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>
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 *ext3_append(handle_t *handle,
57 struct buffer_head *bh;
59 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
61 if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
62 inode->i_size += inode->i_sb->s_blocksize;
63 EXT3_I(inode)->i_disksize = inode->i_size;
64 ext3_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_EXT3_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 ext3_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 ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
168 struct dx_map_entry *offsets, int count);
169 static struct ext3_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 ext3_htree_next_block(struct inode *dir, __u32 hash,
172 struct dx_frame *frame,
173 struct dx_frame *frames,
175 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
176 struct ext3_dir_entry_2 **res_dir, int *err);
177 static int ext3_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 - EXT3_DIR_REC_LEN(1) -
228 EXT3_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 - EXT3_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++)
248 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
260 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_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 ext3fs_dirhash(de->name, de->name_len, &h);
278 printk(":%x.%u ", h.hash,
279 ((char *) de - base));
281 space += EXT3_DIR_REC_LEN(de->name_len);
284 de = (struct ext3_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 = ext3_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 ext3_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 = ext3_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 ext3_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 = EXT3_SB(dir->i_sb)->s_hash_seed;
360 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
363 if (root->info.unused_flags & 1) {
364 ext3_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 ext3_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 = ext3_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 ext3_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 = ext3_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 ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
529 return (struct ext3_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 ext3_dir_entry_2 *de, *top;
546 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
547 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
550 de = (struct ext3_dir_entry_2 *) bh->b_data;
551 top = (struct ext3_dir_entry_2 *) ((char *) de +
552 dir->i_sb->s_blocksize -
553 EXT3_DIR_REC_LEN(0));
554 for (; de < top; de = ext3_next_entry(de)) {
555 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
556 (block<<EXT3_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 ext3fs_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 = ext3_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 ext3_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 ext3_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_dentry->d_inode;
606 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
607 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
608 hinfo.seed = EXT3_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_dentry->d_inode, &hinfo, frames, &err);
620 /* Add '.' and '..' from the htree header */
621 if (!start_hash && !start_minor_hash) {
622 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
623 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
627 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
628 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
629 de = ext3_next_entry(de);
630 if ((err = ext3_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 = ext3_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 ext3_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 ext3fs_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 ext3_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 */
706 if (count - 9 < 2) /* 9, 10 -> 11 */
708 for (p = top, q = p - count; q >= map; p--, q--)
709 if (p->hash < q->hash)
712 /* Garden variety bubble sort */
718 if (q[1].hash >= q[0].hash)
726 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
728 struct dx_entry *entries = frame->entries;
729 struct dx_entry *old = frame->at, *new = old + 1;
730 int count = dx_get_count(entries);
732 assert(count < dx_get_limit(entries));
733 assert(old < entries + count);
734 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
735 dx_set_hash(new, hash);
736 dx_set_block(new, block);
737 dx_set_count(entries, count + 1);
742 static void ext3_update_dx_flag(struct inode *inode)
744 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
745 EXT3_FEATURE_COMPAT_DIR_INDEX))
746 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
750 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
752 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
753 * `de != NULL' is guaranteed by caller.
755 static inline int ext3_match (int len, const char * const name,
756 struct ext3_dir_entry_2 * de)
758 if (len != de->name_len)
762 return !memcmp(name, de->name, len);
766 * Returns 0 if not found, -1 on failure, and 1 on success
768 static inline int search_dirblock(struct buffer_head * bh,
770 struct dentry *dentry,
771 unsigned long offset,
772 struct ext3_dir_entry_2 ** res_dir)
774 struct ext3_dir_entry_2 * de;
777 const char *name = dentry->d_name.name;
778 int namelen = dentry->d_name.len;
780 de = (struct ext3_dir_entry_2 *) bh->b_data;
781 dlimit = bh->b_data + dir->i_sb->s_blocksize;
782 while ((char *) de < dlimit) {
783 /* this code is executed quadratically often */
784 /* do minimal checking `by hand' */
786 if ((char *) de + namelen <= dlimit &&
787 ext3_match (namelen, name, de)) {
788 /* found a match - just to be sure, do a full check */
789 if (!ext3_check_dir_entry("ext3_find_entry",
790 dir, de, bh, offset))
795 /* prevent looping on a bad block */
796 de_len = le16_to_cpu(de->rec_len);
800 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
809 * finds an entry in the specified directory with the wanted name. It
810 * returns the cache buffer in which the entry was found, and the entry
811 * itself (as a parameter - res_dir). It does NOT read the inode of the
812 * entry - you'll have to do that yourself if you want to.
814 * The returned buffer_head has ->b_count elevated. The caller is expected
815 * to brelse() it when appropriate.
817 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
818 struct ext3_dir_entry_2 ** res_dir)
820 struct super_block * sb;
821 struct buffer_head * bh_use[NAMEI_RA_SIZE];
822 struct buffer_head * bh, *ret = NULL;
823 unsigned long start, block, b;
824 int ra_max = 0; /* Number of bh's in the readahead
826 int ra_ptr = 0; /* Current index into readahead
830 struct inode *dir = dentry->d_parent->d_inode;
837 blocksize = sb->s_blocksize;
838 namelen = dentry->d_name.len;
839 name = dentry->d_name.name;
840 if (namelen > EXT3_NAME_LEN)
842 #ifdef CONFIG_EXT3_INDEX
844 bh = ext3_dx_find_entry(dentry, res_dir, &err);
846 * On success, or if the error was file not found,
847 * return. Otherwise, fall back to doing a search the
850 if (bh || (err != ERR_BAD_DX_DIR))
852 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
855 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
856 start = EXT3_I(dir)->i_dir_start_lookup;
857 if (start >= nblocks)
863 * We deal with the read-ahead logic here.
