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"
44 * define how far ahead to read directories while searching them.
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
49 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
51 static struct buffer_head *ext4_append(handle_t *handle,
53 ext4_lblk_t *block, int *err)
55 struct buffer_head *bh;
57 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59 bh = ext4_bread(handle, inode, *block, 1, err);
61 inode->i_size += inode->i_sb->s_blocksize;
62 EXT4_I(inode)->i_disksize = inode->i_size;
63 *err = ext4_journal_get_write_access(handle, bh);
73 #define assert(test) J_ASSERT(test)
77 #define dxtrace(command) command
79 #define dxtrace(command)
103 * dx_root_info is laid out so that if it should somehow get overlaid by a
104 * dirent the two low bits of the hash version will be zero. Therefore, the
105 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
110 struct fake_dirent dot;
112 struct fake_dirent dotdot;
116 __le32 reserved_zero;
118 u8 info_length; /* 8 */
123 struct dx_entry entries[0];
128 struct fake_dirent fake;
129 struct dx_entry entries[0];
135 struct buffer_head *bh;
136 struct dx_entry *entries;
147 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
148 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
149 static inline unsigned dx_get_hash(struct dx_entry *entry);
150 static void dx_set_hash(struct dx_entry *entry, unsigned value);
151 static unsigned dx_get_count(struct dx_entry *entries);
152 static unsigned dx_get_limit(struct dx_entry *entries);
153 static void dx_set_count(struct dx_entry *entries, unsigned value);
154 static void dx_set_limit(struct dx_entry *entries, unsigned value);
155 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
156 static unsigned dx_node_limit(struct inode *dir);
157 static struct dx_frame *dx_probe(const struct qstr *d_name,
159 struct dx_hash_info *hinfo,
160 struct dx_frame *frame,
162 static void dx_release(struct dx_frame *frames);
163 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
164 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
165 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
166 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
167 struct dx_map_entry *offsets, int count, unsigned blocksize);
168 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
169 static void dx_insert_block(struct dx_frame *frame,
170 u32 hash, ext4_lblk_t block);
171 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
172 struct dx_frame *frame,
173 struct dx_frame *frames,
175 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
176 const struct qstr *d_name,
177 struct ext4_dir_entry_2 **res_dir,
179 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
180 struct inode *inode);
182 unsigned int ext4_rec_len_from_disk(__le16 dlen, unsigned blocksize)
184 unsigned len = le16_to_cpu(dlen);
186 if (len == EXT4_MAX_REC_LEN || len == 0)
188 return (len & 65532) | ((len & 3) << 16);
191 __le16 ext4_rec_len_to_disk(unsigned len, unsigned blocksize)
193 if ((len > blocksize) || (blocksize > (1 << 18)) || (len & 3))
196 return cpu_to_le16(len);
197 if (len == blocksize) {
198 if (blocksize == 65536)
199 return cpu_to_le16(EXT4_MAX_REC_LEN);
201 return cpu_to_le16(0);
203 return cpu_to_le16((len & 65532) | ((len >> 16) & 3));
207 * p is at least 6 bytes before the end of page
209 static inline struct ext4_dir_entry_2 *
210 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
212 return (struct ext4_dir_entry_2 *)((char *)p +
213 ext4_rec_len_from_disk(p->rec_len, blocksize));
217 * Future: use high four bits of block for coalesce-on-delete flags
218 * Mask them off for now.
221 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
223 return le32_to_cpu(entry->block) & 0x00ffffff;
226 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
228 entry->block = cpu_to_le32(value);
231 static inline unsigned dx_get_hash(struct dx_entry *entry)
233 return le32_to_cpu(entry->hash);
236 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
238 entry->hash = cpu_to_le32(value);
241 static inline unsigned dx_get_count(struct dx_entry *entries)
243 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
246 static inline unsigned dx_get_limit(struct dx_entry *entries)
248 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
251 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
253 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
256 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
258 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
261 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
263 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
264 EXT4_DIR_REC_LEN(2) - infosize;
265 return entry_space / sizeof(struct dx_entry);
268 static inline unsigned dx_node_limit(struct inode *dir)
270 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
271 return entry_space / sizeof(struct dx_entry);
278 static void dx_show_index(char * label, struct dx_entry *entries)
280 int i, n = dx_get_count (entries);
281 printk(KERN_DEBUG "%s index ", label);
282 for (i = 0; i < n; i++) {
283 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
284 0, (unsigned long)dx_get_block(entries + i));
296 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
297 int size, int show_names)
299 unsigned names = 0, space = 0;
300 char *base = (char *) de;
301 struct dx_hash_info h = *hinfo;
304 while ((char *) de < base + size)
310 int len = de->name_len;
311 char *name = de->name;
312 while (len--) printk("%c", *name++);
313 ext4fs_dirhash(de->name, de->name_len, &h);
314 printk(":%x.%u ", h.hash,
315 ((char *) de - base));
317 space += EXT4_DIR_REC_LEN(de->name_len);
320 de = ext4_next_entry(de, size);
322 printk("(%i)\n", names);
323 return (struct stats) { names, space, 1 };
326 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
327 struct dx_entry *entries, int levels)
329 unsigned blocksize = dir->i_sb->s_blocksize;
330 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
332 struct buffer_head *bh;
334 printk("%i indexed blocks...\n", count);
335 for (i = 0; i < count; i++, entries++)
337 ext4_lblk_t block = dx_get_block(entries);
338 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
339 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
341 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
342 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
344 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
345 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
346 names += stats.names;
347 space += stats.space;
348 bcount += stats.bcount;
352 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
353 levels ? "" : " ", names, space/bcount,
354 (space/bcount)*100/blocksize);
355 return (struct stats) { names, space, bcount};
357 #endif /* DX_DEBUG */
360 * Probe for a directory leaf block to search.
