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>
45 * define how far ahead to read directories while searching them.
47 #define NAMEI_RA_CHUNKS 2
48 #define NAMEI_RA_BLOCKS 4
49 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
52 static struct buffer_head *ext3_append(handle_t *handle,
56 struct buffer_head *bh;
58 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
60 if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
61 inode->i_size += inode->i_sb->s_blocksize;
62 EXT3_I(inode)->i_disksize = inode->i_size;
63 ext3_journal_get_write_access(handle,bh);
69 #define assert(test) J_ASSERT(test)
73 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
77 #define dxtrace(command) command
79 #define dxtrace(command)
103 * dx_root_info is laid out so that if it should somehow get overlaid by a
104 * dirent the two low bits of the hash version will be zero. Therefore, the
105 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
110 struct fake_dirent dot;
112 struct fake_dirent dotdot;
116 __le32 reserved_zero;
118 u8 info_length; /* 8 */
123 struct dx_entry entries[0];
128 struct fake_dirent fake;
129 struct dx_entry entries[0];
135 struct buffer_head *bh;
136 struct dx_entry *entries;
146 #ifdef CONFIG_EXT3_INDEX
147 static inline unsigned dx_get_block (struct dx_entry *entry);
148 static void dx_set_block (struct dx_entry *entry, unsigned value);
149 static inline unsigned dx_get_hash (struct dx_entry *entry);
150 static void dx_set_hash (struct dx_entry *entry, unsigned value);
151 static unsigned dx_get_count (struct dx_entry *entries);
152 static unsigned dx_get_limit (struct dx_entry *entries);
153 static void dx_set_count (struct dx_entry *entries, unsigned value);
154 static void dx_set_limit (struct dx_entry *entries, unsigned value);
155 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
156 static unsigned dx_node_limit (struct inode *dir);
157 static struct dx_frame *dx_probe(struct dentry *dentry,
159 struct dx_hash_info *hinfo,
160 struct dx_frame *frame,
162 static void dx_release (struct dx_frame *frames);
163 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
164 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
165 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
166 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
167 struct dx_map_entry *offsets, int count);
168 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
169 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
170 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
171 struct dx_frame *frame,
172 struct dx_frame *frames,
174 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
175 struct ext3_dir_entry_2 **res_dir, int *err);
176 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
177 struct inode *inode);
180 * Future: use high four bits of block for coalesce-on-delete flags
181 * Mask them off for now.
184 static inline unsigned dx_get_block (struct dx_entry *entry)
186 return le32_to_cpu(entry->block) & 0x00ffffff;
189 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
191 entry->block = cpu_to_le32(value);
194 static inline unsigned dx_get_hash (struct dx_entry *entry)
196 return le32_to_cpu(entry->hash);
199 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
201 entry->hash = cpu_to_le32(value);
204 static inline unsigned dx_get_count (struct dx_entry *entries)
206 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
209 static inline unsigned dx_get_limit (struct dx_entry *entries)
211 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
214 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
216 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
219 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
221 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
224 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
226 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
227 EXT3_DIR_REC_LEN(2) - infosize;
228 return 0? 20: entry_space / sizeof(struct dx_entry);
231 static inline unsigned dx_node_limit (struct inode *dir)
233 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
234 return 0? 22: entry_space / sizeof(struct dx_entry);
241 static void dx_show_index (char * label, struct dx_entry *entries)
243 int i, n = dx_get_count (entries);
244 printk("%s index ", label);
245 for (i = 0; i < n; i++)
247 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
259 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
260 int size, int show_names)
262 unsigned names = 0, space = 0;
263 char *base = (char *) de;
264 struct dx_hash_info h = *hinfo;
267 while ((char *) de < base + size)
273 int len = de->name_len;
274 char *name = de->name;
275 while (len--) printk("%c", *name++);
276 ext3fs_dirhash(de->name, de->name_len, &h);
277 printk(":%x.%u ", h.hash,
278 ((char *) de - base));
280 space += EXT3_DIR_REC_LEN(de->name_len);
283 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
285 printk("(%i)\n", names);
286 return (struct stats) { names, space, 1 };
289 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
290 struct dx_entry *entries, int levels)
292 unsigned blocksize = dir->i_sb->s_blocksize;
293 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
295 struct buffer_head *bh;
297 printk("%i indexed blocks...\n", count);
298 for (i = 0; i < count; i++, entries++)
300 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
301 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
303 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
304 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
306 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
307 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
308 names += stats.names;
309 space += stats.space;
310 bcount += stats.bcount;
314 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
315 names, space/bcount,(space/bcount)*100/blocksize);
316 return (struct stats) { names, space, bcount};
318 #endif /* DX_DEBUG */
321 * Probe for a directory leaf block to search.
323 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
324 * error in the directory index, and the caller should fall back to
325 * searching the directory normally. The callers of dx_probe **MUST**
326 * check for this error code, and make sure it never gets reflected
329 static struct dx_frame *
330 dx_probe(struct dentry *dentry, struct inode *dir,
331 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
333 unsigned count, indirect;
334 struct dx_entry *at, *entries, *p, *q, *m;
335 struct dx_root *root;
336 struct buffer_head *bh;
337 struct dx_frame *frame = frame_in;
342 dir = dentry->d_parent->d_inode;
343 if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
345 root = (struct dx_root *) bh->b_data;
346 if (root->info.hash_version != DX_HASH_TEA &&
347 root->info.hash_version != DX_HASH_HALF_MD4 &&
348 root->info.hash_version != DX_HASH_LEGACY) {
349 ext3_warning(dir->i_sb, __FUNCTION__,
350 "Unrecognised inode hash code %d",
351 root->info.hash_version);
353 *err = ERR_BAD_DX_DIR;
356 hinfo->hash_version = root->info.hash_version;
357 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
359 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
362 if (root->info.unused_flags & 1) {
363 ext3_warning(dir->i_sb, __FUNCTION__,
364 "Unimplemented inode hash flags: %#06x",
365 root->info.unused_flags);
367 *err = ERR_BAD_DX_DIR;
371 if ((indirect = root->info.indirect_levels) > 1) {
372 ext3_warning(dir->i_sb, __FUNCTION__,
373 "Unimplemented inode hash depth: %#06x",
374 root->info.indirect_levels);
376 *err = ERR_BAD_DX_DIR;
380 entries = (struct dx_entry *) (((char *)&root->info) +
381 root->info.info_length);
383 if (dx_get_limit(entries) != dx_root_limit(dir,
384 root->info.info_length)) {
385 ext3_warning(dir->i_sb, __FUNCTION__,
386 "dx entry: limit != root limit");
388 *err = ERR_BAD_DX_DIR;
392 dxtrace (printk("Look up %x", hash));
395 count = dx_get_count(entries);
396 if (!count || count > dx_get_limit(entries)) {
397 ext3_warning(dir->i_sb, __FUNCTION__,
398 "dx entry: no count or count > limit");
400 *err = ERR_BAD_DX_DIR;
405 q = entries + count - 1;
409 dxtrace(printk("."));
410 if (dx_get_hash(m) > hash)
416 if (0) // linear search cross check
418 unsigned n = count - 1;
422 dxtrace(printk(","));
423 if (dx_get_hash(++at) > hash)
429 assert (at == p - 1);
433 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
435 frame->entries = entries;
437 if (!indirect--) return frame;
438 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
440 at = entries = ((struct dx_node *) bh->b_data)->entries;
441 if (dx_get_limit(entries) != dx_node_limit (dir)) {
442 ext3_warning(dir->i_sb, __FUNCTION__,
443 "dx entry: limit != node limit");
445 *err = ERR_BAD_DX_DIR;
452 while (frame >= frame_in) {
457 if (*err == ERR_BAD_DX_DIR)
458 ext3_warning(dir->i_sb, __FUNCTION__,
459 "Corrupt dir inode %ld, running e2fsck is "
460 "recommended.", dir->i_ino);
464 static void dx_release (struct dx_frame *frames)
466 if (frames[0].bh == NULL)
469 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
470 brelse(frames[1].bh);
471 brelse(frames[0].bh);
475 * This function increments the frame pointer to search the next leaf
476 * block, and reads in the necessary intervening nodes if the search
477 * should be necessary. Whether or not the search is necessary is
478 * controlled by the hash parameter. If the hash value is even, then
479 * the search is only continued if the next block starts with that
480 * hash value. This is used if we are searching for a specific file.
