1 // SPDX-License-Identifier: GPL-2.0
3 * f2fs compress support
5 * Copyright (c) 2019 Chao Yu <chao@kernel.org>
9 #include <linux/f2fs_fs.h>
10 #include <linux/writeback.h>
11 #include <linux/backing-dev.h>
12 #include <linux/lzo.h>
13 #include <linux/lz4.h>
14 #include <linux/zstd.h>
18 #include <trace/events/f2fs.h>
20 static struct kmem_cache *cic_entry_slab;
21 static struct kmem_cache *dic_entry_slab;
23 static void *page_array_alloc(struct inode *inode, int nr)
25 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
26 unsigned int size = sizeof(struct page *) * nr;
28 if (likely(size <= sbi->page_array_slab_size))
29 return kmem_cache_zalloc(sbi->page_array_slab, GFP_NOFS);
30 return f2fs_kzalloc(sbi, size, GFP_NOFS);
33 static void page_array_free(struct inode *inode, void *pages, int nr)
35 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
36 unsigned int size = sizeof(struct page *) * nr;
41 if (likely(size <= sbi->page_array_slab_size))
42 kmem_cache_free(sbi->page_array_slab, pages);
47 struct f2fs_compress_ops {
48 int (*init_compress_ctx)(struct compress_ctx *cc);
49 void (*destroy_compress_ctx)(struct compress_ctx *cc);
50 int (*compress_pages)(struct compress_ctx *cc);
51 int (*init_decompress_ctx)(struct decompress_io_ctx *dic);
52 void (*destroy_decompress_ctx)(struct decompress_io_ctx *dic);
53 int (*decompress_pages)(struct decompress_io_ctx *dic);
56 static unsigned int offset_in_cluster(struct compress_ctx *cc, pgoff_t index)
58 return index & (cc->cluster_size - 1);
61 static pgoff_t cluster_idx(struct compress_ctx *cc, pgoff_t index)
63 return index >> cc->log_cluster_size;
66 static pgoff_t start_idx_of_cluster(struct compress_ctx *cc)
68 return cc->cluster_idx << cc->log_cluster_size;
71 bool f2fs_is_compressed_page(struct page *page)
73 if (!PagePrivate(page))
75 if (!page_private(page))
77 if (IS_ATOMIC_WRITTEN_PAGE(page) || IS_DUMMY_WRITTEN_PAGE(page))
80 f2fs_bug_on(F2FS_M_SB(page->mapping),
81 *((u32 *)page_private(page)) != F2FS_COMPRESSED_PAGE_MAGIC);
85 static void f2fs_set_compressed_page(struct page *page,
86 struct inode *inode, pgoff_t index, void *data)
89 set_page_private(page, (unsigned long)data);
91 /* i_crypto_info and iv index */
93 page->mapping = inode->i_mapping;
96 static void f2fs_drop_rpages(struct compress_ctx *cc, int len, bool unlock)
100 for (i = 0; i < len; i++) {
104 unlock_page(cc->rpages[i]);
106 put_page(cc->rpages[i]);
110 static void f2fs_put_rpages(struct compress_ctx *cc)
112 f2fs_drop_rpages(cc, cc->cluster_size, false);
115 static void f2fs_unlock_rpages(struct compress_ctx *cc, int len)
117 f2fs_drop_rpages(cc, len, true);
120 static void f2fs_put_rpages_wbc(struct compress_ctx *cc,
121 struct writeback_control *wbc, bool redirty, int unlock)
125 for (i = 0; i < cc->cluster_size; i++) {
129 redirty_page_for_writepage(wbc, cc->rpages[i]);
130 f2fs_put_page(cc->rpages[i], unlock);
134 struct page *f2fs_compress_control_page(struct page *page)
136 return ((struct compress_io_ctx *)page_private(page))->rpages[0];
139 int f2fs_init_compress_ctx(struct compress_ctx *cc)
144 cc->rpages = page_array_alloc(cc->inode, cc->cluster_size);
145 return cc->rpages ? 0 : -ENOMEM;
148 void f2fs_destroy_compress_ctx(struct compress_ctx *cc, bool reuse)
150 page_array_free(cc->inode, cc->rpages, cc->cluster_size);
155 cc->cluster_idx = NULL_CLUSTER;
158 void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct page *page)
160 unsigned int cluster_ofs;
162 if (!f2fs_cluster_can_merge_page(cc, page->index))
163 f2fs_bug_on(F2FS_I_SB(cc->inode), 1);
165 cluster_ofs = offset_in_cluster(cc, page->index);
166 cc->rpages[cluster_ofs] = page;
168 cc->cluster_idx = cluster_idx(cc, page->index);
171 #ifdef CONFIG_F2FS_FS_LZO
172 static int lzo_init_compress_ctx(struct compress_ctx *cc)
174 cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
175 LZO1X_MEM_COMPRESS, GFP_NOFS);
179 cc->clen = lzo1x_worst_compress(PAGE_SIZE << cc->log_cluster_size);
183 static void lzo_destroy_compress_ctx(struct compress_ctx *cc)
189 static int lzo_compress_pages(struct compress_ctx *cc)
193 ret = lzo1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
194 &cc->clen, cc->private);
195 if (ret != LZO_E_OK) {
196 printk_ratelimited("%sF2FS-fs (%s): lzo compress failed, ret:%d\n",
197 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
203 static int lzo_decompress_pages(struct decompress_io_ctx *dic)
207 ret = lzo1x_decompress_safe(dic->cbuf->cdata, dic->clen,
208 dic->rbuf, &dic->rlen);
209 if (ret != LZO_E_OK) {
210 printk_ratelimited("%sF2FS-fs (%s): lzo decompress failed, ret:%d\n",
211 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
215 if (dic->rlen != PAGE_SIZE << dic->log_cluster_size) {
216 printk_ratelimited("%sF2FS-fs (%s): lzo invalid rlen:%zu, "
217 "expected:%lu\n", KERN_ERR,
218 F2FS_I_SB(dic->inode)->sb->s_id,
220 PAGE_SIZE << dic->log_cluster_size);
226 static const struct f2fs_compress_ops f2fs_lzo_ops = {
227 .init_compress_ctx = lzo_init_compress_ctx,
228 .destroy_compress_ctx = lzo_destroy_compress_ctx,
229 .compress_pages = lzo_compress_pages,
230 .decompress_pages = lzo_decompress_pages,
234 #ifdef CONFIG_F2FS_FS_LZ4
235 static int lz4_init_compress_ctx(struct compress_ctx *cc)
237 unsigned int size = LZ4_MEM_COMPRESS;
239 #ifdef CONFIG_F2FS_FS_LZ4HC
240 if (F2FS_I(cc->inode)->i_compress_flag >> COMPRESS_LEVEL_OFFSET)
241 size = LZ4HC_MEM_COMPRESS;
244 cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), size, GFP_NOFS);
249 * we do not change cc->clen to LZ4_compressBound(inputsize) to
250 * adapt worst compress case, because lz4 compressor can handle
251 * output budget properly.
