xref: /openbmc/linux/fs/ext4/namei.c (revision 5c73cc4b6c83e88863a5de869cc5df3b913aef4a)
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, 1);
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), __LINE__)
88 
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
90 					      ext4_lblk_t block,
91 					      dirblock_type_t type,
92 					      unsigned int line)
93 {
94 	struct buffer_head *bh;
95 	struct ext4_dir_entry *dirent;
96 	int is_dx_block = 0;
97 
98 	bh = ext4_bread(NULL, inode, block, 0);
99 	if (IS_ERR(bh)) {
100 		__ext4_warning(inode->i_sb, __func__, line,
101 			       "error %ld reading directory block "
102 			       "(ino %lu, block %lu)", PTR_ERR(bh), inode->i_ino,
103 			       (unsigned long) block);
104 
105 		return bh;
106 	}
107 	if (!bh) {
108 		ext4_error_inode(inode, __func__, line, block, "Directory hole found");
109 		return ERR_PTR(-EIO);
110 	}
111 	dirent = (struct ext4_dir_entry *) bh->b_data;
112 	/* Determine whether or not we have an index block */
113 	if (is_dx(inode)) {
114 		if (block == 0)
115 			is_dx_block = 1;
116 		else if (ext4_rec_len_from_disk(dirent->rec_len,
117 						inode->i_sb->s_blocksize) ==
118 			 inode->i_sb->s_blocksize)
119 			is_dx_block = 1;
120 	}
121 	if (!is_dx_block && type == INDEX) {
122 		ext4_error_inode(inode, __func__, line, block,
123 		       "directory leaf block found instead of index block");
124 		return ERR_PTR(-EIO);
125 	}
126 	if (!ext4_has_metadata_csum(inode->i_sb) ||
127 	    buffer_verified(bh))
128 		return bh;
129 
130 	/*
131 	 * An empty leaf block can get mistaken for a index block; for
132 	 * this reason, we can only check the index checksum when the
133 	 * caller is sure it should be an index block.
134 	 */
135 	if (is_dx_block && type == INDEX) {
136 		if (ext4_dx_csum_verify(inode, dirent))
137 			set_buffer_verified(bh);
138 		else {
139 			ext4_error_inode(inode, __func__, line, block,
140 				"Directory index failed checksum");
141 			brelse(bh);
142 			return ERR_PTR(-EIO);
143 		}
144 	}
145 	if (!is_dx_block) {
146 		if (ext4_dirent_csum_verify(inode, dirent))
147 			set_buffer_verified(bh);
148 		else {
149 			ext4_error_inode(inode, __func__, line, block,
150 				"Directory block failed checksum");
151 			brelse(bh);
152 			return ERR_PTR(-EIO);
153 		}
154 	}
155 	return bh;
156 }
157 
158 #ifndef assert
159 #define assert(test) J_ASSERT(test)
160 #endif
161 
162 #ifdef DX_DEBUG
163 #define dxtrace(command) command
164 #else
165 #define dxtrace(command)
166 #endif
167 
168 struct fake_dirent
169 {
170 	__le32 inode;
171 	__le16 rec_len;
172 	u8 name_len;
173 	u8 file_type;
174 };
175 
176 struct dx_countlimit
177 {
178 	__le16 limit;
179 	__le16 count;
180 };
181 
182 struct dx_entry
183 {
184 	__le32 hash;
185 	__le32 block;
186 };
187 
188 /*
189  * dx_root_info is laid out so that if it should somehow get overlaid by a
190  * dirent the two low bits of the hash version will be zero.  Therefore, the
191  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
192  */
193 
194 struct dx_root
195 {
196 	struct fake_dirent dot;
197 	char dot_name[4];
198 	struct fake_dirent dotdot;
199 	char dotdot_name[4];
200 	struct dx_root_info
201 	{
202 		__le32 reserved_zero;
203 		u8 hash_version;
204 		u8 info_length; /* 8 */
205 		u8 indirect_levels;
206 		u8 unused_flags;
207 	}
208 	info;
209 	struct dx_entry	entries[0];
210 };
211 
212 struct dx_node
213 {
214 	struct fake_dirent fake;
215 	struct dx_entry	entries[0];
216 };
217 
218 
219 struct dx_frame
220 {
221 	struct buffer_head *bh;
222 	struct dx_entry *entries;
223 	struct dx_entry *at;
224 };
225 
226 struct dx_map_entry
227 {
228 	u32 hash;
229 	u16 offs;
230 	u16 size;
231 };
232 
233 /*
234  * This goes at the end of each htree block.
235  */
236 struct dx_tail {
237 	u32 dt_reserved;
238 	__le32 dt_checksum;	/* crc32c(uuid+inum+dirblock) */
239 };
240 
241 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
242 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
243 static inline unsigned dx_get_hash(struct dx_entry *entry);
244 static void dx_set_hash(struct dx_entry *entry, unsigned value);
245 static unsigned dx_get_count(struct dx_entry *entries);
246 static unsigned dx_get_limit(struct dx_entry *entries);
247 static void dx_set_count(struct dx_entry *entries, unsigned value);
248 static void dx_set_limit(struct dx_entry *entries, unsigned value);
249 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
250 static unsigned dx_node_limit(struct inode *dir);
251 static struct dx_frame *dx_probe(const struct qstr *d_name,
252 				 struct inode *dir,
253 				 struct dx_hash_info *hinfo,
254 				 struct dx_frame *frame);
255 static void dx_release(struct dx_frame *frames);
256 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
257 		       unsigned blocksize, struct dx_hash_info *hinfo,
258 		       struct dx_map_entry map[]);
259 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
260 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
261 		struct dx_map_entry *offsets, int count, unsigned blocksize);
262 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
263 static void dx_insert_block(struct dx_frame *frame,
264 					u32 hash, ext4_lblk_t block);
265 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
266 				 struct dx_frame *frame,
267 				 struct dx_frame *frames,
268 				 __u32 *start_hash);
269 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
270 		const struct qstr *d_name,
271 		struct ext4_dir_entry_2 **res_dir);
272 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
273 			     struct inode *inode);
274 
275 /* checksumming functions */
276 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
277 			    unsigned int blocksize)
278 {
279 	memset(t, 0, sizeof(struct ext4_dir_entry_tail));
280 	t->det_rec_len = ext4_rec_len_to_disk(
281 			sizeof(struct ext4_dir_entry_tail), blocksize);
282 	t->det_reserved_ft = EXT4_FT_DIR_CSUM;
283 }
284 
285 /* Walk through a dirent block to find a checksum "dirent" at the tail */
286 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
287 						   struct ext4_dir_entry *de)
288 {
289 	struct ext4_dir_entry_tail *t;
290 
291 #ifdef PARANOID
292 	struct ext4_dir_entry *d, *top;
293 
294 	d = de;
295 	top = (struct ext4_dir_entry *)(((void *)de) +
296 		(EXT4_BLOCK_SIZE(inode->i_sb) -
297 		sizeof(struct ext4_dir_entry_tail)));
298 	while (d < top && d->rec_len)
299 		d = (struct ext4_dir_entry *)(((void *)d) +
300 		    le16_to_cpu(d->rec_len));
301 
302 	if (d != top)
303 		return NULL;
304 
305 	t = (struct ext4_dir_entry_tail *)d;
306 #else
307 	t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
308 #endif
309 
310 	if (t->det_reserved_zero1 ||
311 	    le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
312 	    t->det_reserved_zero2 ||
313 	    t->det_reserved_ft != EXT4_FT_DIR_CSUM)
314 		return NULL;
315 
316 	return t;
317 }
318 
319 static __le32 ext4_dirent_csum(struct inode *inode,
320 			       struct ext4_dir_entry *dirent, int size)
321 {
322 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
323 	struct ext4_inode_info *ei = EXT4_I(inode);
324 	__u32 csum;
325 
326 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
327 	return cpu_to_le32(csum);
328 }
329 
330 static void warn_no_space_for_csum(struct inode *inode)
331 {
332 	ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
333 		     "checksum.  Please run e2fsck -D.", inode->i_ino);
334 }
335 
336 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
337 {
338 	struct ext4_dir_entry_tail *t;
339 
340 	if (!ext4_has_metadata_csum(inode->i_sb))
341 		return 1;
342 
343 	t = get_dirent_tail(inode, dirent);
344 	if (!t) {
345 		warn_no_space_for_csum(inode);
346 		return 0;
347 	}
348 
349 	if (t->det_checksum != ext4_dirent_csum(inode, dirent,
350 						(void *)t - (void *)dirent))
351 		return 0;
352 
353 	return 1;
354 }
355 
356 static void ext4_dirent_csum_set(struct inode *inode,
357 				 struct ext4_dir_entry *dirent)
358 {
359 	struct ext4_dir_entry_tail *t;
360 
361 	if (!ext4_has_metadata_csum(inode->i_sb))
362 		return;
363 
364 	t = get_dirent_tail(inode, dirent);
365 	if (!t) {
366 		warn_no_space_for_csum(inode);
367 		return;
368 	}
369 
370 	t->det_checksum = ext4_dirent_csum(inode, dirent,
371 					   (void *)t - (void *)dirent);
372 }
373 
374 int ext4_handle_dirty_dirent_node(handle_t *handle,
375 				  struct inode *inode,
376 				  struct buffer_head *bh)
377 {
378 	ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
379 	return ext4_handle_dirty_metadata(handle, inode, bh);
380 }
381 
382 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
383 					       struct ext4_dir_entry *dirent,
384 					       int *offset)
385 {
386 	struct ext4_dir_entry *dp;
387 	struct dx_root_info *root;
388 	int count_offset;
389 
390 	if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
391 		count_offset = 8;
392 	else if (le16_to_cpu(dirent->rec_len) == 12) {
393 		dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
394 		if (le16_to_cpu(dp->rec_len) !=
395 		    EXT4_BLOCK_SIZE(inode->i_sb) - 12)
396 			return NULL;
397 		root = (struct dx_root_info *)(((void *)dp + 12));
398 		if (root->reserved_zero ||
399 		    root->info_length != sizeof(struct dx_root_info))
400 			return NULL;
401 		count_offset = 32;
402 	} else
403 		return NULL;
404 
405 	if (offset)
406 		*offset = count_offset;
407 	return (struct dx_countlimit *)(((void *)dirent) + count_offset);
408 }
409 
410 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
411 			   int count_offset, int count, struct dx_tail *t)
412 {
413 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
414 	struct ext4_inode_info *ei = EXT4_I(inode);
415 	__u32 csum;
416 	__le32 save_csum;
417 	int size;
418 
419 	size = count_offset + (count * sizeof(struct dx_entry));
420 	save_csum = t->dt_checksum;
421 	t->dt_checksum = 0;
422 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
423 	csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
424 	t->dt_checksum = save_csum;
425 
426 	return cpu_to_le32(csum);
427 }
428 
429 static int ext4_dx_csum_verify(struct inode *inode,
430 			       struct ext4_dir_entry *dirent)
431 {
432 	struct dx_countlimit *c;
433 	struct dx_tail *t;
434 	int count_offset, limit, count;
435 
436 	if (!ext4_has_metadata_csum(inode->i_sb))
437 		return 1;
438 
439 	c = get_dx_countlimit(inode, dirent, &count_offset);
440 	if (!c) {
441 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
442 		return 1;
443 	}
444 	limit = le16_to_cpu(c->limit);
445 	count = le16_to_cpu(c->count);
446 	if (count_offset + (limit * sizeof(struct dx_entry)) >
447 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
448 		warn_no_space_for_csum(inode);
449 		return 1;
450 	}
451 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
452 
453 	if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
454 					    count, t))
455 		return 0;
456 	return 1;
457 }
458 
459 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
460 {
461 	struct dx_countlimit *c;
462 	struct dx_tail *t;
463 	int count_offset, limit, count;
464 
465 	if (!ext4_has_metadata_csum(inode->i_sb))
466 		return;
467 
468 	c = get_dx_countlimit(inode, dirent, &count_offset);
469 	if (!c) {
470 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
471 		return;
472 	}
473 	limit = le16_to_cpu(c->limit);
474 	count = le16_to_cpu(c->count);
475 	if (count_offset + (limit * sizeof(struct dx_entry)) >
476 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
477 		warn_no_space_for_csum(inode);
478 		return;
479 	}
480 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
481 
482 	t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
483 }
484 
485 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
486 					    struct inode *inode,
487 					    struct buffer_head *bh)
488 {
489 	ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
490 	return ext4_handle_dirty_metadata(handle, inode, bh);
491 }
492 
493 /*
494  * p is at least 6 bytes before the end of page
495  */
496 static inline struct ext4_dir_entry_2 *
497 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
498 {
499 	return (struct ext4_dir_entry_2 *)((char *)p +
500 		ext4_rec_len_from_disk(p->rec_len, blocksize));
501 }
502 
503 /*
504  * Future: use high four bits of block for coalesce-on-delete flags
505  * Mask them off for now.
506  */
507 
508 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
509 {
510 	return le32_to_cpu(entry->block) & 0x00ffffff;
511 }
512 
513 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
514 {
515 	entry->block = cpu_to_le32(value);
516 }
517 
518 static inline unsigned dx_get_hash(struct dx_entry *entry)
519 {
520 	return le32_to_cpu(entry->hash);
521 }
522 
523 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
524 {
525 	entry->hash = cpu_to_le32(value);
526 }
527 
528 static inline unsigned dx_get_count(struct dx_entry *entries)
529 {
530 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
531 }
532 
533 static inline unsigned dx_get_limit(struct dx_entry *entries)
534 {
535 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
536 }
537 
538 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
539 {
540 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
541 }
542 
543 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
544 {
545 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
546 }
547 
548 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
549 {
550 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
551 		EXT4_DIR_REC_LEN(2) - infosize;
552 
553 	if (ext4_has_metadata_csum(dir->i_sb))
554 		entry_space -= sizeof(struct dx_tail);
555 	return entry_space / sizeof(struct dx_entry);
556 }
557 
558 static inline unsigned dx_node_limit(struct inode *dir)
559 {
560 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
561 
562 	if (ext4_has_metadata_csum(dir->i_sb))
563 		entry_space -= sizeof(struct dx_tail);
564 	return entry_space / sizeof(struct dx_entry);
565 }
566 
567 /*
568  * Debug
569  */
570 #ifdef DX_DEBUG
571 static void dx_show_index(char * label, struct dx_entry *entries)
572 {
573 	int i, n = dx_get_count (entries);
574 	printk(KERN_DEBUG "%s index ", label);
575 	for (i = 0; i < n; i++) {
576 		printk("%x->%lu ", i ? dx_get_hash(entries + i) :
577 				0, (unsigned long)dx_get_block(entries + i));
578 	}
579 	printk("\n");
580 }
581 
582 struct stats
583 {
584 	unsigned names;
585 	unsigned space;
586 	unsigned bcount;
587 };
588 
589 static struct stats dx_show_leaf(struct inode *dir,
590 				struct dx_hash_info *hinfo,
591 				struct ext4_dir_entry_2 *de,
592 				int size, int show_names)
593 {
594 	unsigned names = 0, space = 0;
595 	char *base = (char *) de;
596 	struct dx_hash_info h = *hinfo;
597 
598 	printk("names: ");
599 	while ((char *) de < base + size)
600 	{
601 		if (de->inode)
602 		{
603 			if (show_names)
604 			{
605 #ifdef CONFIG_EXT4_FS_ENCRYPTION
606 				int len;
607 				char *name;
608 				struct ext4_str fname_crypto_str
609 					= {.name = NULL, .len = 0};
610 				struct ext4_fname_crypto_ctx *ctx = NULL;
611 				int res;
612 
613 				name  = de->name;
614 				len = de->name_len;
615 				ctx = ext4_get_fname_crypto_ctx(dir,
616 								EXT4_NAME_LEN);
617 				if (IS_ERR(ctx)) {
618 					printk(KERN_WARNING "Error acquiring"
619 					" crypto ctxt--skipping crypto\n");
620 					ctx = NULL;
621 				}
622 				if (ctx == NULL) {
623 					/* Directory is not encrypted */
624 					ext4fs_dirhash(de->name,
625 						de->name_len, &h);
626 					printk("%*.s:(U)%x.%u ", len,
627 					       name, h.hash,
628 					       (unsigned) ((char *) de
629 							   - base));
630 				} else {
631 					/* Directory is encrypted */
632 					res = ext4_fname_crypto_alloc_buffer(
633 						ctx, de->name_len,
634 						&fname_crypto_str);
635 					if (res < 0) {
636 						printk(KERN_WARNING "Error "
637 							"allocating crypto "
638 							"buffer--skipping "
639 							"crypto\n");
640 						ext4_put_fname_crypto_ctx(&ctx);
641 						ctx = NULL;
642 					}
643 					res = ext4_fname_disk_to_usr(ctx, de,
644 							&fname_crypto_str);
645 					if (res < 0) {
646 						printk(KERN_WARNING "Error "
647 							"converting filename "
648 							"from disk to usr"
649 							"\n");
650 						name = "??";
651 						len = 2;
652 					} else {
653 						name = fname_crypto_str.name;
654 						len = fname_crypto_str.len;
655 					}
656 					res = ext4_fname_disk_to_hash(ctx, de,
657 								      &h);
658 					if (res < 0) {
659 						printk(KERN_WARNING "Error "
660 							"converting filename "
661 							"from disk to htree"
662 							"\n");
663 						h.hash = 0xDEADBEEF;
664 					}
665 					printk("%*.s:(E)%x.%u ", len, name,
666 					       h.hash, (unsigned) ((char *) de
667 								   - base));
668 					ext4_put_fname_crypto_ctx(&ctx);
669 					ext4_fname_crypto_free_buffer(
670 						&fname_crypto_str);
671 				}
672 #else
673 				int len = de->name_len;
674 				char *name = de->name;
675 				ext4fs_dirhash(de->name, de->name_len, &h);
676 				printk("%*.s:%x.%u ", len, name, h.hash,
677 				       (unsigned) ((char *) de - base));
678 #endif
679 			}
680 			space += EXT4_DIR_REC_LEN(de->name_len);
681 			names++;
682 		}
683 		de = ext4_next_entry(de, size);
684 	}
685 	printk("(%i)\n", names);
686 	return (struct stats) { names, space, 1 };
687 }
688 
689 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
690 			     struct dx_entry *entries, int levels)
691 {
692 	unsigned blocksize = dir->i_sb->s_blocksize;
693 	unsigned count = dx_get_count(entries), names = 0, space = 0, i;
694 	unsigned bcount = 0;
695 	struct buffer_head *bh;
696 	printk("%i indexed blocks...\n", count);
697 	for (i = 0; i < count; i++, entries++)
698 	{
699 		ext4_lblk_t block = dx_get_block(entries);
700 		ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
701 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
702 		struct stats stats;
703 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
704 		bh = ext4_bread(NULL,dir, block, 0);
705 		if (!bh || IS_ERR(bh))
706 			continue;
707 		stats = levels?
