FRV: Use generic show_interrupts()
[cris-mirror.git] / fs / ext4 / namei.c
blob67fd0b0258589ae64428d26530807b898e79854b
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"
43 #include <trace/events/ext4.h>
45 * define how far ahead to read directories while searching them.
47 #define NAMEI_RA_CHUNKS 2
48 #define NAMEI_RA_BLOCKS 4
49 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
52 static struct buffer_head *ext4_append(handle_t *handle,
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);
184 * p is at least 6 bytes before the end of page
186 static inline struct ext4_dir_entry_2 *
187 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
189 return (struct ext4_dir_entry_2 *)((char *)p +
190 ext4_rec_len_from_disk(p->rec_len, blocksize));
194 * Future: use high four bits of block for coalesce-on-delete flags
195 * Mask them off for now.
198 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
200 return le32_to_cpu(entry->block) & 0x00ffffff;
203 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
205 entry->block = cpu_to_le32(value);
208 static inline unsigned dx_get_hash(struct dx_entry *entry)
210 return le32_to_cpu(entry->hash);
213 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
215 entry->hash = cpu_to_le32(value);
218 static inline unsigned dx_get_count(struct dx_entry *entries)
220 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
223 static inline unsigned dx_get_limit(struct dx_entry *entries)
225 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
228 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
230 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
233 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
235 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
238 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
240 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
241 EXT4_DIR_REC_LEN(2) - infosize;
242 return entry_space / sizeof(struct dx_entry);
245 static inline unsigned dx_node_limit(struct inode *dir)
247 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
248 return entry_space / sizeof(struct dx_entry);
252 * Debug
254 #ifdef DX_DEBUG
255 static void dx_show_index(char * label, struct dx_entry *entries)
257 int i, n = dx_get_count (entries);
258 printk(KERN_DEBUG "%s index ", label);
259 for (i = 0; i < n; i++) {
260 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
261 0, (unsigned long)dx_get_block(entries + i));
263 printk("\n");
266 struct stats
268 unsigned names;
269 unsigned space;
270 unsigned bcount;
273 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
274 int size, int show_names)
276 unsigned names = 0, space = 0;
277 char *base = (char *) de;
278 struct dx_hash_info h = *hinfo;
280 printk("names: ");
281 while ((char *) de < base + size)
283 if (de->inode)
285 if (show_names)
287 int len = de->name_len;
288 char *name = de->name;
289 while (len--) printk("%c", *name++);
290 ext4fs_dirhash(de->name, de->name_len, &h);
291 printk(":%x.%u ", h.hash,
292 ((char *) de - base));
294 space += EXT4_DIR_REC_LEN(de->name_len);
295 names++;
297 de = ext4_next_entry(de, size);
299 printk("(%i)\n", names);
300 return (struct stats) { names, space, 1 };
303 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
304 struct dx_entry *entries, int levels)
306 unsigned blocksize = dir->i_sb->s_blocksize;
307 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
308 unsigned bcount = 0;
309 struct buffer_head *bh;
310 int err;
311 printk("%i indexed blocks...\n", count);
312 for (i = 0; i < count; i++, entries++)
314 ext4_lblk_t block = dx_get_block(entries);
315 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
316 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
317 struct stats stats;
318 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
319 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
320 stats = levels?
321 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
322 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
323 names += stats.names;
324 space += stats.space;
325 bcount += stats.bcount;
326 brelse(bh);
328 if (bcount)
329 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
330 levels ? "" : " ", names, space/bcount,
331 (space/bcount)*100/blocksize);
332 return (struct stats) { names, space, bcount};
334 #endif /* DX_DEBUG */
337 * Probe for a directory leaf block to search.
339 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
340 * error in the directory index, and the caller should fall back to
341 * searching the directory normally. The callers of dx_probe **MUST**
342 * check for this error code, and make sure it never gets reflected
343 * back to userspace.
345 static struct dx_frame *
346 dx_probe(const struct qstr *d_name, struct inode *dir,
347 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
349 unsigned count, indirect;
350 struct dx_entry *at, *entries, *p, *q, *m;
351 struct dx_root *root;
352 struct buffer_head *bh;
353 struct dx_frame *frame = frame_in;
354 u32 hash;
356 frame->bh = NULL;
357 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
358 goto fail;
359 root = (struct dx_root *) bh->b_data;
360 if (root->info.hash_version != DX_HASH_TEA &&
361 root->info.hash_version != DX_HASH_HALF_MD4 &&
362 root->info.hash_version != DX_HASH_LEGACY) {
363 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
364 root->info.hash_version);
365 brelse(bh);
366 *err = ERR_BAD_DX_DIR;
367 goto fail;
369 hinfo->hash_version = root->info.hash_version;
370 if (hinfo->hash_version <= DX_HASH_TEA)
371 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
372 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
373 if (d_name)
374 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
375 hash = hinfo->hash;
377 if (root->info.unused_flags & 1) {
378 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
379 root->info.unused_flags);
380 brelse(bh);
381 *err = ERR_BAD_DX_DIR;
382 goto fail;
385 if ((indirect = root->info.indirect_levels) > 1) {
386 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
387 root->info.indirect_levels);
388 brelse(bh);
389 *err = ERR_BAD_DX_DIR;
390 goto fail;
393 entries = (struct dx_entry *) (((char *)&root->info) +
394 root->info.info_length);
396 if (dx_get_limit(entries) != dx_root_limit(dir,
397 root->info.info_length)) {
398 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
399 brelse(bh);
400 *err = ERR_BAD_DX_DIR;
401 goto fail;
404 dxtrace(printk("Look up %x", hash));
405 while (1)
407 count = dx_get_count(entries);
408 if (!count || count > dx_get_limit(entries)) {
409 ext4_warning(dir->i_sb,
410 "dx entry: no count or count > limit");
411 brelse(bh);
412 *err = ERR_BAD_DX_DIR;
413 goto fail2;
416 p = entries + 1;
417 q = entries + count - 1;
418 while (p <= q)
420 m = p + (q - p)/2;
421 dxtrace(printk("."));
422 if (dx_get_hash(m) > hash)
423 q = m - 1;
424 else
425 p = m + 1;
428 if (0) // linear search cross check
430 unsigned n = count - 1;
431 at = entries;
432 while (n--)
434 dxtrace(printk(","));
435 if (dx_get_hash(++at) > hash)
437 at--;
438 break;
441 assert (at == p - 1);
444 at = p - 1;
445 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
446 frame->bh = bh;
447 frame->entries = entries;
448 frame->at = at;
449 if (!indirect--) return frame;
450 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
451 goto fail2;
452 at = entries = ((struct dx_node *) bh->b_data)->entries;
453 if (dx_get_limit(entries) != dx_node_limit (dir)) {
454 ext4_warning(dir->i_sb,
455 "dx entry: limit != node limit");
456 brelse(bh);
457 *err = ERR_BAD_DX_DIR;
458 goto fail2;
460 frame++;
461 frame->bh = NULL;
463 fail2:
464 while (frame >= frame_in) {
465 brelse(frame->bh);
466 frame--;
468 fail:
469 if (*err == ERR_BAD_DX_DIR)
470 ext4_warning(dir->i_sb,
471 "Corrupt dir inode %ld, running e2fsck is "
472 "recommended.", dir->i_ino);
473 return NULL;
476 static void dx_release (struct dx_frame *frames)
478 if (frames[0].bh == NULL)
479 return;
481 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
482 brelse(frames[1].bh);
483 brelse(frames[0].bh);
487 * This function increments the frame pointer to search the next leaf
488 * block, and reads in the necessary intervening nodes if the search
489 * should be necessary. Whether or not the search is necessary is
490 * controlled by the hash parameter. If the hash value is even, then
491 * the search is only continued if the next block starts with that
492 * hash value. This is used if we are searching for a specific file.
494 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
496 * This function returns 1 if the caller should continue to search,
497 * or 0 if it should not. If there is an error reading one of the
498 * index blocks, it will a negative error code.
500 * If start_hash is non-null, it will be filled in with the starting
501 * hash of the next page.
503 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
504 struct dx_frame *frame,
505 struct dx_frame *frames,
506 __u32 *start_hash)
508 struct dx_frame *p;
509 struct buffer_head *bh;
510 int err, num_frames = 0;
511 __u32 bhash;
513 p = frame;
515 * Find the next leaf page by incrementing the frame pointer.
516 * If we run out of entries in the interior node, loop around and
517 * increment pointer in the parent node. When we break out of
518 * this loop, num_frames indicates the number of interior
519 * nodes need to be read.
