xref: /linux/fs/ext4/namei.c (revision 56d06fa29edd58c448766014afd833b7ff51247b)
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/time.h>
30 #include <linux/fcntl.h>
31 #include <linux/stat.h>
32 #include <linux/string.h>
33 #include <linux/quotaops.h>
34 #include <linux/buffer_head.h>
35 #include <linux/bio.h>
36 #include "ext4.h"
37 #include "ext4_jbd2.h"
38 
39 #include "xattr.h"
40 #include "acl.h"
41 
42 #include <trace/events/ext4.h>
43 /*
44  * define how far ahead to read directories while searching them.
45  */
46 #define NAMEI_RA_CHUNKS  2
47 #define NAMEI_RA_BLOCKS  4
48 #define NAMEI_RA_SIZE	     (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
49 
50 static struct buffer_head *ext4_append(handle_t *handle,
51 					struct inode *inode,
52 					ext4_lblk_t *block)
53 {
54 	struct buffer_head *bh;
55 	int err;
56 
57 	if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
58 		     ((inode->i_size >> 10) >=
59 		      EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
60 		return ERR_PTR(-ENOSPC);
61 
62 	*block = inode->i_size >> inode->i_sb->s_blocksize_bits;
63 
64 	bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
65 	if (IS_ERR(bh))
66 		return bh;
67 	inode->i_size += inode->i_sb->s_blocksize;
68 	EXT4_I(inode)->i_disksize = inode->i_size;
69 	BUFFER_TRACE(bh, "get_write_access");
70 	err = ext4_journal_get_write_access(handle, bh);
71 	if (err) {
72 		brelse(bh);
73 		ext4_std_error(inode->i_sb, err);
74 		return ERR_PTR(err);
75 	}
76 	return bh;
77 }
78 
79 static int ext4_dx_csum_verify(struct inode *inode,
80 			       struct ext4_dir_entry *dirent);
81 
82 typedef enum {
83 	EITHER, INDEX, DIRENT
84 } dirblock_type_t;
85 
86 #define ext4_read_dirblock(inode, block, type) \
87 	__ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
88 
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
90 						ext4_lblk_t block,
91 						dirblock_type_t type,
92 						const char *func,
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 			       "inode #%lu: lblock %lu: comm %s: "
103 			       "error %ld reading directory block",
104 			       inode->i_ino, (unsigned long)block,
105 			       current->comm, PTR_ERR(bh));
106 
107 		return bh;
108 	}
109 	if (!bh) {
110 		ext4_error_inode(inode, func, line, block,
111 				 "Directory hole found");
112 		return ERR_PTR(-EFSCORRUPTED);
113 	}
114 	dirent = (struct ext4_dir_entry *) bh->b_data;
115 	/* Determine whether or not we have an index block */
116 	if (is_dx(inode)) {
117 		if (block == 0)
118 			is_dx_block = 1;
119 		else if (ext4_rec_len_from_disk(dirent->rec_len,
120 						inode->i_sb->s_blocksize) ==
121 			 inode->i_sb->s_blocksize)
122 			is_dx_block = 1;
123 	}
124 	if (!is_dx_block && type == INDEX) {
125 		ext4_error_inode(inode, func, line, block,
126 		       "directory leaf block found instead of index block");
127 		return ERR_PTR(-EFSCORRUPTED);
128 	}
129 	if (!ext4_has_metadata_csum(inode->i_sb) ||
130 	    buffer_verified(bh))
131 		return bh;
132 
133 	/*
134 	 * An empty leaf block can get mistaken for a index block; for
135 	 * this reason, we can only check the index checksum when the
136 	 * caller is sure it should be an index block.
137 	 */
138 	if (is_dx_block && type == INDEX) {
139 		if (ext4_dx_csum_verify(inode, dirent))
140 			set_buffer_verified(bh);
141 		else {
142 			ext4_error_inode(inode, func, line, block,
143 					 "Directory index failed checksum");
144 			brelse(bh);
145 			return ERR_PTR(-EFSBADCRC);
146 		}
147 	}
148 	if (!is_dx_block) {
149 		if (ext4_dirent_csum_verify(inode, dirent))
150 			set_buffer_verified(bh);
151 		else {
152 			ext4_error_inode(inode, func, line, block,
153 					 "Directory block failed checksum");
154 			brelse(bh);
155 			return ERR_PTR(-EFSBADCRC);
156 		}
157 	}
158 	return bh;
159 }
160 
161 #ifndef assert
162 #define assert(test) J_ASSERT(test)
163 #endif
164 
165 #ifdef DX_DEBUG
166 #define dxtrace(command) command
167 #else
168 #define dxtrace(command)
169 #endif
170 
171 struct fake_dirent
172 {
173 	__le32 inode;
174 	__le16 rec_len;
175 	u8 name_len;
176 	u8 file_type;
177 };
178 
179 struct dx_countlimit
180 {
181 	__le16 limit;
182 	__le16 count;
183 };
184 
185 struct dx_entry
186 {
187 	__le32 hash;
188 	__le32 block;
189 };
190 
191 /*
192  * dx_root_info is laid out so that if it should somehow get overlaid by a
193  * dirent the two low bits of the hash version will be zero.  Therefore, the
194  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
195  */
196 
197 struct dx_root
198 {
199 	struct fake_dirent dot;
200 	char dot_name[4];
201 	struct fake_dirent dotdot;
202 	char dotdot_name[4];
203 	struct dx_root_info
204 	{
205 		__le32 reserved_zero;
206 		u8 hash_version;
207 		u8 info_length; /* 8 */
208 		u8 indirect_levels;
209 		u8 unused_flags;
210 	}
211 	info;
212 	struct dx_entry	entries[0];
213 };
214 
215 struct dx_node
216 {
217 	struct fake_dirent fake;
218 	struct dx_entry	entries[0];
219 };
220 
221 
222 struct dx_frame
223 {
224 	struct buffer_head *bh;
225 	struct dx_entry *entries;
226 	struct dx_entry *at;
227 };
228 
229 struct dx_map_entry
230 {
231 	u32 hash;
232 	u16 offs;
233 	u16 size;
234 };
235 
236 /*
237  * This goes at the end of each htree block.
238  */
239 struct dx_tail {
240 	u32 dt_reserved;
241 	__le32 dt_checksum;	/* crc32c(uuid+inum+dirblock) */
242 };
243 
244 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
245 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
246 static inline unsigned dx_get_hash(struct dx_entry *entry);
247 static void dx_set_hash(struct dx_entry *entry, unsigned value);
248 static unsigned dx_get_count(struct dx_entry *entries);
249 static unsigned dx_get_limit(struct dx_entry *entries);
250 static void dx_set_count(struct dx_entry *entries, unsigned value);
251 static void dx_set_limit(struct dx_entry *entries, unsigned value);
252 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
253 static unsigned dx_node_limit(struct inode *dir);
254 static struct dx_frame *dx_probe(struct ext4_filename *fname,
255 				 struct inode *dir,
256 				 struct dx_hash_info *hinfo,
257 				 struct dx_frame *frame);
258 static void dx_release(struct dx_frame *frames);
259 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
260 		       unsigned blocksize, struct dx_hash_info *hinfo,
261 		       struct dx_map_entry map[]);
262 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
263 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
264 		struct dx_map_entry *offsets, int count, unsigned blocksize);
265 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
266 static void dx_insert_block(struct dx_frame *frame,
267 					u32 hash, ext4_lblk_t block);
268 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
269 				 struct dx_frame *frame,
270 				 struct dx_frame *frames,
271 				 __u32 *start_hash);
272 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
273 		struct ext4_filename *fname,
274 		struct ext4_dir_entry_2 **res_dir);
275 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
276 			     struct inode *dir, struct inode *inode);
277 
278 /* checksumming functions */
279 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
280 			    unsigned int blocksize)
281 {
282 	memset(t, 0, sizeof(struct ext4_dir_entry_tail));
283 	t->det_rec_len = ext4_rec_len_to_disk(
284 			sizeof(struct ext4_dir_entry_tail), blocksize);
285 	t->det_reserved_ft = EXT4_FT_DIR_CSUM;
286 }
287 
288 /* Walk through a dirent block to find a checksum "dirent" at the tail */
289 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
290 						   struct ext4_dir_entry *de)
291 {
292 	struct ext4_dir_entry_tail *t;
293 
294 #ifdef PARANOID
295 	struct ext4_dir_entry *d, *top;
296 
297 	d = de;
298 	top = (struct ext4_dir_entry *)(((void *)de) +
299 		(EXT4_BLOCK_SIZE(inode->i_sb) -
300 		sizeof(struct ext4_dir_entry_tail)));
301 	while (d < top && d->rec_len)
302 		d = (struct ext4_dir_entry *)(((void *)d) +
303 		    le16_to_cpu(d->rec_len));
304 
305 	if (d != top)
306 		return NULL;
307 
308 	t = (struct ext4_dir_entry_tail *)d;
309 #else
310 	t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
311 #endif
312 
313 	if (t->det_reserved_zero1 ||
314 	    le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
315 	    t->det_reserved_zero2 ||
316 	    t->det_reserved_ft != EXT4_FT_DIR_CSUM)
317 		return NULL;
318 
319 	return t;
320 }
321 
322 static __le32 ext4_dirent_csum(struct inode *inode,
323 			       struct ext4_dir_entry *dirent, int size)
324 {
325 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
326 	struct ext4_inode_info *ei = EXT4_I(inode);
327 	__u32 csum;
328 
329 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
330 	return cpu_to_le32(csum);
331 }
332 
333 #define warn_no_space_for_csum(inode)					\
334 	__warn_no_space_for_csum((inode), __func__, __LINE__)
335 
336 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
337 				     unsigned int line)
338 {
339 	__ext4_warning_inode(inode, func, line,
340 		"No space for directory leaf checksum. Please run e2fsck -D.");
341 }
342 
343 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
344 {
345 	struct ext4_dir_entry_tail *t;
346 
347 	if (!ext4_has_metadata_csum(inode->i_sb))
348 		return 1;
349 
350 	t = get_dirent_tail(inode, dirent);
351 	if (!t) {
352 		warn_no_space_for_csum(inode);
353 		return 0;
354 	}
355 
356 	if (t->det_checksum != ext4_dirent_csum(inode, dirent,
357 						(void *)t - (void *)dirent))
358 		return 0;
359 
360 	return 1;
361 }
362 
363 static void ext4_dirent_csum_set(struct inode *inode,
364 				 struct ext4_dir_entry *dirent)
365 {
366 	struct ext4_dir_entry_tail *t;
367 
368 	if (!ext4_has_metadata_csum(inode->i_sb))
369 		return;
370 
371 	t = get_dirent_tail(inode, dirent);
372 	if (!t) {
373 		warn_no_space_for_csum(inode);
374 		return;
375 	}
376 
377 	t->det_checksum = ext4_dirent_csum(inode, dirent,
378 					   (void *)t - (void *)dirent);
379 }
380 
381 int ext4_handle_dirty_dirent_node(handle_t *handle,
382 				  struct inode *inode,
383 				  struct buffer_head *bh)
384 {
385 	ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
386 	return ext4_handle_dirty_metadata(handle, inode, bh);
387 }
388 
389 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
390 					       struct ext4_dir_entry *dirent,
391 					       int *offset)
392 {
393 	struct ext4_dir_entry *dp;
394 	struct dx_root_info *root;
395 	int count_offset;
396 
397 	if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
398 		count_offset = 8;
399 	else if (le16_to_cpu(dirent->rec_len) == 12) {
400 		dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
401 		if (le16_to_cpu(dp->rec_len) !=
402 		    EXT4_BLOCK_SIZE(inode->i_sb) - 12)
403 			return NULL;
404 		root = (struct dx_root_info *)(((void *)dp + 12));
405 		if (root->reserved_zero ||
406 		    root->info_length != sizeof(struct dx_root_info))
407 			return NULL;
408 		count_offset = 32;
409 	} else
410 		return NULL;
411 
412 	if (offset)
413 		*offset = count_offset;
414 	return (struct dx_countlimit *)(((void *)dirent) + count_offset);
415 }
416 
417 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
418 			   int count_offset, int count, struct dx_tail *t)
419 {
420 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
421 	struct ext4_inode_info *ei = EXT4_I(inode);
422 	__u32 csum;
423 	__le32 save_csum;
424 	int size;
425 
426 	size = count_offset + (count * sizeof(struct dx_entry));
427 	save_csum = t->dt_checksum;
428 	t->dt_checksum = 0;
429 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
430 	csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
431 	t->dt_checksum = save_csum;
432 
433 	return cpu_to_le32(csum);
434 }
435 
436 static int ext4_dx_csum_verify(struct inode *inode,
437 			       struct ext4_dir_entry *dirent)
438 {
439 	struct dx_countlimit *c;
440 	struct dx_tail *t;
441 	int count_offset, limit, count;
442 
443 	if (!ext4_has_metadata_csum(inode->i_sb))
444 		return 1;
445 
446 	c = get_dx_countlimit(inode, dirent, &count_offset);
447 	if (!c) {
448 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
449 		return 1;
450 	}
451 	limit = le16_to_cpu(c->limit);
452 	count = le16_to_cpu(c->count);
453 	if (count_offset + (limit * sizeof(struct dx_entry)) >
454 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
455 		warn_no_space_for_csum(inode);
456 		return 1;
457 	}
458 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
459 
460 	if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
461 					    count, t))
462 		return 0;
463 	return 1;
464 }
465 
466 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
467 {
468 	struct dx_countlimit *c;
469 	struct dx_tail *t;
470 	int count_offset, limit, count;
471 
472 	if (!ext4_has_metadata_csum(inode->i_sb))
473 		return;
474 
475 	c = get_dx_countlimit(inode, dirent, &count_offset);
476 	if (!c) {
477 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
478 		return;
479 	}
480 	limit = le16_to_cpu(c->limit);
481 	count = le16_to_cpu(c->count);
482 	if (count_offset + (limit * sizeof(struct dx_entry)) >
483 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
484 		warn_no_space_for_csum(inode);
485 		return;
486 	}
487 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
488 
489 	t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
490 }
491 
492 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
493 					    struct inode *inode,
494 					    struct buffer_head *bh)
495 {
496 	ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
497 	return ext4_handle_dirty_metadata(handle, inode, bh);
498 }
499 
500 /*
501  * p is at least 6 bytes before the end of page
502  */
503 static inline struct ext4_dir_entry_2 *
504 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
505 {
506 	return (struct ext4_dir_entry_2 *)((char *)p +
507 		ext4_rec_len_from_disk(p->rec_len, blocksize));
508 }
509 
510 /*
511  * Future: use high four bits of block for coalesce-on-delete flags
512  * Mask them off for now.
