1
2
3
4
5
6
7
8#include <linux/bio.h>
9#include <linux/bitmap.h>
10#include <linux/err.h>
11#include <linux/init.h>
12#include <linux/kernel.h>
13#include <linux/mm.h>
14#include <linux/sched/mm.h>
15#include <linux/pagemap.h>
16#include <linux/refcount.h>
17#include <linux/sched.h>
18#include <linux/slab.h>
19#include <linux/zstd.h>
20#include "compression.h"
21#include "ctree.h"
22
23#define ZSTD_BTRFS_MAX_WINDOWLOG 17
24#define ZSTD_BTRFS_MAX_INPUT (1 << ZSTD_BTRFS_MAX_WINDOWLOG)
25#define ZSTD_BTRFS_DEFAULT_LEVEL 3
26#define ZSTD_BTRFS_MAX_LEVEL 15
27
28#define ZSTD_BTRFS_RECLAIM_JIFFIES (307 * HZ)
29
30static ZSTD_parameters zstd_get_btrfs_parameters(unsigned int level,
31 size_t src_len)
32{
33 ZSTD_parameters params = ZSTD_getParams(level, src_len, 0);
34
35 if (params.cParams.windowLog > ZSTD_BTRFS_MAX_WINDOWLOG)
36 params.cParams.windowLog = ZSTD_BTRFS_MAX_WINDOWLOG;
37 WARN_ON(src_len > ZSTD_BTRFS_MAX_INPUT);
38 return params;
39}
40
41struct workspace {
42 void *mem;
43 size_t size;
44 char *buf;
45 unsigned int level;
46 unsigned int req_level;
47 unsigned long last_used;
48 struct list_head list;
49 struct list_head lru_list;
50 ZSTD_inBuffer in_buf;
51 ZSTD_outBuffer out_buf;
52};
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74struct zstd_workspace_manager {
75 const struct btrfs_compress_op *ops;
76 spinlock_t lock;
77 struct list_head lru_list;
78 struct list_head idle_ws[ZSTD_BTRFS_MAX_LEVEL];
79 unsigned long active_map;
80 wait_queue_head_t wait;
81 struct timer_list timer;
82};
83
84static struct zstd_workspace_manager wsm;
85
86static size_t zstd_ws_mem_sizes[ZSTD_BTRFS_MAX_LEVEL];
87
88static inline struct workspace *list_to_workspace(struct list_head *list)
89{
90 return container_of(list, struct workspace, list);
91}
92
93static void zstd_free_workspace(struct list_head *ws);
94static struct list_head *zstd_alloc_workspace(unsigned int level);
95
96
97
98
99
100
101
102
103static void zstd_reclaim_timer_fn(struct timer_list *timer)
104{
105 unsigned long reclaim_threshold = jiffies - ZSTD_BTRFS_RECLAIM_JIFFIES;
106 struct list_head *pos, *next;
107
108 spin_lock_bh(&wsm.lock);
109
110 if (list_empty(&wsm.lru_list)) {
111 spin_unlock_bh(&wsm.lock);
112 return;
113 }
114
115 list_for_each_prev_safe(pos, next, &wsm.lru_list) {
116 struct workspace *victim = container_of(pos, struct workspace,
117 lru_list);
118 unsigned int level;
119
120 if (time_after(victim->last_used, reclaim_threshold))
121 break;
122
123
124 if (victim->req_level)
125 continue;
126
127 level = victim->level;
128 list_del(&victim->lru_list);
129 list_del(&victim->list);
130 zstd_free_workspace(&victim->list);
131
132 if (list_empty(&wsm.idle_ws[level - 1]))
133 clear_bit(level - 1, &wsm.active_map);
134
135 }
136
137 if (!list_empty(&wsm.lru_list))
138 mod_timer(&wsm.timer, jiffies + ZSTD_BTRFS_RECLAIM_JIFFIES);
139
140 spin_unlock_bh(&wsm.lock);
141}
142
143
144
145
146
147
148
149
150
151
152static void zstd_calc_ws_mem_sizes(void)
153{
154 size_t max_size = 0;
155 unsigned int level;
156
157 for (level = 1; level <= ZSTD_BTRFS_MAX_LEVEL; level++) {
158 ZSTD_parameters params =
159 zstd_get_btrfs_parameters(level, ZSTD_BTRFS_MAX_INPUT);
160 size_t level_size =
161 max_t(size_t,
162 ZSTD_CStreamWorkspaceBound(params.cParams),
163 ZSTD_DStreamWorkspaceBound(ZSTD_BTRFS_MAX_INPUT));
164
165 max_size = max_t(size_t, max_size, level_size);
166 zstd_ws_mem_sizes[level - 1] = max_size;
167 }
168}
169