865 if (ra_ptr >= ra_max) {
866 /* Refill the readahead buffer */
869 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
871 * Terminate if we reach the end of the
872 * directory and must wrap, or if our
873 * search has finished at this block.
875 if (b >= nblocks || (num && block == start)) {
876 bh_use[ra_max] = NULL;
880 bh = ext3_getblk(NULL, dir, b++, 0, &err);
883 ll_rw_block(READ, 1, &bh);
886 if ((bh = bh_use[ra_ptr++]) == NULL)
889 if (!buffer_uptodate(bh)) {
890 /* read error, skip block & hope for the best */
891 ext3_error(sb, __FUNCTION__, "reading directory #%lu "
892 "offset %lu", dir->i_ino, block);
896 i = search_dirblock(bh, dir, dentry,
897 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
899 EXT3_I(dir)->i_dir_start_lookup = block;
901 goto cleanup_and_exit;
905 goto cleanup_and_exit;
908 if (++block >= nblocks)
910 } while (block != start);
913 * If the directory has grown while we were searching, then
914 * search the last part of the directory before giving up.
917 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
918 if (block < nblocks) {
924 /* Clean up the read-ahead blocks */
925 for (; ra_ptr < ra_max; ra_ptr++)
926 brelse (bh_use[ra_ptr]);
930 #ifdef CONFIG_EXT3_INDEX
931 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
932 struct ext3_dir_entry_2 **res_dir, int *err)
934 struct super_block * sb;
935 struct dx_hash_info hinfo;
937 struct dx_frame frames[2], *frame;
938 struct ext3_dir_entry_2 *de, *top;
939 struct buffer_head *bh;
942 int namelen = dentry->d_name.len;
943 const u8 *name = dentry->d_name.name;
944 struct inode *dir = dentry->d_parent->d_inode;
947 /* NFS may look up ".." - look at dx_root directory block */
948 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
949 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
953 frame->bh = NULL; /* for dx_release() */
954 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
955 dx_set_block(frame->at, 0); /* dx_root block is 0 */
959 block = dx_get_block(frame->at);
960 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
962 de = (struct ext3_dir_entry_2 *) bh->b_data;
963 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
964 EXT3_DIR_REC_LEN(0));
965 for (; de < top; de = ext3_next_entry(de))
966 if (ext3_match (namelen, name, de)) {
967 if (!ext3_check_dir_entry("ext3_find_entry",
969 (block<<EXT3_BLOCK_SIZE_BITS(sb))
970 +((char *)de - bh->b_data))) {
979 /* Check to see if we should continue to search */
980 retval = ext3_htree_next_block(dir, hash, frame,
983 ext3_warning(sb, __FUNCTION__,
984 "error reading index page in directory #%lu",
989 } while (retval == 1);
993 dxtrace(printk("%s not found\n", name));
999 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1001 struct inode * inode;
1002 struct ext3_dir_entry_2 * de;
1003 struct buffer_head * bh;
1005 if (dentry->d_name.len > EXT3_NAME_LEN)
1006 return ERR_PTR(-ENAMETOOLONG);
1008 bh = ext3_find_entry(dentry, &de);
1011 unsigned long ino = le32_to_cpu(de->inode);
1013 if (!ext3_valid_inum(dir->i_sb, ino)) {
1014 ext3_error(dir->i_sb, "ext3_lookup",
1015 "bad inode number: %lu", ino);
1018 inode = iget(dir->i_sb, ino);
1021 return ERR_PTR(-EACCES);
1022 vx_propagate_xid(nd, inode);
1024 return d_splice_alias(inode, dentry);
1028 struct dentry *ext3_get_parent(struct dentry *child)
1031 struct dentry *parent;
1032 struct inode *inode;
1033 struct dentry dotdot;
1034 struct ext3_dir_entry_2 * de;
1035 struct buffer_head *bh;
1037 dotdot.d_name.name = "..";
1038 dotdot.d_name.len = 2;
1039 dotdot.d_parent = child; /* confusing, isn't it! */
1041 bh = ext3_find_entry(&dotdot, &de);
1044 return ERR_PTR(-ENOENT);
1045 ino = le32_to_cpu(de->inode);
1048 if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1049 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1050 "bad inode number: %lu", ino);
1053 inode = iget(child->d_inode->i_sb, ino);
1056 return ERR_PTR(-EACCES);
1058 parent = d_alloc_anon(inode);
1061 parent = ERR_PTR(-ENOMEM);
1067 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1068 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1069 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1070 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1071 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1072 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1073 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1074 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1077 static inline void ext3_set_de_type(struct super_block *sb,
1078 struct ext3_dir_entry_2 *de,
1080 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1081 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1084 #ifdef CONFIG_EXT3_INDEX
1085 static struct ext3_dir_entry_2 *
1086 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1088 unsigned rec_len = 0;
1091 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1092 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1093 memcpy (to, de, rec_len);
1094 ((struct ext3_dir_entry_2 *) to)->rec_len =
1095 cpu_to_le16(rec_len);
1100 return (struct ext3_dir_entry_2 *) (to - rec_len);
