2 * linux/fs/ext3/namei.c
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)
11 * linux/fs/minix/namei.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
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
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>
45 * define how far ahead to read directories while searching them.
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))
52 static struct buffer_head *ext3_append(handle_t *handle,
56 struct buffer_head *bh;
58 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
60 bh = ext3_bread(handle, inode, *block, 1, err);
62 inode->i_size += inode->i_sb->s_blocksize;
63 EXT3_I(inode)->i_disksize = inode->i_size;
64 *err = ext3_journal_get_write_access(handle, bh);
74 #define assert(test) J_ASSERT(test)
78 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
82 #define dxtrace(command) command
84 #define dxtrace(command)
108 * dx_root_info is laid out so that if it should somehow get overlaid by a
109 * dirent the two low bits of the hash version will be zero. Therefore, the
110 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
115 struct fake_dirent dot;
117 struct fake_dirent dotdot;
121 __le32 reserved_zero;
123 u8 info_length; /* 8 */
128 struct dx_entry entries[0];
133 struct fake_dirent fake;
134 struct dx_entry entries[0];
140 struct buffer_head *bh;
141 struct dx_entry *entries;
152 static inline unsigned dx_get_block (struct dx_entry *entry);
153 static void dx_set_block (struct dx_entry *entry, unsigned value);
154 static inline unsigned dx_get_hash (struct dx_entry *entry);
155 static void dx_set_hash (struct dx_entry *entry, unsigned value);
156 static unsigned dx_get_count (struct dx_entry *entries);
157 static unsigned dx_get_limit (struct dx_entry *entries);
158 static void dx_set_count (struct dx_entry *entries, unsigned value);
159 static void dx_set_limit (struct dx_entry *entries, unsigned value);
160 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
161 static unsigned dx_node_limit (struct inode *dir);
162 static struct dx_frame *dx_probe(struct dentry *dentry,
164 struct dx_hash_info *hinfo,
165 struct dx_frame *frame,
167 static void dx_release (struct dx_frame *frames);
168 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
169 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
170 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
171 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
172 struct dx_map_entry *offsets, int count);
173 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
174 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
175 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
176 struct dx_frame *frame,
177 struct dx_frame *frames,
179 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
180 struct ext3_dir_entry_2 **res_dir, int *err);
181 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
182 struct inode *inode);
185 * p is at least 6 bytes before the end of page
187 static inline struct ext3_dir_entry_2 *
188 ext3_next_entry(struct ext3_dir_entry_2 *p)
190 return (struct ext3_dir_entry_2 *)((char *)p +
191 ext3_rec_len_from_disk(p->rec_len));
195 * Future: use high four bits of block for coalesce-on-delete flags
196 * Mask them off for now.
199 static inline unsigned dx_get_block (struct dx_entry *entry)
201 return le32_to_cpu(entry->block) & 0x00ffffff;
204 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
206 entry->block = cpu_to_le32(value);
209 static inline unsigned dx_get_hash (struct dx_entry *entry)
211 return le32_to_cpu(entry->hash);
214 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
216 entry->hash = cpu_to_le32(value);
219 static inline unsigned dx_get_count (struct dx_entry *entries)
221 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
224 static inline unsigned dx_get_limit (struct dx_entry *entries)
226 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
229 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
231 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
234 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
236 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
239 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
241 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
242 EXT3_DIR_REC_LEN(2) - infosize;
243 return entry_space / sizeof(struct dx_entry);
246 static inline unsigned dx_node_limit (struct inode *dir)
248 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
249 return entry_space / sizeof(struct dx_entry);
256 static void dx_show_index (char * label, struct dx_entry *entries)
258 int i, n = dx_get_count (entries);
259 printk("%s index ", label);
260 for (i = 0; i < n; i++)
262 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
274 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
275 int size, int show_names)
277 unsigned names = 0, space = 0;
278 char *base = (char *) de;
279 struct dx_hash_info h = *hinfo;
282 while ((char *) de < base + size)
288 int len = de->name_len;
289 char *name = de->name;
290 while (len--) printk("%c", *name++);
291 ext3fs_dirhash(de->name, de->name_len, &h);
292 printk(":%x.%u ", h.hash,
293 ((char *) de - base));
295 space += EXT3_DIR_REC_LEN(de->name_len);
298 de = ext3_next_entry(de);
300 printk("(%i)\n", names);
301 return (struct stats) { names, space, 1 };
304 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
305 struct dx_entry *entries, int levels)
307 unsigned blocksize = dir->i_sb->s_blocksize;
308 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
310 struct buffer_head *bh;
312 printk("%i indexed blocks...\n", count);
313 for (i = 0; i < count; i++, entries++)
315 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
316 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
318 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
319 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
321 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
322 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
323 names += stats.names;
324 space += stats.space;
325 bcount += stats.bcount;
329 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
330 names, space/bcount,(space/bcount)*100/blocksize);
331 return (struct stats) { names, space, bcount};
333 #endif /* DX_DEBUG */
336 * Probe for a directory leaf block to search.
