1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2011 Novell Inc. 4 * Copyright (C) 2016 Red Hat, Inc. 5 */ 6 7 #include <linux/fs.h> 8 #include <linux/mount.h> 9 #include <linux/slab.h> 10 #include <linux/cred.h> 11 #include <linux/xattr.h> 12 #include <linux/exportfs.h> 13 #include <linux/fileattr.h> 14 #include <linux/uuid.h> 15 #include <linux/namei.h> 16 #include <linux/ratelimit.h> 17 #include "overlayfs.h" 18 19 int ovl_want_write(struct dentry *dentry) 20 { 21 struct ovl_fs *ofs = dentry->d_sb->s_fs_info; 22 return mnt_want_write(ovl_upper_mnt(ofs)); 23 } 24 25 void ovl_drop_write(struct dentry *dentry) 26 { 27 struct ovl_fs *ofs = dentry->d_sb->s_fs_info; 28 mnt_drop_write(ovl_upper_mnt(ofs)); 29 } 30 31 struct dentry *ovl_workdir(struct dentry *dentry) 32 { 33 struct ovl_fs *ofs = dentry->d_sb->s_fs_info; 34 return ofs->workdir; 35 } 36 37 const struct cred *ovl_override_creds(struct super_block *sb) 38 { 39 struct ovl_fs *ofs = sb->s_fs_info; 40 41 return override_creds(ofs->creator_cred); 42 } 43 44 /* 45 * Check if underlying fs supports file handles and try to determine encoding 46 * type, in order to deduce maximum inode number used by fs. 47 * 48 * Return 0 if file handles are not supported. 49 * Return 1 (FILEID_INO32_GEN) if fs uses the default 32bit inode encoding. 50 * Return -1 if fs uses a non default encoding with unknown inode size. 51 */ 52 int ovl_can_decode_fh(struct super_block *sb) 53 { 54 if (!capable(CAP_DAC_READ_SEARCH)) 55 return 0; 56 57 if (!sb->s_export_op || !sb->s_export_op->fh_to_dentry) 58 return 0; 59 60 return sb->s_export_op->encode_fh ? -1 : FILEID_INO32_GEN; 61 } 62 63 struct dentry *ovl_indexdir(struct super_block *sb) 64 { 65 struct ovl_fs *ofs = sb->s_fs_info; 66 67 return ofs->indexdir; 68 } 69 70 /* Index all files on copy up. For now only enabled for NFS export */ 71 bool ovl_index_all(struct super_block *sb) 72 { 73 struct ovl_fs *ofs = sb->s_fs_info; 74 75 return ofs->config.nfs_export && ofs->config.index; 76 } 77 78 /* Verify lower origin on lookup. For now only enabled for NFS export */ 79 bool ovl_verify_lower(struct super_block *sb) 80 { 81 struct ovl_fs *ofs = sb->s_fs_info; 82 83 return ofs->config.nfs_export && ofs->config.index; 84 } 85 86 struct ovl_path *ovl_stack_alloc(unsigned int n) 87 { 88 return kcalloc(n, sizeof(struct ovl_path), GFP_KERNEL); 89 } 90 91 void ovl_stack_cpy(struct ovl_path *dst, struct ovl_path *src, unsigned int n) 92 { 93 unsigned int i; 94 95 memcpy(dst, src, sizeof(struct ovl_path) * n); 96 for (i = 0; i < n; i++) 97 dget(src[i].dentry); 98 } 99 100 void ovl_stack_put(struct ovl_path *stack, unsigned int n) 101 { 102 unsigned int i; 103 104 for (i = 0; stack && i < n; i++) 105 dput(stack[i].dentry); 106 } 107 108 void ovl_stack_free(struct ovl_path *stack, unsigned int n) 109 { 110 ovl_stack_put(stack, n); 111 kfree(stack); 112 } 113 114 struct ovl_entry *ovl_alloc_entry(unsigned int numlower) 115 { 116 size_t size = offsetof(struct ovl_entry, __lowerstack[numlower]); 117 struct ovl_entry *oe = kzalloc(size, GFP_KERNEL); 118 119 if (oe) 120 oe->__numlower = numlower; 121 122 return oe; 123 } 124 125 void ovl_free_entry(struct ovl_entry *oe) 126 { 127 ovl_stack_put(ovl_lowerstack(oe), ovl_numlower(oe)); 128 kfree(oe); 129 } 130 131 #define OVL_D_REVALIDATE (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE) 132 133 bool ovl_dentry_remote(struct dentry *dentry) 134 { 135 return dentry->d_flags & OVL_D_REVALIDATE; 136 } 137 138 void ovl_dentry_update_reval(struct dentry *dentry, struct dentry *realdentry) 139 { 140 if (!ovl_dentry_remote(realdentry)) 141 return; 142 143 spin_lock(&dentry->d_lock); 144 dentry->d_flags |= realdentry->d_flags & OVL_D_REVALIDATE; 145 spin_unlock(&dentry->d_lock); 146 } 147 148 void ovl_dentry_init_reval(struct dentry *dentry, struct dentry *upperdentry, 149 struct ovl_entry *oe) 150 { 151 return ovl_dentry_init_flags(dentry, upperdentry, oe, OVL_D_REVALIDATE); 152 } 153 154 void ovl_dentry_init_flags(struct dentry *dentry, struct dentry *upperdentry, 155 struct ovl_entry *oe, unsigned int mask) 