2 * linux/fs/ext4/namei.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/namei.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
28 #include <linux/pagemap.h>
29 #include <linux/time.h>
30 #include <linux/fcntl.h>
31 #include <linux/stat.h>
32 #include <linux/string.h>
33 #include <linux/quotaops.h>
34 #include <linux/buffer_head.h>
35 #include <linux/bio.h>
37 #include "ext4_jbd2.h"
42 #include <trace/events/ext4.h>
44 * define how far ahead to read directories while searching them.
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 static struct buffer_head *ext4_append(handle_t *handle,
54 struct buffer_head *bh;
57 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
58 ((inode->i_size >> 10) >=
59 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
60 return ERR_PTR(-ENOSPC);
62 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
64 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
67 inode->i_size += inode->i_sb->s_blocksize;
68 EXT4_I(inode)->i_disksize = inode->i_size;
69 BUFFER_TRACE(bh, "get_write_access");
70 err = ext4_journal_get_write_access(handle, bh);
73 ext4_std_error(inode->i_sb, err);
79 static int ext4_dx_csum_verify(struct inode *inode,
80 struct ext4_dir_entry *dirent);
86 #define ext4_read_dirblock(inode, block, type) \
87 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
95 struct buffer_head *bh;
96 struct ext4_dir_entry *dirent;
99 bh = ext4_bread(NULL, inode, block, 0);
101 __ext4_warning(inode->i_sb, func, line,
102 "inode #%lu: lblock %lu: comm %s: "
103 "error %ld reading directory block",
104 inode->i_ino, (unsigned long)block,
105 current->comm, PTR_ERR(bh));
110 ext4_error_inode(inode, func, line, block,
111 "Directory hole found");
112 return ERR_PTR(-EFSCORRUPTED);
114 dirent = (struct ext4_dir_entry *) bh->b_data;
115 /* Determine whether or not we have an index block */
119 else if (ext4_rec_len_from_disk(dirent->rec_len,
120 inode->i_sb->s_blocksize) ==
121 inode->i_sb->s_blocksize)
124 if (!is_dx_block && type == INDEX) {
125 ext4_error_inode(inode, func, line, block,
126 "directory leaf block found instead of index block");
127 return ERR_PTR(-EFSCORRUPTED);
129 if (!ext4_has_metadata_csum(inode->i_sb) ||
134 * An empty leaf block can get mistaken for a index block; for
135 * this reason, we can only check the index checksum when the
136 * caller is sure it should be an index block.
138 if (is_dx_block && type == INDEX) {
139 if (ext4_dx_csum_verify(inode, dirent))
140 set_buffer_verified(bh);
142 ext4_error_inode(inode, func, line, block,
143 "Directory index failed checksum");
145 return ERR_PTR(-EFSBADCRC);
149 if (ext4_dirent_csum_verify(inode, dirent))
150 set_buffer_verified(bh);
152 ext4_error_inode(inode, func, line, block,
153 "Directory block failed checksum");
155 return ERR_PTR(-EFSBADCRC);
162 #define assert(test) J_ASSERT(test)
166 #define dxtrace(command) command
168 #define dxtrace(command)
192 * dx_root_info is laid out so that if it should somehow get overlaid by a
193 * dirent the two low bits of the hash version will be zero. Therefore, the
194 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
199 struct fake_dirent dot;
201 struct fake_dirent dotdot;
205 __le32 reserved_zero;
207 u8 info_length; /* 8 */
212 struct dx_entry entries[0];
217 struct fake_dirent fake;
218 struct dx_entry entries[0];
224 struct buffer_head *bh;
225 struct dx_entry *entries;
237 * This goes at the end of each htree block.
241 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
244 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
245 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
246 static inline unsigned dx_get_hash(struct dx_entry *entry);
247 static void dx_set_hash(struct dx_entry *entry, unsigned value);
248 static unsigned dx_get_count(struct dx_entry *entries);
249 static unsigned dx_get_limit(struct dx_entry *entries);
250 static void dx_set_count(struct dx_entry *entries, unsigned value);
251 static void dx_set_limit(struct dx_entry *entries, unsigned value);
252 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
253 static unsigned dx_node_limit(struct inode *dir);
254 static struct dx_frame *dx_probe(struct ext4_filename *fname,
256 struct dx_hash_info *hinfo,
257 struct dx_frame *frame);
258 static void dx_release(struct dx_frame *frames);
259 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
260 unsigned blocksize, struct dx_hash_info *hinfo,
261 struct dx_map_entry map[]);
262 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
263 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
264 struct dx_map_entry *offsets, int count, unsigned blocksize);
265 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
266 static void dx_insert_block(struct dx_frame *frame,
267 u32 hash, ext4_lblk_t block);
268 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
269 struct dx_frame *frame,
270 struct dx_frame *frames,
272 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
273 struct ext4_filename *fname,
274 struct ext4_dir_entry_2 **res_dir);
275 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
276 struct inode *dir, struct inode *inode);
278 /* checksumming functions */
279 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
280 unsigned int blocksize)
282 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
283 t->det_rec_len = ext4_rec_len_to_disk(
284 sizeof(struct ext4_dir_entry_tail), blocksize);
285 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
288 /* Walk through a dirent block to find a checksum "dirent" at the tail */
289 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
290 struct ext4_dir_entry *de)
292 struct ext4_dir_entry_tail *t;
295 struct ext4_dir_entry *d, *top;
298 top = (struct ext4_dir_entry *)(((void *)de) +
299 (EXT4_BLOCK_SIZE(inode->i_sb) -
300 sizeof(struct ext4_dir_entry_tail)));
301 while (d < top && d->rec_len)
302 d = (struct ext4_dir_entry *)(((void *)d) +
303 le16_to_cpu(d->rec_len));
308 t = (struct ext4_dir_entry_tail *)d;
310 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
313 if (t->det_reserved_zero1 ||
314 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
315 t->det_reserved_zero2 ||
316 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
322 static __le32 ext4_dirent_csum(struct inode *inode,
323 struct ext4_dir_entry *dirent, int size)
325 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
326 struct ext4_inode_info *ei = EXT4_I(inode);
329 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
330 return cpu_to_le32(csum);
333 #define warn_no_space_for_csum(inode) \
334 __warn_no_space_for_csum((inode), __func__, __LINE__)
336 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
339 __ext4_warning_inode(inode, func, line,
340 "No space for directory leaf checksum. Please run e2fsck -D.");
343 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
345 struct ext4_dir_entry_tail *t;
347 if (!ext4_has_metadata_csum(inode->i_sb))
350 t = get_dirent_tail(inode, dirent);
352 warn_no_space_for_csum(inode);
356 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
357 (void *)t - (void *)dirent))
363 static void ext4_dirent_csum_set(struct inode *inode,
364 struct ext4_dir_entry *dirent)
366 struct ext4_dir_entry_tail *t;
368 if (!ext4_has_metadata_csum(inode->i_sb))
371 t = get_dirent_tail(inode, dirent);
373 warn_no_space_for_csum(inode);
377 t->det_checksum = ext4_dirent_csum(inode, dirent,
378 (void *)t - (void *)dirent);
381 int ext4_handle_dirty_dirent_node(handle_t *handle,
383 struct buffer_head *bh)
385 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
386 return ext4_handle_dirty_metadata(handle, inode, bh);
389 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
390 struct ext4_dir_entry *dirent,
393 struct ext4_dir_entry *dp;
394 struct dx_root_info *root;
397 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
399 else if (le16_to_cpu(dirent->rec_len) == 12) {
400 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
401 if (le16_to_cpu(dp->rec_len) !=
402 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
404 root = (struct dx_root_info *)(((void *)dp + 12));
405 if (root->reserved_zero ||
406 root->info_length != sizeof(struct dx_root_info))
413 *offset = count_offset;
414 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
417 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
418 int count_offset, int count, struct dx_tail *t)
420 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
421 struct ext4_inode_info *ei = EXT4_I(inode);
426 size = count_offset + (count * sizeof(struct dx_entry));
427 save_csum = t->dt_checksum;
429 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
430 csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
431 t->dt_checksum = save_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 if (fatal_signal_pending(current)) {
1115 block = dx_get_block(frame->at);
1116 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1117 start_hash, start_minor_hash);
1124 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1125 frame, frames, &hashval);
1126 *next_hash = hashval;
1132 * Stop if: (a) there are no more entries, or
1133 * (b) we have inserted at least one entry and the
1134 * next hash value is not a continuation
1137 (count && ((hashval & 1) == 0)))
1141 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1142 "next hash: %x\n", count, *next_hash));
1149 static inline int search_dirblock(struct buffer_head *bh,
1151 struct ext4_filename *fname,
1152 const struct qstr *d_name,
1153 unsigned int offset,
1154 struct ext4_dir_entry_2 **res_dir)
1156 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1157 fname, d_name, offset, res_dir);
1161 * Directory block splitting, compacting
1165 * Create map of hash values, offsets, and sizes, stored at end of block.
