xref: /linux/fs/ext4/namei.c (revision b889fcf63cb62e7fdb7816565e28f44dbe4a76a5)
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/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include "ext4.h"
38 #include "ext4_jbd2.h"
39 
40 #include "xattr.h"
41 #include "acl.h"
42 
43 #include <trace/events/ext4.h>
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 					ext4_lblk_t *block, int *err)
55 {
56 	struct buffer_head *bh;
57 
58 	if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
59 		     ((inode->i_size >> 10) >=
60 		      EXT4_SB(inode->i_sb)->s_max_dir_size_kb))) {
61 		*err = -ENOSPC;
62 		return NULL;
63 	}
64 
65 	*block = inode->i_size >> inode->i_sb->s_blocksize_bits;
66 
67 	bh = ext4_bread(handle, inode, *block, 1, err);
68 	if (bh) {
69 		inode->i_size += inode->i_sb->s_blocksize;
70 		EXT4_I(inode)->i_disksize = inode->i_size;
71 		*err = ext4_journal_get_write_access(handle, bh);
72 		if (*err) {
73 			brelse(bh);
74 			bh = NULL;
75 		}
76 	}
77 	if (!bh && !(*err)) {
78 		*err = -EIO;
79 		ext4_error(inode->i_sb,
80 			   "Directory hole detected on inode %lu\n",
81 			   inode->i_ino);
82 	}
83 	return bh;
84 }
85 
86 #ifndef assert
87 #define assert(test) J_ASSERT(test)
88 #endif
89 
90 #ifdef DX_DEBUG
91 #define dxtrace(command) command
92 #else
93 #define dxtrace(command)
94 #endif
95 
96 struct fake_dirent
97 {
98 	__le32 inode;
99 	__le16 rec_len;
100 	u8 name_len;
101 	u8 file_type;
102 };
103 
104 struct dx_countlimit
105 {
106 	__le16 limit;
107 	__le16 count;
108 };
109 
110 struct dx_entry
111 {
112 	__le32 hash;
113 	__le32 block;
114 };
115 
116 /*
117  * dx_root_info is laid out so that if it should somehow get overlaid by a
118  * dirent the two low bits of the hash version will be zero.  Therefore, the
119  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
120  */
121 
122 struct dx_root
123 {
124 	struct fake_dirent dot;
125 	char dot_name[4];
126 	struct fake_dirent dotdot;
127 	char dotdot_name[4];
128 	struct dx_root_info
129 	{
130 		__le32 reserved_zero;
131 		u8 hash_version;
132 		u8 info_length; /* 8 */
133 		u8 indirect_levels;
134 		u8 unused_flags;
135 	}
136 	info;
137 	struct dx_entry	entries[0];
138 };
139 
140 struct dx_node
141 {
142 	struct fake_dirent fake;
143 	struct dx_entry	entries[0];
144 };
145 
146 
147 struct dx_frame
148 {
149 	struct buffer_head *bh;
150 	struct dx_entry *entries;
151 	struct dx_entry *at;
152 };
153 
154 struct dx_map_entry
155 {
156 	u32 hash;
157 	u16 offs;
158 	u16 size;
159 };
160 
161 /*
162  * This goes at the end of each htree block.
163  */
164 struct dx_tail {
165 	u32 dt_reserved;
166 	__le32 dt_checksum;	/* crc32c(uuid+inum+dirblock) */
167 };
168 
169 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
170 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
171 static inline unsigned dx_get_hash(struct dx_entry *entry);
172 static void dx_set_hash(struct dx_entry *entry, unsigned value);
173 static unsigned dx_get_count(struct dx_entry *entries);
174 static unsigned dx_get_limit(struct dx_entry *entries);
175 static void dx_set_count(struct dx_entry *entries, unsigned value);
176 static void dx_set_limit(struct dx_entry *entries, unsigned value);
177 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
178 static unsigned dx_node_limit(struct inode *dir);
179 static struct dx_frame *dx_probe(const struct qstr *d_name,
180 				 struct inode *dir,
181 				 struct dx_hash_info *hinfo,
182 				 struct dx_frame *frame,
183 				 int *err);
184 static void dx_release(struct dx_frame *frames);
185 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
186 		       struct dx_hash_info *hinfo, struct dx_map_entry map[]);
187 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
188 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
189 		struct dx_map_entry *offsets, int count, unsigned blocksize);
190 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
191 static void dx_insert_block(struct dx_frame *frame,
192 					u32 hash, ext4_lblk_t block);
193 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
194 				 struct dx_frame *frame,
195 				 struct dx_frame *frames,
196 				 __u32 *start_hash);
197 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
198 		const struct qstr *d_name,
199 		struct ext4_dir_entry_2 **res_dir,
200 		int *err);
201 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
202 			     struct inode *inode);
203 
204 /* checksumming functions */
205 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
206 			    unsigned int blocksize)
207 {
208 	memset(t, 0, sizeof(struct ext4_dir_entry_tail));
209 	t->det_rec_len = ext4_rec_len_to_disk(
210 			sizeof(struct ext4_dir_entry_tail), blocksize);
211 	t->det_reserved_ft = EXT4_FT_DIR_CSUM;
212 }
213 
214 /* Walk through a dirent block to find a checksum "dirent" at the tail */
215 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
216 						   struct ext4_dir_entry *de)
217 {
218 	struct ext4_dir_entry_tail *t;
219 
220 #ifdef PARANOID
221 	struct ext4_dir_entry *d, *top;
222 
223 	d = de;
224 	top = (struct ext4_dir_entry *)(((void *)de) +
225 		(EXT4_BLOCK_SIZE(inode->i_sb) -
226 		sizeof(struct ext4_dir_entry_tail)));
227 	while (d < top && d->rec_len)
228 		d = (struct ext4_dir_entry *)(((void *)d) +
229 		    le16_to_cpu(d->rec_len));
230 
231 	if (d != top)
232 		return NULL;
233 
234 	t = (struct ext4_dir_entry_tail *)d;
235 #else
236 	t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
237 #endif
238 
239 	if (t->det_reserved_zero1 ||
240 	    le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
241 	    t->det_reserved_zero2 ||
242 	    t->det_reserved_ft != EXT4_FT_DIR_CSUM)
243 		return NULL;
244 
245 	return t;
246 }
247 
248 static __le32 ext4_dirent_csum(struct inode *inode,
249 			       struct ext4_dir_entry *dirent, int size)
250 {
251 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
252 	struct ext4_inode_info *ei = EXT4_I(inode);
253 	__u32 csum;
254 
255 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
256 	return cpu_to_le32(csum);
257 }
258 
259 static void warn_no_space_for_csum(struct inode *inode)
260 {
261 	ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
262 		     "checksum.  Please run e2fsck -D.", inode->i_ino);
263 }
264 
265 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
266 {
267 	struct ext4_dir_entry_tail *t;
268 
269 	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
270 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
271 		return 1;
272 
273 	t = get_dirent_tail(inode, dirent);
274 	if (!t) {
275 		warn_no_space_for_csum(inode);
276 		return 0;
277 	}
278 
279 	if (t->det_checksum != ext4_dirent_csum(inode, dirent,
280 						(void *)t - (void *)dirent))
281 		return 0;
282 
283 	return 1;
284 }
285 
286 static void ext4_dirent_csum_set(struct inode *inode,
287 				 struct ext4_dir_entry *dirent)
288 {
289 	struct ext4_dir_entry_tail *t;
290 
291 	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
292 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
293 		return;
294 
295 	t = get_dirent_tail(inode, dirent);
296 	if (!t) {
297 		warn_no_space_for_csum(inode);
298 		return;
299 	}
300 
301 	t->det_checksum = ext4_dirent_csum(inode, dirent,
302 					   (void *)t - (void *)dirent);
303 }
304 
305 int ext4_handle_dirty_dirent_node(handle_t *handle,
306 				  struct inode *inode,
307 				  struct buffer_head *bh)
308 {
309 	ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
310 	return ext4_handle_dirty_metadata(handle, inode, bh);
311 }
312 
313 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
314 					       struct ext4_dir_entry *dirent,
315 					       int *offset)
316 {
317 	struct ext4_dir_entry *dp;
318 	struct dx_root_info *root;
319 	int count_offset;
320 
321 	if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
322 		count_offset = 8;
323 	else if (le16_to_cpu(dirent->rec_len) == 12) {
324 		dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
325 		if (le16_to_cpu(dp->rec_len) !=
326 		    EXT4_BLOCK_SIZE(inode->i_sb) - 12)
327 			return NULL;
328 		root = (struct dx_root_info *)(((void *)dp + 12));
329 		if (root->reserved_zero ||
330 		    root->info_length != sizeof(struct dx_root_info))
331 			return NULL;
332 		count_offset = 32;
333 	} else
334 		return NULL;
335 
336 	if (offset)
337 		*offset = count_offset;
338 	return (struct dx_countlimit *)(((void *)dirent) + count_offset);
339 }
340 
341 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
342 			   int count_offset, int count, struct dx_tail *t)
343 {
344 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
345 	struct ext4_inode_info *ei = EXT4_I(inode);
346 	__u32 csum, old_csum;
347 	int size;
348 
349 	size = count_offset + (count * sizeof(struct dx_entry));
350 	old_csum = t->dt_checksum;
351 	t->dt_checksum = 0;
352 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
353 	csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
354 	t->dt_checksum = old_csum;
355 
356 	return cpu_to_le32(csum);
357 }
358 
359 static int ext4_dx_csum_verify(struct inode *inode,
360 			       struct ext4_dir_entry *dirent)
361 {
362 	struct dx_countlimit *c;
363 	struct dx_tail *t;
364 	int count_offset, limit, count;
365 
366 	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
367 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
368 		return 1;
369 
370 	c = get_dx_countlimit(inode, dirent, &count_offset);
371 	if (!c) {
372 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
373 		return 1;
374 	}
375 	limit = le16_to_cpu(c->limit);
376 	count = le16_to_cpu(c->count);
377 	if (count_offset + (limit * sizeof(struct dx_entry)) >
378 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
379 		warn_no_space_for_csum(inode);
380 		return 1;
381 	}
382 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
383 
384 	if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
385 					    count, t))
386 		return 0;
387 	return 1;
388 }
389 
390 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
391 {
392 	struct dx_countlimit *c;
393 	struct dx_tail *t;
394 	int count_offset, limit, count;
395 
396 	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
397 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
398 		return;
399 
400 	c = get_dx_countlimit(inode, dirent, &count_offset);
401 	if (!c) {
402 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
403 		return;
404 	}
405 	limit = le16_to_cpu(c->limit);
406 	count = le16_to_cpu(c->count);
407 	if (count_offset + (limit * sizeof(struct dx_entry)) >
408 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
409 		warn_no_space_for_csum(inode);
410 		return;
411 	}
412 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
413 
414 	t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
415 }
416 
417 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
418 					    struct inode *inode,
419 					    struct buffer_head *bh)
420 {
421 	ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
422 	return ext4_handle_dirty_metadata(handle, inode, bh);
423 }
424 
425 /*
426  * p is at least 6 bytes before the end of page
427  */
428 static inline struct ext4_dir_entry_2 *
429 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
430 {
431 	return (struct ext4_dir_entry_2 *)((char *)p +
432 		ext4_rec_len_from_disk(p->rec_len, blocksize));
433 }
434 
435 /*
436  * Future: use high four bits of block for coalesce-on-delete flags
437  * Mask them off for now.
438  */
439 
440 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
441 {
442 	return le32_to_cpu(entry->block) & 0x00ffffff;
443 }
444 
445 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
446 {
447 	entry->block = cpu_to_le32(value);
448 }
449 
450 static inline unsigned dx_get_hash(struct dx_entry *entry)
451 {
452 	return le32_to_cpu(entry->hash);
453 }
454 
455 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
456 {
457 	entry->hash = cpu_to_le32(value);
458 }
459 
460 static inline unsigned dx_get_count(struct dx_entry *entries)
461 {
462 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
463 }
464 
465 static inline unsigned dx_get_limit(struct dx_entry *entries)
466 {
467 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
468 }
469 
470 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
471 {
472 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
473 }
474 
475 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
476 {
477 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
478 }
479 
480 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
481 {
482 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
483 		EXT4_DIR_REC_LEN(2) - infosize;
484 
485 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
486 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
487 		entry_space -= sizeof(struct dx_tail);
488 	return entry_space / sizeof(struct dx_entry);
489 }
490 
491 static inline unsigned dx_node_limit(struct inode *dir)
492 {
493 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
494 
495 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
496 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
497 		entry_space -= sizeof(struct dx_tail);
498 	return entry_space / sizeof(struct dx_entry);
499 }
500 
501 /*
502  * Debug
503  */
504 #ifdef DX_DEBUG
505 static void dx_show_index(char * label, struct dx_entry *entries)
506 {
507 	int i, n = dx_get_count (entries);
508 	printk(KERN_DEBUG "%s index ", label);
509 	for (i = 0; i < n; i++) {
510 		printk("%x->%lu ", i ? dx_get_hash(entries + i) :
511 				0, (unsigned long)dx_get_block(entries + i));
512 	}
513 	printk("\n");
514 }
515 
516 struct stats
517 {
518 	unsigned names;
519 	unsigned space;
520 	unsigned bcount;
521 };
522 
523 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
524 				 int size, int show_names)
525 {
526 	unsigned names = 0, space = 0;
527 	char *base = (char *) de;
528 	struct dx_hash_info h = *hinfo;
529 
530 	printk("names: ");
531 	while ((char *) de < base + size)
532 	{
533 		if (de->inode)
534 		{
535 			if (show_names)
536 			{
537 				int len = de->name_len;
538 				char *name = de->name;
539 				while (len--) printk("%c", *name++);
540 				ext4fs_dirhash(de->name, de->name_len, &h);
541 				printk(":%x.%u ", h.hash,
542 				       (unsigned) ((char *) de - base));
543 			}
544 			space += EXT4_DIR_REC_LEN(de->name_len);
545 			names++;
546 		}
547 		de = ext4_next_entry(de, size);
548 	}
549 	printk("(%i)\n", names);
550 	return (struct stats) { names, space, 1 };
551 }
552 
553 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
554 			     struct dx_entry *entries, int levels)
555 {
556 	unsigned blocksize = dir->i_sb->s_blocksize;
557 	unsigned count = dx_get_count(entries), names = 0, space = 0, i;
558 	unsigned bcount = 0;
559 	struct buffer_head *bh;
560 	int err;
561 	printk("%i indexed blocks...\n", count);
562 	for (i = 0; i < count; i++, entries++)
563 	{
564 		ext4_lblk_t block = dx_get_block(entries);
565 		ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
566 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
567 		struct stats stats;
568 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
569 		if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
570 		stats = levels?
