~hc/RK356X_SDK_RELEASE.git

..	..	@@ -12,11 +12,39 @@
12	12	#include <linux/lzo.h>
13	13	#include <linux/lz4.h>
14	14	#include <linux/zstd.h>
15		-#include <linux/moduleparam.h>
	15	+#include <linux/pagevec.h>
16	16
17	17	#include "f2fs.h"
18	18	#include "node.h"
	19	+#include "segment.h"
19	20	#include <trace/events/f2fs.h>
	21	+
	22	+static struct kmem_cache *cic_entry_slab;
	23	+static struct kmem_cache *dic_entry_slab;
	24	+
	25	+static void page_array_alloc(struct inode inode, int nr)
	26	+{
	27	+ struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
	28	+ unsigned int size = sizeof(struct page ) nr;
	29	+
	30	+ if (likely(size <= sbi->page_array_slab_size))
	31	+ return kmem_cache_zalloc(sbi->page_array_slab, GFP_NOFS);
	32	+ return f2fs_kzalloc(sbi, size, GFP_NOFS);
	33	+}
	34	+
	35	+static void page_array_free(struct inode inode, void pages, int nr)
	36	+{
	37	+ struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
	38	+ unsigned int size = sizeof(struct page ) nr;
	39	+
	40	+ if (!pages)
	41	+ return;
	42	+
	43	+ if (likely(size <= sbi->page_array_slab_size))
	44	+ kmem_cache_free(sbi->page_array_slab, pages);
	45	+ else
	46	+ kfree(pages);
	47	+}
20	48
21	49	struct f2fs_compress_ops {
22	50	int (init_compress_ctx)(struct compress_ctx cc);
..	..	@@ -48,8 +76,9 @@
48	76	return false;
49	77	if (!page_private(page))
50	78	return false;
51		- if (IS_ATOMIC_WRITTEN_PAGE(page) \|\| IS_DUMMY_WRITTEN_PAGE(page))
	79	+ if (page_private_nonpointer(page))
52	80	return false;
	81	+
53	82	f2fs_bug_on(F2FS_M_SB(page->mapping),
54	83	((u32 )page_private(page)) != F2FS_COMPRESSED_PAGE_MAGIC);
55	84	return true;
..	..	@@ -58,8 +87,7 @@
58	87	static void f2fs_set_compressed_page(struct page *page,
59	88	struct inode inode, pgoff_t index, void data)
60	89	{
61		- SetPagePrivate(page);
62		- set_page_private(page, (unsigned long)data);
	90	+ attach_page_private(page, (void *)data);
63	91
64	92	/* i_crypto_info and iv index */
65	93	page->index = index;
..	..	@@ -90,19 +118,6 @@
90	118	f2fs_drop_rpages(cc, len, true);
91	119	}
92	120
93		-static void f2fs_put_rpages_mapping(struct address_space *mapping,
94		- pgoff_t start, int len)
95		-{
96		- int i;
97		-
98		- for (i = 0; i < len; i++) {
99		- struct page *page = find_get_page(mapping, start + i);
100		-
101		- put_page(page);
102		- put_page(page);
103		- }
104		-}
105		-
106	121	static void f2fs_put_rpages_wbc(struct compress_ctx *cc,
107	122	struct writeback_control *wbc, bool redirty, int unlock)
108	123	{
..	..	@@ -124,23 +139,21 @@
124	139
125	140	int f2fs_init_compress_ctx(struct compress_ctx *cc)
126	141	{
127		- struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
128		-
129		- if (cc->nr_rpages)
	142	+ if (cc->rpages)
130	143	return 0;
131	144
132		- cc->rpages = f2fs_kzalloc(sbi, sizeof(struct page *) <<
133		- cc->log_cluster_size, GFP_NOFS);
	145	+ cc->rpages = page_array_alloc(cc->inode, cc->cluster_size);
134	146	return cc->rpages ? 0 : -ENOMEM;
135	147	}
136	148
137		-void f2fs_destroy_compress_ctx(struct compress_ctx *cc)
	149	+void f2fs_destroy_compress_ctx(struct compress_ctx *cc, bool reuse)
138	150	{
139		- kfree(cc->rpages);
	151	+ page_array_free(cc->inode, cc->rpages, cc->cluster_size);
140	152	cc->rpages = NULL;
141	153	cc->nr_rpages = 0;
142	154	cc->nr_cpages = 0;
143		- cc->cluster_idx = NULL_CLUSTER;
	155	+ if (!reuse)
	156	+ cc->cluster_idx = NULL_CLUSTER;
144	157	}
145	158
146	159	void f2fs_compress_ctx_add_page(struct compress_ctx cc, struct page page)
..	..	@@ -222,8 +235,14 @@
222	235	#ifdef CONFIG_F2FS_FS_LZ4
223	236	static int lz4_init_compress_ctx(struct compress_ctx *cc)
224	237	{
225		- cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
226		- LZ4_MEM_COMPRESS, GFP_NOFS);
	238	+ unsigned int size = LZ4_MEM_COMPRESS;
	239	+
	240	+#ifdef CONFIG_F2FS_FS_LZ4HC
	241	+ if (F2FS_I(cc->inode)->i_compress_flag >> COMPRESS_LEVEL_OFFSET)
	242	+ size = LZ4HC_MEM_COMPRESS;
	243	+#endif
	244	+
	245	+ cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), size, GFP_NOFS);
227	246	if (!cc->private)
228	247	return -ENOMEM;
229	248
..	..	@@ -242,10 +261,34 @@
242	261	cc->private = NULL;
243	262	}
244	263
	264	+#ifdef CONFIG_F2FS_FS_LZ4HC
	265	+static int lz4hc_compress_pages(struct compress_ctx *cc)
	266	+{
	267	+ unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
	268	+ COMPRESS_LEVEL_OFFSET;
	269	+ int len;
	270	+
	271	+ if (level)
	272	+ len = LZ4_compress_HC(cc->rbuf, cc->cbuf->cdata, cc->rlen,
	273	+ cc->clen, level, cc->private);
	274	+ else
	275	+ len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
	276	+ cc->clen, cc->private);
	277	+ if (!len)
	278	+ return -EAGAIN;
	279	+
	280	+ cc->clen = len;
	281	+ return 0;
	282	+}
	283	+#endif
	284	+
245	285	static int lz4_compress_pages(struct compress_ctx *cc)
246	286	{
247	287	int len;
248	288
	289	+#ifdef CONFIG_F2FS_FS_LZ4HC
	290	+ return lz4hc_compress_pages(cc);
	291	+#endif
249	292	len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
250	293	cc->clen, cc->private);
251	294	if (!len)
..	..	@@ -268,10 +311,9 @@
268	311	}
269	312
270	313	if (ret != PAGE_SIZE << dic->log_cluster_size) {
271		- printk_ratelimited("%sF2FS-fs (%s): lz4 invalid rlen:%zu, "
	314	+ printk_ratelimited("%sF2FS-fs (%s): lz4 invalid ret:%d, "
272	315	"expected:%lu\n", KERN_ERR,
273		- F2FS_I_SB(dic->inode)->sb->s_id,
274		- dic->rlen,
	316	+ F2FS_I_SB(dic->inode)->sb->s_id, ret,
275	317	PAGE_SIZE << dic->log_cluster_size);
276	318	return -EIO;
277	319	}
..	..	@@ -295,8 +337,13 @@
295	337	ZSTD_CStream *stream;
296	338	void *workspace;
297	339	unsigned int workspace_size;
	340	+ unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
	341	+ COMPRESS_LEVEL_OFFSET;
298	342
299		- params = ZSTD_getParams(F2FS_ZSTD_DEFAULT_CLEVEL, cc->rlen, 0);
	343	+ if (!level)
	344	+ level = F2FS_ZSTD_DEFAULT_CLEVEL;
	345	+
	346	+ params = ZSTD_getParams(level, cc->rlen, 0);
300	347	workspace_size = ZSTD_CStreamWorkspaceBound(params.cParams);
301	348
302	349	workspace = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
..	..	@@ -376,16 +423,17 @@
376	423	ZSTD_DStream *stream;
377	424	void *workspace;
378	425	unsigned int workspace_size;
	426	+ unsigned int max_window_size =
	427	+ MAX_COMPRESS_WINDOW_SIZE(dic->log_cluster_size);
379	428
380		- workspace_size = ZSTD_DStreamWorkspaceBound(MAX_COMPRESS_WINDOW_SIZE);
	429	+ workspace_size = ZSTD_DStreamWorkspaceBound(max_window_size);
381	430
