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 (page_private_nonpointer(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)
88 attach_page_private(page, (void *)data);
90 /* i_crypto_info and iv index */
92 page->mapping = inode->i_mapping;
95 static void f2fs_drop_rpages(struct compress_ctx *cc, int len, bool unlock)
99 for (i = 0; i < len; i++) {
103 unlock_page(cc->rpages[i]);
105 put_page(cc->rpages[i]);
109 static void f2fs_put_rpages(struct compress_ctx *cc)
111 f2fs_drop_rpages(cc, cc->cluster_size, false);
114 static void f2fs_unlock_rpages(struct compress_ctx *cc, int len)
116 f2fs_drop_rpages(cc, len, true);
119 static void f2fs_put_rpages_wbc(struct compress_ctx *cc,
120 struct writeback_control *wbc, bool redirty, int unlock)
124 for (i = 0; i < cc->cluster_size; i++) {
128 redirty_page_for_writepage(wbc, cc->rpages[i]);
129 f2fs_put_page(cc->rpages[i], unlock);
133 struct page *f2fs_compress_control_page(struct page *page)
135 return ((struct compress_io_ctx *)page_private(page))->rpages[0];
138 int f2fs_init_compress_ctx(struct compress_ctx *cc)
143 cc->rpages = page_array_alloc(cc->inode, cc->cluster_size);
144 return cc->rpages ? 0 : -ENOMEM;
147 void f2fs_destroy_compress_ctx(struct compress_ctx *cc, bool reuse)
149 page_array_free(cc->inode, cc->rpages, cc->cluster_size);
154 cc->cluster_idx = NULL_CLUSTER;
157 void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct page *page)
159 unsigned int cluster_ofs;
161 if (!f2fs_cluster_can_merge_page(cc, page->index))
162 f2fs_bug_on(F2FS_I_SB(cc->inode), 1);
164 cluster_ofs = offset_in_cluster(cc, page->index);
165 cc->rpages[cluster_ofs] = page;
167 cc->cluster_idx = cluster_idx(cc, page->index);
170 #ifdef CONFIG_F2FS_FS_LZO
171 static int lzo_init_compress_ctx(struct compress_ctx *cc)
173 cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
174 LZO1X_MEM_COMPRESS, GFP_NOFS);
178 cc->clen = lzo1x_worst_compress(PAGE_SIZE << cc->log_cluster_size);
182 static void lzo_destroy_compress_ctx(struct compress_ctx *cc)
188 static int lzo_compress_pages(struct compress_ctx *cc)
192 ret = lzo1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
193 &cc->clen, cc->private);
194 if (ret != LZO_E_OK) {
195 printk_ratelimited("%sF2FS-fs (%s): lzo compress failed, ret:%d\n",
196 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
202 static int lzo_decompress_pages(struct decompress_io_ctx *dic)
206 ret = lzo1x_decompress_safe(dic->cbuf->cdata, dic->clen,
207 dic->rbuf, &dic->rlen);
208 if (ret != LZO_E_OK) {
209 printk_ratelimited("%sF2FS-fs (%s): lzo decompress failed, ret:%d\n",
210 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
214 if (dic->rlen != PAGE_SIZE << dic->log_cluster_size) {
215 printk_ratelimited("%sF2FS-fs (%s): lzo invalid rlen:%zu, "
216 "expected:%lu\n", KERN_ERR,
217 F2FS_I_SB(dic->inode)->sb->s_id,
219 PAGE_SIZE << dic->log_cluster_size);
225 static const struct f2fs_compress_ops f2fs_lzo_ops = {
226 .init_compress_ctx = lzo_init_compress_ctx,
227 .destroy_compress_ctx = lzo_destroy_compress_ctx,
228 .compress_pages = lzo_compress_pages,
229 .decompress_pages = lzo_decompress_pages,
233 #ifdef CONFIG_F2FS_FS_LZ4
234 static int lz4_init_compress_ctx(struct compress_ctx *cc)
236 unsigned int size = LZ4_MEM_COMPRESS;
238 #ifdef CONFIG_F2FS_FS_LZ4HC
239 if (F2FS_I(cc->inode)->i_compress_flag >> COMPRESS_LEVEL_OFFSET)
240 size = LZ4HC_MEM_COMPRESS;
243 cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), size, GFP_NOFS);
248 * we do not change cc->clen to LZ4_compressBound(inputsize) to
249 * adapt worst compress case, because lz4 compressor can handle
250 * output budget properly.
