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 #include <trace/events/ext4.h>
46 * define how far ahead to read directories while searching them.
48 #define NAMEI_RA_CHUNKS 2
49 #define NAMEI_RA_BLOCKS 4
50 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
51 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
53 static struct buffer_head *ext4_append(handle_t *handle,
55 ext4_lblk_t *block, int *err)
57 struct buffer_head *bh;
59 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
61 bh = ext4_bread(handle, inode, *block, 1, err);
63 inode->i_size += inode->i_sb->s_blocksize;
64 EXT4_I(inode)->i_disksize = inode->i_size;
65 *err = ext4_journal_get_write_access(handle, bh);
75 #define assert(test) J_ASSERT(test)
79 #define dxtrace(command) command
81 #define dxtrace(command)
105 * dx_root_info is laid out so that if it should somehow get overlaid by a
106 * dirent the two low bits of the hash version will be zero. Therefore, the
107 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
112 struct fake_dirent dot;
114 struct fake_dirent dotdot;
118 __le32 reserved_zero;
120 u8 info_length; /* 8 */
125 struct dx_entry entries[0];
130 struct fake_dirent fake;
131 struct dx_entry entries[0];
137 struct buffer_head *bh;
138 struct dx_entry *entries;
149 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
150 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
151 static inline unsigned dx_get_hash(struct dx_entry *entry);
152 static void dx_set_hash(struct dx_entry *entry, unsigned value);
153 static unsigned dx_get_count(struct dx_entry *entries);
154 static unsigned dx_get_limit(struct dx_entry *entries);
155 static void dx_set_count(struct dx_entry *entries, unsigned value);
156 static void dx_set_limit(struct dx_entry *entries, unsigned value);
157 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
158 static unsigned dx_node_limit(struct inode *dir);
159 static struct dx_frame *dx_probe(const struct qstr *d_name,
161 struct dx_hash_info *hinfo,
162 struct dx_frame *frame,
164 static void dx_release(struct dx_frame *frames);
165 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
166 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
167 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
168 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
169 struct dx_map_entry *offsets, int count, unsigned blocksize);
170 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
171 static void dx_insert_block(struct dx_frame *frame,
172 u32 hash, ext4_lblk_t block);
173 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
174 struct dx_frame *frame,
175 struct dx_frame *frames,
177 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
178 const struct qstr *d_name,
179 struct ext4_dir_entry_2 **res_dir,
181 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
182 struct inode *inode);
185 * p is at least 6 bytes before the end of page
187 static inline struct ext4_dir_entry_2 *
188 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
190 return (struct ext4_dir_entry_2 *)((char *)p +
191 ext4_rec_len_from_disk(p->rec_len, blocksize));
195 * Future: use high four bits of block for coalesce-on-delete flags
196 * Mask them off for now.
199 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
201 return le32_to_cpu(entry->block) & 0x00ffffff;
204 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
206 entry->block = cpu_to_le32(value);
209 static inline unsigned dx_get_hash(struct dx_entry *entry)
211 return le32_to_cpu(entry->hash);
214 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
216 entry->hash = cpu_to_le32(value);
219 static inline unsigned dx_get_count(struct dx_entry *entries)
221 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
224 static inline unsigned dx_get_limit(struct dx_entry *entries)
226 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
229 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
231 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
234 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
236 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
239 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
241 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
242 EXT4_DIR_REC_LEN(2) - infosize;
243 return entry_space / sizeof(struct dx_entry);
246 static inline unsigned dx_node_limit(struct inode *dir)
248 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
249 return entry_space / sizeof(struct dx_entry);
256 static void dx_show_index(char * label, struct dx_entry *entries)
258 int i, n = dx_get_count (entries);
259 printk(KERN_DEBUG "%s index ", label);
260 for (i = 0; i < n; i++) {
261 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
262 0, (unsigned long)dx_get_block(entries + i));
274 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
275 int size, int show_names)
277 unsigned names = 0, space = 0;
278 char *base = (char *) de;
279 struct dx_hash_info h = *hinfo;
282 while ((char *) de < base + size)
288 int len = de->name_len;
289 char *name = de->name;
290 while (len--) printk("%c", *name++);
291 ext4fs_dirhash(de->name, de->name_len, &h);
292 printk(":%x.%u ", h.hash,
293 (unsigned) ((char *) de - base));
295 space += EXT4_DIR_REC_LEN(de->name_len);
298 de = ext4_next_entry(de, size);
300 printk("(%i)\n", names);
301 return (struct stats) { names, space, 1 };
304 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
305 struct dx_entry *entries, int levels)
307 unsigned blocksize = dir->i_sb->s_blocksize;
308 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
310 struct buffer_head *bh;
312 printk("%i indexed blocks...\n", count);
313 for (i = 0; i < count; i++, entries++)
315 ext4_lblk_t block = dx_get_block(entries);
316 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
317 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
319 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
320 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
322 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
323 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
324 names += stats.names;
325 space += stats.space;
326 bcount += stats.bcount;
330 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
331 levels ? "" : " ", names, space/bcount,
332 (space/bcount)*100/blocksize);
333 return (struct stats) { names, space, bcount};
335 #endif /* DX_DEBUG */
338 * Probe for a directory leaf block to search.
340 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
341 * error in the directory index, and the caller should fall back to
342 * searching the directory normally. The callers of dx_probe **MUST**
343 * check for this error code, and make sure it never gets reflected
346 static struct dx_frame *
347 dx_probe(const struct qstr *d_name, struct inode *dir,
348 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
350 unsigned count, indirect;
351 struct dx_entry *at, *entries, *p, *q, *m;
352 struct dx_root *root;
353 struct buffer_head *bh;
354 struct dx_frame *frame = frame_in;
358 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
360 root = (struct dx_root *) bh->b_data;
361 if (root->info.hash_version != DX_HASH_TEA &&
362 root->info.hash_version != DX_HASH_HALF_MD4 &&
363 root->info.hash_version != DX_HASH_LEGACY) {
364 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
365 root->info.hash_version);
367 *err = ERR_BAD_DX_DIR;
370 hinfo->hash_version = root->info.hash_version;
371 if (hinfo->hash_version <= DX_HASH_TEA)
372 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
373 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
375 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
378 if (root->info.unused_flags & 1) {
379 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
380 root->info.unused_flags);
382 *err = ERR_BAD_DX_DIR;
386 if ((indirect = root->info.indirect_levels) > 1) {
387 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
388 root->info.indirect_levels);
390 *err = ERR_BAD_DX_DIR;
394 entries = (struct dx_entry *) (((char *)&root->info) +
395 root->info.info_length);
397 if (dx_get_limit(entries) != dx_root_limit(dir,
398 root->info.info_length)) {
399 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
401 *err = ERR_BAD_DX_DIR;
405 dxtrace(printk("Look up %x", hash));
408 count = dx_get_count(entries);
409 if (!count || count > dx_get_limit(entries)) {
410 ext4_warning(dir->i_sb,
411 "dx entry: no count or count > limit");
413 *err = ERR_BAD_DX_DIR;
418 q = entries + count - 1;
422 dxtrace(printk("."));
423 if (dx_get_hash(m) > hash)
429 if (0) // linear search cross check
431 unsigned n = count - 1;
435 dxtrace(printk(","));
436 if (dx_get_hash(++at) > hash)
442 assert (at == p - 1);
446 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
448 frame->entries = entries;
450 if (!indirect--) return frame;
451 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
453 at = entries = ((struct dx_node *) bh->b_data)->entries;
454 if (dx_get_limit(entries) != dx_node_limit (dir)) {
455 ext4_warning(dir->i_sb,
456 "dx entry: limit != node limit");
458 *err = ERR_BAD_DX_DIR;
465 while (frame >= frame_in) {
470 if (*err == ERR_BAD_DX_DIR)
471 ext4_warning(dir->i_sb,
472 "Corrupt dir inode %ld, running e2fsck is "
473 "recommended.", dir->i_ino);
477 static void dx_release (struct dx_frame *frames)
479 if (frames[0].bh == NULL)
482 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
483 brelse(frames[1].bh);
484 brelse(frames[0].bh);
488 * This function increments the frame pointer to search the next leaf
489 * block, and reads in the necessary intervening nodes if the search
490 * should be necessary. Whether or not the search is necessary is
491 * controlled by the hash parameter. If the hash value is even, then
492 * the search is only continued if the next block starts with that
493 * hash value. This is used if we are searching for a specific file.
