xref: /openbmc/linux/fs/ext4/namei.c (revision 034f90b3)
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 
51 static struct buffer_head *ext4_append(handle_t *handle,
52 					struct inode *inode,
53 					ext4_lblk_t *block)
54 {
55 	struct buffer_head *bh;
56 	int err;
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 		return ERR_PTR(-ENOSPC);
62 
63 	*block = inode->i_size >> inode->i_sb->s_blocksize_bits;
64 
65 	bh = ext4_bread(handle, inode, *block, 1);
66 	if (IS_ERR(bh))
67 		return bh;
68 	inode->i_size += inode->i_sb->s_blocksize;
69 	EXT4_I(inode)->i_disksize = inode->i_size;
70 	BUFFER_TRACE(bh, "get_write_access");
71 	err = ext4_journal_get_write_access(handle, bh);
72 	if (err) {
73 		brelse(bh);
74 		ext4_std_error(inode->i_sb, err);
75 		return ERR_PTR(err);
76 	}
77 	return bh;
78 }
79 
80 static int ext4_dx_csum_verify(struct inode *inode,
81 			       struct ext4_dir_entry *dirent);
82 
83 typedef enum {
84 	EITHER, INDEX, DIRENT
85 } dirblock_type_t;
86 
87 #define ext4_read_dirblock(inode, block, type) \
88 	__ext4_read_dirblock((inode), (block), (type), __LINE__)
89 
90 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
91 					      ext4_lblk_t block,
92 					      dirblock_type_t type,
93 					      unsigned int line)
94 {
95 	struct buffer_head *bh;
96 	struct ext4_dir_entry *dirent;
97 	int is_dx_block = 0;
98 
99 	bh = ext4_bread(NULL, inode, block, 0);
100 	if (IS_ERR(bh)) {
101 		__ext4_warning(inode->i_sb, __func__, line,
102 			       "error %ld reading directory block "
103 			       "(ino %lu, block %lu)", PTR_ERR(bh), inode->i_ino,
104 			       (unsigned long) block);
105 
106 		return bh;
107 	}
108 	if (!bh) {
109 		ext4_error_inode(inode, __func__, line, block, "Directory hole found");
110 		return ERR_PTR(-EIO);
111 	}
112 	dirent = (struct ext4_dir_entry *) bh->b_data;
113 	/* Determine whether or not we have an index block */
114 	if (is_dx(inode)) {
115 		if (block == 0)
116 			is_dx_block = 1;
117 		else if (ext4_rec_len_from_disk(dirent->rec_len,
118 						inode->i_sb->s_blocksize) ==
119 			 inode->i_sb->s_blocksize)
120 			is_dx_block = 1;
121 	}
122 	if (!is_dx_block && type == INDEX) {
123 		ext4_error_inode(inode, __func__, line, block,
124 		       "directory leaf block found instead of index block");
125 		return ERR_PTR(-EIO);
126 	}
127 	if (!ext4_has_metadata_csum(inode->i_sb) ||
128 	    buffer_verified(bh))
129 		return bh;
130 
131 	/*
132 	 * An empty leaf block can get mistaken for a index block; for
133 	 * this reason, we can only check the index checksum when the
134 	 * caller is sure it should be an index block.
135 	 */
136 	if (is_dx_block && type == INDEX) {
137 		if (ext4_dx_csum_verify(inode, dirent))
138 			set_buffer_verified(bh);
139 		else {
140 			ext4_error_inode(inode, __func__, line, block,
141 				"Directory index failed checksum");
142 			brelse(bh);
143 			return ERR_PTR(-EIO);
144 		}
145 	}
146 	if (!is_dx_block) {
147 		if (ext4_dirent_csum_verify(inode, dirent))
148 			set_buffer_verified(bh);
149 		else {
150 			ext4_error_inode(inode, __func__, line, block,
151 				"Directory block failed checksum");
152 			brelse(bh);
153 			return ERR_PTR(-EIO);
154 		}
155 	}
156 	return bh;
157 }
158 
159 #ifndef assert
160 #define assert(test) J_ASSERT(test)
161 #endif
162 
163 #ifdef DX_DEBUG
164 #define dxtrace(command) command
165 #else
166 #define dxtrace(command)
167 #endif
168 
169 struct fake_dirent
170 {
171 	__le32 inode;
172 	__le16 rec_len;
173 	u8 name_len;
174 	u8 file_type;
175 };
176 
177 struct dx_countlimit
178 {
179 	__le16 limit;
180 	__le16 count;
181 };
182 
183 struct dx_entry
184 {
185 	__le32 hash;
186 	__le32 block;
187 };
188 
189 /*
190  * dx_root_info is laid out so that if it should somehow get overlaid by a
191  * dirent the two low bits of the hash version will be zero.  Therefore, the
192  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
193  */
194 
195 struct dx_root
196 {
197 	struct fake_dirent dot;
198 	char dot_name[4];
199 	struct fake_dirent dotdot;
200 	char dotdot_name[4];
201 	struct dx_root_info
202 	{
203 		__le32 reserved_zero;
204 		u8 hash_version;
205 		u8 info_length; /* 8 */
206 		u8 indirect_levels;
207 		u8 unused_flags;
208 	}
209 	info;
210 	struct dx_entry	entries[0];
211 };
212 
213 struct dx_node
214 {
215 	struct fake_dirent fake;
216 	struct dx_entry	entries[0];
217 };
218 
219 
220 struct dx_frame
221 {
222 	struct buffer_head *bh;
223 	struct dx_entry *entries;
224 	struct dx_entry *at;
225 };
226 
227 struct dx_map_entry
228 {
229 	u32 hash;
230 	u16 offs;
231 	u16 size;
232 };
233 
234 /*
235  * This goes at the end of each htree block.
236  */
237 struct dx_tail {
238 	u32 dt_reserved;
239 	__le32 dt_checksum;	/* crc32c(uuid+inum+dirblock) */
240 };
241 
242 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
243 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
244 static inline unsigned dx_get_hash(struct dx_entry *entry);
245 static void dx_set_hash(struct dx_entry *entry, unsigned value);
246 static unsigned dx_get_count(struct dx_entry *entries);
247 static unsigned dx_get_limit(struct dx_entry *entries);
248 static void dx_set_count(struct dx_entry *entries, unsigned value);
249 static void dx_set_limit(struct dx_entry *entries, unsigned value);
250 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
251 static unsigned dx_node_limit(struct inode *dir);
252 static struct dx_frame *dx_probe(const struct qstr *d_name,
253 				 struct inode *dir,
254 				 struct dx_hash_info *hinfo,
255 				 struct dx_frame *frame);
256 static void dx_release(struct dx_frame *frames);
257 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
258 		       struct dx_hash_info *hinfo, struct dx_map_entry map[]);
259 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
260 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
261 		struct dx_map_entry *offsets, int count, unsigned blocksize);
262 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
263 static void dx_insert_block(struct dx_frame *frame,
264 					u32 hash, ext4_lblk_t block);
265 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
266 				 struct dx_frame *frame,
267 				 struct dx_frame *frames,
268 				 __u32 *start_hash);
269 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
270 		const struct qstr *d_name,
271 		struct ext4_dir_entry_2 **res_dir);
272 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
273 			     struct inode *inode);
274 
275 /* checksumming functions */
276 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
277 			    unsigned int blocksize)
278 {
279 	memset(t, 0, sizeof(struct ext4_dir_entry_tail));
280 	t->det_rec_len = ext4_rec_len_to_disk(
281 			sizeof(struct ext4_dir_entry_tail), blocksize);
282 	t->det_reserved_ft = EXT4_FT_DIR_CSUM;
283 }
284 
285 /* Walk through a dirent block to find a checksum "dirent" at the tail */
286 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
287 						   struct ext4_dir_entry *de)
288 {
289 	struct ext4_dir_entry_tail *t;
290 
291 #ifdef PARANOID
292 	struct ext4_dir_entry *d, *top;
293 
294 	d = de;
295 	top = (struct ext4_dir_entry *)(((void *)de) +
296 		(EXT4_BLOCK_SIZE(inode->i_sb) -
297 		sizeof(struct ext4_dir_entry_tail)));
298 	while (d < top && d->rec_len)
299 		d = (struct ext4_dir_entry *)(((void *)d) +
300 		    le16_to_cpu(d->rec_len));
301 
302 	if (d != top)
303 		return NULL;
304 
305 	t = (struct ext4_dir_entry_tail *)d;
306 #else
307 	t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
308 #endif
309 
310 	if (t->det_reserved_zero1 ||
311 	    le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
312 	    t->det_reserved_zero2 ||
313 	    t->det_reserved_ft != EXT4_FT_DIR_CSUM)
314 		return NULL;
315 
316 	return t;
317 }
318 
319 static __le32 ext4_dirent_csum(struct inode *inode,
320 			       struct ext4_dir_entry *dirent, int size)
321 {
322 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
323 	struct ext4_inode_info *ei = EXT4_I(inode);
324 	__u32 csum;
325 
326 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
327 	return cpu_to_le32(csum);
328 }
329 
330 static void warn_no_space_for_csum(struct inode *inode)
331 {
332 	ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
333 		     "checksum.  Please run e2fsck -D.", inode->i_ino);
334 }
335 
336 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
337 {
338 	struct ext4_dir_entry_tail *t;
339 
340 	if (!ext4_has_metadata_csum(inode->i_sb))
341 		return 1;
342 
343 	t = get_dirent_tail(inode, dirent);
344 	if (!t) {
345 		warn_no_space_for_csum(inode);
346 		return 0;
347 	}
348 
349 	if (t->det_checksum != ext4_dirent_csum(inode, dirent,
350 						(void *)t - (void *)dirent))
351 		return 0;
352 
353 	return 1;
354 }
355 
356 static void ext4_dirent_csum_set(struct inode *inode,
357 				 struct ext4_dir_entry *dirent)
358 {
359 	struct ext4_dir_entry_tail *t;
360 
361 	if (!ext4_has_metadata_csum(inode->i_sb))
362 		return;
363 
364 	t = get_dirent_tail(inode, dirent);
365 	if (!t) {
366 		warn_no_space_for_csum(inode);
367 		return;
368 	}
369 
370 	t->det_checksum = ext4_dirent_csum(inode, dirent,
371 					   (void *)t - (void *)dirent);
372 }
373 
374 int ext4_handle_dirty_dirent_node(handle_t *handle,
375 				  struct inode *inode,
376 				  struct buffer_head *bh)
377 {
378 	ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
379 	return ext4_handle_dirty_metadata(handle, inode, bh);
380 }
381 
382 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
383 					       struct ext4_dir_entry *dirent,
384 					       int *offset)
385 {
386 	struct ext4_dir_entry *dp;
387 	struct dx_root_info *root;
388 	int count_offset;
389 
390 	if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
391 		count_offset = 8;
392 	else if (le16_to_cpu(dirent->rec_len) == 12) {
393 		dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
394 		if (le16_to_cpu(dp->rec_len) !=
395 		    EXT4_BLOCK_SIZE(inode->i_sb) - 12)
396 			return NULL;
397 		root = (struct dx_root_info *)(((void *)dp + 12));
398 		if (root->reserved_zero ||
399 		    root->info_length != sizeof(struct dx_root_info))
400 			return NULL;
401 		count_offset = 32;
402 	} else
403 		return NULL;
404 
405 	if (offset)
406 		*offset = count_offset;
407 	return (struct dx_countlimit *)(((void *)dirent) + count_offset);
408 }
409 
410 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
411 			   int count_offset, int count, struct dx_tail *t)
412 {
413 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
414 	struct ext4_inode_info *ei = EXT4_I(inode);
415 	__u32 csum;
416 	__le32 save_csum;
417 	int size;
418 
419 	size = count_offset + (count * sizeof(struct dx_entry));
420 	save_csum = t->dt_checksum;
421 	t->dt_checksum = 0;
422 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
423 	csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
424 	t->dt_checksum = save_csum;
425 
426 	return cpu_to_le32(csum);
427 }
428 
429 static int ext4_dx_csum_verify(struct inode *inode,
430 			       struct ext4_dir_entry *dirent)
431 {
432 	struct dx_countlimit *c;
433 	struct dx_tail *t;
434 	int count_offset, limit, count;
435 
436 	if (!ext4_has_metadata_csum(inode->i_sb))
437 		return 1;
438 
439 	c = get_dx_countlimit(inode, dirent, &count_offset);
440 	if (!c) {
441 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
442 		return 1;
443 	}
444 	limit = le16_to_cpu(c->limit);
445 	count = le16_to_cpu(c->count);
446 	if (count_offset + (limit * sizeof(struct dx_entry)) >
447 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
448 		warn_no_space_for_csum(inode);
449 		return 1;
450 	}
451 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
452 
453 	if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
454 					    count, t))
455 		return 0;
456 	return 1;
457 }
458 
459 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
460 {
461 	struct dx_countlimit *c;
462 	struct dx_tail *t;
463 	int count_offset, limit, count;
464 
465 	if (!ext4_has_metadata_csum(inode->i_sb))
466 		return;
467 
468 	c = get_dx_countlimit(inode, dirent, &count_offset);
469 	if (!c) {
470 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
471 		return;
472 	}
473 	limit = le16_to_cpu(c->limit);
474 	count = le16_to_cpu(c->count);
475 	if (count_offset + (limit * sizeof(struct dx_entry)) >
476 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
477 		warn_no_space_for_csum(inode);
478 		return;
479 	}
480 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
481 
482 	t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
483 }
484 
485 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
486 					    struct inode *inode,
487 					    struct buffer_head *bh)
488 {
489 	ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
490 	return ext4_handle_dirty_metadata(handle, inode, bh);
491 }
492 
493 /*
494  * p is at least 6 bytes before the end of page
495  */
496 static inline struct ext4_dir_entry_2 *
497 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
498 {
499 	return (struct ext4_dir_entry_2 *)((char *)p +
500 		ext4_rec_len_from_disk(p->rec_len, blocksize));
501 }
502 
503 /*
504  * Future: use high four bits of block for coalesce-on-delete flags
505  * Mask them off for now.
