2 * linux/fs/ext4/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/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/ext4_fs.h>
32 #include <linux/ext4_jbd2.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 *ext4_append(handle_t *handle,
56 struct buffer_head *bh;
58 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
60 if ((bh = ext4_bread(handle, inode, *block, 1, err))) {
61 inode->i_size += inode->i_sb->s_blocksize;
62 EXT4_I(inode)->i_disksize = inode->i_size;
63 ext4_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;
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 ext4_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 ext4_dir_entry_2 *dx_move_dirents (char *from, char *to,
167 struct dx_map_entry *offsets, int count);
168 static struct ext4_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 ext4_htree_next_block(struct inode *dir, __u32 hash,
171 struct dx_frame *frame,
172 struct dx_frame *frames,
174 static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
175 struct ext4_dir_entry_2 **res_dir, int *err);
176 static int ext4_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 - EXT4_DIR_REC_LEN(1) -
227 EXT4_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 - EXT4_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++) {
246 printk("%x->%u ", i? dx_get_hash(entries + i) :
247 0, dx_get_block(entries + i));
259 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_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 ext4fs_dirhash(de->name, de->name_len, &h);
277 printk(":%x.%u ", h.hash,
278 ((char *) de - base));
280 space += EXT4_DIR_REC_LEN(de->name_len);
283 de = ext4_next_entry(de);
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 = ext4_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 ext4_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 = ext4_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 ext4_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 = EXT4_SB(dir->i_sb)->s_hash_seed;
359 ext4fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
362 if (root->info.unused_flags & 1) {
363 ext4_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 ext4_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 ext4_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 ext4_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 = ext4_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 ext4_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 ext4_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 ext4_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 = ext4_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 ext4_dir_entry_2 *ext4_next_entry(struct ext4_dir_entry_2 *p)
554 return (struct ext4_dir_entry_2 *)((char *)p +
555 ext4_rec_len_from_disk(p->rec_len));
559 * This function fills a red-black tree with information from a
560 * directory block. It returns the number directory entries loaded
561 * into the tree. If there is an error it is returned in err.
563 static int htree_dirblock_to_tree(struct file *dir_file,
564 struct inode *dir, int block,
565 struct dx_hash_info *hinfo,
566 __u32 start_hash, __u32 start_minor_hash)
568 struct buffer_head *bh;
569 struct ext4_dir_entry_2 *de, *top;
572 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
573 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
576 de = (struct ext4_dir_entry_2 *) bh->b_data;
577 top = (struct ext4_dir_entry_2 *) ((char *) de +
578 dir->i_sb->s_blocksize -
579 EXT4_DIR_REC_LEN(0));
580 for (; de < top; de = ext4_next_entry(de)) {
581 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
582 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
583 +((char *)de - bh->b_data))) {
584 /* On error, skip the f_pos to the next block. */
585 dir_file->f_pos = (dir_file->f_pos |
586 (dir->i_sb->s_blocksize - 1)) + 1;
590 ext4fs_dirhash(de->name, de->name_len, hinfo);
591 if ((hinfo->hash < start_hash) ||
592 ((hinfo->hash == start_hash) &&
593 (hinfo->minor_hash < start_minor_hash)))
597 if ((err = ext4_htree_store_dirent(dir_file,
598 hinfo->hash, hinfo->minor_hash, de)) != 0) {
610 * This function fills a red-black tree with information from a
611 * directory. We start scanning the directory in hash order, starting
612 * at start_hash and start_minor_hash.
614 * This function returns the number of entries inserted into the tree,
615 * or a negative error code.
617 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
618 __u32 start_minor_hash, __u32 *next_hash)
620 struct dx_hash_info hinfo;
621 struct ext4_dir_entry_2 *de;
622 struct dx_frame frames[2], *frame;
629 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
631 dir = dir_file->f_path.dentry->d_inode;
632 if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
633 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
634 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
635 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
636 start_hash, start_minor_hash);
640 hinfo.hash = start_hash;
641 hinfo.minor_hash = 0;
642 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
646 /* Add '.' and '..' from the htree header */
647 if (!start_hash && !start_minor_hash) {
648 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
649 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
653 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
654 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
655 de = ext4_next_entry(de);
656 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
662 block = dx_get_block(frame->at);
663 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
664 start_hash, start_minor_hash);
671 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
672 frame, frames, &hashval);
673 *next_hash = hashval;
679 * Stop if: (a) there are no more entries, or
680 * (b) we have inserted at least one entry and the
681 * next hash value is not a continuation
684 (count && ((hashval & 1) == 0)))
688 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
698 * Directory block splitting, compacting
702 * Create map of hash values, offsets, and sizes, stored at end of block.
703 * Returns number of entries mapped.
705 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
706 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
709 char *base = (char *) de;
710 struct dx_hash_info h = *hinfo;
712 while ((char *) de < base + size)
714 if (de->name_len && de->inode) {
715 ext4fs_dirhash(de->name, de->name_len, &h);
717 map_tail->hash = h.hash;
718 map_tail->offs = (u16) ((char *) de - base);
719 map_tail->size = le16_to_cpu(de->rec_len);
723 /* XXX: do we need to check rec_len == 0 case? -Chris */
724 de = ext4_next_entry(de);
729 /* Sort map by hash value */
730 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
732 struct dx_map_entry *p, *q, *top = map + count - 1;
734 /* Combsort until bubble sort doesn't suck */
737 if (count - 9 < 2) /* 9, 10 -> 11 */
739 for (p = top, q = p - count; q >= map; p--, q--)
740 if (p->hash < q->hash)
743 /* Garden variety bubble sort */
748 if (q[1].hash >= q[0].hash)
756 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
758 struct dx_entry *entries = frame->entries;
759 struct dx_entry *old = frame->at, *new = old + 1;
760 int count = dx_get_count(entries);
762 assert(count < dx_get_limit(entries));
763 assert(old < entries + count);
764 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
765 dx_set_hash(new, hash);
766 dx_set_block(new, block);
767 dx_set_count(entries, count + 1);
770 static void ext4_update_dx_flag(struct inode *inode)
772 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
773 EXT4_FEATURE_COMPAT_DIR_INDEX))
774 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
778 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
780 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
781 * `de != NULL' is guaranteed by caller.
