2 * linux/fs/ext4/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/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
38 #include "ext4_jbd2.h"
45 * define how far ahead to read directories while searching them.
47 #define NAMEI_RA_CHUNKS 2
48 #define NAMEI_RA_BLOCKS 4
49 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
52 static struct buffer_head *ext4_append(handle_t *handle,
54 ext4_lblk_t *block, int *err)
56 struct buffer_head *bh;
58 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
60 bh = ext4_bread(handle, inode, *block, 1, err);
62 inode->i_size += inode->i_sb->s_blocksize;
63 EXT4_I(inode)->i_disksize = inode->i_size;
64 *err = ext4_journal_get_write_access(handle, bh);
74 #define assert(test) J_ASSERT(test)
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 __le32 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;
148 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
149 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t 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(const struct qstr *d_name,
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 ext4_dir_entry_2 *de, unsigned blocksize,
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 ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
168 struct dx_map_entry *offsets, int count, unsigned blocksize);
169 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
170 static void dx_insert_block(struct dx_frame *frame,
171 u32 hash, ext4_lblk_t block);
172 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
173 struct dx_frame *frame,
174 struct dx_frame *frames,
176 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
177 const struct qstr *d_name,
178 struct ext4_dir_entry_2 **res_dir,
180 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
181 struct inode *inode);
183 unsigned int ext4_rec_len_from_disk(__le16 dlen, unsigned blocksize)
185 unsigned len = le16_to_cpu(dlen);
187 if (len == EXT4_MAX_REC_LEN || len == 0)
189 return (len & 65532) | ((len & 3) << 16);
192 __le16 ext4_rec_len_to_disk(unsigned len, unsigned blocksize)
194 if ((len > blocksize) || (blocksize > (1 << 18)) || (len & 3))
197 return cpu_to_le16(len);
198 if (len == blocksize) {
199 if (blocksize == 65536)
200 return cpu_to_le16(EXT4_MAX_REC_LEN);
202 return cpu_to_le16(0);
204 return cpu_to_le16((len & 65532) | ((len >> 16) & 3));
208 * p is at least 6 bytes before the end of page
210 static inline struct ext4_dir_entry_2 *
211 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
213 return (struct ext4_dir_entry_2 *)((char *)p +
214 ext4_rec_len_from_disk(p->rec_len, blocksize));
218 * Future: use high four bits of block for coalesce-on-delete flags
219 * Mask them off for now.
222 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
224 return le32_to_cpu(entry->block) & 0x00ffffff;
227 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
229 entry->block = cpu_to_le32(value);
232 static inline unsigned dx_get_hash(struct dx_entry *entry)
234 return le32_to_cpu(entry->hash);
237 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
239 entry->hash = cpu_to_le32(value);
242 static inline unsigned dx_get_count(struct dx_entry *entries)
244 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
247 static inline unsigned dx_get_limit(struct dx_entry *entries)
249 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
252 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
254 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
257 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
259 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
262 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
264 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
265 EXT4_DIR_REC_LEN(2) - infosize;
266 return entry_space / sizeof(struct dx_entry);
269 static inline unsigned dx_node_limit(struct inode *dir)
271 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
272 return entry_space / sizeof(struct dx_entry);
279 static void dx_show_index(char * label, struct dx_entry *entries)
281 int i, n = dx_get_count (entries);
282 printk(KERN_DEBUG "%s index ", label);
283 for (i = 0; i < n; i++) {
284 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
285 0, (unsigned long)dx_get_block(entries + i));
297 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
298 int size, int show_names)
300 unsigned names = 0, space = 0;
301 char *base = (char *) de;
302 struct dx_hash_info h = *hinfo;
305 while ((char *) de < base + size)
311 int len = de->name_len;
312 char *name = de->name;
313 while (len--) printk("%c", *name++);
314 ext4fs_dirhash(de->name, de->name_len, &h);
315 printk(":%x.%u ", h.hash,
316 ((char *) de - base));
318 space += EXT4_DIR_REC_LEN(de->name_len);
321 de = ext4_next_entry(de, size);
323 printk("(%i)\n", names);
324 return (struct stats) { names, space, 1 };
327 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
328 struct dx_entry *entries, int levels)
330 unsigned blocksize = dir->i_sb->s_blocksize;
331 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
333 struct buffer_head *bh;
335 printk("%i indexed blocks...\n", count);
336 for (i = 0; i < count; i++, entries++)
338 ext4_lblk_t block = dx_get_block(entries);
339 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
340 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
342 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
343 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
345 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
346 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
347 names += stats.names;
348 space += stats.space;
349 bcount += stats.bcount;
353 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
354 levels ? "" : " ", names, space/bcount,
355 (space/bcount)*100/blocksize);
356 return (struct stats) { names, space, bcount};
358 #endif /* DX_DEBUG */
361 * Probe for a directory leaf block to search.
363 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
364 * error in the directory index, and the caller should fall back to
365 * searching the directory normally. The callers of dx_probe **MUST**
366 * check for this error code, and make sure it never gets reflected
369 static struct dx_frame *
370 dx_probe(const struct qstr *d_name, struct inode *dir,
371 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
373 unsigned count, indirect;
374 struct dx_entry *at, *entries, *p, *q, *m;
375 struct dx_root *root;
376 struct buffer_head *bh;
377 struct dx_frame *frame = frame_in;
381 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
383 root = (struct dx_root *) bh->b_data;
384 if (root->info.hash_version != DX_HASH_TEA &&
385 root->info.hash_version != DX_HASH_HALF_MD4 &&
386 root->info.hash_version != DX_HASH_LEGACY) {
387 ext4_warning(dir->i_sb, __func__,
388 "Unrecognised inode hash code %d",
389 root->info.hash_version);
391 *err = ERR_BAD_DX_DIR;
394 hinfo->hash_version = root->info.hash_version;
395 if (hinfo->hash_version <= DX_HASH_TEA)
396 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
397 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
399 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
402 if (root->info.unused_flags & 1) {
403 ext4_warning(dir->i_sb, __func__,
404 "Unimplemented inode hash flags: %#06x",
405 root->info.unused_flags);
407 *err = ERR_BAD_DX_DIR;
411 if ((indirect = root->info.indirect_levels) > 1) {
412 ext4_warning(dir->i_sb, __func__,
413 "Unimplemented inode hash depth: %#06x",
414 root->info.indirect_levels);
416 *err = ERR_BAD_DX_DIR;
420 entries = (struct dx_entry *) (((char *)&root->info) +
421 root->info.info_length);
423 if (dx_get_limit(entries) != dx_root_limit(dir,
424 root->info.info_length)) {
425 ext4_warning(dir->i_sb, __func__,
426 "dx entry: limit != root limit");
428 *err = ERR_BAD_DX_DIR;
432 dxtrace(printk("Look up %x", hash));
435 count = dx_get_count(entries);
436 if (!count || count > dx_get_limit(entries)) {
437 ext4_warning(dir->i_sb, __func__,
438 "dx entry: no count or count > limit");
440 *err = ERR_BAD_DX_DIR;
445 q = entries + count - 1;
449 dxtrace(printk("."));
450 if (dx_get_hash(m) > hash)
456 if (0) // linear search cross check
458 unsigned n = count - 1;
462 dxtrace(printk(","));
463 if (dx_get_hash(++at) > hash)
469 assert (at == p - 1);
473 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
475 frame->entries = entries;
477 if (!indirect--) return frame;
478 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
480 at = entries = ((struct dx_node *) bh->b_data)->entries;
481 if (dx_get_limit(entries) != dx_node_limit (dir)) {
482 ext4_warning(dir->i_sb, __func__,
483 "dx entry: limit != node limit");
485 *err = ERR_BAD_DX_DIR;
492 while (frame >= frame_in) {
497 if (*err == ERR_BAD_DX_DIR)
498 ext4_warning(dir->i_sb, __func__,
499 "Corrupt dir inode %ld, running e2fsck is "
500 "recommended.", dir->i_ino);
504 static void dx_release (struct dx_frame *frames)
506 if (frames[0].bh == NULL)
509 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
510 brelse(frames[1].bh);
511 brelse(frames[0].bh);
515 * This function increments the frame pointer to search the next leaf
516 * block, and reads in the necessary intervening nodes if the search
517 * should be necessary. Whether or not the search is necessary is
518 * controlled by the hash parameter. If the hash value is even, then
519 * the search is only continued if the next block starts with that
520 * hash value. This is used if we are searching for a specific file.
