ext4: partial revert to fix double brelse WARNING()
[linux-2.6/mini2440.git] / fs / ext4 / namei.c
blob6d2c1b897fc778bd6a08d680aef414a97ac5aa5e
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 "xattr.h"
41 #include "acl.h"
44 * define how far ahead to read directories while searching them.
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
49 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
51 static struct buffer_head *ext4_append(handle_t *handle,
52 struct inode *inode,
53 ext4_lblk_t *block, int *err)
55 struct buffer_head *bh;
57 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59 bh = ext4_bread(handle, inode, *block, 1, err);
60 if (bh) {
61 inode->i_size += inode->i_sb->s_blocksize;
62 EXT4_I(inode)->i_disksize = inode->i_size;
63 *err = ext4_journal_get_write_access(handle, bh);
64 if (*err) {
65 brelse(bh);
66 bh = NULL;
69 return bh;
72 #ifndef assert
73 #define assert(test) J_ASSERT(test)
74 #endif
76 #ifdef DX_DEBUG
77 #define dxtrace(command) command
78 #else
79 #define dxtrace(command)
80 #endif
82 struct fake_dirent
84 __le32 inode;
85 __le16 rec_len;
86 u8 name_len;
87 u8 file_type;
90 struct dx_countlimit
92 __le16 limit;
93 __le16 count;
96 struct dx_entry
98 __le32 hash;
99 __le32 block;
103 * dx_root_info is laid out so that if it should somehow get overlaid by a
104 * dirent the two low bits of the hash version will be zero. Therefore, the
105 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
108 struct dx_root
110 struct fake_dirent dot;
111 char dot_name[4];
112 struct fake_dirent dotdot;
113 char dotdot_name[4];
114 struct dx_root_info
116 __le32 reserved_zero;
117 u8 hash_version;
118 u8 info_length; /* 8 */
119 u8 indirect_levels;
120 u8 unused_flags;
122 info;
123 struct dx_entry entries[0];
126 struct dx_node
128 struct fake_dirent fake;
129 struct dx_entry entries[0];
133 struct dx_frame
135 struct buffer_head *bh;
136 struct dx_entry *entries;
137 struct dx_entry *at;
140 struct dx_map_entry
142 u32 hash;
143 u16 offs;
144 u16 size;
147 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
148 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
149 static inline unsigned dx_get_hash(struct dx_entry *entry);
150 static void dx_set_hash(struct dx_entry *entry, unsigned value);
151 static unsigned dx_get_count(struct dx_entry *entries);
152 static unsigned dx_get_limit(struct dx_entry *entries);
153 static void dx_set_count(struct dx_entry *entries, unsigned value);
154 static void dx_set_limit(struct dx_entry *entries, unsigned value);
155 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
156 static unsigned dx_node_limit(struct inode *dir);
157 static struct dx_frame *dx_probe(const struct qstr *d_name,
158 struct inode *dir,
159 struct dx_hash_info *hinfo,
160 struct dx_frame *frame,
161 int *err);
162 static void dx_release(struct dx_frame *frames);
163 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
164 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
165 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
166 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
167 struct dx_map_entry *offsets, int count, unsigned blocksize);
168 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
169 static void dx_insert_block(struct dx_frame *frame,
170 u32 hash, ext4_lblk_t block);
171 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
172 struct dx_frame *frame,
173 struct dx_frame *frames,
174 __u32 *start_hash);
175 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
176 const struct qstr *d_name,
177 struct ext4_dir_entry_2 **res_dir,
178 int *err);
179 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
180 struct inode *inode);
182 unsigned int ext4_rec_len_from_disk(__le16 dlen, unsigned blocksize)
184 unsigned len = le16_to_cpu(dlen);
186 if (len == EXT4_MAX_REC_LEN || len == 0)
187 return blocksize;
188 return (len & 65532) | ((len & 3) << 16);
191 __le16 ext4_rec_len_to_disk(unsigned len, unsigned blocksize)
193 if ((len > blocksize) || (blocksize > (1 << 18)) || (len & 3))
194 BUG();
195 if (len < 65536)
196 return cpu_to_le16(len);
197 if (len == blocksize) {
198 if (blocksize == 65536)
199 return cpu_to_le16(EXT4_MAX_REC_LEN);
200 else
201 return cpu_to_le16(0);
203 return cpu_to_le16((len & 65532) | ((len >> 16) & 3));
207 * p is at least 6 bytes before the end of page
209 static inline struct ext4_dir_entry_2 *
210 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
212 return (struct ext4_dir_entry_2 *)((char *)p +
213 ext4_rec_len_from_disk(p->rec_len, blocksize));
217 * Future: use high four bits of block for coalesce-on-delete flags
218 * Mask them off for now.
221 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
223 return le32_to_cpu(entry->block) & 0x00ffffff;
226 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
228 entry->block = cpu_to_le32(value);
231 static inline unsigned dx_get_hash(struct dx_entry *entry)
233 return le32_to_cpu(entry->hash);
236 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
238 entry->hash = cpu_to_le32(value);
241 static inline unsigned dx_get_count(struct dx_entry *entries)
243 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
246 static inline unsigned dx_get_limit(struct dx_entry *entries)
248 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
251 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
253 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
256 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
258 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
261 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
263 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
264 EXT4_DIR_REC_LEN(2) - infosize;
265 return entry_space / sizeof(struct dx_entry);
268 static inline unsigned dx_node_limit(struct inode *dir)
270 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
271 return entry_space / sizeof(struct dx_entry);
275 * Debug
277 #ifdef DX_DEBUG
278 static void dx_show_index(char * label, struct dx_entry *entries)
280 int i, n = dx_get_count (entries);
281 printk(KERN_DEBUG "%s index ", label);
282 for (i = 0; i < n; i++) {
283 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
284 0, (unsigned long)dx_get_block(entries + i));
286 printk("\n");
289 struct stats
291 unsigned names;
292 unsigned space;
293 unsigned bcount;
296 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
297 int size, int show_names)
299 unsigned names = 0, space = 0;
300 char *base = (char *) de;
301 struct dx_hash_info h = *hinfo;
303 printk("names: ");
304 while ((char *) de < base + size)
306 if (de->inode)
308 if (show_names)
310 int len = de->name_len;
311 char *name = de->name;
312 while (len--) printk("%c", *name++);
313 ext4fs_dirhash(de->name, de->name_len, &h);
314 printk(":%x.%u ", h.hash,
315 ((char *) de - base));
317 space += EXT4_DIR_REC_LEN(de->name_len);
318 names++;
320 de = ext4_next_entry(de, size);
322 printk("(%i)\n", names);
323 return (struct stats) { names, space, 1 };
326 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
327 struct dx_entry *entries, int levels)
329 unsigned blocksize = dir->i_sb->s_blocksize;
330 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
331 unsigned bcount = 0;
332 struct buffer_head *bh;
333 int err;
334 printk("%i indexed blocks...\n", count);
335 for (i = 0; i < count; i++, entries++)
337 ext4_lblk_t block = dx_get_block(entries);
338 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
339 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
340 struct stats stats;
341 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
342 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
343 stats = levels?
