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>
39 #include <trace/events/ext3.h>
46 * define how far ahead to read directories while searching them.
48 #define NAMEI_RA_CHUNKS 2
49 #define NAMEI_RA_BLOCKS 4
50 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
51 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
53 static struct buffer_head *ext3_append(handle_t *handle,
57 struct buffer_head *bh;
59 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
61 bh = ext3_bread(handle, inode, *block, 1, err);
63 inode->i_size += inode->i_sb->s_blocksize;
64 EXT3_I(inode)->i_disksize = inode->i_size;
65 *err = ext3_journal_get_write_access(handle, bh);
75 #define assert(test) J_ASSERT(test)
79 #define dxtrace(command) command
81 #define dxtrace(command)
105 * dx_root_info is laid out so that if it should somehow get overlaid by a
106 * dirent the two low bits of the hash version will be zero. Therefore, the
107 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
112 struct fake_dirent dot;
114 struct fake_dirent dotdot;
118 __le32 reserved_zero;
120 u8 info_length; /* 8 */
125 struct dx_entry entries[0];
130 struct fake_dirent fake;
131 struct dx_entry entries[0];
137 struct buffer_head *bh;
138 struct dx_entry *entries;
149 static inline unsigned dx_get_block (struct dx_entry *entry);
150 static void dx_set_block (struct dx_entry *entry, unsigned value);
151 static inline unsigned dx_get_hash (struct dx_entry *entry);
152 static void dx_set_hash (struct dx_entry *entry, unsigned value);
153 static unsigned dx_get_count (struct dx_entry *entries);
154 static unsigned dx_get_limit (struct dx_entry *entries);
155 static void dx_set_count (struct dx_entry *entries, unsigned value);
156 static void dx_set_limit (struct dx_entry *entries, unsigned value);
157 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
158 static unsigned dx_node_limit (struct inode *dir);
159 static struct dx_frame *dx_probe(struct qstr *entry,
161 struct dx_hash_info *hinfo,
162 struct dx_frame *frame,
164 static void dx_release (struct dx_frame *frames);
165 static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
166 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
167 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
168 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
169 struct dx_map_entry *offsets, int count);
170 static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize);
171 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
172 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
173 struct dx_frame *frame,
174 struct dx_frame *frames,
176 static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
177 struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
179 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
180 struct inode *inode);
183 * p is at least 6 bytes before the end of page
185 static inline struct ext3_dir_entry_2 *
186 ext3_next_entry(struct ext3_dir_entry_2 *p)
188 return (struct ext3_dir_entry_2 *)((char *)p +
189 ext3_rec_len_from_disk(p->rec_len));
193 * Future: use high four bits of block for coalesce-on-delete flags
194 * Mask them off for now.
197 static inline unsigned dx_get_block (struct dx_entry *entry)
199 return le32_to_cpu(entry->block) & 0x00ffffff;
202 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
204 entry->block = cpu_to_le32(value);
207 static inline unsigned dx_get_hash (struct dx_entry *entry)
209 return le32_to_cpu(entry->hash);
212 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
214 entry->hash = cpu_to_le32(value);
217 static inline unsigned dx_get_count (struct dx_entry *entries)
219 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
222 static inline unsigned dx_get_limit (struct dx_entry *entries)
224 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
227 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
229 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
232 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
234 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
237 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
239 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
240 EXT3_DIR_REC_LEN(2) - infosize;
241 return entry_space / sizeof(struct dx_entry);
244 static inline unsigned dx_node_limit (struct inode *dir)
246 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
247 return entry_space / sizeof(struct dx_entry);
254 static void dx_show_index (char * label, struct dx_entry *entries)
256 int i, n = dx_get_count (entries);
257 printk("%s index ", label);
258 for (i = 0; i < n; i++)
260 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
272 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
273 int size, int show_names)
275 unsigned names = 0, space = 0;
276 char *base = (char *) de;
277 struct dx_hash_info h = *hinfo;
280 while ((char *) de < base + size)
286 int len = de->name_len;
287 char *name = de->name;
288 while (len--) printk("%c", *name++);
289 ext3fs_dirhash(de->name, de->name_len, &h);
290 printk(":%x.%u ", h.hash,
291 (unsigned) ((char *) de - base));
293 space += EXT3_DIR_REC_LEN(de->name_len);
296 de = ext3_next_entry(de);
298 printk("(%i)\n", names);
299 return (struct stats) { names, space, 1 };
302 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
303 struct dx_entry *entries, int levels)
305 unsigned blocksize = dir->i_sb->s_blocksize;
306 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
308 struct buffer_head *bh;
310 printk("%i indexed blocks...\n", count);
311 for (i = 0; i < count; i++, entries++)
313 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
314 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
316 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
317 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
319 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
320 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
321 names += stats.names;
322 space += stats.space;
323 bcount += stats.bcount;
327 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
328 names, space/bcount,(space/bcount)*100/blocksize);
329 return (struct stats) { names, space, bcount};
331 #endif /* DX_DEBUG */
334 * Probe for a directory leaf block to search.
336 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
337 * error in the directory index, and the caller should fall back to
338 * searching the directory normally. The callers of dx_probe **MUST**
339 * check for this error code, and make sure it never gets reflected
342 static struct dx_frame *
343 dx_probe(struct qstr *entry, struct inode *dir,
344 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
346 unsigned count, indirect;
347 struct dx_entry *at, *entries, *p, *q, *m;
348 struct dx_root *root;
349 struct buffer_head *bh;
350 struct dx_frame *frame = frame_in;
354 if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
356 root = (struct dx_root *) bh->b_data;
357 if (root->info.hash_version != DX_HASH_TEA &&
358 root->info.hash_version != DX_HASH_HALF_MD4 &&
359 root->info.hash_version != DX_HASH_LEGACY) {
360 ext3_warning(dir->i_sb, __func__,
361 "Unrecognised inode hash code %d",
362 root->info.hash_version);
364 *err = ERR_BAD_DX_DIR;
367 hinfo->hash_version = root->info.hash_version;
368 if (hinfo->hash_version <= DX_HASH_TEA)
369 hinfo->hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
370 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
372 ext3fs_dirhash(entry->name, entry->len, hinfo);
375 if (root->info.unused_flags & 1) {
376 ext3_warning(dir->i_sb, __func__,
377 "Unimplemented inode hash flags: %#06x",
378 root->info.unused_flags);
380 *err = ERR_BAD_DX_DIR;
384 if ((indirect = root->info.indirect_levels) > 1) {
385 ext3_warning(dir->i_sb, __func__,
386 "Unimplemented inode hash depth: %#06x",
387 root->info.indirect_levels);
389 *err = ERR_BAD_DX_DIR;
393 entries = (struct dx_entry *) (((char *)&root->info) +
394 root->info.info_length);
396 if (dx_get_limit(entries) != dx_root_limit(dir,
397 root->info.info_length)) {
398 ext3_warning(dir->i_sb, __func__,
399 "dx entry: limit != root limit");
401 *err = ERR_BAD_DX_DIR;
405 dxtrace (printk("Look up %x", hash));
408 count = dx_get_count(entries);
409 if (!count || count > dx_get_limit(entries)) {
410 ext3_warning(dir->i_sb, __func__,
411 "dx entry: no count or count > limit");
413 *err = ERR_BAD_DX_DIR;
418 q = entries + count - 1;
422 dxtrace(printk("."));
423 if (dx_get_hash(m) > hash)
429 if (0) // linear search cross check
431 unsigned n = count - 1;
435 dxtrace(printk(","));
436 if (dx_get_hash(++at) > hash)
442 assert (at == p - 1);
446 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
448 frame->entries = entries;
450 if (!indirect--) return frame;
451 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
453 at = entries = ((struct dx_node *) bh->b_data)->entries;
454 if (dx_get_limit(entries) != dx_node_limit (dir)) {
455 ext3_warning(dir->i_sb, __func__,
456 "dx entry: limit != node limit");
458 *err = ERR_BAD_DX_DIR;
465 while (frame >= frame_in) {
470 if (*err == ERR_BAD_DX_DIR)
471 ext3_warning(dir->i_sb, __func__,
472 "Corrupt dir inode %ld, running e2fsck is "
473 "recommended.", dir->i_ino);
477 static void dx_release (struct dx_frame *frames)
479 if (frames[0].bh == NULL)
482 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
483 brelse(frames[1].bh);
484 brelse(frames[0].bh);
488 * This function increments the frame pointer to search the next leaf
489 * block, and reads in the necessary intervening nodes if the search
490 * should be necessary. Whether or not the search is necessary is
491 * controlled by the hash parameter. If the hash value is even, then
492 * the search is only continued if the next block starts with that
493 * hash value. This is used if we are searching for a specific file.
