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/time.h>
30 #include <linux/fcntl.h>
31 #include <linux/stat.h>
32 #include <linux/string.h>
33 #include <linux/quotaops.h>
34 #include <linux/buffer_head.h>
35 #include <linux/bio.h>
37 #include "ext4_jbd2.h"
42 #include <trace/events/ext4.h>
44 * define how far ahead to read directories while searching them.
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 static struct buffer_head *ext4_append(handle_t *handle,
54 struct buffer_head *bh;
57 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
58 ((inode->i_size >> 10) >=
59 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
60 return ERR_PTR(-ENOSPC);
62 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
64 bh = ext4_bread(handle, inode, *block, 1);
67 inode->i_size += inode->i_sb->s_blocksize;
68 EXT4_I(inode)->i_disksize = inode->i_size;
69 BUFFER_TRACE(bh, "get_write_access");
70 err = ext4_journal_get_write_access(handle, bh);
73 ext4_std_error(inode->i_sb, err);
79 static int ext4_dx_csum_verify(struct inode *inode,
80 struct ext4_dir_entry *dirent);
86 #define ext4_read_dirblock(inode, block, type) \
87 __ext4_read_dirblock((inode), (block), (type), __LINE__)
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
94 struct buffer_head *bh;
95 struct ext4_dir_entry *dirent;
98 bh = ext4_bread(NULL, inode, block, 0);
100 __ext4_warning(inode->i_sb, __func__, line,
101 "error %ld reading directory block "
102 "(ino %lu, block %lu)", PTR_ERR(bh), inode->i_ino,
103 (unsigned long) block);
108 ext4_error_inode(inode, __func__, line, block, "Directory hole found");
109 return ERR_PTR(-EIO);
111 dirent = (struct ext4_dir_entry *) bh->b_data;
112 /* Determine whether or not we have an index block */
116 else if (ext4_rec_len_from_disk(dirent->rec_len,
117 inode->i_sb->s_blocksize) ==
118 inode->i_sb->s_blocksize)
121 if (!is_dx_block && type == INDEX) {
122 ext4_error_inode(inode, __func__, line, block,
123 "directory leaf block found instead of index block");
124 return ERR_PTR(-EIO);
126 if (!ext4_has_metadata_csum(inode->i_sb) ||
131 * An empty leaf block can get mistaken for a index block; for
132 * this reason, we can only check the index checksum when the
133 * caller is sure it should be an index block.
135 if (is_dx_block && type == INDEX) {
136 if (ext4_dx_csum_verify(inode, dirent))
137 set_buffer_verified(bh);
139 ext4_error_inode(inode, __func__, line, block,
140 "Directory index failed checksum");
142 return ERR_PTR(-EIO);
146 if (ext4_dirent_csum_verify(inode, dirent))
147 set_buffer_verified(bh);
149 ext4_error_inode(inode, __func__, line, block,
150 "Directory block failed checksum");
152 return ERR_PTR(-EIO);
159 #define assert(test) J_ASSERT(test)
163 #define dxtrace(command) command
165 #define dxtrace(command)
189 * dx_root_info is laid out so that if it should somehow get overlaid by a
190 * dirent the two low bits of the hash version will be zero. Therefore, the
191 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
196 struct fake_dirent dot;
198 struct fake_dirent dotdot;
202 __le32 reserved_zero;
204 u8 info_length; /* 8 */
209 struct dx_entry entries[0];
214 struct fake_dirent fake;
215 struct dx_entry entries[0];
221 struct buffer_head *bh;
222 struct dx_entry *entries;
234 * This goes at the end of each htree block.
238 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
241 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
242 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
243 static inline unsigned dx_get_hash(struct dx_entry *entry);
244 static void dx_set_hash(struct dx_entry *entry, unsigned value);
245 static unsigned dx_get_count(struct dx_entry *entries);
246 static unsigned dx_get_limit(struct dx_entry *entries);
247 static void dx_set_count(struct dx_entry *entries, unsigned value);
248 static void dx_set_limit(struct dx_entry *entries, unsigned value);
249 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
250 static unsigned dx_node_limit(struct inode *dir);
251 static struct dx_frame *dx_probe(const struct qstr *d_name,
253 struct dx_hash_info *hinfo,
254 struct dx_frame *frame);
255 static void dx_release(struct dx_frame *frames);
256 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
257 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
258 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
259 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
260 struct dx_map_entry *offsets, int count, unsigned blocksize);
261 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
262 static void dx_insert_block(struct dx_frame *frame,
263 u32 hash, ext4_lblk_t block);
264 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
265 struct dx_frame *frame,
266 struct dx_frame *frames,
268 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
269 const struct qstr *d_name,
270 struct ext4_dir_entry_2 **res_dir);
271 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
272 struct inode *inode);
274 /* checksumming functions */
275 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
276 unsigned int blocksize)
278 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
279 t->det_rec_len = ext4_rec_len_to_disk(
280 sizeof(struct ext4_dir_entry_tail), blocksize);
281 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
284 /* Walk through a dirent block to find a checksum "dirent" at the tail */
285 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
286 struct ext4_dir_entry *de)
288 struct ext4_dir_entry_tail *t;
291 struct ext4_dir_entry *d, *top;
294 top = (struct ext4_dir_entry *)(((void *)de) +
295 (EXT4_BLOCK_SIZE(inode->i_sb) -
296 sizeof(struct ext4_dir_entry_tail)));
297 while (d < top && d->rec_len)
298 d = (struct ext4_dir_entry *)(((void *)d) +
299 le16_to_cpu(d->rec_len));
304 t = (struct ext4_dir_entry_tail *)d;
306 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
309 if (t->det_reserved_zero1 ||
310 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
311 t->det_reserved_zero2 ||
312 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
318 static __le32 ext4_dirent_csum(struct inode *inode,
319 struct ext4_dir_entry *dirent, int size)
321 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
322 struct ext4_inode_info *ei = EXT4_I(inode);
325 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
326 return cpu_to_le32(csum);
329 static void warn_no_space_for_csum(struct inode *inode)
331 ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
332 "checksum. Please run e2fsck -D.", inode->i_ino);
335 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
337 struct ext4_dir_entry_tail *t;
339 if (!ext4_has_metadata_csum(inode->i_sb))
342 t = get_dirent_tail(inode, dirent);
344 warn_no_space_for_csum(inode);
348 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
349 (void *)t - (void *)dirent))
355 static void ext4_dirent_csum_set(struct inode *inode,
356 struct ext4_dir_entry *dirent)
358 struct ext4_dir_entry_tail *t;
360 if (!ext4_has_metadata_csum(inode->i_sb))
363 t = get_dirent_tail(inode, dirent);
365 warn_no_space_for_csum(inode);
369 t->det_checksum = ext4_dirent_csum(inode, dirent,
370 (void *)t - (void *)dirent);
373 int ext4_handle_dirty_dirent_node(handle_t *handle,
375 struct buffer_head *bh)
377 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
378 return ext4_handle_dirty_metadata(handle, inode, bh);
381 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
382 struct ext4_dir_entry *dirent,
385 struct ext4_dir_entry *dp;
386 struct dx_root_info *root;
389 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
391 else if (le16_to_cpu(dirent->rec_len) == 12) {
392 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
393 if (le16_to_cpu(dp->rec_len) !=
394 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
396 root = (struct dx_root_info *)(((void *)dp + 12));
397 if (root->reserved_zero ||
398 root->info_length != sizeof(struct dx_root_info))
405 *offset = count_offset;
406 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
409 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
410 int count_offset, int count, struct dx_tail *t)
412 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
413 struct ext4_inode_info *ei = EXT4_I(inode);
418 size = count_offset + (count * sizeof(struct dx_entry));
419 save_csum = t->dt_checksum;
421 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
422 csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
423 t->dt_checksum = save_csum;
425 return cpu_to_le32(csum);
428 static int ext4_dx_csum_verify(struct inode *inode,
429 struct ext4_dir_entry *dirent)
431 struct dx_countlimit *c;
433 int count_offset, limit, count;
435 if (!ext4_has_metadata_csum(inode->i_sb))
438 c = get_dx_countlimit(inode, dirent, &count_offset);
440 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
443 limit = le16_to_cpu(c->limit);
444 count = le16_to_cpu(c->count);
445 if (count_offset + (limit * sizeof(struct dx_entry)) >
446 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
447 warn_no_space_for_csum(inode);
450 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
452 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
458 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
460 struct dx_countlimit *c;
462 int count_offset, limit, count;
464 if (!ext4_has_metadata_csum(inode->i_sb))
467 c = get_dx_countlimit(inode, dirent, &count_offset);
469 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
472 limit = le16_to_cpu(c->limit);
473 count = le16_to_cpu(c->count);
474 if (count_offset + (limit * sizeof(struct dx_entry)) >
475 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
476 warn_no_space_for_csum(inode);
479 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
481 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
484 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
486 struct buffer_head *bh)
488 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
489 return ext4_handle_dirty_metadata(handle, inode, bh);
493 * p is at least 6 bytes before the end of page
495 static inline struct ext4_dir_entry_2 *
496 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
498 return (struct ext4_dir_entry_2 *)((char *)p +
499 ext4_rec_len_from_disk(p->rec_len, blocksize));
503 * Future: use high four bits of block for coalesce-on-delete flags
504 * Mask them off for now.