252 cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
256 static void lz4_destroy_compress_ctx(struct compress_ctx *cc)
262 #ifdef CONFIG_F2FS_FS_LZ4HC
263 static int lz4hc_compress_pages(struct compress_ctx *cc)
265 unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
266 COMPRESS_LEVEL_OFFSET;
270 len = LZ4_compress_HC(cc->rbuf, cc->cbuf->cdata, cc->rlen,
271 cc->clen, level, cc->private);
273 len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
274 cc->clen, cc->private);
283 static int lz4_compress_pages(struct compress_ctx *cc)
287 #ifdef CONFIG_F2FS_FS_LZ4HC
288 return lz4hc_compress_pages(cc);
290 len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
291 cc->clen, cc->private);
299 static int lz4_decompress_pages(struct decompress_io_ctx *dic)
303 ret = LZ4_decompress_safe(dic->cbuf->cdata, dic->rbuf,
304 dic->clen, dic->rlen);
306 printk_ratelimited("%sF2FS-fs (%s): lz4 decompress failed, ret:%d\n",
307 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
311 if (ret != PAGE_SIZE << dic->log_cluster_size) {
312 printk_ratelimited("%sF2FS-fs (%s): lz4 invalid rlen:%zu, "
313 "expected:%lu\n", KERN_ERR,
314 F2FS_I_SB(dic->inode)->sb->s_id,
316 PAGE_SIZE << dic->log_cluster_size);
322 static const struct f2fs_compress_ops f2fs_lz4_ops = {
323 .init_compress_ctx = lz4_init_compress_ctx,
324 .destroy_compress_ctx = lz4_destroy_compress_ctx,
325 .compress_pages = lz4_compress_pages,
326 .decompress_pages = lz4_decompress_pages,
330 #ifdef CONFIG_F2FS_FS_ZSTD
331 #define F2FS_ZSTD_DEFAULT_CLEVEL 1
333 static int zstd_init_compress_ctx(struct compress_ctx *cc)
335 ZSTD_parameters params;
336 ZSTD_CStream *stream;
338 unsigned int workspace_size;
339 unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
340 COMPRESS_LEVEL_OFFSET;
343 level = F2FS_ZSTD_DEFAULT_CLEVEL;
345 params = ZSTD_getParams(level, cc->rlen, 0);
346 workspace_size = ZSTD_CStreamWorkspaceBound(params.cParams);
348 workspace = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
349 workspace_size, GFP_NOFS);
353 stream = ZSTD_initCStream(params, 0, workspace, workspace_size);
355 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initCStream failed\n",
356 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
362 cc->private = workspace;
363 cc->private2 = stream;
365 cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
369 static void zstd_destroy_compress_ctx(struct compress_ctx *cc)
376 static int zstd_compress_pages(struct compress_ctx *cc)
378 ZSTD_CStream *stream = cc->private2;
380 ZSTD_outBuffer outbuf;
381 int src_size = cc->rlen;
382 int dst_size = src_size - PAGE_SIZE - COMPRESS_HEADER_SIZE;
386 inbuf.src = cc->rbuf;
387 inbuf.size = src_size;
390 outbuf.dst = cc->cbuf->cdata;
391 outbuf.size = dst_size;
393 ret = ZSTD_compressStream(stream, &outbuf, &inbuf);
394 if (ZSTD_isError(ret)) {
395 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
396 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
397 __func__, ZSTD_getErrorCode(ret));
401 ret = ZSTD_endStream(stream, &outbuf);
402 if (ZSTD_isError(ret)) {
403 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_endStream returned %d\n",
404 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
405 __func__, ZSTD_getErrorCode(ret));
410 * there is compressed data remained in intermediate buffer due to
411 * no more space in cbuf.cdata
416 cc->clen = outbuf.pos;
420 static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
422 ZSTD_DStream *stream;
424 unsigned int workspace_size;
425 unsigned int max_window_size =
426 MAX_COMPRESS_WINDOW_SIZE(dic->log_cluster_size);
428 workspace_size = ZSTD_DStreamWorkspaceBound(max_window_size);
430 workspace = f2fs_kvmalloc(F2FS_I_SB(dic->inode),
431 workspace_size, GFP_NOFS);
435 stream = ZSTD_initDStream(max_window_size, workspace, workspace_size);
437 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initDStream failed\n",
438 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
444 dic->private = workspace;
445 dic->private2 = stream;
450 static void zstd_destroy_decompress_ctx(struct decompress_io_ctx *dic)
452 kvfree(dic->private);
454 dic->private2 = NULL;
457 static int zstd_decompress_pages(struct decompress_io_ctx *dic)
459 ZSTD_DStream *stream = dic->private2;
461 ZSTD_outBuffer outbuf;
465 inbuf.src = dic->cbuf->cdata;
466 inbuf.size = dic->clen;
469 outbuf.dst = dic->rbuf;
470 outbuf.size = dic->rlen;
472 ret = ZSTD_decompressStream(stream, &outbuf, &inbuf);
473 if (ZSTD_isError(ret)) {
