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/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
38 #include "ext4_jbd2.h"
43 #include <trace/events/ext4.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)
51 static struct buffer_head *ext4_append(handle_t *handle,
55 struct buffer_head *bh;
58 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
59 ((inode->i_size >> 10) >=
60 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
61 return ERR_PTR(-ENOSPC);
63 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
65 bh = ext4_bread(handle, inode, *block, 1, &err);
68 inode->i_size += inode->i_sb->s_blocksize;
69 EXT4_I(inode)->i_disksize = inode->i_size;
70 BUFFER_TRACE(bh, "get_write_access");
71 err = ext4_journal_get_write_access(handle, bh);
74 ext4_std_error(inode->i_sb, err);
80 static int ext4_dx_csum_verify(struct inode *inode,
81 struct ext4_dir_entry *dirent);
87 #define ext4_read_dirblock(inode, block, type) \
88 __ext4_read_dirblock((inode), (block), (type), __LINE__)
90 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
95 struct buffer_head *bh;
96 struct ext4_dir_entry *dirent;
97 int err = 0, is_dx_block = 0;
99 bh = ext4_bread(NULL, inode, block, 0, &err);
102 ext4_error_inode(inode, __func__, line, block,
103 "Directory hole found");
104 return ERR_PTR(-EIO);
106 __ext4_warning(inode->i_sb, __func__, line,
107 "error reading directory block "
108 "(ino %lu, block %lu)", inode->i_ino,
109 (unsigned long) block);
112 dirent = (struct ext4_dir_entry *) bh->b_data;
113 /* Determine whether or not we have an index block */
117 else if (ext4_rec_len_from_disk(dirent->rec_len,
118 inode->i_sb->s_blocksize) ==
119 inode->i_sb->s_blocksize)
122 if (!is_dx_block && type == INDEX) {
123 ext4_error_inode(inode, __func__, line, block,
124 "directory leaf block found instead of index block");
125 return ERR_PTR(-EIO);
127 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
128 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) ||
133 * An empty leaf block can get mistaken for a index block; for
134 * this reason, we can only check the index checksum when the
135 * caller is sure it should be an index block.
137 if (is_dx_block && type == INDEX) {
138 if (ext4_dx_csum_verify(inode, dirent))
139 set_buffer_verified(bh);
141 ext4_error_inode(inode, __func__, line, block,
142 "Directory index failed checksum");
144 return ERR_PTR(-EIO);
148 if (ext4_dirent_csum_verify(inode, dirent))
149 set_buffer_verified(bh);
151 ext4_error_inode(inode, __func__, line, block,
152 "Directory block failed checksum");
154 return ERR_PTR(-EIO);
161 #define assert(test) J_ASSERT(test)
165 #define dxtrace(command) command
167 #define dxtrace(command)
191 * dx_root_info is laid out so that if it should somehow get overlaid by a
192 * dirent the two low bits of the hash version will be zero. Therefore, the
193 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
198 struct fake_dirent dot;
200 struct fake_dirent dotdot;
204 __le32 reserved_zero;
206 u8 info_length; /* 8 */
211 struct dx_entry entries[0];
216 struct fake_dirent fake;
217 struct dx_entry entries[0];
223 struct buffer_head *bh;
224 struct dx_entry *entries;
236 * This goes at the end of each htree block.
240 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
243 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
244 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
245 static inline unsigned dx_get_hash(struct dx_entry *entry);
246 static void dx_set_hash(struct dx_entry *entry, unsigned value);
247 static unsigned dx_get_count(struct dx_entry *entries);
248 static unsigned dx_get_limit(struct dx_entry *entries);
249 static void dx_set_count(struct dx_entry *entries, unsigned value);
250 static void dx_set_limit(struct dx_entry *entries, unsigned value);
251 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
252 static unsigned dx_node_limit(struct inode *dir);
253 static struct dx_frame *dx_probe(const struct qstr *d_name,
255 struct dx_hash_info *hinfo,
256 struct dx_frame *frame,
258 static void dx_release(struct dx_frame *frames);
259 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
260 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
261 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
262 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
263 struct dx_map_entry *offsets, int count, unsigned blocksize);
264 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
265 static void dx_insert_block(struct dx_frame *frame,
266 u32 hash, ext4_lblk_t block);
267 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
268 struct dx_frame *frame,
269 struct dx_frame *frames,
271 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
272 const struct qstr *d_name,
273 struct ext4_dir_entry_2 **res_dir);
274 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
275 struct inode *inode);
277 /* checksumming functions */
278 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
279 unsigned int blocksize)
281 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
282 t->det_rec_len = ext4_rec_len_to_disk(
283 sizeof(struct ext4_dir_entry_tail), blocksize);
284 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
287 /* Walk through a dirent block to find a checksum "dirent" at the tail */
288 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
289 struct ext4_dir_entry *de)
291 struct ext4_dir_entry_tail *t;
294 struct ext4_dir_entry *d, *top;
297 top = (struct ext4_dir_entry *)(((void *)de) +
298 (EXT4_BLOCK_SIZE(inode->i_sb) -
299 sizeof(struct ext4_dir_entry_tail)));
300 while (d < top && d->rec_len)
301 d = (struct ext4_dir_entry *)(((void *)d) +
302 le16_to_cpu(d->rec_len));
307 t = (struct ext4_dir_entry_tail *)d;
309 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
312 if (t->det_reserved_zero1 ||
313 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
314 t->det_reserved_zero2 ||
315 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
321 static __le32 ext4_dirent_csum(struct inode *inode,
322 struct ext4_dir_entry *dirent, int size)
324 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
325 struct ext4_inode_info *ei = EXT4_I(inode);
328 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
329 return cpu_to_le32(csum);
332 static void warn_no_space_for_csum(struct inode *inode)
334 ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
335 "checksum. Please run e2fsck -D.", inode->i_ino);
338 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
340 struct ext4_dir_entry_tail *t;
342 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
343 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
346 t = get_dirent_tail(inode, dirent);
348 warn_no_space_for_csum(inode);
352 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
353 (void *)t - (void *)dirent))
359 static void ext4_dirent_csum_set(struct inode *inode,
360 struct ext4_dir_entry *dirent)
362 struct ext4_dir_entry_tail *t;
364 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
365 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
368 t = get_dirent_tail(inode, dirent);
370 warn_no_space_for_csum(inode);
374 t->det_checksum = ext4_dirent_csum(inode, dirent,
375 (void *)t - (void *)dirent);
378 int ext4_handle_dirty_dirent_node(handle_t *handle,
380 struct buffer_head *bh)
382 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
383 return ext4_handle_dirty_metadata(handle, inode, bh);
386 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
387 struct ext4_dir_entry *dirent,
390 struct ext4_dir_entry *dp;
391 struct dx_root_info *root;
394 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
396 else if (le16_to_cpu(dirent->rec_len) == 12) {
397 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
398 if (le16_to_cpu(dp->rec_len) !=
399 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
401 root = (struct dx_root_info *)(((void *)dp + 12));
402 if (root->reserved_zero ||
403 root->info_length != sizeof(struct dx_root_info))
410 *offset = count_offset;
411 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
414 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
415 int count_offset, int count, struct dx_tail *t)
417 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
418 struct ext4_inode_info *ei = EXT4_I(inode);
423 size = count_offset + (count * sizeof(struct dx_entry));
424 save_csum = t->dt_checksum;
426 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
427 csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
428 t->dt_checksum = save_csum;
430 return cpu_to_le32(csum);
433 static int ext4_dx_csum_verify(struct inode *inode,
434 struct ext4_dir_entry *dirent)
436 struct dx_countlimit *c;
438 int count_offset, limit, count;
440 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
441 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
444 c = get_dx_countlimit(inode, dirent, &count_offset);
446 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
449 limit = le16_to_cpu(c->limit);
450 count = le16_to_cpu(c->count);
451 if (count_offset + (limit * sizeof(struct dx_entry)) >
452 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
453 warn_no_space_for_csum(inode);
456 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
458 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
464 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
466 struct dx_countlimit *c;
468 int count_offset, limit, count;
470 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
471 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
474 c = get_dx_countlimit(inode, dirent, &count_offset);
476 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
479 limit = le16_to_cpu(c->limit);
480 count = le16_to_cpu(c->count);
481 if (count_offset + (limit * sizeof(struct dx_entry)) >
482 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
483 warn_no_space_for_csum(inode);
486 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
488 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
491 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
493 struct buffer_head *bh)
495 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
496 return ext4_handle_dirty_metadata(handle, inode, bh);
500 * p is at least 6 bytes before the end of page
502 static inline struct ext4_dir_entry_2 *
503 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
505 return (struct ext4_dir_entry_2 *)((char *)p +
506 ext4_rec_len_from_disk(p->rec_len, blocksize));
510 * Future: use high four bits of block for coalesce-on-delete flags
511 * Mask them off for now.
