1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext4/namei.c 4 * 5 * Copyright (C) 1992, 1993, 1994, 1995 6 * Remy Card (card@masi.ibp.fr) 7 * Laboratoire MASI - Institut Blaise Pascal 8 * Universite Pierre et Marie Curie (Paris VI) 9 * 10 * from 11 * 12 * linux/fs/minix/namei.c 13 * 14 * Copyright (C) 1991, 1992 Linus Torvalds 15 * 16 * Big-endian to little-endian byte-swapping/bitmaps by 17 * David S. Miller (davem@caip.rutgers.edu), 1995 18 * Directory entry file type support and forward compatibility hooks 19 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998 20 * Hash Tree Directory indexing (c) 21 * Daniel Phillips, 2001 22 * Hash Tree Directory indexing porting 23 * Christopher Li, 2002 24 * Hash Tree Directory indexing cleanup 25 * Theodore Ts'o, 2002 26 */ 27 28 #include <linux/fs.h> 29 #include <linux/pagemap.h> 30 #include <linux/time.h> 31 #include <linux/fcntl.h> 32 #include <linux/stat.h> 33 #include <linux/string.h> 34 #include <linux/quotaops.h> 35 #include <linux/buffer_head.h> 36 #include <linux/bio.h> 37 #include <linux/iversion.h> 38 #include <linux/unicode.h> 39 #include "ext4.h" 40 #include "ext4_jbd2.h" 41 42 #include "xattr.h" 43 #include "acl.h" 44 45 #include <trace/events/ext4.h> 46 /* 47 * define how far ahead to read directories while searching them. 48 */ 49 #define NAMEI_RA_CHUNKS 2 50 #define NAMEI_RA_BLOCKS 4 51 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS) 52 53 static struct buffer_head *ext4_append(handle_t *handle, 54 struct inode *inode, 55 ext4_lblk_t *block) 56 { 57 struct buffer_head *bh; 58 int err; 59 60 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb && 61 ((inode->i_size >> 10) >= 62 EXT4_SB(inode->i_sb)->s_max_dir_size_kb))) 63 return ERR_PTR(-ENOSPC); 64 65 *block = inode->i_size >> inode->i_sb->s_blocksize_bits; 66 67 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE); 68 if (IS_ERR(bh)) 69 return bh; 70 inode->i_size += inode->i_sb->s_blocksize; 71 EXT4_I(inode)->i_disksize = inode->i_size; 72 BUFFER_TRACE(bh, "get_write_access"); 73 err = ext4_journal_get_write_access(handle, bh); 74 if (err) { 75 brelse(bh); 76 ext4_std_error(inode->i_sb, err); 77 return ERR_PTR(err); 78 } 79 return bh; 80 } 81 82 static int ext4_dx_csum_verify(struct inode *inode, 83 struct ext4_dir_entry *dirent); 84 85 /* 86 * Hints to ext4_read_dirblock regarding whether we expect a directory 87 * block being read to be an index block, or a block containing 88 * directory entries (and if the latter, whether it was found via a 89 * logical block in an htree index block). This is used to control 90 * what sort of sanity checkinig ext4_read_dirblock() will do on the 91 * directory block read from the storage device. EITHER will means 92 * the caller doesn't know what kind of directory block will be read, 93 * so no specific verification will be done. 94 */ 95 typedef enum { 96 EITHER, INDEX, DIRENT, DIRENT_HTREE 97 } dirblock_type_t; 98 99 #define ext4_read_dirblock(inode, block, type) \ 100 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__) 101 102 static struct buffer_head *__ext4_read_dirblock(struct inode *inode, 103 ext4_lblk_t block, 104 dirblock_type_t type, 105 const char *func, 106 unsigned int line) 107 { 108 struct buffer_head *bh; 109 struct ext4_dir_entry *dirent; 110 int is_dx_block = 0; 111 112 if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO)) 113 bh = ERR_PTR(-EIO); 114 else 115 bh = ext4_bread(NULL, inode, block, 0); 116 if (IS_ERR(bh)) { 117 __ext4_warning(inode->i_sb, func, line, 118 "inode #%lu: lblock %lu: comm %s: " 119 "error %ld reading directory block", 120 inode->i_ino, (unsigned long)block, 121 current->comm, PTR_ERR(bh)); 122 123 return bh; 124 } 125 if (!bh && (type == INDEX || type == DIRENT_HTREE)) { 126 ext4_error_inode(inode, func, line, block, 127 "Directory hole found for htree %s block", 128 (type == INDEX) ? "index" : "leaf"); 129 return ERR_PTR(-EFSCORRUPTED); 130 } 131 if (!bh) 132 return NULL; 133 dirent = (struct ext4_dir_entry *) bh->b_data; 134 /* Determine whether or not we have an index block */ 135 if (is_dx(inode)) { 136 if (block == 0) 137 is_dx_block = 1; 138 else if (ext4_rec_len_from_disk(dirent->rec_len, 139 inode->i_sb->s_blocksize) == 140 inode->i_sb->s_blocksize) 141 is_dx_block = 1; 142 } 143 if (!is_dx_block && type == INDEX) { 144 ext4_error_inode(inode, func, line, block, 145 "directory leaf block found instead of index block"); 146 brelse(bh); 147 return ERR_PTR(-EFSCORRUPTED); 148 } 149 if (!ext4_has_metadata_csum(inode->i_sb) || 150 buffer_verified(bh)) 151 return bh; 152 153 /* 154 * An empty leaf block can get mistaken for a index block; for 155 * this reason, we can only check the index checksum when the 156 * caller is sure it should be an index block. 157 */ 158 if (is_dx_block && type == INDEX) { 159 if (ext4_dx_csum_verify(inode, dirent) && 160 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC)) 161 set_buffer_verified(bh); 162 else { 163 ext4_error_inode_err(inode, func, line, block, 164 EFSBADCRC, 165 "Directory index failed checksum"); 166 brelse(bh); 167 return ERR_PTR(-EFSBADCRC); 168 } 169 } 170 if (!is_dx_block) { 171 if (ext4_dirblock_csum_verify(inode, bh) && 172 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC)) 173 set_buffer_verified(bh); 174 else { 175 ext4_error_inode_err(inode, func, line, block, 176 EFSBADCRC, 177 "Directory block failed checksum"); 178 brelse(bh); 179 return ERR_PTR(-EFSBADCRC); 180 } 181 } 182 return bh; 183 } 184 185 #ifdef DX_DEBUG 186 #define dxtrace(command) command 187 #else 188 #define dxtrace(command) 189 #endif 190 191 struct fake_dirent 192 { 193 __le32 inode; 194 __le16 rec_len; 195 u8 name_len; 196 u8 file_type; 197 }; 198 199 struct dx_countlimit 200 { 201 __le16 limit; 202 __le16 count; 203 }; 204 205 struct dx_entry 206 { 207 __le32 hash; 208 __le32 block; 209 }; 210 211 /* 212 * dx_root_info is laid out so that if it should somehow get overlaid by a 213 * dirent the two low bits of the hash version will be zero. Therefore, the 214 * hash version mod 4 should never be 0. Sincerely, the paranoia department. 215 */ 216 217 struct dx_root 218 { 219 struct fake_dirent dot; 220 char dot_name[4]; 221 struct fake_dirent dotdot; 222 char dotdot_name[4]; 223 struct dx_root_info 224 { 225 __le32 reserved_zero; 226 u8 hash_version; 227 u8 info_length; /* 8 */ 228 u8 indirect_levels; 229 u8 unused_flags; 230 } 231 info; 232 struct dx_entry entries[]; 233 }; 234 235 struct dx_node 236 { 237 struct fake_dirent fake; 238 struct dx_entry entries[]; 239 }; 240 241 242 struct dx_frame 243 { 244 struct buffer_head *bh; 245 struct dx_entry *entries; 246 struct dx_entry *at; 247 }; 248 249 struct dx_map_entry 250 { 251 u32 hash; 252 u16 offs; 253 u16 size; 254 }; 255 256 /* 257 * This goes at the end of each htree block. 258 */ 259 struct dx_tail { 260 u32 dt_reserved; 261 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */ 262 }; 263 264 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry); 265 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value); 266 static inline unsigned dx_get_hash(struct dx_entry *entry); 267 static void dx_set_hash(struct dx_entry *entry, unsigned value); 268 static unsigned dx_get_count(struct dx_entry *entries); 269 static unsigned dx_get_limit(struct dx_entry *entries); 270 static void dx_set_count(struct dx_entry *entries, unsigned value); 271 static void dx_set_limit(struct dx_entry *entries, unsigned value); 272 static unsigned dx_root_limit(struct inode *dir, unsigned infosize); 273 static unsigned dx_node_limit(struct inode *dir); 274 static struct dx_frame *dx_probe(struct ext4_filename *fname, 275 struct inode *dir, 276 struct dx_hash_info *hinfo, 277 struct dx_frame *frame); 278 static void dx_release(struct dx_frame *frames); 279 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de, 280 unsigned blocksize, struct dx_hash_info *hinfo, 281 struct dx_map_entry map[]); 282 static void dx_sort_map(struct dx_map_entry *map, unsigned count); 283 static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from, 284 char *to, struct dx_map_entry *offsets, 285 int count, unsigned int blocksize); 286 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base, 287 unsigned int blocksize); 288 static void dx_insert_block(struct dx_frame *frame, 289 u32 hash, ext4_lblk_t block); 290 static int ext4_htree_next_block(struct inode *dir, __u32 hash, 291 struct dx_frame *frame, 292 struct dx_frame *frames, 293 __u32 *start_hash); 294 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, 295 struct ext4_filename *fname, 296 struct ext4_dir_entry_2 **res_dir); 297 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname, 298 struct inode *dir, struct inode *inode); 299 300 /* checksumming functions */ 301 void ext4_initialize_dirent_tail(struct buffer_head *bh, 302 unsigned int blocksize) 303 { 304 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize); 305 306 memset(t, 0, sizeof(struct ext4_dir_entry_tail)); 307 t->det_rec_len = ext4_rec_len_to_disk( 308 sizeof(struct ext4_dir_entry_tail), blocksize); 309 t->det_reserved_ft = EXT4_FT_DIR_CSUM; 310 } 311 312 /* Walk through a dirent block to find a checksum "dirent" at the tail */ 313 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode, 314 struct buffer_head *bh) 315 { 316 struct ext4_dir_entry_tail *t; 317 318 #ifdef PARANOID 319 struct ext4_dir_entry *d, *top; 320 321 d = (struct ext4_dir_entry *)bh->b_data; 322 top = (struct ext4_dir_entry *)(bh->b_data + 323 (EXT4_BLOCK_SIZE(inode->i_sb) - 324 sizeof(struct ext4_dir_entry_tail))); 325 while (d < top && d->rec_len) 326 d = (struct ext4_dir_entry *)(((void *)d) + 327 le16_to_cpu(d->rec_len)); 328 329 if (d != top) 330 return NULL; 331 332 t = (struct ext4_dir_entry_tail *)d; 333 #else 334 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb)); 335 #endif 336 337 if (t->det_reserved_zero1 || 338 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) || 339 t->det_reserved_zero2 || 340 t->det_reserved_ft != EXT4_FT_DIR_CSUM) 341 return NULL; 342 343 return t; 344 } 345 346 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size) 347 { 348 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 349 struct ext4_inode_info *ei = EXT4_I(inode); 350 __u32 csum; 351 352 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size); 353 return cpu_to_le32(csum); 354 } 355 356 #define warn_no_space_for_csum(inode) \ 357 __warn_no_space_for_csum((inode), __func__, __LINE__) 358 359 static void __warn_no_space_for_csum(struct inode *inode, const char *func, 360 unsigned int line) 361 { 362 __ext4_warning_inode(inode, func, line, 363 "No space for directory leaf checksum. Please run e2fsck -D."); 364 } 365 366 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh) 367 { 368 struct ext4_dir_entry_tail *t; 369 370 if (!ext4_has_metadata_csum(inode->i_sb)) 371 return 1; 372 373 t = get_dirent_tail(inode, bh); 374 if (!t) { 375 warn_no_space_for_csum(inode); 376 return 0; 377 } 378 379 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data, 380 (char *)t - bh->b_data)) 381 return 0; 382 383 return 1; 384 } 385 386 static void ext4_dirblock_csum_set(struct inode *inode, 387 struct buffer_head *bh) 388 { 389 struct ext4_dir_entry_tail *t; 390 391 if (!ext4_has_metadata_csum(inode->i_sb)) 392 return; 393 394 t = get_dirent_tail(inode, bh); 395 if (!t) { 396 warn_no_space_for_csum(inode); 397 return; 398 } 399 400 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data, 401 (char *)t - bh->b_data); 402 } 403 404 int ext4_handle_dirty_dirblock(handle_t *handle, 405 struct inode *inode, 406 struct buffer_head *bh) 407 { 408 ext4_dirblock_csum_set(inode, bh); 409 return ext4_handle_dirty_metadata(handle, inode, bh); 410 } 411 412 static struct dx_countlimit *get_dx_countlimit(struct inode *inode, 413 struct ext4_dir_entry *dirent, 414 int *offset) 415 { 416 struct ext4_dir_entry *dp; 417 struct dx_root_info *root; 418 int count_offset; 419 420 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb)) 421 count_offset = 8; 422 else if (le16_to_cpu(dirent->rec_len) == 12) { 423 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12); 424 if (le16_to_cpu(dp->rec_len) != 425 EXT4_BLOCK_SIZE(inode->i_sb) - 12) 426 return NULL; 427 root = (struct dx_root_info *)(((void *)dp + 12)); 428 if (root->reserved_zero || 429 root->info_length != sizeof(struct dx_root_info)) 430 return NULL; 431 count_offset = 32; 432 } else 433 return NULL; 434 435 if (offset) 436 *offset = count_offset; 437 return (struct dx_countlimit *)(((void *)dirent) + count_offset); 438 } 439 440 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent, 441 int count_offset, int count, struct dx_tail *t) 442 { 443 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 444 struct ext4_inode_info *ei = EXT4_I(inode); 445 __u32 csum; 446 int size; 447 __u32 dummy_csum = 0; 448 int offset = offsetof(struct dx_tail, dt_checksum); 449 450 size = count_offset + (count * sizeof(struct dx_entry)); 451 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size); 452 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset); 453 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum)); 454 455 return cpu_to_le32(csum); 456 } 457 458 static int ext4_dx_csum_verify(struct inode *inode, 459 struct ext4_dir_entry *dirent) 460 { 461 struct dx_countlimit *c; 462 struct dx_tail *t; 463 int count_offset, limit, count; 464 465 if (!ext4_has_metadata_csum(inode->i_sb)) 466 return 1; 467 468 c = get_dx_countlimit(inode, dirent, &count_offset); 469 if (!c) { 470 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D."); 471 return 0; 472 } 473 limit = le16_to_cpu(c->limit); 474 count = le16_to_cpu(c->count); 475 if (count_offset + (limit * sizeof(struct dx_entry)) > 476 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) { 477 warn_no_space_for_csum(inode); 478 return 0; 479 } 480 t = (struct dx_tail *)(((struct dx_entry *)c) + limit); 481 482 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset, 483 count, t)) 484 return 0; 485 return 1; 486 } 487 488 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent) 489 { 490 struct dx_countlimit *c; 491 struct dx_tail *t; 492 int count_offset, limit, count; 493 494 if (!ext4_has_metadata_csum(inode->i_sb)) 495 return; 496 497 c = get_dx_countlimit(inode, dirent, &count_offset); 498 if (!c) { 499 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D."); 500 return; 501 } 502 limit = le16_to_cpu(c->limit); 503 count = le16_to_cpu(c->count); 504 if (count_offset + (limit * sizeof(struct dx_entry)) > 505 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) { 506 warn_no_space_for_csum(inode); 507 return; 508 } 509 t = (struct dx_tail *)(((struct dx_entry *)c) + limit); 510 511 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t); 512 } 513 514 static inline int ext4_handle_dirty_dx_node(handle_t *handle, 515 struct inode *inode, 516 struct buffer_head *bh) 517 { 518 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data); 519 return ext4_handle_dirty_metadata(handle, inode, bh); 520 } 521 522 /* 523 * p is at least 6 bytes before the end of page 524 */ 525 static inline struct ext4_dir_entry_2 * 526 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize) 527 { 528 return (struct ext4_dir_entry_2 *)((char *)p + 529 ext4_rec_len_from_disk(p->rec_len, blocksize)); 530 } 531 532 /* 533 * Future: use high four bits of block for coalesce-on-delete flags 534 * Mask them off for now. 535 */ 536 537 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry) 538 { 539 return le32_to_cpu(entry->block) & 0x0fffffff; 540 } 541 542 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value) 543 { 544 entry->block = cpu_to_le32(value); 545 } 546 547 static inline unsigned dx_get_hash(struct dx_entry *entry) 548 { 549 return le32_to_cpu(entry->hash); 550 } 551 552 