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 "fs.h"
22#include "btrfs_inode.h"
23#include "compression.h"
24#include "super.h"
25
26#define ZSTD_BTRFS_MAX_WINDOWLOG 17
27#define ZSTD_BTRFS_MAX_INPUT (1 << ZSTD_BTRFS_MAX_WINDOWLOG)
28#define ZSTD_BTRFS_DEFAULT_LEVEL 3
29#define ZSTD_BTRFS_MAX_LEVEL 15
30/* 307s to avoid pathologically clashing with transaction commit */
31#define ZSTD_BTRFS_RECLAIM_JIFFIES (307 * HZ)
32
33static zstd_parameters zstd_get_btrfs_parameters(unsigned int level,
34 size_t src_len)
35{
36 zstd_parameters params = zstd_get_params(level, src_len);
37
38 if (params.cParams.windowLog > ZSTD_BTRFS_MAX_WINDOWLOG)
39 params.cParams.windowLog = ZSTD_BTRFS_MAX_WINDOWLOG;
40 WARN_ON(src_len > ZSTD_BTRFS_MAX_INPUT);
41 return params;
42}
43
44struct workspace {
45 void *mem;
46 size_t size;
47 char *buf;
48 unsigned int level;
49 unsigned int req_level;
50 unsigned long last_used; /* jiffies */
51 struct list_head list;
52 struct list_head lru_list;
53 zstd_in_buffer in_buf;
54 zstd_out_buffer out_buf;
55};
56
57/*
58 * Zstd Workspace Management
59 *
60 * Zstd workspaces have different memory requirements depending on the level.
61 * The zstd workspaces are managed by having individual lists for each level
62 * and a global lru. Forward progress is maintained by protecting a max level
63 * workspace.
64 *
65 * Getting a workspace is done by using the bitmap to identify the levels that
66 * have available workspaces and scans up. This lets us recycle higher level
67 * workspaces because of the monotonic memory guarantee. A workspace's
68 * last_used is only updated if it is being used by the corresponding memory
69 * level. Putting a workspace involves adding it back to the appropriate places
70 * and adding it back to the lru if necessary.
71 *
72 * A timer is used to reclaim workspaces if they have not been used for
73 * ZSTD_BTRFS_RECLAIM_JIFFIES. This helps keep only active workspaces around.
74 * The upper bound is provided by the workqueue limit which is 2 (percpu limit).
75 */
76
77struct zstd_workspace_manager {
78 const struct btrfs_compress_op *ops;
79 spinlock_t lock;
80 struct list_head lru_list;
81 struct list_head idle_ws[ZSTD_BTRFS_MAX_LEVEL];
82 unsigned long active_map;
83 wait_queue_head_t wait;
84 struct timer_list timer;
85};
86
87static struct zstd_workspace_manager wsm;
88
89static size_t zstd_ws_mem_sizes[ZSTD_BTRFS_MAX_LEVEL];
90
91static inline struct workspace *list_to_workspace(struct list_head *list)
92{
93 return container_of(list, struct workspace, list);
94}
95
96void zstd_free_workspace(struct list_head *ws);
97struct list_head *zstd_alloc_workspace(unsigned int level);
98
99/*
100 * Timer callback to free unused workspaces.
101 *
102 * @t: timer
103 *
104 * This scans the lru_list and attempts to reclaim any workspace that hasn't
105 * been used for ZSTD_BTRFS_RECLAIM_JIFFIES.
106 *
107 * The context is softirq and does not need the _bh locking primitives.
108 */
109static void zstd_reclaim_timer_fn(struct timer_list *timer)
110{
111 unsigned long reclaim_threshold = jiffies - ZSTD_BTRFS_RECLAIM_JIFFIES;
112 struct list_head *pos, *next;
113
114 ASSERT(timer == &wsm.timer);
115
116 spin_lock(&wsm.lock);
117
118 if (list_empty(&wsm.lru_list)) {
119 spin_unlock(&wsm.lock);
120 return;
121 }
122
123 list_for_each_prev_safe(pos, next, &wsm.lru_list) {
124 struct workspace *victim = container_of(pos, struct workspace,
125 lru_list);
126 unsigned int level;
127
128 if (time_after(victim->last_used, reclaim_threshold))
129 break;
130
131 /* workspace is in use */
132 if (victim->req_level)
133 continue;
134
135 level = victim->level;
136 list_del(&victim->lru_list);
137 list_del(&victim->list);
138 zstd_free_workspace(&victim->list);
139
140 if (list_empty(&wsm.idle_ws[level - 1]))
141 clear_bit(level - 1, &wsm.active_map);
142
143 }
144
145 if (!list_empty(&wsm.lru_list))
146 mod_timer(&wsm.timer, jiffies + ZSTD_BTRFS_RECLAIM_JIFFIES);
147
148 spin_unlock(&wsm.lock);
149}
150
151/*
152 * Calculate monotonic memory bounds.
