xref: /openbmc/linux/fs/ext4/namei.c (revision c21b37f6)
1 /*
2  *  linux/fs/ext4/namei.c
3  *
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)
8  *
9  *  from
10  *
11  *  linux/fs/minix/namei.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
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
25  */
26 
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/ext4_fs.h>
32 #include <linux/ext4_jbd2.h>
33 #include <linux/fcntl.h>
34 #include <linux/stat.h>
35 #include <linux/string.h>
36 #include <linux/quotaops.h>
37 #include <linux/buffer_head.h>
38 #include <linux/bio.h>
39 
40 #include "namei.h"
41 #include "xattr.h"
42 #include "acl.h"
43 
44 /*
45  * define how far ahead to read directories while searching them.
46  */
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))
51 
52 static struct buffer_head *ext4_append(handle_t *handle,
53 					struct inode *inode,
54 					u32 *block, int *err)
55 {
56 	struct buffer_head *bh;
57 
58 	*block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59 
60 	if ((bh = ext4_bread(handle, inode, *block, 1, err))) {
61 		inode->i_size += inode->i_sb->s_blocksize;
62 		EXT4_I(inode)->i_disksize = inode->i_size;
63 		ext4_journal_get_write_access(handle,bh);
64 	}
65 	return bh;
66 }
67 
68 #ifndef assert
69 #define assert(test) J_ASSERT(test)
70 #endif
71 
72 #ifndef swap
73 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
74 #endif
75 
76 #ifdef DX_DEBUG
77 #define dxtrace(command) command
78 #else
79 #define dxtrace(command)
80 #endif
81 
82 struct fake_dirent
83 {
84 	__le32 inode;
85 	__le16 rec_len;
86 	u8 name_len;
87 	u8 file_type;
88 };
89 
90 struct dx_countlimit
91 {
92 	__le16 limit;
93 	__le16 count;
94 };
95 
96 struct dx_entry
97 {
98 	__le32 hash;
99 	__le32 block;
100 };
101 
102 /*
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.
106  */
107 
108 struct dx_root
109 {
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
115 	{
116 		__le32 reserved_zero;
117 		u8 hash_version;
118 		u8 info_length; /* 8 */
119 		u8 indirect_levels;
120 		u8 unused_flags;
121 	}
122 	info;
123 	struct dx_entry	entries[0];
124 };
125 
126 struct dx_node
127 {
128 	struct fake_dirent fake;
129 	struct dx_entry	entries[0];
130 };
131 
132 
133 struct dx_frame
134 {
135 	struct buffer_head *bh;
136 	struct dx_entry *entries;
137 	struct dx_entry *at;
138 };
139 
140 struct dx_map_entry
141 {
142 	u32 hash;
143 	u32 offs;
144 };
145 
146 #ifdef CONFIG_EXT4_INDEX
147 static inline unsigned dx_get_block (struct dx_entry *entry);
148 static void dx_set_block (struct dx_entry *entry, unsigned 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(struct dentry *dentry,
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, int size,
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);
168 static struct ext4_dir_entry_2* dx_pack_dirents (char *base, int size);
169 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
170 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
171 				 struct dx_frame *frame,
172 				 struct dx_frame *frames,
173 				 __u32 *start_hash);
174 static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
175 		       struct ext4_dir_entry_2 **res_dir, int *err);
176 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
177 			     struct inode *inode);
178 
179 /*
180  * Future: use high four bits of block for coalesce-on-delete flags
181  * Mask them off for now.
182  */
183 
184 static inline unsigned dx_get_block (struct dx_entry *entry)
185 {
186 	return le32_to_cpu(entry->block) & 0x00ffffff;
187 }
188 
189 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
190 {
191 	entry->block = cpu_to_le32(value);
192 }
193 
194 static inline unsigned dx_get_hash (struct dx_entry *entry)
195 {
196 	return le32_to_cpu(entry->hash);
197 }
198 
199 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
200 {
201 	entry->hash = cpu_to_le32(value);
202 }
203 
204 static inline unsigned dx_get_count (struct dx_entry *entries)
205 {
206 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
207 }
208 
209 static inline unsigned dx_get_limit (struct dx_entry *entries)
210 {
211 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
212 }
213 
214 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
215 {
216 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
217 }
218 
219 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
220 {
221 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
222 }
223 
224 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
225 {
226 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
227 		EXT4_DIR_REC_LEN(2) - infosize;
228 	return 0? 20: entry_space / sizeof(struct dx_entry);
229 }
230 
231 static inline unsigned dx_node_limit (struct inode *dir)
232 {
233 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
234 	return 0? 22: entry_space / sizeof(struct dx_entry);
235 }
236 
237 /*
238  * Debug
239  */
240 #ifdef DX_DEBUG
241 static void dx_show_index (char * label, struct dx_entry *entries)
242 {
243 	int i, n = dx_get_count (entries);
244 	printk("%s index ", label);
245 	for (i = 0; i < n; i++) {
246 		printk("%x->%u ", i? dx_get_hash(entries + i) :
247 				0, dx_get_block(entries + i));
248 	}
249 	printk("\n");
250 }
251 
252 struct stats
253 {
254 	unsigned names;
255 	unsigned space;
256 	unsigned bcount;
257 };
258 
259 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
260 				 int size, int show_names)
261 {
262 	unsigned names = 0, space = 0;
263 	char *base = (char *) de;
264 	struct dx_hash_info h = *hinfo;
265 
266 	printk("names: ");
267 	while ((char *) de < base + size)
268 	{
269 		if (de->inode)
270 		{
271 			if (show_names)
272 			{
273 				int len = de->name_len;
274 				char *name = de->name;
275 				while (len--) printk("%c", *name++);
276 				ext4fs_dirhash(de->name, de->name_len, &h);
277 				printk(":%x.%u ", h.hash,
278 				       ((char *) de - base));
279 			}
280 			space += EXT4_DIR_REC_LEN(de->name_len);
281 			names++;
282 		}
283 		de = (struct ext4_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
284 	}
285 	printk("(%i)\n", names);
286 	return (struct stats) { names, space, 1 };
287 }
288 
289 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
290 			     struct dx_entry *entries, int levels)
291 {
292 	unsigned blocksize = dir->i_sb->s_blocksize;
293 	unsigned count = dx_get_count (entries), names = 0, space = 0, i;
294 	unsigned bcount = 0;
295 	struct buffer_head *bh;
296 	int err;
297 	printk("%i indexed blocks...\n", count);
298 	for (i = 0; i < count; i++, entries++)
299 	{
300 		u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
301 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
302 		struct stats stats;
303 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
304 		if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
305 		stats = levels?
306 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
307 		   dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
308 		names += stats.names;
309 		space += stats.space;
310 		bcount += stats.bcount;
311 		brelse (bh);
312 	}
313 	if (bcount)
314 		printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
315 			names, space/bcount,(space/bcount)*100/blocksize);
316 	return (struct stats) { names, space, bcount};
317 }
318 #endif /* DX_DEBUG */
319 
320 /*
321  * Probe for a directory leaf block to search.
322  *
323  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
324  * error in the directory index, and the caller should fall back to
325  * searching the directory normally.  The callers of dx_probe **MUST**
326  * check for this error code, and make sure it never gets reflected
327  * back to userspace.
328  */
329 static struct dx_frame *
330 dx_probe(struct dentry *dentry, struct inode *dir,
331 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
332 {
333 	unsigned count, indirect;
334 	struct dx_entry *at, *entries, *p, *q, *m;
335 	struct dx_root *root;
336 	struct buffer_head *bh;
337 	struct dx_frame *frame = frame_in;
338 	u32 hash;
339 
340 	frame->bh = NULL;
341 	if (dentry)
342 		dir = dentry->d_parent->d_inode;
343 	if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
344 		goto fail;
345 	root = (struct dx_root *) bh->b_data;
346 	if (root->info.hash_version != DX_HASH_TEA &&
347 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
348 	    root->info.hash_version != DX_HASH_LEGACY) {
349 		ext4_warning(dir->i_sb, __FUNCTION__,
350 			     "Unrecognised inode hash code %d",
351 			     root->info.hash_version);
352 		brelse(bh);
353 		*err = ERR_BAD_DX_DIR;
354 		goto fail;
355 	}
356 	hinfo->hash_version = root->info.hash_version;
357 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
358 	if (dentry)
359 		ext4fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
360 	hash = hinfo->hash;
361 
362 	if (root->info.unused_flags & 1) {
363 		ext4_warning(dir->i_sb, __FUNCTION__,
364 			     "Unimplemented inode hash flags: %#06x",
365 			     root->info.unused_flags);
366 		brelse(bh);
367 		*err = ERR_BAD_DX_DIR;
368 		goto fail;
369 	}
370 
371 	if ((indirect = root->info.indirect_levels) > 1) {
372 		ext4_warning(dir->i_sb, __FUNCTION__,
373 			     "Unimplemented inode hash depth: %#06x",
374 			     root->info.indirect_levels);
375 		brelse(bh);
376 		*err = ERR_BAD_DX_DIR;
377 		goto fail;
378 	}
379 
380 	entries = (struct dx_entry *) (((char *)&root->info) +
381 				       root->info.info_length);
382 	assert(dx_get_limit(entries) == dx_root_limit(dir,
383 						      root->info.info_length));
384 	dxtrace (printk("Look up %x", hash));
385 	while (1)
386 	{
387 		count = dx_get_count(entries);
388 		assert (count && count <= dx_get_limit(entries));
389 		p = entries + 1;
390 		q = entries + count - 1;
391 		while (p <= q)
392 		{
393 			m = p + (q - p)/2;
394 			dxtrace(printk("."));
395 			if (dx_get_hash(m) > hash)
396 				q = m - 1;
397 			else
398 				p = m + 1;
399 		}
400 
401 		if (0) // linear search cross check
402 		{
403 			unsigned n = count - 1;
404 			at = entries;
405 			while (n--)
406 			{
407 				dxtrace(printk(","));
408 				if (dx_get_hash(++at) > hash)
409 				{
410 					at--;
411 					break;
412 				}
413 			}
414 			assert (at == p - 1);
415 		}
416 
417 		at = p - 1;
418 		dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
419 		frame->bh = bh;
420 		frame->entries = entries;
421 		frame->at = at;
422 		if (!indirect--) return frame;
423 		if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
424 			goto fail2;
425 		at = entries = ((struct dx_node *) bh->b_data)->entries;
426 		assert (dx_get_limit(entries) == dx_node_limit (dir));
427 		frame++;
428 	}
429 fail2:
430 	while (frame >= frame_in) {
431 		brelse(frame->bh);
432 		frame--;
433 	}
434 fail:
435 	return NULL;
436 }
437 
438 static void dx_release (struct dx_frame *frames)
439 {
440 	if (frames[0].bh == NULL)
441 		return;
442 
443 	if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
444 		brelse(frames[1].bh);
445 	brelse(frames[0].bh);
446 }
447 
448 /*
449  * This function increments the frame pointer to search the next leaf
450  * block, and reads in the necessary intervening nodes if the search
451  * should be necessary.  Whether or not the search is necessary is
452  * controlled by the hash parameter.  If the hash value is even, then
453  * the search is only continued if the next block starts with that
454  * hash value.  This is used if we are searching for a specific file.
