2 * linux/fs/ext3/namei.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/namei.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
28 #include <linux/pagemap.h>
29 #include <linux/jbd.h>
30 #include <linux/time.h>
31 #include <linux/ext3_fs.h>
32 #include <linux/ext3_jbd.h>
33 #include <linux/fcntl.h>
34 #include <linux/stat.h>
35 #include <linux/string.h>
36 #include <linux/quotaops.h>
37 #include <linux/buffer_head.h>
38 #include <linux/smp_lock.h>
39 #include <linux/vserver/xid.h>
44 * define how far ahead to read directories while searching them.
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
49 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
51 static struct buffer_head *ext3_append(handle_t *handle,
55 struct buffer_head *bh;
57 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59 if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
60 inode->i_size += inode->i_sb->s_blocksize;
61 EXT3_I(inode)->i_disksize = inode->i_size;
62 ext3_journal_get_write_access(handle,bh);
68 #define assert(test) J_ASSERT(test)
72 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
76 #define dxtrace(command) command
78 #define dxtrace(command)
102 * dx_root_info is laid out so that if it should somehow get overlaid by a
103 * dirent the two low bits of the hash version will be zero. Therefore, the
104 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
109 struct fake_dirent dot;
111 struct fake_dirent dotdot;
115 __le32 reserved_zero;
117 u8 info_length; /* 8 */
122 struct dx_entry entries[0];
127 struct fake_dirent fake;
128 struct dx_entry entries[0];
134 struct buffer_head *bh;
135 struct dx_entry *entries;
145 #ifdef CONFIG_EXT3_INDEX
146 static inline unsigned dx_get_block (struct dx_entry *entry);
147 static void dx_set_block (struct dx_entry *entry, unsigned value);
148 static inline unsigned dx_get_hash (struct dx_entry *entry);
149 static void dx_set_hash (struct dx_entry *entry, unsigned value);
150 static unsigned dx_get_count (struct dx_entry *entries);
151 static unsigned dx_get_limit (struct dx_entry *entries);
152 static void dx_set_count (struct dx_entry *entries, unsigned value);
153 static void dx_set_limit (struct dx_entry *entries, unsigned value);
154 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
155 static unsigned dx_node_limit (struct inode *dir);
156 static struct dx_frame *dx_probe(struct dentry *dentry,
158 struct dx_hash_info *hinfo,
159 struct dx_frame *frame,
161 static void dx_release (struct dx_frame *frames);
162 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
163 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
164 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
165 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
166 struct dx_map_entry *offsets, int count);
167 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
168 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
169 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
170 struct dx_frame *frame,
171 struct dx_frame *frames,
173 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
174 struct ext3_dir_entry_2 **res_dir, int *err);
175 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
176 struct inode *inode);
179 * Future: use high four bits of block for coalesce-on-delete flags
180 * Mask them off for now.
183 static inline unsigned dx_get_block (struct dx_entry *entry)
185 return le32_to_cpu(entry->block) & 0x00ffffff;
188 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
190 entry->block = cpu_to_le32(value);
193 static inline unsigned dx_get_hash (struct dx_entry *entry)
195 return le32_to_cpu(entry->hash);
198 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
200 entry->hash = cpu_to_le32(value);
203 static inline unsigned dx_get_count (struct dx_entry *entries)
205 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
208 static inline unsigned dx_get_limit (struct dx_entry *entries)
210 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
213 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
215 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
218 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
220 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
223 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
225 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
226 EXT3_DIR_REC_LEN(2) - infosize;
227 return 0? 20: entry_space / sizeof(struct dx_entry);
230 static inline unsigned dx_node_limit (struct inode *dir)
232 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
233 return 0? 22: entry_space / sizeof(struct dx_entry);
240 static void dx_show_index (char * label, struct dx_entry *entries)
242 int i, n = dx_get_count (entries);
243 printk("%s index ", label);
244 for (i = 0; i < n; i++)
246 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
258 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
259 int size, int show_names)
261 unsigned names = 0, space = 0;
262 char *base = (char *) de;
263 struct dx_hash_info h = *hinfo;
266 while ((char *) de < base + size)
272 int len = de->name_len;
273 char *name = de->name;
274 while (len--) printk("%c", *name++);
275 ext3fs_dirhash(de->name, de->name_len, &h);
276 printk(":%x.%u ", h.hash,
277 ((char *) de - base));
279 space += EXT3_DIR_REC_LEN(de->name_len);
282 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
284 printk("(%i)\n", names);
285 return (struct stats) { names, space, 1 };
288 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
289 struct dx_entry *entries, int levels)
291 unsigned blocksize = dir->i_sb->s_blocksize;
292 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
294 struct buffer_head *bh;
296 printk("%i indexed blocks...\n", count);
297 for (i = 0; i < count; i++, entries++)
299 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
300 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
302 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
303 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
305 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
306 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
307 names += stats.names;
308 space += stats.space;
309 bcount += stats.bcount;
313 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
314 names, space/bcount,(space/bcount)*100/blocksize);
315 return (struct stats) { names, space, bcount};
317 #endif /* DX_DEBUG */
320 * Probe for a directory leaf block to search.