522 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
524 * This function returns 1 if the caller should continue to search,
525 * or 0 if it should not. If there is an error reading one of the
526 * index blocks, it will a negative error code.
528 * If start_hash is non-null, it will be filled in with the starting
529 * hash of the next page.
531 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
532 struct dx_frame *frame,
533 struct dx_frame *frames,
537 struct buffer_head *bh;
538 int err, num_frames = 0;
543 * Find the next leaf page by incrementing the frame pointer.
544 * If we run out of entries in the interior node, loop around and
545 * increment pointer in the parent node. When we break out of
546 * this loop, num_frames indicates the number of interior
547 * nodes need to be read.
550 if (++(p->at) < p->entries + dx_get_count(p->entries))
559 * If the hash is 1, then continue only if the next page has a
560 * continuation hash of any value. This is used for readdir
561 * handling. Otherwise, check to see if the hash matches the
562 * desired contiuation hash. If it doesn't, return since
563 * there's no point to read in the successive index pages.
565 bhash = dx_get_hash(p->at);
568 if ((hash & 1) == 0) {
569 if ((bhash & ~1) != hash)
573 * If the hash is HASH_NB_ALWAYS, we always go to the next
574 * block so no check is necessary
576 while (num_frames--) {
577 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
579 return err; /* Failure */
583 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
590 * This function fills a red-black tree with information from a
591 * directory block. It returns the number directory entries loaded
592 * into the tree. If there is an error it is returned in err.
594 static int htree_dirblock_to_tree(struct file *dir_file,
595 struct inode *dir, ext4_lblk_t block,
596 struct dx_hash_info *hinfo,
597 __u32 start_hash, __u32 start_minor_hash)
599 struct buffer_head *bh;
600 struct ext4_dir_entry_2 *de, *top;
603 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
604 (unsigned long)block));
605 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
608 de = (struct ext4_dir_entry_2 *) bh->b_data;
609 top = (struct ext4_dir_entry_2 *) ((char *) de +
610 dir->i_sb->s_blocksize -
611 EXT4_DIR_REC_LEN(0));
612 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
613 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
614 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
615 +((char *)de - bh->b_data))) {
616 /* On error, skip the f_pos to the next block. */
617 dir_file->f_pos = (dir_file->f_pos |
618 (dir->i_sb->s_blocksize - 1)) + 1;
622 ext4fs_dirhash(de->name, de->name_len, hinfo);
623 if ((hinfo->hash < start_hash) ||
624 ((hinfo->hash == start_hash) &&
625 (hinfo->minor_hash < start_minor_hash)))
629 if ((err = ext4_htree_store_dirent(dir_file,
630 hinfo->hash, hinfo->minor_hash, de)) != 0) {
642 * This function fills a red-black tree with information from a
643 * directory. We start scanning the directory in hash order, starting
644 * at start_hash and start_minor_hash.
646 * This function returns the number of entries inserted into the tree,
647 * or a negative error code.
649 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
650 __u32 start_minor_hash, __u32 *next_hash)
652 struct dx_hash_info hinfo;
653 struct ext4_dir_entry_2 *de;
654 struct dx_frame frames[2], *frame;
661 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
662 start_hash, start_minor_hash));
663 dir = dir_file->f_path.dentry->d_inode;
664 if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
665 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
666 if (hinfo.hash_version <= DX_HASH_TEA)
667 hinfo.hash_version +=
668 EXT4_SB(dir->i_sb)->s_hash_unsigned;
669 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
670 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
671 start_hash, start_minor_hash);
675 hinfo.hash = start_hash;
676 hinfo.minor_hash = 0;
677 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
681 /* Add '.' and '..' from the htree header */
682 if (!start_hash && !start_minor_hash) {
683 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
684 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
688 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
689 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
690 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
691 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
697 block = dx_get_block(frame->at);
698 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
699 start_hash, start_minor_hash);
706 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
707 frame, frames, &hashval);
708 *next_hash = hashval;
714 * Stop if: (a) there are no more entries, or
715 * (b) we have inserted at least one entry and the
716 * next hash value is not a continuation
719 (count && ((hashval & 1) == 0)))
723 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
724 "next hash: %x\n", count, *next_hash));
733 * Directory block splitting, compacting
737 * Create map of hash values, offsets, and sizes, stored at end of block.
738 * Returns number of entries mapped.
740 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
741 struct dx_hash_info *hinfo,
742 struct dx_map_entry *map_tail)
745 char *base = (char *) de;
746 struct dx_hash_info h = *hinfo;
748 while ((char *) de < base + blocksize) {
749 if (de->name_len && de->inode) {
750 ext4fs_dirhash(de->name, de->name_len, &h);
752 map_tail->hash = h.hash;
753 map_tail->offs = (u16) ((char *) de - base);
754 map_tail->size = le16_to_cpu(de->rec_len);
758 /* XXX: do we need to check rec_len == 0 case? -Chris */
759 de = ext4_next_entry(de, blocksize);
764 /* Sort map by hash value */
765 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
767 struct dx_map_entry *p, *q, *top = map + count - 1;
769 /* Combsort until bubble sort doesn't suck */
772 if (count - 9 < 2) /* 9, 10 -> 11 */
774 for (p = top, q = p - count; q >= map; p--, q--)
775 if (p->hash < q->hash)
778 /* Garden variety bubble sort */
783 if (q[1].hash >= q[0].hash)
791 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
793 struct dx_entry *entries = frame->entries;
794 struct dx_entry *old = frame->at, *new = old + 1;
795 int count = dx_get_count(entries);
797 assert(count < dx_get_limit(entries));
798 assert(old < entries + count);
799 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
800 dx_set_hash(new, hash);
801 dx_set_block(new, block);
802 dx_set_count(entries, count + 1);
805 static void ext4_update_dx_flag(struct inode *inode)
807 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
808 EXT4_FEATURE_COMPAT_DIR_INDEX))
809 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
813 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
815 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
816 * `de != NULL' is guaranteed by caller.