362 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
363 * error in the directory index, and the caller should fall back to
364 * searching the directory normally. The callers of dx_probe **MUST**
365 * check for this error code, and make sure it never gets reflected
368 static struct dx_frame *
369 dx_probe(const struct qstr *d_name, struct inode *dir,
370 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
372 unsigned count, indirect;
373 struct dx_entry *at, *entries, *p, *q, *m;
374 struct dx_root *root;
375 struct buffer_head *bh;
376 struct dx_frame *frame = frame_in;
380 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
382 root = (struct dx_root *) bh->b_data;
383 if (root->info.hash_version != DX_HASH_TEA &&
384 root->info.hash_version != DX_HASH_HALF_MD4 &&
385 root->info.hash_version != DX_HASH_LEGACY) {
386 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
387 root->info.hash_version);
389 *err = ERR_BAD_DX_DIR;
392 hinfo->hash_version = root->info.hash_version;
393 if (hinfo->hash_version <= DX_HASH_TEA)
394 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
395 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
397 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
400 if (root->info.unused_flags & 1) {
401 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
402 root->info.unused_flags);
404 *err = ERR_BAD_DX_DIR;
408 if ((indirect = root->info.indirect_levels) > 1) {
409 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
410 root->info.indirect_levels);
412 *err = ERR_BAD_DX_DIR;
416 entries = (struct dx_entry *) (((char *)&root->info) +
417 root->info.info_length);
419 if (dx_get_limit(entries) != dx_root_limit(dir,
420 root->info.info_length)) {
421 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
423 *err = ERR_BAD_DX_DIR;
427 dxtrace(printk("Look up %x", hash));
430 count = dx_get_count(entries);
431 if (!count || count > dx_get_limit(entries)) {
432 ext4_warning(dir->i_sb,
433 "dx entry: no count or count > limit");
435 *err = ERR_BAD_DX_DIR;
440 q = entries + count - 1;
444 dxtrace(printk("."));
445 if (dx_get_hash(m) > hash)
451 if (0) // linear search cross check
453 unsigned n = count - 1;
457 dxtrace(printk(","));
458 if (dx_get_hash(++at) > hash)
464 assert (at == p - 1);
468 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
470 frame->entries = entries;
472 if (!indirect--) return frame;
473 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
475 at = entries = ((struct dx_node *) bh->b_data)->entries;
476 if (dx_get_limit(entries) != dx_node_limit (dir)) {
477 ext4_warning(dir->i_sb,
478 "dx entry: limit != node limit");
480 *err = ERR_BAD_DX_DIR;
487 while (frame >= frame_in) {
492 if (*err == ERR_BAD_DX_DIR)
493 ext4_warning(dir->i_sb,
494 "Corrupt dir inode %ld, running e2fsck is "
495 "recommended.", dir->i_ino);
499 static void dx_release (struct dx_frame *frames)
501 if (frames[0].bh == NULL)
504 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
505 brelse(frames[1].bh);
506 brelse(frames[0].bh);
510 * This function increments the frame pointer to search the next leaf
511 * block, and reads in the necessary intervening nodes if the search
512 * should be necessary. Whether or not the search is necessary is
513 * controlled by the hash parameter. If the hash value is even, then
514 * the search is only continued if the next block starts with that
515 * hash value. This is used if we are searching for a specific file.
517 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
519 * This function returns 1 if the caller should continue to search,
520 * or 0 if it should not. If there is an error reading one of the
521 * index blocks, it will a negative error code.
523 * If start_hash is non-null, it will be filled in with the starting
524 * hash of the next page.
526 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
527 struct dx_frame *frame,
528 struct dx_frame *frames,
532 struct buffer_head *bh;
533 int err, num_frames = 0;
538 * Find the next leaf page by incrementing the frame pointer.
539 * If we run out of entries in the interior node, loop around and
540 * increment pointer in the parent node. When we break out of
541 * this loop, num_frames indicates the number of interior
542 * nodes need to be read.
545 if (++(p->at) < p->entries + dx_get_count(p->entries))
554 * If the hash is 1, then continue only if the next page has a
555 * continuation hash of any value. This is used for readdir
556 * handling. Otherwise, check to see if the hash matches the
557 * desired contiuation hash. If it doesn't, return since
558 * there's no point to read in the successive index pages.
560 bhash = dx_get_hash(p->at);
563 if ((hash & 1) == 0) {
564 if ((bhash & ~1) != hash)
568 * If the hash is HASH_NB_ALWAYS, we always go to the next
569 * block so no check is necessary
571 while (num_frames--) {
572 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
574 return err; /* Failure */
578 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
585 * This function fills a red-black tree with information from a
586 * directory block. It returns the number directory entries loaded
587 * into the tree. If there is an error it is returned in err.
589 static int htree_dirblock_to_tree(struct file *dir_file,
590 struct inode *dir, ext4_lblk_t block,
591 struct dx_hash_info *hinfo,
592 __u32 start_hash, __u32 start_minor_hash)
594 struct buffer_head *bh;
595 struct ext4_dir_entry_2 *de, *top;
598 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
599 (unsigned long)block));
600 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
603 de = (struct ext4_dir_entry_2 *) bh->b_data;
604 top = (struct ext4_dir_entry_2 *) ((char *) de +
605 dir->i_sb->s_blocksize -
606 EXT4_DIR_REC_LEN(0));
607 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
608 if (!ext4_check_dir_entry(dir, de, bh,
609 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
610 +((char *)de - bh->b_data))) {
611 /* On error, skip the f_pos to the next block. */
612 dir_file->f_pos = (dir_file->f_pos |
613 (dir->i_sb->s_blocksize - 1)) + 1;
617 ext4fs_dirhash(de->name, de->name_len, hinfo);
618 if ((hinfo->hash < start_hash) ||
619 ((hinfo->hash == start_hash) &&
620 (hinfo->minor_hash < start_minor_hash)))
624 if ((err = ext4_htree_store_dirent(dir_file,
625 hinfo->hash, hinfo->minor_hash, de)) != 0) {
637 * This function fills a red-black tree with information from a
638 * directory. We start scanning the directory in hash order, starting
639 * at start_hash and start_minor_hash.
641 * This function returns the number of entries inserted into the tree,
642 * or a negative error code.
644 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
645 __u32 start_minor_hash, __u32 *next_hash)
647 struct dx_hash_info hinfo;
648 struct ext4_dir_entry_2 *de;
649 struct dx_frame frames[2], *frame;
656 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
657 start_hash, start_minor_hash));
658 dir = dir_file->f_path.dentry->d_inode;
659 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
660 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
661 if (hinfo.hash_version <= DX_HASH_TEA)
662 hinfo.hash_version +=
663 EXT4_SB(dir->i_sb)->s_hash_unsigned;
664 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
665 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
666 start_hash, start_minor_hash);
670 hinfo.hash = start_hash;
671 hinfo.minor_hash = 0;
672 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
676 /* Add '.' and '..' from the htree header */
677 if (!start_hash && !start_minor_hash) {
678 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
679 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
683 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
684 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
685 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
686 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
692 block = dx_get_block(frame->at);
693 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
694 start_hash, start_minor_hash);
701 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
702 frame, frames, &hashval);
703 *next_hash = hashval;
709 * Stop if: (a) there are no more entries, or
710 * (b) we have inserted at least one entry and the
711 * next hash value is not a continuation
714 (count && ((hashval & 1) == 0)))
718 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
719 "next hash: %x\n", count, *next_hash));
728 * Directory block splitting, compacting
732 * Create map of hash values, offsets, and sizes, stored at end of block.
733 * Returns number of entries mapped.
