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/bio.h>
39 #include <linux/smp_lock.h>
40 #include <linux/vs_tag.h>
47 * define how far ahead to read directories while searching them.
49 #define NAMEI_RA_CHUNKS 2
50 #define NAMEI_RA_BLOCKS 4
51 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
52 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
54 static struct buffer_head *ext3_append(handle_t *handle,
58 struct buffer_head *bh;
60 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
62 if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
63 inode->i_size += inode->i_sb->s_blocksize;
64 EXT3_I(inode)->i_disksize = inode->i_size;
65 ext3_journal_get_write_access(handle,bh);
71 #define assert(test) J_ASSERT(test)
75 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
79 #define dxtrace(command) command
81 #define dxtrace(command)
105 * dx_root_info is laid out so that if it should somehow get overlaid by a
106 * dirent the two low bits of the hash version will be zero. Therefore, the
107 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
112 struct fake_dirent dot;
114 struct fake_dirent dotdot;
118 __le32 reserved_zero;
120 u8 info_length; /* 8 */
125 struct dx_entry entries[0];
130 struct fake_dirent fake;
131 struct dx_entry entries[0];
137 struct buffer_head *bh;
138 struct dx_entry *entries;
148 #ifdef CONFIG_EXT3_INDEX
149 static inline unsigned dx_get_block (struct dx_entry *entry);
150 static void dx_set_block (struct dx_entry *entry, unsigned value);
151 static inline unsigned dx_get_hash (struct dx_entry *entry);
152 static void dx_set_hash (struct dx_entry *entry, unsigned value);
153 static unsigned dx_get_count (struct dx_entry *entries);
154 static unsigned dx_get_limit (struct dx_entry *entries);
155 static void dx_set_count (struct dx_entry *entries, unsigned value);
156 static void dx_set_limit (struct dx_entry *entries, unsigned value);
157 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
158 static unsigned dx_node_limit (struct inode *dir);
159 static struct dx_frame *dx_probe(struct dentry *dentry,
161 struct dx_hash_info *hinfo,
162 struct dx_frame *frame,
164 static void dx_release (struct dx_frame *frames);
165 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
166 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
167 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
168 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
169 struct dx_map_entry *offsets, int count);
170 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
171 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
172 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
173 struct dx_frame *frame,
174 struct dx_frame *frames,
176 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
177 struct ext3_dir_entry_2 **res_dir, int *err);
178 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
179 struct inode *inode);
182 * Future: use high four bits of block for coalesce-on-delete flags
183 * Mask them off for now.
186 static inline unsigned dx_get_block (struct dx_entry *entry)
188 return le32_to_cpu(entry->block) & 0x00ffffff;
191 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
193 entry->block = cpu_to_le32(value);
196 static inline unsigned dx_get_hash (struct dx_entry *entry)
198 return le32_to_cpu(entry->hash);
201 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
203 entry->hash = cpu_to_le32(value);
206 static inline unsigned dx_get_count (struct dx_entry *entries)
208 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
211 static inline unsigned dx_get_limit (struct dx_entry *entries)
213 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
216 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
218 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
221 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
223 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
226 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
228 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
229 EXT3_DIR_REC_LEN(2) - infosize;
230 return 0? 20: entry_space / sizeof(struct dx_entry);
233 static inline unsigned dx_node_limit (struct inode *dir)
235 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
236 return 0? 22: entry_space / sizeof(struct dx_entry);
243 static void dx_show_index (char * label, struct dx_entry *entries)
245 int i, n = dx_get_count (entries);
246 printk("%s index ", label);
247 for (i = 0; i < n; i++)
249 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
261 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
262 int size, int show_names)
264 unsigned names = 0, space = 0;
265 char *base = (char *) de;
266 struct dx_hash_info h = *hinfo;
269 while ((char *) de < base + size)
275 int len = de->name_len;
276 char *name = de->name;
277 while (len--) printk("%c", *name++);
278 ext3fs_dirhash(de->name, de->name_len, &h);
279 printk(":%x.%u ", h.hash,
280 ((char *) de - base));
282 space += EXT3_DIR_REC_LEN(de->name_len);
285 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
287 printk("(%i)\n", names);
288 return (struct stats) { names, space, 1 };
291 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
292 struct dx_entry *entries, int levels)
294 unsigned blocksize = dir->i_sb->s_blocksize;
295 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
297 struct buffer_head *bh;
299 printk("%i indexed blocks...\n", count);
300 for (i = 0; i < count; i++, entries++)
302 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
303 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
305 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
306 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
308 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
309 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
310 names += stats.names;
311 space += stats.space;
312 bcount += stats.bcount;
316 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
317 names, space/bcount,(space/bcount)*100/blocksize);
318 return (struct stats) { names, space, bcount};
320 #endif /* DX_DEBUG */
323 * Probe for a directory leaf block to search.
325 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
326 * error in the directory index, and the caller should fall back to
327 * searching the directory normally. The callers of dx_probe **MUST**
328 * check for this error code, and make sure it never gets reflected
331 static struct dx_frame *
332 dx_probe(struct dentry *dentry, struct inode *dir,
333 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
335 unsigned count, indirect;
336 struct dx_entry *at, *entries, *p, *q, *m;
337 struct dx_root *root;
338 struct buffer_head *bh;
339 struct dx_frame *frame = frame_in;
344 dir = dentry->d_parent->d_inode;
345 if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
347 root = (struct dx_root *) bh->b_data;
348 if (root->info.hash_version != DX_HASH_TEA &&
349 root->info.hash_version != DX_HASH_HALF_MD4 &&
350 root->info.hash_version != DX_HASH_LEGACY) {
351 ext3_warning(dir->i_sb, __FUNCTION__,
352 "Unrecognised inode hash code %d",
353 root->info.hash_version);
355 *err = ERR_BAD_DX_DIR;
358 hinfo->hash_version = root->info.hash_version;
359 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
361 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
364 if (root->info.unused_flags & 1) {
365 ext3_warning(dir->i_sb, __FUNCTION__,
366 "Unimplemented inode hash flags: %#06x",
367 root->info.unused_flags);
369 *err = ERR_BAD_DX_DIR;
373 if ((indirect = root->info.indirect_levels) > 1) {
374 ext3_warning(dir->i_sb, __FUNCTION__,
375 "Unimplemented inode hash depth: %#06x",
376 root->info.indirect_levels);
378 *err = ERR_BAD_DX_DIR;
382 entries = (struct dx_entry *) (((char *)&root->info) +
383 root->info.info_length);
384 assert(dx_get_limit(entries) == dx_root_limit(dir,
385 root->info.info_length));
386 dxtrace (printk("Look up %x", hash));
389 count = dx_get_count(entries);
390 assert (count && count <= dx_get_limit(entries));
392 q = entries + count - 1;
396 dxtrace(printk("."));
397 if (dx_get_hash(m) > hash)
403 if (0) // linear search cross check
405 unsigned n = count - 1;
409 dxtrace(printk(","));
410 if (dx_get_hash(++at) > hash)
416 assert (at == p - 1);
420 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
422 frame->entries = entries;
424 if (!indirect--) return frame;
425 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
427 at = entries = ((struct dx_node *) bh->b_data)->entries;
428 assert (dx_get_limit(entries) == dx_node_limit (dir));
432 while (frame >= frame_in) {
440 static void dx_release (struct dx_frame *frames)
442 if (frames[0].bh == NULL)
445 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
446 brelse(frames[1].bh);
447 brelse(frames[0].bh);
451 * This function increments the frame pointer to search the next leaf
452 * block, and reads in the necessary intervening nodes if the search
453 * should be necessary. Whether or not the search is necessary is
454 * controlled by the hash parameter. If the hash value is even, then
455 * the search is only continued if the next block starts with that
456 * hash value. This is used if we are searching for a specific file.
