1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2016-present, Facebook, Inc.
9 #include <linux/bitmap.h>
10 #include <linux/err.h>
11 #include <linux/init.h>
12 #include <linux/kernel.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>
21 #include "compression.h"
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)
31 static ZSTD_parameters zstd_get_btrfs_parameters(unsigned int level,
34 ZSTD_parameters params = ZSTD_getParams(level, src_len, 0);
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);
47 unsigned int req_level;
48 unsigned long last_used; /* jiffies */
49 struct list_head list;
50 struct list_head lru_list;
52 ZSTD_outBuffer out_buf;
56 * Zstd Workspace Management
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
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.
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).
75 struct zstd_workspace_manager {
76 const struct btrfs_compress_op *ops;
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;
85 static struct zstd_workspace_manager wsm;
87 static size_t zstd_ws_mem_sizes[ZSTD_BTRFS_MAX_LEVEL];
89 static inline struct workspace *list_to_workspace(struct list_head *list)
91 return container_of(list, struct workspace, list);
94 void zstd_free_workspace(struct list_head *ws);
95 struct list_head *zstd_alloc_workspace(unsigned int level);
97 * zstd_reclaim_timer_fn - reclaim timer
100 * This scans the lru_list and attempts to reclaim any workspace that hasn't
101 * been used for ZSTD_BTRFS_RECLAIM_JIFFIES.
103 static void zstd_reclaim_timer_fn(struct timer_list *timer)
105 unsigned long reclaim_threshold = jiffies - ZSTD_BTRFS_RECLAIM_JIFFIES;
106 struct list_head *pos, *next;
108 spin_lock_bh(&wsm.lock);
110 if (list_empty(&wsm.lru_list)) {
111 spin_unlock_bh(&wsm.lock);
115 list_for_each_prev_safe(pos, next, &wsm.lru_list) {
116 struct workspace *victim = container_of(pos, struct workspace,
120 if (time_after(victim->last_used, reclaim_threshold))
123 /* workspace is in use */
124 if (victim->req_level)
127 level = victim->level;
128 list_del(&victim->lru_list);
129 list_del(&victim->list);
130 zstd_free_workspace(&victim->list);
132 if (list_empty(&wsm.idle_ws[level - 1]))
133 clear_bit(level - 1, &wsm.active_map);
137 if (!list_empty(&wsm.lru_list))
138 mod_timer(&wsm.timer, jiffies + ZSTD_BTRFS_RECLAIM_JIFFIES);
140 spin_unlock_bh(&wsm.lock);
144 * zstd_calc_ws_mem_sizes - calculate monotonic memory bounds
146 * It is possible based on the level configurations that a higher level
147 * workspace uses less memory than a lower level workspace. In order to reuse
148 * workspaces, this must be made a monotonic relationship. This precomputes
149 * the required memory for each level and enforces the monotonicity between
150 * level and memory required.
152 static void zstd_calc_ws_mem_sizes(void)
157 for (level = 1; level <= ZSTD_BTRFS_MAX_LEVEL; level++) {
158 ZSTD_parameters params =
159 zstd_get_btrfs_parameters(level, ZSTD_BTRFS_MAX_INPUT);
162 ZSTD_CStreamWorkspaceBound(params.cParams),
163 ZSTD_DStreamWorkspaceBound(ZSTD_BTRFS_MAX_INPUT));
165 max_size = max_t(size_t, max_size, level_size);
166 zstd_ws_mem_sizes[level - 1] = max_size;
170 void zstd_init_workspace_manager(void)
172 struct list_head *ws;
175 zstd_calc_ws_mem_sizes();
177 wsm.ops = &btrfs_zstd_compress;
178 spin_lock_init(&wsm.lock);
179 init_waitqueue_head(&wsm.wait);
180 timer_setup(&wsm.timer, zstd_reclaim_timer_fn, 0);
182 INIT_LIST_HEAD(&wsm.lru_list);
183 for (i = 0; i < ZSTD_BTRFS_MAX_LEVEL; i++)
184 INIT_LIST_HEAD(&wsm.idle_ws[i]);
186 ws = zstd_alloc_workspace(ZSTD_BTRFS_MAX_LEVEL);
189 "BTRFS: cannot preallocate zstd compression workspace\n");
191 set_bit(ZSTD_BTRFS_MAX_LEVEL - 1, &wsm.active_map);
192 list_add(ws, &wsm.idle_ws[ZSTD_BTRFS_MAX_LEVEL - 1]);
196 void zstd_cleanup_workspace_manager(void)
198 struct workspace *workspace;
201 spin_lock_bh(&wsm.lock);
202 for (i = 0; i < ZSTD_BTRFS_MAX_LEVEL; i++) {
203 while (!list_empty(&wsm.idle_ws[i])) {
204 workspace = container_of(wsm.idle_ws[i].next,
205 struct workspace, list);
206 list_del(&workspace->list);
207 list_del(&workspace->lru_list);
208 zstd_free_workspace(&workspace->list);
211 spin_unlock_bh(&wsm.lock);
213 del_timer_sync(&wsm.timer);
217 * zstd_find_workspace - find workspace
218 * @level: compression level
220 * This iterates over the set bits in the active_map beginning at the requested
221 * compression level. This lets us utilize already allocated workspaces before
222 * allocating a new one. If the workspace is of a larger size, it is used, but
223 * the place in the lru_list and last_used times are not updated. This is to
224 * offer the opportunity to reclaim the workspace in favor of allocating an
225 * appropriately sized one in the future.