783 static inline int ext4_match (int len, const char * const name,
784 struct ext4_dir_entry_2 * de)
786 if (len != de->name_len)
790 return !memcmp(name, de->name, len);
794 * Returns 0 if not found, -1 on failure, and 1 on success
796 static inline int search_dirblock(struct buffer_head * bh,
798 struct dentry *dentry,
799 unsigned long offset,
800 struct ext4_dir_entry_2 ** res_dir)
802 struct ext4_dir_entry_2 * de;
805 const char *name = dentry->d_name.name;
806 int namelen = dentry->d_name.len;
808 de = (struct ext4_dir_entry_2 *) bh->b_data;
809 dlimit = bh->b_data + dir->i_sb->s_blocksize;
810 while ((char *) de < dlimit) {
811 /* this code is executed quadratically often */
812 /* do minimal checking `by hand' */
814 if ((char *) de + namelen <= dlimit &&
815 ext4_match (namelen, name, de)) {
816 /* found a match - just to be sure, do a full check */
817 if (!ext4_check_dir_entry("ext4_find_entry",
818 dir, de, bh, offset))
823 /* prevent looping on a bad block */
824 de_len = ext4_rec_len_from_disk(de->rec_len);
828 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
837 * finds an entry in the specified directory with the wanted name. It
838 * returns the cache buffer in which the entry was found, and the entry
839 * itself (as a parameter - res_dir). It does NOT read the inode of the
840 * entry - you'll have to do that yourself if you want to.
842 * The returned buffer_head has ->b_count elevated. The caller is expected
843 * to brelse() it when appropriate.
845 static struct buffer_head * ext4_find_entry (struct dentry *dentry,
846 struct ext4_dir_entry_2 ** res_dir)
848 struct super_block * sb;
849 struct buffer_head * bh_use[NAMEI_RA_SIZE];
850 struct buffer_head * bh, *ret = NULL;
851 unsigned long start, block, b;
852 int ra_max = 0; /* Number of bh's in the readahead
854 int ra_ptr = 0; /* Current index into readahead
858 struct inode *dir = dentry->d_parent->d_inode;
865 blocksize = sb->s_blocksize;
866 namelen = dentry->d_name.len;
867 name = dentry->d_name.name;
868 if (namelen > EXT4_NAME_LEN)
871 bh = ext4_dx_find_entry(dentry, res_dir, &err);
873 * On success, or if the error was file not found,
874 * return. Otherwise, fall back to doing a search the
877 if (bh || (err != ERR_BAD_DX_DIR))
879 dxtrace(printk("ext4_find_entry: dx failed, falling back\n"));
881 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
882 start = EXT4_I(dir)->i_dir_start_lookup;
883 if (start >= nblocks)
889 * We deal with the read-ahead logic here.
891 if (ra_ptr >= ra_max) {
892 /* Refill the readahead buffer */
895 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
897 * Terminate if we reach the end of the
898 * directory and must wrap, or if our
899 * search has finished at this block.
901 if (b >= nblocks || (num && block == start)) {
902 bh_use[ra_max] = NULL;
906 bh = ext4_getblk(NULL, dir, b++, 0, &err);
909 ll_rw_block(READ_META, 1, &bh);
912 if ((bh = bh_use[ra_ptr++]) == NULL)
915 if (!buffer_uptodate(bh)) {
916 /* read error, skip block & hope for the best */
917 ext4_error(sb, __FUNCTION__, "reading directory #%lu "
918 "offset %lu", dir->i_ino, block);
922 i = search_dirblock(bh, dir, dentry,
923 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
925 EXT4_I(dir)->i_dir_start_lookup = block;
927 goto cleanup_and_exit;
931 goto cleanup_and_exit;
934 if (++block >= nblocks)
936 } while (block != start);
939 * If the directory has grown while we were searching, then
940 * search the last part of the directory before giving up.
943 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
944 if (block < nblocks) {
950 /* Clean up the read-ahead blocks */
951 for (; ra_ptr < ra_max; ra_ptr++)
952 brelse (bh_use[ra_ptr]);
956 static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
957 struct ext4_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 ext4_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 = ext4_bread (NULL,dir, block, 0, err)))
987 de = (struct ext4_dir_entry_2 *) bh->b_data;
988 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
989 EXT4_DIR_REC_LEN(0));
990 for (; de < top; de = ext4_next_entry(de))
991 if (ext4_match (namelen, name, de)) {
992 if (!ext4_check_dir_entry("ext4_find_entry",
994 (block<<EXT4_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 = ext4_htree_next_block(dir, hash, frame,
1009 ext4_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);
1024 static struct dentry *ext4_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1026 struct inode * inode;
1027 struct ext4_dir_entry_2 * de;
1028 struct buffer_head * bh;
1030 if (dentry->d_name.len > EXT4_NAME_LEN)
1031 return ERR_PTR(-ENAMETOOLONG);
1033 bh = ext4_find_entry(dentry, &de);
1036 unsigned long ino = le32_to_cpu(de->inode);
1038 if (!ext4_valid_inum(dir->i_sb, ino)) {
1039 ext4_error(dir->i_sb, "ext4_lookup",
1040 "bad inode number: %lu", ino);
1043 inode = iget(dir->i_sb, ino);
1046 return ERR_PTR(-EACCES);
1048 if (is_bad_inode(inode)) {
1050 return ERR_PTR(-ENOENT);
1053 return d_splice_alias(inode, dentry);
1057 struct dentry *ext4_get_parent(struct dentry *child)
1060 struct dentry *parent;
1061 struct inode *inode;
1062 struct dentry dotdot;
1063 struct ext4_dir_entry_2 * de;
1064 struct buffer_head *bh;
1066 dotdot.d_name.name = "..";
1067 dotdot.d_name.len = 2;
1068 dotdot.d_parent = child; /* confusing, isn't it! */
1070 bh = ext4_find_entry(&dotdot, &de);
1073 return ERR_PTR(-ENOENT);
1074 ino = le32_to_cpu(de->inode);
1077 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1078 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1079 "bad inode number: %lu", ino);
1082 inode = iget(child->d_inode->i_sb, ino);
1085 return ERR_PTR(-EACCES);
1087 if (is_bad_inode(inode)) {
1089 return ERR_PTR(-ENOENT);
1092 parent = d_alloc_anon(inode);
1095 parent = ERR_PTR(-ENOMEM);
1101 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1102 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1103 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1104 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1105 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1106 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1107 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1108 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
1111 static inline void ext4_set_de_type(struct super_block *sb,
1112 struct ext4_dir_entry_2 *de,
1114 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1115 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1119 * Move count entries from end of map between two memory locations.