458 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
460 * This function returns 1 if the caller should continue to search,
461 * or 0 if it should not. If there is an error reading one of the
462 * index blocks, it will a negative error code.
464 * If start_hash is non-null, it will be filled in with the starting
465 * hash of the next page.
467 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
468 struct dx_frame *frame,
469 struct dx_frame *frames,
473 struct buffer_head *bh;
474 int err, num_frames = 0;
479 * Find the next leaf page by incrementing the frame pointer.
480 * If we run out of entries in the interior node, loop around and
481 * increment pointer in the parent node. When we break out of
482 * this loop, num_frames indicates the number of interior
483 * nodes need to be read.
486 if (++(p->at) < p->entries + dx_get_count(p->entries))
495 * If the hash is 1, then continue only if the next page has a
496 * continuation hash of any value. This is used for readdir
497 * handling. Otherwise, check to see if the hash matches the
498 * desired contiuation hash. If it doesn't, return since
499 * there's no point to read in the successive index pages.
501 bhash = dx_get_hash(p->at);
504 if ((hash & 1) == 0) {
505 if ((bhash & ~1) != hash)
509 * If the hash is HASH_NB_ALWAYS, we always go to the next
510 * block so no check is necessary
512 while (num_frames--) {
513 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
515 return err; /* Failure */
519 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
526 * p is at least 6 bytes before the end of page
528 static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
530 return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
534 * This function fills a red-black tree with information from a
535 * directory block. It returns the number directory entries loaded
536 * into the tree. If there is an error it is returned in err.
538 static int htree_dirblock_to_tree(struct file *dir_file,
539 struct inode *dir, int block,
540 struct dx_hash_info *hinfo,
541 __u32 start_hash, __u32 start_minor_hash)
543 struct buffer_head *bh;
544 struct ext3_dir_entry_2 *de, *top;
547 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
548 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
551 de = (struct ext3_dir_entry_2 *) bh->b_data;
552 top = (struct ext3_dir_entry_2 *) ((char *) de +
553 dir->i_sb->s_blocksize -
554 EXT3_DIR_REC_LEN(0));
555 for (; de < top; de = ext3_next_entry(de)) {
556 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
557 (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
558 +((char *)de - bh->b_data))) {
559 /* On error, skip the f_pos to the next block. */
560 dir_file->f_pos = (dir_file->f_pos |
561 (dir->i_sb->s_blocksize - 1)) + 1;
565 ext3fs_dirhash(de->name, de->name_len, hinfo);
566 if ((hinfo->hash < start_hash) ||
567 ((hinfo->hash == start_hash) &&
568 (hinfo->minor_hash < start_minor_hash)))
572 if ((err = ext3_htree_store_dirent(dir_file,
573 hinfo->hash, hinfo->minor_hash, de)) != 0) {
585 * This function fills a red-black tree with information from a
586 * directory. We start scanning the directory in hash order, starting
587 * at start_hash and start_minor_hash.
589 * This function returns the number of entries inserted into the tree,
590 * or a negative error code.
592 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
593 __u32 start_minor_hash, __u32 *next_hash)
595 struct dx_hash_info hinfo;
596 struct ext3_dir_entry_2 *de;
597 struct dx_frame frames[2], *frame;
604 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
606 dir = dir_file->f_path.dentry->d_inode;
607 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
608 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
609 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
610 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
611 start_hash, start_minor_hash);
615 hinfo.hash = start_hash;
616 hinfo.minor_hash = 0;
617 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
621 /* Add '.' and '..' from the htree header */
622 if (!start_hash && !start_minor_hash) {
623 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
624 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
628 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
629 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
630 de = ext3_next_entry(de);
631 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
637 block = dx_get_block(frame->at);
638 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
639 start_hash, start_minor_hash);
646 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
647 frame, frames, &hashval);
648 *next_hash = hashval;
654 * Stop if: (a) there are no more entries, or
655 * (b) we have inserted at least one entry and the
656 * next hash value is not a continuation
659 (count && ((hashval & 1) == 0)))
663 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
673 * Directory block splitting, compacting
676 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
677 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
680 char *base = (char *) de;
681 struct dx_hash_info h = *hinfo;
683 while ((char *) de < base + size)
685 if (de->name_len && de->inode) {
686 ext3fs_dirhash(de->name, de->name_len, &h);
688 map_tail->hash = h.hash;
689 map_tail->offs = (u32) ((char *) de - base);
693 /* XXX: do we need to check rec_len == 0 case? -Chris */
694 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
699 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
701 struct dx_map_entry *p, *q, *top = map + count - 1;
703 /* Combsort until bubble sort doesn't suck */
707 if (count - 9 < 2) /* 9, 10 -> 11 */
709 for (p = top, q = p - count; q >= map; p--, q--)
710 if (p->hash < q->hash)
713 /* Garden variety bubble sort */
719 if (q[1].hash >= q[0].hash)
727 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
729 struct dx_entry *entries = frame->entries;
730 struct dx_entry *old = frame->at, *new = old + 1;
731 int count = dx_get_count(entries);
733 assert(count < dx_get_limit(entries));
734 assert(old < entries + count);
735 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
736 dx_set_hash(new, hash);
737 dx_set_block(new, block);
738 dx_set_count(entries, count + 1);
743 static void ext3_update_dx_flag(struct inode *inode)
745 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
746 EXT3_FEATURE_COMPAT_DIR_INDEX))
747 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
751 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
753 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
754 * `de != NULL' is guaranteed by caller.