1103 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1105 struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1106 unsigned rec_len = 0;
1109 while ((char*)de < base + size) {
1110 next = (struct ext3_dir_entry_2 *) ((char *) de +
1111 le16_to_cpu(de->rec_len));
1112 if (de->inode && de->name_len) {
1113 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1115 memmove(to, de, rec_len);
1116 to->rec_len = cpu_to_le16(rec_len);
1118 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1125 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1126 struct buffer_head **bh,struct dx_frame *frame,
1127 struct dx_hash_info *hinfo, int *error)
1129 unsigned blocksize = dir->i_sb->s_blocksize;
1130 unsigned count, continued;
1131 struct buffer_head *bh2;
1134 struct dx_map_entry *map;
1135 char *data1 = (*bh)->b_data, *data2;
1137 struct ext3_dir_entry_2 *de = NULL, *de2;
1140 bh2 = ext3_append (handle, dir, &newblock, error);
1147 BUFFER_TRACE(*bh, "get_write_access");
1148 err = ext3_journal_get_write_access(handle, *bh);
1154 ext3_std_error(dir->i_sb, err);
1157 BUFFER_TRACE(frame->bh, "get_write_access");
1158 err = ext3_journal_get_write_access(handle, frame->bh);
1162 data2 = bh2->b_data;
1164 /* create map in the end of data2 block */
1165 map = (struct dx_map_entry *) (data2 + blocksize);
1166 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1167 blocksize, hinfo, map);
1169 split = count/2; // need to adjust to actual middle
1170 dx_sort_map (map, count);
1171 hash2 = map[split].hash;
1172 continued = hash2 == map[split - 1].hash;
1173 dxtrace(printk("Split block %i at %x, %i/%i\n",
1174 dx_get_block(frame->at), hash2, split, count-split));
1176 /* Fancy dance to stay within two buffers */
1177 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1178 de = dx_pack_dirents(data1,blocksize);
1179 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1180 de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1181 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1182 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1184 /* Which block gets the new entry? */
1185 if (hinfo->hash >= hash2)
1190 dx_insert_block (frame, hash2 + continued, newblock);
1191 err = ext3_journal_dirty_metadata (handle, bh2);
1194 err = ext3_journal_dirty_metadata (handle, frame->bh);
1198 dxtrace(dx_show_index ("frame", frame->entries));
1206 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1207 * it points to a directory entry which is guaranteed to be large
1208 * enough for new directory entry. If de is NULL, then
1209 * add_dirent_to_buf will attempt search the directory block for
1210 * space. It will return -ENOSPC if no space is available, and -EIO
1211 * and -EEXIST if directory entry already exists.
1213 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1214 * all other cases bh is released.
1216 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1217 struct inode *inode, struct ext3_dir_entry_2 *de,
1218 struct buffer_head * bh)
1220 struct inode *dir = dentry->d_parent->d_inode;
1221 const char *name = dentry->d_name.name;
1222 int namelen = dentry->d_name.len;
1223 unsigned long offset = 0;
1224 unsigned short reclen;
1225 int nlen, rlen, err;
1228 reclen = EXT3_DIR_REC_LEN(namelen);
1230 de = (struct ext3_dir_entry_2 *)bh->b_data;
1231 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1232 while ((char *) de <= top) {
1233 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1238 if (ext3_match (namelen, name, de)) {
1242 nlen = EXT3_DIR_REC_LEN(de->name_len);
1243 rlen = le16_to_cpu(de->rec_len);
1244 if ((de->inode? rlen - nlen: rlen) >= reclen)
1246 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1249 if ((char *) de > top)
1252 BUFFER_TRACE(bh, "get_write_access");
1253 err = ext3_journal_get_write_access(handle, bh);
1255 ext3_std_error(dir->i_sb, err);
1260 /* By now the buffer is marked for journaling */
1261 nlen = EXT3_DIR_REC_LEN(de->name_len);
1262 rlen = le16_to_cpu(de->rec_len);
1264 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1265 de1->rec_len = cpu_to_le16(rlen - nlen);
1266 de->rec_len = cpu_to_le16(nlen);
1269 de->file_type = EXT3_FT_UNKNOWN;
1271 de->inode = cpu_to_le32(inode->i_ino);
1272 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1275 de->name_len = namelen;
1276 memcpy (de->name, name, namelen);
1278 * XXX shouldn't update any times until successful
1279 * completion of syscall, but too many callers depend
1282 * XXX similarly, too many callers depend on
1283 * ext3_new_inode() setting the times, but error
1284 * recovery deletes the inode, so the worst that can
1285 * happen is that the times are slightly out of date
1286 * and/or different from the directory change time.
1288 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1289 ext3_update_dx_flag(dir);
1291 ext3_mark_inode_dirty(handle, dir);
1292 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1293 err = ext3_journal_dirty_metadata(handle, bh);
1295 ext3_std_error(dir->i_sb, err);
1300 #ifdef CONFIG_EXT3_INDEX
1302 * This converts a one block unindexed directory to a 3 block indexed
1303 * directory, and adds the dentry to the indexed directory.
