1/* 2 * linux/fs/ext3/namei.c 3 * 4 * Copyright (C) 1992, 1993, 1994, 1995 5 * Remy Card (card@masi.ibp.fr) 6 * Laboratoire MASI - Institut Blaise Pascal 7 * Universite Pierre et Marie Curie (Paris VI) 8 * 9 * from 10 * 11 * linux/fs/minix/namei.c 12 * 13 * Copyright (C) 1991, 1992 Linus Torvalds 14 * 15 * Big-endian to little-endian byte-swapping/bitmaps by 16 * David S. Miller (davem@caip.rutgers.edu), 1995 17 * Directory entry file type support and forward compatibility hooks 18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998 19 * Hash Tree Directory indexing (c) 20 * Daniel Phillips, 2001 21 * Hash Tree Directory indexing porting 22 * Christopher Li, 2002 23 * Hash Tree Directory indexing cleanup 24 * Theodore Ts'o, 2002 25 */ 26 27#include <linux/fs.h> 28#include <linux/pagemap.h> 29#include <linux/jbd.h> 30#include <linux/time.h> 31#include <linux/ext3_fs.h> 32#include <linux/ext3_jbd.h> 33#include <linux/fcntl.h> 34#include <linux/stat.h> 35#include <linux/string.h> 36#include <linux/quotaops.h> 37#include <linux/buffer_head.h> 38#include <linux/bio.h> 39 40#include "namei.h" 41#include "xattr.h" 42#include "acl.h" 43 44/* 45 * define how far ahead to read directories while searching them. 46 */ 47#define NAMEI_RA_CHUNKS 2 48#define NAMEI_RA_BLOCKS 4 49#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS) 50#define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b)) 51 52static struct buffer_head *ext3_append(handle_t *handle, 53 struct inode *inode, 54 u32 *block, int *err) 55{ 56 struct buffer_head *bh; 57 58 *block = inode->i_size >> inode->i_sb->s_blocksize_bits; 59 60 if ((bh = ext3_bread(handle, inode, *block, 1, err))) { 61 inode->i_size += inode->i_sb->s_blocksize; 62 EXT3_I(inode)->i_disksize = inode->i_size; 63 ext3_journal_get_write_access(handle,bh); 64 } 65 return bh; 66} 67 68#ifndef assert 69#define assert(test) J_ASSERT(test) 70#endif 71 72#ifndef swap 73#define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0) 74#endif 75 76#ifdef DX_DEBUG 77#define dxtrace(command) command 78#else 79#define dxtrace(command) 80#endif 81 82struct fake_dirent 83{ 84 __le32 inode; 85 __le16 rec_len; 86 u8 name_len; 87 u8 file_type; 88}; 89 90struct dx_countlimit 91{ 92 __le16 limit; 93 __le16 count; 94}; 95 96struct dx_entry 97{ 98 __le32 hash; 99 __le32 block; 100}; 101 102/* 103 * dx_root_info is laid out so that if it should somehow get overlaid by a 104 * dirent the two low bits of the hash version will be zero. Therefore, the 105 * hash version mod 4 should never be 0. Sincerely, the paranoia department. 106 */ 107 108struct dx_root 109{ 110 struct fake_dirent dot; 111 char dot_name[4]; 112 struct fake_dirent dotdot; 113 char dotdot_name[4]; 114 struct dx_root_info 115 { 116 __le32 reserved_zero; 117 u8 hash_version; 118 u8 info_length; /* 8 */ 119 u8 indirect_levels; 120 u8 unused_flags; 121 } 122 info; 123 struct dx_entry entries[0]; 124}; 125 126struct dx_node 127{ 128 struct fake_dirent fake; 129 struct dx_entry entries[0]; 130}; 131 132 133struct dx_frame 134{ 135 struct buffer_head *bh; 136 struct dx_entry *entries; 137 struct dx_entry *at; 138}; 139 140struct dx_map_entry 141{ 142 u32 hash; 143 u32 offs; 144}; 145 146#ifdef CONFIG_EXT3_INDEX 147static inline unsigned dx_get_block (struct dx_entry *entry); 148static void dx_set_block (struct dx_entry *entry, unsigned value); 149static inline unsigned dx_get_hash (struct dx_entry *entry); 150static void dx_set_hash (struct dx_entry *entry, unsigned value); 151static unsigned dx_get_count (struct dx_entry *entries); 152static unsigned dx_get_limit (struct dx_entry *entries); 153static void dx_set_count (struct dx_entry *entries, unsigned value); 154static void dx_set_limit (struct dx_entry *entries, unsigned value); 155static unsigned dx_root_limit (struct inode *dir, unsigned infosize); 156static unsigned dx_node_limit (struct inode *dir); 157static struct dx_frame *dx_probe(struct dentry *dentry, 158 struct inode *dir, 159 struct dx_hash_info *hinfo, 160 struct dx_frame *frame, 161 int *err); 162static void dx_release (struct dx_frame *frames); 163static int dx_make_map (struct ext3_dir_entry_2 *de, int size, 164 struct dx_hash_info *hinfo, struct dx_map_entry map[]); 165static void dx_sort_map(struct dx_map_entry *map, unsigned count); 166static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to, 167 struct dx_map_entry *offsets, int count); 168static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size); 169static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block); 170static int ext3_htree_next_block(struct inode *dir, __u32 hash, 171 struct dx_frame *frame, 172 struct dx_frame *frames, 173 __u32 *start_hash); 174static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry, 175 struct ext3_dir_entry_2 **res_dir, int *err); 176static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry, 177 struct inode *inode); 178 179/* 180 * Future: use high four bits of block for coalesce-on-delete flags 181 * Mask them off for now. 182 */ 183 184static inline unsigned dx_get_block (struct dx_entry *entry) 185{ 186 return le32_to_cpu(entry->block) & 0x00ffffff; 187} 188 189static inline void dx_set_block (struct dx_entry *entry, unsigned value) 190{ 191 entry->block = cpu_to_le32(value); 192} 193 194static inline unsigned dx_get_hash (struct dx_entry *entry) 195{ 196 return le32_to_cpu(entry->hash); 197} 198 199static inline void dx_set_hash (struct dx_entry *entry, unsigned value) 200{ 201 entry->hash = cpu_to_le32(value); 202} 203 204static inline unsigned dx_get_count (struct dx_entry *entries) 205{ 206 return le16_to_cpu(((struct dx_countlimit *) entries)->count); 207} 208 209static inline unsigned dx_get_limit (struct dx_entry *entries) 210{ 211 return le16_to_cpu(((struct dx_countlimit *) entries)->limit); 212} 213 214static inline void dx_set_count (struct dx_entry *entries, unsigned value) 215{ 216 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value); 217} 218 219static inline void dx_set_limit (struct dx_entry *entries, unsigned value) 220{ 221 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value); 222} 223 224static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize) 225{ 226 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) - 227 EXT3_DIR_REC_LEN(2) - infosize; 228 return 0? 20: entry_space / sizeof(struct dx_entry); 229} 230 231static inline unsigned dx_node_limit (struct inode *dir) 232{ 233 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0); 234 return 0? 22: entry_space / sizeof(struct dx_entry); 235} 236 237/* 238 * Debug 239 */ 240#ifdef DX_DEBUG 241static void dx_show_index (char * label, struct dx_entry *entries) 242{ 243 int i, n = dx_get_count (entries); 244 printk("%s index ", label); 245 for (i = 0; i < n; i++) 246 { 247 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i)); 248 } 249 printk("\n"); 250} 251 252struct stats 253{ 254 unsigned names; 255 unsigned space; 256 unsigned bcount; 257}; 258 259static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de, 260 int size, int show_names) 261{ 262 unsigned names = 0, space = 0; 263 char *base = (char *) de; 264 struct dx_hash_info h = *hinfo; 265 266 printk("names: "); 267 while ((char *) de < base + size) 268 { 269 if (de->inode) 270 { 271 if (show_names) 272 { 273 int len = de->name_len; 274 char *name = de->name; 275 while (len--) printk("%c", *name++); 276 ext3fs_dirhash(de->name, de->name_len, &h); 277 printk(":%x.%u ", h.hash, 278 ((char *) de - base)); 279 } 280 space += EXT3_DIR_REC_LEN(de->name_len); 281 names++; 282 } 283 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len)); 284 } 285 printk("(%i)\n", names); 286 return (struct stats) { names, space, 1 }; 287} 288 289struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir, 290 struct dx_entry *entries, int levels) 291{ 292 unsigned blocksize = dir->i_sb->s_blocksize; 293 unsigned count = dx_get_count (entries), names = 0, space = 0, i; 294 unsigned bcount = 0; 295 struct buffer_head *bh; 296 int err; 297 printk("%i indexed blocks...\n", count); 298 for (i = 0; i < count; i++, entries++) 299 { 300 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0; 301 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash; 302 struct stats stats; 303 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range); 304 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue; 305 stats = levels? 306 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1): 307 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0); 308 names += stats.names; 309 space += stats.space; 310 bcount += stats.bcount; 311 brelse (bh); 312 } 313 if (bcount) 314 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ", 315 names, space/bcount,(space/bcount)*100/blocksize); 316 return (struct stats) { names, space, bcount}; 317} 318#endif /* DX_DEBUG */ 319 320/* 321 * Probe for a directory leaf block to search. 