474 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
475 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
476 __func__, ZSTD_getErrorCode(ret));
480 if (dic->rlen != outbuf.pos) {
481 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD invalid rlen:%zu, "
482 "expected:%lu\n", KERN_ERR,
483 F2FS_I_SB(dic->inode)->sb->s_id,
485 PAGE_SIZE << dic->log_cluster_size);
492 static const struct f2fs_compress_ops f2fs_zstd_ops = {
493 .init_compress_ctx = zstd_init_compress_ctx,
494 .destroy_compress_ctx = zstd_destroy_compress_ctx,
495 .compress_pages = zstd_compress_pages,
496 .init_decompress_ctx = zstd_init_decompress_ctx,
497 .destroy_decompress_ctx = zstd_destroy_decompress_ctx,
498 .decompress_pages = zstd_decompress_pages,
502 #ifdef CONFIG_F2FS_FS_LZO
503 #ifdef CONFIG_F2FS_FS_LZORLE
504 static int lzorle_compress_pages(struct compress_ctx *cc)
508 ret = lzorle1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
509 &cc->clen, cc->private);
510 if (ret != LZO_E_OK) {
511 printk_ratelimited("%sF2FS-fs (%s): lzo-rle compress failed, ret:%d\n",
512 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
518 static const struct f2fs_compress_ops f2fs_lzorle_ops = {
519 .init_compress_ctx = lzo_init_compress_ctx,
520 .destroy_compress_ctx = lzo_destroy_compress_ctx,
521 .compress_pages = lzorle_compress_pages,
522 .decompress_pages = lzo_decompress_pages,
527 static const struct f2fs_compress_ops *f2fs_cops[COMPRESS_MAX] = {
528 #ifdef CONFIG_F2FS_FS_LZO
533 #ifdef CONFIG_F2FS_FS_LZ4
538 #ifdef CONFIG_F2FS_FS_ZSTD
543 #if defined(CONFIG_F2FS_FS_LZO) && defined(CONFIG_F2FS_FS_LZORLE)
550 bool f2fs_is_compress_backend_ready(struct inode *inode)
552 if (!f2fs_compressed_file(inode))
554 return f2fs_cops[F2FS_I(inode)->i_compress_algorithm];
557 static mempool_t *compress_page_pool;
558 static int num_compress_pages = 512;
559 module_param(num_compress_pages, uint, 0444);
560 MODULE_PARM_DESC(num_compress_pages,
561 "Number of intermediate compress pages to preallocate");
563 int f2fs_init_compress_mempool(void)
565 compress_page_pool = mempool_create_page_pool(num_compress_pages, 0);
566 if (!compress_page_pool)
572 void f2fs_destroy_compress_mempool(void)
574 mempool_destroy(compress_page_pool);
577 static struct page *f2fs_compress_alloc_page(void)
581 page = mempool_alloc(compress_page_pool, GFP_NOFS);
587 static void f2fs_compress_free_page(struct page *page)
591 detach_page_private(page);
592 page->mapping = NULL;
594 mempool_free(page, compress_page_pool);
597 #define MAX_VMAP_RETRIES 3
599 static void *f2fs_vmap(struct page **pages, unsigned int count)
604 for (i = 0; i < MAX_VMAP_RETRIES; i++) {
605 buf = vm_map_ram(pages, count, -1);
613 static int f2fs_compress_pages(struct compress_ctx *cc)
615 struct f2fs_inode_info *fi = F2FS_I(cc->inode);
616 const struct f2fs_compress_ops *cops =
617 f2fs_cops[fi->i_compress_algorithm];
618 unsigned int max_len, new_nr_cpages;
619 struct page **new_cpages;
623 trace_f2fs_compress_pages_start(cc->inode, cc->cluster_idx,
624 cc->cluster_size, fi->i_compress_algorithm);
626 if (cops->init_compress_ctx) {
627 ret = cops->init_compress_ctx(cc);
632 max_len = COMPRESS_HEADER_SIZE + cc->clen;
633 cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE);
635 cc->cpages = page_array_alloc(cc->inode, cc->nr_cpages);
638 goto destroy_compress_ctx;
641 for (i = 0; i < cc->nr_cpages; i++) {
642 cc->cpages[i] = f2fs_compress_alloc_page();
643 if (!cc->cpages[i]) {
645 goto out_free_cpages;
649 cc->rbuf = f2fs_vmap(cc->rpages, cc->cluster_size);
652 goto out_free_cpages;
655 cc->cbuf = f2fs_vmap(cc->cpages, cc->nr_cpages);
658 goto out_vunmap_rbuf;
661 ret = cops->compress_pages(cc);
663 goto out_vunmap_cbuf;
665 max_len = PAGE_SIZE * (cc->cluster_size - 1) - COMPRESS_HEADER_SIZE;
667 if (cc->clen > max_len) {
669 goto out_vunmap_cbuf;
672 cc->cbuf->clen = cpu_to_le32(cc->clen);
674 if (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)
675 chksum = f2fs_crc32(F2FS_I_SB(cc->inode),
676 cc->cbuf->cdata, cc->clen);
677 cc->cbuf->chksum = cpu_to_le32(chksum);
679 for (i = 0; i < COMPRESS_DATA_RESERVED_SIZE; i++)
680 cc->cbuf->reserved[i] = cpu_to_le32(0);
682 new_nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE);
684 /* Now we're going to cut unnecessary tail pages */
685 new_cpages = page_array_alloc(cc->inode, new_nr_cpages);
688 goto out_vunmap_cbuf;
691 /* zero out any unused part of the last page */
692 memset(&cc->cbuf->cdata[cc->clen], 0,
693 (new_nr_cpages * PAGE_SIZE) -