455  *
456  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
457  *
458  * This function returns 1 if the caller should continue to search,
459  * or 0 if it should not.  If there is an error reading one of the
460  * index blocks, it will a negative error code.
461  *
462  * If start_hash is non-null, it will be filled in with the starting
463  * hash of the next page.
464  */
465 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
466 				 struct dx_frame *frame,
467 				 struct dx_frame *frames,
468 				 __u32 *start_hash)
469 {
470 	struct dx_frame *p;
471 	struct buffer_head *bh;
472 	int err, num_frames = 0;
473 	__u32 bhash;
474 
475 	p = frame;
476 	/*
477 	 * Find the next leaf page by incrementing the frame pointer.
478 	 * If we run out of entries in the interior node, loop around and
479 	 * increment pointer in the parent node.  When we break out of
480 	 * this loop, num_frames indicates the number of interior
481 	 * nodes need to be read.
482 	 */
483 	while (1) {
484 		if (++(p->at) < p->entries + dx_get_count(p->entries))
485 			break;
486 		if (p == frames)
487 			return 0;
488 		num_frames++;
489 		p--;
490 	}
491 
492 	/*
493 	 * If the hash is 1, then continue only if the next page has a
494 	 * continuation hash of any value.  This is used for readdir
495 	 * handling.  Otherwise, check to see if the hash matches the
496 	 * desired contiuation hash.  If it doesn't, return since
497 	 * there's no point to read in the successive index pages.
498 	 */
499 	bhash = dx_get_hash(p->at);
500 	if (start_hash)
501 		*start_hash = bhash;
502 	if ((hash & 1) == 0) {
503 		if ((bhash & ~1) != hash)
504 			return 0;
505 	}
506 	/*
507 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
508 	 * block so no check is necessary
509 	 */
510 	while (num_frames--) {
511 		if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
512 				      0, &err)))
513 			return err; /* Failure */
514 		p++;
515 		brelse (p->bh);
516 		p->bh = bh;
517 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
518 	}
519 	return 1;
520 }
521 
522 
523 /*
524  * p is at least 6 bytes before the end of page
525  */
526 static inline struct ext4_dir_entry_2 *ext4_next_entry(struct ext4_dir_entry_2 *p)
527 {
528 	return (struct ext4_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
529 }
530 
531 /*
532  * This function fills a red-black tree with information from a
533  * directory block.  It returns the number directory entries loaded
534  * into the tree.  If there is an error it is returned in err.
535  */
536 static int htree_dirblock_to_tree(struct file *dir_file,
537 				  struct inode *dir, int block,
538 				  struct dx_hash_info *hinfo,
539 				  __u32 start_hash, __u32 start_minor_hash)
540 {
541 	struct buffer_head *bh;
542 	struct ext4_dir_entry_2 *de, *top;
543 	int err, count = 0;
544 
545 	dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
546 	if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
547 		return err;
548 
549 	de = (struct ext4_dir_entry_2 *) bh->b_data;
550 	top = (struct ext4_dir_entry_2 *) ((char *) de +
551 					   dir->i_sb->s_blocksize -
552 					   EXT4_DIR_REC_LEN(0));
553 	for (; de < top; de = ext4_next_entry(de)) {
554 		if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
555 					(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
556 						+((char *)de - bh->b_data))) {
557 			/* On error, skip the f_pos to the next block. */
558 			dir_file->f_pos = (dir_file->f_pos |
559 					(dir->i_sb->s_blocksize - 1)) + 1;
560 			brelse (bh);
561 			return count;
562 		}
563 		ext4fs_dirhash(de->name, de->name_len, hinfo);
564 		if ((hinfo->hash < start_hash) ||
565 		    ((hinfo->hash == start_hash) &&
566 		     (hinfo->minor_hash < start_minor_hash)))
567 			continue;
568 		if (de->inode == 0)
569 			continue;
570 		if ((err = ext4_htree_store_dirent(dir_file,
571 				   hinfo->hash, hinfo->minor_hash, de)) != 0) {
572 			brelse(bh);
573 			return err;
574 		}
575 		count++;
576 	}
577 	brelse(bh);
578 	return count;
579 }
580 
581 
582 /*
583  * This function fills a red-black tree with information from a
584  * directory.  We start scanning the directory in hash order, starting
585  * at start_hash and start_minor_hash.
586  *
587  * This function returns the number of entries inserted into the tree,
588  * or a negative error code.
589  */
590 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
591 			 __u32 start_minor_hash, __u32 *next_hash)
592 {
593 	struct dx_hash_info hinfo;
594 	struct ext4_dir_entry_2 *de;
595 	struct dx_frame frames[2], *frame;
596 	struct inode *dir;
597 	int block, err;
598 	int count = 0;
599 	int ret;
600 	__u32 hashval;
601 
602 	dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
603 		       start_minor_hash));
604 	dir = dir_file->f_path.dentry->d_inode;
605 	if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
606 		hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
607 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
608 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
609 					       start_hash, start_minor_hash);
610 		*next_hash = ~0;
611 		return count;
612 	}
613 	hinfo.hash = start_hash;
614 	hinfo.minor_hash = 0;
615 	frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
616 	if (!frame)
617 		return err;
618 
619 	/* Add '.' and '..' from the htree header */
620 	if (!start_hash && !start_minor_hash) {
621 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
622 		if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
623 			goto errout;
624 		count++;
625 	}
626 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
627 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
628 		de = ext4_next_entry(de);
629 		if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
630 			goto errout;
631 		count++;
632 	}
633 
634 	while (1) {
635 		block = dx_get_block(frame->at);
636 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
637 					     start_hash, start_minor_hash);
638 		if (ret < 0) {
639 			err = ret;
640 			goto errout;
641 		}
642 		count += ret;
643 		hashval = ~0;
644 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
645 					    frame, frames, &hashval);
646 		*next_hash = hashval;
647 		if (ret < 0) {
648 			err = ret;
649 			goto errout;
650 		}
651 		/*
652 		 * Stop if:  (a) there are no more entries, or
653 		 * (b) we have inserted at least one entry and the
654 		 * next hash value is not a continuation
655 		 */
656 		if ((ret == 0) ||
657 		    (count && ((hashval & 1) == 0)))
658 			break;
659 	}
660 	dx_release(frames);
661 	dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
662 		       count, *next_hash));
663 	return count;
664 errout:
665 	dx_release(frames);
666 	return (err);
667 }
668 
669 
670 /*
671  * Directory block splitting, compacting
672  */
673 
674 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
675 			struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
676 {
677 	int count = 0;
678 	char *base = (char *) de;
679 	struct dx_hash_info h = *hinfo;
680 
681 	while ((char *) de < base + size)
682 	{
683 		if (de->name_len && de->inode) {
684 			ext4fs_dirhash(de->name, de->name_len, &h);
685 			map_tail--;
686 			map_tail->hash = h.hash;
687 			map_tail->offs = (u32) ((char *) de - base);
688 			count++;
689 			cond_resched();
690 		}
691 		/* XXX: do we need to check rec_len == 0 case? -Chris */
692 		de = (struct ext4_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
693 	}
694 	return count;
695 }
696 
697 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
698 {
699 	struct dx_map_entry *p, *q, *top = map + count - 1;
700 	int more;
701 	/* Combsort until bubble sort doesn't suck */
702 	while (count > 2) {
703 		count = count*10/13;
704 		if (count - 9 < 2) /* 9, 10 -> 11 */
705 			count = 11;
706 		for (p = top, q = p - count; q >= map; p--, q--)
707 			if (p->hash < q->hash)
708 				swap(*p, *q);
709 	}
710 	/* Garden variety bubble sort */
711 	do {
712 		more = 0;
713 		q = top;
714 		while (q-- > map) {
715 			if (q[1].hash >= q[0].hash)
716 				continue;
717 			swap(*(q+1), *q);
718 			more = 1;
719 		}
720 	} while(more);
721 }
722 
723 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
724 {
725 	struct dx_entry *entries = frame->entries;
726 	struct dx_entry *old = frame->at, *new = old + 1;
727 	int count = dx_get_count(entries);
728 
729 	assert(count < dx_get_limit(entries));
730 	assert(old < entries + count);
731 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
732 	dx_set_hash(new, hash);
733 	dx_set_block(new, block);
734 	dx_set_count(entries, count + 1);
735 }
736 #endif
737 
738 
739 static void ext4_update_dx_flag(struct inode *inode)
740 {
741 	if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
742 				     EXT4_FEATURE_COMPAT_DIR_INDEX))
743 		EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
744 }
745 
746 /*
747  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
748  *
749  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
750  * `de != NULL' is guaranteed by caller.
