2 * linux/fs/ext3/namei.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/namei.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
28 #include <linux/pagemap.h>
29 #include <linux/jbd.h>
30 #include <linux/time.h>
31 #include <linux/ext3_fs.h>
32 #include <linux/ext3_jbd.h>
33 #include <linux/fcntl.h>
34 #include <linux/stat.h>
35 #include <linux/string.h>
36 #include <linux/quotaops.h>
37 #include <linux/buffer_head.h>
38 #include <linux/smp_lock.h>
43 * define how far ahead to read directories while searching them.
45 #define NAMEI_RA_CHUNKS 2
46 #define NAMEI_RA_BLOCKS 4
47 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
48 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
50 static struct buffer_head *ext3_append(handle_t *handle,
54 struct buffer_head *bh;
56 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
58 if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
59 inode->i_size += inode->i_sb->s_blocksize;
60 EXT3_I(inode)->i_disksize = inode->i_size;
61 ext3_journal_get_write_access(handle,bh);
67 #define assert(test) J_ASSERT(test)
71 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
74 typedef struct { u32 v; } le_u32;
75 typedef struct { u16 v; } le_u16;
78 #define dxtrace(command) command
80 #define dxtrace(command)
104 * dx_root_info is laid out so that if it should somehow get overlaid by a
105 * dirent the two low bits of the hash version will be zero. Therefore, the
106 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
111 struct fake_dirent dot;
113 struct fake_dirent dotdot;
117 le_u32 reserved_zero;
119 u8 info_length; /* 8 */
124 struct dx_entry entries[0];
129 struct fake_dirent fake;
130 struct dx_entry entries[0];
136 struct buffer_head *bh;
137 struct dx_entry *entries;
147 #ifdef CONFIG_EXT3_INDEX
148 static inline unsigned dx_get_block (struct dx_entry *entry);
149 static void dx_set_block (struct dx_entry *entry, unsigned value);
150 static inline unsigned dx_get_hash (struct dx_entry *entry);
151 static void dx_set_hash (struct dx_entry *entry, unsigned value);
152 static unsigned dx_get_count (struct dx_entry *entries);
153 static unsigned dx_get_limit (struct dx_entry *entries);
154 static void dx_set_count (struct dx_entry *entries, unsigned value);
155 static void dx_set_limit (struct dx_entry *entries, unsigned value);
156 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
157 static unsigned dx_node_limit (struct inode *dir);
158 static struct dx_frame *dx_probe(struct dentry *dentry,
160 struct dx_hash_info *hinfo,
161 struct dx_frame *frame,
163 static void dx_release (struct dx_frame *frames);
164 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
165 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
166 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
167 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
168 struct dx_map_entry *offsets, int count);
169 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
170 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
171 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
172 struct dx_frame *frame,
173 struct dx_frame *frames,
175 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
176 struct ext3_dir_entry_2 **res_dir, int *err);
177 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
178 struct inode *inode);
181 * Future: use high four bits of block for coalesce-on-delete flags
182 * Mask them off for now.
185 static inline unsigned dx_get_block (struct dx_entry *entry)
187 return le32_to_cpu(entry->block.v) & 0x00ffffff;
190 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
192 entry->block.v = cpu_to_le32(value);
195 static inline unsigned dx_get_hash (struct dx_entry *entry)
197 return le32_to_cpu(entry->hash.v);
200 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
202 entry->hash.v = cpu_to_le32(value);
205 static inline unsigned dx_get_count (struct dx_entry *entries)
207 return le16_to_cpu(((struct dx_countlimit *) entries)->count.v);
210 static inline unsigned dx_get_limit (struct dx_entry *entries)
212 return le16_to_cpu(((struct dx_countlimit *) entries)->limit.v);
215 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
217 ((struct dx_countlimit *) entries)->count.v = cpu_to_le16(value);
220 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
222 ((struct dx_countlimit *) entries)->limit.v = cpu_to_le16(value);
225 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
227 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
228 EXT3_DIR_REC_LEN(2) - infosize;
229 return 0? 20: entry_space / sizeof(struct dx_entry);
232 static inline unsigned dx_node_limit (struct inode *dir)
234 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
235 return 0? 22: entry_space / sizeof(struct dx_entry);
242 static void dx_show_index (char * label, struct dx_entry *entries)
244 int i, n = dx_get_count (entries);
245 printk("%s index ", label);
246 for (i = 0; i < n; i++)
248 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
260 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
261 int size, int show_names)
263 unsigned names = 0, space = 0;
264 char *base = (char *) de;
265 struct dx_hash_info h = *hinfo;
268 while ((char *) de < base + size)
274 int len = de->name_len;
275 char *name = de->name;
276 while (len--) printk("%c", *name++);
277 ext3fs_dirhash(de->name, de->name_len, &h);
278 printk(":%x.%u ", h.hash,
279 ((char *) de - base));
281 space += EXT3_DIR_REC_LEN(de->name_len);
284 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
286 printk("(%i)\n", names);
287 return (struct stats) { names, space, 1 };
290 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
291 struct dx_entry *entries, int levels)
293 unsigned blocksize = dir->i_sb->s_blocksize;
294 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
296 struct buffer_head *bh;
298 printk("%i indexed blocks...\n", count);
299 for (i = 0; i < count; i++, entries++)
301 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
302 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
304 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
305 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
307 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
308 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
309 names += stats.names;
310 space += stats.space;
311 bcount += stats.bcount;
315 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
316 names, space/bcount,(space/bcount)*100/blocksize);
317 return (struct stats) { names, space, bcount};
319 #endif /* DX_DEBUG */
322 * Probe for a directory leaf block to search.
