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/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
38 #include "ext4_jbd2.h"
43 #include <trace/events/ext4.h>
45 * define how far ahead to read directories while searching them.
47 #define NAMEI_RA_CHUNKS 2
48 #define NAMEI_RA_BLOCKS 4
49 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
52 static struct buffer_head *ext4_append(handle_t *handle,
54 ext4_lblk_t *block, int *err)
56 struct buffer_head *bh;
58 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
59 ((inode->i_size >> 10) >=
60 EXT4_SB(inode->i_sb)->s_max_dir_size_kb))) {
65 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
67 bh = ext4_bread(handle, inode, *block, 1, err);
69 inode->i_size += inode->i_sb->s_blocksize;
70 EXT4_I(inode)->i_disksize = inode->i_size;
71 *err = ext4_journal_get_write_access(handle, bh);
79 ext4_error(inode->i_sb,
80 "Directory hole detected on inode %lu\n",
87 #define assert(test) J_ASSERT(test)
91 #define dxtrace(command) command
93 #define dxtrace(command)
117 * dx_root_info is laid out so that if it should somehow get overlaid by a
118 * dirent the two low bits of the hash version will be zero. Therefore, the
119 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
124 struct fake_dirent dot;
126 struct fake_dirent dotdot;
130 __le32 reserved_zero;
132 u8 info_length; /* 8 */
137 struct dx_entry entries[0];
142 struct fake_dirent fake;
143 struct dx_entry entries[0];
149 struct buffer_head *bh;
150 struct dx_entry *entries;
162 * This goes at the end of each htree block.
166 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
169 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
170 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
171 static inline unsigned dx_get_hash(struct dx_entry *entry);
172 static void dx_set_hash(struct dx_entry *entry, unsigned value);
173 static unsigned dx_get_count(struct dx_entry *entries);
174 static unsigned dx_get_limit(struct dx_entry *entries);
175 static void dx_set_count(struct dx_entry *entries, unsigned value);
176 static void dx_set_limit(struct dx_entry *entries, unsigned value);
177 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
178 static unsigned dx_node_limit(struct inode *dir);
179 static struct dx_frame *dx_probe(const struct qstr *d_name,
181 struct dx_hash_info *hinfo,
182 struct dx_frame *frame,
184 static void dx_release(struct dx_frame *frames);
185 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
186 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
187 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
188 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
189 struct dx_map_entry *offsets, int count, unsigned blocksize);
190 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
191 static void dx_insert_block(struct dx_frame *frame,
192 u32 hash, ext4_lblk_t block);
193 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
194 struct dx_frame *frame,
195 struct dx_frame *frames,
197 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
198 const struct qstr *d_name,
199 struct ext4_dir_entry_2 **res_dir,
201 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
202 struct inode *inode);
204 /* checksumming functions */
205 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
206 unsigned int blocksize)
208 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
209 t->det_rec_len = ext4_rec_len_to_disk(
210 sizeof(struct ext4_dir_entry_tail), blocksize);
211 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
214 /* Walk through a dirent block to find a checksum "dirent" at the tail */
215 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
216 struct ext4_dir_entry *de)
218 struct ext4_dir_entry_tail *t;
221 struct ext4_dir_entry *d, *top;
224 top = (struct ext4_dir_entry *)(((void *)de) +
225 (EXT4_BLOCK_SIZE(inode->i_sb) -
226 sizeof(struct ext4_dir_entry_tail)));
227 while (d < top && d->rec_len)
228 d = (struct ext4_dir_entry *)(((void *)d) +
229 le16_to_cpu(d->rec_len));
234 t = (struct ext4_dir_entry_tail *)d;
236 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
239 if (t->det_reserved_zero1 ||
240 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
241 t->det_reserved_zero2 ||
242 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
248 static __le32 ext4_dirent_csum(struct inode *inode,
249 struct ext4_dir_entry *dirent, int size)
251 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
252 struct ext4_inode_info *ei = EXT4_I(inode);
255 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
256 return cpu_to_le32(csum);
259 static void warn_no_space_for_csum(struct inode *inode)
261 ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
262 "checksum. Please run e2fsck -D.", inode->i_ino);
265 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
267 struct ext4_dir_entry_tail *t;
269 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
270 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
273 t = get_dirent_tail(inode, dirent);
275 warn_no_space_for_csum(inode);
279 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
280 (void *)t - (void *)dirent))
286 static void ext4_dirent_csum_set(struct inode *inode,
287 struct ext4_dir_entry *dirent)
289 struct ext4_dir_entry_tail *t;
291 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
292 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
295 t = get_dirent_tail(inode, dirent);
297 warn_no_space_for_csum(inode);
301 t->det_checksum = ext4_dirent_csum(inode, dirent,
302 (void *)t - (void *)dirent);
305 int ext4_handle_dirty_dirent_node(handle_t *handle,
307 struct buffer_head *bh)
309 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
310 return ext4_handle_dirty_metadata(handle, inode, bh);
313 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
314 struct ext4_dir_entry *dirent,
317 struct ext4_dir_entry *dp;
318 struct dx_root_info *root;
321 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
323 else if (le16_to_cpu(dirent->rec_len) == 12) {
324 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
325 if (le16_to_cpu(dp->rec_len) !=
326 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
328 root = (struct dx_root_info *)(((void *)dp + 12));
329 if (root->reserved_zero ||
330 root->info_length != sizeof(struct dx_root_info))
337 *offset = count_offset;
338 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
341 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
342 int count_offset, int count, struct dx_tail *t)
344 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
345 struct ext4_inode_info *ei = EXT4_I(inode);
346 __u32 csum, old_csum;
349 size = count_offset + (count * sizeof(struct dx_entry));
350 old_csum = t->dt_checksum;
352 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
353 csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
354 t->dt_checksum = old_csum;
356 return cpu_to_le32(csum);
359 static int ext4_dx_csum_verify(struct inode *inode,
360 struct ext4_dir_entry *dirent)
362 struct dx_countlimit *c;
364 int count_offset, limit, count;
366 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
367 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
370 c = get_dx_countlimit(inode, dirent, &count_offset);
372 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
375 limit = le16_to_cpu(c->limit);
376 count = le16_to_cpu(c->count);
377 if (count_offset + (limit * sizeof(struct dx_entry)) >
378 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
379 warn_no_space_for_csum(inode);
382 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
384 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
390 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
392 struct dx_countlimit *c;
394 int count_offset, limit, count;
396 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
397 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
400 c = get_dx_countlimit(inode, dirent, &count_offset);
402 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
405 limit = le16_to_cpu(c->limit);
406 count = le16_to_cpu(c->count);
407 if (count_offset + (limit * sizeof(struct dx_entry)) >
408 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
409 warn_no_space_for_csum(inode);
412 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
414 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
417 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
419 struct buffer_head *bh)
421 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
422 return ext4_handle_dirty_metadata(handle, inode, bh);
426 * p is at least 6 bytes before the end of page
428 static inline struct ext4_dir_entry_2 *
429 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
431 return (struct ext4_dir_entry_2 *)((char *)p +
432 ext4_rec_len_from_disk(p->rec_len, blocksize));
436 * Future: use high four bits of block for coalesce-on-delete flags
437 * Mask them off for now.
440 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
442 return le32_to_cpu(entry->block) & 0x00ffffff;
445 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
447 entry->block = cpu_to_le32(value);
450 static inline unsigned dx_get_hash(struct dx_entry *entry)
452 return le32_to_cpu(entry->hash);
455 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
457 entry->hash = cpu_to_le32(value);
460 static inline unsigned dx_get_count(struct dx_entry *entries)
462 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
465 static inline unsigned dx_get_limit(struct dx_entry *entries)
467 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
470 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
472 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
475 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
477 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
480 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
482 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
483 EXT4_DIR_REC_LEN(2) - infosize;
485 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
486 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
487 entry_space -= sizeof(struct dx_tail);
488 return entry_space / sizeof(struct dx_entry);
491 static inline unsigned dx_node_limit(struct inode *dir)
493 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
495 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
496 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
497 entry_space -= sizeof(struct dx_tail);
498 return entry_space / sizeof(struct dx_entry);
505 static void dx_show_index(char * label, struct dx_entry *entries)
507 int i, n = dx_get_count (entries);
508 printk(KERN_DEBUG "%s index ", label);
509 for (i = 0; i < n; i++) {
510 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
511 0, (unsigned long)dx_get_block(entries + i));
523 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
524 int size, int show_names)
526 unsigned names = 0, space = 0;
527 char *base = (char *) de;
528 struct dx_hash_info h = *hinfo;
531 while ((char *) de < base + size)
537 int len = de->name_len;
538 char *name = de->name;
539 while (len--) printk("%c", *name++);
540 ext4fs_dirhash(de->name, de->name_len, &h);
541 printk(":%x.%u ", h.hash,
542 (unsigned) ((char *) de - base));
544 space += EXT4_DIR_REC_LEN(de->name_len);
547 de = ext4_next_entry(de, size);
549 printk("(%i)\n", names);
550 return (struct stats) { names, space, 1 };
553 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
554 struct dx_entry *entries, int levels)
556 unsigned blocksize = dir->i_sb->s_blocksize;
557 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
559 struct buffer_head *bh;
561 printk("%i indexed blocks...\n", count);
562 for (i = 0; i < count; i++, entries++)
564 ext4_lblk_t block = dx_get_block(entries);
565 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
566 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
568 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
569 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
571 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
572 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
573 names += stats.names;
574 space += stats.space;
575 bcount += stats.bcount;
579 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
580 levels ? "" : " ", names, space/bcount,
581 (space/bcount)*100/blocksize);
582 return (struct stats) { names, space, bcount};
584 #endif /* DX_DEBUG */
587 * Probe for a directory leaf block to search.
