Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2016-present, Facebook, Inc.
4 * All rights reserved.
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 "misc.h"
21#include "compression.h"
22#include "ctree.h"
23
24#define ZSTD_BTRFS_MAX_WINDOWLOG 17
25#define ZSTD_BTRFS_MAX_INPUT (1 << ZSTD_BTRFS_MAX_WINDOWLOG)
26#define ZSTD_BTRFS_DEFAULT_LEVEL 3
27#define ZSTD_BTRFS_MAX_LEVEL 15
28/* 307s to avoid pathologically clashing with transaction commit */
29#define ZSTD_BTRFS_RECLAIM_JIFFIES (307 * HZ)
30
31static ZSTD_parameters zstd_get_btrfs_parameters(unsigned int level,
32 size_t src_len)
33{
34 ZSTD_parameters params = ZSTD_getParams(level, src_len, 0);
35
36 if (params.cParams.windowLog > ZSTD_BTRFS_MAX_WINDOWLOG)
37 params.cParams.windowLog = ZSTD_BTRFS_MAX_WINDOWLOG;
38 WARN_ON(src_len > ZSTD_BTRFS_MAX_INPUT);
39 return params;
40}
41
42struct workspace {
43 void *mem;
44 size_t size;
45 char *buf;
46 unsigned int level;
47 unsigned int req_level;
48 unsigned long last_used; /* jiffies */
49 struct list_head list;
50 struct list_head lru_list;
51 ZSTD_inBuffer in_buf;
52 ZSTD_outBuffer out_buf;
53};
54
55/*
56 * Zstd Workspace Management
57 *
58 * Zstd workspaces have different memory requirements depending on the level.
59 * The zstd workspaces are managed by having individual lists for each level
60 * and a global lru. Forward progress is maintained by protecting a max level
61 * workspace.
62 *
63 * Getting a workspace is done by using the bitmap to identify the levels that
64 * have available workspaces and scans up. This lets us recycle higher level
65 * workspaces because of the monotonic memory guarantee. A workspace's
66 * last_used is only updated if it is being used by the corresponding memory
67 * level. Putting a workspace involves adding it back to the appropriate places
68 * and adding it back to the lru if necessary.
69 *
70 * A timer is used to reclaim workspaces if they have not been used for
71 * ZSTD_BTRFS_RECLAIM_JIFFIES. This helps keep only active workspaces around.
72 * The upper bound is provided by the workqueue limit which is 2 (percpu limit).
73 */
74
75struct zstd_workspace_manager {
76 const struct btrfs_compress_op *ops;
77 spinlock_t lock;
78 struct list_head lru_list;
79 struct list_head idle_ws[ZSTD_BTRFS_MAX_LEVEL];
80 unsigned long active_map;
81 wait_queue_head_t wait;
82 struct timer_list timer;
83};
84
85static struct zstd_workspace_manager wsm;
86
87static size_t zstd_ws_mem_sizes[ZSTD_BTRFS_MAX_LEVEL];
88
89static inline struct workspace *list_to_workspace(struct list_head *list)
90{
91 return container_of(list, struct workspace, list);
92}
93
94static void zstd_free_workspace(struct list_head *ws);
95static struct list_head *zstd_alloc_workspace(unsigned int level);
96
97/*
98 * zstd_reclaim_timer_fn - reclaim timer
99 * @t: timer
100 *
101 * This scans the lru_list and attempts to reclaim any workspace that hasn't
102 * been used for ZSTD_BTRFS_RECLAIM_JIFFIES.
103 */
104static void zstd_reclaim_timer_fn(struct timer_list *timer)
105{
106 unsigned long reclaim_threshold = jiffies - ZSTD_BTRFS_RECLAIM_JIFFIES;
107 struct list_head *pos, *next;
108
109 spin_lock_bh(&wsm.lock);
110
111 if (list_empty(&wsm.lru_list)) {
112 spin_unlock_bh(&wsm.lock);
113 return;
114 }
115
116 list_for_each_prev_safe(pos, next, &wsm.lru_list) {
117 struct workspace *victim = container_of(pos, struct workspace,
118 lru_list);
119 unsigned int level;
120
121 if (time_after(victim->last_used, reclaim_threshold))
122 break;
123
124 /* workspace is in use */
125 if (victim->req_level)
126 continue;
127
128 level = victim->level;
129 list_del(&victim->lru_list);
130 list_del(&victim->list);
131 zstd_free_workspace(&victim->list);
132
133 if (list_empty(&wsm.idle_ws[level - 1]))
134 clear_bit(level - 1, &wsm.active_map);
135
136 }
137
138 if (!list_empty(&wsm.lru_list))
139 mod_timer(&wsm.timer, jiffies + ZSTD_BTRFS_RECLAIM_JIFFIES);
140
141 spin_unlock_bh(&wsm.lock);
142}
143
144/*
145 * zstd_calc_ws_mem_sizes - calculate monotonic memory bounds
146 *
147 * It is possible based on the level configurations that a higher level
148 * workspace uses less memory than a lower level workspace. In order to reuse
149 * workspaces, this must be made a monotonic relationship. This precomputes
150 * the required memory for each level and enforces the monotonicity between
151 * level and memory required.
