xref: /openbmc/linux/fs/ext4/namei.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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 	return d_splice_alias(inode, dentry);
1022 }
1023 
1024 
1025 struct dentry *ext4_get_parent(struct dentry *child)
1026 {
1027 	unsigned long ino;
1028 	struct dentry *parent;
1029 	struct inode *inode;
1030 	struct dentry dotdot;
1031 	struct ext4_dir_entry_2 * de;
1032 	struct buffer_head *bh;
1033 
1034 	dotdot.d_name.name = "..";
1035 	dotdot.d_name.len = 2;
1036 	dotdot.d_parent = child; /* confusing, isn't it! */
1037 
1038 	bh = ext4_find_entry(&dotdot, &de);
1039 	inode = NULL;
1040 	if (!bh)
1041 		return ERR_PTR(-ENOENT);
1042 	ino = le32_to_cpu(de->inode);
1043 	brelse(bh);
1044 
1045 	if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1046 		ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1047 			   "bad inode number: %lu", ino);
1048 		inode = NULL;
1049 	} else
1050 		inode = iget(child->d_inode->i_sb, ino);
1051 
1052 	if (!inode)
1053 		return ERR_PTR(-EACCES);
1054 
1055 	parent = d_alloc_anon(inode);
1056 	if (!parent) {
1057 		iput(inode);
1058 		parent = ERR_PTR(-ENOMEM);
1059 	}
1060 	return parent;
1061 }
1062 
1063 #define S_SHIFT 12
1064 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1065 	[S_IFREG >> S_SHIFT]	= EXT4_FT_REG_FILE,
1066 	[S_IFDIR >> S_SHIFT]	= EXT4_FT_DIR,
1067 	[S_IFCHR >> S_SHIFT]	= EXT4_FT_CHRDEV,
1068 	[S_IFBLK >> S_SHIFT]	= EXT4_FT_BLKDEV,
1069 	[S_IFIFO >> S_SHIFT]	= EXT4_FT_FIFO,
1070 	[S_IFSOCK >> S_SHIFT]	= EXT4_FT_SOCK,
1071 	[S_IFLNK >> S_SHIFT]	= EXT4_FT_SYMLINK,
1072 };
1073 
1074 static inline void ext4_set_de_type(struct super_block *sb,
1075 				struct ext4_dir_entry_2 *de,
1076 				umode_t mode) {
1077 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1078 		de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1079 }
1080 
1081 #ifdef CONFIG_EXT4_INDEX
1082 static struct ext4_dir_entry_2 *
1083 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1084 {
1085 	unsigned rec_len = 0;
1086 
1087 	while (count--) {
1088 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1089 		rec_len = EXT4_DIR_REC_LEN(de->name_len);
1090 		memcpy (to, de, rec_len);
1091 		((struct ext4_dir_entry_2 *) to)->rec_len =
1092 				cpu_to_le16(rec_len);
1093 		de->inode = 0;
1094 		map++;
1095 		to += rec_len;
1096 	}
1097 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1098 }
1099 
1100 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
1101 {
1102 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1103 	unsigned rec_len = 0;
1104 
1105 	prev = to = de;
1106 	while ((char*)de < base + size) {
1107 		next = (struct ext4_dir_entry_2 *) ((char *) de +
1108 						    le16_to_cpu(de->rec_len));
1109 		if (de->inode && de->name_len) {
1110 			rec_len = EXT4_DIR_REC_LEN(de->name_len);
1111 			if (de > to)
1112 				memmove(to, de, rec_len);
1113 			to->rec_len = cpu_to_le16(rec_len);
1114 			prev = to;
1115 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1116 		}
1117 		de = next;
1118 	}
1119 	return prev;
1120 }
1121 
1122 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1123 			struct buffer_head **bh,struct dx_frame *frame,
1124 			struct dx_hash_info *hinfo, int *error)
1125 {
1126 	unsigned blocksize = dir->i_sb->s_blocksize;
1127 	unsigned count, continued;
1128 	struct buffer_head *bh2;
1129 	u32 newblock;
1130 	u32 hash2;
1131 	struct dx_map_entry *map;
1132 	char *data1 = (*bh)->b_data, *data2;
1133 	unsigned split;
1134 	struct ext4_dir_entry_2 *de = NULL, *de2;
1135 	int	err = 0;
1136 
1137 	bh2 = ext4_append (handle, dir, &newblock, &err);
1138 	if (!(bh2)) {
1139 		brelse(*bh);
1140 		*bh = NULL;
1141 		goto errout;
1142 	}
1143 
1144 	BUFFER_TRACE(*bh, "get_write_access");
1145 	err = ext4_journal_get_write_access(handle, *bh);
1146 	if (err)
1147 		goto journal_error;
1148 
1149 	BUFFER_TRACE(frame->bh, "get_write_access");
1150 	err = ext4_journal_get_write_access(handle, frame->bh);
1151 	if (err)
1152 		goto journal_error;
1153 
1154 	data2 = bh2->b_data;
1155 
1156 	/* create map in the end of data2 block */
1157 	map = (struct dx_map_entry *) (data2 + blocksize);
1158 	count = dx_make_map ((struct ext4_dir_entry_2 *) data1,
1159 			     blocksize, hinfo, map);
1160 	map -= count;
1161 	split = count/2; // need to adjust to actual middle
1162 	dx_sort_map (map, count);
1163 	hash2 = map[split].hash;
1164 	continued = hash2 == map[split - 1].hash;
1165 	dxtrace(printk("Split block %i at %x, %i/%i\n",
1166 		dx_get_block(frame->at), hash2, split, count-split));
1167 
1168 	/* Fancy dance to stay within two buffers */
1169 	de2 = dx_move_dirents(data1, data2, map + split, count - split);
1170 	de = dx_pack_dirents(data1,blocksize);
1171 	de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1172 	de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1173 	dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1174 	dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1175 
1176 	/* Which block gets the new entry? */
1177 	if (hinfo->hash >= hash2)
1178 	{
1179 		swap(*bh, bh2);
1180 		de = de2;
1181 	}
1182 	dx_insert_block (frame, hash2 + continued, newblock);
1183 	err = ext4_journal_dirty_metadata (handle, bh2);
1184 	if (err)
1185 		goto journal_error;
1186 	err = ext4_journal_dirty_metadata (handle, frame->bh);
1187 	if (err)
1188 		goto journal_error;
1189 	brelse (bh2);
1190 	dxtrace(dx_show_index ("frame", frame->entries));
1191 	return de;
1192 
1193 journal_error:
1194 	brelse(*bh);
1195 	brelse(bh2);
1196 	*bh = NULL;
1197 	ext4_std_error(dir->i_sb, err);
1198 errout:
1199 	*error = err;
1200 	return NULL;
1201 }
1202 #endif
1203 
1204 
1205 /*
1206  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1207  * it points to a directory entry which is guaranteed to be large
1208  * enough for new directory entry.  If de is NULL, then
1209  * add_dirent_to_buf will attempt search the directory block for
1210  * space.  It will return -ENOSPC if no space is available, and -EIO
1211  * and -EEXIST if directory entry already exists.
