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 brelse(bh); 130 return ERR_PTR(-EFSCORRUPTED); 131 } 132 if (!ext4_has_metadata_csum(inode->i_sb) || 133 buffer_verified(bh)) 134 return bh; 135 136 /* 137 * An empty leaf block can get mistaken for a index block; for 138 * this reason, we can only check the index checksum when the 139 * caller is sure it should be an index block. 140 */ 141 if (is_dx_block && type == INDEX) { 142 if (ext4_dx_csum_verify(inode, dirent)) 143 set_buffer_verified(bh); 144 else { 145 ext4_error_inode(inode, func, line, block, 146 "Directory index failed checksum"); 147 brelse(bh); 148 return ERR_PTR(-EFSBADCRC); 149 } 150 } 151 if (!is_dx_block) { 152 if (ext4_dirent_csum_verify(inode, dirent)) 153 set_buffer_verified(bh); 154 else { 155 ext4_error_inode(inode, func, line, block, 156 "Directory block failed checksum"); 157 brelse(bh); 158 return ERR_PTR(-EFSBADCRC); 159 } 160 } 161 return bh; 162 } 163 164 #ifndef assert 165 #define assert(test) J_ASSERT(test) 166 #endif 167 168 #ifdef DX_DEBUG 169 #define dxtrace(command) command 170 #else 171 #define dxtrace(command) 172 #endif 173 174 struct fake_dirent 175 { 176 __le32 inode; 177 __le16 rec_len; 178 u8 name_len; 179 u8 file_type; 180 }; 181 182 struct dx_countlimit 183 { 184 __le16 limit; 185 __le16 count; 186 }; 187 188 struct dx_entry 189 { 190 __le32 hash; 191 __le32 block; 192 }; 193 194 /* 195 * dx_root_info is laid out so that if it should somehow get overlaid by a 196 * dirent the two low bits of the hash version will be zero. Therefore, the 197 * hash version mod 4 should never be 0. Sincerely, the paranoia department. 198 */ 199 200 struct dx_root 201 { 202 struct fake_dirent dot; 203 char dot_name[4]; 204 struct fake_dirent dotdot; 205 char dotdot_name[4]; 206 struct dx_root_info 207 { 208 __le32 reserved_zero; 209 u8 hash_version; 210 u8 info_length; /* 8 */ 211 u8 indirect_levels; 212 u8 unused_flags; 213 } 214 info; 215 struct dx_entry entries[0]; 216 }; 217 218 struct dx_node 219 { 220 struct fake_dirent fake; 221 struct dx_entry entries[0]; 222 }; 223 224 225 struct dx_frame 226 { 227 struct buffer_head *bh; 228 struct dx_entry *entries; 229 struct dx_entry *at; 230 }; 231 232 struct dx_map_entry 233 { 234 u32 hash; 235 u16 offs; 236 u16 size; 237 }; 238 239 /* 240 * This goes at the end of each htree block. 241 */ 242 struct dx_tail { 243 u32 dt_reserved; 244 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */ 245 }; 246 247 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry); 248 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value); 249 static inline unsigned dx_get_hash(struct dx_entry *entry); 250 static void dx_set_hash(struct dx_entry *entry, unsigned value); 251 static unsigned dx_get_count(struct dx_entry *entries); 252 static unsigned dx_get_limit(struct dx_entry *entries); 253 static void dx_set_count(struct dx_entry *entries, unsigned value); 254 static void dx_set_limit(struct dx_entry *entries, unsigned value); 255 static unsigned dx_root_limit(struct inode *dir, unsigned infosize); 256 static unsigned dx_node_limit(struct inode *dir); 257 static struct dx_frame *dx_probe(struct ext4_filename *fname, 258 struct inode *dir, 259 struct dx_hash_info *hinfo, 260 struct dx_frame *frame); 261 static void dx_release(struct dx_frame *frames); 262 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de, 263 unsigned blocksize, struct dx_hash_info *hinfo, 264 struct dx_map_entry map[]); 265 static void dx_sort_map(struct dx_map_entry *map, unsigned count); 266 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to, 267 struct dx_map_entry *offsets, int count, unsigned blocksize); 268 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize); 269 static void dx_insert_block(struct dx_frame *frame, 270 u32 hash, ext4_lblk_t block); 271 static int ext4_htree_next_block(struct inode *dir, __u32 hash, 272 struct dx_frame *frame, 273 struct dx_frame *frames, 274 __u32 *start_hash); 275 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, 276 struct ext4_filename *fname, 277 struct ext4_dir_entry_2 **res_dir); 278 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname, 279 struct inode *dir, struct inode *inode); 280 281 /* checksumming functions */ 282 void initialize_dirent_tail(struct ext4_dir_entry_tail *t, 283 unsigned int blocksize) 284 { 285 memset(t, 0, sizeof(struct ext4_dir_entry_tail)); 286 t->det_rec_len = ext4_rec_len_to_disk( 287 sizeof(struct ext4_dir_entry_tail), blocksize); 288 t->det_reserved_ft = EXT4_FT_DIR_CSUM; 289 } 290 291 /* Walk through a dirent block to find a checksum "dirent" at the tail */ 292 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode, 293 struct ext4_dir_entry *de) 294 { 295 struct ext4_dir_entry_tail *t; 296 297 #ifdef PARANOID 298 struct ext4_dir_entry *d, *top; 299 300 d = de; 301 top = (struct ext4_dir_entry *)(((void *)de) + 302 (EXT4_BLOCK_SIZE(inode->i_sb) - 303 sizeof(struct ext4_dir_entry_tail))); 304 while (d < top && d->rec_len) 305 d = (struct ext4_dir_entry *)(((void *)d) + 306 le16_to_cpu(d->rec_len)); 307 308 if (d != top) 309 return NULL; 310 311 t = (struct ext4_dir_entry_tail *)d; 312 #else 313 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb)); 314 #endif 315 316 if (t->det_reserved_zero1 || 317 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) || 318 t->det_reserved_zero2 || 319 t->det_reserved_ft != EXT4_FT_DIR_CSUM) 320 return NULL; 321 322 return t; 323 } 324 325 static __le32 ext4_dirent_csum(struct inode *inode, 326 struct ext4_dir_entry *dirent, int size) 327 { 328 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 329 struct ext4_inode_info *ei = EXT4_I(inode); 330 __u32 csum; 331 332 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size); 333 return cpu_to_le32(csum); 334 } 335 336 #define warn_no_space_for_csum(inode) \ 337 __warn_no_space_for_csum((inode), __func__, __LINE__) 338 339 static void __warn_no_space_for_csum(struct inode *inode, const char *func, 340 unsigned int line) 341 { 342 __ext4_warning_inode(inode, func, line, 343 "No space for directory leaf checksum. Please run e2fsck -D."); 344 } 345 346 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent) 347 { 348 struct ext4_dir_entry_tail *t; 349 350 if (!ext4_has_metadata_csum(inode->i_sb)) 351 return 1; 352 353 t = get_dirent_tail(inode, dirent); 354 if (!t) { 355 warn_no_space_for_csum(inode); 356 return 0; 357 } 358 359 if (t->det_checksum != ext4_dirent_csum(inode, dirent, 360 (void *)t - (void *)dirent)) 361 return 0; 362 363 return 1; 364 } 365 366 static void ext4_dirent_csum_set(struct inode *inode, 367 struct ext4_dir_entry *dirent) 368 { 369 struct ext4_dir_entry_tail *t; 370 371 if (!ext4_has_metadata_csum(inode->i_sb)) 372 return; 373 374 t = get_dirent_tail(inode, dirent); 375 if (!t) { 376 warn_no_space_for_csum(inode); 377 return; 378 } 379 380 t->det_checksum = ext4_dirent_csum(inode, dirent, 381 (void *)t - (void *)dirent); 382 } 383 384 int ext4_handle_dirty_dirent_node(handle_t *handle, 385 struct inode *inode, 386 struct buffer_head *bh) 387 { 388 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data); 389 return ext4_handle_dirty_metadata(handle, inode, bh); 390 } 391 392 static struct dx_countlimit *get_dx_countlimit(struct inode *inode, 393 struct ext4_dir_entry *dirent, 394 int *offset) 395 { 396 struct ext4_dir_entry *dp; 397 struct dx_root_info *root; 398 int count_offset; 399 400 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb)) 401 count_offset = 8; 402 else if (le16_to_cpu(dirent->rec_len) == 12) { 403 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12); 404 if (le16_to_cpu(dp->rec_len) != 405 EXT4_BLOCK_SIZE(inode->i_sb) - 12) 406 return NULL; 407 root = (struct dx_root_info *)(((void *)dp + 12)); 408 if (root->reserved_zero || 409 root->info_length != sizeof(struct dx_root_info)) 410 return NULL; 411 count_offset = 32; 412 } else 413 return NULL; 414 415 if (offset) 416 *offset = count_offset; 417 return (struct dx_countlimit *)(((void *)dirent) + count_offset); 418 } 419 420 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent, 421 int count_offset, int count, struct dx_tail *t) 422 { 423 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 424 struct ext4_inode_info *ei = EXT4_I(inode); 425 __u32 csum; 426 int size; 427 __u32 dummy_csum = 0; 428 int offset = offsetof(struct dx_tail, dt_checksum); 429 430 size = count_offset + (count * sizeof(struct dx_entry)); 431 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size); 432 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset); 433 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum)); 434 435 return cpu_to_le32(csum); 436 } 437 438 static int ext4_dx_csum_verify(struct inode *inode, 439 struct ext4_dir_entry *dirent) 440 { 441 struct dx_countlimit *c; 442 struct dx_tail *t; 443 int count_offset, limit, count; 444 445 if (!ext4_has_metadata_csum(inode->i_sb)) 446 return 1; 447 448 c = get_dx_countlimit(inode, dirent, &count_offset); 449 if (!c) { 450 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D."); 451 return 0; 452 } 453 limit = le16_to_cpu(c->limit); 454 count = le16_to_cpu(c->count); 455 if (count_offset + (limit * sizeof(struct dx_entry)) > 456 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) { 457 warn_no_space_for_csum(inode); 458 return 0; 459 } 460 t = (struct dx_tail *)(((struct dx_entry *)c) + limit); 461 462 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset, 463 count, t)) 464 return 0; 465 return 1; 466 } 467 468 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent) 469 { 470 struct dx_countlimit *c; 471 struct dx_tail *t; 472 int count_offset, limit, count; 473 474 if (!ext4_has_metadata_csum(inode->i_sb)) 475 return; 476 477 c = get_dx_countlimit(inode, dirent, &count_offset); 478 if (!c) { 479 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D."); 480 return; 481 } 482 limit = le16_to_cpu(c->limit); 483 count = le16_to_cpu(c->count); 484 if (count_offset + (limit * sizeof(struct dx_entry)) > 485 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) { 486 warn_no_space_for_csum(inode); 487 return; 488 } 489 t = (struct dx_tail *)(((struct dx_entry *)c) + limit); 490 491 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t); 492 } 493 494 static inline int ext4_handle_dirty_dx_node(handle_t *handle, 495 struct inode *inode, 496 struct buffer_head *bh) 497 { 498 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data); 499 return ext4_handle_dirty_metadata(handle, inode, bh); 500 } 501 502 /* 503 * p is at least 6 bytes before the end of page 504 */ 505 static inline struct ext4_dir_entry_2 * 506 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize) 507 { 508 return (struct ext4_dir_entry_2 *)((char *)p + 509 ext4_rec_len_from_disk(p->rec_len, blocksize)); 510 } 511 512 /* 513 * Future: use high four bits of block for coalesce-on-delete flags 514 * Mask them off for now. 515 */ 516 517 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry) 518 { 519 return le32_to_cpu(entry->block) & 0x0fffffff; 520 } 521 522 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value) 523 { 524 entry->block = cpu_to_le32(value); 525 } 526 527 static inline unsigned dx_get_hash(struct dx_entry *entry) 528 { 529 return le32_to_cpu(entry->hash); 530 } 531 532 static inline void dx_set_hash(struct dx_entry *entry, unsigned value) 533 { 534 entry->hash = cpu_to_le32(value); 535 } 536 537 static inline unsigned dx_get_count(struct dx_entry *entries) 538 { 539 return le16_to_cpu(((struct dx_countlimit *) entries)->count); 540 } 541 542 static inline unsigned dx_get_limit(struct dx_entry *entries) 543 { 544 return le16_to_cpu(((struct dx_countlimit *) entries)->limit); 545 } 546 547 static inline void dx_set_count(struct dx_entry *entries, unsigned value) 548 { 549 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value); 550 } 551 552 static inline void dx_set_limit(struct dx_entry *entries, unsigned value) 553 { 554 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value); 555 } 556 557 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize) 558 { 559 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) - 560 EXT4_DIR_REC_LEN(2) - infosize; 561 562 if (ext4_has_metadata_csum(dir->i_sb)) 563 entry_space -= sizeof(struct dx_tail); 564 return entry_space / sizeof(struct dx_entry); 565 } 566 567 static inline unsigned dx_node_limit(struct inode *dir) 568 { 569 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0); 570 571 if (ext4_has_metadata_csum(dir->i_sb)) 572 entry_space -= sizeof(struct dx_tail); 573 return entry_space / sizeof(struct dx_entry); 574 } 575 576 /* 577 * Debug 578 */ 579 #ifdef DX_DEBUG 580 static void dx_show_index(char * label, struct dx_entry *entries) 581 { 582 int i, n = dx_get_count (entries); 583 printk(KERN_DEBUG "%s index", label); 584 for (i = 0; i < n; i++) { 585 printk(KERN_CONT " %x->%lu", 586 i ? dx_get_hash(entries + i) : 0, 587 (unsigned long)dx_get_block(entries + i)); 588 } 589 printk(KERN_CONT "\n"); 590 } 591 592 struct stats 593 { 594 unsigned names; 595 unsigned space; 596 unsigned bcount; 597 }; 598 599 static struct stats dx_show_leaf(struct inode *dir, 600 struct dx_hash_info *hinfo, 601 struct ext4_dir_entry_2 *de, 602 int size, int show_names) 603 { 604 unsigned names = 0, space = 0; 605 char *base = (char *) de; 606 struct dx_hash_info h = *hinfo; 607 608 printk("names: "); 609 while ((char *) de < base + size) 610 { 611 if (de->inode) 612 { 613 if (show_names) 614 { 615 #ifdef CONFIG_EXT4_FS_ENCRYPTION 616 int len; 617 char *name; 618 struct fscrypt_str fname_crypto_str = 619 FSTR_INIT(NULL, 0); 620 int res = 0; 621 622 name = de->name; 623 len = de->name_len; 624 if (ext4_encrypted_inode(dir)) 625 res = fscrypt_get_encryption_info(dir); 626 if (res) { 627 printk(KERN_WARNING "Error setting up" 628 " fname crypto: %d\n", res); 629 } 630 if (!fscrypt_has_encryption_key(dir)) { 631 /* Directory is not encrypted */ 632 ext4fs_dirhash(de->name, 633 de->name_len, &h); 634 printk("%*.s:(U)%x.