156 { 157 struct ovl_path *lowerstack = ovl_lowerstack(oe); 158 unsigned int i, flags = 0; 159 160 if (upperdentry) 161 flags |= upperdentry->d_flags; 162 for (i = 0; i < ovl_numlower(oe); i++) 163 flags |= lowerstack[i].dentry->d_flags; 164 165 spin_lock(&dentry->d_lock); 166 dentry->d_flags &= ~mask; 167 dentry->d_flags |= flags & mask; 168 spin_unlock(&dentry->d_lock); 169 } 170 171 bool ovl_dentry_weird(struct dentry *dentry) 172 { 173 return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT | 174 DCACHE_MANAGE_TRANSIT | 175 DCACHE_OP_HASH | 176 DCACHE_OP_COMPARE); 177 } 178 179 enum ovl_path_type ovl_path_type(struct dentry *dentry) 180 { 181 struct ovl_entry *oe = OVL_E(dentry); 182 enum ovl_path_type type = 0; 183 184 if (ovl_dentry_upper(dentry)) { 185 type = __OVL_PATH_UPPER; 186 187 /* 188 * Non-dir dentry can hold lower dentry of its copy up origin. 189 */ 190 if (ovl_numlower(oe)) { 191 if (ovl_test_flag(OVL_CONST_INO, d_inode(dentry))) 192 type |= __OVL_PATH_ORIGIN; 193 if (d_is_dir(dentry) || 194 !ovl_has_upperdata(d_inode(dentry))) 195 type |= __OVL_PATH_MERGE; 196 } 197 } else { 198 if (ovl_numlower(oe) > 1) 199 type |= __OVL_PATH_MERGE; 200 } 201 return type; 202 } 203 204 void ovl_path_upper(struct dentry *dentry, struct path *path) 205 { 206 struct ovl_fs *ofs = dentry->d_sb->s_fs_info; 207 208 path->mnt = ovl_upper_mnt(ofs); 209 path->dentry = ovl_dentry_upper(dentry); 210 } 211 212 void ovl_path_lower(struct dentry *dentry, struct path *path) 213 { 214 struct ovl_entry *oe = OVL_E(dentry); 215 struct ovl_path *lowerpath = ovl_lowerstack(oe); 216 217 if (ovl_numlower(oe)) { 218 path->mnt = lowerpath->layer->mnt; 219 path->dentry = lowerpath->dentry; 220 } else { 221 *path = (struct path) { }; 222 } 223 } 224 225 void ovl_path_lowerdata(struct dentry *dentry, struct path *path) 226 { 227 struct ovl_entry *oe = OVL_E(dentry); 228 struct ovl_path *lowerstack = ovl_lowerstack(oe); 229 230 if (ovl_numlower(oe)) { 231 path->mnt = lowerstack[ovl_numlower(oe) - 1].layer->mnt; 232 path->dentry = lowerstack[ovl_numlower(oe) - 1].dentry; 233 } else { 234 *path = (struct path) { }; 235 } 236 } 237 238 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path) 239 { 240 enum ovl_path_type type = ovl_path_type(dentry); 241 242 if (!OVL_TYPE_UPPER(type)) 243 ovl_path_lower(dentry, path); 244 else 245 ovl_path_upper(dentry, path); 246 247 return type; 248 } 249 250 enum ovl_path_type ovl_path_realdata(struct dentry *dentry, struct path *path) 251 { 252 enum ovl_path_type type = ovl_path_type(dentry); 253 254 WARN_ON_ONCE(d_is_dir(dentry)); 255 256 if (!OVL_TYPE_UPPER(type) || OVL_TYPE_MERGE(type)) 257 ovl_path_lowerdata(dentry, path); 258 else 259 ovl_path_upper(dentry, path); 260 261 return type; 262 } 263 264 struct dentry *ovl_dentry_upper(struct dentry *dentry) 265 { 266 return ovl_upperdentry_dereference(OVL_I(d_inode(dentry))); 267 } 268 269 struct dentry *ovl_dentry_lower(struct dentry *dentry) 270 { 271 struct ovl_entry *oe = OVL_E(dentry); 272 273 return ovl_numlower(oe) ? ovl_lowerstack(oe)->dentry : NULL; 274 } 275 276 const struct ovl_layer *ovl_layer_lower(struct dentry *dentry) 277 { 278 struct ovl_entry *oe = OVL_E(dentry); 279 280 return ovl_numlower(oe) ? ovl_lowerstack(oe)->layer : NULL; 281 } 282 283 /* 284 * ovl_dentry_lower() could return either a data dentry or metacopy dentry 285 * depending on what is stored in lowerstack[0]. At times we need to find 286 * lower dentry which has data (and not metacopy dentry). This helper 287 * returns the lower data dentry. 288 */ 289 struct dentry *ovl_dentry_lowerdata(struct dentry *dentry) 290 { 291 struct ovl_entry *oe = OVL_E(dentry); 292 293 return ovl_numlower(oe) ? 