xref: /linux/fs/ext4/namei.c (revision 71dfa617ea9f18e4585fe78364217cd32b1fc382)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/namei.c
4  *
5  * Copyright (C) 1992, 1993, 1994, 1995
6  * Remy Card (card@masi.ibp.fr)
7  * Laboratoire MASI - Institut Blaise Pascal
8  * Universite Pierre et Marie Curie (Paris VI)
9  *
10  *  from
11  *
12  *  linux/fs/minix/namei.c
13  *
14  *  Copyright (C) 1991, 1992  Linus Torvalds
15  *
16  *  Big-endian to little-endian byte-swapping/bitmaps by
17  *        David S. Miller (davem@caip.rutgers.edu), 1995
18  *  Directory entry file type support and forward compatibility hooks
19  *	for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20  *  Hash Tree Directory indexing (c)
21  *	Daniel Phillips, 2001
22  *  Hash Tree Directory indexing porting
23  *	Christopher Li, 2002
24  *  Hash Tree Directory indexing cleanup
25  *	Theodore Ts'o, 2002
26  */
27 
28 #include <linux/fs.h>
29 #include <linux/pagemap.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 <linux/iversion.h>
38 #include <linux/unicode.h>
39 #include "ext4.h"
40 #include "ext4_jbd2.h"
41 
42 #include "xattr.h"
43 #include "acl.h"
44 
45 #include <trace/events/ext4.h>
46 /*
47  * define how far ahead to read directories while searching them.
48  */
49 #define NAMEI_RA_CHUNKS  2
50 #define NAMEI_RA_BLOCKS  4
51 #define NAMEI_RA_SIZE	     (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
52 
53 static struct buffer_head *ext4_append(handle_t *handle,
54 					struct inode *inode,
55 					ext4_lblk_t *block)
56 {
57 	struct ext4_map_blocks map;
58 	struct buffer_head *bh;
59 	int err;
60 
61 	if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
62 		     ((inode->i_size >> 10) >=
63 		      EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
64 		return ERR_PTR(-ENOSPC);
65 
66 	*block = inode->i_size >> inode->i_sb->s_blocksize_bits;
67 	map.m_lblk = *block;
68 	map.m_len = 1;
69 
70 	/*
71 	 * We're appending new directory block. Make sure the block is not
72 	 * allocated yet, otherwise we will end up corrupting the
73 	 * directory.
74 	 */
75 	err = ext4_map_blocks(NULL, inode, &map, 0);
76 	if (err < 0)
77 		return ERR_PTR(err);
78 	if (err) {
79 		EXT4_ERROR_INODE(inode, "Logical block already allocated");
80 		return ERR_PTR(-EFSCORRUPTED);
81 	}
82 
83 	bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
84 	if (IS_ERR(bh))
85 		return bh;
86 	inode->i_size += inode->i_sb->s_blocksize;
87 	EXT4_I(inode)->i_disksize = inode->i_size;
88 	err = ext4_mark_inode_dirty(handle, inode);
89 	if (err)
90 		goto out;
91 	BUFFER_TRACE(bh, "get_write_access");
92 	err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
93 					    EXT4_JTR_NONE);
94 	if (err)
95 		goto out;
96 	return bh;
97 
98 out:
99 	brelse(bh);
100 	ext4_std_error(inode->i_sb, err);
101 	return ERR_PTR(err);
102 }
103 
104 static int ext4_dx_csum_verify(struct inode *inode,
105 			       struct ext4_dir_entry *dirent);
106 
107 /*
108  * Hints to ext4_read_dirblock regarding whether we expect a directory
109  * block being read to be an index block, or a block containing
110  * directory entries (and if the latter, whether it was found via a
111  * logical block in an htree index block).  This is used to control
112  * what sort of sanity checkinig ext4_read_dirblock() will do on the
113  * directory block read from the storage device.  EITHER will means
114  * the caller doesn't know what kind of directory block will be read,
115  * so no specific verification will be done.
116  */
117 typedef enum {
118 	EITHER, INDEX, DIRENT, DIRENT_HTREE
119 } dirblock_type_t;
120 
121 #define ext4_read_dirblock(inode, block, type) \
122 	__ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
123 
124 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
125 						ext4_lblk_t block,
126 						dirblock_type_t type,
127 						const char *func,
128 						unsigned int line)
129 {
130 	struct buffer_head *bh;
131 	struct ext4_dir_entry *dirent;
132 	int is_dx_block = 0;
133 
134 	if (block >= inode->i_size >> inode->i_blkbits) {
135 		ext4_error_inode(inode, func, line, block,
136 		       "Attempting to read directory block (%u) that is past i_size (%llu)",
137 		       block, inode->i_size);
138 		return ERR_PTR(-EFSCORRUPTED);
139 	}
140 
141 	if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
142 		bh = ERR_PTR(-EIO);
143 	else
144 		bh = ext4_bread(NULL, inode, block, 0);
145 	if (IS_ERR(bh)) {
146 		__ext4_warning(inode->i_sb, func, line,
147 			       "inode #%lu: lblock %lu: comm %s: "
148 			       "error %ld reading directory block",
149 			       inode->i_ino, (unsigned long)block,
150 			       current->comm, PTR_ERR(bh));
151 
152 		return bh;
153 	}
154 	if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
155 		ext4_error_inode(inode, func, line, block,
156 				 "Directory hole found for htree %s block",
157 				 (type == INDEX) ? "index" : "leaf");
158 		return ERR_PTR(-EFSCORRUPTED);
159 	}
160 	if (!bh)
161 		return NULL;
162 	dirent = (struct ext4_dir_entry *) bh->b_data;
163 	/* Determine whether or not we have an index block */
164 	if (is_dx(inode)) {
165 		if (block == 0)
166 			is_dx_block = 1;
167 		else if (ext4_rec_len_from_disk(dirent->rec_len,
168 						inode->i_sb->s_blocksize) ==
169 			 inode->i_sb->s_blocksize)
170 			is_dx_block = 1;
171 	}
172 	if (!is_dx_block && type == INDEX) {
173 		ext4_error_inode(inode, func, line, block,
174 		       "directory leaf block found instead of index block");
175 		brelse(bh);
176 		return ERR_PTR(-EFSCORRUPTED);
177 	}
178 	if (!ext4_has_metadata_csum(inode->i_sb) ||
179 	    buffer_verified(bh))
180 		return bh;
181 
182 	/*
183 	 * An empty leaf block can get mistaken for a index block; for
184 	 * this reason, we can only check the index checksum when the
185 	 * caller is sure it should be an index block.
186 	 */
187 	if (is_dx_block && type == INDEX) {
188 		if (ext4_dx_csum_verify(inode, dirent) &&
189 		    !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
190 			set_buffer_verified(bh);
191 		else {
192 			ext4_error_inode_err(inode, func, line, block,
193 					     EFSBADCRC,
194 					     "Directory index failed checksum");
195 			brelse(bh);
196 			return ERR_PTR(-EFSBADCRC);
197 		}
198 	}
199 	if (!is_dx_block) {
200 		if (ext4_dirblock_csum_verify(inode, bh) &&
201 		    !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
202 			set_buffer_verified(bh);
203 		else {
204 			ext4_error_inode_err(inode, func, line, block,
205 					     EFSBADCRC,
206 					     "Directory block failed checksum");
207 			brelse(bh);
208 			return ERR_PTR(-EFSBADCRC);
209 		}
210 	}
211 	return bh;
212 }
213 
214 #ifdef DX_DEBUG
215 #define dxtrace(command) command
216 #else
217 #define dxtrace(command)
218 #endif
219 
220 struct fake_dirent
221 {
222 	__le32 inode;
223 	__le16 rec_len;
224 	u8 name_len;
225 	u8 file_type;
226 };
227 
228 struct dx_countlimit
229 {
230 	__le16 limit;
231 	__le16 count;
232 };
233 
234 struct dx_entry
235 {
236 	__le32 hash;
237 	__le32 block;
238 };
239 
240 /*
241  * dx_root_info is laid out so that if it should somehow get overlaid by a
242  * dirent the two low bits of the hash version will be zero.  Therefore, the
243  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
244  */
245 
246 struct dx_root
247 {
248 	struct fake_dirent dot;
249 	char dot_name[4];
250 	struct fake_dirent dotdot;
251 	char dotdot_name[4];
252 	struct dx_root_info
253 	{
254 		__le32 reserved_zero;
255 		u8 hash_version;
256 		u8 info_length; /* 8 */
257 		u8 indirect_levels;
258 		u8 unused_flags;
259 	}
260 	info;
261 	struct dx_entry	entries[];
262 };
263 
264 struct dx_node
265 {
266 	struct fake_dirent fake;
267 	struct dx_entry	entries[];
268 };
269 
270 
271 struct dx_frame
272 {
273 	struct buffer_head *bh;
274 	struct dx_entry *entries;
275 	struct dx_entry *at;
276 };
277 
278 struct dx_map_entry
279 {
280 	u32 hash;
281 	u16 offs;
282 	u16 size;
283 };
284 
285 /*
286  * This goes at the end of each htree block.
287  */
288 struct dx_tail {
289 	u32 dt_reserved;
290 	__le32 dt_checksum;	/* crc32c(uuid+inum+dirblock) */
291 };
292 
293 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
294 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
295 static inline unsigned dx_get_hash(struct dx_entry *entry);
296 static void dx_set_hash(struct dx_entry *entry, unsigned value);
297 static unsigned dx_get_count(struct dx_entry *entries);
298 static unsigned dx_get_limit(struct dx_entry *entries);
299 static void dx_set_count(struct dx_entry *entries, unsigned value);
300 static void dx_set_limit(struct dx_entry *entries, unsigned value);
301 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
302 static unsigned dx_node_limit(struct inode *dir);
303 static struct dx_frame *dx_probe(struct ext4_filename *fname,
304 				 struct inode *dir,
305 				 struct dx_hash_info *hinfo,
306 				 struct dx_frame *frame);
307 static void dx_release(struct dx_frame *frames);
308 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
309 		       struct dx_hash_info *hinfo,
310 		       struct dx_map_entry *map_tail);
311 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
312 static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from,
313 					char *to, struct dx_map_entry *offsets,
314 					int count, unsigned int blocksize);
315 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
316 						unsigned int blocksize);
317 static void dx_insert_block(struct dx_frame *frame,
318 					u32 hash, ext4_lblk_t block);
319 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
320 				 struct dx_frame *frame,
321 				 struct dx_frame *frames,
322 				 __u32 *start_hash);
323 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
324 		struct ext4_filename *fname,
325 		struct ext4_dir_entry_2 **res_dir);
326 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
327 			     struct inode *dir, struct inode *inode);
328 
329 /* checksumming functions */
330 void ext4_initialize_dirent_tail(struct buffer_head *bh,
331 				 unsigned int blocksize)
332 {
333 	struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
334 
335 	memset(t, 0, sizeof(struct ext4_dir_entry_tail));
336 	t->det_rec_len = ext4_rec_len_to_disk(
337 			sizeof(struct ext4_dir_entry_tail), blocksize);
338 	t->det_reserved_ft = EXT4_FT_DIR_CSUM;
339 }
340 
341 /* Walk through a dirent block to find a checksum "dirent" at the tail */
342 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
343 						   struct buffer_head *bh)
344 {
345 	struct ext4_dir_entry_tail *t;
346 	int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
347 
348 #ifdef PARANOID
349 	struct ext4_dir_entry *d, *top;
350 
351 	d = (struct ext4_dir_entry *)bh->b_data;
352 	top = (struct ext4_dir_entry *)(bh->b_data +
353 		(blocksize - sizeof(struct ext4_dir_entry_tail)));
354 	while (d < top && ext4_rec_len_from_disk(d->rec_len, blocksize))
355 		d = (struct ext4_dir_entry *)(((void *)d) +
356 		    ext4_rec_len_from_disk(d->rec_len, blocksize));
357 
358 	if (d != top)
359 		return NULL;
360 
361 	t = (struct ext4_dir_entry_tail *)d;
362 #else
363 	t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
364 #endif
365 
366 	if (t->det_reserved_zero1 ||
367 	    (ext4_rec_len_from_disk(t->det_rec_len, blocksize) !=
368 	     sizeof(struct ext4_dir_entry_tail)) ||
369 	    t->det_reserved_zero2 ||
370 	    t->det_reserved_ft != EXT4_FT_DIR_CSUM)
371 		return NULL;
372 
373 	return t;
374 }
375 
376 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
377 {
378 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
379 	struct ext4_inode_info *ei = EXT4_I(inode);
380 	__u32 csum;
381 
382 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
383 	return cpu_to_le32(csum);
384 }
385 
386 #define warn_no_space_for_csum(inode)					\
387 	__warn_no_space_for_csum((inode), __func__, __LINE__)
388 
389 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
390 				     unsigned int line)
391 {
392 	__ext4_warning_inode(inode, func, line,
393 		"No space for directory leaf checksum. Please run e2fsck -D.");
394 }
395 
396 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
397 {
398 	struct ext4_dir_entry_tail *t;
399 
400 	if (!ext4_has_metadata_csum(inode->i_sb))
401 		return 1;
402 
403 	t = get_dirent_tail(inode, bh);
404 	if (!t) {
405 		warn_no_space_for_csum(inode);
406 		return 0;
407 	}
408 
409 	if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
410 						  (char *)t - bh->b_data))
411 		return 0;
412 
413 	return 1;
414 }
415 
416 static void ext4_dirblock_csum_set(struct inode *inode,
417 				 struct buffer_head *bh)
418 {
419 	struct ext4_dir_entry_tail *t;
420 
421 	if (!ext4_has_metadata_csum(inode->i_sb))
422 		return;
423 
424 	t = get_dirent_tail(inode, bh);
425 	if (!t) {
426 		warn_no_space_for_csum(inode);
427 		return;
428 	}
429 
430 	t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
431 					     (char *)t - bh->b_data);
432 }
433 
434 int ext4_handle_dirty_dirblock(handle_t *handle,
435 			       struct inode *inode,
436 			       struct buffer_head *bh)
437 {
438 	ext4_dirblock_csum_set(inode, bh);
439 	return ext4_handle_dirty_metadata(handle, inode, bh);
440 }
441 
442 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
443 					       struct ext4_dir_entry *dirent,
444 					       int *offset)
445 {
446 	struct ext4_dir_entry *dp;
447 	struct dx_root_info *root;
448 	int count_offset;
449 	int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
450 	unsigned int rlen = ext4_rec_len_from_disk(dirent->rec_len, blocksize);
451 
452 	if (rlen == blocksize)
453 		count_offset = 8;
454 	else if (rlen == 12) {
455 		dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
456 		if (ext4_rec_len_from_disk(dp->rec_len, blocksize) != blocksize - 12)
457 			return NULL;
458 		root = (struct dx_root_info *)(((void *)dp + 12));
459 		if (root->reserved_zero ||
460 		    root->info_length != sizeof(struct dx_root_info))
461 			return NULL;
462 		count_offset = 32;
463 	} else
464 		return NULL;
465 
466 	if (offset)
467 		*offset = count_offset;
468 	return (struct dx_countlimit *)(((void *)dirent) + count_offset);
469 }
470 
471 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
472 			   int count_offset, int count, struct dx_tail *t)
473 {
474 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
475 	struct ext4_inode_info *ei = EXT4_I(inode);
476 	__u32 csum;
477 	int size;
478 	__u32 dummy_csum = 0;
479 	int offset = offsetof(struct dx_tail, dt_checksum);
480 
481 	size = count_offset + (count * sizeof(struct dx_entry));
482 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
483 	csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
484 	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
485 
486 	return cpu_to_le32(csum);
487 }
488 
489 static int ext4_dx_csum_verify(struct inode *inode,
490 			       struct ext4_dir_entry *dirent)
491 {
492 	struct dx_countlimit *c;
493 	struct dx_tail *t;
494 	int count_offset, limit, count;
495 
496 	if (!ext4_has_metadata_csum(inode->i_sb))
497 		return 1;
498 
499 	c = get_dx_countlimit(inode, dirent, &count_offset);
500 	if (!c) {
501 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
502 		return 0;
503 	}
504 	limit = le16_to_cpu(c->limit);
505 	count = le16_to_cpu(c->count);
506 	if (count_offset + (limit * sizeof(struct dx_entry)) >
507 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
508 		warn_no_space_for_csum(inode);
509 		return 0;
510 	}
511 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
512 
513 	if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
514 					    count, t))
515 		return 0;
516 	return 1;
517 }
518 
519 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
520 {
521 	struct dx_countlimit *c;
522 	struct dx_tail *t;
523 	int count_offset, limit, count;
524 
525 	if (!ext4_has_metadata_csum(inode->i_sb))
526 		return;
527 
528 	c = get_dx_countlimit(inode, dirent, &count_offset);
529 	if (!c) {
530 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
531 		return;
532 	}
533 	limit = le16_to_cpu(c->limit);
534 	count = le16_to_cpu(c->count);
535 	if (count_offset + (limit * sizeof(struct dx_entry)) >
536 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
537 		warn_no_space_for_csum(inode);
538 		return;
539 	}
540 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
541 
542 	t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
543 }
544 
545 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
546 					    struct inode *inode,
547 					    struct buffer_head *bh)
548 {
549 	ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
550 	return ext4_handle_dirty_metadata(handle, inode, bh);
551 }
552 
553 /*
554  * p is at least 6 bytes before the end of page
555  */
556 static inline struct ext4_dir_entry_2 *
557 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
558 {
559 	return (struct ext4_dir_entry_2 *)((char *)p +
560 		ext4_rec_len_from_disk(p->rec_len, blocksize));
561 }
562 
563 /*
564  * Future: use high four bits of block for coalesce-on-delete flags
565  * Mask them off for now.