1212  *
1213  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1214  * all other cases bh is released.
1215  */
1216 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1217 			     struct inode *inode, struct ext4_dir_entry_2 *de,
1218 			     struct buffer_head * bh)
1219 {
1220 	struct inode	*dir = dentry->d_parent->d_inode;
1221 	const char	*name = dentry->d_name.name;
1222 	int		namelen = dentry->d_name.len;
1223 	unsigned long	offset = 0;
1224 	unsigned short	reclen;
1225 	int		nlen, rlen, err;
1226 	char		*top;
1227 
1228 	reclen = EXT4_DIR_REC_LEN(namelen);
1229 	if (!de) {
1230 		de = (struct ext4_dir_entry_2 *)bh->b_data;
1231 		top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1232 		while ((char *) de <= top) {
1233 			if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1234 						  bh, offset)) {
1235 				brelse (bh);
1236 				return -EIO;
1237 			}
1238 			if (ext4_match (namelen, name, de)) {
1239 				brelse (bh);
1240 				return -EEXIST;
1241 			}
1242 			nlen = EXT4_DIR_REC_LEN(de->name_len);
1243 			rlen = le16_to_cpu(de->rec_len);
1244 			if ((de->inode? rlen - nlen: rlen) >= reclen)
1245 				break;
1246 			de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1247 			offset += rlen;
1248 		}
1249 		if ((char *) de > top)
1250 			return -ENOSPC;
1251 	}
1252 	BUFFER_TRACE(bh, "get_write_access");
1253 	err = ext4_journal_get_write_access(handle, bh);
1254 	if (err) {
1255 		ext4_std_error(dir->i_sb, err);
1256 		brelse(bh);
1257 		return err;
1258 	}
1259 
1260 	/* By now the buffer is marked for journaling */
1261 	nlen = EXT4_DIR_REC_LEN(de->name_len);
1262 	rlen = le16_to_cpu(de->rec_len);
1263 	if (de->inode) {
1264 		struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1265 		de1->rec_len = cpu_to_le16(rlen - nlen);
1266 		de->rec_len = cpu_to_le16(nlen);
1267 		de = de1;
1268 	}
1269 	de->file_type = EXT4_FT_UNKNOWN;
1270 	if (inode) {
1271 		de->inode = cpu_to_le32(inode->i_ino);
1272 		ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1273 	} else
1274 		de->inode = 0;
1275 	de->name_len = namelen;
1276 	memcpy (de->name, name, namelen);
1277 	/*
1278 	 * XXX shouldn't update any times until successful
1279 	 * completion of syscall, but too many callers depend
1280 	 * on this.
1281 	 *
1282 	 * XXX similarly, too many callers depend on
1283 	 * ext4_new_inode() setting the times, but error
1284 	 * recovery deletes the inode, so the worst that can
1285 	 * happen is that the times are slightly out of date
1286 	 * and/or different from the directory change time.
1287 	 */
1288 	dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1289 	ext4_update_dx_flag(dir);
1290 	dir->i_version++;
1291 	ext4_mark_inode_dirty(handle, dir);
1292 	BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1293 	err = ext4_journal_dirty_metadata(handle, bh);
1294 	if (err)
1295 		ext4_std_error(dir->i_sb, err);
1296 	brelse(bh);
1297 	return 0;
1298 }
1299 
1300 #ifdef CONFIG_EXT4_INDEX
1301 /*
1302  * This converts a one block unindexed directory to a 3 block indexed
1303  * directory, and adds the dentry to the indexed directory.
1304  */
1305 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1306 			    struct inode *inode, struct buffer_head *bh)
1307 {
1308 	struct inode	*dir = dentry->d_parent->d_inode;
1309 	const char	*name = dentry->d_name.name;
1310 	int		namelen = dentry->d_name.len;
1311 	struct buffer_head *bh2;
1312 	struct dx_root	*root;
1313 	struct dx_frame	frames[2], *frame;
1314 	struct dx_entry *entries;
1315 	struct ext4_dir_entry_2	*de, *de2;
1316 	char		*data1, *top;
1317 	unsigned	len;
1318 	int		retval;
1319 	unsigned	blocksize;
1320 	struct dx_hash_info hinfo;
1321 	u32		block;
1322 	struct fake_dirent *fde;
1323 
1324 	blocksize =  dir->i_sb->s_blocksize;
1325 	dxtrace(printk("Creating index\n"));
1326 	retval = ext4_journal_get_write_access(handle, bh);
1327 	if (retval) {
1328 		ext4_std_error(dir->i_sb, retval);
1329 		brelse(bh);
1330 		return retval;
1331 	}
1332 	root = (struct dx_root *) bh->b_data;
1333 
1334 	bh2 = ext4_append (handle, dir, &block, &retval);
1335 	if (!(bh2)) {
1336 		brelse(bh);
1337 		return retval;
1338 	}
1339 	EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1340 	data1 = bh2->b_data;
1341 
1342 	/* The 0th block becomes the root, move the dirents out */
1343 	fde = &root->dotdot;
1344 	de = (struct ext4_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1345 	len = ((char *) root) + blocksize - (char *) de;
1346 	memcpy (data1, de, len);
1347 	de = (struct ext4_dir_entry_2 *) data1;
1348 	top = data1 + len;
1349 	while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1350 		de = de2;
1351 	de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1352 	/* Initialize the root; the dot dirents already exist */
1353 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1354 	de->rec_len = cpu_to_le16(blocksize - EXT4_DIR_REC_LEN(2));
1355 	memset (&root->info, 0, sizeof(root->info));
1356 	root->info.info_length = sizeof(root->info);
1357 	root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1358 	entries = root->entries;
1359 	dx_set_block (entries, 1);
1360 	dx_set_count (entries, 1);
1361 	dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1362 
1363 	/* Initialize as for dx_probe */
1364 	hinfo.hash_version = root->info.hash_version;
1365 	hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1366 	ext4fs_dirhash(name, namelen, &hinfo);
1367 	frame = frames;
1368 	frame->entries = entries;
1369 	frame->at = entries;
1370 	frame->bh = bh;
1371 	bh = bh2;
1372 	de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1373 	dx_release (frames);
1374 	if (!(de))
1375 		return retval;
1376 
1377 	return add_dirent_to_buf(handle, dentry, inode, de, bh);
1378 }
1379 #endif
1380 
1381 /*
1382  *	ext4_add_entry()
1383  *
1384  * adds a file entry to the specified directory, using the same
1385  * semantics as ext4_find_entry(). It returns NULL if it failed.