694 (cc->clen + COMPRESS_HEADER_SIZE));
696 vm_unmap_ram(cc->cbuf, cc->nr_cpages);
697 vm_unmap_ram(cc->rbuf, cc->cluster_size);
699 for (i = 0; i < cc->nr_cpages; i++) {
700 if (i < new_nr_cpages) {
701 new_cpages[i] = cc->cpages[i];
704 f2fs_compress_free_page(cc->cpages[i]);
705 cc->cpages[i] = NULL;
708 if (cops->destroy_compress_ctx)
709 cops->destroy_compress_ctx(cc);
711 page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
712 cc->cpages = new_cpages;
713 cc->nr_cpages = new_nr_cpages;
715 trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
720 vm_unmap_ram(cc->cbuf, cc->nr_cpages);
722 vm_unmap_ram(cc->rbuf, cc->cluster_size);
724 for (i = 0; i < cc->nr_cpages; i++) {
726 f2fs_compress_free_page(cc->cpages[i]);
728 page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
730 destroy_compress_ctx:
731 if (cops->destroy_compress_ctx)
732 cops->destroy_compress_ctx(cc);
734 trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
739 static void f2fs_decompress_cluster(struct decompress_io_ctx *dic)
741 struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
742 struct f2fs_inode_info *fi = F2FS_I(dic->inode);
743 const struct f2fs_compress_ops *cops =
744 f2fs_cops[fi->i_compress_algorithm];
748 trace_f2fs_decompress_pages_start(dic->inode, dic->cluster_idx,
749 dic->cluster_size, fi->i_compress_algorithm);
756 dic->tpages = page_array_alloc(dic->inode, dic->cluster_size);
762 for (i = 0; i < dic->cluster_size; i++) {
763 if (dic->rpages[i]) {
764 dic->tpages[i] = dic->rpages[i];
768 dic->tpages[i] = f2fs_compress_alloc_page();
769 if (!dic->tpages[i]) {
775 if (cops->init_decompress_ctx) {
776 ret = cops->init_decompress_ctx(dic);
781 dic->rbuf = f2fs_vmap(dic->tpages, dic->cluster_size);
784 goto out_destroy_decompress_ctx;
787 dic->cbuf = f2fs_vmap(dic->cpages, dic->nr_cpages);
790 goto out_vunmap_rbuf;
793 dic->clen = le32_to_cpu(dic->cbuf->clen);
794 dic->rlen = PAGE_SIZE << dic->log_cluster_size;
796 if (dic->clen > PAGE_SIZE * dic->nr_cpages - COMPRESS_HEADER_SIZE) {
798 goto out_vunmap_cbuf;
801 ret = cops->decompress_pages(dic);
803 if (!ret && (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)) {
804 u32 provided = le32_to_cpu(dic->cbuf->chksum);
805 u32 calculated = f2fs_crc32(sbi, dic->cbuf->cdata, dic->clen);
807 if (provided != calculated) {
808 if (!is_inode_flag_set(dic->inode, FI_COMPRESS_CORRUPT)) {
809 set_inode_flag(dic->inode, FI_COMPRESS_CORRUPT);
811 "%sF2FS-fs (%s): checksum invalid, nid = %lu, %x vs %x",
812 KERN_INFO, sbi->sb->s_id, dic->inode->i_ino,
813 provided, calculated);
815 set_sbi_flag(sbi, SBI_NEED_FSCK);
820 vm_unmap_ram(dic->cbuf, dic->nr_cpages);
822 vm_unmap_ram(dic->rbuf, dic->cluster_size);
823 out_destroy_decompress_ctx:
824 if (cops->destroy_decompress_ctx)
825 cops->destroy_decompress_ctx(dic);
827 trace_f2fs_decompress_pages_end(dic->inode, dic->cluster_idx,
829 f2fs_decompress_end_io(dic, ret);
833 * This is called when a page of a compressed cluster has been read from disk
834 * (or failed to be read from disk). It checks whether this page was the last
835 * page being waited on in the cluster, and if so, it decompresses the cluster
836 * (or in the case of a failure, cleans up without actually decompressing).
838 void f2fs_end_read_compressed_page(struct page *page, bool failed)
840 struct decompress_io_ctx *dic =
841 (struct decompress_io_ctx *)page_private(page);
842 struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
844 dec_page_count(sbi, F2FS_RD_DATA);
847 WRITE_ONCE(dic->failed, true);
849 if (atomic_dec_and_test(&dic->remaining_pages))
850 f2fs_decompress_cluster(dic);
853 static bool is_page_in_cluster(struct compress_ctx *cc, pgoff_t index)
855 if (cc->cluster_idx == NULL_CLUSTER)
857 return cc->cluster_idx == cluster_idx(cc, index);
860 bool f2fs_cluster_is_empty(struct compress_ctx *cc)
862 return cc->nr_rpages == 0;
865 static bool f2fs_cluster_is_full(struct compress_ctx *cc)
867 return cc->cluster_size == cc->nr_rpages;
870 bool f2fs_cluster_can_merge_page(struct compress_ctx *cc, pgoff_t index)
872 if (f2fs_cluster_is_empty(cc))
874 return is_page_in_cluster(cc, index);
877 static bool cluster_has_invalid_data(struct compress_ctx *cc)
879 loff_t i_size = i_size_read(cc->inode);
880 unsigned nr_pages = DIV_ROUND_UP(i_size, PAGE_SIZE);
883 for (i = 0; i < cc->cluster_size; i++) {
884 struct page *page = cc->rpages[i];
886 f2fs_bug_on(F2FS_I_SB(cc->inode), !page);
889 if (page->index >= nr_pages)