253 cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
257 static void lz4_destroy_compress_ctx(struct compress_ctx *cc)
263 #ifdef CONFIG_F2FS_FS_LZ4HC
264 static int lz4hc_compress_pages(struct compress_ctx *cc)
266 unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
267 COMPRESS_LEVEL_OFFSET;
271 len = LZ4_compress_HC(cc->rbuf, cc->cbuf->cdata, cc->rlen,
272 cc->clen, level, cc->private);
274 len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
275 cc->clen, cc->private);
284 static int lz4_compress_pages(struct compress_ctx *cc)
288 #ifdef CONFIG_F2FS_FS_LZ4HC
289 return lz4hc_compress_pages(cc);
291 len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
292 cc->clen, cc->private);
300 static int lz4_decompress_pages(struct decompress_io_ctx *dic)
304 ret = LZ4_decompress_safe(dic->cbuf->cdata, dic->rbuf,
305 dic->clen, dic->rlen);
307 printk_ratelimited("%sF2FS-fs (%s): lz4 decompress failed, ret:%d\n",
308 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
312 if (ret != PAGE_SIZE << dic->log_cluster_size) {
313 printk_ratelimited("%sF2FS-fs (%s): lz4 invalid rlen:%zu, "
314 "expected:%lu\n", KERN_ERR,
315 F2FS_I_SB(dic->inode)->sb->s_id,
317 PAGE_SIZE << dic->log_cluster_size);
323 static const struct f2fs_compress_ops f2fs_lz4_ops = {
324 .init_compress_ctx = lz4_init_compress_ctx,
325 .destroy_compress_ctx = lz4_destroy_compress_ctx,
326 .compress_pages = lz4_compress_pages,
327 .decompress_pages = lz4_decompress_pages,
331 #ifdef CONFIG_F2FS_FS_ZSTD
332 #define F2FS_ZSTD_DEFAULT_CLEVEL 1
334 static int zstd_init_compress_ctx(struct compress_ctx *cc)
336 ZSTD_parameters params;
337 ZSTD_CStream *stream;
339 unsigned int workspace_size;
340 unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
341 COMPRESS_LEVEL_OFFSET;
344 level = F2FS_ZSTD_DEFAULT_CLEVEL;
346 params = ZSTD_getParams(level, cc->rlen, 0);
347 workspace_size = ZSTD_CStreamWorkspaceBound(params.cParams);
349 workspace = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
350 workspace_size, GFP_NOFS);
354 stream = ZSTD_initCStream(params, 0, workspace, workspace_size);
356 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initCStream failed\n",
357 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
363 cc->private = workspace;
364 cc->private2 = stream;
366 cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
370 static void zstd_destroy_compress_ctx(struct compress_ctx *cc)
377 static int zstd_compress_pages(struct compress_ctx *cc)
379 ZSTD_CStream *stream = cc->private2;
381 ZSTD_outBuffer outbuf;
382 int src_size = cc->rlen;
383 int dst_size = src_size - PAGE_SIZE - COMPRESS_HEADER_SIZE;
387 inbuf.src = cc->rbuf;
388 inbuf.size = src_size;
391 outbuf.dst = cc->cbuf->cdata;
392 outbuf.size = dst_size;
394 ret = ZSTD_compressStream(stream, &outbuf, &inbuf);
395 if (ZSTD_isError(ret)) {
396 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
397 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
398 __func__, ZSTD_getErrorCode(ret));
402 ret = ZSTD_endStream(stream, &outbuf);
403 if (ZSTD_isError(ret)) {
404 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_endStream returned %d\n",
405 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
406 __func__, ZSTD_getErrorCode(ret));
411 * there is compressed data remained in intermediate buffer due to
412 * no more space in cbuf.cdata
417 cc->clen = outbuf.pos;
421 static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
423 ZSTD_DStream *stream;
425 unsigned int workspace_size;
426 unsigned int max_window_size =
427 MAX_COMPRESS_WINDOW_SIZE(dic->log_cluster_size);
429 workspace_size = ZSTD_DStreamWorkspaceBound(max_window_size);
431 workspace = f2fs_kvmalloc(F2FS_I_SB(dic->inode),
432 workspace_size, GFP_NOFS);
436 stream = ZSTD_initDStream(max_window_size, workspace, workspace_size);
438 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initDStream failed\n",
439 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
445 dic->private = workspace;
446 dic->private2 = stream;
451 static void zstd_destroy_decompress_ctx(struct decompress_io_ctx *dic)
