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_get_params(level, src_len);
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_in_buffer in_buf;
52 zstd_out_buffer 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
94void zstd_free_workspace(struct list_head *ws);
95struct list_head *zstd_alloc_workspace(unsigned int level);
96
97/*
98 * Timer callback to free unused workspaces.
99 *
100 * @t: timer
101 *
102 * This scans the lru_list and attempts to reclaim any workspace that hasn't
103 * been used for ZSTD_BTRFS_RECLAIM_JIFFIES.
104 *
105 * The context is softirq and does not need the _bh locking primitives.
106 */
107static void zstd_reclaim_timer_fn(struct timer_list *timer)
108{
109 unsigned long reclaim_threshold = jiffies - ZSTD_BTRFS_RECLAIM_JIFFIES;
110 struct list_head *pos, *next;
111
112 spin_lock(&wsm.lock);
113
114 if (list_empty(&wsm.lru_list)) {
115 spin_unlock(&wsm.lock);
116 return;
117 }
118
119 list_for_each_prev_safe(pos, next, &wsm.lru_list) {
120 struct workspace *victim = container_of(pos, struct workspace,
121 lru_list);
122 unsigned int level;
123
124 if (time_after(victim->last_used, reclaim_threshold))
125 break;
126
127 /* workspace is in use */
128 if (victim->req_level)
129 continue;
130
131 level = victim->level;
132 list_del(&victim->lru_list);
133 list_del(&victim->list);
134 zstd_free_workspace(&victim->list);
135
136 if (list_empty(&wsm.idle_ws[level - 1]))
137 clear_bit(level - 1, &wsm.active_map);
138
139 }
140
141 if (!list_empty(&wsm.lru_list))
142 mod_timer(&wsm.timer, jiffies + ZSTD_BTRFS_RECLAIM_JIFFIES);
143
144 spin_unlock(&wsm.lock);
145}
146
147/*
148 * zstd_calc_ws_mem_sizes - calculate monotonic memory bounds
149 *
150 * It is possible based on the level configurations that a higher level
151 * workspace uses less memory than a lower level workspace. In order to reuse
152 * workspaces, this must be made a monotonic relationship. This precomputes
153 * the required memory for each level and enforces the monotonicity between
154 * level and memory required.
155 */
156static void zstd_calc_ws_mem_sizes(void)
157{
158 size_t max_size = 0;
159 unsigned int level;
160
161 for (level = 1; level <= ZSTD_BTRFS_MAX_LEVEL; level++) {
162 zstd_parameters params =
163 zstd_get_btrfs_parameters(level, ZSTD_BTRFS_MAX_INPUT);
164 size_t level_size =
165 max_t(size_t,
166 zstd_cstream_workspace_bound(¶ms.cParams),
167 zstd_dstream_workspace_bound(ZSTD_BTRFS_MAX_INPUT));
168
169 max_size = max_t(size_t, max_size, level_size);
170 zstd_ws_mem_sizes[level - 1] = max_size;
171 }
172}
173
174void zstd_init_workspace_manager(void)
175{
176 struct list_head *ws;
177 int i;
178
179 zstd_calc_ws_mem_sizes();
180
181 wsm.ops = &btrfs_zstd_compress;
182 spin_lock_init(&wsm.lock);
183 init_waitqueue_head(&wsm.wait);
184 timer_setup(&wsm.timer, zstd_reclaim_timer_fn, 0);
185
186 INIT_LIST_HEAD(&wsm.lru_list);
187 for (i = 0; i < ZSTD_BTRFS_MAX_LEVEL; i++)
188 INIT_LIST_HEAD(&wsm.idle_ws[i]);
189
190 ws = zstd_alloc_workspace(ZSTD_BTRFS_MAX_LEVEL);
191 if (IS_ERR(ws)) {
192 pr_warn(
193 "BTRFS: cannot preallocate zstd compression workspace\n");
194 } else {
195 set_bit(ZSTD_BTRFS_MAX_LEVEL - 1, &wsm.active_map);
196 list_add(ws, &wsm.idle_ws[ZSTD_BTRFS_MAX_LEVEL - 1]);
197 }
198}
199
200void zstd_cleanup_workspace_manager(void)
201{
202 struct workspace *workspace;
203 int i;
204
205 spin_lock_bh(&wsm.lock);
206 for (i = 0; i < ZSTD_BTRFS_MAX_LEVEL; i++) {
207 while (!list_empty(&wsm.idle_ws[i])) {
208 workspace = container_of(wsm.idle_ws[i].next,
209 struct workspace, list);
210 list_del(&workspace->list);
211 list_del(&workspace->lru_list);
212 zstd_free_workspace(&workspace->list);
213 }
214 }
215 spin_unlock_bh(&wsm.lock);
216
217 del_timer_sync(&wsm.timer);
218}
219
220/*
221 * zstd_find_workspace - find workspace
222 * @level: compression level
223 *
224 * This iterates over the set bits in the active_map beginning at the requested
225 * compression level. This lets us utilize already allocated workspaces before
226 * allocating a new one. If the workspace is of a larger size, it is used, but
227 * the place in the lru_list and last_used times are not updated. This is to
228 * offer the opportunity to reclaim the workspace in favor of allocating an
229 * appropriately sized one in the future.