513  */
514 
515 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
516 {
517 	return le32_to_cpu(entry->block) & 0x00ffffff;
518 }
519 
520 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
521 {
522 	entry->block = cpu_to_le32(value);
523 }
524 
525 static inline unsigned dx_get_hash(struct dx_entry *entry)
526 {
527 	return le32_to_cpu(entry->hash);
528 }
529 
530 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
531 {
532 	entry->hash = cpu_to_le32(value);
533 }
534 
535 static inline unsigned dx_get_count(struct dx_entry *entries)
536 {
537 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
538 }
539 
540 static inline unsigned dx_get_limit(struct dx_entry *entries)
541 {
542 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
543 }
544 
545 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
546 {
547 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
548 }
549 
550 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
551 {
552 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
553 }
554 
555 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
556 {
557 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
558 		EXT4_DIR_REC_LEN(2) - infosize;
559 
560 	if (ext4_has_metadata_csum(dir->i_sb))
561 		entry_space -= sizeof(struct dx_tail);
562 	return entry_space / sizeof(struct dx_entry);
563 }
564 
565 static inline unsigned dx_node_limit(struct inode *dir)
566 {
567 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
568 
569 	if (ext4_has_metadata_csum(dir->i_sb))
570 		entry_space -= sizeof(struct dx_tail);
571 	return entry_space / sizeof(struct dx_entry);
572 }
573 
574 /*
575  * Debug
576  */
577 #ifdef DX_DEBUG
578 static void dx_show_index(char * label, struct dx_entry *entries)
579 {
580 	int i, n = dx_get_count (entries);
581 	printk(KERN_DEBUG "%s index ", label);
582 	for (i = 0; i < n; i++) {
583 		printk("%x->%lu ", i ? dx_get_hash(entries + i) :
584 				0, (unsigned long)dx_get_block(entries + i));
585 	}
586 	printk("\n");
587 }
588 
589 struct stats
590 {
591 	unsigned names;
592 	unsigned space;
593 	unsigned bcount;
594 };
595 
596 static struct stats dx_show_leaf(struct inode *dir,
597 				struct dx_hash_info *hinfo,
598 				struct ext4_dir_entry_2 *de,
599 				int size, int show_names)
600 {
601 	unsigned names = 0, space = 0;
602 	char *base = (char *) de;
603 	struct dx_hash_info h = *hinfo;
604 
605 	printk("names: ");
606 	while ((char *) de < base + size)
607 	{
608 		if (de->inode)
609 		{
610 			if (show_names)
611 			{
612 #ifdef CONFIG_EXT4_FS_ENCRYPTION
613 				int len;
614 				char *name;
615 				struct ext4_str fname_crypto_str
616 					= {.name = NULL, .len = 0};
617 				int res = 0;
618 
619 				name  = de->name;
620 				len = de->name_len;
621 				if (ext4_encrypted_inode(inode))
622 					res = ext4_get_encryption_info(dir);
623 				if (res) {
624 					printk(KERN_WARNING "Error setting up"
625 					       " fname crypto: %d\n", res);
626 				}
627 				if (ctx == NULL) {
628 					/* Directory is not encrypted */
629 					ext4fs_dirhash(de->name,
630 						de->name_len, &h);
631 					printk("%*.s:(U)%x.%u ", len,
632 					       name, h.hash,
633 					       (unsigned) ((char *) de
634 							   - base));
635 				} else {
636 					/* Directory is encrypted */
637 					res = ext4_fname_crypto_alloc_buffer(
638 						ctx, de->name_len,
639 						&fname_crypto_str);
640 					if (res < 0) {
641 						printk(KERN_WARNING "Error "
642 							"allocating crypto "
643 							"buffer--skipping "
644 							"crypto\n");
645 						ctx = NULL;
646 					}
647 					res = ext4_fname_disk_to_usr(ctx, NULL, de,
648 							&fname_crypto_str);
649 					if (res < 0) {
650 						printk(KERN_WARNING "Error "
651 							"converting filename "
652 							"from disk to usr"
653 							"\n");
654 						name = "??";
655 						len = 2;
656 					} else {
657 						name = fname_crypto_str.name;
658 						len = fname_crypto_str.len;
659 					}
660 					ext4fs_dirhash(de->name, de->name_len,
661 						       &h);
662 					printk("%*.s:(E)%x.%u ", len, name,
663 					       h.hash, (unsigned) ((char *) de
664 								   - base));
665 					ext4_fname_crypto_free_buffer(
666 						&fname_crypto_str);
667 				}
668 #else
669 				int len = de->name_len;
670 				char *name = de->name;
671 				ext4fs_dirhash(de->name, de->name_len, &h);
672 				printk("%*.s:%x.%u ", len, name, h.hash,
673 				       (unsigned) ((char *) de - base));
674 #endif
675 			}
676 			space += EXT4_DIR_REC_LEN(de->name_len);
677 			names++;
678 		}
679 		de = ext4_next_entry(de, size);
680 	}
681 	printk("(%i)\n", names);
682 	return (struct stats) { names, space, 1 };
683 }
684 
685 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
686 			     struct dx_entry *entries, int levels)
687 {
688 	unsigned blocksize = dir->i_sb->s_blocksize;
689 	unsigned count = dx_get_count(entries), names = 0, space = 0, i;
690 	unsigned bcount = 0;
691 	struct buffer_head *bh;
692 	printk("%i indexed blocks...\n", count);
693 	for (i = 0; i < count; i++, entries++)
694 	{
695 		ext4_lblk_t block = dx_get_block(entries);
696 		ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
697 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
698 		struct stats stats;
699 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
700 		bh = ext4_bread(NULL,dir, block, 0);
701 		if (!bh || IS_ERR(bh))
702 			continue;
703 		stats = levels?
704 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
705 		   dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
706 			bh->b_data, blocksize, 0);
707 		names += stats.names;
708 		space += stats.space;
709 		bcount += stats.bcount;
710 		brelse(bh);
711 	}
712 	if (bcount)
713 		printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
714 		       levels ? "" : "   ", names, space/bcount,
715 		       (space/bcount)*100/blocksize);
716 	return (struct stats) { names, space, bcount};
717 }
718 #endif /* DX_DEBUG */
719 
720 /*
721  * Probe for a directory leaf block to search.
722  *
723  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
724  * error in the directory index, and the caller should fall back to
725  * searching the directory normally.  The callers of dx_probe **MUST**
726  * check for this error code, and make sure it never gets reflected
727  * back to userspace.
728  */
729 static struct dx_frame *
730 dx_probe(struct ext4_filename *fname, struct inode *dir,
731 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
732 {
733 	unsigned count, indirect;
734 	struct dx_entry *at, *entries, *p, *q, *m;
735 	struct dx_root *root;
736 	struct dx_frame *frame = frame_in;
737 	struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
738 	u32 hash;
739 
740 	frame->bh = ext4_read_dirblock(dir, 0, INDEX);
741 	if (IS_ERR(frame->bh))
742 		return (struct dx_frame *) frame->bh;
743 
744 	root = (struct dx_root *) frame->bh->b_data;
745 	if (root->info.hash_version != DX_HASH_TEA &&
746 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
747 	    root->info.hash_version != DX_HASH_LEGACY) {
748 		ext4_warning_inode(dir, "Unrecognised inode hash code %u",
749 				   root->info.hash_version);
750 		goto fail;
751 	}
752 	if (fname)
753 		hinfo = &fname->hinfo;
754 	hinfo->hash_version = root->info.hash_version;
755 	if (hinfo->hash_version <= DX_HASH_TEA)
756 		hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
757 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
758 	if (fname && fname_name(fname))
759 		ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
760 	hash = hinfo->hash;
761 
762 	if (root->info.unused_flags & 1) {
763 		ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
764 				   root->info.unused_flags);
765 		goto fail;
766 	}
767 
768 	indirect = root->info.indirect_levels;
769 	if (indirect > 1) {
770 		ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
771 				   root->info.indirect_levels);
772 		goto fail;
773 	}
774 
775 	entries = (struct dx_entry *)(((char *)&root->info) +
776 				      root->info.info_length);
777 
778 	if (dx_get_limit(entries) != dx_root_limit(dir,
779 						   root->info.info_length)) {
780 		ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
781 				   dx_get_limit(entries),
782 				   dx_root_limit(dir, root->info.info_length));
783 		goto fail;
784 	}
785 
786 	dxtrace(printk("Look up %x", hash));
787 	while (1) {
788 		count = dx_get_count(entries);
789 		if (!count || count > dx_get_limit(entries)) {
790 			ext4_warning_inode(dir,
791 					   "dx entry: count %u beyond limit %u",
792 					   count, dx_get_limit(entries));
793 			goto fail;
794 		}
795 
796 		p = entries + 1;
797 		q = entries + count - 1;
798 		while (p <= q) {
799 			m = p + (q - p) / 2;
800 			dxtrace(printk("."));
801 			if (dx_get_hash(m) > hash)
802 				q = m - 1;
803 			else
804 				p = m + 1;
805 		}
806 
807 		if (0) { // linear search cross check
808 			unsigned n = count - 1;
809 			at = entries;
810 			while (n--)
811 			{
812 				dxtrace(printk(","));
813 				if (dx_get_hash(++at) > hash)
814 				{
815 					at--;
816 					break;
817 				}
818 			}
819 			assert (at == p - 1);
820 		}
821 
822 		at = p - 1;
823 		dxtrace(printk(" %x->%u\n", at == entries ? 0 : dx_get_hash(at),
824 			       dx_get_block(at)));
825 		frame->entries = entries;
826 		frame->at = at;
827 		if (!indirect--)
828 			return frame;
829 		frame++;
830 		frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
831 		if (IS_ERR(frame->bh)) {
832 			ret_err = (struct dx_frame *) frame->bh;
833 			frame->bh = NULL;
834 			goto fail;
835 		}
836 		entries = ((struct dx_node *) frame->bh->b_data)->entries;
837 
838 		if (dx_get_limit(entries) != dx_node_limit(dir)) {
839 			ext4_warning_inode(dir,
840 				"dx entry: limit %u != node limit %u",
841 				dx_get_limit(entries), dx_node_limit(dir));
842 			goto fail;
843 		}
844 	}
845 fail:
846 	while (frame >= frame_in) {
847 		brelse(frame->bh);
848 		frame--;
849 	}
850 
851 	if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
852 		ext4_warning_inode(dir,
853 			"Corrupt directory, running e2fsck is recommended");
854 	return ret_err;
855 }
856 
857 static void dx_release(struct dx_frame *frames)
858 {
859 	if (frames[0].bh == NULL)
860 		return;
861 
862 	if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
863 		brelse(frames[1].bh);
864 	brelse(frames[0].bh);
865 }
866 
867 /*
868  * This function increments the frame pointer to search the next leaf
869  * block, and reads in the necessary intervening nodes if the search
870  * should be necessary.  Whether or not the search is necessary is
871  * controlled by the hash parameter.  If the hash value is even, then
872  * the search is only continued if the next block starts with that
873  * hash value.  This is used if we are searching for a specific file.
874  *
875  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
876  *
877  * This function returns 1 if the caller should continue to search,
878  * or 0 if it should not.  If there is an error reading one of the
879  * index blocks, it will a negative error code.
880  *
881  * If start_hash is non-null, it will be filled in with the starting
882  * hash of the next page.
883  */
884 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
885 				 struct dx_frame *frame,
886 				 struct dx_frame *frames,
887 				 __u32 *start_hash)
888 {
889 	struct dx_frame *p;
890 	struct buffer_head *bh;
891 	int num_frames = 0;
892 	__u32 bhash;
893 
894 	p = frame;
895 	/*
896 	 * Find the next leaf page by incrementing the frame pointer.
897 	 * If we run out of entries in the interior node, loop around and
898 	 * increment pointer in the parent node.  When we break out of
899 	 * this loop, num_frames indicates the number of interior
900 	 * nodes need to be read.
901 	 */
902 	while (1) {
903 		if (++(p->at) < p->entries + dx_get_count(p->entries))
904 			break;
905 		if (p == frames)
906 			return 0;
907 		num_frames++;
908 		p--;
909 	}
910 
911 	/*
912 	 * If the hash is 1, then continue only if the next page has a
913 	 * continuation hash of any value.  This is used for readdir
914 	 * handling.  Otherwise, check to see if the hash matches the
915 	 * desired contiuation hash.  If it doesn't, return since
916 	 * there's no point to read in the successive index pages.
917 	 */
918 	bhash = dx_get_hash(p->at);
919 	if (start_hash)
920 		*start_hash = bhash;
921 	if ((hash & 1) == 0) {
922 		if ((bhash & ~1) != hash)
923 			return 0;
924 	}
925 	/*
926 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
927 	 * block so no check is necessary
928 	 */
929 	while (num_frames--) {
930 		bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
931 		if (IS_ERR(bh))
932 			return PTR_ERR(bh);
933 		p++;
934 		brelse(p->bh);
935 		p->bh = bh;
936 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
937 	}
938 	return 1;
939 }
940 
941 
942 /*
943  * This function fills a red-black tree with information from a
944  * directory block.  It returns the number directory entries loaded
945  * into the tree.  If there is an error it is returned in err.
946  */
947 static int htree_dirblock_to_tree(struct file *dir_file,
948 				  struct inode *dir, ext4_lblk_t block,
949 				  struct dx_hash_info *hinfo,
950 				  __u32 start_hash, __u32 start_minor_hash)
951 {
952 	struct buffer_head *bh;
953 	struct ext4_dir_entry_2 *de, *top;
954 	int err = 0, count = 0;
955 	struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}, tmp_str;
956 
957 	dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
958 							(unsigned long)block));
959 	bh = ext4_read_dirblock(dir, block, DIRENT);
960 	if (IS_ERR(bh))
961 		return PTR_ERR(bh);
962 
963 	de = (struct ext4_dir_entry_2 *) bh->b_data;
964 	top = (struct ext4_dir_entry_2 *) ((char *) de +
965 					   dir->i_sb->s_blocksize -
966 					   EXT4_DIR_REC_LEN(0));
967 #ifdef CONFIG_EXT4_FS_ENCRYPTION
968 	/* Check if the directory is encrypted */
969 	if (ext4_encrypted_inode(dir)) {
970 		err = ext4_get_encryption_info(dir);
971 		if (err < 0) {
972 			brelse(bh);
973 			return err;
974 		}
975 		err = ext4_fname_crypto_alloc_buffer(dir, EXT4_NAME_LEN,
976 						     &fname_crypto_str);
977 		if (err < 0) {
978 			brelse(bh);
979 			return err;
980 		}
981 	}
982 #endif
983 	for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
984 		if (ext4_check_dir_entry(dir, NULL, de, bh,
985 				bh->b_data, bh->b_size,
986 				(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
987 					 + ((char *)de - bh->b_data))) {
988 			/* silently ignore the rest of the block */
989 			break;
990 		}
991 		ext4fs_dirhash(de->name, de->name_len, hinfo);
992 		if ((hinfo->hash < start_hash) ||
993 		    ((hinfo->hash == start_hash) &&
994 		     (hinfo->minor_hash < start_minor_hash)))
995 			continue;
996 		if (de->inode == 0)
997 			continue;
998 		if (!ext4_encrypted_inode(dir)) {
999 			tmp_str.name = de->name;
1000 			tmp_str.len = de->name_len;
1001 			err = ext4_htree_store_dirent(dir_file,
1002 				   hinfo->hash, hinfo->minor_hash, de,
1003 				   &tmp_str);
1004 		} else {
1005 			int save_len = fname_crypto_str.len;
1006 
1007 			/* Directory is encrypted */
1008 			err = ext4_fname_disk_to_usr(dir, hinfo, de,
1009 						     &fname_crypto_str);
1010 			if (err < 0) {
1011 				count = err;
1012 				goto errout;
1013 			}
1014 			err = ext4_htree_store_dirent(dir_file,
1015 				   hinfo->hash, hinfo->minor_hash, de,
1016 					&fname_crypto_str);
1017 			fname_crypto_str.len = save_len;
1018 		}
1019 		if (err != 0) {
1020 			count = err;
1021 			goto errout;
1022 		}
1023 		count++;
1024 	}
1025 errout:
1026 	brelse(bh);
1027 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1028 	ext4_fname_crypto_free_buffer(&fname_crypto_str);
1029 #endif
1030 	return count;
1031 }
1032 
1033 
1034 /*
1035  * This function fills a red-black tree with information from a
1036  * directory.  We start scanning the directory in hash order, starting
1037  * at start_hash and start_minor_hash.
1038  *
1039  * This function returns the number of entries inserted into the tree,
1040  * or a negative error code.