507 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
509 return le32_to_cpu(entry->block) & 0x00ffffff;
512 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
514 entry->block = cpu_to_le32(value);
517 static inline unsigned dx_get_hash(struct dx_entry *entry)
519 return le32_to_cpu(entry->hash);
522 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
524 entry->hash = cpu_to_le32(value);
527 static inline unsigned dx_get_count(struct dx_entry *entries)
529 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
532 static inline unsigned dx_get_limit(struct dx_entry *entries)
534 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
537 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
539 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
542 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
544 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
547 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
549 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
550 EXT4_DIR_REC_LEN(2) - infosize;
552 if (ext4_has_metadata_csum(dir->i_sb))
553 entry_space -= sizeof(struct dx_tail);
554 return entry_space / sizeof(struct dx_entry);
557 static inline unsigned dx_node_limit(struct inode *dir)
559 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
561 if (ext4_has_metadata_csum(dir->i_sb))
562 entry_space -= sizeof(struct dx_tail);
563 return entry_space / sizeof(struct dx_entry);
570 static void dx_show_index(char * label, struct dx_entry *entries)
572 int i, n = dx_get_count (entries);
573 printk(KERN_DEBUG "%s index ", label);
574 for (i = 0; i < n; i++) {
575 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
576 0, (unsigned long)dx_get_block(entries + i));
588 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
589 int size, int show_names)
591 unsigned names = 0, space = 0;
592 char *base = (char *) de;
593 struct dx_hash_info h = *hinfo;
596 while ((char *) de < base + size)
602 int len = de->name_len;
603 char *name = de->name;
604 while (len--) printk("%c", *name++);
605 ext4fs_dirhash(de->name, de->name_len, &h);
606 printk(":%x.%u ", h.hash,
607 (unsigned) ((char *) de - base));
609 space += EXT4_DIR_REC_LEN(de->name_len);
612 de = ext4_next_entry(de, size);
614 printk("(%i)\n", names);
615 return (struct stats) { names, space, 1 };
618 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
619 struct dx_entry *entries, int levels)
621 unsigned blocksize = dir->i_sb->s_blocksize;
622 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
624 struct buffer_head *bh;
626 printk("%i indexed blocks...\n", count);
627 for (i = 0; i < count; i++, entries++)
629 ext4_lblk_t block = dx_get_block(entries);
630 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
631 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
633 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
634 bh = ext4_bread(NULL,dir, block, 0);
635 if (!bh || IS_ERR(bh))
638 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
639 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
640 names += stats.names;
641 space += stats.space;
642 bcount += stats.bcount;
646 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
647 levels ? "" : " ", names, space/bcount,
648 (space/bcount)*100/blocksize);
649 return (struct stats) { names, space, bcount};
651 #endif /* DX_DEBUG */
654 * Probe for a directory leaf block to search.
656 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
657 * error in the directory index, and the caller should fall back to
658 * searching the directory normally. The callers of dx_probe **MUST**
659 * check for this error code, and make sure it never gets reflected
662 static struct dx_frame *
663 dx_probe(const struct qstr *d_name, struct inode *dir,
664 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
666 unsigned count, indirect;
667 struct dx_entry *at, *entries, *p, *q, *m;
668 struct dx_root *root;
669 struct dx_frame *frame = frame_in;
670 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
673 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
674 if (IS_ERR(frame->bh))
675 return (struct dx_frame *) frame->bh;
677 root = (struct dx_root *) frame->bh->b_data;
678 if (root->info.hash_version != DX_HASH_TEA &&
679 root->info.hash_version != DX_HASH_HALF_MD4 &&
680 root->info.hash_version != DX_HASH_LEGACY) {
681 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
682 root->info.hash_version);
685 hinfo->hash_version = root->info.hash_version;
686 if (hinfo->hash_version <= DX_HASH_TEA)
687 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
688 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
690 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
693 if (root->info.unused_flags & 1) {
694 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
695 root->info.unused_flags);
699 if ((indirect = root->info.indirect_levels) > 1) {
700 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
701 root->info.indirect_levels);
705 entries = (struct dx_entry *) (((char *)&root->info) +
706 root->info.info_length);
708 if (dx_get_limit(entries) != dx_root_limit(dir,
709 root->info.info_length)) {
710 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
714 dxtrace(printk("Look up %x", hash));
716 count = dx_get_count(entries);
717 if (!count || count > dx_get_limit(entries)) {
718 ext4_warning(dir->i_sb,
719 "dx entry: no count or count > limit");
724 q = entries + count - 1;
727 dxtrace(printk("."));
728 if (dx_get_hash(m) > hash)
734 if (0) { // linear search cross check
735 unsigned n = count - 1;
739 dxtrace(printk(","));
740 if (dx_get_hash(++at) > hash)
746 assert (at == p - 1);
750 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
751 frame->entries = entries;
756 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
757 if (IS_ERR(frame->bh)) {
758 ret_err = (struct dx_frame *) frame->bh;
762 entries = ((struct dx_node *) frame->bh->b_data)->entries;
764 if (dx_get_limit(entries) != dx_node_limit (dir)) {
765 ext4_warning(dir->i_sb,
766 "dx entry: limit != node limit");
771 while (frame >= frame_in) {
775 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
776 ext4_warning(dir->i_sb,
777 "Corrupt dir inode %lu, running e2fsck is "
778 "recommended.", dir->i_ino);
782 static void dx_release (struct dx_frame *frames)
784 if (frames[0].bh == NULL)
787 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
788 brelse(frames[1].bh);
789 brelse(frames[0].bh);
793 * This function increments the frame pointer to search the next leaf
794 * block, and reads in the necessary intervening nodes if the search
795 * should be necessary. Whether or not the search is necessary is
796 * controlled by the hash parameter. If the hash value is even, then
797 * the search is only continued if the next block starts with that
798 * hash value. This is used if we are searching for a specific file.
800 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
802 * This function returns 1 if the caller should continue to search,
803 * or 0 if it should not. If there is an error reading one of the
804 * index blocks, it will a negative error code.
806 * If start_hash is non-null, it will be filled in with the starting
807 * hash of the next page.
809 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
810 struct dx_frame *frame,
811 struct dx_frame *frames,
815 struct buffer_head *bh;
821 * Find the next leaf page by incrementing the frame pointer.
822 * If we run out of entries in the interior node, loop around and
823 * increment pointer in the parent node. When we break out of
824 * this loop, num_frames indicates the number of interior
825 * nodes need to be read.
828 if (++(p->at) < p->entries + dx_get_count(p->entries))
837 * If the hash is 1, then continue only if the next page has a
838 * continuation hash of any value. This is used for readdir
839 * handling. Otherwise, check to see if the hash matches the
840 * desired contiuation hash. If it doesn't, return since
841 * there's no point to read in the successive index pages.
843 bhash = dx_get_hash(p->at);
846 if ((hash & 1) == 0) {
847 if ((bhash & ~1) != hash)
851 * If the hash is HASH_NB_ALWAYS, we always go to the next
852 * block so no check is necessary
854 while (num_frames--) {
855 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
861 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
868 * This function fills a red-black tree with information from a
869 * directory block. It returns the number directory entries loaded
870 * into the tree. If there is an error it is returned in err.
872 static int htree_dirblock_to_tree(struct file *dir_file,
873 struct inode *dir, ext4_lblk_t block,
874 struct dx_hash_info *hinfo,
875 __u32 start_hash, __u32 start_minor_hash)
877 struct buffer_head *bh;
878 struct ext4_dir_entry_2 *de, *top;
879 int err = 0, count = 0;
881 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
882 (unsigned long)block));
883 bh = ext4_read_dirblock(dir, block, DIRENT);
887 de = (struct ext4_dir_entry_2 *) bh->b_data;
888 top = (struct ext4_dir_entry_2 *) ((char *) de +
889 dir->i_sb->s_blocksize -
890 EXT4_DIR_REC_LEN(0));
891 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
892 if (ext4_check_dir_entry(dir, NULL, de, bh,
893 bh->b_data, bh->b_size,
894 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
895 + ((char *)de - bh->b_data))) {
896 /* silently ignore the rest of the block */
899 ext4fs_dirhash(de->name, de->name_len, hinfo);
900 if ((hinfo->hash < start_hash) ||
901 ((hinfo->hash == start_hash) &&
902 (hinfo->minor_hash < start_minor_hash)))
906 if ((err = ext4_htree_store_dirent(dir_file,
907 hinfo->hash, hinfo->minor_hash, de)) != 0) {
919 * This function fills a red-black tree with information from a
920 * directory. We start scanning the directory in hash order, starting
921 * at start_hash and start_minor_hash.
923 * This function returns the number of entries inserted into the tree,
924 * or a negative error code.
926 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
927 __u32 start_minor_hash, __u32 *next_hash)
929 struct dx_hash_info hinfo;
930 struct ext4_dir_entry_2 *de;
931 struct dx_frame frames[2], *frame;
938 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
939 start_hash, start_minor_hash));
940 dir = file_inode(dir_file);
941 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
942 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
943 if (hinfo.hash_version <= DX_HASH_TEA)
944 hinfo.hash_version +=
945 EXT4_SB(dir->i_sb)->s_hash_unsigned;
946 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
947 if (ext4_has_inline_data(dir)) {
948 int has_inline_data = 1;
949 count = htree_inlinedir_to_tree(dir_file, dir, 0,
953 if (has_inline_data) {
958 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
959 start_hash, start_minor_hash);
963 hinfo.hash = start_hash;
964 hinfo.minor_hash = 0;
965 frame = dx_probe(NULL, dir, &hinfo, frames);
967 return PTR_ERR(frame);
969 /* Add '.' and '..' from the htree header */
970 if (!start_hash && !start_minor_hash) {
971 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
972 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
976 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
977 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
978 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
979 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
985 block = dx_get_block(frame->at);
986 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
987 start_hash, start_minor_hash);
994 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
995 frame, frames, &hashval);
996 *next_hash = hashval;
1002 * Stop if: (a) there are no more entries, or
1003 * (b) we have inserted at least one entry and the
1004 * next hash value is not a continuation
1007 (count && ((hashval & 1) == 0)))
1011 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1012 "next hash: %x\n", count, *next_hash));
1019 static inline int search_dirblock(struct buffer_head *bh,
1021 const struct qstr *d_name,
1022 unsigned int offset,
1023 struct ext4_dir_entry_2 **res_dir)
1025 return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1026 d_name, offset, res_dir);
1030 * Directory block splitting, compacting
1034 * Create map of hash values, offsets, and sizes, stored at end of block.
