davinci: DM644x: add support for SFFSDR board
[linux-ginger.git] / fs / ext4 / namei.c
blob22098e1cd085911a7aeffc975ac7a0773a32bbc0
1 /*
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)
9 * from
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
24 * Theodore Ts'o, 2002
27 #include <linux/fs.h>
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>
37 #include "ext4.h"
38 #include "ext4_jbd2.h"
40 #include "namei.h"
41 #include "xattr.h"
42 #include "acl.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,
53 struct inode *inode,
54 ext4_lblk_t *block, int *err)
56 struct buffer_head *bh;
58 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
60 bh = ext4_bread(handle, inode, *block, 1, err);
61 if (bh) {
62 inode->i_size += inode->i_sb->s_blocksize;
63 EXT4_I(inode)->i_disksize = inode->i_size;
64 *err = ext4_journal_get_write_access(handle, bh);
65 if (*err) {
66 brelse(bh);
67 bh = NULL;
70 return bh;
73 #ifndef assert
74 #define assert(test) J_ASSERT(test)
75 #endif
77 #ifdef DX_DEBUG
78 #define dxtrace(command) command
79 #else
80 #define dxtrace(command)
81 #endif
83 struct fake_dirent
85 __le32 inode;
86 __le16 rec_len;
87 u8 name_len;
88 u8 file_type;
91 struct dx_countlimit
93 __le16 limit;
94 __le16 count;
97 struct dx_entry
99 __le32 hash;
100 __le32 block;
104 * dx_root_info is laid out so that if it should somehow get overlaid by a
105 * dirent the two low bits of the hash version will be zero. Therefore, the
106 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
109 struct dx_root
111 struct fake_dirent dot;
112 char dot_name[4];
113 struct fake_dirent dotdot;
114 char dotdot_name[4];
115 struct dx_root_info
117 __le32 reserved_zero;
118 u8 hash_version;
119 u8 info_length; /* 8 */
120 u8 indirect_levels;
121 u8 unused_flags;
123 info;
124 struct dx_entry entries[0];
127 struct dx_node
129 struct fake_dirent fake;
130 struct dx_entry entries[0];
134 struct dx_frame
136 struct buffer_head *bh;
137 struct dx_entry *entries;
138 struct dx_entry *at;
141 struct dx_map_entry
143 u32 hash;
144 u16 offs;
145 u16 size;
148 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
149 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
150 static inline unsigned dx_get_hash(struct dx_entry *entry);
151 static void dx_set_hash(struct dx_entry *entry, unsigned value);
152 static unsigned dx_get_count(struct dx_entry *entries);
153 static unsigned dx_get_limit(struct dx_entry *entries);
154 static void dx_set_count(struct dx_entry *entries, unsigned value);
155 static void dx_set_limit(struct dx_entry *entries, unsigned value);
156 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
157 static unsigned dx_node_limit(struct inode *dir);
158 static struct dx_frame *dx_probe(const struct qstr *d_name,
159 struct inode *dir,
160 struct dx_hash_info *hinfo,
161 struct dx_frame *frame,
162 int *err);
163 static void dx_release(struct dx_frame *frames);
164 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
165 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
166 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
167 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
168 struct dx_map_entry *offsets, int count, unsigned blocksize);
169 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
170 static void dx_insert_block(struct dx_frame *frame,
171 u32 hash, ext4_lblk_t block);
172 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
173 struct dx_frame *frame,
174 struct dx_frame *frames,
175 __u32 *start_hash);
176 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
177 const struct qstr *d_name,
178 struct ext4_dir_entry_2 **res_dir,
179 int *err);
180 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
181 struct inode *inode);
183 unsigned int ext4_rec_len_from_disk(__le16 dlen, unsigned blocksize)
185 unsigned len = le16_to_cpu(dlen);
187 if (len == EXT4_MAX_REC_LEN || len == 0)
188 return blocksize;
189 return (len & 65532) | ((len & 3) << 16);
192 __le16 ext4_rec_len_to_disk(unsigned len, unsigned blocksize)
194 if ((len > blocksize) || (blocksize > (1 << 18)) || (len & 3))
195 BUG();
196 if (len < 65536)
197 return cpu_to_le16(len);
198 if (len == blocksize) {
199 if (blocksize == 65536)
200 return cpu_to_le16(EXT4_MAX_REC_LEN);
201 else
202 return cpu_to_le16(0);
204 return cpu_to_le16((len & 65532) | ((len >> 16) & 3));
208 * p is at least 6 bytes before the end of page
210 static inline struct ext4_dir_entry_2 *
211 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
213 return (struct ext4_dir_entry_2 *)((char *)p +
214 ext4_rec_len_from_disk(p->rec_len, blocksize));
218 * Future: use high four bits of block for coalesce-on-delete flags
219 * Mask them off for now.
222 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
224 return le32_to_cpu(entry->block) & 0x00ffffff;
227 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
229 entry->block = cpu_to_le32(value);
232 static inline unsigned dx_get_hash(struct dx_entry *entry)
234 return le32_to_cpu(entry->hash);
237 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
239 entry->hash = cpu_to_le32(value);
242 static inline unsigned dx_get_count(struct dx_entry *entries)
244 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
247 static inline unsigned dx_get_limit(struct dx_entry *entries)
249 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
252 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
254 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
257 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
259 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
262 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
264 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
265 EXT4_DIR_REC_LEN(2) - infosize;
266 return entry_space / sizeof(struct dx_entry);
269 static inline unsigned dx_node_limit(struct inode *dir)
271 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
272 return entry_space / sizeof(struct dx_entry);
276 * Debug
278 #ifdef DX_DEBUG
279 static void dx_show_index(char * label, struct dx_entry *entries)
281 int i, n = dx_get_count (entries);
282 printk(KERN_DEBUG "%s index ", label);
283 for (i = 0; i < n; i++) {
284 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
285 0, (unsigned long)dx_get_block(entries + i));
287 printk("\n");
290 struct stats
292 unsigned names;
293 unsigned space;
294 unsigned bcount;
297 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
298 int size, int show_names)
300 unsigned names = 0, space = 0;
301 char *base = (char *) de;
302 struct dx_hash_info h = *hinfo;
304 printk("names: ");
305 while ((char *) de < base + size)
307 if (de->inode)
309 if (show_names)
311 int len = de->name_len;
312 char *name = de->name;
313 while (len--) printk("%c", *name++);
314 ext4fs_dirhash(de->name, de->name_len, &h);
315 printk(":%x.%u ", h.hash,
316 ((char *) de - base));
318 space += EXT4_DIR_REC_LEN(de->name_len);
319 names++;
321 de = ext4_next_entry(de, size);
323 printk("(%i)\n", names);
324 return (struct stats) { names, space, 1 };
327 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
328 struct dx_entry *entries, int levels)
330 unsigned blocksize = dir->i_sb->s_blocksize;
331 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
332 unsigned bcount = 0;
333 struct buffer_head *bh;
334 int err;
335 printk("%i indexed blocks...\n", count);
336 for (i = 0; i < count; i++, entries++)
338 ext4_lblk_t block = dx_get_block(entries);
339 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
340 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
341 struct stats stats;
342 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
343 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
344 stats = levels?
345 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
346 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
347 names += stats.names;
348 space += stats.space;
349 bcount += stats.bcount;
350 brelse(bh);
352 if (bcount)
353 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
354 levels ? "" : " ", names, space/bcount,
355 (space/bcount)*100/blocksize);
356 return (struct stats) { names, space, bcount};
358 #endif /* DX_DEBUG */
361 * Probe for a directory leaf block to search.
363 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
364 * error in the directory index, and the caller should fall back to
365 * searching the directory normally. The callers of dx_probe **MUST**
366 * check for this error code, and make sure it never gets reflected
367 * back to userspace.
