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 inode *dir, struct ext4_dir_entry_2 *de,
257 unsigned blocksize, struct dx_hash_info *hinfo,
258 struct dx_map_entry map[]);
259 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
260 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
261 struct dx_map_entry *offsets, int count, unsigned blocksize);
262 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
263 static void dx_insert_block(struct dx_frame *frame,
264 u32 hash, ext4_lblk_t block);
265 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
266 struct dx_frame *frame,
267 struct dx_frame *frames,
269 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
270 const struct qstr *d_name,
271 struct ext4_dir_entry_2 **res_dir);
272 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
273 struct inode *inode);
275 /* checksumming functions */
276 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
277 unsigned int blocksize)
279 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
280 t->det_rec_len = ext4_rec_len_to_disk(
281 sizeof(struct ext4_dir_entry_tail), blocksize);
282 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
285 /* Walk through a dirent block to find a checksum "dirent" at the tail */
286 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
287 struct ext4_dir_entry *de)
289 struct ext4_dir_entry_tail *t;
292 struct ext4_dir_entry *d, *top;
295 top = (struct ext4_dir_entry *)(((void *)de) +
296 (EXT4_BLOCK_SIZE(inode->i_sb) -
297 sizeof(struct ext4_dir_entry_tail)));
298 while (d < top && d->rec_len)
299 d = (struct ext4_dir_entry *)(((void *)d) +
300 le16_to_cpu(d->rec_len));
305 t = (struct ext4_dir_entry_tail *)d;
307 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
310 if (t->det_reserved_zero1 ||
311 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
312 t->det_reserved_zero2 ||
313 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
319 static __le32 ext4_dirent_csum(struct inode *inode,
320 struct ext4_dir_entry *dirent, int size)
322 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
323 struct ext4_inode_info *ei = EXT4_I(inode);
326 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
327 return cpu_to_le32(csum);
330 static void warn_no_space_for_csum(struct inode *inode)
332 ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
333 "checksum. Please run e2fsck -D.", inode->i_ino);
336 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
338 struct ext4_dir_entry_tail *t;
340 if (!ext4_has_metadata_csum(inode->i_sb))
343 t = get_dirent_tail(inode, dirent);
345 warn_no_space_for_csum(inode);
349 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
350 (void *)t - (void *)dirent))
356 static void ext4_dirent_csum_set(struct inode *inode,
357 struct ext4_dir_entry *dirent)
359 struct ext4_dir_entry_tail *t;
361 if (!ext4_has_metadata_csum(inode->i_sb))
364 t = get_dirent_tail(inode, dirent);
366 warn_no_space_for_csum(inode);
370 t->det_checksum = ext4_dirent_csum(inode, dirent,
371 (void *)t - (void *)dirent);
374 int ext4_handle_dirty_dirent_node(handle_t *handle,
376 struct buffer_head *bh)
378 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
379 return ext4_handle_dirty_metadata(handle, inode, bh);
382 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
383 struct ext4_dir_entry *dirent,
386 struct ext4_dir_entry *dp;
387 struct dx_root_info *root;
390 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
392 else if (le16_to_cpu(dirent->rec_len) == 12) {
393 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
394 if (le16_to_cpu(dp->rec_len) !=
395 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
397 root = (struct dx_root_info *)(((void *)dp + 12));
398 if (root->reserved_zero ||
399 root->info_length != sizeof(struct dx_root_info))
406 *offset = count_offset;
407 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
410 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
411 int count_offset, int count, struct dx_tail *t)
413 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
414 struct ext4_inode_info *ei = EXT4_I(inode);
419 size = count_offset + (count * sizeof(struct dx_entry));
420 save_csum = t->dt_checksum;
422 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
423 csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
424 t->dt_checksum = save_csum;
426 return cpu_to_le32(csum);
429 static int ext4_dx_csum_verify(struct inode *inode,
430 struct ext4_dir_entry *dirent)
432 struct dx_countlimit *c;
434 int count_offset, limit, count;
436 if (!ext4_has_metadata_csum(inode->i_sb))
439 c = get_dx_countlimit(inode, dirent, &count_offset);
441 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
444 limit = le16_to_cpu(c->limit);
445 count = le16_to_cpu(c->count);
446 if (count_offset + (limit * sizeof(struct dx_entry)) >
447 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
448 warn_no_space_for_csum(inode);
451 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
453 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
459 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
461 struct dx_countlimit *c;
463 int count_offset, limit, count;
465 if (!ext4_has_metadata_csum(inode->i_sb))
468 c = get_dx_countlimit(inode, dirent, &count_offset);
470 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
473 limit = le16_to_cpu(c->limit);
474 count = le16_to_cpu(c->count);
475 if (count_offset + (limit * sizeof(struct dx_entry)) >
476 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
477 warn_no_space_for_csum(inode);
480 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
482 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
485 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
487 struct buffer_head *bh)
489 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
490 return ext4_handle_dirty_metadata(handle, inode, bh);
494 * p is at least 6 bytes before the end of page
496 static inline struct ext4_dir_entry_2 *
497 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
499 return (struct ext4_dir_entry_2 *)((char *)p +
500 ext4_rec_len_from_disk(p->rec_len, blocksize));
504 * Future: use high four bits of block for coalesce-on-delete flags
505 * Mask them off for now.
508 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
510 return le32_to_cpu(entry->block) & 0x00ffffff;
513 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
515 entry->block = cpu_to_le32(value);
518 static inline unsigned dx_get_hash(struct dx_entry *entry)
520 return le32_to_cpu(entry->hash);
523 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
525 entry->hash = cpu_to_le32(value);
528 static inline unsigned dx_get_count(struct dx_entry *entries)
530 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
533 static inline unsigned dx_get_limit(struct dx_entry *entries)
535 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
538 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
540 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
543 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
545 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
548 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
550 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
551 EXT4_DIR_REC_LEN(2) - infosize;
553 if (ext4_has_metadata_csum(dir->i_sb))
554 entry_space -= sizeof(struct dx_tail);
555 return entry_space / sizeof(struct dx_entry);
558 static inline unsigned dx_node_limit(struct inode *dir)
560 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
562 if (ext4_has_metadata_csum(dir->i_sb))
563 entry_space -= sizeof(struct dx_tail);
564 return entry_space / sizeof(struct dx_entry);
571 static void dx_show_index(char * label, struct dx_entry *entries)
573 int i, n = dx_get_count (entries);
574 printk(KERN_DEBUG "%s index ", label);
575 for (i = 0; i < n; i++) {
576 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
577 0, (unsigned long)dx_get_block(entries + i));
589 static struct stats dx_show_leaf(struct inode *dir,
590 struct dx_hash_info *hinfo,
591 struct ext4_dir_entry_2 *de,
592 int size, int show_names)
594 unsigned names = 0, space = 0;
595 char *base = (char *) de;
596 struct dx_hash_info h = *hinfo;
599 while ((char *) de < base + size)
605 #ifdef CONFIG_EXT4_FS_ENCRYPTION
608 struct ext4_str fname_crypto_str
609 = {.name = NULL, .len = 0};
610 struct ext4_fname_crypto_ctx *ctx = NULL;
615 ctx = ext4_get_fname_crypto_ctx(dir,
618 printk(KERN_WARNING "Error acquiring"
619 " crypto ctxt--skipping crypto\n");
623 /* Directory is not encrypted */
624 ext4fs_dirhash(de->name,
626 printk("%*.s:(U)%x.%u ", len,
628 (unsigned) ((char *) de
631 /* Directory is encrypted */
632 res = ext4_fname_crypto_alloc_buffer(
636 printk(KERN_WARNING "Error "
640 ext4_put_fname_crypto_ctx(&ctx);
643 res = ext4_fname_disk_to_usr(ctx, NULL, de,
646 printk(KERN_WARNING "Error "
647 "converting filename "
653 name = fname_crypto_str.name;
654 len = fname_crypto_str.len;
656 ext4fs_dirhash(de->name, de->name_len,
658 printk("%*.s:(E)%x.%u ", len, name,
659 h.hash, (unsigned) ((char *) de
661 ext4_put_fname_crypto_ctx(&ctx);
662 ext4_fname_crypto_free_buffer(
666 int len = de->name_len;
667 char *name = de->name;
668 ext4fs_dirhash(de->name, de->name_len, &h);
669 printk("%*.s:%x.%u ", len, name, h.hash,
670 (unsigned) ((char *) de - base));
673 space += EXT4_DIR_REC_LEN(de->name_len);
676 de = ext4_next_entry(de, size);
678 printk("(%i)\n", names);
679 return (struct stats) { names, space, 1 };
682 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
683 struct dx_entry *entries, int levels)
685 unsigned blocksize = dir->i_sb->s_blocksize;
686 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
688 struct buffer_head *bh;
689 printk("%i indexed blocks...\n", count);
690 for (i = 0; i < count; i++, entries++)
692 ext4_lblk_t block = dx_get_block(entries);
693 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
694 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
696 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
697 bh = ext4_bread(NULL,dir, block, 0);
698 if (!bh || IS_ERR(bh))
701 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
702 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
703 bh->b_data, blocksize, 0);
704 names += stats.names;
705 space += stats.space;
706 bcount += stats.bcount;
710 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
711 levels ? "" : " ", names, space/bcount,
712 (space/bcount)*100/blocksize);
713 return (struct stats) { names, space, bcount};
715 #endif /* DX_DEBUG */
718 * Probe for a directory leaf block to search.
