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/ext4_fs.h>
32 #include <linux/ext4_jbd2.h>
33 #include <linux/fcntl.h>
34 #include <linux/stat.h>
35 #include <linux/string.h>
36 #include <linux/quotaops.h>
37 #include <linux/buffer_head.h>
38 #include <linux/bio.h>
39 #include <linux/smp_lock.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,
57 struct buffer_head *bh;
59 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
61 if ((bh = ext4_bread(handle, inode, *block, 1, err))) {
62 inode->i_size += inode->i_sb->s_blocksize;
63 EXT4_I(inode)->i_disksize = inode->i_size;
64 ext4_journal_get_write_access(handle,bh);
70 #define assert(test) J_ASSERT(test)
74 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
78 #define dxtrace(command) command
80 #define dxtrace(command)
104 * dx_root_info is laid out so that if it should somehow get overlaid by a
105 * dirent the two low bits of the hash version will be zero. Therefore, the
106 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
111 struct fake_dirent dot;
113 struct fake_dirent dotdot;
117 __le32 reserved_zero;
119 u8 info_length; /* 8 */
124 struct dx_entry entries[0];
129 struct fake_dirent fake;
130 struct dx_entry entries[0];
136 struct buffer_head *bh;
137 struct dx_entry *entries;
147 #ifdef CONFIG_EXT4_INDEX
148 static inline unsigned dx_get_block (struct dx_entry *entry);
149 static void dx_set_block (struct dx_entry *entry, unsigned value);
150 static inline unsigned dx_get_hash (struct dx_entry *entry);
151 static void dx_set_hash (struct dx_entry *entry, unsigned value);
152 static unsigned dx_get_count (struct dx_entry *entries);
153 static unsigned dx_get_limit (struct dx_entry *entries);
154 static void dx_set_count (struct dx_entry *entries, unsigned value);
155 static void dx_set_limit (struct dx_entry *entries, unsigned value);
156 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
157 static unsigned dx_node_limit (struct inode *dir);
158 static struct dx_frame *dx_probe(struct dentry *dentry,
160 struct dx_hash_info *hinfo,
161 struct dx_frame *frame,
163 static void dx_release (struct dx_frame *frames);
164 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
165 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
166 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
167 static struct ext4_dir_entry_2 *dx_move_dirents (char *from, char *to,
168 struct dx_map_entry *offsets, int count);
169 static struct ext4_dir_entry_2* dx_pack_dirents (char *base, int size);
170 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
171 static int 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 dentry *dentry,
176 struct ext4_dir_entry_2 **res_dir, int *err);
177 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
178 struct inode *inode);
181 * Future: use high four bits of block for coalesce-on-delete flags
182 * Mask them off for now.
185 static inline unsigned dx_get_block (struct dx_entry *entry)
187 return le32_to_cpu(entry->block) & 0x00ffffff;
190 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
192 entry->block = cpu_to_le32(value);
195 static inline unsigned dx_get_hash (struct dx_entry *entry)
197 return le32_to_cpu(entry->hash);
200 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
202 entry->hash = cpu_to_le32(value);
205 static inline unsigned dx_get_count (struct dx_entry *entries)
207 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
210 static inline unsigned dx_get_limit (struct dx_entry *entries)
212 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
215 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
217 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
220 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
222 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
225 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
227 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
228 EXT4_DIR_REC_LEN(2) - infosize;
229 return 0? 20: entry_space / sizeof(struct dx_entry);
232 static inline unsigned dx_node_limit (struct inode *dir)
234 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
235 return 0? 22: entry_space / sizeof(struct dx_entry);
242 static void dx_show_index (char * label, struct dx_entry *entries)
244 int i, n = dx_get_count (entries);
245 printk("%s index ", label);
246 for (i = 0; i < n; i++) {
247 printk("%x->%u ", i? dx_get_hash(entries + i) :
248 0, dx_get_block(entries + i));
260 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
261 int size, int show_names)
263 unsigned names = 0, space = 0;
264 char *base = (char *) de;
265 struct dx_hash_info h = *hinfo;
268 while ((char *) de < base + size)
274 int len = de->name_len;
275 char *name = de->name;
276 while (len--) printk("%c", *name++);
277 ext4fs_dirhash(de->name, de->name_len, &h);
278 printk(":%x.%u ", h.hash,
279 ((char *) de - base));
281 space += EXT4_DIR_REC_LEN(de->name_len);
284 de = (struct ext4_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
286 printk("(%i)\n", names);
287 return (struct stats) { names, space, 1 };
290 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
291 struct dx_entry *entries, int levels)
293 unsigned blocksize = dir->i_sb->s_blocksize;
294 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
296 struct buffer_head *bh;
298 printk("%i indexed blocks...\n", count);
299 for (i = 0; i < count; i++, entries++)
301 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
302 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
304 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
305 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
307 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
308 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
309 names += stats.names;
310 space += stats.space;
311 bcount += stats.bcount;
315 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
316 names, space/bcount,(space/bcount)*100/blocksize);
317 return (struct stats) { names, space, bcount};
319 #endif /* DX_DEBUG */
322 * Probe for a directory leaf block to search.
324 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
325 * error in the directory index, and the caller should fall back to
326 * searching the directory normally. The callers of dx_probe **MUST**
327 * check for this error code, and make sure it never gets reflected
330 static struct dx_frame *
331 dx_probe(struct dentry *dentry, struct inode *dir,
332 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
334 unsigned count, indirect;
335 struct dx_entry *at, *entries, *p, *q, *m;
336 struct dx_root *root;
337 struct buffer_head *bh;
338 struct dx_frame *frame = frame_in;
343 dir = dentry->d_parent->d_inode;
344 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
346 root = (struct dx_root *) bh->b_data;
347 if (root->info.hash_version != DX_HASH_TEA &&
348 root->info.hash_version != DX_HASH_HALF_MD4 &&
349 root->info.hash_version != DX_HASH_LEGACY) {
350 ext4_warning(dir->i_sb, __FUNCTION__,
351 "Unrecognised inode hash code %d",
352 root->info.hash_version);
354 *err = ERR_BAD_DX_DIR;
357 hinfo->hash_version = root->info.hash_version;
358 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
360 ext4fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
363 if (root->info.unused_flags & 1) {
364 ext4_warning(dir->i_sb, __FUNCTION__,
365 "Unimplemented inode hash flags: %#06x",
366 root->info.unused_flags);
368 *err = ERR_BAD_DX_DIR;
372 if ((indirect = root->info.indirect_levels) > 1) {
373 ext4_warning(dir->i_sb, __FUNCTION__,
374 "Unimplemented inode hash depth: %#06x",
375 root->info.indirect_levels);
377 *err = ERR_BAD_DX_DIR;
381 entries = (struct dx_entry *) (((char *)&root->info) +
382 root->info.info_length);
383 assert(dx_get_limit(entries) == dx_root_limit(dir,
384 root->info.info_length));
385 dxtrace (printk("Look up %x", hash));
388 count = dx_get_count(entries);
389 assert (count && count <= dx_get_limit(entries));
391 q = entries + count - 1;
395 dxtrace(printk("."));
396 if (dx_get_hash(m) > hash)
402 if (0) // linear search cross check
404 unsigned n = count - 1;
408 dxtrace(printk(","));
409 if (dx_get_hash(++at) > hash)
415 assert (at == p - 1);
419 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
421 frame->entries = entries;
423 if (!indirect--) return frame;
424 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
426 at = entries = ((struct dx_node *) bh->b_data)->entries;
427 assert (dx_get_limit(entries) == dx_node_limit (dir));
431 while (frame >= frame_in) {
439 static void dx_release (struct dx_frame *frames)
441 if (frames[0].bh == NULL)
444 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
445 brelse(frames[1].bh);
446 brelse(frames[0].bh);
450 * This function increments the frame pointer to search the next leaf
451 * block, and reads in the necessary intervening nodes if the search
452 * should be necessary. Whether or not the search is necessary is
453 * controlled by the hash parameter. If the hash value is even, then
454 * the search is only continued if the next block starts with that
455 * hash value. This is used if we are searching for a specific file.