521 while (1) {
522 if (++(p->at) < p->entries + dx_get_count(p->entries))
523 break;
524 if (p == frames)
525 return 0;
526 num_frames++;
527 p--;
531 * If the hash is 1, then continue only if the next page has a
532 * continuation hash of any value. This is used for readdir
533 * handling. Otherwise, check to see if the hash matches the
534 * desired contiuation hash. If it doesn't, return since
535 * there's no point to read in the successive index pages.
537 bhash = dx_get_hash(p->at);
538 if (start_hash)
539 *start_hash = bhash;
540 if ((hash & 1) == 0) {
541 if ((bhash & ~1) != hash)
542 return 0;
545 * If the hash is HASH_NB_ALWAYS, we always go to the next
546 * block so no check is necessary
548 while (num_frames--) {
549 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
550 0, &err)))
551 return err; /* Failure */
552 p++;
553 brelse(p->bh);
554 p->bh = bh;
555 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
557 return 1;
562 * This function fills a red-black tree with information from a
563 * directory block. It returns the number directory entries loaded
564 * into the tree. If there is an error it is returned in err.
566 static int htree_dirblock_to_tree(struct file *dir_file,
567 struct inode *dir, ext4_lblk_t block,
568 struct dx_hash_info *hinfo,
569 __u32 start_hash, __u32 start_minor_hash)
571 struct buffer_head *bh;
572 struct ext4_dir_entry_2 *de, *top;
573 int err, count = 0;
575 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
576 (unsigned long)block));
577 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
578 return err;
580 de = (struct ext4_dir_entry_2 *) bh->b_data;
581 top = (struct ext4_dir_entry_2 *) ((char *) de +
582 dir->i_sb->s_blocksize -
583 EXT4_DIR_REC_LEN(0));
584 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
585 if (ext4_check_dir_entry(dir, NULL, de, bh,
586 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
587 + ((char *)de - bh->b_data))) {
588 /* On error, skip the f_pos to the next block. */
589 dir_file->f_pos = (dir_file->f_pos |
590 (dir->i_sb->s_blocksize - 1)) + 1;
591 brelse(bh);
592 return count;
594 ext4fs_dirhash(de->name, de->name_len, hinfo);
595 if ((hinfo->hash < start_hash) ||
596 ((hinfo->hash == start_hash) &&
597 (hinfo->minor_hash < start_minor_hash)))
598 continue;
599 if (de->inode == 0)
600 continue;
601 if ((err = ext4_htree_store_dirent(dir_file,
602 hinfo->hash, hinfo->minor_hash, de)) != 0) {
603 brelse(bh);
604 return err;
606 count++;
608 brelse(bh);
609 return count;
614 * This function fills a red-black tree with information from a
615 * directory. We start scanning the directory in hash order, starting
616 * at start_hash and start_minor_hash.
618 * This function returns the number of entries inserted into the tree,
619 * or a negative error code.
621 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
622 __u32 start_minor_hash, __u32 *next_hash)
624 struct dx_hash_info hinfo;
625 struct ext4_dir_entry_2 *de;
626 struct dx_frame frames[2], *frame;
627 struct inode *dir;
628 ext4_lblk_t block;
629 int count = 0;
630 int ret, err;
631 __u32 hashval;
633 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
634 start_hash, start_minor_hash));
635 dir = dir_file->f_path.dentry->d_inode;
636 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
637 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
638 if (hinfo.hash_version <= DX_HASH_TEA)
639 hinfo.hash_version +=
640 EXT4_SB(dir->i_sb)->s_hash_unsigned;
641 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
642 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
643 start_hash, start_minor_hash);
644 *next_hash = ~0;
645 return count;
647 hinfo.hash = start_hash;
648 hinfo.minor_hash = 0;
649 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
650 if (!frame)
651 return err;
653 /* Add '.' and '..' from the htree header */
654 if (!start_hash && !start_minor_hash) {
655 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
656 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
657 goto errout;
658 count++;
660 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
661 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
662 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
663 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
664 goto errout;
665 count++;
668 while (1) {
669 block = dx_get_block(frame->at);
670 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
671 start_hash, start_minor_hash);
672 if (ret < 0) {
673 err = ret;
674 goto errout;
676 count += ret;
677 hashval = ~0;
678 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
679 frame, frames, &hashval);
680 *next_hash = hashval;
681 if (ret < 0) {
682 err = ret;
683 goto errout;
686 * Stop if: (a) there are no more entries, or
687 * (b) we have inserted at least one entry and the
688 * next hash value is not a continuation
690 if ((ret == 0) ||
691 (count && ((hashval & 1) == 0)))
692 break;
694 dx_release(frames);
695 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
696 "next hash: %x\n", count, *next_hash));
697 return count;
698 errout:
699 dx_release(frames);
700 return (err);
705 * Directory block splitting, compacting
709 * Create map of hash values, offsets, and sizes, stored at end of block.
710 * Returns number of entries mapped.
712 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
713 struct dx_hash_info *hinfo,
714 struct dx_map_entry *map_tail)
716 int count = 0;
717 char *base = (char *) de;
718 struct dx_hash_info h = *hinfo;
720 while ((char *) de < base + blocksize) {
721 if (de->name_len && de->inode) {
722 ext4fs_dirhash(de->name, de->name_len, &h);
723 map_tail--;
724 map_tail->hash = h.hash;
725 map_tail->offs = ((char *) de - base)>>2;
726 map_tail->size = le16_to_cpu(de->rec_len);
727 count++;
728 cond_resched();
730 /* XXX: do we need to check rec_len == 0 case? -Chris */
731 de = ext4_next_entry(de, blocksize);
733 return count;
736 /* Sort map by hash value */
737 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
739 struct dx_map_entry *p, *q, *top = map + count - 1;
740 int more;
741 /* Combsort until bubble sort doesn't suck */
742 while (count > 2) {
743 count = count*10/13;
744 if (count - 9 < 2) /* 9, 10 -> 11 */
745 count = 11;
746 for (p = top, q = p - count; q >= map; p--, q--)
747 if (p->hash < q->hash)
748 swap(*p, *q);
750 /* Garden variety bubble sort */
751 do {
752 more = 0;
753 q = top;
754 while (q-- > map) {
755 if (q[1].hash >= q[0].hash)
756 continue;
757 swap(*(q+1), *q);
758 more = 1;
760 } while(more);
763 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
765 struct dx_entry *entries = frame->entries;
766 struct dx_entry *old = frame->at, *new = old + 1;
767 int count = dx_get_count(entries);
769 assert(count < dx_get_limit(entries));
770 assert(old < entries + count);
771 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
772 dx_set_hash(new, hash);
773 dx_set_block(new, block);
774 dx_set_count(entries, count + 1);
777 static void ext4_update_dx_flag(struct inode *inode)
779 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
780 EXT4_FEATURE_COMPAT_DIR_INDEX))
781 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
785 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
787 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
788 * `de != NULL' is guaranteed by caller.
790 static inline int ext4_match (int len, const char * const name,
791 struct ext4_dir_entry_2 * de)
793 if (len != de->name_len)
794 return 0;
795 if (!de->inode)
796 return 0;
797 return !memcmp(name, de->name, len);
801 * Returns 0 if not found, -1 on failure, and 1 on success
803 static inline int search_dirblock(struct buffer_head *bh,
804 struct inode *dir,
805 const struct qstr *d_name,
806 unsigned int offset,
807 struct ext4_dir_entry_2 ** res_dir)
809 struct ext4_dir_entry_2 * de;
810 char * dlimit;
811 int de_len;
812 const char *name = d_name->name;
813 int namelen = d_name->len;
815 de = (struct ext4_dir_entry_2 *) bh->b_data;
816 dlimit = bh->b_data + dir->i_sb->s_blocksize;
817 while ((char *) de < dlimit) {
818 /* this code is executed quadratically often */
819 /* do minimal checking `by hand' */
821 if ((char *) de + namelen <= dlimit &&
822 ext4_match (namelen, name, de)) {
823 /* found a match - just to be sure, do a full check */
824 if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
825 return -1;
826 *res_dir = de;
827 return 1;
829 /* prevent looping on a bad block */
830 de_len = ext4_rec_len_from_disk(de->rec_len,
831 dir->i_sb->s_blocksize);
832 if (de_len <= 0)
833 return -1;
834 offset += de_len;
835 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
837 return 0;
842 * ext4_find_entry()
844 * finds an entry in the specified directory with the wanted name. It
845 * returns the cache buffer in which the entry was found, and the entry
846 * itself (as a parameter - res_dir). It does NOT read the inode of the
847 * entry - you'll have to do that yourself if you want to.