708 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
709 		   dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
710 			bh->b_data, blocksize, 0);
711 		names += stats.names;
712 		space += stats.space;
713 		bcount += stats.bcount;
714 		brelse(bh);
715 	}
716 	if (bcount)
717 		printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
718 		       levels ? "" : "   ", names, space/bcount,
719 		       (space/bcount)*100/blocksize);
720 	return (struct stats) { names, space, bcount};
721 }
722 #endif /* DX_DEBUG */
723 
724 /*
725  * Probe for a directory leaf block to search.
726  *
727  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
728  * error in the directory index, and the caller should fall back to
729  * searching the directory normally.  The callers of dx_probe **MUST**
730  * check for this error code, and make sure it never gets reflected
731  * back to userspace.
732  */
733 static struct dx_frame *
734 dx_probe(const struct qstr *d_name, struct inode *dir,
735 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
736 {
737 	unsigned count, indirect;
738 	struct dx_entry *at, *entries, *p, *q, *m;
739 	struct dx_root *root;
740 	struct dx_frame *frame = frame_in;
741 	struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
742 	u32 hash;
743 
744 	frame->bh = ext4_read_dirblock(dir, 0, INDEX);
745 	if (IS_ERR(frame->bh))
746 		return (struct dx_frame *) frame->bh;
747 
748 	root = (struct dx_root *) frame->bh->b_data;
749 	if (root->info.hash_version != DX_HASH_TEA &&
750 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
751 	    root->info.hash_version != DX_HASH_LEGACY) {
752 		ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
753 			     root->info.hash_version);
754 		goto fail;
755 	}
756 	hinfo->hash_version = root->info.hash_version;
757 	if (hinfo->hash_version <= DX_HASH_TEA)
758 		hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
759 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
760 #ifdef CONFIG_EXT4_FS_ENCRYPTION
761 	if (d_name) {
762 		struct ext4_fname_crypto_ctx *ctx = NULL;
763 		int res;
764 
765 		/* Check if the directory is encrypted */
766 		ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
767 		if (IS_ERR(ctx)) {
768 			ret_err = ERR_PTR(PTR_ERR(ctx));
769 			goto fail;
770 		}
771 		res = ext4_fname_usr_to_hash(ctx, d_name, hinfo);
772 		if (res < 0) {
773 			ret_err = ERR_PTR(res);
774 			goto fail;
775 		}
776 		ext4_put_fname_crypto_ctx(&ctx);
777 	}
778 #else
779 	if (d_name)
780 		ext4fs_dirhash(d_name->name, d_name->len, hinfo);
781 #endif
782 	hash = hinfo->hash;
783 
784 	if (root->info.unused_flags & 1) {
785 		ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
786 			     root->info.unused_flags);
787 		goto fail;
788 	}
789 
790 	if ((indirect = root->info.indirect_levels) > 1) {
791 		ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
792 			     root->info.indirect_levels);
793 		goto fail;
794 	}
795 
796 	entries = (struct dx_entry *) (((char *)&root->info) +
797 				       root->info.info_length);
798 
799 	if (dx_get_limit(entries) != dx_root_limit(dir,
800 						   root->info.info_length)) {
801 		ext4_warning(dir->i_sb, "dx entry: limit != root limit");
802 		goto fail;
803 	}
804 
805 	dxtrace(printk("Look up %x", hash));
806 	while (1) {
807 		count = dx_get_count(entries);
808 		if (!count || count > dx_get_limit(entries)) {
809 			ext4_warning(dir->i_sb,
810 				     "dx entry: no count or count > limit");
811 			goto fail;
812 		}
813 
814 		p = entries + 1;
815 		q = entries + count - 1;
816 		while (p <= q) {
817 			m = p + (q - p)/2;
818 			dxtrace(printk("."));
819 			if (dx_get_hash(m) > hash)
820 				q = m - 1;
821 			else
822 				p = m + 1;
823 		}
824 
825 		if (0) { // linear search cross check
826 			unsigned n = count - 1;
827 			at = entries;
828 			while (n--)
829 			{
830 				dxtrace(printk(","));
831 				if (dx_get_hash(++at) > hash)
832 				{
833 					at--;
834 					break;
835 				}
836 			}
837 			assert (at == p - 1);
838 		}
839 
840 		at = p - 1;
841 		dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
842 		frame->entries = entries;
843 		frame->at = at;
844 		if (!indirect--)
845 			return frame;
846 		frame++;
847 		frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
848 		if (IS_ERR(frame->bh)) {
849 			ret_err = (struct dx_frame *) frame->bh;
850 			frame->bh = NULL;
851 			goto fail;
852 		}
853 		entries = ((struct dx_node *) frame->bh->b_data)->entries;
854 
855 		if (dx_get_limit(entries) != dx_node_limit (dir)) {
856 			ext4_warning(dir->i_sb,
857 				     "dx entry: limit != node limit");
858 			goto fail;
859 		}
860 	}
861 fail:
862 	while (frame >= frame_in) {
863 		brelse(frame->bh);
864 		frame--;
865 	}
866 
867 	if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
868 		ext4_warning(dir->i_sb,
869 			     "Corrupt dir inode %lu, running e2fsck is "
870 			     "recommended.", dir->i_ino);
871 	return ret_err;
872 }
873 
874 static void dx_release (struct dx_frame *frames)
875 {
876 	if (frames[0].bh == NULL)
877 		return;
878 
879 	if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
880 		brelse(frames[1].bh);
881 	brelse(frames[0].bh);
882 }
883 
884 /*
885  * This function increments the frame pointer to search the next leaf
886  * block, and reads in the necessary intervening nodes if the search
887  * should be necessary.  Whether or not the search is necessary is
888  * controlled by the hash parameter.  If the hash value is even, then
889  * the search is only continued if the next block starts with that
890  * hash value.  This is used if we are searching for a specific file.
891  *
892  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
893  *
894  * This function returns 1 if the caller should continue to search,
895  * or 0 if it should not.  If there is an error reading one of the
896  * index blocks, it will a negative error code.
897  *
898  * If start_hash is non-null, it will be filled in with the starting
899  * hash of the next page.
900  */
901 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
902 				 struct dx_frame *frame,
903 				 struct dx_frame *frames,
904 				 __u32 *start_hash)
905 {
906 	struct dx_frame *p;
907 	struct buffer_head *bh;
908 	int num_frames = 0;
909 	__u32 bhash;
910 
911 	p = frame;
912 	/*
913 	 * Find the next leaf page by incrementing the frame pointer.
914 	 * If we run out of entries in the interior node, loop around and
915 	 * increment pointer in the parent node.  When we break out of
916 	 * this loop, num_frames indicates the number of interior
917 	 * nodes need to be read.
918 	 */
919 	while (1) {
920 		if (++(p->at) < p->entries + dx_get_count(p->entries))
921 			break;
922 		if (p == frames)
923 			return 0;
924 		num_frames++;
925 		p--;
926 	}
927 
928 	/*
929 	 * If the hash is 1, then continue only if the next page has a
930 	 * continuation hash of any value.  This is used for readdir
931 	 * handling.  Otherwise, check to see if the hash matches the
932 	 * desired contiuation hash.  If it doesn't, return since
933 	 * there's no point to read in the successive index pages.
934 	 */
935 	bhash = dx_get_hash(p->at);
936 	if (start_hash)
937 		*start_hash = bhash;
938 	if ((hash & 1) == 0) {
939 		if ((bhash & ~1) != hash)
940 			return 0;
941 	}
942 	/*
943 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
944 	 * block so no check is necessary
945 	 */
946 	while (num_frames--) {
947 		bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
948 		if (IS_ERR(bh))
949 			return PTR_ERR(bh);
950 		p++;
951 		brelse(p->bh);
952 		p->bh = bh;
953 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
954 	}
955 	return 1;
956 }
957 
958 
959 /*
960  * This function fills a red-black tree with information from a
961  * directory block.  It returns the number directory entries loaded
962  * into the tree.  If there is an error it is returned in err.
963  */
964 static int htree_dirblock_to_tree(struct file *dir_file,
965 				  struct inode *dir, ext4_lblk_t block,
966 				  struct dx_hash_info *hinfo,
967 				  __u32 start_hash, __u32 start_minor_hash)
968 {
969 	struct buffer_head *bh;
970 	struct ext4_dir_entry_2 *de, *top;
971 	int err = 0, count = 0;
972 	struct ext4_fname_crypto_ctx *ctx = NULL;
973 	struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}, tmp_str;
974 
975 	dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
976 							(unsigned long)block));
977 	bh = ext4_read_dirblock(dir, block, DIRENT);
978 	if (IS_ERR(bh))
979 		return PTR_ERR(bh);
980 
981 	de = (struct ext4_dir_entry_2 *) bh->b_data;
982 	top = (struct ext4_dir_entry_2 *) ((char *) de +
983 					   dir->i_sb->s_blocksize -
984 					   EXT4_DIR_REC_LEN(0));
985 #ifdef CONFIG_EXT4_FS_ENCRYPTION
986 	/* Check if the directory is encrypted */
987 	ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
988 	if (IS_ERR(ctx)) {
989 		err = PTR_ERR(ctx);
990 		brelse(bh);
991 		return err;
992 	}
993 	if (ctx != NULL) {
994 		err = ext4_fname_crypto_alloc_buffer(ctx, EXT4_NAME_LEN,
995 						     &fname_crypto_str);
996 		if (err < 0) {
997 			ext4_put_fname_crypto_ctx(&ctx);
998 			brelse(bh);
999 			return err;
1000 		}
1001 	}
1002 #endif
1003 	for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1004 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1005 				bh->b_data, bh->b_size,
1006 				(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1007 					 + ((char *)de - bh->b_data))) {
1008 			/* silently ignore the rest of the block */
1009 			break;
1010 		}
1011 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1012 		err = ext4_fname_disk_to_hash(ctx, de, hinfo);
1013 		if (err < 0) {
1014 			count = err;
1015 			goto errout;
1016 		}
1017 #else
1018 		ext4fs_dirhash(de->name, de->name_len, hinfo);
1019 #endif
1020 		if ((hinfo->hash < start_hash) ||
1021 		    ((hinfo->hash == start_hash) &&
1022 		     (hinfo->minor_hash < start_minor_hash)))
1023 			continue;
1024 		if (de->inode == 0)
1025 			continue;
1026 		if (ctx == NULL) {
1027 			/* Directory is not encrypted */
1028 			tmp_str.name = de->name;
1029 			tmp_str.len = de->name_len;
1030 			err = ext4_htree_store_dirent(dir_file,
1031 				   hinfo->hash, hinfo->minor_hash, de,
1032 				   &tmp_str);
1033 		} else {
1034 			/* Directory is encrypted */
1035 			err = ext4_fname_disk_to_usr(ctx, de,
1036 						     &fname_crypto_str);
1037 			if (err < 0) {
1038 				count = err;
1039 				goto errout;
1040 			}
1041 			err = ext4_htree_store_dirent(dir_file,
1042 				   hinfo->hash, hinfo->minor_hash, de,
1043 					&fname_crypto_str);
1044 		}
1045 		if (err != 0) {
1046 			count = err;
1047 			goto errout;
1048 		}
1049 		count++;
1050 	}
1051 errout:
1052 	brelse(bh);
1053 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1054 	ext4_put_fname_crypto_ctx(&ctx);
1055 	ext4_fname_crypto_free_buffer(&fname_crypto_str);
1056 #endif
1057 	return count;
1058 }
1059 
1060 
1061 /*
1062  * This function fills a red-black tree with information from a
1063  * directory.  We start scanning the directory in hash order, starting
1064  * at start_hash and start_minor_hash.
1065  *
1066  * This function returns the number of entries inserted into the tree,
1067  * or a negative error code.
1068  */
1069 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1070 			 __u32 start_minor_hash, __u32 *next_hash)
1071 {
1072 	struct dx_hash_info hinfo;
1073 	struct ext4_dir_entry_2 *de;
1074 	struct dx_frame frames[2], *frame;
1075 	struct inode *dir;
1076 	ext4_lblk_t block;
1077 	int count = 0;
1078 	int ret, err;
1079 	__u32 hashval;
1080 	struct ext4_str tmp_str;
1081 
1082 	dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1083 		       start_hash, start_minor_hash));
1084 	dir = file_inode(dir_file);
1085 	if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1086 		hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1087 		if (hinfo.hash_version <= DX_HASH_TEA)
1088 			hinfo.hash_version +=
1089 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
1090 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1091 		if (ext4_has_inline_data(dir)) {
1092 			int has_inline_data = 1;
1093 			count = htree_inlinedir_to_tree(dir_file, dir, 0,
1094 							&hinfo, start_hash,
1095 							start_minor_hash,
1096 							&has_inline_data);
1097 			if (has_inline_data) {
1098 				*next_hash = ~0;
1099 				return count;
1100 			}
1101 		}
1102 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1103 					       start_hash, start_minor_hash);
1104 		*next_hash = ~0;
1105 		return count;
1106 	}
1107 	hinfo.hash = start_hash;
1108 	hinfo.minor_hash = 0;
1109 	frame = dx_probe(NULL, dir, &hinfo, frames);
1110 	if (IS_ERR(frame))
1111 		return PTR_ERR(frame);
1112 
1113 	/* Add '.' and '..' from the htree header */
1114 	if (!start_hash && !start_minor_hash) {
1115 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1116 		tmp_str.name = de->name;
1117 		tmp_str.len = de->name_len;
1118 		err = ext4_htree_store_dirent(dir_file, 0, 0,
1119 					      de, &tmp_str);
1120 		if (err != 0)
1121 			goto errout;
1122 		count++;
1123 	}
1124 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1125 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1126 		de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1127 		tmp_str.name = de->name;
1128 		tmp_str.len = de->name_len;
1129 		err = ext4_htree_store_dirent(dir_file, 2, 0,
1130 					      de, &tmp_str);
1131 		if (err != 0)
1132 			goto errout;
1133 		count++;
1134 	}
1135 
1136 	while (1) {
1137 		block = dx_get_block(frame->at);
1138 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1139 					     start_hash, start_minor_hash);
1140 		if (ret < 0) {
1141 			err = ret;
1142 			goto errout;
1143 		}
1144 		count += ret;
1145 		hashval = ~0;
1146 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1147 					    frame, frames, &hashval);
1148 		*next_hash = hashval;
1149 		if (ret < 0) {
1150 			err = ret;
1151 			goto errout;
1152 		}
1153 		/*
1154 		 * Stop if:  (a) there are no more entries, or
1155 		 * (b) we have inserted at least one entry and the
1156 		 * next hash value is not a continuation
1157 		 */
1158 		if ((ret == 0) ||
1159 		    (count && ((hashval & 1) == 0)))
1160 			break;
1161 	}
1162 	dx_release(frames);
1163 	dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1164 		       "next hash: %x\n", count, *next_hash));
1165 	return count;
1166 errout:
1167 	dx_release(frames);
1168 	return (err);
1169 }
1170 
1171 static inline int search_dirblock(struct buffer_head *bh,
1172 				  struct inode *dir,
1173 				  const struct qstr *d_name,
1174 				  unsigned int offset,
1175 				  struct ext4_dir_entry_2 **res_dir)
1176 {
1177 	return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1178 			  d_name, offset, res_dir);
1179 }
1180 
1181 /*
1182  * Directory block splitting, compacting
1183  */
1184 
1185 /*
1186  * Create map of hash values, offsets, and sizes, stored at end of block.
