GNU Linux-libre 6.1.86-gnu
[releases.git] / fs / ntfs3 / bitmap.c
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
25 struct rb_node_key {
26         struct rb_node node;
27         size_t key;
28 };
29
30 struct e_node {
31         struct rb_node_key start; /* Tree sorted by start. */
32         struct rb_node_key count; /* Tree sorted by len. */
33 };
34
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);
38
39 static struct kmem_cache *ntfs_enode_cachep;
40
41 int __init ntfs3_init_bitmap(void)
42 {
43         ntfs_enode_cachep =
44                 kmem_cache_create("ntfs3_enode_cache", sizeof(struct e_node), 0,
45                                   SLAB_RECLAIM_ACCOUNT, NULL);
46         return ntfs_enode_cachep ? 0 : -ENOMEM;
47 }
48
49 void 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  */
62 static size_t wnd_scan(const ulong *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(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(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  */
123 void wnd_close(struct wnd_bitmap *wnd)
124 {
125         struct rb_node *node, *next;
126
127         kfree(wnd->free_bits);
128         run_close(&wnd->run);
129
130         node = rb_first(&wnd->start_tree);
131
132         while (node) {
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));
137                 node = next;
138         }
139 }
140
141 static struct rb_node *rb_lookup(struct rb_root *root, size_t v)
142 {
143         struct rb_node **p = &root->rb_node;
144         struct rb_node *r = NULL;
145
146         while (*p) {
147                 struct rb_node_key *k;
148
149                 k = rb_entry(*p, struct rb_node_key, node);
150                 if (v < k->key) {
151                         p = &(*p)->rb_left;
152                 } else if (v > k->key) {
153                         r = &k->node;
154                         p = &(*p)->rb_right;
155                 } else {
156                         return &k->node;
157                 }
158         }
159
160         return r;
161 }
162
163 /*
164  * rb_insert_count - Helper function to insert special kind of 'count' tree.
165  */
166 static inline bool rb_insert_count(struct rb_root *root, struct e_node *e)
167 {
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;
172
173         while (*p) {
174                 struct e_node *k =
175                         rb_entry(parent = *p, struct e_node, count.node);
176
177                 if (e_ckey > k->count.key) {
178                         p = &(*p)->rb_left;
179                 } else if (e_ckey < k->count.key) {
180                         p = &(*p)->rb_right;
181                 } else if (e_skey < k->start.key) {
182                         p = &(*p)->rb_left;
183                 } else if (e_skey > k->start.key) {
184                         p = &(*p)->rb_right;
185                 } else {
186                         WARN_ON(1);
187                         return false;
188                 }
189         }
190
191         rb_link_node(&e->count.node, parent, p);
192         rb_insert_color(&e->count.node, root);
193         return true;
194 }
195
196 /*
197  * rb_insert_start - Helper function to insert special kind of 'count' tree.
198  */
199 static inline bool rb_insert_start(struct rb_root *root, struct e_node *e)
200 {
201         struct rb_node **p = &root->rb_node;
202         struct rb_node *parent = NULL;
203         size_t e_skey = e->start.key;
204
205         while (*p) {
206                 struct e_node *k;
207
208                 parent = *p;
209
210                 k = rb_entry(parent, struct e_node, start.node);
211                 if (e_skey < k->start.key) {
212                         p = &(*p)->rb_left;
213                 } else if (e_skey > k->start.key) {
214                         p = &(*p)->rb_right;
215                 } else {
216                         WARN_ON(1);
217                         return false;
218                 }
219         }
220
221         rb_link_node(&e->start.node, parent, p);
222         rb_insert_color(&e->start.node, root);
223         return true;
224 }
225
226 /*
227  * wnd_add_free_ext - Adds a new extent of free space.
228  * @build:      1 when building tree.
229  */
230 static void wnd_add_free_ext(struct wnd_bitmap *wnd, size_t bit, size_t len,
231                              bool build)
232 {
233         struct e_node *e, *e0 = NULL;
234         size_t ib, end_in = bit + len;
235         struct rb_node *n;
236
237         if (build) {
238                 /* Use extent_min to filter too short extents. */
239                 if (wnd->count >= NTFS_MAX_WND_EXTENTS &&
240                     len <= wnd->extent_min) {
241                         wnd->uptodated = -1;
242                         return;
243                 }
244         } else {
245                 /* Try to find extent before 'bit'. */
246                 n = rb_lookup(&wnd->start_tree, bit);
247
248                 if (!n) {
249                         n = rb_first(&wnd->start_tree);
250                 } else {
251                         e = rb_entry(n, struct e_node, start.node);
252                         n = rb_next(n);
253                         if (e->start.key + e->count.key == bit) {
254                                 /* Remove left. */
255                                 bit = e->start.key;
256                                 len += e->count.key;
257                                 rb_erase(&e->start.node, &wnd->start_tree);
258                                 rb_erase(&e->count.node, &wnd->count_tree);
259                                 wnd->count -= 1;
260                                 e0 = e;
261                         }
262                 }
263
264                 while (n) {
265                         size_t next_end;
266
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)
270                                 break;
271
272                         /* Remove right. */
273                         n = rb_next(n);
274                         len += next_end - end_in;
275                         end_in = next_end;
276                         rb_erase(&e->start.node, &wnd->start_tree);
277                         rb_erase(&e->count.node, &wnd->count_tree);
278                         wnd->count -= 1;
279
280                         if (!e0)
281                                 e0 = e;
282                         else
283                                 kmem_cache_free(ntfs_enode_cachep, e);
284                 }
285
286                 if (wnd->uptodated != 1) {
287                         /* Check bits before 'bit'. */
288                         ib = wnd->zone_bit == wnd->zone_end ||
289                                              bit < wnd->zone_end