849 * The returned buffer_head has ->b_count elevated. The caller is expected
850 * to brelse() it when appropriate.
852 static struct buffer_head * ext4_find_entry (struct inode *dir,
853 const struct qstr *d_name,
854 struct ext4_dir_entry_2 ** res_dir)
856 struct super_block *sb;
857 struct buffer_head *bh_use[NAMEI_RA_SIZE];
858 struct buffer_head *bh, *ret = NULL;
859 ext4_lblk_t start, block, b;
860 const u8 *name = d_name->name;
861 int ra_max = 0; /* Number of bh's in the readahead
862 buffer, bh_use[] */
863 int ra_ptr = 0; /* Current index into readahead
864 buffer */
865 int num = 0;
866 ext4_lblk_t nblocks;
867 int i, err;
868 int namelen;
870 *res_dir = NULL;
871 sb = dir->i_sb;
872 namelen = d_name->len;
873 if (namelen > EXT4_NAME_LEN)
874 return NULL;
875 if ((namelen <= 2) && (name[0] == '.') &&
876 (name[1] == '.' || name[1] == '\0')) {
878 * "." or ".." will only be in the first block
879 * NFS may look up ".."; "." should be handled by the VFS
881 block = start = 0;
882 nblocks = 1;
883 goto restart;
885 if (is_dx(dir)) {
886 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
888 * On success, or if the error was file not found,
889 * return. Otherwise, fall back to doing a search the
890 * old fashioned way.
892 if (bh || (err != ERR_BAD_DX_DIR))
893 return bh;
894 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
895 "falling back\n"));
897 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
898 start = EXT4_I(dir)->i_dir_start_lookup;
899 if (start >= nblocks)
900 start = 0;
901 block = start;
902 restart:
903 do {
905 * We deal with the read-ahead logic here.
907 if (ra_ptr >= ra_max) {
908 /* Refill the readahead buffer */
909 ra_ptr = 0;
910 b = block;
911 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
913 * Terminate if we reach the end of the
914 * directory and must wrap, or if our
915 * search has finished at this block.
917 if (b >= nblocks || (num && block == start)) {
918 bh_use[ra_max] = NULL;
919 break;
921 num++;
922 bh = ext4_getblk(NULL, dir, b++, 0, &err);
923 bh_use[ra_max] = bh;
924 if (bh)
925 ll_rw_block(READ_META, 1, &bh);
928 if ((bh = bh_use[ra_ptr++]) == NULL)
929 goto next;
930 wait_on_buffer(bh);
931 if (!buffer_uptodate(bh)) {
932 /* read error, skip block & hope for the best */
933 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
934 (unsigned long) block);
935 brelse(bh);
936 goto next;
938 i = search_dirblock(bh, dir, d_name,
939 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
940 if (i == 1) {
941 EXT4_I(dir)->i_dir_start_lookup = block;
942 ret = bh;
943 goto cleanup_and_exit;
944 } else {
945 brelse(bh);
946 if (i < 0)
947 goto cleanup_and_exit;
949 next:
950 if (++block >= nblocks)
951 block = 0;
952 } while (block != start);
955 * If the directory has grown while we were searching, then
956 * search the last part of the directory before giving up.
958 block = nblocks;
959 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
960 if (block < nblocks) {
961 start = 0;
962 goto restart;
965 cleanup_and_exit:
966 /* Clean up the read-ahead blocks */
967 for (; ra_ptr < ra_max; ra_ptr++)
968 brelse(bh_use[ra_ptr]);
969 return ret;
972 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
973 struct ext4_dir_entry_2 **res_dir, int *err)
975 struct super_block * sb = dir->i_sb;
976 struct dx_hash_info hinfo;
977 struct dx_frame frames[2], *frame;
978 struct buffer_head *bh;
979 ext4_lblk_t block;
980 int retval;
982 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
983 return NULL;
984 do {
985 block = dx_get_block(frame->at);
986 if (!(bh = ext4_bread(NULL, dir, block, 0, err)))
987 goto errout;
989 retval = search_dirblock(bh, dir, d_name,
990 block << EXT4_BLOCK_SIZE_BITS(sb),
991 res_dir);
992 if (retval == 1) { /* Success! */
993 dx_release(frames);
994 return bh;
996 brelse(bh);
997 if (retval == -1) {
998 *err = ERR_BAD_DX_DIR;
999 goto errout;
1002 /* Check to see if we should continue to search */
1003 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1004 frames, NULL);
1005 if (retval < 0) {
1006 ext4_warning(sb,
1007 "error reading index page in directory #%lu",
1008 dir->i_ino);
1009 *err = retval;
1010 goto errout;
1012 } while (retval == 1);
1014 *err = -ENOENT;
1015 errout:
1016 dxtrace(printk(KERN_DEBUG "%s not found\n", name));
1017 dx_release (frames);
1018 return NULL;
1021 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1023 struct inode *inode;
1024 struct ext4_dir_entry_2 *de;
1025 struct buffer_head *bh;
1027 if (dentry->d_name.len > EXT4_NAME_LEN)
1028 return ERR_PTR(-ENAMETOOLONG);
1030 bh = ext4_find_entry(dir, &dentry->d_name, &de);
1031 inode = NULL;
1032 if (bh) {
1033 __u32 ino = le32_to_cpu(de->inode);
1034 brelse(bh);
1035 if (!ext4_valid_inum(dir->i_sb, ino)) {
1036 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1037 return ERR_PTR(-EIO);
1039 inode = ext4_iget(dir->i_sb, ino);
1040 if (IS_ERR(inode)) {
1041 if (PTR_ERR(inode) == -ESTALE) {
1042 EXT4_ERROR_INODE(dir,
1043 "deleted inode referenced: %u",
1044 ino);
1045 return ERR_PTR(-EIO);
1046 } else {
1047 return ERR_CAST(inode);
1051 return d_splice_alias(inode, dentry);
1055 struct dentry *ext4_get_parent(struct dentry *child)
1057 __u32 ino;
1058 static const struct qstr dotdot = {
1059 .name = "..",
1060 .len = 2,
1062 struct ext4_dir_entry_2 * de;
1063 struct buffer_head *bh;
1065 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1066 if (!bh)
1067 return ERR_PTR(-ENOENT);
1068 ino = le32_to_cpu(de->inode);
1069 brelse(bh);
1071 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1072 EXT4_ERROR_INODE(child->d_inode,
1073 "bad parent inode number: %u", ino);
1074 return ERR_PTR(-EIO);
1077 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1080 #define S_SHIFT 12
1081 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1082 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1083 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1084 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1085 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1086 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1087 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1088 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
1091 static inline void ext4_set_de_type(struct super_block *sb,
1092 struct ext4_dir_entry_2 *de,
1093 umode_t mode) {
1094 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1095 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1099 * Move count entries from end of map between two memory locations.
1100 * Returns pointer to last entry moved.
1102 static struct ext4_dir_entry_2 *
1103 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1104 unsigned blocksize)
1106 unsigned rec_len = 0;
1108 while (count--) {
1109 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1110 (from + (map->offs<<2));
1111 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1112 memcpy (to, de, rec_len);
1113 ((struct ext4_dir_entry_2 *) to)->rec_len =
1114 ext4_rec_len_to_disk(rec_len, blocksize);
1115 de->inode = 0;
1116 map++;
1117 to += rec_len;
1119 return (struct ext4_dir_entry_2 *) (to - rec_len);
1123 * Compact each dir entry in the range to the minimal rec_len.
1124 * Returns pointer to last entry in range.
1126 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1128 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1129 unsigned rec_len = 0;
1131 prev = to = de;
1132 while ((char*)de < base + blocksize) {
1133 next = ext4_next_entry(de, blocksize);
1134 if (de->inode && de->name_len) {
1135 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1136 if (de > to)
1137 memmove(to, de, rec_len);
1138 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1139 prev = to;
1140 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1142 de = next;
1144 return prev;
1148 * Split a full leaf block to make room for a new dir entry.
1149 * Allocate a new block, and move entries so that they are approx. equally full.
1150 * Returns pointer to de in block into which the new entry will be inserted.