495 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
497 * This function returns 1 if the caller should continue to search,
498 * or 0 if it should not. If there is an error reading one of the
499 * index blocks, it will a negative error code.
501 * If start_hash is non-null, it will be filled in with the starting
502 * hash of the next page.
504 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
505 struct dx_frame *frame,
506 struct dx_frame *frames,
510 struct buffer_head *bh;
511 int err, num_frames = 0;
516 * Find the next leaf page by incrementing the frame pointer.
517 * If we run out of entries in the interior node, loop around and
518 * increment pointer in the parent node. When we break out of
519 * this loop, num_frames indicates the number of interior
520 * nodes need to be read.
523 if (++(p->at) < p->entries + dx_get_count(p->entries))
532 * If the hash is 1, then continue only if the next page has a
533 * continuation hash of any value. This is used for readdir
534 * handling. Otherwise, check to see if the hash matches the
535 * desired contiuation hash. If it doesn't, return since
536 * there's no point to read in the successive index pages.
538 bhash = dx_get_hash(p->at);
541 if ((hash & 1) == 0) {
542 if ((bhash & ~1) != hash)
546 * If the hash is HASH_NB_ALWAYS, we always go to the next
547 * block so no check is necessary
549 while (num_frames--) {
550 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
552 return err; /* Failure */
556 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
563 * This function fills a red-black tree with information from a
564 * directory block. It returns the number directory entries loaded
565 * into the tree. If there is an error it is returned in err.
567 static int htree_dirblock_to_tree(struct file *dir_file,
568 struct inode *dir, int block,
569 struct dx_hash_info *hinfo,
570 __u32 start_hash, __u32 start_minor_hash)
572 struct buffer_head *bh;
573 struct ext3_dir_entry_2 *de, *top;
576 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
577 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
580 de = (struct ext3_dir_entry_2 *) bh->b_data;
581 top = (struct ext3_dir_entry_2 *) ((char *) de +
582 dir->i_sb->s_blocksize -
583 EXT3_DIR_REC_LEN(0));
584 for (; de < top; de = ext3_next_entry(de)) {
585 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
586 (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
587 +((char *)de - bh->b_data))) {
588 /* silently ignore the rest of the block */
591 ext3fs_dirhash(de->name, de->name_len, hinfo);
592 if ((hinfo->hash < start_hash) ||
593 ((hinfo->hash == start_hash) &&
594 (hinfo->minor_hash < start_minor_hash)))
598 if ((err = ext3_htree_store_dirent(dir_file,
599 hinfo->hash, hinfo->minor_hash, de)) != 0) {
611 * This function fills a red-black tree with information from a
612 * directory. We start scanning the directory in hash order, starting
613 * at start_hash and start_minor_hash.
615 * This function returns the number of entries inserted into the tree,
616 * or a negative error code.
618 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
619 __u32 start_minor_hash, __u32 *next_hash)
621 struct dx_hash_info hinfo;
622 struct ext3_dir_entry_2 *de;
623 struct dx_frame frames[2], *frame;
630 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
632 dir = dir_file->f_path.dentry->d_inode;
633 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
634 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
635 if (hinfo.hash_version <= DX_HASH_TEA)
636 hinfo.hash_version +=
637 EXT3_SB(dir->i_sb)->s_hash_unsigned;
638 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
639 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
640 start_hash, start_minor_hash);
644 hinfo.hash = start_hash;
645 hinfo.minor_hash = 0;
646 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
650 /* Add '.' and '..' from the htree header */
651 if (!start_hash && !start_minor_hash) {
652 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
653 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
657 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
658 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
659 de = ext3_next_entry(de);
660 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
666 block = dx_get_block(frame->at);
667 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
668 start_hash, start_minor_hash);
675 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
676 frame, frames, &hashval);
677 *next_hash = hashval;
683 * Stop if: (a) there are no more entries, or
684 * (b) we have inserted at least one entry and the
685 * next hash value is not a continuation
688 (count && ((hashval & 1) == 0)))
692 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
702 * Directory block splitting, compacting
706 * Create map of hash values, offsets, and sizes, stored at end of block.
707 * Returns number of entries mapped.
709 static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
710 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
713 char *base = (char *) de;
714 struct dx_hash_info h = *hinfo;
716 while ((char *) de < base + blocksize)
718 if (de->name_len && de->inode) {
719 ext3fs_dirhash(de->name, de->name_len, &h);
721 map_tail->hash = h.hash;
722 map_tail->offs = (u16) ((char *) de - base);
723 map_tail->size = le16_to_cpu(de->rec_len);
727 /* XXX: do we need to check rec_len == 0 case? -Chris */
728 de = ext3_next_entry(de);
733 /* Sort map by hash value */
734 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
736 struct dx_map_entry *p, *q, *top = map + count - 1;
738 /* Combsort until bubble sort doesn't suck */
742 if (count - 9 < 2) /* 9, 10 -> 11 */
744 for (p = top, q = p - count; q >= map; p--, q--)
745 if (p->hash < q->hash)
748 /* Garden variety bubble sort */
754 if (q[1].hash >= q[0].hash)
762 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
764 struct dx_entry *entries = frame->entries;
765 struct dx_entry *old = frame->at, *new = old + 1;
766 int count = dx_get_count(entries);
768 assert(count < dx_get_limit(entries));
769 assert(old < entries + count);
770 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
771 dx_set_hash(new, hash);
772 dx_set_block(new, block);
773 dx_set_count(entries, count + 1);
776 static void ext3_update_dx_flag(struct inode *inode)
778 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
779 EXT3_FEATURE_COMPAT_DIR_INDEX))
780 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
784 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
786 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
787 * `de != NULL' is guaranteed by caller.