571 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
572 		   dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
573 		names += stats.names;
574 		space += stats.space;
575 		bcount += stats.bcount;
576 		brelse(bh);
577 	}
578 	if (bcount)
579 		printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
580 		       levels ? "" : "   ", names, space/bcount,
581 		       (space/bcount)*100/blocksize);
582 	return (struct stats) { names, space, bcount};
583 }
584 #endif /* DX_DEBUG */
585 
586 /*
587  * Probe for a directory leaf block to search.
588  *
589  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
590  * error in the directory index, and the caller should fall back to
591  * searching the directory normally.  The callers of dx_probe **MUST**
592  * check for this error code, and make sure it never gets reflected
593  * back to userspace.
594  */
595 static struct dx_frame *
596 dx_probe(const struct qstr *d_name, struct inode *dir,
597 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
598 {
599 	unsigned count, indirect;
600 	struct dx_entry *at, *entries, *p, *q, *m;
601 	struct dx_root *root;
602 	struct buffer_head *bh;
603 	struct dx_frame *frame = frame_in;
604 	u32 hash;
605 
606 	frame->bh = NULL;
607 	if (!(bh = ext4_bread(NULL, dir, 0, 0, err))) {
608 		if (*err == 0)
609 			*err = ERR_BAD_DX_DIR;
610 		goto fail;
611 	}
612 	root = (struct dx_root *) bh->b_data;
613 	if (root->info.hash_version != DX_HASH_TEA &&
614 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
615 	    root->info.hash_version != DX_HASH_LEGACY) {
616 		ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
617 			     root->info.hash_version);
618 		brelse(bh);
619 		*err = ERR_BAD_DX_DIR;
620 		goto fail;
621 	}
622 	hinfo->hash_version = root->info.hash_version;
623 	if (hinfo->hash_version <= DX_HASH_TEA)
624 		hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
625 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
626 	if (d_name)
627 		ext4fs_dirhash(d_name->name, d_name->len, hinfo);
628 	hash = hinfo->hash;
629 
630 	if (root->info.unused_flags & 1) {
631 		ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
632 			     root->info.unused_flags);
633 		brelse(bh);
634 		*err = ERR_BAD_DX_DIR;
635 		goto fail;
636 	}
637 
638 	if ((indirect = root->info.indirect_levels) > 1) {
639 		ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
640 			     root->info.indirect_levels);
641 		brelse(bh);
642 		*err = ERR_BAD_DX_DIR;
643 		goto fail;
644 	}
645 
646 	if (!buffer_verified(bh) &&
647 	    !ext4_dx_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data)) {
648 		ext4_warning(dir->i_sb, "Root failed checksum");
649 		brelse(bh);
650 		*err = ERR_BAD_DX_DIR;
651 		goto fail;
652 	}
653 	set_buffer_verified(bh);
654 
655 	entries = (struct dx_entry *) (((char *)&root->info) +
656 				       root->info.info_length);
657 
658 	if (dx_get_limit(entries) != dx_root_limit(dir,
659 						   root->info.info_length)) {
660 		ext4_warning(dir->i_sb, "dx entry: limit != root limit");
661 		brelse(bh);
662 		*err = ERR_BAD_DX_DIR;
663 		goto fail;
664 	}
665 
666 	dxtrace(printk("Look up %x", hash));
667 	while (1)
668 	{
669 		count = dx_get_count(entries);
670 		if (!count || count > dx_get_limit(entries)) {
671 			ext4_warning(dir->i_sb,
672 				     "dx entry: no count or count > limit");
673 			brelse(bh);
674 			*err = ERR_BAD_DX_DIR;
675 			goto fail2;
676 		}
677 
678 		p = entries + 1;
679 		q = entries + count - 1;
680 		while (p <= q)
681 		{
682 			m = p + (q - p)/2;
683 			dxtrace(printk("."));
684 			if (dx_get_hash(m) > hash)
685 				q = m - 1;
686 			else
687 				p = m + 1;
688 		}
689 
690 		if (0) // linear search cross check
691 		{
692 			unsigned n = count - 1;
693 			at = entries;
694 			while (n--)
695 			{
696 				dxtrace(printk(","));
697 				if (dx_get_hash(++at) > hash)
698 				{
699 					at--;
700 					break;
701 				}
702 			}
703 			assert (at == p - 1);
704 		}
705 
706 		at = p - 1;
707 		dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
708 		frame->bh = bh;
709 		frame->entries = entries;
710 		frame->at = at;
711 		if (!indirect--) return frame;
712 		if (!(bh = ext4_bread(NULL, dir, dx_get_block(at), 0, err))) {
713 			if (!(*err))
714 				*err = ERR_BAD_DX_DIR;
715 			goto fail2;
716 		}
717 		at = entries = ((struct dx_node *) bh->b_data)->entries;
718 
719 		if (!buffer_verified(bh) &&
720 		    !ext4_dx_csum_verify(dir,
721 					 (struct ext4_dir_entry *)bh->b_data)) {
722 			ext4_warning(dir->i_sb, "Node failed checksum");
723 			brelse(bh);
724 			*err = ERR_BAD_DX_DIR;
725 			goto fail;
726 		}
727 		set_buffer_verified(bh);
728 
729 		if (dx_get_limit(entries) != dx_node_limit (dir)) {
730 			ext4_warning(dir->i_sb,
731 				     "dx entry: limit != node limit");
732 			brelse(bh);
733 			*err = ERR_BAD_DX_DIR;
734 			goto fail2;
735 		}
736 		frame++;
737 		frame->bh = NULL;
738 	}
739 fail2:
740 	while (frame >= frame_in) {
741 		brelse(frame->bh);
742 		frame--;
743 	}
744 fail:
745 	if (*err == ERR_BAD_DX_DIR)
746 		ext4_warning(dir->i_sb,
747 			     "Corrupt dir inode %lu, running e2fsck is "
748 			     "recommended.", dir->i_ino);
749 	return NULL;
750 }
751 
752 static void dx_release (struct dx_frame *frames)
753 {
754 	if (frames[0].bh == NULL)
755 		return;
756 
757 	if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
758 		brelse(frames[1].bh);
759 	brelse(frames[0].bh);
760 }
761 
762 /*
763  * This function increments the frame pointer to search the next leaf
764  * block, and reads in the necessary intervening nodes if the search
765  * should be necessary.  Whether or not the search is necessary is
766  * controlled by the hash parameter.  If the hash value is even, then
767  * the search is only continued if the next block starts with that
768  * hash value.  This is used if we are searching for a specific file.
769  *
770  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
771  *
772  * This function returns 1 if the caller should continue to search,
773  * or 0 if it should not.  If there is an error reading one of the
774  * index blocks, it will a negative error code.
775  *
776  * If start_hash is non-null, it will be filled in with the starting
777  * hash of the next page.
778  */
779 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
780 				 struct dx_frame *frame,
781 				 struct dx_frame *frames,
782 				 __u32 *start_hash)
783 {
784 	struct dx_frame *p;
785 	struct buffer_head *bh;
786 	int err, num_frames = 0;
787 	__u32 bhash;
788 
789 	p = frame;
790 	/*
791 	 * Find the next leaf page by incrementing the frame pointer.
792 	 * If we run out of entries in the interior node, loop around and
793 	 * increment pointer in the parent node.  When we break out of
794 	 * this loop, num_frames indicates the number of interior
795 	 * nodes need to be read.
796 	 */
797 	while (1) {
798 		if (++(p->at) < p->entries + dx_get_count(p->entries))
799 			break;
800 		if (p == frames)
801 			return 0;
802 		num_frames++;
803 		p--;
804 	}
805 
806 	/*
807 	 * If the hash is 1, then continue only if the next page has a
808 	 * continuation hash of any value.  This is used for readdir
809 	 * handling.  Otherwise, check to see if the hash matches the
810 	 * desired contiuation hash.  If it doesn't, return since
811 	 * there's no point to read in the successive index pages.
812 	 */
813 	bhash = dx_get_hash(p->at);
814 	if (start_hash)
815 		*start_hash = bhash;
816 	if ((hash & 1) == 0) {
817 		if ((bhash & ~1) != hash)
818 			return 0;
819 	}
820 	/*
821 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
822 	 * block so no check is necessary
823 	 */
824 	while (num_frames--) {
825 		if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
826 				      0, &err))) {
827 			if (!err) {
828 				ext4_error(dir->i_sb,
829 					   "Directory hole detected on inode %lu\n",
830 					   dir->i_ino);
831 				return -EIO;
832 			}
833 			return err; /* Failure */
834 		}
835 
836 		if (!buffer_verified(bh) &&
837 		    !ext4_dx_csum_verify(dir,
838 					 (struct ext4_dir_entry *)bh->b_data)) {
839 			ext4_warning(dir->i_sb, "Node failed checksum");
840 			return -EIO;
841 		}
842 		set_buffer_verified(bh);
843 
844 		p++;
845 		brelse(p->bh);
846 		p->bh = bh;
847 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
848 	}
849 	return 1;
850 }
851 
852 
853 /*
854  * This function fills a red-black tree with information from a
855  * directory block.  It returns the number directory entries loaded
856  * into the tree.  If there is an error it is returned in err.
857  */
858 static int htree_dirblock_to_tree(struct file *dir_file,
859 				  struct inode *dir, ext4_lblk_t block,
860 				  struct dx_hash_info *hinfo,
861 				  __u32 start_hash, __u32 start_minor_hash)
862 {
863 	struct buffer_head *bh;
864 	struct ext4_dir_entry_2 *de, *top;
865 	int err = 0, count = 0;
866 
867 	dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
868 							(unsigned long)block));
869 	if (!(bh = ext4_bread(NULL, dir, block, 0, &err))) {
870 		if (!err) {
871 			err = -EIO;
872 			ext4_error(dir->i_sb,
873 				   "Directory hole detected on inode %lu\n",
874 				   dir->i_ino);
875 		}
876 		return err;
877 	}
878 
879 	if (!buffer_verified(bh) &&
880 	    !ext4_dirent_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data))
881 		return -EIO;
882 	set_buffer_verified(bh);
883 
884 	de = (struct ext4_dir_entry_2 *) bh->b_data;
885 	top = (struct ext4_dir_entry_2 *) ((char *) de +
886 					   dir->i_sb->s_blocksize -
887 					   EXT4_DIR_REC_LEN(0));
888 	for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
889 		if (ext4_check_dir_entry(dir, NULL, de, bh,
890 				bh->b_data, bh->b_size,
891 				(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
892 					 + ((char *)de - bh->b_data))) {
893 			/* On error, skip the f_pos to the next block. */
894 			dir_file->f_pos = (dir_file->f_pos |
895 					(dir->i_sb->s_blocksize - 1)) + 1;
896 			brelse(bh);
897 			return count;
898 		}
899 		ext4fs_dirhash(de->name, de->name_len, hinfo);
900 		if ((hinfo->hash < start_hash) ||
901 		    ((hinfo->hash == start_hash) &&
902 		     (hinfo->minor_hash < start_minor_hash)))
903 			continue;
904 		if (de->inode == 0)
905 			continue;
906 		if ((err = ext4_htree_store_dirent(dir_file,
907 				   hinfo->hash, hinfo->minor_hash, de)) != 0) {
908 			brelse(bh);
909 			return err;
910 		}
911 		count++;
912 	}
913 	brelse(bh);
914 	return count;
915 }
916 
917 
918 /*
919  * This function fills a red-black tree with information from a
920  * directory.  We start scanning the directory in hash order, starting
921  * at start_hash and start_minor_hash.
922  *
923  * This function returns the number of entries inserted into the tree,
924  * or a negative error code.
