xref: /linux/fs/ntfs3/index.c (revision 52990390f91c1c39ca742fc8f390b29891d95127)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *
4  * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
5  *
6  */
7 
8 #include <linux/blkdev.h>
9 #include <linux/buffer_head.h>
10 #include <linux/fs.h>
11 #include <linux/kernel.h>
12 
13 #include "debug.h"
14 #include "ntfs.h"
15 #include "ntfs_fs.h"
16 
17 static const struct INDEX_NAMES {
18 	const __le16 *name;
19 	u8 name_len;
20 } s_index_names[INDEX_MUTEX_TOTAL] = {
21 	{ I30_NAME, ARRAY_SIZE(I30_NAME) }, { SII_NAME, ARRAY_SIZE(SII_NAME) },
22 	{ SDH_NAME, ARRAY_SIZE(SDH_NAME) }, { SO_NAME, ARRAY_SIZE(SO_NAME) },
23 	{ SQ_NAME, ARRAY_SIZE(SQ_NAME) },   { SR_NAME, ARRAY_SIZE(SR_NAME) },
24 };
25 
26 /*
27  * cmp_fnames - Compare two names in index.
28  *
29  * if l1 != 0
30  *   Both names are little endian on-disk ATTR_FILE_NAME structs.
31  * else
32  *   key1 - cpu_str, key2 - ATTR_FILE_NAME
33  */
34 static int cmp_fnames(const void *key1, size_t l1, const void *key2, size_t l2,
35 		      const void *data)
36 {
37 	const struct ATTR_FILE_NAME *f2 = key2;
38 	const struct ntfs_sb_info *sbi = data;
39 	const struct ATTR_FILE_NAME *f1;
40 	u16 fsize2;
41 	bool both_case;
42 
43 	if (l2 <= offsetof(struct ATTR_FILE_NAME, name))
44 		return -1;
45 
46 	fsize2 = fname_full_size(f2);
47 	if (l2 < fsize2)
48 		return -1;
49 
50 	both_case = f2->type != FILE_NAME_DOS && !sbi->options->nocase;
51 	if (!l1) {
52 		const struct le_str *s2 = (struct le_str *)&f2->name_len;
53 
54 		/*
55 		 * If names are equal (case insensitive)
56 		 * try to compare it case sensitive.
57 		 */
58 		return ntfs_cmp_names_cpu(key1, s2, sbi->upcase, both_case);
59 	}
60 
61 	f1 = key1;
62 	return ntfs_cmp_names(f1->name, f1->name_len, f2->name, f2->name_len,
63 			      sbi->upcase, both_case);
64 }
65 
66 /*
67  * cmp_uint - $SII of $Secure and $Q of Quota
68  */
69 static int cmp_uint(const void *key1, size_t l1, const void *key2, size_t l2,
70 		    const void *data)
71 {
72 	const u32 *k1 = key1;
73 	const u32 *k2 = key2;
74 
75 	if (l2 < sizeof(u32))
76 		return -1;
77 
78 	if (*k1 < *k2)
79 		return -1;
80 	if (*k1 > *k2)
81 		return 1;
82 	return 0;
83 }
84 
85 /*
86  * cmp_sdh - $SDH of $Secure
87  */
88 static int cmp_sdh(const void *key1, size_t l1, const void *key2, size_t l2,
89 		   const void *data)
90 {
91 	const struct SECURITY_KEY *k1 = key1;
92 	const struct SECURITY_KEY *k2 = key2;
93 	u32 t1, t2;
94 
95 	if (l2 < sizeof(struct SECURITY_KEY))
96 		return -1;
97 
98 	t1 = le32_to_cpu(k1->hash);
99 	t2 = le32_to_cpu(k2->hash);
100 
101 	/* First value is a hash value itself. */
102 	if (t1 < t2)
103 		return -1;
104 	if (t1 > t2)
105 		return 1;
106 
107 	/* Second value is security Id. */
108 	if (data) {
109 		t1 = le32_to_cpu(k1->sec_id);
110 		t2 = le32_to_cpu(k2->sec_id);
111 		if (t1 < t2)
112 			return -1;
113 		if (t1 > t2)
114 			return 1;
115 	}
116 
117 	return 0;
118 }
119 
120 /*
121  * cmp_uints - $O of ObjId and "$R" for Reparse.
122  */
123 static int cmp_uints(const void *key1, size_t l1, const void *key2, size_t l2,
124 		     const void *data)
125 {
126 	const __le32 *k1 = key1;
127 	const __le32 *k2 = key2;
128 	size_t count;
129 
130 	if ((size_t)data == 1) {
131 		/*
132 		 * ni_delete_all -> ntfs_remove_reparse ->
133 		 * delete all with this reference.
134 		 * k1, k2 - pointers to REPARSE_KEY
135 		 */
136 
137 		k1 += 1; // Skip REPARSE_KEY.ReparseTag
138 		k2 += 1; // Skip REPARSE_KEY.ReparseTag
139 		if (l2 <= sizeof(int))
140 			return -1;
141 		l2 -= sizeof(int);
142 		if (l1 <= sizeof(int))
143 			return 1;
144 		l1 -= sizeof(int);
145 	}
146 
147 	if (l2 < sizeof(int))
148 		return -1;
149 
150 	for (count = min(l1, l2) >> 2; count > 0; --count, ++k1, ++k2) {
151 		u32 t1 = le32_to_cpu(*k1);
152 		u32 t2 = le32_to_cpu(*k2);
153 
154 		if (t1 > t2)
155 			return 1;
156 		if (t1 < t2)
157 			return -1;
158 	}
159 
160 	if (l1 > l2)
161 		return 1;
162 	if (l1 < l2)
163 		return -1;
164 
165 	return 0;
166 }
167 
168 static inline NTFS_CMP_FUNC get_cmp_func(const struct INDEX_ROOT *root)
169 {
170 	switch (root->type) {
171 	case ATTR_NAME:
172 		if (root->rule == NTFS_COLLATION_TYPE_FILENAME)
173 			return &cmp_fnames;
174 		break;
175 	case ATTR_ZERO:
176 		switch (root->rule) {
177 		case NTFS_COLLATION_TYPE_UINT:
178 			return &cmp_uint;
179 		case NTFS_COLLATION_TYPE_SECURITY_HASH:
180 			return &cmp_sdh;
181 		case NTFS_COLLATION_TYPE_UINTS:
182 			return &cmp_uints;
183 		default:
184 			break;
185 		}
186 		break;
187 	default:
188 		break;
189 	}
190 
191 	return NULL;
192 }
193 
194 struct bmp_buf {
195 	struct ATTRIB *b;
196 	struct mft_inode *mi;
197 	struct buffer_head *bh;
198 	ulong *buf;
199 	size_t bit;
200 	u32 nbits;
201 	u64 new_valid;
202 };
203 
204 static int bmp_buf_get(struct ntfs_index *indx, struct ntfs_inode *ni,
205 		       size_t bit, struct bmp_buf *bbuf)
206 {
207 	struct ATTRIB *b;
208 	size_t data_size, valid_size, vbo, off = bit >> 3;
209 	struct ntfs_sb_info *sbi = ni->mi.sbi;
210 	CLST vcn = off >> sbi->cluster_bits;
211 	struct ATTR_LIST_ENTRY *le = NULL;
212 	struct buffer_head *bh;
213 	struct super_block *sb;
214 	u32 blocksize;
215 	const struct INDEX_NAMES *in = &s_index_names[indx->type];
216 
217 	bbuf->bh = NULL;
218 
219 	b = ni_find_attr(ni, NULL, &le, ATTR_BITMAP, in->name, in->name_len,
220 			 &vcn, &bbuf->mi);
221 	bbuf->b = b;
222 	if (!b)
223 		return -EINVAL;
224 
225 	if (!b->non_res) {
226 		data_size = le32_to_cpu(b->res.data_size);
227 
228 		if (off >= data_size)
229 			return -EINVAL;
230 
231 		bbuf->buf = (ulong *)resident_data(b);
232 		bbuf->bit = 0;
233 		bbuf->nbits = data_size * 8;
234 
235 		return 0;
236 	}
237 
238 	data_size = le64_to_cpu(b->nres.data_size);
239 	if (WARN_ON(off >= data_size)) {
240 		/* Looks like filesystem error. */
241 		return -EINVAL;
242 	}
243 
244 	valid_size = le64_to_cpu(b->nres.valid_size);
245 
246 	bh = ntfs_bread_run(sbi, &indx->bitmap_run, off);
247 	if (!bh)
248 		return -EIO;
249 
250 	if (IS_ERR(bh))
251 		return PTR_ERR(bh);
252 
253 	bbuf->bh = bh;
254 
255 	if (buffer_locked(bh))
256 		__wait_on_buffer(bh);
257 
258 	lock_buffer(bh);
259 
260 	sb = sbi->sb;
261 	blocksize = sb->s_blocksize;
262 
263 	vbo = off & ~(size_t)sbi->block_mask;
264 
265 	bbuf->new_valid = vbo + blocksize;
266 	if (bbuf->new_valid <= valid_size)
267 		bbuf->new_valid = 0;
268 	else if (bbuf->new_valid > data_size)
269 		bbuf->new_valid = data_size;
270 
271 	if (vbo >= valid_size) {
272 		memset(bh->b_data, 0, blocksize);
273 	} else if (vbo + blocksize > valid_size) {
274 		u32 voff = valid_size & sbi->block_mask;
275 
276 		memset(bh->b_data + voff, 0, blocksize - voff);
277 	}
278 
279 	bbuf->buf = (ulong *)bh->b_data;
280 	bbuf->bit = 8 * (off & ~(size_t)sbi->block_mask);
281 	bbuf->nbits = 8 * blocksize;
282 
283 	return 0;
284 }
285 
286 static void bmp_buf_put(struct bmp_buf *bbuf, bool dirty)
287 {
288 	struct buffer_head *bh = bbuf->bh;
289 	struct ATTRIB *b = bbuf->b;
290 
291 	if (!bh) {
292 		if (b && !b->non_res && dirty)
293 			bbuf->mi->dirty = true;
294 		return;
295 	}
296 
297 	if (!dirty)
298 		goto out;
299 
300 	if (bbuf->new_valid) {
301 		b->nres.valid_size = cpu_to_le64(bbuf->new_valid);
302 		bbuf->mi->dirty = true;
303 	}
304 
305 	set_buffer_uptodate(bh);
306 	mark_buffer_dirty(bh);
307 
308 out:
309 	unlock_buffer(bh);
310 	put_bh(bh);
311 }
312 
313 /*
314  * indx_mark_used - Mark the bit @bit as used.
315  */
316 static int indx_mark_used(struct ntfs_index *indx, struct ntfs_inode *ni,
317 			  size_t bit)
318 {
319 	int err;
320 	struct bmp_buf bbuf;
321 
322 	err = bmp_buf_get(indx, ni, bit, &bbuf);
323 	if (err)
324 		return err;
325 
326 	__set_bit_le(bit - bbuf.bit, bbuf.buf);
327 
328 	bmp_buf_put(&bbuf, true);
329 
330 	return 0;
331 }
332 
333 /*
334  * indx_mark_free - Mark the bit @bit as free.
335  */
336 static int indx_mark_free(struct ntfs_index *indx, struct ntfs_inode *ni,
337 			  size_t bit)
338 {
339 	int err;
340 	struct bmp_buf bbuf;
341 
342 	err = bmp_buf_get(indx, ni, bit, &bbuf);
343 	if (err)
344 		return err;
345 
346 	__clear_bit_le(bit - bbuf.bit, bbuf.buf);
347 
348 	bmp_buf_put(&bbuf, true);
349 
350 	return 0;
351 }
352 
353 /*
354  * scan_nres_bitmap
355  *
356  * If ntfs_readdir calls this function (indx_used_bit -> scan_nres_bitmap),
357  * inode is shared locked and no ni_lock.
358  * Use rw_semaphore for read/write access to bitmap_run.
