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/smp_lock.h>
43 * define how far ahead to read directories while searching them.
45 #define NAMEI_RA_CHUNKS 2
46 #define NAMEI_RA_BLOCKS 4
47 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
48 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
50 static struct buffer_head *ext3_append(handle_t *handle,
54 struct buffer_head *bh;
56 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
58 if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
59 inode->i_size += inode->i_sb->s_blocksize;
60 EXT3_I(inode)->i_disksize = inode->i_size;
61 ext3_journal_get_write_access(handle,bh);
67 #define assert(test) J_ASSERT(test)
71 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
74 typedef struct { u32 v; } le_u32;
75 typedef struct { u16 v; } le_u16;
78 #define dxtrace(command) command
80 #define dxtrace(command)
104 * dx_root_info is laid out so that if it should somehow get overlaid by a
105 * dirent the two low bits of the hash version will be zero. Therefore, the
106 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
111 struct fake_dirent dot;
113 struct fake_dirent dotdot;
117 le_u32 reserved_zero;
119 u8 info_length; /* 8 */
124 struct dx_entry entries[0];
129 struct fake_dirent fake;
130 struct dx_entry entries[0];
136 struct buffer_head *bh;
137 struct dx_entry *entries;
147 #ifdef CONFIG_EXT3_INDEX
148 static inline unsigned dx_get_block (struct dx_entry *entry);
149 static void dx_set_block (struct dx_entry *entry, unsigned value);
150 static inline unsigned dx_get_hash (struct dx_entry *entry);
151 static void dx_set_hash (struct dx_entry *entry, unsigned value);
152 static unsigned dx_get_count (struct dx_entry *entries);
153 static unsigned dx_get_limit (struct dx_entry *entries);
154 static void dx_set_count (struct dx_entry *entries, unsigned value);
155 static void dx_set_limit (struct dx_entry *entries, unsigned value);
156 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
157 static unsigned dx_node_limit (struct inode *dir);
158 static struct dx_frame *dx_probe(struct dentry *dentry,
160 struct dx_hash_info *hinfo,
161 struct dx_frame *frame,
163 static void dx_release (struct dx_frame *frames);
164 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
165 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
166 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
167 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
168 struct dx_map_entry *offsets, int count);
169 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
170 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
171 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
172 struct dx_frame *frame,
173 struct dx_frame *frames,
175 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
176 struct ext3_dir_entry_2 **res_dir, int *err);
177 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
178 struct inode *inode);
181 * Future: use high four bits of block for coalesce-on-delete flags
182 * Mask them off for now.
185 static inline unsigned dx_get_block (struct dx_entry *entry)
187 return le32_to_cpu(entry->block.v) & 0x00ffffff;
190 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
192 entry->block.v = cpu_to_le32(value);
195 static inline unsigned dx_get_hash (struct dx_entry *entry)
197 return le32_to_cpu(entry->hash.v);
200 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
202 entry->hash.v = cpu_to_le32(value);
205 static inline unsigned dx_get_count (struct dx_entry *entries)
207 return le16_to_cpu(((struct dx_countlimit *) entries)->count.v);
210 static inline unsigned dx_get_limit (struct dx_entry *entries)
212 return le16_to_cpu(((struct dx_countlimit *) entries)->limit.v);
215 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
217 ((struct dx_countlimit *) entries)->count.v = cpu_to_le16(value);
220 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
222 ((struct dx_countlimit *) entries)->limit.v = cpu_to_le16(value);
225 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
227 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
228 EXT3_DIR_REC_LEN(2) - infosize;
229 return 0? 20: entry_space / sizeof(struct dx_entry);
232 static inline unsigned dx_node_limit (struct inode *dir)
234 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
235 return 0? 22: entry_space / sizeof(struct dx_entry);
242 static void dx_show_index (char * label, struct dx_entry *entries)
244 int i, n = dx_get_count (entries);
245 printk("%s index ", label);
246 for (i = 0; i < n; i++)
248 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
260 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
261 int size, int show_names)
263 unsigned names = 0, space = 0;
264 char *base = (char *) de;
265 struct dx_hash_info h = *hinfo;
268 while ((char *) de < base + size)
274 int len = de->name_len;
275 char *name = de->name;
276 while (len--) printk("%c", *name++);
277 ext3fs_dirhash(de->name, de->name_len, &h);
278 printk(":%x.%u ", h.hash,
279 ((char *) de - base));
281 space += EXT3_DIR_REC_LEN(de->name_len);
284 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
286 printk("(%i)\n", names);
287 return (struct stats) { names, space, 1 };
290 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
291 struct dx_entry *entries, int levels)
293 unsigned blocksize = dir->i_sb->s_blocksize;
294 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
296 struct buffer_head *bh;
298 printk("%i indexed blocks...\n", count);
299 for (i = 0; i < count; i++, entries++)
301 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
302 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
304 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
305 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
307 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
308 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
309 names += stats.names;
310 space += stats.space;
311 bcount += stats.bcount;
315 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
316 names, space/bcount,(space/bcount)*100/blocksize);
317 return (struct stats) { names, space, bcount};
319 #endif /* DX_DEBUG */
322 * Probe for a directory leaf block to search.