566  */
567 
568 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
569 {
570 	return le32_to_cpu(entry->block) & 0x0fffffff;
571 }
572 
573 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
574 {
575 	entry->block = cpu_to_le32(value);
576 }
577 
578 static inline unsigned dx_get_hash(struct dx_entry *entry)
579 {
580 	return le32_to_cpu(entry->hash);
581 }
582 
583 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
584 {
585 	entry->hash = cpu_to_le32(value);
586 }
587 
588 static inline unsigned dx_get_count(struct dx_entry *entries)
589 {
590 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
591 }
592 
593 static inline unsigned dx_get_limit(struct dx_entry *entries)
594 {
595 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
596 }
597 
598 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
599 {
600 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
601 }
602 
603 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
604 {
605 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
606 }
607 
608 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
609 {
610 	unsigned int entry_space = dir->i_sb->s_blocksize -
611 			ext4_dir_rec_len(1, NULL) -
612 			ext4_dir_rec_len(2, NULL) - infosize;
613 
614 	if (ext4_has_metadata_csum(dir->i_sb))
615 		entry_space -= sizeof(struct dx_tail);
616 	return entry_space / sizeof(struct dx_entry);
617 }
618 
619 static inline unsigned dx_node_limit(struct inode *dir)
620 {
621 	unsigned int entry_space = dir->i_sb->s_blocksize -
622 			ext4_dir_rec_len(0, dir);
623 
624 	if (ext4_has_metadata_csum(dir->i_sb))
625 		entry_space -= sizeof(struct dx_tail);
626 	return entry_space / sizeof(struct dx_entry);
627 }
628 
629 /*
630  * Debug
631  */
632 #ifdef DX_DEBUG
633 static void dx_show_index(char * label, struct dx_entry *entries)
634 {
635 	int i, n = dx_get_count (entries);
636 	printk(KERN_DEBUG "%s index", label);
637 	for (i = 0; i < n; i++) {
638 		printk(KERN_CONT " %x->%lu",
639 		       i ? dx_get_hash(entries + i) : 0,
640 		       (unsigned long)dx_get_block(entries + i));
641 	}
642 	printk(KERN_CONT "\n");
643 }
644 
645 struct stats
646 {
647 	unsigned names;
648 	unsigned space;
649 	unsigned bcount;
650 };
651 
652 static struct stats dx_show_leaf(struct inode *dir,
653 				struct dx_hash_info *hinfo,
654 				struct ext4_dir_entry_2 *de,
655 				int size, int show_names)
656 {
657 	unsigned names = 0, space = 0;
658 	char *base = (char *) de;
659 	struct dx_hash_info h = *hinfo;
660 
661 	printk("names: ");
662 	while ((char *) de < base + size)
663 	{
664 		if (de->inode)
665 		{
666 			if (show_names)
667 			{
668 #ifdef CONFIG_FS_ENCRYPTION
669 				int len;
670 				char *name;
671 				struct fscrypt_str fname_crypto_str =
672 					FSTR_INIT(NULL, 0);
673 				int res = 0;
674 
675 				name  = de->name;
676 				len = de->name_len;
677 				if (!IS_ENCRYPTED(dir)) {
678 					/* Directory is not encrypted */
679 					(void) ext4fs_dirhash(dir, de->name,
680 						de->name_len, &h);
681 					printk("%*.s:(U)%x.%u ", len,
682 					       name, h.hash,
683 					       (unsigned) ((char *) de
684 							   - base));
685 				} else {
686 					struct fscrypt_str de_name =
687 						FSTR_INIT(name, len);
688 
689 					/* Directory is encrypted */
690 					res = fscrypt_fname_alloc_buffer(
691 						len, &fname_crypto_str);
692 					if (res)
693 						printk(KERN_WARNING "Error "
694 							"allocating crypto "
695 							"buffer--skipping "
696 							"crypto\n");
697 					res = fscrypt_fname_disk_to_usr(dir,
698 						0, 0, &de_name,
699 						&fname_crypto_str);
700 					if (res) {
701 						printk(KERN_WARNING "Error "
702 							"converting filename "
703 							"from disk to usr"
704 							"\n");
705 						name = "??";
706 						len = 2;
707 					} else {
708 						name = fname_crypto_str.name;
709 						len = fname_crypto_str.len;
710 					}
711 					if (IS_CASEFOLDED(dir))
712 						h.hash = EXT4_DIRENT_HASH(de);
713 					else
714 						(void) ext4fs_dirhash(dir,
715 							de->name,
716 							de->name_len, &h);
717 					printk("%*.s:(E)%x.%u ", len, name,
718 					       h.hash, (unsigned) ((char *) de
719 								   - base));
720 					fscrypt_fname_free_buffer(
721 							&fname_crypto_str);
722 				}
723 #else
724 				int len = de->name_len;
725 				char *name = de->name;
726 				(void) ext4fs_dirhash(dir, de->name,
727 						      de->name_len, &h);
728 				printk("%*.s:%x.%u ", len, name, h.hash,
729 				       (unsigned) ((char *) de - base));
730 #endif
731 			}
732 			space += ext4_dir_rec_len(de->name_len, dir);
733 			names++;
734 		}
735 		de = ext4_next_entry(de, size);
736 	}
737 	printk(KERN_CONT "(%i)\n", names);
738 	return (struct stats) { names, space, 1 };
739 }
740 
741 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
742 			     struct dx_entry *entries, int levels)
743 {
744 	unsigned blocksize = dir->i_sb->s_blocksize;
745 	unsigned count = dx_get_count(entries), names = 0, space = 0, i;
746 	unsigned bcount = 0;
747 	struct buffer_head *bh;
748 	printk("%i indexed blocks...\n", count);
749 	for (i = 0; i < count; i++, entries++)
750 	{
751 		ext4_lblk_t block = dx_get_block(entries);
752 		ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
753 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
754 		struct stats stats;
755 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
756 		bh = ext4_bread(NULL,dir, block, 0);
757 		if (!bh || IS_ERR(bh))
758 			continue;
759 		stats = levels?
760 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
761 		   dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
762 			bh->b_data, blocksize, 0);
763 		names += stats.names;
764 		space += stats.space;
765 		bcount += stats.bcount;
766 		brelse(bh);
767 	}
768 	if (bcount)
769 		printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
770 		       levels ? "" : "   ", names, space/bcount,
771 		       (space/bcount)*100/blocksize);
772 	return (struct stats) { names, space, bcount};
773 }
774 
775 /*
776  * Linear search cross check
777  */
778 static inline void htree_rep_invariant_check(struct dx_entry *at,
779 					     struct dx_entry *target,
780 					     u32 hash, unsigned int n)
781 {
782 	while (n--) {
783 		dxtrace(printk(KERN_CONT ","));
784 		if (dx_get_hash(++at) > hash) {
785 			at--;
786 			break;
787 		}
788 	}
789 	ASSERT(at == target - 1);
790 }
791 #else /* DX_DEBUG */
792 static inline void htree_rep_invariant_check(struct dx_entry *at,
793 					     struct dx_entry *target,
794 					     u32 hash, unsigned int n)
795 {
796 }
797 #endif /* DX_DEBUG */
798 
799 /*
800  * Probe for a directory leaf block to search.
801  *
802  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
803  * error in the directory index, and the caller should fall back to
804  * searching the directory normally.  The callers of dx_probe **MUST**
805  * check for this error code, and make sure it never gets reflected
806  * back to userspace.
807  */
808 static struct dx_frame *
809 dx_probe(struct ext4_filename *fname, struct inode *dir,
810 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
811 {
812 	unsigned count, indirect, level, i;
813 	struct dx_entry *at, *entries, *p, *q, *m;
814 	struct dx_root *root;
815 	struct dx_frame *frame = frame_in;
816 	struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
817 	u32 hash;
818 	ext4_lblk_t block;
819 	ext4_lblk_t blocks[EXT4_HTREE_LEVEL];
820 
821 	memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
822 	frame->bh = ext4_read_dirblock(dir, 0, INDEX);
823 	if (IS_ERR(frame->bh))
824 		return (struct dx_frame *) frame->bh;
825 
826 	root = (struct dx_root *) frame->bh->b_data;
827 	if (root->info.hash_version != DX_HASH_TEA &&
828 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
829 	    root->info.hash_version != DX_HASH_LEGACY &&
830 	    root->info.hash_version != DX_HASH_SIPHASH) {
831 		ext4_warning_inode(dir, "Unrecognised inode hash code %u",
832 				   root->info.hash_version);
833 		goto fail;
834 	}
835 	if (ext4_hash_in_dirent(dir)) {
836 		if (root->info.hash_version != DX_HASH_SIPHASH) {
837 			ext4_warning_inode(dir,
838 				"Hash in dirent, but hash is not SIPHASH");
839 			goto fail;
840 		}
841 	} else {
842 		if (root->info.hash_version == DX_HASH_SIPHASH) {
843 			ext4_warning_inode(dir,
844 				"Hash code is SIPHASH, but hash not in dirent");
845 			goto fail;
846 		}
847 	}
848 	if (fname)
849 		hinfo = &fname->hinfo;
850 	hinfo->hash_version = root->info.hash_version;
851 	if (hinfo->hash_version <= DX_HASH_TEA)
852 		hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
853 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
854 	/* hash is already computed for encrypted casefolded directory */
855 	if (fname && fname_name(fname) &&
856 	    !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir))) {
857 		int ret = ext4fs_dirhash(dir, fname_name(fname),
858 					 fname_len(fname), hinfo);
859 		if (ret < 0) {
860 			ret_err = ERR_PTR(ret);
861 			goto fail;
862 		}
863 	}
864 	hash = hinfo->hash;
865 
866 	if (root->info.unused_flags & 1) {
867 		ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
868 				   root->info.unused_flags);
869 		goto fail;
870 	}
871 
872 	indirect = root->info.indirect_levels;
873 	if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
874 		ext4_warning(dir->i_sb,
875 			     "Directory (ino: %lu) htree depth %#06x exceed"
876 			     "supported value", dir->i_ino,
877 			     ext4_dir_htree_level(dir->i_sb));
878 		if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
879 			ext4_warning(dir->i_sb, "Enable large directory "
880 						"feature to access it");
881 		}
882 		goto fail;
883 	}
884 
885 	entries = (struct dx_entry *)(((char *)&root->info) +
886 				      root->info.info_length);
887 
888 	if (dx_get_limit(entries) != dx_root_limit(dir,
889 						   root->info.info_length)) {
890 		ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
891 				   dx_get_limit(entries),
892 				   dx_root_limit(dir, root->info.info_length));
893 		goto fail;
894 	}
895 
896 	dxtrace(printk("Look up %x", hash));
897 	level = 0;
898 	blocks[0] = 0;
899 	while (1) {
900 		count = dx_get_count(entries);
901 		if (!count || count > dx_get_limit(entries)) {
902 			ext4_warning_inode(dir,
903 					   "dx entry: count %u beyond limit %u",
904 					   count, dx_get_limit(entries));
905 			goto fail;
906 		}
907 
908 		p = entries + 1;
909 		q = entries + count - 1;
910 		while (p <= q) {
911 			m = p + (q - p) / 2;
912 			dxtrace(printk(KERN_CONT "."));
913 			if (dx_get_hash(m) > hash)
914 				q = m - 1;
915 			else
916 				p = m + 1;
917 		}
918 
919 		htree_rep_invariant_check(entries, p, hash, count - 1);
920 
921 		at = p - 1;
922 		dxtrace(printk(KERN_CONT " %x->%u\n",
923 			       at == entries ? 0 : dx_get_hash(at),
924 			       dx_get_block(at)));
925 		frame->entries = entries;
926 		frame->at = at;
927 
928 		block = dx_get_block(at);
929 		for (i = 0; i <= level; i++) {
930 			if (blocks[i] == block) {
931 				ext4_warning_inode(dir,
932 					"dx entry: tree cycle block %u points back to block %u",
933 					blocks[level], block);
934 				goto fail;
935 			}
936 		}
937 		if (++level > indirect)
938 			return frame;
939 		blocks[level] = block;
940 		frame++;
941 		frame->bh = ext4_read_dirblock(dir, block, INDEX);
942 		if (IS_ERR(frame->bh)) {
943 			ret_err = (struct dx_frame *) frame->bh;
944 			frame->bh = NULL;
945 			goto fail;
946 		}
947 
948 		entries = ((struct dx_node *) frame->bh->b_data)->entries;
949 
950 		if (dx_get_limit(entries) != dx_node_limit(dir)) {
951 			ext4_warning_inode(dir,
952 				"dx entry: limit %u != node limit %u",
953 				dx_get_limit(entries), dx_node_limit(dir));
954 			goto fail;
955 		}
956 	}
957 fail:
958 	while (frame >= frame_in) {
959 		brelse(frame->bh);
960 		frame--;
961 	}
962 
963 	if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
964 		ext4_warning_inode(dir,
965 			"Corrupt directory, running e2fsck is recommended");
966 	return ret_err;
967 }
968 
969 static void dx_release(struct dx_frame *frames)
970 {
971 	struct dx_root_info *info;
972 	int i;
973 	unsigned int indirect_levels;
974 
975 	if (frames[0].bh == NULL)
976 		return;
977 
978 	info = &((struct dx_root *)frames[0].bh->b_data)->info;
979 	/* save local copy, "info" may be freed after brelse() */
980 	indirect_levels = info->indirect_levels;
981 	for (i = 0; i <= indirect_levels; i++) {
982 		if (frames[i].bh == NULL)
983 			break;
984 		brelse(frames[i].bh);
985 		frames[i].bh = NULL;
986 	}
987 }
988 
989 /*
990  * This function increments the frame pointer to search the next leaf
991  * block, and reads in the necessary intervening nodes if the search
992  * should be necessary.  Whether or not the search is necessary is
993  * controlled by the hash parameter.  If the hash value is even, then
994  * the search is only continued if the next block starts with that
995  * hash value.  This is used if we are searching for a specific file.
996  *
997  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
998  *
999  * This function returns 1 if the caller should continue to search,
1000  * or 0 if it should not.  If there is an error reading one of the
1001  * index blocks, it will a negative error code.
1002  *
1003  * If start_hash is non-null, it will be filled in with the starting
1004  * hash of the next page.
1005  */
1006 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
1007 				 struct dx_frame *frame,
1008 				 struct dx_frame *frames,
1009 				 __u32 *start_hash)
1010 {
1011 	struct dx_frame *p;
1012 	struct buffer_head *bh;
1013 	int num_frames = 0;
1014 	__u32 bhash;
1015 
1016 	p = frame;
1017 	/*
1018 	 * Find the next leaf page by incrementing the frame pointer.
1019 	 * If we run out of entries in the interior node, loop around and
1020 	 * increment pointer in the parent node.  When we break out of
1021 	 * this loop, num_frames indicates the number of interior
1022 	 * nodes need to be read.
1023 	 */
1024 	while (1) {
1025 		if (++(p->at) < p->entries + dx_get_count(p->entries))
1026 			break;
1027 		if (p == frames)
1028 			return 0;
1029 		num_frames++;
1030 		p--;
1031 	}
1032 
1033 	/*
1034 	 * If the hash is 1, then continue only if the next page has a
1035 	 * continuation hash of any value.  This is used for readdir
1036 	 * handling.  Otherwise, check to see if the hash matches the
1037 	 * desired continuation hash.  If it doesn't, return since
1038 	 * there's no point to read in the successive index pages.
1039 	 */
1040 	bhash = dx_get_hash(p->at);
1041 	if (start_hash)
1042 		*start_hash = bhash;
1043 	if ((hash & 1) == 0) {
1044 		if ((bhash & ~1) != hash)
1045 			return 0;
1046 	}
1047 	/*
1048 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
1049 	 * block so no check is necessary
1050 	 */
1051 	while (num_frames--) {
1052 		bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
1053 		if (IS_ERR(bh))
1054 			return PTR_ERR(bh);
1055 		p++;
1056 		brelse(p->bh);
1057 		p->bh = bh;
1058 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1059 	}
1060 	return 1;
1061 }
1062 
1063 
1064 /*
1065  * This function fills a red-black tree with information from a
1066  * directory block.  It returns the number directory entries loaded
1067  * into the tree.  If there is an error it is returned in err.
