namei.c revision ba7fe369875075c3b0def9887afb7a44fa88175e
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/smp_lock.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 ext3fs_dirhash(de->name, de->name_len, hinfo); 555 if ((hinfo->hash < start_hash) || 556 ((hinfo->hash == start_hash) && 557 (hinfo->minor_hash < start_minor_hash))) 558 continue; 559 if (de->inode == 0) 560 continue; 561 if ((err = ext3_htree_store_dirent(dir_file, 562 hinfo->hash, hinfo->minor_hash, de)) != 0) { 563 brelse(bh); 564 return err; 565 } 566 count++; 567 } 568 brelse(bh); 569 return count; 570} 571 572 573/* 574 * This function fills a red-black tree with information from a 575 * directory. We start scanning the directory in hash order, starting 576 * at start_hash and start_minor_hash. 577 * 578 * This function returns the number of entries inserted into the tree, 579 * or a negative error code. 580 */ 581int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash, 582 __u32 start_minor_hash, __u32 *next_hash) 583{ 584 struct dx_hash_info hinfo; 585 struct ext3_dir_entry_2 *de; 586 struct dx_frame frames[2], *frame; 587 struct inode *dir; 588 int block, err; 589 int count = 0; 590 int ret; 591 __u32 hashval; 592 593 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash, 594 start_minor_hash)); 595 dir = dir_file->f_dentry->d_inode; 596 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) { 597 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version; 598 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed; 599 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo, 600 start_hash, start_minor_hash); 601 *next_hash = ~0; 602 return count; 603 } 604 hinfo.hash = start_hash; 605 hinfo.minor_hash = 0; 606 frame = dx_probe(NULL, dir_file->f_dentry->d_inode, &hinfo, frames, &err); 607 if (!frame) 608 return err; 609 610 /* Add '.' and '..' from the htree header */ 611 if (!start_hash && !start_minor_hash) { 612 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data; 613 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0) 614 goto errout; 615 count++; 616 } 617 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) { 618 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data; 619 de = ext3_next_entry(de); 620 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0) 621 goto errout; 622 count++; 623 } 624 625 while (1) { 626 block = dx_get_block(frame->at); 627 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo, 628 start_hash, start_minor_hash); 629 if (ret < 0) { 630 err = ret; 631 goto errout; 632 } 633 count += ret; 634 hashval = ~0; 635 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS, 636 frame, frames, &hashval); 637 *next_hash = hashval; 638 if (ret < 0) { 639 err = ret; 640 goto errout; 641 } 642 /* 643 * Stop if: (a) there are no more entries, or 644 * (b) we have inserted at least one entry and the 645 * next hash value is not a continuation 646 */ 647 if ((ret == 0) || 648 (count && ((hashval & 1) == 0))) 649 break; 650 } 651 dx_release(frames); 652 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n", 653 count, *next_hash)); 654 return count; 655errout: 656 dx_release(frames); 657 return (err); 658} 659 660 661/* 662 * Directory block splitting, compacting 663 */ 664 665static int dx_make_map (struct ext3_dir_entry_2 *de, int size, 666 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail) 667{ 668 int count = 0; 669 char *base = (char *) de; 670 struct dx_hash_info h = *hinfo; 671 672 while ((char *) de < base + size) 673 { 674 if (de->name_len && de->inode) { 675 ext3fs_dirhash(de->name, de->name_len, &h); 676 map_tail--; 677 map_tail->hash = h.hash; 678 map_tail->offs = (u32) ((char *) de - base); 679 count++; 680 cond_resched(); 681 } 682 /* XXX: do we need to check rec_len == 0 case? -Chris */ 683 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len)); 684 } 685 return count; 686} 687 688static void dx_sort_map (struct dx_map_entry *map, unsigned count) 689{ 690 struct dx_map_entry *p, *q, *top = map + count - 1; 691 int more; 692 /* Combsort until bubble sort doesn't suck */ 693 while (count > 2) 694 { 695 count = count*10/13; 696 if (count - 9 < 2) /* 9, 10 -> 11 */ 697 count = 11; 698 for (p = top, q = p - count; q >= map; p--, q--) 699 if (p->hash < q->hash) 700 swap(*p, *q); 701 } 702 /* Garden variety bubble sort */ 703 do { 704 more = 0; 705 q = top; 706 while (q-- > map) 707 { 708 if (q[1].hash >= q[0].hash) 709 continue; 710 swap(*(q+1), *q); 711 more = 1; 712 } 713 } while(more); 714} 715 716static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block) 717{ 718 struct dx_entry *entries = frame->entries; 719 struct dx_entry *old = frame->at, *new = old + 1; 720 int count = dx_get_count(entries); 721 722 assert(count < dx_get_limit(entries)); 723 assert(old < entries + count); 724 memmove(new + 1, new, (char *)(entries + count) - (char *)(new)); 725 dx_set_hash(new, hash); 726 dx_set_block(new, block); 727 dx_set_count(entries, count + 1); 728} 729#endif 730 731 732static void ext3_update_dx_flag(struct inode *inode) 733{ 734 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb, 735 EXT3_FEATURE_COMPAT_DIR_INDEX)) 736 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL; 737} 738 739/* 740 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure. 741 * 742 * `len <= EXT3_NAME_LEN' is guaranteed by caller. 743 * `de != NULL' is guaranteed by caller. 744 */ 745static inline int ext3_match (int len, const char * const name, 746 struct ext3_dir_entry_2 * de) 747{ 748 if (len != de->name_len) 749 return 0; 750 if (!de->inode) 751 return 0; 752 return !memcmp(name, de->name, len); 753} 754 755/* 756 * Returns 0 if not found, -1 on failure, and 1 on success 757 */ 758static inline int search_dirblock(struct buffer_head * bh, 759 struct inode *dir, 760 struct dentry *dentry, 761 unsigned long offset, 762 struct ext3_dir_entry_2 ** res_dir) 763{ 764 struct ext3_dir_entry_2 * de; 765 char * dlimit; 766 int de_len; 767 const char *name = dentry->d_name.name; 768 int namelen = dentry->d_name.len; 769 770 de = (struct ext3_dir_entry_2 *) bh->b_data; 771 dlimit = bh->b_data + dir->i_sb->s_blocksize; 772 while ((char *) de < dlimit) { 773 /* this code is executed quadratically often */ 774 /* do minimal checking `by hand' */ 775 776 if ((char *) de + namelen <= dlimit && 777 ext3_match (namelen, name, de)) { 778 /* found a match - just to be sure, do a full check */ 779 if (!ext3_check_dir_entry("ext3_find_entry", 780 dir, de, bh, offset)) 781 return -1; 782 *res_dir = de; 783 return 1; 784 } 785 /* prevent looping on a bad block */ 786 de_len = le16_to_cpu(de->rec_len); 787 if (de_len <= 0) 788 return -1; 789 offset += de_len; 790 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len); 791 } 792 return 0; 793} 794 795 796/* 797 * ext3_find_entry() 798 * 799 * finds an entry in the specified directory with the wanted name. It 800 * returns the cache buffer in which the entry was found, and the entry 801 * itself (as a parameter - res_dir). It does NOT read the inode of the 802 * entry - you'll have to do that yourself if you want to. 803 * 804 * The returned buffer_head has ->b_count elevated. The caller is expected 805 * to brelse() it when appropriate. 