457 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
459 * This function returns 1 if the caller should continue to search,
460 * or 0 if it should not. If there is an error reading one of the
461 * index blocks, it will a negative error code.
463 * If start_hash is non-null, it will be filled in with the starting
464 * hash of the next page.
466 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
467 struct dx_frame *frame,
468 struct dx_frame *frames,
472 struct buffer_head *bh;
473 int err, num_frames = 0;
478 * Find the next leaf page by incrementing the frame pointer.
479 * If we run out of entries in the interior node, loop around and
480 * increment pointer in the parent node. When we break out of
481 * this loop, num_frames indicates the number of interior
482 * nodes need to be read.
485 if (++(p->at) < p->entries + dx_get_count(p->entries))
494 * If the hash is 1, then continue only if the next page has a
495 * continuation hash of any value. This is used for readdir
496 * handling. Otherwise, check to see if the hash matches the
497 * desired contiuation hash. If it doesn't, return since
498 * there's no point to read in the successive index pages.
500 bhash = dx_get_hash(p->at);
503 if ((hash & 1) == 0) {
504 if ((bhash & ~1) != hash)
508 * If the hash is HASH_NB_ALWAYS, we always go to the next
509 * block so no check is necessary
511 while (num_frames--) {
512 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
514 return err; /* Failure */
518 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
525 * p is at least 6 bytes before the end of page
527 static inline struct ext4_dir_entry_2 *ext4_next_entry(struct ext4_dir_entry_2 *p)
529 return (struct ext4_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
533 * This function fills a red-black tree with information from a
534 * directory block. It returns the number directory entries loaded
535 * into the tree. If there is an error it is returned in err.
537 static int htree_dirblock_to_tree(struct file *dir_file,
538 struct inode *dir, int block,
539 struct dx_hash_info *hinfo,
540 __u32 start_hash, __u32 start_minor_hash)
542 struct buffer_head *bh;
543 struct ext4_dir_entry_2 *de, *top;
546 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
547 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
550 de = (struct ext4_dir_entry_2 *) bh->b_data;
551 top = (struct ext4_dir_entry_2 *) ((char *) de +
552 dir->i_sb->s_blocksize -
553 EXT4_DIR_REC_LEN(0));
554 for (; de < top; de = ext4_next_entry(de)) {
555 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
556 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
557 +((char *)de - bh->b_data))) {
558 /* On error, skip the f_pos to the next block. */
559 dir_file->f_pos = (dir_file->f_pos |
560 (dir->i_sb->s_blocksize - 1)) + 1;
564 ext4fs_dirhash(de->name, de->name_len, hinfo);
565 if ((hinfo->hash < start_hash) ||
566 ((hinfo->hash == start_hash) &&
567 (hinfo->minor_hash < start_minor_hash)))
571 if ((err = ext4_htree_store_dirent(dir_file,
572 hinfo->hash, hinfo->minor_hash, de)) != 0) {
584 * This function fills a red-black tree with information from a
585 * directory. We start scanning the directory in hash order, starting
586 * at start_hash and start_minor_hash.
588 * This function returns the number of entries inserted into the tree,
589 * or a negative error code.
591 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
592 __u32 start_minor_hash, __u32 *next_hash)
594 struct dx_hash_info hinfo;
595 struct ext4_dir_entry_2 *de;
596 struct dx_frame frames[2], *frame;
603 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
605 dir = dir_file->f_path.dentry->d_inode;
606 if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
607 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
608 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
609 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
610 start_hash, start_minor_hash);
614 hinfo.hash = start_hash;
615 hinfo.minor_hash = 0;
616 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
620 /* Add '.' and '..' from the htree header */
621 if (!start_hash && !start_minor_hash) {
622 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
623 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
627 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
628 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
629 de = ext4_next_entry(de);
630 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
636 block = dx_get_block(frame->at);
637 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
638 start_hash, start_minor_hash);
645 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
646 frame, frames, &hashval);
647 *next_hash = hashval;
653 * Stop if: (a) there are no more entries, or
654 * (b) we have inserted at least one entry and the
655 * next hash value is not a continuation
658 (count && ((hashval & 1) == 0)))
662 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
672 * Directory block splitting, compacting
675 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
676 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
679 char *base = (char *) de;
680 struct dx_hash_info h = *hinfo;
682 while ((char *) de < base + size)
684 if (de->name_len && de->inode) {
685 ext4fs_dirhash(de->name, de->name_len, &h);
687 map_tail->hash = h.hash;
688 map_tail->offs = (u32) ((char *) de - base);
692 /* XXX: do we need to check rec_len == 0 case? -Chris */
693 de = (struct ext4_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
698 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
700 struct dx_map_entry *p, *q, *top = map + count - 1;
702 /* Combsort until bubble sort doesn't suck */
705 if (count - 9 < 2) /* 9, 10 -> 11 */
707 for (p = top, q = p - count; q >= map; p--, q--)
708 if (p->hash < q->hash)
711 /* Garden variety bubble sort */
716 if (q[1].hash >= q[0].hash)
724 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
726 struct dx_entry *entries = frame->entries;
727 struct dx_entry *old = frame->at, *new = old + 1;
728 int count = dx_get_count(entries);
730 assert(count < dx_get_limit(entries));
731 assert(old < entries + count);
732 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
733 dx_set_hash(new, hash);
734 dx_set_block(new, block);
735 dx_set_count(entries, count + 1);
740 static void ext4_update_dx_flag(struct inode *inode)
742 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
743 EXT4_FEATURE_COMPAT_DIR_INDEX))
744 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
748 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
750 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
751 * `de != NULL' is guaranteed by caller.