1068  */
1069 static int htree_dirblock_to_tree(struct file *dir_file,
1070 				  struct inode *dir, ext4_lblk_t block,
1071 				  struct dx_hash_info *hinfo,
1072 				  __u32 start_hash, __u32 start_minor_hash)
1073 {
1074 	struct buffer_head *bh;
1075 	struct ext4_dir_entry_2 *de, *top;
1076 	int err = 0, count = 0;
1077 	struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1078 	int csum = ext4_has_metadata_csum(dir->i_sb);
1079 
1080 	dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1081 							(unsigned long)block));
1082 	bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1083 	if (IS_ERR(bh))
1084 		return PTR_ERR(bh);
1085 
1086 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1087 	/* csum entries are not larger in the casefolded encrypted case */
1088 	top = (struct ext4_dir_entry_2 *) ((char *) de +
1089 					   dir->i_sb->s_blocksize -
1090 					   ext4_dir_rec_len(0,
1091 							   csum ? NULL : dir));
1092 	/* Check if the directory is encrypted */
1093 	if (IS_ENCRYPTED(dir)) {
1094 		err = fscrypt_prepare_readdir(dir);
1095 		if (err < 0) {
1096 			brelse(bh);
1097 			return err;
1098 		}
1099 		err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1100 						 &fname_crypto_str);
1101 		if (err < 0) {
1102 			brelse(bh);
1103 			return err;
1104 		}
1105 	}
1106 
1107 	for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1108 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1109 				bh->b_data, bh->b_size,
1110 				(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1111 					 + ((char *)de - bh->b_data))) {
1112 			/* silently ignore the rest of the block */
1113 			break;
1114 		}
1115 		if (ext4_hash_in_dirent(dir)) {
1116 			if (de->name_len && de->inode) {
1117 				hinfo->hash = EXT4_DIRENT_HASH(de);
1118 				hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1119 			} else {
1120 				hinfo->hash = 0;
1121 				hinfo->minor_hash = 0;
1122 			}
1123 		} else {
1124 			err = ext4fs_dirhash(dir, de->name,
1125 					     de->name_len, hinfo);
1126 			if (err < 0) {
1127 				count = err;
1128 				goto errout;
1129 			}
1130 		}
1131 		if ((hinfo->hash < start_hash) ||
1132 		    ((hinfo->hash == start_hash) &&
1133 		     (hinfo->minor_hash < start_minor_hash)))
1134 			continue;
1135 		if (de->inode == 0)
1136 			continue;
1137 		if (!IS_ENCRYPTED(dir)) {
1138 			tmp_str.name = de->name;
1139 			tmp_str.len = de->name_len;
1140 			err = ext4_htree_store_dirent(dir_file,
1141 				   hinfo->hash, hinfo->minor_hash, de,
1142 				   &tmp_str);
1143 		} else {
1144 			int save_len = fname_crypto_str.len;
1145 			struct fscrypt_str de_name = FSTR_INIT(de->name,
1146 								de->name_len);
1147 
1148 			/* Directory is encrypted */
1149 			err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1150 					hinfo->minor_hash, &de_name,
1151 					&fname_crypto_str);
1152 			if (err) {
1153 				count = err;
1154 				goto errout;
1155 			}
1156 			err = ext4_htree_store_dirent(dir_file,
1157 				   hinfo->hash, hinfo->minor_hash, de,
1158 					&fname_crypto_str);
1159 			fname_crypto_str.len = save_len;
1160 		}
1161 		if (err != 0) {
1162 			count = err;
1163 			goto errout;
1164 		}
1165 		count++;
1166 	}
1167 errout:
1168 	brelse(bh);
1169 	fscrypt_fname_free_buffer(&fname_crypto_str);
1170 	return count;
1171 }
1172 
1173 
1174 /*
1175  * This function fills a red-black tree with information from a
1176  * directory.  We start scanning the directory in hash order, starting
1177  * at start_hash and start_minor_hash.
1178  *
1179  * This function returns the number of entries inserted into the tree,
1180  * or a negative error code.
1181  */
1182 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1183 			 __u32 start_minor_hash, __u32 *next_hash)
1184 {
1185 	struct dx_hash_info hinfo;
1186 	struct ext4_dir_entry_2 *de;
1187 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1188 	struct inode *dir;
1189 	ext4_lblk_t block;
1190 	int count = 0;
1191 	int ret, err;
1192 	__u32 hashval;
1193 	struct fscrypt_str tmp_str;
1194 
1195 	dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1196 		       start_hash, start_minor_hash));
1197 	dir = file_inode(dir_file);
1198 	if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1199 		if (ext4_hash_in_dirent(dir))
1200 			hinfo.hash_version = DX_HASH_SIPHASH;
1201 		else
1202 			hinfo.hash_version =
1203 					EXT4_SB(dir->i_sb)->s_def_hash_version;
1204 		if (hinfo.hash_version <= DX_HASH_TEA)
1205 			hinfo.hash_version +=
1206 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
1207 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1208 		if (ext4_has_inline_data(dir)) {
1209 			int has_inline_data = 1;
1210 			count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1211 						       &hinfo, start_hash,
1212 						       start_minor_hash,
1213 						       &has_inline_data);
1214 			if (has_inline_data) {
1215 				*next_hash = ~0;
1216 				return count;
1217 			}
1218 		}
1219 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1220 					       start_hash, start_minor_hash);
1221 		*next_hash = ~0;
1222 		return count;
1223 	}
1224 	hinfo.hash = start_hash;
1225 	hinfo.minor_hash = 0;
1226 	frame = dx_probe(NULL, dir, &hinfo, frames);
1227 	if (IS_ERR(frame))
1228 		return PTR_ERR(frame);
1229 
1230 	/* Add '.' and '..' from the htree header */
1231 	if (!start_hash && !start_minor_hash) {
1232 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1233 		tmp_str.name = de->name;
1234 		tmp_str.len = de->name_len;
1235 		err = ext4_htree_store_dirent(dir_file, 0, 0,
1236 					      de, &tmp_str);
1237 		if (err != 0)
1238 			goto errout;
1239 		count++;
1240 	}
1241 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1242 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1243 		de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1244 		tmp_str.name = de->name;
1245 		tmp_str.len = de->name_len;
1246 		err = ext4_htree_store_dirent(dir_file, 2, 0,
1247 					      de, &tmp_str);
1248 		if (err != 0)
1249 			goto errout;
1250 		count++;
1251 	}
1252 
1253 	while (1) {
1254 		if (fatal_signal_pending(current)) {
1255 			err = -ERESTARTSYS;
1256 			goto errout;
1257 		}
1258 		cond_resched();
1259 		block = dx_get_block(frame->at);
1260 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1261 					     start_hash, start_minor_hash);
1262 		if (ret < 0) {
1263 			err = ret;
1264 			goto errout;
1265 		}
1266 		count += ret;
1267 		hashval = ~0;
1268 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1269 					    frame, frames, &hashval);
1270 		*next_hash = hashval;
1271 		if (ret < 0) {
1272 			err = ret;
1273 			goto errout;
1274 		}
1275 		/*
1276 		 * Stop if:  (a) there are no more entries, or
1277 		 * (b) we have inserted at least one entry and the
1278 		 * next hash value is not a continuation
1279 		 */
1280 		if ((ret == 0) ||
1281 		    (count && ((hashval & 1) == 0)))
1282 			break;
1283 	}
1284 	dx_release(frames);
1285 	dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1286 		       "next hash: %x\n", count, *next_hash));
1287 	return count;
1288 errout:
1289 	dx_release(frames);
1290 	return (err);
1291 }
1292 
1293 static inline int search_dirblock(struct buffer_head *bh,
1294 				  struct inode *dir,
1295 				  struct ext4_filename *fname,
1296 				  unsigned int offset,
1297 				  struct ext4_dir_entry_2 **res_dir)
1298 {
1299 	return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1300 			       fname, offset, res_dir);
1301 }
1302 
1303 /*
1304  * Directory block splitting, compacting
1305  */
1306 
1307 /*
1308  * Create map of hash values, offsets, and sizes, stored at end of block.
1309  * Returns number of entries mapped.
1310  */
1311 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
1312 		       struct dx_hash_info *hinfo,
1313 		       struct dx_map_entry *map_tail)
1314 {
1315 	int count = 0;
1316 	struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)bh->b_data;
1317 	unsigned int buflen = bh->b_size;
1318 	char *base = bh->b_data;
1319 	struct dx_hash_info h = *hinfo;
1320 	int blocksize = EXT4_BLOCK_SIZE(dir->i_sb);
1321 
1322 	if (ext4_has_metadata_csum(dir->i_sb))
1323 		buflen -= sizeof(struct ext4_dir_entry_tail);
1324 
1325 	while ((char *) de < base + buflen) {
1326 		if (ext4_check_dir_entry(dir, NULL, de, bh, base, buflen,
1327 					 ((char *)de) - base))
1328 			return -EFSCORRUPTED;
1329 		if (de->name_len && de->inode) {
1330 			if (ext4_hash_in_dirent(dir))
1331 				h.hash = EXT4_DIRENT_HASH(de);
1332 			else {
1333 				int err = ext4fs_dirhash(dir, de->name,
1334 						     de->name_len, &h);
1335 				if (err < 0)
1336 					return err;
1337 			}
1338 			map_tail--;
1339 			map_tail->hash = h.hash;
1340 			map_tail->offs = ((char *) de - base)>>2;
1341 			map_tail->size = ext4_rec_len_from_disk(de->rec_len,
1342 								blocksize);
1343 			count++;
1344 			cond_resched();
1345 		}
1346 		de = ext4_next_entry(de, blocksize);
1347 	}
1348 	return count;
1349 }
1350 
1351 /* Sort map by hash value */
1352 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1353 {
1354 	struct dx_map_entry *p, *q, *top = map + count - 1;
1355 	int more;
1356 	/* Combsort until bubble sort doesn't suck */
1357 	while (count > 2) {
1358 		count = count*10/13;
1359 		if (count - 9 < 2) /* 9, 10 -> 11 */
1360 			count = 11;
1361 		for (p = top, q = p - count; q >= map; p--, q--)
1362 			if (p->hash < q->hash)
1363 				swap(*p, *q);
1364 	}
1365 	/* Garden variety bubble sort */
1366 	do {
1367 		more = 0;
1368 		q = top;
1369 		while (q-- > map) {
1370 			if (q[1].hash >= q[0].hash)
1371 				continue;
1372 			swap(*(q+1), *q);
1373 			more = 1;
1374 		}
1375 	} while(more);
1376 }
1377 
1378 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1379 {
1380 	struct dx_entry *entries = frame->entries;
1381 	struct dx_entry *old = frame->at, *new = old + 1;
1382 	int count = dx_get_count(entries);
1383 
1384 	ASSERT(count < dx_get_limit(entries));
1385 	ASSERT(old < entries + count);
1386 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1387 	dx_set_hash(new, hash);
1388 	dx_set_block(new, block);
1389 	dx_set_count(entries, count + 1);
1390 }
1391 
1392 #if IS_ENABLED(CONFIG_UNICODE)
1393 /*
1394  * Test whether a case-insensitive directory entry matches the filename
1395  * being searched for.  If quick is set, assume the name being looked up
1396  * is already in the casefolded form.
1397  *
1398  * Returns: 0 if the directory entry matches, more than 0 if it
1399  * doesn't match or less than zero on error.
1400  */
1401 static int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1402 			   u8 *de_name, size_t de_name_len, bool quick)
1403 {
1404 	const struct super_block *sb = parent->i_sb;
1405 	const struct unicode_map *um = sb->s_encoding;
1406 	struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1407 	struct qstr entry = QSTR_INIT(de_name, de_name_len);
1408 	int ret;
1409 
1410 	if (IS_ENCRYPTED(parent)) {
1411 		const struct fscrypt_str encrypted_name =
1412 				FSTR_INIT(de_name, de_name_len);
1413 
1414 		decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1415 		if (!decrypted_name.name)
1416 			return -ENOMEM;
1417 		ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1418 						&decrypted_name);
1419 		if (ret < 0)
1420 			goto out;
1421 		entry.name = decrypted_name.name;
1422 		entry.len = decrypted_name.len;
1423 	}
1424 
1425 	if (quick)
1426 		ret = utf8_strncasecmp_folded(um, name, &entry);
1427 	else
1428 		ret = utf8_strncasecmp(um, name, &entry);
1429 	if (ret < 0) {
1430 		/* Handle invalid character sequence as either an error
1431 		 * or as an opaque byte sequence.
1432 		 */
1433 		if (sb_has_strict_encoding(sb))
1434 			ret = -EINVAL;
1435 		else if (name->len != entry.len)
1436 			ret = 1;
1437 		else
1438 			ret = !!memcmp(name->name, entry.name, entry.len);
1439 	}
1440 out:
1441 	kfree(decrypted_name.name);
1442 	return ret;
1443 }
1444 
1445 int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1446 				  struct ext4_filename *name)
1447 {
1448 	struct fscrypt_str *cf_name = &name->cf_name;
1449 	struct dx_hash_info *hinfo = &name->hinfo;
1450 	int len;
1451 
1452 	if (!IS_CASEFOLDED(dir) ||
1453 	    (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1454 		cf_name->name = NULL;
1455 		return 0;
1456 	}
1457 
1458 	cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1459 	if (!cf_name->name)
1460 		return -ENOMEM;
1461 
1462 	len = utf8_casefold(dir->i_sb->s_encoding,
1463 			    iname, cf_name->name,
1464 			    EXT4_NAME_LEN);
1465 	if (len <= 0) {
1466 		kfree(cf_name->name);
1467 		cf_name->name = NULL;
1468 	}
1469 	cf_name->len = (unsigned) len;
1470 	if (!IS_ENCRYPTED(dir))
1471 		return 0;
1472 
1473 	hinfo->hash_version = DX_HASH_SIPHASH;
1474 	hinfo->seed = NULL;
1475 	if (cf_name->name)
1476 		return ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1477 	else
1478 		return ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1479 }
1480 #endif
1481 
1482 /*
1483  * Test whether a directory entry matches the filename being searched for.
1484  *
1485  * Return: %true if the directory entry matches, otherwise %false.
1486  */
1487 static bool ext4_match(struct inode *parent,
1488 			      const struct ext4_filename *fname,
1489 			      struct ext4_dir_entry_2 *de)
1490 {
1491 	struct fscrypt_name f;
1492 
1493 	if (!de->inode)
1494 		return false;
1495 
1496 	f.usr_fname = fname->usr_fname;
1497 	f.disk_name = fname->disk_name;
1498 #ifdef CONFIG_FS_ENCRYPTION
1499 	f.crypto_buf = fname->crypto_buf;
1500 #endif
1501 
1502 #if IS_ENABLED(CONFIG_UNICODE)
1503 	if (IS_CASEFOLDED(parent) &&
1504 	    (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1505 		if (fname->cf_name.name) {
1506 			struct qstr cf = {.name = fname->cf_name.name,
1507 					  .len = fname->cf_name.len};
1508 			if (IS_ENCRYPTED(parent)) {
1509 				if (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1510 					fname->hinfo.minor_hash !=
1511 						EXT4_DIRENT_MINOR_HASH(de)) {
1512 
1513 					return false;
1514 				}
1515 			}
1516 			return !ext4_ci_compare(parent, &cf, de->name,
1517 							de->name_len, true);
1518 		}
1519 		return !ext4_ci_compare(parent, fname->usr_fname, de->name,
1520 						de->name_len, false);
1521 	}
1522 #endif
1523 
1524 	return fscrypt_match_name(&f, de->name, de->name_len);
1525 }
1526 
1527 /*
1528  * Returns 0 if not found, -1 on failure, and 1 on success
1529  */
1530 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1531 		    struct inode *dir, struct ext4_filename *fname,
1532 		    unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1533 {
1534 	struct ext4_dir_entry_2 * de;
1535 	char * dlimit;
1536 	int de_len;
1537 
1538 	de = (struct ext4_dir_entry_2 *)search_buf;
1539 	dlimit = search_buf + buf_size;
1540 	while ((char *) de < dlimit - EXT4_BASE_DIR_LEN) {
1541 		/* this code is executed quadratically often */
1542 		/* do minimal checking `by hand' */
1543 		if (de->name + de->name_len <= dlimit &&
1544 		    ext4_match(dir, fname, de)) {
1545 			/* found a match - just to be sure, do
1546 			 * a full check */
1547 			if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1548 						 buf_size, offset))
1549 				return -1;
1550 			*res_dir = de;
1551 			return 1;
1552 		}
1553 		/* prevent looping on a bad block */
1554 		de_len = ext4_rec_len_from_disk(de->rec_len,
1555 						dir->i_sb->s_blocksize);
1556 		if (de_len <= 0)
1557 			return -1;
1558 		offset += de_len;
1559 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1560 	}
1561 	return 0;
1562 }
1563 
1564 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1565 			       struct ext4_dir_entry *de)
1566 {
1567 	struct super_block *sb = dir->i_sb;
1568 
1569 	if (!is_dx(dir))
1570 		return 0;
1571 	if (block == 0)
1572 		return 1;
1573 	if (de->inode == 0 &&
1574 	    ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1575 			sb->s_blocksize)
1576 		return 1;
1577 	return 0;
1578 }
1579 
1580 /*
1581  *	__ext4_find_entry()
1582  *
1583  * finds an entry in the specified directory with the wanted name. It
1584  * returns the cache buffer in which the entry was found, and the entry
1585  * itself (as a parameter - res_dir). It does NOT read the inode of the
1586  * entry - you'll have to do that yourself if you want to.
1587  *
1588  * The returned buffer_head has ->b_count elevated.  The caller is expected
1589  * to brelse() it when appropriate.
1590  */
1591 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1592 					     struct ext4_filename *fname,
1593 					     struct ext4_dir_entry_2 **res_dir,
1594 					     int *inlined)
1595 {
1596 	struct super_block *sb;
1597 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1598 	struct buffer_head *bh, *ret = NULL;
1599 	ext4_lblk_t start, block;
1600 	const u8 *name = fname->usr_fname->name;
1601 	size_t ra_max = 0;	/* Number of bh's in the readahead
1602 				   buffer, bh_use[] */
1603 	size_t ra_ptr = 0;	/* Current index into readahead
1604 				   buffer */
1605 	ext4_lblk_t  nblocks;
1606 	int i, namelen, retval;
1607 
1608 	*res_dir = NULL;
1609 	sb = dir->i_sb;
1610 	namelen = fname->usr_fname->len;
1611 	if (namelen > EXT4_NAME_LEN)
1612 		return NULL;
1613 
1614 	if (ext4_has_inline_data(dir)) {
1615 		int has_inline_data = 1;
1616 		ret = ext4_find_inline_entry(dir, fname, res_dir,
1617 					     &has_inline_data);
1618 		if (inlined)
1619 			*inlined = has_inline_data;
1620 		if (has_inline_data)
1621 			goto cleanup_and_exit;
1622 	}
1623 
1624 	if ((namelen <= 2) && (name[0] == '.') &&
1625 	    (name[1] == '.' || name[1] == '\0')) {
1626 		/*
1627 		 * "." or ".." will only be in the first block
1628 		 * NFS may look up ".."; "." should be handled by the VFS
1629 		 */
1630 		block = start = 0;
1631 		nblocks = 1;
1632 		goto restart;
1633 	}
1634 	if (is_dx(dir)) {
1635 		ret = ext4_dx_find_entry(dir, fname, res_dir);
1636 		/*
1637 		 * On success, or if the error was file not found,
1638 		 * return.  Otherwise, fall back to doing a search the
1639 		 * old fashioned way.