1041  */
1042 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1043 			 __u32 start_minor_hash, __u32 *next_hash)
1044 {
1045 	struct dx_hash_info hinfo;
1046 	struct ext4_dir_entry_2 *de;
1047 	struct dx_frame frames[2], *frame;
1048 	struct inode *dir;
1049 	ext4_lblk_t block;
1050 	int count = 0;
1051 	int ret, err;
1052 	__u32 hashval;
1053 	struct ext4_str tmp_str;
1054 
1055 	dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1056 		       start_hash, start_minor_hash));
1057 	dir = file_inode(dir_file);
1058 	if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1059 		hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1060 		if (hinfo.hash_version <= DX_HASH_TEA)
1061 			hinfo.hash_version +=
1062 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
1063 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1064 		if (ext4_has_inline_data(dir)) {
1065 			int has_inline_data = 1;
1066 			count = htree_inlinedir_to_tree(dir_file, dir, 0,
1067 							&hinfo, start_hash,
1068 							start_minor_hash,
1069 							&has_inline_data);
1070 			if (has_inline_data) {
1071 				*next_hash = ~0;
1072 				return count;
1073 			}
1074 		}
1075 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1076 					       start_hash, start_minor_hash);
1077 		*next_hash = ~0;
1078 		return count;
1079 	}
1080 	hinfo.hash = start_hash;
1081 	hinfo.minor_hash = 0;
1082 	frame = dx_probe(NULL, dir, &hinfo, frames);
1083 	if (IS_ERR(frame))
1084 		return PTR_ERR(frame);
1085 
1086 	/* Add '.' and '..' from the htree header */
1087 	if (!start_hash && !start_minor_hash) {
1088 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1089 		tmp_str.name = de->name;
1090 		tmp_str.len = de->name_len;
1091 		err = ext4_htree_store_dirent(dir_file, 0, 0,
1092 					      de, &tmp_str);
1093 		if (err != 0)
1094 			goto errout;
1095 		count++;
1096 	}
1097 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1098 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1099 		de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1100 		tmp_str.name = de->name;
1101 		tmp_str.len = de->name_len;
1102 		err = ext4_htree_store_dirent(dir_file, 2, 0,
1103 					      de, &tmp_str);
1104 		if (err != 0)
1105 			goto errout;
1106 		count++;
1107 	}
1108 
1109 	while (1) {
1110 		block = dx_get_block(frame->at);
1111 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1112 					     start_hash, start_minor_hash);
1113 		if (ret < 0) {
1114 			err = ret;
1115 			goto errout;
1116 		}
1117 		count += ret;
1118 		hashval = ~0;
1119 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1120 					    frame, frames, &hashval);
1121 		*next_hash = hashval;
1122 		if (ret < 0) {
1123 			err = ret;
1124 			goto errout;
1125 		}
1126 		/*
1127 		 * Stop if:  (a) there are no more entries, or
1128 		 * (b) we have inserted at least one entry and the
1129 		 * next hash value is not a continuation
1130 		 */
1131 		if ((ret == 0) ||
1132 		    (count && ((hashval & 1) == 0)))
1133 			break;
1134 	}
1135 	dx_release(frames);
1136 	dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1137 		       "next hash: %x\n", count, *next_hash));
1138 	return count;
1139 errout:
1140 	dx_release(frames);
1141 	return (err);
1142 }
1143 
1144 static inline int search_dirblock(struct buffer_head *bh,
1145 				  struct inode *dir,
1146 				  struct ext4_filename *fname,
1147 				  const struct qstr *d_name,
1148 				  unsigned int offset,
1149 				  struct ext4_dir_entry_2 **res_dir)
1150 {
1151 	return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1152 			       fname, d_name, offset, res_dir);
1153 }
1154 
1155 /*
1156  * Directory block splitting, compacting
1157  */
1158 
1159 /*
1160  * Create map of hash values, offsets, and sizes, stored at end of block.
1161  * Returns number of entries mapped.
1162  */
1163 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1164 		       unsigned blocksize, struct dx_hash_info *hinfo,
1165 		       struct dx_map_entry *map_tail)
1166 {
1167 	int count = 0;
1168 	char *base = (char *) de;
1169 	struct dx_hash_info h = *hinfo;
1170 
1171 	while ((char *) de < base + blocksize) {
1172 		if (de->name_len && de->inode) {
1173 			ext4fs_dirhash(de->name, de->name_len, &h);
1174 			map_tail--;
1175 			map_tail->hash = h.hash;
1176 			map_tail->offs = ((char *) de - base)>>2;
1177 			map_tail->size = le16_to_cpu(de->rec_len);
1178 			count++;
1179 			cond_resched();
1180 		}
1181 		/* XXX: do we need to check rec_len == 0 case? -Chris */
1182 		de = ext4_next_entry(de, blocksize);
1183 	}
1184 	return count;
1185 }
1186 
1187 /* Sort map by hash value */
1188 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1189 {
1190 	struct dx_map_entry *p, *q, *top = map + count - 1;
1191 	int more;
1192 	/* Combsort until bubble sort doesn't suck */
1193 	while (count > 2) {
1194 		count = count*10/13;
1195 		if (count - 9 < 2) /* 9, 10 -> 11 */
1196 			count = 11;
1197 		for (p = top, q = p - count; q >= map; p--, q--)
1198 			if (p->hash < q->hash)
1199 				swap(*p, *q);
1200 	}
1201 	/* Garden variety bubble sort */
1202 	do {
1203 		more = 0;
1204 		q = top;
1205 		while (q-- > map) {
1206 			if (q[1].hash >= q[0].hash)
1207 				continue;
1208 			swap(*(q+1), *q);
1209 			more = 1;
1210 		}
1211 	} while(more);
1212 }
1213 
1214 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1215 {
1216 	struct dx_entry *entries = frame->entries;
1217 	struct dx_entry *old = frame->at, *new = old + 1;
1218 	int count = dx_get_count(entries);
1219 
1220 	assert(count < dx_get_limit(entries));
1221 	assert(old < entries + count);
1222 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1223 	dx_set_hash(new, hash);
1224 	dx_set_block(new, block);
1225 	dx_set_count(entries, count + 1);
1226 }
1227 
1228 /*
1229  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1230  *
1231  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1232  * `de != NULL' is guaranteed by caller.
1233  */
1234 static inline int ext4_match(struct ext4_filename *fname,
1235 			     struct ext4_dir_entry_2 *de)
1236 {
1237 	const void *name = fname_name(fname);
1238 	u32 len = fname_len(fname);
1239 
1240 	if (!de->inode)
1241 		return 0;
1242 
1243 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1244 	if (unlikely(!name)) {
1245 		if (fname->usr_fname->name[0] == '_') {
1246 			int ret;
1247 			if (de->name_len < 16)
1248 				return 0;
1249 			ret = memcmp(de->name + de->name_len - 16,
1250 				     fname->crypto_buf.name + 8, 16);
1251 			return (ret == 0) ? 1 : 0;
1252 		}
1253 		name = fname->crypto_buf.name;
1254 		len = fname->crypto_buf.len;
1255 	}
1256 #endif
1257 	if (de->name_len != len)
1258 		return 0;
1259 	return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1260 }
1261 
1262 /*
1263  * Returns 0 if not found, -1 on failure, and 1 on success
1264  */
1265 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1266 		    struct inode *dir, struct ext4_filename *fname,
1267 		    const struct qstr *d_name,
1268 		    unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1269 {
1270 	struct ext4_dir_entry_2 * de;
1271 	char * dlimit;
1272 	int de_len;
1273 	int res;
1274 
1275 	de = (struct ext4_dir_entry_2 *)search_buf;
1276 	dlimit = search_buf + buf_size;
1277 	while ((char *) de < dlimit) {
1278 		/* this code is executed quadratically often */
1279 		/* do minimal checking `by hand' */
1280 		if ((char *) de + de->name_len <= dlimit) {
1281 			res = ext4_match(fname, de);
1282 			if (res < 0) {
1283 				res = -1;
1284 				goto return_result;
1285 			}
1286 			if (res > 0) {
1287 				/* found a match - just to be sure, do
1288 				 * a full check */
1289 				if (ext4_check_dir_entry(dir, NULL, de, bh,
1290 						bh->b_data,
1291 						 bh->b_size, offset)) {
1292 					res = -1;
1293 					goto return_result;
1294 				}
1295 				*res_dir = de;
1296 				res = 1;
1297 				goto return_result;
1298 			}
1299 
1300 		}
1301 		/* prevent looping on a bad block */
1302 		de_len = ext4_rec_len_from_disk(de->rec_len,
1303 						dir->i_sb->s_blocksize);
1304 		if (de_len <= 0) {
1305 			res = -1;
1306 			goto return_result;
1307 		}
1308 		offset += de_len;
1309 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1310 	}
1311 
1312 	res = 0;
1313 return_result:
1314 	return res;
1315 }
1316 
1317 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1318 			       struct ext4_dir_entry *de)
1319 {
1320 	struct super_block *sb = dir->i_sb;
1321 
1322 	if (!is_dx(dir))
1323 		return 0;
1324 	if (block == 0)
1325 		return 1;
1326 	if (de->inode == 0 &&
1327 	    ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1328 			sb->s_blocksize)
1329 		return 1;
1330 	return 0;
1331 }
1332 
1333 /*
1334  *	ext4_find_entry()
1335  *
1336  * finds an entry in the specified directory with the wanted name. It
1337  * returns the cache buffer in which the entry was found, and the entry
1338  * itself (as a parameter - res_dir). It does NOT read the inode of the
1339  * entry - you'll have to do that yourself if you want to.
1340  *
1341  * The returned buffer_head has ->b_count elevated.  The caller is expected
1342  * to brelse() it when appropriate.
1343  */
1344 static struct buffer_head * ext4_find_entry (struct inode *dir,
1345 					const struct qstr *d_name,
1346 					struct ext4_dir_entry_2 **res_dir,
1347 					int *inlined)
1348 {
1349 	struct super_block *sb;
1350 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1351 	struct buffer_head *bh, *ret = NULL;
1352 	ext4_lblk_t start, block, b;
1353 	const u8 *name = d_name->name;
1354 	int ra_max = 0;		/* Number of bh's in the readahead
1355 				   buffer, bh_use[] */
1356 	int ra_ptr = 0;		/* Current index into readahead
1357 				   buffer */
1358 	int num = 0;
1359 	ext4_lblk_t  nblocks;
1360 	int i, namelen, retval;
1361 	struct ext4_filename fname;
1362 
1363 	*res_dir = NULL;
1364 	sb = dir->i_sb;
1365 	namelen = d_name->len;
1366 	if (namelen > EXT4_NAME_LEN)
1367 		return NULL;
1368 
1369 	retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1370 	if (retval)
1371 		return ERR_PTR(retval);
1372 
1373 	if (ext4_has_inline_data(dir)) {
1374 		int has_inline_data = 1;
1375 		ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1376 					     &has_inline_data);
1377 		if (has_inline_data) {
1378 			if (inlined)
1379 				*inlined = 1;
1380 			goto cleanup_and_exit;
1381 		}
1382 	}
1383 
1384 	if ((namelen <= 2) && (name[0] == '.') &&
1385 	    (name[1] == '.' || name[1] == '\0')) {
1386 		/*
1387 		 * "." or ".." will only be in the first block
1388 		 * NFS may look up ".."; "." should be handled by the VFS
1389 		 */
1390 		block = start = 0;
1391 		nblocks = 1;
1392 		goto restart;
1393 	}
1394 	if (is_dx(dir)) {
1395 		ret = ext4_dx_find_entry(dir, &fname, res_dir);
1396 		/*
1397 		 * On success, or if the error was file not found,
1398 		 * return.  Otherwise, fall back to doing a search the
1399 		 * old fashioned way.
1400 		 */
1401 		if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1402 			goto cleanup_and_exit;
1403 		dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1404 			       "falling back\n"));
1405 	}
1406 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1407 	start = EXT4_I(dir)->i_dir_start_lookup;
1408 	if (start >= nblocks)
1409 		start = 0;
1410 	block = start;
1411 restart:
1412 	do {
1413 		/*
1414 		 * We deal with the read-ahead logic here.
1415 		 */
1416 		if (ra_ptr >= ra_max) {
1417 			/* Refill the readahead buffer */
1418 			ra_ptr = 0;
1419 			b = block;
1420 			for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1421 				/*
1422 				 * Terminate if we reach the end of the
1423 				 * directory and must wrap, or if our
1424 				 * search has finished at this block.
1425 				 */
1426 				if (b >= nblocks || (num && block == start)) {
1427 					bh_use[ra_max] = NULL;
1428 					break;
1429 				}
1430 				num++;
1431 				bh = ext4_getblk(NULL, dir, b++, 0);
1432 				if (IS_ERR(bh)) {
1433 					if (ra_max == 0) {
1434 						ret = bh;
1435 						goto cleanup_and_exit;
1436 					}
1437 					break;
1438 				}
1439 				bh_use[ra_max] = bh;
1440 				if (bh)
1441 					ll_rw_block(READ | REQ_META | REQ_PRIO,
1442 						    1, &bh);
1443 			}
1444 		}
1445 		if ((bh = bh_use[ra_ptr++]) == NULL)
1446 			goto next;
1447 		wait_on_buffer(bh);
1448 		if (!buffer_uptodate(bh)) {
1449 			/* read error, skip block & hope for the best */
1450 			EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1451 					 (unsigned long) block);
1452 			brelse(bh);
1453 			goto next;
1454 		}
1455 		if (!buffer_verified(bh) &&
1456 		    !is_dx_internal_node(dir, block,
1457 					 (struct ext4_dir_entry *)bh->b_data) &&
1458 		    !ext4_dirent_csum_verify(dir,
1459 				(struct ext4_dir_entry *)bh->b_data)) {
1460 			EXT4_ERROR_INODE(dir, "checksumming directory "
1461 					 "block %lu", (unsigned long)block);
1462 			brelse(bh);
1463 			goto next;
1464 		}
1465 		set_buffer_verified(bh);
1466 		i = search_dirblock(bh, dir, &fname, d_name,
1467 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1468 		if (i == 1) {
1469 			EXT4_I(dir)->i_dir_start_lookup = block;
1470 			ret = bh;
1471 			goto cleanup_and_exit;
1472 		} else {
1473 			brelse(bh);
1474 			if (i < 0)
1475 				goto cleanup_and_exit;
1476 		}
1477 	next:
1478 		if (++block >= nblocks)
1479 			block = 0;
1480 	} while (block != start);
1481 
1482 	/*
1483 	 * If the directory has grown while we were searching, then
1484 	 * search the last part of the directory before giving up.
1485 	 */
1486 	block = nblocks;
1487 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1488 	if (block < nblocks) {
1489 		start = 0;
1490 		goto restart;
1491 	}
1492 
1493 cleanup_and_exit:
1494 	/* Clean up the read-ahead blocks */
1495 	for (; ra_ptr < ra_max; ra_ptr++)
1496 		brelse(bh_use[ra_ptr]);
1497 	ext4_fname_free_filename(&fname);
1498 	return ret;
1499 }
1500 
1501 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1502 			struct ext4_filename *fname,
1503 			struct ext4_dir_entry_2 **res_dir)
1504 {
1505 	struct super_block * sb = dir->i_sb;
1506 	struct dx_frame frames[2], *frame;
1507 	const struct qstr *d_name = fname->usr_fname;
1508 	struct buffer_head *bh;
1509 	ext4_lblk_t block;
1510 	int retval;
1511 
1512 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1513 	*res_dir = NULL;
1514 #endif
1515 	frame = dx_probe(fname, dir, NULL, frames);
1516 	if (IS_ERR(frame))
1517 		return (struct buffer_head *) frame;
1518 	do {
1519 		block = dx_get_block(frame->at);
1520 		bh = ext4_read_dirblock(dir, block, DIRENT);
1521 		if (IS_ERR(bh))
1522 			goto errout;
1523 
1524 		retval = search_dirblock(bh, dir, fname, d_name,
1525 					 block << EXT4_BLOCK_SIZE_BITS(sb),
1526 					 res_dir);
1527 		if (retval == 1)
1528 			goto success;
1529 		brelse(bh);
1530 		if (retval == -1) {
1531 			bh = ERR_PTR(ERR_BAD_DX_DIR);
1532 			goto errout;
1533 		}
1534 
1535 		/* Check to see if we should continue to search */
1536 		retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1537 					       frames, NULL);
1538 		if (retval < 0) {
1539 			ext4_warning_inode(dir,
1540 				"error %d reading directory index block",
1541 				retval);
1542 			bh = ERR_PTR(retval);
1543 			goto errout;
1544 		}
1545 	} while (retval == 1);
1546 
1547 	bh = NULL;
1548 errout:
1549 	dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1550 success:
1551 	dx_release(frames);
1552 	return bh;
1553 }
1554 
1555 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1556 {
1557 	struct inode *inode;
1558 	struct ext4_dir_entry_2 *de;
1559 	struct buffer_head *bh;
1560 
1561        if (ext4_encrypted_inode(dir)) {
1562                int res = ext4_get_encryption_info(dir);
1563 
1564 		/*
1565 		 * This should be a properly defined flag for
1566 		 * dentry->d_flags when we uplift this to the VFS.