514 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
516 return le32_to_cpu(entry->block) & 0x00ffffff;
519 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
521 entry->block = cpu_to_le32(value);
524 static inline unsigned dx_get_hash(struct dx_entry *entry)
526 return le32_to_cpu(entry->hash);
529 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
531 entry->hash = cpu_to_le32(value);
534 static inline unsigned dx_get_count(struct dx_entry *entries)
536 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
539 static inline unsigned dx_get_limit(struct dx_entry *entries)
541 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
544 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
546 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
549 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
551 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
554 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
556 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
557 EXT4_DIR_REC_LEN(2) - infosize;
559 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
560 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
561 entry_space -= sizeof(struct dx_tail);
562 return entry_space / sizeof(struct dx_entry);
565 static inline unsigned dx_node_limit(struct inode *dir)
567 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
569 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
570 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
571 entry_space -= sizeof(struct dx_tail);
572 return entry_space / sizeof(struct dx_entry);
579 static void dx_show_index(char * label, struct dx_entry *entries)
581 int i, n = dx_get_count (entries);
582 printk(KERN_DEBUG "%s index ", label);
583 for (i = 0; i < n; i++) {
584 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
585 0, (unsigned long)dx_get_block(entries + i));
597 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
598 int size, int show_names)
600 unsigned names = 0, space = 0;
601 char *base = (char *) de;
602 struct dx_hash_info h = *hinfo;
605 while ((char *) de < base + size)
611 int len = de->name_len;
612 char *name = de->name;
613 while (len--) printk("%c", *name++);
614 ext4fs_dirhash(de->name, de->name_len, &h);
615 printk(":%x.%u ", h.hash,
616 (unsigned) ((char *) de - base));
618 space += EXT4_DIR_REC_LEN(de->name_len);
621 de = ext4_next_entry(de, size);
623 printk("(%i)\n", names);
624 return (struct stats) { names, space, 1 };
627 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
628 struct dx_entry *entries, int levels)
630 unsigned blocksize = dir->i_sb->s_blocksize;
631 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
633 struct buffer_head *bh;
635 printk("%i indexed blocks...\n", count);
636 for (i = 0; i < count; i++, entries++)
638 ext4_lblk_t block = dx_get_block(entries);
639 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
640 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
642 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
643 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
645 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
646 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
647 names += stats.names;
648 space += stats.space;
649 bcount += stats.bcount;
653 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
654 levels ? "" : " ", names, space/bcount,
655 (space/bcount)*100/blocksize);
656 return (struct stats) { names, space, bcount};
658 #endif /* DX_DEBUG */
661 * Probe for a directory leaf block to search.
663 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
664 * error in the directory index, and the caller should fall back to
665 * searching the directory normally. The callers of dx_probe **MUST**
666 * check for this error code, and make sure it never gets reflected
669 static struct dx_frame *
670 dx_probe(const struct qstr *d_name, struct inode *dir,
671 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
673 unsigned count, indirect;
674 struct dx_entry *at, *entries, *p, *q, *m;
675 struct dx_root *root;
676 struct buffer_head *bh;
677 struct dx_frame *frame = frame_in;
681 bh = ext4_read_dirblock(dir, 0, INDEX);
686 root = (struct dx_root *) bh->b_data;
687 if (root->info.hash_version != DX_HASH_TEA &&
688 root->info.hash_version != DX_HASH_HALF_MD4 &&
689 root->info.hash_version != DX_HASH_LEGACY) {
690 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
691 root->info.hash_version);
693 *err = ERR_BAD_DX_DIR;
696 hinfo->hash_version = root->info.hash_version;
697 if (hinfo->hash_version <= DX_HASH_TEA)
698 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
699 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
701 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
704 if (root->info.unused_flags & 1) {
705 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
706 root->info.unused_flags);
708 *err = ERR_BAD_DX_DIR;
712 if ((indirect = root->info.indirect_levels) > 1) {
713 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
714 root->info.indirect_levels);
716 *err = ERR_BAD_DX_DIR;
720 entries = (struct dx_entry *) (((char *)&root->info) +
721 root->info.info_length);
723 if (dx_get_limit(entries) != dx_root_limit(dir,
724 root->info.info_length)) {
725 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
727 *err = ERR_BAD_DX_DIR;
731 dxtrace(printk("Look up %x", hash));
734 count = dx_get_count(entries);
735 if (!count || count > dx_get_limit(entries)) {
736 ext4_warning(dir->i_sb,
737 "dx entry: no count or count > limit");
739 *err = ERR_BAD_DX_DIR;
744 q = entries + count - 1;
748 dxtrace(printk("."));
749 if (dx_get_hash(m) > hash)
755 if (0) // linear search cross check
757 unsigned n = count - 1;
761 dxtrace(printk(","));
762 if (dx_get_hash(++at) > hash)
768 assert (at == p - 1);
772 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
774 frame->entries = entries;
776 if (!indirect--) return frame;
777 bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
782 entries = ((struct dx_node *) bh->b_data)->entries;
784 if (dx_get_limit(entries) != dx_node_limit (dir)) {
785 ext4_warning(dir->i_sb,
786 "dx entry: limit != node limit");
788 *err = ERR_BAD_DX_DIR;
795 while (frame >= frame_in) {
800 if (*err == ERR_BAD_DX_DIR)
801 ext4_warning(dir->i_sb,
802 "Corrupt dir inode %lu, running e2fsck is "
803 "recommended.", dir->i_ino);
807 static void dx_release (struct dx_frame *frames)
809 if (frames[0].bh == NULL)
812 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
813 brelse(frames[1].bh);
814 brelse(frames[0].bh);
818 * This function increments the frame pointer to search the next leaf
819 * block, and reads in the necessary intervening nodes if the search
820 * should be necessary. Whether or not the search is necessary is
821 * controlled by the hash parameter. If the hash value is even, then
822 * the search is only continued if the next block starts with that
823 * hash value. This is used if we are searching for a specific file.
825 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
827 * This function returns 1 if the caller should continue to search,
828 * or 0 if it should not. If there is an error reading one of the
829 * index blocks, it will a negative error code.
831 * If start_hash is non-null, it will be filled in with the starting
832 * hash of the next page.
834 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
835 struct dx_frame *frame,
836 struct dx_frame *frames,
840 struct buffer_head *bh;
846 * Find the next leaf page by incrementing the frame pointer.
847 * If we run out of entries in the interior node, loop around and
848 * increment pointer in the parent node. When we break out of
849 * this loop, num_frames indicates the number of interior
850 * nodes need to be read.
853 if (++(p->at) < p->entries + dx_get_count(p->entries))
862 * If the hash is 1, then continue only if the next page has a
863 * continuation hash of any value. This is used for readdir
864 * handling. Otherwise, check to see if the hash matches the
865 * desired contiuation hash. If it doesn't, return since
866 * there's no point to read in the successive index pages.
868 bhash = dx_get_hash(p->at);
871 if ((hash & 1) == 0) {
872 if ((bhash & ~1) != hash)
876 * If the hash is HASH_NB_ALWAYS, we always go to the next
877 * block so no check is necessary
879 while (num_frames--) {
880 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
886 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
893 * This function fills a red-black tree with information from a
894 * directory block. It returns the number directory entries loaded
895 * into the tree. If there is an error it is returned in err.
897 static int htree_dirblock_to_tree(struct file *dir_file,
898 struct inode *dir, ext4_lblk_t block,
899 struct dx_hash_info *hinfo,
900 __u32 start_hash, __u32 start_minor_hash)
902 struct buffer_head *bh;
903 struct ext4_dir_entry_2 *de, *top;
904 int err = 0, count = 0;
906 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
907 (unsigned long)block));
908 bh = ext4_read_dirblock(dir, block, DIRENT);
912 de = (struct ext4_dir_entry_2 *) bh->b_data;
913 top = (struct ext4_dir_entry_2 *) ((char *) de +
914 dir->i_sb->s_blocksize -
915 EXT4_DIR_REC_LEN(0));
916 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
917 if (ext4_check_dir_entry(dir, NULL, de, bh,
918 bh->b_data, bh->b_size,
919 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
920 + ((char *)de - bh->b_data))) {
921 /* silently ignore the rest of the block */
924 ext4fs_dirhash(de->name, de->name_len, hinfo);
925 if ((hinfo->hash < start_hash) ||
926 ((hinfo->hash == start_hash) &&
927 (hinfo->minor_hash < start_minor_hash)))
931 if ((err = ext4_htree_store_dirent(dir_file,
932 hinfo->hash, hinfo->minor_hash, de)) != 0) {
944 * This function fills a red-black tree with information from a
945 * directory. We start scanning the directory in hash order, starting
946 * at start_hash and start_minor_hash.
948 * This function returns the number of entries inserted into the tree,
949 * or a negative error code.
951 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
952 __u32 start_minor_hash, __u32 *next_hash)
954 struct dx_hash_info hinfo;
955 struct ext4_dir_entry_2 *de;
956 struct dx_frame frames[2], *frame;
963 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
964 start_hash, start_minor_hash));
965 dir = file_inode(dir_file);
966 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
967 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
968 if (hinfo.hash_version <= DX_HASH_TEA)
969 hinfo.hash_version +=
970 EXT4_SB(dir->i_sb)->s_hash_unsigned;
971 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
972 if (ext4_has_inline_data(dir)) {
973 int has_inline_data = 1;
974 count = htree_inlinedir_to_tree(dir_file, dir, 0,
978 if (has_inline_data) {
983 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
984 start_hash, start_minor_hash);
988 hinfo.hash = start_hash;
989 hinfo.minor_hash = 0;
990 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
994 /* Add '.' and '..' from the htree header */
995 if (!start_hash && !start_minor_hash) {
996 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
997 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
1001 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1002 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1003 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1004 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
1010 block = dx_get_block(frame->at);
1011 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1012 start_hash, start_minor_hash);
1019 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1020 frame, frames, &hashval);
1021 *next_hash = hashval;
1027 * Stop if: (a) there are no more entries, or
1028 * (b) we have inserted at least one entry and the
1029 * next hash value is not a continuation
1032 (count && ((hashval & 1) == 0)))
1036 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1037 "next hash: %x\n", count, *next_hash));
1044 static inline int search_dirblock(struct buffer_head *bh,
1046 const struct qstr *d_name,
1047 unsigned int offset,
1048 struct ext4_dir_entry_2 **res_dir)
1050 return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1051 d_name, offset, res_dir);
1055 * Directory block splitting, compacting
1059 * Create map of hash values, offsets, and sizes, stored at end of block.