925  */
926 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
927 			 __u32 start_minor_hash, __u32 *next_hash)
928 {
929 	struct dx_hash_info hinfo;
930 	struct ext4_dir_entry_2 *de;
931 	struct dx_frame frames[2], *frame;
932 	struct inode *dir;
933 	ext4_lblk_t block;
934 	int count = 0;
935 	int ret, err;
936 	__u32 hashval;
937 
938 	dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
939 		       start_hash, start_minor_hash));
940 	dir = dir_file->f_path.dentry->d_inode;
941 	if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
942 		hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
943 		if (hinfo.hash_version <= DX_HASH_TEA)
944 			hinfo.hash_version +=
945 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
946 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
947 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
948 					       start_hash, start_minor_hash);
949 		*next_hash = ~0;
950 		return count;
951 	}
952 	hinfo.hash = start_hash;
953 	hinfo.minor_hash = 0;
954 	frame = dx_probe(NULL, dir, &hinfo, frames, &err);
955 	if (!frame)
956 		return err;
957 
958 	/* Add '.' and '..' from the htree header */
959 	if (!start_hash && !start_minor_hash) {
960 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
961 		if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
962 			goto errout;
963 		count++;
964 	}
965 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
966 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
967 		de = ext4_next_entry(de, dir->i_sb->s_blocksize);
968 		if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
969 			goto errout;
970 		count++;
971 	}
972 
973 	while (1) {
974 		block = dx_get_block(frame->at);
975 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
976 					     start_hash, start_minor_hash);
977 		if (ret < 0) {
978 			err = ret;
979 			goto errout;
980 		}
981 		count += ret;
982 		hashval = ~0;
983 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
984 					    frame, frames, &hashval);
985 		*next_hash = hashval;
986 		if (ret < 0) {
987 			err = ret;
988 			goto errout;
989 		}
990 		/*
991 		 * Stop if:  (a) there are no more entries, or
992 		 * (b) we have inserted at least one entry and the
993 		 * next hash value is not a continuation
994 		 */
995 		if ((ret == 0) ||
996 		    (count && ((hashval & 1) == 0)))
997 			break;
998 	}
999 	dx_release(frames);
1000 	dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1001 		       "next hash: %x\n", count, *next_hash));
1002 	return count;
1003 errout:
1004 	dx_release(frames);
1005 	return (err);
1006 }
1007 
1008 static inline int search_dirblock(struct buffer_head *bh,
1009 				  struct inode *dir,
1010 				  const struct qstr *d_name,
1011 				  unsigned int offset,
1012 				  struct ext4_dir_entry_2 **res_dir)
1013 {
1014 	return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1015 			  d_name, offset, res_dir);
1016 }
1017 
1018 /*
1019  * Directory block splitting, compacting
1020  */
1021 
1022 /*
1023  * Create map of hash values, offsets, and sizes, stored at end of block.
1024  * Returns number of entries mapped.
1025  */
1026 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1027 		       struct dx_hash_info *hinfo,
1028 		       struct dx_map_entry *map_tail)
1029 {
1030 	int count = 0;
1031 	char *base = (char *) de;
1032 	struct dx_hash_info h = *hinfo;
1033 
1034 	while ((char *) de < base + blocksize) {
1035 		if (de->name_len && de->inode) {
1036 			ext4fs_dirhash(de->name, de->name_len, &h);
1037 			map_tail--;
1038 			map_tail->hash = h.hash;
1039 			map_tail->offs = ((char *) de - base)>>2;
1040 			map_tail->size = le16_to_cpu(de->rec_len);
1041 			count++;
1042 			cond_resched();
1043 		}
1044 		/* XXX: do we need to check rec_len == 0 case? -Chris */
1045 		de = ext4_next_entry(de, blocksize);
1046 	}
1047 	return count;
1048 }
1049 
1050 /* Sort map by hash value */
1051 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1052 {
1053 	struct dx_map_entry *p, *q, *top = map + count - 1;
1054 	int more;
1055 	/* Combsort until bubble sort doesn't suck */
1056 	while (count > 2) {
1057 		count = count*10/13;
1058 		if (count - 9 < 2) /* 9, 10 -> 11 */
1059 			count = 11;
1060 		for (p = top, q = p - count; q >= map; p--, q--)
1061 			if (p->hash < q->hash)
1062 				swap(*p, *q);
1063 	}
1064 	/* Garden variety bubble sort */
1065 	do {
1066 		more = 0;
1067 		q = top;
1068 		while (q-- > map) {
1069 			if (q[1].hash >= q[0].hash)
1070 				continue;
1071 			swap(*(q+1), *q);
1072 			more = 1;
1073 		}
1074 	} while(more);
1075 }
1076 
1077 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1078 {
1079 	struct dx_entry *entries = frame->entries;
1080 	struct dx_entry *old = frame->at, *new = old + 1;
1081 	int count = dx_get_count(entries);
1082 
1083 	assert(count < dx_get_limit(entries));
1084 	assert(old < entries + count);
1085 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1086 	dx_set_hash(new, hash);
1087 	dx_set_block(new, block);
1088 	dx_set_count(entries, count + 1);
1089 }
1090 
1091 /*
1092  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1093  *
1094  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1095  * `de != NULL' is guaranteed by caller.
1096  */
1097 static inline int ext4_match (int len, const char * const name,
1098 			      struct ext4_dir_entry_2 * de)
1099 {
1100 	if (len != de->name_len)
1101 		return 0;
1102 	if (!de->inode)
1103 		return 0;
1104 	return !memcmp(name, de->name, len);
1105 }
1106 
1107 /*
1108  * Returns 0 if not found, -1 on failure, and 1 on success
1109  */
1110 int search_dir(struct buffer_head *bh,
1111 	       char *search_buf,
1112 	       int buf_size,
1113 	       struct inode *dir,
1114 	       const struct qstr *d_name,
1115 	       unsigned int offset,
1116 	       struct ext4_dir_entry_2 **res_dir)
1117 {
1118 	struct ext4_dir_entry_2 * de;
1119 	char * dlimit;
1120 	int de_len;
1121 	const char *name = d_name->name;
1122 	int namelen = d_name->len;
1123 
1124 	de = (struct ext4_dir_entry_2 *)search_buf;
1125 	dlimit = search_buf + buf_size;
1126 	while ((char *) de < dlimit) {
1127 		/* this code is executed quadratically often */
1128 		/* do minimal checking `by hand' */
1129 
1130 		if ((char *) de + namelen <= dlimit &&
1131 		    ext4_match (namelen, name, de)) {
1132 			/* found a match - just to be sure, do a full check */
1133 			if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1134 						 bh->b_size, offset))
1135 				return -1;
1136 			*res_dir = de;
1137 			return 1;
1138 		}
1139 		/* prevent looping on a bad block */
1140 		de_len = ext4_rec_len_from_disk(de->rec_len,
1141 						dir->i_sb->s_blocksize);
1142 		if (de_len <= 0)
1143 			return -1;
1144 		offset += de_len;
1145 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1146 	}
1147 	return 0;
1148 }
1149 
1150 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1151 			       struct ext4_dir_entry *de)
1152 {
1153 	struct super_block *sb = dir->i_sb;
1154 
1155 	if (!is_dx(dir))
1156 		return 0;
1157 	if (block == 0)
1158 		return 1;
1159 	if (de->inode == 0 &&
1160 	    ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1161 			sb->s_blocksize)
1162 		return 1;
1163 	return 0;
1164 }
1165 
1166 /*
1167  *	ext4_find_entry()
1168  *
1169  * finds an entry in the specified directory with the wanted name. It
1170  * returns the cache buffer in which the entry was found, and the entry
1171  * itself (as a parameter - res_dir). It does NOT read the inode of the
1172  * entry - you'll have to do that yourself if you want to.
1173  *
1174  * The returned buffer_head has ->b_count elevated.  The caller is expected
1175  * to brelse() it when appropriate.
1176  */
1177 static struct buffer_head * ext4_find_entry (struct inode *dir,
1178 					const struct qstr *d_name,
1179 					struct ext4_dir_entry_2 **res_dir,
1180 					int *inlined)
1181 {
1182 	struct super_block *sb;
1183 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1184 	struct buffer_head *bh, *ret = NULL;
1185 	ext4_lblk_t start, block, b;
1186 	const u8 *name = d_name->name;
1187 	int ra_max = 0;		/* Number of bh's in the readahead
1188 				   buffer, bh_use[] */
1189 	int ra_ptr = 0;		/* Current index into readahead
1190 				   buffer */
1191 	int num = 0;
1192 	ext4_lblk_t  nblocks;
1193 	int i, err;
1194 	int namelen;
1195 
1196 	*res_dir = NULL;
1197 	sb = dir->i_sb;
1198 	namelen = d_name->len;
1199 	if (namelen > EXT4_NAME_LEN)
1200 		return NULL;
1201 
1202 	if (ext4_has_inline_data(dir)) {
1203 		int has_inline_data = 1;
1204 		ret = ext4_find_inline_entry(dir, d_name, res_dir,
1205 					     &has_inline_data);
1206 		if (has_inline_data) {
1207 			if (inlined)
1208 				*inlined = 1;
1209 			return ret;
1210 		}
1211 	}
1212 
1213 	if ((namelen <= 2) && (name[0] == '.') &&
1214 	    (name[1] == '.' || name[1] == '\0')) {
1215 		/*
1216 		 * "." or ".." will only be in the first block
1217 		 * NFS may look up ".."; "." should be handled by the VFS
1218 		 */
1219 		block = start = 0;
1220 		nblocks = 1;
1221 		goto restart;
1222 	}
1223 	if (is_dx(dir)) {
1224 		bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
1225 		/*
1226 		 * On success, or if the error was file not found,
1227 		 * return.  Otherwise, fall back to doing a search the
1228 		 * old fashioned way.
1229 		 */
1230 		if (bh || (err != ERR_BAD_DX_DIR))
1231 			return bh;
1232 		dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1233 			       "falling back\n"));
1234 	}
1235 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1236 	start = EXT4_I(dir)->i_dir_start_lookup;
1237 	if (start >= nblocks)
1238 		start = 0;
1239 	block = start;
1240 restart:
1241 	do {
1242 		/*
1243 		 * We deal with the read-ahead logic here.
1244 		 */
1245 		if (ra_ptr >= ra_max) {
1246 			/* Refill the readahead buffer */
1247 			ra_ptr = 0;
1248 			b = block;
1249 			for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1250 				/*
1251 				 * Terminate if we reach the end of the
1252 				 * directory and must wrap, or if our
1253 				 * search has finished at this block.
1254 				 */
1255 				if (b >= nblocks || (num && block == start)) {
1256 					bh_use[ra_max] = NULL;
1257 					break;
1258 				}
1259 				num++;
1260 				bh = ext4_getblk(NULL, dir, b++, 0, &err);
1261 				bh_use[ra_max] = bh;
1262 				if (bh)
1263 					ll_rw_block(READ | REQ_META | REQ_PRIO,
1264 						    1, &bh);
1265 			}
1266 		}
1267 		if ((bh = bh_use[ra_ptr++]) == NULL)
1268 			goto next;
1269 		wait_on_buffer(bh);
1270 		if (!buffer_uptodate(bh)) {
1271 			/* read error, skip block & hope for the best */
1272 			EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1273 					 (unsigned long) block);
1274 			brelse(bh);
1275 			goto next;
1276 		}
1277 		if (!buffer_verified(bh) &&
1278 		    !is_dx_internal_node(dir, block,
1279 					 (struct ext4_dir_entry *)bh->b_data) &&
1280 		    !ext4_dirent_csum_verify(dir,
1281 				(struct ext4_dir_entry *)bh->b_data)) {
1282 			EXT4_ERROR_INODE(dir, "checksumming directory "
1283 					 "block %lu", (unsigned long)block);
1284 			brelse(bh);
1285 			goto next;
1286 		}
1287 		set_buffer_verified(bh);
1288 		i = search_dirblock(bh, dir, d_name,
1289 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1290 		if (i == 1) {
1291 			EXT4_I(dir)->i_dir_start_lookup = block;
1292 			ret = bh;
1293 			goto cleanup_and_exit;
1294 		} else {
1295 			brelse(bh);
1296 			if (i < 0)
1297 				goto cleanup_and_exit;
1298 		}
1299 	next:
1300 		if (++block >= nblocks)
1301 			block = 0;
1302 	} while (block != start);
1303 
1304 	/*
1305 	 * If the directory has grown while we were searching, then
1306 	 * search the last part of the directory before giving up.
