1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
6 * This code builds two trees of free clusters extents.
7 * Trees are sorted by start of extent and by length of extent.
8 * NTFS_MAX_WND_EXTENTS defines the maximum number of elements in trees.
9 * In extreme case code reads on-disk bitmap to find free clusters.
13 #include <linux/buffer_head.h>
15 #include <linux/kernel.h>
21 * Maximum number of extents in tree.
23 #define NTFS_MAX_WND_EXTENTS (32u * 1024u)
31 struct rb_node_key start; /* Tree sorted by start. */
32 struct rb_node_key count; /* Tree sorted by len. */
35 static int wnd_rescan(struct wnd_bitmap *wnd);
36 static struct buffer_head *wnd_map(struct wnd_bitmap *wnd, size_t iw);
37 static bool wnd_is_free_hlp(struct wnd_bitmap *wnd, size_t bit, size_t bits);
39 static struct kmem_cache *ntfs_enode_cachep;
41 int __init ntfs3_init_bitmap(void)
43 ntfs_enode_cachep = kmem_cache_create("ntfs3_enode_cache",
44 sizeof(struct e_node), 0,
45 SLAB_RECLAIM_ACCOUNT, NULL);
46 return ntfs_enode_cachep ? 0 : -ENOMEM;
49 void ntfs3_exit_bitmap(void)
51 kmem_cache_destroy(ntfs_enode_cachep);
57 * b_pos + b_len - biggest fragment.
58 * Scan range [wpos wbits) window @buf.
60 * Return: -1 if not found.
62 static size_t wnd_scan(const void *buf, size_t wbit, u32 wpos, u32 wend,
63 size_t to_alloc, size_t *prev_tail, size_t *b_pos,
69 u32 used = find_next_zero_bit_le(buf, wend, wpos);
72 if (*b_len < *prev_tail) {
73 *b_pos = wbit - *prev_tail;
83 if (*b_len < *prev_tail) {
84 *b_pos = wbit - *prev_tail;
92 * Now we have a fragment [wpos, wend) staring with 0.
94 end = wpos + to_alloc - *prev_tail;
95 free_bits = find_next_bit_le(buf, min(end, wend), wpos);
97 free_len = *prev_tail + free_bits - wpos;
99 if (*b_len < free_len) {
100 *b_pos = wbit + wpos - *prev_tail;
104 if (free_len >= to_alloc)
105 return wbit + wpos - *prev_tail;
107 if (free_bits >= wend) {
108 *prev_tail += free_bits - wpos;
112 wpos = free_bits + 1;
121 * wnd_close - Frees all resources.
123 void wnd_close(struct wnd_bitmap *wnd)
125 struct rb_node *node, *next;
127 kfree(wnd->free_bits);
128 run_close(&wnd->run);
130 node = rb_first(&wnd->start_tree);
133 next = rb_next(node);
134 rb_erase(node, &wnd->start_tree);
135 kmem_cache_free(ntfs_enode_cachep,
136 rb_entry(node, struct e_node, start.node));
141 static struct rb_node *rb_lookup(struct rb_root *root, size_t v)
143 struct rb_node **p = &root->rb_node;
144 struct rb_node *r = NULL;
147 struct rb_node_key *k;
149 k = rb_entry(*p, struct rb_node_key, node);
152 } else if (v > k->key) {
164 * rb_insert_count - Helper function to insert special kind of 'count' tree.
166 static inline bool rb_insert_count(struct rb_root *root, struct e_node *e)
168 struct rb_node **p = &root->rb_node;
169 struct rb_node *parent = NULL;
170 size_t e_ckey = e->count.key;
171 size_t e_skey = e->start.key;
175 rb_entry(parent = *p, struct e_node, count.node);
177 if (e_ckey > k->count.key) {
179 } else if (e_ckey < k->count.key) {
181 } else if (e_skey < k->start.key) {
183 } else if (e_skey > k->start.key) {
191 rb_link_node(&e->count.node, parent, p);
192 rb_insert_color(&e->count.node, root);
197 * rb_insert_start - Helper function to insert special kind of 'count' tree.
199 static inline bool rb_insert_start(struct rb_root *root, struct e_node *e)
201 struct rb_node **p = &root->rb_node;
202 struct rb_node *parent = NULL;
203 size_t e_skey = e->start.key;
210 k = rb_entry(parent, struct e_node, start.node);
211 if (e_skey < k->start.key) {
213 } else if (e_skey > k->start.key) {
221 rb_link_node(&e->start.node, parent, p);
222 rb_insert_color(&e->start.node, root);
227 * wnd_add_free_ext - Adds a new extent of free space.
