1 /* 2 * linux/fs/ext4/namei.c 3 * 4 * Copyright (C) 1992, 1993, 1994, 1995 5 * Remy Card (card@masi.ibp.fr) 6 * Laboratoire MASI - Institut Blaise Pascal 7 * Universite Pierre et Marie Curie (Paris VI) 8 * 9 * from 10 * 11 * linux/fs/minix/namei.c 12 * 13 * Copyright (C) 1991, 1992 Linus Torvalds 14 * 15 * Big-endian to little-endian byte-swapping/bitmaps by 16 * David S. Miller (davem@caip.rutgers.edu), 1995 17 * Directory entry file type support and forward compatibility hooks 18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998 19 * Hash Tree Directory indexing (c) 20 * Daniel Phillips, 2001 21 * Hash Tree Directory indexing porting 22 * Christopher Li, 2002 23 * Hash Tree Directory indexing cleanup 24 * Theodore Ts'o, 2002 25 */ 26 27 #include <linux/fs.h> 28 #include <linux/pagemap.h> 29 #include <linux/time.h> 30 #include <linux/fcntl.h> 31 #include <linux/stat.h> 32 #include <linux/string.h> 33 #include <linux/quotaops.h> 34 #include <linux/buffer_head.h> 35 #include <linux/bio.h> 36 #include "ext4.h" 37 #include "ext4_jbd2.h" 38 39 #include "xattr.h" 40 #include "acl.h" 41 42 #include <trace/events/ext4.h> 43 /* 44 * define how far ahead to read directories while searching them. 45 */ 46 #define NAMEI_RA_CHUNKS 2 47 #define NAMEI_RA_BLOCKS 4 48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS) 49 50 static struct buffer_head *ext4_append(handle_t *handle, 51 struct inode *inode, 52 ext4_lblk_t *block) 53 { 54 struct buffer_head *bh; 55 int err; 56 57 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb && 58 ((inode->i_size >> 10) >= 59 EXT4_SB(inode->i_sb)->s_max_dir_size_kb))) 60 return ERR_PTR(-ENOSPC); 61 62 *block = inode->i_size >> inode->i_sb->s_blocksize_bits; 63 64 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE); 65 if (IS_ERR(bh)) 66 return bh; 67 inode->i_size += inode->i_sb->s_blocksize; 68 EXT4_I(inode)->i_disksize = inode->i_size; 69 BUFFER_TRACE(bh, "get_write_access"); 70 err = ext4_journal_get_write_access(handle, bh); 71 if (err) { 72 brelse(bh); 73 ext4_std_error(inode->i_sb, err); 74 return ERR_PTR(err); 75 } 76 return bh; 77 } 78 79 static int ext4_dx_csum_verify(struct inode *inode, 80 struct ext4_dir_entry *dirent); 81 82 typedef enum { 83 EITHER, INDEX, DIRENT 84 } dirblock_type_t; 85 86 #define ext4_read_dirblock(inode, block, type) \ 87 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__) 88 89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode, 90 ext4_lblk_t block, 91 dirblock_type_t type, 92 const char *func, 93 unsigned int line) 94 { 95 struct buffer_head *bh; 96 struct ext4_dir_entry *dirent; 97 int is_dx_block = 0; 98 99 bh = ext4_bread(NULL, inode, block, 0); 100 if (IS_ERR(bh)) { 101 __ext4_warning(inode->i_sb, func, line, 102 "inode #%lu: lblock %lu: comm %s: " 103 "error %ld reading directory block", 104 inode->i_ino, (unsigned long)block, 105 current->comm, PTR_ERR(bh)); 106 107 return bh; 108 } 109 if (!bh) { 110 ext4_error_inode(inode, func, line, block, 111 "Directory hole found"); 112 return ERR_PTR(-EFSCORRUPTED); 113 } 114 dirent = (struct ext4_dir_entry *) bh->b_data; 115 /* Determine whether or not we have an index block */ 116 if (is_dx(inode)) { 117 if (block == 0) 118 is_dx_block = 1; 119 else if (ext4_rec_len_from_disk(dirent->rec_len, 120 inode->i_sb->s_blocksize) == 121 inode->i_sb->s_blocksize) 122 is_dx_block = 1; 123 } 124 if (!is_dx_block && type == INDEX) { 125 ext4_error_inode(inode, func, line, block, 126 "directory leaf block found instead of index block"); 127 return ERR_PTR(-EFSCORRUPTED); 128 } 129 if (!ext4_has_metadata_csum(inode->i_sb) || 130 buffer_verified(bh)) 131 return bh; 132 133 /* 134 * An empty leaf block can get mistaken for a index block; for 135 * this reason, we can only check the index checksum when the 136 * caller is sure it should be an index block. 137 */ 138 if (is_dx_block && type == INDEX) { 139 if (ext4_dx_csum_verify(inode, dirent)) 140 set_buffer_verified(bh); 141 else { 142 ext4_error_inode(inode, func, line, block, 143 "Directory index failed checksum"); 144 brelse(bh); 145 return ERR_PTR(-EFSBADCRC); 146 } 147 } 148 if (!is_dx_block) { 149 if (ext4_dirent_csum_verify(inode, dirent)) 150 set_buffer_verified(bh); 151 else { 152 ext4_error_inode(inode, func, line, block, 153 "Directory block failed checksum"); 154 brelse(bh); 155 return ERR_PTR(-EFSBADCRC); 156 } 157 } 158 return bh; 159 } 160 161 #ifndef assert 162 #define assert(test) J_ASSERT(test) 163 #endif 164 165 #ifdef DX_DEBUG 166 #define dxtrace(command) command 167 #else 168 #define dxtrace(command) 169 #endif 170 171 struct fake_dirent 172 { 173 __le32 inode; 174 __le16 rec_len; 175 u8 name_len; 176 u8 file_type; 177 }; 178 179 struct dx_countlimit 180 { 181 __le16 limit; 182 __le16 count; 183 }; 184 185 struct dx_entry 186 { 187 __le32 hash; 188 __le32 block; 189 }; 190 191 /* 192 * dx_root_info is laid out so that if it should somehow get overlaid by a 193 * dirent the two low bits of the hash version will be zero. Therefore, the 194 * hash version mod 4 should never be 0. Sincerely, the paranoia department. 195 */ 196 197 struct dx_root 198 { 199 struct fake_dirent dot; 200 char dot_name[4]; 201 struct fake_dirent dotdot; 202 char dotdot_name[4]; 203 struct dx_root_info 204 { 205 __le32 reserved_zero; 206 u8 hash_version; 207 u8 info_length; /* 8 */ 208 u8 indirect_levels; 209 u8 unused_flags; 210 } 211 info; 212 struct dx_entry entries[0]; 213 }; 214 215 struct dx_node 216 { 217 struct fake_dirent fake; 218 struct dx_entry entries[0]; 219 }; 220 221 222 struct dx_frame 223 { 224 struct buffer_head *bh; 225 struct dx_entry *entries; 226 struct dx_entry *at; 227 }; 228 229 struct dx_map_entry 230 { 231 u32 hash; 232 u16 offs; 233 u16 size; 234 }; 235 236 /* 237 * This goes at the end of each htree block. 238 */ 239 struct dx_tail { 240 u32 dt_reserved; 241 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */ 242 }; 243 244 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry); 245 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value); 246 static inline unsigned dx_get_hash(struct dx_entry *entry); 247 static void dx_set_hash(struct dx_entry *entry, unsigned value); 248 static unsigned dx_get_count(struct dx_entry *entries); 249 static unsigned dx_get_limit(struct dx_entry *entries); 250 static void dx_set_count(struct dx_entry *entries, unsigned value); 251 static void dx_set_limit(struct dx_entry *entries, unsigned value); 252 static unsigned dx_root_limit(struct inode *dir, unsigned infosize); 253 static unsigned dx_node_limit(struct inode *dir); 254 static struct dx_frame *dx_probe(struct ext4_filename *fname, 255 struct inode *dir, 256 struct dx_hash_info *hinfo, 257 struct dx_frame *frame); 258 static void dx_release(struct dx_frame *frames); 259 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de, 260 unsigned blocksize, struct dx_hash_info *hinfo, 261 struct dx_map_entry map[]); 262 static void dx_sort_map(struct dx_map_entry *map, unsigned count); 263 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to, 264 struct dx_map_entry *offsets, int count, unsigned blocksize); 265 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize); 266 static void dx_insert_block(struct dx_frame *frame, 267 u32 hash, ext4_lblk_t block); 268 static int ext4_htree_next_block(struct inode *dir, __u32 hash, 269 struct dx_frame *frame, 270 struct dx_frame *frames, 271 __u32 *start_hash); 272 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, 273 struct ext4_filename *fname, 274 struct ext4_dir_entry_2 **res_dir); 275 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname, 276 struct inode *dir, struct inode *inode); 277 278 /* checksumming functions */ 279 void initialize_dirent_tail(struct ext4_dir_entry_tail *t, 280 unsigned int blocksize) 281 { 282 memset(t, 0, sizeof(struct ext4_dir_entry_tail)); 283 t->det_rec_len = ext4_rec_len_to_disk( 284 sizeof(struct ext4_dir_entry_tail), blocksize); 285 t->det_reserved_ft = EXT4_FT_DIR_CSUM; 286 } 287 288 /* Walk through a dirent block to find a checksum "dirent" at the tail */ 289 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode, 290 struct ext4_dir_entry *de) 291 { 292 struct ext4_dir_entry_tail *t; 293 294 #ifdef PARANOID 295 struct ext4_dir_entry *d, *top; 296 297 d = de; 298 top = (struct ext4_dir_entry *)(((void *)de) + 299 (EXT4_BLOCK_SIZE(inode->i_sb) - 300 sizeof(struct ext4_dir_entry_tail))); 301 while (d < top && d->rec_len) 302 d = (struct ext4_dir_entry *)(((void *)d) + 303 le16_to_cpu(d->rec_len)); 304 305 if (d != top) 306 return NULL; 307 308 t = (struct ext4_dir_entry_tail *)d; 309 #else 310 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb)); 311 #endif 312 313 if (t->det_reserved_zero1 || 314 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) || 315 t->det_reserved_zero2 || 316 t->det_reserved_ft != EXT4_FT_DIR_CSUM) 317 return NULL; 318 319 return t; 320 } 321 322 static __le32 ext4_dirent_csum(struct inode *inode, 323 struct ext4_dir_entry *dirent, int size) 324 { 325 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 326 struct ext4_inode_info *ei = EXT4_I(inode); 327 __u32 csum; 328 329 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size); 330 return cpu_to_le32(csum); 331 } 332 333 #define warn_no_space_for_csum(inode) \ 334 __warn_no_space_for_csum((inode), __func__, __LINE__) 335 336 static void __warn_no_space_for_csum(struct inode *inode, const char *func, 337 unsigned int line) 338 { 339 __ext4_warning_inode(inode, func, line, 340 "No space for directory leaf checksum. Please run e2fsck -D."); 341 } 342 343 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent) 344 { 345 struct ext4_dir_entry_tail *t; 346 347 if (!ext4_has_metadata_csum(inode->i_sb)) 348 return 1; 349 350 t = get_dirent_tail(inode, dirent); 351 if (!t) { 352 warn_no_space_for_csum(inode); 353 return 0; 354 } 355 356 if (t->det_checksum != ext4_dirent_csum(inode, dirent, 357 (void *)t - (void *)dirent)) 358 return 0; 359 360 return 1; 361 } 362 363 static void ext4_dirent_csum_set(struct inode *inode, 364 struct ext4_dir_entry *dirent) 365 { 366 struct ext4_dir_entry_tail *t; 367 368 if (!ext4_has_metadata_csum(inode->i_sb)) 369 return; 370 371 t = get_dirent_tail(inode, dirent); 372 if (!t) { 373 warn_no_space_for_csum(inode); 374 return; 375 } 376 377 t->det_checksum = ext4_dirent_csum(inode, dirent, 378 (void *)t - (void *)dirent); 379 } 380 381 int ext4_handle_dirty_dirent_node(handle_t *handle, 382 struct inode *inode, 383 struct buffer_head *bh) 384 { 385 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data); 386 return ext4_handle_dirty_metadata(handle, inode, bh); 387 } 388 389 static struct dx_countlimit *get_dx_countlimit(struct inode *inode, 390 struct ext4_dir_entry *dirent, 391 int *offset) 392 { 393 struct ext4_dir_entry *dp; 394 struct dx_root_info *root; 395 int count_offset; 396 397 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb)) 398 count_offset = 8; 399 else if (le16_to_cpu(dirent->rec_len) == 12) { 400 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12); 401 if (le16_to_cpu(dp->rec_len) != 402 EXT4_BLOCK_SIZE(inode->i_sb) - 12) 403 return NULL; 404 root = (struct dx_root_info *)(((void *)dp + 12)); 405 if (root->reserved_zero || 406 root->info_length != sizeof(struct dx_root_info)) 407 return NULL; 408 count_offset = 32; 409 } else 410 return NULL; 411 412 if (offset) 413 *offset = count_offset; 414 return (struct dx_countlimit *)(((void *)dirent) + count_offset); 415 } 416 417 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent, 418 int count_offset, int count, struct dx_tail *t) 419 { 420 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 421 struct ext4_inode_info *ei = EXT4_I(inode); 422 __u32 csum; 423 __le32 save_csum; 424 int size; 425 426 size = count_offset + (count * sizeof(struct dx_entry)); 427 save_csum = t->dt_checksum; 428 t->dt_checksum = 0; 429 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size); 430 csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail)); 431 t->dt_checksum = save_csum; 432 433 return cpu_to_le32(csum); 434 } 435 436 static int ext4_dx_csum_verify(struct inode *inode, 437 struct ext4_dir_entry *dirent) 438 { 439 struct dx_countlimit *c; 440 struct dx_tail *t; 441 int count_offset, limit, count; 442 443 if (!ext4_has_metadata_csum(inode->i_sb)) 444 return 1; 445 446 c = get_dx_countlimit(inode, dirent, &count_offset); 447 if (!c) { 448 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D."); 449 return 1; 450 } 451 limit = le16_to_cpu(c->limit); 452 count = le16_to_cpu(c->count); 453 if (count_offset + (limit * sizeof(struct dx_entry)) > 454 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) { 455 warn_no_space_for_csum(inode); 456 return 1; 457 } 458 t = (struct dx_tail *)(((struct dx_entry *)c) + limit); 459 460 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset, 461 count, t)) 462 return 0; 463 return 1; 464 } 465 466 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent) 467 { 468 struct dx_countlimit *c; 469 struct dx_tail *t; 470 int count_offset, limit, count; 471 472 if (!ext4_has_metadata_csum(inode->i_sb)) 473 return; 474 475 c = get_dx_countlimit(inode, dirent, &count_offset); 476 if (!c) { 477 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D."); 478 return; 479 } 480 limit = le16_to_cpu(c->limit); 481 count = le16_to_cpu(c->count); 482 if (count_offset + (limit * sizeof(struct dx_entry)) > 483 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) { 484 warn_no_space_for_csum(inode); 485 return; 486 } 487 t = (struct dx_tail *)(((struct dx_entry *)c) + limit); 488 489 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t); 490 } 491 492 static inline int ext4_handle_dirty_dx_node(handle_t *handle, 493 struct inode *inode, 494 struct buffer_head *bh) 495 { 496 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data); 497 return ext4_handle_dirty_metadata(handle, inode, bh); 498 } 499 500 /* 501 * p is at least 6 bytes before the end of page 502 */ 503 static inline struct ext4_dir_entry_2 * 504 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize) 505 { 506 return (struct ext4_dir_entry_2 *)((char *)p + 507 