789 static inline int ext3_match (int len, const char * const name,
790 struct ext3_dir_entry_2 * de)
792 if (len != de->name_len)
796 return !memcmp(name, de->name, len);
800 * Returns 0 if not found, -1 on failure, and 1 on success
802 static inline int search_dirblock(struct buffer_head * bh,
805 unsigned long offset,
806 struct ext3_dir_entry_2 ** res_dir)
808 struct ext3_dir_entry_2 * de;
811 const char *name = child->name;
812 int namelen = child->len;
814 de = (struct ext3_dir_entry_2 *) bh->b_data;
815 dlimit = bh->b_data + dir->i_sb->s_blocksize;
816 while ((char *) de < dlimit) {
817 /* this code is executed quadratically often */
818 /* do minimal checking `by hand' */
820 if ((char *) de + namelen <= dlimit &&
821 ext3_match (namelen, name, de)) {
822 /* found a match - just to be sure, do a full check */
823 if (!ext3_check_dir_entry("ext3_find_entry",
824 dir, de, bh, offset))
829 /* prevent looping on a bad block */
830 de_len = ext3_rec_len_from_disk(de->rec_len);
834 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
843 * finds an entry in the specified directory with the wanted name. It
844 * returns the cache buffer in which the entry was found, and the entry
845 * itself (as a parameter - res_dir). It does NOT read the inode of the
846 * entry - you'll have to do that yourself if you want to.
848 * The returned buffer_head has ->b_count elevated. The caller is expected
849 * to brelse() it when appropriate.
851 static struct buffer_head *ext3_find_entry(struct inode *dir,
853 struct ext3_dir_entry_2 **res_dir)
855 struct super_block * sb;
856 struct buffer_head * bh_use[NAMEI_RA_SIZE];
857 struct buffer_head * bh, *ret = NULL;
858 unsigned long start, block, b;
859 const u8 *name = entry->name;
860 int ra_max = 0; /* Number of bh's in the readahead
862 int ra_ptr = 0; /* Current index into readahead
870 namelen = entry->len;
871 if (namelen > EXT3_NAME_LEN)
873 if ((namelen <= 2) && (name[0] == '.') &&
874 (name[1] == '.' || name[1] == 0)) {
876 * "." or ".." will only be in the first block
877 * NFS may look up ".."; "." should be handled by the VFS
884 bh = ext3_dx_find_entry(dir, entry, res_dir, &err);
886 * On success, or if the error was file not found,
887 * return. Otherwise, fall back to doing a search the
890 if (bh || (err != ERR_BAD_DX_DIR))
892 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
894 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
895 start = EXT3_I(dir)->i_dir_start_lookup;
896 if (start >= nblocks)
902 * We deal with the read-ahead logic here.
904 if (ra_ptr >= ra_max) {
905 /* Refill the readahead buffer */
908 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
910 * Terminate if we reach the end of the
911 * directory and must wrap, or if our
912 * search has finished at this block.
914 if (b >= nblocks || (num && block == start)) {
915 bh_use[ra_max] = NULL;
919 bh = ext3_getblk(NULL, dir, b++, 0, &err);
922 ll_rw_block(READ | REQ_META | REQ_PRIO,
926 if ((bh = bh_use[ra_ptr++]) == NULL)
929 if (!buffer_uptodate(bh)) {
930 /* read error, skip block & hope for the best */
931 ext3_error(sb, __func__, "reading directory #%lu "
932 "offset %lu", dir->i_ino, block);
936 i = search_dirblock(bh, dir, entry,
937 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
939 EXT3_I(dir)->i_dir_start_lookup = block;
941 goto cleanup_and_exit;
945 goto cleanup_and_exit;
948 if (++block >= nblocks)
950 } while (block != start);
953 * If the directory has grown while we were searching, then
954 * search the last part of the directory before giving up.
957 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
958 if (block < nblocks) {
964 /* Clean up the read-ahead blocks */
965 for (; ra_ptr < ra_max; ra_ptr++)
966 brelse (bh_use[ra_ptr]);
970 static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
971 struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
974 struct super_block *sb = dir->i_sb;
975 struct dx_hash_info hinfo;
976 struct dx_frame frames[2], *frame;
977 struct buffer_head *bh;
981 if (!(frame = dx_probe(entry, dir, &hinfo, frames, err)))
984 block = dx_get_block(frame->at);
985 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
988 retval = search_dirblock(bh, dir, entry,
989 block << EXT3_BLOCK_SIZE_BITS(sb),
997 *err = ERR_BAD_DX_DIR;
1001 /* Check to see if we should continue to search */
1002 retval = ext3_htree_next_block(dir, hinfo.hash, frame,
1005 ext3_warning(sb, __func__,
1006 "error reading index page in directory #%lu",
1011 } while (retval == 1);
1015 dxtrace(printk("%s not found\n", entry->name));
1016 dx_release (frames);
1020 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1022 struct inode * inode;
1023 struct ext3_dir_entry_2 * de;
1024 struct buffer_head * bh;
1026 if (dentry->d_name.len > EXT3_NAME_LEN)
1027 return ERR_PTR(-ENAMETOOLONG);
1029 bh = ext3_find_entry(dir, &dentry->d_name, &de);
1032 unsigned long ino = le32_to_cpu(de->inode);
1034 if (!ext3_valid_inum(dir->i_sb, ino)) {
1035 ext3_error(dir->i_sb, "ext3_lookup",
1036 "bad inode number: %lu", ino);
1037 return ERR_PTR(-EIO);
1039 inode = ext3_iget(dir->i_sb, ino);
1040 if (inode == ERR_PTR(-ESTALE)) {
1041 ext3_error(dir->i_sb, __func__,
1042 "deleted inode referenced: %lu",
1044 return ERR_PTR(-EIO);
1047 return d_splice_alias(inode, dentry);
1051 struct dentry *ext3_get_parent(struct dentry *child)
1054 struct qstr dotdot = {.name = "..", .len = 2};
1055 struct ext3_dir_entry_2 * de;
1056 struct buffer_head *bh;
1058 bh = ext3_find_entry(child->d_inode, &dotdot, &de);
1060 return ERR_PTR(-ENOENT);
1061 ino = le32_to_cpu(de->inode);
1064 if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1065 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1066 "bad inode number: %lu", ino);
1067 return ERR_PTR(-EIO);
1070 return d_obtain_alias(ext3_iget(child->d_inode->i_sb, ino));
1074 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1075 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1076 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1077 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1078 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1079 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1080 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1081 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1084 static inline void ext3_set_de_type(struct super_block *sb,
1085 struct ext3_dir_entry_2 *de,
1087 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1088 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1092 * Move count entries from end of map between two memory locations.
1093 * Returns pointer to last entry moved.
1095 static struct ext3_dir_entry_2 *
1096 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1098 unsigned rec_len = 0;
1101 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1102 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1103 memcpy (to, de, rec_len);
1104 ((struct ext3_dir_entry_2 *) to)->rec_len =
1105 ext3_rec_len_to_disk(rec_len);
1110 return (struct ext3_dir_entry_2 *) (to - rec_len);
1114 * Compact each dir entry in the range to the minimal rec_len.