322 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
323 * error in the directory index, and the caller should fall back to
324 * searching the directory normally. The callers of dx_probe **MUST**
325 * check for this error code, and make sure it never gets reflected
328 static struct dx_frame *
329 dx_probe(struct dentry *dentry, struct inode *dir,
330 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
332 unsigned count, indirect;
333 struct dx_entry *at, *entries, *p, *q, *m;
334 struct dx_root *root;
335 struct buffer_head *bh;
336 struct dx_frame *frame = frame_in;
341 dir = dentry->d_parent->d_inode;
342 if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
344 root = (struct dx_root *) bh->b_data;
345 if (root->info.hash_version != DX_HASH_TEA &&
346 root->info.hash_version != DX_HASH_HALF_MD4 &&
347 root->info.hash_version != DX_HASH_LEGACY) {
348 ext3_warning(dir->i_sb, __FUNCTION__,
349 "Unrecognised inode hash code %d",
350 root->info.hash_version);
352 *err = ERR_BAD_DX_DIR;
355 hinfo->hash_version = root->info.hash_version;
356 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
358 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
361 if (root->info.unused_flags & 1) {
362 ext3_warning(dir->i_sb, __FUNCTION__,
363 "Unimplemented inode hash flags: %#06x",
364 root->info.unused_flags);
366 *err = ERR_BAD_DX_DIR;
370 if ((indirect = root->info.indirect_levels) > 1) {
371 ext3_warning(dir->i_sb, __FUNCTION__,
372 "Unimplemented inode hash depth: %#06x",
373 root->info.indirect_levels);
375 *err = ERR_BAD_DX_DIR;
379 entries = (struct dx_entry *) (((char *)&root->info) +
380 root->info.info_length);
381 assert(dx_get_limit(entries) == dx_root_limit(dir,
382 root->info.info_length));
383 dxtrace (printk("Look up %x", hash));
386 count = dx_get_count(entries);
387 assert (count && count <= dx_get_limit(entries));
389 q = entries + count - 1;
393 dxtrace(printk("."));
394 if (dx_get_hash(m) > hash)
400 if (0) // linear search cross check
402 unsigned n = count - 1;
406 dxtrace(printk(","));
407 if (dx_get_hash(++at) > hash)
413 assert (at == p - 1);
417 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
419 frame->entries = entries;
421 if (!indirect--) return frame;
422 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
424 at = entries = ((struct dx_node *) bh->b_data)->entries;
425 assert (dx_get_limit(entries) == dx_node_limit (dir));
429 while (frame >= frame_in) {
437 static void dx_release (struct dx_frame *frames)
439 if (frames[0].bh == NULL)
442 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
443 brelse(frames[1].bh);
444 brelse(frames[0].bh);
448 * This function increments the frame pointer to search the next leaf
449 * block, and reads in the necessary intervening nodes if the search
450 * should be necessary. Whether or not the search is necessary is
451 * controlled by the hash parameter. If the hash value is even, then
452 * the search is only continued if the next block starts with that
453 * hash value. This is used if we are searching for a specific file.
455 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
457 * This function returns 1 if the caller should continue to search,
458 * or 0 if it should not. If there is an error reading one of the
459 * index blocks, it will a negative error code.
461 * If start_hash is non-null, it will be filled in with the starting
462 * hash of the next page.
464 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
465 struct dx_frame *frame,
466 struct dx_frame *frames,
470 struct buffer_head *bh;
471 int err, num_frames = 0;
476 * Find the next leaf page by incrementing the frame pointer.
477 * If we run out of entries in the interior node, loop around and
478 * increment pointer in the parent node. When we break out of
479 * this loop, num_frames indicates the number of interior
480 * nodes need to be read.
483 if (++(p->at) < p->entries + dx_get_count(p->entries))
492 * If the hash is 1, then continue only if the next page has a
493 * continuation hash of any value. This is used for readdir
494 * handling. Otherwise, check to see if the hash matches the
495 * desired contiuation hash. If it doesn't, return since
496 * there's no point to read in the successive index pages.
498 bhash = dx_get_hash(p->at);
501 if ((hash & 1) == 0) {
502 if ((bhash & ~1) != hash)
506 * If the hash is HASH_NB_ALWAYS, we always go to the next
507 * block so no check is necessary
509 while (num_frames--) {
510 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
512 return err; /* Failure */
516 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
523 * p is at least 6 bytes before the end of page
525 static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
527 return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
531 * This function fills a red-black tree with information from a
532 * directory block. It returns the number directory entries loaded
533 * into the tree. If there is an error it is returned in err.
535 static int htree_dirblock_to_tree(struct file *dir_file,
536 struct inode *dir, int block,
537 struct dx_hash_info *hinfo,
538 __u32 start_hash, __u32 start_minor_hash)
540 struct buffer_head *bh;
541 struct ext3_dir_entry_2 *de, *top;
544 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
545 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
548 de = (struct ext3_dir_entry_2 *) bh->b_data;
549 top = (struct ext3_dir_entry_2 *) ((char *) de +
550 dir->i_sb->s_blocksize -
551 EXT3_DIR_REC_LEN(0));
552 for (; de < top; de = ext3_next_entry(de)) {
553 ext3fs_dirhash(de->name, de->name_len, hinfo);
554 if ((hinfo->hash < start_hash) ||
555 ((hinfo->hash == start_hash) &&
556 (hinfo->minor_hash < start_minor_hash)))
560 if ((err = ext3_htree_store_dirent(dir_file,
561 hinfo->hash, hinfo->minor_hash, de)) != 0) {
573 * This function fills a red-black tree with information from a
574 * directory. We start scanning the directory in hash order, starting
575 * at start_hash and start_minor_hash.
577 * This function returns the number of entries inserted into the tree,
578 * or a negative error code.
580 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
581 __u32 start_minor_hash, __u32 *next_hash)
583 struct dx_hash_info hinfo;
584 struct ext3_dir_entry_2 *de;
585 struct dx_frame frames[2], *frame;
592 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
594 dir = dir_file->f_dentry->d_inode;
595 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
596 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
597 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
598 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
599 start_hash, start_minor_hash);
603 hinfo.hash = start_hash;
604 hinfo.minor_hash = 0;
605 frame = dx_probe(NULL, dir_file->f_dentry->d_inode, &hinfo, frames, &err);
609 /* Add '.' and '..' from the htree header */
610 if (!start_hash && !start_minor_hash) {
611 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
612 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
616 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
617 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
618 de = ext3_next_entry(de);
619 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
625 block = dx_get_block(frame->at);
626 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
627 start_hash, start_minor_hash);
634 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
635 frame, frames, &hashval);
636 *next_hash = hashval;
642 * Stop if: (a) there are no more entries, or
643 * (b) we have inserted at least one entry and the
644 * next hash value is not a continuation
647 (count && ((hashval & 1) == 0)))
651 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
661 * Directory block splitting, compacting
664 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
665 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
668 char *base = (char *) de;
669 struct dx_hash_info h = *hinfo;
671 while ((char *) de < base + size)
673 if (de->name_len && de->inode) {
674 ext3fs_dirhash(de->name, de->name_len, &h);
676 map_tail->hash = h.hash;
677 map_tail->offs = (u32) ((char *) de - base);
681 /* XXX: do we need to check rec_len == 0 case? -Chris */
682 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
687 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
689 struct dx_map_entry *p, *q, *top = map + count - 1;
691 /* Combsort until bubble sort doesn't suck */
695 if (count - 9 < 2) /* 9, 10 -> 11 */
697 for (p = top, q = p - count; q >= map; p--, q--)
698 if (p->hash < q->hash)
701 /* Garden variety bubble sort */
707 if (q[1].hash >= q[0].hash)
715 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
717 struct dx_entry *entries = frame->entries;
718 struct dx_entry *old = frame->at, *new = old + 1;
719 int count = dx_get_count(entries);
721 assert(count < dx_get_limit(entries));
722 assert(old < entries + count);
723 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
724 dx_set_hash(new, hash);
725 dx_set_block(new, block);
726 dx_set_count(entries, count + 1);
731 static void ext3_update_dx_flag(struct inode *inode)
733 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
734 EXT3_FEATURE_COMPAT_DIR_INDEX))
735 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
739 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
741 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
742 * `de != NULL' is guaranteed by caller.