895 static int __f2fs_cluster_blocks(struct compress_ctx *cc, bool compr)
897 struct dnode_of_data dn;
900 set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
901 ret = f2fs_get_dnode_of_data(&dn, start_idx_of_cluster(cc),
909 if (dn.data_blkaddr == COMPRESS_ADDR) {
913 for (i = 1; i < cc->cluster_size; i++) {
916 blkaddr = data_blkaddr(dn.inode,
917 dn.node_page, dn.ofs_in_node + i);
919 if (__is_valid_data_blkaddr(blkaddr))
922 if (blkaddr != NULL_ADDR)
932 /* return # of compressed blocks in compressed cluster */
933 static int f2fs_compressed_blocks(struct compress_ctx *cc)
935 return __f2fs_cluster_blocks(cc, true);
938 /* return # of valid blocks in compressed cluster */
939 static int f2fs_cluster_blocks(struct compress_ctx *cc)
941 return __f2fs_cluster_blocks(cc, false);
944 int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index)
946 struct compress_ctx cc = {
948 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
949 .cluster_size = F2FS_I(inode)->i_cluster_size,
950 .cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
953 return f2fs_cluster_blocks(&cc);
956 static bool cluster_may_compress(struct compress_ctx *cc)
958 if (!f2fs_need_compress_data(cc->inode))
960 if (f2fs_is_atomic_file(cc->inode))
962 if (f2fs_is_mmap_file(cc->inode))
964 if (!f2fs_cluster_is_full(cc))
966 if (unlikely(f2fs_cp_error(F2FS_I_SB(cc->inode))))
968 return !cluster_has_invalid_data(cc);
971 static void set_cluster_writeback(struct compress_ctx *cc)
975 for (i = 0; i < cc->cluster_size; i++) {
977 set_page_writeback(cc->rpages[i]);
981 static void set_cluster_dirty(struct compress_ctx *cc)
985 for (i = 0; i < cc->cluster_size; i++)
987 set_page_dirty(cc->rpages[i]);
990 static int prepare_compress_overwrite(struct compress_ctx *cc,
991 struct page **pagep, pgoff_t index, void **fsdata)
993 struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
994 struct address_space *mapping = cc->inode->i_mapping;
996 struct dnode_of_data dn;
997 sector_t last_block_in_bio;
998 unsigned fgp_flag = FGP_LOCK | FGP_WRITE | FGP_CREAT;
999 pgoff_t start_idx = start_idx_of_cluster(cc);
1004 ret = f2fs_cluster_blocks(cc);
1008 /* compressed case */
1009 prealloc = (ret < cc->cluster_size);
1011 ret = f2fs_init_compress_ctx(cc);
1015 /* keep page reference to avoid page reclaim */
1016 for (i = 0; i < cc->cluster_size; i++) {
1017 page = f2fs_pagecache_get_page(mapping, start_idx + i,
1018 fgp_flag, GFP_NOFS);
1024 if (PageUptodate(page))
1025 f2fs_put_page(page, 1);
1027 f2fs_compress_ctx_add_page(cc, page);
1030 if (!f2fs_cluster_is_empty(cc)) {
1031 struct bio *bio = NULL;
1033 ret = f2fs_read_multi_pages(cc, &bio, cc->cluster_size,
1034 &last_block_in_bio, false, true);
1035 f2fs_put_rpages(cc);
1036 f2fs_destroy_compress_ctx(cc, true);
1040 f2fs_submit_bio(sbi, bio, DATA);
1042 ret = f2fs_init_compress_ctx(cc);
1047 for (i = 0; i < cc->cluster_size; i++) {
1048 f2fs_bug_on(sbi, cc->rpages[i]);
1050 page = find_lock_page(mapping, start_idx + i);
1052 /* page can be truncated */
1053 goto release_and_retry;
1056 f2fs_wait_on_page_writeback(page, DATA, true, true);
1057 f2fs_compress_ctx_add_page(cc, page);
1059 if (!PageUptodate(page)) {
1061 f2fs_put_rpages(cc);
1062 f2fs_unlock_rpages(cc, i + 1);
1063 f2fs_destroy_compress_ctx(cc, true);
1069 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
1071 set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1073 for (i = cc->cluster_size - 1; i > 0; i--) {
1074 ret = f2fs_get_block(&dn, start_idx + i);
1076 i = cc->cluster_size;
1080 if (dn.data_blkaddr != NEW_ADDR)
1084 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
1088 *fsdata = cc->rpages;
1089 *pagep = cc->rpages[offset_in_cluster(cc, index)];
1090 return cc->cluster_size;
1094 f2fs_put_rpages(cc);
1095 f2fs_unlock_rpages(cc, i);
1096 f2fs_destroy_compress_ctx(cc, true);
1101 int f2fs_prepare_compress_overwrite(struct inode *inode,
1102 struct page **pagep, pgoff_t index, void **fsdata)
1104 struct compress_ctx cc = {
1106 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1107 .cluster_size = F2FS_I(inode)->i_cluster_size,
1108 .cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
1113 return prepare_compress_overwrite(&cc, pagep, index, fsdata);
1116 bool f2fs_compress_write_end(struct inode *inode, void *fsdata,
1117 pgoff_t index, unsigned copied)
1120 struct compress_ctx cc = {
1122 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1123 .cluster_size = F2FS_I(inode)->i_cluster_size,
1126 bool first_index = (index == cc.rpages[0]->index);
