fs/overlayfs/super.c

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