170static void zstd_init_workspace_manager(void)
171{
172 struct list_head *ws;
173 int i;
174
175 zstd_calc_ws_mem_sizes();
176
177 wsm.ops = &btrfs_zstd_compress;
178 spin_lock_init(&wsm.lock);
179 init_waitqueue_head(&wsm.wait);
180 timer_setup(&wsm.timer, zstd_reclaim_timer_fn, 0);
181
182 INIT_LIST_HEAD(&wsm.lru_list);
183 for (i = 0; i < ZSTD_BTRFS_MAX_LEVEL; i++)
184 INIT_LIST_HEAD(&wsm.idle_ws[i]);
185
186 ws = zstd_alloc_workspace(ZSTD_BTRFS_MAX_LEVEL);
187 if (IS_ERR(ws)) {
188 pr_warn(
189 "BTRFS: cannot preallocate zstd compression workspace\n");
190 } else {
191 set_bit(ZSTD_BTRFS_MAX_LEVEL - 1, &wsm.active_map);
192 list_add(ws, &wsm.idle_ws[ZSTD_BTRFS_MAX_LEVEL - 1]);
193 }
194}
195
196static void zstd_cleanup_workspace_manager(void)
197{
198 struct workspace *workspace;
199 int i;
200
201 spin_lock_bh(&wsm.lock);
202 for (i = 0; i < ZSTD_BTRFS_MAX_LEVEL; i++) {
203 while (!list_empty(&wsm.idle_ws[i])) {
204 workspace = container_of(wsm.idle_ws[i].next,
205 struct workspace, list);
206 list_del(&workspace->list);
207 list_del(&workspace->lru_list);
208 zstd_free_workspace(&workspace->list);
209 }
210 }
211 spin_unlock_bh(&wsm.lock);
212
213 del_timer_sync(&wsm.timer);
214}
215
216
217
218
219
220
221
222
223
224
225
226
227static struct list_head *zstd_find_workspace(unsigned int level)
228{
229 struct list_head *ws;
230 struct workspace *workspace;
231 int i = level - 1;
232
233 spin_lock_bh(&wsm.lock);
234 for_each_set_bit_from(i, &wsm.active_map, ZSTD_BTRFS_MAX_LEVEL) {
235 if (!list_empty(&wsm.idle_ws[i])) {
236 ws = wsm.idle_ws[i].next;
237 workspace = list_to_workspace(ws);
238 list_del_init(ws);
239
240 workspace->req_level = level;
241 if (level == workspace->level)
242 list_del(&workspace->lru_list);
243 if (list_empty(&wsm.idle_ws[i]))
244 clear_bit(i, &wsm.active_map);
245 spin_unlock_bh(&wsm.lock);
246 return ws;
247 }
248 }
249 spin_unlock_bh(&wsm.lock);
250
251 return NULL;
252}
253
254
255
256
257
258
259
260
261
262
263static struct list_head *zstd_get_workspace(unsigned int level)
264{
265 struct list_head *ws;
266 unsigned int nofs_flag;
267
268
269 if (!level)
270 level = 1;
271
272again:
273 ws = zstd_find_workspace(level);
274 if (ws)
275 return ws;
276
277 nofs_flag = memalloc_nofs_save();
278 ws = zstd_alloc_workspace(level);
279 memalloc_nofs_restore(nofs_flag);
280
281 if (IS_ERR(ws)) {
282 DEFINE_WAIT(wait);
283
284 prepare_to_wait(&wsm.wait, &wait, TASK_UNINTERRUPTIBLE);
285 schedule();
286 finish_wait(&wsm.wait, &wait);
287
288 goto again;
289 }
290
291 return ws;
292}
293
294
295
296
297
298
299
300
301
302
303
304static void zstd_put_workspace(struct list_head *ws)
305{
306 struct workspace *workspace = list_to_workspace(ws);
307
308 spin_lock_bh(&wsm.lock);
309
310
311 if (workspace->req_level == workspace->level) {
312
313 if (list_empty(&wsm.idle_ws[ZSTD_BTRFS_MAX_LEVEL - 1])) {
314 INIT_LIST_HEAD(&workspace->lru_list);
315 } else {
316 workspace->last_used = jiffies;
317 list_add(&workspace->lru_list, &wsm.lru_list);
318 if (!timer_pending(&wsm.timer))
319 mod_timer(&wsm.timer,
320 jiffies + ZSTD_BTRFS_RECLAIM_JIFFIES);
321 }
322 }
323
324 set_bit(workspace->level - 1, &wsm.active_map);
325 list_add(&workspace->list, &wsm.idle_ws[workspace->level - 1]);
326 workspace->req_level = 0;
327
328 spin_unlock_bh(&wsm.lock);
329
330 if (workspace->level == ZSTD_BTRFS_MAX_LEVEL)
331 cond_wake_up(&wsm.wait);
332}
333
334static void zstd_free_workspace(struct list_head *ws)
