2 * linux/fs/ext4/namei.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/namei.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
28 #include <linux/pagemap.h>
29 #include <linux/time.h>
30 #include <linux/fcntl.h>
31 #include <linux/stat.h>
32 #include <linux/string.h>
33 #include <linux/quotaops.h>
34 #include <linux/buffer_head.h>
35 #include <linux/bio.h>
37 #include "ext4_jbd2.h"
42 #include <trace/events/ext4.h>
44 * define how far ahead to read directories while searching them.
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 static struct buffer_head *ext4_append(handle_t *handle,
54 struct buffer_head *bh;
57 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
58 ((inode->i_size >> 10) >=
59 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
60 return ERR_PTR(-ENOSPC);
62 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
64 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
67 inode->i_size += inode->i_sb->s_blocksize;
68 EXT4_I(inode)->i_disksize = inode->i_size;
69 BUFFER_TRACE(bh, "get_write_access");
70 err = ext4_journal_get_write_access(handle, bh);
73 ext4_std_error(inode->i_sb, err);
79 static int ext4_dx_csum_verify(struct inode *inode,
80 struct ext4_dir_entry *dirent);
86 #define ext4_read_dirblock(inode, block, type) \
87 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
95 struct buffer_head *bh;
96 struct ext4_dir_entry *dirent;
99 bh = ext4_bread(NULL, inode, block, 0);
101 __ext4_warning(inode->i_sb, func, line,
102 "inode #%lu: lblock %lu: comm %s: "
103 "error %ld reading directory block",
104 inode->i_ino, (unsigned long)block,
105 current->comm, PTR_ERR(bh));
110 ext4_error_inode(inode, func, line, block,
111 "Directory hole found");
112 return ERR_PTR(-EFSCORRUPTED);
114 dirent = (struct ext4_dir_entry *) bh->b_data;
115 /* Determine whether or not we have an index block */
119 else if (ext4_rec_len_from_disk(dirent->rec_len,
120 inode->i_sb->s_blocksize) ==
121 inode->i_sb->s_blocksize)
124 if (!is_dx_block && type == INDEX) {
125 ext4_error_inode(inode, func, line, block,
126 "directory leaf block found instead of index block");
128 return ERR_PTR(-EFSCORRUPTED);
130 if (!ext4_has_metadata_csum(inode->i_sb) ||
135 * An empty leaf block can get mistaken for a index block; for
136 * this reason, we can only check the index checksum when the
137 * caller is sure it should be an index block.
139 if (is_dx_block && type == INDEX) {
140 if (ext4_dx_csum_verify(inode, dirent))
141 set_buffer_verified(bh);
143 ext4_error_inode(inode, func, line, block,
144 "Directory index failed checksum");
146 return ERR_PTR(-EFSBADCRC);
150 if (ext4_dirent_csum_verify(inode, dirent))
151 set_buffer_verified(bh);
153 ext4_error_inode(inode, func, line, block,
154 "Directory block failed checksum");
156 return ERR_PTR(-EFSBADCRC);
163 #define assert(test) J_ASSERT(test)
167 #define dxtrace(command) command
169 #define dxtrace(command)
193 * dx_root_info is laid out so that if it should somehow get overlaid by a
194 * dirent the two low bits of the hash version will be zero. Therefore, the
195 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
200 struct fake_dirent dot;
202 struct fake_dirent dotdot;
206 __le32 reserved_zero;
208 u8 info_length; /* 8 */
213 struct dx_entry entries[0];
218 struct fake_dirent fake;
219 struct dx_entry entries[0];
225 struct buffer_head *bh;
226 struct dx_entry *entries;
238 * This goes at the end of each htree block.
242 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
245 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
246 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
247 static inline unsigned dx_get_hash(struct dx_entry *entry);
248 static void dx_set_hash(struct dx_entry *entry, unsigned value);
249 static unsigned dx_get_count(struct dx_entry *entries);
250 static unsigned dx_get_limit(struct dx_entry *entries);
251 static void dx_set_count(struct dx_entry *entries, unsigned value);
252 static void dx_set_limit(struct dx_entry *entries, unsigned value);
253 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
254 static unsigned dx_node_limit(struct inode *dir);
255 static struct dx_frame *dx_probe(struct ext4_filename *fname,
257 struct dx_hash_info *hinfo,
258 struct dx_frame *frame);
259 static void dx_release(struct dx_frame *frames);
260 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
261 unsigned blocksize, struct dx_hash_info *hinfo,
262 struct dx_map_entry map[]);
263 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
264 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
265 struct dx_map_entry *offsets, int count, unsigned blocksize);
266 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
267 static void dx_insert_block(struct dx_frame *frame,
268 u32 hash, ext4_lblk_t block);
269 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
270 struct dx_frame *frame,
271 struct dx_frame *frames,
273 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
274 struct ext4_filename *fname,
275 struct ext4_dir_entry_2 **res_dir);
276 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
277 struct dentry *dentry, struct inode *inode);
279 /* checksumming functions */
280 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
281 unsigned int blocksize)
283 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
284 t->det_rec_len = ext4_rec_len_to_disk(
285 sizeof(struct ext4_dir_entry_tail), blocksize);
286 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
289 /* Walk through a dirent block to find a checksum "dirent" at the tail */
290 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
291 struct ext4_dir_entry *de)
293 struct ext4_dir_entry_tail *t;
296 struct ext4_dir_entry *d, *top;
299 top = (struct ext4_dir_entry *)(((void *)de) +
300 (EXT4_BLOCK_SIZE(inode->i_sb) -
301 sizeof(struct ext4_dir_entry_tail)));
302 while (d < top && d->rec_len)
303 d = (struct ext4_dir_entry *)(((void *)d) +
304 le16_to_cpu(d->rec_len));
309 t = (struct ext4_dir_entry_tail *)d;
311 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
314 if (t->det_reserved_zero1 ||
315 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
316 t->det_reserved_zero2 ||
317 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
323 static __le32 ext4_dirent_csum(struct inode *inode,
324 struct ext4_dir_entry *dirent, int size)
326 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
327 struct ext4_inode_info *ei = EXT4_I(inode);
330 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
331 return cpu_to_le32(csum);
334 #define warn_no_space_for_csum(inode) \
335 __warn_no_space_for_csum((inode), __func__, __LINE__)
337 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
340 __ext4_warning_inode(inode, func, line,
341 "No space for directory leaf checksum. Please run e2fsck -D.");
344 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
346 struct ext4_dir_entry_tail *t;
348 if (!ext4_has_metadata_csum(inode->i_sb))
351 t = get_dirent_tail(inode, dirent);
353 warn_no_space_for_csum(inode);
357 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
358 (void *)t - (void *)dirent))
364 static void ext4_dirent_csum_set(struct inode *inode,
365 struct ext4_dir_entry *dirent)
367 struct ext4_dir_entry_tail *t;
369 if (!ext4_has_metadata_csum(inode->i_sb))
372 t = get_dirent_tail(inode, dirent);
374 warn_no_space_for_csum(inode);
378 t->det_checksum = ext4_dirent_csum(inode, dirent,
379 (void *)t - (void *)dirent);
382 int ext4_handle_dirty_dirent_node(handle_t *handle,
384 struct buffer_head *bh)
386 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
387 return ext4_handle_dirty_metadata(handle, inode, bh);
390 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
391 struct ext4_dir_entry *dirent,
394 struct ext4_dir_entry *dp;
395 struct dx_root_info *root;
398 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
400 else if (le16_to_cpu(dirent->rec_len) == 12) {
401 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
402 if (le16_to_cpu(dp->rec_len) !=
403 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
405 root = (struct dx_root_info *)(((void *)dp + 12));
406 if (root->reserved_zero ||
407 root->info_length != sizeof(struct dx_root_info))
414 *offset = count_offset;
415 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
418 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
419 int count_offset, int count, struct dx_tail *t)
421 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
422 struct ext4_inode_info *ei = EXT4_I(inode);
425 __u32 dummy_csum = 0;
426 int offset = offsetof(struct dx_tail, dt_checksum);
428 size = count_offset + (count * sizeof(struct dx_entry));
429 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
430 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
431 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
433 return cpu_to_le32(csum);
436 static int ext4_dx_csum_verify(struct inode *inode,
437 struct ext4_dir_entry *dirent)
439 struct dx_countlimit *c;
441 int count_offset, limit, count;
443 if (!ext4_has_metadata_csum(inode->i_sb))
446 c = get_dx_countlimit(inode, dirent, &count_offset);
448 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
451 limit = le16_to_cpu(c->limit);
452 count = le16_to_cpu(c->count);
453 if (count_offset + (limit * sizeof(struct dx_entry)) >
454 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
455 warn_no_space_for_csum(inode);
458 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
460 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
466 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
468 struct dx_countlimit *c;
470 int count_offset, limit, count;
472 if (!ext4_has_metadata_csum(inode->i_sb))
475 c = get_dx_countlimit(inode, dirent, &count_offset);
477 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
480 limit = le16_to_cpu(c->limit);
481 count = le16_to_cpu(c->count);
482 if (count_offset + (limit * sizeof(struct dx_entry)) >
483 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
484 warn_no_space_for_csum(inode);
487 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
489 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
492 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
494 struct buffer_head *bh)
496 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
497 return ext4_handle_dirty_metadata(handle, inode, bh);
501 * p is at least 6 bytes before the end of page
503 static inline struct ext4_dir_entry_2 *
504 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
506 return (struct ext4_dir_entry_2 *)((char *)p +
507 ext4_rec_len_from_disk(p->rec_len, blocksize));
511 * Future: use high four bits of block for coalesce-on-delete flags
512 * Mask them off for now.