152 */
153static void zstd_calc_ws_mem_sizes(void)
154{
155 size_t max_size = 0;
156 unsigned int level;
157
158 for (level = 1; level <= ZSTD_BTRFS_MAX_LEVEL; level++) {
159 ZSTD_parameters params =
160 zstd_get_btrfs_parameters(level, ZSTD_BTRFS_MAX_INPUT);
161 size_t level_size =
162 max_t(size_t,
163 ZSTD_CStreamWorkspaceBound(params.cParams),
164 ZSTD_DStreamWorkspaceBound(ZSTD_BTRFS_MAX_INPUT));
165
166 max_size = max_t(size_t, max_size, level_size);
167 zstd_ws_mem_sizes[level - 1] = max_size;
168 }
169}
170
171static void zstd_init_workspace_manager(void)
172{
173 struct list_head *ws;
174 int i;
175
176 zstd_calc_ws_mem_sizes();
177
178 wsm.ops = &btrfs_zstd_compress;
179 spin_lock_init(&wsm.lock);
180 init_waitqueue_head(&wsm.wait);
181 timer_setup(&wsm.timer, zstd_reclaim_timer_fn, 0);
182
183 INIT_LIST_HEAD(&wsm.lru_list);
184 for (i = 0; i < ZSTD_BTRFS_MAX_LEVEL; i++)
185 INIT_LIST_HEAD(&wsm.idle_ws[i]);
186
187 ws = zstd_alloc_workspace(ZSTD_BTRFS_MAX_LEVEL);
188 if (IS_ERR(ws)) {
189 pr_warn(
190 "BTRFS: cannot preallocate zstd compression workspace\n");
191 } else {
192 set_bit(ZSTD_BTRFS_MAX_LEVEL - 1, &wsm.active_map);
193 list_add(ws, &wsm.idle_ws[ZSTD_BTRFS_MAX_LEVEL - 1]);
194 }
195}
196
197static void zstd_cleanup_workspace_manager(void)
198{
199 struct workspace *workspace;
200 int i;
201
202 spin_lock_bh(&wsm.lock);
203 for (i = 0; i < ZSTD_BTRFS_MAX_LEVEL; i++) {
204 while (!list_empty(&wsm.idle_ws[i])) {
205 workspace = container_of(wsm.idle_ws[i].next,
206 struct workspace, list);
207 list_del(&workspace->list);
208 list_del(&workspace->lru_list);
209 zstd_free_workspace(&workspace->list);
210 }
211 }
212 spin_unlock_bh(&wsm.lock);
213
214 del_timer_sync(&wsm.timer);
215}
216
217/*
218 * zstd_find_workspace - find workspace
219 * @level: compression level
220 *
221 * This iterates over the set bits in the active_map beginning at the requested
222 * compression level. This lets us utilize already allocated workspaces before
223 * allocating a new one. If the workspace is of a larger size, it is used, but
224 * the place in the lru_list and last_used times are not updated. This is to
225 * offer the opportunity to reclaim the workspace in favor of allocating an
226 * appropriately sized one in the future.
227 */
228static struct list_head *zstd_find_workspace(unsigned int level)
229{
230 struct list_head *ws;
231 struct workspace *workspace;
232 int i = level - 1;
233
234 spin_lock_bh(&wsm.lock);
235 for_each_set_bit_from(i, &wsm.active_map, ZSTD_BTRFS_MAX_LEVEL) {
236 if (!list_empty(&wsm.idle_ws[i])) {
237 ws = wsm.idle_ws[i].next;
238 workspace = list_to_workspace(ws);
239 list_del_init(ws);
240 /* keep its place if it's a lower level using this */
241 workspace->req_level = level;
242 if (level == workspace->level)
243 list_del(&workspace->lru_list);
244 if (list_empty(&wsm.idle_ws[i]))
245 clear_bit(i, &wsm.active_map);
246 spin_unlock_bh(&wsm.lock);
247 return ws;
248 }
249 }
250 spin_unlock_bh(&wsm.lock);
251
252 return NULL;
253}
254
255/*
256 * zstd_get_workspace - zstd's get_workspace
257 * @level: compression level
258 *
259 * If @level is 0, then any compression level can be used. Therefore, we begin
260 * scanning from 1. We first scan through possible workspaces and then after
261 * attempt to allocate a new workspace. If we fail to allocate one due to
262 * memory pressure, go to sleep waiting for the max level workspace to free up.