735 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
736 struct dx_hash_info *hinfo,
737 struct dx_map_entry *map_tail)
740 char *base = (char *) de;
741 struct dx_hash_info h = *hinfo;
743 while ((char *) de < base + blocksize) {
744 if (de->name_len && de->inode) {
745 ext4fs_dirhash(de->name, de->name_len, &h);
747 map_tail->hash = h.hash;
748 map_tail->offs = ((char *) de - base)>>2;
749 map_tail->size = le16_to_cpu(de->rec_len);
753 /* XXX: do we need to check rec_len == 0 case? -Chris */
754 de = ext4_next_entry(de, blocksize);
759 /* Sort map by hash value */
760 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
762 struct dx_map_entry *p, *q, *top = map + count - 1;
764 /* Combsort until bubble sort doesn't suck */
767 if (count - 9 < 2) /* 9, 10 -> 11 */
769 for (p = top, q = p - count; q >= map; p--, q--)
770 if (p->hash < q->hash)
773 /* Garden variety bubble sort */
778 if (q[1].hash >= q[0].hash)
786 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
788 struct dx_entry *entries = frame->entries;
789 struct dx_entry *old = frame->at, *new = old + 1;
790 int count = dx_get_count(entries);
792 assert(count < dx_get_limit(entries));
793 assert(old < entries + count);
794 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
795 dx_set_hash(new, hash);
796 dx_set_block(new, block);
797 dx_set_count(entries, count + 1);
800 static void ext4_update_dx_flag(struct inode *inode)
802 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
803 EXT4_FEATURE_COMPAT_DIR_INDEX))
804 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
808 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
810 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
811 * `de != NULL' is guaranteed by caller.
813 static inline int ext4_match (int len, const char * const name,
814 struct ext4_dir_entry_2 * de)
816 if (len != de->name_len)
820 return !memcmp(name, de->name, len);
824 * Returns 0 if not found, -1 on failure, and 1 on success
826 static inline int search_dirblock(struct buffer_head *bh,
828 const struct qstr *d_name,
830 struct ext4_dir_entry_2 ** res_dir)
832 struct ext4_dir_entry_2 * de;
835 const char *name = d_name->name;
836 int namelen = d_name->len;
838 de = (struct ext4_dir_entry_2 *) bh->b_data;
839 dlimit = bh->b_data + dir->i_sb->s_blocksize;
840 while ((char *) de < dlimit) {
841 /* this code is executed quadratically often */
842 /* do minimal checking `by hand' */
844 if ((char *) de + namelen <= dlimit &&
845 ext4_match (namelen, name, de)) {
846 /* found a match - just to be sure, do a full check */
847 if (!ext4_check_dir_entry(dir, de, bh, offset))
852 /* prevent looping on a bad block */
853 de_len = ext4_rec_len_from_disk(de->rec_len,
854 dir->i_sb->s_blocksize);
858 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
867 * finds an entry in the specified directory with the wanted name. It
868 * returns the cache buffer in which the entry was found, and the entry
869 * itself (as a parameter - res_dir). It does NOT read the inode of the
870 * entry - you'll have to do that yourself if you want to.
872 * The returned buffer_head has ->b_count elevated. The caller is expected
873 * to brelse() it when appropriate.
875 static struct buffer_head * ext4_find_entry (struct inode *dir,
876 const struct qstr *d_name,
877 struct ext4_dir_entry_2 ** res_dir)
879 struct super_block *sb;
880 struct buffer_head *bh_use[NAMEI_RA_SIZE];
881 struct buffer_head *bh, *ret = NULL;
882 ext4_lblk_t start, block, b;
883 int ra_max = 0; /* Number of bh's in the readahead
885 int ra_ptr = 0; /* Current index into readahead
894 namelen = d_name->len;
895 if (namelen > EXT4_NAME_LEN)
898 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
900 * On success, or if the error was file not found,
901 * return. Otherwise, fall back to doing a search the
904 if (bh || (err != ERR_BAD_DX_DIR))
906 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
909 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
910 start = EXT4_I(dir)->i_dir_start_lookup;
911 if (start >= nblocks)
917 * We deal with the read-ahead logic here.
919 if (ra_ptr >= ra_max) {
920 /* Refill the readahead buffer */
923 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
925 * Terminate if we reach the end of the
926 * directory and must wrap, or if our
927 * search has finished at this block.
929 if (b >= nblocks || (num && block == start)) {
930 bh_use[ra_max] = NULL;
934 bh = ext4_getblk(NULL, dir, b++, 0, &err);
937 ll_rw_block(READ_META, 1, &bh);
940 if ((bh = bh_use[ra_ptr++]) == NULL)
943 if (!buffer_uptodate(bh)) {
944 /* read error, skip block & hope for the best */
945 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
946 (unsigned long) block);
950 i = search_dirblock(bh, dir, d_name,
951 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
953 EXT4_I(dir)->i_dir_start_lookup = block;
955 goto cleanup_and_exit;
959 goto cleanup_and_exit;
962 if (++block >= nblocks)
964 } while (block != start);
967 * If the directory has grown while we were searching, then
968 * search the last part of the directory before giving up.
971 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
972 if (block < nblocks) {
978 /* Clean up the read-ahead blocks */
979 for (; ra_ptr < ra_max; ra_ptr++)
980 brelse(bh_use[ra_ptr]);
984 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
985 struct ext4_dir_entry_2 **res_dir, int *err)
987 struct super_block * sb;
988 struct dx_hash_info hinfo;
990 struct dx_frame frames[2], *frame;
991 struct ext4_dir_entry_2 *de, *top;
992 struct buffer_head *bh;
995 int namelen = d_name->len;
996 const u8 *name = d_name->name;
999 /* NFS may look up ".." - look at dx_root directory block */
1000 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
1001 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1005 frame->bh = NULL; /* for dx_release() */
1006 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
1007 dx_set_block(frame->at, 0); /* dx_root block is 0 */
1011 block = dx_get_block(frame->at);
1012 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
1014 de = (struct ext4_dir_entry_2 *) bh->b_data;
1015 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
1016 EXT4_DIR_REC_LEN(0));
1017 for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) {
1018 int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
1019 + ((char *) de - bh->b_data);
1021 if (!ext4_check_dir_entry(dir, de, bh, off)) {
1023 *err = ERR_BAD_DX_DIR;
1027 if (ext4_match(namelen, name, de)) {
1034 /* Check to see if we should continue to search */
1035 retval = ext4_htree_next_block(dir, hash, frame,
1039 "error reading index page in directory #%lu",
1044 } while (retval == 1);
1048 dxtrace(printk(KERN_DEBUG "%s not found\n", name));
1049 dx_release (frames);
1053 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1055 struct inode *inode;
1056 struct ext4_dir_entry_2 *de;
1057 struct buffer_head *bh;
1059 if (dentry->d_name.len > EXT4_NAME_LEN)
1060 return ERR_PTR(-ENAMETOOLONG);
1062 bh = ext4_find_entry(dir, &dentry->d_name, &de);
1065 __u32 ino = le32_to_cpu(de->inode);
1067 if (!ext4_valid_inum(dir->i_sb, ino)) {
1068 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1069 return ERR_PTR(-EIO);
1071 inode = ext4_iget(dir->i_sb, ino);
1072 if (unlikely(IS_ERR(inode))) {
1073 if (PTR_ERR(inode) == -ESTALE) {
1074 EXT4_ERROR_INODE(dir,
1075 "deleted inode referenced: %u",
1077 return ERR_PTR(-EIO);
1079 return ERR_CAST(inode);
1083 return d_splice_alias(inode, dentry);
1087 struct dentry *ext4_get_parent(struct dentry *child)
1090 static const struct qstr dotdot = {
1094 struct ext4_dir_entry_2 * de;
1095 struct buffer_head *bh;
1097 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1099 return ERR_PTR(-ENOENT);
1100 ino = le32_to_cpu(de->inode);
1103 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1104 EXT4_ERROR_INODE(child->d_inode,
1105 "bad parent inode number: %u", ino);
1106 return ERR_PTR(-EIO);
1109 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1113 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1114 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1115 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1116 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1117 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1118 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1119 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1120 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
1123 static inline void ext4_set_de_type(struct super_block *sb,
1124 struct ext4_dir_entry_2 *de,
1126 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1127 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1131 * Move count entries from end of map between two memory locations.