290                                      ? 0
291                                      : wnd->zone_end;
292
293                         while (bit > ib && wnd_is_free_hlp(wnd, bit - 1, 1)) {
294                                 bit -= 1;
295                                 len += 1;
296                         }
297
298                         /* Check bits after 'end_in'. */
299                         ib = wnd->zone_bit == wnd->zone_end ||
300                                              end_in > wnd->zone_bit
301                                      ? wnd->nbits
302                                      : wnd->zone_bit;
303
304                         while (end_in < ib && wnd_is_free_hlp(wnd, end_in, 1)) {
305                                 end_in += 1;
306                                 len += 1;
307                         }
308                 }
309         }
310         /* Insert new fragment. */
311         if (wnd->count >= NTFS_MAX_WND_EXTENTS) {
312                 if (e0)
313                         kmem_cache_free(ntfs_enode_cachep, e0);
314
315                 wnd->uptodated = -1;
316
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. */
322
323                 if (build) {
324                         struct e_node *e2;
325
326                         n = rb_prev(n);
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;
330                 }
331
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);
335                 wnd->count -= 1;
336         } else {
337                 e = e0 ? e0 : kmem_cache_alloc(ntfs_enode_cachep, GFP_ATOMIC);
338                 if (!e) {
339                         wnd->uptodated = -1;
340                         goto out;
341                 }
342
343                 if (build && len <= wnd->extent_min)
344                         wnd->extent_min = len;
345         }
346         e->start.key = bit;
347         e->count.key = len;
348         if (len > wnd->extent_max)
349                 wnd->extent_max = len;
350
351         rb_insert_start(&wnd->start_tree, e);
352         rb_insert_count(&wnd->count_tree, e);
353         wnd->count += 1;
354
355 out:;
356 }
357
358 /*
359  * wnd_remove_free_ext - Remove a run from the cached free space.
360  */
361 static void wnd_remove_free_ext(struct wnd_bitmap *wnd, size_t bit, size_t len)
362 {
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;
367
368         /* Try to find extent before 'bit'. */
369         n = rb_lookup(&wnd->start_tree, bit);
370
371         if (!n)
372                 return;
373
374         e = rb_entry(n, struct e_node, start.node);
375         end = e->start.key + e->count.key;
376
377         new_key = new_len = 0;
378         len = e->count.key;
379
380         /* Range [bit,end_in) must be inside 'e' or outside 'e' and 'n'. */
381         if (e->start.key > bit)
382                 ;
383         else if (end_in <= end) {
384                 /* Range [bit,end_in) inside 'e'. */
385                 new_key = end_in;
386                 new_len = end - end_in;
387                 len = bit - e->start.key;
388         } else if (bit > end) {
389                 bool bmax = false;
390
391                 n3 = rb_next(n);
392
393                 while (n3) {
394                         e3 = rb_entry(n3, struct e_node, start.node);
395                         if (e3->start.key >= end_in)
396                                 break;
397
398                         if (e3->count.key == wnd->extent_max)
399                                 bmax = true;
400
401                         end3 = e3->start.key + e3->count.key;
402                         if (end3 > end_in) {
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);
407                                 break;
408                         }
409
410                         n3 = rb_next(n3);
411                         rb_erase(&e3->start.node, &wnd->start_tree);
412                         rb_erase(&e3->count.node, &wnd->count_tree);
413                         wnd->count -= 1;
414                         kmem_cache_free(ntfs_enode_cachep, e3);
415                 }
416                 if (!bmax)
417                         return;
418                 n3 = rb_first(&wnd->count_tree);
419                 wnd->extent_max =
420                         n3 ? rb_entry(n3, struct e_node, count.node)->count.key
421                            : 0;
422                 return;
423         }
424
425         if (e->count.key != wnd->extent_max) {
426                 ;
427         } else if (rb_prev(&e->count.node)) {
428                 ;
429         } else {
430                 n3 = rb_next(&e->count.node);
431                 max_new_len = max(len, new_len);
432                 if (!n3) {
433                         wnd->extent_max = max_new_len;
434                 } else {
435                         e3 = rb_entry(n3, struct e_node, count.node);
436                         wnd->extent_max = max(e3->count.key, max_new_len);
437                 }
438         }
439
440         if (!len) {
441                 if (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);
446                 } else {
447                         rb_erase(&e->start.node, &wnd->start_tree);
448                         rb_erase(&e->count.node, &wnd->count_tree);
449                         wnd->count -= 1;
450                         kmem_cache_free(ntfs_enode_cachep, e);
451                 }
452                 goto out;
453         }
454         rb_erase(&e->count.node, &wnd->count_tree);
455         e->count.key = len;
456         rb_insert_count(&wnd->count_tree, e);
457
458         if (!new_len)
459                 goto out;
460
461         if (wnd->count >= NTFS_MAX_WND_EXTENTS) {
462                 wnd->uptodated = -1;
463
464                 /* Get minimal extent. */
465                 e = rb_entry(rb_last(&wnd->count_tree), struct e_node,
466                              count.node);
467                 if (e->count.key > new_len)
468                         goto out;
469
470                 /* Replace minimum. */
471                 rb_erase(&e->start.node, &wnd->start_tree);
472                 rb_erase(&e->count.node, &wnd->count_tree);
473                 wnd->count -= 1;
474         } else {
475                 e = kmem_cache_alloc(ntfs_enode_cachep, GFP_ATOMIC);
476                 if (!e)
477                         wnd->uptodated = -1;
478         }
479
480         if (e) {
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);
485                 wnd->count += 1;
486         }
487
488 out:
489         if (!wnd->count && 1 != wnd->uptodated)
490                 wnd_rescan(wnd);
491 }
492
493 /*
494  * wnd_rescan - Scan all bitmap. Used while initialization.