1386  *
1387  * NOTE!! The inode part of 'de' is left at 0 - which means you
1388  * may not sleep between calling this and putting something into
1389  * the entry, as someone else might have used it while you slept.
1390  */
1391 static int ext4_add_entry (handle_t *handle, struct dentry *dentry,
1392 	struct inode *inode)
1393 {
1394 	struct inode *dir = dentry->d_parent->d_inode;
1395 	unsigned long offset;
1396 	struct buffer_head * bh;
1397 	struct ext4_dir_entry_2 *de;
1398 	struct super_block * sb;
1399 	int	retval;
1400 #ifdef CONFIG_EXT4_INDEX
1401 	int	dx_fallback=0;
1402 #endif
1403 	unsigned blocksize;
1404 	u32 block, blocks;
1405 
1406 	sb = dir->i_sb;
1407 	blocksize = sb->s_blocksize;
1408 	if (!dentry->d_name.len)
1409 		return -EINVAL;
1410 #ifdef CONFIG_EXT4_INDEX
1411 	if (is_dx(dir)) {
1412 		retval = ext4_dx_add_entry(handle, dentry, inode);
1413 		if (!retval || (retval != ERR_BAD_DX_DIR))
1414 			return retval;
1415 		EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1416 		dx_fallback++;
1417 		ext4_mark_inode_dirty(handle, dir);
1418 	}
1419 #endif
1420 	blocks = dir->i_size >> sb->s_blocksize_bits;
1421 	for (block = 0, offset = 0; block < blocks; block++) {
1422 		bh = ext4_bread(handle, dir, block, 0, &retval);
1423 		if(!bh)
1424 			return retval;
1425 		retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1426 		if (retval != -ENOSPC)
1427 			return retval;
1428 
1429 #ifdef CONFIG_EXT4_INDEX
1430 		if (blocks == 1 && !dx_fallback &&
1431 		    EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1432 			return make_indexed_dir(handle, dentry, inode, bh);
1433 #endif
1434 		brelse(bh);
1435 	}
1436 	bh = ext4_append(handle, dir, &block, &retval);
1437 	if (!bh)
1438 		return retval;
1439 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1440 	de->inode = 0;
1441 	de->rec_len = cpu_to_le16(blocksize);
1442 	return add_dirent_to_buf(handle, dentry, inode, de, bh);
1443 }
1444 
1445 #ifdef CONFIG_EXT4_INDEX
1446 /*
1447  * Returns 0 for success, or a negative error value
1448  */
1449 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1450 			     struct inode *inode)
1451 {
1452 	struct dx_frame frames[2], *frame;
1453 	struct dx_entry *entries, *at;
1454 	struct dx_hash_info hinfo;
1455 	struct buffer_head * bh;
1456 	struct inode *dir = dentry->d_parent->d_inode;
1457 	struct super_block * sb = dir->i_sb;
1458 	struct ext4_dir_entry_2 *de;
1459 	int err;
1460 
1461 	frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1462 	if (!frame)
1463 		return err;
1464 	entries = frame->entries;
1465 	at = frame->at;
1466 
1467 	if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1468 		goto cleanup;
1469 
1470 	BUFFER_TRACE(bh, "get_write_access");
1471 	err = ext4_journal_get_write_access(handle, bh);
1472 	if (err)
1473 		goto journal_error;
1474 
1475 	err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1476 	if (err != -ENOSPC) {
1477 		bh = NULL;
1478 		goto cleanup;
1479 	}
1480 
1481 	/* Block full, should compress but for now just split */
1482 	dxtrace(printk("using %u of %u node entries\n",
1483 		       dx_get_count(entries), dx_get_limit(entries)));
1484 	/* Need to split index? */
1485 	if (dx_get_count(entries) == dx_get_limit(entries)) {
1486 		u32 newblock;
1487 		unsigned icount = dx_get_count(entries);
1488 		int levels = frame - frames;
1489 		struct dx_entry *entries2;
1490 		struct dx_node *node2;
1491 		struct buffer_head *bh2;
1492 
1493 		if (levels && (dx_get_count(frames->entries) ==
1494 			       dx_get_limit(frames->entries))) {
1495 			ext4_warning(sb, __FUNCTION__,
1496 				     "Directory index full!");
1497 			err = -ENOSPC;
1498 			goto cleanup;
1499 		}
1500 		bh2 = ext4_append (handle, dir, &newblock, &err);
1501 		if (!(bh2))
1502 			goto cleanup;
1503 		node2 = (struct dx_node *)(bh2->b_data);
1504 		entries2 = node2->entries;
1505 		node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1506 		node2->fake.inode = 0;
1507 		BUFFER_TRACE(frame->bh, "get_write_access");
1508 		err = ext4_journal_get_write_access(handle, frame->bh);
1509 		if (err)
1510 			goto journal_error;
1511 		if (levels) {
1512 			unsigned icount1 = icount/2, icount2 = icount - icount1;
1513 			unsigned hash2 = dx_get_hash(entries + icount1);
1514 			dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1515 
1516 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1517 			err = ext4_journal_get_write_access(handle,
1518 							     frames[0].bh);
1519 			if (err)
1520 				goto journal_error;
1521 
1522 			memcpy ((char *) entries2, (char *) (entries + icount1),
1523 				icount2 * sizeof(struct dx_entry));
1524 			dx_set_count (entries, icount1);
1525 			dx_set_count (entries2, icount2);
1526 			dx_set_limit (entries2, dx_node_limit(dir));
1527 
1528 			/* Which index block gets the new entry? */
1529 			if (at - entries >= icount1) {
1530 				frame->at = at = at - entries - icount1 + entries2;
1531 				frame->entries = entries = entries2;
1532 				swap(frame->bh, bh2);
1533 			}
1534 			dx_insert_block (frames + 0, hash2, newblock);
1535 			dxtrace(dx_show_index ("node", frames[1].entries));
1536 			dxtrace(dx_show_index ("node",
1537 			       ((struct dx_node *) bh2->b_data)->entries));
1538 			err = ext4_journal_dirty_metadata(handle, bh2);
1539 			if (err)
1540 				goto journal_error;
1541 			brelse (bh2);
1542 		} else {
1543 			dxtrace(printk("Creating second level index...\n"));
1544 			memcpy((char *) entries2, (char *) entries,
1545 			       icount * sizeof(struct dx_entry));