1132 * Returns pointer to last entry moved.
1134 static struct ext4_dir_entry_2 *
1135 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1138 unsigned rec_len = 0;
1141 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1142 (from + (map->offs<<2));
1143 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1144 memcpy (to, de, rec_len);
1145 ((struct ext4_dir_entry_2 *) to)->rec_len =
1146 ext4_rec_len_to_disk(rec_len, blocksize);
1151 return (struct ext4_dir_entry_2 *) (to - rec_len);
1155 * Compact each dir entry in the range to the minimal rec_len.
1156 * Returns pointer to last entry in range.
1158 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1160 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1161 unsigned rec_len = 0;
1164 while ((char*)de < base + blocksize) {
1165 next = ext4_next_entry(de, blocksize);
1166 if (de->inode && de->name_len) {
1167 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1169 memmove(to, de, rec_len);
1170 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1172 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1180 * Split a full leaf block to make room for a new dir entry.
1181 * Allocate a new block, and move entries so that they are approx. equally full.
1182 * Returns pointer to de in block into which the new entry will be inserted.
1184 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1185 struct buffer_head **bh,struct dx_frame *frame,
1186 struct dx_hash_info *hinfo, int *error)
1188 unsigned blocksize = dir->i_sb->s_blocksize;
1189 unsigned count, continued;
1190 struct buffer_head *bh2;
1191 ext4_lblk_t newblock;
1193 struct dx_map_entry *map;
1194 char *data1 = (*bh)->b_data, *data2;
1195 unsigned split, move, size;
1196 struct ext4_dir_entry_2 *de = NULL, *de2;
1199 bh2 = ext4_append (handle, dir, &newblock, &err);
1206 BUFFER_TRACE(*bh, "get_write_access");
1207 err = ext4_journal_get_write_access(handle, *bh);
1211 BUFFER_TRACE(frame->bh, "get_write_access");
1212 err = ext4_journal_get_write_access(handle, frame->bh);
1216 data2 = bh2->b_data;
1218 /* create map in the end of data2 block */
1219 map = (struct dx_map_entry *) (data2 + blocksize);
1220 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1221 blocksize, hinfo, map);
1223 dx_sort_map(map, count);
1224 /* Split the existing block in the middle, size-wise */
1227 for (i = count-1; i >= 0; i--) {
1228 /* is more than half of this entry in 2nd half of the block? */
1229 if (size + map[i].size/2 > blocksize/2)
1231 size += map[i].size;
1234 /* map index at which we will split */
1235 split = count - move;
1236 hash2 = map[split].hash;
1237 continued = hash2 == map[split - 1].hash;
1238 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1239 (unsigned long)dx_get_block(frame->at),
1240 hash2, split, count-split));
1242 /* Fancy dance to stay within two buffers */
1243 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1244 de = dx_pack_dirents(data1, blocksize);
1245 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1247 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1249 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1250 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1252 /* Which block gets the new entry? */
1253 if (hinfo->hash >= hash2)
1258 dx_insert_block(frame, hash2 + continued, newblock);
1259 err = ext4_handle_dirty_metadata(handle, dir, bh2);
1262 err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1266 dxtrace(dx_show_index("frame", frame->entries));
1273 ext4_std_error(dir->i_sb, err);
1280 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1281 * it points to a directory entry which is guaranteed to be large
1282 * enough for new directory entry. If de is NULL, then
1283 * add_dirent_to_buf will attempt search the directory block for
1284 * space. It will return -ENOSPC if no space is available, and -EIO
1285 * and -EEXIST if directory entry already exists.
1287 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1288 struct inode *inode, struct ext4_dir_entry_2 *de,
1289 struct buffer_head *bh)
1291 struct inode *dir = dentry->d_parent->d_inode;
1292 const char *name = dentry->d_name.name;
1293 int namelen = dentry->d_name.len;
1294 unsigned int offset = 0;
1295 unsigned int blocksize = dir->i_sb->s_blocksize;
1296 unsigned short reclen;
1297 int nlen, rlen, err;
1300 reclen = EXT4_DIR_REC_LEN(namelen);
1302 de = (struct ext4_dir_entry_2 *)bh->b_data;
1303 top = bh->b_data + blocksize - reclen;
1304 while ((char *) de <= top) {
1305 if (!ext4_check_dir_entry(dir, de, bh, offset))
1307 if (ext4_match(namelen, name, de))
1309 nlen = EXT4_DIR_REC_LEN(de->name_len);
1310 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1311 if ((de->inode? rlen - nlen: rlen) >= reclen)
1313 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1316 if ((char *) de > top)
1319 BUFFER_TRACE(bh, "get_write_access");
1320 err = ext4_journal_get_write_access(handle, bh);
1322 ext4_std_error(dir->i_sb, err);
1326 /* By now the buffer is marked for journaling */
1327 nlen = EXT4_DIR_REC_LEN(de->name_len);
1328 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1330 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1331 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1332 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1335 de->file_type = EXT4_FT_UNKNOWN;
1337 de->inode = cpu_to_le32(inode->i_ino);
1338 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1341 de->name_len = namelen;
1342 memcpy(de->name, name, namelen);
1344 * XXX shouldn't update any times until successful
1345 * completion of syscall, but too many callers depend
1348 * XXX similarly, too many callers depend on
1349 * ext4_new_inode() setting the times, but error
1350 * recovery deletes the inode, so the worst that can
1351 * happen is that the times are slightly out of date
1352 * and/or different from the directory change time.
1354 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1355 ext4_update_dx_flag(dir);
1357 ext4_mark_inode_dirty(handle, dir);
1358 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1359 err = ext4_handle_dirty_metadata(handle, dir, bh);
1361 ext4_std_error(dir->i_sb, err);
1366 * This converts a one block unindexed directory to a 3 block indexed
1367 * directory, and adds the dentry to the indexed directory.