1187  * Returns number of entries mapped.
1188  */
1189 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1190 		       unsigned blocksize, struct dx_hash_info *hinfo,
1191 		       struct dx_map_entry *map_tail)
1192 {
1193 	int count = 0;
1194 	char *base = (char *) de;
1195 	struct dx_hash_info h = *hinfo;
1196 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1197 	struct ext4_fname_crypto_ctx *ctx = NULL;
1198 	int err;
1199 
1200 	ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
1201 	if (IS_ERR(ctx))
1202 		return PTR_ERR(ctx);
1203 #endif
1204 
1205 	while ((char *) de < base + blocksize) {
1206 		if (de->name_len && de->inode) {
1207 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1208 			err = ext4_fname_disk_to_hash(ctx, de, &h);
1209 			if (err < 0) {
1210 				ext4_put_fname_crypto_ctx(&ctx);
1211 				return err;
1212 			}
1213 #else
1214 			ext4fs_dirhash(de->name, de->name_len, &h);
1215 #endif
1216 			map_tail--;
1217 			map_tail->hash = h.hash;
1218 			map_tail->offs = ((char *) de - base)>>2;
1219 			map_tail->size = le16_to_cpu(de->rec_len);
1220 			count++;
1221 			cond_resched();
1222 		}
1223 		/* XXX: do we need to check rec_len == 0 case? -Chris */
1224 		de = ext4_next_entry(de, blocksize);
1225 	}
1226 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1227 	ext4_put_fname_crypto_ctx(&ctx);
1228 #endif
1229 	return count;
1230 }
1231 
1232 /* Sort map by hash value */
1233 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1234 {
1235 	struct dx_map_entry *p, *q, *top = map + count - 1;
1236 	int more;
1237 	/* Combsort until bubble sort doesn't suck */
1238 	while (count > 2) {
1239 		count = count*10/13;
1240 		if (count - 9 < 2) /* 9, 10 -> 11 */
1241 			count = 11;
1242 		for (p = top, q = p - count; q >= map; p--, q--)
1243 			if (p->hash < q->hash)
1244 				swap(*p, *q);
1245 	}
1246 	/* Garden variety bubble sort */
1247 	do {
1248 		more = 0;
1249 		q = top;
1250 		while (q-- > map) {
1251 			if (q[1].hash >= q[0].hash)
1252 				continue;
1253 			swap(*(q+1), *q);
1254 			more = 1;
1255 		}
1256 	} while(more);
1257 }
1258 
1259 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1260 {
1261 	struct dx_entry *entries = frame->entries;
1262 	struct dx_entry *old = frame->at, *new = old + 1;
1263 	int count = dx_get_count(entries);
1264 
1265 	assert(count < dx_get_limit(entries));
1266 	assert(old < entries + count);
1267 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1268 	dx_set_hash(new, hash);
1269 	dx_set_block(new, block);
1270 	dx_set_count(entries, count + 1);
1271 }
1272 
1273 /*
1274  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1275  *
1276  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1277  * `de != NULL' is guaranteed by caller.
1278  */
1279 static inline int ext4_match(struct ext4_fname_crypto_ctx *ctx,
1280 			     struct ext4_str *fname_crypto_str,
1281 			     int len, const char * const name,
1282 			     struct ext4_dir_entry_2 *de)
1283 {
1284 	int res;
1285 
1286 	if (!de->inode)
1287 		return 0;
1288 
1289 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1290 	if (ctx) {
1291 		/* Directory is encrypted */
1292 		res = ext4_fname_disk_to_usr(ctx, de, fname_crypto_str);
1293 		if (res < 0)
1294 			return res;
1295 		if (len != res)
1296 			return 0;
1297 		res = memcmp(name, fname_crypto_str->name, len);
1298 		return (res == 0) ? 1 : 0;
1299 	}
1300 #endif
1301 	if (len != de->name_len)
1302 		return 0;
1303 	res = memcmp(name, de->name, len);
1304 	return (res == 0) ? 1 : 0;
1305 }
1306 
1307 /*
1308  * Returns 0 if not found, -1 on failure, and 1 on success
1309  */
1310 int search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1311 	       struct inode *dir, const struct qstr *d_name,
1312 	       unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1313 {
1314 	struct ext4_dir_entry_2 * de;
1315 	char * dlimit;
1316 	int de_len;
1317 	const char *name = d_name->name;
1318 	int namelen = d_name->len;
1319 	struct ext4_fname_crypto_ctx *ctx = NULL;
1320 	struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
1321 	int res;
1322 
1323 	ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
1324 	if (IS_ERR(ctx))
1325 		return -1;
1326 
1327 	if (ctx != NULL) {
1328 		/* Allocate buffer to hold maximum name length */
1329 		res = ext4_fname_crypto_alloc_buffer(ctx, EXT4_NAME_LEN,
1330 						     &fname_crypto_str);
1331 		if (res < 0) {
1332 			ext4_put_fname_crypto_ctx(&ctx);
1333 			return -1;
1334 		}
1335 	}
1336 
1337 	de = (struct ext4_dir_entry_2 *)search_buf;
1338 	dlimit = search_buf + buf_size;
1339 	while ((char *) de < dlimit) {
1340 		/* this code is executed quadratically often */
1341 		/* do minimal checking `by hand' */
1342 		if ((char *) de + de->name_len <= dlimit) {
1343 			res = ext4_match(ctx, &fname_crypto_str, namelen,
1344 					 name, de);
1345 			if (res < 0) {
1346 				res = -1;
1347 				goto return_result;
1348 			}
1349 			if (res > 0) {
1350 				/* found a match - just to be sure, do
1351 				 * a full check */
1352 				if (ext4_check_dir_entry(dir, NULL, de, bh,
1353 						bh->b_data,
1354 						 bh->b_size, offset)) {
1355 					res = -1;
1356 					goto return_result;
1357 				}
1358 				*res_dir = de;
1359 				res = 1;
1360 				goto return_result;
1361 			}
1362 
1363 		}
1364 		/* prevent looping on a bad block */
1365 		de_len = ext4_rec_len_from_disk(de->rec_len,
1366 						dir->i_sb->s_blocksize);
1367 		if (de_len <= 0) {
1368 			res = -1;
1369 			goto return_result;
1370 		}
1371 		offset += de_len;
1372 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1373 	}
1374 
1375 	res = 0;
1376 return_result:
1377 	ext4_put_fname_crypto_ctx(&ctx);
1378 	ext4_fname_crypto_free_buffer(&fname_crypto_str);
1379 	return res;
1380 }
1381 
1382 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1383 			       struct ext4_dir_entry *de)
1384 {
1385 	struct super_block *sb = dir->i_sb;
1386 
1387 	if (!is_dx(dir))
1388 		return 0;
1389 	if (block == 0)
1390 		return 1;
1391 	if (de->inode == 0 &&
1392 	    ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1393 			sb->s_blocksize)
1394 		return 1;
1395 	return 0;
1396 }
1397 
1398 /*
1399  *	ext4_find_entry()
1400  *
1401  * finds an entry in the specified directory with the wanted name. It
1402  * returns the cache buffer in which the entry was found, and the entry
1403  * itself (as a parameter - res_dir). It does NOT read the inode of the
1404  * entry - you'll have to do that yourself if you want to.
1405  *
1406  * The returned buffer_head has ->b_count elevated.  The caller is expected
1407  * to brelse() it when appropriate.
1408  */
1409 static struct buffer_head * ext4_find_entry (struct inode *dir,
1410 					const struct qstr *d_name,
1411 					struct ext4_dir_entry_2 **res_dir,
1412 					int *inlined)
1413 {
1414 	struct super_block *sb;
1415 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1416 	struct buffer_head *bh, *ret = NULL;
1417 	ext4_lblk_t start, block, b;
1418 	const u8 *name = d_name->name;
1419 	int ra_max = 0;		/* Number of bh's in the readahead
1420 				   buffer, bh_use[] */
1421 	int ra_ptr = 0;		/* Current index into readahead
1422 				   buffer */
1423 	int num = 0;
1424 	ext4_lblk_t  nblocks;
1425 	int i, namelen;
1426 
1427 	*res_dir = NULL;
1428 	sb = dir->i_sb;
1429 	namelen = d_name->len;
1430 	if (namelen > EXT4_NAME_LEN)
1431 		return NULL;
1432 
1433 	if (ext4_has_inline_data(dir)) {
1434 		int has_inline_data = 1;
1435 		ret = ext4_find_inline_entry(dir, d_name, res_dir,
1436 					     &has_inline_data);
1437 		if (has_inline_data) {
1438 			if (inlined)
1439 				*inlined = 1;
1440 			return ret;
1441 		}
1442 	}
1443 
1444 	if ((namelen <= 2) && (name[0] == '.') &&
1445 	    (name[1] == '.' || name[1] == '\0')) {
1446 		/*
1447 		 * "." or ".." will only be in the first block
1448 		 * NFS may look up ".."; "." should be handled by the VFS
1449 		 */
1450 		block = start = 0;
1451 		nblocks = 1;
1452 		goto restart;
1453 	}
1454 	if (is_dx(dir)) {
1455 		bh = ext4_dx_find_entry(dir, d_name, res_dir);
1456 		/*
1457 		 * On success, or if the error was file not found,
1458 		 * return.  Otherwise, fall back to doing a search the
1459 		 * old fashioned way.
1460 		 */
1461 		if (!IS_ERR(bh) || PTR_ERR(bh) != ERR_BAD_DX_DIR)
1462 			return bh;
1463 		dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1464 			       "falling back\n"));
1465 	}
1466 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1467 	start = EXT4_I(dir)->i_dir_start_lookup;
1468 	if (start >= nblocks)
1469 		start = 0;
1470 	block = start;
1471 restart:
1472 	do {
1473 		/*
1474 		 * We deal with the read-ahead logic here.
1475 		 */
1476 		if (ra_ptr >= ra_max) {
1477 			/* Refill the readahead buffer */
1478 			ra_ptr = 0;
1479 			b = block;
1480 			for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1481 				/*
1482 				 * Terminate if we reach the end of the
1483 				 * directory and must wrap, or if our
1484 				 * search has finished at this block.
1485 				 */
1486 				if (b >= nblocks || (num && block == start)) {
1487 					bh_use[ra_max] = NULL;
1488 					break;
1489 				}
1490 				num++;
1491 				bh = ext4_getblk(NULL, dir, b++, 0);
1492 				if (unlikely(IS_ERR(bh))) {
1493 					if (ra_max == 0)
1494 						return bh;
1495 					break;
1496 				}
1497 				bh_use[ra_max] = bh;
1498 				if (bh)
1499 					ll_rw_block(READ | REQ_META | REQ_PRIO,
1500 						    1, &bh);
1501 			}
1502 		}
1503 		if ((bh = bh_use[ra_ptr++]) == NULL)
1504 			goto next;
1505 		wait_on_buffer(bh);
1506 		if (!buffer_uptodate(bh)) {
1507 			/* read error, skip block & hope for the best */
1508 			EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1509 					 (unsigned long) block);
1510 			brelse(bh);
1511 			goto next;
1512 		}
1513 		if (!buffer_verified(bh) &&
1514 		    !is_dx_internal_node(dir, block,
1515 					 (struct ext4_dir_entry *)bh->b_data) &&
1516 		    !ext4_dirent_csum_verify(dir,
1517 				(struct ext4_dir_entry *)bh->b_data)) {
1518 			EXT4_ERROR_INODE(dir, "checksumming directory "
1519 					 "block %lu", (unsigned long)block);
1520 			brelse(bh);
1521 			goto next;
1522 		}
1523 		set_buffer_verified(bh);
1524 		i = search_dirblock(bh, dir, d_name,
1525 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1526 		if (i == 1) {
1527 			EXT4_I(dir)->i_dir_start_lookup = block;
1528 			ret = bh;
1529 			goto cleanup_and_exit;
1530 		} else {
1531 			brelse(bh);
1532 			if (i < 0)
1533 				goto cleanup_and_exit;
1534 		}
1535 	next:
1536 		if (++block >= nblocks)
1537 			block = 0;
1538 	} while (block != start);
1539 
1540 	/*
1541 	 * If the directory has grown while we were searching, then
1542 	 * search the last part of the directory before giving up.
1543 	 */
1544 	block = nblocks;
1545 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1546 	if (block < nblocks) {
1547 		start = 0;
1548 		goto restart;
1549 	}
1550 
1551 cleanup_and_exit:
1552 	/* Clean up the read-ahead blocks */
1553 	for (; ra_ptr < ra_max; ra_ptr++)
1554 		brelse(bh_use[ra_ptr]);
1555 	return ret;
1556 }
1557 
1558 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1559 		       struct ext4_dir_entry_2 **res_dir)
1560 {
1561 	struct super_block * sb = dir->i_sb;
1562 	struct dx_hash_info	hinfo;
1563 	struct dx_frame frames[2], *frame;
1564 	struct buffer_head *bh;
1565 	ext4_lblk_t block;
1566 	int retval;
1567 
1568 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1569 	*res_dir = NULL;
1570 #endif
1571 	frame = dx_probe(d_name, dir, &hinfo, frames);
1572 	if (IS_ERR(frame))
1573 		return (struct buffer_head *) frame;
1574 	do {
1575 		block = dx_get_block(frame->at);
1576 		bh = ext4_read_dirblock(dir, block, DIRENT);
1577 		if (IS_ERR(bh))
1578 			goto errout;
1579 
1580 		retval = search_dirblock(bh, dir, d_name,
1581 					 block << EXT4_BLOCK_SIZE_BITS(sb),
1582 					 res_dir);
1583 		if (retval == 1)
1584 			goto success;
1585 		brelse(bh);
1586 		if (retval == -1) {
1587 			bh = ERR_PTR(ERR_BAD_DX_DIR);
1588 			goto errout;
1589 		}
1590 
1591 		/* Check to see if we should continue to search */
1592 		retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1593 					       frames, NULL);
1594 		if (retval < 0) {
1595 			ext4_warning(sb,
1596 			     "error %d reading index page in directory #%lu",
1597 			     retval, dir->i_ino);
1598 			bh = ERR_PTR(retval);
1599 			goto errout;
1600 		}
1601 	} while (retval == 1);
1602 
1603 	bh = NULL;
1604 errout:
1605 	dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1606 success:
1607 	dx_release(frames);
1608 	return bh;
1609 }
1610 
1611 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1612 {
1613 	struct inode *inode;
1614 	struct ext4_dir_entry_2 *de;
1615 	struct buffer_head *bh;
1616 
1617 	if (dentry->d_name.len > EXT4_NAME_LEN)
1618 		return ERR_PTR(-ENAMETOOLONG);
1619 
1620 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1621 	if (IS_ERR(bh))
1622 		return (struct dentry *) bh;
1623 	inode = NULL;
1624 	if (bh) {
1625 		__u32 ino = le32_to_cpu(de->inode);
1626 		brelse(bh);
1627 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1628 			EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1629 			return ERR_PTR(-EIO);
1630 		}
1631 		if (unlikely(ino == dir->i_ino)) {
1632 			EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1633 					 dentry);
1634 			return ERR_PTR(-EIO);
1635 		}
1636 		inode = ext4_iget_normal(dir->i_sb, ino);
1637 		if (inode == ERR_PTR(-ESTALE)) {
1638 			EXT4_ERROR_INODE(dir,
1639 					 "deleted inode referenced: %u",
1640 					 ino);
1641 			return ERR_PTR(-EIO);
1642 		}
1643 		if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1644 		    (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1645 		     S_ISLNK(inode->i_mode)) &&
1646 		    !ext4_is_child_context_consistent_with_parent(dir,
1647 								  inode)) {
1648 			iput(inode);
1649 			ext4_warning(inode->i_sb,
1650 				     "Inconsistent encryption contexts: %lu/%lu\n",
1651 				     (unsigned long) dir->i_ino,
1652 				     (unsigned long) inode->i_ino);
1653 			return ERR_PTR(-EPERM);
1654 		}
1655 	}
1656 	return d_splice_alias(inode, dentry);
1657 }
1658 
1659 
1660 struct dentry *ext4_get_parent(struct dentry *child)
1661 {
1662 	__u32 ino;
1663 	static const struct qstr dotdot = QSTR_INIT("..", 2);
1664 	struct ext4_dir_entry_2 * de;
1665 	struct buffer_head *bh;
1666 
1667 	bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1668 	if (IS_ERR(bh))
1669 		return (struct dentry *) bh;
1670 	if (!bh)
1671 		return ERR_PTR(-ENOENT);
1672 	ino = le32_to_cpu(de->inode);
1673 	brelse(bh);
1674 
1675 	if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1676 		EXT4_ERROR_INODE(child->d_inode,
1677 				 "bad parent inode number: %u", ino);
1678 		return ERR_PTR(-EIO);
1679 	}
1680 
1681 	return d_obtain_alias(ext4_iget_normal(child->d_inode->i_sb, ino));
1682 }
1683 
1684 /*
1685  * Move count entries from end of map between two memory locations.
