Loading...
Note: File does not exist in v5.4.
1// SPDX-License-Identifier: GPL-2.0
2/*
3 *
4 * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
5 *
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.
10 *
11 */
12
13#include <linux/buffer_head.h>
14#include <linux/fs.h>
15#include <linux/kernel.h>
16
17#include "ntfs.h"
18#include "ntfs_fs.h"
19
20/*
21 * Maximum number of extents in tree.
22 */
23#define NTFS_MAX_WND_EXTENTS (32u * 1024u)
24
25struct rb_node_key {
26 struct rb_node node;
27 size_t key;
28};
29
30struct e_node {
31 struct rb_node_key start; /* Tree sorted by start. */
32 struct rb_node_key count; /* Tree sorted by len. */
33};
34
35static int wnd_rescan(struct wnd_bitmap *wnd);
36static struct buffer_head *wnd_map(struct wnd_bitmap *wnd, size_t iw);
37static bool wnd_is_free_hlp(struct wnd_bitmap *wnd, size_t bit, size_t bits);
38
39static struct kmem_cache *ntfs_enode_cachep;
40
41int __init ntfs3_init_bitmap(void)
42{
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;
47}
48
49void ntfs3_exit_bitmap(void)
50{
51 kmem_cache_destroy(ntfs_enode_cachep);
52}
53
54/*
55 * wnd_scan
56 *
57 * b_pos + b_len - biggest fragment.
58 * Scan range [wpos wbits) window @buf.
59 *
60 * Return: -1 if not found.
61 */
62static 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,
64 size_t *b_len)
65{
66 while (wpos < wend) {
67 size_t free_len;
68 u32 free_bits, end;
69 u32 used = find_next_zero_bit_le(buf, wend, wpos);
70
71 if (used >= wend) {
72 if (*b_len < *prev_tail) {
73 *b_pos = wbit - *prev_tail;
74 *b_len = *prev_tail;
75 }
76
77 *prev_tail = 0;
78 return -1;
79 }
80
81 if (used > wpos) {
82 wpos = used;
83 if (*b_len < *prev_tail) {
84 *b_pos = wbit - *prev_tail;
85 *b_len = *prev_tail;
86 }
87
88 *prev_tail = 0;
89 }
90
91 /*
92 * Now we have a fragment [wpos, wend) staring with 0.
93 */
94 end = wpos + to_alloc - *prev_tail;
95 free_bits = find_next_bit_le(buf, min(end, wend), wpos);
96
97 free_len = *prev_tail + free_bits - wpos;
98
99 if (*b_len < free_len) {
100 *b_pos = wbit + wpos - *prev_tail;
101 *b_len = free_len;
102 }
103
104 if (free_len >= to_alloc)
105 return wbit + wpos - *prev_tail;
106
107 if (free_bits >= wend) {
108 *prev_tail += free_bits - wpos;
109 return -1;
110 }
111
112 wpos = free_bits + 1;
113
114 *prev_tail = 0;
115 }
116
117 return -1;
118}
119
120/*
121 * wnd_close - Frees all resources.
122 */
123void wnd_close(struct wnd_bitmap *wnd)
124{
125 struct rb_node *node, *next;
126
127 kvfree(wnd->free_bits);
128 wnd->free_bits = NULL;
129 run_close(&wnd->run);
130
131 node = rb_first(&wnd->start_tree);
132
133 while (node) {
134 next = rb_next(node);
135 rb_erase(node, &wnd->start_tree);
136 kmem_cache_free(ntfs_enode_cachep,
137 rb_entry(node, struct e_node, start.node));
138 node = next;
139 }
140}
141
142static struct rb_node *rb_lookup(struct rb_root *root, size_t v)
143{
144 struct rb_node **p = &root->rb_node;
145 struct rb_node *r = NULL;
146
147 while (*p) {
148 struct rb_node_key *k;
149
150 k = rb_entry(*p, struct rb_node_key, node);
151 if (v < k->key) {
152 p = &(*p)->rb_left;
153 } else if (v > k->key) {
154 r = &k->node;
155 p = &(*p)->rb_right;
156 } else {
157 return &k->node;
158 }
159 }
160
161 return r;
162}
163
164/*
165 * rb_insert_count - Helper function to insert special kind of 'count' tree.
166 */
167static inline bool rb_insert_count(struct rb_root *root, struct e_node *e)
168{
169 struct rb_node **p = &root->rb_node;
170 struct rb_node *parent = NULL;
171 size_t e_ckey = e->count.key;
172 size_t e_skey = e->start.key;
173
174 while (*p) {
175 struct e_node *k =
176 rb_entry(parent = *p, struct e_node, count.node);
177
178 if (e_ckey > k->count.key) {
179 p = &(*p)->rb_left;
180 } else if (e_ckey < k->count.key) {
181 p = &(*p)->rb_right;
182 } else if (e_skey < k->start.key) {
183 p = &(*p)->rb_left;
184 } else if (e_skey > k->start.key) {
185 p = &(*p)->rb_right;
186 } else {
187 WARN_ON(1);
188 return false;
189 }
190 }
191
192 rb_link_node(&e->count.node, parent, p);
193 rb_insert_color(&e->count.node, root);
194 return true;
195}
196
197/*
198 * rb_insert_start - Helper function to insert special kind of 'count' tree.
199 */
200static inline bool rb_insert_start(struct rb_root *root, struct e_node *e)
201{
202 struct rb_node **p = &root->rb_node;
203 struct rb_node *parent = NULL;
204 size_t e_skey = e->start.key;
205
206 while (*p) {
207 struct e_node *k;
208
209 parent = *p;
210
211 k = rb_entry(parent, struct e_node, start.node);
212 if (e_skey < k->start.key) {
213 p = &(*p)->rb_left;
214 } else if (e_skey > k->start.key) {
215 p = &(*p)->rb_right;
216 } else {
217 WARN_ON(1);
218 return false;
219 }
220 }
221
222 rb_link_node(&e->start.node, parent, p);
223 rb_insert_color(&e->start.node, root);
224 return true;
225}
226
227/*
228 * wnd_add_free_ext - Adds a new extent of free space.
229 * @build: 1 when building tree.