1152 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1153 struct buffer_head **bh,struct dx_frame *frame,
1154 struct dx_hash_info *hinfo, int *error)
1156 unsigned blocksize = dir->i_sb->s_blocksize;
1157 unsigned count, continued;
1158 struct buffer_head *bh2;
1159 ext4_lblk_t newblock;
1160 u32 hash2;
1161 struct dx_map_entry *map;
1162 char *data1 = (*bh)->b_data, *data2;
1163 unsigned split, move, size;
1164 struct ext4_dir_entry_2 *de = NULL, *de2;
1165 int err = 0, i;
1167 bh2 = ext4_append (handle, dir, &newblock, &err);
1168 if (!(bh2)) {
1169 brelse(*bh);
1170 *bh = NULL;
1171 goto errout;
1174 BUFFER_TRACE(*bh, "get_write_access");
1175 err = ext4_journal_get_write_access(handle, *bh);
1176 if (err)
1177 goto journal_error;
1179 BUFFER_TRACE(frame->bh, "get_write_access");
1180 err = ext4_journal_get_write_access(handle, frame->bh);
1181 if (err)
1182 goto journal_error;
1184 data2 = bh2->b_data;
1186 /* create map in the end of data2 block */
1187 map = (struct dx_map_entry *) (data2 + blocksize);
1188 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1189 blocksize, hinfo, map);
1190 map -= count;
1191 dx_sort_map(map, count);
1192 /* Split the existing block in the middle, size-wise */
1193 size = 0;
1194 move = 0;
1195 for (i = count-1; i >= 0; i--) {
1196 /* is more than half of this entry in 2nd half of the block? */
1197 if (size + map[i].size/2 > blocksize/2)
1198 break;
1199 size += map[i].size;
1200 move++;
1202 /* map index at which we will split */
1203 split = count - move;
1204 hash2 = map[split].hash;
1205 continued = hash2 == map[split - 1].hash;
1206 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1207 (unsigned long)dx_get_block(frame->at),
1208 hash2, split, count-split));
1210 /* Fancy dance to stay within two buffers */
1211 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1212 de = dx_pack_dirents(data1, blocksize);
1213 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1214 blocksize);
1215 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1216 blocksize);
1217 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1218 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1220 /* Which block gets the new entry? */
1221 if (hinfo->hash >= hash2)
1223 swap(*bh, bh2);
1224 de = de2;
1226 dx_insert_block(frame, hash2 + continued, newblock);
1227 err = ext4_handle_dirty_metadata(handle, dir, bh2);
1228 if (err)
1229 goto journal_error;
1230 err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1231 if (err)
1232 goto journal_error;
1233 brelse(bh2);
1234 dxtrace(dx_show_index("frame", frame->entries));
1235 return de;
1237 journal_error:
1238 brelse(*bh);
1239 brelse(bh2);
1240 *bh = NULL;
1241 ext4_std_error(dir->i_sb, err);
1242 errout:
1243 *error = err;
1244 return NULL;
1248 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1249 * it points to a directory entry which is guaranteed to be large
1250 * enough for new directory entry. If de is NULL, then
1251 * add_dirent_to_buf will attempt search the directory block for
1252 * space. It will return -ENOSPC if no space is available, and -EIO
1253 * and -EEXIST if directory entry already exists.
1255 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1256 struct inode *inode, struct ext4_dir_entry_2 *de,
1257 struct buffer_head *bh)
1259 struct inode *dir = dentry->d_parent->d_inode;
1260 const char *name = dentry->d_name.name;
1261 int namelen = dentry->d_name.len;
1262 unsigned int offset = 0;
1263 unsigned int blocksize = dir->i_sb->s_blocksize;
1264 unsigned short reclen;
1265 int nlen, rlen, err;
1266 char *top;
1268 reclen = EXT4_DIR_REC_LEN(namelen);
1269 if (!de) {
1270 de = (struct ext4_dir_entry_2 *)bh->b_data;
1271 top = bh->b_data + blocksize - reclen;
1272 while ((char *) de <= top) {
1273 if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
1274 return -EIO;
1275 if (ext4_match(namelen, name, de))
1276 return -EEXIST;
1277 nlen = EXT4_DIR_REC_LEN(de->name_len);
1278 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1279 if ((de->inode? rlen - nlen: rlen) >= reclen)
1280 break;
1281 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1282 offset += rlen;
1284 if ((char *) de > top)
1285 return -ENOSPC;
1287 BUFFER_TRACE(bh, "get_write_access");
1288 err = ext4_journal_get_write_access(handle, bh);
1289 if (err) {
1290 ext4_std_error(dir->i_sb, err);
1291 return err;
1294 /* By now the buffer is marked for journaling */
1295 nlen = EXT4_DIR_REC_LEN(de->name_len);
1296 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1297 if (de->inode) {
1298 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1299 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1300 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1301 de = de1;
1303 de->file_type = EXT4_FT_UNKNOWN;
1304 if (inode) {
1305 de->inode = cpu_to_le32(inode->i_ino);
1306 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1307 } else
1308 de->inode = 0;
1309 de->name_len = namelen;
1310 memcpy(de->name, name, namelen);
1312 * XXX shouldn't update any times until successful
1313 * completion of syscall, but too many callers depend
1314 * on this.
1316 * XXX similarly, too many callers depend on
1317 * ext4_new_inode() setting the times, but error
1318 * recovery deletes the inode, so the worst that can
1319 * happen is that the times are slightly out of date
1320 * and/or different from the directory change time.
1322 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1323 ext4_update_dx_flag(dir);
1324 dir->i_version++;
1325 ext4_mark_inode_dirty(handle, dir);
1326 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1327 err = ext4_handle_dirty_metadata(handle, dir, bh);
1328 if (err)
1329 ext4_std_error(dir->i_sb, err);
1330 return 0;
1334 * This converts a one block unindexed directory to a 3 block indexed
1335 * directory, and adds the dentry to the indexed directory.
1337 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1338 struct inode *inode, struct buffer_head *bh)
1340 struct inode *dir = dentry->d_parent->d_inode;
1341 const char *name = dentry->d_name.name;
1342 int namelen = dentry->d_name.len;
1343 struct buffer_head *bh2;
1344 struct dx_root *root;
1345 struct dx_frame frames[2], *frame;
1346 struct dx_entry *entries;
1347 struct ext4_dir_entry_2 *de, *de2;
1348 char *data1, *top;
1349 unsigned len;
1350 int retval;
1351 unsigned blocksize;
1352 struct dx_hash_info hinfo;
1353 ext4_lblk_t block;
1354 struct fake_dirent *fde;
1356 blocksize = dir->i_sb->s_blocksize;
1357 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1358 retval = ext4_journal_get_write_access(handle, bh);
1359 if (retval) {
1360 ext4_std_error(dir->i_sb, retval);
1361 brelse(bh);
1362 return retval;
1364 root = (struct dx_root *) bh->b_data;
1366 /* The 0th block becomes the root, move the dirents out */
1367 fde = &root->dotdot;
1368 de = (struct ext4_dir_entry_2 *)((char *)fde +
1369 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1370 if ((char *) de >= (((char *) root) + blocksize)) {
1371 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1372 brelse(bh);
1373 return -EIO;
1375 len = ((char *) root) + blocksize - (char *) de;
1377 /* Allocate new block for the 0th block's dirents */
1378 bh2 = ext4_append(handle, dir, &block, &retval);
1379 if (!(bh2)) {
1380 brelse(bh);
1381 return retval;
1383 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1384 data1 = bh2->b_data;
1386 memcpy (data1, de, len);
1387 de = (struct ext4_dir_entry_2 *) data1;
1388 top = data1 + len;
1389 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1390 de = de2;
1391 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1392 blocksize);
1393 /* Initialize the root; the dot dirents already exist */
1394 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1395 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1396 blocksize);
1397 memset (&root->info, 0, sizeof(root->info));
1398 root->info.info_length = sizeof(root->info);
1399 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1400 entries = root->entries;
1401 dx_set_block(entries, 1);
1402 dx_set_count(entries, 1);
1403 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1405 /* Initialize as for dx_probe */
1406 hinfo.hash_version = root->info.hash_version;
1407 if (hinfo.hash_version <= DX_HASH_TEA)
1408 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1409 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1410 ext4fs_dirhash(name, namelen, &hinfo);
1411 frame = frames;
1412 frame->entries = entries;
1413 frame->at = entries;
1414 frame->bh = bh;
1415 bh = bh2;
1416 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1417 dx_release (frames);
1418 if (!(de))
1419 return retval;
1421 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1422 brelse(bh);
1423 return retval;
1427 * ext4_add_entry()
1429 * adds a file entry to the specified directory, using the same
1430 * semantics as ext4_find_entry(). It returns NULL if it failed.