263 */
264static struct list_head *zstd_get_workspace(unsigned int level)
265{
266 struct list_head *ws;
267 unsigned int nofs_flag;
268
269 /* level == 0 means we can use any workspace */
270 if (!level)
271 level = 1;
272
273again:
274 ws = zstd_find_workspace(level);
275 if (ws)
276 return ws;
277
278 nofs_flag = memalloc_nofs_save();
279 ws = zstd_alloc_workspace(level);
280 memalloc_nofs_restore(nofs_flag);
281
282 if (IS_ERR(ws)) {
283 DEFINE_WAIT(wait);
284
285 prepare_to_wait(&wsm.wait, &wait, TASK_UNINTERRUPTIBLE);
286 schedule();
287 finish_wait(&wsm.wait, &wait);
288
289 goto again;
290 }
291
292 return ws;
293}
294
295/*
296 * zstd_put_workspace - zstd put_workspace
297 * @ws: list_head for the workspace
298 *
299 * When putting back a workspace, we only need to update the LRU if we are of
300 * the requested compression level. Here is where we continue to protect the
301 * max level workspace or update last_used accordingly. If the reclaim timer
302 * isn't set, it is also set here. Only the max level workspace tries and wakes
303 * up waiting workspaces.
304 */
305static void zstd_put_workspace(struct list_head *ws)
306{
307 struct workspace *workspace = list_to_workspace(ws);
308
309 spin_lock_bh(&wsm.lock);
310
311 /* A node is only taken off the lru if we are the corresponding level */
312 if (workspace->req_level == workspace->level) {
313 /* Hide a max level workspace from reclaim */
314 if (list_empty(&wsm.idle_ws[ZSTD_BTRFS_MAX_LEVEL - 1])) {
315 INIT_LIST_HEAD(&workspace->lru_list);
316 } else {
317 workspace->last_used = jiffies;
318 list_add(&workspace->lru_list, &wsm.lru_list);
319 if (!timer_pending(&wsm.timer))
320 mod_timer(&wsm.timer,
321 jiffies + ZSTD_BTRFS_RECLAIM_JIFFIES);
322 }
323 }
324
325 set_bit(workspace->level - 1, &wsm.active_map);
326 list_add(&workspace->list, &wsm.idle_ws[workspace->level - 1]);
327 workspace->req_level = 0;
328
329 spin_unlock_bh(&wsm.lock);
330
331 if (workspace->level == ZSTD_BTRFS_MAX_LEVEL)
332 cond_wake_up(&wsm.wait);
333}
334
335static void zstd_free_workspace(struct list_head *ws)
336{
337 struct workspace *workspace = list_entry(ws, struct workspace, list);
338
339 kvfree(workspace->mem);
340 kfree(workspace->buf);
341 kfree(workspace);
342}
343
344static struct list_head *zstd_alloc_workspace(unsigned int level)
345{
346 struct workspace *workspace;
347
348 workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
349 if (!workspace)
350 return ERR_PTR(-ENOMEM);
351
352 workspace->size = zstd_ws_mem_sizes[level - 1];
353 workspace->level = level;
354 workspace->req_level = level;
355 workspace->last_used = jiffies;
356 workspace->mem = kvmalloc(workspace->size, GFP_KERNEL);
357 workspace->buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
358 if (!workspace->mem || !workspace->buf)
359 goto fail;
360
361 INIT_LIST_HEAD(&workspace->list);
362 INIT_LIST_HEAD(&workspace->lru_list);
363
364 return &workspace->list;
365fail:
366 zstd_free_workspace(&workspace->list);
367 return ERR_PTR(-ENOMEM);
368}
369
370static int zstd_compress_pages(struct list_head *ws,
371 struct address_space *mapping,
372 u64 start,
373 struct page **pages,
374 unsigned long *out_pages,
375 unsigned long *total_in,
376 unsigned long *total_out)
377{
378 struct workspace *workspace = list_entry(ws, struct workspace, list);
379 ZSTD_CStream *stream;
380 int ret = 0;
381 int nr_pages = 0;
382 struct page *in_page = NULL; /* The current page to read */
383 struct page *out_page = NULL; /* The current page to write to */
384 unsigned long tot_in = 0;
385 unsigned long tot_out = 0;
386 unsigned long len = *total_out;
387 const unsigned long nr_dest_pages = *out_pages;
388 unsigned long max_out = nr_dest_pages * PAGE_SIZE;
389 ZSTD_parameters params = zstd_get_btrfs_parameters(workspace->req_level,
390 len);
391
392 *out_pages = 0;
393 *total_out = 0;
394 *total_in = 0;
395
396 /* Initialize the stream */
397 stream = ZSTD_initCStream(params, len, workspace->mem,
398 workspace->size);
399 if (!stream) {
400 pr_warn("BTRFS: ZSTD_initCStream failed\n");
401 ret = -EIO;
402 goto out;
403 }
404
405 /* map in the first page of input data */
406 in_page = find_get_page(mapping, start >> PAGE_SHIFT);
407 workspace->in_buf.src = kmap(in_page);
408 workspace->in_buf.pos = 0;