369 static struct dx_frame *
370 dx_probe(const struct qstr *d_name, struct inode *dir,
371 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
373 unsigned count, indirect;
374 struct dx_entry *at, *entries, *p, *q, *m;
375 struct dx_root *root;
376 struct buffer_head *bh;
377 struct dx_frame *frame = frame_in;
378 u32 hash;
380 frame->bh = NULL;
381 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
382 goto fail;
383 root = (struct dx_root *) bh->b_data;
384 if (root->info.hash_version != DX_HASH_TEA &&
385 root->info.hash_version != DX_HASH_HALF_MD4 &&
386 root->info.hash_version != DX_HASH_LEGACY) {
387 ext4_warning(dir->i_sb, __func__,
388 "Unrecognised inode hash code %d",
389 root->info.hash_version);
390 brelse(bh);
391 *err = ERR_BAD_DX_DIR;
392 goto fail;
394 hinfo->hash_version = root->info.hash_version;
395 if (hinfo->hash_version <= DX_HASH_TEA)
396 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
397 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
398 if (d_name)
399 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
400 hash = hinfo->hash;
402 if (root->info.unused_flags & 1) {
403 ext4_warning(dir->i_sb, __func__,
404 "Unimplemented inode hash flags: %#06x",
405 root->info.unused_flags);
406 brelse(bh);
407 *err = ERR_BAD_DX_DIR;
408 goto fail;
411 if ((indirect = root->info.indirect_levels) > 1) {
412 ext4_warning(dir->i_sb, __func__,
413 "Unimplemented inode hash depth: %#06x",
414 root->info.indirect_levels);
415 brelse(bh);
416 *err = ERR_BAD_DX_DIR;
417 goto fail;
420 entries = (struct dx_entry *) (((char *)&root->info) +
421 root->info.info_length);
423 if (dx_get_limit(entries) != dx_root_limit(dir,
424 root->info.info_length)) {
425 ext4_warning(dir->i_sb, __func__,
426 "dx entry: limit != root limit");
427 brelse(bh);
428 *err = ERR_BAD_DX_DIR;
429 goto fail;
432 dxtrace(printk("Look up %x", hash));
433 while (1)
435 count = dx_get_count(entries);
436 if (!count || count > dx_get_limit(entries)) {
437 ext4_warning(dir->i_sb, __func__,
438 "dx entry: no count or count > limit");
439 brelse(bh);
440 *err = ERR_BAD_DX_DIR;
441 goto fail2;
444 p = entries + 1;
445 q = entries + count - 1;
446 while (p <= q)
448 m = p + (q - p)/2;
449 dxtrace(printk("."));
450 if (dx_get_hash(m) > hash)
451 q = m - 1;
452 else
453 p = m + 1;
456 if (0) // linear search cross check
458 unsigned n = count - 1;
459 at = entries;
460 while (n--)
462 dxtrace(printk(","));
463 if (dx_get_hash(++at) > hash)
465 at--;
466 break;
469 assert (at == p - 1);
472 at = p - 1;
473 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
474 frame->bh = bh;
475 frame->entries = entries;
476 frame->at = at;
477 if (!indirect--) return frame;
478 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
479 goto fail2;
480 at = entries = ((struct dx_node *) bh->b_data)->entries;
481 if (dx_get_limit(entries) != dx_node_limit (dir)) {
482 ext4_warning(dir->i_sb, __func__,
483 "dx entry: limit != node limit");
484 brelse(bh);
485 *err = ERR_BAD_DX_DIR;
486 goto fail2;
488 frame++;
489 frame->bh = NULL;
491 fail2:
492 while (frame >= frame_in) {
493 brelse(frame->bh);
494 frame--;
496 fail:
497 if (*err == ERR_BAD_DX_DIR)
498 ext4_warning(dir->i_sb, __func__,
499 "Corrupt dir inode %ld, running e2fsck is "
500 "recommended.", dir->i_ino);
501 return NULL;
504 static void dx_release (struct dx_frame *frames)
506 if (frames[0].bh == NULL)
507 return;
509 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
510 brelse(frames[1].bh);
511 brelse(frames[0].bh);
515 * This function increments the frame pointer to search the next leaf
516 * block, and reads in the necessary intervening nodes if the search
517 * should be necessary. Whether or not the search is necessary is
518 * controlled by the hash parameter. If the hash value is even, then
519 * the search is only continued if the next block starts with that
520 * hash value. This is used if we are searching for a specific file.
522 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
524 * This function returns 1 if the caller should continue to search,
525 * or 0 if it should not. If there is an error reading one of the
526 * index blocks, it will a negative error code.
528 * If start_hash is non-null, it will be filled in with the starting
529 * hash of the next page.
531 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
532 struct dx_frame *frame,
533 struct dx_frame *frames,
534 __u32 *start_hash)
536 struct dx_frame *p;
537 struct buffer_head *bh;
538 int err, num_frames = 0;
539 __u32 bhash;
541 p = frame;
543 * Find the next leaf page by incrementing the frame pointer.
544 * If we run out of entries in the interior node, loop around and
545 * increment pointer in the parent node. When we break out of
546 * this loop, num_frames indicates the number of interior
547 * nodes need to be read.
549 while (1) {
550 if (++(p->at) < p->entries + dx_get_count(p->entries))
551 break;
552 if (p == frames)
553 return 0;
554 num_frames++;
555 p--;
559 * If the hash is 1, then continue only if the next page has a
560 * continuation hash of any value. This is used for readdir
561 * handling. Otherwise, check to see if the hash matches the
562 * desired contiuation hash. If it doesn't, return since
563 * there's no point to read in the successive index pages.
565 bhash = dx_get_hash(p->at);
566 if (start_hash)
567 *start_hash = bhash;
568 if ((hash & 1) == 0) {
569 if ((bhash & ~1) != hash)
570 return 0;
573 * If the hash is HASH_NB_ALWAYS, we always go to the next
574 * block so no check is necessary
576 while (num_frames--) {
577 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
578 0, &err)))
579 return err; /* Failure */
580 p++;
581 brelse(p->bh);
582 p->bh = bh;
583 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
585 return 1;
590 * This function fills a red-black tree with information from a
591 * directory block. It returns the number directory entries loaded
592 * into the tree. If there is an error it is returned in err.
594 static int htree_dirblock_to_tree(struct file *dir_file,
595 struct inode *dir, ext4_lblk_t block,
596 struct dx_hash_info *hinfo,
597 __u32 start_hash, __u32 start_minor_hash)
599 struct buffer_head *bh;
600 struct ext4_dir_entry_2 *de, *top;
601 int err, count = 0;
603 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
604 (unsigned long)block));
605 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
606 return err;
608 de = (struct ext4_dir_entry_2 *) bh->b_data;
609 top = (struct ext4_dir_entry_2 *) ((char *) de +
610 dir->i_sb->s_blocksize -
611 EXT4_DIR_REC_LEN(0));
612 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
613 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
614 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
615 +((char *)de - bh->b_data))) {
616 /* On error, skip the f_pos to the next block. */
617 dir_file->f_pos = (dir_file->f_pos |
618 (dir->i_sb->s_blocksize - 1)) + 1;
619 brelse(bh);
620 return count;
622 ext4fs_dirhash(de->name, de->name_len, hinfo);
623 if ((hinfo->hash < start_hash) ||
624 ((hinfo->hash == start_hash) &&
625 (hinfo->minor_hash < start_minor_hash)))
626 continue;
627 if (de->inode == 0)
628 continue;
629 if ((err = ext4_htree_store_dirent(dir_file,
630 hinfo->hash, hinfo->minor_hash, de)) != 0) {
631 brelse(bh);
632 return err;
634 count++;
636 brelse(bh);
637 return count;
642 * This function fills a red-black tree with information from a
643 * directory. We start scanning the directory in hash order, starting
644 * at start_hash and start_minor_hash.
646 * This function returns the number of entries inserted into the tree,
647 * or a negative error code.
649 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
650 __u32 start_minor_hash, __u32 *next_hash)
652 struct dx_hash_info hinfo;
653 struct ext4_dir_entry_2 *de;
654 struct dx_frame frames[2], *frame;
655 struct inode *dir;
656 ext4_lblk_t block;
657 int count = 0;
658 int ret, err;
659 __u32 hashval;
661 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
662 start_hash, start_minor_hash));
663 dir = dir_file->f_path.dentry->d_inode;
664 if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
665 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
666 if (hinfo.hash_version <= DX_HASH_TEA)
667 hinfo.hash_version +=
668 EXT4_SB(dir->i_sb)->s_hash_unsigned;
669 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
670 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
671 start_hash, start_minor_hash);
672 *next_hash = ~0;
673 return count;
675 hinfo.hash = start_hash;
676 hinfo.minor_hash = 0;
677 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
678 if (!frame)
679 return err;
681 /* Add '.' and '..' from the htree header */
682 if (!start_hash && !start_minor_hash) {
683 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
684 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
685 goto errout;
686 count++;
688 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
689 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
690 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
691 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
692 goto errout;
693 count++;
696 while (1) {
697 block = dx_get_block(frame->at);
698 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
699 start_hash, start_minor_hash);
700 if (ret < 0) {
701 err = ret;
702 goto errout;
704 count += ret;
705 hashval = ~0;
706 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
707 frame, frames, &hashval);
708 *next_hash = hashval;
709 if (ret < 0) {
710 err = ret;
711 goto errout;
714 * Stop if: (a) there are no more entries, or
715 * (b) we have inserted at least one entry and the
716 * next hash value is not a continuation
718 if ((ret == 0) ||
719 (count && ((hashval & 1) == 0)))
720 break;
722 dx_release(frames);
723 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
724 "next hash: %x\n", count, *next_hash));
725 return count;
726 errout:
727 dx_release(frames);
728 return (err);
733 * Directory block splitting, compacting
737 * Create map of hash values, offsets, and sizes, stored at end of block.
738 * Returns number of entries mapped.
