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, de,
646 printk(KERN_WARNING "Error "
647 "converting filename "
653 name = fname_crypto_str.name;
654 len = fname_crypto_str.len;
656 res = ext4_fname_disk_to_hash(ctx, de,
659 printk(KERN_WARNING "Error "
660 "converting filename "
665 printk("%*.s:(E)%x.%u ", len, name,
666 h.hash, (unsigned) ((char *) de
668 ext4_put_fname_crypto_ctx(&ctx);
669 ext4_fname_crypto_free_buffer(
673 int len = de->name_len;
674 char *name = de->name;
675 ext4fs_dirhash(de->name, de->name_len, &h);
676 printk("%*.s:%x.%u ", len, name, h.hash,
677 (unsigned) ((char *) de - base));
680 space += EXT4_DIR_REC_LEN(de->name_len);
683 de = ext4_next_entry(de, size);
685 printk("(%i)\n", names);
686 return (struct stats) { names, space, 1 };
689 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
690 struct dx_entry *entries, int levels)
692 unsigned blocksize = dir->i_sb->s_blocksize;
693 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
695 struct buffer_head *bh;
696 printk("%i indexed blocks...\n", count);
697 for (i = 0; i < count; i++, entries++)
699 ext4_lblk_t block = dx_get_block(entries);
700 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
701 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
703 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
704 bh = ext4_bread(NULL,dir, block, 0);
705 if (!bh || IS_ERR(bh))
708 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
709 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
710 bh->b_data, blocksize, 0);
711 names += stats.names;
712 space += stats.space;
713 bcount += stats.bcount;
717 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
718 levels ? "" : " ", names, space/bcount,
719 (space/bcount)*100/blocksize);
720 return (struct stats) { names, space, bcount};
722 #endif /* DX_DEBUG */
725 * Probe for a directory leaf block to search.
727 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
728 * error in the directory index, and the caller should fall back to
729 * searching the directory normally. The callers of dx_probe **MUST**
730 * check for this error code, and make sure it never gets reflected
733 static struct dx_frame *
734 dx_probe(const struct qstr *d_name, struct inode *dir,
735 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
737 unsigned count, indirect;
738 struct dx_entry *at, *entries, *p, *q, *m;
739 struct dx_root *root;
740 struct dx_frame *frame = frame_in;
741 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
744 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
745 if (IS_ERR(frame->bh))
746 return (struct dx_frame *) frame->bh;
748 root = (struct dx_root *) frame->bh->b_data;
749 if (root->info.hash_version != DX_HASH_TEA &&
750 root->info.hash_version != DX_HASH_HALF_MD4 &&
751 root->info.hash_version != DX_HASH_LEGACY) {
752 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
753 root->info.hash_version);
756 hinfo->hash_version = root->info.hash_version;
757 if (hinfo->hash_version <= DX_HASH_TEA)
758 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
759 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
760 #ifdef CONFIG_EXT4_FS_ENCRYPTION
762 struct ext4_fname_crypto_ctx *ctx = NULL;
765 /* Check if the directory is encrypted */
766 ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
768 ret_err = ERR_PTR(PTR_ERR(ctx));
771 res = ext4_fname_usr_to_hash(ctx, d_name, hinfo);
773 ret_err = ERR_PTR(res);
776 ext4_put_fname_crypto_ctx(&ctx);
780 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
784 if (root->info.unused_flags & 1) {
785 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
786 root->info.unused_flags);
790 if ((indirect = root->info.indirect_levels) > 1) {
791 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
792 root->info.indirect_levels);
796 entries = (struct dx_entry *) (((char *)&root->info) +
797 root->info.info_length);
799 if (dx_get_limit(entries) != dx_root_limit(dir,
800 root->info.info_length)) {
801 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
805 dxtrace(printk("Look up %x", hash));
807 count = dx_get_count(entries);
808 if (!count || count > dx_get_limit(entries)) {
809 ext4_warning(dir->i_sb,
810 "dx entry: no count or count > limit");
815 q = entries + count - 1;
818 dxtrace(printk("."));
819 if (dx_get_hash(m) > hash)
825 if (0) { // linear search cross check
826 unsigned n = count - 1;
830 dxtrace(printk(","));
831 if (dx_get_hash(++at) > hash)
837 assert (at == p - 1);
841 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
842 frame->entries = entries;
847 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
848 if (IS_ERR(frame->bh)) {
849 ret_err = (struct dx_frame *) frame->bh;
853 entries = ((struct dx_node *) frame->bh->b_data)->entries;
855 if (dx_get_limit(entries) != dx_node_limit (dir)) {
856 ext4_warning(dir->i_sb,
857 "dx entry: limit != node limit");
862 while (frame >= frame_in) {
867 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
868 ext4_warning(dir->i_sb,
869 "Corrupt dir inode %lu, running e2fsck is "
870 "recommended.", dir->i_ino);
874 static void dx_release (struct dx_frame *frames)
876 if (frames[0].bh == NULL)
879 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
880 brelse(frames[1].bh);
881 brelse(frames[0].bh);
885 * This function increments the frame pointer to search the next leaf
886 * block, and reads in the necessary intervening nodes if the search
887 * should be necessary. Whether or not the search is necessary is
888 * controlled by the hash parameter. If the hash value is even, then
889 * the search is only continued if the next block starts with that
890 * hash value. This is used if we are searching for a specific file.
892 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
894 * This function returns 1 if the caller should continue to search,
895 * or 0 if it should not. If there is an error reading one of the
896 * index blocks, it will a negative error code.
898 * If start_hash is non-null, it will be filled in with the starting
899 * hash of the next page.
901 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
902 struct dx_frame *frame,
903 struct dx_frame *frames,
907 struct buffer_head *bh;
913 * Find the next leaf page by incrementing the frame pointer.
914 * If we run out of entries in the interior node, loop around and
915 * increment pointer in the parent node. When we break out of
916 * this loop, num_frames indicates the number of interior
917 * nodes need to be read.
920 if (++(p->at) < p->entries + dx_get_count(p->entries))
929 * If the hash is 1, then continue only if the next page has a
930 * continuation hash of any value. This is used for readdir
931 * handling. Otherwise, check to see if the hash matches the
932 * desired contiuation hash. If it doesn't, return since
933 * there's no point to read in the successive index pages.
935 bhash = dx_get_hash(p->at);
938 if ((hash & 1) == 0) {
939 if ((bhash & ~1) != hash)
943 * If the hash is HASH_NB_ALWAYS, we always go to the next
944 * block so no check is necessary
946 while (num_frames--) {
947 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
953 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
960 * This function fills a red-black tree with information from a
961 * directory block. It returns the number directory entries loaded
962 * into the tree. If there is an error it is returned in err.
964 static int htree_dirblock_to_tree(struct file *dir_file,
965 struct inode *dir, ext4_lblk_t block,
966 struct dx_hash_info *hinfo,
967 __u32 start_hash, __u32 start_minor_hash)
969 struct buffer_head *bh;
970 struct ext4_dir_entry_2 *de, *top;
971 int err = 0, count = 0;
972 struct ext4_fname_crypto_ctx *ctx = NULL;
973 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}, tmp_str;
975 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
976 (unsigned long)block));
977 bh = ext4_read_dirblock(dir, block, DIRENT);
981 de = (struct ext4_dir_entry_2 *) bh->b_data;
982 top = (struct ext4_dir_entry_2 *) ((char *) de +
983 dir->i_sb->s_blocksize -
984 EXT4_DIR_REC_LEN(0));
985 #ifdef CONFIG_EXT4_FS_ENCRYPTION
986 /* Check if the directory is encrypted */
987 ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
994 err = ext4_fname_crypto_alloc_buffer(ctx, EXT4_NAME_LEN,
997 ext4_put_fname_crypto_ctx(&ctx);
1003 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1004 if (ext4_check_dir_entry(dir, NULL, de, bh,
1005 bh->b_data, bh->b_size,
1006 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1007 + ((char *)de - bh->b_data))) {
1008 /* silently ignore the rest of the block */
1011 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1012 err = ext4_fname_disk_to_hash(ctx, de, hinfo);
1018 ext4fs_dirhash(de->name, de->name_len, hinfo);
1020 if ((hinfo->hash < start_hash) ||
1021 ((hinfo->hash == start_hash) &&
1022 (hinfo->minor_hash < start_minor_hash)))
1027 /* Directory is not encrypted */
1028 tmp_str.name = de->name;
1029 tmp_str.len = de->name_len;
1030 err = ext4_htree_store_dirent(dir_file,
1031 hinfo->hash, hinfo->minor_hash, de,
1034 /* Directory is encrypted */
1035 err = ext4_fname_disk_to_usr(ctx, de,
1041 err = ext4_htree_store_dirent(dir_file,
1042 hinfo->hash, hinfo->minor_hash, de,
1053 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1054 ext4_put_fname_crypto_ctx(&ctx);
1055 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1062 * This function fills a red-black tree with information from a
1063 * directory. We start scanning the directory in hash order, starting
1064 * at start_hash and start_minor_hash.
1066 * This function returns the number of entries inserted into the tree,
1067 * or a negative error code.
