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, EXT4_GET_BLOCKS_CREATE);
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), __func__, __LINE__)
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
95 struct buffer_head *bh;
96 struct ext4_dir_entry *dirent;
99 bh = ext4_bread(NULL, inode, block, 0);
101 __ext4_warning(inode->i_sb, func, line,
102 "inode #%lu: lblock %lu: comm %s: "
103 "error %ld reading directory block",
104 inode->i_ino, (unsigned long)block,
105 current->comm, PTR_ERR(bh));
110 ext4_error_inode(inode, func, line, block,
111 "Directory hole found");
112 return ERR_PTR(-EFSCORRUPTED);
114 dirent = (struct ext4_dir_entry *) bh->b_data;
115 /* Determine whether or not we have an index block */
119 else if (ext4_rec_len_from_disk(dirent->rec_len,
120 inode->i_sb->s_blocksize) ==
121 inode->i_sb->s_blocksize)
124 if (!is_dx_block && type == INDEX) {
125 ext4_error_inode(inode, func, line, block,
126 "directory leaf block found instead of index block");
127 return ERR_PTR(-EFSCORRUPTED);
129 if (!ext4_has_metadata_csum(inode->i_sb) ||
134 * An empty leaf block can get mistaken for a index block; for
135 * this reason, we can only check the index checksum when the
136 * caller is sure it should be an index block.
138 if (is_dx_block && type == INDEX) {
139 if (ext4_dx_csum_verify(inode, dirent))
140 set_buffer_verified(bh);
142 ext4_error_inode(inode, func, line, block,
143 "Directory index failed checksum");
145 return ERR_PTR(-EFSBADCRC);
149 if (ext4_dirent_csum_verify(inode, dirent))
150 set_buffer_verified(bh);
152 ext4_error_inode(inode, func, line, block,
153 "Directory block failed checksum");
155 return ERR_PTR(-EFSBADCRC);
162 #define assert(test) J_ASSERT(test)
166 #define dxtrace(command) command
168 #define dxtrace(command)
192 * dx_root_info is laid out so that if it should somehow get overlaid by a
193 * dirent the two low bits of the hash version will be zero. Therefore, the
194 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
199 struct fake_dirent dot;
201 struct fake_dirent dotdot;
205 __le32 reserved_zero;
207 u8 info_length; /* 8 */
212 struct dx_entry entries[0];
217 struct fake_dirent fake;
218 struct dx_entry entries[0];
224 struct buffer_head *bh;
225 struct dx_entry *entries;
237 * This goes at the end of each htree block.
241 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
244 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
245 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
246 static inline unsigned dx_get_hash(struct dx_entry *entry);
247 static void dx_set_hash(struct dx_entry *entry, unsigned value);
248 static unsigned dx_get_count(struct dx_entry *entries);
249 static unsigned dx_get_limit(struct dx_entry *entries);
250 static void dx_set_count(struct dx_entry *entries, unsigned value);
251 static void dx_set_limit(struct dx_entry *entries, unsigned value);
252 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
253 static unsigned dx_node_limit(struct inode *dir);
254 static struct dx_frame *dx_probe(struct ext4_filename *fname,
256 struct dx_hash_info *hinfo,
257 struct dx_frame *frame);
258 static void dx_release(struct dx_frame *frames);
259 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
260 unsigned blocksize, struct dx_hash_info *hinfo,
261 struct dx_map_entry map[]);
262 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
263 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
264 struct dx_map_entry *offsets, int count, unsigned blocksize);
265 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
266 static void dx_insert_block(struct dx_frame *frame,
267 u32 hash, ext4_lblk_t block);
268 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
269 struct dx_frame *frame,
270 struct dx_frame *frames,
272 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
273 struct ext4_filename *fname,
274 struct ext4_dir_entry_2 **res_dir);
275 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
276 struct dentry *dentry, struct inode *inode);
278 /* checksumming functions */
279 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
280 unsigned int blocksize)
282 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
283 t->det_rec_len = ext4_rec_len_to_disk(
284 sizeof(struct ext4_dir_entry_tail), blocksize);
285 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
288 /* Walk through a dirent block to find a checksum "dirent" at the tail */
289 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
290 struct ext4_dir_entry *de)
292 struct ext4_dir_entry_tail *t;
295 struct ext4_dir_entry *d, *top;
298 top = (struct ext4_dir_entry *)(((void *)de) +
299 (EXT4_BLOCK_SIZE(inode->i_sb) -
300 sizeof(struct ext4_dir_entry_tail)));
301 while (d < top && d->rec_len)
302 d = (struct ext4_dir_entry *)(((void *)d) +
303 le16_to_cpu(d->rec_len));
308 t = (struct ext4_dir_entry_tail *)d;
310 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
313 if (t->det_reserved_zero1 ||
314 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
315 t->det_reserved_zero2 ||
316 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
322 static __le32 ext4_dirent_csum(struct inode *inode,
323 struct ext4_dir_entry *dirent, int size)
325 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
326 struct ext4_inode_info *ei = EXT4_I(inode);
329 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
330 return cpu_to_le32(csum);
333 #define warn_no_space_for_csum(inode) \
334 __warn_no_space_for_csum((inode), __func__, __LINE__)
336 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
339 __ext4_warning_inode(inode, func, line,
340 "No space for directory leaf checksum. Please run e2fsck -D.");
343 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
345 struct ext4_dir_entry_tail *t;
347 if (!ext4_has_metadata_csum(inode->i_sb))
350 t = get_dirent_tail(inode, dirent);
352 warn_no_space_for_csum(inode);
356 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
357 (void *)t - (void *)dirent))
363 static void ext4_dirent_csum_set(struct inode *inode,
364 struct ext4_dir_entry *dirent)
366 struct ext4_dir_entry_tail *t;
368 if (!ext4_has_metadata_csum(inode->i_sb))
371 t = get_dirent_tail(inode, dirent);
373 warn_no_space_for_csum(inode);
377 t->det_checksum = ext4_dirent_csum(inode, dirent,
378 (void *)t - (void *)dirent);
381 int ext4_handle_dirty_dirent_node(handle_t *handle,
383 struct buffer_head *bh)
385 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
386 return ext4_handle_dirty_metadata(handle, inode, bh);
389 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
390 struct ext4_dir_entry *dirent,
393 struct ext4_dir_entry *dp;
394 struct dx_root_info *root;
397 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
399 else if (le16_to_cpu(dirent->rec_len) == 12) {
400 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
401 if (le16_to_cpu(dp->rec_len) !=
402 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
404 root = (struct dx_root_info *)(((void *)dp + 12));
405 if (root->reserved_zero ||
406 root->info_length != sizeof(struct dx_root_info))
413 *offset = count_offset;
414 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
417 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
418 int count_offset, int count, struct dx_tail *t)
420 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
421 struct ext4_inode_info *ei = EXT4_I(inode);
424 __u32 dummy_csum = 0;
425 int offset = offsetof(struct dx_tail, dt_checksum);
427 size = count_offset + (count * sizeof(struct dx_entry));
428 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
429 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
430 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
432 return cpu_to_le32(csum);
435 static int ext4_dx_csum_verify(struct inode *inode,
436 struct ext4_dir_entry *dirent)
438 struct dx_countlimit *c;
440 int count_offset, limit, count;
442 if (!ext4_has_metadata_csum(inode->i_sb))
445 c = get_dx_countlimit(inode, dirent, &count_offset);
447 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
450 limit = le16_to_cpu(c->limit);
451 count = le16_to_cpu(c->count);
452 if (count_offset + (limit * sizeof(struct dx_entry)) >
453 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
454 warn_no_space_for_csum(inode);
457 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
459 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
465 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
467 struct dx_countlimit *c;
469 int count_offset, limit, count;
471 if (!ext4_has_metadata_csum(inode->i_sb))
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_metadata_csum(dir->i_sb))
560 entry_space -= sizeof(struct dx_tail);
561 return entry_space / sizeof(struct dx_entry);
564 static inline unsigned dx_node_limit(struct inode *dir)
566 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
568 if (ext4_has_metadata_csum(dir->i_sb))
569 entry_space -= sizeof(struct dx_tail);
570 return entry_space / sizeof(struct dx_entry);
577 static void dx_show_index(char * label, struct dx_entry *entries)
579 int i, n = dx_get_count (entries);
580 printk(KERN_DEBUG "%s index ", label);
581 for (i = 0; i < n; i++) {
582 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
583 0, (unsigned long)dx_get_block(entries + i));
595 static struct stats dx_show_leaf(struct inode *dir,
596 struct dx_hash_info *hinfo,
597 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 #ifdef CONFIG_EXT4_FS_ENCRYPTION
614 struct ext4_str fname_crypto_str
615 = {.name = NULL, .len = 0};
620 if (ext4_encrypted_inode(inode))
621 res = ext4_get_encryption_info(dir);
623 printk(KERN_WARNING "Error setting up"
624 " fname crypto: %d\n", res);
627 /* Directory is not encrypted */
628 ext4fs_dirhash(de->name,
630 printk("%*.s:(U)%x.%u ", len,
632 (unsigned) ((char *) de
635 /* Directory is encrypted */
636 res = ext4_fname_crypto_alloc_buffer(
640 printk(KERN_WARNING "Error "
646 res = ext4_fname_disk_to_usr(ctx, NULL, de,
649 printk(KERN_WARNING "Error "
650 "converting filename "
656 name = fname_crypto_str.name;
657 len = fname_crypto_str.len;
659 ext4fs_dirhash(de->name, de->name_len,
661 printk("%*.s:(E)%x.%u ", len, name,
662 h.hash, (unsigned) ((char *) de
664 ext4_fname_crypto_free_buffer(
668 int len = de->name_len;
669 char *name = de->name;
670 ext4fs_dirhash(de->name, de->name_len, &h);
671 printk("%*.s:%x.%u ", len, name, h.hash,
672 (unsigned) ((char *) de - base));
675 space += EXT4_DIR_REC_LEN(de->name_len);
678 de = ext4_next_entry(de, size);
680 printk("(%i)\n", names);
681 return (struct stats) { names, space, 1 };
684 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
685 struct dx_entry *entries, int levels)
687 unsigned blocksize = dir->i_sb->s_blocksize;
688 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
690 struct buffer_head *bh;
691 printk("%i indexed blocks...\n", count);
692 for (i = 0; i < count; i++, entries++)
694 ext4_lblk_t block = dx_get_block(entries);
695 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
696 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
698 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
699 bh = ext4_bread(NULL,dir, block, 0);
700 if (!bh || IS_ERR(bh))
703 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
704 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
705 bh->b_data, blocksize, 0);
706 names += stats.names;
707 space += stats.space;
708 bcount += stats.bcount;
712 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
713 levels ? "" : " ", names, space/bcount,
714 (space/bcount)*100/blocksize);
715 return (struct stats) { names, space, bcount};
717 #endif /* DX_DEBUG */
720 * Probe for a directory leaf block to search.
