xref: /linux/fs/ext4/namei.c (revision 3630400697a3d334a391c1dba1b601d852145f2c)
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 	/* The first directory block must not be a hole. */
155 	if (!bh && (type == INDEX || type == DIRENT_HTREE || block == 0)) {
156 		ext4_error_inode(inode, func, line, block,
157 				 "Directory hole found for htree %s block %u",
158 				 (type == INDEX) ? "index" : "leaf", block);
159 		return ERR_PTR(-EFSCORRUPTED);
160 	}
161 	if (!bh)
162 		return NULL;
163 	dirent = (struct ext4_dir_entry *) bh->b_data;
164 	/* Determine whether or not we have an index block */
165 	if (is_dx(inode)) {
166 		if (block == 0)
167 			is_dx_block = 1;
168 		else if (ext4_rec_len_from_disk(dirent->rec_len,
169 						inode->i_sb->s_blocksize) ==
170 			 inode->i_sb->s_blocksize)
171 			is_dx_block = 1;
172 	}
173 	if (!is_dx_block && type == INDEX) {
174 		ext4_error_inode(inode, func, line, block,
175 		       "directory leaf block found instead of index block");
176 		brelse(bh);
177 		return ERR_PTR(-EFSCORRUPTED);
178 	}
179 	if (!ext4_has_metadata_csum(inode->i_sb) ||
180 	    buffer_verified(bh))
181 		return bh;
182 
183 	/*
184 	 * An empty leaf block can get mistaken for a index block; for
185 	 * this reason, we can only check the index checksum when the
186 	 * caller is sure it should be an index block.
187 	 */
188 	if (is_dx_block && type == INDEX) {
189 		if (ext4_dx_csum_verify(inode, dirent) &&
190 		    !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
191 			set_buffer_verified(bh);
192 		else {
193 			ext4_error_inode_err(inode, func, line, block,
194 					     EFSBADCRC,
195 					     "Directory index failed checksum");
196 			brelse(bh);
197 			return ERR_PTR(-EFSBADCRC);
198 		}
199 	}
200 	if (!is_dx_block) {
201 		if (ext4_dirblock_csum_verify(inode, bh) &&
202 		    !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
203 			set_buffer_verified(bh);
204 		else {
205 			ext4_error_inode_err(inode, func, line, block,
206 					     EFSBADCRC,
207 					     "Directory block failed checksum");
208 			brelse(bh);
209 			return ERR_PTR(-EFSBADCRC);
210 		}
211 	}
212 	return bh;
213 }
214 
215 #ifdef DX_DEBUG
216 #define dxtrace(command) command
217 #else
218 #define dxtrace(command)
219 #endif
220 
221 struct fake_dirent
222 {
223 	__le32 inode;
224 	__le16 rec_len;
225 	u8 name_len;
226 	u8 file_type;
227 };
228 
229 struct dx_countlimit
230 {
231 	__le16 limit;
232 	__le16 count;
233 };
234 
235 struct dx_entry
236 {
237 	__le32 hash;
238 	__le32 block;
239 };
240 
241 /*
242  * dx_root_info is laid out so that if it should somehow get overlaid by a
243  * dirent the two low bits of the hash version will be zero.  Therefore, the
244  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
245  */
246 
247 struct dx_root
248 {
249 	struct fake_dirent dot;
250 	char dot_name[4];
251 	struct fake_dirent dotdot;
252 	char dotdot_name[4];
253 	struct dx_root_info
254 	{
255 		__le32 reserved_zero;
256 		u8 hash_version;
257 		u8 info_length; /* 8 */
258 		u8 indirect_levels;
259 		u8 unused_flags;
260 	}
261 	info;
262 	struct dx_entry	entries[];
263 };
264 
265 struct dx_node
266 {
267 	struct fake_dirent fake;
268 	struct dx_entry	entries[];
269 };
270 
271 
272 struct dx_frame
273 {
274 	struct buffer_head *bh;
275 	struct dx_entry *entries;
276 	struct dx_entry *at;
277 };
278 
279 struct dx_map_entry
280 {
281 	u32 hash;
282 	u16 offs;
283 	u16 size;
284 };
285 
286 /*
287  * This goes at the end of each htree block.
288  */
289 struct dx_tail {
290 	u32 dt_reserved;
291 	__le32 dt_checksum;	/* crc32c(uuid+inum+dirblock) */
292 };
293 
294 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
295 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
296 static inline unsigned dx_get_hash(struct dx_entry *entry);
297 static void dx_set_hash(struct dx_entry *entry, unsigned value);
298 static unsigned dx_get_count(struct dx_entry *entries);
299 static unsigned dx_get_limit(struct dx_entry *entries);
300 static void dx_set_count(struct dx_entry *entries, unsigned value);
301 static void dx_set_limit(struct dx_entry *entries, unsigned value);
302 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
303 static unsigned dx_node_limit(struct inode *dir);
304 static struct dx_frame *dx_probe(struct ext4_filename *fname,
305 				 struct inode *dir,
306 				 struct dx_hash_info *hinfo,
307 				 struct dx_frame *frame);
308 static void dx_release(struct dx_frame *frames);
309 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
310 		       struct dx_hash_info *hinfo,
311 		       struct dx_map_entry *map_tail);
312 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
313 static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from,
314 					char *to, struct dx_map_entry *offsets,
315 					int count, unsigned int blocksize);
316 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
317 						unsigned int blocksize);
318 static void dx_insert_block(struct dx_frame *frame,
319 					u32 hash, ext4_lblk_t block);
320 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
321 				 struct dx_frame *frame,
322 				 struct dx_frame *frames,
323 				 __u32 *start_hash);
324 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
325 		struct ext4_filename *fname,
326 		struct ext4_dir_entry_2 **res_dir);
327 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
328 			     struct inode *dir, struct inode *inode);
329 
330 /* checksumming functions */
331 void ext4_initialize_dirent_tail(struct buffer_head *bh,
332 				 unsigned int blocksize)
333 {
334 	struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
335 
336 	memset(t, 0, sizeof(struct ext4_dir_entry_tail));
337 	t->det_rec_len = ext4_rec_len_to_disk(
338 			sizeof(struct ext4_dir_entry_tail), blocksize);
339 	t->det_reserved_ft = EXT4_FT_DIR_CSUM;
340 }
341 
342 /* Walk through a dirent block to find a checksum "dirent" at the tail */
343 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
344 						   struct buffer_head *bh)
345 {
346 	struct ext4_dir_entry_tail *t;
347 	int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
348 
349 #ifdef PARANOID
350 	struct ext4_dir_entry *d, *top;
351 
352 	d = (struct ext4_dir_entry *)bh->b_data;
353 	top = (struct ext4_dir_entry *)(bh->b_data +
354 		(blocksize - sizeof(struct ext4_dir_entry_tail)));
355 	while (d < top && ext4_rec_len_from_disk(d->rec_len, blocksize))
356 		d = (struct ext4_dir_entry *)(((void *)d) +
357 		    ext4_rec_len_from_disk(d->rec_len, blocksize));
358 
359 	if (d != top)
360 		return NULL;
361 
362 	t = (struct ext4_dir_entry_tail *)d;
363 #else
364 	t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
365 #endif
366 
367 	if (t->det_reserved_zero1 ||
368 	    (ext4_rec_len_from_disk(t->det_rec_len, blocksize) !=
369 	     sizeof(struct ext4_dir_entry_tail)) ||
370 	    t->det_reserved_zero2 ||
371 	    t->det_reserved_ft != EXT4_FT_DIR_CSUM)
372 		return NULL;
373 
374 	return t;
375 }
376 
377 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
378 {
379 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
380 	struct ext4_inode_info *ei = EXT4_I(inode);
381 	__u32 csum;
382 
383 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
384 	return cpu_to_le32(csum);
385 }
386 
387 #define warn_no_space_for_csum(inode)					\
388 	__warn_no_space_for_csum((inode), __func__, __LINE__)
389 
390 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
391 				     unsigned int line)
392 {
393 	__ext4_warning_inode(inode, func, line,
394 		"No space for directory leaf checksum. Please run e2fsck -D.");
395 }
396 
397 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
398 {
399 	struct ext4_dir_entry_tail *t;
400 
401 	if (!ext4_has_metadata_csum(inode->i_sb))
402 		return 1;
403 
404 	t = get_dirent_tail(inode, bh);
405 	if (!t) {
406 		warn_no_space_for_csum(inode);
407 		return 0;
408 	}
409 
410 	if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
411 						  (char *)t - bh->b_data))
412 		return 0;
413 
414 	return 1;
415 }
416 
417 static void ext4_dirblock_csum_set(struct inode *inode,
418 				 struct buffer_head *bh)
419 {
420 	struct ext4_dir_entry_tail *t;
421 
422 	if (!ext4_has_metadata_csum(inode->i_sb))
423 		return;
424 
425 	t = get_dirent_tail(inode, bh);
426 	if (!t) {
427 		warn_no_space_for_csum(inode);
428 		return;
429 	}
430 
431 	t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
432 					     (char *)t - bh->b_data);
433 }
434 
435 int ext4_handle_dirty_dirblock(handle_t *handle,
436 			       struct inode *inode,
437 			       struct buffer_head *bh)
438 {
439 	ext4_dirblock_csum_set(inode, bh);
440 	return ext4_handle_dirty_metadata(handle, inode, bh);
441 }
442 
443 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
444 					       struct ext4_dir_entry *dirent,
445 					       int *offset)
446 {
447 	struct ext4_dir_entry *dp;
448 	struct dx_root_info *root;
449 	int count_offset;
450 	int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
451 	unsigned int rlen = ext4_rec_len_from_disk(dirent->rec_len, blocksize);
452 
453 	if (rlen == blocksize)
454 		count_offset = 8;
455 	else if (rlen == 12) {
456 		dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
457 		if (ext4_rec_len_from_disk(dp->rec_len, blocksize) != blocksize - 12)
458 			return NULL;
459 		root = (struct dx_root_info *)(((void *)dp + 12));
460 		if (root->reserved_zero ||
461 		    root->info_length != sizeof(struct dx_root_info))
462 			return NULL;
463 		count_offset = 32;
464 	} else
465 		return NULL;
466 
467 	if (offset)
468 		*offset = count_offset;
469 	return (struct dx_countlimit *)(((void *)dirent) + count_offset);
470 }
471 
472 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
473 			   int count_offset, int count, struct dx_tail *t)
474 {
475 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
476 	struct ext4_inode_info *ei = EXT4_I(inode);
477 	__u32 csum;
478 	int size;
479 	__u32 dummy_csum = 0;
480 	int offset = offsetof(struct dx_tail, dt_checksum);
481 
482 	size = count_offset + (count * sizeof(struct dx_entry));
483 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
484 	csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
485 	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
486 
487 	return cpu_to_le32(csum);
488 }
489 
490 static int ext4_dx_csum_verify(struct inode *inode,
491 			       struct ext4_dir_entry *dirent)
492 {
493 	struct dx_countlimit *c;
494 	struct dx_tail *t;
495 	int count_offset, limit, count;
496 
497 	if (!ext4_has_metadata_csum(inode->i_sb))
498 		return 1;
499 
500 	c = get_dx_countlimit(inode, dirent, &count_offset);
501 	if (!c) {
502 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
503 		return 0;
504 	}
505 	limit = le16_to_cpu(c->limit);
506 	count = le16_to_cpu(c->count);
507 	if (count_offset + (limit * sizeof(struct dx_entry)) >
508 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
509 		warn_no_space_for_csum(inode);
510 		return 0;
511 	}
512 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
513 
514 	if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
515 					    count, t))
516 		return 0;
517 	return 1;
518 }
519 
520 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
521 {
522 	struct dx_countlimit *c;
523 	struct dx_tail *t;
524 	int count_offset, limit, count;
525 
526 	if (!ext4_has_metadata_csum(inode->i_sb))
527 		return;
528 
529 	c = get_dx_countlimit(inode, dirent, &count_offset);
530 	if (!c) {
531 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
532 		return;
533 	}
534 	limit = le16_to_cpu(c->limit);
535 	count = le16_to_cpu(c->count);
536 	if (count_offset + (limit * sizeof(struct dx_entry)) >
537 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
538 		warn_no_space_for_csum(inode);
539 		return;
540 	}
541 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
542 
543 	t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
544 }
545 
546 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
547 					    struct inode *inode,
548 					    struct buffer_head *bh)
549 {
550 	ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
551 	return ext4_handle_dirty_metadata(handle, inode, bh);
552 }
553 
554 /*
555  * p is at least 6 bytes before the end of page
556  */
557 static inline struct ext4_dir_entry_2 *
558 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
559 {
560 	return (struct ext4_dir_entry_2 *)((char *)p +
561 		ext4_rec_len_from_disk(p->rec_len, blocksize));
562 }
563 
564 /*
565  * Future: use high four bits of block for coalesce-on-delete flags
566  * Mask them off for now.
567  */
568 
569 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
570 {
571 	return le32_to_cpu(entry->block) & 0x0fffffff;
572 }
573 
574 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
575 {
576 	entry->block = cpu_to_le32(value);
577 }
578 
579 static inline unsigned dx_get_hash(struct dx_entry *entry)
580 {
581 	return le32_to_cpu(entry->hash);
582 }
583 
584 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
585 {
586 	entry->hash = cpu_to_le32(value);
587 }
588 
589 static inline unsigned dx_get_count(struct dx_entry *entries)
590 {
591 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
592 }
593 
594 static inline unsigned dx_get_limit(struct dx_entry *entries)
595 {
596 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
597 }
598 
599 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
600 {
601 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
602 }
603 
604 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
605 {
606 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
607 }
608 
609 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
610 {
611 	unsigned int entry_space = dir->i_sb->s_blocksize -
612 			ext4_dir_rec_len(1, NULL) -
613 			ext4_dir_rec_len(2, NULL) - infosize;
614 
615 	if (ext4_has_metadata_csum(dir->i_sb))
616 		entry_space -= sizeof(struct dx_tail);
617 	return entry_space / sizeof(struct dx_entry);
618 }
619 
620 static inline unsigned dx_node_limit(struct inode *dir)
621 {
622 	unsigned int entry_space = dir->i_sb->s_blocksize -
623 			ext4_dir_rec_len(0, dir);
624 
625 	if (ext4_has_metadata_csum(dir->i_sb))
626 		entry_space -= sizeof(struct dx_tail);
627 	return entry_space / sizeof(struct dx_entry);
628 }
629 
630 /*
631  * Debug
632  */
633 #ifdef DX_DEBUG
634 static void dx_show_index(char * label, struct dx_entry *entries)
635 {
636 	int i, n = dx_get_count (entries);
637 	printk(KERN_DEBUG "%s index", label);
638 	for (i = 0; i < n; i++) {
639 		printk(KERN_CONT " %x->%lu",
640 		       i ? dx_get_hash(entries + i) : 0,
641 		       (unsigned long)dx_get_block(entries + i));
642 	}
643 	printk(KERN_CONT "\n");
644 }
645 
646 struct stats
647 {
648 	unsigned names;
649 	unsigned space;
650 	unsigned bcount;
651 };
652 
653 static struct stats dx_show_leaf(struct inode *dir,
654 				struct dx_hash_info *hinfo,
655 				struct ext4_dir_entry_2 *de,
656 				int size, int show_names)
657 {
658 	unsigned names = 0, space = 0;
659 	char *base = (char *) de;
660 	struct dx_hash_info h = *hinfo;
661 
662 	printk("names: ");
663 	while ((char *) de < base + size)
664 	{
665 		if (de->inode)
666 		{
667 			if (show_names)
668 			{
669 #ifdef CONFIG_FS_ENCRYPTION
670 				int len;
671 				char *name;
672 				struct fscrypt_str fname_crypto_str =
673 					FSTR_INIT(NULL, 0);
674 				int res = 0;
675 
676 				name  = de->name;
677 				len = de->name_len;
678 				if (!IS_ENCRYPTED(dir)) {
679 					/* Directory is not encrypted */
680 					(void) ext4fs_dirhash(dir, de->name,
681 						de->name_len, &h);
682 					printk("%*.s:(U)%x.%u ", len,
683 					       name, h.hash,
684 					       (unsigned) ((char *) de
685 							   - base));
686 				} else {
687 					struct fscrypt_str de_name =
688 						FSTR_INIT(name, len);
689 
690 					/* Directory is encrypted */
691 					res = fscrypt_fname_alloc_buffer(
692 						len, &fname_crypto_str);
693 					if (res)
694 						printk(KERN_WARNING "Error "
695 							"allocating crypto "
696 							"buffer--skipping "
697 							"crypto\n");
698 					res = fscrypt_fname_disk_to_usr(dir,
699 						0, 0, &de_name,
700 						&fname_crypto_str);
701 					if (res) {
702 						printk(KERN_WARNING "Error "
703 							"converting filename "
704 							"from disk to usr"
705 							"\n");
706 						name = "??";
707 						len = 2;
708 					} else {
709 						name = fname_crypto_str.name;
710 						len = fname_crypto_str.len;
711 					}
712 					if (IS_CASEFOLDED(dir))
713 						h.hash = EXT4_DIRENT_HASH(de);
714 					else
715 						(void) ext4fs_dirhash(dir,
716 							de->name,
717 							de->name_len, &h);
718 					printk("%*.s:(E)%x.%u ", len, name,
719 					       h.hash, (unsigned) ((char *) de
720 								   - base));
721 					fscrypt_fname_free_buffer(
722 							&fname_crypto_str);
723 				}
724 #else
725 				int len = de->name_len;
726 				char *name = de->name;
727 				(void) ext4fs_dirhash(dir, de->name,
728 						      de->name_len, &h);
729 				printk("%*.s:%x.%u ", len, name, h.hash,
730 				       (unsigned) ((char *) de - base));
731 #endif
732 			}
733 			space += ext4_dir_rec_len(de->name_len, dir);
734 			names++;
735 		}
736 		de = ext4_next_entry(de, size);
737 	}
738 	printk(KERN_CONT "(%i)\n", names);
739 	return (struct stats) { names, space, 1 };
740 }
741 
742 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
743 			     struct dx_entry *entries, int levels)
744 {
745 	unsigned blocksize = dir->i_sb->s_blocksize;
746 	unsigned count = dx_get_count(entries), names = 0, space = 0, i;
747 	unsigned bcount = 0;
748 	struct buffer_head *bh;
749 	printk("%i indexed blocks...\n", count);
750 	for (i = 0; i < count; i++, entries++)
751 	{
752 		ext4_lblk_t block = dx_get_block(entries);
753 		ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
754 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
755 		struct stats stats;
756 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
757 		bh = ext4_bread(NULL,dir, block, 0);
758 		if (!bh || IS_ERR(bh))
759 			continue;
760 		stats = levels?