1115 * Returns pointer to last entry in range.
1117 static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize)
1119 struct ext3_dir_entry_2 *next, *to, *prev;
1120 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *)base;
1121 unsigned rec_len = 0;
1124 while ((char *)de < base + blocksize) {
1125 next = ext3_next_entry(de);
1126 if (de->inode && de->name_len) {
1127 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1129 memmove(to, de, rec_len);
1130 to->rec_len = ext3_rec_len_to_disk(rec_len);
1132 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1140 * Split a full leaf block to make room for a new dir entry.
1141 * Allocate a new block, and move entries so that they are approx. equally full.
1142 * Returns pointer to de in block into which the new entry will be inserted.
1144 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1145 struct buffer_head **bh,struct dx_frame *frame,
1146 struct dx_hash_info *hinfo, int *error)
1148 unsigned blocksize = dir->i_sb->s_blocksize;
1149 unsigned count, continued;
1150 struct buffer_head *bh2;
1153 struct dx_map_entry *map;
1154 char *data1 = (*bh)->b_data, *data2;
1155 unsigned split, move, size;
1156 struct ext3_dir_entry_2 *de = NULL, *de2;
1159 bh2 = ext3_append (handle, dir, &newblock, &err);
1166 BUFFER_TRACE(*bh, "get_write_access");
1167 err = ext3_journal_get_write_access(handle, *bh);
1171 BUFFER_TRACE(frame->bh, "get_write_access");
1172 err = ext3_journal_get_write_access(handle, frame->bh);
1176 data2 = bh2->b_data;
1178 /* create map in the end of data2 block */
1179 map = (struct dx_map_entry *) (data2 + blocksize);
1180 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1181 blocksize, hinfo, map);
1183 dx_sort_map (map, count);
1184 /* Split the existing block in the middle, size-wise */
1187 for (i = count-1; i >= 0; i--) {
1188 /* is more than half of this entry in 2nd half of the block? */
1189 if (size + map[i].size/2 > blocksize/2)
1191 size += map[i].size;
1194 /* map index at which we will split */
1195 split = count - move;
1196 hash2 = map[split].hash;
1197 continued = hash2 == map[split - 1].hash;
1198 dxtrace(printk("Split block %i at %x, %i/%i\n",
1199 dx_get_block(frame->at), hash2, split, count-split));
1201 /* Fancy dance to stay within two buffers */
1202 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1203 de = dx_pack_dirents(data1,blocksize);
1204 de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1205 de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
1206 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1207 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1209 /* Which block gets the new entry? */
1210 if (hinfo->hash >= hash2)
1215 dx_insert_block (frame, hash2 + continued, newblock);
1216 err = ext3_journal_dirty_metadata (handle, bh2);
1219 err = ext3_journal_dirty_metadata (handle, frame->bh);
1223 dxtrace(dx_show_index ("frame", frame->entries));
1230 ext3_std_error(dir->i_sb, err);
1238 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1239 * it points to a directory entry which is guaranteed to be large
1240 * enough for new directory entry. If de is NULL, then
1241 * add_dirent_to_buf will attempt search the directory block for
1242 * space. It will return -ENOSPC if no space is available, and -EIO
1243 * and -EEXIST if directory entry already exists.
1245 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1246 * all other cases bh is released.
1248 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1249 struct inode *inode, struct ext3_dir_entry_2 *de,
1250 struct buffer_head * bh)
1252 struct inode *dir = dentry->d_parent->d_inode;
1253 const char *name = dentry->d_name.name;
1254 int namelen = dentry->d_name.len;
1255 unsigned long offset = 0;
1256 unsigned short reclen;
1257 int nlen, rlen, err;
1260 reclen = EXT3_DIR_REC_LEN(namelen);
1262 de = (struct ext3_dir_entry_2 *)bh->b_data;
1263 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1264 while ((char *) de <= top) {
1265 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1270 if (ext3_match (namelen, name, de)) {
1274 nlen = EXT3_DIR_REC_LEN(de->name_len);
1275 rlen = ext3_rec_len_from_disk(de->rec_len);
1276 if ((de->inode? rlen - nlen: rlen) >= reclen)
1278 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1281 if ((char *) de > top)
1284 BUFFER_TRACE(bh, "get_write_access");
1285 err = ext3_journal_get_write_access(handle, bh);
1287 ext3_std_error(dir->i_sb, err);
1292 /* By now the buffer is marked for journaling */
1293 nlen = EXT3_DIR_REC_LEN(de->name_len);
1294 rlen = ext3_rec_len_from_disk(de->rec_len);
1296 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1297 de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1298 de->rec_len = ext3_rec_len_to_disk(nlen);
1301 de->file_type = EXT3_FT_UNKNOWN;
1303 de->inode = cpu_to_le32(inode->i_ino);
1304 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1307 de->name_len = namelen;
1308 memcpy (de->name, name, namelen);
1310 * XXX shouldn't update any times until successful
1311 * completion of syscall, but too many callers depend
1314 * XXX similarly, too many callers depend on
1315 * ext3_new_inode() setting the times, but error
1316 * recovery deletes the inode, so the worst that can
1317 * happen is that the times are slightly out of date
1318 * and/or different from the directory change time.
1320 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1321 ext3_update_dx_flag(dir);
1323 ext3_mark_inode_dirty(handle, dir);
1324 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1325 err = ext3_journal_dirty_metadata(handle, bh);
1327 ext3_std_error(dir->i_sb, err);
1333 * This converts a one block unindexed directory to a 3 block indexed
1334 * directory, and adds the dentry to the indexed directory.
1336 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1337 struct inode *inode, struct buffer_head *bh)
1339 struct inode *dir = dentry->d_parent->d_inode;
1340 const char *name = dentry->d_name.name;
1341 int namelen = dentry->d_name.len;
1342 struct buffer_head *bh2;
1343 struct dx_root *root;
1344 struct dx_frame frames[2], *frame;
1345 struct dx_entry *entries;
1346 struct ext3_dir_entry_2 *de, *de2;
1351 struct dx_hash_info hinfo;
1353 struct fake_dirent *fde;
1355 blocksize = dir->i_sb->s_blocksize;
1356 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1357 retval = ext3_journal_get_write_access(handle, bh);
1359 ext3_std_error(dir->i_sb, retval);
1363 root = (struct dx_root *) bh->b_data;
1365 /* The 0th block becomes the root, move the dirents out */
1366 fde = &root->dotdot;
1367 de = (struct ext3_dir_entry_2 *)((char *)fde +
1368 ext3_rec_len_from_disk(fde->rec_len));
1369 if ((char *) de >= (((char *) root) + blocksize)) {
1370 ext3_error(dir->i_sb, __func__,
1371 "invalid rec_len for '..' in inode %lu",
1376 len = ((char *) root) + blocksize - (char *) de;
1378 bh2 = ext3_append (handle, dir, &block, &retval);
1383 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1384 data1 = bh2->b_data;
1386 memcpy (data1, de, len);
1387 de = (struct ext3_dir_entry_2 *) data1;
1389 while ((char *)(de2 = ext3_next_entry(de)) < top)
1391 de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1392 /* Initialize the root; the dot dirents already exist */
1393 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1394 de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
1395 memset (&root->info, 0, sizeof(root->info));
1396 root->info.info_length = sizeof(root->info);
1397 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1398 entries = root->entries;
1399 dx_set_block (entries, 1);
1400 dx_set_count (entries, 1);
1401 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1403 /* Initialize as for dx_probe */
1404 hinfo.hash_version = root->info.hash_version;
1405 if (hinfo.hash_version <= DX_HASH_TEA)
1406 hinfo.hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
1407 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1408 ext3fs_dirhash(name, namelen, &hinfo);
1410 frame->entries = entries;
1411 frame->at = entries;
1415 * Mark buffers dirty here so that if do_split() fails we write a
1416 * consistent set of buffers to disk.