1120 * Returns pointer to last entry moved.
1122 static struct ext4_dir_entry_2 *
1123 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1125 unsigned rec_len = 0;
1128 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1129 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1130 memcpy (to, de, rec_len);
1131 ((struct ext4_dir_entry_2 *) to)->rec_len =
1132 ext4_rec_len_to_disk(rec_len);
1137 return (struct ext4_dir_entry_2 *) (to - rec_len);
1141 * Compact each dir entry in the range to the minimal rec_len.
1142 * Returns pointer to last entry in range.
1144 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
1146 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1147 unsigned rec_len = 0;
1150 while ((char*)de < base + size) {
1151 next = ext4_next_entry(de);
1152 if (de->inode && de->name_len) {
1153 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1155 memmove(to, de, rec_len);
1156 to->rec_len = ext4_rec_len_to_disk(rec_len);
1158 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1166 * Split a full leaf block to make room for a new dir entry.
1167 * Allocate a new block, and move entries so that they are approx. equally full.
1168 * Returns pointer to de in block into which the new entry will be inserted.
1170 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1171 struct buffer_head **bh,struct dx_frame *frame,
1172 struct dx_hash_info *hinfo, int *error)
1174 unsigned blocksize = dir->i_sb->s_blocksize;
1175 unsigned count, continued;
1176 struct buffer_head *bh2;
1179 struct dx_map_entry *map;
1180 char *data1 = (*bh)->b_data, *data2;
1181 unsigned split, move, size, i;
1182 struct ext4_dir_entry_2 *de = NULL, *de2;
1185 bh2 = ext4_append (handle, dir, &newblock, &err);
1192 BUFFER_TRACE(*bh, "get_write_access");
1193 err = ext4_journal_get_write_access(handle, *bh);
1197 BUFFER_TRACE(frame->bh, "get_write_access");
1198 err = ext4_journal_get_write_access(handle, frame->bh);
1202 data2 = bh2->b_data;
1204 /* create map in the end of data2 block */
1205 map = (struct dx_map_entry *) (data2 + blocksize);
1206 count = dx_make_map ((struct ext4_dir_entry_2 *) data1,
1207 blocksize, hinfo, map);
1209 dx_sort_map (map, count);
1210 /* Split the existing block in the middle, size-wise */
1213 for (i = count-1; i >= 0; i--) {
1214 /* is more than half of this entry in 2nd half of the block? */
1215 if (size + map[i].size/2 > blocksize/2)
1217 size += map[i].size;
1220 /* map index at which we will split */
1221 split = count - move;
1222 hash2 = map[split].hash;
1223 continued = hash2 == map[split - 1].hash;
1224 dxtrace(printk("Split block %i at %x, %i/%i\n",
1225 dx_get_block(frame->at), hash2, split, count-split));
1227 /* Fancy dance to stay within two buffers */
1228 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1229 de = dx_pack_dirents(data1,blocksize);
1230 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
1231 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2);
1232 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1233 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1235 /* Which block gets the new entry? */
1236 if (hinfo->hash >= hash2)
1241 dx_insert_block (frame, hash2 + continued, newblock);
1242 err = ext4_journal_dirty_metadata (handle, bh2);
1245 err = ext4_journal_dirty_metadata (handle, frame->bh);
1249 dxtrace(dx_show_index ("frame", frame->entries));
1256 ext4_std_error(dir->i_sb, err);
1263 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1264 * it points to a directory entry which is guaranteed to be large
1265 * enough for new directory entry. If de is NULL, then
1266 * add_dirent_to_buf will attempt search the directory block for
1267 * space. It will return -ENOSPC if no space is available, and -EIO
1268 * and -EEXIST if directory entry already exists.
1270 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1271 * all other cases bh is released.
1273 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1274 struct inode *inode, struct ext4_dir_entry_2 *de,
1275 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 unsigned long offset = 0;
1281 unsigned short reclen;
1282 int nlen, rlen, err;
1285 reclen = EXT4_DIR_REC_LEN(namelen);
1287 de = (struct ext4_dir_entry_2 *)bh->b_data;
1288 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1289 while ((char *) de <= top) {
1290 if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1295 if (ext4_match (namelen, name, de)) {
1299 nlen = EXT4_DIR_REC_LEN(de->name_len);
1300 rlen = ext4_rec_len_from_disk(de->rec_len);
1301 if ((de->inode? rlen - nlen: rlen) >= reclen)
1303 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1306 if ((char *) de > top)
1309 BUFFER_TRACE(bh, "get_write_access");
1310 err = ext4_journal_get_write_access(handle, bh);
1312 ext4_std_error(dir->i_sb, err);
1317 /* By now the buffer is marked for journaling */
1318 nlen = EXT4_DIR_REC_LEN(de->name_len);
1319 rlen = ext4_rec_len_from_disk(de->rec_len);
1321 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1322 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen);
1323 de->rec_len = ext4_rec_len_to_disk(nlen);
1326 de->file_type = EXT4_FT_UNKNOWN;
1328 de->inode = cpu_to_le32(inode->i_ino);
1329 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1332 de->name_len = namelen;
1333 memcpy (de->name, name, namelen);
1335 * XXX shouldn't update any times until successful
1336 * completion of syscall, but too many callers depend
1339 * XXX similarly, too many callers depend on
1340 * ext4_new_inode() setting the times, but error
1341 * recovery deletes the inode, so the worst that can
1342 * happen is that the times are slightly out of date
1343 * and/or different from the directory change time.