409 workspace->in_buf.size = min_t(size_t, len, PAGE_SIZE);
410
411
412 /* Allocate and map in the output buffer */
413 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
414 if (out_page == NULL) {
415 ret = -ENOMEM;
416 goto out;
417 }
418 pages[nr_pages++] = out_page;
419 workspace->out_buf.dst = kmap(out_page);
420 workspace->out_buf.pos = 0;
421 workspace->out_buf.size = min_t(size_t, max_out, PAGE_SIZE);
422
423 while (1) {
424 size_t ret2;
425
426 ret2 = ZSTD_compressStream(stream, &workspace->out_buf,
427 &workspace->in_buf);
428 if (ZSTD_isError(ret2)) {
429 pr_debug("BTRFS: ZSTD_compressStream returned %d\n",
430 ZSTD_getErrorCode(ret2));
431 ret = -EIO;
432 goto out;
433 }
434
435 /* Check to see if we are making it bigger */
436 if (tot_in + workspace->in_buf.pos > 8192 &&
437 tot_in + workspace->in_buf.pos <
438 tot_out + workspace->out_buf.pos) {
439 ret = -E2BIG;
440 goto out;
441 }
442
443 /* We've reached the end of our output range */
444 if (workspace->out_buf.pos >= max_out) {
445 tot_out += workspace->out_buf.pos;
446 ret = -E2BIG;
447 goto out;
448 }
449
450 /* Check if we need more output space */
451 if (workspace->out_buf.pos == workspace->out_buf.size) {
452 tot_out += PAGE_SIZE;
453 max_out -= PAGE_SIZE;
454 kunmap(out_page);
455 if (nr_pages == nr_dest_pages) {
456 out_page = NULL;
457 ret = -E2BIG;
458 goto out;
459 }
460 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
461 if (out_page == NULL) {
462 ret = -ENOMEM;
463 goto out;
464 }
465 pages[nr_pages++] = out_page;
466 workspace->out_buf.dst = kmap(out_page);
467 workspace->out_buf.pos = 0;
468 workspace->out_buf.size = min_t(size_t, max_out,
469 PAGE_SIZE);
470 }
471
472 /* We've reached the end of the input */
473 if (workspace->in_buf.pos >= len) {
474 tot_in += workspace->in_buf.pos;
475 break;
476 }
477
478 /* Check if we need more input */
479 if (workspace->in_buf.pos == workspace->in_buf.size) {
480 tot_in += PAGE_SIZE;
481 kunmap(in_page);
482 put_page(in_page);
483
484 start += PAGE_SIZE;
485 len -= PAGE_SIZE;
486 in_page = find_get_page(mapping, start >> PAGE_SHIFT);
487 workspace->in_buf.src = kmap(in_page);
488 workspace->in_buf.pos = 0;
489 workspace->in_buf.size = min_t(size_t, len, PAGE_SIZE);
490 }
491 }
492 while (1) {
493 size_t ret2;
494
495 ret2 = ZSTD_endStream(stream, &workspace->out_buf);
496 if (ZSTD_isError(ret2)) {
497 pr_debug("BTRFS: ZSTD_endStream returned %d\n",
498 ZSTD_getErrorCode(ret2));
499 ret = -EIO;
500 goto out;
501 }
502 if (ret2 == 0) {
503 tot_out += workspace->out_buf.pos;
504 break;
505 }
506 if (workspace->out_buf.pos >= max_out) {
507 tot_out += workspace->out_buf.pos;
508 ret = -E2BIG;
509 goto out;
510 }
511
512 tot_out += PAGE_SIZE;
513 max_out -= PAGE_SIZE;
514 kunmap(out_page);
515 if (nr_pages == nr_dest_pages) {
516 out_page = NULL;
517 ret = -E2BIG;
518 goto out;
519 }
520 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
521 if (out_page == NULL) {
522 ret = -ENOMEM;
523 goto out;
524 }
525 pages[nr_pages++] = out_page;
526 workspace->out_buf.dst = kmap(out_page);
527 workspace->out_buf.pos = 0;
528 workspace->out_buf.size = min_t(size_t, max_out, PAGE_SIZE);
529 }
530
531 if (tot_out >= tot_in) {
532 ret = -E2BIG;
533 goto out;
534 }
535
536 ret = 0;
537 *total_in = tot_in;
538 *total_out = tot_out;
539out:
540 *out_pages = nr_pages;
541 /* Cleanup */
542 if (in_page) {
543 kunmap(in_page);
544 put_page(in_page);
545 }
546 if (out_page)
547 kunmap(out_page);
548 return ret;
549}
550
551static int zstd_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
552{
553 struct workspace *workspace = list_entry(ws, struct workspace, list);
554 struct page **pages_in = cb->compressed_pages;
555 u64 disk_start = cb->start;
556 struct bio *orig_bio = cb->orig_bio;
557 size_t srclen = cb->compressed_len;
558 ZSTD_DStream *stream;
559 int ret = 0;
560 unsigned long page_in_index = 0;
561 unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE);
562 unsigned long buf_start;
563 unsigned long total_out = 0;
564
565 stream = ZSTD_initDStream(
566 ZSTD_BTRFS_MAX_INPUT, workspace->mem, workspace->size);
567 if (!stream) {
568 pr_debug("BTRFS: ZSTD_initDStream failed\n");
569 ret = -EIO;
570 goto done;
571 }
572
573 workspace->in_buf.src = kmap(pages_in[page_in_index]);
574 workspace->in_buf.pos = 0;
