xref: /openbmc/linux/fs/ext4/namei.c (revision 7f877908)
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 buffer_head *bh;
58 	int err;
59 
60 	if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
61 		     ((inode->i_size >> 10) >=
62 		      EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
63 		return ERR_PTR(-ENOSPC);
64 
65 	*block = inode->i_size >> inode->i_sb->s_blocksize_bits;
66 
67 	bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
68 	if (IS_ERR(bh))
69 		return bh;
70 	inode->i_size += inode->i_sb->s_blocksize;
71 	EXT4_I(inode)->i_disksize = inode->i_size;
72 	BUFFER_TRACE(bh, "get_write_access");
73 	err = ext4_journal_get_write_access(handle, bh);
74 	if (err) {
75 		brelse(bh);
76 		ext4_std_error(inode->i_sb, err);
77 		return ERR_PTR(err);
78 	}
79 	return bh;
80 }
81 
82 static int ext4_dx_csum_verify(struct inode *inode,
83 			       struct ext4_dir_entry *dirent);
84 
85 /*
86  * Hints to ext4_read_dirblock regarding whether we expect a directory
87  * block being read to be an index block, or a block containing
88  * directory entries (and if the latter, whether it was found via a
89  * logical block in an htree index block).  This is used to control
90  * what sort of sanity checkinig ext4_read_dirblock() will do on the
91  * directory block read from the storage device.  EITHER will means
92  * the caller doesn't know what kind of directory block will be read,
93  * so no specific verification will be done.
94  */
95 typedef enum {
96 	EITHER, INDEX, DIRENT, DIRENT_HTREE
97 } dirblock_type_t;
98 
99 #define ext4_read_dirblock(inode, block, type) \
100 	__ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
101 
102 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
103 						ext4_lblk_t block,
104 						dirblock_type_t type,
105 						const char *func,
106 						unsigned int line)
107 {
108 	struct buffer_head *bh;
109 	struct ext4_dir_entry *dirent;
110 	int is_dx_block = 0;
111 
112 	if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
113 		bh = ERR_PTR(-EIO);
114 	else
115 		bh = ext4_bread(NULL, inode, block, 0);
116 	if (IS_ERR(bh)) {
117 		__ext4_warning(inode->i_sb, func, line,
118 			       "inode #%lu: lblock %lu: comm %s: "
119 			       "error %ld reading directory block",
120 			       inode->i_ino, (unsigned long)block,
121 			       current->comm, PTR_ERR(bh));
122 
123 		return bh;
124 	}
125 	if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
126 		ext4_error_inode(inode, func, line, block,
127 				 "Directory hole found for htree %s block",
128 				 (type == INDEX) ? "index" : "leaf");
129 		return ERR_PTR(-EFSCORRUPTED);
130 	}
131 	if (!bh)
132 		return NULL;
133 	dirent = (struct ext4_dir_entry *) bh->b_data;
134 	/* Determine whether or not we have an index block */
135 	if (is_dx(inode)) {
136 		if (block == 0)
137 			is_dx_block = 1;
138 		else if (ext4_rec_len_from_disk(dirent->rec_len,
139 						inode->i_sb->s_blocksize) ==
140 			 inode->i_sb->s_blocksize)
141 			is_dx_block = 1;
142 	}
143 	if (!is_dx_block && type == INDEX) {
144 		ext4_error_inode(inode, func, line, block,
145 		       "directory leaf block found instead of index block");
146 		brelse(bh);
147 		return ERR_PTR(-EFSCORRUPTED);
148 	}
149 	if (!ext4_has_metadata_csum(inode->i_sb) ||
150 	    buffer_verified(bh))
151 		return bh;
152 
153 	/*
154 	 * An empty leaf block can get mistaken for a index block; for
155 	 * this reason, we can only check the index checksum when the
156 	 * caller is sure it should be an index block.
157 	 */
158 	if (is_dx_block && type == INDEX) {
159 		if (ext4_dx_csum_verify(inode, dirent) &&
160 		    !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
161 			set_buffer_verified(bh);
162 		else {
163 			ext4_set_errno(inode->i_sb, EFSBADCRC);
164 			ext4_error_inode(inode, func, line, block,
165 					 "Directory index failed checksum");
166 			brelse(bh);
167 			return ERR_PTR(-EFSBADCRC);
168 		}
169 	}
170 	if (!is_dx_block) {
171 		if (ext4_dirblock_csum_verify(inode, bh) &&
172 		    !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
173 			set_buffer_verified(bh);
174 		else {
175 			ext4_set_errno(inode->i_sb, EFSBADCRC);
176 			ext4_error_inode(inode, func, line, block,
177 					 "Directory block failed checksum");
178 			brelse(bh);
179 			return ERR_PTR(-EFSBADCRC);
180 		}
181 	}
182 	return bh;
183 }
184 
185 #ifndef assert
186 #define assert(test) J_ASSERT(test)
187 #endif
188 
189 #ifdef DX_DEBUG
190 #define dxtrace(command) command
191 #else
192 #define dxtrace(command)
193 #endif
194 
195 struct fake_dirent
196 {
197 	__le32 inode;
198 	__le16 rec_len;
199 	u8 name_len;
200 	u8 file_type;
201 };
202 
203 struct dx_countlimit
204 {
205 	__le16 limit;
206 	__le16 count;
207 };
208 
209 struct dx_entry
210 {
211 	__le32 hash;
212 	__le32 block;
213 };
214 
215 /*
216  * dx_root_info is laid out so that if it should somehow get overlaid by a
217  * dirent the two low bits of the hash version will be zero.  Therefore, the
218  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
219  */
220 
221 struct dx_root
222 {
223 	struct fake_dirent dot;
224 	char dot_name[4];
225 	struct fake_dirent dotdot;
226 	char dotdot_name[4];
227 	struct dx_root_info
228 	{
229 		__le32 reserved_zero;
230 		u8 hash_version;
231 		u8 info_length; /* 8 */
232 		u8 indirect_levels;
233 		u8 unused_flags;
234 	}
235 	info;
236 	struct dx_entry	entries[0];
237 };
238 
239 struct dx_node
240 {
241 	struct fake_dirent fake;
242 	struct dx_entry	entries[0];
243 };
244 
245 
246 struct dx_frame
247 {
248 	struct buffer_head *bh;
249 	struct dx_entry *entries;
250 	struct dx_entry *at;
251 };
252 
253 struct dx_map_entry
254 {
255 	u32 hash;
256 	u16 offs;
257 	u16 size;
258 };
259 
260 /*
261  * This goes at the end of each htree block.
262  */
263 struct dx_tail {
264 	u32 dt_reserved;
265 	__le32 dt_checksum;	/* crc32c(uuid+inum+dirblock) */
266 };
267 
268 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
269 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
270 static inline unsigned dx_get_hash(struct dx_entry *entry);
271 static void dx_set_hash(struct dx_entry *entry, unsigned value);
272 static unsigned dx_get_count(struct dx_entry *entries);
273 static unsigned dx_get_limit(struct dx_entry *entries);
274 static void dx_set_count(struct dx_entry *entries, unsigned value);
275 static void dx_set_limit(struct dx_entry *entries, unsigned value);
276 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
277 static unsigned dx_node_limit(struct inode *dir);
278 static struct dx_frame *dx_probe(struct ext4_filename *fname,
279 				 struct inode *dir,
280 				 struct dx_hash_info *hinfo,
281 				 struct dx_frame *frame);
282 static void dx_release(struct dx_frame *frames);
283 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
284 		       unsigned blocksize, struct dx_hash_info *hinfo,
285 		       struct dx_map_entry map[]);
286 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
287 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
288 		struct dx_map_entry *offsets, int count, unsigned blocksize);
289 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
290 static void dx_insert_block(struct dx_frame *frame,
291 					u32 hash, ext4_lblk_t block);
292 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
293 				 struct dx_frame *frame,
294 				 struct dx_frame *frames,
295 				 __u32 *start_hash);
296 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
297 		struct ext4_filename *fname,
298 		struct ext4_dir_entry_2 **res_dir);
299 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
300 			     struct inode *dir, struct inode *inode);
301 
302 /* checksumming functions */
303 void ext4_initialize_dirent_tail(struct buffer_head *bh,
304 				 unsigned int blocksize)
305 {
306 	struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
307 
308 	memset(t, 0, sizeof(struct ext4_dir_entry_tail));
309 	t->det_rec_len = ext4_rec_len_to_disk(
310 			sizeof(struct ext4_dir_entry_tail), blocksize);
311 	t->det_reserved_ft = EXT4_FT_DIR_CSUM;
312 }
313 
314 /* Walk through a dirent block to find a checksum "dirent" at the tail */
315 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
316 						   struct buffer_head *bh)
317 {
318 	struct ext4_dir_entry_tail *t;
319 
320 #ifdef PARANOID
321 	struct ext4_dir_entry *d, *top;
322 
323 	d = (struct ext4_dir_entry *)bh->b_data;
324 	top = (struct ext4_dir_entry *)(bh->b_data +
325 		(EXT4_BLOCK_SIZE(inode->i_sb) -
326 		 sizeof(struct ext4_dir_entry_tail)));
327 	while (d < top && d->rec_len)
328 		d = (struct ext4_dir_entry *)(((void *)d) +
329 		    le16_to_cpu(d->rec_len));
330 
331 	if (d != top)
332 		return NULL;
333 
334 	t = (struct ext4_dir_entry_tail *)d;
335 #else
336 	t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
337 #endif
338 
339 	if (t->det_reserved_zero1 ||
340 	    le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
341 	    t->det_reserved_zero2 ||
342 	    t->det_reserved_ft != EXT4_FT_DIR_CSUM)
343 		return NULL;
344 
345 	return t;
346 }
347 
348 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
349 {
350 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
351 	struct ext4_inode_info *ei = EXT4_I(inode);
352 	__u32 csum;
353 
354 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
355 	return cpu_to_le32(csum);
356 }
357 
358 #define warn_no_space_for_csum(inode)					\
359 	__warn_no_space_for_csum((inode), __func__, __LINE__)
360 
361 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
362 				     unsigned int line)
363 {
364 	__ext4_warning_inode(inode, func, line,
365 		"No space for directory leaf checksum. Please run e2fsck -D.");
366 }
367 
368 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
369 {
370 	struct ext4_dir_entry_tail *t;
371 
372 	if (!ext4_has_metadata_csum(inode->i_sb))
373 		return 1;
374 
375 	t = get_dirent_tail(inode, bh);
376 	if (!t) {
377 		warn_no_space_for_csum(inode);
378 		return 0;
379 	}
380 
381 	if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
382 						  (char *)t - bh->b_data))
383 		return 0;
384 
385 	return 1;
386 }
387 
388 static void ext4_dirblock_csum_set(struct inode *inode,
389 				 struct buffer_head *bh)
390 {
391 	struct ext4_dir_entry_tail *t;
392 
393 	if (!ext4_has_metadata_csum(inode->i_sb))
394 		return;
395 
396 	t = get_dirent_tail(inode, bh);
397 	if (!t) {
398 		warn_no_space_for_csum(inode);
399 		return;
400 	}
401 
402 	t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
403 					     (char *)t - bh->b_data);
404 }
405 
406 int ext4_handle_dirty_dirblock(handle_t *handle,
407 			       struct inode *inode,
408 			       struct buffer_head *bh)
409 {
410 	ext4_dirblock_csum_set(inode, bh);
411 	return ext4_handle_dirty_metadata(handle, inode, bh);
412 }
413 
414 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
415 					       struct ext4_dir_entry *dirent,
416 					       int *offset)
417 {
418 	struct ext4_dir_entry *dp;
419 	struct dx_root_info *root;
420 	int count_offset;
421 
422 	if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
423 		count_offset = 8;
424 	else if (le16_to_cpu(dirent->rec_len) == 12) {
425 		dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
426 		if (le16_to_cpu(dp->rec_len) !=
427 		    EXT4_BLOCK_SIZE(inode->i_sb) - 12)
428 			return NULL;
429 		root = (struct dx_root_info *)(((void *)dp + 12));
430 		if (root->reserved_zero ||
431 		    root->info_length != sizeof(struct dx_root_info))
432 			return NULL;
433 		count_offset = 32;
434 	} else
435 		return NULL;
436 
437 	if (offset)
438 		*offset = count_offset;
439 	return (struct dx_countlimit *)(((void *)dirent) + count_offset);
440 }
441 
442 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
443 			   int count_offset, int count, struct dx_tail *t)
444 {
445 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
446 	struct ext4_inode_info *ei = EXT4_I(inode);
447 	__u32 csum;
448 	int size;
449 	__u32 dummy_csum = 0;
450 	int offset = offsetof(struct dx_tail, dt_checksum);
451 
452 	size = count_offset + (count * sizeof(struct dx_entry));
453 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
454 	csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
455 	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
456 
457 	return cpu_to_le32(csum);
458 }
459 
460 static int ext4_dx_csum_verify(struct inode *inode,
461 			       struct ext4_dir_entry *dirent)
462 {
463 	struct dx_countlimit *c;
464 	struct dx_tail *t;
465 	int count_offset, limit, count;
466 
467 	if (!ext4_has_metadata_csum(inode->i_sb))
468 		return 1;
469 
470 	c = get_dx_countlimit(inode, dirent, &count_offset);
471 	if (!c) {
472 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
473 		return 0;
474 	}
475 	limit = le16_to_cpu(c->limit);
476 	count = le16_to_cpu(c->count);
477 	if (count_offset + (limit * sizeof(struct dx_entry)) >
478 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
479 		warn_no_space_for_csum(inode);
480 		return 0;
481 	}
482 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
483 
484 	if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
485 					    count, t))
486 		return 0;
487 	return 1;
488 }
489 
490 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
491 {
492 	struct dx_countlimit *c;
493 	struct dx_tail *t;
494 	int count_offset, limit, count;
495 
496 	if (!ext4_has_metadata_csum(inode->i_sb))
497 		return;
498 
499 	c = get_dx_countlimit(inode, dirent, &count_offset);
500 	if (!c) {
501 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
502 		return;
503 	}
504 	limit = le16_to_cpu(c->limit);
505 	count = le16_to_cpu(c->count);
506 	if (count_offset + (limit * sizeof(struct dx_entry)) >
507 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
508 		warn_no_space_for_csum(inode);
509 		return;
510 	}
511 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
512 
513 	t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
514 }
515 
516 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
517 					    struct inode *inode,
518 					    struct buffer_head *bh)
519 {
520 	ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
521 	return ext4_handle_dirty_metadata(handle, inode, bh);
522 }
523 
524 /*
525  * p is at least 6 bytes before the end of page
526  */
527 static inline struct ext4_dir_entry_2 *
528 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
529 {
530 	return (struct ext4_dir_entry_2 *)((char *)p +
531 		ext4_rec_len_from_disk(p->rec_len, blocksize));
532 }
533 
534 /*
535  * Future: use high four bits of block for coalesce-on-delete flags
536  * Mask them off for now.
