Check if upper/work fs supports RENAME_WHITEOUT */
1408         err = ovl_check_rename_whiteout(ofs);
1409         if (err < 0)
1410                 goto out;
1411
1412         rename_whiteout = err;
1413         if (!rename_whiteout)
1414                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1415
1416         /*
1417          * Check if upper/work fs supports (trusted|user).overlay.* xattr
1418          */
1419         err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1420         if (err) {
1421                 ofs->noxattr = true;
1422                 if (ofs->config.index || ofs->config.metacopy) {
1423                         ofs->config.index = false;
1424                         ofs->config.metacopy = false;
1425                         pr_warn("upper fs does not support xattr, falling back to index=off,metacopy=off.\n");
1426                 }
1427                 /*
1428                  * xattr support is required for persistent st_ino.
1429                  * Without persistent st_ino, xino=auto falls back to xino=off.
1430                  */
1431                 if (ofs->config.xino == OVL_XINO_AUTO) {
1432                         ofs->config.xino = OVL_XINO_OFF;
1433                         pr_warn("upper fs does not support xattr, falling back to xino=off.\n");
1434                 }
1435                 err = 0;
1436         } else {
1437                 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1438         }
1439
1440         /*
1441          * We allowed sub-optimal upper fs configuration and don't want to break
1442          * users over kernel upgrade, but we never allowed remote upper fs, so
1443          * we can enforce strict requirements for remote upper fs.
1444          */
1445         if (ovl_dentry_remote(ofs->workdir) &&
1446             (!d_type || !rename_whiteout || ofs->noxattr)) {
1447                 pr_err("upper fs missing required features.\n");
1448                 err = -EINVAL;
1449                 goto out;
1450         }
1451
1452         /*
1453          * For volatile mount, create a incompat/volatile/dirty file to keep
1454          * track of it.
1455          */
1456         if (ofs->config.ovl_volatile) {
1457                 err = ovl_create_volatile_dirty(ofs);
1458                 if (err < 0) {
1459                         pr_err("Failed to create volatile/dirty file.\n");
1460                         goto out;
1461                 }
1462         }
1463
1464         /* Check if upper/work fs supports file handles */
1465         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1466         if (ofs->config.index && !fh_type) {
1467                 ofs->config.index = false;
1468                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1469         }
1470
1471         /* Check if upper fs has 32bit inode numbers */
1472         if (fh_type != FILEID_INO32_GEN)
1473                 ofs->xino_mode = -1;
1474
1475         /* NFS export of r/w mount depends on index */
1476         if (ofs->config.nfs_export && !ofs->config.index) {
1477                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1478                 ofs->config.nfs_export = false;
1479         }
1480 out:
1481         mnt_drop_write(mnt);
1482         return err;
1483 }
1484
1485 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1486                            struct path *upperpath)
1487 {
1488         int err;
1489         struct path workpath = { };
1490
1491         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1492         if (err)
1493                 goto out;
1494
1495         err = -EINVAL;
1496         if (upperpath->mnt != workpath.mnt) {
1497                 pr_err("workdir and upperdir must reside under the same mount\n");
1498                 goto out;
1499         }
1500         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1501                 pr_err("workdir and upperdir must be separate subtrees\n");
1502                 goto out;
1503         }
1504
1505         ofs->workbasedir = dget(workpath.dentry);
1506
1507         if (ovl_inuse_trylock(ofs->workbasedir)) {
1508                 ofs->workdir_locked = true;
1509         } else {
1510                 err = ovl_report_in_use(ofs, "workdir");
1511                 if (err)
1512                         goto out;
1513         }
1514
1515         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1516                              "workdir");
1517         if (err)
1518                 goto out;
1519
1520         err = ovl_make_workdir(sb, ofs, &workpath);
1521
1522 out:
1523         path_put(&workpath);
1524
1525         return err;
1526 }
1527
1528 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1529                             struct ovl_entry *oe, struct path *upperpath)
1530 {
1531         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1532         struct dentry *indexdir;
1533         int err;
1534
1535         err = mnt_want_write(mnt);
1536         if (err)
1537                 return err;
1538
1539         /* Verify lower root is upper root origin */
1540         err = ovl_verify_origin(ofs, upperpath->dentry,
1541                                 oe->lowerstack[0].dentry, true);
1542         if (err) {
1543                 pr_err("failed to verify upper root origin\n");
1544                 goto out;
1545         }
1546
1547         /* index dir will act also as workdir */
1548         iput(ofs->workdir_trap);
1549         ofs->workdir_trap = NULL;
1550         dput(ofs->workdir);
1551         ofs->workdir = NULL;
1552         indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1553         if (IS_ERR(indexdir)) {
1554                 err = PTR_ERR(indexdir);
1555         } else if (indexdir) {
1556                 ofs->indexdir = indexdir;
1557                 ofs->workdir = dget(indexdir);
1558
1559                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1560                                      "indexdir");
1561                 if (err)
1562                         goto out;
1563
1564                 /*
1565                  * Verify upper root is exclusively associated with index dir.