1686  * Returns pointer to last entry moved.
1687  */
1688 static struct ext4_dir_entry_2 *
1689 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1690 		unsigned blocksize)
1691 {
1692 	unsigned rec_len = 0;
1693 
1694 	while (count--) {
1695 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1696 						(from + (map->offs<<2));
1697 		rec_len = EXT4_DIR_REC_LEN(de->name_len);
1698 		memcpy (to, de, rec_len);
1699 		((struct ext4_dir_entry_2 *) to)->rec_len =
1700 				ext4_rec_len_to_disk(rec_len, blocksize);
1701 		de->inode = 0;
1702 		map++;
1703 		to += rec_len;
1704 	}
1705 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1706 }
1707 
1708 /*
1709  * Compact each dir entry in the range to the minimal rec_len.
1710  * Returns pointer to last entry in range.
1711  */
1712 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1713 {
1714 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1715 	unsigned rec_len = 0;
1716 
1717 	prev = to = de;
1718 	while ((char*)de < base + blocksize) {
1719 		next = ext4_next_entry(de, blocksize);
1720 		if (de->inode && de->name_len) {
1721 			rec_len = EXT4_DIR_REC_LEN(de->name_len);
1722 			if (de > to)
1723 				memmove(to, de, rec_len);
1724 			to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1725 			prev = to;
1726 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1727 		}
1728 		de = next;
1729 	}
1730 	return prev;
1731 }
1732 
1733 /*
1734  * Split a full leaf block to make room for a new dir entry.
1735  * Allocate a new block, and move entries so that they are approx. equally full.
1736  * Returns pointer to de in block into which the new entry will be inserted.
1737  */
1738 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1739 			struct buffer_head **bh,struct dx_frame *frame,
1740 			struct dx_hash_info *hinfo)
1741 {
1742 	unsigned blocksize = dir->i_sb->s_blocksize;
1743 	unsigned count, continued;
1744 	struct buffer_head *bh2;
1745 	ext4_lblk_t newblock;
1746 	u32 hash2;
1747 	struct dx_map_entry *map;
1748 	char *data1 = (*bh)->b_data, *data2;
1749 	unsigned split, move, size;
1750 	struct ext4_dir_entry_2 *de = NULL, *de2;
1751 	struct ext4_dir_entry_tail *t;
1752 	int	csum_size = 0;
1753 	int	err = 0, i;
1754 
1755 	if (ext4_has_metadata_csum(dir->i_sb))
1756 		csum_size = sizeof(struct ext4_dir_entry_tail);
1757 
1758 	bh2 = ext4_append(handle, dir, &newblock);
1759 	if (IS_ERR(bh2)) {
1760 		brelse(*bh);
1761 		*bh = NULL;
1762 		return (struct ext4_dir_entry_2 *) bh2;
1763 	}
1764 
1765 	BUFFER_TRACE(*bh, "get_write_access");
1766 	err = ext4_journal_get_write_access(handle, *bh);
1767 	if (err)
1768 		goto journal_error;
1769 
1770 	BUFFER_TRACE(frame->bh, "get_write_access");
1771 	err = ext4_journal_get_write_access(handle, frame->bh);
1772 	if (err)
1773 		goto journal_error;
1774 
1775 	data2 = bh2->b_data;
1776 
1777 	/* create map in the end of data2 block */
1778 	map = (struct dx_map_entry *) (data2 + blocksize);
1779 	count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1780 			     blocksize, hinfo, map);
1781 	map -= count;
1782 	dx_sort_map(map, count);
1783 	/* Split the existing block in the middle, size-wise */
1784 	size = 0;
1785 	move = 0;
1786 	for (i = count-1; i >= 0; i--) {
1787 		/* is more than half of this entry in 2nd half of the block? */
1788 		if (size + map[i].size/2 > blocksize/2)
1789 			break;
1790 		size += map[i].size;
1791 		move++;
1792 	}
1793 	/* map index at which we will split */
1794 	split = count - move;
1795 	hash2 = map[split].hash;
1796 	continued = hash2 == map[split - 1].hash;
1797 	dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1798 			(unsigned long)dx_get_block(frame->at),
1799 					hash2, split, count-split));
1800 
1801 	/* Fancy dance to stay within two buffers */
1802 	de2 = dx_move_dirents(data1, data2, map + split, count - split,
1803 			      blocksize);
1804 	de = dx_pack_dirents(data1, blocksize);
1805 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1806 					   (char *) de,
1807 					   blocksize);
1808 	de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1809 					    (char *) de2,
1810 					    blocksize);
1811 	if (csum_size) {
1812 		t = EXT4_DIRENT_TAIL(data2, blocksize);
1813 		initialize_dirent_tail(t, blocksize);
1814 
1815 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1816 		initialize_dirent_tail(t, blocksize);
1817 	}
1818 
1819 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1820 			blocksize, 1));
1821 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1822 			blocksize, 1));
1823 
1824 	/* Which block gets the new entry? */
1825 	if (hinfo->hash >= hash2) {
1826 		swap(*bh, bh2);
1827 		de = de2;
1828 	}
1829 	dx_insert_block(frame, hash2 + continued, newblock);
1830 	err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1831 	if (err)
1832 		goto journal_error;
1833 	err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1834 	if (err)
1835 		goto journal_error;
1836 	brelse(bh2);
1837 	dxtrace(dx_show_index("frame", frame->entries));
1838 	return de;
1839 
1840 journal_error:
1841 	brelse(*bh);
1842 	brelse(bh2);
1843 	*bh = NULL;
1844 	ext4_std_error(dir->i_sb, err);
1845 	return ERR_PTR(err);
1846 }
1847 
1848 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1849 		      struct buffer_head *bh,
1850 		      void *buf, int buf_size,
1851 		      const char *name, int namelen,
1852 		      struct ext4_dir_entry_2 **dest_de)
1853 {
1854 	struct ext4_dir_entry_2 *de;
1855 	unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1856 	int nlen, rlen;
1857 	unsigned int offset = 0;
1858 	char *top;
1859 	struct ext4_fname_crypto_ctx *ctx = NULL;
1860 	struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
1861 	int res;
1862 
1863 	ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
1864 	if (IS_ERR(ctx))
1865 		return -1;
1866 
1867 	if (ctx != NULL) {
1868 		/* Calculate record length needed to store the entry */
1869 		res = ext4_fname_crypto_namelen_on_disk(ctx, namelen);
1870 		if (res < 0) {
1871 			ext4_put_fname_crypto_ctx(&ctx);
1872 			return res;
1873 		}
1874 		reclen = EXT4_DIR_REC_LEN(res);
1875 
1876 		/* Allocate buffer to hold maximum name length */
1877 		res = ext4_fname_crypto_alloc_buffer(ctx, EXT4_NAME_LEN,
1878 						     &fname_crypto_str);
1879 		if (res < 0) {
1880 			ext4_put_fname_crypto_ctx(&ctx);
1881 			return -1;
1882 		}
1883 	}
1884 
1885 	de = (struct ext4_dir_entry_2 *)buf;
1886 	top = buf + buf_size - reclen;
1887 	while ((char *) de <= top) {
1888 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1889 					 buf, buf_size, offset)) {
1890 			res = -EIO;
1891 			goto return_result;
1892 		}
1893 		/* Provide crypto context and crypto buffer to ext4 match */
1894 		res = ext4_match(ctx, &fname_crypto_str, namelen, name, de);
1895 		if (res < 0)
1896 			goto return_result;
1897 		if (res > 0) {
1898 			res = -EEXIST;
1899 			goto return_result;
1900 		}
1901 		nlen = EXT4_DIR_REC_LEN(de->name_len);
1902 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1903 		if ((de->inode ? rlen - nlen : rlen) >= reclen)
1904 			break;
1905 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1906 		offset += rlen;
1907 	}
1908 
1909 	if ((char *) de > top)
1910 		res = -ENOSPC;
1911 	else {
1912 		*dest_de = de;
1913 		res = 0;
1914 	}
1915 return_result:
1916 	ext4_put_fname_crypto_ctx(&ctx);
1917 	ext4_fname_crypto_free_buffer(&fname_crypto_str);
1918 	return res;
1919 }
1920 
1921 int ext4_insert_dentry(struct inode *dir,
1922 		       struct inode *inode,
1923 		       struct ext4_dir_entry_2 *de,
1924 		       int buf_size,
1925 		       const struct qstr *iname,
1926 		       const char *name, int namelen)
1927 {
1928 
1929 	int nlen, rlen;
1930 	struct ext4_fname_crypto_ctx *ctx = NULL;
1931 	struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
1932 	struct ext4_str tmp_str;
1933 	int res;
1934 
1935 	ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
1936 	if (IS_ERR(ctx))
1937 		return -EIO;
1938 	/* By default, the input name would be written to the disk */
1939 	tmp_str.name = (unsigned char *)name;
1940 	tmp_str.len = namelen;
1941 	if (ctx != NULL) {
1942 		/* Directory is encrypted */
1943 		res = ext4_fname_crypto_alloc_buffer(ctx, EXT4_NAME_LEN,
1944 						     &fname_crypto_str);
1945 		if (res < 0) {
1946 			ext4_put_fname_crypto_ctx(&ctx);
1947 			return -ENOMEM;
1948 		}
1949 		res = ext4_fname_usr_to_disk(ctx, iname, &fname_crypto_str);
1950 		if (res < 0) {
1951 			ext4_put_fname_crypto_ctx(&ctx);
1952 			ext4_fname_crypto_free_buffer(&fname_crypto_str);
1953 			return res;
1954 		}
1955 		tmp_str.name = fname_crypto_str.name;
1956 		tmp_str.len = fname_crypto_str.len;
1957 	}
1958 
1959 	nlen = EXT4_DIR_REC_LEN(de->name_len);
1960 	rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1961 	if (de->inode) {
1962 		struct ext4_dir_entry_2 *de1 =
1963 			(struct ext4_dir_entry_2 *)((char *)de + nlen);
1964 		de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1965 		de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1966 		de = de1;
1967 	}
1968 	de->file_type = EXT4_FT_UNKNOWN;
1969 	de->inode = cpu_to_le32(inode->i_ino);
1970 	ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1971 	de->name_len = tmp_str.len;
1972 
1973 	memcpy(de->name, tmp_str.name, tmp_str.len);
1974 	ext4_put_fname_crypto_ctx(&ctx);
1975 	ext4_fname_crypto_free_buffer(&fname_crypto_str);
1976 	return 0;
1977 }
1978 
1979 /*
1980  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1981  * it points to a directory entry which is guaranteed to be large
1982  * enough for new directory entry.  If de is NULL, then
1983  * add_dirent_to_buf will attempt search the directory block for
1984  * space.  It will return -ENOSPC if no space is available, and -EIO
1985  * and -EEXIST if directory entry already exists.
1986  */
1987 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1988 			     struct inode *inode, struct ext4_dir_entry_2 *de,
1989 			     struct buffer_head *bh)
1990 {
1991 	struct inode	*dir = dentry->d_parent->d_inode;
1992 	const char	*name = dentry->d_name.name;
1993 	int		namelen = dentry->d_name.len;
1994 	unsigned int	blocksize = dir->i_sb->s_blocksize;
1995 	int		csum_size = 0;
1996 	int		err;
1997 
1998 	if (ext4_has_metadata_csum(inode->i_sb))
1999 		csum_size = sizeof(struct ext4_dir_entry_tail);
2000 
2001 	if (!de) {
2002 		err = ext4_find_dest_de(dir, inode,
2003 					bh, bh->b_data, blocksize - csum_size,
2004 					name, namelen, &de);
2005 		if (err)
2006 			return err;
2007 	}
2008 	BUFFER_TRACE(bh, "get_write_access");
2009 	err = ext4_journal_get_write_access(handle, bh);
2010 	if (err) {
2011 		ext4_std_error(dir->i_sb, err);
2012 		return err;
2013 	}
2014 
2015 	/* By now the buffer is marked for journaling. Due to crypto operations,
2016 	 * the following function call may fail */
2017 	err = ext4_insert_dentry(dir, inode, de, blocksize, &dentry->d_name,
2018 				 name, namelen);
2019 	if (err < 0)
2020 		return err;
2021 
2022 	/*
2023 	 * XXX shouldn't update any times until successful
2024 	 * completion of syscall, but too many callers depend
2025 	 * on this.
2026 	 *
2027 	 * XXX similarly, too many callers depend on
2028 	 * ext4_new_inode() setting the times, but error
2029 	 * recovery deletes the inode, so the worst that can
2030 	 * happen is that the times are slightly out of date
2031 	 * and/or different from the directory change time.
2032 	 */
2033 	dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
2034 	ext4_update_dx_flag(dir);
2035 	dir->i_version++;
2036 	ext4_mark_inode_dirty(handle, dir);
2037 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2038 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2039 	if (err)
2040 		ext4_std_error(dir->i_sb, err);
2041 	return 0;
2042 }
2043 
2044 /*
2045  * This converts a one block unindexed directory to a 3 block indexed
2046  * directory, and adds the dentry to the indexed directory.