1567 		 * d_fsdata is set to (void *) 1 if if the dentry is
1568 		 * created while the directory was encrypted and we
1569 		 * don't have access to the key.
1570 		 */
1571 	       dentry->d_fsdata = NULL;
1572 	       if (ext4_encryption_info(dir))
1573 		       dentry->d_fsdata = (void *) 1;
1574 	       d_set_d_op(dentry, &ext4_encrypted_d_ops);
1575 	       if (res && res != -ENOKEY)
1576 		       return ERR_PTR(res);
1577        }
1578 
1579 	if (dentry->d_name.len > EXT4_NAME_LEN)
1580 		return ERR_PTR(-ENAMETOOLONG);
1581 
1582 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1583 	if (IS_ERR(bh))
1584 		return (struct dentry *) bh;
1585 	inode = NULL;
1586 	if (bh) {
1587 		__u32 ino = le32_to_cpu(de->inode);
1588 		brelse(bh);
1589 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1590 			EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1591 			return ERR_PTR(-EFSCORRUPTED);
1592 		}
1593 		if (unlikely(ino == dir->i_ino)) {
1594 			EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1595 					 dentry);
1596 			return ERR_PTR(-EFSCORRUPTED);
1597 		}
1598 		inode = ext4_iget_normal(dir->i_sb, ino);
1599 		if (inode == ERR_PTR(-ESTALE)) {
1600 			EXT4_ERROR_INODE(dir,
1601 					 "deleted inode referenced: %u",
1602 					 ino);
1603 			return ERR_PTR(-EFSCORRUPTED);
1604 		}
1605 		if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1606 		    (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1607 		    !ext4_is_child_context_consistent_with_parent(dir,
1608 								  inode)) {
1609 			int nokey = ext4_encrypted_inode(inode) &&
1610 				!ext4_encryption_info(inode);
1611 
1612 			iput(inode);
1613 			if (nokey)
1614 				return ERR_PTR(-ENOKEY);
1615 			ext4_warning(inode->i_sb,
1616 				     "Inconsistent encryption contexts: %lu/%lu\n",
1617 				     (unsigned long) dir->i_ino,
1618 				     (unsigned long) inode->i_ino);
1619 			return ERR_PTR(-EPERM);
1620 		}
1621 	}
1622 	return d_splice_alias(inode, dentry);
1623 }
1624 
1625 
1626 struct dentry *ext4_get_parent(struct dentry *child)
1627 {
1628 	__u32 ino;
1629 	static const struct qstr dotdot = QSTR_INIT("..", 2);
1630 	struct ext4_dir_entry_2 * de;
1631 	struct buffer_head *bh;
1632 
1633 	bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1634 	if (IS_ERR(bh))
1635 		return (struct dentry *) bh;
1636 	if (!bh)
1637 		return ERR_PTR(-ENOENT);
1638 	ino = le32_to_cpu(de->inode);
1639 	brelse(bh);
1640 
1641 	if (!ext4_valid_inum(d_inode(child)->i_sb, ino)) {
1642 		EXT4_ERROR_INODE(d_inode(child),
1643 				 "bad parent inode number: %u", ino);
1644 		return ERR_PTR(-EFSCORRUPTED);
1645 	}
1646 
1647 	return d_obtain_alias(ext4_iget_normal(d_inode(child)->i_sb, ino));
1648 }
1649 
1650 /*
1651  * Move count entries from end of map between two memory locations.
1652  * Returns pointer to last entry moved.
1653  */
1654 static struct ext4_dir_entry_2 *
1655 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1656 		unsigned blocksize)
1657 {
1658 	unsigned rec_len = 0;
1659 
1660 	while (count--) {
1661 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1662 						(from + (map->offs<<2));
1663 		rec_len = EXT4_DIR_REC_LEN(de->name_len);
1664 		memcpy (to, de, rec_len);
1665 		((struct ext4_dir_entry_2 *) to)->rec_len =
1666 				ext4_rec_len_to_disk(rec_len, blocksize);
1667 		de->inode = 0;
1668 		map++;
1669 		to += rec_len;
1670 	}
1671 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1672 }
1673 
1674 /*
1675  * Compact each dir entry in the range to the minimal rec_len.
1676  * Returns pointer to last entry in range.
1677  */
1678 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1679 {
1680 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1681 	unsigned rec_len = 0;
1682 
1683 	prev = to = de;
1684 	while ((char*)de < base + blocksize) {
1685 		next = ext4_next_entry(de, blocksize);
1686 		if (de->inode && de->name_len) {
1687 			rec_len = EXT4_DIR_REC_LEN(de->name_len);
1688 			if (de > to)
1689 				memmove(to, de, rec_len);
1690 			to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1691 			prev = to;
1692 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1693 		}
1694 		de = next;
1695 	}
1696 	return prev;
1697 }
1698 
1699 /*
1700  * Split a full leaf block to make room for a new dir entry.
1701  * Allocate a new block, and move entries so that they are approx. equally full.
1702  * Returns pointer to de in block into which the new entry will be inserted.
1703  */
1704 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1705 			struct buffer_head **bh,struct dx_frame *frame,
1706 			struct dx_hash_info *hinfo)
1707 {
1708 	unsigned blocksize = dir->i_sb->s_blocksize;
1709 	unsigned count, continued;
1710 	struct buffer_head *bh2;
1711 	ext4_lblk_t newblock;
1712 	u32 hash2;
1713 	struct dx_map_entry *map;
1714 	char *data1 = (*bh)->b_data, *data2;
1715 	unsigned split, move, size;
1716 	struct ext4_dir_entry_2 *de = NULL, *de2;
1717 	struct ext4_dir_entry_tail *t;
1718 	int	csum_size = 0;
1719 	int	err = 0, i;
1720 
1721 	if (ext4_has_metadata_csum(dir->i_sb))
1722 		csum_size = sizeof(struct ext4_dir_entry_tail);
1723 
1724 	bh2 = ext4_append(handle, dir, &newblock);
1725 	if (IS_ERR(bh2)) {
1726 		brelse(*bh);
1727 		*bh = NULL;
1728 		return (struct ext4_dir_entry_2 *) bh2;
1729 	}
1730 
1731 	BUFFER_TRACE(*bh, "get_write_access");
1732 	err = ext4_journal_get_write_access(handle, *bh);
1733 	if (err)
1734 		goto journal_error;
1735 
1736 	BUFFER_TRACE(frame->bh, "get_write_access");
1737 	err = ext4_journal_get_write_access(handle, frame->bh);
1738 	if (err)
1739 		goto journal_error;
1740 
1741 	data2 = bh2->b_data;
1742 
1743 	/* create map in the end of data2 block */
1744 	map = (struct dx_map_entry *) (data2 + blocksize);
1745 	count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1746 			     blocksize, hinfo, map);
1747 	map -= count;
1748 	dx_sort_map(map, count);
1749 	/* Split the existing block in the middle, size-wise */
1750 	size = 0;
1751 	move = 0;
1752 	for (i = count-1; i >= 0; i--) {
1753 		/* is more than half of this entry in 2nd half of the block? */
1754 		if (size + map[i].size/2 > blocksize/2)
1755 			break;
1756 		size += map[i].size;
1757 		move++;
1758 	}
1759 	/* map index at which we will split */
1760 	split = count - move;
1761 	hash2 = map[split].hash;
1762 	continued = hash2 == map[split - 1].hash;
1763 	dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1764 			(unsigned long)dx_get_block(frame->at),
1765 					hash2, split, count-split));
1766 
1767 	/* Fancy dance to stay within two buffers */
1768 	de2 = dx_move_dirents(data1, data2, map + split, count - split,
1769 			      blocksize);
1770 	de = dx_pack_dirents(data1, blocksize);
1771 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1772 					   (char *) de,
1773 					   blocksize);
1774 	de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1775 					    (char *) de2,
1776 					    blocksize);
1777 	if (csum_size) {
1778 		t = EXT4_DIRENT_TAIL(data2, blocksize);
1779 		initialize_dirent_tail(t, blocksize);
1780 
1781 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1782 		initialize_dirent_tail(t, blocksize);
1783 	}
1784 
1785 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1786 			blocksize, 1));
1787 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1788 			blocksize, 1));
1789 
1790 	/* Which block gets the new entry? */
1791 	if (hinfo->hash >= hash2) {
1792 		swap(*bh, bh2);
1793 		de = de2;
1794 	}
1795 	dx_insert_block(frame, hash2 + continued, newblock);
1796 	err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1797 	if (err)
1798 		goto journal_error;
1799 	err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1800 	if (err)
1801 		goto journal_error;
1802 	brelse(bh2);
1803 	dxtrace(dx_show_index("frame", frame->entries));
1804 	return de;
1805 
1806 journal_error:
1807 	brelse(*bh);
1808 	brelse(bh2);
1809 	*bh = NULL;
1810 	ext4_std_error(dir->i_sb, err);
1811 	return ERR_PTR(err);
1812 }
1813 
1814 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1815 		      struct buffer_head *bh,
1816 		      void *buf, int buf_size,
1817 		      struct ext4_filename *fname,
1818 		      struct ext4_dir_entry_2 **dest_de)
1819 {
1820 	struct ext4_dir_entry_2 *de;
1821 	unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1822 	int nlen, rlen;
1823 	unsigned int offset = 0;
1824 	char *top;
1825 	int res;
1826 
1827 	de = (struct ext4_dir_entry_2 *)buf;
1828 	top = buf + buf_size - reclen;
1829 	while ((char *) de <= top) {
1830 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1831 					 buf, buf_size, offset)) {
1832 			res = -EFSCORRUPTED;
1833 			goto return_result;
1834 		}
1835 		/* Provide crypto context and crypto buffer to ext4 match */
1836 		res = ext4_match(fname, de);
1837 		if (res < 0)
1838 			goto return_result;
1839 		if (res > 0) {
1840 			res = -EEXIST;
1841 			goto return_result;
1842 		}
1843 		nlen = EXT4_DIR_REC_LEN(de->name_len);
1844 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1845 		if ((de->inode ? rlen - nlen : rlen) >= reclen)
1846 			break;
1847 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1848 		offset += rlen;
1849 	}
1850 
1851 	if ((char *) de > top)
1852 		res = -ENOSPC;
1853 	else {
1854 		*dest_de = de;
1855 		res = 0;
1856 	}
1857 return_result:
1858 	return res;
1859 }
1860 
1861 int ext4_insert_dentry(struct inode *dir,
1862 		       struct inode *inode,
1863 		       struct ext4_dir_entry_2 *de,
1864 		       int buf_size,
1865 		       struct ext4_filename *fname)
1866 {
1867 
1868 	int nlen, rlen;
1869 
1870 	nlen = EXT4_DIR_REC_LEN(de->name_len);
1871 	rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1872 	if (de->inode) {
1873 		struct ext4_dir_entry_2 *de1 =
1874 			(struct ext4_dir_entry_2 *)((char *)de + nlen);
1875 		de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1876 		de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1877 		de = de1;
1878 	}
1879 	de->file_type = EXT4_FT_UNKNOWN;
1880 	de->inode = cpu_to_le32(inode->i_ino);
1881 	ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1882 	de->name_len = fname_len(fname);
1883 	memcpy(de->name, fname_name(fname), fname_len(fname));
1884 	return 0;
1885 }
1886 
1887 /*
1888  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1889  * it points to a directory entry which is guaranteed to be large
1890  * enough for new directory entry.  If de is NULL, then
1891  * add_dirent_to_buf will attempt search the directory block for
1892  * space.  It will return -ENOSPC if no space is available, and -EIO
1893  * and -EEXIST if directory entry already exists.
1894  */
1895 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1896 			     struct inode *dir,
1897 			     struct inode *inode, struct ext4_dir_entry_2 *de,
1898 			     struct buffer_head *bh)
1899 {
1900 	unsigned int	blocksize = dir->i_sb->s_blocksize;
1901 	int		csum_size = 0;
1902 	int		err;
1903 
1904 	if (ext4_has_metadata_csum(inode->i_sb))
1905 		csum_size = sizeof(struct ext4_dir_entry_tail);
1906 
1907 	if (!de) {
1908 		err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1909 					blocksize - csum_size, fname, &de);
1910 		if (err)
1911 			return err;
1912 	}
1913 	BUFFER_TRACE(bh, "get_write_access");
1914 	err = ext4_journal_get_write_access(handle, bh);
1915 	if (err) {
1916 		ext4_std_error(dir->i_sb, err);
1917 		return err;
1918 	}
1919 
1920 	/* By now the buffer is marked for journaling. Due to crypto operations,
1921 	 * the following function call may fail */
1922 	err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1923 	if (err < 0)
1924 		return err;
1925 
1926 	/*
1927 	 * XXX shouldn't update any times until successful
1928 	 * completion of syscall, but too many callers depend
1929 	 * on this.
1930 	 *
1931 	 * XXX similarly, too many callers depend on
1932 	 * ext4_new_inode() setting the times, but error
1933 	 * recovery deletes the inode, so the worst that can
1934 	 * happen is that the times are slightly out of date
1935 	 * and/or different from the directory change time.
1936 	 */
1937 	dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1938 	ext4_update_dx_flag(dir);
1939 	dir->i_version++;
1940 	ext4_mark_inode_dirty(handle, dir);
1941 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1942 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1943 	if (err)
1944 		ext4_std_error(dir->i_sb, err);
1945 	return 0;
1946 }
1947 
1948 /*
1949  * This converts a one block unindexed directory to a 3 block indexed
1950  * directory, and adds the dentry to the indexed directory.
1951  */
1952 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1953 			    struct inode *dir,
1954 			    struct inode *inode, struct buffer_head *bh)
1955 {
1956 	struct buffer_head *bh2;
1957 	struct dx_root	*root;
1958 	struct dx_frame	frames[2], *frame;
1959 	struct dx_entry *entries;
1960 	struct ext4_dir_entry_2	*de, *de2;
1961 	struct ext4_dir_entry_tail *t;
1962 	char		*data1, *top;
1963 	unsigned	len;
1964 	int		retval;
1965 	unsigned	blocksize;
1966 	ext4_lblk_t  block;
1967 	struct fake_dirent *fde;
1968 	int csum_size = 0;
1969 
1970 	if (ext4_has_metadata_csum(inode->i_sb))
1971 		csum_size = sizeof(struct ext4_dir_entry_tail);
1972 
1973 	blocksize =  dir->i_sb->s_blocksize;
1974 	dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1975 	BUFFER_TRACE(bh, "get_write_access");
1976 	retval = ext4_journal_get_write_access(handle, bh);
1977 	if (retval) {
1978 		ext4_std_error(dir->i_sb, retval);
1979 		brelse(bh);
1980 		return retval;
1981 	}
1982 	root = (struct dx_root *) bh->b_data;
1983 
1984 	/* The 0th block becomes the root, move the dirents out */
1985 	fde = &root->dotdot;
1986 	de = (struct ext4_dir_entry_2 *)((char *)fde +
1987 		ext4_rec_len_from_disk(fde->rec_len, blocksize));
1988 	if ((char *) de >= (((char *) root) + blocksize)) {
1989 		EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1990 		brelse(bh);
1991 		return -EFSCORRUPTED;
1992 	}
1993 	len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1994 
1995 	/* Allocate new block for the 0th block's dirents */
1996 	bh2 = ext4_append(handle, dir, &block);
1997 	if (IS_ERR(bh2)) {
1998 		brelse(bh);
1999 		return PTR_ERR(bh2);
2000 	}
2001 	ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2002 	data1 = bh2->b_data;
2003 
2004 	memcpy (data1, de, len);
2005 	de = (struct ext4_dir_entry_2 *) data1;
2006 	top = data1 + len;
2007 	while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2008 		de = de2;
2009 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2010 					   (char *) de,
2011 					   blocksize);
2012 
2013 	if (csum_size) {
2014 		t = EXT4_DIRENT_TAIL(data1, blocksize);
2015 		initialize_dirent_tail(t, blocksize);
2016 	}
2017 
2018 	/* Initialize the root; the dot dirents already exist */
2019 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2020 	de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2021 					   blocksize);
2022 	memset (&root->info, 0, sizeof(root->info));
2023 	root->info.info_length = sizeof(root->info);
2024 	root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2025 	entries = root->entries;
2026 	dx_set_block(entries, 1);
2027 	dx_set_count(entries, 1);
2028 	dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2029 
2030 	/* Initialize as for dx_probe */
2031 	fname->hinfo.hash_version = root->info.hash_version;
2032 	if (fname->hinfo.hash_version <= DX_HASH_TEA)
2033 		fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2034 	fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2035 	ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2036 
2037 	memset(frames, 0, sizeof(frames));
2038 	frame = frames;
2039 	frame->entries = entries;
2040 	frame->at = entries;
2041 	frame->bh = bh;
2042 	bh = bh2;
2043 
2044 	retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2045 	if (retval)
2046 		goto out_frames;
2047 	retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
2048 	if (retval)
2049 		goto out_frames;
2050 
2051 	de = do_split(handle,dir, &bh, frame, &fname->hinfo);
2052 	if (IS_ERR(de)) {
2053 		retval = PTR_ERR(de);
2054 		goto out_frames;
2055 	}
2056 	dx_release(frames);
2057 
2058 	retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2059 	brelse(bh);
2060 	return retval;
2061 out_frames:
2062 	/*
2063 	 * Even if the block split failed, we have to properly write
2064 	 * out all the changes we did so far. Otherwise we can end up
2065 	 * with corrupted filesystem.