344 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
345 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
346 names += stats.names;
347 space += stats.space;
348 bcount += stats.bcount;
349 brelse(bh);
351 if (bcount)
352 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
353 levels ? "" : " ", names, space/bcount,
354 (space/bcount)*100/blocksize);
355 return (struct stats) { names, space, bcount};
357 #endif /* DX_DEBUG */
360 * Probe for a directory leaf block to search.
362 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
363 * error in the directory index, and the caller should fall back to
364 * searching the directory normally. The callers of dx_probe **MUST**
365 * check for this error code, and make sure it never gets reflected
366 * back to userspace.
368 static struct dx_frame *
369 dx_probe(const struct qstr *d_name, struct inode *dir,
370 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
372 unsigned count, indirect;
373 struct dx_entry *at, *entries, *p, *q, *m;
374 struct dx_root *root;
375 struct buffer_head *bh;
376 struct dx_frame *frame = frame_in;
377 u32 hash;
379 frame->bh = NULL;
380 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
381 goto fail;
382 root = (struct dx_root *) bh->b_data;
383 if (root->info.hash_version != DX_HASH_TEA &&
384 root->info.hash_version != DX_HASH_HALF_MD4 &&
385 root->info.hash_version != DX_HASH_LEGACY) {
386 ext4_warning(dir->i_sb, __func__,
387 "Unrecognised inode hash code %d",
388 root->info.hash_version);
389 brelse(bh);
390 *err = ERR_BAD_DX_DIR;
391 goto fail;
393 hinfo->hash_version = root->info.hash_version;
394 if (hinfo->hash_version <= DX_HASH_TEA)
395 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
396 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
397 if (d_name)
398 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
399 hash = hinfo->hash;
401 if (root->info.unused_flags & 1) {
402 ext4_warning(dir->i_sb, __func__,
403 "Unimplemented inode hash flags: %#06x",
404 root->info.unused_flags);
405 brelse(bh);
406 *err = ERR_BAD_DX_DIR;
407 goto fail;
410 if ((indirect = root->info.indirect_levels) > 1) {
411 ext4_warning(dir->i_sb, __func__,
412 "Unimplemented inode hash depth: %#06x",
413 root->info.indirect_levels);
414 brelse(bh);
415 *err = ERR_BAD_DX_DIR;
416 goto fail;
419 entries = (struct dx_entry *) (((char *)&root->info) +
420 root->info.info_length);
422 if (dx_get_limit(entries) != dx_root_limit(dir,
423 root->info.info_length)) {
424 ext4_warning(dir->i_sb, __func__,
425 "dx entry: limit != root limit");
426 brelse(bh);
427 *err = ERR_BAD_DX_DIR;
428 goto fail;
431 dxtrace(printk("Look up %x", hash));
432 while (1)
434 count = dx_get_count(entries);
435 if (!count || count > dx_get_limit(entries)) {
436 ext4_warning(dir->i_sb, __func__,
437 "dx entry: no count or count > limit");
438 brelse(bh);
439 *err = ERR_BAD_DX_DIR;
440 goto fail2;
443 p = entries + 1;
444 q = entries + count - 1;
445 while (p <= q)
447 m = p + (q - p)/2;
448 dxtrace(printk("."));
449 if (dx_get_hash(m) > hash)
450 q = m - 1;
451 else
452 p = m + 1;
455 if (0) // linear search cross check
457 unsigned n = count - 1;
458 at = entries;
459 while (n--)
461 dxtrace(printk(","));
462 if (dx_get_hash(++at) > hash)
464 at--;
465 break;
468 assert (at == p - 1);
471 at = p - 1;
472 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
473 frame->bh = bh;
474 frame->entries = entries;
475 frame->at = at;
476 if (!indirect--) return frame;
477 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
478 goto fail2;
479 at = entries = ((struct dx_node *) bh->b_data)->entries;
480 if (dx_get_limit(entries) != dx_node_limit (dir)) {
481 ext4_warning(dir->i_sb, __func__,
482 "dx entry: limit != node limit");
483 brelse(bh);
484 *err = ERR_BAD_DX_DIR;
485 goto fail2;
487 frame++;
488 frame->bh = NULL;
490 fail2:
491 while (frame >= frame_in) {
492 brelse(frame->bh);
493 frame--;
495 fail:
496 if (*err == ERR_BAD_DX_DIR)
497 ext4_warning(dir->i_sb, __func__,
498 "Corrupt dir inode %ld, running e2fsck is "
499 "recommended.", dir->i_ino);
500 return NULL;
503 static void dx_release (struct dx_frame *frames)
505 if (frames[0].bh == NULL)
506 return;
508 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
509 brelse(frames[1].bh);
510 brelse(frames[0].bh);
514 * This function increments the frame pointer to search the next leaf
515 * block, and reads in the necessary intervening nodes if the search
516 * should be necessary. Whether or not the search is necessary is
517 * controlled by the hash parameter. If the hash value is even, then
518 * the search is only continued if the next block starts with that
519 * hash value. This is used if we are searching for a specific file.
521 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
523 * This function returns 1 if the caller should continue to search,
524 * or 0 if it should not. If there is an error reading one of the
525 * index blocks, it will a negative error code.
527 * If start_hash is non-null, it will be filled in with the starting
528 * hash of the next page.
530 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
531 struct dx_frame *frame,
532 struct dx_frame *frames,
533 __u32 *start_hash)
535 struct dx_frame *p;
536 struct buffer_head *bh;
537 int err, num_frames = 0;
538 __u32 bhash;
540 p = frame;
542 * Find the next leaf page by incrementing the frame pointer.
543 * If we run out of entries in the interior node, loop around and
544 * increment pointer in the parent node. When we break out of
545 * this loop, num_frames indicates the number of interior
546 * nodes need to be read.