740 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
741 struct dx_hash_info *hinfo,
742 struct dx_map_entry *map_tail)
744 int count = 0;
745 char *base = (char *) de;
746 struct dx_hash_info h = *hinfo;
748 while ((char *) de < base + blocksize) {
749 if (de->name_len && de->inode) {
750 ext4fs_dirhash(de->name, de->name_len, &h);
751 map_tail--;
752 map_tail->hash = h.hash;
753 map_tail->offs = (u16) ((char *) de - base);
754 map_tail->size = le16_to_cpu(de->rec_len);
755 count++;
756 cond_resched();
758 /* XXX: do we need to check rec_len == 0 case? -Chris */
759 de = ext4_next_entry(de, blocksize);
761 return count;
764 /* Sort map by hash value */
765 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
767 struct dx_map_entry *p, *q, *top = map + count - 1;
768 int more;
769 /* Combsort until bubble sort doesn't suck */
770 while (count > 2) {
771 count = count*10/13;
772 if (count - 9 < 2) /* 9, 10 -> 11 */
773 count = 11;
774 for (p = top, q = p - count; q >= map; p--, q--)
775 if (p->hash < q->hash)
776 swap(*p, *q);
778 /* Garden variety bubble sort */
779 do {
780 more = 0;
781 q = top;
782 while (q-- > map) {
783 if (q[1].hash >= q[0].hash)
784 continue;
785 swap(*(q+1), *q);
786 more = 1;
788 } while(more);
791 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
793 struct dx_entry *entries = frame->entries;
794 struct dx_entry *old = frame->at, *new = old + 1;
795 int count = dx_get_count(entries);
797 assert(count < dx_get_limit(entries));
798 assert(old < entries + count);
799 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
800 dx_set_hash(new, hash);
801 dx_set_block(new, block);
802 dx_set_count(entries, count + 1);
805 static void ext4_update_dx_flag(struct inode *inode)
807 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
808 EXT4_FEATURE_COMPAT_DIR_INDEX))
809 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
813 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
815 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
816 * `de != NULL' is guaranteed by caller.
818 static inline int ext4_match (int len, const char * const name,
819 struct ext4_dir_entry_2 * de)
821 if (len != de->name_len)
822 return 0;
823 if (!de->inode)
824 return 0;
825 return !memcmp(name, de->name, len);
829 * Returns 0 if not found, -1 on failure, and 1 on success
831 static inline int search_dirblock(struct buffer_head *bh,
832 struct inode *dir,
833 const struct qstr *d_name,
834 unsigned int offset,
835 struct ext4_dir_entry_2 ** res_dir)
837 struct ext4_dir_entry_2 * de;
838 char * dlimit;
839 int de_len;
840 const char *name = d_name->name;
841 int namelen = d_name->len;
843 de = (struct ext4_dir_entry_2 *) bh->b_data;
844 dlimit = bh->b_data + dir->i_sb->s_blocksize;
845 while ((char *) de < dlimit) {
846 /* this code is executed quadratically often */
847 /* do minimal checking `by hand' */
849 if ((char *) de + namelen <= dlimit &&
850 ext4_match (namelen, name, de)) {
851 /* found a match - just to be sure, do a full check */
852 if (!ext4_check_dir_entry("ext4_find_entry",
853 dir, de, bh, offset))
854 return -1;
855 *res_dir = de;
856 return 1;
858 /* prevent looping on a bad block */
859 de_len = ext4_rec_len_from_disk(de->rec_len,
860 dir->i_sb->s_blocksize);
861 if (de_len <= 0)
862 return -1;
863 offset += de_len;
864 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
866 return 0;
871 * ext4_find_entry()
873 * finds an entry in the specified directory with the wanted name. It
874 * returns the cache buffer in which the entry was found, and the entry
875 * itself (as a parameter - res_dir). It does NOT read the inode of the
876 * entry - you'll have to do that yourself if you want to.
878 * The returned buffer_head has ->b_count elevated. The caller is expected
879 * to brelse() it when appropriate.
881 static struct buffer_head * ext4_find_entry (struct inode *dir,
882 const struct qstr *d_name,
883 struct ext4_dir_entry_2 ** res_dir)
885 struct super_block *sb;
886 struct buffer_head *bh_use[NAMEI_RA_SIZE];
887 struct buffer_head *bh, *ret = NULL;
888 ext4_lblk_t start, block, b;
889 int ra_max = 0; /* Number of bh's in the readahead
890 buffer, bh_use[] */
891 int ra_ptr = 0; /* Current index into readahead
892 buffer */
893 int num = 0;
894 ext4_lblk_t nblocks;
895 int i, err;
896 int namelen;
898 *res_dir = NULL;
899 sb = dir->i_sb;
900 namelen = d_name->len;
901 if (namelen > EXT4_NAME_LEN)
902 return NULL;
903 if (is_dx(dir)) {
904 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
906 * On success, or if the error was file not found,
907 * return. Otherwise, fall back to doing a search the
908 * old fashioned way.
910 if (bh || (err != ERR_BAD_DX_DIR))
911 return bh;
912 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
913 "falling back\n"));
915 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
916 start = EXT4_I(dir)->i_dir_start_lookup;
917 if (start >= nblocks)
918 start = 0;
919 block = start;
920 restart:
921 do {
923 * We deal with the read-ahead logic here.
925 if (ra_ptr >= ra_max) {
926 /* Refill the readahead buffer */
927 ra_ptr = 0;
928 b = block;
929 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
931 * Terminate if we reach the end of the
932 * directory and must wrap, or if our
933 * search has finished at this block.
935 if (b >= nblocks || (num && block == start)) {
936 bh_use[ra_max] = NULL;
937 break;
939 num++;
940 bh = ext4_getblk(NULL, dir, b++, 0, &err);
941 bh_use[ra_max] = bh;
942 if (bh)
943 ll_rw_block(READ_META, 1, &bh);
946 if ((bh = bh_use[ra_ptr++]) == NULL)
947 goto next;
948 wait_on_buffer(bh);
949 if (!buffer_uptodate(bh)) {
950 /* read error, skip block & hope for the best */
951 ext4_error(sb, __func__, "reading directory #%lu "
952 "offset %lu", dir->i_ino,
953 (unsigned long)block);
954 brelse(bh);
955 goto next;
957 i = search_dirblock(bh, dir, d_name,
958 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
959 if (i == 1) {
960 EXT4_I(dir)->i_dir_start_lookup = block;
961 ret = bh;
962 goto cleanup_and_exit;
963 } else {
964 brelse(bh);
965 if (i < 0)
966 goto cleanup_and_exit;
968 next:
969 if (++block >= nblocks)
970 block = 0;
971 } while (block != start);
974 * If the directory has grown while we were searching, then
975 * search the last part of the directory before giving up.
977 block = nblocks;
978 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
979 if (block < nblocks) {
980 start = 0;
981 goto restart;
984 cleanup_and_exit:
985 /* Clean up the read-ahead blocks */
986 for (; ra_ptr < ra_max; ra_ptr++)
987 brelse(bh_use[ra_ptr]);
988 return ret;
991 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
992 struct ext4_dir_entry_2 **res_dir, int *err)
994 struct super_block * sb;
995 struct dx_hash_info hinfo;
996 u32 hash;
997 struct dx_frame frames[2], *frame;
998 struct ext4_dir_entry_2 *de, *top;
999 struct buffer_head *bh;
1000 ext4_lblk_t block;
1001 int retval;
1002 int namelen = d_name->len;
1003 const u8 *name = d_name->name;
1005 sb = dir->i_sb;
1006 /* NFS may look up ".." - look at dx_root directory block */
1007 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
1008 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1009 return NULL;
1010 } else {
1011 frame = frames;
1012 frame->bh = NULL; /* for dx_release() */
1013 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
1014 dx_set_block(frame->at, 0); /* dx_root block is 0 */
1016 hash = hinfo.hash;
1017 do {
1018 block = dx_get_block(frame->at);
1019 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
1020 goto errout;
1021 de = (struct ext4_dir_entry_2 *) bh->b_data;
1022 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
1023 EXT4_DIR_REC_LEN(0));
1024 for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) {
1025 int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
1026 + ((char *) de - bh->b_data);
1028 if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) {
1029 brelse(bh);
1030 *err = ERR_BAD_DX_DIR;
1031 goto errout;
1034 if (ext4_match(namelen, name, de)) {
1035 *res_dir = de;
1036 dx_release(frames);
1037 return bh;
1040 brelse(bh);
1041 /* Check to see if we should continue to search */
1042 retval = ext4_htree_next_block(dir, hash, frame,
1043 frames, NULL);
1044 if (retval < 0) {
1045 ext4_warning(sb, __func__,
1046 "error reading index page in directory #%lu",
1047 dir->i_ino);
1048 *err = retval;
1049 goto errout;
1051 } while (retval == 1);
1053 *err = -ENOENT;
1054 errout:
1055 dxtrace(printk(KERN_DEBUG "%s not found\n", name));
1056 dx_release (frames);
1057 return NULL;
1060 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1062 struct inode *inode;
1063 struct ext4_dir_entry_2 *de;
1064 struct buffer_head *bh;
1066 if (dentry->d_name.len > EXT4_NAME_LEN)
1067 return ERR_PTR(-ENAMETOOLONG);
1069 bh = ext4_find_entry(dir, &dentry->d_name, &de);
1070 inode = NULL;
1071 if (bh) {
1072 __u32 ino = le32_to_cpu(de->inode);
1073 brelse(bh);
1074 if (!ext4_valid_inum(dir->i_sb, ino)) {
1075 ext4_error(dir->i_sb, "ext4_lookup",
1076 "bad inode number: %u", ino);
1077 return ERR_PTR(-EIO);
1079 inode = ext4_iget(dir->i_sb, ino);
1080 if (unlikely(IS_ERR(inode))) {
1081 if (PTR_ERR(inode) == -ESTALE) {
1082 ext4_error(dir->i_sb, __func__,
1083 "deleted inode referenced: %u",
1084 ino);
1085 return ERR_PTR(-EIO);
1086 } else {
1087 return ERR_CAST(inode);
1091 return d_splice_alias(inode, dentry);
1095 struct dentry *ext4_get_parent(struct dentry *child)
1097 __u32 ino;
1098 struct inode *inode;
1099 static const struct qstr dotdot = {
1100 .name = "..",
1101 .len = 2,
1103 struct ext4_dir_entry_2 * de;
1104 struct buffer_head *bh;
1106 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1107 inode = NULL;
1108 if (!bh)
1109 return ERR_PTR(-ENOENT);
1110 ino = le32_to_cpu(de->inode);
1111 brelse(bh);
1113 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1114 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1115 "bad inode number: %u", ino);
1116 return ERR_PTR(-EIO);
1119 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1122 #define S_SHIFT 12
1123 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1124 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1125 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1126 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1127 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1128 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1129 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1130 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
1133 static inline void ext4_set_de_type(struct super_block *sb,
1134 struct ext4_dir_entry_2 *de,
1135 umode_t mode) {
1136 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1137 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1141 * Move count entries from end of map between two memory locations.