359  */
360 static int scan_nres_bitmap(struct ntfs_inode *ni, struct ATTRIB *bitmap,
361 			    struct ntfs_index *indx, size_t from,
362 			    bool (*fn)(const ulong *buf, u32 bit, u32 bits,
363 				       size_t *ret),
364 			    size_t *ret)
365 {
366 	struct ntfs_sb_info *sbi = ni->mi.sbi;
367 	struct super_block *sb = sbi->sb;
368 	struct runs_tree *run = &indx->bitmap_run;
369 	struct rw_semaphore *lock = &indx->run_lock;
370 	u32 nbits = sb->s_blocksize * 8;
371 	u32 blocksize = sb->s_blocksize;
372 	u64 valid_size = le64_to_cpu(bitmap->nres.valid_size);
373 	u64 data_size = le64_to_cpu(bitmap->nres.data_size);
374 	sector_t eblock = bytes_to_block(sb, data_size);
375 	size_t vbo = from >> 3;
376 	sector_t blk = (vbo & sbi->cluster_mask) >> sb->s_blocksize_bits;
377 	sector_t vblock = vbo >> sb->s_blocksize_bits;
378 	sector_t blen, block;
379 	CLST lcn, clen, vcn, vcn_next;
380 	size_t idx;
381 	struct buffer_head *bh;
382 	bool ok;
383 
384 	*ret = MINUS_ONE_T;
385 
386 	if (vblock >= eblock)
387 		return 0;
388 
389 	from &= nbits - 1;
390 	vcn = vbo >> sbi->cluster_bits;
391 
392 	down_read(lock);
393 	ok = run_lookup_entry(run, vcn, &lcn, &clen, &idx);
394 	up_read(lock);
395 
396 next_run:
397 	if (!ok) {
398 		int err;
399 		const struct INDEX_NAMES *name = &s_index_names[indx->type];
400 
401 		down_write(lock);
402 		err = attr_load_runs_vcn(ni, ATTR_BITMAP, name->name,
403 					 name->name_len, run, vcn);
404 		up_write(lock);
405 		if (err)
406 			return err;
407 		down_read(lock);
408 		ok = run_lookup_entry(run, vcn, &lcn, &clen, &idx);
409 		up_read(lock);
410 		if (!ok)
411 			return -EINVAL;
412 	}
413 
414 	blen = (sector_t)clen * sbi->blocks_per_cluster;
415 	block = (sector_t)lcn * sbi->blocks_per_cluster;
416 
417 	for (; blk < blen; blk++, from = 0) {
418 		bh = ntfs_bread(sb, block + blk);
419 		if (!bh)
420 			return -EIO;
421 
422 		vbo = (u64)vblock << sb->s_blocksize_bits;
423 		if (vbo >= valid_size) {
424 			memset(bh->b_data, 0, blocksize);
425 		} else if (vbo + blocksize > valid_size) {
426 			u32 voff = valid_size & sbi->block_mask;
427 
428 			memset(bh->b_data + voff, 0, blocksize - voff);
429 		}
430 
431 		if (vbo + blocksize > data_size)
432 			nbits = 8 * (data_size - vbo);
433 
434 		ok = nbits > from ?
435 				   (*fn)((ulong *)bh->b_data, from, nbits, ret) :
436 				   false;
437 		put_bh(bh);
438 
439 		if (ok) {
440 			*ret += 8 * vbo;
441 			return 0;
442 		}
443 
444 		if (++vblock >= eblock) {
445 			*ret = MINUS_ONE_T;
446 			return 0;
447 		}
448 	}
449 	blk = 0;
450 	vcn_next = vcn + clen;
451 	down_read(lock);
452 	ok = run_get_entry(run, ++idx, &vcn, &lcn, &clen) && vcn == vcn_next;
453 	if (!ok)
454 		vcn = vcn_next;
455 	up_read(lock);
456 	goto next_run;
457 }
458 
459 static bool scan_for_free(const ulong *buf, u32 bit, u32 bits, size_t *ret)
460 {
461 	size_t pos = find_next_zero_bit_le(buf, bits, bit);
462 
463 	if (pos >= bits)
464 		return false;
465 	*ret = pos;
466 	return true;
467 }
468 
469 /*
470  * indx_find_free - Look for free bit.
471  *
472  * Return: -1 if no free bits.
473  */
474 static int indx_find_free(struct ntfs_index *indx, struct ntfs_inode *ni,
475 			  size_t *bit, struct ATTRIB **bitmap)
476 {
477 	struct ATTRIB *b;
478 	struct ATTR_LIST_ENTRY *le = NULL;
479 	const struct INDEX_NAMES *in = &s_index_names[indx->type];
480 	int err;
481 
482 	b = ni_find_attr(ni, NULL, &le, ATTR_BITMAP, in->name, in->name_len,
483 			 NULL, NULL);
484 
485 	if (!b)
486 		return -ENOENT;
487 
488 	*bitmap = b;
489 	*bit = MINUS_ONE_T;
490 
491 	if (!b->non_res) {
492 		u32 nbits = 8 * le32_to_cpu(b->res.data_size);
493 		size_t pos = find_next_zero_bit_le(resident_data(b), nbits, 0);
494 
495 		if (pos < nbits)
496 			*bit = pos;
497 	} else {
498 		err = scan_nres_bitmap(ni, b, indx, 0, &scan_for_free, bit);
499 
500 		if (err)
501 			return err;
502 	}
503 
504 	return 0;
505 }
506 
507 static bool scan_for_used(const ulong *buf, u32 bit, u32 bits, size_t *ret)
508 {
509 	size_t pos = find_next_bit_le(buf, bits, bit);
510 
511 	if (pos >= bits)
512 		return false;
513 	*ret = pos;
514 	return true;
515 }
516 
517 /*
518  * indx_used_bit - Look for used bit.
519  *
520  * Return: MINUS_ONE_T if no used bits.
521  */
522 int indx_used_bit(struct ntfs_index *indx, struct ntfs_inode *ni, size_t *bit)
523 {
524 	struct ATTRIB *b;
525 	struct ATTR_LIST_ENTRY *le = NULL;
526 	size_t from = *bit;
527 	const struct INDEX_NAMES *in = &s_index_names[indx->type];
528 	int err;
529 
530 	b = ni_find_attr(ni, NULL, &le, ATTR_BITMAP, in->name, in->name_len,
531 			 NULL, NULL);
532 
533 	if (!b)
534 		return -ENOENT;
535 
536 	*bit = MINUS_ONE_T;
537 
538 	if (!b->non_res) {
539 		u32 nbits = le32_to_cpu(b->res.data_size) * 8;
540 		size_t pos = find_next_bit_le(resident_data(b), nbits, from);
541 
542 		if (pos < nbits)
543 			*bit = pos;
544 	} else {
545 		err = scan_nres_bitmap(ni, b, indx, from, &scan_for_used, bit);
546 		if (err)
547 			return err;
548 	}
549 
550 	return 0;
551 }
552 
553 /*
554  * hdr_find_split
555  *
556  * Find a point at which the index allocation buffer would like to be split.
557  * NOTE: This function should never return 'END' entry NULL returns on error.
558  */
559 static const struct NTFS_DE *hdr_find_split(const struct INDEX_HDR *hdr)
560 {
561 	size_t o;
562 	const struct NTFS_DE *e = hdr_first_de(hdr);
563 	u32 used_2 = le32_to_cpu(hdr->used) >> 1;
564 	u16 esize;
565 
566 	if (!e || de_is_last(e))
567 		return NULL;
568 
569 	esize = le16_to_cpu(e->size);
570 	for (o = le32_to_cpu(hdr->de_off) + esize; o < used_2; o += esize) {
571 		const struct NTFS_DE *p = e;
572 
573 		e = Add2Ptr(hdr, o);
574 
575 		/* We must not return END entry. */
576 		if (de_is_last(e))
577 			return p;
578 
579 		esize = le16_to_cpu(e->size);
580 	}
581 
582 	return e;
583 }
584 
585 /*
586  * hdr_insert_head - Insert some entries at the beginning of the buffer.
587  *
588  * It is used to insert entries into a newly-created buffer.
589  */
590 static const struct NTFS_DE *hdr_insert_head(struct INDEX_HDR *hdr,
591 					     const void *ins, u32 ins_bytes)
592 {
593 	u32 to_move;
594 	struct NTFS_DE *e = hdr_first_de(hdr);
595 	u32 used = le32_to_cpu(hdr->used);
596 
597 	if (!e)
598 		return NULL;
599 
600 	/* Now we just make room for the inserted entries and jam it in. */
601 	to_move = used - le32_to_cpu(hdr->de_off);
602 	memmove(Add2Ptr(e, ins_bytes), e, to_move);
603 	memcpy(e, ins, ins_bytes);
604 	hdr->used = cpu_to_le32(used + ins_bytes);
605 
606 	return e;
607 }
608 
609 /*
610  * index_hdr_check
611  *
612  * return true if INDEX_HDR is valid
613  */
614 static bool index_hdr_check(const struct INDEX_HDR *hdr, u32 bytes)
615 {
616 	u32 end = le32_to_cpu(hdr->used);
617 	u32 tot = le32_to_cpu(hdr->total);
618 	u32 off = le32_to_cpu(hdr->de_off);
619 
620 	if (!IS_ALIGNED(off, 8) || tot > bytes || end > tot ||
621 	    off + sizeof(struct NTFS_DE) > end) {
622 		/* incorrect index buffer. */
623 		return false;
624 	}
625 
626 	return true;
627 }
628 
629 /*
630  * index_buf_check
631  *
632  * return true if INDEX_BUFFER seems is valid
633  */
634 static bool index_buf_check(const struct INDEX_BUFFER *ib, u32 bytes,
635 			    const CLST *vbn)
636 {
637 	const struct NTFS_RECORD_HEADER *rhdr = &ib->rhdr;
638 	u16 fo = le16_to_cpu(rhdr->fix_off);
639 	u16 fn = le16_to_cpu(rhdr->fix_num);
640 
641 	if (bytes <= offsetof(struct INDEX_BUFFER, ihdr) ||
642 	    rhdr->sign != NTFS_INDX_SIGNATURE ||
643 	    fo < sizeof(struct INDEX_BUFFER)
644 	    /* Check index buffer vbn. */
645 	    || (vbn && *vbn != le64_to_cpu(ib->vbn)) || (fo % sizeof(short)) ||
646 	    fo + fn * sizeof(short) >= bytes ||
647 	    fn != ((bytes >> SECTOR_SHIFT) + 1)) {
648 		/* incorrect index buffer. */
649 		return false;
650 	}
651 
652 	return index_hdr_check(&ib->ihdr,
653 			       bytes - offsetof(struct INDEX_BUFFER, ihdr));
654 }
655 
656 void fnd_clear(struct ntfs_fnd *fnd)
657 {
658 	int i;
659 
660 	for (i = fnd->level - 1; i >= 0; i--) {
661 		struct indx_node *n = fnd->nodes[i];
662 
663 		if (!n)
664 			continue;
665 
666 		put_indx_node(n);
667 		fnd->nodes[i] = NULL;
668 	}
669 	fnd->level = 0;
670 	fnd->root_de = NULL;
671 }
672 
673 static int fnd_push(struct ntfs_fnd *fnd, struct indx_node *n,
674 		    struct NTFS_DE *e)
675 {
676 	int i = fnd->level;
677 
678 	if (i < 0 || i >= ARRAY_SIZE(fnd->nodes))
679 		return -EINVAL;
680 	fnd->nodes[i] = n;
681 	fnd->de[i] = e;
682 	fnd->level += 1;
683 	return 0;
684 }
685 
686 static struct indx_node *fnd_pop(struct ntfs_fnd *fnd)
687 {
688 	struct indx_node *n;
689 	int i = fnd->level;
690 
691 	i -= 1;
692 	n = fnd->nodes[i];
693 	fnd->nodes[i] = NULL;
694 	fnd->level = i;
695 
696 	return n;
697 }
698 
699 static bool fnd_is_empty(struct ntfs_fnd *fnd)
700 {
701 	if (!fnd->level)
702 		return !fnd->root_de;
703 
704 	return !fnd->de[fnd->level - 1];
705 }
706 
707 /*
708  * hdr_find_e - Locate an entry the index buffer.