230 */
231static void wnd_add_free_ext(struct wnd_bitmap *wnd, size_t bit, size_t len,
232 bool build)
233{
234 struct e_node *e, *e0 = NULL;
235 size_t ib, end_in = bit + len;
236 struct rb_node *n;
237
238 if (build) {
239 /* Use extent_min to filter too short extents. */
240 if (wnd->count >= NTFS_MAX_WND_EXTENTS &&
241 len <= wnd->extent_min) {
242 wnd->uptodated = -1;
243 return;
244 }
245 } else {
246 /* Try to find extent before 'bit'. */
247 n = rb_lookup(&wnd->start_tree, bit);
248
249 if (!n) {
250 n = rb_first(&wnd->start_tree);
251 } else {
252 e = rb_entry(n, struct e_node, start.node);
253 n = rb_next(n);
254 if (e->start.key + e->count.key == bit) {
255 /* Remove left. */
256 bit = e->start.key;
257 len += e->count.key;
258 rb_erase(&e->start.node, &wnd->start_tree);
259 rb_erase(&e->count.node, &wnd->count_tree);
260 wnd->count -= 1;
261 e0 = e;
262 }
263 }
264
265 while (n) {
266 size_t next_end;
267
268 e = rb_entry(n, struct e_node, start.node);
269 next_end = e->start.key + e->count.key;
270 if (e->start.key > end_in)
271 break;
272
273 /* Remove right. */
274 n = rb_next(n);
275 len += next_end - end_in;
276 end_in = next_end;
277 rb_erase(&e->start.node, &wnd->start_tree);
278 rb_erase(&e->count.node, &wnd->count_tree);
279 wnd->count -= 1;
280
281 if (!e0)
282 e0 = e;
283 else
284 kmem_cache_free(ntfs_enode_cachep, e);
285 }
286
287 if (wnd->uptodated != 1) {
288 /* Check bits before 'bit'. */
289 ib = wnd->zone_bit == wnd->zone_end ||
290 bit < wnd->zone_end ?
291 0 :
292 wnd->zone_end;
293
294 while (bit > ib && wnd_is_free_hlp(wnd, bit - 1, 1)) {
295 bit -= 1;
296 len += 1;
297 }
298
299 /* Check bits after 'end_in'. */
300 ib = wnd->zone_bit == wnd->zone_end ||
301 end_in > wnd->zone_bit ?
302 wnd->nbits :
303 wnd->zone_bit;
304
305 while (end_in < ib && wnd_is_free_hlp(wnd, end_in, 1)) {
306 end_in += 1;
307 len += 1;
308 }
309 }
310 }
311 /* Insert new fragment. */
312 if (wnd->count >= NTFS_MAX_WND_EXTENTS) {
313 if (e0)
314 kmem_cache_free(ntfs_enode_cachep, e0);
315
316 wnd->uptodated = -1;
317
318 /* Compare with smallest fragment. */
319 n = rb_last(&wnd->count_tree);
320 e = rb_entry(n, struct e_node, count.node);
321 if (len <= e->count.key)
322 goto out; /* Do not insert small fragments. */
323
324 if (build) {
325 struct e_node *e2;
326
327 n = rb_prev(n);
328 e2 = rb_entry(n, struct e_node, count.node);
329 /* Smallest fragment will be 'e2->count.key'. */
330 wnd->extent_min = e2->count.key;
331 }
332
333 /* Replace smallest fragment by new one. */
334 rb_erase(&e->start.node, &wnd->start_tree);
335 rb_erase(&e->count.node, &wnd->count_tree);
336 wnd->count -= 1;
337 } else {
338 e = e0 ? e0 : kmem_cache_alloc(ntfs_enode_cachep, GFP_ATOMIC);
339 if (!e) {
340 wnd->uptodated = -1;
341 goto out;
342 }
343
344 if (build && len <= wnd->extent_min)
345 wnd->extent_min = len;
346 }
347 e->start.key = bit;
348 e->count.key = len;
349 if (len > wnd->extent_max)
350 wnd->extent_max = len;
351
352 rb_insert_start(&wnd->start_tree, e);
353 rb_insert_count(&wnd->count_tree, e);
354 wnd->count += 1;
355
356out:;
357}
358
359/*
360 * wnd_remove_free_ext - Remove a run from the cached free space.
361 */
362static void wnd_remove_free_ext(struct wnd_bitmap *wnd, size_t bit, size_t len)
363{
364 struct rb_node *n, *n3;
365 struct e_node *e, *e3;
366 size_t end_in = bit + len;
367 size_t end3, end, new_key, new_len, max_new_len;
368
369 /* Try to find extent before 'bit'. */
370 n = rb_lookup(&wnd->start_tree, bit);
371
372 if (!n)
373 return;
374
375 e = rb_entry(n, struct e_node, start.node);
376 end = e->start.key + e->count.key;
377
378 new_key = new_len = 0;
379 len = e->count.key;
380
381 /* Range [bit,end_in) must be inside 'e' or outside 'e' and 'n'. */
382 if (e->start.key > bit)
383 ;
384 else if (end_in <= end) {
385 /* Range [bit,end_in) inside 'e'. */
386 new_key = end_in;
387 new_len = end - end_in;
388 len = bit - e->start.key;
389 } else if (bit > end) {
390 bool bmax = false;
391
392 n3 = rb_next(n);
393
394 while (n3) {
395 e3 = rb_entry(n3, struct e_node, start.node);
396 if (e3->start.key >= end_in)
397 break;
398
399 if (e3->count.key == wnd->extent_max)
400 bmax = true;
401
402 end3 = e3->start.key + e3->count.key;
403 if (end3 > end_in) {
404 e3->start.key = end_in;
405 rb_erase(&e3->count.node, &wnd->count_tree);
406 e3->count.key = end3 - end_in;
407 rb_insert_count(&wnd->count_tree, e3);
408 break;
409 }
410
411 n3 = rb_next(n3);
412 rb_erase(&e3->start.node, &wnd->start_tree);
413 rb_erase(&e3->count.node, &wnd->count_tree);
414 wnd->count -= 1;
415 kmem_cache_free(ntfs_enode_cachep, e3);
416 }
417 if (!bmax)