1369 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1370 struct inode *inode, struct buffer_head *bh)
1372 struct inode *dir = dentry->d_parent->d_inode;
1373 const char *name = dentry->d_name.name;
1374 int namelen = dentry->d_name.len;
1375 struct buffer_head *bh2;
1376 struct dx_root *root;
1377 struct dx_frame frames[2], *frame;
1378 struct dx_entry *entries;
1379 struct ext4_dir_entry_2 *de, *de2;
1384 struct dx_hash_info hinfo;
1386 struct fake_dirent *fde;
1388 blocksize = dir->i_sb->s_blocksize;
1389 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1390 retval = ext4_journal_get_write_access(handle, bh);
1392 ext4_std_error(dir->i_sb, retval);
1396 root = (struct dx_root *) bh->b_data;
1398 /* The 0th block becomes the root, move the dirents out */
1399 fde = &root->dotdot;
1400 de = (struct ext4_dir_entry_2 *)((char *)fde +
1401 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1402 if ((char *) de >= (((char *) root) + blocksize)) {
1403 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1407 len = ((char *) root) + blocksize - (char *) de;
1409 /* Allocate new block for the 0th block's dirents */
1410 bh2 = ext4_append(handle, dir, &block, &retval);
1415 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1416 data1 = bh2->b_data;
1418 memcpy (data1, de, len);
1419 de = (struct ext4_dir_entry_2 *) data1;
1421 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1423 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1425 /* Initialize the root; the dot dirents already exist */
1426 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1427 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1429 memset (&root->info, 0, sizeof(root->info));
1430 root->info.info_length = sizeof(root->info);
1431 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1432 entries = root->entries;
1433 dx_set_block(entries, 1);
1434 dx_set_count(entries, 1);
1435 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1437 /* Initialize as for dx_probe */
1438 hinfo.hash_version = root->info.hash_version;
1439 if (hinfo.hash_version <= DX_HASH_TEA)
1440 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1441 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1442 ext4fs_dirhash(name, namelen, &hinfo);
1444 frame->entries = entries;
1445 frame->at = entries;
1448 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1449 dx_release (frames);
1453 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1461 * adds a file entry to the specified directory, using the same
1462 * semantics as ext4_find_entry(). It returns NULL if it failed.
1464 * NOTE!! The inode part of 'de' is left at 0 - which means you
1465 * may not sleep between calling this and putting something into
1466 * the entry, as someone else might have used it while you slept.
1468 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1469 struct inode *inode)
1471 struct inode *dir = dentry->d_parent->d_inode;
1472 struct buffer_head *bh;
1473 struct ext4_dir_entry_2 *de;
1474 struct super_block *sb;
1478 ext4_lblk_t block, blocks;
1481 blocksize = sb->s_blocksize;
1482 if (!dentry->d_name.len)
1485 retval = ext4_dx_add_entry(handle, dentry, inode);
1486 if (!retval || (retval != ERR_BAD_DX_DIR))
1488 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1490 ext4_mark_inode_dirty(handle, dir);
1492 blocks = dir->i_size >> sb->s_blocksize_bits;
1493 for (block = 0; block < blocks; block++) {
1494 bh = ext4_bread(handle, dir, block, 0, &retval);
1497 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1498 if (retval != -ENOSPC) {
1503 if (blocks == 1 && !dx_fallback &&
1504 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1505 return make_indexed_dir(handle, dentry, inode, bh);
1508 bh = ext4_append(handle, dir, &block, &retval);
1511 de = (struct ext4_dir_entry_2 *) bh->b_data;
1513 de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1514 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1517 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1522 * Returns 0 for success, or a negative error value
1524 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1525 struct inode *inode)
1527 struct dx_frame frames[2], *frame;
1528 struct dx_entry *entries, *at;
1529 struct dx_hash_info hinfo;
1530 struct buffer_head *bh;
1531 struct inode *dir = dentry->d_parent->d_inode;
1532 struct super_block *sb = dir->i_sb;
1533 struct ext4_dir_entry_2 *de;
1536 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1539 entries = frame->entries;
1542 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1545 BUFFER_TRACE(bh, "get_write_access");
1546 err = ext4_journal_get_write_access(handle, bh);
1550 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1554 /* Block full, should compress but for now just split */
1555 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1556 dx_get_count(entries), dx_get_limit(entries)));
1557 /* Need to split index? */
1558 if (dx_get_count(entries) == dx_get_limit(entries)) {
1559 ext4_lblk_t newblock;
1560 unsigned icount = dx_get_count(entries);
1561 int levels = frame - frames;
1562 struct dx_entry *entries2;
1563 struct dx_node *node2;
1564 struct buffer_head *bh2;
1566 if (levels && (dx_get_count(frames->entries) ==
1567 dx_get_limit(frames->entries))) {
1568 ext4_warning(sb, "Directory index full!");
1572 bh2 = ext4_append (handle, dir, &newblock, &err);
1575 node2 = (struct dx_node *)(bh2->b_data);
1576 entries2 = node2->entries;
1577 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1578 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1580 BUFFER_TRACE(frame->bh, "get_write_access");
1581 err = ext4_journal_get_write_access(handle, frame->bh);
1585 unsigned icount1 = icount/2, icount2 = icount - icount1;
1586 unsigned hash2 = dx_get_hash(entries + icount1);
1587 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1590 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1591 err = ext4_journal_get_write_access(handle,
1596 memcpy((char *) entries2, (char *) (entries + icount1),
1597 icount2 * sizeof(struct dx_entry));
1598 dx_set_count(entries, icount1);
1599 dx_set_count(entries2, icount2);
1600 dx_set_limit(entries2, dx_node_limit(dir));
1602 /* Which index block gets the new entry? */
1603 if (at - entries >= icount1) {
1604 frame->at = at = at - entries - icount1 + entries2;
1605 frame->entries = entries = entries2;
1606 swap(frame->bh, bh2);
1608 dx_insert_block(frames + 0, hash2, newblock);
1609 dxtrace(dx_show_index("node", frames[1].entries));
1610 dxtrace(dx_show_index("node",
1611 ((struct dx_node *) bh2->b_data)->entries));
1612 err = ext4_handle_dirty_metadata(handle, inode, bh2);
1617 dxtrace(printk(KERN_DEBUG
1618 "Creating second level index...\n"));
1619 memcpy((char *) entries2, (char *) entries,
1620 icount * sizeof(struct dx_entry));
1621 dx_set_limit(entries2, dx_node_limit(dir));
1624 dx_set_count(entries, 1);
1625 dx_set_block(entries + 0, newblock);
1626 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1628 /* Add new access path frame */
1630 frame->at = at = at - entries + entries2;
1631 frame->entries = entries = entries2;
1633 err = ext4_journal_get_write_access(handle,
1638 ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
1640 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1643 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1647 ext4_std_error(dir->i_sb, err);
1656 * ext4_delete_entry deletes a directory entry by merging it with the
1659 static int ext4_delete_entry(handle_t *handle,
1661 struct ext4_dir_entry_2 *de_del,
1662 struct buffer_head *bh)
1664 struct ext4_dir_entry_2 *de, *pde;
1665 unsigned int blocksize = dir->i_sb->s_blocksize;
1670 de = (struct ext4_dir_entry_2 *) bh->b_data;
1671 while (i < bh->b_size) {
1672 if (!ext4_check_dir_entry(dir, de, bh, i))
1675 BUFFER_TRACE(bh, "get_write_access");
1676 ext4_journal_get_write_access(handle, bh);
1678 pde->rec_len = ext4_rec_len_to_disk(
1679 ext4_rec_len_from_disk(pde->rec_len,
1681 ext4_rec_len_from_disk(de->rec_len,
1687 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1688 ext4_handle_dirty_metadata(handle, dir, bh);
1691 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
1693 de = ext4_next_entry(de, blocksize);
1699 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1700 * since this indicates that nlinks count was previously 1.