495 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
497 * This function returns 1 if the caller should continue to search,
498 * or 0 if it should not. If there is an error reading one of the
499 * index blocks, it will a negative error code.
501 * If start_hash is non-null, it will be filled in with the starting
502 * hash of the next page.
504 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
505 struct dx_frame *frame,
506 struct dx_frame *frames,
510 struct buffer_head *bh;
511 int err, num_frames = 0;
516 * Find the next leaf page by incrementing the frame pointer.
517 * If we run out of entries in the interior node, loop around and
518 * increment pointer in the parent node. When we break out of
519 * this loop, num_frames indicates the number of interior
520 * nodes need to be read.
523 if (++(p->at) < p->entries + dx_get_count(p->entries))
532 * If the hash is 1, then continue only if the next page has a
533 * continuation hash of any value. This is used for readdir
534 * handling. Otherwise, check to see if the hash matches the
535 * desired contiuation hash. If it doesn't, return since
536 * there's no point to read in the successive index pages.
538 bhash = dx_get_hash(p->at);
541 if ((hash & 1) == 0) {
542 if ((bhash & ~1) != hash)
546 * If the hash is HASH_NB_ALWAYS, we always go to the next
547 * block so no check is necessary
549 while (num_frames--) {
550 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
552 return err; /* Failure */
556 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
563 * This function fills a red-black tree with information from a
564 * directory block. It returns the number directory entries loaded
565 * into the tree. If there is an error it is returned in err.
567 static int htree_dirblock_to_tree(struct file *dir_file,
568 struct inode *dir, ext4_lblk_t block,
569 struct dx_hash_info *hinfo,
570 __u32 start_hash, __u32 start_minor_hash)
572 struct buffer_head *bh;
573 struct ext4_dir_entry_2 *de, *top;
576 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
577 (unsigned long)block));
578 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
581 de = (struct ext4_dir_entry_2 *) bh->b_data;
582 top = (struct ext4_dir_entry_2 *) ((char *) de +
583 dir->i_sb->s_blocksize -
584 EXT4_DIR_REC_LEN(0));
585 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
586 if (ext4_check_dir_entry(dir, NULL, de, bh,
587 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
588 + ((char *)de - bh->b_data))) {
589 /* On error, skip the f_pos to the next block. */
590 dir_file->f_pos = (dir_file->f_pos |
591 (dir->i_sb->s_blocksize - 1)) + 1;
595 ext4fs_dirhash(de->name, de->name_len, hinfo);
596 if ((hinfo->hash < start_hash) ||
597 ((hinfo->hash == start_hash) &&
598 (hinfo->minor_hash < start_minor_hash)))
602 if ((err = ext4_htree_store_dirent(dir_file,
603 hinfo->hash, hinfo->minor_hash, de)) != 0) {
615 * This function fills a red-black tree with information from a
616 * directory. We start scanning the directory in hash order, starting
617 * at start_hash and start_minor_hash.
619 * This function returns the number of entries inserted into the tree,
620 * or a negative error code.
622 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
623 __u32 start_minor_hash, __u32 *next_hash)
625 struct dx_hash_info hinfo;
626 struct ext4_dir_entry_2 *de;
627 struct dx_frame frames[2], *frame;
634 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
635 start_hash, start_minor_hash));
636 dir = dir_file->f_path.dentry->d_inode;
637 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
638 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
639 if (hinfo.hash_version <= DX_HASH_TEA)
640 hinfo.hash_version +=
641 EXT4_SB(dir->i_sb)->s_hash_unsigned;
642 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
643 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
644 start_hash, start_minor_hash);
648 hinfo.hash = start_hash;
649 hinfo.minor_hash = 0;
650 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
654 /* Add '.' and '..' from the htree header */
655 if (!start_hash && !start_minor_hash) {
656 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
657 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
661 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
662 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
663 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
664 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
670 block = dx_get_block(frame->at);
671 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
672 start_hash, start_minor_hash);
679 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
680 frame, frames, &hashval);
681 *next_hash = hashval;
687 * Stop if: (a) there are no more entries, or
688 * (b) we have inserted at least one entry and the
689 * next hash value is not a continuation
692 (count && ((hashval & 1) == 0)))
696 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
697 "next hash: %x\n", count, *next_hash));
706 * Directory block splitting, compacting
710 * Create map of hash values, offsets, and sizes, stored at end of block.
711 * Returns number of entries mapped.
713 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
714 struct dx_hash_info *hinfo,
715 struct dx_map_entry *map_tail)
718 char *base = (char *) de;
719 struct dx_hash_info h = *hinfo;
721 while ((char *) de < base + blocksize) {
722 if (de->name_len && de->inode) {
723 ext4fs_dirhash(de->name, de->name_len, &h);
725 map_tail->hash = h.hash;
726 map_tail->offs = ((char *) de - base)>>2;
727 map_tail->size = le16_to_cpu(de->rec_len);
731 /* XXX: do we need to check rec_len == 0 case? -Chris */
732 de = ext4_next_entry(de, blocksize);
737 /* Sort map by hash value */
738 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
740 struct dx_map_entry *p, *q, *top = map + count - 1;
742 /* Combsort until bubble sort doesn't suck */
745 if (count - 9 < 2) /* 9, 10 -> 11 */
747 for (p = top, q = p - count; q >= map; p--, q--)
748 if (p->hash < q->hash)
751 /* Garden variety bubble sort */
756 if (q[1].hash >= q[0].hash)
764 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
766 struct dx_entry *entries = frame->entries;
767 struct dx_entry *old = frame->at, *new = old + 1;
768 int count = dx_get_count(entries);
770 assert(count < dx_get_limit(entries));
771 assert(old < entries + count);
772 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
773 dx_set_hash(new, hash);
774 dx_set_block(new, block);
775 dx_set_count(entries, count + 1);
778 static void ext4_update_dx_flag(struct inode *inode)
780 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
781 EXT4_FEATURE_COMPAT_DIR_INDEX))
782 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
786 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
788 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
789 * `de != NULL' is guaranteed by caller.
