xref: /openbmc/linux/fs/f2fs/compress.c (revision 5b448065)
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 
16 #include "f2fs.h"
17 #include "node.h"
18 #include <trace/events/f2fs.h>
19 
20 static struct kmem_cache *cic_entry_slab;
21 static struct kmem_cache *dic_entry_slab;
22 
23 static void *page_array_alloc(struct inode *inode, int nr)
24 {
25 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
26 	unsigned int size = sizeof(struct page *) * nr;
27 
28 	if (likely(size <= sbi->page_array_slab_size))
29 		return kmem_cache_zalloc(sbi->page_array_slab, GFP_NOFS);
30 	return f2fs_kzalloc(sbi, size, GFP_NOFS);
31 }
32 
33 static void page_array_free(struct inode *inode, void *pages, int nr)
34 {
35 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
36 	unsigned int size = sizeof(struct page *) * nr;
37 
38 	if (!pages)
39 		return;
40 
41 	if (likely(size <= sbi->page_array_slab_size))
42 		kmem_cache_free(sbi->page_array_slab, pages);
43 	else
44 		kfree(pages);
45 }
46 
47 struct f2fs_compress_ops {
48 	int (*init_compress_ctx)(struct compress_ctx *cc);
49 	void (*destroy_compress_ctx)(struct compress_ctx *cc);
50 	int (*compress_pages)(struct compress_ctx *cc);
51 	int (*init_decompress_ctx)(struct decompress_io_ctx *dic);
52 	void (*destroy_decompress_ctx)(struct decompress_io_ctx *dic);
53 	int (*decompress_pages)(struct decompress_io_ctx *dic);
54 };
55 
56 static unsigned int offset_in_cluster(struct compress_ctx *cc, pgoff_t index)
57 {
58 	return index & (cc->cluster_size - 1);
59 }
60 
61 static pgoff_t cluster_idx(struct compress_ctx *cc, pgoff_t index)
62 {
63 	return index >> cc->log_cluster_size;
64 }
65 
66 static pgoff_t start_idx_of_cluster(struct compress_ctx *cc)
67 {
68 	return cc->cluster_idx << cc->log_cluster_size;
69 }
70 
71 bool f2fs_is_compressed_page(struct page *page)
72 {
73 	if (!PagePrivate(page))
74 		return false;
75 	if (!page_private(page))
76 		return false;
77 	if (IS_ATOMIC_WRITTEN_PAGE(page) || IS_DUMMY_WRITTEN_PAGE(page))
78 		return false;
79 
80 	f2fs_bug_on(F2FS_M_SB(page->mapping),
81 		*((u32 *)page_private(page)) != F2FS_COMPRESSED_PAGE_MAGIC);
82 	return true;
83 }
84 
85 static void f2fs_set_compressed_page(struct page *page,
86 		struct inode *inode, pgoff_t index, void *data)
87 {
88 	SetPagePrivate(page);
89 	set_page_private(page, (unsigned long)data);
90 
91 	/* i_crypto_info and iv index */
92 	page->index = index;
93 	page->mapping = inode->i_mapping;
94 }
95 
96 static void f2fs_drop_rpages(struct compress_ctx *cc, int len, bool unlock)
97 {
98 	int i;
99 
100 	for (i = 0; i < len; i++) {
101 		if (!cc->rpages[i])
102 			continue;
103 		if (unlock)
104 			unlock_page(cc->rpages[i]);
105 		else
106 			put_page(cc->rpages[i]);
107 	}
108 }
109 
110 static void f2fs_put_rpages(struct compress_ctx *cc)
111 {
112 	f2fs_drop_rpages(cc, cc->cluster_size, false);
113 }
114 
115 static void f2fs_unlock_rpages(struct compress_ctx *cc, int len)
116 {
117 	f2fs_drop_rpages(cc, len, true);
118 }
119 
120 static void f2fs_put_rpages_wbc(struct compress_ctx *cc,
121 		struct writeback_control *wbc, bool redirty, int unlock)
122 {
123 	unsigned int i;
124 
125 	for (i = 0; i < cc->cluster_size; i++) {
126 		if (!cc->rpages[i])
127 			continue;
128 		if (redirty)
129 			redirty_page_for_writepage(wbc, cc->rpages[i]);
130 		f2fs_put_page(cc->rpages[i], unlock);
131 	}
132 }
133 
134 struct page *f2fs_compress_control_page(struct page *page)
135 {
136 	return ((struct compress_io_ctx *)page_private(page))->rpages[0];
137 }
138 
139 int f2fs_init_compress_ctx(struct compress_ctx *cc)
140 {
141 	if (cc->rpages)
142 		return 0;
143 
144 	cc->rpages = page_array_alloc(cc->inode, cc->cluster_size);
145 	return cc->rpages ? 0 : -ENOMEM;
146 }
147 
148 void f2fs_destroy_compress_ctx(struct compress_ctx *cc, bool reuse)
149 {
150 	page_array_free(cc->inode, cc->rpages, cc->cluster_size);
151 	cc->rpages = NULL;
152 	cc->nr_rpages = 0;
153 	cc->nr_cpages = 0;
154 	if (!reuse)
155 		cc->cluster_idx = NULL_CLUSTER;
156 }
157 
158 void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct page *page)
159 {
160 	unsigned int cluster_ofs;
161 
162 	if (!f2fs_cluster_can_merge_page(cc, page->index))
163 		f2fs_bug_on(F2FS_I_SB(cc->inode), 1);
164 
165 	cluster_ofs = offset_in_cluster(cc, page->index);
166 	cc->rpages[cluster_ofs] = page;
167 	cc->nr_rpages++;