294 ovl_lowerstack(oe)[ovl_numlower(oe) - 1].dentry : NULL; 295 } 296 297 struct dentry *ovl_dentry_real(struct dentry *dentry) 298 { 299 return ovl_dentry_upper(dentry) ?: ovl_dentry_lower(dentry); 300 } 301 302 struct dentry *ovl_i_dentry_upper(struct inode *inode) 303 { 304 return ovl_upperdentry_dereference(OVL_I(inode)); 305 } 306 307 struct inode *ovl_i_path_real(struct inode *inode, struct path *path) 308 { 309 struct ovl_path *lowerpath = ovl_lowerpath(OVL_I_E(inode)); 310 311 path->dentry = ovl_i_dentry_upper(inode); 312 if (!path->dentry) { 313 path->dentry = lowerpath->dentry; 314 path->mnt = lowerpath->layer->mnt; 315 } else { 316 path->mnt = ovl_upper_mnt(OVL_FS(inode->i_sb)); 317 } 318 319 return path->dentry ? d_inode_rcu(path->dentry) : NULL; 320 } 321 322 struct inode *ovl_inode_upper(struct inode *inode) 323 { 324 struct dentry *upperdentry = ovl_i_dentry_upper(inode); 325 326 return upperdentry ? d_inode(upperdentry) : NULL; 327 } 328 329 struct inode *ovl_inode_lower(struct inode *inode) 330 { 331 struct ovl_path *lowerpath = ovl_lowerpath(OVL_I_E(inode)); 332 333 return lowerpath ? d_inode(lowerpath->dentry) : NULL; 334 } 335 336 struct inode *ovl_inode_real(struct inode *inode) 337 { 338 return ovl_inode_upper(inode) ?: ovl_inode_lower(inode); 339 } 340 341 /* Return inode which contains lower data. Do not return metacopy */ 342 struct inode *ovl_inode_lowerdata(struct inode *inode) 343 { 344 if (WARN_ON(!S_ISREG(inode->i_mode))) 345 return NULL; 346 347 return OVL_I(inode)->lowerdata ?: ovl_inode_lower(inode); 348 } 349 350 /* Return real inode which contains data. Does not return metacopy inode */ 351 struct inode *ovl_inode_realdata(struct inode *inode) 352 { 353 struct inode *upperinode; 354 355 upperinode = ovl_inode_upper(inode); 356 if (upperinode && ovl_has_upperdata(inode)) 357 return upperinode; 358 359 return ovl_inode_lowerdata(inode); 360 } 361 362 struct ovl_dir_cache *ovl_dir_cache(struct inode *inode) 363 { 364 return OVL_I(inode)->cache; 365 } 366 367 void ovl_set_dir_cache(struct inode *inode, struct ovl_dir_cache *cache) 368 { 369 OVL_I(inode)->cache = cache; 370 } 371 372 void ovl_dentry_set_flag(unsigned long flag, struct dentry *dentry) 373 { 374 set_bit(flag, OVL_E_FLAGS(dentry)); 375 } 376 377 void ovl_dentry_clear_flag(unsigned long flag, struct dentry *dentry) 378 { 379 clear_bit(flag, OVL_E_FLAGS(dentry)); 380 } 381 382 bool ovl_dentry_test_flag(unsigned long flag, struct dentry *dentry) 383 { 384 return test_bit(flag, OVL_E_FLAGS(dentry)); 385 } 386 387 bool ovl_dentry_is_opaque(struct dentry *dentry) 388 { 389 return ovl_dentry_test_flag(OVL_E_OPAQUE, dentry); 390 } 391 392 bool ovl_dentry_is_whiteout(struct dentry *dentry) 393 { 394 return !dentry->d_inode && ovl_dentry_is_opaque(dentry); 395 } 396 397 void ovl_dentry_set_opaque(struct dentry *dentry) 398 { 399 ovl_dentry_set_flag(OVL_E_OPAQUE, dentry); 400 } 401 402 /* 403 * For hard links and decoded file handles, it's possible for ovl_dentry_upper() 404 * to return positive, while there's no actual upper alias for the inode. 405 * Copy up code needs to know about the existence of the upper alias, so it 406 * can't use ovl_dentry_upper(). 407 */ 408 bool ovl_dentry_has_upper_alias(struct dentry *dentry) 409 { 410 return ovl_dentry_test_flag(OVL_E_UPPER_ALIAS, dentry); 411 } 412 413 void ovl_dentry_set_upper_alias(struct dentry *dentry) 414 { 415 ovl_dentry_set_flag(OVL_E_UPPER_ALIAS, dentry); 416 } 417 418 static bool ovl_should_check_upperdata(struct inode *inode) 419 { 420 if (!S_ISREG(inode->i_mode)) 421 return false; 422 423 if (!ovl_inode_lower(inode)) 424 return false; 425 426 return true; 427 } 428 429 bool ovl_has_upperdata(struct inode *inode) 430 { 431 if (!ovl_should_check_upperdata(inode)) 432 return true; 433 434 if (!ovl_test_flag(OVL_UPPERDATA, inode)) 435 return false; 436 /* 437 * Pairs with smp_wmb() in ovl_set_upperdata(). Main user of 438 * ovl_has_upperdata() is ovl_copy_up_meta_inode_data(). Make sure 439 * if setting of OVL_UPPERDATA is visible, then effects of writes 440 * before that are visible too. 441 */ 442 smp_rmb(); 443 return true; 444 } 445 446 void ovl_set_upperdata(struct inode *inode) 447 { 448 /* 449 * Pairs with smp_rmb() in ovl_has_upperdata(). Make sure 450 * if OVL_UPPERDATA flag is visible, then effects of write operations 451 * before it are visible as well. 452 */ 453 smp_wmb(); 454 ovl_set_flag(OVL_UPPERDATA, inode); 455 } 456 457 /* Caller should hold