495  */
496 static int wnd_rescan(struct wnd_bitmap *wnd)
497 {
498         int err = 0;
499         size_t prev_tail = 0;
500         struct super_block *sb = wnd->sb;
501         struct ntfs_sb_info *sbi = sb->s_fs_info;
502         u64 lbo, len = 0;
503         u32 blocksize = sb->s_blocksize;
504         u8 cluster_bits = sbi->cluster_bits;
505         u32 wbits = 8 * sb->s_blocksize;
506         u32 used, frb;
507         const ulong *buf;
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                 buf = (ulong *)bh->b_data;
562
563                 used = bitmap_weight(buf, wbits);
564                 if (used < wbits) {
565                         frb = wbits - used;
566                         wnd->free_bits[iw] = frb;
567                         wnd->total_zeroes += frb;
568                 }
569
570                 wpos = 0;
571                 wbit = vbo * 8;
572
573                 if (wbit + wbits > wnd->nbits)
574                         wbits = wnd->nbits - wbit;
575
576                 do {
577                         used = find_next_zero_bit(buf, wbits, wpos);
578
579                         if (used > wpos && prev_tail) {
580                                 wnd_add_free_ext(wnd, wbit + wpos - prev_tail,
581                                                  prev_tail, true);
582                                 prev_tail = 0;
583                         }
584
585                         wpos = used;
586
587                         if (wpos >= wbits) {
588                                 /* No free blocks. */
589                                 prev_tail = 0;
590                                 break;
591                         }
592
593                         frb = find_next_bit(buf, wbits, wpos);
594                         if (frb >= wbits) {
595                                 /* Keep last free block. */
596                                 prev_tail += frb - wpos;
597                                 break;
598                         }
599
600                         wnd_add_free_ext(wnd, wbit + wpos - prev_tail,
601                                          frb + prev_tail - wpos, true);
602
603                         /* Skip free block and first '1'. */
604                         wpos = frb + 1;
605                         /* Reset previous tail. */
606                         prev_tail = 0;
607                 } while (wpos < wbits);
608
609 next_wnd:
610
611                 if (bh)
612                         put_bh(bh);
613                 bh = NULL;
614
615                 vbo += blocksize;
616                 if (len) {
617                         len -= blocksize;
618                         lbo += blocksize;
619                 }
620         }
621
622         /* Add last block. */
623         if (prev_tail)
624                 wnd_add_free_ext(wnd, wnd->nbits - prev_tail, prev_tail, true);
625
626         /*
627          * Before init cycle wnd->uptodated was 0.
628          * If any errors or limits occurs while initialization then
629          * wnd->uptodated will be -1.
630          * If 'uptodated' is still 0 then Tree is really updated.
631          */
632         if (!wnd->uptodated)
633                 wnd->uptodated = 1;
634
635         if (wnd->zone_bit != wnd->zone_end) {
636                 size_t zlen = wnd->zone_end - wnd->zone_bit;
637
638                 wnd->zone_end = wnd->zone_bit;
639                 wnd_zone_set(wnd, wnd->zone_bit, zlen);
640         }
641
642 out:
643         return err;
644 }
645
646 int wnd_init(struct wnd_bitmap *wnd, struct super_block *sb, size_t nbits)
647 {
648         int err;
649         u32 blocksize = sb->s_blocksize;
650         u32 wbits = blocksize * 8;
651
652         init_rwsem(&wnd->rw_lock);
653
654         wnd->sb = sb;
655         wnd->nbits = nbits;
656         wnd->total_zeroes = nbits;
657         wnd->extent_max = MINUS_ONE_T;
658         wnd->zone_bit = wnd->zone_end = 0;
659         wnd->nwnd = bytes_to_block(sb, bitmap_size(nbits));
660         wnd->bits_last = nbits & (wbits - 1);
661         if (!wnd->bits_last)
662                 wnd->bits_last = wbits;
663
664         wnd->free_bits =
665                 kvmalloc_array(wnd->nwnd, sizeof(u16), GFP_KERNEL | __GFP_ZERO);
666
667         if (!wnd->free_bits)
668                 return -ENOMEM;
669
670         err = wnd_rescan(wnd);
671         if (err)
672                 return err;
673
674         wnd->inited = true;
675
676         return 0;
677 }
678
679 /*
680  * wnd_map - Call sb_bread for requested window.