761 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
762 		   dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
763 			bh->b_data, blocksize, 0);
764 		names += stats.names;
765 		space += stats.space;
766 		bcount += stats.bcount;
767 		brelse(bh);
768 	}
769 	if (bcount)
770 		printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
771 		       levels ? "" : "   ", names, space/bcount,
772 		       (space/bcount)*100/blocksize);
773 	return (struct stats) { names, space, bcount};
774 }
775 
776 /*
777  * Linear search cross check
778  */
779 static inline void htree_rep_invariant_check(struct dx_entry *at,
780 					     struct dx_entry *target,
781 					     u32 hash, unsigned int n)
782 {
783 	while (n--) {
784 		dxtrace(printk(KERN_CONT ","));
785 		if (dx_get_hash(++at) > hash) {
786 			at--;
787 			break;
788 		}
789 	}
790 	ASSERT(at == target - 1);
791 }
792 #else /* DX_DEBUG */
793 static inline void htree_rep_invariant_check(struct dx_entry *at,
794 					     struct dx_entry *target,
795 					     u32 hash, unsigned int n)
796 {
797 }
798 #endif /* DX_DEBUG */
799 
800 /*
801  * Probe for a directory leaf block to search.
802  *
803  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
804  * error in the directory index, and the caller should fall back to
805  * searching the directory normally.  The callers of dx_probe **MUST**
806  * check for this error code, and make sure it never gets reflected
807  * back to userspace.
808  */
809 static struct dx_frame *
810 dx_probe(struct ext4_filename *fname, struct inode *dir,
811 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
812 {
813 	unsigned count, indirect, level, i;
814 	struct dx_entry *at, *entries, *p, *q, *m;
815 	struct dx_root *root;
816 	struct dx_frame *frame = frame_in;
817 	struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
818 	u32 hash;
819 	ext4_lblk_t block;
820 	ext4_lblk_t blocks[EXT4_HTREE_LEVEL];
821 
822 	memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
823 	frame->bh = ext4_read_dirblock(dir, 0, INDEX);
824 	if (IS_ERR(frame->bh))
825 		return (struct dx_frame *) frame->bh;
826 
827 	root = (struct dx_root *) frame->bh->b_data;
828 	if (root->info.hash_version != DX_HASH_TEA &&
829 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
830 	    root->info.hash_version != DX_HASH_LEGACY &&
831 	    root->info.hash_version != DX_HASH_SIPHASH) {
832 		ext4_warning_inode(dir, "Unrecognised inode hash code %u",
833 				   root->info.hash_version);
834 		goto fail;
835 	}
836 	if (ext4_hash_in_dirent(dir)) {
837 		if (root->info.hash_version != DX_HASH_SIPHASH) {
838 			ext4_warning_inode(dir,
839 				"Hash in dirent, but hash is not SIPHASH");
840 			goto fail;
841 		}
842 	} else {
843 		if (root->info.hash_version == DX_HASH_SIPHASH) {
844 			ext4_warning_inode(dir,
845 				"Hash code is SIPHASH, but hash not in dirent");
846 			goto fail;
847 		}
848 	}
849 	if (fname)
850 		hinfo = &fname->hinfo;
851 	hinfo->hash_version = root->info.hash_version;
852 	if (hinfo->hash_version <= DX_HASH_TEA)
853 		hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
854 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
855 	/* hash is already computed for encrypted casefolded directory */
856 	if (fname && fname_name(fname) &&
857 	    !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir))) {
858 		int ret = ext4fs_dirhash(dir, fname_name(fname),
859 					 fname_len(fname), hinfo);
860 		if (ret < 0) {
861 			ret_err = ERR_PTR(ret);
862 			goto fail;
863 		}
864 	}
865 	hash = hinfo->hash;
866 
867 	if (root->info.unused_flags & 1) {
868 		ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
869 				   root->info.unused_flags);
870 		goto fail;
871 	}
872 
873 	indirect = root->info.indirect_levels;
874 	if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
875 		ext4_warning(dir->i_sb,
876 			     "Directory (ino: %lu) htree depth %#06x exceed"
877 			     "supported value", dir->i_ino,
878 			     ext4_dir_htree_level(dir->i_sb));
879 		if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
880 			ext4_warning(dir->i_sb, "Enable large directory "
881 						"feature to access it");
882 		}
883 		goto fail;
884 	}
885 
886 	entries = (struct dx_entry *)(((char *)&root->info) +
887 				      root->info.info_length);
888 
889 	if (dx_get_limit(entries) != dx_root_limit(dir,
890 						   root->info.info_length)) {
891 		ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
892 				   dx_get_limit(entries),
893 				   dx_root_limit(dir, root->info.info_length));
894 		goto fail;
895 	}
896 
897 	dxtrace(printk("Look up %x", hash));
898 	level = 0;
899 	blocks[0] = 0;
900 	while (1) {
901 		count = dx_get_count(entries);
902 		if (!count || count > dx_get_limit(entries)) {
903 			ext4_warning_inode(dir,
904 					   "dx entry: count %u beyond limit %u",
905 					   count, dx_get_limit(entries));
906 			goto fail;
907 		}
908 
909 		p = entries + 1;
910 		q = entries + count - 1;
911 		while (p <= q) {
912 			m = p + (q - p) / 2;
913 			dxtrace(printk(KERN_CONT "."));
914 			if (dx_get_hash(m) > hash)
915 				q = m - 1;
916 			else
917 				p = m + 1;
918 		}
919 
920 		htree_rep_invariant_check(entries, p, hash, count - 1);
921 
922 		at = p - 1;
923 		dxtrace(printk(KERN_CONT " %x->%u\n",
924 			       at == entries ? 0 : dx_get_hash(at),
925 			       dx_get_block(at)));
926 		frame->entries = entries;
927 		frame->at = at;
928 
929 		block = dx_get_block(at);
930 		for (i = 0; i <= level; i++) {
931 			if (blocks[i] == block) {
932 				ext4_warning_inode(dir,
933 					"dx entry: tree cycle block %u points back to block %u",
934 					blocks[level], block);
935 				goto fail;
936 			}
937 		}
938 		if (++level > indirect)
939 			return frame;
940 		blocks[level] = block;
941 		frame++;
942 		frame->bh = ext4_read_dirblock(dir, block, INDEX);
943 		if (IS_ERR(frame->bh)) {
944 			ret_err = (struct dx_frame *) frame->bh;
945 			frame->bh = NULL;
946 			goto fail;
947 		}
948 
949 		entries = ((struct dx_node *) frame->bh->b_data)->entries;
950 
951 		if (dx_get_limit(entries) != dx_node_limit(dir)) {
952 			ext4_warning_inode(dir,
953 				"dx entry: limit %u != node limit %u",
954 				dx_get_limit(entries), dx_node_limit(dir));
955 			goto fail;
956 		}
957 	}
958 fail:
959 	while (frame >= frame_in) {
960 		brelse(frame->bh);
961 		frame--;
962 	}
963 
964 	if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
965 		ext4_warning_inode(dir,
966 			"Corrupt directory, running e2fsck is recommended");
967 	return ret_err;
968 }
969 
970 static void dx_release(struct dx_frame *frames)
971 {
972 	struct dx_root_info *info;
973 	int i;
974 	unsigned int indirect_levels;
975 
976 	if (frames[0].bh == NULL)
977 		return;
978 
979 	info = &((struct dx_root *)frames[0].bh->b_data)->info;
980 	/* save local copy, "info" may be freed after brelse() */
981 	indirect_levels = info->indirect_levels;
982 	for (i = 0; i <= indirect_levels; i++) {
983 		if (frames[i].bh == NULL)
984 			break;
985 		brelse(frames[i].bh);
986 		frames[i].bh = NULL;
987 	}
988 }
989 
990 /*
991  * This function increments the frame pointer to search the next leaf
992  * block, and reads in the necessary intervening nodes if the search
993  * should be necessary.  Whether or not the search is necessary is
994  * controlled by the hash parameter.  If the hash value is even, then
995  * the search is only continued if the next block starts with that
996  * hash value.  This is used if we are searching for a specific file.
997  *
998  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
999  *
1000  * This function returns 1 if the caller should continue to search,
1001  * or 0 if it should not.  If there is an error reading one of the
1002  * index blocks, it will a negative error code.
1003  *
1004  * If start_hash is non-null, it will be filled in with the starting
1005  * hash of the next page.
1006  */
1007 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
1008 				 struct dx_frame *frame,
1009 				 struct dx_frame *frames,
1010 				 __u32 *start_hash)
1011 {
1012 	struct dx_frame *p;
1013 	struct buffer_head *bh;
1014 	int num_frames = 0;
1015 	__u32 bhash;
1016 
1017 	p = frame;
1018 	/*
1019 	 * Find the next leaf page by incrementing the frame pointer.
1020 	 * If we run out of entries in the interior node, loop around and
1021 	 * increment pointer in the parent node.  When we break out of
1022 	 * this loop, num_frames indicates the number of interior
1023 	 * nodes need to be read.
1024 	 */
1025 	while (1) {
1026 		if (++(p->at) < p->entries + dx_get_count(p->entries))
1027 			break;
1028 		if (p == frames)
1029 			return 0;
1030 		num_frames++;
1031 		p--;
1032 	}
1033 
1034 	/*
1035 	 * If the hash is 1, then continue only if the next page has a
1036 	 * continuation hash of any value.  This is used for readdir
1037 	 * handling.  Otherwise, check to see if the hash matches the
1038 	 * desired continuation hash.  If it doesn't, return since
1039 	 * there's no point to read in the successive index pages.
1040 	 */
1041 	bhash = dx_get_hash(p->at);
1042 	if (start_hash)
1043 		*start_hash = bhash;
1044 	if ((hash & 1) == 0) {
1045 		if ((bhash & ~1) != hash)
1046 			return 0;
1047 	}
1048 	/*
1049 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
1050 	 * block so no check is necessary
1051 	 */
1052 	while (num_frames--) {
1053 		bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
1054 		if (IS_ERR(bh))
1055 			return PTR_ERR(bh);
1056 		p++;
1057 		brelse(p->bh);
1058 		p->bh = bh;
1059 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1060 	}
1061 	return 1;
1062 }
1063 
1064 
1065 /*
1066  * This function fills a red-black tree with information from a
1067  * directory block.  It returns the number directory entries loaded
1068  * into the tree.  If there is an error it is returned in err.
1069  */
1070 static int htree_dirblock_to_tree(struct file *dir_file,
1071 				  struct inode *dir, ext4_lblk_t block,
1072 				  struct dx_hash_info *hinfo,
1073 				  __u32 start_hash, __u32 start_minor_hash)
1074 {
1075 	struct buffer_head *bh;
1076 	struct ext4_dir_entry_2 *de, *top;
1077 	int err = 0, count = 0;
1078 	struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1079 	int csum = ext4_has_metadata_csum(dir->i_sb);
1080 
1081 	dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1082 							(unsigned long)block));
1083 	bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1084 	if (IS_ERR(bh))
1085 		return PTR_ERR(bh);
1086 
1087 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1088 	/* csum entries are not larger in the casefolded encrypted case */
1089 	top = (struct ext4_dir_entry_2 *) ((char *) de +
1090 					   dir->i_sb->s_blocksize -
1091 					   ext4_dir_rec_len(0,
1092 							   csum ? NULL : dir));
1093 	/* Check if the directory is encrypted */
1094 	if (IS_ENCRYPTED(dir)) {
1095 		err = fscrypt_prepare_readdir(dir);
1096 		if (err < 0) {
1097 			brelse(bh);
1098 			return err;
1099 		}
1100 		err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1101 						 &fname_crypto_str);
1102 		if (err < 0) {
1103 			brelse(bh);
1104 			return err;
1105 		}
1106 	}
1107 
1108 	for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1109 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1110 				bh->b_data, bh->b_size,
1111 				(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1112 					 + ((char *)de - bh->b_data))) {
1113 			/* silently ignore the rest of the block */
1114 			break;
1115 		}
1116 		if (ext4_hash_in_dirent(dir)) {
1117 			if (de->name_len && de->inode) {
1118 				hinfo->hash = EXT4_DIRENT_HASH(de);
1119 				hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1120 			} else {
1121 				hinfo->hash = 0;
1122 				hinfo->minor_hash = 0;
1123 			}
1124 		} else {
1125 			err = ext4fs_dirhash(dir, de->name,
1126 					     de->name_len, hinfo);
1127 			if (err < 0) {
1128 				count = err;
1129 				goto errout;
1130 			}
1131 		}
1132 		if ((hinfo->hash < start_hash) ||
1133 		    ((hinfo->hash == start_hash) &&
1134 		     (hinfo->minor_hash < start_minor_hash)))
1135 			continue;
1136 		if (de->inode == 0)
1137 			continue;
1138 		if (!IS_ENCRYPTED(dir)) {
1139 			tmp_str.name = de->name;
1140 			tmp_str.len = de->name_len;
1141 			err = ext4_htree_store_dirent(dir_file,
1142 				   hinfo->hash, hinfo->minor_hash, de,
1143 				   &tmp_str);
1144 		} else {
1145 			int save_len = fname_crypto_str.len;
1146 			struct fscrypt_str de_name = FSTR_INIT(de->name,
1147 								de->name_len);
1148 
1149 			/* Directory is encrypted */
1150 			err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1151 					hinfo->minor_hash, &de_name,
1152 					&fname_crypto_str);
1153 			if (err) {
1154 				count = err;
1155 				goto errout;
1156 			}
1157 			err = ext4_htree_store_dirent(dir_file,
1158 				   hinfo->hash, hinfo->minor_hash, de,
1159 					&fname_crypto_str);
1160 			fname_crypto_str.len = save_len;
1161 		}
1162 		if (err != 0) {
1163 			count = err;
1164 			goto errout;
1165 		}
1166 		count++;
1167 	}
1168 errout:
1169 	brelse(bh);
1170 	fscrypt_fname_free_buffer(&fname_crypto_str);
1171 	return count;
1172 }
1173 
1174 
1175 /*
1176  * This function fills a red-black tree with information from a
1177  * directory.  We start scanning the directory in hash order, starting
1178  * at start_hash and start_minor_hash.
1179  *
1180  * This function returns the number of entries inserted into the tree,
1181  * or a negative error code.