575 workspace->in_buf.size = min_t(size_t, srclen, PAGE_SIZE);
576
577 workspace->out_buf.dst = workspace->buf;
578 workspace->out_buf.pos = 0;
579 workspace->out_buf.size = PAGE_SIZE;
580
581 while (1) {
582 size_t ret2;
583
584 ret2 = ZSTD_decompressStream(stream, &workspace->out_buf,
585 &workspace->in_buf);
586 if (ZSTD_isError(ret2)) {
587 pr_debug("BTRFS: ZSTD_decompressStream returned %d\n",
588 ZSTD_getErrorCode(ret2));
589 ret = -EIO;
590 goto done;
591 }
592 buf_start = total_out;
593 total_out += workspace->out_buf.pos;
594 workspace->out_buf.pos = 0;
595
596 ret = btrfs_decompress_buf2page(workspace->out_buf.dst,
597 buf_start, total_out, disk_start, orig_bio);
598 if (ret == 0)
599 break;
600
601 if (workspace->in_buf.pos >= srclen)
602 break;
603
604 /* Check if we've hit the end of a frame */
605 if (ret2 == 0)
606 break;
607
608 if (workspace->in_buf.pos == workspace->in_buf.size) {
609 kunmap(pages_in[page_in_index++]);
610 if (page_in_index >= total_pages_in) {
611 workspace->in_buf.src = NULL;
612 ret = -EIO;
613 goto done;
614 }
615 srclen -= PAGE_SIZE;
616 workspace->in_buf.src = kmap(pages_in[page_in_index]);
617 workspace->in_buf.pos = 0;
618 workspace->in_buf.size = min_t(size_t, srclen, PAGE_SIZE);
619 }
620 }
621 ret = 0;
622 zero_fill_bio(orig_bio);
623done:
624 if (workspace->in_buf.src)
625 kunmap(pages_in[page_in_index]);
626 return ret;
627}
628
629static int zstd_decompress(struct list_head *ws, unsigned char *data_in,
630 struct page *dest_page,
631 unsigned long start_byte,
632 size_t srclen, size_t destlen)
633{
634 struct workspace *workspace = list_entry(ws, struct workspace, list);
635 ZSTD_DStream *stream;
636 int ret = 0;
637 size_t ret2;
638 unsigned long total_out = 0;
639 unsigned long pg_offset = 0;
640 char *kaddr;
641
642 stream = ZSTD_initDStream(
643 ZSTD_BTRFS_MAX_INPUT, workspace->mem, workspace->size);
644 if (!stream) {
645 pr_warn("BTRFS: ZSTD_initDStream failed\n");
646 ret = -EIO;
647 goto finish;
648 }
649
650 destlen = min_t(size_t, destlen, PAGE_SIZE);
651
652 workspace->in_buf.src = data_in;
653 workspace->in_buf.pos = 0;
654 workspace->in_buf.size = srclen;
655
656 workspace->out_buf.dst = workspace->buf;
657 workspace->out_buf.pos = 0;
658 workspace->out_buf.size = PAGE_SIZE;
659
660 ret2 = 1;
661 while (pg_offset < destlen
662 && workspace->in_buf.pos < workspace->in_buf.size) {
663 unsigned long buf_start;
664 unsigned long buf_offset;
665 unsigned long bytes;
666
667 /* Check if the frame is over and we still need more input */
668 if (ret2 == 0) {
669 pr_debug("BTRFS: ZSTD_decompressStream ended early\n");
670 ret = -EIO;
671 goto finish;
672 }
673 ret2 = ZSTD_decompressStream(stream, &workspace->out_buf,
674 &workspace->in_buf);
675 if (ZSTD_isError(ret2)) {
676 pr_debug("BTRFS: ZSTD_decompressStream returned %d\n",
677 ZSTD_getErrorCode(ret2));
678 ret = -EIO;
679 goto finish;
680 }
681
682 buf_start = total_out;
683 total_out += workspace->out_buf.pos;
684 workspace->out_buf.pos = 0;
685
686 if (total_out <= start_byte)
687 continue;
688
689 if (total_out > start_byte && buf_start < start_byte)
690 buf_offset = start_byte - buf_start;
691 else
692 buf_offset = 0;
693
694 bytes = min_t(unsigned long, destlen - pg_offset,
695 workspace->out_buf.size - buf_offset);
696
697 kaddr = kmap_atomic(dest_page);
698 memcpy(kaddr + pg_offset, workspace->out_buf.dst + buf_offset,
699 bytes);
700 kunmap_atomic(kaddr);
701
702 pg_offset += bytes;
703 }
704 ret = 0;
705finish:
706 if (pg_offset < destlen) {
707 kaddr = kmap_atomic(dest_page);
708 memset(kaddr + pg_offset, 0, destlen - pg_offset);
709 kunmap_atomic(kaddr);
710 }
711 return ret;
712}
713
714const struct btrfs_compress_op btrfs_zstd_compress = {
715 .init_workspace_manager = zstd_init_workspace_manager,
716 .cleanup_workspace_manager = zstd_cleanup_workspace_manager,
717 .get_workspace = zstd_get_workspace,
718 .put_workspace = zstd_put_workspace,
719 .alloc_workspace = zstd_alloc_workspace,
720 .free_workspace = zstd_free_workspace,
721 .compress_pages = zstd_compress_pages,
722 .decompress_bio = zstd_decompress_bio,
723 .decompress = zstd_decompress,
724 .max_level = ZSTD_BTRFS_MAX_LEVEL,
725 .default_level = ZSTD_BTRFS_DEFAULT_LEVEL,
726};
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2016-present, Facebook, Inc.