753 static inline int ext4_match (int len, const char * const name,
754 struct ext4_dir_entry_2 * de)
756 if (len != de->name_len)
760 return !memcmp(name, de->name, len);
764 * Returns 0 if not found, -1 on failure, and 1 on success
766 static inline int search_dirblock(struct buffer_head * bh,
768 struct dentry *dentry,
769 unsigned long offset,
770 struct ext4_dir_entry_2 ** res_dir)
772 struct ext4_dir_entry_2 * de;
775 const char *name = dentry->d_name.name;
776 int namelen = dentry->d_name.len;
778 de = (struct ext4_dir_entry_2 *) bh->b_data;
779 dlimit = bh->b_data + dir->i_sb->s_blocksize;
780 while ((char *) de < dlimit) {
781 /* this code is executed quadratically often */
782 /* do minimal checking `by hand' */
784 if ((char *) de + namelen <= dlimit &&
785 ext4_match (namelen, name, de)) {
786 /* found a match - just to be sure, do a full check */
787 if (!ext4_check_dir_entry("ext4_find_entry",
788 dir, de, bh, offset))
793 /* prevent looping on a bad block */
794 de_len = le16_to_cpu(de->rec_len);
798 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
807 * finds an entry in the specified directory with the wanted name. It
808 * returns the cache buffer in which the entry was found, and the entry
809 * itself (as a parameter - res_dir). It does NOT read the inode of the
810 * entry - you'll have to do that yourself if you want to.
812 * The returned buffer_head has ->b_count elevated. The caller is expected
813 * to brelse() it when appropriate.
815 static struct buffer_head * ext4_find_entry (struct dentry *dentry,
816 struct ext4_dir_entry_2 ** res_dir)
818 struct super_block * sb;
819 struct buffer_head * bh_use[NAMEI_RA_SIZE];
820 struct buffer_head * bh, *ret = NULL;
821 unsigned long start, block, b;
822 int ra_max = 0; /* Number of bh's in the readahead
824 int ra_ptr = 0; /* Current index into readahead
828 struct inode *dir = dentry->d_parent->d_inode;
835 blocksize = sb->s_blocksize;
836 namelen = dentry->d_name.len;
837 name = dentry->d_name.name;
838 if (namelen > EXT4_NAME_LEN)
840 #ifdef CONFIG_EXT4_INDEX
842 bh = ext4_dx_find_entry(dentry, res_dir, &err);
844 * On success, or if the error was file not found,
845 * return. Otherwise, fall back to doing a search the
848 if (bh || (err != ERR_BAD_DX_DIR))
850 dxtrace(printk("ext4_find_entry: dx failed, falling back\n"));
853 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
854 start = EXT4_I(dir)->i_dir_start_lookup;
855 if (start >= nblocks)
861 * We deal with the read-ahead logic here.
863 if (ra_ptr >= ra_max) {
864 /* Refill the readahead buffer */
867 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
869 * Terminate if we reach the end of the
870 * directory and must wrap, or if our
871 * search has finished at this block.
873 if (b >= nblocks || (num && block == start)) {
874 bh_use[ra_max] = NULL;
878 bh = ext4_getblk(NULL, dir, b++, 0, &err);
881 ll_rw_block(READ_META, 1, &bh);
884 if ((bh = bh_use[ra_ptr++]) == NULL)
887 if (!buffer_uptodate(bh)) {
888 /* read error, skip block & hope for the best */
889 ext4_error(sb, __FUNCTION__, "reading directory #%lu "
890 "offset %lu", dir->i_ino, block);
894 i = search_dirblock(bh, dir, dentry,
895 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
897 EXT4_I(dir)->i_dir_start_lookup = block;
899 goto cleanup_and_exit;
903 goto cleanup_and_exit;
906 if (++block >= nblocks)
908 } while (block != start);
911 * If the directory has grown while we were searching, then
912 * search the last part of the directory before giving up.
915 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
916 if (block < nblocks) {
922 /* Clean up the read-ahead blocks */
923 for (; ra_ptr < ra_max; ra_ptr++)
924 brelse (bh_use[ra_ptr]);
928 #ifdef CONFIG_EXT4_INDEX
929 static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
930 struct ext4_dir_entry_2 **res_dir, int *err)
932 struct super_block * sb;
933 struct dx_hash_info hinfo;
935 struct dx_frame frames[2], *frame;
936 struct ext4_dir_entry_2 *de, *top;
937 struct buffer_head *bh;
940 int namelen = dentry->d_name.len;
941 const u8 *name = dentry->d_name.name;
942 struct inode *dir = dentry->d_parent->d_inode;
945 /* NFS may look up ".." - look at dx_root directory block */
946 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
947 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
951 frame->bh = NULL; /* for dx_release() */
952 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
953 dx_set_block(frame->at, 0); /* dx_root block is 0 */
957 block = dx_get_block(frame->at);
958 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
960 de = (struct ext4_dir_entry_2 *) bh->b_data;
961 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
962 EXT4_DIR_REC_LEN(0));
963 for (; de < top; de = ext4_next_entry(de))
964 if (ext4_match (namelen, name, de)) {
965 if (!ext4_check_dir_entry("ext4_find_entry",
967 (block<<EXT4_BLOCK_SIZE_BITS(sb))
968 +((char *)de - bh->b_data))) {
970 *err = ERR_BAD_DX_DIR;
978 /* Check to see if we should continue to search */
979 retval = ext4_htree_next_block(dir, hash, frame,
982 ext4_warning(sb, __FUNCTION__,
983 "error reading index page in directory #%lu",
988 } while (retval == 1);
992 dxtrace(printk("%s not found\n", name));
998 static struct dentry *ext4_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1000 struct inode * inode;
1001 struct ext4_dir_entry_2 * de;
1002 struct buffer_head * bh;
1004 if (dentry->d_name.len > EXT4_NAME_LEN)
1005 return ERR_PTR(-ENAMETOOLONG);
1007 bh = ext4_find_entry(dentry, &de);
1010 unsigned long ino = le32_to_cpu(de->inode);
1012 if (!ext4_valid_inum(dir->i_sb, ino)) {
1013 ext4_error(dir->i_sb, "ext4_lookup",
1014 "bad inode number: %lu", ino);
1017 inode = iget(dir->i_sb, ino);
1020 return ERR_PTR(-EACCES);
1022 return d_splice_alias(inode, dentry);
1026 struct dentry *ext4_get_parent(struct dentry *child)
1029 struct dentry *parent;
1030 struct inode *inode;
1031 struct dentry dotdot;
1032 struct ext4_dir_entry_2 * de;
1033 struct buffer_head *bh;
1035 dotdot.d_name.name = "..";
1036 dotdot.d_name.len = 2;
1037 dotdot.d_parent = child; /* confusing, isn't it! */
1039 bh = ext4_find_entry(&dotdot, &de);
1042 return ERR_PTR(-ENOENT);
1043 ino = le32_to_cpu(de->inode);
1046 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1047 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1048 "bad inode number: %lu", ino);