1142 * Returns pointer to last entry moved.
1144 static struct ext4_dir_entry_2 *
1145 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1146 unsigned blocksize)
1148 unsigned rec_len = 0;
1150 while (count--) {
1151 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1152 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1153 memcpy (to, de, rec_len);
1154 ((struct ext4_dir_entry_2 *) to)->rec_len =
1155 ext4_rec_len_to_disk(rec_len, blocksize);
1156 de->inode = 0;
1157 map++;
1158 to += rec_len;
1160 return (struct ext4_dir_entry_2 *) (to - rec_len);
1164 * Compact each dir entry in the range to the minimal rec_len.
1165 * Returns pointer to last entry in range.
1167 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1169 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1170 unsigned rec_len = 0;
1172 prev = to = de;
1173 while ((char*)de < base + blocksize) {
1174 next = ext4_next_entry(de, blocksize);
1175 if (de->inode && de->name_len) {
1176 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1177 if (de > to)
1178 memmove(to, de, rec_len);
1179 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1180 prev = to;
1181 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1183 de = next;
1185 return prev;
1189 * Split a full leaf block to make room for a new dir entry.
1190 * Allocate a new block, and move entries so that they are approx. equally full.
1191 * Returns pointer to de in block into which the new entry will be inserted.
1193 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1194 struct buffer_head **bh,struct dx_frame *frame,
1195 struct dx_hash_info *hinfo, int *error)
1197 unsigned blocksize = dir->i_sb->s_blocksize;
1198 unsigned count, continued;
1199 struct buffer_head *bh2;
1200 ext4_lblk_t newblock;
1201 u32 hash2;
1202 struct dx_map_entry *map;
1203 char *data1 = (*bh)->b_data, *data2;
1204 unsigned split, move, size;
1205 struct ext4_dir_entry_2 *de = NULL, *de2;
1206 int err = 0, i;
1208 bh2 = ext4_append (handle, dir, &newblock, &err);
1209 if (!(bh2)) {
1210 brelse(*bh);
1211 *bh = NULL;
1212 goto errout;
1215 BUFFER_TRACE(*bh, "get_write_access");
1216 err = ext4_journal_get_write_access(handle, *bh);
1217 if (err)
1218 goto journal_error;
1220 BUFFER_TRACE(frame->bh, "get_write_access");
1221 err = ext4_journal_get_write_access(handle, frame->bh);
1222 if (err)
1223 goto journal_error;
1225 data2 = bh2->b_data;
1227 /* create map in the end of data2 block */
1228 map = (struct dx_map_entry *) (data2 + blocksize);
1229 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1230 blocksize, hinfo, map);
1231 map -= count;
1232 dx_sort_map(map, count);
1233 /* Split the existing block in the middle, size-wise */
1234 size = 0;
1235 move = 0;
1236 for (i = count-1; i >= 0; i--) {
1237 /* is more than half of this entry in 2nd half of the block? */
1238 if (size + map[i].size/2 > blocksize/2)
1239 break;
1240 size += map[i].size;
1241 move++;
1243 /* map index at which we will split */
1244 split = count - move;
1245 hash2 = map[split].hash;
1246 continued = hash2 == map[split - 1].hash;
1247 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1248 (unsigned long)dx_get_block(frame->at),
1249 hash2, split, count-split));
1251 /* Fancy dance to stay within two buffers */
1252 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1253 de = dx_pack_dirents(data1, blocksize);
1254 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1255 blocksize);
1256 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1257 blocksize);
1258 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1259 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1261 /* Which block gets the new entry? */
1262 if (hinfo->hash >= hash2)
1264 swap(*bh, bh2);
1265 de = de2;
1267 dx_insert_block(frame, hash2 + continued, newblock);
1268 err = ext4_handle_dirty_metadata(handle, dir, bh2);
1269 if (err)
1270 goto journal_error;
1271 err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1272 if (err)
1273 goto journal_error;
1274 brelse(bh2);
1275 dxtrace(dx_show_index("frame", frame->entries));
1276 return de;
1278 journal_error:
1279 brelse(*bh);
1280 brelse(bh2);
1281 *bh = NULL;
1282 ext4_std_error(dir->i_sb, err);
1283 errout:
1284 *error = err;
1285 return NULL;
1289 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1290 * it points to a directory entry which is guaranteed to be large
1291 * enough for new directory entry. If de is NULL, then
1292 * add_dirent_to_buf will attempt search the directory block for
1293 * space. It will return -ENOSPC if no space is available, and -EIO
1294 * and -EEXIST if directory entry already exists.
1296 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1297 * all other cases bh is released.
1299 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1300 struct inode *inode, struct ext4_dir_entry_2 *de,
1301 struct buffer_head *bh)
1303 struct inode *dir = dentry->d_parent->d_inode;
1304 const char *name = dentry->d_name.name;
1305 int namelen = dentry->d_name.len;
1306 unsigned int offset = 0;
1307 unsigned int blocksize = dir->i_sb->s_blocksize;
1308 unsigned short reclen;
1309 int nlen, rlen, err;
1310 char *top;
1312 reclen = EXT4_DIR_REC_LEN(namelen);
1313 if (!de) {
1314 de = (struct ext4_dir_entry_2 *)bh->b_data;
1315 top = bh->b_data + blocksize - reclen;
1316 while ((char *) de <= top) {
1317 if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1318 bh, offset)) {
1319 brelse(bh);
1320 return -EIO;
1322 if (ext4_match(namelen, name, de)) {
1323 brelse(bh);
1324 return -EEXIST;
1326 nlen = EXT4_DIR_REC_LEN(de->name_len);
1327 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1328 if ((de->inode? rlen - nlen: rlen) >= reclen)
1329 break;
1330 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1331 offset += rlen;
1333 if ((char *) de > top)
1334 return -ENOSPC;
1336 BUFFER_TRACE(bh, "get_write_access");
1337 err = ext4_journal_get_write_access(handle, bh);
1338 if (err) {
1339 ext4_std_error(dir->i_sb, err);
1340 brelse(bh);
1341 return err;
1344 /* By now the buffer is marked for journaling */
1345 nlen = EXT4_DIR_REC_LEN(de->name_len);
1346 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1347 if (de->inode) {
1348 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1349 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1350 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1351 de = de1;
1353 de->file_type = EXT4_FT_UNKNOWN;
1354 if (inode) {
1355 de->inode = cpu_to_le32(inode->i_ino);
1356 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1357 } else
1358 de->inode = 0;
1359 de->name_len = namelen;
1360 memcpy(de->name, name, namelen);
1362 * XXX shouldn't update any times until successful
1363 * completion of syscall, but too many callers depend
1364 * on this.
1366 * XXX similarly, too many callers depend on
1367 * ext4_new_inode() setting the times, but error
1368 * recovery deletes the inode, so the worst that can
1369 * happen is that the times are slightly out of date
1370 * and/or different from the directory change time.
1372 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1373 ext4_update_dx_flag(dir);
1374 dir->i_version++;
1375 ext4_mark_inode_dirty(handle, dir);
1376 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1377 err = ext4_handle_dirty_metadata(handle, dir, bh);
1378 if (err)
1379 ext4_std_error(dir->i_sb, err);
1380 brelse(bh);
1381 return 0;
1385 * This converts a one block unindexed directory to a 3 block indexed
1386 * directory, and adds the dentry to the indexed directory.