ext4_rec_len_from_disk(p->rec_len, blocksize)); 508 } 509 510 /* 511 * Future: use high four bits of block for coalesce-on-delete flags 512 * Mask them off for now. 513 */ 514 515 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry) 516 { 517 return le32_to_cpu(entry->block) & 0x00ffffff; 518 } 519 520 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value) 521 { 522 entry->block = cpu_to_le32(value); 523 } 524 525 static inline unsigned dx_get_hash(struct dx_entry *entry) 526 { 527 return le32_to_cpu(entry->hash); 528 } 529 530 static inline void dx_set_hash(struct dx_entry *entry, unsigned value) 531 { 532 entry->hash = cpu_to_le32(value); 533 } 534 535 static inline unsigned dx_get_count(struct dx_entry *entries) 536 { 537 return le16_to_cpu(((struct dx_countlimit *) entries)->count); 538 } 539 540 static inline unsigned dx_get_limit(struct dx_entry *entries) 541 { 542 return le16_to_cpu(((struct dx_countlimit *) entries)->limit); 543 } 544 545 static inline void dx_set_count(struct dx_entry *entries, unsigned value) 546 { 547 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value); 548 } 549 550 static inline void dx_set_limit(struct dx_entry *entries, unsigned value) 551 { 552 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value); 553 } 554 555 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize) 556 { 557 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) - 558 EXT4_DIR_REC_LEN(2) - infosize; 559 560 if (ext4_has_metadata_csum(dir->i_sb)) 561 entry_space -= sizeof(struct dx_tail); 562 return entry_space / sizeof(struct dx_entry); 563 } 564 565 static inline unsigned dx_node_limit(struct inode *dir) 566 { 567 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0); 568 569 if (ext4_has_metadata_csum(dir->i_sb)) 570 entry_space -= sizeof(struct dx_tail); 571 return entry_space / sizeof(struct dx_entry); 572 } 573 574 /* 575 * Debug 576 */ 577 #ifdef DX_DEBUG 578 static void dx_show_index(char * label, struct dx_entry *entries) 579 { 580 int i, n = dx_get_count (entries); 581 printk(KERN_DEBUG "%s index ", label); 582 for (i = 0; i < n; i++) { 583 printk("%x->%lu ", i ? dx_get_hash(entries + i) : 584 0, (unsigned long)dx_get_block(entries + i)); 585 } 586 printk("\n"); 587 } 588 589 struct stats 590 { 591 unsigned names; 592 unsigned space; 593 unsigned bcount; 594 }; 595 596 static struct stats dx_show_leaf(struct inode *dir, 597 struct dx_hash_info *hinfo, 598 struct ext4_dir_entry_2 *de, 599 int size, int show_names) 600 { 601 unsigned names = 0, space = 0; 602 char *base = (char *) de; 603 struct dx_hash_info h = *hinfo; 604 605 printk("names: "); 606 while ((char *) de < base + size) 607 { 608 if (de->inode) 609 { 610 if (show_names) 611 { 612 #ifdef CONFIG_EXT4_FS_ENCRYPTION 613 int len; 614 char *name; 615 struct ext4_str fname_crypto_str 616 = {.name = NULL, .len = 0}; 617 int res = 0; 618 619 name = de->name; 620 len = de->name_len; 621 if (ext4_encrypted_inode(inode)) 622 res = ext4_get_encryption_info(dir); 623 if (res) { 624 printk(KERN_WARNING "Error setting up" 625 " fname crypto: %d\n", res); 626 } 627 if (ctx == NULL) { 628 /* Directory is not encrypted */ 629 ext4fs_dirhash(de->name, 630 de->name_len, &h); 631 printk("%*.s:(U)%x.%u ", len, 632 name, h.hash, 633 (unsigned) ((char *) de 634 - base)); 635 } else { 636 /* Directory is encrypted */ 637 res = ext4_fname_crypto_alloc_buffer( 638 ctx, de->name_len, 639 &fname_crypto_str); 640 if (res < 0) { 641 printk(KERN_WARNING "Error " 642 "allocating crypto " 643 "buffer--skipping " 644 "crypto\n"); 645 ctx = NULL; 646 } 647 res = ext4_fname_disk_to_usr(ctx, NULL, de, 648 &fname_crypto_str); 649 if (res < 0) { 650 printk(KERN_WARNING "Error " 651 "converting filename " 652 "from disk to usr" 653 "\n"); 654 name = "??"; 655 len = 2; 656 } else { 657 name = fname_crypto_str.name; 658 len = fname_crypto_str.len; 659 } 660 ext4fs_dirhash(de->name, de->name_len, 661 &h); 662 printk("%*.s:(E)%x.%u ", len, name, 663 h.hash, (unsigned) ((char *) de 664 - base)); 665 ext4_fname_crypto_free_buffer( 666 &fname_crypto_str); 667 } 668 #else 669 int len = de->name_len; 670 char *name = de->name; 671 ext4fs_dirhash(de->name, de->name_len, &h); 672 printk("%*.s:%x.%u ", len, name, h.hash, 673 (unsigned) ((char *) de - base)); 674 #endif 675 } 676 space += EXT4_DIR_REC_LEN(de->name_len); 677 names++; 678 } 679 de = ext4_next_entry(de, size); 680 } 681 printk("(%i)\n", names); 682 return (struct stats) { names, space, 1 }; 683 } 684 685 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir, 686 struct dx_entry *entries, int levels) 687 { 688 unsigned blocksize = dir->i_sb->s_blocksize; 689 unsigned count = dx_get_count(entries), names = 0, space = 0, i; 690 unsigned bcount = 0; 691 struct buffer_head *bh; 692 printk("%i indexed blocks...\n", count); 693 for (i = 0; i < count; i++, entries++) 694 { 695 ext4_lblk_t block = dx_get_block(entries); 696 ext4_lblk_t hash = i ? dx_get_hash(entries): 0; 697 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash; 698 struct stats stats; 699 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range); 700 bh = ext4_bread(NULL,dir, block, 0); 701 if (!bh || IS_ERR(bh)) 702 continue; 703 stats = levels? 704 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1): 705 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) 706 bh->b_data, blocksize, 0); 707 names += stats.names; 708 space += stats.space; 709 bcount += stats.bcount; 710 brelse(bh); 711 } 712 if (bcount) 713 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n", 714 levels ? "" : " ", names, space/bcount, 715 (space/bcount)*100/blocksize); 716 return (struct stats) { names, space, bcount}; 717 } 718 #endif /* DX_DEBUG */ 719 720 /* 721 * Probe for a directory leaf block to search. 722 * 723 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format 724 * error in the directory index, and the caller should fall back to 725 * searching the directory normally. The callers of dx_probe **MUST** 726 * check for this error code, and make sure it never gets reflected 727 * back to userspace. 728 */ 729 static struct dx_frame * 730 dx_probe(struct ext4_filename *fname, struct inode *dir, 731 struct dx_hash_info *hinfo, struct dx_frame *frame_in) 732 { 733 unsigned count, indirect; 734 struct dx_entry *at, *entries, *p, *q, *m; 735 struct dx_root *root; 736 struct dx_frame *frame = frame_in; 737 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR); 738 u32 hash; 739 740 frame->bh = ext4_read_dirblock(dir, 0, INDEX); 741 if (IS_ERR(frame->bh)) 742 return (struct dx_frame *) frame->bh; 743 744 root = (struct dx_root *) frame->bh->b_data; 745 if (root->info.hash_version != DX_HASH_TEA && 746 root->info.hash_version != DX_HASH_HALF_MD4 && 747 root->info.hash_version != DX_HASH_LEGACY) { 748 ext4_warning_inode(dir, "Unrecognised inode hash code %u", 749 root->info.hash_version); 750 goto fail; 751 } 752 if (fname) 753 hinfo = &fname->hinfo; 754 hinfo->hash_version = root->info.hash_version; 755 if (hinfo->hash_version <= DX_HASH_TEA) 756 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; 757 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed; 758 if (fname && fname_name(fname)) 759 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo); 760 hash = hinfo->hash; 761 762 if (root->info.unused_flags & 1) { 763 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x", 764 root->info.unused_flags); 765 goto fail; 766 } 767 768 indirect = root->info.indirect_levels; 769 if (indirect > 1) { 770 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x", 771 root->info.indirect_levels); 772 goto fail; 773 } 774 775 entries = (struct dx_entry *)(((char *)&root->info) + 776 root->info.info_length); 777 778 if (dx_get_limit(entries) != dx_root_limit(dir, 779 root->info.info_length)) { 780 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u", 781 dx_get_limit(entries), 782 dx_root_limit(dir, root->info.info_length)); 783 goto fail; 784 } 785 786 dxtrace(printk("Look up %x", hash)); 787 while (1) { 788 count = dx_get_count(entries); 789 if (!count || count > dx_get_limit(entries)) { 790 ext4_warning_inode(dir, 791 "dx entry: count %u beyond limit %u", 792 count, dx_get_limit(entries)); 793 goto fail; 794 } 795 796 p = entries + 1; 797 q = entries + count - 1; 798 while (p <= q) { 799 m = p + (q - p) / 2; 800 dxtrace(printk(".")); 801 if (dx_get_hash(m) > hash) 802 q = m - 1; 803 else 804 p = m + 1; 805 } 806 807 if (0) { // linear search cross check 808 unsigned n = count - 1; 809 at = entries; 810 while (n--) 811 { 812 dxtrace(printk(",")); 813 if (dx_get_hash(++at) > hash) 814 { 815 at--; 816 break; 817 } 818 } 819 assert (at == p - 1); 820 } 821 822 at = p - 1; 823 dxtrace(printk(" %x->%u\n", at == entries ? 0 : dx_get_hash(at), 824 dx_get_block(at))); 825 frame->entries = entries; 826 frame->at = at; 827 if (!indirect--) 828 return frame; 829 frame++; 830 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX); 831 if (IS_ERR(frame->bh)) { 832 ret_err = (struct dx_frame *) frame->bh; 833 frame->bh = NULL; 834 goto fail; 835 } 836 entries = ((struct dx_node *) frame->bh->b_data)->entries; 837 838 if (dx_get_limit(entries) != dx_node_limit(dir)) { 839 ext4_warning_inode(dir, 840 "dx entry: limit %u != node limit %u", 841 dx_get_limit(entries), dx_node_limit(dir)); 842 goto fail; 843 } 844 } 845 fail: 846 while (frame >= frame_in) { 847 brelse(frame->bh); 848 frame--; 849 } 850 851 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR)) 852 ext4_warning_inode(dir, 853 "Corrupt directory, running e2fsck is recommended"); 854 return ret_err; 855 } 856 857 static void dx_release(struct dx_frame *frames) 858 { 859 if (frames[0].bh == NULL) 860 return; 861 862 if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels) 863 brelse(frames[1].bh); 864 brelse(frames[0].bh); 865 } 866 867 /* 868 * This function increments the frame pointer to search the next leaf 869 * block, and reads in the necessary intervening nodes if the search 870 * should be necessary. Whether or not the search is necessary is 871 * controlled by the hash parameter. If the hash value is even, then 872 * the search is only continued if the next block starts with that 873 * hash value. This is used if we are searching for a specific file. 874 * 875 * If the hash value is HASH_NB_ALWAYS, then always go to the next block. 876 * 877 * This function returns 1 if the caller should continue to search, 878 * or 0 if it should not. If there is an error reading one of the 879 * index blocks, it will a negative error code. 880 * 881 * If start_hash is non-null, it will be filled in with the starting 882 * hash of the next page. 883 */ 884 static int ext4_htree_next_block(struct inode *dir, __u32 hash, 885 struct dx_frame *frame, 886 struct dx_frame *frames, 887 __u32 *start_hash) 888 { 889 struct dx_frame *p; 890 struct buffer_head *bh; 891 int num_frames = 0; 892 __u32 bhash; 893 894 p = frame; 895 /* 896 * Find the next leaf page by incrementing the frame pointer. 897 * If we run out of entries in the interior node, loop around and 898 * increment pointer in the parent node. When we break out of 899 * this loop, num_frames indicates the number of interior 900 * nodes need to be read. 901 */ 902 while (1) { 903 if (++(p->at) < p->entries + dx_get_count(p->entries)) 904 break; 905 if (p == frames) 906 return 0; 907 num_frames++; 908 p--; 909 } 910 911 /* 912 * If the hash is 1, then continue only if the next page has a 913 * continuation hash of any value. This is used for readdir 914 * handling. Otherwise, check to see if the hash matches the 915 * desired contiuation hash. If it doesn't, return since 916 * there's no point to read in the successive index pages. 917 */ 918 bhash = dx_get_hash(p->at); 919 if (start_hash) 920 *start_hash = bhash; 921 if ((hash & 1) == 0) { 922 if ((bhash & ~1) != hash) 923 return 0; 924 } 925 /* 926 * If the hash is HASH_NB_ALWAYS, we always go to the next 927 * block so no check is necessary 928 */ 929 while (num_frames--) { 930 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX); 931 if (IS_ERR(bh)) 932 return PTR_ERR(bh); 933 p++; 934 brelse(p->bh); 935 p->bh = bh; 936 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries; 937 } 938 return 1; 939 } 940 941 942 /* 943 * This function fills a red-black tree with information from a 944 * directory block. It returns the number directory entries loaded 945 * into the tree. If there is an error it is returned in err. 946 */ 947 static int htree_dirblock_to_tree(struct file *dir_file, 948 struct inode *dir, ext4_lblk_t block, 949 struct dx_hash_info *hinfo, 950 __u32 start_hash, __u32 start_minor_hash) 951 { 952 struct buffer_head *bh; 953 struct ext4_dir_entry_2 *de, *top; 954 int err = 0, count = 0; 955 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}, tmp_str; 956 957 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n", 958 (unsigned long)block)); 959 bh = ext4_read_dirblock(dir, block, DIRENT); 960 if (IS_ERR(bh)) 961 return PTR_ERR(bh); 962 963 de = (struct ext4_dir_entry_2 *) bh->b_data; 964 top = (struct ext4_dir_entry_2 *) ((char *) de + 965 dir->i_sb->s_blocksize - 966 EXT4_DIR_REC_LEN(0)); 967 #ifdef CONFIG_EXT4_FS_ENCRYPTION 968 /* Check if the directory is encrypted */ 969 if (ext4_encrypted_inode(dir)) { 970 err = ext4_get_encryption_info(dir); 971 if (err < 0) { 972 brelse(bh); 973 return err; 974 } 975 err = ext4_fname_crypto_alloc_buffer(dir, EXT4_NAME_LEN, 976 &fname_crypto_str); 977 if (err < 0) { 978 brelse(bh); 979 return err; 980 } 981 } 982 #endif 983 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) { 984 if (ext4_check_dir_entry(dir, NULL, de, bh, 985 bh->b_data, bh->b_size, 986 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb)) 987 + ((char *)de - bh->b_data))) { 988 /* silently ignore the rest of the block */ 989 break; 990 } 991 ext4fs_dirhash(de->name, de->name_len, hinfo); 992 if ((hinfo->hash < start_hash) || 993 ((hinfo->hash == start_hash) && 994 (hinfo->minor_hash < start_minor_hash))) 995 continue; 996 if (de->inode == 0) 997 continue; 998 if (!ext4_encrypted_inode(dir)) { 999 tmp_str.name = de->name; 1000 tmp_str.len = de->name_len; 1001 err = ext4_htree_store_dirent(dir_file, 1002 