1546 			dx_set_limit(entries2, dx_node_limit(dir));
1547 
1548 			/* Set up root */
1549 			dx_set_count(entries, 1);
1550 			dx_set_block(entries + 0, newblock);
1551 			((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1552 
1553 			/* Add new access path frame */
1554 			frame = frames + 1;
1555 			frame->at = at = at - entries + entries2;
1556 			frame->entries = entries = entries2;
1557 			frame->bh = bh2;
1558 			err = ext4_journal_get_write_access(handle,
1559 							     frame->bh);
1560 			if (err)
1561 				goto journal_error;
1562 		}
1563 		ext4_journal_dirty_metadata(handle, frames[0].bh);
1564 	}
1565 	de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1566 	if (!de)
1567 		goto cleanup;
1568 	err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1569 	bh = NULL;
1570 	goto cleanup;
1571 
1572 journal_error:
1573 	ext4_std_error(dir->i_sb, err);
1574 cleanup:
1575 	if (bh)
1576 		brelse(bh);
1577 	dx_release(frames);
1578 	return err;
1579 }
1580 #endif
1581 
1582 /*
1583  * ext4_delete_entry deletes a directory entry by merging it with the
1584  * previous entry
1585  */
1586 static int ext4_delete_entry (handle_t *handle,
1587 			      struct inode * dir,
1588 			      struct ext4_dir_entry_2 * de_del,
1589 			      struct buffer_head * bh)
1590 {
1591 	struct ext4_dir_entry_2 * de, * pde;
1592 	int i;
1593 
1594 	i = 0;
1595 	pde = NULL;
1596 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1597 	while (i < bh->b_size) {
1598 		if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1599 			return -EIO;
1600 		if (de == de_del)  {
1601 			BUFFER_TRACE(bh, "get_write_access");
1602 			ext4_journal_get_write_access(handle, bh);
1603 			if (pde)
1604 				pde->rec_len =
1605 					cpu_to_le16(le16_to_cpu(pde->rec_len) +
1606 						    le16_to_cpu(de->rec_len));
1607 			else
1608 				de->inode = 0;
1609 			dir->i_version++;
1610 			BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1611 			ext4_journal_dirty_metadata(handle, bh);
1612 			return 0;
1613 		}
1614 		i += le16_to_cpu(de->rec_len);
1615 		pde = de;
1616 		de = (struct ext4_dir_entry_2 *)
1617 			((char *) de + le16_to_cpu(de->rec_len));
1618 	}
1619 	return -ENOENT;
1620 }
1621 
1622 static int ext4_add_nondir(handle_t *handle,
1623 		struct dentry *dentry, struct inode *inode)
1624 {
1625 	int err = ext4_add_entry(handle, dentry, inode);
1626 	if (!err) {
1627 		ext4_mark_inode_dirty(handle, inode);
1628 		d_instantiate(dentry, inode);
1629 		return 0;
1630 	}
1631 	drop_nlink(inode);
1632 	iput(inode);
1633 	return err;
1634 }
1635 
1636 /*
1637  * By the time this is called, we already have created
1638  * the directory cache entry for the new file, but it
1639  * is so far negative - it has no inode.
1640  *
1641  * If the create succeeds, we fill in the inode information
1642  * with d_instantiate().
1643  */
1644 static int ext4_create (struct inode * dir, struct dentry * dentry, int mode,
1645 		struct nameidata *nd)
1646 {
1647 	handle_t *handle;
1648 	struct inode * inode;
1649 	int err, retries = 0;
1650 
1651 retry:
1652 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1653 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1654 					2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1655 	if (IS_ERR(handle))
1656 		return PTR_ERR(handle);
1657 
1658 	if (IS_DIRSYNC(dir))
1659 		handle->h_sync = 1;
1660 
1661 	inode = ext4_new_inode (handle, dir, mode);
1662 	err = PTR_ERR(inode);
1663 	if (!IS_ERR(inode)) {
1664 		inode->i_op = &ext4_file_inode_operations;
1665 		inode->i_fop = &ext4_file_operations;
1666 		ext4_set_aops(inode);
1667 		err = ext4_add_nondir(handle, dentry, inode);
1668 	}
1669 	ext4_journal_stop(handle);
1670 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1671 		goto retry;
1672 	return err;
1673 }
1674 
1675 static int ext4_mknod (struct inode * dir, struct dentry *dentry,
1676 			int mode, dev_t rdev)
1677 {
1678 	handle_t *handle;
1679 	struct inode *inode;
1680 	int err, retries = 0;
1681 
1682 	if (!new_valid_dev(rdev))
1683 		return -EINVAL;
1684 
1685 retry:
1686 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1687 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1688 					2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1689 	if (IS_ERR(handle))
1690 		return PTR_ERR(handle);
1691 
1692 	if (IS_DIRSYNC(dir))
1693 		handle->h_sync = 1;
1694 
1695 	inode = ext4_new_inode (handle, dir, mode);
1696 	err = PTR_ERR(inode);
1697 	if (!IS_ERR(inode)) {
1698 		init_special_inode(inode, inode->i_mode, rdev);
1699 #ifdef CONFIG_EXT4DEV_FS_XATTR
1700 		inode->i_op = &ext4_special_inode_operations;
1701 #endif
1702 		err = ext4_add_nondir(handle, dentry, inode);
1703 	}
1704 	ext4_journal_stop(handle);
1705 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1706 		goto retry;
1707 	return err;
1708 }
1709 
1710 static int ext4_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1711 {
1712 	handle_t *handle;
1713 	struct inode * inode;
1714 	struct buffer_head * dir_block;
1715 	struct ext4_dir_entry_2 * de;
1716 	int err, retries = 0;
1717 
1718 	if (dir->i_nlink >= EXT4_LINK_MAX)
1719 		return -EMLINK;
1720 
1721 retry:
1722 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1723 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1724 					2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1725 	if (IS_ERR(handle))
1726 		return PTR_ERR(handle);
1727 
1728 	if (IS_DIRSYNC(dir))
1729 		handle->h_sync = 1;
1730 
1731 	inode = ext4_new_inode (handle, dir, S_IFDIR | mode);