324 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
325 * error in the directory index, and the caller should fall back to
326 * searching the directory normally. The callers of dx_probe **MUST**
327 * check for this error code, and make sure it never gets reflected
330 static struct dx_frame *
331 dx_probe(struct dentry *dentry, struct inode *dir,
332 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
334 unsigned count, indirect;
335 struct dx_entry *at, *entries, *p, *q, *m;
336 struct dx_root *root;
337 struct buffer_head *bh;
338 struct dx_frame *frame = frame_in;
343 dir = dentry->d_parent->d_inode;
344 if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
346 root = (struct dx_root *) bh->b_data;
347 if (root->info.hash_version != DX_HASH_TEA &&
348 root->info.hash_version != DX_HASH_HALF_MD4 &&
349 root->info.hash_version != DX_HASH_LEGACY) {
350 ext3_warning(dir->i_sb, __FUNCTION__,
351 "Unrecognised inode hash code %d",
352 root->info.hash_version);
354 *err = ERR_BAD_DX_DIR;
357 hinfo->hash_version = root->info.hash_version;
358 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
360 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
363 if (root->info.unused_flags & 1) {
364 ext3_warning(dir->i_sb, __FUNCTION__,
365 "Unimplemented inode hash flags: %#06x",
366 root->info.unused_flags);
368 *err = ERR_BAD_DX_DIR;
372 if ((indirect = root->info.indirect_levels) > 1) {
373 ext3_warning(dir->i_sb, __FUNCTION__,
374 "Unimplemented inode hash depth: %#06x",
375 root->info.indirect_levels);
377 *err = ERR_BAD_DX_DIR;
381 entries = (struct dx_entry *) (((char *)&root->info) +
382 root->info.info_length);
383 assert(dx_get_limit(entries) == dx_root_limit(dir,
384 root->info.info_length));
385 dxtrace (printk("Look up %x", hash));
388 count = dx_get_count(entries);
389 assert (count && count <= dx_get_limit(entries));
391 q = entries + count - 1;
395 dxtrace(printk("."));
396 if (dx_get_hash(m) > hash)
402 if (0) // linear search cross check
404 unsigned n = count - 1;
408 dxtrace(printk(","));
409 if (dx_get_hash(++at) > hash)
415 assert (at == p - 1);
419 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
421 frame->entries = entries;
423 if (!indirect--) return frame;
424 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
426 at = entries = ((struct dx_node *) bh->b_data)->entries;
427 assert (dx_get_limit(entries) == dx_node_limit (dir));
431 while (frame >= frame_in) {
439 static void dx_release (struct dx_frame *frames)
441 if (frames[0].bh == NULL)
444 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
445 brelse(frames[1].bh);
446 brelse(frames[0].bh);
450 * This function increments the frame pointer to search the next leaf
451 * block, and reads in the necessary intervening nodes if the search
452 * should be necessary. Whether or not the search is necessary is
453 * controlled by the hash parameter. If the hash value is even, then
454 * the search is only continued if the next block starts with that
455 * hash value. This is used if we are searching for a specific file.
457 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
459 * This function returns 1 if the caller should continue to search,
460 * or 0 if it should not. If there is an error reading one of the
461 * index blocks, it will a negative error code.
463 * If start_hash is non-null, it will be filled in with the starting
464 * hash of the next page.
466 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
467 struct dx_frame *frame,
468 struct dx_frame *frames,
472 struct buffer_head *bh;
473 int err, num_frames = 0;
478 * Find the next leaf page by incrementing the frame pointer.
479 * If we run out of entries in the interior node, loop around and
480 * increment pointer in the parent node. When we break out of
481 * this loop, num_frames indicates the number of interior
482 * nodes need to be read.
485 if (++(p->at) < p->entries + dx_get_count(p->entries))
494 * If the hash is 1, then continue only if the next page has a
495 * continuation hash of any value. This is used for readdir
496 * handling. Otherwise, check to see if the hash matches the
497 * desired contiuation hash. If it doesn't, return since
498 * there's no point to read in the successive index pages.
500 bhash = dx_get_hash(p->at);
503 if ((hash & 1) == 0) {
504 if ((bhash & ~1) != hash)
508 * If the hash is HASH_NB_ALWAYS, we always go to the next
509 * block so no check is necessary
511 while (num_frames--) {
512 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
514 return err; /* Failure */
518 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
525 * p is at least 6 bytes before the end of page
527 static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
529 return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
533 * This function fills a red-black tree with information from a
534 * directory block. It returns the number directory entries loaded
535 * into the tree. If there is an error it is returned in err.
537 static int htree_dirblock_to_tree(struct file *dir_file,
538 struct inode *dir, int block,
539 struct dx_hash_info *hinfo,
540 __u32 start_hash, __u32 start_minor_hash)
542 struct buffer_head *bh;
543 struct ext3_dir_entry_2 *de, *top;
546 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
547 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
550 de = (struct ext3_dir_entry_2 *) bh->b_data;
551 top = (struct ext3_dir_entry_2 *) ((char *) de +
552 dir->i_sb->s_blocksize -
553 EXT3_DIR_REC_LEN(0));
554 for (; de < top; de = ext3_next_entry(de)) {
555 ext3fs_dirhash(de->name, de->name_len, hinfo);
556 if ((hinfo->hash < start_hash) ||
557 ((hinfo->hash == start_hash) &&
558 (hinfo->minor_hash < start_minor_hash)))
560 if ((err = ext3_htree_store_dirent(dir_file,
561 hinfo->hash, hinfo->minor_hash, de)) != 0) {
573 * This function fills a red-black tree with information from a
574 * directory. We start scanning the directory in hash order, starting
575 * at start_hash and start_minor_hash.
577 * This function returns the number of entries inserted into the tree,
578 * or a negative error code.
580 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
581 __u32 start_minor_hash, __u32 *next_hash)
583 struct dx_hash_info hinfo;
584 struct ext3_dir_entry_2 *de;
585 struct dx_frame frames[2], *frame;
592 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
594 dir = dir_file->f_dentry->d_inode;
595 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
596 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
597 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
598 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
599 start_hash, start_minor_hash);
603 hinfo.hash = start_hash;
604 hinfo.minor_hash = 0;
605 frame = dx_probe(0, dir_file->f_dentry->d_inode, &hinfo, frames, &err);
609 /* Add '.' and '..' from the htree header */
610 if (!start_hash && !start_minor_hash) {
611 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
612 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
614 de = ext3_next_entry(de);
615 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
621 block = dx_get_block(frame->at);
622 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
623 start_hash, start_minor_hash);
630 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
631 frame, frames, &hashval);
632 *next_hash = hashval;
638 * Stop if: (a) there are no more entries, or
639 * (b) we have inserted at least one entry and the
640 * next hash value is not a continuation
643 (count && ((hashval & 1) == 0)))
647 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
657 * Directory block splitting, compacting
660 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
661 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
664 char *base = (char *) de;
665 struct dx_hash_info h = *hinfo;
667 while ((char *) de < base + size)
669 if (de->name_len && de->inode) {
670 ext3fs_dirhash(de->name, de->name_len, &h);
672 map_tail->hash = h.hash;
673 map_tail->offs = (u32) ((char *) de - base);
676 /* XXX: do we need to check rec_len == 0 case? -Chris */
677 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
682 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
684 struct dx_map_entry *p, *q, *top = map + count - 1;
686 /* Combsort until bubble sort doesn't suck */
690 if (count - 9 < 2) /* 9, 10 -> 11 */
692 for (p = top, q = p - count; q >= map; p--, q--)
693 if (p->hash < q->hash)
696 /* Garden variety bubble sort */
702 if (q[1].hash >= q[0].hash)
710 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
712 struct dx_entry *entries = frame->entries;
713 struct dx_entry *old = frame->at, *new = old + 1;
714 int count = dx_get_count(entries);
716 assert(count < dx_get_limit(entries));
717 assert(old < entries + count);
718 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
719 dx_set_hash(new, hash);
720 dx_set_block(new, block);
721 dx_set_count(entries, count + 1);
726 static void ext3_update_dx_flag(struct inode *inode)
728 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
729 EXT3_FEATURE_COMPAT_DIR_INDEX))
730 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
734 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
736 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
737 * `de != NULL' is guaranteed by caller.