756 static inline int ext3_match (int len, const char * const name,
757 struct ext3_dir_entry_2 * de)
759 if (len != de->name_len)
763 return !memcmp(name, de->name, len);
767 * Returns 0 if not found, -1 on failure, and 1 on success
769 static inline int search_dirblock(struct buffer_head * bh,
771 struct dentry *dentry,
772 unsigned long offset,
773 struct ext3_dir_entry_2 ** res_dir)
775 struct ext3_dir_entry_2 * de;
778 const char *name = dentry->d_name.name;
779 int namelen = dentry->d_name.len;
781 de = (struct ext3_dir_entry_2 *) bh->b_data;
782 dlimit = bh->b_data + dir->i_sb->s_blocksize;
783 while ((char *) de < dlimit) {
784 /* this code is executed quadratically often */
785 /* do minimal checking `by hand' */
787 if ((char *) de + namelen <= dlimit &&
788 ext3_match (namelen, name, de)) {
789 /* found a match - just to be sure, do a full check */
790 if (!ext3_check_dir_entry("ext3_find_entry",
791 dir, de, bh, offset))
796 /* prevent looping on a bad block */
797 de_len = le16_to_cpu(de->rec_len);
801 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
810 * finds an entry in the specified directory with the wanted name. It
811 * returns the cache buffer in which the entry was found, and the entry
812 * itself (as a parameter - res_dir). It does NOT read the inode of the
813 * entry - you'll have to do that yourself if you want to.
815 * The returned buffer_head has ->b_count elevated. The caller is expected
816 * to brelse() it when appropriate.
818 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
819 struct ext3_dir_entry_2 ** res_dir)
821 struct super_block * sb;
822 struct buffer_head * bh_use[NAMEI_RA_SIZE];
823 struct buffer_head * bh, *ret = NULL;
824 unsigned long start, block, b;
825 int ra_max = 0; /* Number of bh's in the readahead
827 int ra_ptr = 0; /* Current index into readahead
831 struct inode *dir = dentry->d_parent->d_inode;
838 blocksize = sb->s_blocksize;
839 namelen = dentry->d_name.len;
840 name = dentry->d_name.name;
841 if (namelen > EXT3_NAME_LEN)
843 #ifdef CONFIG_EXT3_INDEX
845 bh = ext3_dx_find_entry(dentry, res_dir, &err);
847 * On success, or if the error was file not found,
848 * return. Otherwise, fall back to doing a search the
851 if (bh || (err != ERR_BAD_DX_DIR))
853 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
856 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
857 start = EXT3_I(dir)->i_dir_start_lookup;
858 if (start >= nblocks)
864 * We deal with the read-ahead logic here.
866 if (ra_ptr >= ra_max) {
867 /* Refill the readahead buffer */
870 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
872 * Terminate if we reach the end of the
873 * directory and must wrap, or if our
874 * search has finished at this block.
876 if (b >= nblocks || (num && block == start)) {
877 bh_use[ra_max] = NULL;
881 bh = ext3_getblk(NULL, dir, b++, 0, &err);
884 ll_rw_block(READ_META, 1, &bh);
887 if ((bh = bh_use[ra_ptr++]) == NULL)
890 if (!buffer_uptodate(bh)) {
891 /* read error, skip block & hope for the best */
892 ext3_error(sb, __FUNCTION__, "reading directory #%lu "
893 "offset %lu", dir->i_ino, block);
897 i = search_dirblock(bh, dir, dentry,
898 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
900 EXT3_I(dir)->i_dir_start_lookup = block;
902 goto cleanup_and_exit;
906 goto cleanup_and_exit;
909 if (++block >= nblocks)
911 } while (block != start);
914 * If the directory has grown while we were searching, then
915 * search the last part of the directory before giving up.
918 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
919 if (block < nblocks) {
925 /* Clean up the read-ahead blocks */
926 for (; ra_ptr < ra_max; ra_ptr++)
927 brelse (bh_use[ra_ptr]);
931 #ifdef CONFIG_EXT3_INDEX
932 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
933 struct ext3_dir_entry_2 **res_dir, int *err)
935 struct super_block * sb;
936 struct dx_hash_info hinfo;
938 struct dx_frame frames[2], *frame;
939 struct ext3_dir_entry_2 *de, *top;
940 struct buffer_head *bh;
943 int namelen = dentry->d_name.len;
944 const u8 *name = dentry->d_name.name;
945 struct inode *dir = dentry->d_parent->d_inode;
948 /* NFS may look up ".." - look at dx_root directory block */
949 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
950 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
954 frame->bh = NULL; /* for dx_release() */
955 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
956 dx_set_block(frame->at, 0); /* dx_root block is 0 */
960 block = dx_get_block(frame->at);
961 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
963 de = (struct ext3_dir_entry_2 *) bh->b_data;
964 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
965 EXT3_DIR_REC_LEN(0));
966 for (; de < top; de = ext3_next_entry(de))
967 if (ext3_match (namelen, name, de)) {
968 if (!ext3_check_dir_entry("ext3_find_entry",
970 (block<<EXT3_BLOCK_SIZE_BITS(sb))
971 +((char *)de - bh->b_data))) {
980 /* Check to see if we should continue to search */
981 retval = ext3_htree_next_block(dir, hash, frame,
984 ext3_warning(sb, __FUNCTION__,
985 "error reading index page in directory #%lu",
990 } while (retval == 1);
994 dxtrace(printk("%s not found\n", name));
1000 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1002 struct inode * inode;
1003 struct ext3_dir_entry_2 * de;
1004 struct buffer_head * bh;
1006 if (dentry->d_name.len > EXT3_NAME_LEN)
1007 return ERR_PTR(-ENAMETOOLONG);
1009 bh = ext3_find_entry(dentry, &de);
1012 unsigned long ino = le32_to_cpu(de->inode);
1014 if (!ext3_valid_inum(dir->i_sb, ino)) {
1015 ext3_error(dir->i_sb, "ext3_lookup",
1016 "bad inode number: %lu", ino);
1019 inode = iget(dir->i_sb, ino);
1022 return ERR_PTR(-EACCES);
1023 dx_propagate_tag(nd, inode);
1025 return d_splice_alias(inode, dentry);
1029 struct dentry *ext3_get_parent(struct dentry *child)
1032 struct dentry *parent;
1033 struct inode *inode;
1034 struct dentry dotdot;
1035 struct ext3_dir_entry_2 * de;
1036 struct buffer_head *bh;
1038 dotdot.d_name.name = "..";
1039 dotdot.d_name.len = 2;
1040 dotdot.d_parent = child; /* confusing, isn't it! */
1042 bh = ext3_find_entry(&dotdot, &de);
1045 return ERR_PTR(-ENOENT);
1046 ino = le32_to_cpu(de->inode);
1049 if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1050 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1051 "bad inode number: %lu", ino);
1054 inode = iget(child->d_inode->i_sb, ino);