2047  */
2048 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
2049 			    struct inode *inode, struct buffer_head *bh)
2050 {
2051 	struct inode	*dir = dentry->d_parent->d_inode;
2052 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2053 	struct ext4_fname_crypto_ctx *ctx = NULL;
2054 	int res;
2055 #else
2056 	const char	*name = dentry->d_name.name;
2057 	int		namelen = dentry->d_name.len;
2058 #endif
2059 	struct buffer_head *bh2;
2060 	struct dx_root	*root;
2061 	struct dx_frame	frames[2], *frame;
2062 	struct dx_entry *entries;
2063 	struct ext4_dir_entry_2	*de, *de2;
2064 	struct ext4_dir_entry_tail *t;
2065 	char		*data1, *top;
2066 	unsigned	len;
2067 	int		retval;
2068 	unsigned	blocksize;
2069 	struct dx_hash_info hinfo;
2070 	ext4_lblk_t  block;
2071 	struct fake_dirent *fde;
2072 	int csum_size = 0;
2073 
2074 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2075 	ctx = ext4_get_fname_crypto_ctx(dir, EXT4_NAME_LEN);
2076 	if (IS_ERR(ctx))
2077 		return PTR_ERR(ctx);
2078 #endif
2079 
2080 	if (ext4_has_metadata_csum(inode->i_sb))
2081 		csum_size = sizeof(struct ext4_dir_entry_tail);
2082 
2083 	blocksize =  dir->i_sb->s_blocksize;
2084 	dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2085 	BUFFER_TRACE(bh, "get_write_access");
2086 	retval = ext4_journal_get_write_access(handle, bh);
2087 	if (retval) {
2088 		ext4_std_error(dir->i_sb, retval);
2089 		brelse(bh);
2090 		return retval;
2091 	}
2092 	root = (struct dx_root *) bh->b_data;
2093 
2094 	/* The 0th block becomes the root, move the dirents out */
2095 	fde = &root->dotdot;
2096 	de = (struct ext4_dir_entry_2 *)((char *)fde +
2097 		ext4_rec_len_from_disk(fde->rec_len, blocksize));
2098 	if ((char *) de >= (((char *) root) + blocksize)) {
2099 		EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2100 		brelse(bh);
2101 		return -EIO;
2102 	}
2103 	len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2104 
2105 	/* Allocate new block for the 0th block's dirents */
2106 	bh2 = ext4_append(handle, dir, &block);
2107 	if (IS_ERR(bh2)) {
2108 		brelse(bh);
2109 		return PTR_ERR(bh2);
2110 	}
2111 	ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2112 	data1 = bh2->b_data;
2113 
2114 	memcpy (data1, de, len);
2115 	de = (struct ext4_dir_entry_2 *) data1;
2116 	top = data1 + len;
2117 	while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2118 		de = de2;
2119 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2120 					   (char *) de,
2121 					   blocksize);
2122 
2123 	if (csum_size) {
2124 		t = EXT4_DIRENT_TAIL(data1, blocksize);
2125 		initialize_dirent_tail(t, blocksize);
2126 	}
2127 
2128 	/* Initialize the root; the dot dirents already exist */
2129 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2130 	de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2131 					   blocksize);
2132 	memset (&root->info, 0, sizeof(root->info));
2133 	root->info.info_length = sizeof(root->info);
2134 	root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2135 	entries = root->entries;
2136 	dx_set_block(entries, 1);
2137 	dx_set_count(entries, 1);
2138 	dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2139 
2140 	/* Initialize as for dx_probe */
2141 	hinfo.hash_version = root->info.hash_version;
2142 	if (hinfo.hash_version <= DX_HASH_TEA)
2143 		hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2144 	hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2145 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2146 	res = ext4_fname_usr_to_hash(ctx, &dentry->d_name, &hinfo);
2147 	if (res < 0) {
2148 		ext4_put_fname_crypto_ctx(&ctx);
2149 		ext4_mark_inode_dirty(handle, dir);
2150 		brelse(bh);
2151 		return res;
2152 	}
2153 	ext4_put_fname_crypto_ctx(&ctx);
2154 #else
2155 	ext4fs_dirhash(name, namelen, &hinfo);
2156 #endif
2157 	memset(frames, 0, sizeof(frames));
2158 	frame = frames;
2159 	frame->entries = entries;
2160 	frame->at = entries;
2161 	frame->bh = bh;
2162 	bh = bh2;
2163 
2164 	retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2165 	if (retval)
2166 		goto out_frames;
2167 	retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
2168 	if (retval)
2169 		goto out_frames;
2170 
2171 	de = do_split(handle,dir, &bh, frame, &hinfo);
2172 	if (IS_ERR(de)) {
2173 		retval = PTR_ERR(de);
2174 		goto out_frames;
2175 	}
2176 	dx_release(frames);
2177 
2178 	retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
2179 	brelse(bh);
2180 	return retval;
2181 out_frames:
2182 	/*
2183 	 * Even if the block split failed, we have to properly write
2184 	 * out all the changes we did so far. Otherwise we can end up
2185 	 * with corrupted filesystem.
2186 	 */
2187 	ext4_mark_inode_dirty(handle, dir);
2188 	dx_release(frames);
2189 	return retval;
2190 }
2191 
2192 /*
2193  *	ext4_add_entry()
2194  *
2195  * adds a file entry to the specified directory, using the same
2196  * semantics as ext4_find_entry(). It returns NULL if it failed.
2197  *
2198  * NOTE!! The inode part of 'de' is left at 0 - which means you
2199  * may not sleep between calling this and putting something into
2200  * the entry, as someone else might have used it while you slept.
2201  */
2202 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2203 			  struct inode *inode)
2204 {
2205 	struct inode *dir = dentry->d_parent->d_inode;
2206 	struct buffer_head *bh = NULL;
2207 	struct ext4_dir_entry_2 *de;
2208 	struct ext4_dir_entry_tail *t;
2209 	struct super_block *sb;
2210 	int	retval;
2211 	int	dx_fallback=0;
2212 	unsigned blocksize;
2213 	ext4_lblk_t block, blocks;
2214 	int	csum_size = 0;
2215 
2216 	if (ext4_has_metadata_csum(inode->i_sb))
2217 		csum_size = sizeof(struct ext4_dir_entry_tail);
2218 
2219 	sb = dir->i_sb;
2220 	blocksize = sb->s_blocksize;
2221 	if (!dentry->d_name.len)
2222 		return -EINVAL;
2223 
2224 	if (ext4_has_inline_data(dir)) {
2225 		retval = ext4_try_add_inline_entry(handle, dentry, inode);
2226 		if (retval < 0)
2227 			return retval;
2228 		if (retval == 1) {
2229 			retval = 0;
2230 			goto out;
2231 		}
2232 	}
2233 
2234 	if (is_dx(dir)) {
2235 		retval = ext4_dx_add_entry(handle, dentry, inode);
2236 		if (!retval || (retval != ERR_BAD_DX_DIR))
2237 			goto out;
2238 		ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2239 		dx_fallback++;
2240 		ext4_mark_inode_dirty(handle, dir);
2241 	}
2242 	blocks = dir->i_size >> sb->s_blocksize_bits;
2243 	for (block = 0; block < blocks; block++) {
2244 		bh = ext4_read_dirblock(dir, block, DIRENT);
2245 		if (IS_ERR(bh))
2246 			return PTR_ERR(bh);
2247 
2248 		retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
2249 		if (retval != -ENOSPC)
2250 			goto out;
2251 
2252 		if (blocks == 1 && !dx_fallback &&
2253 		    EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
2254 			retval = make_indexed_dir(handle, dentry, inode, bh);
2255 			bh = NULL; /* make_indexed_dir releases bh */
2256 			goto out;
2257 		}
2258 		brelse(bh);
2259 	}
2260 	bh = ext4_append(handle, dir, &block);
2261 	if (IS_ERR(bh))
2262 		return PTR_ERR(bh);
2263 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2264 	de->inode = 0;
2265 	de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2266 
2267 	if (csum_size) {
2268 		t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2269 		initialize_dirent_tail(t, blocksize);
2270 	}
2271 
2272 	retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
2273 out:
2274 	brelse(bh);
2275 	if (retval == 0)
2276 		ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2277 	return retval;
2278 }
2279 
2280 /*
2281  * Returns 0 for success, or a negative error value
2282  */
2283 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
2284 			     struct inode *inode)
2285 {
2286 	struct dx_frame frames[2], *frame;
2287 	struct dx_entry *entries, *at;
2288 	struct dx_hash_info hinfo;
2289 	struct buffer_head *bh;
2290 	struct inode *dir = dentry->d_parent->d_inode;
2291 	struct super_block *sb = dir->i_sb;
2292 	struct ext4_dir_entry_2 *de;
2293 	int err;
2294 
2295 	frame = dx_probe(&dentry->d_name, dir, &hinfo, frames);
2296 	if (IS_ERR(frame))
2297 		return PTR_ERR(frame);
2298 	entries = frame->entries;
2299 	at = frame->at;
2300 	bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2301 	if (IS_ERR(bh)) {
2302 		err = PTR_ERR(bh);
2303 		bh = NULL;
2304 		goto cleanup;
2305 	}
2306 
2307 	BUFFER_TRACE(bh, "get_write_access");
2308 	err = ext4_journal_get_write_access(handle, bh);
2309 	if (err)
2310 		goto journal_error;
2311 
2312 	err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
2313 	if (err != -ENOSPC)
2314 		goto cleanup;
2315 
2316 	/* Block full, should compress but for now just split */
2317 	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2318 		       dx_get_count(entries), dx_get_limit(entries)));
2319 	/* Need to split index? */
2320 	if (dx_get_count(entries) == dx_get_limit(entries)) {
2321 		ext4_lblk_t newblock;
2322 		unsigned icount = dx_get_count(entries);
2323 		int levels = frame - frames;
2324 		struct dx_entry *entries2;
2325 		struct dx_node *node2;
2326 		struct buffer_head *bh2;
2327 
2328 		if (levels && (dx_get_count(frames->entries) ==
2329 			       dx_get_limit(frames->entries))) {
2330 			ext4_warning(sb, "Directory index full!");
2331 			err = -ENOSPC;
2332 			goto cleanup;
2333 		}
2334 		bh2 = ext4_append(handle, dir, &newblock);
2335 		if (IS_ERR(bh2)) {
2336 			err = PTR_ERR(bh2);
2337 			goto cleanup;
2338 		}
2339 		node2 = (struct dx_node *)(bh2->b_data);
2340 		entries2 = node2->entries;
2341 		memset(&node2->fake, 0, sizeof(struct fake_dirent));
2342 		node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2343 							   sb->s_blocksize);
2344 		BUFFER_TRACE(frame->bh, "get_write_access");
2345 		err = ext4_journal_get_write_access(handle, frame->bh);
2346 		if (err)
2347 			goto journal_error;
2348 		if (levels) {
2349 			unsigned icount1 = icount/2, icount2 = icount - icount1;
2350 			unsigned hash2 = dx_get_hash(entries + icount1);
2351 			dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2352 				       icount1, icount2));
2353 
2354 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2355 			err = ext4_journal_get_write_access(handle,
2356 							     frames[0].bh);
2357 			if (err)
2358 				goto journal_error;
2359 
2360 			memcpy((char *) entries2, (char *) (entries + icount1),
2361 			       icount2 * sizeof(struct dx_entry));
2362 			dx_set_count(entries, icount1);
2363 			dx_set_count(entries2, icount2);
2364 			dx_set_limit(entries2, dx_node_limit(dir));
2365 
2366 			/* Which index block gets the new entry? */
2367 			if (at - entries >= icount1) {
2368 				frame->at = at = at - entries - icount1 + entries2;
2369 				frame->entries = entries = entries2;
2370 				swap(frame->bh, bh2);
2371 			}
2372 			dx_insert_block(frames + 0, hash2, newblock);
2373 			dxtrace(dx_show_index("node", frames[1].entries));
2374 			dxtrace(dx_show_index("node",
2375 			       ((struct dx_node *) bh2->b_data)->entries));
2376 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2377 			if (err)
2378 				goto journal_error;
2379 			brelse (bh2);
2380 		} else {
2381 			dxtrace(printk(KERN_DEBUG
2382 				       "Creating second level index...\n"));
2383 			memcpy((char *) entries2, (char *) entries,
2384 			       icount * sizeof(struct dx_entry));
2385 			dx_set_limit(entries2, dx_node_limit(dir));
2386 
2387 			/* Set up root */
2388 			dx_set_count(entries, 1);
2389 			dx_set_block(entries + 0, newblock);
2390 			((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2391 
2392 			/* Add new access path frame */
2393 			frame = frames + 1;
2394 			frame->at = at = at - entries + entries2;
2395 			frame->entries = entries = entries2;
2396 			frame->bh = bh2;
2397 			err = ext4_journal_get_write_access(handle,
2398 							     frame->bh);
2399 			if (err)
2400 				goto journal_error;
2401 		}
2402 		err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2403 		if (err) {
2404 			ext4_std_error(inode->i_sb, err);
2405 			goto cleanup;
2406 		}
2407 	}
2408 	de = do_split(handle, dir, &bh, frame, &hinfo);
2409 	if (IS_ERR(de)) {
2410 		err = PTR_ERR(de);
2411 		goto cleanup;
2412 	}
2413 	err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2414 	goto cleanup;
2415 
2416 journal_error:
2417 	ext4_std_error(dir->i_sb, err);
2418 cleanup:
2419 	brelse(bh);
2420 	dx_release(frames);
2421 	return err;
2422 }
2423 
2424 /*
2425  * ext4_generic_delete_entry deletes a directory entry by merging it
2426  * with the previous entry
2427  */
2428 int ext4_generic_delete_entry(handle_t *handle,
2429 			      struct inode *dir,
2430 			      struct ext4_dir_entry_2 *de_del,
2431 			      struct buffer_head *bh,
2432 			      void *entry_buf,
2433 			      int buf_size,
2434 			      int csum_size)
2435 {
2436 	struct ext4_dir_entry_2 *de, *pde;
2437 	unsigned int blocksize = dir->i_sb->s_blocksize;
2438 	int i;
2439 
2440 	i = 0;
2441 	pde = NULL;
2442 	de = (struct ext4_dir_entry_2 *)entry_buf;
2443 	while (i < buf_size - csum_size) {
2444 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2445 					 bh->b_data, bh->b_size, i))
2446 			return -EIO;
2447 		if (de == de_del)  {
2448 			if (pde)
2449 				pde->rec_len = ext4_rec_len_to_disk(
2450 					ext4_rec_len_from_disk(pde->rec_len,
2451 							       blocksize) +
2452 					ext4_rec_len_from_disk(de->rec_len,
2453 							       blocksize),
2454 					blocksize);
2455 			else
2456 				de->inode = 0;
2457 			dir->i_version++;
2458 			return 0;
2459 		}
2460 		i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2461 		pde = de;
2462 		de = ext4_next_entry(de, blocksize);
2463 	}
2464 	return -ENOENT;
2465 }
2466 
2467 static int ext4_delete_entry(handle_t *handle,
2468 			     struct inode *dir,
2469 			     struct ext4_dir_entry_2 *de_del,
2470 			     struct buffer_head *bh)
2471 {
2472 	int err, csum_size = 0;
2473 
2474 	if (ext4_has_inline_data(dir)) {
2475 		int has_inline_data = 1;
2476 		err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2477 					       &has_inline_data);
2478 		if (has_inline_data)
2479 			return err;
2480 	}
2481 
2482 	if (ext4_has_metadata_csum(dir->i_sb))
2483 		csum_size = sizeof(struct ext4_dir_entry_tail);
2484 
2485 	BUFFER_TRACE(bh, "get_write_access");
2486 	err = ext4_journal_get_write_access(handle, bh);
2487 	if (unlikely(err))
2488 		goto out;
2489 
2490 	err = ext4_generic_delete_entry(handle, dir, de_del,
2491 					bh, bh->b_data,
2492 					dir->i_sb->s_blocksize, csum_size);
2493 	if (err)
2494 		goto out;
2495 
2496 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2497 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2498 	if (unlikely(err))
2499 		goto out;
2500 
2501 	return 0;
2502 out:
2503 	if (err != -ENOENT)
2504 		ext4_std_error(dir->i_sb, err);
2505 	return err;
2506 }
2507 
2508 /*
2509  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2510  * since this indicates that nlinks count was previously 1.
2511  */
2512 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2513 {
2514 	inc_nlink(inode);
2515 	if (is_dx(inode) && inode->i_nlink > 1) {
2516 		/* limit is 16-bit i_links_count */
2517 		if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2518 			set_nlink(inode, 1);
2519 			EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2520 					      EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2521 		}
2522 	}
2523 }
2524 
2525 /*
2526  * If a directory had nlink == 1, then we should let it be 1. This indicates
2527  * directory has >EXT4_LINK_MAX subdirs.
2528  */
2529 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2530 {
2531 	if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2532 		drop_nlink(inode);
2533 }
2534 
2535 
2536 static int ext4_add_nondir(handle_t *handle,
2537 		struct dentry *dentry, struct inode *inode)
2538 {
2539 	int err = ext4_add_entry(handle, dentry, inode);
2540 	if (!err) {
2541 		ext4_mark_inode_dirty(handle, inode);
2542 		unlock_new_inode(inode);
2543 		d_instantiate(dentry, inode);
2544 		return 0;
2545 	}
2546 	drop_nlink(inode);
2547 	unlock_new_inode(inode);
2548 	iput(inode);
2549 	return err;
2550 }
2551 
2552 /*
2553  * By the time this is called, we already have created
2554  * the directory cache entry for the new file, but it
2555  * is so far negative - it has no inode.
