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