1057 return ERR_PTR(-EACCES);
1059 parent = d_alloc_anon(inode);
1062 parent = ERR_PTR(-ENOMEM);
1068 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1069 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1070 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1071 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1072 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1073 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1074 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1075 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1078 static inline void ext3_set_de_type(struct super_block *sb,
1079 struct ext3_dir_entry_2 *de,
1081 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1082 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1085 #ifdef CONFIG_EXT3_INDEX
1086 static struct ext3_dir_entry_2 *
1087 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1089 unsigned rec_len = 0;
1092 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1093 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1094 memcpy (to, de, rec_len);
1095 ((struct ext3_dir_entry_2 *) to)->rec_len =
1096 cpu_to_le16(rec_len);
1101 return (struct ext3_dir_entry_2 *) (to - rec_len);
1104 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1106 struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1107 unsigned rec_len = 0;
1110 while ((char*)de < base + size) {
1111 next = (struct ext3_dir_entry_2 *) ((char *) de +
1112 le16_to_cpu(de->rec_len));
1113 if (de->inode && de->name_len) {
1114 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1116 memmove(to, de, rec_len);
1117 to->rec_len = cpu_to_le16(rec_len);
1119 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1126 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1127 struct buffer_head **bh,struct dx_frame *frame,
1128 struct dx_hash_info *hinfo, int *error)
1130 unsigned blocksize = dir->i_sb->s_blocksize;
1131 unsigned count, continued;
1132 struct buffer_head *bh2;
1135 struct dx_map_entry *map;
1136 char *data1 = (*bh)->b_data, *data2;
1138 struct ext3_dir_entry_2 *de = NULL, *de2;
1141 bh2 = ext3_append (handle, dir, &newblock, error);
1148 BUFFER_TRACE(*bh, "get_write_access");
1149 err = ext3_journal_get_write_access(handle, *bh);
1155 ext3_std_error(dir->i_sb, err);
1158 BUFFER_TRACE(frame->bh, "get_write_access");
1159 err = ext3_journal_get_write_access(handle, frame->bh);
1163 data2 = bh2->b_data;
1165 /* create map in the end of data2 block */
1166 map = (struct dx_map_entry *) (data2 + blocksize);
1167 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1168 blocksize, hinfo, map);
1170 split = count/2; // need to adjust to actual middle
1171 dx_sort_map (map, count);
1172 hash2 = map[split].hash;
1173 continued = hash2 == map[split - 1].hash;
1174 dxtrace(printk("Split block %i at %x, %i/%i\n",
1175 dx_get_block(frame->at), hash2, split, count-split));
1177 /* Fancy dance to stay within two buffers */
1178 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1179 de = dx_pack_dirents(data1,blocksize);
1180 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1181 de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1182 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1183 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1185 /* Which block gets the new entry? */
1186 if (hinfo->hash >= hash2)
1191 dx_insert_block (frame, hash2 + continued, newblock);
1192 err = ext3_journal_dirty_metadata (handle, bh2);
1195 err = ext3_journal_dirty_metadata (handle, frame->bh);
1199 dxtrace(dx_show_index ("frame", frame->entries));
1207 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1208 * it points to a directory entry which is guaranteed to be large
1209 * enough for new directory entry. If de is NULL, then
1210 * add_dirent_to_buf will attempt search the directory block for
1211 * space. It will return -ENOSPC if no space is available, and -EIO
1212 * and -EEXIST if directory entry already exists.
1214 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1215 * all other cases bh is released.
1217 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1218 struct inode *inode, struct ext3_dir_entry_2 *de,
1219 struct buffer_head * bh)
1221 struct inode *dir = dentry->d_parent->d_inode;
1222 const char *name = dentry->d_name.name;
1223 int namelen = dentry->d_name.len;
1224 unsigned long offset = 0;
1225 unsigned short reclen;
1226 int nlen, rlen, err;
1229 reclen = EXT3_DIR_REC_LEN(namelen);
1231 de = (struct ext3_dir_entry_2 *)bh->b_data;
1232 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1233 while ((char *) de <= top) {
1234 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1239 if (ext3_match (namelen, name, de)) {
1243 nlen = EXT3_DIR_REC_LEN(de->name_len);
1244 rlen = le16_to_cpu(de->rec_len);
1245 if ((de->inode? rlen - nlen: rlen) >= reclen)
1247 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1250 if ((char *) de > top)
1253 BUFFER_TRACE(bh, "get_write_access");
1254 err = ext3_journal_get_write_access(handle, bh);
1256 ext3_std_error(dir->i_sb, err);
1261 /* By now the buffer is marked for journaling */
1262 nlen = EXT3_DIR_REC_LEN(de->name_len);
1263 rlen = le16_to_cpu(de->rec_len);
1265 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1266 de1->rec_len = cpu_to_le16(rlen - nlen);
1267 de->rec_len = cpu_to_le16(nlen);
1270 de->file_type = EXT3_FT_UNKNOWN;
1272 de->inode = cpu_to_le32(inode->i_ino);
1273 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1276 de->name_len = namelen;
1277 memcpy (de->name, name, namelen);
1279 * XXX shouldn't update any times until successful
1280 * completion of syscall, but too many callers depend
1283 * XXX similarly, too many callers depend on
1284 * ext3_new_inode() setting the times, but error
1285 * recovery deletes the inode, so the worst that can
1286 * happen is that the times are slightly out of date
1287 * and/or different from the directory change time.
1289 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1290 ext3_update_dx_flag(dir);
1292 ext3_mark_inode_dirty(handle, dir);
1293 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1294 err = ext3_journal_dirty_metadata(handle, bh);
1296 ext3_std_error(dir->i_sb, err);
1301 #ifdef CONFIG_EXT3_INDEX
1303 * This converts a one block unindexed directory to a 3 block indexed
1304 * directory, and adds the dentry to the indexed directory.