228 * @build: 1 when building tree.
230 static void wnd_add_free_ext(struct wnd_bitmap *wnd, size_t bit, size_t len,
233 struct e_node *e, *e0 = NULL;
234 size_t ib, end_in = bit + len;
238 /* Use extent_min to filter too short extents. */
239 if (wnd->count >= NTFS_MAX_WND_EXTENTS &&
240 len <= wnd->extent_min) {
245 /* Try to find extent before 'bit'. */
246 n = rb_lookup(&wnd->start_tree, bit);
249 n = rb_first(&wnd->start_tree);
251 e = rb_entry(n, struct e_node, start.node);
253 if (e->start.key + e->count.key == bit) {
257 rb_erase(&e->start.node, &wnd->start_tree);
258 rb_erase(&e->count.node, &wnd->count_tree);
267 e = rb_entry(n, struct e_node, start.node);
268 next_end = e->start.key + e->count.key;
269 if (e->start.key > end_in)
274 len += next_end - end_in;
276 rb_erase(&e->start.node, &wnd->start_tree);
277 rb_erase(&e->count.node, &wnd->count_tree);
283 kmem_cache_free(ntfs_enode_cachep, e);
286 if (wnd->uptodated != 1) {
287 /* Check bits before 'bit'. */
288 ib = wnd->zone_bit == wnd->zone_end ||
289 bit < wnd->zone_end ?
293 while (bit > ib && wnd_is_free_hlp(wnd, bit - 1, 1)) {
298 /* Check bits after 'end_in'. */
299 ib = wnd->zone_bit == wnd->zone_end ||
300 end_in > wnd->zone_bit ?
304 while (end_in < ib && wnd_is_free_hlp(wnd, end_in, 1)) {
310 /* Insert new fragment. */
311 if (wnd->count >= NTFS_MAX_WND_EXTENTS) {
313 kmem_cache_free(ntfs_enode_cachep, e0);
317 /* Compare with smallest fragment. */
318 n = rb_last(&wnd->count_tree);
319 e = rb_entry(n, struct e_node, count.node);
320 if (len <= e->count.key)
321 goto out; /* Do not insert small fragments. */
327 e2 = rb_entry(n, struct e_node, count.node);
328 /* Smallest fragment will be 'e2->count.key'. */
329 wnd->extent_min = e2->count.key;
332 /* Replace smallest fragment by new one. */
333 rb_erase(&e->start.node, &wnd->start_tree);
334 rb_erase(&e->count.node, &wnd->count_tree);
337 e = e0 ? e0 : kmem_cache_alloc(ntfs_enode_cachep, GFP_ATOMIC);
343 if (build && len <= wnd->extent_min)
344 wnd->extent_min = len;
348 if (len > wnd->extent_max)
349 wnd->extent_max = len;
351 rb_insert_start(&wnd->start_tree, e);
352 rb_insert_count(&wnd->count_tree, e);
359 * wnd_remove_free_ext - Remove a run from the cached free space.
361 static void wnd_remove_free_ext(struct wnd_bitmap *wnd, size_t bit, size_t len)
363 struct rb_node *n, *n3;
364 struct e_node *e, *e3;
365 size_t end_in = bit + len;
366 size_t end3, end, new_key, new_len, max_new_len;
368 /* Try to find extent before 'bit'. */
369 n = rb_lookup(&wnd->start_tree, bit);
374 e = rb_entry(n, struct e_node, start.node);
375 end = e->start.key + e->count.key;
377 new_key = new_len = 0;
380 /* Range [bit,end_in) must be inside 'e' or outside 'e' and 'n'. */
381 if (e->start.key > bit)
383 else if (end_in <= end) {
384 /* Range [bit,end_in) inside 'e'. */
386 new_len = end - end_in;
387 len = bit - e->start.key;
388 } else if (bit > end) {
394 e3 = rb_entry(n3, struct e_node, start.node);
395 if (e3->start.key >= end_in)
398 if (e3->count.key == wnd->extent_max)
401 end3 = e3->start.key + e3->count.key;
403 e3->start.key = end_in;
404 rb_erase(&e3->count.node, &wnd->count_tree);
405 e3->count.key = end3 - end_in;
406 rb_insert_count(&wnd->count_tree, e3);
411 rb_erase(&e3->start.node, &wnd->start_tree);