1640 		 */
1641 		if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1642 			goto cleanup_and_exit;
1643 		dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1644 			       "falling back\n"));
1645 		ret = NULL;
1646 	}
1647 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1648 	if (!nblocks) {
1649 		ret = NULL;
1650 		goto cleanup_and_exit;
1651 	}
1652 	start = EXT4_I(dir)->i_dir_start_lookup;
1653 	if (start >= nblocks)
1654 		start = 0;
1655 	block = start;
1656 restart:
1657 	do {
1658 		/*
1659 		 * We deal with the read-ahead logic here.
1660 		 */
1661 		cond_resched();
1662 		if (ra_ptr >= ra_max) {
1663 			/* Refill the readahead buffer */
1664 			ra_ptr = 0;
1665 			if (block < start)
1666 				ra_max = start - block;
1667 			else
1668 				ra_max = nblocks - block;
1669 			ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1670 			retval = ext4_bread_batch(dir, block, ra_max,
1671 						  false /* wait */, bh_use);
1672 			if (retval) {
1673 				ret = ERR_PTR(retval);
1674 				ra_max = 0;
1675 				goto cleanup_and_exit;
1676 			}
1677 		}
1678 		if ((bh = bh_use[ra_ptr++]) == NULL)
1679 			goto next;
1680 		wait_on_buffer(bh);
1681 		if (!buffer_uptodate(bh)) {
1682 			EXT4_ERROR_INODE_ERR(dir, EIO,
1683 					     "reading directory lblock %lu",
1684 					     (unsigned long) block);
1685 			brelse(bh);
1686 			ret = ERR_PTR(-EIO);
1687 			goto cleanup_and_exit;
1688 		}
1689 		if (!buffer_verified(bh) &&
1690 		    !is_dx_internal_node(dir, block,
1691 					 (struct ext4_dir_entry *)bh->b_data) &&
1692 		    !ext4_dirblock_csum_verify(dir, bh)) {
1693 			EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1694 					     "checksumming directory "
1695 					     "block %lu", (unsigned long)block);
1696 			brelse(bh);
1697 			ret = ERR_PTR(-EFSBADCRC);
1698 			goto cleanup_and_exit;
1699 		}
1700 		set_buffer_verified(bh);
1701 		i = search_dirblock(bh, dir, fname,
1702 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1703 		if (i == 1) {
1704 			EXT4_I(dir)->i_dir_start_lookup = block;
1705 			ret = bh;
1706 			goto cleanup_and_exit;
1707 		} else {
1708 			brelse(bh);
1709 			if (i < 0)
1710 				goto cleanup_and_exit;
1711 		}
1712 	next:
1713 		if (++block >= nblocks)
1714 			block = 0;
1715 	} while (block != start);
1716 
1717 	/*
1718 	 * If the directory has grown while we were searching, then
1719 	 * search the last part of the directory before giving up.
1720 	 */
1721 	block = nblocks;
1722 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1723 	if (block < nblocks) {
1724 		start = 0;
1725 		goto restart;
1726 	}
1727 
1728 cleanup_and_exit:
1729 	/* Clean up the read-ahead blocks */
1730 	for (; ra_ptr < ra_max; ra_ptr++)
1731 		brelse(bh_use[ra_ptr]);
1732 	return ret;
1733 }
1734 
1735 static struct buffer_head *ext4_find_entry(struct inode *dir,
1736 					   const struct qstr *d_name,
1737 					   struct ext4_dir_entry_2 **res_dir,
1738 					   int *inlined)
1739 {
1740 	int err;
1741 	struct ext4_filename fname;
1742 	struct buffer_head *bh;
1743 
1744 	err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1745 	if (err == -ENOENT)
1746 		return NULL;
1747 	if (err)
1748 		return ERR_PTR(err);
1749 
1750 	bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1751 
1752 	ext4_fname_free_filename(&fname);
1753 	return bh;
1754 }
1755 
1756 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1757 					     struct dentry *dentry,
1758 					     struct ext4_dir_entry_2 **res_dir)
1759 {
1760 	int err;
1761 	struct ext4_filename fname;
1762 	struct buffer_head *bh;
1763 
1764 	err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1765 	if (err == -ENOENT)
1766 		return NULL;
1767 	if (err)
1768 		return ERR_PTR(err);
1769 
1770 	bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1771 
1772 	ext4_fname_free_filename(&fname);
1773 	return bh;
1774 }
1775 
1776 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1777 			struct ext4_filename *fname,
1778 			struct ext4_dir_entry_2 **res_dir)
1779 {
1780 	struct super_block * sb = dir->i_sb;
1781 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1782 	struct buffer_head *bh;
1783 	ext4_lblk_t block;
1784 	int retval;
1785 
1786 #ifdef CONFIG_FS_ENCRYPTION
1787 	*res_dir = NULL;
1788 #endif
1789 	frame = dx_probe(fname, dir, NULL, frames);
1790 	if (IS_ERR(frame))
1791 		return (struct buffer_head *) frame;
1792 	do {
1793 		block = dx_get_block(frame->at);
1794 		bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1795 		if (IS_ERR(bh))
1796 			goto errout;
1797 
1798 		retval = search_dirblock(bh, dir, fname,
1799 					 block << EXT4_BLOCK_SIZE_BITS(sb),
1800 					 res_dir);
1801 		if (retval == 1)
1802 			goto success;
1803 		brelse(bh);
1804 		if (retval == -1) {
1805 			bh = ERR_PTR(ERR_BAD_DX_DIR);
1806 			goto errout;
1807 		}
1808 
1809 		/* Check to see if we should continue to search */
1810 		retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1811 					       frames, NULL);
1812 		if (retval < 0) {
1813 			ext4_warning_inode(dir,
1814 				"error %d reading directory index block",
1815 				retval);
1816 			bh = ERR_PTR(retval);
1817 			goto errout;
1818 		}
1819 	} while (retval == 1);
1820 
1821 	bh = NULL;
1822 errout:
1823 	dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1824 success:
1825 	dx_release(frames);
1826 	return bh;
1827 }
1828 
1829 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1830 {
1831 	struct inode *inode;
1832 	struct ext4_dir_entry_2 *de;
1833 	struct buffer_head *bh;
1834 
1835 	if (dentry->d_name.len > EXT4_NAME_LEN)
1836 		return ERR_PTR(-ENAMETOOLONG);
1837 
1838 	bh = ext4_lookup_entry(dir, dentry, &de);
1839 	if (IS_ERR(bh))
1840 		return ERR_CAST(bh);
1841 	inode = NULL;
1842 	if (bh) {
1843 		__u32 ino = le32_to_cpu(de->inode);
1844 		brelse(bh);
1845 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1846 			EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1847 			return ERR_PTR(-EFSCORRUPTED);
1848 		}
1849 		if (unlikely(ino == dir->i_ino)) {
1850 			EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1851 					 dentry);
1852 			return ERR_PTR(-EFSCORRUPTED);
1853 		}
1854 		inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1855 		if (inode == ERR_PTR(-ESTALE)) {
1856 			EXT4_ERROR_INODE(dir,
1857 					 "deleted inode referenced: %u",
1858 					 ino);
1859 			return ERR_PTR(-EFSCORRUPTED);
1860 		}
1861 		if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1862 		    (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1863 		    !fscrypt_has_permitted_context(dir, inode)) {
1864 			ext4_warning(inode->i_sb,
1865 				     "Inconsistent encryption contexts: %lu/%lu",
1866 				     dir->i_ino, inode->i_ino);
1867 			iput(inode);
1868 			return ERR_PTR(-EPERM);
1869 		}
1870 	}
1871 
1872 #if IS_ENABLED(CONFIG_UNICODE)
1873 	if (!inode && IS_CASEFOLDED(dir)) {
1874 		/* Eventually we want to call d_add_ci(dentry, NULL)
1875 		 * for negative dentries in the encoding case as
1876 		 * well.  For now, prevent the negative dentry
1877 		 * from being cached.
1878 		 */
1879 		return NULL;
1880 	}
1881 #endif
1882 	return d_splice_alias(inode, dentry);
1883 }
1884 
1885 
1886 struct dentry *ext4_get_parent(struct dentry *child)
1887 {
1888 	__u32 ino;
1889 	struct ext4_dir_entry_2 * de;
1890 	struct buffer_head *bh;
1891 
1892 	bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1893 	if (IS_ERR(bh))
1894 		return ERR_CAST(bh);
1895 	if (!bh)
1896 		return ERR_PTR(-ENOENT);
1897 	ino = le32_to_cpu(de->inode);
1898 	brelse(bh);
1899 
1900 	if (!ext4_valid_inum(child->d_sb, ino)) {
1901 		EXT4_ERROR_INODE(d_inode(child),
1902 				 "bad parent inode number: %u", ino);
1903 		return ERR_PTR(-EFSCORRUPTED);
1904 	}
1905 
1906 	return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1907 }
1908 
1909 /*
1910  * Move count entries from end of map between two memory locations.
1911  * Returns pointer to last entry moved.
1912  */
1913 static struct ext4_dir_entry_2 *
1914 dx_move_dirents(struct inode *dir, char *from, char *to,
1915 		struct dx_map_entry *map, int count,
1916 		unsigned blocksize)
1917 {
1918 	unsigned rec_len = 0;
1919 
1920 	while (count--) {
1921 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1922 						(from + (map->offs<<2));
1923 		rec_len = ext4_dir_rec_len(de->name_len, dir);
1924 
1925 		memcpy (to, de, rec_len);
1926 		((struct ext4_dir_entry_2 *) to)->rec_len =
1927 				ext4_rec_len_to_disk(rec_len, blocksize);
1928 
1929 		/* wipe dir_entry excluding the rec_len field */
1930 		de->inode = 0;
1931 		memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1932 								blocksize) -
1933 					 offsetof(struct ext4_dir_entry_2,
1934 								name_len));
1935 
1936 		map++;
1937 		to += rec_len;
1938 	}
1939 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1940 }
1941 
1942 /*
1943  * Compact each dir entry in the range to the minimal rec_len.
1944  * Returns pointer to last entry in range.
1945  */
1946 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1947 							unsigned int blocksize)
1948 {
1949 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1950 	unsigned rec_len = 0;
1951 
1952 	prev = to = de;
1953 	while ((char*)de < base + blocksize) {
1954 		next = ext4_next_entry(de, blocksize);
1955 		if (de->inode && de->name_len) {
1956 			rec_len = ext4_dir_rec_len(de->name_len, dir);
1957 			if (de > to)
1958 				memmove(to, de, rec_len);
1959 			to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1960 			prev = to;
1961 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1962 		}
1963 		de = next;
1964 	}
1965 	return prev;
1966 }
1967 
1968 /*
1969  * Split a full leaf block to make room for a new dir entry.
1970  * Allocate a new block, and move entries so that they are approx. equally full.
1971  * Returns pointer to de in block into which the new entry will be inserted.
1972  */
1973 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1974 			struct buffer_head **bh,struct dx_frame *frame,
1975 			struct dx_hash_info *hinfo)
1976 {
1977 	unsigned blocksize = dir->i_sb->s_blocksize;
1978 	unsigned continued;
1979 	int count;
1980 	struct buffer_head *bh2;
1981 	ext4_lblk_t newblock;
1982 	u32 hash2;
1983 	struct dx_map_entry *map;
1984 	char *data1 = (*bh)->b_data, *data2;
1985 	unsigned split, move, size;
1986 	struct ext4_dir_entry_2 *de = NULL, *de2;
1987 	int	csum_size = 0;
1988 	int	err = 0, i;
1989 
1990 	if (ext4_has_metadata_csum(dir->i_sb))
1991 		csum_size = sizeof(struct ext4_dir_entry_tail);
1992 
1993 	bh2 = ext4_append(handle, dir, &newblock);
1994 	if (IS_ERR(bh2)) {
1995 		brelse(*bh);
1996 		*bh = NULL;
1997 		return (struct ext4_dir_entry_2 *) bh2;
1998 	}
1999 
2000 	BUFFER_TRACE(*bh, "get_write_access");
2001 	err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
2002 					    EXT4_JTR_NONE);
2003 	if (err)
2004 		goto journal_error;
2005 
2006 	BUFFER_TRACE(frame->bh, "get_write_access");
2007 	err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
2008 					    EXT4_JTR_NONE);
2009 	if (err)
2010 		goto journal_error;
2011 
2012 	data2 = bh2->b_data;
2013 
2014 	/* create map in the end of data2 block */
2015 	map = (struct dx_map_entry *) (data2 + blocksize);
2016 	count = dx_make_map(dir, *bh, hinfo, map);
2017 	if (count < 0) {
2018 		err = count;
2019 		goto journal_error;
2020 	}
2021 	map -= count;
2022 	dx_sort_map(map, count);
2023 	/* Ensure that neither split block is over half full */
2024 	size = 0;
2025 	move = 0;
2026 	for (i = count-1; i >= 0; i--) {
2027 		/* is more than half of this entry in 2nd half of the block? */
2028 		if (size + map[i].size/2 > blocksize/2)
2029 			break;
2030 		size += map[i].size;
2031 		move++;
2032 	}
2033 	/*
2034 	 * map index at which we will split
2035 	 *
2036 	 * If the sum of active entries didn't exceed half the block size, just
2037 	 * split it in half by count; each resulting block will have at least
2038 	 * half the space free.
2039 	 */
2040 	if (i > 0)
2041 		split = count - move;
2042 	else
2043 		split = count/2;
2044 
2045 	hash2 = map[split].hash;
2046 	continued = hash2 == map[split - 1].hash;
2047 	dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
2048 			(unsigned long)dx_get_block(frame->at),
2049 					hash2, split, count-split));
2050 
2051 	/* Fancy dance to stay within two buffers */
2052 	de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
2053 			      blocksize);
2054 	de = dx_pack_dirents(dir, data1, blocksize);
2055 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2056 					   (char *) de,
2057 					   blocksize);
2058 	de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2059 					    (char *) de2,
2060 					    blocksize);
2061 	if (csum_size) {
2062 		ext4_initialize_dirent_tail(*bh, blocksize);
2063 		ext4_initialize_dirent_tail(bh2, blocksize);
2064 	}
2065 
2066 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
2067 			blocksize, 1));
2068 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
2069 			blocksize, 1));
2070 
2071 	/* Which block gets the new entry? */
2072 	if (hinfo->hash >= hash2) {
2073 		swap(*bh, bh2);
2074 		de = de2;
2075 	}
2076 	dx_insert_block(frame, hash2 + continued, newblock);
2077 	err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2078 	if (err)
2079 		goto journal_error;
2080 	err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2081 	if (err)
2082 		goto journal_error;
2083 	brelse(bh2);
2084 	dxtrace(dx_show_index("frame", frame->entries));
2085 	return de;
2086 
2087 journal_error:
2088 	brelse(*bh);
2089 	brelse(bh2);
2090 	*bh = NULL;
2091 	ext4_std_error(dir->i_sb, err);
2092 	return ERR_PTR(err);
2093 }
2094 
2095 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2096 		      struct buffer_head *bh,
2097 		      void *buf, int buf_size,
2098 		      struct ext4_filename *fname,
2099 		      struct ext4_dir_entry_2 **dest_de)
2100 {
2101 	struct ext4_dir_entry_2 *de;
2102 	unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2103 	int nlen, rlen;
2104 	unsigned int offset = 0;
2105 	char *top;
2106 
2107 	de = buf;
2108 	top = buf + buf_size - reclen;
2109 	while ((char *) de <= top) {
2110 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2111 					 buf, buf_size, offset))
2112 			return -EFSCORRUPTED;
2113 		if (ext4_match(dir, fname, de))
2114 			return -EEXIST;
2115 		nlen = ext4_dir_rec_len(de->name_len, dir);
2116 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2117 		if ((de->inode ? rlen - nlen : rlen) >= reclen)
2118 			break;
2119 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2120 		offset += rlen;
2121 	}
2122 	if ((char *) de > top)
2123 		return -ENOSPC;
2124 
2125 	*dest_de = de;
2126 	return 0;
2127 }
2128 
2129 void ext4_insert_dentry(struct inode *dir,
2130 			struct inode *inode,
2131 			struct ext4_dir_entry_2 *de,
2132 			int buf_size,
2133 			struct ext4_filename *fname)
2134 {
2135 
2136 	int nlen, rlen;
2137 
2138 	nlen = ext4_dir_rec_len(de->name_len, dir);
2139 	rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2140 	if (de->inode) {
2141 		struct ext4_dir_entry_2 *de1 =
2142 			(struct ext4_dir_entry_2 *)((char *)de + nlen);
2143 		de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2144 		de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2145 		de = de1;
2146 	}
2147 	de->file_type = EXT4_FT_UNKNOWN;
2148 	de->inode = cpu_to_le32(inode->i_ino);
2149 	ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2150 	de->name_len = fname_len(fname);
2151 	memcpy(de->name, fname_name(fname), fname_len(fname));
2152 	if (ext4_hash_in_dirent(dir)) {
2153 		struct dx_hash_info *hinfo = &fname->hinfo;
2154 
2155 		EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2156 		EXT4_DIRENT_HASHES(de)->minor_hash =
2157 						cpu_to_le32(hinfo->minor_hash);
2158 	}
2159 }
2160 
2161 /*
2162  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
2163  * it points to a directory entry which is guaranteed to be large
2164  * enough for new directory entry.  If de is NULL, then
2165  * add_dirent_to_buf will attempt search the directory block for
2166  * space.  It will return -ENOSPC if no space is available, and -EIO
2167  * and -EEXIST if directory entry already exists.