722 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
723 * error in the directory index, and the caller should fall back to
724 * searching the directory normally. The callers of dx_probe **MUST**
725 * check for this error code, and make sure it never gets reflected
728 static struct dx_frame *
729 dx_probe(struct ext4_filename *fname, struct inode *dir,
730 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
732 unsigned count, indirect;
733 struct dx_entry *at, *entries, *p, *q, *m;
734 struct dx_root *root;
735 struct dx_frame *frame = frame_in;
736 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
739 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
740 if (IS_ERR(frame->bh))
741 return (struct dx_frame *) frame->bh;
743 root = (struct dx_root *) frame->bh->b_data;
744 if (root->info.hash_version != DX_HASH_TEA &&
745 root->info.hash_version != DX_HASH_HALF_MD4 &&
746 root->info.hash_version != DX_HASH_LEGACY) {
747 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
748 root->info.hash_version);
752 hinfo = &fname->hinfo;
753 hinfo->hash_version = root->info.hash_version;
754 if (hinfo->hash_version <= DX_HASH_TEA)
755 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
756 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
757 if (fname && fname_name(fname))
758 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
761 if (root->info.unused_flags & 1) {
762 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
763 root->info.unused_flags);
767 indirect = root->info.indirect_levels;
769 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
770 root->info.indirect_levels);
774 entries = (struct dx_entry *)(((char *)&root->info) +
775 root->info.info_length);
777 if (dx_get_limit(entries) != dx_root_limit(dir,
778 root->info.info_length)) {
779 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
780 dx_get_limit(entries),
781 dx_root_limit(dir, root->info.info_length));
785 dxtrace(printk("Look up %x", hash));
787 count = dx_get_count(entries);
788 if (!count || count > dx_get_limit(entries)) {
789 ext4_warning_inode(dir,
790 "dx entry: count %u beyond limit %u",
791 count, dx_get_limit(entries));
796 q = entries + count - 1;
799 dxtrace(printk("."));
800 if (dx_get_hash(m) > hash)
806 if (0) { // linear search cross check
807 unsigned n = count - 1;
811 dxtrace(printk(","));
812 if (dx_get_hash(++at) > hash)
818 assert (at == p - 1);
822 dxtrace(printk(" %x->%u\n", at == entries ? 0 : dx_get_hash(at),
824 frame->entries = entries;
829 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
830 if (IS_ERR(frame->bh)) {
831 ret_err = (struct dx_frame *) frame->bh;
835 entries = ((struct dx_node *) frame->bh->b_data)->entries;
837 if (dx_get_limit(entries) != dx_node_limit(dir)) {
838 ext4_warning_inode(dir,
839 "dx entry: limit %u != node limit %u",
840 dx_get_limit(entries), dx_node_limit(dir));
845 while (frame >= frame_in) {
850 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
851 ext4_warning_inode(dir,
852 "Corrupt directory, running e2fsck is recommended");
856 static void dx_release(struct dx_frame *frames)
858 if (frames[0].bh == NULL)
861 if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
862 brelse(frames[1].bh);
863 brelse(frames[0].bh);
867 * This function increments the frame pointer to search the next leaf
868 * block, and reads in the necessary intervening nodes if the search
869 * should be necessary. Whether or not the search is necessary is
870 * controlled by the hash parameter. If the hash value is even, then
871 * the search is only continued if the next block starts with that
872 * hash value. This is used if we are searching for a specific file.
874 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
876 * This function returns 1 if the caller should continue to search,
877 * or 0 if it should not. If there is an error reading one of the
878 * index blocks, it will a negative error code.
880 * If start_hash is non-null, it will be filled in with the starting
881 * hash of the next page.
883 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
884 struct dx_frame *frame,
885 struct dx_frame *frames,
889 struct buffer_head *bh;
895 * Find the next leaf page by incrementing the frame pointer.
896 * If we run out of entries in the interior node, loop around and
897 * increment pointer in the parent node. When we break out of
898 * this loop, num_frames indicates the number of interior
899 * nodes need to be read.
902 if (++(p->at) < p->entries + dx_get_count(p->entries))
911 * If the hash is 1, then continue only if the next page has a
912 * continuation hash of any value. This is used for readdir
913 * handling. Otherwise, check to see if the hash matches the
914 * desired contiuation hash. If it doesn't, return since
915 * there's no point to read in the successive index pages.
917 bhash = dx_get_hash(p->at);
920 if ((hash & 1) == 0) {
921 if ((bhash & ~1) != hash)
925 * If the hash is HASH_NB_ALWAYS, we always go to the next
926 * block so no check is necessary
928 while (num_frames--) {
929 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
935 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
942 * This function fills a red-black tree with information from a
943 * directory block. It returns the number directory entries loaded
944 * into the tree. If there is an error it is returned in err.
946 static int htree_dirblock_to_tree(struct file *dir_file,
947 struct inode *dir, ext4_lblk_t block,
948 struct dx_hash_info *hinfo,
949 __u32 start_hash, __u32 start_minor_hash)
951 struct buffer_head *bh;
952 struct ext4_dir_entry_2 *de, *top;
953 int err = 0, count = 0;
954 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}, tmp_str;
956 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
957 (unsigned long)block));
958 bh = ext4_read_dirblock(dir, block, DIRENT);
962 de = (struct ext4_dir_entry_2 *) bh->b_data;
963 top = (struct ext4_dir_entry_2 *) ((char *) de +
964 dir->i_sb->s_blocksize -
965 EXT4_DIR_REC_LEN(0));
966 #ifdef CONFIG_EXT4_FS_ENCRYPTION
967 /* Check if the directory is encrypted */
968 if (ext4_encrypted_inode(dir)) {
969 err = ext4_get_encryption_info(dir);
974 err = ext4_fname_crypto_alloc_buffer(dir, EXT4_NAME_LEN,
982 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
983 if (ext4_check_dir_entry(dir, NULL, de, bh,
984 bh->b_data, bh->b_size,
985 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
986 + ((char *)de - bh->b_data))) {
987 /* silently ignore the rest of the block */
990 ext4fs_dirhash(de->name, de->name_len, hinfo);
991 if ((hinfo->hash < start_hash) ||
992 ((hinfo->hash == start_hash) &&
993 (hinfo->minor_hash < start_minor_hash)))
997 if (!ext4_encrypted_inode(dir)) {
998 tmp_str.name = de->name;
999 tmp_str.len = de->name_len;
1000 err = ext4_htree_store_dirent(dir_file,
1001 hinfo->hash, hinfo->minor_hash, de,
1004 int save_len = fname_crypto_str.len;
1006 /* Directory is encrypted */
1007 err = ext4_fname_disk_to_usr(dir, hinfo, de,
1013 err = ext4_htree_store_dirent(dir_file,
1014 hinfo->hash, hinfo->minor_hash, de,
1016 fname_crypto_str.len = save_len;
1026 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1027 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1034 * This function fills a red-black tree with information from a
1035 * directory. We start scanning the directory in hash order, starting
1036 * at start_hash and start_minor_hash.
1038 * This function returns the number of entries inserted into the tree,
1039 * or a negative error code.
1041 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1042 __u32 start_minor_hash, __u32 *next_hash)
1044 struct dx_hash_info hinfo;
1045 struct ext4_dir_entry_2 *de;
1046 struct dx_frame frames[2], *frame;
1052 struct ext4_str tmp_str;
1054 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1055 start_hash, start_minor_hash));
1056 dir = file_inode(dir_file);
1057 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1058 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1059 if (hinfo.hash_version <= DX_HASH_TEA)
1060 hinfo.hash_version +=
1061 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1062 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1063 if (ext4_has_inline_data(dir)) {
1064 int has_inline_data = 1;
1065 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1069 if (has_inline_data) {
1074 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1075 start_hash, start_minor_hash);
1079 hinfo.hash = start_hash;
1080 hinfo.minor_hash = 0;
1081 frame = dx_probe(NULL, dir, &hinfo, frames);
1083 return PTR_ERR(frame);
1085 /* Add '.' and '..' from the htree header */
1086 if (!start_hash && !start_minor_hash) {
1087 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1088 tmp_str.name = de->name;
1089 tmp_str.len = de->name_len;
1090 err = ext4_htree_store_dirent(dir_file, 0, 0,
1096 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1097 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1098 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1099 tmp_str.name = de->name;
1100 tmp_str.len = de->name_len;
1101 err = ext4_htree_store_dirent(dir_file, 2, 0,
1109 block = dx_get_block(frame->at);
1110 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1111 start_hash, start_minor_hash);
1118 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1119 frame, frames, &hashval);
1120 *next_hash = hashval;
1126 * Stop if: (a) there are no more entries, or
1127 * (b) we have inserted at least one entry and the
1128 * next hash value is not a continuation
1131 (count && ((hashval & 1) == 0)))
1135 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1136 "next hash: %x\n", count, *next_hash));
1143 static inline int search_dirblock(struct buffer_head *bh,
1145 struct ext4_filename *fname,
1146 const struct qstr *d_name,
1147 unsigned int offset,
1148 struct ext4_dir_entry_2 **res_dir)
1150 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1151 fname, d_name, offset, res_dir);
1155 * Directory block splitting, compacting
1159 * Create map of hash values, offsets, and sizes, stored at end of block.