720 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
721 * error in the directory index, and the caller should fall back to
722 * searching the directory normally. The callers of dx_probe **MUST**
723 * check for this error code, and make sure it never gets reflected
726 static struct dx_frame *
727 dx_probe(const struct qstr *d_name, struct inode *dir,
728 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
730 unsigned count, indirect;
731 struct dx_entry *at, *entries, *p, *q, *m;
732 struct dx_root *root;
733 struct dx_frame *frame = frame_in;
734 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
737 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
738 if (IS_ERR(frame->bh))
739 return (struct dx_frame *) frame->bh;
741 root = (struct dx_root *) frame->bh->b_data;
742 if (root->info.hash_version != DX_HASH_TEA &&
743 root->info.hash_version != DX_HASH_HALF_MD4 &&
744 root->info.hash_version != DX_HASH_LEGACY) {
745 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
746 root->info.hash_version);
749 hinfo->hash_version = root->info.hash_version;
750 if (hinfo->hash_version <= DX_HASH_TEA)
751 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
752 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
753 #ifdef CONFIG_EXT4_FS_ENCRYPTION
755 struct ext4_fname_crypto_ctx *ctx = NULL;
758 /* Check if the directory is encrypted */
759 ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
761 ret_err = ERR_PTR(PTR_ERR(ctx));
764 res = ext4_fname_usr_to_hash(ctx, d_name, hinfo);
766 ret_err = ERR_PTR(res);
769 ext4_put_fname_crypto_ctx(&ctx);
773 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
777 if (root->info.unused_flags & 1) {
778 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
779 root->info.unused_flags);
783 if ((indirect = root->info.indirect_levels) > 1) {
784 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
785 root->info.indirect_levels);
789 entries = (struct dx_entry *) (((char *)&root->info) +
790 root->info.info_length);
792 if (dx_get_limit(entries) != dx_root_limit(dir,
793 root->info.info_length)) {
794 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
798 dxtrace(printk("Look up %x", hash));
800 count = dx_get_count(entries);
801 if (!count || count > dx_get_limit(entries)) {
802 ext4_warning(dir->i_sb,
803 "dx entry: no count or count > limit");
808 q = entries + count - 1;
811 dxtrace(printk("."));
812 if (dx_get_hash(m) > hash)
818 if (0) { // linear search cross check
819 unsigned n = count - 1;
823 dxtrace(printk(","));
824 if (dx_get_hash(++at) > hash)
830 assert (at == p - 1);
834 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
835 frame->entries = entries;
840 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
841 if (IS_ERR(frame->bh)) {
842 ret_err = (struct dx_frame *) frame->bh;
846 entries = ((struct dx_node *) frame->bh->b_data)->entries;
848 if (dx_get_limit(entries) != dx_node_limit (dir)) {
849 ext4_warning(dir->i_sb,
850 "dx entry: limit != node limit");
855 while (frame >= frame_in) {
860 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
861 ext4_warning(dir->i_sb,
862 "Corrupt dir inode %lu, running e2fsck is "
863 "recommended.", dir->i_ino);
867 static void dx_release (struct dx_frame *frames)
869 if (frames[0].bh == NULL)
872 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
873 brelse(frames[1].bh);
874 brelse(frames[0].bh);
878 * This function increments the frame pointer to search the next leaf
879 * block, and reads in the necessary intervening nodes if the search
880 * should be necessary. Whether or not the search is necessary is
881 * controlled by the hash parameter. If the hash value is even, then
882 * the search is only continued if the next block starts with that
883 * hash value. This is used if we are searching for a specific file.
885 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
887 * This function returns 1 if the caller should continue to search,
888 * or 0 if it should not. If there is an error reading one of the
889 * index blocks, it will a negative error code.
891 * If start_hash is non-null, it will be filled in with the starting
892 * hash of the next page.
894 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
895 struct dx_frame *frame,
896 struct dx_frame *frames,
900 struct buffer_head *bh;
906 * Find the next leaf page by incrementing the frame pointer.
907 * If we run out of entries in the interior node, loop around and
908 * increment pointer in the parent node. When we break out of
909 * this loop, num_frames indicates the number of interior
910 * nodes need to be read.
913 if (++(p->at) < p->entries + dx_get_count(p->entries))
922 * If the hash is 1, then continue only if the next page has a
923 * continuation hash of any value. This is used for readdir
924 * handling. Otherwise, check to see if the hash matches the
925 * desired contiuation hash. If it doesn't, return since
926 * there's no point to read in the successive index pages.
928 bhash = dx_get_hash(p->at);
931 if ((hash & 1) == 0) {
932 if ((bhash & ~1) != hash)
936 * If the hash is HASH_NB_ALWAYS, we always go to the next
937 * block so no check is necessary
939 while (num_frames--) {
940 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
946 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
953 * This function fills a red-black tree with information from a
954 * directory block. It returns the number directory entries loaded
955 * into the tree. If there is an error it is returned in err.
957 static int htree_dirblock_to_tree(struct file *dir_file,
958 struct inode *dir, ext4_lblk_t block,
959 struct dx_hash_info *hinfo,
960 __u32 start_hash, __u32 start_minor_hash)
962 struct buffer_head *bh;
963 struct ext4_dir_entry_2 *de, *top;
964 int err = 0, count = 0;
965 struct ext4_fname_crypto_ctx *ctx = NULL;
966 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}, tmp_str;
968 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
969 (unsigned long)block));
970 bh = ext4_read_dirblock(dir, block, DIRENT);
974 de = (struct ext4_dir_entry_2 *) bh->b_data;
975 top = (struct ext4_dir_entry_2 *) ((char *) de +
976 dir->i_sb->s_blocksize -
977 EXT4_DIR_REC_LEN(0));
978 #ifdef CONFIG_EXT4_FS_ENCRYPTION
979 /* Check if the directory is encrypted */
980 ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
987 err = ext4_fname_crypto_alloc_buffer(ctx, EXT4_NAME_LEN,
990 ext4_put_fname_crypto_ctx(&ctx);
996 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
997 if (ext4_check_dir_entry(dir, NULL, de, bh,
998 bh->b_data, bh->b_size,
999 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1000 + ((char *)de - bh->b_data))) {
1001 /* silently ignore the rest of the block */
1004 ext4fs_dirhash(de->name, de->name_len, hinfo);
1005 if ((hinfo->hash < start_hash) ||
1006 ((hinfo->hash == start_hash) &&
1007 (hinfo->minor_hash < start_minor_hash)))
1012 /* Directory is not encrypted */
1013 tmp_str.name = de->name;
1014 tmp_str.len = de->name_len;
1015 err = ext4_htree_store_dirent(dir_file,
1016 hinfo->hash, hinfo->minor_hash, de,
1019 /* Directory is encrypted */
1020 err = ext4_fname_disk_to_usr(ctx, hinfo, de,
1026 err = ext4_htree_store_dirent(dir_file,
1027 hinfo->hash, hinfo->minor_hash, de,
1038 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1039 ext4_put_fname_crypto_ctx(&ctx);
1040 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1047 * This function fills a red-black tree with information from a
1048 * directory. We start scanning the directory in hash order, starting
1049 * at start_hash and start_minor_hash.
1051 * This function returns the number of entries inserted into the tree,
1052 * or a negative error code.
1054 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1055 __u32 start_minor_hash, __u32 *next_hash)
1057 struct dx_hash_info hinfo;
1058 struct ext4_dir_entry_2 *de;
1059 struct dx_frame frames[2], *frame;
1065 struct ext4_str tmp_str;
1067 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1068 start_hash, start_minor_hash));
1069 dir = file_inode(dir_file);
1070 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1071 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1072 if (hinfo.hash_version <= DX_HASH_TEA)
1073 hinfo.hash_version +=
1074 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1075 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1076 if (ext4_has_inline_data(dir)) {
1077 int has_inline_data = 1;
1078 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1082 if (has_inline_data) {
1087 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1088 start_hash, start_minor_hash);
1092 hinfo.hash = start_hash;
1093 hinfo.minor_hash = 0;
1094 frame = dx_probe(NULL, dir, &hinfo, frames);
1096 return PTR_ERR(frame);
1098 /* Add '.' and '..' from the htree header */
1099 if (!start_hash && !start_minor_hash) {
1100 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1101 tmp_str.name = de->name;
1102 tmp_str.len = de->name_len;
1103 err = ext4_htree_store_dirent(dir_file, 0, 0,
1109 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1110 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1111 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1112 tmp_str.name = de->name;
1113 tmp_str.len = de->name_len;
1114 err = ext4_htree_store_dirent(dir_file, 2, 0,
1122 block = dx_get_block(frame->at);
1123 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1124 start_hash, start_minor_hash);
1131 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1132 frame, frames, &hashval);
1133 *next_hash = hashval;
1139 * Stop if: (a) there are no more entries, or
1140 * (b) we have inserted at least one entry and the
1141 * next hash value is not a continuation
1144 (count && ((hashval & 1) == 0)))
1148 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1149 "next hash: %x\n", count, *next_hash));
1156 static inline int search_dirblock(struct buffer_head *bh,
1158 const struct qstr *d_name,
1159 unsigned int offset,
1160 struct ext4_dir_entry_2 **res_dir)
1162 return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1163 d_name, offset, res_dir);
1167 * Directory block splitting, compacting
1171 * Create map of hash values, offsets, and sizes, stored at end of block.
1172 * Returns number of entries mapped.
1174 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1175 unsigned blocksize, struct dx_hash_info *hinfo,
1176 struct dx_map_entry *map_tail)
1179 char *base = (char *) de;
1180 struct dx_hash_info h = *hinfo;
1182 while ((char *) de < base + blocksize) {
1183 if (de->name_len && de->inode) {
1184 ext4fs_dirhash(de->name, de->name_len, &h);
1186 map_tail->hash = h.hash;
1187 map_tail->offs = ((char *) de - base)>>2;
1188 map_tail->size = le16_to_cpu(de->rec_len);
1192 /* XXX: do we need to check rec_len == 0 case? -Chris */
1193 de = ext4_next_entry(de, blocksize);
1198 /* Sort map by hash value */
1199 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1201 struct dx_map_entry *p, *q, *top = map + count - 1;
1203 /* Combsort until bubble sort doesn't suck */
1205 count = count*10/13;
1206 if (count - 9 < 2) /* 9, 10 -> 11 */
1208 for (p = top, q = p - count; q >= map; p--, q--)
1209 if (p->hash < q->hash)
1212 /* Garden variety bubble sort */
1217 if (q[1].hash >= q[0].hash)
1225 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1227 struct dx_entry *entries = frame->entries;
1228 struct dx_entry *old = frame->at, *new = old + 1;
1229 int count = dx_get_count(entries);
1231 assert(count < dx_get_limit(entries));
1232 assert(old < entries + count);
1233 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1234 dx_set_hash(new, hash);
1235 dx_set_block(new, block);
1236 dx_set_count(entries, count + 1);
1240 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1242 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1243 * `de != NULL' is guaranteed by caller.
