1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext4/namei.c 4 * 5 * Copyright (C) 1992, 1993, 1994, 1995 6 * Remy Card (card@masi.ibp.fr) 7 * Laboratoire MASI - Institut Blaise Pascal 8 * Universite Pierre et Marie Curie (Paris VI) 9 * 10 * from 11 * 12 * linux/fs/minix/namei.c 13 * 14 * Copyright (C) 1991, 1992 Linus Torvalds 15 * 16 * Big-endian to little-endian byte-swapping/bitmaps by 17 * David S. Miller (davem@caip.rutgers.edu), 1995 18 * Directory entry file type support and forward compatibility hooks 19 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998 20 * Hash Tree Directory indexing (c) 21 * Daniel Phillips, 2001 22 * Hash Tree Directory indexing porting 23 * Christopher Li, 2002 24 * Hash Tree Directory indexing cleanup 25 * Theodore Ts'o, 2002 26 */ 27 28 #include <linux/fs.h> 29 #include <linux/pagemap.h> 30 #include <linux/time.h> 31 #include <linux/fcntl.h> 32 #include <linux/stat.h> 33 #include <linux/string.h> 34 #include <linux/quotaops.h> 35 #include <linux/buffer_head.h> 36 #include <linux/bio.h> 37 #include <linux/iversion.h> 38 #include "ext4.h" 39 #include "ext4_jbd2.h" 40 41 #include "xattr.h" 42 #include "acl.h" 43 44 #include <trace/events/ext4.h> 45 /* 46 * define how far ahead to read directories while searching them. 47 */ 48 #define NAMEI_RA_CHUNKS 2 49 #define NAMEI_RA_BLOCKS 4 50 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS) 51 52 static struct buffer_head *ext4_append(handle_t *handle, 53 struct inode *inode, 54 ext4_lblk_t *block) 55 { 56 struct buffer_head *bh; 57 int err; 58 59 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb && 60 ((inode->i_size >> 10) >= 61 EXT4_SB(inode->i_sb)->s_max_dir_size_kb))) 62 return ERR_PTR(-ENOSPC); 63 64 *block = inode->i_size >> inode->i_sb->s_blocksize_bits; 65 66 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE); 67 if (IS_ERR(bh)) 68 return bh; 69 inode->i_size += inode->i_sb->s_blocksize; 70 EXT4_I(inode)->i_disksize = inode->i_size; 71 BUFFER_TRACE(bh, "get_write_access"); 72 err = ext4_journal_get_write_access(handle, bh); 73 if (err) { 74 brelse(bh); 75 ext4_std_error(inode->i_sb, err); 76 return ERR_PTR(err); 77 } 78 return bh; 79 } 80 81 static int ext4_dx_csum_verify(struct inode *inode, 82 struct ext4_dir_entry *dirent); 83 84 typedef enum { 85 EITHER, INDEX, DIRENT 86 } dirblock_type_t; 87 88 #define ext4_read_dirblock(inode, block, type) \ 89 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__) 90 91 static struct buffer_head *__ext4_read_dirblock(struct inode *inode, 92 ext4_lblk_t block, 93 dirblock_type_t type, 94 const char *func, 95 unsigned int line) 96 { 97 struct buffer_head *bh; 98 struct ext4_dir_entry *dirent; 99 int is_dx_block = 0; 100 101 bh = ext4_bread(NULL, inode, block, 0); 102 if (IS_ERR(bh)) { 103 __ext4_warning(inode->i_sb, func, line, 104 "inode #%lu: lblock %lu: comm %s: " 105 "error %ld reading directory block", 106 inode->i_ino, (unsigned long)block, 107 current->comm, PTR_ERR(bh)); 108 109 return bh; 110 } 111 if (!bh) { 112 ext4_error_inode(inode, func, line, block, 113 "Directory hole found"); 114 return ERR_PTR(-EFSCORRUPTED); 115 } 116 dirent = (struct ext4_dir_entry *) bh->b_data; 117 /* Determine whether or not we have an index block */ 118 if (is_dx(inode)) { 119 if (block == 0) 120 is_dx_block = 1; 121 else if (ext4_rec_len_from_disk(dirent->rec_len, 122 inode->i_sb->s_blocksize) == 123 inode->i_sb->s_blocksize) 124 is_dx_block = 1; 125 } 126 if (!is_dx_block && type == INDEX) { 127 ext4_error_inode(inode, func, line, block, 128 "directory leaf block found instead of index block"); 129 return ERR_PTR(-EFSCORRUPTED); 130 } 131 if (!ext4_has_metadata_csum(inode->i_sb) || 132 buffer_verified(bh)) 133 return bh; 134 135 /* 136 * An empty leaf block can get mistaken for a index block; for 137 * this reason, we can only check the index checksum when the 138 * caller is sure it should be an index block. 139 */ 140 if (is_dx_block && type == INDEX) { 141 if (ext4_dx_csum_verify(inode, dirent)) 142 set_buffer_verified(bh); 143 else { 144 ext4_error_inode(inode, func, line, block, 145 "Directory index failed checksum"); 146 brelse(bh); 147 return ERR_PTR(-EFSBADCRC); 148 } 149 } 150 if (!is_dx_block) { 151 if (ext4_dirent_csum_verify(inode, dirent)) 152 set_buffer_verified(bh); 153 else { 154 ext4_error_inode(inode, func, line, block, 155 "Directory block failed checksum"); 156 brelse(bh); 157 return ERR_PTR(-EFSBADCRC); 158 } 159 } 160 return bh; 161 } 162 163 #ifndef assert 164 #define assert(test) J_ASSERT(test) 165 #endif 166 167 #ifdef DX_DEBUG 168 #define dxtrace(command) command 169 #else 170 #define dxtrace(command) 171 #endif 172 173 struct fake_dirent 174 { 175 __le32 inode; 176 __le16 rec_len; 177 u8 name_len; 178 u8 file_type; 179 }; 180 181 struct dx_countlimit 182 { 183 __le16 limit; 184 __le16 count; 185 }; 186 187 struct dx_entry 188 { 189 __le32 hash; 190 __le32 block; 191 }; 192 193 /* 194 * dx_root_info is laid out so that if it should somehow get overlaid by a 195 * dirent the two low bits of the hash version will be zero. Therefore, the 196 * hash version mod 4 should never be 0. Sincerely, the paranoia department. 197 */ 198 199 struct dx_root 200 { 201 struct fake_dirent dot; 202 char dot_name[4]; 203 struct fake_dirent dotdot; 204 char dotdot_name[4]; 205 struct dx_root_info 206 { 207 __le32 reserved_zero; 208 u8 hash_version; 209 u8 info_length; /* 8 */ 210 u8 indirect_levels; 211 u8 unused_flags; 212 } 213 info; 214 struct dx_entry entries[0]; 215 }; 216 217 struct dx_node 218 { 219 struct fake_dirent fake; 220 struct dx_entry entries[0]; 221 }; 222 223 224 struct dx_frame 225 { 226 struct buffer_head *bh; 227 struct dx_entry *entries; 228 struct dx_entry *at; 229 }; 230 231 struct dx_map_entry 232 { 233 u32 hash; 234 u16 offs; 235 u16 size; 236 }; 237 238 /* 239 * This goes at the end of each htree block. 240 */ 241 struct dx_tail { 242 u32 dt_reserved; 243 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */ 244 }; 245 246 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry); 247 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value); 248 static inline unsigned dx_get_hash(struct dx_entry *entry); 249 static void dx_set_hash(struct dx_entry *entry, unsigned value); 250 static unsigned dx_get_count(struct dx_entry *entries); 251 static unsigned dx_get_limit(struct dx_entry *entries); 252 static void dx_set_count(struct dx_entry *entries, unsigned value); 253 static void dx_set_limit(struct dx_entry *entries, unsigned value); 254 static unsigned dx_root_limit(struct inode *dir, unsigned infosize); 255 static unsigned dx_node_limit(struct inode *dir); 256 static struct dx_frame *dx_probe(struct ext4_filename *fname, 257 struct inode *dir, 258 struct dx_hash_info *hinfo, 259 struct dx_frame *frame); 260 static void dx_release(struct dx_frame *frames); 261 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de, 262 unsigned blocksize, struct dx_hash_info *hinfo, 263 struct dx_map_entry map[]); 264 static void dx_sort_map(struct dx_map_entry *map, unsigned count); 265 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to, 266 struct dx_map_entry *offsets, int count, unsigned blocksize); 267 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize); 268 static void dx_insert_block(struct dx_frame *frame, 269 u32 hash, ext4_lblk_t block); 270 static int ext4_htree_next_block(struct inode *dir, __u32 hash, 271 struct dx_frame *frame, 272 struct dx_frame *frames, 273 __u32 *start_hash); 274 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, 275 struct ext4_filename *fname, 276 struct ext4_dir_entry_2 **res_dir); 277 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname, 278 struct inode *dir, struct inode *inode); 279 280 /* checksumming functions */ 281 void initialize_dirent_tail(struct ext4_dir_entry_tail *t, 282 unsigned int blocksize) 283 { 284 memset(t, 0, sizeof(struct ext4_dir_entry_tail)); 285 t->det_rec_len = ext4_rec_len_to_disk( 286 sizeof(struct ext4_dir_entry_tail), blocksize); 287 t->det_reserved_ft = EXT4_FT_DIR_CSUM; 288 } 289 290 /* Walk through a dirent block to find a checksum "dirent" at the tail */ 291 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode, 292 struct ext4_dir_entry *de) 293 { 294 struct ext4_dir_entry_tail *t; 295 296 #ifdef PARANOID 297 struct ext4_dir_entry *d, *top; 298 299 d = de; 300 top = (struct ext4_dir_entry *)(((void *)de) + 301 (EXT4_BLOCK_SIZE(inode->i_sb) - 302 sizeof(struct ext4_dir_entry_tail))); 303 while (d < top && d->rec_len) 304 d = (struct ext4_dir_entry *)(((void *)d) + 305 le16_to_cpu(d->rec_len)); 306 307 if (d != top) 308 return NULL; 309 310 t = (struct ext4_dir_entry_tail *)d; 311 #else 312 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb)); 313 #endif 314 315 if (t->det_reserved_zero1 || 316 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) || 317 t->det_reserved_zero2 || 318 t->det_reserved_ft != EXT4_FT_DIR_CSUM) 319 return NULL; 320 321 return t; 322 } 323 324 static __le32 ext4_dirent_csum(struct inode *inode, 325 struct ext4_dir_entry *dirent, int size) 326 { 327 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 328 struct ext4_inode_info *ei = EXT4_I(inode); 329 __u32 csum; 330 331 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size); 332 return cpu_to_le32(csum); 333 } 334 335 #define warn_no_space_for_csum(inode) \ 336 __warn_no_space_for_csum((inode), __func__, __LINE__) 337 338 static void __warn_no_space_for_csum(struct inode *inode, const char *func, 339 unsigned int line) 340 { 341 __ext4_warning_inode(inode, func, line, 342 "No space for directory leaf checksum. Please run e2fsck -D."); 343 } 344 345 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent) 346 { 347 struct ext4_dir_entry_tail *t; 348 349 if (!ext4_has_metadata_csum(inode->i_sb)) 350 return 1; 351 352 t = get_dirent_tail(inode, dirent); 353 if (!t) { 354 warn_no_space_for_csum(inode); 355 return 0; 356 } 357 358 if (t->det_checksum != ext4_dirent_csum(inode, dirent, 359 (void *)t - (void *)dirent)) 360 return 0; 361 362 return 1; 363 } 364 365 static void ext4_dirent_csum_set(struct inode *inode, 366 struct ext4_dir_entry *dirent) 367 { 368 struct ext4_dir_entry_tail *t; 369 370 if (!ext4_has_metadata_csum(inode->i_sb)) 371 return; 372 373 t = get_dirent_tail(inode, dirent); 374 if (!t) { 375 warn_no_space_for_csum(inode); 376 return; 377 } 378 379 t->det_checksum = ext4_dirent_csum(inode, dirent, 380 (void *)t - (void *)dirent); 381 } 382 383 int ext4_handle_dirty_dirent_node(handle_t *handle, 384 struct inode *inode, 385 struct buffer_head *bh) 386 { 387 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data); 388 return ext4_handle_dirty_metadata(handle, inode, bh); 389 } 390 391 static struct dx_countlimit *get_dx_countlimit(struct inode *inode, 392 struct ext4_dir_entry *dirent, 393 int *offset) 394 { 395 struct ext4_dir_entry *dp; 396 struct dx_root_info *root; 397 int count_offset; 398 399 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb)) 400 count_offset = 8; 401 else if (le16_to_cpu(dirent->rec_len) == 12) { 402 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12); 403 if (le16_to_cpu(dp->rec_len) != 404 EXT4_BLOCK_SIZE(inode->i_sb) - 12) 405 return NULL; 406 root = (struct dx_root_info *)(((void *)dp + 12)); 407 if (root->reserved_zero || 408 root->info_length != sizeof(struct dx_root_info)) 409 return NULL; 410 count_offset = 32; 411 } else 412 return NULL; 413 414 if (offset) 415 *offset = count_offset; 416 return (struct dx_countlimit *)(((void *)dirent) + count_offset); 417 } 418 419 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent, 420 int count_offset, int count, struct dx_tail *t) 421 { 422 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 423 struct ext4_inode_info *ei = EXT4_I(inode); 424 __u32 csum; 425 int size; 426 __u32 dummy_csum = 0; 427 int offset = offsetof(struct dx_tail, dt_checksum); 428 429 size = count_offset + (count * sizeof(struct dx_entry)); 430 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size); 431 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset); 432 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum)); 433 434 return cpu_to_le32(csum); 435 } 436 437 static int ext4_dx_csum_verify(struct inode *inode, 438 struct ext4_dir_entry *dirent) 439 { 440 struct dx_countlimit *c; 441 struct dx_tail *t; 442 int count_offset, limit, count; 443 444 if (!ext4_has_metadata_csum(inode->i_sb)) 445 return 1; 446 447 c = get_dx_countlimit(inode, dirent, &count_offset); 448 if (!c) { 449 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D."); 450 return 0; 451 } 452 limit = le16_to_cpu(c->limit); 453 count = le16_to_cpu(c->count); 454 if (count_offset + (limit * sizeof(struct dx_entry)) > 455 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) { 456 warn_no_space_for_csum(inode); 457 return 0; 458 } 459 t = (struct dx_tail *)(((struct dx_entry *)c) + limit); 460 461 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset, 462 count, t)) 463 return 0; 464 return 1; 465 } 466 467 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent) 468 { 469 struct dx_countlimit *c; 470 struct dx_tail *t; 471 int count_offset, limit, count; 472 473 if (!ext4_has_metadata_csum(inode->i_sb)) 474 return; 475 476 c = get_dx_countlimit(inode, dirent, &count_offset); 477 if (!c) { 478 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D."); 479 return; 480 } 481 limit = le16_to_cpu(c->limit); 482 count = le16_to_cpu(c->count); 483 if (count_offset + (limit * sizeof(struct dx_entry)) > 484 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) { 485 warn_no_space_for_csum(inode); 486 return; 487 } 488 t = (struct dx_tail *)(((struct dx_entry *)c) + limit); 489 490 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t); 491 } 492 493 static inline int ext4_handle_dirty_dx_node(handle_t *handle, 494 struct inode *inode, 495 struct buffer_head *bh) 496 { 497 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data); 498 return ext4_handle_dirty_metadata(handle, inode, bh); 499 } 500 501 /* 502 * p is at least 6 bytes before the end of page 503 */ 504 static inline struct ext4_dir_entry_2 * 505 