1160 * Returns number of entries mapped.
1162 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1163 unsigned blocksize, struct dx_hash_info *hinfo,
1164 struct dx_map_entry *map_tail)
1167 char *base = (char *) de;
1168 struct dx_hash_info h = *hinfo;
1170 while ((char *) de < base + blocksize) {
1171 if (de->name_len && de->inode) {
1172 ext4fs_dirhash(de->name, de->name_len, &h);
1174 map_tail->hash = h.hash;
1175 map_tail->offs = ((char *) de - base)>>2;
1176 map_tail->size = le16_to_cpu(de->rec_len);
1180 /* XXX: do we need to check rec_len == 0 case? -Chris */
1181 de = ext4_next_entry(de, blocksize);
1186 /* Sort map by hash value */
1187 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1189 struct dx_map_entry *p, *q, *top = map + count - 1;
1191 /* Combsort until bubble sort doesn't suck */
1193 count = count*10/13;
1194 if (count - 9 < 2) /* 9, 10 -> 11 */
1196 for (p = top, q = p - count; q >= map; p--, q--)
1197 if (p->hash < q->hash)
1200 /* Garden variety bubble sort */
1205 if (q[1].hash >= q[0].hash)
1213 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1215 struct dx_entry *entries = frame->entries;
1216 struct dx_entry *old = frame->at, *new = old + 1;
1217 int count = dx_get_count(entries);
1219 assert(count < dx_get_limit(entries));
1220 assert(old < entries + count);
1221 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1222 dx_set_hash(new, hash);
1223 dx_set_block(new, block);
1224 dx_set_count(entries, count + 1);
1228 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1230 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1231 * `de != NULL' is guaranteed by caller.
1233 static inline int ext4_match(struct ext4_filename *fname,
1234 struct ext4_dir_entry_2 *de)
1236 const void *name = fname_name(fname);
1237 u32 len = fname_len(fname);
1242 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1243 if (unlikely(!name)) {
1244 if (fname->usr_fname->name[0] == '_') {
1246 if (de->name_len < 16)
1248 ret = memcmp(de->name + de->name_len - 16,
1249 fname->crypto_buf.name + 8, 16);
1250 return (ret == 0) ? 1 : 0;
1252 name = fname->crypto_buf.name;
1253 len = fname->crypto_buf.len;
1256 if (de->name_len != len)
1258 return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1262 * Returns 0 if not found, -1 on failure, and 1 on success
1264 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1265 struct inode *dir, struct ext4_filename *fname,
1266 const struct qstr *d_name,
1267 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1269 struct ext4_dir_entry_2 * de;
1274 de = (struct ext4_dir_entry_2 *)search_buf;
1275 dlimit = search_buf + buf_size;
1276 while ((char *) de < dlimit) {
1277 /* this code is executed quadratically often */
1278 /* do minimal checking `by hand' */
1279 if ((char *) de + de->name_len <= dlimit) {
1280 res = ext4_match(fname, de);
1286 /* found a match - just to be sure, do
1288 if (ext4_check_dir_entry(dir, NULL, de, bh,
1290 bh->b_size, offset)) {
1300 /* prevent looping on a bad block */
1301 de_len = ext4_rec_len_from_disk(de->rec_len,
1302 dir->i_sb->s_blocksize);
1308 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1316 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1317 struct ext4_dir_entry *de)
1319 struct super_block *sb = dir->i_sb;
1325 if (de->inode == 0 &&
1326 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1335 * finds an entry in the specified directory with the wanted name. It
1336 * returns the cache buffer in which the entry was found, and the entry
1337 * itself (as a parameter - res_dir). It does NOT read the inode of the
1338 * entry - you'll have to do that yourself if you want to.
1340 * The returned buffer_head has ->b_count elevated. The caller is expected
1341 * to brelse() it when appropriate.
1343 static struct buffer_head * ext4_find_entry (struct inode *dir,
1344 const struct qstr *d_name,
1345 struct ext4_dir_entry_2 **res_dir,
1348 struct super_block *sb;
1349 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1350 struct buffer_head *bh, *ret = NULL;
1351 ext4_lblk_t start, block, b;
1352 const u8 *name = d_name->name;
1353 int ra_max = 0; /* Number of bh's in the readahead
1355 int ra_ptr = 0; /* Current index into readahead
1358 ext4_lblk_t nblocks;
1359 int i, namelen, retval;
1360 struct ext4_filename fname;
1364 namelen = d_name->len;
1365 if (namelen > EXT4_NAME_LEN)
1368 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1370 return ERR_PTR(retval);
1372 if (ext4_has_inline_data(dir)) {
1373 int has_inline_data = 1;
1374 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1376 if (has_inline_data) {
1379 goto cleanup_and_exit;
1383 if ((namelen <= 2) && (name[0] == '.') &&
1384 (name[1] == '.' || name[1] == '\0')) {
1386 * "." or ".." will only be in the first block
1387 * NFS may look up ".."; "." should be handled by the VFS
1394 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1396 * On success, or if the error was file not found,
1397 * return. Otherwise, fall back to doing a search the
1398 * old fashioned way.
1400 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1401 goto cleanup_and_exit;
1402 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1405 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1406 start = EXT4_I(dir)->i_dir_start_lookup;
1407 if (start >= nblocks)
1413 * We deal with the read-ahead logic here.
1415 if (ra_ptr >= ra_max) {
1416 /* Refill the readahead buffer */
1419 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1421 * Terminate if we reach the end of the
1422 * directory and must wrap, or if our
1423 * search has finished at this block.
1425 if (b >= nblocks || (num && block == start)) {
1426 bh_use[ra_max] = NULL;
1430 bh = ext4_getblk(NULL, dir, b++, 0);
1434 goto cleanup_and_exit;
1438 bh_use[ra_max] = bh;
1440 ll_rw_block(READ | REQ_META | REQ_PRIO,
1444 if ((bh = bh_use[ra_ptr++]) == NULL)
1447 if (!buffer_uptodate(bh)) {
1448 /* read error, skip block & hope for the best */
1449 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1450 (unsigned long) block);
1454 if (!buffer_verified(bh) &&
1455 !is_dx_internal_node(dir, block,
1456 (struct ext4_dir_entry *)bh->b_data) &&
1457 !ext4_dirent_csum_verify(dir,
1458 (struct ext4_dir_entry *)bh->b_data)) {
1459 EXT4_ERROR_INODE(dir, "checksumming directory "
1460 "block %lu", (unsigned long)block);
1464 set_buffer_verified(bh);
1465 i = search_dirblock(bh, dir, &fname, d_name,
1466 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1468 EXT4_I(dir)->i_dir_start_lookup = block;
1470 goto cleanup_and_exit;
1474 goto cleanup_and_exit;
1477 if (++block >= nblocks)
1479 } while (block != start);
1482 * If the directory has grown while we were searching, then
1483 * search the last part of the directory before giving up.
1486 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1487 if (block < nblocks) {
1493 /* Clean up the read-ahead blocks */
1494 for (; ra_ptr < ra_max; ra_ptr++)
1495 brelse(bh_use[ra_ptr]);
1496 ext4_fname_free_filename(&fname);
1500 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1501 struct ext4_filename *fname,
1502 struct ext4_dir_entry_2 **res_dir)
1504 struct super_block * sb = dir->i_sb;
1505 struct dx_frame frames[2], *frame;
1506 const struct qstr *d_name = fname->usr_fname;
1507 struct buffer_head *bh;
1511 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1514 frame = dx_probe(fname, dir, NULL, frames);
1516 return (struct buffer_head *) frame;
1518 block = dx_get_block(frame->at);
1519 bh = ext4_read_dirblock(dir, block, DIRENT);
1523 retval = search_dirblock(bh, dir, fname, d_name,
1524 block << EXT4_BLOCK_SIZE_BITS(sb),
1530 bh = ERR_PTR(ERR_BAD_DX_DIR);
1534 /* Check to see if we should continue to search */
1535 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1538 ext4_warning_inode(dir,
1539 "error %d reading directory index block",
1541 bh = ERR_PTR(retval);
1544 } while (retval == 1);
1548 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1554 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1556 struct inode *inode;
1557 struct ext4_dir_entry_2 *de;
1558 struct buffer_head *bh;
1560 if (dentry->d_name.len > EXT4_NAME_LEN)
1561 return ERR_PTR(-ENAMETOOLONG);
1563 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1565 return (struct dentry *) bh;
1568 __u32 ino = le32_to_cpu(de->inode);
1570 if (!ext4_valid_inum(dir->i_sb, ino)) {
1571 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1572 return ERR_PTR(-EFSCORRUPTED);
1574 if (unlikely(ino == dir->i_ino)) {
1575 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1577 return ERR_PTR(-EFSCORRUPTED);
1579 inode = ext4_iget_normal(dir->i_sb, ino);
1580 if (inode == ERR_PTR(-ESTALE)) {
1581 EXT4_ERROR_INODE(dir,
1582 "deleted inode referenced: %u",
1584 return ERR_PTR(-EFSCORRUPTED);
1586 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1587 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1588 S_ISLNK(inode->i_mode)) &&
1589 !ext4_is_child_context_consistent_with_parent(dir,
1592 ext4_warning(inode->i_sb,
1593 "Inconsistent encryption contexts: %lu/%lu\n",
1594 (unsigned long) dir->i_ino,
1595 (unsigned long) inode->i_ino);
1596 return ERR_PTR(-EPERM);
1599 return d_splice_alias(inode, dentry);
1603 struct dentry *ext4_get_parent(struct dentry *child)
1606 static const struct qstr dotdot = QSTR_INIT("..", 2);
1607 struct ext4_dir_entry_2 * de;
1608 struct buffer_head *bh;
1610 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1612 return (struct dentry *) bh;
1614 return ERR_PTR(-ENOENT);
1615 ino = le32_to_cpu(de->inode);
1618 if (!ext4_valid_inum(d_inode(child)->i_sb, ino)) {
1619 EXT4_ERROR_INODE(d_inode(child),
1620 "bad parent inode number: %u", ino);
1621 return ERR_PTR(-EFSCORRUPTED);
1624 return d_obtain_alias(ext4_iget_normal(d_inode(child)->i_sb, ino));
1628 * Move count entries from end of map between two memory locations.