168 	cc->cluster_idx = cluster_idx(cc, page->index);
169 }
170 
171 #ifdef CONFIG_F2FS_FS_LZO
172 static int lzo_init_compress_ctx(struct compress_ctx *cc)
173 {
174 	cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
175 				LZO1X_MEM_COMPRESS, GFP_NOFS);
176 	if (!cc->private)
177 		return -ENOMEM;
178 
179 	cc->clen = lzo1x_worst_compress(PAGE_SIZE << cc->log_cluster_size);
180 	return 0;
181 }
182 
183 static void lzo_destroy_compress_ctx(struct compress_ctx *cc)
184 {
185 	kvfree(cc->private);
186 	cc->private = NULL;
187 }
188 
189 static int lzo_compress_pages(struct compress_ctx *cc)
190 {
191 	int ret;
192 
193 	ret = lzo1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
194 					&cc->clen, cc->private);
195 	if (ret != LZO_E_OK) {
196 		printk_ratelimited("%sF2FS-fs (%s): lzo compress failed, ret:%d\n",
197 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
198 		return -EIO;
199 	}
200 	return 0;
201 }
202 
203 static int lzo_decompress_pages(struct decompress_io_ctx *dic)
204 {
205 	int ret;
206 
207 	ret = lzo1x_decompress_safe(dic->cbuf->cdata, dic->clen,
208 						dic->rbuf, &dic->rlen);
209 	if (ret != LZO_E_OK) {
210 		printk_ratelimited("%sF2FS-fs (%s): lzo decompress failed, ret:%d\n",
211 				KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
212 		return -EIO;
213 	}
214 
215 	if (dic->rlen != PAGE_SIZE << dic->log_cluster_size) {
216 		printk_ratelimited("%sF2FS-fs (%s): lzo invalid rlen:%zu, "
217 					"expected:%lu\n", KERN_ERR,
218 					F2FS_I_SB(dic->inode)->sb->s_id,
219 					dic->rlen,
220 					PAGE_SIZE << dic->log_cluster_size);
221 		return -EIO;
222 	}
223 	return 0;
224 }
225 
226 static const struct f2fs_compress_ops f2fs_lzo_ops = {
227 	.init_compress_ctx	= lzo_init_compress_ctx,
228 	.destroy_compress_ctx	= lzo_destroy_compress_ctx,
229 	.compress_pages		= lzo_compress_pages,
230 	.decompress_pages	= lzo_decompress_pages,
231 };
232 #endif
233 
234 #ifdef CONFIG_F2FS_FS_LZ4
235 static int lz4_init_compress_ctx(struct compress_ctx *cc)
236 {
237 	unsigned int size = LZ4_MEM_COMPRESS;
238 
239 #ifdef CONFIG_F2FS_FS_LZ4HC
240 	if (F2FS_I(cc->inode)->i_compress_flag >> COMPRESS_LEVEL_OFFSET)
241 		size = LZ4HC_MEM_COMPRESS;
242 #endif
243 
244 	cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), size, GFP_NOFS);
245 	if (!cc->private)
246 		return -ENOMEM;
247 
248 	/*
249 	 * we do not change cc->clen to LZ4_compressBound(inputsize) to
250 	 * adapt worst compress case, because lz4 compressor can handle
251 	 * output budget properly.
252 	 */
253 	cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
254 	return 0;
255 }
256 
257 static void lz4_destroy_compress_ctx(struct compress_ctx *cc)
258 {
259 	kvfree(cc->private);
260 	cc->private = NULL;
261 }
262 
263 #ifdef CONFIG_F2FS_FS_LZ4HC
264 static int lz4hc_compress_pages(struct compress_ctx *cc)
265 {
266 	unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
267 						COMPRESS_LEVEL_OFFSET;
268 	int len;
269 
270 	if (level)
271 		len = LZ4_compress_HC(cc->rbuf, cc->cbuf->cdata, cc->rlen,
272 					cc->clen, level, cc->private);
273 	else
274 		len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
275 						cc->clen, cc->private);
276 	if (!len)
277 		return -EAGAIN;
278 
279 	cc->clen = len;
280 	return 0;
281 }
282 #endif
283 
284 static int lz4_compress_pages(struct compress_ctx *cc)
285 {
286 	int len;
287 
288 #ifdef CONFIG_F2FS_FS_LZ4HC
289 	return lz4hc_compress_pages(cc);
290 #endif
291 	len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
292 						cc->clen, cc->private);
293 	if (!len)
294 		return -EAGAIN;
295 
296 	cc->clen = len;
297 	return 0;
298 }
299 
300 static int lz4_decompress_pages(struct decompress_io_ctx *dic)
301 {
302 	int ret;
303 
304 	ret = LZ4_decompress_safe(dic->cbuf->cdata, dic->rbuf,
305 						dic->clen, dic->rlen);
306 	if (ret < 0) {
307 		printk_ratelimited("%sF2FS-fs (%s): lz4 decompress failed, ret:%d\n",
308 				KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
309 		return -EIO;
310 	}
311 
312 	if (ret != PAGE_SIZE << dic->log_cluster_size) {
313 		printk_ratelimited("%sF2FS-fs (%s): lz4 invalid rlen:%zu, "
314 					"expected:%lu\n", KERN_ERR,
315 					F2FS_I_SB(dic->inode)->sb->s_id,
316 					dic->rlen,
317 					PAGE_SIZE << dic->log_cluster_size);
318 		return -EIO;
319 	}
320 	return 0;
321 }
322 
323 static const struct f2fs_compress_ops f2fs_lz4_ops = {