382	431	workspace = f2fs_kvmalloc(F2FS_I_SB(dic->inode),
383	432	workspace_size, GFP_NOFS);
384	433	if (!workspace)
385	434	return -ENOMEM;
386	435
387		- stream = ZSTD_initDStream(MAX_COMPRESS_WINDOW_SIZE,
388		- workspace, workspace_size);
	436	+ stream = ZSTD_initDStream(max_window_size, workspace, workspace_size);
389	437	if (!stream) {
390	438	printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initDStream failed\n",
391	439	KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
..	..	@@ -452,6 +500,31 @@
452	500	};
453	501	#endif
454	502
	503	+#ifdef CONFIG_F2FS_FS_LZO
	504	+#ifdef CONFIG_F2FS_FS_LZORLE
	505	+static int lzorle_compress_pages(struct compress_ctx *cc)
	506	+{
	507	+ int ret;
	508	+
	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);
	514	+ return -EIO;
	515	+ }
	516	+ return 0;
	517	+}
	518	+
	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,
	524	+};
	525	+#endif
	526	+#endif
	527	+
455	528	static const struct f2fs_compress_ops *f2fs_cops[COMPRESS_MAX] = {
456	529	#ifdef CONFIG_F2FS_FS_LZO
457	530	&f2fs_lzo_ops,
..	..	@@ -468,6 +541,11 @@
468	541	#else
469	542	NULL,
470	543	#endif
	544	+#if defined(CONFIG_F2FS_FS_LZO) && defined(CONFIG_F2FS_FS_LZORLE)
	545	+ &f2fs_lzorle_ops,
	546	+#else
	547	+ NULL,
	548	+#endif
471	549	};
472	550
473	551	bool f2fs_is_compress_backend_ready(struct inode *inode)
..	..	@@ -477,7 +555,7 @@
477	555	return f2fs_cops[F2FS_I(inode)->i_compress_algorithm];
478	556	}
479	557
480		-static mempool_t *compress_page_pool = NULL;
	558	+static mempool_t *compress_page_pool;
481	559	static int num_compress_pages = 512;
482	560	module_param(num_compress_pages, uint, 0444);
483	561	MODULE_PARM_DESC(num_compress_pages,
..	..	@@ -511,20 +589,36 @@
511	589	{
512	590	if (!page)
513	591	return;
514		- set_page_private(page, (unsigned long)NULL);
515		- ClearPagePrivate(page);
	592	+ detach_page_private(page);
516	593	page->mapping = NULL;
517	594	unlock_page(page);
518	595	mempool_free(page, compress_page_pool);
519	596	}
520	597
	598	+#define MAX_VMAP_RETRIES 3
	599	+
	600	+static void f2fs_vmap(struct page *pages, unsigned int count)
	601	+{
	602	+ int i;
	603	+ void *buf = NULL;
	604	+
	605	+ for (i = 0; i < MAX_VMAP_RETRIES; i++) {
	606	+ buf = vm_map_ram(pages, count, -1);
	607	+ if (buf)
	608	+ break;
	609	+ vm_unmap_aliases();
	610	+ }
	611	+ return buf;
	612	+}
	613	+
521	614	static int f2fs_compress_pages(struct compress_ctx *cc)
522	615	{
523		- struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
524	616	struct f2fs_inode_info *fi = F2FS_I(cc->inode);
525	617	const struct f2fs_compress_ops *cops =
526	618	f2fs_cops[fi->i_compress_algorithm];
527		- unsigned int max_len, nr_cpages;
	619	+ unsigned int max_len, new_nr_cpages;
	620	+ struct page **new_cpages;
	621	+ u32 chksum = 0;
528	622	int i, ret;
529	623
530	624	trace_f2fs_compress_pages_start(cc->inode, cc->cluster_idx,
..	..	@@ -539,8 +633,7 @@
539	633	max_len = COMPRESS_HEADER_SIZE + cc->clen;
540	634	cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE);
541	635
542		- cc->cpages = f2fs_kzalloc(sbi, sizeof(struct page )
543		- cc->nr_cpages, GFP_NOFS);
	636	+ cc->cpages = page_array_alloc(cc->inode, cc->nr_cpages);
544	637	if (!cc->cpages) {
545	638	ret = -ENOMEM;
546	639	goto destroy_compress_ctx;
..	..	@@ -554,13 +647,13 @@
554	647	}
555	648	}
556	649
557		- cc->rbuf = vmap(cc->rpages, cc->cluster_size, VM_MAP, PAGE_KERNEL_RO);
	650	+ cc->rbuf = f2fs_vmap(cc->rpages, cc->cluster_size);
558	651	if (!cc->rbuf) {
559	652	ret = -ENOMEM;
560	653	goto out_free_cpages;
561	654	}
562	655
563		- cc->cbuf = vmap(cc->cpages, cc->nr_cpages, VM_MAP, PAGE_KERNEL);
	656	+ cc->cbuf = f2fs_vmap(cc->cpages, cc->nr_cpages);
564	657	if (!cc->cbuf) {
565	658	ret = -ENOMEM;
566	659	goto out_vunmap_rbuf;
..	..	@@ -579,19 +672,36 @@
579	672
580	673	cc->cbuf->clen = cpu_to_le32(cc->clen);
581	674
	675	+ if (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)
	676	+ chksum = f2fs_crc32(F2FS_I_SB(cc->inode),
	677	+ cc->cbuf->cdata, cc->clen);
	678	+ cc->cbuf->chksum = cpu_to_le32(chksum);
	679	+
582	680	for (i = 0; i < COMPRESS_DATA_RESERVED_SIZE; i++)
583	681	cc->cbuf->reserved[i] = cpu_to_le32(0);
584	682
585		- nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE);
	683	+ new_nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE);
	684	+
	685	+ /* Now we're going to cut unnecessary tail pages */
	686	+ new_cpages = page_array_alloc(cc->inode, new_nr_cpages);
	687	+ if (!new_cpages) {
	688	+ ret = -ENOMEM;
	689	+ goto out_vunmap_cbuf;
	690	+ }
586	691
587	692	/* zero out any unused part of the last page */
588	693	memset(&cc->cbuf->cdata[cc->clen], 0,
589		- (nr_cpages * PAGE_SIZE) - (cc->clen + COMPRESS_HEADER_SIZE));
	694	+ (new_nr_cpages * PAGE_SIZE) -
	695	+ (cc->clen + COMPRESS_HEADER_SIZE));
590	696
591		- vunmap(cc->cbuf);
592		- vunmap(cc->rbuf);
	697	+ vm_unmap_ram(cc->cbuf, cc->nr_cpages);
	698	+ vm_unmap_ram(cc->rbuf, cc->cluster_size);
593	699
594		- for (i = nr_cpages; i < cc->nr_cpages; i++) {
	700	+ for (i = 0; i < cc->nr_cpages; i++) {
	701	+ if (i < new_nr_cpages) {
	702	+ new_cpages[i] = cc->cpages[i];
	703	+ continue;
	704	+ }
595	705	f2fs_compress_free_page(cc->cpages[i]);
596	706	cc->cpages[i] = NULL;
597	707	}
..	..	@@ -599,22 +709,24 @@
599	709	if (cops->destroy_compress_ctx)
600	710	cops->destroy_compress_ctx(cc);
601	711
602		- cc->nr_cpages = nr_cpages;
	712	+ page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
	713	+ cc->cpages = new_cpages;
	714	+ cc->nr_cpages = new_nr_cpages;
603	715
604	716	trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
605	717	cc->clen, ret);
606	718	return 0;
607	719
608	720	out_vunmap_cbuf:
609		- vunmap(cc->cbuf);
	721	+ vm_unmap_ram(cc->cbuf, cc->nr_cpages);
610	722	out_vunmap_rbuf:
611		- vunmap(cc->rbuf);
	723	+ vm_unmap_ram(cc->rbuf, cc->cluster_size);
612	724	out_free_cpages:
613	725	for (i = 0; i < cc->nr_cpages; i++) {
614	726	if (cc->cpages[i])
615	727	f2fs_compress_free_page(cc->cpages[i]);
616	728	}
617		- kfree(cc->cpages);
	729	+ page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
618	730	cc->cpages = NULL;
619	731	destroy_compress_ctx:
620	732	if (cops->destroy_compress_ctx)
..	..	@@ -625,49 +737,32 @@
625	737	return ret;
626	738	}
627	739
628		-void f2fs_decompress_pages(struct bio bio, struct page page, bool verity)
	740	+static int f2fs_prepare_decomp_mem(struct decompress_io_ctx *dic,
	741	+ bool pre_alloc);