static inline void dx_set_hash(struct dx_entry *entry, unsigned value) 553 { 554 entry->hash = cpu_to_le32(value); 555 } 556 557 static inline unsigned dx_get_count(struct dx_entry *entries) 558 { 559 return le16_to_cpu(((struct dx_countlimit *) entries)->count); 560 } 561 562 static inline unsigned dx_get_limit(struct dx_entry *entries) 563 { 564 return le16_to_cpu(((struct dx_countlimit *) entries)->limit); 565 } 566 567 static inline void dx_set_count(struct dx_entry *entries, unsigned value) 568 { 569 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value); 570 } 571 572 static inline void dx_set_limit(struct dx_entry *entries, unsigned value) 573 { 574 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value); 575 } 576 577 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize) 578 { 579 unsigned int entry_space = dir->i_sb->s_blocksize - 580 ext4_dir_rec_len(1, NULL) - 581 ext4_dir_rec_len(2, NULL) - infosize; 582 583 if (ext4_has_metadata_csum(dir->i_sb)) 584 entry_space -= sizeof(struct dx_tail); 585 return entry_space / sizeof(struct dx_entry); 586 } 587 588 static inline unsigned dx_node_limit(struct inode *dir) 589 { 590 unsigned int entry_space = dir->i_sb->s_blocksize - 591 ext4_dir_rec_len(0, dir); 592 593 if (ext4_has_metadata_csum(dir->i_sb)) 594 entry_space -= sizeof(struct dx_tail); 595 return entry_space / sizeof(struct dx_entry); 596 } 597 598 /* 599 * Debug 600 */ 601 #ifdef DX_DEBUG 602 static void dx_show_index(char * label, struct dx_entry *entries) 603 { 604 int i, n = dx_get_count (entries); 605 printk(KERN_DEBUG "%s index", label); 606 for (i = 0; i < n; i++) { 607 printk(KERN_CONT " %x->%lu", 608 i ? dx_get_hash(entries + i) : 0, 609 (unsigned long)dx_get_block(entries + i)); 610 } 611 printk(KERN_CONT "\n"); 612 } 613 614 struct stats 615 { 616 unsigned names; 617 unsigned space; 618 unsigned bcount; 619 }; 620 621 static struct stats dx_show_leaf(struct inode *dir, 622 struct dx_hash_info *hinfo, 623 struct ext4_dir_entry_2 *de, 624 int size, int show_names) 625 { 626 unsigned names = 0, space = 0; 627 char *base = (char *) de; 628 struct dx_hash_info h = *hinfo; 629 630 printk("names: "); 631 while ((char *) de < base + size) 632 { 633 if (de->inode) 634 { 635 if (show_names) 636 { 637 #ifdef CONFIG_FS_ENCRYPTION 638 int len; 639 char *name; 640 struct fscrypt_str fname_crypto_str = 641 FSTR_INIT(NULL, 0); 642 int res = 0; 643 644 name = de->name; 645 len = de->name_len; 646 if (!IS_ENCRYPTED(dir)) { 647 /* Directory is not encrypted */ 648 ext4fs_dirhash(dir, de->name, 649 de->name_len, &h); 650 printk("%*.s:(U)%x.%u ", len, 651 name, h.hash, 652 (unsigned) ((char *) de 653 - base)); 654 } else { 655 struct fscrypt_str de_name = 656 FSTR_INIT(name, len); 657 658 /* Directory is encrypted */ 659 res = fscrypt_fname_alloc_buffer( 660 len, &fname_crypto_str); 661 if (res) 662 printk(KERN_WARNING "Error " 663 "allocating crypto " 664 "buffer--skipping " 665 "crypto\n"); 666 res = fscrypt_fname_disk_to_usr(dir, 667 0, 0, &de_name, 668 &fname_crypto_str); 669 if (res) { 670 printk(KERN_WARNING "Error " 671 "converting filename " 672 "from disk to usr" 673 "\n"); 674 name = "??"; 675 len = 2; 676 } else { 677 name = fname_crypto_str.name; 678 len = fname_crypto_str.len; 679 } 680 if (IS_CASEFOLDED(dir)) 681 h.hash = EXT4_DIRENT_HASH(de); 682 else 683 ext4fs_dirhash(dir, de->name, 684 de->name_len, &h); 685 printk("%*.s:(E)%x.%u ", len, name, 686 h.hash, (unsigned) ((char *) de 687 - base)); 688 fscrypt_fname_free_buffer( 689 &fname_crypto_str); 690 } 691 #else 692 int len = de->name_len; 693 char *name = de->name; 694 ext4fs_dirhash(dir, de->name, de->name_len, &h); 695 printk("%*.s:%x.%u ", len, name, h.hash, 696 (unsigned) ((char *) de - base)); 697 #endif 698 } 699 space += ext4_dir_rec_len(de->name_len, dir); 700 names++; 701 } 702 de = ext4_next_entry(de, size); 703 } 704 printk(KERN_CONT "(%i)\n", names); 705 return (struct stats) { names, space, 1 }; 706 } 707 708 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir, 709 struct dx_entry *entries, int levels) 710 { 711 unsigned blocksize = dir->i_sb->s_blocksize; 712 unsigned count = dx_get_count(entries), names = 0, space = 0, i; 713 unsigned bcount = 0; 714 struct buffer_head *bh; 715 printk("%i indexed blocks...\n", count); 716 for (i = 0; i < count; i++, entries++) 717 { 718 ext4_lblk_t block = dx_get_block(entries); 719 ext4_lblk_t hash = i ? dx_get_hash(entries): 0; 720 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash; 721 struct stats stats; 722 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range); 723 bh = ext4_bread(NULL,dir, block, 0); 724 if (!bh || IS_ERR(bh)) 725 continue; 726 stats = levels? 727 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1): 728 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) 729 bh->b_data, blocksize, 0); 730 names += stats.names; 731 space += stats.space; 732 bcount += stats.bcount; 733 brelse(bh); 734 } 735 if (bcount) 736 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n", 737 levels ? "" : " ", names, space/bcount, 738 (space/bcount)*100/blocksize); 739 return (struct stats) { names, space, bcount}; 740 } 741 742 /* 743 * Linear search cross check 744 */ 745 static inline void htree_rep_invariant_check(struct dx_entry *at, 746 struct dx_entry *target, 747 u32 hash, unsigned int n) 748 { 749 while (n--) { 750 dxtrace(printk(KERN_CONT ",")); 751 if (dx_get_hash(++at) > hash) { 752 at--; 753 break; 754 } 755 } 756 ASSERT(at == target - 1); 757 } 758 #else /* DX_DEBUG */ 759 static inline void htree_rep_invariant_check(struct dx_entry *at, 760 struct dx_entry *target, 761 u32 hash, unsigned int n) 762 { 763 } 764 #endif /* DX_DEBUG */ 765 766 /* 767 * Probe for a directory leaf block to search. 768 * 769 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format 770 * error in the directory index, and the caller should fall back to 771 * searching the directory normally. The callers of dx_probe **MUST** 772 * check for this error code, and make sure it never gets reflected 773 * back to userspace. 774 */ 775 static struct dx_frame * 776 dx_probe(struct ext4_filename *fname, struct inode *dir, 777 struct dx_hash_info *hinfo, struct dx_frame *frame_in) 778 { 779 unsigned count, indirect; 780 struct dx_entry *at, *entries, *p, *q, *m; 781 struct dx_root *root; 782 struct dx_frame *frame = frame_in; 783 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR); 784 u32 hash; 785 786 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0])); 787 frame->bh = ext4_read_dirblock(dir, 0, INDEX); 788 if (IS_ERR(frame->bh)) 789 return (struct dx_frame *) frame->bh; 790 791 root = (struct dx_root *) frame->bh->b_data; 792 if (root->info.hash_version != DX_HASH_TEA && 793 root->info.hash_version != DX_HASH_HALF_MD4 && 794 root->info.hash_version != DX_HASH_LEGACY && 795 root->info.hash_version != DX_HASH_SIPHASH) { 796 ext4_warning_inode(dir, "Unrecognised inode hash code %u", 797 root->info.hash_version); 798 goto fail; 799 } 800 if (ext4_hash_in_dirent(dir)) { 801 if (root->info.hash_version != DX_HASH_SIPHASH) { 802 ext4_warning_inode(dir, 803 "Hash in dirent, but hash is not SIPHASH"); 804 goto fail; 805 } 806 } else { 807 if (root->info.hash_version == DX_HASH_SIPHASH) { 808 ext4_warning_inode(dir, 809 "Hash code is SIPHASH, but hash not in dirent"); 810 goto fail; 811 } 812 } 813 if (fname) 814 hinfo = &fname->hinfo; 815 hinfo->hash_version = root->info.hash_version; 816 if (hinfo->hash_version <= DX_HASH_TEA) 817 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; 818 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed; 819 if (fname && fname_name(fname)) 820 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo); 821 hash = hinfo->hash; 822 823 if (root->info.unused_flags & 1) { 824 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x", 825 root->info.unused_flags); 826 goto fail; 827 } 828 829 indirect = root->info.indirect_levels; 830 if (indirect >= ext4_dir_htree_level(dir->i_sb)) { 831 ext4_warning(dir->i_sb, 832 "Directory (ino: %lu) htree depth %#06x exceed" 833 "supported value", dir->i_ino, 834 ext4_dir_htree_level(dir->i_sb)); 835 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) { 836 ext4_warning(dir->i_sb, "Enable large directory " 837 "feature to access it"); 838 } 839 goto fail; 840 } 841 842 entries = (struct dx_entry *)(((char *)&root->info) + 843 root->info.info_length); 844 845 if (dx_get_limit(entries) != dx_root_limit(dir, 846 root->info.info_length)) { 847 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u", 848 dx_get_limit(entries), 849 dx_root_limit(dir, root->info.info_length)); 850 goto fail; 851 } 852 853 dxtrace(printk("Look up %x", hash)); 854 while (1) { 855 count = dx_get_count(entries); 856 if (!count || count > dx_get_limit(entries)) { 857 ext4_warning_inode(dir, 858 "dx entry: count %u beyond limit %u", 859 count, dx_get_limit(entries)); 860 goto fail; 861 } 862 863 p = entries + 1; 864 q = entries + count - 1; 865 while (p <= q) { 866 m = p + (q - p) / 2; 867 dxtrace(printk(KERN_CONT ".")); 868 if (dx_get_hash(m) > hash) 869 q = m - 1; 870 else 871 p = m + 1; 872 } 873 874 htree_rep_invariant_check(entries, p, hash, count - 1); 875 876 at = p - 1; 877 dxtrace(printk(KERN_CONT " %x->%u\n", 878 at == entries ? 0 : dx_get_hash(at), 879 dx_get_block(at))); 880 frame->entries = entries; 881 frame->at = at; 882 if (!indirect--) 883 return frame; 884 frame++; 885 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX); 886 if (IS_ERR(frame->bh)) { 887 ret_err = (struct dx_frame *) frame->bh; 888 frame->bh = NULL; 889 goto fail; 890 } 891 entries = ((struct dx_node *) frame->bh->b_data)->entries; 892 893 if (dx_get_limit(entries) != dx_node_limit(dir)) { 894 ext4_warning_inode(dir, 895 "dx entry: limit %u != node limit %u", 896 dx_get_limit(entries), dx_node_limit(dir)); 897 goto fail; 898 } 899 } 900 fail: 901 while (frame >= frame_in) { 902 brelse(frame->bh); 903 frame--; 904 } 905 906 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR)) 907 ext4_warning_inode(dir, 908 "Corrupt directory, running e2fsck is recommended"); 909 return ret_err; 910 } 911 912 static void dx_release(struct dx_frame *frames) 913 { 914 struct dx_root_info *info; 915 int i; 916 unsigned int indirect_levels; 917 918 if (frames[0].bh == NULL) 919 return; 920 921 info = &((struct dx_root *)frames[0].bh->b_data)->info; 922 /* save local copy, "info" may be freed after brelse() */ 923 indirect_levels = info->indirect_levels; 924 for (i = 0; i <= indirect_levels; i++) { 925 if (frames[i].bh == NULL) 926 break; 927 brelse(frames[i].bh); 928 frames[i].bh = NULL; 929 } 930 } 931 932 /* 933 * This function increments the frame pointer to search the next leaf 934 * block, and reads in the necessary intervening nodes if the search 935 * should be necessary. Whether or not the search is necessary is 936 * controlled by the hash parameter. If the hash value is even, then 937 * the search is only continued if the next block starts with that 938 * hash value. This is used if we are searching for a specific file. 939 * 940 * If the hash value is HASH_NB_ALWAYS, then always go to the next block. 941 * 942 * This function returns 1 if the caller should continue to search, 943 * or 0 if it should not. If there is an error reading one of the 944 * index blocks, it will a negative error code. 945 * 946 * If start_hash is non-null, it will be filled in with the starting 947 * hash of the next page. 948 */ 949 static int ext4_htree_next_block(struct inode *dir, __u32 hash, 950 struct dx_frame *frame, 951 struct dx_frame *frames, 952 __u32 *start_hash) 953 { 954 struct dx_frame *p; 955 struct buffer_head *bh; 956 int num_frames = 0; 957 __u32 bhash; 958 959 p = frame; 960 /* 961 * Find the next leaf page by incrementing the frame pointer. 962 * If we run out of entries in the interior node, loop around and 963 * increment pointer in the parent node. When we break out of 964 * this loop, num_frames indicates the number of interior 965 * nodes need to be read. 966 */ 967 while (1) { 968 if (++(p->at) < p->entries + dx_get_count(p->entries)) 969 break; 970 if (p == frames) 971 return 0; 972 num_frames++; 973 p--; 974 } 975 976 /* 977 * If the hash is 1, then continue only if the next page has a 978 * continuation hash of any value. This is used for readdir 979 * handling. Otherwise, check to see if the hash matches the 980 * desired contiuation hash. If it doesn't, return since 981 * there's no point to read in the successive index pages. 982 */ 983 bhash = dx_get_hash(p->at); 984 if (start_hash) 985 *start_hash = bhash; 986 if ((hash & 1) == 0) { 987 if ((bhash & ~1) != hash) 988 return 0; 989 } 990 /* 991 * If the hash is HASH_NB_ALWAYS, we always go to the next 992 * block so no check is necessary 993 */ 994 while (num_frames--) { 995 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX); 996 if (IS_ERR(bh)) 997 return PTR_ERR(bh); 998 p++; 999 brelse(p->bh); 1000 p->bh = bh; 1001 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries; 1002 } 1003 return 1; 1004 } 1005 1006 1007 /* 1008 * This function fills a red-black tree with information from a 1009 * directory block. It returns the number directory entries loaded 1010 * into the tree. If there is an error it is returned in err. 1011 */ 1012 static int htree_dirblock_to_tree(struct file *dir_file, 1013 struct inode *dir, ext4_lblk_t block, 1014 struct dx_hash_info *hinfo, 1015 __u32 start_hash, __u32 start_minor_hash) 1016 { 1017 struct buffer_head *bh; 1018 struct ext4_dir_entry_2 *de, *top; 1019 int err = 0, count = 0; 1020 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str; 1021 int csum = ext4_has_metadata_csum(dir->i_sb); 1022 1023 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n", 1024 (unsigned long)block)); 1025 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE); 1026 if (IS_ERR(bh)) 1027 return PTR_ERR(bh); 1028 1029 de = (struct ext4_dir_entry_2 *) bh->b_data; 1030 /* csum entries are not larger in the casefolded encrypted case */ 1031 top = (struct ext4_dir_entry_2 *) ((char *) de + 1032 dir->i_sb->s_blocksize - 1033 ext4_dir_rec_len(0, 1034 csum ? NULL : dir)); 1035 /* Check if the directory is encrypted */ 1036 if (IS_ENCRYPTED(dir)) { 1037 err = fscrypt_prepare_readdir(dir); 1038 if (err < 0) { 1039 brelse(bh); 1040 return err; 1041 } 1042 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN, 1043 &fname_crypto_str); 1044 if (err < 0) { 1045 brelse(bh); 1046 return err; 1047 } 1048 } 1049 1050 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) { 1051 if (ext4_check_dir_entry(dir, NULL, de, bh, 1052 bh->b_data, bh->b_size, 1053 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb)) 1054 + ((char *)de - bh->b_data))) { 1055 /* silently ignore the rest of the block */ 1056 break; 1057 } 1058 if (ext4_hash_in_dirent(dir)) { 1059 if (de->name_len && de->inode) { 1060 hinfo->hash = EXT4_DIRENT_HASH(de); 1061 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de); 1062 } else { 1063 hinfo->hash = 0; 1064 hinfo->minor_hash = 0; 1065 } 1066 } else { 1067 ext4fs_dirhash(dir, de->name, de->name_len, hinfo); 1068 } 1069 if ((hinfo->hash < start_hash) || 1070 ((hinfo->hash == start_hash) && 1071 (hinfo->minor_hash < start_minor_hash))) 1072 continue; 1073 if (de->inode == 0) 1074 continue; 1075 if (!IS_ENCRYPTED(dir)) { 1076 tmp_str.name = de->name; 1077 tmp_str.len = de->name_len; 1078 err = ext4_htree_store_dirent(dir_file, 1079 hinfo->hash, hinfo->minor_hash, de, 1080 &tmp_str); 1081 } else { 1082 int save_len = fname_crypto_str.len; 1083 struct fscrypt_str de_name = FSTR_INIT(de->name, 1084 de->name_len); 1085 1086 /* Directory is encrypted */ 1087 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash, 1088 hinfo->minor_hash, &de_name, 1089 &fname_crypto_str); 1090 if (err) { 1091 count = err; 1092 goto errout; 1093 } 1094 err = ext4_htree_store_dirent(dir_file, 1095 hinfo->hash, hinfo->minor_hash, de, 1096 &fname_crypto_str); 1097 fname_crypto_str.len = save_len; 1098 } 1099 if (err != 0) { 1100 count = err; 1101 goto errout; 1102 } 1103 count++; 1104 } 1105 errout: 1106 brelse(bh); 1107 fscrypt_fname_free_buffer(&fname_crypto_str); 1108 return count; 1109 } 1110 1111 1112 /* 1113 * This function fills a red-black tree with information from a 1114 * directory. We start scanning the directory in hash order, starting 1115 * at start_hash and start_minor_hash. 1116 * 1117 * This function returns the number of entries inserted into the tree, 1118 * or a negative error code. 1119 */ 1120 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash, 1121 __u32 start_minor_hash, __u32 *next_hash) 1122 { 1123 struct dx_hash_info hinfo; 1124 struct ext4_dir_entry_2 *de; 1125 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; 1126 struct inode *dir; 1127 ext4_lblk_t block; 1128 int count = 0; 1129 int ret, err; 1130 __u32 hashval; 1131 struct fscrypt_str tmp_str; 1132 1133 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n", 1134 start_hash, start_minor_hash)); 1135 dir = file_inode(dir_file); 1136 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) { 1137 if (ext4_hash_in_dirent(dir)) 1138 hinfo.hash_version = DX_HASH_SIPHASH; 1139 else 1140 hinfo.hash_version = 1141 EXT4_SB(dir->i_sb)->s_def_hash_version; 1142 if (hinfo.hash_version <= DX_HASH_TEA) 1143 hinfo.hash_version += 1144 EXT4_SB(dir->i_sb)->s_hash_unsigned; 1145 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; 1146 if (ext4_has_inline_data(dir)) { 1147 int has_inline_data = 1; 1148 count = ext4_inlinedir_to_tree(dir_file, dir, 0, 1149 &hinfo, start_hash, 1150 start_minor_hash, 1151 &has_inline_data); 1152 if (has_inline_data) { 1153 *next_hash = ~0; 1154 return count; 1155 } 1156 } 1157 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo, 1158 start_hash, start_minor_hash); 1159 *next_hash = ~0; 1160 return count; 1161 } 1162 hinfo.hash = start_hash; 1163 hinfo.minor_hash = 0; 1164 frame = dx_probe(NULL, dir, &hinfo, frames); 1165 if (IS_ERR(frame)) 1166 return PTR_ERR(frame); 1167 1168 /* Add '.' and '..' from the htree header */ 1169 if (!start_hash && !start_minor_hash) { 1170 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; 1171 tmp_str.name = de->name; 1172 tmp_str.len = de->name_len; 1173 err = ext4_htree_store_dirent(dir_file, 0, 0, 1174 de, &tmp_str); 1175 if (err != 0) 1176 goto errout; 1177 count++; 1178 } 1179 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) { 1180 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; 1181 de = ext4_next_entry(de, dir->i_sb->s_blocksize); 1182 tmp_str.name = de->name; 1183 tmp_str.len = de->name_len; 1184 err = ext4_htree_store_dirent(dir_file, 2, 0, 1185 de, &tmp_str); 1186 if (err != 0) 1187 goto errout; 1188 count++; 1189 } 1190 1191 while (1) { 1192 if (fatal_signal_pending(current)) { 1193 err = -ERESTARTSYS; 1194 goto errout; 1195 } 1196 cond_resched(); 1197 block = dx_get_block(frame->at); 1198 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo, 1199 start_hash, start_minor_hash); 1200 if (ret < 0) { 1201 err = ret; 1202 goto errout; 1203 } 1204 count += ret; 1205 hashval = ~0; 1206 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS, 1207 frame, frames, &hashval); 1208 *next_hash = hashval; 1209 if (ret < 0) { 1210 err = ret; 1211 goto errout; 1212 } 1213 /* 1214 * Stop if: (a) there are no more entries, or 1215 * (b) we have inserted at least one entry and the 1216 * next hash value is not a continuation 1217 */ 1218 if ((ret == 0) || 1219 (count && ((hashval & 1) == 0))) 1220 break; 1221 } 1222 dx_release(frames); 1223 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, " 1224 "next hash: %x\n", count, *next_hash)); 1225 return count; 1226 errout: 1227 dx_release(frames); 1228 return (err); 1229 } 1230 1231 static inline int search_dirblock(struct buffer_head *bh, 1232 struct inode *dir, 1233 struct ext4_filename *fname, 1234 unsigned int offset, 1235 struct ext4_dir_entry_2 **res_dir) 1236 { 1237 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir, 1238 fname, offset, res_dir); 1239 } 1240 1241 /* 1242 * Directory block splitting, compacting 1243 */ 1244 1245 /* 1246 * Create map of hash values, offsets, and sizes, stored at end of block. 1247 * Returns number of entries mapped. 1248 */ 1249 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de, 1250 unsigned blocksize, struct dx_hash_info *hinfo, 1251 struct dx_map_entry *map_tail) 1252 { 1253 int count = 0; 1254 char *base = (char *) de; 1255 struct dx_hash_info h = *hinfo; 1256 1257 while ((char *) de < base + blocksize) { 1258 if (de->name_len && de->inode) { 1259 if (ext4_hash_in_dirent(dir)) 1260 h.hash = EXT4_DIRENT_HASH(de); 1261 else 1262 ext4fs_dirhash(dir, de->name, de->name_len, &h); 1263 map_tail--; 1264 map_tail->hash = h.hash; 1265 map_tail->offs = ((char *) de - base)>>2; 1266 map_tail->size = le16_to_cpu(de->rec_len); 1267 count++; 1268 cond_resched(); 1269 } 1270 /* XXX: do we need to check rec_len == 0 case? -Chris */ 1271 de = ext4_next_entry(de, blocksize); 1272 } 1273 return count; 1274 } 1275 1276 /* Sort map by hash value */ 1277 static void dx_sort_map (struct dx_map_entry *map, unsigned count) 1278 { 1279 struct dx_map_entry *p, *q, *top = map + count - 1; 1280 int more; 1281 /* Combsort until bubble sort doesn't suck */ 1282 while (count > 2) { 1283 count = count*10/13; 1284 if (count - 9 < 2) /* 9, 10 -> 11 */ 1285 count = 11; 1286 for (p = top, q = p - count; q >= map; p--, q--) 1287 if (p->hash < q->hash) 1288 swap(*p, *q); 1289 } 1290 /* Garden variety bubble sort */ 1291 do { 1292 more = 0; 1293 q = top; 1294 while (q-- > map) { 1295 if (q[1].hash >= q[0].hash) 1296 continue; 1297 swap(*(q+1), *q); 1298 more = 1; 1299 } 1300 } while(more); 1301 } 1302 1303 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block) 1304 { 1305 struct dx_entry *entries = frame->entries; 1306 struct dx_entry *old = frame->at, *new = old + 1; 1307 int count = dx_get_count(entries); 1308 1309 ASSERT(count < dx_get_limit(entries)); 1310 ASSERT(old < entries + count); 1311 memmove(new + 1, new, (char *)(entries + count) - (char *)(new)); 1312 dx_set_hash(new, hash); 1313 dx_set_block(new, block); 1314 dx_set_count(entries, count + 1); 1315 } 1316 1317 #ifdef CONFIG_UNICODE 1318 /* 1319 * Test whether a case-insensitive directory entry matches the filename 1320 * being searched for. If quick is set, assume the name being looked up 1321 * is already in the casefolded form. 1322 * 1323 * Returns: 0 if the directory entry matches, more than 0 if it 1324 * doesn't match or less than zero on error. 1325 */ 1326 static int ext4_ci_compare(const struct inode *parent, const struct qstr *name, 1327 u8 *de_name, size_t de_name_len, bool quick) 1328 { 1329 const struct super_block *sb = parent->i_sb; 1330 const struct unicode_map *um = sb->s_encoding; 1331 struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len); 1332 struct qstr entry = QSTR_INIT(de_name, de_name_len); 1333 int ret; 1334 1335 if (IS_ENCRYPTED(parent)) { 1336 const struct fscrypt_str encrypted_name = 1337 FSTR_INIT(de_name, de_name_len); 1338 1339 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL); 1340 if (!decrypted_name.name) 1341 return -ENOMEM; 1342 ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name, 1343 &decrypted_name); 1344 if (ret < 0) 1345 goto out; 1346 entry.name = decrypted_name.name; 1347 entry.len = decrypted_name.len; 1348 } 1349 1350 if (quick) 1351 ret = utf8_strncasecmp_folded(um, name, &entry); 1352 else 1353 ret = utf8_strncasecmp(um, name, &entry); 1354 if (ret < 0) { 1355 /* Handle invalid character sequence as either an error 1356 * or as an opaque byte sequence. 1357 */ 1358 if (sb_has_strict_encoding(sb)) 1359 ret = -EINVAL; 1360 else if (name->len != entry.len) 1361 ret = 1; 1362 else 1363 ret = !!memcmp(name->name, entry.name, entry.len); 1364 } 1365 out: 1366 kfree(decrypted_name.name); 1367 return ret; 1368 } 1369 1370 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname, 1371 struct fscrypt_str *cf_name) 1372 { 1373 int len; 1374 1375 if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding) { 1376 cf_name->name = NULL; 1377 return; 1378 } 1379 1380 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS); 1381 if (!cf_name->name) 1382 return; 1383 1384 len = utf8_casefold(dir->i_sb->s_encoding, 1385 iname, cf_name->name, 1386 EXT4_NAME_LEN); 1387 if (len <= 0) { 1388 kfree(cf_name->name); 1389 cf_name->name = NULL; 1390 return; 1391 } 1392 cf_name->len = (unsigned) len; 1393 1394 } 1395 #endif 1396 1397 /* 1398 * Test whether a directory entry matches the filename being searched for. 1399 * 1400 * Return: %true if the directory entry matches, otherwise %false. 1401 */ 1402 static bool ext4_match(struct inode *parent, 1403 const struct ext4_filename *fname, 1404 struct ext4_dir_entry_2 *de) 1405 { 1406 struct fscrypt_name f; 1407 1408 if (!de->inode) 1409 return false; 1410 1411 f.usr_fname = fname->usr_fname; 1412 f.disk_name = fname->disk_name; 1413 #ifdef CONFIG_FS_ENCRYPTION 1414 f.crypto_buf = fname->crypto_buf; 1415 #endif 1416 1417 #ifdef CONFIG_UNICODE 1418 if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent)) { 1419 if (fname->cf_name.name) { 1420 struct qstr cf = {.name = fname->cf_name.name, 1421 .len = fname->cf_name.len}; 1422 if (IS_ENCRYPTED(parent)) { 1423 struct dx_hash_info hinfo; 1424 1425 hinfo.hash_version = DX_HASH_SIPHASH; 1426 hinfo.seed = NULL; 1427 ext4fs_dirhash(parent, fname->cf_name.name, 1428 fname_len(fname), &hinfo); 1429 if (hinfo.hash != EXT4_DIRENT_HASH(de) || 1430 hinfo.minor_hash != 1431 EXT4_DIRENT_MINOR_HASH(de)) 1432 return 0; 1433 } 1434 return !ext4_ci_compare(parent, &cf, de->name, 1435 de->name_len, true); 1436 } 1437 return !ext4_ci_compare(parent, fname->usr_fname, de->name, 1438 de->name_len, false); 1439 } 1440 #endif 1441 1442 return fscrypt_match_name(&f, de->name, de->name_len); 1443 } 1444 1445 /* 1446 * Returns 0 if not found, -1 on failure, and 1 on success 1447 */ 1448 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size, 1449 struct inode *dir, struct ext4_filename *fname, 1450 unsigned int offset, struct ext4_dir_entry_2 **res_dir) 1451 { 1452 struct ext4_dir_entry_2 * de; 1453 char * dlimit; 1454 int de_len; 1455 1456 de = (struct ext4_dir_entry_2 *)search_buf; 1457 dlimit = search_buf + buf_size; 1458 while ((char *) de < dlimit) { 1459 /* this code is executed quadratically often */ 1460 /* do minimal checking `by hand' */ 1461 if ((char *) de + de->name_len <= dlimit && 1462 ext4_match(dir, fname, de)) { 1463 /* found a match - just to be sure, do 1464 * a full check */ 1465 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf, 1466 buf_size, offset)) 1467 return -1; 1468 *res_dir = de; 1469 return 1; 1470 } 1471 /* prevent looping on a bad block */ 1472 de_len = ext4_rec_len_from_disk(de->rec_len, 1473 dir->i_sb->s_blocksize); 1474 if (de_len <= 0) 1475 return -1; 1476 offset += de_len; 1477 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len); 1478 } 1479 return 0; 1480 } 1481 1482 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block, 1483 struct ext4_dir_entry *de) 1484 { 1485 struct super_block *sb = dir->i_sb; 1486 1487 if (!is_dx(dir)) 1488 return 0; 1489 if (block == 0) 1490 return 1; 1491 if (de->inode == 0 && 1492 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) == 1493 sb->s_blocksize) 1494 return 1; 1495 return 0; 1496 } 1497 1498 /* 1499 * __ext4_find_entry() 1500 * 1501 * finds an entry in the specified directory with the wanted name. It 1502 * returns the cache buffer in which the entry was found, and the entry 1503 * itself (as a parameter - res_dir). It does NOT read the inode of the 1504 * entry - you'll have to do that yourself if you want to. 1505 * 1506 * The returned buffer_head has ->b_count elevated. The caller is expected 1507 * to brelse() it when appropriate. 1508 */ 1509 static struct buffer_head *__ext4_find_entry(struct inode *dir, 1510 struct ext4_filename *fname, 1511 struct ext4_dir_entry_2 **res_dir, 1512 int *inlined) 1513 { 1514 struct super_block *sb; 1515 struct buffer_head *bh_use[NAMEI_RA_SIZE]; 1516 struct buffer_head *bh, *ret = NULL; 1517 ext4_lblk_t start, block; 1518 const u8 *name = fname->usr_fname->name; 1519 size_t ra_max = 0; /* Number of bh's in the readahead 1520 buffer, bh_use[] */ 1521 size_t ra_ptr = 0; /* Current index into readahead 1522 buffer */ 1523 ext4_lblk_t nblocks; 1524 int i, namelen, retval; 1525 1526 *res_dir = NULL; 1527 sb = dir->i_sb; 1528 namelen = fname->usr_fname->len; 1529 if (namelen > EXT4_NAME_LEN) 1530 return NULL; 1531 1532 if (ext4_has_inline_data(dir)) { 1533 int has_inline_data = 1; 1534 ret = ext4_find_inline_entry(dir, fname, res_dir, 1535 &has_inline_data); 1536 if (has_inline_data) { 1537 if (inlined) 1538 *inlined = 1; 1539 goto cleanup_and_exit; 1540 } 1541 } 1542 1543 if ((namelen <= 2) && (name[0] == '.') && 1544 (name[1] == '.' || name[1] == '\0')) { 1545 /* 1546 * "." or ".." will only be in the first block 1547 * NFS may look up ".."; "." should be handled by the VFS 1548 */ 1549 block = start = 0; 1550 nblocks = 1; 1551 goto restart; 1552 } 1553 if (is_dx(dir)) { 1554 ret = ext4_dx_find_entry(dir, fname, res_dir); 1555 /* 1556 * On success, or if the error was file not found, 1557 * return. Otherwise, fall back to doing a search the 1558 * old fashioned way. 1559 */ 1560 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR) 1561 goto cleanup_and_exit; 1562 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, " 1563 "falling back\n")); 1564 ret = NULL; 1565 } 1566 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); 1567 if (!nblocks) { 1568 ret = NULL; 1569 goto cleanup_and_exit; 1570 } 1571 start = EXT4_I(dir)->i_dir_start_lookup; 1572 if (start >= nblocks) 1573 start = 0; 1574 block = start; 1575 restart: 1576 do { 1577 /* 1578 * We deal with the read-ahead logic here. 1579 */ 1580 cond_resched(); 1581 if (ra_ptr >= ra_max) { 1582 /* Refill the readahead buffer */ 1583 ra_ptr = 0; 1584 if (block < start) 1585 ra_max = start - block; 1586 else 1587 ra_max = nblocks - block; 1588 ra_max = min(ra_max, ARRAY_SIZE(bh_use)); 1589 retval = ext4_bread_batch(dir, block, ra_max, 1590 false /* wait */, bh_use); 1591 if (retval) { 1592 ret = ERR_PTR(retval); 1593 ra_max = 0; 1594 goto cleanup_and_exit; 1595 } 1596 } 1597 if ((bh = bh_use[ra_ptr++]) == NULL) 1598 goto next; 1599 wait_on_buffer(bh); 1600 if (!buffer_uptodate(bh)) { 1601 EXT4_ERROR_INODE_ERR(dir, EIO, 1602 "reading directory lblock %lu", 1603 (unsigned long) block); 1604 brelse(bh); 1605 ret = ERR_PTR(-EIO); 1606 goto cleanup_and_exit; 1607 } 1608 if (!buffer_verified(bh) && 1609 !is_dx_internal_node(dir, block, 1610 (struct ext4_dir_entry *)bh->b_data) && 1611 !ext4_dirblock_csum_verify(dir, bh)) { 1612 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC, 1613 "checksumming directory " 1614 "block %lu", (unsigned long)block); 1615 brelse(bh); 1616 ret = ERR_PTR(-EFSBADCRC); 1617 goto cleanup_and_exit; 1618 } 1619 set_buffer_verified(bh); 1620 i = search_dirblock(bh, dir, fname, 1621 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir); 1622 if (i == 1) { 1623 EXT4_I(dir)->i_dir_start_lookup = block; 1624 ret = bh; 1625 goto cleanup_and_exit; 1626 } else { 1627 brelse(bh); 1628 if (i < 0) 1629 goto cleanup_and_exit; 1630 } 1631 next: 1632 if (++block >= nblocks) 1633 block = 0; 1634 } while (block != start); 1635 1636 /* 1637 * If the directory has grown while we were searching, then 1638 * search the last part of the directory before giving up. 1639 */ 1640 block = nblocks; 1641 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); 1642 if (block < nblocks) { 1643 start = 0; 1644 goto restart; 1645 } 1646 1647 cleanup_and_exit: 1648 /* Clean up the read-ahead blocks */ 1649 for (; ra_ptr < ra_max; ra_ptr++) 1650 brelse(bh_use[ra_ptr]); 1651 return ret; 1652 } 1653 1654 static struct buffer_head *ext4_find_entry(struct inode *dir, 1655 const struct qstr *d_name, 1656 struct ext4_dir_entry_2 **res_dir, 1657 int *inlined) 1658 { 1659 int err; 1660 struct ext4_filename fname; 1661 struct buffer_head *bh; 1662 1663 err = ext4_fname_setup_filename(dir, d_name, 1, &fname); 1664 if (err == -ENOENT) 1665 return NULL; 1666 if (err) 1667 return ERR_PTR(err); 1668 1669 bh = __ext4_find_entry(dir, &fname, res_dir, inlined); 1670 1671 ext4_fname_free_filename(&fname); 1672 return bh; 1673 } 1674 1675 static struct buffer_head *ext4_lookup_entry(struct inode *dir, 1676 struct dentry *dentry, 1677 struct ext4_dir_entry_2 **res_dir) 1678 { 1679 int err; 1680 struct ext4_filename fname; 1681 struct buffer_head *bh; 1682 1683 err = ext4_fname_prepare_lookup(dir, dentry, &fname); 1684 generic_set_encrypted_ci_d_ops(dentry); 1685 if (err == -ENOENT) 1686 return NULL; 1687 if (err) 1688 return ERR_PTR(err); 1689 1690 bh = __ext4_find_entry(dir, &fname, res_dir, NULL); 1691 1692 ext4_fname_free_filename(&fname); 1693 return bh; 1694 } 1695 1696 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, 1697 struct ext4_filename *fname, 1698 struct ext4_dir_entry_2 **res_dir) 1699 { 1700 struct super_block * sb = dir->i_sb; 1701 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; 1702 struct buffer_head *bh; 1703 ext4_lblk_t block; 1704 int retval; 1705 1706 #ifdef CONFIG_FS_ENCRYPTION 1707 *res_dir = NULL; 1708 #endif 1709 frame = dx_probe(fname, dir, NULL, frames); 1710 if (IS_ERR(frame)) 1711 return (struct buffer_head *) frame; 1712 do { 1713 block = dx_get_block(frame->at); 1714 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE); 1715 if (IS_ERR(bh)) 1716 goto errout; 1717 1718 retval = search_dirblock(bh, dir, fname, 1719 block << EXT4_BLOCK_SIZE_BITS(sb), 1720 res_dir); 1721 if (retval == 1) 1722 goto success; 1723 brelse(bh); 1724 if (retval == -1) { 1725 bh = ERR_PTR(ERR_BAD_DX_DIR); 1726 goto errout; 1727 } 1728 1729 /* Check to see if we should continue to search */ 1730 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame, 1731 frames, NULL); 1732 if (retval < 0) { 1733 ext4_warning_inode(dir, 1734 "error %d reading directory index block", 1735 retval); 1736 bh = ERR_PTR(retval); 1737 goto errout; 1738 } 1739 } while (retval == 1); 1740 1741 bh = NULL; 1742 errout: 1743 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name)); 1744 success: 1745 dx_release(frames); 1746 return bh; 1747 } 1748 1749 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) 1750 { 1751 struct inode *inode; 1752 struct ext4_dir_entry_2 *de; 1753 struct buffer_head *bh; 1754 1755 if (dentry->d_name.len > EXT4_NAME_LEN) 1756 return ERR_PTR(-ENAMETOOLONG); 1757 1758 bh = ext4_lookup_entry(dir, dentry, &de); 1759 if (IS_ERR(bh)) 1760 return ERR_CAST(bh); 1761 inode = NULL; 1762 if (bh) { 1763 __u32 ino = le32_to_cpu(de->inode); 1764 brelse(bh); 1765 if (!ext4_valid_inum(dir->i_sb, ino)) { 1766 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino); 1767 return ERR_PTR(-EFSCORRUPTED); 1768 } 1769 if (unlikely(ino == dir->i_ino)) { 1770 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir", 1771 dentry); 1772 return ERR_PTR(-EFSCORRUPTED); 1773 } 1774 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL); 1775 if (inode == ERR_PTR(-ESTALE)) { 1776 EXT4_ERROR_INODE(dir, 1777 "deleted inode referenced: %u", 1778 ino); 1779 return ERR_PTR(-EFSCORRUPTED); 1780 } 1781 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) && 1782 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) && 1783 !fscrypt_has_permitted_context(dir, inode)) { 1784 ext4_warning(inode->i_sb, 1785 "Inconsistent encryption contexts: %lu/%lu", 1786 dir->i_ino, inode->i_ino); 1787 iput(inode); 1788 return ERR_PTR(-EPERM); 1789 } 1790 } 1791 1792 #ifdef CONFIG_UNICODE 1793 if (!inode && IS_CASEFOLDED(dir)) { 1794 /* Eventually we want to call d_add_ci(dentry, NULL) 1795 * for negative dentries in the encoding case as 1796 * well. For now, prevent the negative dentry 1797 * from being cached. 1798 */ 1799 return NULL; 1800 } 1801 #endif 1802 return d_splice_alias(inode, dentry); 1803 } 1804 1805 1806 struct dentry *ext4_get_parent(struct dentry *child) 1807 { 1808 __u32 ino; 1809 static const struct qstr dotdot = QSTR_INIT("..", 2); 1810 struct ext4_dir_entry_2 * de; 1811 struct buffer_head *bh; 1812 1813 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL); 1814 if (IS_ERR(bh)) 1815 return ERR_CAST(bh); 1816 if (!bh) 1817 return ERR_PTR(-ENOENT); 1818 ino = le32_to_cpu(de->inode); 1819 brelse(bh); 1820 1821 if (!ext4_valid_inum(child->d_sb, ino)) { 1822 EXT4_ERROR_INODE(d_inode(child), 1823 "bad parent inode number: %u", ino); 1824 return ERR_PTR(-EFSCORRUPTED); 1825 } 1826 1827 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL)); 1828 } 1829 1830 /* 1831 * Move count entries from end of map between two memory locations. 1832 * Returns pointer to last entry moved. 1833 */ 1834 static struct ext4_dir_entry_2 * 1835 dx_move_dirents(struct inode *dir, char *from, char *to, 1836 struct dx_map_entry *map, int count, 1837 unsigned blocksize) 1838 { 1839 unsigned rec_len = 0; 1840 1841 while (count--) { 1842 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) 1843 (from + (map->offs<<2)); 1844 rec_len = ext4_dir_rec_len(de->name_len, dir); 1845 1846 memcpy (to, de, rec_len); 1847 ((struct ext4_dir_entry_2 *) to)->rec_len = 1848 ext4_rec_len_to_disk(rec_len, blocksize); 1849 de->inode = 0; 1850 map++; 1851 to += rec_len; 1852 } 1853 return (struct ext4_dir_entry_2 *) (to - rec_len); 1854 } 1855 1856 /* 1857 * Compact each dir entry in the range to the minimal rec_len. 1858 * Returns pointer to last entry in range. 1859 */ 1860 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base, 1861 unsigned int blocksize) 1862 { 1863 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base; 1864 unsigned rec_len = 0; 1865 1866 prev = to = de; 1867 while ((char*)de < base + blocksize) { 1868 next = ext4_next_entry(de, blocksize); 1869 if (de->inode && de->name_len) { 1870 rec_len = ext4_dir_rec_len(de->name_len, dir); 1871 if (de > to) 1872 memmove(to, de, rec_len); 1873 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize); 1874 prev = to; 1875 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len); 1876 } 1877 de = next; 1878 } 1879 return prev; 1880 } 1881 1882 /* 1883 * Split a full leaf block to make room for a new dir entry. 1884 * Allocate a new block, and move entries so that they are approx. equally full. 1885 * Returns pointer to de in block into which the new entry will be inserted. 1886 */ 1887 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, 1888 struct buffer_head **bh,struct dx_frame *frame, 1889 struct dx_hash_info *hinfo) 1890 { 1891 unsigned blocksize = dir->i_sb->s_blocksize; 1892 unsigned count, continued; 1893 struct buffer_head *bh2; 1894 ext4_lblk_t newblock; 1895 u32 hash2; 1896 struct dx_map_entry *map; 1897 char *data1 = (*bh)->b_data, *data2; 1898 unsigned split, move, size; 1899 struct ext4_dir_entry_2 *de = NULL, *de2; 1900 int csum_size = 0; 1901 int err = 0, i; 1902 1903 if (ext4_has_metadata_csum(dir->i_sb)) 1904 csum_size = sizeof(struct ext4_dir_entry_tail); 1905 1906 bh2 = ext4_append(handle, dir, &newblock); 1907 if (IS_ERR(bh2)) { 1908 brelse(*bh); 1909 *bh = NULL; 1910 return (struct ext4_dir_entry_2 *) bh2; 1911 } 1912 1913 BUFFER_TRACE(*bh, "get_write_access"); 1914 err = ext4_journal_get_write_access(handle, *bh); 1915 if (err) 1916 goto journal_error; 1917 1918 BUFFER_TRACE(frame->bh, "get_write_access"); 1919 err = ext4_journal_get_write_access(handle, frame->bh); 1920 if (err) 1921 goto journal_error; 1922 1923 data2 = bh2->b_data; 1924 1925 /* create map in the end of data2 block */ 1926 map = (struct dx_map_entry *) (data2 + blocksize); 1927 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1, 1928 blocksize, hinfo, map); 1929 map -= count; 1930 dx_sort_map(map, count); 1931 /* Ensure that neither split block is over half full */ 1932 size = 0; 1933 move = 0; 1934 for (i = count-1; i >= 0; i--) { 1935 /* is more than half of this entry in 2nd half of the block? */ 1936 if (size + map[i].size/2 > blocksize/2) 1937 break; 1938 size += map[i].size; 1939 move++; 1940 } 1941 /* 1942 * map index at which we will split 1943 * 1944 * If the sum of active entries didn't exceed half the block size, just 1945 * split it in half by count; each resulting block will have at least 1946 * half the space free. 1947 */ 1948 if (i > 0) 1949 split = count - move; 1950 else 1951 split = count/2; 1952 1953 hash2 = map[split].hash; 1954 continued = hash2 == map[split - 1].hash; 1955 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n", 1956 (unsigned long)dx_get_block(frame->at), 1957 hash2, split, count-split)); 1958 1959 /* Fancy dance to stay within two buffers */ 1960 de2 = dx_move_dirents(dir, data1, data2, map + split, count - split, 1961 blocksize); 1962 de = dx_pack_dirents(dir, data1, blocksize); 1963 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) - 1964 (char *) de, 1965 blocksize); 1966 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) - 1967 (char *) de2, 1968 blocksize); 1969 if (csum_size) { 1970 ext4_initialize_dirent_tail(*bh, blocksize); 1971 ext4_initialize_dirent_tail(bh2, blocksize); 1972 } 1973 1974 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1, 1975 blocksize, 1)); 1976 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2, 1977 blocksize, 1)); 1978 1979 /* Which block gets the new entry? */ 1980 if (hinfo->hash >= hash2) { 1981 swap(*bh, bh2); 1982 de = de2; 1983 } 1984 dx_insert_block(frame, hash2 + continued, newblock); 1985 err = ext4_handle_dirty_dirblock(handle, dir, bh2); 1986 if (err) 1987 goto journal_error; 1988 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh); 1989 if (err) 1990 goto journal_error; 1991 brelse(bh2); 1992 dxtrace(dx_show_index("frame", frame->entries)); 1993 return de; 1994 1995 journal_error: 1996 brelse(*bh); 1997 brelse(bh2); 1998 *bh = NULL; 1999 ext4_std_error(dir->i_sb, err); 2000 return ERR_PTR(err); 2001 } 2002 2003 int ext4_find_dest_de(struct inode *dir, struct inode *inode, 2004 struct buffer_head *bh, 2005 void *buf, int buf_size, 2006 struct ext4_filename *fname, 2007 struct ext4_dir_entry_2 **dest_de) 2008 { 2009 struct ext4_dir_entry_2 *de; 2010 unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir); 2011 int nlen, rlen; 2012 unsigned int offset = 0; 2013 char *top; 2014 2015 de = (struct ext4_dir_entry_2 *)buf; 2016 top = buf + buf_size - reclen; 2017 while ((char *) de <= top) { 2018 if (ext4_check_dir_entry(dir, NULL, de, bh, 2019 buf, buf_size, offset)) 2020 return -EFSCORRUPTED; 2021 if (ext4_match(dir, fname, de)) 2022 return -EEXIST; 2023 nlen = ext4_dir_rec_len(de->name_len, dir); 2024 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 2025 if ((de->inode ? rlen - nlen : rlen) >= reclen) 2026 break; 2027 de = (struct ext4_dir_entry_2 *)((char *)de + rlen); 2028 offset += rlen; 2029 } 2030 if ((char *) de > top) 2031 return -ENOSPC; 2032 2033 *dest_de = de; 2034 return 0; 2035 } 2036 2037 void ext4_insert_dentry(struct inode *dir, 2038 struct inode *inode, 2039 struct ext4_dir_entry_2 *de, 2040 int buf_size, 2041 struct ext4_filename *fname) 2042 { 2043 2044 int nlen, rlen; 2045 2046 nlen = ext4_dir_rec_len(de->name_len, dir); 2047 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 2048 if (de->inode) { 2049 struct ext4_dir_entry_2 *de1 = 2050 (struct ext4_dir_entry_2 *)((char *)de + nlen); 2051 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size); 2052 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size); 2053 de = de1; 2054 } 2055 de->file_type = EXT4_FT_UNKNOWN; 2056 de->inode = cpu_to_le32(inode->i_ino); 2057 ext4_set_de_type(inode->i_sb, de, inode->i_mode); 2058 de->name_len = fname_len(fname); 2059 memcpy(de->name, fname_name(fname), fname_len(fname)); 2060 if (ext4_hash_in_dirent(dir)) { 2061 struct dx_hash_info hinfo; 2062 2063 hinfo.hash_version = DX_HASH_SIPHASH; 2064 hinfo.seed = NULL; 2065 ext4fs_dirhash(dir, fname_usr_name(fname), 2066 fname_len(fname), &hinfo); 2067 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo.hash); 2068 EXT4_DIRENT_HASHES(de)->minor_hash = 2069 cpu_to_le32(hinfo.minor_hash); 2070 } 2071 } 2072 2073 /* 2074 * Add a new entry into a directory (leaf) block. If de is non-NULL, 2075 * it points to a directory entry which is guaranteed to be large 2076 * enough for new directory entry. If de is NULL, then 2077 * add_dirent_to_buf will attempt search the directory block for 2078 * space. It will return -ENOSPC if no space is available, and -EIO 2079 * and -EEXIST if directory entry already exists. 2080 */ 2081 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname, 2082 struct inode *dir, 2083 struct inode *inode, struct ext4_dir_entry_2 *de, 2084 struct buffer_head *bh) 2085 { 2086 unsigned int blocksize = dir->i_sb->s_blocksize; 2087 int csum_size = 0; 2088 int err, err2; 2089 2090 if (ext4_has_metadata_csum(inode->i_sb)) 2091 csum_size = sizeof(struct ext4_dir_entry_tail); 2092 2093 if (!de) { 2094 err = ext4_find_dest_de(dir, inode, bh, bh->b_data, 2095 blocksize - csum_size, fname, &de); 2096 if (err) 2097 return err; 2098 } 2099 BUFFER_TRACE(bh, "get_write_access"); 2100 err = ext4_journal_get_write_access(handle, bh); 2101 if (err) { 2102 ext4_std_error(dir->i_sb, err); 2103 return err; 2104 } 2105 2106 /* By now the buffer is marked for journaling */ 2107 ext4_insert_dentry(dir, inode, de, blocksize, fname); 2108 2109 /* 2110 * XXX shouldn't update any times until successful 2111 * completion of syscall, but too many callers depend 2112 * on this. 2113 * 2114 * XXX similarly, too many callers depend on 2115 * ext4_new_inode() setting the times, but error 2116 * recovery deletes the inode, so the worst that can 2117 * happen is that the times are slightly out of date 2118 * and/or different from the directory change time. 2119 */ 2120 dir->i_mtime = dir->i_ctime = current_time(dir); 2121 ext4_update_dx_flag(dir); 2122 inode_inc_iversion(dir); 2123 err2 = ext4_mark_inode_dirty(handle, dir); 2124 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 2125 err = ext4_handle_dirty_dirblock(handle, dir, bh); 2126 if (err) 2127 ext4_std_error(dir->i_sb, err); 2128 return err ? err : err2; 2129 } 2130 2131 /* 2132 * This converts a one block unindexed directory to a 3 block indexed 2133 * directory, and adds the dentry to the indexed directory. 