548 while (1) {
549 if (++(p->at) < p->entries + dx_get_count(p->entries))
550 break;
551 if (p == frames)
552 return 0;
553 num_frames++;
554 p--;
558 * If the hash is 1, then continue only if the next page has a
559 * continuation hash of any value. This is used for readdir
560 * handling. Otherwise, check to see if the hash matches the
561 * desired contiuation hash. If it doesn't, return since
562 * there's no point to read in the successive index pages.
564 bhash = dx_get_hash(p->at);
565 if (start_hash)
566 *start_hash = bhash;
567 if ((hash & 1) == 0) {
568 if ((bhash & ~1) != hash)
569 return 0;
572 * If the hash is HASH_NB_ALWAYS, we always go to the next
573 * block so no check is necessary
575 while (num_frames--) {
576 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
577 0, &err)))
578 return err; /* Failure */
579 p++;
580 brelse(p->bh);
581 p->bh = bh;
582 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
584 return 1;
589 * This function fills a red-black tree with information from a
590 * directory block. It returns the number directory entries loaded
591 * into the tree. If there is an error it is returned in err.
593 static int htree_dirblock_to_tree(struct file *dir_file,
594 struct inode *dir, ext4_lblk_t block,
595 struct dx_hash_info *hinfo,
596 __u32 start_hash, __u32 start_minor_hash)
598 struct buffer_head *bh;
599 struct ext4_dir_entry_2 *de, *top;
600 int err, count = 0;
602 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
603 (unsigned long)block));
604 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
605 return err;
607 de = (struct ext4_dir_entry_2 *) bh->b_data;
608 top = (struct ext4_dir_entry_2 *) ((char *) de +
609 dir->i_sb->s_blocksize -
610 EXT4_DIR_REC_LEN(0));
611 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
612 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
613 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
614 +((char *)de - bh->b_data))) {
615 /* On error, skip the f_pos to the next block. */
616 dir_file->f_pos = (dir_file->f_pos |
617 (dir->i_sb->s_blocksize - 1)) + 1;
618 brelse(bh);
619 return count;
621 ext4fs_dirhash(de->name, de->name_len, hinfo);
622 if ((hinfo->hash < start_hash) ||
623 ((hinfo->hash == start_hash) &&
624 (hinfo->minor_hash < start_minor_hash)))
625 continue;
626 if (de->inode == 0)
627 continue;
628 if ((err = ext4_htree_store_dirent(dir_file,
629 hinfo->hash, hinfo->minor_hash, de)) != 0) {
630 brelse(bh);
631 return err;
633 count++;
635 brelse(bh);
636 return count;
641 * This function fills a red-black tree with information from a
642 * directory. We start scanning the directory in hash order, starting
643 * at start_hash and start_minor_hash.
645 * This function returns the number of entries inserted into the tree,
646 * or a negative error code.
648 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
649 __u32 start_minor_hash, __u32 *next_hash)
651 struct dx_hash_info hinfo;
652 struct ext4_dir_entry_2 *de;
653 struct dx_frame frames[2], *frame;
654 struct inode *dir;
655 ext4_lblk_t block;
656 int count = 0;
657 int ret, err;
658 __u32 hashval;
660 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
661 start_hash, start_minor_hash));
662 dir = dir_file->f_path.dentry->d_inode;
663 if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
664 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
665 if (hinfo.hash_version <= DX_HASH_TEA)
666 hinfo.hash_version +=
667 EXT4_SB(dir->i_sb)->s_hash_unsigned;
668 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
669 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
670 start_hash, start_minor_hash);
671 *next_hash = ~0;
672 return count;
674 hinfo.hash = start_hash;
675 hinfo.minor_hash = 0;
676 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
677 if (!frame)
678 return err;
680 /* Add '.' and '..' from the htree header */
681 if (!start_hash && !start_minor_hash) {
682 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
683 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
684 goto errout;
685 count++;
687 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
688 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
689 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
690 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
691 goto errout;
692 count++;
695 while (1) {
696 block = dx_get_block(frame->at);
697 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
698 start_hash, start_minor_hash);
699 if (ret < 0) {
700 err = ret;
701 goto errout;
703 count += ret;
704 hashval = ~0;
705 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
706 frame, frames, &hashval);
707 *next_hash = hashval;
708 if (ret < 0) {
709 err = ret;
710 goto errout;
713 * Stop if: (a) there are no more entries, or
714 * (b) we have inserted at least one entry and the
715 * next hash value is not a continuation
717 if ((ret == 0) ||
718 (count && ((hashval & 1) == 0)))
719 break;
721 dx_release(frames);
722 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
723 "next hash: %x\n", count, *next_hash));
724 return count;
725 errout:
726 dx_release(frames);
727 return (err);
732 * Directory block splitting, compacting
736 * Create map of hash values, offsets, and sizes, stored at end of block.
737 * Returns number of entries mapped.
739 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
740 struct dx_hash_info *hinfo,
741 struct dx_map_entry *map_tail)
743 int count = 0;
744 char *base = (char *) de;
745 struct dx_hash_info h = *hinfo;
747 while ((char *) de < base + blocksize) {
748 if (de->name_len && de->inode) {
749 ext4fs_dirhash(de->name, de->name_len, &h);
750 map_tail--;
751 map_tail->hash = h.hash;
752 map_tail->offs = ((char *) de - base)>>2;
753 map_tail->size = le16_to_cpu(de->rec_len);
754 count++;
755 cond_resched();
757 /* XXX: do we need to check rec_len == 0 case? -Chris */
758 de = ext4_next_entry(de, blocksize);
760 return count;
763 /* Sort map by hash value */
764 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
766 struct dx_map_entry *p, *q, *top = map + count - 1;
767 int more;
768 /* Combsort until bubble sort doesn't suck */
769 while (count > 2) {
770 count = count*10/13;
771 if (count - 9 < 2) /* 9, 10 -> 11 */
772 count = 11;
773 for (p = top, q = p - count; q >= map; p--, q--)
774 if (p->hash < q->hash)
775 swap(*p, *q);
777 /* Garden variety bubble sort */
778 do {
779 more = 0;
780 q = top;
781 while (q-- > map) {
782 if (q[1].hash >= q[0].hash)
783 continue;
784 swap(*(q+1), *q);
785 more = 1;
787 } while(more);
790 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
792 struct dx_entry *entries = frame->entries;
793 struct dx_entry *old = frame->at, *new = old + 1;
794 int count = dx_get_count(entries);
796 assert(count < dx_get_limit(entries));
797 assert(old < entries + count);
798 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
799 dx_set_hash(new, hash);
800 dx_set_block(new, block);
801 dx_set_count(entries, count + 1);
804 static void ext4_update_dx_flag(struct inode *inode)
806 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
807 EXT4_FEATURE_COMPAT_DIR_INDEX))
808 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
812 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
814 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
815 * `de != NULL' is guaranteed by caller.