806 */ 807static struct buffer_head * ext3_find_entry (struct dentry *dentry, 808 struct ext3_dir_entry_2 ** res_dir) 809{ 810 struct super_block * sb; 811 struct buffer_head * bh_use[NAMEI_RA_SIZE]; 812 struct buffer_head * bh, *ret = NULL; 813 unsigned long start, block, b; 814 int ra_max = 0; /* Number of bh's in the readahead 815 buffer, bh_use[] */ 816 int ra_ptr = 0; /* Current index into readahead 817 buffer */ 818 int num = 0; 819 int nblocks, i, err; 820 struct inode *dir = dentry->d_parent->d_inode; 821 int namelen; 822 const u8 *name; 823 unsigned blocksize; 824 825 *res_dir = NULL; 826 sb = dir->i_sb; 827 blocksize = sb->s_blocksize; 828 namelen = dentry->d_name.len; 829 name = dentry->d_name.name; 830 if (namelen > EXT3_NAME_LEN) 831 return NULL; 832#ifdef CONFIG_EXT3_INDEX 833 if (is_dx(dir)) { 834 bh = ext3_dx_find_entry(dentry, res_dir, &err); 835 /* 836 * On success, or if the error was file not found, 837 * return. Otherwise, fall back to doing a search the 838 * old fashioned way. 839 */ 840 if (bh || (err != ERR_BAD_DX_DIR)) 841 return bh; 842 dxtrace(printk("ext3_find_entry: dx failed, falling back\n")); 843 } 844#endif 845 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb); 846 start = EXT3_I(dir)->i_dir_start_lookup; 847 if (start >= nblocks) 848 start = 0; 849 block = start; 850restart: 851 do { 852 /* 853 * We deal with the read-ahead logic here. 854 */ 855 if (ra_ptr >= ra_max) { 856 /* Refill the readahead buffer */ 857 ra_ptr = 0; 858 b = block; 859 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) { 860 /* 861 * Terminate if we reach the end of the 862 * directory and must wrap, or if our 863 * search has finished at this block. 864 */ 865 if (b >= nblocks || (num && block == start)) { 866 bh_use[ra_max] = NULL; 867 break; 868 } 869 num++; 870 bh = ext3_getblk(NULL, dir, b++, 0, &err); 871 bh_use[ra_max] = bh; 872 if (bh) 873 ll_rw_block(READ, 1, &bh); 874 } 875 } 876 if ((bh = bh_use[ra_ptr++]) == NULL) 877 goto next; 878 wait_on_buffer(bh); 879 if (!buffer_uptodate(bh)) { 880 /* read error, skip block & hope for the best */ 881 ext3_error(sb, __FUNCTION__, "reading directory #%lu " 882 "offset %lu", dir->i_ino, block); 883 brelse(bh); 884 goto next; 885 } 886 i = search_dirblock(bh, dir, dentry, 887 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir); 888 if (i == 1) { 889 EXT3_I(dir)->i_dir_start_lookup = block; 890 ret = bh; 891 goto cleanup_and_exit; 892 } else { 893 brelse(bh); 894 if (i < 0) 895 goto cleanup_and_exit; 896 } 897 next: 898 if (++block >= nblocks) 899 block = 0; 900 } while (block != start); 901 902 /* 903 * If the directory has grown while we were searching, then 904 * search the last part of the directory before giving up. 905 */ 906 block = nblocks; 907 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb); 908 if (block < nblocks) { 909 start = 0; 910 goto restart; 911 } 912 913cleanup_and_exit: 914 /* Clean up the read-ahead blocks */ 915 for (; ra_ptr < ra_max; ra_ptr++) 916 brelse (bh_use[ra_ptr]); 917 return ret; 918} 919 920#ifdef CONFIG_EXT3_INDEX 921static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry, 922 struct ext3_dir_entry_2 **res_dir, int *err) 923{ 924 struct super_block * sb; 925 struct dx_hash_info hinfo; 926 u32 hash; 927 struct dx_frame frames[2], *frame; 928 struct ext3_dir_entry_2 *de, *top; 929 struct buffer_head *bh; 930 unsigned long block; 931 int retval; 932 int namelen = dentry->d_name.len; 933 const u8 *name = dentry->d_name.name; 934 struct inode *dir = dentry->d_parent->d_inode; 935 936 sb = dir->i_sb; 937 /* NFS may look up ".." - look at dx_root directory block */ 938 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){ 939 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err))) 940 return NULL; 941 } else { 942 frame = frames; 943 frame->bh = NULL; /* for dx_release() */ 944 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/ 945 dx_set_block(frame->at, 0); /* dx_root block is 0 */ 946 } 947 hash = hinfo.hash; 948 do { 949 block = dx_get_block(frame->at); 950 if (!(bh = ext3_bread (NULL,dir, block, 0, err))) 951 goto errout; 952 de = (struct ext3_dir_entry_2 *) bh->b_data; 953 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize - 954 EXT3_DIR_REC_LEN(0)); 955 for (; de < top; de = ext3_next_entry(de)) 956 if (ext3_match (namelen, name, de)) { 957 if (!ext3_check_dir_entry("ext3_find_entry", 958 dir, de, bh, 959 (block<<EXT3_BLOCK_SIZE_BITS(sb)) 960 +((char *)de - bh->b_data))) { 961 brelse (bh); 962 goto errout; 963 } 964 *res_dir = de; 965 dx_release (frames); 966 return bh; 967 } 968 brelse (bh); 969 /* Check to see if we should continue to search */ 970 retval = ext3_htree_next_block(dir, hash, frame, 971 frames, NULL); 972 if (retval < 0) { 973 ext3_warning(sb, __FUNCTION__, 974 "error reading index page in directory #%lu", 975 dir->i_ino); 976 *err = retval; 977 goto errout; 978 } 979 } while (retval == 1); 980 981 *err = -ENOENT; 982errout: 983 dxtrace(printk("%s not found\n", name)); 984 dx_release (frames); 985 return NULL; 986} 987#endif 988 989static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd) 990{ 991 struct inode * inode; 992 struct ext3_dir_entry_2 * de; 993 struct buffer_head * bh; 994 995 if (dentry->d_name.len > EXT3_NAME_LEN) 996 return ERR_PTR(-ENAMETOOLONG); 997 998 bh = ext3_find_entry(dentry, &de); 999 inode = NULL; 1000 if (bh) { 1001 unsigned long ino = le32_to_cpu(de->inode); 1002 brelse (bh); 1003 inode = iget(dir->i_sb, ino); 1004 1005 if (!inode) 1006 return ERR_PTR(-EACCES); 1007 } 1008 return d_splice_alias(inode, dentry); 1009} 1010 1011 1012struct dentry *ext3_get_parent(struct dentry *child) 1013{ 1014 unsigned long ino; 1015 struct dentry *parent; 1016 struct inode *inode; 1017 struct dentry dotdot; 1018 struct ext3_dir_entry_2 * de; 1019 struct buffer_head *bh; 1020 1021 dotdot.d_name.name = ".."; 1022 dotdot.d_name.len = 2; 1023 dotdot.d_parent = child; /* confusing, isn't it! */ 1024 1025 bh = ext3_find_entry(&dotdot, &de); 1026 inode = NULL; 1027 if (!bh) 1028 return ERR_PTR(-ENOENT); 1029 ino = le32_to_cpu(de->inode); 1030 brelse(bh); 1031 inode = iget(child->d_inode->i_sb, ino); 1032 1033 if (!inode) 1034 return ERR_PTR(-EACCES); 1035 1036 parent = d_alloc_anon(inode); 1037 if (!parent) { 1038 iput(inode); 1039 parent = ERR_PTR(-ENOMEM); 1040 } 1041 return parent; 1042} 1043 1044#define S_SHIFT 12 1045static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = { 1046 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE, 1047 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR, 1048 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV, 1049 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV, 