1566                  * Older kernels stored upper fh in ".overlay.origin"
1567                  * xattr. If that xattr exists, verify that it is a match to
1568                  * upper dir file handle. In any case, verify or set xattr
1569                  * ".overlay.upper" to indicate that index may have
1570                  * directory entries.
1571                  */
1572                 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1573                         err = ovl_verify_set_fh(ofs, ofs->indexdir,
1574                                                 OVL_XATTR_ORIGIN,
1575                                                 upperpath->dentry, true, false);
1576                         if (err)
1577                                 pr_err("failed to verify index dir 'origin' xattr\n");
1578                 }
1579                 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1580                                        true);
1581                 if (err)
1582                         pr_err("failed to verify index dir 'upper' xattr\n");
1583
1584                 /* Cleanup bad/stale/orphan index entries */
1585                 if (!err)
1586                         err = ovl_indexdir_cleanup(ofs);
1587         }
1588         if (err || !ofs->indexdir)
1589                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1590
1591 out:
1592         mnt_drop_write(mnt);
1593         return err;
1594 }
1595
1596 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1597 {
1598         unsigned int i;
1599
1600         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1601                 return true;
1602
1603         /*
1604          * We allow using single lower with null uuid for index and nfs_export
1605          * for example to support those features with single lower squashfs.
1606          * To avoid regressions in setups of overlay with re-formatted lower
1607          * squashfs, do not allow decoding origin with lower null uuid unless
1608          * user opted-in to one of the new features that require following the
1609          * lower inode of non-dir upper.
1610          */
1611         if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
1612                 return false;
1613
1614         for (i = 0; i < ofs->numfs; i++) {
1615                 /*
1616                  * We use uuid to associate an overlay lower file handle with a
1617                  * lower layer, so we can accept lower fs with null uuid as long
1618                  * as all lower layers with null uuid are on the same fs.
1619                  * if we detect multiple lower fs with the same uuid, we
1620                  * disable lower file handle decoding on all of them.
1621                  */
1622                 if (ofs->fs[i].is_lower &&
1623                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1624                         ofs->fs[i].bad_uuid = true;
1625                         return false;
1626                 }
1627         }
1628         return true;
1629 }
1630
1631 /* Get a unique fsid for the layer */
1632 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1633 {
1634         struct super_block *sb = path->mnt->mnt_sb;
1635         unsigned int i;
1636         dev_t dev;
1637         int err;
1638         bool bad_uuid = false;
1639         bool warn = false;
1640
1641         for (i = 0; i < ofs->numfs; i++) {
1642                 if (ofs->fs[i].sb == sb)
1643                         return i;
1644         }
1645
1646         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1647                 bad_uuid = true;
1648                 if (ofs->config.xino == OVL_XINO_AUTO) {
1649                         ofs->config.xino = OVL_XINO_OFF;
1650                         warn = true;
1651                 }
1652                 if (ofs->config.index || ofs->config.nfs_export) {
1653                         ofs->config.index = false;
1654                         ofs->config.nfs_export = false;
1655                         warn = true;
1656                 }
1657                 if (warn) {
1658                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1659                                 uuid_is_null(&sb->s_uuid) ? "null" :
1660                                                             "conflicting",
1661                                 path->dentry, ovl_xino_str[ofs->config.xino]);
1662                 }
1663         }
1664
1665         err = get_anon_bdev(&dev);
1666         if (err) {
1667                 pr_err("failed to get anonymous bdev for lowerpath\n");
1668                 return err;
1669         }
1670
1671         ofs->fs[ofs->numfs].sb = sb;
1672         ofs->fs[ofs->numfs].pseudo_dev = dev;
1673         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1674
1675         return ofs->numfs++;
1676 }
1677
1678 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1679                           struct path *stack, unsigned int numlower,
1680                           struct ovl_layer *layers)
1681 {
1682         int err;
1683         unsigned int i;
1684
1685         err = -ENOMEM;
1686         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1687         if (ofs->fs == NULL)
1688                 goto out;
1689
1690         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1691         ofs->numfs++;
1692
1693         /*
1694          * All lower layers that share the same fs as upper layer, use the same
1695          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1696          * only overlay to simplify ovl_fs_free().