817 static inline int ext4_match (int len, const char * const name,
818 struct ext4_dir_entry_2 * de)
820 if (len != de->name_len)
821 return 0;
822 if (!de->inode)
823 return 0;
824 return !memcmp(name, de->name, len);
828 * Returns 0 if not found, -1 on failure, and 1 on success
830 static inline int search_dirblock(struct buffer_head *bh,
831 struct inode *dir,
832 const struct qstr *d_name,
833 unsigned int offset,
834 struct ext4_dir_entry_2 ** res_dir)
836 struct ext4_dir_entry_2 * de;
837 char * dlimit;
838 int de_len;
839 const char *name = d_name->name;
840 int namelen = d_name->len;
842 de = (struct ext4_dir_entry_2 *) bh->b_data;
843 dlimit = bh->b_data + dir->i_sb->s_blocksize;
844 while ((char *) de < dlimit) {
845 /* this code is executed quadratically often */
846 /* do minimal checking `by hand' */
848 if ((char *) de + namelen <= dlimit &&
849 ext4_match (namelen, name, de)) {
850 /* found a match - just to be sure, do a full check */
851 if (!ext4_check_dir_entry("ext4_find_entry",
852 dir, de, bh, offset))
853 return -1;
854 *res_dir = de;
855 return 1;
857 /* prevent looping on a bad block */
858 de_len = ext4_rec_len_from_disk(de->rec_len,
859 dir->i_sb->s_blocksize);
860 if (de_len <= 0)
861 return -1;
862 offset += de_len;
863 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
865 return 0;
870 * ext4_find_entry()
872 * finds an entry in the specified directory with the wanted name. It
873 * returns the cache buffer in which the entry was found, and the entry
874 * itself (as a parameter - res_dir). It does NOT read the inode of the
875 * entry - you'll have to do that yourself if you want to.
877 * The returned buffer_head has ->b_count elevated. The caller is expected
878 * to brelse() it when appropriate.
880 static struct buffer_head * ext4_find_entry (struct inode *dir,
881 const struct qstr *d_name,
882 struct ext4_dir_entry_2 ** res_dir)
884 struct super_block *sb;
885 struct buffer_head *bh_use[NAMEI_RA_SIZE];
886 struct buffer_head *bh, *ret = NULL;
887 ext4_lblk_t start, block, b;
888 int ra_max = 0; /* Number of bh's in the readahead
889 buffer, bh_use[] */
890 int ra_ptr = 0; /* Current index into readahead
891 buffer */
892 int num = 0;
893 ext4_lblk_t nblocks;
894 int i, err;
895 int namelen;
897 *res_dir = NULL;
898 sb = dir->i_sb;
899 namelen = d_name->len;
900 if (namelen > EXT4_NAME_LEN)
901 return NULL;
902 if (is_dx(dir)) {
903 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
905 * On success, or if the error was file not found,
906 * return. Otherwise, fall back to doing a search the
907 * old fashioned way.
909 if (bh || (err != ERR_BAD_DX_DIR))
910 return bh;
911 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
912 "falling back\n"));
914 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
915 start = EXT4_I(dir)->i_dir_start_lookup;
916 if (start >= nblocks)
917 start = 0;
918 block = start;
919 restart:
920 do {
922 * We deal with the read-ahead logic here.
924 if (ra_ptr >= ra_max) {
925 /* Refill the readahead buffer */
926 ra_ptr = 0;
927 b = block;
928 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
930 * Terminate if we reach the end of the
931 * directory and must wrap, or if our
932 * search has finished at this block.
934 if (b >= nblocks || (num && block == start)) {
935 bh_use[ra_max] = NULL;
936 break;
938 num++;
939 bh = ext4_getblk(NULL, dir, b++, 0, &err);
940 bh_use[ra_max] = bh;
941 if (bh)
942 ll_rw_block(READ_META, 1, &bh);
945 if ((bh = bh_use[ra_ptr++]) == NULL)
946 goto next;
947 wait_on_buffer(bh);
948 if (!buffer_uptodate(bh)) {
949 /* read error, skip block & hope for the best */
950 ext4_error(sb, __func__, "reading directory #%lu "
951 "offset %lu", dir->i_ino,
952 (unsigned long)block);
953 brelse(bh);
954 goto next;
956 i = search_dirblock(bh, dir, d_name,
957 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
958 if (i == 1) {
959 EXT4_I(dir)->i_dir_start_lookup = block;
960 ret = bh;
961 goto cleanup_and_exit;
962 } else {
963 brelse(bh);
964 if (i < 0)
965 goto cleanup_and_exit;
967 next:
968 if (++block >= nblocks)
969 block = 0;
970 } while (block != start);
973 * If the directory has grown while we were searching, then
974 * search the last part of the directory before giving up.