1069 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1070 __u32 start_minor_hash, __u32 *next_hash)
1072 struct dx_hash_info hinfo;
1073 struct ext4_dir_entry_2 *de;
1074 struct dx_frame frames[2], *frame;
1080 struct ext4_str tmp_str;
1082 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1083 start_hash, start_minor_hash));
1084 dir = file_inode(dir_file);
1085 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1086 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1087 if (hinfo.hash_version <= DX_HASH_TEA)
1088 hinfo.hash_version +=
1089 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1090 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1091 if (ext4_has_inline_data(dir)) {
1092 int has_inline_data = 1;
1093 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1097 if (has_inline_data) {
1102 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1103 start_hash, start_minor_hash);
1107 hinfo.hash = start_hash;
1108 hinfo.minor_hash = 0;
1109 frame = dx_probe(NULL, dir, &hinfo, frames);
1111 return PTR_ERR(frame);
1113 /* Add '.' and '..' from the htree header */
1114 if (!start_hash && !start_minor_hash) {
1115 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1116 tmp_str.name = de->name;
1117 tmp_str.len = de->name_len;
1118 err = ext4_htree_store_dirent(dir_file, 0, 0,
1124 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1125 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1126 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1127 tmp_str.name = de->name;
1128 tmp_str.len = de->name_len;
1129 err = ext4_htree_store_dirent(dir_file, 2, 0,
1137 block = dx_get_block(frame->at);
1138 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1139 start_hash, start_minor_hash);
1146 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1147 frame, frames, &hashval);
1148 *next_hash = hashval;
1154 * Stop if: (a) there are no more entries, or
1155 * (b) we have inserted at least one entry and the
1156 * next hash value is not a continuation
1159 (count && ((hashval & 1) == 0)))
1163 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1164 "next hash: %x\n", count, *next_hash));
1171 static inline int search_dirblock(struct buffer_head *bh,
1173 const struct qstr *d_name,
1174 unsigned int offset,
1175 struct ext4_dir_entry_2 **res_dir)
1177 return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1178 d_name, offset, res_dir);
1182 * Directory block splitting, compacting
1186 * Create map of hash values, offsets, and sizes, stored at end of block.
1187 * Returns number of entries mapped.
1189 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1190 unsigned blocksize, struct dx_hash_info *hinfo,
1191 struct dx_map_entry *map_tail)
1194 char *base = (char *) de;
1195 struct dx_hash_info h = *hinfo;
1196 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1197 struct ext4_fname_crypto_ctx *ctx = NULL;
1200 ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
1202 return PTR_ERR(ctx);
1205 while ((char *) de < base + blocksize) {
1206 if (de->name_len && de->inode) {
1207 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1208 err = ext4_fname_disk_to_hash(ctx, de, &h);
1210 ext4_put_fname_crypto_ctx(&ctx);
1214 ext4fs_dirhash(de->name, de->name_len, &h);
1217 map_tail->hash = h.hash;
1218 map_tail->offs = ((char *) de - base)>>2;
1219 map_tail->size = le16_to_cpu(de->rec_len);
1223 /* XXX: do we need to check rec_len == 0 case? -Chris */
1224 de = ext4_next_entry(de, blocksize);
1226 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1227 ext4_put_fname_crypto_ctx(&ctx);
1232 /* Sort map by hash value */
1233 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1235 struct dx_map_entry *p, *q, *top = map + count - 1;
1237 /* Combsort until bubble sort doesn't suck */
1239 count = count*10/13;
1240 if (count - 9 < 2) /* 9, 10 -> 11 */
1242 for (p = top, q = p - count; q >= map; p--, q--)
1243 if (p->hash < q->hash)
1246 /* Garden variety bubble sort */
1251 if (q[1].hash >= q[0].hash)
1259 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1261 struct dx_entry *entries = frame->entries;
1262 struct dx_entry *old = frame->at, *new = old + 1;
1263 int count = dx_get_count(entries);
1265 assert(count < dx_get_limit(entries));
1266 assert(old < entries + count);
1267 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1268 dx_set_hash(new, hash);
1269 dx_set_block(new, block);
1270 dx_set_count(entries, count + 1);
1274 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1276 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1277 * `de != NULL' is guaranteed by caller.
1279 static inline int ext4_match(struct ext4_fname_crypto_ctx *ctx,
1280 struct ext4_str *fname_crypto_str,
1281 int len, const char * const name,
1282 struct ext4_dir_entry_2 *de)
1289 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1291 /* Directory is encrypted */
1292 res = ext4_fname_disk_to_usr(ctx, de, fname_crypto_str);
1297 res = memcmp(name, fname_crypto_str->name, len);
1298 return (res == 0) ? 1 : 0;
1301 if (len != de->name_len)
1303 res = memcmp(name, de->name, len);
1304 return (res == 0) ? 1 : 0;
1308 * Returns 0 if not found, -1 on failure, and 1 on success
1310 int search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1311 struct inode *dir, const struct qstr *d_name,
1312 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1314 struct ext4_dir_entry_2 * de;
1317 const char *name = d_name->name;
1318 int namelen = d_name->len;
1319 struct ext4_fname_crypto_ctx *ctx = NULL;
1320 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
1323 ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
1328 /* Allocate buffer to hold maximum name length */
1329 res = ext4_fname_crypto_alloc_buffer(ctx, EXT4_NAME_LEN,
1332 ext4_put_fname_crypto_ctx(&ctx);
1337 de = (struct ext4_dir_entry_2 *)search_buf;
1338 dlimit = search_buf + buf_size;
1339 while ((char *) de < dlimit) {
1340 /* this code is executed quadratically often */
1341 /* do minimal checking `by hand' */
1342 if ((char *) de + de->name_len <= dlimit) {
1343 res = ext4_match(ctx, &fname_crypto_str, namelen,
1350 /* found a match - just to be sure, do
1352 if (ext4_check_dir_entry(dir, NULL, de, bh,
1354 bh->b_size, offset)) {
1364 /* prevent looping on a bad block */
1365 de_len = ext4_rec_len_from_disk(de->rec_len,
1366 dir->i_sb->s_blocksize);
1372 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1377 ext4_put_fname_crypto_ctx(&ctx);
1378 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1382 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1383 struct ext4_dir_entry *de)
1385 struct super_block *sb = dir->i_sb;
1391 if (de->inode == 0 &&
1392 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1401 * finds an entry in the specified directory with the wanted name. It
1402 * returns the cache buffer in which the entry was found, and the entry
1403 * itself (as a parameter - res_dir). It does NOT read the inode of the
1404 * entry - you'll have to do that yourself if you want to.
1406 * The returned buffer_head has ->b_count elevated. The caller is expected
1407 * to brelse() it when appropriate.
1409 static struct buffer_head * ext4_find_entry (struct inode *dir,
1410 const struct qstr *d_name,
1411 struct ext4_dir_entry_2 **res_dir,
1414 struct super_block *sb;
1415 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1416 struct buffer_head *bh, *ret = NULL;
1417 ext4_lblk_t start, block, b;
1418 const u8 *name = d_name->name;
1419 int ra_max = 0; /* Number of bh's in the readahead
1421 int ra_ptr = 0; /* Current index into readahead
1424 ext4_lblk_t nblocks;
1429 namelen = d_name->len;
1430 if (namelen > EXT4_NAME_LEN)
1433 if (ext4_has_inline_data(dir)) {
1434 int has_inline_data = 1;
1435 ret = ext4_find_inline_entry(dir, d_name, res_dir,
1437 if (has_inline_data) {
1444 if ((namelen <= 2) && (name[0] == '.') &&
1445 (name[1] == '.' || name[1] == '\0')) {
1447 * "." or ".." will only be in the first block
1448 * NFS may look up ".."; "." should be handled by the VFS
1455 bh = ext4_dx_find_entry(dir, d_name, res_dir);
1457 * On success, or if the error was file not found,
1458 * return. Otherwise, fall back to doing a search the
1459 * old fashioned way.
1461 if (!IS_ERR(bh) || PTR_ERR(bh) != ERR_BAD_DX_DIR)
1463 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1466 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1467 start = EXT4_I(dir)->i_dir_start_lookup;
1468 if (start >= nblocks)
1474 * We deal with the read-ahead logic here.
1476 if (ra_ptr >= ra_max) {
1477 /* Refill the readahead buffer */
1480 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1482 * Terminate if we reach the end of the
1483 * directory and must wrap, or if our
1484 * search has finished at this block.
1486 if (b >= nblocks || (num && block == start)) {
1487 bh_use[ra_max] = NULL;
1491 bh = ext4_getblk(NULL, dir, b++, 0);
1492 if (unlikely(IS_ERR(bh))) {
1497 bh_use[ra_max] = bh;
1499 ll_rw_block(READ | REQ_META | REQ_PRIO,
1503 if ((bh = bh_use[ra_ptr++]) == NULL)
1506 if (!buffer_uptodate(bh)) {
1507 /* read error, skip block & hope for the best */
1508 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1509 (unsigned long) block);
1513 if (!buffer_verified(bh) &&
1514 !is_dx_internal_node(dir, block,
1515 (struct ext4_dir_entry *)bh->b_data) &&
1516 !ext4_dirent_csum_verify(dir,
1517 (struct ext4_dir_entry *)bh->b_data)) {
1518 EXT4_ERROR_INODE(dir, "checksumming directory "
1519 "block %lu", (unsigned long)block);
1523 set_buffer_verified(bh);
1524 i = search_dirblock(bh, dir, d_name,
1525 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1527 EXT4_I(dir)->i_dir_start_lookup = block;
1529 goto cleanup_and_exit;
1533 goto cleanup_and_exit;
1536 if (++block >= nblocks)
1538 } while (block != start);
1541 * If the directory has grown while we were searching, then
1542 * search the last part of the directory before giving up.