791 static inline int ext4_match (int len, const char * const name,
792 struct ext4_dir_entry_2 * de)
794 if (len != de->name_len)
798 return !memcmp(name, de->name, len);
802 * Returns 0 if not found, -1 on failure, and 1 on success
804 static inline int search_dirblock(struct buffer_head *bh,
806 const struct qstr *d_name,
808 struct ext4_dir_entry_2 ** res_dir)
810 struct ext4_dir_entry_2 * de;
813 const char *name = d_name->name;
814 int namelen = d_name->len;
816 de = (struct ext4_dir_entry_2 *) bh->b_data;
817 dlimit = bh->b_data + dir->i_sb->s_blocksize;
818 while ((char *) de < dlimit) {
819 /* this code is executed quadratically often */
820 /* do minimal checking `by hand' */
822 if ((char *) de + namelen <= dlimit &&
823 ext4_match (namelen, name, de)) {
824 /* found a match - just to be sure, do a full check */
825 if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
830 /* prevent looping on a bad block */
831 de_len = ext4_rec_len_from_disk(de->rec_len,
832 dir->i_sb->s_blocksize);
836 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
845 * finds an entry in the specified directory with the wanted name. It
846 * returns the cache buffer in which the entry was found, and the entry
847 * itself (as a parameter - res_dir). It does NOT read the inode of the
848 * entry - you'll have to do that yourself if you want to.
850 * The returned buffer_head has ->b_count elevated. The caller is expected
851 * to brelse() it when appropriate.
853 static struct buffer_head * ext4_find_entry (struct inode *dir,
854 const struct qstr *d_name,
855 struct ext4_dir_entry_2 ** res_dir)
857 struct super_block *sb;
858 struct buffer_head *bh_use[NAMEI_RA_SIZE];
859 struct buffer_head *bh, *ret = NULL;
860 ext4_lblk_t start, block, b;
861 const u8 *name = d_name->name;
862 int ra_max = 0; /* Number of bh's in the readahead
864 int ra_ptr = 0; /* Current index into readahead
873 namelen = d_name->len;
874 if (namelen > EXT4_NAME_LEN)
876 if ((namelen <= 2) && (name[0] == '.') &&
877 (name[1] == '.' || name[1] == '\0')) {
879 * "." or ".." will only be in the first block
880 * NFS may look up ".."; "." should be handled by the VFS
887 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
889 * On success, or if the error was file not found,
890 * return. Otherwise, fall back to doing a search the
893 if (bh || (err != ERR_BAD_DX_DIR))
895 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
898 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
899 start = EXT4_I(dir)->i_dir_start_lookup;
900 if (start >= nblocks)
906 * We deal with the read-ahead logic here.
908 if (ra_ptr >= ra_max) {
909 /* Refill the readahead buffer */
912 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
914 * Terminate if we reach the end of the
915 * directory and must wrap, or if our
916 * search has finished at this block.
918 if (b >= nblocks || (num && block == start)) {
919 bh_use[ra_max] = NULL;
923 bh = ext4_getblk(NULL, dir, b++, 0, &err);
926 ll_rw_block(READ | REQ_META | REQ_PRIO,
930 if ((bh = bh_use[ra_ptr++]) == NULL)
933 if (!buffer_uptodate(bh)) {
934 /* read error, skip block & hope for the best */
935 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
936 (unsigned long) block);
940 i = search_dirblock(bh, dir, d_name,
941 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
943 EXT4_I(dir)->i_dir_start_lookup = block;
945 goto cleanup_and_exit;
949 goto cleanup_and_exit;
952 if (++block >= nblocks)
954 } while (block != start);
957 * If the directory has grown while we were searching, then
958 * search the last part of the directory before giving up.
961 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
962 if (block < nblocks) {
968 /* Clean up the read-ahead blocks */
969 for (; ra_ptr < ra_max; ra_ptr++)
970 brelse(bh_use[ra_ptr]);
974 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
975 struct ext4_dir_entry_2 **res_dir, int *err)
977 struct super_block * sb = dir->i_sb;
978 struct dx_hash_info hinfo;
979 struct dx_frame frames[2], *frame;
980 struct buffer_head *bh;
984 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
987 block = dx_get_block(frame->at);
988 if (!(bh = ext4_bread(NULL, dir, block, 0, err)))
991 retval = search_dirblock(bh, dir, d_name,
992 block << EXT4_BLOCK_SIZE_BITS(sb),
994 if (retval == 1) { /* Success! */
1000 *err = ERR_BAD_DX_DIR;
1004 /* Check to see if we should continue to search */
1005 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1009 "error reading index page in directory #%lu",
1014 } while (retval == 1);
1018 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1019 dx_release (frames);
1023 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1025 struct inode *inode;
1026 struct ext4_dir_entry_2 *de;
1027 struct buffer_head *bh;
1029 if (dentry->d_name.len > EXT4_NAME_LEN)
1030 return ERR_PTR(-ENAMETOOLONG);
1032 bh = ext4_find_entry(dir, &dentry->d_name, &de);
1035 __u32 ino = le32_to_cpu(de->inode);
1037 if (!ext4_valid_inum(dir->i_sb, ino)) {
1038 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1039 return ERR_PTR(-EIO);
1041 inode = ext4_iget(dir->i_sb, ino);
1042 if (inode == ERR_PTR(-ESTALE)) {
1043 EXT4_ERROR_INODE(dir,
1044 "deleted inode referenced: %u",
1046 return ERR_PTR(-EIO);
1049 return d_splice_alias(inode, dentry);
1053 struct dentry *ext4_get_parent(struct dentry *child)
1056 static const struct qstr dotdot = {
1060 struct ext4_dir_entry_2 * de;
1061 struct buffer_head *bh;
1063 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1065 return ERR_PTR(-ENOENT);
1066 ino = le32_to_cpu(de->inode);
1069 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1070 EXT4_ERROR_INODE(child->d_inode,
1071 "bad parent inode number: %u", ino);
1072 return ERR_PTR(-EIO);
1075 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1079 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1080 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1081 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1082 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1083 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1084 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1085 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1086 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
1089 static inline void ext4_set_de_type(struct super_block *sb,
1090 struct ext4_dir_entry_2 *de,
1092 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1093 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1097 * Move count entries from end of map between two memory locations.
1098 * Returns pointer to last entry moved.
1100 static struct ext4_dir_entry_2 *
1101 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1104 unsigned rec_len = 0;
1107 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1108 (from + (map->offs<<2));
1109 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1110 memcpy (to, de, rec_len);
1111 ((struct ext4_dir_entry_2 *) to)->rec_len =
1112 ext4_rec_len_to_disk(rec_len, blocksize);
1117 return (struct ext4_dir_entry_2 *) (to - rec_len);
1121 * Compact each dir entry in the range to the minimal rec_len.
1122 * Returns pointer to last entry in range.