589 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
590 * error in the directory index, and the caller should fall back to
591 * searching the directory normally. The callers of dx_probe **MUST**
592 * check for this error code, and make sure it never gets reflected
595 static struct dx_frame *
596 dx_probe(const struct qstr *d_name, struct inode *dir,
597 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
599 unsigned count, indirect;
600 struct dx_entry *at, *entries, *p, *q, *m;
601 struct dx_root *root;
602 struct buffer_head *bh;
603 struct dx_frame *frame = frame_in;
607 if (!(bh = ext4_bread(NULL, dir, 0, 0, err))) {
609 *err = ERR_BAD_DX_DIR;
612 root = (struct dx_root *) bh->b_data;
613 if (root->info.hash_version != DX_HASH_TEA &&
614 root->info.hash_version != DX_HASH_HALF_MD4 &&
615 root->info.hash_version != DX_HASH_LEGACY) {
616 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
617 root->info.hash_version);
619 *err = ERR_BAD_DX_DIR;
622 hinfo->hash_version = root->info.hash_version;
623 if (hinfo->hash_version <= DX_HASH_TEA)
624 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
625 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
627 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
630 if (root->info.unused_flags & 1) {
631 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
632 root->info.unused_flags);
634 *err = ERR_BAD_DX_DIR;
638 if ((indirect = root->info.indirect_levels) > 1) {
639 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
640 root->info.indirect_levels);
642 *err = ERR_BAD_DX_DIR;
646 if (!buffer_verified(bh) &&
647 !ext4_dx_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data)) {
648 ext4_warning(dir->i_sb, "Root failed checksum");
650 *err = ERR_BAD_DX_DIR;
653 set_buffer_verified(bh);
655 entries = (struct dx_entry *) (((char *)&root->info) +
656 root->info.info_length);
658 if (dx_get_limit(entries) != dx_root_limit(dir,
659 root->info.info_length)) {
660 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
662 *err = ERR_BAD_DX_DIR;
666 dxtrace(printk("Look up %x", hash));
669 count = dx_get_count(entries);
670 if (!count || count > dx_get_limit(entries)) {
671 ext4_warning(dir->i_sb,
672 "dx entry: no count or count > limit");
674 *err = ERR_BAD_DX_DIR;
679 q = entries + count - 1;
683 dxtrace(printk("."));
684 if (dx_get_hash(m) > hash)
690 if (0) // linear search cross check
692 unsigned n = count - 1;
696 dxtrace(printk(","));
697 if (dx_get_hash(++at) > hash)
703 assert (at == p - 1);
707 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
709 frame->entries = entries;
711 if (!indirect--) return frame;
712 if (!(bh = ext4_bread(NULL, dir, dx_get_block(at), 0, err))) {
714 *err = ERR_BAD_DX_DIR;
717 at = entries = ((struct dx_node *) bh->b_data)->entries;
719 if (!buffer_verified(bh) &&
720 !ext4_dx_csum_verify(dir,
721 (struct ext4_dir_entry *)bh->b_data)) {
722 ext4_warning(dir->i_sb, "Node failed checksum");
724 *err = ERR_BAD_DX_DIR;
727 set_buffer_verified(bh);
729 if (dx_get_limit(entries) != dx_node_limit (dir)) {
730 ext4_warning(dir->i_sb,
731 "dx entry: limit != node limit");
733 *err = ERR_BAD_DX_DIR;
740 while (frame >= frame_in) {
745 if (*err == ERR_BAD_DX_DIR)
746 ext4_warning(dir->i_sb,
747 "Corrupt dir inode %lu, running e2fsck is "
748 "recommended.", dir->i_ino);
752 static void dx_release (struct dx_frame *frames)
754 if (frames[0].bh == NULL)
757 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
758 brelse(frames[1].bh);
759 brelse(frames[0].bh);
763 * This function increments the frame pointer to search the next leaf
764 * block, and reads in the necessary intervening nodes if the search
765 * should be necessary. Whether or not the search is necessary is
766 * controlled by the hash parameter. If the hash value is even, then
767 * the search is only continued if the next block starts with that
768 * hash value. This is used if we are searching for a specific file.
770 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
772 * This function returns 1 if the caller should continue to search,
773 * or 0 if it should not. If there is an error reading one of the
774 * index blocks, it will a negative error code.
776 * If start_hash is non-null, it will be filled in with the starting
777 * hash of the next page.
779 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
780 struct dx_frame *frame,
781 struct dx_frame *frames,
785 struct buffer_head *bh;
786 int err, num_frames = 0;
791 * Find the next leaf page by incrementing the frame pointer.
792 * If we run out of entries in the interior node, loop around and
793 * increment pointer in the parent node. When we break out of
794 * this loop, num_frames indicates the number of interior
795 * nodes need to be read.
798 if (++(p->at) < p->entries + dx_get_count(p->entries))
807 * If the hash is 1, then continue only if the next page has a
808 * continuation hash of any value. This is used for readdir
809 * handling. Otherwise, check to see if the hash matches the
810 * desired contiuation hash. If it doesn't, return since
811 * there's no point to read in the successive index pages.
813 bhash = dx_get_hash(p->at);
816 if ((hash & 1) == 0) {
817 if ((bhash & ~1) != hash)
821 * If the hash is HASH_NB_ALWAYS, we always go to the next
822 * block so no check is necessary
824 while (num_frames--) {
825 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
828 ext4_error(dir->i_sb,
829 "Directory hole detected on inode %lu\n",
833 return err; /* Failure */
836 if (!buffer_verified(bh) &&
837 !ext4_dx_csum_verify(dir,
838 (struct ext4_dir_entry *)bh->b_data)) {
839 ext4_warning(dir->i_sb, "Node failed checksum");
842 set_buffer_verified(bh);
847 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
854 * This function fills a red-black tree with information from a
855 * directory block. It returns the number directory entries loaded
856 * into the tree. If there is an error it is returned in err.
858 static int htree_dirblock_to_tree(struct file *dir_file,
859 struct inode *dir, ext4_lblk_t block,
860 struct dx_hash_info *hinfo,
861 __u32 start_hash, __u32 start_minor_hash)
863 struct buffer_head *bh;
864 struct ext4_dir_entry_2 *de, *top;
865 int err = 0, count = 0;
867 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
868 (unsigned long)block));
869 if (!(bh = ext4_bread(NULL, dir, block, 0, &err))) {
872 ext4_error(dir->i_sb,
873 "Directory hole detected on inode %lu\n",
879 if (!buffer_verified(bh) &&
880 !ext4_dirent_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data))
882 set_buffer_verified(bh);
884 de = (struct ext4_dir_entry_2 *) bh->b_data;
885 top = (struct ext4_dir_entry_2 *) ((char *) de +
886 dir->i_sb->s_blocksize -
887 EXT4_DIR_REC_LEN(0));
888 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
889 if (ext4_check_dir_entry(dir, NULL, de, bh,
890 bh->b_data, bh->b_size,
891 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
892 + ((char *)de - bh->b_data))) {
893 /* On error, skip the f_pos to the next block. */
894 dir_file->f_pos = (dir_file->f_pos |
895 (dir->i_sb->s_blocksize - 1)) + 1;
899 ext4fs_dirhash(de->name, de->name_len, hinfo);
900 if ((hinfo->hash < start_hash) ||
901 ((hinfo->hash == start_hash) &&
902 (hinfo->minor_hash < start_minor_hash)))
906 if ((err = ext4_htree_store_dirent(dir_file,
907 hinfo->hash, hinfo->minor_hash, de)) != 0) {
919 * This function fills a red-black tree with information from a
920 * directory. We start scanning the directory in hash order, starting
921 * at start_hash and start_minor_hash.