1702 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1705 if (is_dx(inode) && inode->i_nlink > 1) {
1706 /* limit is 16-bit i_links_count */
1707 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1709 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1710 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1716 * If a directory had nlink == 1, then we should let it be 1. This indicates
1717 * directory has >EXT4_LINK_MAX subdirs.
1719 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1722 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1727 static int ext4_add_nondir(handle_t *handle,
1728 struct dentry *dentry, struct inode *inode)
1730 int err = ext4_add_entry(handle, dentry, inode);
1732 ext4_mark_inode_dirty(handle, inode);
1733 d_instantiate(dentry, inode);
1734 unlock_new_inode(inode);
1738 unlock_new_inode(inode);
1744 * By the time this is called, we already have created
1745 * the directory cache entry for the new file, but it
1746 * is so far negative - it has no inode.
1748 * If the create succeeds, we fill in the inode information
1749 * with d_instantiate().
1751 static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1752 struct nameidata *nd)
1755 struct inode *inode;
1756 int err, retries = 0;
1758 dquot_initialize(dir);
1761 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1762 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1763 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1765 return PTR_ERR(handle);
1767 if (IS_DIRSYNC(dir))
1768 ext4_handle_sync(handle);
1770 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1771 err = PTR_ERR(inode);
1772 if (!IS_ERR(inode)) {
1773 inode->i_op = &ext4_file_inode_operations;
1774 inode->i_fop = &ext4_file_operations;
1775 ext4_set_aops(inode);
1776 err = ext4_add_nondir(handle, dentry, inode);
1778 ext4_journal_stop(handle);
1779 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1784 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1785 int mode, dev_t rdev)
1788 struct inode *inode;
1789 int err, retries = 0;
1791 if (!new_valid_dev(rdev))
1794 dquot_initialize(dir);
1797 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1798 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1799 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1801 return PTR_ERR(handle);
1803 if (IS_DIRSYNC(dir))
1804 ext4_handle_sync(handle);
1806 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1807 err = PTR_ERR(inode);
1808 if (!IS_ERR(inode)) {
1809 init_special_inode(inode, inode->i_mode, rdev);
1810 #ifdef CONFIG_EXT4_FS_XATTR
1811 inode->i_op = &ext4_special_inode_operations;
1813 err = ext4_add_nondir(handle, dentry, inode);
1815 ext4_journal_stop(handle);
1816 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1821 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1824 struct inode *inode;
1825 struct buffer_head *dir_block;
1826 struct ext4_dir_entry_2 *de;
1827 unsigned int blocksize = dir->i_sb->s_blocksize;
1828 int err, retries = 0;
1830 if (EXT4_DIR_LINK_MAX(dir))
1833 dquot_initialize(dir);
1836 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1837 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1838 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1840 return PTR_ERR(handle);
1842 if (IS_DIRSYNC(dir))
1843 ext4_handle_sync(handle);
1845 inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1846 &dentry->d_name, 0);
1847 err = PTR_ERR(inode);
1851 inode->i_op = &ext4_dir_inode_operations;
1852 inode->i_fop = &ext4_dir_operations;
1853 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1854 dir_block = ext4_bread(handle, inode, 0, 1, &err);
1856 goto out_clear_inode;
1857 BUFFER_TRACE(dir_block, "get_write_access");
1858 ext4_journal_get_write_access(handle, dir_block);
1859 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1860 de->inode = cpu_to_le32(inode->i_ino);
1862 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1864 strcpy(de->name, ".");
1865 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1866 de = ext4_next_entry(de, blocksize);
1867 de->inode = cpu_to_le32(dir->i_ino);
1868 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1871 strcpy(de->name, "..");
1872 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1874 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1875 ext4_handle_dirty_metadata(handle, dir, dir_block);
1877 ext4_mark_inode_dirty(handle, inode);
1878 err = ext4_add_entry(handle, dentry, inode);
1882 unlock_new_inode(inode);
1883 ext4_mark_inode_dirty(handle, inode);
1887 ext4_inc_count(handle, dir);
1888 ext4_update_dx_flag(dir);
1889 ext4_mark_inode_dirty(handle, dir);
1890 d_instantiate(dentry, inode);
1891 unlock_new_inode(inode);
1893 ext4_journal_stop(handle);
1894 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1900 * routine to check that the specified directory is empty (for rmdir)
1902 static int empty_dir(struct inode *inode)
1904 unsigned int offset;
1905 struct buffer_head *bh;
1906 struct ext4_dir_entry_2 *de, *de1;
1907 struct super_block *sb;
1911 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1912 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1914 EXT4_ERROR_INODE(inode,
1915 "error %d reading directory lblock 0", err);
1917 ext4_warning(inode->i_sb,
1918 "bad directory (dir #%lu) - no data block",
1922 de = (struct ext4_dir_entry_2 *) bh->b_data;
1923 de1 = ext4_next_entry(de, sb->s_blocksize);
1924 if (le32_to_cpu(de->inode) != inode->i_ino ||
1925 !le32_to_cpu(de1->inode) ||
1926 strcmp(".", de->name) ||
1927 strcmp("..", de1->name)) {
1928 ext4_warning(inode->i_sb,
1929 "bad directory (dir #%lu) - no `.' or `..'",
1934 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1935 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1936 de = ext4_next_entry(de1, sb->s_blocksize);
1937 while (offset < inode->i_size) {
1939 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1940 unsigned int lblock;
1943 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1944 bh = ext4_bread(NULL, inode, lblock, 0, &err);
1947 EXT4_ERROR_INODE(inode,
1948 "error %d reading directory "
1949 "lblock %u", err, lblock);
1950 offset += sb->s_blocksize;
1953 de = (struct ext4_dir_entry_2 *) bh->b_data;
1955 if (!ext4_check_dir_entry(inode, de, bh, offset)) {
1956 de = (struct ext4_dir_entry_2 *)(bh->b_data +
1958 offset = (offset | (sb->s_blocksize - 1)) + 1;
1961 if (le32_to_cpu(de->inode)) {
1965 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1966 de = ext4_next_entry(de, sb->s_blocksize);
1972 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1973 * such inodes, starting at the superblock, in case we crash before the
1974 * file is closed/deleted, or in case the inode truncate spans multiple
1975 * transactions and the last transaction is not recovered after a crash.