1545 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1546 if (block < nblocks) {
1552 /* Clean up the read-ahead blocks */
1553 for (; ra_ptr < ra_max; ra_ptr++)
1554 brelse(bh_use[ra_ptr]);
1558 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1559 struct ext4_dir_entry_2 **res_dir)
1561 struct super_block * sb = dir->i_sb;
1562 struct dx_hash_info hinfo;
1563 struct dx_frame frames[2], *frame;
1564 struct buffer_head *bh;
1568 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1571 frame = dx_probe(d_name, dir, &hinfo, frames);
1573 return (struct buffer_head *) frame;
1575 block = dx_get_block(frame->at);
1576 bh = ext4_read_dirblock(dir, block, DIRENT);
1580 retval = search_dirblock(bh, dir, d_name,
1581 block << EXT4_BLOCK_SIZE_BITS(sb),
1587 bh = ERR_PTR(ERR_BAD_DX_DIR);
1591 /* Check to see if we should continue to search */
1592 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1596 "error %d reading index page in directory #%lu",
1597 retval, dir->i_ino);
1598 bh = ERR_PTR(retval);
1601 } while (retval == 1);
1605 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1611 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1613 struct inode *inode;
1614 struct ext4_dir_entry_2 *de;
1615 struct buffer_head *bh;
1617 if (dentry->d_name.len > EXT4_NAME_LEN)
1618 return ERR_PTR(-ENAMETOOLONG);
1620 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1622 return (struct dentry *) bh;
1625 __u32 ino = le32_to_cpu(de->inode);
1627 if (!ext4_valid_inum(dir->i_sb, ino)) {
1628 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1629 return ERR_PTR(-EIO);
1631 if (unlikely(ino == dir->i_ino)) {
1632 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1634 return ERR_PTR(-EIO);
1636 inode = ext4_iget_normal(dir->i_sb, ino);
1637 if (inode == ERR_PTR(-ESTALE)) {
1638 EXT4_ERROR_INODE(dir,
1639 "deleted inode referenced: %u",
1641 return ERR_PTR(-EIO);
1643 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1644 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1645 S_ISLNK(inode->i_mode)) &&
1646 !ext4_is_child_context_consistent_with_parent(dir,
1649 ext4_warning(inode->i_sb,
1650 "Inconsistent encryption contexts: %lu/%lu\n",
1651 (unsigned long) dir->i_ino,
1652 (unsigned long) inode->i_ino);
1653 return ERR_PTR(-EPERM);
1656 return d_splice_alias(inode, dentry);
1660 struct dentry *ext4_get_parent(struct dentry *child)
1663 static const struct qstr dotdot = QSTR_INIT("..", 2);
1664 struct ext4_dir_entry_2 * de;
1665 struct buffer_head *bh;
1667 bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1669 return (struct dentry *) bh;
1671 return ERR_PTR(-ENOENT);
1672 ino = le32_to_cpu(de->inode);
1675 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1676 EXT4_ERROR_INODE(child->d_inode,
1677 "bad parent inode number: %u", ino);
1678 return ERR_PTR(-EIO);
1681 return d_obtain_alias(ext4_iget_normal(child->d_inode->i_sb, ino));
1685 * Move count entries from end of map between two memory locations.
1686 * Returns pointer to last entry moved.
1688 static struct ext4_dir_entry_2 *
1689 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1692 unsigned rec_len = 0;
1695 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1696 (from + (map->offs<<2));
1697 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1698 memcpy (to, de, rec_len);
1699 ((struct ext4_dir_entry_2 *) to)->rec_len =
1700 ext4_rec_len_to_disk(rec_len, blocksize);
1705 return (struct ext4_dir_entry_2 *) (to - rec_len);
1709 * Compact each dir entry in the range to the minimal rec_len.
1710 * Returns pointer to last entry in range.
1712 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1714 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1715 unsigned rec_len = 0;
1718 while ((char*)de < base + blocksize) {
1719 next = ext4_next_entry(de, blocksize);
1720 if (de->inode && de->name_len) {
1721 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1723 memmove(to, de, rec_len);
1724 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1726 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1734 * Split a full leaf block to make room for a new dir entry.
1735 * Allocate a new block, and move entries so that they are approx. equally full.
1736 * Returns pointer to de in block into which the new entry will be inserted.
1738 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1739 struct buffer_head **bh,struct dx_frame *frame,
1740 struct dx_hash_info *hinfo)
1742 unsigned blocksize = dir->i_sb->s_blocksize;
1743 unsigned count, continued;
1744 struct buffer_head *bh2;
1745 ext4_lblk_t newblock;
1747 struct dx_map_entry *map;
1748 char *data1 = (*bh)->b_data, *data2;
1749 unsigned split, move, size;
1750 struct ext4_dir_entry_2 *de = NULL, *de2;
1751 struct ext4_dir_entry_tail *t;
1755 if (ext4_has_metadata_csum(dir->i_sb))
1756 csum_size = sizeof(struct ext4_dir_entry_tail);
1758 bh2 = ext4_append(handle, dir, &newblock);
1762 return (struct ext4_dir_entry_2 *) bh2;
1765 BUFFER_TRACE(*bh, "get_write_access");
1766 err = ext4_journal_get_write_access(handle, *bh);
1770 BUFFER_TRACE(frame->bh, "get_write_access");
1771 err = ext4_journal_get_write_access(handle, frame->bh);
1775 data2 = bh2->b_data;
1777 /* create map in the end of data2 block */
1778 map = (struct dx_map_entry *) (data2 + blocksize);
1779 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1780 blocksize, hinfo, map);
1782 dx_sort_map(map, count);
1783 /* Split the existing block in the middle, size-wise */
1786 for (i = count-1; i >= 0; i--) {
1787 /* is more than half of this entry in 2nd half of the block? */
1788 if (size + map[i].size/2 > blocksize/2)
1790 size += map[i].size;
1793 /* map index at which we will split */
1794 split = count - move;
1795 hash2 = map[split].hash;
1796 continued = hash2 == map[split - 1].hash;
1797 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1798 (unsigned long)dx_get_block(frame->at),
1799 hash2, split, count-split));
1801 /* Fancy dance to stay within two buffers */
1802 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1804 de = dx_pack_dirents(data1, blocksize);
1805 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1808 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1812 t = EXT4_DIRENT_TAIL(data2, blocksize);
1813 initialize_dirent_tail(t, blocksize);
1815 t = EXT4_DIRENT_TAIL(data1, blocksize);
1816 initialize_dirent_tail(t, blocksize);
1819 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1821 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1824 /* Which block gets the new entry? */
1825 if (hinfo->hash >= hash2) {
1829 dx_insert_block(frame, hash2 + continued, newblock);
1830 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1833 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1837 dxtrace(dx_show_index("frame", frame->entries));
1844 ext4_std_error(dir->i_sb, err);
1845 return ERR_PTR(err);
1848 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1849 struct buffer_head *bh,
1850 void *buf, int buf_size,
1851 const char *name, int namelen,
1852 struct ext4_dir_entry_2 **dest_de)
1854 struct ext4_dir_entry_2 *de;
1855 unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1857 unsigned int offset = 0;
1859 struct ext4_fname_crypto_ctx *ctx = NULL;
1860 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
1863 ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
1868 /* Calculate record length needed to store the entry */
1869 res = ext4_fname_crypto_namelen_on_disk(ctx, namelen);
1871 ext4_put_fname_crypto_ctx(&ctx);
1874 reclen = EXT4_DIR_REC_LEN(res);
1876 /* Allocate buffer to hold maximum name length */
1877 res = ext4_fname_crypto_alloc_buffer(ctx, EXT4_NAME_LEN,
1880 ext4_put_fname_crypto_ctx(&ctx);
1885 de = (struct ext4_dir_entry_2 *)buf;
1886 top = buf + buf_size - reclen;
1887 while ((char *) de <= top) {
1888 if (ext4_check_dir_entry(dir, NULL, de, bh,
1889 buf, buf_size, offset)) {
1893 /* Provide crypto context and crypto buffer to ext4 match */
1894 res = ext4_match(ctx, &fname_crypto_str, namelen, name, de);
1901 nlen = EXT4_DIR_REC_LEN(de->name_len);
1902 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1903 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1905 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1909 if ((char *) de > top)
1916 ext4_put_fname_crypto_ctx(&ctx);
1917 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1921 int ext4_insert_dentry(struct inode *dir,
1922 struct inode *inode,
1923 struct ext4_dir_entry_2 *de,
1925 const struct qstr *iname,
1926 const char *name, int namelen)
1930 struct ext4_fname_crypto_ctx *ctx = NULL;
1931 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
1932 struct ext4_str tmp_str;
1935 ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
1938 /* By default, the input name would be written to the disk */
1939 tmp_str.name = (unsigned char *)name;
1940 tmp_str.len = namelen;
1942 /* Directory is encrypted */
1943 res = ext4_fname_crypto_alloc_buffer(ctx, EXT4_NAME_LEN,
1946 ext4_put_fname_crypto_ctx(&ctx);
1949 res = ext4_fname_usr_to_disk(ctx, iname, &fname_crypto_str);
1951 ext4_put_fname_crypto_ctx(&ctx);
1952 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1955 tmp_str.name = fname_crypto_str.name;
1956 tmp_str.len = fname_crypto_str.len;
1959 nlen = EXT4_DIR_REC_LEN(de->name_len);
1960 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1962 struct ext4_dir_entry_2 *de1 =
1963 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1964 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1965 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1968 de->file_type = EXT4_FT_UNKNOWN;
1969 de->inode = cpu_to_le32(inode->i_ino);
1970 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1971 de->name_len = tmp_str.len;
1973 memcpy(de->name, tmp_str.name, tmp_str.len);
1974 ext4_put_fname_crypto_ctx(&ctx);
1975 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1980 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1981 * it points to a directory entry which is guaranteed to be large
1982 * enough for new directory entry. If de is NULL, then
1983 * add_dirent_to_buf will attempt search the directory block for
1984 * space. It will return -ENOSPC if no space is available, and -EIO
1985 * and -EEXIST if directory entry already exists.
