xref: /openbmc/linux/fs/ext4/namei.c (revision afb46f79)
1 /*
2  *  linux/fs/ext4/namei.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/namei.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  *  Directory entry file type support and forward compatibility hooks
18  *	for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19  *  Hash Tree Directory indexing (c)
20  *	Daniel Phillips, 2001
21  *  Hash Tree Directory indexing porting
22  *	Christopher Li, 2002
23  *  Hash Tree Directory indexing cleanup
24  *	Theodore Ts'o, 2002
25  */
26 
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include "ext4.h"
38 #include "ext4_jbd2.h"
39 
40 #include "xattr.h"
41 #include "acl.h"
42 
43 #include <trace/events/ext4.h>
44 /*
45  * define how far ahead to read directories while searching them.
46  */
47 #define NAMEI_RA_CHUNKS  2
48 #define NAMEI_RA_BLOCKS  4
49 #define NAMEI_RA_SIZE	     (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 
51 static struct buffer_head *ext4_append(handle_t *handle,
52 					struct inode *inode,
53 					ext4_lblk_t *block)
54 {
55 	struct buffer_head *bh;
56 	int err = 0;
57 
58 	if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
59 		     ((inode->i_size >> 10) >=
60 		      EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
61 		return ERR_PTR(-ENOSPC);
62 
63 	*block = inode->i_size >> inode->i_sb->s_blocksize_bits;
64 
65 	bh = ext4_bread(handle, inode, *block, 1, &err);
66 	if (!bh)
67 		return ERR_PTR(err);
68 	inode->i_size += inode->i_sb->s_blocksize;
69 	EXT4_I(inode)->i_disksize = inode->i_size;
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 err = 0, is_dx_block = 0;
97 
98 	bh = ext4_bread(NULL, inode, block, 0, &err);
99 	if (!bh) {
100 		if (err == 0) {
101 			ext4_error_inode(inode, __func__, line, block,
102 					       "Directory hole found");
103 			return ERR_PTR(-EIO);
104 		}
105 		__ext4_warning(inode->i_sb, __func__, line,
106 			       "error reading directory block "
107 			       "(ino %lu, block %lu)", inode->i_ino,
108 			       (unsigned long) block);
109 		return ERR_PTR(err);
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_RO_COMPAT_FEATURE(inode->i_sb,
127 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) ||
128 	    buffer_verified(bh))
129 		return bh;
130 
131 	/*
132 	 * An empty leaf block can get mistaken for a index block; for
133 	 * this reason, we can only check the index checksum when the
134 	 * caller is sure it should be an index block.
135 	 */
136 	if (is_dx_block && type == INDEX) {
137 		if (ext4_dx_csum_verify(inode, dirent))
138 			set_buffer_verified(bh);
139 		else {
140 			ext4_error_inode(inode, __func__, line, block,
141 				"Directory index failed checksum");
142 			brelse(bh);
143 			return ERR_PTR(-EIO);
144 		}
145 	}
146 	if (!is_dx_block) {
147 		if (ext4_dirent_csum_verify(inode, dirent))
148 			set_buffer_verified(bh);
149 		else {
150 			ext4_error_inode(inode, __func__, line, block,
151 				"Directory block failed checksum");
152 			brelse(bh);
153 			return ERR_PTR(-EIO);
154 		}
155 	}
156 	return bh;
157 }
158 
159 #ifndef assert
160 #define assert(test) J_ASSERT(test)
161 #endif
162 
163 #ifdef DX_DEBUG
164 #define dxtrace(command) command
165 #else
166 #define dxtrace(command)
167 #endif
168 
169 struct fake_dirent
170 {
171 	__le32 inode;
172 	__le16 rec_len;
173 	u8 name_len;
174 	u8 file_type;
175 };
176 
177 struct dx_countlimit
178 {
179 	__le16 limit;
180 	__le16 count;
181 };
182 
183 struct dx_entry
184 {
185 	__le32 hash;
186 	__le32 block;
187 };
188 
189 /*
190  * dx_root_info is laid out so that if it should somehow get overlaid by a
191  * dirent the two low bits of the hash version will be zero.  Therefore, the
192  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
193  */
194 
195 struct dx_root
196 {
197 	struct fake_dirent dot;
198 	char dot_name[4];
199 	struct fake_dirent dotdot;
200 	char dotdot_name[4];
201 	struct dx_root_info
202 	{
203 		__le32 reserved_zero;
204 		u8 hash_version;
205 		u8 info_length; /* 8 */
206 		u8 indirect_levels;
207 		u8 unused_flags;
208 	}
209 	info;
210 	struct dx_entry	entries[0];
211 };
212 
213 struct dx_node
214 {
215 	struct fake_dirent fake;
216 	struct dx_entry	entries[0];
217 };
218 
219 
220 struct dx_frame
221 {
222 	struct buffer_head *bh;
223 	struct dx_entry *entries;
224 	struct dx_entry *at;
225 };
226 
227 struct dx_map_entry
228 {
229 	u32 hash;
230 	u16 offs;
231 	u16 size;
232 };
233 
234 /*
235  * This goes at the end of each htree block.
236  */
237 struct dx_tail {
238 	u32 dt_reserved;
239 	__le32 dt_checksum;	/* crc32c(uuid+inum+dirblock) */
240 };
241 
242 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
243 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
244 static inline unsigned dx_get_hash(struct dx_entry *entry);
245 static void dx_set_hash(struct dx_entry *entry, unsigned value);
246 static unsigned dx_get_count(struct dx_entry *entries);
247 static unsigned dx_get_limit(struct dx_entry *entries);
248 static void dx_set_count(struct dx_entry *entries, unsigned value);
249 static void dx_set_limit(struct dx_entry *entries, unsigned value);
250 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
251 static unsigned dx_node_limit(struct inode *dir);
252 static struct dx_frame *dx_probe(const struct qstr *d_name,
253 				 struct inode *dir,
254 				 struct dx_hash_info *hinfo,
255 				 struct dx_frame *frame,
256 				 int *err);
257 static void dx_release(struct dx_frame *frames);
258 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
259 		       struct dx_hash_info *hinfo, struct dx_map_entry map[]);
260 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
261 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
262 		struct dx_map_entry *offsets, int count, unsigned blocksize);
263 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
264 static void dx_insert_block(struct dx_frame *frame,
265 					u32 hash, ext4_lblk_t block);
266 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
267 				 struct dx_frame *frame,
268 				 struct dx_frame *frames,
269 				 __u32 *start_hash);
270 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
271 		const struct qstr *d_name,
272 		struct ext4_dir_entry_2 **res_dir,
273 		int *err);
274 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
275 			     struct inode *inode);
276 
277 /* checksumming functions */
278 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
279 			    unsigned int blocksize)
280 {
281 	memset(t, 0, sizeof(struct ext4_dir_entry_tail));
282 	t->det_rec_len = ext4_rec_len_to_disk(
283 			sizeof(struct ext4_dir_entry_tail), blocksize);
284 	t->det_reserved_ft = EXT4_FT_DIR_CSUM;
285 }
286 
287 /* Walk through a dirent block to find a checksum "dirent" at the tail */
288 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
289 						   struct ext4_dir_entry *de)
290 {
291 	struct ext4_dir_entry_tail *t;
292 
293 #ifdef PARANOID
294 	struct ext4_dir_entry *d, *top;
295 
296 	d = de;
297 	top = (struct ext4_dir_entry *)(((void *)de) +
298 		(EXT4_BLOCK_SIZE(inode->i_sb) -
299 		sizeof(struct ext4_dir_entry_tail)));
300 	while (d < top && d->rec_len)
301 		d = (struct ext4_dir_entry *)(((void *)d) +
302 		    le16_to_cpu(d->rec_len));
303 
304 	if (d != top)
305 		return NULL;
306 
307 	t = (struct ext4_dir_entry_tail *)d;
308 #else
309 	t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
310 #endif
311 
312 	if (t->det_reserved_zero1 ||
313 	    le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
314 	    t->det_reserved_zero2 ||
315 	    t->det_reserved_ft != EXT4_FT_DIR_CSUM)
316 		return NULL;
317 
318 	return t;
319 }
320 
321 static __le32 ext4_dirent_csum(struct inode *inode,
322 			       struct ext4_dir_entry *dirent, int size)
323 {
324 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
325 	struct ext4_inode_info *ei = EXT4_I(inode);
326 	__u32 csum;
327 
328 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
329 	return cpu_to_le32(csum);
330 }
331 
332 static void warn_no_space_for_csum(struct inode *inode)
333 {
334 	ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
335 		     "checksum.  Please run e2fsck -D.", inode->i_ino);
336 }
337 
338 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
339 {
340 	struct ext4_dir_entry_tail *t;
341 
342 	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
343 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
344 		return 1;
345 
346 	t = get_dirent_tail(inode, dirent);
347 	if (!t) {
348 		warn_no_space_for_csum(inode);
349 		return 0;
350 	}
351 
352 	if (t->det_checksum != ext4_dirent_csum(inode, dirent,
353 						(void *)t - (void *)dirent))
354 		return 0;
355 
356 	return 1;
357 }
358 
359 static void ext4_dirent_csum_set(struct inode *inode,
360 				 struct ext4_dir_entry *dirent)
361 {
362 	struct ext4_dir_entry_tail *t;
363 
364 	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
365 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
366 		return;
367 
368 	t = get_dirent_tail(inode, dirent);
369 	if (!t) {
370 		warn_no_space_for_csum(inode);
371 		return;
372 	}
373 
374 	t->det_checksum = ext4_dirent_csum(inode, dirent,
375 					   (void *)t - (void *)dirent);
376 }
377 
378 int ext4_handle_dirty_dirent_node(handle_t *handle,
379 				  struct inode *inode,
380 				  struct buffer_head *bh)
381 {
382 	ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
383 	return ext4_handle_dirty_metadata(handle, inode, bh);
384 }
385 
386 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
387 					       struct ext4_dir_entry *dirent,
388 					       int *offset)
389 {
390 	struct ext4_dir_entry *dp;
391 	struct dx_root_info *root;
392 	int count_offset;
393 
394 	if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
395 		count_offset = 8;
396 	else if (le16_to_cpu(dirent->rec_len) == 12) {
397 		dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
398 		if (le16_to_cpu(dp->rec_len) !=
399 		    EXT4_BLOCK_SIZE(inode->i_sb) - 12)
400 			return NULL;
401 		root = (struct dx_root_info *)(((void *)dp + 12));
402 		if (root->reserved_zero ||
403 		    root->info_length != sizeof(struct dx_root_info))
404 			return NULL;
405 		count_offset = 32;
406 	} else
407 		return NULL;
408 
409 	if (offset)
410 		*offset = count_offset;
411 	return (struct dx_countlimit *)(((void *)dirent) + count_offset);
412 }
413 
414 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
415 			   int count_offset, int count, struct dx_tail *t)
416 {
417 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
418 	struct ext4_inode_info *ei = EXT4_I(inode);
419 	__u32 csum;
420 	__le32 save_csum;
421 	int size;
422 
423 	size = count_offset + (count * sizeof(struct dx_entry));
424 	save_csum = t->dt_checksum;
425 	t->dt_checksum = 0;
426 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
427 	csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
428 	t->dt_checksum = save_csum;
429 
430 	return cpu_to_le32(csum);
431 }
432 
433 static int ext4_dx_csum_verify(struct inode *inode,
434 			       struct ext4_dir_entry *dirent)
435 {
436 	struct dx_countlimit *c;
437 	struct dx_tail *t;
438 	int count_offset, limit, count;
439 
440 	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
441 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
442 		return 1;
443 
444 	c = get_dx_countlimit(inode, dirent, &count_offset);
445 	if (!c) {
446 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
447 		return 1;
448 	}
449 	limit = le16_to_cpu(c->limit);
450 	count = le16_to_cpu(c->count);
451 	if (count_offset + (limit * sizeof(struct dx_entry)) >
452 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
453 		warn_no_space_for_csum(inode);
454 		return 1;
455 	}
456 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
457 
458 	if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
459 					    count, t))
460 		return 0;
461 	return 1;
462 }
463 
464 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
465 {
466 	struct dx_countlimit *c;
467 	struct dx_tail *t;
468 	int count_offset, limit, count;
469 
470 	if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
471 					EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
472 		return;
473 
474 	c = get_dx_countlimit(inode, dirent, &count_offset);
475 	if (!c) {
476 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
477 		return;
478 	}
479 	limit = le16_to_cpu(c->limit);
480 	count = le16_to_cpu(c->count);
481 	if (count_offset + (limit * sizeof(struct dx_entry)) >
482 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
483 		warn_no_space_for_csum(inode);
484 		return;
485 	}
486 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
487 
488 	t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
489 }
490 
491 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
492 					    struct inode *inode,
493 					    struct buffer_head *bh)
494 {
495 	ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
496 	return ext4_handle_dirty_metadata(handle, inode, bh);
497 }
498 
499 /*
500  * p is at least 6 bytes before the end of page
501  */
502 static inline struct ext4_dir_entry_2 *
503 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
504 {
505 	return (struct ext4_dir_entry_2 *)((char *)p +
506 		ext4_rec_len_from_disk(p->rec_len, blocksize));
507 }
508 
509 /*
510  * Future: use high four bits of block for coalesce-on-delete flags
511  * Mask them off for now.