1060 * Returns number of entries mapped.
1062 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1063 struct dx_hash_info *hinfo,
1064 struct dx_map_entry *map_tail)
1067 char *base = (char *) de;
1068 struct dx_hash_info h = *hinfo;
1070 while ((char *) de < base + blocksize) {
1071 if (de->name_len && de->inode) {
1072 ext4fs_dirhash(de->name, de->name_len, &h);
1074 map_tail->hash = h.hash;
1075 map_tail->offs = ((char *) de - base)>>2;
1076 map_tail->size = le16_to_cpu(de->rec_len);
1080 /* XXX: do we need to check rec_len == 0 case? -Chris */
1081 de = ext4_next_entry(de, blocksize);
1086 /* Sort map by hash value */
1087 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1089 struct dx_map_entry *p, *q, *top = map + count - 1;
1091 /* Combsort until bubble sort doesn't suck */
1093 count = count*10/13;
1094 if (count - 9 < 2) /* 9, 10 -> 11 */
1096 for (p = top, q = p - count; q >= map; p--, q--)
1097 if (p->hash < q->hash)
1100 /* Garden variety bubble sort */
1105 if (q[1].hash >= q[0].hash)
1113 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1115 struct dx_entry *entries = frame->entries;
1116 struct dx_entry *old = frame->at, *new = old + 1;
1117 int count = dx_get_count(entries);
1119 assert(count < dx_get_limit(entries));
1120 assert(old < entries + count);
1121 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1122 dx_set_hash(new, hash);
1123 dx_set_block(new, block);
1124 dx_set_count(entries, count + 1);
1128 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1130 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1131 * `de != NULL' is guaranteed by caller.
1133 static inline int ext4_match (int len, const char * const name,
1134 struct ext4_dir_entry_2 * de)
1136 if (len != de->name_len)
1140 return !memcmp(name, de->name, len);
1144 * Returns 0 if not found, -1 on failure, and 1 on success
1146 int search_dir(struct buffer_head *bh,
1150 const struct qstr *d_name,
1151 unsigned int offset,
1152 struct ext4_dir_entry_2 **res_dir)
1154 struct ext4_dir_entry_2 * de;
1157 const char *name = d_name->name;
1158 int namelen = d_name->len;
1160 de = (struct ext4_dir_entry_2 *)search_buf;
1161 dlimit = search_buf + buf_size;
1162 while ((char *) de < dlimit) {
1163 /* this code is executed quadratically often */
1164 /* do minimal checking `by hand' */
1166 if ((char *) de + namelen <= dlimit &&
1167 ext4_match (namelen, name, de)) {
1168 /* found a match - just to be sure, do a full check */
1169 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1170 bh->b_size, offset))
1175 /* prevent looping on a bad block */
1176 de_len = ext4_rec_len_from_disk(de->rec_len,
1177 dir->i_sb->s_blocksize);
1181 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1186 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1187 struct ext4_dir_entry *de)
1189 struct super_block *sb = dir->i_sb;
1195 if (de->inode == 0 &&
1196 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1205 * finds an entry in the specified directory with the wanted name. It
1206 * returns the cache buffer in which the entry was found, and the entry
1207 * itself (as a parameter - res_dir). It does NOT read the inode of the
1208 * entry - you'll have to do that yourself if you want to.
1210 * The returned buffer_head has ->b_count elevated. The caller is expected
1211 * to brelse() it when appropriate.
1213 static struct buffer_head * ext4_find_entry (struct inode *dir,
1214 const struct qstr *d_name,
1215 struct ext4_dir_entry_2 **res_dir,
1218 struct super_block *sb;
1219 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1220 struct buffer_head *bh, *ret = NULL;
1221 ext4_lblk_t start, block, b;
1222 const u8 *name = d_name->name;
1223 int ra_max = 0; /* Number of bh's in the readahead
1225 int ra_ptr = 0; /* Current index into readahead
1228 ext4_lblk_t nblocks;
1234 namelen = d_name->len;
1235 if (namelen > EXT4_NAME_LEN)
1238 if (ext4_has_inline_data(dir)) {
1239 int has_inline_data = 1;
1240 ret = ext4_find_inline_entry(dir, d_name, res_dir,
1242 if (has_inline_data) {
1249 if ((namelen <= 2) && (name[0] == '.') &&
1250 (name[1] == '.' || name[1] == '\0')) {
1252 * "." or ".." will only be in the first block
1253 * NFS may look up ".."; "." should be handled by the VFS
1260 bh = ext4_dx_find_entry(dir, d_name, res_dir);
1262 * On success, or if the error was file not found,
1263 * return. Otherwise, fall back to doing a search the
1264 * old fashioned way.
1266 if (!IS_ERR(bh) || PTR_ERR(bh) != ERR_BAD_DX_DIR)
1268 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1271 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1272 start = EXT4_I(dir)->i_dir_start_lookup;
1273 if (start >= nblocks)
1279 * We deal with the read-ahead logic here.
1281 if (ra_ptr >= ra_max) {
1282 /* Refill the readahead buffer */
1285 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1287 * Terminate if we reach the end of the
1288 * directory and must wrap, or if our
1289 * search has finished at this block.
1291 if (b >= nblocks || (num && block == start)) {
1292 bh_use[ra_max] = NULL;
1296 bh = ext4_getblk(NULL, dir, b++, 0, &err);
1297 if (unlikely(err)) {
1299 return ERR_PTR(err);
1302 bh_use[ra_max] = bh;
1304 ll_rw_block(READ | REQ_META | REQ_PRIO,
1308 if ((bh = bh_use[ra_ptr++]) == NULL)
1311 if (!buffer_uptodate(bh)) {
1312 /* read error, skip block & hope for the best */
1313 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1314 (unsigned long) block);
1318 if (!buffer_verified(bh) &&
1319 !is_dx_internal_node(dir, block,
1320 (struct ext4_dir_entry *)bh->b_data) &&
1321 !ext4_dirent_csum_verify(dir,
1322 (struct ext4_dir_entry *)bh->b_data)) {
1323 EXT4_ERROR_INODE(dir, "checksumming directory "
1324 "block %lu", (unsigned long)block);
1328 set_buffer_verified(bh);
1329 i = search_dirblock(bh, dir, d_name,
1330 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1332 EXT4_I(dir)->i_dir_start_lookup = block;
1334 goto cleanup_and_exit;
1338 goto cleanup_and_exit;
1341 if (++block >= nblocks)
1343 } while (block != start);
1346 * If the directory has grown while we were searching, then
1347 * search the last part of the directory before giving up.
1350 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1351 if (block < nblocks) {
1357 /* Clean up the read-ahead blocks */
1358 for (; ra_ptr < ra_max; ra_ptr++)
1359 brelse(bh_use[ra_ptr]);
1363 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1364 struct ext4_dir_entry_2 **res_dir)
1366 struct super_block * sb = dir->i_sb;
1367 struct dx_hash_info hinfo;
1368 struct dx_frame frames[2], *frame;
1369 struct buffer_head *bh;
1371 int err = 0, retval;
1373 frame = dx_probe(d_name, dir, &hinfo, frames, &err);
1375 return ERR_PTR(err);
1377 block = dx_get_block(frame->at);
1378 bh = ext4_read_dirblock(dir, block, DIRENT);
1382 retval = search_dirblock(bh, dir, d_name,
1383 block << EXT4_BLOCK_SIZE_BITS(sb),
1389 bh = ERR_PTR(ERR_BAD_DX_DIR);
1393 /* Check to see if we should continue to search */
1394 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1398 "error %d reading index page in directory #%lu",
1399 retval, dir->i_ino);
1400 bh = ERR_PTR(retval);
1403 } while (retval == 1);
1407 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1413 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1415 struct inode *inode;
1416 struct ext4_dir_entry_2 *de;
1417 struct buffer_head *bh;
1419 if (dentry->d_name.len > EXT4_NAME_LEN)
1420 return ERR_PTR(-ENAMETOOLONG);
1422 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1424 return (struct dentry *) bh;
1427 __u32 ino = le32_to_cpu(de->inode);
1429 if (!ext4_valid_inum(dir->i_sb, ino)) {
1430 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1431 return ERR_PTR(-EIO);
1433 if (unlikely(ino == dir->i_ino)) {
1434 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1436 return ERR_PTR(-EIO);
1438 inode = ext4_iget(dir->i_sb, ino);
1439 if (inode == ERR_PTR(-ESTALE)) {
1440 EXT4_ERROR_INODE(dir,
1441 "deleted inode referenced: %u",
1443 return ERR_PTR(-EIO);
1446 return d_splice_alias(inode, dentry);
1450 struct dentry *ext4_get_parent(struct dentry *child)
1453 static const struct qstr dotdot = QSTR_INIT("..", 2);
1454 struct ext4_dir_entry_2 * de;
1455 struct buffer_head *bh;
1457 bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1459 return (struct dentry *) bh;
1461 return ERR_PTR(-ENOENT);
1462 ino = le32_to_cpu(de->inode);
1465 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1466 EXT4_ERROR_INODE(child->d_inode,
1467 "bad parent inode number: %u", ino);
1468 return ERR_PTR(-EIO);
1471 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1475 * Move count entries from end of map between two memory locations.