1050 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO, 1051 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK, 1052 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK, 1053}; 1054 1055static inline void ext3_set_de_type(struct super_block *sb, 1056 struct ext3_dir_entry_2 *de, 1057 umode_t mode) { 1058 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE)) 1059 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT]; 1060} 1061 1062#ifdef CONFIG_EXT3_INDEX 1063static struct ext3_dir_entry_2 * 1064dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count) 1065{ 1066 unsigned rec_len = 0; 1067 1068 while (count--) { 1069 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs); 1070 rec_len = EXT3_DIR_REC_LEN(de->name_len); 1071 memcpy (to, de, rec_len); 1072 ((struct ext3_dir_entry_2 *) to)->rec_len = 1073 cpu_to_le16(rec_len); 1074 de->inode = 0; 1075 map++; 1076 to += rec_len; 1077 } 1078 return (struct ext3_dir_entry_2 *) (to - rec_len); 1079} 1080 1081static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size) 1082{ 1083 struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base; 1084 unsigned rec_len = 0; 1085 1086 prev = to = de; 1087 while ((char*)de < base + size) { 1088 next = (struct ext3_dir_entry_2 *) ((char *) de + 1089 le16_to_cpu(de->rec_len)); 1090 if (de->inode && de->name_len) { 1091 rec_len = EXT3_DIR_REC_LEN(de->name_len); 1092 if (de > to) 1093 memmove(to, de, rec_len); 1094 to->rec_len = cpu_to_le16(rec_len); 1095 prev = to; 1096 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len); 1097 } 1098 de = next; 1099 } 1100 return prev; 1101} 1102 1103static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, 1104 struct buffer_head **bh,struct dx_frame *frame, 1105 struct dx_hash_info *hinfo, int *error) 1106{ 1107 unsigned blocksize = dir->i_sb->s_blocksize; 1108 unsigned count, continued; 1109 struct buffer_head *bh2; 1110 u32 newblock; 1111 u32 hash2; 1112 struct dx_map_entry *map; 1113 char *data1 = (*bh)->b_data, *data2; 1114 unsigned split; 1115 struct ext3_dir_entry_2 *de = NULL, *de2; 1116 int err; 1117 1118 bh2 = ext3_append (handle, dir, &newblock, error); 1119 if (!(bh2)) { 1120 brelse(*bh); 1121 *bh = NULL; 1122 goto errout; 1123 } 1124 1125 BUFFER_TRACE(*bh, "get_write_access"); 1126 err = ext3_journal_get_write_access(handle, *bh); 1127 if (err) { 1128 journal_error: 1129 brelse(*bh); 1130 brelse(bh2); 1131 *bh = NULL; 1132 ext3_std_error(dir->i_sb, err); 1133 goto errout; 1134 } 1135 BUFFER_TRACE(frame->bh, "get_write_access"); 1136 err = ext3_journal_get_write_access(handle, frame->bh); 1137 if (err) 1138 goto journal_error; 1139 1140 data2 = bh2->b_data; 1141 1142 /* create map in the end of data2 block */ 1143 map = (struct dx_map_entry *) (data2 + blocksize); 1144 count = dx_make_map ((struct ext3_dir_entry_2 *) data1, 1145 blocksize, hinfo, map); 1146 map -= count; 1147 split = count/2; // need to adjust to actual middle 1148 dx_sort_map (map, count); 1149 hash2 = map[split].hash; 1150 continued = hash2 == map[split - 1].hash; 1151 dxtrace(printk("Split block %i at %x, %i/%i\n", 1152 dx_get_block(frame->at), hash2, split, count-split)); 1153 1154 /* Fancy dance to stay within two buffers */ 1155 de2 = dx_move_dirents(data1, data2, map + split, count - split); 1156 de = dx_pack_dirents(data1,blocksize); 1157 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de); 1158 de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2); 1159 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1)); 1160 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1)); 1161 1162 /* Which block gets the new entry? */ 1163 if (hinfo->hash >= hash2) 1164 { 1165 swap(*bh, bh2); 1166 de = de2; 1167 } 1168 dx_insert_block (frame, hash2 + continued, newblock); 1169 err = ext3_journal_dirty_metadata (handle, bh2); 1170 if (err) 1171 goto journal_error; 1172 err = ext3_journal_dirty_metadata (handle, frame->bh); 1173 if (err) 1174 goto journal_error; 1175 brelse (bh2); 1176 dxtrace(dx_show_index ("frame", frame->entries)); 1177errout: 1178 return de; 1179} 1180#endif 1181 1182 1183/* 1184 * Add a new entry into a directory (leaf) block. If de is non-NULL, 1185 * it points to a directory entry which is guaranteed to be large 1186 * enough for new directory entry. If de is NULL, then 1187 * add_dirent_to_buf will attempt search the directory block for 1188 * space. It will return -ENOSPC if no space is available, and -EIO 1189 * and -EEXIST if directory entry already exists. 1190 * 1191 * NOTE! bh is NOT released in the case where ENOSPC is returned. In 1192 * all other cases bh is released. 1193 */ 1194static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry, 1195 struct inode *inode, struct ext3_dir_entry_2 *de, 1196 struct buffer_head * bh) 1197{ 1198 struct inode *dir = dentry->d_parent->d_inode; 1199 const char *name = dentry->d_name.name; 1200 int namelen = dentry->d_name.len; 1201 unsigned long offset = 0; 1202 unsigned short reclen; 1203 int nlen, rlen, err; 1204 char *top; 1205 1206 reclen = EXT3_DIR_REC_LEN(namelen); 1207 if (!de) { 1208 de = (struct ext3_dir_entry_2 *)bh->b_data; 1209 top = bh->b_data + dir->i_sb->s_blocksize - reclen; 1210 while ((char *) de <= top) { 1211 if (!ext3_check_dir_entry("ext3_add_entry", dir, de, 1212 bh, offset)) { 1213 brelse (bh); 1214 return -EIO; 1215 } 1216 if (ext3_match (namelen, name, de)) { 1217 brelse (bh); 1218 return -EEXIST; 1219 } 1220 nlen = EXT3_DIR_REC_LEN(de->name_len); 1221 rlen = le16_to_cpu(de->rec_len); 1222 if ((de->inode? rlen - nlen: rlen) >= reclen) 1223 break; 1224 de = (struct ext3_dir_entry_2 *)((char *)de + rlen); 1225 offset += rlen; 1226 } 1227 if ((char *) de > top) 1228 return -ENOSPC; 1229 } 1230 BUFFER_TRACE(bh, "get_write_access"); 1231 err = ext3_journal_get_write_access(handle, bh); 1232 if (err) { 1233 ext3_std_error(dir->i_sb, err); 1234 brelse(bh); 1235 return err; 1236 } 1237 1238 /* By now the buffer is marked for journaling */ 1239 nlen = EXT3_DIR_REC_LEN(de->name_len); 1240 rlen = le16_to_cpu(de->rec_len); 1241 if (de->inode) { 1242 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen); 1243 de1->rec_len = cpu_to_le16(rlen - nlen); 1244 de->rec_len = cpu_to_le16(nlen); 1245 de = de1; 1246 } 1247 de->file_type = EXT3_FT_UNKNOWN; 1248 if (inode) { 1249 de->inode = cpu_to_le32(inode->i_ino); 1250 ext3_set_de_type(dir->i_sb, de, inode->i_mode); 1251 } else 1252 de->inode = 0; 1253 de->name_len = namelen; 1254 memcpy (de->name, name, namelen); 1255 /* 1256 * XXX shouldn't update any times until successful 1257 * completion of syscall, but too many callers depend 1258 * on this. 1259 * 1260 * XXX similarly, too many callers depend on 1261 * ext3_new_inode() setting the times, but error 1262 * recovery deletes the inode, so the worst that can 1263 * happen is that the times are slightly out of date 1264 * and/or different from the directory change time. 1265 */ 1266 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC; 1267 ext3_update_dx_flag(dir); 1268 dir->i_version++; 1269 ext3_mark_inode_dirty(handle, dir); 1270 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); 1271 err = ext3_journal_dirty_metadata(handle, bh); 1272 if (err) 1273 ext3_std_error(dir->i_sb, err); 1274 brelse(bh); 1275 return 0; 1276} 1277 1278#ifdef CONFIG_EXT3_INDEX 1279/* 1280 * This converts a one block unindexed directory to a 3 block indexed 1281 * directory, and adds the dentry to the indexed directory. 