xref: /linux/fs/f2fs/compress.c (revision bb5b94f5bbe75470912b70fb08880fc5273aa62d)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * f2fs compress support
4  *
5  * Copyright (c) 2019 Chao Yu <chao@kernel.org>
6  */
7 
8 #include <linux/fs.h>
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>
15 #include <linux/pagevec.h>
16 
17 #include "f2fs.h"
18 #include "node.h"
19 #include "segment.h"
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 }
48 
49 struct f2fs_compress_ops {
50 	int (*init_compress_ctx)(struct compress_ctx *cc);
51 	void (*destroy_compress_ctx)(struct compress_ctx *cc);
52 	int (*compress_pages)(struct compress_ctx *cc);
53 	int (*init_decompress_ctx)(struct decompress_io_ctx *dic);
54 	void (*destroy_decompress_ctx)(struct decompress_io_ctx *dic);
55 	int (*decompress_pages)(struct decompress_io_ctx *dic);
56 };
57 
58 static unsigned int offset_in_cluster(struct compress_ctx *cc, pgoff_t index)
59 {
60 	return index & (cc->cluster_size - 1);
61 }
62 
63 static pgoff_t cluster_idx(struct compress_ctx *cc, pgoff_t index)
64 {
65 	return index >> cc->log_cluster_size;
66 }
67 
68 static pgoff_t start_idx_of_cluster(struct compress_ctx *cc)
69 {
70 	return cc->cluster_idx << cc->log_cluster_size;
71 }
72 
73 bool f2fs_is_compressed_page(struct page *page)
74 {
75 	if (!PagePrivate(page))
76 		return false;
77 	if (!page_private(page))
78 		return false;
79 	if (page_private_nonpointer(page))
80 		return false;
81 
82 	f2fs_bug_on(F2FS_M_SB(page->mapping),
83 		*((u32 *)page_private(page)) != F2FS_COMPRESSED_PAGE_MAGIC);
84 	return true;
85 }
86 
87 static void f2fs_set_compressed_page(struct page *page,
88 		struct inode *inode, pgoff_t index, void *data)
89 {
90 	attach_page_private(page, (void *)data);
91 
92 	/* i_crypto_info and iv index */
93 	page->index = index;
94 	page->mapping = inode->i_mapping;
95 }
96 
97 static void f2fs_drop_rpages(struct compress_ctx *cc, int len, bool unlock)
98 {
99 	int i;
100 
101 	for (i = 0; i < len; i++) {
102 		if (!cc->rpages[i])
103 			continue;
104 		if (unlock)
105 			unlock_page(cc->rpages[i]);
106 		else
107 			put_page(cc->rpages[i]);
108 	}
109 }
110 
111 static void f2fs_put_rpages(struct compress_ctx *cc)
112 {
113 	f2fs_drop_rpages(cc, cc->cluster_size, false);
114 }
115 
116 static void f2fs_unlock_rpages(struct compress_ctx *cc, int len)
117 {
118 	f2fs_drop_rpages(cc, len, true);
119 }
120 
121 static void f2fs_put_rpages_wbc(struct compress_ctx *cc,
122 		struct writeback_control *wbc, bool redirty, int unlock)
123 {
124 	unsigned int i;
125 
126 	for (i = 0; i < cc->cluster_size; i++) {
127 		if (!cc->rpages[i])
128 			continue;
129 		if (redirty)
130 			redirty_page_for_writepage(wbc, cc->rpages[i]);
131 		f2fs_put_page(cc->rpages[i], unlock);
132 	}
133 }
134 
135 struct page *f2fs_compress_control_page(struct page *page)
136 {
137 	return ((struct compress_io_ctx *)page_private(page))->rpages[0];
138 }
139 
140 int f2fs_init_compress_ctx(struct compress_ctx *cc)
141 {
142 	if (cc->rpages)
143 		return 0;
144 
145 	cc->rpages = page_array_alloc(cc->inode, cc->cluster_size);
146 	return cc->rpages ? 0 : -ENOMEM;
147 }
148 
149 void f2fs_destroy_compress_ctx(struct compress_ctx *cc, bool reuse)
150 {
151 	page_array_free(cc->inode, cc->rpages, cc->cluster_size);
152 	cc->rpages = NULL;
153 	cc->nr_rpages = 0;
154 	cc->nr_cpages = 0;
155 	if (!reuse)
156 		cc->cluster_idx = NULL_CLUSTER;
157 }
158 
159 void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct page *page)
160 {
161 	unsigned int cluster_ofs;
162 
163 	if (!f2fs_cluster_can_merge_page(cc, page->index))
164 		f2fs_bug_on(F2FS_I_SB(cc->inode), 1);
165 
166 	cluster_ofs = offset_in_cluster(cc, page->index);
167 	cc->rpages[cluster_ofs] = page;
168 	cc->nr_rpages++;
169 	cc->cluster_idx = cluster_idx(cc, page->index);
170 }
171 
172 #ifdef CONFIG_F2FS_FS_LZO
173 static int lzo_init_compress_ctx(struct compress_ctx *cc)
174 {
175 	cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
176 				LZO1X_MEM_COMPRESS, GFP_NOFS);
177 	if (!cc->private)
178 		return -ENOMEM;
179 
180 	cc->clen = lzo1x_worst_compress(PAGE_SIZE << cc->log_cluster_size);
181 	return 0;
182 }
183 
184 static void lzo_destroy_compress_ctx(struct compress_ctx *cc)
185 {
186 	kvfree(cc->private);
187 	cc->private = NULL;
188 }
189 
190 static int lzo_compress_pages(struct compress_ctx *cc)
191 {
192 	int ret;
193 
194 	ret = lzo1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
195 					&cc->clen, cc->private);
196 	if (ret != LZO_E_OK) {
197 		printk_ratelimited("%sF2FS-fs (%s): lzo compress failed, ret:%d\n",
198 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
199 		return -EIO;
200 	}
201 	return 0;
202 }
203 
204 static int lzo_decompress_pages(struct decompress_io_ctx *dic)
205 {
206 	int ret;
207 
208 	ret = lzo1x_decompress_safe(dic->cbuf->cdata, dic->clen,
209 						dic->rbuf, &dic->rlen);
210 	if (ret != LZO_E_OK) {
211 		printk_ratelimited("%sF2FS-fs (%s): lzo decompress failed, ret:%d\n",
212 				KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
213 		return -EIO;
214 	}
215 
216 	if (dic->rlen != PAGE_SIZE << dic->log_cluster_size) {
217 		printk_ratelimited("%sF2FS-fs (%s): lzo invalid rlen:%zu, "
218 					"expected:%lu\n", KERN_ERR,
219 					F2FS_I_SB(dic->inode)->sb->s_id,
220 					dic->rlen,
221 					PAGE_SIZE << dic->log_cluster_size);
222 		return -EIO;
223 	}
224 	return 0;
225 }
226 
227 static const struct f2fs_compress_ops f2fs_lzo_ops = {
228 	.init_compress_ctx	= lzo_init_compress_ctx,
229 	.destroy_compress_ctx	= lzo_destroy_compress_ctx,
230 	.compress_pages		= lzo_compress_pages,
231 	.decompress_pages	= lzo_decompress_pages,
232 };
233 #endif
234 
235 #ifdef CONFIG_F2FS_FS_LZ4
236 static int lz4_init_compress_ctx(struct compress_ctx *cc)
237 {
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);
246 	if (!cc->private)
247 		return -ENOMEM;
248 
249 	/*
250 	 * we do not change cc->clen to LZ4_compressBound(inputsize) to
251 	 * adapt worst compress case, because lz4 compressor can handle
252 	 * output budget properly.