515 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
517 return le32_to_cpu(entry->block) & 0x00ffffff;
520 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
522 entry->block = cpu_to_le32(value);
525 static inline unsigned dx_get_hash(struct dx_entry *entry)
527 return le32_to_cpu(entry->hash);
530 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
532 entry->hash = cpu_to_le32(value);
535 static inline unsigned dx_get_count(struct dx_entry *entries)
537 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
540 static inline unsigned dx_get_limit(struct dx_entry *entries)
542 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
545 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
547 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
550 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
552 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
555 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
557 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
558 EXT4_DIR_REC_LEN(2) - infosize;
560 if (ext4_has_metadata_csum(dir->i_sb))
561 entry_space -= sizeof(struct dx_tail);
562 return entry_space / sizeof(struct dx_entry);
565 static inline unsigned dx_node_limit(struct inode *dir)
567 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
569 if (ext4_has_metadata_csum(dir->i_sb))
570 entry_space -= sizeof(struct dx_tail);
571 return entry_space / sizeof(struct dx_entry);
578 static void dx_show_index(char * label, struct dx_entry *entries)
580 int i, n = dx_get_count (entries);
581 printk(KERN_DEBUG "%s index ", label);
582 for (i = 0; i < n; i++) {
583 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
584 0, (unsigned long)dx_get_block(entries + i));
596 static struct stats dx_show_leaf(struct inode *dir,
597 struct dx_hash_info *hinfo,
598 struct ext4_dir_entry_2 *de,
599 int size, int show_names)
601 unsigned names = 0, space = 0;
602 char *base = (char *) de;
603 struct dx_hash_info h = *hinfo;
606 while ((char *) de < base + size)
612 #ifdef CONFIG_EXT4_FS_ENCRYPTION
615 struct ext4_str fname_crypto_str
616 = {.name = NULL, .len = 0};
621 if (ext4_encrypted_inode(inode))
622 res = ext4_get_encryption_info(dir);
624 printk(KERN_WARNING "Error setting up"
625 " fname crypto: %d\n", res);
628 /* Directory is not encrypted */
629 ext4fs_dirhash(de->name,
631 printk("%*.s:(U)%x.%u ", len,
633 (unsigned) ((char *) de
636 /* Directory is encrypted */
637 res = ext4_fname_crypto_alloc_buffer(
641 printk(KERN_WARNING "Error "
647 res = ext4_fname_disk_to_usr(ctx, NULL, de,
650 printk(KERN_WARNING "Error "
651 "converting filename "
657 name = fname_crypto_str.name;
658 len = fname_crypto_str.len;
660 ext4fs_dirhash(de->name, de->name_len,
662 printk("%*.s:(E)%x.%u ", len, name,
663 h.hash, (unsigned) ((char *) de
665 ext4_fname_crypto_free_buffer(
669 int len = de->name_len;
670 char *name = de->name;
671 ext4fs_dirhash(de->name, de->name_len, &h);
672 printk("%*.s:%x.%u ", len, name, h.hash,
673 (unsigned) ((char *) de - base));
676 space += EXT4_DIR_REC_LEN(de->name_len);
679 de = ext4_next_entry(de, size);
681 printk("(%i)\n", names);
682 return (struct stats) { names, space, 1 };
685 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
686 struct dx_entry *entries, int levels)
688 unsigned blocksize = dir->i_sb->s_blocksize;
689 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
691 struct buffer_head *bh;
692 printk("%i indexed blocks...\n", count);
693 for (i = 0; i < count; i++, entries++)
695 ext4_lblk_t block = dx_get_block(entries);
696 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
697 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
699 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
700 bh = ext4_bread(NULL,dir, block, 0);
701 if (!bh || IS_ERR(bh))
704 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
705 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
706 bh->b_data, blocksize, 0);
707 names += stats.names;
708 space += stats.space;
709 bcount += stats.bcount;
713 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
714 levels ? "" : " ", names, space/bcount,
715 (space/bcount)*100/blocksize);
716 return (struct stats) { names, space, bcount};
718 #endif /* DX_DEBUG */
721 * Probe for a directory leaf block to search.
723 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
724 * error in the directory index, and the caller should fall back to
725 * searching the directory normally. The callers of dx_probe **MUST**
726 * check for this error code, and make sure it never gets reflected
729 static struct dx_frame *
730 dx_probe(struct ext4_filename *fname, struct inode *dir,
731 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
733 unsigned count, indirect;
734 struct dx_entry *at, *entries, *p, *q, *m;
735 struct dx_root *root;
736 struct dx_frame *frame = frame_in;
737 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
740 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
741 if (IS_ERR(frame->bh))
742 return (struct dx_frame *) frame->bh;
744 root = (struct dx_root *) frame->bh->b_data;
745 if (root->info.hash_version != DX_HASH_TEA &&
746 root->info.hash_version != DX_HASH_HALF_MD4 &&
747 root->info.hash_version != DX_HASH_LEGACY) {
748 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
749 root->info.hash_version);
753 hinfo = &fname->hinfo;
754 hinfo->hash_version = root->info.hash_version;
755 if (hinfo->hash_version <= DX_HASH_TEA)
756 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
757 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
758 if (fname && fname_name(fname))
759 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
762 if (root->info.unused_flags & 1) {
763 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
764 root->info.unused_flags);
768 indirect = root->info.indirect_levels;
770 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
771 root->info.indirect_levels);
775 entries = (struct dx_entry *)(((char *)&root->info) +
776 root->info.info_length);
778 if (dx_get_limit(entries) != dx_root_limit(dir,
779 root->info.info_length)) {
780 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
781 dx_get_limit(entries),
782 dx_root_limit(dir, root->info.info_length));
786 dxtrace(printk("Look up %x", hash));
788 count = dx_get_count(entries);
789 if (!count || count > dx_get_limit(entries)) {
790 ext4_warning_inode(dir,
791 "dx entry: count %u beyond limit %u",
792 count, dx_get_limit(entries));
797 q = entries + count - 1;
800 dxtrace(printk("."));
801 if (dx_get_hash(m) > hash)
807 if (0) { // linear search cross check
808 unsigned n = count - 1;
812 dxtrace(printk(","));
813 if (dx_get_hash(++at) > hash)
819 assert (at == p - 1);
823 dxtrace(printk(" %x->%u\n", at == entries ? 0 : dx_get_hash(at),
825 frame->entries = entries;
830 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
831 if (IS_ERR(frame->bh)) {
832 ret_err = (struct dx_frame *) frame->bh;
836 entries = ((struct dx_node *) frame->bh->b_data)->entries;
838 if (dx_get_limit(entries) != dx_node_limit(dir)) {
839 ext4_warning_inode(dir,
840 "dx entry: limit %u != node limit %u",
841 dx_get_limit(entries), dx_node_limit(dir));
846 while (frame >= frame_in) {
851 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
852 ext4_warning_inode(dir,
853 "Corrupt directory, running e2fsck is recommended");
857 static void dx_release(struct dx_frame *frames)
859 if (frames[0].bh == NULL)
862 if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
863 brelse(frames[1].bh);
864 brelse(frames[0].bh);
868 * This function increments the frame pointer to search the next leaf
869 * block, and reads in the necessary intervening nodes if the search
870 * should be necessary. Whether or not the search is necessary is
871 * controlled by the hash parameter. If the hash value is even, then
872 * the search is only continued if the next block starts with that
873 * hash value. This is used if we are searching for a specific file.
875 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
877 * This function returns 1 if the caller should continue to search,
878 * or 0 if it should not. If there is an error reading one of the
879 * index blocks, it will a negative error code.
881 * If start_hash is non-null, it will be filled in with the starting
882 * hash of the next page.
884 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
885 struct dx_frame *frame,
886 struct dx_frame *frames,
890 struct buffer_head *bh;
896 * Find the next leaf page by incrementing the frame pointer.
897 * If we run out of entries in the interior node, loop around and
898 * increment pointer in the parent node. When we break out of
899 * this loop, num_frames indicates the number of interior
900 * nodes need to be read.
903 if (++(p->at) < p->entries + dx_get_count(p->entries))
912 * If the hash is 1, then continue only if the next page has a
913 * continuation hash of any value. This is used for readdir
914 * handling. Otherwise, check to see if the hash matches the
915 * desired contiuation hash. If it doesn't, return since
916 * there's no point to read in the successive index pages.
918 bhash = dx_get_hash(p->at);
921 if ((hash & 1) == 0) {
922 if ((bhash & ~1) != hash)
926 * If the hash is HASH_NB_ALWAYS, we always go to the next
927 * block so no check is necessary
929 while (num_frames--) {
930 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
936 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
943 * This function fills a red-black tree with information from a
944 * directory block. It returns the number directory entries loaded
945 * into the tree. If there is an error it is returned in err.
947 static int htree_dirblock_to_tree(struct file *dir_file,
948 struct inode *dir, ext4_lblk_t block,
949 struct dx_hash_info *hinfo,
950 __u32 start_hash, __u32 start_minor_hash)
952 struct buffer_head *bh;
953 struct ext4_dir_entry_2 *de, *top;
954 int err = 0, count = 0;
955 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}, tmp_str;
957 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
958 (unsigned long)block));
959 bh = ext4_read_dirblock(dir, block, DIRENT);
963 de = (struct ext4_dir_entry_2 *) bh->b_data;
964 top = (struct ext4_dir_entry_2 *) ((char *) de +
965 dir->i_sb->s_blocksize -
966 EXT4_DIR_REC_LEN(0));
967 #ifdef CONFIG_EXT4_FS_ENCRYPTION
968 /* Check if the directory is encrypted */
969 if (ext4_encrypted_inode(dir)) {
970 err = ext4_get_encryption_info(dir);
975 err = ext4_fname_crypto_alloc_buffer(dir, EXT4_NAME_LEN,
983 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
984 if (ext4_check_dir_entry(dir, NULL, de, bh,
985 bh->b_data, bh->b_size,
986 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
987 + ((char *)de - bh->b_data))) {
988 /* silently ignore the rest of the block */
991 ext4fs_dirhash(de->name, de->name_len, hinfo);
992 if ((hinfo->hash < start_hash) ||
993 ((hinfo->hash == start_hash) &&
994 (hinfo->minor_hash < start_minor_hash)))
998 if (!ext4_encrypted_inode(dir)) {
999 tmp_str.name = de->name;
1000 tmp_str.len = de->name_len;
1001 err = ext4_htree_store_dirent(dir_file,
1002 hinfo->hash, hinfo->minor_hash, de,
1005 int save_len = fname_crypto_str.len;
1007 /* Directory is encrypted */
1008 err = ext4_fname_disk_to_usr(dir, hinfo, de,
1014 err = ext4_htree_store_dirent(dir_file,
1015 hinfo->hash, hinfo->minor_hash, de,
1017 fname_crypto_str.len = save_len;
1027 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1028 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1035 * This function fills a red-black tree with information from a
1036 * directory. We start scanning the directory in hash order, starting
1037 * at start_hash and start_minor_hash.
1039 * This function returns the number of entries inserted into the tree,
1040 * or a negative error code.