1182  */
1183 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1184 			 __u32 start_minor_hash, __u32 *next_hash)
1185 {
1186 	struct dx_hash_info hinfo;
1187 	struct ext4_dir_entry_2 *de;
1188 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1189 	struct inode *dir;
1190 	ext4_lblk_t block;
1191 	int count = 0;
1192 	int ret, err;
1193 	__u32 hashval;
1194 	struct fscrypt_str tmp_str;
1195 
1196 	dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1197 		       start_hash, start_minor_hash));
1198 	dir = file_inode(dir_file);
1199 	if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1200 		if (ext4_hash_in_dirent(dir))
1201 			hinfo.hash_version = DX_HASH_SIPHASH;
1202 		else
1203 			hinfo.hash_version =
1204 					EXT4_SB(dir->i_sb)->s_def_hash_version;
1205 		if (hinfo.hash_version <= DX_HASH_TEA)
1206 			hinfo.hash_version +=
1207 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
1208 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1209 		if (ext4_has_inline_data(dir)) {
1210 			int has_inline_data = 1;
1211 			count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1212 						       &hinfo, start_hash,
1213 						       start_minor_hash,
1214 						       &has_inline_data);
1215 			if (has_inline_data) {
1216 				*next_hash = ~0;
1217 				return count;
1218 			}
1219 		}
1220 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1221 					       start_hash, start_minor_hash);
1222 		*next_hash = ~0;
1223 		return count;
1224 	}
1225 	hinfo.hash = start_hash;
1226 	hinfo.minor_hash = 0;
1227 	frame = dx_probe(NULL, dir, &hinfo, frames);
1228 	if (IS_ERR(frame))
1229 		return PTR_ERR(frame);
1230 
1231 	/* Add '.' and '..' from the htree header */
1232 	if (!start_hash && !start_minor_hash) {
1233 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1234 		tmp_str.name = de->name;
1235 		tmp_str.len = de->name_len;
1236 		err = ext4_htree_store_dirent(dir_file, 0, 0,
1237 					      de, &tmp_str);
1238 		if (err != 0)
1239 			goto errout;
1240 		count++;
1241 	}
1242 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1243 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1244 		de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1245 		tmp_str.name = de->name;
1246 		tmp_str.len = de->name_len;
1247 		err = ext4_htree_store_dirent(dir_file, 2, 0,
1248 					      de, &tmp_str);
1249 		if (err != 0)
1250 			goto errout;
1251 		count++;
1252 	}
1253 
1254 	while (1) {
1255 		if (fatal_signal_pending(current)) {
1256 			err = -ERESTARTSYS;
1257 			goto errout;
1258 		}
1259 		cond_resched();
1260 		block = dx_get_block(frame->at);
1261 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1262 					     start_hash, start_minor_hash);
1263 		if (ret < 0) {
1264 			err = ret;
1265 			goto errout;
1266 		}
1267 		count += ret;
1268 		hashval = ~0;
1269 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1270 					    frame, frames, &hashval);
1271 		*next_hash = hashval;
1272 		if (ret < 0) {
1273 			err = ret;
1274 			goto errout;
1275 		}
1276 		/*
1277 		 * Stop if:  (a) there are no more entries, or
1278 		 * (b) we have inserted at least one entry and the
1279 		 * next hash value is not a continuation
1280 		 */
1281 		if ((ret == 0) ||
1282 		    (count && ((hashval & 1) == 0)))
1283 			break;
1284 	}
1285 	dx_release(frames);
1286 	dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1287 		       "next hash: %x\n", count, *next_hash));
1288 	return count;
1289 errout:
1290 	dx_release(frames);
1291 	return (err);
1292 }
1293 
1294 static inline int search_dirblock(struct buffer_head *bh,
1295 				  struct inode *dir,
1296 				  struct ext4_filename *fname,
1297 				  unsigned int offset,
1298 				  struct ext4_dir_entry_2 **res_dir)
1299 {
1300 	return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1301 			       fname, offset, res_dir);
1302 }
1303 
1304 /*
1305  * Directory block splitting, compacting
1306  */
1307 
1308 /*
1309  * Create map of hash values, offsets, and sizes, stored at end of block.
1310  * Returns number of entries mapped.
1311  */
1312 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
1313 		       struct dx_hash_info *hinfo,
1314 		       struct dx_map_entry *map_tail)
1315 {
1316 	int count = 0;
1317 	struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)bh->b_data;
1318 	unsigned int buflen = bh->b_size;
1319 	char *base = bh->b_data;
1320 	struct dx_hash_info h = *hinfo;
1321 	int blocksize = EXT4_BLOCK_SIZE(dir->i_sb);
1322 
1323 	if (ext4_has_metadata_csum(dir->i_sb))
1324 		buflen -= sizeof(struct ext4_dir_entry_tail);
1325 
1326 	while ((char *) de < base + buflen) {
1327 		if (ext4_check_dir_entry(dir, NULL, de, bh, base, buflen,
1328 					 ((char *)de) - base))
1329 			return -EFSCORRUPTED;
1330 		if (de->name_len && de->inode) {
1331 			if (ext4_hash_in_dirent(dir))
1332 				h.hash = EXT4_DIRENT_HASH(de);
1333 			else {
1334 				int err = ext4fs_dirhash(dir, de->name,
1335 						     de->name_len, &h);
1336 				if (err < 0)
1337 					return err;
1338 			}
1339 			map_tail--;
1340 			map_tail->hash = h.hash;
1341 			map_tail->offs = ((char *) de - base)>>2;
1342 			map_tail->size = ext4_rec_len_from_disk(de->rec_len,
1343 								blocksize);
1344 			count++;
1345 			cond_resched();
1346 		}
1347 		de = ext4_next_entry(de, blocksize);
1348 	}
1349 	return count;
1350 }
1351 
1352 /* Sort map by hash value */
1353 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1354 {
1355 	struct dx_map_entry *p, *q, *top = map + count - 1;
1356 	int more;
1357 	/* Combsort until bubble sort doesn't suck */
1358 	while (count > 2) {
1359 		count = count*10/13;
1360 		if (count - 9 < 2) /* 9, 10 -> 11 */
1361 			count = 11;
1362 		for (p = top, q = p - count; q >= map; p--, q--)
1363 			if (p->hash < q->hash)
1364 				swap(*p, *q);
1365 	}
1366 	/* Garden variety bubble sort */
1367 	do {
1368 		more = 0;
1369 		q = top;
1370 		while (q-- > map) {
1371 			if (q[1].hash >= q[0].hash)
1372 				continue;
1373 			swap(*(q+1), *q);
1374 			more = 1;
1375 		}
1376 	} while(more);
1377 }
1378 
1379 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1380 {
1381 	struct dx_entry *entries = frame->entries;
1382 	struct dx_entry *old = frame->at, *new = old + 1;
1383 	int count = dx_get_count(entries);
1384 
1385 	ASSERT(count < dx_get_limit(entries));
1386 	ASSERT(old < entries + count);
1387 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1388 	dx_set_hash(new, hash);
1389 	dx_set_block(new, block);
1390 	dx_set_count(entries, count + 1);
1391 }
1392 
1393 #if IS_ENABLED(CONFIG_UNICODE)
1394 int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1395 				  struct ext4_filename *name)
1396 {
1397 	struct qstr *cf_name = &name->cf_name;
1398 	unsigned char *buf;
1399 	struct dx_hash_info *hinfo = &name->hinfo;
1400 	int len;
1401 
1402 	if (!IS_CASEFOLDED(dir) ||
1403 	    (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1404 		cf_name->name = NULL;
1405 		return 0;
1406 	}
1407 
1408 	buf = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1409 	if (!buf)
1410 		return -ENOMEM;
1411 
1412 	len = utf8_casefold(dir->i_sb->s_encoding, iname, buf, EXT4_NAME_LEN);
1413 	if (len <= 0) {
1414 		kfree(buf);
1415 		buf = NULL;
1416 	}
1417 	cf_name->name = buf;
1418 	cf_name->len = (unsigned) len;
1419 
1420 	if (!IS_ENCRYPTED(dir))
1421 		return 0;
1422 
1423 	hinfo->hash_version = DX_HASH_SIPHASH;
1424 	hinfo->seed = NULL;
1425 	if (cf_name->name)
1426 		return ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1427 	else
1428 		return ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1429 }
1430 #endif
1431 
1432 /*
1433  * Test whether a directory entry matches the filename being searched for.
1434  *
1435  * Return: %true if the directory entry matches, otherwise %false.
1436  */
1437 static bool ext4_match(struct inode *parent,
1438 			      const struct ext4_filename *fname,
1439 			      struct ext4_dir_entry_2 *de)
1440 {
1441 	struct fscrypt_name f;
1442 
1443 	if (!de->inode)
1444 		return false;
1445 
1446 	f.usr_fname = fname->usr_fname;
1447 	f.disk_name = fname->disk_name;
1448 #ifdef CONFIG_FS_ENCRYPTION
1449 	f.crypto_buf = fname->crypto_buf;
1450 #endif
1451 
1452 #if IS_ENABLED(CONFIG_UNICODE)
1453 	if (IS_CASEFOLDED(parent) &&
1454 	    (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1455 		/*
1456 		 * Just checking IS_ENCRYPTED(parent) below is not
1457 		 * sufficient to decide whether one can use the hash for
1458 		 * skipping the string comparison, because the key might
1459 		 * have been added right after
1460 		 * ext4_fname_setup_ci_filename().  In this case, a hash
1461 		 * mismatch will be a false negative.  Therefore, make
1462 		 * sure cf_name was properly initialized before
1463 		 * considering the calculated hash.
1464 		 */
1465 		if (IS_ENCRYPTED(parent) && fname->cf_name.name &&
1466 		    (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1467 		     fname->hinfo.minor_hash != EXT4_DIRENT_MINOR_HASH(de)))
1468 			return false;
1469 		/*
1470 		 * Treat comparison errors as not a match.  The
1471 		 * only case where it happens is on a disk
1472 		 * corruption or ENOMEM.
1473 		 */
1474 
1475 		return generic_ci_match(parent, fname->usr_fname,
1476 					&fname->cf_name, de->name,
1477 					de->name_len) > 0;
1478 	}
1479 #endif
1480 
1481 	return fscrypt_match_name(&f, de->name, de->name_len);
1482 }
1483 
1484 /*
1485  * Returns 0 if not found, -EFSCORRUPTED on failure, and 1 on success
1486  */
1487 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1488 		    struct inode *dir, struct ext4_filename *fname,
1489 		    unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1490 {
1491 	struct ext4_dir_entry_2 * de;
1492 	char * dlimit;
1493 	int de_len;
1494 
1495 	de = (struct ext4_dir_entry_2 *)search_buf;
1496 	dlimit = search_buf + buf_size;
1497 	while ((char *) de < dlimit - EXT4_BASE_DIR_LEN) {
1498 		/* this code is executed quadratically often */
1499 		/* do minimal checking `by hand' */
1500 		if (de->name + de->name_len <= dlimit &&
1501 		    ext4_match(dir, fname, de)) {
1502 			/* found a match - just to be sure, do
1503 			 * a full check */
1504 			if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1505 						 buf_size, offset))
1506 				return -EFSCORRUPTED;
1507 			*res_dir = de;
1508 			return 1;
1509 		}
1510 		/* prevent looping on a bad block */
1511 		de_len = ext4_rec_len_from_disk(de->rec_len,
1512 						dir->i_sb->s_blocksize);
1513 		if (de_len <= 0)
1514 			return -EFSCORRUPTED;
1515 		offset += de_len;
1516 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1517 	}
1518 	return 0;
1519 }
1520 
1521 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1522 			       struct ext4_dir_entry *de)
1523 {
1524 	struct super_block *sb = dir->i_sb;
1525 
1526 	if (!is_dx(dir))
1527 		return 0;
1528 	if (block == 0)
1529 		return 1;
1530 	if (de->inode == 0 &&
1531 	    ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1532 			sb->s_blocksize)
1533 		return 1;
1534 	return 0;
1535 }
1536 
1537 /*
1538  *	__ext4_find_entry()
1539  *
1540  * finds an entry in the specified directory with the wanted name. It
1541  * returns the cache buffer in which the entry was found, and the entry
1542  * itself (as a parameter - res_dir). It does NOT read the inode of the
1543  * entry - you'll have to do that yourself if you want to.
1544  *
1545  * The returned buffer_head has ->b_count elevated.  The caller is expected
1546  * to brelse() it when appropriate.
1547  */
1548 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1549 					     struct ext4_filename *fname,
1550 					     struct ext4_dir_entry_2 **res_dir,
1551 					     int *inlined)
1552 {
1553 	struct super_block *sb;
1554 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1555 	struct buffer_head *bh, *ret = NULL;
1556 	ext4_lblk_t start, block;
1557 	const u8 *name = fname->usr_fname->name;
1558 	size_t ra_max = 0;	/* Number of bh's in the readahead
1559 				   buffer, bh_use[] */
1560 	size_t ra_ptr = 0;	/* Current index into readahead
1561 				   buffer */
1562 	ext4_lblk_t  nblocks;
1563 	int i, namelen, retval;
1564 
1565 	*res_dir = NULL;
1566 	sb = dir->i_sb;
1567 	namelen = fname->usr_fname->len;
1568 	if (namelen > EXT4_NAME_LEN)
1569 		return NULL;
1570 
1571 	if (ext4_has_inline_data(dir)) {
1572 		int has_inline_data = 1;
1573 		ret = ext4_find_inline_entry(dir, fname, res_dir,
1574 					     &has_inline_data);
1575 		if (inlined)
1576 			*inlined = has_inline_data;
1577 		if (has_inline_data || IS_ERR(ret))
1578 			goto cleanup_and_exit;
1579 	}
1580 
1581 	if ((namelen <= 2) && (name[0] == '.') &&
1582 	    (name[1] == '.' || name[1] == '\0')) {
1583 		/*
1584 		 * "." or ".." will only be in the first block
1585 		 * NFS may look up ".."; "." should be handled by the VFS
1586 		 */
1587 		block = start = 0;
1588 		nblocks = 1;
1589 		goto restart;
1590 	}
1591 	if (is_dx(dir)) {
1592 		ret = ext4_dx_find_entry(dir, fname, res_dir);
1593 		/*
1594 		 * On success, or if the error was file not found,
1595 		 * return.  Otherwise, fall back to doing a search the
1596 		 * old fashioned way.
1597 		 */
1598 		if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1599 			goto cleanup_and_exit;
1600 		dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1601 			       "falling back\n"));
1602 		ret = NULL;
1603 	}
1604 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1605 	if (!nblocks) {
1606 		ret = NULL;
1607 		goto cleanup_and_exit;
1608 	}
1609 	start = EXT4_I(dir)->i_dir_start_lookup;
1610 	if (start >= nblocks)
1611 		start = 0;
1612 	block = start;
1613 restart:
1614 	do {
1615 		/*
1616 		 * We deal with the read-ahead logic here.
1617 		 */
1618 		cond_resched();
1619 		if (ra_ptr >= ra_max) {
1620 			/* Refill the readahead buffer */
1621 			ra_ptr = 0;
1622 			if (block < start)
1623 				ra_max = start - block;
1624 			else
1625 				ra_max = nblocks - block;
1626 			ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1627 			retval = ext4_bread_batch(dir, block, ra_max,
1628 						  false /* wait */, bh_use);
1629 			if (retval) {
1630 				ret = ERR_PTR(retval);
1631 				ra_max = 0;
1632 				goto cleanup_and_exit;
1633 			}
1634 		}
1635 		if ((bh = bh_use[ra_ptr++]) == NULL)
1636 			goto next;
1637 		wait_on_buffer(bh);
1638 		if (!buffer_uptodate(bh)) {
1639 			EXT4_ERROR_INODE_ERR(dir, EIO,
1640 					     "reading directory lblock %lu",
1641 					     (unsigned long) block);
1642 			brelse(bh);
1643 			ret = ERR_PTR(-EIO);
1644 			goto cleanup_and_exit;
1645 		}
1646 		if (!buffer_verified(bh) &&
1647 		    !is_dx_internal_node(dir, block,
1648 					 (struct ext4_dir_entry *)bh->b_data) &&
1649 		    !ext4_dirblock_csum_verify(dir, bh)) {
1650 			EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1651 					     "checksumming directory "
1652 					     "block %lu", (unsigned long)block);
1653 			brelse(bh);
1654 			ret = ERR_PTR(-EFSBADCRC);
1655 			goto cleanup_and_exit;
1656 		}
1657 		set_buffer_verified(bh);
1658 		i = search_dirblock(bh, dir, fname,
1659 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1660 		if (i == 1) {
1661 			EXT4_I(dir)->i_dir_start_lookup = block;
1662 			ret = bh;
1663 			goto cleanup_and_exit;
1664 		} else {
1665 			brelse(bh);
1666 			if (i < 0) {
1667 				ret = ERR_PTR(i);
1668 				goto cleanup_and_exit;
1669 			}
1670 		}
1671 	next:
1672 		if (++block >= nblocks)
1673 			block = 0;
1674 	} while (block != start);
1675 
1676 	/*
1677 	 * If the directory has grown while we were searching, then
1678 	 * search the last part of the directory before giving up.