709  *
710  * If no matching entry is found, it returns the first entry which is greater
711  * than the desired entry If the search key is greater than all the entries the
712  * buffer, it returns the 'end' entry. This function does a binary search of the
713  * current index buffer, for the first entry that is <= to the search value.
714  *
715  * Return: NULL if error.
716  */
717 static struct NTFS_DE *hdr_find_e(const struct ntfs_index *indx,
718 				  const struct INDEX_HDR *hdr, const void *key,
719 				  size_t key_len, const void *ctx, int *diff)
720 {
721 	struct NTFS_DE *e, *found = NULL;
722 	NTFS_CMP_FUNC cmp = indx->cmp;
723 	int min_idx = 0, mid_idx, max_idx = 0;
724 	int diff2;
725 	int table_size = 8;
726 	u32 e_size, e_key_len;
727 	u32 end = le32_to_cpu(hdr->used);
728 	u32 off = le32_to_cpu(hdr->de_off);
729 	u32 total = le32_to_cpu(hdr->total);
730 	u16 offs[128];
731 
732 fill_table:
733 	if (end > total)
734 		return NULL;
735 
736 	if (off + sizeof(struct NTFS_DE) > end)
737 		return NULL;
738 
739 	e = Add2Ptr(hdr, off);
740 	e_size = le16_to_cpu(e->size);
741 
742 	if (e_size < sizeof(struct NTFS_DE) || off + e_size > end)
743 		return NULL;
744 
745 	if (!de_is_last(e)) {
746 		offs[max_idx] = off;
747 		off += e_size;
748 
749 		max_idx++;
750 		if (max_idx < table_size)
751 			goto fill_table;
752 
753 		max_idx--;
754 	}
755 
756 binary_search:
757 	e_key_len = le16_to_cpu(e->key_size);
758 
759 	diff2 = (*cmp)(key, key_len, e + 1, e_key_len, ctx);
760 	if (diff2 > 0) {
761 		if (found) {
762 			min_idx = mid_idx + 1;
763 		} else {
764 			if (de_is_last(e))
765 				return NULL;
766 
767 			max_idx = 0;
768 			table_size = min(table_size * 2, (int)ARRAY_SIZE(offs));
769 			goto fill_table;
770 		}
771 	} else if (diff2 < 0) {
772 		if (found)
773 			max_idx = mid_idx - 1;
774 		else
775 			max_idx--;
776 
777 		found = e;
778 	} else {
779 		*diff = 0;
780 		return e;
781 	}
782 
783 	if (min_idx > max_idx) {
784 		*diff = -1;
785 		return found;
786 	}
787 
788 	mid_idx = (min_idx + max_idx) >> 1;
789 	e = Add2Ptr(hdr, offs[mid_idx]);
790 
791 	goto binary_search;
792 }
793 
794 /*
795  * hdr_insert_de - Insert an index entry into the buffer.
796  *
797  * 'before' should be a pointer previously returned from hdr_find_e.
798  */
799 static struct NTFS_DE *hdr_insert_de(const struct ntfs_index *indx,
800 				     struct INDEX_HDR *hdr,
801 				     const struct NTFS_DE *de,
802 				     struct NTFS_DE *before, const void *ctx)
803 {
804 	int diff;
805 	size_t off = PtrOffset(hdr, before);
806 	u32 used = le32_to_cpu(hdr->used);
807 	u32 total = le32_to_cpu(hdr->total);
808 	u16 de_size = le16_to_cpu(de->size);
809 
810 	/* First, check to see if there's enough room. */
811 	if (used + de_size > total)
812 		return NULL;
813 
814 	/* We know there's enough space, so we know we'll succeed. */
815 	if (before) {
816 		/* Check that before is inside Index. */
817 		if (off >= used || off < le32_to_cpu(hdr->de_off) ||
818 		    off + le16_to_cpu(before->size) > total) {
819 			return NULL;
820 		}
821 		goto ok;
822 	}
823 	/* No insert point is applied. Get it manually. */
824 	before = hdr_find_e(indx, hdr, de + 1, le16_to_cpu(de->key_size), ctx,
825 			    &diff);
826 	if (!before)
827 		return NULL;
828 	off = PtrOffset(hdr, before);
829 
830 ok:
831 	/* Now we just make room for the entry and jam it in. */
832 	memmove(Add2Ptr(before, de_size), before, used - off);
833 
834 	hdr->used = cpu_to_le32(used + de_size);
835 	memcpy(before, de, de_size);
836 
837 	return before;
838 }
839 
840 /*
841  * hdr_delete_de - Remove an entry from the index buffer.
842  */
843 static inline struct NTFS_DE *hdr_delete_de(struct INDEX_HDR *hdr,
844 					    struct NTFS_DE *re)
845 {
846 	u32 used = le32_to_cpu(hdr->used);
847 	u16 esize = le16_to_cpu(re->size);
848 	u32 off = PtrOffset(hdr, re);
849 	int bytes = used - (off + esize);
850 
851 	/* check INDEX_HDR valid before using INDEX_HDR */
852 	if (!check_index_header(hdr, le32_to_cpu(hdr->total)))
853 		return NULL;
854 
855 	if (off >= used || esize < sizeof(struct NTFS_DE) ||
856 	    bytes < sizeof(struct NTFS_DE))
857 		return NULL;
858 
859 	hdr->used = cpu_to_le32(used - esize);
860 	memmove(re, Add2Ptr(re, esize), bytes);
861 
862 	return re;
863 }
864 
865 void indx_clear(struct ntfs_index *indx)
866 {
867 	run_close(&indx->alloc_run);
868 	run_close(&indx->bitmap_run);
869 }
870 
871 int indx_init(struct ntfs_index *indx, struct ntfs_sb_info *sbi,
872 	      const struct ATTRIB *attr, enum index_mutex_classed type)
873 {
874 	u32 t32;
875 	const struct INDEX_ROOT *root = resident_data(attr);
876 
877 	t32 = le32_to_cpu(attr->res.data_size);
878 	if (t32 <= offsetof(struct INDEX_ROOT, ihdr) ||
879 	    !index_hdr_check(&root->ihdr,
880 			     t32 - offsetof(struct INDEX_ROOT, ihdr))) {
881 		goto out;
882 	}
883 
884 	/* Check root fields. */
885 	if (!root->index_block_clst)
886 		goto out;
887 
888 	indx->type = type;
889 	indx->idx2vbn_bits = __ffs(root->index_block_clst);
890 
891 	t32 = le32_to_cpu(root->index_block_size);
892 	indx->index_bits = blksize_bits(t32);
893 
894 	/* Check index record size. */
895 	if (t32 < sbi->cluster_size) {
896 		/* Index record is smaller than a cluster, use 512 blocks. */
897 		if (t32 != root->index_block_clst * SECTOR_SIZE)
898 			goto out;
899 
900 		/* Check alignment to a cluster. */
901 		if ((sbi->cluster_size >> SECTOR_SHIFT) &
902 		    (root->index_block_clst - 1)) {
903 			goto out;
904 		}
905 
906 		indx->vbn2vbo_bits = SECTOR_SHIFT;
907 	} else {
908 		/* Index record must be a multiple of cluster size. */
909 		if (t32 != root->index_block_clst << sbi->cluster_bits)
910 			goto out;
911 
912 		indx->vbn2vbo_bits = sbi->cluster_bits;
913 	}
914 
915 	init_rwsem(&indx->run_lock);
916 
917 	indx->cmp = get_cmp_func(root);
918 	if (!indx->cmp)
919 		goto out;
920 
921 	return 0;
922 
923 out:
924 	ntfs_set_state(sbi, NTFS_DIRTY_DIRTY);
925 	return -EINVAL;
926 }
927 
928 static struct indx_node *indx_new(struct ntfs_index *indx,
929 				  struct ntfs_inode *ni, CLST vbn,
930 				  const __le64 *sub_vbn)
931 {
932 	int err;
933 	struct NTFS_DE *e;
934 	struct indx_node *r;
935 	struct INDEX_HDR *hdr;
936 	struct INDEX_BUFFER *index;
937 	u64 vbo = (u64)vbn << indx->vbn2vbo_bits;
938 	u32 bytes = 1u << indx->index_bits;
939 	u16 fn;
940 	u32 eo;
941 
942 	r = kzalloc(sizeof(struct indx_node), GFP_NOFS);
943 	if (!r)
944 		return ERR_PTR(-ENOMEM);
945 
946 	index = kzalloc(bytes, GFP_NOFS);
947 	if (!index) {
948 		kfree(r);
949 		return ERR_PTR(-ENOMEM);
950 	}
951 
952 	err = ntfs_get_bh(ni->mi.sbi, &indx->alloc_run, vbo, bytes, &r->nb);
953 
954 	if (err) {
955 		kfree(index);
956 		kfree(r);
957 		return ERR_PTR(err);
958 	}
959 
960 	/* Create header. */
961 	index->rhdr.sign = NTFS_INDX_SIGNATURE;
962 	index->rhdr.fix_off = cpu_to_le16(sizeof(struct INDEX_BUFFER)); // 0x28
963 	fn = (bytes >> SECTOR_SHIFT) + 1; // 9
964 	index->rhdr.fix_num = cpu_to_le16(fn);
965 	index->vbn = cpu_to_le64(vbn);
966 	hdr = &index->ihdr;
967 	eo = ALIGN(sizeof(struct INDEX_BUFFER) + fn * sizeof(short), 8);
968 	hdr->de_off = cpu_to_le32(eo);
969 
970 	e = Add2Ptr(hdr, eo);
971 
972 	if (sub_vbn) {
973 		e->flags = NTFS_IE_LAST | NTFS_IE_HAS_SUBNODES;
974 		e->size = cpu_to_le16(sizeof(struct NTFS_DE) + sizeof(u64));
975 		hdr->used =
976 			cpu_to_le32(eo + sizeof(struct NTFS_DE) + sizeof(u64));
977 		de_set_vbn_le(e, *sub_vbn);
978 		hdr->flags = 1;
979 	} else {
980 		e->size = cpu_to_le16(sizeof(struct NTFS_DE));
981 		hdr->used = cpu_to_le32(eo + sizeof(struct NTFS_DE));
982 		e->flags = NTFS_IE_LAST;
983 	}
984 
985 	hdr->total = cpu_to_le32(bytes - offsetof(struct INDEX_BUFFER, ihdr));
986 
987 	r->index = index;
988 	return r;
989 }
990 
991 struct INDEX_ROOT *indx_get_root(struct ntfs_index *indx, struct ntfs_inode *ni,
992 				 struct ATTRIB **attr, struct mft_inode **mi)
993 {
994 	struct ATTR_LIST_ENTRY *le = NULL;
995 	struct ATTRIB *a;
996 	const struct INDEX_NAMES *in = &s_index_names[indx->type];
997 	struct INDEX_ROOT *root;
998 
999 	a = ni_find_attr(ni, NULL, &le, ATTR_ROOT, in->name, in->name_len, NULL,
1000 			 mi);
1001 	if (!a)
1002 		return NULL;
1003 
1004 	if (attr)
1005 		*attr = a;
1006 
1007 	root = resident_data_ex(a, sizeof(struct INDEX_ROOT));
1008 
1009 	/* length check */
1010 	if (root &&
1011 	    offsetof(struct INDEX_ROOT, ihdr) + le32_to_cpu(root->ihdr.used) >
1012 		    le32_to_cpu(a->res.data_size)) {
1013 		return NULL;
1014 	}
1015 
1016 	return root;
1017 }
1018 
1019 static int indx_write(struct ntfs_index *indx, struct ntfs_inode *ni,
1020 		      struct indx_node *node, int sync)
1021 {
1022 	struct INDEX_BUFFER *ib = node->index;
1023 
1024 	return ntfs_write_bh(ni->mi.sbi, &ib->rhdr, &node->nb, sync);
1025 }
1026 
1027 /*
1028  * indx_read
1029  *
1030  * If ntfs_readdir calls this function
1031  * inode is shared locked and no ni_lock.