ovl_inode->lock */ 458 bool ovl_dentry_needs_data_copy_up_locked(struct dentry *dentry, int flags) 459 { 460 if (!ovl_open_flags_need_copy_up(flags)) 461 return false; 462 463 return !ovl_test_flag(OVL_UPPERDATA, d_inode(dentry)); 464 } 465 466 bool ovl_dentry_needs_data_copy_up(struct dentry *dentry, int flags) 467 { 468 if (!ovl_open_flags_need_copy_up(flags)) 469 return false; 470 471 return !ovl_has_upperdata(d_inode(dentry)); 472 } 473 474 bool ovl_redirect_dir(struct super_block *sb) 475 { 476 struct ovl_fs *ofs = sb->s_fs_info; 477 478 return ofs->config.redirect_dir && !ofs->noxattr; 479 } 480 481 const char *ovl_dentry_get_redirect(struct dentry *dentry) 482 { 483 return OVL_I(d_inode(dentry))->redirect; 484 } 485 486 void ovl_dentry_set_redirect(struct dentry *dentry, const char *redirect) 487 { 488 struct ovl_inode *oi = OVL_I(d_inode(dentry)); 489 490 kfree(oi->redirect); 491 oi->redirect = redirect; 492 } 493 494 void ovl_inode_update(struct inode *inode, struct dentry *upperdentry) 495 { 496 struct inode *upperinode = d_inode(upperdentry); 497 498 WARN_ON(OVL_I(inode)->__upperdentry); 499 500 /* 501 * Make sure upperdentry is consistent before making it visible 502 */ 503 smp_wmb(); 504 OVL_I(inode)->__upperdentry = upperdentry; 505 if (inode_unhashed(inode)) { 506 inode->i_private = upperinode; 507 __insert_inode_hash(inode, (unsigned long) upperinode); 508 } 509 } 510 511 static void ovl_dir_version_inc(struct dentry *dentry, bool impurity) 512 { 513 struct inode *inode = d_inode(dentry); 514 515 WARN_ON(!inode_is_locked(inode)); 516 WARN_ON(!d_is_dir(dentry)); 517 /* 518 * Version is used by readdir code to keep cache consistent. 519 * For merge dirs (or dirs with origin) all changes need to be noted. 520 * For non-merge dirs, cache contains only impure entries (i.e. ones 521 * which have been copied up and have origins), so only need to note 522 * changes to impure entries. 523 */ 524 if (!ovl_dir_is_real(inode) || impurity) 525 OVL_I(inode)->version++; 526 } 527 528 void ovl_dir_modified(struct dentry *dentry, bool impurity) 529 { 530 /* Copy mtime/ctime */ 531 ovl_copyattr(d_inode(dentry)); 532 533 ovl_dir_version_inc(dentry, impurity); 534 } 535 536 u64 ovl_inode_version_get(struct inode *inode) 537 { 538 WARN_ON(!inode_is_locked(inode)); 539 return OVL_I(inode)->version; 540 } 541 542 bool ovl_is_whiteout(struct dentry *dentry) 543 { 544 struct inode *inode = dentry->d_inode; 545 546 return inode && IS_WHITEOUT(inode); 547 } 548 549 struct file *ovl_path_open(const struct path *path, int flags) 550 { 551 struct inode *inode = d_inode(path->dentry); 552 struct mnt_idmap *real_idmap = mnt_idmap(path->mnt); 553 int err, acc_mode; 554 555 if (flags & ~(O_ACCMODE | O_LARGEFILE)) 556 BUG(); 557 558 switch (flags & O_ACCMODE) { 559 case O_RDONLY: 560 acc_mode = MAY_READ; 561 break; 562 case O_WRONLY: 563 acc_mode = MAY_WRITE; 564 break; 565 default: 566 BUG(); 567 } 568 569 err = inode_permission(real_idmap, inode, acc_mode | MAY_OPEN); 570 if (err) 571 return ERR_PTR(err); 572 573 /* O_NOATIME is an optimization, don't fail if not permitted */ 574 if (inode_owner_or_capable(real_idmap, inode)) 575 flags |= O_NOATIME; 576 577 return dentry_open(path, flags, current_cred()); 578 } 579 580 /* Caller should hold ovl_inode->lock */ 581 static bool ovl_already_copied_up_locked(struct dentry *dentry, int flags) 582 { 583 bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED; 584 585 if (ovl_dentry_upper(dentry) && 586 (ovl_dentry_has_upper_alias(dentry) || disconnected) && 587 !ovl_dentry_needs_data_copy_up_locked(dentry, flags)) 588 return true; 589 590 return false; 591 } 592 593 bool ovl_already_copied_up(struct dentry *dentry, int flags) 594 { 595 bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED; 596 597 /* 598 * Check if copy-up has happened as well as for upper alias (in 599 * case of hard links) is there. 600 * 601 * Both checks are lockless: 602 * - false negatives: will recheck under oi->lock 603 * - false positives: 604 * + ovl_dentry_upper() uses memory barriers to ensure the 605 * upper dentry is up-to-date 606 * + ovl_dentry_has_upper_alias() relies on locking of 607 * upper parent i_rwsem to prevent reordering copy-up 608 * with rename. 