751  */
752 static inline int ext4_match (int len, const char * const name,
753 			      struct ext4_dir_entry_2 * de)
754 {
755 	if (len != de->name_len)
756 		return 0;
757 	if (!de->inode)
758 		return 0;
759 	return !memcmp(name, de->name, len);
760 }
761 
762 /*
763  * Returns 0 if not found, -1 on failure, and 1 on success
764  */
765 static inline int search_dirblock(struct buffer_head * bh,
766 				  struct inode *dir,
767 				  struct dentry *dentry,
768 				  unsigned long offset,
769 				  struct ext4_dir_entry_2 ** res_dir)
770 {
771 	struct ext4_dir_entry_2 * de;
772 	char * dlimit;
773 	int de_len;
774 	const char *name = dentry->d_name.name;
775 	int namelen = dentry->d_name.len;
776 
777 	de = (struct ext4_dir_entry_2 *) bh->b_data;
778 	dlimit = bh->b_data + dir->i_sb->s_blocksize;
779 	while ((char *) de < dlimit) {
780 		/* this code is executed quadratically often */
781 		/* do minimal checking `by hand' */
782 
783 		if ((char *) de + namelen <= dlimit &&
784 		    ext4_match (namelen, name, de)) {
785 			/* found a match - just to be sure, do a full check */
786 			if (!ext4_check_dir_entry("ext4_find_entry",
787 						  dir, de, bh, offset))
788 				return -1;
789 			*res_dir = de;
790 			return 1;
791 		}
792 		/* prevent looping on a bad block */
793 		de_len = le16_to_cpu(de->rec_len);
794 		if (de_len <= 0)
795 			return -1;
796 		offset += de_len;
797 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
798 	}
799 	return 0;
800 }
801 
802 
803 /*
804  *	ext4_find_entry()
805  *
806  * finds an entry in the specified directory with the wanted name. It
807  * returns the cache buffer in which the entry was found, and the entry
808  * itself (as a parameter - res_dir). It does NOT read the inode of the
809  * entry - you'll have to do that yourself if you want to.
810  *
811  * The returned buffer_head has ->b_count elevated.  The caller is expected
812  * to brelse() it when appropriate.
813  */
814 static struct buffer_head * ext4_find_entry (struct dentry *dentry,
815 					struct ext4_dir_entry_2 ** res_dir)
816 {
817 	struct super_block * sb;
818 	struct buffer_head * bh_use[NAMEI_RA_SIZE];
819 	struct buffer_head * bh, *ret = NULL;
820 	unsigned long start, block, b;
821 	int ra_max = 0;		/* Number of bh's in the readahead
822 				   buffer, bh_use[] */
823 	int ra_ptr = 0;		/* Current index into readahead
824 				   buffer */
825 	int num = 0;
826 	int nblocks, i, err;
827 	struct inode *dir = dentry->d_parent->d_inode;
828 	int namelen;
829 	const u8 *name;
830 	unsigned blocksize;
831 
832 	*res_dir = NULL;
833 	sb = dir->i_sb;
834 	blocksize = sb->s_blocksize;
835 	namelen = dentry->d_name.len;
836 	name = dentry->d_name.name;
837 	if (namelen > EXT4_NAME_LEN)
838 		return NULL;
839 #ifdef CONFIG_EXT4_INDEX
840 	if (is_dx(dir)) {
841 		bh = ext4_dx_find_entry(dentry, res_dir, &err);
842 		/*
843 		 * On success, or if the error was file not found,
844 		 * return.  Otherwise, fall back to doing a search the
845 		 * old fashioned way.
846 		 */
847 		if (bh || (err != ERR_BAD_DX_DIR))
848 			return bh;
849 		dxtrace(printk("ext4_find_entry: dx failed, falling back\n"));
850 	}
851 #endif
852 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
853 	start = EXT4_I(dir)->i_dir_start_lookup;
854 	if (start >= nblocks)
855 		start = 0;
856 	block = start;
857 restart:
858 	do {
859 		/*
860 		 * We deal with the read-ahead logic here.
861 		 */
862 		if (ra_ptr >= ra_max) {
863 			/* Refill the readahead buffer */
864 			ra_ptr = 0;
865 			b = block;
866 			for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
867 				/*
868 				 * Terminate if we reach the end of the
869 				 * directory and must wrap, or if our
870 				 * search has finished at this block.
871 				 */
872 				if (b >= nblocks || (num && block == start)) {
873 					bh_use[ra_max] = NULL;
874 					break;
875 				}
876 				num++;
877 				bh = ext4_getblk(NULL, dir, b++, 0, &err);
878 				bh_use[ra_max] = bh;
879 				if (bh)
880 					ll_rw_block(READ_META, 1, &bh);
881 			}
882 		}
883 		if ((bh = bh_use[ra_ptr++]) == NULL)
884 			goto next;
885 		wait_on_buffer(bh);
886 		if (!buffer_uptodate(bh)) {
887 			/* read error, skip block & hope for the best */
888 			ext4_error(sb, __FUNCTION__, "reading directory #%lu "
889 				   "offset %lu", dir->i_ino, block);
890 			brelse(bh);
891 			goto next;
892 		}
893 		i = search_dirblock(bh, dir, dentry,
894 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
895 		if (i == 1) {
896 			EXT4_I(dir)->i_dir_start_lookup = block;
897 			ret = bh;
898 			goto cleanup_and_exit;
899 		} else {
900 			brelse(bh);
901 			if (i < 0)
902 				goto cleanup_and_exit;
903 		}
904 	next:
905 		if (++block >= nblocks)
906 			block = 0;
907 	} while (block != start);
908 
909 	/*
910 	 * If the directory has grown while we were searching, then
911 	 * search the last part of the directory before giving up.
912 	 */
913 	block = nblocks;
914 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
915 	if (block < nblocks) {
916 		start = 0;
917 		goto restart;
918 	}
919 
920 cleanup_and_exit:
921 	/* Clean up the read-ahead blocks */
922 	for (; ra_ptr < ra_max; ra_ptr++)
923 		brelse (bh_use[ra_ptr]);
924 	return ret;
925 }
926 
927 #ifdef CONFIG_EXT4_INDEX
928 static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
929 		       struct ext4_dir_entry_2 **res_dir, int *err)
930 {
931 	struct super_block * sb;
932 	struct dx_hash_info	hinfo;
933 	u32 hash;
934 	struct dx_frame frames[2], *frame;
935 	struct ext4_dir_entry_2 *de, *top;
936 	struct buffer_head *bh;
937 	unsigned long block;
938 	int retval;
939 	int namelen = dentry->d_name.len;
940 	const u8 *name = dentry->d_name.name;
941 	struct inode *dir = dentry->d_parent->d_inode;
942 
943 	sb = dir->i_sb;
944 	/* NFS may look up ".." - look at dx_root directory block */
945 	if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
946 		if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
947 			return NULL;
948 	} else {
949 		frame = frames;
950 		frame->bh = NULL;			/* for dx_release() */
951 		frame->at = (struct dx_entry *)frames;	/* hack for zero entry*/
952 		dx_set_block(frame->at, 0);		/* dx_root block is 0 */
953 	}
954 	hash = hinfo.hash;
955 	do {
956 		block = dx_get_block(frame->at);
957 		if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
958 			goto errout;
959 		de = (struct ext4_dir_entry_2 *) bh->b_data;
960 		top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
961 				       EXT4_DIR_REC_LEN(0));
962 		for (; de < top; de = ext4_next_entry(de))
963 		if (ext4_match (namelen, name, de)) {
964 			if (!ext4_check_dir_entry("ext4_find_entry",
965 						  dir, de, bh,
966 				  (block<<EXT4_BLOCK_SIZE_BITS(sb))
967 					  +((char *)de - bh->b_data))) {
968 				brelse (bh);
969 				*err = ERR_BAD_DX_DIR;
970 				goto errout;
971 			}
972 			*res_dir = de;
973 			dx_release (frames);
974 			return bh;
975 		}
976 		brelse (bh);
977 		/* Check to see if we should continue to search */
978 		retval = ext4_htree_next_block(dir, hash, frame,
979 					       frames, NULL);
980 		if (retval < 0) {
981 			ext4_warning(sb, __FUNCTION__,
982 			     "error reading index page in directory #%lu",
983 			     dir->i_ino);
984 			*err = retval;
985 			goto errout;
986 		}
987 	} while (retval == 1);
988 
989 	*err = -ENOENT;
990 errout:
991 	dxtrace(printk("%s not found\n", name));
992 	dx_release (frames);
993 	return NULL;
994 }
995 #endif
996 
997 static struct dentry *ext4_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
998 {
999 	struct inode * inode;
1000 	struct ext4_dir_entry_2 * de;
1001 	struct buffer_head * bh;
1002 
1003 	if (dentry->d_name.len > EXT4_NAME_LEN)
1004 		return ERR_PTR(-ENAMETOOLONG);
1005 
1006 	bh = ext4_find_entry(dentry, &de);
1007 	inode = NULL;
1008 	if (bh) {
1009 		unsigned long ino = le32_to_cpu(de->inode);
1010 		brelse (bh);
1011 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1012 			ext4_error(dir->i_sb, "ext4_lookup",
1013 				   "bad inode number: %lu", ino);
1014 			inode = NULL;
1015 		} else
1016 			inode = iget(dir->i_sb, ino);
1017 
1018 		if (!inode)
1019 			return ERR_PTR(-EACCES);
1020 
1021 		if (is_bad_inode(inode)) {
1022 			iput(inode);
1023 			return ERR_PTR(-ENOENT);
1024 		}
1025 	}
1026 	return d_splice_alias(inode, dentry);
1027 }
1028 
1029 
1030 struct dentry *ext4_get_parent(struct dentry *child)
1031 {
1032 	unsigned long ino;
1033 	struct dentry *parent;
1034 	struct inode *inode;
1035 	struct dentry dotdot;
1036 	struct ext4_dir_entry_2 * de;
1037 	struct buffer_head *bh;
1038 
1039 	dotdot.d_name.name = "..";
1040 	dotdot.d_name.len = 2;
1041 	dotdot.d_parent = child; /* confusing, isn't it! */