1166 * Returns number of entries mapped.
1168 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1169 unsigned blocksize, struct dx_hash_info *hinfo,
1170 struct dx_map_entry *map_tail)
1173 char *base = (char *) de;
1174 struct dx_hash_info h = *hinfo;
1176 while ((char *) de < base + blocksize) {
1177 if (de->name_len && de->inode) {
1178 ext4fs_dirhash(de->name, de->name_len, &h);
1180 map_tail->hash = h.hash;
1181 map_tail->offs = ((char *) de - base)>>2;
1182 map_tail->size = le16_to_cpu(de->rec_len);
1186 /* XXX: do we need to check rec_len == 0 case? -Chris */
1187 de = ext4_next_entry(de, blocksize);
1192 /* Sort map by hash value */
1193 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1195 struct dx_map_entry *p, *q, *top = map + count - 1;
1197 /* Combsort until bubble sort doesn't suck */
1199 count = count*10/13;
1200 if (count - 9 < 2) /* 9, 10 -> 11 */
1202 for (p = top, q = p - count; q >= map; p--, q--)
1203 if (p->hash < q->hash)
1206 /* Garden variety bubble sort */
1211 if (q[1].hash >= q[0].hash)
1219 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1221 struct dx_entry *entries = frame->entries;
1222 struct dx_entry *old = frame->at, *new = old + 1;
1223 int count = dx_get_count(entries);
1225 assert(count < dx_get_limit(entries));
1226 assert(old < entries + count);
1227 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1228 dx_set_hash(new, hash);
1229 dx_set_block(new, block);
1230 dx_set_count(entries, count + 1);
1234 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1236 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1237 * `de != NULL' is guaranteed by caller.
1239 static inline int ext4_match(struct ext4_filename *fname,
1240 struct ext4_dir_entry_2 *de)
1242 const void *name = fname_name(fname);
1243 u32 len = fname_len(fname);
1248 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1249 if (unlikely(!name)) {
1250 if (fname->usr_fname->name[0] == '_') {
1252 if (de->name_len < 16)
1254 ret = memcmp(de->name + de->name_len - 16,
1255 fname->crypto_buf.name + 8, 16);
1256 return (ret == 0) ? 1 : 0;
1258 name = fname->crypto_buf.name;
1259 len = fname->crypto_buf.len;
1262 if (de->name_len != len)
1264 return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1268 * Returns 0 if not found, -1 on failure, and 1 on success
1270 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1271 struct inode *dir, struct ext4_filename *fname,
1272 const struct qstr *d_name,
1273 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1275 struct ext4_dir_entry_2 * de;
1280 de = (struct ext4_dir_entry_2 *)search_buf;
1281 dlimit = search_buf + buf_size;
1282 while ((char *) de < dlimit) {
1283 /* this code is executed quadratically often */
1284 /* do minimal checking `by hand' */
1285 if ((char *) de + de->name_len <= dlimit) {
1286 res = ext4_match(fname, de);
1292 /* found a match - just to be sure, do
1294 if (ext4_check_dir_entry(dir, NULL, de, bh,
1296 bh->b_size, offset)) {
1306 /* prevent looping on a bad block */
1307 de_len = ext4_rec_len_from_disk(de->rec_len,
1308 dir->i_sb->s_blocksize);
1314 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1322 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1323 struct ext4_dir_entry *de)
1325 struct super_block *sb = dir->i_sb;
1331 if (de->inode == 0 &&
1332 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1341 * finds an entry in the specified directory with the wanted name. It
1342 * returns the cache buffer in which the entry was found, and the entry
1343 * itself (as a parameter - res_dir). It does NOT read the inode of the
1344 * entry - you'll have to do that yourself if you want to.
1346 * The returned buffer_head has ->b_count elevated. The caller is expected
1347 * to brelse() it when appropriate.
1349 static struct buffer_head * ext4_find_entry (struct inode *dir,
1350 const struct qstr *d_name,
1351 struct ext4_dir_entry_2 **res_dir,
1354 struct super_block *sb;
1355 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1356 struct buffer_head *bh, *ret = NULL;
1357 ext4_lblk_t start, block, b;
1358 const u8 *name = d_name->name;
1359 int ra_max = 0; /* Number of bh's in the readahead
1361 int ra_ptr = 0; /* Current index into readahead
1364 ext4_lblk_t nblocks;
1365 int i, namelen, retval;
1366 struct ext4_filename fname;
1370 namelen = d_name->len;
1371 if (namelen > EXT4_NAME_LEN)
1374 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1376 return ERR_PTR(retval);
1378 if (ext4_has_inline_data(dir)) {
1379 int has_inline_data = 1;
1380 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1382 if (has_inline_data) {
1385 goto cleanup_and_exit;
1389 if ((namelen <= 2) && (name[0] == '.') &&
1390 (name[1] == '.' || name[1] == '\0')) {
1392 * "." or ".." will only be in the first block
1393 * NFS may look up ".."; "." should be handled by the VFS
1400 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1402 * On success, or if the error was file not found,
1403 * return. Otherwise, fall back to doing a search the
1404 * old fashioned way.
1406 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1407 goto cleanup_and_exit;
1408 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1411 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1412 start = EXT4_I(dir)->i_dir_start_lookup;
1413 if (start >= nblocks)
1419 * We deal with the read-ahead logic here.
1421 if (ra_ptr >= ra_max) {
1422 /* Refill the readahead buffer */
1425 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1427 * Terminate if we reach the end of the
1428 * directory and must wrap, or if our
1429 * search has finished at this block.
1431 if (b >= nblocks || (num && block == start)) {
1432 bh_use[ra_max] = NULL;
1436 bh = ext4_getblk(NULL, dir, b++, 0);
1440 goto cleanup_and_exit;
1444 bh_use[ra_max] = bh;
1446 ll_rw_block(REQ_OP_READ,
1447 REQ_META | REQ_PRIO,
1451 if ((bh = bh_use[ra_ptr++]) == NULL)
1454 if (!buffer_uptodate(bh)) {
1455 /* read error, skip block & hope for the best */
1456 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1457 (unsigned long) block);
1461 if (!buffer_verified(bh) &&
1462 !is_dx_internal_node(dir, block,
1463 (struct ext4_dir_entry *)bh->b_data) &&
1464 !ext4_dirent_csum_verify(dir,
1465 (struct ext4_dir_entry *)bh->b_data)) {
1466 EXT4_ERROR_INODE(dir, "checksumming directory "
1467 "block %lu", (unsigned long)block);
1471 set_buffer_verified(bh);
1472 i = search_dirblock(bh, dir, &fname, d_name,
1473 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1475 EXT4_I(dir)->i_dir_start_lookup = block;
1477 goto cleanup_and_exit;
1481 goto cleanup_and_exit;
1484 if (++block >= nblocks)
1486 } while (block != start);
1489 * If the directory has grown while we were searching, then
1490 * search the last part of the directory before giving up.
1493 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1494 if (block < nblocks) {
1500 /* Clean up the read-ahead blocks */
1501 for (; ra_ptr < ra_max; ra_ptr++)
1502 brelse(bh_use[ra_ptr]);
1503 ext4_fname_free_filename(&fname);
1507 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1508 struct ext4_filename *fname,
1509 struct ext4_dir_entry_2 **res_dir)
1511 struct super_block * sb = dir->i_sb;
1512 struct dx_frame frames[2], *frame;
1513 const struct qstr *d_name = fname->usr_fname;
1514 struct buffer_head *bh;
1518 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1521 frame = dx_probe(fname, dir, NULL, frames);
1523 return (struct buffer_head *) frame;
1525 block = dx_get_block(frame->at);
1526 bh = ext4_read_dirblock(dir, block, DIRENT);
1530 retval = search_dirblock(bh, dir, fname, d_name,
1531 block << EXT4_BLOCK_SIZE_BITS(sb),
1537 bh = ERR_PTR(ERR_BAD_DX_DIR);
1541 /* Check to see if we should continue to search */
1542 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1545 ext4_warning_inode(dir,
1546 "error %d reading directory index block",
1548 bh = ERR_PTR(retval);
1551 } while (retval == 1);
1555 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1561 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1563 struct inode *inode;
1564 struct ext4_dir_entry_2 *de;
1565 struct buffer_head *bh;
1567 if (ext4_encrypted_inode(dir)) {
1568 int res = ext4_get_encryption_info(dir);
1571 * This should be a properly defined flag for
1572 * dentry->d_flags when we uplift this to the VFS.