1035 * Returns number of entries mapped.
1037 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1038 struct dx_hash_info *hinfo,
1039 struct dx_map_entry *map_tail)
1042 char *base = (char *) de;
1043 struct dx_hash_info h = *hinfo;
1045 while ((char *) de < base + blocksize) {
1046 if (de->name_len && de->inode) {
1047 ext4fs_dirhash(de->name, de->name_len, &h);
1049 map_tail->hash = h.hash;
1050 map_tail->offs = ((char *) de - base)>>2;
1051 map_tail->size = le16_to_cpu(de->rec_len);
1055 /* XXX: do we need to check rec_len == 0 case? -Chris */
1056 de = ext4_next_entry(de, blocksize);
1061 /* Sort map by hash value */
1062 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1064 struct dx_map_entry *p, *q, *top = map + count - 1;
1066 /* Combsort until bubble sort doesn't suck */
1068 count = count*10/13;
1069 if (count - 9 < 2) /* 9, 10 -> 11 */
1071 for (p = top, q = p - count; q >= map; p--, q--)
1072 if (p->hash < q->hash)
1075 /* Garden variety bubble sort */
1080 if (q[1].hash >= q[0].hash)
1088 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1090 struct dx_entry *entries = frame->entries;
1091 struct dx_entry *old = frame->at, *new = old + 1;
1092 int count = dx_get_count(entries);
1094 assert(count < dx_get_limit(entries));
1095 assert(old < entries + count);
1096 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1097 dx_set_hash(new, hash);
1098 dx_set_block(new, block);
1099 dx_set_count(entries, count + 1);
1103 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1105 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1106 * `de != NULL' is guaranteed by caller.
1108 static inline int ext4_match (int len, const char * const name,
1109 struct ext4_dir_entry_2 * de)
1111 if (len != de->name_len)
1115 return !memcmp(name, de->name, len);
1119 * Returns 0 if not found, -1 on failure, and 1 on success
1121 int search_dir(struct buffer_head *bh,
1125 const struct qstr *d_name,
1126 unsigned int offset,
1127 struct ext4_dir_entry_2 **res_dir)
1129 struct ext4_dir_entry_2 * de;
1132 const char *name = d_name->name;
1133 int namelen = d_name->len;
1135 de = (struct ext4_dir_entry_2 *)search_buf;
1136 dlimit = search_buf + buf_size;
1137 while ((char *) de < dlimit) {
1138 /* this code is executed quadratically often */
1139 /* do minimal checking `by hand' */
1141 if ((char *) de + namelen <= dlimit &&
1142 ext4_match (namelen, name, de)) {
1143 /* found a match - just to be sure, do a full check */
1144 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1145 bh->b_size, offset))
1150 /* prevent looping on a bad block */
1151 de_len = ext4_rec_len_from_disk(de->rec_len,
1152 dir->i_sb->s_blocksize);
1156 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1161 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1162 struct ext4_dir_entry *de)
1164 struct super_block *sb = dir->i_sb;
1170 if (de->inode == 0 &&
1171 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1180 * finds an entry in the specified directory with the wanted name. It
1181 * returns the cache buffer in which the entry was found, and the entry
1182 * itself (as a parameter - res_dir). It does NOT read the inode of the
1183 * entry - you'll have to do that yourself if you want to.
1185 * The returned buffer_head has ->b_count elevated. The caller is expected
1186 * to brelse() it when appropriate.
1188 static struct buffer_head * ext4_find_entry (struct inode *dir,
1189 const struct qstr *d_name,
1190 struct ext4_dir_entry_2 **res_dir,
1193 struct super_block *sb;
1194 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1195 struct buffer_head *bh, *ret = NULL;
1196 ext4_lblk_t start, block, b;
1197 const u8 *name = d_name->name;
1198 int ra_max = 0; /* Number of bh's in the readahead
1200 int ra_ptr = 0; /* Current index into readahead
1203 ext4_lblk_t nblocks;
1208 namelen = d_name->len;
1209 if (namelen > EXT4_NAME_LEN)
1212 if (ext4_has_inline_data(dir)) {
1213 int has_inline_data = 1;
1214 ret = ext4_find_inline_entry(dir, d_name, res_dir,
1216 if (has_inline_data) {
1223 if ((namelen <= 2) && (name[0] == '.') &&
1224 (name[1] == '.' || name[1] == '\0')) {
1226 * "." or ".." will only be in the first block
1227 * NFS may look up ".."; "." should be handled by the VFS
1234 bh = ext4_dx_find_entry(dir, d_name, res_dir);
1236 * On success, or if the error was file not found,
1237 * return. Otherwise, fall back to doing a search the
1238 * old fashioned way.
1240 if (!IS_ERR(bh) || PTR_ERR(bh) != ERR_BAD_DX_DIR)
1242 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1245 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1246 start = EXT4_I(dir)->i_dir_start_lookup;
1247 if (start >= nblocks)
1253 * We deal with the read-ahead logic here.
1255 if (ra_ptr >= ra_max) {
1256 /* Refill the readahead buffer */
1259 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1261 * Terminate if we reach the end of the
1262 * directory and must wrap, or if our
1263 * search has finished at this block.
1265 if (b >= nblocks || (num && block == start)) {
1266 bh_use[ra_max] = NULL;
1270 bh = ext4_getblk(NULL, dir, b++, 0);
1271 if (unlikely(IS_ERR(bh))) {
1276 bh_use[ra_max] = bh;
1278 ll_rw_block(READ | REQ_META | REQ_PRIO,
1282 if ((bh = bh_use[ra_ptr++]) == NULL)
1285 if (!buffer_uptodate(bh)) {
1286 /* read error, skip block & hope for the best */
1287 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1288 (unsigned long) block);
1292 if (!buffer_verified(bh) &&
1293 !is_dx_internal_node(dir, block,
1294 (struct ext4_dir_entry *)bh->b_data) &&
1295 !ext4_dirent_csum_verify(dir,
1296 (struct ext4_dir_entry *)bh->b_data)) {
1297 EXT4_ERROR_INODE(dir, "checksumming directory "
1298 "block %lu", (unsigned long)block);
1302 set_buffer_verified(bh);
1303 i = search_dirblock(bh, dir, d_name,
1304 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1306 EXT4_I(dir)->i_dir_start_lookup = block;
1308 goto cleanup_and_exit;
1312 goto cleanup_and_exit;
1315 if (++block >= nblocks)
1317 } while (block != start);
1320 * If the directory has grown while we were searching, then
1321 * search the last part of the directory before giving up.
1324 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1325 if (block < nblocks) {
1331 /* Clean up the read-ahead blocks */
1332 for (; ra_ptr < ra_max; ra_ptr++)
1333 brelse(bh_use[ra_ptr]);
1337 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1338 struct ext4_dir_entry_2 **res_dir)
1340 struct super_block * sb = dir->i_sb;
1341 struct dx_hash_info hinfo;
1342 struct dx_frame frames[2], *frame;
1343 struct buffer_head *bh;
1347 frame = dx_probe(d_name, dir, &hinfo, frames);
1349 return (struct buffer_head *) frame;
1351 block = dx_get_block(frame->at);
1352 bh = ext4_read_dirblock(dir, block, DIRENT);
1356 retval = search_dirblock(bh, dir, d_name,
1357 block << EXT4_BLOCK_SIZE_BITS(sb),
1363 bh = ERR_PTR(ERR_BAD_DX_DIR);
1367 /* Check to see if we should continue to search */
1368 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1372 "error %d reading index page in directory #%lu",
1373 retval, dir->i_ino);
1374 bh = ERR_PTR(retval);
1377 } while (retval == 1);
1381 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1387 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1389 struct inode *inode;
1390 struct ext4_dir_entry_2 *de;
1391 struct buffer_head *bh;
1393 if (dentry->d_name.len > EXT4_NAME_LEN)
1394 return ERR_PTR(-ENAMETOOLONG);
1396 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1398 return (struct dentry *) bh;
1401 __u32 ino = le32_to_cpu(de->inode);
1403 if (!ext4_valid_inum(dir->i_sb, ino)) {
1404 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1405 return ERR_PTR(-EIO);
1407 if (unlikely(ino == dir->i_ino)) {
1408 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1410 return ERR_PTR(-EIO);
1412 inode = ext4_iget_normal(dir->i_sb, ino);
1413 if (inode == ERR_PTR(-ESTALE)) {
1414 EXT4_ERROR_INODE(dir,
1415 "deleted inode referenced: %u",
1417 return ERR_PTR(-EIO);
1419 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1420 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1421 S_ISLNK(inode->i_mode)) &&
1422 !ext4_is_child_context_consistent_with_parent(dir,
1425 ext4_warning(inode->i_sb,
1426 "Inconsistent encryption contexts: %lu/%lu\n",
1427 (unsigned long) dir->i_ino,
1428 (unsigned long) inode->i_ino);
1429 return ERR_PTR(-EPERM);
1432 return d_splice_alias(inode, dentry);
1436 struct dentry *ext4_get_parent(struct dentry *child)
1439 static const struct qstr dotdot = QSTR_INIT("..", 2);
1440 struct ext4_dir_entry_2 * de;
1441 struct buffer_head *bh;
1443 bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1445 return (struct dentry *) bh;
1447 return ERR_PTR(-ENOENT);
1448 ino = le32_to_cpu(de->inode);
1451 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1452 EXT4_ERROR_INODE(child->d_inode,
1453 "bad parent inode number: %u", ino);
1454 return ERR_PTR(-EIO);
1457 return d_obtain_alias(ext4_iget_normal(child->d_inode->i_sb, ino));
1461 * Move count entries from end of map between two memory locations.