%u ", len, 635 name, h.hash, 636 (unsigned) ((char *) de 637 - base)); 638 } else { 639 struct fscrypt_str de_name = 640 FSTR_INIT(name, len); 641 642 /* Directory is encrypted */ 643 res = fscrypt_fname_alloc_buffer( 644 dir, len, 645 &fname_crypto_str); 646 if (res) 647 printk(KERN_WARNING "Error " 648 "allocating crypto " 649 "buffer--skipping " 650 "crypto\n"); 651 res = fscrypt_fname_disk_to_usr(dir, 652 0, 0, &de_name, 653 &fname_crypto_str); 654 if (res) { 655 printk(KERN_WARNING "Error " 656 "converting filename " 657 "from disk to usr" 658 "\n"); 659 name = "??"; 660 len = 2; 661 } else { 662 name = fname_crypto_str.name; 663 len = fname_crypto_str.len; 664 } 665 ext4fs_dirhash(de->name, de->name_len, 666 &h); 667 printk("%*.s:(E)%x.%u ", len, name, 668 h.hash, (unsigned) ((char *) de 669 - base)); 670 fscrypt_fname_free_buffer( 671 &fname_crypto_str); 672 } 673 #else 674 int len = de->name_len; 675 char *name = de->name; 676 ext4fs_dirhash(de->name, de->name_len, &h); 677 printk("%*.s:%x.%u ", len, name, h.hash, 678 (unsigned) ((char *) de - base)); 679 #endif 680 } 681 space += EXT4_DIR_REC_LEN(de->name_len); 682 names++; 683 } 684 de = ext4_next_entry(de, size); 685 } 686 printk(KERN_CONT "(%i)\n", names); 687 return (struct stats) { names, space, 1 }; 688 } 689 690 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir, 691 struct dx_entry *entries, int levels) 692 { 693 unsigned blocksize = dir->i_sb->s_blocksize; 694 unsigned count = dx_get_count(entries), names = 0, space = 0, i; 695 unsigned bcount = 0; 696 struct buffer_head *bh; 697 printk("%i indexed blocks...\n", count); 698 for (i = 0; i < count; i++, entries++) 699 { 700 ext4_lblk_t block = dx_get_block(entries); 701 ext4_lblk_t hash = i ? dx_get_hash(entries): 0; 702 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash; 703 struct stats stats; 704 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range); 705 bh = ext4_bread(NULL,dir, block, 0); 706 if (!bh || IS_ERR(bh)) 707 continue; 708 stats = levels? 709 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1): 710 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) 711 bh->b_data, blocksize, 0); 712 names += stats.names; 713 space += stats.space; 714 bcount += stats.bcount; 715 brelse(bh); 716 } 717 if (bcount) 718 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n", 719 levels ? "" : " ", names, space/bcount, 720 (space/bcount)*100/blocksize); 721 return (struct stats) { names, space, bcount}; 722 } 723 #endif /* DX_DEBUG */ 724 725 /* 726 * Probe for a directory leaf block to search. 727 * 728 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format 729 * error in the directory index, and the caller should fall back to 730 * searching the directory normally. The callers of dx_probe **MUST** 731 * check for this error code, and make sure it never gets reflected 732 * back to userspace. 733 */ 734 static struct dx_frame * 735 dx_probe(struct ext4_filename *fname, struct inode *dir, 736 struct dx_hash_info *hinfo, struct dx_frame *frame_in) 737 { 738 unsigned count, indirect; 739 struct dx_entry *at, *entries, *p, *q, *m; 740 struct dx_root *root; 741 struct dx_frame *frame = frame_in; 742 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR); 743 u32 hash; 744 745 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0])); 746 frame->bh = ext4_read_dirblock(dir, 0, INDEX); 747 if (IS_ERR(frame->bh)) 748 return (struct dx_frame *) frame->bh; 749 750 root = (struct dx_root *) frame->bh->b_data; 751 if (root->info.hash_version != DX_HASH_TEA && 752 root->info.hash_version != DX_HASH_HALF_MD4 && 753 root->info.hash_version != DX_HASH_LEGACY) { 754 ext4_warning_inode(dir, "Unrecognised inode hash code %u", 755 root->info.hash_version); 756 goto fail; 757 } 758 if (fname) 759 hinfo = &fname->hinfo; 760 hinfo->hash_version = root->info.hash_version; 761 if (hinfo->hash_version <= DX_HASH_TEA) 762 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; 763 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed; 764 if (fname && fname_name(fname)) 765 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo); 766 hash = hinfo->hash; 767 768 if (root->info.unused_flags & 1) { 769 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x", 770 root->info.unused_flags); 771 goto fail; 772 } 773 774 indirect = root->info.indirect_levels; 775 if (indirect >= ext4_dir_htree_level(dir->i_sb)) { 776 ext4_warning(dir->i_sb, 777 "Directory (ino: %lu) htree depth %#06x exceed" 778 "supported value", dir->i_ino, 779 ext4_dir_htree_level(dir->i_sb)); 780 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) { 781 ext4_warning(dir->i_sb, "Enable large directory " 782 "feature to access it"); 783 } 784 goto fail; 785 } 786 787 entries = (struct dx_entry *)(((char *)&root->info) + 788 root->info.info_length); 789 790 if (dx_get_limit(entries) != dx_root_limit(dir, 791 root->info.info_length)) { 792 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u", 793 dx_get_limit(entries), 794 dx_root_limit(dir, root->info.info_length)); 795 goto fail; 796 } 797 798 dxtrace(printk("Look up %x", hash)); 799 while (1) { 800 count = dx_get_count(entries); 801 if (!count || count > dx_get_limit(entries)) { 802 ext4_warning_inode(dir, 803 "dx entry: count %u beyond limit %u", 804 count, dx_get_limit(entries)); 805 goto fail; 806 } 807 808 p = entries + 1; 809 q = entries + count - 1; 810 while (p <= q) { 811 m = p + (q - p) / 2; 812 dxtrace(printk(KERN_CONT ".")); 813 if (dx_get_hash(m) > hash) 814 q = m - 1; 815 else 816 p = m + 1; 817 } 818 819 if (0) { // linear search cross check 820 unsigned n = count - 1; 821 at = entries; 822 while (n--) 823 { 824 dxtrace(printk(KERN_CONT ",")); 825 if (dx_get_hash(++at) > hash) 826 { 827 at--; 828 break; 829 } 830 } 831 assert (at == p - 1); 832 } 833 834 at = p - 1; 835 dxtrace(printk(KERN_CONT " %x->%u\n", 836 at == entries ? 0 : dx_get_hash(at), 837 dx_get_block(at))); 838 frame->entries = entries; 839 frame->at = at; 840 if (!indirect--) 841 return frame; 842 frame++; 843 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX); 844 if (IS_ERR(frame->bh)) { 845 ret_err = (struct dx_frame *) frame->bh; 846 frame->bh = NULL; 847 goto fail; 848 } 849 entries = ((struct dx_node *) frame->bh->b_data)->entries; 850 851 if (dx_get_limit(entries) != dx_node_limit(dir)) { 852 ext4_warning_inode(dir, 853 "dx entry: limit %u != node limit %u", 854 dx_get_limit(entries), dx_node_limit(dir)); 855 goto fail; 856 } 857 } 858 fail: 859 while (frame >= frame_in) { 860 brelse(frame->bh); 861 frame--; 862 } 863 864 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR)) 865 ext4_warning_inode(dir, 866 "Corrupt directory, running e2fsck is recommended"); 867 return ret_err; 868 } 869 870 static void dx_release(struct dx_frame *frames) 871 { 872 struct dx_root_info *info; 873 int i; 874 875 if (frames[0].bh == NULL) 876 return; 877 878 info = &((struct dx_root *)frames[0].bh->b_data)->info; 879 for (i = 0; i <= info->indirect_levels; i++) { 880 if (frames[i].bh == NULL) 881 break; 882 brelse(frames[i].bh); 883 frames[i].bh = NULL; 884 } 885 } 886 887 /* 888 * This function increments the frame pointer to search the next leaf 889 * block, and reads in the necessary intervening nodes if the search 890 * should be necessary. Whether or not the search is necessary is 891 * controlled by the hash parameter. If the hash value is even, then 892 * the search is only continued if the next block starts with that 893 * hash value. This is used if we are searching for a specific file. 894 * 895 * If the hash value is HASH_NB_ALWAYS, then always go to the next block. 896 * 897 * This function returns 1 if the caller should continue to search, 898 * or 0 if it should not. If there is an error reading one of the 899 * index blocks, it will a negative error code. 900 * 901 * If start_hash is non-null, it will be filled in with the starting 902 * hash of the next page. 903 */ 904 static int ext4_htree_next_block(struct inode *dir, __u32 hash, 905 struct dx_frame *frame, 906 struct dx_frame *frames, 907 __u32 *start_hash) 908 { 909 struct dx_frame *p; 910 struct buffer_head *bh; 911 int num_frames = 0; 912 __u32 bhash; 913 914 p = frame; 915 /* 916 * Find the next leaf page by incrementing the frame pointer. 917 * If we run out of entries in the interior node, loop around and 918 * increment pointer in the parent node. When we break out of 919 * this loop, num_frames indicates the number of interior 920 * nodes need to be read. 921 */ 922 while (1) { 923 if (++(p->at) < p->entries + dx_get_count(p->entries)) 924 break; 925 if (p == frames) 926 return 0; 927 num_frames++; 928 p--; 929 } 930 931 /* 932 * If the hash is 1, then continue only if the next page has a 933 * continuation hash of any value. This is used for readdir 934 * handling. Otherwise, check to see if the hash matches the 935 * desired contiuation hash. If it doesn't, return since 936 * there's no point to read in the successive index pages. 937 */ 938 bhash = dx_get_hash(p->at); 939 if (start_hash) 940 *start_hash = bhash; 941 if ((hash & 1) == 0) { 942 if ((bhash & ~1) != hash) 943 return 0; 944 } 945 /* 946 * If the hash is HASH_NB_ALWAYS, we always go to the next 947 * block so no check is necessary 948 */ 949 while (num_frames--) { 950 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX); 951 if (IS_ERR(bh)) 952 return PTR_ERR(bh); 953 p++; 954 brelse(p->bh); 955 p->bh = bh; 956 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries; 957 } 958 return 1; 959 } 960 961 962 /* 963 * This function fills a red-black tree with information from a 964 * directory block. It returns the number directory entries loaded 965 * into the tree. If there is an error it is returned in err. 966 */ 967 static int htree_dirblock_to_tree(struct file *dir_file, 968 struct inode *dir, ext4_lblk_t block, 969 struct dx_hash_info *hinfo, 970 __u32 start_hash, __u32 start_minor_hash) 971 { 972 struct buffer_head *bh; 973 struct ext4_dir_entry_2 *de, *top; 974 int err = 0, count = 0; 975 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str; 976 977 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n", 978 (unsigned long)block)); 979 bh = ext4_read_dirblock(dir, block, DIRENT); 980 if (IS_ERR(bh)) 981 return PTR_ERR(bh); 982 983 de = (struct ext4_dir_entry_2 *) bh->b_data; 984 top = (struct ext4_dir_entry_2 *) ((char *) de + 985 dir->i_sb->s_blocksize - 986 EXT4_DIR_REC_LEN(0)); 987 #ifdef CONFIG_EXT4_FS_ENCRYPTION 988 /* Check if the directory is encrypted */ 989 if (ext4_encrypted_inode(dir)) { 990 err = fscrypt_get_encryption_info(dir); 991 if (err < 0) { 992 brelse(bh); 993 return err; 994 } 995 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN, 996 &fname_crypto_str); 997 if (err < 0) { 998 brelse(bh); 999 return err; 1000 } 1001 } 1002 #endif 1003 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) { 1004 if (ext4_check_dir_entry(dir, NULL, de, bh, 1005 bh->b_data, bh->b_size, 1006 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb)) 1007 + ((char *)de - bh->b_data))) { 1008 /* silently ignore the rest of the block */ 1009 break; 1010 } 1011 ext4fs_dirhash(de->name, de->name_len, hinfo); 1012 if ((hinfo->hash < start_hash) || 1013 ((hinfo->hash == start_hash) && 1014 (hinfo->minor_hash < start_minor_hash))) 1015 continue; 1016 if (de->inode == 0) 1017 continue; 1018 if (!ext4_encrypted_inode(dir)) { 1019 tmp_str.name = de->name; 1020 tmp_str.len = de->name_len; 1021 err = ext4_htree_store_dirent(dir_file, 1022 hinfo->hash, hinfo->minor_hash, de, 1023 &tmp_str); 1024 } else { 1025 int save_len = fname_crypto_str.len; 1026 struct fscrypt_str de_name = FSTR_INIT(de->name, 1027 de->name_len); 1028 1029 /* Directory is encrypted */ 1030 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash, 1031 hinfo->minor_hash, &de_name, 1032 &fname_crypto_str); 1033 if (err) { 1034 count = err; 1035 goto errout; 1036 } 1037 err = ext4_htree_store_dirent(dir_file, 1038 hinfo->hash, hinfo->minor_hash, de, 1039 &fname_crypto_str); 1040 fname_crypto_str.len = save_len; 1041 } 1042 if (err != 0) { 1043 count = err; 1044 goto errout; 1045 } 1046 count++; 1047 } 1048 errout: 1049 brelse(bh); 1050 #ifdef CONFIG_EXT4_FS_ENCRYPTION 1051 fscrypt_fname_free_buffer(&fname_crypto_str); 1052 #endif 1053 return count; 1054 } 1055 1056 1057 /* 1058 * This function fills a red-black tree with information from a 1059 * directory. We start scanning the directory in hash order, starting 1060 * at start_hash and start_minor_hash. 1061 * 1062 * This function returns the number of entries inserted into the tree, 1063 * or a negative error code. 1064 */ 1065 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash, 1066 __u32 start_minor_hash, __u32 *next_hash) 1067 { 1068 struct dx_hash_info hinfo; 1069 struct ext4_dir_entry_2 *de; 1070 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; 1071 struct inode *dir; 1072 ext4_lblk_t block; 1073 int count = 0; 1074 int ret, err; 1075 __u32 hashval; 1076 struct fscrypt_str tmp_str; 1077 1078 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n", 1079 start_hash, start_minor_hash)); 1080 dir = file_inode(dir_file); 1081 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) { 1082 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; 1083 if (hinfo.hash_version <= DX_HASH_TEA) 1084 hinfo.hash_version += 1085 EXT4_SB(dir->i_sb)->s_hash_unsigned; 1086 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; 1087 if (ext4_has_inline_data(dir)) { 1088 int has_inline_data = 1; 1089 count = htree_inlinedir_to_tree(dir_file, dir, 0, 1090 &hinfo, start_hash, 1091 start_minor_hash, 1092 &has_inline_data); 1093 if (has_inline_data) { 1094 *next_hash = ~0; 1095 return count; 1096 } 1097 } 1098 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo, 1099 start_hash, start_minor_hash); 1100 *next_hash = ~0; 1101 return count; 1102 } 1103 hinfo.hash = start_hash; 1104 hinfo.minor_hash = 0; 1105 frame = dx_probe(NULL, dir, &hinfo, frames); 1106 if (IS_ERR(frame)) 1107 return PTR_ERR(frame); 1108 1109 /* Add '.' and '..' from the htree header */ 1110 if (!start_hash && !start_minor_hash) { 1111 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; 1112 tmp_str.name = de->name; 1113 tmp_str.len = de->name_len; 1114 err = ext4_htree_store_dirent(dir_file, 0, 0, 1115 de, &tmp_str); 1116 if (err != 0) 1117 goto errout; 1118 count++; 1119 } 1120 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) { 1121 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; 1122 de = ext4_next_entry(de, dir->i_sb->s_blocksize); 1123 tmp_str.name = de->name; 1124 tmp_str.len = de->name_len; 1125 err = ext4_htree_store_dirent(dir_file, 2, 0, 1126 de, &tmp_str); 1127 if (err != 0) 1128 goto errout; 1129 count++; 1130 } 1131 1132 while (1) { 1133 if (fatal_signal_pending(current)) { 1134 err = -ERESTARTSYS; 1135 goto errout; 1136 } 1137 cond_resched(); 1138 block = dx_get_block(frame->at); 1139 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo, 1140 start_hash, start_minor_hash); 1141 if (ret < 0) { 1142 err = ret; 1143 goto errout; 1144 } 1145 count += ret; 1146 hashval = ~0; 1147 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS, 1148 frame, frames, &hashval); 1149 *next_hash = hashval; 1150 if (ret < 0) { 1151 err = ret; 1152 goto errout; 1153 } 1154 /* 1155 * Stop if: (a) there are no more entries, or 1156 * (b) we have inserted at least one entry and the 1157 * next hash value is not a continuation 1158 */ 1159 if ((ret == 0) || 1160 (count && ((hashval & 1) == 0))) 1161 break; 1162 } 1163 dx_release(frames); 1164 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, " 1165 "next hash: %x\n", count, *next_hash)); 1166 return count; 1167 errout: 1168 dx_release(frames); 1169 return (err); 1170 } 1171 1172 static inline int search_dirblock(struct buffer_head *bh, 1173 struct inode *dir, 1174 struct ext4_filename *fname, 1175 unsigned int offset, 1176 struct ext4_dir_entry_2 **res_dir) 1177 { 1178 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir, 1179 fname, offset, res_dir); 1180 } 1181 1182 /* 1183 * Directory block splitting, compacting 1184 */ 1185 1186 /* 1187 * Create map of hash values, offsets, and sizes, stored at end of block. 1188 * Returns number of entries mapped. 1189 */ 1190 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de, 1191 unsigned blocksize, struct dx_hash_info *hinfo, 1192 struct dx_map_entry *map_tail) 1193 { 1194 int count = 0; 1195 char *base = (char *) de; 1196 struct dx_hash_info h = *hinfo; 1197 1198 while ((char *) de < base + blocksize) { 1199 if (de->name_len && de->inode) { 1200 ext4fs_dirhash(de->name, de->name_len, &h); 1201 map_tail--; 1202 map_tail->hash = h.hash; 1203 map_tail->offs = ((char *) de - base)>>2; 1204 map_tail->size = le16_to_cpu(de->rec_len); 1205 count++; 1206 cond_resched(); 1207 } 1208 /* XXX: do we need to check rec_len == 0 case? -Chris */ 1209 de = ext4_next_entry(de, blocksize); 1210 } 1211 return count; 1212 } 1213 1214 /* Sort map by hash value */ 1215 static void dx_sort_map (struct dx_map_entry *map, unsigned count) 1216 { 1217 struct dx_map_entry *p, *q, *top = map + count - 1; 1218 int more; 1219 /* Combsort until bubble sort doesn't suck */ 1220 while (count > 2) { 1221 count = count*10/13; 1222 if (count - 9 < 2) /* 9, 10 -> 11 */ 1223 count = 11; 1224 for (p = top, q = p - count; q >= map; p--, q--) 1225 if (p->hash < q->hash) 1226 swap(*p, *q); 1227 } 1228 /* Garden variety bubble sort */ 1229 do { 1230 more = 0; 1231 q = top; 1232 while (q-- > map) { 1233 if (q[1].hash >= q[0].hash) 1234 continue; 1235 swap(*(q+1), *q); 1236 more = 1; 1237 } 1238 } while(more); 1239 } 1240 1241 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block) 1242 { 1243 struct dx_entry *entries = frame->entries; 1244 struct dx_entry *old = frame->at, *new = old + 1; 1245 int count = dx_get_count(entries); 1246 1247 assert(count < dx_get_limit(entries)); 1248 assert(old < entries + count); 1249 memmove(new + 1, new, (char *)(entries + count) - (char *)(new)); 1250 dx_set_hash(new, hash); 1251 dx_set_block(new, block); 1252 dx_set_count(entries, count + 1); 1253 } 1254 1255 /* 1256 * Test whether a directory entry matches the filename being searched for. 1257 * 1258 * Return: %true if the directory entry matches, otherwise %false. 1259 */ 1260 static inline bool ext4_match(const struct ext4_filename *fname, 1261 const struct ext4_dir_entry_2 *de) 1262 { 1263 struct fscrypt_name f; 1264 1265 if (!de->inode) 1266 return false; 1267 1268 f.usr_fname = fname->usr_fname; 1269 f.disk_name = fname->disk_name; 1270 #ifdef CONFIG_EXT4_FS_ENCRYPTION 1271 f.crypto_buf = fname->crypto_buf; 1272 #endif 1273 return fscrypt_match_name(&f, de->name, de->name_len); 1274 } 1275 1276 /* 1277 * Returns 0 if not found, -1 on failure, and 1 on success 1278 */ 1279 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size, 1280 struct inode *dir, struct ext4_filename *fname, 1281 unsigned int offset, struct ext4_dir_entry_2 **res_dir) 1282 { 1283 struct ext4_dir_entry_2 * de; 1284 char * dlimit; 1285 int de_len; 1286 1287 de = (struct ext4_dir_entry_2 *)search_buf; 1288 dlimit = search_buf + buf_size; 1289 while ((char *) de < dlimit) { 1290 /* this code is executed quadratically often */ 1291 /* do minimal checking `by hand' */ 1292 if ((char *) de + de->name_len <= dlimit && 1293 ext4_match(fname, de)) { 1294 /* found a match - just to be sure, do 1295 * a full check */ 1296 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data, 1297 bh->b_size, offset)) 1298 return -1; 1299 *res_dir = de; 1300 return 1; 1301 } 1302 /* prevent looping on a bad block */ 1303 de_len = ext4_rec_len_from_disk(de->rec_len, 1304 dir->i_sb->s_blocksize); 1305 if (de_len <= 0) 1306 return -1; 1307 offset += de_len; 1308 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len); 1309 } 1310 return 0; 1311 } 1312 1313 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block, 1314 struct ext4_dir_entry *de) 1315 { 1316 struct super_block *sb = dir->i_sb; 1317 1318 if (!is_dx(dir)) 1319 return 0; 1320 if (block == 0) 1321 return 1; 1322 if (de->inode == 0 && 1323 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) == 1324 sb->s_blocksize) 1325 return 1; 1326 return 0; 1327 } 1328 1329 /* 1330 * ext4_find_entry() 1331 * 1332 * finds an entry in the specified directory with the wanted name. It 1333 * returns the cache buffer in which the entry was found, and the entry 1334 * itself (as a parameter - res_dir). It does NOT read the inode of the 1335 * entry - you'll have to do that yourself if you want to. 1336 * 1337 * The returned buffer_head has ->b_count elevated. The caller is expected 1338 * to brelse() it when appropriate. 1339 */ 1340 static struct buffer_head * ext4_find_entry (struct inode *dir, 1341 const struct qstr *d_name, 1342 struct ext4_dir_entry_2 **res_dir, 1343 int *inlined) 1344 { 1345 struct super_block *sb; 1346 struct buffer_head *bh_use[NAMEI_RA_SIZE]; 1347 struct buffer_head *bh, *ret = NULL; 1348 ext4_lblk_t start, block; 1349 const u8 *name = d_name->name; 1350 size_t ra_max = 0; /* Number of bh's in the readahead 1351 buffer, bh_use[] */ 1352 size_t ra_ptr = 0; /* Current index into readahead 1353 buffer */ 1354 ext4_lblk_t nblocks; 1355 int i, namelen, retval; 1356 struct ext4_filename fname; 1357 1358 *res_dir = NULL; 1359 sb = dir->i_sb; 1360 namelen = d_name->len; 1361 if (namelen > EXT4_NAME_LEN) 1362 return NULL; 1363 1364 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname); 1365 if (retval == -ENOENT) 1366 return NULL; 1367 if (retval) 1368 return ERR_PTR(retval); 1369 1370 if (ext4_has_inline_data(dir)) { 1371 int has_inline_data = 1; 1372 ret = ext4_find_inline_entry(dir, &fname, res_dir, 1373 &has_inline_data); 1374 if (has_inline_data) { 1375 if (inlined) 1376 *inlined = 1; 1377 goto cleanup_and_exit; 1378 } 1379 } 1380 1381 if ((namelen <= 2) && (name[0] == '.') && 1382 (name[1] == '.' || name[1] == '\0')) { 1383 /* 1384 * "." or ".." will only be in the first block 1385 * NFS may look up ".."; "." should be handled by the VFS 1386 */ 1387 block = start = 0; 1388 nblocks = 1; 1389 goto restart; 1390 } 1391 if (is_dx(dir)) { 1392 ret = ext4_dx_find_entry(dir, &fname, res_dir); 1393 /* 1394 * On success, or if the error was file not found, 1395 * return. Otherwise, fall back to doing a search the 1396 * old fashioned way. 1397 */ 1398 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR) 1399 goto cleanup_and_exit; 1400 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, " 1401 "falling back\n")); 1402 ret = NULL; 1403 } 1404 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); 1405 if (!nblocks) { 1406 ret = NULL; 1407 goto cleanup_and_exit; 1408 } 1409 start = EXT4_I(dir)->i_dir_start_lookup; 1410 if (start >= nblocks) 1411 start = 0; 1412 block = start; 1413 restart: 1414 do { 1415 /* 1416 * We deal with the read-ahead logic here. 1417 */ 1418 if (ra_ptr >= ra_max) { 1419 /* Refill the readahead buffer */ 1420 ra_ptr = 0; 1421 if (block < start) 1422 ra_max = start - block; 1423 else 1424 ra_max = nblocks - block; 1425 ra_max = min(ra_max, ARRAY_SIZE(bh_use)); 1426 retval = ext4_bread_batch(dir, block, ra_max, 1427 false /* wait */, bh_use); 1428 if (retval) { 1429 ret = ERR_PTR(retval); 1430 ra_max = 0; 1431 goto cleanup_and_exit; 1432 } 1433 } 1434 if ((bh = bh_use[ra_ptr++]) == NULL) 1435 goto next; 1436 wait_on_buffer(bh); 1437 if (!buffer_uptodate(bh)) { 1438 EXT4_ERROR_INODE(dir, "reading directory lblock %lu", 1439 (unsigned long) block); 1440 brelse(bh); 1441 ret = ERR_PTR(-EIO); 1442 goto cleanup_and_exit; 1443 } 1444 if (!buffer_verified(bh) && 1445 !is_dx_internal_node(dir, block, 1446 (struct ext4_dir_entry *)bh->b_data) && 1447 !ext4_dirent_csum_verify(dir, 1448 (struct ext4_dir_entry *)bh->b_data)) { 1449 EXT4_ERROR_INODE(dir, "checksumming directory " 1450 "block %lu", (unsigned long)block); 1451 brelse(bh); 1452 ret = ERR_PTR(-EFSBADCRC); 1453 goto cleanup_and_exit; 1454 } 1455 set_buffer_verified(bh); 1456 i = search_dirblock(bh, dir, &fname, 1457 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir); 1458 if (i == 1) { 1459 EXT4_I(dir)->i_dir_start_lookup = block; 1460 ret = bh; 1461 goto cleanup_and_exit; 1462 } else { 1463 brelse(bh); 1464 if (i < 0) 1465 goto cleanup_and_exit; 1466 } 1467 next: 1468 if (++block >= nblocks) 1469 block = 0; 1470 } while (block != start); 1471 1472 /* 1473 * If the directory has grown while we were searching, then 1474 * search the last part of the directory before giving up. 1475 */ 1476 block = nblocks; 1477 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); 1478 if (block < nblocks) { 1479 start = 0; 1480 goto restart; 1481 } 1482 1483 cleanup_and_exit: 1484 /* Clean up the read-ahead blocks */ 1485 for (; ra_ptr < ra_max; ra_ptr++) 1486 brelse(bh_use[ra_ptr]); 1487 ext4_fname_free_filename(&fname); 1488 return ret; 1489 } 1490 1491 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, 1492 struct ext4_filename *fname, 1493 struct ext4_dir_entry_2 **res_dir) 1494 { 1495 struct super_block * sb = dir->i_sb; 1496 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; 1497 struct buffer_head *bh; 1498 ext4_lblk_t block; 1499 int retval; 1500 1501 #ifdef CONFIG_EXT4_FS_ENCRYPTION 1502 *res_dir = NULL; 1503 #endif 1504 frame = dx_probe(fname, dir, NULL, frames); 1505 if (IS_ERR(frame)) 1506 return (struct buffer_head *) frame; 1507 do { 1508 block = dx_get_block(frame->at); 1509 bh = ext4_read_dirblock(dir, block, DIRENT); 1510 if (IS_ERR(bh)) 1511 goto errout; 1512 1513 retval = search_dirblock(bh, dir, fname, 1514 block << EXT4_BLOCK_SIZE_BITS(sb), 1515 res_dir); 1516 if (retval == 1) 1517 goto success; 1518 brelse(bh); 1519 if (retval == -1) { 1520 bh = ERR_PTR(ERR_BAD_DX_DIR); 1521 goto errout; 1522 } 1523 1524 /* Check to see if we should continue to search */ 1525 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame, 1526 frames, NULL); 1527 if (retval < 0) { 1528 ext4_warning_inode(dir, 1529 "error %d reading directory index block", 1530 retval); 1531 bh = ERR_PTR(retval); 1532 goto errout; 1533 } 1534 } while (retval == 1); 1535 1536 bh = NULL; 1537 errout: 1538 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name)); 1539 success: 1540 dx_release(frames); 1541 return bh; 1542 } 1543 1544 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) 1545 { 1546 struct inode *inode; 1547 struct ext4_dir_entry_2 *de; 1548 struct buffer_head *bh; 1549 int err; 1550 1551 err = fscrypt_prepare_lookup(dir, dentry, flags); 1552 if (err) 1553 return ERR_PTR(err); 1554 1555 if (dentry->d_name.len > EXT4_NAME_LEN) 1556 return ERR_PTR(-ENAMETOOLONG); 1557 1558 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); 1559 if (IS_ERR(bh)) 1560 return ERR_CAST(bh); 1561 inode = NULL; 1562 if (bh) { 1563 __u32 ino = le32_to_cpu(de->inode); 1564 brelse(bh); 1565 if (!ext4_valid_inum(dir->i_sb, ino)) { 1566 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino); 1567 return ERR_PTR(-EFSCORRUPTED); 1568 } 1569 if (unlikely(ino == dir->i_ino)) { 1570 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir", 1571 dentry); 1572 return ERR_PTR(-EFSCORRUPTED); 1573 } 1574 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL); 1575 if (inode == ERR_PTR(-ESTALE)) { 1576 EXT4_ERROR_INODE(dir, 1577 "deleted inode referenced: %u", 1578 ino); 1579 return ERR_PTR(-EFSCORRUPTED); 1580 } 1581 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) && 1582 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) && 1583 !fscrypt_has_permitted_context(dir, inode)) { 1584 ext4_warning(inode->i_sb, 1585 "Inconsistent encryption contexts: %lu/%lu", 1586 dir->i_ino, inode->i_ino); 1587 iput(inode); 1588 return ERR_PTR(-EPERM); 1589 } 1590 } 1591 return d_splice_alias(inode, dentry); 1592 } 1593 1594 1595 struct dentry *ext4_get_parent(struct dentry *child) 1596 { 1597 __u32 ino; 1598 static const struct qstr dotdot = QSTR_INIT("..", 2); 1599 struct ext4_dir_entry_2 * de; 1600 struct buffer_head *bh; 1601 1602 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL); 1603 if (IS_ERR(bh)) 1604 return ERR_CAST(bh); 1605 if (!bh) 1606 return ERR_PTR(-ENOENT); 1607 ino = le32_to_cpu(de->inode); 1608 brelse(bh); 1609 1610 if (!ext4_valid_inum(child->d_sb, ino)) { 1611 EXT4_ERROR_INODE(d_inode(child), 1612 "bad parent inode number: %u", ino); 1613 return ERR_PTR(-EFSCORRUPTED); 1614 } 1615 1616 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL)); 1617 } 1618 1619 /* 1620 * Move count entries from end of map between two memory locations. 