1629 * Returns pointer to last entry moved.
1631 static struct ext4_dir_entry_2 *
1632 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1635 unsigned rec_len = 0;
1638 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1639 (from + (map->offs<<2));
1640 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1641 memcpy (to, de, rec_len);
1642 ((struct ext4_dir_entry_2 *) to)->rec_len =
1643 ext4_rec_len_to_disk(rec_len, blocksize);
1648 return (struct ext4_dir_entry_2 *) (to - rec_len);
1652 * Compact each dir entry in the range to the minimal rec_len.
1653 * Returns pointer to last entry in range.
1655 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1657 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1658 unsigned rec_len = 0;
1661 while ((char*)de < base + blocksize) {
1662 next = ext4_next_entry(de, blocksize);
1663 if (de->inode && de->name_len) {
1664 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1666 memmove(to, de, rec_len);
1667 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1669 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1677 * Split a full leaf block to make room for a new dir entry.
1678 * Allocate a new block, and move entries so that they are approx. equally full.
1679 * Returns pointer to de in block into which the new entry will be inserted.
1681 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1682 struct buffer_head **bh,struct dx_frame *frame,
1683 struct dx_hash_info *hinfo)
1685 unsigned blocksize = dir->i_sb->s_blocksize;
1686 unsigned count, continued;
1687 struct buffer_head *bh2;
1688 ext4_lblk_t newblock;
1690 struct dx_map_entry *map;
1691 char *data1 = (*bh)->b_data, *data2;
1692 unsigned split, move, size;
1693 struct ext4_dir_entry_2 *de = NULL, *de2;
1694 struct ext4_dir_entry_tail *t;
1698 if (ext4_has_metadata_csum(dir->i_sb))
1699 csum_size = sizeof(struct ext4_dir_entry_tail);
1701 bh2 = ext4_append(handle, dir, &newblock);
1705 return (struct ext4_dir_entry_2 *) bh2;
1708 BUFFER_TRACE(*bh, "get_write_access");
1709 err = ext4_journal_get_write_access(handle, *bh);
1713 BUFFER_TRACE(frame->bh, "get_write_access");
1714 err = ext4_journal_get_write_access(handle, frame->bh);
1718 data2 = bh2->b_data;
1720 /* create map in the end of data2 block */
1721 map = (struct dx_map_entry *) (data2 + blocksize);
1722 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1723 blocksize, hinfo, map);
1725 dx_sort_map(map, count);
1726 /* Split the existing block in the middle, size-wise */
1729 for (i = count-1; i >= 0; i--) {
1730 /* is more than half of this entry in 2nd half of the block? */
1731 if (size + map[i].size/2 > blocksize/2)
1733 size += map[i].size;
1736 /* map index at which we will split */
1737 split = count - move;
1738 hash2 = map[split].hash;
1739 continued = hash2 == map[split - 1].hash;
1740 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1741 (unsigned long)dx_get_block(frame->at),
1742 hash2, split, count-split));
1744 /* Fancy dance to stay within two buffers */
1745 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1747 de = dx_pack_dirents(data1, blocksize);
1748 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1751 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1755 t = EXT4_DIRENT_TAIL(data2, blocksize);
1756 initialize_dirent_tail(t, blocksize);
1758 t = EXT4_DIRENT_TAIL(data1, blocksize);
1759 initialize_dirent_tail(t, blocksize);
1762 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1764 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1767 /* Which block gets the new entry? */
1768 if (hinfo->hash >= hash2) {
1772 dx_insert_block(frame, hash2 + continued, newblock);
1773 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1776 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1780 dxtrace(dx_show_index("frame", frame->entries));
1787 ext4_std_error(dir->i_sb, err);
1788 return ERR_PTR(err);
1791 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1792 struct buffer_head *bh,
1793 void *buf, int buf_size,
1794 struct ext4_filename *fname,
1795 struct ext4_dir_entry_2 **dest_de)
1797 struct ext4_dir_entry_2 *de;
1798 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1800 unsigned int offset = 0;
1804 de = (struct ext4_dir_entry_2 *)buf;
1805 top = buf + buf_size - reclen;
1806 while ((char *) de <= top) {
1807 if (ext4_check_dir_entry(dir, NULL, de, bh,
1808 buf, buf_size, offset)) {
1809 res = -EFSCORRUPTED;
1812 /* Provide crypto context and crypto buffer to ext4 match */
1813 res = ext4_match(fname, de);
1820 nlen = EXT4_DIR_REC_LEN(de->name_len);
1821 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1822 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1824 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1828 if ((char *) de > top)
1838 int ext4_insert_dentry(struct inode *dir,
1839 struct inode *inode,
1840 struct ext4_dir_entry_2 *de,
1842 struct ext4_filename *fname)
1847 nlen = EXT4_DIR_REC_LEN(de->name_len);
1848 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1850 struct ext4_dir_entry_2 *de1 =
1851 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1852 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1853 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1856 de->file_type = EXT4_FT_UNKNOWN;
1857 de->inode = cpu_to_le32(inode->i_ino);
1858 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1859 de->name_len = fname_len(fname);
1860 memcpy(de->name, fname_name(fname), fname_len(fname));
1865 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1866 * it points to a directory entry which is guaranteed to be large
1867 * enough for new directory entry. If de is NULL, then
1868 * add_dirent_to_buf will attempt search the directory block for
1869 * space. It will return -ENOSPC if no space is available, and -EIO
1870 * and -EEXIST if directory entry already exists.
1872 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1874 struct inode *inode, struct ext4_dir_entry_2 *de,
1875 struct buffer_head *bh)
1877 unsigned int blocksize = dir->i_sb->s_blocksize;
1881 if (ext4_has_metadata_csum(inode->i_sb))
1882 csum_size = sizeof(struct ext4_dir_entry_tail);
1885 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1886 blocksize - csum_size, fname, &de);
1890 BUFFER_TRACE(bh, "get_write_access");
1891 err = ext4_journal_get_write_access(handle, bh);
1893 ext4_std_error(dir->i_sb, err);
1897 /* By now the buffer is marked for journaling. Due to crypto operations,
1898 * the following function call may fail */
1899 err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1904 * XXX shouldn't update any times until successful
1905 * completion of syscall, but too many callers depend
1908 * XXX similarly, too many callers depend on
1909 * ext4_new_inode() setting the times, but error
1910 * recovery deletes the inode, so the worst that can
1911 * happen is that the times are slightly out of date
1912 * and/or different from the directory change time.
1914 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1915 ext4_update_dx_flag(dir);
1917 ext4_mark_inode_dirty(handle, dir);
1918 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1919 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1921 ext4_std_error(dir->i_sb, err);
1926 * This converts a one block unindexed directory to a 3 block indexed
1927 * directory, and adds the dentry to the indexed directory.
1929 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1930 struct dentry *dentry,
1931 struct inode *inode, struct buffer_head *bh)
1933 struct inode *dir = d_inode(dentry->d_parent);
1934 struct buffer_head *bh2;
1935 struct dx_root *root;
1936 struct dx_frame frames[2], *frame;
1937 struct dx_entry *entries;
1938 struct ext4_dir_entry_2 *de, *de2;
1939 struct ext4_dir_entry_tail *t;
1945 struct fake_dirent *fde;
1948 if (ext4_has_metadata_csum(inode->i_sb))
1949 csum_size = sizeof(struct ext4_dir_entry_tail);
1951 blocksize = dir->i_sb->s_blocksize;
1952 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1953 BUFFER_TRACE(bh, "get_write_access");
1954 retval = ext4_journal_get_write_access(handle, bh);
1956 ext4_std_error(dir->i_sb, retval);
1960 root = (struct dx_root *) bh->b_data;
1962 /* The 0th block becomes the root, move the dirents out */
1963 fde = &root->dotdot;
1964 de = (struct ext4_dir_entry_2 *)((char *)fde +
1965 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1966 if ((char *) de >= (((char *) root) + blocksize)) {
1967 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1969 return -EFSCORRUPTED;
1971 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1973 /* Allocate new block for the 0th block's dirents */
1974 bh2 = ext4_append(handle, dir, &block);
1977 return PTR_ERR(bh2);
1979 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1980 data1 = bh2->b_data;
1982 memcpy (data1, de, len);
1983 de = (struct ext4_dir_entry_2 *) data1;
1985 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1987 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1992 t = EXT4_DIRENT_TAIL(data1, blocksize);
1993 initialize_dirent_tail(t, blocksize);
1996 /* Initialize the root; the dot dirents already exist */
1997 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1998 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2000 memset (&root->info, 0, sizeof(root->info));
2001 root->info.info_length = sizeof(root->info);
2002 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2003 entries = root->entries;
2004 dx_set_block(entries, 1);
2005 dx_set_count(entries, 1);
2006 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2008 /* Initialize as for dx_probe */
2009 fname->hinfo.hash_version = root->info.hash_version;
2010 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2011 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2012 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2013 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2015 memset(frames, 0, sizeof(frames));
2017 frame->entries = entries;
2018 frame->at = entries;
2022 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2025 retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
2029 de = do_split(handle,dir, &bh, frame, &fname->hinfo);
2031 retval = PTR_ERR(de);
2036 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2041 * Even if the block split failed, we have to properly write
2042 * out all the changes we did so far. Otherwise we can end up
2043 * with corrupted filesystem.
2045 ext4_mark_inode_dirty(handle, dir);
2053 * adds a file entry to the specified directory, using the same
2054 * semantics as ext4_find_entry(). It returns NULL if it failed.