2134 */ 2135 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname, 2136 struct inode *dir, 2137 struct inode *inode, struct buffer_head *bh) 2138 { 2139 struct buffer_head *bh2; 2140 struct dx_root *root; 2141 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; 2142 struct dx_entry *entries; 2143 struct ext4_dir_entry_2 *de, *de2; 2144 char *data2, *top; 2145 unsigned len; 2146 int retval; 2147 unsigned blocksize; 2148 ext4_lblk_t block; 2149 struct fake_dirent *fde; 2150 int csum_size = 0; 2151 2152 if (ext4_has_metadata_csum(inode->i_sb)) 2153 csum_size = sizeof(struct ext4_dir_entry_tail); 2154 2155 blocksize = dir->i_sb->s_blocksize; 2156 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino)); 2157 BUFFER_TRACE(bh, "get_write_access"); 2158 retval = ext4_journal_get_write_access(handle, bh); 2159 if (retval) { 2160 ext4_std_error(dir->i_sb, retval); 2161 brelse(bh); 2162 return retval; 2163 } 2164 root = (struct dx_root *) bh->b_data; 2165 2166 /* The 0th block becomes the root, move the dirents out */ 2167 fde = &root->dotdot; 2168 de = (struct ext4_dir_entry_2 *)((char *)fde + 2169 ext4_rec_len_from_disk(fde->rec_len, blocksize)); 2170 if ((char *) de >= (((char *) root) + blocksize)) { 2171 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'"); 2172 brelse(bh); 2173 return -EFSCORRUPTED; 2174 } 2175 len = ((char *) root) + (blocksize - csum_size) - (char *) de; 2176 2177 /* Allocate new block for the 0th block's dirents */ 2178 bh2 = ext4_append(handle, dir, &block); 2179 if (IS_ERR(bh2)) { 2180 brelse(bh); 2181 return PTR_ERR(bh2); 2182 } 2183 ext4_set_inode_flag(dir, EXT4_INODE_INDEX); 2184 data2 = bh2->b_data; 2185 2186 memcpy(data2, de, len); 2187 de = (struct ext4_dir_entry_2 *) data2; 2188 top = data2 + len; 2189 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) 2190 de = de2; 2191 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) - 2192 (char *) de, blocksize); 2193 2194 if (csum_size) 2195 ext4_initialize_dirent_tail(bh2, blocksize); 2196 2197 /* Initialize the root; the dot dirents already exist */ 2198 de = (struct ext4_dir_entry_2 *) (&root->dotdot); 2199 de->rec_len = ext4_rec_len_to_disk( 2200 blocksize - ext4_dir_rec_len(2, NULL), blocksize); 2201 memset (&root->info, 0, sizeof(root->info)); 2202 root->info.info_length = sizeof(root->info); 2203 if (ext4_hash_in_dirent(dir)) 2204 root->info.hash_version = DX_HASH_SIPHASH; 2205 else 2206 root->info.hash_version = 2207 EXT4_SB(dir->i_sb)->s_def_hash_version; 2208 2209 entries = root->entries; 2210 dx_set_block(entries, 1); 2211 dx_set_count(entries, 1); 2212 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info))); 2213 2214 /* Initialize as for dx_probe */ 2215 fname->hinfo.hash_version = root->info.hash_version; 2216 if (fname->hinfo.hash_version <= DX_HASH_TEA) 2217 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; 2218 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; 2219 if (ext4_hash_in_dirent(dir)) 2220 ext4fs_dirhash(dir, fname_usr_name(fname), 2221 fname_len(fname), &fname->hinfo); 2222 else 2223 ext4fs_dirhash(dir, fname_name(fname), 2224 fname_len(fname), &fname->hinfo); 2225 2226 memset(frames, 0, sizeof(frames)); 2227 frame = frames; 2228 frame->entries = entries; 2229 frame->at = entries; 2230 frame->bh = bh; 2231 2232 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh); 2233 if (retval) 2234 goto out_frames; 2235 retval = ext4_handle_dirty_dirblock(handle, dir, bh2); 2236 if (retval) 2237 goto out_frames; 2238 2239 de = do_split(handle,dir, &bh2, frame, &fname->hinfo); 2240 if (IS_ERR(de)) { 2241 retval = PTR_ERR(de); 2242 goto out_frames; 2243 } 2244 2245 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2); 2246 out_frames: 2247 /* 2248 * Even if the block split failed, we have to properly write 2249 * out all the changes we did so far. Otherwise we can end up 2250 * with corrupted filesystem. 2251 */ 2252 if (retval) 2253 ext4_mark_inode_dirty(handle, dir); 2254 dx_release(frames); 2255 brelse(bh2); 2256 return retval; 2257 } 2258 2259 /* 2260 * ext4_add_entry() 2261 * 2262 * adds a file entry to the specified directory, using the same 2263 * semantics as ext4_find_entry(). It returns NULL if it failed. 2264 * 2265 * NOTE!! The inode part of 'de' is left at 0 - which means you 2266 * may not sleep between calling this and putting something into 2267 * the entry, as someone else might have used it while you slept. 2268 */ 2269 static int ext4_add_entry(handle_t *handle, struct dentry *dentry, 2270 struct inode *inode) 2271 { 2272 struct inode *dir = d_inode(dentry->d_parent); 2273 struct buffer_head *bh = NULL; 2274 struct ext4_dir_entry_2 *de; 2275 struct super_block *sb; 2276 struct ext4_filename fname; 2277 int retval; 2278 int dx_fallback=0; 2279 unsigned blocksize; 2280 ext4_lblk_t block, blocks; 2281 int csum_size = 0; 2282 2283 if (ext4_has_metadata_csum(inode->i_sb)) 2284 csum_size = sizeof(struct ext4_dir_entry_tail); 2285 2286 sb = dir->i_sb; 2287 blocksize = sb->s_blocksize; 2288 if (!dentry->d_name.len) 2289 return -EINVAL; 2290 2291 if (fscrypt_is_nokey_name(dentry)) 2292 return -ENOKEY; 2293 2294 #ifdef CONFIG_UNICODE 2295 if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) && 2296 sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name)) 2297 return -EINVAL; 2298 #endif 2299 2300 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname); 2301 if (retval) 2302 return retval; 2303 2304 if (ext4_has_inline_data(dir)) { 2305 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode); 2306 if (retval < 0) 2307 goto out; 2308 if (retval == 1) { 2309 retval = 0; 2310 goto out; 2311 } 2312 } 2313 2314 if (is_dx(dir)) { 2315 retval = ext4_dx_add_entry(handle, &fname, dir, inode); 2316 if (!retval || (retval != ERR_BAD_DX_DIR)) 2317 goto out; 2318 /* Can we just ignore htree data? */ 2319 if (ext4_has_metadata_csum(sb)) { 2320 EXT4_ERROR_INODE(dir, 2321 "Directory has corrupted htree index."); 2322 retval = -EFSCORRUPTED; 2323 goto out; 2324 } 2325 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX); 2326 dx_fallback++; 2327 retval = ext4_mark_inode_dirty(handle, dir); 2328 if (unlikely(retval)) 2329 goto out; 2330 } 2331 blocks = dir->i_size >> sb->s_blocksize_bits; 2332 for (block = 0; block < blocks; block++) { 2333 bh = ext4_read_dirblock(dir, block, DIRENT); 2334 if (bh == NULL) { 2335 bh = ext4_bread(handle, dir, block, 2336 EXT4_GET_BLOCKS_CREATE); 2337 goto add_to_new_block; 2338 } 2339 if (IS_ERR(bh)) { 2340 retval = PTR_ERR(bh); 2341 bh = NULL; 2342 goto out; 2343 } 2344 retval = add_dirent_to_buf(handle, &fname, dir, inode, 2345 NULL, bh); 2346 if (retval != -ENOSPC) 2347 goto out; 2348 2349 if (blocks == 1 && !dx_fallback && 2350 ext4_has_feature_dir_index(sb)) { 2351 retval = make_indexed_dir(handle, &fname, dir, 2352 inode, bh); 2353 bh = NULL; /* make_indexed_dir releases bh */ 2354 goto out; 2355 } 2356 brelse(bh); 2357 } 2358 bh = ext4_append(handle, dir, &block); 2359 add_to_new_block: 2360 if (IS_ERR(bh)) { 2361 retval = PTR_ERR(bh); 2362 bh = NULL; 2363 goto out; 2364 } 2365 de = (struct ext4_dir_entry_2 *) bh->b_data; 2366 de->inode = 0; 2367 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize); 2368 2369 if (csum_size) 2370 ext4_initialize_dirent_tail(bh, blocksize); 2371 2372 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh); 2373 out: 2374 ext4_fname_free_filename(&fname); 2375 brelse(bh); 2376 if (retval == 0) 2377 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY); 2378 return retval; 2379 } 2380 2381 /* 2382 * Returns 0 for success, or a negative error value 2383 */ 2384 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname, 2385 struct inode *dir, struct inode *inode) 2386 { 2387 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame; 2388 struct dx_entry *entries, *at; 2389 struct buffer_head *bh; 2390 struct super_block *sb = dir->i_sb; 2391 struct ext4_dir_entry_2 *de; 2392 int restart; 2393 int err; 2394 2395 again: 2396 restart = 0; 2397 frame = dx_probe(fname, dir, NULL, frames); 2398 if (IS_ERR(frame)) 2399 return PTR_ERR(frame); 2400 entries = frame->entries; 2401 at = frame->at; 2402 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE); 2403 if (IS_ERR(bh)) { 2404 err = PTR_ERR(bh); 2405 bh = NULL; 2406 goto cleanup; 2407 } 2408 2409 BUFFER_TRACE(bh, "get_write_access"); 2410 err = ext4_journal_get_write_access(handle, bh); 2411 if (err) 2412 goto journal_error; 2413 2414 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh); 2415 if (err != -ENOSPC) 2416 goto cleanup; 2417 2418 err = 0; 2419 /* Block full, should compress but for now just split */ 2420 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n", 2421 dx_get_count(entries), dx_get_limit(entries))); 2422 /* Need to split index? */ 2423 if (dx_get_count(entries) == dx_get_limit(entries)) { 2424 ext4_lblk_t newblock; 2425 int levels = frame - frames + 1; 2426 unsigned int icount; 2427 int add_level = 1; 2428 struct dx_entry *entries2; 2429 struct dx_node *node2; 2430 struct buffer_head *bh2; 2431 2432 while (frame > frames) { 2433 if (dx_get_count((frame - 1)->entries) < 2434 dx_get_limit((frame - 1)->entries)) { 2435 add_level = 0; 2436 break; 2437 } 2438 frame--; /* split higher index block */ 2439 at = frame->at; 2440 entries = frame->entries; 2441 restart = 1; 2442 } 2443 if (add_level && levels == ext4_dir_htree_level(sb)) { 2444 ext4_warning(sb, "Directory (ino: %lu) index full, " 2445 "reach max htree level :%d", 2446 dir->i_ino, levels); 2447 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) { 2448 ext4_warning(sb, "Large directory feature is " 2449 "not enabled on this " 2450 "filesystem"); 2451 } 2452 err = -ENOSPC; 2453 goto cleanup; 2454 } 2455 icount = dx_get_count(entries); 2456 bh2 = ext4_append(handle, dir, &newblock); 2457 if (IS_ERR(bh2)) { 2458 err = PTR_ERR(bh2); 2459 goto cleanup; 2460 } 2461 node2 = (struct dx_node *)(bh2->b_data); 2462 entries2 = node2->entries; 2463 memset(&node2->fake, 0, sizeof(struct fake_dirent)); 2464 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize, 2465 sb->s_blocksize); 2466 BUFFER_TRACE(frame->bh, "get_write_access"); 2467 err = ext4_journal_get_write_access(handle, frame->bh); 2468 if (err) 2469 goto journal_error; 2470 if (!add_level) { 2471 unsigned icount1 = icount/2, icount2 = icount - icount1; 2472 unsigned hash2 = dx_get_hash(entries + icount1); 2473 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n", 2474 icount1, icount2)); 2475 2476 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */ 2477 err = ext4_journal_get_write_access(handle, 2478 (frame - 1)->bh); 2479 if (err) 2480 goto journal_error; 2481 2482 memcpy((char *) entries2, (char *) (entries + icount1), 2483 icount2 * sizeof(struct dx_entry)); 2484 dx_set_count(entries, icount1); 2485 dx_set_count(entries2, icount2); 2486 dx_set_limit(entries2, dx_node_limit(dir)); 2487 2488 /* Which index block gets the new entry? */ 2489 if (at - entries >= icount1) { 2490 frame->at = at = at - entries - icount1 + entries2; 2491 frame->entries = entries = entries2; 2492 swap(frame->bh, bh2); 2493 } 2494 dx_insert_block((frame - 1), hash2, newblock); 2495 dxtrace(dx_show_index("node", frame->entries)); 2496 dxtrace(dx_show_index("node", 2497 ((struct dx_node *) bh2->b_data)->entries)); 2498 err = ext4_handle_dirty_dx_node(handle, dir, bh2); 2499 if (err) 2500 goto journal_error; 2501 brelse (bh2); 2502 err = ext4_handle_dirty_dx_node(handle, dir, 2503 (frame - 1)->bh); 2504 if (err) 2505 goto journal_error; 2506 err = ext4_handle_dirty_dx_node(handle, dir, 2507 frame->bh); 2508 if (err) 2509 goto journal_error; 2510 } else { 2511 struct dx_root *dxroot; 2512 memcpy((char *) entries2, (char *) entries, 2513 icount * sizeof(struct dx_entry)); 2514 dx_set_limit(entries2, dx_node_limit(dir)); 2515 2516 /* Set up root */ 2517 dx_set_count(entries, 1); 2518 dx_set_block(entries + 0, newblock); 2519 dxroot = (struct dx_root *)frames[0].bh->b_data; 2520 dxroot->info.indirect_levels += 1; 2521 dxtrace(printk(KERN_DEBUG 2522 "Creating %d level index...\n", 2523 dxroot->info.indirect_levels)); 2524 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh); 2525 if (err) 2526 goto journal_error; 2527 err = ext4_handle_dirty_dx_node(handle, dir, bh2); 2528 brelse(bh2); 2529 restart = 1; 2530 goto journal_error; 2531 } 2532 } 2533 de = do_split(handle, dir, &bh, frame, &fname->hinfo); 2534 if (IS_ERR(de)) { 2535 err = PTR_ERR(de); 2536 goto cleanup; 2537 } 2538 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh); 2539 goto cleanup; 2540 2541 journal_error: 2542 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */ 2543 cleanup: 2544 brelse(bh); 2545 dx_release(frames); 2546 /* @restart is true means htree-path has been changed, we need to 2547 * repeat dx_probe() to find out valid htree-path 2548 */ 2549 if (restart && err == 0) 2550 goto again; 2551 return err; 2552 } 2553 2554 /* 2555 * ext4_generic_delete_entry deletes a directory entry by merging it 2556 * with the previous entry 2557 */ 2558 int ext4_generic_delete_entry(struct inode *dir, 2559 struct ext4_dir_entry_2 *de_del, 2560 struct buffer_head *bh, 2561 void *entry_buf, 2562 int buf_size, 2563 int csum_size) 2564 { 2565 struct ext4_dir_entry_2 *de, *pde; 2566 unsigned int blocksize = dir->i_sb->s_blocksize; 2567 int i; 2568 2569 i = 0; 2570 pde = NULL; 2571 de = (struct ext4_dir_entry_2 *)entry_buf; 2572 while (i < buf_size - csum_size) { 2573 if (ext4_check_dir_entry(dir, NULL, de, bh, 2574 entry_buf, buf_size, i)) 2575 return -EFSCORRUPTED; 2576 if (de == de_del) { 2577 if (pde) 2578 pde->rec_len = ext4_rec_len_to_disk( 2579 ext4_rec_len_from_disk(pde->rec_len, 2580 blocksize) + 2581 ext4_rec_len_from_disk(de->rec_len, 2582 blocksize), 2583 blocksize); 2584 else 2585 de->inode = 0; 2586 inode_inc_iversion(dir); 2587 return 0; 2588 } 2589 i += ext4_rec_len_from_disk(de->rec_len, blocksize); 2590 pde = de; 2591 de = ext4_next_entry(de, blocksize); 2592 } 2593 return -ENOENT; 2594 } 2595 2596 static int ext4_delete_entry(handle_t *handle, 2597 struct inode *dir, 2598 struct ext4_dir_entry_2 *de_del, 2599 struct buffer_head *bh) 2600 { 2601 int err, csum_size = 0; 2602 2603 if (ext4_has_inline_data(dir)) { 2604 int has_inline_data = 1; 2605 err = ext4_delete_inline_entry(handle, dir, de_del, bh, 2606 &has_inline_data); 2607 if (has_inline_data) 2608 return err; 2609 } 2610 2611 if (ext4_has_metadata_csum(dir->i_sb)) 2612 csum_size = sizeof(struct ext4_dir_entry_tail); 2613 2614 BUFFER_TRACE(bh, "get_write_access"); 2615 err = ext4_journal_get_write_access(handle, bh); 2616 if (unlikely(err)) 2617 goto out; 2618 2619 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data, 2620 dir->i_sb->s_blocksize, csum_size); 2621 if (err) 2622 goto out; 2623 2624 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 2625 err = ext4_handle_dirty_dirblock(handle, dir, bh); 2626 if (unlikely(err)) 2627 goto out; 2628 2629 return 0; 2630 out: 2631 if (err != -ENOENT) 2632 ext4_std_error(dir->i_sb, err); 2633 return err; 2634 } 2635 2636 /* 2637 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2 2638 * since this indicates that nlinks count was previously 1 to avoid overflowing 2639 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean 2640 * that subdirectory link counts are not being maintained accurately. 2641 * 2642 * The caller has already checked for i_nlink overflow in case the DIR_LINK 2643 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy 2644 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set 2645 * on regular files) and to avoid creating huge/slow non-HTREE directories. 2646 */ 2647 static void ext4_inc_count(struct inode *inode) 2648 { 2649 inc_nlink(inode); 2650 if (is_dx(inode) && 2651 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2)) 2652 set_nlink(inode, 1); 2653 } 2654 2655 /* 2656 * If a directory had nlink == 1, then we should let it be 1. This indicates 2657 * directory has >EXT4_LINK_MAX subdirs. 2658 */ 2659 static void ext4_dec_count(struct inode *inode) 2660 { 2661 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2) 2662 drop_nlink(inode); 2663 } 2664 2665 2666 /* 2667 * Add non-directory inode to a directory. On success, the inode reference is 2668 * consumed by dentry is instantiation. This is also indicated by clearing of 2669 * *inodep pointer. On failure, the caller is responsible for dropping the 2670 * inode reference in the safe context. 2671 */ 2672 static int ext4_add_nondir(handle_t *handle, 2673 struct dentry *dentry, struct inode **inodep) 2674 { 2675 struct inode *dir = d_inode(dentry->d_parent); 2676 struct inode *inode = *inodep; 2677 int err = ext4_add_entry(handle, dentry, inode); 2678 if (!err) { 2679 err = ext4_mark_inode_dirty(handle, inode); 2680 if (IS_DIRSYNC(dir)) 2681 ext4_handle_sync(handle); 2682 d_instantiate_new(dentry, inode); 2683 *inodep = NULL; 2684 return err; 2685 } 2686 drop_nlink(inode); 2687 ext4_orphan_add(handle, inode); 2688 unlock_new_inode(inode); 2689 return err; 2690 } 2691 2692 /* 2693 * By the time this is called, we already have created 2694 * the directory cache entry for the new file, but it 2695 * is so far negative - it has no inode. 