1042 
1043 	bh = ext4_find_entry(&dotdot, &de);
1044 	inode = NULL;
1045 	if (!bh)
1046 		return ERR_PTR(-ENOENT);
1047 	ino = le32_to_cpu(de->inode);
1048 	brelse(bh);
1049 
1050 	if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1051 		ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1052 			   "bad inode number: %lu", ino);
1053 		inode = NULL;
1054 	} else
1055 		inode = iget(child->d_inode->i_sb, ino);
1056 
1057 	if (!inode)
1058 		return ERR_PTR(-EACCES);
1059 
1060 	if (is_bad_inode(inode)) {
1061 		iput(inode);
1062 		return ERR_PTR(-ENOENT);
1063 	}
1064 
1065 	parent = d_alloc_anon(inode);
1066 	if (!parent) {
1067 		iput(inode);
1068 		parent = ERR_PTR(-ENOMEM);
1069 	}
1070 	return parent;
1071 }
1072 
1073 #define S_SHIFT 12
1074 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1075 	[S_IFREG >> S_SHIFT]	= EXT4_FT_REG_FILE,
1076 	[S_IFDIR >> S_SHIFT]	= EXT4_FT_DIR,
1077 	[S_IFCHR >> S_SHIFT]	= EXT4_FT_CHRDEV,
1078 	[S_IFBLK >> S_SHIFT]	= EXT4_FT_BLKDEV,
1079 	[S_IFIFO >> S_SHIFT]	= EXT4_FT_FIFO,
1080 	[S_IFSOCK >> S_SHIFT]	= EXT4_FT_SOCK,
1081 	[S_IFLNK >> S_SHIFT]	= EXT4_FT_SYMLINK,
1082 };
1083 
1084 static inline void ext4_set_de_type(struct super_block *sb,
1085 				struct ext4_dir_entry_2 *de,
1086 				umode_t mode) {
1087 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1088 		de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1089 }
1090 
1091 #ifdef CONFIG_EXT4_INDEX
1092 static struct ext4_dir_entry_2 *
1093 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1094 {
1095 	unsigned rec_len = 0;
1096 
1097 	while (count--) {
1098 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1099 		rec_len = EXT4_DIR_REC_LEN(de->name_len);
1100 		memcpy (to, de, rec_len);
1101 		((struct ext4_dir_entry_2 *) to)->rec_len =
1102 				cpu_to_le16(rec_len);
1103 		de->inode = 0;
1104 		map++;
1105 		to += rec_len;
1106 	}
1107 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1108 }
1109 
1110 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
1111 {
1112 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1113 	unsigned rec_len = 0;
1114 
1115 	prev = to = de;
1116 	while ((char*)de < base + size) {
1117 		next = (struct ext4_dir_entry_2 *) ((char *) de +
1118 						    le16_to_cpu(de->rec_len));
1119 		if (de->inode && de->name_len) {
1120 			rec_len = EXT4_DIR_REC_LEN(de->name_len);
1121 			if (de > to)
1122 				memmove(to, de, rec_len);
1123 			to->rec_len = cpu_to_le16(rec_len);
1124 			prev = to;
1125 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1126 		}
1127 		de = next;
1128 	}
1129 	return prev;
1130 }
1131 
1132 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1133 			struct buffer_head **bh,struct dx_frame *frame,
1134 			struct dx_hash_info *hinfo, int *error)
1135 {
1136 	unsigned blocksize = dir->i_sb->s_blocksize;
1137 	unsigned count, continued;
1138 	struct buffer_head *bh2;
1139 	u32 newblock;
1140 	u32 hash2;
1141 	struct dx_map_entry *map;
1142 	char *data1 = (*bh)->b_data, *data2;
1143 	unsigned split;
1144 	struct ext4_dir_entry_2 *de = NULL, *de2;
1145 	int	err = 0;
1146 
1147 	bh2 = ext4_append (handle, dir, &newblock, &err);
1148 	if (!(bh2)) {
1149 		brelse(*bh);
1150 		*bh = NULL;
1151 		goto errout;
1152 	}
1153 
1154 	BUFFER_TRACE(*bh, "get_write_access");
1155 	err = ext4_journal_get_write_access(handle, *bh);
1156 	if (err)
1157 		goto journal_error;
1158 
1159 	BUFFER_TRACE(frame->bh, "get_write_access");
1160 	err = ext4_journal_get_write_access(handle, frame->bh);
1161 	if (err)
1162 		goto journal_error;
1163 
1164 	data2 = bh2->b_data;
1165 
1166 	/* create map in the end of data2 block */
1167 	map = (struct dx_map_entry *) (data2 + blocksize);
1168 	count = dx_make_map ((struct ext4_dir_entry_2 *) data1,
1169 			     blocksize, hinfo, map);
1170 	map -= count;
1171 	split = count/2; // need to adjust to actual middle
1172 	dx_sort_map (map, count);
1173 	hash2 = map[split].hash;
1174 	continued = hash2 == map[split - 1].hash;
1175 	dxtrace(printk("Split block %i at %x, %i/%i\n",
1176 		dx_get_block(frame->at), hash2, split, count-split));
1177 
1178 	/* Fancy dance to stay within two buffers */
1179 	de2 = dx_move_dirents(data1, data2, map + split, count - split);
1180 	de = dx_pack_dirents(data1,blocksize);
1181 	de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1182 	de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1183 	dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1184 	dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1185 
1186 	/* Which block gets the new entry? */
1187 	if (hinfo->hash >= hash2)
1188 	{
1189 		swap(*bh, bh2);
1190 		de = de2;
1191 	}
1192 	dx_insert_block (frame, hash2 + continued, newblock);
1193 	err = ext4_journal_dirty_metadata (handle, bh2);
1194 	if (err)
1195 		goto journal_error;
1196 	err = ext4_journal_dirty_metadata (handle, frame->bh);
1197 	if (err)
1198 		goto journal_error;
1199 	brelse (bh2);
1200 	dxtrace(dx_show_index ("frame", frame->entries));
1201 	return de;
1202 
1203 journal_error:
1204 	brelse(*bh);
1205 	brelse(bh2);
1206 	*bh = NULL;
1207 	ext4_std_error(dir->i_sb, err);
1208 errout:
1209 	*error = err;
1210 	return NULL;
1211 }
1212 #endif
1213 
1214 
1215 /*
1216  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1217  * it points to a directory entry which is guaranteed to be large
1218  * enough for new directory entry.  If de is NULL, then
1219  * add_dirent_to_buf will attempt search the directory block for
1220  * space.  It will return -ENOSPC if no space is available, and -EIO
1221  * and -EEXIST if directory entry already exists.
1222  *
1223  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1224  * all other cases bh is released.
1225  */
1226 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1227 			     struct inode *inode, struct ext4_dir_entry_2 *de,
1228 			     struct buffer_head * bh)
1229 {
1230 	struct inode	*dir = dentry->d_parent->d_inode;
1231 	const char	*name = dentry->d_name.name;
1232 	int		namelen = dentry->d_name.len;
1233 	unsigned long	offset = 0;
1234 	unsigned short	reclen;
1235 	int		nlen, rlen, err;
1236 	char		*top;
1237 
1238 	reclen = EXT4_DIR_REC_LEN(namelen);
1239 	if (!de) {
1240 		de = (struct ext4_dir_entry_2 *)bh->b_data;
1241 		top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1242 		while ((char *) de <= top) {
1243 			if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1244 						  bh, offset)) {
1245 				brelse (bh);
1246 				return -EIO;
1247 			}
1248 			if (ext4_match (namelen, name, de)) {
1249 				brelse (bh);
1250 				return -EEXIST;
1251 			}
1252 			nlen = EXT4_DIR_REC_LEN(de->name_len);
1253 			rlen = le16_to_cpu(de->rec_len);
1254 			if ((de->inode? rlen - nlen: rlen) >= reclen)
1255 				break;
1256 			de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1257 			offset += rlen;
1258 		}
1259 		if ((char *) de > top)
1260 			return -ENOSPC;
1261 	}
1262 	BUFFER_TRACE(bh, "get_write_access");
1263 	err = ext4_journal_get_write_access(handle, bh);
1264 	if (err) {
1265 		ext4_std_error(dir->i_sb, err);
1266 		brelse(bh);
1267 		return err;
1268 	}
1269 
1270 	/* By now the buffer is marked for journaling */
1271 	nlen = EXT4_DIR_REC_LEN(de->name_len);
1272 	rlen = le16_to_cpu(de->rec_len);
1273 	if (de->inode) {
1274 		struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1275 		de1->rec_len = cpu_to_le16(rlen - nlen);
1276 		de->rec_len = cpu_to_le16(nlen);
1277 		de = de1;
1278 	}
1279 	de->file_type = EXT4_FT_UNKNOWN;
1280 	if (inode) {
1281 		de->inode = cpu_to_le32(inode->i_ino);
1282 		ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1283 	} else
1284 		de->inode = 0;
1285 	de->name_len = namelen;
1286 	memcpy (de->name, name, namelen);
1287 	/*
1288 	 * XXX shouldn't update any times until successful
1289 	 * completion of syscall, but too many callers depend
1290 	 * on this.
1291 	 *
1292 	 * XXX similarly, too many callers depend on
1293 	 * ext4_new_inode() setting the times, but error
1294 	 * recovery deletes the inode, so the worst that can
1295 	 * happen is that the times are slightly out of date
1296 	 * and/or different from the directory change time.
1297 	 */
1298 	dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1299 	ext4_update_dx_flag(dir);
1300 	dir->i_version++;
1301 	ext4_mark_inode_dirty(handle, dir);
1302 	BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1303 	err = ext4_journal_dirty_metadata(handle, bh);
1304 	if (err)
1305 		ext4_std_error(dir->i_sb, err);
1306 	brelse(bh);
1307 	return 0;
1308 }
1309 
1310 #ifdef CONFIG_EXT4_INDEX
1311 /*
1312  * This converts a one block unindexed directory to a 3 block indexed
1313  * directory, and adds the dentry to the indexed directory.