412 rb_erase(&e3->count.node, &wnd->count_tree);
414 kmem_cache_free(ntfs_enode_cachep, e3);
418 n3 = rb_first(&wnd->count_tree);
420 n3 ? rb_entry(n3, struct e_node, count.node)->count.key :
425 if (e->count.key != wnd->extent_max) {
427 } else if (rb_prev(&e->count.node)) {
430 n3 = rb_next(&e->count.node);
431 max_new_len = max(len, new_len);
433 wnd->extent_max = max_new_len;
435 e3 = rb_entry(n3, struct e_node, count.node);
436 wnd->extent_max = max(e3->count.key, max_new_len);
442 e->start.key = new_key;
443 rb_erase(&e->count.node, &wnd->count_tree);
444 e->count.key = new_len;
445 rb_insert_count(&wnd->count_tree, e);
447 rb_erase(&e->start.node, &wnd->start_tree);
448 rb_erase(&e->count.node, &wnd->count_tree);
450 kmem_cache_free(ntfs_enode_cachep, e);
454 rb_erase(&e->count.node, &wnd->count_tree);
456 rb_insert_count(&wnd->count_tree, e);
461 if (wnd->count >= NTFS_MAX_WND_EXTENTS) {
464 /* Get minimal extent. */
465 e = rb_entry(rb_last(&wnd->count_tree), struct e_node,
467 if (e->count.key > new_len)
470 /* Replace minimum. */
471 rb_erase(&e->start.node, &wnd->start_tree);
472 rb_erase(&e->count.node, &wnd->count_tree);
475 e = kmem_cache_alloc(ntfs_enode_cachep, GFP_ATOMIC);
481 e->start.key = new_key;
482 e->count.key = new_len;
483 rb_insert_start(&wnd->start_tree, e);
484 rb_insert_count(&wnd->count_tree, e);
489 if (!wnd->count && 1 != wnd->uptodated)
494 * wnd_rescan - Scan all bitmap. Used while initialization.
496 static int wnd_rescan(struct wnd_bitmap *wnd)
499 size_t prev_tail = 0;
500 struct super_block *sb = wnd->sb;
501 struct ntfs_sb_info *sbi = sb->s_fs_info;
503 u32 blocksize = sb->s_blocksize;
504 u8 cluster_bits = sbi->cluster_bits;
505 u32 wbits = 8 * sb->s_blocksize;
507 size_t wpos, wbit, iw, vbo;
508 struct buffer_head *bh = NULL;
513 wnd->extent_min = MINUS_ONE_T;
514 wnd->total_zeroes = 0;
518 for (iw = 0; iw < wnd->nwnd; iw++) {
519 if (iw + 1 == wnd->nwnd)
520 wbits = wnd->bits_last;
523 if (!wnd->free_bits[iw]) {
526 wnd_add_free_ext(wnd,
533 if (wbits == wnd->free_bits[iw]) {
536 wnd->total_zeroes += wbits;
542 u32 off = vbo & sbi->cluster_mask;
544 if (!run_lookup_entry(&wnd->run, vbo >> cluster_bits,
545 &lcn, &clen, NULL)) {
550 lbo = ((u64)lcn << cluster_bits) + off;
551 len = ((u64)clen << cluster_bits) - off;
554 bh = ntfs_bread(sb, lbo >> sb->s_blocksize_bits);
560 used = ntfs_bitmap_weight_le(bh->b_data, wbits);
563 wnd->free_bits[iw] = frb;
564 wnd->total_zeroes += frb;
570 if (wbit + wbits > wnd->nbits)
571 wbits = wnd->nbits - wbit;
574 used = find_next_zero_bit_le(bh->b_data, wbits, wpos);
576 if (used > wpos && prev_tail) {
577 wnd_add_free_ext(wnd, wbit + wpos - prev_tail,
585 /* No free blocks. */
590 frb = find_next_bit_le(bh->b_data, wbits, wpos);
592 /* Keep last free block. */
593 prev_tail += frb - wpos;
597 wnd_add_free_ext(wnd, wbit + wpos - prev_tail,
598 frb + prev_tail - wpos, true);
600 /* Skip free block and first '1'. */
602 /* Reset previous tail. */
604 } while (wpos < wbits);
619 /* Add last block. */
621 wnd_add_free_ext(wnd, wnd->nbits - prev_tail, prev_tail, true);
624 * Before init cycle wnd->uptodated was 0.
625 * If any errors or limits occurs while initialization then
626 * wnd->uptodated will be -1.