744 static inline int ext3_match (int len, const char * const name,
745 struct ext3_dir_entry_2 * de)
747 if (len != de->name_len)
751 return !memcmp(name, de->name, len);
755 * Returns 0 if not found, -1 on failure, and 1 on success
757 static inline int search_dirblock(struct buffer_head * bh,
759 struct dentry *dentry,
760 unsigned long offset,
761 struct ext3_dir_entry_2 ** res_dir)
763 struct ext3_dir_entry_2 * de;
766 const char *name = dentry->d_name.name;
767 int namelen = dentry->d_name.len;
769 de = (struct ext3_dir_entry_2 *) bh->b_data;
770 dlimit = bh->b_data + dir->i_sb->s_blocksize;
771 while ((char *) de < dlimit) {
772 /* this code is executed quadratically often */
773 /* do minimal checking `by hand' */
775 if ((char *) de + namelen <= dlimit &&
776 ext3_match (namelen, name, de)) {
777 /* found a match - just to be sure, do a full check */
778 if (!ext3_check_dir_entry("ext3_find_entry",
779 dir, de, bh, offset))
784 /* prevent looping on a bad block */
785 de_len = le16_to_cpu(de->rec_len);
789 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
798 * finds an entry in the specified directory with the wanted name. It
799 * returns the cache buffer in which the entry was found, and the entry
800 * itself (as a parameter - res_dir). It does NOT read the inode of the
801 * entry - you'll have to do that yourself if you want to.
803 * The returned buffer_head has ->b_count elevated. The caller is expected
804 * to brelse() it when appropriate.
806 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
807 struct ext3_dir_entry_2 ** res_dir)
809 struct super_block * sb;
810 struct buffer_head * bh_use[NAMEI_RA_SIZE];
811 struct buffer_head * bh, *ret = NULL;
812 unsigned long start, block, b;
813 int ra_max = 0; /* Number of bh's in the readahead
815 int ra_ptr = 0; /* Current index into readahead
819 struct inode *dir = dentry->d_parent->d_inode;
826 blocksize = sb->s_blocksize;
827 namelen = dentry->d_name.len;
828 name = dentry->d_name.name;
829 if (namelen > EXT3_NAME_LEN)
831 #ifdef CONFIG_EXT3_INDEX
833 bh = ext3_dx_find_entry(dentry, res_dir, &err);
835 * On success, or if the error was file not found,
836 * return. Otherwise, fall back to doing a search the
839 if (bh || (err != ERR_BAD_DX_DIR))
841 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
844 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
845 start = EXT3_I(dir)->i_dir_start_lookup;
846 if (start >= nblocks)
852 * We deal with the read-ahead logic here.
854 if (ra_ptr >= ra_max) {
855 /* Refill the readahead buffer */
858 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
860 * Terminate if we reach the end of the
861 * directory and must wrap, or if our
862 * search has finished at this block.
864 if (b >= nblocks || (num && block == start)) {
865 bh_use[ra_max] = NULL;
869 bh = ext3_getblk(NULL, dir, b++, 0, &err);
872 ll_rw_block(READ, 1, &bh);
875 if ((bh = bh_use[ra_ptr++]) == NULL)
878 if (!buffer_uptodate(bh)) {
879 /* read error, skip block & hope for the best */
880 ext3_error(sb, __FUNCTION__, "reading directory #%lu "
881 "offset %lu", dir->i_ino, block);
885 i = search_dirblock(bh, dir, dentry,
886 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
888 EXT3_I(dir)->i_dir_start_lookup = block;
890 goto cleanup_and_exit;
894 goto cleanup_and_exit;
897 if (++block >= nblocks)
899 } while (block != start);
902 * If the directory has grown while we were searching, then
903 * search the last part of the directory before giving up.
906 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
907 if (block < nblocks) {
913 /* Clean up the read-ahead blocks */
914 for (; ra_ptr < ra_max; ra_ptr++)
915 brelse (bh_use[ra_ptr]);
919 #ifdef CONFIG_EXT3_INDEX
920 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
921 struct ext3_dir_entry_2 **res_dir, int *err)
923 struct super_block * sb;
924 struct dx_hash_info hinfo;
926 struct dx_frame frames[2], *frame;
927 struct ext3_dir_entry_2 *de, *top;
928 struct buffer_head *bh;
931 int namelen = dentry->d_name.len;
932 const u8 *name = dentry->d_name.name;
933 struct inode *dir = dentry->d_parent->d_inode;
936 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
940 block = dx_get_block(frame->at);
941 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
943 de = (struct ext3_dir_entry_2 *) bh->b_data;
944 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
945 EXT3_DIR_REC_LEN(0));
946 for (; de < top; de = ext3_next_entry(de))
947 if (ext3_match (namelen, name, de)) {
948 if (!ext3_check_dir_entry("ext3_find_entry",
950 (block<<EXT3_BLOCK_SIZE_BITS(sb))
951 +((char *)de - bh->b_data))) {
960 /* Check to see if we should continue to search */
961 retval = ext3_htree_next_block(dir, hash, frame,
964 ext3_warning(sb, __FUNCTION__,
965 "error reading index page in directory #%lu",
970 } while (retval == 1);
974 dxtrace(printk("%s not found\n", name));
980 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
982 struct inode * inode;
983 struct ext3_dir_entry_2 * de;
984 struct buffer_head * bh;
986 if (dentry->d_name.len > EXT3_NAME_LEN)
987 return ERR_PTR(-ENAMETOOLONG);
989 bh = ext3_find_entry(dentry, &de);
992 unsigned long ino = le32_to_cpu(de->inode);
994 inode = iget(dir->i_sb, ino);
997 return ERR_PTR(-EACCES);
998 vx_propagate_xid(nd, inode);
1001 return d_splice_alias(inode, dentry);
1002 d_add(dentry, inode);
1007 struct dentry *ext3_get_parent(struct dentry *child)
1010 struct dentry *parent;
1011 struct inode *inode;
1012 struct dentry dotdot;
1013 struct ext3_dir_entry_2 * de;
1014 struct buffer_head *bh;
1016 dotdot.d_name.name = "..";
1017 dotdot.d_name.len = 2;
1018 dotdot.d_parent = child; /* confusing, isn't it! */
1020 bh = ext3_find_entry(&dotdot, &de);
1023 return ERR_PTR(-ENOENT);
1024 ino = le32_to_cpu(de->inode);
1026 inode = iget(child->d_inode->i_sb, ino);
1029 return ERR_PTR(-EACCES);
1031 parent = d_alloc_anon(inode);
1034 parent = ERR_PTR(-ENOMEM);