1697          * is_lower will be set if upper fs is shared with a lower layer.
1698          */
1699         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1700         if (err) {
1701                 pr_err("failed to get anonymous bdev for upper fs\n");
1702                 goto out;
1703         }
1704
1705         if (ovl_upper_mnt(ofs)) {
1706                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1707                 ofs->fs[0].is_lower = false;
1708         }
1709
1710         for (i = 0; i < numlower; i++) {
1711                 struct vfsmount *mnt;
1712                 struct inode *trap;
1713                 int fsid;
1714
1715                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1716                 if (err < 0)
1717                         goto out;
1718
1719                 /*
1720                  * Check if lower root conflicts with this overlay layers before
1721                  * checking if it is in-use as upperdir/workdir of "another"
1722                  * mount, because we do not bother to check in ovl_is_inuse() if
1723                  * the upperdir/workdir is in fact in-use by our
1724                  * upperdir/workdir.
1725                  */
1726                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1727                 if (err)
1728                         goto out;
1729
1730                 if (ovl_is_inuse(stack[i].dentry)) {
1731                         err = ovl_report_in_use(ofs, "lowerdir");
1732                         if (err) {
1733                                 iput(trap);
1734                                 goto out;
1735                         }
1736                 }
1737
1738                 mnt = clone_private_mount(&stack[i]);
1739                 err = PTR_ERR(mnt);
1740                 if (IS_ERR(mnt)) {
1741                         pr_err("failed to clone lowerpath\n");
1742                         iput(trap);
1743                         goto out;
1744                 }
1745
1746                 /*
1747                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1748                  * will fail instead of modifying lower fs.
1749                  */
1750                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1751
1752                 layers[ofs->numlayer].trap = trap;
1753                 layers[ofs->numlayer].mnt = mnt;
1754                 layers[ofs->numlayer].idx = ofs->numlayer;
1755                 layers[ofs->numlayer].fsid = fsid;
1756                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1757                 ofs->numlayer++;
1758                 ofs->fs[fsid].is_lower = true;
1759         }
1760
1761         /*
1762          * When all layers on same fs, overlay can use real inode numbers.
1763          * With mount option "xino=<on|auto>", mounter declares that there are
1764          * enough free high bits in underlying fs to hold the unique fsid.
1765          * If overlayfs does encounter underlying inodes using the high xino
1766          * bits reserved for fsid, it emits a warning and uses the original
1767          * inode number or a non persistent inode number allocated from a
1768          * dedicated range.
1769          */
1770         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1771                 if (ofs->config.xino == OVL_XINO_ON)
1772                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1773                 ofs->xino_mode = 0;
1774         } else if (ofs->config.xino == OVL_XINO_OFF) {
1775                 ofs->xino_mode = -1;
1776         } else if (ofs->xino_mode < 0) {
1777                 /*
1778                  * This is a roundup of number of bits needed for encoding
1779                  * fsid, where fsid 0 is reserved for upper fs (even with
1780                  * lower only overlay) +1 extra bit is reserved for the non
1781                  * persistent inode number range that is used for resolving
1782                  * xino lower bits overflow.