1306 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1307 struct inode *inode, struct buffer_head *bh)
1309 struct inode *dir = dentry->d_parent->d_inode;
1310 const char *name = dentry->d_name.name;
1311 int namelen = dentry->d_name.len;
1312 struct buffer_head *bh2;
1313 struct dx_root *root;
1314 struct dx_frame frames[2], *frame;
1315 struct dx_entry *entries;
1316 struct ext3_dir_entry_2 *de, *de2;
1321 struct dx_hash_info hinfo;
1323 struct fake_dirent *fde;
1325 blocksize = dir->i_sb->s_blocksize;
1326 dxtrace(printk("Creating index\n"));
1327 retval = ext3_journal_get_write_access(handle, bh);
1329 ext3_std_error(dir->i_sb, retval);
1333 root = (struct dx_root *) bh->b_data;
1335 bh2 = ext3_append (handle, dir, &block, &retval);
1340 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1341 data1 = bh2->b_data;
1343 /* The 0th block becomes the root, move the dirents out */
1344 fde = &root->dotdot;
1345 de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1346 len = ((char *) root) + blocksize - (char *) de;
1347 memcpy (data1, de, len);
1348 de = (struct ext3_dir_entry_2 *) data1;
1350 while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1352 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1353 /* Initialize the root; the dot dirents already exist */
1354 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1355 de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1356 memset (&root->info, 0, sizeof(root->info));
1357 root->info.info_length = sizeof(root->info);
1358 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1359 entries = root->entries;
1360 dx_set_block (entries, 1);
1361 dx_set_count (entries, 1);
1362 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1364 /* Initialize as for dx_probe */
1365 hinfo.hash_version = root->info.hash_version;
1366 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1367 ext3fs_dirhash(name, namelen, &hinfo);
1369 frame->entries = entries;
1370 frame->at = entries;
1373 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1374 dx_release (frames);
1378 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1385 * adds a file entry to the specified directory, using the same
1386 * semantics as ext3_find_entry(). It returns NULL if it failed.
1388 * NOTE!! The inode part of 'de' is left at 0 - which means you
1389 * may not sleep between calling this and putting something into
1390 * the entry, as someone else might have used it while you slept.
1392 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1393 struct inode *inode)
1395 struct inode *dir = dentry->d_parent->d_inode;
1396 unsigned long offset;
1397 struct buffer_head * bh;
1398 struct ext3_dir_entry_2 *de;
1399 struct super_block * sb;
1401 #ifdef CONFIG_EXT3_INDEX
1408 blocksize = sb->s_blocksize;
1409 if (!dentry->d_name.len)
1411 #ifdef CONFIG_EXT3_INDEX
1413 retval = ext3_dx_add_entry(handle, dentry, inode);
1414 if (!retval || (retval != ERR_BAD_DX_DIR))
1416 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1418 ext3_mark_inode_dirty(handle, dir);
1421 blocks = dir->i_size >> sb->s_blocksize_bits;
1422 for (block = 0, offset = 0; block < blocks; block++) {
1423 bh = ext3_bread(handle, dir, block, 0, &retval);
1426 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1427 if (retval != -ENOSPC)
1430 #ifdef CONFIG_EXT3_INDEX
1431 if (blocks == 1 && !dx_fallback &&
1432 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1433 return make_indexed_dir(handle, dentry, inode, bh);
1437 bh = ext3_append(handle, dir, &block, &retval);
1440 de = (struct ext3_dir_entry_2 *) bh->b_data;
1442 de->rec_len = cpu_to_le16(blocksize);
1443 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1446 #ifdef CONFIG_EXT3_INDEX
1448 * Returns 0 for success, or a negative error value
1450 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1451 struct inode *inode)
1453 struct dx_frame frames[2], *frame;
1454 struct dx_entry *entries, *at;
1455 struct dx_hash_info hinfo;
1456 struct buffer_head * bh;
1457 struct inode *dir = dentry->d_parent->d_inode;
1458 struct super_block * sb = dir->i_sb;
1459 struct ext3_dir_entry_2 *de;
1462 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1465 entries = frame->entries;
1468 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1471 BUFFER_TRACE(bh, "get_write_access");
1472 err = ext3_journal_get_write_access(handle, bh);
1476 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1477 if (err != -ENOSPC) {
1482 /* Block full, should compress but for now just split */
1483 dxtrace(printk("using %u of %u node entries\n",
1484 dx_get_count(entries), dx_get_limit(entries)));
1485 /* Need to split index? */
1486 if (dx_get_count(entries) == dx_get_limit(entries)) {
1488 unsigned icount = dx_get_count(entries);
1489 int levels = frame - frames;
1490 struct dx_entry *entries2;
1491 struct dx_node *node2;
1492 struct buffer_head *bh2;
1494 if (levels && (dx_get_count(frames->entries) ==
1495 dx_get_limit(frames->entries))) {
1496 ext3_warning(sb, __FUNCTION__,
1497 "Directory index full!");
1501 bh2 = ext3_append (handle, dir, &newblock, &err);
1504 node2 = (struct dx_node *)(bh2->b_data);
1505 entries2 = node2->entries;
1506 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1507 node2->fake.inode = 0;
1508 BUFFER_TRACE(frame->bh, "get_write_access");
1509 err = ext3_journal_get_write_access(handle, frame->bh);
1513 unsigned icount1 = icount/2, icount2 = icount - icount1;
1514 unsigned hash2 = dx_get_hash(entries + icount1);
1515 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1517 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1518 err = ext3_journal_get_write_access(handle,
1523 memcpy ((char *) entries2, (char *) (entries + icount1),
1524 icount2 * sizeof(struct dx_entry));
1525 dx_set_count (entries, icount1);
1526 dx_set_count (entries2, icount2);
1527 dx_set_limit (entries2, dx_node_limit(dir));
1529 /* Which index block gets the new entry? */
1530 if (at - entries >= icount1) {
1531 frame->at = at = at - entries - icount1 + entries2;
1532 frame->entries = entries = entries2;
1533 swap(frame->bh, bh2);
1535 dx_insert_block (frames + 0, hash2, newblock);
1536 dxtrace(dx_show_index ("node", frames[1].entries));
1537 dxtrace(dx_show_index ("node",
1538 ((struct dx_node *) bh2->b_data)->entries));
1539 err = ext3_journal_dirty_metadata(handle, bh2);
1544 dxtrace(printk("Creating second level index...\n"));
1545 memcpy((char *) entries2, (char *) entries,
1546 icount * sizeof(struct dx_entry));
1547 dx_set_limit(entries2, dx_node_limit(dir));
1550 dx_set_count(entries, 1);
1551 dx_set_block(entries + 0, newblock);
1552 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1554 /* Add new access path frame */
1556 frame->at = at = at - entries + entries2;
1557 frame->entries = entries = entries2;
1559 err = ext3_journal_get_write_access(handle,
1564 ext3_journal_dirty_metadata(handle, frames[0].bh);
1566 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1569 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1574 ext3_std_error(dir->i_sb, err);
1584 * ext3_delete_entry deletes a directory entry by merging it with the
1587 static int ext3_delete_entry (handle_t *handle,
1589 struct ext3_dir_entry_2 * de_del,
1590 struct buffer_head * bh)
1592 struct ext3_dir_entry_2 * de, * pde;
1597 de = (struct ext3_dir_entry_2 *) bh->b_data;
1598 while (i < bh->b_size) {
1599 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1602 BUFFER_TRACE(bh, "get_write_access");
1603 ext3_journal_get_write_access(handle, bh);
1606 cpu_to_le16(le16_to_cpu(pde->rec_len) +
1607 le16_to_cpu(de->rec_len));
1611 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1612 ext3_journal_dirty_metadata(handle, bh);
1615 i += le16_to_cpu(de->rec_len);
1617 de = (struct ext3_dir_entry_2 *)
1618 ((char *) de + le16_to_cpu(de->rec_len));
1624 * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1625 * do not perform it in these functions. We perform it at the call site,
1628 static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1633 static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1638 static int ext3_add_nondir(handle_t *handle,
1639 struct dentry *dentry, struct inode *inode)
1641 int err = ext3_add_entry(handle, dentry, inode);
1643 ext3_mark_inode_dirty(handle, inode);
1644 d_instantiate(dentry, inode);
1647 ext3_dec_count(handle, inode);
1653 * By the time this is called, we already have created
1654 * the directory cache entry for the new file, but it
1655 * is so far negative - it has no inode.