681  */
682 static struct buffer_head *wnd_map(struct wnd_bitmap *wnd, size_t iw)
683 {
684         size_t vbo;
685         CLST lcn, clen;
686         struct super_block *sb = wnd->sb;
687         struct ntfs_sb_info *sbi;
688         struct buffer_head *bh;
689         u64 lbo;
690
691         sbi = sb->s_fs_info;
692         vbo = (u64)iw << sb->s_blocksize_bits;
693
694         if (!run_lookup_entry(&wnd->run, vbo >> sbi->cluster_bits, &lcn, &clen,
695                               NULL)) {
696                 return ERR_PTR(-ENOENT);
697         }
698
699         lbo = ((u64)lcn << sbi->cluster_bits) + (vbo & sbi->cluster_mask);
700
701         bh = ntfs_bread(wnd->sb, lbo >> sb->s_blocksize_bits);
702         if (!bh)
703                 return ERR_PTR(-EIO);
704
705         return bh;
706 }
707
708 /*
709  * wnd_set_free - Mark the bits range from bit to bit + bits as free.
710  */
711 int wnd_set_free(struct wnd_bitmap *wnd, size_t bit, size_t bits)
712 {
713         int err = 0;
714         struct super_block *sb = wnd->sb;
715         size_t bits0 = bits;
716         u32 wbits = 8 * sb->s_blocksize;
717         size_t iw = bit >> (sb->s_blocksize_bits + 3);
718         u32 wbit = bit & (wbits - 1);
719         struct buffer_head *bh;
720
721         while (iw < wnd->nwnd && bits) {
722                 u32 tail, op;
723                 ulong *buf;
724
725                 if (iw + 1 == wnd->nwnd)
726                         wbits = wnd->bits_last;
727
728                 tail = wbits - wbit;
729                 op = min_t(u32, tail, bits);
730
731                 bh = wnd_map(wnd, iw);
732                 if (IS_ERR(bh)) {
733                         err = PTR_ERR(bh);
734                         break;
735                 }
736
737                 buf = (ulong *)bh->b_data;
738
739                 lock_buffer(bh);
740
741                 __bitmap_clear(buf, wbit, op);
742
743                 wnd->free_bits[iw] += op;
744
745                 set_buffer_uptodate(bh);
746                 mark_buffer_dirty(bh);
747                 unlock_buffer(bh);
748                 put_bh(bh);
749
750                 wnd->total_zeroes += op;
751                 bits -= op;
752                 wbit = 0;
753                 iw += 1;
754         }
755
756         wnd_add_free_ext(wnd, bit, bits0, false);
757
758         return err;
759 }
760
761 /*
762  * wnd_set_used - Mark the bits range from bit to bit + bits as used.
763  */
764 int wnd_set_used(struct wnd_bitmap *wnd, size_t bit, size_t bits)
765 {
766         int err = 0;
767         struct super_block *sb = wnd->sb;
768         size_t bits0 = bits;
769         size_t iw = bit >> (sb->s_blocksize_bits + 3);
770         u32 wbits = 8 * sb->s_blocksize;
771         u32 wbit = bit & (wbits - 1);
772         struct buffer_head *bh;
773
774         while (iw < wnd->nwnd && bits) {
775                 u32 tail, op;
776                 ulong *buf;
777
778                 if (unlikely(iw + 1 == wnd->nwnd))
779                         wbits = wnd->bits_last;
780
781                 tail = wbits - wbit;
782                 op = min_t(u32, tail, bits);
783
784                 bh = wnd_map(wnd, iw);
785                 if (IS_ERR(bh)) {
786                         err = PTR_ERR(bh);
787                         break;
788                 }
789                 buf = (ulong *)bh->b_data;
790
791                 lock_buffer(bh);
792
793                 __bitmap_set(buf, wbit, op);
794                 wnd->free_bits[iw] -= op;
795
796                 set_buffer_uptodate(bh);
797                 mark_buffer_dirty(bh);
798                 unlock_buffer(bh);
799                 put_bh(bh);
800
801                 wnd->total_zeroes -= op;
802                 bits -= op;
803                 wbit = 0;
804                 iw += 1;
805         }
806
807         if (!RB_EMPTY_ROOT(&wnd->start_tree))
808                 wnd_remove_free_ext(wnd, bit, bits0);
809
810         return err;
811 }
812
813 /*
814  * wnd_is_free_hlp
815  *
816  * Return: True if all clusters [bit, bit+bits) are free (bitmap only).
817  */
818 static bool wnd_is_free_hlp(struct wnd_bitmap *wnd, size_t bit, size_t bits)
819 {
820         struct super_block *sb = wnd->sb;
821         size_t iw = bit >> (sb->s_blocksize_bits + 3);
822         u32 wbits = 8 * sb->s_blocksize;
823         u32 wbit = bit & (wbits - 1);
824
825         while (iw < wnd->nwnd && bits) {
826                 u32 tail, op;
827
828                 if (unlikely(iw + 1 == wnd->nwnd))
829                         wbits = wnd->bits_last;
830
831                 tail = wbits - wbit;
832                 op = min_t(u32, tail, bits);
833
834                 if (wbits != wnd->free_bits[iw]) {
835                         bool ret;
836                         struct buffer_head *bh = wnd_map(wnd, iw);
837
838                         if (IS_ERR(bh))
839                                 return false;
840
841                         ret = are_bits_clear((ulong *)bh->b_data, wbit, op);
842
843                         put_bh(bh);
844                         if (!ret)
845                                 return false;
846                 }
847
848                 bits -= op;
849                 wbit = 0;
850                 iw += 1;
851         }
852
853         return true;
854 }
855
856 /*
857  * wnd_is_free
858  *
859  * Return: True if all clusters [bit, bit+bits) are free.