338 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
339 * error in the directory index, and the caller should fall back to
340 * searching the directory normally. The callers of dx_probe **MUST**
341 * check for this error code, and make sure it never gets reflected
344 static struct dx_frame *
345 dx_probe(struct dentry *dentry, struct inode *dir,
346 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
348 unsigned count, indirect;
349 struct dx_entry *at, *entries, *p, *q, *m;
350 struct dx_root *root;
351 struct buffer_head *bh;
352 struct dx_frame *frame = frame_in;
357 dir = dentry->d_parent->d_inode;
358 if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
360 root = (struct dx_root *) bh->b_data;
361 if (root->info.hash_version != DX_HASH_TEA &&
362 root->info.hash_version != DX_HASH_HALF_MD4 &&
363 root->info.hash_version != DX_HASH_LEGACY) {
364 ext3_warning(dir->i_sb, __func__,
365 "Unrecognised inode hash code %d",
366 root->info.hash_version);
368 *err = ERR_BAD_DX_DIR;
371 hinfo->hash_version = root->info.hash_version;
372 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
374 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
377 if (root->info.unused_flags & 1) {
378 ext3_warning(dir->i_sb, __func__,
379 "Unimplemented inode hash flags: %#06x",
380 root->info.unused_flags);
382 *err = ERR_BAD_DX_DIR;
386 if ((indirect = root->info.indirect_levels) > 1) {
387 ext3_warning(dir->i_sb, __func__,
388 "Unimplemented inode hash depth: %#06x",
389 root->info.indirect_levels);
391 *err = ERR_BAD_DX_DIR;
395 entries = (struct dx_entry *) (((char *)&root->info) +
396 root->info.info_length);
398 if (dx_get_limit(entries) != dx_root_limit(dir,
399 root->info.info_length)) {
400 ext3_warning(dir->i_sb, __func__,
401 "dx entry: limit != root limit");
403 *err = ERR_BAD_DX_DIR;
407 dxtrace (printk("Look up %x", hash));
410 count = dx_get_count(entries);
411 if (!count || count > dx_get_limit(entries)) {
412 ext3_warning(dir->i_sb, __func__,
413 "dx entry: no count or count > limit");
415 *err = ERR_BAD_DX_DIR;
420 q = entries + count - 1;
424 dxtrace(printk("."));
425 if (dx_get_hash(m) > hash)
431 if (0) // linear search cross check
433 unsigned n = count - 1;
437 dxtrace(printk(","));
438 if (dx_get_hash(++at) > hash)
444 assert (at == p - 1);
448 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
450 frame->entries = entries;
452 if (!indirect--) return frame;
453 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
455 at = entries = ((struct dx_node *) bh->b_data)->entries;
456 if (dx_get_limit(entries) != dx_node_limit (dir)) {
457 ext3_warning(dir->i_sb, __func__,
458 "dx entry: limit != node limit");
460 *err = ERR_BAD_DX_DIR;
467 while (frame >= frame_in) {
472 if (*err == ERR_BAD_DX_DIR)
473 ext3_warning(dir->i_sb, __func__,
474 "Corrupt dir inode %ld, running e2fsck is "
475 "recommended.", dir->i_ino);
479 static void dx_release (struct dx_frame *frames)
481 if (frames[0].bh == NULL)
484 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
485 brelse(frames[1].bh);
486 brelse(frames[0].bh);
490 * This function increments the frame pointer to search the next leaf
491 * block, and reads in the necessary intervening nodes if the search
492 * should be necessary. Whether or not the search is necessary is
493 * controlled by the hash parameter. If the hash value is even, then
494 * the search is only continued if the next block starts with that
495 * hash value. This is used if we are searching for a specific file.
497 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
499 * This function returns 1 if the caller should continue to search,
500 * or 0 if it should not. If there is an error reading one of the
501 * index blocks, it will a negative error code.
503 * If start_hash is non-null, it will be filled in with the starting
504 * hash of the next page.
506 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
507 struct dx_frame *frame,
508 struct dx_frame *frames,
512 struct buffer_head *bh;
513 int err, num_frames = 0;
518 * Find the next leaf page by incrementing the frame pointer.
519 * If we run out of entries in the interior node, loop around and
520 * increment pointer in the parent node. When we break out of
521 * this loop, num_frames indicates the number of interior
522 * nodes need to be read.
525 if (++(p->at) < p->entries + dx_get_count(p->entries))
534 * If the hash is 1, then continue only if the next page has a
535 * continuation hash of any value. This is used for readdir
536 * handling. Otherwise, check to see if the hash matches the
537 * desired contiuation hash. If it doesn't, return since
538 * there's no point to read in the successive index pages.
540 bhash = dx_get_hash(p->at);
543 if ((hash & 1) == 0) {
544 if ((bhash & ~1) != hash)
548 * If the hash is HASH_NB_ALWAYS, we always go to the next
549 * block so no check is necessary
551 while (num_frames--) {
552 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
554 return err; /* Failure */
558 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
565 * This function fills a red-black tree with information from a
566 * directory block. It returns the number directory entries loaded
567 * into the tree. If there is an error it is returned in err.
569 static int htree_dirblock_to_tree(struct file *dir_file,
570 struct inode *dir, int block,
571 struct dx_hash_info *hinfo,
572 __u32 start_hash, __u32 start_minor_hash)
574 struct buffer_head *bh;
575 struct ext3_dir_entry_2 *de, *top;
578 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
579 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
582 de = (struct ext3_dir_entry_2 *) bh->b_data;
583 top = (struct ext3_dir_entry_2 *) ((char *) de +
584 dir->i_sb->s_blocksize -
585 EXT3_DIR_REC_LEN(0));
586 for (; de < top; de = ext3_next_entry(de)) {
587 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
588 (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
589 +((char *)de - bh->b_data))) {
590 /* On error, skip the f_pos to the next block. */
591 dir_file->f_pos = (dir_file->f_pos |
592 (dir->i_sb->s_blocksize - 1)) + 1;
596 ext3fs_dirhash(de->name, de->name_len, hinfo);
597 if ((hinfo->hash < start_hash) ||
598 ((hinfo->hash == start_hash) &&
599 (hinfo->minor_hash < start_minor_hash)))
603 if ((err = ext3_htree_store_dirent(dir_file,
604 hinfo->hash, hinfo->minor_hash, de)) != 0) {
616 * This function fills a red-black tree with information from a
617 * directory. We start scanning the directory in hash order, starting
618 * at start_hash and start_minor_hash.
620 * This function returns the number of entries inserted into the tree,
621 * or a negative error code.
623 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
624 __u32 start_minor_hash, __u32 *next_hash)
626 struct dx_hash_info hinfo;
627 struct ext3_dir_entry_2 *de;
628 struct dx_frame frames[2], *frame;
635 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
637 dir = dir_file->f_path.dentry->d_inode;
638 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
639 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
640 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
641 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
642 start_hash, start_minor_hash);
646 hinfo.hash = start_hash;
647 hinfo.minor_hash = 0;
648 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
652 /* Add '.' and '..' from the htree header */
653 if (!start_hash && !start_minor_hash) {
654 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
655 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
659 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
660 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
661 de = ext3_next_entry(de);
662 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
668 block = dx_get_block(frame->at);
669 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
670 start_hash, start_minor_hash);
677 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
678 frame, frames, &hashval);
679 *next_hash = hashval;
685 * Stop if: (a) there are no more entries, or
686 * (b) we have inserted at least one entry and the
687 * next hash value is not a continuation
690 (count && ((hashval & 1) == 0)))
694 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
704 * Directory block splitting, compacting
708 * Create map of hash values, offsets, and sizes, stored at end of block.
709 * Returns number of entries mapped.