1657 * If the create succeeds, we fill in the inode information
1658 * with d_instantiate().
1660 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1661 struct nameidata *nd)
1664 struct inode * inode;
1665 int err, retries = 0;
1668 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1669 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1670 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1672 return PTR_ERR(handle);
1674 if (IS_DIRSYNC(dir))
1677 inode = ext3_new_inode (handle, dir, mode);
1678 err = PTR_ERR(inode);
1679 if (!IS_ERR(inode)) {
1680 inode->i_op = &ext3_file_inode_operations;
1681 inode->i_fop = &ext3_file_operations;
1682 ext3_set_aops(inode);
1683 err = ext3_add_nondir(handle, dentry, inode);
1685 ext3_journal_stop(handle);
1686 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1691 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1692 int mode, dev_t rdev)
1695 struct inode *inode;
1696 int err, retries = 0;
1698 if (!new_valid_dev(rdev))
1702 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1703 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1704 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1706 return PTR_ERR(handle);
1708 if (IS_DIRSYNC(dir))
1711 inode = ext3_new_inode (handle, dir, mode);
1712 err = PTR_ERR(inode);
1713 if (!IS_ERR(inode)) {
1714 init_special_inode(inode, inode->i_mode, rdev);
1715 #ifdef CONFIG_EXT3_FS_XATTR
1716 inode->i_op = &ext3_special_inode_operations;
1718 err = ext3_add_nondir(handle, dentry, inode);
1720 ext3_journal_stop(handle);
1721 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1726 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1729 struct inode * inode;
1730 struct buffer_head * dir_block;
1731 struct ext3_dir_entry_2 * de;
1732 int err, retries = 0;
1734 if (dir->i_nlink >= EXT3_LINK_MAX)
1738 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1739 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1740 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1742 return PTR_ERR(handle);
1744 if (IS_DIRSYNC(dir))
1747 inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1748 err = PTR_ERR(inode);
1752 inode->i_op = &ext3_dir_inode_operations;
1753 inode->i_fop = &ext3_dir_operations;
1754 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1755 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1757 drop_nlink(inode); /* is this nlink == 0? */
1758 ext3_mark_inode_dirty(handle, inode);
1762 BUFFER_TRACE(dir_block, "get_write_access");
1763 ext3_journal_get_write_access(handle, dir_block);
1764 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1765 de->inode = cpu_to_le32(inode->i_ino);
1767 de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1768 strcpy (de->name, ".");
1769 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1770 de = (struct ext3_dir_entry_2 *)
1771 ((char *) de + le16_to_cpu(de->rec_len));
1772 de->inode = cpu_to_le32(dir->i_ino);
1773 de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1775 strcpy (de->name, "..");
1776 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1778 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1779 ext3_journal_dirty_metadata(handle, dir_block);
1781 ext3_mark_inode_dirty(handle, inode);
1782 err = ext3_add_entry (handle, dentry, inode);
1785 ext3_mark_inode_dirty(handle, inode);
1790 ext3_update_dx_flag(dir);
1791 ext3_mark_inode_dirty(handle, dir);
1792 d_instantiate(dentry, inode);
1794 ext3_journal_stop(handle);
1795 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1801 * routine to check that the specified directory is empty (for rmdir)
1803 static int empty_dir (struct inode * inode)
1805 unsigned long offset;
1806 struct buffer_head * bh;
1807 struct ext3_dir_entry_2 * de, * de1;
1808 struct super_block * sb;
1812 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1813 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1815 ext3_error(inode->i_sb, __FUNCTION__,
1816 "error %d reading directory #%lu offset 0",
1819 ext3_warning(inode->i_sb, __FUNCTION__,
1820 "bad directory (dir #%lu) - no data block",
1824 de = (struct ext3_dir_entry_2 *) bh->b_data;
1825 de1 = (struct ext3_dir_entry_2 *)
1826 ((char *) de + le16_to_cpu(de->rec_len));
1827 if (le32_to_cpu(de->inode) != inode->i_ino ||
1828 !le32_to_cpu(de1->inode) ||
1829 strcmp (".", de->name) ||
1830 strcmp ("..", de1->name)) {
1831 ext3_warning (inode->i_sb, "empty_dir",
1832 "bad directory (dir #%lu) - no `.' or `..'",
1837 offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1838 de = (struct ext3_dir_entry_2 *)
1839 ((char *) de1 + le16_to_cpu(de1->rec_len));
1840 while (offset < inode->i_size ) {
1842 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1845 bh = ext3_bread (NULL, inode,
1846 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1849 ext3_error(sb, __FUNCTION__,
1850 "error %d reading directory"
1852 err, inode->i_ino, offset);
1853 offset += sb->s_blocksize;
1856 de = (struct ext3_dir_entry_2 *) bh->b_data;
1858 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1859 de = (struct ext3_dir_entry_2 *)(bh->b_data +
1861 offset = (offset | (sb->s_blocksize - 1)) + 1;
1864 if (le32_to_cpu(de->inode)) {
1868 offset += le16_to_cpu(de->rec_len);
1869 de = (struct ext3_dir_entry_2 *)
1870 ((char *) de + le16_to_cpu(de->rec_len));
1876 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1877 * such inodes, starting at the superblock, in case we crash before the
1878 * file is closed/deleted, or in case the inode truncate spans multiple
1879 * transactions and the last transaction is not recovered after a crash.