230 */
231static struct list_head *zstd_find_workspace(unsigned int level)
232{
233 struct list_head *ws;
234 struct workspace *workspace;
235 int i = level - 1;
236
237 spin_lock_bh(&wsm.lock);
238 for_each_set_bit_from(i, &wsm.active_map, ZSTD_BTRFS_MAX_LEVEL) {
239 if (!list_empty(&wsm.idle_ws[i])) {
240 ws = wsm.idle_ws[i].next;
241 workspace = list_to_workspace(ws);
242 list_del_init(ws);
243 /* keep its place if it's a lower level using this */
244 workspace->req_level = level;
245 if (level == workspace->level)
246 list_del(&workspace->lru_list);
247 if (list_empty(&wsm.idle_ws[i]))
248 clear_bit(i, &wsm.active_map);
249 spin_unlock_bh(&wsm.lock);
250 return ws;
251 }
252 }
253 spin_unlock_bh(&wsm.lock);
254
255 return NULL;
256}
257
258/*
259 * zstd_get_workspace - zstd's get_workspace
260 * @level: compression level
261 *
262 * If @level is 0, then any compression level can be used. Therefore, we begin
263 * scanning from 1. We first scan through possible workspaces and then after
264 * attempt to allocate a new workspace. If we fail to allocate one due to
265 * memory pressure, go to sleep waiting for the max level workspace to free up.
266 */
267struct list_head *zstd_get_workspace(unsigned int level)
268{
269 struct list_head *ws;
270 unsigned int nofs_flag;
271
272 /* level == 0 means we can use any workspace */
273 if (!level)
274 level = 1;
275
276again:
277 ws = zstd_find_workspace(level);
278 if (ws)
279 return ws;
280
281 nofs_flag = memalloc_nofs_save();
282 ws = zstd_alloc_workspace(level);
283 memalloc_nofs_restore(nofs_flag);
284
285 if (IS_ERR(ws)) {
286 DEFINE_WAIT(wait);
287
288 prepare_to_wait(&wsm.wait, &wait, TASK_UNINTERRUPTIBLE);
289 schedule();
290 finish_wait(&wsm.wait, &wait);
291
292 goto again;
293 }
294
295 return ws;
296}
297
298/*
299 * zstd_put_workspace - zstd put_workspace
300 * @ws: list_head for the workspace
301 *
302 * When putting back a workspace, we only need to update the LRU if we are of
303 * the requested compression level. Here is where we continue to protect the
304 * max level workspace or update last_used accordingly. If the reclaim timer
305 * isn't set, it is also set here. Only the max level workspace tries and wakes
306 * up waiting workspaces.