1345 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1346 ext4_update_dx_flag(dir);
1348 ext4_mark_inode_dirty(handle, dir);
1349 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1350 err = ext4_journal_dirty_metadata(handle, bh);
1352 ext4_std_error(dir->i_sb, err);
1358 * This converts a one block unindexed directory to a 3 block indexed
1359 * directory, and adds the dentry to the indexed directory.
1361 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1362 struct inode *inode, struct buffer_head *bh)
1364 struct inode *dir = dentry->d_parent->d_inode;
1365 const char *name = dentry->d_name.name;
1366 int namelen = dentry->d_name.len;
1367 struct buffer_head *bh2;
1368 struct dx_root *root;
1369 struct dx_frame frames[2], *frame;
1370 struct dx_entry *entries;
1371 struct ext4_dir_entry_2 *de, *de2;
1376 struct dx_hash_info hinfo;
1378 struct fake_dirent *fde;
1380 blocksize = dir->i_sb->s_blocksize;
1381 dxtrace(printk("Creating index\n"));
1382 retval = ext4_journal_get_write_access(handle, bh);
1384 ext4_std_error(dir->i_sb, retval);
1388 root = (struct dx_root *) bh->b_data;
1390 bh2 = ext4_append (handle, dir, &block, &retval);
1395 EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1396 data1 = bh2->b_data;
1398 /* The 0th block becomes the root, move the dirents out */
1399 fde = &root->dotdot;
1400 de = (struct ext4_dir_entry_2 *)((char *)fde +
1401 ext4_rec_len_from_disk(fde->rec_len));
1402 len = ((char *) root) + blocksize - (char *) de;
1403 memcpy (data1, de, len);
1404 de = (struct ext4_dir_entry_2 *) data1;
1406 while ((char *)(de2 = ext4_next_entry(de)) < top)
1408 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
1409 /* Initialize the root; the dot dirents already exist */
1410 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1411 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2));
1412 memset (&root->info, 0, sizeof(root->info));
1413 root->info.info_length = sizeof(root->info);
1414 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1415 entries = root->entries;
1416 dx_set_block (entries, 1);
1417 dx_set_count (entries, 1);
1418 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1420 /* Initialize as for dx_probe */
1421 hinfo.hash_version = root->info.hash_version;
1422 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1423 ext4fs_dirhash(name, namelen, &hinfo);
1425 frame->entries = entries;
1426 frame->at = entries;
1429 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1430 dx_release (frames);
1434 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1440 * adds a file entry to the specified directory, using the same
1441 * semantics as ext4_find_entry(). It returns NULL if it failed.
1443 * NOTE!! The inode part of 'de' is left at 0 - which means you
1444 * may not sleep between calling this and putting something into
1445 * the entry, as someone else might have used it while you slept.
1447 static int ext4_add_entry (handle_t *handle, struct dentry *dentry,
1448 struct inode *inode)
1450 struct inode *dir = dentry->d_parent->d_inode;
1451 unsigned long offset;
1452 struct buffer_head * bh;
1453 struct ext4_dir_entry_2 *de;
1454 struct super_block * sb;
1461 blocksize = sb->s_blocksize;
1462 if (!dentry->d_name.len)
1465 retval = ext4_dx_add_entry(handle, dentry, inode);
1466 if (!retval || (retval != ERR_BAD_DX_DIR))
1468 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1470 ext4_mark_inode_dirty(handle, dir);
1472 blocks = dir->i_size >> sb->s_blocksize_bits;
1473 for (block = 0, offset = 0; block < blocks; block++) {
1474 bh = ext4_bread(handle, dir, block, 0, &retval);
1477 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1478 if (retval != -ENOSPC)
1481 if (blocks == 1 && !dx_fallback &&
1482 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1483 return make_indexed_dir(handle, dentry, inode, bh);
1486 bh = ext4_append(handle, dir, &block, &retval);
1489 de = (struct ext4_dir_entry_2 *) bh->b_data;
1491 de->rec_len = ext4_rec_len_to_disk(blocksize);
1492 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1496 * Returns 0 for success, or a negative error value
1498 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1499 struct inode *inode)
1501 struct dx_frame frames[2], *frame;
1502 struct dx_entry *entries, *at;
1503 struct dx_hash_info hinfo;
1504 struct buffer_head * bh;
1505 struct inode *dir = dentry->d_parent->d_inode;
1506 struct super_block * sb = dir->i_sb;
1507 struct ext4_dir_entry_2 *de;
1510 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1513 entries = frame->entries;
1516 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1519 BUFFER_TRACE(bh, "get_write_access");