2556  *
2557  * If the create succeeds, we fill in the inode information
2558  * with d_instantiate().
2559  */
2560 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2561 		       bool excl)
2562 {
2563 	handle_t *handle;
2564 	struct inode *inode;
2565 	int err, credits, retries = 0;
2566 
2567 	dquot_initialize(dir);
2568 
2569 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2570 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2571 retry:
2572 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2573 					    NULL, EXT4_HT_DIR, credits);
2574 	handle = ext4_journal_current_handle();
2575 	err = PTR_ERR(inode);
2576 	if (!IS_ERR(inode)) {
2577 		inode->i_op = &ext4_file_inode_operations;
2578 		inode->i_fop = &ext4_file_operations;
2579 		ext4_set_aops(inode);
2580 		err = 0;
2581 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2582 		if (!err && (ext4_encrypted_inode(dir) ||
2583 			     DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)))) {
2584 			err = ext4_inherit_context(dir, inode);
2585 			if (err) {
2586 				clear_nlink(inode);
2587 				unlock_new_inode(inode);
2588 				iput(inode);
2589 			}
2590 		}
2591 #endif
2592 		if (!err)
2593 			err = ext4_add_nondir(handle, dentry, inode);
2594 		if (!err && IS_DIRSYNC(dir))
2595 			ext4_handle_sync(handle);
2596 	}
2597 	if (handle)
2598 		ext4_journal_stop(handle);
2599 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2600 		goto retry;
2601 	return err;
2602 }
2603 
2604 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2605 		      umode_t mode, dev_t rdev)
2606 {
2607 	handle_t *handle;
2608 	struct inode *inode;
2609 	int err, credits, retries = 0;
2610 
2611 	if (!new_valid_dev(rdev))
2612 		return -EINVAL;
2613 
2614 	dquot_initialize(dir);
2615 
2616 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2617 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2618 retry:
2619 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2620 					    NULL, EXT4_HT_DIR, credits);
2621 	handle = ext4_journal_current_handle();
2622 	err = PTR_ERR(inode);
2623 	if (!IS_ERR(inode)) {
2624 		init_special_inode(inode, inode->i_mode, rdev);
2625 		inode->i_op = &ext4_special_inode_operations;
2626 		err = ext4_add_nondir(handle, dentry, inode);
2627 		if (!err && IS_DIRSYNC(dir))
2628 			ext4_handle_sync(handle);
2629 	}
2630 	if (handle)
2631 		ext4_journal_stop(handle);
2632 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2633 		goto retry;
2634 	return err;
2635 }
2636 
2637 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2638 {
2639 	handle_t *handle;
2640 	struct inode *inode;
2641 	int err, retries = 0;
2642 
2643 	dquot_initialize(dir);
2644 
2645 retry:
2646 	inode = ext4_new_inode_start_handle(dir, mode,
2647 					    NULL, 0, NULL,
2648 					    EXT4_HT_DIR,
2649 			EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2650 			  4 + EXT4_XATTR_TRANS_BLOCKS);
2651 	handle = ext4_journal_current_handle();
2652 	err = PTR_ERR(inode);
2653 	if (!IS_ERR(inode)) {
2654 		inode->i_op = &ext4_file_inode_operations;
2655 		inode->i_fop = &ext4_file_operations;
2656 		ext4_set_aops(inode);
2657 		d_tmpfile(dentry, inode);
2658 		err = ext4_orphan_add(handle, inode);
2659 		if (err)
2660 			goto err_unlock_inode;
2661 		mark_inode_dirty(inode);
2662 		unlock_new_inode(inode);
2663 	}
2664 	if (handle)
2665 		ext4_journal_stop(handle);
2666 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2667 		goto retry;
2668 	return err;
2669 err_unlock_inode:
2670 	ext4_journal_stop(handle);
2671 	unlock_new_inode(inode);
2672 	return err;
2673 }
2674 
2675 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2676 			  struct ext4_dir_entry_2 *de,
2677 			  int blocksize, int csum_size,
2678 			  unsigned int parent_ino, int dotdot_real_len)
2679 {
2680 	de->inode = cpu_to_le32(inode->i_ino);
2681 	de->name_len = 1;
2682 	de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2683 					   blocksize);
2684 	strcpy(de->name, ".");
2685 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2686 
2687 	de = ext4_next_entry(de, blocksize);
2688 	de->inode = cpu_to_le32(parent_ino);
2689 	de->name_len = 2;
2690 	if (!dotdot_real_len)
2691 		de->rec_len = ext4_rec_len_to_disk(blocksize -
2692 					(csum_size + EXT4_DIR_REC_LEN(1)),
2693 					blocksize);
2694 	else
2695 		de->rec_len = ext4_rec_len_to_disk(
2696 				EXT4_DIR_REC_LEN(de->name_len), blocksize);
2697 	strcpy(de->name, "..");
2698 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2699 
2700 	return ext4_next_entry(de, blocksize);
2701 }
2702 
2703 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2704 			     struct inode *inode)
2705 {
2706 	struct buffer_head *dir_block = NULL;
2707 	struct ext4_dir_entry_2 *de;
2708 	struct ext4_dir_entry_tail *t;
2709 	ext4_lblk_t block = 0;
2710 	unsigned int blocksize = dir->i_sb->s_blocksize;
2711 	int csum_size = 0;
2712 	int err;
2713 
2714 	if (ext4_has_metadata_csum(dir->i_sb))
2715 		csum_size = sizeof(struct ext4_dir_entry_tail);
2716 
2717 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2718 		err = ext4_try_create_inline_dir(handle, dir, inode);
2719 		if (err < 0 && err != -ENOSPC)
2720 			goto out;
2721 		if (!err)
2722 			goto out;
2723 	}
2724 
2725 	inode->i_size = 0;
2726 	dir_block = ext4_append(handle, inode, &block);
2727 	if (IS_ERR(dir_block))
2728 		return PTR_ERR(dir_block);
2729 	de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2730 	ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2731 	set_nlink(inode, 2);
2732 	if (csum_size) {
2733 		t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2734 		initialize_dirent_tail(t, blocksize);
2735 	}
2736 
2737 	BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2738 	err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2739 	if (err)
2740 		goto out;
2741 	set_buffer_verified(dir_block);
2742 out:
2743 	brelse(dir_block);
2744 	return err;
2745 }
2746 
2747 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2748 {
2749 	handle_t *handle;
2750 	struct inode *inode;
2751 	int err, credits, retries = 0;
2752 
2753 	if (EXT4_DIR_LINK_MAX(dir))
2754 		return -EMLINK;
2755 
2756 	dquot_initialize(dir);
2757 
2758 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2759 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2760 retry:
2761 	inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2762 					    &dentry->d_name,
2763 					    0, NULL, EXT4_HT_DIR, credits);
2764 	handle = ext4_journal_current_handle();
2765 	err = PTR_ERR(inode);
2766 	if (IS_ERR(inode))
2767 		goto out_stop;
2768 
2769 	inode->i_op = &ext4_dir_inode_operations;
2770 	inode->i_fop = &ext4_dir_operations;
2771 	err = ext4_init_new_dir(handle, dir, inode);
2772 	if (err)
2773 		goto out_clear_inode;
2774 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2775 	if (ext4_encrypted_inode(dir) ||
2776 	    DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) {
2777 		err = ext4_inherit_context(dir, inode);
2778 		if (err)
2779 			goto out_clear_inode;
2780 	}
2781 #endif
2782 	err = ext4_mark_inode_dirty(handle, inode);
2783 	if (!err)
2784 		err = ext4_add_entry(handle, dentry, inode);
2785 	if (err) {
2786 out_clear_inode:
2787 		clear_nlink(inode);
2788 		unlock_new_inode(inode);
2789 		ext4_mark_inode_dirty(handle, inode);
2790 		iput(inode);
2791 		goto out_stop;
2792 	}
2793 	ext4_inc_count(handle, dir);
2794 	ext4_update_dx_flag(dir);
2795 	err = ext4_mark_inode_dirty(handle, dir);
2796 	if (err)
2797 		goto out_clear_inode;
2798 	unlock_new_inode(inode);
2799 	d_instantiate(dentry, inode);
2800 	if (IS_DIRSYNC(dir))
2801 		ext4_handle_sync(handle);
2802 
2803 out_stop:
2804 	if (handle)
2805 		ext4_journal_stop(handle);
2806 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2807 		goto retry;
2808 	return err;
2809 }
2810 
2811 /*
2812  * routine to check that the specified directory is empty (for rmdir)
2813  */
2814 int ext4_empty_dir(struct inode *inode)
2815 {
2816 	unsigned int offset;
2817 	struct buffer_head *bh;
2818 	struct ext4_dir_entry_2 *de, *de1;
2819 	struct super_block *sb;
2820 	int err = 0;
2821 
2822 	if (ext4_has_inline_data(inode)) {
2823 		int has_inline_data = 1;
2824 
2825 		err = empty_inline_dir(inode, &has_inline_data);
2826 		if (has_inline_data)
2827 			return err;
2828 	}
2829 
2830 	sb = inode->i_sb;
2831 	if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2832 		EXT4_ERROR_INODE(inode, "invalid size");
2833 		return 1;
2834 	}
2835 	bh = ext4_read_dirblock(inode, 0, EITHER);
2836 	if (IS_ERR(bh))
2837 		return 1;
2838 
2839 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2840 	de1 = ext4_next_entry(de, sb->s_blocksize);
2841 	if (le32_to_cpu(de->inode) != inode->i_ino ||
2842 			!le32_to_cpu(de1->inode) ||
2843 			strcmp(".", de->name) ||
2844 			strcmp("..", de1->name)) {
2845 		ext4_warning(inode->i_sb,
2846 			     "bad directory (dir #%lu) - no `.' or `..'",
2847 			     inode->i_ino);
2848 		brelse(bh);
2849 		return 1;
2850 	}
2851 	offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2852 		 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2853 	de = ext4_next_entry(de1, sb->s_blocksize);
2854 	while (offset < inode->i_size) {
2855 		if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2856 			unsigned int lblock;
2857 			err = 0;
2858 			brelse(bh);
2859 			lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2860 			bh = ext4_read_dirblock(inode, lblock, EITHER);
2861 			if (IS_ERR(bh))
2862 				return 1;
2863 			de = (struct ext4_dir_entry_2 *) bh->b_data;
2864 		}
2865 		if (ext4_check_dir_entry(inode, NULL, de, bh,
2866 					 bh->b_data, bh->b_size, offset)) {
2867 			de = (struct ext4_dir_entry_2 *)(bh->b_data +
2868 							 sb->s_blocksize);
2869 			offset = (offset | (sb->s_blocksize - 1)) + 1;
2870 			continue;
2871 		}
2872 		if (le32_to_cpu(de->inode)) {
2873 			brelse(bh);
2874 			return 0;
2875 		}
2876 		offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2877 		de = ext4_next_entry(de, sb->s_blocksize);
2878 	}
2879 	brelse(bh);
2880 	return 1;
2881 }
2882 
2883 /*
2884  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2885  * such inodes, starting at the superblock, in case we crash before the
2886  * file is closed/deleted, or in case the inode truncate spans multiple
2887  * transactions and the last transaction is not recovered after a crash.
2888  *
2889  * At filesystem recovery time, we walk this list deleting unlinked
2890  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2891  *
2892  * Orphan list manipulation functions must be called under i_mutex unless
2893  * we are just creating the inode or deleting it.
2894  */
2895 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2896 {
2897 	struct super_block *sb = inode->i_sb;
2898 	struct ext4_sb_info *sbi = EXT4_SB(sb);
2899 	struct ext4_iloc iloc;
2900 	int err = 0, rc;
2901 	bool dirty = false;
2902 
2903 	if (!sbi->s_journal || is_bad_inode(inode))
2904 		return 0;
2905 
2906 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2907 		     !mutex_is_locked(&inode->i_mutex));
2908 	/*
2909 	 * Exit early if inode already is on orphan list. This is a big speedup
2910 	 * since we don't have to contend on the global s_orphan_lock.
2911 	 */
2912 	if (!list_empty(&EXT4_I(inode)->i_orphan))
2913 		return 0;
2914 
2915 	/*
2916 	 * Orphan handling is only valid for files with data blocks
2917 	 * being truncated, or files being unlinked. Note that we either
2918 	 * hold i_mutex, or the inode can not be referenced from outside,
2919 	 * so i_nlink should not be bumped due to race
2920 	 */
2921 	J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2922 		  S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2923 
2924 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2925 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2926 	if (err)
2927 		goto out;
2928 
2929 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2930 	if (err)
2931 		goto out;
2932 
2933 	mutex_lock(&sbi->s_orphan_lock);
2934 	/*
2935 	 * Due to previous errors inode may be already a part of on-disk
2936 	 * orphan list. If so skip on-disk list modification.
2937 	 */
2938 	if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2939 	    (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2940 		/* Insert this inode at the head of the on-disk orphan list */
2941 		NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2942 		sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2943 		dirty = true;
2944 	}
2945 	list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2946 	mutex_unlock(&sbi->s_orphan_lock);
2947 
2948 	if (dirty) {
2949 		err = ext4_handle_dirty_super(handle, sb);
2950 		rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2951 		if (!err)
2952 			err = rc;
2953 		if (err) {
2954 			/*
2955 			 * We have to remove inode from in-memory list if
2956 			 * addition to on disk orphan list failed. Stray orphan
2957 			 * list entries can cause panics at unmount time.
2958 			 */
2959 			mutex_lock(&sbi->s_orphan_lock);
2960 			list_del(&EXT4_I(inode)->i_orphan);
2961 			mutex_unlock(&sbi->s_orphan_lock);
2962 		}
2963 	}
2964 	jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2965 	jbd_debug(4, "orphan inode %lu will point to %d\n",
2966 			inode->i_ino, NEXT_ORPHAN(inode));
2967 out:
2968 	ext4_std_error(sb, err);
2969 	return err;
2970 }
2971 
2972 /*
2973  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2974  * of such inodes stored on disk, because it is finally being cleaned up.