324 	.init_compress_ctx	= lz4_init_compress_ctx,
325 	.destroy_compress_ctx	= lz4_destroy_compress_ctx,
326 	.compress_pages		= lz4_compress_pages,
327 	.decompress_pages	= lz4_decompress_pages,
328 };
329 #endif
330 
331 #ifdef CONFIG_F2FS_FS_ZSTD
332 #define F2FS_ZSTD_DEFAULT_CLEVEL	1
333 
334 static int zstd_init_compress_ctx(struct compress_ctx *cc)
335 {
336 	ZSTD_parameters params;
337 	ZSTD_CStream *stream;
338 	void *workspace;
339 	unsigned int workspace_size;
340 	unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
341 						COMPRESS_LEVEL_OFFSET;
342 
343 	if (!level)
344 		level = F2FS_ZSTD_DEFAULT_CLEVEL;
345 
346 	params = ZSTD_getParams(level, cc->rlen, 0);
347 	workspace_size = ZSTD_CStreamWorkspaceBound(params.cParams);
348 
349 	workspace = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
350 					workspace_size, GFP_NOFS);
351 	if (!workspace)
352 		return -ENOMEM;
353 
354 	stream = ZSTD_initCStream(params, 0, workspace, workspace_size);
355 	if (!stream) {
356 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initCStream failed\n",
357 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
358 				__func__);
359 		kvfree(workspace);
360 		return -EIO;
361 	}
362 
363 	cc->private = workspace;
364 	cc->private2 = stream;
365 
366 	cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
367 	return 0;
368 }
369 
370 static void zstd_destroy_compress_ctx(struct compress_ctx *cc)
371 {
372 	kvfree(cc->private);
373 	cc->private = NULL;
374 	cc->private2 = NULL;
375 }
376 
377 static int zstd_compress_pages(struct compress_ctx *cc)
378 {
379 	ZSTD_CStream *stream = cc->private2;
380 	ZSTD_inBuffer inbuf;
381 	ZSTD_outBuffer outbuf;
382 	int src_size = cc->rlen;
383 	int dst_size = src_size - PAGE_SIZE - COMPRESS_HEADER_SIZE;
384 	int ret;
385 
386 	inbuf.pos = 0;
387 	inbuf.src = cc->rbuf;
388 	inbuf.size = src_size;
389 
390 	outbuf.pos = 0;
391 	outbuf.dst = cc->cbuf->cdata;
392 	outbuf.size = dst_size;
393 
394 	ret = ZSTD_compressStream(stream, &outbuf, &inbuf);
395 	if (ZSTD_isError(ret)) {
396 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
397 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
398 				__func__, ZSTD_getErrorCode(ret));
399 		return -EIO;
400 	}
401 
402 	ret = ZSTD_endStream(stream, &outbuf);
403 	if (ZSTD_isError(ret)) {
404 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_endStream returned %d\n",
405 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
406 				__func__, ZSTD_getErrorCode(ret));
407 		return -EIO;
408 	}
409 
410 	/*
411 	 * there is compressed data remained in intermediate buffer due to
412 	 * no more space in cbuf.cdata
413 	 */
414 	if (ret)
415 		return -EAGAIN;
416 
417 	cc->clen = outbuf.pos;
418 	return 0;
419 }
420 
421 static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
422 {
423 	ZSTD_DStream *stream;
424 	void *workspace;
425 	unsigned int workspace_size;
426 	unsigned int max_window_size =
427 			MAX_COMPRESS_WINDOW_SIZE(dic->log_cluster_size);
428 
429 	workspace_size = ZSTD_DStreamWorkspaceBound(max_window_size);
430 
431 	workspace = f2fs_kvmalloc(F2FS_I_SB(dic->inode),
432 					workspace_size, GFP_NOFS);
433 	if (!workspace)
434 		return -ENOMEM;
435 
436 	stream = ZSTD_initDStream(max_window_size, workspace, workspace_size);
437 	if (!stream) {
438 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initDStream failed\n",
439 				KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
440 				__func__);
441 		kvfree(workspace);
442 		return -EIO;
443 	}
444 
445 	dic->private = workspace;
446 	dic->private2 = stream;
447 
448 	return 0;
449 }
450 
451 static void zstd_destroy_decompress_ctx(struct decompress_io_ctx *dic)
452 {
453 	kvfree(dic->private);
454 	dic->private = NULL;
455 	dic->private2 = NULL;
456 }
457 
458 static int zstd_decompress_pages(struct decompress_io_ctx *dic)
459 {
460 	ZSTD_DStream *stream = dic->private2;
461 	ZSTD_inBuffer inbuf;
462 	ZSTD_outBuffer outbuf;
463 	int ret;
464 
465 	inbuf.pos = 0;
466 	inbuf.src = dic->cbuf->cdata;
467 	inbuf.size = dic->clen;
468 
469 	outbuf.pos = 0;
470 	outbuf.dst = dic->rbuf;
471 	outbuf.size = dic->rlen;
472 
473 	ret = ZSTD_decompressStream(stream, &outbuf, &inbuf);
474 	if (ZSTD_isError(ret)) {
475 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
476 				KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