739 static inline int ext3_match (int len, const char * const name,
740 struct ext3_dir_entry_2 * de)
742 if (len != de->name_len)
746 return !memcmp(name, de->name, len);
750 * Returns 0 if not found, -1 on failure, and 1 on success
752 static inline int search_dirblock(struct buffer_head * bh,
754 struct dentry *dentry,
755 unsigned long offset,
756 struct ext3_dir_entry_2 ** res_dir)
758 struct ext3_dir_entry_2 * de;
761 const char *name = dentry->d_name.name;
762 int namelen = dentry->d_name.len;
764 de = (struct ext3_dir_entry_2 *) bh->b_data;
765 dlimit = bh->b_data + dir->i_sb->s_blocksize;
766 while ((char *) de < dlimit) {
767 /* this code is executed quadratically often */
768 /* do minimal checking `by hand' */
770 if ((char *) de + namelen <= dlimit &&
771 ext3_match (namelen, name, de)) {
772 /* found a match - just to be sure, do a full check */
773 if (!ext3_check_dir_entry("ext3_find_entry",
774 dir, de, bh, offset))
779 /* prevent looping on a bad block */
780 de_len = le16_to_cpu(de->rec_len);
784 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
793 * finds an entry in the specified directory with the wanted name. It
794 * returns the cache buffer in which the entry was found, and the entry
795 * itself (as a parameter - res_dir). It does NOT read the inode of the
796 * entry - you'll have to do that yourself if you want to.
798 * The returned buffer_head has ->b_count elevated. The caller is expected
799 * to brelse() it when appropriate.
801 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
802 struct ext3_dir_entry_2 ** res_dir)
804 struct super_block * sb;
805 struct buffer_head * bh_use[NAMEI_RA_SIZE];
806 struct buffer_head * bh, *ret = NULL;
807 unsigned long start, block, b;
808 int ra_max = 0; /* Number of bh's in the readahead
810 int ra_ptr = 0; /* Current index into readahead
814 struct inode *dir = dentry->d_parent->d_inode;
821 blocksize = sb->s_blocksize;
822 namelen = dentry->d_name.len;
823 name = dentry->d_name.name;
824 if (namelen > EXT3_NAME_LEN)
826 #ifdef CONFIG_EXT3_INDEX
828 bh = ext3_dx_find_entry(dentry, res_dir, &err);
830 * On success, or if the error was file not found,
831 * return. Otherwise, fall back to doing a search the
834 if (bh || (err != ERR_BAD_DX_DIR))
836 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
839 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
840 start = EXT3_I(dir)->i_dir_start_lookup;
841 if (start >= nblocks)
847 * We deal with the read-ahead logic here.
849 if (ra_ptr >= ra_max) {
850 /* Refill the readahead buffer */
853 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
855 * Terminate if we reach the end of the
856 * directory and must wrap, or if our
857 * search has finished at this block.
859 if (b >= nblocks || (num && block == start)) {
860 bh_use[ra_max] = NULL;
864 bh = ext3_getblk(NULL, dir, b++, 0, &err);
867 ll_rw_block(READ, 1, &bh);
870 if ((bh = bh_use[ra_ptr++]) == NULL)
873 if (!buffer_uptodate(bh)) {
874 /* read error, skip block & hope for the best */
878 i = search_dirblock(bh, dir, dentry,
879 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
881 EXT3_I(dir)->i_dir_start_lookup = block;
883 goto cleanup_and_exit;
887 goto cleanup_and_exit;
890 if (++block >= nblocks)
892 } while (block != start);
895 * If the directory has grown while we were searching, then
896 * search the last part of the directory before giving up.
899 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
900 if (block < nblocks) {
906 /* Clean up the read-ahead blocks */
907 for (; ra_ptr < ra_max; ra_ptr++)
908 brelse (bh_use[ra_ptr]);
912 #ifdef CONFIG_EXT3_INDEX
913 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
914 struct ext3_dir_entry_2 **res_dir, int *err)
916 struct super_block * sb;
917 struct dx_hash_info hinfo;
919 struct dx_frame frames[2], *frame;
920 struct ext3_dir_entry_2 *de, *top;
921 struct buffer_head *bh;
924 int namelen = dentry->d_name.len;
925 const u8 *name = dentry->d_name.name;
926 struct inode *dir = dentry->d_parent->d_inode;
929 if (!(frame = dx_probe (dentry, 0, &hinfo, frames, err)))
933 block = dx_get_block(frame->at);
934 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
936 de = (struct ext3_dir_entry_2 *) bh->b_data;
937 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
938 EXT3_DIR_REC_LEN(0));
939 for (; de < top; de = ext3_next_entry(de))
940 if (ext3_match (namelen, name, de)) {
941 if (!ext3_check_dir_entry("ext3_find_entry",
943 (block<<EXT3_BLOCK_SIZE_BITS(sb))
944 +((char *)de - bh->b_data))) {
953 /* Check to see if we should continue to search */
954 retval = ext3_htree_next_block(dir, hash, frame,
957 ext3_warning(sb, __FUNCTION__,
958 "error reading index page in directory #%lu",
963 } while (retval == 1);
967 dxtrace(printk("%s not found\n", name));
973 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
975 struct inode * inode;
976 struct ext3_dir_entry_2 * de;
977 struct buffer_head * bh;
979 if (dentry->d_name.len > EXT3_NAME_LEN)
980 return ERR_PTR(-ENAMETOOLONG);
982 bh = ext3_find_entry(dentry, &de);
985 unsigned long ino = le32_to_cpu(de->inode);
987 inode = iget(dir->i_sb, ino);
990 return ERR_PTR(-EACCES);
993 return d_splice_alias(inode, dentry);
994 d_add(dentry, inode);
999 struct dentry *ext3_get_parent(struct dentry *child)
1002 struct dentry *parent;
1003 struct inode *inode;
1004 struct dentry dotdot;
1005 struct ext3_dir_entry_2 * de;
1006 struct buffer_head *bh;
1008 dotdot.d_name.name = "..";
1009 dotdot.d_name.len = 2;
1010 dotdot.d_parent = child; /* confusing, isn't it! */
1012 bh = ext3_find_entry(&dotdot, &de);
1015 return ERR_PTR(-ENOENT);
1016 ino = le32_to_cpu(de->inode);
1018 inode = iget(child->d_inode->i_sb, ino);
1021 return ERR_PTR(-EACCES);
1023 parent = d_alloc_anon(inode);
1026 parent = ERR_PTR(-ENOMEM);