153 *
154 * It is possible based on the level configurations that a higher level
155 * workspace uses less memory than a lower level workspace. In order to reuse
156 * workspaces, this must be made a monotonic relationship. This precomputes
157 * the required memory for each level and enforces the monotonicity between
158 * level and memory required.
159 */
160static void zstd_calc_ws_mem_sizes(void)
161{
162 size_t max_size = 0;
163 unsigned int level;
164
165 for (level = 1; level <= ZSTD_BTRFS_MAX_LEVEL; level++) {
166 zstd_parameters params =
167 zstd_get_btrfs_parameters(level, ZSTD_BTRFS_MAX_INPUT);
168 size_t level_size =
169 max_t(size_t,
170 zstd_cstream_workspace_bound(¶ms.cParams),
171 zstd_dstream_workspace_bound(ZSTD_BTRFS_MAX_INPUT));
172
173 max_size = max_t(size_t, max_size, level_size);
174 zstd_ws_mem_sizes[level - 1] = max_size;
175 }
176}
177
178void zstd_init_workspace_manager(void)
179{
180 struct list_head *ws;
181 int i;
182
183 zstd_calc_ws_mem_sizes();
184
185 wsm.ops = &btrfs_zstd_compress;
186 spin_lock_init(&wsm.lock);
187 init_waitqueue_head(&wsm.wait);
188 timer_setup(&wsm.timer, zstd_reclaim_timer_fn, 0);
189
190 INIT_LIST_HEAD(&wsm.lru_list);
191 for (i = 0; i < ZSTD_BTRFS_MAX_LEVEL; i++)
192 INIT_LIST_HEAD(&wsm.idle_ws[i]);
193
194 ws = zstd_alloc_workspace(ZSTD_BTRFS_MAX_LEVEL);
195 if (IS_ERR(ws)) {
196 pr_warn(
197 "BTRFS: cannot preallocate zstd compression workspace\n");
198 } else {
199 set_bit(ZSTD_BTRFS_MAX_LEVEL - 1, &wsm.active_map);
200 list_add(ws, &wsm.idle_ws[ZSTD_BTRFS_MAX_LEVEL - 1]);
201 }
202}
203
204void zstd_cleanup_workspace_manager(void)
205{
206 struct workspace *workspace;
207 int i;
208
209 spin_lock_bh(&wsm.lock);
210 for (i = 0; i < ZSTD_BTRFS_MAX_LEVEL; i++) {
211 while (!list_empty(&wsm.idle_ws[i])) {
212 workspace = container_of(wsm.idle_ws[i].next,
213 struct workspace, list);
214 list_del(&workspace->list);
215 list_del(&workspace->lru_list);
216 zstd_free_workspace(&workspace->list);
217 }
218 }
219 spin_unlock_bh(&wsm.lock);
220
221 del_timer_sync(&wsm.timer);
222}
223
224/*
225 * Find workspace for given level.
226 *
227 * @level: compression level
228 *
229 * This iterates over the set bits in the active_map beginning at the requested
230 * compression level. This lets us utilize already allocated workspaces before
231 * allocating a new one. If the workspace is of a larger size, it is used, but
232 * the place in the lru_list and last_used times are not updated. This is to
233 * offer the opportunity to reclaim the workspace in favor of allocating an
234 * appropriately sized one in the future.
235 */
236static struct list_head *zstd_find_workspace(unsigned int level)
237{
238 struct list_head *ws;
239 struct workspace *workspace;
240 int i = level - 1;
241
242 spin_lock_bh(&wsm.lock);
243 for_each_set_bit_from(i, &wsm.active_map, ZSTD_BTRFS_MAX_LEVEL) {
244 if (!list_empty(&wsm.idle_ws[i])) {
245 ws = wsm.idle_ws[i].next;
246 workspace = list_to_workspace(ws);
247 list_del_init(ws);
248 /* keep its place if it's a lower level using this */
249 workspace->req_level = level;
250 if (level == workspace->level)
251 list_del(&workspace->lru_list);
252 if (list_empty(&wsm.idle_ws[i]))
253 clear_bit(i, &wsm.active_map);
254 spin_unlock_bh(&wsm.lock);
255 return ws;
256 }
257 }
258 spin_unlock_bh(&wsm.lock);
259
260 return NULL;
261}
262
263/*
264 * Zstd get_workspace for level.