1124 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1126 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1127 unsigned rec_len = 0;
1130 while ((char*)de < base + blocksize) {
1131 next = ext4_next_entry(de, blocksize);
1132 if (de->inode && de->name_len) {
1133 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1135 memmove(to, de, rec_len);
1136 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1138 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1146 * Split a full leaf block to make room for a new dir entry.
1147 * Allocate a new block, and move entries so that they are approx. equally full.
1148 * Returns pointer to de in block into which the new entry will be inserted.
1150 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1151 struct buffer_head **bh,struct dx_frame *frame,
1152 struct dx_hash_info *hinfo, int *error)
1154 unsigned blocksize = dir->i_sb->s_blocksize;
1155 unsigned count, continued;
1156 struct buffer_head *bh2;
1157 ext4_lblk_t newblock;
1159 struct dx_map_entry *map;
1160 char *data1 = (*bh)->b_data, *data2;
1161 unsigned split, move, size;
1162 struct ext4_dir_entry_2 *de = NULL, *de2;
1165 bh2 = ext4_append (handle, dir, &newblock, &err);
1172 BUFFER_TRACE(*bh, "get_write_access");
1173 err = ext4_journal_get_write_access(handle, *bh);
1177 BUFFER_TRACE(frame->bh, "get_write_access");
1178 err = ext4_journal_get_write_access(handle, frame->bh);
1182 data2 = bh2->b_data;
1184 /* create map in the end of data2 block */
1185 map = (struct dx_map_entry *) (data2 + blocksize);
1186 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1187 blocksize, hinfo, map);
1189 dx_sort_map(map, count);
1190 /* Split the existing block in the middle, size-wise */
1193 for (i = count-1; i >= 0; i--) {
1194 /* is more than half of this entry in 2nd half of the block? */
1195 if (size + map[i].size/2 > blocksize/2)
1197 size += map[i].size;
1200 /* map index at which we will split */
1201 split = count - move;
1202 hash2 = map[split].hash;
1203 continued = hash2 == map[split - 1].hash;
1204 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1205 (unsigned long)dx_get_block(frame->at),
1206 hash2, split, count-split));
1208 /* Fancy dance to stay within two buffers */
1209 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1210 de = dx_pack_dirents(data1, blocksize);
1211 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1213 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1215 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1216 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1218 /* Which block gets the new entry? */
1219 if (hinfo->hash >= hash2)
1224 dx_insert_block(frame, hash2 + continued, newblock);
1225 err = ext4_handle_dirty_metadata(handle, dir, bh2);
1228 err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1232 dxtrace(dx_show_index("frame", frame->entries));
1239 ext4_std_error(dir->i_sb, err);
1246 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1247 * it points to a directory entry which is guaranteed to be large
1248 * enough for new directory entry. If de is NULL, then
1249 * add_dirent_to_buf will attempt search the directory block for
1250 * space. It will return -ENOSPC if no space is available, and -EIO
1251 * and -EEXIST if directory entry already exists.
1253 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1254 struct inode *inode, struct ext4_dir_entry_2 *de,
1255 struct buffer_head *bh)
1257 struct inode *dir = dentry->d_parent->d_inode;
1258 const char *name = dentry->d_name.name;
1259 int namelen = dentry->d_name.len;
1260 unsigned int offset = 0;
1261 unsigned int blocksize = dir->i_sb->s_blocksize;
1262 unsigned short reclen;
1263 int nlen, rlen, err;
1266 reclen = EXT4_DIR_REC_LEN(namelen);
1268 de = (struct ext4_dir_entry_2 *)bh->b_data;
1269 top = bh->b_data + blocksize - reclen;
1270 while ((char *) de <= top) {
1271 if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
1273 if (ext4_match(namelen, name, de))
1275 nlen = EXT4_DIR_REC_LEN(de->name_len);
1276 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1277 if ((de->inode? rlen - nlen: rlen) >= reclen)
1279 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1282 if ((char *) de > top)
1285 BUFFER_TRACE(bh, "get_write_access");
1286 err = ext4_journal_get_write_access(handle, bh);
1288 ext4_std_error(dir->i_sb, err);
1292 /* By now the buffer is marked for journaling */
1293 nlen = EXT4_DIR_REC_LEN(de->name_len);
1294 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1296 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1297 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1298 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1301 de->file_type = EXT4_FT_UNKNOWN;
1303 de->inode = cpu_to_le32(inode->i_ino);
1304 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1307 de->name_len = namelen;
1308 memcpy(de->name, name, namelen);
1310 * XXX shouldn't update any times until successful
1311 * completion of syscall, but too many callers depend
1314 * XXX similarly, too many callers depend on
1315 * ext4_new_inode() setting the times, but error
1316 * recovery deletes the inode, so the worst that can
1317 * happen is that the times are slightly out of date
1318 * and/or different from the directory change time.
1320 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1321 ext4_update_dx_flag(dir);
1323 ext4_mark_inode_dirty(handle, dir);
1324 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1325 err = ext4_handle_dirty_metadata(handle, dir, bh);
1327 ext4_std_error(dir->i_sb, err);
1332 * This converts a one block unindexed directory to a 3 block indexed
1333 * directory, and adds the dentry to the indexed directory.
1335 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1336 struct inode *inode, struct buffer_head *bh)
1338 struct inode *dir = dentry->d_parent->d_inode;
1339 const char *name = dentry->d_name.name;
1340 int namelen = dentry->d_name.len;
1341 struct buffer_head *bh2;
1342 struct dx_root *root;
1343 struct dx_frame frames[2], *frame;
1344 struct dx_entry *entries;
1345 struct ext4_dir_entry_2 *de, *de2;
1350 struct dx_hash_info hinfo;
1352 struct fake_dirent *fde;
1354 blocksize = dir->i_sb->s_blocksize;
1355 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1356 retval = ext4_journal_get_write_access(handle, bh);
1358 ext4_std_error(dir->i_sb, retval);
1362 root = (struct dx_root *) bh->b_data;
1364 /* The 0th block becomes the root, move the dirents out */
1365 fde = &root->dotdot;
1366 de = (struct ext4_dir_entry_2 *)((char *)fde +
1367 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1368 if ((char *) de >= (((char *) root) + blocksize)) {
1369 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1373 len = ((char *) root) + blocksize - (char *) de;
1375 /* Allocate new block for the 0th block's dirents */
1376 bh2 = ext4_append(handle, dir, &block, &retval);
1381 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1382 data1 = bh2->b_data;
1384 memcpy (data1, de, len);
1385 de = (struct ext4_dir_entry_2 *) data1;
1387 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1389 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1391 /* Initialize the root; the dot dirents already exist */
1392 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1393 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1395 memset (&root->info, 0, sizeof(root->info));
1396 root->info.info_length = sizeof(root->info);
1397 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1398 entries = root->entries;
1399 dx_set_block(entries, 1);
1400 dx_set_count(entries, 1);
1401 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1403 /* Initialize as for dx_probe */
1404 hinfo.hash_version = root->info.hash_version;
1405 if (hinfo.hash_version <= DX_HASH_TEA)
1406 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1407 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1408 ext4fs_dirhash(name, namelen, &hinfo);
1410 frame->entries = entries;
1411 frame->at = entries;
1415 ext4_handle_dirty_metadata(handle, dir, frame->bh);
1416 ext4_handle_dirty_metadata(handle, dir, bh);
1418 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1421 * Even if the block split failed, we have to properly write
1422 * out all the changes we did so far. Otherwise we can end up
1423 * with corrupted filesystem.