1462 * Returns pointer to last entry moved.
1464 static struct ext4_dir_entry_2 *
1465 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1468 unsigned rec_len = 0;
1471 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1472 (from + (map->offs<<2));
1473 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1474 memcpy (to, de, rec_len);
1475 ((struct ext4_dir_entry_2 *) to)->rec_len =
1476 ext4_rec_len_to_disk(rec_len, blocksize);
1481 return (struct ext4_dir_entry_2 *) (to - rec_len);
1485 * Compact each dir entry in the range to the minimal rec_len.
1486 * Returns pointer to last entry in range.
1488 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1490 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1491 unsigned rec_len = 0;
1494 while ((char*)de < base + blocksize) {
1495 next = ext4_next_entry(de, blocksize);
1496 if (de->inode && de->name_len) {
1497 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1499 memmove(to, de, rec_len);
1500 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1502 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1510 * Split a full leaf block to make room for a new dir entry.
1511 * Allocate a new block, and move entries so that they are approx. equally full.
1512 * Returns pointer to de in block into which the new entry will be inserted.
1514 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1515 struct buffer_head **bh,struct dx_frame *frame,
1516 struct dx_hash_info *hinfo)
1518 unsigned blocksize = dir->i_sb->s_blocksize;
1519 unsigned count, continued;
1520 struct buffer_head *bh2;
1521 ext4_lblk_t newblock;
1523 struct dx_map_entry *map;
1524 char *data1 = (*bh)->b_data, *data2;
1525 unsigned split, move, size;
1526 struct ext4_dir_entry_2 *de = NULL, *de2;
1527 struct ext4_dir_entry_tail *t;
1531 if (ext4_has_metadata_csum(dir->i_sb))
1532 csum_size = sizeof(struct ext4_dir_entry_tail);
1534 bh2 = ext4_append(handle, dir, &newblock);
1538 return (struct ext4_dir_entry_2 *) bh2;
1541 BUFFER_TRACE(*bh, "get_write_access");
1542 err = ext4_journal_get_write_access(handle, *bh);
1546 BUFFER_TRACE(frame->bh, "get_write_access");
1547 err = ext4_journal_get_write_access(handle, frame->bh);
1551 data2 = bh2->b_data;
1553 /* create map in the end of data2 block */
1554 map = (struct dx_map_entry *) (data2 + blocksize);
1555 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1556 blocksize, hinfo, map);
1558 dx_sort_map(map, count);
1559 /* Split the existing block in the middle, size-wise */
1562 for (i = count-1; i >= 0; i--) {
1563 /* is more than half of this entry in 2nd half of the block? */
1564 if (size + map[i].size/2 > blocksize/2)
1566 size += map[i].size;
1569 /* map index at which we will split */
1570 split = count - move;
1571 hash2 = map[split].hash;
1572 continued = hash2 == map[split - 1].hash;
1573 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1574 (unsigned long)dx_get_block(frame->at),
1575 hash2, split, count-split));
1577 /* Fancy dance to stay within two buffers */
1578 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1579 de = dx_pack_dirents(data1, blocksize);
1580 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1583 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1587 t = EXT4_DIRENT_TAIL(data2, blocksize);
1588 initialize_dirent_tail(t, blocksize);
1590 t = EXT4_DIRENT_TAIL(data1, blocksize);
1591 initialize_dirent_tail(t, blocksize);
1594 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1595 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1597 /* Which block gets the new entry? */
1598 if (hinfo->hash >= hash2) {
1602 dx_insert_block(frame, hash2 + continued, newblock);
1603 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1606 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1610 dxtrace(dx_show_index("frame", frame->entries));
1617 ext4_std_error(dir->i_sb, err);
1618 return ERR_PTR(err);
1621 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1622 struct buffer_head *bh,
1623 void *buf, int buf_size,
1624 const char *name, int namelen,
1625 struct ext4_dir_entry_2 **dest_de)
1627 struct ext4_dir_entry_2 *de;
1628 unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1630 unsigned int offset = 0;
1633 de = (struct ext4_dir_entry_2 *)buf;
1634 top = buf + buf_size - reclen;
1635 while ((char *) de <= top) {
1636 if (ext4_check_dir_entry(dir, NULL, de, bh,
1637 buf, buf_size, offset))
1639 if (ext4_match(namelen, name, de))
1641 nlen = EXT4_DIR_REC_LEN(de->name_len);
1642 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1643 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1645 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1648 if ((char *) de > top)
1655 void ext4_insert_dentry(struct inode *inode,
1656 struct ext4_dir_entry_2 *de,
1658 const char *name, int namelen)
1663 nlen = EXT4_DIR_REC_LEN(de->name_len);
1664 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1666 struct ext4_dir_entry_2 *de1 =
1667 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1668 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1669 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1672 de->file_type = EXT4_FT_UNKNOWN;
1673 de->inode = cpu_to_le32(inode->i_ino);
1674 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1675 de->name_len = namelen;
1676 memcpy(de->name, name, namelen);
1679 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1680 * it points to a directory entry which is guaranteed to be large
1681 * enough for new directory entry. If de is NULL, then
1682 * add_dirent_to_buf will attempt search the directory block for
1683 * space. It will return -ENOSPC if no space is available, and -EIO
1684 * and -EEXIST if directory entry already exists.
1686 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1687 struct inode *inode, struct ext4_dir_entry_2 *de,
1688 struct buffer_head *bh)
1690 struct inode *dir = dentry->d_parent->d_inode;
1691 const char *name = dentry->d_name.name;
1692 int namelen = dentry->d_name.len;
1693 unsigned int blocksize = dir->i_sb->s_blocksize;
1697 if (ext4_has_metadata_csum(inode->i_sb))
1698 csum_size = sizeof(struct ext4_dir_entry_tail);
1701 err = ext4_find_dest_de(dir, inode,
1702 bh, bh->b_data, blocksize - csum_size,
1703 name, namelen, &de);
1707 BUFFER_TRACE(bh, "get_write_access");
1708 err = ext4_journal_get_write_access(handle, bh);
1710 ext4_std_error(dir->i_sb, err);
1714 /* By now the buffer is marked for journaling */
1715 ext4_insert_dentry(inode, de, blocksize, name, namelen);
1718 * XXX shouldn't update any times until successful
1719 * completion of syscall, but too many callers depend
1722 * XXX similarly, too many callers depend on
1723 * ext4_new_inode() setting the times, but error
1724 * recovery deletes the inode, so the worst that can
1725 * happen is that the times are slightly out of date
1726 * and/or different from the directory change time.
1728 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1729 ext4_update_dx_flag(dir);
1731 ext4_mark_inode_dirty(handle, dir);
1732 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1733 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1735 ext4_std_error(dir->i_sb, err);
1740 * This converts a one block unindexed directory to a 3 block indexed
1741 * directory, and adds the dentry to the indexed directory.
1743 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1744 struct inode *inode, struct buffer_head *bh)
1746 struct inode *dir = dentry->d_parent->d_inode;
1747 const char *name = dentry->d_name.name;
1748 int namelen = dentry->d_name.len;
1749 struct buffer_head *bh2;
1750 struct dx_root *root;
1751 struct dx_frame frames[2], *frame;
1752 struct dx_entry *entries;
1753 struct ext4_dir_entry_2 *de, *de2;
1754 struct ext4_dir_entry_tail *t;
1759 struct dx_hash_info hinfo;
1761 struct fake_dirent *fde;
1764 if (ext4_has_metadata_csum(inode->i_sb))
1765 csum_size = sizeof(struct ext4_dir_entry_tail);
1767 blocksize = dir->i_sb->s_blocksize;
1768 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1769 BUFFER_TRACE(bh, "get_write_access");
1770 retval = ext4_journal_get_write_access(handle, bh);
1772 ext4_std_error(dir->i_sb, retval);
1776 root = (struct dx_root *) bh->b_data;
1778 /* The 0th block becomes the root, move the dirents out */
1779 fde = &root->dotdot;
1780 de = (struct ext4_dir_entry_2 *)((char *)fde +
1781 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1782 if ((char *) de >= (((char *) root) + blocksize)) {
1783 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1787 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1789 /* Allocate new block for the 0th block's dirents */
1790 bh2 = ext4_append(handle, dir, &block);
1793 return PTR_ERR(bh2);
1795 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1796 data1 = bh2->b_data;
1798 memcpy (data1, de, len);
1799 de = (struct ext4_dir_entry_2 *) data1;
1801 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1803 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1808 t = EXT4_DIRENT_TAIL(data1, blocksize);
1809 initialize_dirent_tail(t, blocksize);
1812 /* Initialize the root; the dot dirents already exist */
1813 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1814 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1816 memset (&root->info, 0, sizeof(root->info));
1817 root->info.info_length = sizeof(root->info);
1818 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1819 entries = root->entries;
1820 dx_set_block(entries, 1);
1821 dx_set_count(entries, 1);
1822 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1824 /* Initialize as for dx_probe */
1825 hinfo.hash_version = root->info.hash_version;
1826 if (hinfo.hash_version <= DX_HASH_TEA)
1827 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1828 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1829 ext4fs_dirhash(name, namelen, &hinfo);
1830 memset(frames, 0, sizeof(frames));
1832 frame->entries = entries;
1833 frame->at = entries;
1837 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1840 retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
1844 de = do_split(handle,dir, &bh, frame, &hinfo);
1846 retval = PTR_ERR(de);
1851 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1856 * Even if the block split failed, we have to properly write
1857 * out all the changes we did so far. Otherwise we can end up
1858 * with corrupted filesystem.