253 	 */
254 	cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
255 	return 0;
256 }
257 
258 static void lz4_destroy_compress_ctx(struct compress_ctx *cc)
259 {
260 	kvfree(cc->private);
261 	cc->private = NULL;
262 }
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 
285 static int lz4_compress_pages(struct compress_ctx *cc)
286 {
287 	int len;
288 
289 #ifdef CONFIG_F2FS_FS_LZ4HC
290 	return lz4hc_compress_pages(cc);
291 #endif
292 	len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
293 						cc->clen, cc->private);
294 	if (!len)
295 		return -EAGAIN;
296 
297 	cc->clen = len;
298 	return 0;
299 }
300 
301 static int lz4_decompress_pages(struct decompress_io_ctx *dic)
302 {
303 	int ret;
304 
305 	ret = LZ4_decompress_safe(dic->cbuf->cdata, dic->rbuf,
306 						dic->clen, dic->rlen);
307 	if (ret < 0) {
308 		printk_ratelimited("%sF2FS-fs (%s): lz4 decompress failed, ret:%d\n",
309 				KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
310 		return -EIO;
311 	}
312 
313 	if (ret != PAGE_SIZE << dic->log_cluster_size) {
314 		printk_ratelimited("%sF2FS-fs (%s): lz4 invalid rlen:%zu, "
315 					"expected:%lu\n", KERN_ERR,
316 					F2FS_I_SB(dic->inode)->sb->s_id,
317 					dic->rlen,
318 					PAGE_SIZE << dic->log_cluster_size);
319 		return -EIO;
320 	}
321 	return 0;
322 }
323 
324 static const struct f2fs_compress_ops f2fs_lz4_ops = {
325 	.init_compress_ctx	= lz4_init_compress_ctx,
326 	.destroy_compress_ctx	= lz4_destroy_compress_ctx,
327 	.compress_pages		= lz4_compress_pages,
328 	.decompress_pages	= lz4_decompress_pages,
329 };
330 #endif
331 
332 #ifdef CONFIG_F2FS_FS_ZSTD
333 #define F2FS_ZSTD_DEFAULT_CLEVEL	1
334 
335 static int zstd_init_compress_ctx(struct compress_ctx *cc)
336 {
337 	ZSTD_parameters params;
338 	ZSTD_CStream *stream;
339 	void *workspace;
340 	unsigned int workspace_size;
341 	unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
342 						COMPRESS_LEVEL_OFFSET;
343 
344 	if (!level)
345 		level = F2FS_ZSTD_DEFAULT_CLEVEL;
346 
347 	params = ZSTD_getParams(level, cc->rlen, 0);
348 	workspace_size = ZSTD_CStreamWorkspaceBound(params.cParams);
349 
350 	workspace = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
351 					workspace_size, GFP_NOFS);
352 	if (!workspace)
353 		return -ENOMEM;
354 
355 	stream = ZSTD_initCStream(params, 0, workspace, workspace_size);
356 	if (!stream) {
357 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initCStream failed\n",
358 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
359 				__func__);
360 		kvfree(workspace);
361 		return -EIO;
362 	}
363 
364 	cc->private = workspace;
365 	cc->private2 = stream;
366 
367 	cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
368 	return 0;
369 }
370 
371 static void zstd_destroy_compress_ctx(struct compress_ctx *cc)
372 {
373 	kvfree(cc->private);
374 	cc->private = NULL;
375 	cc->private2 = NULL;
376 }
377 
378 static int zstd_compress_pages(struct compress_ctx *cc)
379 {
380 	ZSTD_CStream *stream = cc->private2;
381 	ZSTD_inBuffer inbuf;
382 	ZSTD_outBuffer outbuf;
383 	int src_size = cc->rlen;
384 	int dst_size = src_size - PAGE_SIZE - COMPRESS_HEADER_SIZE;
385 	int ret;
386 
387 	inbuf.pos = 0;
388 	inbuf.src = cc->rbuf;
389 	inbuf.size = src_size;
390 
391 	outbuf.pos = 0;
392 	outbuf.dst = cc->cbuf->cdata;
393 	outbuf.size = dst_size;
394 
395 	ret = ZSTD_compressStream(stream, &outbuf, &inbuf);
396 	if (ZSTD_isError(ret)) {
397 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
398 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
399 				__func__, ZSTD_getErrorCode(ret));
400 		return -EIO;
401 	}
402 
403 	ret = ZSTD_endStream(stream, &outbuf);
404 	if (ZSTD_isError(ret)) {
405 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_endStream returned %d\n",
406 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
407 				__func__, ZSTD_getErrorCode(ret));
408 		return -EIO;
409 	}
410 
411 	/*
412 	 * there is compressed data remained in intermediate buffer due to
413 	 * no more space in cbuf.cdata
414 	 */
415 	if (ret)
416 		return -EAGAIN;
417 
418 	cc->clen = outbuf.pos;
419 	return 0;
420 }
421 
422 static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
423 {
424 	ZSTD_DStream *stream;
425 	void *workspace;
426 	unsigned int workspace_size;
427 	unsigned int max_window_size =
428 			MAX_COMPRESS_WINDOW_SIZE(dic->log_cluster_size);
429 
430 	workspace_size = ZSTD_DStreamWorkspaceBound(max_window_size);
431 
432 	workspace = f2fs_kvmalloc(F2FS_I_SB(dic->inode),
433 					workspace_size, GFP_NOFS);
434 	if (!workspace)
435 		return -ENOMEM;
436 
437 	stream = ZSTD_initDStream(max_window_size, workspace, workspace_size);
438 	if (!stream) {
439 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initDStream failed\n",
440 				KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
441 				__func__);
442 		kvfree(workspace);
443 		return -EIO;
444 	}
445 
446 	dic->private = workspace;
447 	dic->private2 = stream;
448 
449 	return 0;
450 }
451 
452 static void zstd_destroy_decompress_ctx(struct decompress_io_ctx *dic)
453 {
454 	kvfree(dic->private);
455 	dic->private = NULL;
456 	dic->private2 = NULL;
457 }
458 
459 static int zstd_decompress_pages(struct decompress_io_ctx *dic)
460 {
461 	ZSTD_DStream *stream = dic->private2;
462 	ZSTD_inBuffer inbuf;
463 	ZSTD_outBuffer outbuf;
464 	int ret;
465 
466 	inbuf.pos = 0;
467 	inbuf.src = dic->cbuf->cdata;
468 	inbuf.size = dic->clen;
469 
470 	outbuf.pos = 0;
471 	outbuf.dst = dic->rbuf;
472 	outbuf.size = dic->rlen;
473 
474 	ret = ZSTD_decompressStream(stream, &outbuf, &inbuf);
475 	if (ZSTD_isError(ret)) {
476 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
477 				KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
478 				__func__, ZSTD_getErrorCode(ret));
479 		return -EIO;
480 	}
481 
482 	if (dic->rlen != outbuf.pos) {
483 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD invalid rlen:%zu, "
484 				"expected:%lu\n", KERN_ERR,
485 				F2FS_I_SB(dic->inode)->sb->s_id,
486 				__func__, dic->rlen,
487 				PAGE_SIZE << dic->log_cluster_size);
488 		return -EIO;
489 	}
490 
491 	return 0;
492 }
493 
494 static const struct f2fs_compress_ops f2fs_zstd_ops = {
495 	.init_compress_ctx	= zstd_init_compress_ctx,
496 	.destroy_compress_ctx	= zstd_destroy_compress_ctx,
497 	.compress_pages		= zstd_compress_pages,
498 	.init_decompress_ctx	= zstd_init_decompress_ctx,
499 	.destroy_decompress_ctx	= zstd_destroy_decompress_ctx,
500 	.decompress_pages	= zstd_decompress_pages,
501 };
502 #endif
503 
504 #ifdef CONFIG_F2FS_FS_LZO
505 #ifdef CONFIG_F2FS_FS_LZORLE
506 static int lzorle_compress_pages(struct compress_ctx *cc)
507 {
508 	int ret;
509 
510 	ret = lzorle1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
511 					&cc->clen, cc->private);
512 	if (ret != LZO_E_OK) {
513 		printk_ratelimited("%sF2FS-fs (%s): lzo-rle compress failed, ret:%d\n",
514 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
515 		return -EIO;
516 	}
517 	return 0;
518 }
519 
520 static const struct f2fs_compress_ops f2fs_lzorle_ops = {
521 	.init_compress_ctx	= lzo_init_compress_ctx,
522 	.destroy_compress_ctx	= lzo_destroy_compress_ctx,
523 	.compress_pages		= lzorle_compress_pages,
524 	.decompress_pages	= lzo_decompress_pages,
525 };
526 #endif
527 #endif
528 
529 static const struct f2fs_compress_ops *f2fs_cops[COMPRESS_MAX] = {