976 block = nblocks;
977 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
978 if (block < nblocks) {
979 start = 0;
980 goto restart;
983 cleanup_and_exit:
984 /* Clean up the read-ahead blocks */
985 for (; ra_ptr < ra_max; ra_ptr++)
986 brelse(bh_use[ra_ptr]);
987 return ret;
990 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
991 struct ext4_dir_entry_2 **res_dir, int *err)
993 struct super_block * sb;
994 struct dx_hash_info hinfo;
995 u32 hash;
996 struct dx_frame frames[2], *frame;
997 struct ext4_dir_entry_2 *de, *top;
998 struct buffer_head *bh;
999 ext4_lblk_t block;
1000 int retval;
1001 int namelen = d_name->len;
1002 const u8 *name = d_name->name;
1004 sb = dir->i_sb;
1005 /* NFS may look up ".." - look at dx_root directory block */
1006 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
1007 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1008 return NULL;
1009 } else {
1010 frame = frames;
1011 frame->bh = NULL; /* for dx_release() */
1012 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
1013 dx_set_block(frame->at, 0); /* dx_root block is 0 */
1015 hash = hinfo.hash;
1016 do {
1017 block = dx_get_block(frame->at);
1018 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
1019 goto errout;
1020 de = (struct ext4_dir_entry_2 *) bh->b_data;
1021 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
1022 EXT4_DIR_REC_LEN(0));
1023 for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) {
1024 int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
1025 + ((char *) de - bh->b_data);
1027 if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) {
1028 brelse(bh);
1029 *err = ERR_BAD_DX_DIR;
1030 goto errout;
1033 if (ext4_match(namelen, name, de)) {
1034 *res_dir = de;
1035 dx_release(frames);
1036 return bh;
1039 brelse(bh);
1040 /* Check to see if we should continue to search */
1041 retval = ext4_htree_next_block(dir, hash, frame,
1042 frames, NULL);
1043 if (retval < 0) {
1044 ext4_warning(sb, __func__,
1045 "error reading index page in directory #%lu",
1046 dir->i_ino);
1047 *err = retval;
1048 goto errout;
1050 } while (retval == 1);
1052 *err = -ENOENT;
1053 errout:
1054 dxtrace(printk(KERN_DEBUG "%s not found\n", name));
1055 dx_release (frames);
1056 return NULL;
1059 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1061 struct inode *inode;
1062 struct ext4_dir_entry_2 *de;
1063 struct buffer_head *bh;
1065 if (dentry->d_name.len > EXT4_NAME_LEN)
1066 return ERR_PTR(-ENAMETOOLONG);
1068 bh = ext4_find_entry(dir, &dentry->d_name, &de);
1069 inode = NULL;
1070 if (bh) {
1071 __u32 ino = le32_to_cpu(de->inode);
1072 brelse(bh);
1073 if (!ext4_valid_inum(dir->i_sb, ino)) {
1074 ext4_error(dir->i_sb, "ext4_lookup",
1075 "bad inode number: %u", ino);
1076 return ERR_PTR(-EIO);
1078 inode = ext4_iget(dir->i_sb, ino);
1079 if (unlikely(IS_ERR(inode))) {
1080 if (PTR_ERR(inode) == -ESTALE) {
1081 ext4_error(dir->i_sb, __func__,
1082 "deleted inode referenced: %u",
1083 ino);
1084 return ERR_PTR(-EIO);
1085 } else {
1086 return ERR_CAST(inode);
1090 return d_splice_alias(inode, dentry);
1094 struct dentry *ext4_get_parent(struct dentry *child)
1096 __u32 ino;
1097 struct inode *inode;
1098 static const struct qstr dotdot = {
1099 .name = "..",
1100 .len = 2,
1102 struct ext4_dir_entry_2 * de;
1103 struct buffer_head *bh;
1105 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1106 inode = NULL;
1107 if (!bh)
1108 return ERR_PTR(-ENOENT);
1109 ino = le32_to_cpu(de->inode);
1110 brelse(bh);
1112 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1113 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1114 "bad inode number: %u", ino);
1115 return ERR_PTR(-EIO);
1118 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1121 #define S_SHIFT 12
1122 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1123 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1124 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1125 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1126 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1127 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1128 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1129 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
1132 static inline void ext4_set_de_type(struct super_block *sb,
1133 struct ext4_dir_entry_2 *de,
1134 umode_t mode) {
1135 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1136 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1140 * Move count entries from end of map between two memory locations.
1141 * Returns pointer to last entry moved.
1143 static struct ext4_dir_entry_2 *
1144 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1145 unsigned blocksize)
1147 unsigned rec_len = 0;
1149 while (count--) {
1150 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1151 (from + (map->offs<<2));
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 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1594 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1595 sb->s_blocksize);
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, &dentry->d_name, 0);
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, &dentry->d_name, 0);
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 &dentry->d_name, 0);
1858 err = PTR_ERR(inode);
1859 if (IS_ERR(inode))
1860 goto out_stop;
1862 inode->i_op = &ext4_dir_inode_operations;
1863 inode->i_fop = &ext4_dir_operations;
1864 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1865 dir_block = ext4_bread(handle, inode, 0, 1, &err);
1866 if (!dir_block)
1867 goto out_clear_inode;
1868 BUFFER_TRACE(dir_block, "get_write_access");
1869 ext4_journal_get_write_access(handle, dir_block);
1870 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1871 de->inode = cpu_to_le32(inode->i_ino);
1872 de->name_len = 1;
1873 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1874 blocksize);
1875 strcpy(de->name, ".");
1876 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1877 de = ext4_next_entry(de, blocksize);
1878 de->inode = cpu_to_le32(dir->i_ino);
1879 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1880 blocksize);
1881 de->name_len = 2;
1882 strcpy(de->name, "..");
1883 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1884 inode->i_nlink = 2;
1885 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1886 ext4_handle_dirty_metadata(handle, dir, dir_block);
1887 brelse(dir_block);
1888 ext4_mark_inode_dirty(handle, inode);
1889 err = ext4_add_entry(handle, dentry, inode);
1890 if (err) {
1891 out_clear_inode:
1892 clear_nlink(inode);
1893 unlock_new_inode(inode);
1894 ext4_mark_inode_dirty(handle, inode);
1895 iput(inode);
1896 goto out_stop;
1898 ext4_inc_count(handle, dir);
1899 ext4_update_dx_flag(dir);
1900 ext4_mark_inode_dirty(handle, dir);
1901 d_instantiate(dentry, inode);
1902 unlock_new_inode(inode);
1903 out_stop:
1904 ext4_journal_stop(handle);
1905 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1906 goto retry;
1907 return err;
1911 * routine to check that the specified directory is empty (for rmdir)
1913 static int empty_dir(struct inode *inode)
1915 unsigned int offset;
1916 struct buffer_head *bh;
1917 struct ext4_dir_entry_2 *de, *de1;
1918 struct super_block *sb;
1919 int err = 0;
1921 sb = inode->i_sb;
1922 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1923 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1924 if (err)
1925 ext4_error(inode->i_sb, __func__,
1926 "error %d reading directory #%lu offset 0",
1927 err, inode->i_ino);
1928 else
1929 ext4_warning(inode->i_sb, __func__,
1930 "bad directory (dir #%lu) - no data block",
1931 inode->i_ino);
1932 return 1;
1934 de = (struct ext4_dir_entry_2 *) bh->b_data;
1935 de1 = ext4_next_entry(de, sb->s_blocksize);
1936 if (le32_to_cpu(de->inode) != inode->i_ino ||
1937 !le32_to_cpu(de1->inode) ||
1938 strcmp(".", de->name) ||
1939 strcmp("..", de1->name)) {
1940 ext4_warning(inode->i_sb, "empty_dir",
1941 "bad directory (dir #%lu) - no `.' or `..'",
1942 inode->i_ino);
1943 brelse(bh);
1944 return 1;
1946 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1947 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1948 de = ext4_next_entry(de1, sb->s_blocksize);
1949 while (offset < inode->i_size) {
1950 if (!bh ||
1951 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1952 err = 0;
1953 brelse(bh);
1954 bh = ext4_bread(NULL, inode,
1955 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1956 if (!bh) {
1957 if (err)
1958 ext4_error(sb, __func__,
1959 "error %d reading directory"
1960 " #%lu offset %u",
1961 err, inode->i_ino, offset);
1962 offset += sb->s_blocksize;
1963 continue;
1965 de = (struct ext4_dir_entry_2 *) bh->b_data;
1967 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1968 de = (struct ext4_dir_entry_2 *)(bh->b_data +
1969 sb->s_blocksize);
1970 offset = (offset | (sb->s_blocksize - 1)) + 1;
1971 continue;
1973 if (le32_to_cpu(de->inode)) {
1974 brelse(bh);
1975 return 0;
1977 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1978 de = ext4_next_entry(de, sb->s_blocksize);
1980 brelse(bh);
1981 return 1;
1984 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1985 * such inodes, starting at the superblock, in case we crash before the
1986 * file is closed/deleted, or in case the inode truncate spans multiple
1987 * transactions and the last transaction is not recovered after a crash.