1051 inode = iget(child->d_inode->i_sb, ino);
1054 return ERR_PTR(-EACCES);
1056 parent = d_alloc_anon(inode);
1059 parent = ERR_PTR(-ENOMEM);
1065 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1066 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1067 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1068 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1069 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1070 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1071 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1072 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
1075 static inline void ext4_set_de_type(struct super_block *sb,
1076 struct ext4_dir_entry_2 *de,
1078 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1079 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1082 #ifdef CONFIG_EXT4_INDEX
1083 static struct ext4_dir_entry_2 *
1084 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1086 unsigned rec_len = 0;
1089 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1090 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1091 memcpy (to, de, rec_len);
1092 ((struct ext4_dir_entry_2 *) to)->rec_len =
1093 cpu_to_le16(rec_len);
1098 return (struct ext4_dir_entry_2 *) (to - rec_len);
1101 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
1103 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1104 unsigned rec_len = 0;
1107 while ((char*)de < base + size) {
1108 next = (struct ext4_dir_entry_2 *) ((char *) de +
1109 le16_to_cpu(de->rec_len));
1110 if (de->inode && de->name_len) {
1111 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1113 memmove(to, de, rec_len);
1114 to->rec_len = cpu_to_le16(rec_len);
1116 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1123 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1124 struct buffer_head **bh,struct dx_frame *frame,
1125 struct dx_hash_info *hinfo, int *error)
1127 unsigned blocksize = dir->i_sb->s_blocksize;
1128 unsigned count, continued;
1129 struct buffer_head *bh2;
1132 struct dx_map_entry *map;
1133 char *data1 = (*bh)->b_data, *data2;
1135 struct ext4_dir_entry_2 *de = NULL, *de2;
1138 bh2 = ext4_append (handle, dir, &newblock, &err);
1145 BUFFER_TRACE(*bh, "get_write_access");
1146 err = ext4_journal_get_write_access(handle, *bh);
1150 BUFFER_TRACE(frame->bh, "get_write_access");
1151 err = ext4_journal_get_write_access(handle, frame->bh);
1155 data2 = bh2->b_data;
1157 /* create map in the end of data2 block */
1158 map = (struct dx_map_entry *) (data2 + blocksize);
1159 count = dx_make_map ((struct ext4_dir_entry_2 *) data1,
1160 blocksize, hinfo, map);
1162 split = count/2; // need to adjust to actual middle
1163 dx_sort_map (map, count);
1164 hash2 = map[split].hash;
1165 continued = hash2 == map[split - 1].hash;
1166 dxtrace(printk("Split block %i at %x, %i/%i\n",
1167 dx_get_block(frame->at), hash2, split, count-split));
1169 /* Fancy dance to stay within two buffers */
1170 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1171 de = dx_pack_dirents(data1,blocksize);
1172 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1173 de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1174 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1175 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1177 /* Which block gets the new entry? */
1178 if (hinfo->hash >= hash2)
1183 dx_insert_block (frame, hash2 + continued, newblock);
1184 err = ext4_journal_dirty_metadata (handle, bh2);
1187 err = ext4_journal_dirty_metadata (handle, frame->bh);
1191 dxtrace(dx_show_index ("frame", frame->entries));
1198 ext4_std_error(dir->i_sb, err);
1207 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1208 * it points to a directory entry which is guaranteed to be large
1209 * enough for new directory entry. If de is NULL, then
1210 * add_dirent_to_buf will attempt search the directory block for
1211 * space. It will return -ENOSPC if no space is available, and -EIO
1212 * and -EEXIST if directory entry already exists.
1214 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1215 * all other cases bh is released.
1217 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1218 struct inode *inode, struct ext4_dir_entry_2 *de,
1219 struct buffer_head * bh)
1221 struct inode *dir = dentry->d_parent->d_inode;
1222 const char *name = dentry->d_name.name;
1223 int namelen = dentry->d_name.len;
1224 unsigned long offset = 0;
1225 unsigned short reclen;
1226 int nlen, rlen, err;
1229 reclen = EXT4_DIR_REC_LEN(namelen);
1231 de = (struct ext4_dir_entry_2 *)bh->b_data;
1232 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1233 while ((char *) de <= top) {
1234 if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1239 if (ext4_match (namelen, name, de)) {
1243 nlen = EXT4_DIR_REC_LEN(de->name_len);
1244 rlen = le16_to_cpu(de->rec_len);
1245 if ((de->inode? rlen - nlen: rlen) >= reclen)
1247 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1250 if ((char *) de > top)
1253 BUFFER_TRACE(bh, "get_write_access");
1254 err = ext4_journal_get_write_access(handle, bh);
1256 ext4_std_error(dir->i_sb, err);
1261 /* By now the buffer is marked for journaling */
1262 nlen = EXT4_DIR_REC_LEN(de->name_len);
1263 rlen = le16_to_cpu(de->rec_len);
1265 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1266 de1->rec_len = cpu_to_le16(rlen - nlen);
1267 de->rec_len = cpu_to_le16(nlen);
1270 de->file_type = EXT4_FT_UNKNOWN;
1272 de->inode = cpu_to_le32(inode->i_ino);
1273 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1276 de->name_len = namelen;
1277 memcpy (de->name, name, namelen);
1279 * XXX shouldn't update any times until successful
1280 * completion of syscall, but too many callers depend
1283 * XXX similarly, too many callers depend on
1284 * ext4_new_inode() setting the times, but error
1285 * recovery deletes the inode, so the worst that can
1286 * happen is that the times are slightly out of date
1287 * and/or different from the directory change time.
1289 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1290 ext4_update_dx_flag(dir);
1292 ext4_mark_inode_dirty(handle, dir);
1293 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1294 err = ext4_journal_dirty_metadata(handle, bh);
1296 ext4_std_error(dir->i_sb, err);
1301 #ifdef CONFIG_EXT4_INDEX
1303 * This converts a one block unindexed directory to a 3 block indexed
1304 * directory, and adds the dentry to the indexed directory.