1040 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1041 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1042 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1043 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1044 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1045 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1046 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1047 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1050 static inline void ext3_set_de_type(struct super_block *sb,
1051 struct ext3_dir_entry_2 *de,
1053 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1054 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1057 #ifdef CONFIG_EXT3_INDEX
1058 static struct ext3_dir_entry_2 *
1059 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1061 unsigned rec_len = 0;
1064 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1065 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1066 memcpy (to, de, rec_len);
1067 ((struct ext3_dir_entry_2 *) to)->rec_len =
1068 cpu_to_le16(rec_len);
1073 return (struct ext3_dir_entry_2 *) (to - rec_len);
1076 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1078 struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1079 unsigned rec_len = 0;
1082 while ((char*)de < base + size) {
1083 next = (struct ext3_dir_entry_2 *) ((char *) de +
1084 le16_to_cpu(de->rec_len));
1085 if (de->inode && de->name_len) {
1086 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1088 memmove(to, de, rec_len);
1089 to->rec_len = cpu_to_le16(rec_len);
1091 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1098 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1099 struct buffer_head **bh,struct dx_frame *frame,
1100 struct dx_hash_info *hinfo, int *error)
1102 unsigned blocksize = dir->i_sb->s_blocksize;
1103 unsigned count, continued;
1104 struct buffer_head *bh2;
1107 struct dx_map_entry *map;
1108 char *data1 = (*bh)->b_data, *data2;
1110 struct ext3_dir_entry_2 *de = NULL, *de2;
1113 bh2 = ext3_append (handle, dir, &newblock, error);
1120 BUFFER_TRACE(*bh, "get_write_access");
1121 err = ext3_journal_get_write_access(handle, *bh);
1127 ext3_std_error(dir->i_sb, err);
1130 BUFFER_TRACE(frame->bh, "get_write_access");
1131 err = ext3_journal_get_write_access(handle, frame->bh);
1135 data2 = bh2->b_data;
1137 /* create map in the end of data2 block */
1138 map = (struct dx_map_entry *) (data2 + blocksize);
1139 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1140 blocksize, hinfo, map);
1142 split = count/2; // need to adjust to actual middle
1143 dx_sort_map (map, count);
1144 hash2 = map[split].hash;
1145 continued = hash2 == map[split - 1].hash;
1146 dxtrace(printk("Split block %i at %x, %i/%i\n",
1147 dx_get_block(frame->at), hash2, split, count-split));
1149 /* Fancy dance to stay within two buffers */
1150 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1151 de = dx_pack_dirents(data1,blocksize);
1152 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1153 de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1154 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1155 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1157 /* Which block gets the new entry? */
1158 if (hinfo->hash >= hash2)
1163 dx_insert_block (frame, hash2 + continued, newblock);
1164 err = ext3_journal_dirty_metadata (handle, bh2);
1167 err = ext3_journal_dirty_metadata (handle, frame->bh);
1171 dxtrace(dx_show_index ("frame", frame->entries));
1179 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1180 * it points to a directory entry which is guaranteed to be large
1181 * enough for new directory entry. If de is NULL, then
1182 * add_dirent_to_buf will attempt search the directory block for
1183 * space. It will return -ENOSPC if no space is available, and -EIO
1184 * and -EEXIST if directory entry already exists.
1186 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1187 * all other cases bh is released.
1189 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1190 struct inode *inode, struct ext3_dir_entry_2 *de,
1191 struct buffer_head * bh)
1193 struct inode *dir = dentry->d_parent->d_inode;
1194 const char *name = dentry->d_name.name;
1195 int namelen = dentry->d_name.len;
1196 unsigned long offset = 0;
1197 unsigned short reclen;
1198 int nlen, rlen, err;
1201 reclen = EXT3_DIR_REC_LEN(namelen);
1203 de = (struct ext3_dir_entry_2 *)bh->b_data;
1204 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1205 while ((char *) de <= top) {
1206 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1211 if (ext3_match (namelen, name, de)) {
1215 nlen = EXT3_DIR_REC_LEN(de->name_len);
1216 rlen = le16_to_cpu(de->rec_len);
1217 if ((de->inode? rlen - nlen: rlen) >= reclen)
1219 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1222 if ((char *) de > top)
1225 BUFFER_TRACE(bh, "get_write_access");
1226 err = ext3_journal_get_write_access(handle, bh);
1228 ext3_std_error(dir->i_sb, err);
1233 /* By now the buffer is marked for journaling */
1234 nlen = EXT3_DIR_REC_LEN(de->name_len);
1235 rlen = le16_to_cpu(de->rec_len);
1237 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1238 de1->rec_len = cpu_to_le16(rlen - nlen);
1239 de->rec_len = cpu_to_le16(nlen);
1242 de->file_type = EXT3_FT_UNKNOWN;
1244 de->inode = cpu_to_le32(inode->i_ino);
1245 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1248 de->name_len = namelen;
1249 memcpy (de->name, name, namelen);
1251 * XXX shouldn't update any times until successful
1252 * completion of syscall, but too many callers depend
1255 * XXX similarly, too many callers depend on
1256 * ext3_new_inode() setting the times, but error
1257 * recovery deletes the inode, so the worst that can
1258 * happen is that the times are slightly out of date
1259 * and/or different from the directory change time.
1261 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1262 ext3_update_dx_flag(dir);
1264 ext3_mark_inode_dirty(handle, dir);
1265 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1266 err = ext3_journal_dirty_metadata(handle, bh);
1268 ext3_std_error(dir->i_sb, err);
1273 #ifdef CONFIG_EXT3_INDEX
1275 * This converts a one block unindexed directory to a 3 block indexed
1276 * directory, and adds the dentry to the indexed directory.