860  */
861 bool wnd_is_free(struct wnd_bitmap *wnd, size_t bit, size_t bits)
862 {
863         bool ret;
864         struct rb_node *n;
865         size_t end;
866         struct e_node *e;
867
868         if (RB_EMPTY_ROOT(&wnd->start_tree))
869                 goto use_wnd;
870
871         n = rb_lookup(&wnd->start_tree, bit);
872         if (!n)
873                 goto use_wnd;
874
875         e = rb_entry(n, struct e_node, start.node);
876
877         end = e->start.key + e->count.key;
878
879         if (bit < end && bit + bits <= end)
880                 return true;
881
882 use_wnd:
883         ret = wnd_is_free_hlp(wnd, bit, bits);
884
885         return ret;
886 }
887
888 /*
889  * wnd_is_used
890  *
891  * Return: True if all clusters [bit, bit+bits) are used.
892  */
893 bool wnd_is_used(struct wnd_bitmap *wnd, size_t bit, size_t bits)
894 {
895         bool ret = false;
896         struct super_block *sb = wnd->sb;
897         size_t iw = bit >> (sb->s_blocksize_bits + 3);
898         u32 wbits = 8 * sb->s_blocksize;
899         u32 wbit = bit & (wbits - 1);
900         size_t end;
901         struct rb_node *n;
902         struct e_node *e;
903
904         if (RB_EMPTY_ROOT(&wnd->start_tree))
905                 goto use_wnd;
906
907         end = bit + bits;
908         n = rb_lookup(&wnd->start_tree, end - 1);
909         if (!n)
910                 goto use_wnd;
911
912         e = rb_entry(n, struct e_node, start.node);
913         if (e->start.key + e->count.key > bit)
914                 return false;
915
916 use_wnd:
917         while (iw < wnd->nwnd && bits) {
918                 u32 tail, op;
919
920                 if (unlikely(iw + 1 == wnd->nwnd))
921                         wbits = wnd->bits_last;
922
923                 tail = wbits - wbit;
924                 op = min_t(u32, tail, bits);
925
926                 if (wnd->free_bits[iw]) {
927                         bool ret;
928                         struct buffer_head *bh = wnd_map(wnd, iw);
929
930                         if (IS_ERR(bh))
931                                 goto out;
932
933                         ret = are_bits_set((ulong *)bh->b_data, wbit, op);
934                         put_bh(bh);
935                         if (!ret)
936                                 goto out;
937                 }
938
939                 bits -= op;
940                 wbit = 0;
941                 iw += 1;
942         }
943         ret = true;
944
945 out:
946         return ret;
947 }
948
949 /*
950  * wnd_find - Look for free space.
951  *
952  * - flags - BITMAP_FIND_XXX flags
953  *
954  * Return: 0 if not found.
955  */
956 size_t wnd_find(struct wnd_bitmap *wnd, size_t to_alloc, size_t hint,
957                 size_t flags, size_t *allocated)
958 {
959         struct super_block *sb;
960         u32 wbits, wpos, wzbit, wzend;
961         size_t fnd, max_alloc, b_len, b_pos;
962         size_t iw, prev_tail, nwnd, wbit, ebit, zbit, zend;
963         size_t to_alloc0 = to_alloc;
964         const ulong *buf;
965         const struct e_node *e;
966         const struct rb_node *pr, *cr;
967         u8 log2_bits;
968         bool fbits_valid;
969         struct buffer_head *bh;
970
971         /* Fast checking for available free space. */
972         if (flags & BITMAP_FIND_FULL) {
973                 size_t zeroes = wnd_zeroes(wnd);
974
975                 zeroes -= wnd->zone_end - wnd->zone_bit;
976                 if (zeroes < to_alloc0)
977                         goto no_space;
978
979                 if (to_alloc0 > wnd->extent_max)
980                         goto no_space;
981         } else {
982                 if (to_alloc > wnd->extent_max)
983                         to_alloc = wnd->extent_max;
984         }
985
986         if (wnd->zone_bit <= hint && hint < wnd->zone_end)
987                 hint = wnd->zone_end;
988
989         max_alloc = wnd->nbits;
990         b_len = b_pos = 0;
991
992         if (hint >= max_alloc)
993                 hint = 0;
994
995         if (RB_EMPTY_ROOT(&wnd->start_tree)) {
996                 if (wnd->uptodated == 1) {
997                         /* Extents tree is updated -> No free space. */
998                         goto no_space;
999                 }
1000                 goto scan_bitmap;
1001         }
1002
1003         e = NULL;
1004         if (!hint)
1005                 goto allocate_biggest;
1006
1007         /* Use hint: Enumerate extents by start >= hint. */
1008         pr = NULL;
1009         cr = wnd->start_tree.rb_node;
1010
1011         for (;;) {
1012                 e = rb_entry(cr, struct e_node, start.node);
1013
1014                 if (e->start.key == hint)
1015                         break;
1016
1017                 if (e->start.key < hint) {
1018                         pr = cr;
1019                         cr = cr->rb_right;
1020                         if (!cr)
1021                                 break;
1022                         continue;
1023                 }
1024
1025                 cr = cr->rb_left;
1026                 if (!cr) {
1027                         e = pr ? rb_entry(pr, struct e_node, start.node) : NULL;
1028                         break;
1029                 }
1030         }
1031
1032         if (!e)
1033                 goto allocate_biggest;
1034
1035         if (e->start.key + e->count.key > hint) {
1036                 /* We have found extension with 'hint' inside. */
1037                 size_t len = e->start.key + e->count.key - hint;
1038