818 static inline int ext4_match (int len, const char * const name,
819 struct ext4_dir_entry_2 * de)
821 if (len != de->name_len)
825 return !memcmp(name, de->name, len);
829 * Returns 0 if not found, -1 on failure, and 1 on success
831 static inline int search_dirblock(struct buffer_head *bh,
833 const struct qstr *d_name,
835 struct ext4_dir_entry_2 ** res_dir)
837 struct ext4_dir_entry_2 * de;
840 const char *name = d_name->name;
841 int namelen = d_name->len;
843 de = (struct ext4_dir_entry_2 *) bh->b_data;
844 dlimit = bh->b_data + dir->i_sb->s_blocksize;
845 while ((char *) de < dlimit) {
846 /* this code is executed quadratically often */
847 /* do minimal checking `by hand' */
849 if ((char *) de + namelen <= dlimit &&
850 ext4_match (namelen, name, de)) {
851 /* found a match - just to be sure, do a full check */
852 if (!ext4_check_dir_entry("ext4_find_entry",
853 dir, de, bh, offset))
858 /* prevent looping on a bad block */
859 de_len = ext4_rec_len_from_disk(de->rec_len,
860 dir->i_sb->s_blocksize);
864 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
873 * finds an entry in the specified directory with the wanted name. It
874 * returns the cache buffer in which the entry was found, and the entry
875 * itself (as a parameter - res_dir). It does NOT read the inode of the
876 * entry - you'll have to do that yourself if you want to.
878 * The returned buffer_head has ->b_count elevated. The caller is expected
879 * to brelse() it when appropriate.
881 static struct buffer_head * ext4_find_entry (struct inode *dir,
882 const struct qstr *d_name,
883 struct ext4_dir_entry_2 ** res_dir)
885 struct super_block *sb;
886 struct buffer_head *bh_use[NAMEI_RA_SIZE];
887 struct buffer_head *bh, *ret = NULL;
888 ext4_lblk_t start, block, b;
889 int ra_max = 0; /* Number of bh's in the readahead
891 int ra_ptr = 0; /* Current index into readahead
900 namelen = d_name->len;
901 if (namelen > EXT4_NAME_LEN)
904 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
906 * On success, or if the error was file not found,
907 * return. Otherwise, fall back to doing a search the
910 if (bh || (err != ERR_BAD_DX_DIR))
912 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
915 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
916 start = EXT4_I(dir)->i_dir_start_lookup;
917 if (start >= nblocks)
923 * We deal with the read-ahead logic here.
925 if (ra_ptr >= ra_max) {
926 /* Refill the readahead buffer */
929 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
931 * Terminate if we reach the end of the
932 * directory and must wrap, or if our
933 * search has finished at this block.
935 if (b >= nblocks || (num && block == start)) {
936 bh_use[ra_max] = NULL;
940 bh = ext4_getblk(NULL, dir, b++, 0, &err);
943 ll_rw_block(READ_META, 1, &bh);
946 if ((bh = bh_use[ra_ptr++]) == NULL)
949 if (!buffer_uptodate(bh)) {
950 /* read error, skip block & hope for the best */
951 ext4_error(sb, __func__, "reading directory #%lu "
952 "offset %lu", dir->i_ino,
953 (unsigned long)block);
957 i = search_dirblock(bh, dir, d_name,
958 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
960 EXT4_I(dir)->i_dir_start_lookup = block;
962 goto cleanup_and_exit;
966 goto cleanup_and_exit;
969 if (++block >= nblocks)
971 } while (block != start);
974 * If the directory has grown while we were searching, then
975 * search the last part of the directory before giving up.
978 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
979 if (block < nblocks) {
985 /* Clean up the read-ahead blocks */
986 for (; ra_ptr < ra_max; ra_ptr++)
987 brelse(bh_use[ra_ptr]);
991 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
992 struct ext4_dir_entry_2 **res_dir, int *err)
994 struct super_block * sb;
995 struct dx_hash_info hinfo;
997 struct dx_frame frames[2], *frame;
998 struct ext4_dir_entry_2 *de, *top;
999 struct buffer_head *bh;
1002 int namelen = d_name->len;
1003 const u8 *name = d_name->name;
1006 /* NFS may look up ".." - look at dx_root directory block */
1007 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
1008 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1012 frame->bh = NULL; /* for dx_release() */
1013 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
1014 dx_set_block(frame->at, 0); /* dx_root block is 0 */
1018 block = dx_get_block(frame->at);
1019 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
1021 de = (struct ext4_dir_entry_2 *) bh->b_data;
1022 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
1023 EXT4_DIR_REC_LEN(0));
1024 for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) {
1025 int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
1026 + ((char *) de - bh->b_data);
1028 if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) {
1030 *err = ERR_BAD_DX_DIR;
1034 if (ext4_match(namelen, name, de)) {
1041 /* Check to see if we should continue to search */
1042 retval = ext4_htree_next_block(dir, hash, frame,
1045 ext4_warning(sb, __func__,
1046 "error reading index page in directory #%lu",
1051 } while (retval == 1);
1055 dxtrace(printk(KERN_DEBUG "%s not found\n", name));
1056 dx_release (frames);
1060 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1062 struct inode *inode;
1063 struct ext4_dir_entry_2 *de;
1064 struct buffer_head *bh;
1066 if (dentry->d_name.len > EXT4_NAME_LEN)
1067 return ERR_PTR(-ENAMETOOLONG);
1069 bh = ext4_find_entry(dir, &dentry->d_name, &de);
1072 __u32 ino = le32_to_cpu(de->inode);
1074 if (!ext4_valid_inum(dir->i_sb, ino)) {
1075 ext4_error(dir->i_sb, "ext4_lookup",
1076 "bad inode number: %u", ino);
1077 return ERR_PTR(-EIO);
1079 inode = ext4_iget(dir->i_sb, ino);
1081 return ERR_CAST(inode);
1083 return d_splice_alias(inode, dentry);
1087 struct dentry *ext4_get_parent(struct dentry *child)
1090 struct inode *inode;
1091 static const struct qstr dotdot = {
1095 struct ext4_dir_entry_2 * de;
1096 struct buffer_head *bh;
1098 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1101 return ERR_PTR(-ENOENT);
1102 ino = le32_to_cpu(de->inode);
1105 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1106 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1107 "bad inode number: %u", ino);
1108 return ERR_PTR(-EIO);
1111 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1115 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1116 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1117 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1118 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1119 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1120 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1121 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1122 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
1125 static inline void ext4_set_de_type(struct super_block *sb,
1126 struct ext4_dir_entry_2 *de,
1128 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1129 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1133 * Move count entries from end of map between two memory locations.