1987 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1988 struct inode *inode, struct ext4_dir_entry_2 *de,
1989 struct buffer_head *bh)
1991 struct inode *dir = dentry->d_parent->d_inode;
1992 const char *name = dentry->d_name.name;
1993 int namelen = dentry->d_name.len;
1994 unsigned int blocksize = dir->i_sb->s_blocksize;
1998 if (ext4_has_metadata_csum(inode->i_sb))
1999 csum_size = sizeof(struct ext4_dir_entry_tail);
2002 err = ext4_find_dest_de(dir, inode,
2003 bh, bh->b_data, blocksize - csum_size,
2004 name, namelen, &de);
2008 BUFFER_TRACE(bh, "get_write_access");
2009 err = ext4_journal_get_write_access(handle, bh);
2011 ext4_std_error(dir->i_sb, err);
2015 /* By now the buffer is marked for journaling. Due to crypto operations,
2016 * the following function call may fail */
2017 err = ext4_insert_dentry(dir, inode, de, blocksize, &dentry->d_name,
2023 * XXX shouldn't update any times until successful
2024 * completion of syscall, but too many callers depend
2027 * XXX similarly, too many callers depend on
2028 * ext4_new_inode() setting the times, but error
2029 * recovery deletes the inode, so the worst that can
2030 * happen is that the times are slightly out of date
2031 * and/or different from the directory change time.
2033 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
2034 ext4_update_dx_flag(dir);
2036 ext4_mark_inode_dirty(handle, dir);
2037 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2038 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2040 ext4_std_error(dir->i_sb, err);
2045 * This converts a one block unindexed directory to a 3 block indexed
2046 * directory, and adds the dentry to the indexed directory.
2048 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
2049 struct inode *inode, struct buffer_head *bh)
2051 struct inode *dir = dentry->d_parent->d_inode;
2052 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2053 struct ext4_fname_crypto_ctx *ctx = NULL;
2056 const char *name = dentry->d_name.name;
2057 int namelen = dentry->d_name.len;
2059 struct buffer_head *bh2;
2060 struct dx_root *root;
2061 struct dx_frame frames[2], *frame;
2062 struct dx_entry *entries;
2063 struct ext4_dir_entry_2 *de, *de2;
2064 struct ext4_dir_entry_tail *t;
2069 struct dx_hash_info hinfo;
2071 struct fake_dirent *fde;
2074 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2075 ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
2077 return PTR_ERR(ctx);
2080 if (ext4_has_metadata_csum(inode->i_sb))
2081 csum_size = sizeof(struct ext4_dir_entry_tail);
2083 blocksize = dir->i_sb->s_blocksize;
2084 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2085 BUFFER_TRACE(bh, "get_write_access");
2086 retval = ext4_journal_get_write_access(handle, bh);
2088 ext4_std_error(dir->i_sb, retval);
2092 root = (struct dx_root *) bh->b_data;
2094 /* The 0th block becomes the root, move the dirents out */
2095 fde = &root->dotdot;
2096 de = (struct ext4_dir_entry_2 *)((char *)fde +
2097 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2098 if ((char *) de >= (((char *) root) + blocksize)) {
2099 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2103 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2105 /* Allocate new block for the 0th block's dirents */
2106 bh2 = ext4_append(handle, dir, &block);
2109 return PTR_ERR(bh2);
2111 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2112 data1 = bh2->b_data;
2114 memcpy (data1, de, len);
2115 de = (struct ext4_dir_entry_2 *) data1;
2117 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2119 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2124 t = EXT4_DIRENT_TAIL(data1, blocksize);
2125 initialize_dirent_tail(t, blocksize);
2128 /* Initialize the root; the dot dirents already exist */
2129 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2130 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2132 memset (&root->info, 0, sizeof(root->info));
2133 root->info.info_length = sizeof(root->info);
2134 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2135 entries = root->entries;
2136 dx_set_block(entries, 1);
2137 dx_set_count(entries, 1);
2138 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2140 /* Initialize as for dx_probe */
2141 hinfo.hash_version = root->info.hash_version;
2142 if (hinfo.hash_version <= DX_HASH_TEA)
2143 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2144 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2145 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2146 res = ext4_fname_usr_to_hash(ctx, &dentry->d_name, &hinfo);
2148 ext4_put_fname_crypto_ctx(&ctx);
2149 ext4_mark_inode_dirty(handle, dir);
2153 ext4_put_fname_crypto_ctx(&ctx);
2155 ext4fs_dirhash(name, namelen, &hinfo);
2157 memset(frames, 0, sizeof(frames));
2159 frame->entries = entries;
2160 frame->at = entries;
2164 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2167 retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
2171 de = do_split(handle,dir, &bh, frame, &hinfo);
2173 retval = PTR_ERR(de);
2178 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
2183 * Even if the block split failed, we have to properly write
2184 * out all the changes we did so far. Otherwise we can end up
2185 * with corrupted filesystem.
2187 ext4_mark_inode_dirty(handle, dir);
2195 * adds a file entry to the specified directory, using the same
2196 * semantics as ext4_find_entry(). It returns NULL if it failed.
2198 * NOTE!! The inode part of 'de' is left at 0 - which means you
2199 * may not sleep between calling this and putting something into
2200 * the entry, as someone else might have used it while you slept.
2202 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2203 struct inode *inode)
2205 struct inode *dir = dentry->d_parent->d_inode;
2206 struct buffer_head *bh = NULL;
2207 struct ext4_dir_entry_2 *de;
2208 struct ext4_dir_entry_tail *t;
2209 struct super_block *sb;
2213 ext4_lblk_t block, blocks;
2216 if (ext4_has_metadata_csum(inode->i_sb))
2217 csum_size = sizeof(struct ext4_dir_entry_tail);
2220 blocksize = sb->s_blocksize;
2221 if (!dentry->d_name.len)
2224 if (ext4_has_inline_data(dir)) {
2225 retval = ext4_try_add_inline_entry(handle, dentry, inode);
2235 retval = ext4_dx_add_entry(handle, dentry, inode);
2236 if (!retval || (retval != ERR_BAD_DX_DIR))
2238 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2240 ext4_mark_inode_dirty(handle, dir);
2242 blocks = dir->i_size >> sb->s_blocksize_bits;
2243 for (block = 0; block < blocks; block++) {
2244 bh = ext4_read_dirblock(dir, block, DIRENT);
2248 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
2249 if (retval != -ENOSPC)
2252 if (blocks == 1 && !dx_fallback &&
2253 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
2254 retval = make_indexed_dir(handle, dentry, inode, bh);
2255 bh = NULL; /* make_indexed_dir releases bh */
2260 bh = ext4_append(handle, dir, &block);
2263 de = (struct ext4_dir_entry_2 *) bh->b_data;
2265 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2268 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2269 initialize_dirent_tail(t, blocksize);
2272 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
2276 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2281 * Returns 0 for success, or a negative error value
2283 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
2284 struct inode *inode)
2286 struct dx_frame frames[2], *frame;
2287 struct dx_entry *entries, *at;
2288 struct dx_hash_info hinfo;
2289 struct buffer_head *bh;
2290 struct inode *dir = dentry->d_parent->d_inode;
2291 struct super_block *sb = dir->i_sb;
2292 struct ext4_dir_entry_2 *de;
2295 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames);
2297 return PTR_ERR(frame);
2298 entries = frame->entries;
2300 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2307 BUFFER_TRACE(bh, "get_write_access");
2308 err = ext4_journal_get_write_access(handle, bh);
2312 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
2316 /* Block full, should compress but for now just split */
2317 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2318 dx_get_count(entries), dx_get_limit(entries)));
2319 /* Need to split index? */
2320 if (dx_get_count(entries) == dx_get_limit(entries)) {
2321 ext4_lblk_t newblock;
2322 unsigned icount = dx_get_count(entries);
2323 int levels = frame - frames;
2324 struct dx_entry *entries2;
2325 struct dx_node *node2;
2326 struct buffer_head *bh2;
2328 if (levels && (dx_get_count(frames->entries) ==
2329 dx_get_limit(frames->entries))) {
2330 ext4_warning(sb, "Directory index full!");
2334 bh2 = ext4_append(handle, dir, &newblock);
2339 node2 = (struct dx_node *)(bh2->b_data);
2340 entries2 = node2->entries;
2341 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2342 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2344 BUFFER_TRACE(frame->bh, "get_write_access");
2345 err = ext4_journal_get_write_access(handle, frame->bh);
2349 unsigned icount1 = icount/2, icount2 = icount - icount1;
2350 unsigned hash2 = dx_get_hash(entries + icount1);
2351 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2354 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2355 err = ext4_journal_get_write_access(handle,
2360 memcpy((char *) entries2, (char *) (entries + icount1),
2361 icount2 * sizeof(struct dx_entry));
2362 dx_set_count(entries, icount1);
2363 dx_set_count(entries2, icount2);
2364 dx_set_limit(entries2, dx_node_limit(dir));
2366 /* Which index block gets the new entry? */
2367 if (at - entries >= icount1) {
2368 frame->at = at = at - entries - icount1 + entries2;
2369 frame->entries = entries = entries2;
2370 swap(frame->bh, bh2);
2372 dx_insert_block(frames + 0, hash2, newblock);
2373 dxtrace(dx_show_index("node", frames[1].entries));
2374 dxtrace(dx_show_index("node",
2375 ((struct dx_node *) bh2->b_data)->entries));