1783                  */
1784                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1785                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1786         }
1787
1788         if (ofs->xino_mode > 0) {
1789                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1790                         ofs->xino_mode);
1791         }
1792
1793         err = 0;
1794 out:
1795         return err;
1796 }
1797
1798 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1799                                 const char *lower, unsigned int numlower,
1800                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1801 {
1802         int err;
1803         struct path *stack = NULL;
1804         unsigned int i;
1805         struct ovl_entry *oe;
1806
1807         if (!ofs->config.upperdir && numlower == 1) {
1808                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1809                 return ERR_PTR(-EINVAL);
1810         }
1811
1812         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1813         if (!stack)
1814                 return ERR_PTR(-ENOMEM);
1815
1816         err = -EINVAL;
1817         for (i = 0; i < numlower; i++) {
1818                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1819                 if (err)
1820                         goto out_err;
1821
1822                 lower = strchr(lower, '\0') + 1;
1823         }
1824
1825         err = -EINVAL;
1826         sb->s_stack_depth++;
1827         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1828                 pr_err("maximum fs stacking depth exceeded\n");
1829                 goto out_err;
1830         }
1831
1832         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1833         if (err)
1834                 goto out_err;
1835
1836         err = -ENOMEM;
1837         oe = ovl_alloc_entry(numlower);
1838         if (!oe)
1839                 goto out_err;
1840
1841         for (i = 0; i < numlower; i++) {
1842                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1843                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1844         }
1845
1846 out:
1847         for (i = 0; i < numlower; i++)
1848                 path_put(&stack[i]);
1849         kfree(stack);
1850
1851         return oe;
1852
1853 out_err:
1854         oe = ERR_PTR(err);
1855         goto out;
1856 }
1857
1858 /*
1859  * Check if this layer root is a descendant of:
1860  * - another layer of this overlayfs instance
1861  * - upper/work dir of any overlayfs instance
1862  */
1863 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1864                            struct dentry *dentry, const char *name,
1865                            bool is_lower)
1866 {
1867         struct dentry *next = dentry, *parent;
1868         int err = 0;
1869
1870         if (!dentry)
1871                 return 0;
1872
1873         parent = dget_parent(next);
1874
1875         /* Walk back ancestors to root (inclusive) looking for traps */
1876         while (!err && parent != next) {
1877                 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1878                         err = -ELOOP;
1879                         pr_err("overlapping %s path\n", name);
1880                 } else if (ovl_is_inuse(parent)) {
1881                         err = ovl_report_in_use(ofs, name);
1882                 }
1883                 next = parent;
1884                 parent = dget_parent(next);
1885                 dput(next);
1886         }
1887
1888         dput(parent);
1889
1890         return err;
1891 }
1892
1893 /*
1894  * Check if any of the layers or work dirs overlap.
1895  */
1896 static int ovl_check_overlapping_layers(struct super_block *sb,
1897                                         struct ovl_fs *ofs)
1898 {
1899         int i, err;
1900
1901         if (ovl_upper_mnt(ofs)) {
1902                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1903                                       "upperdir", false);
1904                 if (err)
1905                         return err;
1906
1907                 /*
1908                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1909                  * this instance and covers overlapping work and index dirs,
1910                  * unless work or index dir have been moved since created inside
1911                  * workbasedir.  In that case, we already have their traps in
1912                  * inode cache and we will catch that case on lookup.
1913                  */
1914                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1915                                       false);
1916                 if (err)
1917                         return err;
1918         }
1919
1920         for (i = 1; i < ofs->numlayer; i++) {
1921                 err = ovl_check_layer(sb, ofs,
1922                                       ofs->layers[i].mnt->mnt_root,
1923                                       "lowerdir", true);
1924                 if (err)
1925                         return err;
1926         }
1927
1928         return 0;
1929 }
1930
1931 static struct dentry *ovl_get_root(struct super_block *sb,
1932                                    struct dentry *upperdentry,
1933                                    struct ovl_entry *oe)
1934 {
1935         struct dentry *root;
1936         struct ovl_path *lowerpath = &oe->lowerstack[0];
1937         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1938         int fsid = lowerpath->layer->fsid;