418 return;
419 n3 = rb_first(&wnd->count_tree);
420 wnd->extent_max =
421 n3 ? rb_entry(n3, struct e_node, count.node)->count.key :
422 0;
423 return;
424 }
425
426 if (e->count.key != wnd->extent_max) {
427 ;
428 } else if (rb_prev(&e->count.node)) {
429 ;
430 } else {
431 n3 = rb_next(&e->count.node);
432 max_new_len = max(len, new_len);
433 if (!n3) {
434 wnd->extent_max = max_new_len;
435 } else {
436 e3 = rb_entry(n3, struct e_node, count.node);
437 wnd->extent_max = max(e3->count.key, max_new_len);
438 }
439 }
440
441 if (!len) {
442 if (new_len) {
443 e->start.key = new_key;
444 rb_erase(&e->count.node, &wnd->count_tree);
445 e->count.key = new_len;
446 rb_insert_count(&wnd->count_tree, e);
447 } else {
448 rb_erase(&e->start.node, &wnd->start_tree);
449 rb_erase(&e->count.node, &wnd->count_tree);
450 wnd->count -= 1;
451 kmem_cache_free(ntfs_enode_cachep, e);
452 }
453 goto out;
454 }
455 rb_erase(&e->count.node, &wnd->count_tree);
456 e->count.key = len;
457 rb_insert_count(&wnd->count_tree, e);
458
459 if (!new_len)
460 goto out;
461
462 if (wnd->count >= NTFS_MAX_WND_EXTENTS) {
463 wnd->uptodated = -1;
464
465 /* Get minimal extent. */
466 e = rb_entry(rb_last(&wnd->count_tree), struct e_node,
467 count.node);
468 if (e->count.key > new_len)
469 goto out;
470
471 /* Replace minimum. */
472 rb_erase(&e->start.node, &wnd->start_tree);
473 rb_erase(&e->count.node, &wnd->count_tree);
474 wnd->count -= 1;
475 } else {
476 e = kmem_cache_alloc(ntfs_enode_cachep, GFP_ATOMIC);
477 if (!e)
478 wnd->uptodated = -1;
479 }
480
481 if (e) {
482 e->start.key = new_key;
483 e->count.key = new_len;
484 rb_insert_start(&wnd->start_tree, e);
485 rb_insert_count(&wnd->count_tree, e);
486 wnd->count += 1;
487 }
488
489out:
490 if (!wnd->count && 1 != wnd->uptodated)
491 wnd_rescan(wnd);
492}
493
494/*
495 * wnd_rescan - Scan all bitmap. Used while initialization.
496 */
497static int wnd_rescan(struct wnd_bitmap *wnd)
498{
499 int err = 0;
500 size_t prev_tail = 0;
501 struct super_block *sb = wnd->sb;
502 struct ntfs_sb_info *sbi = sb->s_fs_info;
503 u64 lbo, len = 0;
504 u32 blocksize = sb->s_blocksize;
505 u8 cluster_bits = sbi->cluster_bits;
506 u32 wbits = 8 * sb->s_blocksize;
507 u32 used, frb;
508 size_t wpos, wbit, iw, vbo;
509 struct buffer_head *bh = NULL;
510 CLST lcn, clen;
511
512 wnd->uptodated = 0;
513 wnd->extent_max = 0;
514 wnd->extent_min = MINUS_ONE_T;
515 wnd->total_zeroes = 0;
516
517 vbo = 0;
518
519 for (iw = 0; iw < wnd->nwnd; iw++) {
520 if (iw + 1 == wnd->nwnd)
521 wbits = wnd->bits_last;
522
523 if (wnd->inited) {
524 if (!wnd->free_bits[iw]) {
525 /* All ones. */
526 if (prev_tail) {
527 wnd_add_free_ext(wnd,
528 vbo * 8 - prev_tail,
529 prev_tail, true);
530 prev_tail = 0;
531 }
532 goto next_wnd;
533 }
534 if (wbits == wnd->free_bits[iw]) {
535 /* All zeroes. */
536 prev_tail += wbits;
537 wnd->total_zeroes += wbits;
538 goto next_wnd;
539 }
540 }
541
542 if (!len) {
543 u32 off = vbo & sbi->cluster_mask;
544
545 if (!run_lookup_entry(&wnd->run, vbo >> cluster_bits,
546 &lcn, &clen, NULL)) {
547 err = -ENOENT;
548 goto out;
549 }
550
551 lbo = ((u64)lcn << cluster_bits) + off;
552 len = ((u64)clen << cluster_bits) - off;
553 }
554
555 bh = ntfs_bread(sb, lbo >> sb->s_blocksize_bits);
556 if (!bh) {
557 err = -EIO;
558 goto out;
559 }
560
561 used = ntfs_bitmap_weight_le(bh->b_data, wbits);
562 if (used < wbits) {
563 frb = wbits - used;
564 wnd->free_bits[iw] = frb;
565 wnd->total_zeroes += frb;
566 }
567
568 wpos = 0;
569 wbit = vbo * 8;
570
571 if (wbit + wbits > wnd->nbits)
572 wbits = wnd->nbits - wbit;
573
574 do {
575 used = find_next_zero_bit_le(bh->b_data, wbits, wpos);
576
577 if (used > wpos && prev_tail) {
578 wnd_add_free_ext(wnd, wbit + wpos - prev_tail,
579 prev_tail, true);
580 prev_tail = 0;
581 }
582
583 wpos = used;
584
585 if (wpos >= wbits) {
586 /* No free blocks. */
587 prev_tail = 0;
588 break;
589 }
590
591 frb = find_next_bit_le(bh->b_data, wbits, wpos);
592 if (frb >= wbits) {
593 /* Keep last free block. */
594 prev_tail += frb - wpos;
595 break;
596 }
597
598 wnd_add_free_ext(wnd, wbit + wpos - prev_tail,
599 frb + prev_tail - wpos, true);
600
601 /* Skip free block and first '1'. */
602 wpos = frb + 1;
603 /* Reset previous tail. */
604 prev_tail = 0;
605 } while (wpos < wbits);
606
607next_wnd:
608
609 if (bh)
610 put_bh(bh);
611 bh = NULL;
612
613 vbo += blocksize;
614 if (len) {
615 len -= blocksize;
616 lbo += blocksize;
617 }
618 }
619
620 /* Add last block. */
621 if (prev_tail)
622 wnd_add_free_ext(wnd, wnd->nbits - prev_tail, prev_tail, true);
623
624 /*
625 * Before init cycle wnd->uptodated was 0.