1573 * d_fsdata is set to (void *) 1 if if the dentry is
1574 * created while the directory was encrypted and we
1575 * don't have access to the key.
1577 dentry->d_fsdata = NULL;
1578 if (ext4_encryption_info(dir))
1579 dentry->d_fsdata = (void *) 1;
1580 d_set_d_op(dentry, &ext4_encrypted_d_ops);
1581 if (res && res != -ENOKEY)
1582 return ERR_PTR(res);
1585 if (dentry->d_name.len > EXT4_NAME_LEN)
1586 return ERR_PTR(-ENAMETOOLONG);
1588 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1590 return (struct dentry *) bh;
1593 __u32 ino = le32_to_cpu(de->inode);
1595 if (!ext4_valid_inum(dir->i_sb, ino)) {
1596 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1597 return ERR_PTR(-EFSCORRUPTED);
1599 if (unlikely(ino == dir->i_ino)) {
1600 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1602 return ERR_PTR(-EFSCORRUPTED);
1604 inode = ext4_iget_normal(dir->i_sb, ino);
1605 if (inode == ERR_PTR(-ESTALE)) {
1606 EXT4_ERROR_INODE(dir,
1607 "deleted inode referenced: %u",
1609 return ERR_PTR(-EFSCORRUPTED);
1611 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1612 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1613 !ext4_is_child_context_consistent_with_parent(dir,
1615 int nokey = ext4_encrypted_inode(inode) &&
1616 !ext4_encryption_info(inode);
1620 return ERR_PTR(-ENOKEY);
1621 ext4_warning(inode->i_sb,
1622 "Inconsistent encryption contexts: %lu/%lu",
1623 (unsigned long) dir->i_ino,
1624 (unsigned long) inode->i_ino);
1625 return ERR_PTR(-EPERM);
1628 return d_splice_alias(inode, dentry);
1632 struct dentry *ext4_get_parent(struct dentry *child)
1635 static const struct qstr dotdot = QSTR_INIT("..", 2);
1636 struct ext4_dir_entry_2 * de;
1637 struct buffer_head *bh;
1639 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1641 return (struct dentry *) bh;
1643 return ERR_PTR(-ENOENT);
1644 ino = le32_to_cpu(de->inode);
1647 if (!ext4_valid_inum(child->d_sb, ino)) {
1648 EXT4_ERROR_INODE(d_inode(child),
1649 "bad parent inode number: %u", ino);
1650 return ERR_PTR(-EFSCORRUPTED);
1653 return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1657 * Move count entries from end of map between two memory locations.
1658 * Returns pointer to last entry moved.
1660 static struct ext4_dir_entry_2 *
1661 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1664 unsigned rec_len = 0;
1667 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1668 (from + (map->offs<<2));
1669 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1670 memcpy (to, de, rec_len);
1671 ((struct ext4_dir_entry_2 *) to)->rec_len =
1672 ext4_rec_len_to_disk(rec_len, blocksize);
1677 return (struct ext4_dir_entry_2 *) (to - rec_len);
1681 * Compact each dir entry in the range to the minimal rec_len.
1682 * Returns pointer to last entry in range.
1684 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1686 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1687 unsigned rec_len = 0;
1690 while ((char*)de < base + blocksize) {
1691 next = ext4_next_entry(de, blocksize);
1692 if (de->inode && de->name_len) {
1693 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1695 memmove(to, de, rec_len);
1696 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1698 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1706 * Split a full leaf block to make room for a new dir entry.
1707 * Allocate a new block, and move entries so that they are approx. equally full.
1708 * Returns pointer to de in block into which the new entry will be inserted.
1710 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1711 struct buffer_head **bh,struct dx_frame *frame,
1712 struct dx_hash_info *hinfo)
1714 unsigned blocksize = dir->i_sb->s_blocksize;
1715 unsigned count, continued;
1716 struct buffer_head *bh2;
1717 ext4_lblk_t newblock;
1719 struct dx_map_entry *map;
1720 char *data1 = (*bh)->b_data, *data2;
1721 unsigned split, move, size;
1722 struct ext4_dir_entry_2 *de = NULL, *de2;
1723 struct ext4_dir_entry_tail *t;
1727 if (ext4_has_metadata_csum(dir->i_sb))
1728 csum_size = sizeof(struct ext4_dir_entry_tail);
1730 bh2 = ext4_append(handle, dir, &newblock);
1734 return (struct ext4_dir_entry_2 *) bh2;
1737 BUFFER_TRACE(*bh, "get_write_access");
1738 err = ext4_journal_get_write_access(handle, *bh);
1742 BUFFER_TRACE(frame->bh, "get_write_access");
1743 err = ext4_journal_get_write_access(handle, frame->bh);
1747 data2 = bh2->b_data;
1749 /* create map in the end of data2 block */
1750 map = (struct dx_map_entry *) (data2 + blocksize);
1751 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1752 blocksize, hinfo, map);
1754 dx_sort_map(map, count);
1755 /* Split the existing block in the middle, size-wise */
1758 for (i = count-1; i >= 0; i--) {
1759 /* is more than half of this entry in 2nd half of the block? */
1760 if (size + map[i].size/2 > blocksize/2)
1762 size += map[i].size;
1765 /* map index at which we will split */
1766 split = count - move;
1767 hash2 = map[split].hash;
1768 continued = hash2 == map[split - 1].hash;
1769 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1770 (unsigned long)dx_get_block(frame->at),
1771 hash2, split, count-split));
1773 /* Fancy dance to stay within two buffers */
1774 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1776 de = dx_pack_dirents(data1, blocksize);
1777 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1780 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1784 t = EXT4_DIRENT_TAIL(data2, blocksize);
1785 initialize_dirent_tail(t, blocksize);
1787 t = EXT4_DIRENT_TAIL(data1, blocksize);
1788 initialize_dirent_tail(t, blocksize);
1791 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1793 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1796 /* Which block gets the new entry? */
1797 if (hinfo->hash >= hash2) {
1801 dx_insert_block(frame, hash2 + continued, newblock);
1802 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1805 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1809 dxtrace(dx_show_index("frame", frame->entries));
1816 ext4_std_error(dir->i_sb, err);
1817 return ERR_PTR(err);
1820 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1821 struct buffer_head *bh,
1822 void *buf, int buf_size,
1823 struct ext4_filename *fname,
1824 struct ext4_dir_entry_2 **dest_de)
1826 struct ext4_dir_entry_2 *de;
1827 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1829 unsigned int offset = 0;
1833 de = (struct ext4_dir_entry_2 *)buf;
1834 top = buf + buf_size - reclen;
1835 while ((char *) de <= top) {
1836 if (ext4_check_dir_entry(dir, NULL, de, bh,
1837 buf, buf_size, offset)) {
1838 res = -EFSCORRUPTED;
1841 /* Provide crypto context and crypto buffer to ext4 match */
1842 res = ext4_match(fname, de);
1849 nlen = EXT4_DIR_REC_LEN(de->name_len);
1850 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1851 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1853 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1857 if ((char *) de > top)
1867 int ext4_insert_dentry(struct inode *dir,
1868 struct inode *inode,
1869 struct ext4_dir_entry_2 *de,
1871 struct ext4_filename *fname)
1876 nlen = EXT4_DIR_REC_LEN(de->name_len);
1877 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1879 struct ext4_dir_entry_2 *de1 =
1880 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1881 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1882 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1885 de->file_type = EXT4_FT_UNKNOWN;
1886 de->inode = cpu_to_le32(inode->i_ino);
1887 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1888 de->name_len = fname_len(fname);
1889 memcpy(de->name, fname_name(fname), fname_len(fname));
1894 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1895 * it points to a directory entry which is guaranteed to be large
1896 * enough for new directory entry. If de is NULL, then
1897 * add_dirent_to_buf will attempt search the directory block for
1898 * space. It will return -ENOSPC if no space is available, and -EIO
1899 * and -EEXIST if directory entry already exists.