2056 * NOTE!! The inode part of 'de' is left at 0 - which means you
2057 * may not sleep between calling this and putting something into
2058 * the entry, as someone else might have used it while you slept.
2060 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2061 struct inode *inode)
2063 struct inode *dir = d_inode(dentry->d_parent);
2064 struct buffer_head *bh = NULL;
2065 struct ext4_dir_entry_2 *de;
2066 struct ext4_dir_entry_tail *t;
2067 struct super_block *sb;
2068 struct ext4_filename fname;
2072 ext4_lblk_t block, blocks;
2075 if (ext4_has_metadata_csum(inode->i_sb))
2076 csum_size = sizeof(struct ext4_dir_entry_tail);
2079 blocksize = sb->s_blocksize;
2080 if (!dentry->d_name.len)
2083 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2087 if (ext4_has_inline_data(dir)) {
2088 retval = ext4_try_add_inline_entry(handle, &fname,
2099 retval = ext4_dx_add_entry(handle, &fname, dentry, inode);
2100 if (!retval || (retval != ERR_BAD_DX_DIR))
2102 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2104 ext4_mark_inode_dirty(handle, dir);
2106 blocks = dir->i_size >> sb->s_blocksize_bits;
2107 for (block = 0; block < blocks; block++) {
2108 bh = ext4_read_dirblock(dir, block, DIRENT);
2110 retval = PTR_ERR(bh);
2114 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2116 if (retval != -ENOSPC)
2119 if (blocks == 1 && !dx_fallback &&
2120 ext4_has_feature_dir_index(sb)) {
2121 retval = make_indexed_dir(handle, &fname, dentry,
2123 bh = NULL; /* make_indexed_dir releases bh */
2128 bh = ext4_append(handle, dir, &block);
2130 retval = PTR_ERR(bh);
2134 de = (struct ext4_dir_entry_2 *) bh->b_data;
2136 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2139 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2140 initialize_dirent_tail(t, blocksize);
2143 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2145 ext4_fname_free_filename(&fname);
2148 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2153 * Returns 0 for success, or a negative error value
2155 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2156 struct dentry *dentry, struct inode *inode)
2158 struct dx_frame frames[2], *frame;
2159 struct dx_entry *entries, *at;
2160 struct buffer_head *bh;
2161 struct inode *dir = d_inode(dentry->d_parent);
2162 struct super_block *sb = dir->i_sb;
2163 struct ext4_dir_entry_2 *de;
2166 frame = dx_probe(fname, dir, NULL, frames);
2168 return PTR_ERR(frame);
2169 entries = frame->entries;
2171 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2178 BUFFER_TRACE(bh, "get_write_access");
2179 err = ext4_journal_get_write_access(handle, bh);
2183 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2187 /* Block full, should compress but for now just split */
2188 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2189 dx_get_count(entries), dx_get_limit(entries)));
2190 /* Need to split index? */
2191 if (dx_get_count(entries) == dx_get_limit(entries)) {
2192 ext4_lblk_t newblock;
2193 unsigned icount = dx_get_count(entries);
2194 int levels = frame - frames;
2195 struct dx_entry *entries2;
2196 struct dx_node *node2;
2197 struct buffer_head *bh2;
2199 if (levels && (dx_get_count(frames->entries) ==
2200 dx_get_limit(frames->entries))) {
2201 ext4_warning_inode(dir, "Directory index full!");
2205 bh2 = ext4_append(handle, dir, &newblock);
2210 node2 = (struct dx_node *)(bh2->b_data);
2211 entries2 = node2->entries;
2212 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2213 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2215 BUFFER_TRACE(frame->bh, "get_write_access");
2216 err = ext4_journal_get_write_access(handle, frame->bh);
2220 unsigned icount1 = icount/2, icount2 = icount - icount1;
2221 unsigned hash2 = dx_get_hash(entries + icount1);
2222 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2225 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2226 err = ext4_journal_get_write_access(handle,
2231 memcpy((char *) entries2, (char *) (entries + icount1),
2232 icount2 * sizeof(struct dx_entry));
2233 dx_set_count(entries, icount1);
2234 dx_set_count(entries2, icount2);
2235 dx_set_limit(entries2, dx_node_limit(dir));
2237 /* Which index block gets the new entry? */
2238 if (at - entries >= icount1) {
2239 frame->at = at = at - entries - icount1 + entries2;
2240 frame->entries = entries = entries2;
2241 swap(frame->bh, bh2);
2243 dx_insert_block(frames + 0, hash2, newblock);
2244 dxtrace(dx_show_index("node", frames[1].entries));
2245 dxtrace(dx_show_index("node",
2246 ((struct dx_node *) bh2->b_data)->entries));
2247 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2252 dxtrace(printk(KERN_DEBUG
2253 "Creating second level index...\n"));
2254 memcpy((char *) entries2, (char *) entries,
2255 icount * sizeof(struct dx_entry));
2256 dx_set_limit(entries2, dx_node_limit(dir));
2259 dx_set_count(entries, 1);
2260 dx_set_block(entries + 0, newblock);
2261 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2263 /* Add new access path frame */
2265 frame->at = at = at - entries + entries2;
2266 frame->entries = entries = entries2;
2268 err = ext4_journal_get_write_access(handle,
2273 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2275 ext4_std_error(inode->i_sb, err);
2279 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2284 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2288 ext4_std_error(dir->i_sb, err);
2296 * ext4_generic_delete_entry deletes a directory entry by merging it
2297 * with the previous entry
2299 int ext4_generic_delete_entry(handle_t *handle,
2301 struct ext4_dir_entry_2 *de_del,
2302 struct buffer_head *bh,
2307 struct ext4_dir_entry_2 *de, *pde;
2308 unsigned int blocksize = dir->i_sb->s_blocksize;
2313 de = (struct ext4_dir_entry_2 *)entry_buf;
2314 while (i < buf_size - csum_size) {
2315 if (ext4_check_dir_entry(dir, NULL, de, bh,
2316 bh->b_data, bh->b_size, i))
2317 return -EFSCORRUPTED;
2320 pde->rec_len = ext4_rec_len_to_disk(
2321 ext4_rec_len_from_disk(pde->rec_len,
2323 ext4_rec_len_from_disk(de->rec_len,
2331 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2333 de = ext4_next_entry(de, blocksize);
2338 static int ext4_delete_entry(handle_t *handle,
2340 struct ext4_dir_entry_2 *de_del,
2341 struct buffer_head *bh)
2343 int err, csum_size = 0;
2345 if (ext4_has_inline_data(dir)) {
2346 int has_inline_data = 1;
2347 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2349 if (has_inline_data)
2353 if (ext4_has_metadata_csum(dir->i_sb))
2354 csum_size = sizeof(struct ext4_dir_entry_tail);
2356 BUFFER_TRACE(bh, "get_write_access");
2357 err = ext4_journal_get_write_access(handle, bh);
2361 err = ext4_generic_delete_entry(handle, dir, de_del,
2363 dir->i_sb->s_blocksize, csum_size);
2367 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2368 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2375 ext4_std_error(dir->i_sb, err);
2380 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2381 * since this indicates that nlinks count was previously 1.
2383 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2386 if (is_dx(inode) && inode->i_nlink > 1) {
2387 /* limit is 16-bit i_links_count */
2388 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2389 set_nlink(inode, 1);
2390 ext4_set_feature_dir_nlink(inode->i_sb);
2396 * If a directory had nlink == 1, then we should let it be 1. This indicates
2397 * directory has >EXT4_LINK_MAX subdirs.
2399 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2401 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2406 static int ext4_add_nondir(handle_t *handle,
2407 struct dentry *dentry, struct inode *inode)
2409 int err = ext4_add_entry(handle, dentry, inode);
2411 ext4_mark_inode_dirty(handle, inode);
2412 unlock_new_inode(inode);
2413 d_instantiate(dentry, inode);
2417 unlock_new_inode(inode);
2423 * By the time this is called, we already have created
2424 * the directory cache entry for the new file, but it
2425 * is so far negative - it has no inode.
2427 * If the create succeeds, we fill in the inode information
2428 * with d_instantiate().