482 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
484 * This function returns 1 if the caller should continue to search,
485 * or 0 if it should not. If there is an error reading one of the
486 * index blocks, it will a negative error code.
488 * If start_hash is non-null, it will be filled in with the starting
489 * hash of the next page.
491 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
492 struct dx_frame *frame,
493 struct dx_frame *frames,
497 struct buffer_head *bh;
498 int err, num_frames = 0;
503 * Find the next leaf page by incrementing the frame pointer.
504 * If we run out of entries in the interior node, loop around and
505 * increment pointer in the parent node. When we break out of
506 * this loop, num_frames indicates the number of interior
507 * nodes need to be read.
510 if (++(p->at) < p->entries + dx_get_count(p->entries))
519 * If the hash is 1, then continue only if the next page has a
520 * continuation hash of any value. This is used for readdir
521 * handling. Otherwise, check to see if the hash matches the
522 * desired contiuation hash. If it doesn't, return since
523 * there's no point to read in the successive index pages.
525 bhash = dx_get_hash(p->at);
528 if ((hash & 1) == 0) {
529 if ((bhash & ~1) != hash)
533 * If the hash is HASH_NB_ALWAYS, we always go to the next
534 * block so no check is necessary
536 while (num_frames--) {
537 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
539 return err; /* Failure */
543 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
550 * p is at least 6 bytes before the end of page
552 static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
554 return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
558 * This function fills a red-black tree with information from a
559 * directory block. It returns the number directory entries loaded
560 * into the tree. If there is an error it is returned in err.
562 static int htree_dirblock_to_tree(struct file *dir_file,
563 struct inode *dir, int block,
564 struct dx_hash_info *hinfo,
565 __u32 start_hash, __u32 start_minor_hash)
567 struct buffer_head *bh;
568 struct ext3_dir_entry_2 *de, *top;
571 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
572 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
575 de = (struct ext3_dir_entry_2 *) bh->b_data;
576 top = (struct ext3_dir_entry_2 *) ((char *) de +
577 dir->i_sb->s_blocksize -
578 EXT3_DIR_REC_LEN(0));
579 for (; de < top; de = ext3_next_entry(de)) {
580 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
581 (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
582 +((char *)de - bh->b_data))) {
583 /* On error, skip the f_pos to the next block. */
584 dir_file->f_pos = (dir_file->f_pos |
585 (dir->i_sb->s_blocksize - 1)) + 1;
589 ext3fs_dirhash(de->name, de->name_len, hinfo);
590 if ((hinfo->hash < start_hash) ||
591 ((hinfo->hash == start_hash) &&
592 (hinfo->minor_hash < start_minor_hash)))
596 if ((err = ext3_htree_store_dirent(dir_file,
597 hinfo->hash, hinfo->minor_hash, de)) != 0) {
609 * This function fills a red-black tree with information from a
610 * directory. We start scanning the directory in hash order, starting
611 * at start_hash and start_minor_hash.
613 * This function returns the number of entries inserted into the tree,
614 * or a negative error code.
616 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
617 __u32 start_minor_hash, __u32 *next_hash)
619 struct dx_hash_info hinfo;
620 struct ext3_dir_entry_2 *de;
621 struct dx_frame frames[2], *frame;
628 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
630 dir = dir_file->f_path.dentry->d_inode;
631 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
632 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
633 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
634 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
635 start_hash, start_minor_hash);
639 hinfo.hash = start_hash;
640 hinfo.minor_hash = 0;
641 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
645 /* Add '.' and '..' from the htree header */
646 if (!start_hash && !start_minor_hash) {
647 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
648 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
652 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
653 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
654 de = ext3_next_entry(de);
655 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
661 block = dx_get_block(frame->at);
662 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
663 start_hash, start_minor_hash);
670 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
671 frame, frames, &hashval);
672 *next_hash = hashval;
678 * Stop if: (a) there are no more entries, or
679 * (b) we have inserted at least one entry and the
680 * next hash value is not a continuation
683 (count && ((hashval & 1) == 0)))
687 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
697 * Directory block splitting, compacting
700 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
701 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
704 char *base = (char *) de;
705 struct dx_hash_info h = *hinfo;
707 while ((char *) de < base + size)
709 if (de->name_len && de->inode) {
710 ext3fs_dirhash(de->name, de->name_len, &h);
712 map_tail->hash = h.hash;
713 map_tail->offs = (u32) ((char *) de - base);
717 /* XXX: do we need to check rec_len == 0 case? -Chris */
718 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
723 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
725 struct dx_map_entry *p, *q, *top = map + count - 1;
727 /* Combsort until bubble sort doesn't suck */
731 if (count - 9 < 2) /* 9, 10 -> 11 */
733 for (p = top, q = p - count; q >= map; p--, q--)
734 if (p->hash < q->hash)
737 /* Garden variety bubble sort */
743 if (q[1].hash >= q[0].hash)
751 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
753 struct dx_entry *entries = frame->entries;
754 struct dx_entry *old = frame->at, *new = old + 1;
755 int count = dx_get_count(entries);
757 assert(count < dx_get_limit(entries));
758 assert(old < entries + count);
759 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
760 dx_set_hash(new, hash);
761 dx_set_block(new, block);
762 dx_set_count(entries, count + 1);
767 static void ext3_update_dx_flag(struct inode *inode)
769 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
770 EXT3_FEATURE_COMPAT_DIR_INDEX))
771 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
775 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
777 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
778 * `de != NULL' is guaranteed by caller.