324 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
325 * error in the directory index, and the caller should fall back to
326 * searching the directory normally. The callers of dx_probe **MUST**
327 * check for this error code, and make sure it never gets reflected
330 static struct dx_frame *
331 dx_probe(struct dentry *dentry, struct inode *dir,
332 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
334 unsigned count, indirect;
335 struct dx_entry *at, *entries, *p, *q, *m;
336 struct dx_root *root;
337 struct buffer_head *bh;
338 struct dx_frame *frame = frame_in;
343 dir = dentry->d_parent->d_inode;
344 if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
346 root = (struct dx_root *) bh->b_data;
347 if (root->info.hash_version != DX_HASH_TEA &&
348 root->info.hash_version != DX_HASH_HALF_MD4 &&
349 root->info.hash_version != DX_HASH_LEGACY) {
350 ext3_warning(dir->i_sb, __FUNCTION__,
351 "Unrecognised inode hash code %d",
352 root->info.hash_version);
354 *err = ERR_BAD_DX_DIR;
357 hinfo->hash_version = root->info.hash_version;
358 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
360 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
363 if (root->info.unused_flags & 1) {
364 ext3_warning(dir->i_sb, __FUNCTION__,
365 "Unimplemented inode hash flags: %#06x",
366 root->info.unused_flags);
368 *err = ERR_BAD_DX_DIR;
372 if ((indirect = root->info.indirect_levels) > 1) {
373 ext3_warning(dir->i_sb, __FUNCTION__,
374 "Unimplemented inode hash depth: %#06x",
375 root->info.indirect_levels);
377 *err = ERR_BAD_DX_DIR;
381 entries = (struct dx_entry *) (((char *)&root->info) +
382 root->info.info_length);
383 assert(dx_get_limit(entries) == dx_root_limit(dir,
384 root->info.info_length));
385 dxtrace (printk("Look up %x", hash));
388 count = dx_get_count(entries);
389 assert (count && count <= dx_get_limit(entries));
391 q = entries + count - 1;
395 dxtrace(printk("."));
396 if (dx_get_hash(m) > hash)
402 if (0) // linear search cross check
404 unsigned n = count - 1;
408 dxtrace(printk(","));
409 if (dx_get_hash(++at) > hash)
415 assert (at == p - 1);
419 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
421 frame->entries = entries;
423 if (!indirect--) return frame;
424 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
426 at = entries = ((struct dx_node *) bh->b_data)->entries;
427 assert (dx_get_limit(entries) == dx_node_limit (dir));
431 while (frame >= frame_in) {
439 static void dx_release (struct dx_frame *frames)
441 if (frames[0].bh == NULL)
444 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
445 brelse(frames[1].bh);
446 brelse(frames[0].bh);
450 * This function increments the frame pointer to search the next leaf
451 * block, and reads in the necessary intervening nodes if the search
452 * should be necessary. Whether or not the search is necessary is
453 * controlled by the hash parameter. If the hash value is even, then
454 * the search is only continued if the next block starts with that
455 * hash value. This is used if we are searching for a specific file.
457 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
459 * This function returns 1 if the caller should continue to search,
460 * or 0 if it should not. If there is an error reading one of the
461 * index blocks, it will a negative error code.
463 * If start_hash is non-null, it will be filled in with the starting
464 * hash of the next page.
466 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
467 struct dx_frame *frame,
468 struct dx_frame *frames,
472 struct buffer_head *bh;
473 int err, num_frames = 0;
478 * Find the next leaf page by incrementing the frame pointer.
479 * If we run out of entries in the interior node, loop around and
480 * increment pointer in the parent node. When we break out of
481 * this loop, num_frames indicates the number of interior
482 * nodes need to be read.
485 if (++(p->at) < p->entries + dx_get_count(p->entries))
494 * If the hash is 1, then continue only if the next page has a
495 * continuation hash of any value. This is used for readdir
496 * handling. Otherwise, check to see if the hash matches the
497 * desired contiuation hash. If it doesn't, return since
498 * there's no point to read in the successive index pages.
500 bhash = dx_get_hash(p->at);
503 if ((hash & 1) == 0) {
504 if ((bhash & ~1) != hash)
508 * If the hash is HASH_NB_ALWAYS, we always go to the next
509 * block so no check is necessary
511 while (num_frames--) {
512 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
514 return err; /* Failure */
518 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
525 * p is at least 6 bytes before the end of page
527 static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
529 return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
533 * This function fills a red-black tree with information from a
534 * directory block. It returns the number directory entries loaded
535 * into the tree. If there is an error it is returned in err.
537 static int htree_dirblock_to_tree(struct file *dir_file,
538 struct inode *dir, int block,
539 struct dx_hash_info *hinfo,
540 __u32 start_hash, __u32 start_minor_hash)
542 struct buffer_head *bh;
543 struct ext3_dir_entry_2 *de, *top;
546 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
547 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
550 de = (struct ext3_dir_entry_2 *) bh->b_data;
551 top = (struct ext3_dir_entry_2 *) ((char *) de +
552 dir->i_sb->s_blocksize -
553 EXT3_DIR_REC_LEN(0));
554 for (; de < top; de = ext3_next_entry(de)) {
555 ext3fs_dirhash(de->name, de->name_len, hinfo);
556 if ((hinfo->hash < start_hash) ||
557 ((hinfo->hash == start_hash) &&
558 (hinfo->minor_hash < start_minor_hash)))
562 if ((err = ext3_htree_store_dirent(dir_file,
563 hinfo->hash, hinfo->minor_hash, de)) != 0) {
575 * This function fills a red-black tree with information from a
576 * directory. We start scanning the directory in hash order, starting
577 * at start_hash and start_minor_hash.
579 * This function returns the number of entries inserted into the tree,
580 * or a negative error code.
582 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
583 __u32 start_minor_hash, __u32 *next_hash)
585 struct dx_hash_info hinfo;
586 struct ext3_dir_entry_2 *de;
587 struct dx_frame frames[2], *frame;
594 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
596 dir = dir_file->f_dentry->d_inode;
597 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
598 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
599 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
600 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
601 start_hash, start_minor_hash);
605 hinfo.hash = start_hash;
606 hinfo.minor_hash = 0;
607 frame = dx_probe(NULL, dir_file->f_dentry->d_inode, &hinfo, frames, &err);
611 /* Add '.' and '..' from the htree header */
612 if (!start_hash && !start_minor_hash) {
613 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
614 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
616 de = ext3_next_entry(de);
617 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
623 block = dx_get_block(frame->at);
624 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
625 start_hash, start_minor_hash);
632 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
633 frame, frames, &hashval);
634 *next_hash = hashval;
640 * Stop if: (a) there are no more entries, or
641 * (b) we have inserted at least one entry and the
642 * next hash value is not a continuation
645 (count && ((hashval & 1) == 0)))
649 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
659 * Directory block splitting, compacting
662 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
663 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
666 char *base = (char *) de;
667 struct dx_hash_info h = *hinfo;
669 while ((char *) de < base + size)
671 if (de->name_len && de->inode) {
672 ext3fs_dirhash(de->name, de->name_len, &h);
674 map_tail->hash = h.hash;
675 map_tail->offs = (u32) ((char *) de - base);
678 /* XXX: do we need to check rec_len == 0 case? -Chris */
679 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
684 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
686 struct dx_map_entry *p, *q, *top = map + count - 1;
688 /* Combsort until bubble sort doesn't suck */
692 if (count - 9 < 2) /* 9, 10 -> 11 */
694 for (p = top, q = p - count; q >= map; p--, q--)
695 if (p->hash < q->hash)
698 /* Garden variety bubble sort */
704 if (q[1].hash >= q[0].hash)
712 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
714 struct dx_entry *entries = frame->entries;
715 struct dx_entry *old = frame->at, *new = old + 1;
716 int count = dx_get_count(entries);
718 assert(count < dx_get_limit(entries));
719 assert(old < entries + count);
720 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
721 dx_set_hash(new, hash);
722 dx_set_block(new, block);
723 dx_set_count(entries, count + 1);
728 static void ext3_update_dx_flag(struct inode *inode)
730 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
731 EXT3_FEATURE_COMPAT_DIR_INDEX))
732 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
736 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
738 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
739 * `de != NULL' is guaranteed by caller.