1860 ext4_mark_inode_dirty(handle, dir);
1868 * adds a file entry to the specified directory, using the same
1869 * semantics as ext4_find_entry(). It returns NULL if it failed.
1871 * NOTE!! The inode part of 'de' is left at 0 - which means you
1872 * may not sleep between calling this and putting something into
1873 * the entry, as someone else might have used it while you slept.
1875 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1876 struct inode *inode)
1878 struct inode *dir = dentry->d_parent->d_inode;
1879 struct buffer_head *bh = NULL;
1880 struct ext4_dir_entry_2 *de;
1881 struct ext4_dir_entry_tail *t;
1882 struct super_block *sb;
1886 ext4_lblk_t block, blocks;
1889 if (ext4_has_metadata_csum(inode->i_sb))
1890 csum_size = sizeof(struct ext4_dir_entry_tail);
1893 blocksize = sb->s_blocksize;
1894 if (!dentry->d_name.len)
1897 if (ext4_has_inline_data(dir)) {
1898 retval = ext4_try_add_inline_entry(handle, dentry, inode);
1908 retval = ext4_dx_add_entry(handle, dentry, inode);
1909 if (!retval || (retval != ERR_BAD_DX_DIR))
1911 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1913 ext4_mark_inode_dirty(handle, dir);
1915 blocks = dir->i_size >> sb->s_blocksize_bits;
1916 for (block = 0; block < blocks; block++) {
1917 bh = ext4_read_dirblock(dir, block, DIRENT);
1921 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1922 if (retval != -ENOSPC)
1925 if (blocks == 1 && !dx_fallback &&
1926 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
1927 retval = make_indexed_dir(handle, dentry, inode, bh);
1928 bh = NULL; /* make_indexed_dir releases bh */
1933 bh = ext4_append(handle, dir, &block);
1936 de = (struct ext4_dir_entry_2 *) bh->b_data;
1938 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1941 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1942 initialize_dirent_tail(t, blocksize);
1945 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1949 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1954 * Returns 0 for success, or a negative error value
1956 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1957 struct inode *inode)
1959 struct dx_frame frames[2], *frame;
1960 struct dx_entry *entries, *at;
1961 struct dx_hash_info hinfo;
1962 struct buffer_head *bh;
1963 struct inode *dir = dentry->d_parent->d_inode;
1964 struct super_block *sb = dir->i_sb;
1965 struct ext4_dir_entry_2 *de;
1968 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames);
1970 return PTR_ERR(frame);
1971 entries = frame->entries;
1973 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
1980 BUFFER_TRACE(bh, "get_write_access");
1981 err = ext4_journal_get_write_access(handle, bh);
1985 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1989 /* Block full, should compress but for now just split */
1990 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1991 dx_get_count(entries), dx_get_limit(entries)));
1992 /* Need to split index? */
1993 if (dx_get_count(entries) == dx_get_limit(entries)) {
1994 ext4_lblk_t newblock;
1995 unsigned icount = dx_get_count(entries);
1996 int levels = frame - frames;
1997 struct dx_entry *entries2;
1998 struct dx_node *node2;
1999 struct buffer_head *bh2;
2001 if (levels && (dx_get_count(frames->entries) ==
2002 dx_get_limit(frames->entries))) {
2003 ext4_warning(sb, "Directory index full!");
2007 bh2 = ext4_append(handle, dir, &newblock);
2012 node2 = (struct dx_node *)(bh2->b_data);
2013 entries2 = node2->entries;
2014 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2015 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2017 BUFFER_TRACE(frame->bh, "get_write_access");
2018 err = ext4_journal_get_write_access(handle, frame->bh);
2022 unsigned icount1 = icount/2, icount2 = icount - icount1;
2023 unsigned hash2 = dx_get_hash(entries + icount1);
2024 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2027 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2028 err = ext4_journal_get_write_access(handle,
2033 memcpy((char *) entries2, (char *) (entries + icount1),
2034 icount2 * sizeof(struct dx_entry));
2035 dx_set_count(entries, icount1);
2036 dx_set_count(entries2, icount2);
2037 dx_set_limit(entries2, dx_node_limit(dir));
2039 /* Which index block gets the new entry? */
2040 if (at - entries >= icount1) {
2041 frame->at = at = at - entries - icount1 + entries2;
2042 frame->entries = entries = entries2;
2043 swap(frame->bh, bh2);
2045 dx_insert_block(frames + 0, hash2, newblock);
2046 dxtrace(dx_show_index("node", frames[1].entries));
2047 dxtrace(dx_show_index("node",
2048 ((struct dx_node *) bh2->b_data)->entries));
2049 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2054 dxtrace(printk(KERN_DEBUG
2055 "Creating second level index...\n"));
2056 memcpy((char *) entries2, (char *) entries,
2057 icount * sizeof(struct dx_entry));
2058 dx_set_limit(entries2, dx_node_limit(dir));
2061 dx_set_count(entries, 1);
2062 dx_set_block(entries + 0, newblock);
2063 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2065 /* Add new access path frame */
2067 frame->at = at = at - entries + entries2;
2068 frame->entries = entries = entries2;
2070 err = ext4_journal_get_write_access(handle,
2075 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2077 ext4_std_error(inode->i_sb, err);
2081 de = do_split(handle, dir, &bh, frame, &hinfo);
2086 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2090 ext4_std_error(dir->i_sb, err);
2098 * ext4_generic_delete_entry deletes a directory entry by merging it
2099 * with the previous entry
2101 int ext4_generic_delete_entry(handle_t *handle,
2103 struct ext4_dir_entry_2 *de_del,
2104 struct buffer_head *bh,
2109 struct ext4_dir_entry_2 *de, *pde;
2110 unsigned int blocksize = dir->i_sb->s_blocksize;
2115 de = (struct ext4_dir_entry_2 *)entry_buf;
2116 while (i < buf_size - csum_size) {
2117 if (ext4_check_dir_entry(dir, NULL, de, bh,
2118 bh->b_data, bh->b_size, i))
2122 pde->rec_len = ext4_rec_len_to_disk(
2123 ext4_rec_len_from_disk(pde->rec_len,
2125 ext4_rec_len_from_disk(de->rec_len,
2133 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2135 de = ext4_next_entry(de, blocksize);
2140 static int ext4_delete_entry(handle_t *handle,
2142 struct ext4_dir_entry_2 *de_del,
2143 struct buffer_head *bh)
2145 int err, csum_size = 0;
2147 if (ext4_has_inline_data(dir)) {
2148 int has_inline_data = 1;
2149 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2151 if (has_inline_data)
2155 if (ext4_has_metadata_csum(dir->i_sb))
2156 csum_size = sizeof(struct ext4_dir_entry_tail);
2158 BUFFER_TRACE(bh, "get_write_access");
2159 err = ext4_journal_get_write_access(handle, bh);
2163 err = ext4_generic_delete_entry(handle, dir, de_del,
2165 dir->i_sb->s_blocksize, csum_size);
2169 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2170 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2177 ext4_std_error(dir->i_sb, err);
2182 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2183 * since this indicates that nlinks count was previously 1.
2185 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2188 if (is_dx(inode) && inode->i_nlink > 1) {
2189 /* limit is 16-bit i_links_count */
2190 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2191 set_nlink(inode, 1);
2192 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2193 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2199 * If a directory had nlink == 1, then we should let it be 1. This indicates
2200 * directory has >EXT4_LINK_MAX subdirs.
2202 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2204 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2209 static int ext4_add_nondir(handle_t *handle,
2210 struct dentry *dentry, struct inode *inode)
2212 int err = ext4_add_entry(handle, dentry, inode);
2214 ext4_mark_inode_dirty(handle, inode);
2215 unlock_new_inode(inode);
2216 d_instantiate(dentry, inode);
2220 unlock_new_inode(inode);
2226 * By the time this is called, we already have created
2227 * the directory cache entry for the new file, but it
2228 * is so far negative - it has no inode.
2230 * If the create succeeds, we fill in the inode information
2231 * with d_instantiate().