322 * 323 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format 324 * error in the directory index, and the caller should fall back to 325 * searching the directory normally. The callers of dx_probe **MUST** 326 * check for this error code, and make sure it never gets reflected 327 * back to userspace. 328 */ 329static struct dx_frame * 330dx_probe(struct dentry *dentry, struct inode *dir, 331 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err) 332{ 333 unsigned count, indirect; 334 struct dx_entry *at, *entries, *p, *q, *m; 335 struct dx_root *root; 336 struct buffer_head *bh; 337 struct dx_frame *frame = frame_in; 338 u32 hash; 339 340 frame->bh = NULL; 341 if (dentry) 342 dir = dentry->d_parent->d_inode; 343 if (!(bh = ext3_bread (NULL,dir, 0, 0, err))) 344 goto fail; 345 root = (struct dx_root *) bh->b_data; 346 if (root->info.hash_version != DX_HASH_TEA && 347 root->info.hash_version != DX_HASH_HALF_MD4 && 348 root->info.hash_version != DX_HASH_LEGACY) { 349 ext3_warning(dir->i_sb, __FUNCTION__, 350 "Unrecognised inode hash code %d", 351 root->info.hash_version); 352 brelse(bh); 353 *err = ERR_BAD_DX_DIR; 354 goto fail; 355 } 356 hinfo->hash_version = root->info.hash_version; 357 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed; 358 if (dentry) 359 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo); 360 hash = hinfo->hash; 361 362 if (root->info.unused_flags & 1) { 363 ext3_warning(dir->i_sb, __FUNCTION__, 364 "Unimplemented inode hash flags: %#06x", 365 root->info.unused_flags); 366 brelse(bh); 367 *err = ERR_BAD_DX_DIR; 368 goto fail; 369 } 370 371 if ((indirect = root->info.indirect_levels) > 1) { 372 ext3_warning(dir->i_sb, __FUNCTION__, 373 "Unimplemented inode hash depth: %#06x", 374 root->info.indirect_levels); 375 brelse(bh); 376 *err = ERR_BAD_DX_DIR; 377 goto fail; 378 } 379 380 entries = (struct dx_entry *) (((char *)&root->info) + 381 root->info.info_length); 382 assert(dx_get_limit(entries) == dx_root_limit(dir, 383 root->info.info_length)); 384 dxtrace (printk("Look up %x", hash)); 385 while (1) 386 { 387 count = dx_get_count(entries); 388 assert (count && count <= dx_get_limit(entries)); 389 p = entries + 1; 390 q = entries + count - 1; 391 while (p <= q) 392 { 393 m = p + (q - p)/2; 394 dxtrace(printk(".")); 395 if (dx_get_hash(m) > hash) 396 q = m - 1; 397 else 398 p = m + 1; 399 } 400 401 if (0) // linear search cross check 402 { 403 unsigned n = count - 1; 404 at = entries; 405 while (n--) 406 { 407 dxtrace(printk(",")); 408 if (dx_get_hash(++at) > hash) 409 { 410 at--; 411 break; 412 } 413 } 414 assert (at == p - 1); 415 } 416 417 at = p - 1; 418 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at))); 419 frame->bh = bh; 420 frame->entries = entries; 421 frame->at = at; 422 if (!indirect--) return frame; 423 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err))) 424 goto fail2; 425 at = entries = ((struct dx_node *) bh->b_data)->entries; 426 assert (dx_get_limit(entries) == dx_node_limit (dir)); 427 frame++; 428 } 429fail2: 430 while (frame >= frame_in) { 431 brelse(frame->bh); 432 frame--; 433 } 434fail: 435 return NULL; 436} 437 438static void dx_release (struct dx_frame *frames) 439{ 440 if (frames[0].bh == NULL) 441 return; 442 443 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels) 444 brelse(frames[1].bh); 445 brelse(frames[0].bh); 446} 447 448/* 449 * This function increments the frame pointer to search the next leaf 450 * block, and reads in the necessary intervening nodes if the search 451 * should be necessary. Whether or not the search is necessary is 452 * controlled by the hash parameter. If the hash value is even, then 453 * the search is only continued if the next block starts with that 454 * hash value. This is used if we are searching for a specific file. 455 * 456 * If the hash value is HASH_NB_ALWAYS, then always go to the next block. 457 * 458 * This function returns 1 if the caller should continue to search, 459 * or 0 if it should not. If there is an error reading one of the 460 * index blocks, it will a negative error code. 461 * 462 * If start_hash is non-null, it will be filled in with the starting 463 * hash of the next page. 464 */ 465static int ext3_htree_next_block(struct inode *dir, __u32 hash, 466 struct dx_frame *frame, 467 struct dx_frame *frames, 468 __u32 *start_hash) 469{ 470 struct dx_frame *p; 471 struct buffer_head *bh; 472 int err, num_frames = 0; 473 __u32 bhash; 474 475 p = frame; 476 /* 477 * Find the next leaf page by incrementing the frame pointer. 478 * If we run out of entries in the interior node, loop around and 479 * increment pointer in the parent node. When we break out of 480 * this loop, num_frames indicates the number of interior 481 * nodes need to be read. 482 */ 483 while (1) { 484 if (++(p->at) < p->entries + dx_get_count(p->entries)) 485 break; 486 if (p == frames) 487 return 0; 488 num_frames++; 489 p--; 490 } 491 492 /* 493 * If the hash is 1, then continue only if the next page has a 494 * continuation hash of any value. This is used for readdir 495 * handling. Otherwise, check to see if the hash matches the 496 * desired contiuation hash. If it doesn't, return since 497 * there's no point to read in the successive index pages. 498 */ 499 bhash = dx_get_hash(p->at); 500 if (start_hash) 501 *start_hash = bhash; 502 if ((hash & 1) == 0) { 503 if ((bhash & ~1) != hash) 504 return 0; 505 } 506 /* 507 * If the hash is HASH_NB_ALWAYS, we always go to the next 508 * block so no check is necessary 509 */ 510 while (num_frames--) { 511 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at), 512 0, &err))) 513 return err; /* Failure */ 514 p++; 515 brelse (p->bh); 516 p->bh = bh; 517 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries; 518 } 519 return 1; 520} 521 522 523/* 524 * p is at least 6 bytes before the end of page 525 */ 526static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p) 527{ 528 return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len)); 529} 530 531/* 532 * This function fills a red-black tree with information from a 533 * directory block. It returns the number directory entries loaded 534 * into the tree. If there is an error it is returned in err. 535 */ 536static int htree_dirblock_to_tree(struct file *dir_file, 537 struct inode *dir, int block, 538 struct dx_hash_info *hinfo, 539 __u32 start_hash, __u32 start_minor_hash) 540{ 541 struct buffer_head *bh; 542 struct ext3_dir_entry_2 *de, *top; 543 int err, count = 0; 544 545 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block)); 546 if (!(bh = ext3_bread (NULL, dir, block, 0, &err))) 547 return err; 548 549 de = (struct ext3_dir_entry_2 *) bh->b_data; 550 top = (struct ext3_dir_entry_2 *) ((char *) de + 551 dir->i_sb->s_blocksize - 552 EXT3_DIR_REC_LEN(0)); 553 for (; de < top; de = ext3_next_entry(de)) { 554 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh, 555 (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb)) 556 +((char *)de - bh->b_data))) { 557 /* On error, skip the f_pos to the next block. */ 558 dir_file->f_pos = (dir_file->f_pos | 559 (dir->i_sb->s_blocksize - 1)) + 1; 560 brelse (bh); 561 return count; 562 } 563 ext3fs_dirhash(de->name, de->name_len, hinfo); 564 if ((hinfo->hash < start_hash) || 565 ((hinfo->hash == start_hash) && 566 (hinfo->minor_hash < start_minor_hash))) 567 continue; 568 if (de->inode == 0) 569 continue; 570 if ((err = ext3_htree_store_dirent(dir_file, 571 hinfo->hash, hinfo->minor_hash, de)) != 0) { 572 brelse(bh); 573 return err; 574 } 575 count++; 576 } 577 brelse(bh); 578 return count; 579} 580 581 582/* 583 * This function fills a red-black tree with information from a 584 * directory. We start scanning the directory in hash order, starting 585 * at start_hash and start_minor_hash. 586 * 587 * This function returns the number of entries inserted into the tree, 588 * or a negative error code. 589 */ 590int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash, 591 __u32 start_minor_hash, __u32 *next_hash) 592{ 593 struct dx_hash_info hinfo; 594 struct ext3_dir_entry_2 *de; 595 struct dx_frame frames[2], *frame; 596 struct inode *dir; 597 int block, err; 598 int count = 0; 599 int ret; 600 __u32 hashval; 601 602 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash, 603 start_minor_hash)); 604 dir = dir_file->f_path.dentry->d_inode; 605 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) { 606 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version; 607 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed; 608 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo, 609 start_hash, start_minor_hash); 610 *next_hash = ~0; 611 return count; 612 } 613 hinfo.hash = start_hash; 614 hinfo.minor_hash = 0; 615 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err); 616 if (!frame) 617 return err; 618 619 /* Add '.' and '..' from the htree header */ 620 if (!start_hash && !start_minor_hash) { 621 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data; 622 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0) 623 goto errout; 624 count++; 625 } 626 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) { 627 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data; 628 de = ext3_next_entry(de); 629 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0) 630 goto errout; 631 count++; 632 } 633 634 while (1) { 635 block = dx_get_block(frame->at); 636 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo, 637 start_hash, start_minor_hash); 638 if (ret < 0) { 639 err = ret; 640 goto errout; 641 } 642 count += ret; 643 hashval = ~0; 644 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS, 645 frame, frames, &hashval); 646 *next_hash = hashval; 647 if (ret < 0) { 648 err = ret; 649 goto errout; 650 } 651 /* 652 * Stop if: (a) there are no more entries, or 653 * (b) we have inserted at least one entry and the 654 * next hash value is not a continuation 655 */ 656 if ((ret == 0) || 657 (count && ((hashval & 1) == 0))) 658 break; 659 } 660 dx_release(frames); 661 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n", 662 count, *next_hash)); 663 return count; 664errout: 665 dx_release(frames); 666 return (err); 667} 668 669 670/* 671 * Directory block splitting, compacting 672 */ 673 674static int dx_make_map (struct ext3_dir_entry_2 *de, int size, 675 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail) 676{ 677 int count = 0; 678 char *base = (char *) de; 679 struct dx_hash_info h = *hinfo; 680 681 while ((char *) de < base + size) 682 { 683 if (de->name_len && de->inode) { 684 ext3fs_dirhash(de->name, de->name_len, &h); 685 map_tail--; 686 map_tail->hash = h.hash; 687 map_tail->offs = (u32) ((char *) de - base); 688 count++; 689 cond_resched(); 690 } 691 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len)); 692 } 693 return count; 694} 695 696static void dx_sort_map (struct dx_map_entry *map, unsigned count) 697{ 698 struct dx_map_entry *p, *q, *top = map + count - 1; 699 int more; 700 /* Combsort until bubble sort doesn't suck */ 701 while (count > 2) 702 { 703 count = count*10/13; 704 if (count - 9 < 2) /* 9, 10 -> 11 */ 705 count = 11; 706 for (p = top, q = p - count; q >= map; p--, q--) 707 if (p->hash < q->hash) 708 swap(*p, *q); 709 } 710 /* Garden variety bubble sort */ 711 do { 712 more = 0; 713 q = top; 714 while (q-- > map) 715 { 716 if (q[1].hash >= q[0].hash) 717 continue; 718 swap(*(q+1), *q); 719 more = 1; 720 } 721 } while(more); 722} 723 724static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block) 725{ 726 struct dx_entry *entries = frame->entries; 727 struct dx_entry *old = frame->at, *new = old + 1; 728 int count = dx_get_count(entries); 729 730 assert(count < dx_get_limit(entries)); 731 assert(old < entries + count); 732 memmove(new + 1, new, (char *)(entries + count) - (char *)(new)); 733 dx_set_hash(new, hash); 734 dx_set_block(new, block); 735 dx_set_count(entries, count + 1); 736} 737#endif 738 739 740static void ext3_update_dx_flag(struct inode *inode) 741{ 742 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb, 743 EXT3_FEATURE_COMPAT_DIR_INDEX)) 744 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL; 745} 746 747/* 748 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure. 749 * 750 * `len <= EXT3_NAME_LEN' is guaranteed by caller. 751 * `de != NULL' is guaranteed by caller. 752 */ 753static inline int ext3_match (int len, const char * const name, 754 struct ext3_dir_entry_2 * de) 755{ 756 if (len != de->name_len) 757 return 0; 758 if (!de->inode) 759 return 0; 760 return !memcmp(name, de->name, len); 761} 762 763/* 764 * Returns 0 if not found, -1 on failure, and 1 on success 765 */ 766static inline int search_dirblock(struct buffer_head * bh, 767 struct inode *dir, 768 struct dentry *dentry, 769 unsigned long offset, 770 struct ext3_dir_entry_2 ** res_dir) 771{ 772 struct ext3_dir_entry_2 * de; 773 char * dlimit; 774 int de_len; 775 const char *name = dentry->d_name.name; 776 int namelen = dentry->d_name.len; 777 778 de = (struct ext3_dir_entry_2 *) bh->b_data; 779 dlimit = bh->b_data + dir->i_sb->s_blocksize; 780 while ((char *) de < dlimit) { 781 /* this code is executed quadratically often */ 782 /* do minimal checking `by hand' */ 783 784 if ((char *) de + namelen <= dlimit && 785 ext3_match (namelen, name, de)) { 786 /* found a match - just to be sure, do a full check */ 787 if (!ext3_check_dir_entry("ext3_find_entry", 788 dir, de, bh, offset)) 789 return -1; 790 *res_dir = de; 791 return 1; 792 } 793 /* prevent looping on a bad block */ 794 de_len = le16_to_cpu(de->rec_len); 795 if (de_len <= 0) 796 return -1; 797 offset += de_len; 798 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len); 799 } 800 return 0; 801} 802 803 804/* 805 * ext3_find_entry() 806 * 807 * finds an entry in the specified directory with the wanted name. It 808 * returns the cache buffer in which the entry was found, and the entry 809 * itself (as a parameter - res_dir). It does NOT read the inode of the 810 * entry - you'll have to do that yourself if you want to. 811 * 812 * The returned buffer_head has ->b_count elevated. The caller is expected 813 * to brelse() it when appropriate. 814 */ 815static struct buffer_head * ext3_find_entry (struct dentry *dentry, 816 struct ext3_dir_entry_2 ** res_dir) 817{ 818 struct super_block * sb; 819 struct buffer_head * bh_use[NAMEI_RA_SIZE]; 820 struct buffer_head * bh, *ret = NULL; 821 unsigned long start, block, b; 822 int ra_max = 0; /* Number of bh's in the readahead 823 buffer, bh_use[] */ 824 int ra_ptr = 0; /* Current index into readahead 825 buffer */ 826 int num = 0; 827 int nblocks, i, err; 828 struct inode *dir = dentry->d_parent->d_inode; 829 int namelen; 830 const u8 *name; 831 unsigned blocksize; 832 833 *res_dir = NULL; 834 sb = dir->i_sb; 835 blocksize = sb->s_blocksize; 836 namelen = dentry->d_name.len; 837 name = dentry->d_name.name; 838 if (namelen > EXT3_NAME_LEN) 839 return NULL; 840#ifdef CONFIG_EXT3_INDEX 841 if (is_dx(dir)) { 842 bh = ext3_dx_find_entry(dentry, res_dir, &err); 843 /* 844 * On success, or if the error was file not found, 845 * return. Otherwise, fall back to doing a search the 846 * old fashioned way. 847 */ 848 if (bh || (err != ERR_BAD_DX_DIR)) 849 return bh; 850 dxtrace(printk("ext3_find_entry: dx failed, falling back\n")); 851 } 852#endif 853 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb); 854 start = EXT3_I(dir)->i_dir_start_lookup; 855 if (start >= nblocks) 856 start = 0; 857 block = start; 858restart: 859 do { 860 /* 861 * We deal with the read-ahead logic here. 862 */ 863 if (ra_ptr >= ra_max) { 864 /* Refill the readahead buffer */ 865 ra_ptr = 0; 866 b = block; 867 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) { 868 /* 869 * Terminate if we reach the end of the 870 * directory and must wrap, or if our 871 * search has finished at this block. 872 */ 873 if (b >= nblocks || (num && block == start)) { 874 bh_use[ra_max] = NULL; 875 break; 876 } 877 num++; 878 bh = ext3_getblk(NULL, dir, b++, 0, &err); 879 bh_use[ra_max] = bh; 880 if (bh) 881 ll_rw_block(READ_META, 1, &bh); 882 } 883 } 884 if ((bh = bh_use[ra_ptr++]) == NULL) 885 goto next; 886 wait_on_buffer(bh); 887 if (!buffer_uptodate(bh)) { 888 /* read error, skip block & hope for the best */ 889 ext3_error(sb, __FUNCTION__, "reading directory #%lu " 890 "offset %lu", dir->i_ino, block); 891 brelse(bh); 892 goto next; 893 } 894 i = search_dirblock(bh, dir, dentry, 895 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir); 896 if (i == 1) { 897 EXT3_I(dir)->i_dir_start_lookup = block; 898 ret = bh; 899 goto cleanup_and_exit; 900 } else { 901 brelse(bh); 902 if (i < 0) 903 goto cleanup_and_exit; 904 } 905 next: 906 if (++block >= nblocks) 907 block = 0; 908 } while (block != start); 909 910 /* 911 * If the directory has grown while we were searching, then 912 * search the last part of the directory before giving up. 913 */ 914 block = nblocks; 915 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb); 916 if (block < nblocks) { 917 start = 0; 918 goto restart; 919 } 920 921cleanup_and_exit: 922 /* Clean up the read-ahead blocks */ 923 for (; ra_ptr < ra_max; ra_ptr++) 924 brelse (bh_use[ra_ptr]); 925 return ret; 926} 927 928#ifdef CONFIG_EXT3_INDEX 929static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry, 930 struct ext3_dir_entry_2 **res_dir, int *err) 931{ 932 struct super_block * sb; 933 struct dx_hash_info hinfo; 934 u32 hash; 935 struct dx_frame frames[2], *frame; 936 struct ext3_dir_entry_2 *de, *top; 937 struct buffer_head *bh; 938 unsigned long block; 939 int retval; 940 int namelen = dentry->d_name.len; 941 const u8 *name = dentry->d_name.name; 942 struct inode *dir = dentry->d_parent->d_inode; 943 944 sb = dir->i_sb; 945 /* NFS may look up ".." - look at dx_root directory block */ 946 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){ 947 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err))) 948 return NULL; 949 } else { 950 frame = frames; 951 frame->bh = NULL; /* for dx_release() */ 952 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/ 953 dx_set_block(frame->at, 0); /* dx_root block is 0 */ 954 } 955 hash = hinfo.hash; 956 do { 957 block = dx_get_block(frame->at); 958 if (!