453 kvfree(dic->private);
455 dic->private2 = NULL;
458 static int zstd_decompress_pages(struct decompress_io_ctx *dic)
460 ZSTD_DStream *stream = dic->private2;
462 ZSTD_outBuffer outbuf;
466 inbuf.src = dic->cbuf->cdata;
467 inbuf.size = dic->clen;
470 outbuf.dst = dic->rbuf;
471 outbuf.size = dic->rlen;
473 ret = ZSTD_decompressStream(stream, &outbuf, &inbuf);
474 if (ZSTD_isError(ret)) {
475 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
476 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
477 __func__, ZSTD_getErrorCode(ret));
481 if (dic->rlen != outbuf.pos) {
482 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD invalid rlen:%zu, "
483 "expected:%lu\n", KERN_ERR,
484 F2FS_I_SB(dic->inode)->sb->s_id,
486 PAGE_SIZE << dic->log_cluster_size);
493 static const struct f2fs_compress_ops f2fs_zstd_ops = {
494 .init_compress_ctx = zstd_init_compress_ctx,
495 .destroy_compress_ctx = zstd_destroy_compress_ctx,
496 .compress_pages = zstd_compress_pages,
497 .init_decompress_ctx = zstd_init_decompress_ctx,
498 .destroy_decompress_ctx = zstd_destroy_decompress_ctx,
499 .decompress_pages = zstd_decompress_pages,
503 #ifdef CONFIG_F2FS_FS_LZO
504 #ifdef CONFIG_F2FS_FS_LZORLE
505 static int lzorle_compress_pages(struct compress_ctx *cc)
509 ret = lzorle1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
510 &cc->clen, cc->private);
511 if (ret != LZO_E_OK) {
512 printk_ratelimited("%sF2FS-fs (%s): lzo-rle compress failed, ret:%d\n",
513 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
519 static const struct f2fs_compress_ops f2fs_lzorle_ops = {
520 .init_compress_ctx = lzo_init_compress_ctx,
521 .destroy_compress_ctx = lzo_destroy_compress_ctx,
522 .compress_pages = lzorle_compress_pages,
523 .decompress_pages = lzo_decompress_pages,
528 static const struct f2fs_compress_ops *f2fs_cops[COMPRESS_MAX] = {
529 #ifdef CONFIG_F2FS_FS_LZO
534 #ifdef CONFIG_F2FS_FS_LZ4
539 #ifdef CONFIG_F2FS_FS_ZSTD
544 #if defined(CONFIG_F2FS_FS_LZO) && defined(CONFIG_F2FS_FS_LZORLE)
551 bool f2fs_is_compress_backend_ready(struct inode *inode)
553 if (!f2fs_compressed_file(inode))
555 return f2fs_cops[F2FS_I(inode)->i_compress_algorithm];
558 static mempool_t *compress_page_pool;
559 static int num_compress_pages = 512;
560 module_param(num_compress_pages, uint, 0444);
561 MODULE_PARM_DESC(num_compress_pages,
562 "Number of intermediate compress pages to preallocate");
564 int f2fs_init_compress_mempool(void)
566 compress_page_pool = mempool_create_page_pool(num_compress_pages, 0);
567 if (!compress_page_pool)
573 void f2fs_destroy_compress_mempool(void)
575 mempool_destroy(compress_page_pool);
578 static struct page *f2fs_compress_alloc_page(void)
582 page = mempool_alloc(compress_page_pool, GFP_NOFS);
588 static void f2fs_compress_free_page(struct page *page)
592 set_page_private(page, (unsigned long)NULL);
593 ClearPagePrivate(page);
594 page->mapping = NULL;
596 mempool_free(page, compress_page_pool);
599 #define MAX_VMAP_RETRIES 3
601 static void *f2fs_vmap(struct page **pages, unsigned int count)
606 for (i = 0; i < MAX_VMAP_RETRIES; i++) {
607 buf = vm_map_ram(pages, count, -1);
615 static int f2fs_compress_pages(struct compress_ctx *cc)
617 struct f2fs_inode_info *fi = F2FS_I(cc->inode);
618 const struct f2fs_compress_ops *cops =
619 f2fs_cops[fi->i_compress_algorithm];
620 unsigned int max_len, new_nr_cpages;
621 struct page **new_cpages;
625 trace_f2fs_compress_pages_start(cc->inode, cc->cluster_idx,
626 cc->cluster_size, fi->i_compress_algorithm);
628 if (cops->init_compress_ctx) {
629 ret = cops->init_compress_ctx(cc);
634 max_len = COMPRESS_HEADER_SIZE + cc->clen;
635 cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE);
637 cc->cpages = page_array_alloc(cc->inode, cc->nr_cpages);
640 goto destroy_compress_ctx;
643 for (i = 0; i < cc->nr_cpages; i++) {
644 cc->cpages[i] = f2fs_compress_alloc_page();
645 if (!cc->cpages[i]) {
647 goto out_free_cpages;
651 cc->rbuf = f2fs_vmap(cc->rpages, cc->cluster_size);
654 goto out_free_cpages;
657 cc->cbuf = f2fs_vmap(cc->cpages, cc->nr_cpages);
660 goto out_vunmap_rbuf;