2233 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2237 struct inode *inode;
2238 int err, credits, retries = 0;
2240 dquot_initialize(dir);
2242 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2243 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2245 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2246 NULL, EXT4_HT_DIR, credits);
2247 handle = ext4_journal_current_handle();
2248 err = PTR_ERR(inode);
2249 if (!IS_ERR(inode)) {
2250 inode->i_op = &ext4_file_inode_operations;
2251 if (test_opt(inode->i_sb, DAX))
2252 inode->i_fop = &ext4_dax_file_operations;
2254 inode->i_fop = &ext4_file_operations;
2255 ext4_set_aops(inode);
2256 err = ext4_add_nondir(handle, dentry, inode);
2257 if (!err && IS_DIRSYNC(dir))
2258 ext4_handle_sync(handle);
2261 ext4_journal_stop(handle);
2262 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2267 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2268 umode_t mode, dev_t rdev)
2271 struct inode *inode;
2272 int err, credits, retries = 0;
2274 if (!new_valid_dev(rdev))
2277 dquot_initialize(dir);
2279 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2280 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2282 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2283 NULL, EXT4_HT_DIR, credits);
2284 handle = ext4_journal_current_handle();
2285 err = PTR_ERR(inode);
2286 if (!IS_ERR(inode)) {
2287 init_special_inode(inode, inode->i_mode, rdev);
2288 inode->i_op = &ext4_special_inode_operations;
2289 err = ext4_add_nondir(handle, dentry, inode);
2290 if (!err && IS_DIRSYNC(dir))
2291 ext4_handle_sync(handle);
2294 ext4_journal_stop(handle);
2295 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2300 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2303 struct inode *inode;
2304 int err, retries = 0;
2306 dquot_initialize(dir);
2309 inode = ext4_new_inode_start_handle(dir, mode,
2312 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2313 4 + EXT4_XATTR_TRANS_BLOCKS);
2314 handle = ext4_journal_current_handle();
2315 err = PTR_ERR(inode);
2316 if (!IS_ERR(inode)) {
2317 inode->i_op = &ext4_file_inode_operations;
2318 if (test_opt(inode->i_sb, DAX))
2319 inode->i_fop = &ext4_dax_file_operations;
2321 inode->i_fop = &ext4_file_operations;
2322 ext4_set_aops(inode);
2323 d_tmpfile(dentry, inode);
2324 err = ext4_orphan_add(handle, inode);
2326 goto err_unlock_inode;
2327 mark_inode_dirty(inode);
2328 unlock_new_inode(inode);
2331 ext4_journal_stop(handle);
2332 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2336 ext4_journal_stop(handle);
2337 unlock_new_inode(inode);
2341 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2342 struct ext4_dir_entry_2 *de,
2343 int blocksize, int csum_size,
2344 unsigned int parent_ino, int dotdot_real_len)
2346 de->inode = cpu_to_le32(inode->i_ino);
2348 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2350 strcpy(de->name, ".");
2351 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2353 de = ext4_next_entry(de, blocksize);
2354 de->inode = cpu_to_le32(parent_ino);
2356 if (!dotdot_real_len)
2357 de->rec_len = ext4_rec_len_to_disk(blocksize -
2358 (csum_size + EXT4_DIR_REC_LEN(1)),
2361 de->rec_len = ext4_rec_len_to_disk(
2362 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2363 strcpy(de->name, "..");
2364 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2366 return ext4_next_entry(de, blocksize);
2369 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2370 struct inode *inode)
2372 struct buffer_head *dir_block = NULL;
2373 struct ext4_dir_entry_2 *de;
2374 struct ext4_dir_entry_tail *t;
2375 ext4_lblk_t block = 0;
2376 unsigned int blocksize = dir->i_sb->s_blocksize;
2380 if (ext4_has_metadata_csum(dir->i_sb))
2381 csum_size = sizeof(struct ext4_dir_entry_tail);
2383 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2384 err = ext4_try_create_inline_dir(handle, dir, inode);
2385 if (err < 0 && err != -ENOSPC)
2392 dir_block = ext4_append(handle, inode, &block);
2393 if (IS_ERR(dir_block))
2394 return PTR_ERR(dir_block);
2395 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2396 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2397 set_nlink(inode, 2);
2399 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2400 initialize_dirent_tail(t, blocksize);
2403 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2404 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2407 set_buffer_verified(dir_block);
2413 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2416 struct inode *inode;
2417 int err, credits, retries = 0;
2419 if (EXT4_DIR_LINK_MAX(dir))
2422 dquot_initialize(dir);
2424 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2425 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2427 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2429 0, NULL, EXT4_HT_DIR, credits);
2430 handle = ext4_journal_current_handle();
2431 err = PTR_ERR(inode);
2435 inode->i_op = &ext4_dir_inode_operations;
2436 inode->i_fop = &ext4_dir_operations;
2437 err = ext4_init_new_dir(handle, dir, inode);
2439 goto out_clear_inode;
2440 err = ext4_mark_inode_dirty(handle, inode);
2442 err = ext4_add_entry(handle, dentry, inode);
2446 unlock_new_inode(inode);
2447 ext4_mark_inode_dirty(handle, inode);
2451 ext4_inc_count(handle, dir);
2452 ext4_update_dx_flag(dir);
2453 err = ext4_mark_inode_dirty(handle, dir);
2455 goto out_clear_inode;
2456 unlock_new_inode(inode);
2457 d_instantiate(dentry, inode);
2458 if (IS_DIRSYNC(dir))
2459 ext4_handle_sync(handle);
2463 ext4_journal_stop(handle);
2464 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2470 * routine to check that the specified directory is empty (for rmdir)
2472 int ext4_empty_dir(struct inode *inode)
2474 unsigned int offset;
2475 struct buffer_head *bh;
2476 struct ext4_dir_entry_2 *de, *de1;
2477 struct super_block *sb;
2480 if (ext4_has_inline_data(inode)) {
2481 int has_inline_data = 1;
2483 err = empty_inline_dir(inode, &has_inline_data);
2484 if (has_inline_data)
2489 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2490 EXT4_ERROR_INODE(inode, "invalid size");
2493 bh = ext4_read_dirblock(inode, 0, EITHER);
2497 de = (struct ext4_dir_entry_2 *) bh->b_data;
2498 de1 = ext4_next_entry(de, sb->s_blocksize);
2499 if (le32_to_cpu(de->inode) != inode->i_ino ||
2500 !le32_to_cpu(de1->inode) ||
2501 strcmp(".", de->name) ||
2502 strcmp("..", de1->name)) {
2503 ext4_warning(inode->i_sb,
2504 "bad directory (dir #%lu) - no `.' or `..'",
2509 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2510 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2511 de = ext4_next_entry(de1, sb->s_blocksize);
2512 while (offset < inode->i_size) {
2513 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2514 unsigned int lblock;
2517 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2518 bh = ext4_read_dirblock(inode, lblock, EITHER);
2521 de = (struct ext4_dir_entry_2 *) bh->b_data;
2523 if (ext4_check_dir_entry(inode, NULL, de, bh,
2524 bh->b_data, bh->b_size, offset)) {
2525 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2527 offset = (offset | (sb->s_blocksize - 1)) + 1;
2530 if (le32_to_cpu(de->inode)) {
2534 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2535 de = ext4_next_entry(de, sb->s_blocksize);
2542 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2543 * such inodes, starting at the superblock, in case we crash before the
2544 * file is closed/deleted, or in case the inode truncate spans multiple
2545 * transactions and the last transaction is not recovered after a crash.
2547 * At filesystem recovery time, we walk this list deleting unlinked
2548 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2550 * Orphan list manipulation functions must be called under i_mutex unless
2551 * we are just creating the inode or deleting it.
2553 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2555 struct super_block *sb = inode->i_sb;
2556 struct ext4_sb_info *sbi = EXT4_SB(sb);
2557 struct ext4_iloc iloc;
2561 if (!sbi->s_journal || is_bad_inode(inode))
2564 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2565 !mutex_is_locked(&inode->i_mutex));
2567 * Exit early if inode already is on orphan list. This is a big speedup
2568 * since we don't have to contend on the global s_orphan_lock.
2570 if (!list_empty(&EXT4_I(inode)->i_orphan))
2574 * Orphan handling is only valid for files with data blocks
2575 * being truncated, or files being unlinked. Note that we either
2576 * hold i_mutex, or the inode can not be referenced from outside,
2577 * so i_nlink should not be bumped due to race
2579 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2580 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2582 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2583 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2587 err = ext4_reserve_inode_write(handle, inode, &iloc);
2591 mutex_lock(&sbi->s_orphan_lock);
2593 * Due to previous errors inode may be already a part of on-disk
2594 * orphan list. If so skip on-disk list modification.
2596 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2597 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2598 /* Insert this inode at the head of the on-disk orphan list */
2599 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2600 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2603 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2604 mutex_unlock(&sbi->s_orphan_lock);
2607 err = ext4_handle_dirty_super(handle, sb);
2608 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2613 * We have to remove inode from in-memory list if
2614 * addition to on disk orphan list failed. Stray orphan
2615 * list entries can cause panics at unmount time.
2617 mutex_lock(&sbi->s_orphan_lock);
2618 list_del(&EXT4_I(inode)->i_orphan);
2619 mutex_unlock(&sbi->s_orphan_lock);
2622 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2623 jbd_debug(4, "orphan inode %lu will point to %d\n",
2624 inode->i_ino, NEXT_ORPHAN(inode));
2626 ext4_std_error(sb, err);
2631 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2632 * of such inodes stored on disk, because it is finally being cleaned up.