1134 * Returns pointer to last entry moved.
1136 static struct ext4_dir_entry_2 *
1137 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1140 unsigned rec_len = 0;
1143 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1144 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1145 memcpy (to, de, rec_len);
1146 ((struct ext4_dir_entry_2 *) to)->rec_len =
1147 ext4_rec_len_to_disk(rec_len, blocksize);
1152 return (struct ext4_dir_entry_2 *) (to - rec_len);
1156 * Compact each dir entry in the range to the minimal rec_len.
1157 * Returns pointer to last entry in range.
1159 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1161 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1162 unsigned rec_len = 0;
1165 while ((char*)de < base + blocksize) {
1166 next = ext4_next_entry(de, blocksize);
1167 if (de->inode && de->name_len) {
1168 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1170 memmove(to, de, rec_len);
1171 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1173 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1181 * Split a full leaf block to make room for a new dir entry.
1182 * Allocate a new block, and move entries so that they are approx. equally full.
1183 * Returns pointer to de in block into which the new entry will be inserted.
1185 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1186 struct buffer_head **bh,struct dx_frame *frame,
1187 struct dx_hash_info *hinfo, int *error)
1189 unsigned blocksize = dir->i_sb->s_blocksize;
1190 unsigned count, continued;
1191 struct buffer_head *bh2;
1192 ext4_lblk_t newblock;
1194 struct dx_map_entry *map;
1195 char *data1 = (*bh)->b_data, *data2;
1196 unsigned split, move, size;
1197 struct ext4_dir_entry_2 *de = NULL, *de2;
1200 bh2 = ext4_append (handle, dir, &newblock, &err);
1207 BUFFER_TRACE(*bh, "get_write_access");
1208 err = ext4_journal_get_write_access(handle, *bh);
1212 BUFFER_TRACE(frame->bh, "get_write_access");
1213 err = ext4_journal_get_write_access(handle, frame->bh);
1217 data2 = bh2->b_data;
1219 /* create map in the end of data2 block */
1220 map = (struct dx_map_entry *) (data2 + blocksize);
1221 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1222 blocksize, hinfo, map);
1224 dx_sort_map(map, count);
1225 /* Split the existing block in the middle, size-wise */
1228 for (i = count-1; i >= 0; i--) {
1229 /* is more than half of this entry in 2nd half of the block? */
1230 if (size + map[i].size/2 > blocksize/2)
1232 size += map[i].size;
1235 /* map index at which we will split */
1236 split = count - move;
1237 hash2 = map[split].hash;
1238 continued = hash2 == map[split - 1].hash;
1239 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1240 (unsigned long)dx_get_block(frame->at),
1241 hash2, split, count-split));
1243 /* Fancy dance to stay within two buffers */
1244 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1245 de = dx_pack_dirents(data1, blocksize);
1246 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1248 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1250 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1251 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1253 /* Which block gets the new entry? */
1254 if (hinfo->hash >= hash2)
1259 dx_insert_block(frame, hash2 + continued, newblock);
1260 err = ext4_handle_dirty_metadata(handle, dir, bh2);
1263 err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1267 dxtrace(dx_show_index("frame", frame->entries));
1274 ext4_std_error(dir->i_sb, err);
1281 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1282 * it points to a directory entry which is guaranteed to be large
1283 * enough for new directory entry. If de is NULL, then
1284 * add_dirent_to_buf will attempt search the directory block for
1285 * space. It will return -ENOSPC if no space is available, and -EIO
1286 * and -EEXIST if directory entry already exists.
1288 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1289 * all other cases bh is released.
1291 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1292 struct inode *inode, struct ext4_dir_entry_2 *de,
1293 struct buffer_head *bh)
1295 struct inode *dir = dentry->d_parent->d_inode;
1296 const char *name = dentry->d_name.name;
1297 int namelen = dentry->d_name.len;
1298 unsigned int offset = 0;
1299 unsigned int blocksize = dir->i_sb->s_blocksize;
1300 unsigned short reclen;
1301 int nlen, rlen, err;
1304 reclen = EXT4_DIR_REC_LEN(namelen);
1306 de = (struct ext4_dir_entry_2 *)bh->b_data;
1307 top = bh->b_data + blocksize - reclen;
1308 while ((char *) de <= top) {
1309 if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1314 if (ext4_match(namelen, name, de)) {
1318 nlen = EXT4_DIR_REC_LEN(de->name_len);
1319 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1320 if ((de->inode? rlen - nlen: rlen) >= reclen)
1322 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1325 if ((char *) de > top)
1328 BUFFER_TRACE(bh, "get_write_access");
1329 err = ext4_journal_get_write_access(handle, bh);
1331 ext4_std_error(dir->i_sb, err);
1336 /* By now the buffer is marked for journaling */
1337 nlen = EXT4_DIR_REC_LEN(de->name_len);
1338 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1340 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1341 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1342 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1345 de->file_type = EXT4_FT_UNKNOWN;
1347 de->inode = cpu_to_le32(inode->i_ino);
1348 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1351 de->name_len = namelen;
1352 memcpy(de->name, name, namelen);
1354 * XXX shouldn't update any times until successful
1355 * completion of syscall, but too many callers depend
1358 * XXX similarly, too many callers depend on
1359 * ext4_new_inode() setting the times, but error
1360 * recovery deletes the inode, so the worst that can
1361 * happen is that the times are slightly out of date
1362 * and/or different from the directory change time.
1364 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1365 ext4_update_dx_flag(dir);
1367 ext4_mark_inode_dirty(handle, dir);
1368 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1369 err = ext4_handle_dirty_metadata(handle, dir, bh);
1371 ext4_std_error(dir->i_sb, err);
1377 * This converts a one block unindexed directory to a 3 block indexed
1378 * directory, and adds the dentry to the indexed directory.