663 ret = cops->compress_pages(cc);
665 goto out_vunmap_cbuf;
667 max_len = PAGE_SIZE * (cc->cluster_size - 1) - COMPRESS_HEADER_SIZE;
669 if (cc->clen > max_len) {
671 goto out_vunmap_cbuf;
674 cc->cbuf->clen = cpu_to_le32(cc->clen);
676 if (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)
677 chksum = f2fs_crc32(F2FS_I_SB(cc->inode),
678 cc->cbuf->cdata, cc->clen);
679 cc->cbuf->chksum = cpu_to_le32(chksum);
681 for (i = 0; i < COMPRESS_DATA_RESERVED_SIZE; i++)
682 cc->cbuf->reserved[i] = cpu_to_le32(0);
684 new_nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE);
686 /* Now we're going to cut unnecessary tail pages */
687 new_cpages = page_array_alloc(cc->inode, new_nr_cpages);
690 goto out_vunmap_cbuf;
693 /* zero out any unused part of the last page */
694 memset(&cc->cbuf->cdata[cc->clen], 0,
695 (new_nr_cpages * PAGE_SIZE) -
696 (cc->clen + COMPRESS_HEADER_SIZE));
698 vm_unmap_ram(cc->cbuf, cc->nr_cpages);
699 vm_unmap_ram(cc->rbuf, cc->cluster_size);
701 for (i = 0; i < cc->nr_cpages; i++) {
702 if (i < new_nr_cpages) {
703 new_cpages[i] = cc->cpages[i];
706 f2fs_compress_free_page(cc->cpages[i]);
707 cc->cpages[i] = NULL;
710 if (cops->destroy_compress_ctx)
711 cops->destroy_compress_ctx(cc);
713 page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
714 cc->cpages = new_cpages;
715 cc->nr_cpages = new_nr_cpages;
717 trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
722 vm_unmap_ram(cc->cbuf, cc->nr_cpages);
724 vm_unmap_ram(cc->rbuf, cc->cluster_size);
726 for (i = 0; i < cc->nr_cpages; i++) {
728 f2fs_compress_free_page(cc->cpages[i]);
730 page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
732 destroy_compress_ctx:
733 if (cops->destroy_compress_ctx)
734 cops->destroy_compress_ctx(cc);
736 trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
741 static void f2fs_decompress_cluster(struct decompress_io_ctx *dic)
743 struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
744 struct f2fs_inode_info *fi = F2FS_I(dic->inode);
745 const struct f2fs_compress_ops *cops =
746 f2fs_cops[fi->i_compress_algorithm];
750 trace_f2fs_decompress_pages_start(dic->inode, dic->cluster_idx,
751 dic->cluster_size, fi->i_compress_algorithm);
758 dic->tpages = page_array_alloc(dic->inode, dic->cluster_size);
764 for (i = 0; i < dic->cluster_size; i++) {
765 if (dic->rpages[i]) {
766 dic->tpages[i] = dic->rpages[i];
770 dic->tpages[i] = f2fs_compress_alloc_page();
771 if (!dic->tpages[i]) {
777 if (cops->init_decompress_ctx) {
778 ret = cops->init_decompress_ctx(dic);
783 dic->rbuf = f2fs_vmap(dic->tpages, dic->cluster_size);
786 goto out_destroy_decompress_ctx;
789 dic->cbuf = f2fs_vmap(dic->cpages, dic->nr_cpages);
792 goto out_vunmap_rbuf;
795 dic->clen = le32_to_cpu(dic->cbuf->clen);
796 dic->rlen = PAGE_SIZE << dic->log_cluster_size;
798 if (dic->clen > PAGE_SIZE * dic->nr_cpages - COMPRESS_HEADER_SIZE) {
800 goto out_vunmap_cbuf;
803 ret = cops->decompress_pages(dic);
805 if (!ret && (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)) {
806 u32 provided = le32_to_cpu(dic->cbuf->chksum);
807 u32 calculated = f2fs_crc32(sbi, dic->cbuf->cdata, dic->clen);
809 if (provided != calculated) {
810 if (!is_inode_flag_set(dic->inode, FI_COMPRESS_CORRUPT)) {
811 set_inode_flag(dic->inode, FI_COMPRESS_CORRUPT);
813 "%sF2FS-fs (%s): checksum invalid, nid = %lu, %x vs %x",
814 KERN_INFO, sbi->sb->s_id, dic->inode->i_ino,
815 provided, calculated);
817 set_sbi_flag(sbi, SBI_NEED_FSCK);
822 vm_unmap_ram(dic->cbuf, dic->nr_cpages);
824 vm_unmap_ram(dic->rbuf, dic->cluster_size);
825 out_destroy_decompress_ctx:
826 if (cops->destroy_decompress_ctx)
827 cops->destroy_decompress_ctx(dic);
829 trace_f2fs_decompress_pages_end(dic->inode, dic->cluster_idx,
831 f2fs_decompress_end_io(dic, ret);
835 * This is called when a page of a compressed cluster has been read from disk
836 * (or failed to be read from disk). It checks whether this page was the last
837 * page being waited on in the cluster, and if so, it decompresses the cluster
838 * (or in the case of a failure, cleans up without actually decompressing).