512  */
513 
514 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
515 {
516 	return le32_to_cpu(entry->block) & 0x00ffffff;
517 }
518 
519 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
520 {
521 	entry->block = cpu_to_le32(value);
522 }
523 
524 static inline unsigned dx_get_hash(struct dx_entry *entry)
525 {
526 	return le32_to_cpu(entry->hash);
527 }
528 
529 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
530 {
531 	entry->hash = cpu_to_le32(value);
532 }
533 
534 static inline unsigned dx_get_count(struct dx_entry *entries)
535 {
536 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
537 }
538 
539 static inline unsigned dx_get_limit(struct dx_entry *entries)
540 {
541 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
542 }
543 
544 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
545 {
546 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
547 }
548 
549 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
550 {
551 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
552 }
553 
554 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
555 {
556 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
557 		EXT4_DIR_REC_LEN(2) - infosize;
558 
559 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
560 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
561 		entry_space -= sizeof(struct dx_tail);
562 	return entry_space / sizeof(struct dx_entry);
563 }
564 
565 static inline unsigned dx_node_limit(struct inode *dir)
566 {
567 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
568 
569 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
570 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
571 		entry_space -= sizeof(struct dx_tail);
572 	return entry_space / sizeof(struct dx_entry);
573 }
574 
575 /*
576  * Debug
577  */
578 #ifdef DX_DEBUG
579 static void dx_show_index(char * label, struct dx_entry *entries)
580 {
581 	int i, n = dx_get_count (entries);
582 	printk(KERN_DEBUG "%s index ", label);
583 	for (i = 0; i < n; i++) {
584 		printk("%x->%lu ", i ? dx_get_hash(entries + i) :
585 				0, (unsigned long)dx_get_block(entries + i));
586 	}
587 	printk("\n");
588 }
589 
590 struct stats
591 {
592 	unsigned names;
593 	unsigned space;
594 	unsigned bcount;
595 };
596 
597 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
598 				 int size, int show_names)
599 {
600 	unsigned names = 0, space = 0;
601 	char *base = (char *) de;
602 	struct dx_hash_info h = *hinfo;
603 
604 	printk("names: ");
605 	while ((char *) de < base + size)
606 	{
607 		if (de->inode)
608 		{
609 			if (show_names)
610 			{
611 				int len = de->name_len;
612 				char *name = de->name;
613 				while (len--) printk("%c", *name++);
614 				ext4fs_dirhash(de->name, de->name_len, &h);
615 				printk(":%x.%u ", h.hash,
616 				       (unsigned) ((char *) de - base));
617 			}
618 			space += EXT4_DIR_REC_LEN(de->name_len);
619 			names++;
620 		}
621 		de = ext4_next_entry(de, size);
622 	}
623 	printk("(%i)\n", names);
624 	return (struct stats) { names, space, 1 };
625 }
626 
627 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
628 			     struct dx_entry *entries, int levels)
629 {
630 	unsigned blocksize = dir->i_sb->s_blocksize;
631 	unsigned count = dx_get_count(entries), names = 0, space = 0, i;
632 	unsigned bcount = 0;
633 	struct buffer_head *bh;
634 	int err;
635 	printk("%i indexed blocks...\n", count);
636 	for (i = 0; i < count; i++, entries++)
637 	{
638 		ext4_lblk_t block = dx_get_block(entries);
639 		ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
640 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
641 		struct stats stats;
642 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
643 		if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
644 		stats = levels?
645 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
646 		   dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
647 		names += stats.names;
648 		space += stats.space;
649 		bcount += stats.bcount;
650 		brelse(bh);
651 	}
652 	if (bcount)
653 		printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
654 		       levels ? "" : "   ", names, space/bcount,
655 		       (space/bcount)*100/blocksize);
656 	return (struct stats) { names, space, bcount};
657 }
658 #endif /* DX_DEBUG */
659 
660 /*
661  * Probe for a directory leaf block to search.
662  *
663  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
664  * error in the directory index, and the caller should fall back to
665  * searching the directory normally.  The callers of dx_probe **MUST**
666  * check for this error code, and make sure it never gets reflected
667  * back to userspace.
668  */
669 static struct dx_frame *
670 dx_probe(const struct qstr *d_name, struct inode *dir,
671 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
672 {
673 	unsigned count, indirect;
674 	struct dx_entry *at, *entries, *p, *q, *m;
675 	struct dx_root *root;
676 	struct buffer_head *bh;
677 	struct dx_frame *frame = frame_in;
678 	u32 hash;
679 
680 	frame->bh = NULL;
681 	bh = ext4_read_dirblock(dir, 0, INDEX);
682 	if (IS_ERR(bh)) {
683 		*err = PTR_ERR(bh);
684 		goto fail;
685 	}
686 	root = (struct dx_root *) bh->b_data;
687 	if (root->info.hash_version != DX_HASH_TEA &&
688 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
689 	    root->info.hash_version != DX_HASH_LEGACY) {
690 		ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
691 			     root->info.hash_version);
692 		brelse(bh);
693 		*err = ERR_BAD_DX_DIR;
694 		goto fail;
695 	}
696 	hinfo->hash_version = root->info.hash_version;
697 	if (hinfo->hash_version <= DX_HASH_TEA)
698 		hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
699 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
700 	if (d_name)
701 		ext4fs_dirhash(d_name->name, d_name->len, hinfo);
702 	hash = hinfo->hash;
703 
704 	if (root->info.unused_flags & 1) {
705 		ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
706 			     root->info.unused_flags);
707 		brelse(bh);
708 		*err = ERR_BAD_DX_DIR;
709 		goto fail;
710 	}
711 
712 	if ((indirect = root->info.indirect_levels) > 1) {
713 		ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
714 			     root->info.indirect_levels);
715 		brelse(bh);
716 		*err = ERR_BAD_DX_DIR;
717 		goto fail;
718 	}
719 
720 	entries = (struct dx_entry *) (((char *)&root->info) +
721 				       root->info.info_length);
722 
723 	if (dx_get_limit(entries) != dx_root_limit(dir,
724 						   root->info.info_length)) {
725 		ext4_warning(dir->i_sb, "dx entry: limit != root limit");
726 		brelse(bh);
727 		*err = ERR_BAD_DX_DIR;
728 		goto fail;
729 	}
730 
731 	dxtrace(printk("Look up %x", hash));
732 	while (1)
733 	{
734 		count = dx_get_count(entries);
735 		if (!count || count > dx_get_limit(entries)) {
736 			ext4_warning(dir->i_sb,
737 				     "dx entry: no count or count > limit");
738 			brelse(bh);
739 			*err = ERR_BAD_DX_DIR;
740 			goto fail2;
741 		}
742 
743 		p = entries + 1;
744 		q = entries + count - 1;
745 		while (p <= q)
746 		{
747 			m = p + (q - p)/2;
748 			dxtrace(printk("."));
749 			if (dx_get_hash(m) > hash)
750 				q = m - 1;
751 			else
752 				p = m + 1;
753 		}
754 
755 		if (0) // linear search cross check
756 		{
757 			unsigned n = count - 1;
758 			at = entries;
759 			while (n--)
760 			{
761 				dxtrace(printk(","));
762 				if (dx_get_hash(++at) > hash)
763 				{
764 					at--;
765 					break;
766 				}
767 			}
768 			assert (at == p - 1);
769 		}
770 
771 		at = p - 1;
772 		dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
773 		frame->bh = bh;
774 		frame->entries = entries;
775 		frame->at = at;
776 		if (!indirect--) return frame;
777 		bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
778 		if (IS_ERR(bh)) {
779 			*err = PTR_ERR(bh);
780 			goto fail2;
781 		}
782 		entries = ((struct dx_node *) bh->b_data)->entries;
783 
784 		if (dx_get_limit(entries) != dx_node_limit (dir)) {
785 			ext4_warning(dir->i_sb,
786 				     "dx entry: limit != node limit");
787 			brelse(bh);
788 			*err = ERR_BAD_DX_DIR;
789 			goto fail2;
790 		}
791 		frame++;
792 		frame->bh = NULL;
793 	}
794 fail2:
795 	while (frame >= frame_in) {
796 		brelse(frame->bh);
797 		frame--;
798 	}
799 fail:
800 	if (*err == ERR_BAD_DX_DIR)
801 		ext4_warning(dir->i_sb,
802 			     "Corrupt dir inode %lu, running e2fsck is "
803 			     "recommended.", dir->i_ino);
804 	return NULL;
805 }
806 
807 static void dx_release (struct dx_frame *frames)
808 {
809 	if (frames[0].bh == NULL)
810 		return;
811 
812 	if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
813 		brelse(frames[1].bh);
814 	brelse(frames[0].bh);
815 }
816 
817 /*
818  * This function increments the frame pointer to search the next leaf
819  * block, and reads in the necessary intervening nodes if the search
820  * should be necessary.  Whether or not the search is necessary is
821  * controlled by the hash parameter.  If the hash value is even, then
822  * the search is only continued if the next block starts with that
823  * hash value.  This is used if we are searching for a specific file.
824  *
825  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
826  *
827  * This function returns 1 if the caller should continue to search,
828  * or 0 if it should not.  If there is an error reading one of the
829  * index blocks, it will a negative error code.
830  *
831  * If start_hash is non-null, it will be filled in with the starting
832  * hash of the next page.
833  */
834 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
835 				 struct dx_frame *frame,
836 				 struct dx_frame *frames,
837 				 __u32 *start_hash)
838 {
839 	struct dx_frame *p;
840 	struct buffer_head *bh;
841 	int num_frames = 0;
842 	__u32 bhash;
843 
844 	p = frame;
845 	/*
846 	 * Find the next leaf page by incrementing the frame pointer.
847 	 * If we run out of entries in the interior node, loop around and
848 	 * increment pointer in the parent node.  When we break out of
849 	 * this loop, num_frames indicates the number of interior
850 	 * nodes need to be read.
851 	 */
852 	while (1) {
853 		if (++(p->at) < p->entries + dx_get_count(p->entries))
854 			break;
855 		if (p == frames)
856 			return 0;
857 		num_frames++;
858 		p--;
859 	}
860 
861 	/*
862 	 * If the hash is 1, then continue only if the next page has a
863 	 * continuation hash of any value.  This is used for readdir
864 	 * handling.  Otherwise, check to see if the hash matches the
865 	 * desired contiuation hash.  If it doesn't, return since
866 	 * there's no point to read in the successive index pages.
867 	 */
868 	bhash = dx_get_hash(p->at);
869 	if (start_hash)
870 		*start_hash = bhash;
871 	if ((hash & 1) == 0) {
872 		if ((bhash & ~1) != hash)
873 			return 0;
874 	}
875 	/*
876 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
877 	 * block so no check is necessary
878 	 */
879 	while (num_frames--) {
880 		bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
881 		if (IS_ERR(bh))
882 			return PTR_ERR(bh);
883 		p++;
884 		brelse(p->bh);
885 		p->bh = bh;
886 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
887 	}
888 	return 1;
889 }
890 
891 
892 /*
893  * This function fills a red-black tree with information from a
894  * directory block.  It returns the number directory entries loaded
895  * into the tree.  If there is an error it is returned in err.
896  */
897 static int htree_dirblock_to_tree(struct file *dir_file,
898 				  struct inode *dir, ext4_lblk_t block,
899 				  struct dx_hash_info *hinfo,
900 				  __u32 start_hash, __u32 start_minor_hash)
901 {
902 	struct buffer_head *bh;
903 	struct ext4_dir_entry_2 *de, *top;
904 	int err = 0, count = 0;
905 
906 	dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
907 							(unsigned long)block));
908 	bh = ext4_read_dirblock(dir, block, DIRENT);
909 	if (IS_ERR(bh))
910 		return PTR_ERR(bh);
911 
912 	de = (struct ext4_dir_entry_2 *) bh->b_data;
913 	top = (struct ext4_dir_entry_2 *) ((char *) de +
914 					   dir->i_sb->s_blocksize -
915 					   EXT4_DIR_REC_LEN(0));
916 	for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
917 		if (ext4_check_dir_entry(dir, NULL, de, bh,
918 				bh->b_data, bh->b_size,
919 				(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
920 					 + ((char *)de - bh->b_data))) {
921 			/* silently ignore the rest of the block */
922 			break;
923 		}
924 		ext4fs_dirhash(de->name, de->name_len, hinfo);
925 		if ((hinfo->hash < start_hash) ||
926 		    ((hinfo->hash == start_hash) &&
927 		     (hinfo->minor_hash < start_minor_hash)))
928 			continue;
929 		if (de->inode == 0)
930 			continue;
931 		if ((err = ext4_htree_store_dirent(dir_file,
932 				   hinfo->hash, hinfo->minor_hash, de)) != 0) {
933 			brelse(bh);
934 			return err;
935 		}
936 		count++;
937 	}
938 	brelse(bh);
939 	return count;
940 }
941 
942 
943 /*
944  * This function fills a red-black tree with information from a
945  * directory.  We start scanning the directory in hash order, starting
946  * at start_hash and start_minor_hash.
947  *
948  * This function returns the number of entries inserted into the tree,
949  * or a negative error code.