1032  * Use rw_semaphore for read/write access to alloc_run.
1033  */
1034 int indx_read(struct ntfs_index *indx, struct ntfs_inode *ni, CLST vbn,
1035 	      struct indx_node **node)
1036 {
1037 	int err;
1038 	struct INDEX_BUFFER *ib;
1039 	struct runs_tree *run = &indx->alloc_run;
1040 	struct rw_semaphore *lock = &indx->run_lock;
1041 	u64 vbo = (u64)vbn << indx->vbn2vbo_bits;
1042 	u32 bytes = 1u << indx->index_bits;
1043 	struct indx_node *in = *node;
1044 	const struct INDEX_NAMES *name;
1045 
1046 	if (!in) {
1047 		in = kzalloc(sizeof(struct indx_node), GFP_NOFS);
1048 		if (!in)
1049 			return -ENOMEM;
1050 	} else {
1051 		nb_put(&in->nb);
1052 	}
1053 
1054 	ib = in->index;
1055 	if (!ib) {
1056 		ib = kmalloc(bytes, GFP_NOFS);
1057 		if (!ib) {
1058 			err = -ENOMEM;
1059 			goto out;
1060 		}
1061 	}
1062 
1063 	down_read(lock);
1064 	err = ntfs_read_bh(ni->mi.sbi, run, vbo, &ib->rhdr, bytes, &in->nb);
1065 	up_read(lock);
1066 	if (!err)
1067 		goto ok;
1068 
1069 	if (err == -E_NTFS_FIXUP)
1070 		goto ok;
1071 
1072 	if (err != -ENOENT)
1073 		goto out;
1074 
1075 	name = &s_index_names[indx->type];
1076 	down_write(lock);
1077 	err = attr_load_runs_range(ni, ATTR_ALLOC, name->name, name->name_len,
1078 				   run, vbo, vbo + bytes);
1079 	up_write(lock);
1080 	if (err)
1081 		goto out;
1082 
1083 	down_read(lock);
1084 	err = ntfs_read_bh(ni->mi.sbi, run, vbo, &ib->rhdr, bytes, &in->nb);
1085 	up_read(lock);
1086 	if (err == -E_NTFS_FIXUP)
1087 		goto ok;
1088 
1089 	if (err)
1090 		goto out;
1091 
1092 ok:
1093 	if (!index_buf_check(ib, bytes, &vbn)) {
1094 		ntfs_inode_err(&ni->vfs_inode, "directory corrupted");
1095 		ntfs_set_state(ni->mi.sbi, NTFS_DIRTY_ERROR);
1096 		err = -EINVAL;
1097 		goto out;
1098 	}
1099 
1100 	if (err == -E_NTFS_FIXUP) {
1101 		ntfs_write_bh(ni->mi.sbi, &ib->rhdr, &in->nb, 0);
1102 		err = 0;
1103 	}
1104 
1105 	/* check for index header length */
1106 	if (offsetof(struct INDEX_BUFFER, ihdr) + le32_to_cpu(ib->ihdr.used) >
1107 	    bytes) {
1108 		err = -EINVAL;
1109 		goto out;
1110 	}
1111 
1112 	in->index = ib;
1113 	*node = in;
1114 
1115 out:
1116 	if (ib != in->index)
1117 		kfree(ib);
1118 
1119 	if (*node != in) {
1120 		nb_put(&in->nb);
1121 		kfree(in);
1122 	}
1123 
1124 	return err;
1125 }
1126 
1127 /*
1128  * indx_find - Scan NTFS directory for given entry.
1129  */
1130 int indx_find(struct ntfs_index *indx, struct ntfs_inode *ni,
1131 	      const struct INDEX_ROOT *root, const void *key, size_t key_len,
1132 	      const void *ctx, int *diff, struct NTFS_DE **entry,
1133 	      struct ntfs_fnd *fnd)
1134 {
1135 	int err;
1136 	struct NTFS_DE *e;
1137 	struct indx_node *node;
1138 
1139 	if (!root)
1140 		root = indx_get_root(&ni->dir, ni, NULL, NULL);
1141 
1142 	if (!root) {
1143 		/* Should not happen. */
1144 		return -EINVAL;
1145 	}
1146 
1147 	/* Check cache. */
1148 	e = fnd->level ? fnd->de[fnd->level - 1] : fnd->root_de;
1149 	if (e && !de_is_last(e) &&
1150 	    !(*indx->cmp)(key, key_len, e + 1, le16_to_cpu(e->key_size), ctx)) {
1151 		*entry = e;
1152 		*diff = 0;
1153 		return 0;
1154 	}
1155 
1156 	/* Soft finder reset. */
1157 	fnd_clear(fnd);
1158 
1159 	/* Lookup entry that is <= to the search value. */
1160 	e = hdr_find_e(indx, &root->ihdr, key, key_len, ctx, diff);
1161 	if (!e)
1162 		return -EINVAL;
1163 
1164 	fnd->root_de = e;
1165 
1166 	for (;;) {
1167 		node = NULL;
1168 		if (*diff >= 0 || !de_has_vcn_ex(e))
1169 			break;
1170 
1171 		/* Read next level. */
1172 		err = indx_read(indx, ni, de_get_vbn(e), &node);
1173 		if (err) {
1174 			/* io error? */
1175 			return err;
1176 		}
1177 
1178 		/* Lookup entry that is <= to the search value. */
1179 		e = hdr_find_e(indx, &node->index->ihdr, key, key_len, ctx,
1180 			       diff);
1181 		if (!e) {
1182 			put_indx_node(node);
1183 			return -EINVAL;
1184 		}
1185 
1186 		fnd_push(fnd, node, e);
1187 	}
1188 
1189 	*entry = e;
1190 	return 0;
1191 }
1192 
1193 int indx_find_sort(struct ntfs_index *indx, struct ntfs_inode *ni,
1194 		   const struct INDEX_ROOT *root, struct NTFS_DE **entry,
1195 		   struct ntfs_fnd *fnd)
1196 {
1197 	int err;
1198 	struct indx_node *n = NULL;
1199 	struct NTFS_DE *e;
1200 	size_t iter = 0;
1201 	int level = fnd->level;
1202 
1203 	if (!*entry) {
1204 		/* Start find. */
1205 		e = hdr_first_de(&root->ihdr);
1206 		if (!e)
1207 			return 0;
1208 		fnd_clear(fnd);
1209 		fnd->root_de = e;
1210 	} else if (!level) {
1211 		if (de_is_last(fnd->root_de)) {
1212 			*entry = NULL;
1213 			return 0;
1214 		}
1215 
1216 		e = hdr_next_de(&root->ihdr, fnd->root_de);
1217 		if (!e)
1218 			return -EINVAL;
1219 		fnd->root_de = e;
1220 	} else {
1221 		n = fnd->nodes[level - 1];
1222 		e = fnd->de[level - 1];
1223 
1224 		if (de_is_last(e))
1225 			goto pop_level;
1226 
1227 		e = hdr_next_de(&n->index->ihdr, e);
1228 		if (!e)
1229 			return -EINVAL;
1230 
1231 		fnd->de[level - 1] = e;
1232 	}
1233 
1234 	/* Just to avoid tree cycle. */
1235 next_iter:
1236 	if (iter++ >= 1000)
1237 		return -EINVAL;
1238 
1239 	while (de_has_vcn_ex(e)) {
1240 		if (le16_to_cpu(e->size) <
1241 		    sizeof(struct NTFS_DE) + sizeof(u64)) {
1242 			if (n) {
1243 				fnd_pop(fnd);
1244 				kfree(n);
1245 			}
1246 			return -EINVAL;
1247 		}
1248 
1249 		/* Read next level. */
1250 		err = indx_read(indx, ni, de_get_vbn(e), &n);
1251 		if (err)
1252 			return err;
1253 
1254 		/* Try next level. */
1255 		e = hdr_first_de(&n->index->ihdr);
1256 		if (!e) {
1257 			kfree(n);
1258 			return -EINVAL;
1259 		}
1260 
1261 		fnd_push(fnd, n, e);
1262 	}
1263 
1264 	if (le16_to_cpu(e->size) > sizeof(struct NTFS_DE)) {
1265 		*entry = e;
1266 		return 0;
1267 	}
1268 
1269 pop_level:
1270 	for (;;) {
1271 		if (!de_is_last(e))
1272 			goto next_iter;
1273 
1274 		/* Pop one level. */
1275 		if (n) {
1276 			fnd_pop(fnd);
1277 			kfree(n);
1278 		}
1279 
1280 		level = fnd->level;
1281 
1282 		if (level) {
1283 			n = fnd->nodes[level - 1];
1284 			e = fnd->de[level - 1];
1285 		} else if (fnd->root_de) {
1286 			n = NULL;
1287 			e = fnd->root_de;
1288 			fnd->root_de = NULL;
1289 		} else {
1290 			*entry = NULL;
1291 			return 0;
1292 		}
1293 
1294 		if (le16_to_cpu(e->size) > sizeof(struct NTFS_DE)) {
1295 			*entry = e;
1296 			if (!fnd->root_de)
1297 				fnd->root_de = e;
1298 			return 0;
1299 		}
1300 	}
1301 }
1302 
1303 int indx_find_raw(struct ntfs_index *indx, struct ntfs_inode *ni,
1304 		  const struct INDEX_ROOT *root, struct NTFS_DE **entry,
1305 		  size_t *off, struct ntfs_fnd *fnd)
1306 {
1307 	int err;
1308 	struct indx_node *n = NULL;
1309 	struct NTFS_DE *e = NULL;
1310 	struct NTFS_DE *e2;
1311 	size_t bit;
1312 	CLST next_used_vbn;
1313 	CLST next_vbn;
1314 	u32 record_size = ni->mi.sbi->record_size;
1315 
1316 	/* Use non sorted algorithm. */
1317 	if (!*entry) {
1318 		/* This is the first call. */
1319 		e = hdr_first_de(&root->ihdr);
1320 		if (!e)
1321 			return 0;
1322 		fnd_clear(fnd);
1323 		fnd->root_de = e;
1324 
1325 		/* The first call with setup of initial element. */
1326 		if (*off >= record_size) {
1327 			next_vbn = (((*off - record_size) >> indx->index_bits))
1328 				   << indx->idx2vbn_bits;
1329 			/* Jump inside cycle 'for'. */
1330 			goto next;
1331 		}
1332 
1333 		/* Start enumeration from root. */
1334 		*off = 0;
1335 	} else if (!fnd->root_de)
1336 		return -EINVAL;
1337 
1338 	for (;;) {
1339 		/* Check if current entry can be used. */
1340 		if (e && le16_to_cpu(e->size) > sizeof(struct NTFS_DE))
1341 			goto ok;
1342 
1343 		if (!fnd->level) {
1344 			/* Continue to enumerate root. */
1345 			if (!de_is_last(fnd->root_de)) {
1346 				e = hdr_next_de(&root->ihdr, fnd->root_de);
1347 				if (!e)
1348 					return -EINVAL;
1349 				fnd->root_de = e;
1350 				continue;
1351 			}
1352 
1353 			/* Start to enumerate indexes from 0. */
1354 			next_vbn = 0;
1355 		} else {
1356 			/* Continue to enumerate indexes. */
1357 			e2 = fnd->de[fnd->level - 1];
1358 
1359 			n = fnd->nodes[fnd->level - 1];
1360 
1361 			if (!de_is_last(e2)) {
1362 				e = hdr_next_de(&n->index->ihdr, e2);
1363 				if (!e)
1364 					return -EINVAL;
1365 				fnd->de[fnd->level - 1] = e;
1366 				continue;
1367 			}
1368 
1369 			/* Continue with next index. */
1370 			next_vbn = le64_to_cpu(n->index->vbn) +
1371 				   root->index_block_clst;
1372 		}
1373 
1374 next:
1375 		/* Release current index. */
1376 		if (n) {
1377 			fnd_pop(fnd);
1378 			put_indx_node(n);
1379 			n = NULL;
1380 		}
1381 
1382 		/* Skip all free indexes. */
1383 		bit = next_vbn >> indx->idx2vbn_bits;
1384 		err = indx_used_bit(indx, ni, &bit);
1385 		if (err == -ENOENT || bit == MINUS_ONE_T) {
1386 			/* No used indexes. */
1387 			*entry = NULL;
1388 			return 0;
1389 		}
1390 
1391 		next_used_vbn = bit << indx->idx2vbn_bits;
1392 
1393 		/* Read buffer into memory. */
1394 		err = indx_read(indx, ni, next_used_vbn, &n);
1395 		if (err)
1396 			return err;
1397 
1398 		e = hdr_first_de(&n->index->ihdr);
1399 		fnd_push(fnd, n, e);
1400 		if (!e)
1401 			return -EINVAL;
1402 	}
1403 
1404 ok:
1405 	/* Return offset to restore enumerator if necessary. */
1406 	if (!n) {
1407 		/* 'e' points in root, */
1408 		*off = PtrOffset(&root->ihdr, e);
1409 	} else {
1410 		/* 'e' points in index, */
1411 		*off = (le64_to_cpu(n->index->vbn) << indx->vbn2vbo_bits) +
1412 		       record_size + PtrOffset(&n->index->ihdr, e);
1413 	}
1414 
1415 	*entry = e;
1416 	return 0;
1417 }
1418 
1419 /*
1420  * indx_create_allocate - Create "Allocation + Bitmap" attributes.