1732 	err = PTR_ERR(inode);
1733 	if (IS_ERR(inode))
1734 		goto out_stop;
1735 
1736 	inode->i_op = &ext4_dir_inode_operations;
1737 	inode->i_fop = &ext4_dir_operations;
1738 	inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1739 	dir_block = ext4_bread (handle, inode, 0, 1, &err);
1740 	if (!dir_block) {
1741 		drop_nlink(inode); /* is this nlink == 0? */
1742 		ext4_mark_inode_dirty(handle, inode);
1743 		iput (inode);
1744 		goto out_stop;
1745 	}
1746 	BUFFER_TRACE(dir_block, "get_write_access");
1747 	ext4_journal_get_write_access(handle, dir_block);
1748 	de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1749 	de->inode = cpu_to_le32(inode->i_ino);
1750 	de->name_len = 1;
1751 	de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de->name_len));
1752 	strcpy (de->name, ".");
1753 	ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1754 	de = (struct ext4_dir_entry_2 *)
1755 			((char *) de + le16_to_cpu(de->rec_len));
1756 	de->inode = cpu_to_le32(dir->i_ino);
1757 	de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT4_DIR_REC_LEN(1));
1758 	de->name_len = 2;
1759 	strcpy (de->name, "..");
1760 	ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1761 	inode->i_nlink = 2;
1762 	BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata");
1763 	ext4_journal_dirty_metadata(handle, dir_block);
1764 	brelse (dir_block);
1765 	ext4_mark_inode_dirty(handle, inode);
1766 	err = ext4_add_entry (handle, dentry, inode);
1767 	if (err) {
1768 		inode->i_nlink = 0;
1769 		ext4_mark_inode_dirty(handle, inode);
1770 		iput (inode);
1771 		goto out_stop;
1772 	}
1773 	inc_nlink(dir);
1774 	ext4_update_dx_flag(dir);
1775 	ext4_mark_inode_dirty(handle, dir);
1776 	d_instantiate(dentry, inode);
1777 out_stop:
1778 	ext4_journal_stop(handle);
1779 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1780 		goto retry;
1781 	return err;
1782 }
1783 
1784 /*
1785  * routine to check that the specified directory is empty (for rmdir)
1786  */
1787 static int empty_dir (struct inode * inode)
1788 {
1789 	unsigned long offset;
1790 	struct buffer_head * bh;
1791 	struct ext4_dir_entry_2 * de, * de1;
1792 	struct super_block * sb;
1793 	int err = 0;
1794 
1795 	sb = inode->i_sb;
1796 	if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1797 	    !(bh = ext4_bread (NULL, inode, 0, 0, &err))) {
1798 		if (err)
1799 			ext4_error(inode->i_sb, __FUNCTION__,
1800 				   "error %d reading directory #%lu offset 0",
1801 				   err, inode->i_ino);
1802 		else
1803 			ext4_warning(inode->i_sb, __FUNCTION__,
1804 				     "bad directory (dir #%lu) - no data block",
1805 				     inode->i_ino);
1806 		return 1;
1807 	}
1808 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1809 	de1 = (struct ext4_dir_entry_2 *)
1810 			((char *) de + le16_to_cpu(de->rec_len));
1811 	if (le32_to_cpu(de->inode) != inode->i_ino ||
1812 			!le32_to_cpu(de1->inode) ||
1813 			strcmp (".", de->name) ||
1814 			strcmp ("..", de1->name)) {
1815 		ext4_warning (inode->i_sb, "empty_dir",
1816 			      "bad directory (dir #%lu) - no `.' or `..'",
1817 			      inode->i_ino);
1818 		brelse (bh);
1819 		return 1;
1820 	}
1821 	offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1822 	de = (struct ext4_dir_entry_2 *)
1823 			((char *) de1 + le16_to_cpu(de1->rec_len));
1824 	while (offset < inode->i_size ) {
1825 		if (!bh ||
1826 			(void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1827 			err = 0;
1828 			brelse (bh);
1829 			bh = ext4_bread (NULL, inode,
1830 				offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1831 			if (!bh) {
1832 				if (err)
1833 					ext4_error(sb, __FUNCTION__,
1834 						   "error %d reading directory"
1835 						   " #%lu offset %lu",
1836 						   err, inode->i_ino, offset);
1837 				offset += sb->s_blocksize;
1838 				continue;
1839 			}
1840 			de = (struct ext4_dir_entry_2 *) bh->b_data;
1841 		}
1842 		if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1843 			de = (struct ext4_dir_entry_2 *)(bh->b_data +
1844 							 sb->s_blocksize);
1845 			offset = (offset | (sb->s_blocksize - 1)) + 1;
1846 			continue;
1847 		}
1848 		if (le32_to_cpu(de->inode)) {
1849 			brelse (bh);
1850 			return 0;
1851 		}
1852 		offset += le16_to_cpu(de->rec_len);
1853 		de = (struct ext4_dir_entry_2 *)
1854 				((char *) de + le16_to_cpu(de->rec_len));
1855 	}
1856 	brelse (bh);
1857 	return 1;
1858 }
1859 
1860 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1861  * such inodes, starting at the superblock, in case we crash before the
1862  * file is closed/deleted, or in case the inode truncate spans multiple
1863  * transactions and the last transaction is not recovered after a crash.
1864  *
1865  * At filesystem recovery time, we walk this list deleting unlinked
1866  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1867  */
1868 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1869 {
1870 	struct super_block *sb = inode->i_sb;
1871 	struct ext4_iloc iloc;
1872 	int err = 0, rc;
1873 
1874 	lock_super(sb);
1875 	if (!list_empty(&EXT4_I(inode)->i_orphan))
1876 		goto out_unlock;
1877 
1878 	/* Orphan handling is only valid for files with data blocks
1879 	 * being truncated, or files being unlinked. */
1880 
1881 	/* @@@ FIXME: Observation from aviro:
1882 	 * I think I can trigger J_ASSERT in ext4_orphan_add().  We block
1883 	 * here (on lock_super()), so race with ext4_link() which might bump
1884 	 * ->i_nlink. For, say it, character device. Not a regular file,
1885 	 * not a directory, not a symlink and ->i_nlink > 0.