609 */ 610 if (ovl_dentry_upper(dentry) && 611 (ovl_dentry_has_upper_alias(dentry) || disconnected) && 612 !ovl_dentry_needs_data_copy_up(dentry, flags)) 613 return true; 614 615 return false; 616 } 617 618 int ovl_copy_up_start(struct dentry *dentry, int flags) 619 { 620 struct inode *inode = d_inode(dentry); 621 int err; 622 623 err = ovl_inode_lock_interruptible(inode); 624 if (!err && ovl_already_copied_up_locked(dentry, flags)) { 625 err = 1; /* Already copied up */ 626 ovl_inode_unlock(inode); 627 } 628 629 return err; 630 } 631 632 void ovl_copy_up_end(struct dentry *dentry) 633 { 634 ovl_inode_unlock(d_inode(dentry)); 635 } 636 637 bool ovl_path_check_origin_xattr(struct ovl_fs *ofs, const struct path *path) 638 { 639 int res; 640 641 res = ovl_path_getxattr(ofs, path, OVL_XATTR_ORIGIN, NULL, 0); 642 643 /* Zero size value means "copied up but origin unknown" */ 644 if (res >= 0) 645 return true; 646 647 return false; 648 } 649 650 bool ovl_path_check_dir_xattr(struct ovl_fs *ofs, const struct path *path, 651 enum ovl_xattr ox) 652 { 653 int res; 654 char val; 655 656 if (!d_is_dir(path->dentry)) 657 return false; 658 659 res = ovl_path_getxattr(ofs, path, ox, &val, 1); 660 if (res == 1 && val == 'y') 661 return true; 662 663 return false; 664 } 665 666 #define OVL_XATTR_OPAQUE_POSTFIX "opaque" 667 #define OVL_XATTR_REDIRECT_POSTFIX "redirect" 668 #define OVL_XATTR_ORIGIN_POSTFIX "origin" 669 #define OVL_XATTR_IMPURE_POSTFIX "impure" 670 #define OVL_XATTR_NLINK_POSTFIX "nlink" 671 #define OVL_XATTR_UPPER_POSTFIX "upper" 672 #define OVL_XATTR_METACOPY_POSTFIX "metacopy" 673 #define OVL_XATTR_PROTATTR_POSTFIX "protattr" 674 675 #define OVL_XATTR_TAB_ENTRY(x) \ 676 [x] = { [false] = OVL_XATTR_TRUSTED_PREFIX x ## _POSTFIX, \ 677 [true] = OVL_XATTR_USER_PREFIX x ## _POSTFIX } 678 679 const char *const ovl_xattr_table[][2] = { 680 OVL_XATTR_TAB_ENTRY(OVL_XATTR_OPAQUE), 681 OVL_XATTR_TAB_ENTRY(OVL_XATTR_REDIRECT), 682 OVL_XATTR_TAB_ENTRY(OVL_XATTR_ORIGIN), 683 OVL_XATTR_TAB_ENTRY(OVL_XATTR_IMPURE), 684 OVL_XATTR_TAB_ENTRY(OVL_XATTR_NLINK), 685 OVL_XATTR_TAB_ENTRY(OVL_XATTR_UPPER), 686 OVL_XATTR_TAB_ENTRY(OVL_XATTR_METACOPY), 687 OVL_XATTR_TAB_ENTRY(OVL_XATTR_PROTATTR), 688 }; 689 690 int ovl_check_setxattr(struct ovl_fs *ofs, struct dentry *upperdentry, 691 enum ovl_xattr ox, const void *value, size_t size, 692 int xerr) 693 { 694 int err; 695 696 if (ofs->noxattr) 697 return xerr; 698 699 err = ovl_setxattr(ofs, upperdentry, ox, value, size); 700 701 if (err == -EOPNOTSUPP) { 702 pr_warn("cannot set %s xattr on upper\n", ovl_xattr(ofs, ox)); 703 ofs->noxattr = true; 704 return xerr; 705 } 706 707 return err; 708 } 709 710 int ovl_set_impure(struct dentry *dentry, struct dentry *upperdentry) 711 { 712 struct ovl_fs *ofs = OVL_FS(dentry->d_sb); 713 int err; 714 715 if (ovl_test_flag(OVL_IMPURE, d_inode(dentry))) 716 return 0; 717 718 /* 719 * Do not fail when upper doesn't support xattrs. 720 * Upper inodes won't have origin nor redirect xattr anyway. 721 */ 722 err = ovl_check_setxattr(ofs, upperdentry, OVL_XATTR_IMPURE, "y", 1, 0); 723 if (!err) 724 ovl_set_flag(OVL_IMPURE, d_inode(dentry)); 725 726 return err; 727 } 728 729 730 #define OVL_PROTATTR_MAX 32 /* Reserved for future flags */ 731 732 void ovl_check_protattr(struct inode *inode, struct dentry *upper) 733 { 734 struct ovl_fs *ofs = OVL_FS(inode->i_sb); 735 u32 iflags = inode->i_flags & OVL_PROT_I_FLAGS_MASK; 736 char buf[OVL_PROTATTR_MAX+1]; 737 int res, n; 738 739 res = ovl_getxattr_upper(ofs, upper, OVL_XATTR_PROTATTR, buf, 740 OVL_PROTATTR_MAX); 741 if (res < 0) 742 return; 743 744 /* 745 * Initialize inode flags from overlay.protattr xattr and upper inode 746 * flags. If upper inode has those fileattr flags set (i.e. from old 747 * kernel), we do not clear them on ovl_get_inode(), but we will clear 748 * them on next fileattr_set(). 