780 static inline int ext3_match (int len, const char * const name,
781 struct ext3_dir_entry_2 * de)
783 if (len != de->name_len)
787 return !memcmp(name, de->name, len);
791 * Returns 0 if not found, -1 on failure, and 1 on success
793 static inline int search_dirblock(struct buffer_head * bh,
795 struct dentry *dentry,
796 unsigned long offset,
797 struct ext3_dir_entry_2 ** res_dir)
799 struct ext3_dir_entry_2 * de;
802 const char *name = dentry->d_name.name;
803 int namelen = dentry->d_name.len;
805 de = (struct ext3_dir_entry_2 *) bh->b_data;
806 dlimit = bh->b_data + dir->i_sb->s_blocksize;
807 while ((char *) de < dlimit) {
808 /* this code is executed quadratically often */
809 /* do minimal checking `by hand' */
811 if ((char *) de + namelen <= dlimit &&
812 ext3_match (namelen, name, de)) {
813 /* found a match - just to be sure, do a full check */
814 if (!ext3_check_dir_entry("ext3_find_entry",
815 dir, de, bh, offset))
820 /* prevent looping on a bad block */
821 de_len = le16_to_cpu(de->rec_len);
825 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
834 * finds an entry in the specified directory with the wanted name. It
835 * returns the cache buffer in which the entry was found, and the entry
836 * itself (as a parameter - res_dir). It does NOT read the inode of the
837 * entry - you'll have to do that yourself if you want to.
839 * The returned buffer_head has ->b_count elevated. The caller is expected
840 * to brelse() it when appropriate.
842 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
843 struct ext3_dir_entry_2 ** res_dir)
845 struct super_block * sb;
846 struct buffer_head * bh_use[NAMEI_RA_SIZE];
847 struct buffer_head * bh, *ret = NULL;
848 unsigned long start, block, b;
849 int ra_max = 0; /* Number of bh's in the readahead
851 int ra_ptr = 0; /* Current index into readahead
855 struct inode *dir = dentry->d_parent->d_inode;
862 blocksize = sb->s_blocksize;
863 namelen = dentry->d_name.len;
864 name = dentry->d_name.name;
865 if (namelen > EXT3_NAME_LEN)
867 #ifdef CONFIG_EXT3_INDEX
869 bh = ext3_dx_find_entry(dentry, res_dir, &err);
871 * On success, or if the error was file not found,
872 * return. Otherwise, fall back to doing a search the
875 if (bh || (err != ERR_BAD_DX_DIR))
877 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
880 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
881 start = EXT3_I(dir)->i_dir_start_lookup;
882 if (start >= nblocks)
888 * We deal with the read-ahead logic here.
890 if (ra_ptr >= ra_max) {
891 /* Refill the readahead buffer */
894 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
896 * Terminate if we reach the end of the
897 * directory and must wrap, or if our
898 * search has finished at this block.
900 if (b >= nblocks || (num && block == start)) {
901 bh_use[ra_max] = NULL;
905 bh = ext3_getblk(NULL, dir, b++, 0, &err);
908 ll_rw_block(READ_META, 1, &bh);
911 if ((bh = bh_use[ra_ptr++]) == NULL)
914 if (!buffer_uptodate(bh)) {
915 /* read error, skip block & hope for the best */
916 ext3_error(sb, __FUNCTION__, "reading directory #%lu "
917 "offset %lu", dir->i_ino, block);
921 i = search_dirblock(bh, dir, dentry,
922 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
924 EXT3_I(dir)->i_dir_start_lookup = block;
926 goto cleanup_and_exit;
930 goto cleanup_and_exit;
933 if (++block >= nblocks)
935 } while (block != start);
938 * If the directory has grown while we were searching, then
939 * search the last part of the directory before giving up.
942 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
943 if (block < nblocks) {
949 /* Clean up the read-ahead blocks */
950 for (; ra_ptr < ra_max; ra_ptr++)
951 brelse (bh_use[ra_ptr]);
955 #ifdef CONFIG_EXT3_INDEX
956 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
957 struct ext3_dir_entry_2 **res_dir, int *err)
959 struct super_block * sb;
960 struct dx_hash_info hinfo;
962 struct dx_frame frames[2], *frame;
963 struct ext3_dir_entry_2 *de, *top;
964 struct buffer_head *bh;
967 int namelen = dentry->d_name.len;
968 const u8 *name = dentry->d_name.name;
969 struct inode *dir = dentry->d_parent->d_inode;
972 /* NFS may look up ".." - look at dx_root directory block */
973 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
974 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
978 frame->bh = NULL; /* for dx_release() */
979 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
980 dx_set_block(frame->at, 0); /* dx_root block is 0 */
984 block = dx_get_block(frame->at);
985 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
987 de = (struct ext3_dir_entry_2 *) bh->b_data;
988 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
989 EXT3_DIR_REC_LEN(0));
990 for (; de < top; de = ext3_next_entry(de))
991 if (ext3_match (namelen, name, de)) {
992 if (!ext3_check_dir_entry("ext3_find_entry",
994 (block<<EXT3_BLOCK_SIZE_BITS(sb))
995 +((char *)de - bh->b_data))) {
997 *err = ERR_BAD_DX_DIR;
1001 dx_release (frames);
1005 /* Check to see if we should continue to search */
1006 retval = ext3_htree_next_block(dir, hash, frame,
1009 ext3_warning(sb, __FUNCTION__,
1010 "error reading index page in directory #%lu",
1015 } while (retval == 1);
1019 dxtrace(printk("%s not found\n", name));
1020 dx_release (frames);
1025 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1027 struct inode * inode;
1028 struct ext3_dir_entry_2 * de;
1029 struct buffer_head * bh;
1031 if (dentry->d_name.len > EXT3_NAME_LEN)
1032 return ERR_PTR(-ENAMETOOLONG);
1034 bh = ext3_find_entry(dentry, &de);
1037 unsigned long ino = le32_to_cpu(de->inode);
1039 if (!ext3_valid_inum(dir->i_sb, ino)) {
1040 ext3_error(dir->i_sb, "ext3_lookup",
1041 "bad inode number: %lu", ino);
1044 inode = iget(dir->i_sb, ino);
1047 return ERR_PTR(-EACCES);
1049 if (is_bad_inode(inode)) {
1051 return ERR_PTR(-ENOENT);
1054 return d_splice_alias(inode, dentry);
1058 struct dentry *ext3_get_parent(struct dentry *child)
1061 struct dentry *parent;
1062 struct inode *inode;
1063 struct dentry dotdot;
1064 struct ext3_dir_entry_2 * de;
1065 struct buffer_head *bh;
1067 dotdot.d_name.name = "..";
1068 dotdot.d_name.len = 2;
1069 dotdot.d_parent = child; /* confusing, isn't it! */
1071 bh = ext3_find_entry(&dotdot, &de);
