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