1039                 if (len >= to_alloc && hint + to_alloc <= max_alloc) {
1040                         fnd = hint;
1041                         goto found;
1042                 }
1043
1044                 if (!(flags & BITMAP_FIND_FULL)) {
1045                         if (len > to_alloc)
1046                                 len = to_alloc;
1047
1048                         if (hint + len <= max_alloc) {
1049                                 fnd = hint;
1050                                 to_alloc = len;
1051                                 goto found;
1052                         }
1053                 }
1054         }
1055
1056 allocate_biggest:
1057         /* Allocate from biggest free extent. */
1058         e = rb_entry(rb_first(&wnd->count_tree), struct e_node, count.node);
1059         if (e->count.key != wnd->extent_max)
1060                 wnd->extent_max = e->count.key;
1061
1062         if (e->count.key < max_alloc) {
1063                 if (e->count.key >= to_alloc) {
1064                         ;
1065                 } else if (flags & BITMAP_FIND_FULL) {
1066                         if (e->count.key < to_alloc0) {
1067                                 /* Biggest free block is less then requested. */
1068                                 goto no_space;
1069                         }
1070                         to_alloc = e->count.key;
1071                 } else if (-1 != wnd->uptodated) {
1072                         to_alloc = e->count.key;
1073                 } else {
1074                         /* Check if we can use more bits. */
1075                         size_t op, max_check;
1076                         struct rb_root start_tree;
1077
1078                         memcpy(&start_tree, &wnd->start_tree,
1079                                sizeof(struct rb_root));
1080                         memset(&wnd->start_tree, 0, sizeof(struct rb_root));
1081
1082                         max_check = e->start.key + to_alloc;
1083                         if (max_check > max_alloc)
1084                                 max_check = max_alloc;
1085                         for (op = e->start.key + e->count.key; op < max_check;
1086                              op++) {
1087                                 if (!wnd_is_free(wnd, op, 1))
1088                                         break;
1089                         }
1090                         memcpy(&wnd->start_tree, &start_tree,
1091                                sizeof(struct rb_root));
1092                         to_alloc = op - e->start.key;
1093                 }
1094
1095                 /* Prepare to return. */
1096                 fnd = e->start.key;
1097                 if (e->start.key + to_alloc > max_alloc)
1098                         to_alloc = max_alloc - e->start.key;
1099                 goto found;
1100         }
1101
1102         if (wnd->uptodated == 1) {
1103                 /* Extents tree is updated -> no free space. */
1104                 goto no_space;
1105         }
1106
1107         b_len = e->count.key;
1108         b_pos = e->start.key;
1109
1110 scan_bitmap:
1111         sb = wnd->sb;
1112         log2_bits = sb->s_blocksize_bits + 3;
1113
1114         /* At most two ranges [hint, max_alloc) + [0, hint). */
1115 Again:
1116
1117         /* TODO: Optimize request for case nbits > wbits. */
1118         iw = hint >> log2_bits;
1119         wbits = sb->s_blocksize * 8;
1120         wpos = hint & (wbits - 1);
1121         prev_tail = 0;
1122         fbits_valid = true;
1123
1124         if (max_alloc == wnd->nbits) {
1125                 nwnd = wnd->nwnd;
1126         } else {
1127                 size_t t = max_alloc + wbits - 1;
1128
1129                 nwnd = likely(t > max_alloc) ? (t >> log2_bits) : wnd->nwnd;
1130         }
1131
1132         /* Enumerate all windows. */
1133         for (; iw < nwnd; iw++) {
1134                 wbit = iw << log2_bits;
1135
1136                 if (!wnd->free_bits[iw]) {
1137                         if (prev_tail > b_len) {
1138                                 b_pos = wbit - prev_tail;
1139                                 b_len = prev_tail;
1140                         }
1141
1142                         /* Skip full used window. */
1143                         prev_tail = 0;
1144                         wpos = 0;
1145                         continue;
1146                 }
1147
1148                 if (unlikely(iw + 1 == nwnd)) {
1149                         if (max_alloc == wnd->nbits) {
1150                                 wbits = wnd->bits_last;
1151                         } else {
1152                                 size_t t = max_alloc & (wbits - 1);
1153
1154                                 if (t) {
1155                                         wbits = t;
1156                                         fbits_valid = false;
1157                                 }
1158                         }
1159                 }
1160
1161                 if (wnd->zone_end > wnd->zone_bit) {
1162                         ebit = wbit + wbits;
1163                         zbit = max(wnd->zone_bit, wbit);
1164                         zend = min(wnd->zone_end, ebit);
1165
1166                         /* Here we have a window [wbit, ebit) and zone [zbit, zend). */
1167                         if (zend <= zbit) {
1168                                 /* Zone does not overlap window. */
1169                         } else {
1170                                 wzbit = zbit - wbit;
1171                                 wzend = zend - wbit;