4 * All rights reserved.
5 *
6 */
7
8#include <linux/bio.h>
9#include <linux/err.h>
10#include <linux/init.h>
11#include <linux/kernel.h>
12#include <linux/mm.h>
13#include <linux/pagemap.h>
14#include <linux/refcount.h>
15#include <linux/sched.h>
16#include <linux/slab.h>
17#include <linux/zstd.h>
18#include "compression.h"
19
20#define ZSTD_BTRFS_MAX_WINDOWLOG 17
21#define ZSTD_BTRFS_MAX_INPUT (1 << ZSTD_BTRFS_MAX_WINDOWLOG)
22#define ZSTD_BTRFS_DEFAULT_LEVEL 3
23
24static ZSTD_parameters zstd_get_btrfs_parameters(size_t src_len)
25{
26 ZSTD_parameters params = ZSTD_getParams(ZSTD_BTRFS_DEFAULT_LEVEL,
27 src_len, 0);
28
29 if (params.cParams.windowLog > ZSTD_BTRFS_MAX_WINDOWLOG)
30 params.cParams.windowLog = ZSTD_BTRFS_MAX_WINDOWLOG;
31 WARN_ON(src_len > ZSTD_BTRFS_MAX_INPUT);
32 return params;
33}
34
35struct workspace {
36 void *mem;
37 size_t size;
38 char *buf;
39 struct list_head list;
40 ZSTD_inBuffer in_buf;
41 ZSTD_outBuffer out_buf;
42};
43
44static void zstd_free_workspace(struct list_head *ws)
45{
46 struct workspace *workspace = list_entry(ws, struct workspace, list);
47
48 kvfree(workspace->mem);
49 kfree(workspace->buf);
50 kfree(workspace);
51}
52
53static struct list_head *zstd_alloc_workspace(void)
54{
55 ZSTD_parameters params =
56 zstd_get_btrfs_parameters(ZSTD_BTRFS_MAX_INPUT);
57 struct workspace *workspace;
58
59 workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
60 if (!workspace)
61 return ERR_PTR(-ENOMEM);
62
63 workspace->size = max_t(size_t,
64 ZSTD_CStreamWorkspaceBound(params.cParams),
65 ZSTD_DStreamWorkspaceBound(ZSTD_BTRFS_MAX_INPUT));
66 workspace->mem = kvmalloc(workspace->size, GFP_KERNEL);
67 workspace->buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
68 if (!workspace->mem || !workspace->buf)
69 goto fail;
70
71 INIT_LIST_HEAD(&workspace->list);
72
73 return &workspace->list;
74fail:
75 zstd_free_workspace(&workspace->list);
76 return ERR_PTR(-ENOMEM);
77}
78
79static int zstd_compress_pages(struct list_head *ws,
80 struct address_space *mapping,
81 u64 start,
82 struct page **pages,
83 unsigned long *out_pages,
84 unsigned long *total_in,
85 unsigned long *total_out)
86{
87 struct workspace *workspace = list_entry(ws, struct workspace, list);
88 ZSTD_CStream *stream;
89 int ret = 0;
90 int nr_pages = 0;
91 struct page *in_page = NULL; /* The current page to read */
92 struct page *out_page = NULL; /* The current page to write to */
93 unsigned long tot_in = 0;
94 unsigned long tot_out = 0;
95 unsigned long len = *total_out;
96 const unsigned long nr_dest_pages = *out_pages;
97 unsigned long max_out = nr_dest_pages * PAGE_SIZE;
98 ZSTD_parameters params = zstd_get_btrfs_parameters(len);
99
100 *out_pages = 0;
101 *total_out = 0;
102 *total_in = 0;
103
104 /* Initialize the stream */
105 stream = ZSTD_initCStream(params, len, workspace->mem,
106 workspace->size);
107 if (!stream) {
108 pr_warn("BTRFS: ZSTD_initCStream failed\n");
109 ret = -EIO;
110 goto out;
111 }
112
113 /* map in the first page of input data */
114 in_page = find_get_page(mapping, start >> PAGE_SHIFT);
115 workspace->in_buf.src = kmap(in_page);
116 workspace->in_buf.pos = 0;
117 workspace->in_buf.size = min_t(size_t, len, PAGE_SIZE);
118
119
120 /* Allocate and map in the output buffer */
121 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
122 if (out_page == NULL) {
123 ret = -ENOMEM;
124 goto out;
125 }
126 pages[nr_pages++] = out_page;
127 workspace->out_buf.dst = kmap(out_page);
128 workspace->out_buf.pos = 0;
129 workspace->out_buf.size = min_t(size_t, max_out, PAGE_SIZE);
130
131 while (1) {
132 size_t ret2;
133
134 ret2 = ZSTD_compressStream(stream, &workspace->out_buf,
135 &workspace->in_buf);
136 if (ZSTD_isError(ret2)) {
137 pr_debug("BTRFS: ZSTD_compressStream returned %d\n",
138 ZSTD_getErrorCode(ret2));
139 ret = -EIO;
140 goto out;