2975  */
2976 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2977 {
2978 	struct list_head *prev;
2979 	struct ext4_inode_info *ei = EXT4_I(inode);
2980 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2981 	__u32 ino_next;
2982 	struct ext4_iloc iloc;
2983 	int err = 0;
2984 
2985 	if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2986 		return 0;
2987 
2988 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2989 		     !mutex_is_locked(&inode->i_mutex));
2990 	/* Do this quick check before taking global s_orphan_lock. */
2991 	if (list_empty(&ei->i_orphan))
2992 		return 0;
2993 
2994 	if (handle) {
2995 		/* Grab inode buffer early before taking global s_orphan_lock */
2996 		err = ext4_reserve_inode_write(handle, inode, &iloc);
2997 	}
2998 
2999 	mutex_lock(&sbi->s_orphan_lock);
3000 	jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3001 
3002 	prev = ei->i_orphan.prev;
3003 	list_del_init(&ei->i_orphan);
3004 
3005 	/* If we're on an error path, we may not have a valid
3006 	 * transaction handle with which to update the orphan list on
3007 	 * disk, but we still need to remove the inode from the linked
3008 	 * list in memory. */
3009 	if (!handle || err) {
3010 		mutex_unlock(&sbi->s_orphan_lock);
3011 		goto out_err;
3012 	}
3013 
3014 	ino_next = NEXT_ORPHAN(inode);
3015 	if (prev == &sbi->s_orphan) {
3016 		jbd_debug(4, "superblock will point to %u\n", ino_next);
3017 		BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3018 		err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3019 		if (err) {
3020 			mutex_unlock(&sbi->s_orphan_lock);
3021 			goto out_brelse;
3022 		}
3023 		sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3024 		mutex_unlock(&sbi->s_orphan_lock);
3025 		err = ext4_handle_dirty_super(handle, inode->i_sb);
3026 	} else {
3027 		struct ext4_iloc iloc2;
3028 		struct inode *i_prev =
3029 			&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3030 
3031 		jbd_debug(4, "orphan inode %lu will point to %u\n",
3032 			  i_prev->i_ino, ino_next);
3033 		err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3034 		if (err) {
3035 			mutex_unlock(&sbi->s_orphan_lock);
3036 			goto out_brelse;
3037 		}
3038 		NEXT_ORPHAN(i_prev) = ino_next;
3039 		err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3040 		mutex_unlock(&sbi->s_orphan_lock);
3041 	}
3042 	if (err)
3043 		goto out_brelse;
3044 	NEXT_ORPHAN(inode) = 0;
3045 	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3046 out_err:
3047 	ext4_std_error(inode->i_sb, err);
3048 	return err;
3049 
3050 out_brelse:
3051 	brelse(iloc.bh);
3052 	goto out_err;
3053 }
3054 
3055 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3056 {
3057 	int retval;
3058 	struct inode *inode;
3059 	struct buffer_head *bh;
3060 	struct ext4_dir_entry_2 *de;
3061 	handle_t *handle = NULL;
3062 
3063 	/* Initialize quotas before so that eventual writes go in
3064 	 * separate transaction */
3065 	dquot_initialize(dir);
3066 	dquot_initialize(dentry->d_inode);
3067 
3068 	retval = -ENOENT;
3069 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3070 	if (IS_ERR(bh))
3071 		return PTR_ERR(bh);
3072 	if (!bh)
3073 		goto end_rmdir;
3074 
3075 	inode = dentry->d_inode;
3076 
3077 	retval = -EIO;
3078 	if (le32_to_cpu(de->inode) != inode->i_ino)
3079 		goto end_rmdir;
3080 
3081 	retval = -ENOTEMPTY;
3082 	if (!ext4_empty_dir(inode))
3083 		goto end_rmdir;
3084 
3085 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3086 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3087 	if (IS_ERR(handle)) {
3088 		retval = PTR_ERR(handle);
3089 		handle = NULL;
3090 		goto end_rmdir;
3091 	}
3092 
3093 	if (IS_DIRSYNC(dir))
3094 		ext4_handle_sync(handle);
3095 
3096 	retval = ext4_delete_entry(handle, dir, de, bh);
3097 	if (retval)
3098 		goto end_rmdir;
3099 	if (!EXT4_DIR_LINK_EMPTY(inode))
3100 		ext4_warning(inode->i_sb,
3101 			     "empty directory has too many links (%d)",
3102 			     inode->i_nlink);
3103 	inode->i_version++;
3104 	clear_nlink(inode);
3105 	/* There's no need to set i_disksize: the fact that i_nlink is
3106 	 * zero will ensure that the right thing happens during any
3107 	 * recovery. */
3108 	inode->i_size = 0;
3109 	ext4_orphan_add(handle, inode);
3110 	inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
3111 	ext4_mark_inode_dirty(handle, inode);
3112 	ext4_dec_count(handle, dir);
3113 	ext4_update_dx_flag(dir);
3114 	ext4_mark_inode_dirty(handle, dir);
3115 
3116 end_rmdir:
3117 	brelse(bh);
3118 	if (handle)
3119 		ext4_journal_stop(handle);
3120 	return retval;
3121 }
3122 
3123 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3124 {
3125 	int retval;
3126 	struct inode *inode;
3127 	struct buffer_head *bh;
3128 	struct ext4_dir_entry_2 *de;
3129 	handle_t *handle = NULL;
3130 
3131 	trace_ext4_unlink_enter(dir, dentry);
3132 	/* Initialize quotas before so that eventual writes go
3133 	 * in separate transaction */
3134 	dquot_initialize(dir);
3135 	dquot_initialize(dentry->d_inode);
3136 
3137 	retval = -ENOENT;
3138 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3139 	if (IS_ERR(bh))
3140 		return PTR_ERR(bh);
3141 	if (!bh)
3142 		goto end_unlink;
3143 
3144 	inode = dentry->d_inode;
3145 
3146 	retval = -EIO;
3147 	if (le32_to_cpu(de->inode) != inode->i_ino)
3148 		goto end_unlink;
3149 
3150 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3151 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3152 	if (IS_ERR(handle)) {
3153 		retval = PTR_ERR(handle);
3154 		handle = NULL;
3155 		goto end_unlink;
3156 	}
3157 
3158 	if (IS_DIRSYNC(dir))
3159 		ext4_handle_sync(handle);
3160 
3161 	if (!inode->i_nlink) {
3162 		ext4_warning(inode->i_sb,
3163 			     "Deleting nonexistent file (%lu), %d",
3164 			     inode->i_ino, inode->i_nlink);
3165 		set_nlink(inode, 1);
3166 	}
3167 	retval = ext4_delete_entry(handle, dir, de, bh);
3168 	if (retval)
3169 		goto end_unlink;
3170 	dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3171 	ext4_update_dx_flag(dir);
3172 	ext4_mark_inode_dirty(handle, dir);
3173 	drop_nlink(inode);
3174 	if (!inode->i_nlink)
3175 		ext4_orphan_add(handle, inode);
3176 	inode->i_ctime = ext4_current_time(inode);
3177 	ext4_mark_inode_dirty(handle, inode);
3178 
3179 end_unlink:
3180 	brelse(bh);
3181 	if (handle)
3182 		ext4_journal_stop(handle);
3183 	trace_ext4_unlink_exit(dentry, retval);
3184 	return retval;
3185 }
3186 
3187 static int ext4_symlink(struct inode *dir,
3188 			struct dentry *dentry, const char *symname)
3189 {
3190 	handle_t *handle;
3191 	struct inode *inode;
3192 	int err, len = strlen(symname);
3193 	int credits;
3194 	bool encryption_required;
3195 	struct ext4_str disk_link;
3196 	struct ext4_encrypted_symlink_data *sd = NULL;
3197 
3198 	disk_link.len = len + 1;
3199 	disk_link.name = (char *) symname;
3200 
3201 	encryption_required = (ext4_encrypted_inode(dir) ||
3202 			       DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3203 	if (encryption_required)
3204 		disk_link.len = encrypted_symlink_data_len(len) + 1;
3205 	if (disk_link.len > dir->i_sb->s_blocksize)
3206 		return -ENAMETOOLONG;
3207 
3208 	dquot_initialize(dir);
3209 
3210 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3211 		/*
3212 		 * For non-fast symlinks, we just allocate inode and put it on
3213 		 * orphan list in the first transaction => we need bitmap,
3214 		 * group descriptor, sb, inode block, quota blocks, and
3215 		 * possibly selinux xattr blocks.
3216 		 */
3217 		credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3218 			  EXT4_XATTR_TRANS_BLOCKS;
3219 	} else {
3220 		/*
3221 		 * Fast symlink. We have to add entry to directory
3222 		 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3223 		 * allocate new inode (bitmap, group descriptor, inode block,
3224 		 * quota blocks, sb is already counted in previous macros).
3225 		 */
3226 		credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3227 			  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3228 	}
3229 
3230 	inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3231 					    &dentry->d_name, 0, NULL,
3232 					    EXT4_HT_DIR, credits);
3233 	handle = ext4_journal_current_handle();
3234 	if (IS_ERR(inode)) {
3235 		if (handle)
3236 			ext4_journal_stop(handle);
3237 		return PTR_ERR(inode);
3238 	}
3239 
3240 	if (encryption_required) {
3241 		struct ext4_fname_crypto_ctx *ctx = NULL;
3242 		struct qstr istr;
3243 		struct ext4_str ostr;
3244 
3245 		sd = kzalloc(disk_link.len, GFP_NOFS);
3246 		if (!sd) {
3247 			err = -ENOMEM;
3248 			goto err_drop_inode;
3249 		}
3250 		err = ext4_inherit_context(dir, inode);
3251 		if (err)
3252 			goto err_drop_inode;
3253 		ctx = ext4_get_fname_crypto_ctx(inode,
3254 						inode->i_sb->s_blocksize);
3255 		if (IS_ERR_OR_NULL(ctx)) {
3256 			/* We just set the policy, so ctx should not be NULL */
3257 			err = (ctx == NULL) ? -EIO : PTR_ERR(ctx);
3258 			goto err_drop_inode;
3259 		}
3260 		istr.name = (const unsigned char *) symname;
3261 		istr.len = len;
3262 		ostr.name = sd->encrypted_path;
3263 		err = ext4_fname_usr_to_disk(ctx, &istr, &ostr);
3264 		ext4_put_fname_crypto_ctx(&ctx);
3265 		if (err < 0)
3266 			goto err_drop_inode;
3267 		sd->len = cpu_to_le16(ostr.len);
3268 		disk_link.name = (char *) sd;
3269 	}
3270 
3271 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3272 		inode->i_op = &ext4_symlink_inode_operations;
3273 		ext4_set_aops(inode);
3274 		/*
3275 		 * We cannot call page_symlink() with transaction started
3276 		 * because it calls into ext4_write_begin() which can wait
3277 		 * for transaction commit if we are running out of space
3278 		 * and thus we deadlock. So we have to stop transaction now
3279 		 * and restart it when symlink contents is written.
3280 		 *
3281 		 * To keep fs consistent in case of crash, we have to put inode
3282 		 * to orphan list in the mean time.
3283 		 */
3284 		drop_nlink(inode);
3285 		err = ext4_orphan_add(handle, inode);
3286 		ext4_journal_stop(handle);
3287 		handle = NULL;
3288 		if (err)
3289 			goto err_drop_inode;
3290 		err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3291 		if (err)
3292 			goto err_drop_inode;
3293 		/*
3294 		 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3295 		 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3296 		 */
3297 		handle = ext4_journal_start(dir, EXT4_HT_DIR,
3298 				EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3299 				EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3300 		if (IS_ERR(handle)) {
3301 			err = PTR_ERR(handle);
3302 			handle = NULL;
3303 			goto err_drop_inode;
3304 		}
3305 		set_nlink(inode, 1);
3306 		err = ext4_orphan_del(handle, inode);
3307 		if (err)
3308 			goto err_drop_inode;
3309 	} else {
3310 		/* clear the extent format for fast symlink */
3311 		ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3312 		inode->i_op = encryption_required ?
3313 			&ext4_symlink_inode_operations :
3314 			&ext4_fast_symlink_inode_operations;
3315 		memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3316 		       disk_link.len);
3317 		inode->i_size = disk_link.len - 1;
3318 	}
3319 	EXT4_I(inode)->i_disksize = inode->i_size;
3320 	err = ext4_add_nondir(handle, dentry, inode);
3321 	if (!err && IS_DIRSYNC(dir))
3322 		ext4_handle_sync(handle);
3323 
3324 	if (handle)
3325 		ext4_journal_stop(handle);
3326 	kfree(sd);
3327 	return err;
3328 err_drop_inode:
3329 	if (handle)
3330 		ext4_journal_stop(handle);
3331 	kfree(sd);
3332 	clear_nlink(inode);
3333 	unlock_new_inode(inode);
3334 	iput(inode);
3335 	return err;
3336 }
3337 
3338 static int ext4_link(struct dentry *old_dentry,
3339 		     struct inode *dir, struct dentry *dentry)
3340 {
3341 	handle_t *handle;
3342 	struct inode *inode = old_dentry->d_inode;
3343 	int err, retries = 0;
3344 
3345 	if (inode->i_nlink >= EXT4_LINK_MAX)
3346 		return -EMLINK;
3347 	if (ext4_encrypted_inode(dir) &&
3348 	    !ext4_is_child_context_consistent_with_parent(dir, inode))
3349 		return -EPERM;
3350 	dquot_initialize(dir);
3351 
3352 retry:
3353 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3354 		(EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3355 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3356 	if (IS_ERR(handle))
3357 		return PTR_ERR(handle);
3358 
3359 	if (IS_DIRSYNC(dir))
3360 		ext4_handle_sync(handle);
3361 
3362 	inode->i_ctime = ext4_current_time(inode);
3363 	ext4_inc_count(handle, inode);
3364 	ihold(inode);
3365 
3366 	err = ext4_add_entry(handle, dentry, inode);
3367 	if (!err) {
3368 		ext4_mark_inode_dirty(handle, inode);
3369 		/* this can happen only for tmpfile being
3370 		 * linked the first time
3371 		 */
3372 		if (inode->i_nlink == 1)
3373 			ext4_orphan_del(handle, inode);
3374 		d_instantiate(dentry, inode);
3375 	} else {
3376 		drop_nlink(inode);
3377 		iput(inode);
3378 	}
3379 	ext4_journal_stop(handle);
3380 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3381 		goto retry;
3382 	return err;
3383 }
3384 
3385 
3386 /*
3387  * Try to find buffer head where contains the parent block.
3388  * It should be the inode block if it is inlined or the 1st block
3389  * if it is a normal dir.
3390  */
3391 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3392 					struct inode *inode,
3393 					int *retval,
3394 					struct ext4_dir_entry_2 **parent_de,
3395 					int *inlined)
3396 {
3397 	struct buffer_head *bh;
3398 
3399 	if (!ext4_has_inline_data(inode)) {
3400 		bh = ext4_read_dirblock(inode, 0, EITHER);
3401 		if (IS_ERR(bh)) {
3402 			*retval = PTR_ERR(bh);
3403 			return NULL;
3404 		}
3405 		*parent_de = ext4_next_entry(
3406 					(struct ext4_dir_entry_2 *)bh->b_data,
3407 					inode->i_sb->s_blocksize);
3408 		return bh;
3409 	}
3410 
3411 	*inlined = 1;
3412 	return ext4_get_first_inline_block(inode, parent_de, retval);
3413 }
3414 
3415 struct ext4_renament {
3416 	struct inode *dir;
3417 	struct dentry *dentry;
3418 	struct inode *inode;
3419 	bool is_dir;
3420 	int dir_nlink_delta;
3421 
3422 	/* entry for "dentry" */
3423 	struct buffer_head *bh;
3424 	struct ext4_dir_entry_2 *de;
3425 	int inlined;
3426 
3427 	/* entry for ".." in inode if it's a directory */
3428 	struct buffer_head *dir_bh;
3429 	struct ext4_dir_entry_2 *parent_de;
3430 	int dir_inlined;
3431 };
3432 
3433 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3434 {
3435 	int retval;
3436 
3437 	ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3438 					      &retval, &ent->parent_de,
3439 					      &ent->dir_inlined);
3440 	if (!ent->dir_bh)
3441 		return retval;
3442 	if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3443 		return -EIO;
3444 	BUFFER_TRACE(ent->dir_bh, "get_write_access");
3445 	return ext4_journal_get_write_access(handle, ent->dir_bh);
3446 }
3447 
3448 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3449 				  unsigned dir_ino)
3450 {
3451 	int retval;
3452 
3453 	ent->parent_de->inode = cpu_to_le32(dir_ino);
3454 	BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3455 	if (!ent->dir_inlined) {
3456 		if (is_dx(ent->inode)) {
3457 			retval = ext4_handle_dirty_dx_node(handle,
3458 							   ent->inode,
3459 							   ent->dir_bh);
3460 		} else {
3461 			retval = ext4_handle_dirty_dirent_node(handle,
3462 							       ent->inode,
3463 							       ent->dir_bh);
3464 		}
3465 	} else {
3466 		retval = ext4_mark_inode_dirty(handle, ent->inode);
3467 	}
3468 	if (retval) {
3469 		ext4_std_error(ent->dir->i_sb, retval);
3470 		return retval;
3471 	}
3472 	return 0;
3473 }
3474 
3475 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3476 		       unsigned ino, unsigned file_type)
3477 {
3478 	int retval;
3479 
3480 	BUFFER_TRACE(ent->bh, "get write access");
3481 	retval = ext4_journal_get_write_access(handle, ent->bh);
3482 	if (retval)
3483 		return retval;
3484 	ent->de->inode = cpu_to_le32(ino);
3485 	if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb,
3486 				      EXT4_FEATURE_INCOMPAT_FILETYPE))
3487 		ent->de->file_type = file_type;
3488 	ent->dir->i_version++;
3489 	ent->dir->i_ctime = ent->dir->i_mtime =
3490 		ext4_current_time(ent->dir);
3491 	ext4_mark_inode_dirty(handle, ent->dir);
3492 	BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3493 	if (!ent->inlined) {
3494 		retval = ext4_handle_dirty_dirent_node(handle,
3495 						       ent->dir, ent->bh);
3496 		if (unlikely(retval)) {
3497 			ext4_std_error(ent->dir->i_sb, retval);
3498 			return retval;
3499 		}
3500 	}
3501 	brelse(ent->bh);
3502 	ent->bh = NULL;
3503 
3504 	return 0;
3505 }
3506 
3507 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3508 				  const struct qstr *d_name)
3509 {
3510 	int retval = -ENOENT;
3511 	struct buffer_head *bh;
3512 	struct ext4_dir_entry_2 *de;
3513 
3514 	bh = ext4_find_entry(dir, d_name, &de, NULL);
3515 	if (IS_ERR(bh))
3516 		return PTR_ERR(bh);
3517 	if (bh) {
3518 		retval = ext4_delete_entry(handle, dir, de, bh);
3519 		brelse(bh);
3520 	}
3521 	return retval;
3522 }
3523 
3524 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3525 			       int force_reread)
3526 {
3527 	int retval;
3528 	/*
3529 	 * ent->de could have moved from under us during htree split, so make
3530 	 * sure that we are deleting the right entry.  We might also be pointing
3531 	 * to a stale entry in the unused part of ent->bh so just checking inum
3532 	 * and the name isn't enough.