1282 */ 1283static int make_indexed_dir(handle_t *handle, struct dentry *dentry, 1284 struct inode *inode, struct buffer_head *bh) 1285{ 1286 struct inode *dir = dentry->d_parent->d_inode; 1287 const char *name = dentry->d_name.name; 1288 int namelen = dentry->d_name.len; 1289 struct buffer_head *bh2; 1290 struct dx_root *root; 1291 struct dx_frame frames[2], *frame; 1292 struct dx_entry *entries; 1293 struct ext3_dir_entry_2 *de, *de2; 1294 char *data1, *top; 1295 unsigned len; 1296 int retval; 1297 unsigned blocksize; 1298 struct dx_hash_info hinfo; 1299 u32 block; 1300 struct fake_dirent *fde; 1301 1302 blocksize = dir->i_sb->s_blocksize; 1303 dxtrace(printk("Creating index\n")); 1304 retval = ext3_journal_get_write_access(handle, bh); 1305 if (retval) { 1306 ext3_std_error(dir->i_sb, retval); 1307 brelse(bh); 1308 return retval; 1309 } 1310 root = (struct dx_root *) bh->b_data; 1311 1312 bh2 = ext3_append (handle, dir, &block, &retval); 1313 if (!(bh2)) { 1314 brelse(bh); 1315 return retval; 1316 } 1317 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL; 1318 data1 = bh2->b_data; 1319 1320 /* The 0th block becomes the root, move the dirents out */ 1321 fde = &root->dotdot; 1322 de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len)); 1323 len = ((char *) root) + blocksize - (char *) de; 1324 memcpy (data1, de, len); 1325 de = (struct ext3_dir_entry_2 *) data1; 1326 top = data1 + len; 1327 while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top) 1328 de = de2; 1329 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de); 1330 /* Initialize the root; the dot dirents already exist */ 1331 de = (struct ext3_dir_entry_2 *) (&root->dotdot); 1332 de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2)); 1333 memset (&root->info, 0, sizeof(root->info)); 1334 root->info.info_length = sizeof(root->info); 1335 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version; 1336 entries = root->entries; 1337 dx_set_block (entries, 1); 1338 dx_set_count (entries, 1); 1339 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info))); 1340 1341 /* Initialize as for dx_probe */ 1342 hinfo.hash_version = root->info.hash_version; 1343 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed; 1344 ext3fs_dirhash(name, namelen, &hinfo); 1345 frame = frames; 1346 frame->entries = entries; 1347 frame->at = entries; 1348 frame->bh = bh; 1349 bh = bh2; 1350 de = do_split(handle,dir, &bh, frame, &hinfo, &retval); 1351 dx_release (frames); 1352 if (!(de)) 1353 return retval; 1354 1355 return add_dirent_to_buf(handle, dentry, inode, de, bh); 1356} 1357#endif 1358 1359/* 1360 * ext3_add_entry() 1361 * 1362 * adds a file entry to the specified directory, using the same 1363 * semantics as ext3_find_entry(). It returns NULL if it failed. 1364 * 1365 * NOTE!! The inode part of 'de' is left at 0 - which means you 1366 * may not sleep between calling this and putting something into 1367 * the entry, as someone else might have used it while you slept. 1368 */ 1369static int ext3_add_entry (handle_t *handle, struct dentry *dentry, 1370 struct inode *inode) 1371{ 1372 struct inode *dir = dentry->d_parent->d_inode; 1373 unsigned long offset; 1374 struct buffer_head * bh; 1375 struct ext3_dir_entry_2 *de; 1376 struct super_block * sb; 1377 int retval; 1378#ifdef CONFIG_EXT3_INDEX 1379 int dx_fallback=0; 1380#endif 1381 unsigned blocksize; 1382 unsigned nlen, rlen; 1383 u32 block, blocks; 1384 1385 sb = dir->i_sb; 1386 blocksize = sb->s_blocksize; 1387 if (!dentry->d_name.len) 1388 return -EINVAL; 1389#ifdef CONFIG_EXT3_INDEX 1390 if (is_dx(dir)) { 1391 retval = ext3_dx_add_entry(handle, dentry, inode); 1392 if (!retval || (retval != ERR_BAD_DX_DIR)) 1393 return retval; 1394 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL; 1395 dx_fallback++; 1396 ext3_mark_inode_dirty(handle, dir); 1397 } 1398#endif 1399 blocks = dir->i_size >> sb->s_blocksize_bits; 1400 for (block = 0, offset = 0; block < blocks; block++) { 1401 bh = ext3_bread(handle, dir, block, 0, &retval); 1402 if(!bh) 1403 return retval; 1404 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh); 1405 if (retval != -ENOSPC) 1406 return retval; 1407 1408#ifdef CONFIG_EXT3_INDEX 1409 if (blocks == 1 && !dx_fallback && 1410 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX)) 1411 return make_indexed_dir(handle, dentry, inode, bh); 1412#endif 1413 brelse(bh); 1414 } 1415 bh = ext3_append(handle, dir, &block, &retval); 1416 if (!bh) 1417 return retval; 1418 de = (struct ext3_dir_entry_2 *) bh->b_data; 1419 de->inode = 0; 1420 de->rec_len = cpu_to_le16(rlen = blocksize); 1421 nlen = 0; 1422 return add_dirent_to_buf(handle, dentry, inode, de, bh); 1423} 1424 1425#ifdef CONFIG_EXT3_INDEX 1426/* 1427 * Returns 0 for success, or a negative error value 1428 */ 1429static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry, 1430 struct inode *inode) 1431{ 1432 struct dx_frame frames[2], *frame; 1433 struct dx_entry *entries, *at; 1434 struct dx_hash_info hinfo; 1435 struct buffer_head * bh; 1436 struct inode *dir = dentry->d_parent->d_inode; 1437 struct super_block * sb = dir->i_sb; 1438 struct ext3_dir_entry_2 *de; 1439 int err; 1440 1441 frame = dx_probe(dentry, NULL, &hinfo, frames, &err); 1442 if (!frame) 1443 return err; 1444 entries = frame->entries; 1445 at = frame->at; 1446 1447 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err))) 1448 goto cleanup; 1449 1450 BUFFER_TRACE(bh, "get_write_access"); 1451 err = ext3_journal_get_write_access(handle, bh); 1452 if (err) 1453 goto journal_error; 1454 1455 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh); 1456 if (err != -ENOSPC) { 1457 bh = NULL; 1458 goto cleanup; 1459 } 1460 1461 /* Block full, should compress but for now just split */ 1462 dxtrace(printk("using %u of %u node entries\n", 1463 dx_get_count(entries), dx_get_limit(entries))); 1464 /* Need to split index? */ 1465 if (dx_get_count(entries) == dx_get_limit(entries)) { 1466 u32 newblock; 1467 unsigned icount = dx_get_count(entries); 1468 int levels = frame - frames; 1469 struct dx_entry *entries2; 1470 struct dx_node *node2; 1471 struct buffer_head *bh2; 1472 1473 if (levels && (dx_get_count(frames->entries) == 1474 dx_get_limit(frames->entries))) { 1475 ext3_warning(sb, __FUNCTION__, 1476 "Directory index full!"); 1477 err = -ENOSPC; 1478 goto cleanup; 1479 } 1480 bh2 = ext3_append (handle, dir, &newblock, &err); 1481 if (!