477 				__func__, ZSTD_getErrorCode(ret));
478 		return -EIO;
479 	}
480 
481 	if (dic->rlen != outbuf.pos) {
482 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD invalid rlen:%zu, "
483 				"expected:%lu\n", KERN_ERR,
484 				F2FS_I_SB(dic->inode)->sb->s_id,
485 				__func__, dic->rlen,
486 				PAGE_SIZE << dic->log_cluster_size);
487 		return -EIO;
488 	}
489 
490 	return 0;
491 }
492 
493 static const struct f2fs_compress_ops f2fs_zstd_ops = {
494 	.init_compress_ctx	= zstd_init_compress_ctx,
495 	.destroy_compress_ctx	= zstd_destroy_compress_ctx,
496 	.compress_pages		= zstd_compress_pages,
497 	.init_decompress_ctx	= zstd_init_decompress_ctx,
498 	.destroy_decompress_ctx	= zstd_destroy_decompress_ctx,
499 	.decompress_pages	= zstd_decompress_pages,
500 };
501 #endif
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 
528 static const struct f2fs_compress_ops *f2fs_cops[COMPRESS_MAX] = {
529 #ifdef CONFIG_F2FS_FS_LZO
530 	&f2fs_lzo_ops,
531 #else
532 	NULL,
533 #endif
534 #ifdef CONFIG_F2FS_FS_LZ4
535 	&f2fs_lz4_ops,
536 #else
537 	NULL,
538 #endif
539 #ifdef CONFIG_F2FS_FS_ZSTD
540 	&f2fs_zstd_ops,
541 #else
542 	NULL,
543 #endif
544 #if defined(CONFIG_F2FS_FS_LZO) && defined(CONFIG_F2FS_FS_LZORLE)
545 	&f2fs_lzorle_ops,
546 #else
547 	NULL,
548 #endif
549 };
550 
551 bool f2fs_is_compress_backend_ready(struct inode *inode)
552 {
553 	if (!f2fs_compressed_file(inode))
554 		return true;
555 	return f2fs_cops[F2FS_I(inode)->i_compress_algorithm];
556 }
557 
558 static mempool_t *compress_page_pool;
559 static int num_compress_pages = 512;
560 module_param(num_compress_pages, uint, 0444);
561 MODULE_PARM_DESC(num_compress_pages,
562 		"Number of intermediate compress pages to preallocate");
563 
564 int f2fs_init_compress_mempool(void)
565 {
566 	compress_page_pool = mempool_create_page_pool(num_compress_pages, 0);
567 	if (!compress_page_pool)
568 		return -ENOMEM;
569 
570 	return 0;
571 }
572 
573 void f2fs_destroy_compress_mempool(void)
574 {
575 	mempool_destroy(compress_page_pool);
576 }
577 
578 static struct page *f2fs_compress_alloc_page(void)
579 {
580 	struct page *page;
581 
582 	page = mempool_alloc(compress_page_pool, GFP_NOFS);
583 	lock_page(page);
584 
585 	return page;
586 }
587 
588 static void f2fs_compress_free_page(struct page *page)
589 {
590 	if (!page)
591 		return;
592 	set_page_private(page, (unsigned long)NULL);
593 	ClearPagePrivate(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 static 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 {
842 	struct decompress_io_ctx *dic =
843 			(struct decompress_io_ctx *)page_private(page);
844 	struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
845 
846 	dec_page_count(sbi, F2FS_RD_DATA);
847 
848 	if (failed)
849 		WRITE_ONCE(dic->failed, true);
850 
851 	if (atomic_dec_and_test(&dic->remaining_pages))
852 		f2fs_decompress_cluster(dic);
853 }
854 
855 static bool is_page_in_cluster(struct compress_ctx *cc, pgoff_t index)
856 {
857 	if (cc->cluster_idx == NULL_CLUSTER)
858 		return true;
859 	return cc->cluster_idx == cluster_idx(cc, index);
860 }
861 
862 bool f2fs_cluster_is_empty(struct compress_ctx *cc)
863 {
864 	return cc->nr_rpages == 0;
865 }
866 
867 static bool f2fs_cluster_is_full(struct compress_ctx *cc)
868 {
869 	return cc->cluster_size == cc->nr_rpages;
870 }
871 
872 bool f2fs_cluster_can_merge_page(struct compress_ctx *cc, pgoff_t index)
873 {
874 	if (f2fs_cluster_is_empty(cc))
875 		return true;
876 	return is_page_in_cluster(cc, index);
877 }
878 
879 static bool __cluster_may_compress(struct compress_ctx *cc)
880 {
881 	loff_t i_size = i_size_read(cc->inode);
882 	unsigned nr_pages = DIV_ROUND_UP(i_size, PAGE_SIZE);
883 	int i;
884 
885 	for (i = 0; i < cc->cluster_size; i++) {
886 		struct page *page = cc->rpages[i];
887 
888 		f2fs_bug_on(F2FS_I_SB(cc->inode), !page);
889 
890 		/* beyond EOF */
891 		if (page->index >= nr_pages)
892 			return false;
893 	}
894 	return true;
895 }
896 
897 static int __f2fs_cluster_blocks(struct compress_ctx *cc, bool compr)
898 {
899 	struct dnode_of_data dn;
900 	int ret;
901 
902 	set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
903 	ret = f2fs_get_dnode_of_data(&dn, start_idx_of_cluster(cc),
904 							LOOKUP_NODE);
905 	if (ret) {
906 		if (ret == -ENOENT)
907 			ret = 0;
908 		goto fail;
909 	}
910 