1901 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1903 struct inode *inode, struct ext4_dir_entry_2 *de,
1904 struct buffer_head *bh)
1906 unsigned int blocksize = dir->i_sb->s_blocksize;
1910 if (ext4_has_metadata_csum(inode->i_sb))
1911 csum_size = sizeof(struct ext4_dir_entry_tail);
1914 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1915 blocksize - csum_size, fname, &de);
1919 BUFFER_TRACE(bh, "get_write_access");
1920 err = ext4_journal_get_write_access(handle, bh);
1922 ext4_std_error(dir->i_sb, err);
1926 /* By now the buffer is marked for journaling. Due to crypto operations,
1927 * the following function call may fail */
1928 err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1933 * XXX shouldn't update any times until successful
1934 * completion of syscall, but too many callers depend
1937 * XXX similarly, too many callers depend on
1938 * ext4_new_inode() setting the times, but error
1939 * recovery deletes the inode, so the worst that can
1940 * happen is that the times are slightly out of date
1941 * and/or different from the directory change time.
1943 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1944 ext4_update_dx_flag(dir);
1946 ext4_mark_inode_dirty(handle, dir);
1947 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1948 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1950 ext4_std_error(dir->i_sb, err);
1955 * This converts a one block unindexed directory to a 3 block indexed
1956 * directory, and adds the dentry to the indexed directory.
1958 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1960 struct inode *inode, struct buffer_head *bh)
1962 struct buffer_head *bh2;
1963 struct dx_root *root;
1964 struct dx_frame frames[2], *frame;
1965 struct dx_entry *entries;
1966 struct ext4_dir_entry_2 *de, *de2;
1967 struct ext4_dir_entry_tail *t;
1973 struct fake_dirent *fde;
1976 if (ext4_has_metadata_csum(inode->i_sb))
1977 csum_size = sizeof(struct ext4_dir_entry_tail);
1979 blocksize = dir->i_sb->s_blocksize;
1980 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1981 BUFFER_TRACE(bh, "get_write_access");
1982 retval = ext4_journal_get_write_access(handle, bh);
1984 ext4_std_error(dir->i_sb, retval);
1988 root = (struct dx_root *) bh->b_data;
1990 /* The 0th block becomes the root, move the dirents out */
1991 fde = &root->dotdot;
1992 de = (struct ext4_dir_entry_2 *)((char *)fde +
1993 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1994 if ((char *) de >= (((char *) root) + blocksize)) {
1995 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1997 return -EFSCORRUPTED;
1999 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2001 /* Allocate new block for the 0th block's dirents */
2002 bh2 = ext4_append(handle, dir, &block);
2005 return PTR_ERR(bh2);
2007 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2008 data1 = bh2->b_data;
2010 memcpy (data1, de, len);
2011 de = (struct ext4_dir_entry_2 *) data1;
2013 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2015 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2020 t = EXT4_DIRENT_TAIL(data1, blocksize);
2021 initialize_dirent_tail(t, blocksize);
2024 /* Initialize the root; the dot dirents already exist */
2025 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2026 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2028 memset (&root->info, 0, sizeof(root->info));
2029 root->info.info_length = sizeof(root->info);
2030 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2031 entries = root->entries;
2032 dx_set_block(entries, 1);
2033 dx_set_count(entries, 1);
2034 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2036 /* Initialize as for dx_probe */
2037 fname->hinfo.hash_version = root->info.hash_version;
2038 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2039 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2040 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2041 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2043 memset(frames, 0, sizeof(frames));
2045 frame->entries = entries;
2046 frame->at = entries;
2050 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2053 retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
2057 de = do_split(handle,dir, &bh, frame, &fname->hinfo);
2059 retval = PTR_ERR(de);
2064 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2069 * Even if the block split failed, we have to properly write
2070 * out all the changes we did so far. Otherwise we can end up
2071 * with corrupted filesystem.
2073 ext4_mark_inode_dirty(handle, dir);
2081 * adds a file entry to the specified directory, using the same
2082 * semantics as ext4_find_entry(). It returns NULL if it failed.
2084 * NOTE!! The inode part of 'de' is left at 0 - which means you
2085 * may not sleep between calling this and putting something into
2086 * the entry, as someone else might have used it while you slept.
2088 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2089 struct inode *inode)
2091 struct inode *dir = d_inode(dentry->d_parent);
2092 struct buffer_head *bh = NULL;
2093 struct ext4_dir_entry_2 *de;
2094 struct ext4_dir_entry_tail *t;
2095 struct super_block *sb;
2096 struct ext4_filename fname;
2100 ext4_lblk_t block, blocks;
2103 if (ext4_has_metadata_csum(inode->i_sb))
2104 csum_size = sizeof(struct ext4_dir_entry_tail);
2107 blocksize = sb->s_blocksize;
2108 if (!dentry->d_name.len)
2111 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2115 if (ext4_has_inline_data(dir)) {
2116 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2126 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2127 if (!retval || (retval != ERR_BAD_DX_DIR))
2129 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2131 ext4_mark_inode_dirty(handle, dir);
2133 blocks = dir->i_size >> sb->s_blocksize_bits;
2134 for (block = 0; block < blocks; block++) {
2135 bh = ext4_read_dirblock(dir, block, DIRENT);
2137 retval = PTR_ERR(bh);
2141 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2143 if (retval != -ENOSPC)
2146 if (blocks == 1 && !dx_fallback &&
2147 ext4_has_feature_dir_index(sb)) {
2148 retval = make_indexed_dir(handle, &fname, dir,
2150 bh = NULL; /* make_indexed_dir releases bh */
2155 bh = ext4_append(handle, dir, &block);
2157 retval = PTR_ERR(bh);
2161 de = (struct ext4_dir_entry_2 *) bh->b_data;
2163 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2166 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2167 initialize_dirent_tail(t, blocksize);
2170 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2172 ext4_fname_free_filename(&fname);
2175 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2180 * Returns 0 for success, or a negative error value
2182 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2183 struct inode *dir, struct inode *inode)
2185 struct dx_frame frames[2], *frame;
2186 struct dx_entry *entries, *at;
2187 struct buffer_head *bh;
2188 struct super_block *sb = dir->i_sb;
2189 struct ext4_dir_entry_2 *de;
2192 frame = dx_probe(fname, dir, NULL, frames);
2194 return PTR_ERR(frame);
2195 entries = frame->entries;
2197 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2204 BUFFER_TRACE(bh, "get_write_access");
2205 err = ext4_journal_get_write_access(handle, bh);
2209 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2213 /* Block full, should compress but for now just split */
2214 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2215 dx_get_count(entries), dx_get_limit(entries)));
2216 /* Need to split index? */
2217 if (dx_get_count(entries) == dx_get_limit(entries)) {
2218 ext4_lblk_t newblock;
2219 unsigned icount = dx_get_count(entries);
2220 int levels = frame - frames;
2221 struct dx_entry *entries2;
2222 struct dx_node *node2;
2223 struct buffer_head *bh2;
2225 if (levels && (dx_get_count(frames->entries) ==
2226 dx_get_limit(frames->entries))) {
2227 ext4_warning_inode(dir, "Directory index full!");
2231 bh2 = ext4_append(handle, dir, &newblock);
2236 node2 = (struct dx_node *)(bh2->b_data);
2237 entries2 = node2->entries;
2238 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2239 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2241 BUFFER_TRACE(frame->bh, "get_write_access");
2242 err = ext4_journal_get_write_access(handle, frame->bh);
2246 unsigned icount1 = icount/2, icount2 = icount - icount1;
2247 unsigned hash2 = dx_get_hash(entries + icount1);
2248 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2251 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2252 err = ext4_journal_get_write_access(handle,
2257 memcpy((char *) entries2, (char *) (entries + icount1),
2258 icount2 * sizeof(struct dx_entry));
2259 dx_set_count(entries, icount1);
2260 dx_set_count(entries2, icount2);
2261 dx_set_limit(entries2, dx_node_limit(dir));
2263 /* Which index block gets the new entry? */
2264 if (at - entries >= icount1) {
2265 frame->at = at = at - entries - icount1 + entries2;
2266 frame->entries = entries = entries2;
2267 swap(frame->bh, bh2);
2269 dx_insert_block(frames + 0, hash2, newblock);
2270 dxtrace(dx_show_index("node", frames[1].entries));
2271 dxtrace(dx_show_index("node",
2272 ((struct dx_node *) bh2->b_data)->entries));
2273 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2278 dxtrace(printk(KERN_DEBUG
2279 "Creating second level index...\n"));
2280 memcpy((char *) entries2, (char *) entries,
2281 icount * sizeof(struct dx_entry));
2282 dx_set_limit(entries2, dx_node_limit(dir));
2285 dx_set_count(entries, 1);
2286 dx_set_block(entries + 0, newblock);
2287 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2289 /* Add new access path frame */
2291 frame->at = at = at - entries + entries2;
2292 frame->entries = entries = entries2;
2294 err = ext4_journal_get_write_access(handle,
2299 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2301 ext4_std_error(inode->i_sb, err);
2305 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2310 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2314 ext4_std_error(dir->i_sb, err);
2322 * ext4_generic_delete_entry deletes a directory entry by merging it
2323 * with the previous entry
2325 int ext4_generic_delete_entry(handle_t *handle,
2327 struct ext4_dir_entry_2 *de_del,
2328 struct buffer_head *bh,
2333 struct ext4_dir_entry_2 *de, *pde;
2334 unsigned int blocksize = dir->i_sb->s_blocksize;
2339 de = (struct ext4_dir_entry_2 *)entry_buf;
2340 while (i < buf_size - csum_size) {
2341 if (ext4_check_dir_entry(dir, NULL, de, bh,
2342 bh->b_data, bh->b_size, i))
2343 return -EFSCORRUPTED;
2346 pde->rec_len = ext4_rec_len_to_disk(
2347 ext4_rec_len_from_disk(pde->rec_len,
2349 ext4_rec_len_from_disk(de->rec_len,
2357 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2359 de = ext4_next_entry(de, blocksize);
2364 static int ext4_delete_entry(handle_t *handle,
2366 struct ext4_dir_entry_2 *de_del,
2367 struct buffer_head *bh)
2369 int err, csum_size = 0;
2371 if (ext4_has_inline_data(dir)) {
2372 int has_inline_data = 1;
2373 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2375 if (has_inline_data)
2379 if (ext4_has_metadata_csum(dir->i_sb))
2380 csum_size = sizeof(struct ext4_dir_entry_tail);
2382 BUFFER_TRACE(bh, "get_write_access");
2383 err = ext4_journal_get_write_access(handle, bh);
2387 err = ext4_generic_delete_entry(handle, dir, de_del,
2389 dir->i_sb->s_blocksize, csum_size);
2393 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2394 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2401 ext4_std_error(dir->i_sb, err);
2406 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2407 * since this indicates that nlinks count was previously 1.