627 * If 'uptodated' is still 0 then Tree is really updated.
632 if (wnd->zone_bit != wnd->zone_end) {
633 size_t zlen = wnd->zone_end - wnd->zone_bit;
635 wnd->zone_end = wnd->zone_bit;
636 wnd_zone_set(wnd, wnd->zone_bit, zlen);
643 int wnd_init(struct wnd_bitmap *wnd, struct super_block *sb, size_t nbits)
646 u32 blocksize = sb->s_blocksize;
647 u32 wbits = blocksize * 8;
649 init_rwsem(&wnd->rw_lock);
653 wnd->total_zeroes = nbits;
654 wnd->extent_max = MINUS_ONE_T;
655 wnd->zone_bit = wnd->zone_end = 0;
656 wnd->nwnd = bytes_to_block(sb, bitmap_size(nbits));
657 wnd->bits_last = nbits & (wbits - 1);
659 wnd->bits_last = wbits;
662 kcalloc(wnd->nwnd, sizeof(u16), GFP_NOFS | __GFP_NOWARN);
666 err = wnd_rescan(wnd);
676 * wnd_map - Call sb_bread for requested window.
678 static struct buffer_head *wnd_map(struct wnd_bitmap *wnd, size_t iw)
682 struct super_block *sb = wnd->sb;
683 struct ntfs_sb_info *sbi;
684 struct buffer_head *bh;
688 vbo = (u64)iw << sb->s_blocksize_bits;
690 if (!run_lookup_entry(&wnd->run, vbo >> sbi->cluster_bits, &lcn, &clen,
692 return ERR_PTR(-ENOENT);
695 lbo = ((u64)lcn << sbi->cluster_bits) + (vbo & sbi->cluster_mask);
697 bh = ntfs_bread(wnd->sb, lbo >> sb->s_blocksize_bits);
699 return ERR_PTR(-EIO);
705 * wnd_set_free - Mark the bits range from bit to bit + bits as free.
707 int wnd_set_free(struct wnd_bitmap *wnd, size_t bit, size_t bits)
710 struct super_block *sb = wnd->sb;
712 u32 wbits = 8 * sb->s_blocksize;
713 size_t iw = bit >> (sb->s_blocksize_bits + 3);
714 u32 wbit = bit & (wbits - 1);
715 struct buffer_head *bh;
717 while (iw < wnd->nwnd && bits) {
720 if (iw + 1 == wnd->nwnd)
721 wbits = wnd->bits_last;
724 op = min_t(u32, tail, bits);
726 bh = wnd_map(wnd, iw);
734 ntfs_bitmap_clear_le(bh->b_data, wbit, op);
736 wnd->free_bits[iw] += op;
738 set_buffer_uptodate(bh);
739 mark_buffer_dirty(bh);
743 wnd->total_zeroes += op;
749 wnd_add_free_ext(wnd, bit, bits0, false);
755 * wnd_set_used - Mark the bits range from bit to bit + bits as used.
757 int wnd_set_used(struct wnd_bitmap *wnd, size_t bit, size_t bits)
760 struct super_block *sb = wnd->sb;
762 size_t iw = bit >> (sb->s_blocksize_bits + 3);
763 u32 wbits = 8 * sb->s_blocksize;
764 u32 wbit = bit & (wbits - 1);
765 struct buffer_head *bh;
767 while (iw < wnd->nwnd && bits) {
770 if (unlikely(iw + 1 == wnd->nwnd))
771 wbits = wnd->bits_last;
774 op = min_t(u32, tail, bits);
776 bh = wnd_map(wnd, iw);
784 ntfs_bitmap_set_le(bh->b_data, wbit, op);
785 wnd->free_bits[iw] -= op;
787 set_buffer_uptodate(bh);
788 mark_buffer_dirty(bh);
792 wnd->total_zeroes -= op;
798 if (!RB_EMPTY_ROOT(&wnd->start_tree))
799 wnd_remove_free_ext(wnd, bit, bits0);
805 * wnd_set_used_safe - Mark the bits range from bit to bit + bits as used.
807 * Unlikely wnd_set_used/wnd_set_free this function is not full trusted.
808 * It scans every bit in bitmap and marks free bit as used.
809 * @done - how many bits were marked as used.
811 * NOTE: normally *done should be 0.
813 int wnd_set_used_safe(struct wnd_bitmap *wnd, size_t bit, size_t bits,
816 size_t i, from = 0, len = 0;
820 for (i = 0; i < bits; i++) {
821 if (wnd_is_free(wnd, bit + i, 1)) {
826 err = wnd_set_used(wnd, from, len);
836 err = wnd_set_used(wnd, from, len);
845 * Return: True if all clusters [bit, bit+bits) are free (bitmap only).