1129 set_cluster_dirty(&cc);
1131 f2fs_put_rpages_wbc(&cc, NULL, false, 1);
1132 f2fs_destroy_compress_ctx(&cc, false);
1137 int f2fs_truncate_partial_cluster(struct inode *inode, u64 from, bool lock)
1139 void *fsdata = NULL;
1141 int log_cluster_size = F2FS_I(inode)->i_log_cluster_size;
1142 pgoff_t start_idx = from >> (PAGE_SHIFT + log_cluster_size) <<
1146 err = f2fs_is_compressed_cluster(inode, start_idx);
1150 /* truncate normal cluster */
1152 return f2fs_do_truncate_blocks(inode, from, lock);
1154 /* truncate compressed cluster */
1155 err = f2fs_prepare_compress_overwrite(inode, &pagep,
1156 start_idx, &fsdata);
1158 /* should not be a normal cluster */
1159 f2fs_bug_on(F2FS_I_SB(inode), err == 0);
1165 struct page **rpages = fsdata;
1166 int cluster_size = F2FS_I(inode)->i_cluster_size;
1169 for (i = cluster_size - 1; i >= 0; i--) {
1170 loff_t start = rpages[i]->index << PAGE_SHIFT;
1172 if (from <= start) {
1173 zero_user_segment(rpages[i], 0, PAGE_SIZE);
1175 zero_user_segment(rpages[i], from - start,
1181 f2fs_compress_write_end(inode, fsdata, start_idx, true);
1186 static int f2fs_write_compressed_pages(struct compress_ctx *cc,
1188 struct writeback_control *wbc,
1189 enum iostat_type io_type)
1191 struct inode *inode = cc->inode;
1192 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1193 struct f2fs_inode_info *fi = F2FS_I(inode);
1194 struct f2fs_io_info fio = {
1196 .ino = cc->inode->i_ino,
1199 .op_flags = wbc_to_write_flags(wbc),
1200 .old_blkaddr = NEW_ADDR,
1202 .encrypted_page = NULL,
1203 .compressed_page = NULL,
1207 .encrypted = fscrypt_inode_uses_fs_layer_crypto(cc->inode),
1209 struct dnode_of_data dn;
1210 struct node_info ni;
1211 struct compress_io_ctx *cic;
1212 pgoff_t start_idx = start_idx_of_cluster(cc);
1213 unsigned int last_index = cc->cluster_size - 1;
1217 /* we should bypass data pages to proceed the kworkder jobs */
1218 if (unlikely(f2fs_cp_error(sbi))) {
1219 mapping_set_error(cc->rpages[0]->mapping, -EIO);
1223 if (IS_NOQUOTA(inode)) {
1225 * We need to wait for node_write to avoid block allocation during
1226 * checkpoint. This can only happen to quota writes which can cause
1227 * the below discard race condition.
1229 down_read(&sbi->node_write);
1230 } else if (!f2fs_trylock_op(sbi)) {
1234 set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1236 err = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
1240 for (i = 0; i < cc->cluster_size; i++) {
1241 if (data_blkaddr(dn.inode, dn.node_page,
1242 dn.ofs_in_node + i) == NULL_ADDR)
1246 psize = (loff_t)(cc->rpages[last_index]->index + 1) << PAGE_SHIFT;
1248 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1252 fio.version = ni.version;
1254 cic = kmem_cache_zalloc(cic_entry_slab, GFP_NOFS);
1258 cic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1260 atomic_set(&cic->pending_pages, cc->nr_cpages);
1261 cic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1265 cic->nr_rpages = cc->cluster_size;
1267 for (i = 0; i < cc->nr_cpages; i++) {
1268 f2fs_set_compressed_page(cc->cpages[i], inode,
1269 cc->rpages[i + 1]->index, cic);
1270 fio.compressed_page = cc->cpages[i];
1272 fio.old_blkaddr = data_blkaddr(dn.inode, dn.node_page,
1273 dn.ofs_in_node + i + 1);
1275 /* wait for GCed page writeback via META_MAPPING */
1276 f2fs_wait_on_block_writeback(inode, fio.old_blkaddr);
1278 if (fio.encrypted) {
1279 fio.page = cc->rpages[i + 1];
1280 err = f2fs_encrypt_one_page(&fio);
1282 goto out_destroy_crypt;
1283 cc->cpages[i] = fio.encrypted_page;
1287 set_cluster_writeback(cc);
1289 for (i = 0; i < cc->cluster_size; i++)
1290 cic->rpages[i] = cc->rpages[i];
1292 for (i = 0; i < cc->cluster_size; i++, dn.ofs_in_node++) {
1295 blkaddr = f2fs_data_blkaddr(&dn);
1296 fio.page = cc->rpages[i];
1297 fio.old_blkaddr = blkaddr;
1299 /* cluster header */
1301 if (blkaddr == COMPRESS_ADDR)
1303 if (__is_valid_data_blkaddr(blkaddr))
1304 f2fs_invalidate_blocks(sbi, blkaddr);
1305 f2fs_update_data_blkaddr(&dn, COMPRESS_ADDR);
1306 goto unlock_continue;
1309 if (fio.compr_blocks && __is_valid_data_blkaddr(blkaddr))
1312 if (i > cc->nr_cpages) {
1313 if (__is_valid_data_blkaddr(blkaddr)) {
1314 f2fs_invalidate_blocks(sbi, blkaddr);
1315 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
1317 goto unlock_continue;
1320 f2fs_bug_on(fio.sbi, blkaddr == NULL_ADDR);
1323 fio.encrypted_page = cc->cpages[i - 1];
1325 fio.compressed_page = cc->cpages[i - 1];
1327 cc->cpages[i - 1] = NULL;
1328 f2fs_outplace_write_data(&dn, &fio);