840 void f2fs_end_read_compressed_page(struct page *page, bool failed)
842 struct decompress_io_ctx *dic =
843 (struct decompress_io_ctx *)page_private(page);
844 struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
846 dec_page_count(sbi, F2FS_RD_DATA);
849 WRITE_ONCE(dic->failed, true);
851 if (atomic_dec_and_test(&dic->remaining_pages))
852 f2fs_decompress_cluster(dic);
855 static bool is_page_in_cluster(struct compress_ctx *cc, pgoff_t index)
857 if (cc->cluster_idx == NULL_CLUSTER)
859 return cc->cluster_idx == cluster_idx(cc, index);
862 bool f2fs_cluster_is_empty(struct compress_ctx *cc)
864 return cc->nr_rpages == 0;
867 static bool f2fs_cluster_is_full(struct compress_ctx *cc)
869 return cc->cluster_size == cc->nr_rpages;
872 bool f2fs_cluster_can_merge_page(struct compress_ctx *cc, pgoff_t index)
874 if (f2fs_cluster_is_empty(cc))
876 return is_page_in_cluster(cc, index);
879 static bool __cluster_may_compress(struct compress_ctx *cc)
881 loff_t i_size = i_size_read(cc->inode);
882 unsigned nr_pages = DIV_ROUND_UP(i_size, PAGE_SIZE);
885 for (i = 0; i < cc->cluster_size; i++) {
886 struct page *page = cc->rpages[i];
888 f2fs_bug_on(F2FS_I_SB(cc->inode), !page);
891 if (page->index >= nr_pages)
897 static int __f2fs_cluster_blocks(struct compress_ctx *cc, bool compr)
899 struct dnode_of_data dn;
902 set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
903 ret = f2fs_get_dnode_of_data(&dn, start_idx_of_cluster(cc),
911 if (dn.data_blkaddr == COMPRESS_ADDR) {
915 for (i = 1; i < cc->cluster_size; i++) {
918 blkaddr = data_blkaddr(dn.inode,
919 dn.node_page, dn.ofs_in_node + i);
921 if (__is_valid_data_blkaddr(blkaddr))
924 if (blkaddr != NULL_ADDR)
934 /* return # of compressed blocks in compressed cluster */
935 static int f2fs_compressed_blocks(struct compress_ctx *cc)
937 return __f2fs_cluster_blocks(cc, true);
940 /* return # of valid blocks in compressed cluster */
941 static int f2fs_cluster_blocks(struct compress_ctx *cc)
943 return __f2fs_cluster_blocks(cc, false);
946 int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index)
948 struct compress_ctx cc = {
950 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
951 .cluster_size = F2FS_I(inode)->i_cluster_size,
952 .cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
955 return f2fs_cluster_blocks(&cc);
958 static bool cluster_may_compress(struct compress_ctx *cc)
960 if (!f2fs_need_compress_data(cc->inode))
962 if (f2fs_is_atomic_file(cc->inode))
964 if (f2fs_is_mmap_file(cc->inode))
966 if (!f2fs_cluster_is_full(cc))
968 if (unlikely(f2fs_cp_error(F2FS_I_SB(cc->inode))))
970 return __cluster_may_compress(cc);
973 static void set_cluster_writeback(struct compress_ctx *cc)
977 for (i = 0; i < cc->cluster_size; i++) {
979 set_page_writeback(cc->rpages[i]);
983 static void set_cluster_dirty(struct compress_ctx *cc)
987 for (i = 0; i < cc->cluster_size; i++)
989 set_page_dirty(cc->rpages[i]);
992 static int prepare_compress_overwrite(struct compress_ctx *cc,
993 struct page **pagep, pgoff_t index, void **fsdata)
995 struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
996 struct address_space *mapping = cc->inode->i_mapping;
998 struct dnode_of_data dn;
999 sector_t last_block_in_bio;
1000 unsigned fgp_flag = FGP_LOCK | FGP_WRITE | FGP_CREAT;
1001 pgoff_t start_idx = start_idx_of_cluster(cc);
1006 ret = f2fs_cluster_blocks(cc);
1010 /* compressed case */
1011 prealloc = (ret < cc->cluster_size);
1013 ret = f2fs_init_compress_ctx(cc);
1017 /* keep page reference to avoid page reclaim */
1018 for (i = 0; i < cc->cluster_size; i++) {
1019 page = f2fs_pagecache_get_page(mapping, start_idx + i,
1020 fgp_flag, GFP_NOFS);
1026 if (PageUptodate(page))
1027 f2fs_put_page(page, 1);
1029 f2fs_compress_ctx_add_page(cc, page);
1032 if (!f2fs_cluster_is_empty(cc)) {
1033 struct bio *bio = NULL;
1035 ret = f2fs_read_multi_pages(cc, &bio, cc->cluster_size,
1036 &last_block_in_bio, false, true);
1037 f2fs_put_rpages(cc);
1038 f2fs_destroy_compress_ctx(cc, true);
1042 f2fs_submit_bio(sbi, bio, DATA);
1044 ret = f2fs_init_compress_ctx(cc);
1049 for (i = 0; i < cc->cluster_size; i++) {
1050 f2fs_bug_on(sbi, cc->rpages[i]);
1052 page = find_lock_page(mapping, start_idx + i);
1054 /* page can be truncated */
1055 goto release_and_retry;
1058 f2fs_wait_on_page_writeback(page, DATA, true, true);
1059 f2fs_compress_ctx_add_page(cc, page);
1061 if (!PageUptodate(page)) {
1063 f2fs_put_rpages(cc);
1064 f2fs_unlock_rpages(cc, i + 1);
1065 f2fs_destroy_compress_ctx(cc, true);
1071 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
1073 set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1075 for (i = cc->cluster_size - 1; i > 0; i--) {
1076 ret = f2fs_get_block(&dn, start_idx + i);
1078 i = cc->cluster_size;