749 */ 750 for (n = 0; n < res; n++) { 751 if (buf[n] == 'a') 752 iflags |= S_APPEND; 753 else if (buf[n] == 'i') 754 iflags |= S_IMMUTABLE; 755 else 756 break; 757 } 758 759 if (!res || n < res) { 760 pr_warn_ratelimited("incompatible overlay.protattr format (%pd2, len=%d)\n", 761 upper, res); 762 } else { 763 inode_set_flags(inode, iflags, OVL_PROT_I_FLAGS_MASK); 764 } 765 } 766 767 int ovl_set_protattr(struct inode *inode, struct dentry *upper, 768 struct fileattr *fa) 769 { 770 struct ovl_fs *ofs = OVL_FS(inode->i_sb); 771 char buf[OVL_PROTATTR_MAX]; 772 int len = 0, err = 0; 773 u32 iflags = 0; 774 775 BUILD_BUG_ON(HWEIGHT32(OVL_PROT_FS_FLAGS_MASK) > OVL_PROTATTR_MAX); 776 777 if (fa->flags & FS_APPEND_FL) { 778 buf[len++] = 'a'; 779 iflags |= S_APPEND; 780 } 781 if (fa->flags & FS_IMMUTABLE_FL) { 782 buf[len++] = 'i'; 783 iflags |= S_IMMUTABLE; 784 } 785 786 /* 787 * Do not allow to set protection flags when upper doesn't support 788 * xattrs, because we do not set those fileattr flags on upper inode. 789 * Remove xattr if it exist and all protection flags are cleared. 790 */ 791 if (len) { 792 err = ovl_check_setxattr(ofs, upper, OVL_XATTR_PROTATTR, 793 buf, len, -EPERM); 794 } else if (inode->i_flags & OVL_PROT_I_FLAGS_MASK) { 795 err = ovl_removexattr(ofs, upper, OVL_XATTR_PROTATTR); 796 if (err == -EOPNOTSUPP || err == -ENODATA) 797 err = 0; 798 } 799 if (err) 800 return err; 801 802 inode_set_flags(inode, iflags, OVL_PROT_I_FLAGS_MASK); 803 804 /* Mask out the fileattr flags that should not be set in upper inode */ 805 fa->flags &= ~OVL_PROT_FS_FLAGS_MASK; 806 fa->fsx_xflags &= ~OVL_PROT_FSX_FLAGS_MASK; 807 808 return 0; 809 } 810 811 /** 812 * Caller must hold a reference to inode to prevent it from being freed while 813 * it is marked inuse. 814 */ 815 bool ovl_inuse_trylock(struct dentry *dentry) 816 { 817 struct inode *inode = d_inode(dentry); 818 bool locked = false; 819 820 spin_lock(&inode->i_lock); 821 if (!(inode->i_state & I_OVL_INUSE)) { 822 inode->i_state |= I_OVL_INUSE; 823 locked = true; 824 } 825 spin_unlock(&inode->i_lock); 826 827 return locked; 828 } 829 830 void ovl_inuse_unlock(struct dentry *dentry) 831 { 832 if (dentry) { 833 struct inode *inode = d_inode(dentry); 834 835 spin_lock(&inode->i_lock); 836 WARN_ON(!(inode->i_state & I_OVL_INUSE)); 837 inode->i_state &= ~I_OVL_INUSE; 838 spin_unlock(&inode->i_lock); 839 } 840 } 841 842 bool ovl_is_inuse(struct dentry *dentry) 843 { 844 struct inode *inode = d_inode(dentry); 845 bool inuse; 846 847 spin_lock(&inode->i_lock); 848 inuse = (inode->i_state & I_OVL_INUSE); 849 spin_unlock(&inode->i_lock); 850 851 return inuse; 852 } 853 854 /* 855 * Does this overlay dentry need to be indexed on copy up? 856 */ 857 bool ovl_need_index(struct dentry *dentry) 858 { 859 struct dentry *lower = ovl_dentry_lower(dentry); 860 861 if (!lower || !ovl_indexdir(dentry->d_sb)) 862 return false; 863 864 /* Index all files for NFS export and consistency verification */ 865 if (ovl_index_all(dentry->d_sb)) 866 return true; 867 868 /* Index only lower hardlinks on copy up */ 869 if (!d_is_dir(lower) && d_inode(lower)->i_nlink > 1) 870 return true; 871 872 return false; 873 } 874 875 /* Caller must hold OVL_I(inode)->lock */ 876 static void ovl_cleanup_index(struct dentry *dentry) 877 { 878 struct ovl_fs *ofs = OVL_FS(dentry->d_sb); 879 struct dentry *indexdir = ovl_indexdir(dentry->d_sb); 880 struct inode *dir = indexdir->d_inode; 881 struct dentry *lowerdentry = ovl_dentry_lower(dentry); 882 struct dentry *upperdentry = ovl_dentry_upper(dentry); 883 struct dentry *index = NULL; 884 struct inode *inode; 885 struct qstr name = { }; 886 int err; 887 888 err = ovl_get_index_name(ofs, lowerdentry, &name); 889 if (err) 890 goto fail; 891 892 inode = d_inode(upperdentry); 893 if (!S_ISDIR(inode->i_mode) && inode->i_nlink != 1) { 894 pr_warn_ratelimited("cleanup linked index (%pd2, ino=%lu, nlink=%u)\n", 895 upperdentry, inode->i_ino, inode->i_nlink); 896 /* 897 * We either have a bug with persistent union nlink or a lower 898 * hardlink was added while overlay is mounted. Adding a lower 899 * hardlink and then unlinking all overlay hardlinks would drop 900 * overlay nlink to zero before all upper inodes are unlinked. 