1425 ext4_mark_inode_dirty(handle, dir);
1431 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1439 * adds a file entry to the specified directory, using the same
1440 * semantics as ext4_find_entry(). It returns NULL if it failed.
1442 * NOTE!! The inode part of 'de' is left at 0 - which means you
1443 * may not sleep between calling this and putting something into
1444 * the entry, as someone else might have used it while you slept.
1446 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1447 struct inode *inode)
1449 struct inode *dir = dentry->d_parent->d_inode;
1450 struct buffer_head *bh;
1451 struct ext4_dir_entry_2 *de;
1452 struct super_block *sb;
1456 ext4_lblk_t block, blocks;
1459 blocksize = sb->s_blocksize;
1460 if (!dentry->d_name.len)
1463 retval = ext4_dx_add_entry(handle, dentry, inode);
1464 if (!retval || (retval != ERR_BAD_DX_DIR))
1466 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1468 ext4_mark_inode_dirty(handle, dir);
1470 blocks = dir->i_size >> sb->s_blocksize_bits;
1471 for (block = 0; block < blocks; block++) {
1472 bh = ext4_bread(handle, dir, block, 0, &retval);
1475 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1476 if (retval != -ENOSPC) {
1481 if (blocks == 1 && !dx_fallback &&
1482 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1483 return make_indexed_dir(handle, dentry, inode, bh);
1486 bh = ext4_append(handle, dir, &block, &retval);
1489 de = (struct ext4_dir_entry_2 *) bh->b_data;
1491 de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1492 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1495 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1500 * Returns 0 for success, or a negative error value
1502 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1503 struct inode *inode)
1505 struct dx_frame frames[2], *frame;
1506 struct dx_entry *entries, *at;
1507 struct dx_hash_info hinfo;
1508 struct buffer_head *bh;
1509 struct inode *dir = dentry->d_parent->d_inode;
1510 struct super_block *sb = dir->i_sb;
1511 struct ext4_dir_entry_2 *de;
1514 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1517 entries = frame->entries;
1520 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1523 BUFFER_TRACE(bh, "get_write_access");
1524 err = ext4_journal_get_write_access(handle, bh);
1528 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1532 /* Block full, should compress but for now just split */
1533 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1534 dx_get_count(entries), dx_get_limit(entries)));
1535 /* Need to split index? */
1536 if (dx_get_count(entries) == dx_get_limit(entries)) {
1537 ext4_lblk_t newblock;
1538 unsigned icount = dx_get_count(entries);
1539 int levels = frame - frames;
1540 struct dx_entry *entries2;
1541 struct dx_node *node2;
1542 struct buffer_head *bh2;
1544 if (levels && (dx_get_count(frames->entries) ==
1545 dx_get_limit(frames->entries))) {
1546 ext4_warning(sb, "Directory index full!");
1550 bh2 = ext4_append (handle, dir, &newblock, &err);
1553 node2 = (struct dx_node *)(bh2->b_data);
1554 entries2 = node2->entries;
1555 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1556 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1558 BUFFER_TRACE(frame->bh, "get_write_access");
1559 err = ext4_journal_get_write_access(handle, frame->bh);
1563 unsigned icount1 = icount/2, icount2 = icount - icount1;
1564 unsigned hash2 = dx_get_hash(entries + icount1);
1565 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1568 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1569 err = ext4_journal_get_write_access(handle,
1574 memcpy((char *) entries2, (char *) (entries + icount1),
1575 icount2 * sizeof(struct dx_entry));
1576 dx_set_count(entries, icount1);
1577 dx_set_count(entries2, icount2);
1578 dx_set_limit(entries2, dx_node_limit(dir));
1580 /* Which index block gets the new entry? */
1581 if (at - entries >= icount1) {
1582 frame->at = at = at - entries - icount1 + entries2;
1583 frame->entries = entries = entries2;
1584 swap(frame->bh, bh2);
1586 dx_insert_block(frames + 0, hash2, newblock);
1587 dxtrace(dx_show_index("node", frames[1].entries));
1588 dxtrace(dx_show_index("node",
1589 ((struct dx_node *) bh2->b_data)->entries));
1590 err = ext4_handle_dirty_metadata(handle, dir, bh2);
1595 dxtrace(printk(KERN_DEBUG
1596 "Creating second level index...\n"));
1597 memcpy((char *) entries2, (char *) entries,
1598 icount * sizeof(struct dx_entry));
1599 dx_set_limit(entries2, dx_node_limit(dir));
1602 dx_set_count(entries, 1);
1603 dx_set_block(entries + 0, newblock);
1604 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1606 /* Add new access path frame */
1608 frame->at = at = at - entries + entries2;
1609 frame->entries = entries = entries2;
1611 err = ext4_journal_get_write_access(handle,
1616 err = ext4_handle_dirty_metadata(handle, dir, frames[0].bh);
1618 ext4_std_error(inode->i_sb, err);
1622 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1625 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1629 ext4_std_error(dir->i_sb, err);
1638 * ext4_delete_entry deletes a directory entry by merging it with the
1641 static int ext4_delete_entry(handle_t *handle,
1643 struct ext4_dir_entry_2 *de_del,
1644 struct buffer_head *bh)
1646 struct ext4_dir_entry_2 *de, *pde;
1647 unsigned int blocksize = dir->i_sb->s_blocksize;
1652 de = (struct ext4_dir_entry_2 *) bh->b_data;
1653 while (i < bh->b_size) {
1654 if (ext4_check_dir_entry(dir, NULL, de, bh, i))
1657 BUFFER_TRACE(bh, "get_write_access");
1658 err = ext4_journal_get_write_access(handle, bh);
1659 if (unlikely(err)) {
1660 ext4_std_error(dir->i_sb, err);
1664 pde->rec_len = ext4_rec_len_to_disk(
1665 ext4_rec_len_from_disk(pde->rec_len,
1667 ext4_rec_len_from_disk(de->rec_len,
1673 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1674 err = ext4_handle_dirty_metadata(handle, dir, bh);
1675 if (unlikely(err)) {
1676 ext4_std_error(dir->i_sb, err);
1681 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
1683 de = ext4_next_entry(de, blocksize);
1689 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1690 * since this indicates that nlinks count was previously 1.
1692 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1695 if (is_dx(inode) && inode->i_nlink > 1) {
1696 /* limit is 16-bit i_links_count */
1697 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1698 set_nlink(inode, 1);
1699 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1700 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1706 * If a directory had nlink == 1, then we should let it be 1. This indicates
1707 * directory has >EXT4_LINK_MAX subdirs.
1709 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1711 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
1716 static int ext4_add_nondir(handle_t *handle,
1717 struct dentry *dentry, struct inode *inode)
1719 int err = ext4_add_entry(handle, dentry, inode);
1721 ext4_mark_inode_dirty(handle, inode);
1722 d_instantiate(dentry, inode);
1723 unlock_new_inode(inode);
1727 unlock_new_inode(inode);
1733 * By the time this is called, we already have created
1734 * the directory cache entry for the new file, but it
1735 * is so far negative - it has no inode.