1432 * NOTE!! The inode part of 'de' is left at 0 - which means you
1433 * may not sleep between calling this and putting something into
1434 * the entry, as someone else might have used it while you slept.
1436 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1437 struct inode *inode)
1439 struct inode *dir = dentry->d_parent->d_inode;
1440 struct buffer_head *bh;
1441 struct ext4_dir_entry_2 *de;
1442 struct super_block *sb;
1443 int retval;
1444 int dx_fallback=0;
1445 unsigned blocksize;
1446 ext4_lblk_t block, blocks;
1448 sb = dir->i_sb;
1449 blocksize = sb->s_blocksize;
1450 if (!dentry->d_name.len)
1451 return -EINVAL;
1452 if (is_dx(dir)) {
1453 retval = ext4_dx_add_entry(handle, dentry, inode);
1454 if (!retval || (retval != ERR_BAD_DX_DIR))
1455 return retval;
1456 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1457 dx_fallback++;
1458 ext4_mark_inode_dirty(handle, dir);
1460 blocks = dir->i_size >> sb->s_blocksize_bits;
1461 for (block = 0; block < blocks; block++) {
1462 bh = ext4_bread(handle, dir, block, 0, &retval);
1463 if(!bh)
1464 return retval;
1465 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1466 if (retval != -ENOSPC) {
1467 brelse(bh);
1468 return retval;
1471 if (blocks == 1 && !dx_fallback &&
1472 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1473 return make_indexed_dir(handle, dentry, inode, bh);
1474 brelse(bh);
1476 bh = ext4_append(handle, dir, &block, &retval);
1477 if (!bh)
1478 return retval;
1479 de = (struct ext4_dir_entry_2 *) bh->b_data;
1480 de->inode = 0;
1481 de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1482 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1483 brelse(bh);
1484 if (retval == 0)
1485 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1486 return retval;
1490 * Returns 0 for success, or a negative error value
1492 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1493 struct inode *inode)
1495 struct dx_frame frames[2], *frame;
1496 struct dx_entry *entries, *at;
1497 struct dx_hash_info hinfo;
1498 struct buffer_head *bh;
1499 struct inode *dir = dentry->d_parent->d_inode;
1500 struct super_block *sb = dir->i_sb;
1501 struct ext4_dir_entry_2 *de;
1502 int err;
1504 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1505 if (!frame)
1506 return err;
1507 entries = frame->entries;
1508 at = frame->at;
1510 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1511 goto cleanup;
1513 BUFFER_TRACE(bh, "get_write_access");
1514 err = ext4_journal_get_write_access(handle, bh);
1515 if (err)
1516 goto journal_error;
1518 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1519 if (err != -ENOSPC)
1520 goto cleanup;
1522 /* Block full, should compress but for now just split */
1523 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1524 dx_get_count(entries), dx_get_limit(entries)));
1525 /* Need to split index? */
1526 if (dx_get_count(entries) == dx_get_limit(entries)) {
1527 ext4_lblk_t newblock;
1528 unsigned icount = dx_get_count(entries);
1529 int levels = frame - frames;
1530 struct dx_entry *entries2;
1531 struct dx_node *node2;
1532 struct buffer_head *bh2;
1534 if (levels && (dx_get_count(frames->entries) ==
1535 dx_get_limit(frames->entries))) {
1536 ext4_warning(sb, "Directory index full!");
1537 err = -ENOSPC;
1538 goto cleanup;
1540 bh2 = ext4_append (handle, dir, &newblock, &err);
1541 if (!(bh2))
1542 goto cleanup;
1543 node2 = (struct dx_node *)(bh2->b_data);
1544 entries2 = node2->entries;
1545 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1546 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1547 sb->s_blocksize);
1548 BUFFER_TRACE(frame->bh, "get_write_access");
1549 err = ext4_journal_get_write_access(handle, frame->bh);
1550 if (err)
1551 goto journal_error;
1552 if (levels) {
1553 unsigned icount1 = icount/2, icount2 = icount - icount1;
1554 unsigned hash2 = dx_get_hash(entries + icount1);
1555 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1556 icount1, icount2));
1558 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1559 err = ext4_journal_get_write_access(handle,
1560 frames[0].bh);
1561 if (err)
1562 goto journal_error;
1564 memcpy((char *) entries2, (char *) (entries + icount1),
1565 icount2 * sizeof(struct dx_entry));
1566 dx_set_count(entries, icount1);
1567 dx_set_count(entries2, icount2);
1568 dx_set_limit(entries2, dx_node_limit(dir));
1570 /* Which index block gets the new entry? */
1571 if (at - entries >= icount1) {
1572 frame->at = at = at - entries - icount1 + entries2;
1573 frame->entries = entries = entries2;
1574 swap(frame->bh, bh2);
1576 dx_insert_block(frames + 0, hash2, newblock);
1577 dxtrace(dx_show_index("node", frames[1].entries));
1578 dxtrace(dx_show_index("node",
1579 ((struct dx_node *) bh2->b_data)->entries));
1580 err = ext4_handle_dirty_metadata(handle, inode, bh2);
1581 if (err)
1582 goto journal_error;
1583 brelse (bh2);
1584 } else {
1585 dxtrace(printk(KERN_DEBUG
1586 "Creating second level index...\n"));
1587 memcpy((char *) entries2, (char *) entries,
1588 icount * sizeof(struct dx_entry));
1589 dx_set_limit(entries2, dx_node_limit(dir));
1591 /* Set up root */
1592 dx_set_count(entries, 1);
1593 dx_set_block(entries + 0, newblock);
1594 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1596 /* Add new access path frame */
1597 frame = frames + 1;
1598 frame->at = at = at - entries + entries2;
1599 frame->entries = entries = entries2;
1600 frame->bh = bh2;
1601 err = ext4_journal_get_write_access(handle,
1602 frame->bh);
1603 if (err)
1604 goto journal_error;
1606 err = ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
1607 if (err) {
1608 ext4_std_error(inode->i_sb, err);
1609 goto cleanup;
1612 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1613 if (!de)
1614 goto cleanup;
1615 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1616 goto cleanup;
1618 journal_error:
1619 ext4_std_error(dir->i_sb, err);
1620 cleanup:
1621 if (bh)
1622 brelse(bh);
1623 dx_release(frames);
1624 return err;
1628 * ext4_delete_entry deletes a directory entry by merging it with the
1629 * previous entry
1631 static int ext4_delete_entry(handle_t *handle,
1632 struct inode *dir,
1633 struct ext4_dir_entry_2 *de_del,
1634 struct buffer_head *bh)
1636 struct ext4_dir_entry_2 *de, *pde;
1637 unsigned int blocksize = dir->i_sb->s_blocksize;
1638 int i, err;
1640 i = 0;
1641 pde = NULL;
1642 de = (struct ext4_dir_entry_2 *) bh->b_data;
1643 while (i < bh->b_size) {
1644 if (ext4_check_dir_entry(dir, NULL, de, bh, i))
1645 return -EIO;
1646 if (de == de_del) {
1647 BUFFER_TRACE(bh, "get_write_access");
1648 err = ext4_journal_get_write_access(handle, bh);
1649 if (unlikely(err)) {
1650 ext4_std_error(dir->i_sb, err);
1651 return err;
1653 if (pde)
1654 pde->rec_len = ext4_rec_len_to_disk(
1655 ext4_rec_len_from_disk(pde->rec_len,
1656 blocksize) +
1657 ext4_rec_len_from_disk(de->rec_len,
1658 blocksize),
1659 blocksize);
1660 else
1661 de->inode = 0;
1662 dir->i_version++;
1663 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1664 err = ext4_handle_dirty_metadata(handle, dir, bh);
1665 if (unlikely(err)) {
1666 ext4_std_error(dir->i_sb, err);
1667 return err;
1669 return 0;
1671 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
1672 pde = de;
1673 de = ext4_next_entry(de, blocksize);
1675 return -ENOENT;
1679 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1680 * since this indicates that nlinks count was previously 1.
1682 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1684 inc_nlink(inode);
1685 if (is_dx(inode) && inode->i_nlink > 1) {
1686 /* limit is 16-bit i_links_count */
1687 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1688 inode->i_nlink = 1;
1689 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1690 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1696 * If a directory had nlink == 1, then we should let it be 1. This indicates
1697 * directory has >EXT4_LINK_MAX subdirs.