506  */
507 
508 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
509 {
510 	return le32_to_cpu(entry->block) & 0x00ffffff;
511 }
512 
513 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
514 {
515 	entry->block = cpu_to_le32(value);
516 }
517 
518 static inline unsigned dx_get_hash(struct dx_entry *entry)
519 {
520 	return le32_to_cpu(entry->hash);
521 }
522 
523 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
524 {
525 	entry->hash = cpu_to_le32(value);
526 }
527 
528 static inline unsigned dx_get_count(struct dx_entry *entries)
529 {
530 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
531 }
532 
533 static inline unsigned dx_get_limit(struct dx_entry *entries)
534 {
535 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
536 }
537 
538 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
539 {
540 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
541 }
542 
543 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
544 {
545 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
546 }
547 
548 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
549 {
550 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
551 		EXT4_DIR_REC_LEN(2) - infosize;
552 
553 	if (ext4_has_metadata_csum(dir->i_sb))
554 		entry_space -= sizeof(struct dx_tail);
555 	return entry_space / sizeof(struct dx_entry);
556 }
557 
558 static inline unsigned dx_node_limit(struct inode *dir)
559 {
560 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
561 
562 	if (ext4_has_metadata_csum(dir->i_sb))
563 		entry_space -= sizeof(struct dx_tail);
564 	return entry_space / sizeof(struct dx_entry);
565 }
566 
567 /*
568  * Debug
569  */
570 #ifdef DX_DEBUG
571 static void dx_show_index(char * label, struct dx_entry *entries)
572 {
573 	int i, n = dx_get_count (entries);
574 	printk(KERN_DEBUG "%s index ", label);
575 	for (i = 0; i < n; i++) {
576 		printk("%x->%lu ", i ? dx_get_hash(entries + i) :
577 				0, (unsigned long)dx_get_block(entries + i));
578 	}
579 	printk("\n");
580 }
581 
582 struct stats
583 {
584 	unsigned names;
585 	unsigned space;
586 	unsigned bcount;
587 };
588 
589 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
590 				 int size, int show_names)
591 {
592 	unsigned names = 0, space = 0;
593 	char *base = (char *) de;
594 	struct dx_hash_info h = *hinfo;
595 
596 	printk("names: ");
597 	while ((char *) de < base + size)
598 	{
599 		if (de->inode)
600 		{
601 			if (show_names)
602 			{
603 				int len = de->name_len;
604 				char *name = de->name;
605 				while (len--) printk("%c", *name++);
606 				ext4fs_dirhash(de->name, de->name_len, &h);
607 				printk(":%x.%u ", h.hash,
608 				       (unsigned) ((char *) de - base));
609 			}
610 			space += EXT4_DIR_REC_LEN(de->name_len);
611 			names++;
612 		}
613 		de = ext4_next_entry(de, size);
614 	}
615 	printk("(%i)\n", names);
616 	return (struct stats) { names, space, 1 };
617 }
618 
619 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
620 			     struct dx_entry *entries, int levels)
621 {
622 	unsigned blocksize = dir->i_sb->s_blocksize;
623 	unsigned count = dx_get_count(entries), names = 0, space = 0, i;
624 	unsigned bcount = 0;
625 	struct buffer_head *bh;
626 	int err;
627 	printk("%i indexed blocks...\n", count);
628 	for (i = 0; i < count; i++, entries++)
629 	{
630 		ext4_lblk_t block = dx_get_block(entries);
631 		ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
632 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
633 		struct stats stats;
634 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
635 		bh = ext4_bread(NULL,dir, block, 0);
636 		if (!bh || IS_ERR(bh))
637 			continue;
638 		stats = levels?
639 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
640 		   dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
641 		names += stats.names;
642 		space += stats.space;
643 		bcount += stats.bcount;
644 		brelse(bh);
645 	}
646 	if (bcount)
647 		printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
648 		       levels ? "" : "   ", names, space/bcount,
649 		       (space/bcount)*100/blocksize);
650 	return (struct stats) { names, space, bcount};
651 }
652 #endif /* DX_DEBUG */
653 
654 /*
655  * Probe for a directory leaf block to search.
656  *
657  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
658  * error in the directory index, and the caller should fall back to
659  * searching the directory normally.  The callers of dx_probe **MUST**
660  * check for this error code, and make sure it never gets reflected
661  * back to userspace.
662  */
663 static struct dx_frame *
664 dx_probe(const struct qstr *d_name, struct inode *dir,
665 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
666 {
667 	unsigned count, indirect;
668 	struct dx_entry *at, *entries, *p, *q, *m;
669 	struct dx_root *root;
670 	struct dx_frame *frame = frame_in;
671 	struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
672 	u32 hash;
673 
674 	frame->bh = ext4_read_dirblock(dir, 0, INDEX);
675 	if (IS_ERR(frame->bh))
676 		return (struct dx_frame *) frame->bh;
677 
678 	root = (struct dx_root *) frame->bh->b_data;
679 	if (root->info.hash_version != DX_HASH_TEA &&
680 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
681 	    root->info.hash_version != DX_HASH_LEGACY) {
682 		ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
683 			     root->info.hash_version);
684 		goto fail;
685 	}
686 	hinfo->hash_version = root->info.hash_version;
687 	if (hinfo->hash_version <= DX_HASH_TEA)
688 		hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
689 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
690 	if (d_name)
691 		ext4fs_dirhash(d_name->name, d_name->len, hinfo);
692 	hash = hinfo->hash;
693 
694 	if (root->info.unused_flags & 1) {
695 		ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
696 			     root->info.unused_flags);
697 		goto fail;
698 	}
699 
700 	if ((indirect = root->info.indirect_levels) > 1) {
701 		ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
702 			     root->info.indirect_levels);
703 		goto fail;
704 	}
705 
706 	entries = (struct dx_entry *) (((char *)&root->info) +
707 				       root->info.info_length);
708 
709 	if (dx_get_limit(entries) != dx_root_limit(dir,
710 						   root->info.info_length)) {
711 		ext4_warning(dir->i_sb, "dx entry: limit != root limit");
712 		goto fail;
713 	}
714 
715 	dxtrace(printk("Look up %x", hash));
716 	while (1) {
717 		count = dx_get_count(entries);
718 		if (!count || count > dx_get_limit(entries)) {
719 			ext4_warning(dir->i_sb,
720 				     "dx entry: no count or count > limit");
721 			goto fail;
722 		}
723 
724 		p = entries + 1;
725 		q = entries + count - 1;
726 		while (p <= q) {
727 			m = p + (q - p)/2;
728 			dxtrace(printk("."));
729 			if (dx_get_hash(m) > hash)
730 				q = m - 1;
731 			else
732 				p = m + 1;
733 		}
734 
735 		if (0) { // linear search cross check
736 			unsigned n = count - 1;
737 			at = entries;
738 			while (n--)
739 			{
740 				dxtrace(printk(","));
741 				if (dx_get_hash(++at) > hash)
742 				{
743 					at--;
744 					break;
745 				}
746 			}
747 			assert (at == p - 1);
748 		}
749 
750 		at = p - 1;
751 		dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
752 		frame->entries = entries;
753 		frame->at = at;
754 		if (!indirect--)
755 			return frame;
756 		frame++;
757 		frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
758 		if (IS_ERR(frame->bh)) {
759 			ret_err = (struct dx_frame *) frame->bh;
760 			frame->bh = NULL;
761 			goto fail;
762 		}
763 		entries = ((struct dx_node *) frame->bh->b_data)->entries;
764 
765 		if (dx_get_limit(entries) != dx_node_limit (dir)) {
766 			ext4_warning(dir->i_sb,
767 				     "dx entry: limit != node limit");
768 			goto fail;
769 		}
770 	}
771 fail:
772 	while (frame >= frame_in) {
773 		brelse(frame->bh);
774 		frame--;
775 	}
776 	if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
777 		ext4_warning(dir->i_sb,
778 			     "Corrupt dir inode %lu, running e2fsck is "
779 			     "recommended.", dir->i_ino);
780 	return ret_err;
781 }
782 
783 static void dx_release (struct dx_frame *frames)
784 {
785 	if (frames[0].bh == NULL)
786 		return;
787 
788 	if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
789 		brelse(frames[1].bh);
790 	brelse(frames[0].bh);
791 }
792 
793 /*
794  * This function increments the frame pointer to search the next leaf
795  * block, and reads in the necessary intervening nodes if the search
796  * should be necessary.  Whether or not the search is necessary is
797  * controlled by the hash parameter.  If the hash value is even, then
798  * the search is only continued if the next block starts with that
799  * hash value.  This is used if we are searching for a specific file.
800  *
801  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
802  *
803  * This function returns 1 if the caller should continue to search,
804  * or 0 if it should not.  If there is an error reading one of the
805  * index blocks, it will a negative error code.
806  *
807  * If start_hash is non-null, it will be filled in with the starting
808  * hash of the next page.
809  */
810 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
811 				 struct dx_frame *frame,
812 				 struct dx_frame *frames,
813 				 __u32 *start_hash)
814 {
815 	struct dx_frame *p;
816 	struct buffer_head *bh;
817 	int num_frames = 0;
818 	__u32 bhash;
819 
820 	p = frame;
821 	/*
822 	 * Find the next leaf page by incrementing the frame pointer.
823 	 * If we run out of entries in the interior node, loop around and
824 	 * increment pointer in the parent node.  When we break out of
825 	 * this loop, num_frames indicates the number of interior
826 	 * nodes need to be read.
827 	 */
828 	while (1) {
829 		if (++(p->at) < p->entries + dx_get_count(p->entries))
830 			break;
831 		if (p == frames)
832 			return 0;
833 		num_frames++;
834 		p--;
835 	}
836 
837 	/*
838 	 * If the hash is 1, then continue only if the next page has a
839 	 * continuation hash of any value.  This is used for readdir
840 	 * handling.  Otherwise, check to see if the hash matches the
841 	 * desired contiuation hash.  If it doesn't, return since
842 	 * there's no point to read in the successive index pages.
843 	 */
844 	bhash = dx_get_hash(p->at);
845 	if (start_hash)
846 		*start_hash = bhash;
847 	if ((hash & 1) == 0) {
848 		if ((bhash & ~1) != hash)
849 			return 0;
850 	}
851 	/*
852 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
853 	 * block so no check is necessary
854 	 */
855 	while (num_frames--) {
856 		bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
857 		if (IS_ERR(bh))
858 			return PTR_ERR(bh);
859 		p++;
860 		brelse(p->bh);
861 		p->bh = bh;
862 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
863 	}
864 	return 1;
865 }
866 
867 
868 /*
869  * This function fills a red-black tree with information from a
870  * directory block.  It returns the number directory entries loaded
871  * into the tree.  If there is an error it is returned in err.
872  */
873 static int htree_dirblock_to_tree(struct file *dir_file,
874 				  struct inode *dir, ext4_lblk_t block,
875 				  struct dx_hash_info *hinfo,
876 				  __u32 start_hash, __u32 start_minor_hash)
877 {
878 	struct buffer_head *bh;
879 	struct ext4_dir_entry_2 *de, *top;
880 	int err = 0, count = 0;
881 
882 	dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
883 							(unsigned long)block));
884 	bh = ext4_read_dirblock(dir, block, DIRENT);
885 	if (IS_ERR(bh))
886 		return PTR_ERR(bh);
887 
888 	de = (struct ext4_dir_entry_2 *) bh->b_data;
889 	top = (struct ext4_dir_entry_2 *) ((char *) de +
890 					   dir->i_sb->s_blocksize -
891 					   EXT4_DIR_REC_LEN(0));
892 	for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
893 		if (ext4_check_dir_entry(dir, NULL, de, bh,
894 				bh->b_data, bh->b_size,
895 				(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
896 					 + ((char *)de - bh->b_data))) {
897 			/* silently ignore the rest of the block */
898 			break;
899 		}
900 		ext4fs_dirhash(de->name, de->name_len, hinfo);
901 		if ((hinfo->hash < start_hash) ||
902 		    ((hinfo->hash == start_hash) &&
903 		     (hinfo->minor_hash < start_minor_hash)))
904 			continue;
905 		if (de->inode == 0)
906 			continue;
907 		if ((err = ext4_htree_store_dirent(dir_file,
908 				   hinfo->hash, hinfo->minor_hash, de)) != 0) {
909 			brelse(bh);
910 			return err;
911 		}
912 		count++;
913 	}
914 	brelse(bh);
915 	return count;
916 }
917 
918 
919 /*
920  * This function fills a red-black tree with information from a
921  * directory.  We start scanning the directory in hash order, starting
922  * at start_hash and start_minor_hash.
923  *
924  * This function returns the number of entries inserted into the tree,
925  * or a negative error code.
926  */
927 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
928 			 __u32 start_minor_hash, __u32 *next_hash)
929 {
930 	struct dx_hash_info hinfo;
931 	struct ext4_dir_entry_2 *de;
932 	struct dx_frame frames[2], *frame;
933 	struct inode *dir;
934 	ext4_lblk_t block;
935 	int count = 0;
936 	int ret, err;
937 	__u32 hashval;
938 
939 	dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
940 		       start_hash, start_minor_hash));
941 	dir = file_inode(dir_file);
942 	if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
943 		hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
944 		if (hinfo.hash_version <= DX_HASH_TEA)
945 			hinfo.hash_version +=
946 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
947 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
948 		if (ext4_has_inline_data(dir)) {
949 			int has_inline_data = 1;
950 			count = htree_inlinedir_to_tree(dir_file, dir, 0,
951 							&hinfo, start_hash,
952 							start_minor_hash,
953 							&has_inline_data);
954 			if (has_inline_data) {
955 				*next_hash = ~0;
956 				return count;
957 			}
958 		}
959 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
960 					       start_hash, start_minor_hash);
961 		*next_hash = ~0;
962 		return count;
963 	}
964 	hinfo.hash = start_hash;
965 	hinfo.minor_hash = 0;
966 	frame = dx_probe(NULL, dir, &hinfo, frames);
967 	if (IS_ERR(frame))
968 		return PTR_ERR(frame);
969 
970 	/* Add '.' and '..' from the htree header */
971 	if (!start_hash && !start_minor_hash) {
972 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
973 		if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
974 			goto errout;
975 		count++;
976 	}
977 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
978 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
979 		de = ext4_next_entry(de, dir->i_sb->s_blocksize);
980 		if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
981 			goto errout;
982 		count++;
983 	}
984 
985 	while (1) {
986 		block = dx_get_block(frame->at);
987 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
988 					     start_hash, start_minor_hash);
989 		if (ret < 0) {
990 			err = ret;
991 			goto errout;
992 		}
993 		count += ret;
994 		hashval = ~0;
995 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
996 					    frame, frames, &hashval);
997 		*next_hash = hashval;
998 		if (ret < 0) {
999 			err = ret;
1000 			goto errout;
1001 		}
1002 		/*
1003 		 * Stop if:  (a) there are no more entries, or
1004 		 * (b) we have inserted at least one entry and the
1005 		 * next hash value is not a continuation
1006 		 */
1007 		if ((ret == 0) ||
1008 		    (count && ((hashval & 1) == 0)))
1009 			break;
1010 	}
1011 	dx_release(frames);
1012 	dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1013 		       "next hash: %x\n", count, *next_hash));
1014 	return count;
1015 errout:
1016 	dx_release(frames);
1017 	return (err);
1018 }
1019 
1020 static inline int search_dirblock(struct buffer_head *bh,
1021 				  struct inode *dir,
1022 				  const struct qstr *d_name,
1023 				  unsigned int offset,
1024 				  struct ext4_dir_entry_2 **res_dir)
1025 {
1026 	return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1027 			  d_name, offset, res_dir);
1028 }
1029 
1030 /*
1031  * Directory block splitting, compacting
1032  */
1033 
1034 /*
1035  * Create map of hash values, offsets, and sizes, stored at end of block.