2066 	 */
2067 	ext4_mark_inode_dirty(handle, dir);
2068 	dx_release(frames);
2069 	return retval;
2070 }
2071 
2072 /*
2073  *	ext4_add_entry()
2074  *
2075  * adds a file entry to the specified directory, using the same
2076  * semantics as ext4_find_entry(). It returns NULL if it failed.
2077  *
2078  * NOTE!! The inode part of 'de' is left at 0 - which means you
2079  * may not sleep between calling this and putting something into
2080  * the entry, as someone else might have used it while you slept.
2081  */
2082 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2083 			  struct inode *inode)
2084 {
2085 	struct inode *dir = d_inode(dentry->d_parent);
2086 	struct buffer_head *bh = NULL;
2087 	struct ext4_dir_entry_2 *de;
2088 	struct ext4_dir_entry_tail *t;
2089 	struct super_block *sb;
2090 	struct ext4_filename fname;
2091 	int	retval;
2092 	int	dx_fallback=0;
2093 	unsigned blocksize;
2094 	ext4_lblk_t block, blocks;
2095 	int	csum_size = 0;
2096 
2097 	if (ext4_has_metadata_csum(inode->i_sb))
2098 		csum_size = sizeof(struct ext4_dir_entry_tail);
2099 
2100 	sb = dir->i_sb;
2101 	blocksize = sb->s_blocksize;
2102 	if (!dentry->d_name.len)
2103 		return -EINVAL;
2104 
2105 	retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2106 	if (retval)
2107 		return retval;
2108 
2109 	if (ext4_has_inline_data(dir)) {
2110 		retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2111 		if (retval < 0)
2112 			goto out;
2113 		if (retval == 1) {
2114 			retval = 0;
2115 			goto out;
2116 		}
2117 	}
2118 
2119 	if (is_dx(dir)) {
2120 		retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2121 		if (!retval || (retval != ERR_BAD_DX_DIR))
2122 			goto out;
2123 		ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2124 		dx_fallback++;
2125 		ext4_mark_inode_dirty(handle, dir);
2126 	}
2127 	blocks = dir->i_size >> sb->s_blocksize_bits;
2128 	for (block = 0; block < blocks; block++) {
2129 		bh = ext4_read_dirblock(dir, block, DIRENT);
2130 		if (IS_ERR(bh)) {
2131 			retval = PTR_ERR(bh);
2132 			bh = NULL;
2133 			goto out;
2134 		}
2135 		retval = add_dirent_to_buf(handle, &fname, dir, inode,
2136 					   NULL, bh);
2137 		if (retval != -ENOSPC)
2138 			goto out;
2139 
2140 		if (blocks == 1 && !dx_fallback &&
2141 		    ext4_has_feature_dir_index(sb)) {
2142 			retval = make_indexed_dir(handle, &fname, dir,
2143 						  inode, bh);
2144 			bh = NULL; /* make_indexed_dir releases bh */
2145 			goto out;
2146 		}
2147 		brelse(bh);
2148 	}
2149 	bh = ext4_append(handle, dir, &block);
2150 	if (IS_ERR(bh)) {
2151 		retval = PTR_ERR(bh);
2152 		bh = NULL;
2153 		goto out;
2154 	}
2155 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2156 	de->inode = 0;
2157 	de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2158 
2159 	if (csum_size) {
2160 		t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2161 		initialize_dirent_tail(t, blocksize);
2162 	}
2163 
2164 	retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2165 out:
2166 	ext4_fname_free_filename(&fname);
2167 	brelse(bh);
2168 	if (retval == 0)
2169 		ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2170 	return retval;
2171 }
2172 
2173 /*
2174  * Returns 0 for success, or a negative error value
2175  */
2176 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2177 			     struct inode *dir, struct inode *inode)
2178 {
2179 	struct dx_frame frames[2], *frame;
2180 	struct dx_entry *entries, *at;
2181 	struct buffer_head *bh;
2182 	struct super_block *sb = dir->i_sb;
2183 	struct ext4_dir_entry_2 *de;
2184 	int err;
2185 
2186 	frame = dx_probe(fname, dir, NULL, frames);
2187 	if (IS_ERR(frame))
2188 		return PTR_ERR(frame);
2189 	entries = frame->entries;
2190 	at = frame->at;
2191 	bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2192 	if (IS_ERR(bh)) {
2193 		err = PTR_ERR(bh);
2194 		bh = NULL;
2195 		goto cleanup;
2196 	}
2197 
2198 	BUFFER_TRACE(bh, "get_write_access");
2199 	err = ext4_journal_get_write_access(handle, bh);
2200 	if (err)
2201 		goto journal_error;
2202 
2203 	err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2204 	if (err != -ENOSPC)
2205 		goto cleanup;
2206 
2207 	/* Block full, should compress but for now just split */
2208 	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2209 		       dx_get_count(entries), dx_get_limit(entries)));
2210 	/* Need to split index? */
2211 	if (dx_get_count(entries) == dx_get_limit(entries)) {
2212 		ext4_lblk_t newblock;
2213 		unsigned icount = dx_get_count(entries);
2214 		int levels = frame - frames;
2215 		struct dx_entry *entries2;
2216 		struct dx_node *node2;
2217 		struct buffer_head *bh2;
2218 
2219 		if (levels && (dx_get_count(frames->entries) ==
2220 			       dx_get_limit(frames->entries))) {
2221 			ext4_warning_inode(dir, "Directory index full!");
2222 			err = -ENOSPC;
2223 			goto cleanup;
2224 		}
2225 		bh2 = ext4_append(handle, dir, &newblock);
2226 		if (IS_ERR(bh2)) {
2227 			err = PTR_ERR(bh2);
2228 			goto cleanup;
2229 		}
2230 		node2 = (struct dx_node *)(bh2->b_data);
2231 		entries2 = node2->entries;
2232 		memset(&node2->fake, 0, sizeof(struct fake_dirent));
2233 		node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2234 							   sb->s_blocksize);
2235 		BUFFER_TRACE(frame->bh, "get_write_access");
2236 		err = ext4_journal_get_write_access(handle, frame->bh);
2237 		if (err)
2238 			goto journal_error;
2239 		if (levels) {
2240 			unsigned icount1 = icount/2, icount2 = icount - icount1;
2241 			unsigned hash2 = dx_get_hash(entries + icount1);
2242 			dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2243 				       icount1, icount2));
2244 
2245 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2246 			err = ext4_journal_get_write_access(handle,
2247 							     frames[0].bh);
2248 			if (err)
2249 				goto journal_error;
2250 
2251 			memcpy((char *) entries2, (char *) (entries + icount1),
2252 			       icount2 * sizeof(struct dx_entry));
2253 			dx_set_count(entries, icount1);
2254 			dx_set_count(entries2, icount2);
2255 			dx_set_limit(entries2, dx_node_limit(dir));
2256 
2257 			/* Which index block gets the new entry? */
2258 			if (at - entries >= icount1) {
2259 				frame->at = at = at - entries - icount1 + entries2;
2260 				frame->entries = entries = entries2;
2261 				swap(frame->bh, bh2);
2262 			}
2263 			dx_insert_block(frames + 0, hash2, newblock);
2264 			dxtrace(dx_show_index("node", frames[1].entries));
2265 			dxtrace(dx_show_index("node",
2266 			       ((struct dx_node *) bh2->b_data)->entries));
2267 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2268 			if (err)
2269 				goto journal_error;
2270 			brelse (bh2);
2271 		} else {
2272 			dxtrace(printk(KERN_DEBUG
2273 				       "Creating second level index...\n"));
2274 			memcpy((char *) entries2, (char *) entries,
2275 			       icount * sizeof(struct dx_entry));
2276 			dx_set_limit(entries2, dx_node_limit(dir));
2277 
2278 			/* Set up root */
2279 			dx_set_count(entries, 1);
2280 			dx_set_block(entries + 0, newblock);
2281 			((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2282 
2283 			/* Add new access path frame */
2284 			frame = frames + 1;
2285 			frame->at = at = at - entries + entries2;
2286 			frame->entries = entries = entries2;
2287 			frame->bh = bh2;
2288 			err = ext4_journal_get_write_access(handle,
2289 							     frame->bh);
2290 			if (err)
2291 				goto journal_error;
2292 		}
2293 		err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2294 		if (err) {
2295 			ext4_std_error(inode->i_sb, err);
2296 			goto cleanup;
2297 		}
2298 	}
2299 	de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2300 	if (IS_ERR(de)) {
2301 		err = PTR_ERR(de);
2302 		goto cleanup;
2303 	}
2304 	err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2305 	goto cleanup;
2306 
2307 journal_error:
2308 	ext4_std_error(dir->i_sb, err);
2309 cleanup:
2310 	brelse(bh);
2311 	dx_release(frames);
2312 	return err;
2313 }
2314 
2315 /*
2316  * ext4_generic_delete_entry deletes a directory entry by merging it
2317  * with the previous entry
2318  */
2319 int ext4_generic_delete_entry(handle_t *handle,
2320 			      struct inode *dir,
2321 			      struct ext4_dir_entry_2 *de_del,
2322 			      struct buffer_head *bh,
2323 			      void *entry_buf,
2324 			      int buf_size,
2325 			      int csum_size)
2326 {
2327 	struct ext4_dir_entry_2 *de, *pde;
2328 	unsigned int blocksize = dir->i_sb->s_blocksize;
2329 	int i;
2330 
2331 	i = 0;
2332 	pde = NULL;
2333 	de = (struct ext4_dir_entry_2 *)entry_buf;
2334 	while (i < buf_size - csum_size) {
2335 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2336 					 bh->b_data, bh->b_size, i))
2337 			return -EFSCORRUPTED;
2338 		if (de == de_del)  {
2339 			if (pde)
2340 				pde->rec_len = ext4_rec_len_to_disk(
2341 					ext4_rec_len_from_disk(pde->rec_len,
2342 							       blocksize) +
2343 					ext4_rec_len_from_disk(de->rec_len,
2344 							       blocksize),
2345 					blocksize);
2346 			else
2347 				de->inode = 0;
2348 			dir->i_version++;
2349 			return 0;
2350 		}
2351 		i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2352 		pde = de;
2353 		de = ext4_next_entry(de, blocksize);
2354 	}
2355 	return -ENOENT;
2356 }
2357 
2358 static int ext4_delete_entry(handle_t *handle,
2359 			     struct inode *dir,
2360 			     struct ext4_dir_entry_2 *de_del,
2361 			     struct buffer_head *bh)
2362 {
2363 	int err, csum_size = 0;
2364 
2365 	if (ext4_has_inline_data(dir)) {
2366 		int has_inline_data = 1;
2367 		err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2368 					       &has_inline_data);
2369 		if (has_inline_data)
2370 			return err;
2371 	}
2372 
2373 	if (ext4_has_metadata_csum(dir->i_sb))
2374 		csum_size = sizeof(struct ext4_dir_entry_tail);
2375 
2376 	BUFFER_TRACE(bh, "get_write_access");
2377 	err = ext4_journal_get_write_access(handle, bh);
2378 	if (unlikely(err))
2379 		goto out;
2380 
2381 	err = ext4_generic_delete_entry(handle, dir, de_del,
2382 					bh, bh->b_data,
2383 					dir->i_sb->s_blocksize, csum_size);
2384 	if (err)
2385 		goto out;
2386 
2387 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2388 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2389 	if (unlikely(err))
2390 		goto out;
2391 
2392 	return 0;
2393 out:
2394 	if (err != -ENOENT)
2395 		ext4_std_error(dir->i_sb, err);
2396 	return err;
2397 }
2398 
2399 /*
2400  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2401  * since this indicates that nlinks count was previously 1.
2402  */
2403 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2404 {
2405 	inc_nlink(inode);
2406 	if (is_dx(inode) && inode->i_nlink > 1) {
2407 		/* limit is 16-bit i_links_count */
2408 		if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2409 			set_nlink(inode, 1);
2410 			ext4_set_feature_dir_nlink(inode->i_sb);
2411 		}
2412 	}
2413 }
2414 
2415 /*
2416  * If a directory had nlink == 1, then we should let it be 1. This indicates
2417  * directory has >EXT4_LINK_MAX subdirs.
2418  */
2419 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2420 {
2421 	if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2422 		drop_nlink(inode);
2423 }
2424 
2425 
2426 static int ext4_add_nondir(handle_t *handle,
2427 		struct dentry *dentry, struct inode *inode)
2428 {
2429 	int err = ext4_add_entry(handle, dentry, inode);
2430 	if (!err) {
2431 		ext4_mark_inode_dirty(handle, inode);
2432 		unlock_new_inode(inode);
2433 		d_instantiate(dentry, inode);
2434 		return 0;
2435 	}
2436 	drop_nlink(inode);
2437 	unlock_new_inode(inode);
2438 	iput(inode);
2439 	return err;
2440 }
2441 
2442 /*
2443  * By the time this is called, we already have created
2444  * the directory cache entry for the new file, but it
2445  * is so far negative - it has no inode.
2446  *
2447  * If the create succeeds, we fill in the inode information
2448  * with d_instantiate().