1042 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1043 __u32 start_minor_hash, __u32 *next_hash)
1045 struct dx_hash_info hinfo;
1046 struct ext4_dir_entry_2 *de;
1047 struct dx_frame frames[2], *frame;
1053 struct ext4_str tmp_str;
1055 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1056 start_hash, start_minor_hash));
1057 dir = file_inode(dir_file);
1058 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1059 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1060 if (hinfo.hash_version <= DX_HASH_TEA)
1061 hinfo.hash_version +=
1062 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1063 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1064 if (ext4_has_inline_data(dir)) {
1065 int has_inline_data = 1;
1066 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1070 if (has_inline_data) {
1075 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1076 start_hash, start_minor_hash);
1080 hinfo.hash = start_hash;
1081 hinfo.minor_hash = 0;
1082 frame = dx_probe(NULL, dir, &hinfo, frames);
1084 return PTR_ERR(frame);
1086 /* Add '.' and '..' from the htree header */
1087 if (!start_hash && !start_minor_hash) {
1088 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1089 tmp_str.name = de->name;
1090 tmp_str.len = de->name_len;
1091 err = ext4_htree_store_dirent(dir_file, 0, 0,
1097 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1098 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1099 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1100 tmp_str.name = de->name;
1101 tmp_str.len = de->name_len;
1102 err = ext4_htree_store_dirent(dir_file, 2, 0,
1110 block = dx_get_block(frame->at);
1111 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1112 start_hash, start_minor_hash);
1119 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1120 frame, frames, &hashval);
1121 *next_hash = hashval;
1127 * Stop if: (a) there are no more entries, or
1128 * (b) we have inserted at least one entry and the
1129 * next hash value is not a continuation
1132 (count && ((hashval & 1) == 0)))
1136 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1137 "next hash: %x\n", count, *next_hash));
1144 static inline int search_dirblock(struct buffer_head *bh,
1146 struct ext4_filename *fname,
1147 const struct qstr *d_name,
1148 unsigned int offset,
1149 struct ext4_dir_entry_2 **res_dir)
1151 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1152 fname, d_name, offset, res_dir);
1156 * Directory block splitting, compacting
1160 * Create map of hash values, offsets, and sizes, stored at end of block.
1161 * Returns number of entries mapped.
1163 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1164 unsigned blocksize, struct dx_hash_info *hinfo,
1165 struct dx_map_entry *map_tail)
1168 char *base = (char *) de;
1169 struct dx_hash_info h = *hinfo;
1171 while ((char *) de < base + blocksize) {
1172 if (de->name_len && de->inode) {
1173 ext4fs_dirhash(de->name, de->name_len, &h);
1175 map_tail->hash = h.hash;
1176 map_tail->offs = ((char *) de - base)>>2;
1177 map_tail->size = le16_to_cpu(de->rec_len);
1181 /* XXX: do we need to check rec_len == 0 case? -Chris */
1182 de = ext4_next_entry(de, blocksize);
1187 /* Sort map by hash value */
1188 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1190 struct dx_map_entry *p, *q, *top = map + count - 1;
1192 /* Combsort until bubble sort doesn't suck */
1194 count = count*10/13;
1195 if (count - 9 < 2) /* 9, 10 -> 11 */
1197 for (p = top, q = p - count; q >= map; p--, q--)
1198 if (p->hash < q->hash)
1201 /* Garden variety bubble sort */
1206 if (q[1].hash >= q[0].hash)
1214 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1216 struct dx_entry *entries = frame->entries;
1217 struct dx_entry *old = frame->at, *new = old + 1;
1218 int count = dx_get_count(entries);
1220 assert(count < dx_get_limit(entries));
1221 assert(old < entries + count);
1222 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1223 dx_set_hash(new, hash);
1224 dx_set_block(new, block);
1225 dx_set_count(entries, count + 1);
1229 * Test whether a directory entry matches the filename being searched for.
1231 * Return: %true if the directory entry matches, otherwise %false.
1233 static inline bool ext4_match(const struct ext4_filename *fname,
1234 const struct ext4_dir_entry_2 *de)
1236 const void *name = fname_name(fname);
1237 u32 len = fname_len(fname);
1242 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1243 if (unlikely(!name)) {
1244 if (fname->usr_fname->name[0] == '_') {
1246 if (de->name_len <= 32)
1248 ret = memcmp(de->name + ((de->name_len - 17) & ~15),
1249 fname->crypto_buf.name + 8, 16);
1250 return (ret == 0) ? 1 : 0;
1252 name = fname->crypto_buf.name;
1253 len = fname->crypto_buf.len;
1256 if (de->name_len != len)
1258 return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1262 * Returns 0 if not found, -1 on failure, and 1 on success
1264 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1265 struct inode *dir, struct ext4_filename *fname,
1266 const struct qstr *d_name,
1267 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1269 struct ext4_dir_entry_2 * de;
1273 de = (struct ext4_dir_entry_2 *)search_buf;
1274 dlimit = search_buf + buf_size;
1275 while ((char *) de < dlimit) {
1276 /* this code is executed quadratically often */
1277 /* do minimal checking `by hand' */
1278 if ((char *) de + de->name_len <= dlimit &&
1279 ext4_match(fname, de)) {
1280 /* found a match - just to be sure, do
1282 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1288 /* prevent looping on a bad block */
1289 de_len = ext4_rec_len_from_disk(de->rec_len,
1290 dir->i_sb->s_blocksize);
1294 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1299 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1300 struct ext4_dir_entry *de)
1302 struct super_block *sb = dir->i_sb;
1308 if (de->inode == 0 &&
1309 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1318 * finds an entry in the specified directory with the wanted name. It
1319 * returns the cache buffer in which the entry was found, and the entry
1320 * itself (as a parameter - res_dir). It does NOT read the inode of the
1321 * entry - you'll have to do that yourself if you want to.
1323 * The returned buffer_head has ->b_count elevated. The caller is expected
1324 * to brelse() it when appropriate.
1326 static struct buffer_head * ext4_find_entry (struct inode *dir,
1327 const struct qstr *d_name,
1328 struct ext4_dir_entry_2 **res_dir,
1331 struct super_block *sb;
1332 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1333 struct buffer_head *bh, *ret = NULL;
1334 ext4_lblk_t start, block, b;
1335 const u8 *name = d_name->name;
1336 int ra_max = 0; /* Number of bh's in the readahead
1338 int ra_ptr = 0; /* Current index into readahead
1341 ext4_lblk_t nblocks;
1342 int i, namelen, retval;
1343 struct ext4_filename fname;
1347 namelen = d_name->len;
1348 if (namelen > EXT4_NAME_LEN)
1351 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1353 return ERR_PTR(retval);
1355 if (ext4_has_inline_data(dir)) {
1356 int has_inline_data = 1;
1357 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1359 if (has_inline_data) {
1362 goto cleanup_and_exit;
1366 if ((namelen <= 2) && (name[0] == '.') &&
1367 (name[1] == '.' || name[1] == '\0')) {
1369 * "." or ".." will only be in the first block
1370 * NFS may look up ".."; "." should be handled by the VFS
1377 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1379 * On success, or if the error was file not found,
1380 * return. Otherwise, fall back to doing a search the
1381 * old fashioned way.
1383 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1384 goto cleanup_and_exit;
1385 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1389 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1392 goto cleanup_and_exit;
1394 start = EXT4_I(dir)->i_dir_start_lookup;
1395 if (start >= nblocks)
1401 * We deal with the read-ahead logic here.
1404 if (ra_ptr >= ra_max) {
1405 /* Refill the readahead buffer */
1408 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1410 * Terminate if we reach the end of the
1411 * directory and must wrap, or if our
1412 * search has finished at this block.
1414 if (b >= nblocks || (num && block == start)) {
1415 bh_use[ra_max] = NULL;
1419 bh = ext4_getblk(NULL, dir, b++, 0);
1423 goto cleanup_and_exit;
1427 bh_use[ra_max] = bh;
1429 ll_rw_block(READ | REQ_META | REQ_PRIO,
1433 if ((bh = bh_use[ra_ptr++]) == NULL)
1436 if (!buffer_uptodate(bh)) {
1437 /* read error, skip block & hope for the best */
1438 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1439 (unsigned long) block);
1443 if (!buffer_verified(bh) &&
1444 !is_dx_internal_node(dir, block,
1445 (struct ext4_dir_entry *)bh->b_data) &&
1446 !ext4_dirent_csum_verify(dir,
1447 (struct ext4_dir_entry *)bh->b_data)) {
1448 EXT4_ERROR_INODE(dir, "checksumming directory "
1449 "block %lu", (unsigned long)block);
1453 set_buffer_verified(bh);
1454 i = search_dirblock(bh, dir, &fname, d_name,
1455 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1457 EXT4_I(dir)->i_dir_start_lookup = block;
1459 goto cleanup_and_exit;
1463 goto cleanup_and_exit;
1466 if (++block >= nblocks)
1468 } while (block != start);
1471 * If the directory has grown while we were searching, then
1472 * search the last part of the directory before giving up.
1475 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1476 if (block < nblocks) {
1482 /* Clean up the read-ahead blocks */
1483 for (; ra_ptr < ra_max; ra_ptr++)
1484 brelse(bh_use[ra_ptr]);
1485 ext4_fname_free_filename(&fname);
1489 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1490 struct ext4_filename *fname,
1491 struct ext4_dir_entry_2 **res_dir)
1493 struct super_block * sb = dir->i_sb;
1494 struct dx_frame frames[2], *frame;
1495 const struct qstr *d_name = fname->usr_fname;
1496 struct buffer_head *bh;
1500 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1503 frame = dx_probe(fname, dir, NULL, frames);
1505 return (struct buffer_head *) frame;
1507 block = dx_get_block(frame->at);
1508 bh = ext4_read_dirblock(dir, block, DIRENT);
1512 retval = search_dirblock(bh, dir, fname, d_name,
1513 block << EXT4_BLOCK_SIZE_BITS(sb),
1519 bh = ERR_PTR(ERR_BAD_DX_DIR);
1523 /* Check to see if we should continue to search */
1524 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1527 ext4_warning_inode(dir,
1528 "error %d reading directory index block",
1530 bh = ERR_PTR(retval);
1533 } while (retval == 1);
1537 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1543 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1545 struct inode *inode;
1546 struct ext4_dir_entry_2 *de;
1547 struct buffer_head *bh;
1549 if (ext4_encrypted_inode(dir)) {
1550 int res = ext4_get_encryption_info(dir);
1553 * This should be a properly defined flag for
1554 * dentry->d_flags when we uplift this to the VFS.
1555 * d_fsdata is set to (void *) 1 if if the dentry is
1556 * created while the directory was encrypted and we
1557 * don't have access to the key.