	742	+static void f2fs_release_decomp_mem(struct decompress_io_ctx *dic,
	743	+ bool bypass_destroy_callback, bool pre_alloc);
	744	+
	745	+void f2fs_decompress_cluster(struct decompress_io_ctx *dic, bool in_task)
629	746	{
630		- struct decompress_io_ctx *dic =
631		- (struct decompress_io_ctx *)page_private(page);
632	747	struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
633		- struct f2fs_inode_info *fi= F2FS_I(dic->inode);
	748	+ struct f2fs_inode_info *fi = F2FS_I(dic->inode);
634	749	const struct f2fs_compress_ops *cops =
635	750	f2fs_cops[fi->i_compress_algorithm];
	751	+ bool bypass_callback = false;
636	752	int ret;
637		-
638		- dec_page_count(sbi, F2FS_RD_DATA);
639		-
640		- if (bio->bi_status \|\| PageError(page))
641		- dic->failed = true;
642		-
643		- if (refcount_dec_not_one(&dic->ref))
644		- return;
645	753
646	754	trace_f2fs_decompress_pages_start(dic->inode, dic->cluster_idx,
647	755	dic->cluster_size, fi->i_compress_algorithm);
648	756
649		- /* submit partial compressed pages */
650	757	if (dic->failed) {
651	758	ret = -EIO;
652		- goto out_free_dic;
	759	+ goto out_end_io;
653	760	}
654	761
655		- if (cops->init_decompress_ctx) {
656		- ret = cops->init_decompress_ctx(dic);
657		- if (ret)
658		- goto out_free_dic;
659		- }
660		-
661		- dic->rbuf = vmap(dic->tpages, dic->cluster_size, VM_MAP, PAGE_KERNEL);
662		- if (!dic->rbuf) {
663		- ret = -ENOMEM;
664		- goto destroy_decompress_ctx;
665		- }
666		-
667		- dic->cbuf = vmap(dic->cpages, dic->nr_cpages, VM_MAP, PAGE_KERNEL_RO);
668		- if (!dic->cbuf) {
669		- ret = -ENOMEM;
670		- goto out_vunmap_rbuf;
	762	+ ret = f2fs_prepare_decomp_mem(dic, false);
	763	+ if (ret) {
	764	+ bypass_callback = true;
	765	+ goto out_release;
671	766	}
672	767
673	768	dic->clen = le32_to_cpu(dic->cbuf->clen);
..	..	@@ -675,29 +770,59 @@
675	770
676	771	if (dic->clen > PAGE_SIZE * dic->nr_cpages - COMPRESS_HEADER_SIZE) {
677	772	ret = -EFSCORRUPTED;
678		- goto out_vunmap_cbuf;
	773	+ goto out_release;
679	774	}
680	775
681	776	ret = cops->decompress_pages(dic);
682	777
683		-out_vunmap_cbuf:
684		- vunmap(dic->cbuf);
685		-out_vunmap_rbuf:
686		- vunmap(dic->rbuf);
687		-destroy_decompress_ctx:
688		- if (cops->destroy_decompress_ctx)
689		- cops->destroy_decompress_ctx(dic);
690		-out_free_dic:
691		- if (verity)
692		- refcount_set(&dic->ref, dic->nr_cpages);
693		- if (!verity)
694		- f2fs_decompress_end_io(dic->rpages, dic->cluster_size,
695		- ret, false);
	778	+ if (!ret && (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)) {
	779	+ u32 provided = le32_to_cpu(dic->cbuf->chksum);
	780	+ u32 calculated = f2fs_crc32(sbi, dic->cbuf->cdata, dic->clen);
696	781
	782	+ if (provided != calculated) {
	783	+ if (!is_inode_flag_set(dic->inode, FI_COMPRESS_CORRUPT)) {
	784	+ set_inode_flag(dic->inode, FI_COMPRESS_CORRUPT);
	785	+ printk_ratelimited(
	786	+ "%sF2FS-fs (%s): checksum invalid, nid = %lu, %x vs %x",
	787	+ KERN_INFO, sbi->sb->s_id, dic->inode->i_ino,
	788	+ provided, calculated);
	789	+ }
	790	+ set_sbi_flag(sbi, SBI_NEED_FSCK);
	791	+ }
	792	+ }
	793	+
	794	+out_release:
	795	+ f2fs_release_decomp_mem(dic, bypass_callback, false);
	796	+
	797	+out_end_io:
697	798	trace_f2fs_decompress_pages_end(dic->inode, dic->cluster_idx,
698	799	dic->clen, ret);
699		- if (!verity)
700		- f2fs_free_dic(dic);
	800	+ f2fs_decompress_end_io(dic, ret, in_task);
	801	+}
	802	+
	803	+/*
	804	+ * This is called when a page of a compressed cluster has been read from disk
	805	+ * (or failed to be read from disk). It checks whether this page was the last
	806	+ * page being waited on in the cluster, and if so, it decompresses the cluster
	807	+ * (or in the case of a failure, cleans up without actually decompressing).
	808	+ */
	809	+void f2fs_end_read_compressed_page(struct page *page, bool failed,
	810	+ block_t blkaddr, bool in_task)
	811	+{
	812	+ struct decompress_io_ctx *dic =
	813	+ (struct decompress_io_ctx *)page_private(page);
	814	+ struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
	815	+
	816	+ dec_page_count(sbi, F2FS_RD_DATA);
	817	+
	818	+ if (failed)
	819	+ WRITE_ONCE(dic->failed, true);
	820	+ else if (blkaddr && in_task)
	821	+ f2fs_cache_compressed_page(sbi, page,
	822	+ dic->inode->i_ino, blkaddr);
	823	+
	824	+ if (atomic_dec_and_test(&dic->remaining_pages))
	825	+ f2fs_decompress_cluster(dic, in_task);
701	826	}
702	827
703	828	static bool is_page_in_cluster(struct compress_ctx *cc, pgoff_t index)
..	..	@@ -724,9 +849,8 @@
724	849	return is_page_in_cluster(cc, index);
725	850	}
726	851
727		-static bool __cluster_may_compress(struct compress_ctx *cc)
	852	+static bool cluster_has_invalid_data(struct compress_ctx *cc)
728	853	{
729		- struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
730	854	loff_t i_size = i_size_read(cc->inode);
731	855	unsigned nr_pages = DIV_ROUND_UP(i_size, PAGE_SIZE);
732	856	int i;
..	..	@@ -734,28 +858,26 @@
734	858	for (i = 0; i < cc->cluster_size; i++) {
735	859	struct page *page = cc->rpages[i];
736	860
737		- f2fs_bug_on(sbi, !page);
738		-
739		- if (unlikely(f2fs_cp_error(sbi)))
740		- return false;
741		- if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
742		- return false;
	861	+ f2fs_bug_on(F2FS_I_SB(cc->inode), !page);
743	862
744	863	/* beyond EOF */
745	864	if (page->index >= nr_pages)
746		- return false;
	865	+ return true;
747	866	}
748		- return true;
	867	+ return false;
749	868	}
750	869
751		-static int __f2fs_cluster_blocks(struct compress_ctx *cc, bool compr)
	870	+static int __f2fs_cluster_blocks(struct inode *inode,
	871	+ unsigned int cluster_idx, bool compr)
752	872	{
753	873	struct dnode_of_data dn;
	874	+ unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
	875	+ unsigned int start_idx = cluster_idx <<
	876	+ F2FS_I(inode)->i_log_cluster_size;
754	877	int ret;
755	878
756		- set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
757		- ret = f2fs_get_dnode_of_data(&dn, start_idx_of_cluster(cc),
758		- LOOKUP_NODE);
	879	+ set_new_dnode(&dn, inode, NULL, NULL, 0);
	880	+ ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
759	881	if (ret) {
760	882	if (ret == -ENOENT)
761	883	ret = 0;
..	..	@@ -766,7 +888,7 @@
766	888	int i;
767	889
768	890	ret = 1;
769		- for (i = 1; i < cc->cluster_size; i++) {
	891	+ for (i = 1; i < cluster_size; i++) {
770	892	block_t blkaddr;
771	893
772	894	blkaddr = data_blkaddr(dn.inode,
..	..	@@ -779,6 +901,10 @@
779	901	ret++;