1380 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1381 struct inode *inode, struct buffer_head *bh)
1383 struct inode *dir = dentry->d_parent->d_inode;
1384 const char *name = dentry->d_name.name;
1385 int namelen = dentry->d_name.len;
1386 struct buffer_head *bh2;
1387 struct dx_root *root;
1388 struct dx_frame frames[2], *frame;
1389 struct dx_entry *entries;
1390 struct ext4_dir_entry_2 *de, *de2;
1395 struct dx_hash_info hinfo;
1397 struct fake_dirent *fde;
1399 blocksize = dir->i_sb->s_blocksize;
1400 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1401 retval = ext4_journal_get_write_access(handle, bh);
1403 ext4_std_error(dir->i_sb, retval);
1407 root = (struct dx_root *) bh->b_data;
1409 /* The 0th block becomes the root, move the dirents out */
1410 fde = &root->dotdot;
1411 de = (struct ext4_dir_entry_2 *)((char *)fde +
1412 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1413 if ((char *) de >= (((char *) root) + blocksize)) {
1414 ext4_error(dir->i_sb, __func__,
1415 "invalid rec_len for '..' in inode %lu",
1420 len = ((char *) root) + blocksize - (char *) de;
1422 /* Allocate new block for the 0th block's dirents */
1423 bh2 = ext4_append(handle, dir, &block, &retval);
1428 EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1429 data1 = bh2->b_data;
1431 memcpy (data1, de, len);
1432 de = (struct ext4_dir_entry_2 *) data1;
1434 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1436 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1438 /* Initialize the root; the dot dirents already exist */
1439 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1440 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1442 memset (&root->info, 0, sizeof(root->info));
1443 root->info.info_length = sizeof(root->info);
1444 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1445 entries = root->entries;
1446 dx_set_block(entries, 1);
1447 dx_set_count(entries, 1);
1448 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1450 /* Initialize as for dx_probe */
1451 hinfo.hash_version = root->info.hash_version;
1452 if (hinfo.hash_version <= DX_HASH_TEA)
1453 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1454 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1455 ext4fs_dirhash(name, namelen, &hinfo);
1457 frame->entries = entries;
1458 frame->at = entries;
1461 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1462 dx_release (frames);
1466 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1472 * adds a file entry to the specified directory, using the same
1473 * semantics as ext4_find_entry(). It returns NULL if it failed.
1475 * NOTE!! The inode part of 'de' is left at 0 - which means you
1476 * may not sleep between calling this and putting something into
1477 * the entry, as someone else might have used it while you slept.
1479 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1480 struct inode *inode)
1482 struct inode *dir = dentry->d_parent->d_inode;
1483 struct buffer_head *bh;
1484 struct ext4_dir_entry_2 *de;
1485 struct super_block *sb;
1489 ext4_lblk_t block, blocks;
1492 blocksize = sb->s_blocksize;
1493 if (!dentry->d_name.len)
1496 retval = ext4_dx_add_entry(handle, dentry, inode);
1497 if (!retval || (retval != ERR_BAD_DX_DIR))
1499 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1501 ext4_mark_inode_dirty(handle, dir);
1503 blocks = dir->i_size >> sb->s_blocksize_bits;
1504 for (block = 0; block < blocks; block++) {
1505 bh = ext4_bread(handle, dir, block, 0, &retval);
1508 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1509 if (retval != -ENOSPC)
1512 if (blocks == 1 && !dx_fallback &&
1513 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1514 return make_indexed_dir(handle, dentry, inode, bh);
1517 bh = ext4_append(handle, dir, &block, &retval);
1520 de = (struct ext4_dir_entry_2 *) bh->b_data;
1522 de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1523 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1527 * Returns 0 for success, or a negative error value
1529 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1530 struct inode *inode)
1532 struct dx_frame frames[2], *frame;
1533 struct dx_entry *entries, *at;
1534 struct dx_hash_info hinfo;
1535 struct buffer_head *bh;
1536 struct inode *dir = dentry->d_parent->d_inode;
1537 struct super_block *sb = dir->i_sb;
1538 struct ext4_dir_entry_2 *de;
1541 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1544 entries = frame->entries;
1547 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1550 BUFFER_TRACE(bh, "get_write_access");
1551 err = ext4_journal_get_write_access(handle, bh);
1555 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1556 if (err != -ENOSPC) {
1561 /* Block full, should compress but for now just split */
1562 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1563 dx_get_count(entries), dx_get_limit(entries)));
1564 /* Need to split index? */
1565 if (dx_get_count(entries) == dx_get_limit(entries)) {
1566 ext4_lblk_t newblock;
1567 unsigned icount = dx_get_count(entries);
1568 int levels = frame - frames;
1569 struct dx_entry *entries2;
1570 struct dx_node *node2;
1571 struct buffer_head *bh2;
1573 if (levels && (dx_get_count(frames->entries) ==
1574 dx_get_limit(frames->entries))) {
1575 ext4_warning(sb, __func__,
1576 "Directory index full!");
1580 bh2 = ext4_append (handle, dir, &newblock, &err);
1583 node2 = (struct dx_node *)(bh2->b_data);
1584 entries2 = node2->entries;
1585 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1587 node2->fake.inode = 0;
1588 BUFFER_TRACE(frame->bh, "get_write_access");
1589 err = ext4_journal_get_write_access(handle, frame->bh);
1593 unsigned icount1 = icount/2, icount2 = icount - icount1;
1594 unsigned hash2 = dx_get_hash(entries + icount1);
1595 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1598 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1599 err = ext4_journal_get_write_access(handle,
1604 memcpy((char *) entries2, (char *) (entries + icount1),
1605 icount2 * sizeof(struct dx_entry));
1606 dx_set_count(entries, icount1);
1607 dx_set_count(entries2, icount2);
1608 dx_set_limit(entries2, dx_node_limit(dir));
1610 /* Which index block gets the new entry? */
1611 if (at - entries >= icount1) {
1612 frame->at = at = at - entries - icount1 + entries2;
1613 frame->entries = entries = entries2;
1614 swap(frame->bh, bh2);
1616 dx_insert_block(frames + 0, hash2, newblock);
1617 dxtrace(dx_show_index("node", frames[1].entries));
1618 dxtrace(dx_show_index("node",
1619 ((struct dx_node *) bh2->b_data)->entries));
1620 err = ext4_handle_dirty_metadata(handle, inode, bh2);
1625 dxtrace(printk(KERN_DEBUG
1626 "Creating second level index...\n"));
1627 memcpy((char *) entries2, (char *) entries,
1628 icount * sizeof(struct dx_entry));
1629 dx_set_limit(entries2, dx_node_limit(dir));
1632 dx_set_count(entries, 1);
1633 dx_set_block(entries + 0, newblock);
1634 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1636 /* Add new access path frame */
1638 frame->at = at = at - entries + entries2;
1639 frame->entries = entries = entries2;
1641 err = ext4_journal_get_write_access(handle,
1646 ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
1648 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1651 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1656 ext4_std_error(dir->i_sb, err);
1665 * ext4_delete_entry deletes a directory entry by merging it with the
1668 static int ext4_delete_entry(handle_t *handle,
1670 struct ext4_dir_entry_2 *de_del,
1671 struct buffer_head *bh)
1673 struct ext4_dir_entry_2 *de, *pde;
1674 unsigned int blocksize = dir->i_sb->s_blocksize;
1679 de = (struct ext4_dir_entry_2 *) bh->b_data;
1680 while (i < bh->b_size) {
1681 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1684 BUFFER_TRACE(bh, "get_write_access");
1685 ext4_journal_get_write_access(handle, bh);
1687 pde->rec_len = ext4_rec_len_to_disk(
1688 ext4_rec_len_from_disk(pde->rec_len,
1690 ext4_rec_len_from_disk(de->rec_len,
1696 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1697 ext4_handle_dirty_metadata(handle, dir, bh);
1700 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
1702 de = ext4_next_entry(de, blocksize);
1708 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1709 * since this indicates that nlinks count was previously 1.