1559 dentry->d_fsdata = NULL;
1560 if (ext4_encryption_info(dir))
1561 dentry->d_fsdata = (void *) 1;
1562 d_set_d_op(dentry, &ext4_encrypted_d_ops);
1563 if (res && res != -ENOKEY)
1564 return ERR_PTR(res);
1567 if (dentry->d_name.len > EXT4_NAME_LEN)
1568 return ERR_PTR(-ENAMETOOLONG);
1570 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1572 return (struct dentry *) bh;
1575 __u32 ino = le32_to_cpu(de->inode);
1577 if (!ext4_valid_inum(dir->i_sb, ino)) {
1578 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1579 return ERR_PTR(-EFSCORRUPTED);
1581 if (unlikely(ino == dir->i_ino)) {
1582 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1584 return ERR_PTR(-EFSCORRUPTED);
1586 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1587 if (inode == ERR_PTR(-ESTALE)) {
1588 EXT4_ERROR_INODE(dir,
1589 "deleted inode referenced: %u",
1591 return ERR_PTR(-EFSCORRUPTED);
1593 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1594 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1595 S_ISLNK(inode->i_mode)) &&
1596 !ext4_is_child_context_consistent_with_parent(dir,
1599 ext4_warning(inode->i_sb,
1600 "Inconsistent encryption contexts: %lu/%lu\n",
1601 (unsigned long) dir->i_ino,
1602 (unsigned long) inode->i_ino);
1603 return ERR_PTR(-EPERM);
1606 return d_splice_alias(inode, dentry);
1610 struct dentry *ext4_get_parent(struct dentry *child)
1613 static const struct qstr dotdot = QSTR_INIT("..", 2);
1614 struct ext4_dir_entry_2 * de;
1615 struct buffer_head *bh;
1617 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1619 return (struct dentry *) bh;
1621 return ERR_PTR(-ENOENT);
1622 ino = le32_to_cpu(de->inode);
1625 if (!ext4_valid_inum(d_inode(child)->i_sb, ino)) {
1626 EXT4_ERROR_INODE(d_inode(child),
1627 "bad parent inode number: %u", ino);
1628 return ERR_PTR(-EFSCORRUPTED);
1631 return d_obtain_alias(ext4_iget(d_inode(child)->i_sb, ino, EXT4_IGET_NORMAL));
1635 * Move count entries from end of map between two memory locations.
1636 * Returns pointer to last entry moved.
1638 static struct ext4_dir_entry_2 *
1639 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1642 unsigned rec_len = 0;
1645 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1646 (from + (map->offs<<2));
1647 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1648 memcpy (to, de, rec_len);
1649 ((struct ext4_dir_entry_2 *) to)->rec_len =
1650 ext4_rec_len_to_disk(rec_len, blocksize);
1655 return (struct ext4_dir_entry_2 *) (to - rec_len);
1659 * Compact each dir entry in the range to the minimal rec_len.
1660 * Returns pointer to last entry in range.
1662 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1664 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1665 unsigned rec_len = 0;
1668 while ((char*)de < base + blocksize) {
1669 next = ext4_next_entry(de, blocksize);
1670 if (de->inode && de->name_len) {
1671 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1673 memmove(to, de, rec_len);
1674 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1676 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1684 * Split a full leaf block to make room for a new dir entry.
1685 * Allocate a new block, and move entries so that they are approx. equally full.
1686 * Returns pointer to de in block into which the new entry will be inserted.
1688 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1689 struct buffer_head **bh,struct dx_frame *frame,
1690 struct dx_hash_info *hinfo)
1692 unsigned blocksize = dir->i_sb->s_blocksize;
1693 unsigned count, continued;
1694 struct buffer_head *bh2;
1695 ext4_lblk_t newblock;
1697 struct dx_map_entry *map;
1698 char *data1 = (*bh)->b_data, *data2;
1699 unsigned split, move, size;
1700 struct ext4_dir_entry_2 *de = NULL, *de2;
1701 struct ext4_dir_entry_tail *t;
1705 if (ext4_has_metadata_csum(dir->i_sb))
1706 csum_size = sizeof(struct ext4_dir_entry_tail);
1708 bh2 = ext4_append(handle, dir, &newblock);
1712 return (struct ext4_dir_entry_2 *) bh2;
1715 BUFFER_TRACE(*bh, "get_write_access");
1716 err = ext4_journal_get_write_access(handle, *bh);
1720 BUFFER_TRACE(frame->bh, "get_write_access");
1721 err = ext4_journal_get_write_access(handle, frame->bh);
1725 data2 = bh2->b_data;
1727 /* create map in the end of data2 block */
1728 map = (struct dx_map_entry *) (data2 + blocksize);
1729 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1730 blocksize, hinfo, map);
1732 dx_sort_map(map, count);
1733 /* Ensure that neither split block is over half full */
1736 for (i = count-1; i >= 0; i--) {
1737 /* is more than half of this entry in 2nd half of the block? */
1738 if (size + map[i].size/2 > blocksize/2)
1740 size += map[i].size;
1744 * map index at which we will split
1746 * If the sum of active entries didn't exceed half the block size, just
1747 * split it in half by count; each resulting block will have at least
1748 * half the space free.
1751 split = count - move;
1755 hash2 = map[split].hash;
1756 continued = hash2 == map[split - 1].hash;
1757 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1758 (unsigned long)dx_get_block(frame->at),
1759 hash2, split, count-split));
1761 /* Fancy dance to stay within two buffers */
1762 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1764 de = dx_pack_dirents(data1, blocksize);
1765 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1768 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1772 t = EXT4_DIRENT_TAIL(data2, blocksize);
1773 initialize_dirent_tail(t, blocksize);
1775 t = EXT4_DIRENT_TAIL(data1, blocksize);
1776 initialize_dirent_tail(t, blocksize);
1779 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1781 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1784 /* Which block gets the new entry? */
1785 if (hinfo->hash >= hash2) {
1789 dx_insert_block(frame, hash2 + continued, newblock);
1790 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1793 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1797 dxtrace(dx_show_index("frame", frame->entries));
1804 ext4_std_error(dir->i_sb, err);
1805 return ERR_PTR(err);
1808 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1809 struct buffer_head *bh,
1810 void *buf, int buf_size,
1811 struct ext4_filename *fname,
1812 struct ext4_dir_entry_2 **dest_de)
1814 struct ext4_dir_entry_2 *de;
1815 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1817 unsigned int offset = 0;
1820 de = (struct ext4_dir_entry_2 *)buf;
1821 top = buf + buf_size - reclen;
1822 while ((char *) de <= top) {
1823 if (ext4_check_dir_entry(dir, NULL, de, bh,
1824 buf, buf_size, offset))
1825 return -EFSCORRUPTED;
1826 if (ext4_match(fname, de))
1828 nlen = EXT4_DIR_REC_LEN(de->name_len);
1829 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1830 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1832 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1835 if ((char *) de > top)
1842 int ext4_insert_dentry(struct inode *dir,
1843 struct inode *inode,
1844 struct ext4_dir_entry_2 *de,
1846 struct ext4_filename *fname)
1851 nlen = EXT4_DIR_REC_LEN(de->name_len);
1852 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1854 struct ext4_dir_entry_2 *de1 =
1855 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1856 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1857 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1860 de->file_type = EXT4_FT_UNKNOWN;
1861 de->inode = cpu_to_le32(inode->i_ino);
1862 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1863 de->name_len = fname_len(fname);
1864 memcpy(de->name, fname_name(fname), fname_len(fname));
1869 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1870 * it points to a directory entry which is guaranteed to be large
1871 * enough for new directory entry. If de is NULL, then
1872 * add_dirent_to_buf will attempt search the directory block for
1873 * space. It will return -ENOSPC if no space is available, and -EIO
1874 * and -EEXIST if directory entry already exists.
1876 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1878 struct inode *inode, struct ext4_dir_entry_2 *de,
1879 struct buffer_head *bh)
1881 unsigned int blocksize = dir->i_sb->s_blocksize;
1885 if (ext4_has_metadata_csum(inode->i_sb))
1886 csum_size = sizeof(struct ext4_dir_entry_tail);
1889 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1890 blocksize - csum_size, fname, &de);
1894 BUFFER_TRACE(bh, "get_write_access");
1895 err = ext4_journal_get_write_access(handle, bh);
1897 ext4_std_error(dir->i_sb, err);
1901 /* By now the buffer is marked for journaling. Due to crypto operations,
1902 * the following function call may fail */
1903 err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1908 * XXX shouldn't update any times until successful
1909 * completion of syscall, but too many callers depend
1912 * XXX similarly, too many callers depend on
1913 * ext4_new_inode() setting the times, but error
1914 * recovery deletes the inode, so the worst that can
1915 * happen is that the times are slightly out of date
1916 * and/or different from the directory change time.
1918 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1919 ext4_update_dx_flag(dir);
1921 ext4_mark_inode_dirty(handle, dir);
1922 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1923 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1925 ext4_std_error(dir->i_sb, err);
1930 * This converts a one block unindexed directory to a 3 block indexed
1931 * directory, and adds the dentry to the indexed directory.
1933 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1934 struct dentry *dentry,
1935 struct inode *inode, struct buffer_head *bh)
1937 struct inode *dir = d_inode(dentry->d_parent);
1938 struct buffer_head *bh2;
1939 struct dx_root *root;
1940 struct dx_frame frames[2], *frame;
1941 struct dx_entry *entries;
1942 struct ext4_dir_entry_2 *de, *de2;
1943 struct ext4_dir_entry_tail *t;
1949 struct fake_dirent *fde;
1952 if (ext4_has_metadata_csum(inode->i_sb))
1953 csum_size = sizeof(struct ext4_dir_entry_tail);
1955 blocksize = dir->i_sb->s_blocksize;
1956 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1957 BUFFER_TRACE(bh, "get_write_access");
1958 retval = ext4_journal_get_write_access(handle, bh);
1960 ext4_std_error(dir->i_sb, retval);
1964 root = (struct dx_root *) bh->b_data;
1966 /* The 0th block becomes the root, move the dirents out */
1967 fde = &root->dotdot;
1968 de = (struct ext4_dir_entry_2 *)((char *)fde +
1969 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1970 if ((char *) de >= (((char *) root) + blocksize)) {
1971 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1973 return -EFSCORRUPTED;
1975 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1977 /* Allocate new block for the 0th block's dirents */
1978 bh2 = ext4_append(handle, dir, &block);
1981 return PTR_ERR(bh2);
1983 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1984 data1 = bh2->b_data;
1986 memcpy (data1, de, len);
1987 de = (struct ext4_dir_entry_2 *) data1;
1989 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1991 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1996 t = EXT4_DIRENT_TAIL(data1, blocksize);
1997 initialize_dirent_tail(t, blocksize);
2000 /* Initialize the root; the dot dirents already exist */
2001 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2002 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2004 memset (&root->info, 0, sizeof(root->info));
2005 root->info.info_length = sizeof(root->info);
2006 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2007 entries = root->entries;
2008 dx_set_block(entries, 1);
2009 dx_set_count(entries, 1);
2010 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2012 /* Initialize as for dx_probe */
2013 fname->hinfo.hash_version = root->info.hash_version;
2014 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2015 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2016 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2017 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2019 memset(frames, 0, sizeof(frames));
2021 frame->entries = entries;
2022 frame->at = entries;
2025 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2028 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2032 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2034 retval = PTR_ERR(de);
2038 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2041 * Even if the block split failed, we have to properly write
2042 * out all the changes we did so far. Otherwise we can end up
2043 * with corrupted filesystem.