hinfo->hash, hinfo->minor_hash, de, 1003 &tmp_str); 1004 } else { 1005 int save_len = fname_crypto_str.len; 1006 1007 /* Directory is encrypted */ 1008 err = ext4_fname_disk_to_usr(dir, hinfo, de, 1009 &fname_crypto_str); 1010 if (err < 0) { 1011 count = err; 1012 goto errout; 1013 } 1014 err = ext4_htree_store_dirent(dir_file, 1015 hinfo->hash, hinfo->minor_hash, de, 1016 &fname_crypto_str); 1017 fname_crypto_str.len = save_len; 1018 } 1019 if (err != 0) { 1020 count = err; 1021 goto errout; 1022 } 1023 count++; 1024 } 1025 errout: 1026 brelse(bh); 1027 #ifdef CONFIG_EXT4_FS_ENCRYPTION 1028 ext4_fname_crypto_free_buffer(&fname_crypto_str); 1029 #endif 1030 return count; 1031 } 1032 1033 1034 /* 1035 * This function fills a red-black tree with information from a 1036 * directory. We start scanning the directory in hash order, starting 1037 * at start_hash and start_minor_hash. 1038 * 1039 * This function returns the number of entries inserted into the tree, 1040 * or a negative error code. 1041 */ 1042 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash, 1043 __u32 start_minor_hash, __u32 *next_hash) 1044 { 1045 struct dx_hash_info hinfo; 1046 struct ext4_dir_entry_2 *de; 1047 struct dx_frame frames[2], *frame; 1048 struct inode *dir; 1049 ext4_lblk_t block; 1050 int count = 0; 1051 int ret, err; 1052 __u32 hashval; 1053 struct ext4_str tmp_str; 1054 1055 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n", 1056 start_hash, start_minor_hash)); 1057 dir = file_inode(dir_file); 1058 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) { 1059 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; 1060 if (hinfo.hash_version <= DX_HASH_TEA) 1061 hinfo.hash_version += 1062 EXT4_SB(dir->i_sb)->s_hash_unsigned; 1063 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; 1064 if (ext4_has_inline_data(dir)) { 1065 int has_inline_data = 1; 1066 count = htree_inlinedir_to_tree(dir_file, dir, 0, 1067 &hinfo, start_hash, 1068 start_minor_hash, 1069 &has_inline_data); 1070 if (has_inline_data) { 1071 *next_hash = ~0; 1072 return count; 1073 } 1074 } 1075 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo, 1076 start_hash, start_minor_hash); 1077 *next_hash = ~0; 1078 return count; 1079 } 1080 hinfo.hash = start_hash; 1081 hinfo.minor_hash = 0; 1082 frame = dx_probe(NULL, dir, &hinfo, frames); 1083 if (IS_ERR(frame)) 1084 return PTR_ERR(frame); 1085 1086 /* Add '.' and '..' from the htree header */ 1087 if (!start_hash && !start_minor_hash) { 1088 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; 1089 tmp_str.name = de->name; 1090 tmp_str.len = de->name_len; 1091 err = ext4_htree_store_dirent(dir_file, 0, 0, 1092 de, &tmp_str); 1093 if (err != 0) 1094 goto errout; 1095 count++; 1096 } 1097 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) { 1098 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; 1099 de = ext4_next_entry(de, dir->i_sb->s_blocksize); 1100 tmp_str.name = de->name; 1101 tmp_str.len = de->name_len; 1102 err = ext4_htree_store_dirent(dir_file, 2, 0, 1103 de, &tmp_str); 1104 if (err != 0) 1105 goto errout; 1106 count++; 1107 } 1108 1109 while (1) { 1110 block = dx_get_block(frame->at); 1111 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo, 1112 start_hash, start_minor_hash); 1113 if (ret < 0) { 1114 err = ret; 1115 goto errout; 1116 } 1117 count += ret; 1118 hashval = ~0; 1119 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS, 1120 frame, frames, &hashval); 1121 *next_hash = hashval; 1122 if (ret < 0) { 1123 err = ret; 1124 goto errout; 1125 } 1126 /* 1127 * Stop if: (a) there are no more entries, or 1128 * (b) we have inserted at least one entry and the 1129 * next hash value is not a continuation 1130 */ 1131 if ((ret == 0) || 1132 (count && ((hashval & 1) == 0))) 1133 break; 1134 } 1135 dx_release(frames); 1136 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, " 1137 "next hash: %x\n", count, *next_hash)); 1138 return count; 1139 errout: 1140 dx_release(frames); 1141 return (err); 1142 } 1143 1144 static inline int search_dirblock(struct buffer_head *bh, 1145 struct inode *dir, 1146 struct ext4_filename *fname, 1147 const struct qstr *d_name, 1148 unsigned int offset, 1149 struct ext4_dir_entry_2 **res_dir) 1150 { 1151 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir, 1152 fname, d_name, offset, res_dir); 1153 } 1154 1155 /* 1156 * Directory block splitting, compacting 1157 */ 1158 1159 /* 1160 * Create map of hash values, offsets, and sizes, stored at end of block. 1161 * Returns number of entries mapped. 1162 */ 1163 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de, 1164 unsigned blocksize, struct dx_hash_info *hinfo, 1165 struct dx_map_entry *map_tail) 1166 { 1167 int count = 0; 1168 char *base = (char *) de; 1169 struct dx_hash_info h = *hinfo; 1170 1171 while ((char *) de < base + blocksize) { 1172 if (de->name_len && de->inode) { 1173 ext4fs_dirhash(de->name, de->name_len, &h); 1174 map_tail--; 1175 map_tail->hash = h.hash; 1176 map_tail->offs = ((char *) de - base)>>2; 1177 map_tail->size = le16_to_cpu(de->rec_len); 1178 count++; 1179 cond_resched(); 1180 } 1181 /* XXX: do we need to check rec_len == 0 case? -Chris */ 1182 de = ext4_next_entry(de, blocksize); 1183 } 1184 return count; 1185 } 1186 1187 /* Sort map by hash value */ 1188 static void dx_sort_map (struct dx_map_entry *map, unsigned count) 1189 { 1190 struct dx_map_entry *p, *q, *top = map + count - 1; 1191 int more; 1192 /* Combsort until bubble sort doesn't suck */ 1193 while (count > 2) { 1194 count = count*10/13; 1195 if (count - 9 < 2) /* 9, 10 -> 11 */ 1196 count = 11; 1197 for (p = top, q = p - count; q >= map; p--, q--) 1198 if (p->hash < q->hash) 1199 swap(*p, *q); 1200 } 1201 /* Garden variety bubble sort */ 1202 do { 1203 more = 0; 1204 q = top; 1205 while (q-- > map) { 1206 if (q[1].hash >= q[0].hash) 1207 continue; 1208 swap(*(q+1), *q); 1209 more = 1; 1210 } 1211 } while(more); 1212 } 1213 1214 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block) 1215 { 1216 struct dx_entry *entries = frame->entries; 1217 struct dx_entry *old = frame->at, *new = old + 1; 1218 int count = dx_get_count(entries); 1219 1220 assert(count < dx_get_limit(entries)); 1221 assert(old < entries + count); 1222 memmove(new + 1, new, (char *)(entries + count) - (char *)(new)); 1223 dx_set_hash(new, hash); 1224 dx_set_block(new, block); 1225 dx_set_count(entries, count + 1); 1226 } 1227 1228 /* 1229 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure. 1230 * 1231 * `len <= EXT4_NAME_LEN' is guaranteed by caller. 1232 * `de != NULL' is guaranteed by caller. 1233 */ 1234 static inline int ext4_match(struct ext4_filename *fname, 1235 struct ext4_dir_entry_2 *de) 1236 { 1237 const void *name = fname_name(fname); 1238 u32 len = fname_len(fname); 1239 1240 if (!de->inode) 1241 return 0; 1242 1243 #ifdef CONFIG_EXT4_FS_ENCRYPTION 1244 if (unlikely(!name)) { 1245 if (fname->usr_fname->name[0] == '_') { 1246 int ret; 1247 if (de->name_len < 16) 1248 return 0; 1249 ret = memcmp(de->name + de->name_len - 16, 1250 fname->crypto_buf.name + 8, 16); 1251 return (ret == 0) ? 1 : 0; 1252 } 1253 name = fname->crypto_buf.name; 1254 len = fname->crypto_buf.len; 1255 } 1256 #endif 1257 if (de->name_len != len) 1258 return 0; 1259 return (memcmp(de->name, name, len) == 0) ? 1 : 0; 1260 } 1261 1262 /* 1263 * Returns 0 if not found, -1 on failure, and 1 on success 1264 */ 1265 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size, 1266 struct inode *dir, struct ext4_filename *fname, 1267 const struct qstr *d_name, 1268 unsigned int offset, struct ext4_dir_entry_2 **res_dir) 1269 { 1270 struct ext4_dir_entry_2 * de; 1271 char * dlimit; 1272 int de_len; 1273 int res; 1274 1275 de = (struct ext4_dir_entry_2 *)search_buf; 1276 dlimit = search_buf + buf_size; 1277 while ((char *) de < dlimit) { 1278 /* this code is executed quadratically often */ 1279 /* do minimal checking `by hand' */ 1280 if ((char *) de + de->name_len <= dlimit) { 1281 res = ext4_match(fname, de); 1282 if (res < 0) { 1283 res = -1; 1284 goto return_result; 1285 } 1286 if (res > 0) { 1287 /* found a match - just to be sure, do 1288 * a full check */ 1289 if (ext4_check_dir_entry(dir, NULL, de, bh, 1290 bh->b_data, 1291 bh->b_size, offset)) { 1292 res = -1; 1293 goto return_result; 1294 } 1295 *res_dir = de; 1296 res = 1; 1297 goto return_result; 1298 } 1299 1300 } 1301 /* prevent looping on a bad block */ 1302 de_len = ext4_rec_len_from_disk(de->rec_len, 1303 dir->i_sb->s_blocksize); 1304 if (de_len <= 0) { 1305 res = -1; 1306 goto return_result; 1307 } 1308 offset += de_len; 1309 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len); 1310 } 1311 1312 res = 0; 1313 return_result: 1314 return res; 1315 } 1316 1317 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block, 1318 struct ext4_dir_entry *de) 1319 { 1320 struct super_block *sb = dir->i_sb; 1321 1322 if (!is_dx(dir)) 1323 return 0; 1324 if (block == 0) 1325 return 1; 1326 if (de->inode == 0 && 1327 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) == 1328 sb->s_blocksize) 1329 return 1; 1330 return 0; 1331 } 1332 1333 /* 1334 * ext4_find_entry() 1335 * 1336 * finds an entry in the specified directory with the wanted name. It 1337 * returns the cache buffer in which the entry was found, and the entry 1338 * itself (as a parameter - res_dir). It does NOT read the inode of the 1339 * entry - you'll have to do that yourself if you want to. 1340 * 1341 * The returned buffer_head has ->b_count elevated. The caller is expected 1342 * to brelse() it when appropriate. 1343 */ 1344 static struct buffer_head * ext4_find_entry (struct inode *dir, 1345 const struct qstr *d_name, 1346 struct ext4_dir_entry_2 **res_dir, 1347 int *inlined) 1348 { 1349 struct super_block *sb; 1350 struct buffer_head *bh_use[NAMEI_RA_SIZE]; 1351 struct buffer_head *bh, *ret = NULL; 1352 ext4_lblk_t start, block, b; 1353 const u8 *name = d_name->name; 1354 int ra_max = 0; /* Number of bh's in the readahead 1355 buffer, bh_use[] */ 1356 int ra_ptr = 0; /* Current index into readahead 1357 buffer */ 1358 int num = 0; 1359 ext4_lblk_t nblocks; 1360 int i, namelen, retval; 1361 struct ext4_filename fname; 1362 1363 *res_dir = NULL; 1364 sb = dir->i_sb; 1365 namelen = d_name->len; 1366 if (namelen > EXT4_NAME_LEN) 1367 return NULL; 1368 1369 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname); 1370 if (retval) 1371 return ERR_PTR(retval); 1372 1373 if (ext4_has_inline_data(dir)) { 1374 int has_inline_data = 1; 1375 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir, 1376 &has_inline_data); 1377 if (has_inline_data) { 1378 if (inlined) 1379 *inlined = 1; 1380 goto cleanup_and_exit; 1381 } 1382 } 1383 1384 if ((namelen <= 2) && (name[0] == '.') && 1385 (name[1] == '.' || name[1] == '\0')) { 1386 /* 1387 * "." or ".." will only be in the first block 1388 * NFS may look up ".."; "." should be handled by the VFS 1389 */ 1390 block = start = 0; 1391 nblocks = 1; 1392 goto restart; 1393 } 1394 if (is_dx(dir)) { 1395 ret = ext4_dx_find_entry(dir, &fname, res_dir); 1396 /* 1397 * On success, or if the error was file not found, 1398 * return. Otherwise, fall back to doing a search the 1399 * old fashioned way. 1400 */ 1401 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR) 1402 goto cleanup_and_exit; 1403 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, " 1404 "falling back\n")); 1405 } 1406 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); 1407 start = EXT4_I(dir)->i_dir_start_lookup; 1408 if (start >= nblocks) 1409 start = 0; 1410 block = start; 1411 restart: 1412 do { 1413 /* 1414 * We deal with the read-ahead logic here. 1415 */ 1416 if (ra_ptr >= ra_max) { 1417 /* Refill the readahead buffer */ 1418 ra_ptr = 0; 1419 b = block; 1420 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) { 1421 /* 1422 * Terminate if we reach the end of the 1423 * directory and must wrap, or if our 1424 * search has finished at this block. 1425 */ 1426 if (b >= nblocks || (num && block == start)) { 1427 bh_use[ra_max] = NULL; 1428 break; 1429 } 1430 num++; 1431 bh = ext4_getblk(NULL, dir, b++, 0); 1432 if (IS_ERR(bh)) { 1433 if (ra_max == 0) { 1434 ret = bh; 1435 goto cleanup_and_exit; 1436 } 1437 break; 1438 } 1439 bh_use[ra_max] = bh; 1440 if (bh) 1441 ll_rw_block(READ | REQ_META | REQ_PRIO, 1442 1, &bh); 1443 } 1444 } 1445 if ((bh = bh_use[ra_ptr++]) == NULL) 1446 goto next; 1447 wait_on_buffer(bh); 1448 if (!buffer_uptodate(bh)) { 1449 /* read error, skip block & hope for the best */ 1450 EXT4_ERROR_INODE(dir, "reading directory lblock %lu", 1451 (unsigned long) block); 1452 brelse(bh); 1453 goto next; 1454 } 1455 if (!buffer_verified(bh) && 1456 !is_dx_internal_node(dir, block, 1457 (struct ext4_dir_entry *)bh->b_data) && 1458 !ext4_dirent_csum_verify(dir, 1459 (struct ext4_dir_entry *)bh->b_data)) { 1460 EXT4_ERROR_INODE(dir, "checksumming directory " 1461 "block %lu", (unsigned long)block); 1462 brelse(bh); 1463 goto next; 1464 } 1465 set_buffer_verified(bh); 1466 i = search_dirblock(bh, dir, &fname, d_name, 1467 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir); 1468 if (i == 1) { 1469 EXT4_I(dir)->i_dir_start_lookup = block; 1470 ret = bh; 1471 goto cleanup_and_exit; 1472 } else { 1473 brelse(bh); 1474 if (i < 0) 1475 goto cleanup_and_exit; 1476 } 1477 next: 1478 if (++block >= nblocks) 1479 block = 0; 1480 } while (block != start); 1481 1482 /* 1483 * If the directory has grown while we were searching, then 1484 * search the last part of the directory before giving up. 1485 */ 1486 block = nblocks; 1487 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); 1488 if (block < nblocks) { 1489 start = 0; 1490 goto restart; 1491 } 1492 1493 cleanup_and_exit: 1494 /* Clean up the read-ahead blocks */ 1495 for (; ra_ptr < ra_max; ra_ptr++) 1496 brelse(bh_use[ra_ptr]); 1497 ext4_fname_free_filename(&fname); 1498 return ret; 1499 } 1500 1501 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, 1502 struct ext4_filename *fname, 1503 struct ext4_dir_entry_2 **res_dir) 1504 { 1505 struct super_block * sb = dir->i_sb; 1506 struct dx_frame frames[2], *frame; 1507 const struct qstr *d_name = fname->usr_fname; 1508 struct buffer_head *bh; 1509 ext4_lblk_t block; 1510 int retval; 1511 1512 #ifdef CONFIG_EXT4_FS_ENCRYPTION 1513 *res_dir = NULL; 1514 #endif 1515 frame = dx_probe(fname, dir, NULL, frames); 1516 if (IS_ERR(frame)) 1517 return (struct buffer_head *) frame; 1518 do { 1519 block = dx_get_block(frame->at); 1520 bh = ext4_read_dirblock(dir, block, DIRENT); 1521 if (IS_ERR(bh)) 1522 goto errout; 1523 1524 retval = search_dirblock(bh, dir, fname, d_name, 1525 block << EXT4_BLOCK_SIZE_BITS(sb), 1526 res_dir); 1527 if (retval == 1) 1528 goto success; 1529 brelse(bh); 1530 if (retval == -1) { 1531 bh = ERR_PTR(ERR_BAD_DX_DIR); 1532 goto errout; 1533 } 1534 1535 /* Check to see if we should continue to search */ 1536 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame, 1537 frames, NULL); 1538 if (retval < 0) { 1539 ext4_warning_inode(dir, 1540 "error %d reading directory index block", 1541 retval); 1542 bh = ERR_PTR(retval); 1543 goto errout; 1544 } 1545 } while (retval == 1); 1546 1547 bh = NULL; 1548 errout: 1549 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name)); 1550 success: 1551 dx_release(frames); 1552 return bh; 1553 } 1554 1555 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) 1556 { 1557 struct inode *inode; 1558 struct ext4_dir_entry_2 *de; 1559 struct buffer_head *bh; 1560 1561 if (dentry->d_name.len > EXT4_NAME_LEN) 1562 return ERR_PTR(-ENAMETOOLONG); 1563 1564 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); 1565 if (IS_ERR(bh)) 1566 return (struct dentry *) bh; 1567 inode = NULL; 1568 if (bh) { 1569 __u32 ino = le32_to_cpu(de->inode); 1570 brelse(bh); 1571 if (!ext4_valid_inum(dir->i_sb, ino)) { 1572 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino); 1573 return ERR_PTR(-EFSCORRUPTED); 1574 } 1575 if (unlikely(ino == dir->i_ino)) { 1576 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir", 1577 dentry); 1578 return ERR_PTR(-EFSCORRUPTED); 1579 } 1580 inode = ext4_iget_normal(dir->i_sb, ino); 1581 if (inode == ERR_PTR(-ESTALE)) { 1582 EXT4_ERROR_INODE(dir, 1583 "deleted inode referenced: %u", 1584 ino); 1585 return ERR_PTR(-EFSCORRUPTED); 1586 } 1587 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) && 1588 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1589 S_ISLNK(inode->i_mode)) && 1590 !ext4_is_child_context_consistent_with_parent(dir, 1591 inode)) { 1592 iput(inode); 1593 ext4_warning(inode->i_sb, 1594 "Inconsistent encryption contexts: %lu/%lu\n", 1595 (unsigned long) dir->i_ino, 1596 (unsigned long) inode->i_ino); 1597 return ERR_PTR(-EPERM); 1598 } 1599 } 1600 return d_splice_alias(inode, dentry); 1601 } 1602 1603 1604 struct dentry *ext4_get_parent(struct dentry *child) 1605 { 1606 __u32 ino; 1607 static const struct qstr dotdot = QSTR_INIT("..", 2); 1608 struct ext4_dir_entry_2 * de; 1609 struct buffer_head *bh; 1610 1611 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL); 1612 if (IS_ERR(bh)) 1613 return (struct dentry *) bh; 1614 if (!bh) 1615 return ERR_PTR(-ENOENT); 1616 ino = le32_to_cpu(de->inode); 1617 brelse(bh); 1618 1619 if (!ext4_valid_inum(d_inode(child)->i_sb, ino)) { 1620 EXT4_ERROR_INODE(d_inode(child), 1621 "bad parent inode number: %u", ino); 1622 return ERR_PTR(-EFSCORRUPTED); 1623 } 1624 1625 return d_obtain_alias(ext4_iget_normal(d_inode(child)->i_sb, ino)); 1626 } 1627 1628 /* 1629 * Move count entries from end of map between two memory locations. 1630 * Returns pointer to last entry moved. 1631 */ 1632 static struct ext4_dir_entry_2 * 1633 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count, 1634 unsigned blocksize) 1635 { 1636 unsigned rec_len = 0; 1637 1638 while (count--) { 1639 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) 1640 (from + (map->offs<<2)); 1641 rec_len = EXT4_DIR_REC_LEN(de->name_len); 1642 memcpy (to, de, rec_len); 1643 ((struct ext4_dir_entry_2 *) to)->rec_len = 1644 ext4_rec_len_to_disk(rec_len, blocksize); 1645 de->inode = 0; 1646 map++; 1647 to += rec_len; 1648 } 1649 return (struct ext4_dir_entry_2 *) (to - rec_len); 1650 } 1651 1652 /* 1653 * Compact each dir entry in the range to the minimal rec_len. 1654 * Returns pointer to last entry in range. 1655 */ 1656 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize) 1657 { 1658 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base; 1659 unsigned rec_len = 0; 1660 1661 prev = to = de; 1662 while ((char*)de < base + blocksize) { 1663 next = ext4_next_entry(de, blocksize); 1664 if (de->inode && de->name_len) { 1665 rec_len = EXT4_DIR_REC_LEN(de->name_len); 1666 if (de > to) 1667 memmove(to, de, rec_len); 1668 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize); 1669 prev = to; 1670 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len); 1671 } 1672 de = next; 1673 } 1674 return prev; 1675 } 1676 1677 /* 1678 * Split a full leaf block to make room for a new dir entry. 1679 * Allocate a new block, and move entries so that they are approx. equally full. 1680 * Returns pointer to de in block into which the new entry will be inserted. 1681 */ 1682 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, 1683 struct buffer_head **bh,struct dx_frame *frame, 1684 struct dx_hash_info *hinfo) 1685 { 1686 unsigned blocksize = dir->i_sb->s_blocksize; 1687 unsigned count, continued; 1688 struct buffer_head *bh2; 1689 ext4_lblk_t newblock; 1690 u32 hash2; 1691 struct dx_map_entry *map; 1692 char *data1 = (*bh)->b_data, *data2; 1693 unsigned split, move, size; 1694 struct ext4_dir_entry_2 *de = NULL, *de2; 1695 struct ext4_dir_entry_tail *t; 1696 int csum_size = 0; 1697 int err = 0, i; 1698 1699 if (ext4_has_metadata_csum(dir->i_sb)) 1700 csum_size = sizeof(struct ext4_dir_entry_tail); 1701 1702 bh2 = ext4_append(handle, dir, &newblock); 1703 if (IS_ERR(bh2)) { 1704 brelse(*bh); 1705 *bh = NULL; 1706 return (struct ext4_dir_entry_2 *) bh2; 1707 } 1708 1709 BUFFER_TRACE(*bh, "get_write_access"); 1710 err = ext4_journal_get_write_access(handle, *bh); 1711 if (err) 1712 goto journal_error; 1713 1714 BUFFER_TRACE(frame->bh, "get_write_access"); 1715 err = ext4_journal_get_write_access(handle, frame->bh); 1716 if (err) 1717 goto journal_error; 1718 1719 data2 = bh2->b_data; 1720 1721 /* create map in the end of data2 block */ 1722 map = (struct dx_map_entry *) (data2 + blocksize); 1723 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1, 1724 blocksize, hinfo, map); 1725 map -= count; 1726 dx_sort_map(map, count); 1727 /* Split the existing block in the middle, size-wise */ 1728 size = 0; 1729 move = 0; 1730 for (i = count-1; i >= 0; i--) { 1731 /* is more than half of this entry in 2nd half of the block? */ 1732 if (size + map[i].size/2 > blocksize/2) 1733 break; 1734 size += map[i].size; 1735 move++; 1736 } 1737 /* map index at which we will split */ 1738 split = count - move; 1739 hash2 = map[split].hash; 1740 continued = hash2 == map[split - 1].hash; 1741 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n", 1742 (unsigned long)dx_get_block(frame->at), 1743 hash2, split, count-split)); 1744 1745 /* Fancy dance to stay within two buffers */ 1746 de2 = dx_move_dirents(data1, data2, map + split, count - split, 1747 blocksize); 1748 de = dx_pack_dirents(data1, blocksize); 1749 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) - 1750 (char *) de, 1751 blocksize); 1752 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) - 1753 (char *) de2, 1754 blocksize); 1755 if (csum_size) { 1756 t = EXT4_DIRENT_TAIL(data2, blocksize); 1757 initialize_dirent_tail(t, blocksize); 1758 1759 t = EXT4_DIRENT_TAIL(data1, blocksize); 1760 initialize_dirent_tail(t, blocksize); 1761 } 1762 1763 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1, 1764 blocksize, 1)); 1765 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2, 1766 blocksize, 1)); 1767 1768 /* Which block gets the new entry? */ 1769 if (hinfo->hash >= hash2) { 1770 swap(*bh, bh2); 1771 de = de2; 1772 } 1773 dx_insert_block(frame, hash2 + continued, newblock); 1774 err = ext4_handle_dirty_dirent_node(handle, dir, bh2); 1775 if (err) 1776 goto journal_error; 1777 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh); 1778 if (err) 1779 goto journal_error; 1780 brelse(bh2); 1781 dxtrace(dx_show_index("frame", frame->entries)); 1782 return de; 1783 1784 journal_error: 1785 brelse(*bh); 1786 brelse(bh2); 1787 *bh = NULL; 1788 ext4_std_error(dir->i_sb, err); 1789 return ERR_PTR(err); 1790 } 1791 1792 int ext4_find_dest_de(struct inode *dir, struct inode *inode, 1793 struct buffer_head *bh, 1794 void *buf, int buf_size, 1795 struct ext4_filename *fname, 1796 struct ext4_dir_entry_2 **dest_de) 1797 { 1798 struct ext4_dir_entry_2 *de; 1799 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname)); 1800 int nlen, rlen; 1801 unsigned int offset = 0; 1802 char *top; 1803 int res; 1804 1805 de = (struct ext4_dir_entry_2 *)buf; 1806 top = buf + buf_size - reclen; 1807 while ((char *) de <= top) { 1808 if (ext4_check_dir_entry(dir, NULL, de, bh, 1809 buf, buf_size, offset)) { 1810 res = -EFSCORRUPTED; 1811 goto return_result; 1812 } 1813 /* Provide crypto context and crypto buffer to ext4 match */ 1814 res = ext4_match(fname, de); 1815 if (res < 0) 1816 goto return_result; 1817 if (res > 0) { 1818 res = -EEXIST; 1819 goto return_result; 1820 } 1821 nlen = EXT4_DIR_REC_LEN(de->name_len); 1822 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 1823 if ((de->inode ? rlen - nlen : rlen) >= reclen) 1824 break; 1825 de = (struct ext4_dir_entry_2 *)((char *)de + rlen); 1826 offset += rlen; 1827 } 1828 1829 if ((char *) de > top) 1830 res = -ENOSPC; 1831 else { 1832 *dest_de = de; 1833 res = 0; 1834 } 1835 return_result: 1836 return res; 1837 } 1838 1839 int ext4_insert_dentry(struct inode *dir, 1840 struct inode *inode, 1841 struct ext4_dir_entry_2 *de, 1842 int buf_size, 1843 struct ext4_filename *fname) 1844 { 1845 1846 int nlen, rlen; 1847 1848 nlen = EXT4_DIR_REC_LEN(de->name_len); 1849 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 1850 if (de->inode) { 1851 struct ext4_dir_entry_2 *de1 = 1852 (struct ext4_dir_entry_2 *)((char *)de + nlen); 1853 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size); 1854 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size); 1855 de = de1; 1856 } 1857 de->file_type = EXT4_FT_UNKNOWN; 1858 de->inode = cpu_to_le32(inode->i_ino); 1859 ext4_set_de_type(inode->i_sb, de, inode->i_mode); 1860 de->name_len = fname_len(fname); 1861 memcpy(de->name, fname_name(fname), fname_len(fname)); 1862 return 0; 1863 } 1864 1865 /* 1866 * Add a new entry into a directory (leaf) block. If de is non-NULL, 1867 * it points to a directory entry which is guaranteed to be large 1868 * enough for new directory entry. If de is NULL, then 1869 * add_dirent_to_buf will attempt search the directory block for 1870 * space. It will return -ENOSPC if no space is available, and -EIO 1871 * and -EEXIST if directory entry already exists. 1872 */ 1873 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname, 1874 struct inode *dir, 1875 struct inode *inode, struct ext4_dir_entry_2 *de, 1876 struct buffer_head *bh) 1877 { 1878 unsigned int blocksize = dir->i_sb->s_blocksize; 1879 int csum_size = 0; 1880 int err; 1881 1882 if (ext4_has_metadata_csum(inode->i_sb)) 1883 csum_size = sizeof(struct ext4_dir_entry_tail); 1884 1885 if (!de) { 1886 err = ext4_find_dest_de(dir, inode, bh, bh->b_data, 1887 blocksize - csum_size, fname, &de); 1888 if (err) 1889 return err; 1890 } 1891 BUFFER_TRACE(bh, "get_write_access"); 1892 err = ext4_journal_get_write_access(handle, bh); 1893 if (err) { 1894 ext4_std_error(dir->i_sb, err); 1895 return err; 1896 } 1897 1898 /* By now the buffer is marked for journaling. Due to crypto operations, 1899 * the following function call may fail */ 1900 err = ext4_insert_dentry(dir, inode, de, blocksize, fname); 1901 if (err < 0) 1902 return err; 1903 1904 /* 1905 * XXX shouldn't update any times until successful 1906 * completion of syscall, but too many callers depend 1907 * on this. 1908 * 1909 * XXX similarly, too many callers depend on 1910 * ext4_new_inode() setting the times, but error 1911 * recovery deletes the inode, so the worst that can 1912 * happen is that the times are slightly out of date 1913 * and/or different from the directory change time. 1914 */ 1915 dir->i_mtime = dir->i_ctime = ext4_current_time(dir); 1916 ext4_update_dx_flag(dir); 1917 dir->i_version++; 1918 ext4_mark_inode_dirty(handle, dir); 1919 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 1920 err = ext4_handle_dirty_dirent_node(handle, dir, bh); 1921 if (err) 1922 ext4_std_error(dir->i_sb, err); 1923 return 0; 1924 } 1925 1926 /* 1927 * This converts a one block unindexed directory to a 3 block indexed 1928 * directory, and adds the dentry to the indexed directory. 