626 * If any errors or limits occurs while initialization then
627 * wnd->uptodated will be -1.
628 * If 'uptodated' is still 0 then Tree is really updated.
629 */
630 if (!wnd->uptodated)
631 wnd->uptodated = 1;
632
633 if (wnd->zone_bit != wnd->zone_end) {
634 size_t zlen = wnd->zone_end - wnd->zone_bit;
635
636 wnd->zone_end = wnd->zone_bit;
637 wnd_zone_set(wnd, wnd->zone_bit, zlen);
638 }
639
640out:
641 return err;
642}
643
644int wnd_init(struct wnd_bitmap *wnd, struct super_block *sb, size_t nbits)
645{
646 int err;
647 u32 blocksize = sb->s_blocksize;
648 u32 wbits = blocksize * 8;
649
650 init_rwsem(&wnd->rw_lock);
651
652 wnd->sb = sb;
653 wnd->nbits = nbits;
654 wnd->total_zeroes = nbits;
655 wnd->extent_max = MINUS_ONE_T;
656 wnd->zone_bit = wnd->zone_end = 0;
657 wnd->nwnd = bytes_to_block(sb, ntfs3_bitmap_size(nbits));
658 wnd->bits_last = nbits & (wbits - 1);
659 if (!wnd->bits_last)
660 wnd->bits_last = wbits;
661
662 wnd->free_bits =
663 kvmalloc_array(wnd->nwnd, sizeof(u16), GFP_KERNEL | __GFP_ZERO);
664
665 if (!wnd->free_bits)
666 return -ENOMEM;
667
668 err = wnd_rescan(wnd);
669 if (err)
670 return err;
671
672 wnd->inited = true;
673
674 return 0;
675}
676
677/*
678 * wnd_map - Call sb_bread for requested window.
679 */
680static struct buffer_head *wnd_map(struct wnd_bitmap *wnd, size_t iw)
681{
682 size_t vbo;
683 CLST lcn, clen;
684 struct super_block *sb = wnd->sb;
685 struct ntfs_sb_info *sbi;
686 struct buffer_head *bh;
687 u64 lbo;
688
689 sbi = sb->s_fs_info;
690 vbo = (u64)iw << sb->s_blocksize_bits;
691
692 if (!run_lookup_entry(&wnd->run, vbo >> sbi->cluster_bits, &lcn, &clen,
693 NULL)) {
694 return ERR_PTR(-ENOENT);
695 }
696
697 lbo = ((u64)lcn << sbi->cluster_bits) + (vbo & sbi->cluster_mask);
698
699 bh = ntfs_bread(wnd->sb, lbo >> sb->s_blocksize_bits);
700 if (!bh)
701 return ERR_PTR(-EIO);
702
703 return bh;
704}
705
706/*
707 * wnd_set_free - Mark the bits range from bit to bit + bits as free.
708 */
709int wnd_set_free(struct wnd_bitmap *wnd, size_t bit, size_t bits)
710{
711 int err = 0;
712 struct super_block *sb = wnd->sb;
713 u32 wbits = 8 * sb->s_blocksize;
714 size_t iw = bit >> (sb->s_blocksize_bits + 3);
715 u32 wbit = bit & (wbits - 1);
716 struct buffer_head *bh;
717 u32 op;
718
719 for (; iw < wnd->nwnd && bits; iw++, bit += op, bits -= op, wbit = 0) {
720 if (iw + 1 == wnd->nwnd)
721 wbits = wnd->bits_last;
722
723 op = min_t(u32, wbits - wbit, bits);
724
725 bh = wnd_map(wnd, iw);
726 if (IS_ERR(bh)) {
727 err = PTR_ERR(bh);
728 break;
729 }
730
731 lock_buffer(bh);
732
733 ntfs_bitmap_clear_le(bh->b_data, wbit, op);
734
735 wnd->free_bits[iw] += op;
736 wnd->total_zeroes += op;
737
738 set_buffer_uptodate(bh);
739 mark_buffer_dirty(bh);
740 unlock_buffer(bh);
741 put_bh(bh);
742
743 wnd_add_free_ext(wnd, bit, op, false);
744 }
745 return err;
746}
747
748/*
749 * wnd_set_used - Mark the bits range from bit to bit + bits as used.
750 */
751int wnd_set_used(struct wnd_bitmap *wnd, size_t bit, size_t bits)
752{
753 int err = 0;
754 struct super_block *sb = wnd->sb;
755 size_t iw = bit >> (sb->s_blocksize_bits + 3);
756 u32 wbits = 8 * sb->s_blocksize;
757 u32 wbit = bit & (wbits - 1);
758 struct buffer_head *bh;
759 u32 op;
760
761 for (; iw < wnd->nwnd && bits; iw++, bit += op, bits -= op, wbit = 0) {
762 if (unlikely(iw + 1 == wnd->nwnd))
763 wbits = wnd->bits_last;
764
765 op = min_t(u32, wbits - wbit, bits);
766
767 bh = wnd_map(wnd, iw);
768 if (IS_ERR(bh)) {
769 err = PTR_ERR(bh);
770 break;
771 }
772
773 lock_buffer(bh);
774
775 ntfs_bitmap_set_le(bh->b_data, wbit, op);
776 wnd->free_bits[iw] -= op;
777 wnd->total_zeroes -= op;
778
779 set_buffer_uptodate(bh);
780 mark_buffer_dirty(bh);
781 unlock_buffer(bh);
782 put_bh(bh);
783
784 if (!RB_EMPTY_ROOT(&wnd->start_tree))
785 wnd_remove_free_ext(wnd, bit, op);
786 }
787 return err;
788}
789
790/*
791 * wnd_set_used_safe - Mark the bits range from bit to bit + bits as used.
792 *
793 * Unlikely wnd_set_used/wnd_set_free this function is not full trusted.
794 * It scans every bit in bitmap and marks free bit as used.
795 * @done - how many bits were marked as used.
796 *
797 * NOTE: normally *done should be 0.