1245 static inline int ext4_match(struct ext4_fname_crypto_ctx *ctx,
1246 struct ext4_str *fname_crypto_str,
1247 int len, const char * const name,
1248 struct ext4_dir_entry_2 *de)
1255 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1257 return ext4_fname_match(ctx, fname_crypto_str, len, name, de);
1259 if (len != de->name_len)
1261 res = memcmp(name, de->name, len);
1262 return (res == 0) ? 1 : 0;
1266 * Returns 0 if not found, -1 on failure, and 1 on success
1268 int search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1269 struct inode *dir, const struct qstr *d_name,
1270 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1272 struct ext4_dir_entry_2 * de;
1275 const char *name = d_name->name;
1276 int namelen = d_name->len;
1277 struct ext4_fname_crypto_ctx *ctx = NULL;
1278 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
1281 ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
1285 de = (struct ext4_dir_entry_2 *)search_buf;
1286 dlimit = search_buf + buf_size;
1287 while ((char *) de < dlimit) {
1288 /* this code is executed quadratically often */
1289 /* do minimal checking `by hand' */
1290 if ((char *) de + de->name_len <= dlimit) {
1291 res = ext4_match(ctx, &fname_crypto_str, namelen,
1298 /* found a match - just to be sure, do
1300 if (ext4_check_dir_entry(dir, NULL, de, bh,
1302 bh->b_size, offset)) {
1312 /* prevent looping on a bad block */
1313 de_len = ext4_rec_len_from_disk(de->rec_len,
1314 dir->i_sb->s_blocksize);
1320 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1325 ext4_put_fname_crypto_ctx(&ctx);
1326 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1330 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1331 struct ext4_dir_entry *de)
1333 struct super_block *sb = dir->i_sb;
1339 if (de->inode == 0 &&
1340 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1349 * finds an entry in the specified directory with the wanted name. It
1350 * returns the cache buffer in which the entry was found, and the entry
1351 * itself (as a parameter - res_dir). It does NOT read the inode of the
1352 * entry - you'll have to do that yourself if you want to.
1354 * The returned buffer_head has ->b_count elevated. The caller is expected
1355 * to brelse() it when appropriate.
1357 static struct buffer_head * ext4_find_entry (struct inode *dir,
1358 const struct qstr *d_name,
1359 struct ext4_dir_entry_2 **res_dir,
1362 struct super_block *sb;
1363 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1364 struct buffer_head *bh, *ret = NULL;
1365 ext4_lblk_t start, block, b;
1366 const u8 *name = d_name->name;
1367 int ra_max = 0; /* Number of bh's in the readahead
1369 int ra_ptr = 0; /* Current index into readahead
1372 ext4_lblk_t nblocks;
1377 namelen = d_name->len;
1378 if (namelen > EXT4_NAME_LEN)
1381 if (ext4_has_inline_data(dir)) {
1382 int has_inline_data = 1;
1383 ret = ext4_find_inline_entry(dir, d_name, res_dir,
1385 if (has_inline_data) {
1392 if ((namelen <= 2) && (name[0] == '.') &&
1393 (name[1] == '.' || name[1] == '\0')) {
1395 * "." or ".." will only be in the first block
1396 * NFS may look up ".."; "." should be handled by the VFS
1403 bh = ext4_dx_find_entry(dir, d_name, res_dir);
1405 * On success, or if the error was file not found,
1406 * return. Otherwise, fall back to doing a search the
1407 * old fashioned way.
1409 if (!IS_ERR(bh) || PTR_ERR(bh) != ERR_BAD_DX_DIR)
1411 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1414 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1415 start = EXT4_I(dir)->i_dir_start_lookup;
1416 if (start >= nblocks)
1422 * We deal with the read-ahead logic here.
1424 if (ra_ptr >= ra_max) {
1425 /* Refill the readahead buffer */
1428 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1430 * Terminate if we reach the end of the
1431 * directory and must wrap, or if our
1432 * search has finished at this block.
1434 if (b >= nblocks || (num && block == start)) {
1435 bh_use[ra_max] = NULL;
1439 bh = ext4_getblk(NULL, dir, b++, 0);
1440 if (unlikely(IS_ERR(bh))) {
1445 bh_use[ra_max] = bh;
1447 ll_rw_block(READ | REQ_META | REQ_PRIO,
1451 if ((bh = bh_use[ra_ptr++]) == NULL)
1454 if (!buffer_uptodate(bh)) {
1455 /* read error, skip block & hope for the best */
1456 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1457 (unsigned long) block);
1461 if (!buffer_verified(bh) &&
1462 !is_dx_internal_node(dir, block,
1463 (struct ext4_dir_entry *)bh->b_data) &&
1464 !ext4_dirent_csum_verify(dir,
1465 (struct ext4_dir_entry *)bh->b_data)) {
1466 EXT4_ERROR_INODE(dir, "checksumming directory "
1467 "block %lu", (unsigned long)block);
1471 set_buffer_verified(bh);
1472 i = search_dirblock(bh, dir, d_name,
1473 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1475 EXT4_I(dir)->i_dir_start_lookup = block;
1477 goto cleanup_and_exit;
1481 goto cleanup_and_exit;
1484 if (++block >= nblocks)
1486 } while (block != start);
1489 * If the directory has grown while we were searching, then
1490 * search the last part of the directory before giving up.
1493 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1494 if (block < nblocks) {
1500 /* Clean up the read-ahead blocks */
1501 for (; ra_ptr < ra_max; ra_ptr++)
1502 brelse(bh_use[ra_ptr]);
1506 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1507 struct ext4_dir_entry_2 **res_dir)
1509 struct super_block * sb = dir->i_sb;
1510 struct dx_hash_info hinfo;
1511 struct dx_frame frames[2], *frame;
1512 struct buffer_head *bh;
1516 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1519 frame = dx_probe(d_name, dir, &hinfo, frames);
1521 return (struct buffer_head *) frame;
1523 block = dx_get_block(frame->at);
1524 bh = ext4_read_dirblock(dir, block, DIRENT);
1528 retval = search_dirblock(bh, dir, d_name,
1529 block << EXT4_BLOCK_SIZE_BITS(sb),
1535 bh = ERR_PTR(ERR_BAD_DX_DIR);
1539 /* Check to see if we should continue to search */
1540 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1544 "error %d reading index page in directory #%lu",
1545 retval, dir->i_ino);
1546 bh = ERR_PTR(retval);
1549 } while (retval == 1);
1553 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1559 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1561 struct inode *inode;
1562 struct ext4_dir_entry_2 *de;
1563 struct buffer_head *bh;
1565 if (dentry->d_name.len > EXT4_NAME_LEN)
1566 return ERR_PTR(-ENAMETOOLONG);
1568 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1570 return (struct dentry *) bh;
1573 __u32 ino = le32_to_cpu(de->inode);
1575 if (!ext4_valid_inum(dir->i_sb, ino)) {
1576 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1577 return ERR_PTR(-EIO);
1579 if (unlikely(ino == dir->i_ino)) {
1580 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1582 return ERR_PTR(-EIO);
1584 inode = ext4_iget_normal(dir->i_sb, ino);
1585 if (inode == ERR_PTR(-ESTALE)) {
1586 EXT4_ERROR_INODE(dir,
1587 "deleted inode referenced: %u",
1589 return ERR_PTR(-EIO);
1591 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1592 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1593 S_ISLNK(inode->i_mode)) &&
1594 !ext4_is_child_context_consistent_with_parent(dir,
1597 ext4_warning(inode->i_sb,
1598 "Inconsistent encryption contexts: %lu/%lu\n",
1599 (unsigned long) dir->i_ino,
1600 (unsigned long) inode->i_ino);
1601 return ERR_PTR(-EPERM);
1604 return d_splice_alias(inode, dentry);
1608 struct dentry *ext4_get_parent(struct dentry *child)
1611 static const struct qstr dotdot = QSTR_INIT("..", 2);
1612 struct ext4_dir_entry_2 * de;
1613 struct buffer_head *bh;
1615 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1617 return (struct dentry *) bh;
1619 return ERR_PTR(-ENOENT);
1620 ino = le32_to_cpu(de->inode);
1623 if (!ext4_valid_inum(d_inode(child)->i_sb, ino)) {
1624 EXT4_ERROR_INODE(d_inode(child),
1625 "bad parent inode number: %u", ino);
1626 return ERR_PTR(-EIO);
1629 return d_obtain_alias(ext4_iget_normal(d_inode(child)->i_sb, ino));
1633 * Move count entries from end of map between two memory locations.
1634 * Returns pointer to last entry moved.
1636 static struct ext4_dir_entry_2 *
1637 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1640 unsigned rec_len = 0;
1643 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1644 (from + (map->offs<<2));
1645 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1646 memcpy (to, de, rec_len);
1647 ((struct ext4_dir_entry_2 *) to)->rec_len =
1648 ext4_rec_len_to_disk(rec_len, blocksize);
1653 return (struct ext4_dir_entry_2 *) (to - rec_len);
1657 * Compact each dir entry in the range to the minimal rec_len.
1658 * Returns pointer to last entry in range.
1660 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1662 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1663 unsigned rec_len = 0;
1666 while ((char*)de < base + blocksize) {
1667 next = ext4_next_entry(de, blocksize);
1668 if (de->inode && de->name_len) {
1669 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1671 memmove(to, de, rec_len);
1672 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1674 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1682 * Split a full leaf block to make room for a new dir entry.
1683 * Allocate a new block, and move entries so that they are approx. equally full.
1684 * Returns pointer to de in block into which the new entry will be inserted.