1305 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1306 struct inode *inode, struct buffer_head *bh)
1308 struct inode *dir = dentry->d_parent->d_inode;
1309 const char *name = dentry->d_name.name;
1310 int namelen = dentry->d_name.len;
1311 struct buffer_head *bh2;
1312 struct dx_root *root;
1313 struct dx_frame frames[2], *frame;
1314 struct dx_entry *entries;
1315 struct ext3_dir_entry_2 *de, *de2;
1320 struct dx_hash_info hinfo;
1322 struct fake_dirent *fde;
1324 blocksize = dir->i_sb->s_blocksize;
1325 dxtrace(printk("Creating index\n"));
1326 retval = ext3_journal_get_write_access(handle, bh);
1328 ext3_std_error(dir->i_sb, retval);
1332 root = (struct dx_root *) bh->b_data;
1334 bh2 = ext3_append (handle, dir, &block, &retval);
1339 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1340 data1 = bh2->b_data;
1342 /* The 0th block becomes the root, move the dirents out */
1343 fde = &root->dotdot;
1344 de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1345 len = ((char *) root) + blocksize - (char *) de;
1346 memcpy (data1, de, len);
1347 de = (struct ext3_dir_entry_2 *) data1;
1349 while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1351 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1352 /* Initialize the root; the dot dirents already exist */
1353 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1354 de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1355 memset (&root->info, 0, sizeof(root->info));
1356 root->info.info_length = sizeof(root->info);
1357 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1358 entries = root->entries;
1359 dx_set_block (entries, 1);
1360 dx_set_count (entries, 1);
1361 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1363 /* Initialize as for dx_probe */
1364 hinfo.hash_version = root->info.hash_version;
1365 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1366 ext3fs_dirhash(name, namelen, &hinfo);
1368 frame->entries = entries;
1369 frame->at = entries;
1372 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1373 dx_release (frames);
1377 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1384 * adds a file entry to the specified directory, using the same
1385 * semantics as ext3_find_entry(). It returns NULL if it failed.
1387 * NOTE!! The inode part of 'de' is left at 0 - which means you
1388 * may not sleep between calling this and putting something into
1389 * the entry, as someone else might have used it while you slept.
1391 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1392 struct inode *inode)
1394 struct inode *dir = dentry->d_parent->d_inode;
1395 unsigned long offset;
1396 struct buffer_head * bh;
1397 struct ext3_dir_entry_2 *de;
1398 struct super_block * sb;
1400 #ifdef CONFIG_EXT3_INDEX
1407 blocksize = sb->s_blocksize;
1408 if (!dentry->d_name.len)
1410 #ifdef CONFIG_EXT3_INDEX
1412 retval = ext3_dx_add_entry(handle, dentry, inode);
1413 if (!retval || (retval != ERR_BAD_DX_DIR))
1415 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1417 ext3_mark_inode_dirty(handle, dir);
1420 blocks = dir->i_size >> sb->s_blocksize_bits;
1421 for (block = 0, offset = 0; block < blocks; block++) {
1422 bh = ext3_bread(handle, dir, block, 0, &retval);
1425 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1426 if (retval != -ENOSPC)
1429 #ifdef CONFIG_EXT3_INDEX
1430 if (blocks == 1 && !dx_fallback &&
1431 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1432 return make_indexed_dir(handle, dentry, inode, bh);
1436 bh = ext3_append(handle, dir, &block, &retval);
1439 de = (struct ext3_dir_entry_2 *) bh->b_data;
1441 de->rec_len = cpu_to_le16(blocksize);
1442 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1445 #ifdef CONFIG_EXT3_INDEX
1447 * Returns 0 for success, or a negative error value
1449 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1450 struct inode *inode)
1452 struct dx_frame frames[2], *frame;
1453 struct dx_entry *entries, *at;
1454 struct dx_hash_info hinfo;
1455 struct buffer_head * bh;
1456 struct inode *dir = dentry->d_parent->d_inode;
1457 struct super_block * sb = dir->i_sb;
1458 struct ext3_dir_entry_2 *de;
1461 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1464 entries = frame->entries;
1467 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1470 BUFFER_TRACE(bh, "get_write_access");
1471 err = ext3_journal_get_write_access(handle, bh);
1475 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1476 if (err != -ENOSPC) {
1481 /* Block full, should compress but for now just split */
1482 dxtrace(printk("using %u of %u node entries\n",
1483 dx_get_count(entries), dx_get_limit(entries)));
1484 /* Need to split index? */
1485 if (dx_get_count(entries) == dx_get_limit(entries)) {
1487 unsigned icount = dx_get_count(entries);
1488 int levels = frame - frames;
1489 struct dx_entry *entries2;
1490 struct dx_node *node2;
1491 struct buffer_head *bh2;
1493 if (levels && (dx_get_count(frames->entries) ==
1494 dx_get_limit(frames->entries))) {
1495 ext3_warning(sb, __FUNCTION__,
1496 "Directory index full!");
1500 bh2 = ext3_append (handle, dir, &newblock, &err);
1503 node2 = (struct dx_node *)(bh2->b_data);
1504 entries2 = node2->entries;
1505 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1506 node2->fake.inode = 0;
1507 BUFFER_TRACE(frame->bh, "get_write_access");
1508 err = ext3_journal_get_write_access(handle, frame->bh);
1512 unsigned icount1 = icount/2, icount2 = icount - icount1;