950  */
951 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
952 			 __u32 start_minor_hash, __u32 *next_hash)
953 {
954 	struct dx_hash_info hinfo;
955 	struct ext4_dir_entry_2 *de;
956 	struct dx_frame frames[2], *frame;
957 	struct inode *dir;
958 	ext4_lblk_t block;
959 	int count = 0;
960 	int ret, err;
961 	__u32 hashval;
962 
963 	dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
964 		       start_hash, start_minor_hash));
965 	dir = file_inode(dir_file);
966 	if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
967 		hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
968 		if (hinfo.hash_version <= DX_HASH_TEA)
969 			hinfo.hash_version +=
970 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
971 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
972 		if (ext4_has_inline_data(dir)) {
973 			int has_inline_data = 1;
974 			count = htree_inlinedir_to_tree(dir_file, dir, 0,
975 							&hinfo, start_hash,
976 							start_minor_hash,
977 							&has_inline_data);
978 			if (has_inline_data) {
979 				*next_hash = ~0;
980 				return count;
981 			}
982 		}
983 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
984 					       start_hash, start_minor_hash);
985 		*next_hash = ~0;
986 		return count;
987 	}
988 	hinfo.hash = start_hash;
989 	hinfo.minor_hash = 0;
990 	frame = dx_probe(NULL, dir, &hinfo, frames, &err);
991 	if (!frame)
992 		return err;
993 
994 	/* Add '.' and '..' from the htree header */
995 	if (!start_hash && !start_minor_hash) {
996 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
997 		if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
998 			goto errout;
999 		count++;
1000 	}
1001 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1002 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1003 		de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1004 		if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
1005 			goto errout;
1006 		count++;
1007 	}
1008 
1009 	while (1) {
1010 		block = dx_get_block(frame->at);
1011 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1012 					     start_hash, start_minor_hash);
1013 		if (ret < 0) {
1014 			err = ret;
1015 			goto errout;
1016 		}
1017 		count += ret;
1018 		hashval = ~0;
1019 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1020 					    frame, frames, &hashval);
1021 		*next_hash = hashval;
1022 		if (ret < 0) {
1023 			err = ret;
1024 			goto errout;
1025 		}
1026 		/*
1027 		 * Stop if:  (a) there are no more entries, or
1028 		 * (b) we have inserted at least one entry and the
1029 		 * next hash value is not a continuation
1030 		 */
1031 		if ((ret == 0) ||
1032 		    (count && ((hashval & 1) == 0)))
1033 			break;
1034 	}
1035 	dx_release(frames);
1036 	dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1037 		       "next hash: %x\n", count, *next_hash));
1038 	return count;
1039 errout:
1040 	dx_release(frames);
1041 	return (err);
1042 }
1043 
1044 static inline int search_dirblock(struct buffer_head *bh,
1045 				  struct inode *dir,
1046 				  const struct qstr *d_name,
1047 				  unsigned int offset,
1048 				  struct ext4_dir_entry_2 **res_dir)
1049 {
1050 	return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1051 			  d_name, offset, res_dir);
1052 }
1053 
1054 /*
1055  * Directory block splitting, compacting
1056  */
1057 
1058 /*
1059  * Create map of hash values, offsets, and sizes, stored at end of block.
1060  * Returns number of entries mapped.
1061  */
1062 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1063 		       struct dx_hash_info *hinfo,
1064 		       struct dx_map_entry *map_tail)
1065 {
1066 	int count = 0;
1067 	char *base = (char *) de;
1068 	struct dx_hash_info h = *hinfo;
1069 
1070 	while ((char *) de < base + blocksize) {
1071 		if (de->name_len && de->inode) {
1072 			ext4fs_dirhash(de->name, de->name_len, &h);
1073 			map_tail--;
1074 			map_tail->hash = h.hash;
1075 			map_tail->offs = ((char *) de - base)>>2;
1076 			map_tail->size = le16_to_cpu(de->rec_len);
1077 			count++;
1078 			cond_resched();
1079 		}
1080 		/* XXX: do we need to check rec_len == 0 case? -Chris */
1081 		de = ext4_next_entry(de, blocksize);
1082 	}
1083 	return count;
1084 }
1085 
1086 /* Sort map by hash value */
1087 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1088 {
1089 	struct dx_map_entry *p, *q, *top = map + count - 1;
1090 	int more;
1091 	/* Combsort until bubble sort doesn't suck */
1092 	while (count > 2) {
1093 		count = count*10/13;
1094 		if (count - 9 < 2) /* 9, 10 -> 11 */
1095 			count = 11;
1096 		for (p = top, q = p - count; q >= map; p--, q--)
1097 			if (p->hash < q->hash)
1098 				swap(*p, *q);
1099 	}
1100 	/* Garden variety bubble sort */
1101 	do {
1102 		more = 0;
1103 		q = top;
1104 		while (q-- > map) {
1105 			if (q[1].hash >= q[0].hash)
1106 				continue;
1107 			swap(*(q+1), *q);
1108 			more = 1;
1109 		}
1110 	} while(more);
1111 }
1112 
1113 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1114 {
1115 	struct dx_entry *entries = frame->entries;
1116 	struct dx_entry *old = frame->at, *new = old + 1;
1117 	int count = dx_get_count(entries);
1118 
1119 	assert(count < dx_get_limit(entries));
1120 	assert(old < entries + count);
1121 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1122 	dx_set_hash(new, hash);
1123 	dx_set_block(new, block);
1124 	dx_set_count(entries, count + 1);
1125 }
1126 
1127 /*
1128  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1129  *
1130  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1131  * `de != NULL' is guaranteed by caller.
1132  */
1133 static inline int ext4_match (int len, const char * const name,
1134 			      struct ext4_dir_entry_2 * de)
1135 {
1136 	if (len != de->name_len)
1137 		return 0;
1138 	if (!de->inode)
1139 		return 0;
1140 	return !memcmp(name, de->name, len);
1141 }
1142 
1143 /*
1144  * Returns 0 if not found, -1 on failure, and 1 on success
1145  */
1146 int search_dir(struct buffer_head *bh,
1147 	       char *search_buf,
1148 	       int buf_size,
1149 	       struct inode *dir,
1150 	       const struct qstr *d_name,
1151 	       unsigned int offset,
1152 	       struct ext4_dir_entry_2 **res_dir)
1153 {
1154 	struct ext4_dir_entry_2 * de;
1155 	char * dlimit;
1156 	int de_len;
1157 	const char *name = d_name->name;
1158 	int namelen = d_name->len;
1159 
1160 	de = (struct ext4_dir_entry_2 *)search_buf;
1161 	dlimit = search_buf + buf_size;
1162 	while ((char *) de < dlimit) {
1163 		/* this code is executed quadratically often */
1164 		/* do minimal checking `by hand' */
1165 
1166 		if ((char *) de + namelen <= dlimit &&
1167 		    ext4_match (namelen, name, de)) {
1168 			/* found a match - just to be sure, do a full check */
1169 			if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1170 						 bh->b_size, offset))
1171 				return -1;
1172 			*res_dir = de;
1173 			return 1;
1174 		}
1175 		/* prevent looping on a bad block */
1176 		de_len = ext4_rec_len_from_disk(de->rec_len,
1177 						dir->i_sb->s_blocksize);
1178 		if (de_len <= 0)
1179 			return -1;
1180 		offset += de_len;
1181 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1182 	}
1183 	return 0;
1184 }
1185 
1186 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1187 			       struct ext4_dir_entry *de)
1188 {
1189 	struct super_block *sb = dir->i_sb;
1190 
1191 	if (!is_dx(dir))
1192 		return 0;
1193 	if (block == 0)
1194 		return 1;
1195 	if (de->inode == 0 &&
1196 	    ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1197 			sb->s_blocksize)
1198 		return 1;
1199 	return 0;
1200 }
1201 
1202 /*
1203  *	ext4_find_entry()
1204  *
1205  * finds an entry in the specified directory with the wanted name. It
1206  * returns the cache buffer in which the entry was found, and the entry
1207  * itself (as a parameter - res_dir). It does NOT read the inode of the
1208  * entry - you'll have to do that yourself if you want to.
1209  *
1210  * The returned buffer_head has ->b_count elevated.  The caller is expected
1211  * to brelse() it when appropriate.
1212  */
1213 static struct buffer_head * ext4_find_entry (struct inode *dir,
1214 					const struct qstr *d_name,
1215 					struct ext4_dir_entry_2 **res_dir,
1216 					int *inlined)
1217 {
1218 	struct super_block *sb;
1219 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1220 	struct buffer_head *bh, *ret = NULL;
1221 	ext4_lblk_t start, block, b;
1222 	const u8 *name = d_name->name;
1223 	int ra_max = 0;		/* Number of bh's in the readahead
1224 				   buffer, bh_use[] */
1225 	int ra_ptr = 0;		/* Current index into readahead
1226 				   buffer */
1227 	int num = 0;
1228 	ext4_lblk_t  nblocks;
1229 	int i, err;
1230 	int namelen;
1231 
1232 	*res_dir = NULL;
1233 	sb = dir->i_sb;
1234 	namelen = d_name->len;
1235 	if (namelen > EXT4_NAME_LEN)
1236 		return NULL;
1237 
1238 	if (ext4_has_inline_data(dir)) {
1239 		int has_inline_data = 1;
1240 		ret = ext4_find_inline_entry(dir, d_name, res_dir,
1241 					     &has_inline_data);
1242 		if (has_inline_data) {
1243 			if (inlined)
1244 				*inlined = 1;
1245 			return ret;
1246 		}
1247 	}
1248 
1249 	if ((namelen <= 2) && (name[0] == '.') &&
1250 	    (name[1] == '.' || name[1] == '\0')) {
1251 		/*
1252 		 * "." or ".." will only be in the first block
1253 		 * NFS may look up ".."; "." should be handled by the VFS
1254 		 */
1255 		block = start = 0;
1256 		nblocks = 1;
1257 		goto restart;
1258 	}
1259 	if (is_dx(dir)) {
1260 		bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
1261 		/*
1262 		 * On success, or if the error was file not found,
1263 		 * return.  Otherwise, fall back to doing a search the
1264 		 * old fashioned way.
1265 		 */
1266 		if (bh || (err != ERR_BAD_DX_DIR))
1267 			return bh;
1268 		dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1269 			       "falling back\n"));
1270 	}
1271 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1272 	start = EXT4_I(dir)->i_dir_start_lookup;
1273 	if (start >= nblocks)
1274 		start = 0;
1275 	block = start;
1276 restart:
1277 	do {
1278 		/*
1279 		 * We deal with the read-ahead logic here.
1280 		 */
1281 		if (ra_ptr >= ra_max) {
1282 			/* Refill the readahead buffer */
1283 			ra_ptr = 0;
1284 			b = block;
1285 			for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1286 				/*
1287 				 * Terminate if we reach the end of the
1288 				 * directory and must wrap, or if our
1289 				 * search has finished at this block.
1290 				 */
1291 				if (b >= nblocks || (num && block == start)) {
1292 					bh_use[ra_max] = NULL;
1293 					break;
1294 				}
1295 				num++;
1296 				bh = ext4_getblk(NULL, dir, b++, 0, &err);
1297 				bh_use[ra_max] = bh;
1298 				if (bh)
1299 					ll_rw_block(READ | REQ_META | REQ_PRIO,
1300 						    1, &bh);
1301 			}
1302 		}
1303 		if ((bh = bh_use[ra_ptr++]) == NULL)
1304 			goto next;
1305 		wait_on_buffer(bh);
1306 		if (!buffer_uptodate(bh)) {
1307 			/* read error, skip block & hope for the best */
1308 			EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1309 					 (unsigned long) block);
1310 			brelse(bh);
1311 			goto next;
1312 		}
1313 		if (!buffer_verified(bh) &&
1314 		    !is_dx_internal_node(dir, block,
1315 					 (struct ext4_dir_entry *)bh->b_data) &&
1316 		    !ext4_dirent_csum_verify(dir,
1317 				(struct ext4_dir_entry *)bh->b_data)) {
1318 			EXT4_ERROR_INODE(dir, "checksumming directory "
1319 					 "block %lu", (unsigned long)block);
1320 			brelse(bh);
1321 			goto next;
1322 		}
1323 		set_buffer_verified(bh);
1324 		i = search_dirblock(bh, dir, d_name,
1325 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1326 		if (i == 1) {
1327 			EXT4_I(dir)->i_dir_start_lookup = block;
1328 			ret = bh;
1329 			goto cleanup_and_exit;
1330 		} else {
1331 			brelse(bh);
1332 			if (i < 0)
1333 				goto cleanup_and_exit;
1334 		}
1335 	next:
1336 		if (++block >= nblocks)
1337 			block = 0;
1338 	} while (block != start);
1339 
1340 	/*
1341 	 * If the directory has grown while we were searching, then
1342 	 * search the last part of the directory before giving up.
1343 	 */
1344 	block = nblocks;
1345 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1346 	if (block < nblocks) {
1347 		start = 0;
1348 		goto restart;
1349 	}
1350 
1351 cleanup_and_exit:
1352 	/* Clean up the read-ahead blocks */
1353 	for (; ra_ptr < ra_max; ra_ptr++)
1354 		brelse(bh_use[ra_ptr]);
1355 	return ret;
1356 }
1357 
1358 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1359 		       struct ext4_dir_entry_2 **res_dir, int *err)
1360 {
1361 	struct super_block * sb = dir->i_sb;
1362 	struct dx_hash_info	hinfo;
1363 	struct dx_frame frames[2], *frame;
1364 	struct buffer_head *bh;
1365 	ext4_lblk_t block;
1366 	int retval;
1367 
1368 	if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1369 		return NULL;
1370 	do {
1371 		block = dx_get_block(frame->at);
1372 		bh = ext4_read_dirblock(dir, block, DIRENT);
1373 		if (IS_ERR(bh)) {
1374 			*err = PTR_ERR(bh);
1375 			goto errout;
1376 		}
1377 		retval = search_dirblock(bh, dir, d_name,
1378 					 block << EXT4_BLOCK_SIZE_BITS(sb),
1379 					 res_dir);
1380 		if (retval == 1) { 	/* Success! */
1381 			dx_release(frames);
1382 			return bh;
1383 		}
1384 		brelse(bh);
1385 		if (retval == -1) {
1386 			*err = ERR_BAD_DX_DIR;
1387 			goto errout;
1388 		}
1389 
1390 		/* Check to see if we should continue to search */
1391 		retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1392 					       frames, NULL);
1393 		if (retval < 0) {
1394 			ext4_warning(sb,
1395 			     "error reading index page in directory #%lu",
1396 			     dir->i_ino);
1397 			*err = retval;
1398 			goto errout;
1399 		}
1400 	} while (retval == 1);
1401 
1402 	*err = -ENOENT;
1403 errout:
1404 	dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1405 	dx_release (frames);
1406 	return NULL;
1407 }
1408 
1409 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1410 {
1411 	struct inode *inode;
1412 	struct ext4_dir_entry_2 *de;
1413 	struct buffer_head *bh;
1414 
1415 	if (dentry->d_name.len > EXT4_NAME_LEN)
1416 		return ERR_PTR(-ENAMETOOLONG);
1417 
1418 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1419 	inode = NULL;
1420 	if (bh) {
1421 		__u32 ino = le32_to_cpu(de->inode);
1422 		brelse(bh);
1423 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1424 			EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1425 			return ERR_PTR(-EIO);
1426 		}
1427 		if (unlikely(ino == dir->i_ino)) {
1428 			EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1429 					 dentry);
1430 			return ERR_PTR(-EIO);
1431 		}
1432 		inode = ext4_iget(dir->i_sb, ino);
1433 		if (inode == ERR_PTR(-ESTALE)) {
1434 			EXT4_ERROR_INODE(dir,
1435 					 "deleted inode referenced: %u",
1436 					 ino);
1437 			return ERR_PTR(-EIO);
1438 		}
1439 	}
1440 	return d_splice_alias(inode, dentry);
1441 }
1442 
1443 
1444 struct dentry *ext4_get_parent(struct dentry *child)
1445 {
1446 	__u32 ino;
1447 	static const struct qstr dotdot = QSTR_INIT("..", 2);
1448 	struct ext4_dir_entry_2 * de;
1449 	struct buffer_head *bh;
1450 
1451 	bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1452 	if (!bh)
1453 		return ERR_PTR(-ENOENT);
1454 	ino = le32_to_cpu(de->inode);
1455 	brelse(bh);
1456 
1457 	if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1458 		EXT4_ERROR_INODE(child->d_inode,
1459 				 "bad parent inode number: %u", ino);
1460 		return ERR_PTR(-EIO);
1461 	}
1462 
1463 	return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1464 }
1465 
1466 /*
1467  * Move count entries from end of map between two memory locations.