711 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
712 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
715 char *base = (char *) de;
716 struct dx_hash_info h = *hinfo;
718 while ((char *) de < base + size)
720 if (de->name_len && de->inode) {
721 ext3fs_dirhash(de->name, de->name_len, &h);
723 map_tail->hash = h.hash;
724 map_tail->offs = (u16) ((char *) de - base);
725 map_tail->size = le16_to_cpu(de->rec_len);
729 /* XXX: do we need to check rec_len == 0 case? -Chris */
730 de = ext3_next_entry(de);
735 /* Sort map by hash value */
736 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
738 struct dx_map_entry *p, *q, *top = map + count - 1;
740 /* Combsort until bubble sort doesn't suck */
744 if (count - 9 < 2) /* 9, 10 -> 11 */
746 for (p = top, q = p - count; q >= map; p--, q--)
747 if (p->hash < q->hash)
750 /* Garden variety bubble sort */
756 if (q[1].hash >= q[0].hash)
764 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
766 struct dx_entry *entries = frame->entries;
767 struct dx_entry *old = frame->at, *new = old + 1;
768 int count = dx_get_count(entries);
770 assert(count < dx_get_limit(entries));
771 assert(old < entries + count);
772 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
773 dx_set_hash(new, hash);
774 dx_set_block(new, block);
775 dx_set_count(entries, count + 1);
778 static void ext3_update_dx_flag(struct inode *inode)
780 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
781 EXT3_FEATURE_COMPAT_DIR_INDEX))
782 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
786 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
788 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
789 * `de != NULL' is guaranteed by caller.
791 static inline int ext3_match (int len, const char * const name,
792 struct ext3_dir_entry_2 * de)
794 if (len != de->name_len)
798 return !memcmp(name, de->name, len);
802 * Returns 0 if not found, -1 on failure, and 1 on success
804 static inline int search_dirblock(struct buffer_head * bh,
806 struct dentry *dentry,
807 unsigned long offset,
808 struct ext3_dir_entry_2 ** res_dir)
810 struct ext3_dir_entry_2 * de;
813 const char *name = dentry->d_name.name;
814 int namelen = dentry->d_name.len;
816 de = (struct ext3_dir_entry_2 *) bh->b_data;
817 dlimit = bh->b_data + dir->i_sb->s_blocksize;
818 while ((char *) de < dlimit) {
819 /* this code is executed quadratically often */
820 /* do minimal checking `by hand' */
822 if ((char *) de + namelen <= dlimit &&
823 ext3_match (namelen, name, de)) {
824 /* found a match - just to be sure, do a full check */
825 if (!ext3_check_dir_entry("ext3_find_entry",
826 dir, de, bh, offset))
831 /* prevent looping on a bad block */
832 de_len = ext3_rec_len_from_disk(de->rec_len);
836 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
845 * finds an entry in the specified directory with the wanted name. It
846 * returns the cache buffer in which the entry was found, and the entry
847 * itself (as a parameter - res_dir). It does NOT read the inode of the
848 * entry - you'll have to do that yourself if you want to.
850 * The returned buffer_head has ->b_count elevated. The caller is expected
851 * to brelse() it when appropriate.
853 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
854 struct ext3_dir_entry_2 ** res_dir)
856 struct super_block * sb;
857 struct buffer_head * bh_use[NAMEI_RA_SIZE];
858 struct buffer_head * bh, *ret = NULL;
859 unsigned long start, block, b;
860 int ra_max = 0; /* Number of bh's in the readahead
862 int ra_ptr = 0; /* Current index into readahead
866 struct inode *dir = dentry->d_parent->d_inode;
871 namelen = dentry->d_name.len;
872 if (namelen > EXT3_NAME_LEN)
875 bh = ext3_dx_find_entry(dentry, res_dir, &err);
877 * On success, or if the error was file not found,
878 * return. Otherwise, fall back to doing a search the
881 if (bh || (err != ERR_BAD_DX_DIR))
883 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
885 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
886 start = EXT3_I(dir)->i_dir_start_lookup;
887 if (start >= nblocks)
893 * We deal with the read-ahead logic here.
895 if (ra_ptr >= ra_max) {
896 /* Refill the readahead buffer */
899 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
901 * Terminate if we reach the end of the
902 * directory and must wrap, or if our
903 * search has finished at this block.
905 if (b >= nblocks || (num && block == start)) {
906 bh_use[ra_max] = NULL;
910 bh = ext3_getblk(NULL, dir, b++, 0, &err);
913 ll_rw_block(READ_META, 1, &bh);
916 if ((bh = bh_use[ra_ptr++]) == NULL)
919 if (!buffer_uptodate(bh)) {
920 /* read error, skip block & hope for the best */
921 ext3_error(sb, __func__, "reading directory #%lu "
922 "offset %lu", dir->i_ino, block);
926 i = search_dirblock(bh, dir, dentry,
927 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
929 EXT3_I(dir)->i_dir_start_lookup = block;
931 goto cleanup_and_exit;
935 goto cleanup_and_exit;
938 if (++block >= nblocks)
940 } while (block != start);
943 * If the directory has grown while we were searching, then
944 * search the last part of the directory before giving up.
947 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
948 if (block < nblocks) {
954 /* Clean up the read-ahead blocks */
955 for (; ra_ptr < ra_max; ra_ptr++)
956 brelse (bh_use[ra_ptr]);
960 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
961 struct ext3_dir_entry_2 **res_dir, int *err)
963 struct super_block * sb;
964 struct dx_hash_info hinfo;
966 struct dx_frame frames[2], *frame;
967 struct ext3_dir_entry_2 *de, *top;
968 struct buffer_head *bh;
971 int namelen = dentry->d_name.len;
972 const u8 *name = dentry->d_name.name;
973 struct inode *dir = dentry->d_parent->d_inode;
976 /* NFS may look up ".." - look at dx_root directory block */
977 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
978 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
982 frame->bh = NULL; /* for dx_release() */
983 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
984 dx_set_block(frame->at, 0); /* dx_root block is 0 */
988 block = dx_get_block(frame->at);
989 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
991 de = (struct ext3_dir_entry_2 *) bh->b_data;
992 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
993 EXT3_DIR_REC_LEN(0));
994 for (; de < top; de = ext3_next_entry(de)) {
995 int off = (block << EXT3_BLOCK_SIZE_BITS(sb))
996 + ((char *) de - bh->b_data);
998 if (!ext3_check_dir_entry(__func__, dir, de, bh, off)) {
1000 *err = ERR_BAD_DX_DIR;
1004 if (ext3_match(namelen, name, de)) {
1011 /* Check to see if we should continue to search */
1012 retval = ext3_htree_next_block(dir, hash, frame,
1015 ext3_warning(sb, __func__,
1016 "error reading index page in directory #%lu",
1021 } while (retval == 1);
1025 dxtrace(printk("%s not found\n", name));
1026 dx_release (frames);
1030 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1032 struct inode * inode;
1033 struct ext3_dir_entry_2 * de;
1034 struct buffer_head * bh;
1036 if (dentry->d_name.len > EXT3_NAME_LEN)
1037 return ERR_PTR(-ENAMETOOLONG);
1039 bh = ext3_find_entry(dentry, &de);
1042 unsigned long ino = le32_to_cpu(de->inode);
1044 if (!ext3_valid_inum(dir->i_sb, ino)) {
1045 ext3_error(dir->i_sb, "ext3_lookup",
1046 "bad inode number: %lu", ino);
1047 return ERR_PTR(-EIO);
1049 inode = ext3_iget(dir->i_sb, ino);
1051 return ERR_CAST(inode);
1053 return d_splice_alias(inode, dentry);
1057 struct dentry *ext3_get_parent(struct dentry *child)
1060 struct dentry dotdot;
1061 struct ext3_dir_entry_2 * de;
1062 struct buffer_head *bh;
1064 dotdot.d_name.name = "..";
1065 dotdot.d_name.len = 2;
1066 dotdot.d_parent = child; /* confusing, isn't it! */
1068 bh = ext3_find_entry(&dotdot, &de);
1070 return ERR_PTR(-ENOENT);
1071 ino = le32_to_cpu(de->inode);
1074 if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1075 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1076 "bad inode number: %lu", ino);
1077 return ERR_PTR(-EIO);
1080 return d_obtain_alias(ext3_iget(child->d_inode->i_sb, ino));
1084 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1085 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1086 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1087 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1088 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1089 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1090 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1091 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1094 static inline void ext3_set_de_type(struct super_block *sb,
1095 struct ext3_dir_entry_2 *de,
1097 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1098 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1102 * Move count entries from end of map between two memory locations.