1032 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1033 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1034 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1035 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1036 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1037 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1038 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1039 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1042 static inline void ext3_set_de_type(struct super_block *sb,
1043 struct ext3_dir_entry_2 *de,
1045 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1046 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1049 #ifdef CONFIG_EXT3_INDEX
1050 static struct ext3_dir_entry_2 *
1051 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1053 unsigned rec_len = 0;
1056 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1057 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1058 memcpy (to, de, rec_len);
1059 ((struct ext3_dir_entry_2 *) to)->rec_len =
1060 cpu_to_le16(rec_len);
1065 return (struct ext3_dir_entry_2 *) (to - rec_len);
1068 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1070 struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1071 unsigned rec_len = 0;
1074 while ((char*)de < base + size) {
1075 next = (struct ext3_dir_entry_2 *) ((char *) de +
1076 le16_to_cpu(de->rec_len));
1077 if (de->inode && de->name_len) {
1078 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1080 memmove(to, de, rec_len);
1081 to->rec_len = cpu_to_le16(rec_len);
1083 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1090 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1091 struct buffer_head **bh,struct dx_frame *frame,
1092 struct dx_hash_info *hinfo, int *error)
1094 unsigned blocksize = dir->i_sb->s_blocksize;
1095 unsigned count, continued;
1096 struct buffer_head *bh2;
1099 struct dx_map_entry *map;
1100 char *data1 = (*bh)->b_data, *data2;
1102 struct ext3_dir_entry_2 *de = NULL, *de2;
1105 bh2 = ext3_append (handle, dir, &newblock, error);
1112 BUFFER_TRACE(*bh, "get_write_access");
1113 err = ext3_journal_get_write_access(handle, *bh);
1119 ext3_std_error(dir->i_sb, err);
1122 BUFFER_TRACE(frame->bh, "get_write_access");
1123 err = ext3_journal_get_write_access(handle, frame->bh);
1127 data2 = bh2->b_data;
1129 /* create map in the end of data2 block */
1130 map = (struct dx_map_entry *) (data2 + blocksize);
1131 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1132 blocksize, hinfo, map);
1134 split = count/2; // need to adjust to actual middle
1135 dx_sort_map (map, count);
1136 hash2 = map[split].hash;
1137 continued = hash2 == map[split - 1].hash;
1138 dxtrace(printk("Split block %i at %x, %i/%i\n",
1139 dx_get_block(frame->at), hash2, split, count-split));
1141 /* Fancy dance to stay within two buffers */
1142 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1143 de = dx_pack_dirents(data1,blocksize);
1144 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1145 de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1146 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1147 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1149 /* Which block gets the new entry? */
1150 if (hinfo->hash >= hash2)
1155 dx_insert_block (frame, hash2 + continued, newblock);
1156 err = ext3_journal_dirty_metadata (handle, bh2);
1159 err = ext3_journal_dirty_metadata (handle, frame->bh);
1163 dxtrace(dx_show_index ("frame", frame->entries));
1171 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1172 * it points to a directory entry which is guaranteed to be large
1173 * enough for new directory entry. If de is NULL, then
1174 * add_dirent_to_buf will attempt search the directory block for
1175 * space. It will return -ENOSPC if no space is available, and -EIO
1176 * and -EEXIST if directory entry already exists.
1178 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1179 * all other cases bh is released.
1181 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1182 struct inode *inode, struct ext3_dir_entry_2 *de,
1183 struct buffer_head * bh)
1185 struct inode *dir = dentry->d_parent->d_inode;
1186 const char *name = dentry->d_name.name;
1187 int namelen = dentry->d_name.len;
1188 unsigned long offset = 0;
1189 unsigned short reclen;
1190 int nlen, rlen, err;
1193 reclen = EXT3_DIR_REC_LEN(namelen);
1195 de = (struct ext3_dir_entry_2 *)bh->b_data;
1196 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1197 while ((char *) de <= top) {
1198 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1203 if (ext3_match (namelen, name, de)) {
1207 nlen = EXT3_DIR_REC_LEN(de->name_len);
1208 rlen = le16_to_cpu(de->rec_len);
1209 if ((de->inode? rlen - nlen: rlen) >= reclen)
1211 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1214 if ((char *) de > top)
1217 BUFFER_TRACE(bh, "get_write_access");
1218 err = ext3_journal_get_write_access(handle, bh);
1220 ext3_std_error(dir->i_sb, err);
1225 /* By now the buffer is marked for journaling */
1226 nlen = EXT3_DIR_REC_LEN(de->name_len);
1227 rlen = le16_to_cpu(de->rec_len);
1229 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1230 de1->rec_len = cpu_to_le16(rlen - nlen);
1231 de->rec_len = cpu_to_le16(nlen);
1234 de->file_type = EXT3_FT_UNKNOWN;
1236 de->inode = cpu_to_le32(inode->i_ino);
1237 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1240 de->name_len = namelen;
1241 memcpy (de->name, name, namelen);
1243 * XXX shouldn't update any times until successful
1244 * completion of syscall, but too many callers depend
1247 * XXX similarly, too many callers depend on
1248 * ext3_new_inode() setting the times, but error
1249 * recovery deletes the inode, so the worst that can
1250 * happen is that the times are slightly out of date
1251 * and/or different from the directory change time.
1253 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1254 ext3_update_dx_flag(dir);
1256 ext3_mark_inode_dirty(handle, dir);
1257 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1258 err = ext3_journal_dirty_metadata(handle, bh);
1260 ext3_std_error(dir->i_sb, err);
1265 #ifdef CONFIG_EXT3_INDEX
1267 * This converts a one block unindexed directory to a 3 block indexed
1268 * directory, and adds the dentry to the indexed directory.