1172
1173                                 /* Zone overlaps window. */
1174                                 if (wnd->free_bits[iw] == wzend - wzbit) {
1175                                         prev_tail = 0;
1176                                         wpos = 0;
1177                                         continue;
1178                                 }
1179
1180                                 /* Scan two ranges window: [wbit, zbit) and [zend, ebit). */
1181                                 bh = wnd_map(wnd, iw);
1182
1183                                 if (IS_ERR(bh)) {
1184                                         /* TODO: Error */
1185                                         prev_tail = 0;
1186                                         wpos = 0;
1187                                         continue;
1188                                 }
1189
1190                                 buf = (ulong *)bh->b_data;
1191
1192                                 /* Scan range [wbit, zbit). */
1193                                 if (wpos < wzbit) {
1194                                         /* Scan range [wpos, zbit). */
1195                                         fnd = wnd_scan(buf, wbit, wpos, wzbit,
1196                                                        to_alloc, &prev_tail,
1197                                                        &b_pos, &b_len);
1198                                         if (fnd != MINUS_ONE_T) {
1199                                                 put_bh(bh);
1200                                                 goto found;
1201                                         }
1202                                 }
1203
1204                                 prev_tail = 0;
1205
1206                                 /* Scan range [zend, ebit). */
1207                                 if (wzend < wbits) {
1208                                         fnd = wnd_scan(buf, wbit,
1209                                                        max(wzend, wpos), wbits,
1210                                                        to_alloc, &prev_tail,
1211                                                        &b_pos, &b_len);
1212                                         if (fnd != MINUS_ONE_T) {
1213                                                 put_bh(bh);
1214                                                 goto found;
1215                                         }
1216                                 }
1217
1218                                 wpos = 0;
1219                                 put_bh(bh);
1220                                 continue;
1221                         }
1222                 }
1223
1224                 /* Current window does not overlap zone. */
1225                 if (!wpos && fbits_valid && wnd->free_bits[iw] == wbits) {
1226                         /* Window is empty. */
1227                         if (prev_tail + wbits >= to_alloc) {
1228                                 fnd = wbit + wpos - prev_tail;
1229                                 goto found;
1230                         }
1231
1232                         /* Increase 'prev_tail' and process next window. */
1233                         prev_tail += wbits;
1234                         wpos = 0;
1235                         continue;
1236                 }
1237
1238                 /* Read window. */
1239                 bh = wnd_map(wnd, iw);
1240                 if (IS_ERR(bh)) {
1241                         // TODO: Error.
1242                         prev_tail = 0;
1243                         wpos = 0;
1244                         continue;
1245                 }
1246
1247                 buf = (ulong *)bh->b_data;
1248
1249                 /* Scan range [wpos, eBits). */
1250                 fnd = wnd_scan(buf, wbit, wpos, wbits, to_alloc, &prev_tail,
1251                                &b_pos, &b_len);
1252                 put_bh(bh);
1253                 if (fnd != MINUS_ONE_T)
1254                         goto found;
1255         }
1256
1257         if (b_len < prev_tail) {
1258                 /* The last fragment. */
1259                 b_len = prev_tail;
1260                 b_pos = max_alloc - prev_tail;
1261         }
1262
1263         if (hint) {
1264                 /*
1265                  * We have scanned range [hint max_alloc).
1266                  * Prepare to scan range [0 hint + to_alloc).
1267                  */
1268                 size_t nextmax = hint + to_alloc;
1269
1270                 if (likely(nextmax >= hint) && nextmax < max_alloc)
1271                         max_alloc = nextmax;
1272                 hint = 0;
1273                 goto Again;
1274         }
1275
1276         if (!b_len)
1277                 goto no_space;
1278
1279         wnd->extent_max = b_len;
1280
1281         if (flags & BITMAP_FIND_FULL)
1282                 goto no_space;
1283
1284         fnd = b_pos;
1285         to_alloc = b_len;
1286
1287 found:
1288         if (flags & BITMAP_FIND_MARK_AS_USED) {
1289                 /* TODO: Optimize remove extent (pass 'e'?). */
1290                 if (wnd_set_used(wnd, fnd, to_alloc))
1291                         goto no_space;
1292         } else if (wnd->extent_max != MINUS_ONE_T &&
1293                    to_alloc > wnd->extent_max) {
1294                 wnd->extent_max = to_alloc;
1295         }
1296
1297         *allocated = fnd;
1298         return to_alloc;
1299
1300 no_space:
1301         return 0;
1302 }
1303
1304 /*
1305  * wnd_extend - Extend bitmap ($MFT bitmap).