1929 */ 1930 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname, 1931 struct inode *dir, 1932 struct inode *inode, struct buffer_head *bh) 1933 { 1934 struct buffer_head *bh2; 1935 struct dx_root *root; 1936 struct dx_frame frames[2], *frame; 1937 struct dx_entry *entries; 1938 struct ext4_dir_entry_2 *de, *de2; 1939 struct ext4_dir_entry_tail *t; 1940 char *data1, *top; 1941 unsigned len; 1942 int retval; 1943 unsigned blocksize; 1944 ext4_lblk_t block; 1945 struct fake_dirent *fde; 1946 int csum_size = 0; 1947 1948 if (ext4_has_metadata_csum(inode->i_sb)) 1949 csum_size = sizeof(struct ext4_dir_entry_tail); 1950 1951 blocksize = dir->i_sb->s_blocksize; 1952 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino)); 1953 BUFFER_TRACE(bh, "get_write_access"); 1954 retval = ext4_journal_get_write_access(handle, bh); 1955 if (retval) { 1956 ext4_std_error(dir->i_sb, retval); 1957 brelse(bh); 1958 return retval; 1959 } 1960 root = (struct dx_root *) bh->b_data; 1961 1962 /* The 0th block becomes the root, move the dirents out */ 1963 fde = &root->dotdot; 1964 de = (struct ext4_dir_entry_2 *)((char *)fde + 1965 ext4_rec_len_from_disk(fde->rec_len, blocksize)); 1966 if ((char *) de >= (((char *) root) + blocksize)) { 1967 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'"); 1968 brelse(bh); 1969 return -EFSCORRUPTED; 1970 } 1971 len = ((char *) root) + (blocksize - csum_size) - (char *) de; 1972 1973 /* Allocate new block for the 0th block's dirents */ 1974 bh2 = ext4_append(handle, dir, &block); 1975 if (IS_ERR(bh2)) { 1976 brelse(bh); 1977 return PTR_ERR(bh2); 1978 } 1979 ext4_set_inode_flag(dir, EXT4_INODE_INDEX); 1980 data1 = bh2->b_data; 1981 1982 memcpy (data1, de, len); 1983 de = (struct ext4_dir_entry_2 *) data1; 1984 top = data1 + len; 1985 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) 1986 de = de2; 1987 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) - 1988 (char *) de, 1989 blocksize); 1990 1991 if (csum_size) { 1992 t = EXT4_DIRENT_TAIL(data1, blocksize); 1993 initialize_dirent_tail(t, blocksize); 1994 } 1995 1996 /* Initialize the root; the dot dirents already exist */ 1997 de = (struct ext4_dir_entry_2 *) (&root->dotdot); 1998 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2), 1999 blocksize); 2000 memset (&root->info, 0, sizeof(root->info)); 2001 root->info.info_length = sizeof(root->info); 2002 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; 2003 entries = root->entries; 2004 dx_set_block(entries, 1); 2005 dx_set_count(entries, 1); 2006 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info))); 2007 2008 /* Initialize as for dx_probe */ 2009 fname->hinfo.hash_version = root->info.hash_version; 2010 if (fname->hinfo.hash_version <= DX_HASH_TEA) 2011 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; 2012 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; 2013 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo); 2014 2015 memset(frames, 0, sizeof(frames)); 2016 frame = frames; 2017 frame->entries = entries; 2018 frame->at = entries; 2019 frame->bh = bh; 2020 bh = bh2; 2021 2022 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh); 2023 if (retval) 2024 goto out_frames; 2025 retval = ext4_handle_dirty_dirent_node(handle, dir, bh); 2026 if (retval) 2027 goto out_frames; 2028 2029 de = do_split(handle,dir, &bh, frame, &fname->hinfo); 2030 if (IS_ERR(de)) { 2031 retval = PTR_ERR(de); 2032 goto out_frames; 2033 } 2034 dx_release(frames); 2035 2036 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh); 2037 brelse(bh); 2038 return retval; 2039 out_frames: 2040 /* 2041 * Even if the block split failed, we have to properly write 2042 * out all the changes we did so far. Otherwise we can end up 2043 * with corrupted filesystem. 2044 */ 2045 ext4_mark_inode_dirty(handle, dir); 2046 dx_release(frames); 2047 return retval; 2048 } 2049 2050 /* 2051 * ext4_add_entry() 2052 * 2053 * adds a file entry to the specified directory, using the same 2054 * semantics as ext4_find_entry(). It returns NULL if it failed. 2055 * 2056 * NOTE!! The inode part of 'de' is left at 0 - which means you 2057 * may not sleep between calling this and putting something into 2058 * the entry, as someone else might have used it while you slept. 2059 */ 2060 static int ext4_add_entry(handle_t *handle, struct dentry *dentry, 2061 struct inode *inode) 2062 { 2063 struct inode *dir = d_inode(dentry->d_parent); 2064 struct buffer_head *bh = NULL; 2065 struct ext4_dir_entry_2 *de; 2066 struct ext4_dir_entry_tail *t; 2067 struct super_block *sb; 2068 struct ext4_filename fname; 2069 int retval; 2070 int dx_fallback=0; 2071 unsigned blocksize; 2072 ext4_lblk_t block, blocks; 2073 int csum_size = 0; 2074 2075 if (ext4_has_metadata_csum(inode->i_sb)) 2076 csum_size = sizeof(struct ext4_dir_entry_tail); 2077 2078 sb = dir->i_sb; 2079 blocksize = sb->s_blocksize; 2080 if (!dentry->d_name.len) 2081 return -EINVAL; 2082 2083 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname); 2084 if (retval) 2085 return retval; 2086 2087 if (ext4_has_inline_data(dir)) { 2088 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode); 2089 if (retval < 0) 2090 goto out; 2091 if (retval == 1) { 2092 retval = 0; 2093 goto out; 2094 } 2095 } 2096 2097 if (is_dx(dir)) { 2098 retval = ext4_dx_add_entry(handle, &fname, dir, inode); 2099 if (!retval || (retval != ERR_BAD_DX_DIR)) 2100 goto out; 2101 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX); 2102 dx_fallback++; 2103 ext4_mark_inode_dirty(handle, dir); 2104 } 2105 blocks = dir->i_size >> sb->s_blocksize_bits; 2106 for (block = 0; block < blocks; block++) { 2107 bh = ext4_read_dirblock(dir, block, DIRENT); 2108 if (IS_ERR(bh)) { 2109 retval = PTR_ERR(bh); 2110 bh = NULL; 2111 goto out; 2112 } 2113 retval = add_dirent_to_buf(handle, &fname, dir, inode, 2114 NULL, bh); 2115 if (retval != -ENOSPC) 2116 goto out; 2117 2118 if (blocks == 1 && !dx_fallback && 2119 ext4_has_feature_dir_index(sb)) { 2120 retval = make_indexed_dir(handle, &fname, dir, 2121 inode, bh); 2122 bh = NULL; /* make_indexed_dir releases bh */ 2123 goto out; 2124 } 2125 brelse(bh); 2126 } 2127 bh = ext4_append(handle, dir, &block); 2128 if (IS_ERR(bh)) { 2129 retval = PTR_ERR(bh); 2130 bh = NULL; 2131 goto out; 2132 } 2133 de = (struct ext4_dir_entry_2 *) bh->b_data; 2134 de->inode = 0; 2135 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize); 2136 2137 if (csum_size) { 2138 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize); 2139 initialize_dirent_tail(t, blocksize); 2140 } 2141 2142 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh); 2143 out: 2144 ext4_fname_free_filename(&fname); 2145 brelse(bh); 2146 if (retval == 0) 2147 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY); 2148 return retval; 2149 } 2150 2151 /* 2152 * Returns 0 for success, or a negative error value 2153 */ 2154 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname, 2155 struct inode *dir, struct inode *inode) 2156 { 2157 struct dx_frame frames[2], *frame; 2158 struct dx_entry *entries, *at; 2159 struct buffer_head *bh; 2160 struct super_block *sb = dir->i_sb; 2161 struct ext4_dir_entry_2 *de; 2162 int err; 2163 2164 frame = dx_probe(fname, dir, NULL, frames); 2165 if (IS_ERR(frame)) 2166 return PTR_ERR(frame); 2167 entries = frame->entries; 2168 at = frame->at; 2169 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT); 2170 if (IS_ERR(bh)) { 2171 err = PTR_ERR(bh); 2172 bh = NULL; 2173 goto cleanup; 2174 } 2175 2176 BUFFER_TRACE(bh, "get_write_access"); 2177 err = ext4_journal_get_write_access(handle, bh); 2178 if (err) 2179 goto journal_error; 2180 2181 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh); 2182 if (err != -ENOSPC) 2183 goto cleanup; 2184 2185 /* Block full, should compress but for now just split */ 2186 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n", 2187 dx_get_count(entries), dx_get_limit(entries))); 2188 /* Need to split index? */ 2189 if (dx_get_count(entries) == dx_get_limit(entries)) { 2190 ext4_lblk_t newblock; 2191 unsigned icount = dx_get_count(entries); 2192 int levels = frame - frames; 2193 struct dx_entry *entries2; 2194 struct dx_node *node2; 2195 struct buffer_head *bh2; 2196 2197 if (levels && (dx_get_count(frames->entries) == 2198 dx_get_limit(frames->entries))) { 2199 ext4_warning_inode(dir, "Directory index full!"); 2200 err = -ENOSPC; 2201 goto cleanup; 2202 } 2203 bh2 = ext4_append(handle, dir, &newblock); 2204 if (IS_ERR(bh2)) { 2205 err = PTR_ERR(bh2); 2206 goto cleanup; 2207 } 2208 node2 = (struct dx_node *)(bh2->b_data); 2209 entries2 = node2->entries; 2210 memset(&node2->fake, 0, sizeof(struct fake_dirent)); 2211 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize, 2212 sb->s_blocksize); 2213 BUFFER_TRACE(frame->bh, "get_write_access"); 2214 err = ext4_journal_get_write_access(handle, frame->bh); 2215 if (err) 2216 goto journal_error; 2217 if (levels) { 2218 unsigned icount1 = icount/2, icount2 = icount - icount1; 2219 unsigned hash2 = dx_get_hash(entries + icount1); 2220 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n", 2221 icount1, icount2)); 2222 2223 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */ 2224 err = ext4_journal_get_write_access(handle, 2225 frames[0].bh); 2226 if (err) 2227 goto journal_error; 2228 2229 memcpy((char *) entries2, (char *) (entries + icount1), 2230 icount2 * sizeof(struct dx_entry)); 2231 dx_set_count(entries, icount1); 2232 dx_set_count(entries2, icount2); 2233 dx_set_limit(entries2, dx_node_limit(dir)); 2234 2235 /* Which index block gets the new entry? */ 2236 if (at - entries >= icount1) { 2237 frame->at = at = at - entries - icount1 + entries2; 2238 frame->entries = entries = entries2; 2239 swap(frame->bh, bh2); 2240 } 2241 dx_insert_block(frames + 0, hash2, newblock); 2242 dxtrace(dx_show_index("node", frames[1].entries)); 2243 dxtrace(dx_show_index("node", 2244 ((struct dx_node *) bh2->b_data)->entries)); 2245 err = ext4_handle_dirty_dx_node(handle, dir, bh2); 2246 if (err) 2247 goto journal_error; 2248 brelse (bh2); 2249 } else { 2250 dxtrace(printk(KERN_DEBUG 2251 "Creating second level index...\n")); 2252 memcpy((char *) entries2, (char *) entries, 2253 icount * sizeof(struct dx_entry)); 2254 dx_set_limit(entries2, dx_node_limit(dir)); 2255 2256 /* Set up root */ 2257 dx_set_count(entries, 1); 2258 dx_set_block(entries + 0, newblock); 2259 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1; 2260 2261 /* Add new access path frame */ 2262 frame = frames + 1; 2263 frame->at = at = at - entries + entries2; 2264 frame->entries = entries = entries2; 2265 frame->bh = bh2; 2266 err = ext4_journal_get_write_access(handle, 2267 frame->bh); 2268 if (err) 2269 goto journal_error; 2270 } 2271 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh); 2272 if (err) { 2273 ext4_std_error(inode->i_sb, err); 2274 goto cleanup; 2275 } 2276 } 2277 de = do_split(handle, dir, &bh, frame, &fname->hinfo); 2278 if (IS_ERR(de)) { 2279 err = PTR_ERR(de); 2280 goto cleanup; 2281 } 2282 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh); 2283 goto cleanup; 2284 2285 journal_error: 2286 ext4_std_error(dir->i_sb, err); 2287 cleanup: 2288 brelse(bh); 2289 dx_release(frames); 2290 return err; 2291 } 2292 2293 /* 2294 * ext4_generic_delete_entry deletes a directory entry by merging it 2295 * with the previous entry 2296 */ 2297 int ext4_generic_delete_entry(handle_t *handle, 2298 struct inode *dir, 2299 struct ext4_dir_entry_2 *de_del, 2300 struct buffer_head *bh, 2301 void *entry_buf, 2302 int buf_size, 2303 int csum_size) 2304 { 2305 struct ext4_dir_entry_2 *de, *pde; 2306 unsigned int blocksize = dir->i_sb->s_blocksize; 2307 int i; 2308 2309 i = 0; 2310 pde = NULL; 2311 de = (struct ext4_dir_entry_2 *)entry_buf; 2312 while (i < buf_size - csum_size) { 2313 if (ext4_check_dir_entry(dir, NULL, de, bh, 2314 bh->b_data, bh->b_size, i)) 2315 return -EFSCORRUPTED; 2316 if (de == de_del) { 2317 if (pde) 2318 pde->rec_len = ext4_rec_len_to_disk( 2319 ext4_rec_len_from_disk(pde->rec_len, 2320 blocksize) + 2321 ext4_rec_len_from_disk(de->rec_len, 2322 blocksize), 2323 blocksize); 2324 else 2325 de->inode = 0; 2326 dir->i_version++; 2327 return 0; 2328 } 2329 i += ext4_rec_len_from_disk(de->rec_len, blocksize); 2330 pde = de; 2331 de = ext4_next_entry(de, blocksize); 2332 } 2333 return -ENOENT; 2334 } 2335 2336 static int ext4_delete_entry(handle_t *handle, 2337 struct inode *dir, 2338 struct ext4_dir_entry_2 *de_del, 2339 struct buffer_head *bh) 2340 { 2341 int err, csum_size = 0; 2342 2343 if (ext4_has_inline_data(dir)) { 2344 int has_inline_data = 1; 2345 err = ext4_delete_inline_entry(handle, dir, de_del, bh, 2346 &has_inline_data); 2347 if (has_inline_data) 2348 return err; 2349 } 2350 2351 if (ext4_has_metadata_csum(dir->i_sb)) 2352 csum_size = sizeof(struct ext4_dir_entry_tail); 2353 2354 BUFFER_TRACE(bh, "get_write_access"); 2355 err = ext4_journal_get_write_access(handle, bh); 2356 if (unlikely(err)) 2357 goto out; 2358 2359 err = ext4_generic_delete_entry(handle, dir, de_del, 2360 bh, bh->b_data, 2361 dir->i_sb->s_blocksize, csum_size); 2362 if (err) 2363 goto out; 2364 2365 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 2366 err = ext4_handle_dirty_dirent_node(handle, dir, bh); 2367 if (unlikely(err)) 2368 goto out; 2369 2370 return 0; 2371 out: 2372 if (err != -ENOENT) 2373 ext4_std_error(dir->i_sb, err); 2374 return err; 2375 } 2376 2377 /* 2378 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2, 2379 * since this indicates that nlinks count was previously 1. 2380 */ 2381 static void ext4_inc_count(handle_t *handle, struct inode *inode) 2382 { 2383 inc_nlink(inode); 2384 if (is_dx(inode) && inode->i_nlink > 1) { 2385 /* limit is 16-bit i_links_count */ 2386 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) { 2387 set_nlink(inode, 1); 2388 ext4_set_feature_dir_nlink(inode->i_sb); 2389 } 2390 } 2391 } 2392 2393 /* 2394 * If a directory had nlink == 1, then we should let it be 1. This indicates 2395 * directory has >EXT4_LINK_MAX subdirs. 2396 */ 2397 static void ext4_dec_count(handle_t *handle, struct inode *inode) 2398 { 2399 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2) 2400 drop_nlink(inode); 2401 } 2402 2403 2404 static int ext4_add_nondir(handle_t *handle, 2405 struct dentry *dentry, struct inode *inode) 2406 { 2407 int err = ext4_add_entry(handle, dentry, inode); 2408 if (!err) { 2409 ext4_mark_inode_dirty(handle, inode); 2410 unlock_new_inode(inode); 2411 d_instantiate(dentry, inode); 2412 return 0; 2413 } 2414 drop_nlink(inode); 2415 unlock_new_inode(inode); 2416 iput(inode); 2417 return err; 2418 } 2419 2420 /* 2421 * By the time this is called, we already have created 2422 * the directory cache entry for the new file, but it 2423 * is so far negative - it has no inode. 2424 * 2425 * If the create succeeds, we fill in the inode information 2426 * with d_instantiate(). 