1270 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1271 struct inode *inode, struct buffer_head *bh)
1273 struct inode *dir = dentry->d_parent->d_inode;
1274 const char *name = dentry->d_name.name;
1275 int namelen = dentry->d_name.len;
1276 struct buffer_head *bh2;
1277 struct dx_root *root;
1278 struct dx_frame frames[2], *frame;
1279 struct dx_entry *entries;
1280 struct ext3_dir_entry_2 *de, *de2;
1285 struct dx_hash_info hinfo;
1287 struct fake_dirent *fde;
1289 blocksize = dir->i_sb->s_blocksize;
1290 dxtrace(printk("Creating index\n"));
1291 retval = ext3_journal_get_write_access(handle, bh);
1293 ext3_std_error(dir->i_sb, retval);
1297 root = (struct dx_root *) bh->b_data;
1299 bh2 = ext3_append (handle, dir, &block, &retval);
1304 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1305 data1 = bh2->b_data;
1307 /* The 0th block becomes the root, move the dirents out */
1308 fde = &root->dotdot;
1309 de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1310 len = ((char *) root) + blocksize - (char *) de;
1311 memcpy (data1, de, len);
1312 de = (struct ext3_dir_entry_2 *) data1;
1314 while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1316 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1317 /* Initialize the root; the dot dirents already exist */
1318 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1319 de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1320 memset (&root->info, 0, sizeof(root->info));
1321 root->info.info_length = sizeof(root->info);
1322 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1323 entries = root->entries;
1324 dx_set_block (entries, 1);
1325 dx_set_count (entries, 1);
1326 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1328 /* Initialize as for dx_probe */
1329 hinfo.hash_version = root->info.hash_version;
1330 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1331 ext3fs_dirhash(name, namelen, &hinfo);
1333 frame->entries = entries;
1334 frame->at = entries;
1337 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1338 dx_release (frames);
1342 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1349 * adds a file entry to the specified directory, using the same
1350 * semantics as ext3_find_entry(). It returns NULL if it failed.
1352 * NOTE!! The inode part of 'de' is left at 0 - which means you
1353 * may not sleep between calling this and putting something into
1354 * the entry, as someone else might have used it while you slept.
1356 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1357 struct inode *inode)
1359 struct inode *dir = dentry->d_parent->d_inode;
1360 unsigned long offset;
1361 struct buffer_head * bh;
1362 struct ext3_dir_entry_2 *de;
1363 struct super_block * sb;
1365 #ifdef CONFIG_EXT3_INDEX
1369 unsigned nlen, rlen;
1373 blocksize = sb->s_blocksize;
1374 if (!dentry->d_name.len)
1376 #ifdef CONFIG_EXT3_INDEX
1378 retval = ext3_dx_add_entry(handle, dentry, inode);
1379 if (!retval || (retval != ERR_BAD_DX_DIR))
1381 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1383 ext3_mark_inode_dirty(handle, dir);
1386 blocks = dir->i_size >> sb->s_blocksize_bits;
1387 for (block = 0, offset = 0; block < blocks; block++) {
1388 bh = ext3_bread(handle, dir, block, 0, &retval);
1391 retval = add_dirent_to_buf(handle, dentry, inode, 0, bh);
1392 if (retval != -ENOSPC)
1395 #ifdef CONFIG_EXT3_INDEX
1396 if (blocks == 1 && !dx_fallback &&
1397 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1398 return make_indexed_dir(handle, dentry, inode, bh);
1402 bh = ext3_append(handle, dir, &block, &retval);
1405 de = (struct ext3_dir_entry_2 *) bh->b_data;
1407 de->rec_len = cpu_to_le16(rlen = blocksize);
1409 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1412 #ifdef CONFIG_EXT3_INDEX
1414 * Returns 0 for success, or a negative error value
1416 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1417 struct inode *inode)
1419 struct dx_frame frames[2], *frame;
1420 struct dx_entry *entries, *at;
1421 struct dx_hash_info hinfo;
1422 struct buffer_head * bh;
1423 struct inode *dir = dentry->d_parent->d_inode;
1424 struct super_block * sb = dir->i_sb;
1425 struct ext3_dir_entry_2 *de;
1428 frame = dx_probe(dentry, 0, &hinfo, frames, &err);
1431 entries = frame->entries;
1434 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1437 BUFFER_TRACE(bh, "get_write_access");
1438 err = ext3_journal_get_write_access(handle, bh);
1442 err = add_dirent_to_buf(handle, dentry, inode, 0, bh);
1443 if (err != -ENOSPC) {
1448 /* Block full, should compress but for now just split */
1449 dxtrace(printk("using %u of %u node entries\n",
1450 dx_get_count(entries), dx_get_limit(entries)));
1451 /* Need to split index? */
1452 if (dx_get_count(entries) == dx_get_limit(entries)) {
1454 unsigned icount = dx_get_count(entries);
1455 int levels = frame - frames;
1456 struct dx_entry *entries2;
1457 struct dx_node *node2;
1458 struct buffer_head *bh2;
1460 if (levels && (dx_get_count(frames->entries) ==
1461 dx_get_limit(frames->entries))) {
1462 ext3_warning(sb, __FUNCTION__,
1463 "Directory index full!\n");
1467 bh2 = ext3_append (handle, dir, &newblock, &err);
1470 node2 = (struct dx_node *)(bh2->b_data);
1471 entries2 = node2->entries;
1472 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1473 node2->fake.inode = 0;
1474 BUFFER_TRACE(frame->bh, "get_write_access");
1475 err = ext3_journal_get_write_access(handle, frame->bh);
1479 unsigned icount1 = icount/2, icount2 = icount - icount1;
1480 unsigned hash2 = dx_get_hash(entries + icount1);
1481 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1483 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1484 err = ext3_journal_get_write_access(handle,
1489 memcpy ((char *) entries2, (char *) (entries + icount1),
1490 icount2 * sizeof(struct dx_entry));
1491 dx_set_count (entries, icount1);
1492 dx_set_count (entries2, icount2);
1493 dx_set_limit (entries2, dx_node_limit(dir));
1495 /* Which index block gets the new entry? */
1496 if (at - entries >= icount1) {
1497 frame->at = at = at - entries - icount1 + entries2;
1498 frame->entries = entries = entries2;
1499 swap(frame->bh, bh2);
1501 dx_insert_block (frames + 0, hash2, newblock);
1502 dxtrace(dx_show_index ("node", frames[1].entries));
1503 dxtrace(dx_show_index ("node",
1504 ((struct dx_node *) bh2->b_data)->entries));
1505 err = ext3_journal_dirty_metadata(handle, bh2);
1510 dxtrace(printk("Creating second level index...\n"));
1511 memcpy((char *) entries2, (char *) entries,
1512 icount * sizeof(struct dx_entry));
1513 dx_set_limit(entries2, dx_node_limit(dir));
1516 dx_set_count(entries, 1);
1517 dx_set_block(entries + 0, newblock);
1518 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1520 /* Add new access path frame */
1522 frame->at = at = at - entries + entries2;