1881 * At filesystem recovery time, we walk this list deleting unlinked
1882 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1884 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1886 struct super_block *sb = inode->i_sb;
1887 struct ext3_iloc iloc;
1891 if (!list_empty(&EXT3_I(inode)->i_orphan))
1894 /* Orphan handling is only valid for files with data blocks
1895 * being truncated, or files being unlinked. */
1897 /* @@@ FIXME: Observation from aviro:
1898 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1899 * here (on lock_super()), so race with ext3_link() which might bump
1900 * ->i_nlink. For, say it, character device. Not a regular file,
1901 * not a directory, not a symlink and ->i_nlink > 0.
1903 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1904 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1906 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1907 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1911 err = ext3_reserve_inode_write(handle, inode, &iloc);
1915 /* Insert this inode at the head of the on-disk orphan list... */
1916 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1917 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1918 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1919 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1923 /* Only add to the head of the in-memory list if all the
1924 * previous operations succeeded. If the orphan_add is going to
1925 * fail (possibly taking the journal offline), we can't risk
1926 * leaving the inode on the orphan list: stray orphan-list
1927 * entries can cause panics at unmount time.
1929 * This is safe: on error we're going to ignore the orphan list
1930 * anyway on the next recovery. */
1932 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1934 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1935 jbd_debug(4, "orphan inode %lu will point to %d\n",
1936 inode->i_ino, NEXT_ORPHAN(inode));
1939 ext3_std_error(inode->i_sb, err);
1944 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1945 * of such inodes stored on disk, because it is finally being cleaned up.
1947 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1949 struct list_head *prev;
1950 struct ext3_inode_info *ei = EXT3_I(inode);
1951 struct ext3_sb_info *sbi;
1952 unsigned long ino_next;
1953 struct ext3_iloc iloc;
1956 lock_super(inode->i_sb);
1957 if (list_empty(&ei->i_orphan)) {
1958 unlock_super(inode->i_sb);
1962 ino_next = NEXT_ORPHAN(inode);
1963 prev = ei->i_orphan.prev;
1964 sbi = EXT3_SB(inode->i_sb);
1966 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1968 list_del_init(&ei->i_orphan);
1970 /* If we're on an error path, we may not have a valid
1971 * transaction handle with which to update the orphan list on
1972 * disk, but we still need to remove the inode from the linked
1973 * list in memory. */
1977 err = ext3_reserve_inode_write(handle, inode, &iloc);
1981 if (prev == &sbi->s_orphan) {
1982 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1983 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1984 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1987 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1988 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1990 struct ext3_iloc iloc2;
1991 struct inode *i_prev =
1992 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1994 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1995 i_prev->i_ino, ino_next);
1996 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1999 NEXT_ORPHAN(i_prev) = ino_next;
2000 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2004 NEXT_ORPHAN(inode) = 0;
2005 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2008 ext3_std_error(inode->i_sb, err);
2010 unlock_super(inode->i_sb);
2018 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2021 struct inode * inode;
2022 struct buffer_head * bh;
2023 struct ext3_dir_entry_2 * de;
2026 /* Initialize quotas before so that eventual writes go in
2027 * separate transaction */
2028 DQUOT_INIT(dentry->d_inode);
2029 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2031 return PTR_ERR(handle);
2034 bh = ext3_find_entry (dentry, &de);
2038 if (IS_DIRSYNC(dir))
2041 inode = dentry->d_inode;
2044 if (le32_to_cpu(de->inode) != inode->i_ino)
2047 retval = -ENOTEMPTY;
2048 if (!empty_dir (inode))
2051 retval = ext3_delete_entry(handle, dir, de, bh);
2054 if (inode->i_nlink != 2)
2055 ext3_warning (inode->i_sb, "ext3_rmdir",
2056 "empty directory has nlink!=2 (%d)",
2060 /* There's no need to set i_disksize: the fact that i_nlink is
2061 * zero will ensure that the right thing happens during any
2064 ext3_orphan_add(handle, inode);
2065 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2066 ext3_mark_inode_dirty(handle, inode);
2068 ext3_update_dx_flag(dir);
2069 ext3_mark_inode_dirty(handle, dir);
2072 ext3_journal_stop(handle);
2077 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2080 struct inode * inode;
2081 struct buffer_head * bh;
2082 struct ext3_dir_entry_2 * de;
2085 /* Initialize quotas before so that eventual writes go
2086 * in separate transaction */
2087 DQUOT_INIT(dentry->d_inode);
2088 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2090 return PTR_ERR(handle);
2092 if (IS_DIRSYNC(dir))
2096 bh = ext3_find_entry (dentry, &de);
2100 inode = dentry->d_inode;
2103 if (le32_to_cpu(de->inode) != inode->i_ino)
2106 if (!inode->i_nlink) {
2107 ext3_warning (inode->i_sb, "ext3_unlink",
2108 "Deleting nonexistent file (%lu), %d",
2109 inode->i_ino, inode->i_nlink);
2112 retval = ext3_delete_entry(handle, dir, de, bh);
2115 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2116 ext3_update_dx_flag(dir);
2117 ext3_mark_inode_dirty(handle, dir);
2119 if (!inode->i_nlink)
2120 ext3_orphan_add(handle, inode);
2121 inode->i_ctime = dir->i_ctime;
2122 ext3_mark_inode_dirty(handle, inode);
2126 ext3_journal_stop(handle);
2131 static int ext3_symlink (struct inode * dir,
2132 struct dentry *dentry, const char * symname)
2135 struct inode * inode;
2136 int l, err, retries = 0;
2138 l = strlen(symname)+1;
2139 if (l > dir->i_sb->s_blocksize)
2140 return -ENAMETOOLONG;
2143 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2144 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2145 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2147 return PTR_ERR(handle);
2149 if (IS_DIRSYNC(dir))
2152 inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2153 err = PTR_ERR(inode);
2157 if (l > sizeof (EXT3_I(inode)->i_data)) {
2158 inode->i_op = &ext3_symlink_inode_operations;
2159 ext3_set_aops(inode);
2161 * page_symlink() calls into ext3_prepare/commit_write.