1939         struct ovl_inode_params oip = {
1940                 .upperdentry = upperdentry,
1941                 .lowerpath = lowerpath,
1942         };
1943
1944         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1945         if (!root)
1946                 return NULL;
1947
1948         root->d_fsdata = oe;
1949
1950         if (upperdentry) {
1951                 /* Root inode uses upper st_ino/i_ino */
1952                 ino = d_inode(upperdentry)->i_ino;
1953                 fsid = 0;
1954                 ovl_dentry_set_upper_alias(root);
1955                 if (ovl_is_impuredir(sb, upperdentry))
1956                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1957         }
1958
1959         /* Root is always merge -> can have whiteouts */
1960         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1961         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1962         ovl_set_upperdata(d_inode(root));
1963         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1964         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1965
1966         return root;
1967 }
1968
1969 static bool ovl_has_idmapped_layers(struct ovl_fs *ofs)
1970 {
1971
1972         unsigned int i;
1973         const struct vfsmount *mnt;
1974
1975         for (i = 0; i < ofs->numlayer; i++) {
1976                 mnt = ofs->layers[i].mnt;
1977                 if (mnt && is_idmapped_mnt(mnt))
1978                         return true;
1979         }
1980         return false;
1981 }
1982
1983 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1984 {
1985         struct path upperpath = { };
1986         struct dentry *root_dentry;
1987         struct ovl_entry *oe;
1988         struct ovl_fs *ofs;
1989         struct ovl_layer *layers;
1990         struct cred *cred;
1991         char *splitlower = NULL;
1992         unsigned int numlower;
1993         int err;
1994
1995         err = -EIO;
1996         if (WARN_ON(sb->s_user_ns != current_user_ns()))
1997                 goto out;
1998
1999         sb->s_d_op = &ovl_dentry_operations;
2000
2001         err = -ENOMEM;
2002         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
2003         if (!ofs)
2004                 goto out;
2005
2006         err = -ENOMEM;
2007         ofs->creator_cred = cred = prepare_creds();
2008         if (!cred)
2009                 goto out_err;
2010
2011         /* Is there a reason anyone would want not to share whiteouts? */
2012         ofs->share_whiteout = true;
2013
2014         ofs->config.index = ovl_index_def;
2015         ofs->config.uuid = true;
2016         ofs->config.nfs_export = ovl_nfs_export_def;
2017         ofs->config.xino = ovl_xino_def();
2018         ofs->config.metacopy = ovl_metacopy_def;
2019         err = ovl_parse_opt((char *) data, &ofs->config);
2020         if (err)
2021                 goto out_err;
2022
2023         err = -EINVAL;
2024         if (!ofs->config.lowerdir) {
2025                 if (!silent)
2026                         pr_err("missing 'lowerdir'\n");
2027                 goto out_err;
2028         }
2029
2030         err = -ENOMEM;
2031         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
2032         if (!splitlower)
2033                 goto out_err;
2034
2035         err = -EINVAL;
2036         numlower = ovl_split_lowerdirs(splitlower);
2037         if (numlower > OVL_MAX_STACK) {
2038                 pr_err("too many lower directories, limit is %d\n",
2039                        OVL_MAX_STACK);
2040                 goto out_err;
2041         }
2042
2043         err = -ENOMEM;
2044         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
2045         if (!layers)
2046                 goto out_err;
2047
2048         ofs->layers = layers;
2049         /* Layer 0 is reserved for upper even if there's no upper */
2050         ofs->numlayer = 1;
2051
2052         sb->s_stack_depth = 0;
2053         sb->s_maxbytes = MAX_LFS_FILESIZE;
2054         atomic_long_set(&ofs->last_ino, 1);
2055         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
2056         if (ofs->config.xino != OVL_XINO_OFF) {
2057                 ofs->xino_mode = BITS_PER_LONG - 32;
2058                 if (!ofs->xino_mode) {
2059                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
2060                         ofs->config.xino = OVL_XINO_OFF;
2061                 }
2062         }
2063
2064         /* alloc/destroy_inode needed for setting up traps in inode cache */
2065         sb->s_op = &ovl_super_operations;
2066
2067         if (ofs->config.upperdir) {
2068                 struct super_block *upper_sb;
2069
2070                 err = -EINVAL;
2071                 if (!ofs->config.workdir) {
2072                         pr_err("missing 'workdir'\n");
2073                         goto out_err;
2074                 }
2075
2076                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
2077                 if (err)
2078                         goto out_err;
2079
2080                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
2081                 if (!ovl_should_sync(ofs)) {
2082                         ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
2083                         if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
2084                                 err = -EIO;
2085                                 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