1307 	 */
1308 	block = nblocks;
1309 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1310 	if (block < nblocks) {
1311 		start = 0;
1312 		goto restart;
1313 	}
1314 
1315 cleanup_and_exit:
1316 	/* Clean up the read-ahead blocks */
1317 	for (; ra_ptr < ra_max; ra_ptr++)
1318 		brelse(bh_use[ra_ptr]);
1319 	return ret;
1320 }
1321 
1322 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1323 		       struct ext4_dir_entry_2 **res_dir, int *err)
1324 {
1325 	struct super_block * sb = dir->i_sb;
1326 	struct dx_hash_info	hinfo;
1327 	struct dx_frame frames[2], *frame;
1328 	struct buffer_head *bh;
1329 	ext4_lblk_t block;
1330 	int retval;
1331 
1332 	if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1333 		return NULL;
1334 	do {
1335 		block = dx_get_block(frame->at);
1336 		if (!(bh = ext4_bread(NULL, dir, block, 0, err))) {
1337 			if (!(*err)) {
1338 				*err = -EIO;
1339 				ext4_error(dir->i_sb,
1340 					   "Directory hole detected on inode %lu\n",
1341 					   dir->i_ino);
1342 			}
1343 			goto errout;
1344 		}
1345 
1346 		if (!buffer_verified(bh) &&
1347 		    !ext4_dirent_csum_verify(dir,
1348 				(struct ext4_dir_entry *)bh->b_data)) {
1349 			EXT4_ERROR_INODE(dir, "checksumming directory "
1350 					 "block %lu", (unsigned long)block);
1351 			brelse(bh);
1352 			*err = -EIO;
1353 			goto errout;
1354 		}
1355 		set_buffer_verified(bh);
1356 		retval = search_dirblock(bh, dir, d_name,
1357 					 block << EXT4_BLOCK_SIZE_BITS(sb),
1358 					 res_dir);
1359 		if (retval == 1) { 	/* Success! */
1360 			dx_release(frames);
1361 			return bh;
1362 		}
1363 		brelse(bh);
1364 		if (retval == -1) {
1365 			*err = ERR_BAD_DX_DIR;
1366 			goto errout;
1367 		}
1368 
1369 		/* Check to see if we should continue to search */
1370 		retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1371 					       frames, NULL);
1372 		if (retval < 0) {
1373 			ext4_warning(sb,
1374 			     "error reading index page in directory #%lu",
1375 			     dir->i_ino);
1376 			*err = retval;
1377 			goto errout;
1378 		}
1379 	} while (retval == 1);
1380 
1381 	*err = -ENOENT;
1382 errout:
1383 	dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1384 	dx_release (frames);
1385 	return NULL;
1386 }
1387 
1388 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1389 {
1390 	struct inode *inode;
1391 	struct ext4_dir_entry_2 *de;
1392 	struct buffer_head *bh;
1393 
1394 	if (dentry->d_name.len > EXT4_NAME_LEN)
1395 		return ERR_PTR(-ENAMETOOLONG);
1396 
1397 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1398 	inode = NULL;
1399 	if (bh) {
1400 		__u32 ino = le32_to_cpu(de->inode);
1401 		brelse(bh);
1402 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1403 			EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1404 			return ERR_PTR(-EIO);
1405 		}
1406 		if (unlikely(ino == dir->i_ino)) {
1407 			EXT4_ERROR_INODE(dir, "'%.*s' linked to parent dir",
1408 					 dentry->d_name.len,
1409 					 dentry->d_name.name);
1410 			return ERR_PTR(-EIO);
1411 		}
1412 		inode = ext4_iget(dir->i_sb, ino);
1413 		if (inode == ERR_PTR(-ESTALE)) {
1414 			EXT4_ERROR_INODE(dir,
1415 					 "deleted inode referenced: %u",
1416 					 ino);
1417 			return ERR_PTR(-EIO);
1418 		}
1419 	}
1420 	return d_splice_alias(inode, dentry);
1421 }
1422 
1423 
1424 struct dentry *ext4_get_parent(struct dentry *child)
1425 {
1426 	__u32 ino;
1427 	static const struct qstr dotdot = QSTR_INIT("..", 2);
1428 	struct ext4_dir_entry_2 * de;
1429 	struct buffer_head *bh;
1430 
1431 	bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1432 	if (!bh)
1433 		return ERR_PTR(-ENOENT);
1434 	ino = le32_to_cpu(de->inode);
1435 	brelse(bh);
1436 
1437 	if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1438 		EXT4_ERROR_INODE(child->d_inode,
1439 				 "bad parent inode number: %u", ino);
1440 		return ERR_PTR(-EIO);
1441 	}
1442 
1443 	return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1444 }
1445 
1446 #define S_SHIFT 12
1447 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1448 	[S_IFREG >> S_SHIFT]	= EXT4_FT_REG_FILE,
1449 	[S_IFDIR >> S_SHIFT]	= EXT4_FT_DIR,
1450 	[S_IFCHR >> S_SHIFT]	= EXT4_FT_CHRDEV,
1451 	[S_IFBLK >> S_SHIFT]	= EXT4_FT_BLKDEV,
1452 	[S_IFIFO >> S_SHIFT]	= EXT4_FT_FIFO,
1453 	[S_IFSOCK >> S_SHIFT]	= EXT4_FT_SOCK,
1454 	[S_IFLNK >> S_SHIFT]	= EXT4_FT_SYMLINK,
1455 };
1456 
1457 static inline void ext4_set_de_type(struct super_block *sb,
1458 				struct ext4_dir_entry_2 *de,
1459 				umode_t mode) {
1460 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1461 		de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1462 }
1463 
1464 /*
1465  * Move count entries from end of map between two memory locations.
1466  * Returns pointer to last entry moved.
1467  */
1468 static struct ext4_dir_entry_2 *
1469 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1470 		unsigned blocksize)
1471 {
1472 	unsigned rec_len = 0;
1473 
1474 	while (count--) {
1475 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1476 						(from + (map->offs<<2));
1477 		rec_len = EXT4_DIR_REC_LEN(de->name_len);
1478 		memcpy (to, de, rec_len);
1479 		((struct ext4_dir_entry_2 *) to)->rec_len =
1480 				ext4_rec_len_to_disk(rec_len, blocksize);
1481 		de->inode = 0;
1482 		map++;
1483 		to += rec_len;
1484 	}
1485 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1486 }
1487 
1488 /*
1489  * Compact each dir entry in the range to the minimal rec_len.
1490  * Returns pointer to last entry in range.
1491  */
1492 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1493 {
1494 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1495 	unsigned rec_len = 0;
1496 
1497 	prev = to = de;
1498 	while ((char*)de < base + blocksize) {
1499 		next = ext4_next_entry(de, blocksize);
1500 		if (de->inode && de->name_len) {
1501 			rec_len = EXT4_DIR_REC_LEN(de->name_len);
1502 			if (de > to)
1503 				memmove(to, de, rec_len);
1504 			to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1505 			prev = to;
1506 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1507 		}
1508 		de = next;
1509 	}
1510 	return prev;
1511 }
1512 
1513 /*
1514  * Split a full leaf block to make room for a new dir entry.
1515  * Allocate a new block, and move entries so that they are approx. equally full.
1516  * Returns pointer to de in block into which the new entry will be inserted.
1517  */
1518 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1519 			struct buffer_head **bh,struct dx_frame *frame,
1520 			struct dx_hash_info *hinfo, int *error)
1521 {
1522 	unsigned blocksize = dir->i_sb->s_blocksize;
1523 	unsigned count, continued;
1524 	struct buffer_head *bh2;
1525 	ext4_lblk_t newblock;
1526 	u32 hash2;
1527 	struct dx_map_entry *map;
1528 	char *data1 = (*bh)->b_data, *data2;
1529 	unsigned split, move, size;
1530 	struct ext4_dir_entry_2 *de = NULL, *de2;
1531 	struct ext4_dir_entry_tail *t;
1532 	int	csum_size = 0;
1533 	int	err = 0, i;
1534 
1535 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1536 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1537 		csum_size = sizeof(struct ext4_dir_entry_tail);
1538 
1539 	bh2 = ext4_append (handle, dir, &newblock, &err);
1540 	if (!(bh2)) {
1541 		brelse(*bh);
1542 		*bh = NULL;
1543 		goto errout;
1544 	}
1545 
1546 	BUFFER_TRACE(*bh, "get_write_access");
1547 	err = ext4_journal_get_write_access(handle, *bh);
1548 	if (err)
1549 		goto journal_error;
1550 
1551 	BUFFER_TRACE(frame->bh, "get_write_access");
1552 	err = ext4_journal_get_write_access(handle, frame->bh);
1553 	if (err)
1554 		goto journal_error;
1555 
1556 	data2 = bh2->b_data;
1557 
1558 	/* create map in the end of data2 block */
1559 	map = (struct dx_map_entry *) (data2 + blocksize);
1560 	count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1561 			     blocksize, hinfo, map);
1562 	map -= count;
1563 	dx_sort_map(map, count);
1564 	/* Split the existing block in the middle, size-wise */
1565 	size = 0;
1566 	move = 0;
1567 	for (i = count-1; i >= 0; i--) {
1568 		/* is more than half of this entry in 2nd half of the block? */
1569 		if (size + map[i].size/2 > blocksize/2)
1570 			break;
1571 		size += map[i].size;
1572 		move++;
1573 	}
1574 	/* map index at which we will split */
1575 	split = count - move;
1576 	hash2 = map[split].hash;
1577 	continued = hash2 == map[split - 1].hash;
1578 	dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1579 			(unsigned long)dx_get_block(frame->at),
1580 					hash2, split, count-split));
1581 
1582 	/* Fancy dance to stay within two buffers */
1583 	de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1584 	de = dx_pack_dirents(data1, blocksize);
1585 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1586 					   (char *) de,
1587 					   blocksize);
1588 	de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1589 					    (char *) de2,
1590 					    blocksize);
1591 	if (csum_size) {
1592 		t = EXT4_DIRENT_TAIL(data2, blocksize);
1593 		initialize_dirent_tail(t, blocksize);
1594 
1595 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1596 		initialize_dirent_tail(t, blocksize);
1597 	}
1598 
1599 	dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1600 	dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1601 
1602 	/* Which block gets the new entry? */
1603 	if (hinfo->hash >= hash2)
1604 	{
1605 		swap(*bh, bh2);
1606 		de = de2;
1607 	}
1608 	dx_insert_block(frame, hash2 + continued, newblock);
1609 	err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1610 	if (err)
1611 		goto journal_error;
1612 	err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1613 	if (err)
1614 		goto journal_error;
1615 	brelse(bh2);
1616 	dxtrace(dx_show_index("frame", frame->entries));
1617 	return de;
1618 
1619 journal_error:
1620 	brelse(*bh);
1621 	brelse(bh2);
1622 	*bh = NULL;
1623 	ext4_std_error(dir->i_sb, err);
1624 errout:
1625 	*error = err;
1626 	return NULL;
1627 }
1628 
1629 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1630 		      struct buffer_head *bh,
1631 		      void *buf, int buf_size,
1632 		      const char *name, int namelen,
1633 		      struct ext4_dir_entry_2 **dest_de)
1634 {
1635 	struct ext4_dir_entry_2 *de;
1636 	unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1637 	int nlen, rlen;
1638 	unsigned int offset = 0;
1639 	char *top;
1640 
1641 	de = (struct ext4_dir_entry_2 *)buf;
1642 	top = buf + buf_size - reclen;
1643 	while ((char *) de <= top) {
1644 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1645 					 buf, buf_size, offset))
1646 			return -EIO;
1647 		if (ext4_match(namelen, name, de))
1648 			return -EEXIST;
1649 		nlen = EXT4_DIR_REC_LEN(de->name_len);
1650 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1651 		if ((de->inode ? rlen - nlen : rlen) >= reclen)
1652 			break;
1653 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1654 		offset += rlen;
1655 	}
1656 	if ((char *) de > top)
1657 		return -ENOSPC;
1658 
1659 	*dest_de = de;
1660 	return 0;
1661 }
1662 
1663 void ext4_insert_dentry(struct inode *inode,
1664 			struct ext4_dir_entry_2 *de,
1665 			int buf_size,
1666 			const char *name, int namelen)
1667 {
1668 
1669 	int nlen, rlen;
1670 
1671 	nlen = EXT4_DIR_REC_LEN(de->name_len);
1672 	rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1673 	if (de->inode) {
1674 		struct ext4_dir_entry_2 *de1 =
1675 				(struct ext4_dir_entry_2 *)((char *)de + nlen);
1676 		de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1677 		de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1678 		de = de1;
1679 	}
1680 	de->file_type = EXT4_FT_UNKNOWN;
1681 	de->inode = cpu_to_le32(inode->i_ino);
1682 	ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1683 	de->name_len = namelen;
1684 	memcpy(de->name, name, namelen);
1685 }
1686 /*
1687  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1688  * it points to a directory entry which is guaranteed to be large
1689  * enough for new directory entry.  If de is NULL, then
1690  * add_dirent_to_buf will attempt search the directory block for
1691  * space.  It will return -ENOSPC if no space is available, and -EIO
1692  * and -EEXIST if directory entry already exists.
1693  */
1694 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1695 			     struct inode *inode, struct ext4_dir_entry_2 *de,
1696 			     struct buffer_head *bh)
1697 {
1698 	struct inode	*dir = dentry->d_parent->d_inode;
1699 	const char	*name = dentry->d_name.name;
1700 	int		namelen = dentry->d_name.len;
1701 	unsigned int	blocksize = dir->i_sb->s_blocksize;
1702 	unsigned short	reclen;
1703 	int		csum_size = 0;
1704 	int		err;
1705 
1706 	if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1707 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1708 		csum_size = sizeof(struct ext4_dir_entry_tail);
1709 
1710 	reclen = EXT4_DIR_REC_LEN(namelen);
1711 	if (!de) {
1712 		err = ext4_find_dest_de(dir, inode,
1713 					bh, bh->b_data, blocksize - csum_size,
1714 					name, namelen, &de);
1715 		if (err)
1716 			return err;
1717 	}
1718 	BUFFER_TRACE(bh, "get_write_access");
1719 	err = ext4_journal_get_write_access(handle, bh);
1720 	if (err) {
1721 		ext4_std_error(dir->i_sb, err);
1722 		return err;
1723 	}
1724 
1725 	/* By now the buffer is marked for journaling */
1726 	ext4_insert_dentry(inode, de, blocksize, name, namelen);
1727 
1728 	/*
1729 	 * XXX shouldn't update any times until successful
1730 	 * completion of syscall, but too many callers depend
1731 	 * on this.