1468  * Returns pointer to last entry moved.
1469  */
1470 static struct ext4_dir_entry_2 *
1471 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1472 		unsigned blocksize)
1473 {
1474 	unsigned rec_len = 0;
1475 
1476 	while (count--) {
1477 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1478 						(from + (map->offs<<2));
1479 		rec_len = EXT4_DIR_REC_LEN(de->name_len);
1480 		memcpy (to, de, rec_len);
1481 		((struct ext4_dir_entry_2 *) to)->rec_len =
1482 				ext4_rec_len_to_disk(rec_len, blocksize);
1483 		de->inode = 0;
1484 		map++;
1485 		to += rec_len;
1486 	}
1487 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1488 }
1489 
1490 /*
1491  * Compact each dir entry in the range to the minimal rec_len.
1492  * Returns pointer to last entry in range.
1493  */
1494 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1495 {
1496 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1497 	unsigned rec_len = 0;
1498 
1499 	prev = to = de;
1500 	while ((char*)de < base + blocksize) {
1501 		next = ext4_next_entry(de, blocksize);
1502 		if (de->inode && de->name_len) {
1503 			rec_len = EXT4_DIR_REC_LEN(de->name_len);
1504 			if (de > to)
1505 				memmove(to, de, rec_len);
1506 			to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1507 			prev = to;
1508 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1509 		}
1510 		de = next;
1511 	}
1512 	return prev;
1513 }
1514 
1515 /*
1516  * Split a full leaf block to make room for a new dir entry.
1517  * Allocate a new block, and move entries so that they are approx. equally full.
1518  * Returns pointer to de in block into which the new entry will be inserted.
1519  */
1520 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1521 			struct buffer_head **bh,struct dx_frame *frame,
1522 			struct dx_hash_info *hinfo, int *error)
1523 {
1524 	unsigned blocksize = dir->i_sb->s_blocksize;
1525 	unsigned count, continued;
1526 	struct buffer_head *bh2;
1527 	ext4_lblk_t newblock;
1528 	u32 hash2;
1529 	struct dx_map_entry *map;
1530 	char *data1 = (*bh)->b_data, *data2;
1531 	unsigned split, move, size;
1532 	struct ext4_dir_entry_2 *de = NULL, *de2;
1533 	struct ext4_dir_entry_tail *t;
1534 	int	csum_size = 0;
1535 	int	err = 0, i;
1536 
1537 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1538 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1539 		csum_size = sizeof(struct ext4_dir_entry_tail);
1540 
1541 	bh2 = ext4_append(handle, dir, &newblock);
1542 	if (IS_ERR(bh2)) {
1543 		brelse(*bh);
1544 		*bh = NULL;
1545 		*error = PTR_ERR(bh2);
1546 		return NULL;
1547 	}
1548 
1549 	BUFFER_TRACE(*bh, "get_write_access");
1550 	err = ext4_journal_get_write_access(handle, *bh);
1551 	if (err)
1552 		goto journal_error;
1553 
1554 	BUFFER_TRACE(frame->bh, "get_write_access");
1555 	err = ext4_journal_get_write_access(handle, frame->bh);
1556 	if (err)
1557 		goto journal_error;
1558 
1559 	data2 = bh2->b_data;
1560 
1561 	/* create map in the end of data2 block */
1562 	map = (struct dx_map_entry *) (data2 + blocksize);
1563 	count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1564 			     blocksize, hinfo, map);
1565 	map -= count;
1566 	dx_sort_map(map, count);
1567 	/* Split the existing block in the middle, size-wise */
1568 	size = 0;
1569 	move = 0;
1570 	for (i = count-1; i >= 0; i--) {
1571 		/* is more than half of this entry in 2nd half of the block? */
1572 		if (size + map[i].size/2 > blocksize/2)
1573 			break;
1574 		size += map[i].size;
1575 		move++;
1576 	}
1577 	/* map index at which we will split */
1578 	split = count - move;
1579 	hash2 = map[split].hash;
1580 	continued = hash2 == map[split - 1].hash;
1581 	dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1582 			(unsigned long)dx_get_block(frame->at),
1583 					hash2, split, count-split));
1584 
1585 	/* Fancy dance to stay within two buffers */
1586 	de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1587 	de = dx_pack_dirents(data1, blocksize);
1588 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1589 					   (char *) de,
1590 					   blocksize);
1591 	de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1592 					    (char *) de2,
1593 					    blocksize);
1594 	if (csum_size) {
1595 		t = EXT4_DIRENT_TAIL(data2, blocksize);
1596 		initialize_dirent_tail(t, blocksize);
1597 
1598 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1599 		initialize_dirent_tail(t, blocksize);
1600 	}
1601 
1602 	dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1603 	dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1604 
1605 	/* Which block gets the new entry? */
1606 	if (hinfo->hash >= hash2)
1607 	{
1608 		swap(*bh, bh2);
1609 		de = de2;
1610 	}
1611 	dx_insert_block(frame, hash2 + continued, newblock);
1612 	err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1613 	if (err)
1614 		goto journal_error;
1615 	err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1616 	if (err)
1617 		goto journal_error;
1618 	brelse(bh2);
1619 	dxtrace(dx_show_index("frame", frame->entries));
1620 	return de;
1621 
1622 journal_error:
1623 	brelse(*bh);
1624 	brelse(bh2);
1625 	*bh = NULL;
1626 	ext4_std_error(dir->i_sb, err);
1627 	*error = err;
1628 	return NULL;
1629 }
1630 
1631 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1632 		      struct buffer_head *bh,
1633 		      void *buf, int buf_size,
1634 		      const char *name, int namelen,
1635 		      struct ext4_dir_entry_2 **dest_de)
1636 {
1637 	struct ext4_dir_entry_2 *de;
1638 	unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1639 	int nlen, rlen;
1640 	unsigned int offset = 0;
1641 	char *top;
1642 
1643 	de = (struct ext4_dir_entry_2 *)buf;
1644 	top = buf + buf_size - reclen;
1645 	while ((char *) de <= top) {
1646 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1647 					 buf, buf_size, offset))
1648 			return -EIO;
1649 		if (ext4_match(namelen, name, de))
1650 			return -EEXIST;
1651 		nlen = EXT4_DIR_REC_LEN(de->name_len);
1652 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1653 		if ((de->inode ? rlen - nlen : rlen) >= reclen)
1654 			break;
1655 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1656 		offset += rlen;
1657 	}
1658 	if ((char *) de > top)
1659 		return -ENOSPC;
1660 
1661 	*dest_de = de;
1662 	return 0;
1663 }
1664 
1665 void ext4_insert_dentry(struct inode *inode,
1666 			struct ext4_dir_entry_2 *de,
1667 			int buf_size,
1668 			const char *name, int namelen)
1669 {
1670 
1671 	int nlen, rlen;
1672 
1673 	nlen = EXT4_DIR_REC_LEN(de->name_len);
1674 	rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1675 	if (de->inode) {
1676 		struct ext4_dir_entry_2 *de1 =
1677 				(struct ext4_dir_entry_2 *)((char *)de + nlen);
1678 		de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1679 		de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1680 		de = de1;
1681 	}
1682 	de->file_type = EXT4_FT_UNKNOWN;
1683 	de->inode = cpu_to_le32(inode->i_ino);
1684 	ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1685 	de->name_len = namelen;
1686 	memcpy(de->name, name, namelen);
1687 }
1688 /*
1689  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1690  * it points to a directory entry which is guaranteed to be large
1691  * enough for new directory entry.  If de is NULL, then
1692  * add_dirent_to_buf will attempt search the directory block for
1693  * space.  It will return -ENOSPC if no space is available, and -EIO
1694  * and -EEXIST if directory entry already exists.
1695  */
1696 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1697 			     struct inode *inode, struct ext4_dir_entry_2 *de,
1698 			     struct buffer_head *bh)
1699 {
1700 	struct inode	*dir = dentry->d_parent->d_inode;
1701 	const char	*name = dentry->d_name.name;
1702 	int		namelen = dentry->d_name.len;
1703 	unsigned int	blocksize = dir->i_sb->s_blocksize;
1704 	int		csum_size = 0;
1705 	int		err;
1706 
1707 	if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1708 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1709 		csum_size = sizeof(struct ext4_dir_entry_tail);
1710 
1711 	if (!de) {
1712 		err = ext4_find_dest_de(dir, inode,
1713 					bh, bh->b_data, blocksize - csum_size,
1714 					name, namelen, &de);
1715 		if (err)
1716 			return err;
1717 	}
1718 	BUFFER_TRACE(bh, "get_write_access");
1719 	err = ext4_journal_get_write_access(handle, bh);
1720 	if (err) {
1721 		ext4_std_error(dir->i_sb, err);
1722 		return err;
1723 	}
1724 
1725 	/* By now the buffer is marked for journaling */
1726 	ext4_insert_dentry(inode, de, blocksize, name, namelen);
1727 
1728 	/*
1729 	 * XXX shouldn't update any times until successful
1730 	 * completion of syscall, but too many callers depend
1731 	 * on this.
1732 	 *
1733 	 * XXX similarly, too many callers depend on
1734 	 * ext4_new_inode() setting the times, but error
1735 	 * recovery deletes the inode, so the worst that can
1736 	 * happen is that the times are slightly out of date
1737 	 * and/or different from the directory change time.
1738 	 */
1739 	dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1740 	ext4_update_dx_flag(dir);
1741 	dir->i_version++;
1742 	ext4_mark_inode_dirty(handle, dir);
1743 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1744 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1745 	if (err)
1746 		ext4_std_error(dir->i_sb, err);
1747 	return 0;
1748 }
1749 
1750 /*
1751  * This converts a one block unindexed directory to a 3 block indexed
1752  * directory, and adds the dentry to the indexed directory.
1753  */
1754 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1755 			    struct inode *inode, struct buffer_head *bh)
1756 {
1757 	struct inode	*dir = dentry->d_parent->d_inode;
1758 	const char	*name = dentry->d_name.name;
1759 	int		namelen = dentry->d_name.len;
1760 	struct buffer_head *bh2;
1761 	struct dx_root	*root;
1762 	struct dx_frame	frames[2], *frame;
1763 	struct dx_entry *entries;
1764 	struct ext4_dir_entry_2	*de, *de2;
1765 	struct ext4_dir_entry_tail *t;
1766 	char		*data1, *top;
1767 	unsigned	len;
1768 	int		retval;
1769 	unsigned	blocksize;
1770 	struct dx_hash_info hinfo;
1771 	ext4_lblk_t  block;
1772 	struct fake_dirent *fde;
1773 	int		csum_size = 0;
1774 
1775 	if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1776 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1777 		csum_size = sizeof(struct ext4_dir_entry_tail);
1778 
1779 	blocksize =  dir->i_sb->s_blocksize;
1780 	dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1781 	retval = ext4_journal_get_write_access(handle, bh);
1782 	if (retval) {
1783 		ext4_std_error(dir->i_sb, retval);
1784 		brelse(bh);
1785 		return retval;
1786 	}
1787 	root = (struct dx_root *) bh->b_data;
1788 
1789 	/* The 0th block becomes the root, move the dirents out */
1790 	fde = &root->dotdot;
1791 	de = (struct ext4_dir_entry_2 *)((char *)fde +
1792 		ext4_rec_len_from_disk(fde->rec_len, blocksize));
1793 	if ((char *) de >= (((char *) root) + blocksize)) {
1794 		EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1795 		brelse(bh);
1796 		return -EIO;
1797 	}
1798 	len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1799 
1800 	/* Allocate new block for the 0th block's dirents */
1801 	bh2 = ext4_append(handle, dir, &block);
1802 	if (IS_ERR(bh2)) {
1803 		brelse(bh);
1804 		return PTR_ERR(bh2);
1805 	}
1806 	ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1807 	data1 = bh2->b_data;
1808 
1809 	memcpy (data1, de, len);
1810 	de = (struct ext4_dir_entry_2 *) data1;
1811 	top = data1 + len;
1812 	while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1813 		de = de2;
1814 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1815 					   (char *) de,
1816 					   blocksize);
1817 
1818 	if (csum_size) {
1819 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1820 		initialize_dirent_tail(t, blocksize);
1821 	}
1822 
1823 	/* Initialize the root; the dot dirents already exist */
1824 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1825 	de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1826 					   blocksize);
1827 	memset (&root->info, 0, sizeof(root->info));
1828 	root->info.info_length = sizeof(root->info);
1829 	root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1830 	entries = root->entries;
1831 	dx_set_block(entries, 1);
1832 	dx_set_count(entries, 1);
1833 	dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1834 
1835 	/* Initialize as for dx_probe */
1836 	hinfo.hash_version = root->info.hash_version;
1837 	if (hinfo.hash_version <= DX_HASH_TEA)
1838 		hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1839 	hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1840 	ext4fs_dirhash(name, namelen, &hinfo);
1841 	frame = frames;
1842 	frame->entries = entries;
1843 	frame->at = entries;
1844 	frame->bh = bh;
1845 	bh = bh2;
1846 
1847 	ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1848 	ext4_handle_dirty_dirent_node(handle, dir, bh);
1849 
1850 	de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1851 	if (!de) {
1852 		/*
1853 		 * Even if the block split failed, we have to properly write
1854 		 * out all the changes we did so far. Otherwise we can end up
1855 		 * with corrupted filesystem.