1520 err = ext4_journal_get_write_access(handle, bh);
1524 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1525 if (err != -ENOSPC) {
1530 /* Block full, should compress but for now just split */
1531 dxtrace(printk("using %u of %u node entries\n",
1532 dx_get_count(entries), dx_get_limit(entries)));
1533 /* Need to split index? */
1534 if (dx_get_count(entries) == dx_get_limit(entries)) {
1536 unsigned icount = dx_get_count(entries);
1537 int levels = frame - frames;
1538 struct dx_entry *entries2;
1539 struct dx_node *node2;
1540 struct buffer_head *bh2;
1542 if (levels && (dx_get_count(frames->entries) ==
1543 dx_get_limit(frames->entries))) {
1544 ext4_warning(sb, __FUNCTION__,
1545 "Directory index full!");
1549 bh2 = ext4_append (handle, dir, &newblock, &err);
1552 node2 = (struct dx_node *)(bh2->b_data);
1553 entries2 = node2->entries;
1554 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize);
1555 node2->fake.inode = 0;
1556 BUFFER_TRACE(frame->bh, "get_write_access");
1557 err = ext4_journal_get_write_access(handle, frame->bh);
1561 unsigned icount1 = icount/2, icount2 = icount - icount1;
1562 unsigned hash2 = dx_get_hash(entries + icount1);
1563 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1565 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1566 err = ext4_journal_get_write_access(handle,
1571 memcpy ((char *) entries2, (char *) (entries + icount1),
1572 icount2 * sizeof(struct dx_entry));
1573 dx_set_count (entries, icount1);
1574 dx_set_count (entries2, icount2);
1575 dx_set_limit (entries2, dx_node_limit(dir));
1577 /* Which index block gets the new entry? */
1578 if (at - entries >= icount1) {
1579 frame->at = at = at - entries - icount1 + entries2;
1580 frame->entries = entries = entries2;
1581 swap(frame->bh, bh2);
1583 dx_insert_block (frames + 0, hash2, newblock);
1584 dxtrace(dx_show_index ("node", frames[1].entries));
1585 dxtrace(dx_show_index ("node",
1586 ((struct dx_node *) bh2->b_data)->entries));
1587 err = ext4_journal_dirty_metadata(handle, bh2);
1592 dxtrace(printk("Creating second level index...\n"));
1593 memcpy((char *) entries2, (char *) entries,
1594 icount * sizeof(struct dx_entry));
1595 dx_set_limit(entries2, dx_node_limit(dir));
1598 dx_set_count(entries, 1);
1599 dx_set_block(entries + 0, newblock);
1600 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1602 /* Add new access path frame */
1604 frame->at = at = at - entries + entries2;
1605 frame->entries = entries = entries2;
1607 err = ext4_journal_get_write_access(handle,
1612 ext4_journal_dirty_metadata(handle, frames[0].bh);
1614 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1617 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1622 ext4_std_error(dir->i_sb, err);
1631 * ext4_delete_entry deletes a directory entry by merging it with the
1634 static int ext4_delete_entry (handle_t *handle,
1636 struct ext4_dir_entry_2 * de_del,
1637 struct buffer_head * bh)
1639 struct ext4_dir_entry_2 * de, * pde;
1644 de = (struct ext4_dir_entry_2 *) bh->b_data;
1645 while (i < bh->b_size) {
1646 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1649 BUFFER_TRACE(bh, "get_write_access");
1650 ext4_journal_get_write_access(handle, bh);
1652 pde->rec_len = ext4_rec_len_to_disk(
1653 ext4_rec_len_from_disk(pde->rec_len) +
1654 ext4_rec_len_from_disk(de->rec_len));
1658 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1659 ext4_journal_dirty_metadata(handle, bh);
1662 i += ext4_rec_len_from_disk(de->rec_len);
1664 de = ext4_next_entry(de);
1670 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1671 * since this indicates that nlinks count was previously 1.
1673 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1676 if (is_dx(inode) && inode->i_nlink > 1) {
1677 /* limit is 16-bit i_links_count */
1678 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1680 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1681 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1687 * If a directory had nlink == 1, then we should let it be 1. This indicates
1688 * directory has >EXT4_LINK_MAX subdirs.
1690 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1693 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1698 static int ext4_add_nondir(handle_t *handle,
1699 struct dentry *dentry, struct inode *inode)
1701 int err = ext4_add_entry(handle, dentry, inode);
1703 ext4_mark_inode_dirty(handle, inode);
1704 d_instantiate(dentry, inode);
1713 * By the time this is called, we already have created
1714 * the directory cache entry for the new file, but it
1715 * is so far negative - it has no inode.
1717 * If the create succeeds, we fill in the inode information
1718 * with d_instantiate().