1513 unsigned hash2 = dx_get_hash(entries + icount1);
1514 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1516 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1517 err = ext3_journal_get_write_access(handle,
1522 memcpy ((char *) entries2, (char *) (entries + icount1),
1523 icount2 * sizeof(struct dx_entry));
1524 dx_set_count (entries, icount1);
1525 dx_set_count (entries2, icount2);
1526 dx_set_limit (entries2, dx_node_limit(dir));
1528 /* Which index block gets the new entry? */
1529 if (at - entries >= icount1) {
1530 frame->at = at = at - entries - icount1 + entries2;
1531 frame->entries = entries = entries2;
1532 swap(frame->bh, bh2);
1534 dx_insert_block (frames + 0, hash2, newblock);
1535 dxtrace(dx_show_index ("node", frames[1].entries));
1536 dxtrace(dx_show_index ("node",
1537 ((struct dx_node *) bh2->b_data)->entries));
1538 err = ext3_journal_dirty_metadata(handle, bh2);
1543 dxtrace(printk("Creating second level index...\n"));
1544 memcpy((char *) entries2, (char *) entries,
1545 icount * sizeof(struct dx_entry));
1546 dx_set_limit(entries2, dx_node_limit(dir));
1549 dx_set_count(entries, 1);
1550 dx_set_block(entries + 0, newblock);
1551 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1553 /* Add new access path frame */
1555 frame->at = at = at - entries + entries2;
1556 frame->entries = entries = entries2;
1558 err = ext3_journal_get_write_access(handle,
1563 ext3_journal_dirty_metadata(handle, frames[0].bh);
1565 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1568 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1573 ext3_std_error(dir->i_sb, err);
1583 * ext3_delete_entry deletes a directory entry by merging it with the
1586 static int ext3_delete_entry (handle_t *handle,
1588 struct ext3_dir_entry_2 * de_del,
1589 struct buffer_head * bh)
1591 struct ext3_dir_entry_2 * de, * pde;
1596 de = (struct ext3_dir_entry_2 *) bh->b_data;
1597 while (i < bh->b_size) {
1598 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1601 BUFFER_TRACE(bh, "get_write_access");
1602 ext3_journal_get_write_access(handle, bh);
1605 cpu_to_le16(le16_to_cpu(pde->rec_len) +
1606 le16_to_cpu(de->rec_len));
1610 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1611 ext3_journal_dirty_metadata(handle, bh);
1614 i += le16_to_cpu(de->rec_len);
1616 de = (struct ext3_dir_entry_2 *)
1617 ((char *) de + le16_to_cpu(de->rec_len));
1623 * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1624 * do not perform it in these functions. We perform it at the call site,
1627 static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1632 static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1637 static int ext3_add_nondir(handle_t *handle,
1638 struct dentry *dentry, struct inode *inode)
1640 int err = ext3_add_entry(handle, dentry, inode);
1642 ext3_mark_inode_dirty(handle, inode);
1643 d_instantiate(dentry, inode);
1646 ext3_dec_count(handle, inode);
1652 * By the time this is called, we already have created
1653 * the directory cache entry for the new file, but it
1654 * is so far negative - it has no inode.
1656 * If the create succeeds, we fill in the inode information
1657 * with d_instantiate().
1659 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1660 struct nameidata *nd)
1663 struct inode * inode;
1664 int err, retries = 0;
1667 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1668 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1669 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1671 return PTR_ERR(handle);
1673 if (IS_DIRSYNC(dir))
1676 inode = ext3_new_inode (handle, dir, mode);
1677 err = PTR_ERR(inode);
1678 if (!IS_ERR(inode)) {
1679 inode->i_op = &ext3_file_inode_operations;
1680 inode->i_fop = &ext3_file_operations;
1681 ext3_set_aops(inode);
1682 err = ext3_add_nondir(handle, dentry, inode);
1684 ext3_journal_stop(handle);
1685 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1690 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1691 int mode, dev_t rdev)
1694 struct inode *inode;
1695 int err, retries = 0;
1697 if (!new_valid_dev(rdev))
1701 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1702 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1703 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1705 return PTR_ERR(handle);
1707 if (IS_DIRSYNC(dir))
1710 inode = ext3_new_inode (handle, dir, mode);
1711 err = PTR_ERR(inode);
1712 if (!IS_ERR(inode)) {
1713 init_special_inode(inode, inode->i_mode, rdev);
1714 #ifdef CONFIG_EXT3_FS_XATTR
1715 inode->i_op = &ext3_special_inode_operations;
1717 err = ext3_add_nondir(handle, dentry, inode);
1719 ext3_journal_stop(handle);
1720 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1725 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1728 struct inode * inode;
1729 struct buffer_head * dir_block;
1730 struct ext3_dir_entry_2 * de;
1731 int err, retries = 0;
1733 if (dir->i_nlink >= EXT3_LINK_MAX)
1737 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1738 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1739 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1741 return PTR_ERR(handle);
1743 if (IS_DIRSYNC(dir))
1746 inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1747 err = PTR_ERR(inode);
1751 inode->i_op = &ext3_dir_inode_operations;
1752 inode->i_fop = &ext3_dir_operations;