2696 * 2697 * If the create succeeds, we fill in the inode information 2698 * with d_instantiate(). 2699 */ 2700 static int ext4_create(struct user_namespace *mnt_userns, struct inode *dir, 2701 struct dentry *dentry, umode_t mode, bool excl) 2702 { 2703 handle_t *handle; 2704 struct inode *inode; 2705 int err, credits, retries = 0; 2706 2707 err = dquot_initialize(dir); 2708 if (err) 2709 return err; 2710 2711 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 2712 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); 2713 retry: 2714 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name, 2715 0, NULL, EXT4_HT_DIR, credits); 2716 handle = ext4_journal_current_handle(); 2717 err = PTR_ERR(inode); 2718 if (!IS_ERR(inode)) { 2719 inode->i_op = &ext4_file_inode_operations; 2720 inode->i_fop = &ext4_file_operations; 2721 ext4_set_aops(inode); 2722 err = ext4_add_nondir(handle, dentry, &inode); 2723 if (!err) 2724 ext4_fc_track_create(handle, dentry); 2725 } 2726 if (handle) 2727 ext4_journal_stop(handle); 2728 if (!IS_ERR_OR_NULL(inode)) 2729 iput(inode); 2730 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2731 goto retry; 2732 return err; 2733 } 2734 2735 static int ext4_mknod(struct user_namespace *mnt_userns, struct inode *dir, 2736 struct dentry *dentry, umode_t mode, dev_t rdev) 2737 { 2738 handle_t *handle; 2739 struct inode *inode; 2740 int err, credits, retries = 0; 2741 2742 err = dquot_initialize(dir); 2743 if (err) 2744 return err; 2745 2746 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 2747 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); 2748 retry: 2749 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name, 2750 0, NULL, EXT4_HT_DIR, credits); 2751 handle = ext4_journal_current_handle(); 2752 err = PTR_ERR(inode); 2753 if (!IS_ERR(inode)) { 2754 init_special_inode(inode, inode->i_mode, rdev); 2755 inode->i_op = &ext4_special_inode_operations; 2756 err = ext4_add_nondir(handle, dentry, &inode); 2757 if (!err) 2758 ext4_fc_track_create(handle, dentry); 2759 } 2760 if (handle) 2761 ext4_journal_stop(handle); 2762 if (!IS_ERR_OR_NULL(inode)) 2763 iput(inode); 2764 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2765 goto retry; 2766 return err; 2767 } 2768 2769 static int ext4_tmpfile(struct user_namespace *mnt_userns, struct inode *dir, 2770 struct dentry *dentry, umode_t mode) 2771 { 2772 handle_t *handle; 2773 struct inode *inode; 2774 int err, retries = 0; 2775 2776 err = dquot_initialize(dir); 2777 if (err) 2778 return err; 2779 2780 retry: 2781 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, 2782 NULL, 0, NULL, 2783 EXT4_HT_DIR, 2784 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) + 2785 4 + EXT4_XATTR_TRANS_BLOCKS); 2786 handle = ext4_journal_current_handle(); 2787 err = PTR_ERR(inode); 2788 if (!IS_ERR(inode)) { 2789 inode->i_op = &ext4_file_inode_operations; 2790 inode->i_fop = &ext4_file_operations; 2791 ext4_set_aops(inode); 2792 d_tmpfile(dentry, inode); 2793 err = ext4_orphan_add(handle, inode); 2794 if (err) 2795 goto err_unlock_inode; 2796 mark_inode_dirty(inode); 2797 unlock_new_inode(inode); 2798 } 2799 if (handle) 2800 ext4_journal_stop(handle); 2801 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2802 goto retry; 2803 return err; 2804 err_unlock_inode: 2805 ext4_journal_stop(handle); 2806 unlock_new_inode(inode); 2807 return err; 2808 } 2809 2810 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode, 2811 struct ext4_dir_entry_2 *de, 2812 int blocksize, int csum_size, 2813 unsigned int parent_ino, int dotdot_real_len) 2814 { 2815 de->inode = cpu_to_le32(inode->i_ino); 2816 de->name_len = 1; 2817 de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL), 2818 blocksize); 2819 strcpy(de->name, "."); 2820 ext4_set_de_type(inode->i_sb, de, S_IFDIR); 2821 2822 de = ext4_next_entry(de, blocksize); 2823 de->inode = cpu_to_le32(parent_ino); 2824 de->name_len = 2; 2825 if (!dotdot_real_len) 2826 de->rec_len = ext4_rec_len_to_disk(blocksize - 2827 (csum_size + ext4_dir_rec_len(1, NULL)), 2828 blocksize); 2829 else 2830 de->rec_len = ext4_rec_len_to_disk( 2831 ext4_dir_rec_len(de->name_len, NULL), 2832 blocksize); 2833 strcpy(de->name, ".."); 2834 ext4_set_de_type(inode->i_sb, de, S_IFDIR); 2835 2836 return ext4_next_entry(de, blocksize); 2837 } 2838 2839 int ext4_init_new_dir(handle_t *handle, struct inode *dir, 2840 struct inode *inode) 2841 { 2842 struct buffer_head *dir_block = NULL; 2843 struct ext4_dir_entry_2 *de; 2844 ext4_lblk_t block = 0; 2845 unsigned int blocksize = dir->i_sb->s_blocksize; 2846 int csum_size = 0; 2847 int err; 2848 2849 if (ext4_has_metadata_csum(dir->i_sb)) 2850 csum_size = sizeof(struct ext4_dir_entry_tail); 2851 2852 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { 2853 err = ext4_try_create_inline_dir(handle, dir, inode); 2854 if (err < 0 && err != -ENOSPC) 2855 goto out; 2856 if (!err) 2857 goto out; 2858 } 2859 2860 inode->i_size = 0; 2861 dir_block = ext4_append(handle, inode, &block); 2862 if (IS_ERR(dir_block)) 2863 return PTR_ERR(dir_block); 2864 de = (struct ext4_dir_entry_2 *)dir_block->b_data; 2865 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0); 2866 set_nlink(inode, 2); 2867 if (csum_size) 2868 ext4_initialize_dirent_tail(dir_block, blocksize); 2869 2870 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata"); 2871 err = ext4_handle_dirty_dirblock(handle, inode, dir_block); 2872 if (err) 2873 goto out; 2874 set_buffer_verified(dir_block); 2875 out: 2876 brelse(dir_block); 2877 return err; 2878 } 2879 2880 static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 2881 struct dentry *dentry, umode_t mode) 2882 { 2883 handle_t *handle; 2884 struct inode *inode; 2885 int err, err2 = 0, credits, retries = 0; 2886 2887 if (EXT4_DIR_LINK_MAX(dir)) 2888 return -EMLINK; 2889 2890 err = dquot_initialize(dir); 2891 if (err) 2892 return err; 2893 2894 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 2895 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); 2896 retry: 2897 inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFDIR | mode, 2898 &dentry->d_name, 2899 0, NULL, EXT4_HT_DIR, credits); 2900 handle = ext4_journal_current_handle(); 2901 err = PTR_ERR(inode); 2902 if (IS_ERR(inode)) 2903 goto out_stop; 2904 2905 inode->i_op = &ext4_dir_inode_operations; 2906 inode->i_fop = &ext4_dir_operations; 2907 err = ext4_init_new_dir(handle, dir, inode); 2908 if (err) 2909 goto out_clear_inode; 2910 err = ext4_mark_inode_dirty(handle, inode); 2911 if (!err) 2912 err = ext4_add_entry(handle, dentry, inode); 2913 if (err) { 2914 out_clear_inode: 2915 clear_nlink(inode); 2916 ext4_orphan_add(handle, inode); 2917 unlock_new_inode(inode); 2918 err2 = ext4_mark_inode_dirty(handle, inode); 2919 if (unlikely(err2)) 2920 err = err2; 2921 ext4_journal_stop(handle); 2922 iput(inode); 2923 goto out_retry; 2924 } 2925 ext4_inc_count(dir); 2926 2927 ext4_update_dx_flag(dir); 2928 err = ext4_mark_inode_dirty(handle, dir); 2929 if (err) 2930 goto out_clear_inode; 2931 d_instantiate_new(dentry, inode); 2932 ext4_fc_track_create(handle, dentry); 2933 if (IS_DIRSYNC(dir)) 2934 ext4_handle_sync(handle); 2935 2936 out_stop: 2937 if (handle) 2938 ext4_journal_stop(handle); 2939 out_retry: 2940 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 2941 goto retry; 2942 return err; 2943 } 2944 2945 /* 2946 * routine to check that the specified directory is empty (for rmdir) 2947 */ 2948 bool ext4_empty_dir(struct inode *inode) 2949 { 2950 unsigned int offset; 2951 struct buffer_head *bh; 2952 struct ext4_dir_entry_2 *de; 2953 struct super_block *sb; 2954 2955 if (ext4_has_inline_data(inode)) { 2956 int has_inline_data = 1; 2957 int ret; 2958 2959 ret = empty_inline_dir(inode, &has_inline_data); 2960 if (has_inline_data) 2961 return ret; 2962 } 2963 2964 sb = inode->i_sb; 2965 if (inode->i_size < ext4_dir_rec_len(1, NULL) + 2966 ext4_dir_rec_len(2, NULL)) { 2967 EXT4_ERROR_INODE(inode, "invalid size"); 2968 return true; 2969 } 2970 /* The first directory block must not be a hole, 2971 * so treat it as DIRENT_HTREE 2972 */ 2973 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE); 2974 if (IS_ERR(bh)) 2975 return true; 2976 2977 de = (struct ext4_dir_entry_2 *) bh->b_data; 2978 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size, 2979 0) || 2980 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) { 2981 ext4_warning_inode(inode, "directory missing '.'"); 2982 brelse(bh); 2983 return true; 2984 } 2985 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize); 2986 de = ext4_next_entry(de, sb->s_blocksize); 2987 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size, 2988 offset) || 2989 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) { 2990 ext4_warning_inode(inode, "directory missing '..'"); 2991 brelse(bh); 2992 return true; 2993 } 2994 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize); 2995 while (offset < inode->i_size) { 2996 if (!(offset & (sb->s_blocksize - 1))) { 2997 unsigned int lblock; 2998 brelse(bh); 2999 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb); 3000 bh = ext4_read_dirblock(inode, lblock, EITHER); 3001 if (bh == NULL) { 3002 offset += sb->s_blocksize; 3003 continue; 3004 } 3005 if (IS_ERR(bh)) 3006 return true; 3007 } 3008 de = (struct ext4_dir_entry_2 *) (bh->b_data + 3009 (offset & (sb->s_blocksize - 1))); 3010 if (ext4_check_dir_entry(inode, NULL, de, bh, 3011 bh->b_data, bh->b_size, offset)) { 3012 offset = (offset | (sb->s_blocksize - 1)) + 1; 3013 continue; 3014 } 3015 if (le32_to_cpu(de->inode)) { 3016 brelse(bh); 3017 return false; 3018 } 3019 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize); 3020 } 3021 brelse(bh); 3022 return true; 3023 } 3024 3025 /* 3026 * ext4_orphan_add() links an unlinked or truncated inode into a list of 3027 * such inodes, starting at the superblock, in case we crash before the 3028 * file is closed/deleted, or in case the inode truncate spans multiple 3029 * transactions and the last transaction is not recovered after a crash. 3030 * 3031 * At filesystem recovery time, we walk this list deleting unlinked 3032 * inodes and truncating linked inodes in ext4_orphan_cleanup(). 3033 * 3034 * Orphan list manipulation functions must be called under i_mutex unless 3035 * we are just creating the inode or deleting it. 3036 */ 3037 int ext4_orphan_add(handle_t *handle, struct inode *inode) 3038 { 3039 struct super_block *sb = inode->i_sb; 3040 struct ext4_sb_info *sbi = EXT4_SB(sb); 3041 struct ext4_iloc iloc; 3042 int err = 0, rc; 3043 bool dirty = false; 3044 3045 if (!sbi->s_journal || is_bad_inode(inode)) 3046 return 0; 3047 3048 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) && 3049 !inode_is_locked(inode)); 3050 /* 3051 * Exit early if inode already is on orphan list. This is a big speedup 3052 * since we don't have to contend on the global s_orphan_lock. 3053 */ 3054 if (!list_empty(&EXT4_I(inode)->i_orphan)) 3055 return 0; 3056 3057 /* 3058 * Orphan handling is only valid for files with data blocks 3059 * being truncated, or files being unlinked. Note that we either 3060 * hold i_mutex, or the inode can not be referenced from outside, 3061 * so i_nlink should not be bumped due to race 3062 */ 3063 ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 3064 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0); 3065 3066 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 3067 err = ext4_journal_get_write_access(handle, sbi->s_sbh); 3068 if (err) 3069 goto out; 3070 3071 err = ext4_reserve_inode_write(handle, inode, &iloc); 3072 if (err) 3073 goto out; 3074 3075 mutex_lock(&sbi->s_orphan_lock); 3076 /* 3077 * Due to previous errors inode may be already a part of on-disk 3078 * orphan list. If so skip on-disk list modification. 3079 */ 3080 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) > 3081 (le32_to_cpu(sbi->s_es->s_inodes_count))) { 3082 /* Insert this inode at the head of the on-disk orphan list */ 3083 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan); 3084 lock_buffer(sbi->s_sbh); 3085 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino); 3086 ext4_superblock_csum_set(sb); 3087 unlock_buffer(sbi->s_sbh); 3088 dirty = true; 3089 } 3090 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan); 3091 mutex_unlock(&sbi->s_orphan_lock); 3092 3093 if (dirty) { 3094 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh); 3095 rc = ext4_mark_iloc_dirty(handle, inode, &iloc); 3096 if (!err) 3097 err = rc; 3098 if (err) { 3099 /* 3100 * We have to remove inode from in-memory list if 3101 * addition to on disk orphan list failed. Stray orphan 3102 * list entries can cause panics at unmount time. 3103 */ 3104 mutex_lock(&sbi->s_orphan_lock); 3105 list_del_init(&EXT4_I(inode)->i_orphan); 3106 mutex_unlock(&sbi->s_orphan_lock); 3107 } 3108 } else 3109 brelse(iloc.bh); 3110 3111 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino); 3112 jbd_debug(4, "orphan inode %lu will point to %d\n", 3113 inode->i_ino, NEXT_ORPHAN(inode)); 3114 out: 3115 ext4_std_error(sb, err); 3116 return err; 3117 } 3118 3119 /* 3120 * ext4_orphan_del() removes an unlinked or truncated inode from the list 3121 * of such inodes stored on disk, because it is finally being cleaned up. 3122 */ 3123 int ext4_orphan_del(handle_t *handle, struct inode *inode) 3124 { 3125 struct list_head *prev; 3126 struct ext4_inode_info *ei = EXT4_I(inode); 3127 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 3128 __u32 ino_next; 3129 struct ext4_iloc iloc; 3130 int err = 0; 3131 3132 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS)) 3133 return 0; 3134 3135 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) && 3136 !inode_is_locked(inode)); 3137 /* Do this quick check before taking global s_orphan_lock. */ 3138 if (list_empty(&ei->i_orphan)) 3139 return 0; 3140 3141 if (handle) { 3142 /* Grab inode buffer early before taking global s_orphan_lock */ 3143 err = ext4_reserve_inode_write(handle, inode, &iloc); 3144 } 3145 3146 mutex_lock(&sbi->s_orphan_lock); 3147 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino); 3148 3149 prev = ei->i_orphan.prev; 3150 list_del_init(&ei->i_orphan); 3151 3152 /* If we're on an error path, we may not have a valid 3153 * transaction handle with which to update the orphan list on 3154 * disk, but we still need to remove the inode from the linked 3155 * list in memory. */ 3156 if (!handle || err) { 3157 mutex_unlock(&sbi->s_orphan_lock); 3158 goto out_err; 3159 } 3160 3161 ino_next = NEXT_ORPHAN(inode); 3162 if (prev == &sbi->s_orphan) { 3163 jbd_debug(4, "superblock will point to %u\n", ino_next); 3164 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 3165 err = ext4_journal_get_write_access(handle, sbi->s_sbh); 3166 if (err) { 3167 mutex_unlock(&sbi->s_orphan_lock); 3168 goto out_brelse; 3169 } 3170 lock_buffer(sbi->s_sbh); 3171 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next); 3172 ext4_superblock_csum_set(inode->i_sb); 3173 unlock_buffer(sbi->s_sbh); 3174 mutex_unlock(&sbi->s_orphan_lock); 3175 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh); 3176 } else { 3177 struct ext4_iloc iloc2; 3178 struct inode *i_prev = 3179 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode; 3180 3181 jbd_debug(4, "orphan inode %lu will point to %u\n", 3182 i_prev->i_ino, ino_next); 3183 err = ext4_reserve_inode_write(handle, i_prev, &iloc2); 3184 if (err) { 3185 mutex_unlock(&sbi->s_orphan_lock); 3186 goto out_brelse; 3187 } 3188 NEXT_ORPHAN(i_prev) = ino_next; 3189 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2); 3190 mutex_unlock(&sbi->s_orphan_lock); 3191 } 3192 if (err) 3193 goto out_brelse; 3194 NEXT_ORPHAN(inode) = 0; 3195 err = ext4_mark_iloc_dirty(handle, inode, &iloc); 3196 out_err: 3197 ext4_std_error(inode->i_sb, err); 3198 return err; 3199 3200 out_brelse: 3201 brelse(iloc.bh); 3202 goto out_err; 3203 } 3204 3205 static int ext4_rmdir(struct inode *dir, struct dentry *dentry) 3206 { 3207 int retval; 3208 struct inode *inode; 3209 struct buffer_head *bh; 3210 struct ext4_dir_entry_2 *de; 3211 handle_t *handle = NULL; 3212 3213 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb)))) 3214 return -EIO; 3215 3216 /* Initialize quotas before so that eventual writes go in 3217 * separate transaction */ 3218 retval = dquot_initialize(dir); 3219 if (retval) 3220 return retval; 3221 retval = dquot_initialize(d_inode(dentry)); 3222 if (retval) 3223 return retval; 3224 3225 retval = -ENOENT; 3226 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); 3227 if (IS_ERR(bh)) 3228 return PTR_ERR(bh); 3229 if (!bh) 3230 goto end_rmdir; 3231 3232 inode = d_inode(dentry); 3233 3234 retval = -EFSCORRUPTED; 3235 if (le32_to_cpu(de->inode) != inode->i_ino) 3236 goto end_rmdir; 3237 3238 retval = -ENOTEMPTY; 3239 if (!ext4_empty_dir(inode)) 3240 goto end_rmdir; 3241 3242 handle = ext4_journal_start(dir, EXT4_HT_DIR, 3243 EXT4_DATA_TRANS_BLOCKS(dir->i_sb)); 3244 if (IS_ERR(handle)) { 3245 retval = PTR_ERR(handle); 3246 handle = NULL; 3247 goto end_rmdir; 3248 } 3249 3250 if (IS_DIRSYNC(dir)) 3251 ext4_handle_sync(handle); 3252 3253 retval = ext4_delete_entry(handle, dir, de, bh); 3254 if (retval) 3255 goto end_rmdir; 3256 if (!EXT4_DIR_LINK_EMPTY(inode)) 3257 ext4_warning_inode(inode, 3258 "empty directory '%.