1082 if (dn.data_blkaddr != NEW_ADDR)
1086 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
1090 *fsdata = cc->rpages;
1091 *pagep = cc->rpages[offset_in_cluster(cc, index)];
1092 return cc->cluster_size;
1096 f2fs_put_rpages(cc);
1097 f2fs_unlock_rpages(cc, i);
1098 f2fs_destroy_compress_ctx(cc, true);
1103 int f2fs_prepare_compress_overwrite(struct inode *inode,
1104 struct page **pagep, pgoff_t index, void **fsdata)
1106 struct compress_ctx cc = {
1108 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1109 .cluster_size = F2FS_I(inode)->i_cluster_size,
1110 .cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
1115 return prepare_compress_overwrite(&cc, pagep, index, fsdata);
1118 bool f2fs_compress_write_end(struct inode *inode, void *fsdata,
1119 pgoff_t index, unsigned copied)
1122 struct compress_ctx cc = {
1124 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1125 .cluster_size = F2FS_I(inode)->i_cluster_size,
1128 bool first_index = (index == cc.rpages[0]->index);
1131 set_cluster_dirty(&cc);
1133 f2fs_put_rpages_wbc(&cc, NULL, false, 1);
1134 f2fs_destroy_compress_ctx(&cc, false);
1139 int f2fs_truncate_partial_cluster(struct inode *inode, u64 from, bool lock)
1141 void *fsdata = NULL;
1143 int log_cluster_size = F2FS_I(inode)->i_log_cluster_size;
1144 pgoff_t start_idx = from >> (PAGE_SHIFT + log_cluster_size) <<
1148 err = f2fs_is_compressed_cluster(inode, start_idx);
1152 /* truncate normal cluster */
1154 return f2fs_do_truncate_blocks(inode, from, lock);
1156 /* truncate compressed cluster */
1157 err = f2fs_prepare_compress_overwrite(inode, &pagep,
1158 start_idx, &fsdata);
1160 /* should not be a normal cluster */
1161 f2fs_bug_on(F2FS_I_SB(inode), err == 0);
1167 struct page **rpages = fsdata;
1168 int cluster_size = F2FS_I(inode)->i_cluster_size;
1171 for (i = cluster_size - 1; i >= 0; i--) {
1172 loff_t start = rpages[i]->index << PAGE_SHIFT;
1174 if (from <= start) {
1175 zero_user_segment(rpages[i], 0, PAGE_SIZE);
1177 zero_user_segment(rpages[i], from - start,
1183 f2fs_compress_write_end(inode, fsdata, start_idx, true);
1188 static int f2fs_write_compressed_pages(struct compress_ctx *cc,
1190 struct writeback_control *wbc,
1191 enum iostat_type io_type)
1193 struct inode *inode = cc->inode;
1194 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1195 struct f2fs_inode_info *fi = F2FS_I(inode);
1196 struct f2fs_io_info fio = {
1198 .ino = cc->inode->i_ino,
1201 .op_flags = wbc_to_write_flags(wbc),
1202 .old_blkaddr = NEW_ADDR,
1204 .encrypted_page = NULL,
1205 .compressed_page = NULL,
1209 .encrypted = fscrypt_inode_uses_fs_layer_crypto(cc->inode),
1211 struct dnode_of_data dn;
1212 struct node_info ni;
1213 struct compress_io_ctx *cic;
1214 pgoff_t start_idx = start_idx_of_cluster(cc);
1215 unsigned int last_index = cc->cluster_size - 1;
1219 if (IS_NOQUOTA(inode)) {
1221 * We need to wait for node_write to avoid block allocation during
1222 * checkpoint. This can only happen to quota writes which can cause
1223 * the below discard race condition.
1225 down_read(&sbi->node_write);
1226 } else if (!f2fs_trylock_op(sbi)) {
1230 set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1232 err = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
1236 for (i = 0; i < cc->cluster_size; i++) {
1237 if (data_blkaddr(dn.inode, dn.node_page,
1238 dn.ofs_in_node + i) == NULL_ADDR)
1242 psize = (loff_t)(cc->rpages[last_index]->index + 1) << PAGE_SHIFT;
1244 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1248 fio.version = ni.version;
1250 cic = kmem_cache_zalloc(cic_entry_slab, GFP_NOFS);
1254 cic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1256 atomic_set(&cic->pending_pages, cc->nr_cpages);
1257 cic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1261 cic->nr_rpages = cc->cluster_size;
1263 for (i = 0; i < cc->nr_cpages; i++) {
1264 f2fs_set_compressed_page(cc->cpages[i], inode,
1265 cc->rpages[i + 1]->index, cic);
1266 fio.compressed_page = cc->cpages[i];
1268 fio.old_blkaddr = data_blkaddr(dn.inode, dn.node_page,
1269 dn.ofs_in_node + i + 1);
1271 /* wait for GCed page writeback via META_MAPPING */
1272 f2fs_wait_on_block_writeback(inode, fio.old_blkaddr);
1274 if (fio.encrypted) {
1275 fio.page = cc->rpages[i + 1];
1276 err = f2fs_encrypt_one_page(&fio);
1278 goto out_destroy_crypt;
1279 cc->cpages[i] = fio.encrypted_page;
1283 set_cluster_writeback(cc);
1285 for (i = 0; i < cc->cluster_size; i++)
1286 cic->rpages[i] = cc->rpages[i];
1288 for (i = 0; i < cc->cluster_size; i++, dn.ofs_in_node++) {
1291 blkaddr = f2fs_data_blkaddr(&dn);
1292 fio.page = cc->rpages[i];
1293 fio.old_blkaddr = blkaddr;
1295 /* cluster header */
1297 if (blkaddr == COMPRESS_ADDR)
1299 if (__is_valid_data_blkaddr(blkaddr))
1300 f2fs_invalidate_blocks(sbi, blkaddr);