1753 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1754 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1756 inode->i_nlink--; /* is this nlink == 0? */
1757 ext3_mark_inode_dirty(handle, inode);
1761 BUFFER_TRACE(dir_block, "get_write_access");
1762 ext3_journal_get_write_access(handle, dir_block);
1763 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1764 de->inode = cpu_to_le32(inode->i_ino);
1766 de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1767 strcpy (de->name, ".");
1768 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1769 de = (struct ext3_dir_entry_2 *)
1770 ((char *) de + le16_to_cpu(de->rec_len));
1771 de->inode = cpu_to_le32(dir->i_ino);
1772 de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1774 strcpy (de->name, "..");
1775 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1777 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1778 ext3_journal_dirty_metadata(handle, dir_block);
1780 ext3_mark_inode_dirty(handle, inode);
1781 err = ext3_add_entry (handle, dentry, inode);
1784 ext3_mark_inode_dirty(handle, inode);
1789 ext3_update_dx_flag(dir);
1790 ext3_mark_inode_dirty(handle, dir);
1791 d_instantiate(dentry, inode);
1793 ext3_journal_stop(handle);
1794 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1800 * routine to check that the specified directory is empty (for rmdir)
1802 static int empty_dir (struct inode * inode)
1804 unsigned long offset;
1805 struct buffer_head * bh;
1806 struct ext3_dir_entry_2 * de, * de1;
1807 struct super_block * sb;
1811 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1812 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1814 ext3_error(inode->i_sb, __FUNCTION__,
1815 "error %d reading directory #%lu offset 0",
1818 ext3_warning(inode->i_sb, __FUNCTION__,
1819 "bad directory (dir #%lu) - no data block",
1823 de = (struct ext3_dir_entry_2 *) bh->b_data;
1824 de1 = (struct ext3_dir_entry_2 *)
1825 ((char *) de + le16_to_cpu(de->rec_len));
1826 if (le32_to_cpu(de->inode) != inode->i_ino ||
1827 !le32_to_cpu(de1->inode) ||
1828 strcmp (".", de->name) ||
1829 strcmp ("..", de1->name)) {
1830 ext3_warning (inode->i_sb, "empty_dir",
1831 "bad directory (dir #%lu) - no `.' or `..'",
1836 offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1837 de = (struct ext3_dir_entry_2 *)
1838 ((char *) de1 + le16_to_cpu(de1->rec_len));
1839 while (offset < inode->i_size ) {
1841 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1844 bh = ext3_bread (NULL, inode,
1845 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1848 ext3_error(sb, __FUNCTION__,
1849 "error %d reading directory"
1851 err, inode->i_ino, offset);
1852 offset += sb->s_blocksize;
1855 de = (struct ext3_dir_entry_2 *) bh->b_data;
1857 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1858 de = (struct ext3_dir_entry_2 *)(bh->b_data +
1860 offset = (offset | (sb->s_blocksize - 1)) + 1;
1863 if (le32_to_cpu(de->inode)) {
1867 offset += le16_to_cpu(de->rec_len);
1868 de = (struct ext3_dir_entry_2 *)
1869 ((char *) de + le16_to_cpu(de->rec_len));
1875 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1876 * such inodes, starting at the superblock, in case we crash before the
1877 * file is closed/deleted, or in case the inode truncate spans multiple
1878 * transactions and the last transaction is not recovered after a crash.
1880 * At filesystem recovery time, we walk this list deleting unlinked
1881 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1883 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1885 struct super_block *sb = inode->i_sb;
1886 struct ext3_iloc iloc;
1890 if (!list_empty(&EXT3_I(inode)->i_orphan))
1893 /* Orphan handling is only valid for files with data blocks
1894 * being truncated, or files being unlinked. */
1896 /* @@@ FIXME: Observation from aviro:
1897 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1898 * here (on lock_super()), so race with ext3_link() which might bump
1899 * ->i_nlink. For, say it, character device. Not a regular file,
1900 * not a directory, not a symlink and ->i_nlink > 0.
1902 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1903 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1905 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1906 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1910 err = ext3_reserve_inode_write(handle, inode, &iloc);
1914 /* Insert this inode at the head of the on-disk orphan list... */
1915 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1916 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1917 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1918 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1922 /* Only add to the head of the in-memory list if all the
1923 * previous operations succeeded. If the orphan_add is going to
1924 * fail (possibly taking the journal offline), we can't risk
1925 * leaving the inode on the orphan list: stray orphan-list
1926 * entries can cause panics at unmount time.
1928 * This is safe: on error we're going to ignore the orphan list
1929 * anyway on the next recovery. */
1931 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1933 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1934 jbd_debug(4, "orphan inode %lu will point to %d\n",
1935 inode->i_ino, NEXT_ORPHAN(inode));
1938 ext3_std_error(inode->i_sb, err);
1943 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1944 * of such inodes stored on disk, because it is finally being cleaned up.