1074 return ERR_PTR(-ENOENT);
1075 ino = le32_to_cpu(de->inode);
1078 if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1079 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1080 "bad inode number: %lu", ino);
1083 inode = iget(child->d_inode->i_sb, ino);
1086 return ERR_PTR(-EACCES);
1088 if (is_bad_inode(inode)) {
1090 return ERR_PTR(-ENOENT);
1093 parent = d_alloc_anon(inode);
1096 parent = ERR_PTR(-ENOMEM);
1102 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1103 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1104 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1105 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1106 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1107 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1108 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1109 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1112 static inline void ext3_set_de_type(struct super_block *sb,
1113 struct ext3_dir_entry_2 *de,
1115 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1116 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1119 #ifdef CONFIG_EXT3_INDEX
1120 static struct ext3_dir_entry_2 *
1121 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1123 unsigned rec_len = 0;
1126 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1127 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1128 memcpy (to, de, rec_len);
1129 ((struct ext3_dir_entry_2 *) to)->rec_len =
1130 cpu_to_le16(rec_len);
1135 return (struct ext3_dir_entry_2 *) (to - rec_len);
1138 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1140 struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1141 unsigned rec_len = 0;
1144 while ((char*)de < base + size) {
1145 next = (struct ext3_dir_entry_2 *) ((char *) de +
1146 le16_to_cpu(de->rec_len));
1147 if (de->inode && de->name_len) {
1148 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1150 memmove(to, de, rec_len);
1151 to->rec_len = cpu_to_le16(rec_len);
1153 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1160 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1161 struct buffer_head **bh,struct dx_frame *frame,
1162 struct dx_hash_info *hinfo, int *error)
1164 unsigned blocksize = dir->i_sb->s_blocksize;
1165 unsigned count, continued;
1166 struct buffer_head *bh2;
1169 struct dx_map_entry *map;
1170 char *data1 = (*bh)->b_data, *data2;
1172 struct ext3_dir_entry_2 *de = NULL, *de2;
1175 bh2 = ext3_append (handle, dir, &newblock, &err);
1182 BUFFER_TRACE(*bh, "get_write_access");
1183 err = ext3_journal_get_write_access(handle, *bh);
1187 BUFFER_TRACE(frame->bh, "get_write_access");
1188 err = ext3_journal_get_write_access(handle, frame->bh);
1192 data2 = bh2->b_data;
1194 /* create map in the end of data2 block */
1195 map = (struct dx_map_entry *) (data2 + blocksize);
1196 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1197 blocksize, hinfo, map);
1199 split = count/2; // need to adjust to actual middle
1200 dx_sort_map (map, count);
1201 hash2 = map[split].hash;
1202 continued = hash2 == map[split - 1].hash;
1203 dxtrace(printk("Split block %i at %x, %i/%i\n",
1204 dx_get_block(frame->at), hash2, split, count-split));
1206 /* Fancy dance to stay within two buffers */
1207 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1208 de = dx_pack_dirents(data1,blocksize);
1209 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1210 de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1211 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1212 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1214 /* Which block gets the new entry? */
1215 if (hinfo->hash >= hash2)
1220 dx_insert_block (frame, hash2 + continued, newblock);
1221 err = ext3_journal_dirty_metadata (handle, bh2);
1224 err = ext3_journal_dirty_metadata (handle, frame->bh);
1228 dxtrace(dx_show_index ("frame", frame->entries));
1235 ext3_std_error(dir->i_sb, err);
1244 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1245 * it points to a directory entry which is guaranteed to be large
1246 * enough for new directory entry. If de is NULL, then
1247 * add_dirent_to_buf will attempt search the directory block for
1248 * space. It will return -ENOSPC if no space is available, and -EIO
1249 * and -EEXIST if directory entry already exists.
1251 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1252 * all other cases bh is released.
1254 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1255 struct inode *inode, struct ext3_dir_entry_2 *de,
1256 struct buffer_head * bh)
1258 struct inode *dir = dentry->d_parent->d_inode;
1259 const char *name = dentry->d_name.name;
1260 int namelen = dentry->d_name.len;
1261 unsigned long offset = 0;
1262 unsigned short reclen;
1263 int nlen, rlen, err;
1266 reclen = EXT3_DIR_REC_LEN(namelen);
1268 de = (struct ext3_dir_entry_2 *)bh->b_data;
1269 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1270 while ((char *) de <= top) {
1271 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1276 if (ext3_match (namelen, name, de)) {
1280 nlen = EXT3_DIR_REC_LEN(de->name_len);
1281 rlen = le16_to_cpu(de->rec_len);
1282 if ((de->inode? rlen - nlen: rlen) >= reclen)
1284 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1287 if ((char *) de > top)
1290 BUFFER_TRACE(bh, "get_write_access");
1291 err = ext3_journal_get_write_access(handle, bh);
1293 ext3_std_error(dir->i_sb, err);
1298 /* By now the buffer is marked for journaling */
1299 nlen = EXT3_DIR_REC_LEN(de->name_len);
1300 rlen = le16_to_cpu(de->rec_len);
1302 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1303 de1->rec_len = cpu_to_le16(rlen - nlen);
1304 de->rec_len = cpu_to_le16(nlen);
1307 de->file_type = EXT3_FT_UNKNOWN;
1309 de->inode = cpu_to_le32(inode->i_ino);
1310 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1313 de->name_len = namelen;
1314 memcpy (de->name, name, namelen);
1316 * XXX shouldn't update any times until successful
1317 * completion of syscall, but too many callers depend
1320 * XXX similarly, too many callers depend on
1321 * ext3_new_inode() setting the times, but error
1322 * recovery deletes the inode, so the worst that can
1323 * happen is that the times are slightly out of date
1324 * and/or different from the directory change time.