1989 * At filesystem recovery time, we walk this list deleting unlinked
1990 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1992 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1994 struct super_block *sb = inode->i_sb;
1995 struct ext4_iloc iloc;
1996 int err = 0, rc;
1998 if (!ext4_handle_valid(handle))
1999 return 0;
2001 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
2002 if (!list_empty(&EXT4_I(inode)->i_orphan))
2003 goto out_unlock;
2005 /* Orphan handling is only valid for files with data blocks
2006 * being truncated, or files being unlinked. */
2008 /* @@@ FIXME: Observation from aviro:
2009 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
2010 * here (on s_orphan_lock), so race with ext4_link() which might bump
2011 * ->i_nlink. For, say it, character device. Not a regular file,
2012 * not a directory, not a symlink and ->i_nlink > 0.
2014 * tytso, 4/25/2009: I'm not sure how that could happen;
2015 * shouldn't the fs core protect us from these sort of
2016 * unlink()/link() races?
2018 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2019 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2021 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2022 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2023 if (err)
2024 goto out_unlock;
2026 err = ext4_reserve_inode_write(handle, inode, &iloc);
2027 if (err)
2028 goto out_unlock;
2030 /* Insert this inode at the head of the on-disk orphan list... */
2031 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2032 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2033 err = ext4_handle_dirty_metadata(handle, inode, EXT4_SB(sb)->s_sbh);
2034 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2035 if (!err)
2036 err = rc;
2038 /* Only add to the head of the in-memory list if all the
2039 * previous operations succeeded. If the orphan_add is going to
2040 * fail (possibly taking the journal offline), we can't risk
2041 * leaving the inode on the orphan list: stray orphan-list
2042 * entries can cause panics at unmount time.
2044 * This is safe: on error we're going to ignore the orphan list
2045 * anyway on the next recovery. */
2046 if (!err)
2047 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2049 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2050 jbd_debug(4, "orphan inode %lu will point to %d\n",
2051 inode->i_ino, NEXT_ORPHAN(inode));
2052 out_unlock:
2053 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2054 ext4_std_error(inode->i_sb, err);
2055 return err;
2059 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2060 * of such inodes stored on disk, because it is finally being cleaned up.
2062 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2064 struct list_head *prev;
2065 struct ext4_inode_info *ei = EXT4_I(inode);
2066 struct ext4_sb_info *sbi;
2067 __u32 ino_next;
2068 struct ext4_iloc iloc;
2069 int err = 0;
2071 /* ext4_handle_valid() assumes a valid handle_t pointer */
2072 if (handle && !ext4_handle_valid(handle))
2073 return 0;
2075 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2076 if (list_empty(&ei->i_orphan))
2077 goto out;
2079 ino_next = NEXT_ORPHAN(inode);
2080 prev = ei->i_orphan.prev;
2081 sbi = EXT4_SB(inode->i_sb);
2083 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2085 list_del_init(&ei->i_orphan);
2087 /* If we're on an error path, we may not have a valid
2088 * transaction handle with which to update the orphan list on
2089 * disk, but we still need to remove the inode from the linked
2090 * list in memory. */
2091 if (sbi->s_journal && !handle)
2092 goto out;
2094 err = ext4_reserve_inode_write(handle, inode, &iloc);
2095 if (err)
2096 goto out_err;
2098 if (prev == &sbi->s_orphan) {
2099 jbd_debug(4, "superblock will point to %u\n", ino_next);
2100 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2101 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2102 if (err)
2103 goto out_brelse;
2104 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2105 err = ext4_handle_dirty_metadata(handle, inode, sbi->s_sbh);
2106 } else {
2107 struct ext4_iloc iloc2;
2108 struct inode *i_prev =
2109 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2111 jbd_debug(4, "orphan inode %lu will point to %u\n",
2112 i_prev->i_ino, ino_next);
2113 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2114 if (err)
2115 goto out_brelse;
2116 NEXT_ORPHAN(i_prev) = ino_next;
2117 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2119 if (err)
2120 goto out_brelse;
2121 NEXT_ORPHAN(inode) = 0;
2122 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2124 out_err:
2125 ext4_std_error(inode->i_sb, err);
2126 out:
2127 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2128 return err;
2130 out_brelse:
2131 brelse(iloc.bh);
2132 goto out_err;
2135 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2137 int retval;
2138 struct inode *inode;
2139 struct buffer_head *bh;
2140 struct ext4_dir_entry_2 *de;
2141 handle_t *handle;
2143 /* Initialize quotas before so that eventual writes go in
2144 * separate transaction */
2145 vfs_dq_init(dentry->d_inode);
2146 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2147 if (IS_ERR(handle))
2148 return PTR_ERR(handle);
2150 retval = -ENOENT;
2151 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2152 if (!bh)
2153 goto end_rmdir;
2155 if (IS_DIRSYNC(dir))
2156 ext4_handle_sync(handle);
2158 inode = dentry->d_inode;