1103 * Returns pointer to last entry moved.
1105 static struct ext3_dir_entry_2 *
1106 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1108 unsigned rec_len = 0;
1111 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1112 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1113 memcpy (to, de, rec_len);
1114 ((struct ext3_dir_entry_2 *) to)->rec_len =
1115 ext3_rec_len_to_disk(rec_len);
1120 return (struct ext3_dir_entry_2 *) (to - rec_len);
1124 * Compact each dir entry in the range to the minimal rec_len.
1125 * Returns pointer to last entry in range.
1127 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1129 struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1130 unsigned rec_len = 0;
1133 while ((char*)de < base + size) {
1134 next = ext3_next_entry(de);
1135 if (de->inode && de->name_len) {
1136 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1138 memmove(to, de, rec_len);
1139 to->rec_len = ext3_rec_len_to_disk(rec_len);
1141 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1149 * Split a full leaf block to make room for a new dir entry.
1150 * Allocate a new block, and move entries so that they are approx. equally full.
1151 * Returns pointer to de in block into which the new entry will be inserted.
1153 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1154 struct buffer_head **bh,struct dx_frame *frame,
1155 struct dx_hash_info *hinfo, int *error)
1157 unsigned blocksize = dir->i_sb->s_blocksize;
1158 unsigned count, continued;
1159 struct buffer_head *bh2;
1162 struct dx_map_entry *map;
1163 char *data1 = (*bh)->b_data, *data2;
1164 unsigned split, move, size, i;
1165 struct ext3_dir_entry_2 *de = NULL, *de2;
1168 bh2 = ext3_append (handle, dir, &newblock, &err);
1175 BUFFER_TRACE(*bh, "get_write_access");
1176 err = ext3_journal_get_write_access(handle, *bh);
1180 BUFFER_TRACE(frame->bh, "get_write_access");
1181 err = ext3_journal_get_write_access(handle, frame->bh);
1185 data2 = bh2->b_data;
1187 /* create map in the end of data2 block */
1188 map = (struct dx_map_entry *) (data2 + blocksize);
1189 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1190 blocksize, hinfo, map);
1192 dx_sort_map (map, count);
1193 /* Split the existing block in the middle, size-wise */
1196 for (i = count-1; i >= 0; i--) {
1197 /* is more than half of this entry in 2nd half of the block? */
1198 if (size + map[i].size/2 > blocksize/2)
1200 size += map[i].size;
1203 /* map index at which we will split */
1204 split = count - move;
1205 hash2 = map[split].hash;
1206 continued = hash2 == map[split - 1].hash;
1207 dxtrace(printk("Split block %i at %x, %i/%i\n",
1208 dx_get_block(frame->at), hash2, split, count-split));
1210 /* Fancy dance to stay within two buffers */
1211 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1212 de = dx_pack_dirents(data1,blocksize);
1213 de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1214 de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
1215 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1216 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1218 /* Which block gets the new entry? */
1219 if (hinfo->hash >= hash2)
1224 dx_insert_block (frame, hash2 + continued, newblock);
1225 err = ext3_journal_dirty_metadata (handle, bh2);
1228 err = ext3_journal_dirty_metadata (handle, frame->bh);
1232 dxtrace(dx_show_index ("frame", frame->entries));
1239 ext3_std_error(dir->i_sb, err);
1247 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1248 * it points to a directory entry which is guaranteed to be large
1249 * enough for new directory entry. If de is NULL, then
1250 * add_dirent_to_buf will attempt search the directory block for
1251 * space. It will return -ENOSPC if no space is available, and -EIO
1252 * and -EEXIST if directory entry already exists.
1254 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1255 * all other cases bh is released.
1257 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1258 struct inode *inode, struct ext3_dir_entry_2 *de,
1259 struct buffer_head * bh)
1261 struct inode *dir = dentry->d_parent->d_inode;
1262 const char *name = dentry->d_name.name;
1263 int namelen = dentry->d_name.len;
1264 unsigned long offset = 0;
1265 unsigned short reclen;
1266 int nlen, rlen, err;
1269 reclen = EXT3_DIR_REC_LEN(namelen);
1271 de = (struct ext3_dir_entry_2 *)bh->b_data;
1272 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1273 while ((char *) de <= top) {
1274 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1279 if (ext3_match (namelen, name, de)) {
1283 nlen = EXT3_DIR_REC_LEN(de->name_len);
1284 rlen = ext3_rec_len_from_disk(de->rec_len);
1285 if ((de->inode? rlen - nlen: rlen) >= reclen)
1287 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1290 if ((char *) de > top)
1293 BUFFER_TRACE(bh, "get_write_access");
1294 err = ext3_journal_get_write_access(handle, bh);
1296 ext3_std_error(dir->i_sb, err);
1301 /* By now the buffer is marked for journaling */
1302 nlen = EXT3_DIR_REC_LEN(de->name_len);
1303 rlen = ext3_rec_len_from_disk(de->rec_len);
1305 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1306 de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1307 de->rec_len = ext3_rec_len_to_disk(nlen);
1310 de->file_type = EXT3_FT_UNKNOWN;
1312 de->inode = cpu_to_le32(inode->i_ino);
1313 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1316 de->name_len = namelen;
1317 memcpy (de->name, name, namelen);
1319 * XXX shouldn't update any times until successful
1320 * completion of syscall, but too many callers depend
1323 * XXX similarly, too many callers depend on
1324 * ext3_new_inode() setting the times, but error
1325 * recovery deletes the inode, so the worst that can
1326 * happen is that the times are slightly out of date
1327 * and/or different from the directory change time.