1306 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1307 struct inode *inode, struct buffer_head *bh)
1309 struct inode *dir = dentry->d_parent->d_inode;
1310 const char *name = dentry->d_name.name;
1311 int namelen = dentry->d_name.len;
1312 struct buffer_head *bh2;
1313 struct dx_root *root;
1314 struct dx_frame frames[2], *frame;
1315 struct dx_entry *entries;
1316 struct ext4_dir_entry_2 *de, *de2;
1321 struct dx_hash_info hinfo;
1323 struct fake_dirent *fde;
1325 blocksize = dir->i_sb->s_blocksize;
1326 dxtrace(printk("Creating index\n"));
1327 retval = ext4_journal_get_write_access(handle, bh);
1329 ext4_std_error(dir->i_sb, retval);
1333 root = (struct dx_root *) bh->b_data;
1335 bh2 = ext4_append (handle, dir, &block, &retval);
1340 EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1341 data1 = bh2->b_data;
1343 /* The 0th block becomes the root, move the dirents out */
1344 fde = &root->dotdot;
1345 de = (struct ext4_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1346 len = ((char *) root) + blocksize - (char *) de;
1347 memcpy (data1, de, len);
1348 de = (struct ext4_dir_entry_2 *) data1;
1350 while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1352 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1353 /* Initialize the root; the dot dirents already exist */
1354 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1355 de->rec_len = cpu_to_le16(blocksize - EXT4_DIR_REC_LEN(2));
1356 memset (&root->info, 0, sizeof(root->info));
1357 root->info.info_length = sizeof(root->info);
1358 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1359 entries = root->entries;
1360 dx_set_block (entries, 1);
1361 dx_set_count (entries, 1);
1362 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1364 /* Initialize as for dx_probe */
1365 hinfo.hash_version = root->info.hash_version;
1366 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1367 ext4fs_dirhash(name, namelen, &hinfo);
1369 frame->entries = entries;
1370 frame->at = entries;
1373 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1374 dx_release (frames);
1378 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1385 * adds a file entry to the specified directory, using the same
1386 * semantics as ext4_find_entry(). It returns NULL if it failed.
1388 * NOTE!! The inode part of 'de' is left at 0 - which means you
1389 * may not sleep between calling this and putting something into
1390 * the entry, as someone else might have used it while you slept.
1392 static int ext4_add_entry (handle_t *handle, struct dentry *dentry,
1393 struct inode *inode)
1395 struct inode *dir = dentry->d_parent->d_inode;
1396 unsigned long offset;
1397 struct buffer_head * bh;
1398 struct ext4_dir_entry_2 *de;
1399 struct super_block * sb;
1401 #ifdef CONFIG_EXT4_INDEX
1408 blocksize = sb->s_blocksize;
1409 if (!dentry->d_name.len)
1411 #ifdef CONFIG_EXT4_INDEX
1413 retval = ext4_dx_add_entry(handle, dentry, inode);
1414 if (!retval || (retval != ERR_BAD_DX_DIR))
1416 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1418 ext4_mark_inode_dirty(handle, dir);
1421 blocks = dir->i_size >> sb->s_blocksize_bits;
1422 for (block = 0, offset = 0; block < blocks; block++) {
1423 bh = ext4_bread(handle, dir, block, 0, &retval);
1426 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1427 if (retval != -ENOSPC)
1430 #ifdef CONFIG_EXT4_INDEX
1431 if (blocks == 1 && !dx_fallback &&
1432 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1433 return make_indexed_dir(handle, dentry, inode, bh);
1437 bh = ext4_append(handle, dir, &block, &retval);
1440 de = (struct ext4_dir_entry_2 *) bh->b_data;
1442 de->rec_len = cpu_to_le16(blocksize);
1443 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1446 #ifdef CONFIG_EXT4_INDEX
1448 * Returns 0 for success, or a negative error value
1450 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1451 struct inode *inode)
1453 struct dx_frame frames[2], *frame;
1454 struct dx_entry *entries, *at;
1455 struct dx_hash_info hinfo;
1456 struct buffer_head * bh;
1457 struct inode *dir = dentry->d_parent->d_inode;
1458 struct super_block * sb = dir->i_sb;
1459 struct ext4_dir_entry_2 *de;
1462 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1465 entries = frame->entries;
1468 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1471 BUFFER_TRACE(bh, "get_write_access");
1472 err = ext4_journal_get_write_access(handle, bh);
1476 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1477 if (err != -ENOSPC) {
1482 /* Block full, should compress but for now just split */
1483 dxtrace(printk("using %u of %u node entries\n",
1484 dx_get_count(entries), dx_get_limit(entries)));
1485 /* Need to split index? */
1486 if (dx_get_count(entries) == dx_get_limit(entries)) {
1488 unsigned icount = dx_get_count(entries);
1489 int levels = frame - frames;
1490 struct dx_entry *entries2;
1491 struct dx_node *node2;
1492 struct buffer_head *bh2;
1494 if (levels && (dx_get_count(frames->entries) ==
1495 dx_get_limit(frames->entries))) {
1496 ext4_warning(sb, __FUNCTION__,
1497 "Directory index full!");
1501 bh2 = ext4_append (handle, dir, &newblock, &err);
1504 node2 = (struct dx_node *)(bh2->b_data);
1505 entries2 = node2->entries;
1506 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1507 node2->fake.inode = 0;
1508 BUFFER_TRACE(frame->bh, "get_write_access");
1509 err = ext4_journal_get_write_access(handle, frame->bh);
1513 unsigned icount1 = icount/2, icount2 = icount - icount1;
1514 unsigned hash2 = dx_get_hash(entries + icount1);
1515 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1517 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1518 err = ext4_journal_get_write_access(handle,
1523 memcpy ((char *) entries2, (char *) (entries + icount1),
1524 icount2 * sizeof(struct dx_entry));
1525 dx_set_count (entries, icount1);
1526 dx_set_count (entries2, icount2);
1527 dx_set_limit (entries2, dx_node_limit(dir));
1529 /* Which index block gets the new entry? */
1530 if (at - entries >= icount1) {
1531 frame->at = at = at - entries - icount1 + entries2;
1532 frame->entries = entries = entries2;
1533 swap(frame->bh, bh2);
1535 dx_insert_block (frames + 0, hash2, newblock);
1536 dxtrace(dx_show_index ("node", frames[1].entries));
1537 dxtrace(dx_show_index ("node",
1538 ((struct dx_node *) bh2->b_data)->entries));
1539 err = ext4_journal_dirty_metadata(handle, bh2);
1544 dxtrace(printk("Creating second level index...\n"));
1545 memcpy((char *) entries2, (char *) entries,
1546 icount * sizeof(struct dx_entry));
1547 dx_set_limit(entries2, dx_node_limit(dir));
1550 dx_set_count(entries, 1);
1551 dx_set_block(entries + 0, newblock);
1552 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1554 /* Add new access path frame */
1556 frame->at = at = at - entries + entries2;
1557 frame->entries = entries = entries2;
1559 err = ext4_journal_get_write_access(handle,
1564 ext4_journal_dirty_metadata(handle, frames[0].bh);
1566 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1569 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1574 ext4_std_error(dir->i_sb, err);
1584 * ext4_delete_entry deletes a directory entry by merging it with the
1587 static int ext4_delete_entry (handle_t *handle,
1589 struct ext4_dir_entry_2 * de_del,
1590 struct buffer_head * bh)
1592 struct ext4_dir_entry_2 * de, * pde;
1597 de = (struct ext4_dir_entry_2 *) bh->b_data;
1598 while (i < bh->b_size) {
1599 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1602 BUFFER_TRACE(bh, "get_write_access");
1603 ext4_journal_get_write_access(handle, bh);
1606 cpu_to_le16(le16_to_cpu(pde->rec_len) +
1607 le16_to_cpu(de->rec_len));
1611 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1612 ext4_journal_dirty_metadata(handle, bh);
1615 i += le16_to_cpu(de->rec_len);
1617 de = (struct ext4_dir_entry_2 *)
1618 ((char *) de + le16_to_cpu(de->rec_len));
1623 static int ext4_add_nondir(handle_t *handle,
1624 struct dentry *dentry, struct inode *inode)
1626 int err = ext4_add_entry(handle, dentry, inode);
1628 ext4_mark_inode_dirty(handle, inode);
1629 d_instantiate(dentry, inode);
1638 * By the time this is called, we already have created
1639 * the directory cache entry for the new file, but it
1640 * is so far negative - it has no inode.