1421  */
1422 static int indx_create_allocate(struct ntfs_index *indx, struct ntfs_inode *ni,
1423 				CLST *vbn)
1424 {
1425 	int err;
1426 	struct ntfs_sb_info *sbi = ni->mi.sbi;
1427 	struct ATTRIB *bitmap;
1428 	struct ATTRIB *alloc;
1429 	u32 data_size = 1u << indx->index_bits;
1430 	u32 alloc_size = ntfs_up_cluster(sbi, data_size);
1431 	CLST len = alloc_size >> sbi->cluster_bits;
1432 	const struct INDEX_NAMES *in = &s_index_names[indx->type];
1433 	CLST alen;
1434 	struct runs_tree run;
1435 
1436 	run_init(&run);
1437 
1438 	err = attr_allocate_clusters(sbi, &run, 0, 0, len, NULL, ALLOCATE_DEF,
1439 				     &alen, 0, NULL, NULL);
1440 	if (err)
1441 		goto out;
1442 
1443 	err = ni_insert_nonresident(ni, ATTR_ALLOC, in->name, in->name_len,
1444 				    &run, 0, len, 0, &alloc, NULL, NULL);
1445 	if (err)
1446 		goto out1;
1447 
1448 	alloc->nres.valid_size = alloc->nres.data_size = cpu_to_le64(data_size);
1449 
1450 	err = ni_insert_resident(ni, bitmap_size(1), ATTR_BITMAP, in->name,
1451 				 in->name_len, &bitmap, NULL, NULL);
1452 	if (err)
1453 		goto out2;
1454 
1455 	if (in->name == I30_NAME) {
1456 		ni->vfs_inode.i_size = data_size;
1457 		inode_set_bytes(&ni->vfs_inode, alloc_size);
1458 	}
1459 
1460 	memcpy(&indx->alloc_run, &run, sizeof(run));
1461 
1462 	*vbn = 0;
1463 
1464 	return 0;
1465 
1466 out2:
1467 	mi_remove_attr(NULL, &ni->mi, alloc);
1468 
1469 out1:
1470 	run_deallocate(sbi, &run, false);
1471 
1472 out:
1473 	return err;
1474 }
1475 
1476 /*
1477  * indx_add_allocate - Add clusters to index.
1478  */
1479 static int indx_add_allocate(struct ntfs_index *indx, struct ntfs_inode *ni,
1480 			     CLST *vbn)
1481 {
1482 	int err;
1483 	size_t bit;
1484 	u64 data_size;
1485 	u64 bmp_size, bmp_size_v;
1486 	struct ATTRIB *bmp, *alloc;
1487 	struct mft_inode *mi;
1488 	const struct INDEX_NAMES *in = &s_index_names[indx->type];
1489 
1490 	err = indx_find_free(indx, ni, &bit, &bmp);
1491 	if (err)
1492 		goto out1;
1493 
1494 	if (bit != MINUS_ONE_T) {
1495 		bmp = NULL;
1496 	} else {
1497 		if (bmp->non_res) {
1498 			bmp_size = le64_to_cpu(bmp->nres.data_size);
1499 			bmp_size_v = le64_to_cpu(bmp->nres.valid_size);
1500 		} else {
1501 			bmp_size = bmp_size_v = le32_to_cpu(bmp->res.data_size);
1502 		}
1503 
1504 		bit = bmp_size << 3;
1505 	}
1506 
1507 	data_size = (u64)(bit + 1) << indx->index_bits;
1508 
1509 	if (bmp) {
1510 		/* Increase bitmap. */
1511 		err = attr_set_size(ni, ATTR_BITMAP, in->name, in->name_len,
1512 				    &indx->bitmap_run, bitmap_size(bit + 1),
1513 				    NULL, true, NULL);
1514 		if (err)
1515 			goto out1;
1516 	}
1517 
1518 	alloc = ni_find_attr(ni, NULL, NULL, ATTR_ALLOC, in->name, in->name_len,
1519 			     NULL, &mi);
1520 	if (!alloc) {
1521 		err = -EINVAL;
1522 		if (bmp)
1523 			goto out2;
1524 		goto out1;
1525 	}
1526 
1527 	/* Increase allocation. */
1528 	err = attr_set_size(ni, ATTR_ALLOC, in->name, in->name_len,
1529 			    &indx->alloc_run, data_size, &data_size, true,
1530 			    NULL);
1531 	if (err) {
1532 		if (bmp)
1533 			goto out2;
1534 		goto out1;
1535 	}
1536 
1537 	if (in->name == I30_NAME)
1538 		ni->vfs_inode.i_size = data_size;
1539 
1540 	*vbn = bit << indx->idx2vbn_bits;
1541 
1542 	return 0;
1543 
1544 out2:
1545 	/* Ops. No space? */
1546 	attr_set_size(ni, ATTR_BITMAP, in->name, in->name_len,
1547 		      &indx->bitmap_run, bmp_size, &bmp_size_v, false, NULL);
1548 
1549 out1:
1550 	return err;
1551 }
1552 
1553 /*
1554  * indx_insert_into_root - Attempt to insert an entry into the index root.
1555  *
1556  * @undo - True if we undoing previous remove.
1557  * If necessary, it will twiddle the index b-tree.
1558  */
1559 static int indx_insert_into_root(struct ntfs_index *indx, struct ntfs_inode *ni,
1560 				 const struct NTFS_DE *new_de,
1561 				 struct NTFS_DE *root_de, const void *ctx,
1562 				 struct ntfs_fnd *fnd, bool undo)
1563 {
1564 	int err = 0;
1565 	struct NTFS_DE *e, *e0, *re;
1566 	struct mft_inode *mi;
1567 	struct ATTRIB *attr;
1568 	struct INDEX_HDR *hdr;
1569 	struct indx_node *n;
1570 	CLST new_vbn;
1571 	__le64 *sub_vbn, t_vbn;
1572 	u16 new_de_size;
1573 	u32 hdr_used, hdr_total, asize, to_move;
1574 	u32 root_size, new_root_size;
1575 	struct ntfs_sb_info *sbi;
1576 	int ds_root;
1577 	struct INDEX_ROOT *root, *a_root;
1578 
1579 	/* Get the record this root placed in. */
1580 	root = indx_get_root(indx, ni, &attr, &mi);
1581 	if (!root)
1582 		return -EINVAL;
1583 
1584 	/*
1585 	 * Try easy case:
1586 	 * hdr_insert_de will succeed if there's
1587 	 * room the root for the new entry.
1588 	 */
1589 	hdr = &root->ihdr;
1590 	sbi = ni->mi.sbi;
1591 	new_de_size = le16_to_cpu(new_de->size);
1592 	hdr_used = le32_to_cpu(hdr->used);
1593 	hdr_total = le32_to_cpu(hdr->total);
1594 	asize = le32_to_cpu(attr->size);
1595 	root_size = le32_to_cpu(attr->res.data_size);
1596 
1597 	ds_root = new_de_size + hdr_used - hdr_total;
1598 
1599 	/* If 'undo' is set then reduce requirements. */
1600 	if ((undo || asize + ds_root < sbi->max_bytes_per_attr) &&
1601 	    mi_resize_attr(mi, attr, ds_root)) {
1602 		hdr->total = cpu_to_le32(hdr_total + ds_root);
1603 		e = hdr_insert_de(indx, hdr, new_de, root_de, ctx);
1604 		WARN_ON(!e);
1605 		fnd_clear(fnd);
1606 		fnd->root_de = e;
1607 
1608 		return 0;
1609 	}
1610 
1611 	/* Make a copy of root attribute to restore if error. */
1612 	a_root = kmemdup(attr, asize, GFP_NOFS);
1613 	if (!a_root)
1614 		return -ENOMEM;
1615 
1616 	/*
1617 	 * Copy all the non-end entries from
1618 	 * the index root to the new buffer.
1619 	 */
1620 	to_move = 0;
1621 	e0 = hdr_first_de(hdr);
1622 
1623 	/* Calculate the size to copy. */
1624 	for (e = e0;; e = hdr_next_de(hdr, e)) {
1625 		if (!e) {
1626 			err = -EINVAL;
1627 			goto out_free_root;
1628 		}
1629 
1630 		if (de_is_last(e))
1631 			break;
1632 		to_move += le16_to_cpu(e->size);
1633 	}
1634 
1635 	if (!to_move) {
1636 		re = NULL;
1637 	} else {
1638 		re = kmemdup(e0, to_move, GFP_NOFS);
1639 		if (!re) {
1640 			err = -ENOMEM;
1641 			goto out_free_root;
1642 		}
1643 	}
1644 
1645 	sub_vbn = NULL;
1646 	if (de_has_vcn(e)) {
1647 		t_vbn = de_get_vbn_le(e);
1648 		sub_vbn = &t_vbn;
1649 	}
1650 
1651 	new_root_size = sizeof(struct INDEX_ROOT) + sizeof(struct NTFS_DE) +
1652 			sizeof(u64);
1653 	ds_root = new_root_size - root_size;
1654 
1655 	if (ds_root > 0 && asize + ds_root > sbi->max_bytes_per_attr) {
1656 		/* Make root external. */
1657 		err = -EOPNOTSUPP;
1658 		goto out_free_re;
1659 	}
1660 
1661 	if (ds_root)
1662 		mi_resize_attr(mi, attr, ds_root);
1663 
1664 	/* Fill first entry (vcn will be set later). */
1665 	e = (struct NTFS_DE *)(root + 1);
1666 	memset(e, 0, sizeof(struct NTFS_DE));
1667 	e->size = cpu_to_le16(sizeof(struct NTFS_DE) + sizeof(u64));
1668 	e->flags = NTFS_IE_HAS_SUBNODES | NTFS_IE_LAST;
1669 
1670 	hdr->flags = 1;
1671 	hdr->used = hdr->total =
1672 		cpu_to_le32(new_root_size - offsetof(struct INDEX_ROOT, ihdr));
1673 
1674 	fnd->root_de = hdr_first_de(hdr);
1675 	mi->dirty = true;
1676 
1677 	/* Create alloc and bitmap attributes (if not). */
1678 	err = run_is_empty(&indx->alloc_run) ?