1301 f2fs_update_data_blkaddr(&dn, COMPRESS_ADDR);
1302 goto unlock_continue;
1305 if (fio.compr_blocks && __is_valid_data_blkaddr(blkaddr))
1308 if (i > cc->nr_cpages) {
1309 if (__is_valid_data_blkaddr(blkaddr)) {
1310 f2fs_invalidate_blocks(sbi, blkaddr);
1311 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
1313 goto unlock_continue;
1316 f2fs_bug_on(fio.sbi, blkaddr == NULL_ADDR);
1319 fio.encrypted_page = cc->cpages[i - 1];
1321 fio.compressed_page = cc->cpages[i - 1];
1323 cc->cpages[i - 1] = NULL;
1324 f2fs_outplace_write_data(&dn, &fio);
1327 inode_dec_dirty_pages(cc->inode);
1328 unlock_page(fio.page);
1331 if (fio.compr_blocks)
1332 f2fs_i_compr_blocks_update(inode, fio.compr_blocks - 1, false);
1333 f2fs_i_compr_blocks_update(inode, cc->nr_cpages, true);
1334 add_compr_block_stat(inode, cc->nr_cpages);
1336 set_inode_flag(cc->inode, FI_APPEND_WRITE);
1337 if (cc->cluster_idx == 0)
1338 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1340 f2fs_put_dnode(&dn);
1341 if (IS_NOQUOTA(inode))
1342 up_read(&sbi->node_write);
1344 f2fs_unlock_op(sbi);
1346 spin_lock(&fi->i_size_lock);
1347 if (fi->last_disk_size < psize)
1348 fi->last_disk_size = psize;
1349 spin_unlock(&fi->i_size_lock);
1351 f2fs_put_rpages(cc);
1352 page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1354 f2fs_destroy_compress_ctx(cc, false);
1358 page_array_free(cc->inode, cic->rpages, cc->cluster_size);
1360 for (--i; i >= 0; i--)
1361 fscrypt_finalize_bounce_page(&cc->cpages[i]);
1362 for (i = 0; i < cc->nr_cpages; i++) {
1365 f2fs_compress_free_page(cc->cpages[i]);
1366 cc->cpages[i] = NULL;
1369 kmem_cache_free(cic_entry_slab, cic);
1371 f2fs_put_dnode(&dn);
1373 if (IS_NOQUOTA(inode))
1374 up_read(&sbi->node_write);
1376 f2fs_unlock_op(sbi);
1378 page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1383 void f2fs_compress_write_end_io(struct bio *bio, struct page *page)
1385 struct f2fs_sb_info *sbi = bio->bi_private;
1386 struct compress_io_ctx *cic =
1387 (struct compress_io_ctx *)page_private(page);
1390 if (unlikely(bio->bi_status))
1391 mapping_set_error(cic->inode->i_mapping, -EIO);
1393 f2fs_compress_free_page(page);
1395 dec_page_count(sbi, F2FS_WB_DATA);
1397 if (atomic_dec_return(&cic->pending_pages))
1400 for (i = 0; i < cic->nr_rpages; i++) {
1401 WARN_ON(!cic->rpages[i]);
1402 clear_cold_data(cic->rpages[i]);
1403 end_page_writeback(cic->rpages[i]);
1406 page_array_free(cic->inode, cic->rpages, cic->nr_rpages);
1407 kmem_cache_free(cic_entry_slab, cic);
1410 static int f2fs_write_raw_pages(struct compress_ctx *cc,
1412 struct writeback_control *wbc,
1413 enum iostat_type io_type)
1415 struct address_space *mapping = cc->inode->i_mapping;
1416 int _submitted, compr_blocks, ret;
1417 int i = -1, err = 0;
1419 compr_blocks = f2fs_compressed_blocks(cc);
1420 if (compr_blocks < 0) {
1425 for (i = 0; i < cc->cluster_size; i++) {
1429 if (cc->rpages[i]->mapping != mapping) {
1430 unlock_page(cc->rpages[i]);
1434 BUG_ON(!PageLocked(cc->rpages[i]));
1436 ret = f2fs_write_single_data_page(cc->rpages[i], &_submitted,
1437 NULL, NULL, wbc, io_type,
1438 compr_blocks, false);
1440 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1441 unlock_page(cc->rpages[i]);
1443 } else if (ret == -EAGAIN) {
1445 * for quota file, just redirty left pages to
1446 * avoid deadlock caused by cluster update race
1447 * from foreground operation.
1449 if (IS_NOQUOTA(cc->inode)) {
1455 congestion_wait(BLK_RW_ASYNC,
1456 DEFAULT_IO_TIMEOUT);
1457 lock_page(cc->rpages[i]);
1459 if (!PageDirty(cc->rpages[i])) {
1460 unlock_page(cc->rpages[i]);
1464 clear_page_dirty_for_io(cc->rpages[i]);
1471 *submitted += _submitted;
1474 f2fs_balance_fs(F2FS_M_SB(mapping), true);
1478 for (++i; i < cc->cluster_size; i++) {
1481 redirty_page_for_writepage(wbc, cc->rpages[i]);
1482 unlock_page(cc->rpages[i]);
1487 int f2fs_write_multi_pages(struct compress_ctx *cc,
1489 struct writeback_control *wbc,
1490 enum iostat_type io_type)
1495 if (cluster_may_compress(cc)) {
1496 err = f2fs_compress_pages(cc);
1497 if (err == -EAGAIN) {
1500 f2fs_put_rpages_wbc(cc, wbc, true, 1);
1504 err = f2fs_write_compressed_pages(cc, submitted,
1508 f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN);
1511 f2fs_bug_on(F2FS_I_SB(cc->inode), *submitted);
1513 err = f2fs_write_raw_pages(cc, submitted, wbc, io_type);
1514 f2fs_put_rpages_wbc(cc, wbc, false, 0);
1516 f2fs_destroy_compress_ctx(cc, false);
1520 static void f2fs_free_dic(struct decompress_io_ctx *dic);
1522 struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
1524 struct decompress_io_ctx *dic;
1525 pgoff_t start_idx = start_idx_of_cluster(cc);
1528 dic = kmem_cache_zalloc(dic_entry_slab, GFP_NOFS);
1530 return ERR_PTR(-ENOMEM);