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize) 506 { 507 return (struct ext4_dir_entry_2 *)((char *)p + 508 ext4_rec_len_from_disk(p->rec_len, blocksize)); 509 } 510 511 /* 512 * Future: use high four bits of block for coalesce-on-delete flags 513 * Mask them off for now. 514 */ 515 516 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry) 517 { 518 return le32_to_cpu(entry->block) & 0x0fffffff; 519 } 520 521 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value) 522 { 523 entry->block = cpu_to_le32(value); 524 } 525 526 static inline unsigned dx_get_hash(struct dx_entry *entry) 527 { 528 return le32_to_cpu(entry->hash); 529 } 530 531 static inline void dx_set_hash(struct dx_entry *entry, unsigned value) 532 { 533 entry->hash = cpu_to_le32(value); 534 } 535 536 static inline unsigned dx_get_count(struct dx_entry *entries) 537 { 538 return le16_to_cpu(((struct dx_countlimit *) entries)->count); 539 } 540 541 static inline unsigned dx_get_limit(struct dx_entry *entries) 542 { 543 return le16_to_cpu(((struct dx_countlimit *) entries)->limit); 544 } 545 546 static inline void dx_set_count(struct dx_entry *entries, unsigned value) 547 { 548 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value); 549 } 550 551 static inline void dx_set_limit(struct dx_entry *entries, unsigned value) 552 { 553 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value); 554 } 555 556 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize) 557 { 558 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) - 559 EXT4_DIR_REC_LEN(2) - infosize; 560 561 if (ext4_has_metadata_csum(dir->i_sb)) 562 entry_space -= sizeof(struct dx_tail); 563 return entry_space / sizeof(struct dx_entry); 564 } 565 566 static inline unsigned dx_node_limit(struct inode *dir) 567 { 568 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0); 569 570 if (ext4_has_metadata_csum(dir->i_sb)) 571 entry_space -= sizeof(struct dx_tail); 572 return entry_space / sizeof(struct dx_entry); 573 } 574 575 /* 576 * Debug 577 */ 578 #ifdef DX_DEBUG 579 static void dx_show_index(char * label, struct dx_entry *entries) 580 { 581 int i, n = dx_get_count (entries); 582 printk(KERN_DEBUG "%s index", label); 583 for (i = 0; i < n; i++) { 584 printk(KERN_CONT " %x->%lu", 585 i ? dx_get_hash(entries + i) : 0, 586 (unsigned long)dx_get_block(entries + i)); 587 } 588 printk(KERN_CONT "\n"); 589 } 590 591 struct stats 592 { 593 unsigned names; 594 unsigned space; 595 unsigned bcount; 596 }; 597 598 static struct stats dx_show_leaf(struct inode *dir, 599 struct dx_hash_info *hinfo, 600 struct ext4_dir_entry_2 *de, 601 int size, int show_names) 602 { 603 unsigned names = 0, space = 0; 604 char *base = (char *) de; 605 struct dx_hash_info h = *hinfo; 606 607 printk("names: "); 608 while ((char *) de < base + size) 609 { 610 if (de->inode) 611 { 612 if (show_names) 613 { 614 #ifdef CONFIG_EXT4_FS_ENCRYPTION 615 int len; 616 char *name; 617 struct fscrypt_str fname_crypto_str = 618 FSTR_INIT(NULL, 0); 619 int res = 0; 620 621 name = de->name; 622 len = de->name_len; 623 if (ext4_encrypted_inode(dir)) 624 res = fscrypt_get_encryption_info(dir); 625 if (res) { 626 printk(KERN_WARNING "Error setting up" 627 " fname crypto: %d\n", res); 628 } 629 if (!fscrypt_has_encryption_key(dir)) { 630 /* Directory is not encrypted */ 631 ext4fs_dirhash(de->name, 632 de->name_len, &h); 633 printk("%*.s:(U)%x.%u ", len, 634 name, h.hash, 635 (unsigned) ((char *) de 636 - base)); 637 } else { 638 struct fscrypt_str de_name = 639 FSTR_INIT(name, len); 640 641 /* Directory is encrypted */ 642 res = fscrypt_fname_alloc_buffer( 643 dir, len, 644 &fname_crypto_str); 645 if (res) 646 printk(KERN_WARNING "Error " 647 "allocating crypto " 648 "buffer--skipping " 649 "crypto\n"); 650 res = fscrypt_fname_disk_to_usr(dir, 651 0, 0, &de_name, 652 &fname_crypto_str); 653 if (res) { 654 printk(KERN_WARNING "Error " 655 "converting filename " 656 "from disk to usr" 657 "\n"); 658 name = "??"; 659 len = 2; 660 } else { 661 name = fname_crypto_str.name; 662 len = fname_crypto_str.len; 663 } 664 ext4fs_dirhash(de->name, de->name_len, 665 &h); 666 printk("%*.s:(E)%x.%u ", len, name, 667 h.hash, (unsigned) ((char *) de 668 - base)); 669 fscrypt_fname_free_buffer( 670 &fname_crypto_str); 671 } 672 #else 673 int len = de->name_len; 674 char *name = de->name; 675 ext4fs_dirhash(de->name, de->name_len, &h); 676 printk("%*.s:%x.%u ", len, name, h.hash, 677 (unsigned) ((char *) de - base)); 678 #endif 679 } 680 space += EXT4_DIR_REC_LEN(de->name_len); 681 names++; 682 } 683 de = ext4_next_entry(de, size); 684 } 685 printk(KERN_CONT "(%i)\n", names); 686 return (struct stats) { names, space, 1 }; 687 } 688 689 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir, 690 struct dx_entry *entries, int levels) 691 { 692 unsigned blocksize = dir->i_sb->s_blocksize; 693 unsigned count = dx_get_count(entries), names = 0, space = 0, i; 694 unsigned bcount = 0; 695 struct buffer_head *bh; 696 printk("%i indexed blocks...\n", count); 697 for (i = 0; i < count; i++, entries++) 698 { 699 ext4_lblk_t block = dx_get_block(entries); 700 ext4_lblk_t hash = i ? dx_get_hash(entries): 0; 701 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash; 702 struct stats stats; 703 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range); 704 bh = ext4_bread(NULL,dir, block, 0); 705 if (!bh || IS_ERR(bh)) 706 continue; 707 stats = levels? 708 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1): 709 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) 710 bh->b_data, blocksize, 0); 711 names += stats.names; 712 space += stats.space; 713 bcount += stats.bcount; 714 brelse(bh); 715 } 716 if (bcount) 717 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n", 718 levels ? "" : " ", names, space/bcount, 719 (space/bcount)*100/blocksize); 720 return (struct stats) { names, space, bcount}; 721 } 722 #endif /* DX_DEBUG */ 723 724 /* 725 * Probe for a directory leaf block to search. 726 * 727 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format 728 * error in the directory index, and the caller should fall back to 729 * searching the directory normally. The callers of dx_probe **MUST** 730 * check for this error code, and make sure it never gets reflected 731 * back to userspace. 732 */ 733 static struct dx_frame * 734 dx_probe(struct ext4_filename *fname, struct inode *dir, 735 struct dx_hash_info *hinfo, struct dx_frame *frame_in) 736 { 737 unsigned count, indirect; 738 struct dx_entry *at, *entries, *p, *q, *m; 739 struct dx_root *root; 740 struct dx_frame *frame = frame_in; 741 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR); 742 u32 hash; 743 744 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0])); 745 frame->bh = ext4_read_dirblock(dir, 0, INDEX); 746 if (IS_ERR(frame->bh)) 747 return (struct dx_frame *) frame->bh; 748 749 root = (struct dx_root *) frame->bh->b_data; 750 if (root->info.hash_version != DX_HASH_TEA && 751 root->info.hash_version != DX_HASH_HALF_MD4 && 752 root->info.hash_version != DX_HASH_LEGACY) { 753 ext4_warning_inode(dir, "Unrecognised inode hash code %u", 754 root->info.hash_version); 755 goto fail; 756 } 757 if (fname) 758 hinfo = &fname->hinfo; 759 hinfo->hash_version = root->info.hash_version; 760 if (hinfo->hash_version <= DX_HASH_TEA) 761 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; 762 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed; 763 if (fname && fname_name(fname)) 764 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo); 765 hash = hinfo->hash; 766 767 if (root->info.unused_flags & 1) { 768 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x", 769 root->info.unused_flags); 770 goto fail; 771 } 772 773 indirect = root->info.indirect_levels; 774 if (indirect >= ext4_dir_htree_level(dir->i_sb)) { 775 ext4_warning(dir->i_sb, 776 "Directory (ino: %lu) htree depth %#06x exceed" 777 "supported value", dir->i_ino, 778 ext4_dir_htree_level(dir->i_sb)); 779 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) { 780 ext4_warning(dir->i_sb, "Enable large directory " 781 "feature to access it"); 782 } 783 goto fail; 784 } 785 786 entries = (struct dx_entry *)(((char *)&root->info) + 787 root->info.info_length); 788 789 if (dx_get_limit(entries) != dx_root_limit(dir, 790 root->info.info_length)) { 791 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u", 792 dx_get_limit(entries), 793 dx_root_limit(dir, root->info.info_length)); 794 goto fail; 795 } 796 797 dxtrace(printk("Look up %x", hash)); 798 while (1) { 799 count = dx_get_count(entries); 800 if (!count || count > dx_get_limit(entries)) { 801 ext4_warning_inode(dir, 802 "dx entry: count %u beyond limit %u", 803 count, dx_get_limit(entries)); 804 goto fail; 805 } 806 807 p = entries + 1; 808 q = entries + count - 1; 809 while (p <= q) { 810 m = p + (q - p) / 2; 811 dxtrace(printk(KERN_CONT ".")); 812 if (dx_get_hash(m) > hash) 813 q = m - 1; 814 else 815 p = m + 1; 816 } 817 818 if (0) { // linear search cross check 819 unsigned n = count - 1; 820 at = entries; 821 while (n--) 822 { 823 dxtrace(printk(KERN_CONT ",")); 824 if (dx_get_hash(++at) > hash) 825 { 826 at--; 827 break; 828 } 829 } 830 assert (at == p - 1); 831 } 832 833 at = p - 1; 834 dxtrace(printk(KERN_CONT " %x->%u\n", 835 at == entries ? 0 : dx_get_hash(at), 836 dx_get_block(at))); 837 frame->entries = entries; 838 frame->at = at; 839 if (!indirect--) 840 return frame; 841 frame++; 842 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX); 843 if (IS_ERR(frame->bh)) { 844 ret_err = (struct dx_frame *) frame->bh; 845 frame->bh = NULL; 846 goto fail; 847 } 848 entries = ((struct dx_node *) frame->bh->b_data)->entries; 849 850 if (dx_get_limit(entries) != dx_node_limit(dir)) { 851 ext4_warning_inode(dir, 852 "dx entry: limit %u != node limit %u", 853 dx_get_limit(entries), dx_node_limit(dir)); 854 goto fail; 855 } 856 } 857 fail: 858 while (frame >= frame_in) { 859 brelse(frame->bh); 860 frame--; 861 } 862 863 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR)) 864 ext4_warning_inode(dir, 865 "Corrupt directory, running e2fsck is recommended"); 866 return ret_err; 867 } 868 869 static void dx_release(struct dx_frame *frames) 870 { 871 struct dx_root_info *info; 872 int i; 873 874 if (frames[0].bh == NULL) 875 return; 876 877 info = &((struct dx_root *)frames[0].bh->b_data)->info; 878 for (i = 0; i <= info->indirect_levels; i++) { 879 if (frames[i].bh == NULL) 880 break; 881 brelse(frames[i].bh); 882 frames[i].bh = NULL; 883 } 884 } 885 886 /* 887 * This function increments the frame pointer to search the next leaf 888 * block, and reads in the necessary intervening nodes if the search 889 * should be necessary. Whether or not the search is necessary is 890 * controlled by the hash parameter. If the hash value is even, then 891 * the search is only continued if the next block starts with that 892 * hash value. This is used if we are searching for a specific file. 893 * 894 * If the hash value is HASH_NB_ALWAYS, then always go to the next block. 895 * 896 * This function returns 1 if the caller should continue to search, 897 * or 0 if it should not. If there is an error reading one of the 898 * index blocks, it will a negative error code. 899 * 900 * If start_hash is non-null, it will be filled in with the starting 901 * hash of the next page. 902 */ 903 static int ext4_htree_next_block(struct inode *dir, __u32 hash, 904 struct dx_frame *frame, 905 struct dx_frame *frames, 906 __u32 *start_hash) 907 { 908 struct dx_frame *p; 909 struct buffer_head *bh; 910 int num_frames = 0; 911 __u32 bhash; 912 913 p = frame; 914 /* 915 * Find the next leaf page by incrementing the frame pointer. 916 * If we run out of entries in the interior node, loop around and 917 * increment pointer in the parent node. When we break out of 918 * this loop, num_frames indicates the number of interior 919 * nodes need to be read. 920 */ 921 while (1) { 922 if (++(p->at) < p->entries + dx_get_count(p->entries)) 923 break; 924 if (p == frames) 925 return 0; 926 num_frames++; 927 p--; 928 } 929 930 /* 931 * If the hash is 1, then continue only if the next page has a 932 * continuation hash of any value. This is used for readdir 933 * handling. Otherwise, check to see if the hash matches the 934 * desired contiuation hash. If it doesn't, return since 935 * there's no point to read in the successive index pages. 