1711 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1714 if (is_dx(inode) && inode->i_nlink > 1) {
1715 /* limit is 16-bit i_links_count */
1716 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1718 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1719 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1725 * If a directory had nlink == 1, then we should let it be 1. This indicates
1726 * directory has >EXT4_LINK_MAX subdirs.
1728 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1731 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1736 static int ext4_add_nondir(handle_t *handle,
1737 struct dentry *dentry, struct inode *inode)
1739 int err = ext4_add_entry(handle, dentry, inode);
1741 ext4_mark_inode_dirty(handle, inode);
1742 d_instantiate(dentry, inode);
1743 unlock_new_inode(inode);
1747 unlock_new_inode(inode);
1753 * By the time this is called, we already have created
1754 * the directory cache entry for the new file, but it
1755 * is so far negative - it has no inode.
1757 * If the create succeeds, we fill in the inode information
1758 * with d_instantiate().
1760 static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1761 struct nameidata *nd)
1764 struct inode *inode;
1765 int err, retries = 0;
1768 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1769 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1770 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1772 return PTR_ERR(handle);
1774 if (IS_DIRSYNC(dir))
1775 ext4_handle_sync(handle);
1777 inode = ext4_new_inode (handle, dir, mode);
1778 err = PTR_ERR(inode);
1779 if (!IS_ERR(inode)) {
1780 inode->i_op = &ext4_file_inode_operations;
1781 inode->i_fop = &ext4_file_operations;
1782 ext4_set_aops(inode);
1783 err = ext4_add_nondir(handle, dentry, inode);
1785 ext4_journal_stop(handle);
1786 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1791 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1792 int mode, dev_t rdev)
1795 struct inode *inode;
1796 int err, retries = 0;
1798 if (!new_valid_dev(rdev))
1802 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1803 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1804 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1806 return PTR_ERR(handle);
1808 if (IS_DIRSYNC(dir))
1809 ext4_handle_sync(handle);
1811 inode = ext4_new_inode(handle, dir, mode);
1812 err = PTR_ERR(inode);
1813 if (!IS_ERR(inode)) {
1814 init_special_inode(inode, inode->i_mode, rdev);
1815 #ifdef CONFIG_EXT4_FS_XATTR
1816 inode->i_op = &ext4_special_inode_operations;
1818 err = ext4_add_nondir(handle, dentry, inode);
1820 ext4_journal_stop(handle);
1821 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1826 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1829 struct inode *inode;
1830 struct buffer_head *dir_block;
1831 struct ext4_dir_entry_2 *de;
1832 unsigned int blocksize = dir->i_sb->s_blocksize;
1833 int err, retries = 0;
1835 if (EXT4_DIR_LINK_MAX(dir))
1839 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1840 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1841 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1843 return PTR_ERR(handle);
1845 if (IS_DIRSYNC(dir))
1846 ext4_handle_sync(handle);
1848 inode = ext4_new_inode(handle, dir, S_IFDIR | mode);
1849 err = PTR_ERR(inode);
1853 inode->i_op = &ext4_dir_inode_operations;
1854 inode->i_fop = &ext4_dir_operations;
1855 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1856 dir_block = ext4_bread(handle, inode, 0, 1, &err);
1858 goto out_clear_inode;
1859 BUFFER_TRACE(dir_block, "get_write_access");
1860 ext4_journal_get_write_access(handle, dir_block);
1861 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1862 de->inode = cpu_to_le32(inode->i_ino);
1864 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1866 strcpy(de->name, ".");
1867 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1868 de = ext4_next_entry(de, blocksize);
1869 de->inode = cpu_to_le32(dir->i_ino);
1870 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1873 strcpy(de->name, "..");
1874 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1876 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1877 ext4_handle_dirty_metadata(handle, dir, dir_block);
1879 ext4_mark_inode_dirty(handle, inode);
1880 err = ext4_add_entry(handle, dentry, inode);
1884 unlock_new_inode(inode);
1885 ext4_mark_inode_dirty(handle, inode);
1889 ext4_inc_count(handle, dir);
1890 ext4_update_dx_flag(dir);
1891 ext4_mark_inode_dirty(handle, dir);
1892 d_instantiate(dentry, inode);
1893 unlock_new_inode(inode);
1895 ext4_journal_stop(handle);
1896 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1902 * routine to check that the specified directory is empty (for rmdir)
1904 static int empty_dir(struct inode *inode)
1906 unsigned int offset;
1907 struct buffer_head *bh;
1908 struct ext4_dir_entry_2 *de, *de1;
1909 struct super_block *sb;
1913 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1914 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1916 ext4_error(inode->i_sb, __func__,
1917 "error %d reading directory #%lu offset 0",
1920 ext4_warning(inode->i_sb, __func__,
1921 "bad directory (dir #%lu) - no data block",
1925 de = (struct ext4_dir_entry_2 *) bh->b_data;
1926 de1 = ext4_next_entry(de, sb->s_blocksize);
1927 if (le32_to_cpu(de->inode) != inode->i_ino ||
1928 !le32_to_cpu(de1->inode) ||
1929 strcmp(".", de->name) ||
1930 strcmp("..", de1->name)) {
1931 ext4_warning(inode->i_sb, "empty_dir",
1932 "bad directory (dir #%lu) - no `.' or `..'",
1937 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1938 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1939 de = ext4_next_entry(de1, sb->s_blocksize);
1940 while (offset < inode->i_size) {
1942 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1945 bh = ext4_bread(NULL, inode,
1946 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1949 ext4_error(sb, __func__,
1950 "error %d reading directory"
1952 err, inode->i_ino, offset);
1953 offset += sb->s_blocksize;
1956 de = (struct ext4_dir_entry_2 *) bh->b_data;
1958 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1959 de = (struct ext4_dir_entry_2 *)(bh->b_data +
1961 offset = (offset | (sb->s_blocksize - 1)) + 1;
1964 if (le32_to_cpu(de->inode)) {
1968 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1969 de = ext4_next_entry(de, sb->s_blocksize);
1975 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1976 * such inodes, starting at the superblock, in case we crash before the
1977 * file is closed/deleted, or in case the inode truncate spans multiple
1978 * transactions and the last transaction is not recovered after a crash.