2046 ext4_mark_inode_dirty(handle, dir);
2055 * adds a file entry to the specified directory, using the same
2056 * semantics as ext4_find_entry(). It returns NULL if it failed.
2058 * NOTE!! The inode part of 'de' is left at 0 - which means you
2059 * may not sleep between calling this and putting something into
2060 * the entry, as someone else might have used it while you slept.
2062 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2063 struct inode *inode)
2065 struct inode *dir = d_inode(dentry->d_parent);
2066 struct buffer_head *bh = NULL;
2067 struct ext4_dir_entry_2 *de;
2068 struct ext4_dir_entry_tail *t;
2069 struct super_block *sb;
2070 struct ext4_filename fname;
2074 ext4_lblk_t block, blocks;
2077 if (ext4_has_metadata_csum(inode->i_sb))
2078 csum_size = sizeof(struct ext4_dir_entry_tail);
2081 blocksize = sb->s_blocksize;
2082 if (!dentry->d_name.len)
2085 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2089 if (ext4_has_inline_data(dir)) {
2090 retval = ext4_try_add_inline_entry(handle, &fname,
2101 retval = ext4_dx_add_entry(handle, &fname, dentry, inode);
2102 if (!retval || (retval != ERR_BAD_DX_DIR))
2104 /* Can we just ignore htree data? */
2105 if (ext4_has_metadata_csum(sb)) {
2106 EXT4_ERROR_INODE(dir,
2107 "Directory has corrupted htree index.");
2108 retval = -EFSCORRUPTED;
2111 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2113 ext4_mark_inode_dirty(handle, dir);
2115 blocks = dir->i_size >> sb->s_blocksize_bits;
2116 for (block = 0; block < blocks; block++) {
2117 bh = ext4_read_dirblock(dir, block, DIRENT);
2119 retval = PTR_ERR(bh);
2123 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2125 if (retval != -ENOSPC)
2128 if (blocks == 1 && !dx_fallback &&
2129 ext4_has_feature_dir_index(sb)) {
2130 retval = make_indexed_dir(handle, &fname, dentry,
2132 bh = NULL; /* make_indexed_dir releases bh */
2137 bh = ext4_append(handle, dir, &block);
2139 retval = PTR_ERR(bh);
2143 de = (struct ext4_dir_entry_2 *) bh->b_data;
2145 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2148 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2149 initialize_dirent_tail(t, blocksize);
2152 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2154 ext4_fname_free_filename(&fname);
2157 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2162 * Returns 0 for success, or a negative error value
2164 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2165 struct dentry *dentry, struct inode *inode)
2167 struct dx_frame frames[2], *frame;
2168 struct dx_entry *entries, *at;
2169 struct buffer_head *bh;
2170 struct inode *dir = d_inode(dentry->d_parent);
2171 struct super_block *sb = dir->i_sb;
2172 struct ext4_dir_entry_2 *de;
2175 frame = dx_probe(fname, dir, NULL, frames);
2177 return PTR_ERR(frame);
2178 entries = frame->entries;
2180 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2187 BUFFER_TRACE(bh, "get_write_access");
2188 err = ext4_journal_get_write_access(handle, bh);
2192 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2196 /* Block full, should compress but for now just split */
2197 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2198 dx_get_count(entries), dx_get_limit(entries)));
2199 /* Need to split index? */
2200 if (dx_get_count(entries) == dx_get_limit(entries)) {
2201 ext4_lblk_t newblock;
2202 unsigned icount = dx_get_count(entries);
2203 int levels = frame - frames;
2204 struct dx_entry *entries2;
2205 struct dx_node *node2;
2206 struct buffer_head *bh2;
2208 if (levels && (dx_get_count(frames->entries) ==
2209 dx_get_limit(frames->entries))) {
2210 ext4_warning_inode(dir, "Directory index full!");
2214 bh2 = ext4_append(handle, dir, &newblock);
2219 node2 = (struct dx_node *)(bh2->b_data);
2220 entries2 = node2->entries;
2221 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2222 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2224 BUFFER_TRACE(frame->bh, "get_write_access");
2225 err = ext4_journal_get_write_access(handle, frame->bh);
2229 unsigned icount1 = icount/2, icount2 = icount - icount1;
2230 unsigned hash2 = dx_get_hash(entries + icount1);
2231 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2234 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2235 err = ext4_journal_get_write_access(handle,
2240 memcpy((char *) entries2, (char *) (entries + icount1),
2241 icount2 * sizeof(struct dx_entry));
2242 dx_set_count(entries, icount1);
2243 dx_set_count(entries2, icount2);
2244 dx_set_limit(entries2, dx_node_limit(dir));
2246 /* Which index block gets the new entry? */
2247 if (at - entries >= icount1) {
2248 frame->at = at = at - entries - icount1 + entries2;
2249 frame->entries = entries = entries2;
2250 swap(frame->bh, bh2);
2252 dx_insert_block(frames + 0, hash2, newblock);
2253 dxtrace(dx_show_index("node", frames[1].entries));
2254 dxtrace(dx_show_index("node",
2255 ((struct dx_node *) bh2->b_data)->entries));
2256 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2261 dxtrace(printk(KERN_DEBUG
2262 "Creating second level index...\n"));
2263 memcpy((char *) entries2, (char *) entries,
2264 icount * sizeof(struct dx_entry));
2265 dx_set_limit(entries2, dx_node_limit(dir));
2268 dx_set_count(entries, 1);
2269 dx_set_block(entries + 0, newblock);
2270 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2272 /* Add new access path frame */
2274 frame->at = at = at - entries + entries2;
2275 frame->entries = entries = entries2;
2277 err = ext4_journal_get_write_access(handle,
2282 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2284 ext4_std_error(inode->i_sb, err);
2288 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2293 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2297 ext4_std_error(dir->i_sb, err);
2305 * ext4_generic_delete_entry deletes a directory entry by merging it
2306 * with the previous entry
2308 int ext4_generic_delete_entry(handle_t *handle,
2310 struct ext4_dir_entry_2 *de_del,
2311 struct buffer_head *bh,
2316 struct ext4_dir_entry_2 *de, *pde;
2317 unsigned int blocksize = dir->i_sb->s_blocksize;
2322 de = (struct ext4_dir_entry_2 *)entry_buf;
2323 while (i < buf_size - csum_size) {
2324 if (ext4_check_dir_entry(dir, NULL, de, bh,
2325 entry_buf, buf_size, i))
2326 return -EFSCORRUPTED;
2329 pde->rec_len = ext4_rec_len_to_disk(
2330 ext4_rec_len_from_disk(pde->rec_len,
2332 ext4_rec_len_from_disk(de->rec_len,
2340 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2342 de = ext4_next_entry(de, blocksize);
2347 static int ext4_delete_entry(handle_t *handle,
2349 struct ext4_dir_entry_2 *de_del,
2350 struct buffer_head *bh)
2352 int err, csum_size = 0;
2354 if (ext4_has_inline_data(dir)) {
2355 int has_inline_data = 1;
2356 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2358 if (has_inline_data)
2362 if (ext4_has_metadata_csum(dir->i_sb))
2363 csum_size = sizeof(struct ext4_dir_entry_tail);
2365 BUFFER_TRACE(bh, "get_write_access");
2366 err = ext4_journal_get_write_access(handle, bh);
2370 err = ext4_generic_delete_entry(handle, dir, de_del,
2372 dir->i_sb->s_blocksize, csum_size);
2376 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2377 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2384 ext4_std_error(dir->i_sb, err);
2389 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2390 * since this indicates that nlinks count was previously 1.
2392 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2395 if (is_dx(inode) && inode->i_nlink > 1) {
2396 /* limit is 16-bit i_links_count */
2397 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2398 set_nlink(inode, 1);
2399 ext4_set_feature_dir_nlink(inode->i_sb);
2405 * If a directory had nlink == 1, then we should let it be 1. This indicates
2406 * directory has >EXT4_LINK_MAX subdirs.
2408 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2410 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2415 static int ext4_add_nondir(handle_t *handle,
2416 struct dentry *dentry, struct inode *inode)
2418 int err = ext4_add_entry(handle, dentry, inode);
2420 ext4_mark_inode_dirty(handle, inode);
2421 d_instantiate_new(dentry, inode);
2425 unlock_new_inode(inode);
2431 * By the time this is called, we already have created
2432 * the directory cache entry for the new file, but it
2433 * is so far negative - it has no inode.
2435 * If the create succeeds, we fill in the inode information
2436 * with d_instantiate().