2409 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2412 if (is_dx(inode) && inode->i_nlink > 1) {
2413 /* limit is 16-bit i_links_count */
2414 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2415 set_nlink(inode, 1);
2416 ext4_set_feature_dir_nlink(inode->i_sb);
2422 * If a directory had nlink == 1, then we should let it be 1. This indicates
2423 * directory has >EXT4_LINK_MAX subdirs.
2425 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2427 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2432 static int ext4_add_nondir(handle_t *handle,
2433 struct dentry *dentry, struct inode *inode)
2435 int err = ext4_add_entry(handle, dentry, inode);
2437 ext4_mark_inode_dirty(handle, inode);
2438 unlock_new_inode(inode);
2439 d_instantiate(dentry, inode);
2443 unlock_new_inode(inode);
2449 * By the time this is called, we already have created
2450 * the directory cache entry for the new file, but it
2451 * is so far negative - it has no inode.
2453 * If the create succeeds, we fill in the inode information
2454 * with d_instantiate().
2456 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2460 struct inode *inode;
2461 int err, credits, retries = 0;
2463 err = dquot_initialize(dir);
2467 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2468 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2470 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2471 NULL, EXT4_HT_DIR, credits);
2472 handle = ext4_journal_current_handle();
2473 err = PTR_ERR(inode);
2474 if (!IS_ERR(inode)) {
2475 inode->i_op = &ext4_file_inode_operations;
2476 inode->i_fop = &ext4_file_operations;
2477 ext4_set_aops(inode);
2478 err = ext4_add_nondir(handle, dentry, inode);
2479 if (!err && IS_DIRSYNC(dir))
2480 ext4_handle_sync(handle);
2483 ext4_journal_stop(handle);
2484 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2489 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2490 umode_t mode, dev_t rdev)
2493 struct inode *inode;
2494 int err, credits, retries = 0;
2496 err = dquot_initialize(dir);
2500 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2501 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2503 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2504 NULL, EXT4_HT_DIR, credits);
2505 handle = ext4_journal_current_handle();
2506 err = PTR_ERR(inode);
2507 if (!IS_ERR(inode)) {
2508 init_special_inode(inode, inode->i_mode, rdev);
2509 inode->i_op = &ext4_special_inode_operations;
2510 err = ext4_add_nondir(handle, dentry, inode);
2511 if (!err && IS_DIRSYNC(dir))
2512 ext4_handle_sync(handle);
2515 ext4_journal_stop(handle);
2516 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2521 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2524 struct inode *inode;
2525 int err, retries = 0;
2527 err = dquot_initialize(dir);
2532 inode = ext4_new_inode_start_handle(dir, mode,
2535 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2536 4 + EXT4_XATTR_TRANS_BLOCKS);
2537 handle = ext4_journal_current_handle();
2538 err = PTR_ERR(inode);
2539 if (!IS_ERR(inode)) {
2540 inode->i_op = &ext4_file_inode_operations;
2541 inode->i_fop = &ext4_file_operations;
2542 ext4_set_aops(inode);
2543 d_tmpfile(dentry, inode);
2544 err = ext4_orphan_add(handle, inode);
2546 goto err_unlock_inode;
2547 mark_inode_dirty(inode);
2548 unlock_new_inode(inode);
2551 ext4_journal_stop(handle);
2552 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2556 ext4_journal_stop(handle);
2557 unlock_new_inode(inode);
2561 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2562 struct ext4_dir_entry_2 *de,
2563 int blocksize, int csum_size,
2564 unsigned int parent_ino, int dotdot_real_len)
2566 de->inode = cpu_to_le32(inode->i_ino);
2568 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2570 strcpy(de->name, ".");
2571 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2573 de = ext4_next_entry(de, blocksize);
2574 de->inode = cpu_to_le32(parent_ino);
2576 if (!dotdot_real_len)
2577 de->rec_len = ext4_rec_len_to_disk(blocksize -
2578 (csum_size + EXT4_DIR_REC_LEN(1)),
2581 de->rec_len = ext4_rec_len_to_disk(
2582 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2583 strcpy(de->name, "..");
2584 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2586 return ext4_next_entry(de, blocksize);
2589 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2590 struct inode *inode)
2592 struct buffer_head *dir_block = NULL;
2593 struct ext4_dir_entry_2 *de;
2594 struct ext4_dir_entry_tail *t;
2595 ext4_lblk_t block = 0;
2596 unsigned int blocksize = dir->i_sb->s_blocksize;
2600 if (ext4_has_metadata_csum(dir->i_sb))
2601 csum_size = sizeof(struct ext4_dir_entry_tail);
2603 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2604 err = ext4_try_create_inline_dir(handle, dir, inode);
2605 if (err < 0 && err != -ENOSPC)
2612 dir_block = ext4_append(handle, inode, &block);
2613 if (IS_ERR(dir_block))
2614 return PTR_ERR(dir_block);
2615 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2616 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2617 set_nlink(inode, 2);
2619 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2620 initialize_dirent_tail(t, blocksize);
2623 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2624 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2627 set_buffer_verified(dir_block);
2633 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2636 struct inode *inode;
2637 int err, credits, retries = 0;
2639 if (EXT4_DIR_LINK_MAX(dir))
2642 err = dquot_initialize(dir);
2646 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2647 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2649 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2651 0, NULL, EXT4_HT_DIR, credits);
2652 handle = ext4_journal_current_handle();
2653 err = PTR_ERR(inode);
2657 inode->i_op = &ext4_dir_inode_operations;
2658 inode->i_fop = &ext4_dir_operations;
2659 err = ext4_init_new_dir(handle, dir, inode);
2661 goto out_clear_inode;
2662 err = ext4_mark_inode_dirty(handle, inode);
2664 err = ext4_add_entry(handle, dentry, inode);
2668 unlock_new_inode(inode);
2669 ext4_mark_inode_dirty(handle, inode);
2673 ext4_inc_count(handle, dir);
2674 ext4_update_dx_flag(dir);
2675 err = ext4_mark_inode_dirty(handle, dir);
2677 goto out_clear_inode;
2678 unlock_new_inode(inode);
2679 d_instantiate(dentry, inode);
2680 if (IS_DIRSYNC(dir))
2681 ext4_handle_sync(handle);
2685 ext4_journal_stop(handle);
2686 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2692 * routine to check that the specified directory is empty (for rmdir)
2694 int ext4_empty_dir(struct inode *inode)
2696 unsigned int offset;
2697 struct buffer_head *bh;
2698 struct ext4_dir_entry_2 *de, *de1;
2699 struct super_block *sb;
2702 if (ext4_has_inline_data(inode)) {
2703 int has_inline_data = 1;
2705 err = empty_inline_dir(inode, &has_inline_data);
2706 if (has_inline_data)
2711 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2712 EXT4_ERROR_INODE(inode, "invalid size");
2715 bh = ext4_read_dirblock(inode, 0, EITHER);
2719 de = (struct ext4_dir_entry_2 *) bh->b_data;
2720 de1 = ext4_next_entry(de, sb->s_blocksize);
2721 if (le32_to_cpu(de->inode) != inode->i_ino ||
2722 le32_to_cpu(de1->inode) == 0 ||
2723 strcmp(".", de->name) || strcmp("..", de1->name)) {
2724 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2728 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2729 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2730 de = ext4_next_entry(de1, sb->s_blocksize);
2731 while (offset < inode->i_size) {
2732 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2733 unsigned int lblock;
2736 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2737 bh = ext4_read_dirblock(inode, lblock, EITHER);
2740 de = (struct ext4_dir_entry_2 *) bh->b_data;
2742 if (ext4_check_dir_entry(inode, NULL, de, bh,
2743 bh->b_data, bh->b_size, offset)) {
2744 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2746 offset = (offset | (sb->s_blocksize - 1)) + 1;
2749 if (le32_to_cpu(de->inode)) {
2753 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2754 de = ext4_next_entry(de, sb->s_blocksize);
2761 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2762 * such inodes, starting at the superblock, in case we crash before the
2763 * file is closed/deleted, or in case the inode truncate spans multiple
2764 * transactions and the last transaction is not recovered after a crash.