530 #ifdef CONFIG_F2FS_FS_LZO
531 	&f2fs_lzo_ops,
532 #else
533 	NULL,
534 #endif
535 #ifdef CONFIG_F2FS_FS_LZ4
536 	&f2fs_lz4_ops,
537 #else
538 	NULL,
539 #endif
540 #ifdef CONFIG_F2FS_FS_ZSTD
541 	&f2fs_zstd_ops,
542 #else
543 	NULL,
544 #endif
545 #if defined(CONFIG_F2FS_FS_LZO) && defined(CONFIG_F2FS_FS_LZORLE)
546 	&f2fs_lzorle_ops,
547 #else
548 	NULL,
549 #endif
550 };
551 
552 bool f2fs_is_compress_backend_ready(struct inode *inode)
553 {
554 	if (!f2fs_compressed_file(inode))
555 		return true;
556 	return f2fs_cops[F2FS_I(inode)->i_compress_algorithm];
557 }
558 
559 static mempool_t *compress_page_pool;
560 static int num_compress_pages = 512;
561 module_param(num_compress_pages, uint, 0444);
562 MODULE_PARM_DESC(num_compress_pages,
563 		"Number of intermediate compress pages to preallocate");
564 
565 int f2fs_init_compress_mempool(void)
566 {
567 	compress_page_pool = mempool_create_page_pool(num_compress_pages, 0);
568 	if (!compress_page_pool)
569 		return -ENOMEM;
570 
571 	return 0;
572 }
573 
574 void f2fs_destroy_compress_mempool(void)
575 {
576 	mempool_destroy(compress_page_pool);
577 }
578 
579 static struct page *f2fs_compress_alloc_page(void)
580 {
581 	struct page *page;
582 
583 	page = mempool_alloc(compress_page_pool, GFP_NOFS);
584 	lock_page(page);
585 
586 	return page;
587 }
588 
589 static void f2fs_compress_free_page(struct page *page)
590 {
591 	if (!page)
592 		return;
593 	detach_page_private(page);
594 	page->mapping = NULL;
595 	unlock_page(page);
596 	mempool_free(page, compress_page_pool);
597 }
598 
599 #define MAX_VMAP_RETRIES	3
600 
601 static void *f2fs_vmap(struct page **pages, unsigned int count)
602 {
603 	int i;
604 	void *buf = NULL;
605 
606 	for (i = 0; i < MAX_VMAP_RETRIES; i++) {
607 		buf = vm_map_ram(pages, count, -1);
608 		if (buf)
609 			break;
610 		vm_unmap_aliases();
611 	}
612 	return buf;
613 }
614 
615 static int f2fs_compress_pages(struct compress_ctx *cc)
616 {
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;
622 	u32 chksum = 0;
623 	int i, ret;
624 
625 	trace_f2fs_compress_pages_start(cc->inode, cc->cluster_idx,
626 				cc->cluster_size, fi->i_compress_algorithm);
627 
628 	if (cops->init_compress_ctx) {
629 		ret = cops->init_compress_ctx(cc);
630 		if (ret)
631 			goto out;
632 	}
633 
634 	max_len = COMPRESS_HEADER_SIZE + cc->clen;
635 	cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE);
636 
637 	cc->cpages = page_array_alloc(cc->inode, cc->nr_cpages);
638 	if (!cc->cpages) {
639 		ret = -ENOMEM;
640 		goto destroy_compress_ctx;
641 	}
642 
643 	for (i = 0; i < cc->nr_cpages; i++) {
644 		cc->cpages[i] = f2fs_compress_alloc_page();
645 		if (!cc->cpages[i]) {
646 			ret = -ENOMEM;
647 			goto out_free_cpages;
648 		}
649 	}
650 
651 	cc->rbuf = f2fs_vmap(cc->rpages, cc->cluster_size);
652 	if (!cc->rbuf) {
653 		ret = -ENOMEM;
654 		goto out_free_cpages;
655 	}
656 
657 	cc->cbuf = f2fs_vmap(cc->cpages, cc->nr_cpages);
658 	if (!cc->cbuf) {
659 		ret = -ENOMEM;
660 		goto out_vunmap_rbuf;
661 	}
662 
663 	ret = cops->compress_pages(cc);
664 	if (ret)
665 		goto out_vunmap_cbuf;
666 
667 	max_len = PAGE_SIZE * (cc->cluster_size - 1) - COMPRESS_HEADER_SIZE;
668 
669 	if (cc->clen > max_len) {
670 		ret = -EAGAIN;
671 		goto out_vunmap_cbuf;
672 	}
673 
674 	cc->cbuf->clen = cpu_to_le32(cc->clen);
675 
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);
680 
681 	for (i = 0; i < COMPRESS_DATA_RESERVED_SIZE; i++)
682 		cc->cbuf->reserved[i] = cpu_to_le32(0);
683 
684 	new_nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE);
685 
686 	/* Now we're going to cut unnecessary tail pages */
687 	new_cpages = page_array_alloc(cc->inode, new_nr_cpages);
688 	if (!new_cpages) {
689 		ret = -ENOMEM;
690 		goto out_vunmap_cbuf;
691 	}
692 
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));
697 
698 	vm_unmap_ram(cc->cbuf, cc->nr_cpages);
699 	vm_unmap_ram(cc->rbuf, cc->cluster_size);
700 
701 	for (i = 0; i < cc->nr_cpages; i++) {
702 		if (i < new_nr_cpages) {
703 			new_cpages[i] = cc->cpages[i];
704 			continue;
705 		}
706 		f2fs_compress_free_page(cc->cpages[i]);
707 		cc->cpages[i] = NULL;
708 	}
709 
710 	if (cops->destroy_compress_ctx)
711 		cops->destroy_compress_ctx(cc);
712 
713 	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
714 	cc->cpages = new_cpages;
715 	cc->nr_cpages = new_nr_cpages;
716 
717 	trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
718 							cc->clen, ret);
719 	return 0;
720 
721 out_vunmap_cbuf:
722 	vm_unmap_ram(cc->cbuf, cc->nr_cpages);
723 out_vunmap_rbuf:
724 	vm_unmap_ram(cc->rbuf, cc->cluster_size);
725 out_free_cpages:
726 	for (i = 0; i < cc->nr_cpages; i++) {
727 		if (cc->cpages[i])
728 			f2fs_compress_free_page(cc->cpages[i]);
729 	}
730 	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
731 	cc->cpages = NULL;
732 destroy_compress_ctx:
733 	if (cops->destroy_compress_ctx)
734 		cops->destroy_compress_ctx(cc);
735 out:
736 	trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
737 							cc->clen, ret);
738 	return ret;
739 }
740 
741 void f2fs_decompress_cluster(struct decompress_io_ctx *dic)
742 {
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];
747 	int ret;
748 	int i;
749 
750 	trace_f2fs_decompress_pages_start(dic->inode, dic->cluster_idx,
751 				dic->cluster_size, fi->i_compress_algorithm);
752 
753 	if (dic->failed) {
754 		ret = -EIO;
755 		goto out_end_io;
756 	}
757 
758 	dic->tpages = page_array_alloc(dic->inode, dic->cluster_size);
759 	if (!dic->tpages) {
760 		ret = -ENOMEM;
761 		goto out_end_io;
762 	}
763 
764 	for (i = 0; i < dic->cluster_size; i++) {
765 		if (dic->rpages[i]) {
766 			dic->tpages[i] = dic->rpages[i];
767 			continue;
768 		}
769 
770 		dic->tpages[i] = f2fs_compress_alloc_page();
771 		if (!dic->tpages[i]) {
772 			ret = -ENOMEM;
773 			goto out_end_io;
774 		}
775 	}
776 
777 	if (cops->init_decompress_ctx) {
778 		ret = cops->init_decompress_ctx(dic);
779 		if (ret)
780 			goto out_end_io;
781 	}
782 
783 	dic->rbuf = f2fs_vmap(dic->tpages, dic->cluster_size);
784 	if (!dic->rbuf) {
785 		ret = -ENOMEM;
786 		goto out_destroy_decompress_ctx;
787 	}
788 
789 	dic->cbuf = f2fs_vmap(dic->cpages, dic->nr_cpages);
790 	if (!dic->cbuf) {
791 		ret = -ENOMEM;
792 		goto out_vunmap_rbuf;
793 	}
794 
795 	dic->clen = le32_to_cpu(dic->cbuf->clen);
796 	dic->rlen = PAGE_SIZE << dic->log_cluster_size;
797 
798 	if (dic->clen > PAGE_SIZE * dic->nr_cpages - COMPRESS_HEADER_SIZE) {
799 		ret = -EFSCORRUPTED;
800 		goto out_vunmap_cbuf;
801 	}
802 
803 	ret = cops->decompress_pages(dic);
804 
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);
808 
809 		if (provided != calculated) {
810 			if (!is_inode_flag_set(dic->inode, FI_COMPRESS_CORRUPT)) {
811 				set_inode_flag(dic->inode, FI_COMPRESS_CORRUPT);
812 				printk_ratelimited(
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);
816 			}
817 			set_sbi_flag(sbi, SBI_NEED_FSCK);
818 		}
819 	}
820 
821 out_vunmap_cbuf:
822 	vm_unmap_ram(dic->cbuf, dic->nr_cpages);
823 out_vunmap_rbuf:
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);
828 out_end_io:
829 	trace_f2fs_decompress_pages_end(dic->inode, dic->cluster_idx,
830 							dic->clen, ret);
831 	f2fs_decompress_end_io(dic, ret);
832 }
833 
834 /*
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).