(bh = ext3_bread (NULL,dir, block, 0, err))) 959 goto errout; 960 de = (struct ext3_dir_entry_2 *) bh->b_data; 961 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize - 962 EXT3_DIR_REC_LEN(0)); 963 for (; de < top; de = ext3_next_entry(de)) 964 if (ext3_match (namelen, name, de)) { 965 if (!ext3_check_dir_entry("ext3_find_entry", 966 dir, de, bh, 967 (block<<EXT3_BLOCK_SIZE_BITS(sb)) 968 +((char *)de - bh->b_data))) { 969 brelse (bh); 970 *err = ERR_BAD_DX_DIR; 971 goto errout; 972 } 973 *res_dir = de; 974 dx_release (frames); 975 return bh; 976 } 977 brelse (bh); 978 /* Check to see if we should continue to search */ 979 retval = ext3_htree_next_block(dir, hash, frame, 980 frames, NULL); 981 if (retval < 0) { 982 ext3_warning(sb, __FUNCTION__, 983 "error reading index page in directory #%lu", 984 dir->i_ino); 985 *err = retval; 986 goto errout; 987 } 988 } while (retval == 1); 989 990 *err = -ENOENT; 991errout: 992 dxtrace(printk("%s not found\n", name)); 993 dx_release (frames); 994 return NULL; 995} 996#endif 997 998static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd) 999{ 1000 struct inode * inode; 1001 struct ext3_dir_entry_2 * de; 1002 struct buffer_head * bh; 1003 1004 if (dentry->d_name.len > EXT3_NAME_LEN) 1005 return ERR_PTR(-ENAMETOOLONG); 1006 1007 bh = ext3_find_entry(dentry, &de); 1008 inode = NULL; 1009 if (bh) { 1010 unsigned long ino = le32_to_cpu(de->inode); 1011 brelse (bh); 1012 if (!ext3_valid_inum(dir->i_sb, ino)) { 1013 ext3_error(dir->i_sb, "ext3_lookup", 1014 "bad inode number: %lu", ino); 1015 inode = NULL; 1016 } else 1017 inode = iget(dir->i_sb, ino); 1018 1019 if (!inode) 1020 return ERR_PTR(-EACCES); 1021 } 1022 return d_splice_alias(inode, dentry); 1023} 1024 1025 1026struct dentry *ext3_get_parent(struct dentry *child) 1027{ 1028 unsigned long ino; 1029 struct dentry *parent; 1030 struct inode *inode; 1031 struct dentry dotdot; 1032 struct ext3_dir_entry_2 * de; 1033 struct buffer_head *bh; 1034 1035 dotdot.d_name.name = ".."; 1036 dotdot.d_name.len = 2; 1037 dotdot.d_parent = child; /* confusing, isn't it! */ 1038 1039 bh = ext3_find_entry(&dotdot, &de); 1040 inode = NULL; 1041 if (!bh) 1042 return ERR_PTR(-ENOENT); 1043 ino = le32_to_cpu(de->inode); 1044 brelse(bh); 1045 1046 if (!ext3_valid_inum(child->d_inode->i_sb, ino)) { 1047 ext3_error(child->d_inode->i_sb, "ext3_get_parent", 1048 "bad inode number: %lu", ino); 1049 inode = NULL; 1050 } else 1051 inode = iget(child->d_inode->i_sb, ino); 1052 1053 if (!inode) 1054 return ERR_PTR(-EACCES); 1055 1056 parent = d_alloc_anon(inode); 1057 if (!parent) { 1058 iput(inode); 1059 parent = ERR_PTR(-ENOMEM); 1060 } 1061 return parent; 1062} 1063 1064#define S_SHIFT 12 1065static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = { 1066 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE, 1067 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR, 1068 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV, 1069 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV, 1070 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO, 1071 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK, 1072 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK, 1073}; 1074 1075static inline void ext3_set_de_type(struct super_block *sb, 1076 struct ext3_dir_entry_2 *de, 1077 umode_t mode) { 1078 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE)) 1079 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT]; 1080} 1081 1082#ifdef CONFIG_EXT3_INDEX 1083static struct ext3_dir_entry_2 * 1084dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count) 1085{ 1086 unsigned rec_len = 0; 1087 1088 while (count--) { 1089 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs); 1090 rec_len = EXT3_DIR_REC_LEN(de->name_len); 1091 memcpy (to, de, rec_len); 1092 ((struct ext3_dir_entry_2 *) to)->rec_len = 1093 cpu_to_le16(rec_len); 1094 de->inode = 0; 1095 map++; 1096 to += rec_len; 1097 } 1098 return (struct ext3_dir_entry_2 *) (to - rec_len); 1099} 1100 1101static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size) 1102{ 1103 struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base; 1104 unsigned rec_len = 0; 1105 1106 prev = to = de; 1107 while ((char*)de < base + size) { 1108 next = (struct ext3_dir_entry_2 *) ((char *) de + 1109 le16_to_cpu(de->rec_len)); 1110 if (de->inode && de->name_len) { 1111 rec_len = EXT3_DIR_REC_LEN(de->name_len); 1112 if (de > to) 1113 memmove(to, de, rec_len); 1114 to->rec_len = cpu_to_le16(rec_len); 1115 prev = to; 1116 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len); 1117 } 1118 de = next; 1119 } 1120 return prev; 1121} 1122 1123static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, 1124 struct buffer_head **bh,struct dx_frame *frame, 1125 struct dx_hash_info *hinfo, int *error) 1126{ 1127 unsigned blocksize = dir->i_sb->s_blocksize; 1128 unsigned count, continued; 1129 struct buffer_head *bh2; 1130 u32 newblock; 1131 u32 hash2; 1132 struct dx_map_entry *map; 1133 char *data1 = (*bh)->b_data, *data2; 1134 unsigned split; 1135 struct ext3_dir_entry_2 *de = NULL, *de2; 1136 int err = 0; 1137 1138 bh2 = ext3_append (handle, dir, &newblock, &err); 1139 if (!(bh2)) { 1140 brelse(*bh); 1141 *bh = NULL; 1142 goto errout; 1143 } 1144 1145 BUFFER_TRACE(*bh, "get_write_access"); 1146 err = ext3_journal_get_write_access(handle, *bh); 1147 if (err) 1148 goto journal_error; 1149 1150 BUFFER_TRACE(frame->bh, "get_write_access"); 1151 err = ext3_journal_get_write_access(handle, frame->bh); 1152 if (err) 1153 goto journal_error; 1154 1155 data2 = bh2->b_data; 1156 1157 /* create map in the end of data2 block */ 1158 map = (struct dx_map_entry *) (data2 + blocksize); 1159 count = dx_make_map ((struct ext3_dir_entry_2 *) data1, 1160 blocksize, hinfo, map); 1161 map -= count; 1162 split = count/2; // need to adjust to actual middle 1163 dx_sort_map (map, count); 1164 hash2 = map[split].hash; 1165 continued = hash2 == map[split - 1].hash; 1166 dxtrace(printk("Split block %i at %x, %i/%i\n", 1167 dx_get_block(frame->at), hash2, split, count-split)); 1168 1169 /* Fancy dance to stay within two buffers */ 1170 de2 = dx_move_dirents(data1, data2, map + split, count - split); 1171 de = dx_pack_dirents(data1,blocksize); 1172 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de); 1173 de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2); 1174 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1)); 1175 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1)); 1176 1177 /* Which block gets the new entry? */ 1178 if (hinfo->hash >= hash2) 1179 { 1180 swap(*bh, bh2); 1181 de = de2; 1182 } 1183 dx_insert_block (frame, hash2 + continued, newblock); 1184 err = ext3_journal_dirty_metadata (handle, bh2); 1185 if (err) 1186 goto journal_error; 1187 err = ext3_journal_dirty_metadata (handle, frame->bh); 1188 if (err) 1189 goto journal_error; 1190 brelse (bh2); 1191 dxtrace(dx_show_index ("frame", frame->entries)); 1192 return de; 1193 1194journal_error: 1195 brelse(*bh); 1196 brelse(bh2); 1197 *bh = NULL; 1198 ext3_std_error(dir->i_sb, err); 1199errout: 1200 *error = err; 1201 return NULL; 1202} 1203#endif 1204 1205 1206/* 1207 * Add a new entry into a directory (leaf) block. If de is non-NULL, 1208 * it points to a directory entry which is guaranteed to be large 1209 * enough for new directory entry. If de is NULL, then 1210 * add_dirent_to_buf will attempt search the directory block for 1211 * space. It will return -ENOSPC if no space is available, and -EIO 1212 * and -EEXIST if directory entry already exists. 1213 * 1214 * NOTE! bh is NOT released in the case where ENOSPC is returned. In 1215 * all other cases bh is released. 