307 */
308void zstd_put_workspace(struct list_head *ws)
309{
310 struct workspace *workspace = list_to_workspace(ws);
311
312 spin_lock_bh(&wsm.lock);
313
314 /* A node is only taken off the lru if we are the corresponding level */
315 if (workspace->req_level == workspace->level) {
316 /* Hide a max level workspace from reclaim */
317 if (list_empty(&wsm.idle_ws[ZSTD_BTRFS_MAX_LEVEL - 1])) {
318 INIT_LIST_HEAD(&workspace->lru_list);
319 } else {
320 workspace->last_used = jiffies;
321 list_add(&workspace->lru_list, &wsm.lru_list);
322 if (!timer_pending(&wsm.timer))
323 mod_timer(&wsm.timer,
324 jiffies + ZSTD_BTRFS_RECLAIM_JIFFIES);
325 }
326 }
327
328 set_bit(workspace->level - 1, &wsm.active_map);
329 list_add(&workspace->list, &wsm.idle_ws[workspace->level - 1]);
330 workspace->req_level = 0;
331
332 spin_unlock_bh(&wsm.lock);
333
334 if (workspace->level == ZSTD_BTRFS_MAX_LEVEL)
335 cond_wake_up(&wsm.wait);
336}
337
338void zstd_free_workspace(struct list_head *ws)
339{
340 struct workspace *workspace = list_entry(ws, struct workspace, list);
341
342 kvfree(workspace->mem);
343 kfree(workspace->buf);
344 kfree(workspace);
345}
346
347struct list_head *zstd_alloc_workspace(unsigned int level)
348{
349 struct workspace *workspace;
350
351 workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
352 if (!workspace)
353 return ERR_PTR(-ENOMEM);
354
355 workspace->size = zstd_ws_mem_sizes[level - 1];
356 workspace->level = level;
357 workspace->req_level = level;
358 workspace->last_used = jiffies;
359 workspace->mem = kvmalloc(workspace->size, GFP_KERNEL);
360 workspace->buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
361 if (!workspace->mem || !workspace->buf)
362 goto fail;
363
364 INIT_LIST_HEAD(&workspace->list);
365 INIT_LIST_HEAD(&workspace->lru_list);
366
367 return &workspace->list;
368fail:
369 zstd_free_workspace(&workspace->list);
370 return ERR_PTR(-ENOMEM);
371}
372
373int zstd_compress_pages(struct list_head *ws, struct address_space *mapping,
374 u64 start, struct page **pages, unsigned long *out_pages,
375 unsigned long *total_in, 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; /* The current page to read */
382 struct page *out_page = NULL; /* The current page to write to */
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 /* Initialize the stream */
396 stream = zstd_init_cstream(¶ms, len, workspace->mem,
397 workspace->size);
398 if (!stream) {
399 pr_warn("BTRFS: zstd_init_cstream failed\n");
400 ret = -EIO;
401 goto out;
402 }
403
404 /* map in the first page of input data */
405 in_page = find_get_page(mapping, start >> PAGE_SHIFT);
406 workspace->in_buf.src = kmap_local_page(in_page);
407 workspace->in_buf.pos = 0;
408 workspace->in_buf.size = min_t(size_t, len, PAGE_SIZE);
409
410
411 /* Allocate and map in the output buffer */
412 out_page = alloc_page(GFP_NOFS);
413 if (out_page == NULL) {
414 ret = -ENOMEM;
415 goto out;
416 }
417 pages[nr_pages++] = out_page;
418 workspace->out_buf.dst = page_address(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_compress_stream(stream, &workspace->out_buf,
426 &workspace->in_buf);
427 if (zstd_is_error(ret2)) {
428 pr_debug("BTRFS: zstd_compress_stream returned %d\n",
429 zstd_get_error_code(ret2));
430 ret = -EIO;
431 goto out;
432 }
433
434 /* Check to see if we are making it bigger */
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 /* We've reached the end of our output range */
443 if (workspace->out_buf.pos >= max_out) {