335{
336 struct workspace *workspace = list_entry(ws, struct workspace, list);
337
338 kvfree(workspace->mem);
339 kfree(workspace->buf);
340 kfree(workspace);
341}
342
343static struct list_head *zstd_alloc_workspace(unsigned int level)
344{
345 struct workspace *workspace;
346
347 workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
348 if (!workspace)
349 return ERR_PTR(-ENOMEM);
350
351 workspace->size = zstd_ws_mem_sizes[level - 1];
352 workspace->level = level;
353 workspace->req_level = level;
354 workspace->last_used = jiffies;
355 workspace->mem = kvmalloc(workspace->size, GFP_KERNEL);
356 workspace->buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
357 if (!workspace->mem || !workspace->buf)
358 goto fail;
359
360 INIT_LIST_HEAD(&workspace->list);
361 INIT_LIST_HEAD(&workspace->lru_list);
362
363 return &workspace->list;
364fail:
365 zstd_free_workspace(&workspace->list);
366 return ERR_PTR(-ENOMEM);
367}
368
369static int zstd_compress_pages(struct list_head *ws,
370 struct address_space *mapping,
371 u64 start,
372 struct page **pages,
373 unsigned long *out_pages,
374 unsigned long *total_in,
375 unsigned long *total_out)
376{
377 struct workspace *workspace = list_entry(ws, struct workspace, list);
378 ZSTD_CStream *stream;
379 int ret = 0;
380 int nr_pages = 0;
381 struct page *in_page = NULL;
382 struct page *out_page = NULL;
383 unsigned long tot_in = 0;
384 unsigned long tot_out = 0;
385 unsigned long len = *total_out;
386 const unsigned long nr_dest_pages = *out_pages;
387 unsigned long max_out = nr_dest_pages * PAGE_SIZE;
388 ZSTD_parameters params = zstd_get_btrfs_parameters(workspace->req_level,
389 len);
390
391 *out_pages = 0;
392 *total_out = 0;
393 *total_in = 0;
394
395
396 stream = ZSTD_initCStream(params, len, workspace->mem,
397 workspace->size);
398 if (!stream) {
399 pr_warn("BTRFS: ZSTD_initCStream failed\n");
400 ret = -EIO;
401 goto out;
402 }
403
404
405 in_page = find_get_page(mapping, start >> PAGE_SHIFT);
406 workspace->in_buf.src = kmap(in_page);
407 workspace->in_buf.pos = 0;
408 workspace->in_buf.size = min_t(size_t, len, PAGE_SIZE);
409
410
411
412 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
413 if (out_page == NULL) {
414 ret = -ENOMEM;
415 goto out;
416 }
417 pages[nr_pages++] = out_page;
418 workspace->out_buf.dst = kmap(out_page);
419 workspace->out_buf.pos = 0;
420 workspace->out_buf.size = min_t(size_t, max_out, PAGE_SIZE);
421
422 while (1) {
423 size_t ret2;
424
425 ret2 = ZSTD_compressStream(stream, &workspace->out_buf,
426 &workspace->in_buf);
427 if (ZSTD_isError(ret2)) {
428 pr_debug("BTRFS: ZSTD_compressStream returned %d\n",
429 ZSTD_getErrorCode(ret2));
430 ret = -EIO;
431 goto out;
432 }
433
434
435 if (tot_in + workspace->in_buf.pos > 8192 &&
436 tot_in + workspace->in_buf.pos <
437 tot_out + workspace->out_buf.pos) {
438 ret = -E2BIG;
439 goto out;
440 }
441
442
443 if (workspace->out_buf.pos >= max_out) {
444 tot_out += workspace->out_buf.pos;
445 ret = -E2BIG;
446 goto out;
447 }
448
449
450 if (workspace->out_buf.pos == workspace->out_buf.size) {
451 tot_out += PAGE_SIZE;
452 max_out -= PAGE_SIZE;
453 kunmap(out_page);
454 if (nr_pages == nr_dest_pages) {
455 out_page = NULL;
456 ret = -E2BIG;
457 goto out;
458 }
459 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
460 if (out_page == NULL) {
461 ret = -ENOMEM;
462 goto out;
463 }
464 pages[nr_pages++] = out_page;
465 workspace->out_buf.dst = kmap(out_page);