1476 * Returns pointer to last entry moved.
1478 static struct ext4_dir_entry_2 *
1479 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1482 unsigned rec_len = 0;
1485 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1486 (from + (map->offs<<2));
1487 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1488 memcpy (to, de, rec_len);
1489 ((struct ext4_dir_entry_2 *) to)->rec_len =
1490 ext4_rec_len_to_disk(rec_len, blocksize);
1495 return (struct ext4_dir_entry_2 *) (to - rec_len);
1499 * Compact each dir entry in the range to the minimal rec_len.
1500 * Returns pointer to last entry in range.
1502 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1504 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1505 unsigned rec_len = 0;
1508 while ((char*)de < base + blocksize) {
1509 next = ext4_next_entry(de, blocksize);
1510 if (de->inode && de->name_len) {
1511 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1513 memmove(to, de, rec_len);
1514 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1516 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1524 * Split a full leaf block to make room for a new dir entry.
1525 * Allocate a new block, and move entries so that they are approx. equally full.
1526 * Returns pointer to de in block into which the new entry will be inserted.
1528 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1529 struct buffer_head **bh,struct dx_frame *frame,
1530 struct dx_hash_info *hinfo, int *error)
1532 unsigned blocksize = dir->i_sb->s_blocksize;
1533 unsigned count, continued;
1534 struct buffer_head *bh2;
1535 ext4_lblk_t newblock;
1537 struct dx_map_entry *map;
1538 char *data1 = (*bh)->b_data, *data2;
1539 unsigned split, move, size;
1540 struct ext4_dir_entry_2 *de = NULL, *de2;
1541 struct ext4_dir_entry_tail *t;
1545 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1546 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1547 csum_size = sizeof(struct ext4_dir_entry_tail);
1549 bh2 = ext4_append(handle, dir, &newblock);
1553 *error = PTR_ERR(bh2);
1557 BUFFER_TRACE(*bh, "get_write_access");
1558 err = ext4_journal_get_write_access(handle, *bh);
1562 BUFFER_TRACE(frame->bh, "get_write_access");
1563 err = ext4_journal_get_write_access(handle, frame->bh);
1567 data2 = bh2->b_data;
1569 /* create map in the end of data2 block */
1570 map = (struct dx_map_entry *) (data2 + blocksize);
1571 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1572 blocksize, hinfo, map);
1574 dx_sort_map(map, count);
1575 /* Split the existing block in the middle, size-wise */
1578 for (i = count-1; i >= 0; i--) {
1579 /* is more than half of this entry in 2nd half of the block? */
1580 if (size + map[i].size/2 > blocksize/2)
1582 size += map[i].size;
1585 /* map index at which we will split */
1586 split = count - move;
1587 hash2 = map[split].hash;
1588 continued = hash2 == map[split - 1].hash;
1589 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1590 (unsigned long)dx_get_block(frame->at),
1591 hash2, split, count-split));
1593 /* Fancy dance to stay within two buffers */
1594 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1595 de = dx_pack_dirents(data1, blocksize);
1596 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1599 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1603 t = EXT4_DIRENT_TAIL(data2, blocksize);
1604 initialize_dirent_tail(t, blocksize);
1606 t = EXT4_DIRENT_TAIL(data1, blocksize);
1607 initialize_dirent_tail(t, blocksize);
1610 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1611 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1613 /* Which block gets the new entry? */
1614 if (hinfo->hash >= hash2)
1619 dx_insert_block(frame, hash2 + continued, newblock);
1620 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1623 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1627 dxtrace(dx_show_index("frame", frame->entries));
1634 ext4_std_error(dir->i_sb, err);
1639 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1640 struct buffer_head *bh,
1641 void *buf, int buf_size,
1642 const char *name, int namelen,
1643 struct ext4_dir_entry_2 **dest_de)
1645 struct ext4_dir_entry_2 *de;
1646 unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1648 unsigned int offset = 0;
1651 de = (struct ext4_dir_entry_2 *)buf;
1652 top = buf + buf_size - reclen;
1653 while ((char *) de <= top) {
1654 if (ext4_check_dir_entry(dir, NULL, de, bh,
1655 buf, buf_size, offset))
1657 if (ext4_match(namelen, name, de))
1659 nlen = EXT4_DIR_REC_LEN(de->name_len);
1660 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1661 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1663 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1666 if ((char *) de > top)
1673 void ext4_insert_dentry(struct inode *inode,
1674 struct ext4_dir_entry_2 *de,
1676 const char *name, int namelen)
1681 nlen = EXT4_DIR_REC_LEN(de->name_len);
1682 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1684 struct ext4_dir_entry_2 *de1 =
1685 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1686 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1687 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1690 de->file_type = EXT4_FT_UNKNOWN;
1691 de->inode = cpu_to_le32(inode->i_ino);
1692 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1693 de->name_len = namelen;
1694 memcpy(de->name, name, namelen);
1697 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1698 * it points to a directory entry which is guaranteed to be large
1699 * enough for new directory entry. If de is NULL, then
1700 * add_dirent_to_buf will attempt search the directory block for
1701 * space. It will return -ENOSPC if no space is available, and -EIO
1702 * and -EEXIST if directory entry already exists.
1704 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1705 struct inode *inode, struct ext4_dir_entry_2 *de,
1706 struct buffer_head *bh)
1708 struct inode *dir = dentry->d_parent->d_inode;
1709 const char *name = dentry->d_name.name;
1710 int namelen = dentry->d_name.len;
1711 unsigned int blocksize = dir->i_sb->s_blocksize;
1715 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1716 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1717 csum_size = sizeof(struct ext4_dir_entry_tail);
1720 err = ext4_find_dest_de(dir, inode,
1721 bh, bh->b_data, blocksize - csum_size,
1722 name, namelen, &de);
1726 BUFFER_TRACE(bh, "get_write_access");
1727 err = ext4_journal_get_write_access(handle, bh);
1729 ext4_std_error(dir->i_sb, err);
1733 /* By now the buffer is marked for journaling */
1734 ext4_insert_dentry(inode, de, blocksize, name, namelen);
1737 * XXX shouldn't update any times until successful
1738 * completion of syscall, but too many callers depend
1741 * XXX similarly, too many callers depend on
1742 * ext4_new_inode() setting the times, but error
1743 * recovery deletes the inode, so the worst that can
1744 * happen is that the times are slightly out of date
1745 * and/or different from the directory change time.
1747 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1748 ext4_update_dx_flag(dir);
1750 ext4_mark_inode_dirty(handle, dir);
1751 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1752 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1754 ext4_std_error(dir->i_sb, err);
1759 * This converts a one block unindexed directory to a 3 block indexed
1760 * directory, and adds the dentry to the indexed directory.
1762 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1763 struct inode *inode, struct buffer_head *bh)
1765 struct inode *dir = dentry->d_parent->d_inode;
1766 const char *name = dentry->d_name.name;
1767 int namelen = dentry->d_name.len;
1768 struct buffer_head *bh2;
1769 struct dx_root *root;
1770 struct dx_frame frames[2], *frame;
1771 struct dx_entry *entries;
1772 struct ext4_dir_entry_2 *de, *de2;
1773 struct ext4_dir_entry_tail *t;
1778 struct dx_hash_info hinfo;
1780 struct fake_dirent *fde;
1783 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1784 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1785 csum_size = sizeof(struct ext4_dir_entry_tail);
1787 blocksize = dir->i_sb->s_blocksize;
1788 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1789 BUFFER_TRACE(bh, "get_write_access");
1790 retval = ext4_journal_get_write_access(handle, bh);
1792 ext4_std_error(dir->i_sb, retval);
1796 root = (struct dx_root *) bh->b_data;
1798 /* The 0th block becomes the root, move the dirents out */
1799 fde = &root->dotdot;
1800 de = (struct ext4_dir_entry_2 *)((char *)fde +
1801 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1802 if ((char *) de >= (((char *) root) + blocksize)) {
1803 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1807 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1809 /* Allocate new block for the 0th block's dirents */
1810 bh2 = ext4_append(handle, dir, &block);
1813 return PTR_ERR(bh2);
1815 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1816 data1 = bh2->b_data;
1818 memcpy (data1, de, len);
1819 de = (struct ext4_dir_entry_2 *) data1;
1821 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1823 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1828 t = EXT4_DIRENT_TAIL(data1, blocksize);
1829 initialize_dirent_tail(t, blocksize);
1832 /* Initialize the root; the dot dirents already exist */
1833 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1834 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1836 memset (&root->info, 0, sizeof(root->info));
1837 root->info.info_length = sizeof(root->info);
1838 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1839 entries = root->entries;
1840 dx_set_block(entries, 1);
1841 dx_set_count(entries, 1);
1842 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1844 /* Initialize as for dx_probe */
1845 hinfo.hash_version = root->info.hash_version;
1846 if (hinfo.hash_version <= DX_HASH_TEA)
1847 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1848 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1849 ext4fs_dirhash(name, namelen, &hinfo);
1851 frame->entries = entries;
1852 frame->at = entries;
1856 ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1857 ext4_handle_dirty_dirent_node(handle, dir, bh);
1859 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1862 * Even if the block split failed, we have to properly write
1863 * out all the changes we did so far. Otherwise we can end up
1864 * with corrupted filesystem.