1679 			    indx_create_allocate(indx, ni, &new_vbn) :
1680 			    indx_add_allocate(indx, ni, &new_vbn);
1681 
1682 	/* Layout of record may be changed, so rescan root. */
1683 	root = indx_get_root(indx, ni, &attr, &mi);
1684 	if (!root) {
1685 		/* Bug? */
1686 		ntfs_set_state(sbi, NTFS_DIRTY_ERROR);
1687 		err = -EINVAL;
1688 		goto out_free_re;
1689 	}
1690 
1691 	if (err) {
1692 		/* Restore root. */
1693 		if (mi_resize_attr(mi, attr, -ds_root)) {
1694 			memcpy(attr, a_root, asize);
1695 		} else {
1696 			/* Bug? */
1697 			ntfs_set_state(sbi, NTFS_DIRTY_ERROR);
1698 		}
1699 		goto out_free_re;
1700 	}
1701 
1702 	e = (struct NTFS_DE *)(root + 1);
1703 	*(__le64 *)(e + 1) = cpu_to_le64(new_vbn);
1704 	mi->dirty = true;
1705 
1706 	/* Now we can create/format the new buffer and copy the entries into. */
1707 	n = indx_new(indx, ni, new_vbn, sub_vbn);
1708 	if (IS_ERR(n)) {
1709 		err = PTR_ERR(n);
1710 		goto out_free_re;
1711 	}
1712 
1713 	hdr = &n->index->ihdr;
1714 	hdr_used = le32_to_cpu(hdr->used);
1715 	hdr_total = le32_to_cpu(hdr->total);
1716 
1717 	/* Copy root entries into new buffer. */
1718 	hdr_insert_head(hdr, re, to_move);
1719 
1720 	/* Update bitmap attribute. */
1721 	indx_mark_used(indx, ni, new_vbn >> indx->idx2vbn_bits);
1722 
1723 	/* Check if we can insert new entry new index buffer. */
1724 	if (hdr_used + new_de_size > hdr_total) {
1725 		/*
1726 		 * This occurs if MFT record is the same or bigger than index
1727 		 * buffer. Move all root new index and have no space to add
1728 		 * new entry classic case when MFT record is 1K and index
1729 		 * buffer 4K the problem should not occurs.
1730 		 */
1731 		kfree(re);
1732 		indx_write(indx, ni, n, 0);
1733 
1734 		put_indx_node(n);
1735 		fnd_clear(fnd);
1736 		err = indx_insert_entry(indx, ni, new_de, ctx, fnd, undo);
1737 		goto out_free_root;
1738 	}
1739 
1740 	/*
1741 	 * Now root is a parent for new index buffer.
1742 	 * Insert NewEntry a new buffer.
1743 	 */
1744 	e = hdr_insert_de(indx, hdr, new_de, NULL, ctx);
1745 	if (!e) {
1746 		err = -EINVAL;
1747 		goto out_put_n;
1748 	}
1749 	fnd_push(fnd, n, e);
1750 
1751 	/* Just write updates index into disk. */
1752 	indx_write(indx, ni, n, 0);
1753 
1754 	n = NULL;
1755 
1756 out_put_n:
1757 	put_indx_node(n);
1758 out_free_re:
1759 	kfree(re);
1760 out_free_root:
1761 	kfree(a_root);
1762 	return err;
1763 }
1764 
1765 /*
1766  * indx_insert_into_buffer
1767  *
1768  * Attempt to insert an entry into an Index Allocation Buffer.
1769  * If necessary, it will split the buffer.
1770  */
1771 static int
1772 indx_insert_into_buffer(struct ntfs_index *indx, struct ntfs_inode *ni,
1773 			struct INDEX_ROOT *root, const struct NTFS_DE *new_de,
1774 			const void *ctx, int level, struct ntfs_fnd *fnd)
1775 {
1776 	int err;
1777 	const struct NTFS_DE *sp;
1778 	struct NTFS_DE *e, *de_t, *up_e;
1779 	struct indx_node *n2;
1780 	struct indx_node *n1 = fnd->nodes[level];
1781 	struct INDEX_HDR *hdr1 = &n1->index->ihdr;
1782 	struct INDEX_HDR *hdr2;
1783 	u32 to_copy, used, used1;
1784 	CLST new_vbn;
1785 	__le64 t_vbn, *sub_vbn;
1786 	u16 sp_size;
1787 	void *hdr1_saved = NULL;
1788 
1789 	/* Try the most easy case. */
1790 	e = fnd->level - 1 == level ? fnd->de[level] : NULL;
1791 	e = hdr_insert_de(indx, hdr1, new_de, e, ctx);
1792 	fnd->de[level] = e;
1793 	if (e) {
1794 		/* Just write updated index into disk. */
1795 		indx_write(indx, ni, n1, 0);
1796 		return 0;
1797 	}
1798 
1799 	/*
1800 	 * No space to insert into buffer. Split it.
1801 	 * To split we:
1802 	 *  - Save split point ('cause index buffers will be changed)
1803 	 * - Allocate NewBuffer and copy all entries <= sp into new buffer
1804 	 * - Remove all entries (sp including) from TargetBuffer
1805 	 * - Insert NewEntry into left or right buffer (depending on sp <=>
1806 	 *     NewEntry)
1807 	 * - Insert sp into parent buffer (or root)
1808 	 * - Make sp a parent for new buffer
1809 	 */
1810 	sp = hdr_find_split(hdr1);
1811 	if (!sp)
1812 		return -EINVAL;
1813 
1814 	sp_size = le16_to_cpu(sp->size);
1815 	up_e = kmalloc(sp_size + sizeof(u64), GFP_NOFS);
1816 	if (!up_e)
1817 		return -ENOMEM;
1818 	memcpy(up_e, sp, sp_size);
1819 
1820 	used1 = le32_to_cpu(hdr1->used);
1821 	hdr1_saved = kmemdup(hdr1, used1, GFP_NOFS);
1822 	if (!hdr1_saved) {
1823 		err = -ENOMEM;
1824 		goto out;
1825 	}
1826 
1827 	if (!hdr1->flags) {
1828 		up_e->flags |= NTFS_IE_HAS_SUBNODES;
1829 		up_e->size = cpu_to_le16(sp_size + sizeof(u64));
1830 		sub_vbn = NULL;
1831 	} else {
1832 		t_vbn = de_get_vbn_le(up_e);
1833 		sub_vbn = &t_vbn;
1834 	}
1835 
1836 	/* Allocate on disk a new index allocation buffer. */
1837 	err = indx_add_allocate(indx, ni, &new_vbn);
1838 	if (err)
1839 		goto out;
1840 
1841 	/* Allocate and format memory a new index buffer. */
1842 	n2 = indx_new(indx, ni, new_vbn, sub_vbn);
1843 	if (IS_ERR(n2)) {
1844 		err = PTR_ERR(n2);
1845 		goto out;
1846 	}
1847 
1848 	hdr2 = &n2->index->ihdr;
1849 
1850 	/* Make sp a parent for new buffer. */
1851 	de_set_vbn(up_e, new_vbn);
1852 
1853 	/* Copy all the entries <= sp into the new buffer. */
1854 	de_t = hdr_first_de(hdr1);
1855 	to_copy = PtrOffset(de_t, sp);
1856 	hdr_insert_head(hdr2, de_t, to_copy);
1857 
1858 	/* Remove all entries (sp including) from hdr1. */
1859 	used = used1 - to_copy - sp_size;
1860 	memmove(de_t, Add2Ptr(sp, sp_size), used - le32_to_cpu(hdr1->de_off));
1861 	hdr1->used = cpu_to_le32(used);
1862 
1863 	/*
1864 	 * Insert new entry into left or right buffer
1865 	 * (depending on sp <=> new_de).
1866 	 */
1867 	hdr_insert_de(indx,
1868 		      (*indx->cmp)(new_de + 1, le16_to_cpu(new_de->key_size),
1869 				   up_e + 1, le16_to_cpu(up_e->key_size),
1870 				   ctx) < 0 ?
1871 				    hdr2 :
1872 				    hdr1,
1873 		      new_de, NULL, ctx);
1874 
1875 	indx_mark_used(indx, ni, new_vbn >> indx->idx2vbn_bits);
1876 
1877 	indx_write(indx, ni, n1, 0);
1878 	indx_write(indx, ni, n2, 0);
1879 
1880 	put_indx_node(n2);
1881 
1882 	/*
1883 	 * We've finished splitting everybody, so we are ready to
1884 	 * insert the promoted entry into the parent.
1885 	 */
1886 	if (!level) {
1887 		/* Insert in root. */
1888 		err = indx_insert_into_root(indx, ni, up_e, NULL, ctx, fnd, 0);
1889 	} else {
1890 		/*
1891 		 * The target buffer's parent is another index buffer.
1892 		 * TODO: Remove recursion.
1893 		 */
1894 		err = indx_insert_into_buffer(indx, ni, root, up_e, ctx,
1895 					      level - 1, fnd);
1896 	}
1897 
1898 	if (err) {
1899 		/*
1900 		 * Undo critical operations.
1901 		 */
1902 		indx_mark_free(indx, ni, new_vbn >> indx->idx2vbn_bits);
1903 		memcpy(hdr1, hdr1_saved, used1);
1904 		indx_write(indx, ni, n1, 0);
1905 	}
1906 
1907 out:
1908 	kfree(up_e);
1909 	kfree(hdr1_saved);
1910 
1911 	return err;
1912 }
1913 
1914 /*
1915  * indx_insert_entry - Insert new entry into index.
1916  *
1917  * @undo - True if we undoing previous remove.
1918  */
1919 int indx_insert_entry(struct ntfs_index *indx, struct ntfs_inode *ni,
1920 		      const struct NTFS_DE *new_de, const void *ctx,
1921 		      struct ntfs_fnd *fnd, bool undo)
1922 {
1923 	int err;
1924 	int diff;
1925 	struct NTFS_DE *e;
1926 	struct ntfs_fnd *fnd_a = NULL;
1927 	struct INDEX_ROOT *root;
1928 
1929 	if (!fnd) {
1930 		fnd_a = fnd_get();
1931 		if (!fnd_a) {
1932 			err = -ENOMEM;
1933 			goto out1;
1934 		}
1935 		fnd = fnd_a;
1936 	}
1937 
1938 	root = indx_get_root(indx, ni, NULL, NULL);
1939 	if (!root) {
1940 		err = -EINVAL;
1941 		goto out;
1942 	}
1943 
1944 	if (fnd_is_empty(fnd)) {
1945 		/*
1946 		 * Find the spot the tree where we want to
1947 		 * insert the new entry.
1948 		 */
1949 		err = indx_find(indx, ni, root, new_de + 1,
1950 				le16_to_cpu(new_de->key_size), ctx, &diff, &e,
1951 				fnd);
1952 		if (err)
1953 			goto out;
1954 
1955 		if (!diff) {
1956 			err = -EEXIST;
1957 			goto out;
1958 		}
1959 	}
1960 
1961 	if (!fnd->level) {
1962 		/*
1963 		 * The root is also a leaf, so we'll insert the
1964 		 * new entry into it.
1965 		 */
1966 		err = indx_insert_into_root(indx, ni, new_de, fnd->root_de, ctx,
1967 					    fnd, undo);
1968 	} else {
1969 		/*
1970 		 * Found a leaf buffer, so we'll insert the new entry into it.
1971 		 */
1972 		err = indx_insert_into_buffer(indx, ni, root, new_de, ctx,
1973 					      fnd->level - 1, fnd);
1974 	}
1975 
1976 out:
1977 	fnd_put(fnd_a);
1978 out1:
1979 	return err;
1980 }
1981 
1982 /*
1983  * indx_find_buffer - Locate a buffer from the tree.
1984  */
1985 static struct indx_node *indx_find_buffer(struct ntfs_index *indx,
1986 					  struct ntfs_inode *ni,
1987 					  const struct INDEX_ROOT *root,
1988 					  __le64 vbn, struct indx_node *n)
1989 {
1990 	int err;
1991 	const struct NTFS_DE *e;
1992 	struct indx_node *r;
1993 	const struct INDEX_HDR *hdr = n ? &n->index->ihdr : &root->ihdr;
1994 
1995 	/* Step 1: Scan one level. */
1996 	for (e = hdr_first_de(hdr);; e = hdr_next_de(hdr, e)) {
1997 		if (!e)
1998 			return ERR_PTR(-EINVAL);
1999 
2000 		if (de_has_vcn(e) && vbn == de_get_vbn_le(e))
2001 			return n;
2002 
2003 		if (de_is_last(e))
2004 			break;
2005 	}
2006 
2007 	/* Step2: Do recursion. */
2008 	e = Add2Ptr(hdr, le32_to_cpu(hdr->de_off));
2009 	for (;;) {
2010 		if (de_has_vcn_ex(e)) {
2011 			err = indx_read(indx, ni, de_get_vbn(e), &n);
2012 			if (err)
2013 				return ERR_PTR(err);
2014 
2015 			r = indx_find_buffer(indx, ni, root, vbn, n);
2016 			if (r)
2017 				return r;
2018 		}
2019 
2020 		if (de_is_last(e))
2021 			break;
2022 
2023 		e = Add2Ptr(e, le16_to_cpu(e->size));
2024 	}
2025 
2026 	return NULL;
2027 }
2028 
2029 /*
2030  * indx_shrink - Deallocate unused tail indexes.