1886 	 */
1887 	J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1888 		S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1889 
1890 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1891 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1892 	if (err)
1893 		goto out_unlock;
1894 
1895 	err = ext4_reserve_inode_write(handle, inode, &iloc);
1896 	if (err)
1897 		goto out_unlock;
1898 
1899 	/* Insert this inode at the head of the on-disk orphan list... */
1900 	NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1901 	EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1902 	err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1903 	rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
1904 	if (!err)
1905 		err = rc;
1906 
1907 	/* Only add to the head of the in-memory list if all the
1908 	 * previous operations succeeded.  If the orphan_add is going to
1909 	 * fail (possibly taking the journal offline), we can't risk
1910 	 * leaving the inode on the orphan list: stray orphan-list
1911 	 * entries can cause panics at unmount time.
1912 	 *
1913 	 * This is safe: on error we're going to ignore the orphan list
1914 	 * anyway on the next recovery. */
1915 	if (!err)
1916 		list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1917 
1918 	jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1919 	jbd_debug(4, "orphan inode %lu will point to %d\n",
1920 			inode->i_ino, NEXT_ORPHAN(inode));
1921 out_unlock:
1922 	unlock_super(sb);
1923 	ext4_std_error(inode->i_sb, err);
1924 	return err;
1925 }
1926 
1927 /*
1928  * ext4_orphan_del() removes an unlinked or truncated inode from the list
1929  * of such inodes stored on disk, because it is finally being cleaned up.
1930  */
1931 int ext4_orphan_del(handle_t *handle, struct inode *inode)
1932 {
1933 	struct list_head *prev;
1934 	struct ext4_inode_info *ei = EXT4_I(inode);
1935 	struct ext4_sb_info *sbi;
1936 	unsigned long ino_next;
1937 	struct ext4_iloc iloc;
1938 	int err = 0;
1939 
1940 	lock_super(inode->i_sb);
1941 	if (list_empty(&ei->i_orphan)) {
1942 		unlock_super(inode->i_sb);
1943 		return 0;
1944 	}
1945 
1946 	ino_next = NEXT_ORPHAN(inode);
1947 	prev = ei->i_orphan.prev;
1948 	sbi = EXT4_SB(inode->i_sb);
1949 
1950 	jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1951 
1952 	list_del_init(&ei->i_orphan);
1953 
1954 	/* If we're on an error path, we may not have a valid
1955 	 * transaction handle with which to update the orphan list on
1956 	 * disk, but we still need to remove the inode from the linked
1957 	 * list in memory. */
1958 	if (!handle)
1959 		goto out;
1960 
1961 	err = ext4_reserve_inode_write(handle, inode, &iloc);
1962 	if (err)
1963 		goto out_err;
1964 
1965 	if (prev == &sbi->s_orphan) {
1966 		jbd_debug(4, "superblock will point to %lu\n", ino_next);
1967 		BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1968 		err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1969 		if (err)
1970 			goto out_brelse;
1971 		sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1972 		err = ext4_journal_dirty_metadata(handle, sbi->s_sbh);
1973 	} else {
1974 		struct ext4_iloc iloc2;
1975 		struct inode *i_prev =
1976 			&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
1977 
1978 		jbd_debug(4, "orphan inode %lu will point to %lu\n",
1979 			  i_prev->i_ino, ino_next);
1980 		err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
1981 		if (err)
1982 			goto out_brelse;
1983 		NEXT_ORPHAN(i_prev) = ino_next;
1984 		err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
1985 	}
1986 	if (err)
1987 		goto out_brelse;
1988 	NEXT_ORPHAN(inode) = 0;
1989 	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
1990 
1991 out_err:
1992 	ext4_std_error(inode->i_sb, err);
1993 out:
1994 	unlock_super(inode->i_sb);
1995 	return err;
1996 
1997 out_brelse:
1998 	brelse(iloc.bh);
1999 	goto out_err;
2000 }
2001 
2002 static int ext4_rmdir (struct inode * dir, struct dentry *dentry)
2003 {
2004 	int retval;
2005 	struct inode * inode;
2006 	struct buffer_head * bh;
2007 	struct ext4_dir_entry_2 * de;
2008 	handle_t *handle;
2009 
2010 	/* Initialize quotas before so that eventual writes go in
2011 	 * separate transaction */
2012 	DQUOT_INIT(dentry->d_inode);
2013 	handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2014 	if (IS_ERR(handle))
2015 		return PTR_ERR(handle);
2016 
2017 	retval = -ENOENT;
2018 	bh = ext4_find_entry (dentry, &de);
2019 	if (!bh)
2020 		goto end_rmdir;
2021 
2022 	if (IS_DIRSYNC(dir))
2023 		handle->h_sync = 1;
2024 
2025 	inode = dentry->d_inode;
2026 
2027 	retval = -EIO;
2028 	if (le32_to_cpu(de->inode) != inode->i_ino)
2029 		goto end_rmdir;
2030 
2031 	retval = -ENOTEMPTY;
2032 	if (!empty_dir (inode))
2033 		goto end_rmdir;
2034 
2035 	retval = ext4_delete_entry(handle, dir, de, bh);
2036 	if (retval)
2037 		goto end_rmdir;
2038 	if (inode->i_nlink != 2)
2039 		ext4_warning (inode->i_sb, "ext4_rmdir",
2040 			      "empty directory has nlink!=2 (%d)",
2041 			      inode->i_nlink);
2042 	inode->i_version++;
2043 	clear_nlink(inode);
2044 	/* There's no need to set i_disksize: the fact that i_nlink is
2045 	 * zero will ensure that the right thing happens during any
2046 	 * recovery. */
2047 	inode->i_size = 0;
2048 	ext4_orphan_add(handle, inode);
2049 	inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2050 	ext4_mark_inode_dirty(handle, inode);
2051 	drop_nlink(dir);
2052 	ext4_update_dx_flag(dir);
2053 	ext4_mark_inode_dirty(handle, dir);
2054 
2055 end_rmdir:
2056 	ext4_journal_stop(handle);
2057 	brelse (bh);
2058 	return retval;
2059 }
2060 
2061 static int ext4_unlink(struct inode * dir, struct dentry *dentry)