1946 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1948 struct list_head *prev;
1949 struct ext3_inode_info *ei = EXT3_I(inode);
1950 struct ext3_sb_info *sbi;
1951 unsigned long ino_next;
1952 struct ext3_iloc iloc;
1955 lock_super(inode->i_sb);
1956 if (list_empty(&ei->i_orphan)) {
1957 unlock_super(inode->i_sb);
1961 ino_next = NEXT_ORPHAN(inode);
1962 prev = ei->i_orphan.prev;
1963 sbi = EXT3_SB(inode->i_sb);
1965 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1967 list_del_init(&ei->i_orphan);
1969 /* If we're on an error path, we may not have a valid
1970 * transaction handle with which to update the orphan list on
1971 * disk, but we still need to remove the inode from the linked
1972 * list in memory. */
1976 err = ext3_reserve_inode_write(handle, inode, &iloc);
1980 if (prev == &sbi->s_orphan) {
1981 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1982 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1983 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1986 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1987 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1989 struct ext3_iloc iloc2;
1990 struct inode *i_prev =
1991 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1993 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1994 i_prev->i_ino, ino_next);
1995 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1998 NEXT_ORPHAN(i_prev) = ino_next;
1999 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2003 NEXT_ORPHAN(inode) = 0;
2004 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2007 ext3_std_error(inode->i_sb, err);
2009 unlock_super(inode->i_sb);
2017 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2020 struct inode * inode;
2021 struct buffer_head * bh;
2022 struct ext3_dir_entry_2 * de;
2025 /* Initialize quotas before so that eventual writes go in
2026 * separate transaction */
2027 DQUOT_INIT(dentry->d_inode);
2028 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2030 return PTR_ERR(handle);
2033 bh = ext3_find_entry (dentry, &de);
2037 if (IS_DIRSYNC(dir))
2040 inode = dentry->d_inode;
2043 if (le32_to_cpu(de->inode) != inode->i_ino)
2046 retval = -ENOTEMPTY;
2047 if (!empty_dir (inode))
2050 retval = ext3_delete_entry(handle, dir, de, bh);
2053 if (inode->i_nlink != 2)
2054 ext3_warning (inode->i_sb, "ext3_rmdir",
2055 "empty directory has nlink!=2 (%d)",
2059 /* There's no need to set i_disksize: the fact that i_nlink is
2060 * zero will ensure that the right thing happens during any
2063 ext3_orphan_add(handle, inode);
2064 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2065 ext3_mark_inode_dirty(handle, inode);
2067 ext3_update_dx_flag(dir);
2068 ext3_mark_inode_dirty(handle, dir);
2071 ext3_journal_stop(handle);
2076 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2079 struct inode * inode;
2080 struct buffer_head * bh;
2081 struct ext3_dir_entry_2 * de;
2084 /* Initialize quotas before so that eventual writes go
2085 * in separate transaction */
2086 DQUOT_INIT(dentry->d_inode);
2087 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2089 return PTR_ERR(handle);
2091 if (IS_DIRSYNC(dir))
2095 bh = ext3_find_entry (dentry, &de);
2099 inode = dentry->d_inode;
2102 if (le32_to_cpu(de->inode) != inode->i_ino)
2105 if (!inode->i_nlink) {
2106 ext3_warning (inode->i_sb, "ext3_unlink",
2107 "Deleting nonexistent file (%lu), %d",
2108 inode->i_ino, inode->i_nlink);
2111 retval = ext3_delete_entry(handle, dir, de, bh);
2114 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2115 ext3_update_dx_flag(dir);
2116 ext3_mark_inode_dirty(handle, dir);
2118 if (!inode->i_nlink)
2119 ext3_orphan_add(handle, inode);
2120 inode->i_ctime = dir->i_ctime;
2121 ext3_mark_inode_dirty(handle, inode);
2125 ext3_journal_stop(handle);
2130 static int ext3_symlink (struct inode * dir,
2131 struct dentry *dentry, const char * symname)
2134 struct inode * inode;
2135 int l, err, retries = 0;
2137 l = strlen(symname)+1;
2138 if (l > dir->i_sb->s_blocksize)
2139 return -ENAMETOOLONG;
2142 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2143 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2144 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2146 return PTR_ERR(handle);
2148 if (IS_DIRSYNC(dir))
2151 inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2152 err = PTR_ERR(inode);
2156 if (l > sizeof (EXT3_I(inode)->i_data)) {
2157 inode->i_op = &ext3_symlink_inode_operations;
2158 ext3_set_aops(inode);
2160 * page_symlink() calls into ext3_prepare/commit_write.
2161 * We have a transaction open. All is sweetness. It also sets
2162 * i_size in generic_commit_write().
2164 err = __page_symlink(inode, symname, l,
2165 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2167 ext3_dec_count(handle, inode);
2168 ext3_mark_inode_dirty(handle, inode);
2173 inode->i_op = &ext3_fast_symlink_inode_operations;
2174 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2175 inode->i_size = l-1;
2177 EXT3_I(inode)->i_disksize = inode->i_size;
2178 err = ext3_add_nondir(handle, dentry, inode);
2180 ext3_journal_stop(handle);
2181 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2186 static int ext3_link (struct dentry * old_dentry,
2187 struct inode * dir, struct dentry *dentry)
2190 struct inode *inode = old_dentry->d_inode;
2191 int err, retries = 0;
2193 if (inode->i_nlink >= EXT3_LINK_MAX)
2197 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2198 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2200 return PTR_ERR(handle);
2202 if (IS_DIRSYNC(dir))
2205 inode->i_ctime = CURRENT_TIME_SEC;
2206 ext3_inc_count(handle, inode);
2207 atomic_inc(&inode->i_count);
2209 err = ext3_add_nondir(handle, dentry, inode);
2210 ext3_journal_stop(handle);
2211 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2216 #define PARENT_INO(buffer) \
2217 ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2218 le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2221 * Anybody can rename anything with this: the permission checks are left to the
2222 * higher-level routines.