537  */
538 
539 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
540 {
541 	return le32_to_cpu(entry->block) & 0x0fffffff;
542 }
543 
544 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
545 {
546 	entry->block = cpu_to_le32(value);
547 }
548 
549 static inline unsigned dx_get_hash(struct dx_entry *entry)
550 {
551 	return le32_to_cpu(entry->hash);
552 }
553 
554 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
555 {
556 	entry->hash = cpu_to_le32(value);
557 }
558 
559 static inline unsigned dx_get_count(struct dx_entry *entries)
560 {
561 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
562 }
563 
564 static inline unsigned dx_get_limit(struct dx_entry *entries)
565 {
566 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
567 }
568 
569 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
570 {
571 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
572 }
573 
574 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
575 {
576 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
577 }
578 
579 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
580 {
581 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
582 		EXT4_DIR_REC_LEN(2) - infosize;
583 
584 	if (ext4_has_metadata_csum(dir->i_sb))
585 		entry_space -= sizeof(struct dx_tail);
586 	return entry_space / sizeof(struct dx_entry);
587 }
588 
589 static inline unsigned dx_node_limit(struct inode *dir)
590 {
591 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
592 
593 	if (ext4_has_metadata_csum(dir->i_sb))
594 		entry_space -= sizeof(struct dx_tail);
595 	return entry_space / sizeof(struct dx_entry);
596 }
597 
598 /*
599  * Debug
600  */
601 #ifdef DX_DEBUG
602 static void dx_show_index(char * label, struct dx_entry *entries)
603 {
604 	int i, n = dx_get_count (entries);
605 	printk(KERN_DEBUG "%s index", label);
606 	for (i = 0; i < n; i++) {
607 		printk(KERN_CONT " %x->%lu",
608 		       i ? dx_get_hash(entries + i) : 0,
609 		       (unsigned long)dx_get_block(entries + i));
610 	}
611 	printk(KERN_CONT "\n");
612 }
613 
614 struct stats
615 {
616 	unsigned names;
617 	unsigned space;
618 	unsigned bcount;
619 };
620 
621 static struct stats dx_show_leaf(struct inode *dir,
622 				struct dx_hash_info *hinfo,
623 				struct ext4_dir_entry_2 *de,
624 				int size, int show_names)
625 {
626 	unsigned names = 0, space = 0;
627 	char *base = (char *) de;
628 	struct dx_hash_info h = *hinfo;
629 
630 	printk("names: ");
631 	while ((char *) de < base + size)
632 	{
633 		if (de->inode)
634 		{
635 			if (show_names)
636 			{
637 #ifdef CONFIG_FS_ENCRYPTION
638 				int len;
639 				char *name;
640 				struct fscrypt_str fname_crypto_str =
641 					FSTR_INIT(NULL, 0);
642 				int res = 0;
643 
644 				name  = de->name;
645 				len = de->name_len;
646 				if (IS_ENCRYPTED(dir))
647 					res = fscrypt_get_encryption_info(dir);
648 				if (res) {
649 					printk(KERN_WARNING "Error setting up"
650 					       " fname crypto: %d\n", res);
651 				}
652 				if (!fscrypt_has_encryption_key(dir)) {
653 					/* Directory is not encrypted */
654 					ext4fs_dirhash(dir, de->name,
655 						de->name_len, &h);
656 					printk("%*.s:(U)%x.%u ", len,
657 					       name, h.hash,
658 					       (unsigned) ((char *) de
659 							   - base));
660 				} else {
661 					struct fscrypt_str de_name =
662 						FSTR_INIT(name, len);
663 
664 					/* Directory is encrypted */
665 					res = fscrypt_fname_alloc_buffer(
666 						dir, len,
667 						&fname_crypto_str);
668 					if (res)
669 						printk(KERN_WARNING "Error "
670 							"allocating crypto "
671 							"buffer--skipping "
672 							"crypto\n");
673 					res = fscrypt_fname_disk_to_usr(dir,
674 						0, 0, &de_name,
675 						&fname_crypto_str);
676 					if (res) {
677 						printk(KERN_WARNING "Error "
678 							"converting filename "
679 							"from disk to usr"
680 							"\n");
681 						name = "??";
682 						len = 2;
683 					} else {
684 						name = fname_crypto_str.name;
685 						len = fname_crypto_str.len;
686 					}
687 					ext4fs_dirhash(dir, de->name,
688 						       de->name_len, &h);
689 					printk("%*.s:(E)%x.%u ", len, name,
690 					       h.hash, (unsigned) ((char *) de
691 								   - base));
692 					fscrypt_fname_free_buffer(
693 							&fname_crypto_str);
694 				}
695 #else
696 				int len = de->name_len;
697 				char *name = de->name;
698 				ext4fs_dirhash(dir, de->name, de->name_len, &h);
699 				printk("%*.s:%x.%u ", len, name, h.hash,
700 				       (unsigned) ((char *) de - base));
701 #endif
702 			}
703 			space += EXT4_DIR_REC_LEN(de->name_len);
704 			names++;
705 		}
706 		de = ext4_next_entry(de, size);
707 	}
708 	printk(KERN_CONT "(%i)\n", names);
709 	return (struct stats) { names, space, 1 };
710 }
711 
712 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
713 			     struct dx_entry *entries, int levels)
714 {
715 	unsigned blocksize = dir->i_sb->s_blocksize;
716 	unsigned count = dx_get_count(entries), names = 0, space = 0, i;
717 	unsigned bcount = 0;
718 	struct buffer_head *bh;
719 	printk("%i indexed blocks...\n", count);
720 	for (i = 0; i < count; i++, entries++)
721 	{
722 		ext4_lblk_t block = dx_get_block(entries);
723 		ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
724 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
725 		struct stats stats;
726 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
727 		bh = ext4_bread(NULL,dir, block, 0);
728 		if (!bh || IS_ERR(bh))
729 			continue;
730 		stats = levels?
731 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
732 		   dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
733 			bh->b_data, blocksize, 0);
734 		names += stats.names;
735 		space += stats.space;
736 		bcount += stats.bcount;
737 		brelse(bh);
738 	}
739 	if (bcount)
740 		printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
741 		       levels ? "" : "   ", names, space/bcount,
742 		       (space/bcount)*100/blocksize);
743 	return (struct stats) { names, space, bcount};
744 }
745 #endif /* DX_DEBUG */
746 
747 /*
748  * Probe for a directory leaf block to search.
749  *
750  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
751  * error in the directory index, and the caller should fall back to
752  * searching the directory normally.  The callers of dx_probe **MUST**
753  * check for this error code, and make sure it never gets reflected
754  * back to userspace.
755  */
756 static struct dx_frame *
757 dx_probe(struct ext4_filename *fname, struct inode *dir,
758 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
759 {
760 	unsigned count, indirect;
761 	struct dx_entry *at, *entries, *p, *q, *m;
762 	struct dx_root *root;
763 	struct dx_frame *frame = frame_in;
764 	struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
765 	u32 hash;
766 
767 	memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
768 	frame->bh = ext4_read_dirblock(dir, 0, INDEX);
769 	if (IS_ERR(frame->bh))
770 		return (struct dx_frame *) frame->bh;
771 
772 	root = (struct dx_root *) frame->bh->b_data;
773 	if (root->info.hash_version != DX_HASH_TEA &&
774 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
775 	    root->info.hash_version != DX_HASH_LEGACY) {
776 		ext4_warning_inode(dir, "Unrecognised inode hash code %u",
777 				   root->info.hash_version);
778 		goto fail;
779 	}
780 	if (fname)
781 		hinfo = &fname->hinfo;
782 	hinfo->hash_version = root->info.hash_version;
783 	if (hinfo->hash_version <= DX_HASH_TEA)
784 		hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
785 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
786 	if (fname && fname_name(fname))
787 		ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
788 	hash = hinfo->hash;
789 
790 	if (root->info.unused_flags & 1) {
791 		ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
792 				   root->info.unused_flags);
793 		goto fail;
794 	}
795 
796 	indirect = root->info.indirect_levels;
797 	if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
798 		ext4_warning(dir->i_sb,
799 			     "Directory (ino: %lu) htree depth %#06x exceed"
800 			     "supported value", dir->i_ino,
801 			     ext4_dir_htree_level(dir->i_sb));
802 		if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
803 			ext4_warning(dir->i_sb, "Enable large directory "
804 						"feature to access it");
805 		}
806 		goto fail;
807 	}
808 
809 	entries = (struct dx_entry *)(((char *)&root->info) +
810 				      root->info.info_length);
811 
812 	if (dx_get_limit(entries) != dx_root_limit(dir,
813 						   root->info.info_length)) {
814 		ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
815 				   dx_get_limit(entries),
816 				   dx_root_limit(dir, root->info.info_length));
817 		goto fail;
818 	}
819 
820 	dxtrace(printk("Look up %x", hash));
821 	while (1) {
822 		count = dx_get_count(entries);
823 		if (!count || count > dx_get_limit(entries)) {
824 			ext4_warning_inode(dir,
825 					   "dx entry: count %u beyond limit %u",
826 					   count, dx_get_limit(entries));
827 			goto fail;
828 		}
829 
830 		p = entries + 1;
831 		q = entries + count - 1;
832 		while (p <= q) {
833 			m = p + (q - p) / 2;
834 			dxtrace(printk(KERN_CONT "."));
835 			if (dx_get_hash(m) > hash)
836 				q = m - 1;
837 			else
838 				p = m + 1;
839 		}
840 
841 		if (0) { // linear search cross check
842 			unsigned n = count - 1;
843 			at = entries;
844 			while (n--)
845 			{
846 				dxtrace(printk(KERN_CONT ","));
847 				if (dx_get_hash(++at) > hash)
848 				{
849 					at--;
850 					break;
851 				}
852 			}
853 			assert (at == p - 1);
854 		}
855 
856 		at = p - 1;
857 		dxtrace(printk(KERN_CONT " %x->%u\n",
858 			       at == entries ? 0 : dx_get_hash(at),
859 			       dx_get_block(at)));
860 		frame->entries = entries;
861 		frame->at = at;
862 		if (!indirect--)
863 			return frame;
864 		frame++;
865 		frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
866 		if (IS_ERR(frame->bh)) {
867 			ret_err = (struct dx_frame *) frame->bh;
868 			frame->bh = NULL;
869 			goto fail;
870 		}
871 		entries = ((struct dx_node *) frame->bh->b_data)->entries;
872 
873 		if (dx_get_limit(entries) != dx_node_limit(dir)) {
874 			ext4_warning_inode(dir,
875 				"dx entry: limit %u != node limit %u",
876 				dx_get_limit(entries), dx_node_limit(dir));
877 			goto fail;
878 		}
879 	}
880 fail:
881 	while (frame >= frame_in) {
882 		brelse(frame->bh);
883 		frame--;
884 	}
885 
886 	if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
887 		ext4_warning_inode(dir,
888 			"Corrupt directory, running e2fsck is recommended");
889 	return ret_err;
890 }
891 
892 static void dx_release(struct dx_frame *frames)
893 {
894 	struct dx_root_info *info;
895 	int i;
896 	unsigned int indirect_levels;
897 
898 	if (frames[0].bh == NULL)
899 		return;
900 
901 	info = &((struct dx_root *)frames[0].bh->b_data)->info;
902 	/* save local copy, "info" may be freed after brelse() */
903 	indirect_levels = info->indirect_levels;
904 	for (i = 0; i <= indirect_levels; i++) {
905 		if (frames[i].bh == NULL)
906 			break;
907 		brelse(frames[i].bh);
908 		frames[i].bh = NULL;
909 	}
910 }
911 
912 /*
913  * This function increments the frame pointer to search the next leaf
914  * block, and reads in the necessary intervening nodes if the search
915  * should be necessary.  Whether or not the search is necessary is
916  * controlled by the hash parameter.  If the hash value is even, then
917  * the search is only continued if the next block starts with that
918  * hash value.  This is used if we are searching for a specific file.
919  *
920  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
921  *
922  * This function returns 1 if the caller should continue to search,
923  * or 0 if it should not.  If there is an error reading one of the
924  * index blocks, it will a negative error code.
925  *
926  * If start_hash is non-null, it will be filled in with the starting
927  * hash of the next page.
928  */
929 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
930 				 struct dx_frame *frame,
931 				 struct dx_frame *frames,
932 				 __u32 *start_hash)
933 {
934 	struct dx_frame *p;
935 	struct buffer_head *bh;
936 	int num_frames = 0;
937 	__u32 bhash;
938 
939 	p = frame;
940 	/*
941 	 * Find the next leaf page by incrementing the frame pointer.
942 	 * If we run out of entries in the interior node, loop around and
943 	 * increment pointer in the parent node.  When we break out of
944 	 * this loop, num_frames indicates the number of interior
945 	 * nodes need to be read.
946 	 */
947 	while (1) {
948 		if (++(p->at) < p->entries + dx_get_count(p->entries))
949 			break;
950 		if (p == frames)
951 			return 0;
952 		num_frames++;
953 		p--;
954 	}
955 
956 	/*
957 	 * If the hash is 1, then continue only if the next page has a
958 	 * continuation hash of any value.  This is used for readdir
959 	 * handling.  Otherwise, check to see if the hash matches the
960 	 * desired contiuation hash.  If it doesn't, return since
961 	 * there's no point to read in the successive index pages.
962 	 */
963 	bhash = dx_get_hash(p->at);
964 	if (start_hash)
965 		*start_hash = bhash;
966 	if ((hash & 1) == 0) {
967 		if ((bhash & ~1) != hash)
968 			return 0;
969 	}
970 	/*
971 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
972 	 * block so no check is necessary
973 	 */
974 	while (num_frames--) {
975 		bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
976 		if (IS_ERR(bh))
977 			return PTR_ERR(bh);
978 		p++;
979 		brelse(p->bh);
980 		p->bh = bh;
981 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
982 	}
983 	return 1;
984 }
985 
986 
987 /*
988  * This function fills a red-black tree with information from a
989  * directory block.  It returns the number directory entries loaded
990  * into the tree.  If there is an error it is returned in err.
991  */
992 static int htree_dirblock_to_tree(struct file *dir_file,
993 				  struct inode *dir, ext4_lblk_t block,
994 				  struct dx_hash_info *hinfo,
995 				  __u32 start_hash, __u32 start_minor_hash)
996 {
997 	struct buffer_head *bh;
998 	struct ext4_dir_entry_2 *de, *top;
999 	int err = 0, count = 0;
1000 	struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1001 
1002 	dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1003 							(unsigned long)block));
1004 	bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1005 	if (IS_ERR(bh))
1006 		return PTR_ERR(bh);
1007 
1008 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1009 	top = (struct ext4_dir_entry_2 *) ((char *) de +
1010 					   dir->i_sb->s_blocksize -
1011 					   EXT4_DIR_REC_LEN(0));
1012 	/* Check if the directory is encrypted */
1013 	if (IS_ENCRYPTED(dir)) {
1014 		err = fscrypt_get_encryption_info(dir);
1015 		if (err < 0) {
1016 			brelse(bh);
1017 			return err;
1018 		}
1019 		err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1020 						     &fname_crypto_str);
1021 		if (err < 0) {
1022 			brelse(bh);
1023 			return err;
1024 		}
1025 	}
1026 
1027 	for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1028 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1029 				bh->b_data, bh->b_size,
1030 				(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1031 					 + ((char *)de - bh->b_data))) {
1032 			/* silently ignore the rest of the block */
1033 			break;
1034 		}
1035 		ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1036 		if ((hinfo->hash < start_hash) ||
1037 		    ((hinfo->hash == start_hash) &&
1038 		     (hinfo->minor_hash < start_minor_hash)))
1039 			continue;
1040 		if (de->inode == 0)
1041 			continue;
1042 		if (!IS_ENCRYPTED(dir)) {
1043 			tmp_str.name = de->name;
1044 			tmp_str.len = de->name_len;
1045 			err = ext4_htree_store_dirent(dir_file,
1046 				   hinfo->hash, hinfo->minor_hash, de,
1047 				   &tmp_str);
1048 		} else {
1049 			int save_len = fname_crypto_str.len;
1050 			struct fscrypt_str de_name = FSTR_INIT(de->name,
1051 								de->name_len);
1052 
1053 			/* Directory is encrypted */
1054 			err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1055 					hinfo->minor_hash, &de_name,
1056 					&fname_crypto_str);
1057 			if (err) {
1058 				count = err;
1059 				goto errout;
1060 			}
1061 			err = ext4_htree_store_dirent(dir_file,
1062 				   hinfo->hash, hinfo->minor_hash, de,
1063 					&fname_crypto_str);
1064 			fname_crypto_str.len = save_len;
1065 		}
1066 		if (err != 0) {
1067 			count = err;
1068 			goto errout;
1069 		}
1070 		count++;
1071 	}
1072 errout:
1073 	brelse(bh);
1074 	fscrypt_fname_free_buffer(&fname_crypto_str);
1075 	return count;
1076 }
1077 
1078 
1079 /*
1080  * This function fills a red-black tree with information from a
1081  * directory.  We start scanning the directory in hash order, starting
1082  * at start_hash and start_minor_hash.
1083  *
1084  * This function returns the number of entries inserted into the tree,
1085  * or a negative error code.