1686 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1687 struct buffer_head **bh,struct dx_frame *frame,
1688 struct dx_hash_info *hinfo)
1690 unsigned blocksize = dir->i_sb->s_blocksize;
1691 unsigned count, continued;
1692 struct buffer_head *bh2;
1693 ext4_lblk_t newblock;
1695 struct dx_map_entry *map;
1696 char *data1 = (*bh)->b_data, *data2;
1697 unsigned split, move, size;
1698 struct ext4_dir_entry_2 *de = NULL, *de2;
1699 struct ext4_dir_entry_tail *t;
1703 if (ext4_has_metadata_csum(dir->i_sb))
1704 csum_size = sizeof(struct ext4_dir_entry_tail);
1706 bh2 = ext4_append(handle, dir, &newblock);
1710 return (struct ext4_dir_entry_2 *) bh2;
1713 BUFFER_TRACE(*bh, "get_write_access");
1714 err = ext4_journal_get_write_access(handle, *bh);
1718 BUFFER_TRACE(frame->bh, "get_write_access");
1719 err = ext4_journal_get_write_access(handle, frame->bh);
1723 data2 = bh2->b_data;
1725 /* create map in the end of data2 block */
1726 map = (struct dx_map_entry *) (data2 + blocksize);
1727 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1728 blocksize, hinfo, map);
1730 dx_sort_map(map, count);
1731 /* Split the existing block in the middle, size-wise */
1734 for (i = count-1; i >= 0; i--) {
1735 /* is more than half of this entry in 2nd half of the block? */
1736 if (size + map[i].size/2 > blocksize/2)
1738 size += map[i].size;
1741 /* map index at which we will split */
1742 split = count - move;
1743 hash2 = map[split].hash;
1744 continued = hash2 == map[split - 1].hash;
1745 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1746 (unsigned long)dx_get_block(frame->at),
1747 hash2, split, count-split));
1749 /* Fancy dance to stay within two buffers */
1750 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1752 de = dx_pack_dirents(data1, blocksize);
1753 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1756 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1760 t = EXT4_DIRENT_TAIL(data2, blocksize);
1761 initialize_dirent_tail(t, blocksize);
1763 t = EXT4_DIRENT_TAIL(data1, blocksize);
1764 initialize_dirent_tail(t, blocksize);
1767 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1769 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1772 /* Which block gets the new entry? */
1773 if (hinfo->hash >= hash2) {
1777 dx_insert_block(frame, hash2 + continued, newblock);
1778 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1781 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1785 dxtrace(dx_show_index("frame", frame->entries));
1792 ext4_std_error(dir->i_sb, err);
1793 return ERR_PTR(err);
1796 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1797 struct buffer_head *bh,
1798 void *buf, int buf_size,
1799 const char *name, int namelen,
1800 struct ext4_dir_entry_2 **dest_de)
1802 struct ext4_dir_entry_2 *de;
1803 unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1805 unsigned int offset = 0;
1807 struct ext4_fname_crypto_ctx *ctx = NULL;
1808 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
1811 ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
1816 /* Calculate record length needed to store the entry */
1817 res = ext4_fname_crypto_namelen_on_disk(ctx, namelen);
1819 ext4_put_fname_crypto_ctx(&ctx);
1822 reclen = EXT4_DIR_REC_LEN(res);
1825 de = (struct ext4_dir_entry_2 *)buf;
1826 top = buf + buf_size - reclen;
1827 while ((char *) de <= top) {
1828 if (ext4_check_dir_entry(dir, NULL, de, bh,
1829 buf, buf_size, offset)) {
1833 /* Provide crypto context and crypto buffer to ext4 match */
1834 res = ext4_match(ctx, &fname_crypto_str, namelen, name, de);
1841 nlen = EXT4_DIR_REC_LEN(de->name_len);
1842 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1843 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1845 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1849 if ((char *) de > top)
1856 ext4_put_fname_crypto_ctx(&ctx);
1857 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1861 int ext4_insert_dentry(struct inode *dir,
1862 struct inode *inode,
1863 struct ext4_dir_entry_2 *de,
1865 const struct qstr *iname,
1866 const char *name, int namelen)
1870 struct ext4_fname_crypto_ctx *ctx = NULL;
1871 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
1872 struct ext4_str tmp_str;
1875 ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
1878 /* By default, the input name would be written to the disk */
1879 tmp_str.name = (unsigned char *)name;
1880 tmp_str.len = namelen;
1882 /* Directory is encrypted */
1883 res = ext4_fname_crypto_alloc_buffer(ctx, EXT4_NAME_LEN,
1886 ext4_put_fname_crypto_ctx(&ctx);
1889 res = ext4_fname_usr_to_disk(ctx, iname, &fname_crypto_str);
1891 ext4_put_fname_crypto_ctx(&ctx);
1892 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1895 tmp_str.name = fname_crypto_str.name;
1896 tmp_str.len = fname_crypto_str.len;
1899 nlen = EXT4_DIR_REC_LEN(de->name_len);
1900 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1902 struct ext4_dir_entry_2 *de1 =
1903 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1904 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1905 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1908 de->file_type = EXT4_FT_UNKNOWN;
1909 de->inode = cpu_to_le32(inode->i_ino);
1910 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1911 de->name_len = tmp_str.len;
1913 memcpy(de->name, tmp_str.name, tmp_str.len);
1914 ext4_put_fname_crypto_ctx(&ctx);
1915 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1920 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1921 * it points to a directory entry which is guaranteed to be large
1922 * enough for new directory entry. If de is NULL, then
1923 * add_dirent_to_buf will attempt search the directory block for
1924 * space. It will return -ENOSPC if no space is available, and -EIO
1925 * and -EEXIST if directory entry already exists.
1927 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1928 struct inode *inode, struct ext4_dir_entry_2 *de,
1929 struct buffer_head *bh)
1931 struct inode *dir = d_inode(dentry->d_parent);
1932 const char *name = dentry->d_name.name;
1933 int namelen = dentry->d_name.len;
1934 unsigned int blocksize = dir->i_sb->s_blocksize;
1938 if (ext4_has_metadata_csum(inode->i_sb))
1939 csum_size = sizeof(struct ext4_dir_entry_tail);
1942 err = ext4_find_dest_de(dir, inode,
1943 bh, bh->b_data, blocksize - csum_size,
1944 name, namelen, &de);
1948 BUFFER_TRACE(bh, "get_write_access");
1949 err = ext4_journal_get_write_access(handle, bh);
1951 ext4_std_error(dir->i_sb, err);
1955 /* By now the buffer is marked for journaling. Due to crypto operations,
1956 * the following function call may fail */
1957 err = ext4_insert_dentry(dir, inode, de, blocksize, &dentry->d_name,
1963 * XXX shouldn't update any times until successful
1964 * completion of syscall, but too many callers depend
1967 * XXX similarly, too many callers depend on
1968 * ext4_new_inode() setting the times, but error
1969 * recovery deletes the inode, so the worst that can
1970 * happen is that the times are slightly out of date
1971 * and/or different from the directory change time.
1973 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1974 ext4_update_dx_flag(dir);
1976 ext4_mark_inode_dirty(handle, dir);
1977 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1978 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1980 ext4_std_error(dir->i_sb, err);
1985 * This converts a one block unindexed directory to a 3 block indexed
1986 * directory, and adds the dentry to the indexed directory.
1988 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1989 struct inode *inode, struct buffer_head *bh)
1991 struct inode *dir = d_inode(dentry->d_parent);
1992 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1993 struct ext4_fname_crypto_ctx *ctx = NULL;
1996 const char *name = dentry->d_name.name;
1997 int namelen = dentry->d_name.len;
1999 struct buffer_head *bh2;
2000 struct dx_root *root;
2001 struct dx_frame frames[2], *frame;
2002 struct dx_entry *entries;
2003 struct ext4_dir_entry_2 *de, *de2;
2004 struct ext4_dir_entry_tail *t;
2009 struct dx_hash_info hinfo;
2011 struct fake_dirent *fde;
2014 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2015 ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
2017 return PTR_ERR(ctx);
2020 if (ext4_has_metadata_csum(inode->i_sb))
2021 csum_size = sizeof(struct ext4_dir_entry_tail);
2023 blocksize = dir->i_sb->s_blocksize;
2024 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2025 BUFFER_TRACE(bh, "get_write_access");
2026 retval = ext4_journal_get_write_access(handle, bh);
2028 ext4_std_error(dir->i_sb, retval);
2032 root = (struct dx_root *) bh->b_data;
2034 /* The 0th block becomes the root, move the dirents out */
2035 fde = &root->dotdot;
2036 de = (struct ext4_dir_entry_2 *)((char *)fde +
2037 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2038 if ((char *) de >= (((char *) root) + blocksize)) {
2039 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2043 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2045 /* Allocate new block for the 0th block's dirents */
2046 bh2 = ext4_append(handle, dir, &block);
2049 return PTR_ERR(bh2);
2051 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2052 data1 = bh2->b_data;
2054 memcpy (data1, de, len);
2055 de = (struct ext4_dir_entry_2 *) data1;
2057 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2059 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2064 t = EXT4_DIRENT_TAIL(data1, blocksize);
2065 initialize_dirent_tail(t, blocksize);
2068 /* Initialize the root; the dot dirents already exist */
2069 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2070 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2072 memset (&root->info, 0, sizeof(root->info));
2073 root->info.info_length = sizeof(root->info);
2074 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2075 entries = root->entries;
2076 dx_set_block(entries, 1);
2077 dx_set_count(entries, 1);
2078 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2080 /* Initialize as for dx_probe */
2081 hinfo.hash_version = root->info.hash_version;
2082 if (hinfo.hash_version <= DX_HASH_TEA)
2083 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2084 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2085 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2086 res = ext4_fname_usr_to_hash(ctx, &dentry->d_name, &hinfo);
2088 ext4_put_fname_crypto_ctx(&ctx);
2089 ext4_mark_inode_dirty(handle, dir);
2093 ext4_put_fname_crypto_ctx(&ctx);
2095 ext4fs_dirhash(name, namelen, &hinfo);
2097 memset(frames, 0, sizeof(frames));
2099 frame->entries = entries;
2100 frame->at = entries;
2104 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2107 retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
2111 de = do_split(handle,dir, &bh, frame, &hinfo);
2113 retval = PTR_ERR(de);
2118 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
2123 * Even if the block split failed, we have to properly write
2124 * out all the changes we did so far. Otherwise we can end up
2125 * with corrupted filesystem.