798 */
799int wnd_set_used_safe(struct wnd_bitmap *wnd, size_t bit, size_t bits,
800 size_t *done)
801{
802 size_t i, from = 0, len = 0;
803 int err = 0;
804
805 *done = 0;
806 for (i = 0; i < bits; i++) {
807 if (wnd_is_free(wnd, bit + i, 1)) {
808 if (!len)
809 from = bit + i;
810 len += 1;
811 } else if (len) {
812 err = wnd_set_used(wnd, from, len);
813 *done += len;
814 len = 0;
815 if (err)
816 break;
817 }
818 }
819
820 if (len) {
821 /* last fragment. */
822 err = wnd_set_used(wnd, from, len);
823 *done += len;
824 }
825 return err;
826}
827
828/*
829 * wnd_is_free_hlp
830 *
831 * Return: True if all clusters [bit, bit+bits) are free (bitmap only).
832 */
833static bool wnd_is_free_hlp(struct wnd_bitmap *wnd, size_t bit, size_t bits)
834{
835 struct super_block *sb = wnd->sb;
836 size_t iw = bit >> (sb->s_blocksize_bits + 3);
837 u32 wbits = 8 * sb->s_blocksize;
838 u32 wbit = bit & (wbits - 1);
839 u32 op;
840
841 for (; iw < wnd->nwnd && bits; iw++, bits -= op, wbit = 0) {
842 if (unlikely(iw + 1 == wnd->nwnd))
843 wbits = wnd->bits_last;
844
845 op = min_t(u32, wbits - wbit, bits);
846
847 if (wbits != wnd->free_bits[iw]) {
848 bool ret;
849 struct buffer_head *bh = wnd_map(wnd, iw);
850
851 if (IS_ERR(bh))
852 return false;
853
854 ret = are_bits_clear(bh->b_data, wbit, op);
855
856 put_bh(bh);
857 if (!ret)
858 return false;
859 }
860 }
861
862 return true;
863}
864
865/*
866 * wnd_is_free
867 *
868 * Return: True if all clusters [bit, bit+bits) are free.
869 */
870bool wnd_is_free(struct wnd_bitmap *wnd, size_t bit, size_t bits)
871{
872 bool ret;
873 struct rb_node *n;
874 size_t end;
875 struct e_node *e;
876
877 if (RB_EMPTY_ROOT(&wnd->start_tree))
878 goto use_wnd;
879
880 n = rb_lookup(&wnd->start_tree, bit);
881 if (!n)
882 goto use_wnd;
883
884 e = rb_entry(n, struct e_node, start.node);
885
886 end = e->start.key + e->count.key;
887
888 if (bit < end && bit + bits <= end)
889 return true;
890
891use_wnd:
892 ret = wnd_is_free_hlp(wnd, bit, bits);
893
894 return ret;
895}
896
897/*
898 * wnd_is_used
899 *
900 * Return: True if all clusters [bit, bit+bits) are used.
901 */
902bool wnd_is_used(struct wnd_bitmap *wnd, size_t bit, size_t bits)
903{
904 bool ret = false;
905 struct super_block *sb = wnd->sb;
906 size_t iw = bit >> (sb->s_blocksize_bits + 3);
907 u32 wbits = 8 * sb->s_blocksize;
908 u32 wbit = bit & (wbits - 1);
909 u32 op;
910 size_t end;
911 struct rb_node *n;
912 struct e_node *e;
913
914 if (RB_EMPTY_ROOT(&wnd->start_tree))
915 goto use_wnd;
916
917 end = bit + bits;
918 n = rb_lookup(&wnd->start_tree, end - 1);
919 if (!n)
920 goto use_wnd;
921
922 e = rb_entry(n, struct e_node, start.node);
923 if (e->start.key + e->count.key > bit)
924 return false;
925
926use_wnd:
927 for (; iw < wnd->nwnd && bits; iw++, bits -= op, wbit = 0) {
928 if (unlikely(iw + 1 == wnd->nwnd))
929 wbits = wnd->bits_last;
930
931 op = min_t(u32, wbits - wbit, bits);
932
933 if (wnd->free_bits[iw]) {
934 bool ret;
935 struct buffer_head *bh = wnd_map(wnd, iw);
936
937 if (IS_ERR(bh))
938 goto out;
939
940 ret = are_bits_set(bh->b_data, wbit, op);
941 put_bh(bh);
942 if (!ret)
943 goto out;
944 }
945 }
946 ret = true;
947
948out:
949 return ret;
950}
951
952/*
953 * wnd_find - Look for free space.
954 *
955 * - flags - BITMAP_FIND_XXX flags
956 *
957 * Return: 0 if not found.
958 */
959size_t wnd_find(struct wnd_bitmap *wnd, size_t to_alloc, size_t hint,
960 size_t flags, size_t *allocated)
961{
962 struct super_block *sb;
963 u32 wbits, wpos, wzbit, wzend;
964 size_t fnd, max_alloc, b_len, b_pos;
965 size_t iw, prev_tail, nwnd, wbit, ebit, zbit, zend;
966 size_t to_alloc0 = to_alloc;
967 const struct e_node *e;
968 const struct rb_node *pr, *cr;
969 u8 log2_bits;
970 bool fbits_valid;
971 struct buffer_head *bh;
972
973 /* Fast checking for available free space. */
974 if (flags & BITMAP_FIND_FULL) {
975 size_t zeroes = wnd_zeroes(wnd);
976
977 zeroes -= wnd->zone_end - wnd->zone_bit;
978 if (zeroes < to_alloc0)
979 goto no_space;
980
981 if (to_alloc0 > wnd->extent_max)
982 goto no_space;
983 } else {
984 if (to_alloc > wnd->extent_max)
985 to_alloc = wnd->extent_max;
986 }
987
988 if (wnd->zone_bit <= hint && hint < wnd->zone_end)
989 hint = wnd->zone_end;
990
991 max_alloc = wnd->nbits;
992 b_len = b_pos = 0;
993
994 if (hint >= max_alloc)
995 hint = 0;
996
997 if (RB_EMPTY_ROOT(&wnd->start_tree)) {
998 if (wnd->uptodated == 1) {
999 /* Extents tree is updated -> No free space. */
1000 goto no_space;
1001 }
1002 goto scan_bitmap;
1003 }
1004
1005 e = NULL;
1006 if (!hint)
1007 goto allocate_biggest;
1008
1009 /* Use hint: Enumerate extents by start >= hint. */
1010 pr = NULL;
1011 cr = wnd->start_tree.rb_node;
1012
1013 for (;;) {
1014 e = rb_entry(cr, struct e_node, start.node);
1015
1016 if (e->start.key == hint)
1017 break;
1018
1019 if (e->start.key < hint) {
1020 pr = cr;
1021 cr = cr->rb_right;
1022 if (!cr)
1023 break;
1024 continue;
1025 }
1026
1027 cr = cr->rb_left;
1028 if (!cr) {
1029 e = pr ? rb_entry(pr, struct e_node, start.node) : NULL;