1331 inode_dec_dirty_pages(cc->inode);
1332 unlock_page(fio.page);
1335 if (fio.compr_blocks)
1336 f2fs_i_compr_blocks_update(inode, fio.compr_blocks - 1, false);
1337 f2fs_i_compr_blocks_update(inode, cc->nr_cpages, true);
1338 add_compr_block_stat(inode, cc->nr_cpages);
1340 set_inode_flag(cc->inode, FI_APPEND_WRITE);
1341 if (cc->cluster_idx == 0)
1342 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1344 f2fs_put_dnode(&dn);
1345 if (IS_NOQUOTA(inode))
1346 up_read(&sbi->node_write);
1348 f2fs_unlock_op(sbi);
1350 spin_lock(&fi->i_size_lock);
1351 if (fi->last_disk_size < psize)
1352 fi->last_disk_size = psize;
1353 spin_unlock(&fi->i_size_lock);
1355 f2fs_put_rpages(cc);
1356 page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1358 f2fs_destroy_compress_ctx(cc, false);
1362 page_array_free(cc->inode, cic->rpages, cc->cluster_size);
1364 for (--i; i >= 0; i--)
1365 fscrypt_finalize_bounce_page(&cc->cpages[i]);
1366 for (i = 0; i < cc->nr_cpages; i++) {
1369 f2fs_compress_free_page(cc->cpages[i]);
1370 cc->cpages[i] = NULL;
1373 kmem_cache_free(cic_entry_slab, cic);
1375 f2fs_put_dnode(&dn);
1377 if (IS_NOQUOTA(inode))
1378 up_read(&sbi->node_write);
1380 f2fs_unlock_op(sbi);
1382 page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1387 void f2fs_compress_write_end_io(struct bio *bio, struct page *page)
1389 struct f2fs_sb_info *sbi = bio->bi_private;
1390 struct compress_io_ctx *cic =
1391 (struct compress_io_ctx *)page_private(page);
1394 if (unlikely(bio->bi_status))
1395 mapping_set_error(cic->inode->i_mapping, -EIO);
1397 f2fs_compress_free_page(page);
1399 dec_page_count(sbi, F2FS_WB_DATA);
1401 if (atomic_dec_return(&cic->pending_pages))
1404 for (i = 0; i < cic->nr_rpages; i++) {
1405 WARN_ON(!cic->rpages[i]);
1406 clear_page_private_gcing(cic->rpages[i]);
1407 end_page_writeback(cic->rpages[i]);
1410 page_array_free(cic->inode, cic->rpages, cic->nr_rpages);
1411 kmem_cache_free(cic_entry_slab, cic);
1414 static int f2fs_write_raw_pages(struct compress_ctx *cc,
1416 struct writeback_control *wbc,
1417 enum iostat_type io_type)
1419 struct address_space *mapping = cc->inode->i_mapping;
1420 int _submitted, compr_blocks, ret;
1421 int i = -1, err = 0;
1423 compr_blocks = f2fs_compressed_blocks(cc);
1424 if (compr_blocks < 0) {
1429 for (i = 0; i < cc->cluster_size; i++) {
1433 if (cc->rpages[i]->mapping != mapping) {
1434 unlock_page(cc->rpages[i]);
1438 BUG_ON(!PageLocked(cc->rpages[i]));
1440 ret = f2fs_write_single_data_page(cc->rpages[i], &_submitted,
1441 NULL, NULL, wbc, io_type,
1442 compr_blocks, false);
1444 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1445 unlock_page(cc->rpages[i]);
1447 } else if (ret == -EAGAIN) {
1449 * for quota file, just redirty left pages to
1450 * avoid deadlock caused by cluster update race
1451 * from foreground operation.
1453 if (IS_NOQUOTA(cc->inode)) {
1459 congestion_wait(BLK_RW_ASYNC,
1460 DEFAULT_IO_TIMEOUT);
1461 lock_page(cc->rpages[i]);
1463 if (!PageDirty(cc->rpages[i])) {
1464 unlock_page(cc->rpages[i]);
1468 clear_page_dirty_for_io(cc->rpages[i]);
1475 *submitted += _submitted;
1478 f2fs_balance_fs(F2FS_M_SB(mapping), true);
1482 for (++i; i < cc->cluster_size; i++) {
1485 redirty_page_for_writepage(wbc, cc->rpages[i]);
1486 unlock_page(cc->rpages[i]);
1491 int f2fs_write_multi_pages(struct compress_ctx *cc,
1493 struct writeback_control *wbc,
1494 enum iostat_type io_type)
1499 if (cluster_may_compress(cc)) {
1500 err = f2fs_compress_pages(cc);
1501 if (err == -EAGAIN) {
1504 f2fs_put_rpages_wbc(cc, wbc, true, 1);
1508 err = f2fs_write_compressed_pages(cc, submitted,
1512 f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN);
1515 f2fs_bug_on(F2FS_I_SB(cc->inode), *submitted);
1517 err = f2fs_write_raw_pages(cc, submitted, wbc, io_type);
1518 f2fs_put_rpages_wbc(cc, wbc, false, 0);
1520 f2fs_destroy_compress_ctx(cc, false);
1524 static void f2fs_free_dic(struct decompress_io_ctx *dic);
1526 struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
1528 struct decompress_io_ctx *dic;
1529 pgoff_t start_idx = start_idx_of_cluster(cc);
1532 dic = kmem_cache_zalloc(dic_entry_slab, GFP_NOFS);
1534 return ERR_PTR(-ENOMEM);
1536 dic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1538 kmem_cache_free(dic_entry_slab, dic);
1539 return ERR_PTR(-ENOMEM);
1542 dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1543 dic->inode = cc->inode;
1544 atomic_set(&dic->remaining_pages, cc->nr_cpages);
1545 dic->cluster_idx = cc->cluster_idx;