2766 * At filesystem recovery time, we walk this list deleting unlinked
2767 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2769 * Orphan list manipulation functions must be called under i_mutex unless
2770 * we are just creating the inode or deleting it.
2772 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2774 struct super_block *sb = inode->i_sb;
2775 struct ext4_sb_info *sbi = EXT4_SB(sb);
2776 struct ext4_iloc iloc;
2780 if (!sbi->s_journal || is_bad_inode(inode))
2783 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2784 !inode_is_locked(inode));
2786 * Exit early if inode already is on orphan list. This is a big speedup
2787 * since we don't have to contend on the global s_orphan_lock.
2789 if (!list_empty(&EXT4_I(inode)->i_orphan))
2793 * Orphan handling is only valid for files with data blocks
2794 * being truncated, or files being unlinked. Note that we either
2795 * hold i_mutex, or the inode can not be referenced from outside,
2796 * so i_nlink should not be bumped due to race
2798 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2799 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2801 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2802 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2806 err = ext4_reserve_inode_write(handle, inode, &iloc);
2810 mutex_lock(&sbi->s_orphan_lock);
2812 * Due to previous errors inode may be already a part of on-disk
2813 * orphan list. If so skip on-disk list modification.
2815 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2816 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2817 /* Insert this inode at the head of the on-disk orphan list */
2818 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2819 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2822 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2823 mutex_unlock(&sbi->s_orphan_lock);
2826 err = ext4_handle_dirty_super(handle, sb);
2827 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2832 * We have to remove inode from in-memory list if
2833 * addition to on disk orphan list failed. Stray orphan
2834 * list entries can cause panics at unmount time.
2836 mutex_lock(&sbi->s_orphan_lock);
2837 list_del_init(&EXT4_I(inode)->i_orphan);
2838 mutex_unlock(&sbi->s_orphan_lock);
2841 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2842 jbd_debug(4, "orphan inode %lu will point to %d\n",
2843 inode->i_ino, NEXT_ORPHAN(inode));
2845 ext4_std_error(sb, err);
2850 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2851 * of such inodes stored on disk, because it is finally being cleaned up.
2853 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2855 struct list_head *prev;
2856 struct ext4_inode_info *ei = EXT4_I(inode);
2857 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2859 struct ext4_iloc iloc;
2862 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2865 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2866 !inode_is_locked(inode));
2867 /* Do this quick check before taking global s_orphan_lock. */
2868 if (list_empty(&ei->i_orphan))
2872 /* Grab inode buffer early before taking global s_orphan_lock */
2873 err = ext4_reserve_inode_write(handle, inode, &iloc);
2876 mutex_lock(&sbi->s_orphan_lock);
2877 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2879 prev = ei->i_orphan.prev;
2880 list_del_init(&ei->i_orphan);
2882 /* If we're on an error path, we may not have a valid
2883 * transaction handle with which to update the orphan list on
2884 * disk, but we still need to remove the inode from the linked
2885 * list in memory. */
2886 if (!handle || err) {
2887 mutex_unlock(&sbi->s_orphan_lock);
2891 ino_next = NEXT_ORPHAN(inode);
2892 if (prev == &sbi->s_orphan) {
2893 jbd_debug(4, "superblock will point to %u\n", ino_next);
2894 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2895 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2897 mutex_unlock(&sbi->s_orphan_lock);
2900 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2901 mutex_unlock(&sbi->s_orphan_lock);
2902 err = ext4_handle_dirty_super(handle, inode->i_sb);
2904 struct ext4_iloc iloc2;
2905 struct inode *i_prev =
2906 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2908 jbd_debug(4, "orphan inode %lu will point to %u\n",
2909 i_prev->i_ino, ino_next);
2910 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2912 mutex_unlock(&sbi->s_orphan_lock);
2915 NEXT_ORPHAN(i_prev) = ino_next;
2916 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2917 mutex_unlock(&sbi->s_orphan_lock);
2921 NEXT_ORPHAN(inode) = 0;
2922 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2924 ext4_std_error(inode->i_sb, err);
2932 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2935 struct inode *inode;
2936 struct buffer_head *bh;
2937 struct ext4_dir_entry_2 *de;
2938 handle_t *handle = NULL;
2940 /* Initialize quotas before so that eventual writes go in
2941 * separate transaction */
2942 retval = dquot_initialize(dir);
2945 retval = dquot_initialize(d_inode(dentry));
2950 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2956 inode = d_inode(dentry);
2958 retval = -EFSCORRUPTED;
2959 if (le32_to_cpu(de->inode) != inode->i_ino)
2962 retval = -ENOTEMPTY;
2963 if (!ext4_empty_dir(inode))
2966 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2967 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2968 if (IS_ERR(handle)) {
2969 retval = PTR_ERR(handle);
2974 if (IS_DIRSYNC(dir))
2975 ext4_handle_sync(handle);
2977 retval = ext4_delete_entry(handle, dir, de, bh);
2980 if (!EXT4_DIR_LINK_EMPTY(inode))
2981 ext4_warning_inode(inode,
2982 "empty directory '%.*s' has too many links (%u)",
2983 dentry->d_name.len, dentry->d_name.name,
2987 /* There's no need to set i_disksize: the fact that i_nlink is
2988 * zero will ensure that the right thing happens during any
2991 ext4_orphan_add(handle, inode);
2992 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2993 ext4_mark_inode_dirty(handle, inode);
2994 ext4_dec_count(handle, dir);
2995 ext4_update_dx_flag(dir);
2996 ext4_mark_inode_dirty(handle, dir);
3001 ext4_journal_stop(handle);
3005 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3008 struct inode *inode;
3009 struct buffer_head *bh;
3010 struct ext4_dir_entry_2 *de;
3011 handle_t *handle = NULL;
3013 trace_ext4_unlink_enter(dir, dentry);
3014 /* Initialize quotas before so that eventual writes go
3015 * in separate transaction */
3016 retval = dquot_initialize(dir);
3019 retval = dquot_initialize(d_inode(dentry));
3024 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3030 inode = d_inode(dentry);
3032 retval = -EFSCORRUPTED;
3033 if (le32_to_cpu(de->inode) != inode->i_ino)
3036 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3037 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3038 if (IS_ERR(handle)) {
3039 retval = PTR_ERR(handle);
3044 if (IS_DIRSYNC(dir))
3045 ext4_handle_sync(handle);
3047 if (inode->i_nlink == 0) {
3048 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3049 dentry->d_name.len, dentry->d_name.name);
3050 set_nlink(inode, 1);
3052 retval = ext4_delete_entry(handle, dir, de, bh);
3055 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3056 ext4_update_dx_flag(dir);
3057 ext4_mark_inode_dirty(handle, dir);
3059 if (!inode->i_nlink)
3060 ext4_orphan_add(handle, inode);
3061 inode->i_ctime = ext4_current_time(inode);
3062 ext4_mark_inode_dirty(handle, inode);
3067 ext4_journal_stop(handle);
3068 trace_ext4_unlink_exit(dentry, retval);
3072 static int ext4_symlink(struct inode *dir,
3073 struct dentry *dentry, const char *symname)
3076 struct inode *inode;
3077 int err, len = strlen(symname);
3079 bool encryption_required;
3080 struct ext4_str disk_link;
3081 struct ext4_encrypted_symlink_data *sd = NULL;
3083 disk_link.len = len + 1;
3084 disk_link.name = (char *) symname;
3086 encryption_required = (ext4_encrypted_inode(dir) ||
3087 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3088 if (encryption_required) {
3089 err = ext4_get_encryption_info(dir);
3092 if (ext4_encryption_info(dir) == NULL)
3094 disk_link.len = (ext4_fname_encrypted_size(dir, len) +
3095 sizeof(struct ext4_encrypted_symlink_data));
3096 sd = kzalloc(disk_link.len, GFP_KERNEL);
3101 if (disk_link.len > dir->i_sb->s_blocksize) {
3102 err = -ENAMETOOLONG;
3106 err = dquot_initialize(dir);
3110 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3112 * For non-fast symlinks, we just allocate inode and put it on
3113 * orphan list in the first transaction => we need bitmap,
3114 * group descriptor, sb, inode block, quota blocks, and
3115 * possibly selinux xattr blocks.