1866 ext4_mark_inode_dirty(handle, dir);
1872 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1880 * adds a file entry to the specified directory, using the same
1881 * semantics as ext4_find_entry(). It returns NULL if it failed.
1883 * NOTE!! The inode part of 'de' is left at 0 - which means you
1884 * may not sleep between calling this and putting something into
1885 * the entry, as someone else might have used it while you slept.
1887 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1888 struct inode *inode)
1890 struct inode *dir = dentry->d_parent->d_inode;
1891 struct buffer_head *bh;
1892 struct ext4_dir_entry_2 *de;
1893 struct ext4_dir_entry_tail *t;
1894 struct super_block *sb;
1898 ext4_lblk_t block, blocks;
1901 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1902 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1903 csum_size = sizeof(struct ext4_dir_entry_tail);
1906 blocksize = sb->s_blocksize;
1907 if (!dentry->d_name.len)
1910 if (ext4_has_inline_data(dir)) {
1911 retval = ext4_try_add_inline_entry(handle, dentry, inode);
1921 retval = ext4_dx_add_entry(handle, dentry, inode);
1922 if (!retval || (retval != ERR_BAD_DX_DIR))
1924 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1926 ext4_mark_inode_dirty(handle, dir);
1928 blocks = dir->i_size >> sb->s_blocksize_bits;
1929 for (block = 0; block < blocks; block++) {
1930 bh = ext4_read_dirblock(dir, block, DIRENT);
1934 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1935 if (retval != -ENOSPC) {
1940 if (blocks == 1 && !dx_fallback &&
1941 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1942 return make_indexed_dir(handle, dentry, inode, bh);
1945 bh = ext4_append(handle, dir, &block);
1948 de = (struct ext4_dir_entry_2 *) bh->b_data;
1950 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1953 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1954 initialize_dirent_tail(t, blocksize);
1957 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1960 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1965 * Returns 0 for success, or a negative error value
1967 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1968 struct inode *inode)
1970 struct dx_frame frames[2], *frame;
1971 struct dx_entry *entries, *at;
1972 struct dx_hash_info hinfo;
1973 struct buffer_head *bh;
1974 struct inode *dir = dentry->d_parent->d_inode;
1975 struct super_block *sb = dir->i_sb;
1976 struct ext4_dir_entry_2 *de;
1979 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1982 entries = frame->entries;
1984 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
1991 BUFFER_TRACE(bh, "get_write_access");
1992 err = ext4_journal_get_write_access(handle, bh);
1996 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
2000 /* Block full, should compress but for now just split */
2001 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2002 dx_get_count(entries), dx_get_limit(entries)));
2003 /* Need to split index? */
2004 if (dx_get_count(entries) == dx_get_limit(entries)) {
2005 ext4_lblk_t newblock;
2006 unsigned icount = dx_get_count(entries);
2007 int levels = frame - frames;
2008 struct dx_entry *entries2;
2009 struct dx_node *node2;
2010 struct buffer_head *bh2;
2012 if (levels && (dx_get_count(frames->entries) ==
2013 dx_get_limit(frames->entries))) {
2014 ext4_warning(sb, "Directory index full!");
2018 bh2 = ext4_append(handle, dir, &newblock);
2023 node2 = (struct dx_node *)(bh2->b_data);
2024 entries2 = node2->entries;
2025 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2026 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2028 BUFFER_TRACE(frame->bh, "get_write_access");
2029 err = ext4_journal_get_write_access(handle, frame->bh);
2033 unsigned icount1 = icount/2, icount2 = icount - icount1;
2034 unsigned hash2 = dx_get_hash(entries + icount1);
2035 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2038 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2039 err = ext4_journal_get_write_access(handle,
2044 memcpy((char *) entries2, (char *) (entries + icount1),
2045 icount2 * sizeof(struct dx_entry));
2046 dx_set_count(entries, icount1);
2047 dx_set_count(entries2, icount2);
2048 dx_set_limit(entries2, dx_node_limit(dir));
2050 /* Which index block gets the new entry? */
2051 if (at - entries >= icount1) {
2052 frame->at = at = at - entries - icount1 + entries2;
2053 frame->entries = entries = entries2;
2054 swap(frame->bh, bh2);
2056 dx_insert_block(frames + 0, hash2, newblock);
2057 dxtrace(dx_show_index("node", frames[1].entries));
2058 dxtrace(dx_show_index("node",
2059 ((struct dx_node *) bh2->b_data)->entries));
2060 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2065 dxtrace(printk(KERN_DEBUG
2066 "Creating second level index...\n"));
2067 memcpy((char *) entries2, (char *) entries,
2068 icount * sizeof(struct dx_entry));
2069 dx_set_limit(entries2, dx_node_limit(dir));
2072 dx_set_count(entries, 1);
2073 dx_set_block(entries + 0, newblock);
2074 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2076 /* Add new access path frame */
2078 frame->at = at = at - entries + entries2;
2079 frame->entries = entries = entries2;
2081 err = ext4_journal_get_write_access(handle,
2086 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2088 ext4_std_error(inode->i_sb, err);
2092 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
2095 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2099 ext4_std_error(dir->i_sb, err);
2107 * ext4_generic_delete_entry deletes a directory entry by merging it
2108 * with the previous entry
2110 int ext4_generic_delete_entry(handle_t *handle,
2112 struct ext4_dir_entry_2 *de_del,
2113 struct buffer_head *bh,
2118 struct ext4_dir_entry_2 *de, *pde;
2119 unsigned int blocksize = dir->i_sb->s_blocksize;
2124 de = (struct ext4_dir_entry_2 *)entry_buf;
2125 while (i < buf_size - csum_size) {
2126 if (ext4_check_dir_entry(dir, NULL, de, bh,
2127 bh->b_data, bh->b_size, i))
2131 pde->rec_len = ext4_rec_len_to_disk(
2132 ext4_rec_len_from_disk(pde->rec_len,
2134 ext4_rec_len_from_disk(de->rec_len,
2142 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2144 de = ext4_next_entry(de, blocksize);
2149 static int ext4_delete_entry(handle_t *handle,
2151 struct ext4_dir_entry_2 *de_del,
2152 struct buffer_head *bh)
2154 int err, csum_size = 0;
2156 if (ext4_has_inline_data(dir)) {
2157 int has_inline_data = 1;
2158 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2160 if (has_inline_data)
2164 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2165 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2166 csum_size = sizeof(struct ext4_dir_entry_tail);
2168 BUFFER_TRACE(bh, "get_write_access");
2169 err = ext4_journal_get_write_access(handle, bh);
2173 err = ext4_generic_delete_entry(handle, dir, de_del,
2175 dir->i_sb->s_blocksize, csum_size);
2179 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2180 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2187 ext4_std_error(dir->i_sb, err);
2192 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2193 * since this indicates that nlinks count was previously 1.
2195 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2198 if (is_dx(inode) && inode->i_nlink > 1) {
2199 /* limit is 16-bit i_links_count */
2200 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2201 set_nlink(inode, 1);
2202 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2203 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2209 * If a directory had nlink == 1, then we should let it be 1. This indicates
2210 * directory has >EXT4_LINK_MAX subdirs.
2212 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2214 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2219 static int ext4_add_nondir(handle_t *handle,
2220 struct dentry *dentry, struct inode *inode)
2222 int err = ext4_add_entry(handle, dentry, inode);
2224 ext4_mark_inode_dirty(handle, inode);
2225 unlock_new_inode(inode);
2226 d_instantiate(dentry, inode);
2230 unlock_new_inode(inode);
2236 * By the time this is called, we already have created
2237 * the directory cache entry for the new file, but it
2238 * is so far negative - it has no inode.
2240 * If the create succeeds, we fill in the inode information
2241 * with d_instantiate().