936 */ 937 bhash = dx_get_hash(p->at); 938 if (start_hash) 939 *start_hash = bhash; 940 if ((hash & 1) == 0) { 941 if ((bhash & ~1) != hash) 942 return 0; 943 } 944 /* 945 * If the hash is HASH_NB_ALWAYS, we always go to the next 946 * block so no check is necessary 947 */ 948 while (num_frames--) { 949 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX); 950 if (IS_ERR(bh)) 951 return PTR_ERR(bh); 952 p++; 953 brelse(p->bh); 954 p->bh = bh; 955 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries; 956 } 957 return 1; 958 } 959 960 961 /* 962 * This function fills a red-black tree with information from a 963 * directory block. It returns the number directory entries loaded 964 * into the tree. If there is an error it is returned in err. 965 */ 966 static int htree_dirblock_to_tree(struct file *dir_file, 967 struct inode *dir, ext4_lblk_t block, 968 struct dx_hash_info *hinfo, 969 __u32 start_hash, __u32 start_minor_hash) 970 { 971 struct buffer_head *bh; 972 struct ext4_dir_entry_2 *de, *top; 973 int err = 0, count = 0; 974 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str; 975 976 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n", 977 (unsigned long)block)); 978 bh = ext4_read_dirblock(dir, block, DIRENT); 979 if (IS_ERR(bh)) 980 return PTR_ERR(bh); 981 982 de = (struct ext4_dir_entry_2 *) bh->b_data; 983 top = (struct ext4_dir_entry_2 *) ((char *) de + 984 dir->i_sb->s_blocksize - 985 EXT4_DIR_REC_LEN(0)); 986 #ifdef CONFIG_EXT4_FS_ENCRYPTION 987 /* Check if the directory is encrypted */ 988 if (ext4_encrypted_inode(dir)) { 989 err = fscrypt_get_encryption_info(dir); 990 if (err < 0) { 991 brelse(bh); 992 return err; 993 } 994 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN, 995 &fname_crypto_str); 996 if (err < 0) { 997 brelse(bh); 998 return err; 999 } 1000 } 1001 #endif 1002 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) { 1003 if (ext4_check_dir_entry(dir, NULL, de, bh, 1004 bh->b_data, bh->b_size, 1005 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb)) 1006 + ((char *)de - bh->b_data))) { 1007 /* silently ignore the rest of the block */ 1008 break; 1009 } 1010 ext4fs_dirhash(de->name, de->name_len, hinfo); 1011 if ((hinfo->hash < start_hash) || 1012 ((hinfo->hash == start_hash) && 1013 (hinfo->minor_hash < start_minor_hash))) 1014 continue; 1015 if (de->inode == 0) 1016 continue; 1017 if (!ext4_encrypted_inode(dir)) { 1018 tmp_str.name = de->name; 1019 tmp_str.len = de->name_len; 1020 err = ext4_htree_store_dirent(dir_file, 1021 hinfo->hash, hinfo->minor_hash, de, 1022 &tmp_str); 1023 } else { 1024 int save_len = fname_crypto_str.len; 1025 struct fscrypt_str de_name = FSTR_INIT(de->name, 1026 de->name_len); 1027 1028 /* Directory is encrypted */ 1029 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash, 1030 hinfo->minor_hash, &de_name, 1031 &fname_crypto_str); 1032 if (err) { 1033 count = err; 1034 goto errout; 1035 } 1036 err = ext4_htree_store_dirent(dir_file, 1037 hinfo->hash, hinfo->minor_hash, de, 1038 &fname_crypto_str); 1039 fname_crypto_str.len = save_len; 1040 } 1041 if (err != 0) { 1042 count = err; 1043 goto errout; 1044 } 1045 count++; 1046 } 1047 errout: 1048 brelse(bh); 1049 #ifdef CONFIG_EXT4_FS_ENCRYPTION 1050 fscrypt_fname_free_buffer(&fname_crypto_str); 1051 #endif 1052 return count; 1053 } 1054 1055 1056 /* 1057 * This function fills a red-black tree with information from a 1058 * directory. We start scanning the directory in hash order, starting 1059 * at start_hash and start_minor_hash. 1060 * 1061 * This function returns the number of entries inserted into the tree, 1062 * or a negative error code. 1063 */ 1064 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash, 1065 __u32 start_minor_hash, __u32 *next_hash) 1066 { 1067 struct dx_hash_info hinfo; 1068 struct ext4_dir_entry_2 *de; 1069 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; 1070 struct inode *dir; 1071 ext4_lblk_t block; 1072 int count = 0; 1073 int ret, err; 1074 __u32 hashval; 1075 struct fscrypt_str tmp_str; 1076 1077 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n", 1078 start_hash, start_minor_hash)); 1079 dir = file_inode(dir_file); 1080 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) { 1081 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; 1082 if (hinfo.hash_version <= DX_HASH_TEA) 1083 hinfo.hash_version += 1084 EXT4_SB(dir->i_sb)->s_hash_unsigned; 1085 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; 1086 if (ext4_has_inline_data(dir)) { 1087 int has_inline_data = 1; 1088 count = htree_inlinedir_to_tree(dir_file, dir, 0, 1089 &hinfo, start_hash, 1090 start_minor_hash, 1091 &has_inline_data); 1092 if (has_inline_data) { 1093 *next_hash = ~0; 1094 return count; 1095 } 1096 } 1097 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo, 1098 start_hash, start_minor_hash); 1099 *next_hash = ~0; 1100 return count; 1101 } 1102 hinfo.hash = start_hash; 1103 hinfo.minor_hash = 0; 1104 frame = dx_probe(NULL, dir, &hinfo, frames); 1105 if (IS_ERR(frame)) 1106 return PTR_ERR(frame); 1107 1108 /* Add '.' and '..' from the htree header */ 1109 if (!start_hash && !start_minor_hash) { 1110 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; 1111 tmp_str.name = de->name; 1112 tmp_str.len = de->name_len; 1113 err = ext4_htree_store_dirent(dir_file, 0, 0, 1114 de, &tmp_str); 1115 if (err != 0) 1116 goto errout; 1117 count++; 1118 } 1119 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) { 1120 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; 1121 de = ext4_next_entry(de, dir->i_sb->s_blocksize); 1122 tmp_str.name = de->name; 1123 tmp_str.len = de->name_len; 1124 err = ext4_htree_store_dirent(dir_file, 2, 0, 1125 de, &tmp_str); 1126 if (err != 0) 1127 goto errout; 1128 count++; 1129 } 1130 1131 while (1) { 1132 if (fatal_signal_pending(current)) { 1133 err = -ERESTARTSYS; 1134 goto errout; 1135 } 1136 cond_resched(); 1137 block = dx_get_block(frame->at); 1138 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo, 1139 start_hash, start_minor_hash); 1140 if (ret < 0) { 1141 err = ret; 1142 goto errout; 1143 } 1144 count += ret; 1145 hashval = ~0; 1146 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS, 1147 frame, frames, &hashval); 1148 *next_hash = hashval; 1149 if (ret < 0) { 1150 err = ret; 1151 goto errout; 1152 } 1153 /* 1154 * Stop if: (a) there are no more entries, or 1155 * (b) we have inserted at least one entry and the 1156 * next hash value is not a continuation 1157 */ 1158 if ((ret == 0) || 1159 (count && ((hashval & 1) == 0))) 1160 break; 1161 } 1162 dx_release(frames); 1163 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, " 1164 "next hash: %x\n", count, *next_hash)); 1165 return count; 1166 errout: 1167 dx_release(frames); 1168 return (err); 1169 } 1170 1171 static inline int search_dirblock(struct buffer_head *bh, 1172 struct inode *dir, 1173 struct ext4_filename *fname, 1174 unsigned int offset, 1175 struct ext4_dir_entry_2 **res_dir) 1176 { 1177 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir, 1178 fname, offset, res_dir); 1179 } 1180 1181 /* 1182 * Directory block splitting, compacting 1183 */ 1184 1185 /* 1186 * Create map of hash values, offsets, and sizes, stored at end of block. 1187 * Returns number of entries mapped. 1188 */ 1189 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de, 1190 unsigned blocksize, struct dx_hash_info *hinfo, 1191 struct dx_map_entry *map_tail) 1192 { 1193 int count = 0; 1194 char *base = (char *) de; 1195 struct dx_hash_info h = *hinfo; 1196 1197 while ((char *) de < base + blocksize) { 1198 if (de->name_len && de->inode) { 1199 ext4fs_dirhash(de->name, de->name_len, &h); 1200 map_tail--; 1201 map_tail->hash = h.hash; 1202 map_tail->offs = ((char *) de - base)>>2; 1203 map_tail->size = le16_to_cpu(de->rec_len); 1204 count++; 1205 cond_resched(); 1206 } 1207 /* XXX: do we need to check rec_len == 0 case? -Chris */ 1208 de = ext4_next_entry(de, blocksize); 1209 } 1210 return count; 1211 } 1212 1213 /* Sort map by hash value */ 1214 static void dx_sort_map (struct dx_map_entry *map, unsigned count) 1215 { 1216 struct dx_map_entry *p, *q, *top = map + count - 1; 1217 int more; 1218 /* Combsort until bubble sort doesn't suck */ 1219 while (count > 2) { 1220 count = count*10/13; 1221 if (count - 9 < 2) /* 9, 10 -> 11 */ 1222 count = 11; 1223 for (p = top, q = p - count; q >= map; p--, q--) 1224 if (p->hash < q->hash) 1225 swap(*p, *q); 1226 } 1227 /* Garden variety bubble sort */ 1228 do { 1229 more = 0; 1230 q = top; 1231 while (q-- > map) { 1232 if (q[1].hash >= q[0].hash) 1233 continue; 1234 swap(*(q+1), *q); 1235 more = 1; 1236 } 1237 } while(more); 1238 } 1239 1240 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block) 1241 { 1242 struct dx_entry *entries = frame->entries; 1243 struct dx_entry *old = frame->at, *new = old + 1; 1244 int count = dx_get_count(entries); 1245 1246 assert(count < dx_get_limit(entries)); 1247 assert(old < entries + count); 1248 memmove(new + 1, new, (char *)(entries + count) - (char *)(new)); 1249 dx_set_hash(new, hash); 1250 dx_set_block(new, block); 1251 dx_set_count(entries, count + 1); 1252 } 1253 1254 /* 1255 * Test whether a directory entry matches the filename being searched for. 1256 * 1257 * Return: %true if the directory entry matches, otherwise %false. 1258 */ 1259 static inline bool ext4_match(const struct ext4_filename *fname, 1260 const struct ext4_dir_entry_2 *de) 1261 { 1262 struct fscrypt_name f; 1263 1264 if (!de->inode) 1265 return false; 1266 1267 f.usr_fname = fname->usr_fname; 1268 f.disk_name = fname->disk_name; 1269 #ifdef CONFIG_EXT4_FS_ENCRYPTION 1270 f.crypto_buf = fname->crypto_buf; 1271 #endif 1272 return fscrypt_match_name(&f, de->name, de->name_len); 1273 } 1274 1275 /* 1276 * Returns 0 if not found, -1 on failure, and 1 on success 1277 */ 1278 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size, 1279 struct inode *dir, struct ext4_filename *fname, 1280 unsigned int offset, struct ext4_dir_entry_2 **res_dir) 1281 { 1282 struct ext4_dir_entry_2 * de; 1283 char * dlimit; 1284 int de_len; 1285 1286 de = (struct ext4_dir_entry_2 *)search_buf; 1287 dlimit = search_buf + buf_size; 1288 while ((char *) de < dlimit) { 1289 /* this code is executed quadratically often */ 1290 /* do minimal checking `by hand' */ 1291 if ((char *) de + de->name_len <= dlimit && 1292 ext4_match(fname, de)) { 1293 /* found a match - just to be sure, do 1294 * a full check */ 1295 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data, 1296 bh->b_size, offset)) 1297 return -1; 1298 *res_dir = de; 1299 return 1; 1300 } 1301 /* prevent looping on a bad block */ 1302 de_len = ext4_rec_len_from_disk(de->rec_len, 1303 dir->i_sb->s_blocksize); 1304 if (de_len <= 0) 1305 return -1; 1306 offset += de_len; 1307 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len); 1308 } 1309 return 0; 1310 } 1311 1312 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block, 1313 struct ext4_dir_entry *de) 1314 { 1315 struct super_block *sb = dir->i_sb; 1316 1317 if (!is_dx(dir)) 1318 return 0; 1319 if (block == 0) 1320 return 1; 1321 if (de->inode == 0 && 1322 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) == 1323 sb->s_blocksize) 1324 return 1; 1325 return 0; 1326 } 1327 1328 /* 1329 * ext4_find_entry() 1330 * 1331 * finds an entry in the specified directory with the wanted name. It 1332 * returns the cache buffer in which the entry was found, and the entry 1333 * itself (as a parameter - res_dir). It does NOT read the inode of the 1334 * entry - you'll have to do that yourself if you want to. 1335 * 1336 * The returned buffer_head has ->b_count elevated. The caller is expected 1337 * to brelse() it when appropriate. 1338 */ 1339 static struct buffer_head * ext4_find_entry (struct inode *dir, 1340 const struct qstr *d_name, 1341 struct ext4_dir_entry_2 **res_dir, 1342 int *inlined) 1343 { 1344 struct super_block *sb; 1345 struct buffer_head *bh_use[NAMEI_RA_SIZE]; 1346 struct buffer_head *bh, *ret = NULL; 1347 ext4_lblk_t start, block; 1348 const u8 *name = d_name->name; 1349 size_t ra_max = 0; /* Number of bh's in the readahead 1350 buffer, bh_use[] */ 1351 size_t ra_ptr = 0; /* Current index into readahead 1352 buffer */ 1353 ext4_lblk_t nblocks; 1354 int i, namelen, retval; 1355 struct ext4_filename fname; 1356 1357 *res_dir = NULL; 1358 sb = dir->i_sb; 1359 namelen = d_name->len; 1360 if (namelen > EXT4_NAME_LEN) 1361 return NULL; 1362 1363 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname); 1364 if (retval == -ENOENT) 1365 return NULL; 1366 if (retval) 1367 return ERR_PTR(retval); 1368 1369 if (ext4_has_inline_data(dir)) { 1370 int has_inline_data = 1; 1371 ret = ext4_find_inline_entry(dir, &fname, res_dir, 1372 &has_inline_data); 1373 if (has_inline_data) { 1374 if (inlined) 1375 *inlined = 1; 1376 goto cleanup_and_exit; 1377 } 1378 } 1379 1380 if ((namelen <= 2) && (name[0] == '.') && 1381 (name[1] == '.' || name[1] == '\0')) { 1382 /* 1383 * "." or ".." will only be in the first block 1384 * NFS may look up ".."; "." should be handled by the VFS 1385 */ 1386 block = start = 0; 1387 nblocks = 1; 1388 goto restart; 1389 } 1390 if (is_dx(dir)) { 1391 ret = ext4_dx_find_entry(dir, &fname, res_dir); 1392 /* 1393 * On success, or if the error was file not found, 1394 * return. Otherwise, fall back to doing a search the 1395 * old fashioned way. 1396 */ 1397 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR) 1398 goto cleanup_and_exit; 1399 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, " 1400 "falling back\n")); 1401 ret = NULL; 1402 } 1403 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); 1404 if (!nblocks) { 1405 ret = NULL; 1406 goto cleanup_and_exit; 1407 } 1408 start = EXT4_I(dir)->i_dir_start_lookup; 1409 if (start >= nblocks) 1410 start = 0; 1411 block = start; 1412 restart: 1413 do { 1414 /* 1415 * We deal with the read-ahead logic here. 1416 */ 1417 if (ra_ptr >= ra_max) { 1418 /* Refill the readahead buffer */ 1419 ra_ptr = 0; 1420 if (block < start) 1421 ra_max = start - block; 1422 else 1423 ra_max = nblocks - block; 1424 ra_max = min(ra_max, ARRAY_SIZE(bh_use)); 1425 retval = ext4_bread_batch(dir, block, ra_max, 1426 false /* wait */, bh_use); 1427 if (retval) { 1428 ret = ERR_PTR(retval); 1429 ra_max = 0; 1430 goto cleanup_and_exit; 1431 } 1432 } 1433 if ((bh = bh_use[ra_ptr++]) == NULL) 1434 goto next; 1435 wait_on_buffer(bh); 1436 if (!buffer_uptodate(bh)) { 1437 EXT4_ERROR_INODE(dir, "reading directory lblock %lu", 1438 (unsigned long) block); 1439 brelse(bh); 1440 ret = ERR_PTR(-EIO); 1441 goto cleanup_and_exit; 1442 } 1443 if (!buffer_verified(bh) && 1444 !is_dx_internal_node(dir, block, 1445 (struct ext4_dir_entry *)bh->b_data) && 1446 !ext4_dirent_csum_verify(dir, 1447 (struct ext4_dir_entry *)bh->b_data)) { 1448 EXT4_ERROR_INODE(dir, "checksumming directory " 1449 "block %lu", (unsigned long)block); 1450 brelse(bh); 1451 ret = ERR_PTR(-EFSBADCRC); 1452 goto cleanup_and_exit; 1453 } 1454 set_buffer_verified(bh); 1455 i = search_dirblock(bh, dir, &fname, 1456 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir); 1457 if (i == 1) { 1458 EXT4_I(dir)->i_dir_start_lookup = block; 1459 ret = bh; 1460 goto cleanup_and_exit; 1461 } else { 1462 brelse(bh); 1463 if (i < 0) 1464 goto cleanup_and_exit; 1465 } 1466 next: 1467 if (++block >= nblocks) 1468 block = 0; 1469 } while (block != start); 1470 1471 /* 1472 * If the directory has grown while we were searching, then 1473 * search the last part of the directory before giving up. 1474 */ 1475 block = nblocks; 1476 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); 1477 if (block < nblocks) { 1478 start = 0; 1479 goto restart; 1480 } 1481 1482 cleanup_and_exit: 1483 /* Clean up the read-ahead blocks */ 1484 for (; ra_ptr < ra_max; ra_ptr++) 1485 brelse(bh_use[ra_ptr]); 1486 ext4_fname_free_filename(&fname); 1487 return ret; 1488 } 1489 1490 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, 1491 struct