2376 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2381 dxtrace(printk(KERN_DEBUG
2382 "Creating second level index...\n"));
2383 memcpy((char *) entries2, (char *) entries,
2384 icount * sizeof(struct dx_entry));
2385 dx_set_limit(entries2, dx_node_limit(dir));
2388 dx_set_count(entries, 1);
2389 dx_set_block(entries + 0, newblock);
2390 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2392 /* Add new access path frame */
2394 frame->at = at = at - entries + entries2;
2395 frame->entries = entries = entries2;
2397 err = ext4_journal_get_write_access(handle,
2402 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2404 ext4_std_error(inode->i_sb, err);
2408 de = do_split(handle, dir, &bh, frame, &hinfo);
2413 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2417 ext4_std_error(dir->i_sb, err);
2425 * ext4_generic_delete_entry deletes a directory entry by merging it
2426 * with the previous entry
2428 int ext4_generic_delete_entry(handle_t *handle,
2430 struct ext4_dir_entry_2 *de_del,
2431 struct buffer_head *bh,
2436 struct ext4_dir_entry_2 *de, *pde;
2437 unsigned int blocksize = dir->i_sb->s_blocksize;
2442 de = (struct ext4_dir_entry_2 *)entry_buf;
2443 while (i < buf_size - csum_size) {
2444 if (ext4_check_dir_entry(dir, NULL, de, bh,
2445 bh->b_data, bh->b_size, i))
2449 pde->rec_len = ext4_rec_len_to_disk(
2450 ext4_rec_len_from_disk(pde->rec_len,
2452 ext4_rec_len_from_disk(de->rec_len,
2460 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2462 de = ext4_next_entry(de, blocksize);
2467 static int ext4_delete_entry(handle_t *handle,
2469 struct ext4_dir_entry_2 *de_del,
2470 struct buffer_head *bh)
2472 int err, csum_size = 0;
2474 if (ext4_has_inline_data(dir)) {
2475 int has_inline_data = 1;
2476 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2478 if (has_inline_data)
2482 if (ext4_has_metadata_csum(dir->i_sb))
2483 csum_size = sizeof(struct ext4_dir_entry_tail);
2485 BUFFER_TRACE(bh, "get_write_access");
2486 err = ext4_journal_get_write_access(handle, bh);
2490 err = ext4_generic_delete_entry(handle, dir, de_del,
2492 dir->i_sb->s_blocksize, csum_size);
2496 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2497 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2504 ext4_std_error(dir->i_sb, err);
2509 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2510 * since this indicates that nlinks count was previously 1.
2512 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2515 if (is_dx(inode) && inode->i_nlink > 1) {
2516 /* limit is 16-bit i_links_count */
2517 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2518 set_nlink(inode, 1);
2519 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2520 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2526 * If a directory had nlink == 1, then we should let it be 1. This indicates
2527 * directory has >EXT4_LINK_MAX subdirs.
2529 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2531 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2536 static int ext4_add_nondir(handle_t *handle,
2537 struct dentry *dentry, struct inode *inode)
2539 int err = ext4_add_entry(handle, dentry, inode);
2541 ext4_mark_inode_dirty(handle, inode);
2542 unlock_new_inode(inode);
2543 d_instantiate(dentry, inode);
2547 unlock_new_inode(inode);
2553 * By the time this is called, we already have created
2554 * the directory cache entry for the new file, but it
2555 * is so far negative - it has no inode.
2557 * If the create succeeds, we fill in the inode information
2558 * with d_instantiate().
2560 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2564 struct inode *inode;
2565 int err, credits, retries = 0;
2567 dquot_initialize(dir);
2569 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2570 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2572 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2573 NULL, EXT4_HT_DIR, credits);
2574 handle = ext4_journal_current_handle();
2575 err = PTR_ERR(inode);
2576 if (!IS_ERR(inode)) {
2577 inode->i_op = &ext4_file_inode_operations;
2578 if (test_opt(inode->i_sb, DAX))
2579 inode->i_fop = &ext4_dax_file_operations;
2581 inode->i_fop = &ext4_file_operations;
2582 ext4_set_aops(inode);
2584 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2585 if (!err && ext4_encrypted_inode(dir)) {
2586 err = ext4_inherit_context(dir, inode);
2589 unlock_new_inode(inode);
2595 err = ext4_add_nondir(handle, dentry, inode);
2596 if (!err && IS_DIRSYNC(dir))
2597 ext4_handle_sync(handle);
2600 ext4_journal_stop(handle);
2601 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2606 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2607 umode_t mode, dev_t rdev)
2610 struct inode *inode;
2611 int err, credits, retries = 0;
2613 if (!new_valid_dev(rdev))
2616 dquot_initialize(dir);
2618 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2619 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2621 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2622 NULL, EXT4_HT_DIR, credits);
2623 handle = ext4_journal_current_handle();
2624 err = PTR_ERR(inode);
2625 if (!IS_ERR(inode)) {
2626 init_special_inode(inode, inode->i_mode, rdev);
2627 inode->i_op = &ext4_special_inode_operations;
2628 err = ext4_add_nondir(handle, dentry, inode);
2629 if (!err && IS_DIRSYNC(dir))
2630 ext4_handle_sync(handle);
2633 ext4_journal_stop(handle);
2634 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2639 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2642 struct inode *inode;
2643 int err, retries = 0;
2645 dquot_initialize(dir);
2648 inode = ext4_new_inode_start_handle(dir, mode,
2651 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2652 4 + EXT4_XATTR_TRANS_BLOCKS);
2653 handle = ext4_journal_current_handle();
2654 err = PTR_ERR(inode);
2655 if (!IS_ERR(inode)) {
2656 inode->i_op = &ext4_file_inode_operations;
2657 if (test_opt(inode->i_sb, DAX))
2658 inode->i_fop = &ext4_dax_file_operations;
2660 inode->i_fop = &ext4_file_operations;
2661 ext4_set_aops(inode);
2662 d_tmpfile(dentry, inode);
2663 err = ext4_orphan_add(handle, inode);
2665 goto err_unlock_inode;
2666 mark_inode_dirty(inode);
2667 unlock_new_inode(inode);
2670 ext4_journal_stop(handle);
2671 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2675 ext4_journal_stop(handle);
2676 unlock_new_inode(inode);
2680 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2681 struct ext4_dir_entry_2 *de,
2682 int blocksize, int csum_size,
2683 unsigned int parent_ino, int dotdot_real_len)
2685 de->inode = cpu_to_le32(inode->i_ino);
2687 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2689 strcpy(de->name, ".");
2690 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2692 de = ext4_next_entry(de, blocksize);
2693 de->inode = cpu_to_le32(parent_ino);
2695 if (!dotdot_real_len)
2696 de->rec_len = ext4_rec_len_to_disk(blocksize -
2697 (csum_size + EXT4_DIR_REC_LEN(1)),
2700 de->rec_len = ext4_rec_len_to_disk(
2701 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2702 strcpy(de->name, "..");
2703 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2705 return ext4_next_entry(de, blocksize);
2708 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2709 struct inode *inode)
2711 struct buffer_head *dir_block = NULL;
2712 struct ext4_dir_entry_2 *de;
2713 struct ext4_dir_entry_tail *t;
2714 ext4_lblk_t block = 0;
2715 unsigned int blocksize = dir->i_sb->s_blocksize;
2719 if (ext4_has_metadata_csum(dir->i_sb))
2720 csum_size = sizeof(struct ext4_dir_entry_tail);
2722 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2723 err = ext4_try_create_inline_dir(handle, dir, inode);
2724 if (err < 0 && err != -ENOSPC)
2731 dir_block = ext4_append(handle, inode, &block);
2732 if (IS_ERR(dir_block))
2733 return PTR_ERR(dir_block);
2734 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2735 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2736 set_nlink(inode, 2);
2738 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2739 initialize_dirent_tail(t, blocksize);
2742 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2743 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2746 set_buffer_verified(dir_block);
2752 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2755 struct inode *inode;
2756 int err, credits, retries = 0;
2758 if (EXT4_DIR_LINK_MAX(dir))
2761 dquot_initialize(dir);
2763 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2764 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2766 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2768 0, NULL, EXT4_HT_DIR, credits);
2769 handle = ext4_journal_current_handle();
2770 err = PTR_ERR(inode);
2774 inode->i_op = &ext4_dir_inode_operations;
2775 inode->i_fop = &ext4_dir_operations;
2776 err = ext4_init_new_dir(handle, dir, inode);
2778 goto out_clear_inode;
2779 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2780 if (ext4_encrypted_inode(dir)) {
2781 err = ext4_inherit_context(dir, inode);
2783 goto out_clear_inode;
2786 err = ext4_mark_inode_dirty(handle, inode);
2788 err = ext4_add_entry(handle, dentry, inode);
2792 unlock_new_inode(inode);
2793 ext4_mark_inode_dirty(handle, inode);
2797 ext4_inc_count(handle, dir);
2798 ext4_update_dx_flag(dir);
2799 err = ext4_mark_inode_dirty(handle, dir);
2801 goto out_clear_inode;
2802 unlock_new_inode(inode);
2803 d_instantiate(dentry, inode);
2804 if (IS_DIRSYNC(dir))
2805 ext4_handle_sync(handle);
2809 ext4_journal_stop(handle);
2810 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2816 * routine to check that the specified directory is empty (for rmdir)
2818 int ext4_empty_dir(struct inode *inode)
2820 unsigned int offset;
2821 struct buffer_head *bh;
2822 struct ext4_dir_entry_2 *de, *de1;
2823 struct super_block *sb;
2826 if (ext4_has_inline_data(inode)) {
2827 int has_inline_data = 1;
2829 err = empty_inline_dir(inode, &has_inline_data);