839  */
840 void f2fs_end_read_compressed_page(struct page *page, bool failed,
841 						block_t blkaddr)
842 {
843 	struct decompress_io_ctx *dic =
844 			(struct decompress_io_ctx *)page_private(page);
845 	struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
846 
847 	dec_page_count(sbi, F2FS_RD_DATA);
848 
849 	if (failed)
850 		WRITE_ONCE(dic->failed, true);
851 	else if (blkaddr)
852 		f2fs_cache_compressed_page(sbi, page,
853 					dic->inode->i_ino, blkaddr);
854 
855 	if (atomic_dec_and_test(&dic->remaining_pages))
856 		f2fs_decompress_cluster(dic);
857 }
858 
859 static bool is_page_in_cluster(struct compress_ctx *cc, pgoff_t index)
860 {
861 	if (cc->cluster_idx == NULL_CLUSTER)
862 		return true;
863 	return cc->cluster_idx == cluster_idx(cc, index);
864 }
865 
866 bool f2fs_cluster_is_empty(struct compress_ctx *cc)
867 {
868 	return cc->nr_rpages == 0;
869 }
870 
871 static bool f2fs_cluster_is_full(struct compress_ctx *cc)
872 {
873 	return cc->cluster_size == cc->nr_rpages;
874 }
875 
876 bool f2fs_cluster_can_merge_page(struct compress_ctx *cc, pgoff_t index)
877 {
878 	if (f2fs_cluster_is_empty(cc))
879 		return true;
880 	return is_page_in_cluster(cc, index);
881 }
882 
883 static bool cluster_has_invalid_data(struct compress_ctx *cc)
884 {
885 	loff_t i_size = i_size_read(cc->inode);
886 	unsigned nr_pages = DIV_ROUND_UP(i_size, PAGE_SIZE);
887 	int i;
888 
889 	for (i = 0; i < cc->cluster_size; i++) {
890 		struct page *page = cc->rpages[i];
891 
892 		f2fs_bug_on(F2FS_I_SB(cc->inode), !page);
893 
894 		/* beyond EOF */
895 		if (page->index >= nr_pages)
896 			return true;
897 	}
898 	return false;
899 }
900 
901 static int __f2fs_cluster_blocks(struct inode *inode,
902 				unsigned int cluster_idx, bool compr)
903 {
904 	struct dnode_of_data dn;
905 	unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
906 	unsigned int start_idx = cluster_idx <<
907 				F2FS_I(inode)->i_log_cluster_size;
908 	int ret;
909 
910 	set_new_dnode(&dn, inode, NULL, NULL, 0);
911 	ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
912 	if (ret) {
913 		if (ret == -ENOENT)
914 			ret = 0;
915 		goto fail;
916 	}
917 
918 	if (dn.data_blkaddr == COMPRESS_ADDR) {
919 		int i;
920 
921 		ret = 1;
922 		for (i = 1; i < cluster_size; i++) {
923 			block_t blkaddr;
924 
925 			blkaddr = data_blkaddr(dn.inode,
926 					dn.node_page, dn.ofs_in_node + i);
927 			if (compr) {
928 				if (__is_valid_data_blkaddr(blkaddr))
929 					ret++;
930 			} else {
931 				if (blkaddr != NULL_ADDR)
932 					ret++;
933 			}
934 		}
935 
936 		f2fs_bug_on(F2FS_I_SB(inode),
937 			!compr && ret != cluster_size &&
938 			!is_inode_flag_set(inode, FI_COMPRESS_RELEASED));
939 	}
940 fail:
941 	f2fs_put_dnode(&dn);
942 	return ret;
943 }
944 
945 /* return # of compressed blocks in compressed cluster */
946 static int f2fs_compressed_blocks(struct compress_ctx *cc)
947 {
948 	return __f2fs_cluster_blocks(cc->inode, cc->cluster_idx, true);
949 }
950 
951 /* return # of valid blocks in compressed cluster */
952 int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index)
953 {
954 	return __f2fs_cluster_blocks(inode,
955 		index >> F2FS_I(inode)->i_log_cluster_size,
956 		false);
957 }
958 
959 static bool cluster_may_compress(struct compress_ctx *cc)
960 {
961 	if (!f2fs_need_compress_data(cc->inode))
962 		return false;
963 	if (f2fs_is_atomic_file(cc->inode))
964 		return false;
965 	if (!f2fs_cluster_is_full(cc))
966 		return false;
967 	if (unlikely(f2fs_cp_error(F2FS_I_SB(cc->inode))))
968 		return false;
969 	return !cluster_has_invalid_data(cc);
970 }
971 
972 static void set_cluster_writeback(struct compress_ctx *cc)
973 {
974 	int i;
975 
976 	for (i = 0; i < cc->cluster_size; i++) {
977 		if (cc->rpages[i])
978 			set_page_writeback(cc->rpages[i]);
979 	}
980 }
981 
982 static void set_cluster_dirty(struct compress_ctx *cc)
983 {
984 	int i;
985 
986 	for (i = 0; i < cc->cluster_size; i++)
987 		if (cc->rpages[i])
988 			set_page_dirty(cc->rpages[i]);
989 }
990 
991 static int prepare_compress_overwrite(struct compress_ctx *cc,
992 		struct page **pagep, pgoff_t index, void **fsdata)
993 {
994 	struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
995 	struct address_space *mapping = cc->inode->i_mapping;
996 	struct page *page;
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);
1000 	int i, ret;
1001 
1002 retry:
1003 	ret = f2fs_is_compressed_cluster(cc->inode, start_idx);
1004 	if (ret <= 0)
1005 		return ret;
1006 
1007 	ret = f2fs_init_compress_ctx(cc);
1008 	if (ret)
1009 		return ret;
1010 
1011 	/* keep page reference to avoid page reclaim */
1012 	for (i = 0; i < cc->cluster_size; i++) {
1013 		page = f2fs_pagecache_get_page(mapping, start_idx + i,
1014 							fgp_flag, GFP_NOFS);
1015 		if (!page) {
1016 			ret = -ENOMEM;
1017 			goto unlock_pages;
1018 		}
1019 
1020 		if (PageUptodate(page))
1021 			f2fs_put_page(page, 1);
1022 		else
1023 			f2fs_compress_ctx_add_page(cc, page);
1024 	}
1025 
1026 	if (!f2fs_cluster_is_empty(cc)) {
1027 		struct bio *bio = NULL;
1028 
1029 		ret = f2fs_read_multi_pages(cc, &bio, cc->cluster_size,