1314  */
1315 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1316 			    struct inode *inode, struct buffer_head *bh)
1317 {
1318 	struct inode	*dir = dentry->d_parent->d_inode;
1319 	const char	*name = dentry->d_name.name;
1320 	int		namelen = dentry->d_name.len;
1321 	struct buffer_head *bh2;
1322 	struct dx_root	*root;
1323 	struct dx_frame	frames[2], *frame;
1324 	struct dx_entry *entries;
1325 	struct ext4_dir_entry_2	*de, *de2;
1326 	char		*data1, *top;
1327 	unsigned	len;
1328 	int		retval;
1329 	unsigned	blocksize;
1330 	struct dx_hash_info hinfo;
1331 	u32		block;
1332 	struct fake_dirent *fde;
1333 
1334 	blocksize =  dir->i_sb->s_blocksize;
1335 	dxtrace(printk("Creating index\n"));
1336 	retval = ext4_journal_get_write_access(handle, bh);
1337 	if (retval) {
1338 		ext4_std_error(dir->i_sb, retval);
1339 		brelse(bh);
1340 		return retval;
1341 	}
1342 	root = (struct dx_root *) bh->b_data;
1343 
1344 	bh2 = ext4_append (handle, dir, &block, &retval);
1345 	if (!(bh2)) {
1346 		brelse(bh);
1347 		return retval;
1348 	}
1349 	EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1350 	data1 = bh2->b_data;
1351 
1352 	/* The 0th block becomes the root, move the dirents out */
1353 	fde = &root->dotdot;
1354 	de = (struct ext4_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1355 	len = ((char *) root) + blocksize - (char *) de;
1356 	memcpy (data1, de, len);
1357 	de = (struct ext4_dir_entry_2 *) data1;
1358 	top = data1 + len;
1359 	while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1360 		de = de2;
1361 	de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1362 	/* Initialize the root; the dot dirents already exist */
1363 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1364 	de->rec_len = cpu_to_le16(blocksize - EXT4_DIR_REC_LEN(2));
1365 	memset (&root->info, 0, sizeof(root->info));
1366 	root->info.info_length = sizeof(root->info);
1367 	root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1368 	entries = root->entries;
1369 	dx_set_block (entries, 1);
1370 	dx_set_count (entries, 1);
1371 	dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1372 
1373 	/* Initialize as for dx_probe */
1374 	hinfo.hash_version = root->info.hash_version;
1375 	hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1376 	ext4fs_dirhash(name, namelen, &hinfo);
1377 	frame = frames;
1378 	frame->entries = entries;
1379 	frame->at = entries;
1380 	frame->bh = bh;
1381 	bh = bh2;
1382 	de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1383 	dx_release (frames);
1384 	if (!(de))
1385 		return retval;
1386 
1387 	return add_dirent_to_buf(handle, dentry, inode, de, bh);
1388 }
1389 #endif
1390 
1391 /*
1392  *	ext4_add_entry()
1393  *
1394  * adds a file entry to the specified directory, using the same
1395  * semantics as ext4_find_entry(). It returns NULL if it failed.
1396  *
1397  * NOTE!! The inode part of 'de' is left at 0 - which means you
1398  * may not sleep between calling this and putting something into
1399  * the entry, as someone else might have used it while you slept.
1400  */
1401 static int ext4_add_entry (handle_t *handle, struct dentry *dentry,
1402 	struct inode *inode)
1403 {
1404 	struct inode *dir = dentry->d_parent->d_inode;
1405 	unsigned long offset;
1406 	struct buffer_head * bh;
1407 	struct ext4_dir_entry_2 *de;
1408 	struct super_block * sb;
1409 	int	retval;
1410 #ifdef CONFIG_EXT4_INDEX
1411 	int	dx_fallback=0;
1412 #endif
1413 	unsigned blocksize;
1414 	u32 block, blocks;
1415 
1416 	sb = dir->i_sb;
1417 	blocksize = sb->s_blocksize;
1418 	if (!dentry->d_name.len)
1419 		return -EINVAL;
1420 #ifdef CONFIG_EXT4_INDEX
1421 	if (is_dx(dir)) {
1422 		retval = ext4_dx_add_entry(handle, dentry, inode);
1423 		if (!retval || (retval != ERR_BAD_DX_DIR))
1424 			return retval;
1425 		EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1426 		dx_fallback++;
1427 		ext4_mark_inode_dirty(handle, dir);
1428 	}
1429 #endif
1430 	blocks = dir->i_size >> sb->s_blocksize_bits;
1431 	for (block = 0, offset = 0; block < blocks; block++) {
1432 		bh = ext4_bread(handle, dir, block, 0, &retval);
1433 		if(!bh)
1434 			return retval;
1435 		retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1436 		if (retval != -ENOSPC)
1437 			return retval;
1438 
1439 #ifdef CONFIG_EXT4_INDEX
1440 		if (blocks == 1 && !dx_fallback &&
1441 		    EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1442 			return make_indexed_dir(handle, dentry, inode, bh);
1443 #endif
1444 		brelse(bh);
1445 	}
1446 	bh = ext4_append(handle, dir, &block, &retval);
1447 	if (!bh)
1448 		return retval;
1449 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1450 	de->inode = 0;
1451 	de->rec_len = cpu_to_le16(blocksize);
1452 	return add_dirent_to_buf(handle, dentry, inode, de, bh);
1453 }
1454 
1455 #ifdef CONFIG_EXT4_INDEX
1456 /*
1457  * Returns 0 for success, or a negative error value
1458  */
1459 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1460 			     struct inode *inode)
1461 {
1462 	struct dx_frame frames[2], *frame;
1463 	struct dx_entry *entries, *at;
1464 	struct dx_hash_info hinfo;
1465 	struct buffer_head * bh;
1466 	struct inode *dir = dentry->d_parent->d_inode;
1467 	struct super_block * sb = dir->i_sb;
1468 	struct ext4_dir_entry_2 *de;
1469 	int err;
1470 
1471 	frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1472 	if (!frame)
1473 		return err;
1474 	entries = frame->entries;
1475 	at = frame->at;
1476 
1477 	if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1478 		goto cleanup;
1479 
1480 	BUFFER_TRACE(bh, "get_write_access");
1481 	err = ext4_journal_get_write_access(handle, bh);
1482 	if (err)
1483 		goto journal_error;
1484 
1485 	err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1486 	if (err != -ENOSPC) {
1487 		bh = NULL;
1488 		goto cleanup;
1489 	}
1490 
1491 	/* Block full, should compress but for now just split */
1492 	dxtrace(printk("using %u of %u node entries\n",
1493 		       dx_get_count(entries), dx_get_limit(entries)));
1494 	/* Need to split index? */
1495 	if (dx_get_count(entries) == dx_get_limit(entries)) {
1496 		u32 newblock;
1497 		unsigned icount = dx_get_count(entries);
1498 		int levels = frame - frames;
1499 		struct dx_entry *entries2;
1500 		struct dx_node *node2;
1501 		struct buffer_head *bh2;
1502 
1503 		if (levels && (dx_get_count(frames->entries) ==
1504 			       dx_get_limit(frames->entries))) {
1505 			ext4_warning(sb, __FUNCTION__,
1506 				     "Directory index full!");
1507 			err = -ENOSPC;
1508 			goto cleanup;
1509 		}
1510 		bh2 = ext4_append (handle, dir, &newblock, &err);
1511 		if (!(bh2))
1512 			goto cleanup;
1513 		node2 = (struct dx_node *)(bh2->b_data);
1514 		entries2 = node2->entries;
1515 		node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1516 		node2->fake.inode = 0;
1517 		BUFFER_TRACE(frame->bh, "get_write_access");
1518 		err = ext4_journal_get_write_access(handle, frame->bh);
1519 		if (err)
1520 			goto journal_error;
1521 		if (levels) {
1522 			unsigned icount1 = icount/2, icount2 = icount - icount1;
1523 			unsigned hash2 = dx_get_hash(entries + icount1);
1524 			dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1525 
1526 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1527 			err = ext4_journal_get_write_access(handle,
1528 							     frames[0].bh);
1529 			if (err)
1530 				goto journal_error;
1531 
1532 			memcpy ((char *) entries2, (char *) (entries + icount1),
1533 				icount2 * sizeof(struct dx_entry));
1534 			dx_set_count (entries, icount1);
1535 			dx_set_count (entries2, icount2);
1536 			dx_set_limit (entries2, dx_node_limit(dir));
1537 
1538 			/* Which index block gets the new entry? */
1539 			if (at - entries >= icount1) {
1540 				frame->at = at = at - entries - icount1 + entries2;
1541 				frame->entries = entries = entries2;
1542 				swap(frame->bh, bh2);
1543 			}
1544 			dx_insert_block (frames + 0, hash2, newblock);
1545 			dxtrace(dx_show_index ("node", frames[1].entries));
1546 			dxtrace(dx_show_index ("node",
1547 			       ((struct dx_node *) bh2->b_data)->entries));
1548 			err = ext4_journal_dirty_metadata(handle, bh2);
1549 			if (err)
1550 				goto journal_error;
1551 			brelse (bh2);
1552 		} else {
1553 			dxtrace(printk("Creating second level index...\n"));
1554 			memcpy((char *) entries2, (char *) entries,
1555 			       icount * sizeof(struct dx_entry));
1556 			dx_set_limit(entries2, dx_node_limit(dir));
1557 
1558 			/* Set up root */
1559 			dx_set_count(entries, 1);
1560 			dx_set_block(entries + 0, newblock);
1561 			((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1562 
1563 			/* Add new access path frame */
1564 			frame = frames + 1;
1565 			frame->at = at = at - entries + entries2;
1566 			frame->entries = entries = entries2;
1567 			frame->bh = bh2;
1568 			err = ext4_journal_get_write_access(handle,
1569 							     frame->bh);
1570 			if (err)
1571 				goto journal_error;
1572 		}
1573 		ext4_journal_dirty_metadata(handle, frames[0].bh);
1574 	}
1575 	de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1576 	if (!de)
1577 		goto cleanup;
1578 	err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1579 	bh = NULL;
1580 	goto cleanup;
1581 
1582 journal_error:
1583 	ext4_std_error(dir->i_sb, err);
1584 cleanup:
1585 	if (bh)
1586 		brelse(bh);
1587 	dx_release(frames);
1588 	return err;
1589 }
1590 #endif
1591 
1592 /*
1593  * ext4_delete_entry deletes a directory entry by merging it with the
1594  * previous entry
1595  */
1596 static int ext4_delete_entry (handle_t *handle,
1597 			      struct inode * dir,
1598 			      struct ext4_dir_entry_2 * de_del,
1599 			      struct buffer_head * bh)
1600 {
1601 	struct ext4_dir_entry_2 * de, * pde;
1602 	int i;
1603 
1604 	i = 0;
1605 	pde = NULL;
1606 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1607 	while (i < bh->b_size) {
1608 		if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1609 			return -EIO;
1610 		if (de == de_del)  {
1611 			BUFFER_TRACE(bh, "get_write_access");
1612 			ext4_journal_get_write_access(handle, bh);
1613 			if (pde)
1614 				pde->rec_len =
1615 					cpu_to_le16(le16_to_cpu(pde->rec_len) +
1616 						    le16_to_cpu(de->rec_len));
1617 			else
1618 				de->inode = 0;
1619 			dir->i_version++;
1620 			BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1621 			ext4_journal_dirty_metadata(handle, bh);
1622 			return 0;
1623 		}
1624 		i += le16_to_cpu(de->rec_len);
1625 		pde = de;
1626 		de = (struct ext4_dir_entry_2 *)
1627 			((char *) de + le16_to_cpu(de->rec_len));
1628 	}
1629 	return -ENOENT;
1630 }
1631 
1632 /*
1633  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1634  * since this indicates that nlinks count was previously 1.