911 	if (dn.data_blkaddr == COMPRESS_ADDR) {
912 		int i;
913 
914 		ret = 1;
915 		for (i = 1; i < cc->cluster_size; i++) {
916 			block_t blkaddr;
917 
918 			blkaddr = data_blkaddr(dn.inode,
919 					dn.node_page, dn.ofs_in_node + i);
920 			if (compr) {
921 				if (__is_valid_data_blkaddr(blkaddr))
922 					ret++;
923 			} else {
924 				if (blkaddr != NULL_ADDR)
925 					ret++;
926 			}
927 		}
928 	}
929 fail:
930 	f2fs_put_dnode(&dn);
931 	return ret;
932 }
933 
934 /* return # of compressed blocks in compressed cluster */
935 static int f2fs_compressed_blocks(struct compress_ctx *cc)
936 {
937 	return __f2fs_cluster_blocks(cc, true);
938 }
939 
940 /* return # of valid blocks in compressed cluster */
941 static int f2fs_cluster_blocks(struct compress_ctx *cc)
942 {
943 	return __f2fs_cluster_blocks(cc, false);
944 }
945 
946 int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index)
947 {
948 	struct compress_ctx cc = {
949 		.inode = inode,
950 		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
951 		.cluster_size = F2FS_I(inode)->i_cluster_size,
952 		.cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
953 	};
954 
955 	return f2fs_cluster_blocks(&cc);
956 }
957 
958 static bool cluster_may_compress(struct compress_ctx *cc)
959 {
960 	if (!f2fs_need_compress_data(cc->inode))
961 		return false;
962 	if (f2fs_is_atomic_file(cc->inode))
963 		return false;
964 	if (f2fs_is_mmap_file(cc->inode))
965 		return false;
966 	if (!f2fs_cluster_is_full(cc))
967 		return false;
968 	if (unlikely(f2fs_cp_error(F2FS_I_SB(cc->inode))))
969 		return false;
970 	return __cluster_may_compress(cc);
971 }
972 
973 static void set_cluster_writeback(struct compress_ctx *cc)
974 {
975 	int i;
976 
977 	for (i = 0; i < cc->cluster_size; i++) {
978 		if (cc->rpages[i])
979 			set_page_writeback(cc->rpages[i]);
980 	}
981 }
982 
983 static void set_cluster_dirty(struct compress_ctx *cc)
984 {
985 	int i;
986 
987 	for (i = 0; i < cc->cluster_size; i++)
988 		if (cc->rpages[i])
989 			set_page_dirty(cc->rpages[i]);
990 }
991 
992 static int prepare_compress_overwrite(struct compress_ctx *cc,
993 		struct page **pagep, pgoff_t index, void **fsdata)
994 {
995 	struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
996 	struct address_space *mapping = cc->inode->i_mapping;
997 	struct page *page;
998 	struct dnode_of_data dn;
999 	sector_t last_block_in_bio;
1000 	unsigned fgp_flag = FGP_LOCK | FGP_WRITE | FGP_CREAT;
1001 	pgoff_t start_idx = start_idx_of_cluster(cc);
1002 	int i, ret;
1003 	bool prealloc;
1004 
1005 retry:
1006 	ret = f2fs_cluster_blocks(cc);
1007 	if (ret <= 0)
1008 		return ret;
1009 
1010 	/* compressed case */
1011 	prealloc = (ret < cc->cluster_size);
1012 
1013 	ret = f2fs_init_compress_ctx(cc);
1014 	if (ret)
1015 		return ret;
1016 
1017 	/* keep page reference to avoid page reclaim */
1018 	for (i = 0; i < cc->cluster_size; i++) {
1019 		page = f2fs_pagecache_get_page(mapping, start_idx + i,
1020 							fgp_flag, GFP_NOFS);
1021 		if (!page) {
1022 			ret = -ENOMEM;
1023 			goto unlock_pages;
1024 		}
1025 
1026 		if (PageUptodate(page))
1027 			f2fs_put_page(page, 1);
1028 		else
1029 			f2fs_compress_ctx_add_page(cc, page);
1030 	}
1031 
1032 	if (!f2fs_cluster_is_empty(cc)) {
1033 		struct bio *bio = NULL;
1034 
1035 		ret = f2fs_read_multi_pages(cc, &bio, cc->cluster_size,
1036 					&last_block_in_bio, false, true);
1037 		f2fs_put_rpages(cc);
1038 		f2fs_destroy_compress_ctx(cc, true);
1039 		if (ret)
1040 			goto out;
1041 		if (bio)
1042 			f2fs_submit_bio(sbi, bio, DATA);
1043 
1044 		ret = f2fs_init_compress_ctx(cc);
1045 		if (ret)
1046 			goto out;
1047 	}
1048 
1049 	for (i = 0; i < cc->cluster_size; i++) {
1050 		f2fs_bug_on(sbi, cc->rpages[i]);
1051 
1052 		page = find_lock_page(mapping, start_idx + i);
1053 		if (!page) {
1054 			/* page can be truncated */
1055 			goto release_and_retry;
1056 		}
1057 
1058 		f2fs_wait_on_page_writeback(page, DATA, true, true);
1059 		f2fs_compress_ctx_add_page(cc, page);
1060 
1061 		if (!PageUptodate(page)) {
1062 release_and_retry:
1063 			f2fs_put_rpages(cc);
1064 			f2fs_unlock_rpages(cc, i + 1);
1065 			f2fs_destroy_compress_ctx(cc, true);
1066 			goto retry;
1067 		}
1068 	}
1069 
1070 	if (prealloc) {
1071 		f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
1072 
1073 		set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1074 