1036  * Returns number of entries mapped.
1037  */
1038 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1039 		       struct dx_hash_info *hinfo,
1040 		       struct dx_map_entry *map_tail)
1041 {
1042 	int count = 0;
1043 	char *base = (char *) de;
1044 	struct dx_hash_info h = *hinfo;
1045 
1046 	while ((char *) de < base + blocksize) {
1047 		if (de->name_len && de->inode) {
1048 			ext4fs_dirhash(de->name, de->name_len, &h);
1049 			map_tail--;
1050 			map_tail->hash = h.hash;
1051 			map_tail->offs = ((char *) de - base)>>2;
1052 			map_tail->size = le16_to_cpu(de->rec_len);
1053 			count++;
1054 			cond_resched();
1055 		}
1056 		/* XXX: do we need to check rec_len == 0 case? -Chris */
1057 		de = ext4_next_entry(de, blocksize);
1058 	}
1059 	return count;
1060 }
1061 
1062 /* Sort map by hash value */
1063 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1064 {
1065 	struct dx_map_entry *p, *q, *top = map + count - 1;
1066 	int more;
1067 	/* Combsort until bubble sort doesn't suck */
1068 	while (count > 2) {
1069 		count = count*10/13;
1070 		if (count - 9 < 2) /* 9, 10 -> 11 */
1071 			count = 11;
1072 		for (p = top, q = p - count; q >= map; p--, q--)
1073 			if (p->hash < q->hash)
1074 				swap(*p, *q);
1075 	}
1076 	/* Garden variety bubble sort */
1077 	do {
1078 		more = 0;
1079 		q = top;
1080 		while (q-- > map) {
1081 			if (q[1].hash >= q[0].hash)
1082 				continue;
1083 			swap(*(q+1), *q);
1084 			more = 1;
1085 		}
1086 	} while(more);
1087 }
1088 
1089 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1090 {
1091 	struct dx_entry *entries = frame->entries;
1092 	struct dx_entry *old = frame->at, *new = old + 1;
1093 	int count = dx_get_count(entries);
1094 
1095 	assert(count < dx_get_limit(entries));
1096 	assert(old < entries + count);
1097 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1098 	dx_set_hash(new, hash);
1099 	dx_set_block(new, block);
1100 	dx_set_count(entries, count + 1);
1101 }
1102 
1103 /*
1104  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1105  *
1106  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1107  * `de != NULL' is guaranteed by caller.
1108  */
1109 static inline int ext4_match (int len, const char * const name,
1110 			      struct ext4_dir_entry_2 * de)
1111 {
1112 	if (len != de->name_len)
1113 		return 0;
1114 	if (!de->inode)
1115 		return 0;
1116 	return !memcmp(name, de->name, len);
1117 }
1118 
1119 /*
1120  * Returns 0 if not found, -1 on failure, and 1 on success
1121  */
1122 int search_dir(struct buffer_head *bh,
1123 	       char *search_buf,
1124 	       int buf_size,
1125 	       struct inode *dir,
1126 	       const struct qstr *d_name,
1127 	       unsigned int offset,
1128 	       struct ext4_dir_entry_2 **res_dir)
1129 {
1130 	struct ext4_dir_entry_2 * de;
1131 	char * dlimit;
1132 	int de_len;
1133 	const char *name = d_name->name;
1134 	int namelen = d_name->len;
1135 
1136 	de = (struct ext4_dir_entry_2 *)search_buf;
1137 	dlimit = search_buf + buf_size;
1138 	while ((char *) de < dlimit) {
1139 		/* this code is executed quadratically often */
1140 		/* do minimal checking `by hand' */
1141 
1142 		if ((char *) de + namelen <= dlimit &&
1143 		    ext4_match (namelen, name, de)) {
1144 			/* found a match - just to be sure, do a full check */
1145 			if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1146 						 bh->b_size, offset))
1147 				return -1;
1148 			*res_dir = de;
1149 			return 1;
1150 		}
1151 		/* prevent looping on a bad block */
1152 		de_len = ext4_rec_len_from_disk(de->rec_len,
1153 						dir->i_sb->s_blocksize);
1154 		if (de_len <= 0)
1155 			return -1;
1156 		offset += de_len;
1157 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1158 	}
1159 	return 0;
1160 }
1161 
1162 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1163 			       struct ext4_dir_entry *de)
1164 {
1165 	struct super_block *sb = dir->i_sb;
1166 
1167 	if (!is_dx(dir))
1168 		return 0;
1169 	if (block == 0)
1170 		return 1;
1171 	if (de->inode == 0 &&
1172 	    ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1173 			sb->s_blocksize)
1174 		return 1;
1175 	return 0;
1176 }
1177 
1178 /*
1179  *	ext4_find_entry()
1180  *
1181  * finds an entry in the specified directory with the wanted name. It
1182  * returns the cache buffer in which the entry was found, and the entry
1183  * itself (as a parameter - res_dir). It does NOT read the inode of the
1184  * entry - you'll have to do that yourself if you want to.
1185  *
1186  * The returned buffer_head has ->b_count elevated.  The caller is expected
1187  * to brelse() it when appropriate.
1188  */
1189 static struct buffer_head * ext4_find_entry (struct inode *dir,
1190 					const struct qstr *d_name,
1191 					struct ext4_dir_entry_2 **res_dir,
1192 					int *inlined)
1193 {
1194 	struct super_block *sb;
1195 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1196 	struct buffer_head *bh, *ret = NULL;
1197 	ext4_lblk_t start, block, b;
1198 	const u8 *name = d_name->name;
1199 	int ra_max = 0;		/* Number of bh's in the readahead
1200 				   buffer, bh_use[] */
1201 	int ra_ptr = 0;		/* Current index into readahead
1202 				   buffer */
1203 	int num = 0;
1204 	ext4_lblk_t  nblocks;
1205 	int i, namelen;
1206 
1207 	*res_dir = NULL;
1208 	sb = dir->i_sb;
1209 	namelen = d_name->len;
1210 	if (namelen > EXT4_NAME_LEN)
1211 		return NULL;
1212 
1213 	if (ext4_has_inline_data(dir)) {
1214 		int has_inline_data = 1;
1215 		ret = ext4_find_inline_entry(dir, d_name, res_dir,
1216 					     &has_inline_data);
1217 		if (has_inline_data) {
1218 			if (inlined)
1219 				*inlined = 1;
1220 			return ret;
1221 		}
1222 	}
1223 
1224 	if ((namelen <= 2) && (name[0] == '.') &&
1225 	    (name[1] == '.' || name[1] == '\0')) {
1226 		/*
1227 		 * "." or ".." will only be in the first block
1228 		 * NFS may look up ".."; "." should be handled by the VFS
1229 		 */
1230 		block = start = 0;
1231 		nblocks = 1;
1232 		goto restart;
1233 	}
1234 	if (is_dx(dir)) {
1235 		bh = ext4_dx_find_entry(dir, d_name, res_dir);
1236 		/*
1237 		 * On success, or if the error was file not found,
1238 		 * return.  Otherwise, fall back to doing a search the
1239 		 * old fashioned way.
1240 		 */
1241 		if (!IS_ERR(bh) || PTR_ERR(bh) != ERR_BAD_DX_DIR)
1242 			return bh;
1243 		dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1244 			       "falling back\n"));
1245 	}
1246 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1247 	start = EXT4_I(dir)->i_dir_start_lookup;
1248 	if (start >= nblocks)
1249 		start = 0;
1250 	block = start;
1251 restart:
1252 	do {
1253 		/*
1254 		 * We deal with the read-ahead logic here.
1255 		 */
1256 		if (ra_ptr >= ra_max) {
1257 			/* Refill the readahead buffer */
1258 			ra_ptr = 0;
1259 			b = block;
1260 			for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1261 				/*
1262 				 * Terminate if we reach the end of the
1263 				 * directory and must wrap, or if our
1264 				 * search has finished at this block.
1265 				 */
1266 				if (b >= nblocks || (num && block == start)) {
1267 					bh_use[ra_max] = NULL;
1268 					break;
1269 				}
1270 				num++;
1271 				bh = ext4_getblk(NULL, dir, b++, 0);
1272 				if (unlikely(IS_ERR(bh))) {
1273 					if (ra_max == 0)
1274 						return bh;
1275 					break;
1276 				}
1277 				bh_use[ra_max] = bh;
1278 				if (bh)
1279 					ll_rw_block(READ | REQ_META | REQ_PRIO,
1280 						    1, &bh);
1281 			}
1282 		}
1283 		if ((bh = bh_use[ra_ptr++]) == NULL)
1284 			goto next;
1285 		wait_on_buffer(bh);
1286 		if (!buffer_uptodate(bh)) {
1287 			/* read error, skip block & hope for the best */
1288 			EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1289 					 (unsigned long) block);
1290 			brelse(bh);
1291 			goto next;
1292 		}
1293 		if (!buffer_verified(bh) &&
1294 		    !is_dx_internal_node(dir, block,
1295 					 (struct ext4_dir_entry *)bh->b_data) &&
1296 		    !ext4_dirent_csum_verify(dir,
1297 				(struct ext4_dir_entry *)bh->b_data)) {
1298 			EXT4_ERROR_INODE(dir, "checksumming directory "
1299 					 "block %lu", (unsigned long)block);
1300 			brelse(bh);
1301 			goto next;
1302 		}
1303 		set_buffer_verified(bh);
1304 		i = search_dirblock(bh, dir, d_name,
1305 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1306 		if (i == 1) {
1307 			EXT4_I(dir)->i_dir_start_lookup = block;
1308 			ret = bh;
1309 			goto cleanup_and_exit;
1310 		} else {
1311 			brelse(bh);
1312 			if (i < 0)
1313 				goto cleanup_and_exit;
1314 		}
1315 	next:
1316 		if (++block >= nblocks)
1317 			block = 0;
1318 	} while (block != start);
1319 
1320 	/*
1321 	 * If the directory has grown while we were searching, then
1322 	 * search the last part of the directory before giving up.
1323 	 */
1324 	block = nblocks;
1325 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1326 	if (block < nblocks) {
1327 		start = 0;
1328 		goto restart;
1329 	}
1330 
1331 cleanup_and_exit:
1332 	/* Clean up the read-ahead blocks */
1333 	for (; ra_ptr < ra_max; ra_ptr++)
1334 		brelse(bh_use[ra_ptr]);
1335 	return ret;
1336 }
1337 
1338 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1339 		       struct ext4_dir_entry_2 **res_dir)
1340 {
1341 	struct super_block * sb = dir->i_sb;
1342 	struct dx_hash_info	hinfo;
1343 	struct dx_frame frames[2], *frame;
1344 	struct buffer_head *bh;
1345 	ext4_lblk_t block;
1346 	int retval;
1347 
1348 	frame = dx_probe(d_name, dir, &hinfo, frames);
1349 	if (IS_ERR(frame))
1350 		return (struct buffer_head *) frame;
1351 	do {
1352 		block = dx_get_block(frame->at);
1353 		bh = ext4_read_dirblock(dir, block, DIRENT);
1354 		if (IS_ERR(bh))
1355 			goto errout;
1356 
1357 		retval = search_dirblock(bh, dir, d_name,
1358 					 block << EXT4_BLOCK_SIZE_BITS(sb),
1359 					 res_dir);
1360 		if (retval == 1)
1361 			goto success;
1362 		brelse(bh);
1363 		if (retval == -1) {
1364 			bh = ERR_PTR(ERR_BAD_DX_DIR);
1365 			goto errout;
1366 		}
1367 
1368 		/* Check to see if we should continue to search */
1369 		retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1370 					       frames, NULL);
1371 		if (retval < 0) {
1372 			ext4_warning(sb,
1373 			     "error %d reading index page in directory #%lu",
1374 			     retval, dir->i_ino);
1375 			bh = ERR_PTR(retval);
1376 			goto errout;
1377 		}
1378 	} while (retval == 1);
1379 
1380 	bh = NULL;
1381 errout:
1382 	dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1383 success:
1384 	dx_release(frames);
1385 	return bh;
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 	if (IS_ERR(bh))
1399 		return (struct dentry *) bh;
1400 	inode = NULL;
1401 	if (bh) {
1402 		__u32 ino = le32_to_cpu(de->inode);
1403 		brelse(bh);
1404 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1405 			EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1406 			return ERR_PTR(-EIO);
1407 		}
1408 		if (unlikely(ino == dir->i_ino)) {
1409 			EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1410 					 dentry);
1411 			return ERR_PTR(-EIO);
1412 		}
1413 		inode = ext4_iget_normal(dir->i_sb, ino);
1414 		if (inode == ERR_PTR(-ESTALE)) {
1415 			EXT4_ERROR_INODE(dir,
1416 					 "deleted inode referenced: %u",
1417 					 ino);
1418 			return ERR_PTR(-EIO);
1419 		}
1420 	}
1421 	return d_splice_alias(inode, dentry);
1422 }
1423 
1424 
1425 struct dentry *ext4_get_parent(struct dentry *child)
1426 {
1427 	__u32 ino;
1428 	static const struct qstr dotdot = QSTR_INIT("..", 2);
1429 	struct ext4_dir_entry_2 * de;
1430 	struct buffer_head *bh;
1431 
1432 	bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1433 	if (IS_ERR(bh))
1434 		return (struct dentry *) bh;
1435 	if (!bh)
1436 		return ERR_PTR(-ENOENT);
1437 	ino = le32_to_cpu(de->inode);
1438 	brelse(bh);
1439 
1440 	if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1441 		EXT4_ERROR_INODE(child->d_inode,
1442 				 "bad parent inode number: %u", ino);
1443 		return ERR_PTR(-EIO);
1444 	}
1445 
1446 	return d_obtain_alias(ext4_iget_normal(child->d_inode->i_sb, ino));
1447 }
1448 
1449 /*
1450  * Move count entries from end of map between two memory locations.