2168  */
2169 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2170 			     struct inode *dir,
2171 			     struct inode *inode, struct ext4_dir_entry_2 *de,
2172 			     struct buffer_head *bh)
2173 {
2174 	unsigned int	blocksize = dir->i_sb->s_blocksize;
2175 	int		csum_size = 0;
2176 	int		err, err2;
2177 
2178 	if (ext4_has_metadata_csum(inode->i_sb))
2179 		csum_size = sizeof(struct ext4_dir_entry_tail);
2180 
2181 	if (!de) {
2182 		err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2183 					blocksize - csum_size, fname, &de);
2184 		if (err)
2185 			return err;
2186 	}
2187 	BUFFER_TRACE(bh, "get_write_access");
2188 	err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2189 					    EXT4_JTR_NONE);
2190 	if (err) {
2191 		ext4_std_error(dir->i_sb, err);
2192 		return err;
2193 	}
2194 
2195 	/* By now the buffer is marked for journaling */
2196 	ext4_insert_dentry(dir, inode, de, blocksize, fname);
2197 
2198 	/*
2199 	 * XXX shouldn't update any times until successful
2200 	 * completion of syscall, but too many callers depend
2201 	 * on this.
2202 	 *
2203 	 * XXX similarly, too many callers depend on
2204 	 * ext4_new_inode() setting the times, but error
2205 	 * recovery deletes the inode, so the worst that can
2206 	 * happen is that the times are slightly out of date
2207 	 * and/or different from the directory change time.
2208 	 */
2209 	inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
2210 	ext4_update_dx_flag(dir);
2211 	inode_inc_iversion(dir);
2212 	err2 = ext4_mark_inode_dirty(handle, dir);
2213 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2214 	err = ext4_handle_dirty_dirblock(handle, dir, bh);
2215 	if (err)
2216 		ext4_std_error(dir->i_sb, err);
2217 	return err ? err : err2;
2218 }
2219 
2220 /*
2221  * This converts a one block unindexed directory to a 3 block indexed
2222  * directory, and adds the dentry to the indexed directory.
2223  */
2224 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2225 			    struct inode *dir,
2226 			    struct inode *inode, struct buffer_head *bh)
2227 {
2228 	struct buffer_head *bh2;
2229 	struct dx_root	*root;
2230 	struct dx_frame	frames[EXT4_HTREE_LEVEL], *frame;
2231 	struct dx_entry *entries;
2232 	struct ext4_dir_entry_2	*de, *de2;
2233 	char		*data2, *top;
2234 	unsigned	len;
2235 	int		retval;
2236 	unsigned	blocksize;
2237 	ext4_lblk_t  block;
2238 	struct fake_dirent *fde;
2239 	int csum_size = 0;
2240 
2241 	if (ext4_has_metadata_csum(inode->i_sb))
2242 		csum_size = sizeof(struct ext4_dir_entry_tail);
2243 
2244 	blocksize =  dir->i_sb->s_blocksize;
2245 	dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2246 	BUFFER_TRACE(bh, "get_write_access");
2247 	retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2248 					       EXT4_JTR_NONE);
2249 	if (retval) {
2250 		ext4_std_error(dir->i_sb, retval);
2251 		brelse(bh);
2252 		return retval;
2253 	}
2254 	root = (struct dx_root *) bh->b_data;
2255 
2256 	/* The 0th block becomes the root, move the dirents out */
2257 	fde = &root->dotdot;
2258 	de = (struct ext4_dir_entry_2 *)((char *)fde +
2259 		ext4_rec_len_from_disk(fde->rec_len, blocksize));
2260 	if ((char *) de >= (((char *) root) + blocksize)) {
2261 		EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2262 		brelse(bh);
2263 		return -EFSCORRUPTED;
2264 	}
2265 	len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2266 
2267 	/* Allocate new block for the 0th block's dirents */
2268 	bh2 = ext4_append(handle, dir, &block);
2269 	if (IS_ERR(bh2)) {
2270 		brelse(bh);
2271 		return PTR_ERR(bh2);
2272 	}
2273 	ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2274 	data2 = bh2->b_data;
2275 
2276 	memcpy(data2, de, len);
2277 	memset(de, 0, len); /* wipe old data */
2278 	de = (struct ext4_dir_entry_2 *) data2;
2279 	top = data2 + len;
2280 	while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) {
2281 		if (ext4_check_dir_entry(dir, NULL, de, bh2, data2, len,
2282 					(char *)de - data2)) {
2283 			brelse(bh2);
2284 			brelse(bh);
2285 			return -EFSCORRUPTED;
2286 		}
2287 		de = de2;
2288 	}
2289 	de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2290 					   (char *) de, blocksize);
2291 
2292 	if (csum_size)
2293 		ext4_initialize_dirent_tail(bh2, blocksize);
2294 
2295 	/* Initialize the root; the dot dirents already exist */
2296 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2297 	de->rec_len = ext4_rec_len_to_disk(
2298 			blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2299 	memset (&root->info, 0, sizeof(root->info));
2300 	root->info.info_length = sizeof(root->info);
2301 	if (ext4_hash_in_dirent(dir))
2302 		root->info.hash_version = DX_HASH_SIPHASH;
2303 	else
2304 		root->info.hash_version =
2305 				EXT4_SB(dir->i_sb)->s_def_hash_version;
2306 
2307 	entries = root->entries;
2308 	dx_set_block(entries, 1);
2309 	dx_set_count(entries, 1);
2310 	dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2311 
2312 	/* Initialize as for dx_probe */
2313 	fname->hinfo.hash_version = root->info.hash_version;
2314 	if (fname->hinfo.hash_version <= DX_HASH_TEA)
2315 		fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2316 	fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2317 
2318 	/* casefolded encrypted hashes are computed on fname setup */
2319 	if (!ext4_hash_in_dirent(dir)) {
2320 		int err = ext4fs_dirhash(dir, fname_name(fname),
2321 					 fname_len(fname), &fname->hinfo);
2322 		if (err < 0) {
2323 			brelse(bh2);
2324 			brelse(bh);
2325 			return err;
2326 		}
2327 	}
2328 	memset(frames, 0, sizeof(frames));
2329 	frame = frames;
2330 	frame->entries = entries;
2331 	frame->at = entries;
2332 	frame->bh = bh;
2333 
2334 	retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2335 	if (retval)
2336 		goto out_frames;
2337 	retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2338 	if (retval)
2339 		goto out_frames;
2340 
2341 	de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2342 	if (IS_ERR(de)) {
2343 		retval = PTR_ERR(de);
2344 		goto out_frames;
2345 	}
2346 
2347 	retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2348 out_frames:
2349 	/*
2350 	 * Even if the block split failed, we have to properly write
2351 	 * out all the changes we did so far. Otherwise we can end up
2352 	 * with corrupted filesystem.
2353 	 */
2354 	if (retval)
2355 		ext4_mark_inode_dirty(handle, dir);
2356 	dx_release(frames);
2357 	brelse(bh2);
2358 	return retval;
2359 }
2360 
2361 /*
2362  *	ext4_add_entry()
2363  *
2364  * adds a file entry to the specified directory, using the same
2365  * semantics as ext4_find_entry(). It returns NULL if it failed.
2366  *
2367  * NOTE!! The inode part of 'de' is left at 0 - which means you
2368  * may not sleep between calling this and putting something into
2369  * the entry, as someone else might have used it while you slept.
2370  */
2371 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2372 			  struct inode *inode)
2373 {
2374 	struct inode *dir = d_inode(dentry->d_parent);
2375 	struct buffer_head *bh = NULL;
2376 	struct ext4_dir_entry_2 *de;
2377 	struct super_block *sb;
2378 	struct ext4_filename fname;
2379 	int	retval;
2380 	int	dx_fallback=0;
2381 	unsigned blocksize;
2382 	ext4_lblk_t block, blocks;
2383 	int	csum_size = 0;
2384 
2385 	if (ext4_has_metadata_csum(inode->i_sb))
2386 		csum_size = sizeof(struct ext4_dir_entry_tail);
2387 
2388 	sb = dir->i_sb;
2389 	blocksize = sb->s_blocksize;
2390 
2391 	if (fscrypt_is_nokey_name(dentry))
2392 		return -ENOKEY;
2393 
2394 #if IS_ENABLED(CONFIG_UNICODE)
2395 	if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2396 	    utf8_validate(sb->s_encoding, &dentry->d_name))
2397 		return -EINVAL;
2398 #endif
2399 
2400 	retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2401 	if (retval)
2402 		return retval;
2403 
2404 	if (ext4_has_inline_data(dir)) {
2405 		retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2406 		if (retval < 0)
2407 			goto out;
2408 		if (retval == 1) {
2409 			retval = 0;
2410 			goto out;
2411 		}
2412 	}
2413 
2414 	if (is_dx(dir)) {
2415 		retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2416 		if (!retval || (retval != ERR_BAD_DX_DIR))
2417 			goto out;
2418 		/* Can we just ignore htree data? */
2419 		if (ext4_has_metadata_csum(sb)) {
2420 			EXT4_ERROR_INODE(dir,
2421 				"Directory has corrupted htree index.");
2422 			retval = -EFSCORRUPTED;
2423 			goto out;
2424 		}
2425 		ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2426 		dx_fallback++;
2427 		retval = ext4_mark_inode_dirty(handle, dir);
2428 		if (unlikely(retval))
2429 			goto out;
2430 	}
2431 	blocks = dir->i_size >> sb->s_blocksize_bits;
2432 	for (block = 0; block < blocks; block++) {
2433 		bh = ext4_read_dirblock(dir, block, DIRENT);
2434 		if (bh == NULL) {
2435 			bh = ext4_bread(handle, dir, block,
2436 					EXT4_GET_BLOCKS_CREATE);
2437 			goto add_to_new_block;
2438 		}
2439 		if (IS_ERR(bh)) {
2440 			retval = PTR_ERR(bh);
2441 			bh = NULL;
2442 			goto out;
2443 		}
2444 		retval = add_dirent_to_buf(handle, &fname, dir, inode,
2445 					   NULL, bh);
2446 		if (retval != -ENOSPC)
2447 			goto out;
2448 
2449 		if (blocks == 1 && !dx_fallback &&
2450 		    ext4_has_feature_dir_index(sb)) {
2451 			retval = make_indexed_dir(handle, &fname, dir,
2452 						  inode, bh);
2453 			bh = NULL; /* make_indexed_dir releases bh */
2454 			goto out;
2455 		}
2456 		brelse(bh);
2457 	}
2458 	bh = ext4_append(handle, dir, &block);
2459 add_to_new_block:
2460 	if (IS_ERR(bh)) {
2461 		retval = PTR_ERR(bh);
2462 		bh = NULL;
2463 		goto out;
2464 	}
2465 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2466 	de->inode = 0;
2467 	de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2468 
2469 	if (csum_size)
2470 		ext4_initialize_dirent_tail(bh, blocksize);
2471 
2472 	retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2473 out:
2474 	ext4_fname_free_filename(&fname);
2475 	brelse(bh);
2476 	if (retval == 0)
2477 		ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2478 	return retval;
2479 }
2480 
2481 /*
2482  * Returns 0 for success, or a negative error value
2483  */
2484 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2485 			     struct inode *dir, struct inode *inode)
2486 {
2487 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2488 	struct dx_entry *entries, *at;
2489 	struct buffer_head *bh;
2490 	struct super_block *sb = dir->i_sb;
2491 	struct ext4_dir_entry_2 *de;
2492 	int restart;
2493 	int err;
2494 
2495 again:
2496 	restart = 0;
2497 	frame = dx_probe(fname, dir, NULL, frames);
2498 	if (IS_ERR(frame))
2499 		return PTR_ERR(frame);
2500 	entries = frame->entries;
2501 	at = frame->at;
2502 	bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2503 	if (IS_ERR(bh)) {
2504 		err = PTR_ERR(bh);
2505 		bh = NULL;
2506 		goto cleanup;
2507 	}
2508 
2509 	BUFFER_TRACE(bh, "get_write_access");
2510 	err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
2511 	if (err)
2512 		goto journal_error;
2513 
2514 	err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2515 	if (err != -ENOSPC)
2516 		goto cleanup;
2517 
2518 	err = 0;
2519 	/* Block full, should compress but for now just split */
2520 	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2521 		       dx_get_count(entries), dx_get_limit(entries)));
2522 	/* Need to split index? */
2523 	if (dx_get_count(entries) == dx_get_limit(entries)) {
2524 		ext4_lblk_t newblock;
2525 		int levels = frame - frames + 1;
2526 		unsigned int icount;
2527 		int add_level = 1;
2528 		struct dx_entry *entries2;
2529 		struct dx_node *node2;
2530 		struct buffer_head *bh2;
2531 
2532 		while (frame > frames) {
2533 			if (dx_get_count((frame - 1)->entries) <
2534 			    dx_get_limit((frame - 1)->entries)) {
2535 				add_level = 0;
2536 				break;
2537 			}
2538 			frame--; /* split higher index block */
2539 			at = frame->at;
2540 			entries = frame->entries;
2541 			restart = 1;
2542 		}
2543 		if (add_level && levels == ext4_dir_htree_level(sb)) {
2544 			ext4_warning(sb, "Directory (ino: %lu) index full, "
2545 					 "reach max htree level :%d",
2546 					 dir->i_ino, levels);
2547 			if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2548 				ext4_warning(sb, "Large directory feature is "
2549 						 "not enabled on this "
2550 						 "filesystem");
2551 			}
2552 			err = -ENOSPC;
2553 			goto cleanup;
2554 		}
2555 		icount = dx_get_count(entries);
2556 		bh2 = ext4_append(handle, dir, &newblock);
2557 		if (IS_ERR(bh2)) {
2558 			err = PTR_ERR(bh2);
2559 			goto cleanup;
2560 		}
2561 		node2 = (struct dx_node *)(bh2->b_data);
2562 		entries2 = node2->entries;
2563 		memset(&node2->fake, 0, sizeof(struct fake_dirent));
2564 		node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2565 							   sb->s_blocksize);
2566 		BUFFER_TRACE(frame->bh, "get_write_access");
2567 		err = ext4_journal_get_write_access(handle, sb, frame->bh,
2568 						    EXT4_JTR_NONE);
2569 		if (err)
2570 			goto journal_error;
2571 		if (!add_level) {
2572 			unsigned icount1 = icount/2, icount2 = icount - icount1;
2573 			unsigned hash2 = dx_get_hash(entries + icount1);
2574 			dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2575 				       icount1, icount2));
2576 
2577 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2578 			err = ext4_journal_get_write_access(handle, sb,
2579 							    (frame - 1)->bh,
2580 							    EXT4_JTR_NONE);
2581 			if (err)
2582 				goto journal_error;
2583 
2584 			memcpy((char *) entries2, (char *) (entries + icount1),
2585 			       icount2 * sizeof(struct dx_entry));
2586 			dx_set_count(entries, icount1);
2587 			dx_set_count(entries2, icount2);
2588 			dx_set_limit(entries2, dx_node_limit(dir));
2589 
2590 			/* Which index block gets the new entry? */
2591 			if (at - entries >= icount1) {
2592 				frame->at = at - entries - icount1 + entries2;
2593 				frame->entries = entries = entries2;
2594 				swap(frame->bh, bh2);
2595 			}
2596 			dx_insert_block((frame - 1), hash2, newblock);
2597 			dxtrace(dx_show_index("node", frame->entries));
2598 			dxtrace(dx_show_index("node",
2599 			       ((struct dx_node *) bh2->b_data)->entries));
2600 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2601 			if (err)
2602 				goto journal_error;
2603 			brelse (bh2);
2604 			err = ext4_handle_dirty_dx_node(handle, dir,
2605 						   (frame - 1)->bh);
2606 			if (err)
2607 				goto journal_error;
2608 			err = ext4_handle_dirty_dx_node(handle, dir,
2609 							frame->bh);
2610 			if (restart || err)
2611 				goto journal_error;
2612 		} else {
2613 			struct dx_root *dxroot;
2614 			memcpy((char *) entries2, (char *) entries,
2615 			       icount * sizeof(struct dx_entry));
2616 			dx_set_limit(entries2, dx_node_limit(dir));
2617 
2618 			/* Set up root */
2619 			dx_set_count(entries, 1);
2620 			dx_set_block(entries + 0, newblock);
2621 			dxroot = (struct dx_root *)frames[0].bh->b_data;
2622 			dxroot->info.indirect_levels += 1;
2623 			dxtrace(printk(KERN_DEBUG
2624 				       "Creating %d level index...\n",
2625 				       dxroot->info.indirect_levels));
2626 			err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2627 			if (err)
2628 				goto journal_error;
2629 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2630 			brelse(bh2);
2631 			restart = 1;
2632 			goto journal_error;
2633 		}
2634 	}
2635 	de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2636 	if (IS_ERR(de)) {
2637 		err = PTR_ERR(de);
2638 		goto cleanup;
2639 	}
2640 	err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2641 	goto cleanup;
2642 
2643 journal_error:
2644 	ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2645 cleanup:
2646 	brelse(bh);
2647 	dx_release(frames);
2648 	/* @restart is true means htree-path has been changed, we need to
2649 	 * repeat dx_probe() to find out valid htree-path
2650 	 */
2651 	if (restart && err == 0)
2652 		goto again;
2653 	return err;
2654 }
2655 
2656 /*
2657  * ext4_generic_delete_entry deletes a directory entry by merging it
2658  * with the previous entry
2659  */
2660 int ext4_generic_delete_entry(struct inode *dir,
2661 			      struct ext4_dir_entry_2 *de_del,
2662 			      struct buffer_head *bh,
2663 			      void *entry_buf,
2664 			      int buf_size,
2665 			      int csum_size)
2666 {
2667 	struct ext4_dir_entry_2 *de, *pde;
2668 	unsigned int blocksize = dir->i_sb->s_blocksize;
2669 	int i;
2670 
2671 	i = 0;
2672 	pde = NULL;
2673 	de = entry_buf;
2674 	while (i < buf_size - csum_size) {
2675 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2676 					 entry_buf, buf_size, i))
2677 			return -EFSCORRUPTED;
2678 		if (de == de_del)  {
2679 			if (pde) {
2680 				pde->rec_len = ext4_rec_len_to_disk(
2681 					ext4_rec_len_from_disk(pde->rec_len,
2682 							       blocksize) +
2683 					ext4_rec_len_from_disk(de->rec_len,
2684 							       blocksize),
2685 					blocksize);
2686 
2687 				/* wipe entire dir_entry */
2688 				memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2689 								blocksize));
2690 			} else {
2691 				/* wipe dir_entry excluding the rec_len field */
2692 				de->inode = 0;
2693 				memset(&de->name_len, 0,
2694 					ext4_rec_len_from_disk(de->rec_len,
2695 								blocksize) -
2696 					offsetof(struct ext4_dir_entry_2,
2697 								name_len));
2698 			}
2699 
2700 			inode_inc_iversion(dir);
2701 			return 0;
2702 		}
2703 		i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2704 		pde = de;
2705 		de = ext4_next_entry(de, blocksize);
2706 	}
2707 	return -ENOENT;
2708 }
2709 
2710 static int ext4_delete_entry(handle_t *handle,
2711 			     struct inode *dir,
2712 			     struct ext4_dir_entry_2 *de_del,
2713 			     struct buffer_head *bh)
2714 {
2715 	int err, csum_size = 0;
2716 
2717 	if (ext4_has_inline_data(dir)) {
2718 		int has_inline_data = 1;
2719 		err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2720 					       &has_inline_data);
2721 		if (has_inline_data)
2722 			return err;
2723 	}
2724 
2725 	if (ext4_has_metadata_csum(dir->i_sb))
2726 		csum_size = sizeof(struct ext4_dir_entry_tail);
2727 
2728 	BUFFER_TRACE(bh, "get_write_access");
2729 	err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2730 					    EXT4_JTR_NONE);
2731 	if (unlikely(err))
2732 		goto out;
2733 
2734 	err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2735 					dir->i_sb->s_blocksize, csum_size);
2736 	if (err)
2737 		goto out;
2738 
2739 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2740 	err = ext4_handle_dirty_dirblock(handle, dir, bh);
2741 	if (unlikely(err))
2742 		goto out;
2743 
2744 	return 0;
2745 out:
2746 	if (err != -ENOENT)
2747 		ext4_std_error(dir->i_sb, err);
2748 	return err;
2749 }
2750 
2751 /*
2752  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2753  * since this indicates that nlinks count was previously 1 to avoid overflowing
2754  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2755  * that subdirectory link counts are not being maintained accurately.