265 *
266 * @level: compression level
267 *
268 * If @level is 0, then any compression level can be used. Therefore, we begin
269 * scanning from 1. We first scan through possible workspaces and then after
270 * attempt to allocate a new workspace. If we fail to allocate one due to
271 * memory pressure, go to sleep waiting for the max level workspace to free up.
272 */
273struct list_head *zstd_get_workspace(unsigned int level)
274{
275 struct list_head *ws;
276 unsigned int nofs_flag;
277
278 /* level == 0 means we can use any workspace */
279 if (!level)
280 level = 1;
281
282again:
283 ws = zstd_find_workspace(level);
284 if (ws)
285 return ws;
286
287 nofs_flag = memalloc_nofs_save();
288 ws = zstd_alloc_workspace(level);
289 memalloc_nofs_restore(nofs_flag);
290
291 if (IS_ERR(ws)) {
292 DEFINE_WAIT(wait);
293
294 prepare_to_wait(&wsm.wait, &wait, TASK_UNINTERRUPTIBLE);
295 schedule();
296 finish_wait(&wsm.wait, &wait);
297
298 goto again;
299 }
300
301 return ws;
302}
303
304/*
305 * Zstd put_workspace.
306 *
307 * @ws: list_head for the workspace
308 *
309 * When putting back a workspace, we only need to update the LRU if we are of
310 * the requested compression level. Here is where we continue to protect the
311 * max level workspace or update last_used accordingly. If the reclaim timer
312 * isn't set, it is also set here. Only the max level workspace tries and wakes
313 * up waiting workspaces.
314 */
315void zstd_put_workspace(struct list_head *ws)
316{
317 struct workspace *workspace = list_to_workspace(ws);
318
319 spin_lock_bh(&wsm.lock);
320
321 /* A node is only taken off the lru if we are the corresponding level */
322 if (workspace->req_level == workspace->level) {
323 /* Hide a max level workspace from reclaim */
324 if (list_empty(&wsm.idle_ws[ZSTD_BTRFS_MAX_LEVEL - 1])) {
325 INIT_LIST_HEAD(&workspace->lru_list);
326 } else {
327 workspace->last_used = jiffies;
328 list_add(&workspace->lru_list, &wsm.lru_list);
329 if (!timer_pending(&wsm.timer))
330 mod_timer(&wsm.timer,
331 jiffies + ZSTD_BTRFS_RECLAIM_JIFFIES);
332 }
333 }
334
335 set_bit(workspace->level - 1, &wsm.active_map);
336 list_add(&workspace->list, &wsm.idle_ws[workspace->level - 1]);
337 workspace->req_level = 0;
338
339 spin_unlock_bh(&wsm.lock);
340
341 if (workspace->level == ZSTD_BTRFS_MAX_LEVEL)
342 cond_wake_up(&wsm.wait);
343}
344
345void zstd_free_workspace(struct list_head *ws)
346{
347 struct workspace *workspace = list_entry(ws, struct workspace, list);
348
349 kvfree(workspace->mem);
350 kfree(workspace->buf);
351 kfree(workspace);
352}
353
354struct list_head *zstd_alloc_workspace(unsigned int level)
355{
356 struct workspace *workspace;
357
358 workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
359 if (!workspace)
360 return ERR_PTR(-ENOMEM);
361
362 workspace->size = zstd_ws_mem_sizes[level - 1];
363 workspace->level = level;
364 workspace->req_level = level;
365 workspace->last_used = jiffies;
366 workspace->mem = kvmalloc(workspace->size, GFP_KERNEL | __GFP_NOWARN);
367 workspace->buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
368 if (!workspace->mem || !workspace->buf)
369 goto fail;
370
371 INIT_LIST_HEAD(&workspace->list);
372 INIT_LIST_HEAD(&workspace->lru_list);
373
374 return &workspace->list;
375fail:
376 zstd_free_workspace(&workspace->list);
377 return ERR_PTR(-ENOMEM);
378}
379
380int zstd_compress_folios(struct list_head *ws, struct address_space *mapping,