1699 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1701 drop_nlink(inode);
1702 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1703 inc_nlink(inode);
1707 static int ext4_add_nondir(handle_t *handle,
1708 struct dentry *dentry, struct inode *inode)
1710 int err = ext4_add_entry(handle, dentry, inode);
1711 if (!err) {
1712 ext4_mark_inode_dirty(handle, inode);
1713 d_instantiate(dentry, inode);
1714 unlock_new_inode(inode);
1715 return 0;
1717 drop_nlink(inode);
1718 unlock_new_inode(inode);
1719 iput(inode);
1720 return err;
1724 * By the time this is called, we already have created
1725 * the directory cache entry for the new file, but it
1726 * is so far negative - it has no inode.
1728 * If the create succeeds, we fill in the inode information
1729 * with d_instantiate().
1731 static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1732 struct nameidata *nd)
1734 handle_t *handle;
1735 struct inode *inode;
1736 int err, retries = 0;
1738 dquot_initialize(dir);
1740 retry:
1741 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1742 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1743 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1744 if (IS_ERR(handle))
1745 return PTR_ERR(handle);
1747 if (IS_DIRSYNC(dir))
1748 ext4_handle_sync(handle);
1750 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1751 err = PTR_ERR(inode);
1752 if (!IS_ERR(inode)) {
1753 inode->i_op = &ext4_file_inode_operations;
1754 inode->i_fop = &ext4_file_operations;
1755 ext4_set_aops(inode);
1756 err = ext4_add_nondir(handle, dentry, inode);
1758 ext4_journal_stop(handle);
1759 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1760 goto retry;
1761 return err;
1764 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1765 int mode, dev_t rdev)
1767 handle_t *handle;
1768 struct inode *inode;
1769 int err, retries = 0;
1771 if (!new_valid_dev(rdev))
1772 return -EINVAL;
1774 dquot_initialize(dir);
1776 retry:
1777 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1778 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1779 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1780 if (IS_ERR(handle))
1781 return PTR_ERR(handle);
1783 if (IS_DIRSYNC(dir))
1784 ext4_handle_sync(handle);
1786 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1787 err = PTR_ERR(inode);
1788 if (!IS_ERR(inode)) {
1789 init_special_inode(inode, inode->i_mode, rdev);
1790 #ifdef CONFIG_EXT4_FS_XATTR
1791 inode->i_op = &ext4_special_inode_operations;
1792 #endif
1793 err = ext4_add_nondir(handle, dentry, inode);
1795 ext4_journal_stop(handle);
1796 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1797 goto retry;
1798 return err;
1801 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1803 handle_t *handle;
1804 struct inode *inode;
1805 struct buffer_head *dir_block = NULL;
1806 struct ext4_dir_entry_2 *de;
1807 unsigned int blocksize = dir->i_sb->s_blocksize;
1808 int err, retries = 0;
1810 if (EXT4_DIR_LINK_MAX(dir))
1811 return -EMLINK;
1813 dquot_initialize(dir);
1815 retry:
1816 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1817 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1818 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1819 if (IS_ERR(handle))
1820 return PTR_ERR(handle);
1822 if (IS_DIRSYNC(dir))
1823 ext4_handle_sync(handle);
1825 inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1826 &dentry->d_name, 0);
1827 err = PTR_ERR(inode);
1828 if (IS_ERR(inode))
1829 goto out_stop;
1831 inode->i_op = &ext4_dir_inode_operations;
1832 inode->i_fop = &ext4_dir_operations;
1833 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1834 dir_block = ext4_bread(handle, inode, 0, 1, &err);
1835 if (!dir_block)
1836 goto out_clear_inode;
1837 BUFFER_TRACE(dir_block, "get_write_access");
1838 err = ext4_journal_get_write_access(handle, dir_block);
1839 if (err)
1840 goto out_clear_inode;
1841 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1842 de->inode = cpu_to_le32(inode->i_ino);
1843 de->name_len = 1;
1844 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1845 blocksize);
1846 strcpy(de->name, ".");
1847 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1848 de = ext4_next_entry(de, blocksize);
1849 de->inode = cpu_to_le32(dir->i_ino);
1850 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1851 blocksize);
1852 de->name_len = 2;
1853 strcpy(de->name, "..");
1854 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1855 inode->i_nlink = 2;
1856 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1857 err = ext4_handle_dirty_metadata(handle, dir, dir_block);
1858 if (err)
1859 goto out_clear_inode;
1860 err = ext4_mark_inode_dirty(handle, inode);
1861 if (!err)
1862 err = ext4_add_entry(handle, dentry, inode);
1863 if (err) {
1864 out_clear_inode:
1865 clear_nlink(inode);
1866 unlock_new_inode(inode);
1867 ext4_mark_inode_dirty(handle, inode);
1868 iput(inode);
1869 goto out_stop;
1871 ext4_inc_count(handle, dir);
1872 ext4_update_dx_flag(dir);
1873 err = ext4_mark_inode_dirty(handle, dir);
1874 if (err)
1875 goto out_clear_inode;
1876 d_instantiate(dentry, inode);
1877 unlock_new_inode(inode);
1878 out_stop:
1879 brelse(dir_block);
1880 ext4_journal_stop(handle);
1881 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1882 goto retry;
1883 return err;
1887 * routine to check that the specified directory is empty (for rmdir)
1889 static int empty_dir(struct inode *inode)
1891 unsigned int offset;
1892 struct buffer_head *bh;
1893 struct ext4_dir_entry_2 *de, *de1;
1894 struct super_block *sb;
1895 int err = 0;
1897 sb = inode->i_sb;
1898 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1899 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1900 if (err)
1901 EXT4_ERROR_INODE(inode,
1902 "error %d reading directory lblock 0", err);
1903 else
1904 ext4_warning(inode->i_sb,
1905 "bad directory (dir #%lu) - no data block",
1906 inode->i_ino);
1907 return 1;
1909 de = (struct ext4_dir_entry_2 *) bh->b_data;
1910 de1 = ext4_next_entry(de, sb->s_blocksize);
1911 if (le32_to_cpu(de->inode) != inode->i_ino ||
1912 !le32_to_cpu(de1->inode) ||
1913 strcmp(".", de->name) ||
1914 strcmp("..", de1->name)) {
1915 ext4_warning(inode->i_sb,
1916 "bad directory (dir #%lu) - no `.' or `..'",
1917 inode->i_ino);
1918 brelse(bh);
1919 return 1;
1921 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1922 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1923 de = ext4_next_entry(de1, sb->s_blocksize);
1924 while (offset < inode->i_size) {
1925 if (!bh ||
1926 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1927 unsigned int lblock;
1928 err = 0;
1929 brelse(bh);
1930 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1931 bh = ext4_bread(NULL, inode, lblock, 0, &err);
1932 if (!bh) {
1933 if (err)
1934 EXT4_ERROR_INODE(inode,
1935 "error %d reading directory "
1936 "lblock %u", err, lblock);
1937 offset += sb->s_blocksize;
1938 continue;
1940 de = (struct ext4_dir_entry_2 *) bh->b_data;
1942 if (ext4_check_dir_entry(inode, NULL, de, bh, offset)) {
1943 de = (struct ext4_dir_entry_2 *)(bh->b_data +
1944 sb->s_blocksize);
1945 offset = (offset | (sb->s_blocksize - 1)) + 1;
1946 continue;
1948 if (le32_to_cpu(de->inode)) {
1949 brelse(bh);
1950 return 0;
1952 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1953 de = ext4_next_entry(de, sb->s_blocksize);
1955 brelse(bh);
1956 return 1;
1959 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1960 * such inodes, starting at the superblock, in case we crash before the
1961 * file is closed/deleted, or in case the inode truncate spans multiple
1962 * transactions and the last transaction is not recovered after a crash.
1964 * At filesystem recovery time, we walk this list deleting unlinked
1965 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1967 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1969 struct super_block *sb = inode->i_sb;
1970 struct ext4_iloc iloc;
1971 int err = 0, rc;
1973 if (!ext4_handle_valid(handle))
1974 return 0;
1976 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
1977 if (!list_empty(&EXT4_I(inode)->i_orphan))
1978 goto out_unlock;
1980 /* Orphan handling is only valid for files with data blocks
1981 * being truncated, or files being unlinked. */
1983 /* @@@ FIXME: Observation from aviro:
1984 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
1985 * here (on s_orphan_lock), so race with ext4_link() which might bump
1986 * ->i_nlink. For, say it, character device. Not a regular file,
1987 * not a directory, not a symlink and ->i_nlink > 0.