1737 * If the create succeeds, we fill in the inode information
1738 * with d_instantiate().
1740 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1741 struct nameidata *nd)
1744 struct inode *inode;
1745 int err, retries = 0;
1747 dquot_initialize(dir);
1750 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1751 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1752 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1754 return PTR_ERR(handle);
1756 if (IS_DIRSYNC(dir))
1757 ext4_handle_sync(handle);
1759 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
1760 err = PTR_ERR(inode);
1761 if (!IS_ERR(inode)) {
1762 inode->i_op = &ext4_file_inode_operations;
1763 inode->i_fop = &ext4_file_operations;
1764 ext4_set_aops(inode);
1765 err = ext4_add_nondir(handle, dentry, inode);
1767 ext4_journal_stop(handle);
1768 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1773 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1774 umode_t mode, dev_t rdev)
1777 struct inode *inode;
1778 int err, retries = 0;
1780 if (!new_valid_dev(rdev))
1783 dquot_initialize(dir);
1786 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1787 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1788 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1790 return PTR_ERR(handle);
1792 if (IS_DIRSYNC(dir))
1793 ext4_handle_sync(handle);
1795 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
1796 err = PTR_ERR(inode);
1797 if (!IS_ERR(inode)) {
1798 init_special_inode(inode, inode->i_mode, rdev);
1799 #ifdef CONFIG_EXT4_FS_XATTR
1800 inode->i_op = &ext4_special_inode_operations;
1802 err = ext4_add_nondir(handle, dentry, inode);
1804 ext4_journal_stop(handle);
1805 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1810 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1813 struct inode *inode;
1814 struct buffer_head *dir_block = NULL;
1815 struct ext4_dir_entry_2 *de;
1816 unsigned int blocksize = dir->i_sb->s_blocksize;
1817 int err, retries = 0;
1819 if (EXT4_DIR_LINK_MAX(dir))
1822 dquot_initialize(dir);
1825 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1826 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1827 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1829 return PTR_ERR(handle);
1831 if (IS_DIRSYNC(dir))
1832 ext4_handle_sync(handle);
1834 inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1835 &dentry->d_name, 0, NULL);
1836 err = PTR_ERR(inode);
1840 inode->i_op = &ext4_dir_inode_operations;
1841 inode->i_fop = &ext4_dir_operations;
1842 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1843 dir_block = ext4_bread(handle, inode, 0, 1, &err);
1845 goto out_clear_inode;
1846 BUFFER_TRACE(dir_block, "get_write_access");
1847 err = ext4_journal_get_write_access(handle, dir_block);
1849 goto out_clear_inode;
1850 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1851 de->inode = cpu_to_le32(inode->i_ino);
1853 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1855 strcpy(de->name, ".");
1856 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1857 de = ext4_next_entry(de, blocksize);
1858 de->inode = cpu_to_le32(dir->i_ino);
1859 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1862 strcpy(de->name, "..");
1863 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1864 set_nlink(inode, 2);
1865 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1866 err = ext4_handle_dirty_metadata(handle, inode, dir_block);
1868 goto out_clear_inode;
1869 err = ext4_mark_inode_dirty(handle, inode);
1871 err = ext4_add_entry(handle, dentry, inode);
1875 unlock_new_inode(inode);
1876 ext4_mark_inode_dirty(handle, inode);
1880 ext4_inc_count(handle, dir);
1881 ext4_update_dx_flag(dir);
1882 err = ext4_mark_inode_dirty(handle, dir);
1884 goto out_clear_inode;
1885 d_instantiate(dentry, inode);
1886 unlock_new_inode(inode);
1889 ext4_journal_stop(handle);
1890 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1896 * routine to check that the specified directory is empty (for rmdir)
1898 static int empty_dir(struct inode *inode)
1900 unsigned int offset;
1901 struct buffer_head *bh;
1902 struct ext4_dir_entry_2 *de, *de1;
1903 struct super_block *sb;
1907 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1908 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1910 EXT4_ERROR_INODE(inode,
1911 "error %d reading directory lblock 0", err);
1913 ext4_warning(inode->i_sb,
1914 "bad directory (dir #%lu) - no data block",
1918 de = (struct ext4_dir_entry_2 *) bh->b_data;
1919 de1 = ext4_next_entry(de, sb->s_blocksize);
1920 if (le32_to_cpu(de->inode) != inode->i_ino ||
1921 !le32_to_cpu(de1->inode) ||
1922 strcmp(".", de->name) ||
1923 strcmp("..", de1->name)) {
1924 ext4_warning(inode->i_sb,
1925 "bad directory (dir #%lu) - no `.' or `..'",
1930 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1931 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1932 de = ext4_next_entry(de1, sb->s_blocksize);
1933 while (offset < inode->i_size) {
1935 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1936 unsigned int lblock;
1939 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1940 bh = ext4_bread(NULL, inode, lblock, 0, &err);
1943 EXT4_ERROR_INODE(inode,
1944 "error %d reading directory "
1945 "lblock %u", err, lblock);
1946 offset += sb->s_blocksize;
1949 de = (struct ext4_dir_entry_2 *) bh->b_data;
1951 if (ext4_check_dir_entry(inode, NULL, de, bh, offset)) {
1952 de = (struct ext4_dir_entry_2 *)(bh->b_data +
1954 offset = (offset | (sb->s_blocksize - 1)) + 1;
1957 if (le32_to_cpu(de->inode)) {
1961 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1962 de = ext4_next_entry(de, sb->s_blocksize);
1968 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1969 * such inodes, starting at the superblock, in case we crash before the
1970 * file is closed/deleted, or in case the inode truncate spans multiple
1971 * transactions and the last transaction is not recovered after a crash.
1973 * At filesystem recovery time, we walk this list deleting unlinked
1974 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1976 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1978 struct super_block *sb = inode->i_sb;
1979 struct ext4_iloc iloc;
1982 if (!ext4_handle_valid(handle))
1985 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
1986 if (!list_empty(&EXT4_I(inode)->i_orphan))
1990 * Orphan handling is only valid for files with data blocks
1991 * being truncated, or files being unlinked. Note that we either
1992 * hold i_mutex, or the inode can not be referenced from outside,
1993 * so i_nlink should not be bumped due to race
1995 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1996 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1998 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1999 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2003 err = ext4_reserve_inode_write(handle, inode, &iloc);
2007 * Due to previous errors inode may be already a part of on-disk
2008 * orphan list. If so skip on-disk list modification.
2010 if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2011 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2014 /* Insert this inode at the head of the on-disk orphan list... */
2015 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2016 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2017 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
2018 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2022 /* Only add to the head of the in-memory list if all the
2023 * previous operations succeeded. If the orphan_add is going to
2024 * fail (possibly taking the journal offline), we can't risk
2025 * leaving the inode on the orphan list: stray orphan-list
2026 * entries can cause panics at unmount time.