2086                                 goto out_err;
2087                         }
2088                 }
2089
2090                 err = ovl_get_workdir(sb, ofs, &upperpath);
2091                 if (err)
2092                         goto out_err;
2093
2094                 if (!ofs->workdir)
2095                         sb->s_flags |= SB_RDONLY;
2096
2097                 sb->s_stack_depth = upper_sb->s_stack_depth;
2098                 sb->s_time_gran = upper_sb->s_time_gran;
2099         }
2100         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2101         err = PTR_ERR(oe);
2102         if (IS_ERR(oe))
2103                 goto out_err;
2104
2105         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2106         if (!ovl_upper_mnt(ofs))
2107                 sb->s_flags |= SB_RDONLY;
2108
2109         if (!ofs->config.uuid && ofs->numfs > 1) {
2110                 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2111                 ofs->config.uuid = true;
2112         }
2113
2114         if (!ovl_force_readonly(ofs) && ofs->config.index) {
2115                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2116                 if (err)
2117                         goto out_free_oe;
2118
2119                 /* Force r/o mount with no index dir */
2120                 if (!ofs->indexdir)
2121                         sb->s_flags |= SB_RDONLY;
2122         }
2123
2124         err = ovl_check_overlapping_layers(sb, ofs);
2125         if (err)
2126                 goto out_free_oe;
2127
2128         /* Show index=off in /proc/mounts for forced r/o mount */
2129         if (!ofs->indexdir) {
2130                 ofs->config.index = false;
2131                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2132                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2133                         ofs->config.nfs_export = false;
2134                 }
2135         }
2136
2137         if (ofs->config.metacopy && ofs->config.nfs_export) {
2138                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2139                 ofs->config.nfs_export = false;
2140         }
2141
2142         if (ofs->config.nfs_export)
2143                 sb->s_export_op = &ovl_export_operations;
2144
2145         /* Never override disk quota limits or use reserved space */
2146         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2147
2148         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2149         sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2150                 ovl_trusted_xattr_handlers;
2151         sb->s_fs_info = ofs;
2152         if (ovl_has_idmapped_layers(ofs))
2153                 pr_warn("POSIX ACLs are not yet supported with idmapped layers, mounting without ACL support.\n");
2154         else
2155                 sb->s_flags |= SB_POSIXACL;
2156         sb->s_iflags |= SB_I_SKIP_SYNC;
2157
2158         err = -ENOMEM;
2159         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2160         if (!root_dentry)
2161                 goto out_free_oe;
2162
2163         mntput(upperpath.mnt);
2164         kfree(splitlower);
2165
2166         sb->s_root = root_dentry;
2167
2168         return 0;
2169
2170 out_free_oe:
2171         ovl_entry_stack_free(oe);
2172         kfree(oe);
2173 out_err:
2174         kfree(splitlower);
2175         path_put(&upperpath);
2176         ovl_free_fs(ofs);
2177 out:
2178         return err;
2179 }
2180
2181 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2182                                 const char *dev_name, void *raw_data)
2183 {
2184         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2185 }
2186
2187 static struct file_system_type ovl_fs_type = {
2188         .owner          = THIS_MODULE,
2189         .name           = "overlay",
2190         .fs_flags       = FS_USERNS_MOUNT,
2191         .mount          = ovl_mount,
2192         .kill_sb        = kill_anon_super,
2193 };
2194 MODULE_ALIAS_FS("overlay");
2195
2196 static void ovl_inode_init_once(void *foo)
2197 {
2198         struct ovl_inode *oi = foo;
2199
2200         inode_init_once(&oi->vfs_inode);
2201 }
2202
2203 static int __init ovl_init(void)
2204 {
2205         int err;
2206
2207         ovl_inode_cachep = kmem_cache_create("ovl_inode",
2208                                              sizeof(struct ovl_inode), 0,
2209                                              (SLAB_RECLAIM_ACCOUNT|
2210                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2211                                              ovl_inode_init_once);
2212         if (ovl_inode_cachep == NULL)
2213                 return -ENOMEM;
2214
2215         err = ovl_aio_request_cache_init();
2216         if (!err) {
2217                 err = register_filesystem(&ovl_fs_type);
2218                 if (!err)
2219                         return 0;
2220
2221                 ovl_aio_request_cache_destroy();
2222         }
2223         kmem_cache_destroy(ovl_inode_cachep);
2224
2225         return err;
2226 }
2227
2228 static void __exit ovl_exit(void)
2229 {
2230         unregister_filesystem(&ovl_fs_type);
2231
2232         /*
2233          * Make sure all delayed rcu free inodes are flushed before we
2234          * destroy cache.
2235          */
2236         rcu_barrier();
2237         kmem_cache_destroy(ovl_inode_cachep);
2238         ovl_aio_request_cache_destroy();
2239 }
2240
2241 module_init(ovl_init);
2242 module_exit(ovl_exit);