1856 		 */
1857 		ext4_mark_inode_dirty(handle, dir);
1858 		dx_release(frames);
1859 		return retval;
1860 	}
1861 	dx_release(frames);
1862 
1863 	retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1864 	brelse(bh);
1865 	return retval;
1866 }
1867 
1868 /*
1869  *	ext4_add_entry()
1870  *
1871  * adds a file entry to the specified directory, using the same
1872  * semantics as ext4_find_entry(). It returns NULL if it failed.
1873  *
1874  * NOTE!! The inode part of 'de' is left at 0 - which means you
1875  * may not sleep between calling this and putting something into
1876  * the entry, as someone else might have used it while you slept.
1877  */
1878 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1879 			  struct inode *inode)
1880 {
1881 	struct inode *dir = dentry->d_parent->d_inode;
1882 	struct buffer_head *bh;
1883 	struct ext4_dir_entry_2 *de;
1884 	struct ext4_dir_entry_tail *t;
1885 	struct super_block *sb;
1886 	int	retval;
1887 	int	dx_fallback=0;
1888 	unsigned blocksize;
1889 	ext4_lblk_t block, blocks;
1890 	int	csum_size = 0;
1891 
1892 	if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1893 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1894 		csum_size = sizeof(struct ext4_dir_entry_tail);
1895 
1896 	sb = dir->i_sb;
1897 	blocksize = sb->s_blocksize;
1898 	if (!dentry->d_name.len)
1899 		return -EINVAL;
1900 
1901 	if (ext4_has_inline_data(dir)) {
1902 		retval = ext4_try_add_inline_entry(handle, dentry, inode);
1903 		if (retval < 0)
1904 			return retval;
1905 		if (retval == 1) {
1906 			retval = 0;
1907 			return retval;
1908 		}
1909 	}
1910 
1911 	if (is_dx(dir)) {
1912 		retval = ext4_dx_add_entry(handle, dentry, inode);
1913 		if (!retval || (retval != ERR_BAD_DX_DIR))
1914 			return retval;
1915 		ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1916 		dx_fallback++;
1917 		ext4_mark_inode_dirty(handle, dir);
1918 	}
1919 	blocks = dir->i_size >> sb->s_blocksize_bits;
1920 	for (block = 0; block < blocks; block++) {
1921 		bh = ext4_read_dirblock(dir, block, DIRENT);
1922 		if (IS_ERR(bh))
1923 			return PTR_ERR(bh);
1924 
1925 		retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1926 		if (retval != -ENOSPC) {
1927 			brelse(bh);
1928 			return retval;
1929 		}
1930 
1931 		if (blocks == 1 && !dx_fallback &&
1932 		    EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1933 			return make_indexed_dir(handle, dentry, inode, bh);
1934 		brelse(bh);
1935 	}
1936 	bh = ext4_append(handle, dir, &block);
1937 	if (IS_ERR(bh))
1938 		return PTR_ERR(bh);
1939 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1940 	de->inode = 0;
1941 	de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1942 
1943 	if (csum_size) {
1944 		t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1945 		initialize_dirent_tail(t, blocksize);
1946 	}
1947 
1948 	retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1949 	brelse(bh);
1950 	if (retval == 0)
1951 		ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1952 	return retval;
1953 }
1954 
1955 /*
1956  * Returns 0 for success, or a negative error value
1957  */
1958 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1959 			     struct inode *inode)
1960 {
1961 	struct dx_frame frames[2], *frame;
1962 	struct dx_entry *entries, *at;
1963 	struct dx_hash_info hinfo;
1964 	struct buffer_head *bh;
1965 	struct inode *dir = dentry->d_parent->d_inode;
1966 	struct super_block *sb = dir->i_sb;
1967 	struct ext4_dir_entry_2 *de;
1968 	int err;
1969 
1970 	frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1971 	if (!frame)
1972 		return err;
1973 	entries = frame->entries;
1974 	at = frame->at;
1975 	bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
1976 	if (IS_ERR(bh)) {
1977 		err = PTR_ERR(bh);
1978 		bh = NULL;
1979 		goto cleanup;
1980 	}
1981 
1982 	BUFFER_TRACE(bh, "get_write_access");
1983 	err = ext4_journal_get_write_access(handle, bh);
1984 	if (err)
1985 		goto journal_error;
1986 
1987 	err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1988 	if (err != -ENOSPC)
1989 		goto cleanup;
1990 
1991 	/* Block full, should compress but for now just split */
1992 	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1993 		       dx_get_count(entries), dx_get_limit(entries)));
1994 	/* Need to split index? */
1995 	if (dx_get_count(entries) == dx_get_limit(entries)) {
1996 		ext4_lblk_t newblock;
1997 		unsigned icount = dx_get_count(entries);
1998 		int levels = frame - frames;
1999 		struct dx_entry *entries2;
2000 		struct dx_node *node2;
2001 		struct buffer_head *bh2;
2002 
2003 		if (levels && (dx_get_count(frames->entries) ==
2004 			       dx_get_limit(frames->entries))) {
2005 			ext4_warning(sb, "Directory index full!");
2006 			err = -ENOSPC;
2007 			goto cleanup;
2008 		}
2009 		bh2 = ext4_append(handle, dir, &newblock);
2010 		if (IS_ERR(bh2)) {
2011 			err = PTR_ERR(bh2);
2012 			goto cleanup;
2013 		}
2014 		node2 = (struct dx_node *)(bh2->b_data);
2015 		entries2 = node2->entries;
2016 		memset(&node2->fake, 0, sizeof(struct fake_dirent));
2017 		node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2018 							   sb->s_blocksize);
2019 		BUFFER_TRACE(frame->bh, "get_write_access");
2020 		err = ext4_journal_get_write_access(handle, frame->bh);
2021 		if (err)
2022 			goto journal_error;
2023 		if (levels) {
2024 			unsigned icount1 = icount/2, icount2 = icount - icount1;
2025 			unsigned hash2 = dx_get_hash(entries + icount1);
2026 			dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2027 				       icount1, icount2));
2028 
2029 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2030 			err = ext4_journal_get_write_access(handle,
2031 							     frames[0].bh);
2032 			if (err)
2033 				goto journal_error;
2034 
2035 			memcpy((char *) entries2, (char *) (entries + icount1),
2036 			       icount2 * sizeof(struct dx_entry));
2037 			dx_set_count(entries, icount1);
2038 			dx_set_count(entries2, icount2);
2039 			dx_set_limit(entries2, dx_node_limit(dir));
2040 
2041 			/* Which index block gets the new entry? */
2042 			if (at - entries >= icount1) {
2043 				frame->at = at = at - entries - icount1 + entries2;
2044 				frame->entries = entries = entries2;
2045 				swap(frame->bh, bh2);
2046 			}
2047 			dx_insert_block(frames + 0, hash2, newblock);
2048 			dxtrace(dx_show_index("node", frames[1].entries));
2049 			dxtrace(dx_show_index("node",
2050 			       ((struct dx_node *) bh2->b_data)->entries));
2051 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2052 			if (err)
2053 				goto journal_error;
2054 			brelse (bh2);
2055 		} else {
2056 			dxtrace(printk(KERN_DEBUG
2057 				       "Creating second level index...\n"));
2058 			memcpy((char *) entries2, (char *) entries,
2059 			       icount * sizeof(struct dx_entry));
2060 			dx_set_limit(entries2, dx_node_limit(dir));
2061 
2062 			/* Set up root */
2063 			dx_set_count(entries, 1);
2064 			dx_set_block(entries + 0, newblock);
2065 			((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2066 
2067 			/* Add new access path frame */
2068 			frame = frames + 1;
2069 			frame->at = at = at - entries + entries2;
2070 			frame->entries = entries = entries2;
2071 			frame->bh = bh2;
2072 			err = ext4_journal_get_write_access(handle,
2073 							     frame->bh);
2074 			if (err)
2075 				goto journal_error;
2076 		}
2077 		err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2078 		if (err) {
2079 			ext4_std_error(inode->i_sb, err);
2080 			goto cleanup;
2081 		}
2082 	}
2083 	de = do_split(handle, dir, &bh, frame, &hinfo, &err);
2084 	if (!de)
2085 		goto cleanup;
2086 	err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2087 	goto cleanup;
2088 
2089 journal_error:
2090 	ext4_std_error(dir->i_sb, err);
2091 cleanup:
2092 	brelse(bh);
2093 	dx_release(frames);
2094 	return err;
2095 }
2096 
2097 /*
2098  * ext4_generic_delete_entry deletes a directory entry by merging it
2099  * with the previous entry
2100  */
2101 int ext4_generic_delete_entry(handle_t *handle,
2102 			      struct inode *dir,
2103 			      struct ext4_dir_entry_2 *de_del,
2104 			      struct buffer_head *bh,
2105 			      void *entry_buf,
2106 			      int buf_size,
2107 			      int csum_size)
2108 {
2109 	struct ext4_dir_entry_2 *de, *pde;
2110 	unsigned int blocksize = dir->i_sb->s_blocksize;
2111 	int i;
2112 
2113 	i = 0;
2114 	pde = NULL;
2115 	de = (struct ext4_dir_entry_2 *)entry_buf;
2116 	while (i < buf_size - csum_size) {
2117 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2118 					 bh->b_data, bh->b_size, i))
2119 			return -EIO;
2120 		if (de == de_del)  {
2121 			if (pde)
2122 				pde->rec_len = ext4_rec_len_to_disk(
2123 					ext4_rec_len_from_disk(pde->rec_len,
2124 							       blocksize) +
2125 					ext4_rec_len_from_disk(de->rec_len,
2126 							       blocksize),
2127 					blocksize);
2128 			else
2129 				de->inode = 0;
2130 			dir->i_version++;
2131 			return 0;
2132 		}
2133 		i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2134 		pde = de;
2135 		de = ext4_next_entry(de, blocksize);
2136 	}
2137 	return -ENOENT;
2138 }
2139 
2140 static int ext4_delete_entry(handle_t *handle,
2141 			     struct inode *dir,
2142 			     struct ext4_dir_entry_2 *de_del,
2143 			     struct buffer_head *bh)
2144 {
2145 	int err, csum_size = 0;
2146 
2147 	if (ext4_has_inline_data(dir)) {
2148 		int has_inline_data = 1;
2149 		err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2150 					       &has_inline_data);
2151 		if (has_inline_data)
2152 			return err;
2153 	}
2154 
2155 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2156 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2157 		csum_size = sizeof(struct ext4_dir_entry_tail);
2158 
2159 	BUFFER_TRACE(bh, "get_write_access");
2160 	err = ext4_journal_get_write_access(handle, bh);
2161 	if (unlikely(err))
2162 		goto out;
2163 
2164 	err = ext4_generic_delete_entry(handle, dir, de_del,
2165 					bh, bh->b_data,
2166 					dir->i_sb->s_blocksize, csum_size);
2167 	if (err)
2168 		goto out;
2169 
2170 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2171 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2172 	if (unlikely(err))
2173 		goto out;
2174 
2175 	return 0;
2176 out:
2177 	if (err != -ENOENT)
2178 		ext4_std_error(dir->i_sb, err);
2179 	return err;
2180 }
2181 
2182 /*
2183  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2184  * since this indicates that nlinks count was previously 1.
2185  */
2186 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2187 {
2188 	inc_nlink(inode);
2189 	if (is_dx(inode) && inode->i_nlink > 1) {
2190 		/* limit is 16-bit i_links_count */
2191 		if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2192 			set_nlink(inode, 1);
2193 			EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2194 					      EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2195 		}
2196 	}
2197 }
2198 
2199 /*
2200  * If a directory had nlink == 1, then we should let it be 1. This indicates
2201  * directory has >EXT4_LINK_MAX subdirs.
2202  */
2203 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2204 {
2205 	if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2206 		drop_nlink(inode);
2207 }
2208 
2209 
2210 static int ext4_add_nondir(handle_t *handle,
2211 		struct dentry *dentry, struct inode *inode)
2212 {
2213 	int err = ext4_add_entry(handle, dentry, inode);
2214 	if (!err) {
2215 		ext4_mark_inode_dirty(handle, inode);
2216 		unlock_new_inode(inode);
2217 		d_instantiate(dentry, inode);
2218 		return 0;
2219 	}
2220 	drop_nlink(inode);
2221 	unlock_new_inode(inode);
2222 	iput(inode);
2223 	return err;
2224 }
2225 
2226 /*
2227  * By the time this is called, we already have created
2228  * the directory cache entry for the new file, but it
2229  * is so far negative - it has no inode.