1326 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1327 ext3_update_dx_flag(dir);
1329 ext3_mark_inode_dirty(handle, dir);
1330 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1331 err = ext3_journal_dirty_metadata(handle, bh);
1333 ext3_std_error(dir->i_sb, err);
1338 #ifdef CONFIG_EXT3_INDEX
1340 * This converts a one block unindexed directory to a 3 block indexed
1341 * directory, and adds the dentry to the indexed directory.
1343 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1344 struct inode *inode, struct buffer_head *bh)
1346 struct inode *dir = dentry->d_parent->d_inode;
1347 const char *name = dentry->d_name.name;
1348 int namelen = dentry->d_name.len;
1349 struct buffer_head *bh2;
1350 struct dx_root *root;
1351 struct dx_frame frames[2], *frame;
1352 struct dx_entry *entries;
1353 struct ext3_dir_entry_2 *de, *de2;
1358 struct dx_hash_info hinfo;
1360 struct fake_dirent *fde;
1362 blocksize = dir->i_sb->s_blocksize;
1363 dxtrace(printk("Creating index\n"));
1364 retval = ext3_journal_get_write_access(handle, bh);
1366 ext3_std_error(dir->i_sb, retval);
1370 root = (struct dx_root *) bh->b_data;
1372 bh2 = ext3_append (handle, dir, &block, &retval);
1377 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1378 data1 = bh2->b_data;
1380 /* The 0th block becomes the root, move the dirents out */
1381 fde = &root->dotdot;
1382 de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1383 len = ((char *) root) + blocksize - (char *) de;
1384 memcpy (data1, de, len);
1385 de = (struct ext3_dir_entry_2 *) data1;
1387 while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1389 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1390 /* Initialize the root; the dot dirents already exist */
1391 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1392 de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1393 memset (&root->info, 0, sizeof(root->info));
1394 root->info.info_length = sizeof(root->info);
1395 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1396 entries = root->entries;
1397 dx_set_block (entries, 1);
1398 dx_set_count (entries, 1);
1399 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1401 /* Initialize as for dx_probe */
1402 hinfo.hash_version = root->info.hash_version;
1403 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1404 ext3fs_dirhash(name, namelen, &hinfo);
1406 frame->entries = entries;
1407 frame->at = entries;
1410 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1411 dx_release (frames);
1415 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1422 * adds a file entry to the specified directory, using the same
1423 * semantics as ext3_find_entry(). It returns NULL if it failed.
1425 * NOTE!! The inode part of 'de' is left at 0 - which means you
1426 * may not sleep between calling this and putting something into
1427 * the entry, as someone else might have used it while you slept.
1429 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1430 struct inode *inode)
1432 struct inode *dir = dentry->d_parent->d_inode;
1433 unsigned long offset;
1434 struct buffer_head * bh;
1435 struct ext3_dir_entry_2 *de;
1436 struct super_block * sb;
1438 #ifdef CONFIG_EXT3_INDEX
1445 blocksize = sb->s_blocksize;
1446 if (!dentry->d_name.len)
1448 #ifdef CONFIG_EXT3_INDEX
1450 retval = ext3_dx_add_entry(handle, dentry, inode);
1451 if (!retval || (retval != ERR_BAD_DX_DIR))
1453 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1455 ext3_mark_inode_dirty(handle, dir);
1458 blocks = dir->i_size >> sb->s_blocksize_bits;
1459 for (block = 0, offset = 0; block < blocks; block++) {
1460 bh = ext3_bread(handle, dir, block, 0, &retval);
1463 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1464 if (retval != -ENOSPC)
1467 #ifdef CONFIG_EXT3_INDEX
1468 if (blocks == 1 && !dx_fallback &&
1469 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1470 return make_indexed_dir(handle, dentry, inode, bh);
1474 bh = ext3_append(handle, dir, &block, &retval);
1477 de = (struct ext3_dir_entry_2 *) bh->b_data;
1479 de->rec_len = cpu_to_le16(blocksize);
1480 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1483 #ifdef CONFIG_EXT3_INDEX
1485 * Returns 0 for success, or a negative error value
1487 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1488 struct inode *inode)
1490 struct dx_frame frames[2], *frame;
1491 struct dx_entry *entries, *at;
1492 struct dx_hash_info hinfo;
1493 struct buffer_head * bh;
1494 struct inode *dir = dentry->d_parent->d_inode;
1495 struct super_block * sb = dir->i_sb;
1496 struct ext3_dir_entry_2 *de;
1499 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1502 entries = frame->entries;
1505 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1508 BUFFER_TRACE(bh, "get_write_access");
1509 err = ext3_journal_get_write_access(handle, bh);
1513 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1514 if (err != -ENOSPC) {
1519 /* Block full, should compress but for now just split */
1520 dxtrace(printk("using %u of %u node entries\n",
1521 dx_get_count(entries), dx_get_limit(entries)));
1522 /* Need to split index? */
1523 if (dx_get_count(entries) == dx_get_limit(entries)) {
1525 unsigned icount = dx_get_count(entries);
1526 int levels = frame - frames;
1527 struct dx_entry *entries2;
1528 struct dx_node *node2;
1529 struct buffer_head *bh2;
1531 if (levels && (dx_get_count(frames->entries) ==
1532 dx_get_limit(frames->entries))) {
1533 ext3_warning(sb, __FUNCTION__,
1534 "Directory index full!");
1538 bh2 = ext3_append (handle, dir, &newblock, &err);
1541 node2 = (struct dx_node *)(bh2->b_data);
1542 entries2 = node2->entries;
1543 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1544 node2->fake.inode = 0;
1545 BUFFER_TRACE(frame->bh, "get_write_access");
1546 err = ext3_journal_get_write_access(handle, frame->bh);
1550 unsigned icount1 = icount/2, icount2 = icount - icount1;
1551 unsigned hash2 = dx_get_hash(entries + icount1);
1552 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1554 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1555 err = ext3_journal_get_write_access(handle,
1560 memcpy ((char *) entries2, (char *) (entries + icount1),
1561 icount2 * sizeof(struct dx_entry));
1562 dx_set_count (entries, icount1);
1563 dx_set_count (entries2, icount2);
1564 dx_set_limit (entries2, dx_node_limit(dir));
1566 /* Which index block gets the new entry? */
1567 if (at - entries >= icount1) {
1568 frame->at = at = at - entries - icount1 + entries2;
1569 frame->entries = entries = entries2;
1570 swap(frame->bh, bh2);
1572 dx_insert_block (frames + 0, hash2, newblock);
1573 dxtrace(dx_show_index ("node", frames[1].entries));
1574 dxtrace(dx_show_index ("node",
1575 ((struct dx_node *) bh2->b_data)->entries));
1576 err = ext3_journal_dirty_metadata(handle, bh2);
1581 dxtrace(printk("Creating second level index...\n"));
1582 memcpy((char *) entries2, (char *) entries,
1583 icount * sizeof(struct dx_entry));
1584 dx_set_limit(entries2, dx_node_limit(dir));
1587 dx_set_count(entries, 1);
1588 dx_set_block(entries + 0, newblock);
1589 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1591 /* Add new access path frame */