923 * This function returns the number of entries inserted into the tree,
924 * or a negative error code.
926 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
927 __u32 start_minor_hash, __u32 *next_hash)
929 struct dx_hash_info hinfo;
930 struct ext4_dir_entry_2 *de;
931 struct dx_frame frames[2], *frame;
938 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
939 start_hash, start_minor_hash));
940 dir = dir_file->f_path.dentry->d_inode;
941 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
942 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
943 if (hinfo.hash_version <= DX_HASH_TEA)
944 hinfo.hash_version +=
945 EXT4_SB(dir->i_sb)->s_hash_unsigned;
946 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
947 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
948 start_hash, start_minor_hash);
952 hinfo.hash = start_hash;
953 hinfo.minor_hash = 0;
954 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
958 /* Add '.' and '..' from the htree header */
959 if (!start_hash && !start_minor_hash) {
960 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
961 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
965 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
966 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
967 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
968 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
974 block = dx_get_block(frame->at);
975 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
976 start_hash, start_minor_hash);
983 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
984 frame, frames, &hashval);
985 *next_hash = hashval;
991 * Stop if: (a) there are no more entries, or
992 * (b) we have inserted at least one entry and the
993 * next hash value is not a continuation
996 (count && ((hashval & 1) == 0)))
1000 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1001 "next hash: %x\n", count, *next_hash));
1010 * Directory block splitting, compacting
1014 * Create map of hash values, offsets, and sizes, stored at end of block.
1015 * Returns number of entries mapped.
1017 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1018 struct dx_hash_info *hinfo,
1019 struct dx_map_entry *map_tail)
1022 char *base = (char *) de;
1023 struct dx_hash_info h = *hinfo;
1025 while ((char *) de < base + blocksize) {
1026 if (de->name_len && de->inode) {
1027 ext4fs_dirhash(de->name, de->name_len, &h);
1029 map_tail->hash = h.hash;
1030 map_tail->offs = ((char *) de - base)>>2;
1031 map_tail->size = le16_to_cpu(de->rec_len);
1035 /* XXX: do we need to check rec_len == 0 case? -Chris */
1036 de = ext4_next_entry(de, blocksize);
1041 /* Sort map by hash value */
1042 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1044 struct dx_map_entry *p, *q, *top = map + count - 1;
1046 /* Combsort until bubble sort doesn't suck */
1048 count = count*10/13;
1049 if (count - 9 < 2) /* 9, 10 -> 11 */
1051 for (p = top, q = p - count; q >= map; p--, q--)
1052 if (p->hash < q->hash)
1055 /* Garden variety bubble sort */
1060 if (q[1].hash >= q[0].hash)
1068 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1070 struct dx_entry *entries = frame->entries;
1071 struct dx_entry *old = frame->at, *new = old + 1;
1072 int count = dx_get_count(entries);
1074 assert(count < dx_get_limit(entries));
1075 assert(old < entries + count);
1076 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1077 dx_set_hash(new, hash);
1078 dx_set_block(new, block);
1079 dx_set_count(entries, count + 1);
1083 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1085 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1086 * `de != NULL' is guaranteed by caller.
1088 static inline int ext4_match (int len, const char * const name,
1089 struct ext4_dir_entry_2 * de)
1091 if (len != de->name_len)
1095 return !memcmp(name, de->name, len);
1099 * Returns 0 if not found, -1 on failure, and 1 on success
1101 static inline int search_dirblock(struct buffer_head *bh,
1103 const struct qstr *d_name,
1104 unsigned int offset,
1105 struct ext4_dir_entry_2 ** res_dir)
1107 struct ext4_dir_entry_2 * de;
1110 const char *name = d_name->name;
1111 int namelen = d_name->len;
1113 de = (struct ext4_dir_entry_2 *) bh->b_data;
1114 dlimit = bh->b_data + dir->i_sb->s_blocksize;
1115 while ((char *) de < dlimit) {
1116 /* this code is executed quadratically often */
1117 /* do minimal checking `by hand' */
1119 if ((char *) de + namelen <= dlimit &&
1120 ext4_match (namelen, name, de)) {
1121 /* found a match - just to be sure, do a full check */
1122 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1123 bh->b_size, offset))
1128 /* prevent looping on a bad block */
1129 de_len = ext4_rec_len_from_disk(de->rec_len,
1130 dir->i_sb->s_blocksize);
1134 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1139 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1140 struct ext4_dir_entry *de)
1142 struct super_block *sb = dir->i_sb;
1148 if (de->inode == 0 &&
1149 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1158 * finds an entry in the specified directory with the wanted name. It
1159 * returns the cache buffer in which the entry was found, and the entry
1160 * itself (as a parameter - res_dir). It does NOT read the inode of the
1161 * entry - you'll have to do that yourself if you want to.
1163 * The returned buffer_head has ->b_count elevated. The caller is expected
1164 * to brelse() it when appropriate.
1166 static struct buffer_head * ext4_find_entry (struct inode *dir,
1167 const struct qstr *d_name,
1168 struct ext4_dir_entry_2 ** res_dir)
1170 struct super_block *sb;
1171 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1172 struct buffer_head *bh, *ret = NULL;
1173 ext4_lblk_t start, block, b;
1174 const u8 *name = d_name->name;
1175 int ra_max = 0; /* Number of bh's in the readahead
1177 int ra_ptr = 0; /* Current index into readahead
1180 ext4_lblk_t nblocks;
1186 namelen = d_name->len;
1187 if (namelen > EXT4_NAME_LEN)
1189 if ((namelen <= 2) && (name[0] == '.') &&
1190 (name[1] == '.' || name[1] == '\0')) {
1192 * "." or ".." will only be in the first block
1193 * NFS may look up ".."; "." should be handled by the VFS
1200 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
1202 * On success, or if the error was file not found,
1203 * return. Otherwise, fall back to doing a search the
1204 * old fashioned way.
1206 if (bh || (err != ERR_BAD_DX_DIR))
1208 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1211 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1212 start = EXT4_I(dir)->i_dir_start_lookup;
1213 if (start >= nblocks)
1219 * We deal with the read-ahead logic here.
1221 if (ra_ptr >= ra_max) {
1222 /* Refill the readahead buffer */
1225 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1227 * Terminate if we reach the end of the
1228 * directory and must wrap, or if our
1229 * search has finished at this block.
1231 if (b >= nblocks || (num && block == start)) {
1232 bh_use[ra_max] = NULL;
1236 bh = ext4_getblk(NULL, dir, b++, 0, &err);
1237 bh_use[ra_max] = bh;
1239 ll_rw_block(READ | REQ_META | REQ_PRIO,
1243 if ((bh = bh_use[ra_ptr++]) == NULL)
1246 if (!buffer_uptodate(bh)) {
1247 /* read error, skip block & hope for the best */
1248 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1249 (unsigned long) block);
1253 if (!buffer_verified(bh) &&
1254 !is_dx_internal_node(dir, block,
1255 (struct ext4_dir_entry *)bh->b_data) &&
1256 !ext4_dirent_csum_verify(dir,
1257 (struct ext4_dir_entry *)bh->b_data)) {
1258 EXT4_ERROR_INODE(dir, "checksumming directory "
1259 "block %lu", (unsigned long)block);
1263 set_buffer_verified(bh);
1264 i = search_dirblock(bh, dir, d_name,
1265 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1267 EXT4_I(dir)->i_dir_start_lookup = block;
1269 goto cleanup_and_exit;
1273 goto cleanup_and_exit;
1276 if (++block >= nblocks)
1278 } while (block != start);
1281 * If the directory has grown while we were searching, then
1282 * search the last part of the directory before giving up.