(bh2)) 1482 goto cleanup; 1483 node2 = (struct dx_node *)(bh2->b_data); 1484 entries2 = node2->entries; 1485 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize); 1486 node2->fake.inode = 0; 1487 BUFFER_TRACE(frame->bh, "get_write_access"); 1488 err = ext3_journal_get_write_access(handle, frame->bh); 1489 if (err) 1490 goto journal_error; 1491 if (levels) { 1492 unsigned icount1 = icount/2, icount2 = icount - icount1; 1493 unsigned hash2 = dx_get_hash(entries + icount1); 1494 dxtrace(printk("Split index %i/%i\n", icount1, icount2)); 1495 1496 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */ 1497 err = ext3_journal_get_write_access(handle, 1498 frames[0].bh); 1499 if (err) 1500 goto journal_error; 1501 1502 memcpy ((char *) entries2, (char *) (entries + icount1), 1503 icount2 * sizeof(struct dx_entry)); 1504 dx_set_count (entries, icount1); 1505 dx_set_count (entries2, icount2); 1506 dx_set_limit (entries2, dx_node_limit(dir)); 1507 1508 /* Which index block gets the new entry? */ 1509 if (at - entries >= icount1) { 1510 frame->at = at = at - entries - icount1 + entries2; 1511 frame->entries = entries = entries2; 1512 swap(frame->bh, bh2); 1513 } 1514 dx_insert_block (frames + 0, hash2, newblock); 1515 dxtrace(dx_show_index ("node", frames[1].entries)); 1516 dxtrace(dx_show_index ("node", 1517 ((struct dx_node *) bh2->b_data)->entries)); 1518 err = ext3_journal_dirty_metadata(handle, bh2); 1519 if (err) 1520 goto journal_error; 1521 brelse (bh2); 1522 } else { 1523 dxtrace(printk("Creating second level index...\n")); 1524 memcpy((char *) entries2, (char *) entries, 1525 icount * sizeof(struct dx_entry)); 1526 dx_set_limit(entries2, dx_node_limit(dir)); 1527 1528 /* Set up root */ 1529 dx_set_count(entries, 1); 1530 dx_set_block(entries + 0, newblock); 1531 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1; 1532 1533 /* Add new access path frame */ 1534 frame = frames + 1; 1535 frame->at = at = at - entries + entries2; 1536 frame->entries = entries = entries2; 1537 frame->bh = bh2; 1538 err = ext3_journal_get_write_access(handle, 1539 frame->bh); 1540 if (err) 1541 goto journal_error; 1542 } 1543 ext3_journal_dirty_metadata(handle, frames[0].bh); 1544 } 1545 de = do_split(handle, dir, &bh, frame, &hinfo, &err); 1546 if (!de) 1547 goto cleanup; 1548 err = add_dirent_to_buf(handle, dentry, inode, de, bh); 1549 bh = NULL; 1550 goto cleanup; 1551 1552journal_error: 1553 ext3_std_error(dir->i_sb, err); 1554cleanup: 1555 if (bh) 1556 brelse(bh); 1557 dx_release(frames); 1558 return err; 1559} 1560#endif 1561 1562/* 1563 * ext3_delete_entry deletes a directory entry by merging it with the 1564 * previous entry 1565 */ 1566static int ext3_delete_entry (handle_t *handle, 1567 struct inode * dir, 1568 struct ext3_dir_entry_2 * de_del, 1569 struct buffer_head * bh) 1570{ 1571 struct ext3_dir_entry_2 * de, * pde; 1572 int i; 1573 1574 i = 0; 1575 pde = NULL; 1576 de = (struct ext3_dir_entry_2 *) bh->b_data; 1577 while (i < bh->b_size) { 1578 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i)) 1579 return -EIO; 1580 if (de == de_del) { 1581 BUFFER_TRACE(bh, "get_write_access"); 1582 ext3_journal_get_write_access(handle, bh); 1583 if (pde) 1584 pde->rec_len = 1585 cpu_to_le16(le16_to_cpu(pde->rec_len) + 1586 le16_to_cpu(de->rec_len)); 1587 else 1588 de->inode = 0; 1589 dir->i_version++; 1590 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); 1591 ext3_journal_dirty_metadata(handle, bh); 1592 return 0; 1593 } 1594 i += le16_to_cpu(de->rec_len); 1595 pde = de; 1596 de = (struct ext3_dir_entry_2 *) 1597 ((char *) de + le16_to_cpu(de->rec_len)); 1598 } 1599 return -ENOENT; 1600} 1601 1602/* 1603 * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we 1604 * do not perform it in these functions. We perform it at the call site, 1605 * if it is needed. 1606 */ 1607static inline void ext3_inc_count(handle_t *handle, struct inode *inode) 1608{ 1609 inode->i_nlink++; 1610} 1611 1612static inline void ext3_dec_count(handle_t *handle, struct inode *inode) 1613{ 1614 inode->i_nlink--; 1615} 1616 1617static int ext3_add_nondir(handle_t *handle, 1618 struct dentry *dentry, struct inode *inode) 1619{ 1620 int err = ext3_add_entry(handle, dentry, inode); 1621 if (!err) { 1622 ext3_mark_inode_dirty(handle, inode); 1623 d_instantiate(dentry, inode); 1624 return 0; 1625 } 1626 ext3_dec_count(handle, inode); 1627 iput(inode); 1628 return err; 1629} 1630 1631/* 1632 * By the time this is called, we already have created 1633 * the directory cache entry for the new file, but it 1634 * is so far negative - it has no inode. 1635 * 1636 * If the create succeeds, we fill in the inode information 1637 * with d_instantiate(). 1638 */ 1639static int ext3_create (struct inode * dir, struct dentry * dentry, int mode, 1640 struct nameidata *nd) 1641{ 1642 handle_t *handle; 1643 struct inode * inode; 1644 int err, retries = 0; 1645 1646retry: 1647 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + 1648 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 + 1649 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb)); 1650 if (IS_ERR(handle)) 1651 return PTR_ERR(handle); 1652 1653 if (IS_DIRSYNC(dir)) 1654 handle->h_sync = 1; 1655 1656 inode = ext3_new_inode (handle, dir, mode); 1657 err = PTR_ERR(inode); 1658 if (!IS_ERR(inode)) { 1659 inode->i_op = &ext3_file_inode_operations; 1660 inode->i_fop = &ext3_file_operations; 1661 ext3_set_aops(inode); 1662 err = ext3_add_nondir(handle, dentry, inode); 1663 } 1664 ext3_journal_stop(handle); 1665 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) 1666 goto retry; 1667 return err; 1668} 1669 1670static int ext3_mknod (struct inode * dir, struct dentry *dentry, 1671 int mode, dev_t rdev) 1672{ 1673 handle_t *handle; 1674 struct inode *inode; 1675 int err, retries = 0; 1676 1677 if (!new_valid_dev(rdev)) 1678 return -EINVAL; 1679 1680retry: 1681 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + 1682 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 + 1683 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb)); 1684 if (IS_ERR(handle)) 1685 return PTR_ERR(handle); 1686 1687 if (IS_DIRSYNC(dir)) 1688 handle->h_sync = 1; 1689 1690 inode = ext3_new_inode (handle, dir, mode); 1691 err = PTR_ERR(inode); 1692 if (!IS_ERR(inode)) { 1693 init_special_inode(inode, inode->i_mode, rdev); 1694#ifdef CONFIG_EXT3_FS_XATTR 1695 inode->i_op = &ext3_special_inode_operations; 1696#endif 1697 err = ext3_add_nondir(handle, dentry, inode); 1698 } 1699 ext3_journal_stop(handle); 1700 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) 1701 goto retry; 1702 return err; 1703} 1704 1705static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode) 1706{ 1707 handle_t *handle; 1708 struct inode * inode; 1709 struct buffer_head * dir_block; 1710 struct ext3_dir_entry_2 * de; 1711 int err, retries = 0; 1712 1713 if (dir->i_nlink >= EXT3_LINK_MAX) 1714 return -EMLINK; 1715 1716retry: 1717 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + 1718 