1679 	 */
1680 	block = nblocks;
1681 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1682 	if (block < nblocks) {
1683 		start = 0;
1684 		goto restart;
1685 	}
1686 
1687 cleanup_and_exit:
1688 	/* Clean up the read-ahead blocks */
1689 	for (; ra_ptr < ra_max; ra_ptr++)
1690 		brelse(bh_use[ra_ptr]);
1691 	return ret;
1692 }
1693 
1694 static struct buffer_head *ext4_find_entry(struct inode *dir,
1695 					   const struct qstr *d_name,
1696 					   struct ext4_dir_entry_2 **res_dir,
1697 					   int *inlined)
1698 {
1699 	int err;
1700 	struct ext4_filename fname;
1701 	struct buffer_head *bh;
1702 
1703 	err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1704 	if (err == -ENOENT)
1705 		return NULL;
1706 	if (err)
1707 		return ERR_PTR(err);
1708 
1709 	bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1710 
1711 	ext4_fname_free_filename(&fname);
1712 	return bh;
1713 }
1714 
1715 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1716 					     struct dentry *dentry,
1717 					     struct ext4_dir_entry_2 **res_dir)
1718 {
1719 	int err;
1720 	struct ext4_filename fname;
1721 	struct buffer_head *bh;
1722 
1723 	err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1724 	if (err == -ENOENT)
1725 		return NULL;
1726 	if (err)
1727 		return ERR_PTR(err);
1728 
1729 	bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1730 
1731 	ext4_fname_free_filename(&fname);
1732 	return bh;
1733 }
1734 
1735 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1736 			struct ext4_filename *fname,
1737 			struct ext4_dir_entry_2 **res_dir)
1738 {
1739 	struct super_block * sb = dir->i_sb;
1740 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1741 	struct buffer_head *bh;
1742 	ext4_lblk_t block;
1743 	int retval;
1744 
1745 #ifdef CONFIG_FS_ENCRYPTION
1746 	*res_dir = NULL;
1747 #endif
1748 	frame = dx_probe(fname, dir, NULL, frames);
1749 	if (IS_ERR(frame))
1750 		return (struct buffer_head *) frame;
1751 	do {
1752 		block = dx_get_block(frame->at);
1753 		bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1754 		if (IS_ERR(bh))
1755 			goto errout;
1756 
1757 		retval = search_dirblock(bh, dir, fname,
1758 					 block << EXT4_BLOCK_SIZE_BITS(sb),
1759 					 res_dir);
1760 		if (retval == 1)
1761 			goto success;
1762 		brelse(bh);
1763 		if (retval < 0) {
1764 			bh = ERR_PTR(ERR_BAD_DX_DIR);
1765 			goto errout;
1766 		}
1767 
1768 		/* Check to see if we should continue to search */
1769 		retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1770 					       frames, NULL);
1771 		if (retval < 0) {
1772 			ext4_warning_inode(dir,
1773 				"error %d reading directory index block",
1774 				retval);
1775 			bh = ERR_PTR(retval);
1776 			goto errout;
1777 		}
1778 	} while (retval == 1);
1779 
1780 	bh = NULL;
1781 errout:
1782 	dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1783 success:
1784 	dx_release(frames);
1785 	return bh;
1786 }
1787 
1788 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1789 {
1790 	struct inode *inode;
1791 	struct ext4_dir_entry_2 *de;
1792 	struct buffer_head *bh;
1793 
1794 	if (dentry->d_name.len > EXT4_NAME_LEN)
1795 		return ERR_PTR(-ENAMETOOLONG);
1796 
1797 	bh = ext4_lookup_entry(dir, dentry, &de);
1798 	if (IS_ERR(bh))
1799 		return ERR_CAST(bh);
1800 	inode = NULL;
1801 	if (bh) {
1802 		__u32 ino = le32_to_cpu(de->inode);
1803 		brelse(bh);
1804 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1805 			EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1806 			return ERR_PTR(-EFSCORRUPTED);
1807 		}
1808 		if (unlikely(ino == dir->i_ino)) {
1809 			EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1810 					 dentry);
1811 			return ERR_PTR(-EFSCORRUPTED);
1812 		}
1813 		inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1814 		if (inode == ERR_PTR(-ESTALE)) {
1815 			EXT4_ERROR_INODE(dir,
1816 					 "deleted inode referenced: %u",
1817 					 ino);
1818 			return ERR_PTR(-EFSCORRUPTED);
1819 		}
1820 		if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1821 		    (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1822 		    !fscrypt_has_permitted_context(dir, inode)) {
1823 			ext4_warning(inode->i_sb,
1824 				     "Inconsistent encryption contexts: %lu/%lu",
1825 				     dir->i_ino, inode->i_ino);
1826 			iput(inode);
1827 			return ERR_PTR(-EPERM);
1828 		}
1829 	}
1830 
1831 	if (IS_ENABLED(CONFIG_UNICODE) && !inode && IS_CASEFOLDED(dir)) {
1832 		/* Eventually we want to call d_add_ci(dentry, NULL)
1833 		 * for negative dentries in the encoding case as
1834 		 * well.  For now, prevent the negative dentry
1835 		 * from being cached.
1836 		 */
1837 		return NULL;
1838 	}
1839 
1840 	return d_splice_alias(inode, dentry);
1841 }
1842 
1843 
1844 struct dentry *ext4_get_parent(struct dentry *child)
1845 {
1846 	__u32 ino;
1847 	struct ext4_dir_entry_2 * de;
1848 	struct buffer_head *bh;
1849 
1850 	bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1851 	if (IS_ERR(bh))
1852 		return ERR_CAST(bh);
1853 	if (!bh)
1854 		return ERR_PTR(-ENOENT);
1855 	ino = le32_to_cpu(de->inode);
1856 	brelse(bh);
1857 
1858 	if (!ext4_valid_inum(child->d_sb, ino)) {
1859 		EXT4_ERROR_INODE(d_inode(child),
1860 				 "bad parent inode number: %u", ino);
1861 		return ERR_PTR(-EFSCORRUPTED);
1862 	}
1863 
1864 	return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1865 }
1866 
1867 /*
1868  * Move count entries from end of map between two memory locations.
1869  * Returns pointer to last entry moved.
1870  */
1871 static struct ext4_dir_entry_2 *
1872 dx_move_dirents(struct inode *dir, char *from, char *to,
1873 		struct dx_map_entry *map, int count,
1874 		unsigned blocksize)
1875 {
1876 	unsigned rec_len = 0;
1877 
1878 	while (count--) {
1879 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1880 						(from + (map->offs<<2));
1881 		rec_len = ext4_dir_rec_len(de->name_len, dir);
1882 
1883 		memcpy (to, de, rec_len);
1884 		((struct ext4_dir_entry_2 *) to)->rec_len =
1885 				ext4_rec_len_to_disk(rec_len, blocksize);
1886 
1887 		/* wipe dir_entry excluding the rec_len field */
1888 		de->inode = 0;
1889 		memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1890 								blocksize) -
1891 					 offsetof(struct ext4_dir_entry_2,
1892 								name_len));
1893 
1894 		map++;
1895 		to += rec_len;
1896 	}
1897 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1898 }
1899 
1900 /*
1901  * Compact each dir entry in the range to the minimal rec_len.
1902  * Returns pointer to last entry in range.
1903  */
1904 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1905 							unsigned int blocksize)
1906 {
1907 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1908 	unsigned rec_len = 0;
1909 
1910 	prev = to = de;
1911 	while ((char*)de < base + blocksize) {
1912 		next = ext4_next_entry(de, blocksize);
1913 		if (de->inode && de->name_len) {
1914 			rec_len = ext4_dir_rec_len(de->name_len, dir);
1915 			if (de > to)
1916 				memmove(to, de, rec_len);
1917 			to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1918 			prev = to;
1919 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1920 		}
1921 		de = next;
1922 	}
1923 	return prev;
1924 }
1925 
1926 /*
1927  * Split a full leaf block to make room for a new dir entry.
1928  * Allocate a new block, and move entries so that they are approx. equally full.
1929  * Returns pointer to de in block into which the new entry will be inserted.
1930  */
1931 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1932 			struct buffer_head **bh,struct dx_frame *frame,
1933 			struct dx_hash_info *hinfo)
1934 {
1935 	unsigned blocksize = dir->i_sb->s_blocksize;
1936 	unsigned continued;
1937 	int count;
1938 	struct buffer_head *bh2;
1939 	ext4_lblk_t newblock;
1940 	u32 hash2;
1941 	struct dx_map_entry *map;
1942 	char *data1 = (*bh)->b_data, *data2;
1943 	unsigned split, move, size;
1944 	struct ext4_dir_entry_2 *de = NULL, *de2;
1945 	int	csum_size = 0;
1946 	int	err = 0, i;
1947 
1948 	if (ext4_has_metadata_csum(dir->i_sb))
1949 		csum_size = sizeof(struct ext4_dir_entry_tail);
1950 
1951 	bh2 = ext4_append(handle, dir, &newblock);
1952 	if (IS_ERR(bh2)) {
1953 		brelse(*bh);
1954 		*bh = NULL;
1955 		return (struct ext4_dir_entry_2 *) bh2;
1956 	}
1957 
1958 	BUFFER_TRACE(*bh, "get_write_access");
1959 	err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
1960 					    EXT4_JTR_NONE);
1961 	if (err)
1962 		goto journal_error;
1963 
1964 	BUFFER_TRACE(frame->bh, "get_write_access");
1965 	err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
1966 					    EXT4_JTR_NONE);
1967 	if (err)
1968 		goto journal_error;
1969 
1970 	data2 = bh2->b_data;
1971 
1972 	/* create map in the end of data2 block */
1973 	map = (struct dx_map_entry *) (data2 + blocksize);
1974 	count = dx_make_map(dir, *bh, hinfo, map);
1975 	if (count < 0) {
1976 		err = count;
1977 		goto journal_error;
1978 	}
1979 	map -= count;
1980 	dx_sort_map(map, count);
1981 	/* Ensure that neither split block is over half full */
1982 	size = 0;
1983 	move = 0;
1984 	for (i = count-1; i >= 0; i--) {
1985 		/* is more than half of this entry in 2nd half of the block? */
1986 		if (size + map[i].size/2 > blocksize/2)
1987 			break;
1988 		size += map[i].size;
1989 		move++;
1990 	}
1991 	/*
1992 	 * map index at which we will split
1993 	 *
1994 	 * If the sum of active entries didn't exceed half the block size, just
1995 	 * split it in half by count; each resulting block will have at least
1996 	 * half the space free.
1997 	 */
1998 	if (i > 0)
1999 		split = count - move;
2000 	else
2001 		split = count/2;
2002 
2003 	hash2 = map[split].hash;
2004 	continued = split > 0 ? hash2 == map[split - 1].hash : 0;
2005 	dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
2006 			(unsigned long)dx_get_block(frame->at),
2007 					hash2, split, count-split));
2008 
2009 	/* Fancy dance to stay within two buffers */
2010 	de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
2011 			      blocksize);
2012 	de = dx_pack_dirents(dir, data1, blocksize);
2013 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2014 					   (char *) de,
2015 					   blocksize);
2016 	de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2017 					    (char *) de2,
2018 					    blocksize);
2019 	if (csum_size) {
2020 		ext4_initialize_dirent_tail(*bh, blocksize);
2021 		ext4_initialize_dirent_tail(bh2, blocksize);
2022 	}
2023 
2024 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
2025 			blocksize, 1));
2026 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
2027 			blocksize, 1));
2028 
2029 	/* Which block gets the new entry? */
2030 	if (hinfo->hash >= hash2) {
2031 		swap(*bh, bh2);
2032 		de = de2;
2033 	}
2034 	dx_insert_block(frame, hash2 + continued, newblock);
2035 	err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2036 	if (err)
2037 		goto journal_error;
2038 	err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2039 	if (err)
2040 		goto journal_error;
2041 	brelse(bh2);
2042 	dxtrace(dx_show_index("frame", frame->entries));
2043 	return de;
2044 
2045 journal_error:
2046 	brelse(*bh);
2047 	brelse(bh2);
2048 	*bh = NULL;
2049 	ext4_std_error(dir->i_sb, err);
2050 	return ERR_PTR(err);
2051 }
2052 
2053 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2054 		      struct buffer_head *bh,
2055 		      void *buf, int buf_size,
2056 		      struct ext4_filename *fname,
2057 		      struct ext4_dir_entry_2 **dest_de)
2058 {
2059 	struct ext4_dir_entry_2 *de;
2060 	unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2061 	int nlen, rlen;
2062 	unsigned int offset = 0;
2063 	char *top;
2064 
2065 	de = buf;
2066 	top = buf + buf_size - reclen;
2067 	while ((char *) de <= top) {
2068 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2069 					 buf, buf_size, offset))
2070 			return -EFSCORRUPTED;
2071 		if (ext4_match(dir, fname, de))
2072 			return -EEXIST;
2073 		nlen = ext4_dir_rec_len(de->name_len, dir);
2074 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2075 		if ((de->inode ? rlen - nlen : rlen) >= reclen)
2076 			break;
2077 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2078 		offset += rlen;
2079 	}
2080 	if ((char *) de > top)
2081 		return -ENOSPC;
2082 
2083 	*dest_de = de;
2084 	return 0;
2085 }
2086 
2087 void ext4_insert_dentry(struct inode *dir,
2088 			struct inode *inode,
2089 			struct ext4_dir_entry_2 *de,
2090 			int buf_size,
2091 			struct ext4_filename *fname)
2092 {
2093 
2094 	int nlen, rlen;
2095 
2096 	nlen = ext4_dir_rec_len(de->name_len, dir);
2097 	rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2098 	if (de->inode) {
2099 		struct ext4_dir_entry_2 *de1 =
2100 			(struct ext4_dir_entry_2 *)((char *)de + nlen);
2101 		de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2102 		de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2103 		de = de1;
2104 	}
2105 	de->file_type = EXT4_FT_UNKNOWN;
2106 	de->inode = cpu_to_le32(inode->i_ino);
2107 	ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2108 	de->name_len = fname_len(fname);
2109 	memcpy(de->name, fname_name(fname), fname_len(fname));
2110 	if (ext4_hash_in_dirent(dir)) {
2111 		struct dx_hash_info *hinfo = &fname->hinfo;
2112 
2113 		EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2114 		EXT4_DIRENT_HASHES(de)->minor_hash =
2115 						cpu_to_le32(hinfo->minor_hash);
2116 	}
2117 }
2118 
2119 /*
2120  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
2121  * it points to a directory entry which is guaranteed to be large
2122  * enough for new directory entry.  If de is NULL, then
2123  * add_dirent_to_buf will attempt search the directory block for
2124  * space.  It will return -ENOSPC if no space is available, and -EIO
2125  * and -EEXIST if directory entry already exists.
2126  */
2127 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2128 			     struct inode *dir,
2129 			     struct inode *inode, struct ext4_dir_entry_2 *de,
2130 			     struct buffer_head *bh)
2131 {
2132 	unsigned int	blocksize = dir->i_sb->s_blocksize;
2133 	int		csum_size = 0;
2134 	int		err, err2;
2135 
2136 	if (ext4_has_metadata_csum(inode->i_sb))
2137 		csum_size = sizeof(struct ext4_dir_entry_tail);
2138 
2139 	if (!de) {
2140 		err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2141 					blocksize - csum_size, fname, &de);
2142 		if (err)
2143 			return err;
2144 	}
2145 	BUFFER_TRACE(bh, "get_write_access");
2146 	err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2147 					    EXT4_JTR_NONE);
2148 	if (err) {
2149 		ext4_std_error(dir->i_sb, err);
2150 		return err;
2151 	}
2152 
2153 	/* By now the buffer is marked for journaling */
2154 	ext4_insert_dentry(dir, inode, de, blocksize, fname);
2155 
2156 	/*
2157 	 * XXX shouldn't update any times until successful
2158 	 * completion of syscall, but too many callers depend
2159 	 * on this.
2160 	 *
2161 	 * XXX similarly, too many callers depend on
2162 	 * ext4_new_inode() setting the times, but error
2163 	 * recovery deletes the inode, so the worst that can
2164 	 * happen is that the times are slightly out of date
2165 	 * and/or different from the directory change time.
2166 	 */
2167 	inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
2168 	ext4_update_dx_flag(dir);
2169 	inode_inc_iversion(dir);
2170 	err2 = ext4_mark_inode_dirty(handle, dir);
2171 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2172 	err = ext4_handle_dirty_dirblock(handle, dir, bh);
2173 	if (err)
2174 		ext4_std_error(dir->i_sb, err);
2175 	return err ? err : err2;
2176 }
2177 
2178 static bool ext4_check_dx_root(struct inode *dir, struct dx_root *root)
2179 {
2180 	struct fake_dirent *fde;
2181 	const char *error_msg;
2182 	unsigned int rlen;
2183 	unsigned int blocksize = dir->i_sb->s_blocksize;
2184 	char *blockend = (char *)root + dir->i_sb->s_blocksize;
2185 
2186 	fde = &root->dot;
2187 	if (unlikely(fde->name_len != 1)) {
2188 		error_msg = "invalid name_len for '.'";
2189 		goto corrupted;
2190 	}
2191 	if (unlikely(strncmp(root->dot_name, ".", fde->name_len))) {
2192 		error_msg = "invalid name for '.'";
2193 		goto corrupted;
2194 	}
2195 	rlen = ext4_rec_len_from_disk(fde->rec_len, blocksize);
2196 	if (unlikely((char *)fde + rlen >= blockend)) {
2197 		error_msg = "invalid rec_len for '.'";
2198 		goto corrupted;
2199 	}
2200 
2201 	fde = &root->dotdot;
2202 	if (unlikely(fde->name_len != 2)) {
2203 		error_msg = "invalid name_len for '..'";
2204 		goto corrupted;
2205 	}
2206 	if (unlikely(strncmp(root->dotdot_name, "..", fde->name_len))) {
2207 		error_msg = "invalid name for '..'";
2208 		goto corrupted;
2209 	}
2210 	rlen = ext4_rec_len_from_disk(fde->rec_len, blocksize);
2211 	if (unlikely((char *)fde + rlen >= blockend)) {
2212 		error_msg = "invalid rec_len for '..'";
2213 		goto corrupted;
2214 	}
2215 
2216 	return true;
2217 
2218 corrupted:
2219 	EXT4_ERROR_INODE(dir, "Corrupt dir, %s, running e2fsck is recommended",
2220 			 error_msg);
2221 	return false;
2222 }
2223 
2224 /*
2225  * This converts a one block unindexed directory to a 3 block indexed
2226  * directory, and adds the dentry to the indexed directory.