1285 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1286 if (block < nblocks) {
1292 /* Clean up the read-ahead blocks */
1293 for (; ra_ptr < ra_max; ra_ptr++)
1294 brelse(bh_use[ra_ptr]);
1298 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1299 struct ext4_dir_entry_2 **res_dir, int *err)
1301 struct super_block * sb = dir->i_sb;
1302 struct dx_hash_info hinfo;
1303 struct dx_frame frames[2], *frame;
1304 struct buffer_head *bh;
1308 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1311 block = dx_get_block(frame->at);
1312 if (!(bh = ext4_bread(NULL, dir, block, 0, err))) {
1315 ext4_error(dir->i_sb,
1316 "Directory hole detected on inode %lu\n",
1322 if (!buffer_verified(bh) &&
1323 !ext4_dirent_csum_verify(dir,
1324 (struct ext4_dir_entry *)bh->b_data)) {
1325 EXT4_ERROR_INODE(dir, "checksumming directory "
1326 "block %lu", (unsigned long)block);
1331 set_buffer_verified(bh);
1332 retval = search_dirblock(bh, dir, d_name,
1333 block << EXT4_BLOCK_SIZE_BITS(sb),
1335 if (retval == 1) { /* Success! */
1341 *err = ERR_BAD_DX_DIR;
1345 /* Check to see if we should continue to search */
1346 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1350 "error reading index page in directory #%lu",
1355 } while (retval == 1);
1359 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1360 dx_release (frames);
1364 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1366 struct inode *inode;
1367 struct ext4_dir_entry_2 *de;
1368 struct buffer_head *bh;
1370 if (dentry->d_name.len > EXT4_NAME_LEN)
1371 return ERR_PTR(-ENAMETOOLONG);
1373 bh = ext4_find_entry(dir, &dentry->d_name, &de);
1376 __u32 ino = le32_to_cpu(de->inode);
1378 if (!ext4_valid_inum(dir->i_sb, ino)) {
1379 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1380 return ERR_PTR(-EIO);
1382 if (unlikely(ino == dir->i_ino)) {
1383 EXT4_ERROR_INODE(dir, "'%.*s' linked to parent dir",
1385 dentry->d_name.name);
1386 return ERR_PTR(-EIO);
1388 inode = ext4_iget(dir->i_sb, ino);
1389 if (inode == ERR_PTR(-ESTALE)) {
1390 EXT4_ERROR_INODE(dir,
1391 "deleted inode referenced: %u",
1393 return ERR_PTR(-EIO);
1396 return d_splice_alias(inode, dentry);
1400 struct dentry *ext4_get_parent(struct dentry *child)
1403 static const struct qstr dotdot = QSTR_INIT("..", 2);
1404 struct ext4_dir_entry_2 * de;
1405 struct buffer_head *bh;
1407 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1409 return ERR_PTR(-ENOENT);
1410 ino = le32_to_cpu(de->inode);
1413 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1414 EXT4_ERROR_INODE(child->d_inode,
1415 "bad parent inode number: %u", ino);
1416 return ERR_PTR(-EIO);
1419 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1423 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1424 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1425 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1426 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1427 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1428 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1429 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1430 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
1433 static inline void ext4_set_de_type(struct super_block *sb,
1434 struct ext4_dir_entry_2 *de,
1436 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1437 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1441 * Move count entries from end of map between two memory locations.
1442 * Returns pointer to last entry moved.
1444 static struct ext4_dir_entry_2 *
1445 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1448 unsigned rec_len = 0;
1451 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1452 (from + (map->offs<<2));
1453 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1454 memcpy (to, de, rec_len);
1455 ((struct ext4_dir_entry_2 *) to)->rec_len =
1456 ext4_rec_len_to_disk(rec_len, blocksize);
1461 return (struct ext4_dir_entry_2 *) (to - rec_len);
1465 * Compact each dir entry in the range to the minimal rec_len.
1466 * Returns pointer to last entry in range.
1468 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1470 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1471 unsigned rec_len = 0;
1474 while ((char*)de < base + blocksize) {
1475 next = ext4_next_entry(de, blocksize);
1476 if (de->inode && de->name_len) {
1477 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1479 memmove(to, de, rec_len);
1480 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1482 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1490 * Split a full leaf block to make room for a new dir entry.
1491 * Allocate a new block, and move entries so that they are approx. equally full.
1492 * Returns pointer to de in block into which the new entry will be inserted.
1494 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1495 struct buffer_head **bh,struct dx_frame *frame,
1496 struct dx_hash_info *hinfo, int *error)
1498 unsigned blocksize = dir->i_sb->s_blocksize;
1499 unsigned count, continued;
1500 struct buffer_head *bh2;
1501 ext4_lblk_t newblock;
1503 struct dx_map_entry *map;
1504 char *data1 = (*bh)->b_data, *data2;
1505 unsigned split, move, size;
1506 struct ext4_dir_entry_2 *de = NULL, *de2;
1507 struct ext4_dir_entry_tail *t;
1511 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1512 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1513 csum_size = sizeof(struct ext4_dir_entry_tail);
1515 bh2 = ext4_append (handle, dir, &newblock, &err);
1522 BUFFER_TRACE(*bh, "get_write_access");
1523 err = ext4_journal_get_write_access(handle, *bh);
1527 BUFFER_TRACE(frame->bh, "get_write_access");
1528 err = ext4_journal_get_write_access(handle, frame->bh);
1532 data2 = bh2->b_data;
1534 /* create map in the end of data2 block */
1535 map = (struct dx_map_entry *) (data2 + blocksize);
1536 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1537 blocksize, hinfo, map);
1539 dx_sort_map(map, count);
1540 /* Split the existing block in the middle, size-wise */
1543 for (i = count-1; i >= 0; i--) {
1544 /* is more than half of this entry in 2nd half of the block? */
1545 if (size + map[i].size/2 > blocksize/2)
1547 size += map[i].size;
1550 /* map index at which we will split */
1551 split = count - move;
1552 hash2 = map[split].hash;
1553 continued = hash2 == map[split - 1].hash;
1554 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1555 (unsigned long)dx_get_block(frame->at),
1556 hash2, split, count-split));
1558 /* Fancy dance to stay within two buffers */
1559 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1560 de = dx_pack_dirents(data1, blocksize);
1561 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1564 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1568 t = EXT4_DIRENT_TAIL(data2, blocksize);
1569 initialize_dirent_tail(t, blocksize);
1571 t = EXT4_DIRENT_TAIL(data1, blocksize);
1572 initialize_dirent_tail(t, blocksize);
1575 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1576 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1578 /* Which block gets the new entry? */
1579 if (hinfo->hash >= hash2)
1584 dx_insert_block(frame, hash2 + continued, newblock);
1585 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1588 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1592 dxtrace(dx_show_index("frame", frame->entries));
1599 ext4_std_error(dir->i_sb, err);
1605 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1606 struct buffer_head *bh,
1607 void *buf, int buf_size,
1608 const char *name, int namelen,
1609 struct ext4_dir_entry_2 **dest_de)
1611 struct ext4_dir_entry_2 *de;
1612 unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1614 unsigned int offset = 0;
1617 de = (struct ext4_dir_entry_2 *)buf;
1618 top = buf + buf_size - reclen;
1619 while ((char *) de <= top) {
1620 if (ext4_check_dir_entry(dir, NULL, de, bh,
1621 buf, buf_size, offset))
1623 if (ext4_match(namelen, name, de))
1625 nlen = EXT4_DIR_REC_LEN(de->name_len);
1626 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1627 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1629 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1632 if ((char *) de > top)
1639 void ext4_insert_dentry(struct inode *inode,
1640 struct ext4_dir_entry_2 *de,
1642 const char *name, int namelen)
1647 nlen = EXT4_DIR_REC_LEN(de->name_len);
1648 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1650 struct ext4_dir_entry_2 *de1 =
1651 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1652 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1653 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1656 de->file_type = EXT4_FT_UNKNOWN;
1657 de->inode = cpu_to_le32(inode->i_ino);
1658 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1659 de->name_len = namelen;
1660 memcpy(de->name, name, namelen);
1663 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1664 * it points to a directory entry which is guaranteed to be large
1665 * enough for new directory entry. If de is NULL, then
1666 * add_dirent_to_buf will attempt search the directory block for
1667 * space. It will return -ENOSPC if no space is available, and -EIO
1668 * and -EEXIST if directory entry already exists.
1670 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1671 struct inode *inode, struct ext4_dir_entry_2 *de,
1672 struct buffer_head *bh)
1674 struct inode *dir = dentry->d_parent->d_inode;
1675 const char *name = dentry->d_name.name;
1676 int namelen = dentry->d_name.len;
1677 unsigned int blocksize = dir->i_sb->s_blocksize;
1678 unsigned short reclen;
1682 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1683 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1684 csum_size = sizeof(struct ext4_dir_entry_tail);
1686 reclen = EXT4_DIR_REC_LEN(namelen);
1688 err = ext4_find_dest_de(dir, inode,
1689 bh, bh->b_data, blocksize - csum_size,
1690 name, namelen, &de);
1694 BUFFER_TRACE(bh, "get_write_access");
1695 err = ext4_journal_get_write_access(handle, bh);
1697 ext4_std_error(dir->i_sb, err);
1701 /* By now the buffer is marked for journaling */
1702 ext4_insert_dentry(inode, de, blocksize, name, namelen);
1705 * XXX shouldn't update any times until successful
1706 * completion of syscall, but too many callers depend
1709 * XXX similarly, too many callers depend on
1710 * ext4_new_inode() setting the times, but error
1711 * recovery deletes the inode, so the worst that can
1712 * happen is that the times are slightly out of date
1713 * and/or different from the directory change time.