2031  */
2032 static int indx_shrink(struct ntfs_index *indx, struct ntfs_inode *ni,
2033 		       size_t bit)
2034 {
2035 	int err = 0;
2036 	u64 bpb, new_data;
2037 	size_t nbits;
2038 	struct ATTRIB *b;
2039 	struct ATTR_LIST_ENTRY *le = NULL;
2040 	const struct INDEX_NAMES *in = &s_index_names[indx->type];
2041 
2042 	b = ni_find_attr(ni, NULL, &le, ATTR_BITMAP, in->name, in->name_len,
2043 			 NULL, NULL);
2044 
2045 	if (!b)
2046 		return -ENOENT;
2047 
2048 	if (!b->non_res) {
2049 		unsigned long pos;
2050 		const unsigned long *bm = resident_data(b);
2051 
2052 		nbits = (size_t)le32_to_cpu(b->res.data_size) * 8;
2053 
2054 		if (bit >= nbits)
2055 			return 0;
2056 
2057 		pos = find_next_bit_le(bm, nbits, bit);
2058 		if (pos < nbits)
2059 			return 0;
2060 	} else {
2061 		size_t used = MINUS_ONE_T;
2062 
2063 		nbits = le64_to_cpu(b->nres.data_size) * 8;
2064 
2065 		if (bit >= nbits)
2066 			return 0;
2067 
2068 		err = scan_nres_bitmap(ni, b, indx, bit, &scan_for_used, &used);
2069 		if (err)
2070 			return err;
2071 
2072 		if (used != MINUS_ONE_T)
2073 			return 0;
2074 	}
2075 
2076 	new_data = (u64)bit << indx->index_bits;
2077 
2078 	err = attr_set_size(ni, ATTR_ALLOC, in->name, in->name_len,
2079 			    &indx->alloc_run, new_data, &new_data, false, NULL);
2080 	if (err)
2081 		return err;
2082 
2083 	if (in->name == I30_NAME)
2084 		ni->vfs_inode.i_size = new_data;
2085 
2086 	bpb = bitmap_size(bit);
2087 	if (bpb * 8 == nbits)
2088 		return 0;
2089 
2090 	err = attr_set_size(ni, ATTR_BITMAP, in->name, in->name_len,
2091 			    &indx->bitmap_run, bpb, &bpb, false, NULL);
2092 
2093 	return err;
2094 }
2095 
2096 static int indx_free_children(struct ntfs_index *indx, struct ntfs_inode *ni,
2097 			      const struct NTFS_DE *e, bool trim)
2098 {
2099 	int err;
2100 	struct indx_node *n = NULL;
2101 	struct INDEX_HDR *hdr;
2102 	CLST vbn = de_get_vbn(e);
2103 	size_t i;
2104 
2105 	err = indx_read(indx, ni, vbn, &n);
2106 	if (err)
2107 		return err;
2108 
2109 	hdr = &n->index->ihdr;
2110 	/* First, recurse into the children, if any. */
2111 	if (hdr_has_subnode(hdr)) {
2112 		for (e = hdr_first_de(hdr); e; e = hdr_next_de(hdr, e)) {
2113 			indx_free_children(indx, ni, e, false);
2114 			if (de_is_last(e))
2115 				break;
2116 		}
2117 	}
2118 
2119 	put_indx_node(n);
2120 
2121 	i = vbn >> indx->idx2vbn_bits;
2122 	/*
2123 	 * We've gotten rid of the children; add this buffer to the free list.
2124 	 */
2125 	indx_mark_free(indx, ni, i);
2126 
2127 	if (!trim)
2128 		return 0;
2129 
2130 	/*
2131 	 * If there are no used indexes after current free index
2132 	 * then we can truncate allocation and bitmap.
2133 	 * Use bitmap to estimate the case.
2134 	 */
2135 	indx_shrink(indx, ni, i + 1);
2136 	return 0;
2137 }
2138 
2139 /*
2140  * indx_get_entry_to_replace
2141  *
2142  * Find a replacement entry for a deleted entry.
2143  * Always returns a node entry:
2144  * NTFS_IE_HAS_SUBNODES is set the flags and the size includes the sub_vcn.
2145  */
2146 static int indx_get_entry_to_replace(struct ntfs_index *indx,
2147 				     struct ntfs_inode *ni,
2148 				     const struct NTFS_DE *de_next,
2149 				     struct NTFS_DE **de_to_replace,
2150 				     struct ntfs_fnd *fnd)
2151 {
2152 	int err;
2153 	int level = -1;
2154 	CLST vbn;
2155 	struct NTFS_DE *e, *te, *re;
2156 	struct indx_node *n;
2157 	struct INDEX_BUFFER *ib;
2158 
2159 	*de_to_replace = NULL;
2160 
2161 	/* Find first leaf entry down from de_next. */
2162 	vbn = de_get_vbn(de_next);
2163 	for (;;) {
2164 		n = NULL;
2165 		err = indx_read(indx, ni, vbn, &n);
2166 		if (err)
2167 			goto out;
2168 
2169 		e = hdr_first_de(&n->index->ihdr);
2170 		fnd_push(fnd, n, e);
2171 
2172 		if (!de_is_last(e)) {
2173 			/*
2174 			 * This buffer is non-empty, so its first entry
2175 			 * could be used as the replacement entry.
2176 			 */
2177 			level = fnd->level - 1;
2178 		}
2179 
2180 		if (!de_has_vcn(e))
2181 			break;
2182 
2183 		/* This buffer is a node. Continue to go down. */
2184 		vbn = de_get_vbn(e);
2185 	}
2186 
2187 	if (level == -1)
2188 		goto out;
2189 
2190 	n = fnd->nodes[level];
2191 	te = hdr_first_de(&n->index->ihdr);
2192 	/* Copy the candidate entry into the replacement entry buffer. */
2193 	re = kmalloc(le16_to_cpu(te->size) + sizeof(u64), GFP_NOFS);
2194 	if (!re) {
2195 		err = -ENOMEM;
2196 		goto out;
2197 	}
2198 
2199 	*de_to_replace = re;
2200 	memcpy(re, te, le16_to_cpu(te->size));
2201 
2202 	if (!de_has_vcn(re)) {
2203 		/*
2204 		 * The replacement entry we found doesn't have a sub_vcn.
2205 		 * increase its size to hold one.
2206 		 */
2207 		le16_add_cpu(&re->size, sizeof(u64));
2208 		re->flags |= NTFS_IE_HAS_SUBNODES;
2209 	} else {
2210 		/*
2211 		 * The replacement entry we found was a node entry, which
2212 		 * means that all its child buffers are empty. Return them
2213 		 * to the free pool.
2214 		 */
2215 		indx_free_children(indx, ni, te, true);
2216 	}
2217 
2218 	/*
2219 	 * Expunge the replacement entry from its former location,
2220 	 * and then write that buffer.
2221 	 */
2222 	ib = n->index;
2223 	e = hdr_delete_de(&ib->ihdr, te);
2224 
2225 	fnd->de[level] = e;
2226 	indx_write(indx, ni, n, 0);
2227 
2228 	if (ib_is_leaf(ib) && ib_is_empty(ib)) {
2229 		/* An empty leaf. */
2230 		return 0;
2231 	}
2232 
2233 out:
2234 	fnd_clear(fnd);
2235 	return err;
2236 }
2237 
2238 /*
2239  * indx_delete_entry - Delete an entry from the index.
2240  */
2241 int indx_delete_entry(struct ntfs_index *indx, struct ntfs_inode *ni,
2242 		      const void *key, u32 key_len, const void *ctx)
2243 {
2244 	int err, diff;
2245 	struct INDEX_ROOT *root;
2246 	struct INDEX_HDR *hdr;
2247 	struct ntfs_fnd *fnd, *fnd2;
2248 	struct INDEX_BUFFER *ib;
2249 	struct NTFS_DE *e, *re, *next, *prev, *me;
2250 	struct indx_node *n, *n2d = NULL;
2251 	__le64 sub_vbn;
2252 	int level, level2;
2253 	struct ATTRIB *attr;
2254 	struct mft_inode *mi;
2255 	u32 e_size, root_size, new_root_size;
2256 	size_t trim_bit;
2257 	const struct INDEX_NAMES *in;
2258 
2259 	fnd = fnd_get();
2260 	if (!fnd) {
2261 		err = -ENOMEM;
2262 		goto out2;
2263 	}
2264 
2265 	fnd2 = fnd_get();
2266 	if (!fnd2) {
2267 		err = -ENOMEM;
2268 		goto out1;
2269 	}
2270 
2271 	root = indx_get_root(indx, ni, &attr, &mi);
2272 	if (!root) {
2273 		err = -EINVAL;
2274 		goto out;
2275 	}
2276 
2277 	/* Locate the entry to remove. */
2278 	err = indx_find(indx, ni, root, key, key_len, ctx, &diff, &e, fnd);
2279 	if (err)
2280 		goto out;
2281 
2282 	if (!e || diff) {
2283 		err = -ENOENT;
2284 		goto out;
2285 	}
2286 
2287 	level = fnd->level;
2288 
2289 	if (level) {
2290 		n = fnd->nodes[level - 1];
2291 		e = fnd->de[level - 1];
2292 		ib = n->index;
2293 		hdr = &ib->ihdr;
2294 	} else {
2295 		hdr = &root->ihdr;
2296 		e = fnd->root_de;
2297 		n = NULL;
2298 	}
2299 
2300 	e_size = le16_to_cpu(e->size);
2301 
2302 	if (!de_has_vcn_ex(e)) {
2303 		/* The entry to delete is a leaf, so we can just rip it out. */
2304 		hdr_delete_de(hdr, e);
2305 
2306 		if (!level) {
2307 			hdr->total = hdr->used;
2308 
2309 			/* Shrink resident root attribute. */
2310 			mi_resize_attr(mi, attr, 0 - e_size);
2311 			goto out;
2312 		}
2313 
2314 		indx_write(indx, ni, n, 0);
2315 
2316 		/*
2317 		 * Check to see if removing that entry made
2318 		 * the leaf empty.
2319 		 */
2320 		if (ib_is_leaf(ib) && ib_is_empty(ib)) {
2321 			fnd_pop(fnd);
2322 			fnd_push(fnd2, n, e);
2323 		}
2324 	} else {
2325 		/*
2326 		 * The entry we wish to delete is a node buffer, so we
2327 		 * have to find a replacement for it.
2328 		 */
2329 		next = de_get_next(e);
2330 
2331 		err = indx_get_entry_to_replace(indx, ni, next, &re, fnd2);
2332 		if (err)
2333 			goto out;
2334 
2335 		if (re) {
2336 			de_set_vbn_le(re, de_get_vbn_le(e));
2337 			hdr_delete_de(hdr, e);
2338 
2339 			err = level ? indx_insert_into_buffer(indx, ni, root,
2340 							      re, ctx,
2341 							      fnd->level - 1,
2342 							      fnd) :
2343 					    indx_insert_into_root(indx, ni, re, e,
2344 							    ctx, fnd, 0);
2345 			kfree(re);
2346 
2347 			if (err)
2348 				goto out;
2349 		} else {
2350 			/*
2351 			 * There is no replacement for the current entry.
2352 			 * This means that the subtree rooted at its node
2353 			 * is empty, and can be deleted, which turn means
2354 			 * that the node can just inherit the deleted
2355 			 * entry sub_vcn.