*s' has too many links (%u)", 3259 dentry->d_name.len, dentry->d_name.name, 3260 inode->i_nlink); 3261 inode_inc_iversion(inode); 3262 clear_nlink(inode); 3263 /* There's no need to set i_disksize: the fact that i_nlink is 3264 * zero will ensure that the right thing happens during any 3265 * recovery. */ 3266 inode->i_size = 0; 3267 ext4_orphan_add(handle, inode); 3268 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); 3269 retval = ext4_mark_inode_dirty(handle, inode); 3270 if (retval) 3271 goto end_rmdir; 3272 ext4_dec_count(dir); 3273 ext4_update_dx_flag(dir); 3274 ext4_fc_track_unlink(handle, dentry); 3275 retval = ext4_mark_inode_dirty(handle, dir); 3276 3277 #ifdef CONFIG_UNICODE 3278 /* VFS negative dentries are incompatible with Encoding and 3279 * Case-insensitiveness. Eventually we'll want avoid 3280 * invalidating the dentries here, alongside with returning the 3281 * negative dentries at ext4_lookup(), when it is better 3282 * supported by the VFS for the CI case. 3283 */ 3284 if (IS_CASEFOLDED(dir)) 3285 d_invalidate(dentry); 3286 #endif 3287 3288 end_rmdir: 3289 brelse(bh); 3290 if (handle) 3291 ext4_journal_stop(handle); 3292 return retval; 3293 } 3294 3295 int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name, 3296 struct inode *inode) 3297 { 3298 int retval = -ENOENT; 3299 struct buffer_head *bh; 3300 struct ext4_dir_entry_2 *de; 3301 int skip_remove_dentry = 0; 3302 3303 bh = ext4_find_entry(dir, d_name, &de, NULL); 3304 if (IS_ERR(bh)) 3305 return PTR_ERR(bh); 3306 3307 if (!bh) 3308 return -ENOENT; 3309 3310 if (le32_to_cpu(de->inode) != inode->i_ino) { 3311 /* 3312 * It's okay if we find dont find dentry which matches 3313 * the inode. That's because it might have gotten 3314 * renamed to a different inode number 3315 */ 3316 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) 3317 skip_remove_dentry = 1; 3318 else 3319 goto out; 3320 } 3321 3322 if (IS_DIRSYNC(dir)) 3323 ext4_handle_sync(handle); 3324 3325 if (!skip_remove_dentry) { 3326 retval = ext4_delete_entry(handle, dir, de, bh); 3327 if (retval) 3328 goto out; 3329 dir->i_ctime = dir->i_mtime = current_time(dir); 3330 ext4_update_dx_flag(dir); 3331 retval = ext4_mark_inode_dirty(handle, dir); 3332 if (retval) 3333 goto out; 3334 } else { 3335 retval = 0; 3336 } 3337 if (inode->i_nlink == 0) 3338 ext4_warning_inode(inode, "Deleting file '%.*s' with no links", 3339 d_name->len, d_name->name); 3340 else 3341 drop_nlink(inode); 3342 if (!inode->i_nlink) 3343 ext4_orphan_add(handle, inode); 3344 inode->i_ctime = current_time(inode); 3345 retval = ext4_mark_inode_dirty(handle, inode); 3346 3347 out: 3348 brelse(bh); 3349 return retval; 3350 } 3351 3352 static int ext4_unlink(struct inode *dir, struct dentry *dentry) 3353 { 3354 handle_t *handle; 3355 int retval; 3356 3357 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb)))) 3358 return -EIO; 3359 3360 trace_ext4_unlink_enter(dir, dentry); 3361 /* 3362 * Initialize quotas before so that eventual writes go 3363 * in separate transaction 3364 */ 3365 retval = dquot_initialize(dir); 3366 if (retval) 3367 goto out_trace; 3368 retval = dquot_initialize(d_inode(dentry)); 3369 if (retval) 3370 goto out_trace; 3371 3372 handle = ext4_journal_start(dir, EXT4_HT_DIR, 3373 EXT4_DATA_TRANS_BLOCKS(dir->i_sb)); 3374 if (IS_ERR(handle)) { 3375 retval = PTR_ERR(handle); 3376 goto out_trace; 3377 } 3378 3379 retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry)); 3380 if (!retval) 3381 ext4_fc_track_unlink(handle, dentry); 3382 #ifdef CONFIG_UNICODE 3383 /* VFS negative dentries are incompatible with Encoding and 3384 * Case-insensitiveness. Eventually we'll want avoid 3385 * invalidating the dentries here, alongside with returning the 3386 * negative dentries at ext4_lookup(), when it is better 3387 * supported by the VFS for the CI case. 3388 */ 3389 if (IS_CASEFOLDED(dir)) 3390 d_invalidate(dentry); 3391 #endif 3392 if (handle) 3393 ext4_journal_stop(handle); 3394 3395 out_trace: 3396 trace_ext4_unlink_exit(dentry, retval); 3397 return retval; 3398 } 3399 3400 static int ext4_symlink(struct user_namespace *mnt_userns, struct inode *dir, 3401 struct dentry *dentry, const char *symname) 3402 { 3403 handle_t *handle; 3404 struct inode *inode; 3405 int err, len = strlen(symname); 3406 int credits; 3407 struct fscrypt_str disk_link; 3408 3409 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb)))) 3410 return -EIO; 3411 3412 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize, 3413 &disk_link); 3414 if (err) 3415 return err; 3416 3417 err = dquot_initialize(dir); 3418 if (err) 3419 return err; 3420 3421 if ((disk_link.len > EXT4_N_BLOCKS * 4)) { 3422 /* 3423 * For non-fast symlinks, we just allocate inode and put it on 3424 * orphan list in the first transaction => we need bitmap, 3425 * group descriptor, sb, inode block, quota blocks, and 3426 * possibly selinux xattr blocks. 3427 */ 3428 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) + 3429 EXT4_XATTR_TRANS_BLOCKS; 3430 } else { 3431 /* 3432 * Fast symlink. We have to add entry to directory 3433 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS), 3434 * allocate new inode (bitmap, group descriptor, inode block, 3435 * quota blocks, sb is already counted in previous macros). 3436 */ 3437 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 3438 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3; 3439 } 3440 3441 inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFLNK|S_IRWXUGO, 3442 &dentry->d_name, 0, NULL, 3443 EXT4_HT_DIR, credits); 3444 handle = ext4_journal_current_handle(); 3445 if (IS_ERR(inode)) { 3446 if (handle) 3447 ext4_journal_stop(handle); 3448 return PTR_ERR(inode); 3449 } 3450 3451 if (IS_ENCRYPTED(inode)) { 3452 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link); 3453 if (err) 3454 goto err_drop_inode; 3455 inode->i_op = &ext4_encrypted_symlink_inode_operations; 3456 } 3457 3458 if ((disk_link.len > EXT4_N_BLOCKS * 4)) { 3459 if (!IS_ENCRYPTED(inode)) 3460 inode->i_op = &ext4_symlink_inode_operations; 3461 inode_nohighmem(inode); 3462 ext4_set_aops(inode); 3463 /* 3464 * We cannot call page_symlink() with transaction started 3465 * because it calls into ext4_write_begin() which can wait 3466 * for transaction commit if we are running out of space 3467 * and thus we deadlock. So we have to stop transaction now 3468 * and restart it when symlink contents is written. 3469 * 3470 * To keep fs consistent in case of crash, we have to put inode 3471 * to orphan list in the mean time. 3472 */ 3473 drop_nlink(inode); 3474 err = ext4_orphan_add(handle, inode); 3475 if (handle) 3476 ext4_journal_stop(handle); 3477 handle = NULL; 3478 if (err) 3479 goto err_drop_inode; 3480 err = __page_symlink(inode, disk_link.name, disk_link.len, 1); 3481 if (err) 3482 goto err_drop_inode; 3483 /* 3484 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS 3485 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified 3486 */ 3487 handle = ext4_journal_start(dir, EXT4_HT_DIR, 3488 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 3489 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1); 3490 if (IS_ERR(handle)) { 3491 err = PTR_ERR(handle); 3492 handle = NULL; 3493 goto err_drop_inode; 3494 } 3495 set_nlink(inode, 1); 3496 err = ext4_orphan_del(handle, inode); 3497 if (err) 3498 goto err_drop_inode; 3499 } else { 3500 /* clear the extent format for fast symlink */ 3501 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS); 3502 if (!IS_ENCRYPTED(inode)) { 3503 inode->i_op = &ext4_fast_symlink_inode_operations; 3504 inode->i_link = (char *)&EXT4_I(inode)->i_data; 3505 } 3506 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name, 3507 disk_link.len); 3508 inode->i_size = disk_link.len - 1; 3509 } 3510 EXT4_I(inode)->i_disksize = inode->i_size; 3511 err = ext4_add_nondir(handle, dentry, &inode); 3512 if (handle) 3513 ext4_journal_stop(handle); 3514 if (inode) 3515 iput(inode); 3516 goto out_free_encrypted_link; 3517 3518 err_drop_inode: 3519 if (handle) 3520 ext4_journal_stop(handle); 3521 clear_nlink(inode); 3522 unlock_new_inode(inode); 3523 iput(inode); 3524 out_free_encrypted_link: 3525 if (disk_link.name != (unsigned char *)symname) 3526 kfree(disk_link.name); 3527 return err; 3528 } 3529 3530 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry) 3531 { 3532 handle_t *handle; 3533 int err, retries = 0; 3534 retry: 3535 handle = ext4_journal_start(dir, EXT4_HT_DIR, 3536 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + 3537 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1); 3538 if (IS_ERR(handle)) 3539 return PTR_ERR(handle); 3540 3541 if (IS_DIRSYNC(dir)) 3542 ext4_handle_sync(handle); 3543 3544 inode->i_ctime = current_time(inode); 3545 ext4_inc_count(inode); 3546 ihold(inode); 3547 3548 err = ext4_add_entry(handle, dentry, inode); 3549 if (!err) { 3550 err = ext4_mark_inode_dirty(handle, inode); 3551 /* this can happen only for tmpfile being 3552 * linked the first time 3553 */ 3554 if (inode->i_nlink == 1) 3555 ext4_orphan_del(handle, inode); 3556 d_instantiate(dentry, inode); 3557 ext4_fc_track_link(handle, dentry); 3558 } else { 3559 drop_nlink(inode); 3560 iput(inode); 3561 } 3562 ext4_journal_stop(handle); 3563 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) 3564 goto retry; 3565 return err; 3566 } 3567 3568 static int ext4_link(struct dentry *old_dentry, 3569 struct inode *dir, struct dentry *dentry) 3570 { 3571 struct inode *inode = d_inode(old_dentry); 3572 int err; 3573 3574 if (inode->i_nlink >= EXT4_LINK_MAX) 3575 return -EMLINK; 3576 3577 err = fscrypt_prepare_link(old_dentry, dir, dentry); 3578 if (err) 3579 return err; 3580 3581 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) && 3582 (!projid_eq(EXT4_I(dir)->i_projid, 3583 EXT4_I(old_dentry->d_inode)->i_projid))) 3584 return -EXDEV; 3585 3586 err = dquot_initialize(dir); 3587 if (err) 3588 return err; 3589 return __ext4_link(dir, inode, dentry); 3590 } 3591 3592 /* 3593 * Try to find buffer head where contains the parent block. 3594 * It should be the inode block if it is inlined or the 1st block 3595 * if it is a normal dir. 3596 */ 3597 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle, 3598 struct inode *inode, 3599 int *retval, 3600 struct ext4_dir_entry_2 **parent_de, 3601 int *inlined) 3602 { 3603 struct buffer_head *bh; 3604 3605 if (!ext4_has_inline_data(inode)) { 3606 /* The first directory block must not be a hole, so 3607 * treat it as DIRENT_HTREE 3608 */ 3609 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE); 3610 if (IS_ERR(bh)) { 3611 *retval = PTR_ERR(bh); 3612 return NULL; 3613 } 3614 *parent_de = ext4_next_entry( 3615 (struct ext4_dir_entry_2 *)bh->b_data, 3616 inode->i_sb->s_blocksize); 3617 return bh; 3618 } 3619 3620 *inlined = 1; 3621 return ext4_get_first_inline_block(inode, parent_de, retval); 3622 } 3623 3624 struct ext4_renament { 3625 struct inode *dir; 3626 struct dentry *dentry; 3627 struct inode *inode; 3628 bool is_dir; 3629 int dir_nlink_delta; 3630 3631 /* entry for "dentry" */ 3632 struct buffer_head *bh; 3633 struct ext4_dir_entry_2 *de; 3634 int inlined; 3635 3636 /* entry for ".." in inode if it's a directory */ 3637 struct buffer_head *dir_bh; 3638 struct ext4_dir_entry_2 *parent_de; 3639 int dir_inlined; 3640 }; 3641 3642 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent) 3643 { 3644 int retval; 3645 3646 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode, 3647 &retval, &ent->parent_de, 3648 &ent->dir_inlined); 3649 if (!ent->dir_bh) 3650 return retval; 3651 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino) 3652 return -EFSCORRUPTED; 3653 BUFFER_TRACE(ent->dir_bh, "get_write_access"); 3654 return ext4_journal_get_write_access(handle, ent->dir_bh); 3655 } 3656 3657 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent, 3658 unsigned dir_ino) 3659 { 3660 int retval; 3661 3662 ent->parent_de->inode = cpu_to_le32(dir_ino); 3663 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata"); 3664 if (!ent->dir_inlined) { 3665 if (is_dx(ent->inode)) { 3666 retval = ext4_handle_dirty_dx_node(handle, 3667 ent->inode, 3668 ent->dir_bh); 3669 } else { 3670 retval = ext4_handle_dirty_dirblock(handle, ent->inode, 3671 ent->dir_bh); 3672 } 3673 } else { 3674 retval = ext4_mark_inode_dirty(handle, ent->inode); 3675 } 3676 if (retval) { 3677 ext4_std_error(ent->dir->i_sb, retval); 3678 return retval; 3679 } 3680 return 0; 3681 } 3682 3683 static int ext4_setent(handle_t *handle, struct ext4_renament *ent, 3684 unsigned ino, unsigned file_type) 3685 { 3686 int retval, retval2; 3687 3688 BUFFER_TRACE(ent->bh, "get write access"); 3689 retval = ext4_journal_get_write_access(handle, ent->bh); 3690 if (retval) 3691 return retval; 3692 ent->de->inode = cpu_to_le32(ino); 3693 if (ext4_has_feature_filetype(ent->dir->i_sb)) 3694 ent->de->file_type = file_type; 3695 inode_inc_iversion(ent->dir); 3696 ent->dir->i_ctime = ent->dir->i_mtime = 3697 current_time(ent->dir); 3698 retval = ext4_mark_inode_dirty(handle, ent->dir); 3699 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata"); 3700 if (!ent->inlined) { 3701 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh); 3702 if (unlikely(retval2)) { 3703 ext4_std_error(ent->dir->i_sb, retval2); 3704 return retval2; 3705 } 3706 } 3707 return retval; 3708 } 3709 3710 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent, 3711 unsigned ino, unsigned file_type) 3712 { 3713 struct ext4_renament old = *ent; 3714 int retval = 0; 3715 3716 /* 3717 * old->de could have moved from under us during make indexed dir, 3718 * so the old->de may no longer valid and need to find it again 3719 * before reset old inode info. 3720 */ 3721 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL); 3722 if (IS_ERR(old.bh)) 3723 retval = PTR_ERR(old.bh); 3724 if (!old.bh) 3725 retval = -ENOENT; 3726 if (retval) { 3727 ext4_std_error(old.dir->i_sb, retval); 3728 return; 3729 } 3730 3731 ext4_setent(handle, &old, ino, file_type); 3732 brelse(old.bh); 3733 } 3734 3735 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir, 3736 const struct qstr *d_name) 3737 { 3738 int retval = -ENOENT; 3739 struct buffer_head *bh; 3740 struct ext4_dir_entry_2 *de; 3741 3742 bh = ext4_find_entry(dir, d_name, &de, NULL); 3743 if (IS_ERR(bh)) 3744 return PTR_ERR(bh); 3745 if (bh) { 3746 retval = ext4_delete_entry(handle, dir, de, bh); 3747 brelse(bh); 3748 } 3749 return retval; 3750 } 3751 3752 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent, 3753 int force_reread) 3754 { 3755 int retval; 3756 /* 3757 * ent->de could have moved from under us during htree split, so make 3758 * sure that we are deleting the right entry. We might also be pointing 3759 * to a stale entry in the unused part of ent->bh so just checking inum 3760 * and the name isn't enough. 