2449  */
2450 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2451 		       bool excl)
2452 {
2453 	handle_t *handle;
2454 	struct inode *inode;
2455 	int err, credits, retries = 0;
2456 
2457 	err = dquot_initialize(dir);
2458 	if (err)
2459 		return err;
2460 
2461 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2462 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2463 retry:
2464 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2465 					    NULL, EXT4_HT_DIR, credits);
2466 	handle = ext4_journal_current_handle();
2467 	err = PTR_ERR(inode);
2468 	if (!IS_ERR(inode)) {
2469 		inode->i_op = &ext4_file_inode_operations;
2470 		inode->i_fop = &ext4_file_operations;
2471 		ext4_set_aops(inode);
2472 		err = ext4_add_nondir(handle, dentry, inode);
2473 		if (!err && IS_DIRSYNC(dir))
2474 			ext4_handle_sync(handle);
2475 	}
2476 	if (handle)
2477 		ext4_journal_stop(handle);
2478 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2479 		goto retry;
2480 	return err;
2481 }
2482 
2483 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2484 		      umode_t mode, dev_t rdev)
2485 {
2486 	handle_t *handle;
2487 	struct inode *inode;
2488 	int err, credits, retries = 0;
2489 
2490 	err = dquot_initialize(dir);
2491 	if (err)
2492 		return err;
2493 
2494 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2495 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2496 retry:
2497 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2498 					    NULL, EXT4_HT_DIR, credits);
2499 	handle = ext4_journal_current_handle();
2500 	err = PTR_ERR(inode);
2501 	if (!IS_ERR(inode)) {
2502 		init_special_inode(inode, inode->i_mode, rdev);
2503 		inode->i_op = &ext4_special_inode_operations;
2504 		err = ext4_add_nondir(handle, dentry, inode);
2505 		if (!err && IS_DIRSYNC(dir))
2506 			ext4_handle_sync(handle);
2507 	}
2508 	if (handle)
2509 		ext4_journal_stop(handle);
2510 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2511 		goto retry;
2512 	return err;
2513 }
2514 
2515 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2516 {
2517 	handle_t *handle;
2518 	struct inode *inode;
2519 	int err, retries = 0;
2520 
2521 	err = dquot_initialize(dir);
2522 	if (err)
2523 		return err;
2524 
2525 retry:
2526 	inode = ext4_new_inode_start_handle(dir, mode,
2527 					    NULL, 0, NULL,
2528 					    EXT4_HT_DIR,
2529 			EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2530 			  4 + EXT4_XATTR_TRANS_BLOCKS);
2531 	handle = ext4_journal_current_handle();
2532 	err = PTR_ERR(inode);
2533 	if (!IS_ERR(inode)) {
2534 		inode->i_op = &ext4_file_inode_operations;
2535 		inode->i_fop = &ext4_file_operations;
2536 		ext4_set_aops(inode);
2537 		d_tmpfile(dentry, inode);
2538 		err = ext4_orphan_add(handle, inode);
2539 		if (err)
2540 			goto err_unlock_inode;
2541 		mark_inode_dirty(inode);
2542 		unlock_new_inode(inode);
2543 	}
2544 	if (handle)
2545 		ext4_journal_stop(handle);
2546 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2547 		goto retry;
2548 	return err;
2549 err_unlock_inode:
2550 	ext4_journal_stop(handle);
2551 	unlock_new_inode(inode);
2552 	return err;
2553 }
2554 
2555 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2556 			  struct ext4_dir_entry_2 *de,
2557 			  int blocksize, int csum_size,
2558 			  unsigned int parent_ino, int dotdot_real_len)
2559 {
2560 	de->inode = cpu_to_le32(inode->i_ino);
2561 	de->name_len = 1;
2562 	de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2563 					   blocksize);
2564 	strcpy(de->name, ".");
2565 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2566 
2567 	de = ext4_next_entry(de, blocksize);
2568 	de->inode = cpu_to_le32(parent_ino);
2569 	de->name_len = 2;
2570 	if (!dotdot_real_len)
2571 		de->rec_len = ext4_rec_len_to_disk(blocksize -
2572 					(csum_size + EXT4_DIR_REC_LEN(1)),
2573 					blocksize);
2574 	else
2575 		de->rec_len = ext4_rec_len_to_disk(
2576 				EXT4_DIR_REC_LEN(de->name_len), blocksize);
2577 	strcpy(de->name, "..");
2578 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2579 
2580 	return ext4_next_entry(de, blocksize);
2581 }
2582 
2583 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2584 			     struct inode *inode)
2585 {
2586 	struct buffer_head *dir_block = NULL;
2587 	struct ext4_dir_entry_2 *de;
2588 	struct ext4_dir_entry_tail *t;
2589 	ext4_lblk_t block = 0;
2590 	unsigned int blocksize = dir->i_sb->s_blocksize;
2591 	int csum_size = 0;
2592 	int err;
2593 
2594 	if (ext4_has_metadata_csum(dir->i_sb))
2595 		csum_size = sizeof(struct ext4_dir_entry_tail);
2596 
2597 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2598 		err = ext4_try_create_inline_dir(handle, dir, inode);
2599 		if (err < 0 && err != -ENOSPC)
2600 			goto out;
2601 		if (!err)
2602 			goto out;
2603 	}
2604 
2605 	inode->i_size = 0;
2606 	dir_block = ext4_append(handle, inode, &block);
2607 	if (IS_ERR(dir_block))
2608 		return PTR_ERR(dir_block);
2609 	de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2610 	ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2611 	set_nlink(inode, 2);
2612 	if (csum_size) {
2613 		t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2614 		initialize_dirent_tail(t, blocksize);
2615 	}
2616 
2617 	BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2618 	err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2619 	if (err)
2620 		goto out;
2621 	set_buffer_verified(dir_block);
2622 out:
2623 	brelse(dir_block);
2624 	return err;
2625 }
2626 
2627 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2628 {
2629 	handle_t *handle;
2630 	struct inode *inode;
2631 	int err, credits, retries = 0;
2632 
2633 	if (EXT4_DIR_LINK_MAX(dir))
2634 		return -EMLINK;
2635 
2636 	err = dquot_initialize(dir);
2637 	if (err)
2638 		return err;
2639 
2640 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2641 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2642 retry:
2643 	inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2644 					    &dentry->d_name,
2645 					    0, NULL, EXT4_HT_DIR, credits);
2646 	handle = ext4_journal_current_handle();
2647 	err = PTR_ERR(inode);
2648 	if (IS_ERR(inode))
2649 		goto out_stop;
2650 
2651 	inode->i_op = &ext4_dir_inode_operations;
2652 	inode->i_fop = &ext4_dir_operations;
2653 	err = ext4_init_new_dir(handle, dir, inode);
2654 	if (err)
2655 		goto out_clear_inode;
2656 	err = ext4_mark_inode_dirty(handle, inode);
2657 	if (!err)
2658 		err = ext4_add_entry(handle, dentry, inode);
2659 	if (err) {
2660 out_clear_inode:
2661 		clear_nlink(inode);
2662 		unlock_new_inode(inode);
2663 		ext4_mark_inode_dirty(handle, inode);
2664 		iput(inode);
2665 		goto out_stop;
2666 	}
2667 	ext4_inc_count(handle, dir);
2668 	ext4_update_dx_flag(dir);
2669 	err = ext4_mark_inode_dirty(handle, dir);
2670 	if (err)
2671 		goto out_clear_inode;
2672 	unlock_new_inode(inode);
2673 	d_instantiate(dentry, inode);
2674 	if (IS_DIRSYNC(dir))
2675 		ext4_handle_sync(handle);
2676 
2677 out_stop:
2678 	if (handle)
2679 		ext4_journal_stop(handle);
2680 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2681 		goto retry;
2682 	return err;
2683 }
2684 
2685 /*
2686  * routine to check that the specified directory is empty (for rmdir)
2687  */
2688 int ext4_empty_dir(struct inode *inode)
2689 {
2690 	unsigned int offset;
2691 	struct buffer_head *bh;
2692 	struct ext4_dir_entry_2 *de, *de1;
2693 	struct super_block *sb;
2694 	int err = 0;
2695 
2696 	if (ext4_has_inline_data(inode)) {
2697 		int has_inline_data = 1;
2698 
2699 		err = empty_inline_dir(inode, &has_inline_data);
2700 		if (has_inline_data)
2701 			return err;
2702 	}
2703 
2704 	sb = inode->i_sb;
2705 	if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2706 		EXT4_ERROR_INODE(inode, "invalid size");
2707 		return 1;
2708 	}
2709 	bh = ext4_read_dirblock(inode, 0, EITHER);
2710 	if (IS_ERR(bh))
2711 		return 1;
2712 
2713 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2714 	de1 = ext4_next_entry(de, sb->s_blocksize);
2715 	if (le32_to_cpu(de->inode) != inode->i_ino ||
2716 			le32_to_cpu(de1->inode) == 0 ||
2717 			strcmp(".", de->name) || strcmp("..", de1->name)) {
2718 		ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2719 		brelse(bh);
2720 		return 1;
2721 	}
2722 	offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2723 		 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2724 	de = ext4_next_entry(de1, sb->s_blocksize);
2725 	while (offset < inode->i_size) {
2726 		if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2727 			unsigned int lblock;
2728 			err = 0;
2729 			brelse(bh);
2730 			lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2731 			bh = ext4_read_dirblock(inode, lblock, EITHER);
2732 			if (IS_ERR(bh))
2733 				return 1;
2734 			de = (struct ext4_dir_entry_2 *) bh->b_data;
2735 		}
2736 		if (ext4_check_dir_entry(inode, NULL, de, bh,
2737 					 bh->b_data, bh->b_size, offset)) {
2738 			de = (struct ext4_dir_entry_2 *)(bh->b_data +
2739 							 sb->s_blocksize);
2740 			offset = (offset | (sb->s_blocksize - 1)) + 1;
2741 			continue;
2742 		}
2743 		if (le32_to_cpu(de->inode)) {
2744 			brelse(bh);
2745 			return 0;
2746 		}
2747 		offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2748 		de = ext4_next_entry(de, sb->s_blocksize);
2749 	}
2750 	brelse(bh);
2751 	return 1;
2752 }
2753 
2754 /*
2755  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2756  * such inodes, starting at the superblock, in case we crash before the
2757  * file is closed/deleted, or in case the inode truncate spans multiple
2758  * transactions and the last transaction is not recovered after a crash.
2759  *
2760  * At filesystem recovery time, we walk this list deleting unlinked
2761  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2762  *
2763  * Orphan list manipulation functions must be called under i_mutex unless
2764  * we are just creating the inode or deleting it.
2765  */
2766 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2767 {
2768 	struct super_block *sb = inode->i_sb;
2769 	struct ext4_sb_info *sbi = EXT4_SB(sb);
2770 	struct ext4_iloc iloc;
2771 	int err = 0, rc;
2772 	bool dirty = false;
2773 
2774 	if (!sbi->s_journal || is_bad_inode(inode))
2775 		return 0;
2776 
2777 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2778 		     !inode_is_locked(inode));
2779 	/*
2780 	 * Exit early if inode already is on orphan list. This is a big speedup
2781 	 * since we don't have to contend on the global s_orphan_lock.
2782 	 */
2783 	if (!list_empty(&EXT4_I(inode)->i_orphan))
2784 		return 0;
2785 
2786 	/*
2787 	 * Orphan handling is only valid for files with data blocks
2788 	 * being truncated, or files being unlinked. Note that we either
2789 	 * hold i_mutex, or the inode can not be referenced from outside,
2790 	 * so i_nlink should not be bumped due to race
2791 	 */
2792 	J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2793 		  S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2794 
2795 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2796 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2797 	if (err)
2798 		goto out;
2799 
2800 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2801 	if (err)
2802 		goto out;
2803 
2804 	mutex_lock(&sbi->s_orphan_lock);
2805 	/*
2806 	 * Due to previous errors inode may be already a part of on-disk
2807 	 * orphan list. If so skip on-disk list modification.
2808 	 */
2809 	if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2810 	    (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2811 		/* Insert this inode at the head of the on-disk orphan list */
2812 		NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2813 		sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2814 		dirty = true;
2815 	}
2816 	list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2817 	mutex_unlock(&sbi->s_orphan_lock);
2818 
2819 	if (dirty) {
2820 		err = ext4_handle_dirty_super(handle, sb);
2821 		rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2822 		if (!err)
2823 			err = rc;
2824 		if (err) {
2825 			/*
2826 			 * We have to remove inode from in-memory list if
2827 			 * addition to on disk orphan list failed. Stray orphan
2828 			 * list entries can cause panics at unmount time.
2829 			 */
2830 			mutex_lock(&sbi->s_orphan_lock);
2831 			list_del(&EXT4_I(inode)->i_orphan);
2832 			mutex_unlock(&sbi->s_orphan_lock);
2833 		}
2834 	}
2835 	jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2836 	jbd_debug(4, "orphan inode %lu will point to %d\n",
2837 			inode->i_ino, NEXT_ORPHAN(inode));
2838 out:
2839 	ext4_std_error(sb, err);
2840 	return err;
2841 }
2842 
2843 /*
2844  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2845  * of such inodes stored on disk, because it is finally being cleaned up.