2127 ext4_mark_inode_dirty(handle, dir);
2135 * adds a file entry to the specified directory, using the same
2136 * semantics as ext4_find_entry(). It returns NULL if it failed.
2138 * NOTE!! The inode part of 'de' is left at 0 - which means you
2139 * may not sleep between calling this and putting something into
2140 * the entry, as someone else might have used it while you slept.
2142 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2143 struct inode *inode)
2145 struct inode *dir = d_inode(dentry->d_parent);
2146 struct buffer_head *bh = NULL;
2147 struct ext4_dir_entry_2 *de;
2148 struct ext4_dir_entry_tail *t;
2149 struct super_block *sb;
2153 ext4_lblk_t block, blocks;
2156 if (ext4_has_metadata_csum(inode->i_sb))
2157 csum_size = sizeof(struct ext4_dir_entry_tail);
2160 blocksize = sb->s_blocksize;
2161 if (!dentry->d_name.len)
2164 if (ext4_has_inline_data(dir)) {
2165 retval = ext4_try_add_inline_entry(handle, dentry, inode);
2175 retval = ext4_dx_add_entry(handle, dentry, inode);
2176 if (!retval || (retval != ERR_BAD_DX_DIR))
2178 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2180 ext4_mark_inode_dirty(handle, dir);
2182 blocks = dir->i_size >> sb->s_blocksize_bits;
2183 for (block = 0; block < blocks; block++) {
2184 bh = ext4_read_dirblock(dir, block, DIRENT);
2188 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
2189 if (retval != -ENOSPC)
2192 if (blocks == 1 && !dx_fallback &&
2193 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
2194 retval = make_indexed_dir(handle, dentry, inode, bh);
2195 bh = NULL; /* make_indexed_dir releases bh */
2200 bh = ext4_append(handle, dir, &block);
2203 de = (struct ext4_dir_entry_2 *) bh->b_data;
2205 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2208 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2209 initialize_dirent_tail(t, blocksize);
2212 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
2216 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2221 * Returns 0 for success, or a negative error value
2223 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
2224 struct inode *inode)
2226 struct dx_frame frames[2], *frame;
2227 struct dx_entry *entries, *at;
2228 struct dx_hash_info hinfo;
2229 struct buffer_head *bh;
2230 struct inode *dir = d_inode(dentry->d_parent);
2231 struct super_block *sb = dir->i_sb;
2232 struct ext4_dir_entry_2 *de;
2235 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames);
2237 return PTR_ERR(frame);
2238 entries = frame->entries;
2240 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2247 BUFFER_TRACE(bh, "get_write_access");
2248 err = ext4_journal_get_write_access(handle, bh);
2252 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
2256 /* Block full, should compress but for now just split */
2257 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2258 dx_get_count(entries), dx_get_limit(entries)));
2259 /* Need to split index? */
2260 if (dx_get_count(entries) == dx_get_limit(entries)) {
2261 ext4_lblk_t newblock;
2262 unsigned icount = dx_get_count(entries);
2263 int levels = frame - frames;
2264 struct dx_entry *entries2;
2265 struct dx_node *node2;
2266 struct buffer_head *bh2;
2268 if (levels && (dx_get_count(frames->entries) ==
2269 dx_get_limit(frames->entries))) {
2270 ext4_warning(sb, "Directory index full!");
2274 bh2 = ext4_append(handle, dir, &newblock);
2279 node2 = (struct dx_node *)(bh2->b_data);
2280 entries2 = node2->entries;
2281 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2282 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2284 BUFFER_TRACE(frame->bh, "get_write_access");
2285 err = ext4_journal_get_write_access(handle, frame->bh);
2289 unsigned icount1 = icount/2, icount2 = icount - icount1;
2290 unsigned hash2 = dx_get_hash(entries + icount1);
2291 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2294 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2295 err = ext4_journal_get_write_access(handle,
2300 memcpy((char *) entries2, (char *) (entries + icount1),
2301 icount2 * sizeof(struct dx_entry));
2302 dx_set_count(entries, icount1);
2303 dx_set_count(entries2, icount2);
2304 dx_set_limit(entries2, dx_node_limit(dir));
2306 /* Which index block gets the new entry? */
2307 if (at - entries >= icount1) {
2308 frame->at = at = at - entries - icount1 + entries2;
2309 frame->entries = entries = entries2;
2310 swap(frame->bh, bh2);
2312 dx_insert_block(frames + 0, hash2, newblock);
2313 dxtrace(dx_show_index("node", frames[1].entries));
2314 dxtrace(dx_show_index("node",
2315 ((struct dx_node *) bh2->b_data)->entries));
2316 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2321 dxtrace(printk(KERN_DEBUG
2322 "Creating second level index...\n"));
2323 memcpy((char *) entries2, (char *) entries,
2324 icount * sizeof(struct dx_entry));
2325 dx_set_limit(entries2, dx_node_limit(dir));
2328 dx_set_count(entries, 1);
2329 dx_set_block(entries + 0, newblock);
2330 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2332 /* Add new access path frame */
2334 frame->at = at = at - entries + entries2;
2335 frame->entries = entries = entries2;
2337 err = ext4_journal_get_write_access(handle,
2342 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2344 ext4_std_error(inode->i_sb, err);
2348 de = do_split(handle, dir, &bh, frame, &hinfo);
2353 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2357 ext4_std_error(dir->i_sb, err);
2365 * ext4_generic_delete_entry deletes a directory entry by merging it
2366 * with the previous entry
2368 int ext4_generic_delete_entry(handle_t *handle,
2370 struct ext4_dir_entry_2 *de_del,
2371 struct buffer_head *bh,
2376 struct ext4_dir_entry_2 *de, *pde;
2377 unsigned int blocksize = dir->i_sb->s_blocksize;
2382 de = (struct ext4_dir_entry_2 *)entry_buf;
2383 while (i < buf_size - csum_size) {
2384 if (ext4_check_dir_entry(dir, NULL, de, bh,
2385 bh->b_data, bh->b_size, i))
2389 pde->rec_len = ext4_rec_len_to_disk(
2390 ext4_rec_len_from_disk(pde->rec_len,
2392 ext4_rec_len_from_disk(de->rec_len,
2400 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2402 de = ext4_next_entry(de, blocksize);
2407 static int ext4_delete_entry(handle_t *handle,
2409 struct ext4_dir_entry_2 *de_del,
2410 struct buffer_head *bh)
2412 int err, csum_size = 0;
2414 if (ext4_has_inline_data(dir)) {
2415 int has_inline_data = 1;
2416 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2418 if (has_inline_data)
2422 if (ext4_has_metadata_csum(dir->i_sb))
2423 csum_size = sizeof(struct ext4_dir_entry_tail);
2425 BUFFER_TRACE(bh, "get_write_access");
2426 err = ext4_journal_get_write_access(handle, bh);
2430 err = ext4_generic_delete_entry(handle, dir, de_del,
2432 dir->i_sb->s_blocksize, csum_size);
2436 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2437 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2444 ext4_std_error(dir->i_sb, err);
2449 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2450 * since this indicates that nlinks count was previously 1.
2452 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2455 if (is_dx(inode) && inode->i_nlink > 1) {
2456 /* limit is 16-bit i_links_count */
2457 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2458 set_nlink(inode, 1);
2459 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2460 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2466 * If a directory had nlink == 1, then we should let it be 1. This indicates
2467 * directory has >EXT4_LINK_MAX subdirs.
2469 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2471 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2476 static int ext4_add_nondir(handle_t *handle,
2477 struct dentry *dentry, struct inode *inode)
2479 int err = ext4_add_entry(handle, dentry, inode);
2481 ext4_mark_inode_dirty(handle, inode);
2482 unlock_new_inode(inode);
2483 d_instantiate(dentry, inode);
2487 unlock_new_inode(inode);
2493 * By the time this is called, we already have created
2494 * the directory cache entry for the new file, but it
2495 * is so far negative - it has no inode.
2497 * If the create succeeds, we fill in the inode information
2498 * with d_instantiate().