1977 * At filesystem recovery time, we walk this list deleting unlinked
1978 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1980 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1982 struct super_block *sb = inode->i_sb;
1983 struct ext4_iloc iloc;
1986 if (!ext4_handle_valid(handle))
1989 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
1990 if (!list_empty(&EXT4_I(inode)->i_orphan))
1993 /* Orphan handling is only valid for files with data blocks
1994 * being truncated, or files being unlinked. */
1996 /* @@@ FIXME: Observation from aviro:
1997 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
1998 * here (on s_orphan_lock), so race with ext4_link() which might bump
1999 * ->i_nlink. For, say it, character device. Not a regular file,
2000 * not a directory, not a symlink and ->i_nlink > 0.
2002 * tytso, 4/25/2009: I'm not sure how that could happen;
2003 * shouldn't the fs core protect us from these sort of
2004 * unlink()/link() races?
2006 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2007 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2009 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2010 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2014 err = ext4_reserve_inode_write(handle, inode, &iloc);
2018 * Due to previous errors inode may be already a part of on-disk
2019 * orphan list. If so skip on-disk list modification.
2021 if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2022 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2025 /* Insert this inode at the head of the on-disk orphan list... */
2026 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2027 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2028 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
2029 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2033 /* Only add to the head of the in-memory list if all the
2034 * previous operations succeeded. If the orphan_add is going to
2035 * fail (possibly taking the journal offline), we can't risk
2036 * leaving the inode on the orphan list: stray orphan-list
2037 * entries can cause panics at unmount time.
2039 * This is safe: on error we're going to ignore the orphan list
2040 * anyway on the next recovery. */
2043 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2045 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2046 jbd_debug(4, "orphan inode %lu will point to %d\n",
2047 inode->i_ino, NEXT_ORPHAN(inode));
2049 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2050 ext4_std_error(inode->i_sb, err);
2055 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2056 * of such inodes stored on disk, because it is finally being cleaned up.
2058 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2060 struct list_head *prev;
2061 struct ext4_inode_info *ei = EXT4_I(inode);
2062 struct ext4_sb_info *sbi;
2064 struct ext4_iloc iloc;
2067 /* ext4_handle_valid() assumes a valid handle_t pointer */
2068 if (handle && !ext4_handle_valid(handle))
2071 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2072 if (list_empty(&ei->i_orphan))
2075 ino_next = NEXT_ORPHAN(inode);
2076 prev = ei->i_orphan.prev;
2077 sbi = EXT4_SB(inode->i_sb);
2079 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2081 list_del_init(&ei->i_orphan);
2083 /* If we're on an error path, we may not have a valid
2084 * transaction handle with which to update the orphan list on
2085 * disk, but we still need to remove the inode from the linked
2086 * list in memory. */
2087 if (sbi->s_journal && !handle)
2090 err = ext4_reserve_inode_write(handle, inode, &iloc);
2094 if (prev == &sbi->s_orphan) {
2095 jbd_debug(4, "superblock will point to %u\n", ino_next);
2096 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2097 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2100 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2101 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
2103 struct ext4_iloc iloc2;
2104 struct inode *i_prev =
2105 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2107 jbd_debug(4, "orphan inode %lu will point to %u\n",
2108 i_prev->i_ino, ino_next);
2109 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2112 NEXT_ORPHAN(i_prev) = ino_next;
2113 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2117 NEXT_ORPHAN(inode) = 0;
2118 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2121 ext4_std_error(inode->i_sb, err);
2123 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2131 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2134 struct inode *inode;
2135 struct buffer_head *bh;
2136 struct ext4_dir_entry_2 *de;
2139 /* Initialize quotas before so that eventual writes go in
2140 * separate transaction */
2141 dquot_initialize(dir);
2142 dquot_initialize(dentry->d_inode);
2144 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2146 return PTR_ERR(handle);
2149 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2153 if (IS_DIRSYNC(dir))
2154 ext4_handle_sync(handle);
2156 inode = dentry->d_inode;
2159 if (le32_to_cpu(de->inode) != inode->i_ino)
2162 retval = -ENOTEMPTY;
2163 if (!empty_dir(inode))
2166 retval = ext4_delete_entry(handle, dir, de, bh);
2169 if (!EXT4_DIR_LINK_EMPTY(inode))
2170 ext4_warning(inode->i_sb,
2171 "empty directory has too many links (%d)",
2175 /* There's no need to set i_disksize: the fact that i_nlink is
2176 * zero will ensure that the right thing happens during any
2179 ext4_orphan_add(handle, inode);
2180 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2181 ext4_mark_inode_dirty(handle, inode);
2182 ext4_dec_count(handle, dir);
2183 ext4_update_dx_flag(dir);
2184 ext4_mark_inode_dirty(handle, dir);
2187 ext4_journal_stop(handle);
2192 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2195 struct inode *inode;
2196 struct buffer_head *bh;
2197 struct ext4_dir_entry_2 *de;
2200 /* Initialize quotas before so that eventual writes go
2201 * in separate transaction */
2202 dquot_initialize(dir);
2203 dquot_initialize(dentry->d_inode);
2205 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2207 return PTR_ERR(handle);
2209 if (IS_DIRSYNC(dir))
2210 ext4_handle_sync(handle);
2213 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2217 inode = dentry->d_inode;
2220 if (le32_to_cpu(de->inode) != inode->i_ino)
2223 if (!inode->i_nlink) {
2224 ext4_warning(inode->i_sb,
2225 "Deleting nonexistent file (%lu), %d",
2226 inode->i_ino, inode->i_nlink);
2229 retval = ext4_delete_entry(handle, dir, de, bh);
2232 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2233 ext4_update_dx_flag(dir);
2234 ext4_mark_inode_dirty(handle, dir);
2236 if (!inode->i_nlink)
2237 ext4_orphan_add(handle, inode);
2238 inode->i_ctime = ext4_current_time(inode);
2239 ext4_mark_inode_dirty(handle, inode);
2243 ext4_journal_stop(handle);
2248 static int ext4_symlink(struct inode *dir,
2249 struct dentry *dentry, const char *symname)
2252 struct inode *inode;
2253 int l, err, retries = 0;
2255 l = strlen(symname)+1;
2256 if (l > dir->i_sb->s_blocksize)
2257 return -ENAMETOOLONG;
2259 dquot_initialize(dir);
2262 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2263 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2264 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2266 return PTR_ERR(handle);
2268 if (IS_DIRSYNC(dir))
2269 ext4_handle_sync(handle);
2271 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2272 &dentry->d_name, 0);
2273 err = PTR_ERR(inode);
2277 if (l > sizeof(EXT4_I(inode)->i_data)) {
2278 inode->i_op = &ext4_symlink_inode_operations;
2279 ext4_set_aops(inode);
2281 * page_symlink() calls into ext4_prepare/commit_write.
2282 * We have a transaction open. All is sweetness. It also sets
2283 * i_size in generic_commit_write().