2846  */
2847 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2848 {
2849 	struct list_head *prev;
2850 	struct ext4_inode_info *ei = EXT4_I(inode);
2851 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2852 	__u32 ino_next;
2853 	struct ext4_iloc iloc;
2854 	int err = 0;
2855 
2856 	if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2857 		return 0;
2858 
2859 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2860 		     !inode_is_locked(inode));
2861 	/* Do this quick check before taking global s_orphan_lock. */
2862 	if (list_empty(&ei->i_orphan))
2863 		return 0;
2864 
2865 	if (handle) {
2866 		/* Grab inode buffer early before taking global s_orphan_lock */
2867 		err = ext4_reserve_inode_write(handle, inode, &iloc);
2868 	}
2869 
2870 	mutex_lock(&sbi->s_orphan_lock);
2871 	jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2872 
2873 	prev = ei->i_orphan.prev;
2874 	list_del_init(&ei->i_orphan);
2875 
2876 	/* If we're on an error path, we may not have a valid
2877 	 * transaction handle with which to update the orphan list on
2878 	 * disk, but we still need to remove the inode from the linked
2879 	 * list in memory. */
2880 	if (!handle || err) {
2881 		mutex_unlock(&sbi->s_orphan_lock);
2882 		goto out_err;
2883 	}
2884 
2885 	ino_next = NEXT_ORPHAN(inode);
2886 	if (prev == &sbi->s_orphan) {
2887 		jbd_debug(4, "superblock will point to %u\n", ino_next);
2888 		BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2889 		err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2890 		if (err) {
2891 			mutex_unlock(&sbi->s_orphan_lock);
2892 			goto out_brelse;
2893 		}
2894 		sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2895 		mutex_unlock(&sbi->s_orphan_lock);
2896 		err = ext4_handle_dirty_super(handle, inode->i_sb);
2897 	} else {
2898 		struct ext4_iloc iloc2;
2899 		struct inode *i_prev =
2900 			&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2901 
2902 		jbd_debug(4, "orphan inode %lu will point to %u\n",
2903 			  i_prev->i_ino, ino_next);
2904 		err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2905 		if (err) {
2906 			mutex_unlock(&sbi->s_orphan_lock);
2907 			goto out_brelse;
2908 		}
2909 		NEXT_ORPHAN(i_prev) = ino_next;
2910 		err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2911 		mutex_unlock(&sbi->s_orphan_lock);
2912 	}
2913 	if (err)
2914 		goto out_brelse;
2915 	NEXT_ORPHAN(inode) = 0;
2916 	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2917 out_err:
2918 	ext4_std_error(inode->i_sb, err);
2919 	return err;
2920 
2921 out_brelse:
2922 	brelse(iloc.bh);
2923 	goto out_err;
2924 }
2925 
2926 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2927 {
2928 	int retval;
2929 	struct inode *inode;
2930 	struct buffer_head *bh;
2931 	struct ext4_dir_entry_2 *de;
2932 	handle_t *handle = NULL;
2933 
2934 	/* Initialize quotas before so that eventual writes go in
2935 	 * separate transaction */
2936 	retval = dquot_initialize(dir);
2937 	if (retval)
2938 		return retval;
2939 	retval = dquot_initialize(d_inode(dentry));
2940 	if (retval)
2941 		return retval;
2942 
2943 	retval = -ENOENT;
2944 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2945 	if (IS_ERR(bh))
2946 		return PTR_ERR(bh);
2947 	if (!bh)
2948 		goto end_rmdir;
2949 
2950 	inode = d_inode(dentry);
2951 
2952 	retval = -EFSCORRUPTED;
2953 	if (le32_to_cpu(de->inode) != inode->i_ino)
2954 		goto end_rmdir;
2955 
2956 	retval = -ENOTEMPTY;
2957 	if (!ext4_empty_dir(inode))
2958 		goto end_rmdir;
2959 
2960 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
2961 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2962 	if (IS_ERR(handle)) {
2963 		retval = PTR_ERR(handle);
2964 		handle = NULL;
2965 		goto end_rmdir;
2966 	}
2967 
2968 	if (IS_DIRSYNC(dir))
2969 		ext4_handle_sync(handle);
2970 
2971 	retval = ext4_delete_entry(handle, dir, de, bh);
2972 	if (retval)
2973 		goto end_rmdir;
2974 	if (!EXT4_DIR_LINK_EMPTY(inode))
2975 		ext4_warning_inode(inode,
2976 			     "empty directory '%.*s' has too many links (%u)",
2977 			     dentry->d_name.len, dentry->d_name.name,
2978 			     inode->i_nlink);
2979 	inode->i_version++;
2980 	clear_nlink(inode);
2981 	/* There's no need to set i_disksize: the fact that i_nlink is
2982 	 * zero will ensure that the right thing happens during any
2983 	 * recovery. */
2984 	inode->i_size = 0;
2985 	ext4_orphan_add(handle, inode);
2986 	inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2987 	ext4_mark_inode_dirty(handle, inode);
2988 	ext4_dec_count(handle, dir);
2989 	ext4_update_dx_flag(dir);
2990 	ext4_mark_inode_dirty(handle, dir);
2991 
2992 end_rmdir:
2993 	brelse(bh);
2994 	if (handle)
2995 		ext4_journal_stop(handle);
2996 	return retval;
2997 }
2998 
2999 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3000 {
3001 	int retval;
3002 	struct inode *inode;
3003 	struct buffer_head *bh;
3004 	struct ext4_dir_entry_2 *de;
3005 	handle_t *handle = NULL;
3006 
3007 	trace_ext4_unlink_enter(dir, dentry);
3008 	/* Initialize quotas before so that eventual writes go
3009 	 * in separate transaction */
3010 	retval = dquot_initialize(dir);
3011 	if (retval)
3012 		return retval;
3013 	retval = dquot_initialize(d_inode(dentry));
3014 	if (retval)
3015 		return retval;
3016 
3017 	retval = -ENOENT;
3018 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3019 	if (IS_ERR(bh))
3020 		return PTR_ERR(bh);
3021 	if (!bh)
3022 		goto end_unlink;
3023 
3024 	inode = d_inode(dentry);
3025 
3026 	retval = -EFSCORRUPTED;
3027 	if (le32_to_cpu(de->inode) != inode->i_ino)
3028 		goto end_unlink;
3029 
3030 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3031 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3032 	if (IS_ERR(handle)) {
3033 		retval = PTR_ERR(handle);
3034 		handle = NULL;
3035 		goto end_unlink;
3036 	}
3037 
3038 	if (IS_DIRSYNC(dir))
3039 		ext4_handle_sync(handle);
3040 
3041 	if (inode->i_nlink == 0) {
3042 		ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3043 				   dentry->d_name.len, dentry->d_name.name);
3044 		set_nlink(inode, 1);
3045 	}
3046 	retval = ext4_delete_entry(handle, dir, de, bh);
3047 	if (retval)
3048 		goto end_unlink;
3049 	dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3050 	ext4_update_dx_flag(dir);
3051 	ext4_mark_inode_dirty(handle, dir);
3052 	drop_nlink(inode);
3053 	if (!inode->i_nlink)
3054 		ext4_orphan_add(handle, inode);
3055 	inode->i_ctime = ext4_current_time(inode);
3056 	ext4_mark_inode_dirty(handle, inode);
3057 
3058 end_unlink:
3059 	brelse(bh);
3060 	if (handle)
3061 		ext4_journal_stop(handle);
3062 	trace_ext4_unlink_exit(dentry, retval);
3063 	return retval;
3064 }
3065 
3066 static int ext4_symlink(struct inode *dir,
3067 			struct dentry *dentry, const char *symname)
3068 {
3069 	handle_t *handle;
3070 	struct inode *inode;
3071 	int err, len = strlen(symname);
3072 	int credits;
3073 	bool encryption_required;
3074 	struct ext4_str disk_link;
3075 	struct ext4_encrypted_symlink_data *sd = NULL;
3076 
3077 	disk_link.len = len + 1;
3078 	disk_link.name = (char *) symname;
3079 
3080 	encryption_required = (ext4_encrypted_inode(dir) ||
3081 			       DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3082 	if (encryption_required) {
3083 		err = ext4_get_encryption_info(dir);
3084 		if (err)
3085 			return err;
3086 		if (ext4_encryption_info(dir) == NULL)
3087 			return -EPERM;
3088 		disk_link.len = (ext4_fname_encrypted_size(dir, len) +
3089 				 sizeof(struct ext4_encrypted_symlink_data));
3090 		sd = kzalloc(disk_link.len, GFP_KERNEL);
3091 		if (!sd)
3092 			return -ENOMEM;
3093 	}
3094 
3095 	if (disk_link.len > dir->i_sb->s_blocksize) {
3096 		err = -ENAMETOOLONG;
3097 		goto err_free_sd;
3098 	}
3099 
3100 	err = dquot_initialize(dir);
3101 	if (err)
3102 		goto err_free_sd;
3103 
3104 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3105 		/*
3106 		 * For non-fast symlinks, we just allocate inode and put it on
3107 		 * orphan list in the first transaction => we need bitmap,
3108 		 * group descriptor, sb, inode block, quota blocks, and
3109 		 * possibly selinux xattr blocks.
3110 		 */
3111 		credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3112 			  EXT4_XATTR_TRANS_BLOCKS;
3113 	} else {
3114 		/*
3115 		 * Fast symlink. We have to add entry to directory
3116 		 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3117 		 * allocate new inode (bitmap, group descriptor, inode block,
3118 		 * quota blocks, sb is already counted in previous macros).
3119 		 */
3120 		credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3121 			  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3122 	}
3123 
3124 	inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3125 					    &dentry->d_name, 0, NULL,
3126 					    EXT4_HT_DIR, credits);
3127 	handle = ext4_journal_current_handle();
3128 	if (IS_ERR(inode)) {
3129 		if (handle)
3130 			ext4_journal_stop(handle);
3131 		err = PTR_ERR(inode);
3132 		goto err_free_sd;
3133 	}
3134 
3135 	if (encryption_required) {
3136 		struct qstr istr;
3137 		struct ext4_str ostr;
3138 
3139 		istr.name = (const unsigned char *) symname;
3140 		istr.len = len;
3141 		ostr.name = sd->encrypted_path;
3142 		ostr.len = disk_link.len;
3143 		err = ext4_fname_usr_to_disk(inode, &istr, &ostr);
3144 		if (err < 0)
3145 			goto err_drop_inode;
3146 		sd->len = cpu_to_le16(ostr.len);
3147 		disk_link.name = (char *) sd;
3148 		inode->i_op = &ext4_encrypted_symlink_inode_operations;
3149 	}
3150 
3151 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3152 		if (!encryption_required)
3153 			inode->i_op = &ext4_symlink_inode_operations;
3154 		inode_nohighmem(inode);
3155 		ext4_set_aops(inode);
3156 		/*
3157 		 * We cannot call page_symlink() with transaction started
3158 		 * because it calls into ext4_write_begin() which can wait
3159 		 * for transaction commit if we are running out of space
3160 		 * and thus we deadlock. So we have to stop transaction now
3161 		 * and restart it when symlink contents is written.
3162 		 *
3163 		 * To keep fs consistent in case of crash, we have to put inode
3164 		 * to orphan list in the mean time.
3165 		 */
3166 		drop_nlink(inode);
3167 		err = ext4_orphan_add(handle, inode);
3168 		ext4_journal_stop(handle);
3169 		handle = NULL;
3170 		if (err)
3171 			goto err_drop_inode;
3172 		err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3173 		if (err)
3174 			goto err_drop_inode;
3175 		/*
3176 		 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3177 		 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3178 		 */
3179 		handle = ext4_journal_start(dir, EXT4_HT_DIR,
3180 				EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3181 				EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3182 		if (IS_ERR(handle)) {
3183 			err = PTR_ERR(handle);
3184 			handle = NULL;
3185 			goto err_drop_inode;
3186 		}
3187 		set_nlink(inode, 1);
3188 		err = ext4_orphan_del(handle, inode);
3189 		if (err)
3190 			goto err_drop_inode;
3191 	} else {
3192 		/* clear the extent format for fast symlink */
3193 		ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3194 		if (!encryption_required) {
3195 			inode->i_op = &ext4_fast_symlink_inode_operations;
3196 			inode->i_link = (char *)&EXT4_I(inode)->i_data;
3197 		}
3198 		memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3199 		       disk_link.len);
3200 		inode->i_size = disk_link.len - 1;
3201 	}
3202 	EXT4_I(inode)->i_disksize = inode->i_size;
3203 	err = ext4_add_nondir(handle, dentry, inode);
3204 	if (!err && IS_DIRSYNC(dir))
3205 		ext4_handle_sync(handle);
3206 
3207 	if (handle)
3208 		ext4_journal_stop(handle);
3209 	kfree(sd);
3210 	return err;
3211 err_drop_inode:
3212 	if (handle)
3213 		ext4_journal_stop(handle);
3214 	clear_nlink(inode);
3215 	unlock_new_inode(inode);
3216 	iput(inode);
3217 err_free_sd:
3218 	kfree(sd);
3219 	return err;
3220 }
3221 
3222 static int ext4_link(struct dentry *old_dentry,
3223 		     struct inode *dir, struct dentry *dentry)
3224 {
3225 	handle_t *handle;
3226 	struct inode *inode = d_inode(old_dentry);
3227 	int err, retries = 0;
3228 
3229 	if (inode->i_nlink >= EXT4_LINK_MAX)
3230 		return -EMLINK;
3231 	if (ext4_encrypted_inode(dir) &&
3232 	    !ext4_is_child_context_consistent_with_parent(dir, inode))
3233 		return -EPERM;
3234 
3235        if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3236 	   (!projid_eq(EXT4_I(dir)->i_projid,
3237 		       EXT4_I(old_dentry->d_inode)->i_projid)))
3238 		return -EXDEV;
3239 
3240 	err = dquot_initialize(dir);
3241 	if (err)
3242 		return err;
3243 
3244 retry:
3245 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3246 		(EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3247 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3248 	if (IS_ERR(handle))
3249 		return PTR_ERR(handle);
3250 
3251 	if (IS_DIRSYNC(dir))
3252 		ext4_handle_sync(handle);
3253 
3254 	inode->i_ctime = ext4_current_time(inode);
3255 	ext4_inc_count(handle, inode);
3256 	ihold(inode);
3257 
3258 	err = ext4_add_entry(handle, dentry, inode);
3259 	if (!err) {
3260 		ext4_mark_inode_dirty(handle, inode);
3261 		/* this can happen only for tmpfile being
3262 		 * linked the first time
3263 		 */
3264 		if (inode->i_nlink == 1)
3265 			ext4_orphan_del(handle, inode);
3266 		d_instantiate(dentry, inode);
3267 	} else {
3268 		drop_nlink(inode);
3269 		iput(inode);
3270 	}
3271 	ext4_journal_stop(handle);
3272 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3273 		goto retry;
3274 	return err;
3275 }
3276 
3277 
3278 /*
3279  * Try to find buffer head where contains the parent block.
3280  * It should be the inode block if it is inlined or the 1st block
3281  * if it is a normal dir.
3282  */
3283 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3284 					struct inode *inode,
3285 					int *retval,
3286 					struct ext4_dir_entry_2 **parent_de,
3287 					int *inlined)
3288 {
3289 	struct buffer_head *bh;
3290 
3291 	if (!ext4_has_inline_data(inode)) {
3292 		bh = ext4_read_dirblock(inode, 0, EITHER);
3293 		if (IS_ERR(bh)) {
3294 			*retval = PTR_ERR(bh);
3295 			return NULL;
3296 		}
3297 		*parent_de = ext4_next_entry(
3298 					(struct ext4_dir_entry_2 *)bh->b_data,
3299 					inode->i_sb->s_blocksize);
3300 		return bh;
3301 	}
3302 
3303 	*inlined = 1;
3304 	return ext4_get_first_inline_block(inode, parent_de, retval);
3305 }
3306 
3307 struct ext4_renament {
3308 	struct inode *dir;
3309 	struct dentry *dentry;
3310 	struct inode *inode;
3311 	bool is_dir;
3312 	int dir_nlink_delta;
3313 
3314 	/* entry for "dentry" */
3315 	struct buffer_head *bh;
3316 	struct ext4_dir_entry_2 *de;
3317 	int inlined;
3318 
3319 	/* entry for ".." in inode if it's a directory */
3320 	struct buffer_head *dir_bh;
3321 	struct ext4_dir_entry_2 *parent_de;
3322 	int dir_inlined;
3323 };
3324 
3325 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3326 {
3327 	int retval;
3328 
3329 	ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3330 					      &retval, &ent->parent_de,
3331 					      &ent->dir_inlined);
3332 	if (!ent->dir_bh)
3333 		return retval;
3334 	if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3335 		return -EFSCORRUPTED;
3336 	BUFFER_TRACE(ent->dir_bh, "get_write_access");
3337 	return ext4_journal_get_write_access(handle, ent->dir_bh);
3338 }
3339 
3340 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3341 				  unsigned dir_ino)
3342 {
3343 	int retval;
3344 
3345 	ent->parent_de->inode = cpu_to_le32(dir_ino);
3346 	BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3347 	if (!ent->dir_inlined) {
3348 		if (is_dx(ent->inode)) {
3349 			retval = ext4_handle_dirty_dx_node(handle,
3350 							   ent->inode,
3351 							   ent->dir_bh);
3352 		} else {
3353 			retval = ext4_handle_dirty_dirent_node(handle,
3354 							       ent->inode,
3355 							       ent->dir_bh);
3356 		}
3357 	} else {
3358 		retval = ext4_mark_inode_dirty(handle, ent->inode);
3359 	}
3360 	if (retval) {
3361 		ext4_std_error(ent->dir->i_sb, retval);
3362 		return retval;
3363 	}
3364 	return 0;
3365 }
3366 
3367 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3368 		       unsigned ino, unsigned file_type)
3369 {
3370 	int retval;
3371 
3372 	BUFFER_TRACE(ent->bh, "get write access");
3373 	retval = ext4_journal_get_write_access(handle, ent->bh);
3374 	if (retval)
3375 		return retval;
3376 	ent->de->inode = cpu_to_le32(ino);
3377 	if (ext4_has_feature_filetype(ent->dir->i_sb))
3378 		ent->de->file_type = file_type;
3379 	ent->dir->i_version++;
3380 	ent->dir->i_ctime = ent->dir->i_mtime =
3381 		ext4_current_time(ent->dir);
3382 	ext4_mark_inode_dirty(handle, ent->dir);
3383 	BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3384 	if (!ent->inlined) {
3385 		retval = ext4_handle_dirty_dirent_node(handle,
3386 						       ent->dir, ent->bh);
3387 		if (unlikely(retval)) {
3388 			ext4_std_error(ent->dir->i_sb, retval);
3389 			return retval;
3390 		}
3391 	}
3392 	brelse(ent->bh);
3393 	ent->bh = NULL;
3394 
3395 	return 0;
3396 }
3397 
3398 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3399 				  const struct qstr *d_name)
3400 {
3401 	int retval = -ENOENT;
3402 	struct buffer_head *bh;
3403 	struct ext4_dir_entry_2 *de;
3404 
3405 	bh = ext4_find_entry(dir, d_name, &de, NULL);
3406 	if (IS_ERR(bh))
3407 		return PTR_ERR(bh);
3408 	if (bh) {
3409 		retval = ext4_delete_entry(handle, dir, de, bh);
3410 		brelse(bh);
3411 	}
3412 	return retval;
3413 }
3414 
3415 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3416 			       int force_reread)
3417 {
3418 	int retval;
3419 	/*
3420 	 * ent->de could have moved from under us during htree split, so make
3421 	 * sure that we are deleting the right entry.  We might also be pointing
3422 	 * to a stale entry in the unused part of ent->bh so just checking inum
3423 	 * and the name isn't enough.