1086  */
1087 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1088 			 __u32 start_minor_hash, __u32 *next_hash)
1089 {
1090 	struct dx_hash_info hinfo;
1091 	struct ext4_dir_entry_2 *de;
1092 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1093 	struct inode *dir;
1094 	ext4_lblk_t block;
1095 	int count = 0;
1096 	int ret, err;
1097 	__u32 hashval;
1098 	struct fscrypt_str tmp_str;
1099 
1100 	dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1101 		       start_hash, start_minor_hash));
1102 	dir = file_inode(dir_file);
1103 	if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1104 		hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1105 		if (hinfo.hash_version <= DX_HASH_TEA)
1106 			hinfo.hash_version +=
1107 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
1108 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1109 		if (ext4_has_inline_data(dir)) {
1110 			int has_inline_data = 1;
1111 			count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1112 						       &hinfo, start_hash,
1113 						       start_minor_hash,
1114 						       &has_inline_data);
1115 			if (has_inline_data) {
1116 				*next_hash = ~0;
1117 				return count;
1118 			}
1119 		}
1120 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1121 					       start_hash, start_minor_hash);
1122 		*next_hash = ~0;
1123 		return count;
1124 	}
1125 	hinfo.hash = start_hash;
1126 	hinfo.minor_hash = 0;
1127 	frame = dx_probe(NULL, dir, &hinfo, frames);
1128 	if (IS_ERR(frame))
1129 		return PTR_ERR(frame);
1130 
1131 	/* Add '.' and '..' from the htree header */
1132 	if (!start_hash && !start_minor_hash) {
1133 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1134 		tmp_str.name = de->name;
1135 		tmp_str.len = de->name_len;
1136 		err = ext4_htree_store_dirent(dir_file, 0, 0,
1137 					      de, &tmp_str);
1138 		if (err != 0)
1139 			goto errout;
1140 		count++;
1141 	}
1142 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1143 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1144 		de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1145 		tmp_str.name = de->name;
1146 		tmp_str.len = de->name_len;
1147 		err = ext4_htree_store_dirent(dir_file, 2, 0,
1148 					      de, &tmp_str);
1149 		if (err != 0)
1150 			goto errout;
1151 		count++;
1152 	}
1153 
1154 	while (1) {
1155 		if (fatal_signal_pending(current)) {
1156 			err = -ERESTARTSYS;
1157 			goto errout;
1158 		}
1159 		cond_resched();
1160 		block = dx_get_block(frame->at);
1161 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1162 					     start_hash, start_minor_hash);
1163 		if (ret < 0) {
1164 			err = ret;
1165 			goto errout;
1166 		}
1167 		count += ret;
1168 		hashval = ~0;
1169 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1170 					    frame, frames, &hashval);
1171 		*next_hash = hashval;
1172 		if (ret < 0) {
1173 			err = ret;
1174 			goto errout;
1175 		}
1176 		/*
1177 		 * Stop if:  (a) there are no more entries, or
1178 		 * (b) we have inserted at least one entry and the
1179 		 * next hash value is not a continuation
1180 		 */
1181 		if ((ret == 0) ||
1182 		    (count && ((hashval & 1) == 0)))
1183 			break;
1184 	}
1185 	dx_release(frames);
1186 	dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1187 		       "next hash: %x\n", count, *next_hash));
1188 	return count;
1189 errout:
1190 	dx_release(frames);
1191 	return (err);
1192 }
1193 
1194 static inline int search_dirblock(struct buffer_head *bh,
1195 				  struct inode *dir,
1196 				  struct ext4_filename *fname,
1197 				  unsigned int offset,
1198 				  struct ext4_dir_entry_2 **res_dir)
1199 {
1200 	return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1201 			       fname, offset, res_dir);
1202 }
1203 
1204 /*
1205  * Directory block splitting, compacting
1206  */
1207 
1208 /*
1209  * Create map of hash values, offsets, and sizes, stored at end of block.
1210  * Returns number of entries mapped.
1211  */
1212 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1213 		       unsigned blocksize, struct dx_hash_info *hinfo,
1214 		       struct dx_map_entry *map_tail)
1215 {
1216 	int count = 0;
1217 	char *base = (char *) de;
1218 	struct dx_hash_info h = *hinfo;
1219 
1220 	while ((char *) de < base + blocksize) {
1221 		if (de->name_len && de->inode) {
1222 			ext4fs_dirhash(dir, de->name, de->name_len, &h);
1223 			map_tail--;
1224 			map_tail->hash = h.hash;
1225 			map_tail->offs = ((char *) de - base)>>2;
1226 			map_tail->size = le16_to_cpu(de->rec_len);
1227 			count++;
1228 			cond_resched();
1229 		}
1230 		/* XXX: do we need to check rec_len == 0 case? -Chris */
1231 		de = ext4_next_entry(de, blocksize);
1232 	}
1233 	return count;
1234 }
1235 
1236 /* Sort map by hash value */
1237 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1238 {
1239 	struct dx_map_entry *p, *q, *top = map + count - 1;
1240 	int more;
1241 	/* Combsort until bubble sort doesn't suck */
1242 	while (count > 2) {
1243 		count = count*10/13;
1244 		if (count - 9 < 2) /* 9, 10 -> 11 */
1245 			count = 11;
1246 		for (p = top, q = p - count; q >= map; p--, q--)
1247 			if (p->hash < q->hash)
1248 				swap(*p, *q);
1249 	}
1250 	/* Garden variety bubble sort */
1251 	do {
1252 		more = 0;
1253 		q = top;
1254 		while (q-- > map) {
1255 			if (q[1].hash >= q[0].hash)
1256 				continue;
1257 			swap(*(q+1), *q);
1258 			more = 1;
1259 		}
1260 	} while(more);
1261 }
1262 
1263 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1264 {
1265 	struct dx_entry *entries = frame->entries;
1266 	struct dx_entry *old = frame->at, *new = old + 1;
1267 	int count = dx_get_count(entries);
1268 
1269 	assert(count < dx_get_limit(entries));
1270 	assert(old < entries + count);
1271 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1272 	dx_set_hash(new, hash);
1273 	dx_set_block(new, block);
1274 	dx_set_count(entries, count + 1);
1275 }
1276 
1277 #ifdef CONFIG_UNICODE
1278 /*
1279  * Test whether a case-insensitive directory entry matches the filename
1280  * being searched for.  If quick is set, assume the name being looked up
1281  * is already in the casefolded form.
1282  *
1283  * Returns: 0 if the directory entry matches, more than 0 if it
1284  * doesn't match or less than zero on error.
1285  */
1286 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1287 		    const struct qstr *entry, bool quick)
1288 {
1289 	const struct ext4_sb_info *sbi = EXT4_SB(parent->i_sb);
1290 	const struct unicode_map *um = sbi->s_encoding;
1291 	int ret;
1292 
1293 	if (quick)
1294 		ret = utf8_strncasecmp_folded(um, name, entry);
1295 	else
1296 		ret = utf8_strncasecmp(um, name, entry);
1297 
1298 	if (ret < 0) {
1299 		/* Handle invalid character sequence as either an error
1300 		 * or as an opaque byte sequence.
1301 		 */
1302 		if (ext4_has_strict_mode(sbi))
1303 			return -EINVAL;
1304 
1305 		if (name->len != entry->len)
1306 			return 1;
1307 
1308 		return !!memcmp(name->name, entry->name, name->len);
1309 	}
1310 
1311 	return ret;
1312 }
1313 
1314 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1315 				  struct fscrypt_str *cf_name)
1316 {
1317 	int len;
1318 
1319 	if (!IS_CASEFOLDED(dir) || !EXT4_SB(dir->i_sb)->s_encoding) {
1320 		cf_name->name = NULL;
1321 		return;
1322 	}
1323 
1324 	cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1325 	if (!cf_name->name)
1326 		return;
1327 
1328 	len = utf8_casefold(EXT4_SB(dir->i_sb)->s_encoding,
1329 			    iname, cf_name->name,
1330 			    EXT4_NAME_LEN);
1331 	if (len <= 0) {
1332 		kfree(cf_name->name);
1333 		cf_name->name = NULL;
1334 		return;
1335 	}
1336 	cf_name->len = (unsigned) len;
1337 
1338 }
1339 #endif
1340 
1341 /*
1342  * Test whether a directory entry matches the filename being searched for.
1343  *
1344  * Return: %true if the directory entry matches, otherwise %false.
1345  */
1346 static inline bool ext4_match(const struct inode *parent,
1347 			      const struct ext4_filename *fname,
1348 			      const struct ext4_dir_entry_2 *de)
1349 {
1350 	struct fscrypt_name f;
1351 #ifdef CONFIG_UNICODE
1352 	const struct qstr entry = {.name = de->name, .len = de->name_len};
1353 #endif
1354 
1355 	if (!de->inode)
1356 		return false;
1357 
1358 	f.usr_fname = fname->usr_fname;
1359 	f.disk_name = fname->disk_name;
1360 #ifdef CONFIG_FS_ENCRYPTION
1361 	f.crypto_buf = fname->crypto_buf;
1362 #endif
1363 
1364 #ifdef CONFIG_UNICODE
1365 	if (EXT4_SB(parent->i_sb)->s_encoding && IS_CASEFOLDED(parent)) {
1366 		if (fname->cf_name.name) {
1367 			struct qstr cf = {.name = fname->cf_name.name,
1368 					  .len = fname->cf_name.len};
1369 			return !ext4_ci_compare(parent, &cf, &entry, true);
1370 		}
1371 		return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1372 					false);
1373 	}
1374 #endif
1375 
1376 	return fscrypt_match_name(&f, de->name, de->name_len);
1377 }
1378 
1379 /*
1380  * Returns 0 if not found, -1 on failure, and 1 on success
1381  */
1382 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1383 		    struct inode *dir, struct ext4_filename *fname,
1384 		    unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1385 {
1386 	struct ext4_dir_entry_2 * de;
1387 	char * dlimit;
1388 	int de_len;
1389 
1390 	de = (struct ext4_dir_entry_2 *)search_buf;
1391 	dlimit = search_buf + buf_size;
1392 	while ((char *) de < dlimit) {
1393 		/* this code is executed quadratically often */
1394 		/* do minimal checking `by hand' */
1395 		if ((char *) de + de->name_len <= dlimit &&
1396 		    ext4_match(dir, fname, de)) {
1397 			/* found a match - just to be sure, do
1398 			 * a full check */
1399 			if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1400 						 bh->b_size, offset))
1401 				return -1;
1402 			*res_dir = de;
1403 			return 1;
1404 		}
1405 		/* prevent looping on a bad block */
1406 		de_len = ext4_rec_len_from_disk(de->rec_len,
1407 						dir->i_sb->s_blocksize);
1408 		if (de_len <= 0)
1409 			return -1;
1410 		offset += de_len;
1411 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1412 	}
1413 	return 0;
1414 }
1415 
1416 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1417 			       struct ext4_dir_entry *de)
1418 {
1419 	struct super_block *sb = dir->i_sb;
1420 
1421 	if (!is_dx(dir))
1422 		return 0;
1423 	if (block == 0)
1424 		return 1;
1425 	if (de->inode == 0 &&
1426 	    ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1427 			sb->s_blocksize)
1428 		return 1;
1429 	return 0;
1430 }
1431 
1432 /*
1433  *	__ext4_find_entry()
1434  *
1435  * finds an entry in the specified directory with the wanted name. It
1436  * returns the cache buffer in which the entry was found, and the entry
1437  * itself (as a parameter - res_dir). It does NOT read the inode of the
1438  * entry - you'll have to do that yourself if you want to.
1439  *
1440  * The returned buffer_head has ->b_count elevated.  The caller is expected
1441  * to brelse() it when appropriate.
1442  */
1443 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1444 					     struct ext4_filename *fname,
1445 					     struct ext4_dir_entry_2 **res_dir,
1446 					     int *inlined)
1447 {
1448 	struct super_block *sb;
1449 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1450 	struct buffer_head *bh, *ret = NULL;
1451 	ext4_lblk_t start, block;
1452 	const u8 *name = fname->usr_fname->name;
1453 	size_t ra_max = 0;	/* Number of bh's in the readahead
1454 				   buffer, bh_use[] */
1455 	size_t ra_ptr = 0;	/* Current index into readahead
1456 				   buffer */
1457 	ext4_lblk_t  nblocks;
1458 	int i, namelen, retval;
1459 
1460 	*res_dir = NULL;
1461 	sb = dir->i_sb;
1462 	namelen = fname->usr_fname->len;
1463 	if (namelen > EXT4_NAME_LEN)
1464 		return NULL;
1465 
1466 	if (ext4_has_inline_data(dir)) {
1467 		int has_inline_data = 1;
1468 		ret = ext4_find_inline_entry(dir, fname, res_dir,
1469 					     &has_inline_data);
1470 		if (has_inline_data) {
1471 			if (inlined)
1472 				*inlined = 1;
1473 			goto cleanup_and_exit;
1474 		}
1475 	}
1476 
1477 	if ((namelen <= 2) && (name[0] == '.') &&
1478 	    (name[1] == '.' || name[1] == '\0')) {
1479 		/*
1480 		 * "." or ".." will only be in the first block
1481 		 * NFS may look up ".."; "." should be handled by the VFS
1482 		 */
1483 		block = start = 0;
1484 		nblocks = 1;
1485 		goto restart;
1486 	}
1487 	if (is_dx(dir)) {
1488 		ret = ext4_dx_find_entry(dir, fname, res_dir);
1489 		/*
1490 		 * On success, or if the error was file not found,
1491 		 * return.  Otherwise, fall back to doing a search the
1492 		 * old fashioned way.
1493 		 */
1494 		if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1495 			goto cleanup_and_exit;
1496 		dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1497 			       "falling back\n"));
1498 		ret = NULL;
1499 	}
1500 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1501 	if (!nblocks) {
1502 		ret = NULL;
1503 		goto cleanup_and_exit;
1504 	}
1505 	start = EXT4_I(dir)->i_dir_start_lookup;
1506 	if (start >= nblocks)
1507 		start = 0;
1508 	block = start;
1509 restart:
1510 	do {
1511 		/*
1512 		 * We deal with the read-ahead logic here.
1513 		 */
1514 		if (ra_ptr >= ra_max) {
1515 			/* Refill the readahead buffer */
1516 			ra_ptr = 0;
1517 			if (block < start)
1518 				ra_max = start - block;
1519 			else
1520 				ra_max = nblocks - block;
1521 			ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1522 			retval = ext4_bread_batch(dir, block, ra_max,
1523 						  false /* wait */, bh_use);
1524 			if (retval) {
1525 				ret = ERR_PTR(retval);
1526 				ra_max = 0;
1527 				goto cleanup_and_exit;
1528 			}
1529 		}
1530 		if ((bh = bh_use[ra_ptr++]) == NULL)
1531 			goto next;
1532 		wait_on_buffer(bh);
1533 		if (!buffer_uptodate(bh)) {
1534 			ext4_set_errno(sb, EIO);
1535 			EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1536 					 (unsigned long) block);
1537 			brelse(bh);
1538 			ret = ERR_PTR(-EIO);
1539 			goto cleanup_and_exit;
1540 		}
1541 		if (!buffer_verified(bh) &&
1542 		    !is_dx_internal_node(dir, block,
1543 					 (struct ext4_dir_entry *)bh->b_data) &&
1544 		    !ext4_dirblock_csum_verify(dir, bh)) {
1545 			ext4_set_errno(sb, EFSBADCRC);
1546 			EXT4_ERROR_INODE(dir, "checksumming directory "
1547 					 "block %lu", (unsigned long)block);
1548 			brelse(bh);
1549 			ret = ERR_PTR(-EFSBADCRC);
1550 			goto cleanup_and_exit;
1551 		}
1552 		set_buffer_verified(bh);
1553 		i = search_dirblock(bh, dir, fname,
1554 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1555 		if (i == 1) {
1556 			EXT4_I(dir)->i_dir_start_lookup = block;
1557 			ret = bh;
1558 			goto cleanup_and_exit;
1559 		} else {
1560 			brelse(bh);
1561 			if (i < 0)
1562 				goto cleanup_and_exit;
1563 		}
1564 	next:
1565 		if (++block >= nblocks)
1566 			block = 0;
1567 	} while (block != start);
1568 
1569 	/*
1570 	 * If the directory has grown while we were searching, then
1571 	 * search the last part of the directory before giving up.