444 tot_out += workspace->out_buf.pos;
445 ret = -E2BIG;
446 goto out;
447 }
448
449 /* Check if we need more output space */
450 if (workspace->out_buf.pos == workspace->out_buf.size) {
451 tot_out += PAGE_SIZE;
452 max_out -= PAGE_SIZE;
453 if (nr_pages == nr_dest_pages) {
454 ret = -E2BIG;
455 goto out;
456 }
457 out_page = alloc_page(GFP_NOFS);
458 if (out_page == NULL) {
459 ret = -ENOMEM;
460 goto out;
461 }
462 pages[nr_pages++] = out_page;
463 workspace->out_buf.dst = page_address(out_page);
464 workspace->out_buf.pos = 0;
465 workspace->out_buf.size = min_t(size_t, max_out,
466 PAGE_SIZE);
467 }
468
469 /* We've reached the end of the input */
470 if (workspace->in_buf.pos >= len) {
471 tot_in += workspace->in_buf.pos;
472 break;
473 }
474
475 /* Check if we need more input */
476 if (workspace->in_buf.pos == workspace->in_buf.size) {
477 tot_in += PAGE_SIZE;
478 kunmap_local(workspace->in_buf.src);
479 put_page(in_page);
480 start += PAGE_SIZE;
481 len -= PAGE_SIZE;
482 in_page = find_get_page(mapping, start >> PAGE_SHIFT);
483 workspace->in_buf.src = kmap_local_page(in_page);
484 workspace->in_buf.pos = 0;
485 workspace->in_buf.size = min_t(size_t, len, PAGE_SIZE);
486 }
487 }
488 while (1) {
489 size_t ret2;
490
491 ret2 = zstd_end_stream(stream, &workspace->out_buf);
492 if (zstd_is_error(ret2)) {
493 pr_debug("BTRFS: zstd_end_stream returned %d\n",
494 zstd_get_error_code(ret2));
495 ret = -EIO;
496 goto out;
497 }
498 if (ret2 == 0) {
499 tot_out += workspace->out_buf.pos;
500 break;
501 }
502 if (workspace->out_buf.pos >= max_out) {
503 tot_out += workspace->out_buf.pos;
504 ret = -E2BIG;
505 goto out;
506 }
507
508 tot_out += PAGE_SIZE;
509 max_out -= PAGE_SIZE;
510 if (nr_pages == nr_dest_pages) {
511 ret = -E2BIG;
512 goto out;
513 }
514 out_page = alloc_page(GFP_NOFS);
515 if (out_page == NULL) {
516 ret = -ENOMEM;
517 goto out;
518 }
519 pages[nr_pages++] = out_page;
520 workspace->out_buf.dst = page_address(out_page);
521 workspace->out_buf.pos = 0;
522 workspace->out_buf.size = min_t(size_t, max_out, PAGE_SIZE);
523 }
524
525 if (tot_out >= tot_in) {
526 ret = -E2BIG;
527 goto out;
528 }
529
530 ret = 0;
531 *total_in = tot_in;
532 *total_out = tot_out;
533out:
534 *out_pages = nr_pages;
535 if (workspace->in_buf.src) {
536 kunmap_local(workspace->in_buf.src);
537 put_page(in_page);
538 }
539 return ret;
540}
541
542int zstd_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
543{
544 struct workspace *workspace = list_entry(ws, struct workspace, list);
545 struct page **pages_in = cb->compressed_pages;
546 size_t srclen = cb->compressed_len;
547 zstd_dstream *stream;
548 int ret = 0;
549 unsigned long page_in_index = 0;
550 unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE);
551 unsigned long buf_start;
552 unsigned long total_out = 0;
553
554 stream = zstd_init_dstream(
555 ZSTD_BTRFS_MAX_INPUT, workspace->mem, workspace->size);
556 if (!stream) {
557 pr_debug("BTRFS: zstd_init_dstream failed\n");
558 ret = -EIO;
559 goto done;
560 }
561
562 workspace->in_buf.src = kmap_local_page(pages_in[page_in_index]);
563 workspace->in_buf.pos = 0;
564 workspace->in_buf.size = min_t(size_t, srclen, PAGE_SIZE);
565
566 workspace->out_buf.dst = workspace->buf;
567 workspace->out_buf.pos = 0;
568 workspace->out_buf.size = PAGE_SIZE;
569
570 while (1) {
571 size_t ret2;
572
573 ret2 = zstd_decompress_stream(stream, &workspace->out_buf,
574 &workspace->in_buf);
575 if (zstd_is_error(ret2)) {