2162 * We have a transaction open. All is sweetness. It also sets
2163 * i_size in generic_commit_write().
2165 err = __page_symlink(inode, symname, l,
2166 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2168 ext3_dec_count(handle, inode);
2169 ext3_mark_inode_dirty(handle, inode);
2174 inode->i_op = &ext3_fast_symlink_inode_operations;
2175 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2176 inode->i_size = l-1;
2178 EXT3_I(inode)->i_disksize = inode->i_size;
2179 err = ext3_add_nondir(handle, dentry, inode);
2181 ext3_journal_stop(handle);
2182 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2187 static int ext3_link (struct dentry * old_dentry,
2188 struct inode * dir, struct dentry *dentry)
2191 struct inode *inode = old_dentry->d_inode;
2192 int err, retries = 0;
2194 if (inode->i_nlink >= EXT3_LINK_MAX)
2198 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2199 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2201 return PTR_ERR(handle);
2203 if (IS_DIRSYNC(dir))
2206 inode->i_ctime = CURRENT_TIME_SEC;
2207 ext3_inc_count(handle, inode);
2208 atomic_inc(&inode->i_count);
2210 err = ext3_add_nondir(handle, dentry, inode);
2211 ext3_journal_stop(handle);
2212 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2217 #define PARENT_INO(buffer) \
2218 ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2219 le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2222 * Anybody can rename anything with this: the permission checks are left to the
2223 * higher-level routines.
2225 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2226 struct inode * new_dir,struct dentry *new_dentry)
2229 struct inode * old_inode, * new_inode;
2230 struct buffer_head * old_bh, * new_bh, * dir_bh;
2231 struct ext3_dir_entry_2 * old_de, * new_de;
2234 old_bh = new_bh = dir_bh = NULL;
2236 /* Initialize quotas before so that eventual writes go
2237 * in separate transaction */
2238 if (new_dentry->d_inode)
2239 DQUOT_INIT(new_dentry->d_inode);
2240 handle = ext3_journal_start(old_dir, 2 *
2241 EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2242 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2244 return PTR_ERR(handle);
2246 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2249 old_bh = ext3_find_entry (old_dentry, &old_de);
2251 * Check for inode number is _not_ due to possible IO errors.
2252 * We might rmdir the source, keep it as pwd of some process
2253 * and merrily kill the link to whatever was created under the
2254 * same name. Goodbye sticky bit ;-<
2256 old_inode = old_dentry->d_inode;
2258 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2261 new_inode = new_dentry->d_inode;
2262 new_bh = ext3_find_entry (new_dentry, &new_de);
2269 if (S_ISDIR(old_inode->i_mode)) {
2271 retval = -ENOTEMPTY;
2272 if (!empty_dir (new_inode))
2276 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2279 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2282 if (!new_inode && new_dir!=old_dir &&
2283 new_dir->i_nlink >= EXT3_LINK_MAX)
2287 retval = ext3_add_entry (handle, new_dentry, old_inode);
2291 BUFFER_TRACE(new_bh, "get write access");
2292 ext3_journal_get_write_access(handle, new_bh);
2293 new_de->inode = cpu_to_le32(old_inode->i_ino);
2294 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2295 EXT3_FEATURE_INCOMPAT_FILETYPE))
2296 new_de->file_type = old_de->file_type;
2297 new_dir->i_version++;
2298 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2299 ext3_journal_dirty_metadata(handle, new_bh);
2305 * Like most other Unix systems, set the ctime for inodes on a
2308 old_inode->i_ctime = CURRENT_TIME_SEC;
2309 ext3_mark_inode_dirty(handle, old_inode);
2314 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2315 old_de->name_len != old_dentry->d_name.len ||
2316 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2317 (retval = ext3_delete_entry(handle, old_dir,
2318 old_de, old_bh)) == -ENOENT) {
2319 /* old_de could have moved from under us during htree split, so
2320 * make sure that we are deleting the right entry. We might
2321 * also be pointing to a stale entry in the unused part of
2322 * old_bh so just checking inum and the name isn't enough. */
2323 struct buffer_head *old_bh2;
2324 struct ext3_dir_entry_2 *old_de2;
2326 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2328 retval = ext3_delete_entry(handle, old_dir,
2334 ext3_warning(old_dir->i_sb, "ext3_rename",
2335 "Deleting old file (%lu), %d, error=%d",
2336 old_dir->i_ino, old_dir->i_nlink, retval);
2340 drop_nlink(new_inode);
2341 new_inode->i_ctime = CURRENT_TIME_SEC;
2343 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2344 ext3_update_dx_flag(old_dir);
2346 BUFFER_TRACE(dir_bh, "get_write_access");
2347 ext3_journal_get_write_access(handle, dir_bh);
2348 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2349 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2350 ext3_journal_dirty_metadata(handle, dir_bh);
2351 drop_nlink(old_dir);
2353 drop_nlink(new_inode);
2356 ext3_update_dx_flag(new_dir);
2357 ext3_mark_inode_dirty(handle, new_dir);
2360 ext3_mark_inode_dirty(handle, old_dir);
2362 ext3_mark_inode_dirty(handle, new_inode);
2363 if (!new_inode->i_nlink)
2364 ext3_orphan_add(handle, new_inode);
2372 ext3_journal_stop(handle);
2377 * directories can handle most operations...
2379 struct inode_operations ext3_dir_inode_operations = {
2380 .create = ext3_create,
2381 .lookup = ext3_lookup,
2383 .unlink = ext3_unlink,
2384 .symlink = ext3_symlink,
2385 .mkdir = ext3_mkdir,
2386 .rmdir = ext3_rmdir,
2387 .mknod = ext3_mknod,
2388 .rename = ext3_rename,
2389 .setattr = ext3_setattr,
2390 #ifdef CONFIG_EXT3_FS_XATTR
2391 .setxattr = generic_setxattr,
2392 .getxattr = generic_getxattr,
2393 .listxattr = ext3_listxattr,
2394 .removexattr = generic_removexattr,
2396 .permission = ext3_permission,
2397 .sync_flags = ext3_sync_flags,
2400 struct inode_operations ext3_special_inode_operations = {
2401 .setattr = ext3_setattr,
2402 #ifdef CONFIG_EXT3_FS_XATTR
2403 .setxattr = generic_setxattr,
2404 .getxattr = generic_getxattr,
2405 .listxattr = ext3_listxattr,
2406 .removexattr = generic_removexattr,
2408 .permission = ext3_permission,
2409 .sync_flags = ext3_sync_flags,