2500 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2504 struct inode *inode;
2505 int err, credits, retries = 0;
2507 dquot_initialize(dir);
2509 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2510 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2512 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2513 NULL, EXT4_HT_DIR, credits);
2514 handle = ext4_journal_current_handle();
2515 err = PTR_ERR(inode);
2516 if (!IS_ERR(inode)) {
2517 inode->i_op = &ext4_file_inode_operations;
2518 inode->i_fop = &ext4_file_operations;
2519 ext4_set_aops(inode);
2521 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2522 if (!err && (ext4_encrypted_inode(dir) ||
2523 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)))) {
2524 err = ext4_inherit_context(dir, inode);
2527 unlock_new_inode(inode);
2533 err = ext4_add_nondir(handle, dentry, inode);
2534 if (!err && IS_DIRSYNC(dir))
2535 ext4_handle_sync(handle);
2538 ext4_journal_stop(handle);
2539 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2544 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2545 umode_t mode, dev_t rdev)
2548 struct inode *inode;
2549 int err, credits, retries = 0;
2551 if (!new_valid_dev(rdev))
2554 dquot_initialize(dir);
2556 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2557 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2559 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2560 NULL, EXT4_HT_DIR, credits);
2561 handle = ext4_journal_current_handle();
2562 err = PTR_ERR(inode);
2563 if (!IS_ERR(inode)) {
2564 init_special_inode(inode, inode->i_mode, rdev);
2565 inode->i_op = &ext4_special_inode_operations;
2566 err = ext4_add_nondir(handle, dentry, inode);
2567 if (!err && IS_DIRSYNC(dir))
2568 ext4_handle_sync(handle);
2571 ext4_journal_stop(handle);
2572 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2577 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2580 struct inode *inode;
2581 int err, retries = 0;
2583 dquot_initialize(dir);
2586 inode = ext4_new_inode_start_handle(dir, mode,
2589 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2590 4 + EXT4_XATTR_TRANS_BLOCKS);
2591 handle = ext4_journal_current_handle();
2592 err = PTR_ERR(inode);
2593 if (!IS_ERR(inode)) {
2594 inode->i_op = &ext4_file_inode_operations;
2595 inode->i_fop = &ext4_file_operations;
2596 ext4_set_aops(inode);
2597 d_tmpfile(dentry, inode);
2598 err = ext4_orphan_add(handle, inode);
2600 goto err_unlock_inode;
2601 mark_inode_dirty(inode);
2602 unlock_new_inode(inode);
2605 ext4_journal_stop(handle);
2606 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2610 ext4_journal_stop(handle);
2611 unlock_new_inode(inode);
2615 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2616 struct ext4_dir_entry_2 *de,
2617 int blocksize, int csum_size,
2618 unsigned int parent_ino, int dotdot_real_len)
2620 de->inode = cpu_to_le32(inode->i_ino);
2622 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2624 strcpy(de->name, ".");
2625 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2627 de = ext4_next_entry(de, blocksize);
2628 de->inode = cpu_to_le32(parent_ino);
2630 if (!dotdot_real_len)
2631 de->rec_len = ext4_rec_len_to_disk(blocksize -
2632 (csum_size + EXT4_DIR_REC_LEN(1)),
2635 de->rec_len = ext4_rec_len_to_disk(
2636 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2637 strcpy(de->name, "..");
2638 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2640 return ext4_next_entry(de, blocksize);
2643 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2644 struct inode *inode)
2646 struct buffer_head *dir_block = NULL;
2647 struct ext4_dir_entry_2 *de;
2648 struct ext4_dir_entry_tail *t;
2649 ext4_lblk_t block = 0;
2650 unsigned int blocksize = dir->i_sb->s_blocksize;
2654 if (ext4_has_metadata_csum(dir->i_sb))
2655 csum_size = sizeof(struct ext4_dir_entry_tail);
2657 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2658 err = ext4_try_create_inline_dir(handle, dir, inode);
2659 if (err < 0 && err != -ENOSPC)
2666 dir_block = ext4_append(handle, inode, &block);
2667 if (IS_ERR(dir_block))
2668 return PTR_ERR(dir_block);
2669 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2670 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2671 set_nlink(inode, 2);
2673 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2674 initialize_dirent_tail(t, blocksize);
2677 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2678 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2681 set_buffer_verified(dir_block);
2687 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2690 struct inode *inode;
2691 int err, credits, retries = 0;
2693 if (EXT4_DIR_LINK_MAX(dir))
2696 dquot_initialize(dir);
2698 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2699 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2701 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2703 0, NULL, EXT4_HT_DIR, credits);
2704 handle = ext4_journal_current_handle();
2705 err = PTR_ERR(inode);
2709 inode->i_op = &ext4_dir_inode_operations;
2710 inode->i_fop = &ext4_dir_operations;
2711 err = ext4_init_new_dir(handle, dir, inode);
2713 goto out_clear_inode;
2714 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2715 if (ext4_encrypted_inode(dir) ||
2716 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) {
2717 err = ext4_inherit_context(dir, inode);
2719 goto out_clear_inode;
2722 err = ext4_mark_inode_dirty(handle, inode);
2724 err = ext4_add_entry(handle, dentry, inode);
2728 unlock_new_inode(inode);
2729 ext4_mark_inode_dirty(handle, inode);
2733 ext4_inc_count(handle, dir);
2734 ext4_update_dx_flag(dir);
2735 err = ext4_mark_inode_dirty(handle, dir);
2737 goto out_clear_inode;
2738 unlock_new_inode(inode);
2739 d_instantiate(dentry, inode);
2740 if (IS_DIRSYNC(dir))
2741 ext4_handle_sync(handle);
2745 ext4_journal_stop(handle);
2746 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2752 * routine to check that the specified directory is empty (for rmdir)
2754 int ext4_empty_dir(struct inode *inode)
2756 unsigned int offset;
2757 struct buffer_head *bh;
2758 struct ext4_dir_entry_2 *de, *de1;
2759 struct super_block *sb;
2762 if (ext4_has_inline_data(inode)) {
2763 int has_inline_data = 1;
2765 err = empty_inline_dir(inode, &has_inline_data);
2766 if (has_inline_data)
2771 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2772 EXT4_ERROR_INODE(inode, "invalid size");
2775 bh = ext4_read_dirblock(inode, 0, EITHER);
2779 de = (struct ext4_dir_entry_2 *) bh->b_data;
2780 de1 = ext4_next_entry(de, sb->s_blocksize);
2781 if (le32_to_cpu(de->inode) != inode->i_ino ||
2782 !le32_to_cpu(de1->inode) ||
2783 strcmp(".", de->name) ||
2784 strcmp("..", de1->name)) {
2785 ext4_warning(inode->i_sb,
2786 "bad directory (dir #%lu) - no `.' or `..'",
2791 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2792 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2793 de = ext4_next_entry(de1, sb->s_blocksize);
2794 while (offset < inode->i_size) {
2795 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2796 unsigned int lblock;
2799 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2800 bh = ext4_read_dirblock(inode, lblock, EITHER);
2803 de = (struct ext4_dir_entry_2 *) bh->b_data;
2805 if (ext4_check_dir_entry(inode, NULL, de, bh,
2806 bh->b_data, bh->b_size, offset)) {
2807 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2809 offset = (offset | (sb->s_blocksize - 1)) + 1;
2812 if (le32_to_cpu(de->inode)) {
2816 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2817 de = ext4_next_entry(de, sb->s_blocksize);
2824 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2825 * such inodes, starting at the superblock, in case we crash before the
2826 * file is closed/deleted, or in case the inode truncate spans multiple
2827 * transactions and the last transaction is not recovered after a crash.
2829 * At filesystem recovery time, we walk this list deleting unlinked
2830 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2832 * Orphan list manipulation functions must be called under i_mutex unless
2833 * we are just creating the inode or deleting it.
2835 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2837 struct super_block *sb = inode->i_sb;
2838 struct ext4_sb_info *sbi = EXT4_SB(sb);
2839 struct ext4_iloc iloc;
2843 if (!sbi->s_journal || is_bad_inode(inode))
2846 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2847 !mutex_is_locked(&inode->i_mutex));
2849 * Exit early if inode already is on orphan list. This is a big speedup
2850 * since we don't have to contend on the global s_orphan_lock.
2852 if (!list_empty(&EXT4_I(inode)->i_orphan))
2856 * Orphan handling is only valid for files with data blocks
2857 * being truncated, or files being unlinked. Note that we either
2858 * hold i_mutex, or the inode can not be referenced from outside,
2859 * so i_nlink should not be bumped due to race
2861 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2862 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2864 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2865 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2869 err = ext4_reserve_inode_write(handle, inode, &iloc);
2873 mutex_lock(&sbi->s_orphan_lock);
2875 * Due to previous errors inode may be already a part of on-disk
2876 * orphan list. If so skip on-disk list modification.
2878 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2879 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2880 /* Insert this inode at the head of the on-disk orphan list */
2881 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2882 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2885 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2886 mutex_unlock(&sbi->s_orphan_lock);
2889 err = ext4_handle_dirty_super(handle, sb);
2890 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2895 * We have to remove inode from in-memory list if
2896 * addition to on disk orphan list failed. Stray orphan
2897 * list entries can cause panics at unmount time.
2899 mutex_lock(&sbi->s_orphan_lock);
2900 list_del(&EXT4_I(inode)->i_orphan);
2901 mutex_unlock(&sbi->s_orphan_lock);
2904 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2905 jbd_debug(4, "orphan inode %lu will point to %d\n",
2906 inode->i_ino, NEXT_ORPHAN(inode));
2908 ext4_std_error(sb, err);
2913 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2914 * of such inodes stored on disk, because it is finally being cleaned up.