576 pr_debug("BTRFS: zstd_decompress_stream returned %d\n",
577 zstd_get_error_code(ret2));
578 ret = -EIO;
579 goto done;
580 }
581 buf_start = total_out;
582 total_out += workspace->out_buf.pos;
583 workspace->out_buf.pos = 0;
584
585 ret = btrfs_decompress_buf2page(workspace->out_buf.dst,
586 total_out - buf_start, cb, buf_start);
587 if (ret == 0)
588 break;
589
590 if (workspace->in_buf.pos >= srclen)
591 break;
592
593 /* Check if we've hit the end of a frame */
594 if (ret2 == 0)
595 break;
596
597 if (workspace->in_buf.pos == workspace->in_buf.size) {
598 kunmap_local(workspace->in_buf.src);
599 page_in_index++;
600 if (page_in_index >= total_pages_in) {
601 workspace->in_buf.src = NULL;
602 ret = -EIO;
603 goto done;
604 }
605 srclen -= PAGE_SIZE;
606 workspace->in_buf.src = kmap_local_page(pages_in[page_in_index]);
607 workspace->in_buf.pos = 0;
608 workspace->in_buf.size = min_t(size_t, srclen, PAGE_SIZE);
609 }
610 }
611 ret = 0;
612 zero_fill_bio(cb->orig_bio);
613done:
614 if (workspace->in_buf.src)
615 kunmap_local(workspace->in_buf.src);
616 return ret;
617}
618
619int zstd_decompress(struct list_head *ws, const u8 *data_in,
620 struct page *dest_page, unsigned long start_byte, size_t srclen,
621 size_t destlen)
622{
623 struct workspace *workspace = list_entry(ws, struct workspace, list);
624 zstd_dstream *stream;
625 int ret = 0;
626 size_t ret2;
627 unsigned long total_out = 0;
628 unsigned long pg_offset = 0;
629
630 stream = zstd_init_dstream(
631 ZSTD_BTRFS_MAX_INPUT, workspace->mem, workspace->size);
632 if (!stream) {
633 pr_warn("BTRFS: zstd_init_dstream failed\n");
634 ret = -EIO;
635 goto finish;
636 }
637
638 destlen = min_t(size_t, destlen, PAGE_SIZE);
639
640 workspace->in_buf.src = data_in;
641 workspace->in_buf.pos = 0;
642 workspace->in_buf.size = srclen;
643
644 workspace->out_buf.dst = workspace->buf;
645 workspace->out_buf.pos = 0;
646 workspace->out_buf.size = PAGE_SIZE;
647
648 ret2 = 1;
649 while (pg_offset < destlen
650 && workspace->in_buf.pos < workspace->in_buf.size) {
651 unsigned long buf_start;
652 unsigned long buf_offset;
653 unsigned long bytes;
654
655 /* Check if the frame is over and we still need more input */
656 if (ret2 == 0) {
657 pr_debug("BTRFS: zstd_decompress_stream ended early\n");
658 ret = -EIO;
659 goto finish;
660 }
661 ret2 = zstd_decompress_stream(stream, &workspace->out_buf,
662 &workspace->in_buf);
663 if (zstd_is_error(ret2)) {
664 pr_debug("BTRFS: zstd_decompress_stream returned %d\n",
665 zstd_get_error_code(ret2));
666 ret = -EIO;
667 goto finish;
668 }
669
670 buf_start = total_out;
671 total_out += workspace->out_buf.pos;
672 workspace->out_buf.pos = 0;
673
674 if (total_out <= start_byte)
675 continue;
676
677 if (total_out > start_byte && buf_start < start_byte)
678 buf_offset = start_byte - buf_start;
679 else
680 buf_offset = 0;
681
682 bytes = min_t(unsigned long, destlen - pg_offset,
683 workspace->out_buf.size - buf_offset);
684
685 memcpy_to_page(dest_page, pg_offset,
686 workspace->out_buf.dst + buf_offset, bytes);
687
688 pg_offset += bytes;
689 }
690 ret = 0;
691finish:
692 if (pg_offset < destlen) {
693 memzero_page(dest_page, pg_offset, destlen - pg_offset);
694 }
695 return ret;
696}
697
698const struct btrfs_compress_op btrfs_zstd_compress = {
699 /* ZSTD uses own workspace manager */
700 .workspace_manager = NULL,
701 .max_level = ZSTD_BTRFS_MAX_LEVEL,
702 .default_level = ZSTD_BTRFS_DEFAULT_LEVEL,
703};
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};