2243 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2247 struct inode *inode;
2248 int err, credits, retries = 0;
2250 dquot_initialize(dir);
2252 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2253 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2255 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2256 NULL, EXT4_HT_DIR, credits);
2257 handle = ext4_journal_current_handle();
2258 err = PTR_ERR(inode);
2259 if (!IS_ERR(inode)) {
2260 inode->i_op = &ext4_file_inode_operations;
2261 inode->i_fop = &ext4_file_operations;
2262 ext4_set_aops(inode);
2263 err = ext4_add_nondir(handle, dentry, inode);
2264 if (!err && IS_DIRSYNC(dir))
2265 ext4_handle_sync(handle);
2268 ext4_journal_stop(handle);
2269 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2274 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2275 umode_t mode, dev_t rdev)
2278 struct inode *inode;
2279 int err, credits, retries = 0;
2281 if (!new_valid_dev(rdev))
2284 dquot_initialize(dir);
2286 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2287 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2289 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2290 NULL, EXT4_HT_DIR, credits);
2291 handle = ext4_journal_current_handle();
2292 err = PTR_ERR(inode);
2293 if (!IS_ERR(inode)) {
2294 init_special_inode(inode, inode->i_mode, rdev);
2295 inode->i_op = &ext4_special_inode_operations;
2296 err = ext4_add_nondir(handle, dentry, inode);
2297 if (!err && IS_DIRSYNC(dir))
2298 ext4_handle_sync(handle);
2301 ext4_journal_stop(handle);
2302 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2307 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2310 struct inode *inode;
2311 int err, retries = 0;
2313 dquot_initialize(dir);
2316 inode = ext4_new_inode_start_handle(dir, mode,
2319 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2320 4 + EXT4_XATTR_TRANS_BLOCKS);
2321 handle = ext4_journal_current_handle();
2322 err = PTR_ERR(inode);
2323 if (!IS_ERR(inode)) {
2324 inode->i_op = &ext4_file_inode_operations;
2325 inode->i_fop = &ext4_file_operations;
2326 ext4_set_aops(inode);
2327 d_tmpfile(dentry, inode);
2328 err = ext4_orphan_add(handle, inode);
2330 goto err_unlock_inode;
2331 mark_inode_dirty(inode);
2332 unlock_new_inode(inode);
2335 ext4_journal_stop(handle);
2336 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2340 ext4_journal_stop(handle);
2341 unlock_new_inode(inode);
2345 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2346 struct ext4_dir_entry_2 *de,
2347 int blocksize, int csum_size,
2348 unsigned int parent_ino, int dotdot_real_len)
2350 de->inode = cpu_to_le32(inode->i_ino);
2352 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2354 strcpy(de->name, ".");
2355 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2357 de = ext4_next_entry(de, blocksize);
2358 de->inode = cpu_to_le32(parent_ino);
2360 if (!dotdot_real_len)
2361 de->rec_len = ext4_rec_len_to_disk(blocksize -
2362 (csum_size + EXT4_DIR_REC_LEN(1)),
2365 de->rec_len = ext4_rec_len_to_disk(
2366 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2367 strcpy(de->name, "..");
2368 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2370 return ext4_next_entry(de, blocksize);
2373 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2374 struct inode *inode)
2376 struct buffer_head *dir_block = NULL;
2377 struct ext4_dir_entry_2 *de;
2378 struct ext4_dir_entry_tail *t;
2379 ext4_lblk_t block = 0;
2380 unsigned int blocksize = dir->i_sb->s_blocksize;
2384 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2385 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2386 csum_size = sizeof(struct ext4_dir_entry_tail);
2388 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2389 err = ext4_try_create_inline_dir(handle, dir, inode);
2390 if (err < 0 && err != -ENOSPC)
2397 dir_block = ext4_append(handle, inode, &block);
2398 if (IS_ERR(dir_block))
2399 return PTR_ERR(dir_block);
2400 BUFFER_TRACE(dir_block, "get_write_access");
2401 err = ext4_journal_get_write_access(handle, dir_block);
2404 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2405 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2406 set_nlink(inode, 2);
2408 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2409 initialize_dirent_tail(t, blocksize);
2412 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2413 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2416 set_buffer_verified(dir_block);
2422 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2425 struct inode *inode;
2426 int err, credits, retries = 0;
2428 if (EXT4_DIR_LINK_MAX(dir))
2431 dquot_initialize(dir);
2433 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2434 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2436 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2438 0, NULL, EXT4_HT_DIR, credits);
2439 handle = ext4_journal_current_handle();
2440 err = PTR_ERR(inode);
2444 inode->i_op = &ext4_dir_inode_operations;
2445 inode->i_fop = &ext4_dir_operations;
2446 err = ext4_init_new_dir(handle, dir, inode);
2448 goto out_clear_inode;
2449 err = ext4_mark_inode_dirty(handle, inode);
2451 err = ext4_add_entry(handle, dentry, inode);
2455 unlock_new_inode(inode);
2456 ext4_mark_inode_dirty(handle, inode);
2460 ext4_inc_count(handle, dir);
2461 ext4_update_dx_flag(dir);
2462 err = ext4_mark_inode_dirty(handle, dir);
2464 goto out_clear_inode;
2465 unlock_new_inode(inode);
2466 d_instantiate(dentry, inode);
2467 if (IS_DIRSYNC(dir))
2468 ext4_handle_sync(handle);
2472 ext4_journal_stop(handle);
2473 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2479 * routine to check that the specified directory is empty (for rmdir)
2481 static int empty_dir(struct inode *inode)
2483 unsigned int offset;
2484 struct buffer_head *bh;
2485 struct ext4_dir_entry_2 *de, *de1;
2486 struct super_block *sb;
2489 if (ext4_has_inline_data(inode)) {
2490 int has_inline_data = 1;
2492 err = empty_inline_dir(inode, &has_inline_data);
2493 if (has_inline_data)
2498 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2499 EXT4_ERROR_INODE(inode, "invalid size");
2502 bh = ext4_read_dirblock(inode, 0, EITHER);
2506 de = (struct ext4_dir_entry_2 *) bh->b_data;
2507 de1 = ext4_next_entry(de, sb->s_blocksize);
2508 if (le32_to_cpu(de->inode) != inode->i_ino ||
2509 !le32_to_cpu(de1->inode) ||
2510 strcmp(".", de->name) ||
2511 strcmp("..", de1->name)) {
2512 ext4_warning(inode->i_sb,
2513 "bad directory (dir #%lu) - no `.' or `..'",
2518 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2519 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2520 de = ext4_next_entry(de1, sb->s_blocksize);
2521 while (offset < inode->i_size) {
2522 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2523 unsigned int lblock;
2526 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2527 bh = ext4_read_dirblock(inode, lblock, EITHER);
2530 de = (struct ext4_dir_entry_2 *) bh->b_data;
2532 if (ext4_check_dir_entry(inode, NULL, de, bh,
2533 bh->b_data, bh->b_size, offset)) {
2534 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2536 offset = (offset | (sb->s_blocksize - 1)) + 1;
2539 if (le32_to_cpu(de->inode)) {
2543 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2544 de = ext4_next_entry(de, sb->s_blocksize);
2551 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2552 * such inodes, starting at the superblock, in case we crash before the
2553 * file is closed/deleted, or in case the inode truncate spans multiple
2554 * transactions and the last transaction is not recovered after a crash.
2556 * At filesystem recovery time, we walk this list deleting unlinked
2557 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2559 * Orphan list manipulation functions must be called under i_mutex unless
2560 * we are just creating the inode or deleting it.
2562 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2564 struct super_block *sb = inode->i_sb;
2565 struct ext4_sb_info *sbi = EXT4_SB(sb);
2566 struct ext4_iloc iloc;
2570 if (!sbi->s_journal)
2573 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2574 !mutex_is_locked(&inode->i_mutex));
2576 * Exit early if inode already is on orphan list. This is a big speedup
2577 * since we don't have to contend on the global s_orphan_lock.
2579 if (!list_empty(&EXT4_I(inode)->i_orphan))
2583 * Orphan handling is only valid for files with data blocks
2584 * being truncated, or files being unlinked. Note that we either
2585 * hold i_mutex, or the inode can not be referenced from outside,
2586 * so i_nlink should not be bumped due to race
2588 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2589 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2591 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2592 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2596 err = ext4_reserve_inode_write(handle, inode, &iloc);
2600 mutex_lock(&sbi->s_orphan_lock);
2602 * Due to previous errors inode may be already a part of on-disk
2603 * orphan list. If so skip on-disk list modification.
2605 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2606 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2607 /* Insert this inode at the head of the on-disk orphan list */
2608 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2609 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2612 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2613 mutex_unlock(&sbi->s_orphan_lock);
2616 err = ext4_handle_dirty_super(handle, sb);
2617 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2622 * We have to remove inode from in-memory list if
2623 * addition to on disk orphan list failed. Stray orphan
2624 * list entries can cause panics at unmount time.
2626 mutex_lock(&sbi->s_orphan_lock);
2627 list_del(&EXT4_I(inode)->i_orphan);
2628 mutex_unlock(&sbi->s_orphan_lock);
2631 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2632 jbd_debug(4, "orphan inode %lu will point to %d\n",
2633 inode->i_ino, NEXT_ORPHAN(inode));
2635 ext4_std_error(sb, err);
2640 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2641 * of such inodes stored on disk, because it is finally being cleaned up.