2756  *
2757  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2758  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2759  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2760  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2761  */
2762 static void ext4_inc_count(struct inode *inode)
2763 {
2764 	inc_nlink(inode);
2765 	if (is_dx(inode) &&
2766 	    (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2767 		set_nlink(inode, 1);
2768 }
2769 
2770 /*
2771  * If a directory had nlink == 1, then we should let it be 1. This indicates
2772  * directory has >EXT4_LINK_MAX subdirs.
2773  */
2774 static void ext4_dec_count(struct inode *inode)
2775 {
2776 	if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2777 		drop_nlink(inode);
2778 }
2779 
2780 
2781 /*
2782  * Add non-directory inode to a directory. On success, the inode reference is
2783  * consumed by dentry is instantiation. This is also indicated by clearing of
2784  * *inodep pointer. On failure, the caller is responsible for dropping the
2785  * inode reference in the safe context.
2786  */
2787 static int ext4_add_nondir(handle_t *handle,
2788 		struct dentry *dentry, struct inode **inodep)
2789 {
2790 	struct inode *dir = d_inode(dentry->d_parent);
2791 	struct inode *inode = *inodep;
2792 	int err = ext4_add_entry(handle, dentry, inode);
2793 	if (!err) {
2794 		err = ext4_mark_inode_dirty(handle, inode);
2795 		if (IS_DIRSYNC(dir))
2796 			ext4_handle_sync(handle);
2797 		d_instantiate_new(dentry, inode);
2798 		*inodep = NULL;
2799 		return err;
2800 	}
2801 	drop_nlink(inode);
2802 	ext4_mark_inode_dirty(handle, inode);
2803 	ext4_orphan_add(handle, inode);
2804 	unlock_new_inode(inode);
2805 	return err;
2806 }
2807 
2808 /*
2809  * By the time this is called, we already have created
2810  * the directory cache entry for the new file, but it
2811  * is so far negative - it has no inode.
2812  *
2813  * If the create succeeds, we fill in the inode information
2814  * with d_instantiate().
2815  */
2816 static int ext4_create(struct mnt_idmap *idmap, struct inode *dir,
2817 		       struct dentry *dentry, umode_t mode, bool excl)
2818 {
2819 	handle_t *handle;
2820 	struct inode *inode;
2821 	int err, credits, retries = 0;
2822 
2823 	err = dquot_initialize(dir);
2824 	if (err)
2825 		return err;
2826 
2827 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2828 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2829 retry:
2830 	inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2831 					    0, NULL, EXT4_HT_DIR, credits);
2832 	handle = ext4_journal_current_handle();
2833 	err = PTR_ERR(inode);
2834 	if (!IS_ERR(inode)) {
2835 		inode->i_op = &ext4_file_inode_operations;
2836 		inode->i_fop = &ext4_file_operations;
2837 		ext4_set_aops(inode);
2838 		err = ext4_add_nondir(handle, dentry, &inode);
2839 		if (!err)
2840 			ext4_fc_track_create(handle, dentry);
2841 	}
2842 	if (handle)
2843 		ext4_journal_stop(handle);
2844 	if (!IS_ERR_OR_NULL(inode))
2845 		iput(inode);
2846 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2847 		goto retry;
2848 	return err;
2849 }
2850 
2851 static int ext4_mknod(struct mnt_idmap *idmap, struct inode *dir,
2852 		      struct dentry *dentry, umode_t mode, dev_t rdev)
2853 {
2854 	handle_t *handle;
2855 	struct inode *inode;
2856 	int err, credits, retries = 0;
2857 
2858 	err = dquot_initialize(dir);
2859 	if (err)
2860 		return err;
2861 
2862 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2863 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2864 retry:
2865 	inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2866 					    0, NULL, EXT4_HT_DIR, credits);
2867 	handle = ext4_journal_current_handle();
2868 	err = PTR_ERR(inode);
2869 	if (!IS_ERR(inode)) {
2870 		init_special_inode(inode, inode->i_mode, rdev);
2871 		inode->i_op = &ext4_special_inode_operations;
2872 		err = ext4_add_nondir(handle, dentry, &inode);
2873 		if (!err)
2874 			ext4_fc_track_create(handle, dentry);
2875 	}
2876 	if (handle)
2877 		ext4_journal_stop(handle);
2878 	if (!IS_ERR_OR_NULL(inode))
2879 		iput(inode);
2880 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2881 		goto retry;
2882 	return err;
2883 }
2884 
2885 static int ext4_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
2886 			struct file *file, umode_t mode)
2887 {
2888 	handle_t *handle;
2889 	struct inode *inode;
2890 	int err, retries = 0;
2891 
2892 	err = dquot_initialize(dir);
2893 	if (err)
2894 		return err;
2895 
2896 retry:
2897 	inode = ext4_new_inode_start_handle(idmap, dir, mode,
2898 					    NULL, 0, NULL,
2899 					    EXT4_HT_DIR,
2900 			EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2901 			  4 + EXT4_XATTR_TRANS_BLOCKS);
2902 	handle = ext4_journal_current_handle();
2903 	err = PTR_ERR(inode);
2904 	if (!IS_ERR(inode)) {
2905 		inode->i_op = &ext4_file_inode_operations;
2906 		inode->i_fop = &ext4_file_operations;
2907 		ext4_set_aops(inode);
2908 		d_tmpfile(file, inode);
2909 		err = ext4_orphan_add(handle, inode);
2910 		if (err)
2911 			goto err_unlock_inode;
2912 		mark_inode_dirty(inode);
2913 		unlock_new_inode(inode);
2914 	}
2915 	if (handle)
2916 		ext4_journal_stop(handle);
2917 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2918 		goto retry;
2919 	return finish_open_simple(file, err);
2920 err_unlock_inode:
2921 	ext4_journal_stop(handle);
2922 	unlock_new_inode(inode);
2923 	return err;
2924 }
2925 
2926 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2927 			  struct ext4_dir_entry_2 *de,
2928 			  int blocksize, int csum_size,
2929 			  unsigned int parent_ino, int dotdot_real_len)
2930 {
2931 	de->inode = cpu_to_le32(inode->i_ino);
2932 	de->name_len = 1;
2933 	de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2934 					   blocksize);
2935 	strcpy(de->name, ".");
2936 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2937 
2938 	de = ext4_next_entry(de, blocksize);
2939 	de->inode = cpu_to_le32(parent_ino);
2940 	de->name_len = 2;
2941 	if (!dotdot_real_len)
2942 		de->rec_len = ext4_rec_len_to_disk(blocksize -
2943 					(csum_size + ext4_dir_rec_len(1, NULL)),
2944 					blocksize);
2945 	else
2946 		de->rec_len = ext4_rec_len_to_disk(
2947 					ext4_dir_rec_len(de->name_len, NULL),
2948 					blocksize);
2949 	strcpy(de->name, "..");
2950 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2951 
2952 	return ext4_next_entry(de, blocksize);
2953 }
2954 
2955 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2956 			     struct inode *inode)
2957 {
2958 	struct buffer_head *dir_block = NULL;
2959 	struct ext4_dir_entry_2 *de;
2960 	ext4_lblk_t block = 0;
2961 	unsigned int blocksize = dir->i_sb->s_blocksize;
2962 	int csum_size = 0;
2963 	int err;
2964 
2965 	if (ext4_has_metadata_csum(dir->i_sb))
2966 		csum_size = sizeof(struct ext4_dir_entry_tail);
2967 
2968 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2969 		err = ext4_try_create_inline_dir(handle, dir, inode);
2970 		if (err < 0 && err != -ENOSPC)
2971 			goto out;
2972 		if (!err)
2973 			goto out;
2974 	}
2975 
2976 	inode->i_size = 0;
2977 	dir_block = ext4_append(handle, inode, &block);
2978 	if (IS_ERR(dir_block))
2979 		return PTR_ERR(dir_block);
2980 	de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2981 	ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2982 	set_nlink(inode, 2);
2983 	if (csum_size)
2984 		ext4_initialize_dirent_tail(dir_block, blocksize);
2985 
2986 	BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2987 	err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2988 	if (err)
2989 		goto out;
2990 	set_buffer_verified(dir_block);
2991 out:
2992 	brelse(dir_block);
2993 	return err;
2994 }
2995 
2996 static int ext4_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2997 		      struct dentry *dentry, umode_t mode)
2998 {
2999 	handle_t *handle;
3000 	struct inode *inode;
3001 	int err, err2 = 0, credits, retries = 0;
3002 
3003 	if (EXT4_DIR_LINK_MAX(dir))
3004 		return -EMLINK;
3005 
3006 	err = dquot_initialize(dir);
3007 	if (err)
3008 		return err;
3009 
3010 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3011 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
3012 retry:
3013 	inode = ext4_new_inode_start_handle(idmap, dir, S_IFDIR | mode,
3014 					    &dentry->d_name,
3015 					    0, NULL, EXT4_HT_DIR, credits);
3016 	handle = ext4_journal_current_handle();
3017 	err = PTR_ERR(inode);
3018 	if (IS_ERR(inode))
3019 		goto out_stop;
3020 
3021 	inode->i_op = &ext4_dir_inode_operations;
3022 	inode->i_fop = &ext4_dir_operations;
3023 	err = ext4_init_new_dir(handle, dir, inode);
3024 	if (err)
3025 		goto out_clear_inode;
3026 	err = ext4_mark_inode_dirty(handle, inode);
3027 	if (!err)
3028 		err = ext4_add_entry(handle, dentry, inode);
3029 	if (err) {
3030 out_clear_inode:
3031 		clear_nlink(inode);
3032 		ext4_orphan_add(handle, inode);
3033 		unlock_new_inode(inode);
3034 		err2 = ext4_mark_inode_dirty(handle, inode);
3035 		if (unlikely(err2))
3036 			err = err2;
3037 		ext4_journal_stop(handle);
3038 		iput(inode);
3039 		goto out_retry;
3040 	}
3041 	ext4_inc_count(dir);
3042 
3043 	ext4_update_dx_flag(dir);
3044 	err = ext4_mark_inode_dirty(handle, dir);
3045 	if (err)
3046 		goto out_clear_inode;
3047 	d_instantiate_new(dentry, inode);
3048 	ext4_fc_track_create(handle, dentry);
3049 	if (IS_DIRSYNC(dir))
3050 		ext4_handle_sync(handle);
3051 
3052 out_stop:
3053 	if (handle)
3054 		ext4_journal_stop(handle);
3055 out_retry:
3056 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3057 		goto retry;
3058 	return err;
3059 }
3060 
3061 /*
3062  * routine to check that the specified directory is empty (for rmdir)
3063  */
3064 bool ext4_empty_dir(struct inode *inode)
3065 {
3066 	unsigned int offset;
3067 	struct buffer_head *bh;
3068 	struct ext4_dir_entry_2 *de;
3069 	struct super_block *sb;
3070 
3071 	if (ext4_has_inline_data(inode)) {
3072 		int has_inline_data = 1;
3073 		int ret;
3074 
3075 		ret = empty_inline_dir(inode, &has_inline_data);
3076 		if (has_inline_data)
3077 			return ret;
3078 	}
3079 
3080 	sb = inode->i_sb;
3081 	if (inode->i_size < ext4_dir_rec_len(1, NULL) +
3082 					ext4_dir_rec_len(2, NULL)) {
3083 		EXT4_ERROR_INODE(inode, "invalid size");
3084 		return false;
3085 	}
3086 	/* The first directory block must not be a hole,
3087 	 * so treat it as DIRENT_HTREE
3088 	 */
3089 	bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3090 	if (IS_ERR(bh))
3091 		return false;
3092 
3093 	de = (struct ext4_dir_entry_2 *) bh->b_data;
3094 	if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3095 				 0) ||
3096 	    le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3097 		ext4_warning_inode(inode, "directory missing '.'");
3098 		brelse(bh);
3099 		return false;
3100 	}
3101 	offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3102 	de = ext4_next_entry(de, sb->s_blocksize);
3103 	if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3104 				 offset) ||
3105 	    le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3106 		ext4_warning_inode(inode, "directory missing '..'");
3107 		brelse(bh);
3108 		return false;
3109 	}
3110 	offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3111 	while (offset < inode->i_size) {
3112 		if (!(offset & (sb->s_blocksize - 1))) {
3113 			unsigned int lblock;
3114 			brelse(bh);
3115 			lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3116 			bh = ext4_read_dirblock(inode, lblock, EITHER);
3117 			if (bh == NULL) {
3118 				offset += sb->s_blocksize;
3119 				continue;
3120 			}
3121 			if (IS_ERR(bh))
3122 				return false;
3123 		}
3124 		de = (struct ext4_dir_entry_2 *) (bh->b_data +
3125 					(offset & (sb->s_blocksize - 1)));
3126 		if (ext4_check_dir_entry(inode, NULL, de, bh,
3127 					 bh->b_data, bh->b_size, offset) ||
3128 		    le32_to_cpu(de->inode)) {
3129 			brelse(bh);
3130 			return false;
3131 		}
3132 		offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3133 	}
3134 	brelse(bh);
3135 	return true;
3136 }
3137 
3138 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3139 {
3140 	int retval;
3141 	struct inode *inode;
3142 	struct buffer_head *bh;
3143 	struct ext4_dir_entry_2 *de;
3144 	handle_t *handle = NULL;
3145 
3146 	if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3147 		return -EIO;
3148 
3149 	/* Initialize quotas before so that eventual writes go in
3150 	 * separate transaction */
3151 	retval = dquot_initialize(dir);
3152 	if (retval)
3153 		return retval;
3154 	retval = dquot_initialize(d_inode(dentry));
3155 	if (retval)
3156 		return retval;
3157 
3158 	retval = -ENOENT;
3159 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3160 	if (IS_ERR(bh))
3161 		return PTR_ERR(bh);
3162 	if (!bh)
3163 		goto end_rmdir;
3164 
3165 	inode = d_inode(dentry);
3166 
3167 	retval = -EFSCORRUPTED;
3168 	if (le32_to_cpu(de->inode) != inode->i_ino)
3169 		goto end_rmdir;
3170 
3171 	retval = -ENOTEMPTY;
3172 	if (!ext4_empty_dir(inode))
3173 		goto end_rmdir;
3174 
3175 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3176 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3177 	if (IS_ERR(handle)) {
3178 		retval = PTR_ERR(handle);
3179 		handle = NULL;
3180 		goto end_rmdir;
3181 	}
3182 
3183 	if (IS_DIRSYNC(dir))
3184 		ext4_handle_sync(handle);
3185 
3186 	retval = ext4_delete_entry(handle, dir, de, bh);
3187 	if (retval)
3188 		goto end_rmdir;
3189 	if (!EXT4_DIR_LINK_EMPTY(inode))
3190 		ext4_warning_inode(inode,
3191 			     "empty directory '%.*s' has too many links (%u)",
3192 			     dentry->d_name.len, dentry->d_name.name,
3193 			     inode->i_nlink);
3194 	inode_inc_iversion(inode);
3195 	clear_nlink(inode);
3196 	/* There's no need to set i_disksize: the fact that i_nlink is
3197 	 * zero will ensure that the right thing happens during any
3198 	 * recovery. */
3199 	inode->i_size = 0;
3200 	ext4_orphan_add(handle, inode);
3201 	inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3202 	inode_set_ctime_current(inode);
3203 	retval = ext4_mark_inode_dirty(handle, inode);
3204 	if (retval)
3205 		goto end_rmdir;
3206 	ext4_dec_count(dir);
3207 	ext4_update_dx_flag(dir);
3208 	ext4_fc_track_unlink(handle, dentry);
3209 	retval = ext4_mark_inode_dirty(handle, dir);
3210 
3211 #if IS_ENABLED(CONFIG_UNICODE)
3212 	/* VFS negative dentries are incompatible with Encoding and
3213 	 * Case-insensitiveness. Eventually we'll want avoid
3214 	 * invalidating the dentries here, alongside with returning the
3215 	 * negative dentries at ext4_lookup(), when it is better
3216 	 * supported by the VFS for the CI case.