901 * As a safety measure, when that situation is detected, set 902 * the overlay nlink to the index inode nlink minus one for the 903 * index entry itself. 904 */ 905 set_nlink(d_inode(dentry), inode->i_nlink - 1); 906 ovl_set_nlink_upper(dentry); 907 goto out; 908 } 909 910 inode_lock_nested(dir, I_MUTEX_PARENT); 911 index = ovl_lookup_upper(ofs, name.name, indexdir, name.len); 912 err = PTR_ERR(index); 913 if (IS_ERR(index)) { 914 index = NULL; 915 } else if (ovl_index_all(dentry->d_sb)) { 916 /* Whiteout orphan index to block future open by handle */ 917 err = ovl_cleanup_and_whiteout(OVL_FS(dentry->d_sb), 918 dir, index); 919 } else { 920 /* Cleanup orphan index entries */ 921 err = ovl_cleanup(ofs, dir, index); 922 } 923 924 inode_unlock(dir); 925 if (err) 926 goto fail; 927 928 out: 929 kfree(name.name); 930 dput(index); 931 return; 932 933 fail: 934 pr_err("cleanup index of '%pd2' failed (%i)\n", dentry, err); 935 goto out; 936 } 937 938 /* 939 * Operations that change overlay inode and upper inode nlink need to be 940 * synchronized with copy up for persistent nlink accounting. 941 */ 942 int ovl_nlink_start(struct dentry *dentry) 943 { 944 struct inode *inode = d_inode(dentry); 945 const struct cred *old_cred; 946 int err; 947 948 if (WARN_ON(!inode)) 949 return -ENOENT; 950 951 /* 952 * With inodes index is enabled, we store the union overlay nlink 953 * in an xattr on the index inode. When whiting out an indexed lower, 954 * we need to decrement the overlay persistent nlink, but before the 955 * first copy up, we have no upper index inode to store the xattr. 956 * 957 * As a workaround, before whiteout/rename over an indexed lower, 958 * copy up to create the upper index. Creating the upper index will 959 * initialize the overlay nlink, so it could be dropped if unlink 960 * or rename succeeds. 961 * 962 * TODO: implement metadata only index copy up when called with 963 * ovl_copy_up_flags(dentry, O_PATH). 964 */ 965 if (ovl_need_index(dentry) && !ovl_dentry_has_upper_alias(dentry)) { 966 err = ovl_copy_up(dentry); 967 if (err) 968 return err; 969 } 970 971 err = ovl_inode_lock_interruptible(inode); 972 if (err) 973 return err; 974 975 if (d_is_dir(dentry) || !ovl_test_flag(OVL_INDEX, inode)) 976 goto out; 977 978 old_cred = ovl_override_creds(dentry->d_sb); 979 /* 980 * The overlay inode nlink should be incremented/decremented IFF the 981 * upper operation succeeds, along with nlink change of upper inode. 982 * Therefore, before link/unlink/rename, we store the union nlink 983 * value relative to the upper inode nlink in an upper inode xattr. 984 */ 985 err = ovl_set_nlink_upper(dentry); 986 revert_creds(old_cred); 987 988 out: 989 if (err) 990 ovl_inode_unlock(inode); 991 992 return err; 993 } 994 995 void ovl_nlink_end(struct dentry *dentry) 996 { 997 struct inode *inode = d_inode(dentry); 998 999 if (ovl_test_flag(OVL_INDEX, inode) && inode->i_nlink == 0) { 1000 const struct cred *old_cred; 1001 1002 old_cred = ovl_override_creds(dentry->d_sb); 1003 ovl_cleanup_index(dentry); 1004 revert_creds(old_cred); 1005 } 1006 1007 ovl_inode_unlock(inode); 1008 } 1009 1010 int ovl_lock_rename_workdir(struct dentry *workdir, struct dentry *upperdir) 1011 { 1012 /* Workdir should not be the same as upperdir */ 1013 if (workdir == upperdir) 1014 goto err; 1015 1016 /* Workdir should not be subdir of upperdir and vice versa */ 1017 if (lock_rename(workdir, upperdir) != NULL) 1018 goto err_unlock; 1019 1020 return 0; 1021 1022 err_unlock: 1023 unlock_rename(workdir, upperdir); 1024 err: 1025 pr_err("failed to lock workdir+upperdir\n"); 1026 return -EIO; 1027 } 1028 1029 /* err < 0, 0 if no metacopy xattr, 1 if metacopy xattr found */ 1030 int ovl_check_metacopy_xattr(struct ovl_fs *ofs, const struct path *path) 1031 { 1032 int res; 1033 1034 /* Only regular files can have metacopy xattr */ 1035 if (!S_ISREG(d_inode(path->dentry)->i_mode)) 1036 return 0; 1037 1038 res = ovl_path_getxattr(ofs, path, OVL_XATTR_METACOPY, NULL, 0); 1039 if (res < 0) { 1040 if (res == -ENODATA || res == -EOPNOTSUPP) 1041 return 0; 1042 /* 1043 * getxattr on user.