ext4_filename *fname, 1492 struct ext4_dir_entry_2 **res_dir) 1493 { 1494 struct super_block * sb = dir->i_sb; 1495 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; 1496 struct buffer_head *bh; 1497 ext4_lblk_t block; 1498 int retval; 1499 1500 #ifdef CONFIG_EXT4_FS_ENCRYPTION 1501 *res_dir = NULL; 1502 #endif 1503 frame = dx_probe(fname, dir, NULL, frames); 1504 if (IS_ERR(frame)) 1505 return (struct buffer_head *) frame; 1506 do { 1507 block = dx_get_block(frame->at); 1508 bh = ext4_read_dirblock(dir, block, DIRENT); 1509 if (IS_ERR(bh)) 1510 goto errout; 1511 1512 retval = search_dirblock(bh, dir, fname, 1513 block << EXT4_BLOCK_SIZE_BITS(sb), 1514 res_dir); 1515 if (retval == 1) 1516 goto success; 1517 brelse(bh); 1518 if (retval == -1) { 1519 bh = ERR_PTR(ERR_BAD_DX_DIR); 1520 goto errout; 1521 } 1522 1523 /* Check to see if we should continue to search */ 1524 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame, 1525 frames, NULL); 1526 if (retval < 0) { 1527 ext4_warning_inode(dir, 1528 "error %d reading directory index block", 1529 retval); 1530 bh = ERR_PTR(retval); 1531 goto errout; 1532 } 1533 } while (retval == 1); 1534 1535 bh = NULL; 1536 errout: 1537 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name)); 1538 success: 1539 dx_release(frames); 1540 return bh; 1541 } 1542 1543 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) 1544 { 1545 struct inode *inode; 1546 struct ext4_dir_entry_2 *de; 1547 struct buffer_head *bh; 1548 int err; 1549 1550 err = fscrypt_prepare_lookup(dir, dentry, flags); 1551 if (err) 1552 return ERR_PTR(err); 1553 1554 if (dentry->d_name.len > EXT4_NAME_LEN) 1555 return ERR_PTR(-ENAMETOOLONG); 1556 1557 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); 1558 if (IS_ERR(bh)) 1559 return (struct dentry *) bh; 1560 inode = NULL; 1561 if (bh) { 1562 __u32 ino = le32_to_cpu(de->inode); 1563 brelse(bh); 1564 if (!ext4_valid_inum(dir->i_sb, ino)) { 1565 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino); 1566 return ERR_PTR(-EFSCORRUPTED); 1567 } 1568 if (unlikely(ino == dir->i_ino)) { 1569 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir", 1570 dentry); 1571 return ERR_PTR(-EFSCORRUPTED); 1572 } 1573 inode = ext4_iget_normal(dir->i_sb, ino); 1574 if (inode == ERR_PTR(-ESTALE)) { 1575 EXT4_ERROR_INODE(dir, 1576 "deleted inode referenced: %u", 1577 ino); 1578 return ERR_PTR(-EFSCORRUPTED); 1579 } 1580 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) && 1581 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) && 1582 !fscrypt_has_permitted_context(dir, inode)) { 1583 ext4_warning(inode->i_sb, 1584 "Inconsistent encryption contexts: %lu/%lu", 1585 dir->i_ino, inode->i_ino); 1586 iput(inode); 1587 return ERR_PTR(-EPERM); 1588 } 1589 } 1590 return d_splice_alias(inode, dentry); 1591 } 1592 1593 1594 struct dentry *ext4_get_parent(struct dentry *child) 1595 { 1596 __u32 ino; 1597 static const struct qstr dotdot = QSTR_INIT("..", 2); 1598 struct ext4_dir_entry_2 * de; 1599 struct buffer_head *bh; 1600 1601 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL); 1602 if (IS_ERR(bh)) 1603 return (struct dentry *) bh; 1604 if (!bh) 1605 return ERR_PTR(-ENOENT); 1606 ino = le32_to_cpu(de->inode); 1607 brelse(bh); 1608 1609 if (!ext4_valid_inum(child->d_sb, ino)) { 1610 EXT4_ERROR_INODE(d_inode(child), 1611 "bad parent inode number: %u", ino); 1612 return ERR_PTR(-EFSCORRUPTED); 1613 } 1614 1615 return d_obtain_alias(ext4_iget_normal(child->d_sb, ino)); 1616 } 1617 1618 /* 1619 * Move count entries from end of map between two memory locations. 1620 * Returns pointer to last entry moved. 1621 */ 1622 static struct ext4_dir_entry_2 * 1623 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count, 1624 unsigned blocksize) 1625 { 1626 unsigned rec_len = 0; 1627 1628 while (count--) { 1629 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) 1630 (from + (map->offs<<2)); 1631 rec_len = EXT4_DIR_REC_LEN(de->name_len); 1632 memcpy (to, de, rec_len); 1633 ((struct ext4_dir_entry_2 *) to)->rec_len = 1634 ext4_rec_len_to_disk(rec_len, blocksize); 1635 de->inode = 0; 1636 map++; 1637 to += rec_len; 1638 } 1639 return (struct ext4_dir_entry_2 *) (to - rec_len); 1640 } 1641 1642 /* 1643 * Compact each dir entry in the range to the minimal rec_len. 1644 * Returns pointer to last entry in range. 1645 */ 1646 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize) 1647 { 1648 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base; 1649 unsigned rec_len = 0; 1650 1651 prev = to = de; 1652 while ((char*)de < base + blocksize) { 1653 next = ext4_next_entry(de, blocksize); 1654 if (de->inode && de->name_len) { 1655 rec_len = EXT4_DIR_REC_LEN(de->name_len); 1656 if (de > to) 1657 memmove(to, de, rec_len); 1658 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize); 1659 prev = to; 1660 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len); 1661 } 1662 de = next; 1663 } 1664 return prev; 1665 } 1666 1667 /* 1668 * Split a full leaf block to make room for a new dir entry. 1669 * Allocate a new block, and move entries so that they are approx. equally full. 1670 * Returns pointer to de in block into which the new entry will be inserted. 1671 */ 1672 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, 1673 struct buffer_head **bh,struct dx_frame *frame, 1674 struct dx_hash_info *hinfo) 1675 { 1676 unsigned blocksize = dir->i_sb->s_blocksize; 1677 unsigned count, continued; 1678 struct buffer_head *bh2; 1679 ext4_lblk_t newblock; 1680 u32 hash2; 1681 struct dx_map_entry *map; 1682 char *data1 = (*bh)->b_data, *data2; 1683 unsigned split, move, size; 1684 struct ext4_dir_entry_2 *de = NULL, *de2; 1685 struct ext4_dir_entry_tail *t; 1686 int csum_size = 0; 1687 int err = 0, i; 1688 1689 if (ext4_has_metadata_csum(dir->i_sb)) 1690 csum_size = sizeof(struct ext4_dir_entry_tail); 1691 1692 bh2 = ext4_append(handle, dir, &newblock); 1693 if (IS_ERR(bh2)) { 1694 brelse(*bh); 1695 *bh = NULL; 1696 return (struct ext4_dir_entry_2 *) bh2; 1697 } 1698 1699 BUFFER_TRACE(*bh, "get_write_access"); 1700 err = ext4_journal_get_write_access(handle, *bh); 1701 if (err) 1702 goto journal_error; 1703 1704 BUFFER_TRACE(frame->bh, "get_write_access"); 1705 err = ext4_journal_get_write_access(handle, frame->bh); 1706 if (err) 1707 goto journal_error; 1708 1709 data2 = bh2->b_data; 1710 1711 /* create map in the end of data2 block */ 1712 map = (struct dx_map_entry *) (data2 + blocksize); 1713 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1, 1714 blocksize, hinfo, map); 1715 map -= count; 1716 dx_sort_map(map, count); 1717 /* Split the existing block in the middle, size-wise */ 1718 size = 0; 1719 move = 0; 1720 for (i = count-1; i >= 0; i--) { 1721 /* is more than half of this entry in 2nd half of the block? */ 1722 if (size + map[i].size/2 > blocksize/2) 1723 break; 1724 size += map[i].size; 1725 move++; 1726 } 1727 /* map index at which we will split */ 1728 split = count - move; 1729 hash2 = map[split].hash; 1730 continued = hash2 == map[split - 1].hash; 1731 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n", 1732 (unsigned long)dx_get_block(frame->at), 1733 hash2, split, count-split)); 1734 1735 /* Fancy dance to stay within two buffers */ 1736 de2 = dx_move_dirents(data1, data2, map + split, count - split, 1737 blocksize); 1738 de = dx_pack_dirents(data1, blocksize); 1739 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) - 1740 (char *) de, 1741 blocksize); 1742 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) - 1743 (char *) de2, 1744 blocksize); 1745 if (csum_size) { 1746 t = EXT4_DIRENT_TAIL(data2, blocksize); 1747 initialize_dirent_tail(t, blocksize); 1748 1749 t = EXT4_DIRENT_TAIL(data1, blocksize); 1750 initialize_dirent_tail(t, blocksize); 1751 } 1752 1753 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1, 1754 blocksize, 1)); 1755 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2, 1756 blocksize, 1)); 1757 1758 /* Which block gets the new entry? */ 1759 if (hinfo->hash >= hash2) { 1760 swap(*bh, bh2); 1761 de = de2; 1762 } 1763 dx_insert_block(frame, hash2 + continued, newblock); 1764 err = ext4_handle_dirty_dirent_node(handle, dir, bh2); 1765 if (err) 1766 goto journal_error; 1767 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh); 1768 if (err) 1769 goto journal_error; 1770 brelse(bh2); 1771 dxtrace(dx_show_index("frame", frame->entries)); 1772 return de; 1773 1774 journal_error: 1775 brelse(*bh); 1776 brelse(bh2); 1777 *bh = NULL; 1778 ext4_std_error(dir->i_sb, err); 1779 return ERR_PTR(err); 1780 } 1781 1782 int ext4_find_dest_de(struct inode *dir, struct inode *inode, 1783 struct buffer_head *bh, 1784 void *buf, int buf_size, 1785 struct ext4_filename *fname, 1786 struct ext4_dir_entry_2 **dest_de) 1787 { 1788 struct ext4_dir_entry_2 *de; 1789 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname)); 1790 int nlen, rlen; 1791 unsigned int offset = 0; 1792 char *top; 1793 1794 de = (struct ext4_dir_entry_2 *)buf; 1795 top = buf + buf_size - reclen; 1796 while ((char *) de <= top) { 1797 if (ext4_check_dir_entry(dir, NULL, de, bh, 1798 buf, buf_size, offset)) 1799 return -EFSCORRUPTED; 1800 if (ext4_match(fname, de)) 1801 return -EEXIST; 1802 nlen = EXT4_DIR_REC_LEN(de->name_len); 1803 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 1804 if ((de->inode ? rlen - nlen : rlen) >= reclen) 1805 break; 1806 de = (struct ext4_dir_entry_2 *)((char *)de + rlen); 1807 offset += rlen; 1808 } 1809 if ((char *) de > top) 1810 return -ENOSPC; 1811 1812 *dest_de = de; 1813 return 0; 1814 } 1815 1816 void ext4_insert_dentry(struct inode *inode, 1817 struct ext4_dir_entry_2 *de, 1818 int buf_size, 1819 struct ext4_filename *fname) 1820 { 1821 1822 int nlen, rlen; 1823 1824 nlen = EXT4_DIR_REC_LEN(de->name_len); 1825 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 1826 if (de->inode) { 1827 struct ext4_dir_entry_2 *de1 = 1828 (struct ext4_dir_entry_2 *)((char *)de + nlen); 1829 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size); 1830 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size); 1831 de = de1; 1832 } 1833 de->file_type = EXT4_FT_UNKNOWN; 1834 de->inode = cpu_to_le32(inode->i_ino); 1835 ext4_set_de_type(inode->i_sb, de, inode->i_mode); 1836 de->name_len = fname_len(fname); 1837 memcpy(de->name, fname_name(fname), fname_len(fname)); 1838 } 1839 1840 /* 1841 * Add a new entry into a directory (leaf) block. If de is non-NULL, 1842 * it points to a directory entry which is guaranteed to be large 1843 * enough for new directory entry. If de is NULL, then 1844 * add_dirent_to_buf will attempt search the directory block for 1845 * space. It will return -ENOSPC if no space is available, and -EIO 1846 * and -EEXIST if directory entry already exists. 1847 */ 1848 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname, 1849 struct inode *dir, 1850 struct inode *inode, struct ext4_dir_entry_2 *de, 1851 struct buffer_head *bh) 1852 { 1853 unsigned int blocksize = dir->i_sb->s_blocksize; 1854 int csum_size = 0; 1855 int err; 1856 1857 if (ext4_has_metadata_csum(inode->i_sb)) 1858 csum_size = sizeof(struct ext4_dir_entry_tail); 1859 1860 if (!de) { 1861 err = ext4_find_dest_de(dir, inode, bh, bh->b_data, 1862 blocksize - csum_size, fname, &de); 1863 if (err) 1864 return err; 1865 } 1866 BUFFER_TRACE(bh, "get_write_access"); 1867 err = ext4_journal_get_write_access(handle, bh); 1868 if (err) { 1869 ext4_std_error(dir->i_sb, err); 1870 return err; 1871 } 1872 1873 /* By now the buffer is marked for journaling */ 1874 ext4_insert_dentry(inode, de, blocksize, fname); 1875 1876 /* 1877 * XXX shouldn't update any times until successful 1878 * completion of syscall, but too many callers depend 1879 * on this. 1880 * 1881 * XXX similarly, too many callers depend on 1882 * ext4_new_inode() setting the times, but error 1883 * recovery deletes the inode, so the worst that can 1884 * happen is that the times are slightly out of date 1885 * and/or different from the directory change time. 1886 */ 1887 dir->i_mtime = dir->i_ctime = current_time(dir); 1888 ext4_update_dx_flag(dir); 1889 inode_inc_iversion(dir); 1890 ext4_mark_inode_dirty(handle, dir); 1891 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 1892 err = ext4_handle_dirty_dirent_node(handle, dir, bh); 1893 if (err) 1894 ext4_std_error(dir->i_sb, err); 1895 return 0; 1896 } 1897 1898 /* 1899 * This converts a one block unindexed directory to a 3 block indexed 1900 * directory, and adds the dentry to the indexed directory. 