1278 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1279 struct inode *inode, struct buffer_head *bh)
1281 struct inode *dir = dentry->d_parent->d_inode;
1282 const char *name = dentry->d_name.name;
1283 int namelen = dentry->d_name.len;
1284 struct buffer_head *bh2;
1285 struct dx_root *root;
1286 struct dx_frame frames[2], *frame;
1287 struct dx_entry *entries;
1288 struct ext3_dir_entry_2 *de, *de2;
1293 struct dx_hash_info hinfo;
1295 struct fake_dirent *fde;
1297 blocksize = dir->i_sb->s_blocksize;
1298 dxtrace(printk("Creating index\n"));
1299 retval = ext3_journal_get_write_access(handle, bh);
1301 ext3_std_error(dir->i_sb, retval);
1305 root = (struct dx_root *) bh->b_data;
1307 bh2 = ext3_append (handle, dir, &block, &retval);
1312 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1313 data1 = bh2->b_data;
1315 /* The 0th block becomes the root, move the dirents out */
1316 fde = &root->dotdot;
1317 de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1318 len = ((char *) root) + blocksize - (char *) de;
1319 memcpy (data1, de, len);
1320 de = (struct ext3_dir_entry_2 *) data1;
1322 while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1324 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1325 /* Initialize the root; the dot dirents already exist */
1326 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1327 de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1328 memset (&root->info, 0, sizeof(root->info));
1329 root->info.info_length = sizeof(root->info);
1330 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1331 entries = root->entries;
1332 dx_set_block (entries, 1);
1333 dx_set_count (entries, 1);
1334 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1336 /* Initialize as for dx_probe */
1337 hinfo.hash_version = root->info.hash_version;
1338 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1339 ext3fs_dirhash(name, namelen, &hinfo);
1341 frame->entries = entries;
1342 frame->at = entries;
1345 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1346 dx_release (frames);
1350 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1357 * adds a file entry to the specified directory, using the same
1358 * semantics as ext3_find_entry(). It returns NULL if it failed.
1360 * NOTE!! The inode part of 'de' is left at 0 - which means you
1361 * may not sleep between calling this and putting something into
1362 * the entry, as someone else might have used it while you slept.
1364 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1365 struct inode *inode)
1367 struct inode *dir = dentry->d_parent->d_inode;
1368 unsigned long offset;
1369 struct buffer_head * bh;
1370 struct ext3_dir_entry_2 *de;
1371 struct super_block * sb;
1373 #ifdef CONFIG_EXT3_INDEX
1377 unsigned nlen, rlen;
1381 blocksize = sb->s_blocksize;
1382 if (!dentry->d_name.len)
1384 #ifdef CONFIG_EXT3_INDEX
1386 retval = ext3_dx_add_entry(handle, dentry, inode);
1387 if (!retval || (retval != ERR_BAD_DX_DIR))
1389 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1391 ext3_mark_inode_dirty(handle, dir);
1394 blocks = dir->i_size >> sb->s_blocksize_bits;
1395 for (block = 0, offset = 0; block < blocks; block++) {
1396 bh = ext3_bread(handle, dir, block, 0, &retval);
1399 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1400 if (retval != -ENOSPC)
1403 #ifdef CONFIG_EXT3_INDEX
1404 if (blocks == 1 && !dx_fallback &&
1405 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1406 return make_indexed_dir(handle, dentry, inode, bh);
1410 bh = ext3_append(handle, dir, &block, &retval);
1413 de = (struct ext3_dir_entry_2 *) bh->b_data;
1415 de->rec_len = cpu_to_le16(rlen = blocksize);
1417 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1420 #ifdef CONFIG_EXT3_INDEX
1422 * Returns 0 for success, or a negative error value
1424 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1425 struct inode *inode)
1427 struct dx_frame frames[2], *frame;
1428 struct dx_entry *entries, *at;
1429 struct dx_hash_info hinfo;
1430 struct buffer_head * bh;
1431 struct inode *dir = dentry->d_parent->d_inode;
1432 struct super_block * sb = dir->i_sb;
1433 struct ext3_dir_entry_2 *de;
1436 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1439 entries = frame->entries;
1442 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1445 BUFFER_TRACE(bh, "get_write_access");
1446 err = ext3_journal_get_write_access(handle, bh);
1450 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1451 if (err != -ENOSPC) {
1456 /* Block full, should compress but for now just split */
1457 dxtrace(printk("using %u of %u node entries\n",
1458 dx_get_count(entries), dx_get_limit(entries)));
1459 /* Need to split index? */
1460 if (dx_get_count(entries) == dx_get_limit(entries)) {
1462 unsigned icount = dx_get_count(entries);
1463 int levels = frame - frames;
1464 struct dx_entry *entries2;
1465 struct dx_node *node2;
1466 struct buffer_head *bh2;
1468 if (levels && (dx_get_count(frames->entries) ==
1469 dx_get_limit(frames->entries))) {
1470 ext3_warning(sb, __FUNCTION__,
1471 "Directory index full!\n");
1475 bh2 = ext3_append (handle, dir, &newblock, &err);
1478 node2 = (struct dx_node *)(bh2->b_data);
1479 entries2 = node2->entries;
1480 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1481 node2->fake.inode = 0;
1482 BUFFER_TRACE(frame->bh, "get_write_access");
1483 err = ext3_journal_get_write_access(handle, frame->bh);
1487 unsigned icount1 = icount/2, icount2 = icount - icount1;
1488 unsigned hash2 = dx_get_hash(entries + icount1);
1489 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1491 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1492 err = ext3_journal_get_write_access(handle,
1497 memcpy ((char *) entries2, (char *) (entries + icount1),
1498 icount2 * sizeof(struct dx_entry));
1499 dx_set_count (entries, icount1);
1500 dx_set_count (entries2, icount2);
1501 dx_set_limit (entries2, dx_node_limit(dir));
1503 /* Which index block gets the new entry? */
1504 if (at - entries >= icount1) {
1505 frame->at = at = at - entries - icount1 + entries2;
1506 frame->entries = entries = entries2;
1507 swap(frame->bh, bh2);
1509 dx_insert_block (frames + 0, hash2, newblock);
1510 dxtrace(dx_show_index ("node", frames[1].entries));
1511 dxtrace(dx_show_index ("node",
1512 ((struct dx_node *) bh2->b_data)->entries));
1513 err = ext3_journal_dirty_metadata(handle, bh2);
1518 dxtrace(printk("Creating second level index...\n"));
1519 memcpy((char *) entries2, (char *) entries,
1520 icount * sizeof(struct dx_entry));
1521 dx_set_limit(entries2, dx_node_limit(dir));
1524 dx_set_count(entries, 1);
1525 dx_set_block(entries + 0, newblock);
1526 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1528 /* Add new access path frame */
1530 frame->at = at = at - entries + entries2;
1531 frame->entries = entries = entries2;
1533 err = ext3_journal_get_write_access(handle,
1538 ext3_journal_dirty_metadata(handle, frames[0].bh);
1540 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1543 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1548 ext3_std_error(dir->i_sb, err);
1558 * ext3_delete_entry deletes a directory entry by merging it with the
1561 static int ext3_delete_entry (handle_t *handle,
1563 struct ext3_dir_entry_2 * de_del,
1564 struct buffer_head * bh)
1566 struct ext3_dir_entry_2 * de, * pde;
1571 de = (struct ext3_dir_entry_2 *) bh->b_data;
1572 while (i < bh->b_size) {
1573 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1576 BUFFER_TRACE(bh, "get_write_access");
1577 ext3_journal_get_write_access(handle, bh);
1580 cpu_to_le16(le16_to_cpu(pde->rec_len) +
1581 le16_to_cpu(de->rec_len));
1585 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1586 ext3_journal_dirty_metadata(handle, bh);
1589 i += le16_to_cpu(de->rec_len);
1591 de = (struct ext3_dir_entry_2 *)
1592 ((char *) de + le16_to_cpu(de->rec_len));
1598 * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1599 * do not perform it in these functions. We perform it at the call site,
1602 static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1607 static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1612 static int ext3_add_nondir(handle_t *handle,
1613 struct dentry *dentry, struct inode *inode)
1615 int err = ext3_add_entry(handle, dentry, inode);
1617 ext3_mark_inode_dirty(handle, inode);
1618 d_instantiate(dentry, inode);
1621 ext3_dec_count(handle, inode);
1627 * By the time this is called, we already have created
1628 * the directory cache entry for the new file, but it
1629 * is so far negative - it has no inode.