1523 frame->entries = entries = entries2;
1525 err = ext3_journal_get_write_access(handle,
1530 ext3_journal_dirty_metadata(handle, frames[0].bh);
1532 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1535 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1540 ext3_std_error(dir->i_sb, err);
1550 * ext3_delete_entry deletes a directory entry by merging it with the
1553 static int ext3_delete_entry (handle_t *handle,
1555 struct ext3_dir_entry_2 * de_del,
1556 struct buffer_head * bh)
1558 struct ext3_dir_entry_2 * de, * pde;
1563 de = (struct ext3_dir_entry_2 *) bh->b_data;
1564 while (i < bh->b_size) {
1565 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1568 BUFFER_TRACE(bh, "get_write_access");
1569 ext3_journal_get_write_access(handle, bh);
1572 cpu_to_le16(le16_to_cpu(pde->rec_len) +
1573 le16_to_cpu(de->rec_len));
1577 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1578 ext3_journal_dirty_metadata(handle, bh);
1581 i += le16_to_cpu(de->rec_len);
1583 de = (struct ext3_dir_entry_2 *)
1584 ((char *) de + le16_to_cpu(de->rec_len));
1590 * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1591 * do not perform it in these functions. We perform it at the call site,
1594 static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1599 static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1604 static int ext3_add_nondir(handle_t *handle,
1605 struct dentry *dentry, struct inode *inode)
1607 int err = ext3_add_entry(handle, dentry, inode);
1609 ext3_mark_inode_dirty(handle, inode);
1610 d_instantiate(dentry, inode);
1613 ext3_dec_count(handle, inode);
1619 * By the time this is called, we already have created
1620 * the directory cache entry for the new file, but it
1621 * is so far negative - it has no inode.
1623 * If the create succeeds, we fill in the inode information
1624 * with d_instantiate().
1626 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1627 struct nameidata *nd)
1630 struct inode * inode;
1633 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1634 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1635 2*EXT3_QUOTA_INIT_BLOCKS);
1637 return PTR_ERR(handle);
1639 if (IS_DIRSYNC(dir))
1642 inode = ext3_new_inode (handle, dir, mode);
1643 err = PTR_ERR(inode);
1644 if (!IS_ERR(inode)) {
1645 inode->i_op = &ext3_file_inode_operations;
1646 inode->i_fop = &ext3_file_operations;
1647 ext3_set_aops(inode);
1648 err = ext3_add_nondir(handle, dentry, inode);
1650 ext3_journal_stop(handle);
1654 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1655 int mode, dev_t rdev)
1658 struct inode *inode;
1661 if (!new_valid_dev(rdev))
1664 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1665 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1666 2*EXT3_QUOTA_INIT_BLOCKS);
1668 return PTR_ERR(handle);
1670 if (IS_DIRSYNC(dir))
1673 inode = ext3_new_inode (handle, dir, mode);
1674 err = PTR_ERR(inode);
1675 if (!IS_ERR(inode)) {
1676 init_special_inode(inode, inode->i_mode, rdev);
1677 #ifdef CONFIG_EXT3_FS_XATTR
1678 inode->i_op = &ext3_special_inode_operations;
1680 err = ext3_add_nondir(handle, dentry, inode);
1682 ext3_journal_stop(handle);
1686 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1689 struct inode * inode;
1690 struct buffer_head * dir_block;
1691 struct ext3_dir_entry_2 * de;
1694 if (dir->i_nlink >= EXT3_LINK_MAX)
1697 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1698 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1699 2*EXT3_QUOTA_INIT_BLOCKS);
1701 return PTR_ERR(handle);
1703 if (IS_DIRSYNC(dir))
1706 inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1707 err = PTR_ERR(inode);
1711 inode->i_op = &ext3_dir_inode_operations;
1712 inode->i_fop = &ext3_dir_operations;
1713 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1714 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1716 inode->i_nlink--; /* is this nlink == 0? */
1717 ext3_mark_inode_dirty(handle, inode);
1721 BUFFER_TRACE(dir_block, "get_write_access");
1722 ext3_journal_get_write_access(handle, dir_block);
1723 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1724 de->inode = cpu_to_le32(inode->i_ino);
1726 de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1727 strcpy (de->name, ".");
1728 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1729 de = (struct ext3_dir_entry_2 *)
1730 ((char *) de + le16_to_cpu(de->rec_len));
1731 de->inode = cpu_to_le32(dir->i_ino);
1732 de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1734 strcpy (de->name, "..");
1735 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1737 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1738 ext3_journal_dirty_metadata(handle, dir_block);
1740 ext3_mark_inode_dirty(handle, inode);
1741 err = ext3_add_entry (handle, dentry, inode);
1744 ext3_mark_inode_dirty(handle, inode);
1749 ext3_update_dx_flag(dir);
1750 ext3_mark_inode_dirty(handle, dir);
1751 d_instantiate(dentry, inode);
1753 ext3_journal_stop(handle);
1758 * routine to check that the specified directory is empty (for rmdir)
1760 static int empty_dir (struct inode * inode)
1762 unsigned long offset;
1763 struct buffer_head * bh;
1764 struct ext3_dir_entry_2 * de, * de1;
1765 struct super_block * sb;
1769 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1770 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1771 ext3_warning (inode->i_sb, "empty_dir",
1772 "bad directory (dir #%lu) - no data block",
1776 de = (struct ext3_dir_entry_2 *) bh->b_data;
1777 de1 = (struct ext3_dir_entry_2 *)
1778 ((char *) de + le16_to_cpu(de->rec_len));
1779 if (le32_to_cpu(de->inode) != inode->i_ino ||
1780 !le32_to_cpu(de1->inode) ||
1781 strcmp (".", de->name) ||
1782 strcmp ("..", de1->name)) {
1783 ext3_warning (inode->i_sb, "empty_dir",
1784 "bad directory (dir #%lu) - no `.' or `..'",
1789 offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1790 de = (struct ext3_dir_entry_2 *)
1791 ((char *) de1 + le16_to_cpu(de1->rec_len));
1792 while (offset < inode->i_size ) {
1794 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1796 bh = ext3_bread (NULL, inode,
1797 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1800 ext3_error (sb, "empty_dir",
1801 "directory #%lu contains a hole at offset %lu",
1802 inode->i_ino, offset);
1804 offset += sb->s_blocksize;
1807 de = (struct ext3_dir_entry_2 *) bh->b_data;
1809 if (!ext3_check_dir_entry ("empty_dir", inode, de, bh,
1814 if (le32_to_cpu(de->inode)) {
1818 offset += le16_to_cpu(de->rec_len);
1819 de = (struct ext3_dir_entry_2 *)
1820 ((char *) de + le16_to_cpu(de->rec_len));
1826 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1827 * such inodes, starting at the superblock, in case we crash before the
1828 * file is closed/deleted, or in case the inode truncate spans multiple
1829 * transactions and the last transaction is not recovered after a crash.