1030 					&last_block_in_bio, false, true);
1031 		f2fs_put_rpages(cc);
1032 		f2fs_destroy_compress_ctx(cc, true);
1033 		if (ret)
1034 			goto out;
1035 		if (bio)
1036 			f2fs_submit_bio(sbi, bio, DATA);
1037 
1038 		ret = f2fs_init_compress_ctx(cc);
1039 		if (ret)
1040 			goto out;
1041 	}
1042 
1043 	for (i = 0; i < cc->cluster_size; i++) {
1044 		f2fs_bug_on(sbi, cc->rpages[i]);
1045 
1046 		page = find_lock_page(mapping, start_idx + i);
1047 		if (!page) {
1048 			/* page can be truncated */
1049 			goto release_and_retry;
1050 		}
1051 
1052 		f2fs_wait_on_page_writeback(page, DATA, true, true);
1053 		f2fs_compress_ctx_add_page(cc, page);
1054 
1055 		if (!PageUptodate(page)) {
1056 release_and_retry:
1057 			f2fs_put_rpages(cc);
1058 			f2fs_unlock_rpages(cc, i + 1);
1059 			f2fs_destroy_compress_ctx(cc, true);
1060 			goto retry;
1061 		}
1062 	}
1063 
1064 	if (likely(!ret)) {
1065 		*fsdata = cc->rpages;
1066 		*pagep = cc->rpages[offset_in_cluster(cc, index)];
1067 		return cc->cluster_size;
1068 	}
1069 
1070 unlock_pages:
1071 	f2fs_put_rpages(cc);
1072 	f2fs_unlock_rpages(cc, i);
1073 	f2fs_destroy_compress_ctx(cc, true);
1074 out:
1075 	return ret;
1076 }
1077 
1078 int f2fs_prepare_compress_overwrite(struct inode *inode,
1079 		struct page **pagep, pgoff_t index, void **fsdata)
1080 {
1081 	struct compress_ctx cc = {
1082 		.inode = inode,
1083 		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1084 		.cluster_size = F2FS_I(inode)->i_cluster_size,
1085 		.cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
1086 		.rpages = NULL,
1087 		.nr_rpages = 0,
1088 	};
1089 
1090 	return prepare_compress_overwrite(&cc, pagep, index, fsdata);
1091 }
1092 
1093 bool f2fs_compress_write_end(struct inode *inode, void *fsdata,
1094 					pgoff_t index, unsigned copied)
1095 
1096 {
1097 	struct compress_ctx cc = {
1098 		.inode = inode,
1099 		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1100 		.cluster_size = F2FS_I(inode)->i_cluster_size,
1101 		.rpages = fsdata,
1102 	};
1103 	bool first_index = (index == cc.rpages[0]->index);
1104 
1105 	if (copied)
1106 		set_cluster_dirty(&cc);
1107 
1108 	f2fs_put_rpages_wbc(&cc, NULL, false, 1);
1109 	f2fs_destroy_compress_ctx(&cc, false);
1110 
1111 	return first_index;
1112 }
1113 
1114 int f2fs_truncate_partial_cluster(struct inode *inode, u64 from, bool lock)
1115 {
1116 	void *fsdata = NULL;
1117 	struct page *pagep;
1118 	int log_cluster_size = F2FS_I(inode)->i_log_cluster_size;
1119 	pgoff_t start_idx = from >> (PAGE_SHIFT + log_cluster_size) <<
1120 							log_cluster_size;
1121 	int err;
1122 
1123 	err = f2fs_is_compressed_cluster(inode, start_idx);
1124 	if (err < 0)
1125 		return err;
1126 
1127 	/* truncate normal cluster */
1128 	if (!err)
1129 		return f2fs_do_truncate_blocks(inode, from, lock);
1130 
1131 	/* truncate compressed cluster */
1132 	err = f2fs_prepare_compress_overwrite(inode, &pagep,
1133 						start_idx, &fsdata);
1134 
1135 	/* should not be a normal cluster */
1136 	f2fs_bug_on(F2FS_I_SB(inode), err == 0);
1137 
1138 	if (err <= 0)
1139 		return err;
1140 
1141 	if (err > 0) {
1142 		struct page **rpages = fsdata;
1143 		int cluster_size = F2FS_I(inode)->i_cluster_size;
1144 		int i;
1145 
1146 		for (i = cluster_size - 1; i >= 0; i--) {
1147 			loff_t start = rpages[i]->index << PAGE_SHIFT;
1148 
1149 			if (from <= start) {
1150 				zero_user_segment(rpages[i], 0, PAGE_SIZE);
1151 			} else {
1152 				zero_user_segment(rpages[i], from - start,
1153 								PAGE_SIZE);
1154 				break;
1155 			}
1156 		}
1157 
1158 		f2fs_compress_write_end(inode, fsdata, start_idx, true);
1159 	}
1160 	return 0;
1161 }
1162 
1163 static int f2fs_write_compressed_pages(struct compress_ctx *cc,
1164 					int *submitted,
1165 					struct writeback_control *wbc,
1166 					enum iostat_type io_type)
1167 {
1168 	struct inode *inode = cc->inode;
1169 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1170 	struct f2fs_inode_info *fi = F2FS_I(inode);
1171 	struct f2fs_io_info fio = {
1172 		.sbi = sbi,
1173 		.ino = cc->inode->i_ino,
1174 		.type = DATA,
1175 		.op = REQ_OP_WRITE,
1176 		.op_flags = wbc_to_write_flags(wbc),
1177 		.old_blkaddr = NEW_ADDR,
1178 		.page = NULL,
1179 		.encrypted_page = NULL,
1180 		.compressed_page = NULL,
1181 		.submitted = false,
1182 		.io_type = io_type,
1183 		.io_wbc = wbc,
1184 		.encrypted = fscrypt_inode_uses_fs_layer_crypto(cc->inode),
1185 	};
1186 	struct dnode_of_data dn;
1187 	struct node_info ni;
1188 	struct compress_io_ctx *cic;
1189 	pgoff_t start_idx = start_idx_of_cluster(cc);
1190 	unsigned int last_index = cc->cluster_size - 1;
1191 	loff_t psize;
1192 	int i, err;
1193 
1194 	/* we should bypass data pages to proceed the kworkder jobs */
1195 	if (unlikely(f2fs_cp_error(sbi))) {
1196 		mapping_set_error(cc->rpages[0]->mapping, -EIO);
1197 		goto out_free;
1198 	}
1199 
1200 	if (IS_NOQUOTA(inode)) {
1201 		/*
1202 		 * We need to wait for node_write to avoid block allocation during
1203 		 * checkpoint. This can only happen to quota writes which can cause
1204 		 * the below discard race condition.