780	902	}
781	903	}
	904	+
	905	+ f2fs_bug_on(F2FS_I_SB(inode),
	906	+ !compr && ret != cluster_size &&
	907	+ !is_inode_flag_set(inode, FI_COMPRESS_RELEASED));
782	908	}
783	909	fail:
784	910	f2fs_put_dnode(&dn);
..	..	@@ -788,40 +914,28 @@
788	914	/* return # of compressed blocks in compressed cluster */
789	915	static int f2fs_compressed_blocks(struct compress_ctx *cc)
790	916	{
791		- return __f2fs_cluster_blocks(cc, true);
	917	+ return __f2fs_cluster_blocks(cc->inode, cc->cluster_idx, true);
792	918	}
793	919
794	920	/* return # of valid blocks in compressed cluster */
795		-static int f2fs_cluster_blocks(struct compress_ctx *cc, bool compr)
796		-{
797		- return __f2fs_cluster_blocks(cc, false);
798		-}
799		-
800	921	int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index)
801	922	{
802		- struct compress_ctx cc = {
803		- .inode = inode,
804		- .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
805		- .cluster_size = F2FS_I(inode)->i_cluster_size,
806		- .cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
807		- };
808		-
809		- return f2fs_cluster_blocks(&cc, false);
	923	+ return __f2fs_cluster_blocks(inode,
	924	+ index >> F2FS_I(inode)->i_log_cluster_size,
	925	+ false);
810	926	}
811	927
812	928	static bool cluster_may_compress(struct compress_ctx *cc)
813	929	{
814		- if (!f2fs_compressed_file(cc->inode))
	930	+ if (!f2fs_need_compress_data(cc->inode))
815	931	return false;
816	932	if (f2fs_is_atomic_file(cc->inode))
817		- return false;
818		- if (f2fs_is_mmap_file(cc->inode))
819	933	return false;
820	934	if (!f2fs_cluster_is_full(cc))
821	935	return false;
822	936	if (unlikely(f2fs_cp_error(F2FS_I_SB(cc->inode))))
823	937	return false;
824		- return __cluster_may_compress(cc);
	938	+ return !cluster_has_invalid_data(cc);
825	939	}
826	940
827	941	static void set_cluster_writeback(struct compress_ctx *cc)
..	..	@@ -849,20 +963,15 @@
849	963	struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
850	964	struct address_space *mapping = cc->inode->i_mapping;
851	965	struct page *page;
852		- struct dnode_of_data dn;
853	966	sector_t last_block_in_bio;
854	967	unsigned fgp_flag = FGP_LOCK \| FGP_WRITE \| FGP_CREAT;
855	968	pgoff_t start_idx = start_idx_of_cluster(cc);
856	969	int i, ret;
857		- bool prealloc;
858	970
859	971	retry:
860		- ret = f2fs_cluster_blocks(cc, false);
	972	+ ret = f2fs_is_compressed_cluster(cc->inode, start_idx);
861	973	if (ret <= 0)
862	974	return ret;
863		-
864		- /* compressed case */
865		- prealloc = (ret < cc->cluster_size);
866	975
867	976	ret = f2fs_init_compress_ctx(cc);
868	977	if (ret)
..	..	@@ -878,7 +987,7 @@
878	987	}
879	988
880	989	if (PageUptodate(page))
881		- unlock_page(page);
	990	+ f2fs_put_page(page, 1);
882	991	else
883	992	f2fs_compress_ctx_add_page(cc, page);
884	993	}
..	..	@@ -888,54 +997,37 @@
888	997
889	998	ret = f2fs_read_multi_pages(cc, &bio, cc->cluster_size,
890	999	&last_block_in_bio, false, true);
891		- f2fs_destroy_compress_ctx(cc);
	1000	+ f2fs_put_rpages(cc);
	1001	+ f2fs_destroy_compress_ctx(cc, true);
892	1002	if (ret)
893		- goto release_pages;
	1003	+ goto out;
894	1004	if (bio)
895	1005	f2fs_submit_bio(sbi, bio, DATA);
896	1006
897	1007	ret = f2fs_init_compress_ctx(cc);
898	1008	if (ret)
899		- goto release_pages;
	1009	+ goto out;
900	1010	}
901	1011
902	1012	for (i = 0; i < cc->cluster_size; i++) {
903	1013	f2fs_bug_on(sbi, cc->rpages[i]);
904	1014
905	1015	page = find_lock_page(mapping, start_idx + i);
906		- f2fs_bug_on(sbi, !page);
	1016	+ if (!page) {
	1017	+ /* page can be truncated */
	1018	+ goto release_and_retry;
	1019	+ }
907	1020
908	1021	f2fs_wait_on_page_writeback(page, DATA, true, true);
909		-
910	1022	f2fs_compress_ctx_add_page(cc, page);
911		- f2fs_put_page(page, 0);
912	1023
913	1024	if (!PageUptodate(page)) {
	1025	+release_and_retry:
	1026	+ f2fs_put_rpages(cc);
914	1027	f2fs_unlock_rpages(cc, i + 1);
915		- f2fs_put_rpages_mapping(mapping, start_idx,
916		- cc->cluster_size);
917		- f2fs_destroy_compress_ctx(cc);
	1028	+ f2fs_destroy_compress_ctx(cc, true);
918	1029	goto retry;
919	1030	}
920		- }
921		-
922		- if (prealloc) {
923		- __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
924		-
925		- set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
926		-
927		- for (i = cc->cluster_size - 1; i > 0; i--) {
928		- ret = f2fs_get_block(&dn, start_idx + i);
929		- if (ret) {
930		- i = cc->cluster_size;
931		- break;
932		- }
933		-
934		- if (dn.data_blkaddr != NEW_ADDR)
935		- break;
936		- }
937		-
938		- __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
939	1031	}
940	1032
941	1033	if (likely(!ret)) {
..	..	@@ -945,10 +1037,10 @@
945	1037	}
946	1038
947	1039	unlock_pages:
	1040	+ f2fs_put_rpages(cc);
948	1041	f2fs_unlock_rpages(cc, i);
949		-release_pages:
950		- f2fs_put_rpages_mapping(mapping, start_idx, i);
951		- f2fs_destroy_compress_ctx(cc);
	1042	+ f2fs_destroy_compress_ctx(cc, true);
	1043	+out:
952	1044	return ret;
953	1045	}
954	1046
..	..	@@ -972,6 +1064,7 @@
972	1064
973	1065	{
974	1066	struct compress_ctx cc = {
	1067	+ .inode = inode,
975	1068	.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
976	1069	.cluster_size = F2FS_I(inode)->i_cluster_size,
977	1070	.rpages = fsdata,
..	..	@@ -982,7 +1075,7 @@
982	1075	set_cluster_dirty(&cc);
983	1076
984	1077	f2fs_put_rpages_wbc(&cc, NULL, false, 1);
985		- f2fs_destroy_compress_ctx(&cc);
	1078	+ f2fs_destroy_compress_ctx(&cc, false);
986	1079
987	1080	return first_index;
988	1081	}
..	..	@@ -1057,7 +1150,7 @@
1057	1150	.submitted = false,
1058	1151	.io_type = io_type,
1059	1152	.io_wbc = wbc,
1060		- .encrypted = f2fs_encrypted_file(cc->inode),
	1153	+ .encrypted = fscrypt_inode_uses_fs_layer_crypto(cc->inode),
1061	1154	};
1062	1155	struct dnode_of_data dn;
1063	1156	struct node_info ni;
..	..	@@ -1067,8 +1160,22 @@
1067	1160	loff_t psize;
1068	1161	int i, err;
1069	1162
1070		- if (!IS_NOQUOTA(inode) && !f2fs_trylock_op(sbi))
1071		- return -EAGAIN;
	1163	+ /* we should bypass data pages to proceed the kworkder jobs */
	1164	+ if (unlikely(f2fs_cp_error(sbi))) {
	1165	+ mapping_set_error(cc->rpages[0]->mapping, -EIO);
	1166	+ goto out_free;
	1167	+ }
	1168	+
	1169	+ if (IS_NOQUOTA(inode)) {
	1170	+ /*
	1171	+ * We need to wait for node_write to avoid block allocation during
	1172	+ * checkpoint. This can only happen to quota writes which can cause
	1173	+ * the below discard race condition.