1831 * At filesystem recovery time, we walk this list deleting unlinked
1832 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1834 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1836 struct super_block *sb = inode->i_sb;
1837 struct ext3_iloc iloc;
1841 if (!list_empty(&EXT3_I(inode)->i_orphan))
1844 /* Orphan handling is only valid for files with data blocks
1845 * being truncated, or files being unlinked. */
1847 /* @@@ FIXME: Observation from aviro:
1848 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1849 * here (on lock_super()), so race with ext3_link() which might bump
1850 * ->i_nlink. For, say it, character device. Not a regular file,
1851 * not a directory, not a symlink and ->i_nlink > 0.
1853 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1854 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1856 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1857 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1861 err = ext3_reserve_inode_write(handle, inode, &iloc);
1865 /* Insert this inode at the head of the on-disk orphan list... */
1866 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1867 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1868 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1869 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1873 /* Only add to the head of the in-memory list if all the
1874 * previous operations succeeded. If the orphan_add is going to
1875 * fail (possibly taking the journal offline), we can't risk
1876 * leaving the inode on the orphan list: stray orphan-list
1877 * entries can cause panics at unmount time.
1879 * This is safe: on error we're going to ignore the orphan list
1880 * anyway on the next recovery. */
1882 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1884 jbd_debug(4, "superblock will point to %ld\n", inode->i_ino);
1885 jbd_debug(4, "orphan inode %ld will point to %d\n",
1886 inode->i_ino, NEXT_ORPHAN(inode));
1889 ext3_std_error(inode->i_sb, err);
1894 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1895 * of such inodes stored on disk, because it is finally being cleaned up.
1897 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1899 struct list_head *prev;
1900 struct ext3_inode_info *ei = EXT3_I(inode);
1901 struct ext3_sb_info *sbi;
1902 unsigned long ino_next;
1903 struct ext3_iloc iloc;
1906 lock_super(inode->i_sb);
1907 if (list_empty(&ei->i_orphan)) {
1908 unlock_super(inode->i_sb);
1912 ino_next = NEXT_ORPHAN(inode);
1913 prev = ei->i_orphan.prev;
1914 sbi = EXT3_SB(inode->i_sb);
1916 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1918 list_del_init(&ei->i_orphan);
1920 /* If we're on an error path, we may not have a valid
1921 * transaction handle with which to update the orphan list on
1922 * disk, but we still need to remove the inode from the linked
1923 * list in memory. */
1927 err = ext3_reserve_inode_write(handle, inode, &iloc);
1931 if (prev == &sbi->s_orphan) {
1932 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1933 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1934 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1937 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1938 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1940 struct ext3_iloc iloc2;
1941 struct inode *i_prev =
1942 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1944 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1945 i_prev->i_ino, ino_next);
1946 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1949 NEXT_ORPHAN(i_prev) = ino_next;
1950 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
1954 NEXT_ORPHAN(inode) = 0;
1955 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
1960 ext3_std_error(inode->i_sb, err);
1962 unlock_super(inode->i_sb);
1970 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
1973 struct inode * inode;
1974 struct buffer_head * bh;
1975 struct ext3_dir_entry_2 * de;
1978 /* Initialize quotas before so that eventual writes go in
1979 * separate transaction */
1980 DQUOT_INIT(dentry->d_inode);
1981 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
1983 return PTR_ERR(handle);
1986 bh = ext3_find_entry (dentry, &de);
1990 if (IS_DIRSYNC(dir))
1993 inode = dentry->d_inode;
1996 if (le32_to_cpu(de->inode) != inode->i_ino)
1999 retval = -ENOTEMPTY;
2000 if (!empty_dir (inode))
2003 retval = ext3_delete_entry(handle, dir, de, bh);
2006 if (inode->i_nlink != 2)
2007 ext3_warning (inode->i_sb, "ext3_rmdir",
2008 "empty directory has nlink!=2 (%d)",
2012 /* There's no need to set i_disksize: the fact that i_nlink is
2013 * zero will ensure that the right thing happens during any
2016 ext3_orphan_add(handle, inode);
2017 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2018 ext3_mark_inode_dirty(handle, inode);
2020 ext3_update_dx_flag(dir);
2021 ext3_mark_inode_dirty(handle, dir);
2024 ext3_journal_stop(handle);
2029 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2032 struct inode * inode;
2033 struct buffer_head * bh;
2034 struct ext3_dir_entry_2 * de;
2037 /* Initialize quotas before so that eventual writes go
2038 * in separate transaction */
2039 DQUOT_INIT(dentry->d_inode);
2040 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
2042 return PTR_ERR(handle);
2044 if (IS_DIRSYNC(dir))
2048 bh = ext3_find_entry (dentry, &de);
2052 inode = dentry->d_inode;
2055 if (le32_to_cpu(de->inode) != inode->i_ino)
2058 if (!inode->i_nlink) {
2059 ext3_warning (inode->i_sb, "ext3_unlink",
2060 "Deleting nonexistent file (%lu), %d",
2061 inode->i_ino, inode->i_nlink);
2064 retval = ext3_delete_entry(handle, dir, de, bh);
2067 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2068 ext3_update_dx_flag(dir);
2069 ext3_mark_inode_dirty(handle, dir);
2071 if (!inode->i_nlink)
2072 ext3_orphan_add(handle, inode);
2073 inode->i_ctime = dir->i_ctime;
2074 ext3_mark_inode_dirty(handle, inode);
2078 ext3_journal_stop(handle);
2083 static int ext3_symlink (struct inode * dir,
2084 struct dentry *dentry, const char * symname)
2087 struct inode * inode;
2090 l = strlen(symname)+1;
2091 if (l > dir->i_sb->s_blocksize)
2092 return -ENAMETOOLONG;
2094 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2095 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2096 2*EXT3_QUOTA_INIT_BLOCKS);
2098 return PTR_ERR(handle);
2100 if (IS_DIRSYNC(dir))
2103 inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2104 err = PTR_ERR(inode);
2108 if (l > sizeof (EXT3_I(inode)->i_data)) {
2109 inode->i_op = &ext3_symlink_inode_operations;
2110 ext3_set_aops(inode);
2112 * page_symlink() calls into ext3_prepare/commit_write.