2227  */
2228 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2229 			    struct inode *dir,
2230 			    struct inode *inode, struct buffer_head *bh)
2231 {
2232 	struct buffer_head *bh2;
2233 	struct dx_root	*root;
2234 	struct dx_frame	frames[EXT4_HTREE_LEVEL], *frame;
2235 	struct dx_entry *entries;
2236 	struct ext4_dir_entry_2	*de, *de2;
2237 	char		*data2, *top;
2238 	unsigned	len;
2239 	int		retval;
2240 	unsigned	blocksize;
2241 	ext4_lblk_t  block;
2242 	struct fake_dirent *fde;
2243 	int csum_size = 0;
2244 
2245 	if (ext4_has_metadata_csum(inode->i_sb))
2246 		csum_size = sizeof(struct ext4_dir_entry_tail);
2247 
2248 	blocksize =  dir->i_sb->s_blocksize;
2249 	dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2250 	BUFFER_TRACE(bh, "get_write_access");
2251 	retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2252 					       EXT4_JTR_NONE);
2253 	if (retval) {
2254 		ext4_std_error(dir->i_sb, retval);
2255 		brelse(bh);
2256 		return retval;
2257 	}
2258 
2259 	root = (struct dx_root *) bh->b_data;
2260 	if (!ext4_check_dx_root(dir, root)) {
2261 		brelse(bh);
2262 		return -EFSCORRUPTED;
2263 	}
2264 
2265 	/* The 0th block becomes the root, move the dirents out */
2266 	fde = &root->dotdot;
2267 	de = (struct ext4_dir_entry_2 *)((char *)fde +
2268 		ext4_rec_len_from_disk(fde->rec_len, blocksize));
2269 	len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2270 
2271 	/* Allocate new block for the 0th block's dirents */
2272 	bh2 = ext4_append(handle, dir, &block);
2273 	if (IS_ERR(bh2)) {
2274 		brelse(bh);
2275 		return PTR_ERR(bh2);
2276 	}
2277 	ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2278 	data2 = bh2->b_data;
2279 
2280 	memcpy(data2, de, len);
2281 	memset(de, 0, len); /* wipe old data */
2282 	de = (struct ext4_dir_entry_2 *) data2;
2283 	top = data2 + len;
2284 	while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) {
2285 		if (ext4_check_dir_entry(dir, NULL, de, bh2, data2, len,
2286 					(char *)de - data2)) {
2287 			brelse(bh2);
2288 			brelse(bh);
2289 			return -EFSCORRUPTED;
2290 		}
2291 		de = de2;
2292 	}
2293 	de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2294 					   (char *) de, blocksize);
2295 
2296 	if (csum_size)
2297 		ext4_initialize_dirent_tail(bh2, blocksize);
2298 
2299 	/* Initialize the root; the dot dirents already exist */
2300 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2301 	de->rec_len = ext4_rec_len_to_disk(
2302 			blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2303 	memset (&root->info, 0, sizeof(root->info));
2304 	root->info.info_length = sizeof(root->info);
2305 	if (ext4_hash_in_dirent(dir))
2306 		root->info.hash_version = DX_HASH_SIPHASH;
2307 	else
2308 		root->info.hash_version =
2309 				EXT4_SB(dir->i_sb)->s_def_hash_version;
2310 
2311 	entries = root->entries;
2312 	dx_set_block(entries, 1);
2313 	dx_set_count(entries, 1);
2314 	dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2315 
2316 	/* Initialize as for dx_probe */
2317 	fname->hinfo.hash_version = root->info.hash_version;
2318 	if (fname->hinfo.hash_version <= DX_HASH_TEA)
2319 		fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2320 	fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2321 
2322 	/* casefolded encrypted hashes are computed on fname setup */
2323 	if (!ext4_hash_in_dirent(dir)) {
2324 		int err = ext4fs_dirhash(dir, fname_name(fname),
2325 					 fname_len(fname), &fname->hinfo);
2326 		if (err < 0) {
2327 			brelse(bh2);
2328 			brelse(bh);
2329 			return err;
2330 		}
2331 	}
2332 	memset(frames, 0, sizeof(frames));
2333 	frame = frames;
2334 	frame->entries = entries;
2335 	frame->at = entries;
2336 	frame->bh = bh;
2337 
2338 	retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2339 	if (retval)
2340 		goto out_frames;
2341 	retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2342 	if (retval)
2343 		goto out_frames;
2344 
2345 	de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2346 	if (IS_ERR(de)) {
2347 		retval = PTR_ERR(de);
2348 		goto out_frames;
2349 	}
2350 
2351 	retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2352 out_frames:
2353 	/*
2354 	 * Even if the block split failed, we have to properly write
2355 	 * out all the changes we did so far. Otherwise we can end up
2356 	 * with corrupted filesystem.
2357 	 */
2358 	if (retval)
2359 		ext4_mark_inode_dirty(handle, dir);
2360 	dx_release(frames);
2361 	brelse(bh2);
2362 	return retval;
2363 }
2364 
2365 /*
2366  *	ext4_add_entry()
2367  *
2368  * adds a file entry to the specified directory, using the same
2369  * semantics as ext4_find_entry(). It returns NULL if it failed.
2370  *
2371  * NOTE!! The inode part of 'de' is left at 0 - which means you
2372  * may not sleep between calling this and putting something into
2373  * the entry, as someone else might have used it while you slept.
2374  */
2375 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2376 			  struct inode *inode)
2377 {
2378 	struct inode *dir = d_inode(dentry->d_parent);
2379 	struct buffer_head *bh = NULL;
2380 	struct ext4_dir_entry_2 *de;
2381 	struct super_block *sb;
2382 	struct ext4_filename fname;
2383 	int	retval;
2384 	int	dx_fallback=0;
2385 	unsigned blocksize;
2386 	ext4_lblk_t block, blocks;
2387 	int	csum_size = 0;
2388 
2389 	if (ext4_has_metadata_csum(inode->i_sb))
2390 		csum_size = sizeof(struct ext4_dir_entry_tail);
2391 
2392 	sb = dir->i_sb;
2393 	blocksize = sb->s_blocksize;
2394 
2395 	if (fscrypt_is_nokey_name(dentry))
2396 		return -ENOKEY;
2397 
2398 #if IS_ENABLED(CONFIG_UNICODE)
2399 	if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2400 	    utf8_validate(sb->s_encoding, &dentry->d_name))
2401 		return -EINVAL;
2402 #endif
2403 
2404 	retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2405 	if (retval)
2406 		return retval;
2407 
2408 	if (ext4_has_inline_data(dir)) {
2409 		retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2410 		if (retval < 0)
2411 			goto out;
2412 		if (retval == 1) {
2413 			retval = 0;
2414 			goto out;
2415 		}
2416 	}
2417 
2418 	if (is_dx(dir)) {
2419 		retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2420 		if (!retval || (retval != ERR_BAD_DX_DIR))
2421 			goto out;
2422 		/* Can we just ignore htree data? */
2423 		if (ext4_has_metadata_csum(sb)) {
2424 			EXT4_ERROR_INODE(dir,
2425 				"Directory has corrupted htree index.");
2426 			retval = -EFSCORRUPTED;
2427 			goto out;
2428 		}
2429 		ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2430 		dx_fallback++;
2431 		retval = ext4_mark_inode_dirty(handle, dir);
2432 		if (unlikely(retval))
2433 			goto out;
2434 	}
2435 	blocks = dir->i_size >> sb->s_blocksize_bits;
2436 	for (block = 0; block < blocks; block++) {
2437 		bh = ext4_read_dirblock(dir, block, DIRENT);
2438 		if (bh == NULL) {
2439 			bh = ext4_bread(handle, dir, block,
2440 					EXT4_GET_BLOCKS_CREATE);
2441 			goto add_to_new_block;
2442 		}
2443 		if (IS_ERR(bh)) {
2444 			retval = PTR_ERR(bh);
2445 			bh = NULL;
2446 			goto out;
2447 		}
2448 		retval = add_dirent_to_buf(handle, &fname, dir, inode,
2449 					   NULL, bh);
2450 		if (retval != -ENOSPC)
2451 			goto out;
2452 
2453 		if (blocks == 1 && !dx_fallback &&
2454 		    ext4_has_feature_dir_index(sb)) {
2455 			retval = make_indexed_dir(handle, &fname, dir,
2456 						  inode, bh);
2457 			bh = NULL; /* make_indexed_dir releases bh */
2458 			goto out;
2459 		}
2460 		brelse(bh);
2461 	}
2462 	bh = ext4_append(handle, dir, &block);
2463 add_to_new_block:
2464 	if (IS_ERR(bh)) {
2465 		retval = PTR_ERR(bh);
2466 		bh = NULL;
2467 		goto out;
2468 	}
2469 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2470 	de->inode = 0;
2471 	de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2472 
2473 	if (csum_size)
2474 		ext4_initialize_dirent_tail(bh, blocksize);
2475 
2476 	retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2477 out:
2478 	ext4_fname_free_filename(&fname);
2479 	brelse(bh);
2480 	if (retval == 0)
2481 		ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2482 	return retval;
2483 }
2484 
2485 /*
2486  * Returns 0 for success, or a negative error value
2487  */
2488 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2489 			     struct inode *dir, struct inode *inode)
2490 {
2491 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2492 	struct dx_entry *entries, *at;
2493 	struct buffer_head *bh;
2494 	struct super_block *sb = dir->i_sb;
2495 	struct ext4_dir_entry_2 *de;
2496 	int restart;
2497 	int err;
2498 
2499 again:
2500 	restart = 0;
2501 	frame = dx_probe(fname, dir, NULL, frames);
2502 	if (IS_ERR(frame))
2503 		return PTR_ERR(frame);
2504 	entries = frame->entries;
2505 	at = frame->at;
2506 	bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2507 	if (IS_ERR(bh)) {
2508 		err = PTR_ERR(bh);
2509 		bh = NULL;
2510 		goto cleanup;
2511 	}
2512 
2513 	BUFFER_TRACE(bh, "get_write_access");
2514 	err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
2515 	if (err)
2516 		goto journal_error;
2517 
2518 	err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2519 	if (err != -ENOSPC)
2520 		goto cleanup;
2521 
2522 	err = 0;
2523 	/* Block full, should compress but for now just split */
2524 	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2525 		       dx_get_count(entries), dx_get_limit(entries)));
2526 	/* Need to split index? */
2527 	if (dx_get_count(entries) == dx_get_limit(entries)) {
2528 		ext4_lblk_t newblock;
2529 		int levels = frame - frames + 1;
2530 		unsigned int icount;
2531 		int add_level = 1;
2532 		struct dx_entry *entries2;
2533 		struct dx_node *node2;
2534 		struct buffer_head *bh2;
2535 
2536 		while (frame > frames) {
2537 			if (dx_get_count((frame - 1)->entries) <
2538 			    dx_get_limit((frame - 1)->entries)) {
2539 				add_level = 0;
2540 				break;
2541 			}
2542 			frame--; /* split higher index block */
2543 			at = frame->at;
2544 			entries = frame->entries;
2545 			restart = 1;
2546 		}
2547 		if (add_level && levels == ext4_dir_htree_level(sb)) {
2548 			ext4_warning(sb, "Directory (ino: %lu) index full, "
2549 					 "reach max htree level :%d",
2550 					 dir->i_ino, levels);
2551 			if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2552 				ext4_warning(sb, "Large directory feature is "
2553 						 "not enabled on this "
2554 						 "filesystem");
2555 			}
2556 			err = -ENOSPC;
2557 			goto cleanup;
2558 		}
2559 		icount = dx_get_count(entries);
2560 		bh2 = ext4_append(handle, dir, &newblock);
2561 		if (IS_ERR(bh2)) {
2562 			err = PTR_ERR(bh2);
2563 			goto cleanup;
2564 		}
2565 		node2 = (struct dx_node *)(bh2->b_data);
2566 		entries2 = node2->entries;
2567 		memset(&node2->fake, 0, sizeof(struct fake_dirent));
2568 		node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2569 							   sb->s_blocksize);
2570 		BUFFER_TRACE(frame->bh, "get_write_access");
2571 		err = ext4_journal_get_write_access(handle, sb, frame->bh,
2572 						    EXT4_JTR_NONE);
2573 		if (err)
2574 			goto journal_error;
2575 		if (!add_level) {
2576 			unsigned icount1 = icount/2, icount2 = icount - icount1;
2577 			unsigned hash2 = dx_get_hash(entries + icount1);
2578 			dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2579 				       icount1, icount2));
2580 
2581 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2582 			err = ext4_journal_get_write_access(handle, sb,
2583 							    (frame - 1)->bh,
2584 							    EXT4_JTR_NONE);
2585 			if (err)
2586 				goto journal_error;
2587 
2588 			memcpy((char *) entries2, (char *) (entries + icount1),
2589 			       icount2 * sizeof(struct dx_entry));
2590 			dx_set_count(entries, icount1);
2591 			dx_set_count(entries2, icount2);
2592 			dx_set_limit(entries2, dx_node_limit(dir));
2593 
2594 			/* Which index block gets the new entry? */
2595 			if (at - entries >= icount1) {
2596 				frame->at = at - entries - icount1 + entries2;
2597 				frame->entries = entries = entries2;
2598 				swap(frame->bh, bh2);
2599 			}
2600 			dx_insert_block((frame - 1), hash2, newblock);
2601 			dxtrace(dx_show_index("node", frame->entries));
2602 			dxtrace(dx_show_index("node",
2603 			       ((struct dx_node *) bh2->b_data)->entries));
2604 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2605 			if (err)
2606 				goto journal_error;
2607 			brelse (bh2);
2608 			err = ext4_handle_dirty_dx_node(handle, dir,
2609 						   (frame - 1)->bh);
2610 			if (err)
2611 				goto journal_error;
2612 			err = ext4_handle_dirty_dx_node(handle, dir,
2613 							frame->bh);
2614 			if (restart || err)
2615 				goto journal_error;
2616 		} else {
2617 			struct dx_root *dxroot;
2618 			memcpy((char *) entries2, (char *) entries,
2619 			       icount * sizeof(struct dx_entry));
2620 			dx_set_limit(entries2, dx_node_limit(dir));
2621 
2622 			/* Set up root */
2623 			dx_set_count(entries, 1);
2624 			dx_set_block(entries + 0, newblock);
2625 			dxroot = (struct dx_root *)frames[0].bh->b_data;
2626 			dxroot->info.indirect_levels += 1;
2627 			dxtrace(printk(KERN_DEBUG
2628 				       "Creating %d level index...\n",
2629 				       dxroot->info.indirect_levels));
2630 			err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2631 			if (err)
2632 				goto journal_error;
2633 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2634 			brelse(bh2);
2635 			restart = 1;
2636 			goto journal_error;
2637 		}
2638 	}
2639 	de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2640 	if (IS_ERR(de)) {
2641 		err = PTR_ERR(de);
2642 		goto cleanup;
2643 	}
2644 	err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2645 	goto cleanup;
2646 
2647 journal_error:
2648 	ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2649 cleanup:
2650 	brelse(bh);
2651 	dx_release(frames);
2652 	/* @restart is true means htree-path has been changed, we need to
2653 	 * repeat dx_probe() to find out valid htree-path
2654 	 */
2655 	if (restart && err == 0)
2656 		goto again;
2657 	return err;
2658 }
2659 
2660 /*
2661  * ext4_generic_delete_entry deletes a directory entry by merging it
2662  * with the previous entry
2663  */
2664 int ext4_generic_delete_entry(struct inode *dir,
2665 			      struct ext4_dir_entry_2 *de_del,
2666 			      struct buffer_head *bh,
2667 			      void *entry_buf,
2668 			      int buf_size,
2669 			      int csum_size)
2670 {
2671 	struct ext4_dir_entry_2 *de, *pde;
2672 	unsigned int blocksize = dir->i_sb->s_blocksize;
2673 	int i;
2674 
2675 	i = 0;
2676 	pde = NULL;
2677 	de = entry_buf;
2678 	while (i < buf_size - csum_size) {
2679 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2680 					 entry_buf, buf_size, i))
2681 			return -EFSCORRUPTED;
2682 		if (de == de_del)  {
2683 			if (pde) {
2684 				pde->rec_len = ext4_rec_len_to_disk(
2685 					ext4_rec_len_from_disk(pde->rec_len,
2686 							       blocksize) +
2687 					ext4_rec_len_from_disk(de->rec_len,
2688 							       blocksize),
2689 					blocksize);
2690 
2691 				/* wipe entire dir_entry */
2692 				memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2693 								blocksize));
2694 			} else {
2695 				/* wipe dir_entry excluding the rec_len field */
2696 				de->inode = 0;
2697 				memset(&de->name_len, 0,
2698 					ext4_rec_len_from_disk(de->rec_len,
2699 								blocksize) -
2700 					offsetof(struct ext4_dir_entry_2,
2701 								name_len));
2702 			}
2703 
2704 			inode_inc_iversion(dir);
2705 			return 0;
2706 		}
2707 		i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2708 		pde = de;
2709 		de = ext4_next_entry(de, blocksize);
2710 	}
2711 	return -ENOENT;
2712 }
2713 
2714 static int ext4_delete_entry(handle_t *handle,
2715 			     struct inode *dir,
2716 			     struct ext4_dir_entry_2 *de_del,
2717 			     struct buffer_head *bh)
2718 {
2719 	int err, csum_size = 0;
2720 
2721 	if (ext4_has_inline_data(dir)) {
2722 		int has_inline_data = 1;
2723 		err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2724 					       &has_inline_data);
2725 		if (has_inline_data)
2726 			return err;
2727 	}
2728 
2729 	if (ext4_has_metadata_csum(dir->i_sb))
2730 		csum_size = sizeof(struct ext4_dir_entry_tail);
2731 
2732 	BUFFER_TRACE(bh, "get_write_access");
2733 	err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2734 					    EXT4_JTR_NONE);
2735 	if (unlikely(err))
2736 		goto out;
2737 
2738 	err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2739 					dir->i_sb->s_blocksize, csum_size);
2740 	if (err)
2741 		goto out;
2742 
2743 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2744 	err = ext4_handle_dirty_dirblock(handle, dir, bh);
2745 	if (unlikely(err))
2746 		goto out;
2747 
2748 	return 0;
2749 out:
2750 	if (err != -ENOENT)
2751 		ext4_std_error(dir->i_sb, err);
2752 	return err;
2753 }
2754 
2755 /*
2756  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2757  * since this indicates that nlinks count was previously 1 to avoid overflowing
2758  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2759  * that subdirectory link counts are not being maintained accurately.