466 workspace->out_buf.pos = 0;
467 workspace->out_buf.size = min_t(size_t, max_out,
468 PAGE_SIZE);
469 }
470
471
472 if (workspace->in_buf.pos >= len) {
473 tot_in += workspace->in_buf.pos;
474 break;
475 }
476
477
478 if (workspace->in_buf.pos == workspace->in_buf.size) {
479 tot_in += PAGE_SIZE;
480 kunmap(in_page);
481 put_page(in_page);
482
483 start += PAGE_SIZE;
484 len -= PAGE_SIZE;
485 in_page = find_get_page(mapping, start >> PAGE_SHIFT);
486 workspace->in_buf.src = kmap(in_page);
487 workspace->in_buf.pos = 0;
488 workspace->in_buf.size = min_t(size_t, len, PAGE_SIZE);
489 }
490 }
491 while (1) {
492 size_t ret2;
493
494 ret2 = ZSTD_endStream(stream, &workspace->out_buf);
495 if (ZSTD_isError(ret2)) {
496 pr_debug("BTRFS: ZSTD_endStream returned %d\n",
497 ZSTD_getErrorCode(ret2));
498 ret = -EIO;
499 goto out;
500 }
501 if (ret2 == 0) {
502 tot_out += workspace->out_buf.pos;
503 break;
504 }
505 if (workspace->out_buf.pos >= max_out) {
506 tot_out += workspace->out_buf.pos;
507 ret = -E2BIG;
508 goto out;
509 }
510
511 tot_out += PAGE_SIZE;
512 max_out -= PAGE_SIZE;
513 kunmap(out_page);
514 if (nr_pages == nr_dest_pages) {
515 out_page = NULL;
516 ret = -E2BIG;
517 goto out;
518 }
519 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
520 if (out_page == NULL) {
521 ret = -ENOMEM;
522 goto out;
523 }
524 pages[nr_pages++] = out_page;
525 workspace->out_buf.dst = kmap(out_page);
526 workspace->out_buf.pos = 0;
527 workspace->out_buf.size = min_t(size_t, max_out, PAGE_SIZE);
528 }
529
530 if (tot_out >= tot_in) {
531 ret = -E2BIG;
532 goto out;
533 }
534
535 ret = 0;
536 *total_in = tot_in;
537 *total_out = tot_out;
538out:
539 *out_pages = nr_pages;
540
541 if (in_page) {
542 kunmap(in_page);
543 put_page(in_page);
544 }
545 if (out_page)
546 kunmap(out_page);
547 return ret;
548}
549
550static int zstd_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
551{
552 struct workspace *workspace = list_entry(ws, struct workspace, list);
553 struct page **pages_in = cb->compressed_pages;
554 u64 disk_start = cb->start;
555 struct bio *orig_bio = cb->orig_bio;
556 size_t srclen = cb->compressed_len;
557 ZSTD_DStream *stream;
558 int ret = 0;
559 unsigned long page_in_index = 0;
560 unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE);
561 unsigned long buf_start;
562 unsigned long total_out = 0;
563
564 stream = ZSTD_initDStream(
565 ZSTD_BTRFS_MAX_INPUT, workspace->mem, workspace->size);
566 if (!stream) {
567 pr_debug("BTRFS: ZSTD_initDStream failed\n");
568 ret = -EIO;
569 goto done;
570 }
571
572 workspace->in_buf.src = kmap(pages_in[page_in_index]);
573 workspace->in_buf.pos = 0;
574 workspace->in_buf.size = min_t(size_t, srclen, PAGE_SIZE);
575
576 workspace->out_buf.dst = workspace->buf;
577 workspace->out_buf.pos = 0;
578 workspace->out_buf.size = PAGE_SIZE;
579
580 while (1) {
581 size_t ret2;
582
583 ret2 = ZSTD_decompressStream(stream, &workspace->out_buf,
584 &workspace->in_buf);
585 if (ZSTD_isError(ret2)) {
586 pr_debug("BTRFS: ZSTD_decompressStream returned %d\n",
587 ZSTD_getErrorCode(ret2));
588 ret = -EIO;
589 goto done;
590 }
591 buf_start = total_out;
592 total_out += workspace->out_buf.pos;
593 workspace->out_buf.pos = 0;
594
595 ret = btrfs_decompress_buf2page(workspace->out_buf.dst,
596 buf_start, total_out, disk_start, orig_bio);
597 if (ret == 0)
598 break;
599
600 if (workspace->in_buf.pos >= srclen)
601 break;
602
603
604 if (ret2 == 0)
605 break;
606