1216 */ 1217static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry, 1218 struct inode *inode, struct ext3_dir_entry_2 *de, 1219 struct buffer_head * bh) 1220{ 1221 struct inode *dir = dentry->d_parent->d_inode; 1222 const char *name = dentry->d_name.name; 1223 int namelen = dentry->d_name.len; 1224 unsigned long offset = 0; 1225 unsigned short reclen; 1226 int nlen, rlen, err; 1227 char *top; 1228 1229 reclen = EXT3_DIR_REC_LEN(namelen); 1230 if (!de) { 1231 de = (struct ext3_dir_entry_2 *)bh->b_data; 1232 top = bh->b_data + dir->i_sb->s_blocksize - reclen; 1233 while ((char *) de <= top) { 1234 if (!ext3_check_dir_entry("ext3_add_entry", dir, de, 1235 bh, offset)) { 1236 brelse (bh); 1237 return -EIO; 1238 } 1239 if (ext3_match (namelen, name, de)) { 1240 brelse (bh); 1241 return -EEXIST; 1242 } 1243 nlen = EXT3_DIR_REC_LEN(de->name_len); 1244 rlen = le16_to_cpu(de->rec_len); 1245 if ((de->inode? rlen - nlen: rlen) >= reclen) 1246 break; 1247 de = (struct ext3_dir_entry_2 *)((char *)de + rlen); 1248 offset += rlen; 1249 } 1250 if ((char *) de > top) 1251 return -ENOSPC; 1252 } 1253 BUFFER_TRACE(bh, "get_write_access"); 1254 err = ext3_journal_get_write_access(handle, bh); 1255 if (err) { 1256 ext3_std_error(dir->i_sb, err); 1257 brelse(bh); 1258 return err; 1259 } 1260 1261 /* By now the buffer is marked for journaling */ 1262 nlen = EXT3_DIR_REC_LEN(de->name_len); 1263 rlen = le16_to_cpu(de->rec_len); 1264 if (de->inode) { 1265 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen); 1266 de1->rec_len = cpu_to_le16(rlen - nlen); 1267 de->rec_len = cpu_to_le16(nlen); 1268 de = de1; 1269 } 1270 de->file_type = EXT3_FT_UNKNOWN; 1271 if (inode) { 1272 de->inode = cpu_to_le32(inode->i_ino); 1273 ext3_set_de_type(dir->i_sb, de, inode->i_mode); 1274 } else 1275 de->inode = 0; 1276 de->name_len = namelen; 1277 memcpy (de->name, name, namelen); 1278 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC; 1279 ext3_update_dx_flag(dir); 1280 dir->i_version++; 1281 ext3_mark_inode_dirty(handle, dir); 1282 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); 1283 err = ext3_journal_dirty_metadata(handle, bh); 1284 if (err) 1285 ext3_std_error(dir->i_sb, err); 1286 brelse(bh); 1287 return 0; 1288} 1289 1290#ifdef CONFIG_EXT3_INDEX 1291/* 1292 * This converts a one block unindexed directory to a 3 block indexed 1293 * directory, and adds the dentry to the indexed directory. 1294 */ 1295static int make_indexed_dir(handle_t *handle, struct dentry *dentry, 1296 struct inode *inode, struct buffer_head *bh) 1297{ 1298 struct inode *dir = dentry->d_parent->d_inode; 1299 const char *name = dentry->d_name.name; 1300 int namelen = dentry->d_name.len; 1301 struct buffer_head *bh2; 1302 struct dx_root *root; 1303 struct dx_frame frames[2], *frame; 1304 struct dx_entry *entries; 1305 struct ext3_dir_entry_2 *de, *de2; 1306 char *data1, *top; 1307 unsigned len; 1308 int retval; 1309 unsigned blocksize; 1310 struct dx_hash_info hinfo; 1311 u32 block; 1312 struct fake_dirent *fde; 1313 1314 blocksize = dir->i_sb->s_blocksize; 1315 dxtrace(printk("Creating index\n")); 1316 retval = ext3_journal_get_write_access(handle, bh); 1317 if (retval) { 1318 ext3_std_error(dir->i_sb, retval); 1319 brelse(bh); 1320 return retval; 1321 } 1322 root = (struct dx_root *) bh->b_data; 1323 1324 bh2 = ext3_append (handle, dir, &block, &retval); 1325 if (!(bh2)) { 1326 brelse(bh); 1327 return retval; 1328 } 1329 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL; 1330 data1 = bh2->b_data; 1331 1332 /* The 0th block becomes the root, move the dirents out */ 1333 fde = &root->dotdot; 1334 de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len)); 1335 len = ((char *) root) + blocksize - (char *) de; 1336 memcpy (data1, de, len); 1337 de = (struct ext3_dir_entry_2 *) data1; 1338 top = data1 + len; 1339 while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top) 1340 de = de2; 1341 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de); 1342 /* Initialize the root; the dot dirents already exist */ 1343 de = (struct ext3_dir_entry_2 *) (&root->dotdot); 1344 de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2)); 1345 memset (&root->info, 0, sizeof(root->info)); 1346 root->info.info_length = sizeof(root->info); 1347 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version; 1348 entries = root->entries; 1349 dx_set_block (entries, 1); 1350 dx_set_count (entries, 1); 1351 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info))); 1352 1353 /* Initialize as for dx_probe */ 1354 hinfo.hash_version = root->info.hash_version; 1355 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed; 1356 ext3fs_dirhash(name, namelen, &hinfo); 1357 frame = frames; 1358 frame->entries = entries; 1359 frame->at = entries; 1360 frame->bh = bh; 1361 bh = bh2; 1362 de = do_split(handle,dir, &bh, frame, &hinfo, &retval); 1363 dx_release (frames); 1364 if (!(de)) 1365 return retval; 1366 1367 return add_dirent_to_buf(handle, dentry, inode, de, bh); 1368} 1369#endif 1370 1371/* 1372 * ext3_add_entry() 1373 * 1374 * adds a file entry to the specified directory, using the same 1375 * semantics as ext3_find_entry(). It returns NULL if it failed. 1376 * 1377 * NOTE!! The inode part of 'de' is left at 0 - which means you 1378 * may not sleep between calling this and putting something into 1379 * the entry, as someone else might have used it while you slept. 1380 */ 1381static int ext3_add_entry (handle_t *handle, struct dentry *dentry, 1382 struct inode *inode) 1383{ 1384 struct inode *dir = dentry->d_parent->d_inode; 1385 unsigned long offset; 1386 struct buffer_head * bh; 1387 struct ext3_dir_entry_2 *de; 1388 struct super_block * sb; 1389 int retval; 1390#ifdef CONFIG_EXT3_INDEX 1391 int dx_fallback=0; 1392#endif 1393 unsigned blocksize; 1394 u32 block, blocks; 1395 1396 sb = dir->i_sb; 1397 blocksize = sb->s_blocksize; 1398 if (!dentry->d_name.len) 1399 return -EINVAL; 1400#ifdef CONFIG_EXT3_INDEX 1401 if (is_dx(dir)) { 1402 retval = ext3_dx_add_entry(handle, dentry, inode); 1403 if (!retval || (retval != ERR_BAD_DX_DIR)) 1404 return retval; 1405 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL; 1406 dx_fallback++; 1407 ext3_mark_inode_dirty(handle, dir); 1408 } 1409#endif 1410 blocks = dir->i_size >> sb->s_blocksize_bits; 1411 for (block = 0, offset = 0; block < blocks; block++) { 1412 bh = ext3_bread(handle, dir, block, 0, &retval); 1413 if(!bh) 1414 return retval; 1415 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh); 1416 if (retval != -ENOSPC) 1417 return retval; 1418 1419#ifdef CONFIG_EXT3_INDEX 1420 if (blocks == 1 && !dx_fallback && 1421 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX)) 1422 return make_indexed_dir(handle, dentry, inode, bh); 1423#endif 1424 brelse(bh); 1425 } 1426 bh = ext3_append(handle, dir, &block, &retval); 1427 if (!bh) 1428 return retval; 1429 de = (struct ext3_dir_entry_2 *) bh->b_data; 1430 de->inode = 0; 1431 de->rec_len = cpu_to_le16(blocksize); 1432 return add_dirent_to_buf(handle, dentry, inode, de, bh); 1433} 1434 1435#ifdef CONFIG_EXT3_INDEX 1436/* 1437 * Returns 0 for success, or a negative error value 1438 */ 1439static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry, 1440 struct inode *inode) 1441{ 1442 struct dx_frame frames[2], *frame; 1443 struct dx_entry *entries, *at; 1444 struct dx_hash_info hinfo; 1445 struct buffer_head * bh; 1446 struct inode *dir = dentry->d_parent->d_inode; 1447 struct super_block * sb = dir->i_sb; 1448 struct ext3_dir_entry_2 *de; 1449 int err; 1450 1451 frame = dx_probe(dentry, NULL, &hinfo, frames, &err); 1452 if (!frame) 1453 return err; 1454 entries = frame->entries; 1455 at = frame->at; 1456 1457 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err))) 1458 goto cleanup; 1459 1460 BUFFER_TRACE(bh, "get_write_access"); 1461 err = ext3_journal_get_write_access(handle, bh); 1462 if (err) 1463 goto journal_error; 1464 1465 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh); 1466 if (err != -ENOSPC) { 1467 bh = NULL; 1468 goto cleanup; 1469 } 1470 1471 /* Block full, should compress but for now just split */ 1472 dxtrace(printk("using %u of %u node entries\n", 1473 dx_get_count(entries), dx_get_limit(entries))); 1474 /* Need to split index? */ 1475 if (dx_get_count(entries) == dx_get_limit(entries)) { 1476 u32 newblock; 1477 unsigned icount = dx_get_count(entries); 1478 int levels = frame - frames; 1479 struct dx_entry *entries2; 1480 struct dx_node *node2; 1481 struct buffer_head *bh2; 1482 1483 if (levels && (dx_get_count(frames->entries) == 1484 dx_get_limit(frames->entries))) { 1485 ext3_warning(sb, __FUNCTION__, 1486 "Directory index full!"); 1487 err = -ENOSPC; 1488 goto cleanup; 1489 } 1490 bh2 = ext3_append (handle, dir, &newblock, &err); 1491 if (!(bh2)) 1492 goto cleanup; 1493 node2 = (struct dx_node *)(bh2->b_data); 1494 entries2 = node2->entries; 1495 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize); 1496 node2->fake.inode = 0; 1497 BUFFER_TRACE(frame->bh, "get_write_access"); 1498 err = ext3_journal_get_write_access(handle, frame->bh); 1499 if (err) 1500 goto journal_error; 1501 if (levels) { 1502 unsigned icount1 = icount/2, icount2 = icount - icount1; 1503 unsigned hash2 = dx_get_hash(entries + icount1); 1504 dxtrace(printk("Split index %i/%i\n", icount1, icount2)); 1505 1506 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */ 1507 err = ext3_journal_get_write_access(handle, 1508 frames[0].bh); 1509 if (err) 1510 goto journal_error; 1511 1512 memcpy ((char *) entries2, (char *) (entries + icount1), 1513 icount2 * sizeof(struct dx_entry)); 1514 dx_set_count (entries, icount1); 1515 dx_set_count (entries2, icount2); 1516 dx_set_limit (entries2, dx_node_limit(dir)); 1517 1518 /* Which index block gets the new entry? */ 1519 if (at - entries >= icount1) { 1520 frame->at = at = at - entries - icount1 + entries2; 1521 frame->entries = entries = entries2; 1522 swap(frame->bh, bh2); 1523 } 1524 dx_insert_block (frames + 0, hash2, newblock); 1525 dxtrace(dx_show_index ("node", frames[1].entries)); 1526 dxtrace(dx_show_index ("node", 1527 ((struct dx_node *) bh2->b_data)->entries)); 1528 err = ext3_journal_dirty_metadata(handle, bh2); 1529 if (err) 1530 goto journal_error; 1531 brelse (bh2); 1532 } else { 1533 dxtrace(printk("Creating second level index...