741 static inline int ext3_match (int len, const char * const name,
742 struct ext3_dir_entry_2 * de)
744 if (len != de->name_len)
748 return !memcmp(name, de->name, len);
752 * Returns 0 if not found, -1 on failure, and 1 on success
754 static inline int search_dirblock(struct buffer_head * bh,
756 struct dentry *dentry,
757 unsigned long offset,
758 struct ext3_dir_entry_2 ** res_dir)
760 struct ext3_dir_entry_2 * de;
763 const char *name = dentry->d_name.name;
764 int namelen = dentry->d_name.len;
766 de = (struct ext3_dir_entry_2 *) bh->b_data;
767 dlimit = bh->b_data + dir->i_sb->s_blocksize;
768 while ((char *) de < dlimit) {
769 /* this code is executed quadratically often */
770 /* do minimal checking `by hand' */
772 if ((char *) de + namelen <= dlimit &&
773 ext3_match (namelen, name, de)) {
774 /* found a match - just to be sure, do a full check */
775 if (!ext3_check_dir_entry("ext3_find_entry",
776 dir, de, bh, offset))
781 /* prevent looping on a bad block */
782 de_len = le16_to_cpu(de->rec_len);
786 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
795 * finds an entry in the specified directory with the wanted name. It
796 * returns the cache buffer in which the entry was found, and the entry
797 * itself (as a parameter - res_dir). It does NOT read the inode of the
798 * entry - you'll have to do that yourself if you want to.
800 * The returned buffer_head has ->b_count elevated. The caller is expected
801 * to brelse() it when appropriate.
803 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
804 struct ext3_dir_entry_2 ** res_dir)
806 struct super_block * sb;
807 struct buffer_head * bh_use[NAMEI_RA_SIZE];
808 struct buffer_head * bh, *ret = NULL;
809 unsigned long start, block, b;
810 int ra_max = 0; /* Number of bh's in the readahead
812 int ra_ptr = 0; /* Current index into readahead
816 struct inode *dir = dentry->d_parent->d_inode;
823 blocksize = sb->s_blocksize;
824 namelen = dentry->d_name.len;
825 name = dentry->d_name.name;
826 if (namelen > EXT3_NAME_LEN)
828 #ifdef CONFIG_EXT3_INDEX
830 bh = ext3_dx_find_entry(dentry, res_dir, &err);
832 * On success, or if the error was file not found,
833 * return. Otherwise, fall back to doing a search the
836 if (bh || (err != ERR_BAD_DX_DIR))
838 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
841 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
842 start = EXT3_I(dir)->i_dir_start_lookup;
843 if (start >= nblocks)
849 * We deal with the read-ahead logic here.
851 if (ra_ptr >= ra_max) {
852 /* Refill the readahead buffer */
855 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
857 * Terminate if we reach the end of the
858 * directory and must wrap, or if our
859 * search has finished at this block.
861 if (b >= nblocks || (num && block == start)) {
862 bh_use[ra_max] = NULL;
866 bh = ext3_getblk(NULL, dir, b++, 0, &err);
869 ll_rw_block(READ, 1, &bh);
872 if ((bh = bh_use[ra_ptr++]) == NULL)
875 if (!buffer_uptodate(bh)) {
876 /* read error, skip block & hope for the best */
877 ext3_error(sb, __FUNCTION__, "reading directory #%lu "
878 "offset %lu\n", dir->i_ino, block);
882 i = search_dirblock(bh, dir, dentry,
883 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
885 EXT3_I(dir)->i_dir_start_lookup = block;
887 goto cleanup_and_exit;
891 goto cleanup_and_exit;
894 if (++block >= nblocks)
896 } while (block != start);
899 * If the directory has grown while we were searching, then
900 * search the last part of the directory before giving up.
903 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
904 if (block < nblocks) {
910 /* Clean up the read-ahead blocks */
911 for (; ra_ptr < ra_max; ra_ptr++)
912 brelse (bh_use[ra_ptr]);
916 #ifdef CONFIG_EXT3_INDEX
917 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
918 struct ext3_dir_entry_2 **res_dir, int *err)
920 struct super_block * sb;
921 struct dx_hash_info hinfo;
923 struct dx_frame frames[2], *frame;
924 struct ext3_dir_entry_2 *de, *top;
925 struct buffer_head *bh;
928 int namelen = dentry->d_name.len;
929 const u8 *name = dentry->d_name.name;
930 struct inode *dir = dentry->d_parent->d_inode;
933 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
937 block = dx_get_block(frame->at);
938 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
940 de = (struct ext3_dir_entry_2 *) bh->b_data;
941 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
942 EXT3_DIR_REC_LEN(0));
943 for (; de < top; de = ext3_next_entry(de))
944 if (ext3_match (namelen, name, de)) {
945 if (!ext3_check_dir_entry("ext3_find_entry",
947 (block<<EXT3_BLOCK_SIZE_BITS(sb))
948 +((char *)de - bh->b_data))) {
957 /* Check to see if we should continue to search */
958 retval = ext3_htree_next_block(dir, hash, frame,
961 ext3_warning(sb, __FUNCTION__,
962 "error reading index page in directory #%lu",
967 } while (retval == 1);
971 dxtrace(printk("%s not found\n", name));
977 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
979 struct inode * inode;
980 struct ext3_dir_entry_2 * de;
981 struct buffer_head * bh;
983 if (dentry->d_name.len > EXT3_NAME_LEN)
984 return ERR_PTR(-ENAMETOOLONG);
986 bh = ext3_find_entry(dentry, &de);
989 unsigned long ino = le32_to_cpu(de->inode);
991 inode = iget(dir->i_sb, ino);
994 return ERR_PTR(-EACCES);
997 return d_splice_alias(inode, dentry);
998 d_add(dentry, inode);
1003 struct dentry *ext3_get_parent(struct dentry *child)
1006 struct dentry *parent;
1007 struct inode *inode;
1008 struct dentry dotdot;
1009 struct ext3_dir_entry_2 * de;
1010 struct buffer_head *bh;
1012 dotdot.d_name.name = "..";
1013 dotdot.d_name.len = 2;
1014 dotdot.d_parent = child; /* confusing, isn't it! */
1016 bh = ext3_find_entry(&dotdot, &de);
1019 return ERR_PTR(-ENOENT);
1020 ino = le32_to_cpu(de->inode);
1022 inode = iget(child->d_inode->i_sb, ino);
1025 return ERR_PTR(-EACCES);
1027 parent = d_alloc_anon(inode);
1030 parent = ERR_PTR(-ENOMEM);