2760  *
2761  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2762  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2763  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2764  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2765  */
2766 static void ext4_inc_count(struct inode *inode)
2767 {
2768 	inc_nlink(inode);
2769 	if (is_dx(inode) &&
2770 	    (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2771 		set_nlink(inode, 1);
2772 }
2773 
2774 /*
2775  * If a directory had nlink == 1, then we should let it be 1. This indicates
2776  * directory has >EXT4_LINK_MAX subdirs.
2777  */
2778 static void ext4_dec_count(struct inode *inode)
2779 {
2780 	if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2781 		drop_nlink(inode);
2782 }
2783 
2784 
2785 /*
2786  * Add non-directory inode to a directory. On success, the inode reference is
2787  * consumed by dentry is instantiation. This is also indicated by clearing of
2788  * *inodep pointer. On failure, the caller is responsible for dropping the
2789  * inode reference in the safe context.
2790  */
2791 static int ext4_add_nondir(handle_t *handle,
2792 		struct dentry *dentry, struct inode **inodep)
2793 {
2794 	struct inode *dir = d_inode(dentry->d_parent);
2795 	struct inode *inode = *inodep;
2796 	int err = ext4_add_entry(handle, dentry, inode);
2797 	if (!err) {
2798 		err = ext4_mark_inode_dirty(handle, inode);
2799 		if (IS_DIRSYNC(dir))
2800 			ext4_handle_sync(handle);
2801 		d_instantiate_new(dentry, inode);
2802 		*inodep = NULL;
2803 		return err;
2804 	}
2805 	drop_nlink(inode);
2806 	ext4_mark_inode_dirty(handle, inode);
2807 	ext4_orphan_add(handle, inode);
2808 	unlock_new_inode(inode);
2809 	return err;
2810 }
2811 
2812 /*
2813  * By the time this is called, we already have created
2814  * the directory cache entry for the new file, but it
2815  * is so far negative - it has no inode.
2816  *
2817  * If the create succeeds, we fill in the inode information
2818  * with d_instantiate().
2819  */
2820 static int ext4_create(struct mnt_idmap *idmap, struct inode *dir,
2821 		       struct dentry *dentry, umode_t mode, bool excl)
2822 {
2823 	handle_t *handle;
2824 	struct inode *inode;
2825 	int err, credits, retries = 0;
2826 
2827 	err = dquot_initialize(dir);
2828 	if (err)
2829 		return err;
2830 
2831 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2832 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2833 retry:
2834 	inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2835 					    0, NULL, EXT4_HT_DIR, credits);
2836 	handle = ext4_journal_current_handle();
2837 	err = PTR_ERR(inode);
2838 	if (!IS_ERR(inode)) {
2839 		inode->i_op = &ext4_file_inode_operations;
2840 		inode->i_fop = &ext4_file_operations;
2841 		ext4_set_aops(inode);
2842 		err = ext4_add_nondir(handle, dentry, &inode);
2843 		if (!err)
2844 			ext4_fc_track_create(handle, dentry);
2845 	}
2846 	if (handle)
2847 		ext4_journal_stop(handle);
2848 	if (!IS_ERR_OR_NULL(inode))
2849 		iput(inode);
2850 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2851 		goto retry;
2852 	return err;
2853 }
2854 
2855 static int ext4_mknod(struct mnt_idmap *idmap, struct inode *dir,
2856 		      struct dentry *dentry, umode_t mode, dev_t rdev)
2857 {
2858 	handle_t *handle;
2859 	struct inode *inode;
2860 	int err, credits, retries = 0;
2861 
2862 	err = dquot_initialize(dir);
2863 	if (err)
2864 		return err;
2865 
2866 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2867 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2868 retry:
2869 	inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2870 					    0, NULL, EXT4_HT_DIR, credits);
2871 	handle = ext4_journal_current_handle();
2872 	err = PTR_ERR(inode);
2873 	if (!IS_ERR(inode)) {
2874 		init_special_inode(inode, inode->i_mode, rdev);
2875 		inode->i_op = &ext4_special_inode_operations;
2876 		err = ext4_add_nondir(handle, dentry, &inode);
2877 		if (!err)
2878 			ext4_fc_track_create(handle, dentry);
2879 	}
2880 	if (handle)
2881 		ext4_journal_stop(handle);
2882 	if (!IS_ERR_OR_NULL(inode))
2883 		iput(inode);
2884 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2885 		goto retry;
2886 	return err;
2887 }
2888 
2889 static int ext4_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
2890 			struct file *file, umode_t mode)
2891 {
2892 	handle_t *handle;
2893 	struct inode *inode;
2894 	int err, retries = 0;
2895 
2896 	err = dquot_initialize(dir);
2897 	if (err)
2898 		return err;
2899 
2900 retry:
2901 	inode = ext4_new_inode_start_handle(idmap, dir, mode,
2902 					    NULL, 0, NULL,
2903 					    EXT4_HT_DIR,
2904 			EXT4_MAXQUOTAS_TRANS_BLOCKS(dir->i_sb) +
2905 			  4 + EXT4_XATTR_TRANS_BLOCKS);
2906 	handle = ext4_journal_current_handle();
2907 	err = PTR_ERR(inode);
2908 	if (!IS_ERR(inode)) {
2909 		inode->i_op = &ext4_file_inode_operations;
2910 		inode->i_fop = &ext4_file_operations;
2911 		ext4_set_aops(inode);
2912 		d_tmpfile(file, inode);
2913 		err = ext4_orphan_add(handle, inode);
2914 		if (err)
2915 			goto err_unlock_inode;
2916 		mark_inode_dirty(inode);
2917 		unlock_new_inode(inode);
2918 	}
2919 	if (handle)
2920 		ext4_journal_stop(handle);
2921 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2922 		goto retry;
2923 	return finish_open_simple(file, err);
2924 err_unlock_inode:
2925 	ext4_journal_stop(handle);
2926 	unlock_new_inode(inode);
2927 	return err;
2928 }
2929 
2930 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2931 			  struct ext4_dir_entry_2 *de,
2932 			  int blocksize, int csum_size,
2933 			  unsigned int parent_ino, int dotdot_real_len)
2934 {
2935 	de->inode = cpu_to_le32(inode->i_ino);
2936 	de->name_len = 1;
2937 	de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2938 					   blocksize);
2939 	strcpy(de->name, ".");
2940 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2941 
2942 	de = ext4_next_entry(de, blocksize);
2943 	de->inode = cpu_to_le32(parent_ino);
2944 	de->name_len = 2;
2945 	if (!dotdot_real_len)
2946 		de->rec_len = ext4_rec_len_to_disk(blocksize -
2947 					(csum_size + ext4_dir_rec_len(1, NULL)),
2948 					blocksize);
2949 	else
2950 		de->rec_len = ext4_rec_len_to_disk(
2951 					ext4_dir_rec_len(de->name_len, NULL),
2952 					blocksize);
2953 	strcpy(de->name, "..");
2954 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2955 
2956 	return ext4_next_entry(de, blocksize);
2957 }
2958 
2959 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2960 			     struct inode *inode)
2961 {
2962 	struct buffer_head *dir_block = NULL;
2963 	struct ext4_dir_entry_2 *de;
2964 	ext4_lblk_t block = 0;
2965 	unsigned int blocksize = dir->i_sb->s_blocksize;
2966 	int csum_size = 0;
2967 	int err;
2968 
2969 	if (ext4_has_metadata_csum(dir->i_sb))
2970 		csum_size = sizeof(struct ext4_dir_entry_tail);
2971 
2972 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2973 		err = ext4_try_create_inline_dir(handle, dir, inode);
2974 		if (err < 0 && err != -ENOSPC)
2975 			goto out;
2976 		if (!err)
2977 			goto out;
2978 	}
2979 
2980 	inode->i_size = 0;
2981 	dir_block = ext4_append(handle, inode, &block);
2982 	if (IS_ERR(dir_block))
2983 		return PTR_ERR(dir_block);
2984 	de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2985 	ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2986 	set_nlink(inode, 2);
2987 	if (csum_size)
2988 		ext4_initialize_dirent_tail(dir_block, blocksize);
2989 
2990 	BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2991 	err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2992 	if (err)
2993 		goto out;
2994 	set_buffer_verified(dir_block);
2995 out:
2996 	brelse(dir_block);
2997 	return err;
2998 }
2999 
3000 static int ext4_mkdir(struct mnt_idmap *idmap, struct inode *dir,
3001 		      struct dentry *dentry, umode_t mode)
3002 {
3003 	handle_t *handle;
3004 	struct inode *inode;
3005 	int err, err2 = 0, credits, retries = 0;
3006 
3007 	if (EXT4_DIR_LINK_MAX(dir))
3008 		return -EMLINK;
3009 
3010 	err = dquot_initialize(dir);
3011 	if (err)
3012 		return err;
3013 
3014 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3015 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
3016 retry:
3017 	inode = ext4_new_inode_start_handle(idmap, dir, S_IFDIR | mode,
3018 					    &dentry->d_name,
3019 					    0, NULL, EXT4_HT_DIR, credits);
3020 	handle = ext4_journal_current_handle();
3021 	err = PTR_ERR(inode);
3022 	if (IS_ERR(inode))
3023 		goto out_stop;
3024 
3025 	inode->i_op = &ext4_dir_inode_operations;
3026 	inode->i_fop = &ext4_dir_operations;
3027 	err = ext4_init_new_dir(handle, dir, inode);
3028 	if (err)
3029 		goto out_clear_inode;
3030 	err = ext4_mark_inode_dirty(handle, inode);
3031 	if (!err)
3032 		err = ext4_add_entry(handle, dentry, inode);
3033 	if (err) {
3034 out_clear_inode:
3035 		clear_nlink(inode);
3036 		ext4_orphan_add(handle, inode);
3037 		unlock_new_inode(inode);
3038 		err2 = ext4_mark_inode_dirty(handle, inode);
3039 		if (unlikely(err2))
3040 			err = err2;
3041 		ext4_journal_stop(handle);
3042 		iput(inode);
3043 		goto out_retry;
3044 	}
3045 	ext4_inc_count(dir);
3046 
3047 	ext4_update_dx_flag(dir);
3048 	err = ext4_mark_inode_dirty(handle, dir);
3049 	if (err)
3050 		goto out_clear_inode;
3051 	d_instantiate_new(dentry, inode);
3052 	ext4_fc_track_create(handle, dentry);
3053 	if (IS_DIRSYNC(dir))
3054 		ext4_handle_sync(handle);
3055 
3056 out_stop:
3057 	if (handle)
3058 		ext4_journal_stop(handle);
3059 out_retry:
3060 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3061 		goto retry;
3062 	return err;
3063 }
3064 
3065 /*
3066  * routine to check that the specified directory is empty (for rmdir)
3067  */
3068 bool ext4_empty_dir(struct inode *inode)
3069 {
3070 	unsigned int offset;
3071 	struct buffer_head *bh;
3072 	struct ext4_dir_entry_2 *de;
3073 	struct super_block *sb;
3074 
3075 	if (ext4_has_inline_data(inode)) {
3076 		int has_inline_data = 1;
3077 		int ret;
3078 
3079 		ret = empty_inline_dir(inode, &has_inline_data);
3080 		if (has_inline_data)
3081 			return ret;
3082 	}
3083 
3084 	sb = inode->i_sb;
3085 	if (inode->i_size < ext4_dir_rec_len(1, NULL) +
3086 					ext4_dir_rec_len(2, NULL)) {
3087 		EXT4_ERROR_INODE(inode, "invalid size");
3088 		return false;
3089 	}
3090 	bh = ext4_read_dirblock(inode, 0, EITHER);
3091 	if (IS_ERR(bh))
3092 		return false;
3093 
3094 	de = (struct ext4_dir_entry_2 *) bh->b_data;
3095 	if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3096 				 0) ||
3097 	    le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3098 		ext4_warning_inode(inode, "directory missing '.'");
3099 		brelse(bh);
3100 		return false;
3101 	}
3102 	offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3103 	de = ext4_next_entry(de, sb->s_blocksize);
3104 	if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3105 				 offset) ||
3106 	    le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3107 		ext4_warning_inode(inode, "directory missing '..'");
3108 		brelse(bh);
3109 		return false;
3110 	}
3111 	offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3112 	while (offset < inode->i_size) {
3113 		if (!(offset & (sb->s_blocksize - 1))) {
3114 			unsigned int lblock;
3115 			brelse(bh);
3116 			lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3117 			bh = ext4_read_dirblock(inode, lblock, EITHER);
3118 			if (bh == NULL) {
3119 				offset += sb->s_blocksize;
3120 				continue;
3121 			}
3122 			if (IS_ERR(bh))
3123 				return false;
3124 		}
3125 		de = (struct ext4_dir_entry_2 *) (bh->b_data +
3126 					(offset & (sb->s_blocksize - 1)));
3127 		if (ext4_check_dir_entry(inode, NULL, de, bh,
3128 					 bh->b_data, bh->b_size, offset) ||
3129 		    le32_to_cpu(de->inode)) {
3130 			brelse(bh);
3131 			return false;
3132 		}
3133 		offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3134 	}
3135 	brelse(bh);
3136 	return true;
3137 }
3138 
3139 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3140 {
3141 	int retval;
3142 	struct inode *inode;
3143 	struct buffer_head *bh;
3144 	struct ext4_dir_entry_2 *de;
3145 	handle_t *handle = NULL;
3146 
3147 	if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3148 		return -EIO;
3149 
3150 	/* Initialize quotas before so that eventual writes go in
3151 	 * separate transaction */
3152 	retval = dquot_initialize(dir);
3153 	if (retval)
3154 		return retval;
3155 	retval = dquot_initialize(d_inode(dentry));
3156 	if (retval)
3157 		return retval;
3158 
3159 	retval = -ENOENT;
3160 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3161 	if (IS_ERR(bh))
3162 		return PTR_ERR(bh);
3163 	if (!bh)
3164 		goto end_rmdir;
3165 
3166 	inode = d_inode(dentry);
3167 
3168 	retval = -EFSCORRUPTED;
3169 	if (le32_to_cpu(de->inode) != inode->i_ino)
3170 		goto end_rmdir;
3171 
3172 	retval = -ENOTEMPTY;
3173 	if (!ext4_empty_dir(inode))
3174 		goto end_rmdir;
3175 
3176 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3177 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3178 	if (IS_ERR(handle)) {
3179 		retval = PTR_ERR(handle);
3180 		handle = NULL;
3181 		goto end_rmdir;
3182 	}
3183 
3184 	if (IS_DIRSYNC(dir))
3185 		ext4_handle_sync(handle);
3186 
3187 	retval = ext4_delete_entry(handle, dir, de, bh);
3188 	if (retval)
3189 		goto end_rmdir;
3190 	if (!EXT4_DIR_LINK_EMPTY(inode))
3191 		ext4_warning_inode(inode,
3192 			     "empty directory '%.*s' has too many links (%u)",
3193 			     dentry->d_name.len, dentry->d_name.name,
3194 			     inode->i_nlink);
3195 	inode_inc_iversion(inode);
3196 	clear_nlink(inode);
3197 	/* There's no need to set i_disksize: the fact that i_nlink is
3198 	 * zero will ensure that the right thing happens during any
3199 	 * recovery. */
3200 	inode->i_size = 0;
3201 	ext4_orphan_add(handle, inode);
3202 	inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3203 	inode_set_ctime_current(inode);
3204 	retval = ext4_mark_inode_dirty(handle, inode);
3205 	if (retval)
3206 		goto end_rmdir;
3207 	ext4_dec_count(dir);
3208 	ext4_update_dx_flag(dir);
3209 	ext4_fc_track_unlink(handle, dentry);
3210 	retval = ext4_mark_inode_dirty(handle, dir);
3211 
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_ENABLED(CONFIG_UNICODE) && IS_CASEFOLDED(dir))
3219 		d_invalidate(dentry);
3220 
3221 end_rmdir:
3222 	brelse(bh);
3223 	if (handle)
3224 		ext4_journal_stop(handle);
3225 	return retval;
3226 }
3227 
3228 int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
3229 		  struct inode *inode,
3230 		  struct dentry *dentry /* NULL during fast_commit recovery */)
3231 {
3232 	int retval = -ENOENT;
3233 	struct buffer_head *bh;
3234 	struct ext4_dir_entry_2 *de;
3235 	handle_t *handle;
3236 	int skip_remove_dentry = 0;
3237 
3238 	/*
3239 	 * Keep this outside the transaction; it may have to set up the
3240 	 * directory's encryption key, which isn't GFP_NOFS-safe.