2224 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2225 struct inode * new_dir,struct dentry *new_dentry)
2228 struct inode * old_inode, * new_inode;
2229 struct buffer_head * old_bh, * new_bh, * dir_bh;
2230 struct ext3_dir_entry_2 * old_de, * new_de;
2233 old_bh = new_bh = dir_bh = NULL;
2235 /* Initialize quotas before so that eventual writes go
2236 * in separate transaction */
2237 if (new_dentry->d_inode)
2238 DQUOT_INIT(new_dentry->d_inode);
2239 handle = ext3_journal_start(old_dir, 2 *
2240 EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2241 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2243 return PTR_ERR(handle);
2245 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2248 old_bh = ext3_find_entry (old_dentry, &old_de);
2250 * Check for inode number is _not_ due to possible IO errors.
2251 * We might rmdir the source, keep it as pwd of some process
2252 * and merrily kill the link to whatever was created under the
2253 * same name. Goodbye sticky bit ;-<
2255 old_inode = old_dentry->d_inode;
2257 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2260 new_inode = new_dentry->d_inode;
2261 new_bh = ext3_find_entry (new_dentry, &new_de);
2268 if (S_ISDIR(old_inode->i_mode)) {
2270 retval = -ENOTEMPTY;
2271 if (!empty_dir (new_inode))
2275 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2278 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2281 if (!new_inode && new_dir!=old_dir &&
2282 new_dir->i_nlink >= EXT3_LINK_MAX)
2286 retval = ext3_add_entry (handle, new_dentry, old_inode);
2290 BUFFER_TRACE(new_bh, "get write access");
2291 ext3_journal_get_write_access(handle, new_bh);
2292 new_de->inode = cpu_to_le32(old_inode->i_ino);
2293 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2294 EXT3_FEATURE_INCOMPAT_FILETYPE))
2295 new_de->file_type = old_de->file_type;
2296 new_dir->i_version++;
2297 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2298 ext3_journal_dirty_metadata(handle, new_bh);
2304 * Like most other Unix systems, set the ctime for inodes on a
2307 old_inode->i_ctime = CURRENT_TIME_SEC;
2308 ext3_mark_inode_dirty(handle, old_inode);
2313 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2314 old_de->name_len != old_dentry->d_name.len ||
2315 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2316 (retval = ext3_delete_entry(handle, old_dir,
2317 old_de, old_bh)) == -ENOENT) {
2318 /* old_de could have moved from under us during htree split, so
2319 * make sure that we are deleting the right entry. We might
2320 * also be pointing to a stale entry in the unused part of
2321 * old_bh so just checking inum and the name isn't enough. */
2322 struct buffer_head *old_bh2;
2323 struct ext3_dir_entry_2 *old_de2;
2325 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2327 retval = ext3_delete_entry(handle, old_dir,
2333 ext3_warning(old_dir->i_sb, "ext3_rename",
2334 "Deleting old file (%lu), %d, error=%d",
2335 old_dir->i_ino, old_dir->i_nlink, retval);
2339 new_inode->i_nlink--;
2340 new_inode->i_ctime = CURRENT_TIME_SEC;
2342 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2343 ext3_update_dx_flag(old_dir);
2345 BUFFER_TRACE(dir_bh, "get_write_access");
2346 ext3_journal_get_write_access(handle, dir_bh);
2347 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2348 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2349 ext3_journal_dirty_metadata(handle, dir_bh);
2352 new_inode->i_nlink--;
2355 ext3_update_dx_flag(new_dir);
2356 ext3_mark_inode_dirty(handle, new_dir);
2359 ext3_mark_inode_dirty(handle, old_dir);
2361 ext3_mark_inode_dirty(handle, new_inode);
2362 if (!new_inode->i_nlink)
2363 ext3_orphan_add(handle, new_inode);
2371 ext3_journal_stop(handle);
2376 * directories can handle most operations...
2378 struct inode_operations ext3_dir_inode_operations = {
2379 .create = ext3_create,
2380 .lookup = ext3_lookup,
2382 .unlink = ext3_unlink,
2383 .symlink = ext3_symlink,
2384 .mkdir = ext3_mkdir,
2385 .rmdir = ext3_rmdir,
2386 .mknod = ext3_mknod,
2387 .rename = ext3_rename,
2388 .setattr = ext3_setattr,
2389 #ifdef CONFIG_EXT3_FS_XATTR
2390 .setxattr = generic_setxattr,
2391 .getxattr = generic_getxattr,
2392 .listxattr = ext3_listxattr,
2393 .removexattr = generic_removexattr,
2395 .permission = ext3_permission,
2396 .sync_flags = ext3_sync_flags,
2399 struct inode_operations ext3_special_inode_operations = {
2400 .setattr = ext3_setattr,
2401 #ifdef CONFIG_EXT3_FS_XATTR
2402 .setxattr = generic_setxattr,
2403 .getxattr = generic_getxattr,
2404 .listxattr = ext3_listxattr,
2405 .removexattr = generic_removexattr,
2407 .permission = ext3_permission,
2408 .sync_flags = ext3_sync_flags,