1036 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1037 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1038 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1039 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1040 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1041 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1042 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1043 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1046 static inline void ext3_set_de_type(struct super_block *sb,
1047 struct ext3_dir_entry_2 *de,
1049 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1050 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1053 #ifdef CONFIG_EXT3_INDEX
1054 static struct ext3_dir_entry_2 *
1055 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1057 unsigned rec_len = 0;
1060 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1061 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1062 memcpy (to, de, rec_len);
1063 ((struct ext3_dir_entry_2 *) to)->rec_len =
1064 cpu_to_le16(rec_len);
1069 return (struct ext3_dir_entry_2 *) (to - rec_len);
1072 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1074 struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1075 unsigned rec_len = 0;
1078 while ((char*)de < base + size) {
1079 next = (struct ext3_dir_entry_2 *) ((char *) de +
1080 le16_to_cpu(de->rec_len));
1081 if (de->inode && de->name_len) {
1082 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1084 memmove(to, de, rec_len);
1085 to->rec_len = cpu_to_le16(rec_len);
1087 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1094 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1095 struct buffer_head **bh,struct dx_frame *frame,
1096 struct dx_hash_info *hinfo, int *error)
1098 unsigned blocksize = dir->i_sb->s_blocksize;
1099 unsigned count, continued;
1100 struct buffer_head *bh2;
1103 struct dx_map_entry *map;
1104 char *data1 = (*bh)->b_data, *data2;
1106 struct ext3_dir_entry_2 *de = NULL, *de2;
1109 bh2 = ext3_append (handle, dir, &newblock, error);
1116 BUFFER_TRACE(*bh, "get_write_access");
1117 err = ext3_journal_get_write_access(handle, *bh);
1123 ext3_std_error(dir->i_sb, err);
1126 BUFFER_TRACE(frame->bh, "get_write_access");
1127 err = ext3_journal_get_write_access(handle, frame->bh);
1131 data2 = bh2->b_data;
1133 /* create map in the end of data2 block */
1134 map = (struct dx_map_entry *) (data2 + blocksize);
1135 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1136 blocksize, hinfo, map);
1138 split = count/2; // need to adjust to actual middle
1139 dx_sort_map (map, count);
1140 hash2 = map[split].hash;
1141 continued = hash2 == map[split - 1].hash;
1142 dxtrace(printk("Split block %i at %x, %i/%i\n",
1143 dx_get_block(frame->at), hash2, split, count-split));
1145 /* Fancy dance to stay within two buffers */
1146 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1147 de = dx_pack_dirents(data1,blocksize);
1148 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1149 de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1150 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1151 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1153 /* Which block gets the new entry? */
1154 if (hinfo->hash >= hash2)
1159 dx_insert_block (frame, hash2 + continued, newblock);
1160 err = ext3_journal_dirty_metadata (handle, bh2);
1163 err = ext3_journal_dirty_metadata (handle, frame->bh);
1167 dxtrace(dx_show_index ("frame", frame->entries));
1175 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1176 * it points to a directory entry which is guaranteed to be large
1177 * enough for new directory entry. If de is NULL, then
1178 * add_dirent_to_buf will attempt search the directory block for
1179 * space. It will return -ENOSPC if no space is available, and -EIO
1180 * and -EEXIST if directory entry already exists.
1182 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1183 * all other cases bh is released.
1185 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1186 struct inode *inode, struct ext3_dir_entry_2 *de,
1187 struct buffer_head * bh)
1189 struct inode *dir = dentry->d_parent->d_inode;
1190 const char *name = dentry->d_name.name;
1191 int namelen = dentry->d_name.len;
1192 unsigned long offset = 0;
1193 unsigned short reclen;
1194 int nlen, rlen, err;
1197 reclen = EXT3_DIR_REC_LEN(namelen);
1199 de = (struct ext3_dir_entry_2 *)bh->b_data;
1200 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1201 while ((char *) de <= top) {
1202 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1207 if (ext3_match (namelen, name, de)) {
1211 nlen = EXT3_DIR_REC_LEN(de->name_len);
1212 rlen = le16_to_cpu(de->rec_len);
1213 if ((de->inode? rlen - nlen: rlen) >= reclen)
1215 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1218 if ((char *) de > top)
1221 BUFFER_TRACE(bh, "get_write_access");
1222 err = ext3_journal_get_write_access(handle, bh);
1224 ext3_std_error(dir->i_sb, err);
1229 /* By now the buffer is marked for journaling */
1230 nlen = EXT3_DIR_REC_LEN(de->name_len);
1231 rlen = le16_to_cpu(de->rec_len);
1233 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1234 de1->rec_len = cpu_to_le16(rlen - nlen);
1235 de->rec_len = cpu_to_le16(nlen);
1238 de->file_type = EXT3_FT_UNKNOWN;
1240 de->inode = cpu_to_le32(inode->i_ino);
1241 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1244 de->name_len = namelen;
1245 memcpy (de->name, name, namelen);
1247 * XXX shouldn't update any times until successful
1248 * completion of syscall, but too many callers depend
1251 * XXX similarly, too many callers depend on
1252 * ext3_new_inode() setting the times, but error
1253 * recovery deletes the inode, so the worst that can
1254 * happen is that the times are slightly out of date
1255 * and/or different from the directory change time.
1257 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1258 ext3_update_dx_flag(dir);
1260 ext3_mark_inode_dirty(handle, dir);
1261 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1262 err = ext3_journal_dirty_metadata(handle, bh);
1264 ext3_std_error(dir->i_sb, err);
1269 #ifdef CONFIG_EXT3_INDEX
1271 * This converts a one block unindexed directory to a 3 block indexed
1272 * directory, and adds the dentry to the indexed directory.