1720 static int ext4_create (struct inode * dir, struct dentry * dentry, int mode,
1721 struct nameidata *nd)
1724 struct inode * inode;
1725 int err, retries = 0;
1728 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1729 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1730 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1732 return PTR_ERR(handle);
1734 if (IS_DIRSYNC(dir))
1737 inode = ext4_new_inode (handle, dir, mode);
1738 err = PTR_ERR(inode);
1739 if (!IS_ERR(inode)) {
1740 inode->i_op = &ext4_file_inode_operations;
1741 inode->i_fop = &ext4_file_operations;
1742 ext4_set_aops(inode);
1743 err = ext4_add_nondir(handle, dentry, inode);
1745 ext4_journal_stop(handle);
1746 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1751 static int ext4_mknod (struct inode * dir, struct dentry *dentry,
1752 int mode, dev_t rdev)
1755 struct inode *inode;
1756 int err, retries = 0;
1758 if (!new_valid_dev(rdev))
1762 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1763 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1764 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1766 return PTR_ERR(handle);
1768 if (IS_DIRSYNC(dir))
1771 inode = ext4_new_inode (handle, dir, mode);
1772 err = PTR_ERR(inode);
1773 if (!IS_ERR(inode)) {
1774 init_special_inode(inode, inode->i_mode, rdev);
1775 #ifdef CONFIG_EXT4DEV_FS_XATTR
1776 inode->i_op = &ext4_special_inode_operations;
1778 err = ext4_add_nondir(handle, dentry, inode);
1780 ext4_journal_stop(handle);
1781 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1786 static int ext4_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1789 struct inode * inode;
1790 struct buffer_head * dir_block;
1791 struct ext4_dir_entry_2 * de;
1792 int err, retries = 0;
1794 if (EXT4_DIR_LINK_MAX(dir))
1798 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1799 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1800 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1802 return PTR_ERR(handle);
1804 if (IS_DIRSYNC(dir))
1807 inode = ext4_new_inode (handle, dir, S_IFDIR | mode);
1808 err = PTR_ERR(inode);
1812 inode->i_op = &ext4_dir_inode_operations;
1813 inode->i_fop = &ext4_dir_operations;
1814 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1815 dir_block = ext4_bread (handle, inode, 0, 1, &err);
1817 ext4_dec_count(handle, inode); /* is this nlink == 0? */
1818 ext4_mark_inode_dirty(handle, inode);
1822 BUFFER_TRACE(dir_block, "get_write_access");
1823 ext4_journal_get_write_access(handle, dir_block);
1824 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1825 de->inode = cpu_to_le32(inode->i_ino);
1827 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len));
1828 strcpy (de->name, ".");
1829 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1830 de = ext4_next_entry(de);
1831 de->inode = cpu_to_le32(dir->i_ino);
1832 de->rec_len = ext4_rec_len_to_disk(inode->i_sb->s_blocksize -
1833 EXT4_DIR_REC_LEN(1));
1835 strcpy (de->name, "..");
1836 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1838 BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata");
1839 ext4_journal_dirty_metadata(handle, dir_block);
1841 ext4_mark_inode_dirty(handle, inode);
1842 err = ext4_add_entry (handle, dentry, inode);
1845 ext4_mark_inode_dirty(handle, inode);
1849 ext4_inc_count(handle, dir);
1850 ext4_update_dx_flag(dir);
1851 ext4_mark_inode_dirty(handle, dir);
1852 d_instantiate(dentry, inode);
1854 ext4_journal_stop(handle);
1855 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1861 * routine to check that the specified directory is empty (for rmdir)
1863 static int empty_dir (struct inode * inode)
1865 unsigned long offset;
1866 struct buffer_head * bh;
1867 struct ext4_dir_entry_2 * de, * de1;
1868 struct super_block * sb;
1872 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1873 !(bh = ext4_bread (NULL, inode, 0, 0, &err))) {
1875 ext4_error(inode->i_sb, __FUNCTION__,
1876 "error %d reading directory #%lu offset 0",
1879 ext4_warning(inode->i_sb, __FUNCTION__,
1880 "bad directory (dir #%lu) - no data block",
1884 de = (struct ext4_dir_entry_2 *) bh->b_data;
1885 de1 = ext4_next_entry(de);
1886 if (le32_to_cpu(de->inode) != inode->i_ino ||
1887 !le32_to_cpu(de1->inode) ||
1888 strcmp (".", de->name) ||
1889 strcmp ("..", de1->name)) {
1890 ext4_warning (inode->i_sb, "empty_dir",
1891 "bad directory (dir #%lu) - no `.' or `..'",
1896 offset = ext4_rec_len_from_disk(de->rec_len) +
1897 ext4_rec_len_from_disk(de1->rec_len);
1898 de = ext4_next_entry(de1);
1899 while (offset < inode->i_size ) {
1901 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1904 bh = ext4_bread (NULL, inode,
1905 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1908 ext4_error(sb, __FUNCTION__,
1909 "error %d reading directory"
1911 err, inode->i_ino, offset);
1912 offset += sb->s_blocksize;
1915 de = (struct ext4_dir_entry_2 *) bh->b_data;
1917 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1918 de = (struct ext4_dir_entry_2 *)(bh->b_data +
1920 offset = (offset | (sb->s_blocksize - 1)) + 1;
1923 if (le32_to_cpu(de->inode)) {
1927 offset += ext4_rec_len_from_disk(de->rec_len);
1928 de = ext4_next_entry(de);
1934 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1935 * such inodes, starting at the superblock, in case we crash before the
1936 * file is closed/deleted, or in case the inode truncate spans multiple
1937 * transactions and the last transaction is not recovered after a crash.
1939 * At filesystem recovery time, we walk this list deleting unlinked
1940 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1942 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1944 struct super_block *sb = inode->i_sb;
1945 struct ext4_iloc iloc;
1949 if (!list_empty(&EXT4_I(inode)->i_orphan))
1952 /* Orphan handling is only valid for files with data blocks
1953 * being truncated, or files being unlinked. */
1955 /* @@@ FIXME: Observation from aviro:
1956 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
1957 * here (on lock_super()), so race with ext4_link() which might bump
1958 * ->i_nlink. For, say it, character device. Not a regular file,
1959 * not a directory, not a symlink and ->i_nlink > 0.
1961 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1962 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1964 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1965 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1969 err = ext4_reserve_inode_write(handle, inode, &iloc);
1973 /* Insert this inode at the head of the on-disk orphan list... */
1974 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1975 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1976 err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1977 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
1981 /* Only add to the head of the in-memory list if all the
1982 * previous operations succeeded. If the orphan_add is going to
1983 * fail (possibly taking the journal offline), we can't risk
1984 * leaving the inode on the orphan list: stray orphan-list
1985 * entries can cause panics at unmount time.
1987 * This is safe: on error we're going to ignore the orphan list
1988 * anyway on the next recovery. */
1990 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1992 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1993 jbd_debug(4, "orphan inode %lu will point to %d\n",
1994 inode->i_ino, NEXT_ORPHAN(inode));
1997 ext4_std_error(inode->i_sb, err);
2002 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2003 * of such inodes stored on disk, because it is finally being cleaned up.