2830 if (has_inline_data)
2835 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2836 EXT4_ERROR_INODE(inode, "invalid size");
2839 bh = ext4_read_dirblock(inode, 0, EITHER);
2843 de = (struct ext4_dir_entry_2 *) bh->b_data;
2844 de1 = ext4_next_entry(de, sb->s_blocksize);
2845 if (le32_to_cpu(de->inode) != inode->i_ino ||
2846 !le32_to_cpu(de1->inode) ||
2847 strcmp(".", de->name) ||
2848 strcmp("..", de1->name)) {
2849 ext4_warning(inode->i_sb,
2850 "bad directory (dir #%lu) - no `.' or `..'",
2855 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2856 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2857 de = ext4_next_entry(de1, sb->s_blocksize);
2858 while (offset < inode->i_size) {
2859 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2860 unsigned int lblock;
2863 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2864 bh = ext4_read_dirblock(inode, lblock, EITHER);
2867 de = (struct ext4_dir_entry_2 *) bh->b_data;
2869 if (ext4_check_dir_entry(inode, NULL, de, bh,
2870 bh->b_data, bh->b_size, offset)) {
2871 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2873 offset = (offset | (sb->s_blocksize - 1)) + 1;
2876 if (le32_to_cpu(de->inode)) {
2880 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2881 de = ext4_next_entry(de, sb->s_blocksize);
2888 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2889 * such inodes, starting at the superblock, in case we crash before the
2890 * file is closed/deleted, or in case the inode truncate spans multiple
2891 * transactions and the last transaction is not recovered after a crash.
2893 * At filesystem recovery time, we walk this list deleting unlinked
2894 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2896 * Orphan list manipulation functions must be called under i_mutex unless
2897 * we are just creating the inode or deleting it.
2899 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2901 struct super_block *sb = inode->i_sb;
2902 struct ext4_sb_info *sbi = EXT4_SB(sb);
2903 struct ext4_iloc iloc;
2907 if (!sbi->s_journal || is_bad_inode(inode))
2910 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2911 !mutex_is_locked(&inode->i_mutex));
2913 * Exit early if inode already is on orphan list. This is a big speedup
2914 * since we don't have to contend on the global s_orphan_lock.
2916 if (!list_empty(&EXT4_I(inode)->i_orphan))
2920 * Orphan handling is only valid for files with data blocks
2921 * being truncated, or files being unlinked. Note that we either
2922 * hold i_mutex, or the inode can not be referenced from outside,
2923 * so i_nlink should not be bumped due to race
2925 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2926 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2928 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2929 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2933 err = ext4_reserve_inode_write(handle, inode, &iloc);
2937 mutex_lock(&sbi->s_orphan_lock);
2939 * Due to previous errors inode may be already a part of on-disk
2940 * orphan list. If so skip on-disk list modification.
2942 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2943 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2944 /* Insert this inode at the head of the on-disk orphan list */
2945 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2946 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2949 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2950 mutex_unlock(&sbi->s_orphan_lock);
2953 err = ext4_handle_dirty_super(handle, sb);
2954 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2959 * We have to remove inode from in-memory list if
2960 * addition to on disk orphan list failed. Stray orphan
2961 * list entries can cause panics at unmount time.
2963 mutex_lock(&sbi->s_orphan_lock);
2964 list_del(&EXT4_I(inode)->i_orphan);
2965 mutex_unlock(&sbi->s_orphan_lock);
2968 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2969 jbd_debug(4, "orphan inode %lu will point to %d\n",
2970 inode->i_ino, NEXT_ORPHAN(inode));
2972 ext4_std_error(sb, err);
2977 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2978 * of such inodes stored on disk, because it is finally being cleaned up.
2980 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2982 struct list_head *prev;
2983 struct ext4_inode_info *ei = EXT4_I(inode);
2984 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2986 struct ext4_iloc iloc;
2989 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2992 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2993 !mutex_is_locked(&inode->i_mutex));
2994 /* Do this quick check before taking global s_orphan_lock. */
2995 if (list_empty(&ei->i_orphan))
2999 /* Grab inode buffer early before taking global s_orphan_lock */
3000 err = ext4_reserve_inode_write(handle, inode, &iloc);
3003 mutex_lock(&sbi->s_orphan_lock);
3004 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3006 prev = ei->i_orphan.prev;
3007 list_del_init(&ei->i_orphan);
3009 /* If we're on an error path, we may not have a valid
3010 * transaction handle with which to update the orphan list on
3011 * disk, but we still need to remove the inode from the linked
3012 * list in memory. */
3013 if (!handle || err) {
3014 mutex_unlock(&sbi->s_orphan_lock);
3018 ino_next = NEXT_ORPHAN(inode);
3019 if (prev == &sbi->s_orphan) {
3020 jbd_debug(4, "superblock will point to %u\n", ino_next);
3021 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3022 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3024 mutex_unlock(&sbi->s_orphan_lock);
3027 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3028 mutex_unlock(&sbi->s_orphan_lock);
3029 err = ext4_handle_dirty_super(handle, inode->i_sb);
3031 struct ext4_iloc iloc2;
3032 struct inode *i_prev =
3033 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3035 jbd_debug(4, "orphan inode %lu will point to %u\n",
3036 i_prev->i_ino, ino_next);
3037 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3039 mutex_unlock(&sbi->s_orphan_lock);
3042 NEXT_ORPHAN(i_prev) = ino_next;
3043 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3044 mutex_unlock(&sbi->s_orphan_lock);
3048 NEXT_ORPHAN(inode) = 0;
3049 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3051 ext4_std_error(inode->i_sb, err);
3059 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3062 struct inode *inode;
3063 struct buffer_head *bh;
3064 struct ext4_dir_entry_2 *de;
3065 handle_t *handle = NULL;
3067 /* Initialize quotas before so that eventual writes go in
3068 * separate transaction */
3069 dquot_initialize(dir);
3070 dquot_initialize(dentry->d_inode);
3073 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3079 inode = dentry->d_inode;
3082 if (le32_to_cpu(de->inode) != inode->i_ino)
3085 retval = -ENOTEMPTY;
3086 if (!ext4_empty_dir(inode))
3089 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3090 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3091 if (IS_ERR(handle)) {
3092 retval = PTR_ERR(handle);
3097 if (IS_DIRSYNC(dir))
3098 ext4_handle_sync(handle);
3100 retval = ext4_delete_entry(handle, dir, de, bh);
3103 if (!EXT4_DIR_LINK_EMPTY(inode))
3104 ext4_warning(inode->i_sb,
3105 "empty directory has too many links (%d)",
3109 /* There's no need to set i_disksize: the fact that i_nlink is
3110 * zero will ensure that the right thing happens during any
3113 ext4_orphan_add(handle, inode);
3114 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
3115 ext4_mark_inode_dirty(handle, inode);
3116 ext4_dec_count(handle, dir);
3117 ext4_update_dx_flag(dir);
3118 ext4_mark_inode_dirty(handle, dir);
3123 ext4_journal_stop(handle);
3127 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3130 struct inode *inode;
3131 struct buffer_head *bh;
3132 struct ext4_dir_entry_2 *de;
3133 handle_t *handle = NULL;
3135 trace_ext4_unlink_enter(dir, dentry);
3136 /* Initialize quotas before so that eventual writes go
3137 * in separate transaction */
3138 dquot_initialize(dir);
3139 dquot_initialize(dentry->d_inode);
3142 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3148 inode = dentry->d_inode;
3151 if (le32_to_cpu(de->inode) != inode->i_ino)
3154 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3155 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3156 if (IS_ERR(handle)) {
3157 retval = PTR_ERR(handle);
3162 if (IS_DIRSYNC(dir))
3163 ext4_handle_sync(handle);
3165 if (!inode->i_nlink) {
3166 ext4_warning(inode->i_sb,
3167 "Deleting nonexistent file (%lu), %d",
3168 inode->i_ino, inode->i_nlink);
3169 set_nlink(inode, 1);
3171 retval = ext4_delete_entry(handle, dir, de, bh);
3174 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3175 ext4_update_dx_flag(dir);
3176 ext4_mark_inode_dirty(handle, dir);
3178 if (!inode->i_nlink)
3179 ext4_orphan_add(handle, inode);
3180 inode->i_ctime = ext4_current_time(inode);
3181 ext4_mark_inode_dirty(handle, inode);
3186 ext4_journal_stop(handle);
3187 trace_ext4_unlink_exit(dentry, retval);
3191 static int ext4_symlink(struct inode *dir,
3192 struct dentry *dentry, const char *symname)
3195 struct inode *inode;
3196 int err, len = strlen(symname);
3198 bool encryption_required;
3199 struct ext4_str disk_link;
3200 struct ext4_encrypted_symlink_data *sd = NULL;
3202 disk_link.len = len + 1;
3203 disk_link.name = (char *) symname;
3205 encryption_required = ext4_encrypted_inode(dir);
3206 if (encryption_required)
3207 disk_link.len = encrypted_symlink_data_len(len) + 1;
3208 if (disk_link.len > dir->i_sb->s_blocksize)
3209 return -ENAMETOOLONG;
3211 dquot_initialize(dir);
3213 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3215 * For non-fast symlinks, we just allocate inode and put it on
3216 * orphan list in the first transaction => we need bitmap,
3217 * group descriptor, sb, inode block, quota blocks, and
3218 * possibly selinux xattr blocks.