1621 * Returns pointer to last entry moved. 1622 */ 1623 static struct ext4_dir_entry_2 * 1624 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count, 1625 unsigned blocksize) 1626 { 1627 unsigned rec_len = 0; 1628 1629 while (count--) { 1630 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) 1631 (from + (map->offs<<2)); 1632 rec_len = EXT4_DIR_REC_LEN(de->name_len); 1633 memcpy (to, de, rec_len); 1634 ((struct ext4_dir_entry_2 *) to)->rec_len = 1635 ext4_rec_len_to_disk(rec_len, blocksize); 1636 de->inode = 0; 1637 map++; 1638 to += rec_len; 1639 } 1640 return (struct ext4_dir_entry_2 *) (to - rec_len); 1641 } 1642 1643 /* 1644 * Compact each dir entry in the range to the minimal rec_len. 1645 * Returns pointer to last entry in range. 1646 */ 1647 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize) 1648 { 1649 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base; 1650 unsigned rec_len = 0; 1651 1652 prev = to = de; 1653 while ((char*)de < base + blocksize) { 1654 next = ext4_next_entry(de, blocksize); 1655 if (de->inode && de->name_len) { 1656 rec_len = EXT4_DIR_REC_LEN(de->name_len); 1657 if (de > to) 1658 memmove(to, de, rec_len); 1659 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize); 1660 prev = to; 1661 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len); 1662 } 1663 de = next; 1664 } 1665 return prev; 1666 } 1667 1668 /* 1669 * Split a full leaf block to make room for a new dir entry. 1670 * Allocate a new block, and move entries so that they are approx. equally full. 1671 * Returns pointer to de in block into which the new entry will be inserted. 1672 */ 1673 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, 1674 struct buffer_head **bh,struct dx_frame *frame, 1675 struct dx_hash_info *hinfo) 1676 { 1677 unsigned blocksize = dir->i_sb->s_blocksize; 1678 unsigned count, continued; 1679 struct buffer_head *bh2; 1680 ext4_lblk_t newblock; 1681 u32 hash2; 1682 struct dx_map_entry *map; 1683 char *data1 = (*bh)->b_data, *data2; 1684 unsigned split, move, size; 1685 struct ext4_dir_entry_2 *de = NULL, *de2; 1686 struct ext4_dir_entry_tail *t; 1687 int csum_size = 0; 1688 int err = 0, i; 1689 1690 if (ext4_has_metadata_csum(dir->i_sb)) 1691 csum_size = sizeof(struct ext4_dir_entry_tail); 1692 1693 bh2 = ext4_append(handle, dir, &newblock); 1694 if (IS_ERR(bh2)) { 1695 brelse(*bh); 1696 *bh = NULL; 1697 return (struct ext4_dir_entry_2 *) bh2; 1698 } 1699 1700 BUFFER_TRACE(*bh, "get_write_access"); 1701 err = ext4_journal_get_write_access(handle, *bh); 1702 if (err) 1703 goto journal_error; 1704 1705 BUFFER_TRACE(frame->bh, "get_write_access"); 1706 err = ext4_journal_get_write_access(handle, frame->bh); 1707 if (err) 1708 goto journal_error; 1709 1710 data2 = bh2->b_data; 1711 1712 /* create map in the end of data2 block */ 1713 map = (struct dx_map_entry *) (data2 + blocksize); 1714 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1, 1715 blocksize, hinfo, map); 1716 map -= count; 1717 dx_sort_map(map, count); 1718 /* Split the existing block in the middle, size-wise */ 1719 size = 0; 1720 move = 0; 1721 for (i = count-1; i >= 0; i--) { 1722 /* is more than half of this entry in 2nd half of the block? */ 1723 if (size + map[i].size/2 > blocksize/2) 1724 break; 1725 size += map[i].size; 1726 move++; 1727 } 1728 /* map index at which we will split */ 1729 split = count - move; 1730 hash2 = map[split].hash; 1731 continued = hash2 == map[split - 1].hash; 1732 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n", 1733 (unsigned long)dx_get_block(frame->at), 1734 hash2, split, count-split)); 1735 1736 /* Fancy dance to stay within two buffers */ 1737 de2 = dx_move_dirents(data1, data2, map + split, count - split, 1738 blocksize); 1739 de = dx_pack_dirents(data1, blocksize); 1740 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) - 1741 (char *) de, 1742 blocksize); 1743 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) - 1744 (char *) de2, 1745 blocksize); 1746 if (csum_size) { 1747 t = EXT4_DIRENT_TAIL(data2, blocksize); 1748 initialize_dirent_tail(t, blocksize); 1749 1750 t = EXT4_DIRENT_TAIL(data1, blocksize); 1751 initialize_dirent_tail(t, blocksize); 1752 } 1753 1754 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1, 1755 blocksize, 1)); 1756 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2, 1757 blocksize, 1)); 1758 1759 /* Which block gets the new entry? */ 1760 if (hinfo->hash >= hash2) { 1761 swap(*bh, bh2); 1762 de = de2; 1763 } 1764 dx_insert_block(frame, hash2 + continued, newblock); 1765 err = ext4_handle_dirty_dirent_node(handle, dir, bh2); 1766 if (err) 1767 goto journal_error; 1768 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh); 1769 if (err) 1770 goto journal_error; 1771 brelse(bh2); 1772 dxtrace(dx_show_index("frame", frame->entries)); 1773 return de; 1774 1775 journal_error: 1776 brelse(*bh); 1777 brelse(bh2); 1778 *bh = NULL; 1779 ext4_std_error(dir->i_sb, err); 1780 return ERR_PTR(err); 1781 } 1782 1783 int ext4_find_dest_de(struct inode *dir, struct inode *inode, 1784 struct buffer_head *bh, 1785 void *buf, int buf_size, 1786 struct ext4_filename *fname, 1787 struct ext4_dir_entry_2 **dest_de) 1788 { 1789 struct ext4_dir_entry_2 *de; 1790 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname)); 1791 int nlen, rlen; 1792 unsigned int offset = 0; 1793 char *top; 1794 1795 de = (struct ext4_dir_entry_2 *)buf; 1796 top = buf + buf_size - reclen; 1797 while ((char *) de <= top) { 1798 if (ext4_check_dir_entry(dir, NULL, de, bh, 1799 buf, buf_size, offset)) 1800 return -EFSCORRUPTED; 1801 if (ext4_match(fname, de)) 1802 return -EEXIST; 1803 nlen = EXT4_DIR_REC_LEN(de->name_len); 1804 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 1805 if ((de->inode ? rlen - nlen : rlen) >= reclen) 1806 break; 1807 de = (struct ext4_dir_entry_2 *)((char *)de + rlen); 1808 offset += rlen; 1809 } 1810 if ((char *) de > top) 1811 return -ENOSPC; 1812 1813 *dest_de = de; 1814 return 0; 1815 } 1816 1817 void ext4_insert_dentry(struct inode *inode, 1818 struct ext4_dir_entry_2 *de, 1819 int buf_size, 1820 struct ext4_filename *fname) 1821 { 1822 1823 int nlen, rlen; 1824 1825 nlen = EXT4_DIR_REC_LEN(de->name_len); 1826 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 1827 if (de->inode) { 1828 struct ext4_dir_entry_2 *de1 = 1829 (struct ext4_dir_entry_2 *)((char *)de + nlen); 1830 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size); 1831 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size); 1832 de = de1; 1833 } 1834 de->file_type = EXT4_FT_UNKNOWN; 1835 de->inode = cpu_to_le32(inode->i_ino); 1836 ext4_set_de_type(inode->i_sb, de, inode->i_mode); 1837 de->name_len = fname_len(fname); 1838 memcpy(de->name, fname_name(fname), fname_len(fname)); 1839 } 1840 1841 /* 1842 * Add a new entry into a directory (leaf) block. If de is non-NULL, 1843 * it points to a directory entry which is guaranteed to be large 1844 * enough for new directory entry. If de is NULL, then 1845 * add_dirent_to_buf will attempt search the directory block for 1846 * space. It will return -ENOSPC if no space is available, and -EIO 1847 * and -EEXIST if directory entry already exists. 1848 */ 1849 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname, 1850 struct inode *dir, 1851 struct inode *inode, struct ext4_dir_entry_2 *de, 1852 struct buffer_head *bh) 1853 { 1854 unsigned int blocksize = dir->i_sb->s_blocksize; 1855 int csum_size = 0; 1856 int err; 1857 1858 if (ext4_has_metadata_csum(inode->i_sb)) 1859 csum_size = sizeof(struct ext4_dir_entry_tail); 1860 1861 if (!de) { 1862 err = ext4_find_dest_de(dir, inode, bh, bh->b_data, 1863 blocksize - csum_size, fname, &de); 1864 if (err) 1865 return err; 1866 } 1867 BUFFER_TRACE(bh, "get_write_access"); 1868 err = ext4_journal_get_write_access(handle, bh); 1869 if (err) { 1870 ext4_std_error(dir->i_sb, err); 1871 return err; 1872 } 1873 1874 /* By now the buffer is marked for journaling */ 1875 ext4_insert_dentry(inode, de, blocksize, fname); 1876 1877 /* 1878 * XXX shouldn't update any times until successful 1879 * completion of syscall, but too many callers depend 1880 * on this. 1881 * 1882 * XXX similarly, too many callers depend on 1883 * ext4_new_inode() setting the times, but error 1884 * recovery deletes the inode, so the worst that can 1885 * happen is that the times are slightly out of date 1886 * and/or different from the directory change time. 1887 */ 1888 dir->i_mtime = dir->i_ctime = current_time(dir); 1889 ext4_update_dx_flag(dir); 1890 inode_inc_iversion(dir); 1891 ext4_mark_inode_dirty(handle, dir); 1892 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 1893 err = ext4_handle_dirty_dirent_node(handle, dir, bh); 1894 if (err) 1895 ext4_std_error(dir->i_sb, err); 1896 return 0; 1897 } 1898 1899 /* 1900 * This converts a one block unindexed directory to a 3 block indexed 1901 * directory, and adds the dentry to the indexed directory. 1902 */ 1903 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname, 1904 struct inode *dir, 1905 struct inode *inode, struct buffer_head *bh) 1906 { 1907 struct buffer_head *bh2; 1908 struct dx_root *root; 1909 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; 1910 struct dx_entry *entries; 1911 struct ext4_dir_entry_2 *de, *de2; 1912 struct ext4_dir_entry_tail *t; 1913 char *data1, *top; 1914 unsigned len; 1915 int retval; 1916 unsigned blocksize; 1917 ext4_lblk_t block; 1918 struct fake_dirent *fde; 1919 int csum_size = 0; 1920 1921 if (ext4_has_metadata_csum(inode->i_sb)) 1922 csum_size = sizeof(struct ext4_dir_entry_tail); 1923 1924 blocksize = dir->i_sb->s_blocksize; 1925 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino)); 1926 BUFFER_TRACE(bh, "get_write_access"); 1927 retval = ext4_journal_get_write_access(handle, bh); 1928 if (retval) { 1929 ext4_std_error(dir->i_sb, retval); 1930 brelse(bh); 1931 return retval; 1932 } 1933 root = (struct dx_root *) bh->b_data; 1934 1935 /* The 0th block becomes the root, move the dirents out */ 1936 fde = &root->dotdot; 1937 de = (struct ext4_dir_entry_2 *)((char *)fde + 1938 ext4_rec_len_from_disk(fde->rec_len, blocksize)); 1939 if ((char *) de >= (((char *) root) + blocksize)) { 1940 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'"); 1941 brelse(bh); 1942 return -EFSCORRUPTED; 1943 } 1944 len = ((char *) root) + (blocksize - csum_size) - (char *) de; 1945 1946 /* Allocate new block for the 0th block's dirents */ 1947 bh2 = ext4_append(handle, dir, &block); 1948 if (IS_ERR(bh2)) { 1949 brelse(bh); 1950 return PTR_ERR(bh2); 1951 } 1952 ext4_set_inode_flag(dir, EXT4_INODE_INDEX); 1953 data1 = bh2->b_data; 1954 1955 memcpy (data1, de, len); 1956 de = (struct ext4_dir_entry_2 *) data1; 1957 top = data1 + len; 1958 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) 1959 de = de2; 1960 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) - 1961 (char *) de, 1962 blocksize); 1963 1964 if (csum_size) { 1965 t = EXT4_DIRENT_TAIL(data1, blocksize); 1966 initialize_dirent_tail(t, blocksize); 1967 } 1968 1969 /* Initialize the root; the dot dirents already exist */ 1970 de = (struct ext4_dir_entry_2 *) (&root->dotdot); 1971 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2), 1972 blocksize); 1973 memset (&root->info, 0, sizeof(root->info)); 1974 root->info.info_length = sizeof(root->info); 1975 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; 1976 entries = root->entries; 1977 dx_set_block(entries, 1); 1978 dx_set_count(entries, 1); 1979 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info))); 1980 1981 /* Initialize as for dx_probe */ 1982 fname->hinfo.hash_version = root->info.hash_version; 1983 if (fname->hinfo.hash_version <= DX_HASH_TEA) 1984 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; 1985 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; 1986 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo); 1987 1988 memset(frames, 0, sizeof(frames)); 1989 frame = frames; 1990 frame->entries = entries; 1991 frame->at = entries; 1992 frame->bh = bh; 1993 1994 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh); 1995 if (retval) 1996 goto out_frames; 1997 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2); 1998 if (retval) 1999 goto out_frames; 2000 2001 de = do_split(handle,dir, &bh2, frame, &fname->hinfo); 2002 if (IS_ERR(de)) { 2003 retval = PTR_ERR(de); 2004 goto out_frames; 2005 } 2006 2007 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2); 2008 out_frames: 2009 /* 2010 * Even if the block split failed, we have to properly write 2011 * out all the changes we did so far. Otherwise we can end up 2012 * with corrupted filesystem. 