\n")); 1534 memcpy((char *) entries2, (char *) entries, 1535 icount * sizeof(struct dx_entry)); 1536 dx_set_limit(entries2, dx_node_limit(dir)); 1537 1538 /* Set up root */ 1539 dx_set_count(entries, 1); 1540 dx_set_block(entries + 0, newblock); 1541 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1; 1542 1543 /* Add new access path frame */ 1544 frame = frames + 1; 1545 frame->at = at = at - entries + entries2; 1546 frame->entries = entries = entries2; 1547 frame->bh = bh2; 1548 err = ext3_journal_get_write_access(handle, 1549 frame->bh); 1550 if (err) 1551 goto journal_error; 1552 } 1553 ext3_journal_dirty_metadata(handle, frames[0].bh); 1554 } 1555 de = do_split(handle, dir, &bh, frame, &hinfo, &err); 1556 if (!de) 1557 goto cleanup; 1558 err = add_dirent_to_buf(handle, dentry, inode, de, bh); 1559 bh = NULL; 1560 goto cleanup; 1561 1562journal_error: 1563 ext3_std_error(dir->i_sb, err); 1564cleanup: 1565 if (bh) 1566 brelse(bh); 1567 dx_release(frames); 1568 return err; 1569} 1570#endif 1571 1572/* 1573 * ext3_delete_entry deletes a directory entry by merging it with the 1574 * previous entry 1575 */ 1576static int ext3_delete_entry (handle_t *handle, 1577 struct inode * dir, 1578 struct ext3_dir_entry_2 * de_del, 1579 struct buffer_head * bh) 1580{ 1581 struct ext3_dir_entry_2 * de, * pde; 1582 int i; 1583 1584 i = 0; 1585 pde = NULL; 1586 de = (struct ext3_dir_entry_2 *) bh->b_data; 1587 while (i < bh->b_size) { 1588 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i)) 1589 return -EIO; 1590 if (de == de_del) { 1591 BUFFER_TRACE(bh, "get_write_access"); 1592 ext3_journal_get_write_access(handle, bh); 1593 if (pde) 1594 pde->rec_len = 1595 cpu_to_le16(le16_to_cpu(pde->rec_len) + 1596 le16_to_cpu(de->rec_len)); 1597 else 1598 de->inode = 0; 1599 dir->i_version++; 1600 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); 1601 ext3_journal_dirty_metadata(handle, bh); 1602 return 0; 1603 } 1604 i += le16_to_cpu(de->rec_len); 1605 pde = de; 1606 de = (struct ext3_dir_entry_2 *) 1607 ((char *) de + le16_to_cpu(de->rec_len)); 1608 } 1609 return -ENOENT; 1610} 1611 1612static int ext3_add_nondir(handle_t *handle, 1613 struct dentry *dentry, struct inode *inode) 1614{ 1615 int err = ext3_add_entry(handle, dentry, inode); 1616 if (!err) { 1617 ext3_mark_inode_dirty(handle, inode); 1618 d_instantiate(dentry, inode); 1619 return 0; 1620 } 1621 drop_nlink(inode); 1622 iput(inode); 1623 return err; 1624} 1625 1626/* 1627 * By the time this is called, we already have created 1628 * the directory cache entry for the new file, but it 1629 * is so far negative - it has no inode. 1630 * 1631 * If the create succeeds, we fill in the inode information 1632 * with d_instantiate(). 1633 */ 1634static int ext3_create (struct inode * dir, struct dentry * dentry, int mode, 1635 struct nameidata *nd) 1636{ 1637 handle_t *handle; 1638 struct inode * inode; 1639 int err, retries = 0; 1640 1641retry: 1642 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + 1643 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 + 1644 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb)); 1645 if (IS_ERR(handle)) 1646 return PTR_ERR(handle); 1647 1648 if (IS_DIRSYNC(dir)) 1649 handle->h_sync = 1; 1650 1651 inode = ext3_new_inode (handle, dir, mode); 1652 err = PTR_ERR(inode); 1653 if (!IS_ERR(inode)) { 1654 inode->i_op = &ext3_file_inode_operations; 1655 inode->i_fop = &ext3_file_operations; 1656 ext3_set_aops(inode); 1657 err = ext3_add_nondir(handle, dentry, inode); 1658 } 1659 ext3_journal_stop(handle); 1660 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) 1661 goto retry; 1662 return err; 1663} 1664 1665static int ext3_mknod (struct inode * dir, struct dentry *dentry, 1666 int mode, dev_t rdev) 1667{ 1668 handle_t *handle; 1669 struct inode *inode; 1670 int err, retries = 0; 1671 1672 if (!new_valid_dev(rdev)) 1673 return -EINVAL; 1674 1675retry: 1676 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + 1677 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 + 1678 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb)); 1679 if (IS_ERR(handle)) 1680 return PTR_ERR(handle); 1681 1682 if (IS_DIRSYNC(dir)) 1683 handle->h_sync = 1; 1684 1685 inode = ext3_new_inode (handle, dir, mode); 1686 err = PTR_ERR(inode); 1687 if (!IS_ERR(inode)) { 1688 init_special_inode(inode, inode->i_mode, rdev); 1689#ifdef CONFIG_EXT3_FS_XATTR 1690 inode->i_op = &ext3_special_inode_operations; 1691#endif 1692 err = ext3_add_nondir(handle, dentry, inode); 1693 } 1694 ext3_journal_stop(handle); 1695 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) 1696 goto retry; 1697 return err; 1698} 1699 1700static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode) 1701{ 1702 handle_t *handle; 1703 struct inode * inode; 1704 struct buffer_head * dir_block; 1705 struct ext3_dir_entry_2 * de; 1706 int err, retries = 0; 1707 1708 if (dir->i_nlink >= EXT3_LINK_MAX) 1709 return -EMLINK; 1710 1711retry: 1712 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + 1713 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 + 1714 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb)); 1715 if (IS_ERR(handle)) 1716 return PTR_ERR(handle); 1717 1718 if (IS_DIRSYNC(dir)) 1719 handle->h_sync = 1; 1720 1721 inode = ext3_new_inode (handle, dir, S_IFDIR | mode); 1722 err = PTR_ERR(inode); 1723 if (IS_ERR(inode)) 1724 goto out_stop; 1725 1726 inode->i_op = &ext3_dir_inode_operations; 1727 inode->i_fop = &ext3_dir_operations; 1728 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize; 1729 dir_block = ext3_bread (handle, inode, 0, 1, &err); 1730 if (!dir_block) { 1731 drop_nlink(inode); /* is this nlink == 0? */ 1732 ext3_mark_inode_dirty(handle, inode); 1733 iput (inode); 1734 goto out_stop; 1735 } 1736 BUFFER_TRACE(dir_block, "get_write_access"); 1737 ext3_journal_get_write_access(handle, dir_block); 1738 de = (struct ext3_dir_entry_2 *) dir_block->b_data; 1739 de->inode = cpu_to_le32(inode->i_ino); 1740 de->name_len = 1; 1741 de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len)); 1742 strcpy (de->name, "."); 1743 ext3_set_de_type(dir->i_sb, de, S_IFDIR); 1744 de = (struct ext3_dir_entry_2 *) 1745 ((char *) de + le16_to_cpu(de->rec_len)); 1746 de->inode = cpu_to_le32(dir->i_ino); 1747 de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1)); 1748 de->name_len = 2; 1749 strcpy (de->name, ".."); 1750 ext3_set_de_type(dir->i_sb, de, S_IFDIR); 1751 inode->i_nlink = 2; 1752 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata"); 1753 ext3_journal_dirty_metadata(handle, dir_block); 1754 brelse (dir_block); 1755 ext3_mark_inode_dirty(handle, inode); 1756 err = ext3_add_entry (handle, dentry, inode); 1757 if (err) { 1758 inode->i_nlink = 0; 1759 ext3_mark_inode_dirty(handle, inode); 1760 iput (inode); 1761 goto out_stop; 1762 } 1763 inc_nlink(dir); 1764 ext3_update_dx_flag(dir); 1765 ext3_mark_inode_dirty(handle, dir); 1766 d_instantiate(dentry, inode); 1767out_stop: 1768 ext3_journal_stop(handle); 1769 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) 1770 goto retry; 1771 return err; 1772} 1773 1774/* 1775 * routine to check that the specified directory is empty (for rmdir) 1776 */ 1777static int empty_dir (struct inode * inode) 1778{ 1779 unsigned long offset; 1780 struct buffer_head * bh; 1781 struct ext3_dir_entry_2 * de, * de1; 1782 struct super_block * sb; 1783 int err = 0; 1784 1785 sb = inode->i_sb; 1786 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) || 1787 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) { 1788 if (err) 1789 ext3_error(inode->i_sb, __FUNCTION__, 1790 "error %d reading directory #%lu offset 0", 1791 err, inode->i_ino); 1792 else 1793 ext3_warning(inode->i_sb, __FUNCTION__, 1794 "bad directory (dir #%lu) - no data block", 1795 inode->i_ino); 1796 return 1; 1797 } 1798 de = (struct ext3_dir_entry_2 *) bh->b_data; 1799 de1 = (struct ext3_dir_entry_2 *) 1800 ((char *) de + le16_to_cpu(de->rec_len)); 1801 if (le32_to_cpu(de->inode) != inode->i_ino || 1802 !le32_to_cpu(de1->inode) || 1803 strcmp (".", de->name) || 1804 strcmp ("..", de1->name)) { 1805 ext3_warning (inode->i_sb, "empty_dir", 1806 "bad directory (dir #%lu) - no `.' or `..'", 1807 inode->i_ino); 1808 brelse (bh); 1809 return 1; 1810 } 1811 offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len); 1812 de = (struct ext3_dir_entry_2 *) 1813 ((char *) de1 + le16_to_cpu(de1->rec_len)); 1814 while (offset < inode->i_size ) { 1815 if (!bh || 1816 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) { 1817 err = 0; 1818 brelse (bh); 1819 bh = ext3_bread (NULL, inode, 1820 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err); 1821 if (!bh) { 1822 if (err) 1823 ext3_error(sb, __FUNCTION__, 1824 "error %d reading directory" 1825 " #%lu offset %lu", 1826 err, inode->i_ino, offset); 1827 offset += sb->s_blocksize; 1828 continue; 1829 } 1830 de = (struct ext3_dir_entry_2 *) bh->b_data; 1831 } 1832 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) { 1833 de = (struct ext3_dir_entry_2 *)(bh->b_data + 1834 sb->s_blocksize); 1835 offset = (offset | (sb->s_blocksize - 1)) + 1; 1836 continue; 1837 } 1838 if (le32_to_cpu(de->inode)) { 1839 brelse (bh); 1840 return 0; 1841 } 1842 offset += le16_to_cpu(de->rec_len); 1843 de = (struct ext3_dir_entry_2 *) 1844 ((char *) de + le16_to_cpu(de->rec_len)); 1845 } 1846 brelse (bh); 1847 return 1; 1848} 1849 1850/* ext3_orphan_add() links an unlinked or truncated inode into a list of 1851 * such inodes, starting at the superblock, in case we crash before the 1852 * file is closed/deleted, or in case the inode truncate spans multiple 1853 * transactions and the last transaction is not recovered after a crash. 1854 * 1855 * At filesystem recovery time, we walk this list deleting unlinked 1856 * inodes and truncating linked inodes in ext3_orphan_cleanup(). 1857 */ 1858int ext3_orphan_add(handle_t *handle, struct inode *inode) 1859{ 1860 struct super_block *sb = inode->i_sb; 1861 struct ext3_iloc iloc; 1862 int err = 0, rc; 1863 1864 lock_super(sb); 1865 if (!list_empty(&EXT3_I(inode)->i_orphan)) 1866 goto out_unlock; 1867 1868 /* Orphan handling is only valid for files with data blocks 1869 * being truncated, or files being unlinked. */ 1870 1871 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1872 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0); 1873 1874 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access"); 1875 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh); 1876 if (err) 1877 goto out_unlock; 1878 1879 err = ext3_reserve_inode_write(handle, inode, &iloc); 1880 if (err) 1881 goto out_unlock; 1882 1883 /* Insert this inode at the head of the on-disk orphan list... */ 1884 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan); 1885 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino); 1886 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh); 1887 rc = ext3_mark_iloc_dirty(handle, inode, &iloc); 1888 if (!err) 1889 err = rc; 1890 1891 /* Only add to the head of the in-memory list if all the 1892 * previous operations succeeded. If the orphan_add is going to 1893 * fail (possibly taking the journal offline), we can't risk 1894 * leaving the inode on the orphan list: stray orphan-list 1895 * entries can cause panics at unmount time. 1896 * 1897 * This is safe: on error we're going to ignore the orphan list 1898 * anyway on the next recovery. */ 1899 if (!err) 1900 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan); 1901 1902 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino); 1903 jbd_debug(4, "orphan inode %lu will point to %d\n", 1904 inode->i_ino, NEXT_ORPHAN(inode)); 1905out_unlock: 1906 unlock_super(sb); 1907 ext3_std_error(inode->i_sb, err); 1908 return err; 1909} 1910 1911/* 1912 * ext3_orphan_del() removes an unlinked or truncated inode from the list 1913 * of such inodes stored on disk, because it is finally being cleaned up. 1914 */ 1915int ext3_orphan_del(handle_t *handle, struct inode *inode) 1916{ 1917 struct list_head *prev; 1918 struct ext3_inode_info *ei = EXT3_I(inode); 1919 struct ext3_sb_info *sbi; 1920 unsigned long ino_next; 1921 struct ext3_iloc iloc; 1922 int err = 0; 1923 1924 lock_super(inode->i_sb); 1925 if (list_empty(&ei->i_orphan)) { 1926 unlock_super(inode->i_sb); 1927 return 0; 1928 } 1929 1930 ino_next = NEXT_ORPHAN(inode); 1931 prev = ei->i_orphan.prev; 1932 sbi = EXT3_SB(inode->i_sb); 1933 1934 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino); 1935 1936 list_del_init(&ei->i_orphan); 1937 1938 /* If we're on an error path, we may not have a valid 1939 * transaction handle with which to update the orphan list on 1940 * disk, but we still need to remove the inode from the linked 1941 * list in memory. */ 1942 if (!handle) 1943 goto out; 1944 1945 err = ext3_reserve_inode_write(handle, inode, &iloc); 1946 if (err) 1947 goto out_err; 1948 1949 if (prev == &sbi->s_orphan) { 1950 jbd_debug(4, "superblock will point to %lu\n", ino_next); 1951 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 1952 err = ext3_journal_get_write_access(handle, sbi->s_sbh); 1953 if (err) 1954 goto out_brelse; 1955 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next); 1956 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh); 1957 } else { 1958 struct ext3_iloc iloc2; 1959 struct inode *i_prev = 1960 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode; 1961 1962 jbd_debug(4, "orphan inode %lu will point to %lu\n", 1963 i_prev->i_ino, ino_next); 1964 err = ext3_reserve_inode_write(handle, i_prev, &iloc2); 1965 if (err) 1966 goto out_brelse; 1967 NEXT_ORPHAN(i_prev) = ino_next; 1968 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2); 1969 } 1970 if (err) 1971 goto out_brelse; 1972 NEXT_ORPHAN(inode) = 0; 1973 err = ext3_mark_iloc_dirty(handle, inode, &iloc); 1974 1975out_err: 1976 ext3_std_error(inode->i_sb, err); 1977out: 1978 unlock_super(inode->i_sb); 1979 return err; 1980 1981out_brelse: 1982 brelse(iloc.bh); 1983 goto out_err; 1984} 1985 1986static int ext3_rmdir (struct inode * dir, struct dentry *dentry) 1987{ 1988 int retval; 1989 struct inode * inode; 1990 struct buffer_head * bh; 1991 struct ext3_dir_entry_2 * de; 1992 handle_t *handle; 1993 1994 /* Initialize quotas before so that eventual writes go in 1995 * separate transaction */ 1996 DQUOT_INIT(dentry->d_inode); 1997 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb)); 1998 if (IS_ERR(handle)) 1999 return PTR_ERR(handle); 2000 2001 retval = -ENOENT; 2002 bh = ext3_find_entry (dentry, &de); 2003 if (!bh) 2004 goto end_rmdir; 2005 2006 if (IS_DIRSYNC(dir)) 2007 handle->h_sync = 1; 2008 2009 inode = dentry->d_inode; 2010 2011 retval = -EIO; 2012 if (le32_to_cpu(de->inode) != inode->i_ino) 2013 goto end_rmdir; 2014 2015 retval = -ENOTEMPTY; 2016 if (!empty_dir (inode)) 2017 goto end_rmdir; 2018 2019 retval = ext3_delete_entry(handle, dir, de, bh); 2020 if (retval) 2021 goto end_rmdir; 2022 if (inode->i_nlink != 2) 2023 ext3_warning (inode->i_sb, "ext3_rmdir", 2024 "empty directory has nlink!=2 (%d)", 2025 inode->i_nlink); 2026 inode->i_version++; 2027 clear_nlink(inode); 2028 /* There's no need to set i_disksize: the fact that i_nlink is 2029 * zero will ensure that the right thing happens during any 2030 * recovery. */ 2031 inode->i_size = 0; 2032 ext3_orphan_add(handle, inode); 2033 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC; 2034 ext3_mark_inode_dirty(handle, inode); 2035 drop_nlink(dir); 2036 ext3_update_dx_flag(dir); 2037 ext3_mark_inode_dirty(handle, dir); 2038 2039end_rmdir: 2040 ext3_journal_stop(handle); 2041 brelse (bh); 2042 return retval; 2043} 2044 2045static int ext3_unlink(struct inode * dir, struct dentry *dentry) 2046{ 2047 int retval; 2048 struct inode * inode; 2049 struct buffer_head * bh; 2050 struct ext3_dir_entry_2 * de; 2051 handle_t *handle; 2052 2053 /* Initialize quotas before so that eventual writes go 2054 * in separate transaction */ 2055 DQUOT_INIT(dentry->d_inode); 2056 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb)); 2057 if (IS_ERR(handle)) 2058 return PTR_ERR(handle); 2059 2060 if (IS_DIRSYNC(dir)) 2061 handle->h_sync = 1; 2062 2063 retval = -ENOENT; 2064 bh = ext3_find_entry (dentry, &de); 2065 if (!bh) 2066 goto end_unlink; 2067 2068 inode = dentry->d_inode; 2069 2070 retval = -EIO; 2071 if (le32_to_cpu(de->inode) != inode->i_ino) 2072 goto end_unlink; 2073 2074 if (!inode->i_nlink) { 2075 ext3_warning (inode->i_sb, "ext3_unlink", 2076 "Deleting nonexistent file (%lu), %d", 2077 inode->i_ino, inode->i_nlink); 2078 inode->i_nlink = 1; 2079 } 2080 retval = ext3_delete_entry(handle, dir, de, bh); 2081 if (retval) 2082 goto end_unlink; 2083 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC; 2084 ext3_update_dx_flag(dir); 2085 ext3_mark_inode_dirty(handle, dir); 2086 drop_nlink(inode); 2087 if (!inode->i_nlink) 2088 ext3_orphan_add(handle, inode); 2089 inode->i_ctime = dir->i_ctime; 2090 ext3_mark_inode_dirty(handle, inode); 2091 retval = 0; 2092 2093end_unlink: 2094 ext3_journal_stop(handle); 2095 brelse (bh); 2096 return retval; 2097} 2098 2099static int ext3_symlink (struct inode * dir, 2100 struct dentry *dentry, const char * symname) 2101{ 2102 handle_t *handle; 2103 struct inode * inode; 2104 int l, err, retries = 0; 2105 2106 l = strlen(symname)+1; 2107 if (l > dir->i_sb->s_blocksize) 2108 return -ENAMETOOLONG; 2109 2110retry: 2111 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + 2112 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 + 2113 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb)); 2114 if (IS_ERR(handle)) 2115 return PTR_ERR(handle); 2116 2117 if (IS_DIRSYNC(dir)) 2118 handle->h_sync = 1; 2119 2120 inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO); 2121 err = PTR_ERR(inode); 2122 if (IS_ERR(inode)) 2123 goto out_stop; 2124 2125 if (l > sizeof (EXT3_I(inode)->i_data)) { 2126 inode->i_op = &ext3_symlink_inode_operations; 2127 ext3_set_aops(inode); 2128 /* 2129 * page_symlink() calls into ext3_prepare/commit_write. 