1388 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1389 struct inode *inode, struct buffer_head *bh)
1391 struct inode *dir = dentry->d_parent->d_inode;
1392 const char *name = dentry->d_name.name;
1393 int namelen = dentry->d_name.len;
1394 struct buffer_head *bh2;
1395 struct dx_root *root;
1396 struct dx_frame frames[2], *frame;
1397 struct dx_entry *entries;
1398 struct ext4_dir_entry_2 *de, *de2;
1399 char *data1, *top;
1400 unsigned len;
1401 int retval;
1402 unsigned blocksize;
1403 struct dx_hash_info hinfo;
1404 ext4_lblk_t block;
1405 struct fake_dirent *fde;
1407 blocksize = dir->i_sb->s_blocksize;
1408 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1409 retval = ext4_journal_get_write_access(handle, bh);
1410 if (retval) {
1411 ext4_std_error(dir->i_sb, retval);
1412 brelse(bh);
1413 return retval;
1415 root = (struct dx_root *) bh->b_data;
1417 /* The 0th block becomes the root, move the dirents out */
1418 fde = &root->dotdot;
1419 de = (struct ext4_dir_entry_2 *)((char *)fde +
1420 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1421 if ((char *) de >= (((char *) root) + blocksize)) {
1422 ext4_error(dir->i_sb, __func__,
1423 "invalid rec_len for '..' in inode %lu",
1424 dir->i_ino);
1425 brelse(bh);
1426 return -EIO;
1428 len = ((char *) root) + blocksize - (char *) de;
1430 /* Allocate new block for the 0th block's dirents */
1431 bh2 = ext4_append(handle, dir, &block, &retval);
1432 if (!(bh2)) {
1433 brelse(bh);
1434 return retval;
1436 EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1437 data1 = bh2->b_data;
1439 memcpy (data1, de, len);
1440 de = (struct ext4_dir_entry_2 *) data1;
1441 top = data1 + len;
1442 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1443 de = de2;
1444 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1445 blocksize);
1446 /* Initialize the root; the dot dirents already exist */
1447 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1448 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1449 blocksize);
1450 memset (&root->info, 0, sizeof(root->info));
1451 root->info.info_length = sizeof(root->info);
1452 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1453 entries = root->entries;
1454 dx_set_block(entries, 1);
1455 dx_set_count(entries, 1);
1456 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1458 /* Initialize as for dx_probe */
1459 hinfo.hash_version = root->info.hash_version;
1460 if (hinfo.hash_version <= DX_HASH_TEA)
1461 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1462 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1463 ext4fs_dirhash(name, namelen, &hinfo);
1464 frame = frames;
1465 frame->entries = entries;
1466 frame->at = entries;
1467 frame->bh = bh;
1468 bh = bh2;
1469 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1470 dx_release (frames);
1471 if (!(de))
1472 return retval;
1474 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1478 * ext4_add_entry()
1480 * adds a file entry to the specified directory, using the same
1481 * semantics as ext4_find_entry(). It returns NULL if it failed.
1483 * NOTE!! The inode part of 'de' is left at 0 - which means you
1484 * may not sleep between calling this and putting something into
1485 * the entry, as someone else might have used it while you slept.
1487 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1488 struct inode *inode)
1490 struct inode *dir = dentry->d_parent->d_inode;
1491 struct buffer_head *bh;
1492 struct ext4_dir_entry_2 *de;
1493 struct super_block *sb;
1494 int retval;
1495 int dx_fallback=0;
1496 unsigned blocksize;
1497 ext4_lblk_t block, blocks;
1499 sb = dir->i_sb;
1500 blocksize = sb->s_blocksize;
1501 if (!dentry->d_name.len)
1502 return -EINVAL;
1503 if (is_dx(dir)) {
1504 retval = ext4_dx_add_entry(handle, dentry, inode);
1505 if (!retval || (retval != ERR_BAD_DX_DIR))
1506 return retval;
1507 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1508 dx_fallback++;
1509 ext4_mark_inode_dirty(handle, dir);
1511 blocks = dir->i_size >> sb->s_blocksize_bits;
1512 for (block = 0; block < blocks; block++) {
1513 bh = ext4_bread(handle, dir, block, 0, &retval);
1514 if(!bh)
1515 return retval;
1516 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1517 if (retval != -ENOSPC)
1518 return retval;
1520 if (blocks == 1 && !dx_fallback &&
1521 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1522 return make_indexed_dir(handle, dentry, inode, bh);
1523 brelse(bh);
1525 bh = ext4_append(handle, dir, &block, &retval);
1526 if (!bh)
1527 return retval;
1528 de = (struct ext4_dir_entry_2 *) bh->b_data;
1529 de->inode = 0;
1530 de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1531 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1535 * Returns 0 for success, or a negative error value
1537 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1538 struct inode *inode)
1540 struct dx_frame frames[2], *frame;
1541 struct dx_entry *entries, *at;
1542 struct dx_hash_info hinfo;
1543 struct buffer_head *bh;
1544 struct inode *dir = dentry->d_parent->d_inode;
1545 struct super_block *sb = dir->i_sb;
1546 struct ext4_dir_entry_2 *de;
1547 int err;
1549 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1550 if (!frame)
1551 return err;
1552 entries = frame->entries;
1553 at = frame->at;
1555 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1556 goto cleanup;
1558 BUFFER_TRACE(bh, "get_write_access");
1559 err = ext4_journal_get_write_access(handle, bh);
1560 if (err)
1561 goto journal_error;
1563 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1564 if (err != -ENOSPC) {
1565 bh = NULL;
1566 goto cleanup;
1569 /* Block full, should compress but for now just split */
1570 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1571 dx_get_count(entries), dx_get_limit(entries)));
1572 /* Need to split index? */
1573 if (dx_get_count(entries) == dx_get_limit(entries)) {
1574 ext4_lblk_t newblock;
1575 unsigned icount = dx_get_count(entries);
1576 int levels = frame - frames;
1577 struct dx_entry *entries2;
1578 struct dx_node *node2;
1579 struct buffer_head *bh2;
1581 if (levels && (dx_get_count(frames->entries) ==
1582 dx_get_limit(frames->entries))) {
1583 ext4_warning(sb, __func__,
1584 "Directory index full!");
1585 err = -ENOSPC;
1586 goto cleanup;
1588 bh2 = ext4_append (handle, dir, &newblock, &err);
1589 if (!(bh2))
1590 goto cleanup;
1591 node2 = (struct dx_node *)(bh2->b_data);
1592 entries2 = node2->entries;
1593 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1594 sb->s_blocksize);
1595 node2->fake.inode = 0;
1596 BUFFER_TRACE(frame->bh, "get_write_access");
1597 err = ext4_journal_get_write_access(handle, frame->bh);
1598 if (err)
1599 goto journal_error;
1600 if (levels) {
1601 unsigned icount1 = icount/2, icount2 = icount - icount1;
1602 unsigned hash2 = dx_get_hash(entries + icount1);
1603 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1604 icount1, icount2));
1606 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1607 err = ext4_journal_get_write_access(handle,
1608 frames[0].bh);
1609 if (err)
1610 goto journal_error;
1612 memcpy((char *) entries2, (char *) (entries + icount1),
1613 icount2 * sizeof(struct dx_entry));
1614 dx_set_count(entries, icount1);
1615 dx_set_count(entries2, icount2);
1616 dx_set_limit(entries2, dx_node_limit(dir));
1618 /* Which index block gets the new entry? */
1619 if (at - entries >= icount1) {
1620 frame->at = at = at - entries - icount1 + entries2;
1621 frame->entries = entries = entries2;
1622 swap(frame->bh, bh2);
1624 dx_insert_block(frames + 0, hash2, newblock);
1625 dxtrace(dx_show_index("node", frames[1].entries));
1626 dxtrace(dx_show_index("node",
1627 ((struct dx_node *) bh2->b_data)->entries));
1628 err = ext4_handle_dirty_metadata(handle, inode, bh2);
1629 if (err)
1630 goto journal_error;
1631 brelse (bh2);
1632 } else {
1633 dxtrace(printk(KERN_DEBUG
1634 "Creating second level index...\n"));
1635 memcpy((char *) entries2, (char *) entries,
1636 icount * sizeof(struct dx_entry));
1637 dx_set_limit(entries2, dx_node_limit(dir));
1639 /* Set up root */
1640 dx_set_count(entries, 1);
1641 dx_set_block(entries + 0, newblock);
1642 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1644 /* Add new access path frame */
1645 frame = frames + 1;
1646 frame->at = at = at - entries + entries2;
1647 frame->entries = entries = entries2;
1648 frame->bh = bh2;
1649 err = ext4_journal_get_write_access(handle,
1650 frame->bh);
1651 if (err)
1652 goto journal_error;
1654 ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
1656 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1657 if (!de)
1658 goto cleanup;
1659 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1660 bh = NULL;
1661 goto cleanup;
1663 journal_error:
1664 ext4_std_error(dir->i_sb, err);
1665 cleanup:
1666 if (bh)
1667 brelse(bh);
1668 dx_release(frames);
1669 return err;
1673 * ext4_delete_entry deletes a directory entry by merging it with the
1674 * previous entry
1676 static int ext4_delete_entry(handle_t *handle,
1677 struct inode *dir,
1678 struct ext4_dir_entry_2 *de_del,
1679 struct buffer_head *bh)
1681 struct ext4_dir_entry_2 *de, *pde;
1682 unsigned int blocksize = dir->i_sb->s_blocksize;
1683 int i;
1685 i = 0;
1686 pde = NULL;
1687 de = (struct ext4_dir_entry_2 *) bh->b_data;
1688 while (i < bh->b_size) {
1689 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1690 return -EIO;
1691 if (de == de_del) {
1692 BUFFER_TRACE(bh, "get_write_access");
1693 ext4_journal_get_write_access(handle, bh);
1694 if (pde)
1695 pde->rec_len = ext4_rec_len_to_disk(
1696 ext4_rec_len_from_disk(pde->rec_len,
1697 blocksize) +
1698 ext4_rec_len_from_disk(de->rec_len,
1699 blocksize),
1700 blocksize);
1701 else
1702 de->inode = 0;
1703 dir->i_version++;
1704 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1705 ext4_handle_dirty_metadata(handle, dir, bh);
1706 return 0;
1708 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
1709 pde = de;
1710 de = ext4_next_entry(de, blocksize);
1712 return -ENOENT;
1716 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1717 * since this indicates that nlinks count was previously 1.