1418 ext3_journal_dirty_metadata(handle, frame->bh);
1419 ext3_journal_dirty_metadata(handle, bh);
1420 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1422 ext3_mark_inode_dirty(handle, dir);
1428 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1434 * adds a file entry to the specified directory, using the same
1435 * semantics as ext3_find_entry(). It returns NULL if it failed.
1437 * NOTE!! The inode part of 'de' is left at 0 - which means you
1438 * may not sleep between calling this and putting something into
1439 * the entry, as someone else might have used it while you slept.
1441 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1442 struct inode *inode)
1444 struct inode *dir = dentry->d_parent->d_inode;
1445 struct buffer_head * bh;
1446 struct ext3_dir_entry_2 *de;
1447 struct super_block * sb;
1454 blocksize = sb->s_blocksize;
1455 if (!dentry->d_name.len)
1458 retval = ext3_dx_add_entry(handle, dentry, inode);
1459 if (!retval || (retval != ERR_BAD_DX_DIR))
1461 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1463 ext3_mark_inode_dirty(handle, dir);
1465 blocks = dir->i_size >> sb->s_blocksize_bits;
1466 for (block = 0; block < blocks; block++) {
1467 bh = ext3_bread(handle, dir, block, 0, &retval);
1470 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1471 if (retval != -ENOSPC)
1474 if (blocks == 1 && !dx_fallback &&
1475 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1476 return make_indexed_dir(handle, dentry, inode, bh);
1479 bh = ext3_append(handle, dir, &block, &retval);
1482 de = (struct ext3_dir_entry_2 *) bh->b_data;
1484 de->rec_len = ext3_rec_len_to_disk(blocksize);
1485 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1489 * Returns 0 for success, or a negative error value
1491 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1492 struct inode *inode)
1494 struct dx_frame frames[2], *frame;
1495 struct dx_entry *entries, *at;
1496 struct dx_hash_info hinfo;
1497 struct buffer_head * bh;
1498 struct inode *dir = dentry->d_parent->d_inode;
1499 struct super_block * sb = dir->i_sb;
1500 struct ext3_dir_entry_2 *de;
1503 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1506 entries = frame->entries;
1509 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1512 BUFFER_TRACE(bh, "get_write_access");
1513 err = ext3_journal_get_write_access(handle, bh);
1517 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1518 if (err != -ENOSPC) {
1523 /* Block full, should compress but for now just split */
1524 dxtrace(printk("using %u of %u node entries\n",
1525 dx_get_count(entries), dx_get_limit(entries)));
1526 /* Need to split index? */
1527 if (dx_get_count(entries) == dx_get_limit(entries)) {
1529 unsigned icount = dx_get_count(entries);
1530 int levels = frame - frames;
1531 struct dx_entry *entries2;
1532 struct dx_node *node2;
1533 struct buffer_head *bh2;
1535 if (levels && (dx_get_count(frames->entries) ==
1536 dx_get_limit(frames->entries))) {
1537 ext3_warning(sb, __func__,
1538 "Directory index full!");
1542 bh2 = ext3_append (handle, dir, &newblock, &err);
1545 node2 = (struct dx_node *)(bh2->b_data);
1546 entries2 = node2->entries;
1547 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1548 node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
1549 BUFFER_TRACE(frame->bh, "get_write_access");
1550 err = ext3_journal_get_write_access(handle, frame->bh);
1554 unsigned icount1 = icount/2, icount2 = icount - icount1;
1555 unsigned hash2 = dx_get_hash(entries + icount1);
1556 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1558 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1559 err = ext3_journal_get_write_access(handle,
1564 memcpy ((char *) entries2, (char *) (entries + icount1),
1565 icount2 * sizeof(struct dx_entry));
1566 dx_set_count (entries, icount1);
1567 dx_set_count (entries2, icount2);
1568 dx_set_limit (entries2, dx_node_limit(dir));
1570 /* Which index block gets the new entry? */
1571 if (at - entries >= icount1) {
1572 frame->at = at = at - entries - icount1 + entries2;
1573 frame->entries = entries = entries2;
1574 swap(frame->bh, bh2);
1576 dx_insert_block (frames + 0, hash2, newblock);
1577 dxtrace(dx_show_index ("node", frames[1].entries));
1578 dxtrace(dx_show_index ("node",
1579 ((struct dx_node *) bh2->b_data)->entries));
1580 err = ext3_journal_dirty_metadata(handle, bh2);
1585 dxtrace(printk("Creating second level index...\n"));
1586 memcpy((char *) entries2, (char *) entries,
1587 icount * sizeof(struct dx_entry));
1588 dx_set_limit(entries2, dx_node_limit(dir));
1591 dx_set_count(entries, 1);
1592 dx_set_block(entries + 0, newblock);
1593 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1595 /* Add new access path frame */
1597 frame->at = at = at - entries + entries2;
1598 frame->entries = entries = entries2;
1600 err = ext3_journal_get_write_access(handle,
1605 err = ext3_journal_dirty_metadata(handle, frames[0].bh);
1609 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1612 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1617 ext3_std_error(dir->i_sb, err);
1626 * ext3_delete_entry deletes a directory entry by merging it with the
1629 static int ext3_delete_entry (handle_t *handle,
1631 struct ext3_dir_entry_2 * de_del,
1632 struct buffer_head * bh)
1634 struct ext3_dir_entry_2 * de, * pde;
1639 de = (struct ext3_dir_entry_2 *) bh->b_data;
1640 while (i < bh->b_size) {
1641 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1646 BUFFER_TRACE(bh, "get_write_access");
1647 err = ext3_journal_get_write_access(handle, bh);
1652 pde->rec_len = ext3_rec_len_to_disk(
1653 ext3_rec_len_from_disk(pde->rec_len) +
1654 ext3_rec_len_from_disk(de->rec_len));
1658 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1659 err = ext3_journal_dirty_metadata(handle, bh);
1662 ext3_std_error(dir->i_sb, err);
1667 i += ext3_rec_len_from_disk(de->rec_len);
1669 de = ext3_next_entry(de);
1674 static int ext3_add_nondir(handle_t *handle,
1675 struct dentry *dentry, struct inode *inode)
1677 int err = ext3_add_entry(handle, dentry, inode);
1679 ext3_mark_inode_dirty(handle, inode);
1680 d_instantiate(dentry, inode);
1681 unlock_new_inode(inode);
1685 unlock_new_inode(inode);
1691 * By the time this is called, we already have created
1692 * the directory cache entry for the new file, but it
1693 * is so far negative - it has no inode.