1274 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1275 struct inode *inode, struct buffer_head *bh)
1277 struct inode *dir = dentry->d_parent->d_inode;
1278 const char *name = dentry->d_name.name;
1279 int namelen = dentry->d_name.len;
1280 struct buffer_head *bh2;
1281 struct dx_root *root;
1282 struct dx_frame frames[2], *frame;
1283 struct dx_entry *entries;
1284 struct ext3_dir_entry_2 *de, *de2;
1289 struct dx_hash_info hinfo;
1291 struct fake_dirent *fde;
1293 blocksize = dir->i_sb->s_blocksize;
1294 dxtrace(printk("Creating index\n"));
1295 retval = ext3_journal_get_write_access(handle, bh);
1297 ext3_std_error(dir->i_sb, retval);
1301 root = (struct dx_root *) bh->b_data;
1303 bh2 = ext3_append (handle, dir, &block, &retval);
1308 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1309 data1 = bh2->b_data;
1311 /* The 0th block becomes the root, move the dirents out */
1312 fde = &root->dotdot;
1313 de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1314 len = ((char *) root) + blocksize - (char *) de;
1315 memcpy (data1, de, len);
1316 de = (struct ext3_dir_entry_2 *) data1;
1318 while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1320 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1321 /* Initialize the root; the dot dirents already exist */
1322 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1323 de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1324 memset (&root->info, 0, sizeof(root->info));
1325 root->info.info_length = sizeof(root->info);
1326 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1327 entries = root->entries;
1328 dx_set_block (entries, 1);
1329 dx_set_count (entries, 1);
1330 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1332 /* Initialize as for dx_probe */
1333 hinfo.hash_version = root->info.hash_version;
1334 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1335 ext3fs_dirhash(name, namelen, &hinfo);
1337 frame->entries = entries;
1338 frame->at = entries;
1341 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1342 dx_release (frames);
1346 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1353 * adds a file entry to the specified directory, using the same
1354 * semantics as ext3_find_entry(). It returns NULL if it failed.
1356 * NOTE!! The inode part of 'de' is left at 0 - which means you
1357 * may not sleep between calling this and putting something into
1358 * the entry, as someone else might have used it while you slept.
1360 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1361 struct inode *inode)
1363 struct inode *dir = dentry->d_parent->d_inode;
1364 unsigned long offset;
1365 struct buffer_head * bh;
1366 struct ext3_dir_entry_2 *de;
1367 struct super_block * sb;
1369 #ifdef CONFIG_EXT3_INDEX
1373 unsigned nlen, rlen;
1377 blocksize = sb->s_blocksize;
1378 if (!dentry->d_name.len)
1380 #ifdef CONFIG_EXT3_INDEX
1382 retval = ext3_dx_add_entry(handle, dentry, inode);
1383 if (!retval || (retval != ERR_BAD_DX_DIR))
1385 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1387 ext3_mark_inode_dirty(handle, dir);
1390 blocks = dir->i_size >> sb->s_blocksize_bits;
1391 for (block = 0, offset = 0; block < blocks; block++) {
1392 bh = ext3_bread(handle, dir, block, 0, &retval);
1395 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1396 if (retval != -ENOSPC)
1399 #ifdef CONFIG_EXT3_INDEX
1400 if (blocks == 1 && !dx_fallback &&
1401 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1402 return make_indexed_dir(handle, dentry, inode, bh);
1406 bh = ext3_append(handle, dir, &block, &retval);
1409 de = (struct ext3_dir_entry_2 *) bh->b_data;
1411 de->rec_len = cpu_to_le16(rlen = blocksize);
1413 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1416 #ifdef CONFIG_EXT3_INDEX
1418 * Returns 0 for success, or a negative error value
1420 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1421 struct inode *inode)
1423 struct dx_frame frames[2], *frame;
1424 struct dx_entry *entries, *at;
1425 struct dx_hash_info hinfo;
1426 struct buffer_head * bh;
1427 struct inode *dir = dentry->d_parent->d_inode;
1428 struct super_block * sb = dir->i_sb;
1429 struct ext3_dir_entry_2 *de;
1432 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1435 entries = frame->entries;
1438 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1441 BUFFER_TRACE(bh, "get_write_access");
1442 err = ext3_journal_get_write_access(handle, bh);
1446 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1447 if (err != -ENOSPC) {
1452 /* Block full, should compress but for now just split */
1453 dxtrace(printk("using %u of %u node entries\n",
1454 dx_get_count(entries), dx_get_limit(entries)));
1455 /* Need to split index? */
1456 if (dx_get_count(entries) == dx_get_limit(entries)) {
1458 unsigned icount = dx_get_count(entries);
1459 int levels = frame - frames;
1460 struct dx_entry *entries2;
1461 struct dx_node *node2;
1462 struct buffer_head *bh2;
1464 if (levels && (dx_get_count(frames->entries) ==
1465 dx_get_limit(frames->entries))) {
1466 ext3_warning(sb, __FUNCTION__,
1467 "Directory index full!\n");
1471 bh2 = ext3_append (handle, dir, &newblock, &err);
1474 node2 = (struct dx_node *)(bh2->b_data);
1475 entries2 = node2->entries;
1476 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1477 node2->fake.inode = 0;
1478 BUFFER_TRACE(frame->bh, "get_write_access");
1479 err = ext3_journal_get_write_access(handle, frame->bh);
1483 unsigned icount1 = icount/2, icount2 = icount - icount1;
1484 unsigned hash2 = dx_get_hash(entries + icount1);
1485 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1487 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1488 err = ext3_journal_get_write_access(handle,
1493 memcpy ((char *) entries2, (char *) (entries + icount1),
1494 icount2 * sizeof(struct dx_entry));
1495 dx_set_count (entries, icount1);
1496 dx_set_count (entries2, icount2);
1497 dx_set_limit (entries2, dx_node_limit(dir));
1499 /* Which index block gets the new entry? */
1500 if (at - entries >= icount1) {
1501 frame->at = at = at - entries - icount1 + entries2;
1502 frame->entries = entries = entries2;
1503 swap(frame->bh, bh2);
1505 dx_insert_block (frames + 0, hash2, newblock);
1506 dxtrace(dx_show_index ("node", frames[1].entries));
1507 dxtrace(dx_show_index ("node",
1508 ((struct dx_node *) bh2->b_data)->entries));
1509 err = ext3_journal_dirty_metadata(handle, bh2);
1514 dxtrace(printk("Creating second level index...\n"));
1515 memcpy((char *) entries2, (char *) entries,
1516 icount * sizeof(struct dx_entry));
1517 dx_set_limit(entries2, dx_node_limit(dir));
1520 dx_set_count(entries, 1);
1521 dx_set_block(entries + 0, newblock);
1522 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1524 /* Add new access path frame */
1526 frame->at = at = at - entries + entries2;
1527 frame->entries = entries = entries2;
1529 err = ext3_journal_get_write_access(handle,
1534 ext3_journal_dirty_metadata(handle, frames[0].bh);
1536 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1539 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1544 ext3_std_error(dir->i_sb, err);
1554 * ext3_delete_entry deletes a directory entry by merging it with the
1557 static int ext3_delete_entry (handle_t *handle,
1559 struct ext3_dir_entry_2 * de_del,
1560 struct buffer_head * bh)
1562 struct ext3_dir_entry_2 * de, * pde;
1567 de = (struct ext3_dir_entry_2 *) bh->b_data;
1568 while (i < bh->b_size) {
1569 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1572 BUFFER_TRACE(bh, "get_write_access");
1573 ext3_journal_get_write_access(handle, bh);
1576 cpu_to_le16(le16_to_cpu(pde->rec_len) +
1577 le16_to_cpu(de->rec_len));
1581 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1582 ext3_journal_dirty_metadata(handle, bh);
1585 i += le16_to_cpu(de->rec_len);
1587 de = (struct ext3_dir_entry_2 *)
1588 ((char *) de + le16_to_cpu(de->rec_len));
1594 * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1595 * do not perform it in these functions. We perform it at the call site,
1598 static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1603 static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1608 static int ext3_add_nondir(handle_t *handle,
1609 struct dentry *dentry, struct inode *inode)
1611 int err = ext3_add_entry(handle, dentry, inode);
1613 ext3_mark_inode_dirty(handle, inode);
1614 d_instantiate(dentry, inode);
1617 ext3_dec_count(handle, inode);
1623 * By the time this is called, we already have created
1624 * the directory cache entry for the new file, but it
1625 * is so far negative - it has no inode.