1719 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1721 inc_nlink(inode);
1722 if (is_dx(inode) && inode->i_nlink > 1) {
1723 /* limit is 16-bit i_links_count */
1724 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1725 inode->i_nlink = 1;
1726 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1727 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1733 * If a directory had nlink == 1, then we should let it be 1. This indicates
1734 * directory has >EXT4_LINK_MAX subdirs.
1736 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1738 drop_nlink(inode);
1739 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1740 inc_nlink(inode);
1744 static int ext4_add_nondir(handle_t *handle,
1745 struct dentry *dentry, struct inode *inode)
1747 int err = ext4_add_entry(handle, dentry, inode);
1748 if (!err) {
1749 ext4_mark_inode_dirty(handle, inode);
1750 d_instantiate(dentry, inode);
1751 unlock_new_inode(inode);
1752 return 0;
1754 drop_nlink(inode);
1755 unlock_new_inode(inode);
1756 iput(inode);
1757 return err;
1761 * By the time this is called, we already have created
1762 * the directory cache entry for the new file, but it
1763 * is so far negative - it has no inode.
1765 * If the create succeeds, we fill in the inode information
1766 * with d_instantiate().
1768 static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1769 struct nameidata *nd)
1771 handle_t *handle;
1772 struct inode *inode;
1773 int err, retries = 0;
1775 retry:
1776 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1777 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1778 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1779 if (IS_ERR(handle))
1780 return PTR_ERR(handle);
1782 if (IS_DIRSYNC(dir))
1783 ext4_handle_sync(handle);
1785 inode = ext4_new_inode (handle, dir, mode);
1786 err = PTR_ERR(inode);
1787 if (!IS_ERR(inode)) {
1788 inode->i_op = &ext4_file_inode_operations;
1789 inode->i_fop = &ext4_file_operations;
1790 ext4_set_aops(inode);
1791 err = ext4_add_nondir(handle, dentry, inode);
1793 ext4_journal_stop(handle);
1794 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1795 goto retry;
1796 return err;
1799 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1800 int mode, dev_t rdev)
1802 handle_t *handle;
1803 struct inode *inode;
1804 int err, retries = 0;
1806 if (!new_valid_dev(rdev))
1807 return -EINVAL;
1809 retry:
1810 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1811 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1812 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1813 if (IS_ERR(handle))
1814 return PTR_ERR(handle);
1816 if (IS_DIRSYNC(dir))
1817 ext4_handle_sync(handle);
1819 inode = ext4_new_inode(handle, dir, mode);
1820 err = PTR_ERR(inode);
1821 if (!IS_ERR(inode)) {
1822 init_special_inode(inode, inode->i_mode, rdev);
1823 #ifdef CONFIG_EXT4_FS_XATTR
1824 inode->i_op = &ext4_special_inode_operations;
1825 #endif
1826 err = ext4_add_nondir(handle, dentry, inode);
1828 ext4_journal_stop(handle);
1829 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1830 goto retry;
1831 return err;
1834 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1836 handle_t *handle;
1837 struct inode *inode;
1838 struct buffer_head *dir_block;
1839 struct ext4_dir_entry_2 *de;
1840 unsigned int blocksize = dir->i_sb->s_blocksize;
1841 int err, retries = 0;
1843 if (EXT4_DIR_LINK_MAX(dir))
1844 return -EMLINK;
1846 retry:
1847 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1848 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1849 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1850 if (IS_ERR(handle))
1851 return PTR_ERR(handle);
1853 if (IS_DIRSYNC(dir))
1854 ext4_handle_sync(handle);
1856 inode = ext4_new_inode(handle, dir, S_IFDIR | mode);
1857 err = PTR_ERR(inode);
1858 if (IS_ERR(inode))
1859 goto out_stop;
1861 inode->i_op = &ext4_dir_inode_operations;
1862 inode->i_fop = &ext4_dir_operations;
1863 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1864 dir_block = ext4_bread(handle, inode, 0, 1, &err);
1865 if (!dir_block)
1866 goto out_clear_inode;
1867 BUFFER_TRACE(dir_block, "get_write_access");
1868 ext4_journal_get_write_access(handle, dir_block);
1869 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1870 de->inode = cpu_to_le32(inode->i_ino);
1871 de->name_len = 1;
1872 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1873 blocksize);
1874 strcpy(de->name, ".");
1875 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1876 de = ext4_next_entry(de, blocksize);
1877 de->inode = cpu_to_le32(dir->i_ino);
1878 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1879 blocksize);
1880 de->name_len = 2;
1881 strcpy(de->name, "..");
1882 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1883 inode->i_nlink = 2;
1884 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1885 ext4_handle_dirty_metadata(handle, dir, dir_block);
1886 brelse(dir_block);
1887 ext4_mark_inode_dirty(handle, inode);
1888 err = ext4_add_entry(handle, dentry, inode);
1889 if (err) {
1890 out_clear_inode:
1891 clear_nlink(inode);
1892 unlock_new_inode(inode);
1893 ext4_mark_inode_dirty(handle, inode);
1894 iput(inode);
1895 goto out_stop;
1897 ext4_inc_count(handle, dir);
1898 ext4_update_dx_flag(dir);
1899 ext4_mark_inode_dirty(handle, dir);
1900 d_instantiate(dentry, inode);
1901 unlock_new_inode(inode);
1902 out_stop:
1903 ext4_journal_stop(handle);
1904 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1905 goto retry;
1906 return err;
1910 * routine to check that the specified directory is empty (for rmdir)
1912 static int empty_dir(struct inode *inode)
1914 unsigned int offset;
1915 struct buffer_head *bh;
1916 struct ext4_dir_entry_2 *de, *de1;
1917 struct super_block *sb;
1918 int err = 0;
1920 sb = inode->i_sb;
1921 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1922 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1923 if (err)
1924 ext4_error(inode->i_sb, __func__,
1925 "error %d reading directory #%lu offset 0",
1926 err, inode->i_ino);
1927 else
1928 ext4_warning(inode->i_sb, __func__,
1929 "bad directory (dir #%lu) - no data block",
1930 inode->i_ino);
1931 return 1;
1933 de = (struct ext4_dir_entry_2 *) bh->b_data;
1934 de1 = ext4_next_entry(de, sb->s_blocksize);
1935 if (le32_to_cpu(de->inode) != inode->i_ino ||
1936 !le32_to_cpu(de1->inode) ||
1937 strcmp(".", de->name) ||
1938 strcmp("..", de1->name)) {
1939 ext4_warning(inode->i_sb, "empty_dir",
1940 "bad directory (dir #%lu) - no `.' or `..'",
1941 inode->i_ino);
1942 brelse(bh);
1943 return 1;
1945 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1946 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1947 de = ext4_next_entry(de1, sb->s_blocksize);
1948 while (offset < inode->i_size) {
1949 if (!bh ||
1950 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1951 err = 0;
1952 brelse(bh);
1953 bh = ext4_bread(NULL, inode,
1954 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1955 if (!bh) {
1956 if (err)
1957 ext4_error(sb, __func__,
1958 "error %d reading directory"
1959 " #%lu offset %u",
1960 err, inode->i_ino, offset);
1961 offset += sb->s_blocksize;
1962 continue;
1964 de = (struct ext4_dir_entry_2 *) bh->b_data;
1966 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1967 de = (struct ext4_dir_entry_2 *)(bh->b_data +
1968 sb->s_blocksize);
1969 offset = (offset | (sb->s_blocksize - 1)) + 1;
1970 continue;
1972 if (le32_to_cpu(de->inode)) {
1973 brelse(bh);
1974 return 0;
1976 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1977 de = ext4_next_entry(de, sb->s_blocksize);
1979 brelse(bh);
1980 return 1;
1983 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1984 * such inodes, starting at the superblock, in case we crash before the
1985 * file is closed/deleted, or in case the inode truncate spans multiple
1986 * transactions and the last transaction is not recovered after a crash.