1635  */
1636 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1637 {
1638 	inc_nlink(inode);
1639 	if (is_dx(inode) && inode->i_nlink > 1) {
1640 		/* limit is 16-bit i_links_count */
1641 		if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1642 			inode->i_nlink = 1;
1643 			EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1644 					      EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1645 		}
1646 	}
1647 }
1648 
1649 /*
1650  * If a directory had nlink == 1, then we should let it be 1. This indicates
1651  * directory has >EXT4_LINK_MAX subdirs.
1652  */
1653 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1654 {
1655 	drop_nlink(inode);
1656 	if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1657 		inc_nlink(inode);
1658 }
1659 
1660 
1661 static int ext4_add_nondir(handle_t *handle,
1662 		struct dentry *dentry, struct inode *inode)
1663 {
1664 	int err = ext4_add_entry(handle, dentry, inode);
1665 	if (!err) {
1666 		ext4_mark_inode_dirty(handle, inode);
1667 		d_instantiate(dentry, inode);
1668 		return 0;
1669 	}
1670 	drop_nlink(inode);
1671 	iput(inode);
1672 	return err;
1673 }
1674 
1675 /*
1676  * By the time this is called, we already have created
1677  * the directory cache entry for the new file, but it
1678  * is so far negative - it has no inode.
1679  *
1680  * If the create succeeds, we fill in the inode information
1681  * with d_instantiate().
1682  */
1683 static int ext4_create (struct inode * dir, struct dentry * dentry, int mode,
1684 		struct nameidata *nd)
1685 {
1686 	handle_t *handle;
1687 	struct inode * inode;
1688 	int err, retries = 0;
1689 
1690 retry:
1691 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1692 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1693 					2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1694 	if (IS_ERR(handle))
1695 		return PTR_ERR(handle);
1696 
1697 	if (IS_DIRSYNC(dir))
1698 		handle->h_sync = 1;
1699 
1700 	inode = ext4_new_inode (handle, dir, mode);
1701 	err = PTR_ERR(inode);
1702 	if (!IS_ERR(inode)) {
1703 		inode->i_op = &ext4_file_inode_operations;
1704 		inode->i_fop = &ext4_file_operations;
1705 		ext4_set_aops(inode);
1706 		err = ext4_add_nondir(handle, dentry, inode);
1707 	}
1708 	ext4_journal_stop(handle);
1709 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1710 		goto retry;
1711 	return err;
1712 }
1713 
1714 static int ext4_mknod (struct inode * dir, struct dentry *dentry,
1715 			int mode, dev_t rdev)
1716 {
1717 	handle_t *handle;
1718 	struct inode *inode;
1719 	int err, retries = 0;
1720 
1721 	if (!new_valid_dev(rdev))
1722 		return -EINVAL;
1723 
1724 retry:
1725 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1726 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1727 					2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1728 	if (IS_ERR(handle))
1729 		return PTR_ERR(handle);
1730 
1731 	if (IS_DIRSYNC(dir))
1732 		handle->h_sync = 1;
1733 
1734 	inode = ext4_new_inode (handle, dir, mode);
1735 	err = PTR_ERR(inode);
1736 	if (!IS_ERR(inode)) {
1737 		init_special_inode(inode, inode->i_mode, rdev);
1738 #ifdef CONFIG_EXT4DEV_FS_XATTR
1739 		inode->i_op = &ext4_special_inode_operations;
1740 #endif
1741 		err = ext4_add_nondir(handle, dentry, inode);
1742 	}
1743 	ext4_journal_stop(handle);
1744 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1745 		goto retry;
1746 	return err;
1747 }
1748 
1749 static int ext4_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1750 {
1751 	handle_t *handle;
1752 	struct inode * inode;
1753 	struct buffer_head * dir_block;
1754 	struct ext4_dir_entry_2 * de;
1755 	int err, retries = 0;
1756 
1757 	if (EXT4_DIR_LINK_MAX(dir))
1758 		return -EMLINK;
1759 
1760 retry:
1761 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1762 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1763 					2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1764 	if (IS_ERR(handle))
1765 		return PTR_ERR(handle);
1766 
1767 	if (IS_DIRSYNC(dir))
1768 		handle->h_sync = 1;
1769 
1770 	inode = ext4_new_inode (handle, dir, S_IFDIR | mode);
1771 	err = PTR_ERR(inode);
1772 	if (IS_ERR(inode))
1773 		goto out_stop;
1774 
1775 	inode->i_op = &ext4_dir_inode_operations;
1776 	inode->i_fop = &ext4_dir_operations;
1777 	inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1778 	dir_block = ext4_bread (handle, inode, 0, 1, &err);
1779 	if (!dir_block) {
1780 		ext4_dec_count(handle, inode); /* is this nlink == 0? */
1781 		ext4_mark_inode_dirty(handle, inode);
1782 		iput (inode);
1783 		goto out_stop;
1784 	}
1785 	BUFFER_TRACE(dir_block, "get_write_access");
1786 	ext4_journal_get_write_access(handle, dir_block);
1787 	de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1788 	de->inode = cpu_to_le32(inode->i_ino);
1789 	de->name_len = 1;
1790 	de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de->name_len));
1791 	strcpy (de->name, ".");
1792 	ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1793 	de = (struct ext4_dir_entry_2 *)
1794 			((char *) de + le16_to_cpu(de->rec_len));
1795 	de->inode = cpu_to_le32(dir->i_ino);
1796 	de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT4_DIR_REC_LEN(1));
1797 	de->name_len = 2;
1798 	strcpy (de->name, "..");
1799 	ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1800 	inode->i_nlink = 2;
1801 	BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata");
1802 	ext4_journal_dirty_metadata(handle, dir_block);
1803 	brelse (dir_block);
1804 	ext4_mark_inode_dirty(handle, inode);
1805 	err = ext4_add_entry (handle, dentry, inode);
1806 	if (err) {
1807 		inode->i_nlink = 0;
1808 		ext4_mark_inode_dirty(handle, inode);
1809 		iput (inode);
1810 		goto out_stop;
1811 	}
1812 	ext4_inc_count(handle, dir);
1813 	ext4_update_dx_flag(dir);
1814 	ext4_mark_inode_dirty(handle, dir);
1815 	d_instantiate(dentry, inode);
1816 out_stop:
1817 	ext4_journal_stop(handle);
1818 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1819 		goto retry;
1820 	return err;
1821 }
1822 
1823 /*
1824  * routine to check that the specified directory is empty (for rmdir)
1825  */
1826 static int empty_dir (struct inode * inode)
1827 {
1828 	unsigned long offset;
1829 	struct buffer_head * bh;
1830 	struct ext4_dir_entry_2 * de, * de1;
1831 	struct super_block * sb;
1832 	int err = 0;
1833 
1834 	sb = inode->i_sb;
1835 	if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1836 	    !(bh = ext4_bread (NULL, inode, 0, 0, &err))) {
1837 		if (err)
1838 			ext4_error(inode->i_sb, __FUNCTION__,
1839 				   "error %d reading directory #%lu offset 0",
1840 				   err, inode->i_ino);
1841 		else
1842 			ext4_warning(inode->i_sb, __FUNCTION__,
1843 				     "bad directory (dir #%lu) - no data block",
1844 				     inode->i_ino);
1845 		return 1;
1846 	}
1847 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1848 	de1 = (struct ext4_dir_entry_2 *)
1849 			((char *) de + le16_to_cpu(de->rec_len));
1850 	if (le32_to_cpu(de->inode) != inode->i_ino ||
1851 			!le32_to_cpu(de1->inode) ||
1852 			strcmp (".", de->name) ||
1853 			strcmp ("..", de1->name)) {
1854 		ext4_warning (inode->i_sb, "empty_dir",
1855 			      "bad directory (dir #%lu) - no `.' or `..'",
1856 			      inode->i_ino);
1857 		brelse (bh);
1858 		return 1;
1859 	}
1860 	offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1861 	de = (struct ext4_dir_entry_2 *)
1862 			((char *) de1 + le16_to_cpu(de1->rec_len));
1863 	while (offset < inode->i_size ) {
1864 		if (!bh ||
1865 			(void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1866 			err = 0;
1867 			brelse (bh);
1868 			bh = ext4_bread (NULL, inode,
1869 				offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1870 			if (!bh) {
1871 				if (err)
1872 					ext4_error(sb, __FUNCTION__,
1873 						   "error %d reading directory"
1874 						   " #%lu offset %lu",
1875 						   err, inode->i_ino, offset);
1876 				offset += sb->s_blocksize;
1877 				continue;
1878 			}
1879 			de = (struct ext4_dir_entry_2 *) bh->b_data;
1880 		}
1881 		if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1882 			de = (struct ext4_dir_entry_2 *)(bh->b_data +
1883 							 sb->s_blocksize);
1884 			offset = (offset | (sb->s_blocksize - 1)) + 1;
1885 			continue;
1886 		}
1887 		if (le32_to_cpu(de->inode)) {
1888 			brelse (bh);
1889 			return 0;
1890 		}
1891 		offset += le16_to_cpu(de->rec_len);
1892 		de = (struct ext4_dir_entry_2 *)
1893 				((char *) de + le16_to_cpu(de->rec_len));
1894 	}
1895 	brelse (bh);
1896 	return 1;
1897 }
1898 
1899 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1900  * such inodes, starting at the superblock, in case we crash before the
1901  * file is closed/deleted, or in case the inode truncate spans multiple
1902  * transactions and the last transaction is not recovered after a crash.