847 static bool wnd_is_free_hlp(struct wnd_bitmap *wnd, size_t bit, size_t bits)
849 struct super_block *sb = wnd->sb;
850 size_t iw = bit >> (sb->s_blocksize_bits + 3);
851 u32 wbits = 8 * sb->s_blocksize;
852 u32 wbit = bit & (wbits - 1);
854 while (iw < wnd->nwnd && bits) {
857 if (unlikely(iw + 1 == wnd->nwnd))
858 wbits = wnd->bits_last;
861 op = min_t(u32, tail, bits);
863 if (wbits != wnd->free_bits[iw]) {
865 struct buffer_head *bh = wnd_map(wnd, iw);
870 ret = are_bits_clear(bh->b_data, wbit, op);
888 * Return: True if all clusters [bit, bit+bits) are free.
890 bool wnd_is_free(struct wnd_bitmap *wnd, size_t bit, size_t bits)
897 if (RB_EMPTY_ROOT(&wnd->start_tree))
900 n = rb_lookup(&wnd->start_tree, bit);
904 e = rb_entry(n, struct e_node, start.node);
906 end = e->start.key + e->count.key;
908 if (bit < end && bit + bits <= end)
912 ret = wnd_is_free_hlp(wnd, bit, bits);
920 * Return: True if all clusters [bit, bit+bits) are used.
922 bool wnd_is_used(struct wnd_bitmap *wnd, size_t bit, size_t bits)
925 struct super_block *sb = wnd->sb;
926 size_t iw = bit >> (sb->s_blocksize_bits + 3);
927 u32 wbits = 8 * sb->s_blocksize;
928 u32 wbit = bit & (wbits - 1);
933 if (RB_EMPTY_ROOT(&wnd->start_tree))
937 n = rb_lookup(&wnd->start_tree, end - 1);
941 e = rb_entry(n, struct e_node, start.node);
942 if (e->start.key + e->count.key > bit)
946 while (iw < wnd->nwnd && bits) {
949 if (unlikely(iw + 1 == wnd->nwnd))
950 wbits = wnd->bits_last;
953 op = min_t(u32, tail, bits);
955 if (wnd->free_bits[iw]) {
957 struct buffer_head *bh = wnd_map(wnd, iw);
962 ret = are_bits_set(bh->b_data, wbit, op);
979 * wnd_find - Look for free space.
981 * - flags - BITMAP_FIND_XXX flags
983 * Return: 0 if not found.
985 size_t wnd_find(struct wnd_bitmap *wnd, size_t to_alloc, size_t hint,
986 size_t flags, size_t *allocated)
988 struct super_block *sb;
989 u32 wbits, wpos, wzbit, wzend;
990 size_t fnd, max_alloc, b_len, b_pos;
991 size_t iw, prev_tail, nwnd, wbit, ebit, zbit, zend;
992 size_t to_alloc0 = to_alloc;
993 const struct e_node *e;
994 const struct rb_node *pr, *cr;
997 struct buffer_head *bh;
999 /* Fast checking for available free space. */
1000 if (flags & BITMAP_FIND_FULL) {
1001 size_t zeroes = wnd_zeroes(wnd);
1003 zeroes -= wnd->zone_end - wnd->zone_bit;
1004 if (zeroes < to_alloc0)
1007 if (to_alloc0 > wnd->extent_max)
1010 if (to_alloc > wnd->extent_max)
1011 to_alloc = wnd->extent_max;
1014 if (wnd->zone_bit <= hint && hint < wnd->zone_end)
1015 hint = wnd->zone_end;
1017 max_alloc = wnd->nbits;
1020 if (hint >= max_alloc)
1023 if (RB_EMPTY_ROOT(&wnd->start_tree)) {
1024 if (wnd->uptodated == 1) {
1025 /* Extents tree is updated -> No free space. */
1033 goto allocate_biggest;
1035 /* Use hint: Enumerate extents by start >= hint. */
1037 cr = wnd->start_tree.rb_node;
1040 e = rb_entry(cr, struct e_node, start.node);
1042 if (e->start.key == hint)
1045 if (e->start.key < hint) {
1055 e = pr ? rb_entry(pr, struct e_node, start.node) : NULL;
1061 goto allocate_biggest;
1063 if (e->start.key + e->count.key > hint) {
1064 /* We have found extension with 'hint' inside. */
1065 size_t len = e->start.key + e->count.key - hint;
1067 if (len >= to_alloc && hint + to_alloc <= max_alloc) {
1072 if (!(flags & BITMAP_FIND_FULL)) {
1076 if (hint + len <= max_alloc) {
1085 /* Allocate from biggest free extent. */
1086 e = rb_entry(rb_first(&wnd->count_tree), struct e_node, count.node);