1631 * If the create succeeds, we fill in the inode information
1632 * with d_instantiate().
1634 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1635 struct nameidata *nd)
1638 struct inode * inode;
1639 int err, retries = 0;
1642 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1643 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1644 2*EXT3_QUOTA_INIT_BLOCKS);
1646 return PTR_ERR(handle);
1648 if (IS_DIRSYNC(dir))
1651 inode = ext3_new_inode (handle, dir, mode);
1652 err = PTR_ERR(inode);
1653 if (!IS_ERR(inode)) {
1654 inode->i_op = &ext3_file_inode_operations;
1655 inode->i_fop = &ext3_file_operations;
1656 ext3_set_aops(inode);
1657 err = ext3_add_nondir(handle, dentry, inode);
1659 ext3_journal_stop(handle);
1660 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1665 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1666 int mode, dev_t rdev)
1669 struct inode *inode;
1670 int err, retries = 0;
1672 if (!new_valid_dev(rdev))
1676 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1677 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1678 2*EXT3_QUOTA_INIT_BLOCKS);
1680 return PTR_ERR(handle);
1682 if (IS_DIRSYNC(dir))
1685 inode = ext3_new_inode (handle, dir, mode);
1686 err = PTR_ERR(inode);
1687 if (!IS_ERR(inode)) {
1688 init_special_inode(inode, inode->i_mode, rdev);
1689 #ifdef CONFIG_EXT3_FS_XATTR
1690 inode->i_op = &ext3_special_inode_operations;
1692 err = ext3_add_nondir(handle, dentry, inode);
1694 ext3_journal_stop(handle);
1695 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1700 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1703 struct inode * inode;
1704 struct buffer_head * dir_block;
1705 struct ext3_dir_entry_2 * de;
1706 int err, retries = 0;
1708 if (dir->i_nlink >= EXT3_LINK_MAX)
1712 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1713 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1714 2*EXT3_QUOTA_INIT_BLOCKS);
1716 return PTR_ERR(handle);
1718 if (IS_DIRSYNC(dir))
1721 inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1722 err = PTR_ERR(inode);
1726 inode->i_op = &ext3_dir_inode_operations;
1727 inode->i_fop = &ext3_dir_operations;
1728 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1729 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1731 inode->i_nlink--; /* is this nlink == 0? */
1732 ext3_mark_inode_dirty(handle, inode);
1736 BUFFER_TRACE(dir_block, "get_write_access");
1737 ext3_journal_get_write_access(handle, dir_block);
1738 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1739 de->inode = cpu_to_le32(inode->i_ino);
1741 de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1742 strcpy (de->name, ".");
1743 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1744 de = (struct ext3_dir_entry_2 *)
1745 ((char *) de + le16_to_cpu(de->rec_len));
1746 de->inode = cpu_to_le32(dir->i_ino);
1747 de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1749 strcpy (de->name, "..");
1750 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1752 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1753 ext3_journal_dirty_metadata(handle, dir_block);
1755 ext3_mark_inode_dirty(handle, inode);
1756 err = ext3_add_entry (handle, dentry, inode);
1759 ext3_mark_inode_dirty(handle, inode);
1764 ext3_update_dx_flag(dir);
1765 ext3_mark_inode_dirty(handle, dir);
1766 d_instantiate(dentry, inode);
1768 ext3_journal_stop(handle);
1769 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1775 * routine to check that the specified directory is empty (for rmdir)
1777 static int empty_dir (struct inode * inode)
1779 unsigned long offset;
1780 struct buffer_head * bh;
1781 struct ext3_dir_entry_2 * de, * de1;
1782 struct super_block * sb;
1786 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1787 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1789 ext3_error(inode->i_sb, __FUNCTION__,
1790 "error %d reading directory #%lu offset 0",
1793 ext3_warning(inode->i_sb, __FUNCTION__,
1794 "bad directory (dir #%lu) - no data block",
1798 de = (struct ext3_dir_entry_2 *) bh->b_data;
1799 de1 = (struct ext3_dir_entry_2 *)
1800 ((char *) de + le16_to_cpu(de->rec_len));
1801 if (le32_to_cpu(de->inode) != inode->i_ino ||
1802 !le32_to_cpu(de1->inode) ||
1803 strcmp (".", de->name) ||
1804 strcmp ("..", de1->name)) {
1805 ext3_warning (inode->i_sb, "empty_dir",
1806 "bad directory (dir #%lu) - no `.' or `..'",
1811 offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1812 de = (struct ext3_dir_entry_2 *)
1813 ((char *) de1 + le16_to_cpu(de1->rec_len));
1814 while (offset < inode->i_size ) {
1816 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1819 bh = ext3_bread (NULL, inode,
1820 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1823 ext3_error(sb, __FUNCTION__,
1824 "error %d reading directory"
1826 err, inode->i_ino, offset);
1827 offset += sb->s_blocksize;
1830 de = (struct ext3_dir_entry_2 *) bh->b_data;
1832 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1833 de = (struct ext3_dir_entry_2 *)(bh->b_data +
1835 offset = (offset | (sb->s_blocksize - 1)) + 1;
1838 if (le32_to_cpu(de->inode)) {
1842 offset += le16_to_cpu(de->rec_len);
1843 de = (struct ext3_dir_entry_2 *)
1844 ((char *) de + le16_to_cpu(de->rec_len));
1850 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1851 * such inodes, starting at the superblock, in case we crash before the
1852 * file is closed/deleted, or in case the inode truncate spans multiple
1853 * transactions and the last transaction is not recovered after a crash.