1901 */ 1902 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname, 1903 struct inode *dir, 1904 struct inode *inode, struct buffer_head *bh) 1905 { 1906 struct buffer_head *bh2; 1907 struct dx_root *root; 1908 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; 1909 struct dx_entry *entries; 1910 struct ext4_dir_entry_2 *de, *de2; 1911 struct ext4_dir_entry_tail *t; 1912 char *data1, *top; 1913 unsigned len; 1914 int retval; 1915 unsigned blocksize; 1916 ext4_lblk_t block; 1917 struct fake_dirent *fde; 1918 int csum_size = 0; 1919 1920 if (ext4_has_metadata_csum(inode->i_sb)) 1921 csum_size = sizeof(struct ext4_dir_entry_tail); 1922 1923 blocksize = dir->i_sb->s_blocksize; 1924 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino)); 1925 BUFFER_TRACE(bh, "get_write_access"); 1926 retval = ext4_journal_get_write_access(handle, bh); 1927 if (retval) { 1928 ext4_std_error(dir->i_sb, retval); 1929 brelse(bh); 1930 return retval; 1931 } 1932 root = (struct dx_root *) bh->b_data; 1933 1934 /* The 0th block becomes the root, move the dirents out */ 1935 fde = &root->dotdot; 1936 de = (struct ext4_dir_entry_2 *)((char *)fde + 1937 ext4_rec_len_from_disk(fde->rec_len, blocksize)); 1938 if ((char *) de >= (((char *) root) + blocksize)) { 1939 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'"); 1940 brelse(bh); 1941 return -EFSCORRUPTED; 1942 } 1943 len = ((char *) root) + (blocksize - csum_size) - (char *) de; 1944 1945 /* Allocate new block for the 0th block's dirents */ 1946 bh2 = ext4_append(handle, dir, &block); 1947 if (IS_ERR(bh2)) { 1948 brelse(bh); 1949 return PTR_ERR(bh2); 1950 } 1951 ext4_set_inode_flag(dir, EXT4_INODE_INDEX); 1952 data1 = bh2->b_data; 1953 1954 memcpy (data1, de, len); 1955 de = (struct ext4_dir_entry_2 *) data1; 1956 top = data1 + len; 1957 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) 1958 de = de2; 1959 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) - 1960 (char *) de, 1961 blocksize); 1962 1963 if (csum_size) { 1964 t = EXT4_DIRENT_TAIL(data1, blocksize); 1965 initialize_dirent_tail(t, blocksize); 1966 } 1967 1968 /* Initialize the root; the dot dirents already exist */ 1969 de = (struct ext4_dir_entry_2 *) (&root->dotdot); 1970 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2), 1971 blocksize); 1972 memset (&root->info, 0, sizeof(root->info)); 1973 root->info.info_length = sizeof(root->info); 1974 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; 1975 entries = root->entries; 1976 dx_set_block(entries, 1); 1977 dx_set_count(entries, 1); 1978 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info))); 1979 1980 /* Initialize as for dx_probe */ 1981 fname->hinfo.hash_version = root->info.hash_version; 1982 if (fname->hinfo.hash_version <= DX_HASH_TEA) 1983 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; 1984 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; 1985 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo); 1986 1987 memset(frames, 0, sizeof(frames)); 1988 frame = frames; 1989 frame->entries = entries; 1990 frame->at = entries; 1991 frame->bh = bh; 1992 1993 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh); 1994 if (retval) 1995 goto out_frames; 1996 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2); 1997 if (retval) 1998 goto out_frames; 1999 2000 de = do_split(handle,dir, &bh2, frame, &fname->hinfo); 2001 if (IS_ERR(de)) { 2002 retval = PTR_ERR(de); 2003 goto out_frames; 2004 } 2005 2006 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2); 2007 out_frames: 2008 /* 2009 * Even if the block split failed, we have to properly write 2010 * out all the changes we did so far. Otherwise we can end up 2011 * with corrupted filesystem. 2012 */ 2013 if (retval) 2014 ext4_mark_inode_dirty(handle, dir); 2015 dx_release(frames); 2016 brelse(bh2); 2017 return retval; 2018 } 2019 2020 /* 2021 * ext4_add_entry() 2022 * 2023 * adds a file entry to the specified directory, using the same 2024 * semantics as ext4_find_entry(). It returns NULL if it failed. 2025 * 2026 * NOTE!! The inode part of 'de' is left at 0 - which means you 2027 * may not sleep between calling this and putting something into 2028 * the entry, as someone else might have used it while you slept. 2029 */ 2030 static int ext4_add_entry(handle_t *handle, struct dentry *dentry, 2031 struct inode *inode) 2032 { 2033 struct inode *dir = d_inode(dentry->d_parent); 2034 struct buffer_head *bh = NULL; 2035 struct ext4_dir_entry_2 *de; 2036 struct ext4_dir_entry_tail *t; 2037 struct super_block *sb; 2038 struct ext4_filename fname; 2039 int retval; 2040 int dx_fallback=0; 2041 unsigned blocksize; 2042 ext4_lblk_t block, blocks; 2043 int csum_size = 0; 2044 2045 if (ext4_has_metadata_csum(inode->i_sb)) 2046 csum_size = sizeof(struct ext4_dir_entry_tail); 2047 2048 sb = dir->i_sb; 2049 blocksize = sb->s_blocksize; 2050 if (!dentry->d_name.len) 2051 return -EINVAL; 2052 2053 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname); 2054 if (retval) 2055 return retval; 2056 2057 if (ext4_has_inline_data(dir)) { 2058 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode); 2059 if (retval < 0) 2060 goto out; 2061 if (retval == 1) { 2062 retval = 0; 2063 goto out; 2064 } 2065 } 2066 2067 if (is_dx(dir)) { 2068 retval = ext4_dx_add_entry(handle, &fname, dir, inode); 2069 if (!retval || (retval != ERR_BAD_DX_DIR)) 2070 goto out; 2071 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX); 2072 dx_fallback++; 2073 ext4_mark_inode_dirty(handle, dir); 2074 } 2075 blocks = dir->i_size >> sb->s_blocksize_bits; 2076 for (block = 0; block < blocks; block++) { 2077 bh = ext4_read_dirblock(dir, block, DIRENT); 2078 if (IS_ERR(bh)) { 2079 retval = PTR_ERR(bh); 2080 bh = NULL; 2081 goto out; 2082 } 2083 retval = add_dirent_to_buf(handle, &fname, dir, inode, 2084 NULL, bh); 2085 if (retval != -ENOSPC) 2086 goto out; 2087 2088 if (blocks == 1 && !dx_fallback && 2089 ext4_has_feature_dir_index(sb)) { 2090 retval = make_indexed_dir(handle, &fname, dir, 2091 inode, bh); 2092 bh = NULL; /* make_indexed_dir releases bh */ 2093 goto out; 2094 } 2095 brelse(bh); 2096 } 2097 bh = ext4_append(handle, dir, &block); 2098 if (IS_ERR(bh)) { 2099 retval = PTR_ERR(bh); 2100 bh = NULL; 2101 goto out; 2102 } 2103 de = (struct ext4_dir_entry_2 *) bh->b_data; 2104 de->inode = 0; 2105 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize); 2106 2107 if (csum_size) { 2108 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize); 2109 initialize_dirent_tail(t, blocksize); 2110 } 2111 2112 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh); 2113 out: 2114 ext4_fname_free_filename(&fname); 2115 brelse(bh); 2116 if (retval == 0) 2117 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY); 2118 return retval; 2119 } 2120 2121 /* 2122 * Returns 0 for success, or a negative error value 2123 */ 2124 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname, 2125 struct inode *dir, struct inode *inode) 2126 { 2127 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; 2128 struct dx_entry *entries, *at; 2129 struct buffer_head *bh; 2130 struct super_block *sb = dir->i_sb; 2131 struct ext4_dir_entry_2 *de; 2132 int restart; 2133 int err; 2134 2135 again: 2136 restart = 0; 2137 frame = dx_probe(fname, dir, NULL, frames); 2138 if (IS_ERR(frame)) 2139 return PTR_ERR(frame); 2140 entries = frame->entries; 2141 at = frame->at; 2142 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT); 2143 if (IS_ERR(bh)) { 2144 err = PTR_ERR(bh); 2145 bh = NULL; 2146 goto cleanup; 2147 } 2148 2149 BUFFER_TRACE(bh, "get_write_access"); 2150 err = ext4_journal_get_write_access(handle, bh); 2151 if (err) 2152 goto journal_error; 2153 2154 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh); 2155 if (err != -ENOSPC) 2156 goto cleanup; 2157 2158 err = 0; 2159 /* Block full, should compress but for now just split */ 2160 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n", 2161 dx_get_count(entries), dx_get_limit(entries))); 2162 /* Need to split index? */ 2163 if (dx_get_count(entries) == dx_get_limit(entries)) { 2164 ext4_lblk_t newblock; 2165 int levels = frame - frames + 1; 2166 unsigned int icount; 2167 int add_level = 1; 2168 struct dx_entry *entries2; 2169 struct dx_node *node2; 2170 struct buffer_head *bh2; 2171 2172 while (frame > frames) { 2173 if (dx_get_count((frame - 1)->entries) < 2174 dx_get_limit((frame - 1)->entries)) { 2175 add_level = 0; 2176 break; 2177 } 2178 frame--; /* split higher index block */ 2179 at = frame->at; 2180 entries = frame->entries; 2181 restart = 1; 2182 } 2183 if (add_level && levels == ext4_dir_htree_level(sb)) { 2184 ext4_warning(sb, "Directory (ino: %lu) index full, " 2185 "reach max htree level :%d", 2186 dir->i_ino, levels); 2187 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) { 2188 ext4_warning(sb, "Large directory feature is " 2189 "not enabled on this " 2190 "filesystem"); 2191 } 2192 err = -ENOSPC; 2193 goto cleanup; 2194 } 2195 icount = dx_get_count(entries); 2196 bh2 = ext4_append(handle, dir, &newblock); 2197 if (IS_ERR(bh2)) { 2198 err = PTR_ERR(bh2); 2199 goto cleanup; 2200 } 2201 node2 = (struct dx_node *)(bh2->b_data); 2202 entries2 = node2->entries; 2203 memset(&node2->fake, 0, sizeof(struct fake_dirent)); 2204 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize, 2205 sb->s_blocksize); 2206 BUFFER_TRACE(frame->bh, "get_write_access"); 2207 err = ext4_journal_get_write_access(handle, frame->bh); 2208 if (err) 2209 goto journal_error; 2210 if (!add_level) { 2211 unsigned icount1 = icount/2, icount2 = icount - icount1; 2212 unsigned hash2 = dx_get_hash(entries + icount1); 2213 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n", 2214 icount1, icount2)); 2215 2216 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */ 2217 err = ext4_journal_get_write_access(handle, 2218 (frame - 1)->bh); 2219 if (err) 2220 goto journal_error; 2221 2222 memcpy((char *) entries2, (char *) (entries + icount1), 2223 icount2 * sizeof(struct dx_entry)); 2224 dx_set_count(entries, icount1); 2225 dx_set_count(entries2, icount2); 2226 dx_set_limit(entries2, dx_node_limit(dir)); 2227 2228 /* Which index block gets the new entry? */ 2229 if (at - entries >= icount1) { 2230 frame->at = at = at - entries - icount1 + entries2; 2231 frame->entries = entries = entries2; 2232 swap(frame->bh, bh2); 2233 } 2234 dx_insert_block((frame - 1), hash2, newblock); 2235 dxtrace(dx_show_index("node", frame->entries)); 2236 dxtrace(dx_show_index("node", 2237 ((struct dx_node *) bh2->b_data)->entries)); 2238 err = ext4_handle_dirty_dx_node(handle, dir, bh2); 2239 if (err) 2240 goto journal_error; 2241 brelse (bh2); 2242 err = ext4_handle_dirty_dx_node(handle, dir, 2243 (frame - 1)->bh); 2244 if (err) 2245 goto journal_error; 2246 if (restart) { 2247 err = ext4_handle_dirty_dx_node(handle, dir, 2248 frame->bh); 2249 goto journal_error; 2250 } 2251 } else { 2252 struct dx_root *dxroot; 2253 memcpy((char *) entries2, (char *) entries, 2254 icount * sizeof(struct dx_entry)); 2255 dx_set_limit(entries2, dx_node_limit(dir)); 2256 2257 /* Set up root */ 2258 dx_set_count(entries, 1); 2259 dx_set_block(entries + 0, newblock); 2260 dxroot = (struct dx_root *)frames[0].bh->b_data; 2261 dxroot->info.indirect_levels += 1; 2262 dxtrace(printk(KERN_DEBUG 2263 "Creating %d level index...\n", 2264 info->indirect_levels)); 2265 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh); 2266 if (err) 2267 goto journal_error; 2268 err = ext4_handle_dirty_dx_node(handle, dir, bh2); 2269 brelse(bh2); 2270 restart = 1; 2271 goto journal_error; 2272 } 2273 } 2274 de = do_split(handle, dir, &bh, frame, &fname->hinfo); 2275 if (IS_ERR(de)) { 2276 err = PTR_ERR(de); 2277 goto cleanup; 2278 } 2279 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh); 2280 goto cleanup; 2281 2282 journal_error: 2283 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */ 2284 cleanup: 2285 brelse(bh); 2286 dx_release(frames); 2287 /* @restart is true means htree-path has been changed, we need to 2288 * repeat dx_probe() to find out valid htree-path 2289 */ 2290 if (restart && err == 0) 2291 goto again; 2292 return err; 2293 } 2294 2295 /* 2296 * ext4_generic_delete_entry deletes a directory entry by merging it 2297 * with the previous entry 2298 */ 2299 int ext4_generic_delete_entry(handle_t *handle, 2300 struct inode *dir, 2301 struct ext4_dir_entry_2 *de_del, 2302 struct buffer_head *bh, 2303 void *entry_buf, 2304 int buf_size, 2305 int csum_size) 2306 { 2307 struct ext4_dir_entry_2 *de, *pde; 2308 unsigned int blocksize = dir->i_sb->s_blocksize; 2309 int i; 2310 2311 i = 0; 2312 pde = NULL; 2313 de = (struct ext4_dir_entry_2 *)entry_buf; 2314 while (i < buf_size - csum_size) { 2315 if (ext4_check_dir_entry(dir, NULL, de, bh, 2316 bh->b_data, bh->b_size, i)) 2317 return -EFSCORRUPTED; 2318 if (de == de_del) { 2319 if (pde) 2320 pde->rec_len = ext4_rec_len_to_disk( 2321 ext4_rec_len_from_disk(pde->rec_len, 2322 blocksize) + 2323 ext4_rec_len_from_disk(de->rec_len, 2324 blocksize), 2325 blocksize); 2326 else 2327 de->inode = 0; 2328 inode_inc_iversion(dir); 2329 return 0; 2330 } 2331 i += ext4_rec_len_from_disk(de->rec_len, blocksize); 2332 pde = de; 2333 de = ext4_next_entry(de, blocksize); 2334 } 2335 return -ENOENT; 2336 } 2337 2338 static int ext4_delete_entry(handle_t *handle, 2339 struct inode *dir, 2340 struct ext4_dir_entry_2 *de_del, 2341 struct buffer_head *bh) 2342 { 2343 int err, csum_size = 0; 2344 2345 if (ext4_has_inline_data(dir)) { 2346 int has_inline_data = 1; 2347 err = ext4_delete_inline_entry(handle, dir, de_del, bh, 2348 &has_inline_data); 2349 if (has_inline_data) 2350 return err; 2351 } 2352 2353 if (ext4_has_metadata_csum(dir->i_sb)) 2354 csum_size = sizeof(struct ext4_dir_entry_tail); 2355 2356 BUFFER_TRACE(bh, "get_write_access"); 2357 err = ext4_journal_get_write_access(handle, bh); 2358 if (unlikely(err)) 2359 goto out; 2360 2361 err = ext4_generic_delete_entry(handle, dir, de_del, 2362 bh, bh->b_data, 2363 dir->i_sb->s_blocksize, csum_size); 2364 if (err) 2365 goto out; 2366 2367 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 2368 err = ext4_handle_dirty_dirent_node(handle, dir, bh); 2369 if (unlikely(err)) 2370 goto out; 2371 2372 return 0; 2373 out: 2374 if (err != -ENOENT) 2375 ext4_std_error(dir->i_sb, err); 2376 return err; 2377 } 2378 2379 /* 2380 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2 2381 * since this indicates that nlinks count was previously 1 to avoid overflowing 2382 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean 2383 * that subdirectory link counts are not being maintained accurately. 2384 * 2385 * The caller has already checked for i_nlink overflow in case the DIR_LINK 2386 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy 2387 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set 2388 * on regular files) and to avoid creating huge/slow non-HTREE directories. 2389 */ 2390 static void ext4_inc_count(handle_t *handle, struct inode *inode) 2391 { 2392 inc_nlink(inode); 2393 if (is_dx(inode) && 2394 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2)) 2395 set_nlink(inode, 1); 2396 } 2397 2398 /* 2399 * If a directory had nlink == 1, then we should let it be 1. This indicates 2400 * directory has >EXT4_LINK_MAX subdirs. 2401 */ 2402 static void ext4_dec_count(handle_t *handle, struct inode *inode) 2403 { 2404 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2) 2405 drop_nlink(inode); 2406 } 2407 2408 2409 static int ext4_add_nondir(handle_t *handle, 2410 struct dentry *dentry, struct inode *inode) 2411 { 2412 int err = ext4_add_entry(handle, dentry, inode); 2413 if (!err) { 2414 ext4_mark_inode_dirty(handle, inode); 2415 d_instantiate_new(dentry, inode); 2416 return 0; 2417 } 2418 drop_nlink(inode); 2419 unlock_new_inode(inode); 2420 iput(inode); 2421 return err; 2422 } 2423 2424 /* 2425 * By the time this is called, we already have created 2426 * the directory cache entry for the new file, but it 2427 * is so far negative - it has no inode. 