1451  * Returns pointer to last entry moved.
1452  */
1453 static struct ext4_dir_entry_2 *
1454 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1455 		unsigned blocksize)
1456 {
1457 	unsigned rec_len = 0;
1458 
1459 	while (count--) {
1460 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1461 						(from + (map->offs<<2));
1462 		rec_len = EXT4_DIR_REC_LEN(de->name_len);
1463 		memcpy (to, de, rec_len);
1464 		((struct ext4_dir_entry_2 *) to)->rec_len =
1465 				ext4_rec_len_to_disk(rec_len, blocksize);
1466 		de->inode = 0;
1467 		map++;
1468 		to += rec_len;
1469 	}
1470 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1471 }
1472 
1473 /*
1474  * Compact each dir entry in the range to the minimal rec_len.
1475  * Returns pointer to last entry in range.
1476  */
1477 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1478 {
1479 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1480 	unsigned rec_len = 0;
1481 
1482 	prev = to = de;
1483 	while ((char*)de < base + blocksize) {
1484 		next = ext4_next_entry(de, blocksize);
1485 		if (de->inode && de->name_len) {
1486 			rec_len = EXT4_DIR_REC_LEN(de->name_len);
1487 			if (de > to)
1488 				memmove(to, de, rec_len);
1489 			to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1490 			prev = to;
1491 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1492 		}
1493 		de = next;
1494 	}
1495 	return prev;
1496 }
1497 
1498 /*
1499  * Split a full leaf block to make room for a new dir entry.
1500  * Allocate a new block, and move entries so that they are approx. equally full.
1501  * Returns pointer to de in block into which the new entry will be inserted.
1502  */
1503 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1504 			struct buffer_head **bh,struct dx_frame *frame,
1505 			struct dx_hash_info *hinfo)
1506 {
1507 	unsigned blocksize = dir->i_sb->s_blocksize;
1508 	unsigned count, continued;
1509 	struct buffer_head *bh2;
1510 	ext4_lblk_t newblock;
1511 	u32 hash2;
1512 	struct dx_map_entry *map;
1513 	char *data1 = (*bh)->b_data, *data2;
1514 	unsigned split, move, size;
1515 	struct ext4_dir_entry_2 *de = NULL, *de2;
1516 	struct ext4_dir_entry_tail *t;
1517 	int	csum_size = 0;
1518 	int	err = 0, i;
1519 
1520 	if (ext4_has_metadata_csum(dir->i_sb))
1521 		csum_size = sizeof(struct ext4_dir_entry_tail);
1522 
1523 	bh2 = ext4_append(handle, dir, &newblock);
1524 	if (IS_ERR(bh2)) {
1525 		brelse(*bh);
1526 		*bh = NULL;
1527 		return (struct ext4_dir_entry_2 *) bh2;
1528 	}
1529 
1530 	BUFFER_TRACE(*bh, "get_write_access");
1531 	err = ext4_journal_get_write_access(handle, *bh);
1532 	if (err)
1533 		goto journal_error;
1534 
1535 	BUFFER_TRACE(frame->bh, "get_write_access");
1536 	err = ext4_journal_get_write_access(handle, frame->bh);
1537 	if (err)
1538 		goto journal_error;
1539 
1540 	data2 = bh2->b_data;
1541 
1542 	/* create map in the end of data2 block */
1543 	map = (struct dx_map_entry *) (data2 + blocksize);
1544 	count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1545 			     blocksize, hinfo, map);
1546 	map -= count;
1547 	dx_sort_map(map, count);
1548 	/* Split the existing block in the middle, size-wise */
1549 	size = 0;
1550 	move = 0;
1551 	for (i = count-1; i >= 0; i--) {
1552 		/* is more than half of this entry in 2nd half of the block? */
1553 		if (size + map[i].size/2 > blocksize/2)
1554 			break;
1555 		size += map[i].size;
1556 		move++;
1557 	}
1558 	/* map index at which we will split */
1559 	split = count - move;
1560 	hash2 = map[split].hash;
1561 	continued = hash2 == map[split - 1].hash;
1562 	dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1563 			(unsigned long)dx_get_block(frame->at),
1564 					hash2, split, count-split));
1565 
1566 	/* Fancy dance to stay within two buffers */
1567 	de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1568 	de = dx_pack_dirents(data1, blocksize);
1569 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1570 					   (char *) de,
1571 					   blocksize);
1572 	de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1573 					    (char *) de2,
1574 					    blocksize);
1575 	if (csum_size) {
1576 		t = EXT4_DIRENT_TAIL(data2, blocksize);
1577 		initialize_dirent_tail(t, blocksize);
1578 
1579 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1580 		initialize_dirent_tail(t, blocksize);
1581 	}
1582 
1583 	dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1584 	dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1585 
1586 	/* Which block gets the new entry? */
1587 	if (hinfo->hash >= hash2) {
1588 		swap(*bh, bh2);
1589 		de = de2;
1590 	}
1591 	dx_insert_block(frame, hash2 + continued, newblock);
1592 	err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1593 	if (err)
1594 		goto journal_error;
1595 	err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1596 	if (err)
1597 		goto journal_error;
1598 	brelse(bh2);
1599 	dxtrace(dx_show_index("frame", frame->entries));
1600 	return de;
1601 
1602 journal_error:
1603 	brelse(*bh);
1604 	brelse(bh2);
1605 	*bh = NULL;
1606 	ext4_std_error(dir->i_sb, err);
1607 	return ERR_PTR(err);
1608 }
1609 
1610 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1611 		      struct buffer_head *bh,
1612 		      void *buf, int buf_size,
1613 		      const char *name, int namelen,
1614 		      struct ext4_dir_entry_2 **dest_de)
1615 {
1616 	struct ext4_dir_entry_2 *de;
1617 	unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1618 	int nlen, rlen;
1619 	unsigned int offset = 0;
1620 	char *top;
1621 
1622 	de = (struct ext4_dir_entry_2 *)buf;
1623 	top = buf + buf_size - reclen;
1624 	while ((char *) de <= top) {
1625 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1626 					 buf, buf_size, offset))
1627 			return -EIO;
1628 		if (ext4_match(namelen, name, de))
1629 			return -EEXIST;
1630 		nlen = EXT4_DIR_REC_LEN(de->name_len);
1631 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1632 		if ((de->inode ? rlen - nlen : rlen) >= reclen)
1633 			break;
1634 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1635 		offset += rlen;
1636 	}
1637 	if ((char *) de > top)
1638 		return -ENOSPC;
1639 
1640 	*dest_de = de;
1641 	return 0;
1642 }
1643 
1644 void ext4_insert_dentry(struct inode *inode,
1645 			struct ext4_dir_entry_2 *de,
1646 			int buf_size,
1647 			const char *name, int namelen)
1648 {
1649 
1650 	int nlen, rlen;
1651 
1652 	nlen = EXT4_DIR_REC_LEN(de->name_len);
1653 	rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1654 	if (de->inode) {
1655 		struct ext4_dir_entry_2 *de1 =
1656 				(struct ext4_dir_entry_2 *)((char *)de + nlen);
1657 		de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1658 		de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1659 		de = de1;
1660 	}
1661 	de->file_type = EXT4_FT_UNKNOWN;
1662 	de->inode = cpu_to_le32(inode->i_ino);
1663 	ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1664 	de->name_len = namelen;
1665 	memcpy(de->name, name, namelen);
1666 }
1667 /*
1668  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1669  * it points to a directory entry which is guaranteed to be large
1670  * enough for new directory entry.  If de is NULL, then
1671  * add_dirent_to_buf will attempt search the directory block for
1672  * space.  It will return -ENOSPC if no space is available, and -EIO
1673  * and -EEXIST if directory entry already exists.
1674  */
1675 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1676 			     struct inode *inode, struct ext4_dir_entry_2 *de,
1677 			     struct buffer_head *bh)
1678 {
1679 	struct inode	*dir = dentry->d_parent->d_inode;
1680 	const char	*name = dentry->d_name.name;
1681 	int		namelen = dentry->d_name.len;
1682 	unsigned int	blocksize = dir->i_sb->s_blocksize;
1683 	int		csum_size = 0;
1684 	int		err;
1685 
1686 	if (ext4_has_metadata_csum(inode->i_sb))
1687 		csum_size = sizeof(struct ext4_dir_entry_tail);
1688 
1689 	if (!de) {
1690 		err = ext4_find_dest_de(dir, inode,
1691 					bh, bh->b_data, blocksize - csum_size,
1692 					name, namelen, &de);
1693 		if (err)
1694 			return err;
1695 	}
1696 	BUFFER_TRACE(bh, "get_write_access");
1697 	err = ext4_journal_get_write_access(handle, bh);
1698 	if (err) {
1699 		ext4_std_error(dir->i_sb, err);
1700 		return err;
1701 	}
1702 
1703 	/* By now the buffer is marked for journaling */
1704 	ext4_insert_dentry(inode, de, blocksize, name, namelen);
1705 
1706 	/*
1707 	 * XXX shouldn't update any times until successful
1708 	 * completion of syscall, but too many callers depend
1709 	 * on this.
1710 	 *
1711 	 * XXX similarly, too many callers depend on
1712 	 * ext4_new_inode() setting the times, but error
1713 	 * recovery deletes the inode, so the worst that can
1714 	 * happen is that the times are slightly out of date
1715 	 * and/or different from the directory change time.
1716 	 */
1717 	dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1718 	ext4_update_dx_flag(dir);
1719 	dir->i_version++;
1720 	ext4_mark_inode_dirty(handle, dir);
1721 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1722 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1723 	if (err)
1724 		ext4_std_error(dir->i_sb, err);
1725 	return 0;
1726 }
1727 
1728 /*
1729  * This converts a one block unindexed directory to a 3 block indexed
1730  * directory, and adds the dentry to the indexed directory.
1731  */
1732 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1733 			    struct inode *inode, struct buffer_head *bh)
1734 {
1735 	struct inode	*dir = dentry->d_parent->d_inode;
1736 	const char	*name = dentry->d_name.name;
1737 	int		namelen = dentry->d_name.len;
1738 	struct buffer_head *bh2;
1739 	struct dx_root	*root;
1740 	struct dx_frame	frames[2], *frame;
1741 	struct dx_entry *entries;
1742 	struct ext4_dir_entry_2	*de, *de2;
1743 	struct ext4_dir_entry_tail *t;
1744 	char		*data1, *top;
1745 	unsigned	len;
1746 	int		retval;
1747 	unsigned	blocksize;
1748 	struct dx_hash_info hinfo;
1749 	ext4_lblk_t  block;
1750 	struct fake_dirent *fde;
1751 	int		csum_size = 0;
1752 
1753 	if (ext4_has_metadata_csum(inode->i_sb))
1754 		csum_size = sizeof(struct ext4_dir_entry_tail);
1755 
1756 	blocksize =  dir->i_sb->s_blocksize;
1757 	dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1758 	BUFFER_TRACE(bh, "get_write_access");
1759 	retval = ext4_journal_get_write_access(handle, bh);
1760 	if (retval) {
1761 		ext4_std_error(dir->i_sb, retval);
1762 		brelse(bh);
1763 		return retval;
1764 	}
1765 	root = (struct dx_root *) bh->b_data;
1766 
1767 	/* The 0th block becomes the root, move the dirents out */
1768 	fde = &root->dotdot;
1769 	de = (struct ext4_dir_entry_2 *)((char *)fde +
1770 		ext4_rec_len_from_disk(fde->rec_len, blocksize));
1771 	if ((char *) de >= (((char *) root) + blocksize)) {
1772 		EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1773 		brelse(bh);
1774 		return -EIO;
1775 	}
1776 	len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1777 
1778 	/* Allocate new block for the 0th block's dirents */
1779 	bh2 = ext4_append(handle, dir, &block);
1780 	if (IS_ERR(bh2)) {
1781 		brelse(bh);
1782 		return PTR_ERR(bh2);
1783 	}
1784 	ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1785 	data1 = bh2->b_data;
1786 
1787 	memcpy (data1, de, len);
1788 	de = (struct ext4_dir_entry_2 *) data1;
1789 	top = data1 + len;
1790 	while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1791 		de = de2;
1792 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1793 					   (char *) de,
1794 					   blocksize);
1795 
1796 	if (csum_size) {
1797 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1798 		initialize_dirent_tail(t, blocksize);
1799 	}
1800 
1801 	/* Initialize the root; the dot dirents already exist */
1802 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1803 	de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1804 					   blocksize);
1805 	memset (&root->info, 0, sizeof(root->info));
1806 	root->info.info_length = sizeof(root->info);
1807 	root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1808 	entries = root->entries;
1809 	dx_set_block(entries, 1);
1810 	dx_set_count(entries, 1);
1811 	dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1812 
1813 	/* Initialize as for dx_probe */
1814 	hinfo.hash_version = root->info.hash_version;
1815 	if (hinfo.hash_version <= DX_HASH_TEA)
1816 		hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1817 	hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1818 	ext4fs_dirhash(name, namelen, &hinfo);
1819 	memset(frames, 0, sizeof(frames));
1820 	frame = frames;
1821 	frame->entries = entries;
1822 	frame->at = entries;
1823 	frame->bh = bh;
1824 	bh = bh2;
1825 
1826 	retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1827 	if (retval)
1828 		goto out_frames;
1829 	retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
1830 	if (retval)
1831 		goto out_frames;
1832 
1833 	de = do_split(handle,dir, &bh, frame, &hinfo);
1834 	if (IS_ERR(de)) {
1835 		retval = PTR_ERR(de);
1836 		goto out_frames;
1837 	}
1838 	dx_release(frames);
1839 
1840 	retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1841 	brelse(bh);
1842 	return retval;
1843 out_frames:
1844 	/*
1845 	 * Even if the block split failed, we have to properly write
1846 	 * out all the changes we did so far. Otherwise we can end up
1847 	 * with corrupted filesystem.
1848 	 */
1849 	ext4_mark_inode_dirty(handle, dir);
1850 	dx_release(frames);
1851 	return retval;
1852 }
1853 
1854 /*
1855  *	ext4_add_entry()
1856  *
1857  * adds a file entry to the specified directory, using the same
1858  * semantics as ext4_find_entry(). It returns NULL if it failed.