	1174	+ */
	1175	+ f2fs_down_read(&sbi->node_write);
	1176	+ } else if (!f2fs_trylock_op(sbi)) {
	1177	+ goto out_free;
	1178	+ }
1072	1179
1073	1180	set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1074	1181
..	..	@@ -1084,21 +1191,20 @@
1084	1191
1085	1192	psize = (loff_t)(cc->rpages[last_index]->index + 1) << PAGE_SHIFT;
1086	1193
1087		- err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
	1194	+ err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1088	1195	if (err)
1089	1196	goto out_put_dnode;
1090	1197
1091	1198	fio.version = ni.version;
1092	1199
1093		- cic = f2fs_kzalloc(sbi, sizeof(struct compress_io_ctx), GFP_NOFS);
	1200	+ cic = kmem_cache_zalloc(cic_entry_slab, GFP_NOFS);
1094	1201	if (!cic)
1095	1202	goto out_put_dnode;
1096	1203
1097	1204	cic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1098	1205	cic->inode = inode;
1099		- refcount_set(&cic->ref, cc->nr_cpages);
1100		- cic->rpages = f2fs_kzalloc(sbi, sizeof(struct page *) <<
1101		- cc->log_cluster_size, GFP_NOFS);
	1206	+ atomic_set(&cic->pending_pages, cc->nr_cpages);
	1207	+ cic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1102	1208	if (!cic->rpages)
1103	1209	goto out_put_cic;
1104	1210
..	..	@@ -1108,13 +1214,19 @@
1108	1214	f2fs_set_compressed_page(cc->cpages[i], inode,
1109	1215	cc->rpages[i + 1]->index, cic);
1110	1216	fio.compressed_page = cc->cpages[i];
	1217	+
	1218	+ fio.old_blkaddr = data_blkaddr(dn.inode, dn.node_page,
	1219	+ dn.ofs_in_node + i + 1);
	1220	+
	1221	+ /* wait for GCed page writeback via META_MAPPING */
	1222	+ f2fs_wait_on_block_writeback(inode, fio.old_blkaddr);
	1223	+
1111	1224	if (fio.encrypted) {
1112	1225	fio.page = cc->rpages[i + 1];
1113	1226	err = f2fs_encrypt_one_page(&fio);
1114	1227	if (err)
1115	1228	goto out_destroy_crypt;
1116		- if (fscrypt_inode_uses_fs_layer_crypto(inode))
1117		- cc->cpages[i] = fio.encrypted_page;
	1229	+ cc->cpages[i] = fio.encrypted_page;
1118	1230	}
1119	1231	}
1120	1232
..	..	@@ -1153,7 +1265,7 @@
1153	1265
1154	1266	f2fs_bug_on(fio.sbi, blkaddr == NULL_ADDR);
1155	1267
1156		- if (fio.encrypted && fscrypt_inode_uses_fs_layer_crypto(inode))
	1268	+ if (fio.encrypted)
1157	1269	fio.encrypted_page = cc->cpages[i - 1];
1158	1270	else
1159	1271	fio.compressed_page = cc->cpages[i - 1];
..	..	@@ -1169,13 +1281,16 @@
1169	1281	if (fio.compr_blocks)
1170	1282	f2fs_i_compr_blocks_update(inode, fio.compr_blocks - 1, false);
1171	1283	f2fs_i_compr_blocks_update(inode, cc->nr_cpages, true);
	1284	+ add_compr_block_stat(inode, cc->nr_cpages);
1172	1285
1173	1286	set_inode_flag(cc->inode, FI_APPEND_WRITE);
1174	1287	if (cc->cluster_idx == 0)
1175	1288	set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1176	1289
1177	1290	f2fs_put_dnode(&dn);
1178		- if (!IS_NOQUOTA(inode))
	1291	+ if (IS_NOQUOTA(inode))
	1292	+ f2fs_up_read(&sbi->node_write);
	1293	+ else
1179	1294	f2fs_unlock_op(sbi);
1180	1295
1181	1296	spin_lock(&fi->i_size_lock);
..	..	@@ -1184,26 +1299,34 @@
1184	1299	spin_unlock(&fi->i_size_lock);
1185	1300
1186	1301	f2fs_put_rpages(cc);
1187		- f2fs_destroy_compress_ctx(cc);
	1302	+ page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
	1303	+ cc->cpages = NULL;
	1304	+ f2fs_destroy_compress_ctx(cc, false);
1188	1305	return 0;
1189	1306
1190	1307	out_destroy_crypt:
1191		- kfree(cic->rpages);
	1308	+ page_array_free(cc->inode, cic->rpages, cc->cluster_size);
1192	1309
1193	1310	for (--i; i >= 0; i--)
1194	1311	fscrypt_finalize_bounce_page(&cc->cpages[i]);
1195		- for (i = 0; i < cc->nr_cpages; i++) {
1196		- if (!cc->cpages[i])
1197		- continue;
1198		- f2fs_put_page(cc->cpages[i], 1);
1199		- }
1200	1312	out_put_cic:
1201		- kfree(cic);
	1313	+ kmem_cache_free(cic_entry_slab, cic);
1202	1314	out_put_dnode:
1203	1315	f2fs_put_dnode(&dn);
1204	1316	out_unlock_op:
1205		- if (!IS_NOQUOTA(inode))
	1317	+ if (IS_NOQUOTA(inode))
	1318	+ f2fs_up_read(&sbi->node_write);
	1319	+ else
1206	1320	f2fs_unlock_op(sbi);
	1321	+out_free:
	1322	+ for (i = 0; i < cc->nr_cpages; i++) {
	1323	+ if (!cc->cpages[i])
	1324	+ continue;
	1325	+ f2fs_compress_free_page(cc->cpages[i]);
	1326	+ cc->cpages[i] = NULL;
	1327	+ }
	1328	+ page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
	1329	+ cc->cpages = NULL;
1207	1330	return -EAGAIN;
1208	1331	}
1209	1332
..	..	@@ -1221,17 +1344,17 @@
1221	1344
1222	1345	dec_page_count(sbi, F2FS_WB_DATA);
1223	1346
1224		- if (refcount_dec_not_one(&cic->ref))
	1347	+ if (atomic_dec_return(&cic->pending_pages))
1225	1348	return;
1226	1349
1227	1350	for (i = 0; i < cic->nr_rpages; i++) {
1228	1351	WARN_ON(!cic->rpages[i]);
1229		- clear_cold_data(cic->rpages[i]);
	1352	+ clear_page_private_gcing(cic->rpages[i]);
1230	1353	end_page_writeback(cic->rpages[i]);
1231	1354	}
1232	1355
1233		- kfree(cic->rpages);
1234		- kfree(cic);
	1356	+ page_array_free(cic->inode, cic->rpages, cic->nr_rpages);
	1357	+ kmem_cache_free(cic_entry_slab, cic);
1235	1358	}
1236	1359
1237	1360	static int f2fs_write_raw_pages(struct compress_ctx *cc,
..	..	@@ -1240,29 +1363,42 @@
1240	1363	enum iostat_type io_type)
1241	1364	{
1242	1365	struct address_space *mapping = cc->inode->i_mapping;
1243		- int _submitted, compr_blocks, ret;
1244		- int i = -1, err = 0;
	1366	+ int _submitted, compr_blocks, ret, i;
1245	1367
1246	1368	compr_blocks = f2fs_compressed_blocks(cc);
1247		- if (compr_blocks < 0) {
1248		- err = compr_blocks;
1249		- goto out_err;
	1369	+
	1370	+ for (i = 0; i < cc->cluster_size; i++) {
	1371	+ if (!cc->rpages[i])
	1372	+ continue;
	1373	+
	1374	+ redirty_page_for_writepage(wbc, cc->rpages[i]);
	1375	+ unlock_page(cc->rpages[i]);
1250	1376	}
	1377	+
	1378	+ if (compr_blocks < 0)
	1379	+ return compr_blocks;
1251	1380
1252	1381	for (i = 0; i < cc->cluster_size; i++) {
1253	1382	if (!cc->rpages[i])
1254	1383	continue;
1255	1384	retry_write:
	1385	+ lock_page(cc->rpages[i]);
	1386	+
1256	1387	if (cc->rpages[i]->mapping != mapping) {
	1388	+continue_unlock:
1257	1389	unlock_page(cc->rpages[i]);
1258	1390	continue;
1259	1391	}
1260	1392
1261		- BUG_ON(!PageLocked(cc->rpages[i]));
	1393	+ if (!PageDirty(cc->rpages[i]))
	1394	+ goto continue_unlock;
	1395	+
	1396	+ if (!clear_page_dirty_for_io(cc->rpages[i]))
	1397	+ goto continue_unlock;
1262	1398
1263	1399	ret = f2fs_write_single_data_page(cc->rpages[i], &_submitted,
1264	1400	NULL, NULL, wbc, io_type,
1265		- compr_blocks);
	1401	+ compr_blocks, false);
1266	1402	if (ret) {
1267	1403	if (ret == AOP_WRITEPAGE_ACTIVATE) {
1268	1404	unlock_page(cc->rpages[i]);
..	..	@@ -1273,33 +1409,23 @@
1273	1409	* avoid deadlock caused by cluster update race
1274	1410	* from foreground operation.