1087 if (e->count.key != wnd->extent_max)
1088 wnd->extent_max = e->count.key;
1090 if (e->count.key < max_alloc) {
1091 if (e->count.key >= to_alloc) {
1093 } else if (flags & BITMAP_FIND_FULL) {
1094 if (e->count.key < to_alloc0) {
1095 /* Biggest free block is less then requested. */
1098 to_alloc = e->count.key;
1099 } else if (-1 != wnd->uptodated) {
1100 to_alloc = e->count.key;
1102 /* Check if we can use more bits. */
1103 size_t op, max_check;
1104 struct rb_root start_tree;
1106 memcpy(&start_tree, &wnd->start_tree,
1107 sizeof(struct rb_root));
1108 memset(&wnd->start_tree, 0, sizeof(struct rb_root));
1110 max_check = e->start.key + to_alloc;
1111 if (max_check > max_alloc)
1112 max_check = max_alloc;
1113 for (op = e->start.key + e->count.key; op < max_check;
1115 if (!wnd_is_free(wnd, op, 1))
1118 memcpy(&wnd->start_tree, &start_tree,
1119 sizeof(struct rb_root));
1120 to_alloc = op - e->start.key;
1123 /* Prepare to return. */
1125 if (e->start.key + to_alloc > max_alloc)
1126 to_alloc = max_alloc - e->start.key;
1130 if (wnd->uptodated == 1) {
1131 /* Extents tree is updated -> no free space. */
1135 b_len = e->count.key;
1136 b_pos = e->start.key;
1140 log2_bits = sb->s_blocksize_bits + 3;
1142 /* At most two ranges [hint, max_alloc) + [0, hint). */
1145 /* TODO: Optimize request for case nbits > wbits. */
1146 iw = hint >> log2_bits;
1147 wbits = sb->s_blocksize * 8;
1148 wpos = hint & (wbits - 1);
1152 if (max_alloc == wnd->nbits) {
1155 size_t t = max_alloc + wbits - 1;
1157 nwnd = likely(t > max_alloc) ? (t >> log2_bits) : wnd->nwnd;
1160 /* Enumerate all windows. */
1161 for (; iw < nwnd; iw++) {
1162 wbit = iw << log2_bits;
1164 if (!wnd->free_bits[iw]) {
1165 if (prev_tail > b_len) {
1166 b_pos = wbit - prev_tail;
1170 /* Skip full used window. */
1176 if (unlikely(iw + 1 == nwnd)) {
1177 if (max_alloc == wnd->nbits) {
1178 wbits = wnd->bits_last;
1180 size_t t = max_alloc & (wbits - 1);
1184 fbits_valid = false;
1189 if (wnd->zone_end > wnd->zone_bit) {
1190 ebit = wbit + wbits;
1191 zbit = max(wnd->zone_bit, wbit);
1192 zend = min(wnd->zone_end, ebit);
1194 /* Here we have a window [wbit, ebit) and zone [zbit, zend). */
1196 /* Zone does not overlap window. */
1198 wzbit = zbit - wbit;
1199 wzend = zend - wbit;
1201 /* Zone overlaps window. */
1202 if (wnd->free_bits[iw] == wzend - wzbit) {
1208 /* Scan two ranges window: [wbit, zbit) and [zend, ebit). */
1209 bh = wnd_map(wnd, iw);
1218 /* Scan range [wbit, zbit). */
1220 /* Scan range [wpos, zbit). */
1221 fnd = wnd_scan(bh->b_data, wbit, wpos,
1225 if (fnd != MINUS_ONE_T) {
1233 /* Scan range [zend, ebit). */
1234 if (wzend < wbits) {
1235 fnd = wnd_scan(bh->b_data, wbit,
1236 max(wzend, wpos), wbits,
1237 to_alloc, &prev_tail,
1239 if (fnd != MINUS_ONE_T) {
1251 /* Current window does not overlap zone. */
1252 if (!wpos && fbits_valid && wnd->free_bits[iw] == wbits) {
1253 /* Window is empty. */
1254 if (prev_tail + wbits >= to_alloc) {
1255 fnd = wbit + wpos - prev_tail;
1259 /* Increase 'prev_tail' and process next window. */
1266 bh = wnd_map(wnd, iw);
1274 /* Scan range [wpos, eBits). */
1275 fnd = wnd_scan(bh->b_data, wbit, wpos, wbits, to_alloc,
1276 &prev_tail, &b_pos, &b_len);
1278 if (fnd != MINUS_ONE_T)
1282 if (b_len < prev_tail) {
1283 /* The last fragment. */
1285 b_pos = max_alloc - prev_tail;
1290 * We have scanned range [hint max_alloc).