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 + 1719 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb)); 1720 if (IS_ERR(handle)) 1721 return PTR_ERR(handle); 1722 1723 if (IS_DIRSYNC(dir)) 1724 handle->h_sync = 1; 1725 1726 inode = ext3_new_inode (handle, dir, S_IFDIR | mode); 1727 err = PTR_ERR(inode); 1728 if (IS_ERR(inode)) 1729 goto out_stop; 1730 1731 inode->i_op = &ext3_dir_inode_operations; 1732 inode->i_fop = &ext3_dir_operations; 1733 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize; 1734 dir_block = ext3_bread (handle, inode, 0, 1, &err); 1735 if (!dir_block) { 1736 inode->i_nlink--; /* is this nlink == 0? */ 1737 ext3_mark_inode_dirty(handle, inode); 1738 iput (inode); 1739 goto out_stop; 1740 } 1741 BUFFER_TRACE(dir_block, "get_write_access"); 1742 ext3_journal_get_write_access(handle, dir_block); 1743 de = (struct ext3_dir_entry_2 *) dir_block->b_data; 1744 de->inode = cpu_to_le32(inode->i_ino); 1745 de->name_len = 1; 1746 de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len)); 1747 strcpy (de->name, "."); 1748 ext3_set_de_type(dir->i_sb, de, S_IFDIR); 1749 de = (struct ext3_dir_entry_2 *) 1750 ((char *) de + le16_to_cpu(de->rec_len)); 1751 de->inode = cpu_to_le32(dir->i_ino); 1752 de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1)); 1753 de->name_len = 2; 1754 strcpy (de->name, ".."); 1755 ext3_set_de_type(dir->i_sb, de, S_IFDIR); 1756 inode->i_nlink = 2; 1757 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata"); 1758 ext3_journal_dirty_metadata(handle, dir_block); 1759 brelse (dir_block); 1760 ext3_mark_inode_dirty(handle, inode); 1761 err = ext3_add_entry (handle, dentry, inode); 1762 if (err) { 1763 inode->i_nlink = 0; 1764 ext3_mark_inode_dirty(handle, inode); 1765 iput (inode); 1766 goto out_stop; 1767 } 1768 dir->i_nlink++; 1769 ext3_update_dx_flag(dir); 1770 ext3_mark_inode_dirty(handle, dir); 1771 d_instantiate(dentry, inode); 1772out_stop: 1773 ext3_journal_stop(handle); 1774 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) 1775 goto retry; 1776 return err; 1777} 1778 1779/* 1780 * routine to check that the specified directory is empty (for rmdir) 1781 */ 1782static int empty_dir (struct inode * inode) 1783{ 1784 unsigned long offset; 1785 struct buffer_head * bh; 1786 struct ext3_dir_entry_2 * de, * de1; 1787 struct super_block * sb; 1788 int err = 0; 1789 1790 sb = inode->i_sb; 1791 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) || 1792 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) { 1793 if (err) 1794 ext3_error(inode->i_sb, __FUNCTION__, 1795 "error %d reading directory #%lu offset 0", 1796 err, inode->i_ino); 1797 else 1798 ext3_warning(inode->i_sb, __FUNCTION__, 1799 "bad directory (dir #%lu) - no data block", 1800 inode->i_ino); 1801 return 1; 1802 } 1803 de = (struct ext3_dir_entry_2 *) bh->b_data; 1804 de1 = (struct ext3_dir_entry_2 *) 1805 ((char *) de + le16_to_cpu(de->rec_len)); 1806 if (le32_to_cpu(de->inode) != inode->i_ino || 1807 !le32_to_cpu(de1->inode) || 1808 strcmp (".", de->name) || 1809 strcmp ("..", de1->name)) { 1810 ext3_warning (inode->i_sb, "empty_dir", 1811 "bad directory (dir #%lu) - no `.' or `..'", 1812 inode->i_ino); 1813 brelse (bh); 1814 return 1; 1815 } 1816 offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len); 1817 de = (struct ext3_dir_entry_2 *) 1818 ((char *) de1 + le16_to_cpu(de1->rec_len)); 1819 while (offset < inode->i_size ) { 1820 if (!bh || 1821 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) { 1822 err = 0; 1823 brelse (bh); 1824 bh = ext3_bread (NULL, inode, 1825 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err); 1826 if (!bh) { 1827 if (err) 1828 ext3_error(sb, __FUNCTION__, 1829 "error %d reading directory" 1830 " #%lu offset %lu", 1831 err, inode->i_ino, offset); 1832 offset += sb->s_blocksize; 1833 continue; 1834 } 1835 de = (struct ext3_dir_entry_2 *) bh->b_data; 1836 } 1837 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) { 1838 de = (struct ext3_dir_entry_2 *)(bh->b_data + 1839 sb->s_blocksize); 1840 offset = (offset | (sb->s_blocksize - 1)) + 1; 1841 continue; 1842 } 1843 if (le32_to_cpu(de->inode)) { 1844 brelse (bh); 1845 return 0; 1846 } 1847 offset += le16_to_cpu(de->rec_len); 1848 de = (struct ext3_dir_entry_2 *) 1849 ((char *) de + le16_to_cpu(de->rec_len)); 1850 } 1851 brelse (bh); 1852 return 1; 1853} 1854 1855/* ext3_orphan_add() links an unlinked or truncated inode into a list of 1856 * such inodes, starting at the superblock, in case we crash before the 1857 * file is closed/deleted, or in case the inode truncate spans multiple 1858 * transactions and the last transaction is not recovered after a crash. 1859 * 1860 * At filesystem recovery time, we walk this list deleting unlinked 1861 * inodes and truncating linked inodes in ext3_orphan_cleanup(). 1862 */ 1863int ext3_orphan_add(handle_t *handle, struct inode *inode) 1864{ 1865 struct super_block *sb = inode->i_sb; 1866 struct ext3_iloc iloc; 1867 int err = 0, rc; 1868 1869 lock_super(sb); 1870 if (!list_empty(&EXT3_I(inode)->i_orphan)) 1871 goto out_unlock; 1872 1873 /* Orphan handling is only valid for files with data blocks 1874 * being truncated, or files being unlinked. */ 1875 1876 /* @@@ FIXME: Observation from aviro: 1877 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block 1878 * here (on lock_super()), so race with ext3_link() which might bump 1879 * ->i_nlink. For, say it, character device. Not a regular file, 1880 * not a directory, not a symlink and ->i_nlink > 0. 1881 */ 1882 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 1883 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0); 1884 1885 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access"); 1886 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh); 1887 if (err) 1888 goto out_unlock; 1889 1890 err = ext3_reserve_inode_write(handle, inode, &iloc); 1891 if (err) 1892 goto out_unlock; 1893 1894 /* Insert this inode at the head of the on-disk orphan list... */ 1895 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan); 1896 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino); 1897 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh); 1898 rc = ext3_mark_iloc_dirty(handle, inode, &iloc); 1899 if (!err) 1900 err = rc; 1901 1902 /* Only add to the head of the in-memory list if all the 1903 * previous operations succeeded. If the orphan_add is going to 1904 * fail (possibly taking the journal offline), we can't risk 1905 * leaving the inode on the orphan list: stray orphan-list 1906 * entries can cause panics at unmount time. 1907 * 1908 * This is safe: on error we're going to ignore the orphan list 1909 * anyway on the next recovery. */ 1910 if (!err) 1911 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan); 1912 1913 jbd_debug(4, "superblock will point to %ld\n", inode->i_ino); 1914 jbd_debug(4, "orphan inode %ld will point to %d\n", 1915 inode->i_ino, NEXT_ORPHAN(inode)); 1916out_unlock: 1917 unlock_super(sb); 