1903  *
1904  * At filesystem recovery time, we walk this list deleting unlinked
1905  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1906  */
1907 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1908 {
1909 	struct super_block *sb = inode->i_sb;
1910 	struct ext4_iloc iloc;
1911 	int err = 0, rc;
1912 
1913 	lock_super(sb);
1914 	if (!list_empty(&EXT4_I(inode)->i_orphan))
1915 		goto out_unlock;
1916 
1917 	/* Orphan handling is only valid for files with data blocks
1918 	 * being truncated, or files being unlinked. */
1919 
1920 	/* @@@ FIXME: Observation from aviro:
1921 	 * I think I can trigger J_ASSERT in ext4_orphan_add().  We block
1922 	 * here (on lock_super()), so race with ext4_link() which might bump
1923 	 * ->i_nlink. For, say it, character device. Not a regular file,
1924 	 * not a directory, not a symlink and ->i_nlink > 0.
1925 	 */
1926 	J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1927 		S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1928 
1929 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1930 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1931 	if (err)
1932 		goto out_unlock;
1933 
1934 	err = ext4_reserve_inode_write(handle, inode, &iloc);
1935 	if (err)
1936 		goto out_unlock;
1937 
1938 	/* Insert this inode at the head of the on-disk orphan list... */
1939 	NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1940 	EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1941 	err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1942 	rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
1943 	if (!err)
1944 		err = rc;
1945 
1946 	/* Only add to the head of the in-memory list if all the
1947 	 * previous operations succeeded.  If the orphan_add is going to
1948 	 * fail (possibly taking the journal offline), we can't risk
1949 	 * leaving the inode on the orphan list: stray orphan-list
1950 	 * entries can cause panics at unmount time.
1951 	 *
1952 	 * This is safe: on error we're going to ignore the orphan list
1953 	 * anyway on the next recovery. */
1954 	if (!err)
1955 		list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1956 
1957 	jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1958 	jbd_debug(4, "orphan inode %lu will point to %d\n",
1959 			inode->i_ino, NEXT_ORPHAN(inode));
1960 out_unlock:
1961 	unlock_super(sb);
1962 	ext4_std_error(inode->i_sb, err);
1963 	return err;
1964 }
1965 
1966 /*
1967  * ext4_orphan_del() removes an unlinked or truncated inode from the list
1968  * of such inodes stored on disk, because it is finally being cleaned up.
1969  */
1970 int ext4_orphan_del(handle_t *handle, struct inode *inode)
1971 {
1972 	struct list_head *prev;
1973 	struct ext4_inode_info *ei = EXT4_I(inode);
1974 	struct ext4_sb_info *sbi;
1975 	unsigned long ino_next;
1976 	struct ext4_iloc iloc;
1977 	int err = 0;
1978 
1979 	lock_super(inode->i_sb);
1980 	if (list_empty(&ei->i_orphan)) {
1981 		unlock_super(inode->i_sb);
1982 		return 0;
1983 	}
1984 
1985 	ino_next = NEXT_ORPHAN(inode);
1986 	prev = ei->i_orphan.prev;
1987 	sbi = EXT4_SB(inode->i_sb);
1988 
1989 	jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1990 
1991 	list_del_init(&ei->i_orphan);
1992 
1993 	/* If we're on an error path, we may not have a valid
1994 	 * transaction handle with which to update the orphan list on
1995 	 * disk, but we still need to remove the inode from the linked
1996 	 * list in memory. */
1997 	if (!handle)
1998 		goto out;
1999 
2000 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2001 	if (err)
2002 		goto out_err;
2003 
2004 	if (prev == &sbi->s_orphan) {
2005 		jbd_debug(4, "superblock will point to %lu\n", ino_next);
2006 		BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2007 		err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2008 		if (err)
2009 			goto out_brelse;
2010 		sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2011 		err = ext4_journal_dirty_metadata(handle, sbi->s_sbh);
2012 	} else {
2013 		struct ext4_iloc iloc2;
2014 		struct inode *i_prev =
2015 			&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2016 
2017 		jbd_debug(4, "orphan inode %lu will point to %lu\n",
2018 			  i_prev->i_ino, ino_next);
2019 		err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2020 		if (err)
2021 			goto out_brelse;
2022 		NEXT_ORPHAN(i_prev) = ino_next;
2023 		err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2024 	}
2025 	if (err)
2026 		goto out_brelse;
2027 	NEXT_ORPHAN(inode) = 0;
2028 	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2029 
2030 out_err:
2031 	ext4_std_error(inode->i_sb, err);
2032 out:
2033 	unlock_super(inode->i_sb);
2034 	return err;
2035 
2036 out_brelse:
2037 	brelse(iloc.bh);
2038 	goto out_err;
2039 }
2040 
2041 static int ext4_rmdir (struct inode * dir, struct dentry *dentry)
2042 {
2043 	int retval;
2044 	struct inode * inode;
2045 	struct buffer_head * bh;
2046 	struct ext4_dir_entry_2 * de;
2047 	handle_t *handle;
2048 
2049 	/* Initialize quotas before so that eventual writes go in
2050 	 * separate transaction */
2051 	DQUOT_INIT(dentry->d_inode);
2052 	handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2053 	if (IS_ERR(handle))
2054 		return PTR_ERR(handle);
2055 
2056 	retval = -ENOENT;
2057 	bh = ext4_find_entry (dentry, &de);
2058 	if (!bh)
2059 		goto end_rmdir;
2060 
2061 	if (IS_DIRSYNC(dir))
2062 		handle->h_sync = 1;
2063 
2064 	inode = dentry->d_inode;
2065 
2066 	retval = -EIO;
2067 	if (le32_to_cpu(de->inode) != inode->i_ino)
2068 		goto end_rmdir;
2069 
2070 	retval = -ENOTEMPTY;
2071 	if (!empty_dir (inode))
2072 		goto end_rmdir;
2073 
2074 	retval = ext4_delete_entry(handle, dir, de, bh);
2075 	if (retval)
2076 		goto end_rmdir;
2077 	if (!EXT4_DIR_LINK_EMPTY(inode))
2078 		ext4_warning (inode->i_sb, "ext4_rmdir",
2079 			      "empty directory has too many links (%d)",
2080 			      inode->i_nlink);
2081 	inode->i_version++;
2082 	clear_nlink(inode);
2083 	/* There's no need to set i_disksize: the fact that i_nlink is
2084 	 * zero will ensure that the right thing happens during any
2085 	 * recovery. */
2086 	inode->i_size = 0;
2087 	ext4_orphan_add(handle, inode);
2088 	inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2089 	ext4_mark_inode_dirty(handle, inode);
2090 	ext4_dec_count(handle, dir);
2091 	ext4_update_dx_flag(dir);
2092 	ext4_mark_inode_dirty(handle, dir);
2093 
2094 end_rmdir:
2095 	ext4_journal_stop(handle);
2096 	brelse (bh);
2097 	return retval;
2098 }
2099 
2100 static int ext4_unlink(struct inode * dir, struct dentry *dentry)
2101 {
2102 	int retval;
2103 	struct inode * inode;
2104 	struct buffer_head * bh;
2105 	struct ext4_dir_entry_2 * de;
2106 	handle_t *handle;
2107 
2108 	/* Initialize quotas before so that eventual writes go
2109 	 * in separate transaction */
2110 	DQUOT_INIT(dentry->d_inode);
2111 	handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2112 	if (IS_ERR(handle))
2113 		return PTR_ERR(handle);
2114 
2115 	if (IS_DIRSYNC(dir))
2116 		handle->h_sync = 1;
2117 
2118 	retval = -ENOENT;
2119 	bh = ext4_find_entry (dentry, &de);
2120 	if (!bh)
2121 		goto end_unlink;
2122 
2123 	inode = dentry->d_inode;
2124 
2125 	retval = -EIO;
2126 	if (le32_to_cpu(de->inode) != inode->i_ino)
2127 		goto end_unlink;
2128 
2129 	if (!inode->i_nlink) {
2130 		ext4_warning (inode->i_sb, "ext4_unlink",
2131 			      "Deleting nonexistent file (%lu), %d",
2132 			      inode->i_ino, inode->i_nlink);
2133 		inode->i_nlink = 1;
2134 	}
2135 	retval = ext4_delete_entry(handle, dir, de, bh);
2136 	if (retval)
2137 		goto end_unlink;
2138 	dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2139 	ext4_update_dx_flag(dir);
2140 	ext4_mark_inode_dirty(handle, dir);
2141 	ext4_dec_count(handle, inode);
2142 	if (!inode->i_nlink)
2143 		ext4_orphan_add(handle, inode);
2144 	inode->i_ctime = ext4_current_time(inode);
2145 	ext4_mark_inode_dirty(handle, inode);
2146 	retval = 0;
2147 
2148 end_unlink:
2149 	ext4_journal_stop(handle);
2150 	brelse (bh);
2151 	return retval;
2152 }
2153 
2154 static int ext4_symlink (struct inode * dir,
2155 		struct dentry *dentry, const char * symname)
2156 {
2157 	handle_t *handle;
2158 	struct inode * inode;
2159 	int l, err, retries = 0;
2160 
2161 	l = strlen(symname)+1;
2162 	if (l > dir->i_sb->s_blocksize)
2163 		return -ENAMETOOLONG;
2164 
2165 retry:
2166 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2167 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2168 					2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2169 	if (IS_ERR(handle))
2170 		return PTR_ERR(handle);
2171 
2172 	if (IS_DIRSYNC(dir))
2173 		handle->h_sync = 1;
2174 
2175 	inode = ext4_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2176 	err = PTR_ERR(inode);
2177 	if (IS_ERR(inode))
2178 		goto out_stop;
2179 
2180 	if (l > sizeof (EXT4_I(inode)->i_data)) {
2181 		inode->i_op = &ext4_symlink_inode_operations;
2182 		ext4_set_aops(inode);
2183 		/*
2184 		 * page_symlink() calls into ext4_prepare/commit_write.