1075 		for (i = cc->cluster_size - 1; i > 0; i--) {
1076 			ret = f2fs_get_block(&dn, start_idx + i);
1077 			if (ret) {
1078 				i = cc->cluster_size;
1079 				break;
1080 			}
1081 
1082 			if (dn.data_blkaddr != NEW_ADDR)
1083 				break;
1084 		}
1085 
1086 		f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
1087 	}
1088 
1089 	if (likely(!ret)) {
1090 		*fsdata = cc->rpages;
1091 		*pagep = cc->rpages[offset_in_cluster(cc, index)];
1092 		return cc->cluster_size;
1093 	}
1094 
1095 unlock_pages:
1096 	f2fs_put_rpages(cc);
1097 	f2fs_unlock_rpages(cc, i);
1098 	f2fs_destroy_compress_ctx(cc, true);
1099 out:
1100 	return ret;
1101 }
1102 
1103 int f2fs_prepare_compress_overwrite(struct inode *inode,
1104 		struct page **pagep, pgoff_t index, void **fsdata)
1105 {
1106 	struct compress_ctx cc = {
1107 		.inode = inode,
1108 		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1109 		.cluster_size = F2FS_I(inode)->i_cluster_size,
1110 		.cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
1111 		.rpages = NULL,
1112 		.nr_rpages = 0,
1113 	};
1114 
1115 	return prepare_compress_overwrite(&cc, pagep, index, fsdata);
1116 }
1117 
1118 bool f2fs_compress_write_end(struct inode *inode, void *fsdata,
1119 					pgoff_t index, unsigned copied)
1120 
1121 {
1122 	struct compress_ctx cc = {
1123 		.inode = inode,
1124 		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1125 		.cluster_size = F2FS_I(inode)->i_cluster_size,
1126 		.rpages = fsdata,
1127 	};
1128 	bool first_index = (index == cc.rpages[0]->index);
1129 
1130 	if (copied)
1131 		set_cluster_dirty(&cc);
1132 
1133 	f2fs_put_rpages_wbc(&cc, NULL, false, 1);
1134 	f2fs_destroy_compress_ctx(&cc, false);
1135 
1136 	return first_index;
1137 }
1138 
1139 int f2fs_truncate_partial_cluster(struct inode *inode, u64 from, bool lock)
1140 {
1141 	void *fsdata = NULL;
1142 	struct page *pagep;
1143 	int log_cluster_size = F2FS_I(inode)->i_log_cluster_size;
1144 	pgoff_t start_idx = from >> (PAGE_SHIFT + log_cluster_size) <<
1145 							log_cluster_size;
1146 	int err;
1147 
1148 	err = f2fs_is_compressed_cluster(inode, start_idx);
1149 	if (err < 0)
1150 		return err;
1151 
1152 	/* truncate normal cluster */
1153 	if (!err)
1154 		return f2fs_do_truncate_blocks(inode, from, lock);
1155 
1156 	/* truncate compressed cluster */
1157 	err = f2fs_prepare_compress_overwrite(inode, &pagep,
1158 						start_idx, &fsdata);
1159 
1160 	/* should not be a normal cluster */
1161 	f2fs_bug_on(F2FS_I_SB(inode), err == 0);
1162 
1163 	if (err <= 0)
1164 		return err;
1165 
1166 	if (err > 0) {
1167 		struct page **rpages = fsdata;
1168 		int cluster_size = F2FS_I(inode)->i_cluster_size;
1169 		int i;
1170 
1171 		for (i = cluster_size - 1; i >= 0; i--) {
1172 			loff_t start = rpages[i]->index << PAGE_SHIFT;
1173 
1174 			if (from <= start) {
1175 				zero_user_segment(rpages[i], 0, PAGE_SIZE);
1176 			} else {
1177 				zero_user_segment(rpages[i], from - start,
1178 								PAGE_SIZE);
1179 				break;
1180 			}
1181 		}
1182 
1183 		f2fs_compress_write_end(inode, fsdata, start_idx, true);
1184 	}
1185 	return 0;
1186 }
1187 
1188 static int f2fs_write_compressed_pages(struct compress_ctx *cc,
1189 					int *submitted,
1190 					struct writeback_control *wbc,
1191 					enum iostat_type io_type)
1192 {
1193 	struct inode *inode = cc->inode;
1194 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1195 	struct f2fs_inode_info *fi = F2FS_I(inode);
1196 	struct f2fs_io_info fio = {
1197 		.sbi = sbi,
1198 		.ino = cc->inode->i_ino,
1199 		.type = DATA,
1200 		.op = REQ_OP_WRITE,
1201 		.op_flags = wbc_to_write_flags(wbc),
1202 		.old_blkaddr = NEW_ADDR,
1203 		.page = NULL,
1204 		.encrypted_page = NULL,
1205 		.compressed_page = NULL,
1206 		.submitted = false,
1207 		.io_type = io_type,
1208 		.io_wbc = wbc,
1209 		.encrypted = fscrypt_inode_uses_fs_layer_crypto(cc->inode),
1210 	};
1211 	struct dnode_of_data dn;
1212 	struct node_info ni;
1213 	struct compress_io_ctx *cic;
1214 	pgoff_t start_idx = start_idx_of_cluster(cc);
1215 	unsigned int last_index = cc->cluster_size - 1;
1216 	loff_t psize;
1217 	int i, err;
1218 
1219 	if (IS_NOQUOTA(inode)) {
1220 		/*
1221 		 * We need to wait for node_write to avoid block allocation during
1222 		 * checkpoint. This can only happen to quota writes which can cause
1223 		 * the below discard race condition.