1572 	 */
1573 	block = nblocks;
1574 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1575 	if (block < nblocks) {
1576 		start = 0;
1577 		goto restart;
1578 	}
1579 
1580 cleanup_and_exit:
1581 	/* Clean up the read-ahead blocks */
1582 	for (; ra_ptr < ra_max; ra_ptr++)
1583 		brelse(bh_use[ra_ptr]);
1584 	return ret;
1585 }
1586 
1587 static struct buffer_head *ext4_find_entry(struct inode *dir,
1588 					   const struct qstr *d_name,
1589 					   struct ext4_dir_entry_2 **res_dir,
1590 					   int *inlined)
1591 {
1592 	int err;
1593 	struct ext4_filename fname;
1594 	struct buffer_head *bh;
1595 
1596 	err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1597 	if (err == -ENOENT)
1598 		return NULL;
1599 	if (err)
1600 		return ERR_PTR(err);
1601 
1602 	bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1603 
1604 	ext4_fname_free_filename(&fname);
1605 	return bh;
1606 }
1607 
1608 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1609 					     struct dentry *dentry,
1610 					     struct ext4_dir_entry_2 **res_dir)
1611 {
1612 	int err;
1613 	struct ext4_filename fname;
1614 	struct buffer_head *bh;
1615 
1616 	err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1617 	if (err == -ENOENT)
1618 		return NULL;
1619 	if (err)
1620 		return ERR_PTR(err);
1621 
1622 	bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1623 
1624 	ext4_fname_free_filename(&fname);
1625 	return bh;
1626 }
1627 
1628 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1629 			struct ext4_filename *fname,
1630 			struct ext4_dir_entry_2 **res_dir)
1631 {
1632 	struct super_block * sb = dir->i_sb;
1633 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1634 	struct buffer_head *bh;
1635 	ext4_lblk_t block;
1636 	int retval;
1637 
1638 #ifdef CONFIG_FS_ENCRYPTION
1639 	*res_dir = NULL;
1640 #endif
1641 	frame = dx_probe(fname, dir, NULL, frames);
1642 	if (IS_ERR(frame))
1643 		return (struct buffer_head *) frame;
1644 	do {
1645 		block = dx_get_block(frame->at);
1646 		bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1647 		if (IS_ERR(bh))
1648 			goto errout;
1649 
1650 		retval = search_dirblock(bh, dir, fname,
1651 					 block << EXT4_BLOCK_SIZE_BITS(sb),
1652 					 res_dir);
1653 		if (retval == 1)
1654 			goto success;
1655 		brelse(bh);
1656 		if (retval == -1) {
1657 			bh = ERR_PTR(ERR_BAD_DX_DIR);
1658 			goto errout;
1659 		}
1660 
1661 		/* Check to see if we should continue to search */
1662 		retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1663 					       frames, NULL);
1664 		if (retval < 0) {
1665 			ext4_warning_inode(dir,
1666 				"error %d reading directory index block",
1667 				retval);
1668 			bh = ERR_PTR(retval);
1669 			goto errout;
1670 		}
1671 	} while (retval == 1);
1672 
1673 	bh = NULL;
1674 errout:
1675 	dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1676 success:
1677 	dx_release(frames);
1678 	return bh;
1679 }
1680 
1681 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1682 {
1683 	struct inode *inode;
1684 	struct ext4_dir_entry_2 *de;
1685 	struct buffer_head *bh;
1686 
1687 	if (dentry->d_name.len > EXT4_NAME_LEN)
1688 		return ERR_PTR(-ENAMETOOLONG);
1689 
1690 	bh = ext4_lookup_entry(dir, dentry, &de);
1691 	if (IS_ERR(bh))
1692 		return ERR_CAST(bh);
1693 	inode = NULL;
1694 	if (bh) {
1695 		__u32 ino = le32_to_cpu(de->inode);
1696 		brelse(bh);
1697 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1698 			EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1699 			return ERR_PTR(-EFSCORRUPTED);
1700 		}
1701 		if (unlikely(ino == dir->i_ino)) {
1702 			EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1703 					 dentry);
1704 			return ERR_PTR(-EFSCORRUPTED);
1705 		}
1706 		inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1707 		if (inode == ERR_PTR(-ESTALE)) {
1708 			EXT4_ERROR_INODE(dir,
1709 					 "deleted inode referenced: %u",
1710 					 ino);
1711 			return ERR_PTR(-EFSCORRUPTED);
1712 		}
1713 		if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1714 		    (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1715 		    !fscrypt_has_permitted_context(dir, inode)) {
1716 			ext4_warning(inode->i_sb,
1717 				     "Inconsistent encryption contexts: %lu/%lu",
1718 				     dir->i_ino, inode->i_ino);
1719 			iput(inode);
1720 			return ERR_PTR(-EPERM);
1721 		}
1722 	}
1723 
1724 #ifdef CONFIG_UNICODE
1725 	if (!inode && IS_CASEFOLDED(dir)) {
1726 		/* Eventually we want to call d_add_ci(dentry, NULL)
1727 		 * for negative dentries in the encoding case as
1728 		 * well.  For now, prevent the negative dentry
1729 		 * from being cached.
1730 		 */
1731 		return NULL;
1732 	}
1733 #endif
1734 	return d_splice_alias(inode, dentry);
1735 }
1736 
1737 
1738 struct dentry *ext4_get_parent(struct dentry *child)
1739 {
1740 	__u32 ino;
1741 	static const struct qstr dotdot = QSTR_INIT("..", 2);
1742 	struct ext4_dir_entry_2 * de;
1743 	struct buffer_head *bh;
1744 
1745 	bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1746 	if (IS_ERR(bh))
1747 		return ERR_CAST(bh);
1748 	if (!bh)
1749 		return ERR_PTR(-ENOENT);
1750 	ino = le32_to_cpu(de->inode);
1751 	brelse(bh);
1752 
1753 	if (!ext4_valid_inum(child->d_sb, ino)) {
1754 		EXT4_ERROR_INODE(d_inode(child),
1755 				 "bad parent inode number: %u", ino);
1756 		return ERR_PTR(-EFSCORRUPTED);
1757 	}
1758 
1759 	return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1760 }
1761 
1762 /*
1763  * Move count entries from end of map between two memory locations.
1764  * Returns pointer to last entry moved.
1765  */
1766 static struct ext4_dir_entry_2 *
1767 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1768 		unsigned blocksize)
1769 {
1770 	unsigned rec_len = 0;
1771 
1772 	while (count--) {
1773 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1774 						(from + (map->offs<<2));
1775 		rec_len = EXT4_DIR_REC_LEN(de->name_len);
1776 		memcpy (to, de, rec_len);
1777 		((struct ext4_dir_entry_2 *) to)->rec_len =
1778 				ext4_rec_len_to_disk(rec_len, blocksize);
1779 		de->inode = 0;
1780 		map++;
1781 		to += rec_len;
1782 	}
1783 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1784 }
1785 
1786 /*
1787  * Compact each dir entry in the range to the minimal rec_len.
1788  * Returns pointer to last entry in range.
1789  */
1790 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1791 {
1792 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1793 	unsigned rec_len = 0;
1794 
1795 	prev = to = de;
1796 	while ((char*)de < base + blocksize) {
1797 		next = ext4_next_entry(de, blocksize);
1798 		if (de->inode && de->name_len) {
1799 			rec_len = EXT4_DIR_REC_LEN(de->name_len);
1800 			if (de > to)
1801 				memmove(to, de, rec_len);
1802 			to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1803 			prev = to;
1804 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1805 		}
1806 		de = next;
1807 	}
1808 	return prev;
1809 }
1810 
1811 /*
1812  * Split a full leaf block to make room for a new dir entry.
1813  * Allocate a new block, and move entries so that they are approx. equally full.
1814  * Returns pointer to de in block into which the new entry will be inserted.
1815  */
1816 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1817 			struct buffer_head **bh,struct dx_frame *frame,
1818 			struct dx_hash_info *hinfo)
1819 {
1820 	unsigned blocksize = dir->i_sb->s_blocksize;
1821 	unsigned count, continued;
1822 	struct buffer_head *bh2;
1823 	ext4_lblk_t newblock;
1824 	u32 hash2;
1825 	struct dx_map_entry *map;
1826 	char *data1 = (*bh)->b_data, *data2;
1827 	unsigned split, move, size;
1828 	struct ext4_dir_entry_2 *de = NULL, *de2;
1829 	int	csum_size = 0;
1830 	int	err = 0, i;
1831 
1832 	if (ext4_has_metadata_csum(dir->i_sb))
1833 		csum_size = sizeof(struct ext4_dir_entry_tail);
1834 
1835 	bh2 = ext4_append(handle, dir, &newblock);
1836 	if (IS_ERR(bh2)) {
1837 		brelse(*bh);
1838 		*bh = NULL;
1839 		return (struct ext4_dir_entry_2 *) bh2;
1840 	}
1841 
1842 	BUFFER_TRACE(*bh, "get_write_access");
1843 	err = ext4_journal_get_write_access(handle, *bh);
1844 	if (err)
1845 		goto journal_error;
1846 
1847 	BUFFER_TRACE(frame->bh, "get_write_access");
1848 	err = ext4_journal_get_write_access(handle, frame->bh);
1849 	if (err)
1850 		goto journal_error;
1851 
1852 	data2 = bh2->b_data;
1853 
1854 	/* create map in the end of data2 block */
1855 	map = (struct dx_map_entry *) (data2 + blocksize);
1856 	count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1857 			     blocksize, hinfo, map);
1858 	map -= count;
1859 	dx_sort_map(map, count);
1860 	/* Split the existing block in the middle, size-wise */
1861 	size = 0;
1862 	move = 0;
1863 	for (i = count-1; i >= 0; i--) {
1864 		/* is more than half of this entry in 2nd half of the block? */
1865 		if (size + map[i].size/2 > blocksize/2)
1866 			break;
1867 		size += map[i].size;
1868 		move++;
1869 	}
1870 	/* map index at which we will split */
1871 	split = count - move;
1872 	hash2 = map[split].hash;
1873 	continued = hash2 == map[split - 1].hash;
1874 	dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1875 			(unsigned long)dx_get_block(frame->at),
1876 					hash2, split, count-split));
1877 
1878 	/* Fancy dance to stay within two buffers */
1879 	de2 = dx_move_dirents(data1, data2, map + split, count - split,
1880 			      blocksize);
1881 	de = dx_pack_dirents(data1, blocksize);
1882 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1883 					   (char *) de,
1884 					   blocksize);
1885 	de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1886 					    (char *) de2,
1887 					    blocksize);
1888 	if (csum_size) {
1889 		ext4_initialize_dirent_tail(*bh, blocksize);
1890 		ext4_initialize_dirent_tail(bh2, blocksize);
1891 	}
1892 
1893 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1894 			blocksize, 1));
1895 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1896 			blocksize, 1));
1897 
1898 	/* Which block gets the new entry? */
1899 	if (hinfo->hash >= hash2) {
1900 		swap(*bh, bh2);
1901 		de = de2;
1902 	}
1903 	dx_insert_block(frame, hash2 + continued, newblock);
1904 	err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1905 	if (err)
1906 		goto journal_error;
1907 	err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1908 	if (err)
1909 		goto journal_error;
1910 	brelse(bh2);
1911 	dxtrace(dx_show_index("frame", frame->entries));
1912 	return de;
1913 
1914 journal_error:
1915 	brelse(*bh);
1916 	brelse(bh2);
1917 	*bh = NULL;
1918 	ext4_std_error(dir->i_sb, err);
1919 	return ERR_PTR(err);
1920 }
1921 
1922 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1923 		      struct buffer_head *bh,
1924 		      void *buf, int buf_size,
1925 		      struct ext4_filename *fname,
1926 		      struct ext4_dir_entry_2 **dest_de)
1927 {
1928 	struct ext4_dir_entry_2 *de;
1929 	unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1930 	int nlen, rlen;
1931 	unsigned int offset = 0;
1932 	char *top;
1933 
1934 	de = (struct ext4_dir_entry_2 *)buf;
1935 	top = buf + buf_size - reclen;
1936 	while ((char *) de <= top) {
1937 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1938 					 buf, buf_size, offset))
1939 			return -EFSCORRUPTED;
1940 		if (ext4_match(dir, fname, de))
1941 			return -EEXIST;
1942 		nlen = EXT4_DIR_REC_LEN(de->name_len);
1943 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1944 		if ((de->inode ? rlen - nlen : rlen) >= reclen)
1945 			break;
1946 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1947 		offset += rlen;
1948 	}
1949 	if ((char *) de > top)
1950 		return -ENOSPC;
1951 
1952 	*dest_de = de;
1953 	return 0;
1954 }
1955 
1956 void ext4_insert_dentry(struct inode *inode,
1957 			struct ext4_dir_entry_2 *de,
1958 			int buf_size,
1959 			struct ext4_filename *fname)
1960 {
1961 
1962 	int nlen, rlen;
1963 
1964 	nlen = EXT4_DIR_REC_LEN(de->name_len);
1965 	rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1966 	if (de->inode) {
1967 		struct ext4_dir_entry_2 *de1 =
1968 			(struct ext4_dir_entry_2 *)((char *)de + nlen);
1969 		de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1970 		de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1971 		de = de1;
1972 	}
1973 	de->file_type = EXT4_FT_UNKNOWN;
1974 	de->inode = cpu_to_le32(inode->i_ino);
1975 	ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1976 	de->name_len = fname_len(fname);
1977 	memcpy(de->name, fname_name(fname), fname_len(fname));
1978 }
1979 
1980 /*
1981  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1982  * it points to a directory entry which is guaranteed to be large
1983  * enough for new directory entry.  If de is NULL, then
1984  * add_dirent_to_buf will attempt search the directory block for
1985  * space.  It will return -ENOSPC if no space is available, and -EIO
1986  * and -EEXIST if directory entry already exists.
1987  */
1988 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1989 			     struct inode *dir,
1990 			     struct inode *inode, struct ext4_dir_entry_2 *de,
1991 			     struct buffer_head *bh)
1992 {
1993 	unsigned int	blocksize = dir->i_sb->s_blocksize;
1994 	int		csum_size = 0;
1995 	int		err;
1996 
1997 	if (ext4_has_metadata_csum(inode->i_sb))
1998 		csum_size = sizeof(struct ext4_dir_entry_tail);
1999 
2000 	if (!de) {
2001 		err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2002 					blocksize - csum_size, fname, &de);
2003 		if (err)
2004 			return err;
2005 	}
2006 	BUFFER_TRACE(bh, "get_write_access");
2007 	err = ext4_journal_get_write_access(handle, bh);
2008 	if (err) {
2009 		ext4_std_error(dir->i_sb, err);
2010 		return err;
2011 	}
2012 
2013 	/* By now the buffer is marked for journaling */
2014 	ext4_insert_dentry(inode, de, blocksize, fname);
2015 
2016 	/*
2017 	 * XXX shouldn't update any times until successful
2018 	 * completion of syscall, but too many callers depend
2019 	 * on this.
2020 	 *
2021 	 * XXX similarly, too many callers depend on
2022 	 * ext4_new_inode() setting the times, but error
2023 	 * recovery deletes the inode, so the worst that can
2024 	 * happen is that the times are slightly out of date
2025 	 * and/or different from the directory change time.
2026 	 */
2027 	dir->i_mtime = dir->i_ctime = current_time(dir);
2028 	ext4_update_dx_flag(dir);
2029 	inode_inc_iversion(dir);
2030 	ext4_mark_inode_dirty(handle, dir);
2031 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2032 	err = ext4_handle_dirty_dirblock(handle, dir, bh);
2033 	if (err)
2034 		ext4_std_error(dir->i_sb, err);
2035 	return 0;
2036 }
2037 
2038 /*
2039  * This converts a one block unindexed directory to a 3 block indexed
2040  * directory, and adds the dentry to the indexed directory.