141 }
142
143 /* Check to see if we are making it bigger */
144 if (tot_in + workspace->in_buf.pos > 8192 &&
145 tot_in + workspace->in_buf.pos <
146 tot_out + workspace->out_buf.pos) {
147 ret = -E2BIG;
148 goto out;
149 }
150
151 /* We've reached the end of our output range */
152 if (workspace->out_buf.pos >= max_out) {
153 tot_out += workspace->out_buf.pos;
154 ret = -E2BIG;
155 goto out;
156 }
157
158 /* Check if we need more output space */
159 if (workspace->out_buf.pos == workspace->out_buf.size) {
160 tot_out += PAGE_SIZE;
161 max_out -= PAGE_SIZE;
162 kunmap(out_page);
163 if (nr_pages == nr_dest_pages) {
164 out_page = NULL;
165 ret = -E2BIG;
166 goto out;
167 }
168 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
169 if (out_page == NULL) {
170 ret = -ENOMEM;
171 goto out;
172 }
173 pages[nr_pages++] = out_page;
174 workspace->out_buf.dst = kmap(out_page);
175 workspace->out_buf.pos = 0;
176 workspace->out_buf.size = min_t(size_t, max_out,
177 PAGE_SIZE);
178 }
179
180 /* We've reached the end of the input */
181 if (workspace->in_buf.pos >= len) {
182 tot_in += workspace->in_buf.pos;
183 break;
184 }
185
186 /* Check if we need more input */
187 if (workspace->in_buf.pos == workspace->in_buf.size) {
188 tot_in += PAGE_SIZE;
189 kunmap(in_page);
190 put_page(in_page);
191
192 start += PAGE_SIZE;
193 len -= PAGE_SIZE;
194 in_page = find_get_page(mapping, start >> PAGE_SHIFT);
195 workspace->in_buf.src = kmap(in_page);
196 workspace->in_buf.pos = 0;
197 workspace->in_buf.size = min_t(size_t, len, PAGE_SIZE);
198 }
199 }
200 while (1) {
201 size_t ret2;
202
203 ret2 = ZSTD_endStream(stream, &workspace->out_buf);
204 if (ZSTD_isError(ret2)) {
205 pr_debug("BTRFS: ZSTD_endStream returned %d\n",
206 ZSTD_getErrorCode(ret2));
207 ret = -EIO;
208 goto out;
209 }
210 if (ret2 == 0) {
211 tot_out += workspace->out_buf.pos;
212 break;
213 }
214 if (workspace->out_buf.pos >= max_out) {
215 tot_out += workspace->out_buf.pos;
216 ret = -E2BIG;
217 goto out;
218 }
219
220 tot_out += PAGE_SIZE;
221 max_out -= PAGE_SIZE;
222 kunmap(out_page);
223 if (nr_pages == nr_dest_pages) {
224 out_page = NULL;
225 ret = -E2BIG;
226 goto out;
227 }
228 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
229 if (out_page == NULL) {
230 ret = -ENOMEM;
231 goto out;
232 }
233 pages[nr_pages++] = out_page;
234 workspace->out_buf.dst = kmap(out_page);
235 workspace->out_buf.pos = 0;
236 workspace->out_buf.size = min_t(size_t, max_out, PAGE_SIZE);
237 }
238
239 if (tot_out >= tot_in) {
240 ret = -E2BIG;
241 goto out;
242 }
243
244 ret = 0;
245 *total_in = tot_in;
246 *total_out = tot_out;
247out:
248 *out_pages = nr_pages;
249 /* Cleanup */
250 if (in_page) {
251 kunmap(in_page);
252 put_page(in_page);
253 }
254 if (out_page)
255 kunmap(out_page);
256 return ret;
257}
258
259static int zstd_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
260{
261 struct workspace *workspace = list_entry(ws, struct workspace, list);
262 struct page **pages_in = cb->compressed_pages;
263 u64 disk_start = cb->start;
264 struct bio *orig_bio = cb->orig_bio;
265 size_t srclen = cb->compressed_len;
266 ZSTD_DStream *stream;
267 int ret = 0;
268 unsigned long page_in_index = 0;
269 unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE);
270 unsigned long buf_start;
271 unsigned long total_out = 0;
272
273 stream = ZSTD_initDStream(
274 ZSTD_BTRFS_MAX_INPUT, workspace->mem, workspace->size);
275 if (!stream) {
276 pr_debug("BTRFS: ZSTD_initDStream failed\n");
277 ret = -EIO;
278 goto done;
279 }
280
281 workspace->in_buf.src = kmap(pages_in[page_in_index]);
282 workspace->in_buf.pos = 0;
283 workspace->in_buf.size = min_t(size_t, srclen, PAGE_SIZE);
284
285 workspace->out_buf.dst = workspace->buf;
286 workspace->out_buf.pos = 0;
287 workspace->out_buf.size = PAGE_SIZE;
288
289 while (1) {
290 size_t ret2;