1329 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1330 ext3_update_dx_flag(dir);
1332 ext3_mark_inode_dirty(handle, dir);
1333 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1334 err = ext3_journal_dirty_metadata(handle, bh);
1336 ext3_std_error(dir->i_sb, err);
1342 * This converts a one block unindexed directory to a 3 block indexed
1343 * directory, and adds the dentry to the indexed directory.
1345 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1346 struct inode *inode, struct buffer_head *bh)
1348 struct inode *dir = dentry->d_parent->d_inode;
1349 const char *name = dentry->d_name.name;
1350 int namelen = dentry->d_name.len;
1351 struct buffer_head *bh2;
1352 struct dx_root *root;
1353 struct dx_frame frames[2], *frame;
1354 struct dx_entry *entries;
1355 struct ext3_dir_entry_2 *de, *de2;
1360 struct dx_hash_info hinfo;
1362 struct fake_dirent *fde;
1364 blocksize = dir->i_sb->s_blocksize;
1365 dxtrace(printk("Creating index\n"));
1366 retval = ext3_journal_get_write_access(handle, bh);
1368 ext3_std_error(dir->i_sb, retval);
1372 root = (struct dx_root *) bh->b_data;
1374 bh2 = ext3_append (handle, dir, &block, &retval);
1379 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1380 data1 = bh2->b_data;
1382 /* The 0th block becomes the root, move the dirents out */
1383 fde = &root->dotdot;
1384 de = (struct ext3_dir_entry_2 *)((char *)fde +
1385 ext3_rec_len_from_disk(fde->rec_len));
1386 len = ((char *) root) + blocksize - (char *) de;
1387 memcpy (data1, de, len);
1388 de = (struct ext3_dir_entry_2 *) data1;
1390 while ((char *)(de2 = ext3_next_entry(de)) < top)
1392 de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1393 /* Initialize the root; the dot dirents already exist */
1394 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1395 de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
1396 memset (&root->info, 0, sizeof(root->info));
1397 root->info.info_length = sizeof(root->info);
1398 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1399 entries = root->entries;
1400 dx_set_block (entries, 1);
1401 dx_set_count (entries, 1);
1402 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1404 /* Initialize as for dx_probe */
1405 hinfo.hash_version = root->info.hash_version;
1406 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1407 ext3fs_dirhash(name, namelen, &hinfo);
1409 frame->entries = entries;
1410 frame->at = entries;
1413 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1414 dx_release (frames);
1418 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1424 * adds a file entry to the specified directory, using the same
1425 * semantics as ext3_find_entry(). It returns NULL if it failed.
1427 * NOTE!! The inode part of 'de' is left at 0 - which means you
1428 * may not sleep between calling this and putting something into
1429 * the entry, as someone else might have used it while you slept.
1431 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1432 struct inode *inode)
1434 struct inode *dir = dentry->d_parent->d_inode;
1435 unsigned long offset;
1436 struct buffer_head * bh;
1437 struct ext3_dir_entry_2 *de;
1438 struct super_block * sb;
1445 blocksize = sb->s_blocksize;
1446 if (!dentry->d_name.len)
1449 retval = ext3_dx_add_entry(handle, dentry, inode);
1450 if (!retval || (retval != ERR_BAD_DX_DIR))
1452 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1454 ext3_mark_inode_dirty(handle, dir);
1456 blocks = dir->i_size >> sb->s_blocksize_bits;
1457 for (block = 0, offset = 0; block < blocks; block++) {
1458 bh = ext3_bread(handle, dir, block, 0, &retval);
1461 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1462 if (retval != -ENOSPC)
1465 if (blocks == 1 && !dx_fallback &&
1466 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1467 return make_indexed_dir(handle, dentry, inode, bh);
1470 bh = ext3_append(handle, dir, &block, &retval);
1473 de = (struct ext3_dir_entry_2 *) bh->b_data;
1475 de->rec_len = ext3_rec_len_to_disk(blocksize);
1476 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1480 * Returns 0 for success, or a negative error value
1482 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1483 struct inode *inode)
1485 struct dx_frame frames[2], *frame;
1486 struct dx_entry *entries, *at;
1487 struct dx_hash_info hinfo;
1488 struct buffer_head * bh;
1489 struct inode *dir = dentry->d_parent->d_inode;
1490 struct super_block * sb = dir->i_sb;
1491 struct ext3_dir_entry_2 *de;
1494 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1497 entries = frame->entries;
1500 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1503 BUFFER_TRACE(bh, "get_write_access");
1504 err = ext3_journal_get_write_access(handle, bh);
1508 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1509 if (err != -ENOSPC) {
1514 /* Block full, should compress but for now just split */
1515 dxtrace(printk("using %u of %u node entries\n",
1516 dx_get_count(entries), dx_get_limit(entries)));
1517 /* Need to split index? */
1518 if (dx_get_count(entries) == dx_get_limit(entries)) {
1520 unsigned icount = dx_get_count(entries);
1521 int levels = frame - frames;
1522 struct dx_entry *entries2;
1523 struct dx_node *node2;
1524 struct buffer_head *bh2;
1526 if (levels && (dx_get_count(frames->entries) ==
1527 dx_get_limit(frames->entries))) {
1528 ext3_warning(sb, __func__,
1529 "Directory index full!");
1533 bh2 = ext3_append (handle, dir, &newblock, &err);
1536 node2 = (struct dx_node *)(bh2->b_data);
1537 entries2 = node2->entries;
1538 node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
1539 node2->fake.inode = 0;
1540 BUFFER_TRACE(frame->bh, "get_write_access");
1541 err = ext3_journal_get_write_access(handle, frame->bh);
1545 unsigned icount1 = icount/2, icount2 = icount - icount1;
1546 unsigned hash2 = dx_get_hash(entries + icount1);
1547 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1549 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1550 err = ext3_journal_get_write_access(handle,
1555 memcpy ((char *) entries2, (char *) (entries + icount1),
1556 icount2 * sizeof(struct dx_entry));
1557 dx_set_count (entries, icount1);
1558 dx_set_count (entries2, icount2);
1559 dx_set_limit (entries2, dx_node_limit(dir));
1561 /* Which index block gets the new entry? */
1562 if (at - entries >= icount1) {
1563 frame->at = at = at - entries - icount1 + entries2;
1564 frame->entries = entries = entries2;
1565 swap(frame->bh, bh2);
1567 dx_insert_block (frames + 0, hash2, newblock);
1568 dxtrace(dx_show_index ("node", frames[1].entries));
1569 dxtrace(dx_show_index ("node",
1570 ((struct dx_node *) bh2->b_data)->entries));
1571 err = ext3_journal_dirty_metadata(handle, bh2);
1576 dxtrace(printk("Creating second level index...\n"));
1577 memcpy((char *) entries2, (char *) entries,
1578 icount * sizeof(struct dx_entry));
1579 dx_set_limit(entries2, dx_node_limit(dir));
1582 dx_set_count(entries, 1);
1583 dx_set_block(entries + 0, newblock);