* may fail with EACCES in case there's no 1044 * read permission on the inode. Not much we can do, other than 1045 * tell the caller that this is not a metacopy inode. 1046 */ 1047 if (ofs->config.userxattr && res == -EACCES) 1048 return 0; 1049 goto out; 1050 } 1051 1052 return 1; 1053 out: 1054 pr_warn_ratelimited("failed to get metacopy (%i)\n", res); 1055 return res; 1056 } 1057 1058 bool ovl_is_metacopy_dentry(struct dentry *dentry) 1059 { 1060 struct ovl_entry *oe = OVL_E(dentry); 1061 1062 if (!d_is_reg(dentry)) 1063 return false; 1064 1065 if (ovl_dentry_upper(dentry)) { 1066 if (!ovl_has_upperdata(d_inode(dentry))) 1067 return true; 1068 return false; 1069 } 1070 1071 return (ovl_numlower(oe) > 1); 1072 } 1073 1074 char *ovl_get_redirect_xattr(struct ovl_fs *ofs, const struct path *path, int padding) 1075 { 1076 int res; 1077 char *s, *next, *buf = NULL; 1078 1079 res = ovl_path_getxattr(ofs, path, OVL_XATTR_REDIRECT, NULL, 0); 1080 if (res == -ENODATA || res == -EOPNOTSUPP) 1081 return NULL; 1082 if (res < 0) 1083 goto fail; 1084 if (res == 0) 1085 goto invalid; 1086 1087 buf = kzalloc(res + padding + 1, GFP_KERNEL); 1088 if (!buf) 1089 return ERR_PTR(-ENOMEM); 1090 1091 res = ovl_path_getxattr(ofs, path, OVL_XATTR_REDIRECT, buf, res); 1092 if (res < 0) 1093 goto fail; 1094 if (res == 0) 1095 goto invalid; 1096 1097 if (buf[0] == '/') { 1098 for (s = buf; *s++ == '/'; s = next) { 1099 next = strchrnul(s, '/'); 1100 if (s == next) 1101 goto invalid; 1102 } 1103 } else { 1104 if (strchr(buf, '/') != NULL) 1105 goto invalid; 1106 } 1107 1108 return buf; 1109 invalid: 1110 pr_warn_ratelimited("invalid redirect (%s)\n", buf); 1111 res = -EINVAL; 1112 goto err_free; 1113 fail: 1114 pr_warn_ratelimited("failed to get redirect (%i)\n", res); 1115 err_free: 1116 kfree(buf); 1117 return ERR_PTR(res); 1118 } 1119 1120 /* 1121 * ovl_sync_status() - Check fs sync status for volatile mounts 1122 * 1123 * Returns 1 if this is not a volatile mount and a real sync is required. 1124 * 1125 * Returns 0 if syncing can be skipped because mount is volatile, and no errors 1126 * have occurred on the upperdir since the mount. 1127 * 1128 * Returns -errno if it is a volatile mount, and the error that occurred since 1129 * the last mount. If the error code changes, it'll return the latest error 1130 * code. 1131 */ 1132 1133 int ovl_sync_status(struct ovl_fs *ofs) 1134 { 1135 struct vfsmount *mnt; 1136 1137 if (ovl_should_sync(ofs)) 1138 return 1; 1139 1140 mnt = ovl_upper_mnt(ofs); 1141 if (!mnt) 1142 return 0; 1143 1144 return errseq_check(&mnt->mnt_sb->s_wb_err, ofs->errseq); 1145 } 1146 1147 /* 1148 * ovl_copyattr() - copy inode attributes from layer to ovl inode 1149 * 1150 * When overlay copies inode information from an upper or lower layer to the 1151 * relevant overlay inode it will apply the idmapping of the upper or lower 1152 * layer when doing so ensuring that the ovl inode ownership will correctly 1153 * reflect the ownership of the idmapped upper or lower layer. For example, an 1154 * idmapped upper or lower layer mapping id 1001 to id 1000 will take care to 1155 * map any lower or upper inode owned by id 1001 to id 1000. These mapping 1156 * helpers are nops when the relevant layer isn't idmapped. 1157 */ 1158 void ovl_copyattr(struct inode *inode) 1159 { 1160 struct path realpath; 1161 struct inode *realinode; 1162 struct mnt_idmap *real_idmap; 1163 vfsuid_t vfsuid; 1164 vfsgid_t vfsgid; 1165 1166 realinode = ovl_i_path_real(inode, &realpath); 1167 real_idmap = mnt_idmap(realpath.mnt); 1168 1169 vfsuid = i_uid_into_vfsuid(real_idmap, realinode); 1170 vfsgid = i_gid_into_vfsgid(real_idmap, realinode); 1171 1172 inode->i_uid = vfsuid_into_kuid(vfsuid); 1173 inode->i_gid = vfsgid_into_kgid(vfsgid); 1174 inode->i_mode = realinode->i_mode; 1175 inode->i_atime = realinode->i_atime; 1176 inode->i_mtime = realinode->i_mtime; 1177 inode->i_ctime = realinode->i_ctime; 1178 i_size_write(inode, i_size_read(realinode)); 1179 } 1180