2438 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2442 struct inode *inode;
2443 int err, credits, retries = 0;
2445 err = dquot_initialize(dir);
2449 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2450 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2452 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2453 NULL, EXT4_HT_DIR, credits);
2454 handle = ext4_journal_current_handle();
2455 err = PTR_ERR(inode);
2456 if (!IS_ERR(inode)) {
2457 inode->i_op = &ext4_file_inode_operations;
2458 inode->i_fop = &ext4_file_operations;
2459 ext4_set_aops(inode);
2460 err = ext4_add_nondir(handle, dentry, inode);
2461 if (!err && IS_DIRSYNC(dir))
2462 ext4_handle_sync(handle);
2465 ext4_journal_stop(handle);
2466 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2471 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2472 umode_t mode, dev_t rdev)
2475 struct inode *inode;
2476 int err, credits, retries = 0;
2478 err = dquot_initialize(dir);
2482 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2483 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2485 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2486 NULL, EXT4_HT_DIR, credits);
2487 handle = ext4_journal_current_handle();
2488 err = PTR_ERR(inode);
2489 if (!IS_ERR(inode)) {
2490 init_special_inode(inode, inode->i_mode, rdev);
2491 inode->i_op = &ext4_special_inode_operations;
2492 err = ext4_add_nondir(handle, dentry, inode);
2493 if (!err && IS_DIRSYNC(dir))
2494 ext4_handle_sync(handle);
2497 ext4_journal_stop(handle);
2498 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2503 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2506 struct inode *inode;
2507 int err, retries = 0;
2509 err = dquot_initialize(dir);
2514 inode = ext4_new_inode_start_handle(dir, mode,
2517 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2518 4 + EXT4_XATTR_TRANS_BLOCKS);
2519 handle = ext4_journal_current_handle();
2520 err = PTR_ERR(inode);
2521 if (!IS_ERR(inode)) {
2522 inode->i_op = &ext4_file_inode_operations;
2523 inode->i_fop = &ext4_file_operations;
2524 ext4_set_aops(inode);
2525 d_tmpfile(dentry, inode);
2526 err = ext4_orphan_add(handle, inode);
2528 goto err_unlock_inode;
2529 mark_inode_dirty(inode);
2530 unlock_new_inode(inode);
2533 ext4_journal_stop(handle);
2534 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2538 ext4_journal_stop(handle);
2539 unlock_new_inode(inode);
2543 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2544 struct ext4_dir_entry_2 *de,
2545 int blocksize, int csum_size,
2546 unsigned int parent_ino, int dotdot_real_len)
2548 de->inode = cpu_to_le32(inode->i_ino);
2550 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2552 strcpy(de->name, ".");
2553 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2555 de = ext4_next_entry(de, blocksize);
2556 de->inode = cpu_to_le32(parent_ino);
2558 if (!dotdot_real_len)
2559 de->rec_len = ext4_rec_len_to_disk(blocksize -
2560 (csum_size + EXT4_DIR_REC_LEN(1)),
2563 de->rec_len = ext4_rec_len_to_disk(
2564 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2565 strcpy(de->name, "..");
2566 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2568 return ext4_next_entry(de, blocksize);
2571 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2572 struct inode *inode)
2574 struct buffer_head *dir_block = NULL;
2575 struct ext4_dir_entry_2 *de;
2576 struct ext4_dir_entry_tail *t;
2577 ext4_lblk_t block = 0;
2578 unsigned int blocksize = dir->i_sb->s_blocksize;
2582 if (ext4_has_metadata_csum(dir->i_sb))
2583 csum_size = sizeof(struct ext4_dir_entry_tail);
2585 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2586 err = ext4_try_create_inline_dir(handle, dir, inode);
2587 if (err < 0 && err != -ENOSPC)
2594 dir_block = ext4_append(handle, inode, &block);
2595 if (IS_ERR(dir_block))
2596 return PTR_ERR(dir_block);
2597 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2598 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2599 set_nlink(inode, 2);
2601 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2602 initialize_dirent_tail(t, blocksize);
2605 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2606 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2609 set_buffer_verified(dir_block);
2615 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2618 struct inode *inode;
2619 int err, credits, retries = 0;
2621 if (EXT4_DIR_LINK_MAX(dir))
2624 err = dquot_initialize(dir);
2628 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2629 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2631 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2633 0, NULL, EXT4_HT_DIR, credits);
2634 handle = ext4_journal_current_handle();
2635 err = PTR_ERR(inode);
2639 inode->i_op = &ext4_dir_inode_operations;
2640 inode->i_fop = &ext4_dir_operations;
2641 err = ext4_init_new_dir(handle, dir, inode);
2643 goto out_clear_inode;
2644 err = ext4_mark_inode_dirty(handle, inode);
2646 err = ext4_add_entry(handle, dentry, inode);
2650 unlock_new_inode(inode);
2651 ext4_mark_inode_dirty(handle, inode);
2655 ext4_inc_count(handle, dir);
2656 ext4_update_dx_flag(dir);
2657 err = ext4_mark_inode_dirty(handle, dir);
2659 goto out_clear_inode;
2660 d_instantiate_new(dentry, inode);
2661 if (IS_DIRSYNC(dir))
2662 ext4_handle_sync(handle);
2666 ext4_journal_stop(handle);
2667 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2673 * routine to check that the specified directory is empty (for rmdir)
2675 int ext4_empty_dir(struct inode *inode)
2677 unsigned int offset;
2678 struct buffer_head *bh;
2679 struct ext4_dir_entry_2 *de, *de1;
2680 struct super_block *sb;
2683 if (ext4_has_inline_data(inode)) {
2684 int has_inline_data = 1;
2686 err = empty_inline_dir(inode, &has_inline_data);
2687 if (has_inline_data)
2692 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2693 EXT4_ERROR_INODE(inode, "invalid size");
2696 bh = ext4_read_dirblock(inode, 0, EITHER);
2700 de = (struct ext4_dir_entry_2 *) bh->b_data;
2701 de1 = ext4_next_entry(de, sb->s_blocksize);
2702 if (le32_to_cpu(de->inode) != inode->i_ino ||
2703 le32_to_cpu(de1->inode) == 0 ||
2704 strcmp(".", de->name) || strcmp("..", de1->name)) {
2705 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2709 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2710 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2711 de = ext4_next_entry(de1, sb->s_blocksize);
2712 while (offset < inode->i_size) {
2713 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2714 unsigned int lblock;
2717 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2718 bh = ext4_read_dirblock(inode, lblock, EITHER);
2721 de = (struct ext4_dir_entry_2 *) bh->b_data;
2723 if (ext4_check_dir_entry(inode, NULL, de, bh,
2724 bh->b_data, bh->b_size, offset)) {
2725 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2727 offset = (offset | (sb->s_blocksize - 1)) + 1;
2730 if (le32_to_cpu(de->inode)) {
2734 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2735 de = ext4_next_entry(de, sb->s_blocksize);
2742 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2743 * such inodes, starting at the superblock, in case we crash before the
2744 * file is closed/deleted, or in case the inode truncate spans multiple
2745 * transactions and the last transaction is not recovered after a crash.
2747 * At filesystem recovery time, we walk this list deleting unlinked
2748 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2750 * Orphan list manipulation functions must be called under i_mutex unless
2751 * we are just creating the inode or deleting it.
2753 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2755 struct super_block *sb = inode->i_sb;
2756 struct ext4_sb_info *sbi = EXT4_SB(sb);
2757 struct ext4_iloc iloc;
2761 if (!sbi->s_journal || is_bad_inode(inode))
2764 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2765 !mutex_is_locked(&inode->i_mutex));
2767 * Exit early if inode already is on orphan list. This is a big speedup
2768 * since we don't have to contend on the global s_orphan_lock.
2770 if (!list_empty(&EXT4_I(inode)->i_orphan))
2774 * Orphan handling is only valid for files with data blocks
2775 * being truncated, or files being unlinked. Note that we either
2776 * hold i_mutex, or the inode can not be referenced from outside,
2777 * so i_nlink should not be bumped due to race
2779 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2780 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2782 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2783 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2787 err = ext4_reserve_inode_write(handle, inode, &iloc);
2791 mutex_lock(&sbi->s_orphan_lock);
2793 * Due to previous errors inode may be already a part of on-disk
2794 * orphan list. If so skip on-disk list modification.
2796 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2797 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2798 /* Insert this inode at the head of the on-disk orphan list */
2799 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2800 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2803 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2804 mutex_unlock(&sbi->s_orphan_lock);
2807 err = ext4_handle_dirty_super(handle, sb);
2808 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2813 * We have to remove inode from in-memory list if
2814 * addition to on disk orphan list failed. Stray orphan
2815 * list entries can cause panics at unmount time.
2817 mutex_lock(&sbi->s_orphan_lock);
2818 list_del_init(&EXT4_I(inode)->i_orphan);
2819 mutex_unlock(&sbi->s_orphan_lock);
2824 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2825 jbd_debug(4, "orphan inode %lu will point to %d\n",
2826 inode->i_ino, NEXT_ORPHAN(inode));
2828 ext4_std_error(sb, err);
2833 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2834 * of such inodes stored on disk, because it is finally being cleaned up.