1715 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1716 ext4_update_dx_flag(dir);
1718 ext4_mark_inode_dirty(handle, dir);
1719 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1720 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1722 ext4_std_error(dir->i_sb, err);
1727 * This converts a one block unindexed directory to a 3 block indexed
1728 * directory, and adds the dentry to the indexed directory.
1730 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1731 struct inode *inode, struct buffer_head *bh)
1733 struct inode *dir = dentry->d_parent->d_inode;
1734 const char *name = dentry->d_name.name;
1735 int namelen = dentry->d_name.len;
1736 struct buffer_head *bh2;
1737 struct dx_root *root;
1738 struct dx_frame frames[2], *frame;
1739 struct dx_entry *entries;
1740 struct ext4_dir_entry_2 *de, *de2;
1741 struct ext4_dir_entry_tail *t;
1746 struct dx_hash_info hinfo;
1748 struct fake_dirent *fde;
1751 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1752 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1753 csum_size = sizeof(struct ext4_dir_entry_tail);
1755 blocksize = dir->i_sb->s_blocksize;
1756 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1757 retval = ext4_journal_get_write_access(handle, bh);
1759 ext4_std_error(dir->i_sb, retval);
1763 root = (struct dx_root *) bh->b_data;
1765 /* The 0th block becomes the root, move the dirents out */
1766 fde = &root->dotdot;
1767 de = (struct ext4_dir_entry_2 *)((char *)fde +
1768 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1769 if ((char *) de >= (((char *) root) + blocksize)) {
1770 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1774 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1776 /* Allocate new block for the 0th block's dirents */
1777 bh2 = ext4_append(handle, dir, &block, &retval);
1782 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1783 data1 = bh2->b_data;
1785 memcpy (data1, de, len);
1786 de = (struct ext4_dir_entry_2 *) data1;
1788 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1790 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1795 t = EXT4_DIRENT_TAIL(data1, blocksize);
1796 initialize_dirent_tail(t, blocksize);
1799 /* Initialize the root; the dot dirents already exist */
1800 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1801 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1803 memset (&root->info, 0, sizeof(root->info));
1804 root->info.info_length = sizeof(root->info);
1805 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1806 entries = root->entries;
1807 dx_set_block(entries, 1);
1808 dx_set_count(entries, 1);
1809 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1811 /* Initialize as for dx_probe */
1812 hinfo.hash_version = root->info.hash_version;
1813 if (hinfo.hash_version <= DX_HASH_TEA)
1814 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1815 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1816 ext4fs_dirhash(name, namelen, &hinfo);
1818 frame->entries = entries;
1819 frame->at = entries;
1823 ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1824 ext4_handle_dirty_dirent_node(handle, dir, bh);
1826 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1829 * Even if the block split failed, we have to properly write
1830 * out all the changes we did so far. Otherwise we can end up
1831 * with corrupted filesystem.
1833 ext4_mark_inode_dirty(handle, dir);
1839 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1847 * adds a file entry to the specified directory, using the same
1848 * semantics as ext4_find_entry(). It returns NULL if it failed.
1850 * NOTE!! The inode part of 'de' is left at 0 - which means you
1851 * may not sleep between calling this and putting something into
1852 * the entry, as someone else might have used it while you slept.
1854 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1855 struct inode *inode)
1857 struct inode *dir = dentry->d_parent->d_inode;
1858 struct buffer_head *bh;
1859 struct ext4_dir_entry_2 *de;
1860 struct ext4_dir_entry_tail *t;
1861 struct super_block *sb;
1865 ext4_lblk_t block, blocks;
1868 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1869 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1870 csum_size = sizeof(struct ext4_dir_entry_tail);
1873 blocksize = sb->s_blocksize;
1874 if (!dentry->d_name.len)
1877 if (ext4_has_inline_data(dir)) {
1878 retval = ext4_try_add_inline_entry(handle, dentry, inode);
1888 retval = ext4_dx_add_entry(handle, dentry, inode);
1889 if (!retval || (retval != ERR_BAD_DX_DIR))
1891 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1893 ext4_mark_inode_dirty(handle, dir);
1895 blocks = dir->i_size >> sb->s_blocksize_bits;
1896 for (block = 0; block < blocks; block++) {
1897 if (!(bh = ext4_bread(handle, dir, block, 0, &retval))) {
1900 ext4_error(inode->i_sb,
1901 "Directory hole detected on inode %lu\n",
1906 if (!buffer_verified(bh) &&
1907 !ext4_dirent_csum_verify(dir,
1908 (struct ext4_dir_entry *)bh->b_data))
1910 set_buffer_verified(bh);
1911 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1912 if (retval != -ENOSPC) {
1917 if (blocks == 1 && !dx_fallback &&
1918 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1919 return make_indexed_dir(handle, dentry, inode, bh);
1922 bh = ext4_append(handle, dir, &block, &retval);
1925 de = (struct ext4_dir_entry_2 *) bh->b_data;
1927 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1930 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1931 initialize_dirent_tail(t, blocksize);
1934 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1937 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1942 * Returns 0 for success, or a negative error value
1944 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1945 struct inode *inode)
1947 struct dx_frame frames[2], *frame;
1948 struct dx_entry *entries, *at;
1949 struct dx_hash_info hinfo;
1950 struct buffer_head *bh;
1951 struct inode *dir = dentry->d_parent->d_inode;
1952 struct super_block *sb = dir->i_sb;
1953 struct ext4_dir_entry_2 *de;
1956 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1959 entries = frame->entries;
1962 if (!(bh = ext4_bread(handle, dir, dx_get_block(frame->at), 0, &err))) {
1965 ext4_error(dir->i_sb,
1966 "Directory hole detected on inode %lu\n",
1972 if (!buffer_verified(bh) &&
1973 !ext4_dirent_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data))
1975 set_buffer_verified(bh);
1977 BUFFER_TRACE(bh, "get_write_access");
1978 err = ext4_journal_get_write_access(handle, bh);
1982 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1986 /* Block full, should compress but for now just split */
1987 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1988 dx_get_count(entries), dx_get_limit(entries)));
1989 /* Need to split index? */
1990 if (dx_get_count(entries) == dx_get_limit(entries)) {
1991 ext4_lblk_t newblock;
1992 unsigned icount = dx_get_count(entries);
1993 int levels = frame - frames;
1994 struct dx_entry *entries2;
1995 struct dx_node *node2;
1996 struct buffer_head *bh2;
1998 if (levels && (dx_get_count(frames->entries) ==
1999 dx_get_limit(frames->entries))) {
2000 ext4_warning(sb, "Directory index full!");
2004 bh2 = ext4_append (handle, dir, &newblock, &err);
2007 node2 = (struct dx_node *)(bh2->b_data);
2008 entries2 = node2->entries;
2009 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2010 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2012 BUFFER_TRACE(frame->bh, "get_write_access");
2013 err = ext4_journal_get_write_access(handle, frame->bh);
2017 unsigned icount1 = icount/2, icount2 = icount - icount1;
2018 unsigned hash2 = dx_get_hash(entries + icount1);
2019 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2022 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2023 err = ext4_journal_get_write_access(handle,
2028 memcpy((char *) entries2, (char *) (entries + icount1),
2029 icount2 * sizeof(struct dx_entry));
2030 dx_set_count(entries, icount1);
2031 dx_set_count(entries2, icount2);
2032 dx_set_limit(entries2, dx_node_limit(dir));
2034 /* Which index block gets the new entry? */
2035 if (at - entries >= icount1) {
2036 frame->at = at = at - entries - icount1 + entries2;
2037 frame->entries = entries = entries2;
2038 swap(frame->bh, bh2);
2040 dx_insert_block(frames + 0, hash2, newblock);
2041 dxtrace(dx_show_index("node", frames[1].entries));
2042 dxtrace(dx_show_index("node",
2043 ((struct dx_node *) bh2->b_data)->entries));
2044 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2049 dxtrace(printk(KERN_DEBUG
2050 "Creating second level index...\n"));
2051 memcpy((char *) entries2, (char *) entries,
2052 icount * sizeof(struct dx_entry));
2053 dx_set_limit(entries2, dx_node_limit(dir));
2056 dx_set_count(entries, 1);
2057 dx_set_block(entries + 0, newblock);
2058 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2060 /* Add new access path frame */
2062 frame->at = at = at - entries + entries2;
2063 frame->entries = entries = entries2;
2065 err = ext4_journal_get_write_access(handle,
2070 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2072 ext4_std_error(inode->i_sb, err);
2076 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
2079 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2083 ext4_std_error(dir->i_sb, err);
2092 * ext4_delete_entry deletes a directory entry by merging it with the
2095 static int ext4_delete_entry(handle_t *handle,
2097 struct ext4_dir_entry_2 *de_del,
2098 struct buffer_head *bh)
2100 struct ext4_dir_entry_2 *de, *pde;
2101 unsigned int blocksize = dir->i_sb->s_blocksize;
2105 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2106 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2107 csum_size = sizeof(struct ext4_dir_entry_tail);
2111 de = (struct ext4_dir_entry_2 *) bh->b_data;
2112 while (i < bh->b_size - csum_size) {
2113 if (ext4_check_dir_entry(dir, NULL, de, bh,
2114 bh->b_data, bh->b_size, i))
2117 BUFFER_TRACE(bh, "get_write_access");
2118 err = ext4_journal_get_write_access(handle, bh);
2119 if (unlikely(err)) {
2120 ext4_std_error(dir->i_sb, err);
2124 pde->rec_len = ext4_rec_len_to_disk(
2125 ext4_rec_len_from_disk(pde->rec_len,
2127 ext4_rec_len_from_disk(de->rec_len,
2133 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2134 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2135 if (unlikely(err)) {
2136 ext4_std_error(dir->i_sb, err);
2141 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2143 de = ext4_next_entry(de, blocksize);
2149 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2150 * since this indicates that nlinks count was previously 1.