1642 * If the create succeeds, we fill in the inode information
1643 * with d_instantiate().
1645 static int ext4_create (struct inode * dir, struct dentry * dentry, int mode,
1646 struct nameidata *nd)
1649 struct inode * inode;
1650 int err, retries = 0;
1653 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1654 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1655 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1657 return PTR_ERR(handle);
1659 if (IS_DIRSYNC(dir))
1662 inode = ext4_new_inode (handle, dir, mode);
1663 err = PTR_ERR(inode);
1664 if (!IS_ERR(inode)) {
1665 inode->i_op = &ext4_file_inode_operations;
1666 inode->i_fop = &ext4_file_operations;
1667 ext4_set_aops(inode);
1668 err = ext4_add_nondir(handle, dentry, inode);
1670 ext4_journal_stop(handle);
1671 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1676 static int ext4_mknod (struct inode * dir, struct dentry *dentry,
1677 int mode, dev_t rdev)
1680 struct inode *inode;
1681 int err, retries = 0;
1683 if (!new_valid_dev(rdev))
1687 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1688 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1689 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1691 return PTR_ERR(handle);
1693 if (IS_DIRSYNC(dir))
1696 inode = ext4_new_inode (handle, dir, mode);
1697 err = PTR_ERR(inode);
1698 if (!IS_ERR(inode)) {
1699 init_special_inode(inode, inode->i_mode, rdev);
1700 #ifdef CONFIG_EXT4DEV_FS_XATTR
1701 inode->i_op = &ext4_special_inode_operations;
1703 err = ext4_add_nondir(handle, dentry, inode);
1705 ext4_journal_stop(handle);
1706 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1711 static int ext4_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1714 struct inode * inode;
1715 struct buffer_head * dir_block;
1716 struct ext4_dir_entry_2 * de;
1717 int err, retries = 0;
1719 if (dir->i_nlink >= EXT4_LINK_MAX)
1723 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1724 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1725 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1727 return PTR_ERR(handle);
1729 if (IS_DIRSYNC(dir))
1732 inode = ext4_new_inode (handle, dir, S_IFDIR | mode);
1733 err = PTR_ERR(inode);
1737 inode->i_op = &ext4_dir_inode_operations;
1738 inode->i_fop = &ext4_dir_operations;
1739 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1740 dir_block = ext4_bread (handle, inode, 0, 1, &err);
1742 drop_nlink(inode); /* is this nlink == 0? */
1743 ext4_mark_inode_dirty(handle, inode);
1747 BUFFER_TRACE(dir_block, "get_write_access");
1748 ext4_journal_get_write_access(handle, dir_block);
1749 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1750 de->inode = cpu_to_le32(inode->i_ino);
1752 de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de->name_len));
1753 strcpy (de->name, ".");
1754 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1755 de = (struct ext4_dir_entry_2 *)
1756 ((char *) de + le16_to_cpu(de->rec_len));
1757 de->inode = cpu_to_le32(dir->i_ino);
1758 de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT4_DIR_REC_LEN(1));
1760 strcpy (de->name, "..");
1761 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1763 BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata");
1764 ext4_journal_dirty_metadata(handle, dir_block);
1766 ext4_mark_inode_dirty(handle, inode);
1767 err = ext4_add_entry (handle, dentry, inode);
1770 ext4_mark_inode_dirty(handle, inode);
1775 ext4_update_dx_flag(dir);
1776 ext4_mark_inode_dirty(handle, dir);
1777 d_instantiate(dentry, inode);
1779 ext4_journal_stop(handle);
1780 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1786 * routine to check that the specified directory is empty (for rmdir)
1788 static int empty_dir (struct inode * inode)
1790 unsigned long offset;
1791 struct buffer_head * bh;
1792 struct ext4_dir_entry_2 * de, * de1;
1793 struct super_block * sb;
1797 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1798 !(bh = ext4_bread (NULL, inode, 0, 0, &err))) {
1800 ext4_error(inode->i_sb, __FUNCTION__,
1801 "error %d reading directory #%lu offset 0",
1804 ext4_warning(inode->i_sb, __FUNCTION__,
1805 "bad directory (dir #%lu) - no data block",
1809 de = (struct ext4_dir_entry_2 *) bh->b_data;
1810 de1 = (struct ext4_dir_entry_2 *)
1811 ((char *) de + le16_to_cpu(de->rec_len));
1812 if (le32_to_cpu(de->inode) != inode->i_ino ||
1813 !le32_to_cpu(de1->inode) ||
1814 strcmp (".", de->name) ||
1815 strcmp ("..", de1->name)) {
1816 ext4_warning (inode->i_sb, "empty_dir",
1817 "bad directory (dir #%lu) - no `.' or `..'",
1822 offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1823 de = (struct ext4_dir_entry_2 *)
1824 ((char *) de1 + le16_to_cpu(de1->rec_len));
1825 while (offset < inode->i_size ) {
1827 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1830 bh = ext4_bread (NULL, inode,
1831 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1834 ext4_error(sb, __FUNCTION__,
1835 "error %d reading directory"
1837 err, inode->i_ino, offset);
1838 offset += sb->s_blocksize;
1841 de = (struct ext4_dir_entry_2 *) bh->b_data;
1843 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1844 de = (struct ext4_dir_entry_2 *)(bh->b_data +
1846 offset = (offset | (sb->s_blocksize - 1)) + 1;
1849 if (le32_to_cpu(de->inode)) {
1853 offset += le16_to_cpu(de->rec_len);
1854 de = (struct ext4_dir_entry_2 *)
1855 ((char *) de + le16_to_cpu(de->rec_len));
1861 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1862 * such inodes, starting at the superblock, in case we crash before the
1863 * file is closed/deleted, or in case the inode truncate spans multiple
1864 * transactions and the last transaction is not recovered after a crash.