2041  */
2042 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2043 			    struct inode *dir,
2044 			    struct inode *inode, struct buffer_head *bh)
2045 {
2046 	struct buffer_head *bh2;
2047 	struct dx_root	*root;
2048 	struct dx_frame	frames[EXT4_HTREE_LEVEL], *frame;
2049 	struct dx_entry *entries;
2050 	struct ext4_dir_entry_2	*de, *de2;
2051 	char		*data2, *top;
2052 	unsigned	len;
2053 	int		retval;
2054 	unsigned	blocksize;
2055 	ext4_lblk_t  block;
2056 	struct fake_dirent *fde;
2057 	int csum_size = 0;
2058 
2059 	if (ext4_has_metadata_csum(inode->i_sb))
2060 		csum_size = sizeof(struct ext4_dir_entry_tail);
2061 
2062 	blocksize =  dir->i_sb->s_blocksize;
2063 	dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2064 	BUFFER_TRACE(bh, "get_write_access");
2065 	retval = ext4_journal_get_write_access(handle, bh);
2066 	if (retval) {
2067 		ext4_std_error(dir->i_sb, retval);
2068 		brelse(bh);
2069 		return retval;
2070 	}
2071 	root = (struct dx_root *) bh->b_data;
2072 
2073 	/* The 0th block becomes the root, move the dirents out */
2074 	fde = &root->dotdot;
2075 	de = (struct ext4_dir_entry_2 *)((char *)fde +
2076 		ext4_rec_len_from_disk(fde->rec_len, blocksize));
2077 	if ((char *) de >= (((char *) root) + blocksize)) {
2078 		EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2079 		brelse(bh);
2080 		return -EFSCORRUPTED;
2081 	}
2082 	len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2083 
2084 	/* Allocate new block for the 0th block's dirents */
2085 	bh2 = ext4_append(handle, dir, &block);
2086 	if (IS_ERR(bh2)) {
2087 		brelse(bh);
2088 		return PTR_ERR(bh2);
2089 	}
2090 	ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2091 	data2 = bh2->b_data;
2092 
2093 	memcpy(data2, de, len);
2094 	de = (struct ext4_dir_entry_2 *) data2;
2095 	top = data2 + len;
2096 	while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2097 		de = de2;
2098 	de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2099 					   (char *) de, blocksize);
2100 
2101 	if (csum_size)
2102 		ext4_initialize_dirent_tail(bh2, blocksize);
2103 
2104 	/* Initialize the root; the dot dirents already exist */
2105 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2106 	de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2107 					   blocksize);
2108 	memset (&root->info, 0, sizeof(root->info));
2109 	root->info.info_length = sizeof(root->info);
2110 	root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2111 	entries = root->entries;
2112 	dx_set_block(entries, 1);
2113 	dx_set_count(entries, 1);
2114 	dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2115 
2116 	/* Initialize as for dx_probe */
2117 	fname->hinfo.hash_version = root->info.hash_version;
2118 	if (fname->hinfo.hash_version <= DX_HASH_TEA)
2119 		fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2120 	fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2121 	ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2122 
2123 	memset(frames, 0, sizeof(frames));
2124 	frame = frames;
2125 	frame->entries = entries;
2126 	frame->at = entries;
2127 	frame->bh = bh;
2128 
2129 	retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2130 	if (retval)
2131 		goto out_frames;
2132 	retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2133 	if (retval)
2134 		goto out_frames;
2135 
2136 	de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2137 	if (IS_ERR(de)) {
2138 		retval = PTR_ERR(de);
2139 		goto out_frames;
2140 	}
2141 
2142 	retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2143 out_frames:
2144 	/*
2145 	 * Even if the block split failed, we have to properly write
2146 	 * out all the changes we did so far. Otherwise we can end up
2147 	 * with corrupted filesystem.
2148 	 */
2149 	if (retval)
2150 		ext4_mark_inode_dirty(handle, dir);
2151 	dx_release(frames);
2152 	brelse(bh2);
2153 	return retval;
2154 }
2155 
2156 /*
2157  *	ext4_add_entry()
2158  *
2159  * adds a file entry to the specified directory, using the same
2160  * semantics as ext4_find_entry(). It returns NULL if it failed.
2161  *
2162  * NOTE!! The inode part of 'de' is left at 0 - which means you
2163  * may not sleep between calling this and putting something into
2164  * the entry, as someone else might have used it while you slept.
2165  */
2166 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2167 			  struct inode *inode)
2168 {
2169 	struct inode *dir = d_inode(dentry->d_parent);
2170 	struct buffer_head *bh = NULL;
2171 	struct ext4_dir_entry_2 *de;
2172 	struct super_block *sb;
2173 #ifdef CONFIG_UNICODE
2174 	struct ext4_sb_info *sbi;
2175 #endif
2176 	struct ext4_filename fname;
2177 	int	retval;
2178 	int	dx_fallback=0;
2179 	unsigned blocksize;
2180 	ext4_lblk_t block, blocks;
2181 	int	csum_size = 0;
2182 
2183 	if (ext4_has_metadata_csum(inode->i_sb))
2184 		csum_size = sizeof(struct ext4_dir_entry_tail);
2185 
2186 	sb = dir->i_sb;
2187 	blocksize = sb->s_blocksize;
2188 	if (!dentry->d_name.len)
2189 		return -EINVAL;
2190 
2191 #ifdef CONFIG_UNICODE
2192 	sbi = EXT4_SB(sb);
2193 	if (ext4_has_strict_mode(sbi) && IS_CASEFOLDED(dir) &&
2194 	    sbi->s_encoding && utf8_validate(sbi->s_encoding, &dentry->d_name))
2195 		return -EINVAL;
2196 #endif
2197 
2198 	retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2199 	if (retval)
2200 		return retval;
2201 
2202 	if (ext4_has_inline_data(dir)) {
2203 		retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2204 		if (retval < 0)
2205 			goto out;
2206 		if (retval == 1) {
2207 			retval = 0;
2208 			goto out;
2209 		}
2210 	}
2211 
2212 	if (is_dx(dir)) {
2213 		retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2214 		if (!retval || (retval != ERR_BAD_DX_DIR))
2215 			goto out;
2216 		ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2217 		dx_fallback++;
2218 		ext4_mark_inode_dirty(handle, dir);
2219 	}
2220 	blocks = dir->i_size >> sb->s_blocksize_bits;
2221 	for (block = 0; block < blocks; block++) {
2222 		bh = ext4_read_dirblock(dir, block, DIRENT);
2223 		if (bh == NULL) {
2224 			bh = ext4_bread(handle, dir, block,
2225 					EXT4_GET_BLOCKS_CREATE);
2226 			goto add_to_new_block;
2227 		}
2228 		if (IS_ERR(bh)) {
2229 			retval = PTR_ERR(bh);
2230 			bh = NULL;
2231 			goto out;
2232 		}
2233 		retval = add_dirent_to_buf(handle, &fname, dir, inode,
2234 					   NULL, bh);
2235 		if (retval != -ENOSPC)
2236 			goto out;
2237 
2238 		if (blocks == 1 && !dx_fallback &&
2239 		    ext4_has_feature_dir_index(sb)) {
2240 			retval = make_indexed_dir(handle, &fname, dir,
2241 						  inode, bh);
2242 			bh = NULL; /* make_indexed_dir releases bh */
2243 			goto out;
2244 		}
2245 		brelse(bh);
2246 	}
2247 	bh = ext4_append(handle, dir, &block);
2248 add_to_new_block:
2249 	if (IS_ERR(bh)) {
2250 		retval = PTR_ERR(bh);
2251 		bh = NULL;
2252 		goto out;
2253 	}
2254 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2255 	de->inode = 0;
2256 	de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2257 
2258 	if (csum_size)
2259 		ext4_initialize_dirent_tail(bh, blocksize);
2260 
2261 	retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2262 out:
2263 	ext4_fname_free_filename(&fname);
2264 	brelse(bh);
2265 	if (retval == 0)
2266 		ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2267 	return retval;
2268 }
2269 
2270 /*
2271  * Returns 0 for success, or a negative error value
2272  */
2273 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2274 			     struct inode *dir, struct inode *inode)
2275 {
2276 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2277 	struct dx_entry *entries, *at;
2278 	struct buffer_head *bh;
2279 	struct super_block *sb = dir->i_sb;
2280 	struct ext4_dir_entry_2 *de;
2281 	int restart;
2282 	int err;
2283 
2284 again:
2285 	restart = 0;
2286 	frame = dx_probe(fname, dir, NULL, frames);
2287 	if (IS_ERR(frame))
2288 		return PTR_ERR(frame);
2289 	entries = frame->entries;
2290 	at = frame->at;
2291 	bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2292 	if (IS_ERR(bh)) {
2293 		err = PTR_ERR(bh);
2294 		bh = NULL;
2295 		goto cleanup;
2296 	}
2297 
2298 	BUFFER_TRACE(bh, "get_write_access");
2299 	err = ext4_journal_get_write_access(handle, bh);
2300 	if (err)
2301 		goto journal_error;
2302 
2303 	err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2304 	if (err != -ENOSPC)
2305 		goto cleanup;
2306 
2307 	err = 0;
2308 	/* Block full, should compress but for now just split */
2309 	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2310 		       dx_get_count(entries), dx_get_limit(entries)));
2311 	/* Need to split index? */
2312 	if (dx_get_count(entries) == dx_get_limit(entries)) {
2313 		ext4_lblk_t newblock;
2314 		int levels = frame - frames + 1;
2315 		unsigned int icount;
2316 		int add_level = 1;
2317 		struct dx_entry *entries2;
2318 		struct dx_node *node2;
2319 		struct buffer_head *bh2;
2320 
2321 		while (frame > frames) {
2322 			if (dx_get_count((frame - 1)->entries) <
2323 			    dx_get_limit((frame - 1)->entries)) {
2324 				add_level = 0;
2325 				break;
2326 			}
2327 			frame--; /* split higher index block */
2328 			at = frame->at;
2329 			entries = frame->entries;
2330 			restart = 1;
2331 		}
2332 		if (add_level && levels == ext4_dir_htree_level(sb)) {
2333 			ext4_warning(sb, "Directory (ino: %lu) index full, "
2334 					 "reach max htree level :%d",
2335 					 dir->i_ino, levels);
2336 			if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2337 				ext4_warning(sb, "Large directory feature is "
2338 						 "not enabled on this "
2339 						 "filesystem");
2340 			}
2341 			err = -ENOSPC;
2342 			goto cleanup;
2343 		}
2344 		icount = dx_get_count(entries);
2345 		bh2 = ext4_append(handle, dir, &newblock);
2346 		if (IS_ERR(bh2)) {
2347 			err = PTR_ERR(bh2);
2348 			goto cleanup;
2349 		}
2350 		node2 = (struct dx_node *)(bh2->b_data);
2351 		entries2 = node2->entries;
2352 		memset(&node2->fake, 0, sizeof(struct fake_dirent));
2353 		node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2354 							   sb->s_blocksize);
2355 		BUFFER_TRACE(frame->bh, "get_write_access");
2356 		err = ext4_journal_get_write_access(handle, frame->bh);
2357 		if (err)
2358 			goto journal_error;
2359 		if (!add_level) {
2360 			unsigned icount1 = icount/2, icount2 = icount - icount1;
2361 			unsigned hash2 = dx_get_hash(entries + icount1);
2362 			dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2363 				       icount1, icount2));
2364 
2365 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2366 			err = ext4_journal_get_write_access(handle,
2367 							     (frame - 1)->bh);
2368 			if (err)
2369 				goto journal_error;
2370 
2371 			memcpy((char *) entries2, (char *) (entries + icount1),
2372 			       icount2 * sizeof(struct dx_entry));
2373 			dx_set_count(entries, icount1);
2374 			dx_set_count(entries2, icount2);
2375 			dx_set_limit(entries2, dx_node_limit(dir));
2376 
2377 			/* Which index block gets the new entry? */
2378 			if (at - entries >= icount1) {
2379 				frame->at = at = at - entries - icount1 + entries2;
2380 				frame->entries = entries = entries2;
2381 				swap(frame->bh, bh2);
2382 			}
2383 			dx_insert_block((frame - 1), hash2, newblock);
2384 			dxtrace(dx_show_index("node", frame->entries));
2385 			dxtrace(dx_show_index("node",
2386 			       ((struct dx_node *) bh2->b_data)->entries));
2387 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2388 			if (err)
2389 				goto journal_error;
2390 			brelse (bh2);
2391 			err = ext4_handle_dirty_dx_node(handle, dir,
2392 						   (frame - 1)->bh);
2393 			if (err)
2394 				goto journal_error;
2395 			if (restart) {
2396 				err = ext4_handle_dirty_dx_node(handle, dir,
2397 							   frame->bh);
2398 				goto journal_error;
2399 			}
2400 		} else {
2401 			struct dx_root *dxroot;
2402 			memcpy((char *) entries2, (char *) entries,
2403 			       icount * sizeof(struct dx_entry));
2404 			dx_set_limit(entries2, dx_node_limit(dir));
2405 
2406 			/* Set up root */
2407 			dx_set_count(entries, 1);
2408 			dx_set_block(entries + 0, newblock);
2409 			dxroot = (struct dx_root *)frames[0].bh->b_data;
2410 			dxroot->info.indirect_levels += 1;
2411 			dxtrace(printk(KERN_DEBUG
2412 				       "Creating %d level index...\n",
2413 				       dxroot->info.indirect_levels));
2414 			err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2415 			if (err)
2416 				goto journal_error;
2417 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2418 			brelse(bh2);
2419 			restart = 1;
2420 			goto journal_error;
2421 		}
2422 	}
2423 	de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2424 	if (IS_ERR(de)) {
2425 		err = PTR_ERR(de);
2426 		goto cleanup;
2427 	}
2428 	err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2429 	goto cleanup;
2430 
2431 journal_error:
2432 	ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2433 cleanup:
2434 	brelse(bh);
2435 	dx_release(frames);
2436 	/* @restart is true means htree-path has been changed, we need to
2437 	 * repeat dx_probe() to find out valid htree-path
2438 	 */
2439 	if (restart && err == 0)
2440 		goto again;
2441 	return err;
2442 }
2443 
2444 /*
2445  * ext4_generic_delete_entry deletes a directory entry by merging it
2446  * with the previous entry
2447  */
2448 int ext4_generic_delete_entry(handle_t *handle,
2449 			      struct inode *dir,
2450 			      struct ext4_dir_entry_2 *de_del,
2451 			      struct buffer_head *bh,
2452 			      void *entry_buf,
2453 			      int buf_size,
2454 			      int csum_size)
2455 {
2456 	struct ext4_dir_entry_2 *de, *pde;
2457 	unsigned int blocksize = dir->i_sb->s_blocksize;
2458 	int i;
2459 
2460 	i = 0;
2461 	pde = NULL;
2462 	de = (struct ext4_dir_entry_2 *)entry_buf;
2463 	while (i < buf_size - csum_size) {
2464 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2465 					 bh->b_data, bh->b_size, i))
2466 			return -EFSCORRUPTED;
2467 		if (de == de_del)  {
2468 			if (pde)
2469 				pde->rec_len = ext4_rec_len_to_disk(
2470 					ext4_rec_len_from_disk(pde->rec_len,
2471 							       blocksize) +
2472 					ext4_rec_len_from_disk(de->rec_len,
2473 							       blocksize),
2474 					blocksize);
2475 			else
2476 				de->inode = 0;
2477 			inode_inc_iversion(dir);
2478 			return 0;
2479 		}
2480 		i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2481 		pde = de;
2482 		de = ext4_next_entry(de, blocksize);
2483 	}
2484 	return -ENOENT;
2485 }
2486 
2487 static int ext4_delete_entry(handle_t *handle,
2488 			     struct inode *dir,
2489 			     struct ext4_dir_entry_2 *de_del,
2490 			     struct buffer_head *bh)
2491 {
2492 	int err, csum_size = 0;
2493 
2494 	if (ext4_has_inline_data(dir)) {
2495 		int has_inline_data = 1;
2496 		err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2497 					       &has_inline_data);
2498 		if (has_inline_data)
2499 			return err;
2500 	}
2501 
2502 	if (ext4_has_metadata_csum(dir->i_sb))
2503 		csum_size = sizeof(struct ext4_dir_entry_tail);
2504 
2505 	BUFFER_TRACE(bh, "get_write_access");
2506 	err = ext4_journal_get_write_access(handle, bh);
2507 	if (unlikely(err))
2508 		goto out;
2509 
2510 	err = ext4_generic_delete_entry(handle, dir, de_del,
2511 					bh, bh->b_data,
2512 					dir->i_sb->s_blocksize, csum_size);
2513 	if (err)
2514 		goto out;
2515 
2516 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2517 	err = ext4_handle_dirty_dirblock(handle, dir, bh);
2518 	if (unlikely(err))
2519 		goto out;
2520 
2521 	return 0;
2522 out:
2523 	if (err != -ENOENT)
2524 		ext4_std_error(dir->i_sb, err);
2525 	return err;
2526 }
2527 
2528 /*
2529  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2530  * since this indicates that nlinks count was previously 1 to avoid overflowing
2531  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2532  * that subdirectory link counts are not being maintained accurately.
2533  *
2534  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2535  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2536  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2537  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2538  */
2539 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2540 {
2541 	inc_nlink(inode);
2542 	if (is_dx(inode) &&
2543 	    (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2544 		set_nlink(inode, 1);
2545 }
2546 
2547 /*
2548  * If a directory had nlink == 1, then we should let it be 1. This indicates
2549  * directory has >EXT4_LINK_MAX subdirs.
2550  */
2551 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2552 {
2553 	if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2554 		drop_nlink(inode);
2555 }
2556 
2557 
2558 /*
2559  * Add non-directory inode to a directory. On success, the inode reference is
2560  * consumed by dentry is instantiation. This is also indicated by clearing of
2561  * *inodep pointer. On failure, the caller is responsible for dropping the
2562  * inode reference in the safe context.