1859  *
1860  * NOTE!! The inode part of 'de' is left at 0 - which means you
1861  * may not sleep between calling this and putting something into
1862  * the entry, as someone else might have used it while you slept.
1863  */
1864 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1865 			  struct inode *inode)
1866 {
1867 	struct inode *dir = dentry->d_parent->d_inode;
1868 	struct buffer_head *bh;
1869 	struct ext4_dir_entry_2 *de;
1870 	struct ext4_dir_entry_tail *t;
1871 	struct super_block *sb;
1872 	int	retval;
1873 	int	dx_fallback=0;
1874 	unsigned blocksize;
1875 	ext4_lblk_t block, blocks;
1876 	int	csum_size = 0;
1877 
1878 	if (ext4_has_metadata_csum(inode->i_sb))
1879 		csum_size = sizeof(struct ext4_dir_entry_tail);
1880 
1881 	sb = dir->i_sb;
1882 	blocksize = sb->s_blocksize;
1883 	if (!dentry->d_name.len)
1884 		return -EINVAL;
1885 
1886 	if (ext4_has_inline_data(dir)) {
1887 		retval = ext4_try_add_inline_entry(handle, dentry, inode);
1888 		if (retval < 0)
1889 			return retval;
1890 		if (retval == 1) {
1891 			retval = 0;
1892 			return retval;
1893 		}
1894 	}
1895 
1896 	if (is_dx(dir)) {
1897 		retval = ext4_dx_add_entry(handle, dentry, inode);
1898 		if (!retval || (retval != ERR_BAD_DX_DIR))
1899 			return retval;
1900 		ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1901 		dx_fallback++;
1902 		ext4_mark_inode_dirty(handle, dir);
1903 	}
1904 	blocks = dir->i_size >> sb->s_blocksize_bits;
1905 	for (block = 0; block < blocks; block++) {
1906 		bh = ext4_read_dirblock(dir, block, DIRENT);
1907 		if (IS_ERR(bh))
1908 			return PTR_ERR(bh);
1909 
1910 		retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1911 		if (retval != -ENOSPC) {
1912 			brelse(bh);
1913 			return retval;
1914 		}
1915 
1916 		if (blocks == 1 && !dx_fallback &&
1917 		    EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1918 			return make_indexed_dir(handle, dentry, inode, bh);
1919 		brelse(bh);
1920 	}
1921 	bh = ext4_append(handle, dir, &block);
1922 	if (IS_ERR(bh))
1923 		return PTR_ERR(bh);
1924 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1925 	de->inode = 0;
1926 	de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1927 
1928 	if (csum_size) {
1929 		t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1930 		initialize_dirent_tail(t, blocksize);
1931 	}
1932 
1933 	retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1934 	brelse(bh);
1935 	if (retval == 0)
1936 		ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1937 	return retval;
1938 }
1939 
1940 /*
1941  * Returns 0 for success, or a negative error value
1942  */
1943 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1944 			     struct inode *inode)
1945 {
1946 	struct dx_frame frames[2], *frame;
1947 	struct dx_entry *entries, *at;
1948 	struct dx_hash_info hinfo;
1949 	struct buffer_head *bh;
1950 	struct inode *dir = dentry->d_parent->d_inode;
1951 	struct super_block *sb = dir->i_sb;
1952 	struct ext4_dir_entry_2 *de;
1953 	int err;
1954 
1955 	frame = dx_probe(&dentry->d_name, dir, &hinfo, frames);
1956 	if (IS_ERR(frame))
1957 		return PTR_ERR(frame);
1958 	entries = frame->entries;
1959 	at = frame->at;
1960 	bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
1961 	if (IS_ERR(bh)) {
1962 		err = PTR_ERR(bh);
1963 		bh = NULL;
1964 		goto cleanup;
1965 	}
1966 
1967 	BUFFER_TRACE(bh, "get_write_access");
1968 	err = ext4_journal_get_write_access(handle, bh);
1969 	if (err)
1970 		goto journal_error;
1971 
1972 	err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1973 	if (err != -ENOSPC)
1974 		goto cleanup;
1975 
1976 	/* Block full, should compress but for now just split */
1977 	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1978 		       dx_get_count(entries), dx_get_limit(entries)));
1979 	/* Need to split index? */
1980 	if (dx_get_count(entries) == dx_get_limit(entries)) {
1981 		ext4_lblk_t newblock;
1982 		unsigned icount = dx_get_count(entries);
1983 		int levels = frame - frames;
1984 		struct dx_entry *entries2;
1985 		struct dx_node *node2;
1986 		struct buffer_head *bh2;
1987 
1988 		if (levels && (dx_get_count(frames->entries) ==
1989 			       dx_get_limit(frames->entries))) {
1990 			ext4_warning(sb, "Directory index full!");
1991 			err = -ENOSPC;
1992 			goto cleanup;
1993 		}
1994 		bh2 = ext4_append(handle, dir, &newblock);
1995 		if (IS_ERR(bh2)) {
1996 			err = PTR_ERR(bh2);
1997 			goto cleanup;
1998 		}
1999 		node2 = (struct dx_node *)(bh2->b_data);
2000 		entries2 = node2->entries;
2001 		memset(&node2->fake, 0, sizeof(struct fake_dirent));
2002 		node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2003 							   sb->s_blocksize);
2004 		BUFFER_TRACE(frame->bh, "get_write_access");
2005 		err = ext4_journal_get_write_access(handle, frame->bh);
2006 		if (err)
2007 			goto journal_error;
2008 		if (levels) {
2009 			unsigned icount1 = icount/2, icount2 = icount - icount1;
2010 			unsigned hash2 = dx_get_hash(entries + icount1);
2011 			dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2012 				       icount1, icount2));
2013 
2014 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2015 			err = ext4_journal_get_write_access(handle,
2016 							     frames[0].bh);
2017 			if (err)
2018 				goto journal_error;
2019 
2020 			memcpy((char *) entries2, (char *) (entries + icount1),
2021 			       icount2 * sizeof(struct dx_entry));
2022 			dx_set_count(entries, icount1);
2023 			dx_set_count(entries2, icount2);
2024 			dx_set_limit(entries2, dx_node_limit(dir));
2025 
2026 			/* Which index block gets the new entry? */
2027 			if (at - entries >= icount1) {
2028 				frame->at = at = at - entries - icount1 + entries2;
2029 				frame->entries = entries = entries2;
2030 				swap(frame->bh, bh2);
2031 			}
2032 			dx_insert_block(frames + 0, hash2, newblock);
2033 			dxtrace(dx_show_index("node", frames[1].entries));
2034 			dxtrace(dx_show_index("node",
2035 			       ((struct dx_node *) bh2->b_data)->entries));
2036 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2037 			if (err)
2038 				goto journal_error;
2039 			brelse (bh2);
2040 		} else {
2041 			dxtrace(printk(KERN_DEBUG
2042 				       "Creating second level index...\n"));
2043 			memcpy((char *) entries2, (char *) entries,
2044 			       icount * sizeof(struct dx_entry));
2045 			dx_set_limit(entries2, dx_node_limit(dir));
2046 
2047 			/* Set up root */
2048 			dx_set_count(entries, 1);
2049 			dx_set_block(entries + 0, newblock);
2050 			((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2051 
2052 			/* Add new access path frame */
2053 			frame = frames + 1;
2054 			frame->at = at = at - entries + entries2;
2055 			frame->entries = entries = entries2;
2056 			frame->bh = bh2;
2057 			err = ext4_journal_get_write_access(handle,
2058 							     frame->bh);
2059 			if (err)
2060 				goto journal_error;
2061 		}
2062 		err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2063 		if (err) {
2064 			ext4_std_error(inode->i_sb, err);
2065 			goto cleanup;
2066 		}
2067 	}
2068 	de = do_split(handle, dir, &bh, frame, &hinfo);
2069 	if (IS_ERR(de)) {
2070 		err = PTR_ERR(de);
2071 		goto cleanup;
2072 	}
2073 	err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2074 	goto cleanup;
2075 
2076 journal_error:
2077 	ext4_std_error(dir->i_sb, err);
2078 cleanup:
2079 	brelse(bh);
2080 	dx_release(frames);
2081 	return err;
2082 }
2083 
2084 /*
2085  * ext4_generic_delete_entry deletes a directory entry by merging it
2086  * with the previous entry
2087  */
2088 int ext4_generic_delete_entry(handle_t *handle,
2089 			      struct inode *dir,
2090 			      struct ext4_dir_entry_2 *de_del,
2091 			      struct buffer_head *bh,
2092 			      void *entry_buf,
2093 			      int buf_size,
2094 			      int csum_size)
2095 {
2096 	struct ext4_dir_entry_2 *de, *pde;
2097 	unsigned int blocksize = dir->i_sb->s_blocksize;
2098 	int i;
2099 
2100 	i = 0;
2101 	pde = NULL;
2102 	de = (struct ext4_dir_entry_2 *)entry_buf;
2103 	while (i < buf_size - csum_size) {
2104 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2105 					 bh->b_data, bh->b_size, i))
2106 			return -EIO;
2107 		if (de == de_del)  {
2108 			if (pde)
2109 				pde->rec_len = ext4_rec_len_to_disk(
2110 					ext4_rec_len_from_disk(pde->rec_len,
2111 							       blocksize) +
2112 					ext4_rec_len_from_disk(de->rec_len,
2113 							       blocksize),
2114 					blocksize);
2115 			else
2116 				de->inode = 0;
2117 			dir->i_version++;
2118 			return 0;
2119 		}
2120 		i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2121 		pde = de;
2122 		de = ext4_next_entry(de, blocksize);
2123 	}
2124 	return -ENOENT;
2125 }
2126 
2127 static int ext4_delete_entry(handle_t *handle,
2128 			     struct inode *dir,
2129 			     struct ext4_dir_entry_2 *de_del,
2130 			     struct buffer_head *bh)
2131 {
2132 	int err, csum_size = 0;
2133 
2134 	if (ext4_has_inline_data(dir)) {
2135 		int has_inline_data = 1;
2136 		err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2137 					       &has_inline_data);
2138 		if (has_inline_data)
2139 			return err;
2140 	}
2141 
2142 	if (ext4_has_metadata_csum(dir->i_sb))
2143 		csum_size = sizeof(struct ext4_dir_entry_tail);
2144 
2145 	BUFFER_TRACE(bh, "get_write_access");
2146 	err = ext4_journal_get_write_access(handle, bh);
2147 	if (unlikely(err))
2148 		goto out;
2149 
2150 	err = ext4_generic_delete_entry(handle, dir, de_del,
2151 					bh, bh->b_data,
2152 					dir->i_sb->s_blocksize, csum_size);
2153 	if (err)
2154 		goto out;
2155 
2156 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2157 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2158 	if (unlikely(err))
2159 		goto out;
2160 
2161 	return 0;
2162 out:
2163 	if (err != -ENOENT)
2164 		ext4_std_error(dir->i_sb, err);
2165 	return err;
2166 }
2167 
2168 /*
2169  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2170  * since this indicates that nlinks count was previously 1.
2171  */
2172 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2173 {
2174 	inc_nlink(inode);
2175 	if (is_dx(inode) && inode->i_nlink > 1) {
2176 		/* limit is 16-bit i_links_count */
2177 		if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2178 			set_nlink(inode, 1);
2179 			EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2180 					      EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2181 		}
2182 	}
2183 }
2184 
2185 /*
2186  * If a directory had nlink == 1, then we should let it be 1. This indicates
2187  * directory has >EXT4_LINK_MAX subdirs.
2188  */
2189 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2190 {
2191 	if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2192 		drop_nlink(inode);
2193 }
2194 
2195 
2196 static int ext4_add_nondir(handle_t *handle,
2197 		struct dentry *dentry, struct inode *inode)
2198 {
2199 	int err = ext4_add_entry(handle, dentry, inode);
2200 	if (!err) {
2201 		ext4_mark_inode_dirty(handle, inode);
2202 		unlock_new_inode(inode);
2203 		d_instantiate(dentry, inode);
2204 		return 0;
2205 	}
2206 	drop_nlink(inode);
2207 	unlock_new_inode(inode);
2208 	iput(inode);
2209 	return err;
2210 }
2211 
2212 /*
2213  * By the time this is called, we already have created
2214  * the directory cache entry for the new file, but it
2215  * is so far negative - it has no inode.
2216  *
2217  * If the create succeeds, we fill in the inode information
2218  * with d_instantiate().