2062 {
2063 	int retval;
2064 	struct inode * inode;
2065 	struct buffer_head * bh;
2066 	struct ext4_dir_entry_2 * de;
2067 	handle_t *handle;
2068 
2069 	/* Initialize quotas before so that eventual writes go
2070 	 * in separate transaction */
2071 	DQUOT_INIT(dentry->d_inode);
2072 	handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2073 	if (IS_ERR(handle))
2074 		return PTR_ERR(handle);
2075 
2076 	if (IS_DIRSYNC(dir))
2077 		handle->h_sync = 1;
2078 
2079 	retval = -ENOENT;
2080 	bh = ext4_find_entry (dentry, &de);
2081 	if (!bh)
2082 		goto end_unlink;
2083 
2084 	inode = dentry->d_inode;
2085 
2086 	retval = -EIO;
2087 	if (le32_to_cpu(de->inode) != inode->i_ino)
2088 		goto end_unlink;
2089 
2090 	if (!inode->i_nlink) {
2091 		ext4_warning (inode->i_sb, "ext4_unlink",
2092 			      "Deleting nonexistent file (%lu), %d",
2093 			      inode->i_ino, inode->i_nlink);
2094 		inode->i_nlink = 1;
2095 	}
2096 	retval = ext4_delete_entry(handle, dir, de, bh);
2097 	if (retval)
2098 		goto end_unlink;
2099 	dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2100 	ext4_update_dx_flag(dir);
2101 	ext4_mark_inode_dirty(handle, dir);
2102 	drop_nlink(inode);
2103 	if (!inode->i_nlink)
2104 		ext4_orphan_add(handle, inode);
2105 	inode->i_ctime = dir->i_ctime;
2106 	ext4_mark_inode_dirty(handle, inode);
2107 	retval = 0;
2108 
2109 end_unlink:
2110 	ext4_journal_stop(handle);
2111 	brelse (bh);
2112 	return retval;
2113 }
2114 
2115 static int ext4_symlink (struct inode * dir,
2116 		struct dentry *dentry, const char * symname)
2117 {
2118 	handle_t *handle;
2119 	struct inode * inode;
2120 	int l, err, retries = 0;
2121 
2122 	l = strlen(symname)+1;
2123 	if (l > dir->i_sb->s_blocksize)
2124 		return -ENAMETOOLONG;
2125 
2126 retry:
2127 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2128 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2129 					2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2130 	if (IS_ERR(handle))
2131 		return PTR_ERR(handle);
2132 
2133 	if (IS_DIRSYNC(dir))
2134 		handle->h_sync = 1;
2135 
2136 	inode = ext4_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2137 	err = PTR_ERR(inode);
2138 	if (IS_ERR(inode))
2139 		goto out_stop;
2140 
2141 	if (l > sizeof (EXT4_I(inode)->i_data)) {
2142 		inode->i_op = &ext4_symlink_inode_operations;
2143 		ext4_set_aops(inode);
2144 		/*
2145 		 * page_symlink() calls into ext4_prepare/commit_write.
2146 		 * We have a transaction open.  All is sweetness.  It also sets
2147 		 * i_size in generic_commit_write().
2148 		 */
2149 		err = __page_symlink(inode, symname, l,
2150 				mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2151 		if (err) {
2152 			drop_nlink(inode);
2153 			ext4_mark_inode_dirty(handle, inode);
2154 			iput (inode);
2155 			goto out_stop;
2156 		}
2157 	} else {
2158 		inode->i_op = &ext4_fast_symlink_inode_operations;
2159 		memcpy((char*)&EXT4_I(inode)->i_data,symname,l);
2160 		inode->i_size = l-1;
2161 	}
2162 	EXT4_I(inode)->i_disksize = inode->i_size;
2163 	err = ext4_add_nondir(handle, dentry, inode);
2164 out_stop:
2165 	ext4_journal_stop(handle);
2166 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2167 		goto retry;
2168 	return err;
2169 }
2170 
2171 static int ext4_link (struct dentry * old_dentry,
2172 		struct inode * dir, struct dentry *dentry)
2173 {
2174 	handle_t *handle;
2175 	struct inode *inode = old_dentry->d_inode;
2176 	int err, retries = 0;
2177 
2178 	if (inode->i_nlink >= EXT4_LINK_MAX)
2179 		return -EMLINK;
2180 	/*
2181 	 * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2182 	 * otherwise has the potential to corrupt the orphan inode list.
2183 	 */
2184 	if (inode->i_nlink == 0)
2185 		return -ENOENT;
2186 
2187 retry:
2188 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2189 					EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2190 	if (IS_ERR(handle))
2191 		return PTR_ERR(handle);
2192 
2193 	if (IS_DIRSYNC(dir))
2194 		handle->h_sync = 1;
2195 
2196 	inode->i_ctime = CURRENT_TIME_SEC;
2197 	inc_nlink(inode);
2198 	atomic_inc(&inode->i_count);
2199 
2200 	err = ext4_add_nondir(handle, dentry, inode);
2201 	ext4_journal_stop(handle);
2202 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2203 		goto retry;
2204 	return err;
2205 }
2206 
2207 #define PARENT_INO(buffer) \
2208 	((struct ext4_dir_entry_2 *) ((char *) buffer + \
2209 	le16_to_cpu(((struct ext4_dir_entry_2 *) buffer)->rec_len)))->inode
2210 
2211 /*
2212  * Anybody can rename anything with this: the permission checks are left to the
2213  * higher-level routines.
2214  */
2215 static int ext4_rename (struct inode * old_dir, struct dentry *old_dentry,
2216 			   struct inode * new_dir,struct dentry *new_dentry)
2217 {
2218 	handle_t *handle;
2219 	struct inode * old_inode, * new_inode;
2220 	struct buffer_head * old_bh, * new_bh, * dir_bh;
2221 	struct ext4_dir_entry_2 * old_de, * new_de;
2222 	int retval;
2223 
2224 	old_bh = new_bh = dir_bh = NULL;
2225 
2226 	/* Initialize quotas before so that eventual writes go
2227 	 * in separate transaction */
2228 	if (new_dentry->d_inode)
2229 		DQUOT_INIT(new_dentry->d_inode);
2230 	handle = ext4_journal_start(old_dir, 2 *
2231 					EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2232 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2233 	if (IS_ERR(handle))
2234 		return PTR_ERR(handle);
2235 
2236 	if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2237 		handle->h_sync = 1;
2238 
2239 	old_bh = ext4_find_entry (old_dentry, &old_de);
2240 	/*
2241 	 *  Check for inode number is _not_ due to possible IO errors.