1593 frame->at = at = at - entries + entries2;
1594 frame->entries = entries = entries2;
1596 err = ext3_journal_get_write_access(handle,
1601 ext3_journal_dirty_metadata(handle, frames[0].bh);
1603 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1606 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1611 ext3_std_error(dir->i_sb, err);
1621 * ext3_delete_entry deletes a directory entry by merging it with the
1624 static int ext3_delete_entry (handle_t *handle,
1626 struct ext3_dir_entry_2 * de_del,
1627 struct buffer_head * bh)
1629 struct ext3_dir_entry_2 * de, * pde;
1634 de = (struct ext3_dir_entry_2 *) bh->b_data;
1635 while (i < bh->b_size) {
1636 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1639 BUFFER_TRACE(bh, "get_write_access");
1640 ext3_journal_get_write_access(handle, bh);
1643 cpu_to_le16(le16_to_cpu(pde->rec_len) +
1644 le16_to_cpu(de->rec_len));
1648 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1649 ext3_journal_dirty_metadata(handle, bh);
1652 i += le16_to_cpu(de->rec_len);
1654 de = (struct ext3_dir_entry_2 *)
1655 ((char *) de + le16_to_cpu(de->rec_len));
1660 static int ext3_add_nondir(handle_t *handle,
1661 struct dentry *dentry, struct inode *inode)
1663 int err = ext3_add_entry(handle, dentry, inode);
1665 ext3_mark_inode_dirty(handle, inode);
1666 d_instantiate(dentry, inode);
1675 * By the time this is called, we already have created
1676 * the directory cache entry for the new file, but it
1677 * is so far negative - it has no inode.
1679 * If the create succeeds, we fill in the inode information
1680 * with d_instantiate().
1682 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1683 struct nameidata *nd)
1686 struct inode * inode;
1687 int err, retries = 0;
1690 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1691 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1692 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1694 return PTR_ERR(handle);
1696 if (IS_DIRSYNC(dir))
1699 inode = ext3_new_inode (handle, dir, mode);
1700 err = PTR_ERR(inode);
1701 if (!IS_ERR(inode)) {
1702 inode->i_op = &ext3_file_inode_operations;
1703 inode->i_fop = &ext3_file_operations;
1704 ext3_set_aops(inode);
1705 err = ext3_add_nondir(handle, dentry, inode);
1707 ext3_journal_stop(handle);
1708 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1713 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1714 int mode, dev_t rdev)
1717 struct inode *inode;
1718 int err, retries = 0;
1720 if (!new_valid_dev(rdev))
1724 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1725 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1726 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1728 return PTR_ERR(handle);
1730 if (IS_DIRSYNC(dir))
1733 inode = ext3_new_inode (handle, dir, mode);
1734 err = PTR_ERR(inode);
1735 if (!IS_ERR(inode)) {
1736 init_special_inode(inode, inode->i_mode, rdev);
1737 #ifdef CONFIG_EXT3_FS_XATTR
1738 inode->i_op = &ext3_special_inode_operations;
1740 err = ext3_add_nondir(handle, dentry, inode);
1742 ext3_journal_stop(handle);
1743 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1748 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1751 struct inode * inode;
1752 struct buffer_head * dir_block;
1753 struct ext3_dir_entry_2 * de;
1754 int err, retries = 0;
1756 if (dir->i_nlink >= EXT3_LINK_MAX)
1760 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1761 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1762 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1764 return PTR_ERR(handle);
1766 if (IS_DIRSYNC(dir))
1769 inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1770 err = PTR_ERR(inode);
1774 inode->i_op = &ext3_dir_inode_operations;
1775 inode->i_fop = &ext3_dir_operations;
1776 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1777 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1779 drop_nlink(inode); /* is this nlink == 0? */
1780 ext3_mark_inode_dirty(handle, inode);
1784 BUFFER_TRACE(dir_block, "get_write_access");
1785 ext3_journal_get_write_access(handle, dir_block);
1786 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1787 de->inode = cpu_to_le32(inode->i_ino);
1789 de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1790 strcpy (de->name, ".");
1791 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1792 de = (struct ext3_dir_entry_2 *)
1793 ((char *) de + le16_to_cpu(de->rec_len));
1794 de->inode = cpu_to_le32(dir->i_ino);
1795 de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1797 strcpy (de->name, "..");
1798 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1800 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1801 ext3_journal_dirty_metadata(handle, dir_block);
1803 ext3_mark_inode_dirty(handle, inode);
1804 err = ext3_add_entry (handle, dentry, inode);
1807 ext3_mark_inode_dirty(handle, inode);
1812 ext3_update_dx_flag(dir);
1813 ext3_mark_inode_dirty(handle, dir);
1814 d_instantiate(dentry, inode);
1816 ext3_journal_stop(handle);
1817 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1823 * routine to check that the specified directory is empty (for rmdir)
1825 static int empty_dir (struct inode * inode)
1827 unsigned long offset;
1828 struct buffer_head * bh;
1829 struct ext3_dir_entry_2 * de, * de1;
1830 struct super_block * sb;
1834 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1835 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1837 ext3_error(inode->i_sb, __FUNCTION__,
1838 "error %d reading directory #%lu offset 0",
1841 ext3_warning(inode->i_sb, __FUNCTION__,
1842 "bad directory (dir #%lu) - no data block",
1846 de = (struct ext3_dir_entry_2 *) bh->b_data;
1847 de1 = (struct ext3_dir_entry_2 *)
1848 ((char *) de + le16_to_cpu(de->rec_len));
1849 if (le32_to_cpu(de->inode) != inode->i_ino ||
1850 !le32_to_cpu(de1->inode) ||
1851 strcmp (".", de->name) ||
1852 strcmp ("..", de1->name)) {
1853 ext3_warning (inode->i_sb, "empty_dir",
1854 "bad directory (dir #%lu) - no `.' or `..'",
1859 offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1860 de = (struct ext3_dir_entry_2 *)
1861 ((char *) de1 + le16_to_cpu(de1->rec_len));
1862 while (offset < inode->i_size ) {
1864 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1867 bh = ext3_bread (NULL, inode,
1868 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1871 ext3_error(sb, __FUNCTION__,
1872 "error %d reading directory"
1874 err, inode->i_ino, offset);
1875 offset += sb->s_blocksize;
1878 de = (struct ext3_dir_entry_2 *) bh->b_data;
1880 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1881 de = (struct ext3_dir_entry_2 *)(bh->b_data +
1883 offset = (offset | (sb->s_blocksize - 1)) + 1;
1886 if (le32_to_cpu(de->inode)) {
1890 offset += le16_to_cpu(de->rec_len);
1891 de = (struct ext3_dir_entry_2 *)
1892 ((char *) de + le16_to_cpu(de->rec_len));
1898 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1899 * such inodes, starting at the superblock, in case we crash before the
1900 * file is closed/deleted, or in case the inode truncate spans multiple
1901 * transactions and the last transaction is not recovered after a crash.