1306  */
1307 int wnd_extend(struct wnd_bitmap *wnd, size_t new_bits)
1308 {
1309         int err;
1310         struct super_block *sb = wnd->sb;
1311         struct ntfs_sb_info *sbi = sb->s_fs_info;
1312         u32 blocksize = sb->s_blocksize;
1313         u32 wbits = blocksize * 8;
1314         u32 b0, new_last;
1315         size_t bits, iw, new_wnd;
1316         size_t old_bits = wnd->nbits;
1317         u16 *new_free;
1318
1319         if (new_bits <= old_bits)
1320                 return -EINVAL;
1321
1322         /* Align to 8 byte boundary. */
1323         new_wnd = bytes_to_block(sb, bitmap_size(new_bits));
1324         new_last = new_bits & (wbits - 1);
1325         if (!new_last)
1326                 new_last = wbits;
1327
1328         if (new_wnd != wnd->nwnd) {
1329                 new_free = kmalloc(new_wnd * sizeof(u16), GFP_NOFS);
1330                 if (!new_free)
1331                         return -ENOMEM;
1332
1333                 memcpy(new_free, wnd->free_bits, wnd->nwnd * sizeof(short));
1334                 memset(new_free + wnd->nwnd, 0,
1335                        (new_wnd - wnd->nwnd) * sizeof(short));
1336                 kfree(wnd->free_bits);
1337                 wnd->free_bits = new_free;
1338         }
1339
1340         /* Zero bits [old_bits,new_bits). */
1341         bits = new_bits - old_bits;
1342         b0 = old_bits & (wbits - 1);
1343
1344         for (iw = old_bits >> (sb->s_blocksize_bits + 3); bits; iw += 1) {
1345                 u32 op;
1346                 size_t frb;
1347                 u64 vbo, lbo, bytes;
1348                 struct buffer_head *bh;
1349                 ulong *buf;
1350
1351                 if (iw + 1 == new_wnd)
1352                         wbits = new_last;
1353
1354                 op = b0 + bits > wbits ? wbits - b0 : bits;
1355                 vbo = (u64)iw * blocksize;
1356
1357                 err = ntfs_vbo_to_lbo(sbi, &wnd->run, vbo, &lbo, &bytes);
1358                 if (err)
1359                         break;
1360
1361                 bh = ntfs_bread(sb, lbo >> sb->s_blocksize_bits);
1362                 if (!bh)
1363                         return -EIO;
1364
1365                 lock_buffer(bh);
1366                 buf = (ulong *)bh->b_data;
1367
1368                 __bitmap_clear(buf, b0, blocksize * 8 - b0);
1369                 frb = wbits - bitmap_weight(buf, wbits);
1370                 wnd->total_zeroes += frb - wnd->free_bits[iw];
1371                 wnd->free_bits[iw] = frb;
1372
1373                 set_buffer_uptodate(bh);
1374                 mark_buffer_dirty(bh);
1375                 unlock_buffer(bh);
1376                 /* err = sync_dirty_buffer(bh); */
1377
1378                 b0 = 0;
1379                 bits -= op;
1380         }
1381
1382         wnd->nbits = new_bits;
1383         wnd->nwnd = new_wnd;
1384         wnd->bits_last = new_last;
1385
1386         wnd_add_free_ext(wnd, old_bits, new_bits - old_bits, false);
1387
1388         return 0;
1389 }
1390
1391 void wnd_zone_set(struct wnd_bitmap *wnd, size_t lcn, size_t len)
1392 {
1393         size_t zlen = wnd->zone_end - wnd->zone_bit;
1394
1395         if (zlen)
1396                 wnd_add_free_ext(wnd, wnd->zone_bit, zlen, false);
1397
1398         if (!RB_EMPTY_ROOT(&wnd->start_tree) && len)
1399                 wnd_remove_free_ext(wnd, lcn, len);
1400
1401         wnd->zone_bit = lcn;
1402         wnd->zone_end = lcn + len;
1403 }
1404
1405 int ntfs_trim_fs(struct ntfs_sb_info *sbi, struct fstrim_range *range)
1406 {
1407         int err = 0;
1408         struct super_block *sb = sbi->sb;
1409         struct wnd_bitmap *wnd = &sbi->used.bitmap;
1410         u32 wbits = 8 * sb->s_blocksize;
1411         CLST len = 0, lcn = 0, done = 0;
1412         CLST minlen = bytes_to_cluster(sbi, range->minlen);
1413         CLST lcn_from = bytes_to_cluster(sbi, range->start);
1414         size_t iw = lcn_from >> (sb->s_blocksize_bits + 3);
1415         u32 wbit = lcn_from & (wbits - 1);
1416         const ulong *buf;
1417         CLST lcn_to;
1418
1419         if (!minlen)
1420                 minlen = 1;
1421
1422         if (range->len == (u64)-1)
1423                 lcn_to = wnd->nbits;
1424         else
1425                 lcn_to = bytes_to_cluster(sbi, range->start + range->len);
1426
1427         down_read_nested(&wnd->rw_lock, BITMAP_MUTEX_CLUSTERS);
1428
1429         for (; iw < wnd->nwnd; iw++, wbit = 0) {
1430                 CLST lcn_wnd = iw * wbits;
1431                 struct buffer_head *bh;
1432
1433                 if (lcn_wnd > lcn_to)
1434                         break;
1435
1436                 if (!wnd->free_bits[iw])
1437                         continue;
1438
1439                 if (iw + 1 == wnd->nwnd)
1440                         wbits = wnd->bits_last;
1441
1442                 if (lcn_wnd + wbits > lcn_to)
1443                         wbits = lcn_to - lcn_wnd;
1444
1445                 bh = wnd_map(wnd, iw);
1446                 if (IS_ERR(bh)) {
1447                         err = PTR_ERR(bh);
1448                         break;
1449                 }
1450
1451                 buf = (ulong *)bh->b_data;
1452
1453                 for (; wbit < wbits; wbit++) {
1454                         if (!test_bit(wbit, buf)) {
1455                                 if (!len)
1456                                         lcn = lcn_wnd + wbit;
1457                                 len += 1;
1458                                 continue;
1459                         }
1460                         if (len >= minlen) {
1461                                 err = ntfs_discard(sbi, lcn, len);
1462                                 if (err)
1463                                         goto out;
1464                                 done += len;
1465                         }
1466                         len = 0;
1467                 }
1468                 put_bh(bh);
1469         }
1470
1471         /* Process the last fragment. */
1472         if (len >= minlen) {
1473                 err = ntfs_discard(sbi, lcn, len);
1474                 if (err)
1475                         goto out;
1476                 done += len;
1477         }
1478
1479 out:
1480         range->len = (u64)done << sbi->cluster_bits;
1481
1482         up_read(&wnd->rw_lock);
1483
1484         return err;
1485 }