2160 retval = -EIO;
2161 if (le32_to_cpu(de->inode) != inode->i_ino)
2162 goto end_rmdir;
2164 retval = -ENOTEMPTY;
2165 if (!empty_dir(inode))
2166 goto end_rmdir;
2168 retval = ext4_delete_entry(handle, dir, de, bh);
2169 if (retval)
2170 goto end_rmdir;
2171 if (!EXT4_DIR_LINK_EMPTY(inode))
2172 ext4_warning(inode->i_sb, "ext4_rmdir",
2173 "empty directory has too many links (%d)",
2174 inode->i_nlink);
2175 inode->i_version++;
2176 clear_nlink(inode);
2177 /* There's no need to set i_disksize: the fact that i_nlink is
2178 * zero will ensure that the right thing happens during any
2179 * recovery. */
2180 inode->i_size = 0;
2181 ext4_orphan_add(handle, inode);
2182 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2183 ext4_mark_inode_dirty(handle, inode);
2184 ext4_dec_count(handle, dir);
2185 ext4_update_dx_flag(dir);
2186 ext4_mark_inode_dirty(handle, dir);
2188 end_rmdir:
2189 ext4_journal_stop(handle);
2190 brelse(bh);
2191 return retval;
2194 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2196 int retval;
2197 struct inode *inode;
2198 struct buffer_head *bh;
2199 struct ext4_dir_entry_2 *de;
2200 handle_t *handle;
2202 /* Initialize quotas before so that eventual writes go
2203 * in separate transaction */
2204 vfs_dq_init(dentry->d_inode);
2205 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2206 if (IS_ERR(handle))
2207 return PTR_ERR(handle);
2209 if (IS_DIRSYNC(dir))
2210 ext4_handle_sync(handle);
2212 retval = -ENOENT;
2213 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2214 if (!bh)
2215 goto end_unlink;
2217 inode = dentry->d_inode;
2219 retval = -EIO;
2220 if (le32_to_cpu(de->inode) != inode->i_ino)
2221 goto end_unlink;
2223 if (!inode->i_nlink) {
2224 ext4_warning(inode->i_sb, "ext4_unlink",
2225 "Deleting nonexistent file (%lu), %d",
2226 inode->i_ino, inode->i_nlink);
2227 inode->i_nlink = 1;
2229 retval = ext4_delete_entry(handle, dir, de, bh);
2230 if (retval)
2231 goto end_unlink;
2232 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2233 ext4_update_dx_flag(dir);
2234 ext4_mark_inode_dirty(handle, dir);
2235 drop_nlink(inode);
2236 if (!inode->i_nlink)
2237 ext4_orphan_add(handle, inode);
2238 inode->i_ctime = ext4_current_time(inode);
2239 ext4_mark_inode_dirty(handle, inode);
2240 retval = 0;
2242 end_unlink:
2243 ext4_journal_stop(handle);
2244 brelse(bh);
2245 return retval;
2248 static int ext4_symlink(struct inode *dir,
2249 struct dentry *dentry, const char *symname)
2251 handle_t *handle;
2252 struct inode *inode;
2253 int l, err, retries = 0;
2255 l = strlen(symname)+1;
2256 if (l > dir->i_sb->s_blocksize)
2257 return -ENAMETOOLONG;
2259 retry:
2260 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2261 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2262 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2263 if (IS_ERR(handle))
2264 return PTR_ERR(handle);
2266 if (IS_DIRSYNC(dir))
2267 ext4_handle_sync(handle);
2269 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2270 &dentry->d_name, 0);
2271 err = PTR_ERR(inode);
2272 if (IS_ERR(inode))
2273 goto out_stop;
2275 if (l > sizeof(EXT4_I(inode)->i_data)) {
2276 inode->i_op = &ext4_symlink_inode_operations;
2277 ext4_set_aops(inode);
2279 * page_symlink() calls into ext4_prepare/commit_write.
2280 * We have a transaction open. All is sweetness. It also sets
2281 * i_size in generic_commit_write().
2283 err = __page_symlink(inode, symname, l, 1);
2284 if (err) {
2285 clear_nlink(inode);
2286 unlock_new_inode(inode);
2287 ext4_mark_inode_dirty(handle, inode);
2288 iput(inode);
2289 goto out_stop;
2291 } else {
2292 /* clear the extent format for fast symlink */
2293 EXT4_I(inode)->i_flags &= ~EXT4_EXTENTS_FL;
2294 inode->i_op = &ext4_fast_symlink_inode_operations;
2295 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2296 inode->i_size = l-1;
2298 EXT4_I(inode)->i_disksize = inode->i_size;
2299 err = ext4_add_nondir(handle, dentry, inode);
2300 out_stop:
2301 ext4_journal_stop(handle);
2302 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2303 goto retry;
2304 return err;
2307 static int ext4_link(struct dentry *old_dentry,
2308 struct inode *dir, struct dentry *dentry)
2310 handle_t *handle;
2311 struct inode *inode = old_dentry->d_inode;
2312 int err, retries = 0;
2314 if (inode->i_nlink >= EXT4_LINK_MAX)
2315 return -EMLINK;
2318 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2319 * otherwise has the potential to corrupt the orphan inode list.
2321 if (inode->i_nlink == 0)
2322 return -ENOENT;
2324 retry:
2325 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2326 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2327 if (IS_ERR(handle))
2328 return PTR_ERR(handle);
2330 if (IS_DIRSYNC(dir))
2331 ext4_handle_sync(handle);
2333 inode->i_ctime = ext4_current_time(inode);
2334 ext4_inc_count(handle, inode);
2335 atomic_inc(&inode->i_count);
2337 err = ext4_add_entry(handle, dentry, inode);
2338 if (!err) {
2339 ext4_mark_inode_dirty(handle, inode);
2340 d_instantiate(dentry, inode);
2341 } else {
2342 drop_nlink(inode);
2343 iput(inode);
2345 ext4_journal_stop(handle);
2346 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2347 goto retry;
2348 return err;
2351 #define PARENT_INO(buffer, size) \
2352 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2355 * Anybody can rename anything with this: the permission checks are left to the
2356 * higher-level routines.