3241 	 */
3242 	bh = ext4_find_entry(dir, d_name, &de, NULL);
3243 	if (IS_ERR(bh))
3244 		return PTR_ERR(bh);
3245 
3246 	if (!bh)
3247 		return -ENOENT;
3248 
3249 	if (le32_to_cpu(de->inode) != inode->i_ino) {
3250 		/*
3251 		 * It's okay if we find dont find dentry which matches
3252 		 * the inode. That's because it might have gotten
3253 		 * renamed to a different inode number
3254 		 */
3255 		if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3256 			skip_remove_dentry = 1;
3257 		else
3258 			goto out_bh;
3259 	}
3260 
3261 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3262 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3263 	if (IS_ERR(handle)) {
3264 		retval = PTR_ERR(handle);
3265 		goto out_bh;
3266 	}
3267 
3268 	if (IS_DIRSYNC(dir))
3269 		ext4_handle_sync(handle);
3270 
3271 	if (!skip_remove_dentry) {
3272 		retval = ext4_delete_entry(handle, dir, de, bh);
3273 		if (retval)
3274 			goto out_handle;
3275 		inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3276 		ext4_update_dx_flag(dir);
3277 		retval = ext4_mark_inode_dirty(handle, dir);
3278 		if (retval)
3279 			goto out_handle;
3280 	} else {
3281 		retval = 0;
3282 	}
3283 	if (inode->i_nlink == 0)
3284 		ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3285 				   d_name->len, d_name->name);
3286 	else
3287 		drop_nlink(inode);
3288 	if (!inode->i_nlink)
3289 		ext4_orphan_add(handle, inode);
3290 	inode_set_ctime_current(inode);
3291 	retval = ext4_mark_inode_dirty(handle, inode);
3292 	if (dentry && !retval)
3293 		ext4_fc_track_unlink(handle, dentry);
3294 out_handle:
3295 	ext4_journal_stop(handle);
3296 out_bh:
3297 	brelse(bh);
3298 	return retval;
3299 }
3300 
3301 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3302 {
3303 	int retval;
3304 
3305 	if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3306 		return -EIO;
3307 
3308 	trace_ext4_unlink_enter(dir, dentry);
3309 	/*
3310 	 * Initialize quotas before so that eventual writes go
3311 	 * in separate transaction
3312 	 */
3313 	retval = dquot_initialize(dir);
3314 	if (retval)
3315 		goto out_trace;
3316 	retval = dquot_initialize(d_inode(dentry));
3317 	if (retval)
3318 		goto out_trace;
3319 
3320 	retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry), dentry);
3321 
3322 	/* VFS negative dentries are incompatible with Encoding and
3323 	 * Case-insensitiveness. Eventually we'll want avoid
3324 	 * invalidating the dentries here, alongside with returning the
3325 	 * negative dentries at ext4_lookup(), when it is  better
3326 	 * supported by the VFS for the CI case.
3327 	 */
3328 	if (IS_ENABLED(CONFIG_UNICODE) && IS_CASEFOLDED(dir))
3329 		d_invalidate(dentry);
3330 
3331 out_trace:
3332 	trace_ext4_unlink_exit(dentry, retval);
3333 	return retval;
3334 }
3335 
3336 static int ext4_init_symlink_block(handle_t *handle, struct inode *inode,
3337 				   struct fscrypt_str *disk_link)
3338 {
3339 	struct buffer_head *bh;
3340 	char *kaddr;
3341 	int err = 0;
3342 
3343 	bh = ext4_bread(handle, inode, 0, EXT4_GET_BLOCKS_CREATE);
3344 	if (IS_ERR(bh))
3345 		return PTR_ERR(bh);
3346 
3347 	BUFFER_TRACE(bh, "get_write_access");
3348 	err = ext4_journal_get_write_access(handle, inode->i_sb, bh, EXT4_JTR_NONE);
3349 	if (err)
3350 		goto out;
3351 
3352 	kaddr = (char *)bh->b_data;
3353 	memcpy(kaddr, disk_link->name, disk_link->len);
3354 	inode->i_size = disk_link->len - 1;
3355 	EXT4_I(inode)->i_disksize = inode->i_size;
3356 	err = ext4_handle_dirty_metadata(handle, inode, bh);
3357 out:
3358 	brelse(bh);
3359 	return err;
3360 }
3361 
3362 static int ext4_symlink(struct mnt_idmap *idmap, struct inode *dir,
3363 			struct dentry *dentry, const char *symname)
3364 {
3365 	handle_t *handle;
3366 	struct inode *inode;
3367 	int err, len = strlen(symname);
3368 	int credits;
3369 	struct fscrypt_str disk_link;
3370 	int retries = 0;
3371 
3372 	if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3373 		return -EIO;
3374 
3375 	err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3376 				      &disk_link);
3377 	if (err)
3378 		return err;
3379 
3380 	err = dquot_initialize(dir);
3381 	if (err)
3382 		return err;
3383 
3384 	/*
3385 	 * EXT4_INDEX_EXTRA_TRANS_BLOCKS for addition of entry into the
3386 	 * directory. +3 for inode, inode bitmap, group descriptor allocation.
3387 	 * EXT4_DATA_TRANS_BLOCKS for the data block allocation and
3388 	 * modification.
3389 	 */
3390 	credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3391 		  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3392 retry:
3393 	inode = ext4_new_inode_start_handle(idmap, dir, S_IFLNK|S_IRWXUGO,
3394 					    &dentry->d_name, 0, NULL,
3395 					    EXT4_HT_DIR, credits);
3396 	handle = ext4_journal_current_handle();
3397 	if (IS_ERR(inode)) {
3398 		if (handle)
3399 			ext4_journal_stop(handle);
3400 		err = PTR_ERR(inode);
3401 		goto out_retry;
3402 	}
3403 
3404 	if (IS_ENCRYPTED(inode)) {
3405 		err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3406 		if (err)
3407 			goto err_drop_inode;
3408 		inode->i_op = &ext4_encrypted_symlink_inode_operations;
3409 	} else {
3410 		if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3411 			inode->i_op = &ext4_symlink_inode_operations;
3412 		} else {
3413 			inode->i_op = &ext4_fast_symlink_inode_operations;
3414 			inode->i_link = (char *)&EXT4_I(inode)->i_data;
3415 		}
3416 	}
3417 
3418 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3419 		/* alloc symlink block and fill it */
3420 		err = ext4_init_symlink_block(handle, inode, &disk_link);
3421 		if (err)
3422 			goto err_drop_inode;
3423 	} else {
3424 		/* clear the extent format for fast symlink */
3425 		ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3426 		memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3427 		       disk_link.len);
3428 		inode->i_size = disk_link.len - 1;
3429 		EXT4_I(inode)->i_disksize = inode->i_size;
3430 	}
3431 	err = ext4_add_nondir(handle, dentry, &inode);
3432 	if (handle)
3433 		ext4_journal_stop(handle);
3434 	iput(inode);
3435 	goto out_retry;
3436 
3437 err_drop_inode:
3438 	clear_nlink(inode);
3439 	ext4_mark_inode_dirty(handle, inode);
3440 	ext4_orphan_add(handle, inode);
3441 	unlock_new_inode(inode);
3442 	if (handle)
3443 		ext4_journal_stop(handle);
3444 	iput(inode);
3445 out_retry:
3446 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3447 		goto retry;
3448 	if (disk_link.name != (unsigned char *)symname)
3449 		kfree(disk_link.name);
3450 	return err;
3451 }
3452 
3453 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3454 {
3455 	handle_t *handle;
3456 	int err, retries = 0;
3457 retry:
3458 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3459 		(EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3460 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3461 	if (IS_ERR(handle))
3462 		return PTR_ERR(handle);
3463 
3464 	if (IS_DIRSYNC(dir))
3465 		ext4_handle_sync(handle);
3466 
3467 	inode_set_ctime_current(inode);
3468 	ext4_inc_count(inode);
3469 	ihold(inode);
3470 
3471 	err = ext4_add_entry(handle, dentry, inode);
3472 	if (!err) {
3473 		err = ext4_mark_inode_dirty(handle, inode);
3474 		/* this can happen only for tmpfile being
3475 		 * linked the first time
3476 		 */
3477 		if (inode->i_nlink == 1)
3478 			ext4_orphan_del(handle, inode);
3479 		d_instantiate(dentry, inode);
3480 		ext4_fc_track_link(handle, dentry);
3481 	} else {
3482 		drop_nlink(inode);
3483 		iput(inode);
3484 	}
3485 	ext4_journal_stop(handle);
3486 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3487 		goto retry;
3488 	return err;
3489 }
3490 
3491 static int ext4_link(struct dentry *old_dentry,
3492 		     struct inode *dir, struct dentry *dentry)
3493 {
3494 	struct inode *inode = d_inode(old_dentry);
3495 	int err;
3496 
3497 	if (inode->i_nlink >= EXT4_LINK_MAX)
3498 		return -EMLINK;
3499 
3500 	err = fscrypt_prepare_link(old_dentry, dir, dentry);
3501 	if (err)
3502 		return err;
3503 
3504 	if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3505 	    (!projid_eq(EXT4_I(dir)->i_projid,
3506 			EXT4_I(old_dentry->d_inode)->i_projid)))
3507 		return -EXDEV;
3508 
3509 	err = dquot_initialize(dir);
3510 	if (err)
3511 		return err;
3512 	return __ext4_link(dir, inode, dentry);
3513 }
3514 
3515 /*
3516  * Try to find buffer head where contains the parent block.
3517  * It should be the inode block if it is inlined or the 1st block
3518  * if it is a normal dir.
3519  */
3520 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3521 					struct inode *inode,
3522 					int *retval,
3523 					struct ext4_dir_entry_2 **parent_de,
3524 					int *inlined)
3525 {
3526 	struct buffer_head *bh;
3527 
3528 	if (!ext4_has_inline_data(inode)) {
3529 		struct ext4_dir_entry_2 *de;
3530 		unsigned int offset;
3531 
3532 		bh = ext4_read_dirblock(inode, 0, EITHER);
3533 		if (IS_ERR(bh)) {
3534 			*retval = PTR_ERR(bh);
3535 			return NULL;
3536 		}
3537 
3538 		de = (struct ext4_dir_entry_2 *) bh->b_data;
3539 		if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3540 					 bh->b_size, 0) ||
3541 		    le32_to_cpu(de->inode) != inode->i_ino ||
3542 		    strcmp(".", de->name)) {
3543 			EXT4_ERROR_INODE(inode, "directory missing '.'");
3544 			brelse(bh);
3545 			*retval = -EFSCORRUPTED;
3546 			return NULL;
3547 		}
3548 		offset = ext4_rec_len_from_disk(de->rec_len,
3549 						inode->i_sb->s_blocksize);
3550 		de = ext4_next_entry(de, inode->i_sb->s_blocksize);
3551 		if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3552 					 bh->b_size, offset) ||
3553 		    le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3554 			EXT4_ERROR_INODE(inode, "directory missing '..'");
3555 			brelse(bh);
3556 			*retval = -EFSCORRUPTED;
3557 			return NULL;
3558 		}
3559 		*parent_de = de;
3560 
3561 		return bh;
3562 	}
3563 
3564 	*inlined = 1;
3565 	return ext4_get_first_inline_block(inode, parent_de, retval);
3566 }
3567 
3568 struct ext4_renament {
3569 	struct inode *dir;
3570 	struct dentry *dentry;
3571 	struct inode *inode;
3572 	bool is_dir;
3573 	int dir_nlink_delta;
3574 
3575 	/* entry for "dentry" */
3576 	struct buffer_head *bh;
3577 	struct ext4_dir_entry_2 *de;
3578 	int inlined;
3579 
3580 	/* entry for ".." in inode if it's a directory */
3581 	struct buffer_head *dir_bh;
3582 	struct ext4_dir_entry_2 *parent_de;
3583 	int dir_inlined;
3584 };
3585 
3586 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent, bool is_cross)
3587 {
3588 	int retval;
3589 
3590 	ent->is_dir = true;
3591 	if (!is_cross)
3592 		return 0;
3593 
3594 	ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3595 					      &retval, &ent->parent_de,
3596 					      &ent->dir_inlined);
3597 	if (!ent->dir_bh)
3598 		return retval;
3599 	if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3600 		return -EFSCORRUPTED;
3601 	BUFFER_TRACE(ent->dir_bh, "get_write_access");
3602 	return ext4_journal_get_write_access(handle, ent->dir->i_sb,
3603 					     ent->dir_bh, EXT4_JTR_NONE);
3604 }
3605 
3606 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3607 				  unsigned dir_ino)
3608 {
3609 	int retval;
3610 
3611 	if (!ent->dir_bh)
3612 		return 0;
3613 
3614 	ent->parent_de->inode = cpu_to_le32(dir_ino);
3615 	BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3616 	if (!ent->dir_inlined) {
3617 		if (is_dx(ent->inode)) {
3618 			retval = ext4_handle_dirty_dx_node(handle,
3619 							   ent->inode,
3620 							   ent->dir_bh);
3621 		} else {
3622 			retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3623 							    ent->dir_bh);
3624 		}
3625 	} else {
3626 		retval = ext4_mark_inode_dirty(handle, ent->inode);
3627 	}
3628 	if (retval) {
3629 		ext4_std_error(ent->dir->i_sb, retval);
3630 		return retval;
3631 	}
3632 	return 0;
3633 }
3634 
3635 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3636 		       unsigned ino, unsigned file_type)
3637 {
3638 	int retval, retval2;
3639 
3640 	BUFFER_TRACE(ent->bh, "get write access");
3641 	retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
3642 					       EXT4_JTR_NONE);
3643 	if (retval)
3644 		return retval;
3645 	ent->de->inode = cpu_to_le32(ino);
3646 	if (ext4_has_feature_filetype(ent->dir->i_sb))
3647 		ent->de->file_type = file_type;
3648 	inode_inc_iversion(ent->dir);
3649 	inode_set_mtime_to_ts(ent->dir, inode_set_ctime_current(ent->dir));
3650 	retval = ext4_mark_inode_dirty(handle, ent->dir);
3651 	BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3652 	if (!ent->inlined) {
3653 		retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3654 		if (unlikely(retval2)) {
3655 			ext4_std_error(ent->dir->i_sb, retval2);
3656 			return retval2;
3657 		}
3658 	}
3659 	return retval;
3660 }
3661 
3662 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3663 			  unsigned ino, unsigned file_type)
3664 {
3665 	struct ext4_renament old = *ent;
3666 	int retval = 0;
3667 
3668 	/*
3669 	 * old->de could have moved from under us during make indexed dir,
3670 	 * so the old->de may no longer valid and need to find it again
3671 	 * before reset old inode info.
3672 	 */
3673 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3674 				 &old.inlined);
3675 	if (IS_ERR(old.bh))
3676 		retval = PTR_ERR(old.bh);
3677 	if (!old.bh)
3678 		retval = -ENOENT;
3679 	if (retval) {
3680 		ext4_std_error(old.dir->i_sb, retval);
3681 		return;
3682 	}
3683 
3684 	ext4_setent(handle, &old, ino, file_type);
3685 	brelse(old.bh);
3686 }
3687 
3688 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3689 				  const struct qstr *d_name)
3690 {
3691 	int retval = -ENOENT;
3692 	struct buffer_head *bh;
3693 	struct ext4_dir_entry_2 *de;
3694 
3695 	bh = ext4_find_entry(dir, d_name, &de, NULL);
3696 	if (IS_ERR(bh))
3697 		return PTR_ERR(bh);
3698 	if (bh) {
3699 		retval = ext4_delete_entry(handle, dir, de, bh);
3700 		brelse(bh);
3701 	}
3702 	return retval;
3703 }
3704 
3705 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3706 			       int force_reread)
3707 {
3708 	int retval;
3709 	/*
3710 	 * ent->de could have moved from under us during htree split, so make
3711 	 * sure that we are deleting the right entry.  We might also be pointing
3712 	 * to a stale entry in the unused part of ent->bh so just checking inum
3713 	 * and the name isn't enough.
3714 	 */
3715 	if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3716 	    ent->de->name_len != ent->dentry->d_name.len ||
3717 	    strncmp(ent->de->name, ent->dentry->d_name.name,
3718 		    ent->de->name_len) ||
3719 	    force_reread) {
3720 		retval = ext4_find_delete_entry(handle, ent->dir,
3721 						&ent->dentry->d_name);
3722 	} else {
3723 		retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3724 		if (retval == -ENOENT) {
3725 			retval = ext4_find_delete_entry(handle, ent->dir,
3726 							&ent->dentry->d_name);
3727 		}
3728 	}
3729 
3730 	if (retval) {
3731 		ext4_warning_inode(ent->dir,
3732 				   "Deleting old file: nlink %d, error=%d",
3733 				   ent->dir->i_nlink, retval);
3734 	}
3735 }
3736 
3737 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3738 {
3739 	if (ent->dir_nlink_delta) {
3740 		if (ent->dir_nlink_delta == -1)
3741 			ext4_dec_count(ent->dir);
3742 		else
3743 			ext4_inc_count(ent->dir);
3744 		ext4_mark_inode_dirty(handle, ent->dir);
3745 	}
3746 }
3747 
3748 static struct inode *ext4_whiteout_for_rename(struct mnt_idmap *idmap,
3749 					      struct ext4_renament *ent,
3750 					      int credits, handle_t **h)
3751 {
3752 	struct inode *wh;
3753 	handle_t *handle;
3754 	int retries = 0;
3755 
3756 	/*
3757 	 * for inode block, sb block, group summaries,
3758 	 * and inode bitmap
3759 	 */
3760 	credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3761 		    EXT4_XATTR_TRANS_BLOCKS + 4);
3762 retry:
3763 	wh = ext4_new_inode_start_handle(idmap, ent->dir,
3764 					 S_IFCHR | WHITEOUT_MODE,
3765 					 &ent->dentry->d_name, 0, NULL,
3766 					 EXT4_HT_DIR, credits);
3767 
3768 	handle = ext4_journal_current_handle();
3769 	if (IS_ERR(wh)) {
3770 		if (handle)
3771 			ext4_journal_stop(handle);
3772 		if (PTR_ERR(wh) == -ENOSPC &&
3773 		    ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3774 			goto retry;
3775 	} else {
3776 		*h = handle;
3777 		init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3778 		wh->i_op = &ext4_special_inode_operations;
3779 	}
3780 	return wh;
3781 }
3782 
3783 /*
3784  * Anybody can rename anything with this: the permission checks are left to the
3785  * higher-level routines.