227 static struct list_head *zstd_find_workspace(unsigned int level)
229 struct list_head *ws;
230 struct workspace *workspace;
233 spin_lock_bh(&wsm.lock);
234 for_each_set_bit_from(i, &wsm.active_map, ZSTD_BTRFS_MAX_LEVEL) {
235 if (!list_empty(&wsm.idle_ws[i])) {
236 ws = wsm.idle_ws[i].next;
237 workspace = list_to_workspace(ws);
239 /* keep its place if it's a lower level using this */
240 workspace->req_level = level;
241 if (level == workspace->level)
242 list_del(&workspace->lru_list);
243 if (list_empty(&wsm.idle_ws[i]))
244 clear_bit(i, &wsm.active_map);
245 spin_unlock_bh(&wsm.lock);
249 spin_unlock_bh(&wsm.lock);
255 * zstd_get_workspace - zstd's get_workspace
256 * @level: compression level
258 * If @level is 0, then any compression level can be used. Therefore, we begin
259 * scanning from 1. We first scan through possible workspaces and then after
260 * attempt to allocate a new workspace. If we fail to allocate one due to
261 * memory pressure, go to sleep waiting for the max level workspace to free up.
263 struct list_head *zstd_get_workspace(unsigned int level)
265 struct list_head *ws;
266 unsigned int nofs_flag;
268 /* level == 0 means we can use any workspace */
273 ws = zstd_find_workspace(level);
277 nofs_flag = memalloc_nofs_save();
278 ws = zstd_alloc_workspace(level);
279 memalloc_nofs_restore(nofs_flag);
284 prepare_to_wait(&wsm.wait, &wait, TASK_UNINTERRUPTIBLE);
286 finish_wait(&wsm.wait, &wait);
295 * zstd_put_workspace - zstd put_workspace
296 * @ws: list_head for the workspace
298 * When putting back a workspace, we only need to update the LRU if we are of
299 * the requested compression level. Here is where we continue to protect the
300 * max level workspace or update last_used accordingly. If the reclaim timer
301 * isn't set, it is also set here. Only the max level workspace tries and wakes
302 * up waiting workspaces.
304 void zstd_put_workspace(struct list_head *ws)
306 struct workspace *workspace = list_to_workspace(ws);
308 spin_lock_bh(&wsm.lock);
310 /* A node is only taken off the lru if we are the corresponding level */
311 if (workspace->req_level == workspace->level) {
312 /* Hide a max level workspace from reclaim */
313 if (list_empty(&wsm.idle_ws[ZSTD_BTRFS_MAX_LEVEL - 1])) {
314 INIT_LIST_HEAD(&workspace->lru_list);
316 workspace->last_used = jiffies;
317 list_add(&workspace->lru_list, &wsm.lru_list);
318 if (!timer_pending(&wsm.timer))
319 mod_timer(&wsm.timer,
320 jiffies + ZSTD_BTRFS_RECLAIM_JIFFIES);
324 set_bit(workspace->level - 1, &wsm.active_map);
325 list_add(&workspace->list, &wsm.idle_ws[workspace->level - 1]);
326 workspace->req_level = 0;
328 spin_unlock_bh(&wsm.lock);
330 if (workspace->level == ZSTD_BTRFS_MAX_LEVEL)
331 cond_wake_up(&wsm.wait);
334 void zstd_free_workspace(struct list_head *ws)
336 struct workspace *workspace = list_entry(ws, struct workspace, list);
338 kvfree(workspace->mem);
339 kfree(workspace->buf);
343 struct list_head *zstd_alloc_workspace(unsigned int level)
345 struct workspace *workspace;
347 workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
349 return ERR_PTR(-ENOMEM);
351 workspace->size = zstd_ws_mem_sizes[level - 1];
352 workspace->level = level;
353 workspace->req_level = level;
354 workspace->last_used = jiffies;
355 workspace->mem = kvmalloc(workspace->size, GFP_KERNEL);
356 workspace->buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
357 if (!workspace->mem || !workspace->buf)
360 INIT_LIST_HEAD(&workspace->list);
361 INIT_LIST_HEAD(&workspace->lru_list);
363 return &workspace->list;
365 zstd_free_workspace(&workspace->list);
366 return ERR_PTR(-ENOMEM);