1584 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1586 /* Add new access path frame */
1588 frame->at = at = at - entries + entries2;
1589 frame->entries = entries = entries2;
1591 err = ext3_journal_get_write_access(handle,
1596 ext3_journal_dirty_metadata(handle, frames[0].bh);
1598 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1601 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1606 ext3_std_error(dir->i_sb, err);
1615 * ext3_delete_entry deletes a directory entry by merging it with the
1618 static int ext3_delete_entry (handle_t *handle,
1620 struct ext3_dir_entry_2 * de_del,
1621 struct buffer_head * bh)
1623 struct ext3_dir_entry_2 * de, * pde;
1628 de = (struct ext3_dir_entry_2 *) bh->b_data;
1629 while (i < bh->b_size) {
1630 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1633 BUFFER_TRACE(bh, "get_write_access");
1634 ext3_journal_get_write_access(handle, bh);
1636 pde->rec_len = ext3_rec_len_to_disk(
1637 ext3_rec_len_from_disk(pde->rec_len) +
1638 ext3_rec_len_from_disk(de->rec_len));
1642 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1643 ext3_journal_dirty_metadata(handle, bh);
1646 i += ext3_rec_len_from_disk(de->rec_len);
1648 de = ext3_next_entry(de);
1653 static int ext3_add_nondir(handle_t *handle,
1654 struct dentry *dentry, struct inode *inode)
1656 int err = ext3_add_entry(handle, dentry, inode);
1658 ext3_mark_inode_dirty(handle, inode);
1659 d_instantiate(dentry, inode);
1668 * By the time this is called, we already have created
1669 * the directory cache entry for the new file, but it
1670 * is so far negative - it has no inode.
1672 * If the create succeeds, we fill in the inode information
1673 * with d_instantiate().
1675 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1676 struct nameidata *nd)
1679 struct inode * inode;
1680 int err, retries = 0;
1683 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1684 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1685 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1687 return PTR_ERR(handle);
1689 if (IS_DIRSYNC(dir))
1692 inode = ext3_new_inode (handle, dir, mode);
1693 err = PTR_ERR(inode);
1694 if (!IS_ERR(inode)) {
1695 inode->i_op = &ext3_file_inode_operations;
1696 inode->i_fop = &ext3_file_operations;
1697 ext3_set_aops(inode);
1698 err = ext3_add_nondir(handle, dentry, inode);
1700 ext3_journal_stop(handle);
1701 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1706 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1707 int mode, dev_t rdev)
1710 struct inode *inode;
1711 int err, retries = 0;
1713 if (!new_valid_dev(rdev))
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));
1721 return PTR_ERR(handle);
1723 if (IS_DIRSYNC(dir))
1726 inode = ext3_new_inode (handle, dir, mode);
1727 err = PTR_ERR(inode);
1728 if (!IS_ERR(inode)) {
1729 init_special_inode(inode, inode->i_mode, rdev);
1730 #ifdef CONFIG_EXT3_FS_XATTR
1731 inode->i_op = &ext3_special_inode_operations;
1733 err = ext3_add_nondir(handle, dentry, inode);
1735 ext3_journal_stop(handle);
1736 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1741 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1744 struct inode * inode;
1745 struct buffer_head * dir_block;
1746 struct ext3_dir_entry_2 * de;
1747 int err, retries = 0;
1749 if (dir->i_nlink >= EXT3_LINK_MAX)
1753 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1754 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1755 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1757 return PTR_ERR(handle);
1759 if (IS_DIRSYNC(dir))
1762 inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1763 err = PTR_ERR(inode);
1767 inode->i_op = &ext3_dir_inode_operations;
1768 inode->i_fop = &ext3_dir_operations;
1769 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1770 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1772 drop_nlink(inode); /* is this nlink == 0? */
1773 ext3_mark_inode_dirty(handle, inode);
1777 BUFFER_TRACE(dir_block, "get_write_access");
1778 ext3_journal_get_write_access(handle, dir_block);
1779 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1780 de->inode = cpu_to_le32(inode->i_ino);
1782 de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
1783 strcpy (de->name, ".");
1784 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1785 de = ext3_next_entry(de);
1786 de->inode = cpu_to_le32(dir->i_ino);
1787 de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1788 EXT3_DIR_REC_LEN(1));
1790 strcpy (de->name, "..");
1791 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1793 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1794 ext3_journal_dirty_metadata(handle, dir_block);
1796 ext3_mark_inode_dirty(handle, inode);
1797 err = ext3_add_entry (handle, dentry, inode);
1800 ext3_mark_inode_dirty(handle, inode);
1805 ext3_update_dx_flag(dir);
1806 ext3_mark_inode_dirty(handle, dir);
1807 d_instantiate(dentry, inode);
1809 ext3_journal_stop(handle);
1810 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1816 * routine to check that the specified directory is empty (for rmdir)
1818 static int empty_dir (struct inode * inode)
1820 unsigned long offset;
1821 struct buffer_head * bh;
1822 struct ext3_dir_entry_2 * de, * de1;
1823 struct super_block * sb;
1827 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1828 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1830 ext3_error(inode->i_sb, __func__,
1831 "error %d reading directory #%lu offset 0",
1834 ext3_warning(inode->i_sb, __func__,
1835 "bad directory (dir #%lu) - no data block",
1839 de = (struct ext3_dir_entry_2 *) bh->b_data;
1840 de1 = ext3_next_entry(de);
1841 if (le32_to_cpu(de->inode) != inode->i_ino ||
1842 !le32_to_cpu(de1->inode) ||
1843 strcmp (".", de->name) ||
1844 strcmp ("..", de1->name)) {
1845 ext3_warning (inode->i_sb, "empty_dir",
1846 "bad directory (dir #%lu) - no `.' or `..'",
1851 offset = ext3_rec_len_from_disk(de->rec_len) +
1852 ext3_rec_len_from_disk(de1->rec_len);
1853 de = ext3_next_entry(de1);
1854 while (offset < inode->i_size ) {
1856 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1859 bh = ext3_bread (NULL, inode,
1860 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1863 ext3_error(sb, __func__,
1864 "error %d reading directory"
1866 err, inode->i_ino, offset);
1867 offset += sb->s_blocksize;
1870 de = (struct ext3_dir_entry_2 *) bh->b_data;
1872 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1873 de = (struct ext3_dir_entry_2 *)(bh->b_data +
1875 offset = (offset | (sb->s_blocksize - 1)) + 1;
1878 if (le32_to_cpu(de->inode)) {
1882 offset += ext3_rec_len_from_disk(de->rec_len);
1883 de = ext3_next_entry(de);
1889 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1890 * such inodes, starting at the superblock, in case we crash before the
1891 * file is closed/deleted, or in case the inode truncate spans multiple
1892 * transactions and the last transaction is not recovered after a crash.