3786  *
3787  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3788  * while new_{dentry,inode) refers to the destination dentry/inode
3789  * This comes from rename(const char *oldpath, const char *newpath)
3790  */
3791 static int ext4_rename(struct mnt_idmap *idmap, struct inode *old_dir,
3792 		       struct dentry *old_dentry, struct inode *new_dir,
3793 		       struct dentry *new_dentry, unsigned int flags)
3794 {
3795 	handle_t *handle = NULL;
3796 	struct ext4_renament old = {
3797 		.dir = old_dir,
3798 		.dentry = old_dentry,
3799 		.inode = d_inode(old_dentry),
3800 	};
3801 	struct ext4_renament new = {
3802 		.dir = new_dir,
3803 		.dentry = new_dentry,
3804 		.inode = d_inode(new_dentry),
3805 	};
3806 	int force_reread;
3807 	int retval;
3808 	struct inode *whiteout = NULL;
3809 	int credits;
3810 	u8 old_file_type;
3811 
3812 	if (new.inode && new.inode->i_nlink == 0) {
3813 		EXT4_ERROR_INODE(new.inode,
3814 				 "target of rename is already freed");
3815 		return -EFSCORRUPTED;
3816 	}
3817 
3818 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3819 	    (!projid_eq(EXT4_I(new_dir)->i_projid,
3820 			EXT4_I(old_dentry->d_inode)->i_projid)))
3821 		return -EXDEV;
3822 
3823 	retval = dquot_initialize(old.dir);
3824 	if (retval)
3825 		return retval;
3826 	retval = dquot_initialize(old.inode);
3827 	if (retval)
3828 		return retval;
3829 	retval = dquot_initialize(new.dir);
3830 	if (retval)
3831 		return retval;
3832 
3833 	/* Initialize quotas before so that eventual writes go
3834 	 * in separate transaction */
3835 	if (new.inode) {
3836 		retval = dquot_initialize(new.inode);
3837 		if (retval)
3838 			return retval;
3839 	}
3840 
3841 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3842 				 &old.inlined);
3843 	if (IS_ERR(old.bh))
3844 		return PTR_ERR(old.bh);
3845 
3846 	/*
3847 	 *  Check for inode number is _not_ due to possible IO errors.
3848 	 *  We might rmdir the source, keep it as pwd of some process
3849 	 *  and merrily kill the link to whatever was created under the
3850 	 *  same name. Goodbye sticky bit ;-<
3851 	 */
3852 	retval = -ENOENT;
3853 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3854 		goto release_bh;
3855 
3856 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3857 				 &new.de, &new.inlined);
3858 	if (IS_ERR(new.bh)) {
3859 		retval = PTR_ERR(new.bh);
3860 		new.bh = NULL;
3861 		goto release_bh;
3862 	}
3863 	if (new.bh) {
3864 		if (!new.inode) {
3865 			brelse(new.bh);
3866 			new.bh = NULL;
3867 		}
3868 	}
3869 	if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3870 		ext4_alloc_da_blocks(old.inode);
3871 
3872 	credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3873 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3874 	if (!(flags & RENAME_WHITEOUT)) {
3875 		handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3876 		if (IS_ERR(handle)) {
3877 			retval = PTR_ERR(handle);
3878 			goto release_bh;
3879 		}
3880 	} else {
3881 		whiteout = ext4_whiteout_for_rename(idmap, &old, credits, &handle);
3882 		if (IS_ERR(whiteout)) {
3883 			retval = PTR_ERR(whiteout);
3884 			goto release_bh;
3885 		}
3886 	}
3887 
3888 	old_file_type = old.de->file_type;
3889 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3890 		ext4_handle_sync(handle);
3891 
3892 	if (S_ISDIR(old.inode->i_mode)) {
3893 		if (new.inode) {
3894 			retval = -ENOTEMPTY;
3895 			if (!ext4_empty_dir(new.inode))
3896 				goto end_rename;
3897 		} else {
3898 			retval = -EMLINK;
3899 			if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3900 				goto end_rename;
3901 		}
3902 		retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
3903 		if (retval)
3904 			goto end_rename;
3905 	}
3906 	/*
3907 	 * If we're renaming a file within an inline_data dir and adding or
3908 	 * setting the new dirent causes a conversion from inline_data to
3909 	 * extents/blockmap, we need to force the dirent delete code to
3910 	 * re-read the directory, or else we end up trying to delete a dirent
3911 	 * from what is now the extent tree root (or a block map).
3912 	 */
3913 	force_reread = (new.dir->i_ino == old.dir->i_ino &&
3914 			ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3915 
3916 	if (whiteout) {
3917 		/*
3918 		 * Do this before adding a new entry, so the old entry is sure
3919 		 * to be still pointing to the valid old entry.
3920 		 */
3921 		retval = ext4_setent(handle, &old, whiteout->i_ino,
3922 				     EXT4_FT_CHRDEV);
3923 		if (retval)
3924 			goto end_rename;
3925 		retval = ext4_mark_inode_dirty(handle, whiteout);
3926 		if (unlikely(retval))
3927 			goto end_rename;
3928 
3929 	}
3930 	if (!new.bh) {
3931 		retval = ext4_add_entry(handle, new.dentry, old.inode);
3932 		if (retval)
3933 			goto end_rename;
3934 	} else {
3935 		retval = ext4_setent(handle, &new,
3936 				     old.inode->i_ino, old_file_type);
3937 		if (retval)
3938 			goto end_rename;
3939 	}
3940 	if (force_reread)
3941 		force_reread = !ext4_test_inode_flag(new.dir,
3942 						     EXT4_INODE_INLINE_DATA);
3943 
3944 	/*
3945 	 * Like most other Unix systems, set the ctime for inodes on a
3946 	 * rename.
3947 	 */
3948 	inode_set_ctime_current(old.inode);
3949 	retval = ext4_mark_inode_dirty(handle, old.inode);
3950 	if (unlikely(retval))
3951 		goto end_rename;
3952 
3953 	if (!whiteout) {
3954 		/*
3955 		 * ok, that's it
3956 		 */
3957 		ext4_rename_delete(handle, &old, force_reread);
3958 	}
3959 
3960 	if (new.inode) {
3961 		ext4_dec_count(new.inode);
3962 		inode_set_ctime_current(new.inode);
3963 	}
3964 	inode_set_mtime_to_ts(old.dir, inode_set_ctime_current(old.dir));
3965 	ext4_update_dx_flag(old.dir);
3966 	if (old.is_dir) {
3967 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3968 		if (retval)
3969 			goto end_rename;
3970 
3971 		ext4_dec_count(old.dir);
3972 		if (new.inode) {
3973 			/* checked ext4_empty_dir above, can't have another
3974 			 * parent, ext4_dec_count() won't work for many-linked
3975 			 * dirs */
3976 			clear_nlink(new.inode);
3977 		} else {
3978 			ext4_inc_count(new.dir);
3979 			ext4_update_dx_flag(new.dir);
3980 			retval = ext4_mark_inode_dirty(handle, new.dir);
3981 			if (unlikely(retval))
3982 				goto end_rename;
3983 		}
3984 	}
3985 	retval = ext4_mark_inode_dirty(handle, old.dir);
3986 	if (unlikely(retval))
3987 		goto end_rename;
3988 
3989 	if (old.is_dir) {
3990 		/*
3991 		 * We disable fast commits here that's because the
3992 		 * replay code is not yet capable of changing dot dot
3993 		 * dirents in directories.
3994 		 */
3995 		ext4_fc_mark_ineligible(old.inode->i_sb,
3996 			EXT4_FC_REASON_RENAME_DIR, handle);
3997 	} else {
3998 		struct super_block *sb = old.inode->i_sb;
3999 
4000 		if (new.inode)
4001 			ext4_fc_track_unlink(handle, new.dentry);
4002 		if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4003 		    !(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4004 		    !(ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE))) {
4005 			__ext4_fc_track_link(handle, old.inode, new.dentry);
4006 			__ext4_fc_track_unlink(handle, old.inode, old.dentry);
4007 			if (whiteout)
4008 				__ext4_fc_track_create(handle, whiteout,
4009 						       old.dentry);
4010 		}
4011 	}
4012 
4013 	if (new.inode) {
4014 		retval = ext4_mark_inode_dirty(handle, new.inode);
4015 		if (unlikely(retval))
4016 			goto end_rename;
4017 		if (!new.inode->i_nlink)
4018 			ext4_orphan_add(handle, new.inode);
4019 	}
4020 	retval = 0;
4021 
4022 end_rename:
4023 	if (whiteout) {
4024 		if (retval) {
4025 			ext4_resetent(handle, &old,
4026 				      old.inode->i_ino, old_file_type);
4027 			drop_nlink(whiteout);
4028 			ext4_mark_inode_dirty(handle, whiteout);
4029 			ext4_orphan_add(handle, whiteout);
4030 		}
4031 		unlock_new_inode(whiteout);
4032 		ext4_journal_stop(handle);
4033 		iput(whiteout);
4034 	} else {
4035 		ext4_journal_stop(handle);
4036 	}
4037 release_bh:
4038 	brelse(old.dir_bh);
4039 	brelse(old.bh);
4040 	brelse(new.bh);
4041 
4042 	return retval;
4043 }
4044 
4045 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4046 			     struct inode *new_dir, struct dentry *new_dentry)
4047 {
4048 	handle_t *handle = NULL;
4049 	struct ext4_renament old = {
4050 		.dir = old_dir,
4051 		.dentry = old_dentry,
4052 		.inode = d_inode(old_dentry),
4053 	};
4054 	struct ext4_renament new = {
4055 		.dir = new_dir,
4056 		.dentry = new_dentry,
4057 		.inode = d_inode(new_dentry),
4058 	};
4059 	u8 new_file_type;
4060 	int retval;
4061 
4062 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4063 	     !projid_eq(EXT4_I(new_dir)->i_projid,
4064 			EXT4_I(old_dentry->d_inode)->i_projid)) ||
4065 	    (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4066 	     !projid_eq(EXT4_I(old_dir)->i_projid,
4067 			EXT4_I(new_dentry->d_inode)->i_projid)))
4068 		return -EXDEV;
4069 
4070 	retval = dquot_initialize(old.dir);
4071 	if (retval)
4072 		return retval;
4073 	retval = dquot_initialize(new.dir);
4074 	if (retval)
4075 		return retval;
4076 
4077 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4078 				 &old.de, &old.inlined);
4079 	if (IS_ERR(old.bh))
4080 		return PTR_ERR(old.bh);
4081 	/*
4082 	 *  Check for inode number is _not_ due to possible IO errors.
4083 	 *  We might rmdir the source, keep it as pwd of some process
4084 	 *  and merrily kill the link to whatever was created under the
4085 	 *  same name. Goodbye sticky bit ;-<
4086 	 */
4087 	retval = -ENOENT;
4088 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4089 		goto end_rename;
4090 
4091 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4092 				 &new.de, &new.inlined);
4093 	if (IS_ERR(new.bh)) {
4094 		retval = PTR_ERR(new.bh);
4095 		new.bh = NULL;
4096 		goto end_rename;
4097 	}
4098 
4099 	/* RENAME_EXCHANGE case: old *and* new must both exist */
4100 	if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4101 		goto end_rename;
4102 
4103 	handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4104 		(2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4105 		 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4106 	if (IS_ERR(handle)) {
4107 		retval = PTR_ERR(handle);
4108 		handle = NULL;
4109 		goto end_rename;
4110 	}
4111 
4112 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4113 		ext4_handle_sync(handle);
4114 
4115 	if (S_ISDIR(old.inode->i_mode)) {
4116 		retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
4117 		if (retval)
4118 			goto end_rename;
4119 	}
4120 	if (S_ISDIR(new.inode->i_mode)) {
4121 		retval = ext4_rename_dir_prepare(handle, &new, new.dir != old.dir);
4122 		if (retval)
4123 			goto end_rename;
4124 	}
4125 
4126 	/*
4127 	 * Other than the special case of overwriting a directory, parents'
4128 	 * nlink only needs to be modified if this is a cross directory rename.
4129 	 */
4130 	if (old.dir != new.dir && old.is_dir != new.is_dir) {
4131 		old.dir_nlink_delta = old.is_dir ? -1 : 1;
4132 		new.dir_nlink_delta = -old.dir_nlink_delta;
4133 		retval = -EMLINK;
4134 		if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4135 		    (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4136 			goto end_rename;
4137 	}
4138 
4139 	new_file_type = new.de->file_type;
4140 	retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4141 	if (retval)
4142 		goto end_rename;
4143 
4144 	retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4145 	if (retval)
4146 		goto end_rename;
4147 
4148 	/*
4149 	 * Like most other Unix systems, set the ctime for inodes on a
4150 	 * rename.
4151 	 */
4152 	inode_set_ctime_current(old.inode);
4153 	inode_set_ctime_current(new.inode);
4154 	retval = ext4_mark_inode_dirty(handle, old.inode);
4155 	if (unlikely(retval))
4156 		goto end_rename;
4157 	retval = ext4_mark_inode_dirty(handle, new.inode);
4158 	if (unlikely(retval))
4159 		goto end_rename;
4160 	ext4_fc_mark_ineligible(new.inode->i_sb,
4161 				EXT4_FC_REASON_CROSS_RENAME, handle);
4162 	if (old.dir_bh) {
4163 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4164 		if (retval)
4165 			goto end_rename;
4166 	}
4167 	if (new.dir_bh) {
4168 		retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4169 		if (retval)
4170 			goto end_rename;
4171 	}
4172 	ext4_update_dir_count(handle, &old);
4173 	ext4_update_dir_count(handle, &new);
4174 	retval = 0;
4175 
4176 end_rename:
4177 	brelse(old.dir_bh);
4178 	brelse(new.dir_bh);
4179 	brelse(old.bh);
4180 	brelse(new.bh);
4181 	if (handle)
4182 		ext4_journal_stop(handle);
4183 	return retval;
4184 }
4185 
4186 static int ext4_rename2(struct mnt_idmap *idmap,
4187 			struct inode *old_dir, struct dentry *old_dentry,
4188 			struct inode *new_dir, struct dentry *new_dentry,
4189 			unsigned int flags)
4190 {
4191 	int err;
4192 
4193 	if (unlikely(ext4_forced_shutdown(old_dir->i_sb)))
4194 		return -EIO;
4195 
4196 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4197 		return -EINVAL;
4198 
4199 	err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4200 				     flags);
4201 	if (err)
4202 		return err;
4203 
4204 	if (flags & RENAME_EXCHANGE) {
4205 		return ext4_cross_rename(old_dir, old_dentry,
4206 					 new_dir, new_dentry);
4207 	}
4208 
4209 	return ext4_rename(idmap, old_dir, old_dentry, new_dir, new_dentry, flags);
4210 }
4211 
4212 /*
4213  * directories can handle most operations...
4214  */
4215 const struct inode_operations ext4_dir_inode_operations = {
4216 	.create		= ext4_create,
4217 	.lookup		= ext4_lookup,
4218 	.link		= ext4_link,
4219 	.unlink		= ext4_unlink,
4220 	.symlink	= ext4_symlink,
4221 	.mkdir		= ext4_mkdir,
4222 	.rmdir		= ext4_rmdir,
4223 	.mknod		= ext4_mknod,
4224 	.tmpfile	= ext4_tmpfile,
4225 	.rename		= ext4_rename2,
4226 	.setattr	= ext4_setattr,
4227 	.getattr	= ext4_getattr,
4228 	.listxattr	= ext4_listxattr,
4229 	.get_inode_acl	= ext4_get_acl,
4230 	.set_acl	= ext4_set_acl,
4231 	.fiemap         = ext4_fiemap,
4232 	.fileattr_get	= ext4_fileattr_get,
4233 	.fileattr_set	= ext4_fileattr_set,
4234 };
4235 
4236 const struct inode_operations ext4_special_inode_operations = {
4237 	.setattr	= ext4_setattr,
4238 	.getattr	= ext4_getattr,
4239 	.listxattr	= ext4_listxattr,
4240 	.get_inode_acl	= ext4_get_acl,
4241 	.set_acl	= ext4_set_acl,
4242 };
4243