2013 */ 2014 if (retval) 2015 ext4_mark_inode_dirty(handle, dir); 2016 dx_release(frames); 2017 brelse(bh2); 2018 return retval; 2019 } 2020 2021 /* 2022 * ext4_add_entry() 2023 * 2024 * adds a file entry to the specified directory, using the same 2025 * semantics as ext4_find_entry(). It returns NULL if it failed. 2026 * 2027 * NOTE!! The inode part of 'de' is left at 0 - which means you 2028 * may not sleep between calling this and putting something into 2029 * the entry, as someone else might have used it while you slept. 2030 */ 2031 static int ext4_add_entry(handle_t *handle, struct dentry *dentry, 2032 struct inode *inode) 2033 { 2034 struct inode *dir = d_inode(dentry->d_parent); 2035 struct buffer_head *bh = NULL; 2036 struct ext4_dir_entry_2 *de; 2037 struct ext4_dir_entry_tail *t; 2038 struct super_block *sb; 2039 struct ext4_filename fname; 2040 int retval; 2041 int dx_fallback=0; 2042 unsigned blocksize; 2043 ext4_lblk_t block, blocks; 2044 int csum_size = 0; 2045 2046 if (ext4_has_metadata_csum(inode->i_sb)) 2047 csum_size = sizeof(struct ext4_dir_entry_tail); 2048 2049 sb = dir->i_sb; 2050 blocksize = sb->s_blocksize; 2051 if (!dentry->d_name.len) 2052 return -EINVAL; 2053 2054 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname); 2055 if (retval) 2056 return retval; 2057 2058 if (ext4_has_inline_data(dir)) { 2059 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode); 2060 if (retval < 0) 2061 goto out; 2062 if (retval == 1) { 2063 retval = 0; 2064 goto out; 2065 } 2066 } 2067 2068 if (is_dx(dir)) { 2069 retval = ext4_dx_add_entry(handle, &fname, dir, inode); 2070 if (!retval || (retval != ERR_BAD_DX_DIR)) 2071 goto out; 2072 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX); 2073 dx_fallback++; 2074 ext4_mark_inode_dirty(handle, dir); 2075 } 2076 blocks = dir->i_size >> sb->s_blocksize_bits; 2077 for (block = 0; block < blocks; block++) { 2078 bh = ext4_read_dirblock(dir, block, DIRENT); 2079 if (IS_ERR(bh)) { 2080 retval = PTR_ERR(bh); 2081 bh = NULL; 2082 goto out; 2083 } 2084 retval = add_dirent_to_buf(handle, &fname, dir, inode, 2085 NULL, bh); 2086 if (retval != -ENOSPC) 2087 goto out; 2088 2089 if (blocks == 1 && !dx_fallback && 2090 ext4_has_feature_dir_index(sb)) { 2091 retval = make_indexed_dir(handle, &fname, dir, 2092 inode, bh); 2093 bh = NULL; /* make_indexed_dir releases bh */ 2094 goto out; 2095 } 2096 brelse(bh); 2097 } 2098 bh = ext4_append(handle, dir, &block); 2099 if (IS_ERR(bh)) { 2100 retval = PTR_ERR(bh); 2101 bh = NULL; 2102 goto out; 2103 } 2104 de = (struct ext4_dir_entry_2 *) bh->b_data; 2105 de->inode = 0; 2106 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize); 2107 2108 if (csum_size) { 2109 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize); 2110 initialize_dirent_tail(t, blocksize); 2111 } 2112 2113 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh); 2114 out: 2115 ext4_fname_free_filename(&fname); 2116 brelse(bh); 2117 if (retval == 0) 2118 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY); 2119 return retval; 2120 } 2121 2122 /* 2123 * Returns 0 for success, or a negative error value 2124 */ 2125 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname, 2126 struct inode *dir, struct inode *inode) 2127 { 2128 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; 2129 struct dx_entry *entries, *at; 2130 struct buffer_head *bh; 2131 struct super_block *sb = dir->i_sb; 2132 struct ext4_dir_entry_2 *de; 2133 int restart; 2134 int err; 2135 2136 again: 2137 restart = 0; 2138 frame = dx_probe(fname, dir, NULL, frames); 2139 if (IS_ERR(frame)) 2140 return PTR_ERR(frame); 2141 entries = frame->entries; 2142 at = frame->at; 2143 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT); 2144 if (IS_ERR(bh)) { 2145 err = PTR_ERR(bh); 2146 bh = NULL; 2147 goto cleanup; 2148 } 2149 2150 BUFFER_TRACE(bh, "get_write_access"); 2151 err = ext4_journal_get_write_access(handle, bh); 2152 if (err) 2153 goto journal_error; 2154 2155 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh); 2156 if (err != -ENOSPC) 2157 goto cleanup; 2158 2159 err = 0; 2160 /* Block full, should compress but for now just split */ 2161 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n", 2162 dx_get_count(entries), dx_get_limit(entries))); 2163 /* Need to split index? */ 2164 if (dx_get_count(entries) == dx_get_limit(entries)) { 2165 ext4_lblk_t newblock; 2166 int levels = frame - frames + 1; 2167 unsigned int icount; 2168 int add_level = 1; 2169 struct dx_entry *entries2; 2170 struct dx_node *node2; 2171 struct buffer_head *bh2; 2172 2173 while (frame > frames) { 2174 if (dx_get_count((frame - 1)->entries) < 2175 dx_get_limit((frame - 1)->entries)) { 2176 add_level = 0; 2177 break; 2178 } 2179 frame--; /* split higher index block */ 2180 at = frame->at; 2181 entries = frame->entries; 2182 restart = 1; 2183 } 2184 if (add_level && levels == ext4_dir_htree_level(sb)) { 2185 ext4_warning(sb, "Directory (ino: %lu) index full, " 2186 "reach max htree level :%d", 2187 dir->i_ino, levels); 2188 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) { 2189 ext4_warning(sb, "Large directory feature is " 2190 "not enabled on this " 2191 "filesystem"); 2192 } 2193 err = -ENOSPC; 2194 goto cleanup; 2195 } 2196 icount = dx_get_count(entries); 2197 bh2 = ext4_append(handle, dir, &newblock); 2198 if (IS_ERR(bh2)) { 2199 err = PTR_ERR(bh2); 2200 goto cleanup; 2201 } 2202 node2 = (struct dx_node *)(bh2->b_data); 2203 entries2 = node2->entries; 2204 memset(&node2->fake, 0, sizeof(struct fake_dirent)); 2205 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize, 2206 sb->s_blocksize); 2207 BUFFER_TRACE(frame->bh, "get_write_access"); 2208 err = ext4_journal_get_write_access(handle, frame->bh); 2209 if (err) 2210 goto journal_error; 2211 if (!add_level) { 2212 unsigned icount1 = icount/2, icount2 = icount - icount1; 2213 unsigned hash2 = dx_get_hash(entries + icount1); 2214 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n", 2215 icount1, icount2)); 2216 2217 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */ 2218 err = ext4_journal_get_write_access(handle, 2219 (frame - 1)->bh); 2220 if (err) 2221 goto journal_error; 2222 2223 memcpy((char *) entries2, (char *) (entries + icount1), 2224 icount2 * sizeof(struct dx_entry)); 2225 dx_set_count(entries, icount1); 2226 dx_set_count(entries2, icount2); 2227 dx_set_limit(entries2, dx_node_limit(dir)); 2228 2229 /* Which index block gets the new entry? */ 2230 if (at - entries >= icount1) { 2231 frame->at = at = at - entries - icount1 + entries2; 2232 frame->entries = entries = entries2; 2233 swap(frame->bh, bh2); 2234 } 2235 dx_insert_block((frame - 1), hash2, newblock); 2236 dxtrace(dx_show_index("node", frame->entries)); 2237 dxtrace(dx_show_index("node", 2238 ((struct dx_node *) bh2->b_data)->entries)); 2239 err = ext4_handle_dirty_dx_node(handle, dir, bh2); 2240 if (err) 2241 goto journal_error; 2242 brelse (bh2); 2243 err = ext4_handle_dirty_dx_node(handle, dir, 2244 (frame - 1)->bh); 2245 if (err) 2246 goto journal_error; 2247 if (restart) { 2248 err = ext4_handle_dirty_dx_node(handle, dir, 2249 frame->bh); 2250 goto journal_error; 2251 } 2252 } else { 2253 struct dx_root *dxroot; 2254 memcpy((char *) entries2, (char *) entries, 2255 icount * sizeof(struct dx_entry)); 2256 dx_set_limit(entries2, dx_node_limit(dir)); 2257 2258 /* Set up root */ 2259 dx_set_count(entries, 1); 2260 dx_set_block(entries + 0, newblock); 2261 dxroot = (struct dx_root *)frames[0].bh->b_data; 2262 dxroot->info.indirect_levels += 1; 2263 dxtrace(printk(KERN_DEBUG 2264 "Creating %d level index...\n", 2265 dxroot->info.indirect_levels)); 2266 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh); 2267 if (err) 2268 goto journal_error; 2269 err = ext4_handle_dirty_dx_node(handle, dir, bh2); 2270 brelse(bh2); 2271 restart = 1; 2272 goto journal_error; 2273 } 2274 } 2275 de = do_split(handle, dir, &bh, frame, &fname->hinfo); 2276 if (IS_ERR(de)) { 2277 err = PTR_ERR(de); 2278 goto cleanup; 2279 } 2280 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh); 2281 goto cleanup; 2282 2283 journal_error: 2284 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */ 2285 cleanup: 2286 brelse(bh); 2287 dx_release(frames); 2288 /* @restart is true means htree-path has been changed, we need to 2289 * repeat dx_probe() to find out valid htree-path 2290 */ 2291 if (restart && err == 0) 2292 goto again; 2293 return err; 2294 } 2295 2296 /* 2297 * ext4_generic_delete_entry deletes a directory entry by merging it 2298 * with the previous entry 2299 */ 2300 int ext4_generic_delete_entry(handle_t *handle, 2301 struct inode *dir, 2302 struct ext4_dir_entry_2 *de_del, 2303 struct buffer_head *bh, 2304 void *entry_buf, 2305 int buf_size, 2306 int csum_size) 2307 { 2308 struct ext4_dir_entry_2 *de, *pde; 2309 unsigned int blocksize = dir->i_sb->s_blocksize; 2310 int i; 2311 2312 i = 0; 2313 pde = NULL; 2314 de = (struct ext4_dir_entry_2 *)entry_buf; 2315 while (i < buf_size - csum_size) { 2316 if (ext4_check_dir_entry(dir, NULL, de, bh, 2317 bh->b_data, bh->b_size, i)) 2318 return -EFSCORRUPTED; 2319 if (de == de_del) { 2320 if (pde) 2321 pde->rec_len = ext4_rec_len_to_disk( 2322 ext4_rec_len_from_disk(pde->rec_len, 2323 blocksize) + 2324 ext4_rec_len_from_disk(de->rec_len, 2325 blocksize), 2326 blocksize); 2327 else 2328 de->inode = 0; 2329 inode_inc_iversion(dir); 2330 return 0; 2331 } 2332 i += ext4_rec_len_from_disk(de->rec_len, blocksize); 2333 pde = de; 2334 de = ext4_next_entry(de, blocksize); 2335 } 2336 return -ENOENT; 2337 } 2338 2339 static int ext4_delete_entry(handle_t *handle, 2340 struct inode *dir, 2341 struct ext4_dir_entry_2 *de_del, 2342 struct buffer_head *bh) 2343 { 2344 int err, csum_size = 0; 2345 2346 if (ext4_has_inline_data(dir)) { 2347 int has_inline_data = 1; 2348 err = ext4_delete_inline_entry(handle, dir, de_del, bh, 2349 &has_inline_data); 2350 if (has_inline_data) 2351 return err; 2352 } 2353 2354 if (ext4_has_metadata_csum(dir->i_sb)) 2355 csum_size = sizeof(struct ext4_dir_entry_tail); 2356 2357 BUFFER_TRACE(bh, "get_write_access"); 2358 err = ext4_journal_get_write_access(handle, bh); 2359 if (unlikely(err)) 2360 goto out; 2361 2362 err = ext4_generic_delete_entry(handle, dir, de_del, 2363 bh, bh->b_data, 2364 dir->i_sb->s_blocksize, csum_size); 2365 if (err) 2366 goto out; 2367 2368 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 2369 err = ext4_handle_dirty_dirent_node(handle, dir, bh); 2370 if (unlikely(err)) 2371 goto out; 2372 2373 return 0; 2374 out: 2375 if (err != -ENOENT) 2376 ext4_std_error(dir->i_sb, err); 2377 return err; 2378 } 2379 2380 /* 2381 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2 2382 * since this indicates that nlinks count was previously 1 to avoid overflowing 2383 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean 2384 * that subdirectory link counts are not being maintained accurately. 2385 * 2386 * The caller has already checked for i_nlink overflow in case the DIR_LINK 2387 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy 2388 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set 2389 * on regular files) and to avoid creating huge/slow non-HTREE directories. 2390 */ 2391 static void ext4_inc_count(handle_t *handle, struct inode *inode) 2392 { 2393 inc_nlink(inode); 2394 if (is_dx(inode) && 2395 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2)) 2396 set_nlink(inode, 1); 2397 } 2398 2399 /* 2400 * If a directory had nlink == 1, then we should let it be 1. This indicates 2401 * directory has >EXT4_LINK_MAX subdirs. 2402 */ 2403 static void ext4_dec_count(handle_t *handle, struct inode *inode) 2404 { 2405 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2) 2406 drop_nlink(inode); 2407 } 2408 2409 2410 static int ext4_add_nondir(handle_t *handle, 2411 struct dentry *dentry, struct inode *inode) 2412 { 2413 int err = ext4_add_entry(handle, dentry, inode); 2414 if (!err) { 2415 ext4_mark_inode_dirty(handle, inode); 2416 d_instantiate_new(dentry, inode); 2417 return 0; 2418 } 2419 drop_nlink(inode); 2420 unlock_new_inode(inode); 2421 iput(inode); 2422 return err; 2423 } 2424 2425 /* 2426 * By the time this is called, we already have created 2427 * the directory cache entry for the new file, but it 2428 * is so far negative - it has no inode. 2429 * 2430 * If the create succeeds, we fill in the inode information 2431 * with d_instantiate(). 