2430 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2434 struct inode *inode;
2435 int err, credits, retries = 0;
2437 err = dquot_initialize(dir);
2441 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2442 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2444 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2445 NULL, EXT4_HT_DIR, credits);
2446 handle = ext4_journal_current_handle();
2447 err = PTR_ERR(inode);
2448 if (!IS_ERR(inode)) {
2449 inode->i_op = &ext4_file_inode_operations;
2450 inode->i_fop = &ext4_file_operations;
2451 ext4_set_aops(inode);
2452 err = ext4_add_nondir(handle, dentry, inode);
2453 if (!err && IS_DIRSYNC(dir))
2454 ext4_handle_sync(handle);
2457 ext4_journal_stop(handle);
2458 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2463 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2464 umode_t mode, dev_t rdev)
2467 struct inode *inode;
2468 int err, credits, retries = 0;
2470 err = dquot_initialize(dir);
2474 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2475 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2477 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2478 NULL, EXT4_HT_DIR, credits);
2479 handle = ext4_journal_current_handle();
2480 err = PTR_ERR(inode);
2481 if (!IS_ERR(inode)) {
2482 init_special_inode(inode, inode->i_mode, rdev);
2483 inode->i_op = &ext4_special_inode_operations;
2484 err = ext4_add_nondir(handle, dentry, inode);
2485 if (!err && IS_DIRSYNC(dir))
2486 ext4_handle_sync(handle);
2489 ext4_journal_stop(handle);
2490 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2495 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2498 struct inode *inode;
2499 int err, retries = 0;
2501 err = dquot_initialize(dir);
2506 inode = ext4_new_inode_start_handle(dir, mode,
2509 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2510 4 + EXT4_XATTR_TRANS_BLOCKS);
2511 handle = ext4_journal_current_handle();
2512 err = PTR_ERR(inode);
2513 if (!IS_ERR(inode)) {
2514 inode->i_op = &ext4_file_inode_operations;
2515 inode->i_fop = &ext4_file_operations;
2516 ext4_set_aops(inode);
2517 d_tmpfile(dentry, inode);
2518 err = ext4_orphan_add(handle, inode);
2520 goto err_unlock_inode;
2521 mark_inode_dirty(inode);
2522 unlock_new_inode(inode);
2525 ext4_journal_stop(handle);
2526 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2530 ext4_journal_stop(handle);
2531 unlock_new_inode(inode);
2535 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2536 struct ext4_dir_entry_2 *de,
2537 int blocksize, int csum_size,
2538 unsigned int parent_ino, int dotdot_real_len)
2540 de->inode = cpu_to_le32(inode->i_ino);
2542 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2544 strcpy(de->name, ".");
2545 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2547 de = ext4_next_entry(de, blocksize);
2548 de->inode = cpu_to_le32(parent_ino);
2550 if (!dotdot_real_len)
2551 de->rec_len = ext4_rec_len_to_disk(blocksize -
2552 (csum_size + EXT4_DIR_REC_LEN(1)),
2555 de->rec_len = ext4_rec_len_to_disk(
2556 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2557 strcpy(de->name, "..");
2558 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2560 return ext4_next_entry(de, blocksize);
2563 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2564 struct inode *inode)
2566 struct buffer_head *dir_block = NULL;
2567 struct ext4_dir_entry_2 *de;
2568 struct ext4_dir_entry_tail *t;
2569 ext4_lblk_t block = 0;
2570 unsigned int blocksize = dir->i_sb->s_blocksize;
2574 if (ext4_has_metadata_csum(dir->i_sb))
2575 csum_size = sizeof(struct ext4_dir_entry_tail);
2577 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2578 err = ext4_try_create_inline_dir(handle, dir, inode);
2579 if (err < 0 && err != -ENOSPC)
2586 dir_block = ext4_append(handle, inode, &block);
2587 if (IS_ERR(dir_block))
2588 return PTR_ERR(dir_block);
2589 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2590 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2591 set_nlink(inode, 2);
2593 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2594 initialize_dirent_tail(t, blocksize);
2597 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2598 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2601 set_buffer_verified(dir_block);
2607 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2610 struct inode *inode;
2611 int err, credits, retries = 0;
2613 if (EXT4_DIR_LINK_MAX(dir))
2616 err = dquot_initialize(dir);
2620 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2621 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2623 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2625 0, NULL, EXT4_HT_DIR, credits);
2626 handle = ext4_journal_current_handle();
2627 err = PTR_ERR(inode);
2631 inode->i_op = &ext4_dir_inode_operations;
2632 inode->i_fop = &ext4_dir_operations;
2633 err = ext4_init_new_dir(handle, dir, inode);
2635 goto out_clear_inode;
2636 err = ext4_mark_inode_dirty(handle, inode);
2638 err = ext4_add_entry(handle, dentry, inode);
2642 unlock_new_inode(inode);
2643 ext4_mark_inode_dirty(handle, inode);
2647 ext4_inc_count(handle, dir);
2648 ext4_update_dx_flag(dir);
2649 err = ext4_mark_inode_dirty(handle, dir);
2651 goto out_clear_inode;
2652 unlock_new_inode(inode);
2653 d_instantiate(dentry, inode);
2654 if (IS_DIRSYNC(dir))
2655 ext4_handle_sync(handle);
2659 ext4_journal_stop(handle);
2660 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2666 * routine to check that the specified directory is empty (for rmdir)
2668 int ext4_empty_dir(struct inode *inode)
2670 unsigned int offset;
2671 struct buffer_head *bh;
2672 struct ext4_dir_entry_2 *de, *de1;
2673 struct super_block *sb;
2676 if (ext4_has_inline_data(inode)) {
2677 int has_inline_data = 1;
2679 err = empty_inline_dir(inode, &has_inline_data);
2680 if (has_inline_data)
2685 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2686 EXT4_ERROR_INODE(inode, "invalid size");
2689 bh = ext4_read_dirblock(inode, 0, EITHER);
2693 de = (struct ext4_dir_entry_2 *) bh->b_data;
2694 de1 = ext4_next_entry(de, sb->s_blocksize);
2695 if (le32_to_cpu(de->inode) != inode->i_ino ||
2696 le32_to_cpu(de1->inode) == 0 ||
2697 strcmp(".", de->name) || strcmp("..", de1->name)) {
2698 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2702 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2703 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2704 de = ext4_next_entry(de1, sb->s_blocksize);
2705 while (offset < inode->i_size) {
2706 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2707 unsigned int lblock;
2710 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2711 bh = ext4_read_dirblock(inode, lblock, EITHER);
2714 de = (struct ext4_dir_entry_2 *) bh->b_data;
2716 if (ext4_check_dir_entry(inode, NULL, de, bh,
2717 bh->b_data, bh->b_size, offset)) {
2718 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2720 offset = (offset | (sb->s_blocksize - 1)) + 1;
2723 if (le32_to_cpu(de->inode)) {
2727 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2728 de = ext4_next_entry(de, sb->s_blocksize);
2735 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2736 * such inodes, starting at the superblock, in case we crash before the
2737 * file is closed/deleted, or in case the inode truncate spans multiple
2738 * transactions and the last transaction is not recovered after a crash.
2740 * At filesystem recovery time, we walk this list deleting unlinked
2741 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2743 * Orphan list manipulation functions must be called under i_mutex unless
2744 * we are just creating the inode or deleting it.
2746 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2748 struct super_block *sb = inode->i_sb;
2749 struct ext4_sb_info *sbi = EXT4_SB(sb);
2750 struct ext4_iloc iloc;
2754 if (!sbi->s_journal || is_bad_inode(inode))
2757 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2758 !mutex_is_locked(&inode->i_mutex));
2760 * Exit early if inode already is on orphan list. This is a big speedup
2761 * since we don't have to contend on the global s_orphan_lock.
2763 if (!list_empty(&EXT4_I(inode)->i_orphan))
2767 * Orphan handling is only valid for files with data blocks
2768 * being truncated, or files being unlinked. Note that we either
2769 * hold i_mutex, or the inode can not be referenced from outside,
2770 * so i_nlink should not be bumped due to race
2772 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2773 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2775 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2776 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2780 err = ext4_reserve_inode_write(handle, inode, &iloc);
2784 mutex_lock(&sbi->s_orphan_lock);
2786 * Due to previous errors inode may be already a part of on-disk
2787 * orphan list. If so skip on-disk list modification.
2789 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2790 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2791 /* Insert this inode at the head of the on-disk orphan list */
2792 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2793 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2796 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2797 mutex_unlock(&sbi->s_orphan_lock);
2800 err = ext4_handle_dirty_super(handle, sb);
2801 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2806 * We have to remove inode from in-memory list if
2807 * addition to on disk orphan list failed. Stray orphan
2808 * list entries can cause panics at unmount time.
2810 mutex_lock(&sbi->s_orphan_lock);
2811 list_del_init(&EXT4_I(inode)->i_orphan);
2812 mutex_unlock(&sbi->s_orphan_lock);
2815 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2816 jbd_debug(4, "orphan inode %lu will point to %d\n",
2817 inode->i_ino, NEXT_ORPHAN(inode));
2819 ext4_std_error(sb, err);
2824 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2825 * of such inodes stored on disk, because it is finally being cleaned up.