2230  *
2231  * If the create succeeds, we fill in the inode information
2232  * with d_instantiate().
2233  */
2234 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2235 		       bool excl)
2236 {
2237 	handle_t *handle;
2238 	struct inode *inode;
2239 	int err, credits, retries = 0;
2240 
2241 	dquot_initialize(dir);
2242 
2243 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2244 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2245 retry:
2246 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2247 					    NULL, EXT4_HT_DIR, credits);
2248 	handle = ext4_journal_current_handle();
2249 	err = PTR_ERR(inode);
2250 	if (!IS_ERR(inode)) {
2251 		inode->i_op = &ext4_file_inode_operations;
2252 		inode->i_fop = &ext4_file_operations;
2253 		ext4_set_aops(inode);
2254 		err = ext4_add_nondir(handle, dentry, inode);
2255 		if (!err && IS_DIRSYNC(dir))
2256 			ext4_handle_sync(handle);
2257 	}
2258 	if (handle)
2259 		ext4_journal_stop(handle);
2260 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2261 		goto retry;
2262 	return err;
2263 }
2264 
2265 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2266 		      umode_t mode, dev_t rdev)
2267 {
2268 	handle_t *handle;
2269 	struct inode *inode;
2270 	int err, credits, retries = 0;
2271 
2272 	if (!new_valid_dev(rdev))
2273 		return -EINVAL;
2274 
2275 	dquot_initialize(dir);
2276 
2277 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2278 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2279 retry:
2280 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2281 					    NULL, EXT4_HT_DIR, credits);
2282 	handle = ext4_journal_current_handle();
2283 	err = PTR_ERR(inode);
2284 	if (!IS_ERR(inode)) {
2285 		init_special_inode(inode, inode->i_mode, rdev);
2286 		inode->i_op = &ext4_special_inode_operations;
2287 		err = ext4_add_nondir(handle, dentry, inode);
2288 		if (!err && IS_DIRSYNC(dir))
2289 			ext4_handle_sync(handle);
2290 	}
2291 	if (handle)
2292 		ext4_journal_stop(handle);
2293 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2294 		goto retry;
2295 	return err;
2296 }
2297 
2298 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2299 {
2300 	handle_t *handle;
2301 	struct inode *inode;
2302 	int err, retries = 0;
2303 
2304 	dquot_initialize(dir);
2305 
2306 retry:
2307 	inode = ext4_new_inode_start_handle(dir, mode,
2308 					    NULL, 0, NULL,
2309 					    EXT4_HT_DIR,
2310 			EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2311 			  4 + EXT4_XATTR_TRANS_BLOCKS);
2312 	handle = ext4_journal_current_handle();
2313 	err = PTR_ERR(inode);
2314 	if (!IS_ERR(inode)) {
2315 		inode->i_op = &ext4_file_inode_operations;
2316 		inode->i_fop = &ext4_file_operations;
2317 		ext4_set_aops(inode);
2318 		d_tmpfile(dentry, inode);
2319 		err = ext4_orphan_add(handle, inode);
2320 		if (err)
2321 			goto err_unlock_inode;
2322 		mark_inode_dirty(inode);
2323 		unlock_new_inode(inode);
2324 	}
2325 	if (handle)
2326 		ext4_journal_stop(handle);
2327 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2328 		goto retry;
2329 	return err;
2330 err_unlock_inode:
2331 	ext4_journal_stop(handle);
2332 	unlock_new_inode(inode);
2333 	return err;
2334 }
2335 
2336 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2337 			  struct ext4_dir_entry_2 *de,
2338 			  int blocksize, int csum_size,
2339 			  unsigned int parent_ino, int dotdot_real_len)
2340 {
2341 	de->inode = cpu_to_le32(inode->i_ino);
2342 	de->name_len = 1;
2343 	de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2344 					   blocksize);
2345 	strcpy(de->name, ".");
2346 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2347 
2348 	de = ext4_next_entry(de, blocksize);
2349 	de->inode = cpu_to_le32(parent_ino);
2350 	de->name_len = 2;
2351 	if (!dotdot_real_len)
2352 		de->rec_len = ext4_rec_len_to_disk(blocksize -
2353 					(csum_size + EXT4_DIR_REC_LEN(1)),
2354 					blocksize);
2355 	else
2356 		de->rec_len = ext4_rec_len_to_disk(
2357 				EXT4_DIR_REC_LEN(de->name_len), blocksize);
2358 	strcpy(de->name, "..");
2359 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2360 
2361 	return ext4_next_entry(de, blocksize);
2362 }
2363 
2364 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2365 			     struct inode *inode)
2366 {
2367 	struct buffer_head *dir_block = NULL;
2368 	struct ext4_dir_entry_2 *de;
2369 	struct ext4_dir_entry_tail *t;
2370 	ext4_lblk_t block = 0;
2371 	unsigned int blocksize = dir->i_sb->s_blocksize;
2372 	int csum_size = 0;
2373 	int err;
2374 
2375 	if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2376 				       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2377 		csum_size = sizeof(struct ext4_dir_entry_tail);
2378 
2379 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2380 		err = ext4_try_create_inline_dir(handle, dir, inode);
2381 		if (err < 0 && err != -ENOSPC)
2382 			goto out;
2383 		if (!err)
2384 			goto out;
2385 	}
2386 
2387 	inode->i_size = 0;
2388 	dir_block = ext4_append(handle, inode, &block);
2389 	if (IS_ERR(dir_block))
2390 		return PTR_ERR(dir_block);
2391 	BUFFER_TRACE(dir_block, "get_write_access");
2392 	err = ext4_journal_get_write_access(handle, dir_block);
2393 	if (err)
2394 		goto out;
2395 	de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2396 	ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2397 	set_nlink(inode, 2);
2398 	if (csum_size) {
2399 		t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2400 		initialize_dirent_tail(t, blocksize);
2401 	}
2402 
2403 	BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2404 	err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2405 	if (err)
2406 		goto out;
2407 	set_buffer_verified(dir_block);
2408 out:
2409 	brelse(dir_block);
2410 	return err;
2411 }
2412 
2413 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2414 {
2415 	handle_t *handle;
2416 	struct inode *inode;
2417 	int err, credits, retries = 0;
2418 
2419 	if (EXT4_DIR_LINK_MAX(dir))
2420 		return -EMLINK;
2421 
2422 	dquot_initialize(dir);
2423 
2424 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2425 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2426 retry:
2427 	inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2428 					    &dentry->d_name,
2429 					    0, NULL, EXT4_HT_DIR, credits);
2430 	handle = ext4_journal_current_handle();
2431 	err = PTR_ERR(inode);
2432 	if (IS_ERR(inode))
2433 		goto out_stop;
2434 
2435 	inode->i_op = &ext4_dir_inode_operations;
2436 	inode->i_fop = &ext4_dir_operations;
2437 	err = ext4_init_new_dir(handle, dir, inode);
2438 	if (err)
2439 		goto out_clear_inode;
2440 	err = ext4_mark_inode_dirty(handle, inode);
2441 	if (!err)
2442 		err = ext4_add_entry(handle, dentry, inode);
2443 	if (err) {
2444 out_clear_inode:
2445 		clear_nlink(inode);
2446 		unlock_new_inode(inode);
2447 		ext4_mark_inode_dirty(handle, inode);
2448 		iput(inode);
2449 		goto out_stop;
2450 	}
2451 	ext4_inc_count(handle, dir);
2452 	ext4_update_dx_flag(dir);
2453 	err = ext4_mark_inode_dirty(handle, dir);
2454 	if (err)
2455 		goto out_clear_inode;
2456 	unlock_new_inode(inode);
2457 	d_instantiate(dentry, inode);
2458 	if (IS_DIRSYNC(dir))
2459 		ext4_handle_sync(handle);
2460 
2461 out_stop:
2462 	if (handle)
2463 		ext4_journal_stop(handle);
2464 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2465 		goto retry;
2466 	return err;
2467 }
2468 
2469 /*
2470  * routine to check that the specified directory is empty (for rmdir)
2471  */
2472 static int empty_dir(struct inode *inode)
2473 {
2474 	unsigned int offset;
2475 	struct buffer_head *bh;
2476 	struct ext4_dir_entry_2 *de, *de1;
2477 	struct super_block *sb;
2478 	int err = 0;
2479 
2480 	if (ext4_has_inline_data(inode)) {
2481 		int has_inline_data = 1;
2482 
2483 		err = empty_inline_dir(inode, &has_inline_data);
2484 		if (has_inline_data)
2485 			return err;
2486 	}
2487 
2488 	sb = inode->i_sb;
2489 	if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2490 		EXT4_ERROR_INODE(inode, "invalid size");
2491 		return 1;
2492 	}
2493 	bh = ext4_read_dirblock(inode, 0, EITHER);
2494 	if (IS_ERR(bh))
2495 		return 1;
2496 
2497 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2498 	de1 = ext4_next_entry(de, sb->s_blocksize);
2499 	if (le32_to_cpu(de->inode) != inode->i_ino ||
2500 			!le32_to_cpu(de1->inode) ||
2501 			strcmp(".", de->name) ||
2502 			strcmp("..", de1->name)) {
2503 		ext4_warning(inode->i_sb,
2504 			     "bad directory (dir #%lu) - no `.' or `..'",
2505 			     inode->i_ino);
2506 		brelse(bh);
2507 		return 1;
2508 	}
2509 	offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2510 		 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2511 	de = ext4_next_entry(de1, sb->s_blocksize);
2512 	while (offset < inode->i_size) {
2513 		if (!bh ||
2514 		    (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2515 			unsigned int lblock;
2516 			err = 0;
2517 			brelse(bh);
2518 			lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2519 			bh = ext4_read_dirblock(inode, lblock, EITHER);
2520 			if (IS_ERR(bh))
2521 				return 1;
2522 			de = (struct ext4_dir_entry_2 *) bh->b_data;
2523 		}
2524 		if (ext4_check_dir_entry(inode, NULL, de, bh,
2525 					 bh->b_data, bh->b_size, offset)) {
2526 			de = (struct ext4_dir_entry_2 *)(bh->b_data +
2527 							 sb->s_blocksize);
2528 			offset = (offset | (sb->s_blocksize - 1)) + 1;
2529 			continue;
2530 		}
2531 		if (le32_to_cpu(de->inode)) {
2532 			brelse(bh);
2533 			return 0;
2534 		}
2535 		offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2536 		de = ext4_next_entry(de, sb->s_blocksize);
2537 	}
2538 	brelse(bh);
2539 	return 1;
2540 }
2541 
2542 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
2543  * such inodes, starting at the superblock, in case we crash before the
2544  * file is closed/deleted, or in case the inode truncate spans multiple
2545  * transactions and the last transaction is not recovered after a crash.
2546  *
2547  * At filesystem recovery time, we walk this list deleting unlinked
2548  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2549  */
2550 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2551 {
2552 	struct super_block *sb = inode->i_sb;
2553 	struct ext4_iloc iloc;
2554 	int err = 0, rc;
2555 
2556 	if (!EXT4_SB(sb)->s_journal)
2557 		return 0;
2558 
2559 	mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
2560 	if (!list_empty(&EXT4_I(inode)->i_orphan))
2561 		goto out_unlock;
2562 
2563 	/*
2564 	 * Orphan handling is only valid for files with data blocks
2565 	 * being truncated, or files being unlinked. Note that we either
2566 	 * hold i_mutex, or the inode can not be referenced from outside,
2567 	 * so i_nlink should not be bumped due to race
2568 	 */
2569 	J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2570 		  S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2571 
2572 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2573 	err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2574 	if (err)
2575 		goto out_unlock;
2576 
2577 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2578 	if (err)
2579 		goto out_unlock;
2580 	/*
2581 	 * Due to previous errors inode may be already a part of on-disk
2582 	 * orphan list. If so skip on-disk list modification.
2583 	 */
2584 	if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2585 		(le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2586 			goto mem_insert;
2587 
2588 	/* Insert this inode at the head of the on-disk orphan list... */
2589 	NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2590 	EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2591 	err = ext4_handle_dirty_super(handle, sb);
2592 	rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2593 	if (!err)
2594 		err = rc;
2595 
2596 	/* Only add to the head of the in-memory list if all the
2597 	 * previous operations succeeded.  If the orphan_add is going to
2598 	 * fail (possibly taking the journal offline), we can't risk
2599 	 * leaving the inode on the orphan list: stray orphan-list
2600 	 * entries can cause panics at unmount time.
2601 	 *
2602 	 * This is safe: on error we're going to ignore the orphan list
2603 	 * anyway on the next recovery. */
2604 mem_insert:
2605 	if (!err)
2606 		list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2607 
2608 	jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2609 	jbd_debug(4, "orphan inode %lu will point to %d\n",
2610 			inode->i_ino, NEXT_ORPHAN(inode));
2611 out_unlock:
2612 	mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2613 	ext4_std_error(inode->i_sb, err);
2614 	return err;
2615 }
2616 
2617 /*
2618  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2619  * of such inodes stored on disk, because it is finally being cleaned up.