1275	1411	*/
1276		- if (IS_NOQUOTA(cc->inode)) {
1277		- err = 0;
1278		- goto out_err;
1279		- }
	1412	+ if (IS_NOQUOTA(cc->inode))
	1413	+ return 0;
1280	1414	ret = 0;
1281	1415	cond_resched();
1282	1416	congestion_wait(BLK_RW_ASYNC,
1283	1417	DEFAULT_IO_TIMEOUT);
1284		- lock_page(cc->rpages[i]);
1285		- clear_page_dirty_for_io(cc->rpages[i]);
1286	1418	goto retry_write;
1287	1419	}
1288		- err = ret;
1289		- goto out_err;
	1420	+ return ret;
1290	1421	}
1291	1422
1292	1423	*submitted += _submitted;
1293	1424	}
	1425	+
	1426	+ f2fs_balance_fs(F2FS_M_SB(mapping), true);
	1427	+
1294	1428	return 0;
1295		-out_err:
1296		- for (++i; i < cc->cluster_size; i++) {
1297		- if (!cc->rpages[i])
1298		- continue;
1299		- redirty_page_for_writepage(wbc, cc->rpages[i]);
1300		- unlock_page(cc->rpages[i]);
1301		- }
1302		- return err;
1303	1429	}
1304	1430
1305	1431	int f2fs_write_multi_pages(struct compress_ctx *cc,
..	..	@@ -1307,15 +1433,13 @@
1307	1433	struct writeback_control *wbc,
1308	1434	enum iostat_type io_type)
1309	1435	{
1310		- struct f2fs_inode_info *fi = F2FS_I(cc->inode);
1311		- const struct f2fs_compress_ops *cops =
1312		- f2fs_cops[fi->i_compress_algorithm];
1313	1436	int err;
1314	1437
1315	1438	*submitted = 0;
1316	1439	if (cluster_may_compress(cc)) {
1317	1440	err = f2fs_compress_pages(cc);
1318	1441	if (err == -EAGAIN) {
	1442	+ add_compr_block_stat(cc->inode, cc->cluster_size);
1319	1443	goto write;
1320	1444	} else if (err) {
1321	1445	f2fs_put_rpages_wbc(cc, wbc, true, 1);
..	..	@@ -1324,7 +1448,6 @@
1324	1448
1325	1449	err = f2fs_write_compressed_pages(cc, submitted,
1326	1450	wbc, io_type);
1327		- cops->destroy_compress_ctx(cc);
1328	1451	if (!err)
1329	1452	return 0;
1330	1453	f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN);
..	..	@@ -1335,84 +1458,149 @@
1335	1458	err = f2fs_write_raw_pages(cc, submitted, wbc, io_type);
1336	1459	f2fs_put_rpages_wbc(cc, wbc, false, 0);
1337	1460	destroy_out:
1338		- f2fs_destroy_compress_ctx(cc);
	1461	+ f2fs_destroy_compress_ctx(cc, false);
1339	1462	return err;
1340	1463	}
1341	1464
1342		-struct decompress_io_ctx f2fs_alloc_dic(struct compress_ctx cc)
	1465	+static inline bool allow_memalloc_for_decomp(struct f2fs_sb_info *sbi,
	1466	+ bool pre_alloc)
1343	1467	{
1344		- struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
1345		- struct decompress_io_ctx *dic;
1346		- pgoff_t start_idx = start_idx_of_cluster(cc);
	1468	+ return pre_alloc ^ f2fs_low_mem_mode(sbi);
	1469	+}
	1470	+
	1471	+static int f2fs_prepare_decomp_mem(struct decompress_io_ctx *dic,
	1472	+ bool pre_alloc)
	1473	+{
	1474	+ const struct f2fs_compress_ops *cops =
	1475	+ f2fs_cops[F2FS_I(dic->inode)->i_compress_algorithm];
1347	1476	int i;
1348	1477
1349		- dic = f2fs_kzalloc(sbi, sizeof(struct decompress_io_ctx), GFP_NOFS);
	1478	+ if (!allow_memalloc_for_decomp(F2FS_I_SB(dic->inode), pre_alloc))
	1479	+ return 0;
	1480	+
	1481	+ dic->tpages = page_array_alloc(dic->inode, dic->cluster_size);
	1482	+ if (!dic->tpages)
	1483	+ return -ENOMEM;
	1484	+
	1485	+ for (i = 0; i < dic->cluster_size; i++) {
	1486	+ if (dic->rpages[i]) {
	1487	+ dic->tpages[i] = dic->rpages[i];
	1488	+ continue;
	1489	+ }
	1490	+
	1491	+ dic->tpages[i] = f2fs_compress_alloc_page();
	1492	+ if (!dic->tpages[i])
	1493	+ return -ENOMEM;
	1494	+ }
	1495	+
	1496	+ dic->rbuf = f2fs_vmap(dic->tpages, dic->cluster_size);
	1497	+ if (!dic->rbuf)
	1498	+ return -ENOMEM;
	1499	+
	1500	+ dic->cbuf = f2fs_vmap(dic->cpages, dic->nr_cpages);
	1501	+ if (!dic->cbuf)
	1502	+ return -ENOMEM;
	1503	+
	1504	+ if (cops->init_decompress_ctx) {
	1505	+ int ret = cops->init_decompress_ctx(dic);
	1506	+
	1507	+ if (ret)
	1508	+ return ret;
	1509	+ }
	1510	+
	1511	+ return 0;
	1512	+}
	1513	+
	1514	+static void f2fs_release_decomp_mem(struct decompress_io_ctx *dic,
	1515	+ bool bypass_destroy_callback, bool pre_alloc)
	1516	+{
	1517	+ const struct f2fs_compress_ops *cops =
	1518	+ f2fs_cops[F2FS_I(dic->inode)->i_compress_algorithm];
	1519	+
	1520	+ if (!allow_memalloc_for_decomp(F2FS_I_SB(dic->inode), pre_alloc))
	1521	+ return;
	1522	+
	1523	+ if (!bypass_destroy_callback && cops->destroy_decompress_ctx)
	1524	+ cops->destroy_decompress_ctx(dic);
	1525	+
	1526	+ if (dic->cbuf)
	1527	+ vm_unmap_ram(dic->cbuf, dic->nr_cpages);
	1528	+
	1529	+ if (dic->rbuf)
	1530	+ vm_unmap_ram(dic->rbuf, dic->cluster_size);
	1531	+}
	1532	+
	1533	+static void f2fs_free_dic(struct decompress_io_ctx *dic,
	1534	+ bool bypass_destroy_callback);
	1535	+
	1536	+struct decompress_io_ctx f2fs_alloc_dic(struct compress_ctx cc)
	1537	+{
	1538	+ struct decompress_io_ctx *dic;
	1539	+ pgoff_t start_idx = start_idx_of_cluster(cc);
	1540	+ int i, ret;
	1541	+
	1542	+ dic = kmem_cache_zalloc(dic_entry_slab, GFP_NOFS);
1350	1543	if (!dic)
1351	1544	return ERR_PTR(-ENOMEM);
1352	1545
1353		- dic->rpages = f2fs_kzalloc(sbi, sizeof(struct page *) <<
1354		- cc->log_cluster_size, GFP_NOFS);
	1546	+ dic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1355	1547	if (!dic->rpages) {
1356		- kfree(dic);
	1548	+ kmem_cache_free(dic_entry_slab, dic);
1357	1549	return ERR_PTR(-ENOMEM);
1358	1550	}
1359	1551
1360	1552	dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1361	1553	dic->inode = cc->inode;
1362		- refcount_set(&dic->ref, cc->nr_cpages);
	1554	+ atomic_set(&dic->remaining_pages, cc->nr_cpages);
1363	1555	dic->cluster_idx = cc->cluster_idx;
1364	1556	dic->cluster_size = cc->cluster_size;
1365	1557	dic->log_cluster_size = cc->log_cluster_size;
1366	1558	dic->nr_cpages = cc->nr_cpages;
	1559	+ refcount_set(&dic->refcnt, 1);
1367	1560	dic->failed = false;
	1561	+ dic->need_verity = f2fs_need_verity(cc->inode, start_idx);
1368	1562
1369	1563	for (i = 0; i < dic->cluster_size; i++)
1370	1564	dic->rpages[i] = cc->rpages[i];
1371	1565	dic->nr_rpages = cc->cluster_size;
1372	1566
1373		- dic->cpages = f2fs_kzalloc(sbi, sizeof(struct page )
1374		- dic->nr_cpages, GFP_NOFS);
1375		- if (!dic->cpages)
	1567	+ dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages);