2634 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2636 struct list_head *prev;
2637 struct ext4_inode_info *ei = EXT4_I(inode);
2638 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2640 struct ext4_iloc iloc;
2643 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2646 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2647 !mutex_is_locked(&inode->i_mutex));
2648 /* Do this quick check before taking global s_orphan_lock. */
2649 if (list_empty(&ei->i_orphan))
2653 /* Grab inode buffer early before taking global s_orphan_lock */
2654 err = ext4_reserve_inode_write(handle, inode, &iloc);
2657 mutex_lock(&sbi->s_orphan_lock);
2658 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2660 prev = ei->i_orphan.prev;
2661 list_del_init(&ei->i_orphan);
2663 /* If we're on an error path, we may not have a valid
2664 * transaction handle with which to update the orphan list on
2665 * disk, but we still need to remove the inode from the linked
2666 * list in memory. */
2667 if (!handle || err) {
2668 mutex_unlock(&sbi->s_orphan_lock);
2672 ino_next = NEXT_ORPHAN(inode);
2673 if (prev == &sbi->s_orphan) {
2674 jbd_debug(4, "superblock will point to %u\n", ino_next);
2675 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2676 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2678 mutex_unlock(&sbi->s_orphan_lock);
2681 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2682 mutex_unlock(&sbi->s_orphan_lock);
2683 err = ext4_handle_dirty_super(handle, inode->i_sb);
2685 struct ext4_iloc iloc2;
2686 struct inode *i_prev =
2687 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2689 jbd_debug(4, "orphan inode %lu will point to %u\n",
2690 i_prev->i_ino, ino_next);
2691 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2693 mutex_unlock(&sbi->s_orphan_lock);
2696 NEXT_ORPHAN(i_prev) = ino_next;
2697 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2698 mutex_unlock(&sbi->s_orphan_lock);
2702 NEXT_ORPHAN(inode) = 0;
2703 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2705 ext4_std_error(inode->i_sb, err);
2713 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2716 struct inode *inode;
2717 struct buffer_head *bh;
2718 struct ext4_dir_entry_2 *de;
2719 handle_t *handle = NULL;
2721 /* Initialize quotas before so that eventual writes go in
2722 * separate transaction */
2723 dquot_initialize(dir);
2724 dquot_initialize(dentry->d_inode);
2727 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2733 inode = dentry->d_inode;
2736 if (le32_to_cpu(de->inode) != inode->i_ino)
2739 retval = -ENOTEMPTY;
2740 if (!ext4_empty_dir(inode))
2743 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2744 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2745 if (IS_ERR(handle)) {
2746 retval = PTR_ERR(handle);
2751 if (IS_DIRSYNC(dir))
2752 ext4_handle_sync(handle);
2754 retval = ext4_delete_entry(handle, dir, de, bh);
2757 if (!EXT4_DIR_LINK_EMPTY(inode))
2758 ext4_warning(inode->i_sb,
2759 "empty directory has too many links (%d)",
2763 /* There's no need to set i_disksize: the fact that i_nlink is
2764 * zero will ensure that the right thing happens during any
2767 ext4_orphan_add(handle, inode);
2768 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2769 ext4_mark_inode_dirty(handle, inode);
2770 ext4_dec_count(handle, dir);
2771 ext4_update_dx_flag(dir);
2772 ext4_mark_inode_dirty(handle, dir);
2777 ext4_journal_stop(handle);
2781 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2784 struct inode *inode;
2785 struct buffer_head *bh;
2786 struct ext4_dir_entry_2 *de;
2787 handle_t *handle = NULL;
2789 trace_ext4_unlink_enter(dir, dentry);
2790 /* Initialize quotas before so that eventual writes go
2791 * in separate transaction */
2792 dquot_initialize(dir);
2793 dquot_initialize(dentry->d_inode);
2796 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2802 inode = dentry->d_inode;
2805 if (le32_to_cpu(de->inode) != inode->i_ino)
2808 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2809 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2810 if (IS_ERR(handle)) {
2811 retval = PTR_ERR(handle);
2816 if (IS_DIRSYNC(dir))
2817 ext4_handle_sync(handle);
2819 if (!inode->i_nlink) {
2820 ext4_warning(inode->i_sb,
2821 "Deleting nonexistent file (%lu), %d",
2822 inode->i_ino, inode->i_nlink);
2823 set_nlink(inode, 1);
2825 retval = ext4_delete_entry(handle, dir, de, bh);
2828 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2829 ext4_update_dx_flag(dir);
2830 ext4_mark_inode_dirty(handle, dir);
2832 if (!inode->i_nlink)
2833 ext4_orphan_add(handle, inode);
2834 inode->i_ctime = ext4_current_time(inode);
2835 ext4_mark_inode_dirty(handle, inode);
2840 ext4_journal_stop(handle);
2841 trace_ext4_unlink_exit(dentry, retval);
2845 static int ext4_symlink(struct inode *dir,
2846 struct dentry *dentry, const char *symname)
2849 struct inode *inode;
2850 int l, err, retries = 0;
2853 l = strlen(symname)+1;
2854 if (l > dir->i_sb->s_blocksize)
2855 return -ENAMETOOLONG;
2857 dquot_initialize(dir);
2859 if (l > EXT4_N_BLOCKS * 4) {
2861 * For non-fast symlinks, we just allocate inode and put it on
2862 * orphan list in the first transaction => we need bitmap,
2863 * group descriptor, sb, inode block, quota blocks, and
2864 * possibly selinux xattr blocks.
2866 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2867 EXT4_XATTR_TRANS_BLOCKS;
2870 * Fast symlink. We have to add entry to directory
2871 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2872 * allocate new inode (bitmap, group descriptor, inode block,
2873 * quota blocks, sb is already counted in previous macros).
2875 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2876 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
2879 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
2880 &dentry->d_name, 0, NULL,
2881 EXT4_HT_DIR, credits);
2882 handle = ext4_journal_current_handle();
2883 err = PTR_ERR(inode);
2887 if (l > EXT4_N_BLOCKS * 4) {
2888 inode->i_op = &ext4_symlink_inode_operations;
2889 ext4_set_aops(inode);
2891 * We cannot call page_symlink() with transaction started
2892 * because it calls into ext4_write_begin() which can wait
2893 * for transaction commit if we are running out of space
2894 * and thus we deadlock. So we have to stop transaction now
2895 * and restart it when symlink contents is written.
2897 * To keep fs consistent in case of crash, we have to put inode
2898 * to orphan list in the mean time.
2901 err = ext4_orphan_add(handle, inode);
2902 ext4_journal_stop(handle);
2904 goto err_drop_inode;
2905 err = __page_symlink(inode, symname, l, 1);
2907 goto err_drop_inode;
2909 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2910 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2912 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2913 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2914 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2915 if (IS_ERR(handle)) {
2916 err = PTR_ERR(handle);
2917 goto err_drop_inode;
2919 set_nlink(inode, 1);
2920 err = ext4_orphan_del(handle, inode);
2922 ext4_journal_stop(handle);
2924 goto err_drop_inode;
2927 /* clear the extent format for fast symlink */
2928 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2929 inode->i_op = &ext4_fast_symlink_inode_operations;
2930 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2931 inode->i_size = l-1;
2933 EXT4_I(inode)->i_disksize = inode->i_size;
2934 err = ext4_add_nondir(handle, dentry, inode);
2935 if (!err && IS_DIRSYNC(dir))
2936 ext4_handle_sync(handle);
2940 ext4_journal_stop(handle);
2941 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2945 unlock_new_inode(inode);
2950 static int ext4_link(struct dentry *old_dentry,
2951 struct inode *dir, struct dentry *dentry)
2954 struct inode *inode = old_dentry->d_inode;
2955 int err, retries = 0;
2957 if (inode->i_nlink >= EXT4_LINK_MAX)
2959 if (ext4_encrypted_inode(dir) &&
2960 !ext4_is_child_context_consistent_with_parent(dir, inode))
2962 dquot_initialize(dir);
2965 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2966 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2967 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
2969 return PTR_ERR(handle);
2971 if (IS_DIRSYNC(dir))
2972 ext4_handle_sync(handle);
2974 inode->i_ctime = ext4_current_time(inode);
2975 ext4_inc_count(handle, inode);
2978 err = ext4_add_entry(handle, dentry, inode);
2980 ext4_mark_inode_dirty(handle, inode);
2981 /* this can happen only for tmpfile being
2982 * linked the first time
2984 if (inode->i_nlink == 1)
2985 ext4_orphan_del(handle, inode);
2986 d_instantiate(dentry, inode);
2991 ext4_journal_stop(handle);
2992 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2999 * Try to find buffer head where contains the parent block.
3000 * It should be the inode block if it is inlined or the 1st block
3001 * if it is a normal dir.
3003 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3004 struct inode *inode,
3006 struct ext4_dir_entry_2 **parent_de,
3009 struct buffer_head *bh;
3011 if (!ext4_has_inline_data(inode)) {
3012 bh = ext4_read_dirblock(inode, 0, EITHER);
3014 *retval = PTR_ERR(bh);
3017 *parent_de = ext4_next_entry(
3018 (struct ext4_dir_entry_2 *)bh->b_data,
3019 inode->i_sb->s_blocksize);
3024 return ext4_get_first_inline_block(inode, parent_de, retval);
3027 struct ext4_renament {
3029 struct dentry *dentry;
3030 struct inode *inode;
3032 int dir_nlink_delta;
3034 /* entry for "dentry" */
3035 struct buffer_head *bh;
3036 struct ext4_dir_entry_2 *de;
3039 /* entry for ".." in inode if it's a directory */
3040 struct buffer_head *dir_bh;
3041 struct ext4_dir_entry_2 *parent_de;
3045 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3049 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3050 &retval, &ent->parent_de,
3054 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3056 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3057 return ext4_journal_get_write_access(handle, ent->dir_bh);
3060 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3065 ent->parent_de->inode = cpu_to_le32(dir_ino);
3066 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3067 if (!ent->dir_inlined) {
3068 if (is_dx(ent->inode)) {
3069 retval = ext4_handle_dirty_dx_node(handle,
3073 retval = ext4_handle_dirty_dirent_node(handle,
3078 retval = ext4_mark_inode_dirty(handle, ent->inode);
3081 ext4_std_error(ent->dir->i_sb, retval);
3087 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3088 unsigned ino, unsigned file_type)
3092 BUFFER_TRACE(ent->bh, "get write access");
3093 retval = ext4_journal_get_write_access(handle, ent->bh);
3096 ent->de->inode = cpu_to_le32(ino);
3097 if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb,
3098 EXT4_FEATURE_INCOMPAT_FILETYPE))
3099 ent->de->file_type = file_type;
3100 ent->dir->i_version++;
3101 ent->dir->i_ctime = ent->dir->i_mtime =
3102 ext4_current_time(ent->dir);
3103 ext4_mark_inode_dirty(handle, ent->dir);
3104 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3105 if (!ent->inlined) {
3106 retval = ext4_handle_dirty_dirent_node(handle,
3108 if (unlikely(retval)) {
3109 ext4_std_error(ent->dir->i_sb, retval);
3119 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3120 const struct qstr *d_name)
3122 int retval = -ENOENT;
3123 struct buffer_head *bh;
3124 struct ext4_dir_entry_2 *de;
3126 bh = ext4_find_entry(dir, d_name, &de, NULL);
3130 retval = ext4_delete_entry(handle, dir, de, bh);
3136 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3141 * ent->de could have moved from under us during htree split, so make
3142 * sure that we are deleting the right entry. We might also be pointing
3143 * to a stale entry in the unused part of ent->bh so just checking inum
3144 * and the name isn't enough.