2219  */
2220 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2221 		       bool excl)
2222 {
2223 	handle_t *handle;
2224 	struct inode *inode;
2225 	int err, credits, retries = 0;
2226 
2227 	dquot_initialize(dir);
2228 
2229 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2230 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2231 retry:
2232 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2233 					    NULL, EXT4_HT_DIR, credits);
2234 	handle = ext4_journal_current_handle();
2235 	err = PTR_ERR(inode);
2236 	if (!IS_ERR(inode)) {
2237 		inode->i_op = &ext4_file_inode_operations;
2238 		if (test_opt(inode->i_sb, DAX))
2239 			inode->i_fop = &ext4_dax_file_operations;
2240 		else
2241 			inode->i_fop = &ext4_file_operations;
2242 		ext4_set_aops(inode);
2243 		err = ext4_add_nondir(handle, dentry, inode);
2244 		if (!err && IS_DIRSYNC(dir))
2245 			ext4_handle_sync(handle);
2246 	}
2247 	if (handle)
2248 		ext4_journal_stop(handle);
2249 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2250 		goto retry;
2251 	return err;
2252 }
2253 
2254 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2255 		      umode_t mode, dev_t rdev)
2256 {
2257 	handle_t *handle;
2258 	struct inode *inode;
2259 	int err, credits, retries = 0;
2260 
2261 	if (!new_valid_dev(rdev))
2262 		return -EINVAL;
2263 
2264 	dquot_initialize(dir);
2265 
2266 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2267 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2268 retry:
2269 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2270 					    NULL, EXT4_HT_DIR, credits);
2271 	handle = ext4_journal_current_handle();
2272 	err = PTR_ERR(inode);
2273 	if (!IS_ERR(inode)) {
2274 		init_special_inode(inode, inode->i_mode, rdev);
2275 		inode->i_op = &ext4_special_inode_operations;
2276 		err = ext4_add_nondir(handle, dentry, inode);
2277 		if (!err && IS_DIRSYNC(dir))
2278 			ext4_handle_sync(handle);
2279 	}
2280 	if (handle)
2281 		ext4_journal_stop(handle);
2282 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2283 		goto retry;
2284 	return err;
2285 }
2286 
2287 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2288 {
2289 	handle_t *handle;
2290 	struct inode *inode;
2291 	int err, retries = 0;
2292 
2293 	dquot_initialize(dir);
2294 
2295 retry:
2296 	inode = ext4_new_inode_start_handle(dir, mode,
2297 					    NULL, 0, NULL,
2298 					    EXT4_HT_DIR,
2299 			EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2300 			  4 + EXT4_XATTR_TRANS_BLOCKS);
2301 	handle = ext4_journal_current_handle();
2302 	err = PTR_ERR(inode);
2303 	if (!IS_ERR(inode)) {
2304 		inode->i_op = &ext4_file_inode_operations;
2305 		if (test_opt(inode->i_sb, DAX))
2306 			inode->i_fop = &ext4_dax_file_operations;
2307 		else
2308 			inode->i_fop = &ext4_file_operations;
2309 		ext4_set_aops(inode);
2310 		d_tmpfile(dentry, inode);
2311 		err = ext4_orphan_add(handle, inode);
2312 		if (err)
2313 			goto err_unlock_inode;
2314 		mark_inode_dirty(inode);
2315 		unlock_new_inode(inode);
2316 	}
2317 	if (handle)
2318 		ext4_journal_stop(handle);
2319 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2320 		goto retry;
2321 	return err;
2322 err_unlock_inode:
2323 	ext4_journal_stop(handle);
2324 	unlock_new_inode(inode);
2325 	return err;
2326 }
2327 
2328 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2329 			  struct ext4_dir_entry_2 *de,
2330 			  int blocksize, int csum_size,
2331 			  unsigned int parent_ino, int dotdot_real_len)
2332 {
2333 	de->inode = cpu_to_le32(inode->i_ino);
2334 	de->name_len = 1;
2335 	de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2336 					   blocksize);
2337 	strcpy(de->name, ".");
2338 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2339 
2340 	de = ext4_next_entry(de, blocksize);
2341 	de->inode = cpu_to_le32(parent_ino);
2342 	de->name_len = 2;
2343 	if (!dotdot_real_len)
2344 		de->rec_len = ext4_rec_len_to_disk(blocksize -
2345 					(csum_size + EXT4_DIR_REC_LEN(1)),
2346 					blocksize);
2347 	else
2348 		de->rec_len = ext4_rec_len_to_disk(
2349 				EXT4_DIR_REC_LEN(de->name_len), blocksize);
2350 	strcpy(de->name, "..");
2351 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2352 
2353 	return ext4_next_entry(de, blocksize);
2354 }
2355 
2356 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2357 			     struct inode *inode)
2358 {
2359 	struct buffer_head *dir_block = NULL;
2360 	struct ext4_dir_entry_2 *de;
2361 	struct ext4_dir_entry_tail *t;
2362 	ext4_lblk_t block = 0;
2363 	unsigned int blocksize = dir->i_sb->s_blocksize;
2364 	int csum_size = 0;
2365 	int err;
2366 
2367 	if (ext4_has_metadata_csum(dir->i_sb))
2368 		csum_size = sizeof(struct ext4_dir_entry_tail);
2369 
2370 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2371 		err = ext4_try_create_inline_dir(handle, dir, inode);
2372 		if (err < 0 && err != -ENOSPC)
2373 			goto out;
2374 		if (!err)
2375 			goto out;
2376 	}
2377 
2378 	inode->i_size = 0;
2379 	dir_block = ext4_append(handle, inode, &block);
2380 	if (IS_ERR(dir_block))
2381 		return PTR_ERR(dir_block);
2382 	de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2383 	ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2384 	set_nlink(inode, 2);
2385 	if (csum_size) {
2386 		t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2387 		initialize_dirent_tail(t, blocksize);
2388 	}
2389 
2390 	BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2391 	err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2392 	if (err)
2393 		goto out;
2394 	set_buffer_verified(dir_block);
2395 out:
2396 	brelse(dir_block);
2397 	return err;
2398 }
2399 
2400 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2401 {
2402 	handle_t *handle;
2403 	struct inode *inode;
2404 	int err, credits, retries = 0;
2405 
2406 	if (EXT4_DIR_LINK_MAX(dir))
2407 		return -EMLINK;
2408 
2409 	dquot_initialize(dir);
2410 
2411 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2412 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2413 retry:
2414 	inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2415 					    &dentry->d_name,
2416 					    0, NULL, EXT4_HT_DIR, credits);
2417 	handle = ext4_journal_current_handle();
2418 	err = PTR_ERR(inode);
2419 	if (IS_ERR(inode))
2420 		goto out_stop;
2421 
2422 	inode->i_op = &ext4_dir_inode_operations;
2423 	inode->i_fop = &ext4_dir_operations;
2424 	err = ext4_init_new_dir(handle, dir, inode);
2425 	if (err)
2426 		goto out_clear_inode;
2427 	err = ext4_mark_inode_dirty(handle, inode);
2428 	if (!err)
2429 		err = ext4_add_entry(handle, dentry, inode);
2430 	if (err) {
2431 out_clear_inode:
2432 		clear_nlink(inode);
2433 		unlock_new_inode(inode);
2434 		ext4_mark_inode_dirty(handle, inode);
2435 		iput(inode);
2436 		goto out_stop;
2437 	}
2438 	ext4_inc_count(handle, dir);
2439 	ext4_update_dx_flag(dir);
2440 	err = ext4_mark_inode_dirty(handle, dir);
2441 	if (err)
2442 		goto out_clear_inode;
2443 	unlock_new_inode(inode);
2444 	d_instantiate(dentry, inode);
2445 	if (IS_DIRSYNC(dir))
2446 		ext4_handle_sync(handle);
2447 
2448 out_stop:
2449 	if (handle)
2450 		ext4_journal_stop(handle);
2451 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2452 		goto retry;
2453 	return err;
2454 }
2455 
2456 /*
2457  * routine to check that the specified directory is empty (for rmdir)
2458  */
2459 static int empty_dir(struct inode *inode)
2460 {
2461 	unsigned int offset;
2462 	struct buffer_head *bh;
2463 	struct ext4_dir_entry_2 *de, *de1;
2464 	struct super_block *sb;
2465 	int err = 0;
2466 
2467 	if (ext4_has_inline_data(inode)) {
2468 		int has_inline_data = 1;
2469 
2470 		err = empty_inline_dir(inode, &has_inline_data);
2471 		if (has_inline_data)
2472 			return err;
2473 	}
2474 
2475 	sb = inode->i_sb;
2476 	if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2477 		EXT4_ERROR_INODE(inode, "invalid size");
2478 		return 1;
2479 	}
2480 	bh = ext4_read_dirblock(inode, 0, EITHER);
2481 	if (IS_ERR(bh))
2482 		return 1;
2483 
2484 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2485 	de1 = ext4_next_entry(de, sb->s_blocksize);
2486 	if (le32_to_cpu(de->inode) != inode->i_ino ||
2487 			!le32_to_cpu(de1->inode) ||
2488 			strcmp(".", de->name) ||
2489 			strcmp("..", de1->name)) {
2490 		ext4_warning(inode->i_sb,
2491 			     "bad directory (dir #%lu) - no `.' or `..'",
2492 			     inode->i_ino);
2493 		brelse(bh);
2494 		return 1;
2495 	}
2496 	offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2497 		 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2498 	de = ext4_next_entry(de1, sb->s_blocksize);
2499 	while (offset < inode->i_size) {
2500 		if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2501 			unsigned int lblock;
2502 			err = 0;
2503 			brelse(bh);
2504 			lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2505 			bh = ext4_read_dirblock(inode, lblock, EITHER);
2506 			if (IS_ERR(bh))
2507 				return 1;
2508 			de = (struct ext4_dir_entry_2 *) bh->b_data;
2509 		}
2510 		if (ext4_check_dir_entry(inode, NULL, de, bh,
2511 					 bh->b_data, bh->b_size, offset)) {
2512 			de = (struct ext4_dir_entry_2 *)(bh->b_data +
2513 							 sb->s_blocksize);
2514 			offset = (offset | (sb->s_blocksize - 1)) + 1;
2515 			continue;
2516 		}
2517 		if (le32_to_cpu(de->inode)) {
2518 			brelse(bh);
2519 			return 0;
2520 		}
2521 		offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2522 		de = ext4_next_entry(de, sb->s_blocksize);
2523 	}
2524 	brelse(bh);
2525 	return 1;
2526 }
2527 
2528 /*
2529  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2530  * such inodes, starting at the superblock, in case we crash before the
2531  * file is closed/deleted, or in case the inode truncate spans multiple
2532  * transactions and the last transaction is not recovered after a crash.
2533  *
2534  * At filesystem recovery time, we walk this list deleting unlinked
2535  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2536  *
2537  * Orphan list manipulation functions must be called under i_mutex unless
2538  * we are just creating the inode or deleting it.
2539  */
2540 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2541 {
2542 	struct super_block *sb = inode->i_sb;
2543 	struct ext4_sb_info *sbi = EXT4_SB(sb);
2544 	struct ext4_iloc iloc;
2545 	int err = 0, rc;
2546 	bool dirty = false;
2547 
2548 	if (!sbi->s_journal || is_bad_inode(inode))
2549 		return 0;
2550 
2551 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2552 		     !mutex_is_locked(&inode->i_mutex));
2553 	/*
2554 	 * Exit early if inode already is on orphan list. This is a big speedup
2555 	 * since we don't have to contend on the global s_orphan_lock.
2556 	 */
2557 	if (!list_empty(&EXT4_I(inode)->i_orphan))
2558 		return 0;
2559 
2560 	/*
2561 	 * Orphan handling is only valid for files with data blocks
2562 	 * being truncated, or files being unlinked. Note that we either
2563 	 * hold i_mutex, or the inode can not be referenced from outside,
2564 	 * so i_nlink should not be bumped due to race
2565 	 */
2566 	J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2567 		  S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2568 
2569 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2570 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2571 	if (err)
2572 		goto out;
2573 
2574 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2575 	if (err)
2576 		goto out;
2577 
2578 	mutex_lock(&sbi->s_orphan_lock);
2579 	/*
2580 	 * Due to previous errors inode may be already a part of on-disk
2581 	 * orphan list. If so skip on-disk list modification.
2582 	 */
2583 	if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2584 	    (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2585 		/* Insert this inode at the head of the on-disk orphan list */
2586 		NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2587 		sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2588 		dirty = true;
2589 	}
2590 	list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2591 	mutex_unlock(&sbi->s_orphan_lock);
2592 
2593 	if (dirty) {
2594 		err = ext4_handle_dirty_super(handle, sb);
2595 		rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2596 		if (!err)
2597 			err = rc;
2598 		if (err) {
2599 			/*
2600 			 * We have to remove inode from in-memory list if
2601 			 * addition to on disk orphan list failed. Stray orphan
2602 			 * list entries can cause panics at unmount time.
2603 			 */
2604 			mutex_lock(&sbi->s_orphan_lock);
2605 			list_del(&EXT4_I(inode)->i_orphan);
2606 			mutex_unlock(&sbi->s_orphan_lock);
2607 		}
2608 	}
2609 	jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2610 	jbd_debug(4, "orphan inode %lu will point to %d\n",
2611 			inode->i_ino, NEXT_ORPHAN(inode));
2612 out:
2613 	ext4_std_error(sb, err);
2614 	return err;
2615 }
2616 
2617 /*
2618  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2619  * of such inodes stored on disk, because it is finally being cleaned up.