	1568	+ if (!dic->cpages) {
	1569	+ ret = -ENOMEM;
1376	1570	goto out_free;
	1571	+ }
1377	1572
1378	1573	for (i = 0; i < dic->nr_cpages; i++) {
1379	1574	struct page *page;
1380	1575
1381	1576	page = f2fs_compress_alloc_page();
1382		- if (!page)
	1577	+ if (!page) {
	1578	+ ret = -ENOMEM;
1383	1579	goto out_free;
	1580	+ }
1384	1581
1385	1582	f2fs_set_compressed_page(page, cc->inode,
1386	1583	start_idx + i + 1, dic);
1387	1584	dic->cpages[i] = page;
1388	1585	}
1389	1586
1390		- dic->tpages = f2fs_kzalloc(sbi, sizeof(struct page )
1391		- dic->cluster_size, GFP_NOFS);
1392		- if (!dic->tpages)
	1587	+ ret = f2fs_prepare_decomp_mem(dic, true);
	1588	+ if (ret)
1393	1589	goto out_free;
1394		-
1395		- for (i = 0; i < dic->cluster_size; i++) {
1396		- if (cc->rpages[i]) {
1397		- dic->tpages[i] = cc->rpages[i];
1398		- continue;
1399		- }
1400		-
1401		- dic->tpages[i] = f2fs_compress_alloc_page();
1402		- if (!dic->tpages[i])
1403		- goto out_free;
1404		- }
1405	1590
1406	1591	return dic;
1407	1592
1408	1593	out_free:
1409		- f2fs_free_dic(dic);
1410		- return ERR_PTR(-ENOMEM);
	1594	+ f2fs_free_dic(dic, true);
	1595	+ return ERR_PTR(ret);
1411	1596	}
1412	1597
1413		-void f2fs_free_dic(struct decompress_io_ctx *dic)
	1598	+static void f2fs_free_dic(struct decompress_io_ctx *dic,
	1599	+ bool bypass_destroy_callback)
1414	1600	{
1415	1601	int i;
	1602	+
	1603	+ f2fs_release_decomp_mem(dic, bypass_destroy_callback, true);
1416	1604
1417	1605	if (dic->tpages) {
1418	1606	for (i = 0; i < dic->cluster_size; i++) {
..	..	@@ -1422,7 +1610,7 @@
1422	1610	continue;
1423	1611	f2fs_compress_free_page(dic->tpages[i]);
1424	1612	}
1425		- kfree(dic->tpages);
	1613	+ page_array_free(dic->inode, dic->tpages, dic->cluster_size);
1426	1614	}
1427	1615
1428	1616	if (dic->cpages) {
..	..	@@ -1431,35 +1619,365 @@
1431	1619	continue;
1432	1620	f2fs_compress_free_page(dic->cpages[i]);
1433	1621	}
1434		- kfree(dic->cpages);
	1622	+ page_array_free(dic->inode, dic->cpages, dic->nr_cpages);
1435	1623	}
1436	1624
1437		- kfree(dic->rpages);
1438		- kfree(dic);
	1625	+ page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
	1626	+ kmem_cache_free(dic_entry_slab, dic);
1439	1627	}
1440	1628
1441		-void f2fs_decompress_end_io(struct page **rpages,
1442		- unsigned int cluster_size, bool err, bool verity)
	1629	+static void f2fs_late_free_dic(struct work_struct *work)
	1630	+{
	1631	+ struct decompress_io_ctx *dic =
	1632	+ container_of(work, struct decompress_io_ctx, free_work);
	1633	+
	1634	+ f2fs_free_dic(dic, false);
	1635	+}
	1636	+
	1637	+static void f2fs_put_dic(struct decompress_io_ctx *dic, bool in_task)
	1638	+{
	1639	+ if (refcount_dec_and_test(&dic->refcnt)) {
	1640	+ if (in_task) {
	1641	+ f2fs_free_dic(dic, false);
	1642	+ } else {
	1643	+ INIT_WORK(&dic->free_work, f2fs_late_free_dic);
	1644	+ queue_work(F2FS_I_SB(dic->inode)->post_read_wq,
	1645	+ &dic->free_work);
	1646	+ }
	1647	+ }
	1648	+}
	1649	+
	1650	+/*
	1651	+ * Update and unlock the cluster's pagecache pages, and release the reference to
	1652	+ * the decompress_io_ctx that was being held for I/O completion.
	1653	+ */
	1654	+static void __f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed,
	1655	+ bool in_task)
1443	1656	{
1444	1657	int i;
1445	1658
1446		- for (i = 0; i < cluster_size; i++) {
1447		- struct page *rpage = rpages[i];
	1659	+ for (i = 0; i < dic->cluster_size; i++) {
	1660	+ struct page *rpage = dic->rpages[i];
1448	1661
1449	1662	if (!rpage)
1450	1663	continue;
1451	1664
1452		- if (err \|\| PageError(rpage))
1453		- goto clear_uptodate;
1454		-
1455		- if (!verity \|\| fsverity_verify_page(rpage)) {
	1665	+ /* PG_error was set if verity failed. */
	1666	+ if (failed \|\| PageError(rpage)) {
	1667	+ ClearPageUptodate(rpage);
	1668	+ /* will re-read again later */
	1669	+ ClearPageError(rpage);
	1670	+ } else {
1456	1671	SetPageUptodate(rpage);
1457		- goto unlock;
1458	1672	}
1459		-clear_uptodate:
1460		- ClearPageUptodate(rpage);
1461		- ClearPageError(rpage);
1462		-unlock:
1463	1673	unlock_page(rpage);
1464	1674	}
	1675	+
	1676	+ f2fs_put_dic(dic, in_task);
	1677	+}
	1678	+
	1679	+static void f2fs_verify_cluster(struct work_struct *work)
	1680	+{
	1681	+ struct decompress_io_ctx *dic =
	1682	+ container_of(work, struct decompress_io_ctx, verity_work);
	1683	+ int i;
	1684	+
	1685	+ /* Verify the cluster's decompressed pages with fs-verity. */
	1686	+ for (i = 0; i < dic->cluster_size; i++) {
	1687	+ struct page *rpage = dic->rpages[i];
	1688	+
	1689	+ if (rpage && !fsverity_verify_page(rpage))
	1690	+ SetPageError(rpage);
	1691	+ }
	1692	+
	1693	+ __f2fs_decompress_end_io(dic, false, true);
	1694	+}
	1695	+
	1696	+/*
	1697	+ * This is called when a compressed cluster has been decompressed
	1698	+ * (or failed to be read and/or decompressed).
	1699	+ */
	1700	+void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed,
	1701	+ bool in_task)
	1702	+{
	1703	+ if (!failed && dic->need_verity) {
	1704	+ /*
	1705	+ * Note that to avoid deadlocks, the verity work can't be done
	1706	+ * on the decompression workqueue. This is because verifying
	1707	+ * the data pages can involve reading metadata pages from the
	1708	+ * file, and these metadata pages may be compressed.
	1709	+ */
	1710	+ INIT_WORK(&dic->verity_work, f2fs_verify_cluster);
	1711	+ fsverity_enqueue_verify_work(&dic->verity_work);
	1712	+ } else {
	1713	+ __f2fs_decompress_end_io(dic, failed, in_task);
	1714	+ }
	1715	+}
	1716	+
	1717	+/*
	1718	+ * Put a reference to a compressed page's decompress_io_ctx.
	1719	+ *
	1720	+ * This is called when the page is no longer needed and can be freed.
	1721	+ */
	1722	+void f2fs_put_page_dic(struct page *page, bool in_task)
	1723	+{
	1724	+ struct decompress_io_ctx *dic =
	1725	+ (struct decompress_io_ctx *)page_private(page);
	1726	+
	1727	+ f2fs_put_dic(dic, in_task);
	1728	+}
	1729	+
	1730	+/*
	1731	+ * check whether cluster blocks are contiguous, and add extent cache entry
	1732	+ * only if cluster blocks are logically and physically contiguous.