607 if (workspace->in_buf.pos == workspace->in_buf.size) {
608 kunmap(pages_in[page_in_index++]);
609 if (page_in_index >= total_pages_in) {
610 workspace->in_buf.src = NULL;
611 ret = -EIO;
612 goto done;
613 }
614 srclen -= PAGE_SIZE;
615 workspace->in_buf.src = kmap(pages_in[page_in_index]);
616 workspace->in_buf.pos = 0;
617 workspace->in_buf.size = min_t(size_t, srclen, PAGE_SIZE);
618 }
619 }
620 ret = 0;
621 zero_fill_bio(orig_bio);
622done:
623 if (workspace->in_buf.src)
624 kunmap(pages_in[page_in_index]);
625 return ret;
626}
627
628static int zstd_decompress(struct list_head *ws, unsigned char *data_in,
629 struct page *dest_page,
630 unsigned long start_byte,
631 size_t srclen, size_t destlen)
632{
633 struct workspace *workspace = list_entry(ws, struct workspace, list);
634 ZSTD_DStream *stream;
635 int ret = 0;
636 size_t ret2;
637 unsigned long total_out = 0;
638 unsigned long pg_offset = 0;
639 char *kaddr;
640
641 stream = ZSTD_initDStream(
642 ZSTD_BTRFS_MAX_INPUT, workspace->mem, workspace->size);
643 if (!stream) {
644 pr_warn("BTRFS: ZSTD_initDStream failed\n");
645 ret = -EIO;
646 goto finish;
647 }
648
649 destlen = min_t(size_t, destlen, PAGE_SIZE);
650
651 workspace->in_buf.src = data_in;
652 workspace->in_buf.pos = 0;
653 workspace->in_buf.size = srclen;
654
655 workspace->out_buf.dst = workspace->buf;
656 workspace->out_buf.pos = 0;
657 workspace->out_buf.size = PAGE_SIZE;
658
659 ret2 = 1;
660 while (pg_offset < destlen
661 && workspace->in_buf.pos < workspace->in_buf.size) {
662 unsigned long buf_start;
663 unsigned long buf_offset;
664 unsigned long bytes;
665
666
667 if (ret2 == 0) {
668 pr_debug("BTRFS: ZSTD_decompressStream ended early\n");
669 ret = -EIO;
670 goto finish;
671 }
672 ret2 = ZSTD_decompressStream(stream, &workspace->out_buf,
673 &workspace->in_buf);
674 if (ZSTD_isError(ret2)) {
675 pr_debug("BTRFS: ZSTD_decompressStream returned %d\n",
676 ZSTD_getErrorCode(ret2));
677 ret = -EIO;
678 goto finish;
679 }
680
681 buf_start = total_out;
682 total_out += workspace->out_buf.pos;
683 workspace->out_buf.pos = 0;
684
685 if (total_out <= start_byte)
686 continue;
687
688 if (total_out > start_byte && buf_start < start_byte)
689 buf_offset = start_byte - buf_start;
690 else
691 buf_offset = 0;
692
693 bytes = min_t(unsigned long, destlen - pg_offset,
694 workspace->out_buf.size - buf_offset);
695
696 kaddr = kmap_atomic(dest_page);
697 memcpy(kaddr + pg_offset, workspace->out_buf.dst + buf_offset,
698 bytes);
699 kunmap_atomic(kaddr);
700
701 pg_offset += bytes;
702 }
703 ret = 0;
704finish:
705 if (pg_offset < destlen) {
706 kaddr = kmap_atomic(dest_page);
707 memset(kaddr + pg_offset, 0, destlen - pg_offset);
708 kunmap_atomic(kaddr);
709 }
710 return ret;
711}
712
713static unsigned int zstd_set_level(unsigned int level)
714{
715 if (!level)
716 return ZSTD_BTRFS_DEFAULT_LEVEL;
717
718 return min_t(unsigned int, level, ZSTD_BTRFS_MAX_LEVEL);
719}
720
721const struct btrfs_compress_op btrfs_zstd_compress = {
722 .init_workspace_manager = zstd_init_workspace_manager,
723 .cleanup_workspace_manager = zstd_cleanup_workspace_manager,
724 .get_workspace = zstd_get_workspace,
725 .put_workspace = zstd_put_workspace,
726 .alloc_workspace = zstd_alloc_workspace,
727 .free_workspace = zstd_free_workspace,
728 .compress_pages = zstd_compress_pages,
729 .decompress_bio = zstd_decompress_bio,
730 .decompress = zstd_decompress,
731 .set_level = zstd_set_level,
732};
733