2028 * This is safe: on error we're going to ignore the orphan list
2029 * anyway on the next recovery. */
2032 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2034 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2035 jbd_debug(4, "orphan inode %lu will point to %d\n",
2036 inode->i_ino, NEXT_ORPHAN(inode));
2038 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2039 ext4_std_error(inode->i_sb, err);
2044 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2045 * of such inodes stored on disk, because it is finally being cleaned up.
2047 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2049 struct list_head *prev;
2050 struct ext4_inode_info *ei = EXT4_I(inode);
2051 struct ext4_sb_info *sbi;
2053 struct ext4_iloc iloc;
2056 /* ext4_handle_valid() assumes a valid handle_t pointer */
2057 if (handle && !ext4_handle_valid(handle))
2060 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2061 if (list_empty(&ei->i_orphan))
2064 ino_next = NEXT_ORPHAN(inode);
2065 prev = ei->i_orphan.prev;
2066 sbi = EXT4_SB(inode->i_sb);
2068 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2070 list_del_init(&ei->i_orphan);
2072 /* If we're on an error path, we may not have a valid
2073 * transaction handle with which to update the orphan list on
2074 * disk, but we still need to remove the inode from the linked
2075 * list in memory. */
2076 if (sbi->s_journal && !handle)
2079 err = ext4_reserve_inode_write(handle, inode, &iloc);
2083 if (prev == &sbi->s_orphan) {
2084 jbd_debug(4, "superblock will point to %u\n", ino_next);
2085 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2086 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2089 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2090 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
2092 struct ext4_iloc iloc2;
2093 struct inode *i_prev =
2094 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2096 jbd_debug(4, "orphan inode %lu will point to %u\n",
2097 i_prev->i_ino, ino_next);
2098 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2101 NEXT_ORPHAN(i_prev) = ino_next;
2102 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2106 NEXT_ORPHAN(inode) = 0;
2107 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2110 ext4_std_error(inode->i_sb, err);
2112 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2120 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2123 struct inode *inode;
2124 struct buffer_head *bh;
2125 struct ext4_dir_entry_2 *de;
2128 /* Initialize quotas before so that eventual writes go in
2129 * separate transaction */
2130 dquot_initialize(dir);
2131 dquot_initialize(dentry->d_inode);
2133 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2135 return PTR_ERR(handle);
2138 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2142 if (IS_DIRSYNC(dir))
2143 ext4_handle_sync(handle);
2145 inode = dentry->d_inode;
2148 if (le32_to_cpu(de->inode) != inode->i_ino)
2151 retval = -ENOTEMPTY;
2152 if (!empty_dir(inode))
2155 retval = ext4_delete_entry(handle, dir, de, bh);
2158 if (!EXT4_DIR_LINK_EMPTY(inode))
2159 ext4_warning(inode->i_sb,
2160 "empty directory has too many links (%d)",
2164 /* There's no need to set i_disksize: the fact that i_nlink is
2165 * zero will ensure that the right thing happens during any
2168 ext4_orphan_add(handle, inode);
2169 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2170 ext4_mark_inode_dirty(handle, inode);
2171 ext4_dec_count(handle, dir);
2172 ext4_update_dx_flag(dir);
2173 ext4_mark_inode_dirty(handle, dir);
2176 ext4_journal_stop(handle);
2181 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2184 struct inode *inode;
2185 struct buffer_head *bh;
2186 struct ext4_dir_entry_2 *de;
2189 trace_ext4_unlink_enter(dir, dentry);
2190 /* Initialize quotas before so that eventual writes go
2191 * in separate transaction */
2192 dquot_initialize(dir);
2193 dquot_initialize(dentry->d_inode);
2195 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2197 return PTR_ERR(handle);
2199 if (IS_DIRSYNC(dir))
2200 ext4_handle_sync(handle);
2203 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2207 inode = dentry->d_inode;
2210 if (le32_to_cpu(de->inode) != inode->i_ino)
2213 if (!inode->i_nlink) {
2214 ext4_warning(inode->i_sb,
2215 "Deleting nonexistent file (%lu), %d",
2216 inode->i_ino, inode->i_nlink);
2217 set_nlink(inode, 1);
2219 retval = ext4_delete_entry(handle, dir, de, bh);
2222 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2223 ext4_update_dx_flag(dir);
2224 ext4_mark_inode_dirty(handle, dir);
2226 if (!inode->i_nlink)
2227 ext4_orphan_add(handle, inode);
2228 inode->i_ctime = ext4_current_time(inode);
2229 ext4_mark_inode_dirty(handle, inode);
2233 ext4_journal_stop(handle);
2235 trace_ext4_unlink_exit(dentry, retval);
2239 static int ext4_symlink(struct inode *dir,
2240 struct dentry *dentry, const char *symname)
2243 struct inode *inode;
2244 int l, err, retries = 0;
2247 l = strlen(symname)+1;
2248 if (l > dir->i_sb->s_blocksize)
2249 return -ENAMETOOLONG;
2251 dquot_initialize(dir);
2253 if (l > EXT4_N_BLOCKS * 4) {
2255 * For non-fast symlinks, we just allocate inode and put it on
2256 * orphan list in the first transaction => we need bitmap,
2257 * group descriptor, sb, inode block, quota blocks, and
2258 * possibly selinux xattr blocks.
2260 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2261 EXT4_XATTR_TRANS_BLOCKS;
2264 * Fast symlink. We have to add entry to directory
2265 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2266 * allocate new inode (bitmap, group descriptor, inode block,
2267 * quota blocks, sb is already counted in previous macros).
2269 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2270 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2271 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2274 handle = ext4_journal_start(dir, credits);
2276 return PTR_ERR(handle);
2278 if (IS_DIRSYNC(dir))
2279 ext4_handle_sync(handle);
2281 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2282 &dentry->d_name, 0, NULL);
2283 err = PTR_ERR(inode);
2287 if (l > EXT4_N_BLOCKS * 4) {
2288 inode->i_op = &ext4_symlink_inode_operations;
2289 ext4_set_aops(inode);
2291 * We cannot call page_symlink() with transaction started
2292 * because it calls into ext4_write_begin() which can wait
2293 * for transaction commit if we are running out of space
2294 * and thus we deadlock. So we have to stop transaction now
2295 * and restart it when symlink contents is written.
2297 * To keep fs consistent in case of crash, we have to put inode
2298 * to orphan list in the mean time.