2185 		 * We have a transaction open.  All is sweetness.  It also sets
2186 		 * i_size in generic_commit_write().
2187 		 */
2188 		err = __page_symlink(inode, symname, l,
2189 				mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2190 		if (err) {
2191 			ext4_dec_count(handle, inode);
2192 			ext4_mark_inode_dirty(handle, inode);
2193 			iput (inode);
2194 			goto out_stop;
2195 		}
2196 	} else {
2197 		inode->i_op = &ext4_fast_symlink_inode_operations;
2198 		memcpy((char*)&EXT4_I(inode)->i_data,symname,l);
2199 		inode->i_size = l-1;
2200 	}
2201 	EXT4_I(inode)->i_disksize = inode->i_size;
2202 	err = ext4_add_nondir(handle, dentry, inode);
2203 out_stop:
2204 	ext4_journal_stop(handle);
2205 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2206 		goto retry;
2207 	return err;
2208 }
2209 
2210 static int ext4_link (struct dentry * old_dentry,
2211 		struct inode * dir, struct dentry *dentry)
2212 {
2213 	handle_t *handle;
2214 	struct inode *inode = old_dentry->d_inode;
2215 	int err, retries = 0;
2216 
2217 	if (EXT4_DIR_LINK_MAX(inode))
2218 		return -EMLINK;
2219 
2220 	/*
2221 	 * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2222 	 * otherwise has the potential to corrupt the orphan inode list.
2223 	 */
2224 	if (inode->i_nlink == 0)
2225 		return -ENOENT;
2226 
2227 retry:
2228 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2229 					EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2230 	if (IS_ERR(handle))
2231 		return PTR_ERR(handle);
2232 
2233 	if (IS_DIRSYNC(dir))
2234 		handle->h_sync = 1;
2235 
2236 	inode->i_ctime = ext4_current_time(inode);
2237 	ext4_inc_count(handle, inode);
2238 	atomic_inc(&inode->i_count);
2239 
2240 	err = ext4_add_nondir(handle, dentry, inode);
2241 	ext4_journal_stop(handle);
2242 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2243 		goto retry;
2244 	return err;
2245 }
2246 
2247 #define PARENT_INO(buffer) \
2248 	((struct ext4_dir_entry_2 *) ((char *) buffer + \
2249 	le16_to_cpu(((struct ext4_dir_entry_2 *) buffer)->rec_len)))->inode
2250 
2251 /*
2252  * Anybody can rename anything with this: the permission checks are left to the
2253  * higher-level routines.
2254  */
2255 static int ext4_rename (struct inode * old_dir, struct dentry *old_dentry,
2256 			   struct inode * new_dir,struct dentry *new_dentry)
2257 {
2258 	handle_t *handle;
2259 	struct inode * old_inode, * new_inode;
2260 	struct buffer_head * old_bh, * new_bh, * dir_bh;
2261 	struct ext4_dir_entry_2 * old_de, * new_de;
2262 	int retval;
2263 
2264 	old_bh = new_bh = dir_bh = NULL;
2265 
2266 	/* Initialize quotas before so that eventual writes go
2267 	 * in separate transaction */
2268 	if (new_dentry->d_inode)
2269 		DQUOT_INIT(new_dentry->d_inode);
2270 	handle = ext4_journal_start(old_dir, 2 *
2271 					EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2272 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2273 	if (IS_ERR(handle))
2274 		return PTR_ERR(handle);
2275 
2276 	if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2277 		handle->h_sync = 1;
2278 
2279 	old_bh = ext4_find_entry (old_dentry, &old_de);
2280 	/*
2281 	 *  Check for inode number is _not_ due to possible IO errors.
2282 	 *  We might rmdir the source, keep it as pwd of some process
2283 	 *  and merrily kill the link to whatever was created under the
2284 	 *  same name. Goodbye sticky bit ;-<
2285 	 */
2286 	old_inode = old_dentry->d_inode;
2287 	retval = -ENOENT;
2288 	if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2289 		goto end_rename;
2290 
2291 	new_inode = new_dentry->d_inode;
2292 	new_bh = ext4_find_entry (new_dentry, &new_de);
2293 	if (new_bh) {
2294 		if (!new_inode) {
2295 			brelse (new_bh);
2296 			new_bh = NULL;
2297 		}
2298 	}
2299 	if (S_ISDIR(old_inode->i_mode)) {
2300 		if (new_inode) {
2301 			retval = -ENOTEMPTY;
2302 			if (!empty_dir (new_inode))
2303 				goto end_rename;
2304 		}
2305 		retval = -EIO;
2306 		dir_bh = ext4_bread (handle, old_inode, 0, 0, &retval);
2307 		if (!dir_bh)
2308 			goto end_rename;
2309 		if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2310 			goto end_rename;
2311 		retval = -EMLINK;
2312 		if (!new_inode && new_dir!=old_dir &&
2313 				new_dir->i_nlink >= EXT4_LINK_MAX)
2314 			goto end_rename;
2315 	}
2316 	if (!new_bh) {
2317 		retval = ext4_add_entry (handle, new_dentry, old_inode);
2318 		if (retval)
2319 			goto end_rename;
2320 	} else {
2321 		BUFFER_TRACE(new_bh, "get write access");
2322 		ext4_journal_get_write_access(handle, new_bh);
2323 		new_de->inode = cpu_to_le32(old_inode->i_ino);
2324 		if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2325 					      EXT4_FEATURE_INCOMPAT_FILETYPE))
2326 			new_de->file_type = old_de->file_type;
2327 		new_dir->i_version++;
2328 		BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata");
2329 		ext4_journal_dirty_metadata(handle, new_bh);
2330 		brelse(new_bh);
2331 		new_bh = NULL;
2332 	}
2333 
2334 	/*
2335 	 * Like most other Unix systems, set the ctime for inodes on a
2336 	 * rename.
2337 	 */
2338 	old_inode->i_ctime = ext4_current_time(old_inode);
2339 	ext4_mark_inode_dirty(handle, old_inode);
2340 
2341 	/*
2342 	 * ok, that's it
2343 	 */
2344 	if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2345 	    old_de->name_len != old_dentry->d_name.len ||
2346 	    strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2347 	    (retval = ext4_delete_entry(handle, old_dir,
2348 					old_de, old_bh)) == -ENOENT) {
2349 		/* old_de could have moved from under us during htree split, so
2350 		 * make sure that we are deleting the right entry.  We might
2351 		 * also be pointing to a stale entry in the unused part of
2352 		 * old_bh so just checking inum and the name isn't enough. */
2353 		struct buffer_head *old_bh2;
2354 		struct ext4_dir_entry_2 *old_de2;
2355 
2356 		old_bh2 = ext4_find_entry(old_dentry, &old_de2);
2357 		if (old_bh2) {
2358 			retval = ext4_delete_entry(handle, old_dir,
2359 						   old_de2, old_bh2);
2360 			brelse(old_bh2);
2361 		}
2362 	}
2363 	if (retval) {
2364 		ext4_warning(old_dir->i_sb, "ext4_rename",
2365 				"Deleting old file (%lu), %d, error=%d",
2366 				old_dir->i_ino, old_dir->i_nlink, retval);
2367 	}
2368 
2369 	if (new_inode) {
2370 		ext4_dec_count(handle, new_inode);
2371 		new_inode->i_ctime = ext4_current_time(new_inode);
2372 	}
2373 	old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2374 	ext4_update_dx_flag(old_dir);
2375 	if (dir_bh) {
2376 		BUFFER_TRACE(dir_bh, "get_write_access");
2377 		ext4_journal_get_write_access(handle, dir_bh);
2378 		PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2379 		BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata");
2380 		ext4_journal_dirty_metadata(handle, dir_bh);
2381 		ext4_dec_count(handle, old_dir);
2382 		if (new_inode) {
2383 			/* checked empty_dir above, can't have another parent,
2384 			 * ext3_dec_count() won't work for many-linked dirs */
2385 			new_inode->i_nlink = 0;
2386 		} else {
2387 			ext4_inc_count(handle, new_dir);
2388 			ext4_update_dx_flag(new_dir);
2389 			ext4_mark_inode_dirty(handle, new_dir);
2390 		}
2391 	}
2392 	ext4_mark_inode_dirty(handle, old_dir);
2393 	if (new_inode) {
2394 		ext4_mark_inode_dirty(handle, new_inode);
2395 		if (!new_inode->i_nlink)
2396 			ext4_orphan_add(handle, new_inode);
2397 	}
2398 	retval = 0;
2399 
2400 end_rename:
2401 	brelse (dir_bh);
2402 	brelse (old_bh);
2403 	brelse (new_bh);
2404 	ext4_journal_stop(handle);
2405 	return retval;
2406 }
2407 
2408 /*
2409  * directories can handle most operations...
2410  */
2411 const struct inode_operations ext4_dir_inode_operations = {
2412 	.create		= ext4_create,
2413 	.lookup		= ext4_lookup,
2414 	.link		= ext4_link,
2415 	.unlink		= ext4_unlink,
2416 	.symlink	= ext4_symlink,
2417 	.mkdir		= ext4_mkdir,
2418 	.rmdir		= ext4_rmdir,
2419 	.mknod		= ext4_mknod,
2420 	.rename		= ext4_rename,
2421 	.setattr	= ext4_setattr,
2422 #ifdef CONFIG_EXT4DEV_FS_XATTR
2423 	.setxattr	= generic_setxattr,
2424 	.getxattr	= generic_getxattr,
2425 	.listxattr	= ext4_listxattr,
2426 	.removexattr	= generic_removexattr,
2427 #endif
2428 	.permission	= ext4_permission,
2429 };
2430 
2431 const struct inode_operations ext4_special_inode_operations = {
2432 	.setattr	= ext4_setattr,
2433 #ifdef CONFIG_EXT4DEV_FS_XATTR
2434 	.setxattr	= generic_setxattr,
2435 	.getxattr	= generic_getxattr,
2436 	.listxattr	= ext4_listxattr,
2437 	.removexattr	= generic_removexattr,
2438 #endif
2439 	.permission	= ext4_permission,
2440 };
2441