3146 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3147 ent->de->name_len != ent->dentry->d_name.len ||
3148 strncmp(ent->de->name, ent->dentry->d_name.name,
3149 ent->de->name_len) ||
3151 retval = ext4_find_delete_entry(handle, ent->dir,
3152 &ent->dentry->d_name);
3154 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3155 if (retval == -ENOENT) {
3156 retval = ext4_find_delete_entry(handle, ent->dir,
3157 &ent->dentry->d_name);
3162 ext4_warning(ent->dir->i_sb,
3163 "Deleting old file (%lu), %d, error=%d",
3164 ent->dir->i_ino, ent->dir->i_nlink, retval);
3168 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3170 if (ent->dir_nlink_delta) {
3171 if (ent->dir_nlink_delta == -1)
3172 ext4_dec_count(handle, ent->dir);
3174 ext4_inc_count(handle, ent->dir);
3175 ext4_mark_inode_dirty(handle, ent->dir);
3179 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3180 int credits, handle_t **h)
3187 * for inode block, sb block, group summaries,
3190 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3191 EXT4_XATTR_TRANS_BLOCKS + 4);
3193 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3194 &ent->dentry->d_name, 0, NULL,
3195 EXT4_HT_DIR, credits);
3197 handle = ext4_journal_current_handle();
3200 ext4_journal_stop(handle);
3201 if (PTR_ERR(wh) == -ENOSPC &&
3202 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3206 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3207 wh->i_op = &ext4_special_inode_operations;
3213 * Anybody can rename anything with this: the permission checks are left to the
3214 * higher-level routines.
3216 * n.b. old_{dentry,inode) refers to the source dentry/inode
3217 * while new_{dentry,inode) refers to the destination dentry/inode
3218 * This comes from rename(const char *oldpath, const char *newpath)
3220 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3221 struct inode *new_dir, struct dentry *new_dentry,
3224 handle_t *handle = NULL;
3225 struct ext4_renament old = {
3227 .dentry = old_dentry,
3228 .inode = old_dentry->d_inode,
3230 struct ext4_renament new = {
3232 .dentry = new_dentry,
3233 .inode = new_dentry->d_inode,
3237 struct inode *whiteout = NULL;
3241 dquot_initialize(old.dir);
3242 dquot_initialize(new.dir);
3244 /* Initialize quotas before so that eventual writes go
3245 * in separate transaction */
3247 dquot_initialize(new.inode);
3249 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3251 return PTR_ERR(old.bh);
3253 * Check for inode number is _not_ due to possible IO errors.
3254 * We might rmdir the source, keep it as pwd of some process
3255 * and merrily kill the link to whatever was created under the
3256 * same name. Goodbye sticky bit ;-<
3259 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3262 if ((old.dir != new.dir) &&
3263 ext4_encrypted_inode(new.dir) &&
3264 !ext4_is_child_context_consistent_with_parent(new.dir,
3270 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3271 &new.de, &new.inlined);
3272 if (IS_ERR(new.bh)) {
3273 retval = PTR_ERR(new.bh);
3283 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3284 ext4_alloc_da_blocks(old.inode);
3286 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3287 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3288 if (!(flags & RENAME_WHITEOUT)) {
3289 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3290 if (IS_ERR(handle)) {
3291 retval = PTR_ERR(handle);
3296 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3297 if (IS_ERR(whiteout)) {
3298 retval = PTR_ERR(whiteout);
3304 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3305 ext4_handle_sync(handle);
3307 if (S_ISDIR(old.inode->i_mode)) {
3309 retval = -ENOTEMPTY;
3310 if (!ext4_empty_dir(new.inode))
3314 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3317 retval = ext4_rename_dir_prepare(handle, &old);
3322 * If we're renaming a file within an inline_data dir and adding or
3323 * setting the new dirent causes a conversion from inline_data to
3324 * extents/blockmap, we need to force the dirent delete code to
3325 * re-read the directory, or else we end up trying to delete a dirent
3326 * from what is now the extent tree root (or a block map).
3328 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3329 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3331 old_file_type = old.de->file_type;
3334 * Do this before adding a new entry, so the old entry is sure
3335 * to be still pointing to the valid old entry.
3337 retval = ext4_setent(handle, &old, whiteout->i_ino,
3341 ext4_mark_inode_dirty(handle, whiteout);
3344 retval = ext4_add_entry(handle, new.dentry, old.inode);
3348 retval = ext4_setent(handle, &new,
3349 old.inode->i_ino, old_file_type);
3354 force_reread = !ext4_test_inode_flag(new.dir,
3355 EXT4_INODE_INLINE_DATA);
3358 * Like most other Unix systems, set the ctime for inodes on a
3361 old.inode->i_ctime = ext4_current_time(old.inode);
3362 ext4_mark_inode_dirty(handle, old.inode);
3368 ext4_rename_delete(handle, &old, force_reread);
3372 ext4_dec_count(handle, new.inode);
3373 new.inode->i_ctime = ext4_current_time(new.inode);
3375 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3376 ext4_update_dx_flag(old.dir);
3378 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3382 ext4_dec_count(handle, old.dir);
3384 /* checked ext4_empty_dir above, can't have another
3385 * parent, ext4_dec_count() won't work for many-linked
3387 clear_nlink(new.inode);
3389 ext4_inc_count(handle, new.dir);
3390 ext4_update_dx_flag(new.dir);
3391 ext4_mark_inode_dirty(handle, new.dir);
3394 ext4_mark_inode_dirty(handle, old.dir);
3396 ext4_mark_inode_dirty(handle, new.inode);
3397 if (!new.inode->i_nlink)
3398 ext4_orphan_add(handle, new.inode);
3408 drop_nlink(whiteout);
3409 unlock_new_inode(whiteout);
3413 ext4_journal_stop(handle);
3417 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3418 struct inode *new_dir, struct dentry *new_dentry)
3420 handle_t *handle = NULL;
3421 struct ext4_renament old = {
3423 .dentry = old_dentry,
3424 .inode = old_dentry->d_inode,
3426 struct ext4_renament new = {
3428 .dentry = new_dentry,
3429 .inode = new_dentry->d_inode,
3434 dquot_initialize(old.dir);
3435 dquot_initialize(new.dir);
3437 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3438 &old.de, &old.inlined);
3440 return PTR_ERR(old.bh);
3442 * Check for inode number is _not_ due to possible IO errors.
3443 * We might rmdir the source, keep it as pwd of some process
3444 * and merrily kill the link to whatever was created under the
3445 * same name. Goodbye sticky bit ;-<
3448 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3451 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3452 &new.de, &new.inlined);
3453 if (IS_ERR(new.bh)) {
3454 retval = PTR_ERR(new.bh);
3459 /* RENAME_EXCHANGE case: old *and* new must both exist */
3460 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3463 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3464 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3465 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3466 if (IS_ERR(handle)) {
3467 retval = PTR_ERR(handle);
3472 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3473 ext4_handle_sync(handle);
3475 if (S_ISDIR(old.inode->i_mode)) {
3477 retval = ext4_rename_dir_prepare(handle, &old);
3481 if (S_ISDIR(new.inode->i_mode)) {
3483 retval = ext4_rename_dir_prepare(handle, &new);
3489 * Other than the special case of overwriting a directory, parents'
3490 * nlink only needs to be modified if this is a cross directory rename.
3492 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3493 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3494 new.dir_nlink_delta = -old.dir_nlink_delta;
3496 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3497 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3501 new_file_type = new.de->file_type;
3502 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3506 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3511 * Like most other Unix systems, set the ctime for inodes on a
3514 old.inode->i_ctime = ext4_current_time(old.inode);
3515 new.inode->i_ctime = ext4_current_time(new.inode);
3516 ext4_mark_inode_dirty(handle, old.inode);
3517 ext4_mark_inode_dirty(handle, new.inode);
3520 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3525 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3529 ext4_update_dir_count(handle, &old);
3530 ext4_update_dir_count(handle, &new);
3539 ext4_journal_stop(handle);
3543 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3544 struct inode *new_dir, struct dentry *new_dentry,
3547 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3550 if (flags & RENAME_EXCHANGE) {
3551 return ext4_cross_rename(old_dir, old_dentry,
3552 new_dir, new_dentry);
3555 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3559 * directories can handle most operations...
3561 const struct inode_operations ext4_dir_inode_operations = {
3562 .create = ext4_create,
3563 .lookup = ext4_lookup,
3565 .unlink = ext4_unlink,
3566 .symlink = ext4_symlink,
3567 .mkdir = ext4_mkdir,
3568 .rmdir = ext4_rmdir,
3569 .mknod = ext4_mknod,
3570 .tmpfile = ext4_tmpfile,
3571 .rename2 = ext4_rename2,
3572 .setattr = ext4_setattr,
3573 .setxattr = generic_setxattr,
3574 .getxattr = generic_getxattr,
3575 .listxattr = ext4_listxattr,
3576 .removexattr = generic_removexattr,
3577 .get_acl = ext4_get_acl,
3578 .set_acl = ext4_set_acl,
3579 .fiemap = ext4_fiemap,
3582 const struct inode_operations ext4_special_inode_operations = {
3583 .setattr = ext4_setattr,
3584 .setxattr = generic_setxattr,
3585 .getxattr = generic_getxattr,
3586 .listxattr = ext4_listxattr,
3587 .removexattr = generic_removexattr,
3588 .get_acl = ext4_get_acl,
3589 .set_acl = ext4_set_acl,