2827 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2829 struct list_head *prev;
2830 struct ext4_inode_info *ei = EXT4_I(inode);
2831 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2833 struct ext4_iloc iloc;
2836 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2839 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2840 !mutex_is_locked(&inode->i_mutex));
2841 /* Do this quick check before taking global s_orphan_lock. */
2842 if (list_empty(&ei->i_orphan))
2846 /* Grab inode buffer early before taking global s_orphan_lock */
2847 err = ext4_reserve_inode_write(handle, inode, &iloc);
2850 mutex_lock(&sbi->s_orphan_lock);
2851 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2853 prev = ei->i_orphan.prev;
2854 list_del_init(&ei->i_orphan);
2856 /* If we're on an error path, we may not have a valid
2857 * transaction handle with which to update the orphan list on
2858 * disk, but we still need to remove the inode from the linked
2859 * list in memory. */
2860 if (!handle || err) {
2861 mutex_unlock(&sbi->s_orphan_lock);
2865 ino_next = NEXT_ORPHAN(inode);
2866 if (prev == &sbi->s_orphan) {
2867 jbd_debug(4, "superblock will point to %u\n", ino_next);
2868 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2869 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2871 mutex_unlock(&sbi->s_orphan_lock);
2874 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2875 mutex_unlock(&sbi->s_orphan_lock);
2876 err = ext4_handle_dirty_super(handle, inode->i_sb);
2878 struct ext4_iloc iloc2;
2879 struct inode *i_prev =
2880 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2882 jbd_debug(4, "orphan inode %lu will point to %u\n",
2883 i_prev->i_ino, ino_next);
2884 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2886 mutex_unlock(&sbi->s_orphan_lock);
2889 NEXT_ORPHAN(i_prev) = ino_next;
2890 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2891 mutex_unlock(&sbi->s_orphan_lock);
2895 NEXT_ORPHAN(inode) = 0;
2896 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2898 ext4_std_error(inode->i_sb, err);
2906 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2909 struct inode *inode;
2910 struct buffer_head *bh;
2911 struct ext4_dir_entry_2 *de;
2912 handle_t *handle = NULL;
2914 /* Initialize quotas before so that eventual writes go in
2915 * separate transaction */
2916 retval = dquot_initialize(dir);
2919 retval = dquot_initialize(d_inode(dentry));
2924 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2930 inode = d_inode(dentry);
2932 retval = -EFSCORRUPTED;
2933 if (le32_to_cpu(de->inode) != inode->i_ino)
2936 retval = -ENOTEMPTY;
2937 if (!ext4_empty_dir(inode))
2940 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2941 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2942 if (IS_ERR(handle)) {
2943 retval = PTR_ERR(handle);
2948 if (IS_DIRSYNC(dir))
2949 ext4_handle_sync(handle);
2951 retval = ext4_delete_entry(handle, dir, de, bh);
2954 if (!EXT4_DIR_LINK_EMPTY(inode))
2955 ext4_warning_inode(inode,
2956 "empty directory '%.*s' has too many links (%u)",
2957 dentry->d_name.len, dentry->d_name.name,
2961 /* There's no need to set i_disksize: the fact that i_nlink is
2962 * zero will ensure that the right thing happens during any
2965 ext4_orphan_add(handle, inode);
2966 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2967 ext4_mark_inode_dirty(handle, inode);
2968 ext4_dec_count(handle, dir);
2969 ext4_update_dx_flag(dir);
2970 ext4_mark_inode_dirty(handle, dir);
2975 ext4_journal_stop(handle);
2979 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2982 struct inode *inode;
2983 struct buffer_head *bh;
2984 struct ext4_dir_entry_2 *de;
2985 handle_t *handle = NULL;
2987 trace_ext4_unlink_enter(dir, dentry);
2988 /* Initialize quotas before so that eventual writes go
2989 * in separate transaction */
2990 retval = dquot_initialize(dir);
2993 retval = dquot_initialize(d_inode(dentry));
2998 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3004 inode = d_inode(dentry);
3006 retval = -EFSCORRUPTED;
3007 if (le32_to_cpu(de->inode) != inode->i_ino)
3010 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3011 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3012 if (IS_ERR(handle)) {
3013 retval = PTR_ERR(handle);
3018 if (IS_DIRSYNC(dir))
3019 ext4_handle_sync(handle);
3021 if (inode->i_nlink == 0) {
3022 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3023 dentry->d_name.len, dentry->d_name.name);
3024 set_nlink(inode, 1);
3026 retval = ext4_delete_entry(handle, dir, de, bh);
3029 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3030 ext4_update_dx_flag(dir);
3031 ext4_mark_inode_dirty(handle, dir);
3033 if (!inode->i_nlink)
3034 ext4_orphan_add(handle, inode);
3035 inode->i_ctime = ext4_current_time(inode);
3036 ext4_mark_inode_dirty(handle, inode);
3041 ext4_journal_stop(handle);
3042 trace_ext4_unlink_exit(dentry, retval);
3046 static int ext4_symlink(struct inode *dir,
3047 struct dentry *dentry, const char *symname)
3050 struct inode *inode;
3051 int err, len = strlen(symname);
3053 bool encryption_required;
3054 struct ext4_str disk_link;
3055 struct ext4_encrypted_symlink_data *sd = NULL;
3057 disk_link.len = len + 1;
3058 disk_link.name = (char *) symname;
3060 encryption_required = (ext4_encrypted_inode(dir) ||
3061 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3062 if (encryption_required) {
3063 err = ext4_get_encryption_info(dir);
3066 if (ext4_encryption_info(dir) == NULL)
3068 disk_link.len = (ext4_fname_encrypted_size(dir, len) +
3069 sizeof(struct ext4_encrypted_symlink_data));
3070 sd = kzalloc(disk_link.len, GFP_KERNEL);
3075 if (disk_link.len > dir->i_sb->s_blocksize) {
3076 err = -ENAMETOOLONG;
3080 err = dquot_initialize(dir);
3084 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3086 * For non-fast symlinks, we just allocate inode and put it on
3087 * orphan list in the first transaction => we need bitmap,
3088 * group descriptor, sb, inode block, quota blocks, and
3089 * possibly selinux xattr blocks.
3091 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3092 EXT4_XATTR_TRANS_BLOCKS;
3095 * Fast symlink. We have to add entry to directory
3096 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3097 * allocate new inode (bitmap, group descriptor, inode block,
3098 * quota blocks, sb is already counted in previous macros).
3100 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3101 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3104 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3105 &dentry->d_name, 0, NULL,
3106 EXT4_HT_DIR, credits);
3107 handle = ext4_journal_current_handle();
3108 if (IS_ERR(inode)) {
3110 ext4_journal_stop(handle);
3111 err = PTR_ERR(inode);
3115 if (encryption_required) {
3117 struct ext4_str ostr;
3119 istr.name = (const unsigned char *) symname;
3121 ostr.name = sd->encrypted_path;
3122 ostr.len = disk_link.len;
3123 err = ext4_fname_usr_to_disk(inode, &istr, &ostr);
3125 goto err_drop_inode;
3126 sd->len = cpu_to_le16(ostr.len);
3127 disk_link.name = (char *) sd;
3128 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3131 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3132 if (!encryption_required)
3133 inode->i_op = &ext4_symlink_inode_operations;
3134 ext4_set_aops(inode);
3136 * We cannot call page_symlink() with transaction started
3137 * because it calls into ext4_write_begin() which can wait
3138 * for transaction commit if we are running out of space
3139 * and thus we deadlock. So we have to stop transaction now
3140 * and restart it when symlink contents is written.
3142 * To keep fs consistent in case of crash, we have to put inode
3143 * to orphan list in the mean time.
3146 err = ext4_orphan_add(handle, inode);
3147 ext4_journal_stop(handle);
3150 goto err_drop_inode;
3151 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3153 goto err_drop_inode;
3155 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3156 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3158 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3159 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3160 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3161 if (IS_ERR(handle)) {
3162 err = PTR_ERR(handle);
3164 goto err_drop_inode;
3166 set_nlink(inode, 1);
3167 err = ext4_orphan_del(handle, inode);
3169 goto err_drop_inode;
3171 /* clear the extent format for fast symlink */
3172 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3173 if (!encryption_required) {
3174 inode->i_op = &ext4_fast_symlink_inode_operations;
3175 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3177 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3179 inode->i_size = disk_link.len - 1;
3181 EXT4_I(inode)->i_disksize = inode->i_size;
3182 err = ext4_add_nondir(handle, dentry, inode);
3183 if (!err && IS_DIRSYNC(dir))
3184 ext4_handle_sync(handle);
3187 ext4_journal_stop(handle);
3192 ext4_journal_stop(handle);
3194 unlock_new_inode(inode);
3201 static int ext4_link(struct dentry *old_dentry,
3202 struct inode *dir, struct dentry *dentry)
3205 struct inode *inode = d_inode(old_dentry);
3206 int err, retries = 0;
3208 if (inode->i_nlink >= EXT4_LINK_MAX)
3210 if (ext4_encrypted_inode(dir) &&
3211 !ext4_is_child_context_consistent_with_parent(dir, inode))
3213 err = dquot_initialize(dir);
3218 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3219 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3220 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3222 return PTR_ERR(handle);
3224 if (IS_DIRSYNC(dir))
3225 ext4_handle_sync(handle);
3227 inode->i_ctime = ext4_current_time(inode);
3228 ext4_inc_count(handle, inode);
3231 err = ext4_add_entry(handle, dentry, inode);
3233 ext4_mark_inode_dirty(handle, inode);
3234 /* this can happen only for tmpfile being
3235 * linked the first time
3237 if (inode->i_nlink == 1)
3238 ext4_orphan_del(handle, inode);
3239 d_instantiate(dentry, inode);
3244 ext4_journal_stop(handle);
3245 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3252 * Try to find buffer head where contains the parent block.
3253 * It should be the inode block if it is inlined or the 1st block
3254 * if it is a normal dir.
3256 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3257 struct inode *inode,
3259 struct ext4_dir_entry_2 **parent_de,
3262 struct buffer_head *bh;
3264 if (!ext4_has_inline_data(inode)) {
3265 bh = ext4_read_dirblock(inode, 0, EITHER);
3267 *retval = PTR_ERR(bh);
3270 *parent_de = ext4_next_entry(
3271 (struct ext4_dir_entry_2 *)bh->b_data,
3272 inode->i_sb->s_blocksize);
3277 return ext4_get_first_inline_block(inode, parent_de, retval);
3280 struct ext4_renament {
3282 struct dentry *dentry;
3283 struct inode *inode;
3285 int dir_nlink_delta;
3287 /* entry for "dentry" */
3288 struct buffer_head *bh;
3289 struct ext4_dir_entry_2 *de;
3292 /* entry for ".." in inode if it's a directory */
3293 struct buffer_head *dir_bh;
3294 struct ext4_dir_entry_2 *parent_de;
3298 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3302 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3303 &retval, &ent->parent_de,
3307 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3308 return -EFSCORRUPTED;
3309 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3310 return ext4_journal_get_write_access(handle, ent->dir_bh);
3313 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3318 ent->parent_de->inode = cpu_to_le32(dir_ino);
3319 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3320 if (!ent->dir_inlined) {
3321 if (is_dx(ent->inode)) {
3322 retval = ext4_handle_dirty_dx_node(handle,
3326 retval = ext4_handle_dirty_dirent_node(handle,
3331 retval = ext4_mark_inode_dirty(handle, ent->inode);
3334 ext4_std_error(ent->dir->i_sb, retval);
3340 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3341 unsigned ino, unsigned file_type)
3345 BUFFER_TRACE(ent->bh, "get write access");
3346 retval = ext4_journal_get_write_access(handle, ent->bh);
3349 ent->de->inode = cpu_to_le32(ino);
3350 if (ext4_has_feature_filetype(ent->dir->i_sb))
3351 ent->de->file_type = file_type;
3352 ent->dir->i_version++;
3353 ent->dir->i_ctime = ent->dir->i_mtime =
3354 ext4_current_time(ent->dir);
3355 ext4_mark_inode_dirty(handle, ent->dir);
3356 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3357 if (!ent->inlined) {
3358 retval = ext4_handle_dirty_dirent_node(handle,
3360 if (unlikely(retval)) {
3361 ext4_std_error(ent->dir->i_sb, retval);
3371 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3372 const struct qstr *d_name)
3374 int retval = -ENOENT;
3375 struct buffer_head *bh;
3376 struct ext4_dir_entry_2 *de;
3378 bh = ext4_find_entry(dir, d_name, &de, NULL);
3382 retval = ext4_delete_entry(handle, dir, de, bh);
3388 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3393 * ent->de could have moved from under us during htree split, so make
3394 * sure that we are deleting the right entry. We might also be pointing
3395 * to a stale entry in the unused part of ent->bh so just checking inum
3396 * and the name isn't enough.