369 int zstd_compress_pages(struct list_head *ws, struct address_space *mapping,
370 u64 start, struct page **pages, unsigned long *out_pages,
371 unsigned long *total_in, unsigned long *total_out)
373 struct workspace *workspace = list_entry(ws, struct workspace, list);
374 ZSTD_CStream *stream;
377 struct page *in_page = NULL; /* The current page to read */
378 struct page *out_page = NULL; /* The current page to write to */
379 unsigned long tot_in = 0;
380 unsigned long tot_out = 0;
381 unsigned long len = *total_out;
382 const unsigned long nr_dest_pages = *out_pages;
383 unsigned long max_out = nr_dest_pages * PAGE_SIZE;
384 ZSTD_parameters params = zstd_get_btrfs_parameters(workspace->req_level,
391 /* Initialize the stream */
392 stream = ZSTD_initCStream(params, len, workspace->mem,
395 pr_warn("BTRFS: ZSTD_initCStream failed\n");
400 /* map in the first page of input data */
401 in_page = find_get_page(mapping, start >> PAGE_SHIFT);
402 workspace->in_buf.src = kmap(in_page);
403 workspace->in_buf.pos = 0;
404 workspace->in_buf.size = min_t(size_t, len, PAGE_SIZE);
407 /* Allocate and map in the output buffer */
408 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
409 if (out_page == NULL) {
413 pages[nr_pages++] = out_page;
414 workspace->out_buf.dst = kmap(out_page);
415 workspace->out_buf.pos = 0;
416 workspace->out_buf.size = min_t(size_t, max_out, PAGE_SIZE);
421 ret2 = ZSTD_compressStream(stream, &workspace->out_buf,
423 if (ZSTD_isError(ret2)) {
424 pr_debug("BTRFS: ZSTD_compressStream returned %d\n",
425 ZSTD_getErrorCode(ret2));
430 /* Check to see if we are making it bigger */
431 if (tot_in + workspace->in_buf.pos > 8192 &&
432 tot_in + workspace->in_buf.pos <
433 tot_out + workspace->out_buf.pos) {
438 /* We've reached the end of our output range */
439 if (workspace->out_buf.pos >= max_out) {
440 tot_out += workspace->out_buf.pos;
445 /* Check if we need more output space */
446 if (workspace->out_buf.pos == workspace->out_buf.size) {
447 tot_out += PAGE_SIZE;
448 max_out -= PAGE_SIZE;
450 if (nr_pages == nr_dest_pages) {
455 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
456 if (out_page == NULL) {
460 pages[nr_pages++] = out_page;
461 workspace->out_buf.dst = kmap(out_page);
462 workspace->out_buf.pos = 0;
463 workspace->out_buf.size = min_t(size_t, max_out,
467 /* We've reached the end of the input */
468 if (workspace->in_buf.pos >= len) {
469 tot_in += workspace->in_buf.pos;
473 /* Check if we need more input */
474 if (workspace->in_buf.pos == workspace->in_buf.size) {
481 in_page = find_get_page(mapping, start >> PAGE_SHIFT);
482 workspace->in_buf.src = kmap(in_page);
483 workspace->in_buf.pos = 0;
484 workspace->in_buf.size = min_t(size_t, len, PAGE_SIZE);
490 ret2 = ZSTD_endStream(stream, &workspace->out_buf);
491 if (ZSTD_isError(ret2)) {
492 pr_debug("BTRFS: ZSTD_endStream returned %d\n",
493 ZSTD_getErrorCode(ret2));
498 tot_out += workspace->out_buf.pos;
501 if (workspace->out_buf.pos >= max_out) {
502 tot_out += workspace->out_buf.pos;
507 tot_out += PAGE_SIZE;
508 max_out -= PAGE_SIZE;
510 if (nr_pages == nr_dest_pages) {
515 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
516 if (out_page == NULL) {
520 pages[nr_pages++] = out_page;
521 workspace->out_buf.dst = kmap(out_page);
522 workspace->out_buf.pos = 0;
523 workspace->out_buf.size = min_t(size_t, max_out, PAGE_SIZE);
526 if (tot_out >= tot_in) {
533 *total_out = tot_out;
535 *out_pages = nr_pages;
546 int zstd_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
548 struct workspace *workspace = list_entry(ws, struct workspace, list);
549 struct page **pages_in = cb->compressed_pages;