2620  */
2621 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2622 {
2623 	struct list_head *prev;
2624 	struct ext4_inode_info *ei = EXT4_I(inode);
2625 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2626 	__u32 ino_next;
2627 	struct ext4_iloc iloc;
2628 	int err = 0;
2629 
2630 	if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2631 		return 0;
2632 
2633 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2634 		     !mutex_is_locked(&inode->i_mutex));
2635 	/* Do this quick check before taking global s_orphan_lock. */
2636 	if (list_empty(&ei->i_orphan))
2637 		return 0;
2638 
2639 	if (handle) {
2640 		/* Grab inode buffer early before taking global s_orphan_lock */
2641 		err = ext4_reserve_inode_write(handle, inode, &iloc);
2642 	}
2643 
2644 	mutex_lock(&sbi->s_orphan_lock);
2645 	jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2646 
2647 	prev = ei->i_orphan.prev;
2648 	list_del_init(&ei->i_orphan);
2649 
2650 	/* If we're on an error path, we may not have a valid
2651 	 * transaction handle with which to update the orphan list on
2652 	 * disk, but we still need to remove the inode from the linked
2653 	 * list in memory. */
2654 	if (!handle || err) {
2655 		mutex_unlock(&sbi->s_orphan_lock);
2656 		goto out_err;
2657 	}
2658 
2659 	ino_next = NEXT_ORPHAN(inode);
2660 	if (prev == &sbi->s_orphan) {
2661 		jbd_debug(4, "superblock will point to %u\n", ino_next);
2662 		BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2663 		err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2664 		if (err) {
2665 			mutex_unlock(&sbi->s_orphan_lock);
2666 			goto out_brelse;
2667 		}
2668 		sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2669 		mutex_unlock(&sbi->s_orphan_lock);
2670 		err = ext4_handle_dirty_super(handle, inode->i_sb);
2671 	} else {
2672 		struct ext4_iloc iloc2;
2673 		struct inode *i_prev =
2674 			&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2675 
2676 		jbd_debug(4, "orphan inode %lu will point to %u\n",
2677 			  i_prev->i_ino, ino_next);
2678 		err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2679 		if (err) {
2680 			mutex_unlock(&sbi->s_orphan_lock);
2681 			goto out_brelse;
2682 		}
2683 		NEXT_ORPHAN(i_prev) = ino_next;
2684 		err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2685 		mutex_unlock(&sbi->s_orphan_lock);
2686 	}
2687 	if (err)
2688 		goto out_brelse;
2689 	NEXT_ORPHAN(inode) = 0;
2690 	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2691 out_err:
2692 	ext4_std_error(inode->i_sb, err);
2693 	return err;
2694 
2695 out_brelse:
2696 	brelse(iloc.bh);
2697 	goto out_err;
2698 }
2699 
2700 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2701 {
2702 	int retval;
2703 	struct inode *inode;
2704 	struct buffer_head *bh;
2705 	struct ext4_dir_entry_2 *de;
2706 	handle_t *handle = NULL;
2707 
2708 	/* Initialize quotas before so that eventual writes go in
2709 	 * separate transaction */
2710 	dquot_initialize(dir);
2711 	dquot_initialize(dentry->d_inode);
2712 
2713 	retval = -ENOENT;
2714 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2715 	if (IS_ERR(bh))
2716 		return PTR_ERR(bh);
2717 	if (!bh)
2718 		goto end_rmdir;
2719 
2720 	inode = dentry->d_inode;
2721 
2722 	retval = -EIO;
2723 	if (le32_to_cpu(de->inode) != inode->i_ino)
2724 		goto end_rmdir;
2725 
2726 	retval = -ENOTEMPTY;
2727 	if (!empty_dir(inode))
2728 		goto end_rmdir;
2729 
2730 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
2731 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2732 	if (IS_ERR(handle)) {
2733 		retval = PTR_ERR(handle);
2734 		handle = NULL;
2735 		goto end_rmdir;
2736 	}
2737 
2738 	if (IS_DIRSYNC(dir))
2739 		ext4_handle_sync(handle);
2740 
2741 	retval = ext4_delete_entry(handle, dir, de, bh);
2742 	if (retval)
2743 		goto end_rmdir;
2744 	if (!EXT4_DIR_LINK_EMPTY(inode))
2745 		ext4_warning(inode->i_sb,
2746 			     "empty directory has too many links (%d)",
2747 			     inode->i_nlink);
2748 	inode->i_version++;
2749 	clear_nlink(inode);
2750 	/* There's no need to set i_disksize: the fact that i_nlink is
2751 	 * zero will ensure that the right thing happens during any
2752 	 * recovery. */
2753 	inode->i_size = 0;
2754 	ext4_orphan_add(handle, inode);
2755 	inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2756 	ext4_mark_inode_dirty(handle, inode);
2757 	ext4_dec_count(handle, dir);
2758 	ext4_update_dx_flag(dir);
2759 	ext4_mark_inode_dirty(handle, dir);
2760 
2761 end_rmdir:
2762 	brelse(bh);
2763 	if (handle)
2764 		ext4_journal_stop(handle);
2765 	return retval;
2766 }
2767 
2768 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2769 {
2770 	int retval;
2771 	struct inode *inode;
2772 	struct buffer_head *bh;
2773 	struct ext4_dir_entry_2 *de;
2774 	handle_t *handle = NULL;
2775 
2776 	trace_ext4_unlink_enter(dir, dentry);
2777 	/* Initialize quotas before so that eventual writes go
2778 	 * in separate transaction */
2779 	dquot_initialize(dir);
2780 	dquot_initialize(dentry->d_inode);
2781 
2782 	retval = -ENOENT;
2783 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2784 	if (IS_ERR(bh))
2785 		return PTR_ERR(bh);
2786 	if (!bh)
2787 		goto end_unlink;
2788 
2789 	inode = dentry->d_inode;
2790 
2791 	retval = -EIO;
2792 	if (le32_to_cpu(de->inode) != inode->i_ino)
2793 		goto end_unlink;
2794 
2795 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
2796 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2797 	if (IS_ERR(handle)) {
2798 		retval = PTR_ERR(handle);
2799 		handle = NULL;
2800 		goto end_unlink;
2801 	}
2802 
2803 	if (IS_DIRSYNC(dir))
2804 		ext4_handle_sync(handle);
2805 
2806 	if (!inode->i_nlink) {
2807 		ext4_warning(inode->i_sb,
2808 			     "Deleting nonexistent file (%lu), %d",
2809 			     inode->i_ino, inode->i_nlink);
2810 		set_nlink(inode, 1);
2811 	}
2812 	retval = ext4_delete_entry(handle, dir, de, bh);
2813 	if (retval)
2814 		goto end_unlink;
2815 	dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2816 	ext4_update_dx_flag(dir);
2817 	ext4_mark_inode_dirty(handle, dir);
2818 	drop_nlink(inode);
2819 	if (!inode->i_nlink)
2820 		ext4_orphan_add(handle, inode);
2821 	inode->i_ctime = ext4_current_time(inode);
2822 	ext4_mark_inode_dirty(handle, inode);
2823 
2824 end_unlink:
2825 	brelse(bh);
2826 	if (handle)
2827 		ext4_journal_stop(handle);
2828 	trace_ext4_unlink_exit(dentry, retval);
2829 	return retval;
2830 }
2831 
2832 static int ext4_symlink(struct inode *dir,
2833 			struct dentry *dentry, const char *symname)
2834 {
2835 	handle_t *handle;
2836 	struct inode *inode;
2837 	int l, err, retries = 0;
2838 	int credits;
2839 
2840 	l = strlen(symname)+1;
2841 	if (l > dir->i_sb->s_blocksize)
2842 		return -ENAMETOOLONG;
2843 
2844 	dquot_initialize(dir);
2845 
2846 	if (l > EXT4_N_BLOCKS * 4) {
2847 		/*
2848 		 * For non-fast symlinks, we just allocate inode and put it on
2849 		 * orphan list in the first transaction => we need bitmap,
2850 		 * group descriptor, sb, inode block, quota blocks, and
2851 		 * possibly selinux xattr blocks.
2852 		 */
2853 		credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2854 			  EXT4_XATTR_TRANS_BLOCKS;
2855 	} else {
2856 		/*
2857 		 * Fast symlink. We have to add entry to directory
2858 		 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2859 		 * allocate new inode (bitmap, group descriptor, inode block,
2860 		 * quota blocks, sb is already counted in previous macros).
2861 		 */
2862 		credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2863 			  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
2864 	}
2865 retry:
2866 	inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
2867 					    &dentry->d_name, 0, NULL,
2868 					    EXT4_HT_DIR, credits);
2869 	handle = ext4_journal_current_handle();
2870 	err = PTR_ERR(inode);
2871 	if (IS_ERR(inode))
2872 		goto out_stop;
2873 
2874 	if (l > EXT4_N_BLOCKS * 4) {
2875 		inode->i_op = &ext4_symlink_inode_operations;
2876 		ext4_set_aops(inode);
2877 		/*
2878 		 * We cannot call page_symlink() with transaction started
2879 		 * because it calls into ext4_write_begin() which can wait
2880 		 * for transaction commit if we are running out of space
2881 		 * and thus we deadlock. So we have to stop transaction now
2882 		 * and restart it when symlink contents is written.
2883 		 *
2884 		 * To keep fs consistent in case of crash, we have to put inode
2885 		 * to orphan list in the mean time.
2886 		 */
2887 		drop_nlink(inode);
2888 		err = ext4_orphan_add(handle, inode);
2889 		ext4_journal_stop(handle);
2890 		if (err)
2891 			goto err_drop_inode;
2892 		err = __page_symlink(inode, symname, l, 1);
2893 		if (err)
2894 			goto err_drop_inode;
2895 		/*
2896 		 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2897 		 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2898 		 */
2899 		handle = ext4_journal_start(dir, EXT4_HT_DIR,
2900 				EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2901 				EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2902 		if (IS_ERR(handle)) {
2903 			err = PTR_ERR(handle);
2904 			goto err_drop_inode;
2905 		}
2906 		set_nlink(inode, 1);
2907 		err = ext4_orphan_del(handle, inode);
2908 		if (err) {
2909 			ext4_journal_stop(handle);
2910 			clear_nlink(inode);
2911 			goto err_drop_inode;
2912 		}
2913 	} else {
2914 		/* clear the extent format for fast symlink */
2915 		ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2916 		inode->i_op = &ext4_fast_symlink_inode_operations;
2917 		memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2918 		inode->i_size = l-1;
2919 	}
2920 	EXT4_I(inode)->i_disksize = inode->i_size;
2921 	err = ext4_add_nondir(handle, dentry, inode);
2922 	if (!err && IS_DIRSYNC(dir))
2923 		ext4_handle_sync(handle);
2924 
2925 out_stop:
2926 	if (handle)
2927 		ext4_journal_stop(handle);
2928 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2929 		goto retry;
2930 	return err;
2931 err_drop_inode:
2932 	unlock_new_inode(inode);
2933 	iput(inode);
2934 	return err;
2935 }
2936 
2937 static int ext4_link(struct dentry *old_dentry,
2938 		     struct inode *dir, struct dentry *dentry)
2939 {
2940 	handle_t *handle;
2941 	struct inode *inode = old_dentry->d_inode;
2942 	int err, retries = 0;
2943 
2944 	if (inode->i_nlink >= EXT4_LINK_MAX)
2945 		return -EMLINK;
2946 
2947 	dquot_initialize(dir);
2948 
2949 retry:
2950 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
2951 		(EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2952 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
2953 	if (IS_ERR(handle))
2954 		return PTR_ERR(handle);
2955 
2956 	if (IS_DIRSYNC(dir))
2957 		ext4_handle_sync(handle);
2958 
2959 	inode->i_ctime = ext4_current_time(inode);
2960 	ext4_inc_count(handle, inode);
2961 	ihold(inode);
2962 
2963 	err = ext4_add_entry(handle, dentry, inode);
2964 	if (!err) {
2965 		ext4_mark_inode_dirty(handle, inode);
2966 		/* this can happen only for tmpfile being
2967 		 * linked the first time
2968 		 */
2969 		if (inode->i_nlink == 1)
2970 			ext4_orphan_del(handle, inode);
2971 		d_instantiate(dentry, inode);
2972 	} else {
2973 		drop_nlink(inode);
2974 		iput(inode);
2975 	}
2976 	ext4_journal_stop(handle);
2977 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2978 		goto retry;
2979 	return err;
2980 }
2981 
2982 
2983 /*
2984  * Try to find buffer head where contains the parent block.
2985  * It should be the inode block if it is inlined or the 1st block
2986  * if it is a normal dir.
2987  */
2988 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
2989 					struct inode *inode,
2990 					int *retval,
2991 					struct ext4_dir_entry_2 **parent_de,
2992 					int *inlined)
2993 {
2994 	struct buffer_head *bh;
2995 
2996 	if (!ext4_has_inline_data(inode)) {
2997 		bh = ext4_read_dirblock(inode, 0, EITHER);
2998 		if (IS_ERR(bh)) {
2999 			*retval = PTR_ERR(bh);
3000 			return NULL;
3001 		}
3002 		*parent_de = ext4_next_entry(
3003 					(struct ext4_dir_entry_2 *)bh->b_data,
3004 					inode->i_sb->s_blocksize);
3005 		return bh;
3006 	}
3007 
3008 	*inlined = 1;
3009 	return ext4_get_first_inline_block(inode, parent_de, retval);
3010 }
3011 
3012 struct ext4_renament {
3013 	struct inode *dir;
3014 	struct dentry *dentry;
3015 	struct inode *inode;
3016 	bool is_dir;
3017 	int dir_nlink_delta;
3018 
3019 	/* entry for "dentry" */
3020 	struct buffer_head *bh;
3021 	struct ext4_dir_entry_2 *de;
3022 	int inlined;
3023 
3024 	/* entry for ".." in inode if it's a directory */
3025 	struct buffer_head *dir_bh;
3026 	struct ext4_dir_entry_2 *parent_de;
3027 	int dir_inlined;
3028 };
3029 
3030 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3031 {
3032 	int retval;
3033 
3034 	ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3035 					      &retval, &ent->parent_de,
3036 					      &ent->dir_inlined);
3037 	if (!ent->dir_bh)
3038 		return retval;
3039 	if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3040 		return -EIO;
3041 	BUFFER_TRACE(ent->dir_bh, "get_write_access");
3042 	return ext4_journal_get_write_access(handle, ent->dir_bh);
3043 }
3044 
3045 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3046 				  unsigned dir_ino)
3047 {
3048 	int retval;
3049 
3050 	ent->parent_de->inode = cpu_to_le32(dir_ino);
3051 	BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3052 	if (!ent->dir_inlined) {
3053 		if (is_dx(ent->inode)) {
3054 			retval = ext4_handle_dirty_dx_node(handle,
3055 							   ent->inode,
3056 							   ent->dir_bh);
3057 		} else {
3058 			retval = ext4_handle_dirty_dirent_node(handle,
3059 							       ent->inode,
3060 							       ent->dir_bh);
3061 		}
3062 	} else {
3063 		retval = ext4_mark_inode_dirty(handle, ent->inode);
3064 	}
3065 	if (retval) {
3066 		ext4_std_error(ent->dir->i_sb, retval);
3067 		return retval;
3068 	}
3069 	return 0;
3070 }
3071 
3072 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3073 		       unsigned ino, unsigned file_type)
3074 {
3075 	int retval;
3076 
3077 	BUFFER_TRACE(ent->bh, "get write access");
3078 	retval = ext4_journal_get_write_access(handle, ent->bh);
3079 	if (retval)
3080 		return retval;
3081 	ent->de->inode = cpu_to_le32(ino);
3082 	if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb,
3083 				      EXT4_FEATURE_INCOMPAT_FILETYPE))
3084 		ent->de->file_type = file_type;
3085 	ent->dir->i_version++;
3086 	ent->dir->i_ctime = ent->dir->i_mtime =
3087 		ext4_current_time(ent->dir);
3088 	ext4_mark_inode_dirty(handle, ent->dir);
3089 	BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3090 	if (!ent->inlined) {
3091 		retval = ext4_handle_dirty_dirent_node(handle,
3092 						       ent->dir, ent->bh);
3093 		if (unlikely(retval)) {
3094 			ext4_std_error(ent->dir->i_sb, retval);
3095 			return retval;
3096 		}
3097 	}
3098 	brelse(ent->bh);
3099 	ent->bh = NULL;
3100 
3101 	return 0;
3102 }
3103 
3104 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3105 				  const struct qstr *d_name)
3106 {
3107 	int retval = -ENOENT;
3108 	struct buffer_head *bh;
3109 	struct ext4_dir_entry_2 *de;
3110 
3111 	bh = ext4_find_entry(dir, d_name, &de, NULL);
3112 	if (IS_ERR(bh))
3113 		return PTR_ERR(bh);
3114 	if (bh) {
3115 		retval = ext4_delete_entry(handle, dir, de, bh);
3116 		brelse(bh);
3117 	}
3118 	return retval;
3119 }
3120 
3121 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3122 			       int force_reread)
3123 {
3124 	int retval;
3125 	/*
3126 	 * ent->de could have moved from under us during htree split, so make
3127 	 * sure that we are deleting the right entry.  We might also be pointing
3128 	 * to a stale entry in the unused part of ent->bh so just checking inum
3129 	 * and the name isn't enough.