1532 dic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1534 kmem_cache_free(dic_entry_slab, dic);
1535 return ERR_PTR(-ENOMEM);
1538 dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1539 dic->inode = cc->inode;
1540 atomic_set(&dic->remaining_pages, cc->nr_cpages);
1541 dic->cluster_idx = cc->cluster_idx;
1542 dic->cluster_size = cc->cluster_size;
1543 dic->log_cluster_size = cc->log_cluster_size;
1544 dic->nr_cpages = cc->nr_cpages;
1545 refcount_set(&dic->refcnt, 1);
1546 dic->failed = false;
1547 dic->need_verity = f2fs_need_verity(cc->inode, start_idx);
1549 for (i = 0; i < dic->cluster_size; i++)
1550 dic->rpages[i] = cc->rpages[i];
1551 dic->nr_rpages = cc->cluster_size;
1553 dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages);
1557 for (i = 0; i < dic->nr_cpages; i++) {
1560 page = f2fs_compress_alloc_page();
1564 f2fs_set_compressed_page(page, cc->inode,
1565 start_idx + i + 1, dic);
1566 dic->cpages[i] = page;
1573 return ERR_PTR(-ENOMEM);
1576 static void f2fs_free_dic(struct decompress_io_ctx *dic)
1581 for (i = 0; i < dic->cluster_size; i++) {
1584 if (!dic->tpages[i])
1586 f2fs_compress_free_page(dic->tpages[i]);
1588 page_array_free(dic->inode, dic->tpages, dic->cluster_size);
1592 for (i = 0; i < dic->nr_cpages; i++) {
1593 if (!dic->cpages[i])
1595 f2fs_compress_free_page(dic->cpages[i]);
1597 page_array_free(dic->inode, dic->cpages, dic->nr_cpages);
1600 page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
1601 kmem_cache_free(dic_entry_slab, dic);
1604 static void f2fs_put_dic(struct decompress_io_ctx *dic)
1606 if (refcount_dec_and_test(&dic->refcnt))
1611 * Update and unlock the cluster's pagecache pages, and release the reference to
1612 * the decompress_io_ctx that was being held for I/O completion.
1614 static void __f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1618 for (i = 0; i < dic->cluster_size; i++) {
1619 struct page *rpage = dic->rpages[i];
1624 /* PG_error was set if verity failed. */
1625 if (failed || PageError(rpage)) {
1626 ClearPageUptodate(rpage);
1627 /* will re-read again later */
1628 ClearPageError(rpage);
1630 SetPageUptodate(rpage);
1638 static void f2fs_verify_cluster(struct work_struct *work)
1640 struct decompress_io_ctx *dic =
1641 container_of(work, struct decompress_io_ctx, verity_work);
1644 /* Verify the cluster's decompressed pages with fs-verity. */
1645 for (i = 0; i < dic->cluster_size; i++) {
1646 struct page *rpage = dic->rpages[i];
1648 if (rpage && !fsverity_verify_page(rpage))
1649 SetPageError(rpage);
1652 __f2fs_decompress_end_io(dic, false);
1656 * This is called when a compressed cluster has been decompressed
1657 * (or failed to be read and/or decompressed).
1659 void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1661 if (!failed && dic->need_verity) {
1663 * Note that to avoid deadlocks, the verity work can't be done
1664 * on the decompression workqueue. This is because verifying
1665 * the data pages can involve reading metadata pages from the
1666 * file, and these metadata pages may be compressed.
1668 INIT_WORK(&dic->verity_work, f2fs_verify_cluster);
1669 fsverity_enqueue_verify_work(&dic->verity_work);
1671 __f2fs_decompress_end_io(dic, failed);
1676 * Put a reference to a compressed page's decompress_io_ctx.
1678 * This is called when the page is no longer needed and can be freed.
1680 void f2fs_put_page_dic(struct page *page)
1682 struct decompress_io_ctx *dic =
1683 (struct decompress_io_ctx *)page_private(page);
1688 int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi)
1690 dev_t dev = sbi->sb->s_bdev->bd_dev;
1693 sprintf(slab_name, "f2fs_page_array_entry-%u:%u", MAJOR(dev), MINOR(dev));
1695 sbi->page_array_slab_size = sizeof(struct page *) <<
1696 F2FS_OPTION(sbi).compress_log_size;
1698 sbi->page_array_slab = f2fs_kmem_cache_create(slab_name,
1699 sbi->page_array_slab_size);
1700 if (!sbi->page_array_slab)
1705 void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi)
1707 kmem_cache_destroy(sbi->page_array_slab);
1710 static int __init f2fs_init_cic_cache(void)
1712 cic_entry_slab = f2fs_kmem_cache_create("f2fs_cic_entry",
1713 sizeof(struct compress_io_ctx));
1714 if (!cic_entry_slab)
1719 static void f2fs_destroy_cic_cache(void)
1721 kmem_cache_destroy(cic_entry_slab);
1724 static int __init f2fs_init_dic_cache(void)
1726 dic_entry_slab = f2fs_kmem_cache_create("f2fs_dic_entry",
1727 sizeof(struct decompress_io_ctx));
1728 if (!dic_entry_slab)
1733 static void f2fs_destroy_dic_cache(void)
1735 kmem_cache_destroy(dic_entry_slab);
1738 int __init f2fs_init_compress_cache(void)
1742 err = f2fs_init_cic_cache();
1745 err = f2fs_init_dic_cache();
1750 f2fs_destroy_cic_cache();
1755 void f2fs_destroy_compress_cache(void)
1757 f2fs_destroy_dic_cache();
1758 f2fs_destroy_cic_cache();