1030 break;
1031 }
1032 }
1033
1034 if (!e)
1035 goto allocate_biggest;
1036
1037 if (e->start.key + e->count.key > hint) {
1038 /* We have found extension with 'hint' inside. */
1039 size_t len = e->start.key + e->count.key - hint;
1040
1041 if (len >= to_alloc && hint + to_alloc <= max_alloc) {
1042 fnd = hint;
1043 goto found;
1044 }
1045
1046 if (!(flags & BITMAP_FIND_FULL)) {
1047 if (len > to_alloc)
1048 len = to_alloc;
1049
1050 if (hint + len <= max_alloc) {
1051 fnd = hint;
1052 to_alloc = len;
1053 goto found;
1054 }
1055 }
1056 }
1057
1058allocate_biggest:
1059 /* Allocate from biggest free extent. */
1060 e = rb_entry(rb_first(&wnd->count_tree), struct e_node, count.node);
1061 if (e->count.key != wnd->extent_max)
1062 wnd->extent_max = e->count.key;
1063
1064 if (e->count.key < max_alloc) {
1065 if (e->count.key >= to_alloc) {
1066 ;
1067 } else if (flags & BITMAP_FIND_FULL) {
1068 if (e->count.key < to_alloc0) {
1069 /* Biggest free block is less then requested. */
1070 goto no_space;
1071 }
1072 to_alloc = e->count.key;
1073 } else if (-1 != wnd->uptodated) {
1074 to_alloc = e->count.key;
1075 } else {
1076 /* Check if we can use more bits. */
1077 size_t op, max_check;
1078 struct rb_root start_tree;
1079
1080 memcpy(&start_tree, &wnd->start_tree,
1081 sizeof(struct rb_root));
1082 memset(&wnd->start_tree, 0, sizeof(struct rb_root));
1083
1084 max_check = e->start.key + to_alloc;
1085 if (max_check > max_alloc)
1086 max_check = max_alloc;
1087 for (op = e->start.key + e->count.key; op < max_check;
1088 op++) {
1089 if (!wnd_is_free(wnd, op, 1))
1090 break;
1091 }
1092 memcpy(&wnd->start_tree, &start_tree,
1093 sizeof(struct rb_root));
1094 to_alloc = op - e->start.key;
1095 }
1096
1097 /* Prepare to return. */
1098 fnd = e->start.key;
1099 if (e->start.key + to_alloc > max_alloc)
1100 to_alloc = max_alloc - e->start.key;
1101 goto found;
1102 }
1103
1104 if (wnd->uptodated == 1) {
1105 /* Extents tree is updated -> no free space. */
1106 goto no_space;
1107 }
1108
1109 b_len = e->count.key;
1110 b_pos = e->start.key;
1111
1112scan_bitmap:
1113 sb = wnd->sb;
1114 log2_bits = sb->s_blocksize_bits + 3;
1115
1116 /* At most two ranges [hint, max_alloc) + [0, hint). */
1117Again:
1118
1119 /* TODO: Optimize request for case nbits > wbits. */
1120 iw = hint >> log2_bits;
1121 wbits = sb->s_blocksize * 8;
1122 wpos = hint & (wbits - 1);
1123 prev_tail = 0;
1124 fbits_valid = true;
1125
1126 if (max_alloc == wnd->nbits) {
1127 nwnd = wnd->nwnd;
1128 } else {
1129 size_t t = max_alloc + wbits - 1;
1130
1131 nwnd = likely(t > max_alloc) ? (t >> log2_bits) : wnd->nwnd;
1132 }
1133
1134 /* Enumerate all windows. */
1135 for (; iw < nwnd; iw++) {
1136 wbit = iw << log2_bits;
1137
1138 if (!wnd->free_bits[iw]) {
1139 if (prev_tail > b_len) {
1140 b_pos = wbit - prev_tail;
1141 b_len = prev_tail;
1142 }
1143
1144 /* Skip full used window. */
1145 prev_tail = 0;
1146 wpos = 0;
1147 continue;
1148 }
1149
1150 if (unlikely(iw + 1 == nwnd)) {
1151 if (max_alloc == wnd->nbits) {
1152 wbits = wnd->bits_last;
1153 } else {
1154 size_t t = max_alloc & (wbits - 1);
1155
1156 if (t) {
1157 wbits = t;
1158 fbits_valid = false;
1159 }
1160 }
1161 }
1162
1163 if (wnd->zone_end > wnd->zone_bit) {
1164 ebit = wbit + wbits;
1165 zbit = max(wnd->zone_bit, wbit);
1166 zend = min(wnd->zone_end, ebit);
1167
1168 /* Here we have a window [wbit, ebit) and zone [zbit, zend). */
1169 if (zend <= zbit) {
1170 /* Zone does not overlap window. */
1171 } else {
1172 wzbit = zbit - wbit;
1173 wzend = zend - wbit;
1174
1175 /* Zone overlaps window. */
1176 if (wnd->free_bits[iw] == wzend - wzbit) {
1177 prev_tail = 0;
1178 wpos = 0;
1179 continue;
1180 }
1181
1182 /* Scan two ranges window: [wbit, zbit) and [zend, ebit). */
1183 bh = wnd_map(wnd, iw);
1184
1185 if (IS_ERR(bh)) {
1186 /* TODO: Error */
1187 prev_tail = 0;
1188 wpos = 0;
1189 continue;
1190 }
1191
1192 /* Scan range [wbit, zbit). */
1193 if (wpos < wzbit) {
1194 /* Scan range [wpos, zbit). */
1195 fnd = wnd_scan(bh->b_data, wbit, wpos,
1196 wzbit, to_alloc,
1197 &prev_tail, &b_pos,
1198 &b_len);
1199 if (fnd != MINUS_ONE_T) {
1200 put_bh(bh);
1201 goto found;
1202 }
1203 }
1204
1205 prev_tail = 0;
1206
1207 /* Scan range [zend, ebit). */
1208 if (wzend < wbits) {
1209 fnd = wnd_scan(bh->b_data, wbit,
1210 max(wzend, wpos), wbits,
1211 to_alloc, &prev_tail,
1212 &b_pos, &b_len);
1213 if (fnd != MINUS_ONE_T) {
1214 put_bh(bh);
1215 goto found;
1216 }
1217 }
1218
1219 wpos = 0;
1220 put_bh(bh);
1221 continue;
1222 }
1223 }
1224
1225 /* Current window does not overlap zone. */
1226 if (!wpos && fbits_valid && wnd->free_bits[iw] == wbits) {
1227 /* Window is empty. */
1228 if (prev_tail + wbits >= to_alloc) {
1229 fnd = wbit + wpos - prev_tail;
1230 goto found;
1231 }
1232
1233 /* Increase 'prev_tail' and process next window. */
1234 prev_tail += wbits;
1235 wpos = 0;
1236 continue;
1237 }
1238
1239 /* Read window. */
1240 bh = wnd_map(wnd, iw);
1241 if (IS_ERR(bh)) {
1242 // TODO: Error.