1695 * If the create succeeds, we fill in the inode information
1696 * with d_instantiate().
1698 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1699 struct nameidata *nd)
1702 struct inode * inode;
1703 int err, retries = 0;
1705 dquot_initialize(dir);
1708 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1709 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1710 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1712 return PTR_ERR(handle);
1714 if (IS_DIRSYNC(dir))
1717 inode = ext3_new_inode (handle, dir, &dentry->d_name, mode);
1718 err = PTR_ERR(inode);
1719 if (!IS_ERR(inode)) {
1720 inode->i_op = &ext3_file_inode_operations;
1721 inode->i_fop = &ext3_file_operations;
1722 ext3_set_aops(inode);
1723 err = ext3_add_nondir(handle, dentry, inode);
1725 ext3_journal_stop(handle);
1726 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1731 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1732 int mode, dev_t rdev)
1735 struct inode *inode;
1736 int err, retries = 0;
1738 if (!new_valid_dev(rdev))
1741 dquot_initialize(dir);
1744 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1745 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1746 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1748 return PTR_ERR(handle);
1750 if (IS_DIRSYNC(dir))
1753 inode = ext3_new_inode (handle, dir, &dentry->d_name, mode);
1754 err = PTR_ERR(inode);
1755 if (!IS_ERR(inode)) {
1756 init_special_inode(inode, inode->i_mode, rdev);
1757 #ifdef CONFIG_EXT3_FS_XATTR
1758 inode->i_op = &ext3_special_inode_operations;
1760 err = ext3_add_nondir(handle, dentry, inode);
1762 ext3_journal_stop(handle);
1763 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1768 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1771 struct inode * inode;
1772 struct buffer_head * dir_block = NULL;
1773 struct ext3_dir_entry_2 * de;
1774 int err, retries = 0;
1776 if (dir->i_nlink >= EXT3_LINK_MAX)
1779 dquot_initialize(dir);
1782 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1783 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1784 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1786 return PTR_ERR(handle);
1788 if (IS_DIRSYNC(dir))
1791 inode = ext3_new_inode (handle, dir, &dentry->d_name, S_IFDIR | mode);
1792 err = PTR_ERR(inode);
1796 inode->i_op = &ext3_dir_inode_operations;
1797 inode->i_fop = &ext3_dir_operations;
1798 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1799 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1801 goto out_clear_inode;
1803 BUFFER_TRACE(dir_block, "get_write_access");
1804 err = ext3_journal_get_write_access(handle, dir_block);
1806 goto out_clear_inode;
1808 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1809 de->inode = cpu_to_le32(inode->i_ino);
1811 de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
1812 strcpy (de->name, ".");
1813 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1814 de = ext3_next_entry(de);
1815 de->inode = cpu_to_le32(dir->i_ino);
1816 de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1817 EXT3_DIR_REC_LEN(1));
1819 strcpy (de->name, "..");
1820 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1821 set_nlink(inode, 2);
1822 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1823 err = ext3_journal_dirty_metadata(handle, dir_block);
1825 goto out_clear_inode;
1827 err = ext3_mark_inode_dirty(handle, inode);
1829 err = ext3_add_entry (handle, dentry, inode);
1834 unlock_new_inode(inode);
1835 ext3_mark_inode_dirty(handle, inode);
1840 ext3_update_dx_flag(dir);
1841 err = ext3_mark_inode_dirty(handle, dir);
1843 goto out_clear_inode;
1845 d_instantiate(dentry, inode);
1846 unlock_new_inode(inode);
1849 ext3_journal_stop(handle);
1850 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1856 * routine to check that the specified directory is empty (for rmdir)
1858 static int empty_dir (struct inode * inode)
1860 unsigned long offset;
1861 struct buffer_head * bh;
1862 struct ext3_dir_entry_2 * de, * de1;
1863 struct super_block * sb;
1867 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1868 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1870 ext3_error(inode->i_sb, __func__,
1871 "error %d reading directory #%lu offset 0",
1874 ext3_warning(inode->i_sb, __func__,
1875 "bad directory (dir #%lu) - no data block",
1879 de = (struct ext3_dir_entry_2 *) bh->b_data;
1880 de1 = ext3_next_entry(de);
1881 if (le32_to_cpu(de->inode) != inode->i_ino ||
1882 !le32_to_cpu(de1->inode) ||
1883 strcmp (".", de->name) ||
1884 strcmp ("..", de1->name)) {
1885 ext3_warning (inode->i_sb, "empty_dir",
1886 "bad directory (dir #%lu) - no `.' or `..'",
1891 offset = ext3_rec_len_from_disk(de->rec_len) +
1892 ext3_rec_len_from_disk(de1->rec_len);
1893 de = ext3_next_entry(de1);
1894 while (offset < inode->i_size ) {
1896 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1899 bh = ext3_bread (NULL, inode,
1900 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1903 ext3_error(sb, __func__,
1904 "error %d reading directory"
1906 err, inode->i_ino, offset);
1907 offset += sb->s_blocksize;
1910 de = (struct ext3_dir_entry_2 *) bh->b_data;
1912 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1913 de = (struct ext3_dir_entry_2 *)(bh->b_data +
1915 offset = (offset | (sb->s_blocksize - 1)) + 1;
1918 if (le32_to_cpu(de->inode)) {
1922 offset += ext3_rec_len_from_disk(de->rec_len);
1923 de = ext3_next_entry(de);
1929 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1930 * such inodes, starting at the superblock, in case we crash before the
1931 * file is closed/deleted, or in case the inode truncate spans multiple
1932 * transactions and the last transaction is not recovered after a crash.
1934 * At filesystem recovery time, we walk this list deleting unlinked
1935 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1937 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1939 struct super_block *sb = inode->i_sb;
1940 struct ext3_iloc iloc;
1943 mutex_lock(&EXT3_SB(sb)->s_orphan_lock);
1944 if (!list_empty(&EXT3_I(inode)->i_orphan))
1947 /* Orphan handling is only valid for files with data blocks
1948 * being truncated, or files being unlinked. */
1950 /* @@@ FIXME: Observation from aviro:
1951 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1952 * here (on s_orphan_lock), so race with ext3_link() which might bump
1953 * ->i_nlink. For, say it, character device. Not a regular file,
1954 * not a directory, not a symlink and ->i_nlink > 0.
1956 * tytso, 4/25/2009: I'm not sure how that could happen;
1957 * shouldn't the fs core protect us from these sort of
1958 * unlink()/link() races?
1960 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1961 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1963 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1964 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1968 err = ext3_reserve_inode_write(handle, inode, &iloc);
1972 /* Insert this inode at the head of the on-disk orphan list... */
1973 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1974 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1975 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1976 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1980 /* Only add to the head of the in-memory list if all the
1981 * previous operations succeeded. If the orphan_add is going to
1982 * fail (possibly taking the journal offline), we can't risk
1983 * leaving the inode on the orphan list: stray orphan-list
1984 * entries can cause panics at unmount time.