2836 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2838 struct list_head *prev;
2839 struct ext4_inode_info *ei = EXT4_I(inode);
2840 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2842 struct ext4_iloc iloc;
2845 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2848 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2849 !mutex_is_locked(&inode->i_mutex));
2850 /* Do this quick check before taking global s_orphan_lock. */
2851 if (list_empty(&ei->i_orphan))
2855 /* Grab inode buffer early before taking global s_orphan_lock */
2856 err = ext4_reserve_inode_write(handle, inode, &iloc);
2859 mutex_lock(&sbi->s_orphan_lock);
2860 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2862 prev = ei->i_orphan.prev;
2863 list_del_init(&ei->i_orphan);
2865 /* If we're on an error path, we may not have a valid
2866 * transaction handle with which to update the orphan list on
2867 * disk, but we still need to remove the inode from the linked
2868 * list in memory. */
2869 if (!handle || err) {
2870 mutex_unlock(&sbi->s_orphan_lock);
2874 ino_next = NEXT_ORPHAN(inode);
2875 if (prev == &sbi->s_orphan) {
2876 jbd_debug(4, "superblock will point to %u\n", ino_next);
2877 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2878 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2880 mutex_unlock(&sbi->s_orphan_lock);
2883 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2884 mutex_unlock(&sbi->s_orphan_lock);
2885 err = ext4_handle_dirty_super(handle, inode->i_sb);
2887 struct ext4_iloc iloc2;
2888 struct inode *i_prev =
2889 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2891 jbd_debug(4, "orphan inode %lu will point to %u\n",
2892 i_prev->i_ino, ino_next);
2893 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2895 mutex_unlock(&sbi->s_orphan_lock);
2898 NEXT_ORPHAN(i_prev) = ino_next;
2899 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2900 mutex_unlock(&sbi->s_orphan_lock);
2904 NEXT_ORPHAN(inode) = 0;
2905 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2907 ext4_std_error(inode->i_sb, err);
2915 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2918 struct inode *inode;
2919 struct buffer_head *bh;
2920 struct ext4_dir_entry_2 *de;
2921 handle_t *handle = NULL;
2923 /* Initialize quotas before so that eventual writes go in
2924 * separate transaction */
2925 retval = dquot_initialize(dir);
2928 retval = dquot_initialize(d_inode(dentry));
2933 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2939 inode = d_inode(dentry);
2941 retval = -EFSCORRUPTED;
2942 if (le32_to_cpu(de->inode) != inode->i_ino)
2945 retval = -ENOTEMPTY;
2946 if (!ext4_empty_dir(inode))
2949 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2950 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2951 if (IS_ERR(handle)) {
2952 retval = PTR_ERR(handle);
2957 if (IS_DIRSYNC(dir))
2958 ext4_handle_sync(handle);
2960 retval = ext4_delete_entry(handle, dir, de, bh);
2963 if (!EXT4_DIR_LINK_EMPTY(inode))
2964 ext4_warning_inode(inode,
2965 "empty directory '%.*s' has too many links (%u)",
2966 dentry->d_name.len, dentry->d_name.name,
2970 /* There's no need to set i_disksize: the fact that i_nlink is
2971 * zero will ensure that the right thing happens during any
2974 ext4_orphan_add(handle, inode);
2975 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2976 ext4_mark_inode_dirty(handle, inode);
2977 ext4_dec_count(handle, dir);
2978 ext4_update_dx_flag(dir);
2979 ext4_mark_inode_dirty(handle, dir);
2984 ext4_journal_stop(handle);
2988 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2991 struct inode *inode;
2992 struct buffer_head *bh;
2993 struct ext4_dir_entry_2 *de;
2994 handle_t *handle = NULL;
2996 trace_ext4_unlink_enter(dir, dentry);
2997 /* Initialize quotas before so that eventual writes go
2998 * in separate transaction */
2999 retval = dquot_initialize(dir);
3002 retval = dquot_initialize(d_inode(dentry));
3007 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3013 inode = d_inode(dentry);
3015 retval = -EFSCORRUPTED;
3016 if (le32_to_cpu(de->inode) != inode->i_ino)
3019 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3020 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3021 if (IS_ERR(handle)) {
3022 retval = PTR_ERR(handle);
3027 if (IS_DIRSYNC(dir))
3028 ext4_handle_sync(handle);
3030 retval = ext4_delete_entry(handle, dir, de, bh);
3033 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3034 ext4_update_dx_flag(dir);
3035 ext4_mark_inode_dirty(handle, dir);
3036 if (inode->i_nlink == 0)
3037 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3038 dentry->d_name.len, dentry->d_name.name);
3041 if (!inode->i_nlink)
3042 ext4_orphan_add(handle, inode);
3043 inode->i_ctime = ext4_current_time(inode);
3044 ext4_mark_inode_dirty(handle, inode);
3049 ext4_journal_stop(handle);
3050 trace_ext4_unlink_exit(dentry, retval);
3054 static int ext4_symlink(struct inode *dir,
3055 struct dentry *dentry, const char *symname)
3058 struct inode *inode;
3059 int err, len = strlen(symname);
3061 bool encryption_required;
3062 struct ext4_str disk_link;
3063 struct ext4_encrypted_symlink_data *sd = NULL;
3065 disk_link.len = len + 1;
3066 disk_link.name = (char *) symname;
3068 encryption_required = (ext4_encrypted_inode(dir) ||
3069 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3070 if (encryption_required) {
3071 err = ext4_get_encryption_info(dir);
3074 if (ext4_encryption_info(dir) == NULL)
3076 disk_link.len = (ext4_fname_encrypted_size(dir, len) +
3077 sizeof(struct ext4_encrypted_symlink_data));
3078 sd = kzalloc(disk_link.len, GFP_KERNEL);
3083 if (disk_link.len > dir->i_sb->s_blocksize) {
3084 err = -ENAMETOOLONG;
3088 err = dquot_initialize(dir);
3092 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3094 * For non-fast symlinks, we just allocate inode and put it on
3095 * orphan list in the first transaction => we need bitmap,
3096 * group descriptor, sb, inode block, quota blocks, and
3097 * possibly selinux xattr blocks.
3099 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3100 EXT4_XATTR_TRANS_BLOCKS;
3103 * Fast symlink. We have to add entry to directory
3104 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3105 * allocate new inode (bitmap, group descriptor, inode block,
3106 * quota blocks, sb is already counted in previous macros).
3108 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3109 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3112 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3113 &dentry->d_name, 0, NULL,
3114 EXT4_HT_DIR, credits);
3115 handle = ext4_journal_current_handle();
3116 if (IS_ERR(inode)) {
3118 ext4_journal_stop(handle);
3119 err = PTR_ERR(inode);
3123 if (encryption_required) {
3125 struct ext4_str ostr;
3127 istr.name = (const unsigned char *) symname;
3129 ostr.name = sd->encrypted_path;
3130 ostr.len = disk_link.len;
3131 err = ext4_fname_usr_to_disk(inode, &istr, &ostr);
3133 goto err_drop_inode;
3134 sd->len = cpu_to_le16(ostr.len);
3135 disk_link.name = (char *) sd;
3136 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3139 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3140 if (!encryption_required)
3141 inode->i_op = &ext4_symlink_inode_operations;
3142 inode_nohighmem(inode);
3143 ext4_set_aops(inode);
3145 * We cannot call page_symlink() with transaction started
3146 * because it calls into ext4_write_begin() which can wait
3147 * for transaction commit if we are running out of space
3148 * and thus we deadlock. So we have to stop transaction now
3149 * and restart it when symlink contents is written.
3151 * To keep fs consistent in case of crash, we have to put inode
3152 * to orphan list in the mean time.
3155 err = ext4_orphan_add(handle, inode);
3156 ext4_journal_stop(handle);
3159 goto err_drop_inode;
3160 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3162 goto err_drop_inode;
3164 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3165 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3167 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3168 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3169 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3170 if (IS_ERR(handle)) {
3171 err = PTR_ERR(handle);
3173 goto err_drop_inode;
3175 set_nlink(inode, 1);
3176 err = ext4_orphan_del(handle, inode);
3178 goto err_drop_inode;
3180 /* clear the extent format for fast symlink */
3181 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3182 if (!encryption_required) {
3183 inode->i_op = &ext4_fast_symlink_inode_operations;
3184 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3186 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3188 inode->i_size = disk_link.len - 1;
3190 EXT4_I(inode)->i_disksize = inode->i_size;
3191 err = ext4_add_nondir(handle, dentry, inode);
3192 if (!err && IS_DIRSYNC(dir))
3193 ext4_handle_sync(handle);
3196 ext4_journal_stop(handle);
3201 ext4_journal_stop(handle);
3203 unlock_new_inode(inode);
3210 static int ext4_link(struct dentry *old_dentry,
3211 struct inode *dir, struct dentry *dentry)
3214 struct inode *inode = d_inode(old_dentry);
3215 int err, retries = 0;
3217 if (inode->i_nlink >= EXT4_LINK_MAX)
3219 if (ext4_encrypted_inode(dir) &&
3220 !ext4_is_child_context_consistent_with_parent(dir, inode))
3222 err = dquot_initialize(dir);
3227 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3228 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3229 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3231 return PTR_ERR(handle);
3233 if (IS_DIRSYNC(dir))
3234 ext4_handle_sync(handle);
3236 inode->i_ctime = ext4_current_time(inode);
3237 ext4_inc_count(handle, inode);
3240 err = ext4_add_entry(handle, dentry, inode);
3242 ext4_mark_inode_dirty(handle, inode);
3243 /* this can happen only for tmpfile being
3244 * linked the first time
3246 if (inode->i_nlink == 1)
3247 ext4_orphan_del(handle, inode);
3248 d_instantiate(dentry, inode);
3253 ext4_journal_stop(handle);
3254 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3261 * Try to find buffer head where contains the parent block.
3262 * It should be the inode block if it is inlined or the 1st block
3263 * if it is a normal dir.
3265 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3266 struct inode *inode,
3268 struct ext4_dir_entry_2 **parent_de,
3271 struct buffer_head *bh;
3273 if (!ext4_has_inline_data(inode)) {
3274 bh = ext4_read_dirblock(inode, 0, EITHER);
3276 *retval = PTR_ERR(bh);
3279 *parent_de = ext4_next_entry(
3280 (struct ext4_dir_entry_2 *)bh->b_data,
3281 inode->i_sb->s_blocksize);
3286 return ext4_get_first_inline_block(inode, parent_de, retval);
3289 struct ext4_renament {
3291 struct dentry *dentry;
3292 struct inode *inode;
3294 int dir_nlink_delta;
3296 /* entry for "dentry" */
3297 struct buffer_head *bh;
3298 struct ext4_dir_entry_2 *de;
3301 /* entry for ".." in inode if it's a directory */
3302 struct buffer_head *dir_bh;
3303 struct ext4_dir_entry_2 *parent_de;
3307 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3311 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3312 &retval, &ent->parent_de,
3316 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3317 return -EFSCORRUPTED;
3318 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3319 return ext4_journal_get_write_access(handle, ent->dir_bh);
3322 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3327 ent->parent_de->inode = cpu_to_le32(dir_ino);
3328 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3329 if (!ent->dir_inlined) {
3330 if (is_dx(ent->inode)) {
3331 retval = ext4_handle_dirty_dx_node(handle,
3335 retval = ext4_handle_dirty_dirent_node(handle,
3340 retval = ext4_mark_inode_dirty(handle, ent->inode);
3343 ext4_std_error(ent->dir->i_sb, retval);
3349 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3350 unsigned ino, unsigned file_type)
3354 BUFFER_TRACE(ent->bh, "get write access");
3355 retval = ext4_journal_get_write_access(handle, ent->bh);
3358 ent->de->inode = cpu_to_le32(ino);
3359 if (ext4_has_feature_filetype(ent->dir->i_sb))
3360 ent->de->file_type = file_type;
3361 ent->dir->i_version++;
3362 ent->dir->i_ctime = ent->dir->i_mtime =
3363 ext4_current_time(ent->dir);
3364 ext4_mark_inode_dirty(handle, ent->dir);
3365 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3366 if (!ent->inlined) {
3367 retval = ext4_handle_dirty_dirent_node(handle,
3369 if (unlikely(retval)) {
3370 ext4_std_error(ent->dir->i_sb, retval);
3378 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3379 unsigned ino, unsigned file_type)
3381 struct ext4_renament old = *ent;
3385 * old->de could have moved from under us during make indexed dir,
3386 * so the old->de may no longer valid and need to find it again
3387 * before reset old inode info.
3389 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3391 retval = PTR_ERR(old.bh);
3395 ext4_std_error(old.dir->i_sb, retval);
3399 ext4_setent(handle, &old, ino, file_type);
3403 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3404 const struct qstr *d_name)
3406 int retval = -ENOENT;
3407 struct buffer_head *bh;
3408 struct ext4_dir_entry_2 *de;
3410 bh = ext4_find_entry(dir, d_name, &de, NULL);
3414 retval = ext4_delete_entry(handle, dir, de, bh);
3420 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3425 * ent->de could have moved from under us during htree split, so make
3426 * sure that we are deleting the right entry. We might also be pointing
3427 * to a stale entry in the unused part of ent->bh so just checking inum
3428 * and the name isn't enough.