3117 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3118 EXT4_XATTR_TRANS_BLOCKS;
3121 * Fast symlink. We have to add entry to directory
3122 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3123 * allocate new inode (bitmap, group descriptor, inode block,
3124 * quota blocks, sb is already counted in previous macros).
3126 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3127 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3130 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3131 &dentry->d_name, 0, NULL,
3132 EXT4_HT_DIR, credits);
3133 handle = ext4_journal_current_handle();
3134 if (IS_ERR(inode)) {
3136 ext4_journal_stop(handle);
3137 err = PTR_ERR(inode);
3141 if (encryption_required) {
3143 struct ext4_str ostr;
3145 istr.name = (const unsigned char *) symname;
3147 ostr.name = sd->encrypted_path;
3148 ostr.len = disk_link.len;
3149 err = ext4_fname_usr_to_disk(inode, &istr, &ostr);
3151 goto err_drop_inode;
3152 sd->len = cpu_to_le16(ostr.len);
3153 disk_link.name = (char *) sd;
3154 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3157 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3158 if (!encryption_required)
3159 inode->i_op = &ext4_symlink_inode_operations;
3160 inode_nohighmem(inode);
3161 ext4_set_aops(inode);
3163 * We cannot call page_symlink() with transaction started
3164 * because it calls into ext4_write_begin() which can wait
3165 * for transaction commit if we are running out of space
3166 * and thus we deadlock. So we have to stop transaction now
3167 * and restart it when symlink contents is written.
3169 * To keep fs consistent in case of crash, we have to put inode
3170 * to orphan list in the mean time.
3173 err = ext4_orphan_add(handle, inode);
3174 ext4_journal_stop(handle);
3177 goto err_drop_inode;
3178 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3180 goto err_drop_inode;
3182 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3183 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3185 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3186 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3187 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3188 if (IS_ERR(handle)) {
3189 err = PTR_ERR(handle);
3191 goto err_drop_inode;
3193 set_nlink(inode, 1);
3194 err = ext4_orphan_del(handle, inode);
3196 goto err_drop_inode;
3198 /* clear the extent format for fast symlink */
3199 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3200 if (!encryption_required) {
3201 inode->i_op = &ext4_fast_symlink_inode_operations;
3202 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3204 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3206 inode->i_size = disk_link.len - 1;
3208 EXT4_I(inode)->i_disksize = inode->i_size;
3209 err = ext4_add_nondir(handle, dentry, inode);
3210 if (!err && IS_DIRSYNC(dir))
3211 ext4_handle_sync(handle);
3214 ext4_journal_stop(handle);
3219 ext4_journal_stop(handle);
3221 unlock_new_inode(inode);
3228 static int ext4_link(struct dentry *old_dentry,
3229 struct inode *dir, struct dentry *dentry)
3232 struct inode *inode = d_inode(old_dentry);
3233 int err, retries = 0;
3235 if (inode->i_nlink >= EXT4_LINK_MAX)
3237 if (ext4_encrypted_inode(dir) &&
3238 !ext4_is_child_context_consistent_with_parent(dir, inode))
3241 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3242 (!projid_eq(EXT4_I(dir)->i_projid,
3243 EXT4_I(old_dentry->d_inode)->i_projid)))
3246 err = dquot_initialize(dir);
3251 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3252 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3253 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3255 return PTR_ERR(handle);
3257 if (IS_DIRSYNC(dir))
3258 ext4_handle_sync(handle);
3260 inode->i_ctime = ext4_current_time(inode);
3261 ext4_inc_count(handle, inode);
3264 err = ext4_add_entry(handle, dentry, inode);
3266 ext4_mark_inode_dirty(handle, inode);
3267 /* this can happen only for tmpfile being
3268 * linked the first time
3270 if (inode->i_nlink == 1)
3271 ext4_orphan_del(handle, inode);
3272 d_instantiate(dentry, inode);
3277 ext4_journal_stop(handle);
3278 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3285 * Try to find buffer head where contains the parent block.
3286 * It should be the inode block if it is inlined or the 1st block
3287 * if it is a normal dir.
3289 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3290 struct inode *inode,
3292 struct ext4_dir_entry_2 **parent_de,
3295 struct buffer_head *bh;
3297 if (!ext4_has_inline_data(inode)) {
3298 bh = ext4_read_dirblock(inode, 0, EITHER);
3300 *retval = PTR_ERR(bh);
3303 *parent_de = ext4_next_entry(
3304 (struct ext4_dir_entry_2 *)bh->b_data,
3305 inode->i_sb->s_blocksize);
3310 return ext4_get_first_inline_block(inode, parent_de, retval);
3313 struct ext4_renament {
3315 struct dentry *dentry;
3316 struct inode *inode;
3318 int dir_nlink_delta;
3320 /* entry for "dentry" */
3321 struct buffer_head *bh;
3322 struct ext4_dir_entry_2 *de;
3325 /* entry for ".." in inode if it's a directory */
3326 struct buffer_head *dir_bh;
3327 struct ext4_dir_entry_2 *parent_de;
3331 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3335 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3336 &retval, &ent->parent_de,
3340 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3341 return -EFSCORRUPTED;
3342 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3343 return ext4_journal_get_write_access(handle, ent->dir_bh);
3346 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3351 ent->parent_de->inode = cpu_to_le32(dir_ino);
3352 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3353 if (!ent->dir_inlined) {
3354 if (is_dx(ent->inode)) {
3355 retval = ext4_handle_dirty_dx_node(handle,
3359 retval = ext4_handle_dirty_dirent_node(handle,
3364 retval = ext4_mark_inode_dirty(handle, ent->inode);
3367 ext4_std_error(ent->dir->i_sb, retval);
3373 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3374 unsigned ino, unsigned file_type)
3378 BUFFER_TRACE(ent->bh, "get write access");
3379 retval = ext4_journal_get_write_access(handle, ent->bh);
3382 ent->de->inode = cpu_to_le32(ino);
3383 if (ext4_has_feature_filetype(ent->dir->i_sb))
3384 ent->de->file_type = file_type;
3385 ent->dir->i_version++;
3386 ent->dir->i_ctime = ent->dir->i_mtime =
3387 ext4_current_time(ent->dir);
3388 ext4_mark_inode_dirty(handle, ent->dir);
3389 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3390 if (!ent->inlined) {
3391 retval = ext4_handle_dirty_dirent_node(handle,
3393 if (unlikely(retval)) {
3394 ext4_std_error(ent->dir->i_sb, retval);
3404 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3405 const struct qstr *d_name)
3407 int retval = -ENOENT;
3408 struct buffer_head *bh;
3409 struct ext4_dir_entry_2 *de;
3411 bh = ext4_find_entry(dir, d_name, &de, NULL);
3415 retval = ext4_delete_entry(handle, dir, de, bh);
3421 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3426 * ent->de could have moved from under us during htree split, so make
3427 * sure that we are deleting the right entry. We might also be pointing
3428 * to a stale entry in the unused part of ent->bh so just checking inum
3429 * and the name isn't enough.