1546 dic->cluster_size = cc->cluster_size;
1547 dic->log_cluster_size = cc->log_cluster_size;
1548 dic->nr_cpages = cc->nr_cpages;
1549 refcount_set(&dic->refcnt, 1);
1550 dic->failed = false;
1551 dic->need_verity = f2fs_need_verity(cc->inode, start_idx);
1553 for (i = 0; i < dic->cluster_size; i++)
1554 dic->rpages[i] = cc->rpages[i];
1555 dic->nr_rpages = cc->cluster_size;
1557 dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages);
1561 for (i = 0; i < dic->nr_cpages; i++) {
1564 page = f2fs_compress_alloc_page();
1568 f2fs_set_compressed_page(page, cc->inode,
1569 start_idx + i + 1, dic);
1570 dic->cpages[i] = page;
1577 return ERR_PTR(-ENOMEM);
1580 static void f2fs_free_dic(struct decompress_io_ctx *dic)
1585 for (i = 0; i < dic->cluster_size; i++) {
1588 if (!dic->tpages[i])
1590 f2fs_compress_free_page(dic->tpages[i]);
1592 page_array_free(dic->inode, dic->tpages, dic->cluster_size);
1596 for (i = 0; i < dic->nr_cpages; i++) {
1597 if (!dic->cpages[i])
1599 f2fs_compress_free_page(dic->cpages[i]);
1601 page_array_free(dic->inode, dic->cpages, dic->nr_cpages);
1604 page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
1605 kmem_cache_free(dic_entry_slab, dic);
1608 static void f2fs_put_dic(struct decompress_io_ctx *dic)
1610 if (refcount_dec_and_test(&dic->refcnt))
1615 * Update and unlock the cluster's pagecache pages, and release the reference to
1616 * the decompress_io_ctx that was being held for I/O completion.
1618 static void __f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1622 for (i = 0; i < dic->cluster_size; i++) {
1623 struct page *rpage = dic->rpages[i];
1628 /* PG_error was set if verity failed. */
1629 if (failed || PageError(rpage)) {
1630 ClearPageUptodate(rpage);
1631 /* will re-read again later */
1632 ClearPageError(rpage);
1634 SetPageUptodate(rpage);
1642 static void f2fs_verify_cluster(struct work_struct *work)
1644 struct decompress_io_ctx *dic =
1645 container_of(work, struct decompress_io_ctx, verity_work);
1648 /* Verify the cluster's decompressed pages with fs-verity. */
1649 for (i = 0; i < dic->cluster_size; i++) {
1650 struct page *rpage = dic->rpages[i];
1652 if (rpage && !fsverity_verify_page(rpage))
1653 SetPageError(rpage);
1656 __f2fs_decompress_end_io(dic, false);
1660 * This is called when a compressed cluster has been decompressed
1661 * (or failed to be read and/or decompressed).
1663 void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1665 if (!failed && dic->need_verity) {
1667 * Note that to avoid deadlocks, the verity work can't be done
1668 * on the decompression workqueue. This is because verifying
1669 * the data pages can involve reading metadata pages from the
1670 * file, and these metadata pages may be compressed.
1672 INIT_WORK(&dic->verity_work, f2fs_verify_cluster);
1673 fsverity_enqueue_verify_work(&dic->verity_work);
1675 __f2fs_decompress_end_io(dic, failed);
1680 * Put a reference to a compressed page's decompress_io_ctx.
1682 * This is called when the page is no longer needed and can be freed.
1684 void f2fs_put_page_dic(struct page *page)
1686 struct decompress_io_ctx *dic =
1687 (struct decompress_io_ctx *)page_private(page);
1692 int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi)
1694 dev_t dev = sbi->sb->s_bdev->bd_dev;
1697 sprintf(slab_name, "f2fs_page_array_entry-%u:%u", MAJOR(dev), MINOR(dev));
1699 sbi->page_array_slab_size = sizeof(struct page *) <<
1700 F2FS_OPTION(sbi).compress_log_size;
1702 sbi->page_array_slab = f2fs_kmem_cache_create(slab_name,
1703 sbi->page_array_slab_size);
1704 if (!sbi->page_array_slab)
1709 void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi)
1711 kmem_cache_destroy(sbi->page_array_slab);
1714 static int __init f2fs_init_cic_cache(void)
1716 cic_entry_slab = f2fs_kmem_cache_create("f2fs_cic_entry",
1717 sizeof(struct compress_io_ctx));
1718 if (!cic_entry_slab)
1723 static void f2fs_destroy_cic_cache(void)
1725 kmem_cache_destroy(cic_entry_slab);
1728 static int __init f2fs_init_dic_cache(void)
1730 dic_entry_slab = f2fs_kmem_cache_create("f2fs_dic_entry",
1731 sizeof(struct decompress_io_ctx));
1732 if (!dic_entry_slab)
1737 static void f2fs_destroy_dic_cache(void)
1739 kmem_cache_destroy(dic_entry_slab);
1742 int __init f2fs_init_compress_cache(void)
1746 err = f2fs_init_cic_cache();
1749 err = f2fs_init_dic_cache();
1754 f2fs_destroy_cic_cache();
1759 void f2fs_destroy_compress_cache(void)
1761 f2fs_destroy_dic_cache();
1762 f2fs_destroy_cic_cache();