2432 */ 2433 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2434 bool excl) 2435 { 2436 handle_t *handle; 2437 struct inode *inode; 2438 int err, credits, retries = 0; 2439 2440 err = dquot_initialize(dir); 2441 if (err) 2442 return err; 2443 2444 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 2445 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); 2446 retry: 2447 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0, 2448 NULL, EXT4_HT_DIR, credits); 2449 handle = ext4_journal_current_handle(); 2450 err = PTR_ERR(inode); 2451 if (!IS_ERR(inode)) { 2452 inode->i_op = &ext4_file_inode_operations; 2453 inode->i_fop = &ext4_file_operations; 2454 ext4_set_aops(inode); 2455 err = ext4_add_nondir(handle, dentry, inode); 2456 if (!err && IS_DIRSYNC(dir)) 2457 ext4_handle_sync(handle); 2458 } 2459 if (handle) 2460 ext4_journal_stop(handle); 2461 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2462 goto retry; 2463 return err; 2464 } 2465 2466 static int ext4_mknod(struct inode *dir, struct dentry *dentry, 2467 umode_t mode, dev_t rdev) 2468 { 2469 handle_t *handle; 2470 struct inode *inode; 2471 int err, credits, retries = 0; 2472 2473 err = dquot_initialize(dir); 2474 if (err) 2475 return err; 2476 2477 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 2478 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); 2479 retry: 2480 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0, 2481 NULL, EXT4_HT_DIR, credits); 2482 handle = ext4_journal_current_handle(); 2483 err = PTR_ERR(inode); 2484 if (!IS_ERR(inode)) { 2485 init_special_inode(inode, inode->i_mode, rdev); 2486 inode->i_op = &ext4_special_inode_operations; 2487 err = ext4_add_nondir(handle, dentry, inode); 2488 if (!err && IS_DIRSYNC(dir)) 2489 ext4_handle_sync(handle); 2490 } 2491 if (handle) 2492 ext4_journal_stop(handle); 2493 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2494 goto retry; 2495 return err; 2496 } 2497 2498 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) 2499 { 2500 handle_t *handle; 2501 struct inode *inode; 2502 int err, retries = 0; 2503 2504 err = dquot_initialize(dir); 2505 if (err) 2506 return err; 2507 2508 retry: 2509 inode = ext4_new_inode_start_handle(dir, mode, 2510 NULL, 0, NULL, 2511 EXT4_HT_DIR, 2512 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) + 2513 4 + EXT4_XATTR_TRANS_BLOCKS); 2514 handle = ext4_journal_current_handle(); 2515 err = PTR_ERR(inode); 2516 if (!IS_ERR(inode)) { 2517 inode->i_op = &ext4_file_inode_operations; 2518 inode->i_fop = &ext4_file_operations; 2519 ext4_set_aops(inode); 2520 d_tmpfile(dentry, inode); 2521 err = ext4_orphan_add(handle, inode); 2522 if (err) 2523 goto err_unlock_inode; 2524 mark_inode_dirty(inode); 2525 unlock_new_inode(inode); 2526 } 2527 if (handle) 2528 ext4_journal_stop(handle); 2529 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2530 goto retry; 2531 return err; 2532 err_unlock_inode: 2533 ext4_journal_stop(handle); 2534 unlock_new_inode(inode); 2535 return err; 2536 } 2537 2538 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode, 2539 struct ext4_dir_entry_2 *de, 2540 int blocksize, int csum_size, 2541 unsigned int parent_ino, int dotdot_real_len) 2542 { 2543 de->inode = cpu_to_le32(inode->i_ino); 2544 de->name_len = 1; 2545 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len), 2546 blocksize); 2547 strcpy(de->name, "."); 2548 ext4_set_de_type(inode->i_sb, de, S_IFDIR); 2549 2550 de = ext4_next_entry(de, blocksize); 2551 de->inode = cpu_to_le32(parent_ino); 2552 de->name_len = 2; 2553 if (!dotdot_real_len) 2554 de->rec_len = ext4_rec_len_to_disk(blocksize - 2555 (csum_size + EXT4_DIR_REC_LEN(1)), 2556 blocksize); 2557 else 2558 de->rec_len = ext4_rec_len_to_disk( 2559 EXT4_DIR_REC_LEN(de->name_len), blocksize); 2560 strcpy(de->name, ".."); 2561 ext4_set_de_type(inode->i_sb, de, S_IFDIR); 2562 2563 return ext4_next_entry(de, blocksize); 2564 } 2565 2566 static int ext4_init_new_dir(handle_t *handle, struct inode *dir, 2567 struct inode *inode) 2568 { 2569 struct buffer_head *dir_block = NULL; 2570 struct ext4_dir_entry_2 *de; 2571 struct ext4_dir_entry_tail *t; 2572 ext4_lblk_t block = 0; 2573 unsigned int blocksize = dir->i_sb->s_blocksize; 2574 int csum_size = 0; 2575 int err; 2576 2577 if (ext4_has_metadata_csum(dir->i_sb)) 2578 csum_size = sizeof(struct ext4_dir_entry_tail); 2579 2580 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { 2581 err = ext4_try_create_inline_dir(handle, dir, inode); 2582 if (err < 0 && err != -ENOSPC) 2583 goto out; 2584 if (!err) 2585 goto out; 2586 } 2587 2588 inode->i_size = 0; 2589 dir_block = ext4_append(handle, inode, &block); 2590 if (IS_ERR(dir_block)) 2591 return PTR_ERR(dir_block); 2592 de = (struct ext4_dir_entry_2 *)dir_block->b_data; 2593 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0); 2594 set_nlink(inode, 2); 2595 if (csum_size) { 2596 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize); 2597 initialize_dirent_tail(t, blocksize); 2598 } 2599 2600 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata"); 2601 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block); 2602 if (err) 2603 goto out; 2604 set_buffer_verified(dir_block); 2605 out: 2606 brelse(dir_block); 2607 return err; 2608 } 2609 2610 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 2611 { 2612 handle_t *handle; 2613 struct inode *inode; 2614 int err, credits, retries = 0; 2615 2616 if (EXT4_DIR_LINK_MAX(dir)) 2617 return -EMLINK; 2618 2619 err = dquot_initialize(dir); 2620 if (err) 2621 return err; 2622 2623 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 2624 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); 2625 retry: 2626 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode, 2627 &dentry->d_name, 2628 0, NULL, EXT4_HT_DIR, credits); 2629 handle = ext4_journal_current_handle(); 2630 err = PTR_ERR(inode); 2631 if (IS_ERR(inode)) 2632 goto out_stop; 2633 2634 inode->i_op = &ext4_dir_inode_operations; 2635 inode->i_fop = &ext4_dir_operations; 2636 err = ext4_init_new_dir(handle, dir, inode); 2637 if (err) 2638 goto out_clear_inode; 2639 err = ext4_mark_inode_dirty(handle, inode); 2640 if (!err) 2641 err = ext4_add_entry(handle, dentry, inode); 2642 if (err) { 2643 out_clear_inode: 2644 clear_nlink(inode); 2645 unlock_new_inode(inode); 2646 ext4_mark_inode_dirty(handle, inode); 2647 iput(inode); 2648 goto out_stop; 2649 } 2650 ext4_inc_count(handle, dir); 2651 ext4_update_dx_flag(dir); 2652 err = ext4_mark_inode_dirty(handle, dir); 2653 if (err) 2654 goto out_clear_inode; 2655 d_instantiate_new(dentry, inode); 2656 if (IS_DIRSYNC(dir)) 2657 ext4_handle_sync(handle); 2658 2659 out_stop: 2660 if (handle) 2661 ext4_journal_stop(handle); 2662 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2663 goto retry; 2664 return err; 2665 } 2666 2667 /* 2668 * routine to check that the specified directory is empty (for rmdir) 2669 */ 2670 bool ext4_empty_dir(struct inode *inode) 2671 { 2672 unsigned int offset; 2673 struct buffer_head *bh; 2674 struct ext4_dir_entry_2 *de, *de1; 2675 struct super_block *sb; 2676 2677 if (ext4_has_inline_data(inode)) { 2678 int has_inline_data = 1; 2679 int ret; 2680 2681 ret = empty_inline_dir(inode, &has_inline_data); 2682 if (has_inline_data) 2683 return ret; 2684 } 2685 2686 sb = inode->i_sb; 2687 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) { 2688 EXT4_ERROR_INODE(inode, "invalid size"); 2689 return true; 2690 } 2691 bh = ext4_read_dirblock(inode, 0, EITHER); 2692 if (IS_ERR(bh)) 2693 return true; 2694 2695 de = (struct ext4_dir_entry_2 *) bh->b_data; 2696 de1 = ext4_next_entry(de, sb->s_blocksize); 2697 if (le32_to_cpu(de->inode) != inode->i_ino || 2698 le32_to_cpu(de1->inode) == 0 || 2699 strcmp(".", de->name) || strcmp("..", de1->name)) { 2700 ext4_warning_inode(inode, "directory missing '.' and/or '..'"); 2701 brelse(bh); 2702 return true; 2703 } 2704 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) + 2705 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize); 2706 de = ext4_next_entry(de1, sb->s_blocksize); 2707 while (offset < inode->i_size) { 2708 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) { 2709 unsigned int lblock; 2710 brelse(bh); 2711 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb); 2712 bh = ext4_read_dirblock(inode, lblock, EITHER); 2713 if (IS_ERR(bh)) 2714 return true; 2715 de = (struct ext4_dir_entry_2 *) bh->b_data; 2716 } 2717 if (ext4_check_dir_entry(inode, NULL, de, bh, 2718 bh->b_data, bh->b_size, offset)) { 2719 de = (struct ext4_dir_entry_2 *)(bh->b_data + 2720 sb->s_blocksize); 2721 offset = (offset | (sb->s_blocksize - 1)) + 1; 2722 continue; 2723 } 2724 if (le32_to_cpu(de->inode)) { 2725 brelse(bh); 2726 return false; 2727 } 2728 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize); 2729 de = ext4_next_entry(de, sb->s_blocksize); 2730 } 2731 brelse(bh); 2732 return true; 2733 } 2734 2735 /* 2736 * ext4_orphan_add() links an unlinked or truncated inode into a list of 2737 * such inodes, starting at the superblock, in case we crash before the 2738 * file is closed/deleted, or in case the inode truncate spans multiple 2739 * transactions and the last transaction is not recovered after a crash. 2740 * 2741 * At filesystem recovery time, we walk this list deleting unlinked 2742 * inodes and truncating linked inodes in ext4_orphan_cleanup(). 2743 * 2744 * Orphan list manipulation functions must be called under i_mutex unless 2745 * we are just creating the inode or deleting it. 2746 */ 2747 int ext4_orphan_add(handle_t *handle, struct inode *inode) 2748 { 2749 struct super_block *sb = inode->i_sb; 2750 struct ext4_sb_info *sbi = EXT4_SB(sb); 2751 struct ext4_iloc iloc; 2752 int err = 0, rc; 2753 bool dirty = false; 2754 2755 if (!sbi->s_journal || is_bad_inode(inode)) 2756 return 0; 2757 2758 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) && 2759 !inode_is_locked(inode)); 2760 /* 2761 * Exit early if inode already is on orphan list. This is a big speedup 2762 * since we don't have to contend on the global s_orphan_lock. 2763 */ 2764 if (!list_empty(&EXT4_I(inode)->i_orphan)) 2765 return 0; 2766 2767 /* 2768 * Orphan handling is only valid for files with data blocks 2769 * being truncated, or files being unlinked. Note that we either 2770 * hold i_mutex, or the inode can not be referenced from outside, 2771 * so i_nlink should not be bumped due to race 2772 */ 2773 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 2774 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0); 2775 2776 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 2777 err = ext4_journal_get_write_access(handle, sbi->s_sbh); 2778 if (err) 2779 goto out; 2780 2781 err = ext4_reserve_inode_write(handle, inode, &iloc); 2782 if (err) 2783 goto out; 2784 2785 mutex_lock(&sbi->s_orphan_lock); 2786 /* 2787 * Due to previous errors inode may be already a part of on-disk 2788 * orphan list. If so skip on-disk list modification. 2789 */ 2790 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) > 2791 (le32_to_cpu(sbi->s_es->s_inodes_count))) { 2792 /* Insert this inode at the head of the on-disk orphan list */ 2793 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan); 2794 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino); 2795 dirty = true; 2796 } 2797 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan); 2798 mutex_unlock(&sbi->s_orphan_lock); 2799 2800 if (dirty) { 2801 err = ext4_handle_dirty_super(handle, sb); 2802 rc = ext4_mark_iloc_dirty(handle, inode, &iloc); 2803 if (!err) 2804 err = rc; 2805 if (err) { 2806 /* 2807 * We have to remove inode from in-memory list if 2808 * addition to on disk orphan list failed. Stray orphan 2809 * list entries can cause panics at unmount time. 2810 */ 2811 mutex_lock(&sbi->s_orphan_lock); 2812 list_del_init(&EXT4_I(inode)->i_orphan); 2813 mutex_unlock(&sbi->s_orphan_lock); 2814 } 2815 } else 2816 brelse(iloc.bh); 2817 2818 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino); 2819 jbd_debug(4, "orphan inode %lu will point to %d\n", 2820 inode->i_ino, NEXT_ORPHAN(inode)); 2821 out: 2822 ext4_std_error(sb, err); 2823 return err; 2824 } 2825 2826 /* 2827 * ext4_orphan_del() removes an unlinked or truncated inode from the list 2828 * of such inodes stored on disk, because it is finally being cleaned up. 2829 */ 2830 int ext4_orphan_del(handle_t *handle, struct inode *inode) 2831 { 2832 struct list_head *prev; 2833 struct ext4_inode_info *ei = EXT4_I(inode); 2834 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 2835 __u32 ino_next; 2836 struct ext4_iloc iloc; 2837 int err = 0; 2838 2839 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS)) 2840 return 0; 2841 2842 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) && 2843 !inode_is_locked(inode)); 2844 /* Do this quick check before taking global s_orphan_lock. */ 2845 if (list_empty(&ei->i_orphan)) 2846 return 0; 2847 2848 if (handle) { 2849 /* Grab inode buffer early before taking global s_orphan_lock */ 2850 err = ext4_reserve_inode_write(handle, inode, &iloc); 2851 } 2852 2853 mutex_lock(&sbi->s_orphan_lock); 2854 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino); 2855 2856 prev = ei->i_orphan.prev; 2857 list_del_init(&ei->i_orphan); 2858 2859 /* If we're on an error path, we may not have a valid 2860 * transaction handle with which to update the orphan list on 2861 * disk, but we still need to remove the inode from the linked 2862 * list in memory. */ 2863 if (!handle || err) { 2864 mutex_unlock(&sbi->s_orphan_lock); 2865 goto out_err; 2866 } 2867 2868 ino_next = NEXT_ORPHAN(inode); 2869 if (prev == &sbi->s_orphan) { 2870 jbd_debug(4, "superblock will point to %u\n", ino_next); 2871 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 2872 err = ext4_journal_get_write_access(handle, sbi->s_sbh); 2873 if (err) { 2874 mutex_unlock(&sbi->s_orphan_lock); 2875 goto out_brelse; 2876 } 2877 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next); 2878 mutex_unlock(&sbi->s_orphan_lock); 2879 err = ext4_handle_dirty_super(handle, inode->i_sb); 2880 } else { 2881 struct ext4_iloc iloc2; 2882 struct inode *i_prev = 2883 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode; 2884 2885 jbd_debug(4, "orphan inode %lu will point to %u\n", 2886 i_prev->i_ino, ino_next); 2887 err = ext4_reserve_inode_write(handle, i_prev, &iloc2); 2888 if (err) { 2889 mutex_unlock(&sbi->s_orphan_lock); 2890 goto out_brelse; 2891 } 2892 NEXT_ORPHAN(i_prev) = ino_next; 2893 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2); 2894 mutex_unlock(&sbi->s_orphan_lock); 2895 } 2896 if (err) 2897 goto out_brelse; 2898 NEXT_ORPHAN(inode) = 0; 2899 err = ext4_mark_iloc_dirty(handle, inode, &iloc); 2900 out_err: 2901 ext4_std_error(inode->i_sb, err); 2902 return err; 2903 2904 out_brelse: 2905 brelse(iloc.bh); 2906 goto out_err; 2907 } 2908 2909 static int ext4_rmdir(struct inode *dir, struct dentry *dentry) 2910 { 2911 int retval; 2912 struct inode *inode; 2913 struct buffer_head *bh; 2914 struct ext4_dir_entry_2 *de; 2915 handle_t *handle = NULL; 2916 2917 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb)))) 2918 return -EIO; 2919 2920 /* Initialize quotas before so that eventual writes go in 2921 * separate transaction */ 2922 retval = dquot_initialize(dir); 2923 if (retval) 2924 return retval; 2925 retval = dquot_initialize(d_inode(dentry)); 2926 if (retval) 2927 return retval; 2928 2929 retval = -ENOENT; 2930 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); 2931 if (IS_ERR(bh)) 2932 return PTR_ERR(bh); 2933 if (!bh) 2934 goto end_rmdir; 2935 2936 inode = d_inode(dentry); 2937 2938 retval = -EFSCORRUPTED; 2939 if (le32_to_cpu(de->inode) != inode->i_ino) 2940 goto end_rmdir; 2941 2942 retval = -ENOTEMPTY; 2943 if (!ext4_empty_dir(inode)) 2944 goto end_rmdir; 2945 2946 handle = ext4_journal_start(dir, EXT4_HT_DIR, 2947 EXT4_DATA_TRANS_BLOCKS(dir->i_sb)); 2948 if (IS_ERR(handle)) { 2949 retval = PTR_ERR(handle); 2950 handle = NULL; 2951 goto end_rmdir; 2952 } 2953 2954 if (IS_DIRSYNC(dir)) 2955 ext4_handle_sync(handle); 2956 2957 retval = ext4_delete_entry(handle, dir, de, bh); 2958 if (retval) 2959 goto end_rmdir; 2960 if (!EXT4_DIR_LINK_EMPTY(inode)) 2961 ext4_warning_inode(inode, 2962 "empty directory '%.