1243 prev_tail = 0;
1244 wpos = 0;
1245 continue;
1246 }
1247
1248 /* Scan range [wpos, eBits). */
1249 fnd = wnd_scan(bh->b_data, wbit, wpos, wbits, to_alloc,
1250 &prev_tail, &b_pos, &b_len);
1251 put_bh(bh);
1252 if (fnd != MINUS_ONE_T)
1253 goto found;
1254 }
1255
1256 if (b_len < prev_tail) {
1257 /* The last fragment. */
1258 b_len = prev_tail;
1259 b_pos = max_alloc - prev_tail;
1260 }
1261
1262 if (hint) {
1263 /*
1264 * We have scanned range [hint max_alloc).
1265 * Prepare to scan range [0 hint + to_alloc).
1266 */
1267 size_t nextmax = hint + to_alloc;
1268
1269 if (likely(nextmax >= hint) && nextmax < max_alloc)
1270 max_alloc = nextmax;
1271 hint = 0;
1272 goto Again;
1273 }
1274
1275 if (!b_len)
1276 goto no_space;
1277
1278 wnd->extent_max = b_len;
1279
1280 if (flags & BITMAP_FIND_FULL)
1281 goto no_space;
1282
1283 fnd = b_pos;
1284 to_alloc = b_len;
1285
1286found:
1287 if (flags & BITMAP_FIND_MARK_AS_USED) {
1288 /* TODO: Optimize remove extent (pass 'e'?). */
1289 if (wnd_set_used(wnd, fnd, to_alloc))
1290 goto no_space;
1291 } else if (wnd->extent_max != MINUS_ONE_T &&
1292 to_alloc > wnd->extent_max) {
1293 wnd->extent_max = to_alloc;
1294 }
1295
1296 *allocated = fnd;
1297 return to_alloc;
1298
1299no_space:
1300 return 0;
1301}
1302
1303/*
1304 * wnd_extend - Extend bitmap ($MFT bitmap).
1305 */
1306int wnd_extend(struct wnd_bitmap *wnd, size_t new_bits)
1307{
1308 int err;
1309 struct super_block *sb = wnd->sb;
1310 struct ntfs_sb_info *sbi = sb->s_fs_info;
1311 u32 blocksize = sb->s_blocksize;
1312 u32 wbits = blocksize * 8;
1313 u32 b0, new_last;
1314 size_t bits, iw, new_wnd;
1315 size_t old_bits = wnd->nbits;
1316 u16 *new_free;
1317
1318 if (new_bits <= old_bits)
1319 return -EINVAL;
1320
1321 /* Align to 8 byte boundary. */
1322 new_wnd = bytes_to_block(sb, ntfs3_bitmap_size(new_bits));
1323 new_last = new_bits & (wbits - 1);
1324 if (!new_last)
1325 new_last = wbits;
1326
1327 if (new_wnd != wnd->nwnd) {
1328 new_free = kmalloc_array(new_wnd, sizeof(u16), GFP_NOFS);
1329 if (!new_free)
1330 return -ENOMEM;
1331
1332 memcpy(new_free, wnd->free_bits, wnd->nwnd * sizeof(short));
1333 memset(new_free + wnd->nwnd, 0,
1334 (new_wnd - wnd->nwnd) * sizeof(short));
1335 kvfree(wnd->free_bits);
1336 wnd->free_bits = new_free;
1337 }
1338
1339 /* Zero bits [old_bits,new_bits). */
1340 bits = new_bits - old_bits;
1341 b0 = old_bits & (wbits - 1);
1342
1343 for (iw = old_bits >> (sb->s_blocksize_bits + 3); bits; iw += 1) {
1344 u32 op;
1345 size_t frb;
1346 u64 vbo, lbo, bytes;
1347 struct buffer_head *bh;
1348
1349 if (iw + 1 == new_wnd)
1350 wbits = new_last;
1351
1352 op = b0 + bits > wbits ? wbits - b0 : bits;
1353 vbo = (u64)iw * blocksize;
1354
1355 err = ntfs_vbo_to_lbo(sbi, &wnd->run, vbo, &lbo, &bytes);
1356 if (err)
1357 return err;
1358
1359 bh = ntfs_bread(sb, lbo >> sb->s_blocksize_bits);
1360 if (!bh)
1361 return -EIO;
1362
1363 lock_buffer(bh);
1364
1365 ntfs_bitmap_clear_le(bh->b_data, b0, blocksize * 8 - b0);
1366 frb = wbits - ntfs_bitmap_weight_le(bh->b_data, wbits);
1367 wnd->total_zeroes += frb - wnd->free_bits[iw];
1368 wnd->free_bits[iw] = frb;
1369
1370 set_buffer_uptodate(bh);
1371 mark_buffer_dirty(bh);
1372 unlock_buffer(bh);
1373 /* err = sync_dirty_buffer(bh); */
1374
1375 b0 = 0;
1376 bits -= op;
1377 }
1378
1379 wnd->nbits = new_bits;
1380 wnd->nwnd = new_wnd;
1381 wnd->bits_last = new_last;
1382
1383 wnd_add_free_ext(wnd, old_bits, new_bits - old_bits, false);
1384
1385 return 0;
1386}
1387
1388void wnd_zone_set(struct wnd_bitmap *wnd, size_t lcn, size_t len)
1389{
1390 size_t zlen = wnd->zone_end - wnd->zone_bit;
1391
1392 if (zlen)
1393 wnd_add_free_ext(wnd, wnd->zone_bit, zlen, false);
1394
1395 if (!RB_EMPTY_ROOT(&wnd->start_tree) && len)
1396 wnd_remove_free_ext(wnd, lcn, len);
1397