1918 ext3_std_error(inode->i_sb, err); 1919 return err; 1920} 1921 1922/* 1923 * ext3_orphan_del() removes an unlinked or truncated inode from the list 1924 * of such inodes stored on disk, because it is finally being cleaned up. 1925 */ 1926int ext3_orphan_del(handle_t *handle, struct inode *inode) 1927{ 1928 struct list_head *prev; 1929 struct ext3_inode_info *ei = EXT3_I(inode); 1930 struct ext3_sb_info *sbi; 1931 unsigned long ino_next; 1932 struct ext3_iloc iloc; 1933 int err = 0; 1934 1935 lock_super(inode->i_sb); 1936 if (list_empty(&ei->i_orphan)) { 1937 unlock_super(inode->i_sb); 1938 return 0; 1939 } 1940 1941 ino_next = NEXT_ORPHAN(inode); 1942 prev = ei->i_orphan.prev; 1943 sbi = EXT3_SB(inode->i_sb); 1944 1945 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino); 1946 1947 list_del_init(&ei->i_orphan); 1948 1949 /* If we're on an error path, we may not have a valid 1950 * transaction handle with which to update the orphan list on 1951 * disk, but we still need to remove the inode from the linked 1952 * list in memory. */ 1953 if (!handle) 1954 goto out; 1955 1956 err = ext3_reserve_inode_write(handle, inode, &iloc); 1957 if (err) 1958 goto out_err; 1959 1960 if (prev == &sbi->s_orphan) { 1961 jbd_debug(4, "superblock will point to %lu\n", ino_next); 1962 BUFFER_TRACE(sbi->s_sbh, "get_write_access"); 1963 err = ext3_journal_get_write_access(handle, sbi->s_sbh); 1964 if (err) 1965 goto out_brelse; 1966 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next); 1967 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh); 1968 } else { 1969 struct ext3_iloc iloc2; 1970 struct inode *i_prev = 1971 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode; 1972 1973 jbd_debug(4, "orphan inode %lu will point to %lu\n", 1974 i_prev->i_ino, ino_next); 1975 err = ext3_reserve_inode_write(handle, i_prev, &iloc2); 1976 if (err) 1977 goto out_brelse; 1978 NEXT_ORPHAN(i_prev) = ino_next; 1979 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2); 1980 } 1981 if (err) 1982 goto out_brelse; 1983 NEXT_ORPHAN(inode) = 0; 1984 err = ext3_mark_iloc_dirty(handle, inode, &iloc); 1985 1986out_err: 1987 ext3_std_error(inode->i_sb, err); 1988out: 1989 unlock_super(inode->i_sb); 1990 return err; 1991 1992out_brelse: 1993 brelse(iloc.bh); 1994 goto out_err; 1995} 1996 1997static int ext3_rmdir (struct inode * dir, struct dentry *dentry) 1998{ 1999 int retval; 2000 struct inode * inode; 2001 struct buffer_head * bh; 2002 struct ext3_dir_entry_2 * de; 2003 handle_t *handle; 2004 2005 /* Initialize quotas before so that eventual writes go in 2006 * separate transaction */ 2007 DQUOT_INIT(dentry->d_inode); 2008 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb)); 2009 if (IS_ERR(handle)) 2010 return PTR_ERR(handle); 2011 2012 retval = -ENOENT; 2013 bh = ext3_find_entry (dentry, &de); 2014 if (!bh) 2015 goto end_rmdir; 2016 2017 if (IS_DIRSYNC(dir)) 2018 handle->h_sync = 1; 2019 2020 inode = dentry->d_inode; 2021 2022 retval = -EIO; 2023 if (le32_to_cpu(de->inode) != inode->i_ino) 2024 goto end_rmdir; 2025 2026 retval = -ENOTEMPTY; 2027 if (!empty_dir (inode)) 2028 goto end_rmdir; 2029 2030 retval = ext3_delete_entry(handle, dir, de, bh); 2031 if (retval) 2032 goto end_rmdir; 2033 if (inode->i_nlink != 2) 2034 ext3_warning (inode->i_sb, "ext3_rmdir", 2035 "empty directory has nlink!=2 (%d)", 2036 inode->i_nlink); 2037 inode->i_version++; 2038 inode->i_nlink = 0; 2039 /* There's no need to set i_disksize: the fact that i_nlink is 2040 * zero will ensure that the right thing happens during any 2041 * recovery. */ 2042 inode->i_size = 0; 2043 ext3_orphan_add(handle, inode); 2044 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC; 2045 ext3_mark_inode_dirty(handle, inode); 2046 dir->i_nlink--; 2047 ext3_update_dx_flag(dir); 2048 ext3_mark_inode_dirty(handle, dir); 2049 2050end_rmdir: 2051 ext3_journal_stop(handle); 2052 brelse (bh); 2053 return retval; 2054} 2055 2056static int ext3_unlink(struct inode * dir, struct dentry *dentry) 2057{ 2058 int retval; 2059 struct inode * inode; 2060 struct buffer_head * bh; 2061 struct ext3_dir_entry_2 * de; 2062 handle_t *handle; 2063 2064 /* Initialize quotas before so that eventual writes go 2065 * in separate transaction */ 2066 DQUOT_INIT(dentry->d_inode); 2067 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb)); 2068 if (IS_ERR(handle)) 2069 return PTR_ERR(handle); 2070 2071 if (IS_DIRSYNC(dir)) 2072 handle->h_sync = 1; 2073 2074 retval = -ENOENT; 2075 bh = ext3_find_entry (dentry, &de); 2076 if (!bh) 2077 goto end_unlink; 2078 2079 inode = dentry->d_inode; 2080 2081 retval = -EIO; 2082 if (le32_to_cpu(de->inode) != inode->i_ino) 2083 goto end_unlink; 2084 2085 if (!inode->i_nlink) { 2086 ext3_warning (inode->i_sb, "ext3_unlink", 2087 "Deleting nonexistent file (%lu), %d", 2088 inode->i_ino, inode->i_nlink); 2089 inode->i_nlink = 1; 2090 } 2091 retval = ext3_delete_entry(handle, dir, de, bh); 2092 if (retval) 2093 goto end_unlink; 2094 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC; 2095 ext3_update_dx_flag(dir); 2096 ext3_mark_inode_dirty(handle, dir); 2097 inode->i_nlink--; 2098 if (!inode->i_nlink) 2099 ext3_orphan_add(handle, inode); 2100 inode->i_ctime = dir->i_ctime; 2101 ext3_mark_inode_dirty(handle, inode); 2102 retval = 0; 2103 2104end_unlink: 2105 ext3_journal_stop(handle); 2106 brelse (bh); 2107 return retval; 2108} 2109 2110static int ext3_symlink (struct inode * dir, 2111 struct dentry *dentry, const char * symname) 2112{ 2113 handle_t *handle; 2114 struct inode * inode; 2115 int l, err, retries = 0; 2116 2117 l = strlen(symname)+1; 2118 if (l > dir->i_sb->s_blocksize) 2119 return -ENAMETOOLONG; 2120 2121retry: 2122 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + 2123 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 + 2124 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb)); 2125 if (IS_ERR(handle)) 2126 return PTR_ERR(handle); 2127 2128 if (IS_DIRSYNC(dir)) 2129 handle->h_sync = 1; 2130 2131 inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO); 2132 err = PTR_ERR(inode); 2133 if (IS_ERR(inode)) 2134 goto out_stop; 2135 2136 if (l > sizeof (EXT3_I(inode)->i_data)) { 2137 inode->i_op = &ext3_symlink_inode_operations; 2138 ext3_set_aops(inode); 2139 /* 2140 * page_symlink() calls into ext3_prepare/commit_write. 2141 * We have a transaction open. All is sweetness. It also sets 2142 * i_size in generic_commit_write(). 