2285 err = __page_symlink(inode, symname, l, 1);
2288 unlock_new_inode(inode);
2289 ext4_mark_inode_dirty(handle, inode);
2294 /* clear the extent format for fast symlink */
2295 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2296 inode->i_op = &ext4_fast_symlink_inode_operations;
2297 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2298 inode->i_size = l-1;
2300 EXT4_I(inode)->i_disksize = inode->i_size;
2301 err = ext4_add_nondir(handle, dentry, inode);
2303 ext4_journal_stop(handle);
2304 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2309 static int ext4_link(struct dentry *old_dentry,
2310 struct inode *dir, struct dentry *dentry)
2313 struct inode *inode = old_dentry->d_inode;
2314 int err, retries = 0;
2316 if (inode->i_nlink >= EXT4_LINK_MAX)
2319 dquot_initialize(dir);
2322 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2323 * otherwise has the potential to corrupt the orphan inode list.
2325 if (inode->i_nlink == 0)
2329 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2330 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2332 return PTR_ERR(handle);
2334 if (IS_DIRSYNC(dir))
2335 ext4_handle_sync(handle);
2337 inode->i_ctime = ext4_current_time(inode);
2338 ext4_inc_count(handle, inode);
2339 atomic_inc(&inode->i_count);
2341 err = ext4_add_entry(handle, dentry, inode);
2343 ext4_mark_inode_dirty(handle, inode);
2344 d_instantiate(dentry, inode);
2349 ext4_journal_stop(handle);
2350 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2355 #define PARENT_INO(buffer, size) \
2356 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2359 * Anybody can rename anything with this: the permission checks are left to the
2360 * higher-level routines.
2362 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2363 struct inode *new_dir, struct dentry *new_dentry)
2366 struct inode *old_inode, *new_inode;
2367 struct buffer_head *old_bh, *new_bh, *dir_bh;
2368 struct ext4_dir_entry_2 *old_de, *new_de;
2369 int retval, force_da_alloc = 0;
2371 dquot_initialize(old_dir);
2372 dquot_initialize(new_dir);
2374 old_bh = new_bh = dir_bh = NULL;
2376 /* Initialize quotas before so that eventual writes go
2377 * in separate transaction */
2378 if (new_dentry->d_inode)
2379 dquot_initialize(new_dentry->d_inode);
2380 handle = ext4_journal_start(old_dir, 2 *
2381 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2382 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2384 return PTR_ERR(handle);
2386 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2387 ext4_handle_sync(handle);
2389 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2391 * Check for inode number is _not_ due to possible IO errors.
2392 * We might rmdir the source, keep it as pwd of some process
2393 * and merrily kill the link to whatever was created under the
2394 * same name. Goodbye sticky bit ;-<
2396 old_inode = old_dentry->d_inode;
2398 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2401 new_inode = new_dentry->d_inode;
2402 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2409 if (S_ISDIR(old_inode->i_mode)) {
2411 retval = -ENOTEMPTY;
2412 if (!empty_dir(new_inode))
2416 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2419 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2420 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2423 if (!new_inode && new_dir != old_dir &&
2424 EXT4_DIR_LINK_MAX(new_dir))
2428 retval = ext4_add_entry(handle, new_dentry, old_inode);
2432 BUFFER_TRACE(new_bh, "get write access");
2433 ext4_journal_get_write_access(handle, new_bh);
2434 new_de->inode = cpu_to_le32(old_inode->i_ino);
2435 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2436 EXT4_FEATURE_INCOMPAT_FILETYPE))
2437 new_de->file_type = old_de->file_type;
2438 new_dir->i_version++;
2439 new_dir->i_ctime = new_dir->i_mtime =
2440 ext4_current_time(new_dir);
2441 ext4_mark_inode_dirty(handle, new_dir);
2442 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2443 ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2449 * Like most other Unix systems, set the ctime for inodes on a
2452 old_inode->i_ctime = ext4_current_time(old_inode);
2453 ext4_mark_inode_dirty(handle, old_inode);
2458 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2459 old_de->name_len != old_dentry->d_name.len ||
2460 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2461 (retval = ext4_delete_entry(handle, old_dir,
2462 old_de, old_bh)) == -ENOENT) {
2463 /* old_de could have moved from under us during htree split, so
2464 * make sure that we are deleting the right entry. We might
2465 * also be pointing to a stale entry in the unused part of
2466 * old_bh so just checking inum and the name isn't enough. */
2467 struct buffer_head *old_bh2;
2468 struct ext4_dir_entry_2 *old_de2;
2470 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2472 retval = ext4_delete_entry(handle, old_dir,
2478 ext4_warning(old_dir->i_sb,
2479 "Deleting old file (%lu), %d, error=%d",
2480 old_dir->i_ino, old_dir->i_nlink, retval);
2484 ext4_dec_count(handle, new_inode);
2485 new_inode->i_ctime = ext4_current_time(new_inode);
2487 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2488 ext4_update_dx_flag(old_dir);
2490 BUFFER_TRACE(dir_bh, "get_write_access");
2491 ext4_journal_get_write_access(handle, dir_bh);
2492 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2493 cpu_to_le32(new_dir->i_ino);
2494 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2495 ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2496 ext4_dec_count(handle, old_dir);
2498 /* checked empty_dir above, can't have another parent,
2499 * ext4_dec_count() won't work for many-linked dirs */
2500 new_inode->i_nlink = 0;
2502 ext4_inc_count(handle, new_dir);
2503 ext4_update_dx_flag(new_dir);
2504 ext4_mark_inode_dirty(handle, new_dir);
2507 ext4_mark_inode_dirty(handle, old_dir);
2509 ext4_mark_inode_dirty(handle, new_inode);
2510 if (!new_inode->i_nlink)
2511 ext4_orphan_add(handle, new_inode);
2512 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2521 ext4_journal_stop(handle);
2522 if (retval == 0 && force_da_alloc)
2523 ext4_alloc_da_blocks(old_inode);
2528 * directories can handle most operations...
2530 const struct inode_operations ext4_dir_inode_operations = {
2531 .create = ext4_create,
2532 .lookup = ext4_lookup,
2534 .unlink = ext4_unlink,
2535 .symlink = ext4_symlink,
2536 .mkdir = ext4_mkdir,
2537 .rmdir = ext4_rmdir,
2538 .mknod = ext4_mknod,
2539 .rename = ext4_rename,
2540 .setattr = ext4_setattr,
2541 #ifdef CONFIG_EXT4_FS_XATTR
2542 .setxattr = generic_setxattr,
2543 .getxattr = generic_getxattr,
2544 .listxattr = ext4_listxattr,
2545 .removexattr = generic_removexattr,
2547 .check_acl = ext4_check_acl,
2548 .fiemap = ext4_fiemap,
2551 const struct inode_operations ext4_special_inode_operations = {
2552 .setattr = ext4_setattr,
2553 #ifdef CONFIG_EXT4_FS_XATTR
2554 .setxattr = generic_setxattr,
2555 .getxattr = generic_getxattr,
2556 .listxattr = ext4_listxattr,
2557 .removexattr = generic_removexattr,
2559 .check_acl = ext4_check_acl,