1291 * Prepare to scan range [0 hint + to_alloc).
1293 size_t nextmax = hint + to_alloc;
1295 if (likely(nextmax >= hint) && nextmax < max_alloc)
1296 max_alloc = nextmax;
1304 wnd->extent_max = b_len;
1306 if (flags & BITMAP_FIND_FULL)
1313 if (flags & BITMAP_FIND_MARK_AS_USED) {
1314 /* TODO: Optimize remove extent (pass 'e'?). */
1315 if (wnd_set_used(wnd, fnd, to_alloc))
1317 } else if (wnd->extent_max != MINUS_ONE_T &&
1318 to_alloc > wnd->extent_max) {
1319 wnd->extent_max = to_alloc;
1330 * wnd_extend - Extend bitmap ($MFT bitmap).
1332 int wnd_extend(struct wnd_bitmap *wnd, size_t new_bits)
1335 struct super_block *sb = wnd->sb;
1336 struct ntfs_sb_info *sbi = sb->s_fs_info;
1337 u32 blocksize = sb->s_blocksize;
1338 u32 wbits = blocksize * 8;
1340 size_t bits, iw, new_wnd;
1341 size_t old_bits = wnd->nbits;
1344 if (new_bits <= old_bits)
1347 /* Align to 8 byte boundary. */
1348 new_wnd = bytes_to_block(sb, bitmap_size(new_bits));
1349 new_last = new_bits & (wbits - 1);
1353 if (new_wnd != wnd->nwnd) {
1354 new_free = kmalloc_array(new_wnd, sizeof(u16), GFP_NOFS);
1358 memcpy(new_free, wnd->free_bits, wnd->nwnd * sizeof(short));
1359 memset(new_free + wnd->nwnd, 0,
1360 (new_wnd - wnd->nwnd) * sizeof(short));
1361 kfree(wnd->free_bits);
1362 wnd->free_bits = new_free;
1365 /* Zero bits [old_bits,new_bits). */
1366 bits = new_bits - old_bits;
1367 b0 = old_bits & (wbits - 1);
1369 for (iw = old_bits >> (sb->s_blocksize_bits + 3); bits; iw += 1) {
1372 u64 vbo, lbo, bytes;
1373 struct buffer_head *bh;
1375 if (iw + 1 == new_wnd)
1378 op = b0 + bits > wbits ? wbits - b0 : bits;
1379 vbo = (u64)iw * blocksize;
1381 err = ntfs_vbo_to_lbo(sbi, &wnd->run, vbo, &lbo, &bytes);
1385 bh = ntfs_bread(sb, lbo >> sb->s_blocksize_bits);
1391 ntfs_bitmap_clear_le(bh->b_data, b0, blocksize * 8 - b0);
1392 frb = wbits - ntfs_bitmap_weight_le(bh->b_data, wbits);
1393 wnd->total_zeroes += frb - wnd->free_bits[iw];
1394 wnd->free_bits[iw] = frb;
1396 set_buffer_uptodate(bh);
1397 mark_buffer_dirty(bh);
1399 /* err = sync_dirty_buffer(bh); */
1405 wnd->nbits = new_bits;
1406 wnd->nwnd = new_wnd;
1407 wnd->bits_last = new_last;
1409 wnd_add_free_ext(wnd, old_bits, new_bits - old_bits, false);
1414 void wnd_zone_set(struct wnd_bitmap *wnd, size_t lcn, size_t len)
1416 size_t zlen = wnd->zone_end - wnd->zone_bit;
1419 wnd_add_free_ext(wnd, wnd->zone_bit, zlen, false);
1421 if (!RB_EMPTY_ROOT(&wnd->start_tree) && len)
1422 wnd_remove_free_ext(wnd, lcn, len);
1424 wnd->zone_bit = lcn;
1425 wnd->zone_end = lcn + len;
1428 int ntfs_trim_fs(struct ntfs_sb_info *sbi, struct fstrim_range *range)
1431 struct super_block *sb = sbi->sb;
1432 struct wnd_bitmap *wnd = &sbi->used.bitmap;
1433 u32 wbits = 8 * sb->s_blocksize;