1205 		 */
1206 		down_read(&sbi->node_write);
1207 	} else if (!f2fs_trylock_op(sbi)) {
1208 		goto out_free;
1209 	}
1210 
1211 	set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1212 
1213 	err = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
1214 	if (err)
1215 		goto out_unlock_op;
1216 
1217 	for (i = 0; i < cc->cluster_size; i++) {
1218 		if (data_blkaddr(dn.inode, dn.node_page,
1219 					dn.ofs_in_node + i) == NULL_ADDR)
1220 			goto out_put_dnode;
1221 	}
1222 
1223 	psize = (loff_t)(cc->rpages[last_index]->index + 1) << PAGE_SHIFT;
1224 
1225 	err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1226 	if (err)
1227 		goto out_put_dnode;
1228 
1229 	fio.version = ni.version;
1230 
1231 	cic = kmem_cache_zalloc(cic_entry_slab, GFP_NOFS);
1232 	if (!cic)
1233 		goto out_put_dnode;
1234 
1235 	cic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1236 	cic->inode = inode;
1237 	atomic_set(&cic->pending_pages, cc->nr_cpages);
1238 	cic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1239 	if (!cic->rpages)
1240 		goto out_put_cic;
1241 
1242 	cic->nr_rpages = cc->cluster_size;
1243 
1244 	for (i = 0; i < cc->nr_cpages; i++) {
1245 		f2fs_set_compressed_page(cc->cpages[i], inode,
1246 					cc->rpages[i + 1]->index, cic);
1247 		fio.compressed_page = cc->cpages[i];
1248 
1249 		fio.old_blkaddr = data_blkaddr(dn.inode, dn.node_page,
1250 						dn.ofs_in_node + i + 1);
1251 
1252 		/* wait for GCed page writeback via META_MAPPING */
1253 		f2fs_wait_on_block_writeback(inode, fio.old_blkaddr);
1254 
1255 		if (fio.encrypted) {
1256 			fio.page = cc->rpages[i + 1];
1257 			err = f2fs_encrypt_one_page(&fio);
1258 			if (err)
1259 				goto out_destroy_crypt;
1260 			cc->cpages[i] = fio.encrypted_page;
1261 		}
1262 	}
1263 
1264 	set_cluster_writeback(cc);
1265 
1266 	for (i = 0; i < cc->cluster_size; i++)
1267 		cic->rpages[i] = cc->rpages[i];
1268 
1269 	for (i = 0; i < cc->cluster_size; i++, dn.ofs_in_node++) {
1270 		block_t blkaddr;
1271 
1272 		blkaddr = f2fs_data_blkaddr(&dn);
1273 		fio.page = cc->rpages[i];
1274 		fio.old_blkaddr = blkaddr;
1275 
1276 		/* cluster header */
1277 		if (i == 0) {
1278 			if (blkaddr == COMPRESS_ADDR)
1279 				fio.compr_blocks++;
1280 			if (__is_valid_data_blkaddr(blkaddr))
1281 				f2fs_invalidate_blocks(sbi, blkaddr);
1282 			f2fs_update_data_blkaddr(&dn, COMPRESS_ADDR);
1283 			goto unlock_continue;
1284 		}
1285 
1286 		if (fio.compr_blocks && __is_valid_data_blkaddr(blkaddr))
1287 			fio.compr_blocks++;
1288 
1289 		if (i > cc->nr_cpages) {
1290 			if (__is_valid_data_blkaddr(blkaddr)) {
1291 				f2fs_invalidate_blocks(sbi, blkaddr);
1292 				f2fs_update_data_blkaddr(&dn, NEW_ADDR);
1293 			}
1294 			goto unlock_continue;
1295 		}
1296 
1297 		f2fs_bug_on(fio.sbi, blkaddr == NULL_ADDR);
1298 
1299 		if (fio.encrypted)
1300 			fio.encrypted_page = cc->cpages[i - 1];
1301 		else
1302 			fio.compressed_page = cc->cpages[i - 1];
1303 
1304 		cc->cpages[i - 1] = NULL;
1305 		f2fs_outplace_write_data(&dn, &fio);
1306 		(*submitted)++;
1307 unlock_continue:
1308 		inode_dec_dirty_pages(cc->inode);
1309 		unlock_page(fio.page);
1310 	}
1311 
1312 	if (fio.compr_blocks)
1313 		f2fs_i_compr_blocks_update(inode, fio.compr_blocks - 1, false);
1314 	f2fs_i_compr_blocks_update(inode, cc->nr_cpages, true);
1315 	add_compr_block_stat(inode, cc->nr_cpages);
1316 
1317 	set_inode_flag(cc->inode, FI_APPEND_WRITE);
1318 	if (cc->cluster_idx == 0)
1319 		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1320 
1321 	f2fs_put_dnode(&dn);
1322 	if (IS_NOQUOTA(inode))
1323 		up_read(&sbi->node_write);
1324 	else
1325 		f2fs_unlock_op(sbi);
1326 
1327 	spin_lock(&fi->i_size_lock);
1328 	if (fi->last_disk_size < psize)
1329 		fi->last_disk_size = psize;
1330 	spin_unlock(&fi->i_size_lock);
1331 
1332 	f2fs_put_rpages(cc);
1333 	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1334 	cc->cpages = NULL;
1335 	f2fs_destroy_compress_ctx(cc, false);
1336 	return 0;
1337 
1338 out_destroy_crypt:
1339 	page_array_free(cc->inode, cic->rpages, cc->cluster_size);
1340 
1341 	for (--i; i >= 0; i--)
1342 		fscrypt_finalize_bounce_page(&cc->cpages[i]);
1343 	for (i = 0; i < cc->nr_cpages; i++) {
1344 		if (!cc->cpages[i])
1345 			continue;
1346 		f2fs_compress_free_page(cc->cpages[i]);
1347 		cc->cpages[i] = NULL;
1348 	}
1349 out_put_cic:
1350 	kmem_cache_free(cic_entry_slab, cic);
1351 out_put_dnode:
1352 	f2fs_put_dnode(&dn);
1353 out_unlock_op:
1354 	if (IS_NOQUOTA(inode))
1355 		up_read(&sbi->node_write);
1356 	else
1357 		f2fs_unlock_op(sbi);
1358 out_free:
1359 	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1360 	cc->cpages = NULL;
1361 	return -EAGAIN;
1362 }
1363 
1364 void f2fs_compress_write_end_io(struct bio *bio, struct page *page)
1365 {
1366 	struct f2fs_sb_info *sbi = bio->bi_private;
1367 	struct compress_io_ctx *cic =
1368 			(struct compress_io_ctx *)page_private(page);
1369 	int i;
1370 
1371 	if (unlikely(bio->bi_status))
1372 		mapping_set_error(cic->inode->i_mapping, -EIO);
1373 
1374 	f2fs_compress_free_page(page);
1375 
1376 	dec_page_count(sbi, F2FS_WB_DATA);
1377 
1378 	if (atomic_dec_return(&cic->pending_pages))
1379 		return;
1380 
1381 	for (i = 0; i < cic->nr_rpages; i++) {
1382 		WARN_ON(!cic->rpages[i]);
1383 		clear_page_private_gcing(cic->rpages[i]);
1384 		end_page_writeback(cic->rpages[i]);
1385 	}
1386 
1387 	page_array_free(cic->inode, cic->rpages, cic->nr_rpages);
1388 	kmem_cache_free(cic_entry_slab, cic);
1389 }
1390 
1391 static int f2fs_write_raw_pages(struct compress_ctx *cc,
1392 					int *submitted,
1393 					struct writeback_control *wbc,
1394 					enum iostat_type io_type)
1395 {
1396 	struct address_space *mapping = cc->inode->i_mapping;
1397 	int _submitted, compr_blocks, ret;
1398 	int i = -1, err = 0;
1399 
1400 	compr_blocks = f2fs_compressed_blocks(cc);
1401 	if (compr_blocks < 0) {
1402 		err = compr_blocks;
1403 		goto out_err;
1404 	}
1405 
1406 	for (i = 0; i < cc->cluster_size; i++) {
1407 		if (!cc->rpages[i])
1408 			continue;
1409 retry_write:
1410 		if (cc->rpages[i]->mapping != mapping) {
1411 			unlock_page(cc->rpages[i]);
1412 			continue;
1413 		}
1414 
1415 		BUG_ON(!PageLocked(cc->rpages[i]));
1416 
1417 		ret = f2fs_write_single_data_page(cc->rpages[i], &_submitted,
1418 						NULL, NULL, wbc, io_type,
1419 						compr_blocks, false);
1420 		if (ret) {
1421 			if (ret == AOP_WRITEPAGE_ACTIVATE) {
1422 				unlock_page(cc->rpages[i]);
1423 				ret = 0;
1424 			} else if (ret == -EAGAIN) {
1425 				/*
1426 				 * for quota file, just redirty left pages to
1427 				 * avoid deadlock caused by cluster update race
1428 				 * from foreground operation.