2358 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2359 struct inode *new_dir, struct dentry *new_dentry)
2361 handle_t *handle;
2362 struct inode *old_inode, *new_inode;
2363 struct buffer_head *old_bh, *new_bh, *dir_bh;
2364 struct ext4_dir_entry_2 *old_de, *new_de;
2365 int retval, force_da_alloc = 0;
2367 old_bh = new_bh = dir_bh = NULL;
2369 /* Initialize quotas before so that eventual writes go
2370 * in separate transaction */
2371 if (new_dentry->d_inode)
2372 vfs_dq_init(new_dentry->d_inode);
2373 handle = ext4_journal_start(old_dir, 2 *
2374 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2375 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2376 if (IS_ERR(handle))
2377 return PTR_ERR(handle);
2379 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2380 ext4_handle_sync(handle);
2382 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2384 * Check for inode number is _not_ due to possible IO errors.
2385 * We might rmdir the source, keep it as pwd of some process
2386 * and merrily kill the link to whatever was created under the
2387 * same name. Goodbye sticky bit ;-<
2389 old_inode = old_dentry->d_inode;
2390 retval = -ENOENT;
2391 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2392 goto end_rename;
2394 new_inode = new_dentry->d_inode;
2395 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2396 if (new_bh) {
2397 if (!new_inode) {
2398 brelse(new_bh);
2399 new_bh = NULL;
2402 if (S_ISDIR(old_inode->i_mode)) {
2403 if (new_inode) {
2404 retval = -ENOTEMPTY;
2405 if (!empty_dir(new_inode))
2406 goto end_rename;
2408 retval = -EIO;
2409 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2410 if (!dir_bh)
2411 goto end_rename;
2412 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2413 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2414 goto end_rename;
2415 retval = -EMLINK;
2416 if (!new_inode && new_dir != old_dir &&
2417 EXT4_DIR_LINK_MAX(new_dir))
2418 goto end_rename;
2420 if (!new_bh) {
2421 retval = ext4_add_entry(handle, new_dentry, old_inode);
2422 if (retval)
2423 goto end_rename;
2424 } else {
2425 BUFFER_TRACE(new_bh, "get write access");
2426 ext4_journal_get_write_access(handle, new_bh);
2427 new_de->inode = cpu_to_le32(old_inode->i_ino);
2428 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2429 EXT4_FEATURE_INCOMPAT_FILETYPE))
2430 new_de->file_type = old_de->file_type;
2431 new_dir->i_version++;
2432 new_dir->i_ctime = new_dir->i_mtime =
2433 ext4_current_time(new_dir);
2434 ext4_mark_inode_dirty(handle, new_dir);
2435 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2436 ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2437 brelse(new_bh);
2438 new_bh = NULL;
2442 * Like most other Unix systems, set the ctime for inodes on a
2443 * rename.
2445 old_inode->i_ctime = ext4_current_time(old_inode);
2446 ext4_mark_inode_dirty(handle, old_inode);
2449 * ok, that's it
2451 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2452 old_de->name_len != old_dentry->d_name.len ||
2453 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2454 (retval = ext4_delete_entry(handle, old_dir,
2455 old_de, old_bh)) == -ENOENT) {
2456 /* old_de could have moved from under us during htree split, so
2457 * make sure that we are deleting the right entry. We might
2458 * also be pointing to a stale entry in the unused part of
2459 * old_bh so just checking inum and the name isn't enough. */
2460 struct buffer_head *old_bh2;
2461 struct ext4_dir_entry_2 *old_de2;
2463 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2464 if (old_bh2) {
2465 retval = ext4_delete_entry(handle, old_dir,
2466 old_de2, old_bh2);
2467 brelse(old_bh2);
2470 if (retval) {
2471 ext4_warning(old_dir->i_sb, "ext4_rename",
2472 "Deleting old file (%lu), %d, error=%d",
2473 old_dir->i_ino, old_dir->i_nlink, retval);
2476 if (new_inode) {
2477 ext4_dec_count(handle, new_inode);
2478 new_inode->i_ctime = ext4_current_time(new_inode);
2480 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2481 ext4_update_dx_flag(old_dir);
2482 if (dir_bh) {
2483 BUFFER_TRACE(dir_bh, "get_write_access");
2484 ext4_journal_get_write_access(handle, dir_bh);
2485 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2486 cpu_to_le32(new_dir->i_ino);
2487 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2488 ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2489 ext4_dec_count(handle, old_dir);
2490 if (new_inode) {
2491 /* checked empty_dir above, can't have another parent,
2492 * ext4_dec_count() won't work for many-linked dirs */
2493 new_inode->i_nlink = 0;
2494 } else {
2495 ext4_inc_count(handle, new_dir);
2496 ext4_update_dx_flag(new_dir);
2497 ext4_mark_inode_dirty(handle, new_dir);
2500 ext4_mark_inode_dirty(handle, old_dir);
2501 if (new_inode) {
2502 ext4_mark_inode_dirty(handle, new_inode);
2503 if (!new_inode->i_nlink)
2504 ext4_orphan_add(handle, new_inode);
2505 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2506 force_da_alloc = 1;
2508 retval = 0;
2510 end_rename:
2511 brelse(dir_bh);
2512 brelse(old_bh);
2513 brelse(new_bh);
2514 ext4_journal_stop(handle);
2515 if (retval == 0 && force_da_alloc)
2516 ext4_alloc_da_blocks(old_inode);
2517 return retval;
2521 * directories can handle most operations...
2523 const struct inode_operations ext4_dir_inode_operations = {
2524 .create = ext4_create,
2525 .lookup = ext4_lookup,
2526 .link = ext4_link,
2527 .unlink = ext4_unlink,
2528 .symlink = ext4_symlink,
2529 .mkdir = ext4_mkdir,
2530 .rmdir = ext4_rmdir,
2531 .mknod = ext4_mknod,
2532 .rename = ext4_rename,
2533 .setattr = ext4_setattr,
2534 #ifdef CONFIG_EXT4_FS_XATTR
2535 .setxattr = generic_setxattr,
2536 .getxattr = generic_getxattr,
2537 .listxattr = ext4_listxattr,
2538 .removexattr = generic_removexattr,
2539 #endif
2540 .check_acl = ext4_check_acl,
2541 .fiemap = ext4_fiemap,
2544 const struct inode_operations ext4_special_inode_operations = {
2545 .setattr = ext4_setattr,
2546 #ifdef CONFIG_EXT4_FS_XATTR
2547 .setxattr = generic_setxattr,
2548 .getxattr = generic_getxattr,
2549 .listxattr = ext4_listxattr,
2550 .removexattr = generic_removexattr,
2551 #endif
2552 .check_acl = ext4_check_acl,