2301 err = ext4_orphan_add(handle, inode);
2302 ext4_journal_stop(handle);
2304 goto err_drop_inode;
2305 err = __page_symlink(inode, symname, l, 1);
2307 goto err_drop_inode;
2309 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2310 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2312 handle = ext4_journal_start(dir,
2313 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2314 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2315 if (IS_ERR(handle)) {
2316 err = PTR_ERR(handle);
2317 goto err_drop_inode;
2319 set_nlink(inode, 1);
2320 err = ext4_orphan_del(handle, inode);
2322 ext4_journal_stop(handle);
2324 goto err_drop_inode;
2327 /* clear the extent format for fast symlink */
2328 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2329 inode->i_op = &ext4_fast_symlink_inode_operations;
2330 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2331 inode->i_size = l-1;
2333 EXT4_I(inode)->i_disksize = inode->i_size;
2334 err = ext4_add_nondir(handle, dentry, inode);
2336 ext4_journal_stop(handle);
2337 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2341 unlock_new_inode(inode);
2346 static int ext4_link(struct dentry *old_dentry,
2347 struct inode *dir, struct dentry *dentry)
2350 struct inode *inode = old_dentry->d_inode;
2351 int err, retries = 0;
2353 if (inode->i_nlink >= EXT4_LINK_MAX)
2356 dquot_initialize(dir);
2359 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2360 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2362 return PTR_ERR(handle);
2364 if (IS_DIRSYNC(dir))
2365 ext4_handle_sync(handle);
2367 inode->i_ctime = ext4_current_time(inode);
2368 ext4_inc_count(handle, inode);
2371 err = ext4_add_entry(handle, dentry, inode);
2373 ext4_mark_inode_dirty(handle, inode);
2374 d_instantiate(dentry, inode);
2379 ext4_journal_stop(handle);
2380 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2385 #define PARENT_INO(buffer, size) \
2386 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2389 * Anybody can rename anything with this: the permission checks are left to the
2390 * higher-level routines.
2392 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2393 struct inode *new_dir, struct dentry *new_dentry)
2396 struct inode *old_inode, *new_inode;
2397 struct buffer_head *old_bh, *new_bh, *dir_bh;
2398 struct ext4_dir_entry_2 *old_de, *new_de;
2399 int retval, force_da_alloc = 0;
2401 dquot_initialize(old_dir);
2402 dquot_initialize(new_dir);
2404 old_bh = new_bh = dir_bh = NULL;
2406 /* Initialize quotas before so that eventual writes go
2407 * in separate transaction */
2408 if (new_dentry->d_inode)
2409 dquot_initialize(new_dentry->d_inode);
2410 handle = ext4_journal_start(old_dir, 2 *
2411 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2412 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2414 return PTR_ERR(handle);
2416 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2417 ext4_handle_sync(handle);
2419 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2421 * Check for inode number is _not_ due to possible IO errors.
2422 * We might rmdir the source, keep it as pwd of some process
2423 * and merrily kill the link to whatever was created under the
2424 * same name. Goodbye sticky bit ;-<
2426 old_inode = old_dentry->d_inode;
2428 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2431 new_inode = new_dentry->d_inode;
2432 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2439 if (S_ISDIR(old_inode->i_mode)) {
2441 retval = -ENOTEMPTY;
2442 if (!empty_dir(new_inode))
2446 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2449 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2450 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2453 if (!new_inode && new_dir != old_dir &&
2454 EXT4_DIR_LINK_MAX(new_dir))
2456 BUFFER_TRACE(dir_bh, "get_write_access");
2457 retval = ext4_journal_get_write_access(handle, dir_bh);
2462 retval = ext4_add_entry(handle, new_dentry, old_inode);
2466 BUFFER_TRACE(new_bh, "get write access");
2467 retval = ext4_journal_get_write_access(handle, new_bh);
2470 new_de->inode = cpu_to_le32(old_inode->i_ino);
2471 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2472 EXT4_FEATURE_INCOMPAT_FILETYPE))
2473 new_de->file_type = old_de->file_type;
2474 new_dir->i_version++;
2475 new_dir->i_ctime = new_dir->i_mtime =
2476 ext4_current_time(new_dir);
2477 ext4_mark_inode_dirty(handle, new_dir);
2478 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2479 retval = ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2480 if (unlikely(retval)) {
2481 ext4_std_error(new_dir->i_sb, retval);
2489 * Like most other Unix systems, set the ctime for inodes on a
2492 old_inode->i_ctime = ext4_current_time(old_inode);
2493 ext4_mark_inode_dirty(handle, old_inode);
2498 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2499 old_de->name_len != old_dentry->d_name.len ||
2500 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2501 (retval = ext4_delete_entry(handle, old_dir,
2502 old_de, old_bh)) == -ENOENT) {
2503 /* old_de could have moved from under us during htree split, so
2504 * make sure that we are deleting the right entry. We might
2505 * also be pointing to a stale entry in the unused part of
2506 * old_bh so just checking inum and the name isn't enough. */
2507 struct buffer_head *old_bh2;
2508 struct ext4_dir_entry_2 *old_de2;
2510 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2512 retval = ext4_delete_entry(handle, old_dir,
2518 ext4_warning(old_dir->i_sb,
2519 "Deleting old file (%lu), %d, error=%d",
2520 old_dir->i_ino, old_dir->i_nlink, retval);
2524 ext4_dec_count(handle, new_inode);
2525 new_inode->i_ctime = ext4_current_time(new_inode);
2527 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2528 ext4_update_dx_flag(old_dir);
2530 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2531 cpu_to_le32(new_dir->i_ino);
2532 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2533 retval = ext4_handle_dirty_metadata(handle, old_inode, dir_bh);
2535 ext4_std_error(old_dir->i_sb, retval);
2538 ext4_dec_count(handle, old_dir);
2540 /* checked empty_dir above, can't have another parent,
2541 * ext4_dec_count() won't work for many-linked dirs */
2542 clear_nlink(new_inode);
2544 ext4_inc_count(handle, new_dir);
2545 ext4_update_dx_flag(new_dir);
2546 ext4_mark_inode_dirty(handle, new_dir);
2549 ext4_mark_inode_dirty(handle, old_dir);
2551 ext4_mark_inode_dirty(handle, new_inode);
2552 if (!new_inode->i_nlink)
2553 ext4_orphan_add(handle, new_inode);
2554 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2563 ext4_journal_stop(handle);
2564 if (retval == 0 && force_da_alloc)
2565 ext4_alloc_da_blocks(old_inode);
2570 * directories can handle most operations...
2572 const struct inode_operations ext4_dir_inode_operations = {
2573 .create = ext4_create,
2574 .lookup = ext4_lookup,
2576 .unlink = ext4_unlink,
2577 .symlink = ext4_symlink,
2578 .mkdir = ext4_mkdir,
2579 .rmdir = ext4_rmdir,
2580 .mknod = ext4_mknod,
2581 .rename = ext4_rename,
2582 .setattr = ext4_setattr,
2583 #ifdef CONFIG_EXT4_FS_XATTR
2584 .setxattr = generic_setxattr,
2585 .getxattr = generic_getxattr,
2586 .listxattr = ext4_listxattr,
2587 .removexattr = generic_removexattr,
2589 .get_acl = ext4_get_acl,
2590 .fiemap = ext4_fiemap,
2591 .permission = ext4_permission,
2592 .may_create = ext4_may_create,
2593 .may_delete = ext4_may_delete,
2596 const struct inode_operations ext4_special_inode_operations = {
2597 .setattr = ext4_setattr,
2598 #ifdef CONFIG_EXT4_FS_XATTR
2599 .setxattr = generic_setxattr,
2600 .getxattr = generic_getxattr,
2601 .listxattr = ext4_listxattr,
2602 .removexattr = generic_removexattr,
2604 .get_acl = ext4_get_acl,
2605 .permission = ext4_permission,
2606 .may_create = ext4_may_create,
2607 .may_delete = ext4_may_delete,