1980 * At filesystem recovery time, we walk this list deleting unlinked
1981 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1983 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1985 struct super_block *sb = inode->i_sb;
1986 struct ext4_iloc iloc;
1989 if (!ext4_handle_valid(handle))
1993 if (!list_empty(&EXT4_I(inode)->i_orphan))
1996 /* Orphan handling is only valid for files with data blocks
1997 * being truncated, or files being unlinked. */
1999 /* @@@ FIXME: Observation from aviro:
2000 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
2001 * here (on lock_super()), so race with ext4_link() which might bump
2002 * ->i_nlink. For, say it, character device. Not a regular file,
2003 * not a directory, not a symlink and ->i_nlink > 0.
2005 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2006 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2008 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2009 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2013 err = ext4_reserve_inode_write(handle, inode, &iloc);
2017 /* Insert this inode at the head of the on-disk orphan list... */
2018 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2019 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2020 err = ext4_handle_dirty_metadata(handle, inode, EXT4_SB(sb)->s_sbh);
2021 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2025 /* Only add to the head of the in-memory list if all the
2026 * previous operations succeeded. If the orphan_add is going to
2027 * fail (possibly taking the journal offline), we can't risk
2028 * leaving the inode on the orphan list: stray orphan-list
2029 * entries can cause panics at unmount time.
2031 * This is safe: on error we're going to ignore the orphan list
2032 * anyway on the next recovery. */
2034 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2036 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2037 jbd_debug(4, "orphan inode %lu will point to %d\n",
2038 inode->i_ino, NEXT_ORPHAN(inode));
2041 ext4_std_error(inode->i_sb, err);
2046 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2047 * of such inodes stored on disk, because it is finally being cleaned up.
2049 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2051 struct list_head *prev;
2052 struct ext4_inode_info *ei = EXT4_I(inode);
2053 struct ext4_sb_info *sbi;
2055 struct ext4_iloc iloc;
2058 if (!ext4_handle_valid(handle))
2061 lock_super(inode->i_sb);
2062 if (list_empty(&ei->i_orphan)) {
2063 unlock_super(inode->i_sb);
2067 ino_next = NEXT_ORPHAN(inode);
2068 prev = ei->i_orphan.prev;
2069 sbi = EXT4_SB(inode->i_sb);
2071 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2073 list_del_init(&ei->i_orphan);
2075 /* If we're on an error path, we may not have a valid
2076 * transaction handle with which to update the orphan list on
2077 * disk, but we still need to remove the inode from the linked
2078 * list in memory. */
2079 if (sbi->s_journal && !handle)
2082 err = ext4_reserve_inode_write(handle, inode, &iloc);
2086 if (prev == &sbi->s_orphan) {
2087 jbd_debug(4, "superblock will point to %u\n", ino_next);
2088 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2089 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2092 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2093 err = ext4_handle_dirty_metadata(handle, inode, sbi->s_sbh);
2095 struct ext4_iloc iloc2;
2096 struct inode *i_prev =
2097 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2099 jbd_debug(4, "orphan inode %lu will point to %u\n",
2100 i_prev->i_ino, ino_next);
2101 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2104 NEXT_ORPHAN(i_prev) = ino_next;
2105 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2109 NEXT_ORPHAN(inode) = 0;
2110 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2113 ext4_std_error(inode->i_sb, err);
2115 unlock_super(inode->i_sb);
2123 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2126 struct inode *inode;
2127 struct buffer_head *bh;
2128 struct ext4_dir_entry_2 *de;
2131 /* Initialize quotas before so that eventual writes go in
2132 * separate transaction */
2133 vfs_dq_init(dentry->d_inode);
2134 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2136 return PTR_ERR(handle);
2139 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2143 if (IS_DIRSYNC(dir))
2144 ext4_handle_sync(handle);
2146 inode = dentry->d_inode;
2149 if (le32_to_cpu(de->inode) != inode->i_ino)
2152 retval = -ENOTEMPTY;
2153 if (!empty_dir(inode))
2156 retval = ext4_delete_entry(handle, dir, de, bh);
2159 if (!EXT4_DIR_LINK_EMPTY(inode))
2160 ext4_warning(inode->i_sb, "ext4_rmdir",
2161 "empty directory has too many links (%d)",
2165 /* There's no need to set i_disksize: the fact that i_nlink is
2166 * zero will ensure that the right thing happens during any
2169 ext4_orphan_add(handle, inode);
2170 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2171 ext4_mark_inode_dirty(handle, inode);
2172 ext4_dec_count(handle, dir);
2173 ext4_update_dx_flag(dir);
2174 ext4_mark_inode_dirty(handle, dir);
2177 ext4_journal_stop(handle);
2182 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2185 struct inode *inode;
2186 struct buffer_head *bh;
2187 struct ext4_dir_entry_2 *de;
2190 /* Initialize quotas before so that eventual writes go
2191 * in separate transaction */
2192 vfs_dq_init(dentry->d_inode);
2193 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2195 return PTR_ERR(handle);
2197 if (IS_DIRSYNC(dir))
2198 ext4_handle_sync(handle);
2201 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2205 inode = dentry->d_inode;
2208 if (le32_to_cpu(de->inode) != inode->i_ino)
2211 if (!inode->i_nlink) {
2212 ext4_warning(inode->i_sb, "ext4_unlink",
2213 "Deleting nonexistent file (%lu), %d",
2214 inode->i_ino, inode->i_nlink);
2217 retval = ext4_delete_entry(handle, dir, de, bh);
2220 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2221 ext4_update_dx_flag(dir);
2222 ext4_mark_inode_dirty(handle, dir);
2224 if (!inode->i_nlink)
2225 ext4_orphan_add(handle, inode);
2226 inode->i_ctime = ext4_current_time(inode);
2227 ext4_mark_inode_dirty(handle, inode);
2231 ext4_journal_stop(handle);
2236 static int ext4_symlink(struct inode *dir,
2237 struct dentry *dentry, const char *symname)
2240 struct inode *inode;
2241 int l, err, retries = 0;
2243 l = strlen(symname)+1;
2244 if (l > dir->i_sb->s_blocksize)
2245 return -ENAMETOOLONG;
2248 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2249 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2250 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2252 return PTR_ERR(handle);
2254 if (IS_DIRSYNC(dir))
2255 ext4_handle_sync(handle);
2257 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO);
2258 err = PTR_ERR(inode);
2262 if (l > sizeof(EXT4_I(inode)->i_data)) {
2263 inode->i_op = &ext4_symlink_inode_operations;
2264 ext4_set_aops(inode);
2266 * page_symlink() calls into ext4_prepare/commit_write.
2267 * We have a transaction open. All is sweetness. It also sets
2268 * i_size in generic_commit_write().