2916 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2918 struct list_head *prev;
2919 struct ext4_inode_info *ei = EXT4_I(inode);
2920 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2922 struct ext4_iloc iloc;
2925 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2928 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2929 !mutex_is_locked(&inode->i_mutex));
2930 /* Do this quick check before taking global s_orphan_lock. */
2931 if (list_empty(&ei->i_orphan))
2935 /* Grab inode buffer early before taking global s_orphan_lock */
2936 err = ext4_reserve_inode_write(handle, inode, &iloc);
2939 mutex_lock(&sbi->s_orphan_lock);
2940 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2942 prev = ei->i_orphan.prev;
2943 list_del_init(&ei->i_orphan);
2945 /* If we're on an error path, we may not have a valid
2946 * transaction handle with which to update the orphan list on
2947 * disk, but we still need to remove the inode from the linked
2948 * list in memory. */
2949 if (!handle || err) {
2950 mutex_unlock(&sbi->s_orphan_lock);
2954 ino_next = NEXT_ORPHAN(inode);
2955 if (prev == &sbi->s_orphan) {
2956 jbd_debug(4, "superblock will point to %u\n", ino_next);
2957 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2958 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2960 mutex_unlock(&sbi->s_orphan_lock);
2963 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2964 mutex_unlock(&sbi->s_orphan_lock);
2965 err = ext4_handle_dirty_super(handle, inode->i_sb);
2967 struct ext4_iloc iloc2;
2968 struct inode *i_prev =
2969 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2971 jbd_debug(4, "orphan inode %lu will point to %u\n",
2972 i_prev->i_ino, ino_next);
2973 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2975 mutex_unlock(&sbi->s_orphan_lock);
2978 NEXT_ORPHAN(i_prev) = ino_next;
2979 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2980 mutex_unlock(&sbi->s_orphan_lock);
2984 NEXT_ORPHAN(inode) = 0;
2985 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2987 ext4_std_error(inode->i_sb, err);
2995 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2998 struct inode *inode;
2999 struct buffer_head *bh;
3000 struct ext4_dir_entry_2 *de;
3001 handle_t *handle = NULL;
3003 /* Initialize quotas before so that eventual writes go in
3004 * separate transaction */
3005 dquot_initialize(dir);
3006 dquot_initialize(d_inode(dentry));
3009 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3015 inode = d_inode(dentry);
3018 if (le32_to_cpu(de->inode) != inode->i_ino)
3021 retval = -ENOTEMPTY;
3022 if (!ext4_empty_dir(inode))
3025 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3026 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3027 if (IS_ERR(handle)) {
3028 retval = PTR_ERR(handle);
3033 if (IS_DIRSYNC(dir))
3034 ext4_handle_sync(handle);
3036 retval = ext4_delete_entry(handle, dir, de, bh);
3039 if (!EXT4_DIR_LINK_EMPTY(inode))
3040 ext4_warning(inode->i_sb,
3041 "empty directory has too many links (%d)",
3045 /* There's no need to set i_disksize: the fact that i_nlink is
3046 * zero will ensure that the right thing happens during any
3049 ext4_orphan_add(handle, inode);
3050 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
3051 ext4_mark_inode_dirty(handle, inode);
3052 ext4_dec_count(handle, dir);
3053 ext4_update_dx_flag(dir);
3054 ext4_mark_inode_dirty(handle, dir);
3059 ext4_journal_stop(handle);
3063 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3066 struct inode *inode;
3067 struct buffer_head *bh;
3068 struct ext4_dir_entry_2 *de;
3069 handle_t *handle = NULL;
3071 trace_ext4_unlink_enter(dir, dentry);
3072 /* Initialize quotas before so that eventual writes go
3073 * in separate transaction */
3074 dquot_initialize(dir);
3075 dquot_initialize(d_inode(dentry));
3078 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3084 inode = d_inode(dentry);
3087 if (le32_to_cpu(de->inode) != inode->i_ino)
3090 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3091 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3092 if (IS_ERR(handle)) {
3093 retval = PTR_ERR(handle);
3098 if (IS_DIRSYNC(dir))
3099 ext4_handle_sync(handle);
3101 if (!inode->i_nlink) {
3102 ext4_warning(inode->i_sb,
3103 "Deleting nonexistent file (%lu), %d",
3104 inode->i_ino, inode->i_nlink);
3105 set_nlink(inode, 1);
3107 retval = ext4_delete_entry(handle, dir, de, bh);
3110 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3111 ext4_update_dx_flag(dir);
3112 ext4_mark_inode_dirty(handle, dir);
3114 if (!inode->i_nlink)
3115 ext4_orphan_add(handle, inode);
3116 inode->i_ctime = ext4_current_time(inode);
3117 ext4_mark_inode_dirty(handle, inode);
3122 ext4_journal_stop(handle);
3123 trace_ext4_unlink_exit(dentry, retval);
3127 static int ext4_symlink(struct inode *dir,
3128 struct dentry *dentry, const char *symname)
3131 struct inode *inode;
3132 int err, len = strlen(symname);
3134 bool encryption_required;
3135 struct ext4_str disk_link;
3136 struct ext4_encrypted_symlink_data *sd = NULL;
3138 disk_link.len = len + 1;
3139 disk_link.name = (char *) symname;
3141 encryption_required = (ext4_encrypted_inode(dir) ||
3142 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3143 if (encryption_required)
3144 disk_link.len = encrypted_symlink_data_len(len) + 1;
3145 if (disk_link.len > dir->i_sb->s_blocksize)
3146 return -ENAMETOOLONG;
3148 dquot_initialize(dir);
3150 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3152 * For non-fast symlinks, we just allocate inode and put it on
3153 * orphan list in the first transaction => we need bitmap,
3154 * group descriptor, sb, inode block, quota blocks, and
3155 * possibly selinux xattr blocks.
3157 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3158 EXT4_XATTR_TRANS_BLOCKS;
3161 * Fast symlink. We have to add entry to directory
3162 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3163 * allocate new inode (bitmap, group descriptor, inode block,
3164 * quota blocks, sb is already counted in previous macros).
3166 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3167 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3170 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3171 &dentry->d_name, 0, NULL,
3172 EXT4_HT_DIR, credits);
3173 handle = ext4_journal_current_handle();
3174 if (IS_ERR(inode)) {
3176 ext4_journal_stop(handle);
3177 return PTR_ERR(inode);
3180 if (encryption_required) {
3181 struct ext4_fname_crypto_ctx *ctx = NULL;
3183 struct ext4_str ostr;
3185 sd = kzalloc(disk_link.len, GFP_NOFS);
3188 goto err_drop_inode;
3190 err = ext4_inherit_context(dir, inode);
3192 goto err_drop_inode;
3193 ctx = ext4_get_fname_crypto_ctx(inode,
3194 inode->i_sb->s_blocksize);
3195 if (IS_ERR_OR_NULL(ctx)) {
3196 /* We just set the policy, so ctx should not be NULL */
3197 err = (ctx == NULL) ? -EIO : PTR_ERR(ctx);
3198 goto err_drop_inode;
3200 istr.name = (const unsigned char *) symname;
3202 ostr.name = sd->encrypted_path;
3203 err = ext4_fname_usr_to_disk(ctx, &istr, &ostr);
3204 ext4_put_fname_crypto_ctx(&ctx);
3206 goto err_drop_inode;
3207 sd->len = cpu_to_le16(ostr.len);
3208 disk_link.name = (char *) sd;
3211 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3212 inode->i_op = &ext4_symlink_inode_operations;
3213 ext4_set_aops(inode);
3215 * We cannot call page_symlink() with transaction started
3216 * because it calls into ext4_write_begin() which can wait
3217 * for transaction commit if we are running out of space
3218 * and thus we deadlock. So we have to stop transaction now
3219 * and restart it when symlink contents is written.
3221 * To keep fs consistent in case of crash, we have to put inode
3222 * to orphan list in the mean time.
3225 err = ext4_orphan_add(handle, inode);
3226 ext4_journal_stop(handle);
3229 goto err_drop_inode;
3230 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3232 goto err_drop_inode;
3234 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3235 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3237 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3238 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3239 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3240 if (IS_ERR(handle)) {
3241 err = PTR_ERR(handle);
3243 goto err_drop_inode;
3245 set_nlink(inode, 1);
3246 err = ext4_orphan_del(handle, inode);
3248 goto err_drop_inode;
3250 /* clear the extent format for fast symlink */
3251 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3252 inode->i_op = encryption_required ?
3253 &ext4_symlink_inode_operations :
3254 &ext4_fast_symlink_inode_operations;
3255 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3257 inode->i_size = disk_link.len - 1;
3259 EXT4_I(inode)->i_disksize = inode->i_size;
3260 err = ext4_add_nondir(handle, dentry, inode);
3261 if (!err && IS_DIRSYNC(dir))
3262 ext4_handle_sync(handle);
3265 ext4_journal_stop(handle);
3270 ext4_journal_stop(handle);
3273 unlock_new_inode(inode);
3278 static int ext4_link(struct dentry *old_dentry,
3279 struct inode *dir, struct dentry *dentry)
3282 struct inode *inode = d_inode(old_dentry);
3283 int err, retries = 0;
3285 if (inode->i_nlink >= EXT4_LINK_MAX)
3287 if (ext4_encrypted_inode(dir) &&
3288 !ext4_is_child_context_consistent_with_parent(dir, inode))
3290 dquot_initialize(dir);
3293 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3294 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3295 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3297 return PTR_ERR(handle);
3299 if (IS_DIRSYNC(dir))
3300 ext4_handle_sync(handle);
3302 inode->i_ctime = ext4_current_time(inode);
3303 ext4_inc_count(handle, inode);
3306 err = ext4_add_entry(handle, dentry, inode);
3308 ext4_mark_inode_dirty(handle, inode);
3309 /* this can happen only for tmpfile being
3310 * linked the first time
3312 if (inode->i_nlink == 1)
3313 ext4_orphan_del(handle, inode);
3314 d_instantiate(dentry, inode);
3319 ext4_journal_stop(handle);
3320 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3327 * Try to find buffer head where contains the parent block.
3328 * It should be the inode block if it is inlined or the 1st block
3329 * if it is a normal dir.
3331 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3332 struct inode *inode,
3334 struct ext4_dir_entry_2 **parent_de,
3337 struct buffer_head *bh;
3339 if (!ext4_has_inline_data(inode)) {
3340 bh = ext4_read_dirblock(inode, 0, EITHER);
3342 *retval = PTR_ERR(bh);
3345 *parent_de = ext4_next_entry(
3346 (struct ext4_dir_entry_2 *)bh->b_data,
3347 inode->i_sb->s_blocksize);
3352 return ext4_get_first_inline_block(inode, parent_de, retval);
3355 struct ext4_renament {
3357 struct dentry *dentry;
3358 struct inode *inode;
3360 int dir_nlink_delta;
3362 /* entry for "dentry" */
3363 struct buffer_head *bh;
3364 struct ext4_dir_entry_2 *de;
3367 /* entry for ".." in inode if it's a directory */
3368 struct buffer_head *dir_bh;
3369 struct ext4_dir_entry_2 *parent_de;
3373 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3377 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3378 &retval, &ent->parent_de,
3382 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3384 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3385 return ext4_journal_get_write_access(handle, ent->dir_bh);
3388 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3393 ent->parent_de->inode = cpu_to_le32(dir_ino);
3394 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3395 if (!ent->dir_inlined) {
3396 if (is_dx(ent->inode)) {
3397 retval = ext4_handle_dirty_dx_node(handle,
3401 retval = ext4_handle_dirty_dirent_node(handle,
3406 retval = ext4_mark_inode_dirty(handle, ent->inode);
3409 ext4_std_error(ent->dir->i_sb, retval);
3415 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3416 unsigned ino, unsigned file_type)
3420 BUFFER_TRACE(ent->bh, "get write access");
3421 retval = ext4_journal_get_write_access(handle, ent->bh);
3424 ent->de->inode = cpu_to_le32(ino);
3425 if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb,
3426 EXT4_FEATURE_INCOMPAT_FILETYPE))
3427 ent->de->file_type = file_type;
3428 ent->dir->i_version++;
3429 ent->dir->i_ctime = ent->dir->i_mtime =
3430 ext4_current_time(ent->dir);
3431 ext4_mark_inode_dirty(handle, ent->dir);
3432 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3433 if (!ent->inlined) {
3434 retval = ext4_handle_dirty_dirent_node(handle,
3436 if (unlikely(retval)) {
3437 ext4_std_error(ent->dir->i_sb, retval);
3447 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3448 const struct qstr *d_name)
3450 int retval = -ENOENT;
3451 struct buffer_head *bh;
3452 struct ext4_dir_entry_2 *de;
3454 bh = ext4_find_entry(dir, d_name, &de, NULL);
3458 retval = ext4_delete_entry(handle, dir, de, bh);
3464 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3469 * ent->de could have moved from under us during htree split, so make
3470 * sure that we are deleting the right entry. We might also be pointing
3471 * to a stale entry in the unused part of ent->bh so just checking inum
3472 * and the name isn't enough.