2005 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2007 struct list_head *prev;
2008 struct ext4_inode_info *ei = EXT4_I(inode);
2009 struct ext4_sb_info *sbi;
2010 unsigned long ino_next;
2011 struct ext4_iloc iloc;
2014 lock_super(inode->i_sb);
2015 if (list_empty(&ei->i_orphan)) {
2016 unlock_super(inode->i_sb);
2020 ino_next = NEXT_ORPHAN(inode);
2021 prev = ei->i_orphan.prev;
2022 sbi = EXT4_SB(inode->i_sb);
2024 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2026 list_del_init(&ei->i_orphan);
2028 /* If we're on an error path, we may not have a valid
2029 * transaction handle with which to update the orphan list on
2030 * disk, but we still need to remove the inode from the linked
2031 * list in memory. */
2035 err = ext4_reserve_inode_write(handle, inode, &iloc);
2039 if (prev == &sbi->s_orphan) {
2040 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2041 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2042 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2045 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2046 err = ext4_journal_dirty_metadata(handle, sbi->s_sbh);
2048 struct ext4_iloc iloc2;
2049 struct inode *i_prev =
2050 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2052 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2053 i_prev->i_ino, ino_next);
2054 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2057 NEXT_ORPHAN(i_prev) = ino_next;
2058 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2062 NEXT_ORPHAN(inode) = 0;
2063 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2066 ext4_std_error(inode->i_sb, err);
2068 unlock_super(inode->i_sb);
2076 static int ext4_rmdir (struct inode * dir, struct dentry *dentry)
2079 struct inode * inode;
2080 struct buffer_head * bh;
2081 struct ext4_dir_entry_2 * de;
2084 /* Initialize quotas before so that eventual writes go in
2085 * separate transaction */
2086 DQUOT_INIT(dentry->d_inode);
2087 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2089 return PTR_ERR(handle);
2092 bh = ext4_find_entry (dentry, &de);
2096 if (IS_DIRSYNC(dir))
2099 inode = dentry->d_inode;
2102 if (le32_to_cpu(de->inode) != inode->i_ino)
2105 retval = -ENOTEMPTY;
2106 if (!empty_dir (inode))
2109 retval = ext4_delete_entry(handle, dir, de, bh);
2112 if (!EXT4_DIR_LINK_EMPTY(inode))
2113 ext4_warning (inode->i_sb, "ext4_rmdir",
2114 "empty directory has too many links (%d)",
2118 /* There's no need to set i_disksize: the fact that i_nlink is
2119 * zero will ensure that the right thing happens during any
2122 ext4_orphan_add(handle, inode);
2123 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2124 ext4_mark_inode_dirty(handle, inode);
2125 ext4_dec_count(handle, dir);
2126 ext4_update_dx_flag(dir);
2127 ext4_mark_inode_dirty(handle, dir);
2130 ext4_journal_stop(handle);
2135 static int ext4_unlink(struct inode * dir, struct dentry *dentry)
2138 struct inode * inode;
2139 struct buffer_head * bh;
2140 struct ext4_dir_entry_2 * de;
2143 /* Initialize quotas before so that eventual writes go
2144 * in separate transaction */
2145 DQUOT_INIT(dentry->d_inode);
2146 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2148 return PTR_ERR(handle);
2150 if (IS_DIRSYNC(dir))
2154 bh = ext4_find_entry (dentry, &de);
2158 inode = dentry->d_inode;
2161 if (le32_to_cpu(de->inode) != inode->i_ino)
2164 if (!inode->i_nlink) {
2165 ext4_warning (inode->i_sb, "ext4_unlink",
2166 "Deleting nonexistent file (%lu), %d",
2167 inode->i_ino, inode->i_nlink);
2170 retval = ext4_delete_entry(handle, dir, de, bh);
2173 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2174 ext4_update_dx_flag(dir);
2175 ext4_mark_inode_dirty(handle, dir);
2176 ext4_dec_count(handle, inode);
2177 if (!inode->i_nlink)
2178 ext4_orphan_add(handle, inode);
2179 inode->i_ctime = ext4_current_time(inode);
2180 ext4_mark_inode_dirty(handle, inode);
2184 ext4_journal_stop(handle);
2189 static int ext4_symlink (struct inode * dir,
2190 struct dentry *dentry, const char * symname)
2193 struct inode * inode;
2194 int l, err, retries = 0;
2196 l = strlen(symname)+1;
2197 if (l > dir->i_sb->s_blocksize)
2198 return -ENAMETOOLONG;
2201 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2202 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2203 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2205 return PTR_ERR(handle);
2207 if (IS_DIRSYNC(dir))
2210 inode = ext4_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2211 err = PTR_ERR(inode);
2215 if (l > sizeof (EXT4_I(inode)->i_data)) {
2216 inode->i_op = &ext4_symlink_inode_operations;
2217 ext4_set_aops(inode);
2219 * page_symlink() calls into ext4_prepare/commit_write.
2220 * We have a transaction open. All is sweetness. It also sets
2221 * i_size in generic_commit_write().
2223 err = __page_symlink(inode, symname, l,
2224 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2226 ext4_dec_count(handle, inode);
2227 ext4_mark_inode_dirty(handle, inode);
2232 inode->i_op = &ext4_fast_symlink_inode_operations;
2233 memcpy((char*)&EXT4_I(inode)->i_data,symname,l);
2234 inode->i_size = l-1;
2236 EXT4_I(inode)->i_disksize = inode->i_size;
2237 err = ext4_add_nondir(handle, dentry, inode);
2239 ext4_journal_stop(handle);
2240 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2245 static int ext4_link (struct dentry * old_dentry,
2246 struct inode * dir, struct dentry *dentry)
2249 struct inode *inode = old_dentry->d_inode;
2250 int err, retries = 0;
2252 if (EXT4_DIR_LINK_MAX(inode))
2256 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2257 * otherwise has the potential to corrupt the orphan inode list.