3430 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3431 ent->de->name_len != ent->dentry->d_name.len ||
3432 strncmp(ent->de->name, ent->dentry->d_name.name,
3433 ent->de->name_len) ||
3435 retval = ext4_find_delete_entry(handle, ent->dir,
3436 &ent->dentry->d_name);
3438 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3439 if (retval == -ENOENT) {
3440 retval = ext4_find_delete_entry(handle, ent->dir,
3441 &ent->dentry->d_name);
3446 ext4_warning_inode(ent->dir,
3447 "Deleting old file: nlink %d, error=%d",
3448 ent->dir->i_nlink, retval);
3452 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3454 if (ent->dir_nlink_delta) {
3455 if (ent->dir_nlink_delta == -1)
3456 ext4_dec_count(handle, ent->dir);
3458 ext4_inc_count(handle, ent->dir);
3459 ext4_mark_inode_dirty(handle, ent->dir);
3463 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3464 int credits, handle_t **h)
3471 * for inode block, sb block, group summaries,
3474 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3475 EXT4_XATTR_TRANS_BLOCKS + 4);
3477 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3478 &ent->dentry->d_name, 0, NULL,
3479 EXT4_HT_DIR, credits);
3481 handle = ext4_journal_current_handle();
3484 ext4_journal_stop(handle);
3485 if (PTR_ERR(wh) == -ENOSPC &&
3486 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3490 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3491 wh->i_op = &ext4_special_inode_operations;
3497 * Anybody can rename anything with this: the permission checks are left to the
3498 * higher-level routines.
3500 * n.b. old_{dentry,inode) refers to the source dentry/inode
3501 * while new_{dentry,inode) refers to the destination dentry/inode
3502 * This comes from rename(const char *oldpath, const char *newpath)
3504 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3505 struct inode *new_dir, struct dentry *new_dentry,
3508 handle_t *handle = NULL;
3509 struct ext4_renament old = {
3511 .dentry = old_dentry,
3512 .inode = d_inode(old_dentry),
3514 struct ext4_renament new = {
3516 .dentry = new_dentry,
3517 .inode = d_inode(new_dentry),
3521 struct inode *whiteout = NULL;
3525 if ((ext4_encrypted_inode(old_dir) &&
3526 !ext4_has_encryption_key(old_dir)) ||
3527 (ext4_encrypted_inode(new_dir) &&
3528 !ext4_has_encryption_key(new_dir)))
3531 retval = dquot_initialize(old.dir);
3534 retval = dquot_initialize(new.dir);
3538 /* Initialize quotas before so that eventual writes go
3539 * in separate transaction */
3541 retval = dquot_initialize(new.inode);
3546 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3548 return PTR_ERR(old.bh);
3550 * Check for inode number is _not_ due to possible IO errors.
3551 * We might rmdir the source, keep it as pwd of some process
3552 * and merrily kill the link to whatever was created under the
3553 * same name. Goodbye sticky bit ;-<
3556 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3559 if ((old.dir != new.dir) &&
3560 ext4_encrypted_inode(new.dir) &&
3561 !ext4_is_child_context_consistent_with_parent(new.dir,
3567 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3568 &new.de, &new.inlined);
3569 if (IS_ERR(new.bh)) {
3570 retval = PTR_ERR(new.bh);
3580 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3581 ext4_alloc_da_blocks(old.inode);
3583 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3584 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3585 if (!(flags & RENAME_WHITEOUT)) {
3586 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3587 if (IS_ERR(handle)) {
3588 retval = PTR_ERR(handle);
3592 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3593 if (IS_ERR(whiteout)) {
3594 retval = PTR_ERR(whiteout);
3599 old_file_type = old.de->file_type;
3600 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3601 ext4_handle_sync(handle);
3603 if (S_ISDIR(old.inode->i_mode)) {
3605 retval = -ENOTEMPTY;
3606 if (!ext4_empty_dir(new.inode))
3610 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3613 retval = ext4_rename_dir_prepare(handle, &old);
3618 * If we're renaming a file within an inline_data dir and adding or
3619 * setting the new dirent causes a conversion from inline_data to
3620 * extents/blockmap, we need to force the dirent delete code to
3621 * re-read the directory, or else we end up trying to delete a dirent
3622 * from what is now the extent tree root (or a block map).
3624 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3625 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3629 * Do this before adding a new entry, so the old entry is sure
3630 * to be still pointing to the valid old entry.
3632 retval = ext4_setent(handle, &old, whiteout->i_ino,
3636 ext4_mark_inode_dirty(handle, whiteout);
3639 retval = ext4_add_entry(handle, new.dentry, old.inode);
3643 retval = ext4_setent(handle, &new,
3644 old.inode->i_ino, old_file_type);
3649 force_reread = !ext4_test_inode_flag(new.dir,
3650 EXT4_INODE_INLINE_DATA);
3653 * Like most other Unix systems, set the ctime for inodes on a
3656 old.inode->i_ctime = ext4_current_time(old.inode);
3657 ext4_mark_inode_dirty(handle, old.inode);
3663 ext4_rename_delete(handle, &old, force_reread);
3667 ext4_dec_count(handle, new.inode);
3668 new.inode->i_ctime = ext4_current_time(new.inode);
3670 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3671 ext4_update_dx_flag(old.dir);
3673 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3677 ext4_dec_count(handle, old.dir);
3679 /* checked ext4_empty_dir above, can't have another
3680 * parent, ext4_dec_count() won't work for many-linked
3682 clear_nlink(new.inode);
3684 ext4_inc_count(handle, new.dir);
3685 ext4_update_dx_flag(new.dir);
3686 ext4_mark_inode_dirty(handle, new.dir);
3689 ext4_mark_inode_dirty(handle, old.dir);
3691 ext4_mark_inode_dirty(handle, new.inode);
3692 if (!new.inode->i_nlink)
3693 ext4_orphan_add(handle, new.inode);
3700 ext4_resetent(handle, &old,
3701 old.inode->i_ino, old_file_type);
3702 drop_nlink(whiteout);
3703 ext4_orphan_add(handle, whiteout);
3705 unlock_new_inode(whiteout);
3706 ext4_journal_stop(handle);
3709 ext4_journal_stop(handle);
3718 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3719 struct inode *new_dir, struct dentry *new_dentry)
3721 handle_t *handle = NULL;
3722 struct ext4_renament old = {
3724 .dentry = old_dentry,
3725 .inode = d_inode(old_dentry),
3727 struct ext4_renament new = {
3729 .dentry = new_dentry,
3730 .inode = d_inode(new_dentry),
3735 if ((ext4_encrypted_inode(old_dir) &&
3736 !ext4_has_encryption_key(old_dir)) ||
3737 (ext4_encrypted_inode(new_dir) &&
3738 !ext4_has_encryption_key(new_dir)))
3741 if ((ext4_encrypted_inode(old_dir) ||
3742 ext4_encrypted_inode(new_dir)) &&
3743 (old_dir != new_dir) &&
3744 (!ext4_is_child_context_consistent_with_parent(new_dir,
3746 !ext4_is_child_context_consistent_with_parent(old_dir,
3750 retval = dquot_initialize(old.dir);
3753 retval = dquot_initialize(new.dir);
3757 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3758 &old.de, &old.inlined);
3760 return PTR_ERR(old.bh);
3762 * Check for inode number is _not_ due to possible IO errors.
3763 * We might rmdir the source, keep it as pwd of some process
3764 * and merrily kill the link to whatever was created under the
3765 * same name. Goodbye sticky bit ;-<
3768 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3771 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3772 &new.de, &new.inlined);
3773 if (IS_ERR(new.bh)) {
3774 retval = PTR_ERR(new.bh);
3779 /* RENAME_EXCHANGE case: old *and* new must both exist */
3780 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3783 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3784 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3785 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3786 if (IS_ERR(handle)) {
3787 retval = PTR_ERR(handle);
3792 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3793 ext4_handle_sync(handle);
3795 if (S_ISDIR(old.inode->i_mode)) {
3797 retval = ext4_rename_dir_prepare(handle, &old);
3801 if (S_ISDIR(new.inode->i_mode)) {
3803 retval = ext4_rename_dir_prepare(handle, &new);
3809 * Other than the special case of overwriting a directory, parents'
3810 * nlink only needs to be modified if this is a cross directory rename.
3812 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3813 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3814 new.dir_nlink_delta = -old.dir_nlink_delta;
3816 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3817 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3821 new_file_type = new.de->file_type;
3822 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3826 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3831 * Like most other Unix systems, set the ctime for inodes on a
3834 old.inode->i_ctime = ext4_current_time(old.inode);
3835 new.inode->i_ctime = ext4_current_time(new.inode);
3836 ext4_mark_inode_dirty(handle, old.inode);
3837 ext4_mark_inode_dirty(handle, new.inode);
3840 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3845 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3849 ext4_update_dir_count(handle, &old);
3850 ext4_update_dir_count(handle, &new);
3859 ext4_journal_stop(handle);
3863 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3864 struct inode *new_dir, struct dentry *new_dentry,
3867 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3870 if (flags & RENAME_EXCHANGE) {
3871 return ext4_cross_rename(old_dir, old_dentry,
3872 new_dir, new_dentry);
3875 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3879 * directories can handle most operations...
3881 const struct inode_operations ext4_dir_inode_operations = {
3882 .create = ext4_create,
3883 .lookup = ext4_lookup,
3885 .unlink = ext4_unlink,
3886 .symlink = ext4_symlink,
3887 .mkdir = ext4_mkdir,
3888 .rmdir = ext4_rmdir,
3889 .mknod = ext4_mknod,
3890 .tmpfile = ext4_tmpfile,
3891 .rename2 = ext4_rename2,
3892 .setattr = ext4_setattr,
3893 .setxattr = generic_setxattr,
3894 .getxattr = generic_getxattr,
3895 .listxattr = ext4_listxattr,
3896 .removexattr = generic_removexattr,
3897 .get_acl = ext4_get_acl,
3898 .set_acl = ext4_set_acl,
3899 .fiemap = ext4_fiemap,
3902 const struct inode_operations ext4_special_inode_operations = {
3903 .setattr = ext4_setattr,
3904 .setxattr = generic_setxattr,
3905 .getxattr = generic_getxattr,
3906 .listxattr = ext4_listxattr,
3907 .removexattr = generic_removexattr,
3908 .get_acl = ext4_get_acl,
3909 .set_acl = ext4_set_acl,