3398 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3399 ent->de->name_len != ent->dentry->d_name.len ||
3400 strncmp(ent->de->name, ent->dentry->d_name.name,
3401 ent->de->name_len) ||
3403 retval = ext4_find_delete_entry(handle, ent->dir,
3404 &ent->dentry->d_name);
3406 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3407 if (retval == -ENOENT) {
3408 retval = ext4_find_delete_entry(handle, ent->dir,
3409 &ent->dentry->d_name);
3414 ext4_warning_inode(ent->dir,
3415 "Deleting old file: nlink %d, error=%d",
3416 ent->dir->i_nlink, retval);
3420 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3422 if (ent->dir_nlink_delta) {
3423 if (ent->dir_nlink_delta == -1)
3424 ext4_dec_count(handle, ent->dir);
3426 ext4_inc_count(handle, ent->dir);
3427 ext4_mark_inode_dirty(handle, ent->dir);
3431 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3432 int credits, handle_t **h)
3439 * for inode block, sb block, group summaries,
3442 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3443 EXT4_XATTR_TRANS_BLOCKS + 4);
3445 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3446 &ent->dentry->d_name, 0, NULL,
3447 EXT4_HT_DIR, credits);
3449 handle = ext4_journal_current_handle();
3452 ext4_journal_stop(handle);
3453 if (PTR_ERR(wh) == -ENOSPC &&
3454 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3458 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3459 wh->i_op = &ext4_special_inode_operations;
3465 * Anybody can rename anything with this: the permission checks are left to the
3466 * higher-level routines.
3468 * n.b. old_{dentry,inode) refers to the source dentry/inode
3469 * while new_{dentry,inode) refers to the destination dentry/inode
3470 * This comes from rename(const char *oldpath, const char *newpath)
3472 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3473 struct inode *new_dir, struct dentry *new_dentry,
3476 handle_t *handle = NULL;
3477 struct ext4_renament old = {
3479 .dentry = old_dentry,
3480 .inode = d_inode(old_dentry),
3482 struct ext4_renament new = {
3484 .dentry = new_dentry,
3485 .inode = d_inode(new_dentry),
3489 struct inode *whiteout = NULL;
3493 retval = dquot_initialize(old.dir);
3496 retval = dquot_initialize(new.dir);
3500 /* Initialize quotas before so that eventual writes go
3501 * in separate transaction */
3503 retval = dquot_initialize(new.inode);
3508 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3510 return PTR_ERR(old.bh);
3512 * Check for inode number is _not_ due to possible IO errors.
3513 * We might rmdir the source, keep it as pwd of some process
3514 * and merrily kill the link to whatever was created under the
3515 * same name. Goodbye sticky bit ;-<
3518 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3521 if ((old.dir != new.dir) &&
3522 ext4_encrypted_inode(new.dir) &&
3523 !ext4_is_child_context_consistent_with_parent(new.dir,
3529 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3530 &new.de, &new.inlined);
3531 if (IS_ERR(new.bh)) {
3532 retval = PTR_ERR(new.bh);
3542 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3543 ext4_alloc_da_blocks(old.inode);
3545 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3546 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3547 if (!(flags & RENAME_WHITEOUT)) {
3548 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3549 if (IS_ERR(handle)) {
3550 retval = PTR_ERR(handle);
3555 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3556 if (IS_ERR(whiteout)) {
3557 retval = PTR_ERR(whiteout);
3563 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3564 ext4_handle_sync(handle);
3566 if (S_ISDIR(old.inode->i_mode)) {
3568 retval = -ENOTEMPTY;
3569 if (!ext4_empty_dir(new.inode))
3573 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3576 retval = ext4_rename_dir_prepare(handle, &old);
3581 * If we're renaming a file within an inline_data dir and adding or
3582 * setting the new dirent causes a conversion from inline_data to
3583 * extents/blockmap, we need to force the dirent delete code to
3584 * re-read the directory, or else we end up trying to delete a dirent
3585 * from what is now the extent tree root (or a block map).
3587 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3588 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3590 old_file_type = old.de->file_type;
3593 * Do this before adding a new entry, so the old entry is sure
3594 * to be still pointing to the valid old entry.
3596 retval = ext4_setent(handle, &old, whiteout->i_ino,
3600 ext4_mark_inode_dirty(handle, whiteout);
3603 retval = ext4_add_entry(handle, new.dentry, old.inode);
3607 retval = ext4_setent(handle, &new,
3608 old.inode->i_ino, old_file_type);
3613 force_reread = !ext4_test_inode_flag(new.dir,
3614 EXT4_INODE_INLINE_DATA);
3617 * Like most other Unix systems, set the ctime for inodes on a
3620 old.inode->i_ctime = ext4_current_time(old.inode);
3621 ext4_mark_inode_dirty(handle, old.inode);
3627 ext4_rename_delete(handle, &old, force_reread);
3631 ext4_dec_count(handle, new.inode);
3632 new.inode->i_ctime = ext4_current_time(new.inode);
3634 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3635 ext4_update_dx_flag(old.dir);
3637 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3641 ext4_dec_count(handle, old.dir);
3643 /* checked ext4_empty_dir above, can't have another
3644 * parent, ext4_dec_count() won't work for many-linked
3646 clear_nlink(new.inode);
3648 ext4_inc_count(handle, new.dir);
3649 ext4_update_dx_flag(new.dir);
3650 ext4_mark_inode_dirty(handle, new.dir);
3653 ext4_mark_inode_dirty(handle, old.dir);
3655 ext4_mark_inode_dirty(handle, new.inode);
3656 if (!new.inode->i_nlink)
3657 ext4_orphan_add(handle, new.inode);
3667 drop_nlink(whiteout);
3668 unlock_new_inode(whiteout);
3672 ext4_journal_stop(handle);
3676 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3677 struct inode *new_dir, struct dentry *new_dentry)
3679 handle_t *handle = NULL;
3680 struct ext4_renament old = {
3682 .dentry = old_dentry,
3683 .inode = d_inode(old_dentry),
3685 struct ext4_renament new = {
3687 .dentry = new_dentry,
3688 .inode = d_inode(new_dentry),
3693 if ((ext4_encrypted_inode(old_dir) ||
3694 ext4_encrypted_inode(new_dir)) &&
3695 (old_dir != new_dir) &&
3696 (!ext4_is_child_context_consistent_with_parent(new_dir,
3698 !ext4_is_child_context_consistent_with_parent(old_dir,
3702 retval = dquot_initialize(old.dir);
3705 retval = dquot_initialize(new.dir);
3709 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3710 &old.de, &old.inlined);
3712 return PTR_ERR(old.bh);
3714 * Check for inode number is _not_ due to possible IO errors.
3715 * We might rmdir the source, keep it as pwd of some process
3716 * and merrily kill the link to whatever was created under the
3717 * same name. Goodbye sticky bit ;-<
3720 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3723 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3724 &new.de, &new.inlined);
3725 if (IS_ERR(new.bh)) {
3726 retval = PTR_ERR(new.bh);
3731 /* RENAME_EXCHANGE case: old *and* new must both exist */
3732 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3735 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3736 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3737 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3738 if (IS_ERR(handle)) {
3739 retval = PTR_ERR(handle);
3744 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3745 ext4_handle_sync(handle);
3747 if (S_ISDIR(old.inode->i_mode)) {
3749 retval = ext4_rename_dir_prepare(handle, &old);
3753 if (S_ISDIR(new.inode->i_mode)) {
3755 retval = ext4_rename_dir_prepare(handle, &new);
3761 * Other than the special case of overwriting a directory, parents'
3762 * nlink only needs to be modified if this is a cross directory rename.
3764 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3765 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3766 new.dir_nlink_delta = -old.dir_nlink_delta;
3768 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3769 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3773 new_file_type = new.de->file_type;
3774 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3778 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3783 * Like most other Unix systems, set the ctime for inodes on a
3786 old.inode->i_ctime = ext4_current_time(old.inode);
3787 new.inode->i_ctime = ext4_current_time(new.inode);
3788 ext4_mark_inode_dirty(handle, old.inode);
3789 ext4_mark_inode_dirty(handle, new.inode);
3792 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3797 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3801 ext4_update_dir_count(handle, &old);
3802 ext4_update_dir_count(handle, &new);
3811 ext4_journal_stop(handle);
3815 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3816 struct inode *new_dir, struct dentry *new_dentry,
3819 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3822 if (flags & RENAME_EXCHANGE) {
3823 return ext4_cross_rename(old_dir, old_dentry,
3824 new_dir, new_dentry);
3827 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3831 * directories can handle most operations...
3833 const struct inode_operations ext4_dir_inode_operations = {
3834 .create = ext4_create,
3835 .lookup = ext4_lookup,
3837 .unlink = ext4_unlink,
3838 .symlink = ext4_symlink,
3839 .mkdir = ext4_mkdir,
3840 .rmdir = ext4_rmdir,
3841 .mknod = ext4_mknod,
3842 .tmpfile = ext4_tmpfile,
3843 .rename2 = ext4_rename2,
3844 .setattr = ext4_setattr,
3845 .setxattr = generic_setxattr,
3846 .getxattr = generic_getxattr,
3847 .listxattr = ext4_listxattr,
3848 .removexattr = generic_removexattr,
3849 .get_acl = ext4_get_acl,
3850 .set_acl = ext4_set_acl,
3851 .fiemap = ext4_fiemap,
3854 const struct inode_operations ext4_special_inode_operations = {
3855 .setattr = ext4_setattr,
3856 .setxattr = generic_setxattr,
3857 .getxattr = generic_getxattr,
3858 .listxattr = ext4_listxattr,
3859 .removexattr = generic_removexattr,
3860 .get_acl = ext4_get_acl,
3861 .set_acl = ext4_set_acl,