1627 * If the create succeeds, we fill in the inode information
1628 * with d_instantiate().
1630 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1631 struct nameidata *nd)
1634 struct inode * inode;
1635 int err, retries = 0;
1638 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1639 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1640 2*EXT3_QUOTA_INIT_BLOCKS);
1642 return PTR_ERR(handle);
1644 if (IS_DIRSYNC(dir))
1647 inode = ext3_new_inode (handle, dir, mode);
1648 err = PTR_ERR(inode);
1649 if (!IS_ERR(inode)) {
1650 inode->i_op = &ext3_file_inode_operations;
1651 inode->i_fop = &ext3_file_operations;
1652 ext3_set_aops(inode);
1653 err = ext3_add_nondir(handle, dentry, inode);
1655 ext3_journal_stop(handle);
1656 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1661 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1662 int mode, dev_t rdev)
1665 struct inode *inode;
1666 int err, retries = 0;
1668 if (!new_valid_dev(rdev))
1672 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1673 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1674 2*EXT3_QUOTA_INIT_BLOCKS);
1676 return PTR_ERR(handle);
1678 if (IS_DIRSYNC(dir))
1681 inode = ext3_new_inode (handle, dir, mode);
1682 err = PTR_ERR(inode);
1683 if (!IS_ERR(inode)) {
1684 init_special_inode(inode, inode->i_mode, rdev);
1685 #ifdef CONFIG_EXT3_FS_XATTR
1686 inode->i_op = &ext3_special_inode_operations;
1688 err = ext3_add_nondir(handle, dentry, inode);
1690 ext3_journal_stop(handle);
1691 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1696 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1699 struct inode * inode;
1700 struct buffer_head * dir_block;
1701 struct ext3_dir_entry_2 * de;
1702 int err, retries = 0;
1704 if (dir->i_nlink >= EXT3_LINK_MAX)
1708 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1709 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1710 2*EXT3_QUOTA_INIT_BLOCKS);
1712 return PTR_ERR(handle);
1714 if (IS_DIRSYNC(dir))
1717 inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1718 err = PTR_ERR(inode);
1722 inode->i_op = &ext3_dir_inode_operations;
1723 inode->i_fop = &ext3_dir_operations;
1724 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1725 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1727 inode->i_nlink--; /* is this nlink == 0? */
1728 ext3_mark_inode_dirty(handle, inode);
1732 BUFFER_TRACE(dir_block, "get_write_access");
1733 ext3_journal_get_write_access(handle, dir_block);
1734 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1735 de->inode = cpu_to_le32(inode->i_ino);
1737 de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1738 strcpy (de->name, ".");
1739 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1740 de = (struct ext3_dir_entry_2 *)
1741 ((char *) de + le16_to_cpu(de->rec_len));
1742 de->inode = cpu_to_le32(dir->i_ino);
1743 de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1745 strcpy (de->name, "..");
1746 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1748 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1749 ext3_journal_dirty_metadata(handle, dir_block);
1751 ext3_mark_inode_dirty(handle, inode);
1752 err = ext3_add_entry (handle, dentry, inode);
1755 ext3_mark_inode_dirty(handle, inode);
1760 ext3_update_dx_flag(dir);
1761 ext3_mark_inode_dirty(handle, dir);
1762 d_instantiate(dentry, inode);
1764 ext3_journal_stop(handle);
1765 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1771 * routine to check that the specified directory is empty (for rmdir)
1773 static int empty_dir (struct inode * inode)
1775 unsigned long offset;
1776 struct buffer_head * bh;
1777 struct ext3_dir_entry_2 * de, * de1;
1778 struct super_block * sb;
1782 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1783 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1785 ext3_error(inode->i_sb, __FUNCTION__,
1786 "error %d reading directory #%lu offset 0",
1789 ext3_warning(inode->i_sb, __FUNCTION__,
1790 "bad directory (dir #%lu) - no data block",
1794 de = (struct ext3_dir_entry_2 *) bh->b_data;
1795 de1 = (struct ext3_dir_entry_2 *)
1796 ((char *) de + le16_to_cpu(de->rec_len));
1797 if (le32_to_cpu(de->inode) != inode->i_ino ||
1798 !le32_to_cpu(de1->inode) ||
1799 strcmp (".", de->name) ||
1800 strcmp ("..", de1->name)) {
1801 ext3_warning (inode->i_sb, "empty_dir",
1802 "bad directory (dir #%lu) - no `.' or `..'",
1807 offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1808 de = (struct ext3_dir_entry_2 *)
1809 ((char *) de1 + le16_to_cpu(de1->rec_len));
1810 while (offset < inode->i_size ) {
1812 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1815 bh = ext3_bread (NULL, inode,
1816 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1819 ext3_error(sb, __FUNCTION__,
1820 "error %d reading directory"
1822 err, inode->i_ino, offset);
1823 offset += sb->s_blocksize;
1826 de = (struct ext3_dir_entry_2 *) bh->b_data;
1828 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1829 de = (struct ext3_dir_entry_2 *)(bh->b_data +
1831 offset = (offset | (sb->s_blocksize - 1)) + 1;
1834 if (le32_to_cpu(de->inode)) {
1838 offset += le16_to_cpu(de->rec_len);
1839 de = (struct ext3_dir_entry_2 *)
1840 ((char *) de + le16_to_cpu(de->rec_len));
1846 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1847 * such inodes, starting at the superblock, in case we crash before the
1848 * file is closed/deleted, or in case the inode truncate spans multiple
1849 * transactions and the last transaction is not recovered after a crash.