3424 	 */
3425 	if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3426 	    ent->de->name_len != ent->dentry->d_name.len ||
3427 	    strncmp(ent->de->name, ent->dentry->d_name.name,
3428 		    ent->de->name_len) ||
3429 	    force_reread) {
3430 		retval = ext4_find_delete_entry(handle, ent->dir,
3431 						&ent->dentry->d_name);
3432 	} else {
3433 		retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3434 		if (retval == -ENOENT) {
3435 			retval = ext4_find_delete_entry(handle, ent->dir,
3436 							&ent->dentry->d_name);
3437 		}
3438 	}
3439 
3440 	if (retval) {
3441 		ext4_warning_inode(ent->dir,
3442 				   "Deleting old file: nlink %d, error=%d",
3443 				   ent->dir->i_nlink, retval);
3444 	}
3445 }
3446 
3447 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3448 {
3449 	if (ent->dir_nlink_delta) {
3450 		if (ent->dir_nlink_delta == -1)
3451 			ext4_dec_count(handle, ent->dir);
3452 		else
3453 			ext4_inc_count(handle, ent->dir);
3454 		ext4_mark_inode_dirty(handle, ent->dir);
3455 	}
3456 }
3457 
3458 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3459 					      int credits, handle_t **h)
3460 {
3461 	struct inode *wh;
3462 	handle_t *handle;
3463 	int retries = 0;
3464 
3465 	/*
3466 	 * for inode block, sb block, group summaries,
3467 	 * and inode bitmap
3468 	 */
3469 	credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3470 		    EXT4_XATTR_TRANS_BLOCKS + 4);
3471 retry:
3472 	wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3473 					 &ent->dentry->d_name, 0, NULL,
3474 					 EXT4_HT_DIR, credits);
3475 
3476 	handle = ext4_journal_current_handle();
3477 	if (IS_ERR(wh)) {
3478 		if (handle)
3479 			ext4_journal_stop(handle);
3480 		if (PTR_ERR(wh) == -ENOSPC &&
3481 		    ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3482 			goto retry;
3483 	} else {
3484 		*h = handle;
3485 		init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3486 		wh->i_op = &ext4_special_inode_operations;
3487 	}
3488 	return wh;
3489 }
3490 
3491 /*
3492  * Anybody can rename anything with this: the permission checks are left to the
3493  * higher-level routines.
3494  *
3495  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3496  * while new_{dentry,inode) refers to the destination dentry/inode
3497  * This comes from rename(const char *oldpath, const char *newpath)
3498  */
3499 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3500 		       struct inode *new_dir, struct dentry *new_dentry,
3501 		       unsigned int flags)
3502 {
3503 	handle_t *handle = NULL;
3504 	struct ext4_renament old = {
3505 		.dir = old_dir,
3506 		.dentry = old_dentry,
3507 		.inode = d_inode(old_dentry),
3508 	};
3509 	struct ext4_renament new = {
3510 		.dir = new_dir,
3511 		.dentry = new_dentry,
3512 		.inode = d_inode(new_dentry),
3513 	};
3514 	int force_reread;
3515 	int retval;
3516 	struct inode *whiteout = NULL;
3517 	int credits;
3518 	u8 old_file_type;
3519 
3520 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3521 	    (!projid_eq(EXT4_I(new_dir)->i_projid,
3522 			EXT4_I(old_dentry->d_inode)->i_projid)))
3523 		return -EXDEV;
3524 
3525 	retval = dquot_initialize(old.dir);
3526 	if (retval)
3527 		return retval;
3528 	retval = dquot_initialize(new.dir);
3529 	if (retval)
3530 		return retval;
3531 
3532 	/* Initialize quotas before so that eventual writes go
3533 	 * in separate transaction */
3534 	if (new.inode) {
3535 		retval = dquot_initialize(new.inode);
3536 		if (retval)
3537 			return retval;
3538 	}
3539 
3540 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3541 	if (IS_ERR(old.bh))
3542 		return PTR_ERR(old.bh);
3543 	/*
3544 	 *  Check for inode number is _not_ due to possible IO errors.
3545 	 *  We might rmdir the source, keep it as pwd of some process
3546 	 *  and merrily kill the link to whatever was created under the
3547 	 *  same name. Goodbye sticky bit ;-<
3548 	 */
3549 	retval = -ENOENT;
3550 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3551 		goto end_rename;
3552 
3553 	if ((old.dir != new.dir) &&
3554 	    ext4_encrypted_inode(new.dir) &&
3555 	    !ext4_is_child_context_consistent_with_parent(new.dir,
3556 							  old.inode)) {
3557 		retval = -EPERM;
3558 		goto end_rename;
3559 	}
3560 
3561 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3562 				 &new.de, &new.inlined);
3563 	if (IS_ERR(new.bh)) {
3564 		retval = PTR_ERR(new.bh);
3565 		new.bh = NULL;
3566 		goto end_rename;
3567 	}
3568 	if (new.bh) {
3569 		if (!new.inode) {
3570 			brelse(new.bh);
3571 			new.bh = NULL;
3572 		}
3573 	}
3574 	if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3575 		ext4_alloc_da_blocks(old.inode);
3576 
3577 	credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3578 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3579 	if (!(flags & RENAME_WHITEOUT)) {
3580 		handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3581 		if (IS_ERR(handle)) {
3582 			retval = PTR_ERR(handle);
3583 			handle = NULL;
3584 			goto end_rename;
3585 		}
3586 	} else {
3587 		whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3588 		if (IS_ERR(whiteout)) {
3589 			retval = PTR_ERR(whiteout);
3590 			whiteout = NULL;
3591 			goto end_rename;
3592 		}
3593 	}
3594 
3595 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3596 		ext4_handle_sync(handle);
3597 
3598 	if (S_ISDIR(old.inode->i_mode)) {
3599 		if (new.inode) {
3600 			retval = -ENOTEMPTY;
3601 			if (!ext4_empty_dir(new.inode))
3602 				goto end_rename;
3603 		} else {
3604 			retval = -EMLINK;
3605 			if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3606 				goto end_rename;
3607 		}
3608 		retval = ext4_rename_dir_prepare(handle, &old);
3609 		if (retval)
3610 			goto end_rename;
3611 	}
3612 	/*
3613 	 * If we're renaming a file within an inline_data dir and adding or
3614 	 * setting the new dirent causes a conversion from inline_data to
3615 	 * extents/blockmap, we need to force the dirent delete code to
3616 	 * re-read the directory, or else we end up trying to delete a dirent
3617 	 * from what is now the extent tree root (or a block map).
3618 	 */
3619 	force_reread = (new.dir->i_ino == old.dir->i_ino &&
3620 			ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3621 
3622 	old_file_type = old.de->file_type;
3623 	if (whiteout) {
3624 		/*
3625 		 * Do this before adding a new entry, so the old entry is sure
3626 		 * to be still pointing to the valid old entry.
3627 		 */
3628 		retval = ext4_setent(handle, &old, whiteout->i_ino,
3629 				     EXT4_FT_CHRDEV);
3630 		if (retval)
3631 			goto end_rename;
3632 		ext4_mark_inode_dirty(handle, whiteout);
3633 	}
3634 	if (!new.bh) {
3635 		retval = ext4_add_entry(handle, new.dentry, old.inode);
3636 		if (retval)
3637 			goto end_rename;
3638 	} else {
3639 		retval = ext4_setent(handle, &new,
3640 				     old.inode->i_ino, old_file_type);
3641 		if (retval)
3642 			goto end_rename;
3643 	}
3644 	if (force_reread)
3645 		force_reread = !ext4_test_inode_flag(new.dir,
3646 						     EXT4_INODE_INLINE_DATA);
3647 
3648 	/*
3649 	 * Like most other Unix systems, set the ctime for inodes on a
3650 	 * rename.
3651 	 */
3652 	old.inode->i_ctime = ext4_current_time(old.inode);
3653 	ext4_mark_inode_dirty(handle, old.inode);
3654 
3655 	if (!whiteout) {
3656 		/*
3657 		 * ok, that's it
3658 		 */
3659 		ext4_rename_delete(handle, &old, force_reread);
3660 	}
3661 
3662 	if (new.inode) {
3663 		ext4_dec_count(handle, new.inode);
3664 		new.inode->i_ctime = ext4_current_time(new.inode);
3665 	}
3666 	old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3667 	ext4_update_dx_flag(old.dir);
3668 	if (old.dir_bh) {
3669 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3670 		if (retval)
3671 			goto end_rename;
3672 
3673 		ext4_dec_count(handle, old.dir);
3674 		if (new.inode) {
3675 			/* checked ext4_empty_dir above, can't have another
3676 			 * parent, ext4_dec_count() won't work for many-linked
3677 			 * dirs */
3678 			clear_nlink(new.inode);
3679 		} else {
3680 			ext4_inc_count(handle, new.dir);
3681 			ext4_update_dx_flag(new.dir);
3682 			ext4_mark_inode_dirty(handle, new.dir);
3683 		}
3684 	}
3685 	ext4_mark_inode_dirty(handle, old.dir);
3686 	if (new.inode) {
3687 		ext4_mark_inode_dirty(handle, new.inode);
3688 		if (!new.inode->i_nlink)
3689 			ext4_orphan_add(handle, new.inode);
3690 	}
3691 	retval = 0;
3692 
3693 end_rename:
3694 	brelse(old.dir_bh);
3695 	brelse(old.bh);
3696 	brelse(new.bh);
3697 	if (whiteout) {
3698 		if (retval)
3699 			drop_nlink(whiteout);
3700 		unlock_new_inode(whiteout);
3701 		iput(whiteout);
3702 	}
3703 	if (handle)
3704 		ext4_journal_stop(handle);
3705 	return retval;
3706 }
3707 
3708 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3709 			     struct inode *new_dir, struct dentry *new_dentry)
3710 {
3711 	handle_t *handle = NULL;
3712 	struct ext4_renament old = {
3713 		.dir = old_dir,
3714 		.dentry = old_dentry,
3715 		.inode = d_inode(old_dentry),
3716 	};
3717 	struct ext4_renament new = {
3718 		.dir = new_dir,
3719 		.dentry = new_dentry,
3720 		.inode = d_inode(new_dentry),
3721 	};
3722 	u8 new_file_type;
3723 	int retval;
3724 
3725 	if ((ext4_encrypted_inode(old_dir) ||
3726 	     ext4_encrypted_inode(new_dir)) &&
3727 	    (old_dir != new_dir) &&
3728 	    (!ext4_is_child_context_consistent_with_parent(new_dir,
3729 							   old.inode) ||
3730 	     !ext4_is_child_context_consistent_with_parent(old_dir,
3731 							   new.inode)))
3732 		return -EPERM;
3733 
3734 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3735 	     !projid_eq(EXT4_I(new_dir)->i_projid,
3736 			EXT4_I(old_dentry->d_inode)->i_projid)) ||
3737 	    (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3738 	     !projid_eq(EXT4_I(old_dir)->i_projid,
3739 			EXT4_I(new_dentry->d_inode)->i_projid)))
3740 		return -EXDEV;
3741 
3742 	retval = dquot_initialize(old.dir);
3743 	if (retval)
3744 		return retval;
3745 	retval = dquot_initialize(new.dir);
3746 	if (retval)
3747 		return retval;
3748 
3749 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3750 				 &old.de, &old.inlined);
3751 	if (IS_ERR(old.bh))
3752 		return PTR_ERR(old.bh);
3753 	/*
3754 	 *  Check for inode number is _not_ due to possible IO errors.
3755 	 *  We might rmdir the source, keep it as pwd of some process
3756 	 *  and merrily kill the link to whatever was created under the
3757 	 *  same name. Goodbye sticky bit ;-<
3758 	 */
3759 	retval = -ENOENT;
3760 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3761 		goto end_rename;
3762 
3763 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3764 				 &new.de, &new.inlined);
3765 	if (IS_ERR(new.bh)) {
3766 		retval = PTR_ERR(new.bh);
3767 		new.bh = NULL;
3768 		goto end_rename;
3769 	}
3770 
3771 	/* RENAME_EXCHANGE case: old *and* new must both exist */
3772 	if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3773 		goto end_rename;
3774 
3775 	handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3776 		(2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3777 		 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3778 	if (IS_ERR(handle)) {
3779 		retval = PTR_ERR(handle);
3780 		handle = NULL;
3781 		goto end_rename;
3782 	}
3783 
3784 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3785 		ext4_handle_sync(handle);
3786 
3787 	if (S_ISDIR(old.inode->i_mode)) {
3788 		old.is_dir = true;
3789 		retval = ext4_rename_dir_prepare(handle, &old);
3790 		if (retval)
3791 			goto end_rename;
3792 	}
3793 	if (S_ISDIR(new.inode->i_mode)) {
3794 		new.is_dir = true;
3795 		retval = ext4_rename_dir_prepare(handle, &new);
3796 		if (retval)
3797 			goto end_rename;
3798 	}
3799 
3800 	/*
3801 	 * Other than the special case of overwriting a directory, parents'
3802 	 * nlink only needs to be modified if this is a cross directory rename.
3803 	 */
3804 	if (old.dir != new.dir && old.is_dir != new.is_dir) {
3805 		old.dir_nlink_delta = old.is_dir ? -1 : 1;
3806 		new.dir_nlink_delta = -old.dir_nlink_delta;
3807 		retval = -EMLINK;
3808 		if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3809 		    (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3810 			goto end_rename;
3811 	}
3812 
3813 	new_file_type = new.de->file_type;
3814 	retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3815 	if (retval)
3816 		goto end_rename;
3817 
3818 	retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3819 	if (retval)
3820 		goto end_rename;
3821 
3822 	/*
3823 	 * Like most other Unix systems, set the ctime for inodes on a
3824 	 * rename.
3825 	 */
3826 	old.inode->i_ctime = ext4_current_time(old.inode);
3827 	new.inode->i_ctime = ext4_current_time(new.inode);
3828 	ext4_mark_inode_dirty(handle, old.inode);
3829 	ext4_mark_inode_dirty(handle, new.inode);
3830 
3831 	if (old.dir_bh) {
3832 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3833 		if (retval)
3834 			goto end_rename;
3835 	}
3836 	if (new.dir_bh) {
3837 		retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3838 		if (retval)
3839 			goto end_rename;
3840 	}
3841 	ext4_update_dir_count(handle, &old);
3842 	ext4_update_dir_count(handle, &new);
3843 	retval = 0;
3844 
3845 end_rename:
3846 	brelse(old.dir_bh);
3847 	brelse(new.dir_bh);
3848 	brelse(old.bh);
3849 	brelse(new.bh);
3850 	if (handle)
3851 		ext4_journal_stop(handle);
3852 	return retval;
3853 }
3854 
3855 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3856 			struct inode *new_dir, struct dentry *new_dentry,
3857 			unsigned int flags)
3858 {
3859 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3860 		return -EINVAL;
3861 
3862 	if (flags & RENAME_EXCHANGE) {
3863 		return ext4_cross_rename(old_dir, old_dentry,
3864 					 new_dir, new_dentry);
3865 	}
3866 
3867 	return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3868 }
3869 
3870 /*
3871  * directories can handle most operations...
3872  */
3873 const struct inode_operations ext4_dir_inode_operations = {
3874 	.create		= ext4_create,
3875 	.lookup		= ext4_lookup,
3876 	.link		= ext4_link,
3877 	.unlink		= ext4_unlink,
3878 	.symlink	= ext4_symlink,
3879 	.mkdir		= ext4_mkdir,
3880 	.rmdir		= ext4_rmdir,
3881 	.mknod		= ext4_mknod,
3882 	.tmpfile	= ext4_tmpfile,
3883 	.rename2	= ext4_rename2,
3884 	.setattr	= ext4_setattr,
3885 	.setxattr	= generic_setxattr,
3886 	.getxattr	= generic_getxattr,
3887 	.listxattr	= ext4_listxattr,
3888 	.removexattr	= generic_removexattr,
3889 	.get_acl	= ext4_get_acl,
3890 	.set_acl	= ext4_set_acl,
3891 	.fiemap         = ext4_fiemap,
3892 };
3893 
3894 const struct inode_operations ext4_special_inode_operations = {
3895 	.setattr	= ext4_setattr,
3896 	.setxattr	= generic_setxattr,
3897 	.getxattr	= generic_getxattr,
3898 	.listxattr	= ext4_listxattr,
3899 	.removexattr	= generic_removexattr,
3900 	.get_acl	= ext4_get_acl,
3901 	.set_acl	= ext4_set_acl,
3902 };
3903