2428 * 2429 * If the create succeeds, we fill in the inode information 2430 * with d_instantiate(). 2431 */ 2432 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2433 bool excl) 2434 { 2435 handle_t *handle; 2436 struct inode *inode; 2437 int err, credits, retries = 0; 2438 2439 err = dquot_initialize(dir); 2440 if (err) 2441 return err; 2442 2443 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 2444 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); 2445 retry: 2446 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0, 2447 NULL, EXT4_HT_DIR, credits); 2448 handle = ext4_journal_current_handle(); 2449 err = PTR_ERR(inode); 2450 if (!IS_ERR(inode)) { 2451 inode->i_op = &ext4_file_inode_operations; 2452 inode->i_fop = &ext4_file_operations; 2453 ext4_set_aops(inode); 2454 err = ext4_add_nondir(handle, dentry, inode); 2455 if (!err && IS_DIRSYNC(dir)) 2456 ext4_handle_sync(handle); 2457 } 2458 if (handle) 2459 ext4_journal_stop(handle); 2460 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2461 goto retry; 2462 return err; 2463 } 2464 2465 static int ext4_mknod(struct inode *dir, struct dentry *dentry, 2466 umode_t mode, dev_t rdev) 2467 { 2468 handle_t *handle; 2469 struct inode *inode; 2470 int err, credits, retries = 0; 2471 2472 err = dquot_initialize(dir); 2473 if (err) 2474 return err; 2475 2476 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 2477 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); 2478 retry: 2479 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0, 2480 NULL, EXT4_HT_DIR, credits); 2481 handle = ext4_journal_current_handle(); 2482 err = PTR_ERR(inode); 2483 if (!IS_ERR(inode)) { 2484 init_special_inode(inode, inode->i_mode, rdev); 2485 inode->i_op = &ext4_special_inode_operations; 2486 err = ext4_add_nondir(handle, dentry, inode); 2487 if (!err && IS_DIRSYNC(dir)) 2488 ext4_handle_sync(handle); 2489 } 2490 if (handle) 2491 ext4_journal_stop(handle); 2492 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2493 goto retry; 2494 return err; 2495 } 2496 2497 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) 2498 { 2499 handle_t *handle; 2500 struct inode *inode; 2501 int err, retries = 0; 2502 2503 err = dquot_initialize(dir); 2504 if (err) 2505 return err; 2506 2507 retry: 2508 inode = ext4_new_inode_start_handle(dir, mode, 2509 NULL, 0, NULL, 2510 EXT4_HT_DIR, 2511 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) + 2512 4 + EXT4_XATTR_TRANS_BLOCKS); 2513 handle = ext4_journal_current_handle(); 2514 err = PTR_ERR(inode); 2515 if (!IS_ERR(inode)) { 2516 inode->i_op = &ext4_file_inode_operations; 2517 inode->i_fop = &ext4_file_operations; 2518 ext4_set_aops(inode); 2519 d_tmpfile(dentry, inode); 2520 err = ext4_orphan_add(handle, inode); 2521 if (err) 2522 goto err_unlock_inode; 2523 mark_inode_dirty(inode); 2524 unlock_new_inode(inode); 2525 } 2526 if (handle) 2527 ext4_journal_stop(handle); 2528 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2529 goto retry; 2530 return err; 2531 err_unlock_inode: 2532 ext4_journal_stop(handle); 2533 unlock_new_inode(inode); 2534 return err; 2535 } 2536 2537 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode, 2538 struct ext4_dir_entry_2 *de, 2539 int blocksize, int csum_size, 2540 unsigned int parent_ino, int dotdot_real_len) 2541 { 2542 de->inode = cpu_to_le32(inode->i_ino); 2543 de->name_len = 1; 2544 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len), 2545 blocksize); 2546 strcpy(de->name, "."); 2547 ext4_set_de_type(inode->i_sb, de, S_IFDIR); 2548 2549 de = ext4_next_entry(de, blocksize); 2550 de->inode = cpu_to_le32(parent_ino); 2551 de->name_len = 2; 2552 if (!dotdot_real_len) 2553 de->rec_len = ext4_rec_len_to_disk(blocksize - 2554 (csum_size + EXT4_DIR_REC_LEN(1)), 2555 blocksize); 2556 else 2557 de->rec_len = ext4_rec_len_to_disk( 2558 EXT4_DIR_REC_LEN(de->name_len), blocksize); 2559 strcpy(de->name, ".."); 2560 ext4_set_de_type(inode->i_sb, de, S_IFDIR); 2561 2562 return ext4_next_entry(de, blocksize); 2563 } 2564 2565 static int ext4_init_new_dir(handle_t *handle, struct inode *dir, 2566 struct inode *inode) 2567 { 2568 struct buffer_head *dir_block = NULL; 2569 struct ext4_dir_entry_2 *de; 2570 struct ext4_dir_entry_tail *t; 2571 ext4_lblk_t block = 0; 2572 unsigned int blocksize = dir->i_sb->s_blocksize; 2573 int csum_size = 0; 2574 int err; 2575 2576 if (ext4_has_metadata_csum(dir->i_sb)) 2577 csum_size = sizeof(struct ext4_dir_entry_tail); 2578 2579 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { 2580 err = ext4_try_create_inline_dir(handle, dir, inode); 2581 if (err < 0 && err != -ENOSPC) 2582 goto out; 2583 if (!err) 2584 goto out; 2585 } 2586 2587 inode->i_size = 0; 2588 dir_block = ext4_append(handle, inode, &block); 2589 if (IS_ERR(dir_block)) 2590 return PTR_ERR(dir_block); 2591 de = (struct ext4_dir_entry_2 *)dir_block->b_data; 2592 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0); 2593 set_nlink(inode, 2); 2594 if (csum_size) { 2595 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize); 2596 initialize_dirent_tail(t, blocksize); 2597 } 2598 2599 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata"); 2600 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block); 2601 if (err) 2602 goto out; 2603 set_buffer_verified(dir_block); 2604 out: 2605 brelse(dir_block); 2606 return err; 2607 } 2608 2609 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 2610 { 2611 handle_t *handle; 2612 struct inode *inode; 2613 int err, credits, retries = 0; 2614 2615 if (EXT4_DIR_LINK_MAX(dir)) 2616 return -EMLINK; 2617 2618 err = dquot_initialize(dir); 2619 if (err) 2620 return err; 2621 2622 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 2623 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); 2624 retry: 2625 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode, 2626 &dentry->d_name, 2627 0, NULL, EXT4_HT_DIR, credits); 2628 handle = ext4_journal_current_handle(); 2629 err = PTR_ERR(inode); 2630 if (IS_ERR(inode)) 2631 goto out_stop; 2632 2633 inode->i_op = &ext4_dir_inode_operations; 2634 inode->i_fop = &ext4_dir_operations; 2635 err = ext4_init_new_dir(handle, dir, inode); 2636 if (err) 2637 goto out_clear_inode; 2638 err = ext4_mark_inode_dirty(handle, inode); 2639 if (!err) 2640 err = ext4_add_entry(handle, dentry, inode); 2641 if (err) { 2642 out_clear_inode: 2643 clear_nlink(inode); 2644 unlock_new_inode(inode); 2645 ext4_mark_inode_dirty(handle, inode); 2646 iput(inode); 2647 goto out_stop; 2648 } 2649 ext4_inc_count(handle, dir); 2650 ext4_update_dx_flag(dir); 2651 err = ext4_mark_inode_dirty(handle, dir); 2652 if (err) 2653 goto out_clear_inode; 2654 d_instantiate_new(dentry, inode); 2655 if (IS_DIRSYNC(dir)) 2656 ext4_handle_sync(handle); 2657 2658 out_stop: 2659 if (handle) 2660 ext4_journal_stop(handle); 2661 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2662 goto retry; 2663 return err; 2664 } 2665 2666 /* 2667 * routine to check that the specified directory is empty (for rmdir) 2668 */ 2669 bool ext4_empty_dir(struct inode *inode) 2670 { 2671 unsigned int offset; 2672 struct buffer_head *bh; 2673 struct ext4_dir_entry_2 *de, *de1; 2674 struct super_block *sb; 2675 2676 if (ext4_has_inline_data(inode)) { 2677 int has_inline_data = 1; 2678 int ret; 2679 2680 ret = empty_inline_dir(inode, &has_inline_data); 2681 if (has_inline_data) 2682 return ret; 2683 } 2684 2685 sb = inode->i_sb; 2686 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) { 2687 EXT4_ERROR_INODE(inode, "invalid size"); 2688 return true; 2689 } 2690 bh = ext4_read_dirblock(inode, 0, EITHER); 2691 if (IS_ERR(bh)) 2692 return true; 2693 2694 de = (struct ext4_dir_entry_2 *) bh->b_data; 2695 de1 = ext4_next_entry(de, sb->s_blocksize); 2696 if (le32_to_cpu(de->inode) != inode->i_ino || 2697 le32_to_cpu(de1->inode) == 0 || 2698 strcmp(".", de->name) || strcmp("..", de1->name)) { 2699 ext4_warning_inode(inode, "directory missing '.' and/or '..'"); 2700 brelse(bh); 2701 return true; 2702 } 2703 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) + 2704 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize); 2705 de = ext4_next_entry(de1, sb->s_blocksize); 2706 while (offset < inode->i_size) { 2707 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) { 2708 unsigned int lblock; 2709 brelse(bh); 2710 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb); 2711 bh = ext4_read_dirblock(inode, lblock, EITHER); 2712 if (IS_ERR(bh)) 2713 return true; 2714 de = (struct ext4_dir_entry_2 *) bh->b_data; 2715 } 2716 if (ext4_check_dir_entry(inode, NULL, de, bh, 2717 bh->b_data, bh->b_size, offset)) { 2718 de = (struct ext4_dir_entry_2 *)(bh->b_data + 2719 sb->s_blocksize); 2720 offset = (offset | (sb->s_blocksize - 1)) + 1; 2721 continue; 2722 } 2723 if (le32_to_cpu(de->inode)) { 2724 brelse(bh); 2725 return false; 2726 } 2727 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize); 2728 de = ext4_next_entry(de, sb->s_blocksize); 2729 } 2730 brelse(bh); 2731 return true; 2732 } 2733 2734 /* 2735 * ext4_orphan_add() links an unlinked or truncated inode into a list of 2736 * such inodes, starting at the superblock, in case we crash before the 2737 * file is closed/deleted, or in case the inode truncate spans multiple 2738 * transactions and the last transaction is not recovered after a crash. 2739 * 2740 * At filesystem recovery time, we walk this list deleting unlinked 2741 * inodes and truncating linked inodes in ext4_orphan_cleanup(). 2742 * 2743 * Orphan list manipulation functions must be called under i_mutex unless 2744 * we are just creating the inode or deleting it. 2745 */ 2746 int ext4_orphan_add(handle_t *handle, struct inode *inode) 2747 { 2748 struct super_block *sb = inode->i_sb; 2749 struct ext4_sb_info *sbi = EXT4_SB(sb); 2750 struct ext4_iloc iloc; 2751 int err = 0, rc; 2752 bool dirty = false; 2753 2754 if (!sbi->s_journal || is_bad_inode(inode)) 2755 return 0; 2756 2757 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) && 2758 !inode_is_locked(inode)); 2759 /* 2760 * Exit early if inode already is on orphan list. This is a big speedup 2761 * since we don't have to contend on the global s_orphan_lock. 2762 */ 2763 if (!list_empty(&EXT4_I(inode)->i_orphan)) 2764 return 0; 2765 2766 /* 2767 * Orphan handling is only valid for files with data blocks 2768 * being truncated, or files being unlinked. Note that we either 2769 * hold i_mutex, or the inode can not be referenced from outside, 2770 * so i_nlink should not be bumped due to race 2771 */ 2772 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 2773 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0); 2774 2775 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 2776 err = ext4_journal_get_write_access(handle, sbi->s_sbh); 2777 if (err) 2778 goto out; 2779 2780 err = ext4_reserve_inode_write(handle, inode, &iloc); 2781 if (err) 2782 goto out; 2783 2784 mutex_lock(&sbi->s_orphan_lock); 2785 /* 2786 * Due to previous errors inode may be already a part of on-disk 2787 * orphan list. If so skip on-disk list modification. 2788 */ 2789 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) > 2790 (le32_to_cpu(sbi->s_es->s_inodes_count))) { 2791 /* Insert this inode at the head of the on-disk orphan list */ 2792 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan); 2793 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino); 2794 dirty = true; 2795 } 2796 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan); 2797 mutex_unlock(&sbi->s_orphan_lock); 2798 2799 if (dirty) { 2800 err = ext4_handle_dirty_super(handle, sb); 2801 rc = ext4_mark_iloc_dirty(handle, inode, &iloc); 2802 if (!err) 2803 err = rc; 2804 if (err) { 2805 /* 2806 * We have to remove inode from in-memory list if 2807 * addition to on disk orphan list failed. Stray orphan 2808 * list entries can cause panics at unmount time. 2809 */ 2810 mutex_lock(&sbi->s_orphan_lock); 2811 list_del_init(&EXT4_I(inode)->i_orphan); 2812 mutex_unlock(&sbi->s_orphan_lock); 2813 } 2814 } 2815 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino); 2816 jbd_debug(4, "orphan inode %lu will point to %d\n", 2817 inode->i_ino, NEXT_ORPHAN(inode)); 2818 out: 2819 ext4_std_error(sb, err); 2820 return err; 2821 } 2822 2823 /* 2824 * ext4_orphan_del() removes an unlinked or truncated inode from the list 2825 * of such inodes stored on disk, because it is finally being cleaned up. 2826 */ 2827 int ext4_orphan_del(handle_t *handle, struct inode *inode) 2828 { 2829 struct list_head *prev; 2830 struct ext4_inode_info *ei = EXT4_I(inode); 2831 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 2832 __u32 ino_next; 2833 struct ext4_iloc iloc; 2834 int err = 0; 2835 2836 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS)) 2837 return 0; 2838 2839 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) && 2840 !inode_is_locked(inode)); 2841 /* Do this quick check before taking global s_orphan_lock. */ 2842 if (list_empty(&ei->i_orphan)) 2843 return 0; 2844 2845 if (handle) { 2846 /* Grab inode buffer early before taking global s_orphan_lock */ 2847 err = ext4_reserve_inode_write(handle, inode, &iloc); 2848 } 2849 2850 mutex_lock(&sbi->s_orphan_lock); 2851 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino); 2852 2853 prev = ei->i_orphan.prev; 2854 list_del_init(&ei->i_orphan); 2855 2856 /* If we're on an error path, we may not have a valid 2857 * transaction handle with which to update the orphan list on 2858 * disk, but we still need to remove the inode from the linked 2859 * list in memory. */ 2860 if (!handle || err) { 2861 mutex_unlock(&sbi->s_orphan_lock); 2862 goto out_err; 2863 } 2864 2865 ino_next = NEXT_ORPHAN(inode); 2866 if (prev == &sbi->s_orphan) { 2867 jbd_debug(4, "superblock will point to %u\n", ino_next); 2868 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 2869 err = ext4_journal_get_write_access(handle, sbi->s_sbh); 2870 if (err) { 2871 mutex_unlock(&sbi->s_orphan_lock); 2872 goto out_brelse; 2873 } 2874 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next); 