1851 * At filesystem recovery time, we walk this list deleting unlinked
1852 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1854 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1856 struct super_block *sb = inode->i_sb;
1857 struct ext3_iloc iloc;
1861 if (!list_empty(&EXT3_I(inode)->i_orphan))
1864 /* Orphan handling is only valid for files with data blocks
1865 * being truncated, or files being unlinked. */
1867 /* @@@ FIXME: Observation from aviro:
1868 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1869 * here (on lock_super()), so race with ext3_link() which might bump
1870 * ->i_nlink. For, say it, character device. Not a regular file,
1871 * not a directory, not a symlink and ->i_nlink > 0.
1873 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1874 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1876 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1877 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1881 err = ext3_reserve_inode_write(handle, inode, &iloc);
1885 /* Insert this inode at the head of the on-disk orphan list... */
1886 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1887 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1888 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1889 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1893 /* Only add to the head of the in-memory list if all the
1894 * previous operations succeeded. If the orphan_add is going to
1895 * fail (possibly taking the journal offline), we can't risk
1896 * leaving the inode on the orphan list: stray orphan-list
1897 * entries can cause panics at unmount time.
1899 * This is safe: on error we're going to ignore the orphan list
1900 * anyway on the next recovery. */
1902 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1904 jbd_debug(4, "superblock will point to %ld\n", inode->i_ino);
1905 jbd_debug(4, "orphan inode %ld will point to %d\n",
1906 inode->i_ino, NEXT_ORPHAN(inode));
1909 ext3_std_error(inode->i_sb, err);
1914 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1915 * of such inodes stored on disk, because it is finally being cleaned up.
1917 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1919 struct list_head *prev;
1920 struct ext3_inode_info *ei = EXT3_I(inode);
1921 struct ext3_sb_info *sbi;
1922 unsigned long ino_next;
1923 struct ext3_iloc iloc;
1926 lock_super(inode->i_sb);
1927 if (list_empty(&ei->i_orphan)) {
1928 unlock_super(inode->i_sb);
1932 ino_next = NEXT_ORPHAN(inode);
1933 prev = ei->i_orphan.prev;
1934 sbi = EXT3_SB(inode->i_sb);
1936 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1938 list_del_init(&ei->i_orphan);
1940 /* If we're on an error path, we may not have a valid
1941 * transaction handle with which to update the orphan list on
1942 * disk, but we still need to remove the inode from the linked
1943 * list in memory. */
1947 err = ext3_reserve_inode_write(handle, inode, &iloc);
1951 if (prev == &sbi->s_orphan) {
1952 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1953 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1954 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1957 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1958 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1960 struct ext3_iloc iloc2;
1961 struct inode *i_prev =
1962 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1964 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1965 i_prev->i_ino, ino_next);
1966 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1969 NEXT_ORPHAN(i_prev) = ino_next;
1970 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
1974 NEXT_ORPHAN(inode) = 0;
1975 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
1978 ext3_std_error(inode->i_sb, err);
1980 unlock_super(inode->i_sb);
1988 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
1991 struct inode * inode;
1992 struct buffer_head * bh;
1993 struct ext3_dir_entry_2 * de;
1996 /* Initialize quotas before so that eventual writes go in
1997 * separate transaction */
1998 DQUOT_INIT(dentry->d_inode);
1999 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
2001 return PTR_ERR(handle);
2004 bh = ext3_find_entry (dentry, &de);
2008 if (IS_DIRSYNC(dir))
2011 inode = dentry->d_inode;
2014 if (le32_to_cpu(de->inode) != inode->i_ino)
2017 retval = -ENOTEMPTY;
2018 if (!empty_dir (inode))
2021 retval = ext3_delete_entry(handle, dir, de, bh);
2024 if (inode->i_nlink != 2)
2025 ext3_warning (inode->i_sb, "ext3_rmdir",
2026 "empty directory has nlink!=2 (%d)",
2030 /* There's no need to set i_disksize: the fact that i_nlink is
2031 * zero will ensure that the right thing happens during any
2034 ext3_orphan_add(handle, inode);
2035 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2036 ext3_mark_inode_dirty(handle, inode);
2038 ext3_update_dx_flag(dir);
2039 ext3_mark_inode_dirty(handle, dir);
2042 ext3_journal_stop(handle);
2047 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2050 struct inode * inode;
2051 struct buffer_head * bh;
2052 struct ext3_dir_entry_2 * de;
2055 /* Initialize quotas before so that eventual writes go
2056 * in separate transaction */
2057 DQUOT_INIT(dentry->d_inode);
2058 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
2060 return PTR_ERR(handle);
2062 if (IS_DIRSYNC(dir))
2066 bh = ext3_find_entry (dentry, &de);
2070 inode = dentry->d_inode;
2073 if (le32_to_cpu(de->inode) != inode->i_ino)
2076 if (!inode->i_nlink) {
2077 ext3_warning (inode->i_sb, "ext3_unlink",
2078 "Deleting nonexistent file (%lu), %d",
2079 inode->i_ino, inode->i_nlink);
2082 retval = ext3_delete_entry(handle, dir, de, bh);
2085 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2086 ext3_update_dx_flag(dir);
2087 ext3_mark_inode_dirty(handle, dir);
2089 if (!inode->i_nlink)
2090 ext3_orphan_add(handle, inode);
2091 inode->i_ctime = dir->i_ctime;
2092 ext3_mark_inode_dirty(handle, inode);
2096 ext3_journal_stop(handle);
2101 static int ext3_symlink (struct inode * dir,
2102 struct dentry *dentry, const char * symname)
2105 struct inode * inode;
2106 int l, err, retries = 0;
2108 l = strlen(symname)+1;
2109 if (l > dir->i_sb->s_blocksize)
2110 return -ENAMETOOLONG;
2113 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2114 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2115 2*EXT3_QUOTA_INIT_BLOCKS);
2117 return PTR_ERR(handle);
2119 if (IS_DIRSYNC(dir))
2122 inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2123 err = PTR_ERR(inode);
2127 if (l > sizeof (EXT3_I(inode)->i_data)) {
2128 inode->i_op = &ext3_symlink_inode_operations;
2129 ext3_set_aops(inode);
2131 * page_symlink() calls into ext3_prepare/commit_write.