2427 */ 2428 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2429 bool excl) 2430 { 2431 handle_t *handle; 2432 struct inode *inode; 2433 int err, credits, retries = 0; 2434 2435 err = dquot_initialize(dir); 2436 if (err) 2437 return err; 2438 2439 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 2440 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); 2441 retry: 2442 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0, 2443 NULL, EXT4_HT_DIR, credits); 2444 handle = ext4_journal_current_handle(); 2445 err = PTR_ERR(inode); 2446 if (!IS_ERR(inode)) { 2447 inode->i_op = &ext4_file_inode_operations; 2448 inode->i_fop = &ext4_file_operations; 2449 ext4_set_aops(inode); 2450 err = ext4_add_nondir(handle, dentry, inode); 2451 if (!err && IS_DIRSYNC(dir)) 2452 ext4_handle_sync(handle); 2453 } 2454 if (handle) 2455 ext4_journal_stop(handle); 2456 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2457 goto retry; 2458 return err; 2459 } 2460 2461 static int ext4_mknod(struct inode *dir, struct dentry *dentry, 2462 umode_t mode, dev_t rdev) 2463 { 2464 handle_t *handle; 2465 struct inode *inode; 2466 int err, credits, retries = 0; 2467 2468 err = dquot_initialize(dir); 2469 if (err) 2470 return err; 2471 2472 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 2473 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); 2474 retry: 2475 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0, 2476 NULL, EXT4_HT_DIR, credits); 2477 handle = ext4_journal_current_handle(); 2478 err = PTR_ERR(inode); 2479 if (!IS_ERR(inode)) { 2480 init_special_inode(inode, inode->i_mode, rdev); 2481 inode->i_op = &ext4_special_inode_operations; 2482 err = ext4_add_nondir(handle, dentry, inode); 2483 if (!err && IS_DIRSYNC(dir)) 2484 ext4_handle_sync(handle); 2485 } 2486 if (handle) 2487 ext4_journal_stop(handle); 2488 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2489 goto retry; 2490 return err; 2491 } 2492 2493 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) 2494 { 2495 handle_t *handle; 2496 struct inode *inode; 2497 int err, retries = 0; 2498 2499 err = dquot_initialize(dir); 2500 if (err) 2501 return err; 2502 2503 retry: 2504 inode = ext4_new_inode_start_handle(dir, mode, 2505 NULL, 0, NULL, 2506 EXT4_HT_DIR, 2507 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) + 2508 4 + EXT4_XATTR_TRANS_BLOCKS); 2509 handle = ext4_journal_current_handle(); 2510 err = PTR_ERR(inode); 2511 if (!IS_ERR(inode)) { 2512 inode->i_op = &ext4_file_inode_operations; 2513 inode->i_fop = &ext4_file_operations; 2514 ext4_set_aops(inode); 2515 d_tmpfile(dentry, inode); 2516 err = ext4_orphan_add(handle, inode); 2517 if (err) 2518 goto err_unlock_inode; 2519 mark_inode_dirty(inode); 2520 unlock_new_inode(inode); 2521 } 2522 if (handle) 2523 ext4_journal_stop(handle); 2524 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2525 goto retry; 2526 return err; 2527 err_unlock_inode: 2528 ext4_journal_stop(handle); 2529 unlock_new_inode(inode); 2530 return err; 2531 } 2532 2533 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode, 2534 struct ext4_dir_entry_2 *de, 2535 int blocksize, int csum_size, 2536 unsigned int parent_ino, int dotdot_real_len) 2537 { 2538 de->inode = cpu_to_le32(inode->i_ino); 2539 de->name_len = 1; 2540 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len), 2541 blocksize); 2542 strcpy(de->name, "."); 2543 ext4_set_de_type(inode->i_sb, de, S_IFDIR); 2544 2545 de = ext4_next_entry(de, blocksize); 2546 de->inode = cpu_to_le32(parent_ino); 2547 de->name_len = 2; 2548 if (!dotdot_real_len) 2549 de->rec_len = ext4_rec_len_to_disk(blocksize - 2550 (csum_size + EXT4_DIR_REC_LEN(1)), 2551 blocksize); 2552 else 2553 de->rec_len = ext4_rec_len_to_disk( 2554 EXT4_DIR_REC_LEN(de->name_len), blocksize); 2555 strcpy(de->name, ".."); 2556 ext4_set_de_type(inode->i_sb, de, S_IFDIR); 2557 2558 return ext4_next_entry(de, blocksize); 2559 } 2560 2561 static int ext4_init_new_dir(handle_t *handle, struct inode *dir, 2562 struct inode *inode) 2563 { 2564 struct buffer_head *dir_block = NULL; 2565 struct ext4_dir_entry_2 *de; 2566 struct ext4_dir_entry_tail *t; 2567 ext4_lblk_t block = 0; 2568 unsigned int blocksize = dir->i_sb->s_blocksize; 2569 int csum_size = 0; 2570 int err; 2571 2572 if (ext4_has_metadata_csum(dir->i_sb)) 2573 csum_size = sizeof(struct ext4_dir_entry_tail); 2574 2575 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { 2576 err = ext4_try_create_inline_dir(handle, dir, inode); 2577 if (err < 0 && err != -ENOSPC) 2578 goto out; 2579 if (!err) 2580 goto out; 2581 } 2582 2583 inode->i_size = 0; 2584 dir_block = ext4_append(handle, inode, &block); 2585 if (IS_ERR(dir_block)) 2586 return PTR_ERR(dir_block); 2587 de = (struct ext4_dir_entry_2 *)dir_block->b_data; 2588 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0); 2589 set_nlink(inode, 2); 2590 if (csum_size) { 2591 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize); 2592 initialize_dirent_tail(t, blocksize); 2593 } 2594 2595 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata"); 2596 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block); 2597 if (err) 2598 goto out; 2599 set_buffer_verified(dir_block); 2600 out: 2601 brelse(dir_block); 2602 return err; 2603 } 2604 2605 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 2606 { 2607 handle_t *handle; 2608 struct inode *inode; 2609 int err, credits, retries = 0; 2610 2611 if (EXT4_DIR_LINK_MAX(dir)) 2612 return -EMLINK; 2613 2614 err = dquot_initialize(dir); 2615 if (err) 2616 return err; 2617 2618 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 2619 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); 2620 retry: 2621 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode, 2622 &dentry->d_name, 2623 0, NULL, EXT4_HT_DIR, credits); 2624 handle = ext4_journal_current_handle(); 2625 err = PTR_ERR(inode); 2626 if (IS_ERR(inode)) 2627 goto out_stop; 2628 2629 inode->i_op = &ext4_dir_inode_operations; 2630 inode->i_fop = &ext4_dir_operations; 2631 err = ext4_init_new_dir(handle, dir, inode); 2632 if (err) 2633 goto out_clear_inode; 2634 err = ext4_mark_inode_dirty(handle, inode); 2635 if (!err) 2636 err = ext4_add_entry(handle, dentry, inode); 2637 if (err) { 2638 out_clear_inode: 2639 clear_nlink(inode); 2640 unlock_new_inode(inode); 2641 ext4_mark_inode_dirty(handle, inode); 2642 iput(inode); 2643 goto out_stop; 2644 } 2645 ext4_inc_count(handle, dir); 2646 ext4_update_dx_flag(dir); 2647 err = ext4_mark_inode_dirty(handle, dir); 2648 if (err) 2649 goto out_clear_inode; 2650 unlock_new_inode(inode); 2651 d_instantiate(dentry, inode); 2652 if (IS_DIRSYNC(dir)) 2653 ext4_handle_sync(handle); 2654 2655 out_stop: 2656 if (handle) 2657 ext4_journal_stop(handle); 2658 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2659 goto retry; 2660 return err; 2661 } 2662 2663 /* 2664 * routine to check that the specified directory is empty (for rmdir) 2665 */ 2666 int ext4_empty_dir(struct inode *inode) 2667 { 2668 unsigned int offset; 2669 struct buffer_head *bh; 2670 struct ext4_dir_entry_2 *de, *de1; 2671 struct super_block *sb; 2672 int err = 0; 2673 2674 if (ext4_has_inline_data(inode)) { 2675 int has_inline_data = 1; 2676 2677 err = empty_inline_dir(inode, &has_inline_data); 2678 if (has_inline_data) 2679 return err; 2680 } 2681 2682 sb = inode->i_sb; 2683 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) { 2684 EXT4_ERROR_INODE(inode, "invalid size"); 2685 return 1; 2686 } 2687 bh = ext4_read_dirblock(inode, 0, EITHER); 2688 if (IS_ERR(bh)) 2689 return 1; 2690 2691 de = (struct ext4_dir_entry_2 *) bh->b_data; 2692 de1 = ext4_next_entry(de, sb->s_blocksize); 2693 if (le32_to_cpu(de->inode) != inode->i_ino || 2694 le32_to_cpu(de1->inode) == 0 || 2695 strcmp(".", de->name) || strcmp("..", de1->name)) { 2696 ext4_warning_inode(inode, "directory missing '.' and/or '..'"); 2697 brelse(bh); 2698 return 1; 2699 } 2700 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) + 2701 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize); 2702 de = ext4_next_entry(de1, sb->s_blocksize); 2703 while (offset < inode->i_size) { 2704 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) { 2705 unsigned int lblock; 2706 err = 0; 2707 brelse(bh); 2708 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb); 2709 bh = ext4_read_dirblock(inode, lblock, EITHER); 2710 if (IS_ERR(bh)) 2711 return 1; 2712 de = (struct ext4_dir_entry_2 *) bh->b_data; 2713 } 2714 if (ext4_check_dir_entry(inode, NULL, de, bh, 2715 bh->b_data, bh->b_size, offset)) { 2716 de = (struct ext4_dir_entry_2 *)(bh->b_data + 2717 sb->s_blocksize); 2718 offset = (offset | (sb->s_blocksize - 1)) + 1; 2719 continue; 2720 } 2721 if (le32_to_cpu(de->inode)) { 2722 brelse(bh); 2723 return 0; 2724 } 2725 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize); 2726 de = ext4_next_entry(de, sb->s_blocksize); 2727 } 2728 brelse(bh); 2729 return 1; 2730 } 2731 2732 /* 2733 * ext4_orphan_add() links an unlinked or truncated inode into a list of 2734 * such inodes, starting at the superblock, in case we crash before the 2735 * file is closed/deleted, or in case the inode truncate spans multiple 2736 * transactions and the last transaction is not recovered after a crash. 2737 * 2738 * At filesystem recovery time, we walk this list deleting unlinked 2739 * inodes and truncating linked inodes in ext4_orphan_cleanup(). 2740 * 2741 * Orphan list manipulation functions must be called under i_mutex unless 2742 * we are just creating the inode or deleting it. 2743 */ 2744 int ext4_orphan_add(handle_t *handle, struct inode *inode) 2745 { 2746 struct super_block *sb = inode->i_sb; 2747 struct ext4_sb_info *sbi = EXT4_SB(sb); 2748 struct ext4_iloc iloc; 2749 int err = 0, rc; 2750 bool dirty = false; 2751 2752 if (!sbi->s_journal || is_bad_inode(inode)) 2753 return 0; 2754 2755 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) && 2756 !inode_is_locked(inode)); 2757 /* 2758 * Exit early if inode already is on orphan list. This is a big speedup 2759 * since we don't have to contend on the global s_orphan_lock. 2760 */ 2761 if (!list_empty(&EXT4_I(inode)->i_orphan)) 2762 return 0; 2763 2764 /* 2765 * Orphan handling is only valid for files with data blocks 2766 * being truncated, or files being unlinked. Note that we either 2767 * hold i_mutex, or the inode can not be referenced from outside, 2768 * so i_nlink should not be bumped due to race 2769 */ 2770 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 2771 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0); 2772 2773 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 2774 err = ext4_journal_get_write_access(handle, sbi->s_sbh); 2775 if (err) 2776 goto out; 2777 2778 err = ext4_reserve_inode_write(handle, inode, &iloc); 2779 if (err) 2780 goto out; 2781 2782 mutex_lock(&sbi->s_orphan_lock); 2783 /* 2784 * Due to previous errors inode may be already a part of on-disk 2785 * orphan list. If so skip on-disk list modification. 2786 */ 2787 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) > 2788 (le32_to_cpu(sbi->s_es->s_inodes_count))) { 2789 /* Insert this inode at the head of the on-disk orphan list */ 2790 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan); 2791 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino); 2792 dirty = true; 2793 } 2794 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan); 2795 mutex_unlock(&sbi->s_orphan_lock); 2796 2797 if (dirty) { 2798 err = ext4_handle_dirty_super(handle, sb); 2799 rc = ext4_mark_iloc_dirty(handle, inode, &iloc); 2800 if (!err) 2801 err = rc; 2802 if (err) { 2803 /* 2804 * We have to remove inode from in-memory list if 2805 * addition to on disk orphan list failed. Stray orphan 2806 * list entries can cause panics at unmount time. 2807 */ 2808 mutex_lock(&sbi->s_orphan_lock); 2809 list_del(&EXT4_I(inode)->i_orphan); 2810 mutex_unlock(&sbi->s_orphan_lock); 2811 } 2812 } 2813 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino); 2814 jbd_debug(4, "orphan inode %lu will point to %d\n", 2815 inode->i_ino, NEXT_ORPHAN(inode)); 2816 out: 2817 ext4_std_error(sb, err); 2818 return err; 2819 } 2820 2821 /* 2822 * ext4_orphan_del() removes an unlinked or truncated inode from the list 2823 * of such inodes stored on disk, because it is finally being cleaned up. 2824 */ 2825 int ext4_orphan_del(handle_t *handle, struct inode *inode) 2826 { 2827 struct list_head *prev; 2828 struct ext4_inode_info *ei = EXT4_I(inode); 2829 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 2830 __u32 ino_next; 2831 struct ext4_iloc iloc; 2832 int err = 0; 2833 2834 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS)) 2835 return 0; 2836 2837 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) && 2838 !inode_is_locked(inode)); 2839 /* Do this quick check before taking global s_orphan_lock. */ 2840 if (list_empty(&ei->i_orphan)) 2841 return 0; 2842 2843 if (handle) { 2844 /* Grab inode buffer early before taking global s_orphan_lock */ 2845 err = ext4_reserve_inode_write(handle, inode, &iloc); 2846 } 2847 2848 mutex_lock(&sbi->s_orphan_lock); 2849 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino); 2850 2851 prev = ei->i_orphan.prev; 2852 list_del_init(&ei->i_orphan); 2853 2854 /* If we're on an error path, we may not have a valid 2855 * transaction handle with which to update the orphan list on 2856 * disk, but we still need to remove the inode from the linked 2857 * list in memory. */ 2858 if (!handle || err) { 2859 mutex_unlock(&sbi->s_orphan_lock); 2860 goto out_err; 2861 } 2862 2863 ino_next = NEXT_ORPHAN(inode); 2864 if (prev == &sbi->s_orphan) { 2865 jbd_debug(4, "superblock will point to %u\n", ino_next); 2866 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 2867 err = ext4_journal_get_write_access(handle, sbi->s_sbh); 2868 if (err) { 2869 mutex_unlock(&sbi->s_orphan_lock); 2870 goto out_brelse; 2871 } 2872 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next); 2873 mutex_unlock(&sbi->s_orphan_lock); 2874 err = ext4_handle_dirty_super(handle, inode->i_sb); 2875 } else { 2876 struct ext4_iloc iloc2; 2877 struct inode *i_prev = 2878 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode; 2879 2880 jbd_debug(4, "orphan inode %lu will point to %u\n", 2881 i_prev->i_ino, ino_next); 2882 err = ext4_reserve_inode_write(handle, i_prev, &iloc2); 2883 if (err) { 2884 mutex_unlock(&sbi->s_orphan_lock); 2885 goto out_brelse; 2886 } 2887 NEXT_ORPHAN(i_prev) = ino_next; 2888 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2); 2889 mutex_unlock(&sbi->s_orphan_lock); 2890 } 2891 if (err) 2892 goto out_brelse; 2893 NEXT_ORPHAN(inode) = 0; 2894 err = ext4_mark_iloc_dirty(handle, inode, &iloc); 2895 out_err: 2896 ext4_std_error(inode->i_sb, err); 2897 return err; 2898 2899 out_brelse: 2900 brelse(iloc.bh); 2901 goto out_err; 2902 } 2903 2904 static int ext4_rmdir(struct inode *dir, struct dentry *dentry) 2905 { 2906 int retval; 2907 struct inode *inode; 2908 struct buffer_head *bh; 2909 struct ext4_dir_entry_2 *de; 2910 handle_t *handle = NULL; 2911 2912 /* Initialize quotas before so that eventual writes go in 2913 * separate transaction */ 2914 retval = dquot_initialize(dir); 2915 if (retval) 2916 return retval; 2917 retval = dquot_initialize(d_inode(dentry)); 2918 if (retval) 2919 return retval; 2920 2921 retval = -ENOENT; 2922 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); 2923 if (IS_ERR(bh)) 2924 return PTR_ERR(bh); 2925 if (!bh) 2926 goto end_rmdir; 2927 2928 inode = d_inode(dentry); 2929 2930 retval = -EFSCORRUPTED; 2931 if (le32_to_cpu(de->inode) != inode->i_ino) 2932 goto end_rmdir; 2933 2934 retval = -ENOTEMPTY; 2935 if (!ext4_empty_dir(inode)) 2936 goto end_rmdir; 2937 2938 handle = ext4_journal_start(dir, EXT4_HT_DIR, 2939 EXT4_DATA_TRANS_BLOCKS(dir->i_sb)); 2940 if (IS_ERR(handle)) { 2941 retval = PTR_ERR(handle); 2942 handle = NULL; 2943 goto end_rmdir; 2944 } 2945 2946 if (IS_DIRSYNC(dir)) 2947 ext4_handle_sync(handle); 2948 2949 retval = ext4_delete_entry(handle, dir, de, bh); 2950 if (retval) 2951 goto end_rmdir; 2952 if (!EXT4_DIR_LINK_EMPTY(inode)) 2953 ext4_warning_inode(inode, 2954 "empty directory '%.