2152 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2155 if (is_dx(inode) && inode->i_nlink > 1) {
2156 /* limit is 16-bit i_links_count */
2157 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2158 set_nlink(inode, 1);
2159 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2160 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2166 * If a directory had nlink == 1, then we should let it be 1. This indicates
2167 * directory has >EXT4_LINK_MAX subdirs.
2169 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2171 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2176 static int ext4_add_nondir(handle_t *handle,
2177 struct dentry *dentry, struct inode *inode)
2179 int err = ext4_add_entry(handle, dentry, inode);
2181 ext4_mark_inode_dirty(handle, inode);
2182 unlock_new_inode(inode);
2183 d_instantiate(dentry, inode);
2187 unlock_new_inode(inode);
2193 * By the time this is called, we already have created
2194 * the directory cache entry for the new file, but it
2195 * is so far negative - it has no inode.
2197 * If the create succeeds, we fill in the inode information
2198 * with d_instantiate().
2200 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2204 struct inode *inode;
2205 int err, retries = 0;
2207 dquot_initialize(dir);
2210 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2211 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2212 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2214 return PTR_ERR(handle);
2216 if (IS_DIRSYNC(dir))
2217 ext4_handle_sync(handle);
2219 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
2220 err = PTR_ERR(inode);
2221 if (!IS_ERR(inode)) {
2222 inode->i_op = &ext4_file_inode_operations;
2223 inode->i_fop = &ext4_file_operations;
2224 ext4_set_aops(inode);
2225 err = ext4_add_nondir(handle, dentry, inode);
2227 ext4_journal_stop(handle);
2228 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2233 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2234 umode_t mode, dev_t rdev)
2237 struct inode *inode;
2238 int err, retries = 0;
2240 if (!new_valid_dev(rdev))
2243 dquot_initialize(dir);
2246 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2247 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2248 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2250 return PTR_ERR(handle);
2252 if (IS_DIRSYNC(dir))
2253 ext4_handle_sync(handle);
2255 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
2256 err = PTR_ERR(inode);
2257 if (!IS_ERR(inode)) {
2258 init_special_inode(inode, inode->i_mode, rdev);
2259 inode->i_op = &ext4_special_inode_operations;
2260 err = ext4_add_nondir(handle, dentry, inode);
2262 ext4_journal_stop(handle);
2263 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2268 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2269 struct ext4_dir_entry_2 *de,
2270 int blocksize, int csum_size,
2271 unsigned int parent_ino, int dotdot_real_len)
2273 de->inode = cpu_to_le32(inode->i_ino);
2275 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2277 strcpy(de->name, ".");
2278 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2280 de = ext4_next_entry(de, blocksize);
2281 de->inode = cpu_to_le32(parent_ino);
2283 if (!dotdot_real_len)
2284 de->rec_len = ext4_rec_len_to_disk(blocksize -
2285 (csum_size + EXT4_DIR_REC_LEN(1)),
2288 de->rec_len = ext4_rec_len_to_disk(
2289 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2290 strcpy(de->name, "..");
2291 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2293 return ext4_next_entry(de, blocksize);
2296 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2297 struct inode *inode)
2299 struct buffer_head *dir_block = NULL;
2300 struct ext4_dir_entry_2 *de;
2301 struct ext4_dir_entry_tail *t;
2302 unsigned int blocksize = dir->i_sb->s_blocksize;
2306 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2307 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2308 csum_size = sizeof(struct ext4_dir_entry_tail);
2310 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2311 err = ext4_try_create_inline_dir(handle, dir, inode);
2312 if (err < 0 && err != -ENOSPC)
2318 inode->i_size = EXT4_I(inode)->i_disksize = blocksize;
2319 dir_block = ext4_bread(handle, inode, 0, 1, &err);
2320 if (!(dir_block = ext4_bread(handle, inode, 0, 1, &err))) {
2323 ext4_error(inode->i_sb,
2324 "Directory hole detected on inode %lu\n",
2329 BUFFER_TRACE(dir_block, "get_write_access");
2330 err = ext4_journal_get_write_access(handle, dir_block);
2333 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2334 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2335 set_nlink(inode, 2);
2337 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2338 initialize_dirent_tail(t, blocksize);
2341 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2342 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2345 set_buffer_verified(dir_block);
2351 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2354 struct inode *inode;
2355 int err, retries = 0;
2357 if (EXT4_DIR_LINK_MAX(dir))
2360 dquot_initialize(dir);
2363 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2364 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2365 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2367 return PTR_ERR(handle);
2369 if (IS_DIRSYNC(dir))
2370 ext4_handle_sync(handle);
2372 inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
2373 &dentry->d_name, 0, NULL);
2374 err = PTR_ERR(inode);
2378 inode->i_op = &ext4_dir_inode_operations;
2379 inode->i_fop = &ext4_dir_operations;
2380 err = ext4_init_new_dir(handle, dir, inode);
2382 goto out_clear_inode;
2383 err = ext4_mark_inode_dirty(handle, inode);
2385 err = ext4_add_entry(handle, dentry, inode);
2389 unlock_new_inode(inode);
2390 ext4_mark_inode_dirty(handle, inode);
2394 ext4_inc_count(handle, dir);
2395 ext4_update_dx_flag(dir);
2396 err = ext4_mark_inode_dirty(handle, dir);
2398 goto out_clear_inode;
2399 unlock_new_inode(inode);
2400 d_instantiate(dentry, inode);
2402 ext4_journal_stop(handle);
2403 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2409 * routine to check that the specified directory is empty (for rmdir)
2411 static int empty_dir(struct inode *inode)
2413 unsigned int offset;
2414 struct buffer_head *bh;
2415 struct ext4_dir_entry_2 *de, *de1;
2416 struct super_block *sb;
2420 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
2421 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
2423 EXT4_ERROR_INODE(inode,
2424 "error %d reading directory lblock 0", err);
2426 ext4_warning(inode->i_sb,
2427 "bad directory (dir #%lu) - no data block",
2431 if (!buffer_verified(bh) &&
2432 !ext4_dirent_csum_verify(inode,
2433 (struct ext4_dir_entry *)bh->b_data)) {
2434 EXT4_ERROR_INODE(inode, "checksum error reading directory "
2438 set_buffer_verified(bh);
2439 de = (struct ext4_dir_entry_2 *) bh->b_data;
2440 de1 = ext4_next_entry(de, sb->s_blocksize);
2441 if (le32_to_cpu(de->inode) != inode->i_ino ||
2442 !le32_to_cpu(de1->inode) ||
2443 strcmp(".", de->name) ||
2444 strcmp("..", de1->name)) {
2445 ext4_warning(inode->i_sb,
2446 "bad directory (dir #%lu) - no `.' or `..'",
2451 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2452 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2453 de = ext4_next_entry(de1, sb->s_blocksize);
2454 while (offset < inode->i_size) {
2456 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2457 unsigned int lblock;
2460 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2461 bh = ext4_bread(NULL, inode, lblock, 0, &err);
2464 EXT4_ERROR_INODE(inode,
2465 "error %d reading directory "
2466 "lblock %u", err, lblock);
2468 ext4_warning(inode->i_sb,
2469 "bad directory (dir #%lu) - no data block",
2472 offset += sb->s_blocksize;
2475 if (!buffer_verified(bh) &&
2476 !ext4_dirent_csum_verify(inode,
2477 (struct ext4_dir_entry *)bh->b_data)) {
2478 EXT4_ERROR_INODE(inode, "checksum error "
2479 "reading directory lblock 0");
2482 set_buffer_verified(bh);
2483 de = (struct ext4_dir_entry_2 *) bh->b_data;
2485 if (ext4_check_dir_entry(inode, NULL, de, bh,
2486 bh->b_data, bh->b_size, offset)) {
2487 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2489 offset = (offset | (sb->s_blocksize - 1)) + 1;
2492 if (le32_to_cpu(de->inode)) {
2496 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2497 de = ext4_next_entry(de, sb->s_blocksize);
2503 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
2504 * such inodes, starting at the superblock, in case we crash before the
2505 * file is closed/deleted, or in case the inode truncate spans multiple
2506 * transactions and the last transaction is not recovered after a crash.