3220 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3221 EXT4_XATTR_TRANS_BLOCKS;
3224 * Fast symlink. We have to add entry to directory
3225 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3226 * allocate new inode (bitmap, group descriptor, inode block,
3227 * quota blocks, sb is already counted in previous macros).
3229 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3230 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3233 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3234 &dentry->d_name, 0, NULL,
3235 EXT4_HT_DIR, credits);
3236 handle = ext4_journal_current_handle();
3237 if (IS_ERR(inode)) {
3239 ext4_journal_stop(handle);
3240 return PTR_ERR(inode);
3243 if (encryption_required) {
3244 struct ext4_fname_crypto_ctx *ctx = NULL;
3246 struct ext4_str ostr;
3248 sd = kzalloc(disk_link.len, GFP_NOFS);
3251 goto err_drop_inode;
3253 err = ext4_inherit_context(dir, inode);
3255 goto err_drop_inode;
3256 ctx = ext4_get_fname_crypto_ctx(inode,
3257 inode->i_sb->s_blocksize);
3258 if (IS_ERR_OR_NULL(ctx)) {
3259 /* We just set the policy, so ctx should not be NULL */
3260 err = (ctx == NULL) ? -EIO : PTR_ERR(ctx);
3261 goto err_drop_inode;
3263 istr.name = (const unsigned char *) symname;
3265 ostr.name = sd->encrypted_path;
3266 err = ext4_fname_usr_to_disk(ctx, &istr, &ostr);
3267 ext4_put_fname_crypto_ctx(&ctx);
3269 goto err_drop_inode;
3270 sd->len = cpu_to_le16(ostr.len);
3271 disk_link.name = (char *) sd;
3274 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3275 inode->i_op = &ext4_symlink_inode_operations;
3276 ext4_set_aops(inode);
3278 * We cannot call page_symlink() with transaction started
3279 * because it calls into ext4_write_begin() which can wait
3280 * for transaction commit if we are running out of space
3281 * and thus we deadlock. So we have to stop transaction now
3282 * and restart it when symlink contents is written.
3284 * To keep fs consistent in case of crash, we have to put inode
3285 * to orphan list in the mean time.
3288 err = ext4_orphan_add(handle, inode);
3289 ext4_journal_stop(handle);
3292 goto err_drop_inode;
3293 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3295 goto err_drop_inode;
3297 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3298 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3300 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3301 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3302 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3303 if (IS_ERR(handle)) {
3304 err = PTR_ERR(handle);
3306 goto err_drop_inode;
3308 set_nlink(inode, 1);
3309 err = ext4_orphan_del(handle, inode);
3311 goto err_drop_inode;
3313 /* clear the extent format for fast symlink */
3314 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3315 inode->i_op = encryption_required ?
3316 &ext4_symlink_inode_operations :
3317 &ext4_fast_symlink_inode_operations;
3318 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3320 inode->i_size = disk_link.len - 1;
3322 EXT4_I(inode)->i_disksize = inode->i_size;
3323 err = ext4_add_nondir(handle, dentry, inode);
3324 if (!err && IS_DIRSYNC(dir))
3325 ext4_handle_sync(handle);
3328 ext4_journal_stop(handle);
3333 ext4_journal_stop(handle);
3336 unlock_new_inode(inode);
3341 static int ext4_link(struct dentry *old_dentry,
3342 struct inode *dir, struct dentry *dentry)
3345 struct inode *inode = old_dentry->d_inode;
3346 int err, retries = 0;
3348 if (inode->i_nlink >= EXT4_LINK_MAX)
3350 if (ext4_encrypted_inode(dir) &&
3351 !ext4_is_child_context_consistent_with_parent(dir, inode))
3353 dquot_initialize(dir);
3356 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3357 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3358 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3360 return PTR_ERR(handle);
3362 if (IS_DIRSYNC(dir))
3363 ext4_handle_sync(handle);
3365 inode->i_ctime = ext4_current_time(inode);
3366 ext4_inc_count(handle, inode);
3369 err = ext4_add_entry(handle, dentry, inode);
3371 ext4_mark_inode_dirty(handle, inode);
3372 /* this can happen only for tmpfile being
3373 * linked the first time
3375 if (inode->i_nlink == 1)
3376 ext4_orphan_del(handle, inode);
3377 d_instantiate(dentry, inode);
3382 ext4_journal_stop(handle);
3383 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3390 * Try to find buffer head where contains the parent block.
3391 * It should be the inode block if it is inlined or the 1st block
3392 * if it is a normal dir.
3394 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3395 struct inode *inode,
3397 struct ext4_dir_entry_2 **parent_de,
3400 struct buffer_head *bh;
3402 if (!ext4_has_inline_data(inode)) {
3403 bh = ext4_read_dirblock(inode, 0, EITHER);
3405 *retval = PTR_ERR(bh);
3408 *parent_de = ext4_next_entry(
3409 (struct ext4_dir_entry_2 *)bh->b_data,
3410 inode->i_sb->s_blocksize);
3415 return ext4_get_first_inline_block(inode, parent_de, retval);
3418 struct ext4_renament {
3420 struct dentry *dentry;
3421 struct inode *inode;
3423 int dir_nlink_delta;
3425 /* entry for "dentry" */
3426 struct buffer_head *bh;
3427 struct ext4_dir_entry_2 *de;
3430 /* entry for ".." in inode if it's a directory */
3431 struct buffer_head *dir_bh;
3432 struct ext4_dir_entry_2 *parent_de;
3436 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3440 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3441 &retval, &ent->parent_de,
3445 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3447 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3448 return ext4_journal_get_write_access(handle, ent->dir_bh);
3451 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3456 ent->parent_de->inode = cpu_to_le32(dir_ino);
3457 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3458 if (!ent->dir_inlined) {
3459 if (is_dx(ent->inode)) {
3460 retval = ext4_handle_dirty_dx_node(handle,
3464 retval = ext4_handle_dirty_dirent_node(handle,
3469 retval = ext4_mark_inode_dirty(handle, ent->inode);
3472 ext4_std_error(ent->dir->i_sb, retval);
3478 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3479 unsigned ino, unsigned file_type)
3483 BUFFER_TRACE(ent->bh, "get write access");
3484 retval = ext4_journal_get_write_access(handle, ent->bh);
3487 ent->de->inode = cpu_to_le32(ino);
3488 if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb,
3489 EXT4_FEATURE_INCOMPAT_FILETYPE))
3490 ent->de->file_type = file_type;
3491 ent->dir->i_version++;
3492 ent->dir->i_ctime = ent->dir->i_mtime =
3493 ext4_current_time(ent->dir);
3494 ext4_mark_inode_dirty(handle, ent->dir);
3495 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3496 if (!ent->inlined) {
3497 retval = ext4_handle_dirty_dirent_node(handle,
3499 if (unlikely(retval)) {
3500 ext4_std_error(ent->dir->i_sb, retval);
3510 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3511 const struct qstr *d_name)
3513 int retval = -ENOENT;
3514 struct buffer_head *bh;
3515 struct ext4_dir_entry_2 *de;
3517 bh = ext4_find_entry(dir, d_name, &de, NULL);
3521 retval = ext4_delete_entry(handle, dir, de, bh);
3527 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3532 * ent->de could have moved from under us during htree split, so make
3533 * sure that we are deleting the right entry. We might also be pointing
3534 * to a stale entry in the unused part of ent->bh so just checking inum
3535 * and the name isn't enough.