2259 if (inode->i_nlink == 0)
2263 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2264 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2266 return PTR_ERR(handle);
2268 if (IS_DIRSYNC(dir))
2271 inode->i_ctime = ext4_current_time(inode);
2272 ext4_inc_count(handle, inode);
2273 atomic_inc(&inode->i_count);
2275 err = ext4_add_nondir(handle, dentry, inode);
2276 ext4_journal_stop(handle);
2277 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2282 #define PARENT_INO(buffer) \
2283 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer))->inode)
2286 * Anybody can rename anything with this: the permission checks are left to the
2287 * higher-level routines.
2289 static int ext4_rename (struct inode * old_dir, struct dentry *old_dentry,
2290 struct inode * new_dir,struct dentry *new_dentry)
2293 struct inode * old_inode, * new_inode;
2294 struct buffer_head * old_bh, * new_bh, * dir_bh;
2295 struct ext4_dir_entry_2 * old_de, * new_de;
2298 old_bh = new_bh = dir_bh = NULL;
2300 /* Initialize quotas before so that eventual writes go
2301 * in separate transaction */
2302 if (new_dentry->d_inode)
2303 DQUOT_INIT(new_dentry->d_inode);
2304 handle = ext4_journal_start(old_dir, 2 *
2305 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2306 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2308 return PTR_ERR(handle);
2310 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2313 old_bh = ext4_find_entry (old_dentry, &old_de);
2315 * Check for inode number is _not_ due to possible IO errors.
2316 * We might rmdir the source, keep it as pwd of some process
2317 * and merrily kill the link to whatever was created under the
2318 * same name. Goodbye sticky bit ;-<
2320 old_inode = old_dentry->d_inode;
2322 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2325 new_inode = new_dentry->d_inode;
2326 new_bh = ext4_find_entry (new_dentry, &new_de);
2333 if (S_ISDIR(old_inode->i_mode)) {
2335 retval = -ENOTEMPTY;
2336 if (!empty_dir (new_inode))
2340 dir_bh = ext4_bread (handle, old_inode, 0, 0, &retval);
2343 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2346 if (!new_inode && new_dir!=old_dir &&
2347 new_dir->i_nlink >= EXT4_LINK_MAX)
2351 retval = ext4_add_entry (handle, new_dentry, old_inode);
2355 BUFFER_TRACE(new_bh, "get write access");
2356 ext4_journal_get_write_access(handle, new_bh);
2357 new_de->inode = cpu_to_le32(old_inode->i_ino);
2358 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2359 EXT4_FEATURE_INCOMPAT_FILETYPE))
2360 new_de->file_type = old_de->file_type;
2361 new_dir->i_version++;
2362 BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata");
2363 ext4_journal_dirty_metadata(handle, new_bh);
2369 * Like most other Unix systems, set the ctime for inodes on a
2372 old_inode->i_ctime = ext4_current_time(old_inode);
2373 ext4_mark_inode_dirty(handle, old_inode);
2378 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2379 old_de->name_len != old_dentry->d_name.len ||
2380 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2381 (retval = ext4_delete_entry(handle, old_dir,
2382 old_de, old_bh)) == -ENOENT) {
2383 /* old_de could have moved from under us during htree split, so
2384 * make sure that we are deleting the right entry. We might
2385 * also be pointing to a stale entry in the unused part of
2386 * old_bh so just checking inum and the name isn't enough. */
2387 struct buffer_head *old_bh2;
2388 struct ext4_dir_entry_2 *old_de2;
2390 old_bh2 = ext4_find_entry(old_dentry, &old_de2);
2392 retval = ext4_delete_entry(handle, old_dir,
2398 ext4_warning(old_dir->i_sb, "ext4_rename",
2399 "Deleting old file (%lu), %d, error=%d",
2400 old_dir->i_ino, old_dir->i_nlink, retval);
2404 ext4_dec_count(handle, new_inode);
2405 new_inode->i_ctime = ext4_current_time(new_inode);
2407 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2408 ext4_update_dx_flag(old_dir);
2410 BUFFER_TRACE(dir_bh, "get_write_access");
2411 ext4_journal_get_write_access(handle, dir_bh);
2412 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2413 BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata");
2414 ext4_journal_dirty_metadata(handle, dir_bh);
2415 ext4_dec_count(handle, old_dir);
2417 /* checked empty_dir above, can't have another parent,
2418 * ext3_dec_count() won't work for many-linked dirs */
2419 new_inode->i_nlink = 0;
2421 ext4_inc_count(handle, new_dir);
2422 ext4_update_dx_flag(new_dir);
2423 ext4_mark_inode_dirty(handle, new_dir);
2426 ext4_mark_inode_dirty(handle, old_dir);
2428 ext4_mark_inode_dirty(handle, new_inode);
2429 if (!new_inode->i_nlink)
2430 ext4_orphan_add(handle, new_inode);
2438 ext4_journal_stop(handle);
2443 * directories can handle most operations...
2445 const struct inode_operations ext4_dir_inode_operations = {
2446 .create = ext4_create,
2447 .lookup = ext4_lookup,
2449 .unlink = ext4_unlink,
2450 .symlink = ext4_symlink,
2451 .mkdir = ext4_mkdir,
2452 .rmdir = ext4_rmdir,
2453 .mknod = ext4_mknod,
2454 .rename = ext4_rename,
2455 .setattr = ext4_setattr,
2456 #ifdef CONFIG_EXT4DEV_FS_XATTR
2457 .setxattr = generic_setxattr,
2458 .getxattr = generic_getxattr,
2459 .listxattr = ext4_listxattr,
2460 .removexattr = generic_removexattr,
2462 .permission = ext4_permission,
2465 const struct inode_operations ext4_special_inode_operations = {
2466 .setattr = ext4_setattr,
2467 #ifdef CONFIG_EXT4DEV_FS_XATTR
2468 .setxattr = generic_setxattr,
2469 .getxattr = generic_getxattr,
2470 .listxattr = ext4_listxattr,
2471 .removexattr = generic_removexattr,
2473 .permission = ext4_permission,