2563  */
2564 static int ext4_add_nondir(handle_t *handle,
2565 		struct dentry *dentry, struct inode **inodep)
2566 {
2567 	struct inode *dir = d_inode(dentry->d_parent);
2568 	struct inode *inode = *inodep;
2569 	int err = ext4_add_entry(handle, dentry, inode);
2570 	if (!err) {
2571 		ext4_mark_inode_dirty(handle, inode);
2572 		if (IS_DIRSYNC(dir))
2573 			ext4_handle_sync(handle);
2574 		d_instantiate_new(dentry, inode);
2575 		*inodep = NULL;
2576 		return 0;
2577 	}
2578 	drop_nlink(inode);
2579 	ext4_orphan_add(handle, inode);
2580 	unlock_new_inode(inode);
2581 	return err;
2582 }
2583 
2584 /*
2585  * By the time this is called, we already have created
2586  * the directory cache entry for the new file, but it
2587  * is so far negative - it has no inode.
2588  *
2589  * If the create succeeds, we fill in the inode information
2590  * with d_instantiate().
2591  */
2592 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2593 		       bool excl)
2594 {
2595 	handle_t *handle;
2596 	struct inode *inode;
2597 	int err, credits, retries = 0;
2598 
2599 	err = dquot_initialize(dir);
2600 	if (err)
2601 		return err;
2602 
2603 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2604 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2605 retry:
2606 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2607 					    NULL, EXT4_HT_DIR, credits);
2608 	handle = ext4_journal_current_handle();
2609 	err = PTR_ERR(inode);
2610 	if (!IS_ERR(inode)) {
2611 		inode->i_op = &ext4_file_inode_operations;
2612 		inode->i_fop = &ext4_file_operations;
2613 		ext4_set_aops(inode);
2614 		err = ext4_add_nondir(handle, dentry, &inode);
2615 	}
2616 	if (handle)
2617 		ext4_journal_stop(handle);
2618 	if (!IS_ERR_OR_NULL(inode))
2619 		iput(inode);
2620 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2621 		goto retry;
2622 	return err;
2623 }
2624 
2625 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2626 		      umode_t mode, dev_t rdev)
2627 {
2628 	handle_t *handle;
2629 	struct inode *inode;
2630 	int err, credits, retries = 0;
2631 
2632 	err = dquot_initialize(dir);
2633 	if (err)
2634 		return err;
2635 
2636 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2637 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2638 retry:
2639 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2640 					    NULL, EXT4_HT_DIR, credits);
2641 	handle = ext4_journal_current_handle();
2642 	err = PTR_ERR(inode);
2643 	if (!IS_ERR(inode)) {
2644 		init_special_inode(inode, inode->i_mode, rdev);
2645 		inode->i_op = &ext4_special_inode_operations;
2646 		err = ext4_add_nondir(handle, dentry, &inode);
2647 	}
2648 	if (handle)
2649 		ext4_journal_stop(handle);
2650 	if (!IS_ERR_OR_NULL(inode))
2651 		iput(inode);
2652 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2653 		goto retry;
2654 	return err;
2655 }
2656 
2657 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2658 {
2659 	handle_t *handle;
2660 	struct inode *inode;
2661 	int err, retries = 0;
2662 
2663 	err = dquot_initialize(dir);
2664 	if (err)
2665 		return err;
2666 
2667 retry:
2668 	inode = ext4_new_inode_start_handle(dir, mode,
2669 					    NULL, 0, NULL,
2670 					    EXT4_HT_DIR,
2671 			EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2672 			  4 + EXT4_XATTR_TRANS_BLOCKS);
2673 	handle = ext4_journal_current_handle();
2674 	err = PTR_ERR(inode);
2675 	if (!IS_ERR(inode)) {
2676 		inode->i_op = &ext4_file_inode_operations;
2677 		inode->i_fop = &ext4_file_operations;
2678 		ext4_set_aops(inode);
2679 		d_tmpfile(dentry, inode);
2680 		err = ext4_orphan_add(handle, inode);
2681 		if (err)
2682 			goto err_unlock_inode;
2683 		mark_inode_dirty(inode);
2684 		unlock_new_inode(inode);
2685 	}
2686 	if (handle)
2687 		ext4_journal_stop(handle);
2688 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2689 		goto retry;
2690 	return err;
2691 err_unlock_inode:
2692 	ext4_journal_stop(handle);
2693 	unlock_new_inode(inode);
2694 	return err;
2695 }
2696 
2697 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2698 			  struct ext4_dir_entry_2 *de,
2699 			  int blocksize, int csum_size,
2700 			  unsigned int parent_ino, int dotdot_real_len)
2701 {
2702 	de->inode = cpu_to_le32(inode->i_ino);
2703 	de->name_len = 1;
2704 	de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2705 					   blocksize);
2706 	strcpy(de->name, ".");
2707 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2708 
2709 	de = ext4_next_entry(de, blocksize);
2710 	de->inode = cpu_to_le32(parent_ino);
2711 	de->name_len = 2;
2712 	if (!dotdot_real_len)
2713 		de->rec_len = ext4_rec_len_to_disk(blocksize -
2714 					(csum_size + EXT4_DIR_REC_LEN(1)),
2715 					blocksize);
2716 	else
2717 		de->rec_len = ext4_rec_len_to_disk(
2718 				EXT4_DIR_REC_LEN(de->name_len), blocksize);
2719 	strcpy(de->name, "..");
2720 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2721 
2722 	return ext4_next_entry(de, blocksize);
2723 }
2724 
2725 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2726 			     struct inode *inode)
2727 {
2728 	struct buffer_head *dir_block = NULL;
2729 	struct ext4_dir_entry_2 *de;
2730 	ext4_lblk_t block = 0;
2731 	unsigned int blocksize = dir->i_sb->s_blocksize;
2732 	int csum_size = 0;
2733 	int err;
2734 
2735 	if (ext4_has_metadata_csum(dir->i_sb))
2736 		csum_size = sizeof(struct ext4_dir_entry_tail);
2737 
2738 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2739 		err = ext4_try_create_inline_dir(handle, dir, inode);
2740 		if (err < 0 && err != -ENOSPC)
2741 			goto out;
2742 		if (!err)
2743 			goto out;
2744 	}
2745 
2746 	inode->i_size = 0;
2747 	dir_block = ext4_append(handle, inode, &block);
2748 	if (IS_ERR(dir_block))
2749 		return PTR_ERR(dir_block);
2750 	de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2751 	ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2752 	set_nlink(inode, 2);
2753 	if (csum_size)
2754 		ext4_initialize_dirent_tail(dir_block, blocksize);
2755 
2756 	BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2757 	err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2758 	if (err)
2759 		goto out;
2760 	set_buffer_verified(dir_block);
2761 out:
2762 	brelse(dir_block);
2763 	return err;
2764 }
2765 
2766 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2767 {
2768 	handle_t *handle;
2769 	struct inode *inode;
2770 	int err, credits, retries = 0;
2771 
2772 	if (EXT4_DIR_LINK_MAX(dir))
2773 		return -EMLINK;
2774 
2775 	err = dquot_initialize(dir);
2776 	if (err)
2777 		return err;
2778 
2779 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2780 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2781 retry:
2782 	inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2783 					    &dentry->d_name,
2784 					    0, NULL, EXT4_HT_DIR, credits);
2785 	handle = ext4_journal_current_handle();
2786 	err = PTR_ERR(inode);
2787 	if (IS_ERR(inode))
2788 		goto out_stop;
2789 
2790 	inode->i_op = &ext4_dir_inode_operations;
2791 	inode->i_fop = &ext4_dir_operations;
2792 	err = ext4_init_new_dir(handle, dir, inode);
2793 	if (err)
2794 		goto out_clear_inode;
2795 	err = ext4_mark_inode_dirty(handle, inode);
2796 	if (!err)
2797 		err = ext4_add_entry(handle, dentry, inode);
2798 	if (err) {
2799 out_clear_inode:
2800 		clear_nlink(inode);
2801 		ext4_orphan_add(handle, inode);
2802 		unlock_new_inode(inode);
2803 		ext4_mark_inode_dirty(handle, inode);
2804 		ext4_journal_stop(handle);
2805 		iput(inode);
2806 		goto out_retry;
2807 	}
2808 	ext4_inc_count(handle, dir);
2809 	ext4_update_dx_flag(dir);
2810 	err = ext4_mark_inode_dirty(handle, dir);
2811 	if (err)
2812 		goto out_clear_inode;
2813 	d_instantiate_new(dentry, inode);
2814 	if (IS_DIRSYNC(dir))
2815 		ext4_handle_sync(handle);
2816 
2817 out_stop:
2818 	if (handle)
2819 		ext4_journal_stop(handle);
2820 out_retry:
2821 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2822 		goto retry;
2823 	return err;
2824 }
2825 
2826 /*
2827  * routine to check that the specified directory is empty (for rmdir)
2828  */
2829 bool ext4_empty_dir(struct inode *inode)
2830 {
2831 	unsigned int offset;
2832 	struct buffer_head *bh;
2833 	struct ext4_dir_entry_2 *de;
2834 	struct super_block *sb;
2835 
2836 	if (ext4_has_inline_data(inode)) {
2837 		int has_inline_data = 1;
2838 		int ret;
2839 
2840 		ret = empty_inline_dir(inode, &has_inline_data);
2841 		if (has_inline_data)
2842 			return ret;
2843 	}
2844 
2845 	sb = inode->i_sb;
2846 	if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2847 		EXT4_ERROR_INODE(inode, "invalid size");
2848 		return true;
2849 	}
2850 	/* The first directory block must not be a hole,
2851 	 * so treat it as DIRENT_HTREE
2852 	 */
2853 	bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2854 	if (IS_ERR(bh))
2855 		return true;
2856 
2857 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2858 	if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2859 				 0) ||
2860 	    le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2861 		ext4_warning_inode(inode, "directory missing '.'");
2862 		brelse(bh);
2863 		return true;
2864 	}
2865 	offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2866 	de = ext4_next_entry(de, sb->s_blocksize);
2867 	if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2868 				 offset) ||
2869 	    le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2870 		ext4_warning_inode(inode, "directory missing '..'");
2871 		brelse(bh);
2872 		return true;
2873 	}
2874 	offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2875 	while (offset < inode->i_size) {
2876 		if (!(offset & (sb->s_blocksize - 1))) {
2877 			unsigned int lblock;
2878 			brelse(bh);
2879 			lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2880 			bh = ext4_read_dirblock(inode, lblock, EITHER);
2881 			if (bh == NULL) {
2882 				offset += sb->s_blocksize;
2883 				continue;
2884 			}
2885 			if (IS_ERR(bh))
2886 				return true;
2887 		}
2888 		de = (struct ext4_dir_entry_2 *) (bh->b_data +
2889 					(offset & (sb->s_blocksize - 1)));
2890 		if (ext4_check_dir_entry(inode, NULL, de, bh,
2891 					 bh->b_data, bh->b_size, offset)) {
2892 			offset = (offset | (sb->s_blocksize - 1)) + 1;
2893 			continue;
2894 		}
2895 		if (le32_to_cpu(de->inode)) {
2896 			brelse(bh);
2897 			return false;
2898 		}
2899 		offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2900 	}
2901 	brelse(bh);
2902 	return true;
2903 }
2904 
2905 /*
2906  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2907  * such inodes, starting at the superblock, in case we crash before the
2908  * file is closed/deleted, or in case the inode truncate spans multiple
2909  * transactions and the last transaction is not recovered after a crash.
2910  *
2911  * At filesystem recovery time, we walk this list deleting unlinked
2912  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2913  *
2914  * Orphan list manipulation functions must be called under i_mutex unless
2915  * we are just creating the inode or deleting it.
2916  */
2917 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2918 {
2919 	struct super_block *sb = inode->i_sb;
2920 	struct ext4_sb_info *sbi = EXT4_SB(sb);
2921 	struct ext4_iloc iloc;
2922 	int err = 0, rc;
2923 	bool dirty = false;
2924 
2925 	if (!sbi->s_journal || is_bad_inode(inode))
2926 		return 0;
2927 
2928 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2929 		     !inode_is_locked(inode));
2930 	/*
2931 	 * Exit early if inode already is on orphan list. This is a big speedup
2932 	 * since we don't have to contend on the global s_orphan_lock.
2933 	 */
2934 	if (!list_empty(&EXT4_I(inode)->i_orphan))
2935 		return 0;
2936 
2937 	/*
2938 	 * Orphan handling is only valid for files with data blocks
2939 	 * being truncated, or files being unlinked. Note that we either
2940 	 * hold i_mutex, or the inode can not be referenced from outside,
2941 	 * so i_nlink should not be bumped due to race
2942 	 */
2943 	J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2944 		  S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2945 
2946 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2947 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2948 	if (err)
2949 		goto out;
2950 
2951 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2952 	if (err)
2953 		goto out;
2954 
2955 	mutex_lock(&sbi->s_orphan_lock);
2956 	/*
2957 	 * Due to previous errors inode may be already a part of on-disk
2958 	 * orphan list. If so skip on-disk list modification.
2959 	 */
2960 	if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2961 	    (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2962 		/* Insert this inode at the head of the on-disk orphan list */
2963 		NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2964 		sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2965 		dirty = true;
2966 	}
2967 	list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2968 	mutex_unlock(&sbi->s_orphan_lock);
2969 
2970 	if (dirty) {
2971 		err = ext4_handle_dirty_super(handle, sb);
2972 		rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2973 		if (!err)
2974 			err = rc;
2975 		if (err) {
2976 			/*
2977 			 * We have to remove inode from in-memory list if
2978 			 * addition to on disk orphan list failed. Stray orphan
2979 			 * list entries can cause panics at unmount time.
2980 			 */
2981 			mutex_lock(&sbi->s_orphan_lock);
2982 			list_del_init(&EXT4_I(inode)->i_orphan);
2983 			mutex_unlock(&sbi->s_orphan_lock);
2984 		}
2985 	} else
2986 		brelse(iloc.bh);
2987 
2988 	jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2989 	jbd_debug(4, "orphan inode %lu will point to %d\n",
2990 			inode->i_ino, NEXT_ORPHAN(inode));
2991 out:
2992 	ext4_std_error(sb, err);
2993 	return err;
2994 }
2995 
2996 /*
2997  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2998  * of such inodes stored on disk, because it is finally being cleaned up.