	1733	+ */
	1734	+unsigned int f2fs_cluster_blocks_are_contiguous(struct dnode_of_data *dn)
	1735	+{
	1736	+ bool compressed = f2fs_data_blkaddr(dn) == COMPRESS_ADDR;
	1737	+ int i = compressed ? 1 : 0;
	1738	+ block_t first_blkaddr = data_blkaddr(dn->inode, dn->node_page,
	1739	+ dn->ofs_in_node + i);
	1740	+
	1741	+ for (i += 1; i < F2FS_I(dn->inode)->i_cluster_size; i++) {
	1742	+ block_t blkaddr = data_blkaddr(dn->inode, dn->node_page,
	1743	+ dn->ofs_in_node + i);
	1744	+
	1745	+ if (!__is_valid_data_blkaddr(blkaddr))
	1746	+ break;
	1747	+ if (first_blkaddr + i - (compressed ? 1 : 0) != blkaddr)
	1748	+ return 0;
	1749	+ }
	1750	+
	1751	+ return compressed ? i - 1 : i;
	1752	+}
	1753	+
	1754	+const struct address_space_operations f2fs_compress_aops = {
	1755	+ .releasepage = f2fs_release_page,
	1756	+ .invalidatepage = f2fs_invalidate_page,
	1757	+};
	1758	+
	1759	+struct address_space COMPRESS_MAPPING(struct f2fs_sb_info sbi)
	1760	+{
	1761	+ return sbi->compress_inode->i_mapping;
	1762	+}
	1763	+
	1764	+void f2fs_invalidate_compress_page(struct f2fs_sb_info *sbi, block_t blkaddr)
	1765	+{
	1766	+ if (!sbi->compress_inode)
	1767	+ return;
	1768	+ invalidate_mapping_pages(COMPRESS_MAPPING(sbi), blkaddr, blkaddr);
	1769	+}
	1770	+
	1771	+void f2fs_cache_compressed_page(struct f2fs_sb_info sbi, struct page page,
	1772	+ nid_t ino, block_t blkaddr)
	1773	+{
	1774	+ struct page *cpage;
	1775	+ int ret;
	1776	+
	1777	+ if (!test_opt(sbi, COMPRESS_CACHE))
	1778	+ return;
	1779	+
	1780	+ if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
	1781	+ return;
	1782	+
	1783	+ if (!f2fs_available_free_memory(sbi, COMPRESS_PAGE))
	1784	+ return;
	1785	+
	1786	+ cpage = find_get_page(COMPRESS_MAPPING(sbi), blkaddr);
	1787	+ if (cpage) {
	1788	+ f2fs_put_page(cpage, 0);
	1789	+ return;
	1790	+ }
	1791	+
	1792	+ cpage = alloc_page(__GFP_NOWARN \| __GFP_IO);
	1793	+ if (!cpage)
	1794	+ return;
	1795	+
	1796	+ ret = add_to_page_cache_lru(cpage, COMPRESS_MAPPING(sbi),
	1797	+ blkaddr, GFP_NOFS);
	1798	+ if (ret) {
	1799	+ f2fs_put_page(cpage, 0);
	1800	+ return;
	1801	+ }
	1802	+
	1803	+ set_page_private_data(cpage, ino);
	1804	+
	1805	+ if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
	1806	+ goto out;
	1807	+
	1808	+ memcpy(page_address(cpage), page_address(page), PAGE_SIZE);
	1809	+ SetPageUptodate(cpage);
	1810	+out:
	1811	+ f2fs_put_page(cpage, 1);
	1812	+}
	1813	+
	1814	+bool f2fs_load_compressed_page(struct f2fs_sb_info sbi, struct page page,
	1815	+ block_t blkaddr)
	1816	+{
	1817	+ struct page *cpage;
	1818	+ bool hitted = false;
	1819	+
	1820	+ if (!test_opt(sbi, COMPRESS_CACHE))
	1821	+ return false;
	1822	+
	1823	+ cpage = f2fs_pagecache_get_page(COMPRESS_MAPPING(sbi),
	1824	+ blkaddr, FGP_LOCK \| FGP_NOWAIT, GFP_NOFS);
	1825	+ if (cpage) {
	1826	+ if (PageUptodate(cpage)) {
	1827	+ atomic_inc(&sbi->compress_page_hit);
	1828	+ memcpy(page_address(page),
	1829	+ page_address(cpage), PAGE_SIZE);
	1830	+ hitted = true;
	1831	+ }
	1832	+ f2fs_put_page(cpage, 1);
	1833	+ }
	1834	+
	1835	+ return hitted;
	1836	+}
	1837	+
	1838	+void f2fs_invalidate_compress_pages(struct f2fs_sb_info *sbi, nid_t ino)
	1839	+{
	1840	+ struct address_space *mapping = sbi->compress_inode->i_mapping;
	1841	+ struct pagevec pvec;
	1842	+ pgoff_t index = 0;
	1843	+ pgoff_t end = MAX_BLKADDR(sbi);
	1844	+
	1845	+ if (!mapping->nrpages)
	1846	+ return;
	1847	+
	1848	+ pagevec_init(&pvec);
	1849	+
	1850	+ do {
	1851	+ unsigned int nr_pages;
	1852	+ int i;
	1853	+
	1854	+ nr_pages = pagevec_lookup_range(&pvec, mapping,
	1855	+ &index, end - 1);
	1856	+ if (!nr_pages)
	1857	+ break;
	1858	+
	1859	+ for (i = 0; i < nr_pages; i++) {
	1860	+ struct page *page = pvec.pages[i];
	1861	+
	1862	+ if (page->index > end)
	1863	+ break;
	1864	+
	1865	+ lock_page(page);
	1866	+ if (page->mapping != mapping) {
	1867	+ unlock_page(page);
	1868	+ continue;
	1869	+ }
	1870	+
	1871	+ if (ino != get_page_private_data(page)) {
	1872	+ unlock_page(page);
	1873	+ continue;
	1874	+ }
	1875	+
	1876	+ generic_error_remove_page(mapping, page);
	1877	+ unlock_page(page);
	1878	+ }
	1879	+ pagevec_release(&pvec);
	1880	+ cond_resched();
	1881	+ } while (index < end);
	1882	+}
	1883	+
	1884	+int f2fs_init_compress_inode(struct f2fs_sb_info *sbi)
	1885	+{
	1886	+ struct inode *inode;
	1887	+
	1888	+ if (!test_opt(sbi, COMPRESS_CACHE))
	1889	+ return 0;
	1890	+
	1891	+ inode = f2fs_iget(sbi->sb, F2FS_COMPRESS_INO(sbi));
	1892	+ if (IS_ERR(inode))
	1893	+ return PTR_ERR(inode);
	1894	+ sbi->compress_inode = inode;
	1895	+
	1896	+ sbi->compress_percent = COMPRESS_PERCENT;
	1897	+ sbi->compress_watermark = COMPRESS_WATERMARK;
	1898	+
	1899	+ atomic_set(&sbi->compress_page_hit, 0);
	1900	+
	1901	+ return 0;
	1902	+}
	1903	+
	1904	+void f2fs_destroy_compress_inode(struct f2fs_sb_info *sbi)
	1905	+{
	1906	+ if (!sbi->compress_inode)
	1907	+ return;
	1908	+ iput(sbi->compress_inode);
	1909	+ sbi->compress_inode = NULL;
	1910	+}
	1911	+
	1912	+int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi)
	1913	+{
	1914	+ dev_t dev = sbi->sb->s_bdev->bd_dev;
	1915	+ char slab_name[32];
	1916	+
	1917	+ sprintf(slab_name, "f2fs_page_array_entry-%u:%u", MAJOR(dev), MINOR(dev));
	1918	+
	1919	+ sbi->page_array_slab_size = sizeof(struct page *) <<
	1920	+ F2FS_OPTION(sbi).compress_log_size;
	1921	+
	1922	+ sbi->page_array_slab = f2fs_kmem_cache_create(slab_name,
	1923	+ sbi->page_array_slab_size);
	1924	+ if (!sbi->page_array_slab)
	1925	+ return -ENOMEM;
	1926	+ return 0;
	1927	+}
	1928	+
	1929	+void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi)
	1930	+{
	1931	+ kmem_cache_destroy(sbi->page_array_slab);
	1932	+}
	1933	+
	1934	+static int __init f2fs_init_cic_cache(void)
	1935	+{
	1936	+ cic_entry_slab = f2fs_kmem_cache_create("f2fs_cic_entry",
	1937	+ sizeof(struct compress_io_ctx));
	1938	+ if (!cic_entry_slab)
	1939	+ return -ENOMEM;
	1940	+ return 0;
	1941	+}
	1942	+
	1943	+static void f2fs_destroy_cic_cache(void)
	1944	+{
	1945	+ kmem_cache_destroy(cic_entry_slab);
	1946	+}
	1947	+
	1948	+static int __init f2fs_init_dic_cache(void)
	1949	+{
	1950	+ dic_entry_slab = f2fs_kmem_cache_create("f2fs_dic_entry",
	1951	+ sizeof(struct decompress_io_ctx));
	1952	+ if (!dic_entry_slab)
	1953	+ return -ENOMEM;
	1954	+ return 0;
	1955	+}
	1956	+
	1957	+static void f2fs_destroy_dic_cache(void)
	1958	+{
	1959	+ kmem_cache_destroy(dic_entry_slab);
	1960	+}
	1961	+
	1962	+int __init f2fs_init_compress_cache(void)
	1963	+{
	1964	+ int err;
	1965	+
	1966	+ err = f2fs_init_cic_cache();
	1967	+ if (err)
	1968	+ goto out;
	1969	+ err = f2fs_init_dic_cache();
	1970	+ if (err)
	1971	+ goto free_cic;
	1972	+ return 0;
	1973	+free_cic:
	1974	+ f2fs_destroy_cic_cache();
	1975	+out:
	1976	+ return -ENOMEM;
	1977	+}
	1978	+
	1979	+void f2fs_destroy_compress_cache(void)
	1980	+{
	1981	+ f2fs_destroy_dic_cache();
	1982	+ f2fs_destroy_cic_cache();
1465	1983	}