2508 * At filesystem recovery time, we walk this list deleting unlinked
2509 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2511 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2513 struct super_block *sb = inode->i_sb;
2514 struct ext4_iloc iloc;
2517 if (!EXT4_SB(sb)->s_journal)
2520 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
2521 if (!list_empty(&EXT4_I(inode)->i_orphan))
2525 * Orphan handling is only valid for files with data blocks
2526 * being truncated, or files being unlinked. Note that we either
2527 * hold i_mutex, or the inode can not be referenced from outside,
2528 * so i_nlink should not be bumped due to race
2530 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2531 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2533 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2534 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2538 err = ext4_reserve_inode_write(handle, inode, &iloc);
2542 * Due to previous errors inode may be already a part of on-disk
2543 * orphan list. If so skip on-disk list modification.
2545 if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2546 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2549 /* Insert this inode at the head of the on-disk orphan list... */
2550 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2551 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2552 err = ext4_handle_dirty_super(handle, sb);
2553 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2557 /* Only add to the head of the in-memory list if all the
2558 * previous operations succeeded. If the orphan_add is going to
2559 * fail (possibly taking the journal offline), we can't risk
2560 * leaving the inode on the orphan list: stray orphan-list
2561 * entries can cause panics at unmount time.
2563 * This is safe: on error we're going to ignore the orphan list
2564 * anyway on the next recovery. */
2567 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2569 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2570 jbd_debug(4, "orphan inode %lu will point to %d\n",
2571 inode->i_ino, NEXT_ORPHAN(inode));
2573 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2574 ext4_std_error(inode->i_sb, err);
2579 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2580 * of such inodes stored on disk, because it is finally being cleaned up.
2582 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2584 struct list_head *prev;
2585 struct ext4_inode_info *ei = EXT4_I(inode);
2586 struct ext4_sb_info *sbi;
2588 struct ext4_iloc iloc;
2591 if (!EXT4_SB(inode->i_sb)->s_journal)
2594 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2595 if (list_empty(&ei->i_orphan))
2598 ino_next = NEXT_ORPHAN(inode);
2599 prev = ei->i_orphan.prev;
2600 sbi = EXT4_SB(inode->i_sb);
2602 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2604 list_del_init(&ei->i_orphan);
2606 /* If we're on an error path, we may not have a valid
2607 * transaction handle with which to update the orphan list on
2608 * disk, but we still need to remove the inode from the linked
2609 * list in memory. */
2613 err = ext4_reserve_inode_write(handle, inode, &iloc);
2617 if (prev == &sbi->s_orphan) {
2618 jbd_debug(4, "superblock will point to %u\n", ino_next);
2619 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2620 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2623 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2624 err = ext4_handle_dirty_super(handle, inode->i_sb);
2626 struct ext4_iloc iloc2;
2627 struct inode *i_prev =
2628 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2630 jbd_debug(4, "orphan inode %lu will point to %u\n",
2631 i_prev->i_ino, ino_next);
2632 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2635 NEXT_ORPHAN(i_prev) = ino_next;
2636 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2640 NEXT_ORPHAN(inode) = 0;
2641 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2644 ext4_std_error(inode->i_sb, err);
2646 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2654 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2657 struct inode *inode;
2658 struct buffer_head *bh;
2659 struct ext4_dir_entry_2 *de;
2662 /* Initialize quotas before so that eventual writes go in
2663 * separate transaction */
2664 dquot_initialize(dir);
2665 dquot_initialize(dentry->d_inode);
2667 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2669 return PTR_ERR(handle);
2672 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2676 if (IS_DIRSYNC(dir))
2677 ext4_handle_sync(handle);
2679 inode = dentry->d_inode;
2682 if (le32_to_cpu(de->inode) != inode->i_ino)
2685 retval = -ENOTEMPTY;
2686 if (!empty_dir(inode))
2689 retval = ext4_delete_entry(handle, dir, de, bh);
2692 if (!EXT4_DIR_LINK_EMPTY(inode))
2693 ext4_warning(inode->i_sb,
2694 "empty directory has too many links (%d)",
2698 /* There's no need to set i_disksize: the fact that i_nlink is
2699 * zero will ensure that the right thing happens during any
2702 ext4_orphan_add(handle, inode);
2703 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2704 ext4_mark_inode_dirty(handle, inode);
2705 ext4_dec_count(handle, dir);
2706 ext4_update_dx_flag(dir);
2707 ext4_mark_inode_dirty(handle, dir);
2710 ext4_journal_stop(handle);
2715 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2718 struct inode *inode;
2719 struct buffer_head *bh;
2720 struct ext4_dir_entry_2 *de;
2723 trace_ext4_unlink_enter(dir, dentry);
2724 /* Initialize quotas before so that eventual writes go
2725 * in separate transaction */
2726 dquot_initialize(dir);
2727 dquot_initialize(dentry->d_inode);
2729 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2731 return PTR_ERR(handle);
2733 if (IS_DIRSYNC(dir))
2734 ext4_handle_sync(handle);
2737 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2741 inode = dentry->d_inode;
2744 if (le32_to_cpu(de->inode) != inode->i_ino)
2747 if (!inode->i_nlink) {
2748 ext4_warning(inode->i_sb,
2749 "Deleting nonexistent file (%lu), %d",
2750 inode->i_ino, inode->i_nlink);
2751 set_nlink(inode, 1);
2753 retval = ext4_delete_entry(handle, dir, de, bh);
2756 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2757 ext4_update_dx_flag(dir);
2758 ext4_mark_inode_dirty(handle, dir);
2760 if (!inode->i_nlink)
2761 ext4_orphan_add(handle, inode);
2762 inode->i_ctime = ext4_current_time(inode);
2763 ext4_mark_inode_dirty(handle, inode);
2767 ext4_journal_stop(handle);
2769 trace_ext4_unlink_exit(dentry, retval);
2773 static int ext4_symlink(struct inode *dir,
2774 struct dentry *dentry, const char *symname)
2777 struct inode *inode;
2778 int l, err, retries = 0;
2781 l = strlen(symname)+1;
2782 if (l > dir->i_sb->s_blocksize)
2783 return -ENAMETOOLONG;
2785 dquot_initialize(dir);
2787 if (l > EXT4_N_BLOCKS * 4) {
2789 * For non-fast symlinks, we just allocate inode and put it on
2790 * orphan list in the first transaction => we need bitmap,
2791 * group descriptor, sb, inode block, quota blocks, and
2792 * possibly selinux xattr blocks.
2794 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2795 EXT4_XATTR_TRANS_BLOCKS;
2798 * Fast symlink. We have to add entry to directory
2799 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2800 * allocate new inode (bitmap, group descriptor, inode block,
2801 * quota blocks, sb is already counted in previous macros).
2803 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2804 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2805 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2808 handle = ext4_journal_start(dir, credits);
2810 return PTR_ERR(handle);
2812 if (IS_DIRSYNC(dir))
2813 ext4_handle_sync(handle);
2815 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2816 &dentry->d_name, 0, NULL);
2817 err = PTR_ERR(inode);
2821 if (l > EXT4_N_BLOCKS * 4) {
2822 inode->i_op = &ext4_symlink_inode_operations;
2823 ext4_set_aops(inode);
2825 * We cannot call page_symlink() with transaction started
2826 * because it calls into ext4_write_begin() which can wait
2827 * for transaction commit if we are running out of space
2828 * and thus we deadlock. So we have to stop transaction now
2829 * and restart it when symlink contents is written.