550 u64 disk_start = cb->start;
551 struct bio *orig_bio = cb->orig_bio;
552 size_t srclen = cb->compressed_len;
553 ZSTD_DStream *stream;
555 unsigned long page_in_index = 0;
556 unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE);
557 unsigned long buf_start;
558 unsigned long total_out = 0;
560 stream = ZSTD_initDStream(
561 ZSTD_BTRFS_MAX_INPUT, workspace->mem, workspace->size);
563 pr_debug("BTRFS: ZSTD_initDStream failed\n");
568 workspace->in_buf.src = kmap(pages_in[page_in_index]);
569 workspace->in_buf.pos = 0;
570 workspace->in_buf.size = min_t(size_t, srclen, PAGE_SIZE);
572 workspace->out_buf.dst = workspace->buf;
573 workspace->out_buf.pos = 0;
574 workspace->out_buf.size = PAGE_SIZE;
579 ret2 = ZSTD_decompressStream(stream, &workspace->out_buf,
581 if (ZSTD_isError(ret2)) {
582 pr_debug("BTRFS: ZSTD_decompressStream returned %d\n",
583 ZSTD_getErrorCode(ret2));
587 buf_start = total_out;
588 total_out += workspace->out_buf.pos;
589 workspace->out_buf.pos = 0;
591 ret = btrfs_decompress_buf2page(workspace->out_buf.dst,
592 buf_start, total_out, disk_start, orig_bio);
596 if (workspace->in_buf.pos >= srclen)
599 /* Check if we've hit the end of a frame */
603 if (workspace->in_buf.pos == workspace->in_buf.size) {
604 kunmap(pages_in[page_in_index++]);
605 if (page_in_index >= total_pages_in) {
606 workspace->in_buf.src = NULL;
611 workspace->in_buf.src = kmap(pages_in[page_in_index]);
612 workspace->in_buf.pos = 0;
613 workspace->in_buf.size = min_t(size_t, srclen, PAGE_SIZE);
617 zero_fill_bio(orig_bio);
619 if (workspace->in_buf.src)
620 kunmap(pages_in[page_in_index]);
624 int zstd_decompress(struct list_head *ws, unsigned char *data_in,
625 struct page *dest_page, unsigned long start_byte, size_t srclen,
628 struct workspace *workspace = list_entry(ws, struct workspace, list);
629 ZSTD_DStream *stream;
632 unsigned long total_out = 0;
633 unsigned long pg_offset = 0;
636 stream = ZSTD_initDStream(
637 ZSTD_BTRFS_MAX_INPUT, workspace->mem, workspace->size);
639 pr_warn("BTRFS: ZSTD_initDStream failed\n");
644 destlen = min_t(size_t, destlen, PAGE_SIZE);
646 workspace->in_buf.src = data_in;
647 workspace->in_buf.pos = 0;
648 workspace->in_buf.size = srclen;
650 workspace->out_buf.dst = workspace->buf;
651 workspace->out_buf.pos = 0;
652 workspace->out_buf.size = PAGE_SIZE;
655 while (pg_offset < destlen
656 && workspace->in_buf.pos < workspace->in_buf.size) {
657 unsigned long buf_start;
658 unsigned long buf_offset;
661 /* Check if the frame is over and we still need more input */
663 pr_debug("BTRFS: ZSTD_decompressStream ended early\n");
667 ret2 = ZSTD_decompressStream(stream, &workspace->out_buf,
669 if (ZSTD_isError(ret2)) {
670 pr_debug("BTRFS: ZSTD_decompressStream returned %d\n",
671 ZSTD_getErrorCode(ret2));
676 buf_start = total_out;
677 total_out += workspace->out_buf.pos;
678 workspace->out_buf.pos = 0;
680 if (total_out <= start_byte)
683 if (total_out > start_byte && buf_start < start_byte)
684 buf_offset = start_byte - buf_start;
688 bytes = min_t(unsigned long, destlen - pg_offset,
689 workspace->out_buf.size - buf_offset);
691 kaddr = kmap_atomic(dest_page);
692 memcpy(kaddr + pg_offset, workspace->out_buf.dst + buf_offset,
694 kunmap_atomic(kaddr);
700 if (pg_offset < destlen) {
701 kaddr = kmap_atomic(dest_page);
702 memset(kaddr + pg_offset, 0, destlen - pg_offset);
703 kunmap_atomic(kaddr);
708 const struct btrfs_compress_op btrfs_zstd_compress = {
709 /* ZSTD uses own workspace manager */
710 .workspace_manager = NULL,
711 .max_level = ZSTD_BTRFS_MAX_LEVEL,
712 .default_level = ZSTD_BTRFS_DEFAULT_LEVEL,