1429 				 */
1430 				if (IS_NOQUOTA(cc->inode)) {
1431 					err = 0;
1432 					goto out_err;
1433 				}
1434 				ret = 0;
1435 				cond_resched();
1436 				congestion_wait(BLK_RW_ASYNC,
1437 						DEFAULT_IO_TIMEOUT);
1438 				lock_page(cc->rpages[i]);
1439 
1440 				if (!PageDirty(cc->rpages[i])) {
1441 					unlock_page(cc->rpages[i]);
1442 					continue;
1443 				}
1444 
1445 				clear_page_dirty_for_io(cc->rpages[i]);
1446 				goto retry_write;
1447 			}
1448 			err = ret;
1449 			goto out_err;
1450 		}
1451 
1452 		*submitted += _submitted;
1453 	}
1454 
1455 	f2fs_balance_fs(F2FS_M_SB(mapping), true);
1456 
1457 	return 0;
1458 out_err:
1459 	for (++i; i < cc->cluster_size; i++) {
1460 		if (!cc->rpages[i])
1461 			continue;
1462 		redirty_page_for_writepage(wbc, cc->rpages[i]);
1463 		unlock_page(cc->rpages[i]);
1464 	}
1465 	return err;
1466 }
1467 
1468 int f2fs_write_multi_pages(struct compress_ctx *cc,
1469 					int *submitted,
1470 					struct writeback_control *wbc,
1471 					enum iostat_type io_type)
1472 {
1473 	int err;
1474 
1475 	*submitted = 0;
1476 	if (cluster_may_compress(cc)) {
1477 		err = f2fs_compress_pages(cc);
1478 		if (err == -EAGAIN) {
1479 			goto write;
1480 		} else if (err) {
1481 			f2fs_put_rpages_wbc(cc, wbc, true, 1);
1482 			goto destroy_out;
1483 		}
1484 
1485 		err = f2fs_write_compressed_pages(cc, submitted,
1486 							wbc, io_type);
1487 		if (!err)
1488 			return 0;
1489 		f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN);
1490 	}
1491 write:
1492 	f2fs_bug_on(F2FS_I_SB(cc->inode), *submitted);
1493 
1494 	err = f2fs_write_raw_pages(cc, submitted, wbc, io_type);
1495 	f2fs_put_rpages_wbc(cc, wbc, false, 0);
1496 destroy_out:
1497 	f2fs_destroy_compress_ctx(cc, false);
1498 	return err;
1499 }
1500 
1501 static void f2fs_free_dic(struct decompress_io_ctx *dic);
1502 
1503 struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
1504 {
1505 	struct decompress_io_ctx *dic;
1506 	pgoff_t start_idx = start_idx_of_cluster(cc);
1507 	int i;
1508 
1509 	dic = kmem_cache_zalloc(dic_entry_slab, GFP_NOFS);
1510 	if (!dic)
1511 		return ERR_PTR(-ENOMEM);
1512 
1513 	dic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1514 	if (!dic->rpages) {
1515 		kmem_cache_free(dic_entry_slab, dic);
1516 		return ERR_PTR(-ENOMEM);
1517 	}
1518 
1519 	dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1520 	dic->inode = cc->inode;
1521 	atomic_set(&dic->remaining_pages, cc->nr_cpages);
1522 	dic->cluster_idx = cc->cluster_idx;
1523 	dic->cluster_size = cc->cluster_size;
1524 	dic->log_cluster_size = cc->log_cluster_size;
1525 	dic->nr_cpages = cc->nr_cpages;
1526 	refcount_set(&dic->refcnt, 1);
1527 	dic->failed = false;
1528 	dic->need_verity = f2fs_need_verity(cc->inode, start_idx);
1529 
1530 	for (i = 0; i < dic->cluster_size; i++)
1531 		dic->rpages[i] = cc->rpages[i];
1532 	dic->nr_rpages = cc->cluster_size;
1533 
1534 	dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages);
1535 	if (!dic->cpages)
1536 		goto out_free;
1537 
1538 	for (i = 0; i < dic->nr_cpages; i++) {
1539 		struct page *page;
1540 
1541 		page = f2fs_compress_alloc_page();
1542 		if (!page)
1543 			goto out_free;
1544 
1545 		f2fs_set_compressed_page(page, cc->inode,
1546 					start_idx + i + 1, dic);
1547 		dic->cpages[i] = page;
1548 	}
1549 
1550 	return dic;
1551 
1552 out_free:
1553 	f2fs_free_dic(dic);
1554 	return ERR_PTR(-ENOMEM);
1555 }
1556 
1557 static void f2fs_free_dic(struct decompress_io_ctx *dic)
1558 {
1559 	int i;
1560 
1561 	if (dic->tpages) {
1562 		for (i = 0; i < dic->cluster_size; i++) {
1563 			if (dic->rpages[i])
1564 				continue;
1565 			if (!dic->tpages[i])
1566 				continue;
1567 			f2fs_compress_free_page(dic->tpages[i]);
1568 		}
1569 		page_array_free(dic->inode, dic->tpages, dic->cluster_size);
1570 	}
1571 
1572 	if (dic->cpages) {
1573 		for (i = 0; i < dic->nr_cpages; i++) {
1574 			if (!dic->cpages[i])
1575 				continue;
1576 			f2fs_compress_free_page(dic->cpages[i]);
1577 		}
1578 		page_array_free(dic->inode, dic->cpages, dic->nr_cpages);
1579 	}
1580 
1581 	page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
1582 	kmem_cache_free(dic_entry_slab, dic);
1583 }
1584 
1585 static void f2fs_put_dic(struct decompress_io_ctx *dic)
1586 {
1587 	if (refcount_dec_and_test(&dic->refcnt))
1588 		f2fs_free_dic(dic);
1589 }
1590 
1591 /*
1592  * Update and unlock the cluster's pagecache pages, and release the reference to
1593  * the decompress_io_ctx that was being held for I/O completion.
1594  */
1595 static void __f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1596 {
1597 	int i;
1598 
1599 	for (i = 0; i < dic->cluster_size; i++) {
1600 		struct page *rpage = dic->rpages[i];
1601 
1602 		if (!rpage)
1603 			continue;
1604 
1605 		/* PG_error was set if verity failed. */
1606 		if (failed || PageError(rpage)) {
1607 			ClearPageUptodate(rpage);
1608 			/* will re-read again later */
1609 			ClearPageError(rpage);
1610 		} else {
1611 			SetPageUptodate(rpage);
1612 		}
1613 		unlock_page(rpage);
1614 	}
1615 
1616 	f2fs_put_dic(dic);
1617 }
1618 
1619 static void f2fs_verify_cluster(struct work_struct *work)
1620 {
1621 	struct decompress_io_ctx *dic =
1622 		container_of(work, struct decompress_io_ctx, verity_work);
1623 	int i;
1624 
1625 	/* Verify the cluster's decompressed pages with fs-verity. */
1626 	for (i = 0; i < dic->cluster_size; i++) {
1627 		struct page *rpage = dic->rpages[i];
1628 
1629 		if (rpage && !fsverity_verify_page(rpage))
1630 			SetPageError(rpage);
1631 	}
1632 
1633 	__f2fs_decompress_end_io(dic, false);
1634 }
1635 
1636 /*
1637  * This is called when a compressed cluster has been decompressed
1638  * (or failed to be read and/or decompressed).
1639  */
1640 void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1641 {
1642 	if (!failed && dic->need_verity) {
1643 		/*
1644 		 * Note that to avoid deadlocks, the verity work can't be done
1645 		 * on the decompression workqueue.  This is because verifying
1646 		 * the data pages can involve reading metadata pages from the
1647 		 * file, and these metadata pages may be compressed.
1648 		 */
1649 		INIT_WORK(&dic->verity_work, f2fs_verify_cluster);
1650 		fsverity_enqueue_verify_work(&dic->verity_work);
1651 	} else {
1652 		__f2fs_decompress_end_io(dic, failed);
1653 	}
1654 }
1655 
1656 /*
1657  * Put a reference to a compressed page's decompress_io_ctx.
1658  *
1659  * This is called when the page is no longer needed and can be freed.