1855 * At filesystem recovery time, we walk this list deleting unlinked
1856 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1858 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1860 struct super_block *sb = inode->i_sb;
1861 struct ext3_iloc iloc;
1865 if (!list_empty(&EXT3_I(inode)->i_orphan))
1868 /* Orphan handling is only valid for files with data blocks
1869 * being truncated, or files being unlinked. */
1871 /* @@@ FIXME: Observation from aviro:
1872 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1873 * here (on lock_super()), so race with ext3_link() which might bump
1874 * ->i_nlink. For, say it, character device. Not a regular file,
1875 * not a directory, not a symlink and ->i_nlink > 0.
1877 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1878 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1880 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1881 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1885 err = ext3_reserve_inode_write(handle, inode, &iloc);
1889 /* Insert this inode at the head of the on-disk orphan list... */
1890 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1891 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1892 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1893 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1897 /* Only add to the head of the in-memory list if all the
1898 * previous operations succeeded. If the orphan_add is going to
1899 * fail (possibly taking the journal offline), we can't risk
1900 * leaving the inode on the orphan list: stray orphan-list
1901 * entries can cause panics at unmount time.
1903 * This is safe: on error we're going to ignore the orphan list
1904 * anyway on the next recovery. */
1906 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1908 jbd_debug(4, "superblock will point to %ld\n", inode->i_ino);
1909 jbd_debug(4, "orphan inode %ld will point to %d\n",
1910 inode->i_ino, NEXT_ORPHAN(inode));
1913 ext3_std_error(inode->i_sb, err);
1918 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1919 * of such inodes stored on disk, because it is finally being cleaned up.
1921 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1923 struct list_head *prev;
1924 struct ext3_inode_info *ei = EXT3_I(inode);
1925 struct ext3_sb_info *sbi;
1926 unsigned long ino_next;
1927 struct ext3_iloc iloc;
1930 lock_super(inode->i_sb);
1931 if (list_empty(&ei->i_orphan)) {
1932 unlock_super(inode->i_sb);
1936 ino_next = NEXT_ORPHAN(inode);
1937 prev = ei->i_orphan.prev;
1938 sbi = EXT3_SB(inode->i_sb);
1940 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1942 list_del_init(&ei->i_orphan);
1944 /* If we're on an error path, we may not have a valid
1945 * transaction handle with which to update the orphan list on
1946 * disk, but we still need to remove the inode from the linked
1947 * list in memory. */
1951 err = ext3_reserve_inode_write(handle, inode, &iloc);
1955 if (prev == &sbi->s_orphan) {
1956 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1957 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1958 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1961 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1962 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1964 struct ext3_iloc iloc2;
1965 struct inode *i_prev =
1966 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1968 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1969 i_prev->i_ino, ino_next);
1970 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1973 NEXT_ORPHAN(i_prev) = ino_next;
1974 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
1978 NEXT_ORPHAN(inode) = 0;
1979 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
1982 ext3_std_error(inode->i_sb, err);
1984 unlock_super(inode->i_sb);
1992 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
1995 struct inode * inode;
1996 struct buffer_head * bh;
1997 struct ext3_dir_entry_2 * de;
2000 /* Initialize quotas before so that eventual writes go in
2001 * separate transaction */
2002 DQUOT_INIT(dentry->d_inode);
2003 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
2005 return PTR_ERR(handle);
2008 bh = ext3_find_entry (dentry, &de);
2012 if (IS_DIRSYNC(dir))
2015 inode = dentry->d_inode;
2018 if (le32_to_cpu(de->inode) != inode->i_ino)
2021 retval = -ENOTEMPTY;
2022 if (!empty_dir (inode))
2025 retval = ext3_delete_entry(handle, dir, de, bh);
2028 if (inode->i_nlink != 2)
2029 ext3_warning (inode->i_sb, "ext3_rmdir",
2030 "empty directory has nlink!=2 (%d)",
2034 /* There's no need to set i_disksize: the fact that i_nlink is
2035 * zero will ensure that the right thing happens during any
2038 ext3_orphan_add(handle, inode);
2039 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2040 ext3_mark_inode_dirty(handle, inode);
2042 ext3_update_dx_flag(dir);
2043 ext3_mark_inode_dirty(handle, dir);
2046 ext3_journal_stop(handle);
2051 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2054 struct inode * inode;
2055 struct buffer_head * bh;
2056 struct ext3_dir_entry_2 * de;
2059 /* Initialize quotas before so that eventual writes go
2060 * in separate transaction */
2061 DQUOT_INIT(dentry->d_inode);
2062 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
2064 return PTR_ERR(handle);
2066 if (IS_DIRSYNC(dir))
2070 bh = ext3_find_entry (dentry, &de);
2074 inode = dentry->d_inode;
2077 if (le32_to_cpu(de->inode) != inode->i_ino)
2080 if (!inode->i_nlink) {
2081 ext3_warning (inode->i_sb, "ext3_unlink",
2082 "Deleting nonexistent file (%lu), %d",
2083 inode->i_ino, inode->i_nlink);
2086 retval = ext3_delete_entry(handle, dir, de, bh);
2089 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2090 ext3_update_dx_flag(dir);
2091 ext3_mark_inode_dirty(handle, dir);
2093 if (!inode->i_nlink)
2094 ext3_orphan_add(handle, inode);
2095 inode->i_ctime = dir->i_ctime;
2096 ext3_mark_inode_dirty(handle, inode);
2100 ext3_journal_stop(handle);
2105 static int ext3_symlink (struct inode * dir,
2106 struct dentry *dentry, const char * symname)
2109 struct inode * inode;
2110 int l, err, retries = 0;
2112 l = strlen(symname)+1;
2113 if (l > dir->i_sb->s_blocksize)
2114 return -ENAMETOOLONG;
2117 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2118 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2119 2*EXT3_QUOTA_INIT_BLOCKS);
2121 return PTR_ERR(handle);
2123 if (IS_DIRSYNC(dir))
2126 inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2127 err = PTR_ERR(inode);
2131 if (l > sizeof (EXT3_I(inode)->i_data)) {
2132 inode->i_op = &ext3_symlink_inode_operations;
2133 ext3_set_aops(inode);
2135 * page_symlink() calls into ext3_prepare/commit_write.