2999  */
3000 int ext4_orphan_del(handle_t *handle, struct inode *inode)
3001 {
3002 	struct list_head *prev;
3003 	struct ext4_inode_info *ei = EXT4_I(inode);
3004 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3005 	__u32 ino_next;
3006 	struct ext4_iloc iloc;
3007 	int err = 0;
3008 
3009 	if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3010 		return 0;
3011 
3012 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3013 		     !inode_is_locked(inode));
3014 	/* Do this quick check before taking global s_orphan_lock. */
3015 	if (list_empty(&ei->i_orphan))
3016 		return 0;
3017 
3018 	if (handle) {
3019 		/* Grab inode buffer early before taking global s_orphan_lock */
3020 		err = ext4_reserve_inode_write(handle, inode, &iloc);
3021 	}
3022 
3023 	mutex_lock(&sbi->s_orphan_lock);
3024 	jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3025 
3026 	prev = ei->i_orphan.prev;
3027 	list_del_init(&ei->i_orphan);
3028 
3029 	/* If we're on an error path, we may not have a valid
3030 	 * transaction handle with which to update the orphan list on
3031 	 * disk, but we still need to remove the inode from the linked
3032 	 * list in memory. */
3033 	if (!handle || err) {
3034 		mutex_unlock(&sbi->s_orphan_lock);
3035 		goto out_err;
3036 	}
3037 
3038 	ino_next = NEXT_ORPHAN(inode);
3039 	if (prev == &sbi->s_orphan) {
3040 		jbd_debug(4, "superblock will point to %u\n", ino_next);
3041 		BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3042 		err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3043 		if (err) {
3044 			mutex_unlock(&sbi->s_orphan_lock);
3045 			goto out_brelse;
3046 		}
3047 		sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3048 		mutex_unlock(&sbi->s_orphan_lock);
3049 		err = ext4_handle_dirty_super(handle, inode->i_sb);
3050 	} else {
3051 		struct ext4_iloc iloc2;
3052 		struct inode *i_prev =
3053 			&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3054 
3055 		jbd_debug(4, "orphan inode %lu will point to %u\n",
3056 			  i_prev->i_ino, ino_next);
3057 		err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3058 		if (err) {
3059 			mutex_unlock(&sbi->s_orphan_lock);
3060 			goto out_brelse;
3061 		}
3062 		NEXT_ORPHAN(i_prev) = ino_next;
3063 		err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3064 		mutex_unlock(&sbi->s_orphan_lock);
3065 	}
3066 	if (err)
3067 		goto out_brelse;
3068 	NEXT_ORPHAN(inode) = 0;
3069 	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3070 out_err:
3071 	ext4_std_error(inode->i_sb, err);
3072 	return err;
3073 
3074 out_brelse:
3075 	brelse(iloc.bh);
3076 	goto out_err;
3077 }
3078 
3079 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3080 {
3081 	int retval;
3082 	struct inode *inode;
3083 	struct buffer_head *bh;
3084 	struct ext4_dir_entry_2 *de;
3085 	handle_t *handle = NULL;
3086 
3087 	if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3088 		return -EIO;
3089 
3090 	/* Initialize quotas before so that eventual writes go in
3091 	 * separate transaction */
3092 	retval = dquot_initialize(dir);
3093 	if (retval)
3094 		return retval;
3095 	retval = dquot_initialize(d_inode(dentry));
3096 	if (retval)
3097 		return retval;
3098 
3099 	retval = -ENOENT;
3100 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3101 	if (IS_ERR(bh))
3102 		return PTR_ERR(bh);
3103 	if (!bh)
3104 		goto end_rmdir;
3105 
3106 	inode = d_inode(dentry);
3107 
3108 	retval = -EFSCORRUPTED;
3109 	if (le32_to_cpu(de->inode) != inode->i_ino)
3110 		goto end_rmdir;
3111 
3112 	retval = -ENOTEMPTY;
3113 	if (!ext4_empty_dir(inode))
3114 		goto end_rmdir;
3115 
3116 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3117 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3118 	if (IS_ERR(handle)) {
3119 		retval = PTR_ERR(handle);
3120 		handle = NULL;
3121 		goto end_rmdir;
3122 	}
3123 
3124 	if (IS_DIRSYNC(dir))
3125 		ext4_handle_sync(handle);
3126 
3127 	retval = ext4_delete_entry(handle, dir, de, bh);
3128 	if (retval)
3129 		goto end_rmdir;
3130 	if (!EXT4_DIR_LINK_EMPTY(inode))
3131 		ext4_warning_inode(inode,
3132 			     "empty directory '%.*s' has too many links (%u)",
3133 			     dentry->d_name.len, dentry->d_name.name,
3134 			     inode->i_nlink);
3135 	inode_inc_iversion(inode);
3136 	clear_nlink(inode);
3137 	/* There's no need to set i_disksize: the fact that i_nlink is
3138 	 * zero will ensure that the right thing happens during any
3139 	 * recovery. */
3140 	inode->i_size = 0;
3141 	ext4_orphan_add(handle, inode);
3142 	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3143 	ext4_mark_inode_dirty(handle, inode);
3144 	ext4_dec_count(handle, dir);
3145 	ext4_update_dx_flag(dir);
3146 	ext4_mark_inode_dirty(handle, dir);
3147 
3148 #ifdef CONFIG_UNICODE
3149 	/* VFS negative dentries are incompatible with Encoding and
3150 	 * Case-insensitiveness. Eventually we'll want avoid
3151 	 * invalidating the dentries here, alongside with returning the
3152 	 * negative dentries at ext4_lookup(), when it is better
3153 	 * supported by the VFS for the CI case.
3154 	 */
3155 	if (IS_CASEFOLDED(dir))
3156 		d_invalidate(dentry);
3157 #endif
3158 
3159 end_rmdir:
3160 	brelse(bh);
3161 	if (handle)
3162 		ext4_journal_stop(handle);
3163 	return retval;
3164 }
3165 
3166 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3167 {
3168 	int retval;
3169 	struct inode *inode;
3170 	struct buffer_head *bh;
3171 	struct ext4_dir_entry_2 *de;
3172 	handle_t *handle = NULL;
3173 
3174 	if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3175 		return -EIO;
3176 
3177 	trace_ext4_unlink_enter(dir, dentry);
3178 	/* Initialize quotas before so that eventual writes go
3179 	 * in separate transaction */
3180 	retval = dquot_initialize(dir);
3181 	if (retval)
3182 		return retval;
3183 	retval = dquot_initialize(d_inode(dentry));
3184 	if (retval)
3185 		return retval;
3186 
3187 	retval = -ENOENT;
3188 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3189 	if (IS_ERR(bh))
3190 		return PTR_ERR(bh);
3191 	if (!bh)
3192 		goto end_unlink;
3193 
3194 	inode = d_inode(dentry);
3195 
3196 	retval = -EFSCORRUPTED;
3197 	if (le32_to_cpu(de->inode) != inode->i_ino)
3198 		goto end_unlink;
3199 
3200 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3201 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3202 	if (IS_ERR(handle)) {
3203 		retval = PTR_ERR(handle);
3204 		handle = NULL;
3205 		goto end_unlink;
3206 	}
3207 
3208 	if (IS_DIRSYNC(dir))
3209 		ext4_handle_sync(handle);
3210 
3211 	retval = ext4_delete_entry(handle, dir, de, bh);
3212 	if (retval)
3213 		goto end_unlink;
3214 	dir->i_ctime = dir->i_mtime = current_time(dir);
3215 	ext4_update_dx_flag(dir);
3216 	ext4_mark_inode_dirty(handle, dir);
3217 	if (inode->i_nlink == 0)
3218 		ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3219 				   dentry->d_name.len, dentry->d_name.name);
3220 	else
3221 		drop_nlink(inode);
3222 	if (!inode->i_nlink)
3223 		ext4_orphan_add(handle, inode);
3224 	inode->i_ctime = current_time(inode);
3225 	ext4_mark_inode_dirty(handle, inode);
3226 
3227 #ifdef CONFIG_UNICODE
3228 	/* VFS negative dentries are incompatible with Encoding and
3229 	 * Case-insensitiveness. Eventually we'll want avoid
3230 	 * invalidating the dentries here, alongside with returning the
3231 	 * negative dentries at ext4_lookup(), when it is  better
3232 	 * supported by the VFS for the CI case.
3233 	 */
3234 	if (IS_CASEFOLDED(dir))
3235 		d_invalidate(dentry);
3236 #endif
3237 
3238 end_unlink:
3239 	brelse(bh);
3240 	if (handle)
3241 		ext4_journal_stop(handle);
3242 	trace_ext4_unlink_exit(dentry, retval);
3243 	return retval;
3244 }
3245 
3246 static int ext4_symlink(struct inode *dir,
3247 			struct dentry *dentry, const char *symname)
3248 {
3249 	handle_t *handle;
3250 	struct inode *inode;
3251 	int err, len = strlen(symname);
3252 	int credits;
3253 	struct fscrypt_str disk_link;
3254 
3255 	if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3256 		return -EIO;
3257 
3258 	err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3259 				      &disk_link);
3260 	if (err)
3261 		return err;
3262 
3263 	err = dquot_initialize(dir);
3264 	if (err)
3265 		return err;
3266 
3267 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3268 		/*
3269 		 * For non-fast symlinks, we just allocate inode and put it on
3270 		 * orphan list in the first transaction => we need bitmap,
3271 		 * group descriptor, sb, inode block, quota blocks, and
3272 		 * possibly selinux xattr blocks.
3273 		 */
3274 		credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3275 			  EXT4_XATTR_TRANS_BLOCKS;
3276 	} else {
3277 		/*
3278 		 * Fast symlink. We have to add entry to directory
3279 		 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3280 		 * allocate new inode (bitmap, group descriptor, inode block,
3281 		 * quota blocks, sb is already counted in previous macros).
3282 		 */
3283 		credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3284 			  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3285 	}
3286 
3287 	inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3288 					    &dentry->d_name, 0, NULL,
3289 					    EXT4_HT_DIR, credits);
3290 	handle = ext4_journal_current_handle();
3291 	if (IS_ERR(inode)) {
3292 		if (handle)
3293 			ext4_journal_stop(handle);
3294 		return PTR_ERR(inode);
3295 	}
3296 
3297 	if (IS_ENCRYPTED(inode)) {
3298 		err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3299 		if (err)
3300 			goto err_drop_inode;
3301 		inode->i_op = &ext4_encrypted_symlink_inode_operations;
3302 	}
3303 
3304 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3305 		if (!IS_ENCRYPTED(inode))
3306 			inode->i_op = &ext4_symlink_inode_operations;
3307 		inode_nohighmem(inode);
3308 		ext4_set_aops(inode);
3309 		/*
3310 		 * We cannot call page_symlink() with transaction started
3311 		 * because it calls into ext4_write_begin() which can wait
3312 		 * for transaction commit if we are running out of space
3313 		 * and thus we deadlock. So we have to stop transaction now
3314 		 * and restart it when symlink contents is written.
3315 		 *
3316 		 * To keep fs consistent in case of crash, we have to put inode
3317 		 * to orphan list in the mean time.
3318 		 */
3319 		drop_nlink(inode);
3320 		err = ext4_orphan_add(handle, inode);
3321 		ext4_journal_stop(handle);
3322 		handle = NULL;
3323 		if (err)
3324 			goto err_drop_inode;
3325 		err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3326 		if (err)
3327 			goto err_drop_inode;
3328 		/*
3329 		 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3330 		 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3331 		 */
3332 		handle = ext4_journal_start(dir, EXT4_HT_DIR,
3333 				EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3334 				EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3335 		if (IS_ERR(handle)) {
3336 			err = PTR_ERR(handle);
3337 			handle = NULL;
3338 			goto err_drop_inode;
3339 		}
3340 		set_nlink(inode, 1);
3341 		err = ext4_orphan_del(handle, inode);
3342 		if (err)
3343 			goto err_drop_inode;
3344 	} else {
3345 		/* clear the extent format for fast symlink */
3346 		ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3347 		if (!IS_ENCRYPTED(inode)) {
3348 			inode->i_op = &ext4_fast_symlink_inode_operations;
3349 			inode->i_link = (char *)&EXT4_I(inode)->i_data;
3350 		}
3351 		memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3352 		       disk_link.len);
3353 		inode->i_size = disk_link.len - 1;
3354 	}
3355 	EXT4_I(inode)->i_disksize = inode->i_size;
3356 	err = ext4_add_nondir(handle, dentry, &inode);
3357 	if (handle)
3358 		ext4_journal_stop(handle);
3359 	if (inode)
3360 		iput(inode);
3361 	goto out_free_encrypted_link;
3362 
3363 err_drop_inode:
3364 	if (handle)
3365 		ext4_journal_stop(handle);
3366 	clear_nlink(inode);
3367 	unlock_new_inode(inode);
3368 	iput(inode);
3369 out_free_encrypted_link:
3370 	if (disk_link.name != (unsigned char *)symname)
3371 		kfree(disk_link.name);
3372 	return err;
3373 }
3374 
3375 static int ext4_link(struct dentry *old_dentry,
3376 		     struct inode *dir, struct dentry *dentry)
3377 {
3378 	handle_t *handle;
3379 	struct inode *inode = d_inode(old_dentry);
3380 	int err, retries = 0;
3381 
3382 	if (inode->i_nlink >= EXT4_LINK_MAX)
3383 		return -EMLINK;
3384 
3385 	err = fscrypt_prepare_link(old_dentry, dir, dentry);
3386 	if (err)
3387 		return err;
3388 
3389 	if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3390 	    (!projid_eq(EXT4_I(dir)->i_projid,
3391 			EXT4_I(old_dentry->d_inode)->i_projid)))
3392 		return -EXDEV;
3393 
3394 	err = dquot_initialize(dir);
3395 	if (err)
3396 		return err;
3397 
3398 retry:
3399 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3400 		(EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3401 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3402 	if (IS_ERR(handle))
3403 		return PTR_ERR(handle);
3404 
3405 	if (IS_DIRSYNC(dir))
3406 		ext4_handle_sync(handle);
3407 
3408 	inode->i_ctime = current_time(inode);
3409 	ext4_inc_count(handle, inode);
3410 	ihold(inode);
3411 
3412 	err = ext4_add_entry(handle, dentry, inode);
3413 	if (!err) {
3414 		ext4_mark_inode_dirty(handle, inode);
3415 		/* this can happen only for tmpfile being
3416 		 * linked the first time
3417 		 */
3418 		if (inode->i_nlink == 1)
3419 			ext4_orphan_del(handle, inode);
3420 		d_instantiate(dentry, inode);
3421 	} else {
3422 		drop_nlink(inode);
3423 		iput(inode);
3424 	}
3425 	ext4_journal_stop(handle);
3426 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3427 		goto retry;
3428 	return err;
3429 }
3430 
3431 
3432 /*
3433  * Try to find buffer head where contains the parent block.
3434  * It should be the inode block if it is inlined or the 1st block
3435  * if it is a normal dir.
3436  */
3437 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3438 					struct inode *inode,
3439 					int *retval,
3440 					struct ext4_dir_entry_2 **parent_de,
3441 					int *inlined)
3442 {
3443 	struct buffer_head *bh;
3444 
3445 	if (!ext4_has_inline_data(inode)) {
3446 		/* The first directory block must not be a hole, so
3447 		 * treat it as DIRENT_HTREE
3448 		 */
3449 		bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3450 		if (IS_ERR(bh)) {
3451 			*retval = PTR_ERR(bh);
3452 			return NULL;
3453 		}
3454 		*parent_de = ext4_next_entry(
3455 					(struct ext4_dir_entry_2 *)bh->b_data,
3456 					inode->i_sb->s_blocksize);
3457 		return bh;
3458 	}
3459 
3460 	*inlined = 1;
3461 	return ext4_get_first_inline_block(inode, parent_de, retval);
3462 }
3463 
3464 struct ext4_renament {
3465 	struct inode *dir;
3466 	struct dentry *dentry;
3467 	struct inode *inode;
3468 	bool is_dir;
3469 	int dir_nlink_delta;
3470 
3471 	/* entry for "dentry" */
3472 	struct buffer_head *bh;
3473 	struct ext4_dir_entry_2 *de;
3474 	int inlined;
3475 
3476 	/* entry for ".." in inode if it's a directory */
3477 	struct buffer_head *dir_bh;
3478 	struct ext4_dir_entry_2 *parent_de;
3479 	int dir_inlined;
3480 };
3481 
3482 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3483 {
3484 	int retval;
3485 
3486 	ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3487 					      &retval, &ent->parent_de,
3488 					      &ent->dir_inlined);
3489 	if (!ent->dir_bh)
3490 		return retval;
3491 	if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3492 		return -EFSCORRUPTED;
3493 	BUFFER_TRACE(ent->dir_bh, "get_write_access");
3494 	return ext4_journal_get_write_access(handle, ent->dir_bh);
3495 }
3496 
3497 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3498 				  unsigned dir_ino)
3499 {
3500 	int retval;
3501 
3502 	ent->parent_de->inode = cpu_to_le32(dir_ino);
3503 	BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3504 	if (!ent->dir_inlined) {
3505 		if (is_dx(ent->inode)) {
3506 			retval = ext4_handle_dirty_dx_node(handle,
3507 							   ent->inode,
3508 							   ent->dir_bh);
3509 		} else {
3510 			retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3511 							    ent->dir_bh);
3512 		}
3513 	} else {
3514 		retval = ext4_mark_inode_dirty(handle, ent->inode);
3515 	}
3516 	if (retval) {
3517 		ext4_std_error(ent->dir->i_sb, retval);
3518 		return retval;
3519 	}
3520 	return 0;
3521 }
3522 
3523 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3524 		       unsigned ino, unsigned file_type)
3525 {
3526 	int retval;
3527 
3528 	BUFFER_TRACE(ent->bh, "get write access");
3529 	retval = ext4_journal_get_write_access(handle, ent->bh);
3530 	if (retval)
3531 		return retval;
3532 	ent->de->inode = cpu_to_le32(ino);
3533 	if (ext4_has_feature_filetype(ent->dir->i_sb))
3534 		ent->de->file_type = file_type;
3535 	inode_inc_iversion(ent->dir);
3536 	ent->dir->i_ctime = ent->dir->i_mtime =
3537 		current_time(ent->dir);
3538 	ext4_mark_inode_dirty(handle, ent->dir);
3539 	BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3540 	if (!ent->inlined) {
3541 		retval = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3542 		if (unlikely(retval)) {
3543 			ext4_std_error(ent->dir->i_sb, retval);
3544 			return retval;
3545 		}
3546 	}
3547 	brelse(ent->bh);
3548 	ent->bh = NULL;
3549 
3550 	return 0;
3551 }
3552 
3553 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3554 				  const struct qstr *d_name)
3555 {
3556 	int retval = -ENOENT;
3557 	struct buffer_head *bh;
3558 	struct ext4_dir_entry_2 *de;
3559 
3560 	bh = ext4_find_entry(dir, d_name, &de, NULL);
3561 	if (IS_ERR(bh))
3562 		return PTR_ERR(bh);
3563 	if (bh) {
3564 		retval = ext4_delete_entry(handle, dir, de, bh);
3565 		brelse(bh);
3566 	}
3567 	return retval;
3568 }
3569 
3570 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3571 			       int force_reread)
3572 {
3573 	int retval;
3574 	/*
3575 	 * ent->de could have moved from under us during htree split, so make
3576 	 * sure that we are deleting the right entry.  We might also be pointing
3577 	 * to a stale entry in the unused part of ent->bh so just checking inum
3578 	 * and the name isn't enough.