3217 	 */
3218 	if (IS_CASEFOLDED(dir))
3219 		d_invalidate(dentry);
3220 #endif
3221 
3222 end_rmdir:
3223 	brelse(bh);
3224 	if (handle)
3225 		ext4_journal_stop(handle);
3226 	return retval;
3227 }
3228 
3229 int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
3230 		  struct inode *inode,
3231 		  struct dentry *dentry /* NULL during fast_commit recovery */)
3232 {
3233 	int retval = -ENOENT;
3234 	struct buffer_head *bh;
3235 	struct ext4_dir_entry_2 *de;
3236 	handle_t *handle;
3237 	int skip_remove_dentry = 0;
3238 
3239 	/*
3240 	 * Keep this outside the transaction; it may have to set up the
3241 	 * directory's encryption key, which isn't GFP_NOFS-safe.
3242 	 */
3243 	bh = ext4_find_entry(dir, d_name, &de, NULL);
3244 	if (IS_ERR(bh))
3245 		return PTR_ERR(bh);
3246 
3247 	if (!bh)
3248 		return -ENOENT;
3249 
3250 	if (le32_to_cpu(de->inode) != inode->i_ino) {
3251 		/*
3252 		 * It's okay if we find dont find dentry which matches
3253 		 * the inode. That's because it might have gotten
3254 		 * renamed to a different inode number
3255 		 */
3256 		if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3257 			skip_remove_dentry = 1;
3258 		else
3259 			goto out_bh;
3260 	}
3261 
3262 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3263 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3264 	if (IS_ERR(handle)) {
3265 		retval = PTR_ERR(handle);
3266 		goto out_bh;
3267 	}
3268 
3269 	if (IS_DIRSYNC(dir))
3270 		ext4_handle_sync(handle);
3271 
3272 	if (!skip_remove_dentry) {
3273 		retval = ext4_delete_entry(handle, dir, de, bh);
3274 		if (retval)
3275 			goto out_handle;
3276 		inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3277 		ext4_update_dx_flag(dir);
3278 		retval = ext4_mark_inode_dirty(handle, dir);
3279 		if (retval)
3280 			goto out_handle;
3281 	} else {
3282 		retval = 0;
3283 	}
3284 	if (inode->i_nlink == 0)
3285 		ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3286 				   d_name->len, d_name->name);
3287 	else
3288 		drop_nlink(inode);
3289 	if (!inode->i_nlink)
3290 		ext4_orphan_add(handle, inode);
3291 	inode_set_ctime_current(inode);
3292 	retval = ext4_mark_inode_dirty(handle, inode);
3293 	if (dentry && !retval)
3294 		ext4_fc_track_unlink(handle, dentry);
3295 out_handle:
3296 	ext4_journal_stop(handle);
3297 out_bh:
3298 	brelse(bh);
3299 	return retval;
3300 }
3301 
3302 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3303 {
3304 	int retval;
3305 
3306 	if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3307 		return -EIO;
3308 
3309 	trace_ext4_unlink_enter(dir, dentry);
3310 	/*
3311 	 * Initialize quotas before so that eventual writes go
3312 	 * in separate transaction
3313 	 */
3314 	retval = dquot_initialize(dir);
3315 	if (retval)
3316 		goto out_trace;
3317 	retval = dquot_initialize(d_inode(dentry));
3318 	if (retval)
3319 		goto out_trace;
3320 
3321 	retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry), dentry);
3322 #if IS_ENABLED(CONFIG_UNICODE)
3323 	/* VFS negative dentries are incompatible with Encoding and
3324 	 * Case-insensitiveness. Eventually we'll want avoid
3325 	 * invalidating the dentries here, alongside with returning the
3326 	 * negative dentries at ext4_lookup(), when it is  better
3327 	 * supported by the VFS for the CI case.
3328 	 */
3329 	if (IS_CASEFOLDED(dir))
3330 		d_invalidate(dentry);
3331 #endif
3332 
3333 out_trace:
3334 	trace_ext4_unlink_exit(dentry, retval);
3335 	return retval;
3336 }
3337 
3338 static int ext4_init_symlink_block(handle_t *handle, struct inode *inode,
3339 				   struct fscrypt_str *disk_link)
3340 {
3341 	struct buffer_head *bh;
3342 	char *kaddr;
3343 	int err = 0;
3344 
3345 	bh = ext4_bread(handle, inode, 0, EXT4_GET_BLOCKS_CREATE);
3346 	if (IS_ERR(bh))
3347 		return PTR_ERR(bh);
3348 
3349 	BUFFER_TRACE(bh, "get_write_access");
3350 	err = ext4_journal_get_write_access(handle, inode->i_sb, bh, EXT4_JTR_NONE);
3351 	if (err)
3352 		goto out;
3353 
3354 	kaddr = (char *)bh->b_data;
3355 	memcpy(kaddr, disk_link->name, disk_link->len);
3356 	inode->i_size = disk_link->len - 1;
3357 	EXT4_I(inode)->i_disksize = inode->i_size;
3358 	err = ext4_handle_dirty_metadata(handle, inode, bh);
3359 out:
3360 	brelse(bh);
3361 	return err;
3362 }
3363 
3364 static int ext4_symlink(struct mnt_idmap *idmap, struct inode *dir,
3365 			struct dentry *dentry, const char *symname)
3366 {
3367 	handle_t *handle;
3368 	struct inode *inode;
3369 	int err, len = strlen(symname);
3370 	int credits;
3371 	struct fscrypt_str disk_link;
3372 	int retries = 0;
3373 
3374 	if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3375 		return -EIO;
3376 
3377 	err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3378 				      &disk_link);
3379 	if (err)
3380 		return err;
3381 
3382 	err = dquot_initialize(dir);
3383 	if (err)
3384 		return err;
3385 
3386 	/*
3387 	 * EXT4_INDEX_EXTRA_TRANS_BLOCKS for addition of entry into the
3388 	 * directory. +3 for inode, inode bitmap, group descriptor allocation.
3389 	 * EXT4_DATA_TRANS_BLOCKS for the data block allocation and
3390 	 * modification.
3391 	 */
3392 	credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3393 		  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3394 retry:
3395 	inode = ext4_new_inode_start_handle(idmap, dir, S_IFLNK|S_IRWXUGO,
3396 					    &dentry->d_name, 0, NULL,
3397 					    EXT4_HT_DIR, credits);
3398 	handle = ext4_journal_current_handle();
3399 	if (IS_ERR(inode)) {
3400 		if (handle)
3401 			ext4_journal_stop(handle);
3402 		err = PTR_ERR(inode);
3403 		goto out_retry;
3404 	}
3405 
3406 	if (IS_ENCRYPTED(inode)) {
3407 		err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3408 		if (err)
3409 			goto err_drop_inode;
3410 		inode->i_op = &ext4_encrypted_symlink_inode_operations;
3411 	} else {
3412 		if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3413 			inode->i_op = &ext4_symlink_inode_operations;
3414 		} else {
3415 			inode->i_op = &ext4_fast_symlink_inode_operations;
3416 			inode->i_link = (char *)&EXT4_I(inode)->i_data;
3417 		}
3418 	}
3419 
3420 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3421 		/* alloc symlink block and fill it */
3422 		err = ext4_init_symlink_block(handle, inode, &disk_link);
3423 		if (err)
3424 			goto err_drop_inode;
3425 	} else {
3426 		/* clear the extent format for fast symlink */
3427 		ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3428 		memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3429 		       disk_link.len);
3430 		inode->i_size = disk_link.len - 1;
3431 		EXT4_I(inode)->i_disksize = inode->i_size;
3432 	}
3433 	err = ext4_add_nondir(handle, dentry, &inode);
3434 	if (handle)
3435 		ext4_journal_stop(handle);
3436 	iput(inode);
3437 	goto out_retry;
3438 
3439 err_drop_inode:
3440 	clear_nlink(inode);
3441 	ext4_mark_inode_dirty(handle, inode);
3442 	ext4_orphan_add(handle, inode);
3443 	unlock_new_inode(inode);
3444 	if (handle)
3445 		ext4_journal_stop(handle);
3446 	iput(inode);
3447 out_retry:
3448 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3449 		goto retry;
3450 	if (disk_link.name != (unsigned char *)symname)
3451 		kfree(disk_link.name);
3452 	return err;
3453 }
3454 
3455 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3456 {
3457 	handle_t *handle;
3458 	int err, retries = 0;
3459 retry:
3460 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3461 		(EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3462 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3463 	if (IS_ERR(handle))
3464 		return PTR_ERR(handle);
3465 
3466 	if (IS_DIRSYNC(dir))
3467 		ext4_handle_sync(handle);
3468 
3469 	inode_set_ctime_current(inode);
3470 	ext4_inc_count(inode);
3471 	ihold(inode);
3472 
3473 	err = ext4_add_entry(handle, dentry, inode);
3474 	if (!err) {
3475 		err = ext4_mark_inode_dirty(handle, inode);
3476 		/* this can happen only for tmpfile being
3477 		 * linked the first time
3478 		 */
3479 		if (inode->i_nlink == 1)
3480 			ext4_orphan_del(handle, inode);
3481 		d_instantiate(dentry, inode);
3482 		ext4_fc_track_link(handle, dentry);
3483 	} else {
3484 		drop_nlink(inode);
3485 		iput(inode);
3486 	}
3487 	ext4_journal_stop(handle);
3488 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3489 		goto retry;
3490 	return err;
3491 }
3492 
3493 static int ext4_link(struct dentry *old_dentry,
3494 		     struct inode *dir, struct dentry *dentry)
3495 {
3496 	struct inode *inode = d_inode(old_dentry);
3497 	int err;
3498 
3499 	if (inode->i_nlink >= EXT4_LINK_MAX)
3500 		return -EMLINK;
3501 
3502 	err = fscrypt_prepare_link(old_dentry, dir, dentry);
3503 	if (err)
3504 		return err;
3505 
3506 	if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3507 	    (!projid_eq(EXT4_I(dir)->i_projid,
3508 			EXT4_I(old_dentry->d_inode)->i_projid)))
3509 		return -EXDEV;
3510 
3511 	err = dquot_initialize(dir);
3512 	if (err)
3513 		return err;
3514 	return __ext4_link(dir, inode, dentry);
3515 }
3516 
3517 /*
3518  * Try to find buffer head where contains the parent block.
3519  * It should be the inode block if it is inlined or the 1st block
3520  * if it is a normal dir.
3521  */
3522 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3523 					struct inode *inode,
3524 					int *retval,
3525 					struct ext4_dir_entry_2 **parent_de,
3526 					int *inlined)
3527 {
3528 	struct buffer_head *bh;
3529 
3530 	if (!ext4_has_inline_data(inode)) {
3531 		struct ext4_dir_entry_2 *de;
3532 		unsigned int offset;
3533 
3534 		/* The first directory block must not be a hole, so
3535 		 * treat it as DIRENT_HTREE
3536 		 */
3537 		bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3538 		if (IS_ERR(bh)) {
3539 			*retval = PTR_ERR(bh);
3540 			return NULL;
3541 		}
3542 
3543 		de = (struct ext4_dir_entry_2 *) bh->b_data;
3544 		if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3545 					 bh->b_size, 0) ||
3546 		    le32_to_cpu(de->inode) != inode->i_ino ||
3547 		    strcmp(".", de->name)) {
3548 			EXT4_ERROR_INODE(inode, "directory missing '.'");
3549 			brelse(bh);
3550 			*retval = -EFSCORRUPTED;
3551 			return NULL;
3552 		}
3553 		offset = ext4_rec_len_from_disk(de->rec_len,
3554 						inode->i_sb->s_blocksize);
3555 		de = ext4_next_entry(de, inode->i_sb->s_blocksize);
3556 		if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3557 					 bh->b_size, offset) ||
3558 		    le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3559 			EXT4_ERROR_INODE(inode, "directory missing '..'");
3560 			brelse(bh);
3561 			*retval = -EFSCORRUPTED;
3562 			return NULL;
3563 		}
3564 		*parent_de = de;
3565 
3566 		return bh;
3567 	}
3568 
3569 	*inlined = 1;
3570 	return ext4_get_first_inline_block(inode, parent_de, retval);
3571 }
3572 
3573 struct ext4_renament {
3574 	struct inode *dir;
3575 	struct dentry *dentry;
3576 	struct inode *inode;
3577 	bool is_dir;
3578 	int dir_nlink_delta;
3579 
3580 	/* entry for "dentry" */
3581 	struct buffer_head *bh;
3582 	struct ext4_dir_entry_2 *de;
3583 	int inlined;
3584 
3585 	/* entry for ".." in inode if it's a directory */
3586 	struct buffer_head *dir_bh;
3587 	struct ext4_dir_entry_2 *parent_de;
3588 	int dir_inlined;
3589 };
3590 
3591 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent, bool is_cross)
3592 {
3593 	int retval;
3594 
3595 	ent->is_dir = true;
3596 	if (!is_cross)
3597 		return 0;
3598 
3599 	ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3600 					      &retval, &ent->parent_de,
3601 					      &ent->dir_inlined);
3602 	if (!ent->dir_bh)
3603 		return retval;
3604 	if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3605 		return -EFSCORRUPTED;
3606 	BUFFER_TRACE(ent->dir_bh, "get_write_access");
3607 	return ext4_journal_get_write_access(handle, ent->dir->i_sb,
3608 					     ent->dir_bh, EXT4_JTR_NONE);
3609 }
3610 
3611 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3612 				  unsigned dir_ino)
3613 {
3614 	int retval;
3615 
3616 	if (!ent->dir_bh)
3617 		return 0;
3618 
3619 	ent->parent_de->inode = cpu_to_le32(dir_ino);
3620 	BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3621 	if (!ent->dir_inlined) {
3622 		if (is_dx(ent->inode)) {
3623 			retval = ext4_handle_dirty_dx_node(handle,
3624 							   ent->inode,
3625 							   ent->dir_bh);
3626 		} else {
3627 			retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3628 							    ent->dir_bh);
3629 		}
3630 	} else {
3631 		retval = ext4_mark_inode_dirty(handle, ent->inode);
3632 	}
3633 	if (retval) {
3634 		ext4_std_error(ent->dir->i_sb, retval);
3635 		return retval;
3636 	}
3637 	return 0;
3638 }
3639 
3640 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3641 		       unsigned ino, unsigned file_type)
3642 {
3643 	int retval, retval2;
3644 
3645 	BUFFER_TRACE(ent->bh, "get write access");
3646 	retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
3647 					       EXT4_JTR_NONE);
3648 	if (retval)
3649 		return retval;
3650 	ent->de->inode = cpu_to_le32(ino);
3651 	if (ext4_has_feature_filetype(ent->dir->i_sb))
3652 		ent->de->file_type = file_type;
3653 	inode_inc_iversion(ent->dir);
3654 	inode_set_mtime_to_ts(ent->dir, inode_set_ctime_current(ent->dir));
3655 	retval = ext4_mark_inode_dirty(handle, ent->dir);
3656 	BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3657 	if (!ent->inlined) {
3658 		retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3659 		if (unlikely(retval2)) {
3660 			ext4_std_error(ent->dir->i_sb, retval2);
3661 			return retval2;
3662 		}
3663 	}
3664 	return retval;
3665 }
3666 
3667 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3668 			  unsigned ino, unsigned file_type)
3669 {
3670 	struct ext4_renament old = *ent;
3671 	int retval = 0;
3672 
3673 	/*
3674 	 * old->de could have moved from under us during make indexed dir,
3675 	 * so the old->de may no longer valid and need to find it again
3676 	 * before reset old inode info.
3677 	 */
3678 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3679 				 &old.inlined);
3680 	if (IS_ERR(old.bh))
3681 		retval = PTR_ERR(old.bh);
3682 	if (!old.bh)
3683 		retval = -ENOENT;
3684 	if (retval) {
3685 		ext4_std_error(old.dir->i_sb, retval);
3686 		return;
3687 	}
3688 
3689 	ext4_setent(handle, &old, ino, file_type);
3690 	brelse(old.bh);
3691 }
3692 
3693 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3694 				  const struct qstr *d_name)
3695 {
3696 	int retval = -ENOENT;
3697 	struct buffer_head *bh;
3698 	struct ext4_dir_entry_2 *de;
3699 
3700 	bh = ext4_find_entry(dir, d_name, &de, NULL);
3701 	if (IS_ERR(bh))
3702 		return PTR_ERR(bh);
3703 	if (bh) {
3704 		retval = ext4_delete_entry(handle, dir, de, bh);
3705 		brelse(bh);
3706 	}
3707 	return retval;
3708 }
3709 
3710 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3711 			       int force_reread)
3712 {
3713 	int retval;
3714 	/*
3715 	 * ent->de could have moved from under us during htree split, so make
3716 	 * sure that we are deleting the right entry.  We might also be pointing
3717 	 * to a stale entry in the unused part of ent->bh so just checking inum
3718 	 * and the name isn't enough.