1866 * At filesystem recovery time, we walk this list deleting unlinked
1867 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1869 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1871 struct super_block *sb = inode->i_sb;
1872 struct ext4_iloc iloc;
1876 if (!list_empty(&EXT4_I(inode)->i_orphan))
1879 /* Orphan handling is only valid for files with data blocks
1880 * being truncated, or files being unlinked. */
1882 /* @@@ FIXME: Observation from aviro:
1883 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
1884 * here (on lock_super()), so race with ext4_link() which might bump
1885 * ->i_nlink. For, say it, character device. Not a regular file,
1886 * not a directory, not a symlink and ->i_nlink > 0.
1888 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1889 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1891 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1892 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1896 err = ext4_reserve_inode_write(handle, inode, &iloc);
1900 /* Insert this inode at the head of the on-disk orphan list... */
1901 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1902 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1903 err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1904 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
1908 /* Only add to the head of the in-memory list if all the
1909 * previous operations succeeded. If the orphan_add is going to
1910 * fail (possibly taking the journal offline), we can't risk
1911 * leaving the inode on the orphan list: stray orphan-list
1912 * entries can cause panics at unmount time.
1914 * This is safe: on error we're going to ignore the orphan list
1915 * anyway on the next recovery. */
1917 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1919 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1920 jbd_debug(4, "orphan inode %lu will point to %d\n",
1921 inode->i_ino, NEXT_ORPHAN(inode));
1924 ext4_std_error(inode->i_sb, err);
1929 * ext4_orphan_del() removes an unlinked or truncated inode from the list
1930 * of such inodes stored on disk, because it is finally being cleaned up.
1932 int ext4_orphan_del(handle_t *handle, struct inode *inode)
1934 struct list_head *prev;
1935 struct ext4_inode_info *ei = EXT4_I(inode);
1936 struct ext4_sb_info *sbi;
1937 unsigned long ino_next;
1938 struct ext4_iloc iloc;
1941 lock_super(inode->i_sb);
1942 if (list_empty(&ei->i_orphan)) {
1943 unlock_super(inode->i_sb);
1947 ino_next = NEXT_ORPHAN(inode);
1948 prev = ei->i_orphan.prev;
1949 sbi = EXT4_SB(inode->i_sb);
1951 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1953 list_del_init(&ei->i_orphan);
1955 /* If we're on an error path, we may not have a valid
1956 * transaction handle with which to update the orphan list on
1957 * disk, but we still need to remove the inode from the linked
1958 * list in memory. */
1962 err = ext4_reserve_inode_write(handle, inode, &iloc);
1966 if (prev == &sbi->s_orphan) {
1967 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1968 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1969 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1972 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1973 err = ext4_journal_dirty_metadata(handle, sbi->s_sbh);
1975 struct ext4_iloc iloc2;
1976 struct inode *i_prev =
1977 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
1979 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1980 i_prev->i_ino, ino_next);
1981 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
1984 NEXT_ORPHAN(i_prev) = ino_next;
1985 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
1989 NEXT_ORPHAN(inode) = 0;
1990 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
1993 ext4_std_error(inode->i_sb, err);
1995 unlock_super(inode->i_sb);
2003 static int ext4_rmdir (struct inode * dir, struct dentry *dentry)
2006 struct inode * inode;
2007 struct buffer_head * bh;
2008 struct ext4_dir_entry_2 * de;
2011 /* Initialize quotas before so that eventual writes go in
2012 * separate transaction */
2013 DQUOT_INIT(dentry->d_inode);
2014 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2016 return PTR_ERR(handle);
2019 bh = ext4_find_entry (dentry, &de);
2023 if (IS_DIRSYNC(dir))
2026 inode = dentry->d_inode;
2029 if (le32_to_cpu(de->inode) != inode->i_ino)
2032 retval = -ENOTEMPTY;
2033 if (!empty_dir (inode))
2036 retval = ext4_delete_entry(handle, dir, de, bh);
2039 if (inode->i_nlink != 2)
2040 ext4_warning (inode->i_sb, "ext4_rmdir",
2041 "empty directory has nlink!=2 (%d)",
2045 /* There's no need to set i_disksize: the fact that i_nlink is
2046 * zero will ensure that the right thing happens during any
2049 ext4_orphan_add(handle, inode);
2050 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2051 ext4_mark_inode_dirty(handle, inode);
2053 ext4_update_dx_flag(dir);
2054 ext4_mark_inode_dirty(handle, dir);
2057 ext4_journal_stop(handle);
2062 static int ext4_unlink(struct inode * dir, struct dentry *dentry)
2065 struct inode * inode;
2066 struct buffer_head * bh;
2067 struct ext4_dir_entry_2 * de;
2070 /* Initialize quotas before so that eventual writes go
2071 * in separate transaction */
2072 DQUOT_INIT(dentry->d_inode);
2073 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2075 return PTR_ERR(handle);
2077 if (IS_DIRSYNC(dir))
2081 bh = ext4_find_entry (dentry, &de);
2085 inode = dentry->d_inode;
2088 if (le32_to_cpu(de->inode) != inode->i_ino)
2091 if (!inode->i_nlink) {
2092 ext4_warning (inode->i_sb, "ext4_unlink",
2093 "Deleting nonexistent file (%lu), %d",
2094 inode->i_ino, inode->i_nlink);
2097 retval = ext4_delete_entry(handle, dir, de, bh);
2100 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2101 ext4_update_dx_flag(dir);
2102 ext4_mark_inode_dirty(handle, dir);
2104 if (!inode->i_nlink)
2105 ext4_orphan_add(handle, inode);
2106 inode->i_ctime = dir->i_ctime;
2107 ext4_mark_inode_dirty(handle, inode);
2111 ext4_journal_stop(handle);
2116 static int ext4_symlink (struct inode * dir,
2117 struct dentry *dentry, const char * symname)
2120 struct inode * inode;
2121 int l, err, retries = 0;
2123 l = strlen(symname)+1;
2124 if (l > dir->i_sb->s_blocksize)
2125 return -ENAMETOOLONG;
2128 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2129 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2130 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2132 return PTR_ERR(handle);
2134 if (IS_DIRSYNC(dir))
2137 inode = ext4_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2138 err = PTR_ERR(inode);
2142 if (l > sizeof (EXT4_I(inode)->i_data)) {
2143 inode->i_op = &ext4_symlink_inode_operations;
2144 ext4_set_aops(inode);
2146 * page_symlink() calls into ext4_prepare/commit_write.
2147 * We have a transaction open. All is sweetness. It also sets
2148 * i_size in generic_commit_write().