2143 */ 2144 err = page_symlink(inode, symname, l); 2145 if (err) { 2146 ext3_dec_count(handle, inode); 2147 ext3_mark_inode_dirty(handle, inode); 2148 iput (inode); 2149 goto out_stop; 2150 } 2151 } else { 2152 inode->i_op = &ext3_fast_symlink_inode_operations; 2153 memcpy((char*)&EXT3_I(inode)->i_data,symname,l); 2154 inode->i_size = l-1; 2155 } 2156 EXT3_I(inode)->i_disksize = inode->i_size; 2157 err = ext3_add_nondir(handle, dentry, inode); 2158out_stop: 2159 ext3_journal_stop(handle); 2160 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) 2161 goto retry; 2162 return err; 2163} 2164 2165static int ext3_link (struct dentry * old_dentry, 2166 struct inode * dir, struct dentry *dentry) 2167{ 2168 handle_t *handle; 2169 struct inode *inode = old_dentry->d_inode; 2170 int err, retries = 0; 2171 2172 if (inode->i_nlink >= EXT3_LINK_MAX) 2173 return -EMLINK; 2174 2175retry: 2176 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + 2177 EXT3_INDEX_EXTRA_TRANS_BLOCKS); 2178 if (IS_ERR(handle)) 2179 return PTR_ERR(handle); 2180 2181 if (IS_DIRSYNC(dir)) 2182 handle->h_sync = 1; 2183 2184 inode->i_ctime = CURRENT_TIME_SEC; 2185 ext3_inc_count(handle, inode); 2186 atomic_inc(&inode->i_count); 2187 2188 err = ext3_add_nondir(handle, dentry, inode); 2189 ext3_journal_stop(handle); 2190 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) 2191 goto retry; 2192 return err; 2193} 2194 2195#define PARENT_INO(buffer) \ 2196 ((struct ext3_dir_entry_2 *) ((char *) buffer + \ 2197 le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode 2198 2199/* 2200 * Anybody can rename anything with this: the permission checks are left to the 2201 * higher-level routines. 2202 */ 2203static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry, 2204 struct inode * new_dir,struct dentry *new_dentry) 2205{ 2206 handle_t *handle; 2207 struct inode * old_inode, * new_inode; 2208 struct buffer_head * old_bh, * new_bh, * dir_bh; 2209 struct ext3_dir_entry_2 * old_de, * new_de; 2210 int retval; 2211 2212 old_bh = new_bh = dir_bh = NULL; 2213 2214 /* Initialize quotas before so that eventual writes go 2215 * in separate transaction */ 2216 if (new_dentry->d_inode) 2217 DQUOT_INIT(new_dentry->d_inode); 2218 handle = ext3_journal_start(old_dir, 2 * 2219 EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) + 2220 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2); 2221 if (IS_ERR(handle)) 2222 return PTR_ERR(handle); 2223 2224 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir)) 2225 handle->h_sync = 1; 2226 2227 old_bh = ext3_find_entry (old_dentry, &old_de); 2228 /* 2229 * Check for inode number is _not_ due to possible IO errors. 2230 * We might rmdir the source, keep it as pwd of some process 2231 * and merrily kill the link to whatever was created under the 2232 * same name. Goodbye sticky bit ;-< 2233 */ 2234 old_inode = old_dentry->d_inode; 2235 retval = -ENOENT; 2236 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino) 2237 goto end_rename; 2238 2239 new_inode = new_dentry->d_inode; 2240 new_bh = ext3_find_entry (new_dentry, &new_de); 2241 if (new_bh) { 2242 if (!new_inode) { 2243 brelse (new_bh); 2244 new_bh = NULL; 2245 } 2246 } 2247 if (S_ISDIR(old_inode->i_mode)) { 2248 if (new_inode) { 2249 retval = -ENOTEMPTY; 2250 if (!empty_dir (new_inode)) 2251 goto end_rename; 2252 } 2253 retval = -EIO; 2254 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval); 2255 if (!dir_bh) 2256 goto end_rename; 2257 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino) 2258 goto end_rename; 2259 retval = -EMLINK; 2260 if (!new_inode && new_dir!=old_dir && 2261 new_dir->i_nlink >= EXT3_LINK_MAX) 2262 goto end_rename; 2263 } 2264 if (!new_bh) { 2265 retval = ext3_add_entry (handle, new_dentry, old_inode); 2266 if (retval) 2267 goto end_rename; 2268 } else { 2269 BUFFER_TRACE(new_bh, "get write access"); 2270 ext3_journal_get_write_access(handle, new_bh); 2271 new_de->inode = cpu_to_le32(old_inode->i_ino); 2272 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb, 2273 EXT3_FEATURE_INCOMPAT_FILETYPE)) 2274 new_de->file_type = old_de->file_type; 2275 new_dir->i_version++; 2276 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata"); 2277 ext3_journal_dirty_metadata(handle, new_bh); 2278 brelse(new_bh); 2279 new_bh = NULL; 2280 } 2281 2282 /* 2283 * Like most other Unix systems, set the ctime for inodes on a 2284 * rename. 2285 */ 2286 old_inode->i_ctime = CURRENT_TIME_SEC; 2287 ext3_mark_inode_dirty(handle, old_inode); 2288 2289 /* 2290 * ok, that's it 2291 */ 2292 if (le32_to_cpu(old_de->inode) != old_inode->i_ino || 2293 old_de->name_len != old_dentry->d_name.len || 2294 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) || 2295 (retval = ext3_delete_entry(handle, old_dir, 2296 old_de, old_bh)) == -ENOENT) { 2297 /* old_de could have moved from under us during htree split, so 2298 * make sure that we are deleting the right entry. We might 2299 * also be pointing to a stale entry in the unused part of 2300 * old_bh so just checking inum and the name isn't enough. */ 2301 struct buffer_head *old_bh2; 2302 struct ext3_dir_entry_2 *old_de2; 2303 2304 old_bh2 = ext3_find_entry(old_dentry, &old_de2); 2305 if (old_bh2) { 2306 retval = ext3_delete_entry(handle, old_dir, 2307 old_de2, old_bh2); 2308 brelse(old_bh2); 2309 } 2310 } 2311 if (retval) { 2312 ext3_warning(old_dir->i_sb, "ext3_rename", 2313 "Deleting old file (%lu), %d, error=%d", 2314 old_dir->i_ino, old_dir->i_nlink, retval); 2315 } 2316 2317 if (new_inode) { 2318 new_inode->i_nlink--; 2319 new_inode->i_ctime = CURRENT_TIME_SEC; 2320 } 2321 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC; 2322 ext3_update_dx_flag(old_dir); 2323 if (dir_bh) { 2324 BUFFER_TRACE(dir_bh, "get_write_access"); 2325 ext3_journal_get_write_access(handle, dir_bh); 2326 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino); 2327 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata"); 2328 ext3_journal_dirty_metadata(handle, dir_bh); 2329 old_dir->i_nlink--; 2330 if (new_inode) { 2331 new_inode->i_nlink--; 2332 } else { 2333 new_dir->i_nlink++; 2334 ext3_update_dx_flag(new_dir); 2335 ext3_mark_inode_dirty(handle, new_dir); 2336 } 2337 } 2338 ext3_mark_inode_dirty(handle, old_dir); 2339 if (new_inode) { 2340 ext3_mark_inode_dirty(handle, new_inode); 2341 if (!new_inode->i_nlink) 2342 ext3_orphan_add(handle, new_inode); 2343 } 2344 retval = 0; 2345 2346end_rename: 2347 brelse (dir_bh); 2348 brelse (old_bh); 2349 brelse (new_bh); 2350 ext3_journal_stop(handle); 2351 return retval; 2352} 2353 2354/* 2355 * directories can handle most operations... 2356 */ 2357struct inode_operations ext3_dir_inode_operations = { 2358 .create = ext3_create, 2359 .lookup = ext3_lookup, 2360 .link = ext3_link, 2361 .unlink = ext3_unlink, 2362 .symlink = ext3_symlink, 2363 .mkdir = ext3_mkdir, 2364 .rmdir = ext3_rmdir, 2365 .mknod = ext3_mknod, 2366 .rename = ext3_rename, 2367 .setattr = ext3_setattr, 2368#ifdef CONFIG_EXT3_FS_XATTR 2369 .setxattr = generic_setxattr, 2370 .getxattr = generic_getxattr, 2371 .listxattr = ext3_listxattr, 2372 .removexattr = generic_removexattr, 2373#endif 2374 .permission = ext3_permission, 2375}; 2376 2377struct inode_operations ext3_special_inode_operations = { 2378 .setattr = ext3_setattr, 2379#ifdef CONFIG_EXT3_FS_XATTR 2380 .setxattr = generic_setxattr, 2381 .getxattr = generic_getxattr, 2382 .listxattr = ext3_listxattr, 2383 .removexattr = generic_removexattr, 2384#endif 2385 .permission = ext3_permission, 2386}; 2387