1434 CLST len = 0, lcn = 0, done = 0;
1435 CLST minlen = bytes_to_cluster(sbi, range->minlen);
1436 CLST lcn_from = bytes_to_cluster(sbi, range->start);
1437 size_t iw = lcn_from >> (sb->s_blocksize_bits + 3);
1438 u32 wbit = lcn_from & (wbits - 1);
1444 if (range->len == (u64)-1)
1445 lcn_to = wnd->nbits;
1447 lcn_to = bytes_to_cluster(sbi, range->start + range->len);
1449 down_read_nested(&wnd->rw_lock, BITMAP_MUTEX_CLUSTERS);
1451 for (; iw < wnd->nwnd; iw++, wbit = 0) {
1452 CLST lcn_wnd = iw * wbits;
1453 struct buffer_head *bh;
1455 if (lcn_wnd > lcn_to)
1458 if (!wnd->free_bits[iw])
1461 if (iw + 1 == wnd->nwnd)
1462 wbits = wnd->bits_last;
1464 if (lcn_wnd + wbits > lcn_to)
1465 wbits = lcn_to - lcn_wnd;
1467 bh = wnd_map(wnd, iw);
1473 for (; wbit < wbits; wbit++) {
1474 if (!test_bit_le(wbit, bh->b_data)) {
1476 lcn = lcn_wnd + wbit;
1480 if (len >= minlen) {
1481 err = ntfs_discard(sbi, lcn, len);
1491 /* Process the last fragment. */
1492 if (len >= minlen) {
1493 err = ntfs_discard(sbi, lcn, len);
1500 range->len = (u64)done << sbi->cluster_bits;
1502 up_read(&wnd->rw_lock);
1507 #if BITS_PER_LONG == 64
1508 typedef __le64 bitmap_ulong;
1509 #define cpu_to_ul(x) cpu_to_le64(x)
1510 #define ul_to_cpu(x) le64_to_cpu(x)
1512 typedef __le32 bitmap_ulong;
1513 #define cpu_to_ul(x) cpu_to_le32(x)
1514 #define ul_to_cpu(x) le32_to_cpu(x)
1517 void ntfs_bitmap_set_le(void *map, unsigned int start, int len)
1519 bitmap_ulong *p = (bitmap_ulong *)map + BIT_WORD(start);
1520 const unsigned int size = start + len;
1521 int bits_to_set = BITS_PER_LONG - (start % BITS_PER_LONG);
1522 bitmap_ulong mask_to_set = cpu_to_ul(BITMAP_FIRST_WORD_MASK(start));
1524 while (len - bits_to_set >= 0) {
1527 bits_to_set = BITS_PER_LONG;
1528 mask_to_set = cpu_to_ul(~0UL);
1532 mask_to_set &= cpu_to_ul(BITMAP_LAST_WORD_MASK(size));
1537 void ntfs_bitmap_clear_le(void *map, unsigned int start, int len)
1539 bitmap_ulong *p = (bitmap_ulong *)map + BIT_WORD(start);
1540 const unsigned int size = start + len;
1541 int bits_to_clear = BITS_PER_LONG - (start % BITS_PER_LONG);
1542 bitmap_ulong mask_to_clear = cpu_to_ul(BITMAP_FIRST_WORD_MASK(start));
1544 while (len - bits_to_clear >= 0) {
1545 *p &= ~mask_to_clear;
1546 len -= bits_to_clear;
1547 bits_to_clear = BITS_PER_LONG;
1548 mask_to_clear = cpu_to_ul(~0UL);
1552 mask_to_clear &= cpu_to_ul(BITMAP_LAST_WORD_MASK(size));
1553 *p &= ~mask_to_clear;
1557 unsigned int ntfs_bitmap_weight_le(const void *bitmap, int bits)
1559 const ulong *bmp = bitmap;
1560 unsigned int k, lim = bits / BITS_PER_LONG;
1563 for (k = 0; k < lim; k++)
1564 w += hweight_long(bmp[k]);
1566 if (bits % BITS_PER_LONG) {
1567 w += hweight_long(ul_to_cpu(((bitmap_ulong *)bitmap)[k]) &
1568 BITMAP_LAST_WORD_MASK(bits));