3761 */ 3762 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino || 3763 ent->de->name_len != ent->dentry->d_name.len || 3764 strncmp(ent->de->name, ent->dentry->d_name.name, 3765 ent->de->name_len) || 3766 force_reread) { 3767 retval = ext4_find_delete_entry(handle, ent->dir, 3768 &ent->dentry->d_name); 3769 } else { 3770 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh); 3771 if (retval == -ENOENT) { 3772 retval = ext4_find_delete_entry(handle, ent->dir, 3773 &ent->dentry->d_name); 3774 } 3775 } 3776 3777 if (retval) { 3778 ext4_warning_inode(ent->dir, 3779 "Deleting old file: nlink %d, error=%d", 3780 ent->dir->i_nlink, retval); 3781 } 3782 } 3783 3784 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent) 3785 { 3786 if (ent->dir_nlink_delta) { 3787 if (ent->dir_nlink_delta == -1) 3788 ext4_dec_count(ent->dir); 3789 else 3790 ext4_inc_count(ent->dir); 3791 ext4_mark_inode_dirty(handle, ent->dir); 3792 } 3793 } 3794 3795 static struct inode *ext4_whiteout_for_rename(struct user_namespace *mnt_userns, 3796 struct ext4_renament *ent, 3797 int credits, handle_t **h) 3798 { 3799 struct inode *wh; 3800 handle_t *handle; 3801 int retries = 0; 3802 3803 /* 3804 * for inode block, sb block, group summaries, 3805 * and inode bitmap 3806 */ 3807 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) + 3808 EXT4_XATTR_TRANS_BLOCKS + 4); 3809 retry: 3810 wh = ext4_new_inode_start_handle(mnt_userns, ent->dir, 3811 S_IFCHR | WHITEOUT_MODE, 3812 &ent->dentry->d_name, 0, NULL, 3813 EXT4_HT_DIR, credits); 3814 3815 handle = ext4_journal_current_handle(); 3816 if (IS_ERR(wh)) { 3817 if (handle) 3818 ext4_journal_stop(handle); 3819 if (PTR_ERR(wh) == -ENOSPC && 3820 ext4_should_retry_alloc(ent->dir->i_sb, &retries)) 3821 goto retry; 3822 } else { 3823 *h = handle; 3824 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV); 3825 wh->i_op = &ext4_special_inode_operations; 3826 } 3827 return wh; 3828 } 3829 3830 /* 3831 * Anybody can rename anything with this: the permission checks are left to the 3832 * higher-level routines. 3833 * 3834 * n.b. old_{dentry,inode) refers to the source dentry/inode 3835 * while new_{dentry,inode) refers to the destination dentry/inode 3836 * This comes from rename(const char *oldpath, const char *newpath) 3837 */ 3838 static int ext4_rename(struct user_namespace *mnt_userns, struct inode *old_dir, 3839 struct dentry *old_dentry, struct inode *new_dir, 3840 struct dentry *new_dentry, unsigned int flags) 3841 { 3842 handle_t *handle = NULL; 3843 struct ext4_renament old = { 3844 .dir = old_dir, 3845 .dentry = old_dentry, 3846 .inode = d_inode(old_dentry), 3847 }; 3848 struct ext4_renament new = { 3849 .dir = new_dir, 3850 .dentry = new_dentry, 3851 .inode = d_inode(new_dentry), 3852 }; 3853 int force_reread; 3854 int retval; 3855 struct inode *whiteout = NULL; 3856 int credits; 3857 u8 old_file_type; 3858 3859 if (new.inode && new.inode->i_nlink == 0) { 3860 EXT4_ERROR_INODE(new.inode, 3861 "target of rename is already freed"); 3862 return -EFSCORRUPTED; 3863 } 3864 3865 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) && 3866 (!projid_eq(EXT4_I(new_dir)->i_projid, 3867 EXT4_I(old_dentry->d_inode)->i_projid))) 3868 return -EXDEV; 3869 3870 retval = dquot_initialize(old.dir); 3871 if (retval) 3872 return retval; 3873 retval = dquot_initialize(new.dir); 3874 if (retval) 3875 return retval; 3876 3877 /* Initialize quotas before so that eventual writes go 3878 * in separate transaction */ 3879 if (new.inode) { 3880 retval = dquot_initialize(new.inode); 3881 if (retval) 3882 return retval; 3883 } 3884 3885 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL); 3886 if (IS_ERR(old.bh)) 3887 return PTR_ERR(old.bh); 3888 /* 3889 * Check for inode number is _not_ due to possible IO errors. 3890 * We might rmdir the source, keep it as pwd of some process 3891 * and merrily kill the link to whatever was created under the 3892 * same name. Goodbye sticky bit ;-< 3893 */ 3894 retval = -ENOENT; 3895 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino) 3896 goto release_bh; 3897 3898 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name, 3899 &new.de, &new.inlined); 3900 if (IS_ERR(new.bh)) { 3901 retval = PTR_ERR(new.bh); 3902 new.bh = NULL; 3903 goto release_bh; 3904 } 3905 if (new.bh) { 3906 if (!new.inode) { 3907 brelse(new.bh); 3908 new.bh = NULL; 3909 } 3910 } 3911 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC)) 3912 ext4_alloc_da_blocks(old.inode); 3913 3914 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) + 3915 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2); 3916 if (!(flags & RENAME_WHITEOUT)) { 3917 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits); 3918 if (IS_ERR(handle)) { 3919 retval = PTR_ERR(handle); 3920 goto release_bh; 3921 } 3922 } else { 3923 whiteout = ext4_whiteout_for_rename(mnt_userns, &old, credits, &handle); 3924 if (IS_ERR(whiteout)) { 3925 retval = PTR_ERR(whiteout); 3926 goto release_bh; 3927 } 3928 } 3929 3930 old_file_type = old.de->file_type; 3931 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir)) 3932 ext4_handle_sync(handle); 3933 3934 if (S_ISDIR(old.inode->i_mode)) { 3935 if (new.inode) { 3936 retval = -ENOTEMPTY; 3937 if (!ext4_empty_dir(new.inode)) 3938 goto end_rename; 3939 } else { 3940 retval = -EMLINK; 3941 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir)) 3942 goto end_rename; 3943 } 3944 retval = ext4_rename_dir_prepare(handle, &old); 3945 if (retval) 3946 goto end_rename; 3947 } 3948 /* 3949 * If we're renaming a file within an inline_data dir and adding or 3950 * setting the new dirent causes a conversion from inline_data to 3951 * extents/blockmap, we need to force the dirent delete code to 3952 * re-read the directory, or else we end up trying to delete a dirent 3953 * from what is now the extent tree root (or a block map). 3954 */ 3955 force_reread = (new.dir->i_ino == old.dir->i_ino && 3956 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA)); 3957 3958 if (whiteout) { 3959 /* 3960 * Do this before adding a new entry, so the old entry is sure 3961 * to be still pointing to the valid old entry. 3962 */ 3963 retval = ext4_setent(handle, &old, whiteout->i_ino, 3964 EXT4_FT_CHRDEV); 3965 if (retval) 3966 goto end_rename; 3967 retval = ext4_mark_inode_dirty(handle, whiteout); 3968 if (unlikely(retval)) 3969 goto end_rename; 3970 3971 } 3972 if (!new.bh) { 3973 retval = ext4_add_entry(handle, new.dentry, old.inode); 3974 if (retval) 3975 goto end_rename; 3976 } else { 3977 retval = ext4_setent(handle, &new, 3978 old.inode->i_ino, old_file_type); 3979 if (retval) 3980 goto end_rename; 3981 } 3982 if (force_reread) 3983 force_reread = !ext4_test_inode_flag(new.dir, 3984 EXT4_INODE_INLINE_DATA); 3985 3986 /* 3987 * Like most other Unix systems, set the ctime for inodes on a 3988 * rename. 3989 */ 3990 old.inode->i_ctime = current_time(old.inode); 3991 retval = ext4_mark_inode_dirty(handle, old.inode); 3992 if (unlikely(retval)) 3993 goto end_rename; 3994 3995 if (!whiteout) { 3996 /* 3997 * ok, that's it 3998 */ 3999 ext4_rename_delete(handle, &old, force_reread); 4000 } 4001 4002 if (new.inode) { 4003 ext4_dec_count(new.inode); 4004 new.inode->i_ctime = current_time(new.inode); 4005 } 4006 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir); 4007 ext4_update_dx_flag(old.dir); 4008 if (old.dir_bh) { 4009 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino); 4010 if (retval) 4011 goto end_rename; 4012 4013 ext4_dec_count(old.dir); 4014 if (new.inode) { 4015 /* checked ext4_empty_dir above, can't have another 4016 * parent, ext4_dec_count() won't work for many-linked 4017 * dirs */ 4018 clear_nlink(new.inode); 4019 } else { 4020 ext4_inc_count(new.dir); 4021 ext4_update_dx_flag(new.dir); 4022 retval = ext4_mark_inode_dirty(handle, new.dir); 4023 if (unlikely(retval)) 4024 goto end_rename; 4025 } 4026 } 4027 retval = ext4_mark_inode_dirty(handle, old.dir); 4028 if (unlikely(retval)) 4029 goto end_rename; 4030 4031 if (S_ISDIR(old.inode->i_mode)) { 4032 /* 4033 * We disable fast commits here that's because the 4034 * replay code is not yet capable of changing dot dot 4035 * dirents in directories. 4036 */ 4037 ext4_fc_mark_ineligible(old.inode->i_sb, 4038 EXT4_FC_REASON_RENAME_DIR); 4039 } else { 4040 if (new.inode) 4041 ext4_fc_track_unlink(handle, new.dentry); 4042 __ext4_fc_track_link(handle, old.inode, new.dentry); 4043 __ext4_fc_track_unlink(handle, old.inode, old.dentry); 4044 if (whiteout) 4045 __ext4_fc_track_create(handle, whiteout, old.dentry); 4046 } 4047 4048 if (new.inode) { 4049 retval = ext4_mark_inode_dirty(handle, new.inode); 4050 if (unlikely(retval)) 4051 goto end_rename; 4052 if (!new.inode->i_nlink) 4053 ext4_orphan_add(handle, new.inode); 4054 } 4055 retval = 0; 4056 4057 end_rename: 4058 if (whiteout) { 4059 if (retval) { 4060 ext4_resetent(handle, &old, 4061 old.inode->i_ino, old_file_type); 4062 drop_nlink(whiteout); 4063 ext4_orphan_add(handle, whiteout); 4064 } 4065 unlock_new_inode(whiteout); 4066 ext4_journal_stop(handle); 4067 iput(whiteout); 4068 } else { 4069 ext4_journal_stop(handle); 4070 } 4071 release_bh: 4072 brelse(old.dir_bh); 4073 brelse(old.bh); 4074 brelse(new.bh); 4075 return retval; 4076 } 4077 4078 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry, 4079 struct inode *new_dir, struct dentry *new_dentry) 4080 { 4081 handle_t *handle = NULL; 4082 struct ext4_renament old = { 4083 .dir = old_dir, 4084 .dentry = old_dentry, 4085 .inode = d_inode(old_dentry), 4086 }; 4087 struct ext4_renament new = { 4088 .dir = new_dir, 4089 .dentry = new_dentry, 4090 .inode = d_inode(new_dentry), 4091 }; 4092 u8 new_file_type; 4093 int retval; 4094 struct timespec64 ctime; 4095 4096 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) && 4097 !projid_eq(EXT4_I(new_dir)->i_projid, 4098 EXT4_I(old_dentry->d_inode)->i_projid)) || 4099 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) && 4100 !projid_eq(EXT4_I(old_dir)->i_projid, 4101 EXT4_I(new_dentry->d_inode)->i_projid))) 4102 return -EXDEV; 4103 4104 retval = dquot_initialize(old.dir); 4105 if (retval) 4106 return retval; 4107 retval = dquot_initialize(new.dir); 4108 if (retval) 4109 return retval; 4110 4111 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, 4112 &old.de, &old.inlined); 4113 if (IS_ERR(old.bh)) 4114 return PTR_ERR(old.bh); 4115 /* 4116 * Check for inode number is _not_ due to possible IO errors. 4117 * We might rmdir the source, keep it as pwd of some process 4118 * and merrily kill the link to whatever was created under the 4119 * same name. Goodbye sticky bit ;-< 4120 */ 4121 retval = -ENOENT; 4122 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino) 4123 goto end_rename; 4124 4125 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name, 4126 &new.de, &new.inlined); 4127 if (IS_ERR(new.bh)) { 4128 retval = PTR_ERR(new.bh); 4129 new.bh = NULL; 4130 goto end_rename; 4131 } 4132 4133 /* RENAME_EXCHANGE case: old *and* new must both exist */ 4134 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino) 4135 goto end_rename; 4136 4137 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, 4138 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) + 4139 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2)); 4140 if (IS_ERR(handle)) { 4141 retval = PTR_ERR(handle); 4142 handle = NULL; 4143 goto end_rename; 4144 } 4145 4146 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir)) 4147 ext4_handle_sync(handle); 4148 4149 if (S_ISDIR(old.inode->i_mode)) { 4150 old.is_dir = true; 4151 retval = ext4_rename_dir_prepare(handle, &old); 4152 if (retval) 4153 goto end_rename; 4154 } 4155 if (S_ISDIR(new.inode->i_mode)) { 4156 new.is_dir = true; 4157 retval = ext4_rename_dir_prepare(handle, &new); 4158 if (retval) 4159 goto end_rename; 4160 } 4161 4162 /* 4163 * Other than the special case of overwriting a directory, parents' 4164 * nlink only needs to be modified if this is a cross directory rename. 4165 */ 4166 if (old.dir != new.dir && old.is_dir != new.is_dir) { 4167 old.dir_nlink_delta = old.is_dir ? -1 : 1; 4168 new.dir_nlink_delta = -old.dir_nlink_delta; 4169 retval = -EMLINK; 4170 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) || 4171 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir))) 4172 goto end_rename; 4173 } 4174 4175 new_file_type = new.de->file_type; 4176 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type); 4177 if (retval) 4178 goto end_rename; 4179 4180 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type); 4181 if (retval) 4182 goto end_rename; 4183 4184 /* 4185 * Like most other Unix systems, set the ctime for inodes on a 4186 * rename. 4187 */ 4188 ctime = current_time(old.inode); 4189 old.inode->i_ctime = ctime; 4190 new.inode->i_ctime = ctime; 4191 retval = ext4_mark_inode_dirty(handle, old.inode); 4192 if (unlikely(retval)) 4193 goto end_rename; 4194 retval = ext4_mark_inode_dirty(handle, new.inode); 4195 if (unlikely(retval)) 4196 goto end_rename; 4197 ext4_fc_mark_ineligible(new.inode->i_sb, 4198 EXT4_FC_REASON_CROSS_RENAME); 4199 if (old.dir_bh) { 4200 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino); 4201 if (retval) 4202 goto end_rename; 4203 } 4204 if (new.dir_bh) { 4205 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino); 4206 if (retval) 4207 goto end_rename; 4208 } 4209 ext4_update_dir_count(handle, &old); 4210 ext4_update_dir_count(handle, &new); 4211 retval = 0; 4212 4213 end_rename: 4214 brelse(old.dir_bh); 4215 brelse(new.dir_bh); 4216 brelse(old.bh); 4217 brelse(new.bh); 4218 if (handle) 4219 ext4_journal_stop(handle); 4220 return retval; 4221 } 4222 4223 static int ext4_rename2(struct user_namespace *mnt_userns, 4224 struct inode *old_dir, struct dentry *old_dentry, 4225 struct inode *new_dir, struct dentry *new_dentry, 4226 unsigned int flags) 4227 { 4228 int err; 4229 4230 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb)))) 4231 return -EIO; 4232 4233 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4234 return -EINVAL; 4235 4236 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry, 4237 flags); 4238 if (err) 4239 return err; 4240 4241 if (flags & RENAME_EXCHANGE) { 4242 return ext4_cross_rename(old_dir, old_dentry, 4243 new_dir, new_dentry); 4244 } 4245 4246 return ext4_rename(mnt_userns, old_dir, old_dentry, new_dir, new_dentry, flags); 4247 } 4248 4249 /* 4250 * directories can handle most operations... 4251 */ 4252 const struct inode_operations ext4_dir_inode_operations = { 4253 .create = ext4_create, 4254 .lookup = ext4_lookup, 4255 .link = ext4_link, 4256 .unlink = ext4_unlink, 4257 .symlink = ext4_symlink, 4258 .mkdir = ext4_mkdir, 4259 .rmdir = ext4_rmdir, 4260 .mknod = ext4_mknod, 4261 .tmpfile = ext4_tmpfile, 4262 .rename = ext4_rename2, 4263 .setattr = ext4_setattr, 4264 .getattr = ext4_getattr, 4265 .listxattr = ext4_listxattr, 4266 .get_acl = ext4_get_acl, 4267 .set_acl = ext4_set_acl, 4268 .fiemap = ext4_fiemap, 4269 }; 4270 4271 const struct inode_operations ext4_special_inode_operations = { 4272 .setattr = ext4_setattr, 4273 .getattr = ext4_getattr, 4274 .listxattr = ext4_listxattr, 4275 .get_acl = ext4_get_acl, 4276 .set_acl = ext4_set_acl, 4277 }; 4278