1660  */
1661 void f2fs_put_page_dic(struct page *page)
1662 {
1663 	struct decompress_io_ctx *dic =
1664 			(struct decompress_io_ctx *)page_private(page);
1665 
1666 	f2fs_put_dic(dic);
1667 }
1668 
1669 const struct address_space_operations f2fs_compress_aops = {
1670 	.releasepage = f2fs_release_page,
1671 	.invalidatepage = f2fs_invalidate_page,
1672 };
1673 
1674 struct address_space *COMPRESS_MAPPING(struct f2fs_sb_info *sbi)
1675 {
1676 	return sbi->compress_inode->i_mapping;
1677 }
1678 
1679 void f2fs_invalidate_compress_page(struct f2fs_sb_info *sbi, block_t blkaddr)
1680 {
1681 	if (!sbi->compress_inode)
1682 		return;
1683 	invalidate_mapping_pages(COMPRESS_MAPPING(sbi), blkaddr, blkaddr);
1684 }
1685 
1686 void f2fs_cache_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
1687 						nid_t ino, block_t blkaddr)
1688 {
1689 	struct page *cpage;
1690 	int ret;
1691 
1692 	if (!test_opt(sbi, COMPRESS_CACHE))
1693 		return;
1694 
1695 	if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
1696 		return;
1697 
1698 	if (!f2fs_available_free_memory(sbi, COMPRESS_PAGE))
1699 		return;
1700 
1701 	cpage = find_get_page(COMPRESS_MAPPING(sbi), blkaddr);
1702 	if (cpage) {
1703 		f2fs_put_page(cpage, 0);
1704 		return;
1705 	}
1706 
1707 	cpage = alloc_page(__GFP_NOWARN | __GFP_IO);
1708 	if (!cpage)
1709 		return;
1710 
1711 	ret = add_to_page_cache_lru(cpage, COMPRESS_MAPPING(sbi),
1712 						blkaddr, GFP_NOFS);
1713 	if (ret) {
1714 		f2fs_put_page(cpage, 0);
1715 		return;
1716 	}
1717 
1718 	set_page_private_data(cpage, ino);
1719 
1720 	if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
1721 		goto out;
1722 
1723 	memcpy(page_address(cpage), page_address(page), PAGE_SIZE);
1724 	SetPageUptodate(cpage);
1725 out:
1726 	f2fs_put_page(cpage, 1);
1727 }
1728 
1729 bool f2fs_load_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
1730 								block_t blkaddr)
1731 {
1732 	struct page *cpage;
1733 	bool hitted = false;
1734 
1735 	if (!test_opt(sbi, COMPRESS_CACHE))
1736 		return false;
1737 
1738 	cpage = f2fs_pagecache_get_page(COMPRESS_MAPPING(sbi),
1739 				blkaddr, FGP_LOCK | FGP_NOWAIT, GFP_NOFS);
1740 	if (cpage) {
1741 		if (PageUptodate(cpage)) {
1742 			atomic_inc(&sbi->compress_page_hit);
1743 			memcpy(page_address(page),
1744 				page_address(cpage), PAGE_SIZE);
1745 			hitted = true;
1746 		}
1747 		f2fs_put_page(cpage, 1);
1748 	}
1749 
1750 	return hitted;
1751 }
1752 
1753 void f2fs_invalidate_compress_pages(struct f2fs_sb_info *sbi, nid_t ino)
1754 {
1755 	struct address_space *mapping = sbi->compress_inode->i_mapping;
1756 	struct pagevec pvec;
1757 	pgoff_t index = 0;
1758 	pgoff_t end = MAX_BLKADDR(sbi);
1759 
1760 	if (!mapping->nrpages)
1761 		return;
1762 
1763 	pagevec_init(&pvec);
1764 
1765 	do {
1766 		unsigned int nr_pages;
1767 		int i;
1768 
1769 		nr_pages = pagevec_lookup_range(&pvec, mapping,
1770 						&index, end - 1);
1771 		if (!nr_pages)
1772 			break;
1773 
1774 		for (i = 0; i < nr_pages; i++) {
1775 			struct page *page = pvec.pages[i];
1776 
1777 			if (page->index > end)
1778 				break;
1779 
1780 			lock_page(page);
1781 			if (page->mapping != mapping) {
1782 				unlock_page(page);
1783 				continue;
1784 			}
1785 
1786 			if (ino != get_page_private_data(page)) {
1787 				unlock_page(page);
1788 				continue;
1789 			}
1790 
1791 			generic_error_remove_page(mapping, page);
1792 			unlock_page(page);
1793 		}
1794 		pagevec_release(&pvec);
1795 		cond_resched();
1796 	} while (index < end);
1797 }
1798 
1799 int f2fs_init_compress_inode(struct f2fs_sb_info *sbi)
1800 {
1801 	struct inode *inode;
1802 
1803 	if (!test_opt(sbi, COMPRESS_CACHE))
1804 		return 0;
1805 
1806 	inode = f2fs_iget(sbi->sb, F2FS_COMPRESS_INO(sbi));
1807 	if (IS_ERR(inode))
1808 		return PTR_ERR(inode);
1809 	sbi->compress_inode = inode;
1810 
1811 	sbi->compress_percent = COMPRESS_PERCENT;
1812 	sbi->compress_watermark = COMPRESS_WATERMARK;
1813 
1814 	atomic_set(&sbi->compress_page_hit, 0);
1815 
1816 	return 0;
1817 }
1818 
1819 void f2fs_destroy_compress_inode(struct f2fs_sb_info *sbi)
1820 {
1821 	if (!sbi->compress_inode)
1822 		return;
1823 	iput(sbi->compress_inode);
1824 	sbi->compress_inode = NULL;
1825 }
1826 
1827 int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi)
1828 {
1829 	dev_t dev = sbi->sb->s_bdev->bd_dev;
1830 	char slab_name[32];
1831 
1832 	sprintf(slab_name, "f2fs_page_array_entry-%u:%u", MAJOR(dev), MINOR(dev));
1833 
1834 	sbi->page_array_slab_size = sizeof(struct page *) <<
1835 					F2FS_OPTION(sbi).compress_log_size;
1836 
1837 	sbi->page_array_slab = f2fs_kmem_cache_create(slab_name,
1838 					sbi->page_array_slab_size);
1839 	if (!sbi->page_array_slab)
1840 		return -ENOMEM;
1841 	return 0;
1842 }
1843 
1844 void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi)
1845 {
1846 	kmem_cache_destroy(sbi->page_array_slab);
1847 }
1848 
1849 static int __init f2fs_init_cic_cache(void)
1850 {
1851 	cic_entry_slab = f2fs_kmem_cache_create("f2fs_cic_entry",
1852 					sizeof(struct compress_io_ctx));
1853 	if (!cic_entry_slab)
1854 		return -ENOMEM;
1855 	return 0;
1856 }
1857 
1858 static void f2fs_destroy_cic_cache(void)
1859 {
1860 	kmem_cache_destroy(cic_entry_slab);
1861 }
1862 
1863 static int __init f2fs_init_dic_cache(void)
1864 {
1865 	dic_entry_slab = f2fs_kmem_cache_create("f2fs_dic_entry",
1866 					sizeof(struct decompress_io_ctx));
1867 	if (!dic_entry_slab)
1868 		return -ENOMEM;
1869 	return 0;
1870 }
1871 
1872 static void f2fs_destroy_dic_cache(void)
1873 {
1874 	kmem_cache_destroy(dic_entry_slab);
1875 }
1876 
1877 int __init f2fs_init_compress_cache(void)
1878 {
1879 	int err;
1880 
1881 	err = f2fs_init_cic_cache();
1882 	if (err)
1883 		goto out;
1884 	err = f2fs_init_dic_cache();
1885 	if (err)
1886 		goto free_cic;
1887 	return 0;
1888 free_cic:
1889 	f2fs_destroy_cic_cache();
1890 out:
1891 	return -ENOMEM;
1892 }
1893 
1894 void f2fs_destroy_compress_cache(void)
1895 {
1896 	f2fs_destroy_dic_cache();
1897 	f2fs_destroy_cic_cache();
1898 }
1899