1732 	 *
1733 	 * XXX similarly, too many callers depend on
1734 	 * ext4_new_inode() setting the times, but error
1735 	 * recovery deletes the inode, so the worst that can
1736 	 * happen is that the times are slightly out of date
1737 	 * and/or different from the directory change time.
1738 	 */
1739 	dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1740 	ext4_update_dx_flag(dir);
1741 	dir->i_version++;
1742 	ext4_mark_inode_dirty(handle, dir);
1743 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1744 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1745 	if (err)
1746 		ext4_std_error(dir->i_sb, err);
1747 	return 0;
1748 }
1749 
1750 /*
1751  * This converts a one block unindexed directory to a 3 block indexed
1752  * directory, and adds the dentry to the indexed directory.
1753  */
1754 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1755 			    struct inode *inode, struct buffer_head *bh)
1756 {
1757 	struct inode	*dir = dentry->d_parent->d_inode;
1758 	const char	*name = dentry->d_name.name;
1759 	int		namelen = dentry->d_name.len;
1760 	struct buffer_head *bh2;
1761 	struct dx_root	*root;
1762 	struct dx_frame	frames[2], *frame;
1763 	struct dx_entry *entries;
1764 	struct ext4_dir_entry_2	*de, *de2;
1765 	struct ext4_dir_entry_tail *t;
1766 	char		*data1, *top;
1767 	unsigned	len;
1768 	int		retval;
1769 	unsigned	blocksize;
1770 	struct dx_hash_info hinfo;
1771 	ext4_lblk_t  block;
1772 	struct fake_dirent *fde;
1773 	int		csum_size = 0;
1774 
1775 	if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1776 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1777 		csum_size = sizeof(struct ext4_dir_entry_tail);
1778 
1779 	blocksize =  dir->i_sb->s_blocksize;
1780 	dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1781 	retval = ext4_journal_get_write_access(handle, bh);
1782 	if (retval) {
1783 		ext4_std_error(dir->i_sb, retval);
1784 		brelse(bh);
1785 		return retval;
1786 	}
1787 	root = (struct dx_root *) bh->b_data;
1788 
1789 	/* The 0th block becomes the root, move the dirents out */
1790 	fde = &root->dotdot;
1791 	de = (struct ext4_dir_entry_2 *)((char *)fde +
1792 		ext4_rec_len_from_disk(fde->rec_len, blocksize));
1793 	if ((char *) de >= (((char *) root) + blocksize)) {
1794 		EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1795 		brelse(bh);
1796 		return -EIO;
1797 	}
1798 	len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1799 
1800 	/* Allocate new block for the 0th block's dirents */
1801 	bh2 = ext4_append(handle, dir, &block, &retval);
1802 	if (!(bh2)) {
1803 		brelse(bh);
1804 		return retval;
1805 	}
1806 	ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1807 	data1 = bh2->b_data;
1808 
1809 	memcpy (data1, de, len);
1810 	de = (struct ext4_dir_entry_2 *) data1;
1811 	top = data1 + len;
1812 	while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1813 		de = de2;
1814 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1815 					   (char *) de,
1816 					   blocksize);
1817 
1818 	if (csum_size) {
1819 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1820 		initialize_dirent_tail(t, blocksize);
1821 	}
1822 
1823 	/* Initialize the root; the dot dirents already exist */
1824 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1825 	de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1826 					   blocksize);
1827 	memset (&root->info, 0, sizeof(root->info));
1828 	root->info.info_length = sizeof(root->info);
1829 	root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1830 	entries = root->entries;
1831 	dx_set_block(entries, 1);
1832 	dx_set_count(entries, 1);
1833 	dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1834 
1835 	/* Initialize as for dx_probe */
1836 	hinfo.hash_version = root->info.hash_version;
1837 	if (hinfo.hash_version <= DX_HASH_TEA)
1838 		hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1839 	hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1840 	ext4fs_dirhash(name, namelen, &hinfo);
1841 	frame = frames;
1842 	frame->entries = entries;
1843 	frame->at = entries;
1844 	frame->bh = bh;
1845 	bh = bh2;
1846 
1847 	ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1848 	ext4_handle_dirty_dirent_node(handle, dir, bh);
1849 
1850 	de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1851 	if (!de) {
1852 		/*
1853 		 * Even if the block split failed, we have to properly write
1854 		 * out all the changes we did so far. Otherwise we can end up
1855 		 * with corrupted filesystem.
1856 		 */
1857 		ext4_mark_inode_dirty(handle, dir);
1858 		dx_release(frames);
1859 		return retval;
1860 	}
1861 	dx_release(frames);
1862 
1863 	retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1864 	brelse(bh);
1865 	return retval;
1866 }
1867 
1868 /*
1869  *	ext4_add_entry()
1870  *
1871  * adds a file entry to the specified directory, using the same
1872  * semantics as ext4_find_entry(). It returns NULL if it failed.
1873  *
1874  * NOTE!! The inode part of 'de' is left at 0 - which means you
1875  * may not sleep between calling this and putting something into
1876  * the entry, as someone else might have used it while you slept.
1877  */
1878 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1879 			  struct inode *inode)
1880 {
1881 	struct inode *dir = dentry->d_parent->d_inode;
1882 	struct buffer_head *bh;
1883 	struct ext4_dir_entry_2 *de;
1884 	struct ext4_dir_entry_tail *t;
1885 	struct super_block *sb;
1886 	int	retval;
1887 	int	dx_fallback=0;
1888 	unsigned blocksize;
1889 	ext4_lblk_t block, blocks;
1890 	int	csum_size = 0;
1891 
1892 	if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1893 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1894 		csum_size = sizeof(struct ext4_dir_entry_tail);
1895 
1896 	sb = dir->i_sb;
1897 	blocksize = sb->s_blocksize;
1898 	if (!dentry->d_name.len)
1899 		return -EINVAL;
1900 
1901 	if (ext4_has_inline_data(dir)) {
1902 		retval = ext4_try_add_inline_entry(handle, dentry, inode);
1903 		if (retval < 0)
1904 			return retval;
1905 		if (retval == 1) {
1906 			retval = 0;
1907 			return retval;
1908 		}
1909 	}
1910 
1911 	if (is_dx(dir)) {
1912 		retval = ext4_dx_add_entry(handle, dentry, inode);
1913 		if (!retval || (retval != ERR_BAD_DX_DIR))
1914 			return retval;
1915 		ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1916 		dx_fallback++;
1917 		ext4_mark_inode_dirty(handle, dir);
1918 	}
1919 	blocks = dir->i_size >> sb->s_blocksize_bits;
1920 	for (block = 0; block < blocks; block++) {
1921 		if (!(bh = ext4_bread(handle, dir, block, 0, &retval))) {
1922 			if (!retval) {
1923 				retval = -EIO;
1924 				ext4_error(inode->i_sb,
1925 					   "Directory hole detected on inode %lu\n",
1926 					   inode->i_ino);
1927 			}
1928 			return retval;
1929 		}
1930 		if (!buffer_verified(bh) &&
1931 		    !ext4_dirent_csum_verify(dir,
1932 				(struct ext4_dir_entry *)bh->b_data))
1933 			return -EIO;
1934 		set_buffer_verified(bh);
1935 		retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1936 		if (retval != -ENOSPC) {
1937 			brelse(bh);
1938 			return retval;
1939 		}
1940 
1941 		if (blocks == 1 && !dx_fallback &&
1942 		    EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1943 			return make_indexed_dir(handle, dentry, inode, bh);
1944 		brelse(bh);
1945 	}
1946 	bh = ext4_append(handle, dir, &block, &retval);
1947 	if (!bh)
1948 		return retval;
1949 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1950 	de->inode = 0;
1951 	de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1952 
1953 	if (csum_size) {
1954 		t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1955 		initialize_dirent_tail(t, blocksize);
1956 	}
1957 
1958 	retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1959 	brelse(bh);
1960 	if (retval == 0)
1961 		ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1962 	return retval;
1963 }
1964 
1965 /*
1966  * Returns 0 for success, or a negative error value
1967  */
1968 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1969 			     struct inode *inode)
1970 {
1971 	struct dx_frame frames[2], *frame;
1972 	struct dx_entry *entries, *at;
1973 	struct dx_hash_info hinfo;
1974 	struct buffer_head *bh;
1975 	struct inode *dir = dentry->d_parent->d_inode;
1976 	struct super_block *sb = dir->i_sb;
1977 	struct ext4_dir_entry_2 *de;
1978 	int err;
1979 
1980 	frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1981 	if (!frame)
1982 		return err;
1983 	entries = frame->entries;
1984 	at = frame->at;
1985 
1986 	if (!(bh = ext4_bread(handle, dir, dx_get_block(frame->at), 0, &err))) {
1987 		if (!err) {
1988 			err = -EIO;
1989 			ext4_error(dir->i_sb,
1990 				   "Directory hole detected on inode %lu\n",
1991 				   dir->i_ino);
1992 		}
1993 		goto cleanup;
1994 	}
1995 
1996 	if (!buffer_verified(bh) &&
1997 	    !ext4_dirent_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data))
1998 		goto journal_error;
1999 	set_buffer_verified(bh);
2000 
2001 	BUFFER_TRACE(bh, "get_write_access");
2002 	err = ext4_journal_get_write_access(handle, bh);
2003 	if (err)
2004 		goto journal_error;
2005 
2006 	err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
2007 	if (err != -ENOSPC)
2008 		goto cleanup;
2009 
2010 	/* Block full, should compress but for now just split */
2011 	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2012 		       dx_get_count(entries), dx_get_limit(entries)));
2013 	/* Need to split index? */
2014 	if (dx_get_count(entries) == dx_get_limit(entries)) {
2015 		ext4_lblk_t newblock;
2016 		unsigned icount = dx_get_count(entries);
2017 		int levels = frame - frames;
2018 		struct dx_entry *entries2;
2019 		struct dx_node *node2;
2020 		struct buffer_head *bh2;
2021 
2022 		if (levels && (dx_get_count(frames->entries) ==
2023 			       dx_get_limit(frames->entries))) {
2024 			ext4_warning(sb, "Directory index full!");
2025 			err = -ENOSPC;
2026 			goto cleanup;
2027 		}
2028 		bh2 = ext4_append (handle, dir, &newblock, &err);
2029 		if (!(bh2))
2030 			goto cleanup;
2031 		node2 = (struct dx_node *)(bh2->b_data);
2032 		entries2 = node2->entries;
2033 		memset(&node2->fake, 0, sizeof(struct fake_dirent));
2034 		node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2035 							   sb->s_blocksize);
2036 		BUFFER_TRACE(frame->bh, "get_write_access");
2037 		err = ext4_journal_get_write_access(handle, frame->bh);
2038 		if (err)
2039 			goto journal_error;
2040 		if (levels) {
2041 			unsigned icount1 = icount/2, icount2 = icount - icount1;
2042 			unsigned hash2 = dx_get_hash(entries + icount1);
2043 			dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2044 				       icount1, icount2));
2045 
2046 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2047 			err = ext4_journal_get_write_access(handle,
2048 							     frames[0].bh);
2049 			if (err)
2050 				goto journal_error;
2051 
2052 			memcpy((char *) entries2, (char *) (entries + icount1),
2053 			       icount2 * sizeof(struct dx_entry));
2054 			dx_set_count(entries, icount1);
2055 			dx_set_count(entries2, icount2);
2056 			dx_set_limit(entries2, dx_node_limit(dir));
2057 
2058 			/* Which index block gets the new entry? */
2059 			if (at - entries >= icount1) {
2060 				frame->at = at = at - entries - icount1 + entries2;
2061 				frame->entries = entries = entries2;
2062 				swap(frame->bh, bh2);
2063 			}
2064 			dx_insert_block(frames + 0, hash2, newblock);
2065 			dxtrace(dx_show_index("node", frames[1].entries));
2066 			dxtrace(dx_show_index("node",
2067 			       ((struct dx_node *) bh2->b_data)->entries));
2068 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2069 			if (err)
2070 				goto journal_error;
2071 			brelse (bh2);
2072 		} else {
2073 			dxtrace(printk(KERN_DEBUG
2074 				       "Creating second level index...\n"));
2075 			memcpy((char *) entries2, (char *) entries,
2076 			       icount * sizeof(struct dx_entry));
2077 			dx_set_limit(entries2, dx_node_limit(dir));
2078 
2079 			/* Set up root */
2080 			dx_set_count(entries, 1);
2081 			dx_set_block(entries + 0, newblock);
2082 			((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2083 
2084 			/* Add new access path frame */
2085 			frame = frames + 1;
2086 			frame->at = at = at - entries + entries2;
2087 			frame->entries = entries = entries2;
2088 			frame->bh = bh2;
2089 			err = ext4_journal_get_write_access(handle,
2090 							     frame->bh);
2091 			if (err)
2092 				goto journal_error;
2093 		}
2094 		err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2095 		if (err) {
2096 			ext4_std_error(inode->i_sb, err);
2097 			goto cleanup;
2098 		}
2099 	}
2100 	de = do_split(handle, dir, &bh, frame, &hinfo, &err);
2101 	if (!de)
2102 		goto cleanup;
2103 	err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2104 	goto cleanup;
2105 
2106 journal_error:
2107 	ext4_std_error(dir->i_sb, err);
2108 cleanup:
2109 	if (bh)
2110 		brelse(bh);
2111 	dx_release(frames);
2112 	return err;
2113 }
2114 
2115 /*
2116  * ext4_generic_delete_entry deletes a directory entry by merging it
2117  * with the previous entry
2118  */
2119 int ext4_generic_delete_entry(handle_t *handle,
2120 			      struct inode *dir,
2121 			      struct ext4_dir_entry_2 *de_del,
2122 			      struct buffer_head *bh,
2123 			      void *entry_buf,
2124 			      int buf_size,
2125 			      int csum_size)
2126 {
2127 	struct ext4_dir_entry_2 *de, *pde;
2128 	unsigned int blocksize = dir->i_sb->s_blocksize;
2129 	int i;
2130 
2131 	i = 0;
2132 	pde = NULL;
2133 	de = (struct ext4_dir_entry_2 *)entry_buf;
2134 	while (i < buf_size - csum_size) {
2135 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2136 					 bh->b_data, bh->b_size, i))
2137 			return -EIO;
2138 		if (de == de_del)  {
2139 			if (pde)
2140 				pde->rec_len = ext4_rec_len_to_disk(
2141 					ext4_rec_len_from_disk(pde->rec_len,
2142 							       blocksize) +
2143 					ext4_rec_len_from_disk(de->rec_len,
2144 							       blocksize),
2145 					blocksize);
2146 			else
2147 				de->inode = 0;
2148 			dir->i_version++;
2149 			return 0;
2150 		}
2151 		i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2152 		pde = de;
2153 		de = ext4_next_entry(de, blocksize);
2154 	}
2155 	return -ENOENT;
2156 }
2157 
2158 static int ext4_delete_entry(handle_t *handle,
2159 			     struct inode *dir,
2160 			     struct ext4_dir_entry_2 *de_del,
2161 			     struct buffer_head *bh)
2162 {
2163 	int err, csum_size = 0;
2164 
2165 	if (ext4_has_inline_data(dir)) {
2166 		int has_inline_data = 1;
2167 		err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2168 					       &has_inline_data);
2169 		if (has_inline_data)
2170 			return err;
2171 	}
2172 
2173 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2174 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2175 		csum_size = sizeof(struct ext4_dir_entry_tail);
2176 
2177 	BUFFER_TRACE(bh, "get_write_access");
2178 	err = ext4_journal_get_write_access(handle, bh);
2179 	if (unlikely(err))
2180 		goto out;
2181 
2182 	err = ext4_generic_delete_entry(handle, dir, de_del,
2183 					bh, bh->b_data,
2184 					dir->i_sb->s_blocksize, csum_size);
2185 	if (err)
2186 		goto out;
2187 
2188 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2189 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2190 	if (unlikely(err))
2191 		goto out;
2192 
2193 	return 0;
2194 out:
2195 	if (err != -ENOENT)
2196 		ext4_std_error(dir->i_sb, err);
2197 	return err;
2198 }
2199 
2200 /*
2201  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2202  * since this indicates that nlinks count was previously 1.