*s' has too many links (%u)", 2955 dentry->d_name.len, dentry->d_name.name, 2956 inode->i_nlink); 2957 inode->i_version++; 2958 clear_nlink(inode); 2959 /* There's no need to set i_disksize: the fact that i_nlink is 2960 * zero will ensure that the right thing happens during any 2961 * recovery. */ 2962 inode->i_size = 0; 2963 ext4_orphan_add(handle, inode); 2964 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode); 2965 ext4_mark_inode_dirty(handle, inode); 2966 ext4_dec_count(handle, dir); 2967 ext4_update_dx_flag(dir); 2968 ext4_mark_inode_dirty(handle, dir); 2969 2970 end_rmdir: 2971 brelse(bh); 2972 if (handle) 2973 ext4_journal_stop(handle); 2974 return retval; 2975 } 2976 2977 static int ext4_unlink(struct inode *dir, struct dentry *dentry) 2978 { 2979 int retval; 2980 struct inode *inode; 2981 struct buffer_head *bh; 2982 struct ext4_dir_entry_2 *de; 2983 handle_t *handle = NULL; 2984 2985 trace_ext4_unlink_enter(dir, dentry); 2986 /* Initialize quotas before so that eventual writes go 2987 * in separate transaction */ 2988 retval = dquot_initialize(dir); 2989 if (retval) 2990 return retval; 2991 retval = dquot_initialize(d_inode(dentry)); 2992 if (retval) 2993 return retval; 2994 2995 retval = -ENOENT; 2996 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); 2997 if (IS_ERR(bh)) 2998 return PTR_ERR(bh); 2999 if (!bh) 3000 goto end_unlink; 3001 3002 inode = d_inode(dentry); 3003 3004 retval = -EFSCORRUPTED; 3005 if (le32_to_cpu(de->inode) != inode->i_ino) 3006 goto end_unlink; 3007 3008 handle = ext4_journal_start(dir, EXT4_HT_DIR, 3009 EXT4_DATA_TRANS_BLOCKS(dir->i_sb)); 3010 if (IS_ERR(handle)) { 3011 retval = PTR_ERR(handle); 3012 handle = NULL; 3013 goto end_unlink; 3014 } 3015 3016 if (IS_DIRSYNC(dir)) 3017 ext4_handle_sync(handle); 3018 3019 if (inode->i_nlink == 0) { 3020 ext4_warning_inode(inode, "Deleting file '%.*s' with no links", 3021 dentry->d_name.len, dentry->d_name.name); 3022 set_nlink(inode, 1); 3023 } 3024 retval = ext4_delete_entry(handle, dir, de, bh); 3025 if (retval) 3026 goto end_unlink; 3027 dir->i_ctime = dir->i_mtime = ext4_current_time(dir); 3028 ext4_update_dx_flag(dir); 3029 ext4_mark_inode_dirty(handle, dir); 3030 drop_nlink(inode); 3031 if (!inode->i_nlink) 3032 ext4_orphan_add(handle, inode); 3033 inode->i_ctime = ext4_current_time(inode); 3034 ext4_mark_inode_dirty(handle, inode); 3035 3036 end_unlink: 3037 brelse(bh); 3038 if (handle) 3039 ext4_journal_stop(handle); 3040 trace_ext4_unlink_exit(dentry, retval); 3041 return retval; 3042 } 3043 3044 static int ext4_symlink(struct inode *dir, 3045 struct dentry *dentry, const char *symname) 3046 { 3047 handle_t *handle; 3048 struct inode *inode; 3049 int err, len = strlen(symname); 3050 int credits; 3051 bool encryption_required; 3052 struct ext4_str disk_link; 3053 struct ext4_encrypted_symlink_data *sd = NULL; 3054 3055 disk_link.len = len + 1; 3056 disk_link.name = (char *) symname; 3057 3058 encryption_required = (ext4_encrypted_inode(dir) || 3059 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))); 3060 if (encryption_required) { 3061 err = ext4_get_encryption_info(dir); 3062 if (err) 3063 return err; 3064 if (ext4_encryption_info(dir) == NULL) 3065 return -EPERM; 3066 disk_link.len = (ext4_fname_encrypted_size(dir, len) + 3067 sizeof(struct ext4_encrypted_symlink_data)); 3068 sd = kzalloc(disk_link.len, GFP_KERNEL); 3069 if (!sd) 3070 return -ENOMEM; 3071 } 3072 3073 if (disk_link.len > dir->i_sb->s_blocksize) { 3074 err = -ENAMETOOLONG; 3075 goto err_free_sd; 3076 } 3077 3078 err = dquot_initialize(dir); 3079 if (err) 3080 goto err_free_sd; 3081 3082 if ((disk_link.len > EXT4_N_BLOCKS * 4)) { 3083 /* 3084 * For non-fast symlinks, we just allocate inode and put it on 3085 * orphan list in the first transaction => we need bitmap, 3086 * group descriptor, sb, inode block, quota blocks, and 3087 * possibly selinux xattr blocks. 3088 */ 3089 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) + 3090 EXT4_XATTR_TRANS_BLOCKS; 3091 } else { 3092 /* 3093 * Fast symlink. We have to add entry to directory 3094 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS), 3095 * allocate new inode (bitmap, group descriptor, inode block, 3096 * quota blocks, sb is already counted in previous macros). 3097 */ 3098 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 3099 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3; 3100 } 3101 3102 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO, 3103 &dentry->d_name, 0, NULL, 3104 EXT4_HT_DIR, credits); 3105 handle = ext4_journal_current_handle(); 3106 if (IS_ERR(inode)) { 3107 if (handle) 3108 ext4_journal_stop(handle); 3109 err = PTR_ERR(inode); 3110 goto err_free_sd; 3111 } 3112 3113 if (encryption_required) { 3114 struct qstr istr; 3115 struct ext4_str ostr; 3116 3117 istr.name = (const unsigned char *) symname; 3118 istr.len = len; 3119 ostr.name = sd->encrypted_path; 3120 ostr.len = disk_link.len; 3121 err = ext4_fname_usr_to_disk(inode, &istr, &ostr); 3122 if (err < 0) 3123 goto err_drop_inode; 3124 sd->len = cpu_to_le16(ostr.len); 3125 disk_link.name = (char *) sd; 3126 inode->i_op = &ext4_encrypted_symlink_inode_operations; 3127 } 3128 3129 if ((disk_link.len > EXT4_N_BLOCKS * 4)) { 3130 if (!encryption_required) 3131 inode->i_op = &ext4_symlink_inode_operations; 3132 inode_nohighmem(inode); 3133 ext4_set_aops(inode); 3134 /* 3135 * We cannot call page_symlink() with transaction started 3136 * because it calls into ext4_write_begin() which can wait 3137 * for transaction commit if we are running out of space 3138 * and thus we deadlock. So we have to stop transaction now 3139 * and restart it when symlink contents is written. 3140 * 3141 * To keep fs consistent in case of crash, we have to put inode 3142 * to orphan list in the mean time. 3143 */ 3144 drop_nlink(inode); 3145 err = ext4_orphan_add(handle, inode); 3146 ext4_journal_stop(handle); 3147 handle = NULL; 3148 if (err) 3149 goto err_drop_inode; 3150 err = __page_symlink(inode, disk_link.name, disk_link.len, 1); 3151 if (err) 3152 goto err_drop_inode; 3153 /* 3154 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS 3155 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified 3156 */ 3157 handle = ext4_journal_start(dir, EXT4_HT_DIR, 3158 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 3159 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1); 3160 if (IS_ERR(handle)) { 3161 err = PTR_ERR(handle); 3162 handle = NULL; 3163 goto err_drop_inode; 3164 } 3165 set_nlink(inode, 1); 3166 err = ext4_orphan_del(handle, inode); 3167 if (err) 3168 goto err_drop_inode; 3169 } else { 3170 /* clear the extent format for fast symlink */ 3171 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS); 3172 if (!encryption_required) { 3173 inode->i_op = &ext4_fast_symlink_inode_operations; 3174 inode->i_link = (char *)&EXT4_I(inode)->i_data; 3175 } 3176 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name, 3177 disk_link.len); 3178 inode->i_size = disk_link.len - 1; 3179 } 3180 EXT4_I(inode)->i_disksize = inode->i_size; 3181 err = ext4_add_nondir(handle, dentry, inode); 3182 if (!err && IS_DIRSYNC(dir)) 3183 ext4_handle_sync(handle); 3184 3185 if (handle) 3186 ext4_journal_stop(handle); 3187 kfree(sd); 3188 return err; 3189 err_drop_inode: 3190 if (handle) 3191 ext4_journal_stop(handle); 3192 clear_nlink(inode); 3193 unlock_new_inode(inode); 3194 iput(inode); 3195 err_free_sd: 3196 kfree(sd); 3197 return err; 3198 } 3199 3200 static int ext4_link(struct dentry *old_dentry, 3201 struct inode *dir, struct dentry *dentry) 3202 { 3203 handle_t *handle; 3204 struct inode *inode = d_inode(old_dentry); 3205 int err, retries = 0; 3206 3207 if (inode->i_nlink >= EXT4_LINK_MAX) 3208 return -EMLINK; 3209 if (ext4_encrypted_inode(dir) && 3210 !ext4_is_child_context_consistent_with_parent(dir, inode)) 3211 return -EPERM; 3212 3213 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) && 3214 (!projid_eq(EXT4_I(dir)->i_projid, 3215 EXT4_I(old_dentry->d_inode)->i_projid))) 3216 return -EXDEV; 3217 3218 err = dquot_initialize(dir); 3219 if (err) 3220 return err; 3221 3222 retry: 3223 handle = ext4_journal_start(dir, EXT4_HT_DIR, 3224 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 3225 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1); 3226 if (IS_ERR(handle)) 3227 return PTR_ERR(handle); 3228 3229 if (IS_DIRSYNC(dir)) 3230 ext4_handle_sync(handle); 3231 3232 inode->i_ctime = ext4_current_time(inode); 3233 ext4_inc_count(handle, inode); 3234 ihold(inode); 3235 3236 err = ext4_add_entry(handle, dentry, inode); 3237 if (!err) { 3238 ext4_mark_inode_dirty(handle, inode); 3239 /* this can happen only for tmpfile being 3240 * linked the first time 3241 */ 3242 if (inode->i_nlink == 1) 3243 ext4_orphan_del(handle, inode); 3244 d_instantiate(dentry, inode); 3245 } else { 3246 drop_nlink(inode); 3247 iput(inode); 3248 } 3249 ext4_journal_stop(handle); 3250 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 3251 goto retry; 3252 return err; 3253 } 3254 3255 3256 /* 3257 * Try to find buffer head where contains the parent block. 3258 * It should be the inode block if it is inlined or the 1st block 3259 * if it is a normal dir. 3260 */ 3261 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle, 3262 struct inode *inode, 3263 int *retval, 3264 struct ext4_dir_entry_2 **parent_de, 3265 int *inlined) 3266 { 3267 struct buffer_head *bh; 3268 3269 if (!ext4_has_inline_data(inode)) { 3270 bh = ext4_read_dirblock(inode, 0, EITHER); 3271 if (IS_ERR(bh)) { 3272 *retval = PTR_ERR(bh); 3273 return NULL; 3274 } 3275 *parent_de = ext4_next_entry( 3276 (struct ext4_dir_entry_2 *)bh->b_data, 3277 inode->i_sb->s_blocksize); 3278 return bh; 3279 } 3280 3281 *inlined = 1; 3282 return ext4_get_first_inline_block(inode, parent_de, retval); 3283 } 3284 3285 struct ext4_renament { 3286 struct inode *dir; 3287 struct dentry *dentry; 3288 struct inode *inode; 3289 bool is_dir; 3290 int dir_nlink_delta; 3291 3292 /* entry for "dentry" */ 3293 struct buffer_head *bh; 3294 struct ext4_dir_entry_2 *de; 3295 int inlined; 3296 3297 /* entry for ".." in inode if it's a directory */ 3298 struct buffer_head *dir_bh; 3299 struct ext4_dir_entry_2 *parent_de; 3300 int dir_inlined; 3301 }; 3302 3303 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent) 3304 { 3305 int retval; 3306 3307 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode, 3308 &retval, &ent->parent_de, 3309 &ent->dir_inlined); 3310 if (!ent->dir_bh) 3311 return retval; 3312 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino) 3313 return -EFSCORRUPTED; 3314 BUFFER_TRACE(ent->dir_bh, "get_write_access"); 3315 return ext4_journal_get_write_access(handle, ent->dir_bh); 3316 } 3317 3318 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent, 3319 unsigned dir_ino) 3320 { 3321 int retval; 3322 3323 ent->parent_de->inode = cpu_to_le32(dir_ino); 3324 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata"); 3325 if (!ent->dir_inlined) { 3326 if (is_dx(ent->inode)) { 3327 retval = ext4_handle_dirty_dx_node(handle, 3328 ent->inode, 3329 ent->dir_bh); 3330 } else { 3331 retval = ext4_handle_dirty_dirent_node(handle, 3332 ent->inode, 3333 ent->dir_bh); 3334 } 3335 } else { 3336 retval = ext4_mark_inode_dirty(handle, ent->inode); 3337 } 3338 if (retval) { 3339 ext4_std_error(ent->dir->i_sb, retval); 3340 return retval; 3341 } 3342 return 0; 3343 } 3344 3345 static int ext4_setent(handle_t *handle, struct ext4_renament *ent, 3346 unsigned ino, unsigned file_type) 3347 { 3348 int retval; 3349 3350 BUFFER_TRACE(ent->bh, "get write access"); 3351 retval = ext4_journal_get_write_access(handle, ent->bh); 3352 if (retval) 3353 return retval; 3354 ent->de->inode = cpu_to_le32(ino); 3355 if (ext4_has_feature_filetype(ent->dir->i_sb)) 3356 ent->de->file_type = file_type; 3357 ent->dir->i_version++; 3358 ent->dir->i_ctime = ent->dir->i_mtime = 3359 ext4_current_time(ent->dir); 3360 ext4_mark_inode_dirty(handle, ent->dir); 3361 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata"); 3362 if (!ent->inlined) { 3363 retval = ext4_handle_dirty_dirent_node(handle, 3364 ent->dir, ent->bh); 3365 if (unlikely(retval)) { 3366 ext4_std_error(ent->dir->i_sb, retval); 3367 return retval; 3368 } 3369 } 3370 brelse(ent->bh); 3371 ent->bh = NULL; 3372 3373 return 0; 3374 } 3375 3376 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir, 3377 const struct qstr *d_name) 3378 { 3379 int retval = -ENOENT; 3380 struct buffer_head *bh; 3381 struct ext4_dir_entry_2 *de; 3382 3383 bh = ext4_find_entry(dir, d_name, &de, NULL); 3384 if (IS_ERR(bh)) 3385 return PTR_ERR(bh); 3386 if (bh) { 3387 retval = ext4_delete_entry(handle, dir, de, bh); 3388 brelse(bh); 3389 } 3390 return retval; 3391 } 3392 3393 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent, 3394 int force_reread) 3395 { 3396 int retval; 3397 /* 3398 * ent->de could have moved from under us during htree split, so make 3399 * sure that we are deleting the right entry. We might also be pointing 3400 * to a stale entry in the unused part of ent->bh so just checking inum 3401 * and the name isn't enough. 