3533 	 */
3534 	if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3535 	    ent->de->name_len != ent->dentry->d_name.len ||
3536 	    strncmp(ent->de->name, ent->dentry->d_name.name,
3537 		    ent->de->name_len) ||
3538 	    force_reread) {
3539 		retval = ext4_find_delete_entry(handle, ent->dir,
3540 						&ent->dentry->d_name);
3541 	} else {
3542 		retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3543 		if (retval == -ENOENT) {
3544 			retval = ext4_find_delete_entry(handle, ent->dir,
3545 							&ent->dentry->d_name);
3546 		}
3547 	}
3548 
3549 	if (retval) {
3550 		ext4_warning(ent->dir->i_sb,
3551 				"Deleting old file (%lu), %d, error=%d",
3552 				ent->dir->i_ino, ent->dir->i_nlink, retval);
3553 	}
3554 }
3555 
3556 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3557 {
3558 	if (ent->dir_nlink_delta) {
3559 		if (ent->dir_nlink_delta == -1)
3560 			ext4_dec_count(handle, ent->dir);
3561 		else
3562 			ext4_inc_count(handle, ent->dir);
3563 		ext4_mark_inode_dirty(handle, ent->dir);
3564 	}
3565 }
3566 
3567 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3568 					      int credits, handle_t **h)
3569 {
3570 	struct inode *wh;
3571 	handle_t *handle;
3572 	int retries = 0;
3573 
3574 	/*
3575 	 * for inode block, sb block, group summaries,
3576 	 * and inode bitmap
3577 	 */
3578 	credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3579 		    EXT4_XATTR_TRANS_BLOCKS + 4);
3580 retry:
3581 	wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3582 					 &ent->dentry->d_name, 0, NULL,
3583 					 EXT4_HT_DIR, credits);
3584 
3585 	handle = ext4_journal_current_handle();
3586 	if (IS_ERR(wh)) {
3587 		if (handle)
3588 			ext4_journal_stop(handle);
3589 		if (PTR_ERR(wh) == -ENOSPC &&
3590 		    ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3591 			goto retry;
3592 	} else {
3593 		*h = handle;
3594 		init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3595 		wh->i_op = &ext4_special_inode_operations;
3596 	}
3597 	return wh;
3598 }
3599 
3600 /*
3601  * Anybody can rename anything with this: the permission checks are left to the
3602  * higher-level routines.
3603  *
3604  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3605  * while new_{dentry,inode) refers to the destination dentry/inode
3606  * This comes from rename(const char *oldpath, const char *newpath)
3607  */
3608 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3609 		       struct inode *new_dir, struct dentry *new_dentry,
3610 		       unsigned int flags)
3611 {
3612 	handle_t *handle = NULL;
3613 	struct ext4_renament old = {
3614 		.dir = old_dir,
3615 		.dentry = old_dentry,
3616 		.inode = old_dentry->d_inode,
3617 	};
3618 	struct ext4_renament new = {
3619 		.dir = new_dir,
3620 		.dentry = new_dentry,
3621 		.inode = new_dentry->d_inode,
3622 	};
3623 	int force_reread;
3624 	int retval;
3625 	struct inode *whiteout = NULL;
3626 	int credits;
3627 	u8 old_file_type;
3628 
3629 	dquot_initialize(old.dir);
3630 	dquot_initialize(new.dir);
3631 
3632 	/* Initialize quotas before so that eventual writes go
3633 	 * in separate transaction */
3634 	if (new.inode)
3635 		dquot_initialize(new.inode);
3636 
3637 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3638 	if (IS_ERR(old.bh))
3639 		return PTR_ERR(old.bh);
3640 	/*
3641 	 *  Check for inode number is _not_ due to possible IO errors.
3642 	 *  We might rmdir the source, keep it as pwd of some process
3643 	 *  and merrily kill the link to whatever was created under the
3644 	 *  same name. Goodbye sticky bit ;-<
3645 	 */
3646 	retval = -ENOENT;
3647 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3648 		goto end_rename;
3649 
3650 	if ((old.dir != new.dir) &&
3651 	    ext4_encrypted_inode(new.dir) &&
3652 	    !ext4_is_child_context_consistent_with_parent(new.dir,
3653 							  old.inode)) {
3654 		retval = -EPERM;
3655 		goto end_rename;
3656 	}
3657 
3658 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3659 				 &new.de, &new.inlined);
3660 	if (IS_ERR(new.bh)) {
3661 		retval = PTR_ERR(new.bh);
3662 		new.bh = NULL;
3663 		goto end_rename;
3664 	}
3665 	if (new.bh) {
3666 		if (!new.inode) {
3667 			brelse(new.bh);
3668 			new.bh = NULL;
3669 		}
3670 	}
3671 	if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3672 		ext4_alloc_da_blocks(old.inode);
3673 
3674 	credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3675 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3676 	if (!(flags & RENAME_WHITEOUT)) {
3677 		handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3678 		if (IS_ERR(handle)) {
3679 			retval = PTR_ERR(handle);
3680 			handle = NULL;
3681 			goto end_rename;
3682 		}
3683 	} else {
3684 		whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3685 		if (IS_ERR(whiteout)) {
3686 			retval = PTR_ERR(whiteout);
3687 			whiteout = NULL;
3688 			goto end_rename;
3689 		}
3690 	}
3691 
3692 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3693 		ext4_handle_sync(handle);
3694 
3695 	if (S_ISDIR(old.inode->i_mode)) {
3696 		if (new.inode) {
3697 			retval = -ENOTEMPTY;
3698 			if (!ext4_empty_dir(new.inode))
3699 				goto end_rename;
3700 		} else {
3701 			retval = -EMLINK;
3702 			if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3703 				goto end_rename;
3704 		}
3705 		retval = ext4_rename_dir_prepare(handle, &old);
3706 		if (retval)
3707 			goto end_rename;
3708 	}
3709 	/*
3710 	 * If we're renaming a file within an inline_data dir and adding or
3711 	 * setting the new dirent causes a conversion from inline_data to
3712 	 * extents/blockmap, we need to force the dirent delete code to
3713 	 * re-read the directory, or else we end up trying to delete a dirent
3714 	 * from what is now the extent tree root (or a block map).
3715 	 */
3716 	force_reread = (new.dir->i_ino == old.dir->i_ino &&
3717 			ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3718 
3719 	old_file_type = old.de->file_type;
3720 	if (whiteout) {
3721 		/*
3722 		 * Do this before adding a new entry, so the old entry is sure
3723 		 * to be still pointing to the valid old entry.
3724 		 */
3725 		retval = ext4_setent(handle, &old, whiteout->i_ino,
3726 				     EXT4_FT_CHRDEV);
3727 		if (retval)
3728 			goto end_rename;
3729 		ext4_mark_inode_dirty(handle, whiteout);
3730 	}
3731 	if (!new.bh) {
3732 		retval = ext4_add_entry(handle, new.dentry, old.inode);
3733 		if (retval)
3734 			goto end_rename;
3735 	} else {
3736 		retval = ext4_setent(handle, &new,
3737 				     old.inode->i_ino, old_file_type);
3738 		if (retval)
3739 			goto end_rename;
3740 	}
3741 	if (force_reread)
3742 		force_reread = !ext4_test_inode_flag(new.dir,
3743 						     EXT4_INODE_INLINE_DATA);
3744 
3745 	/*
3746 	 * Like most other Unix systems, set the ctime for inodes on a
3747 	 * rename.
3748 	 */
3749 	old.inode->i_ctime = ext4_current_time(old.inode);
3750 	ext4_mark_inode_dirty(handle, old.inode);
3751 
3752 	if (!whiteout) {
3753 		/*
3754 		 * ok, that's it
3755 		 */
3756 		ext4_rename_delete(handle, &old, force_reread);
3757 	}
3758 
3759 	if (new.inode) {
3760 		ext4_dec_count(handle, new.inode);
3761 		new.inode->i_ctime = ext4_current_time(new.inode);
3762 	}
3763 	old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3764 	ext4_update_dx_flag(old.dir);
3765 	if (old.dir_bh) {
3766 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3767 		if (retval)
3768 			goto end_rename;
3769 
3770 		ext4_dec_count(handle, old.dir);
3771 		if (new.inode) {
3772 			/* checked ext4_empty_dir above, can't have another
3773 			 * parent, ext4_dec_count() won't work for many-linked
3774 			 * dirs */
3775 			clear_nlink(new.inode);
3776 		} else {
3777 			ext4_inc_count(handle, new.dir);
3778 			ext4_update_dx_flag(new.dir);
3779 			ext4_mark_inode_dirty(handle, new.dir);
3780 		}
3781 	}
3782 	ext4_mark_inode_dirty(handle, old.dir);
3783 	if (new.inode) {
3784 		ext4_mark_inode_dirty(handle, new.inode);
3785 		if (!new.inode->i_nlink)
3786 			ext4_orphan_add(handle, new.inode);
3787 	}
3788 	retval = 0;
3789 
3790 end_rename:
3791 	brelse(old.dir_bh);
3792 	brelse(old.bh);
3793 	brelse(new.bh);
3794 	if (whiteout) {
3795 		if (retval)
3796 			drop_nlink(whiteout);
3797 		unlock_new_inode(whiteout);
3798 		iput(whiteout);
3799 	}
3800 	if (handle)
3801 		ext4_journal_stop(handle);
3802 	return retval;
3803 }
3804 
3805 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3806 			     struct inode *new_dir, struct dentry *new_dentry)
3807 {
3808 	handle_t *handle = NULL;
3809 	struct ext4_renament old = {
3810 		.dir = old_dir,
3811 		.dentry = old_dentry,
3812 		.inode = old_dentry->d_inode,
3813 	};
3814 	struct ext4_renament new = {
3815 		.dir = new_dir,
3816 		.dentry = new_dentry,
3817 		.inode = new_dentry->d_inode,
3818 	};
3819 	u8 new_file_type;
3820 	int retval;
3821 
3822 	dquot_initialize(old.dir);
3823 	dquot_initialize(new.dir);
3824 
3825 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3826 				 &old.de, &old.inlined);
3827 	if (IS_ERR(old.bh))
3828 		return PTR_ERR(old.bh);
3829 	/*
3830 	 *  Check for inode number is _not_ due to possible IO errors.
3831 	 *  We might rmdir the source, keep it as pwd of some process
3832 	 *  and merrily kill the link to whatever was created under the
3833 	 *  same name. Goodbye sticky bit ;-<
3834 	 */
3835 	retval = -ENOENT;
3836 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3837 		goto end_rename;
3838 
3839 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3840 				 &new.de, &new.inlined);
3841 	if (IS_ERR(new.bh)) {
3842 		retval = PTR_ERR(new.bh);
3843 		new.bh = NULL;
3844 		goto end_rename;
3845 	}
3846 
3847 	/* RENAME_EXCHANGE case: old *and* new must both exist */
3848 	if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3849 		goto end_rename;
3850 
3851 	handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3852 		(2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3853 		 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3854 	if (IS_ERR(handle)) {
3855 		retval = PTR_ERR(handle);
3856 		handle = NULL;
3857 		goto end_rename;
3858 	}
3859 
3860 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3861 		ext4_handle_sync(handle);
3862 
3863 	if (S_ISDIR(old.inode->i_mode)) {
3864 		old.is_dir = true;
3865 		retval = ext4_rename_dir_prepare(handle, &old);
3866 		if (retval)
3867 			goto end_rename;
3868 	}
3869 	if (S_ISDIR(new.inode->i_mode)) {
3870 		new.is_dir = true;
3871 		retval = ext4_rename_dir_prepare(handle, &new);
3872 		if (retval)
3873 			goto end_rename;
3874 	}
3875 
3876 	/*
3877 	 * Other than the special case of overwriting a directory, parents'
3878 	 * nlink only needs to be modified if this is a cross directory rename.
3879 	 */
3880 	if (old.dir != new.dir && old.is_dir != new.is_dir) {
3881 		old.dir_nlink_delta = old.is_dir ? -1 : 1;
3882 		new.dir_nlink_delta = -old.dir_nlink_delta;
3883 		retval = -EMLINK;
3884 		if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3885 		    (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3886 			goto end_rename;
3887 	}
3888 
3889 	new_file_type = new.de->file_type;
3890 	retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3891 	if (retval)
3892 		goto end_rename;
3893 
3894 	retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3895 	if (retval)
3896 		goto end_rename;
3897 
3898 	/*
3899 	 * Like most other Unix systems, set the ctime for inodes on a
3900 	 * rename.
3901 	 */
3902 	old.inode->i_ctime = ext4_current_time(old.inode);
3903 	new.inode->i_ctime = ext4_current_time(new.inode);
3904 	ext4_mark_inode_dirty(handle, old.inode);
3905 	ext4_mark_inode_dirty(handle, new.inode);
3906 
3907 	if (old.dir_bh) {
3908 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3909 		if (retval)
3910 			goto end_rename;
3911 	}
3912 	if (new.dir_bh) {
3913 		retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3914 		if (retval)
3915 			goto end_rename;
3916 	}
3917 	ext4_update_dir_count(handle, &old);
3918 	ext4_update_dir_count(handle, &new);
3919 	retval = 0;
3920 
3921 end_rename:
3922 	brelse(old.dir_bh);
3923 	brelse(new.dir_bh);
3924 	brelse(old.bh);
3925 	brelse(new.bh);
3926 	if (handle)
3927 		ext4_journal_stop(handle);
3928 	return retval;
3929 }
3930 
3931 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3932 			struct inode *new_dir, struct dentry *new_dentry,
3933 			unsigned int flags)
3934 {
3935 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3936 		return -EINVAL;
3937 
3938 	if (flags & RENAME_EXCHANGE) {
3939 		return ext4_cross_rename(old_dir, old_dentry,
3940 					 new_dir, new_dentry);
3941 	}
3942 
3943 	return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3944 }
3945 
3946 /*
3947  * directories can handle most operations...
3948  */
3949 const struct inode_operations ext4_dir_inode_operations = {
3950 	.create		= ext4_create,
3951 	.lookup		= ext4_lookup,
3952 	.link		= ext4_link,
3953 	.unlink		= ext4_unlink,
3954 	.symlink	= ext4_symlink,
3955 	.mkdir		= ext4_mkdir,
3956 	.rmdir		= ext4_rmdir,
3957 	.mknod		= ext4_mknod,
3958 	.tmpfile	= ext4_tmpfile,
3959 	.rename2	= ext4_rename2,
3960 	.setattr	= ext4_setattr,
3961 	.setxattr	= generic_setxattr,
3962 	.getxattr	= generic_getxattr,
3963 	.listxattr	= ext4_listxattr,
3964 	.removexattr	= generic_removexattr,
3965 	.get_acl	= ext4_get_acl,
3966 	.set_acl	= ext4_set_acl,
3967 	.fiemap         = ext4_fiemap,
3968 };
3969 
3970 const struct inode_operations ext4_special_inode_operations = {
3971 	.setattr	= ext4_setattr,
3972 	.setxattr	= generic_setxattr,
3973 	.getxattr	= generic_getxattr,
3974 	.listxattr	= ext4_listxattr,
3975 	.removexattr	= generic_removexattr,
3976 	.get_acl	= ext4_get_acl,
3977 	.set_acl	= ext4_set_acl,
3978 };
3979