3719 	 */
3720 	if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3721 	    ent->de->name_len != ent->dentry->d_name.len ||
3722 	    strncmp(ent->de->name, ent->dentry->d_name.name,
3723 		    ent->de->name_len) ||
3724 	    force_reread) {
3725 		retval = ext4_find_delete_entry(handle, ent->dir,
3726 						&ent->dentry->d_name);
3727 	} else {
3728 		retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3729 		if (retval == -ENOENT) {
3730 			retval = ext4_find_delete_entry(handle, ent->dir,
3731 							&ent->dentry->d_name);
3732 		}
3733 	}
3734 
3735 	if (retval) {
3736 		ext4_warning_inode(ent->dir,
3737 				   "Deleting old file: nlink %d, error=%d",
3738 				   ent->dir->i_nlink, retval);
3739 	}
3740 }
3741 
3742 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3743 {
3744 	if (ent->dir_nlink_delta) {
3745 		if (ent->dir_nlink_delta == -1)
3746 			ext4_dec_count(ent->dir);
3747 		else
3748 			ext4_inc_count(ent->dir);
3749 		ext4_mark_inode_dirty(handle, ent->dir);
3750 	}
3751 }
3752 
3753 static struct inode *ext4_whiteout_for_rename(struct mnt_idmap *idmap,
3754 					      struct ext4_renament *ent,
3755 					      int credits, handle_t **h)
3756 {
3757 	struct inode *wh;
3758 	handle_t *handle;
3759 	int retries = 0;
3760 
3761 	/*
3762 	 * for inode block, sb block, group summaries,
3763 	 * and inode bitmap
3764 	 */
3765 	credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3766 		    EXT4_XATTR_TRANS_BLOCKS + 4);
3767 retry:
3768 	wh = ext4_new_inode_start_handle(idmap, ent->dir,
3769 					 S_IFCHR | WHITEOUT_MODE,
3770 					 &ent->dentry->d_name, 0, NULL,
3771 					 EXT4_HT_DIR, credits);
3772 
3773 	handle = ext4_journal_current_handle();
3774 	if (IS_ERR(wh)) {
3775 		if (handle)
3776 			ext4_journal_stop(handle);
3777 		if (PTR_ERR(wh) == -ENOSPC &&
3778 		    ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3779 			goto retry;
3780 	} else {
3781 		*h = handle;
3782 		init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3783 		wh->i_op = &ext4_special_inode_operations;
3784 	}
3785 	return wh;
3786 }
3787 
3788 /*
3789  * Anybody can rename anything with this: the permission checks are left to the
3790  * higher-level routines.
3791  *
3792  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3793  * while new_{dentry,inode) refers to the destination dentry/inode
3794  * This comes from rename(const char *oldpath, const char *newpath)
3795  */
3796 static int ext4_rename(struct mnt_idmap *idmap, struct inode *old_dir,
3797 		       struct dentry *old_dentry, struct inode *new_dir,
3798 		       struct dentry *new_dentry, unsigned int flags)
3799 {
3800 	handle_t *handle = NULL;
3801 	struct ext4_renament old = {
3802 		.dir = old_dir,
3803 		.dentry = old_dentry,
3804 		.inode = d_inode(old_dentry),
3805 	};
3806 	struct ext4_renament new = {
3807 		.dir = new_dir,
3808 		.dentry = new_dentry,
3809 		.inode = d_inode(new_dentry),
3810 	};
3811 	int force_reread;
3812 	int retval;
3813 	struct inode *whiteout = NULL;
3814 	int credits;
3815 	u8 old_file_type;
3816 
3817 	if (new.inode && new.inode->i_nlink == 0) {
3818 		EXT4_ERROR_INODE(new.inode,
3819 				 "target of rename is already freed");
3820 		return -EFSCORRUPTED;
3821 	}
3822 
3823 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3824 	    (!projid_eq(EXT4_I(new_dir)->i_projid,
3825 			EXT4_I(old_dentry->d_inode)->i_projid)))
3826 		return -EXDEV;
3827 
3828 	retval = dquot_initialize(old.dir);
3829 	if (retval)
3830 		return retval;
3831 	retval = dquot_initialize(old.inode);
3832 	if (retval)
3833 		return retval;
3834 	retval = dquot_initialize(new.dir);
3835 	if (retval)
3836 		return retval;
3837 
3838 	/* Initialize quotas before so that eventual writes go
3839 	 * in separate transaction */
3840 	if (new.inode) {
3841 		retval = dquot_initialize(new.inode);
3842 		if (retval)
3843 			return retval;
3844 	}
3845 
3846 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3847 				 &old.inlined);
3848 	if (IS_ERR(old.bh))
3849 		return PTR_ERR(old.bh);
3850 
3851 	/*
3852 	 *  Check for inode number is _not_ due to possible IO errors.
3853 	 *  We might rmdir the source, keep it as pwd of some process
3854 	 *  and merrily kill the link to whatever was created under the
3855 	 *  same name. Goodbye sticky bit ;-<
3856 	 */
3857 	retval = -ENOENT;
3858 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3859 		goto release_bh;
3860 
3861 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3862 				 &new.de, &new.inlined);
3863 	if (IS_ERR(new.bh)) {
3864 		retval = PTR_ERR(new.bh);
3865 		new.bh = NULL;
3866 		goto release_bh;
3867 	}
3868 	if (new.bh) {
3869 		if (!new.inode) {
3870 			brelse(new.bh);
3871 			new.bh = NULL;
3872 		}
3873 	}
3874 	if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3875 		ext4_alloc_da_blocks(old.inode);
3876 
3877 	credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3878 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3879 	if (!(flags & RENAME_WHITEOUT)) {
3880 		handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3881 		if (IS_ERR(handle)) {
3882 			retval = PTR_ERR(handle);
3883 			goto release_bh;
3884 		}
3885 	} else {
3886 		whiteout = ext4_whiteout_for_rename(idmap, &old, credits, &handle);
3887 		if (IS_ERR(whiteout)) {
3888 			retval = PTR_ERR(whiteout);
3889 			goto release_bh;
3890 		}
3891 	}
3892 
3893 	old_file_type = old.de->file_type;
3894 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3895 		ext4_handle_sync(handle);
3896 
3897 	if (S_ISDIR(old.inode->i_mode)) {
3898 		if (new.inode) {
3899 			retval = -ENOTEMPTY;
3900 			if (!ext4_empty_dir(new.inode))
3901 				goto end_rename;
3902 		} else {
3903 			retval = -EMLINK;
3904 			if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3905 				goto end_rename;
3906 		}
3907 		retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
3908 		if (retval)
3909 			goto end_rename;
3910 	}
3911 	/*
3912 	 * If we're renaming a file within an inline_data dir and adding or
3913 	 * setting the new dirent causes a conversion from inline_data to
3914 	 * extents/blockmap, we need to force the dirent delete code to
3915 	 * re-read the directory, or else we end up trying to delete a dirent
3916 	 * from what is now the extent tree root (or a block map).
3917 	 */
3918 	force_reread = (new.dir->i_ino == old.dir->i_ino &&
3919 			ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3920 
3921 	if (whiteout) {
3922 		/*
3923 		 * Do this before adding a new entry, so the old entry is sure
3924 		 * to be still pointing to the valid old entry.
3925 		 */
3926 		retval = ext4_setent(handle, &old, whiteout->i_ino,
3927 				     EXT4_FT_CHRDEV);
3928 		if (retval)
3929 			goto end_rename;
3930 		retval = ext4_mark_inode_dirty(handle, whiteout);
3931 		if (unlikely(retval))
3932 			goto end_rename;
3933 
3934 	}
3935 	if (!new.bh) {
3936 		retval = ext4_add_entry(handle, new.dentry, old.inode);
3937 		if (retval)
3938 			goto end_rename;
3939 	} else {
3940 		retval = ext4_setent(handle, &new,
3941 				     old.inode->i_ino, old_file_type);
3942 		if (retval)
3943 			goto end_rename;
3944 	}
3945 	if (force_reread)
3946 		force_reread = !ext4_test_inode_flag(new.dir,
3947 						     EXT4_INODE_INLINE_DATA);
3948 
3949 	/*
3950 	 * Like most other Unix systems, set the ctime for inodes on a
3951 	 * rename.
3952 	 */
3953 	inode_set_ctime_current(old.inode);
3954 	retval = ext4_mark_inode_dirty(handle, old.inode);
3955 	if (unlikely(retval))
3956 		goto end_rename;
3957 
3958 	if (!whiteout) {
3959 		/*
3960 		 * ok, that's it
3961 		 */
3962 		ext4_rename_delete(handle, &old, force_reread);
3963 	}
3964 
3965 	if (new.inode) {
3966 		ext4_dec_count(new.inode);
3967 		inode_set_ctime_current(new.inode);
3968 	}
3969 	inode_set_mtime_to_ts(old.dir, inode_set_ctime_current(old.dir));
3970 	ext4_update_dx_flag(old.dir);
3971 	if (old.is_dir) {
3972 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3973 		if (retval)
3974 			goto end_rename;
3975 
3976 		ext4_dec_count(old.dir);
3977 		if (new.inode) {
3978 			/* checked ext4_empty_dir above, can't have another
3979 			 * parent, ext4_dec_count() won't work for many-linked
3980 			 * dirs */
3981 			clear_nlink(new.inode);
3982 		} else {
3983 			ext4_inc_count(new.dir);
3984 			ext4_update_dx_flag(new.dir);
3985 			retval = ext4_mark_inode_dirty(handle, new.dir);
3986 			if (unlikely(retval))
3987 				goto end_rename;
3988 		}
3989 	}
3990 	retval = ext4_mark_inode_dirty(handle, old.dir);
3991 	if (unlikely(retval))
3992 		goto end_rename;
3993 
3994 	if (old.is_dir) {
3995 		/*
3996 		 * We disable fast commits here that's because the
3997 		 * replay code is not yet capable of changing dot dot
3998 		 * dirents in directories.
3999 		 */
4000 		ext4_fc_mark_ineligible(old.inode->i_sb,
4001 			EXT4_FC_REASON_RENAME_DIR, handle);
4002 	} else {
4003 		struct super_block *sb = old.inode->i_sb;
4004 
4005 		if (new.inode)
4006 			ext4_fc_track_unlink(handle, new.dentry);
4007 		if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4008 		    !(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4009 		    !(ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE))) {
4010 			__ext4_fc_track_link(handle, old.inode, new.dentry);
4011 			__ext4_fc_track_unlink(handle, old.inode, old.dentry);
4012 			if (whiteout)
4013 				__ext4_fc_track_create(handle, whiteout,
4014 						       old.dentry);
4015 		}
4016 	}
4017 
4018 	if (new.inode) {
4019 		retval = ext4_mark_inode_dirty(handle, new.inode);
4020 		if (unlikely(retval))
4021 			goto end_rename;
4022 		if (!new.inode->i_nlink)
4023 			ext4_orphan_add(handle, new.inode);
4024 	}
4025 	retval = 0;
4026 
4027 end_rename:
4028 	if (whiteout) {
4029 		if (retval) {
4030 			ext4_resetent(handle, &old,
4031 				      old.inode->i_ino, old_file_type);
4032 			drop_nlink(whiteout);
4033 			ext4_mark_inode_dirty(handle, whiteout);
4034 			ext4_orphan_add(handle, whiteout);
4035 		}
4036 		unlock_new_inode(whiteout);
4037 		ext4_journal_stop(handle);
4038 		iput(whiteout);
4039 	} else {
4040 		ext4_journal_stop(handle);
4041 	}
4042 release_bh:
4043 	brelse(old.dir_bh);
4044 	brelse(old.bh);
4045 	brelse(new.bh);
4046 
4047 	return retval;
4048 }
4049 
4050 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4051 			     struct inode *new_dir, struct dentry *new_dentry)
4052 {
4053 	handle_t *handle = NULL;
4054 	struct ext4_renament old = {
4055 		.dir = old_dir,
4056 		.dentry = old_dentry,
4057 		.inode = d_inode(old_dentry),
4058 	};
4059 	struct ext4_renament new = {
4060 		.dir = new_dir,
4061 		.dentry = new_dentry,
4062 		.inode = d_inode(new_dentry),
4063 	};
4064 	u8 new_file_type;
4065 	int retval;
4066 
4067 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4068 	     !projid_eq(EXT4_I(new_dir)->i_projid,
4069 			EXT4_I(old_dentry->d_inode)->i_projid)) ||
4070 	    (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4071 	     !projid_eq(EXT4_I(old_dir)->i_projid,
4072 			EXT4_I(new_dentry->d_inode)->i_projid)))
4073 		return -EXDEV;
4074 
4075 	retval = dquot_initialize(old.dir);
4076 	if (retval)
4077 		return retval;
4078 	retval = dquot_initialize(new.dir);
4079 	if (retval)
4080 		return retval;
4081 
4082 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4083 				 &old.de, &old.inlined);
4084 	if (IS_ERR(old.bh))
4085 		return PTR_ERR(old.bh);
4086 	/*
4087 	 *  Check for inode number is _not_ due to possible IO errors.
4088 	 *  We might rmdir the source, keep it as pwd of some process
4089 	 *  and merrily kill the link to whatever was created under the
4090 	 *  same name. Goodbye sticky bit ;-<
4091 	 */
4092 	retval = -ENOENT;
4093 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4094 		goto end_rename;
4095 
4096 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4097 				 &new.de, &new.inlined);
4098 	if (IS_ERR(new.bh)) {
4099 		retval = PTR_ERR(new.bh);
4100 		new.bh = NULL;
4101 		goto end_rename;
4102 	}
4103 
4104 	/* RENAME_EXCHANGE case: old *and* new must both exist */
4105 	if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4106 		goto end_rename;
4107 
4108 	handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4109 		(2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4110 		 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4111 	if (IS_ERR(handle)) {
4112 		retval = PTR_ERR(handle);
4113 		handle = NULL;
4114 		goto end_rename;
4115 	}
4116 
4117 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4118 		ext4_handle_sync(handle);
4119 
4120 	if (S_ISDIR(old.inode->i_mode)) {
4121 		retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
4122 		if (retval)
4123 			goto end_rename;
4124 	}
4125 	if (S_ISDIR(new.inode->i_mode)) {
4126 		retval = ext4_rename_dir_prepare(handle, &new, new.dir != old.dir);
4127 		if (retval)
4128 			goto end_rename;
4129 	}
4130 
4131 	/*
4132 	 * Other than the special case of overwriting a directory, parents'
4133 	 * nlink only needs to be modified if this is a cross directory rename.
4134 	 */
4135 	if (old.dir != new.dir && old.is_dir != new.is_dir) {
4136 		old.dir_nlink_delta = old.is_dir ? -1 : 1;
4137 		new.dir_nlink_delta = -old.dir_nlink_delta;
4138 		retval = -EMLINK;
4139 		if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4140 		    (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4141 			goto end_rename;
4142 	}
4143 
4144 	new_file_type = new.de->file_type;
4145 	retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4146 	if (retval)
4147 		goto end_rename;
4148 
4149 	retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4150 	if (retval)
4151 		goto end_rename;
4152 
4153 	/*
4154 	 * Like most other Unix systems, set the ctime for inodes on a
4155 	 * rename.
4156 	 */
4157 	inode_set_ctime_current(old.inode);
4158 	inode_set_ctime_current(new.inode);
4159 	retval = ext4_mark_inode_dirty(handle, old.inode);
4160 	if (unlikely(retval))
4161 		goto end_rename;
4162 	retval = ext4_mark_inode_dirty(handle, new.inode);
4163 	if (unlikely(retval))
4164 		goto end_rename;
4165 	ext4_fc_mark_ineligible(new.inode->i_sb,
4166 				EXT4_FC_REASON_CROSS_RENAME, handle);
4167 	if (old.dir_bh) {
4168 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4169 		if (retval)
4170 			goto end_rename;
4171 	}
4172 	if (new.dir_bh) {
4173 		retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4174 		if (retval)
4175 			goto end_rename;
4176 	}
4177 	ext4_update_dir_count(handle, &old);
4178 	ext4_update_dir_count(handle, &new);
4179 	retval = 0;
4180 
4181 end_rename:
4182 	brelse(old.dir_bh);
4183 	brelse(new.dir_bh);
4184 	brelse(old.bh);
4185 	brelse(new.bh);
4186 	if (handle)
4187 		ext4_journal_stop(handle);
4188 	return retval;
4189 }
4190 
4191 static int ext4_rename2(struct mnt_idmap *idmap,
4192 			struct inode *old_dir, struct dentry *old_dentry,
4193 			struct inode *new_dir, struct dentry *new_dentry,
4194 			unsigned int flags)
4195 {
4196 	int err;
4197 
4198 	if (unlikely(ext4_forced_shutdown(old_dir->i_sb)))
4199 		return -EIO;
4200 
4201 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4202 		return -EINVAL;
4203 
4204 	err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4205 				     flags);
4206 	if (err)
4207 		return err;
4208 
4209 	if (flags & RENAME_EXCHANGE) {
4210 		return ext4_cross_rename(old_dir, old_dentry,
4211 					 new_dir, new_dentry);
4212 	}
4213 
4214 	return ext4_rename(idmap, old_dir, old_dentry, new_dir, new_dentry, flags);
4215 }
4216 
4217 /*
4218  * directories can handle most operations...
4219  */
4220 const struct inode_operations ext4_dir_inode_operations = {
4221 	.create		= ext4_create,
4222 	.lookup		= ext4_lookup,
4223 	.link		= ext4_link,
4224 	.unlink		= ext4_unlink,
4225 	.symlink	= ext4_symlink,
4226 	.mkdir		= ext4_mkdir,
4227 	.rmdir		= ext4_rmdir,
4228 	.mknod		= ext4_mknod,
4229 	.tmpfile	= ext4_tmpfile,
4230 	.rename		= ext4_rename2,
4231 	.setattr	= ext4_setattr,
4232 	.getattr	= ext4_getattr,
4233 	.listxattr	= ext4_listxattr,
4234 	.get_inode_acl	= ext4_get_acl,
4235 	.set_acl	= ext4_set_acl,
4236 	.fiemap         = ext4_fiemap,
4237 	.fileattr_get	= ext4_fileattr_get,
4238 	.fileattr_set	= ext4_fileattr_set,
4239 };
4240 
4241 const struct inode_operations ext4_special_inode_operations = {
4242 	.setattr	= ext4_setattr,
4243 	.getattr	= ext4_getattr,
4244 	.listxattr	= ext4_listxattr,
4245 	.get_inode_acl	= ext4_get_acl,
4246 	.set_acl	= ext4_set_acl,
4247 };
4248