2203  */
2204 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2205 {
2206 	inc_nlink(inode);
2207 	if (is_dx(inode) && inode->i_nlink > 1) {
2208 		/* limit is 16-bit i_links_count */
2209 		if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2210 			set_nlink(inode, 1);
2211 			EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2212 					      EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2213 		}
2214 	}
2215 }
2216 
2217 /*
2218  * If a directory had nlink == 1, then we should let it be 1. This indicates
2219  * directory has >EXT4_LINK_MAX subdirs.
2220  */
2221 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2222 {
2223 	if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2224 		drop_nlink(inode);
2225 }
2226 
2227 
2228 static int ext4_add_nondir(handle_t *handle,
2229 		struct dentry *dentry, struct inode *inode)
2230 {
2231 	int err = ext4_add_entry(handle, dentry, inode);
2232 	if (!err) {
2233 		ext4_mark_inode_dirty(handle, inode);
2234 		unlock_new_inode(inode);
2235 		d_instantiate(dentry, inode);
2236 		return 0;
2237 	}
2238 	drop_nlink(inode);
2239 	unlock_new_inode(inode);
2240 	iput(inode);
2241 	return err;
2242 }
2243 
2244 /*
2245  * By the time this is called, we already have created
2246  * the directory cache entry for the new file, but it
2247  * is so far negative - it has no inode.
2248  *
2249  * If the create succeeds, we fill in the inode information
2250  * with d_instantiate().
2251  */
2252 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2253 		       bool excl)
2254 {
2255 	handle_t *handle;
2256 	struct inode *inode;
2257 	int err, retries = 0;
2258 
2259 	dquot_initialize(dir);
2260 
2261 retry:
2262 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2263 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2264 					EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2265 	if (IS_ERR(handle))
2266 		return PTR_ERR(handle);
2267 
2268 	if (IS_DIRSYNC(dir))
2269 		ext4_handle_sync(handle);
2270 
2271 	inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
2272 	err = PTR_ERR(inode);
2273 	if (!IS_ERR(inode)) {
2274 		inode->i_op = &ext4_file_inode_operations;
2275 		inode->i_fop = &ext4_file_operations;
2276 		ext4_set_aops(inode);
2277 		err = ext4_add_nondir(handle, dentry, inode);
2278 	}
2279 	ext4_journal_stop(handle);
2280 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2281 		goto retry;
2282 	return err;
2283 }
2284 
2285 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2286 		      umode_t mode, dev_t rdev)
2287 {
2288 	handle_t *handle;
2289 	struct inode *inode;
2290 	int err, retries = 0;
2291 
2292 	if (!new_valid_dev(rdev))
2293 		return -EINVAL;
2294 
2295 	dquot_initialize(dir);
2296 
2297 retry:
2298 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2299 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2300 					EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2301 	if (IS_ERR(handle))
2302 		return PTR_ERR(handle);
2303 
2304 	if (IS_DIRSYNC(dir))
2305 		ext4_handle_sync(handle);
2306 
2307 	inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
2308 	err = PTR_ERR(inode);
2309 	if (!IS_ERR(inode)) {
2310 		init_special_inode(inode, inode->i_mode, rdev);
2311 		inode->i_op = &ext4_special_inode_operations;
2312 		err = ext4_add_nondir(handle, dentry, inode);
2313 	}
2314 	ext4_journal_stop(handle);
2315 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2316 		goto retry;
2317 	return err;
2318 }
2319 
2320 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2321 			  struct ext4_dir_entry_2 *de,
2322 			  int blocksize, int csum_size,
2323 			  unsigned int parent_ino, int dotdot_real_len)
2324 {
2325 	de->inode = cpu_to_le32(inode->i_ino);
2326 	de->name_len = 1;
2327 	de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2328 					   blocksize);
2329 	strcpy(de->name, ".");
2330 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2331 
2332 	de = ext4_next_entry(de, blocksize);
2333 	de->inode = cpu_to_le32(parent_ino);
2334 	de->name_len = 2;
2335 	if (!dotdot_real_len)
2336 		de->rec_len = ext4_rec_len_to_disk(blocksize -
2337 					(csum_size + EXT4_DIR_REC_LEN(1)),
2338 					blocksize);
2339 	else
2340 		de->rec_len = ext4_rec_len_to_disk(
2341 				EXT4_DIR_REC_LEN(de->name_len), blocksize);
2342 	strcpy(de->name, "..");
2343 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2344 
2345 	return ext4_next_entry(de, blocksize);
2346 }
2347 
2348 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2349 			     struct inode *inode)
2350 {
2351 	struct buffer_head *dir_block = NULL;
2352 	struct ext4_dir_entry_2 *de;
2353 	struct ext4_dir_entry_tail *t;
2354 	unsigned int blocksize = dir->i_sb->s_blocksize;
2355 	int csum_size = 0;
2356 	int err;
2357 
2358 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2359 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2360 		csum_size = sizeof(struct ext4_dir_entry_tail);
2361 
2362 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2363 		err = ext4_try_create_inline_dir(handle, dir, inode);
2364 		if (err < 0 && err != -ENOSPC)
2365 			goto out;
2366 		if (!err)
2367 			goto out;
2368 	}
2369 
2370 	inode->i_size = EXT4_I(inode)->i_disksize = blocksize;
2371 	dir_block = ext4_bread(handle, inode, 0, 1, &err);
2372 	if (!(dir_block = ext4_bread(handle, inode, 0, 1, &err))) {
2373 		if (!err) {
2374 			err = -EIO;
2375 			ext4_error(inode->i_sb,
2376 				   "Directory hole detected on inode %lu\n",
2377 				   inode->i_ino);
2378 		}
2379 		goto out;
2380 	}
2381 	BUFFER_TRACE(dir_block, "get_write_access");
2382 	err = ext4_journal_get_write_access(handle, dir_block);
2383 	if (err)
2384 		goto out;
2385 	de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2386 	ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2387 	set_nlink(inode, 2);
2388 	if (csum_size) {
2389 		t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2390 		initialize_dirent_tail(t, blocksize);
2391 	}
2392 
2393 	BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2394 	err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2395 	if (err)
2396 		goto out;
2397 	set_buffer_verified(dir_block);
2398 out:
2399 	brelse(dir_block);
2400 	return err;
2401 }
2402 
2403 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2404 {
2405 	handle_t *handle;
2406 	struct inode *inode;
2407 	int err, retries = 0;
2408 
2409 	if (EXT4_DIR_LINK_MAX(dir))
2410 		return -EMLINK;
2411 
2412 	dquot_initialize(dir);
2413 
2414 retry:
2415 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2416 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2417 					EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2418 	if (IS_ERR(handle))
2419 		return PTR_ERR(handle);
2420 
2421 	if (IS_DIRSYNC(dir))
2422 		ext4_handle_sync(handle);
2423 
2424 	inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
2425 			       &dentry->d_name, 0, NULL);
2426 	err = PTR_ERR(inode);
2427 	if (IS_ERR(inode))
2428 		goto out_stop;
2429 
2430 	inode->i_op = &ext4_dir_inode_operations;
2431 	inode->i_fop = &ext4_dir_operations;
2432 	err = ext4_init_new_dir(handle, dir, inode);
2433 	if (err)
2434 		goto out_clear_inode;
2435 	err = ext4_mark_inode_dirty(handle, inode);
2436 	if (!err)
2437 		err = ext4_add_entry(handle, dentry, inode);
2438 	if (err) {
2439 out_clear_inode:
2440 		clear_nlink(inode);
2441 		unlock_new_inode(inode);
2442 		ext4_mark_inode_dirty(handle, inode);
2443 		iput(inode);
2444 		goto out_stop;
2445 	}
2446 	ext4_inc_count(handle, dir);
2447 	ext4_update_dx_flag(dir);
2448 	err = ext4_mark_inode_dirty(handle, dir);
2449 	if (err)
2450 		goto out_clear_inode;
2451 	unlock_new_inode(inode);
2452 	d_instantiate(dentry, inode);
2453 out_stop:
2454 	ext4_journal_stop(handle);
2455 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2456 		goto retry;
2457 	return err;
2458 }
2459 
2460 /*
2461  * routine to check that the specified directory is empty (for rmdir)
2462  */
2463 static int empty_dir(struct inode *inode)
2464 {
2465 	unsigned int offset;
2466 	struct buffer_head *bh;
2467 	struct ext4_dir_entry_2 *de, *de1;
2468 	struct super_block *sb;
2469 	int err = 0;
2470 
2471 	if (ext4_has_inline_data(inode)) {
2472 		int has_inline_data = 1;
2473 
2474 		err = empty_inline_dir(inode, &has_inline_data);
2475 		if (has_inline_data)
2476 			return err;
2477 	}
2478 
2479 	sb = inode->i_sb;
2480 	if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
2481 	    !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
2482 		if (err)
2483 			EXT4_ERROR_INODE(inode,
2484 				"error %d reading directory lblock 0", err);
2485 		else
2486 			ext4_warning(inode->i_sb,
2487 				     "bad directory (dir #%lu) - no data block",
2488 				     inode->i_ino);
2489 		return 1;
2490 	}
2491 	if (!buffer_verified(bh) &&
2492 	    !ext4_dirent_csum_verify(inode,
2493 			(struct ext4_dir_entry *)bh->b_data)) {
2494 		EXT4_ERROR_INODE(inode, "checksum error reading directory "
2495 				 "lblock 0");
2496 		return -EIO;
2497 	}
2498 	set_buffer_verified(bh);
2499 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2500 	de1 = ext4_next_entry(de, sb->s_blocksize);
2501 	if (le32_to_cpu(de->inode) != inode->i_ino ||
2502 			!le32_to_cpu(de1->inode) ||
2503 			strcmp(".", de->name) ||
2504 			strcmp("..", de1->name)) {
2505 		ext4_warning(inode->i_sb,
2506 			     "bad directory (dir #%lu) - no `.' or `..'",
2507 			     inode->i_ino);
2508 		brelse(bh);
2509 		return 1;
2510 	}
2511 	offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2512 		 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2513 	de = ext4_next_entry(de1, sb->s_blocksize);
2514 	while (offset < inode->i_size) {
2515 		if (!bh ||
2516 		    (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2517 			unsigned int lblock;
2518 			err = 0;
2519 			brelse(bh);
2520 			lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2521 			bh = ext4_bread(NULL, inode, lblock, 0, &err);
2522 			if (!bh) {
2523 				if (err)
2524 					EXT4_ERROR_INODE(inode,
2525 						"error %d reading directory "
2526 						"lblock %u", err, lblock);
2527 				else
2528 					ext4_warning(inode->i_sb,
2529 						"bad directory (dir #%lu) - no data block",
2530 						inode->i_ino);
2531 
2532 				offset += sb->s_blocksize;
2533 				continue;
2534 			}
2535 			if (!buffer_verified(bh) &&
2536 			    !ext4_dirent_csum_verify(inode,
2537 					(struct ext4_dir_entry *)bh->b_data)) {
2538 				EXT4_ERROR_INODE(inode, "checksum error "
2539 						 "reading directory lblock 0");
2540 				return -EIO;
2541 			}
2542 			set_buffer_verified(bh);
2543 			de = (struct ext4_dir_entry_2 *) bh->b_data;
2544 		}
2545 		if (ext4_check_dir_entry(inode, NULL, de, bh,
2546 					 bh->b_data, bh->b_size, offset)) {
2547 			de = (struct ext4_dir_entry_2 *)(bh->b_data +
2548 							 sb->s_blocksize);
2549 			offset = (offset | (sb->s_blocksize - 1)) + 1;
2550 			continue;
2551 		}
2552 		if (le32_to_cpu(de->inode)) {
2553 			brelse(bh);
2554 			return 0;
2555 		}
2556 		offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2557 		de = ext4_next_entry(de, sb->s_blocksize);
2558 	}
2559 	brelse(bh);
2560 	return 1;
2561 }
2562 
2563 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
2564  * such inodes, starting at the superblock, in case we crash before the
2565  * file is closed/deleted, or in case the inode truncate spans multiple
2566  * transactions and the last transaction is not recovered after a crash.