2113 * We have a transaction open. All is sweetness. It also sets
2114 * i_size in generic_commit_write().
2116 err = page_symlink(inode, symname, l);
2118 ext3_dec_count(handle, inode);
2119 ext3_mark_inode_dirty(handle, inode);
2124 inode->i_op = &ext3_fast_symlink_inode_operations;
2125 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2126 inode->i_size = l-1;
2128 EXT3_I(inode)->i_disksize = inode->i_size;
2129 err = ext3_add_nondir(handle, dentry, inode);
2131 ext3_journal_stop(handle);
2135 static int ext3_link (struct dentry * old_dentry,
2136 struct inode * dir, struct dentry *dentry)
2139 struct inode *inode = old_dentry->d_inode;
2142 if (inode->i_nlink >= EXT3_LINK_MAX)
2145 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2146 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2148 return PTR_ERR(handle);
2150 if (IS_DIRSYNC(dir))
2153 inode->i_ctime = CURRENT_TIME;
2154 ext3_inc_count(handle, inode);
2155 atomic_inc(&inode->i_count);
2157 err = ext3_add_nondir(handle, dentry, inode);
2158 ext3_journal_stop(handle);
2162 #define PARENT_INO(buffer) \
2163 ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2164 le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2167 * Anybody can rename anything with this: the permission checks are left to the
2168 * higher-level routines.
2170 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2171 struct inode * new_dir,struct dentry *new_dentry)
2174 struct inode * old_inode, * new_inode;
2175 struct buffer_head * old_bh, * new_bh, * dir_bh;
2176 struct ext3_dir_entry_2 * old_de, * new_de;
2179 old_bh = new_bh = dir_bh = NULL;
2181 /* Initialize quotas before so that eventual writes go
2182 * in separate transaction */
2183 if (new_dentry->d_inode)
2184 DQUOT_INIT(new_dentry->d_inode);
2185 handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS +
2186 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2188 return PTR_ERR(handle);
2190 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2193 old_bh = ext3_find_entry (old_dentry, &old_de);
2195 * Check for inode number is _not_ due to possible IO errors.
2196 * We might rmdir the source, keep it as pwd of some process
2197 * and merrily kill the link to whatever was created under the
2198 * same name. Goodbye sticky bit ;-<
2200 old_inode = old_dentry->d_inode;
2202 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2205 new_inode = new_dentry->d_inode;
2206 new_bh = ext3_find_entry (new_dentry, &new_de);
2213 if (S_ISDIR(old_inode->i_mode)) {
2215 retval = -ENOTEMPTY;
2216 if (!empty_dir (new_inode))
2220 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2223 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2226 if (!new_inode && new_dir!=old_dir &&
2227 new_dir->i_nlink >= EXT3_LINK_MAX)
2231 retval = ext3_add_entry (handle, new_dentry, old_inode);
2235 BUFFER_TRACE(new_bh, "get write access");
2236 ext3_journal_get_write_access(handle, new_bh);
2237 new_de->inode = le32_to_cpu(old_inode->i_ino);
2238 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2239 EXT3_FEATURE_INCOMPAT_FILETYPE))
2240 new_de->file_type = old_de->file_type;
2241 new_dir->i_version++;
2242 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2243 ext3_journal_dirty_metadata(handle, new_bh);
2249 * Like most other Unix systems, set the ctime for inodes on a
2252 old_inode->i_ctime = CURRENT_TIME;
2253 ext3_mark_inode_dirty(handle, old_inode);
2258 retval = ext3_delete_entry(handle, old_dir, old_de, old_bh);
2259 if (retval == -ENOENT) {
2261 * old_de could have moved out from under us.
2263 struct buffer_head *old_bh2;
2264 struct ext3_dir_entry_2 *old_de2;
2266 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2268 retval = ext3_delete_entry(handle, old_dir,
2274 ext3_warning(old_dir->i_sb, "ext3_rename",
2275 "Deleting old file (%lu), %d, error=%d",
2276 old_dir->i_ino, old_dir->i_nlink, retval);
2280 new_inode->i_nlink--;
2281 new_inode->i_ctime = CURRENT_TIME;
2283 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
2284 ext3_update_dx_flag(old_dir);
2286 BUFFER_TRACE(dir_bh, "get_write_access");
2287 ext3_journal_get_write_access(handle, dir_bh);
2288 PARENT_INO(dir_bh->b_data) = le32_to_cpu(new_dir->i_ino);
2289 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2290 ext3_journal_dirty_metadata(handle, dir_bh);
2293 new_inode->i_nlink--;
2296 ext3_update_dx_flag(new_dir);
2297 ext3_mark_inode_dirty(handle, new_dir);
2300 ext3_mark_inode_dirty(handle, old_dir);
2302 ext3_mark_inode_dirty(handle, new_inode);
2303 if (!new_inode->i_nlink)
2304 ext3_orphan_add(handle, new_inode);
2312 ext3_journal_stop(handle);
2317 * directories can handle most operations...
2319 struct inode_operations ext3_dir_inode_operations = {
2320 .create = ext3_create,
2321 .lookup = ext3_lookup,
2323 .unlink = ext3_unlink,
2324 .symlink = ext3_symlink,
2325 .mkdir = ext3_mkdir,
2326 .rmdir = ext3_rmdir,
2327 .mknod = ext3_mknod,
2328 .rename = ext3_rename,
2329 .setattr = ext3_setattr,
2330 .setxattr = ext3_setxattr,
2331 .getxattr = ext3_getxattr,
2332 .listxattr = ext3_listxattr,
2333 .removexattr = ext3_removexattr,
2334 .permission = ext3_permission,
2337 struct inode_operations ext3_special_inode_operations = {
2338 .setattr = ext3_setattr,
2339 .setxattr = ext3_setxattr,
2340 .getxattr = ext3_getxattr,
2341 .listxattr = ext3_listxattr,
2342 .removexattr = ext3_removexattr,
2343 .permission = ext3_permission,