2643 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2645 struct list_head *prev;
2646 struct ext4_inode_info *ei = EXT4_I(inode);
2647 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2649 struct ext4_iloc iloc;
2652 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2655 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2656 !mutex_is_locked(&inode->i_mutex));
2657 /* Do this quick check before taking global s_orphan_lock. */
2658 if (list_empty(&ei->i_orphan))
2662 /* Grab inode buffer early before taking global s_orphan_lock */
2663 err = ext4_reserve_inode_write(handle, inode, &iloc);
2666 mutex_lock(&sbi->s_orphan_lock);
2667 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2669 prev = ei->i_orphan.prev;
2670 list_del_init(&ei->i_orphan);
2672 /* If we're on an error path, we may not have a valid
2673 * transaction handle with which to update the orphan list on
2674 * disk, but we still need to remove the inode from the linked
2675 * list in memory. */
2676 if (!handle || err) {
2677 mutex_unlock(&sbi->s_orphan_lock);
2681 ino_next = NEXT_ORPHAN(inode);
2682 if (prev == &sbi->s_orphan) {
2683 jbd_debug(4, "superblock will point to %u\n", ino_next);
2684 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2685 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2687 mutex_unlock(&sbi->s_orphan_lock);
2690 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2691 mutex_unlock(&sbi->s_orphan_lock);
2692 err = ext4_handle_dirty_super(handle, inode->i_sb);
2694 struct ext4_iloc iloc2;
2695 struct inode *i_prev =
2696 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2698 jbd_debug(4, "orphan inode %lu will point to %u\n",
2699 i_prev->i_ino, ino_next);
2700 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2702 mutex_unlock(&sbi->s_orphan_lock);
2705 NEXT_ORPHAN(i_prev) = ino_next;
2706 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2707 mutex_unlock(&sbi->s_orphan_lock);
2711 NEXT_ORPHAN(inode) = 0;
2712 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2714 ext4_std_error(inode->i_sb, err);
2722 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2725 struct inode *inode;
2726 struct buffer_head *bh;
2727 struct ext4_dir_entry_2 *de;
2728 handle_t *handle = NULL;
2730 /* Initialize quotas before so that eventual writes go in
2731 * separate transaction */
2732 dquot_initialize(dir);
2733 dquot_initialize(dentry->d_inode);
2736 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2742 inode = dentry->d_inode;
2745 if (le32_to_cpu(de->inode) != inode->i_ino)
2748 retval = -ENOTEMPTY;
2749 if (!empty_dir(inode))
2752 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2753 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2754 if (IS_ERR(handle)) {
2755 retval = PTR_ERR(handle);
2760 if (IS_DIRSYNC(dir))
2761 ext4_handle_sync(handle);
2763 retval = ext4_delete_entry(handle, dir, de, bh);
2766 if (!EXT4_DIR_LINK_EMPTY(inode))
2767 ext4_warning(inode->i_sb,
2768 "empty directory has too many links (%d)",
2772 /* There's no need to set i_disksize: the fact that i_nlink is
2773 * zero will ensure that the right thing happens during any
2776 ext4_orphan_add(handle, inode);
2777 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2778 ext4_mark_inode_dirty(handle, inode);
2779 ext4_dec_count(handle, dir);
2780 ext4_update_dx_flag(dir);
2781 ext4_mark_inode_dirty(handle, dir);
2786 ext4_journal_stop(handle);
2790 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2793 struct inode *inode;
2794 struct buffer_head *bh;
2795 struct ext4_dir_entry_2 *de;
2796 handle_t *handle = NULL;
2798 trace_ext4_unlink_enter(dir, dentry);
2799 /* Initialize quotas before so that eventual writes go
2800 * in separate transaction */
2801 dquot_initialize(dir);
2802 dquot_initialize(dentry->d_inode);
2805 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2811 inode = dentry->d_inode;
2814 if (le32_to_cpu(de->inode) != inode->i_ino)
2817 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2818 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2819 if (IS_ERR(handle)) {
2820 retval = PTR_ERR(handle);
2825 if (IS_DIRSYNC(dir))
2826 ext4_handle_sync(handle);
2828 if (!inode->i_nlink) {
2829 ext4_warning(inode->i_sb,
2830 "Deleting nonexistent file (%lu), %d",
2831 inode->i_ino, inode->i_nlink);
2832 set_nlink(inode, 1);
2834 retval = ext4_delete_entry(handle, dir, de, bh);
2837 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2838 ext4_update_dx_flag(dir);
2839 ext4_mark_inode_dirty(handle, dir);
2841 if (!inode->i_nlink)
2842 ext4_orphan_add(handle, inode);
2843 inode->i_ctime = ext4_current_time(inode);
2844 ext4_mark_inode_dirty(handle, inode);
2850 ext4_journal_stop(handle);
2851 trace_ext4_unlink_exit(dentry, retval);
2855 static int ext4_symlink(struct inode *dir,
2856 struct dentry *dentry, const char *symname)
2859 struct inode *inode;
2860 int l, err, retries = 0;
2863 l = strlen(symname)+1;
2864 if (l > dir->i_sb->s_blocksize)
2865 return -ENAMETOOLONG;
2867 dquot_initialize(dir);
2869 if (l > EXT4_N_BLOCKS * 4) {
2871 * For non-fast symlinks, we just allocate inode and put it on
2872 * orphan list in the first transaction => we need bitmap,
2873 * group descriptor, sb, inode block, quota blocks, and
2874 * possibly selinux xattr blocks.
2876 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2877 EXT4_XATTR_TRANS_BLOCKS;
2880 * Fast symlink. We have to add entry to directory
2881 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2882 * allocate new inode (bitmap, group descriptor, inode block,
2883 * quota blocks, sb is already counted in previous macros).
2885 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2886 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
2889 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
2890 &dentry->d_name, 0, NULL,
2891 EXT4_HT_DIR, credits);
2892 handle = ext4_journal_current_handle();
2893 err = PTR_ERR(inode);
2897 if (l > EXT4_N_BLOCKS * 4) {
2898 inode->i_op = &ext4_symlink_inode_operations;
2899 ext4_set_aops(inode);
2901 * We cannot call page_symlink() with transaction started
2902 * because it calls into ext4_write_begin() which can wait
2903 * for transaction commit if we are running out of space
2904 * and thus we deadlock. So we have to stop transaction now
2905 * and restart it when symlink contents is written.
2907 * To keep fs consistent in case of crash, we have to put inode
2908 * to orphan list in the mean time.
2911 err = ext4_orphan_add(handle, inode);
2912 ext4_journal_stop(handle);
2914 goto err_drop_inode;
2915 err = __page_symlink(inode, symname, l, 1);
2917 goto err_drop_inode;
2919 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2920 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2922 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2923 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2924 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2925 if (IS_ERR(handle)) {
2926 err = PTR_ERR(handle);
2927 goto err_drop_inode;
2929 set_nlink(inode, 1);
2930 err = ext4_orphan_del(handle, inode);
2932 ext4_journal_stop(handle);
2934 goto err_drop_inode;
2937 /* clear the extent format for fast symlink */
2938 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2939 inode->i_op = &ext4_fast_symlink_inode_operations;
2940 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2941 inode->i_size = l-1;
2943 EXT4_I(inode)->i_disksize = inode->i_size;
2944 err = ext4_add_nondir(handle, dentry, inode);
2945 if (!err && IS_DIRSYNC(dir))
2946 ext4_handle_sync(handle);
2950 ext4_journal_stop(handle);
2951 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2955 unlock_new_inode(inode);
2960 static int ext4_link(struct dentry *old_dentry,
2961 struct inode *dir, struct dentry *dentry)
2964 struct inode *inode = old_dentry->d_inode;
2965 int err, retries = 0;
2967 if (inode->i_nlink >= EXT4_LINK_MAX)
2970 dquot_initialize(dir);
2973 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2974 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2975 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
2977 return PTR_ERR(handle);
2979 if (IS_DIRSYNC(dir))
2980 ext4_handle_sync(handle);
2982 inode->i_ctime = ext4_current_time(inode);
2983 ext4_inc_count(handle, inode);
2986 err = ext4_add_entry(handle, dentry, inode);
2988 ext4_mark_inode_dirty(handle, inode);
2989 /* this can happen only for tmpfile being
2990 * linked the first time
2992 if (inode->i_nlink == 1)
2993 ext4_orphan_del(handle, inode);
2994 d_instantiate(dentry, inode);
2999 ext4_journal_stop(handle);
3000 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3007 * Try to find buffer head where contains the parent block.
3008 * It should be the inode block if it is inlined or the 1st block
3009 * if it is a normal dir.
3011 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3012 struct inode *inode,
3014 struct ext4_dir_entry_2 **parent_de,
3017 struct buffer_head *bh;
3019 if (!ext4_has_inline_data(inode)) {
3020 bh = ext4_read_dirblock(inode, 0, EITHER);
3022 *retval = PTR_ERR(bh);
3025 *parent_de = ext4_next_entry(
3026 (struct ext4_dir_entry_2 *)bh->b_data,
3027 inode->i_sb->s_blocksize);
3032 return ext4_get_first_inline_block(inode, parent_de, retval);
3035 struct ext4_renament {
3037 struct dentry *dentry;
3038 struct inode *inode;
3040 int dir_nlink_delta;
3042 /* entry for "dentry" */
3043 struct buffer_head *bh;
3044 struct ext4_dir_entry_2 *de;
3047 /* entry for ".." in inode if it's a directory */
3048 struct buffer_head *dir_bh;
3049 struct ext4_dir_entry_2 *parent_de;
3053 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3057 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3058 &retval, &ent->parent_de,
3062 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3064 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3065 return ext4_journal_get_write_access(handle, ent->dir_bh);
3068 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3073 ent->parent_de->inode = cpu_to_le32(dir_ino);
3074 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3075 if (!ent->dir_inlined) {
3076 if (is_dx(ent->inode)) {
3077 retval = ext4_handle_dirty_dx_node(handle,
3081 retval = ext4_handle_dirty_dirent_node(handle,
3086 retval = ext4_mark_inode_dirty(handle, ent->inode);
3089 ext4_std_error(ent->dir->i_sb, retval);
3095 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3096 unsigned ino, unsigned file_type)
3100 BUFFER_TRACE(ent->bh, "get write access");
3101 retval = ext4_journal_get_write_access(handle, ent->bh);
3104 ent->de->inode = cpu_to_le32(ino);
3105 if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb,
3106 EXT4_FEATURE_INCOMPAT_FILETYPE))
3107 ent->de->file_type = file_type;
3108 ent->dir->i_version++;
3109 ent->dir->i_ctime = ent->dir->i_mtime =
3110 ext4_current_time(ent->dir);
3111 ext4_mark_inode_dirty(handle, ent->dir);
3112 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3113 if (!ent->inlined) {
3114 retval = ext4_handle_dirty_dirent_node(handle,
3116 if (unlikely(retval)) {
3117 ext4_std_error(ent->dir->i_sb, retval);
3127 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3128 const struct qstr *d_name)
3130 int retval = -ENOENT;
3131 struct buffer_head *bh;
3132 struct ext4_dir_entry_2 *de;
3134 bh = ext4_find_entry(dir, d_name, &de, NULL);
3138 retval = ext4_delete_entry(handle, dir, de, bh);
3144 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3149 * ent->de could have moved from under us during htree split, so make
3150 * sure that we are deleting the right entry. We might also be pointing
3151 * to a stale entry in the unused part of ent->bh so just checking inum
3152 * and the name isn't enough.