1988 * At filesystem recovery time, we walk this list deleting unlinked
1989 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1991 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1993 struct super_block *sb = inode->i_sb;
1994 struct ext4_iloc iloc;
1995 int err = 0, rc;
1997 if (!ext4_handle_valid(handle))
1998 return 0;
2000 lock_super(sb);
2001 if (!list_empty(&EXT4_I(inode)->i_orphan))
2002 goto out_unlock;
2004 /* Orphan handling is only valid for files with data blocks
2005 * being truncated, or files being unlinked. */
2007 /* @@@ FIXME: Observation from aviro:
2008 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
2009 * here (on lock_super()), so race with ext4_link() which might bump
2010 * ->i_nlink. For, say it, character device. Not a regular file,
2011 * not a directory, not a symlink and ->i_nlink > 0.
2013 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2014 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2016 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2017 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2018 if (err)
2019 goto out_unlock;
2021 err = ext4_reserve_inode_write(handle, inode, &iloc);
2022 if (err)
2023 goto out_unlock;
2025 /* Insert this inode at the head of the on-disk orphan list... */
2026 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2027 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2028 err = ext4_handle_dirty_metadata(handle, inode, EXT4_SB(sb)->s_sbh);
2029 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2030 if (!err)
2031 err = rc;
2033 /* Only add to the head of the in-memory list if all the
2034 * previous operations succeeded. If the orphan_add is going to
2035 * fail (possibly taking the journal offline), we can't risk
2036 * leaving the inode on the orphan list: stray orphan-list
2037 * entries can cause panics at unmount time.
2039 * This is safe: on error we're going to ignore the orphan list
2040 * anyway on the next recovery. */
2041 if (!err)
2042 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2044 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2045 jbd_debug(4, "orphan inode %lu will point to %d\n",
2046 inode->i_ino, NEXT_ORPHAN(inode));
2047 out_unlock:
2048 unlock_super(sb);
2049 ext4_std_error(inode->i_sb, err);
2050 return err;
2054 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2055 * of such inodes stored on disk, because it is finally being cleaned up.
2057 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2059 struct list_head *prev;
2060 struct ext4_inode_info *ei = EXT4_I(inode);
2061 struct ext4_sb_info *sbi;
2062 __u32 ino_next;
2063 struct ext4_iloc iloc;
2064 int err = 0;
2066 if (!ext4_handle_valid(handle))
2067 return 0;
2069 lock_super(inode->i_sb);
2070 if (list_empty(&ei->i_orphan)) {
2071 unlock_super(inode->i_sb);
2072 return 0;
2075 ino_next = NEXT_ORPHAN(inode);
2076 prev = ei->i_orphan.prev;
2077 sbi = EXT4_SB(inode->i_sb);
2079 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2081 list_del_init(&ei->i_orphan);
2083 /* If we're on an error path, we may not have a valid
2084 * transaction handle with which to update the orphan list on
2085 * disk, but we still need to remove the inode from the linked
2086 * list in memory. */
2087 if (sbi->s_journal && !handle)
2088 goto out;
2090 err = ext4_reserve_inode_write(handle, inode, &iloc);
2091 if (err)
2092 goto out_err;
2094 if (prev == &sbi->s_orphan) {
2095 jbd_debug(4, "superblock will point to %u\n", ino_next);
2096 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2097 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2098 if (err)
2099 goto out_brelse;
2100 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2101 err = ext4_handle_dirty_metadata(handle, inode, sbi->s_sbh);
2102 } else {
2103 struct ext4_iloc iloc2;
2104 struct inode *i_prev =
2105 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2107 jbd_debug(4, "orphan inode %lu will point to %u\n",
2108 i_prev->i_ino, ino_next);
2109 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2110 if (err)
2111 goto out_brelse;
2112 NEXT_ORPHAN(i_prev) = ino_next;
2113 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2115 if (err)
2116 goto out_brelse;
2117 NEXT_ORPHAN(inode) = 0;
2118 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2120 out_err:
2121 ext4_std_error(inode->i_sb, err);
2122 out:
2123 unlock_super(inode->i_sb);
2124 return err;
2126 out_brelse:
2127 brelse(iloc.bh);
2128 goto out_err;
2131 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2133 int retval;
2134 struct inode *inode;
2135 struct buffer_head *bh;
2136 struct ext4_dir_entry_2 *de;
2137 handle_t *handle;
2139 /* Initialize quotas before so that eventual writes go in
2140 * separate transaction */
2141 vfs_dq_init(dentry->d_inode);
2142 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2143 if (IS_ERR(handle))
2144 return PTR_ERR(handle);
2146 retval = -ENOENT;
2147 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2148 if (!bh)
2149 goto end_rmdir;
2151 if (IS_DIRSYNC(dir))
2152 ext4_handle_sync(handle);
2154 inode = dentry->d_inode;
2156 retval = -EIO;
2157 if (le32_to_cpu(de->inode) != inode->i_ino)
2158 goto end_rmdir;
2160 retval = -ENOTEMPTY;
2161 if (!empty_dir(inode))
2162 goto end_rmdir;
2164 retval = ext4_delete_entry(handle, dir, de, bh);
2165 if (retval)
2166 goto end_rmdir;
2167 if (!EXT4_DIR_LINK_EMPTY(inode))
2168 ext4_warning(inode->i_sb, "ext4_rmdir",
2169 "empty directory has too many links (%d)",
2170 inode->i_nlink);
2171 inode->i_version++;
2172 clear_nlink(inode);
2173 /* There's no need to set i_disksize: the fact that i_nlink is
2174 * zero will ensure that the right thing happens during any
2175 * recovery. */
2176 inode->i_size = 0;
2177 ext4_orphan_add(handle, inode);
2178 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2179 ext4_mark_inode_dirty(handle, inode);
2180 ext4_dec_count(handle, dir);
2181 ext4_update_dx_flag(dir);
2182 ext4_mark_inode_dirty(handle, dir);
2184 end_rmdir:
2185 ext4_journal_stop(handle);
2186 brelse(bh);
2187 return retval;
2190 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2192 int retval;
2193 struct inode *inode;
2194 struct buffer_head *bh;
2195 struct ext4_dir_entry_2 *de;
2196 handle_t *handle;
2198 /* Initialize quotas before so that eventual writes go
2199 * in separate transaction */
2200 vfs_dq_init(dentry->d_inode);
2201 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2202 if (IS_ERR(handle))
2203 return PTR_ERR(handle);
2205 if (IS_DIRSYNC(dir))
2206 ext4_handle_sync(handle);
2208 retval = -ENOENT;
2209 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2210 if (!bh)
2211 goto end_unlink;
2213 inode = dentry->d_inode;
2215 retval = -EIO;
2216 if (le32_to_cpu(de->inode) != inode->i_ino)
2217 goto end_unlink;
2219 if (!inode->i_nlink) {
2220 ext4_warning(inode->i_sb, "ext4_unlink",
2221 "Deleting nonexistent file (%lu), %d",
2222 inode->i_ino, inode->i_nlink);
2223 inode->i_nlink = 1;
2225 retval = ext4_delete_entry(handle, dir, de, bh);
2226 if (retval)
2227 goto end_unlink;
2228 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2229 ext4_update_dx_flag(dir);
2230 ext4_mark_inode_dirty(handle, dir);
2231 drop_nlink(inode);
2232 if (!inode->i_nlink)
2233 ext4_orphan_add(handle, inode);
2234 inode->i_ctime = ext4_current_time(inode);
2235 ext4_mark_inode_dirty(handle, inode);
2236 retval = 0;
2238 end_unlink:
2239 ext4_journal_stop(handle);
2240 brelse(bh);
2241 return retval;
2244 static int ext4_symlink(struct inode *dir,
2245 struct dentry *dentry, const char *symname)
2247 handle_t *handle;
2248 struct inode *inode;
2249 int l, err, retries = 0;
2251 l = strlen(symname)+1;
2252 if (l > dir->i_sb->s_blocksize)
2253 return -ENAMETOOLONG;
2255 retry:
2256 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2257 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2258 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2259 if (IS_ERR(handle))
2260 return PTR_ERR(handle);
2262 if (IS_DIRSYNC(dir))
2263 ext4_handle_sync(handle);
2265 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO);
2266 err = PTR_ERR(inode);
2267 if (IS_ERR(inode))
2268 goto out_stop;
2270 if (l > sizeof(EXT4_I(inode)->i_data)) {
2271 inode->i_op = &ext4_symlink_inode_operations;
2272 ext4_set_aops(inode);
2274 * page_symlink() calls into ext4_prepare/commit_write.