381 u64 start, struct folio **folios, unsigned long *out_folios,
382 unsigned long *total_in, unsigned long *total_out)
383{
384 struct workspace *workspace = list_entry(ws, struct workspace, list);
385 zstd_cstream *stream;
386 int ret = 0;
387 int nr_folios = 0;
388 struct folio *in_folio = NULL; /* The current folio to read. */
389 struct folio *out_folio = NULL; /* The current folio to write to. */
390 unsigned long tot_in = 0;
391 unsigned long tot_out = 0;
392 unsigned long len = *total_out;
393 const unsigned long nr_dest_folios = *out_folios;
394 const u64 orig_end = start + len;
395 unsigned long max_out = nr_dest_folios * PAGE_SIZE;
396 unsigned int pg_off;
397 unsigned int cur_len;
398 zstd_parameters params = zstd_get_btrfs_parameters(workspace->req_level,
399 len);
400
401 *out_folios = 0;
402 *total_out = 0;
403 *total_in = 0;
404
405 /* Initialize the stream */
406 stream = zstd_init_cstream(¶ms, len, workspace->mem,
407 workspace->size);
408 if (unlikely(!stream)) {
409 struct btrfs_inode *inode = BTRFS_I(mapping->host);
410
411 btrfs_err(inode->root->fs_info,
412 "zstd compression init level %d failed, root %llu inode %llu offset %llu",
413 workspace->req_level, btrfs_root_id(inode->root),
414 btrfs_ino(inode), start);
415 ret = -EIO;
416 goto out;
417 }
418
419 /* map in the first page of input data */
420 ret = btrfs_compress_filemap_get_folio(mapping, start, &in_folio);
421 if (ret < 0)
422 goto out;
423 pg_off = offset_in_page(start);
424 cur_len = btrfs_calc_input_length(orig_end, start);
425 workspace->in_buf.src = kmap_local_folio(in_folio, pg_off);
426 workspace->in_buf.pos = 0;
427 workspace->in_buf.size = cur_len;
428
429 /* Allocate and map in the output buffer */
430 out_folio = btrfs_alloc_compr_folio();
431 if (out_folio == NULL) {
432 ret = -ENOMEM;
433 goto out;
434 }
435 folios[nr_folios++] = out_folio;
436 workspace->out_buf.dst = folio_address(out_folio);
437 workspace->out_buf.pos = 0;
438 workspace->out_buf.size = min_t(size_t, max_out, PAGE_SIZE);
439
440 while (1) {
441 size_t ret2;
442
443 ret2 = zstd_compress_stream(stream, &workspace->out_buf,
444 &workspace->in_buf);
445 if (unlikely(zstd_is_error(ret2))) {
446 struct btrfs_inode *inode = BTRFS_I(mapping->host);
447
448 btrfs_warn(inode->root->fs_info,
449"zstd compression level %d failed, error %d root %llu inode %llu offset %llu",
450 workspace->req_level, zstd_get_error_code(ret2),
451 btrfs_root_id(inode->root), btrfs_ino(inode),
452 start);
453 ret = -EIO;
454 goto out;
455 }
456
457 /* Check to see if we are making it bigger */
458 if (tot_in + workspace->in_buf.pos > 8192 &&
459 tot_in + workspace->in_buf.pos <
460 tot_out + workspace->out_buf.pos) {
461 ret = -E2BIG;
462 goto out;
463 }
464
465 /* We've reached the end of our output range */
466 if (workspace->out_buf.pos >= max_out) {
467 tot_out += workspace->out_buf.pos;
468 ret = -E2BIG;
469 goto out;
470 }
471
472 /* Check if we need more output space */
473 if (workspace->out_buf.pos == workspace->out_buf.size) {
474 tot_out += PAGE_SIZE;
475 max_out -= PAGE_SIZE;
476 if (nr_folios == nr_dest_folios) {
477 ret = -E2BIG;
478 goto out;
479 }
480 out_folio = btrfs_alloc_compr_folio();
481 if (out_folio == NULL) {
482 ret = -ENOMEM;
483 goto out;
484 }
485 folios[nr_folios++] = out_folio;
486 workspace->out_buf.dst = folio_address(out_folio);
487 workspace->out_buf.pos = 0;
488 workspace->out_buf.size = min_t(size_t, max_out,
489 PAGE_SIZE);
490 }
491