2242 	 *  We might rmdir the source, keep it as pwd of some process
2243 	 *  and merrily kill the link to whatever was created under the
2244 	 *  same name. Goodbye sticky bit ;-<
2245 	 */
2246 	old_inode = old_dentry->d_inode;
2247 	retval = -ENOENT;
2248 	if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2249 		goto end_rename;
2250 
2251 	new_inode = new_dentry->d_inode;
2252 	new_bh = ext4_find_entry (new_dentry, &new_de);
2253 	if (new_bh) {
2254 		if (!new_inode) {
2255 			brelse (new_bh);
2256 			new_bh = NULL;
2257 		}
2258 	}
2259 	if (S_ISDIR(old_inode->i_mode)) {
2260 		if (new_inode) {
2261 			retval = -ENOTEMPTY;
2262 			if (!empty_dir (new_inode))
2263 				goto end_rename;
2264 		}
2265 		retval = -EIO;
2266 		dir_bh = ext4_bread (handle, old_inode, 0, 0, &retval);
2267 		if (!dir_bh)
2268 			goto end_rename;
2269 		if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2270 			goto end_rename;
2271 		retval = -EMLINK;
2272 		if (!new_inode && new_dir!=old_dir &&
2273 				new_dir->i_nlink >= EXT4_LINK_MAX)
2274 			goto end_rename;
2275 	}
2276 	if (!new_bh) {
2277 		retval = ext4_add_entry (handle, new_dentry, old_inode);
2278 		if (retval)
2279 			goto end_rename;
2280 	} else {
2281 		BUFFER_TRACE(new_bh, "get write access");
2282 		ext4_journal_get_write_access(handle, new_bh);
2283 		new_de->inode = cpu_to_le32(old_inode->i_ino);
2284 		if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2285 					      EXT4_FEATURE_INCOMPAT_FILETYPE))
2286 			new_de->file_type = old_de->file_type;
2287 		new_dir->i_version++;
2288 		BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata");
2289 		ext4_journal_dirty_metadata(handle, new_bh);
2290 		brelse(new_bh);
2291 		new_bh = NULL;
2292 	}
2293 
2294 	/*
2295 	 * Like most other Unix systems, set the ctime for inodes on a
2296 	 * rename.
2297 	 */
2298 	old_inode->i_ctime = CURRENT_TIME_SEC;
2299 	ext4_mark_inode_dirty(handle, old_inode);
2300 
2301 	/*
2302 	 * ok, that's it
2303 	 */
2304 	if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2305 	    old_de->name_len != old_dentry->d_name.len ||
2306 	    strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2307 	    (retval = ext4_delete_entry(handle, old_dir,
2308 					old_de, old_bh)) == -ENOENT) {
2309 		/* old_de could have moved from under us during htree split, so
2310 		 * make sure that we are deleting the right entry.  We might
2311 		 * also be pointing to a stale entry in the unused part of
2312 		 * old_bh so just checking inum and the name isn't enough. */
2313 		struct buffer_head *old_bh2;
2314 		struct ext4_dir_entry_2 *old_de2;
2315 
2316 		old_bh2 = ext4_find_entry(old_dentry, &old_de2);
2317 		if (old_bh2) {
2318 			retval = ext4_delete_entry(handle, old_dir,
2319 						   old_de2, old_bh2);
2320 			brelse(old_bh2);
2321 		}
2322 	}
2323 	if (retval) {
2324 		ext4_warning(old_dir->i_sb, "ext4_rename",
2325 				"Deleting old file (%lu), %d, error=%d",
2326 				old_dir->i_ino, old_dir->i_nlink, retval);
2327 	}
2328 
2329 	if (new_inode) {
2330 		drop_nlink(new_inode);
2331 		new_inode->i_ctime = CURRENT_TIME_SEC;
2332 	}
2333 	old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2334 	ext4_update_dx_flag(old_dir);
2335 	if (dir_bh) {
2336 		BUFFER_TRACE(dir_bh, "get_write_access");
2337 		ext4_journal_get_write_access(handle, dir_bh);
2338 		PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2339 		BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata");
2340 		ext4_journal_dirty_metadata(handle, dir_bh);
2341 		drop_nlink(old_dir);
2342 		if (new_inode) {
2343 			drop_nlink(new_inode);
2344 		} else {
2345 			inc_nlink(new_dir);
2346 			ext4_update_dx_flag(new_dir);
2347 			ext4_mark_inode_dirty(handle, new_dir);
2348 		}
2349 	}
2350 	ext4_mark_inode_dirty(handle, old_dir);
2351 	if (new_inode) {
2352 		ext4_mark_inode_dirty(handle, new_inode);
2353 		if (!new_inode->i_nlink)
2354 			ext4_orphan_add(handle, new_inode);
2355 	}
2356 	retval = 0;
2357 
2358 end_rename:
2359 	brelse (dir_bh);
2360 	brelse (old_bh);
2361 	brelse (new_bh);
2362 	ext4_journal_stop(handle);
2363 	return retval;
2364 }
2365 
2366 /*
2367  * directories can handle most operations...
2368  */
2369 const struct inode_operations ext4_dir_inode_operations = {
2370 	.create		= ext4_create,
2371 	.lookup		= ext4_lookup,
2372 	.link		= ext4_link,
2373 	.unlink		= ext4_unlink,
2374 	.symlink	= ext4_symlink,
2375 	.mkdir		= ext4_mkdir,
2376 	.rmdir		= ext4_rmdir,
2377 	.mknod		= ext4_mknod,
2378 	.rename		= ext4_rename,
2379 	.setattr	= ext4_setattr,
2380 #ifdef CONFIG_EXT4DEV_FS_XATTR
2381 	.setxattr	= generic_setxattr,
2382 	.getxattr	= generic_getxattr,
2383 	.listxattr	= ext4_listxattr,
2384 	.removexattr	= generic_removexattr,
2385 #endif
2386 	.permission	= ext4_permission,
2387 };
2388 
2389 const struct inode_operations ext4_special_inode_operations = {
2390 	.setattr	= ext4_setattr,
2391 #ifdef CONFIG_EXT4DEV_FS_XATTR
2392 	.setxattr	= generic_setxattr,
2393 	.getxattr	= generic_getxattr,
2394 	.listxattr	= ext4_listxattr,
2395 	.removexattr	= generic_removexattr,
2396 #endif
2397 	.permission	= ext4_permission,
2398 };
2399