*s' has too many links (%u)", 2963 dentry->d_name.len, dentry->d_name.name, 2964 inode->i_nlink); 2965 inode_inc_iversion(inode); 2966 clear_nlink(inode); 2967 /* There's no need to set i_disksize: the fact that i_nlink is 2968 * zero will ensure that the right thing happens during any 2969 * recovery. */ 2970 inode->i_size = 0; 2971 ext4_orphan_add(handle, inode); 2972 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); 2973 ext4_mark_inode_dirty(handle, inode); 2974 ext4_dec_count(handle, dir); 2975 ext4_update_dx_flag(dir); 2976 ext4_mark_inode_dirty(handle, dir); 2977 2978 end_rmdir: 2979 brelse(bh); 2980 if (handle) 2981 ext4_journal_stop(handle); 2982 return retval; 2983 } 2984 2985 static int ext4_unlink(struct inode *dir, struct dentry *dentry) 2986 { 2987 int retval; 2988 struct inode *inode; 2989 struct buffer_head *bh; 2990 struct ext4_dir_entry_2 *de; 2991 handle_t *handle = NULL; 2992 2993 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb)))) 2994 return -EIO; 2995 2996 trace_ext4_unlink_enter(dir, dentry); 2997 /* Initialize quotas before so that eventual writes go 2998 * in separate transaction */ 2999 retval = dquot_initialize(dir); 3000 if (retval) 3001 return retval; 3002 retval = dquot_initialize(d_inode(dentry)); 3003 if (retval) 3004 return retval; 3005 3006 retval = -ENOENT; 3007 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); 3008 if (IS_ERR(bh)) 3009 return PTR_ERR(bh); 3010 if (!bh) 3011 goto end_unlink; 3012 3013 inode = d_inode(dentry); 3014 3015 retval = -EFSCORRUPTED; 3016 if (le32_to_cpu(de->inode) != inode->i_ino) 3017 goto end_unlink; 3018 3019 handle = ext4_journal_start(dir, EXT4_HT_DIR, 3020 EXT4_DATA_TRANS_BLOCKS(dir->i_sb)); 3021 if (IS_ERR(handle)) { 3022 retval = PTR_ERR(handle); 3023 handle = NULL; 3024 goto end_unlink; 3025 } 3026 3027 if (IS_DIRSYNC(dir)) 3028 ext4_handle_sync(handle); 3029 3030 if (inode->i_nlink == 0) { 3031 ext4_warning_inode(inode, "Deleting file '%.*s' with no links", 3032 dentry->d_name.len, dentry->d_name.name); 3033 set_nlink(inode, 1); 3034 } 3035 retval = ext4_delete_entry(handle, dir, de, bh); 3036 if (retval) 3037 goto end_unlink; 3038 dir->i_ctime = dir->i_mtime = current_time(dir); 3039 ext4_update_dx_flag(dir); 3040 ext4_mark_inode_dirty(handle, dir); 3041 drop_nlink(inode); 3042 if (!inode->i_nlink) 3043 ext4_orphan_add(handle, inode); 3044 inode->i_ctime = current_time(inode); 3045 ext4_mark_inode_dirty(handle, inode); 3046 3047 end_unlink: 3048 brelse(bh); 3049 if (handle) 3050 ext4_journal_stop(handle); 3051 trace_ext4_unlink_exit(dentry, retval); 3052 return retval; 3053 } 3054 3055 static int ext4_symlink(struct inode *dir, 3056 struct dentry *dentry, const char *symname) 3057 { 3058 handle_t *handle; 3059 struct inode *inode; 3060 int err, len = strlen(symname); 3061 int credits; 3062 struct fscrypt_str disk_link; 3063 3064 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb)))) 3065 return -EIO; 3066 3067 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize, 3068 &disk_link); 3069 if (err) 3070 return err; 3071 3072 err = dquot_initialize(dir); 3073 if (err) 3074 return err; 3075 3076 if ((disk_link.len > EXT4_N_BLOCKS * 4)) { 3077 /* 3078 * For non-fast symlinks, we just allocate inode and put it on 3079 * orphan list in the first transaction => we need bitmap, 3080 * group descriptor, sb, inode block, quota blocks, and 3081 * possibly selinux xattr blocks. 3082 */ 3083 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) + 3084 EXT4_XATTR_TRANS_BLOCKS; 3085 } else { 3086 /* 3087 * Fast symlink. We have to add entry to directory 3088 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS), 3089 * allocate new inode (bitmap, group descriptor, inode block, 3090 * quota blocks, sb is already counted in previous macros). 3091 */ 3092 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 3093 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3; 3094 } 3095 3096 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO, 3097 &dentry->d_name, 0, NULL, 3098 EXT4_HT_DIR, credits); 3099 handle = ext4_journal_current_handle(); 3100 if (IS_ERR(inode)) { 3101 if (handle) 3102 ext4_journal_stop(handle); 3103 return PTR_ERR(inode); 3104 } 3105 3106 if (IS_ENCRYPTED(inode)) { 3107 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link); 3108 if (err) 3109 goto err_drop_inode; 3110 inode->i_op = &ext4_encrypted_symlink_inode_operations; 3111 } 3112 3113 if ((disk_link.len > EXT4_N_BLOCKS * 4)) { 3114 if (!IS_ENCRYPTED(inode)) 3115 inode->i_op = &ext4_symlink_inode_operations; 3116 inode_nohighmem(inode); 3117 ext4_set_aops(inode); 3118 /* 3119 * We cannot call page_symlink() with transaction started 3120 * because it calls into ext4_write_begin() which can wait 3121 * for transaction commit if we are running out of space 3122 * and thus we deadlock. So we have to stop transaction now 3123 * and restart it when symlink contents is written. 3124 * 3125 * To keep fs consistent in case of crash, we have to put inode 3126 * to orphan list in the mean time. 3127 */ 3128 drop_nlink(inode); 3129 err = ext4_orphan_add(handle, inode); 3130 ext4_journal_stop(handle); 3131 handle = NULL; 3132 if (err) 3133 goto err_drop_inode; 3134 err = __page_symlink(inode, disk_link.name, disk_link.len, 1); 3135 if (err) 3136 goto err_drop_inode; 3137 /* 3138 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS 3139 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified 3140 */ 3141 handle = ext4_journal_start(dir, EXT4_HT_DIR, 3142 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 3143 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1); 3144 if (IS_ERR(handle)) { 3145 err = PTR_ERR(handle); 3146 handle = NULL; 3147 goto err_drop_inode; 3148 } 3149 set_nlink(inode, 1); 3150 err = ext4_orphan_del(handle, inode); 3151 if (err) 3152 goto err_drop_inode; 3153 } else { 3154 /* clear the extent format for fast symlink */ 3155 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS); 3156 if (!IS_ENCRYPTED(inode)) { 3157 inode->i_op = &ext4_fast_symlink_inode_operations; 3158 inode->i_link = (char *)&EXT4_I(inode)->i_data; 3159 } 3160 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name, 3161 disk_link.len); 3162 inode->i_size = disk_link.len - 1; 3163 } 3164 EXT4_I(inode)->i_disksize = inode->i_size; 3165 err = ext4_add_nondir(handle, dentry, inode); 3166 if (!err && IS_DIRSYNC(dir)) 3167 ext4_handle_sync(handle); 3168 3169 if (handle) 3170 ext4_journal_stop(handle); 3171 goto out_free_encrypted_link; 3172 3173 err_drop_inode: 3174 if (handle) 3175 ext4_journal_stop(handle); 3176 clear_nlink(inode); 3177 unlock_new_inode(inode); 3178 iput(inode); 3179 out_free_encrypted_link: 3180 if (disk_link.name != (unsigned char *)symname) 3181 kfree(disk_link.name); 3182 return err; 3183 } 3184 3185 static int ext4_link(struct dentry *old_dentry, 3186 struct inode *dir, struct dentry *dentry) 3187 { 3188 handle_t *handle; 3189 struct inode *inode = d_inode(old_dentry); 3190 int err, retries = 0; 3191 3192 if (inode->i_nlink >= EXT4_LINK_MAX) 3193 return -EMLINK; 3194 3195 err = fscrypt_prepare_link(old_dentry, dir, dentry); 3196 if (err) 3197 return err; 3198 3199 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) && 3200 (!projid_eq(EXT4_I(dir)->i_projid, 3201 EXT4_I(old_dentry->d_inode)->i_projid))) 3202 return -EXDEV; 3203 3204 err = dquot_initialize(dir); 3205 if (err) 3206 return err; 3207 3208 retry: 3209 handle = ext4_journal_start(dir, EXT4_HT_DIR, 3210 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 3211 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1); 3212 if (IS_ERR(handle)) 3213 return PTR_ERR(handle); 3214 3215 if (IS_DIRSYNC(dir)) 3216 ext4_handle_sync(handle); 3217 3218 inode->i_ctime = current_time(inode); 3219 ext4_inc_count(handle, inode); 3220 ihold(inode); 3221 3222 err = ext4_add_entry(handle, dentry, inode); 3223 if (!err) { 3224 ext4_mark_inode_dirty(handle, inode); 3225 /* this can happen only for tmpfile being 3226 * linked the first time 3227 */ 3228 if (inode->i_nlink == 1) 3229 ext4_orphan_del(handle, inode); 3230 d_instantiate(dentry, inode); 3231 } else { 3232 drop_nlink(inode); 3233 iput(inode); 3234 } 3235 ext4_journal_stop(handle); 3236 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 3237 goto retry; 3238 return err; 3239 } 3240 3241 3242 /* 3243 * Try to find buffer head where contains the parent block. 3244 * It should be the inode block if it is inlined or the 1st block 3245 * if it is a normal dir. 3246 */ 3247 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle, 3248 struct inode *inode, 3249 int *retval, 3250 struct ext4_dir_entry_2 **parent_de, 3251 int *inlined) 3252 { 3253 struct buffer_head *bh; 3254 3255 if (!ext4_has_inline_data(inode)) { 3256 bh = ext4_read_dirblock(inode, 0, EITHER); 3257 if (IS_ERR(bh)) { 3258 *retval = PTR_ERR(bh); 3259 return NULL; 3260 } 3261 *parent_de = ext4_next_entry( 3262 (struct ext4_dir_entry_2 *)bh->b_data, 3263 inode->i_sb->s_blocksize); 3264 return bh; 3265 } 3266 3267 *inlined = 1; 3268 return ext4_get_first_inline_block(inode, parent_de, retval); 3269 } 3270 3271 struct ext4_renament { 3272 struct inode *dir; 3273 struct dentry *dentry; 3274 struct inode *inode; 3275 bool is_dir; 3276 int dir_nlink_delta; 3277 3278 /* entry for "dentry" */ 3279 struct buffer_head *bh; 3280 struct ext4_dir_entry_2 *de; 3281 int inlined; 3282 3283 /* entry for ".." in inode if it's a directory */ 3284 struct buffer_head *dir_bh; 3285 struct ext4_dir_entry_2 *parent_de; 3286 int dir_inlined; 3287 }; 3288 3289 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent) 3290 { 3291 int retval; 3292 3293 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode, 3294 &retval, &ent->parent_de, 3295 &ent->dir_inlined); 3296 if (!ent->dir_bh) 3297 return retval; 3298 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino) 3299 return -EFSCORRUPTED; 3300 BUFFER_TRACE(ent->dir_bh, "get_write_access"); 3301 return ext4_journal_get_write_access(handle, ent->dir_bh); 3302 } 3303 3304 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent, 3305 unsigned dir_ino) 3306 { 3307 int retval; 3308 3309 ent->parent_de->inode = cpu_to_le32(dir_ino); 3310 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata"); 3311 if (!ent->dir_inlined) { 3312 if (is_dx(ent->inode)) { 3313 retval = ext4_handle_dirty_dx_node(handle, 3314 ent->inode, 3315 ent->dir_bh); 3316 } else { 3317 retval = ext4_handle_dirty_dirent_node(handle, 3318 ent->inode, 3319 ent->dir_bh); 3320 } 3321 } else { 3322 retval = ext4_mark_inode_dirty(handle, ent->inode); 3323 } 3324 if (retval) { 3325 ext4_std_error(ent->dir->i_sb, retval); 3326 return retval; 3327 } 3328 return 0; 3329 } 3330 3331 static int ext4_setent(handle_t *handle, struct ext4_renament *ent, 3332 unsigned ino, unsigned file_type) 3333 { 3334 int retval; 3335 3336 BUFFER_TRACE(ent->bh, "get write access"); 3337 retval = ext4_journal_get_write_access(handle, ent->bh); 3338 if (retval) 3339 return retval; 3340 ent->de->inode = cpu_to_le32(ino); 3341 if (ext4_has_feature_filetype(ent->dir->i_sb)) 3342 ent->de->file_type = file_type; 3343 inode_inc_iversion(ent->dir); 3344 ent->dir->i_ctime = ent->dir->i_mtime = 3345 current_time(ent->dir); 3346 ext4_mark_inode_dirty(handle, ent->dir); 3347 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata"); 3348 if (!ent->inlined) { 3349 retval = ext4_handle_dirty_dirent_node(handle, 3350 ent->dir, ent->bh); 3351 if (unlikely(retval)) { 3352 ext4_std_error(ent->dir->i_sb, retval); 3353 return retval; 3354 } 3355 } 3356 brelse(ent->bh); 3357 ent->bh = NULL; 3358 3359 return 0; 3360 } 3361 3362 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir, 3363 const struct qstr *d_name) 3364 { 3365 int retval = -ENOENT; 3366 struct buffer_head *bh; 3367 struct ext4_dir_entry_2 *de; 3368 3369 bh = ext4_find_entry(dir, d_name, &de, NULL); 3370 if (IS_ERR(bh)) 3371 return PTR_ERR(bh); 3372 if (bh) { 3373 retval = ext4_delete_entry(handle, dir, de, bh); 3374 brelse(bh); 3375 } 3376 return retval; 3377 } 3378 3379 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent, 3380 int force_reread) 3381 { 3382 int retval; 3383 /* 3384 * ent->de could have moved from under us during htree split, so make 3385 * sure that we are deleting the right entry. We might also be pointing 3386 * to a stale entry in the unused part of ent->bh so just checking inum 3387 * and the name isn't enough. 