1398 wnd->zone_bit = lcn;
1399 wnd->zone_end = lcn + len;
1400}
1401
1402int ntfs_trim_fs(struct ntfs_sb_info *sbi, struct fstrim_range *range)
1403{
1404 int err = 0;
1405 struct super_block *sb = sbi->sb;
1406 struct wnd_bitmap *wnd = &sbi->used.bitmap;
1407 u32 wbits = 8 * sb->s_blocksize;
1408 CLST len = 0, lcn = 0, done = 0;
1409 CLST minlen = bytes_to_cluster(sbi, range->minlen);
1410 CLST lcn_from = bytes_to_cluster(sbi, range->start);
1411 size_t iw = lcn_from >> (sb->s_blocksize_bits + 3);
1412 u32 wbit = lcn_from & (wbits - 1);
1413 CLST lcn_to;
1414
1415 if (!minlen)
1416 minlen = 1;
1417
1418 if (range->len == (u64)-1)
1419 lcn_to = wnd->nbits;
1420 else
1421 lcn_to = bytes_to_cluster(sbi, range->start + range->len);
1422
1423 down_read_nested(&wnd->rw_lock, BITMAP_MUTEX_CLUSTERS);
1424
1425 for (; iw < wnd->nwnd; iw++, wbit = 0) {
1426 CLST lcn_wnd = iw * wbits;
1427 struct buffer_head *bh;
1428
1429 if (lcn_wnd > lcn_to)
1430 break;
1431
1432 if (!wnd->free_bits[iw])
1433 continue;
1434
1435 if (iw + 1 == wnd->nwnd)
1436 wbits = wnd->bits_last;
1437
1438 if (lcn_wnd + wbits > lcn_to)
1439 wbits = lcn_to - lcn_wnd;
1440
1441 bh = wnd_map(wnd, iw);
1442 if (IS_ERR(bh)) {
1443 err = PTR_ERR(bh);
1444 break;
1445 }
1446
1447 for (; wbit < wbits; wbit++) {
1448 if (!test_bit_le(wbit, bh->b_data)) {
1449 if (!len)
1450 lcn = lcn_wnd + wbit;
1451 len += 1;
1452 continue;
1453 }
1454 if (len >= minlen) {
1455 err = ntfs_discard(sbi, lcn, len);
1456 if (err)
1457 goto out;
1458 done += len;
1459 }
1460 len = 0;
1461 }
1462 put_bh(bh);
1463 }
1464
1465 /* Process the last fragment. */
1466 if (len >= minlen) {
1467 err = ntfs_discard(sbi, lcn, len);
1468 if (err)
1469 goto out;
1470 done += len;
1471 }
1472
1473out:
1474 range->len = (u64)done << sbi->cluster_bits;
1475
1476 up_read(&wnd->rw_lock);
1477
1478 return err;
1479}
1480
1481#if BITS_PER_LONG == 64
1482typedef __le64 bitmap_ulong;
1483#define cpu_to_ul(x) cpu_to_le64(x)
1484#define ul_to_cpu(x) le64_to_cpu(x)
1485#else
1486typedef __le32 bitmap_ulong;
1487#define cpu_to_ul(x) cpu_to_le32(x)
1488#define ul_to_cpu(x) le32_to_cpu(x)
1489#endif
1490
1491void ntfs_bitmap_set_le(void *map, unsigned int start, int len)
1492{
1493 bitmap_ulong *p = (bitmap_ulong *)map + BIT_WORD(start);
1494 const unsigned int size = start + len;
1495 int bits_to_set = BITS_PER_LONG - (start % BITS_PER_LONG);
1496 bitmap_ulong mask_to_set = cpu_to_ul(BITMAP_FIRST_WORD_MASK(start));
1497
1498 while (len - bits_to_set >= 0) {
1499 *p |= mask_to_set;
1500 len -= bits_to_set;
1501 bits_to_set = BITS_PER_LONG;
1502 mask_to_set = cpu_to_ul(~0UL);
1503 p++;
1504 }
1505 if (len) {
1506 mask_to_set &= cpu_to_ul(BITMAP_LAST_WORD_MASK(size));
1507 *p |= mask_to_set;
1508 }
1509}
1510
1511void ntfs_bitmap_clear_le(void *map, unsigned int start, int len)
1512{
1513 bitmap_ulong *p = (bitmap_ulong *)map + BIT_WORD(start);
1514 const unsigned int size = start + len;
1515 int bits_to_clear = BITS_PER_LONG - (start % BITS_PER_LONG);
1516 bitmap_ulong mask_to_clear = cpu_to_ul(BITMAP_FIRST_WORD_MASK(start));
1517
1518 while (len - bits_to_clear >= 0) {
1519 *p &= ~mask_to_clear;
1520 len -= bits_to_clear;
1521 bits_to_clear = BITS_PER_LONG;
1522 mask_to_clear = cpu_to_ul(~0UL);
1523 p++;
1524 }
1525 if (len) {
1526 mask_to_clear &= cpu_to_ul(BITMAP_LAST_WORD_MASK(size));
1527 *p &= ~mask_to_clear;
1528 }
1529}
1530
1531unsigned int ntfs_bitmap_weight_le(const void *bitmap, int bits)
1532{
1533 const ulong *bmp = bitmap;
1534 unsigned int k, lim = bits / BITS_PER_LONG;
1535 unsigned int w = 0;
1536
1537 for (k = 0; k < lim; k++)
1538 w += hweight_long(bmp[k]);
1539
1540 if (bits % BITS_PER_LONG) {
1541 w += hweight_long(ul_to_cpu(((bitmap_ulong *)bitmap)[k]) &
1542 BITMAP_LAST_WORD_MASK(bits));
1543 }
1544
1545 return w;
1546}