2356 			 */
2357 			indx_free_children(indx, ni, next, true);
2358 
2359 			de_set_vbn_le(next, de_get_vbn_le(e));
2360 			hdr_delete_de(hdr, e);
2361 			if (level) {
2362 				indx_write(indx, ni, n, 0);
2363 			} else {
2364 				hdr->total = hdr->used;
2365 
2366 				/* Shrink resident root attribute. */
2367 				mi_resize_attr(mi, attr, 0 - e_size);
2368 			}
2369 		}
2370 	}
2371 
2372 	/* Delete a branch of tree. */
2373 	if (!fnd2 || !fnd2->level)
2374 		goto out;
2375 
2376 	/* Reinit root 'cause it can be changed. */
2377 	root = indx_get_root(indx, ni, &attr, &mi);
2378 	if (!root) {
2379 		err = -EINVAL;
2380 		goto out;
2381 	}
2382 
2383 	n2d = NULL;
2384 	sub_vbn = fnd2->nodes[0]->index->vbn;
2385 	level2 = 0;
2386 	level = fnd->level;
2387 
2388 	hdr = level ? &fnd->nodes[level - 1]->index->ihdr : &root->ihdr;
2389 
2390 	/* Scan current level. */
2391 	for (e = hdr_first_de(hdr);; e = hdr_next_de(hdr, e)) {
2392 		if (!e) {
2393 			err = -EINVAL;
2394 			goto out;
2395 		}
2396 
2397 		if (de_has_vcn(e) && sub_vbn == de_get_vbn_le(e))
2398 			break;
2399 
2400 		if (de_is_last(e)) {
2401 			e = NULL;
2402 			break;
2403 		}
2404 	}
2405 
2406 	if (!e) {
2407 		/* Do slow search from root. */
2408 		struct indx_node *in;
2409 
2410 		fnd_clear(fnd);
2411 
2412 		in = indx_find_buffer(indx, ni, root, sub_vbn, NULL);
2413 		if (IS_ERR(in)) {
2414 			err = PTR_ERR(in);
2415 			goto out;
2416 		}
2417 
2418 		if (in)
2419 			fnd_push(fnd, in, NULL);
2420 	}
2421 
2422 	/* Merge fnd2 -> fnd. */
2423 	for (level = 0; level < fnd2->level; level++) {
2424 		fnd_push(fnd, fnd2->nodes[level], fnd2->de[level]);
2425 		fnd2->nodes[level] = NULL;
2426 	}
2427 	fnd2->level = 0;
2428 
2429 	hdr = NULL;
2430 	for (level = fnd->level; level; level--) {
2431 		struct indx_node *in = fnd->nodes[level - 1];
2432 
2433 		ib = in->index;
2434 		if (ib_is_empty(ib)) {
2435 			sub_vbn = ib->vbn;
2436 		} else {
2437 			hdr = &ib->ihdr;
2438 			n2d = in;
2439 			level2 = level;
2440 			break;
2441 		}
2442 	}
2443 
2444 	if (!hdr)
2445 		hdr = &root->ihdr;
2446 
2447 	e = hdr_first_de(hdr);
2448 	if (!e) {
2449 		err = -EINVAL;
2450 		goto out;
2451 	}
2452 
2453 	if (hdr != &root->ihdr || !de_is_last(e)) {
2454 		prev = NULL;
2455 		while (!de_is_last(e)) {
2456 			if (de_has_vcn(e) && sub_vbn == de_get_vbn_le(e))
2457 				break;
2458 			prev = e;
2459 			e = hdr_next_de(hdr, e);
2460 			if (!e) {
2461 				err = -EINVAL;
2462 				goto out;
2463 			}
2464 		}
2465 
2466 		if (sub_vbn != de_get_vbn_le(e)) {
2467 			/*
2468 			 * Didn't find the parent entry, although this buffer
2469 			 * is the parent trail. Something is corrupt.
2470 			 */
2471 			err = -EINVAL;
2472 			goto out;
2473 		}
2474 
2475 		if (de_is_last(e)) {
2476 			/*
2477 			 * Since we can't remove the end entry, we'll remove
2478 			 * its predecessor instead. This means we have to
2479 			 * transfer the predecessor's sub_vcn to the end entry.
2480 			 * Note: This index block is not empty, so the
2481 			 * predecessor must exist.
2482 			 */
2483 			if (!prev) {
2484 				err = -EINVAL;
2485 				goto out;
2486 			}
2487 
2488 			if (de_has_vcn(prev)) {
2489 				de_set_vbn_le(e, de_get_vbn_le(prev));
2490 			} else if (de_has_vcn(e)) {
2491 				le16_sub_cpu(&e->size, sizeof(u64));
2492 				e->flags &= ~NTFS_IE_HAS_SUBNODES;
2493 				le32_sub_cpu(&hdr->used, sizeof(u64));
2494 			}
2495 			e = prev;
2496 		}
2497 
2498 		/*
2499 		 * Copy the current entry into a temporary buffer (stripping
2500 		 * off its down-pointer, if any) and delete it from the current
2501 		 * buffer or root, as appropriate.
2502 		 */
2503 		e_size = le16_to_cpu(e->size);
2504 		me = kmemdup(e, e_size, GFP_NOFS);
2505 		if (!me) {
2506 			err = -ENOMEM;
2507 			goto out;
2508 		}
2509 
2510 		if (de_has_vcn(me)) {
2511 			me->flags &= ~NTFS_IE_HAS_SUBNODES;
2512 			le16_sub_cpu(&me->size, sizeof(u64));
2513 		}
2514 
2515 		hdr_delete_de(hdr, e);
2516 
2517 		if (hdr == &root->ihdr) {
2518 			level = 0;
2519 			hdr->total = hdr->used;
2520 
2521 			/* Shrink resident root attribute. */
2522 			mi_resize_attr(mi, attr, 0 - e_size);
2523 		} else {
2524 			indx_write(indx, ni, n2d, 0);
2525 			level = level2;
2526 		}
2527 
2528 		/* Mark unused buffers as free. */
2529 		trim_bit = -1;
2530 		for (; level < fnd->level; level++) {
2531 			ib = fnd->nodes[level]->index;
2532 			if (ib_is_empty(ib)) {
2533 				size_t k = le64_to_cpu(ib->vbn) >>
2534 					   indx->idx2vbn_bits;
2535 
2536 				indx_mark_free(indx, ni, k);
2537 				if (k < trim_bit)
2538 					trim_bit = k;
2539 			}
2540 		}
2541 
2542 		fnd_clear(fnd);
2543 		/*fnd->root_de = NULL;*/
2544 
2545 		/*
2546 		 * Re-insert the entry into the tree.
2547 		 * Find the spot the tree where we want to insert the new entry.
2548 		 */
2549 		err = indx_insert_entry(indx, ni, me, ctx, fnd, 0);
2550 		kfree(me);
2551 		if (err)
2552 			goto out;
2553 
2554 		if (trim_bit != -1)
2555 			indx_shrink(indx, ni, trim_bit);
2556 	} else {
2557 		/*
2558 		 * This tree needs to be collapsed down to an empty root.
2559 		 * Recreate the index root as an empty leaf and free all
2560 		 * the bits the index allocation bitmap.
2561 		 */
2562 		fnd_clear(fnd);
2563 		fnd_clear(fnd2);
2564 
2565 		in = &s_index_names[indx->type];
2566 
2567 		err = attr_set_size(ni, ATTR_ALLOC, in->name, in->name_len,
2568 				    &indx->alloc_run, 0, NULL, false, NULL);
2569 		if (in->name == I30_NAME)
2570 			ni->vfs_inode.i_size = 0;
2571 
2572 		err = ni_remove_attr(ni, ATTR_ALLOC, in->name, in->name_len,
2573 				     false, NULL);
2574 		run_close(&indx->alloc_run);
2575 
2576 		err = attr_set_size(ni, ATTR_BITMAP, in->name, in->name_len,
2577 				    &indx->bitmap_run, 0, NULL, false, NULL);
2578 		err = ni_remove_attr(ni, ATTR_BITMAP, in->name, in->name_len,
2579 				     false, NULL);
2580 		run_close(&indx->bitmap_run);
2581 
2582 		root = indx_get_root(indx, ni, &attr, &mi);
2583 		if (!root) {
2584 			err = -EINVAL;
2585 			goto out;
2586 		}
2587 
2588 		root_size = le32_to_cpu(attr->res.data_size);
2589 		new_root_size =
2590 			sizeof(struct INDEX_ROOT) + sizeof(struct NTFS_DE);
2591 
2592 		if (new_root_size != root_size &&
2593 		    !mi_resize_attr(mi, attr, new_root_size - root_size)) {
2594 			err = -EINVAL;
2595 			goto out;
2596 		}
2597 
2598 		/* Fill first entry. */
2599 		e = (struct NTFS_DE *)(root + 1);
2600 		e->ref.low = 0;
2601 		e->ref.high = 0;
2602 		e->ref.seq = 0;
2603 		e->size = cpu_to_le16(sizeof(struct NTFS_DE));
2604 		e->flags = NTFS_IE_LAST; // 0x02
2605 		e->key_size = 0;
2606 		e->res = 0;
2607 
2608 		hdr = &root->ihdr;
2609 		hdr->flags = 0;
2610 		hdr->used = hdr->total = cpu_to_le32(
2611 			new_root_size - offsetof(struct INDEX_ROOT, ihdr));
2612 		mi->dirty = true;
2613 	}
2614 
2615 out:
2616 	fnd_put(fnd2);
2617 out1:
2618 	fnd_put(fnd);
2619 out2:
2620 	return err;
2621 }
2622 
2623 /*
2624  * Update duplicated information in directory entry
2625  * 'dup' - info from MFT record
2626  */
2627 int indx_update_dup(struct ntfs_inode *ni, struct ntfs_sb_info *sbi,
2628 		    const struct ATTR_FILE_NAME *fname,
2629 		    const struct NTFS_DUP_INFO *dup, int sync)
2630 {
2631 	int err, diff;
2632 	struct NTFS_DE *e = NULL;
2633 	struct ATTR_FILE_NAME *e_fname;
2634 	struct ntfs_fnd *fnd;
2635 	struct INDEX_ROOT *root;
2636 	struct mft_inode *mi;
2637 	struct ntfs_index *indx = &ni->dir;
2638 
2639 	fnd = fnd_get();
2640 	if (!fnd)
2641 		return -ENOMEM;
2642 
2643 	root = indx_get_root(indx, ni, NULL, &mi);
2644 	if (!root) {
2645 		err = -EINVAL;
2646 		goto out;
2647 	}
2648 
2649 	/* Find entry in directory. */
2650 	err = indx_find(indx, ni, root, fname, fname_full_size(fname), sbi,
2651 			&diff, &e, fnd);
2652 	if (err)
2653 		goto out;
2654 
2655 	if (!e) {
2656 		err = -EINVAL;
2657 		goto out;
2658 	}
2659 
2660 	if (diff) {
2661 		err = -EINVAL;
2662 		goto out;
2663 	}
2664 
2665 	e_fname = (struct ATTR_FILE_NAME *)(e + 1);
2666 
2667 	if (!memcmp(&e_fname->dup, dup, sizeof(*dup))) {
2668 		/*
2669 		 * Nothing to update in index! Try to avoid this call.
2670 		 */
2671 		goto out;
2672 	}
2673 
2674 	memcpy(&e_fname->dup, dup, sizeof(*dup));
2675 
2676 	if (fnd->level) {
2677 		/* Directory entry in index. */
2678 		err = indx_write(indx, ni, fnd->nodes[fnd->level - 1], sync);
2679 	} else {
2680 		/* Directory entry in directory MFT record. */
2681 		mi->dirty = true;
2682 		if (sync)
2683 			err = mi_write(mi, 1);
2684 		else
2685 			mark_inode_dirty(&ni->vfs_inode);
2686 	}
2687 
2688 out:
2689 	fnd_put(fnd);
2690 	return err;
2691 }
2692