3431 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3432 ent->de->name_len != ent->dentry->d_name.len ||
3433 strncmp(ent->de->name, ent->dentry->d_name.name,
3434 ent->de->name_len) ||
3436 retval = ext4_find_delete_entry(handle, ent->dir,
3437 &ent->dentry->d_name);
3439 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3440 if (retval == -ENOENT) {
3441 retval = ext4_find_delete_entry(handle, ent->dir,
3442 &ent->dentry->d_name);
3447 ext4_warning_inode(ent->dir,
3448 "Deleting old file: nlink %d, error=%d",
3449 ent->dir->i_nlink, retval);
3453 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3455 if (ent->dir_nlink_delta) {
3456 if (ent->dir_nlink_delta == -1)
3457 ext4_dec_count(handle, ent->dir);
3459 ext4_inc_count(handle, ent->dir);
3460 ext4_mark_inode_dirty(handle, ent->dir);
3464 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3465 int credits, handle_t **h)
3472 * for inode block, sb block, group summaries,
3475 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3476 EXT4_XATTR_TRANS_BLOCKS + 4);
3478 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3479 &ent->dentry->d_name, 0, NULL,
3480 EXT4_HT_DIR, credits);
3482 handle = ext4_journal_current_handle();
3485 ext4_journal_stop(handle);
3486 if (PTR_ERR(wh) == -ENOSPC &&
3487 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3491 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3492 wh->i_op = &ext4_special_inode_operations;
3498 * Anybody can rename anything with this: the permission checks are left to the
3499 * higher-level routines.
3501 * n.b. old_{dentry,inode) refers to the source dentry/inode
3502 * while new_{dentry,inode) refers to the destination dentry/inode
3503 * This comes from rename(const char *oldpath, const char *newpath)
3505 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3506 struct inode *new_dir, struct dentry *new_dentry,
3509 handle_t *handle = NULL;
3510 struct ext4_renament old = {
3512 .dentry = old_dentry,
3513 .inode = d_inode(old_dentry),
3515 struct ext4_renament new = {
3517 .dentry = new_dentry,
3518 .inode = d_inode(new_dentry),
3522 struct inode *whiteout = NULL;
3526 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3527 (!projid_eq(EXT4_I(new_dir)->i_projid,
3528 EXT4_I(old_dentry->d_inode)->i_projid)))
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);
3593 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3594 if (IS_ERR(whiteout)) {
3595 retval = PTR_ERR(whiteout);
3601 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3602 ext4_handle_sync(handle);
3604 if (S_ISDIR(old.inode->i_mode)) {
3606 retval = -ENOTEMPTY;
3607 if (!ext4_empty_dir(new.inode))
3611 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3614 retval = ext4_rename_dir_prepare(handle, &old);
3619 * If we're renaming a file within an inline_data dir and adding or
3620 * setting the new dirent causes a conversion from inline_data to
3621 * extents/blockmap, we need to force the dirent delete code to
3622 * re-read the directory, or else we end up trying to delete a dirent
3623 * from what is now the extent tree root (or a block map).
3625 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3626 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3628 old_file_type = old.de->file_type;
3631 * Do this before adding a new entry, so the old entry is sure
3632 * to be still pointing to the valid old entry.
3634 retval = ext4_setent(handle, &old, whiteout->i_ino,
3638 ext4_mark_inode_dirty(handle, whiteout);
3641 retval = ext4_add_entry(handle, new.dentry, old.inode);
3645 retval = ext4_setent(handle, &new,
3646 old.inode->i_ino, old_file_type);
3651 force_reread = !ext4_test_inode_flag(new.dir,
3652 EXT4_INODE_INLINE_DATA);
3655 * Like most other Unix systems, set the ctime for inodes on a
3658 old.inode->i_ctime = ext4_current_time(old.inode);
3659 ext4_mark_inode_dirty(handle, old.inode);
3665 ext4_rename_delete(handle, &old, force_reread);
3669 ext4_dec_count(handle, new.inode);
3670 new.inode->i_ctime = ext4_current_time(new.inode);
3672 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3673 ext4_update_dx_flag(old.dir);
3675 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3679 ext4_dec_count(handle, old.dir);
3681 /* checked ext4_empty_dir above, can't have another
3682 * parent, ext4_dec_count() won't work for many-linked
3684 clear_nlink(new.inode);
3686 ext4_inc_count(handle, new.dir);
3687 ext4_update_dx_flag(new.dir);
3688 ext4_mark_inode_dirty(handle, new.dir);
3691 ext4_mark_inode_dirty(handle, old.dir);
3693 ext4_mark_inode_dirty(handle, new.inode);
3694 if (!new.inode->i_nlink)
3695 ext4_orphan_add(handle, new.inode);
3705 drop_nlink(whiteout);
3706 unlock_new_inode(whiteout);
3710 ext4_journal_stop(handle);
3714 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3715 struct inode *new_dir, struct dentry *new_dentry)
3717 handle_t *handle = NULL;
3718 struct ext4_renament old = {
3720 .dentry = old_dentry,
3721 .inode = d_inode(old_dentry),
3723 struct ext4_renament new = {
3725 .dentry = new_dentry,
3726 .inode = d_inode(new_dentry),
3731 if ((ext4_encrypted_inode(old_dir) ||
3732 ext4_encrypted_inode(new_dir)) &&
3733 (old_dir != new_dir) &&
3734 (!ext4_is_child_context_consistent_with_parent(new_dir,
3736 !ext4_is_child_context_consistent_with_parent(old_dir,
3740 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3741 !projid_eq(EXT4_I(new_dir)->i_projid,
3742 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3743 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3744 !projid_eq(EXT4_I(old_dir)->i_projid,
3745 EXT4_I(new_dentry->d_inode)->i_projid)))
3748 retval = dquot_initialize(old.dir);
3751 retval = dquot_initialize(new.dir);
3755 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3756 &old.de, &old.inlined);
3758 return PTR_ERR(old.bh);
3760 * Check for inode number is _not_ due to possible IO errors.
3761 * We might rmdir the source, keep it as pwd of some process
3762 * and merrily kill the link to whatever was created under the
3763 * same name. Goodbye sticky bit ;-<
3766 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3769 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3770 &new.de, &new.inlined);
3771 if (IS_ERR(new.bh)) {
3772 retval = PTR_ERR(new.bh);
3777 /* RENAME_EXCHANGE case: old *and* new must both exist */
3778 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3781 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3782 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3783 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3784 if (IS_ERR(handle)) {
3785 retval = PTR_ERR(handle);
3790 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3791 ext4_handle_sync(handle);
3793 if (S_ISDIR(old.inode->i_mode)) {
3795 retval = ext4_rename_dir_prepare(handle, &old);
3799 if (S_ISDIR(new.inode->i_mode)) {
3801 retval = ext4_rename_dir_prepare(handle, &new);
3807 * Other than the special case of overwriting a directory, parents'
3808 * nlink only needs to be modified if this is a cross directory rename.
3810 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3811 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3812 new.dir_nlink_delta = -old.dir_nlink_delta;
3814 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3815 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3819 new_file_type = new.de->file_type;
3820 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3824 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3829 * Like most other Unix systems, set the ctime for inodes on a
3832 old.inode->i_ctime = ext4_current_time(old.inode);
3833 new.inode->i_ctime = ext4_current_time(new.inode);
3834 ext4_mark_inode_dirty(handle, old.inode);
3835 ext4_mark_inode_dirty(handle, new.inode);
3838 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3843 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3847 ext4_update_dir_count(handle, &old);
3848 ext4_update_dir_count(handle, &new);
3857 ext4_journal_stop(handle);
3861 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3862 struct inode *new_dir, struct dentry *new_dentry,
3865 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3868 if (flags & RENAME_EXCHANGE) {
3869 return ext4_cross_rename(old_dir, old_dentry,
3870 new_dir, new_dentry);
3873 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3877 * directories can handle most operations...
3879 const struct inode_operations ext4_dir_inode_operations = {
3880 .create = ext4_create,
3881 .lookup = ext4_lookup,
3883 .unlink = ext4_unlink,
3884 .symlink = ext4_symlink,
3885 .mkdir = ext4_mkdir,
3886 .rmdir = ext4_rmdir,
3887 .mknod = ext4_mknod,
3888 .tmpfile = ext4_tmpfile,
3889 .rename2 = ext4_rename2,
3890 .setattr = ext4_setattr,
3891 .setxattr = generic_setxattr,
3892 .getxattr = generic_getxattr,
3893 .listxattr = ext4_listxattr,
3894 .removexattr = generic_removexattr,
3895 .get_acl = ext4_get_acl,
3896 .set_acl = ext4_set_acl,
3897 .fiemap = ext4_fiemap,
3900 const struct inode_operations ext4_special_inode_operations = {
3901 .setattr = ext4_setattr,
3902 .setxattr = generic_setxattr,
3903 .getxattr = generic_getxattr,
3904 .listxattr = ext4_listxattr,
3905 .removexattr = generic_removexattr,
3906 .get_acl = ext4_get_acl,
3907 .set_acl = ext4_set_acl,