1903 * At filesystem recovery time, we walk this list deleting unlinked
1904 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1906 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1908 struct super_block *sb = inode->i_sb;
1909 struct ext3_iloc iloc;
1913 if (!list_empty(&EXT3_I(inode)->i_orphan))
1916 /* Orphan handling is only valid for files with data blocks
1917 * being truncated, or files being unlinked. */
1919 /* @@@ FIXME: Observation from aviro:
1920 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1921 * here (on lock_super()), so race with ext3_link() which might bump
1922 * ->i_nlink. For, say it, character device. Not a regular file,
1923 * not a directory, not a symlink and ->i_nlink > 0.
1925 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1926 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1928 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1929 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1933 err = ext3_reserve_inode_write(handle, inode, &iloc);
1937 /* Insert this inode at the head of the on-disk orphan list... */
1938 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1939 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1940 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1941 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1945 /* Only add to the head of the in-memory list if all the
1946 * previous operations succeeded. If the orphan_add is going to
1947 * fail (possibly taking the journal offline), we can't risk
1948 * leaving the inode on the orphan list: stray orphan-list
1949 * entries can cause panics at unmount time.
1951 * This is safe: on error we're going to ignore the orphan list
1952 * anyway on the next recovery. */
1954 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1956 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1957 jbd_debug(4, "orphan inode %lu will point to %d\n",
1958 inode->i_ino, NEXT_ORPHAN(inode));
1961 ext3_std_error(inode->i_sb, err);
1966 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1967 * of such inodes stored on disk, because it is finally being cleaned up.
1969 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1971 struct list_head *prev;
1972 struct ext3_inode_info *ei = EXT3_I(inode);
1973 struct ext3_sb_info *sbi;
1974 unsigned long ino_next;
1975 struct ext3_iloc iloc;
1978 lock_super(inode->i_sb);
1979 if (list_empty(&ei->i_orphan)) {
1980 unlock_super(inode->i_sb);
1984 ino_next = NEXT_ORPHAN(inode);
1985 prev = ei->i_orphan.prev;
1986 sbi = EXT3_SB(inode->i_sb);
1988 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1990 list_del_init(&ei->i_orphan);
1992 /* If we're on an error path, we may not have a valid
1993 * transaction handle with which to update the orphan list on
1994 * disk, but we still need to remove the inode from the linked
1995 * list in memory. */
1999 err = ext3_reserve_inode_write(handle, inode, &iloc);
2003 if (prev == &sbi->s_orphan) {
2004 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2005 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2006 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2009 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2010 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2012 struct ext3_iloc iloc2;
2013 struct inode *i_prev =
2014 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2016 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2017 i_prev->i_ino, ino_next);
2018 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2021 NEXT_ORPHAN(i_prev) = ino_next;
2022 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2026 NEXT_ORPHAN(inode) = 0;
2027 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2030 ext3_std_error(inode->i_sb, err);
2032 unlock_super(inode->i_sb);
2040 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2043 struct inode * inode;
2044 struct buffer_head * bh;
2045 struct ext3_dir_entry_2 * de;
2048 /* Initialize quotas before so that eventual writes go in
2049 * separate transaction */
2050 DQUOT_INIT(dentry->d_inode);
2051 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2053 return PTR_ERR(handle);
2056 bh = ext3_find_entry (dentry, &de);
2060 if (IS_DIRSYNC(dir))
2063 inode = dentry->d_inode;
2066 if (le32_to_cpu(de->inode) != inode->i_ino)
2069 retval = -ENOTEMPTY;
2070 if (!empty_dir (inode))
2073 retval = ext3_delete_entry(handle, dir, de, bh);
2076 if (inode->i_nlink != 2)
2077 ext3_warning (inode->i_sb, "ext3_rmdir",
2078 "empty directory has nlink!=2 (%d)",
2082 /* There's no need to set i_disksize: the fact that i_nlink is
2083 * zero will ensure that the right thing happens during any
2086 ext3_orphan_add(handle, inode);
2087 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2088 ext3_mark_inode_dirty(handle, inode);
2090 ext3_update_dx_flag(dir);
2091 ext3_mark_inode_dirty(handle, dir);
2094 ext3_journal_stop(handle);
2099 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2102 struct inode * inode;
2103 struct buffer_head * bh;
2104 struct ext3_dir_entry_2 * de;
2107 /* Initialize quotas before so that eventual writes go
2108 * in separate transaction */
2109 DQUOT_INIT(dentry->d_inode);
2110 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2112 return PTR_ERR(handle);
2114 if (IS_DIRSYNC(dir))
2118 bh = ext3_find_entry (dentry, &de);
2122 inode = dentry->d_inode;
2125 if (le32_to_cpu(de->inode) != inode->i_ino)
2128 if (!inode->i_nlink) {
2129 ext3_warning (inode->i_sb, "ext3_unlink",
2130 "Deleting nonexistent file (%lu), %d",
2131 inode->i_ino, inode->i_nlink);
2134 retval = ext3_delete_entry(handle, dir, de, bh);
2137 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2138 ext3_update_dx_flag(dir);
2139 ext3_mark_inode_dirty(handle, dir);
2141 if (!inode->i_nlink)
2142 ext3_orphan_add(handle, inode);
2143 inode->i_ctime = dir->i_ctime;
2144 ext3_mark_inode_dirty(handle, inode);
2148 ext3_journal_stop(handle);
2153 static int ext3_symlink (struct inode * dir,
2154 struct dentry *dentry, const char * symname)
2157 struct inode * inode;
2158 int l, err, retries = 0;
2160 l = strlen(symname)+1;
2161 if (l > dir->i_sb->s_blocksize)
2162 return -ENAMETOOLONG;
2165 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2166 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2167 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2169 return PTR_ERR(handle);
2171 if (IS_DIRSYNC(dir))
2174 inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2175 err = PTR_ERR(inode);
2179 if (l > sizeof (EXT3_I(inode)->i_data)) {
2180 inode->i_op = &ext3_symlink_inode_operations;
2181 ext3_set_aops(inode);
2183 * page_symlink() calls into ext3_prepare/commit_write.