492 /* We've reached the end of the input */
493 if (workspace->in_buf.pos >= len) {
494 tot_in += workspace->in_buf.pos;
495 break;
496 }
497
498 /* Check if we need more input */
499 if (workspace->in_buf.pos == workspace->in_buf.size) {
500 tot_in += workspace->in_buf.size;
501 kunmap_local(workspace->in_buf.src);
502 workspace->in_buf.src = NULL;
503 folio_put(in_folio);
504 start += cur_len;
505 len -= cur_len;
506 ret = btrfs_compress_filemap_get_folio(mapping, start, &in_folio);
507 if (ret < 0)
508 goto out;
509 pg_off = offset_in_page(start);
510 cur_len = btrfs_calc_input_length(orig_end, start);
511 workspace->in_buf.src = kmap_local_folio(in_folio, pg_off);
512 workspace->in_buf.pos = 0;
513 workspace->in_buf.size = cur_len;
514 }
515 }
516 while (1) {
517 size_t ret2;
518
519 ret2 = zstd_end_stream(stream, &workspace->out_buf);
520 if (unlikely(zstd_is_error(ret2))) {
521 struct btrfs_inode *inode = BTRFS_I(mapping->host);
522
523 btrfs_err(inode->root->fs_info,
524"zstd compression end level %d failed, error %d root %llu inode %llu offset %llu",
525 workspace->req_level, zstd_get_error_code(ret2),
526 btrfs_root_id(inode->root), btrfs_ino(inode),
527 start);
528 ret = -EIO;
529 goto out;
530 }
531 if (ret2 == 0) {
532 tot_out += workspace->out_buf.pos;
533 break;
534 }
535 if (workspace->out_buf.pos >= max_out) {
536 tot_out += workspace->out_buf.pos;
537 ret = -E2BIG;
538 goto out;
539 }
540
541 tot_out += PAGE_SIZE;
542 max_out -= PAGE_SIZE;
543 if (nr_folios == nr_dest_folios) {
544 ret = -E2BIG;
545 goto out;
546 }
547 out_folio = btrfs_alloc_compr_folio();
548 if (out_folio == NULL) {
549 ret = -ENOMEM;
550 goto out;
551 }
552 folios[nr_folios++] = out_folio;
553 workspace->out_buf.dst = folio_address(out_folio);
554 workspace->out_buf.pos = 0;
555 workspace->out_buf.size = min_t(size_t, max_out, PAGE_SIZE);
556 }
557
558 if (tot_out >= tot_in) {
559 ret = -E2BIG;
560 goto out;
561 }
562
563 ret = 0;
564 *total_in = tot_in;
565 *total_out = tot_out;
566out:
567 *out_folios = nr_folios;
568 if (workspace->in_buf.src) {
569 kunmap_local(workspace->in_buf.src);
570 folio_put(in_folio);
571 }
572 return ret;
573}
574
575int zstd_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
576{
577 struct workspace *workspace = list_entry(ws, struct workspace, list);
578 struct folio **folios_in = cb->compressed_folios;
579 size_t srclen = cb->compressed_len;
580 zstd_dstream *stream;
581 int ret = 0;
582 unsigned long folio_in_index = 0;
583 unsigned long total_folios_in = DIV_ROUND_UP(srclen, PAGE_SIZE);
584 unsigned long buf_start;
585 unsigned long total_out = 0;
586
587 stream = zstd_init_dstream(
588 ZSTD_BTRFS_MAX_INPUT, workspace->mem, workspace->size);
589 if (unlikely(!stream)) {
590 struct btrfs_inode *inode = cb->bbio.inode;
591
592 btrfs_err(inode->root->fs_info,
593 "zstd decompression init failed, root %llu inode %llu offset %llu",
594 btrfs_root_id(inode->root), btrfs_ino(inode), cb->start);
595 ret = -EIO;
596 goto done;
597 }
598
599 workspace->in_buf.src = kmap_local_folio(folios_in[folio_in_index], 0);
600 workspace->in_buf.pos = 0;
601 workspace->in_buf.size = min_t(size_t, srclen, PAGE_SIZE);
602
603 workspace->out_buf.dst = workspace->buf;
604 workspace->out_buf.pos = 0;
605 workspace->out_buf.size = PAGE_SIZE;
606
607 while (1) {
608 size_t ret2;
609
610 ret2 = zstd_decompress_stream(stream, &workspace->out_buf,
611 &workspace->in_buf);
612 if (unlikely(zstd_is_error(ret2))) {