2620  */
2621 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2622 {
2623 	struct list_head *prev;
2624 	struct ext4_inode_info *ei = EXT4_I(inode);
2625 	struct ext4_sb_info *sbi;
2626 	__u32 ino_next;
2627 	struct ext4_iloc iloc;
2628 	int err = 0;
2629 
2630 	if ((!EXT4_SB(inode->i_sb)->s_journal) &&
2631 	    !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS))
2632 		return 0;
2633 
2634 	mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2635 	if (list_empty(&ei->i_orphan))
2636 		goto out;
2637 
2638 	ino_next = NEXT_ORPHAN(inode);
2639 	prev = ei->i_orphan.prev;
2640 	sbi = EXT4_SB(inode->i_sb);
2641 
2642 	jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2643 
2644 	list_del_init(&ei->i_orphan);
2645 
2646 	/* If we're on an error path, we may not have a valid
2647 	 * transaction handle with which to update the orphan list on
2648 	 * disk, but we still need to remove the inode from the linked
2649 	 * list in memory. */
2650 	if (!handle)
2651 		goto out;
2652 
2653 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2654 	if (err)
2655 		goto out_err;
2656 
2657 	if (prev == &sbi->s_orphan) {
2658 		jbd_debug(4, "superblock will point to %u\n", ino_next);
2659 		BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2660 		err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2661 		if (err)
2662 			goto out_brelse;
2663 		sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2664 		err = ext4_handle_dirty_super(handle, inode->i_sb);
2665 	} else {
2666 		struct ext4_iloc iloc2;
2667 		struct inode *i_prev =
2668 			&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2669 
2670 		jbd_debug(4, "orphan inode %lu will point to %u\n",
2671 			  i_prev->i_ino, ino_next);
2672 		err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2673 		if (err)
2674 			goto out_brelse;
2675 		NEXT_ORPHAN(i_prev) = ino_next;
2676 		err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2677 	}
2678 	if (err)
2679 		goto out_brelse;
2680 	NEXT_ORPHAN(inode) = 0;
2681 	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2682 
2683 out_err:
2684 	ext4_std_error(inode->i_sb, err);
2685 out:
2686 	mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2687 	return err;
2688 
2689 out_brelse:
2690 	brelse(iloc.bh);
2691 	goto out_err;
2692 }
2693 
2694 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2695 {
2696 	int retval;
2697 	struct inode *inode;
2698 	struct buffer_head *bh;
2699 	struct ext4_dir_entry_2 *de;
2700 	handle_t *handle = NULL;
2701 
2702 	/* Initialize quotas before so that eventual writes go in
2703 	 * separate transaction */
2704 	dquot_initialize(dir);
2705 	dquot_initialize(dentry->d_inode);
2706 
2707 	retval = -ENOENT;
2708 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2709 	if (!bh)
2710 		goto end_rmdir;
2711 
2712 	inode = dentry->d_inode;
2713 
2714 	retval = -EIO;
2715 	if (le32_to_cpu(de->inode) != inode->i_ino)
2716 		goto end_rmdir;
2717 
2718 	retval = -ENOTEMPTY;
2719 	if (!empty_dir(inode))
2720 		goto end_rmdir;
2721 
2722 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
2723 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2724 	if (IS_ERR(handle)) {
2725 		retval = PTR_ERR(handle);
2726 		handle = NULL;
2727 		goto end_rmdir;
2728 	}
2729 
2730 	if (IS_DIRSYNC(dir))
2731 		ext4_handle_sync(handle);
2732 
2733 	retval = ext4_delete_entry(handle, dir, de, bh);
2734 	if (retval)
2735 		goto end_rmdir;
2736 	if (!EXT4_DIR_LINK_EMPTY(inode))
2737 		ext4_warning(inode->i_sb,
2738 			     "empty directory has too many links (%d)",
2739 			     inode->i_nlink);
2740 	inode->i_version++;
2741 	clear_nlink(inode);
2742 	/* There's no need to set i_disksize: the fact that i_nlink is
2743 	 * zero will ensure that the right thing happens during any
2744 	 * recovery. */
2745 	inode->i_size = 0;
2746 	ext4_orphan_add(handle, inode);
2747 	inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2748 	ext4_mark_inode_dirty(handle, inode);
2749 	ext4_dec_count(handle, dir);
2750 	ext4_update_dx_flag(dir);
2751 	ext4_mark_inode_dirty(handle, dir);
2752 
2753 end_rmdir:
2754 	brelse(bh);
2755 	if (handle)
2756 		ext4_journal_stop(handle);
2757 	return retval;
2758 }
2759 
2760 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2761 {
2762 	int retval;
2763 	struct inode *inode;
2764 	struct buffer_head *bh;
2765 	struct ext4_dir_entry_2 *de;
2766 	handle_t *handle = NULL;
2767 
2768 	trace_ext4_unlink_enter(dir, dentry);
2769 	/* Initialize quotas before so that eventual writes go
2770 	 * in separate transaction */
2771 	dquot_initialize(dir);
2772 	dquot_initialize(dentry->d_inode);
2773 
2774 	retval = -ENOENT;
2775 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2776 	if (!bh)
2777 		goto end_unlink;
2778 
2779 	inode = dentry->d_inode;
2780 
2781 	retval = -EIO;
2782 	if (le32_to_cpu(de->inode) != inode->i_ino)
2783 		goto end_unlink;
2784 
2785 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
2786 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2787 	if (IS_ERR(handle)) {
2788 		retval = PTR_ERR(handle);
2789 		handle = NULL;
2790 		goto end_unlink;
2791 	}
2792 
2793 	if (IS_DIRSYNC(dir))
2794 		ext4_handle_sync(handle);
2795 
2796 	if (!inode->i_nlink) {
2797 		ext4_warning(inode->i_sb,
2798 			     "Deleting nonexistent file (%lu), %d",
2799 			     inode->i_ino, inode->i_nlink);
2800 		set_nlink(inode, 1);
2801 	}
2802 	retval = ext4_delete_entry(handle, dir, de, bh);
2803 	if (retval)
2804 		goto end_unlink;
2805 	dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2806 	ext4_update_dx_flag(dir);
2807 	ext4_mark_inode_dirty(handle, dir);
2808 	drop_nlink(inode);
2809 	if (!inode->i_nlink)
2810 		ext4_orphan_add(handle, inode);
2811 	inode->i_ctime = ext4_current_time(inode);
2812 	ext4_mark_inode_dirty(handle, inode);
2813 	retval = 0;
2814 
2815 end_unlink:
2816 	brelse(bh);
2817 	if (handle)
2818 		ext4_journal_stop(handle);
2819 	trace_ext4_unlink_exit(dentry, retval);
2820 	return retval;
2821 }
2822 
2823 static int ext4_symlink(struct inode *dir,
2824 			struct dentry *dentry, const char *symname)
2825 {
2826 	handle_t *handle;
2827 	struct inode *inode;
2828 	int l, err, retries = 0;
2829 	int credits;
2830 
2831 	l = strlen(symname)+1;
2832 	if (l > dir->i_sb->s_blocksize)
2833 		return -ENAMETOOLONG;
2834 
2835 	dquot_initialize(dir);
2836 
2837 	if (l > EXT4_N_BLOCKS * 4) {
2838 		/*
2839 		 * For non-fast symlinks, we just allocate inode and put it on
2840 		 * orphan list in the first transaction => we need bitmap,
2841 		 * group descriptor, sb, inode block, quota blocks, and
2842 		 * possibly selinux xattr blocks.
2843 		 */
2844 		credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2845 			  EXT4_XATTR_TRANS_BLOCKS;
2846 	} else {
2847 		/*
2848 		 * Fast symlink. We have to add entry to directory
2849 		 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2850 		 * allocate new inode (bitmap, group descriptor, inode block,
2851 		 * quota blocks, sb is already counted in previous macros).
2852 		 */
2853 		credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2854 			  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
2855 	}
2856 retry:
2857 	inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
2858 					    &dentry->d_name, 0, NULL,
2859 					    EXT4_HT_DIR, credits);
2860 	handle = ext4_journal_current_handle();
2861 	err = PTR_ERR(inode);
2862 	if (IS_ERR(inode))
2863 		goto out_stop;
2864 
2865 	if (l > EXT4_N_BLOCKS * 4) {
2866 		inode->i_op = &ext4_symlink_inode_operations;
2867 		ext4_set_aops(inode);
2868 		/*
2869 		 * We cannot call page_symlink() with transaction started
2870 		 * because it calls into ext4_write_begin() which can wait
2871 		 * for transaction commit if we are running out of space
2872 		 * and thus we deadlock. So we have to stop transaction now
2873 		 * and restart it when symlink contents is written.
2874 		 *
2875 		 * To keep fs consistent in case of crash, we have to put inode
2876 		 * to orphan list in the mean time.
2877 		 */
2878 		drop_nlink(inode);
2879 		err = ext4_orphan_add(handle, inode);
2880 		ext4_journal_stop(handle);
2881 		if (err)
2882 			goto err_drop_inode;
2883 		err = __page_symlink(inode, symname, l, 1);
2884 		if (err)
2885 			goto err_drop_inode;
2886 		/*
2887 		 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2888 		 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2889 		 */
2890 		handle = ext4_journal_start(dir, EXT4_HT_DIR,
2891 				EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2892 				EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2893 		if (IS_ERR(handle)) {
2894 			err = PTR_ERR(handle);
2895 			goto err_drop_inode;
2896 		}
2897 		set_nlink(inode, 1);
2898 		err = ext4_orphan_del(handle, inode);
2899 		if (err) {
2900 			ext4_journal_stop(handle);
2901 			clear_nlink(inode);
2902 			goto err_drop_inode;
2903 		}
2904 	} else {
2905 		/* clear the extent format for fast symlink */
2906 		ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2907 		inode->i_op = &ext4_fast_symlink_inode_operations;
2908 		memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2909 		inode->i_size = l-1;
2910 	}
2911 	EXT4_I(inode)->i_disksize = inode->i_size;
2912 	err = ext4_add_nondir(handle, dentry, inode);
2913 	if (!err && IS_DIRSYNC(dir))
2914 		ext4_handle_sync(handle);
2915 
2916 out_stop:
2917 	if (handle)
2918 		ext4_journal_stop(handle);
2919 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2920 		goto retry;
2921 	return err;
2922 err_drop_inode:
2923 	unlock_new_inode(inode);
2924 	iput(inode);
2925 	return err;
2926 }
2927 
2928 static int ext4_link(struct dentry *old_dentry,
2929 		     struct inode *dir, struct dentry *dentry)
2930 {
2931 	handle_t *handle;
2932 	struct inode *inode = old_dentry->d_inode;
2933 	int err, retries = 0;
2934 
2935 	if (inode->i_nlink >= EXT4_LINK_MAX)
2936 		return -EMLINK;
2937 
2938 	dquot_initialize(dir);
2939 
2940 retry:
2941 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
2942 		(EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2943 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
2944 	if (IS_ERR(handle))
2945 		return PTR_ERR(handle);
2946 
2947 	if (IS_DIRSYNC(dir))
2948 		ext4_handle_sync(handle);
2949 
2950 	inode->i_ctime = ext4_current_time(inode);
2951 	ext4_inc_count(handle, inode);
2952 	ihold(inode);
2953 
2954 	err = ext4_add_entry(handle, dentry, inode);
2955 	if (!err) {
2956 		ext4_mark_inode_dirty(handle, inode);
2957 		/* this can happen only for tmpfile being
2958 		 * linked the first time
2959 		 */
2960 		if (inode->i_nlink == 1)
2961 			ext4_orphan_del(handle, inode);
2962 		d_instantiate(dentry, inode);
2963 	} else {
2964 		drop_nlink(inode);
2965 		iput(inode);
2966 	}
2967 	ext4_journal_stop(handle);
2968 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2969 		goto retry;
2970 	return err;
2971 }
2972 
2973 
2974 /*
2975  * Try to find buffer head where contains the parent block.
2976  * It should be the inode block if it is inlined or the 1st block
2977  * if it is a normal dir.
2978  */
2979 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
2980 					struct inode *inode,
2981 					int *retval,
2982 					struct ext4_dir_entry_2 **parent_de,
2983 					int *inlined)
2984 {
2985 	struct buffer_head *bh;
2986 
2987 	if (!ext4_has_inline_data(inode)) {
2988 		bh = ext4_read_dirblock(inode, 0, EITHER);
2989 		if (IS_ERR(bh)) {
2990 			*retval = PTR_ERR(bh);
2991 			return NULL;
2992 		}
2993 		*parent_de = ext4_next_entry(
2994 					(struct ext4_dir_entry_2 *)bh->b_data,
2995 					inode->i_sb->s_blocksize);
2996 		return bh;
2997 	}
2998 
2999 	*inlined = 1;
3000 	return ext4_get_first_inline_block(inode, parent_de, retval);
3001 }
3002 
3003 struct ext4_renament {
3004 	struct inode *dir;
3005 	struct dentry *dentry;
3006 	struct inode *inode;
3007 	bool is_dir;
3008 	int dir_nlink_delta;
3009 
3010 	/* entry for "dentry" */
3011 	struct buffer_head *bh;
3012 	struct ext4_dir_entry_2 *de;
3013 	int inlined;
3014 
3015 	/* entry for ".." in inode if it's a directory */
3016 	struct buffer_head *dir_bh;
3017 	struct ext4_dir_entry_2 *parent_de;
3018 	int dir_inlined;
3019 };
3020 
3021 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3022 {
3023 	int retval;
3024 
3025 	ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3026 					      &retval, &ent->parent_de,
3027 					      &ent->dir_inlined);
3028 	if (!ent->dir_bh)
3029 		return retval;
3030 	if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3031 		return -EIO;
3032 	BUFFER_TRACE(ent->dir_bh, "get_write_access");
3033 	return ext4_journal_get_write_access(handle, ent->dir_bh);
3034 }
3035 
3036 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3037 				  unsigned dir_ino)
3038 {
3039 	int retval;
3040 
3041 	ent->parent_de->inode = cpu_to_le32(dir_ino);
3042 	BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3043 	if (!ent->dir_inlined) {
3044 		if (is_dx(ent->inode)) {
3045 			retval = ext4_handle_dirty_dx_node(handle,
3046 							   ent->inode,
3047 							   ent->dir_bh);
3048 		} else {
3049 			retval = ext4_handle_dirty_dirent_node(handle,
3050 							       ent->inode,
3051 							       ent->dir_bh);
3052 		}
3053 	} else {
3054 		retval = ext4_mark_inode_dirty(handle, ent->inode);
3055 	}
3056 	if (retval) {
3057 		ext4_std_error(ent->dir->i_sb, retval);
3058 		return retval;
3059 	}
3060 	return 0;
3061 }
3062 
3063 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3064 		       unsigned ino, unsigned file_type)
3065 {
3066 	int retval;
3067 
3068 	BUFFER_TRACE(ent->bh, "get write access");
3069 	retval = ext4_journal_get_write_access(handle, ent->bh);
3070 	if (retval)
3071 		return retval;
3072 	ent->de->inode = cpu_to_le32(ino);
3073 	if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb,
3074 				      EXT4_FEATURE_INCOMPAT_FILETYPE))
3075 		ent->de->file_type = file_type;
3076 	ent->dir->i_version++;
3077 	ent->dir->i_ctime = ent->dir->i_mtime =
3078 		ext4_current_time(ent->dir);
3079 	ext4_mark_inode_dirty(handle, ent->dir);
3080 	BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3081 	if (!ent->inlined) {
3082 		retval = ext4_handle_dirty_dirent_node(handle,
3083 						       ent->dir, ent->bh);
3084 		if (unlikely(retval)) {
3085 			ext4_std_error(ent->dir->i_sb, retval);
3086 			return retval;
3087 		}
3088 	}
3089 	brelse(ent->bh);
3090 	ent->bh = NULL;
3091 
3092 	return 0;
3093 }
3094 
3095 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3096 				  const struct qstr *d_name)
3097 {
3098 	int retval = -ENOENT;
3099 	struct buffer_head *bh;
3100 	struct ext4_dir_entry_2 *de;
3101 
3102 	bh = ext4_find_entry(dir, d_name, &de, NULL);
3103 	if (bh) {
3104 		retval = ext4_delete_entry(handle, dir, de, bh);
3105 		brelse(bh);
3106 	}
3107 	return retval;
3108 }
3109 
3110 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent)
3111 {
3112 	int retval;
3113 	/*
3114 	 * ent->de could have moved from under us during htree split, so make
3115 	 * sure that we are deleting the right entry.  We might also be pointing
3116 	 * to a stale entry in the unused part of ent->bh so just checking inum
3117 	 * and the name isn't enough.