2130 * We have a transaction open. All is sweetness. It also sets 2131 * i_size in generic_commit_write(). 2132 */ 2133 err = __page_symlink(inode, symname, l, 2134 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS); 2135 if (err) { 2136 drop_nlink(inode); 2137 ext3_mark_inode_dirty(handle, inode); 2138 iput (inode); 2139 goto out_stop; 2140 } 2141 } else { 2142 inode->i_op = &ext3_fast_symlink_inode_operations; 2143 memcpy((char*)&EXT3_I(inode)->i_data,symname,l); 2144 inode->i_size = l-1; 2145 } 2146 EXT3_I(inode)->i_disksize = inode->i_size; 2147 err = ext3_add_nondir(handle, dentry, inode); 2148out_stop: 2149 ext3_journal_stop(handle); 2150 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) 2151 goto retry; 2152 return err; 2153} 2154 2155static int ext3_link (struct dentry * old_dentry, 2156 struct inode * dir, struct dentry *dentry) 2157{ 2158 handle_t *handle; 2159 struct inode *inode = old_dentry->d_inode; 2160 int err, retries = 0; 2161 2162 if (inode->i_nlink >= EXT3_LINK_MAX) 2163 return -EMLINK; 2164 /* 2165 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing 2166 * otherwise has the potential to corrupt the orphan inode list. 2167 */ 2168 if (inode->i_nlink == 0) 2169 return -ENOENT; 2170 2171retry: 2172 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + 2173 EXT3_INDEX_EXTRA_TRANS_BLOCKS); 2174 if (IS_ERR(handle)) 2175 return PTR_ERR(handle); 2176 2177 if (IS_DIRSYNC(dir)) 2178 handle->h_sync = 1; 2179 2180 inode->i_ctime = CURRENT_TIME_SEC; 2181 inc_nlink(inode); 2182 atomic_inc(&inode->i_count); 2183 2184 err = ext3_add_nondir(handle, dentry, inode); 2185 ext3_journal_stop(handle); 2186 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) 2187 goto retry; 2188 return err; 2189} 2190 2191#define PARENT_INO(buffer) \ 2192 ((struct ext3_dir_entry_2 *) ((char *) buffer + \ 2193 le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode 2194 2195/* 2196 * Anybody can rename anything with this: the permission checks are left to the 2197 * higher-level routines. 2198 */ 2199static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry, 2200 struct inode * new_dir,struct dentry *new_dentry) 2201{ 2202 handle_t *handle; 2203 struct inode * old_inode, * new_inode; 2204 struct buffer_head * old_bh, * new_bh, * dir_bh; 2205 struct ext3_dir_entry_2 * old_de, * new_de; 2206 int retval; 2207 2208 old_bh = new_bh = dir_bh = NULL; 2209 2210 /* Initialize quotas before so that eventual writes go 2211 * in separate transaction */ 2212 if (new_dentry->d_inode) 2213 DQUOT_INIT(new_dentry->d_inode); 2214 handle = ext3_journal_start(old_dir, 2 * 2215 EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) + 2216 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2); 2217 if (IS_ERR(handle)) 2218 return PTR_ERR(handle); 2219 2220 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir)) 2221 handle->h_sync = 1; 2222 2223 old_bh = ext3_find_entry (old_dentry, &old_de); 2224 /* 2225 * Check for inode number is _not_ due to possible IO errors. 2226 * We might rmdir the source, keep it as pwd of some process 2227 * and merrily kill the link to whatever was created under the 2228 * same name. Goodbye sticky bit ;-< 2229 */ 2230 old_inode = old_dentry->d_inode; 2231 retval = -ENOENT; 2232 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino) 2233 goto end_rename; 2234 2235 new_inode = new_dentry->d_inode; 2236 new_bh = ext3_find_entry (new_dentry, &new_de); 2237 if (new_bh) { 2238 if (!new_inode) { 2239 brelse (new_bh); 2240 new_bh = NULL; 2241 } 2242 } 2243 if (S_ISDIR(old_inode->i_mode)) { 2244 if (new_inode) { 2245 retval = -ENOTEMPTY; 2246 if (!empty_dir (new_inode)) 2247 goto end_rename; 2248 } 2249 retval = -EIO; 2250 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval); 2251 if (!dir_bh) 2252 goto end_rename; 2253 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino) 2254 goto end_rename; 2255 retval = -EMLINK; 2256 if (!new_inode && new_dir!=old_dir && 2257 new_dir->i_nlink >= EXT3_LINK_MAX) 2258 goto end_rename; 2259 } 2260 if (!new_bh) { 2261 retval = ext3_add_entry (handle, new_dentry, old_inode); 2262 if (retval) 2263 goto end_rename; 2264 } else { 2265 BUFFER_TRACE(new_bh, "get write access"); 2266 ext3_journal_get_write_access(handle, new_bh); 2267 new_de->inode = cpu_to_le32(old_inode->i_ino); 2268 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb, 2269 EXT3_FEATURE_INCOMPAT_FILETYPE)) 2270 new_de->file_type = old_de->file_type; 2271 new_dir->i_version++; 2272 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata"); 2273 ext3_journal_dirty_metadata(handle, new_bh); 2274 brelse(new_bh); 2275 new_bh = NULL; 2276 } 2277 2278 /* 2279 * Like most other Unix systems, set the ctime for inodes on a 2280 * rename. 2281 */ 2282 old_inode->i_ctime = CURRENT_TIME_SEC; 2283 ext3_mark_inode_dirty(handle, old_inode); 2284 2285 /* 2286 * ok, that's it 2287 */ 2288 if (le32_to_cpu(old_de->inode) != old_inode->i_ino || 2289 old_de->name_len != old_dentry->d_name.len || 2290 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) || 2291 (retval = ext3_delete_entry(handle, old_dir, 2292 old_de, old_bh)) == -ENOENT) { 2293 /* old_de could have moved from under us during htree split, so 2294 * make sure that we are deleting the right entry. We might 2295 * also be pointing to a stale entry in the unused part of 2296 * old_bh so just checking inum and the name isn't enough. */ 2297 struct buffer_head *old_bh2; 2298 struct ext3_dir_entry_2 *old_de2; 2299 2300 old_bh2 = ext3_find_entry(old_dentry, &old_de2); 2301 if (old_bh2) { 2302 retval = ext3_delete_entry(handle, old_dir, 2303 old_de2, old_bh2); 2304 brelse(old_bh2); 2305 } 2306 } 2307 if (retval) { 2308 ext3_warning(old_dir->i_sb, "ext3_rename", 2309 "Deleting old file (%lu), %d, error=%d", 2310 old_dir->i_ino, old_dir->i_nlink, retval); 2311 } 2312 2313 if (new_inode) { 2314 drop_nlink(new_inode); 2315 new_inode->i_ctime = CURRENT_TIME_SEC; 2316 } 2317 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC; 2318 ext3_update_dx_flag(old_dir); 2319 if (dir_bh) { 2320 BUFFER_TRACE(dir_bh, "get_write_access"); 2321 ext3_journal_get_write_access(handle, dir_bh); 2322 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino); 2323 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata"); 2324 ext3_journal_dirty_metadata(handle, dir_bh); 2325 drop_nlink(old_dir); 2326 if (new_inode) { 2327 drop_nlink(new_inode); 2328 } else { 2329 inc_nlink(new_dir); 2330 ext3_update_dx_flag(new_dir); 2331 ext3_mark_inode_dirty(handle, new_dir); 2332 } 2333 } 2334 ext3_mark_inode_dirty(handle, old_dir); 2335 if (new_inode) { 2336 ext3_mark_inode_dirty(handle, new_inode); 2337 if (!new_inode->i_nlink) 2338 ext3_orphan_add(handle, new_inode); 2339 } 2340 retval = 0; 2341 2342end_rename: 2343 brelse (dir_bh); 2344 brelse (old_bh); 2345 brelse (new_bh); 2346 ext3_journal_stop(handle); 2347 return retval; 2348} 2349 2350/* 2351 * directories can handle most operations... 2352 */ 2353const struct inode_operations ext3_dir_inode_operations = { 2354 .create = ext3_create, 2355 .lookup = ext3_lookup, 2356 .link = ext3_link, 2357 .unlink = ext3_unlink, 2358 .symlink = ext3_symlink, 2359 .mkdir = ext3_mkdir, 2360 .rmdir = ext3_rmdir, 2361 .mknod = ext3_mknod, 2362 .rename = ext3_rename, 2363 .setattr = ext3_setattr, 2364#ifdef CONFIG_EXT3_FS_XATTR 2365 .setxattr = generic_setxattr, 2366 .getxattr = generic_getxattr, 2367 .listxattr = ext3_listxattr, 2368 .removexattr = generic_removexattr, 2369#endif 2370 .permission = ext3_permission, 2371}; 2372 2373const struct inode_operations ext3_special_inode_operations = { 2374 .setattr = ext3_setattr, 2375#ifdef CONFIG_EXT3_FS_XATTR 2376 .setxattr = generic_setxattr, 2377 .getxattr = generic_getxattr, 2378 .listxattr = ext3_listxattr, 2379 .removexattr = generic_removexattr, 2380#endif 2381 .permission = ext3_permission, 2382}; 2383