291
292 ret2 = ZSTD_decompressStream(stream, &workspace->out_buf,
293 &workspace->in_buf);
294 if (ZSTD_isError(ret2)) {
295 pr_debug("BTRFS: ZSTD_decompressStream returned %d\n",
296 ZSTD_getErrorCode(ret2));
297 ret = -EIO;
298 goto done;
299 }
300 buf_start = total_out;
301 total_out += workspace->out_buf.pos;
302 workspace->out_buf.pos = 0;
303
304 ret = btrfs_decompress_buf2page(workspace->out_buf.dst,
305 buf_start, total_out, disk_start, orig_bio);
306 if (ret == 0)
307 break;
308
309 if (workspace->in_buf.pos >= srclen)
310 break;
311
312 /* Check if we've hit the end of a frame */
313 if (ret2 == 0)
314 break;
315
316 if (workspace->in_buf.pos == workspace->in_buf.size) {
317 kunmap(pages_in[page_in_index++]);
318 if (page_in_index >= total_pages_in) {
319 workspace->in_buf.src = NULL;
320 ret = -EIO;
321 goto done;
322 }
323 srclen -= PAGE_SIZE;
324 workspace->in_buf.src = kmap(pages_in[page_in_index]);
325 workspace->in_buf.pos = 0;
326 workspace->in_buf.size = min_t(size_t, srclen, PAGE_SIZE);
327 }
328 }
329 ret = 0;
330 zero_fill_bio(orig_bio);
331done:
332 if (workspace->in_buf.src)
333 kunmap(pages_in[page_in_index]);
334 return ret;
335}
336
337static int zstd_decompress(struct list_head *ws, unsigned char *data_in,
338 struct page *dest_page,
339 unsigned long start_byte,
340 size_t srclen, size_t destlen)
341{
342 struct workspace *workspace = list_entry(ws, struct workspace, list);
343 ZSTD_DStream *stream;
344 int ret = 0;
345 size_t ret2;
346 unsigned long total_out = 0;
347 unsigned long pg_offset = 0;
348 char *kaddr;
349
350 stream = ZSTD_initDStream(
351 ZSTD_BTRFS_MAX_INPUT, workspace->mem, workspace->size);
352 if (!stream) {
353 pr_warn("BTRFS: ZSTD_initDStream failed\n");
354 ret = -EIO;
355 goto finish;
356 }
357
358 destlen = min_t(size_t, destlen, PAGE_SIZE);
359
360 workspace->in_buf.src = data_in;
361 workspace->in_buf.pos = 0;
362 workspace->in_buf.size = srclen;
363
364 workspace->out_buf.dst = workspace->buf;
365 workspace->out_buf.pos = 0;
366 workspace->out_buf.size = PAGE_SIZE;
367
368 ret2 = 1;
369 while (pg_offset < destlen
370 && workspace->in_buf.pos < workspace->in_buf.size) {
371 unsigned long buf_start;
372 unsigned long buf_offset;
373 unsigned long bytes;
374
375 /* Check if the frame is over and we still need more input */
376 if (ret2 == 0) {
377 pr_debug("BTRFS: ZSTD_decompressStream ended early\n");
378 ret = -EIO;
379 goto finish;
380 }
381 ret2 = ZSTD_decompressStream(stream, &workspace->out_buf,
382 &workspace->in_buf);
383 if (ZSTD_isError(ret2)) {
384 pr_debug("BTRFS: ZSTD_decompressStream returned %d\n",
385 ZSTD_getErrorCode(ret2));
386 ret = -EIO;
387 goto finish;
388 }
389
390 buf_start = total_out;
391 total_out += workspace->out_buf.pos;
392 workspace->out_buf.pos = 0;
393
394 if (total_out <= start_byte)
395 continue;
396
397 if (total_out > start_byte && buf_start < start_byte)
398 buf_offset = start_byte - buf_start;
399 else
400 buf_offset = 0;
401
402 bytes = min_t(unsigned long, destlen - pg_offset,
403 workspace->out_buf.size - buf_offset);
404
405 kaddr = kmap_atomic(dest_page);
406 memcpy(kaddr + pg_offset, workspace->out_buf.dst + buf_offset,
407 bytes);
408 kunmap_atomic(kaddr);
409
410 pg_offset += bytes;
411 }
412 ret = 0;
413finish:
414 if (pg_offset < destlen) {
415 kaddr = kmap_atomic(dest_page);
416 memset(kaddr + pg_offset, 0, destlen - pg_offset);
417 kunmap_atomic(kaddr);
418 }
419 return ret;
420}
421
422static void zstd_set_level(struct list_head *ws, unsigned int type)
423{
424}
425
426const struct btrfs_compress_op btrfs_zstd_compress = {
427 .alloc_workspace = zstd_alloc_workspace,
428 .free_workspace = zstd_free_workspace,
429 .compress_pages = zstd_compress_pages,
430 .decompress_bio = zstd_decompress_bio,
431 .decompress = zstd_decompress,
432 .set_level = zstd_set_level,
433};