1894 * At filesystem recovery time, we walk this list deleting unlinked
1895 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1897 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1899 struct super_block *sb = inode->i_sb;
1900 struct ext3_iloc iloc;
1904 if (!list_empty(&EXT3_I(inode)->i_orphan))
1907 /* Orphan handling is only valid for files with data blocks
1908 * being truncated, or files being unlinked. */
1910 /* @@@ FIXME: Observation from aviro:
1911 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1912 * here (on lock_super()), so race with ext3_link() which might bump
1913 * ->i_nlink. For, say it, character device. Not a regular file,
1914 * not a directory, not a symlink and ->i_nlink > 0.
1916 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1917 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1919 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1920 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1924 err = ext3_reserve_inode_write(handle, inode, &iloc);
1928 /* Insert this inode at the head of the on-disk orphan list... */
1929 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1930 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1931 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1932 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1936 /* Only add to the head of the in-memory list if all the
1937 * previous operations succeeded. If the orphan_add is going to
1938 * fail (possibly taking the journal offline), we can't risk
1939 * leaving the inode on the orphan list: stray orphan-list
1940 * entries can cause panics at unmount time.
1942 * This is safe: on error we're going to ignore the orphan list
1943 * anyway on the next recovery. */
1945 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1947 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1948 jbd_debug(4, "orphan inode %lu will point to %d\n",
1949 inode->i_ino, NEXT_ORPHAN(inode));
1952 ext3_std_error(inode->i_sb, err);
1957 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1958 * of such inodes stored on disk, because it is finally being cleaned up.
1960 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1962 struct list_head *prev;
1963 struct ext3_inode_info *ei = EXT3_I(inode);
1964 struct ext3_sb_info *sbi;
1965 unsigned long ino_next;
1966 struct ext3_iloc iloc;
1969 lock_super(inode->i_sb);
1970 if (list_empty(&ei->i_orphan)) {
1971 unlock_super(inode->i_sb);
1975 ino_next = NEXT_ORPHAN(inode);
1976 prev = ei->i_orphan.prev;
1977 sbi = EXT3_SB(inode->i_sb);
1979 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1981 list_del_init(&ei->i_orphan);
1983 /* If we're on an error path, we may not have a valid
1984 * transaction handle with which to update the orphan list on
1985 * disk, but we still need to remove the inode from the linked
1986 * list in memory. */
1990 err = ext3_reserve_inode_write(handle, inode, &iloc);
1994 if (prev == &sbi->s_orphan) {
1995 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1996 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1997 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2000 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2001 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2003 struct ext3_iloc iloc2;
2004 struct inode *i_prev =
2005 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2007 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2008 i_prev->i_ino, ino_next);
2009 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2012 NEXT_ORPHAN(i_prev) = ino_next;
2013 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2017 NEXT_ORPHAN(inode) = 0;
2018 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2021 ext3_std_error(inode->i_sb, err);
2023 unlock_super(inode->i_sb);
2031 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2034 struct inode * inode;
2035 struct buffer_head * bh;
2036 struct ext3_dir_entry_2 * de;
2039 /* Initialize quotas before so that eventual writes go in
2040 * separate transaction */
2041 DQUOT_INIT(dentry->d_inode);
2042 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2044 return PTR_ERR(handle);
2047 bh = ext3_find_entry (dentry, &de);
2051 if (IS_DIRSYNC(dir))
2054 inode = dentry->d_inode;
2057 if (le32_to_cpu(de->inode) != inode->i_ino)
2060 retval = -ENOTEMPTY;
2061 if (!empty_dir (inode))
2064 retval = ext3_delete_entry(handle, dir, de, bh);
2067 if (inode->i_nlink != 2)
2068 ext3_warning (inode->i_sb, "ext3_rmdir",
2069 "empty directory has nlink!=2 (%d)",
2073 /* There's no need to set i_disksize: the fact that i_nlink is
2074 * zero will ensure that the right thing happens during any
2077 ext3_orphan_add(handle, inode);
2078 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2079 ext3_mark_inode_dirty(handle, inode);
2081 ext3_update_dx_flag(dir);
2082 ext3_mark_inode_dirty(handle, dir);
2085 ext3_journal_stop(handle);
2090 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2093 struct inode * inode;
2094 struct buffer_head * bh;
2095 struct ext3_dir_entry_2 * de;
2098 /* Initialize quotas before so that eventual writes go
2099 * in separate transaction */
2100 DQUOT_INIT(dentry->d_inode);
2101 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2103 return PTR_ERR(handle);
2105 if (IS_DIRSYNC(dir))
2109 bh = ext3_find_entry (dentry, &de);
2113 inode = dentry->d_inode;
2116 if (le32_to_cpu(de->inode) != inode->i_ino)
2119 if (!inode->i_nlink) {
2120 ext3_warning (inode->i_sb, "ext3_unlink",
2121 "Deleting nonexistent file (%lu), %d",
2122 inode->i_ino, inode->i_nlink);
2125 retval = ext3_delete_entry(handle, dir, de, bh);
2128 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2129 ext3_update_dx_flag(dir);
2130 ext3_mark_inode_dirty(handle, dir);
2132 if (!inode->i_nlink)
2133 ext3_orphan_add(handle, inode);
2134 inode->i_ctime = dir->i_ctime;
2135 ext3_mark_inode_dirty(handle, inode);
2139 ext3_journal_stop(handle);
2144 static int ext3_symlink (struct inode * dir,
2145 struct dentry *dentry, const char * symname)
2148 struct inode * inode;
2149 int l, err, retries = 0;
2151 l = strlen(symname)+1;
2152 if (l > dir->i_sb->s_blocksize)
2153 return -ENAMETOOLONG;
2156 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2157 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2158 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2160 return PTR_ERR(handle);
2162 if (IS_DIRSYNC(dir))
2165 inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2166 err = PTR_ERR(inode);
2170 if (l > sizeof (EXT3_I(inode)->i_data)) {
2171 inode->i_op = &ext3_symlink_inode_operations;
2172 ext3_set_aops(inode);
2174 * page_symlink() calls into ext3_prepare/commit_write.