3388 */ 3389 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino || 3390 ent->de->name_len != ent->dentry->d_name.len || 3391 strncmp(ent->de->name, ent->dentry->d_name.name, 3392 ent->de->name_len) || 3393 force_reread) { 3394 retval = ext4_find_delete_entry(handle, ent->dir, 3395 &ent->dentry->d_name); 3396 } else { 3397 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh); 3398 if (retval == -ENOENT) { 3399 retval = ext4_find_delete_entry(handle, ent->dir, 3400 &ent->dentry->d_name); 3401 } 3402 } 3403 3404 if (retval) { 3405 ext4_warning_inode(ent->dir, 3406 "Deleting old file: nlink %d, error=%d", 3407 ent->dir->i_nlink, retval); 3408 } 3409 } 3410 3411 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent) 3412 { 3413 if (ent->dir_nlink_delta) { 3414 if (ent->dir_nlink_delta == -1) 3415 ext4_dec_count(handle, ent->dir); 3416 else 3417 ext4_inc_count(handle, ent->dir); 3418 ext4_mark_inode_dirty(handle, ent->dir); 3419 } 3420 } 3421 3422 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent, 3423 int credits, handle_t **h) 3424 { 3425 struct inode *wh; 3426 handle_t *handle; 3427 int retries = 0; 3428 3429 /* 3430 * for inode block, sb block, group summaries, 3431 * and inode bitmap 3432 */ 3433 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) + 3434 EXT4_XATTR_TRANS_BLOCKS + 4); 3435 retry: 3436 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE, 3437 &ent->dentry->d_name, 0, NULL, 3438 EXT4_HT_DIR, credits); 3439 3440 handle = ext4_journal_current_handle(); 3441 if (IS_ERR(wh)) { 3442 if (handle) 3443 ext4_journal_stop(handle); 3444 if (PTR_ERR(wh) == -ENOSPC && 3445 ext4_should_retry_alloc(ent->dir->i_sb, &retries)) 3446 goto retry; 3447 } else { 3448 *h = handle; 3449 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV); 3450 wh->i_op = &ext4_special_inode_operations; 3451 } 3452 return wh; 3453 } 3454 3455 /* 3456 * Anybody can rename anything with this: the permission checks are left to the 3457 * higher-level routines. 3458 * 3459 * n.b. old_{dentry,inode) refers to the source dentry/inode 3460 * while new_{dentry,inode) refers to the destination dentry/inode 3461 * This comes from rename(const char *oldpath, const char *newpath) 3462 */ 3463 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry, 3464 struct inode *new_dir, struct dentry *new_dentry, 3465 unsigned int flags) 3466 { 3467 handle_t *handle = NULL; 3468 struct ext4_renament old = { 3469 .dir = old_dir, 3470 .dentry = old_dentry, 3471 .inode = d_inode(old_dentry), 3472 }; 3473 struct ext4_renament new = { 3474 .dir = new_dir, 3475 .dentry = new_dentry, 3476 .inode = d_inode(new_dentry), 3477 }; 3478 int force_reread; 3479 int retval; 3480 struct inode *whiteout = NULL; 3481 int credits; 3482 u8 old_file_type; 3483 3484 if (new.inode && new.inode->i_nlink == 0) { 3485 EXT4_ERROR_INODE(new.inode, 3486 "target of rename is already freed"); 3487 return -EFSCORRUPTED; 3488 } 3489 3490 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) && 3491 (!projid_eq(EXT4_I(new_dir)->i_projid, 3492 EXT4_I(old_dentry->d_inode)->i_projid))) 3493 return -EXDEV; 3494 3495 retval = dquot_initialize(old.dir); 3496 if (retval) 3497 return retval; 3498 retval = dquot_initialize(new.dir); 3499 if (retval) 3500 return retval; 3501 3502 /* Initialize quotas before so that eventual writes go 3503 * in separate transaction */ 3504 if (new.inode) { 3505 retval = dquot_initialize(new.inode); 3506 if (retval) 3507 return retval; 3508 } 3509 3510 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL); 3511 if (IS_ERR(old.bh)) 3512 return PTR_ERR(old.bh); 3513 /* 3514 * Check for inode number is _not_ due to possible IO errors. 3515 * We might rmdir the source, keep it as pwd of some process 3516 * and merrily kill the link to whatever was created under the 3517 * same name. Goodbye sticky bit ;-< 3518 */ 3519 retval = -ENOENT; 3520 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino) 3521 goto end_rename; 3522 3523 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name, 3524 &new.de, &new.inlined); 3525 if (IS_ERR(new.bh)) { 3526 retval = PTR_ERR(new.bh); 3527 new.bh = NULL; 3528 goto end_rename; 3529 } 3530 if (new.bh) { 3531 if (!new.inode) { 3532 brelse(new.bh); 3533 new.bh = NULL; 3534 } 3535 } 3536 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC)) 3537 ext4_alloc_da_blocks(old.inode); 3538 3539 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) + 3540 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2); 3541 if (!(flags & RENAME_WHITEOUT)) { 3542 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits); 3543 if (IS_ERR(handle)) { 3544 retval = PTR_ERR(handle); 3545 handle = NULL; 3546 goto end_rename; 3547 } 3548 } else { 3549 whiteout = ext4_whiteout_for_rename(&old, credits, &handle); 3550 if (IS_ERR(whiteout)) { 3551 retval = PTR_ERR(whiteout); 3552 whiteout = NULL; 3553 goto end_rename; 3554 } 3555 } 3556 3557 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir)) 3558 ext4_handle_sync(handle); 3559 3560 if (S_ISDIR(old.inode->i_mode)) { 3561 if (new.inode) { 3562 retval = -ENOTEMPTY; 3563 if (!ext4_empty_dir(new.inode)) 3564 goto end_rename; 3565 } else { 3566 retval = -EMLINK; 3567 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir)) 3568 goto end_rename; 3569 } 3570 retval = ext4_rename_dir_prepare(handle, &old); 3571 if (retval) 3572 goto end_rename; 3573 } 3574 /* 3575 * If we're renaming a file within an inline_data dir and adding or 3576 * setting the new dirent causes a conversion from inline_data to 3577 * extents/blockmap, we need to force the dirent delete code to 3578 * re-read the directory, or else we end up trying to delete a dirent 3579 * from what is now the extent tree root (or a block map). 3580 */ 3581 force_reread = (new.dir->i_ino == old.dir->i_ino && 3582 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA)); 3583 3584 old_file_type = old.de->file_type; 3585 if (whiteout) { 3586 /* 3587 * Do this before adding a new entry, so the old entry is sure 3588 * to be still pointing to the valid old entry. 3589 */ 3590 retval = ext4_setent(handle, &old, whiteout->i_ino, 3591 EXT4_FT_CHRDEV); 3592 if (retval) 3593 goto end_rename; 3594 ext4_mark_inode_dirty(handle, whiteout); 3595 } 3596 if (!new.bh) { 3597 retval = ext4_add_entry(handle, new.dentry, old.inode); 3598 if (retval) 3599 goto end_rename; 3600 } else { 3601 retval = ext4_setent(handle, &new, 3602 old.inode->i_ino, old_file_type); 3603 if (retval) 3604 goto end_rename; 3605 } 3606 if (force_reread) 3607 force_reread = !ext4_test_inode_flag(new.dir, 3608 EXT4_INODE_INLINE_DATA); 3609 3610 /* 3611 * Like most other Unix systems, set the ctime for inodes on a 3612 * rename. 3613 */ 3614 old.inode->i_ctime = current_time(old.inode); 3615 ext4_mark_inode_dirty(handle, old.inode); 3616 3617 if (!whiteout) { 3618 /* 3619 * ok, that's it 3620 */ 3621 ext4_rename_delete(handle, &old, force_reread); 3622 } 3623 3624 if (new.inode) { 3625 ext4_dec_count(handle, new.inode); 3626 new.inode->i_ctime = current_time(new.inode); 3627 } 3628 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir); 3629 ext4_update_dx_flag(old.dir); 3630 if (old.dir_bh) { 3631 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino); 3632 if (retval) 3633 goto end_rename; 3634 3635 ext4_dec_count(handle, old.dir); 3636 if (new.inode) { 3637 /* checked ext4_empty_dir above, can't have another 3638 * parent, ext4_dec_count() won't work for many-linked 3639 * dirs */ 3640 clear_nlink(new.inode); 3641 } else { 3642 ext4_inc_count(handle, new.dir); 3643 ext4_update_dx_flag(new.dir); 3644 ext4_mark_inode_dirty(handle, new.dir); 3645 } 3646 } 3647 ext4_mark_inode_dirty(handle, old.dir); 3648 if (new.inode) { 3649 ext4_mark_inode_dirty(handle, new.inode); 3650 if (!new.inode->i_nlink) 3651 ext4_orphan_add(handle, new.inode); 3652 } 3653 retval = 0; 3654 3655 end_rename: 3656 brelse(old.dir_bh); 3657 brelse(old.bh); 3658 brelse(new.bh); 3659 if (whiteout) { 3660 if (retval) 3661 drop_nlink(whiteout); 3662 unlock_new_inode(whiteout); 3663 iput(whiteout); 3664 } 3665 if (handle) 3666 ext4_journal_stop(handle); 3667 return retval; 3668 } 3669 3670 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry, 3671 struct inode *new_dir, struct dentry *new_dentry) 3672 { 3673 handle_t *handle = NULL; 3674 struct ext4_renament old = { 3675 .dir = old_dir, 3676 .dentry = old_dentry, 3677 .inode = d_inode(old_dentry), 3678 }; 3679 struct ext4_renament new = { 3680 .dir = new_dir, 3681 .dentry = new_dentry, 3682 .inode = d_inode(new_dentry), 3683 }; 3684 u8 new_file_type; 3685 int retval; 3686 struct timespec64 ctime; 3687 3688 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) && 3689 !projid_eq(EXT4_I(new_dir)->i_projid, 3690 EXT4_I(old_dentry->d_inode)->i_projid)) || 3691 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) && 3692 !projid_eq(EXT4_I(old_dir)->i_projid, 3693 EXT4_I(new_dentry->d_inode)->i_projid))) 3694 return -EXDEV; 3695 3696 retval = dquot_initialize(old.dir); 3697 if (retval) 3698 return retval; 3699 retval = dquot_initialize(new.dir); 3700 if (retval) 3701 return retval; 3702 3703 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, 3704 &old.de, &old.inlined); 3705 if (IS_ERR(old.bh)) 3706 return PTR_ERR(old.bh); 3707 /* 3708 * Check for inode number is _not_ due to possible IO errors. 3709 * We might rmdir the source, keep it as pwd of some process 3710 * and merrily kill the link to whatever was created under the 3711 * same name. Goodbye sticky bit ;-< 3712 */ 3713 retval = -ENOENT; 3714 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino) 3715 goto end_rename; 3716 3717 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name, 3718 &new.de, &new.inlined); 3719 if (IS_ERR(new.bh)) { 3720 retval = PTR_ERR(new.bh); 3721 new.bh = NULL; 3722 goto end_rename; 3723 } 3724 3725 /* RENAME_EXCHANGE case: old *and* new must both exist */ 3726 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino) 3727 goto end_rename; 3728 3729 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, 3730 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) + 3731 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2)); 3732 if (IS_ERR(handle)) { 3733 retval = PTR_ERR(handle); 3734 handle = NULL; 3735 goto end_rename; 3736 } 3737 3738 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir)) 3739 ext4_handle_sync(handle); 3740 3741 if (S_ISDIR(old.inode->i_mode)) { 3742 old.is_dir = true; 3743 retval = ext4_rename_dir_prepare(handle, &old); 3744 if (retval) 3745 goto end_rename; 3746 } 3747 if (S_ISDIR(new.inode->i_mode)) { 3748 new.is_dir = true; 3749 retval = ext4_rename_dir_prepare(handle, &new); 3750 if (retval) 3751 goto end_rename; 3752 } 3753 3754 /* 3755 * Other than the special case of overwriting a directory, parents' 3756 * nlink only needs to be modified if this is a cross directory rename. 3757 */ 3758 if (old.dir != new.dir && old.is_dir != new.is_dir) { 3759 old.dir_nlink_delta = old.is_dir ? -1 : 1; 3760 new.dir_nlink_delta = -old.dir_nlink_delta; 3761 retval = -EMLINK; 3762 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) || 3763 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir))) 3764 goto end_rename; 3765 } 3766 3767 new_file_type = new.de->file_type; 3768 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type); 3769 if (retval) 3770 goto end_rename; 3771 3772 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type); 3773 if (retval) 3774 goto end_rename; 3775 3776 /* 3777 * Like most other Unix systems, set the ctime for inodes on a 3778 * rename. 3779 */ 3780 ctime = current_time(old.inode); 3781 old.inode->i_ctime = ctime; 3782 new.inode->i_ctime = ctime; 3783 ext4_mark_inode_dirty(handle, old.inode); 3784 ext4_mark_inode_dirty(handle, new.inode); 3785 3786 if (old.dir_bh) { 3787 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino); 3788 if (retval) 3789 goto end_rename; 3790 } 3791 if (new.dir_bh) { 3792 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino); 3793 if (retval) 3794 goto end_rename; 3795 } 3796 ext4_update_dir_count(handle, &old); 3797 ext4_update_dir_count(handle, &new); 3798 retval = 0; 3799 3800 end_rename: 3801 brelse(old.dir_bh); 3802 brelse(new.dir_bh); 3803 brelse(old.bh); 3804 brelse(new.bh); 3805 if (handle) 3806 ext4_journal_stop(handle); 3807 return retval; 3808 } 3809 3810 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry, 3811 struct inode *new_dir, struct dentry *new_dentry, 3812 unsigned int flags) 3813 { 3814 int err; 3815 3816 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb)))) 3817 return -EIO; 3818 3819 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 3820 return -EINVAL; 3821 3822 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry, 3823 flags); 3824 if (err) 3825 return err; 3826 3827 if (flags & RENAME_EXCHANGE) { 3828 return ext4_cross_rename(old_dir, old_dentry, 3829 new_dir, new_dentry); 3830 } 3831 3832 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags); 3833 } 3834 3835 /* 3836 * directories can handle most operations... 3837 */ 3838 const struct inode_operations ext4_dir_inode_operations = { 3839 .create = ext4_create, 3840 .lookup = ext4_lookup, 3841 .link = ext4_link, 3842 .unlink = ext4_unlink, 3843 .symlink = ext4_symlink, 3844 .mkdir = ext4_mkdir, 3845 .rmdir = ext4_rmdir, 3846 .mknod = ext4_mknod, 3847 .tmpfile = ext4_tmpfile, 3848 .rename = ext4_rename2, 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 .fiemap = ext4_fiemap, 3855 }; 3856 3857 const struct inode_operations ext4_special_inode_operations = { 3858 .setattr = ext4_setattr, 3859 .getattr = ext4_getattr, 3860 .listxattr = ext4_listxattr, 3861 .get_acl = ext4_get_acl, 3862 .set_acl = ext4_set_acl, 3863 }; 3864