2831 * To keep fs consistent in case of crash, we have to put inode
2832 * to orphan list in the mean time.
2835 err = ext4_orphan_add(handle, inode);
2836 ext4_journal_stop(handle);
2838 goto err_drop_inode;
2839 err = __page_symlink(inode, symname, l, 1);
2841 goto err_drop_inode;
2843 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2844 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2846 handle = ext4_journal_start(dir,
2847 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2848 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2849 if (IS_ERR(handle)) {
2850 err = PTR_ERR(handle);
2851 goto err_drop_inode;
2853 set_nlink(inode, 1);
2854 err = ext4_orphan_del(handle, inode);
2856 ext4_journal_stop(handle);
2858 goto err_drop_inode;
2861 /* clear the extent format for fast symlink */
2862 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2863 inode->i_op = &ext4_fast_symlink_inode_operations;
2864 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2865 inode->i_size = l-1;
2867 EXT4_I(inode)->i_disksize = inode->i_size;
2868 err = ext4_add_nondir(handle, dentry, inode);
2870 ext4_journal_stop(handle);
2871 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2875 unlock_new_inode(inode);
2880 static int ext4_link(struct dentry *old_dentry,
2881 struct inode *dir, struct dentry *dentry)
2884 struct inode *inode = old_dentry->d_inode;
2885 int err, retries = 0;
2887 if (inode->i_nlink >= EXT4_LINK_MAX)
2890 dquot_initialize(dir);
2893 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2894 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2896 return PTR_ERR(handle);
2898 if (IS_DIRSYNC(dir))
2899 ext4_handle_sync(handle);
2901 inode->i_ctime = ext4_current_time(inode);
2902 ext4_inc_count(handle, inode);
2905 err = ext4_add_entry(handle, dentry, inode);
2907 ext4_mark_inode_dirty(handle, inode);
2908 d_instantiate(dentry, inode);
2913 ext4_journal_stop(handle);
2914 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2919 #define PARENT_INO(buffer, size) \
2920 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2923 * Anybody can rename anything with this: the permission checks are left to the
2924 * higher-level routines.
2926 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2927 struct inode *new_dir, struct dentry *new_dentry)
2930 struct inode *old_inode, *new_inode;
2931 struct buffer_head *old_bh, *new_bh, *dir_bh;
2932 struct ext4_dir_entry_2 *old_de, *new_de;
2933 int retval, force_da_alloc = 0;
2935 dquot_initialize(old_dir);
2936 dquot_initialize(new_dir);
2938 old_bh = new_bh = dir_bh = NULL;
2940 /* Initialize quotas before so that eventual writes go
2941 * in separate transaction */
2942 if (new_dentry->d_inode)
2943 dquot_initialize(new_dentry->d_inode);
2944 handle = ext4_journal_start(old_dir, 2 *
2945 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2946 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2948 return PTR_ERR(handle);
2950 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2951 ext4_handle_sync(handle);
2953 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2955 * Check for inode number is _not_ due to possible IO errors.
2956 * We might rmdir the source, keep it as pwd of some process
2957 * and merrily kill the link to whatever was created under the
2958 * same name. Goodbye sticky bit ;-<
2960 old_inode = old_dentry->d_inode;
2962 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2965 new_inode = new_dentry->d_inode;
2966 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2973 if (S_ISDIR(old_inode->i_mode)) {
2975 retval = -ENOTEMPTY;
2976 if (!empty_dir(new_inode))
2980 if (!(dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval))) {
2983 ext4_error(old_inode->i_sb,
2984 "Directory hole detected on inode %lu\n",
2989 if (!buffer_verified(dir_bh) &&
2990 !ext4_dirent_csum_verify(old_inode,
2991 (struct ext4_dir_entry *)dir_bh->b_data))
2993 set_buffer_verified(dir_bh);
2994 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2995 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2998 if (!new_inode && new_dir != old_dir &&
2999 EXT4_DIR_LINK_MAX(new_dir))
3001 BUFFER_TRACE(dir_bh, "get_write_access");
3002 retval = ext4_journal_get_write_access(handle, dir_bh);
3007 retval = ext4_add_entry(handle, new_dentry, old_inode);
3011 BUFFER_TRACE(new_bh, "get write access");
3012 retval = ext4_journal_get_write_access(handle, new_bh);
3015 new_de->inode = cpu_to_le32(old_inode->i_ino);
3016 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
3017 EXT4_FEATURE_INCOMPAT_FILETYPE))
3018 new_de->file_type = old_de->file_type;
3019 new_dir->i_version++;
3020 new_dir->i_ctime = new_dir->i_mtime =
3021 ext4_current_time(new_dir);
3022 ext4_mark_inode_dirty(handle, new_dir);
3023 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
3024 retval = ext4_handle_dirty_dirent_node(handle, new_dir, new_bh);
3025 if (unlikely(retval)) {
3026 ext4_std_error(new_dir->i_sb, retval);
3034 * Like most other Unix systems, set the ctime for inodes on a
3037 old_inode->i_ctime = ext4_current_time(old_inode);
3038 ext4_mark_inode_dirty(handle, old_inode);
3043 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
3044 old_de->name_len != old_dentry->d_name.len ||
3045 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
3046 (retval = ext4_delete_entry(handle, old_dir,
3047 old_de, old_bh)) == -ENOENT) {
3048 /* old_de could have moved from under us during htree split, so
3049 * make sure that we are deleting the right entry. We might
3050 * also be pointing to a stale entry in the unused part of
3051 * old_bh so just checking inum and the name isn't enough. */
3052 struct buffer_head *old_bh2;
3053 struct ext4_dir_entry_2 *old_de2;
3055 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
3057 retval = ext4_delete_entry(handle, old_dir,
3063 ext4_warning(old_dir->i_sb,
3064 "Deleting old file (%lu), %d, error=%d",
3065 old_dir->i_ino, old_dir->i_nlink, retval);
3069 ext4_dec_count(handle, new_inode);
3070 new_inode->i_ctime = ext4_current_time(new_inode);
3072 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
3073 ext4_update_dx_flag(old_dir);
3075 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
3076 cpu_to_le32(new_dir->i_ino);
3077 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
3078 if (is_dx(old_inode)) {
3079 retval = ext4_handle_dirty_dx_node(handle,
3083 retval = ext4_handle_dirty_dirent_node(handle,
3088 ext4_std_error(old_dir->i_sb, retval);
3091 ext4_dec_count(handle, old_dir);
3093 /* checked empty_dir above, can't have another parent,
3094 * ext4_dec_count() won't work for many-linked dirs */
3095 clear_nlink(new_inode);
3097 ext4_inc_count(handle, new_dir);
3098 ext4_update_dx_flag(new_dir);
3099 ext4_mark_inode_dirty(handle, new_dir);
3102 ext4_mark_inode_dirty(handle, old_dir);
3104 ext4_mark_inode_dirty(handle, new_inode);
3105 if (!new_inode->i_nlink)
3106 ext4_orphan_add(handle, new_inode);
3107 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
3116 ext4_journal_stop(handle);
3117 if (retval == 0 && force_da_alloc)
3118 ext4_alloc_da_blocks(old_inode);
3123 * directories can handle most operations...
3125 const struct inode_operations ext4_dir_inode_operations = {
3126 .create = ext4_create,
3127 .lookup = ext4_lookup,
3129 .unlink = ext4_unlink,
3130 .symlink = ext4_symlink,
3131 .mkdir = ext4_mkdir,
3132 .rmdir = ext4_rmdir,
3133 .mknod = ext4_mknod,
3134 .rename = ext4_rename,
3135 .setattr = ext4_setattr,
3136 #ifdef CONFIG_EXT4_FS_XATTR
3137 .setxattr = generic_setxattr,
3138 .getxattr = generic_getxattr,
3139 .listxattr = ext4_listxattr,
3140 .removexattr = generic_removexattr,
3142 .get_acl = ext4_get_acl,
3143 .fiemap = ext4_fiemap,
3146 const struct inode_operations ext4_special_inode_operations = {
3147 .setattr = ext4_setattr,
3148 #ifdef CONFIG_EXT4_FS_XATTR
3149 .setxattr = generic_setxattr,
3150 .getxattr = generic_getxattr,
3151 .listxattr = ext4_listxattr,
3152 .removexattr = generic_removexattr,
3154 .get_acl = ext4_get_acl,