2150 err = __page_symlink(inode, symname, l,
2151 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2154 ext4_mark_inode_dirty(handle, inode);
2159 inode->i_op = &ext4_fast_symlink_inode_operations;
2160 memcpy((char*)&EXT4_I(inode)->i_data,symname,l);
2161 inode->i_size = l-1;
2163 EXT4_I(inode)->i_disksize = inode->i_size;
2164 err = ext4_add_nondir(handle, dentry, inode);
2166 ext4_journal_stop(handle);
2167 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2172 static int ext4_link (struct dentry * old_dentry,
2173 struct inode * dir, struct dentry *dentry)
2176 struct inode *inode = old_dentry->d_inode;
2177 int err, retries = 0;
2179 if (inode->i_nlink >= EXT4_LINK_MAX)
2182 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2183 * otherwise has the potential to corrupt the orphan inode list.
2185 if (inode->i_nlink == 0)
2189 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2190 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2192 return PTR_ERR(handle);
2194 if (IS_DIRSYNC(dir))
2197 inode->i_ctime = CURRENT_TIME_SEC;
2199 atomic_inc(&inode->i_count);
2201 err = ext4_add_nondir(handle, dentry, inode);
2202 ext4_journal_stop(handle);
2203 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2208 #define PARENT_INO(buffer) \
2209 ((struct ext4_dir_entry_2 *) ((char *) buffer + \
2210 le16_to_cpu(((struct ext4_dir_entry_2 *) buffer)->rec_len)))->inode
2213 * Anybody can rename anything with this: the permission checks are left to the
2214 * higher-level routines.
2216 static int ext4_rename (struct inode * old_dir, struct dentry *old_dentry,
2217 struct inode * new_dir,struct dentry *new_dentry)
2220 struct inode * old_inode, * new_inode;
2221 struct buffer_head * old_bh, * new_bh, * dir_bh;
2222 struct ext4_dir_entry_2 * old_de, * new_de;
2225 old_bh = new_bh = dir_bh = NULL;
2227 /* Initialize quotas before so that eventual writes go
2228 * in separate transaction */
2229 if (new_dentry->d_inode)
2230 DQUOT_INIT(new_dentry->d_inode);
2231 handle = ext4_journal_start(old_dir, 2 *
2232 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2233 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2235 return PTR_ERR(handle);
2237 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2240 old_bh = ext4_find_entry (old_dentry, &old_de);
2242 * Check for inode number is _not_ due to possible IO errors.
2243 * We might rmdir the source, keep it as pwd of some process
2244 * and merrily kill the link to whatever was created under the
2245 * same name. Goodbye sticky bit ;-<
2247 old_inode = old_dentry->d_inode;
2249 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2252 new_inode = new_dentry->d_inode;
2253 new_bh = ext4_find_entry (new_dentry, &new_de);
2260 if (S_ISDIR(old_inode->i_mode)) {
2262 retval = -ENOTEMPTY;
2263 if (!empty_dir (new_inode))
2267 dir_bh = ext4_bread (handle, old_inode, 0, 0, &retval);
2270 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2273 if (!new_inode && new_dir!=old_dir &&
2274 new_dir->i_nlink >= EXT4_LINK_MAX)
2278 retval = ext4_add_entry (handle, new_dentry, old_inode);
2282 BUFFER_TRACE(new_bh, "get write access");
2283 ext4_journal_get_write_access(handle, new_bh);
2284 new_de->inode = cpu_to_le32(old_inode->i_ino);
2285 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2286 EXT4_FEATURE_INCOMPAT_FILETYPE))
2287 new_de->file_type = old_de->file_type;
2288 new_dir->i_version++;
2289 BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata");
2290 ext4_journal_dirty_metadata(handle, new_bh);
2296 * Like most other Unix systems, set the ctime for inodes on a
2299 old_inode->i_ctime = CURRENT_TIME_SEC;
2300 ext4_mark_inode_dirty(handle, old_inode);
2305 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2306 old_de->name_len != old_dentry->d_name.len ||
2307 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2308 (retval = ext4_delete_entry(handle, old_dir,
2309 old_de, old_bh)) == -ENOENT) {
2310 /* old_de could have moved from under us during htree split, so
2311 * make sure that we are deleting the right entry. We might
2312 * also be pointing to a stale entry in the unused part of
2313 * old_bh so just checking inum and the name isn't enough. */
2314 struct buffer_head *old_bh2;
2315 struct ext4_dir_entry_2 *old_de2;
2317 old_bh2 = ext4_find_entry(old_dentry, &old_de2);
2319 retval = ext4_delete_entry(handle, old_dir,
2325 ext4_warning(old_dir->i_sb, "ext4_rename",
2326 "Deleting old file (%lu), %d, error=%d",
2327 old_dir->i_ino, old_dir->i_nlink, retval);
2331 drop_nlink(new_inode);
2332 new_inode->i_ctime = CURRENT_TIME_SEC;
2334 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2335 ext4_update_dx_flag(old_dir);
2337 BUFFER_TRACE(dir_bh, "get_write_access");
2338 ext4_journal_get_write_access(handle, dir_bh);
2339 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2340 BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata");
2341 ext4_journal_dirty_metadata(handle, dir_bh);
2342 drop_nlink(old_dir);
2344 drop_nlink(new_inode);
2347 ext4_update_dx_flag(new_dir);
2348 ext4_mark_inode_dirty(handle, new_dir);
2351 ext4_mark_inode_dirty(handle, old_dir);
2353 ext4_mark_inode_dirty(handle, new_inode);
2354 if (!new_inode->i_nlink)
2355 ext4_orphan_add(handle, new_inode);
2363 ext4_journal_stop(handle);
2368 * directories can handle most operations...
2370 const struct inode_operations ext4_dir_inode_operations = {
2371 .create = ext4_create,
2372 .lookup = ext4_lookup,
2374 .unlink = ext4_unlink,
2375 .symlink = ext4_symlink,
2376 .mkdir = ext4_mkdir,
2377 .rmdir = ext4_rmdir,
2378 .mknod = ext4_mknod,
2379 .rename = ext4_rename,
2380 .setattr = ext4_setattr,
2381 #ifdef CONFIG_EXT4DEV_FS_XATTR
2382 .setxattr = generic_setxattr,
2383 .getxattr = generic_getxattr,
2384 .listxattr = ext4_listxattr,
2385 .removexattr = generic_removexattr,
2387 .permission = ext4_permission,
2390 const struct inode_operations ext4_special_inode_operations = {
2391 .setattr = ext4_setattr,
2392 #ifdef CONFIG_EXT4DEV_FS_XATTR
2393 .setxattr = generic_setxattr,
2394 .getxattr = generic_getxattr,
2395 .listxattr = ext4_listxattr,
2396 .removexattr = generic_removexattr,
2398 .permission = ext4_permission,