2275 * We have a transaction open. All is sweetness. It also sets
2276 * i_size in generic_commit_write().
2278 err = __page_symlink(inode, symname, l, 1);
2279 if (err) {
2280 clear_nlink(inode);
2281 unlock_new_inode(inode);
2282 ext4_mark_inode_dirty(handle, inode);
2283 iput(inode);
2284 goto out_stop;
2286 } else {
2287 /* clear the extent format for fast symlink */
2288 EXT4_I(inode)->i_flags &= ~EXT4_EXTENTS_FL;
2289 inode->i_op = &ext4_fast_symlink_inode_operations;
2290 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2291 inode->i_size = l-1;
2293 EXT4_I(inode)->i_disksize = inode->i_size;
2294 err = ext4_add_nondir(handle, dentry, inode);
2295 out_stop:
2296 ext4_journal_stop(handle);
2297 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2298 goto retry;
2299 return err;
2302 static int ext4_link(struct dentry *old_dentry,
2303 struct inode *dir, struct dentry *dentry)
2305 handle_t *handle;
2306 struct inode *inode = old_dentry->d_inode;
2307 int err, retries = 0;
2309 if (EXT4_DIR_LINK_MAX(inode))
2310 return -EMLINK;
2313 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2314 * otherwise has the potential to corrupt the orphan inode list.
2316 if (inode->i_nlink == 0)
2317 return -ENOENT;
2319 retry:
2320 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2321 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2322 if (IS_ERR(handle))
2323 return PTR_ERR(handle);
2325 if (IS_DIRSYNC(dir))
2326 ext4_handle_sync(handle);
2328 inode->i_ctime = ext4_current_time(inode);
2329 ext4_inc_count(handle, inode);
2330 atomic_inc(&inode->i_count);
2332 err = ext4_add_entry(handle, dentry, inode);
2333 if (!err) {
2334 ext4_mark_inode_dirty(handle, inode);
2335 d_instantiate(dentry, inode);
2336 } else {
2337 drop_nlink(inode);
2338 iput(inode);
2340 ext4_journal_stop(handle);
2341 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2342 goto retry;
2343 return err;
2346 #define PARENT_INO(buffer, size) \
2347 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2350 * Anybody can rename anything with this: the permission checks are left to the
2351 * higher-level routines.
2353 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2354 struct inode *new_dir, struct dentry *new_dentry)
2356 handle_t *handle;
2357 struct inode *old_inode, *new_inode;
2358 struct buffer_head *old_bh, *new_bh, *dir_bh;
2359 struct ext4_dir_entry_2 *old_de, *new_de;
2360 int retval, force_da_alloc = 0;
2362 old_bh = new_bh = dir_bh = NULL;
2364 /* Initialize quotas before so that eventual writes go
2365 * in separate transaction */
2366 if (new_dentry->d_inode)
2367 vfs_dq_init(new_dentry->d_inode);
2368 handle = ext4_journal_start(old_dir, 2 *
2369 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2370 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2371 if (IS_ERR(handle))
2372 return PTR_ERR(handle);
2374 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2375 ext4_handle_sync(handle);
2377 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2379 * Check for inode number is _not_ due to possible IO errors.
2380 * We might rmdir the source, keep it as pwd of some process
2381 * and merrily kill the link to whatever was created under the
2382 * same name. Goodbye sticky bit ;-<
2384 old_inode = old_dentry->d_inode;
2385 retval = -ENOENT;
2386 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2387 goto end_rename;
2389 new_inode = new_dentry->d_inode;
2390 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2391 if (new_bh) {
2392 if (!new_inode) {
2393 brelse(new_bh);
2394 new_bh = NULL;
2397 if (S_ISDIR(old_inode->i_mode)) {
2398 if (new_inode) {
2399 retval = -ENOTEMPTY;
2400 if (!empty_dir(new_inode))
2401 goto end_rename;
2403 retval = -EIO;
2404 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2405 if (!dir_bh)
2406 goto end_rename;
2407 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2408 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2409 goto end_rename;
2410 retval = -EMLINK;
2411 if (!new_inode && new_dir != old_dir &&
2412 new_dir->i_nlink >= EXT4_LINK_MAX)
2413 goto end_rename;
2415 if (!new_bh) {
2416 retval = ext4_add_entry(handle, new_dentry, old_inode);
2417 if (retval)
2418 goto end_rename;
2419 } else {
2420 BUFFER_TRACE(new_bh, "get write access");
2421 ext4_journal_get_write_access(handle, new_bh);
2422 new_de->inode = cpu_to_le32(old_inode->i_ino);
2423 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2424 EXT4_FEATURE_INCOMPAT_FILETYPE))
2425 new_de->file_type = old_de->file_type;
2426 new_dir->i_version++;
2427 new_dir->i_ctime = new_dir->i_mtime =
2428 ext4_current_time(new_dir);
2429 ext4_mark_inode_dirty(handle, new_dir);
2430 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2431 ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2432 brelse(new_bh);
2433 new_bh = NULL;
2437 * Like most other Unix systems, set the ctime for inodes on a
2438 * rename.
2440 old_inode->i_ctime = ext4_current_time(old_inode);
2441 ext4_mark_inode_dirty(handle, old_inode);
2444 * ok, that's it
2446 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2447 old_de->name_len != old_dentry->d_name.len ||
2448 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2449 (retval = ext4_delete_entry(handle, old_dir,
2450 old_de, old_bh)) == -ENOENT) {
2451 /* old_de could have moved from under us during htree split, so
2452 * make sure that we are deleting the right entry. We might
2453 * also be pointing to a stale entry in the unused part of
2454 * old_bh so just checking inum and the name isn't enough. */
2455 struct buffer_head *old_bh2;
2456 struct ext4_dir_entry_2 *old_de2;
2458 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2459 if (old_bh2) {
2460 retval = ext4_delete_entry(handle, old_dir,
2461 old_de2, old_bh2);
2462 brelse(old_bh2);
2465 if (retval) {
2466 ext4_warning(old_dir->i_sb, "ext4_rename",
2467 "Deleting old file (%lu), %d, error=%d",
2468 old_dir->i_ino, old_dir->i_nlink, retval);
2471 if (new_inode) {
2472 ext4_dec_count(handle, new_inode);
2473 new_inode->i_ctime = ext4_current_time(new_inode);
2475 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2476 ext4_update_dx_flag(old_dir);
2477 if (dir_bh) {
2478 BUFFER_TRACE(dir_bh, "get_write_access");
2479 ext4_journal_get_write_access(handle, dir_bh);
2480 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2481 cpu_to_le32(new_dir->i_ino);
2482 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2483 ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2484 ext4_dec_count(handle, old_dir);
2485 if (new_inode) {
2486 /* checked empty_dir above, can't have another parent,
2487 * ext4_dec_count() won't work for many-linked dirs */
2488 new_inode->i_nlink = 0;
2489 } else {
2490 ext4_inc_count(handle, new_dir);
2491 ext4_update_dx_flag(new_dir);
2492 ext4_mark_inode_dirty(handle, new_dir);
2495 ext4_mark_inode_dirty(handle, old_dir);
2496 if (new_inode) {
2497 ext4_mark_inode_dirty(handle, new_inode);
2498 if (!new_inode->i_nlink)
2499 ext4_orphan_add(handle, new_inode);
2500 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2501 force_da_alloc = 1;
2503 retval = 0;
2505 end_rename:
2506 brelse(dir_bh);
2507 brelse(old_bh);
2508 brelse(new_bh);
2509 ext4_journal_stop(handle);
2510 if (retval == 0 && force_da_alloc)
2511 ext4_alloc_da_blocks(old_inode);
2512 return retval;
2516 * directories can handle most operations...
2518 const struct inode_operations ext4_dir_inode_operations = {
2519 .create = ext4_create,
2520 .lookup = ext4_lookup,
2521 .link = ext4_link,
2522 .unlink = ext4_unlink,
2523 .symlink = ext4_symlink,
2524 .mkdir = ext4_mkdir,
2525 .rmdir = ext4_rmdir,
2526 .mknod = ext4_mknod,
2527 .rename = ext4_rename,
2528 .setattr = ext4_setattr,
2529 #ifdef CONFIG_EXT4_FS_XATTR
2530 .setxattr = generic_setxattr,
2531 .getxattr = generic_getxattr,
2532 .listxattr = ext4_listxattr,
2533 .removexattr = generic_removexattr,
2534 #endif
2535 .permission = ext4_permission,
2538 const struct inode_operations ext4_special_inode_operations = {
2539 .setattr = ext4_setattr,
2540 #ifdef CONFIG_EXT4_FS_XATTR
2541 .setxattr = generic_setxattr,
2542 .getxattr = generic_getxattr,
2543 .listxattr = ext4_listxattr,
2544 .removexattr = generic_removexattr,
2545 #endif
2546 .permission = ext4_permission,