2567  *
2568  * At filesystem recovery time, we walk this list deleting unlinked
2569  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2570  */
2571 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2572 {
2573 	struct super_block *sb = inode->i_sb;
2574 	struct ext4_iloc iloc;
2575 	int err = 0, rc;
2576 
2577 	if (!EXT4_SB(sb)->s_journal)
2578 		return 0;
2579 
2580 	mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
2581 	if (!list_empty(&EXT4_I(inode)->i_orphan))
2582 		goto out_unlock;
2583 
2584 	/*
2585 	 * Orphan handling is only valid for files with data blocks
2586 	 * being truncated, or files being unlinked. Note that we either
2587 	 * hold i_mutex, or the inode can not be referenced from outside,
2588 	 * so i_nlink should not be bumped due to race
2589 	 */
2590 	J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2591 		  S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2592 
2593 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2594 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2595 	if (err)
2596 		goto out_unlock;
2597 
2598 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2599 	if (err)
2600 		goto out_unlock;
2601 	/*
2602 	 * Due to previous errors inode may be already a part of on-disk
2603 	 * orphan list. If so skip on-disk list modification.
2604 	 */
2605 	if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2606 		(le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2607 			goto mem_insert;
2608 
2609 	/* Insert this inode at the head of the on-disk orphan list... */
2610 	NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2611 	EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2612 	err = ext4_handle_dirty_super(handle, sb);
2613 	rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2614 	if (!err)
2615 		err = rc;
2616 
2617 	/* Only add to the head of the in-memory list if all the
2618 	 * previous operations succeeded.  If the orphan_add is going to
2619 	 * fail (possibly taking the journal offline), we can't risk
2620 	 * leaving the inode on the orphan list: stray orphan-list
2621 	 * entries can cause panics at unmount time.
2622 	 *
2623 	 * This is safe: on error we're going to ignore the orphan list
2624 	 * anyway on the next recovery. */
2625 mem_insert:
2626 	if (!err)
2627 		list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2628 
2629 	jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2630 	jbd_debug(4, "orphan inode %lu will point to %d\n",
2631 			inode->i_ino, NEXT_ORPHAN(inode));
2632 out_unlock:
2633 	mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2634 	ext4_std_error(inode->i_sb, err);
2635 	return err;
2636 }
2637 
2638 /*
2639  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2640  * of such inodes stored on disk, because it is finally being cleaned up.
2641  */
2642 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2643 {
2644 	struct list_head *prev;
2645 	struct ext4_inode_info *ei = EXT4_I(inode);
2646 	struct ext4_sb_info *sbi;
2647 	__u32 ino_next;
2648 	struct ext4_iloc iloc;
2649 	int err = 0;
2650 
2651 	if (!EXT4_SB(inode->i_sb)->s_journal)
2652 		return 0;
2653 
2654 	mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2655 	if (list_empty(&ei->i_orphan))
2656 		goto out;
2657 
2658 	ino_next = NEXT_ORPHAN(inode);
2659 	prev = ei->i_orphan.prev;
2660 	sbi = EXT4_SB(inode->i_sb);
2661 
2662 	jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2663 
2664 	list_del_init(&ei->i_orphan);
2665 
2666 	/* If we're on an error path, we may not have a valid
2667 	 * transaction handle with which to update the orphan list on
2668 	 * disk, but we still need to remove the inode from the linked
2669 	 * list in memory. */
2670 	if (!handle)
2671 		goto out;
2672 
2673 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2674 	if (err)
2675 		goto out_err;
2676 
2677 	if (prev == &sbi->s_orphan) {
2678 		jbd_debug(4, "superblock will point to %u\n", ino_next);
2679 		BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2680 		err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2681 		if (err)
2682 			goto out_brelse;
2683 		sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2684 		err = ext4_handle_dirty_super(handle, inode->i_sb);
2685 	} else {
2686 		struct ext4_iloc iloc2;
2687 		struct inode *i_prev =
2688 			&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2689 
2690 		jbd_debug(4, "orphan inode %lu will point to %u\n",
2691 			  i_prev->i_ino, ino_next);
2692 		err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2693 		if (err)
2694 			goto out_brelse;
2695 		NEXT_ORPHAN(i_prev) = ino_next;
2696 		err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2697 	}
2698 	if (err)
2699 		goto out_brelse;
2700 	NEXT_ORPHAN(inode) = 0;
2701 	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2702 
2703 out_err:
2704 	ext4_std_error(inode->i_sb, err);
2705 out:
2706 	mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2707 	return err;
2708 
2709 out_brelse:
2710 	brelse(iloc.bh);
2711 	goto out_err;
2712 }
2713 
2714 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2715 {
2716 	int retval;
2717 	struct inode *inode;
2718 	struct buffer_head *bh;
2719 	struct ext4_dir_entry_2 *de;
2720 	handle_t *handle;
2721 
2722 	/* Initialize quotas before so that eventual writes go in
2723 	 * separate transaction */
2724 	dquot_initialize(dir);
2725 	dquot_initialize(dentry->d_inode);
2726 
2727 	handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2728 	if (IS_ERR(handle))
2729 		return PTR_ERR(handle);
2730 
2731 	retval = -ENOENT;
2732 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2733 	if (!bh)
2734 		goto end_rmdir;
2735 
2736 	if (IS_DIRSYNC(dir))
2737 		ext4_handle_sync(handle);
2738 
2739 	inode = dentry->d_inode;
2740 
2741 	retval = -EIO;
2742 	if (le32_to_cpu(de->inode) != inode->i_ino)
2743 		goto end_rmdir;
2744 
2745 	retval = -ENOTEMPTY;
2746 	if (!empty_dir(inode))
2747 		goto end_rmdir;
2748 
2749 	retval = ext4_delete_entry(handle, dir, de, bh);
2750 	if (retval)
2751 		goto end_rmdir;
2752 	if (!EXT4_DIR_LINK_EMPTY(inode))
2753 		ext4_warning(inode->i_sb,
2754 			     "empty directory has too many links (%d)",
2755 			     inode->i_nlink);
2756 	inode->i_version++;
2757 	clear_nlink(inode);
2758 	/* There's no need to set i_disksize: the fact that i_nlink is
2759 	 * zero will ensure that the right thing happens during any
2760 	 * recovery. */
2761 	inode->i_size = 0;
2762 	ext4_orphan_add(handle, inode);
2763 	inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2764 	ext4_mark_inode_dirty(handle, inode);
2765 	ext4_dec_count(handle, dir);
2766 	ext4_update_dx_flag(dir);
2767 	ext4_mark_inode_dirty(handle, dir);
2768 
2769 end_rmdir:
2770 	ext4_journal_stop(handle);
2771 	brelse(bh);
2772 	return retval;
2773 }
2774 
2775 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2776 {
2777 	int retval;
2778 	struct inode *inode;
2779 	struct buffer_head *bh;
2780 	struct ext4_dir_entry_2 *de;
2781 	handle_t *handle;
2782 
2783 	trace_ext4_unlink_enter(dir, dentry);
2784 	/* Initialize quotas before so that eventual writes go
2785 	 * in separate transaction */
2786 	dquot_initialize(dir);
2787 	dquot_initialize(dentry->d_inode);
2788 
2789 	handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2790 	if (IS_ERR(handle))
2791 		return PTR_ERR(handle);
2792 
2793 	if (IS_DIRSYNC(dir))
2794 		ext4_handle_sync(handle);
2795 
2796 	retval = -ENOENT;
2797 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2798 	if (!bh)
2799 		goto end_unlink;
2800 
2801 	inode = dentry->d_inode;
2802 
2803 	retval = -EIO;
2804 	if (le32_to_cpu(de->inode) != inode->i_ino)
2805 		goto end_unlink;
2806 
2807 	if (!inode->i_nlink) {
2808 		ext4_warning(inode->i_sb,
2809 			     "Deleting nonexistent file (%lu), %d",
2810 			     inode->i_ino, inode->i_nlink);
2811 		set_nlink(inode, 1);
2812 	}
2813 	retval = ext4_delete_entry(handle, dir, de, bh);
2814 	if (retval)
2815 		goto end_unlink;
2816 	dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2817 	ext4_update_dx_flag(dir);
2818 	ext4_mark_inode_dirty(handle, dir);
2819 	drop_nlink(inode);
2820 	if (!inode->i_nlink)
2821 		ext4_orphan_add(handle, inode);
2822 	inode->i_ctime = ext4_current_time(inode);
2823 	ext4_mark_inode_dirty(handle, inode);
2824 	retval = 0;
2825 
2826 end_unlink:
2827 	ext4_journal_stop(handle);
2828 	brelse(bh);
2829 	trace_ext4_unlink_exit(dentry, retval);
2830 	return retval;
2831 }
2832 
2833 static int ext4_symlink(struct inode *dir,
2834 			struct dentry *dentry, const char *symname)
2835 {
2836 	handle_t *handle;
2837 	struct inode *inode;
2838 	int l, err, retries = 0;
2839 	int credits;
2840 
2841 	l = strlen(symname)+1;
2842 	if (l > dir->i_sb->s_blocksize)
2843 		return -ENAMETOOLONG;
2844 
2845 	dquot_initialize(dir);
2846 
2847 	if (l > EXT4_N_BLOCKS * 4) {
2848 		/*
2849 		 * For non-fast symlinks, we just allocate inode and put it on
2850 		 * orphan list in the first transaction => we need bitmap,
2851 		 * group descriptor, sb, inode block, quota blocks, and
2852 		 * possibly selinux xattr blocks.
2853 		 */
2854 		credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2855 			  EXT4_XATTR_TRANS_BLOCKS;
2856 	} else {
2857 		/*
2858 		 * Fast symlink. We have to add entry to directory
2859 		 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2860 		 * allocate new inode (bitmap, group descriptor, inode block,
2861 		 * quota blocks, sb is already counted in previous macros).
2862 		 */
2863 		credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2864 			  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2865 			  EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2866 	}
2867 retry:
2868 	handle = ext4_journal_start(dir, credits);
2869 	if (IS_ERR(handle))
2870 		return PTR_ERR(handle);
2871 
2872 	if (IS_DIRSYNC(dir))
2873 		ext4_handle_sync(handle);
2874 
2875 	inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2876 			       &dentry->d_name, 0, NULL);
2877 	err = PTR_ERR(inode);
2878 	if (IS_ERR(inode))
2879 		goto out_stop;
2880 
2881 	if (l > EXT4_N_BLOCKS * 4) {
2882 		inode->i_op = &ext4_symlink_inode_operations;
2883 		ext4_set_aops(inode);
2884 		/*
2885 		 * We cannot call page_symlink() with transaction started
2886 		 * because it calls into ext4_write_begin() which can wait
2887 		 * for transaction commit if we are running out of space
2888 		 * and thus we deadlock. So we have to stop transaction now
2889 		 * and restart it when symlink contents is written.
2890 		 *
2891 		 * To keep fs consistent in case of crash, we have to put inode
2892 		 * to orphan list in the mean time.