613 struct btrfs_inode *inode = cb->bbio.inode;
614
615 btrfs_err(inode->root->fs_info,
616 "zstd decompression failed, error %d root %llu inode %llu offset %llu",
617 zstd_get_error_code(ret2), btrfs_root_id(inode->root),
618 btrfs_ino(inode), cb->start);
619 ret = -EIO;
620 goto done;
621 }
622 buf_start = total_out;
623 total_out += workspace->out_buf.pos;
624 workspace->out_buf.pos = 0;
625
626 ret = btrfs_decompress_buf2page(workspace->out_buf.dst,
627 total_out - buf_start, cb, buf_start);
628 if (ret == 0)
629 break;
630
631 if (workspace->in_buf.pos >= srclen)
632 break;
633
634 /* Check if we've hit the end of a frame */
635 if (ret2 == 0)
636 break;
637
638 if (workspace->in_buf.pos == workspace->in_buf.size) {
639 kunmap_local(workspace->in_buf.src);
640 folio_in_index++;
641 if (folio_in_index >= total_folios_in) {
642 workspace->in_buf.src = NULL;
643 ret = -EIO;
644 goto done;
645 }
646 srclen -= PAGE_SIZE;
647 workspace->in_buf.src =
648 kmap_local_folio(folios_in[folio_in_index], 0);
649 workspace->in_buf.pos = 0;
650 workspace->in_buf.size = min_t(size_t, srclen, PAGE_SIZE);
651 }
652 }
653 ret = 0;
654done:
655 if (workspace->in_buf.src)
656 kunmap_local(workspace->in_buf.src);
657 return ret;
658}
659
660int zstd_decompress(struct list_head *ws, const u8 *data_in,
661 struct folio *dest_folio, unsigned long dest_pgoff, size_t srclen,
662 size_t destlen)
663{
664 struct workspace *workspace = list_entry(ws, struct workspace, list);
665 struct btrfs_fs_info *fs_info = btrfs_sb(folio_inode(dest_folio)->i_sb);
666 const u32 sectorsize = fs_info->sectorsize;
667 zstd_dstream *stream;
668 int ret = 0;
669 unsigned long to_copy = 0;
670
671 stream = zstd_init_dstream(
672 ZSTD_BTRFS_MAX_INPUT, workspace->mem, workspace->size);
673 if (unlikely(!stream)) {
674 struct btrfs_inode *inode = folio_to_inode(dest_folio);
675
676 btrfs_err(inode->root->fs_info,
677 "zstd decompression init failed, root %llu inode %llu offset %llu",
678 btrfs_root_id(inode->root), btrfs_ino(inode),
679 folio_pos(dest_folio));
680 ret = -EIO;
681 goto finish;
682 }
683
684 workspace->in_buf.src = data_in;
685 workspace->in_buf.pos = 0;
686 workspace->in_buf.size = srclen;
687
688 workspace->out_buf.dst = workspace->buf;
689 workspace->out_buf.pos = 0;
690 workspace->out_buf.size = sectorsize;
691
692 /*
693 * Since both input and output buffers should not exceed one sector,
694 * one call should end the decompression.
695 */
696 ret = zstd_decompress_stream(stream, &workspace->out_buf, &workspace->in_buf);
697 if (unlikely(zstd_is_error(ret))) {
698 struct btrfs_inode *inode = folio_to_inode(dest_folio);
699
700 btrfs_err(inode->root->fs_info,
701 "zstd decompression failed, error %d root %llu inode %llu offset %llu",
702 zstd_get_error_code(ret), btrfs_root_id(inode->root),
703 btrfs_ino(inode), folio_pos(dest_folio));
704 goto finish;
705 }
706 to_copy = workspace->out_buf.pos;
707 memcpy_to_folio(dest_folio, dest_pgoff, workspace->out_buf.dst, to_copy);
708finish:
709 /* Error or early end. */
710 if (unlikely(to_copy < destlen)) {
711 ret = -EIO;
712 folio_zero_range(dest_folio, dest_pgoff + to_copy, destlen - to_copy);
713 }
714 return ret;
715}
716
717const struct btrfs_compress_op btrfs_zstd_compress = {
718 /* ZSTD uses own workspace manager */
719 .workspace_manager = NULL,
720 .max_level = ZSTD_BTRFS_MAX_LEVEL,
721 .default_level = ZSTD_BTRFS_DEFAULT_LEVEL,
722};
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};