3118 	 */
3119 	if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3120 	    ent->de->name_len != ent->dentry->d_name.len ||
3121 	    strncmp(ent->de->name, ent->dentry->d_name.name,
3122 		    ent->de->name_len)) {
3123 		retval = ext4_find_delete_entry(handle, ent->dir,
3124 						&ent->dentry->d_name);
3125 	} else {
3126 		retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3127 		if (retval == -ENOENT) {
3128 			retval = ext4_find_delete_entry(handle, ent->dir,
3129 							&ent->dentry->d_name);
3130 		}
3131 	}
3132 
3133 	if (retval) {
3134 		ext4_warning(ent->dir->i_sb,
3135 				"Deleting old file (%lu), %d, error=%d",
3136 				ent->dir->i_ino, ent->dir->i_nlink, retval);
3137 	}
3138 }
3139 
3140 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3141 {
3142 	if (ent->dir_nlink_delta) {
3143 		if (ent->dir_nlink_delta == -1)
3144 			ext4_dec_count(handle, ent->dir);
3145 		else
3146 			ext4_inc_count(handle, ent->dir);
3147 		ext4_mark_inode_dirty(handle, ent->dir);
3148 	}
3149 }
3150 
3151 /*
3152  * Anybody can rename anything with this: the permission checks are left to the
3153  * higher-level routines.
3154  *
3155  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3156  * while new_{dentry,inode) refers to the destination dentry/inode
3157  * This comes from rename(const char *oldpath, const char *newpath)
3158  */
3159 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3160 		       struct inode *new_dir, struct dentry *new_dentry)
3161 {
3162 	handle_t *handle = NULL;
3163 	struct ext4_renament old = {
3164 		.dir = old_dir,
3165 		.dentry = old_dentry,
3166 		.inode = old_dentry->d_inode,
3167 	};
3168 	struct ext4_renament new = {
3169 		.dir = new_dir,
3170 		.dentry = new_dentry,
3171 		.inode = new_dentry->d_inode,
3172 	};
3173 	int retval;
3174 
3175 	dquot_initialize(old.dir);
3176 	dquot_initialize(new.dir);
3177 
3178 	/* Initialize quotas before so that eventual writes go
3179 	 * in separate transaction */
3180 	if (new.inode)
3181 		dquot_initialize(new.inode);
3182 
3183 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3184 	/*
3185 	 *  Check for inode number is _not_ due to possible IO errors.
3186 	 *  We might rmdir the source, keep it as pwd of some process
3187 	 *  and merrily kill the link to whatever was created under the
3188 	 *  same name. Goodbye sticky bit ;-<
3189 	 */
3190 	retval = -ENOENT;
3191 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3192 		goto end_rename;
3193 
3194 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3195 				 &new.de, &new.inlined);
3196 	if (new.bh) {
3197 		if (!new.inode) {
3198 			brelse(new.bh);
3199 			new.bh = NULL;
3200 		}
3201 	}
3202 	if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3203 		ext4_alloc_da_blocks(old.inode);
3204 
3205 	handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3206 		(2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3207 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3208 	if (IS_ERR(handle))
3209 		return PTR_ERR(handle);
3210 
3211 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3212 		ext4_handle_sync(handle);
3213 
3214 	if (S_ISDIR(old.inode->i_mode)) {
3215 		if (new.inode) {
3216 			retval = -ENOTEMPTY;
3217 			if (!empty_dir(new.inode))
3218 				goto end_rename;
3219 		} else {
3220 			retval = -EMLINK;
3221 			if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3222 				goto end_rename;
3223 		}
3224 		retval = ext4_rename_dir_prepare(handle, &old);
3225 		if (retval)
3226 			goto end_rename;
3227 	}
3228 	if (!new.bh) {
3229 		retval = ext4_add_entry(handle, new.dentry, old.inode);
3230 		if (retval)
3231 			goto end_rename;
3232 	} else {
3233 		retval = ext4_setent(handle, &new,
3234 				     old.inode->i_ino, old.de->file_type);
3235 		if (retval)
3236 			goto end_rename;
3237 	}
3238 
3239 	/*
3240 	 * Like most other Unix systems, set the ctime for inodes on a
3241 	 * rename.
3242 	 */
3243 	old.inode->i_ctime = ext4_current_time(old.inode);
3244 	ext4_mark_inode_dirty(handle, old.inode);
3245 
3246 	/*
3247 	 * ok, that's it
3248 	 */
3249 	ext4_rename_delete(handle, &old);
3250 
3251 	if (new.inode) {
3252 		ext4_dec_count(handle, new.inode);
3253 		new.inode->i_ctime = ext4_current_time(new.inode);
3254 	}
3255 	old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3256 	ext4_update_dx_flag(old.dir);
3257 	if (old.dir_bh) {
3258 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3259 		if (retval)
3260 			goto end_rename;
3261 
3262 		ext4_dec_count(handle, old.dir);
3263 		if (new.inode) {
3264 			/* checked empty_dir above, can't have another parent,
3265 			 * ext4_dec_count() won't work for many-linked dirs */
3266 			clear_nlink(new.inode);
3267 		} else {
3268 			ext4_inc_count(handle, new.dir);
3269 			ext4_update_dx_flag(new.dir);
3270 			ext4_mark_inode_dirty(handle, new.dir);
3271 		}
3272 	}
3273 	ext4_mark_inode_dirty(handle, old.dir);
3274 	if (new.inode) {
3275 		ext4_mark_inode_dirty(handle, new.inode);
3276 		if (!new.inode->i_nlink)
3277 			ext4_orphan_add(handle, new.inode);
3278 	}
3279 	retval = 0;
3280 
3281 end_rename:
3282 	brelse(old.dir_bh);
3283 	brelse(old.bh);
3284 	brelse(new.bh);
3285 	if (handle)
3286 		ext4_journal_stop(handle);
3287 	return retval;
3288 }
3289 
3290 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3291 			     struct inode *new_dir, struct dentry *new_dentry)
3292 {
3293 	handle_t *handle = NULL;
3294 	struct ext4_renament old = {
3295 		.dir = old_dir,
3296 		.dentry = old_dentry,
3297 		.inode = old_dentry->d_inode,
3298 	};
3299 	struct ext4_renament new = {
3300 		.dir = new_dir,
3301 		.dentry = new_dentry,
3302 		.inode = new_dentry->d_inode,
3303 	};
3304 	u8 new_file_type;
3305 	int retval;
3306 
3307 	dquot_initialize(old.dir);
3308 	dquot_initialize(new.dir);
3309 
3310 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3311 				 &old.de, &old.inlined);
3312 	/*
3313 	 *  Check for inode number is _not_ due to possible IO errors.
3314 	 *  We might rmdir the source, keep it as pwd of some process
3315 	 *  and merrily kill the link to whatever was created under the
3316 	 *  same name. Goodbye sticky bit ;-<
3317 	 */
3318 	retval = -ENOENT;
3319 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3320 		goto end_rename;
3321 
3322 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3323 				 &new.de, &new.inlined);
3324 
3325 	/* RENAME_EXCHANGE case: old *and* new must both exist */
3326 	if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3327 		goto end_rename;
3328 
3329 	handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3330 		(2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3331 		 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3332 	if (IS_ERR(handle))
3333 		return PTR_ERR(handle);
3334 
3335 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3336 		ext4_handle_sync(handle);
3337 
3338 	if (S_ISDIR(old.inode->i_mode)) {
3339 		old.is_dir = true;
3340 		retval = ext4_rename_dir_prepare(handle, &old);
3341 		if (retval)
3342 			goto end_rename;
3343 	}
3344 	if (S_ISDIR(new.inode->i_mode)) {
3345 		new.is_dir = true;
3346 		retval = ext4_rename_dir_prepare(handle, &new);
3347 		if (retval)
3348 			goto end_rename;
3349 	}
3350 
3351 	/*
3352 	 * Other than the special case of overwriting a directory, parents'
3353 	 * nlink only needs to be modified if this is a cross directory rename.
3354 	 */
3355 	if (old.dir != new.dir && old.is_dir != new.is_dir) {
3356 		old.dir_nlink_delta = old.is_dir ? -1 : 1;
3357 		new.dir_nlink_delta = -old.dir_nlink_delta;
3358 		retval = -EMLINK;
3359 		if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3360 		    (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3361 			goto end_rename;
3362 	}
3363 
3364 	new_file_type = new.de->file_type;
3365 	retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3366 	if (retval)
3367 		goto end_rename;
3368 
3369 	retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3370 	if (retval)
3371 		goto end_rename;
3372 
3373 	/*
3374 	 * Like most other Unix systems, set the ctime for inodes on a
3375 	 * rename.
3376 	 */
3377 	old.inode->i_ctime = ext4_current_time(old.inode);
3378 	new.inode->i_ctime = ext4_current_time(new.inode);
3379 	ext4_mark_inode_dirty(handle, old.inode);
3380 	ext4_mark_inode_dirty(handle, new.inode);
3381 
3382 	if (old.dir_bh) {
3383 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3384 		if (retval)
3385 			goto end_rename;
3386 	}
3387 	if (new.dir_bh) {
3388 		retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3389 		if (retval)
3390 			goto end_rename;
3391 	}
3392 	ext4_update_dir_count(handle, &old);
3393 	ext4_update_dir_count(handle, &new);
3394 	retval = 0;
3395 
3396 end_rename:
3397 	brelse(old.dir_bh);
3398 	brelse(new.dir_bh);
3399 	brelse(old.bh);
3400 	brelse(new.bh);
3401 	if (handle)
3402 		ext4_journal_stop(handle);
3403 	return retval;
3404 }
3405 
3406 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3407 			struct inode *new_dir, struct dentry *new_dentry,
3408 			unsigned int flags)
3409 {
3410 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
3411 		return -EINVAL;
3412 
3413 	if (flags & RENAME_EXCHANGE) {
3414 		return ext4_cross_rename(old_dir, old_dentry,
3415 					 new_dir, new_dentry);
3416 	}
3417 	/*
3418 	 * Existence checking was done by the VFS, otherwise "RENAME_NOREPLACE"
3419 	 * is equivalent to regular rename.
3420 	 */
3421 	return ext4_rename(old_dir, old_dentry, new_dir, new_dentry);
3422 }
3423 
3424 /*
3425  * directories can handle most operations...
3426  */
3427 const struct inode_operations ext4_dir_inode_operations = {
3428 	.create		= ext4_create,
3429 	.lookup		= ext4_lookup,
3430 	.link		= ext4_link,
3431 	.unlink		= ext4_unlink,
3432 	.symlink	= ext4_symlink,
3433 	.mkdir		= ext4_mkdir,
3434 	.rmdir		= ext4_rmdir,
3435 	.mknod		= ext4_mknod,
3436 	.tmpfile	= ext4_tmpfile,
3437 	.rename		= ext4_rename,
3438 	.rename2	= ext4_rename2,
3439 	.setattr	= ext4_setattr,
3440 	.setxattr	= generic_setxattr,
3441 	.getxattr	= generic_getxattr,
3442 	.listxattr	= ext4_listxattr,
3443 	.removexattr	= generic_removexattr,
3444 	.get_acl	= ext4_get_acl,
3445 	.set_acl	= ext4_set_acl,
3446 	.fiemap         = ext4_fiemap,
3447 };
3448 
3449 const struct inode_operations ext4_special_inode_operations = {
3450 	.setattr	= ext4_setattr,
3451 	.setxattr	= generic_setxattr,
3452 	.getxattr	= generic_getxattr,
3453 	.listxattr	= ext4_listxattr,
3454 	.removexattr	= generic_removexattr,
3455 	.get_acl	= ext4_get_acl,
3456 	.set_acl	= ext4_set_acl,
3457 };
3458