2875 mutex_unlock(&sbi->s_orphan_lock); 2876 err = ext4_handle_dirty_super(handle, inode->i_sb); 2877 } else { 2878 struct ext4_iloc iloc2; 2879 struct inode *i_prev = 2880 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode; 2881 2882 jbd_debug(4, "orphan inode %lu will point to %u\n", 2883 i_prev->i_ino, ino_next); 2884 err = ext4_reserve_inode_write(handle, i_prev, &iloc2); 2885 if (err) { 2886 mutex_unlock(&sbi->s_orphan_lock); 2887 goto out_brelse; 2888 } 2889 NEXT_ORPHAN(i_prev) = ino_next; 2890 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2); 2891 mutex_unlock(&sbi->s_orphan_lock); 2892 } 2893 if (err) 2894 goto out_brelse; 2895 NEXT_ORPHAN(inode) = 0; 2896 err = ext4_mark_iloc_dirty(handle, inode, &iloc); 2897 out_err: 2898 ext4_std_error(inode->i_sb, err); 2899 return err; 2900 2901 out_brelse: 2902 brelse(iloc.bh); 2903 goto out_err; 2904 } 2905 2906 static int ext4_rmdir(struct inode *dir, struct dentry *dentry) 2907 { 2908 int retval; 2909 struct inode *inode; 2910 struct buffer_head *bh; 2911 struct ext4_dir_entry_2 *de; 2912 handle_t *handle = NULL; 2913 2914 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb)))) 2915 return -EIO; 2916 2917 /* Initialize quotas before so that eventual writes go in 2918 * separate transaction */ 2919 retval = dquot_initialize(dir); 2920 if (retval) 2921 return retval; 2922 retval = dquot_initialize(d_inode(dentry)); 2923 if (retval) 2924 return retval; 2925 2926 retval = -ENOENT; 2927 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); 2928 if (IS_ERR(bh)) 2929 return PTR_ERR(bh); 2930 if (!bh) 2931 goto end_rmdir; 2932 2933 inode = d_inode(dentry); 2934 2935 retval = -EFSCORRUPTED; 2936 if (le32_to_cpu(de->inode) != inode->i_ino) 2937 goto end_rmdir; 2938 2939 retval = -ENOTEMPTY; 2940 if (!ext4_empty_dir(inode)) 2941 goto end_rmdir; 2942 2943 handle = ext4_journal_start(dir, EXT4_HT_DIR, 2944 EXT4_DATA_TRANS_BLOCKS(dir->i_sb)); 2945 if (IS_ERR(handle)) { 2946 retval = PTR_ERR(handle); 2947 handle = NULL; 2948 goto end_rmdir; 2949 } 2950 2951 if (IS_DIRSYNC(dir)) 2952 ext4_handle_sync(handle); 2953 2954 retval = ext4_delete_entry(handle, dir, de, bh); 2955 if (retval) 2956 goto end_rmdir; 2957 if (!EXT4_DIR_LINK_EMPTY(inode)) 2958 ext4_warning_inode(inode, 2959 "empty directory '%.*s' has too many links (%u)", 2960 dentry->d_name.len, dentry->d_name.name, 2961 inode->i_nlink); 2962 inode_inc_iversion(inode); 2963 clear_nlink(inode); 2964 /* There's no need to set i_disksize: the fact that i_nlink is 2965 * zero will ensure that the right thing happens during any 2966 * recovery. */ 2967 inode->i_size = 0; 2968 ext4_orphan_add(handle, inode); 2969 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); 2970 ext4_mark_inode_dirty(handle, inode); 2971 ext4_dec_count(handle, dir); 2972 ext4_update_dx_flag(dir); 2973 ext4_mark_inode_dirty(handle, dir); 2974 2975 end_rmdir: 2976 brelse(bh); 2977 if (handle) 2978 ext4_journal_stop(handle); 2979 return retval; 2980 } 2981 2982 static int ext4_unlink(struct inode *dir, struct dentry *dentry) 2983 { 2984 int retval; 2985 struct inode *inode; 2986 struct buffer_head *bh; 2987 struct ext4_dir_entry_2 *de; 2988 handle_t *handle = NULL; 2989 2990 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb)))) 2991 return -EIO; 2992 2993 trace_ext4_unlink_enter(dir, dentry); 2994 /* Initialize quotas before so that eventual writes go 2995 * in separate transaction */ 2996 retval = dquot_initialize(dir); 2997 if (retval) 2998 return retval; 2999 retval = dquot_initialize(d_inode(dentry)); 3000 if (retval) 3001 return retval; 3002 3003 retval = -ENOENT; 3004 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); 3005 if (IS_ERR(bh)) 3006 return PTR_ERR(bh); 3007 if (!bh) 3008 goto end_unlink; 3009 3010 inode = d_inode(dentry); 3011 3012 retval = -EFSCORRUPTED; 3013 if (le32_to_cpu(de->inode) != inode->i_ino) 3014 goto end_unlink; 3015 3016 handle = ext4_journal_start(dir, EXT4_HT_DIR, 3017 EXT4_DATA_TRANS_BLOCKS(dir->i_sb)); 3018 if (IS_ERR(handle)) { 3019 retval = PTR_ERR(handle); 3020 handle = NULL; 3021 goto end_unlink; 3022 } 3023 3024 if (IS_DIRSYNC(dir)) 3025 ext4_handle_sync(handle); 3026 3027 if (inode->i_nlink == 0) { 3028 ext4_warning_inode(inode, "Deleting file '%.*s' with no links", 3029 dentry->d_name.len, dentry->d_name.name); 3030 set_nlink(inode, 1); 3031 } 3032 retval = ext4_delete_entry(handle, dir, de, bh); 3033 if (retval) 3034 goto end_unlink; 3035 dir->i_ctime = dir->i_mtime = current_time(dir); 3036 ext4_update_dx_flag(dir); 3037 ext4_mark_inode_dirty(handle, dir); 3038 drop_nlink(inode); 3039 if (!inode->i_nlink) 3040 ext4_orphan_add(handle, inode); 3041 inode->i_ctime = current_time(inode); 3042 ext4_mark_inode_dirty(handle, inode); 3043 3044 end_unlink: 3045 brelse(bh); 3046 if (handle) 3047 ext4_journal_stop(handle); 3048 trace_ext4_unlink_exit(dentry, retval); 3049 return retval; 3050 } 3051 3052 static int ext4_symlink(struct inode *dir, 3053 struct dentry *dentry, const char *symname) 3054 { 3055 handle_t *handle; 3056 struct inode *inode; 3057 int err, len = strlen(symname); 3058 int credits; 3059 struct fscrypt_str disk_link; 3060 3061 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb)))) 3062 return -EIO; 3063 3064 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize, 3065 &disk_link); 3066 if (err) 3067 return err; 3068 3069 err = dquot_initialize(dir); 3070 if (err) 3071 return err; 3072 3073 if ((disk_link.len > EXT4_N_BLOCKS * 4)) { 3074 /* 3075 * For non-fast symlinks, we just allocate inode and put it on 3076 * orphan list in the first transaction => we need bitmap, 3077 * group descriptor, sb, inode block, quota blocks, and 3078 * possibly selinux xattr blocks. 3079 */ 3080 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) + 3081 EXT4_XATTR_TRANS_BLOCKS; 3082 } else { 3083 /* 3084 * Fast symlink. We have to add entry to directory 3085 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS), 3086 * allocate new inode (bitmap, group descriptor, inode block, 3087 * quota blocks, sb is already counted in previous macros). 3088 */ 3089 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 3090 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3; 3091 } 3092 3093 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO, 3094 &dentry->d_name, 0, NULL, 3095 EXT4_HT_DIR, credits); 3096 handle = ext4_journal_current_handle(); 3097 if (IS_ERR(inode)) { 3098 if (handle) 3099 ext4_journal_stop(handle); 3100 return PTR_ERR(inode); 3101 } 3102 3103 if (IS_ENCRYPTED(inode)) { 3104 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link); 3105 if (err) 3106 goto err_drop_inode; 3107 inode->i_op = &ext4_encrypted_symlink_inode_operations; 3108 } 3109 3110 if ((disk_link.len > EXT4_N_BLOCKS * 4)) { 3111 if (!IS_ENCRYPTED(inode)) 3112 inode->i_op = &ext4_symlink_inode_operations; 3113 inode_nohighmem(inode); 3114 ext4_set_aops(inode); 3115 /* 3116 * We cannot call page_symlink() with transaction started 3117 * because it calls into ext4_write_begin() which can wait 3118 * for transaction commit if we are running out of space 3119 * and thus we deadlock. So we have to stop transaction now 3120 * and restart it when symlink contents is written. 3121 * 3122 * To keep fs consistent in case of crash, we have to put inode 3123 * to orphan list in the mean time. 3124 */ 3125 drop_nlink(inode); 3126 err = ext4_orphan_add(handle, inode); 3127 ext4_journal_stop(handle); 3128 handle = NULL; 3129 if (err) 3130 goto err_drop_inode; 3131 err = __page_symlink(inode, disk_link.name, disk_link.len, 1); 3132 if (err) 3133 goto err_drop_inode; 3134 /* 3135 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS 3136 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified 3137 */ 3138 handle = ext4_journal_start(dir, EXT4_HT_DIR, 3139 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 3140 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1); 3141 if (IS_ERR(handle)) { 3142 err = PTR_ERR(handle); 3143 handle = NULL; 3144 goto err_drop_inode; 3145 } 3146 set_nlink(inode, 1); 3147 err = ext4_orphan_del(handle, inode); 3148 if (err) 3149 goto err_drop_inode; 3150 } else { 3151 /* clear the extent format for fast symlink */ 3152 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS); 3153 if (!IS_ENCRYPTED(inode)) { 3154 inode->i_op = &ext4_fast_symlink_inode_operations; 3155 inode->i_link = (char *)&EXT4_I(inode)->i_data; 3156 } 3157 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name, 3158 disk_link.len); 3159 inode->i_size = disk_link.len - 1; 3160 } 3161 EXT4_I(inode)->i_disksize = inode->i_size; 3162 err = ext4_add_nondir(handle, dentry, inode); 3163 if (!err && IS_DIRSYNC(dir)) 3164 ext4_handle_sync(handle); 3165 3166 if (handle) 3167 ext4_journal_stop(handle); 3168 goto out_free_encrypted_link; 3169 3170 err_drop_inode: 3171 if (handle) 3172 ext4_journal_stop(handle); 3173 clear_nlink(inode); 3174 unlock_new_inode(inode); 3175 iput(inode); 3176 out_free_encrypted_link: 3177 if (disk_link.name != (unsigned char *)symname) 3178 kfree(disk_link.name); 3179 return err; 3180 } 3181 3182 static int ext4_link(struct dentry *old_dentry, 3183 struct inode *dir, struct dentry *dentry) 3184 { 3185 handle_t *handle; 3186 struct inode *inode = d_inode(old_dentry); 3187 int err, retries = 0; 3188 3189 if (inode->i_nlink >= EXT4_LINK_MAX) 3190 return -EMLINK; 3191 3192 err = fscrypt_prepare_link(old_dentry, dir, dentry); 3193 if (err) 3194 return err; 3195 3196 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) && 3197 (!projid_eq(EXT4_I(dir)->i_projid, 3198 EXT4_I(old_dentry->d_inode)->i_projid))) 3199 return -EXDEV; 3200 3201 err = dquot_initialize(dir); 3202 if (err) 3203 return err; 3204 3205 retry: 3206 handle = ext4_journal_start(dir, EXT4_HT_DIR, 3207 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 3208 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1); 3209 if (IS_ERR(handle)) 3210 return PTR_ERR(handle); 3211 3212 if (IS_DIRSYNC(dir)) 3213 ext4_handle_sync(handle); 3214 3215 inode->i_ctime = current_time(inode); 3216 ext4_inc_count(handle, inode); 3217 ihold(inode); 3218 3219 err = ext4_add_entry(handle, dentry, inode); 3220 if (!err) { 3221 ext4_mark_inode_dirty(handle, inode); 3222 /* this can happen only for tmpfile being 3223 * linked the first time 3224 */ 3225 if (inode->i_nlink == 1) 3226 ext4_orphan_del(handle, inode); 3227 d_instantiate(dentry, inode); 3228 } else { 3229 drop_nlink(inode); 3230 iput(inode); 3231 } 3232 ext4_journal_stop(handle); 3233 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 3234 goto retry; 3235 return err; 3236 } 3237 3238 3239 /* 3240 * Try to find buffer head where contains the parent block. 3241 * It should be the inode block if it is inlined or the 1st block 3242 * if it is a normal dir. 3243 */ 3244 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle, 3245 struct inode *inode, 3246 int *retval, 3247 struct ext4_dir_entry_2 **parent_de, 3248 int *inlined) 3249 { 3250 struct buffer_head *bh; 3251 3252 if (!ext4_has_inline_data(inode)) { 3253 bh = ext4_read_dirblock(inode, 0, EITHER); 3254 if (IS_ERR(bh)) { 3255 *retval = PTR_ERR(bh); 3256 return NULL; 3257 } 3258 *parent_de = ext4_next_entry( 3259 (struct ext4_dir_entry_2 *)bh->b_data, 3260 inode->i_sb->s_blocksize); 3261 return bh; 3262 } 3263 3264 *inlined = 1; 3265 return ext4_get_first_inline_block(inode, parent_de, retval); 3266 } 3267 3268 struct ext4_renament { 3269 struct inode *dir; 3270 struct dentry *dentry; 3271 struct inode *inode; 3272 bool is_dir; 3273 int dir_nlink_delta; 3274 3275 /* entry for "dentry" */ 3276 struct buffer_head *bh; 3277 struct ext4_dir_entry_2 *de; 3278 int inlined; 3279 3280 /* entry for ".." in inode if it's a directory */ 3281 struct buffer_head *dir_bh; 3282 struct ext4_dir_entry_2 *parent_de; 3283 int dir_inlined; 3284 }; 3285 3286 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent) 3287 { 3288 int retval; 3289 3290 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode, 3291 &retval, &ent->parent_de, 3292 &ent->dir_inlined); 3293 if (!ent->dir_bh) 3294 return retval; 3295 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino) 3296 return -EFSCORRUPTED; 3297 BUFFER_TRACE(ent->dir_bh, "get_write_access"); 3298 return ext4_journal_get_write_access(handle, ent->dir_bh); 3299 } 3300 3301 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent, 3302 unsigned dir_ino) 3303 { 3304 int retval; 3305 3306 ent->parent_de->inode = cpu_to_le32(dir_ino); 3307 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata"); 3308 if (!ent->dir_inlined) { 3309 if (is_dx(ent->inode)) { 3310 retval = ext4_handle_dirty_dx_node(handle, 3311 ent->inode, 3312 ent->dir_bh); 3313 } else { 3314 retval = ext4_handle_dirty_dirent_node(handle, 3315 ent->inode, 3316 ent->dir_bh); 3317 } 3318 } else { 3319 retval = ext4_mark_inode_dirty(handle, ent->inode); 3320 } 3321 if (retval) { 3322 ext4_std_error(ent->dir->i_sb, retval); 3323 return retval; 3324 } 3325 return 0; 3326 } 3327 3328 static int ext4_setent(handle_t *handle, struct ext4_renament *ent, 3329 unsigned ino, unsigned file_type) 3330 { 3331 int retval; 3332 3333 BUFFER_TRACE(ent->bh, "get write access"); 3334 retval = ext4_journal_get_write_access(handle, ent->bh); 3335 if (retval) 3336 return retval; 3337 ent->de->inode = cpu_to_le32(ino); 3338 if (ext4_has_feature_filetype(ent->dir->i_sb)) 3339 ent->de->file_type = file_type; 3340 inode_inc_iversion(ent->dir); 3341 ent->dir->i_ctime = ent->dir->i_mtime = 3342 current_time(ent->dir); 3343 ext4_mark_inode_dirty(handle, ent->dir); 3344 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata"); 3345 if (!ent->inlined) { 3346 retval = ext4_handle_dirty_dirent_node(handle, 3347 ent->dir, ent->bh); 3348 if (unlikely(retval)) { 3349 ext4_std_error(ent->dir->i_sb, retval); 3350 return retval; 3351 } 3352 } 3353 brelse(ent->bh); 3354 ent->bh = NULL; 3355 3356 return 0; 3357 } 3358 3359 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir, 3360 const struct qstr *d_name) 3361 { 3362 int retval = -ENOENT; 3363 struct buffer_head *bh; 3364 struct ext4_dir_entry_2 *de; 3365 3366 bh = ext4_find_entry(dir, d_name, &de, NULL); 3367 if (IS_ERR(bh)) 3368 return PTR_ERR(bh); 3369 if (bh) { 3370 retval = ext4_delete_entry(handle, dir, de, bh); 3371 brelse(bh); 3372 } 3373 return retval; 3374 } 3375 3376 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent, 3377 int force_reread) 3378 { 3379 int retval; 3380 /* 3381 * ent->de could have moved from under us during htree split, so make 3382 * sure that we are deleting the right entry. We might also be pointing 3383 * to a stale entry in the unused part of ent->bh so just checking inum 3384 * and the name isn't enough. 