1986 * This is safe: on error we're going to ignore the orphan list
1987 * anyway on the next recovery. */
1989 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1991 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1992 jbd_debug(4, "orphan inode %lu will point to %d\n",
1993 inode->i_ino, NEXT_ORPHAN(inode));
1995 mutex_unlock(&EXT3_SB(sb)->s_orphan_lock);
1996 ext3_std_error(inode->i_sb, err);
2001 * ext3_orphan_del() removes an unlinked or truncated inode from the list
2002 * of such inodes stored on disk, because it is finally being cleaned up.
2004 int ext3_orphan_del(handle_t *handle, struct inode *inode)
2006 struct list_head *prev;
2007 struct ext3_inode_info *ei = EXT3_I(inode);
2008 struct ext3_sb_info *sbi;
2009 unsigned long ino_next;
2010 struct ext3_iloc iloc;
2013 mutex_lock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
2014 if (list_empty(&ei->i_orphan))
2017 ino_next = NEXT_ORPHAN(inode);
2018 prev = ei->i_orphan.prev;
2019 sbi = EXT3_SB(inode->i_sb);
2021 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2023 list_del_init(&ei->i_orphan);
2025 /* If we're on an error path, we may not have a valid
2026 * transaction handle with which to update the orphan list on
2027 * disk, but we still need to remove the inode from the linked
2028 * list in memory. */
2032 err = ext3_reserve_inode_write(handle, inode, &iloc);
2036 if (prev == &sbi->s_orphan) {
2037 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2038 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2039 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2042 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2043 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2045 struct ext3_iloc iloc2;
2046 struct inode *i_prev =
2047 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2049 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2050 i_prev->i_ino, ino_next);
2051 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2054 NEXT_ORPHAN(i_prev) = ino_next;
2055 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2059 NEXT_ORPHAN(inode) = 0;
2060 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2063 ext3_std_error(inode->i_sb, err);
2065 mutex_unlock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
2073 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2076 struct inode * inode;
2077 struct buffer_head * bh;
2078 struct ext3_dir_entry_2 * de;
2081 /* Initialize quotas before so that eventual writes go in
2082 * separate transaction */
2083 dquot_initialize(dir);
2084 dquot_initialize(dentry->d_inode);
2086 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2088 return PTR_ERR(handle);
2091 bh = ext3_find_entry(dir, &dentry->d_name, &de);
2095 if (IS_DIRSYNC(dir))
2098 inode = dentry->d_inode;
2101 if (le32_to_cpu(de->inode) != inode->i_ino)
2104 retval = -ENOTEMPTY;
2105 if (!empty_dir (inode))
2108 retval = ext3_delete_entry(handle, dir, de, bh);
2111 if (inode->i_nlink != 2)
2112 ext3_warning (inode->i_sb, "ext3_rmdir",
2113 "empty directory has nlink!=2 (%d)",
2117 /* There's no need to set i_disksize: the fact that i_nlink is
2118 * zero will ensure that the right thing happens during any
2121 ext3_orphan_add(handle, inode);
2122 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2123 ext3_mark_inode_dirty(handle, inode);
2125 ext3_update_dx_flag(dir);
2126 ext3_mark_inode_dirty(handle, dir);
2129 ext3_journal_stop(handle);
2134 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2137 struct inode * inode;
2138 struct buffer_head * bh;
2139 struct ext3_dir_entry_2 * de;
2142 trace_ext3_unlink_enter(dir, dentry);
2143 /* Initialize quotas before so that eventual writes go
2144 * in separate transaction */
2145 dquot_initialize(dir);
2146 dquot_initialize(dentry->d_inode);
2148 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2150 return PTR_ERR(handle);
2152 if (IS_DIRSYNC(dir))
2156 bh = ext3_find_entry(dir, &dentry->d_name, &de);
2160 inode = dentry->d_inode;
2163 if (le32_to_cpu(de->inode) != inode->i_ino)
2166 if (!inode->i_nlink) {
2167 ext3_warning (inode->i_sb, "ext3_unlink",
2168 "Deleting nonexistent file (%lu), %d",
2169 inode->i_ino, inode->i_nlink);
2170 set_nlink(inode, 1);
2172 retval = ext3_delete_entry(handle, dir, de, bh);
2175 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2176 ext3_update_dx_flag(dir);
2177 ext3_mark_inode_dirty(handle, dir);
2179 if (!inode->i_nlink)
2180 ext3_orphan_add(handle, inode);
2181 inode->i_ctime = dir->i_ctime;
2182 ext3_mark_inode_dirty(handle, inode);
2186 ext3_journal_stop(handle);
2188 trace_ext3_unlink_exit(dentry, retval);
2192 static int ext3_symlink (struct inode * dir,
2193 struct dentry *dentry, const char * symname)
2196 struct inode * inode;
2197 int l, err, retries = 0;
2200 l = strlen(symname)+1;
2201 if (l > dir->i_sb->s_blocksize)
2202 return -ENAMETOOLONG;
2204 dquot_initialize(dir);
2206 if (l > EXT3_N_BLOCKS * 4) {
2208 * For non-fast symlinks, we just allocate inode and put it on
2209 * orphan list in the first transaction => we need bitmap,
2210 * group descriptor, sb, inode block, quota blocks, and
2211 * possibly selinux xattr blocks.
2213 credits = 4 + EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2214 EXT3_XATTR_TRANS_BLOCKS;
2217 * Fast symlink. We have to add entry to directory
2218 * (EXT3_DATA_TRANS_BLOCKS + EXT3_INDEX_EXTRA_TRANS_BLOCKS),
2219 * allocate new inode (bitmap, group descriptor, inode block,
2220 * quota blocks, sb is already counted in previous macros).
2222 credits = EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2223 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2224 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2227 handle = ext3_journal_start(dir, credits);
2229 return PTR_ERR(handle);
2231 if (IS_DIRSYNC(dir))
2234 inode = ext3_new_inode (handle, dir, &dentry->d_name, S_IFLNK|S_IRWXUGO);
2235 err = PTR_ERR(inode);
2239 if (l > EXT3_N_BLOCKS * 4) {
2240 inode->i_op = &ext3_symlink_inode_operations;
2241 ext3_set_aops(inode);
2243 * We cannot call page_symlink() with transaction started
2244 * because it calls into ext3_write_begin() which acquires page
2245 * lock which ranks below transaction start (and it can also
2246 * wait for journal commit if we are running out of space). So
2247 * we have to stop transaction now and restart it when symlink
2248 * contents is written.
2250 * To keep fs consistent in case of crash, we have to put inode
2251 * to orphan list in the mean time.