1989 * tytso, 4/25/2009: I'm not sure how that could happen;
1990 * shouldn't the fs core protect us from these sort of
1991 * unlink()/link() races?
1993 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1994 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1996 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1997 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1998 if (err)
1999 goto out_unlock;
2001 err = ext4_reserve_inode_write(handle, inode, &iloc);
2002 if (err)
2003 goto out_unlock;
2005 * Due to previous errors inode may be already a part of on-disk
2006 * orphan list. If so skip on-disk list modification.
2008 if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2009 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2010 goto mem_insert;
2012 /* Insert this inode at the head of the on-disk orphan list... */
2013 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2014 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2015 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
2016 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2017 if (!err)
2018 err = rc;
2020 /* Only add to the head of the in-memory list if all the
2021 * previous operations succeeded. If the orphan_add is going to
2022 * fail (possibly taking the journal offline), we can't risk
2023 * leaving the inode on the orphan list: stray orphan-list
2024 * entries can cause panics at unmount time.
2026 * This is safe: on error we're going to ignore the orphan list
2027 * anyway on the next recovery. */
2028 mem_insert:
2029 if (!err)
2030 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2032 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2033 jbd_debug(4, "orphan inode %lu will point to %d\n",
2034 inode->i_ino, NEXT_ORPHAN(inode));
2035 out_unlock:
2036 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2037 ext4_std_error(inode->i_sb, err);
2038 return err;
2042 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2043 * of such inodes stored on disk, because it is finally being cleaned up.
2045 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2047 struct list_head *prev;
2048 struct ext4_inode_info *ei = EXT4_I(inode);
2049 struct ext4_sb_info *sbi;
2050 __u32 ino_next;
2051 struct ext4_iloc iloc;
2052 int err = 0;
2054 /* ext4_handle_valid() assumes a valid handle_t pointer */
2055 if (handle && !ext4_handle_valid(handle))
2056 return 0;
2058 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2059 if (list_empty(&ei->i_orphan))
2060 goto out;
2062 ino_next = NEXT_ORPHAN(inode);
2063 prev = ei->i_orphan.prev;
2064 sbi = EXT4_SB(inode->i_sb);
2066 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2068 list_del_init(&ei->i_orphan);
2070 /* If we're on an error path, we may not have a valid
2071 * transaction handle with which to update the orphan list on
2072 * disk, but we still need to remove the inode from the linked
2073 * list in memory. */
2074 if (sbi->s_journal && !handle)
2075 goto out;
2077 err = ext4_reserve_inode_write(handle, inode, &iloc);
2078 if (err)
2079 goto out_err;
2081 if (prev == &sbi->s_orphan) {
2082 jbd_debug(4, "superblock will point to %u\n", ino_next);
2083 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2084 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2085 if (err)
2086 goto out_brelse;
2087 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2088 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
2089 } else {
2090 struct ext4_iloc iloc2;
2091 struct inode *i_prev =
2092 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2094 jbd_debug(4, "orphan inode %lu will point to %u\n",
2095 i_prev->i_ino, ino_next);
2096 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2097 if (err)
2098 goto out_brelse;
2099 NEXT_ORPHAN(i_prev) = ino_next;
2100 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2102 if (err)
2103 goto out_brelse;
2104 NEXT_ORPHAN(inode) = 0;
2105 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2107 out_err:
2108 ext4_std_error(inode->i_sb, err);
2109 out:
2110 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2111 return err;
2113 out_brelse:
2114 brelse(iloc.bh);
2115 goto out_err;
2118 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2120 int retval;
2121 struct inode *inode;
2122 struct buffer_head *bh;
2123 struct ext4_dir_entry_2 *de;
2124 handle_t *handle;
2126 /* Initialize quotas before so that eventual writes go in
2127 * separate transaction */
2128 dquot_initialize(dir);
2129 dquot_initialize(dentry->d_inode);
2131 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2132 if (IS_ERR(handle))
2133 return PTR_ERR(handle);
2135 retval = -ENOENT;
2136 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2137 if (!bh)
2138 goto end_rmdir;
2140 if (IS_DIRSYNC(dir))
2141 ext4_handle_sync(handle);
2143 inode = dentry->d_inode;
2145 retval = -EIO;
2146 if (le32_to_cpu(de->inode) != inode->i_ino)
2147 goto end_rmdir;
2149 retval = -ENOTEMPTY;
2150 if (!empty_dir(inode))
2151 goto end_rmdir;
2153 retval = ext4_delete_entry(handle, dir, de, bh);
2154 if (retval)
2155 goto end_rmdir;
2156 if (!EXT4_DIR_LINK_EMPTY(inode))
2157 ext4_warning(inode->i_sb,
2158 "empty directory has too many links (%d)",
2159 inode->i_nlink);
2160 inode->i_version++;
2161 clear_nlink(inode);
2162 /* There's no need to set i_disksize: the fact that i_nlink is
2163 * zero will ensure that the right thing happens during any
2164 * recovery. */
2165 inode->i_size = 0;
2166 ext4_orphan_add(handle, inode);
2167 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2168 ext4_mark_inode_dirty(handle, inode);
2169 ext4_dec_count(handle, dir);
2170 ext4_update_dx_flag(dir);
2171 ext4_mark_inode_dirty(handle, dir);
2173 end_rmdir:
2174 ext4_journal_stop(handle);
2175 brelse(bh);
2176 return retval;
2179 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2181 int retval;
2182 struct inode *inode;
2183 struct buffer_head *bh;
2184 struct ext4_dir_entry_2 *de;
2185 handle_t *handle;
2187 trace_ext4_unlink_enter(dir, dentry);
2188 /* Initialize quotas before so that eventual writes go
2189 * in separate transaction */
2190 dquot_initialize(dir);
2191 dquot_initialize(dentry->d_inode);
2193 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2194 if (IS_ERR(handle))
2195 return PTR_ERR(handle);
2197 if (IS_DIRSYNC(dir))
2198 ext4_handle_sync(handle);
2200 retval = -ENOENT;
2201 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2202 if (!bh)
2203 goto end_unlink;
2205 inode = dentry->d_inode;
2207 retval = -EIO;
2208 if (le32_to_cpu(de->inode) != inode->i_ino)
2209 goto end_unlink;
2211 if (!inode->i_nlink) {
2212 ext4_warning(inode->i_sb,
2213 "Deleting nonexistent file (%lu), %d",
2214 inode->i_ino, inode->i_nlink);
2215 inode->i_nlink = 1;
2217 retval = ext4_delete_entry(handle, dir, de, bh);
2218 if (retval)
2219 goto end_unlink;
2220 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2221 ext4_update_dx_flag(dir);
2222 ext4_mark_inode_dirty(handle, dir);
2223 drop_nlink(inode);
2224 if (!inode->i_nlink)
2225 ext4_orphan_add(handle, inode);
2226 inode->i_ctime = ext4_current_time(inode);
2227 ext4_mark_inode_dirty(handle, inode);
2228 retval = 0;
2230 end_unlink:
2231 ext4_journal_stop(handle);
2232 brelse(bh);
2233 trace_ext4_unlink_exit(dentry, retval);
2234 return retval;
2237 static int ext4_symlink(struct inode *dir,
2238 struct dentry *dentry, const char *symname)
2240 handle_t *handle;
2241 struct inode *inode;
2242 int l, err, retries = 0;
2244 l = strlen(symname)+1;
2245 if (l > dir->i_sb->s_blocksize)
2246 return -ENAMETOOLONG;
2248 dquot_initialize(dir);
2250 retry:
2251 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2252 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2253 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2254 if (IS_ERR(handle))
2255 return PTR_ERR(handle);
2257 if (IS_DIRSYNC(dir))
2258 ext4_handle_sync(handle);
2260 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2261 &dentry->d_name, 0);
2262 err = PTR_ERR(inode);
2263 if (IS_ERR(inode))
2264 goto out_stop;
2266 if (l > sizeof(EXT4_I(inode)->i_data)) {
2267 inode->i_op = &ext4_symlink_inode_operations;
2268 ext4_set_aops(inode);
2270 * page_symlink() calls into ext4_prepare/commit_write.
2271 * We have a transaction open. All is sweetness. It also sets
2272 * i_size in generic_commit_write().