2184 * We have a transaction open. All is sweetness. It also sets
2185 * i_size in generic_commit_write().
2187 err = __page_symlink(inode, symname, l,
2188 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2191 ext3_mark_inode_dirty(handle, inode);
2196 inode->i_op = &ext3_fast_symlink_inode_operations;
2197 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2198 inode->i_size = l-1;
2200 EXT3_I(inode)->i_disksize = inode->i_size;
2201 err = ext3_add_nondir(handle, dentry, inode);
2203 ext3_journal_stop(handle);
2204 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2209 static int ext3_link (struct dentry * old_dentry,
2210 struct inode * dir, struct dentry *dentry)
2213 struct inode *inode = old_dentry->d_inode;
2214 int err, retries = 0;
2216 if (inode->i_nlink >= EXT3_LINK_MAX)
2219 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2220 * otherwise has the potential to corrupt the orphan inode list.
2222 if (inode->i_nlink == 0)
2226 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2227 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2229 return PTR_ERR(handle);
2231 if (IS_DIRSYNC(dir))
2234 inode->i_ctime = CURRENT_TIME_SEC;
2236 atomic_inc(&inode->i_count);
2238 err = ext3_add_nondir(handle, dentry, inode);
2239 ext3_journal_stop(handle);
2240 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2245 #define PARENT_INO(buffer) \
2246 ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2247 le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2250 * Anybody can rename anything with this: the permission checks are left to the
2251 * higher-level routines.
2253 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2254 struct inode * new_dir,struct dentry *new_dentry)
2257 struct inode * old_inode, * new_inode;
2258 struct buffer_head * old_bh, * new_bh, * dir_bh;
2259 struct ext3_dir_entry_2 * old_de, * new_de;
2262 old_bh = new_bh = dir_bh = NULL;
2264 /* Initialize quotas before so that eventual writes go
2265 * in separate transaction */
2266 if (new_dentry->d_inode)
2267 DQUOT_INIT(new_dentry->d_inode);
2268 handle = ext3_journal_start(old_dir, 2 *
2269 EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2270 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2272 return PTR_ERR(handle);
2274 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2277 old_bh = ext3_find_entry (old_dentry, &old_de);
2279 * Check for inode number is _not_ due to possible IO errors.
2280 * We might rmdir the source, keep it as pwd of some process
2281 * and merrily kill the link to whatever was created under the
2282 * same name. Goodbye sticky bit ;-<
2284 old_inode = old_dentry->d_inode;
2286 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2289 new_inode = new_dentry->d_inode;
2290 new_bh = ext3_find_entry (new_dentry, &new_de);
2297 if (S_ISDIR(old_inode->i_mode)) {
2299 retval = -ENOTEMPTY;
2300 if (!empty_dir (new_inode))
2304 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2307 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2310 if (!new_inode && new_dir!=old_dir &&
2311 new_dir->i_nlink >= EXT3_LINK_MAX)
2315 retval = ext3_add_entry (handle, new_dentry, old_inode);
2319 BUFFER_TRACE(new_bh, "get write access");
2320 ext3_journal_get_write_access(handle, new_bh);
2321 new_de->inode = cpu_to_le32(old_inode->i_ino);
2322 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2323 EXT3_FEATURE_INCOMPAT_FILETYPE))
2324 new_de->file_type = old_de->file_type;
2325 new_dir->i_version++;
2326 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2327 ext3_journal_dirty_metadata(handle, new_bh);
2333 * Like most other Unix systems, set the ctime for inodes on a
2336 old_inode->i_ctime = CURRENT_TIME_SEC;
2337 ext3_mark_inode_dirty(handle, old_inode);
2342 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2343 old_de->name_len != old_dentry->d_name.len ||
2344 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2345 (retval = ext3_delete_entry(handle, old_dir,
2346 old_de, old_bh)) == -ENOENT) {
2347 /* old_de could have moved from under us during htree split, so
2348 * make sure that we are deleting the right entry. We might
2349 * also be pointing to a stale entry in the unused part of
2350 * old_bh so just checking inum and the name isn't enough. */
2351 struct buffer_head *old_bh2;
2352 struct ext3_dir_entry_2 *old_de2;
2354 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2356 retval = ext3_delete_entry(handle, old_dir,
2362 ext3_warning(old_dir->i_sb, "ext3_rename",
2363 "Deleting old file (%lu), %d, error=%d",
2364 old_dir->i_ino, old_dir->i_nlink, retval);
2368 drop_nlink(new_inode);
2369 new_inode->i_ctime = CURRENT_TIME_SEC;
2371 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2372 ext3_update_dx_flag(old_dir);
2374 BUFFER_TRACE(dir_bh, "get_write_access");
2375 ext3_journal_get_write_access(handle, dir_bh);
2376 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2377 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2378 ext3_journal_dirty_metadata(handle, dir_bh);
2379 drop_nlink(old_dir);
2381 drop_nlink(new_inode);
2384 ext3_update_dx_flag(new_dir);
2385 ext3_mark_inode_dirty(handle, new_dir);
2388 ext3_mark_inode_dirty(handle, old_dir);
2390 ext3_mark_inode_dirty(handle, new_inode);
2391 if (!new_inode->i_nlink)
2392 ext3_orphan_add(handle, new_inode);
2400 ext3_journal_stop(handle);
2405 * directories can handle most operations...
2407 const struct inode_operations ext3_dir_inode_operations = {
2408 .create = ext3_create,
2409 .lookup = ext3_lookup,
2411 .unlink = ext3_unlink,
2412 .symlink = ext3_symlink,
2413 .mkdir = ext3_mkdir,
2414 .rmdir = ext3_rmdir,
2415 .mknod = ext3_mknod,
2416 .rename = ext3_rename,
2417 .setattr = ext3_setattr,
2418 #ifdef CONFIG_EXT3_FS_XATTR
2419 .setxattr = generic_setxattr,
2420 .getxattr = generic_getxattr,
2421 .listxattr = ext3_listxattr,
2422 .removexattr = generic_removexattr,
2424 .permission = ext3_permission,
2427 const struct inode_operations ext3_special_inode_operations = {
2428 .setattr = ext3_setattr,
2429 #ifdef CONFIG_EXT3_FS_XATTR
2430 .setxattr = generic_setxattr,
2431 .getxattr = generic_getxattr,
2432 .listxattr = ext3_listxattr,
2433 .removexattr = generic_removexattr,
2435 .permission = ext3_permission,