2132 * We have a transaction open. All is sweetness. It also sets
2133 * i_size in generic_commit_write().
2135 err = page_symlink(inode, symname, l);
2137 ext3_dec_count(handle, inode);
2138 ext3_mark_inode_dirty(handle, inode);
2143 inode->i_op = &ext3_fast_symlink_inode_operations;
2144 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2145 inode->i_size = l-1;
2147 EXT3_I(inode)->i_disksize = inode->i_size;
2148 err = ext3_add_nondir(handle, dentry, inode);
2150 ext3_journal_stop(handle);
2151 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2156 static int ext3_link (struct dentry * old_dentry,
2157 struct inode * dir, struct dentry *dentry)
2160 struct inode *inode = old_dentry->d_inode;
2161 int err, retries = 0;
2163 if (inode->i_nlink >= EXT3_LINK_MAX)
2167 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2168 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2170 return PTR_ERR(handle);
2172 if (IS_DIRSYNC(dir))
2175 inode->i_ctime = CURRENT_TIME;
2176 ext3_inc_count(handle, inode);
2177 atomic_inc(&inode->i_count);
2179 err = ext3_add_nondir(handle, dentry, inode);
2180 ext3_journal_stop(handle);
2181 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2186 #define PARENT_INO(buffer) \
2187 ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2188 le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2191 * Anybody can rename anything with this: the permission checks are left to the
2192 * higher-level routines.
2194 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2195 struct inode * new_dir,struct dentry *new_dentry)
2198 struct inode * old_inode, * new_inode;
2199 struct buffer_head * old_bh, * new_bh, * dir_bh;
2200 struct ext3_dir_entry_2 * old_de, * new_de;
2203 old_bh = new_bh = dir_bh = NULL;
2205 /* Initialize quotas before so that eventual writes go
2206 * in separate transaction */
2207 if (new_dentry->d_inode)
2208 DQUOT_INIT(new_dentry->d_inode);
2209 handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS +
2210 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2212 return PTR_ERR(handle);
2214 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2217 old_bh = ext3_find_entry (old_dentry, &old_de);
2219 * Check for inode number is _not_ due to possible IO errors.
2220 * We might rmdir the source, keep it as pwd of some process
2221 * and merrily kill the link to whatever was created under the
2222 * same name. Goodbye sticky bit ;-<
2224 old_inode = old_dentry->d_inode;
2226 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2229 new_inode = new_dentry->d_inode;
2230 new_bh = ext3_find_entry (new_dentry, &new_de);
2237 if (S_ISDIR(old_inode->i_mode)) {
2239 retval = -ENOTEMPTY;
2240 if (!empty_dir (new_inode))
2244 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2247 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2250 if (!new_inode && new_dir!=old_dir &&
2251 new_dir->i_nlink >= EXT3_LINK_MAX)
2255 retval = ext3_add_entry (handle, new_dentry, old_inode);
2259 BUFFER_TRACE(new_bh, "get write access");
2260 ext3_journal_get_write_access(handle, new_bh);
2261 new_de->inode = le32_to_cpu(old_inode->i_ino);
2262 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2263 EXT3_FEATURE_INCOMPAT_FILETYPE))
2264 new_de->file_type = old_de->file_type;
2265 new_dir->i_version++;
2266 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2267 ext3_journal_dirty_metadata(handle, new_bh);
2273 * Like most other Unix systems, set the ctime for inodes on a
2276 old_inode->i_ctime = CURRENT_TIME;
2277 ext3_mark_inode_dirty(handle, old_inode);
2282 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2283 old_de->name_len != old_dentry->d_name.len ||
2284 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2285 (retval = ext3_delete_entry(handle, old_dir,
2286 old_de, old_bh)) == -ENOENT) {
2287 /* old_de could have moved from under us during htree split, so
2288 * make sure that we are deleting the right entry. We might
2289 * also be pointing to a stale entry in the unused part of
2290 * old_bh so just checking inum and the name isn't enough. */
2291 struct buffer_head *old_bh2;
2292 struct ext3_dir_entry_2 *old_de2;
2294 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2296 retval = ext3_delete_entry(handle, old_dir,
2302 ext3_warning(old_dir->i_sb, "ext3_rename",
2303 "Deleting old file (%lu), %d, error=%d",
2304 old_dir->i_ino, old_dir->i_nlink, retval);
2308 new_inode->i_nlink--;
2309 new_inode->i_ctime = CURRENT_TIME;
2311 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
2312 ext3_update_dx_flag(old_dir);
2314 BUFFER_TRACE(dir_bh, "get_write_access");
2315 ext3_journal_get_write_access(handle, dir_bh);
2316 PARENT_INO(dir_bh->b_data) = le32_to_cpu(new_dir->i_ino);
2317 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2318 ext3_journal_dirty_metadata(handle, dir_bh);
2321 new_inode->i_nlink--;
2324 ext3_update_dx_flag(new_dir);
2325 ext3_mark_inode_dirty(handle, new_dir);
2328 ext3_mark_inode_dirty(handle, old_dir);
2330 ext3_mark_inode_dirty(handle, new_inode);
2331 if (!new_inode->i_nlink)
2332 ext3_orphan_add(handle, new_inode);
2340 ext3_journal_stop(handle);
2345 * directories can handle most operations...
2347 struct inode_operations ext3_dir_inode_operations = {
2348 .create = ext3_create,
2349 .lookup = ext3_lookup,
2351 .unlink = ext3_unlink,
2352 .symlink = ext3_symlink,
2353 .mkdir = ext3_mkdir,
2354 .rmdir = ext3_rmdir,
2355 .mknod = ext3_mknod,
2356 .rename = ext3_rename,
2357 .setattr = ext3_setattr,
2358 .setxattr = ext3_setxattr,
2359 .getxattr = ext3_getxattr,
2360 .listxattr = ext3_listxattr,
2361 .removexattr = ext3_removexattr,
2362 .permission = ext3_permission,
2365 struct inode_operations ext3_special_inode_operations = {
2366 .setattr = ext3_setattr,
2367 .setxattr = ext3_setxattr,
2368 .getxattr = ext3_getxattr,
2369 .listxattr = ext3_listxattr,
2370 .removexattr = ext3_removexattr,
2371 .permission = ext3_permission,