3402 */ 3403 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino || 3404 ent->de->name_len != ent->dentry->d_name.len || 3405 strncmp(ent->de->name, ent->dentry->d_name.name, 3406 ent->de->name_len) || 3407 force_reread) { 3408 retval = ext4_find_delete_entry(handle, ent->dir, 3409 &ent->dentry->d_name); 3410 } else { 3411 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh); 3412 if (retval == -ENOENT) { 3413 retval = ext4_find_delete_entry(handle, ent->dir, 3414 &ent->dentry->d_name); 3415 } 3416 } 3417 3418 if (retval) { 3419 ext4_warning_inode(ent->dir, 3420 "Deleting old file: nlink %d, error=%d", 3421 ent->dir->i_nlink, retval); 3422 } 3423 } 3424 3425 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent) 3426 { 3427 if (ent->dir_nlink_delta) { 3428 if (ent->dir_nlink_delta == -1) 3429 ext4_dec_count(handle, ent->dir); 3430 else 3431 ext4_inc_count(handle, ent->dir); 3432 ext4_mark_inode_dirty(handle, ent->dir); 3433 } 3434 } 3435 3436 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent, 3437 int credits, handle_t **h) 3438 { 3439 struct inode *wh; 3440 handle_t *handle; 3441 int retries = 0; 3442 3443 /* 3444 * for inode block, sb block, group summaries, 3445 * and inode bitmap 3446 */ 3447 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) + 3448 EXT4_XATTR_TRANS_BLOCKS + 4); 3449 retry: 3450 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE, 3451 &ent->dentry->d_name, 0, NULL, 3452 EXT4_HT_DIR, credits); 3453 3454 handle = ext4_journal_current_handle(); 3455 if (IS_ERR(wh)) { 3456 if (handle) 3457 ext4_journal_stop(handle); 3458 if (PTR_ERR(wh) == -ENOSPC && 3459 ext4_should_retry_alloc(ent->dir->i_sb, &retries)) 3460 goto retry; 3461 } else { 3462 *h = handle; 3463 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV); 3464 wh->i_op = &ext4_special_inode_operations; 3465 } 3466 return wh; 3467 } 3468 3469 /* 3470 * Anybody can rename anything with this: the permission checks are left to the 3471 * higher-level routines. 3472 * 3473 * n.b. old_{dentry,inode) refers to the source dentry/inode 3474 * while new_{dentry,inode) refers to the destination dentry/inode 3475 * This comes from rename(const char *oldpath, const char *newpath) 3476 */ 3477 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry, 3478 struct inode *new_dir, struct dentry *new_dentry, 3479 unsigned int flags) 3480 { 3481 handle_t *handle = NULL; 3482 struct ext4_renament old = { 3483 .dir = old_dir, 3484 .dentry = old_dentry, 3485 .inode = d_inode(old_dentry), 3486 }; 3487 struct ext4_renament new = { 3488 .dir = new_dir, 3489 .dentry = new_dentry, 3490 .inode = d_inode(new_dentry), 3491 }; 3492 int force_reread; 3493 int retval; 3494 struct inode *whiteout = NULL; 3495 int credits; 3496 u8 old_file_type; 3497 3498 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) && 3499 (!projid_eq(EXT4_I(new_dir)->i_projid, 3500 EXT4_I(old_dentry->d_inode)->i_projid))) 3501 return -EXDEV; 3502 3503 retval = dquot_initialize(old.dir); 3504 if (retval) 3505 return retval; 3506 retval = dquot_initialize(new.dir); 3507 if (retval) 3508 return retval; 3509 3510 /* Initialize quotas before so that eventual writes go 3511 * in separate transaction */ 3512 if (new.inode) { 3513 retval = dquot_initialize(new.inode); 3514 if (retval) 3515 return retval; 3516 } 3517 3518 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL); 3519 if (IS_ERR(old.bh)) 3520 return PTR_ERR(old.bh); 3521 /* 3522 * Check for inode number is _not_ due to possible IO errors. 3523 * We might rmdir the source, keep it as pwd of some process 3524 * and merrily kill the link to whatever was created under the 3525 * same name. Goodbye sticky bit ;-< 3526 */ 3527 retval = -ENOENT; 3528 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino) 3529 goto end_rename; 3530 3531 if ((old.dir != new.dir) && 3532 ext4_encrypted_inode(new.dir) && 3533 !ext4_is_child_context_consistent_with_parent(new.dir, 3534 old.inode)) { 3535 retval = -EPERM; 3536 goto end_rename; 3537 } 3538 3539 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name, 3540 &new.de, &new.inlined); 3541 if (IS_ERR(new.bh)) { 3542 retval = PTR_ERR(new.bh); 3543 new.bh = NULL; 3544 goto end_rename; 3545 } 3546 if (new.bh) { 3547 if (!new.inode) { 3548 brelse(new.bh); 3549 new.bh = NULL; 3550 } 3551 } 3552 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC)) 3553 ext4_alloc_da_blocks(old.inode); 3554 3555 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) + 3556 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2); 3557 if (!(flags & RENAME_WHITEOUT)) { 3558 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits); 3559 if (IS_ERR(handle)) { 3560 retval = PTR_ERR(handle); 3561 handle = NULL; 3562 goto end_rename; 3563 } 3564 } else { 3565 whiteout = ext4_whiteout_for_rename(&old, credits, &handle); 3566 if (IS_ERR(whiteout)) { 3567 retval = PTR_ERR(whiteout); 3568 whiteout = NULL; 3569 goto end_rename; 3570 } 3571 } 3572 3573 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir)) 3574 ext4_handle_sync(handle); 3575 3576 if (S_ISDIR(old.inode->i_mode)) { 3577 if (new.inode) { 3578 retval = -ENOTEMPTY; 3579 if (!ext4_empty_dir(new.inode)) 3580 goto end_rename; 3581 } else { 3582 retval = -EMLINK; 3583 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir)) 3584 goto end_rename; 3585 } 3586 retval = ext4_rename_dir_prepare(handle, &old); 3587 if (retval) 3588 goto end_rename; 3589 } 3590 /* 3591 * If we're renaming a file within an inline_data dir and adding or 3592 * setting the new dirent causes a conversion from inline_data to 3593 * extents/blockmap, we need to force the dirent delete code to 3594 * re-read the directory, or else we end up trying to delete a dirent 3595 * from what is now the extent tree root (or a block map). 3596 */ 3597 force_reread = (new.dir->i_ino == old.dir->i_ino && 3598 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA)); 3599 3600 old_file_type = old.de->file_type; 3601 if (whiteout) { 3602 /* 3603 * Do this before adding a new entry, so the old entry is sure 3604 * to be still pointing to the valid old entry. 3605 */ 3606 retval = ext4_setent(handle, &old, whiteout->i_ino, 3607 EXT4_FT_CHRDEV); 3608 if (retval) 3609 goto end_rename; 3610 ext4_mark_inode_dirty(handle, whiteout); 3611 } 3612 if (!new.bh) { 3613 retval = ext4_add_entry(handle, new.dentry, old.inode); 3614 if (retval) 3615 goto end_rename; 3616 } else { 3617 retval = ext4_setent(handle, &new, 3618 old.inode->i_ino, old_file_type); 3619 if (retval) 3620 goto end_rename; 3621 } 3622 if (force_reread) 3623 force_reread = !ext4_test_inode_flag(new.dir, 3624 EXT4_INODE_INLINE_DATA); 3625 3626 /* 3627 * Like most other Unix systems, set the ctime for inodes on a 3628 * rename. 3629 */ 3630 old.inode->i_ctime = ext4_current_time(old.inode); 3631 ext4_mark_inode_dirty(handle, old.inode); 3632 3633 if (!whiteout) { 3634 /* 3635 * ok, that's it 3636 */ 3637 ext4_rename_delete(handle, &old, force_reread); 3638 } 3639 3640 if (new.inode) { 3641 ext4_dec_count(handle, new.inode); 3642 new.inode->i_ctime = ext4_current_time(new.inode); 3643 } 3644 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir); 3645 ext4_update_dx_flag(old.dir); 3646 if (old.dir_bh) { 3647 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino); 3648 if (retval) 3649 goto end_rename; 3650 3651 ext4_dec_count(handle, old.dir); 3652 if (new.inode) { 3653 /* checked ext4_empty_dir above, can't have another 3654 * parent, ext4_dec_count() won't work for many-linked 3655 * dirs */ 3656 clear_nlink(new.inode); 3657 } else { 3658 ext4_inc_count(handle, new.dir); 3659 ext4_update_dx_flag(new.dir); 3660 ext4_mark_inode_dirty(handle, new.dir); 3661 } 3662 } 3663 ext4_mark_inode_dirty(handle, old.dir); 3664 if (new.inode) { 3665 ext4_mark_inode_dirty(handle, new.inode); 3666 if (!new.inode->i_nlink) 3667 ext4_orphan_add(handle, new.inode); 3668 } 3669 retval = 0; 3670 3671 end_rename: 3672 brelse(old.dir_bh); 3673 brelse(old.bh); 3674 brelse(new.bh); 3675 if (whiteout) { 3676 if (retval) 3677 drop_nlink(whiteout); 3678 unlock_new_inode(whiteout); 3679 iput(whiteout); 3680 } 3681 if (handle) 3682 ext4_journal_stop(handle); 3683 return retval; 3684 } 3685 3686 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry, 3687 struct inode *new_dir, struct dentry *new_dentry) 3688 { 3689 handle_t *handle = NULL; 3690 struct ext4_renament old = { 3691 .dir = old_dir, 3692 .dentry = old_dentry, 3693 .inode = d_inode(old_dentry), 3694 }; 3695 struct ext4_renament new = { 3696 .dir = new_dir, 3697 .dentry = new_dentry, 3698 .inode = d_inode(new_dentry), 3699 }; 3700 u8 new_file_type; 3701 int retval; 3702 3703 if ((ext4_encrypted_inode(old_dir) || 3704 ext4_encrypted_inode(new_dir)) && 3705 (old_dir != new_dir) && 3706 (!ext4_is_child_context_consistent_with_parent(new_dir, 3707 old.inode) || 3708 !ext4_is_child_context_consistent_with_parent(old_dir, 3709 new.inode))) 3710 return -EPERM; 3711 3712 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) && 3713 !projid_eq(EXT4_I(new_dir)->i_projid, 3714 EXT4_I(old_dentry->d_inode)->i_projid)) || 3715 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) && 3716 !projid_eq(EXT4_I(old_dir)->i_projid, 3717 EXT4_I(new_dentry->d_inode)->i_projid))) 3718 return -EXDEV; 3719 3720 retval = dquot_initialize(old.dir); 3721 if (retval) 3722 return retval; 3723 retval = dquot_initialize(new.dir); 3724 if (retval) 3725 return retval; 3726 3727 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, 3728 &old.de, &old.inlined); 3729 if (IS_ERR(old.bh)) 3730 return PTR_ERR(old.bh); 3731 /* 3732 * Check for inode number is _not_ due to possible IO errors. 3733 * We might rmdir the source, keep it as pwd of some process 3734 * and merrily kill the link to whatever was created under the 3735 * same name. Goodbye sticky bit ;-< 3736 */ 3737 retval = -ENOENT; 3738 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino) 3739 goto end_rename; 3740 3741 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name, 3742 &new.de, &new.inlined); 3743 if (IS_ERR(new.bh)) { 3744 retval = PTR_ERR(new.bh); 3745 new.bh = NULL; 3746 goto end_rename; 3747 } 3748 3749 /* RENAME_EXCHANGE case: old *and* new must both exist */ 3750 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino) 3751 goto end_rename; 3752 3753 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, 3754 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) + 3755 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2)); 3756 if (IS_ERR(handle)) { 3757 retval = PTR_ERR(handle); 3758 handle = NULL; 3759 goto end_rename; 3760 } 3761 3762 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir)) 3763 ext4_handle_sync(handle); 3764 3765 if (S_ISDIR(old.inode->i_mode)) { 3766 old.is_dir = true; 3767 retval = ext4_rename_dir_prepare(handle, &old); 3768 if (retval) 3769 goto end_rename; 3770 } 3771 if (S_ISDIR(new.inode->i_mode)) { 3772 new.is_dir = true; 3773 retval = ext4_rename_dir_prepare(handle, &new); 3774 if (retval) 3775 goto end_rename; 3776 } 3777 3778 /* 3779 * Other than the special case of overwriting a directory, parents' 3780 * nlink only needs to be modified if this is a cross directory rename. 3781 */ 3782 if (old.dir != new.dir && old.is_dir != new.is_dir) { 3783 old.dir_nlink_delta = old.is_dir ? -1 : 1; 3784 new.dir_nlink_delta = -old.dir_nlink_delta; 3785 retval = -EMLINK; 3786 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) || 3787 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir))) 3788 goto end_rename; 3789 } 3790 3791 new_file_type = new.de->file_type; 3792 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type); 3793 if (retval) 3794 goto end_rename; 3795 3796 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type); 3797 if (retval) 3798 goto end_rename; 3799 3800 /* 3801 * Like most other Unix systems, set the ctime for inodes on a 3802 * rename. 3803 */ 3804 old.inode->i_ctime = ext4_current_time(old.inode); 3805 new.inode->i_ctime = ext4_current_time(new.inode); 3806 ext4_mark_inode_dirty(handle, old.inode); 3807 ext4_mark_inode_dirty(handle, new.inode); 3808 3809 if (old.dir_bh) { 3810 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino); 3811 if (retval) 3812 goto end_rename; 3813 } 3814 if (new.dir_bh) { 3815 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino); 3816 if (retval) 3817 goto end_rename; 3818 } 3819 ext4_update_dir_count(handle, &old); 3820 ext4_update_dir_count(handle, &new); 3821 retval = 0; 3822 3823 end_rename: 3824 brelse(old.dir_bh); 3825 brelse(new.dir_bh); 3826 brelse(old.bh); 3827 brelse(new.bh); 3828 if (handle) 3829 ext4_journal_stop(handle); 3830 return retval; 3831 } 3832 3833 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry, 3834 struct inode *new_dir, struct dentry *new_dentry, 3835 unsigned int flags) 3836 { 3837 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 3838 return -EINVAL; 3839 3840 if (flags & RENAME_EXCHANGE) { 3841 return ext4_cross_rename(old_dir, old_dentry, 3842 new_dir, new_dentry); 3843 } 3844 3845 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags); 3846 } 3847 3848 /* 3849 * directories can handle most operations... 3850 */ 3851 const struct inode_operations ext4_dir_inode_operations = { 3852 .create = ext4_create, 3853 .lookup = ext4_lookup, 3854 .link = ext4_link, 3855 .unlink = ext4_unlink, 3856 .symlink = ext4_symlink, 3857 .mkdir = ext4_mkdir, 3858 .rmdir = ext4_rmdir, 3859 .mknod = ext4_mknod, 3860 .tmpfile = ext4_tmpfile, 3861 .rename2 = ext4_rename2, 3862 .setattr = ext4_setattr, 3863 .setxattr = generic_setxattr, 3864 .getxattr = generic_getxattr, 3865 .listxattr = ext4_listxattr, 3866 .removexattr = generic_removexattr, 3867 .get_acl = ext4_get_acl, 3868 .set_acl = ext4_set_acl, 3869 .fiemap = ext4_fiemap, 3870 }; 3871 3872 const struct inode_operations ext4_special_inode_operations = { 3873 .setattr = ext4_setattr, 3874 .setxattr = generic_setxattr, 3875 .getxattr = generic_getxattr, 3876 .listxattr = ext4_listxattr, 3877 .removexattr = generic_removexattr, 3878 .get_acl = ext4_get_acl, 3879 .set_acl = ext4_set_acl, 3880 }; 3881