2254 err = ext3_orphan_add(handle, inode);
2255 ext3_journal_stop(handle);
2257 goto err_drop_inode;
2258 err = __page_symlink(inode, symname, l, 1);
2260 goto err_drop_inode;
2262 * Now inode is being linked into dir (EXT3_DATA_TRANS_BLOCKS
2263 * + EXT3_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2265 handle = ext3_journal_start(dir,
2266 EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2267 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 1);
2268 if (IS_ERR(handle)) {
2269 err = PTR_ERR(handle);
2270 goto err_drop_inode;
2273 err = ext3_orphan_del(handle, inode);
2275 ext3_journal_stop(handle);
2277 goto err_drop_inode;
2280 inode->i_op = &ext3_fast_symlink_inode_operations;
2281 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2282 inode->i_size = l-1;
2284 EXT3_I(inode)->i_disksize = inode->i_size;
2285 err = ext3_add_nondir(handle, dentry, inode);
2287 ext3_journal_stop(handle);
2288 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2292 unlock_new_inode(inode);
2297 static int ext3_link (struct dentry * old_dentry,
2298 struct inode * dir, struct dentry *dentry)
2301 struct inode *inode = old_dentry->d_inode;
2302 int err, retries = 0;
2304 if (inode->i_nlink >= EXT3_LINK_MAX)
2307 dquot_initialize(dir);
2310 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2311 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2313 return PTR_ERR(handle);
2315 if (IS_DIRSYNC(dir))
2318 inode->i_ctime = CURRENT_TIME_SEC;
2322 err = ext3_add_entry(handle, dentry, inode);
2324 ext3_mark_inode_dirty(handle, inode);
2325 d_instantiate(dentry, inode);
2330 ext3_journal_stop(handle);
2331 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2336 #define PARENT_INO(buffer) \
2337 (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2340 * Anybody can rename anything with this: the permission checks are left to the
2341 * higher-level routines.
2343 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2344 struct inode * new_dir,struct dentry *new_dentry)
2347 struct inode * old_inode, * new_inode;
2348 struct buffer_head * old_bh, * new_bh, * dir_bh;
2349 struct ext3_dir_entry_2 * old_de, * new_de;
2350 int retval, flush_file = 0;
2352 dquot_initialize(old_dir);
2353 dquot_initialize(new_dir);
2355 old_bh = new_bh = dir_bh = NULL;
2357 /* Initialize quotas before so that eventual writes go
2358 * in separate transaction */
2359 if (new_dentry->d_inode)
2360 dquot_initialize(new_dentry->d_inode);
2361 handle = ext3_journal_start(old_dir, 2 *
2362 EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2363 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2365 return PTR_ERR(handle);
2367 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2370 old_bh = ext3_find_entry(old_dir, &old_dentry->d_name, &old_de);
2372 * Check for inode number is _not_ due to possible IO errors.
2373 * We might rmdir the source, keep it as pwd of some process
2374 * and merrily kill the link to whatever was created under the
2375 * same name. Goodbye sticky bit ;-<
2377 old_inode = old_dentry->d_inode;
2379 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2382 new_inode = new_dentry->d_inode;
2383 new_bh = ext3_find_entry(new_dir, &new_dentry->d_name, &new_de);
2390 if (S_ISDIR(old_inode->i_mode)) {
2392 retval = -ENOTEMPTY;
2393 if (!empty_dir (new_inode))
2397 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2400 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2403 if (!new_inode && new_dir!=old_dir &&
2404 new_dir->i_nlink >= EXT3_LINK_MAX)
2408 retval = ext3_add_entry (handle, new_dentry, old_inode);
2412 BUFFER_TRACE(new_bh, "get write access");
2413 retval = ext3_journal_get_write_access(handle, new_bh);
2416 new_de->inode = cpu_to_le32(old_inode->i_ino);
2417 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2418 EXT3_FEATURE_INCOMPAT_FILETYPE))
2419 new_de->file_type = old_de->file_type;
2420 new_dir->i_version++;
2421 new_dir->i_ctime = new_dir->i_mtime = CURRENT_TIME_SEC;
2422 ext3_mark_inode_dirty(handle, new_dir);
2423 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2424 retval = ext3_journal_dirty_metadata(handle, new_bh);
2432 * Like most other Unix systems, set the ctime for inodes on a
2435 old_inode->i_ctime = CURRENT_TIME_SEC;
2436 ext3_mark_inode_dirty(handle, old_inode);
2441 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2442 old_de->name_len != old_dentry->d_name.len ||
2443 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2444 (retval = ext3_delete_entry(handle, old_dir,
2445 old_de, old_bh)) == -ENOENT) {
2446 /* old_de could have moved from under us during htree split, so
2447 * make sure that we are deleting the right entry. We might
2448 * also be pointing to a stale entry in the unused part of
2449 * old_bh so just checking inum and the name isn't enough. */
2450 struct buffer_head *old_bh2;
2451 struct ext3_dir_entry_2 *old_de2;
2453 old_bh2 = ext3_find_entry(old_dir, &old_dentry->d_name,
2456 retval = ext3_delete_entry(handle, old_dir,
2462 ext3_warning(old_dir->i_sb, "ext3_rename",
2463 "Deleting old file (%lu), %d, error=%d",
2464 old_dir->i_ino, old_dir->i_nlink, retval);
2468 drop_nlink(new_inode);
2469 new_inode->i_ctime = CURRENT_TIME_SEC;
2471 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2472 ext3_update_dx_flag(old_dir);
2474 BUFFER_TRACE(dir_bh, "get_write_access");
2475 retval = ext3_journal_get_write_access(handle, dir_bh);
2478 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2479 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2480 retval = ext3_journal_dirty_metadata(handle, dir_bh);
2483 ext3_std_error(new_dir->i_sb, retval);
2486 drop_nlink(old_dir);
2488 drop_nlink(new_inode);
2491 ext3_update_dx_flag(new_dir);
2492 ext3_mark_inode_dirty(handle, new_dir);
2495 ext3_mark_inode_dirty(handle, old_dir);
2497 ext3_mark_inode_dirty(handle, new_inode);
2498 if (!new_inode->i_nlink)
2499 ext3_orphan_add(handle, new_inode);
2500 if (ext3_should_writeback_data(new_inode))
2509 ext3_journal_stop(handle);
2510 if (retval == 0 && flush_file)
2511 filemap_flush(old_inode->i_mapping);
2516 * directories can handle most operations...
2518 const struct inode_operations ext3_dir_inode_operations = {
2519 .create = ext3_create,
2520 .lookup = ext3_lookup,
2522 .unlink = ext3_unlink,
2523 .symlink = ext3_symlink,
2524 .mkdir = ext3_mkdir,
2525 .rmdir = ext3_rmdir,
2526 .mknod = ext3_mknod,
2527 .rename = ext3_rename,
2528 .setattr = ext3_setattr,
2529 #ifdef CONFIG_EXT3_FS_XATTR
2530 .setxattr = generic_setxattr,
2531 .getxattr = generic_getxattr,
2532 .listxattr = ext3_listxattr,
2533 .removexattr = generic_removexattr,
2535 .get_acl = ext3_get_acl,
2538 const struct inode_operations ext3_special_inode_operations = {
2539 .setattr = ext3_setattr,
2540 #ifdef CONFIG_EXT3_FS_XATTR
2541 .setxattr = generic_setxattr,
2542 .getxattr = generic_getxattr,
2543 .listxattr = ext3_listxattr,
2544 .removexattr = generic_removexattr,
2546 .get_acl = ext3_get_acl,