3130 	 */
3131 	if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3132 	    ent->de->name_len != ent->dentry->d_name.len ||
3133 	    strncmp(ent->de->name, ent->dentry->d_name.name,
3134 		    ent->de->name_len) ||
3135 	    force_reread) {
3136 		retval = ext4_find_delete_entry(handle, ent->dir,
3137 						&ent->dentry->d_name);
3138 	} else {
3139 		retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3140 		if (retval == -ENOENT) {
3141 			retval = ext4_find_delete_entry(handle, ent->dir,
3142 							&ent->dentry->d_name);
3143 		}
3144 	}
3145 
3146 	if (retval) {
3147 		ext4_warning(ent->dir->i_sb,
3148 				"Deleting old file (%lu), %d, error=%d",
3149 				ent->dir->i_ino, ent->dir->i_nlink, retval);
3150 	}
3151 }
3152 
3153 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3154 {
3155 	if (ent->dir_nlink_delta) {
3156 		if (ent->dir_nlink_delta == -1)
3157 			ext4_dec_count(handle, ent->dir);
3158 		else
3159 			ext4_inc_count(handle, ent->dir);
3160 		ext4_mark_inode_dirty(handle, ent->dir);
3161 	}
3162 }
3163 
3164 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3165 					      int credits, handle_t **h)
3166 {
3167 	struct inode *wh;
3168 	handle_t *handle;
3169 	int retries = 0;
3170 
3171 	/*
3172 	 * for inode block, sb block, group summaries,
3173 	 * and inode bitmap
3174 	 */
3175 	credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3176 		    EXT4_XATTR_TRANS_BLOCKS + 4);
3177 retry:
3178 	wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3179 					 &ent->dentry->d_name, 0, NULL,
3180 					 EXT4_HT_DIR, credits);
3181 
3182 	handle = ext4_journal_current_handle();
3183 	if (IS_ERR(wh)) {
3184 		if (handle)
3185 			ext4_journal_stop(handle);
3186 		if (PTR_ERR(wh) == -ENOSPC &&
3187 		    ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3188 			goto retry;
3189 	} else {
3190 		*h = handle;
3191 		init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3192 		wh->i_op = &ext4_special_inode_operations;
3193 	}
3194 	return wh;
3195 }
3196 
3197 /*
3198  * Anybody can rename anything with this: the permission checks are left to the
3199  * higher-level routines.
3200  *
3201  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3202  * while new_{dentry,inode) refers to the destination dentry/inode
3203  * This comes from rename(const char *oldpath, const char *newpath)
3204  */
3205 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3206 		       struct inode *new_dir, struct dentry *new_dentry,
3207 		       unsigned int flags)
3208 {
3209 	handle_t *handle = NULL;
3210 	struct ext4_renament old = {
3211 		.dir = old_dir,
3212 		.dentry = old_dentry,
3213 		.inode = old_dentry->d_inode,
3214 	};
3215 	struct ext4_renament new = {
3216 		.dir = new_dir,
3217 		.dentry = new_dentry,
3218 		.inode = new_dentry->d_inode,
3219 	};
3220 	int force_reread;
3221 	int retval;
3222 	struct inode *whiteout = NULL;
3223 	int credits;
3224 	u8 old_file_type;
3225 
3226 	dquot_initialize(old.dir);
3227 	dquot_initialize(new.dir);
3228 
3229 	/* Initialize quotas before so that eventual writes go
3230 	 * in separate transaction */
3231 	if (new.inode)
3232 		dquot_initialize(new.inode);
3233 
3234 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3235 	if (IS_ERR(old.bh))
3236 		return PTR_ERR(old.bh);
3237 	/*
3238 	 *  Check for inode number is _not_ due to possible IO errors.
3239 	 *  We might rmdir the source, keep it as pwd of some process
3240 	 *  and merrily kill the link to whatever was created under the
3241 	 *  same name. Goodbye sticky bit ;-<
3242 	 */
3243 	retval = -ENOENT;
3244 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3245 		goto end_rename;
3246 
3247 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3248 				 &new.de, &new.inlined);
3249 	if (IS_ERR(new.bh)) {
3250 		retval = PTR_ERR(new.bh);
3251 		new.bh = NULL;
3252 		goto end_rename;
3253 	}
3254 	if (new.bh) {
3255 		if (!new.inode) {
3256 			brelse(new.bh);
3257 			new.bh = NULL;
3258 		}
3259 	}
3260 	if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3261 		ext4_alloc_da_blocks(old.inode);
3262 
3263 	credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3264 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3265 	if (!(flags & RENAME_WHITEOUT)) {
3266 		handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3267 		if (IS_ERR(handle))
3268 			return PTR_ERR(handle);
3269 	} else {
3270 		whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3271 		if (IS_ERR(whiteout))
3272 			return PTR_ERR(whiteout);
3273 	}
3274 
3275 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3276 		ext4_handle_sync(handle);
3277 
3278 	if (S_ISDIR(old.inode->i_mode)) {
3279 		if (new.inode) {
3280 			retval = -ENOTEMPTY;
3281 			if (!empty_dir(new.inode))
3282 				goto end_rename;
3283 		} else {
3284 			retval = -EMLINK;
3285 			if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3286 				goto end_rename;
3287 		}
3288 		retval = ext4_rename_dir_prepare(handle, &old);
3289 		if (retval)
3290 			goto end_rename;
3291 	}
3292 	/*
3293 	 * If we're renaming a file within an inline_data dir and adding or
3294 	 * setting the new dirent causes a conversion from inline_data to
3295 	 * extents/blockmap, we need to force the dirent delete code to
3296 	 * re-read the directory, or else we end up trying to delete a dirent
3297 	 * from what is now the extent tree root (or a block map).
3298 	 */
3299 	force_reread = (new.dir->i_ino == old.dir->i_ino &&
3300 			ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3301 
3302 	old_file_type = old.de->file_type;
3303 	if (whiteout) {
3304 		/*
3305 		 * Do this before adding a new entry, so the old entry is sure
3306 		 * to be still pointing to the valid old entry.
3307 		 */
3308 		retval = ext4_setent(handle, &old, whiteout->i_ino,
3309 				     EXT4_FT_CHRDEV);
3310 		if (retval)
3311 			goto end_rename;
3312 		ext4_mark_inode_dirty(handle, whiteout);
3313 	}
3314 	if (!new.bh) {
3315 		retval = ext4_add_entry(handle, new.dentry, old.inode);
3316 		if (retval)
3317 			goto end_rename;
3318 	} else {
3319 		retval = ext4_setent(handle, &new,
3320 				     old.inode->i_ino, old_file_type);
3321 		if (retval)
3322 			goto end_rename;
3323 	}
3324 	if (force_reread)
3325 		force_reread = !ext4_test_inode_flag(new.dir,
3326 						     EXT4_INODE_INLINE_DATA);
3327 
3328 	/*
3329 	 * Like most other Unix systems, set the ctime for inodes on a
3330 	 * rename.
3331 	 */
3332 	old.inode->i_ctime = ext4_current_time(old.inode);
3333 	ext4_mark_inode_dirty(handle, old.inode);
3334 
3335 	if (!whiteout) {
3336 		/*
3337 		 * ok, that's it
3338 		 */
3339 		ext4_rename_delete(handle, &old, force_reread);
3340 	}
3341 
3342 	if (new.inode) {
3343 		ext4_dec_count(handle, new.inode);
3344 		new.inode->i_ctime = ext4_current_time(new.inode);
3345 	}
3346 	old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3347 	ext4_update_dx_flag(old.dir);
3348 	if (old.dir_bh) {
3349 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3350 		if (retval)
3351 			goto end_rename;
3352 
3353 		ext4_dec_count(handle, old.dir);
3354 		if (new.inode) {
3355 			/* checked empty_dir above, can't have another parent,
3356 			 * ext4_dec_count() won't work for many-linked dirs */
3357 			clear_nlink(new.inode);
3358 		} else {
3359 			ext4_inc_count(handle, new.dir);
3360 			ext4_update_dx_flag(new.dir);
3361 			ext4_mark_inode_dirty(handle, new.dir);
3362 		}
3363 	}
3364 	ext4_mark_inode_dirty(handle, old.dir);
3365 	if (new.inode) {
3366 		ext4_mark_inode_dirty(handle, new.inode);
3367 		if (!new.inode->i_nlink)
3368 			ext4_orphan_add(handle, new.inode);
3369 	}
3370 	retval = 0;
3371 
3372 end_rename:
3373 	brelse(old.dir_bh);
3374 	brelse(old.bh);
3375 	brelse(new.bh);
3376 	if (whiteout) {
3377 		if (retval)
3378 			drop_nlink(whiteout);
3379 		unlock_new_inode(whiteout);
3380 		iput(whiteout);
3381 	}
3382 	if (handle)
3383 		ext4_journal_stop(handle);
3384 	return retval;
3385 }
3386 
3387 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3388 			     struct inode *new_dir, struct dentry *new_dentry)
3389 {
3390 	handle_t *handle = NULL;
3391 	struct ext4_renament old = {
3392 		.dir = old_dir,
3393 		.dentry = old_dentry,
3394 		.inode = old_dentry->d_inode,
3395 	};
3396 	struct ext4_renament new = {
3397 		.dir = new_dir,
3398 		.dentry = new_dentry,
3399 		.inode = new_dentry->d_inode,
3400 	};
3401 	u8 new_file_type;
3402 	int retval;
3403 
3404 	dquot_initialize(old.dir);
3405 	dquot_initialize(new.dir);
3406 
3407 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3408 				 &old.de, &old.inlined);
3409 	if (IS_ERR(old.bh))
3410 		return PTR_ERR(old.bh);
3411 	/*
3412 	 *  Check for inode number is _not_ due to possible IO errors.
3413 	 *  We might rmdir the source, keep it as pwd of some process
3414 	 *  and merrily kill the link to whatever was created under the
3415 	 *  same name. Goodbye sticky bit ;-<
3416 	 */
3417 	retval = -ENOENT;
3418 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3419 		goto end_rename;
3420 
3421 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3422 				 &new.de, &new.inlined);
3423 	if (IS_ERR(new.bh)) {
3424 		retval = PTR_ERR(new.bh);
3425 		new.bh = NULL;
3426 		goto end_rename;
3427 	}
3428 
3429 	/* RENAME_EXCHANGE case: old *and* new must both exist */
3430 	if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3431 		goto end_rename;
3432 
3433 	handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3434 		(2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3435 		 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3436 	if (IS_ERR(handle))
3437 		return PTR_ERR(handle);
3438 
3439 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3440 		ext4_handle_sync(handle);
3441 
3442 	if (S_ISDIR(old.inode->i_mode)) {
3443 		old.is_dir = true;
3444 		retval = ext4_rename_dir_prepare(handle, &old);
3445 		if (retval)
3446 			goto end_rename;
3447 	}
3448 	if (S_ISDIR(new.inode->i_mode)) {
3449 		new.is_dir = true;
3450 		retval = ext4_rename_dir_prepare(handle, &new);
3451 		if (retval)
3452 			goto end_rename;
3453 	}
3454 
3455 	/*
3456 	 * Other than the special case of overwriting a directory, parents'
3457 	 * nlink only needs to be modified if this is a cross directory rename.
3458 	 */
3459 	if (old.dir != new.dir && old.is_dir != new.is_dir) {
3460 		old.dir_nlink_delta = old.is_dir ? -1 : 1;
3461 		new.dir_nlink_delta = -old.dir_nlink_delta;
3462 		retval = -EMLINK;
3463 		if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3464 		    (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3465 			goto end_rename;
3466 	}
3467 
3468 	new_file_type = new.de->file_type;
3469 	retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3470 	if (retval)
3471 		goto end_rename;
3472 
3473 	retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3474 	if (retval)
3475 		goto end_rename;
3476 
3477 	/*
3478 	 * Like most other Unix systems, set the ctime for inodes on a
3479 	 * rename.
3480 	 */
3481 	old.inode->i_ctime = ext4_current_time(old.inode);
3482 	new.inode->i_ctime = ext4_current_time(new.inode);
3483 	ext4_mark_inode_dirty(handle, old.inode);
3484 	ext4_mark_inode_dirty(handle, new.inode);
3485 
3486 	if (old.dir_bh) {
3487 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3488 		if (retval)
3489 			goto end_rename;
3490 	}
3491 	if (new.dir_bh) {
3492 		retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3493 		if (retval)
3494 			goto end_rename;
3495 	}
3496 	ext4_update_dir_count(handle, &old);
3497 	ext4_update_dir_count(handle, &new);
3498 	retval = 0;
3499 
3500 end_rename:
3501 	brelse(old.dir_bh);
3502 	brelse(new.dir_bh);
3503 	brelse(old.bh);
3504 	brelse(new.bh);
3505 	if (handle)
3506 		ext4_journal_stop(handle);
3507 	return retval;
3508 }
3509 
3510 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3511 			struct inode *new_dir, struct dentry *new_dentry,
3512 			unsigned int flags)
3513 {
3514 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3515 		return -EINVAL;
3516 
3517 	if (flags & RENAME_EXCHANGE) {
3518 		return ext4_cross_rename(old_dir, old_dentry,
3519 					 new_dir, new_dentry);
3520 	}
3521 
3522 	return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3523 }
3524 
3525 /*
3526  * directories can handle most operations...
3527  */
3528 const struct inode_operations ext4_dir_inode_operations = {
3529 	.create		= ext4_create,
3530 	.lookup		= ext4_lookup,
3531 	.link		= ext4_link,
3532 	.unlink		= ext4_unlink,
3533 	.symlink	= ext4_symlink,
3534 	.mkdir		= ext4_mkdir,
3535 	.rmdir		= ext4_rmdir,
3536 	.mknod		= ext4_mknod,
3537 	.tmpfile	= ext4_tmpfile,
3538 	.rename2	= ext4_rename2,
3539 	.setattr	= ext4_setattr,
3540 	.setxattr	= generic_setxattr,
3541 	.getxattr	= generic_getxattr,
3542 	.listxattr	= ext4_listxattr,
3543 	.removexattr	= generic_removexattr,
3544 	.get_acl	= ext4_get_acl,
3545 	.set_acl	= ext4_set_acl,
3546 	.fiemap         = ext4_fiemap,
3547 };
3548 
3549 const struct inode_operations ext4_special_inode_operations = {
3550 	.setattr	= ext4_setattr,
3551 	.setxattr	= generic_setxattr,
3552 	.getxattr	= generic_getxattr,
3553 	.listxattr	= ext4_listxattr,
3554 	.removexattr	= generic_removexattr,
3555 	.get_acl	= ext4_get_acl,
3556 	.set_acl	= ext4_set_acl,
3557 };
3558