3385 */ 3386 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino || 3387 ent->de->name_len != ent->dentry->d_name.len || 3388 strncmp(ent->de->name, ent->dentry->d_name.name, 3389 ent->de->name_len) || 3390 force_reread) { 3391 retval = ext4_find_delete_entry(handle, ent->dir, 3392 &ent->dentry->d_name); 3393 } else { 3394 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh); 3395 if (retval == -ENOENT) { 3396 retval = ext4_find_delete_entry(handle, ent->dir, 3397 &ent->dentry->d_name); 3398 } 3399 } 3400 3401 if (retval) { 3402 ext4_warning_inode(ent->dir, 3403 "Deleting old file: nlink %d, error=%d", 3404 ent->dir->i_nlink, retval); 3405 } 3406 } 3407 3408 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent) 3409 { 3410 if (ent->dir_nlink_delta) { 3411 if (ent->dir_nlink_delta == -1) 3412 ext4_dec_count(handle, ent->dir); 3413 else 3414 ext4_inc_count(handle, ent->dir); 3415 ext4_mark_inode_dirty(handle, ent->dir); 3416 } 3417 } 3418 3419 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent, 3420 int credits, handle_t **h) 3421 { 3422 struct inode *wh; 3423 handle_t *handle; 3424 int retries = 0; 3425 3426 /* 3427 * for inode block, sb block, group summaries, 3428 * and inode bitmap 3429 */ 3430 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) + 3431 EXT4_XATTR_TRANS_BLOCKS + 4); 3432 retry: 3433 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE, 3434 &ent->dentry->d_name, 0, NULL, 3435 EXT4_HT_DIR, credits); 3436 3437 handle = ext4_journal_current_handle(); 3438 if (IS_ERR(wh)) { 3439 if (handle) 3440 ext4_journal_stop(handle); 3441 if (PTR_ERR(wh) == -ENOSPC && 3442 ext4_should_retry_alloc(ent->dir->i_sb, &retries)) 3443 goto retry; 3444 } else { 3445 *h = handle; 3446 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV); 3447 wh->i_op = &ext4_special_inode_operations; 3448 } 3449 return wh; 3450 } 3451 3452 /* 3453 * Anybody can rename anything with this: the permission checks are left to the 3454 * higher-level routines. 3455 * 3456 * n.b. old_{dentry,inode) refers to the source dentry/inode 3457 * while new_{dentry,inode) refers to the destination dentry/inode 3458 * This comes from rename(const char *oldpath, const char *newpath) 3459 */ 3460 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry, 3461 struct inode *new_dir, struct dentry *new_dentry, 3462 unsigned int flags) 3463 { 3464 handle_t *handle = NULL; 3465 struct ext4_renament old = { 3466 .dir = old_dir, 3467 .dentry = old_dentry, 3468 .inode = d_inode(old_dentry), 3469 }; 3470 struct ext4_renament new = { 3471 .dir = new_dir, 3472 .dentry = new_dentry, 3473 .inode = d_inode(new_dentry), 3474 }; 3475 int force_reread; 3476 int retval; 3477 struct inode *whiteout = NULL; 3478 int credits; 3479 u8 old_file_type; 3480 3481 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) && 3482 (!projid_eq(EXT4_I(new_dir)->i_projid, 3483 EXT4_I(old_dentry->d_inode)->i_projid))) 3484 return -EXDEV; 3485 3486 retval = dquot_initialize(old.dir); 3487 if (retval) 3488 return retval; 3489 retval = dquot_initialize(new.dir); 3490 if (retval) 3491 return retval; 3492 3493 /* Initialize quotas before so that eventual writes go 3494 * in separate transaction */ 3495 if (new.inode) { 3496 retval = dquot_initialize(new.inode); 3497 if (retval) 3498 return retval; 3499 } 3500 3501 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL); 3502 if (IS_ERR(old.bh)) 3503 return PTR_ERR(old.bh); 3504 /* 3505 * Check for inode number is _not_ due to possible IO errors. 3506 * We might rmdir the source, keep it as pwd of some process 3507 * and merrily kill the link to whatever was created under the 3508 * same name. Goodbye sticky bit ;-< 3509 */ 3510 retval = -ENOENT; 3511 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino) 3512 goto end_rename; 3513 3514 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name, 3515 &new.de, &new.inlined); 3516 if (IS_ERR(new.bh)) { 3517 retval = PTR_ERR(new.bh); 3518 new.bh = NULL; 3519 goto end_rename; 3520 } 3521 if (new.bh) { 3522 if (!new.inode) { 3523 brelse(new.bh); 3524 new.bh = NULL; 3525 } 3526 } 3527 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC)) 3528 ext4_alloc_da_blocks(old.inode); 3529 3530 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) + 3531 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2); 3532 if (!(flags & RENAME_WHITEOUT)) { 3533 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits); 3534 if (IS_ERR(handle)) { 3535 retval = PTR_ERR(handle); 3536 handle = NULL; 3537 goto end_rename; 3538 } 3539 } else { 3540 whiteout = ext4_whiteout_for_rename(&old, credits, &handle); 3541 if (IS_ERR(whiteout)) { 3542 retval = PTR_ERR(whiteout); 3543 whiteout = NULL; 3544 goto end_rename; 3545 } 3546 } 3547 3548 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir)) 3549 ext4_handle_sync(handle); 3550 3551 if (S_ISDIR(old.inode->i_mode)) { 3552 if (new.inode) { 3553 retval = -ENOTEMPTY; 3554 if (!ext4_empty_dir(new.inode)) 3555 goto end_rename; 3556 } else { 3557 retval = -EMLINK; 3558 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir)) 3559 goto end_rename; 3560 } 3561 retval = ext4_rename_dir_prepare(handle, &old); 3562 if (retval) 3563 goto end_rename; 3564 } 3565 /* 3566 * If we're renaming a file within an inline_data dir and adding or 3567 * setting the new dirent causes a conversion from inline_data to 3568 * extents/blockmap, we need to force the dirent delete code to 3569 * re-read the directory, or else we end up trying to delete a dirent 3570 * from what is now the extent tree root (or a block map). 3571 */ 3572 force_reread = (new.dir->i_ino == old.dir->i_ino && 3573 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA)); 3574 3575 old_file_type = old.de->file_type; 3576 if (whiteout) { 3577 /* 3578 * Do this before adding a new entry, so the old entry is sure 3579 * to be still pointing to the valid old entry. 3580 */ 3581 retval = ext4_setent(handle, &old, whiteout->i_ino, 3582 EXT4_FT_CHRDEV); 3583 if (retval) 3584 goto end_rename; 3585 ext4_mark_inode_dirty(handle, whiteout); 3586 } 3587 if (!new.bh) { 3588 retval = ext4_add_entry(handle, new.dentry, old.inode); 3589 if (retval) 3590 goto end_rename; 3591 } else { 3592 retval = ext4_setent(handle, &new, 3593 old.inode->i_ino, old_file_type); 3594 if (retval) 3595 goto end_rename; 3596 } 3597 if (force_reread) 3598 force_reread = !ext4_test_inode_flag(new.dir, 3599 EXT4_INODE_INLINE_DATA); 3600 3601 /* 3602 * Like most other Unix systems, set the ctime for inodes on a 3603 * rename. 3604 */ 3605 old.inode->i_ctime = current_time(old.inode); 3606 ext4_mark_inode_dirty(handle, old.inode); 3607 3608 if (!whiteout) { 3609 /* 3610 * ok, that's it 3611 */ 3612 ext4_rename_delete(handle, &old, force_reread); 3613 } 3614 3615 if (new.inode) { 3616 ext4_dec_count(handle, new.inode); 3617 new.inode->i_ctime = current_time(new.inode); 3618 } 3619 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir); 3620 ext4_update_dx_flag(old.dir); 3621 if (old.dir_bh) { 3622 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino); 3623 if (retval) 3624 goto end_rename; 3625 3626 ext4_dec_count(handle, old.dir); 3627 if (new.inode) { 3628 /* checked ext4_empty_dir above, can't have another 3629 * parent, ext4_dec_count() won't work for many-linked 3630 * dirs */ 3631 clear_nlink(new.inode); 3632 } else { 3633 ext4_inc_count(handle, new.dir); 3634 ext4_update_dx_flag(new.dir); 3635 ext4_mark_inode_dirty(handle, new.dir); 3636 } 3637 } 3638 ext4_mark_inode_dirty(handle, old.dir); 3639 if (new.inode) { 3640 ext4_mark_inode_dirty(handle, new.inode); 3641 if (!new.inode->i_nlink) 3642 ext4_orphan_add(handle, new.inode); 3643 } 3644 retval = 0; 3645 3646 end_rename: 3647 brelse(old.dir_bh); 3648 brelse(old.bh); 3649 brelse(new.bh); 3650 if (whiteout) { 3651 if (retval) 3652 drop_nlink(whiteout); 3653 unlock_new_inode(whiteout); 3654 iput(whiteout); 3655 } 3656 if (handle) 3657 ext4_journal_stop(handle); 3658 return retval; 3659 } 3660 3661 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry, 3662 struct inode *new_dir, struct dentry *new_dentry) 3663 { 3664 handle_t *handle = NULL; 3665 struct ext4_renament old = { 3666 .dir = old_dir, 3667 .dentry = old_dentry, 3668 .inode = d_inode(old_dentry), 3669 }; 3670 struct ext4_renament new = { 3671 .dir = new_dir, 3672 .dentry = new_dentry, 3673 .inode = d_inode(new_dentry), 3674 }; 3675 u8 new_file_type; 3676 int retval; 3677 struct timespec64 ctime; 3678 3679 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) && 3680 !projid_eq(EXT4_I(new_dir)->i_projid, 3681 EXT4_I(old_dentry->d_inode)->i_projid)) || 3682 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) && 3683 !projid_eq(EXT4_I(old_dir)->i_projid, 3684 EXT4_I(new_dentry->d_inode)->i_projid))) 3685 return -EXDEV; 3686 3687 retval = dquot_initialize(old.dir); 3688 if (retval) 3689 return retval; 3690 retval = dquot_initialize(new.dir); 3691 if (retval) 3692 return retval; 3693 3694 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, 3695 &old.de, &old.inlined); 3696 if (IS_ERR(old.bh)) 3697 return PTR_ERR(old.bh); 3698 /* 3699 * Check for inode number is _not_ due to possible IO errors. 3700 * We might rmdir the source, keep it as pwd of some process 3701 * and merrily kill the link to whatever was created under the 3702 * same name. Goodbye sticky bit ;-< 3703 */ 3704 retval = -ENOENT; 3705 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino) 3706 goto end_rename; 3707 3708 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name, 3709 &new.de, &new.inlined); 3710 if (IS_ERR(new.bh)) { 3711 retval = PTR_ERR(new.bh); 3712 new.bh = NULL; 3713 goto end_rename; 3714 } 3715 3716 /* RENAME_EXCHANGE case: old *and* new must both exist */ 3717 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino) 3718 goto end_rename; 3719 3720 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, 3721 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) + 3722 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2)); 3723 if (IS_ERR(handle)) { 3724 retval = PTR_ERR(handle); 3725 handle = NULL; 3726 goto end_rename; 3727 } 3728 3729 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir)) 3730 ext4_handle_sync(handle); 3731 3732 if (S_ISDIR(old.inode->i_mode)) { 3733 old.is_dir = true; 3734 retval = ext4_rename_dir_prepare(handle, &old); 3735 if (retval) 3736 goto end_rename; 3737 } 3738 if (S_ISDIR(new.inode->i_mode)) { 3739 new.is_dir = true; 3740 retval = ext4_rename_dir_prepare(handle, &new); 3741 if (retval) 3742 goto end_rename; 3743 } 3744 3745 /* 3746 * Other than the special case of overwriting a directory, parents' 3747 * nlink only needs to be modified if this is a cross directory rename. 3748 */ 3749 if (old.dir != new.dir && old.is_dir != new.is_dir) { 3750 old.dir_nlink_delta = old.is_dir ? -1 : 1; 3751 new.dir_nlink_delta = -old.dir_nlink_delta; 3752 retval = -EMLINK; 3753 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) || 3754 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir))) 3755 goto end_rename; 3756 } 3757 3758 new_file_type = new.de->file_type; 3759 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type); 3760 if (retval) 3761 goto end_rename; 3762 3763 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type); 3764 if (retval) 3765 goto end_rename; 3766 3767 /* 3768 * Like most other Unix systems, set the ctime for inodes on a 3769 * rename. 3770 */ 3771 ctime = current_time(old.inode); 3772 old.inode->i_ctime = ctime; 3773 new.inode->i_ctime = ctime; 3774 ext4_mark_inode_dirty(handle, old.inode); 3775 ext4_mark_inode_dirty(handle, new.inode); 3776 3777 if (old.dir_bh) { 3778 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino); 3779 if (retval) 3780 goto end_rename; 3781 } 3782 if (new.dir_bh) { 3783 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino); 3784 if (retval) 3785 goto end_rename; 3786 } 3787 ext4_update_dir_count(handle, &old); 3788 ext4_update_dir_count(handle, &new); 3789 retval = 0; 3790 3791 end_rename: 3792 brelse(old.dir_bh); 3793 brelse(new.dir_bh); 3794 brelse(old.bh); 3795 brelse(new.bh); 3796 if (handle) 3797 ext4_journal_stop(handle); 3798 return retval; 3799 } 3800 3801 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry, 3802 struct inode *new_dir, struct dentry *new_dentry, 3803 unsigned int flags) 3804 { 3805 int err; 3806 3807 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb)))) 3808 return -EIO; 3809 3810 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 3811 return -EINVAL; 3812 3813 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry, 3814 flags); 3815 if (err) 3816 return err; 3817 3818 if (flags & RENAME_EXCHANGE) { 3819 return ext4_cross_rename(old_dir, old_dentry, 3820 new_dir, new_dentry); 3821 } 3822 3823 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags); 3824 } 3825 3826 /* 3827 * directories can handle most operations... 3828 */ 3829 const struct inode_operations ext4_dir_inode_operations = { 3830 .create = ext4_create, 3831 .lookup = ext4_lookup, 3832 .link = ext4_link, 3833 .unlink = ext4_unlink, 3834 .symlink = ext4_symlink, 3835 .mkdir = ext4_mkdir, 3836 .rmdir = ext4_rmdir, 3837 .mknod = ext4_mknod, 3838 .tmpfile = ext4_tmpfile, 3839 .rename = ext4_rename2, 3840 .setattr = ext4_setattr, 3841 .getattr = ext4_getattr, 3842 .listxattr = ext4_listxattr, 3843 .get_acl = ext4_get_acl, 3844 .set_acl = ext4_set_acl, 3845 .fiemap = ext4_fiemap, 3846 }; 3847 3848 const struct inode_operations ext4_special_inode_operations = { 3849 .setattr = ext4_setattr, 3850 .getattr = ext4_getattr, 3851 .listxattr = ext4_listxattr, 3852 .get_acl = ext4_get_acl, 3853 .set_acl = ext4_set_acl, 3854 }; 3855