3579 	 */
3580 	if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3581 	    ent->de->name_len != ent->dentry->d_name.len ||
3582 	    strncmp(ent->de->name, ent->dentry->d_name.name,
3583 		    ent->de->name_len) ||
3584 	    force_reread) {
3585 		retval = ext4_find_delete_entry(handle, ent->dir,
3586 						&ent->dentry->d_name);
3587 	} else {
3588 		retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3589 		if (retval == -ENOENT) {
3590 			retval = ext4_find_delete_entry(handle, ent->dir,
3591 							&ent->dentry->d_name);
3592 		}
3593 	}
3594 
3595 	if (retval) {
3596 		ext4_warning_inode(ent->dir,
3597 				   "Deleting old file: nlink %d, error=%d",
3598 				   ent->dir->i_nlink, retval);
3599 	}
3600 }
3601 
3602 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3603 {
3604 	if (ent->dir_nlink_delta) {
3605 		if (ent->dir_nlink_delta == -1)
3606 			ext4_dec_count(handle, ent->dir);
3607 		else
3608 			ext4_inc_count(handle, ent->dir);
3609 		ext4_mark_inode_dirty(handle, ent->dir);
3610 	}
3611 }
3612 
3613 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3614 					      int credits, handle_t **h)
3615 {
3616 	struct inode *wh;
3617 	handle_t *handle;
3618 	int retries = 0;
3619 
3620 	/*
3621 	 * for inode block, sb block, group summaries,
3622 	 * and inode bitmap
3623 	 */
3624 	credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3625 		    EXT4_XATTR_TRANS_BLOCKS + 4);
3626 retry:
3627 	wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3628 					 &ent->dentry->d_name, 0, NULL,
3629 					 EXT4_HT_DIR, credits);
3630 
3631 	handle = ext4_journal_current_handle();
3632 	if (IS_ERR(wh)) {
3633 		if (handle)
3634 			ext4_journal_stop(handle);
3635 		if (PTR_ERR(wh) == -ENOSPC &&
3636 		    ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3637 			goto retry;
3638 	} else {
3639 		*h = handle;
3640 		init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3641 		wh->i_op = &ext4_special_inode_operations;
3642 	}
3643 	return wh;
3644 }
3645 
3646 /*
3647  * Anybody can rename anything with this: the permission checks are left to the
3648  * higher-level routines.
3649  *
3650  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3651  * while new_{dentry,inode) refers to the destination dentry/inode
3652  * This comes from rename(const char *oldpath, const char *newpath)
3653  */
3654 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3655 		       struct inode *new_dir, struct dentry *new_dentry,
3656 		       unsigned int flags)
3657 {
3658 	handle_t *handle = NULL;
3659 	struct ext4_renament old = {
3660 		.dir = old_dir,
3661 		.dentry = old_dentry,
3662 		.inode = d_inode(old_dentry),
3663 	};
3664 	struct ext4_renament new = {
3665 		.dir = new_dir,
3666 		.dentry = new_dentry,
3667 		.inode = d_inode(new_dentry),
3668 	};
3669 	int force_reread;
3670 	int retval;
3671 	struct inode *whiteout = NULL;
3672 	int credits;
3673 	u8 old_file_type;
3674 
3675 	if (new.inode && new.inode->i_nlink == 0) {
3676 		EXT4_ERROR_INODE(new.inode,
3677 				 "target of rename is already freed");
3678 		return -EFSCORRUPTED;
3679 	}
3680 
3681 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3682 	    (!projid_eq(EXT4_I(new_dir)->i_projid,
3683 			EXT4_I(old_dentry->d_inode)->i_projid)))
3684 		return -EXDEV;
3685 
3686 	retval = dquot_initialize(old.dir);
3687 	if (retval)
3688 		return retval;
3689 	retval = dquot_initialize(new.dir);
3690 	if (retval)
3691 		return retval;
3692 
3693 	/* Initialize quotas before so that eventual writes go
3694 	 * in separate transaction */
3695 	if (new.inode) {
3696 		retval = dquot_initialize(new.inode);
3697 		if (retval)
3698 			return retval;
3699 	}
3700 
3701 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3702 	if (IS_ERR(old.bh))
3703 		return PTR_ERR(old.bh);
3704 	/*
3705 	 *  Check for inode number is _not_ due to possible IO errors.
3706 	 *  We might rmdir the source, keep it as pwd of some process
3707 	 *  and merrily kill the link to whatever was created under the
3708 	 *  same name. Goodbye sticky bit ;-<
3709 	 */
3710 	retval = -ENOENT;
3711 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3712 		goto end_rename;
3713 
3714 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3715 				 &new.de, &new.inlined);
3716 	if (IS_ERR(new.bh)) {
3717 		retval = PTR_ERR(new.bh);
3718 		new.bh = NULL;
3719 		goto end_rename;
3720 	}
3721 	if (new.bh) {
3722 		if (!new.inode) {
3723 			brelse(new.bh);
3724 			new.bh = NULL;
3725 		}
3726 	}
3727 	if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3728 		ext4_alloc_da_blocks(old.inode);
3729 
3730 	credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3731 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3732 	if (!(flags & RENAME_WHITEOUT)) {
3733 		handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3734 		if (IS_ERR(handle)) {
3735 			retval = PTR_ERR(handle);
3736 			handle = NULL;
3737 			goto end_rename;
3738 		}
3739 	} else {
3740 		whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3741 		if (IS_ERR(whiteout)) {
3742 			retval = PTR_ERR(whiteout);
3743 			whiteout = NULL;
3744 			goto end_rename;
3745 		}
3746 	}
3747 
3748 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3749 		ext4_handle_sync(handle);
3750 
3751 	if (S_ISDIR(old.inode->i_mode)) {
3752 		if (new.inode) {
3753 			retval = -ENOTEMPTY;
3754 			if (!ext4_empty_dir(new.inode))
3755 				goto end_rename;
3756 		} else {
3757 			retval = -EMLINK;
3758 			if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3759 				goto end_rename;
3760 		}
3761 		retval = ext4_rename_dir_prepare(handle, &old);
3762 		if (retval)
3763 			goto end_rename;
3764 	}
3765 	/*
3766 	 * If we're renaming a file within an inline_data dir and adding or
3767 	 * setting the new dirent causes a conversion from inline_data to
3768 	 * extents/blockmap, we need to force the dirent delete code to
3769 	 * re-read the directory, or else we end up trying to delete a dirent
3770 	 * from what is now the extent tree root (or a block map).
3771 	 */
3772 	force_reread = (new.dir->i_ino == old.dir->i_ino &&
3773 			ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3774 
3775 	old_file_type = old.de->file_type;
3776 	if (whiteout) {
3777 		/*
3778 		 * Do this before adding a new entry, so the old entry is sure
3779 		 * to be still pointing to the valid old entry.
3780 		 */
3781 		retval = ext4_setent(handle, &old, whiteout->i_ino,
3782 				     EXT4_FT_CHRDEV);
3783 		if (retval)
3784 			goto end_rename;
3785 		ext4_mark_inode_dirty(handle, whiteout);
3786 	}
3787 	if (!new.bh) {
3788 		retval = ext4_add_entry(handle, new.dentry, old.inode);
3789 		if (retval)
3790 			goto end_rename;
3791 	} else {
3792 		retval = ext4_setent(handle, &new,
3793 				     old.inode->i_ino, old_file_type);
3794 		if (retval)
3795 			goto end_rename;
3796 	}
3797 	if (force_reread)
3798 		force_reread = !ext4_test_inode_flag(new.dir,
3799 						     EXT4_INODE_INLINE_DATA);
3800 
3801 	/*
3802 	 * Like most other Unix systems, set the ctime for inodes on a
3803 	 * rename.
3804 	 */
3805 	old.inode->i_ctime = current_time(old.inode);
3806 	ext4_mark_inode_dirty(handle, old.inode);
3807 
3808 	if (!whiteout) {
3809 		/*
3810 		 * ok, that's it
3811 		 */
3812 		ext4_rename_delete(handle, &old, force_reread);
3813 	}
3814 
3815 	if (new.inode) {
3816 		ext4_dec_count(handle, new.inode);
3817 		new.inode->i_ctime = current_time(new.inode);
3818 	}
3819 	old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3820 	ext4_update_dx_flag(old.dir);
3821 	if (old.dir_bh) {
3822 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3823 		if (retval)
3824 			goto end_rename;
3825 
3826 		ext4_dec_count(handle, old.dir);
3827 		if (new.inode) {
3828 			/* checked ext4_empty_dir above, can't have another
3829 			 * parent, ext4_dec_count() won't work for many-linked
3830 			 * dirs */
3831 			clear_nlink(new.inode);
3832 		} else {
3833 			ext4_inc_count(handle, new.dir);
3834 			ext4_update_dx_flag(new.dir);
3835 			ext4_mark_inode_dirty(handle, new.dir);
3836 		}
3837 	}
3838 	ext4_mark_inode_dirty(handle, old.dir);
3839 	if (new.inode) {
3840 		ext4_mark_inode_dirty(handle, new.inode);
3841 		if (!new.inode->i_nlink)
3842 			ext4_orphan_add(handle, new.inode);
3843 	}
3844 	retval = 0;
3845 
3846 end_rename:
3847 	brelse(old.dir_bh);
3848 	brelse(old.bh);
3849 	brelse(new.bh);
3850 	if (whiteout) {
3851 		if (retval)
3852 			drop_nlink(whiteout);
3853 		unlock_new_inode(whiteout);
3854 		iput(whiteout);
3855 	}
3856 	if (handle)
3857 		ext4_journal_stop(handle);
3858 	return retval;
3859 }
3860 
3861 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3862 			     struct inode *new_dir, struct dentry *new_dentry)
3863 {
3864 	handle_t *handle = NULL;
3865 	struct ext4_renament old = {
3866 		.dir = old_dir,
3867 		.dentry = old_dentry,
3868 		.inode = d_inode(old_dentry),
3869 	};
3870 	struct ext4_renament new = {
3871 		.dir = new_dir,
3872 		.dentry = new_dentry,
3873 		.inode = d_inode(new_dentry),
3874 	};
3875 	u8 new_file_type;
3876 	int retval;
3877 	struct timespec64 ctime;
3878 
3879 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3880 	     !projid_eq(EXT4_I(new_dir)->i_projid,
3881 			EXT4_I(old_dentry->d_inode)->i_projid)) ||
3882 	    (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3883 	     !projid_eq(EXT4_I(old_dir)->i_projid,
3884 			EXT4_I(new_dentry->d_inode)->i_projid)))
3885 		return -EXDEV;
3886 
3887 	retval = dquot_initialize(old.dir);
3888 	if (retval)
3889 		return retval;
3890 	retval = dquot_initialize(new.dir);
3891 	if (retval)
3892 		return retval;
3893 
3894 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3895 				 &old.de, &old.inlined);
3896 	if (IS_ERR(old.bh))
3897 		return PTR_ERR(old.bh);
3898 	/*
3899 	 *  Check for inode number is _not_ due to possible IO errors.
3900 	 *  We might rmdir the source, keep it as pwd of some process
3901 	 *  and merrily kill the link to whatever was created under the
3902 	 *  same name. Goodbye sticky bit ;-<
3903 	 */
3904 	retval = -ENOENT;
3905 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3906 		goto end_rename;
3907 
3908 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3909 				 &new.de, &new.inlined);
3910 	if (IS_ERR(new.bh)) {
3911 		retval = PTR_ERR(new.bh);
3912 		new.bh = NULL;
3913 		goto end_rename;
3914 	}
3915 
3916 	/* RENAME_EXCHANGE case: old *and* new must both exist */
3917 	if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3918 		goto end_rename;
3919 
3920 	handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3921 		(2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3922 		 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3923 	if (IS_ERR(handle)) {
3924 		retval = PTR_ERR(handle);
3925 		handle = NULL;
3926 		goto end_rename;
3927 	}
3928 
3929 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3930 		ext4_handle_sync(handle);
3931 
3932 	if (S_ISDIR(old.inode->i_mode)) {
3933 		old.is_dir = true;
3934 		retval = ext4_rename_dir_prepare(handle, &old);
3935 		if (retval)
3936 			goto end_rename;
3937 	}
3938 	if (S_ISDIR(new.inode->i_mode)) {
3939 		new.is_dir = true;
3940 		retval = ext4_rename_dir_prepare(handle, &new);
3941 		if (retval)
3942 			goto end_rename;
3943 	}
3944 
3945 	/*
3946 	 * Other than the special case of overwriting a directory, parents'
3947 	 * nlink only needs to be modified if this is a cross directory rename.
3948 	 */
3949 	if (old.dir != new.dir && old.is_dir != new.is_dir) {
3950 		old.dir_nlink_delta = old.is_dir ? -1 : 1;
3951 		new.dir_nlink_delta = -old.dir_nlink_delta;
3952 		retval = -EMLINK;
3953 		if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3954 		    (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3955 			goto end_rename;
3956 	}
3957 
3958 	new_file_type = new.de->file_type;
3959 	retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3960 	if (retval)
3961 		goto end_rename;
3962 
3963 	retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3964 	if (retval)
3965 		goto end_rename;
3966 
3967 	/*
3968 	 * Like most other Unix systems, set the ctime for inodes on a
3969 	 * rename.
3970 	 */
3971 	ctime = current_time(old.inode);
3972 	old.inode->i_ctime = ctime;
3973 	new.inode->i_ctime = ctime;
3974 	ext4_mark_inode_dirty(handle, old.inode);
3975 	ext4_mark_inode_dirty(handle, new.inode);
3976 
3977 	if (old.dir_bh) {
3978 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3979 		if (retval)
3980 			goto end_rename;
3981 	}
3982 	if (new.dir_bh) {
3983 		retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3984 		if (retval)
3985 			goto end_rename;
3986 	}
3987 	ext4_update_dir_count(handle, &old);
3988 	ext4_update_dir_count(handle, &new);
3989 	retval = 0;
3990 
3991 end_rename:
3992 	brelse(old.dir_bh);
3993 	brelse(new.dir_bh);
3994 	brelse(old.bh);
3995 	brelse(new.bh);
3996 	if (handle)
3997 		ext4_journal_stop(handle);
3998 	return retval;
3999 }
4000 
4001 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
4002 			struct inode *new_dir, struct dentry *new_dentry,
4003 			unsigned int flags)
4004 {
4005 	int err;
4006 
4007 	if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4008 		return -EIO;
4009 
4010 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4011 		return -EINVAL;
4012 
4013 	err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4014 				     flags);
4015 	if (err)
4016 		return err;
4017 
4018 	if (flags & RENAME_EXCHANGE) {
4019 		return ext4_cross_rename(old_dir, old_dentry,
4020 					 new_dir, new_dentry);
4021 	}
4022 
4023 	return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4024 }
4025 
4026 /*
4027  * directories can handle most operations...
4028  */
4029 const struct inode_operations ext4_dir_inode_operations = {
4030 	.create		= ext4_create,
4031 	.lookup		= ext4_lookup,
4032 	.link		= ext4_link,
4033 	.unlink		= ext4_unlink,
4034 	.symlink	= ext4_symlink,
4035 	.mkdir		= ext4_mkdir,
4036 	.rmdir		= ext4_rmdir,
4037 	.mknod		= ext4_mknod,
4038 	.tmpfile	= ext4_tmpfile,
4039 	.rename		= ext4_rename2,
4040 	.setattr	= ext4_setattr,
4041 	.getattr	= ext4_getattr,
4042 	.listxattr	= ext4_listxattr,
4043 	.get_acl	= ext4_get_acl,
4044 	.set_acl	= ext4_set_acl,
4045 	.fiemap         = ext4_fiemap,
4046 };
4047 
4048 const struct inode_operations ext4_special_inode_operations = {
4049 	.setattr	= ext4_setattr,
4050 	.getattr	= ext4_getattr,
4051 	.listxattr	= ext4_listxattr,
4052 	.get_acl	= ext4_get_acl,
4053 	.set_acl	= ext4_set_acl,
4054 };
4055