2175 * We have a transaction open. All is sweetness. It also sets
2176 * i_size in generic_commit_write().
2178 err = __page_symlink(inode, symname, l,
2179 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2182 ext3_mark_inode_dirty(handle, inode);
2187 inode->i_op = &ext3_fast_symlink_inode_operations;
2188 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2189 inode->i_size = l-1;
2191 EXT3_I(inode)->i_disksize = inode->i_size;
2192 err = ext3_add_nondir(handle, dentry, inode);
2194 ext3_journal_stop(handle);
2195 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2200 static int ext3_link (struct dentry * old_dentry,
2201 struct inode * dir, struct dentry *dentry)
2204 struct inode *inode = old_dentry->d_inode;
2205 int err, retries = 0;
2207 if (inode->i_nlink >= EXT3_LINK_MAX)
2210 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2211 * otherwise has the potential to corrupt the orphan inode list.
2213 if (inode->i_nlink == 0)
2217 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2218 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2220 return PTR_ERR(handle);
2222 if (IS_DIRSYNC(dir))
2225 inode->i_ctime = CURRENT_TIME_SEC;
2227 atomic_inc(&inode->i_count);
2229 err = ext3_add_nondir(handle, dentry, inode);
2230 ext3_journal_stop(handle);
2231 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2236 #define PARENT_INO(buffer) \
2237 (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2240 * Anybody can rename anything with this: the permission checks are left to the
2241 * higher-level routines.
2243 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2244 struct inode * new_dir,struct dentry *new_dentry)
2247 struct inode * old_inode, * new_inode;
2248 struct buffer_head * old_bh, * new_bh, * dir_bh;
2249 struct ext3_dir_entry_2 * old_de, * new_de;
2252 old_bh = new_bh = dir_bh = NULL;
2254 /* Initialize quotas before so that eventual writes go
2255 * in separate transaction */
2256 if (new_dentry->d_inode)
2257 DQUOT_INIT(new_dentry->d_inode);
2258 handle = ext3_journal_start(old_dir, 2 *
2259 EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2260 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2262 return PTR_ERR(handle);
2264 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2267 old_bh = ext3_find_entry (old_dentry, &old_de);
2269 * Check for inode number is _not_ due to possible IO errors.
2270 * We might rmdir the source, keep it as pwd of some process
2271 * and merrily kill the link to whatever was created under the
2272 * same name. Goodbye sticky bit ;-<
2274 old_inode = old_dentry->d_inode;
2276 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2279 new_inode = new_dentry->d_inode;
2280 new_bh = ext3_find_entry (new_dentry, &new_de);
2287 if (S_ISDIR(old_inode->i_mode)) {
2289 retval = -ENOTEMPTY;
2290 if (!empty_dir (new_inode))
2294 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2297 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2300 if (!new_inode && new_dir!=old_dir &&
2301 new_dir->i_nlink >= EXT3_LINK_MAX)
2305 retval = ext3_add_entry (handle, new_dentry, old_inode);
2309 BUFFER_TRACE(new_bh, "get write access");
2310 ext3_journal_get_write_access(handle, new_bh);
2311 new_de->inode = cpu_to_le32(old_inode->i_ino);
2312 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2313 EXT3_FEATURE_INCOMPAT_FILETYPE))
2314 new_de->file_type = old_de->file_type;
2315 new_dir->i_version++;
2316 new_dir->i_ctime = new_dir->i_mtime = CURRENT_TIME_SEC;
2317 ext3_mark_inode_dirty(handle, new_dir);
2318 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2319 ext3_journal_dirty_metadata(handle, new_bh);
2325 * Like most other Unix systems, set the ctime for inodes on a
2328 old_inode->i_ctime = CURRENT_TIME_SEC;
2329 ext3_mark_inode_dirty(handle, old_inode);
2334 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2335 old_de->name_len != old_dentry->d_name.len ||
2336 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2337 (retval = ext3_delete_entry(handle, old_dir,
2338 old_de, old_bh)) == -ENOENT) {
2339 /* old_de could have moved from under us during htree split, so
2340 * make sure that we are deleting the right entry. We might
2341 * also be pointing to a stale entry in the unused part of
2342 * old_bh so just checking inum and the name isn't enough. */
2343 struct buffer_head *old_bh2;
2344 struct ext3_dir_entry_2 *old_de2;
2346 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2348 retval = ext3_delete_entry(handle, old_dir,
2354 ext3_warning(old_dir->i_sb, "ext3_rename",
2355 "Deleting old file (%lu), %d, error=%d",
2356 old_dir->i_ino, old_dir->i_nlink, retval);
2360 drop_nlink(new_inode);
2361 new_inode->i_ctime = CURRENT_TIME_SEC;
2363 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2364 ext3_update_dx_flag(old_dir);
2366 BUFFER_TRACE(dir_bh, "get_write_access");
2367 ext3_journal_get_write_access(handle, dir_bh);
2368 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2369 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2370 ext3_journal_dirty_metadata(handle, dir_bh);
2371 drop_nlink(old_dir);
2373 drop_nlink(new_inode);
2376 ext3_update_dx_flag(new_dir);
2377 ext3_mark_inode_dirty(handle, new_dir);
2380 ext3_mark_inode_dirty(handle, old_dir);
2382 ext3_mark_inode_dirty(handle, new_inode);
2383 if (!new_inode->i_nlink)
2384 ext3_orphan_add(handle, new_inode);
2392 ext3_journal_stop(handle);
2397 * directories can handle most operations...
2399 const struct inode_operations ext3_dir_inode_operations = {
2400 .create = ext3_create,
2401 .lookup = ext3_lookup,
2403 .unlink = ext3_unlink,
2404 .symlink = ext3_symlink,
2405 .mkdir = ext3_mkdir,
2406 .rmdir = ext3_rmdir,
2407 .mknod = ext3_mknod,
2408 .rename = ext3_rename,
2409 .setattr = ext3_setattr,
2410 #ifdef CONFIG_EXT3_FS_XATTR
2411 .setxattr = generic_setxattr,
2412 .getxattr = generic_getxattr,
2413 .listxattr = ext3_listxattr,
2414 .removexattr = generic_removexattr,
2416 .permission = ext3_permission,
2419 const struct inode_operations ext3_special_inode_operations = {
2420 .setattr = ext3_setattr,
2421 #ifdef CONFIG_EXT3_FS_XATTR
2422 .setxattr = generic_setxattr,
2423 .getxattr = generic_getxattr,
2424 .listxattr = ext3_listxattr,
2425 .removexattr = generic_removexattr,
2427 .permission = ext3_permission,