1224 		 */
1225 		down_read(&sbi->node_write);
1226 	} else if (!f2fs_trylock_op(sbi)) {
1227 		goto out_free;
1228 	}
1229 
1230 	set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1231 
1232 	err = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
1233 	if (err)
1234 		goto out_unlock_op;
1235 
1236 	for (i = 0; i < cc->cluster_size; i++) {
1237 		if (data_blkaddr(dn.inode, dn.node_page,
1238 					dn.ofs_in_node + i) == NULL_ADDR)
1239 			goto out_put_dnode;
1240 	}
1241 
1242 	psize = (loff_t)(cc->rpages[last_index]->index + 1) << PAGE_SHIFT;
1243 
1244 	err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1245 	if (err)
1246 		goto out_put_dnode;
1247 
1248 	fio.version = ni.version;
1249 
1250 	cic = kmem_cache_zalloc(cic_entry_slab, GFP_NOFS);
1251 	if (!cic)
1252 		goto out_put_dnode;
1253 
1254 	cic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1255 	cic->inode = inode;
1256 	atomic_set(&cic->pending_pages, cc->nr_cpages);
1257 	cic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1258 	if (!cic->rpages)
1259 		goto out_put_cic;
1260 
1261 	cic->nr_rpages = cc->cluster_size;
1262 
1263 	for (i = 0; i < cc->nr_cpages; i++) {
1264 		f2fs_set_compressed_page(cc->cpages[i], inode,
1265 					cc->rpages[i + 1]->index, cic);
1266 		fio.compressed_page = cc->cpages[i];
1267 
1268 		fio.old_blkaddr = data_blkaddr(dn.inode, dn.node_page,
1269 						dn.ofs_in_node + i + 1);
1270 
1271 		/* wait for GCed page writeback via META_MAPPING */
1272 		f2fs_wait_on_block_writeback(inode, fio.old_blkaddr);
1273 
1274 		if (fio.encrypted) {
1275 			fio.page = cc->rpages[i + 1];
1276 			err = f2fs_encrypt_one_page(&fio);
1277 			if (err)
1278 				goto out_destroy_crypt;
1279 			cc->cpages[i] = fio.encrypted_page;
1280 		}
1281 	}
1282 
1283 	set_cluster_writeback(cc);
1284 
1285 	for (i = 0; i < cc->cluster_size; i++)
1286 		cic->rpages[i] = cc->rpages[i];
1287 
1288 	for (i = 0; i < cc->cluster_size; i++, dn.ofs_in_node++) {
1289 		block_t blkaddr;
1290 
1291 		blkaddr = f2fs_data_blkaddr(&dn);
1292 		fio.page = cc->rpages[i];
1293 		fio.old_blkaddr = blkaddr;
1294 
1295 		/* cluster header */
1296 		if (i == 0) {
1297 			if (blkaddr == COMPRESS_ADDR)
1298 				fio.compr_blocks++;
1299 			if (__is_valid_data_blkaddr(blkaddr))
1300 				f2fs_invalidate_blocks(sbi, blkaddr);
1301 			f2fs_update_data_blkaddr(&dn, COMPRESS_ADDR);
1302 			goto unlock_continue;
1303 		}
1304 
1305 		if (fio.compr_blocks && __is_valid_data_blkaddr(blkaddr))
1306 			fio.compr_blocks++;
1307 
1308 		if (i > cc->nr_cpages) {
1309 			if (__is_valid_data_blkaddr(blkaddr)) {
1310 				f2fs_invalidate_blocks(sbi, blkaddr);
1311 				f2fs_update_data_blkaddr(&dn, NEW_ADDR);
1312 			}
1313 			goto unlock_continue;
1314 		}
1315 
1316 		f2fs_bug_on(fio.sbi, blkaddr == NULL_ADDR);
1317 
1318 		if (fio.encrypted)
1319 			fio.encrypted_page = cc->cpages[i - 1];
1320 		else
1321 			fio.compressed_page = cc->cpages[i - 1];
1322 
1323 		cc->cpages[i - 1] = NULL;
1324 		f2fs_outplace_write_data(&dn, &fio);
1325 		(*submitted)++;
1326 unlock_continue:
1327 		inode_dec_dirty_pages(cc->inode);
1328 		unlock_page(fio.page);
1329 	}
1330 
1331 	if (fio.compr_blocks)
1332 		f2fs_i_compr_blocks_update(inode, fio.compr_blocks - 1, false);
1333 	f2fs_i_compr_blocks_update(inode, cc->nr_cpages, true);
1334 	add_compr_block_stat(inode, cc->nr_cpages);
1335 
1336 	set_inode_flag(cc->inode, FI_APPEND_WRITE);
1337 	if (cc->cluster_idx == 0)
1338 		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1339 
1340 	f2fs_put_dnode(&dn);
1341 	if (IS_NOQUOTA(inode))
1342 		up_read(&sbi->node_write);
1343 	else
1344 		f2fs_unlock_op(sbi);
1345 
1346 	spin_lock(&fi->i_size_lock);
1347 	if (fi->last_disk_size < psize)
1348 		fi->last_disk_size = psize;
1349 	spin_unlock(&fi->i_size_lock);
1350 
1351 	f2fs_put_rpages(cc);