2274 err = __page_symlink(inode, symname, l, 1);
2275 if (err) {
2276 clear_nlink(inode);
2277 unlock_new_inode(inode);
2278 ext4_mark_inode_dirty(handle, inode);
2279 iput(inode);
2280 goto out_stop;
2282 } else {
2283 /* clear the extent format for fast symlink */
2284 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2285 inode->i_op = &ext4_fast_symlink_inode_operations;
2286 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2287 inode->i_size = l-1;
2289 EXT4_I(inode)->i_disksize = inode->i_size;
2290 err = ext4_add_nondir(handle, dentry, inode);
2291 out_stop:
2292 ext4_journal_stop(handle);
2293 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2294 goto retry;
2295 return err;
2298 static int ext4_link(struct dentry *old_dentry,
2299 struct inode *dir, struct dentry *dentry)
2301 handle_t *handle;
2302 struct inode *inode = old_dentry->d_inode;
2303 int err, retries = 0;
2305 if (inode->i_nlink >= EXT4_LINK_MAX)
2306 return -EMLINK;
2308 dquot_initialize(dir);
2310 retry:
2311 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2312 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2313 if (IS_ERR(handle))
2314 return PTR_ERR(handle);
2316 if (IS_DIRSYNC(dir))
2317 ext4_handle_sync(handle);
2319 inode->i_ctime = ext4_current_time(inode);
2320 ext4_inc_count(handle, inode);
2321 ihold(inode);
2323 err = ext4_add_entry(handle, dentry, inode);
2324 if (!err) {
2325 ext4_mark_inode_dirty(handle, inode);
2326 d_instantiate(dentry, inode);
2327 } else {
2328 drop_nlink(inode);
2329 iput(inode);
2331 ext4_journal_stop(handle);
2332 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2333 goto retry;
2334 return err;
2337 #define PARENT_INO(buffer, size) \
2338 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2341 * Anybody can rename anything with this: the permission checks are left to the
2342 * higher-level routines.
2344 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2345 struct inode *new_dir, struct dentry *new_dentry)
2347 handle_t *handle;
2348 struct inode *old_inode, *new_inode;
2349 struct buffer_head *old_bh, *new_bh, *dir_bh;
2350 struct ext4_dir_entry_2 *old_de, *new_de;
2351 int retval, force_da_alloc = 0;
2353 dquot_initialize(old_dir);
2354 dquot_initialize(new_dir);
2356 old_bh = new_bh = dir_bh = NULL;
2358 /* Initialize quotas before so that eventual writes go
2359 * in separate transaction */
2360 if (new_dentry->d_inode)
2361 dquot_initialize(new_dentry->d_inode);
2362 handle = ext4_journal_start(old_dir, 2 *
2363 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2364 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2365 if (IS_ERR(handle))
2366 return PTR_ERR(handle);
2368 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2369 ext4_handle_sync(handle);
2371 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2373 * Check for inode number is _not_ due to possible IO errors.
2374 * We might rmdir the source, keep it as pwd of some process
2375 * and merrily kill the link to whatever was created under the
2376 * same name. Goodbye sticky bit ;-<
2378 old_inode = old_dentry->d_inode;
2379 retval = -ENOENT;
2380 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2381 goto end_rename;
2383 new_inode = new_dentry->d_inode;
2384 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2385 if (new_bh) {
2386 if (!new_inode) {
2387 brelse(new_bh);
2388 new_bh = NULL;
2391 if (S_ISDIR(old_inode->i_mode)) {
2392 if (new_inode) {
2393 retval = -ENOTEMPTY;
2394 if (!empty_dir(new_inode))
2395 goto end_rename;
2397 retval = -EIO;
2398 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2399 if (!dir_bh)
2400 goto end_rename;
2401 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2402 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2403 goto end_rename;
2404 retval = -EMLINK;
2405 if (!new_inode && new_dir != old_dir &&
2406 EXT4_DIR_LINK_MAX(new_dir))
2407 goto end_rename;
2408 BUFFER_TRACE(dir_bh, "get_write_access");
2409 retval = ext4_journal_get_write_access(handle, dir_bh);
2410 if (retval)
2411 goto end_rename;
2413 if (!new_bh) {
2414 retval = ext4_add_entry(handle, new_dentry, old_inode);
2415 if (retval)
2416 goto end_rename;
2417 } else {
2418 BUFFER_TRACE(new_bh, "get write access");
2419 retval = ext4_journal_get_write_access(handle, new_bh);
2420 if (retval)
2421 goto end_rename;
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 retval = ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2432 if (unlikely(retval)) {
2433 ext4_std_error(new_dir->i_sb, retval);
2434 goto end_rename;
2436 brelse(new_bh);
2437 new_bh = NULL;
2441 * Like most other Unix systems, set the ctime for inodes on a
2442 * rename.
2444 old_inode->i_ctime = ext4_current_time(old_inode);
2445 ext4_mark_inode_dirty(handle, old_inode);
2448 * ok, that's it
2450 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2451 old_de->name_len != old_dentry->d_name.len ||
2452 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2453 (retval = ext4_delete_entry(handle, old_dir,
2454 old_de, old_bh)) == -ENOENT) {
2455 /* old_de could have moved from under us during htree split, so
2456 * make sure that we are deleting the right entry. We might
2457 * also be pointing to a stale entry in the unused part of
2458 * old_bh so just checking inum and the name isn't enough. */
2459 struct buffer_head *old_bh2;
2460 struct ext4_dir_entry_2 *old_de2;
2462 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2463 if (old_bh2) {
2464 retval = ext4_delete_entry(handle, old_dir,
2465 old_de2, old_bh2);
2466 brelse(old_bh2);
2469 if (retval) {
2470 ext4_warning(old_dir->i_sb,
2471 "Deleting old file (%lu), %d, error=%d",
2472 old_dir->i_ino, old_dir->i_nlink, retval);
2475 if (new_inode) {
2476 ext4_dec_count(handle, new_inode);
2477 new_inode->i_ctime = ext4_current_time(new_inode);
2479 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2480 ext4_update_dx_flag(old_dir);
2481 if (dir_bh) {
2482 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2483 cpu_to_le32(new_dir->i_ino);
2484 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2485 retval = ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2486 if (retval) {
2487 ext4_std_error(old_dir->i_sb, retval);
2488 goto end_rename;
2490 ext4_dec_count(handle, old_dir);
2491 if (new_inode) {
2492 /* checked empty_dir above, can't have another parent,
2493 * ext4_dec_count() won't work for many-linked dirs */
2494 new_inode->i_nlink = 0;
2495 } else {
2496 ext4_inc_count(handle, new_dir);
2497 ext4_update_dx_flag(new_dir);
2498 ext4_mark_inode_dirty(handle, new_dir);
2501 ext4_mark_inode_dirty(handle, old_dir);
2502 if (new_inode) {
2503 ext4_mark_inode_dirty(handle, new_inode);
2504 if (!new_inode->i_nlink)
2505 ext4_orphan_add(handle, new_inode);
2506 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2507 force_da_alloc = 1;
2509 retval = 0;
2511 end_rename:
2512 brelse(dir_bh);
2513 brelse(old_bh);
2514 brelse(new_bh);
2515 ext4_journal_stop(handle);
2516 if (retval == 0 && force_da_alloc)
2517 ext4_alloc_da_blocks(old_inode);
2518 return retval;
2522 * directories can handle most operations...
2524 const struct inode_operations ext4_dir_inode_operations = {
2525 .create = ext4_create,
2526 .lookup = ext4_lookup,
2527 .link = ext4_link,
2528 .unlink = ext4_unlink,
2529 .symlink = ext4_symlink,
2530 .mkdir = ext4_mkdir,
2531 .rmdir = ext4_rmdir,
2532 .mknod = ext4_mknod,
2533 .rename = ext4_rename,
2534 .setattr = ext4_setattr,
2535 #ifdef CONFIG_EXT4_FS_XATTR
2536 .setxattr = generic_setxattr,
2537 .getxattr = generic_getxattr,
2538 .listxattr = ext4_listxattr,
2539 .removexattr = generic_removexattr,
2540 #endif
2541 .check_acl = ext4_check_acl,
2542 .fiemap = ext4_fiemap,
2545 const struct inode_operations ext4_special_inode_operations = {
2546 .setattr = ext4_setattr,
2547 #ifdef CONFIG_EXT4_FS_XATTR
2548 .setxattr = generic_setxattr,
2549 .getxattr = generic_getxattr,
2550 .listxattr = ext4_listxattr,
2551 .removexattr = generic_removexattr,
2552 #endif
2553 .check_acl = ext4_check_acl,