2136 * We have a transaction open. All is sweetness. It also sets
2137 * i_size in generic_commit_write().
2139 err = page_symlink(inode, symname, l);
2141 ext3_dec_count(handle, inode);
2142 ext3_mark_inode_dirty(handle, inode);
2147 inode->i_op = &ext3_fast_symlink_inode_operations;
2148 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2149 inode->i_size = l-1;
2151 EXT3_I(inode)->i_disksize = inode->i_size;
2152 err = ext3_add_nondir(handle, dentry, inode);
2154 ext3_journal_stop(handle);
2155 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2160 static int ext3_link (struct dentry * old_dentry,
2161 struct inode * dir, struct dentry *dentry)
2164 struct inode *inode = old_dentry->d_inode;
2165 int err, retries = 0;
2167 if (inode->i_nlink >= EXT3_LINK_MAX)
2171 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2172 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2174 return PTR_ERR(handle);
2176 if (IS_DIRSYNC(dir))
2179 inode->i_ctime = CURRENT_TIME_SEC;
2180 ext3_inc_count(handle, inode);
2181 atomic_inc(&inode->i_count);
2183 err = ext3_add_nondir(handle, dentry, inode);
2184 ext3_journal_stop(handle);
2185 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2190 #define PARENT_INO(buffer) \
2191 ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2192 le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2195 * Anybody can rename anything with this: the permission checks are left to the
2196 * higher-level routines.
2198 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2199 struct inode * new_dir,struct dentry *new_dentry)
2202 struct inode * old_inode, * new_inode;
2203 struct buffer_head * old_bh, * new_bh, * dir_bh;
2204 struct ext3_dir_entry_2 * old_de, * new_de;
2207 old_bh = new_bh = dir_bh = NULL;
2209 /* Initialize quotas before so that eventual writes go
2210 * in separate transaction */
2211 if (new_dentry->d_inode)
2212 DQUOT_INIT(new_dentry->d_inode);
2213 handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS +
2214 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2216 return PTR_ERR(handle);
2218 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2221 old_bh = ext3_find_entry (old_dentry, &old_de);
2223 * Check for inode number is _not_ due to possible IO errors.
2224 * We might rmdir the source, keep it as pwd of some process
2225 * and merrily kill the link to whatever was created under the
2226 * same name. Goodbye sticky bit ;-<
2228 old_inode = old_dentry->d_inode;
2230 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2233 new_inode = new_dentry->d_inode;
2234 new_bh = ext3_find_entry (new_dentry, &new_de);
2241 if (S_ISDIR(old_inode->i_mode)) {
2243 retval = -ENOTEMPTY;
2244 if (!empty_dir (new_inode))
2248 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2251 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2254 if (!new_inode && new_dir!=old_dir &&
2255 new_dir->i_nlink >= EXT3_LINK_MAX)
2259 retval = ext3_add_entry (handle, new_dentry, old_inode);
2263 BUFFER_TRACE(new_bh, "get write access");
2264 ext3_journal_get_write_access(handle, new_bh);
2265 new_de->inode = cpu_to_le32(old_inode->i_ino);
2266 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2267 EXT3_FEATURE_INCOMPAT_FILETYPE))
2268 new_de->file_type = old_de->file_type;
2269 new_dir->i_version++;
2270 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2271 ext3_journal_dirty_metadata(handle, new_bh);
2277 * Like most other Unix systems, set the ctime for inodes on a
2280 old_inode->i_ctime = CURRENT_TIME_SEC;
2281 ext3_mark_inode_dirty(handle, old_inode);
2286 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2287 old_de->name_len != old_dentry->d_name.len ||
2288 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2289 (retval = ext3_delete_entry(handle, old_dir,
2290 old_de, old_bh)) == -ENOENT) {
2291 /* old_de could have moved from under us during htree split, so
2292 * make sure that we are deleting the right entry. We might
2293 * also be pointing to a stale entry in the unused part of
2294 * old_bh so just checking inum and the name isn't enough. */
2295 struct buffer_head *old_bh2;
2296 struct ext3_dir_entry_2 *old_de2;
2298 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2300 retval = ext3_delete_entry(handle, old_dir,
2306 ext3_warning(old_dir->i_sb, "ext3_rename",
2307 "Deleting old file (%lu), %d, error=%d",
2308 old_dir->i_ino, old_dir->i_nlink, retval);
2312 new_inode->i_nlink--;
2313 new_inode->i_ctime = CURRENT_TIME_SEC;
2315 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2316 ext3_update_dx_flag(old_dir);
2318 BUFFER_TRACE(dir_bh, "get_write_access");
2319 ext3_journal_get_write_access(handle, dir_bh);
2320 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2321 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2322 ext3_journal_dirty_metadata(handle, dir_bh);
2325 new_inode->i_nlink--;
2328 ext3_update_dx_flag(new_dir);
2329 ext3_mark_inode_dirty(handle, new_dir);
2332 ext3_mark_inode_dirty(handle, old_dir);
2334 ext3_mark_inode_dirty(handle, new_inode);
2335 if (!new_inode->i_nlink)
2336 ext3_orphan_add(handle, new_inode);
2344 ext3_journal_stop(handle);
2349 * directories can handle most operations...
2351 struct inode_operations ext3_dir_inode_operations = {
2352 .create = ext3_create,
2353 .lookup = ext3_lookup,
2355 .unlink = ext3_unlink,
2356 .symlink = ext3_symlink,
2357 .mkdir = ext3_mkdir,
2358 .rmdir = ext3_rmdir,
2359 .mknod = ext3_mknod,
2360 .rename = ext3_rename,
2361 .setattr = ext3_setattr,
2362 #ifdef CONFIG_EXT3_FS_XATTR
2363 .setxattr = generic_setxattr,
2364 .getxattr = generic_getxattr,
2365 .listxattr = ext3_listxattr,
2366 .removexattr = generic_removexattr,
2368 .permission = ext3_permission,
2371 struct inode_operations ext3_special_inode_operations = {
2372 .setattr = ext3_setattr,
2373 #ifdef CONFIG_EXT3_FS_XATTR
2374 .setxattr = generic_setxattr,
2375 .getxattr = generic_getxattr,
2376 .listxattr = ext3_listxattr,
2377 .removexattr = generic_removexattr,
2379 .permission = ext3_permission,