3537 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3538 ent->de->name_len != ent->dentry->d_name.len ||
3539 strncmp(ent->de->name, ent->dentry->d_name.name,
3540 ent->de->name_len) ||
3542 retval = ext4_find_delete_entry(handle, ent->dir,
3543 &ent->dentry->d_name);
3545 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3546 if (retval == -ENOENT) {
3547 retval = ext4_find_delete_entry(handle, ent->dir,
3548 &ent->dentry->d_name);
3553 ext4_warning(ent->dir->i_sb,
3554 "Deleting old file (%lu), %d, error=%d",
3555 ent->dir->i_ino, ent->dir->i_nlink, retval);
3559 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3561 if (ent->dir_nlink_delta) {
3562 if (ent->dir_nlink_delta == -1)
3563 ext4_dec_count(handle, ent->dir);
3565 ext4_inc_count(handle, ent->dir);
3566 ext4_mark_inode_dirty(handle, ent->dir);
3570 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3571 int credits, handle_t **h)
3578 * for inode block, sb block, group summaries,
3581 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3582 EXT4_XATTR_TRANS_BLOCKS + 4);
3584 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3585 &ent->dentry->d_name, 0, NULL,
3586 EXT4_HT_DIR, credits);
3588 handle = ext4_journal_current_handle();
3591 ext4_journal_stop(handle);
3592 if (PTR_ERR(wh) == -ENOSPC &&
3593 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3597 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3598 wh->i_op = &ext4_special_inode_operations;
3604 * Anybody can rename anything with this: the permission checks are left to the
3605 * higher-level routines.
3607 * n.b. old_{dentry,inode) refers to the source dentry/inode
3608 * while new_{dentry,inode) refers to the destination dentry/inode
3609 * This comes from rename(const char *oldpath, const char *newpath)
3611 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3612 struct inode *new_dir, struct dentry *new_dentry,
3615 handle_t *handle = NULL;
3616 struct ext4_renament old = {
3618 .dentry = old_dentry,
3619 .inode = old_dentry->d_inode,
3621 struct ext4_renament new = {
3623 .dentry = new_dentry,
3624 .inode = new_dentry->d_inode,
3628 struct inode *whiteout = NULL;
3632 dquot_initialize(old.dir);
3633 dquot_initialize(new.dir);
3635 /* Initialize quotas before so that eventual writes go
3636 * in separate transaction */
3638 dquot_initialize(new.inode);
3640 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3642 return PTR_ERR(old.bh);
3644 * Check for inode number is _not_ due to possible IO errors.
3645 * We might rmdir the source, keep it as pwd of some process
3646 * and merrily kill the link to whatever was created under the
3647 * same name. Goodbye sticky bit ;-<
3650 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3653 if ((old.dir != new.dir) &&
3654 ext4_encrypted_inode(new.dir) &&
3655 !ext4_is_child_context_consistent_with_parent(new.dir,
3661 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3662 &new.de, &new.inlined);
3663 if (IS_ERR(new.bh)) {
3664 retval = PTR_ERR(new.bh);
3674 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3675 ext4_alloc_da_blocks(old.inode);
3677 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3678 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3679 if (!(flags & RENAME_WHITEOUT)) {
3680 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3681 if (IS_ERR(handle)) {
3682 retval = PTR_ERR(handle);
3687 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3688 if (IS_ERR(whiteout)) {
3689 retval = PTR_ERR(whiteout);
3695 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3696 ext4_handle_sync(handle);
3698 if (S_ISDIR(old.inode->i_mode)) {
3700 retval = -ENOTEMPTY;
3701 if (!ext4_empty_dir(new.inode))
3705 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3708 retval = ext4_rename_dir_prepare(handle, &old);
3713 * If we're renaming a file within an inline_data dir and adding or
3714 * setting the new dirent causes a conversion from inline_data to
3715 * extents/blockmap, we need to force the dirent delete code to
3716 * re-read the directory, or else we end up trying to delete a dirent
3717 * from what is now the extent tree root (or a block map).
3719 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3720 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3722 old_file_type = old.de->file_type;
3725 * Do this before adding a new entry, so the old entry is sure
3726 * to be still pointing to the valid old entry.
3728 retval = ext4_setent(handle, &old, whiteout->i_ino,
3732 ext4_mark_inode_dirty(handle, whiteout);
3735 retval = ext4_add_entry(handle, new.dentry, old.inode);
3739 retval = ext4_setent(handle, &new,
3740 old.inode->i_ino, old_file_type);
3745 force_reread = !ext4_test_inode_flag(new.dir,
3746 EXT4_INODE_INLINE_DATA);
3749 * Like most other Unix systems, set the ctime for inodes on a
3752 old.inode->i_ctime = ext4_current_time(old.inode);
3753 ext4_mark_inode_dirty(handle, old.inode);
3759 ext4_rename_delete(handle, &old, force_reread);
3763 ext4_dec_count(handle, new.inode);
3764 new.inode->i_ctime = ext4_current_time(new.inode);
3766 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3767 ext4_update_dx_flag(old.dir);
3769 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3773 ext4_dec_count(handle, old.dir);
3775 /* checked ext4_empty_dir above, can't have another
3776 * parent, ext4_dec_count() won't work for many-linked
3778 clear_nlink(new.inode);
3780 ext4_inc_count(handle, new.dir);
3781 ext4_update_dx_flag(new.dir);
3782 ext4_mark_inode_dirty(handle, new.dir);
3785 ext4_mark_inode_dirty(handle, old.dir);
3787 ext4_mark_inode_dirty(handle, new.inode);
3788 if (!new.inode->i_nlink)
3789 ext4_orphan_add(handle, new.inode);
3799 drop_nlink(whiteout);
3800 unlock_new_inode(whiteout);
3804 ext4_journal_stop(handle);
3808 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3809 struct inode *new_dir, struct dentry *new_dentry)
3811 handle_t *handle = NULL;
3812 struct ext4_renament old = {
3814 .dentry = old_dentry,
3815 .inode = old_dentry->d_inode,
3817 struct ext4_renament new = {
3819 .dentry = new_dentry,
3820 .inode = new_dentry->d_inode,
3825 dquot_initialize(old.dir);
3826 dquot_initialize(new.dir);
3828 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3829 &old.de, &old.inlined);
3831 return PTR_ERR(old.bh);
3833 * Check for inode number is _not_ due to possible IO errors.
3834 * We might rmdir the source, keep it as pwd of some process
3835 * and merrily kill the link to whatever was created under the
3836 * same name. Goodbye sticky bit ;-<
3839 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3842 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3843 &new.de, &new.inlined);
3844 if (IS_ERR(new.bh)) {
3845 retval = PTR_ERR(new.bh);
3850 /* RENAME_EXCHANGE case: old *and* new must both exist */
3851 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3854 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3855 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3856 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3857 if (IS_ERR(handle)) {
3858 retval = PTR_ERR(handle);
3863 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3864 ext4_handle_sync(handle);
3866 if (S_ISDIR(old.inode->i_mode)) {
3868 retval = ext4_rename_dir_prepare(handle, &old);
3872 if (S_ISDIR(new.inode->i_mode)) {
3874 retval = ext4_rename_dir_prepare(handle, &new);
3880 * Other than the special case of overwriting a directory, parents'
3881 * nlink only needs to be modified if this is a cross directory rename.
3883 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3884 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3885 new.dir_nlink_delta = -old.dir_nlink_delta;
3887 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3888 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3892 new_file_type = new.de->file_type;
3893 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3897 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3902 * Like most other Unix systems, set the ctime for inodes on a
3905 old.inode->i_ctime = ext4_current_time(old.inode);
3906 new.inode->i_ctime = ext4_current_time(new.inode);
3907 ext4_mark_inode_dirty(handle, old.inode);
3908 ext4_mark_inode_dirty(handle, new.inode);
3911 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3916 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3920 ext4_update_dir_count(handle, &old);
3921 ext4_update_dir_count(handle, &new);
3930 ext4_journal_stop(handle);
3934 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3935 struct inode *new_dir, struct dentry *new_dentry,
3938 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3941 if (flags & RENAME_EXCHANGE) {
3942 return ext4_cross_rename(old_dir, old_dentry,
3943 new_dir, new_dentry);
3946 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3950 * directories can handle most operations...
3952 const struct inode_operations ext4_dir_inode_operations = {
3953 .create = ext4_create,
3954 .lookup = ext4_lookup,
3956 .unlink = ext4_unlink,
3957 .symlink = ext4_symlink,
3958 .mkdir = ext4_mkdir,
3959 .rmdir = ext4_rmdir,
3960 .mknod = ext4_mknod,
3961 .tmpfile = ext4_tmpfile,
3962 .rename2 = ext4_rename2,
3963 .setattr = ext4_setattr,
3964 .setxattr = generic_setxattr,
3965 .getxattr = generic_getxattr,
3966 .listxattr = ext4_listxattr,
3967 .removexattr = generic_removexattr,
3968 .get_acl = ext4_get_acl,
3969 .set_acl = ext4_set_acl,
3970 .fiemap = ext4_fiemap,
3973 const struct inode_operations ext4_special_inode_operations = {
3974 .setattr = ext4_setattr,
3975 .setxattr = generic_setxattr,
3976 .getxattr = generic_getxattr,
3977 .listxattr = ext4_listxattr,
3978 .removexattr = generic_removexattr,
3979 .get_acl = ext4_get_acl,
3980 .set_acl = ext4_set_acl,