2893 		 */
2894 		drop_nlink(inode);
2895 		err = ext4_orphan_add(handle, inode);
2896 		ext4_journal_stop(handle);
2897 		if (err)
2898 			goto err_drop_inode;
2899 		err = __page_symlink(inode, symname, l, 1);
2900 		if (err)
2901 			goto err_drop_inode;
2902 		/*
2903 		 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2904 		 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2905 		 */
2906 		handle = ext4_journal_start(dir,
2907 				EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2908 				EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2909 		if (IS_ERR(handle)) {
2910 			err = PTR_ERR(handle);
2911 			goto err_drop_inode;
2912 		}
2913 		set_nlink(inode, 1);
2914 		err = ext4_orphan_del(handle, inode);
2915 		if (err) {
2916 			ext4_journal_stop(handle);
2917 			clear_nlink(inode);
2918 			goto err_drop_inode;
2919 		}
2920 	} else {
2921 		/* clear the extent format for fast symlink */
2922 		ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2923 		inode->i_op = &ext4_fast_symlink_inode_operations;
2924 		memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2925 		inode->i_size = l-1;
2926 	}
2927 	EXT4_I(inode)->i_disksize = inode->i_size;
2928 	err = ext4_add_nondir(handle, dentry, inode);
2929 out_stop:
2930 	ext4_journal_stop(handle);
2931 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2932 		goto retry;
2933 	return err;
2934 err_drop_inode:
2935 	unlock_new_inode(inode);
2936 	iput(inode);
2937 	return err;
2938 }
2939 
2940 static int ext4_link(struct dentry *old_dentry,
2941 		     struct inode *dir, struct dentry *dentry)
2942 {
2943 	handle_t *handle;
2944 	struct inode *inode = old_dentry->d_inode;
2945 	int err, retries = 0;
2946 
2947 	if (inode->i_nlink >= EXT4_LINK_MAX)
2948 		return -EMLINK;
2949 
2950 	dquot_initialize(dir);
2951 
2952 retry:
2953 	handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2954 					EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2955 	if (IS_ERR(handle))
2956 		return PTR_ERR(handle);
2957 
2958 	if (IS_DIRSYNC(dir))
2959 		ext4_handle_sync(handle);
2960 
2961 	inode->i_ctime = ext4_current_time(inode);
2962 	ext4_inc_count(handle, inode);
2963 	ihold(inode);
2964 
2965 	err = ext4_add_entry(handle, dentry, inode);
2966 	if (!err) {
2967 		ext4_mark_inode_dirty(handle, inode);
2968 		d_instantiate(dentry, inode);
2969 	} else {
2970 		drop_nlink(inode);
2971 		iput(inode);
2972 	}
2973 	ext4_journal_stop(handle);
2974 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2975 		goto retry;
2976 	return err;
2977 }
2978 
2979 
2980 /*
2981  * Try to find buffer head where contains the parent block.
2982  * It should be the inode block if it is inlined or the 1st block
2983  * if it is a normal dir.
2984  */
2985 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
2986 					struct inode *inode,
2987 					int *retval,
2988 					struct ext4_dir_entry_2 **parent_de,
2989 					int *inlined)
2990 {
2991 	struct buffer_head *bh;
2992 
2993 	if (!ext4_has_inline_data(inode)) {
2994 		if (!(bh = ext4_bread(handle, inode, 0, 0, retval))) {
2995 			if (!*retval) {
2996 				*retval = -EIO;
2997 				ext4_error(inode->i_sb,
2998 					   "Directory hole detected on inode %lu\n",
2999 					   inode->i_ino);
3000 			}
3001 			return NULL;
3002 		}
3003 		*parent_de = ext4_next_entry(
3004 					(struct ext4_dir_entry_2 *)bh->b_data,
3005 					inode->i_sb->s_blocksize);
3006 		return bh;
3007 	}
3008 
3009 	*inlined = 1;
3010 	return ext4_get_first_inline_block(inode, parent_de, retval);
3011 }
3012 
3013 /*
3014  * Anybody can rename anything with this: the permission checks are left to the
3015  * higher-level routines.
3016  */
3017 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3018 		       struct inode *new_dir, struct dentry *new_dentry)
3019 {
3020 	handle_t *handle;
3021 	struct inode *old_inode, *new_inode;
3022 	struct buffer_head *old_bh, *new_bh, *dir_bh;
3023 	struct ext4_dir_entry_2 *old_de, *new_de;
3024 	int retval, force_da_alloc = 0;
3025 	int inlined = 0, new_inlined = 0;
3026 	struct ext4_dir_entry_2 *parent_de;
3027 
3028 	dquot_initialize(old_dir);
3029 	dquot_initialize(new_dir);
3030 
3031 	old_bh = new_bh = dir_bh = NULL;
3032 
3033 	/* Initialize quotas before so that eventual writes go
3034 	 * in separate transaction */
3035 	if (new_dentry->d_inode)
3036 		dquot_initialize(new_dentry->d_inode);
3037 	handle = ext4_journal_start(old_dir, 2 *
3038 					EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
3039 					EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3040 	if (IS_ERR(handle))
3041 		return PTR_ERR(handle);
3042 
3043 	if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
3044 		ext4_handle_sync(handle);
3045 
3046 	old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de, NULL);
3047 	/*
3048 	 *  Check for inode number is _not_ due to possible IO errors.
3049 	 *  We might rmdir the source, keep it as pwd of some process
3050 	 *  and merrily kill the link to whatever was created under the
3051 	 *  same name. Goodbye sticky bit ;-<
3052 	 */
3053 	old_inode = old_dentry->d_inode;
3054 	retval = -ENOENT;
3055 	if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
3056 		goto end_rename;
3057 
3058 	new_inode = new_dentry->d_inode;
3059 	new_bh = ext4_find_entry(new_dir, &new_dentry->d_name,
3060 				 &new_de, &new_inlined);
3061 	if (new_bh) {
3062 		if (!new_inode) {
3063 			brelse(new_bh);
3064 			new_bh = NULL;
3065 		}
3066 	}
3067 	if (S_ISDIR(old_inode->i_mode)) {
3068 		if (new_inode) {
3069 			retval = -ENOTEMPTY;
3070 			if (!empty_dir(new_inode))
3071 				goto end_rename;
3072 		}
3073 		retval = -EIO;
3074 		dir_bh = ext4_get_first_dir_block(handle, old_inode,
3075 						  &retval, &parent_de,
3076 						  &inlined);
3077 		if (!dir_bh)
3078 			goto end_rename;
3079 		if (!inlined && !buffer_verified(dir_bh) &&
3080 		    !ext4_dirent_csum_verify(old_inode,
3081 				(struct ext4_dir_entry *)dir_bh->b_data))
3082 			goto end_rename;
3083 		set_buffer_verified(dir_bh);
3084 		if (le32_to_cpu(parent_de->inode) != old_dir->i_ino)
3085 			goto end_rename;
3086 		retval = -EMLINK;
3087 		if (!new_inode && new_dir != old_dir &&
3088 		    EXT4_DIR_LINK_MAX(new_dir))
3089 			goto end_rename;
3090 		BUFFER_TRACE(dir_bh, "get_write_access");
3091 		retval = ext4_journal_get_write_access(handle, dir_bh);
3092 		if (retval)
3093 			goto end_rename;
3094 	}
3095 	if (!new_bh) {
3096 		retval = ext4_add_entry(handle, new_dentry, old_inode);
3097 		if (retval)
3098 			goto end_rename;
3099 	} else {
3100 		BUFFER_TRACE(new_bh, "get write access");
3101 		retval = ext4_journal_get_write_access(handle, new_bh);
3102 		if (retval)
3103 			goto end_rename;
3104 		new_de->inode = cpu_to_le32(old_inode->i_ino);
3105 		if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
3106 					      EXT4_FEATURE_INCOMPAT_FILETYPE))
3107 			new_de->file_type = old_de->file_type;
3108 		new_dir->i_version++;
3109 		new_dir->i_ctime = new_dir->i_mtime =
3110 					ext4_current_time(new_dir);
3111 		ext4_mark_inode_dirty(handle, new_dir);
3112 		BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
3113 		if (!new_inlined) {
3114 			retval = ext4_handle_dirty_dirent_node(handle,
3115 							       new_dir, new_bh);
3116 			if (unlikely(retval)) {
3117 				ext4_std_error(new_dir->i_sb, retval);
3118 				goto end_rename;
3119 			}
3120 		}
3121 		brelse(new_bh);
3122 		new_bh = NULL;
3123 	}
3124 
3125 	/*
3126 	 * Like most other Unix systems, set the ctime for inodes on a
3127 	 * rename.
3128 	 */
3129 	old_inode->i_ctime = ext4_current_time(old_inode);
3130 	ext4_mark_inode_dirty(handle, old_inode);
3131 
3132 	/*
3133 	 * ok, that's it
3134 	 */
3135 	if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
3136 	    old_de->name_len != old_dentry->d_name.len ||
3137 	    strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
3138 	    (retval = ext4_delete_entry(handle, old_dir,
3139 					old_de, old_bh)) == -ENOENT) {
3140 		/* old_de could have moved from under us during htree split, so
3141 		 * make sure that we are deleting the right entry.  We might
3142 		 * also be pointing to a stale entry in the unused part of
3143 		 * old_bh so just checking inum and the name isn't enough. */
3144 		struct buffer_head *old_bh2;
3145 		struct ext4_dir_entry_2 *old_de2;
3146 
3147 		old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name,
3148 					  &old_de2, NULL);
3149 		if (old_bh2) {
3150 			retval = ext4_delete_entry(handle, old_dir,
3151 						   old_de2, old_bh2);
3152 			brelse(old_bh2);
3153 		}
3154 	}
3155 	if (retval) {
3156 		ext4_warning(old_dir->i_sb,
3157 				"Deleting old file (%lu), %d, error=%d",
3158 				old_dir->i_ino, old_dir->i_nlink, retval);
3159 	}
3160 
3161 	if (new_inode) {
3162 		ext4_dec_count(handle, new_inode);
3163 		new_inode->i_ctime = ext4_current_time(new_inode);
3164 	}
3165 	old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
3166 	ext4_update_dx_flag(old_dir);
3167 	if (dir_bh) {
3168 		parent_de->inode = cpu_to_le32(new_dir->i_ino);
3169 		BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
3170 		if (!inlined) {
3171 			if (is_dx(old_inode)) {
3172 				retval = ext4_handle_dirty_dx_node(handle,
3173 								   old_inode,
3174 								   dir_bh);
3175 			} else {
3176 				retval = ext4_handle_dirty_dirent_node(handle,
3177 							old_inode, dir_bh);
3178 			}
3179 		} else {
3180 			retval = ext4_mark_inode_dirty(handle, old_inode);
3181 		}
3182 		if (retval) {
3183 			ext4_std_error(old_dir->i_sb, retval);
3184 			goto end_rename;
3185 		}
3186 		ext4_dec_count(handle, old_dir);
3187 		if (new_inode) {
3188 			/* checked empty_dir above, can't have another parent,
3189 			 * ext4_dec_count() won't work for many-linked dirs */
3190 			clear_nlink(new_inode);
3191 		} else {
3192 			ext4_inc_count(handle, new_dir);
3193 			ext4_update_dx_flag(new_dir);
3194 			ext4_mark_inode_dirty(handle, new_dir);
3195 		}
3196 	}
3197 	ext4_mark_inode_dirty(handle, old_dir);
3198 	if (new_inode) {
3199 		ext4_mark_inode_dirty(handle, new_inode);
3200 		if (!new_inode->i_nlink)
3201 			ext4_orphan_add(handle, new_inode);
3202 		if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
3203 			force_da_alloc = 1;
3204 	}
3205 	retval = 0;
3206 
3207 end_rename:
3208 	brelse(dir_bh);
3209 	brelse(old_bh);
3210 	brelse(new_bh);
3211 	ext4_journal_stop(handle);
3212 	if (retval == 0 && force_da_alloc)
3213 		ext4_alloc_da_blocks(old_inode);
3214 	return retval;
3215 }
3216 
3217 /*
3218  * directories can handle most operations...
3219  */
3220 const struct inode_operations ext4_dir_inode_operations = {
3221 	.create		= ext4_create,
3222 	.lookup		= ext4_lookup,
3223 	.link		= ext4_link,
3224 	.unlink		= ext4_unlink,
3225 	.symlink	= ext4_symlink,
3226 	.mkdir		= ext4_mkdir,
3227 	.rmdir		= ext4_rmdir,
3228 	.mknod		= ext4_mknod,
3229 	.rename		= ext4_rename,
3230 	.setattr	= ext4_setattr,
3231 	.setxattr	= generic_setxattr,
3232 	.getxattr	= generic_getxattr,
3233 	.listxattr	= ext4_listxattr,
3234 	.removexattr	= generic_removexattr,
3235 	.get_acl	= ext4_get_acl,
3236 	.fiemap         = ext4_fiemap,
3237 };
3238 
3239 const struct inode_operations ext4_special_inode_operations = {
3240 	.setattr	= ext4_setattr,
3241 	.setxattr	= generic_setxattr,
3242 	.getxattr	= generic_getxattr,
3243 	.listxattr	= ext4_listxattr,
3244 	.removexattr	= generic_removexattr,
3245 	.get_acl	= ext4_get_acl,
3246 };
3247