2270 err = __page_symlink(inode, symname, l, 1);
2273 unlock_new_inode(inode);
2274 ext4_mark_inode_dirty(handle, inode);
2279 /* clear the extent format for fast symlink */
2280 EXT4_I(inode)->i_flags &= ~EXT4_EXTENTS_FL;
2281 inode->i_op = &ext4_fast_symlink_inode_operations;
2282 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2283 inode->i_size = l-1;
2285 EXT4_I(inode)->i_disksize = inode->i_size;
2286 err = ext4_add_nondir(handle, dentry, inode);
2288 ext4_journal_stop(handle);
2289 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2294 static int ext4_link(struct dentry *old_dentry,
2295 struct inode *dir, struct dentry *dentry)
2298 struct inode *inode = old_dentry->d_inode;
2299 int err, retries = 0;
2301 if (EXT4_DIR_LINK_MAX(inode))
2305 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2306 * otherwise has the potential to corrupt the orphan inode list.
2308 if (inode->i_nlink == 0)
2312 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2313 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2315 return PTR_ERR(handle);
2317 if (IS_DIRSYNC(dir))
2318 ext4_handle_sync(handle);
2320 inode->i_ctime = ext4_current_time(inode);
2321 ext4_inc_count(handle, inode);
2322 atomic_inc(&inode->i_count);
2324 err = ext4_add_entry(handle, dentry, inode);
2326 ext4_mark_inode_dirty(handle, inode);
2327 d_instantiate(dentry, inode);
2332 ext4_journal_stop(handle);
2333 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2338 #define PARENT_INO(buffer, size) \
2339 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2342 * Anybody can rename anything with this: the permission checks are left to the
2343 * higher-level routines.
2345 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2346 struct inode *new_dir, struct dentry *new_dentry)
2349 struct inode *old_inode, *new_inode;
2350 struct buffer_head *old_bh, *new_bh, *dir_bh;
2351 struct ext4_dir_entry_2 *old_de, *new_de;
2354 old_bh = new_bh = dir_bh = NULL;
2356 /* Initialize quotas before so that eventual writes go
2357 * in separate transaction */
2358 if (new_dentry->d_inode)
2359 vfs_dq_init(new_dentry->d_inode);
2360 handle = ext4_journal_start(old_dir, 2 *
2361 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2362 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2364 return PTR_ERR(handle);
2366 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2367 ext4_handle_sync(handle);
2369 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2371 * Check for inode number is _not_ due to possible IO errors.
2372 * We might rmdir the source, keep it as pwd of some process
2373 * and merrily kill the link to whatever was created under the
2374 * same name. Goodbye sticky bit ;-<
2376 old_inode = old_dentry->d_inode;
2378 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2381 new_inode = new_dentry->d_inode;
2382 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2389 if (S_ISDIR(old_inode->i_mode)) {
2391 retval = -ENOTEMPTY;
2392 if (!empty_dir(new_inode))
2396 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2399 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2400 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2403 if (!new_inode && new_dir != old_dir &&
2404 new_dir->i_nlink >= EXT4_LINK_MAX)
2408 retval = ext4_add_entry(handle, new_dentry, old_inode);
2412 BUFFER_TRACE(new_bh, "get write access");
2413 ext4_journal_get_write_access(handle, new_bh);
2414 new_de->inode = cpu_to_le32(old_inode->i_ino);
2415 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2416 EXT4_FEATURE_INCOMPAT_FILETYPE))
2417 new_de->file_type = old_de->file_type;
2418 new_dir->i_version++;
2419 new_dir->i_ctime = new_dir->i_mtime =
2420 ext4_current_time(new_dir);
2421 ext4_mark_inode_dirty(handle, new_dir);
2422 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2423 ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2429 * Like most other Unix systems, set the ctime for inodes on a
2432 old_inode->i_ctime = ext4_current_time(old_inode);
2433 ext4_mark_inode_dirty(handle, old_inode);
2438 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2439 old_de->name_len != old_dentry->d_name.len ||
2440 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2441 (retval = ext4_delete_entry(handle, old_dir,
2442 old_de, old_bh)) == -ENOENT) {
2443 /* old_de could have moved from under us during htree split, so
2444 * make sure that we are deleting the right entry. We might
2445 * also be pointing to a stale entry in the unused part of
2446 * old_bh so just checking inum and the name isn't enough. */
2447 struct buffer_head *old_bh2;
2448 struct ext4_dir_entry_2 *old_de2;
2450 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2452 retval = ext4_delete_entry(handle, old_dir,
2458 ext4_warning(old_dir->i_sb, "ext4_rename",
2459 "Deleting old file (%lu), %d, error=%d",
2460 old_dir->i_ino, old_dir->i_nlink, retval);
2464 ext4_dec_count(handle, new_inode);
2465 new_inode->i_ctime = ext4_current_time(new_inode);
2467 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2468 ext4_update_dx_flag(old_dir);
2470 BUFFER_TRACE(dir_bh, "get_write_access");
2471 ext4_journal_get_write_access(handle, dir_bh);
2472 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2473 cpu_to_le32(new_dir->i_ino);
2474 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2475 ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2476 ext4_dec_count(handle, old_dir);
2478 /* checked empty_dir above, can't have another parent,
2479 * ext4_dec_count() won't work for many-linked dirs */
2480 new_inode->i_nlink = 0;
2482 ext4_inc_count(handle, new_dir);
2483 ext4_update_dx_flag(new_dir);
2484 ext4_mark_inode_dirty(handle, new_dir);
2487 ext4_mark_inode_dirty(handle, old_dir);
2489 ext4_mark_inode_dirty(handle, new_inode);
2490 if (!new_inode->i_nlink)
2491 ext4_orphan_add(handle, new_inode);
2499 ext4_journal_stop(handle);
2504 * directories can handle most operations...
2506 const struct inode_operations ext4_dir_inode_operations = {
2507 .create = ext4_create,
2508 .lookup = ext4_lookup,
2510 .unlink = ext4_unlink,
2511 .symlink = ext4_symlink,
2512 .mkdir = ext4_mkdir,
2513 .rmdir = ext4_rmdir,
2514 .mknod = ext4_mknod,
2515 .rename = ext4_rename,
2516 .setattr = ext4_setattr,
2517 #ifdef CONFIG_EXT4_FS_XATTR
2518 .setxattr = generic_setxattr,
2519 .getxattr = generic_getxattr,
2520 .listxattr = ext4_listxattr,
2521 .removexattr = generic_removexattr,
2523 .permission = ext4_permission,
2526 const struct inode_operations ext4_special_inode_operations = {
2527 .setattr = ext4_setattr,
2528 #ifdef CONFIG_EXT4_FS_XATTR
2529 .setxattr = generic_setxattr,
2530 .getxattr = generic_getxattr,
2531 .listxattr = ext4_listxattr,
2532 .removexattr = generic_removexattr,
2534 .permission = ext4_permission,