3154 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3155 ent->de->name_len != ent->dentry->d_name.len ||
3156 strncmp(ent->de->name, ent->dentry->d_name.name,
3157 ent->de->name_len) ||
3159 retval = ext4_find_delete_entry(handle, ent->dir,
3160 &ent->dentry->d_name);
3162 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3163 if (retval == -ENOENT) {
3164 retval = ext4_find_delete_entry(handle, ent->dir,
3165 &ent->dentry->d_name);
3170 ext4_warning(ent->dir->i_sb,
3171 "Deleting old file (%lu), %d, error=%d",
3172 ent->dir->i_ino, ent->dir->i_nlink, retval);
3176 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3178 if (ent->dir_nlink_delta) {
3179 if (ent->dir_nlink_delta == -1)
3180 ext4_dec_count(handle, ent->dir);
3182 ext4_inc_count(handle, ent->dir);
3183 ext4_mark_inode_dirty(handle, ent->dir);
3188 * Anybody can rename anything with this: the permission checks are left to the
3189 * higher-level routines.
3191 * n.b. old_{dentry,inode) refers to the source dentry/inode
3192 * while new_{dentry,inode) refers to the destination dentry/inode
3193 * This comes from rename(const char *oldpath, const char *newpath)
3195 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3196 struct inode *new_dir, struct dentry *new_dentry)
3198 handle_t *handle = NULL;
3199 struct ext4_renament old = {
3201 .dentry = old_dentry,
3202 .inode = old_dentry->d_inode,
3204 struct ext4_renament new = {
3206 .dentry = new_dentry,
3207 .inode = new_dentry->d_inode,
3212 dquot_initialize(old.dir);
3213 dquot_initialize(new.dir);
3215 /* Initialize quotas before so that eventual writes go
3216 * in separate transaction */
3218 dquot_initialize(new.inode);
3220 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3222 return PTR_ERR(old.bh);
3224 * Check for inode number is _not_ due to possible IO errors.
3225 * We might rmdir the source, keep it as pwd of some process
3226 * and merrily kill the link to whatever was created under the
3227 * same name. Goodbye sticky bit ;-<
3230 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3233 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3234 &new.de, &new.inlined);
3235 if (IS_ERR(new.bh)) {
3236 retval = PTR_ERR(new.bh);
3245 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3246 ext4_alloc_da_blocks(old.inode);
3248 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3249 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3250 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3252 return PTR_ERR(handle);
3254 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3255 ext4_handle_sync(handle);
3257 if (S_ISDIR(old.inode->i_mode)) {
3259 retval = -ENOTEMPTY;
3260 if (!empty_dir(new.inode))
3264 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3267 retval = ext4_rename_dir_prepare(handle, &old);
3272 * If we're renaming a file within an inline_data dir and adding or
3273 * setting the new dirent causes a conversion from inline_data to
3274 * extents/blockmap, we need to force the dirent delete code to
3275 * re-read the directory, or else we end up trying to delete a dirent
3276 * from what is now the extent tree root (or a block map).
3278 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3279 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3281 retval = ext4_add_entry(handle, new.dentry, old.inode);
3285 retval = ext4_setent(handle, &new,
3286 old.inode->i_ino, old.de->file_type);
3291 force_reread = !ext4_test_inode_flag(new.dir,
3292 EXT4_INODE_INLINE_DATA);
3295 * Like most other Unix systems, set the ctime for inodes on a
3298 old.inode->i_ctime = ext4_current_time(old.inode);
3299 ext4_mark_inode_dirty(handle, old.inode);
3304 ext4_rename_delete(handle, &old, force_reread);
3307 ext4_dec_count(handle, new.inode);
3308 new.inode->i_ctime = ext4_current_time(new.inode);
3310 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3311 ext4_update_dx_flag(old.dir);
3313 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3317 ext4_dec_count(handle, old.dir);
3319 /* checked empty_dir above, can't have another parent,
3320 * ext4_dec_count() won't work for many-linked dirs */
3321 clear_nlink(new.inode);
3323 ext4_inc_count(handle, new.dir);
3324 ext4_update_dx_flag(new.dir);
3325 ext4_mark_inode_dirty(handle, new.dir);
3328 ext4_mark_inode_dirty(handle, old.dir);
3330 ext4_mark_inode_dirty(handle, new.inode);
3331 if (!new.inode->i_nlink)
3332 ext4_orphan_add(handle, new.inode);
3341 ext4_journal_stop(handle);
3345 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3346 struct inode *new_dir, struct dentry *new_dentry)
3348 handle_t *handle = NULL;
3349 struct ext4_renament old = {
3351 .dentry = old_dentry,
3352 .inode = old_dentry->d_inode,
3354 struct ext4_renament new = {
3356 .dentry = new_dentry,
3357 .inode = new_dentry->d_inode,
3362 dquot_initialize(old.dir);
3363 dquot_initialize(new.dir);
3365 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3366 &old.de, &old.inlined);
3368 return PTR_ERR(old.bh);
3370 * Check for inode number is _not_ due to possible IO errors.
3371 * We might rmdir the source, keep it as pwd of some process
3372 * and merrily kill the link to whatever was created under the
3373 * same name. Goodbye sticky bit ;-<
3376 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3379 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3380 &new.de, &new.inlined);
3381 if (IS_ERR(new.bh)) {
3382 retval = PTR_ERR(new.bh);
3386 /* RENAME_EXCHANGE case: old *and* new must both exist */
3387 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3390 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3391 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3392 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3394 return PTR_ERR(handle);
3396 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3397 ext4_handle_sync(handle);
3399 if (S_ISDIR(old.inode->i_mode)) {
3401 retval = ext4_rename_dir_prepare(handle, &old);
3405 if (S_ISDIR(new.inode->i_mode)) {
3407 retval = ext4_rename_dir_prepare(handle, &new);
3413 * Other than the special case of overwriting a directory, parents'
3414 * nlink only needs to be modified if this is a cross directory rename.
3416 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3417 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3418 new.dir_nlink_delta = -old.dir_nlink_delta;
3420 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3421 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3425 new_file_type = new.de->file_type;
3426 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3430 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3435 * Like most other Unix systems, set the ctime for inodes on a
3438 old.inode->i_ctime = ext4_current_time(old.inode);
3439 new.inode->i_ctime = ext4_current_time(new.inode);
3440 ext4_mark_inode_dirty(handle, old.inode);
3441 ext4_mark_inode_dirty(handle, new.inode);
3444 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3449 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3453 ext4_update_dir_count(handle, &old);
3454 ext4_update_dir_count(handle, &new);
3463 ext4_journal_stop(handle);
3467 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3468 struct inode *new_dir, struct dentry *new_dentry,
3471 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
3474 if (flags & RENAME_EXCHANGE) {
3475 return ext4_cross_rename(old_dir, old_dentry,
3476 new_dir, new_dentry);
3479 * Existence checking was done by the VFS, otherwise "RENAME_NOREPLACE"
3480 * is equivalent to regular rename.
3482 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry);
3486 * directories can handle most operations...
3488 const struct inode_operations ext4_dir_inode_operations = {
3489 .create = ext4_create,
3490 .lookup = ext4_lookup,
3492 .unlink = ext4_unlink,
3493 .symlink = ext4_symlink,
3494 .mkdir = ext4_mkdir,
3495 .rmdir = ext4_rmdir,
3496 .mknod = ext4_mknod,
3497 .tmpfile = ext4_tmpfile,
3498 .rename2 = ext4_rename2,
3499 .setattr = ext4_setattr,
3500 .setxattr = generic_setxattr,
3501 .getxattr = generic_getxattr,
3502 .listxattr = ext4_listxattr,
3503 .removexattr = generic_removexattr,
3504 .get_acl = ext4_get_acl,
3505 .set_acl = ext4_set_acl,
3506 .fiemap = ext4_fiemap,
3509 const struct inode_operations ext4_special_inode_operations = {
3510 .setattr = ext4_setattr,
3511 .setxattr = generic_setxattr,
3512 .getxattr = generic_getxattr,
3513 .listxattr = ext4_listxattr,
3514 .removexattr = generic_removexattr,
3515 .get_acl = ext4_get_acl,
3516 .set_acl = ext4_set_acl,