3474 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3475 ent->de->name_len != ent->dentry->d_name.len ||
3476 strncmp(ent->de->name, ent->dentry->d_name.name,
3477 ent->de->name_len) ||
3479 retval = ext4_find_delete_entry(handle, ent->dir,
3480 &ent->dentry->d_name);
3482 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3483 if (retval == -ENOENT) {
3484 retval = ext4_find_delete_entry(handle, ent->dir,
3485 &ent->dentry->d_name);
3490 ext4_warning(ent->dir->i_sb,
3491 "Deleting old file (%lu), %d, error=%d",
3492 ent->dir->i_ino, ent->dir->i_nlink, retval);
3496 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3498 if (ent->dir_nlink_delta) {
3499 if (ent->dir_nlink_delta == -1)
3500 ext4_dec_count(handle, ent->dir);
3502 ext4_inc_count(handle, ent->dir);
3503 ext4_mark_inode_dirty(handle, ent->dir);
3507 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3508 int credits, handle_t **h)
3515 * for inode block, sb block, group summaries,
3518 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3519 EXT4_XATTR_TRANS_BLOCKS + 4);
3521 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3522 &ent->dentry->d_name, 0, NULL,
3523 EXT4_HT_DIR, credits);
3525 handle = ext4_journal_current_handle();
3528 ext4_journal_stop(handle);
3529 if (PTR_ERR(wh) == -ENOSPC &&
3530 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3534 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3535 wh->i_op = &ext4_special_inode_operations;
3541 * Anybody can rename anything with this: the permission checks are left to the
3542 * higher-level routines.
3544 * n.b. old_{dentry,inode) refers to the source dentry/inode
3545 * while new_{dentry,inode) refers to the destination dentry/inode
3546 * This comes from rename(const char *oldpath, const char *newpath)
3548 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3549 struct inode *new_dir, struct dentry *new_dentry,
3552 handle_t *handle = NULL;
3553 struct ext4_renament old = {
3555 .dentry = old_dentry,
3556 .inode = d_inode(old_dentry),
3558 struct ext4_renament new = {
3560 .dentry = new_dentry,
3561 .inode = d_inode(new_dentry),
3565 struct inode *whiteout = NULL;
3569 dquot_initialize(old.dir);
3570 dquot_initialize(new.dir);
3572 /* Initialize quotas before so that eventual writes go
3573 * in separate transaction */
3575 dquot_initialize(new.inode);
3577 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3579 return PTR_ERR(old.bh);
3581 * Check for inode number is _not_ due to possible IO errors.
3582 * We might rmdir the source, keep it as pwd of some process
3583 * and merrily kill the link to whatever was created under the
3584 * same name. Goodbye sticky bit ;-<
3587 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3590 if ((old.dir != new.dir) &&
3591 ext4_encrypted_inode(new.dir) &&
3592 !ext4_is_child_context_consistent_with_parent(new.dir,
3598 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3599 &new.de, &new.inlined);
3600 if (IS_ERR(new.bh)) {
3601 retval = PTR_ERR(new.bh);
3611 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3612 ext4_alloc_da_blocks(old.inode);
3614 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3615 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3616 if (!(flags & RENAME_WHITEOUT)) {
3617 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3618 if (IS_ERR(handle)) {
3619 retval = PTR_ERR(handle);
3624 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3625 if (IS_ERR(whiteout)) {
3626 retval = PTR_ERR(whiteout);
3632 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3633 ext4_handle_sync(handle);
3635 if (S_ISDIR(old.inode->i_mode)) {
3637 retval = -ENOTEMPTY;
3638 if (!ext4_empty_dir(new.inode))
3642 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3645 retval = ext4_rename_dir_prepare(handle, &old);
3650 * If we're renaming a file within an inline_data dir and adding or
3651 * setting the new dirent causes a conversion from inline_data to
3652 * extents/blockmap, we need to force the dirent delete code to
3653 * re-read the directory, or else we end up trying to delete a dirent
3654 * from what is now the extent tree root (or a block map).
3656 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3657 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3659 old_file_type = old.de->file_type;
3662 * Do this before adding a new entry, so the old entry is sure
3663 * to be still pointing to the valid old entry.
3665 retval = ext4_setent(handle, &old, whiteout->i_ino,
3669 ext4_mark_inode_dirty(handle, whiteout);
3672 retval = ext4_add_entry(handle, new.dentry, old.inode);
3676 retval = ext4_setent(handle, &new,
3677 old.inode->i_ino, old_file_type);
3682 force_reread = !ext4_test_inode_flag(new.dir,
3683 EXT4_INODE_INLINE_DATA);
3686 * Like most other Unix systems, set the ctime for inodes on a
3689 old.inode->i_ctime = ext4_current_time(old.inode);
3690 ext4_mark_inode_dirty(handle, old.inode);
3696 ext4_rename_delete(handle, &old, force_reread);
3700 ext4_dec_count(handle, new.inode);
3701 new.inode->i_ctime = ext4_current_time(new.inode);
3703 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3704 ext4_update_dx_flag(old.dir);
3706 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3710 ext4_dec_count(handle, old.dir);
3712 /* checked ext4_empty_dir above, can't have another
3713 * parent, ext4_dec_count() won't work for many-linked
3715 clear_nlink(new.inode);
3717 ext4_inc_count(handle, new.dir);
3718 ext4_update_dx_flag(new.dir);
3719 ext4_mark_inode_dirty(handle, new.dir);
3722 ext4_mark_inode_dirty(handle, old.dir);
3724 ext4_mark_inode_dirty(handle, new.inode);
3725 if (!new.inode->i_nlink)
3726 ext4_orphan_add(handle, new.inode);
3736 drop_nlink(whiteout);
3737 unlock_new_inode(whiteout);
3741 ext4_journal_stop(handle);
3745 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3746 struct inode *new_dir, struct dentry *new_dentry)
3748 handle_t *handle = NULL;
3749 struct ext4_renament old = {
3751 .dentry = old_dentry,
3752 .inode = d_inode(old_dentry),
3754 struct ext4_renament new = {
3756 .dentry = new_dentry,
3757 .inode = d_inode(new_dentry),
3762 dquot_initialize(old.dir);
3763 dquot_initialize(new.dir);
3765 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3766 &old.de, &old.inlined);
3768 return PTR_ERR(old.bh);
3770 * Check for inode number is _not_ due to possible IO errors.
3771 * We might rmdir the source, keep it as pwd of some process
3772 * and merrily kill the link to whatever was created under the
3773 * same name. Goodbye sticky bit ;-<
3776 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3779 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3780 &new.de, &new.inlined);
3781 if (IS_ERR(new.bh)) {
3782 retval = PTR_ERR(new.bh);
3787 /* RENAME_EXCHANGE case: old *and* new must both exist */
3788 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3791 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3792 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3793 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3794 if (IS_ERR(handle)) {
3795 retval = PTR_ERR(handle);
3800 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3801 ext4_handle_sync(handle);
3803 if (S_ISDIR(old.inode->i_mode)) {
3805 retval = ext4_rename_dir_prepare(handle, &old);
3809 if (S_ISDIR(new.inode->i_mode)) {
3811 retval = ext4_rename_dir_prepare(handle, &new);
3817 * Other than the special case of overwriting a directory, parents'
3818 * nlink only needs to be modified if this is a cross directory rename.
3820 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3821 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3822 new.dir_nlink_delta = -old.dir_nlink_delta;
3824 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3825 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3829 new_file_type = new.de->file_type;
3830 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3834 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3839 * Like most other Unix systems, set the ctime for inodes on a
3842 old.inode->i_ctime = ext4_current_time(old.inode);
3843 new.inode->i_ctime = ext4_current_time(new.inode);
3844 ext4_mark_inode_dirty(handle, old.inode);
3845 ext4_mark_inode_dirty(handle, new.inode);
3848 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3853 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3857 ext4_update_dir_count(handle, &old);
3858 ext4_update_dir_count(handle, &new);
3867 ext4_journal_stop(handle);
3871 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3872 struct inode *new_dir, struct dentry *new_dentry,
3875 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3878 if (flags & RENAME_EXCHANGE) {
3879 return ext4_cross_rename(old_dir, old_dentry,
3880 new_dir, new_dentry);
3883 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3887 * directories can handle most operations...
3889 const struct inode_operations ext4_dir_inode_operations = {
3890 .create = ext4_create,
3891 .lookup = ext4_lookup,
3893 .unlink = ext4_unlink,
3894 .symlink = ext4_symlink,
3895 .mkdir = ext4_mkdir,
3896 .rmdir = ext4_rmdir,
3897 .mknod = ext4_mknod,
3898 .tmpfile = ext4_tmpfile,
3899 .rename2 = ext4_rename2,
3900 .setattr = ext4_setattr,
3901 .setxattr = generic_setxattr,
3902 .getxattr = generic_getxattr,
3903 .listxattr = ext4_listxattr,
3904 .removexattr = generic_removexattr,
3905 .get_acl = ext4_get_acl,
3906 .set_acl = ext4_set_acl,
3907 .fiemap = ext4_fiemap,
3910 const struct inode_operations ext4_special_inode_operations = {
3911 .setattr = ext4_setattr,
3912 .setxattr = generic_setxattr,
3913 .getxattr = generic_getxattr,
3914 .listxattr = ext4_listxattr,
3915 .removexattr = generic_removexattr,
3916 .get_acl = ext4_get_acl,
3917 .set_acl = ext4_set_acl,