1352 	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1353 	cc->cpages = NULL;
1354 	f2fs_destroy_compress_ctx(cc, false);
1355 	return 0;
1356 
1357 out_destroy_crypt:
1358 	page_array_free(cc->inode, cic->rpages, cc->cluster_size);
1359 
1360 	for (--i; i >= 0; i--)
1361 		fscrypt_finalize_bounce_page(&cc->cpages[i]);
1362 	for (i = 0; i < cc->nr_cpages; i++) {
1363 		if (!cc->cpages[i])
1364 			continue;
1365 		f2fs_compress_free_page(cc->cpages[i]);
1366 		cc->cpages[i] = NULL;
1367 	}
1368 out_put_cic:
1369 	kmem_cache_free(cic_entry_slab, cic);
1370 out_put_dnode:
1371 	f2fs_put_dnode(&dn);
1372 out_unlock_op:
1373 	if (IS_NOQUOTA(inode))
1374 		up_read(&sbi->node_write);
1375 	else
1376 		f2fs_unlock_op(sbi);
1377 out_free:
1378 	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1379 	cc->cpages = NULL;
1380 	return -EAGAIN;
1381 }
1382 
1383 void f2fs_compress_write_end_io(struct bio *bio, struct page *page)
1384 {
1385 	struct f2fs_sb_info *sbi = bio->bi_private;
1386 	struct compress_io_ctx *cic =
1387 			(struct compress_io_ctx *)page_private(page);
1388 	int i;
1389 
1390 	if (unlikely(bio->bi_status))
1391 		mapping_set_error(cic->inode->i_mapping, -EIO);
1392 
1393 	f2fs_compress_free_page(page);
1394 
1395 	dec_page_count(sbi, F2FS_WB_DATA);
1396 
1397 	if (atomic_dec_return(&cic->pending_pages))
1398 		return;
1399 
1400 	for (i = 0; i < cic->nr_rpages; i++) {
1401 		WARN_ON(!cic->rpages[i]);
1402 		clear_cold_data(cic->rpages[i]);
1403 		end_page_writeback(cic->rpages[i]);
1404 	}
1405 
1406 	page_array_free(cic->inode, cic->rpages, cic->nr_rpages);
1407 	kmem_cache_free(cic_entry_slab, cic);
1408 }
1409 
1410 static int f2fs_write_raw_pages(struct compress_ctx *cc,
1411 					int *submitted,
1412 					struct writeback_control *wbc,
1413 					enum iostat_type io_type)
1414 {
1415 	struct address_space *mapping = cc->inode->i_mapping;
1416 	int _submitted, compr_blocks, ret;
1417 	int i = -1, err = 0;
1418 
1419 	compr_blocks = f2fs_compressed_blocks(cc);
1420 	if (compr_blocks < 0) {
1421 		err = compr_blocks;
1422 		goto out_err;
1423 	}
1424 
1425 	for (i = 0; i < cc->cluster_size; i++) {
1426 		if (!cc->rpages[i])
1427 			continue;
1428 retry_write:
1429 		if (cc->rpages[i]->mapping != mapping) {
1430 			unlock_page(cc->rpages[i]);
1431 			continue;
1432 		}
1433 
1434 		BUG_ON(!PageLocked(cc->rpages[i]));
1435 
1436 		ret = f2fs_write_single_data_page(cc->rpages[i], &_submitted,
1437 						NULL, NULL, wbc, io_type,
1438 						compr_blocks, false);
1439 		if (ret) {
1440 			if (ret == AOP_WRITEPAGE_ACTIVATE) {
1441 				unlock_page(cc->rpages[i]);
1442 				ret = 0;
1443 			} else if (ret == -EAGAIN) {
1444 				/*
1445 				 * for quota file, just redirty left pages to
1446 				 * avoid deadlock caused by cluster update race
1447 				 * from foreground operation.
1448 				 */
1449 				if (IS_NOQUOTA(cc->inode)) {
1450 					err = 0;
1451 					goto out_err;
1452 				}
1453 				ret = 0;
1454 				cond_resched();
1455 				congestion_wait(BLK_RW_ASYNC,
1456 						DEFAULT_IO_TIMEOUT);
1457 				lock_page(cc->rpages[i]);
1458 
1459 				if (!PageDirty(cc->rpages[i])) {
1460 					unlock_page(cc->rpages[i]);
1461 					continue;
1462 				}
1463 
1464 				clear_page_dirty_for_io(cc->rpages[i]);
1465 				goto retry_write;
1466 			}
1467 			err = ret;
1468 			goto out_err;
1469 		}
1470 
1471 		*submitted += _submitted;
1472 	}
1473 
1474 	f2fs_balance_fs(F2FS_M_SB(mapping), true);
1475 
1476 	return 0;
1477 out_err:
1478 	for (++i; i < cc->cluster_size; i++) {
1479 		if (!cc->rpages[i])
1480 			continue;
1481 		redirty_page_for_writepage(wbc, cc->rpages[i]);
1482 		unlock_page(cc->rpages[i]);
1483 	}
1484 	return err;
1485 }
1486 
1487 int f2fs_write_multi_pages(struct compress_ctx *cc,
1488 					int *submitted,
1489 					struct writeback_control *wbc,
1490 					enum iostat_type io_type)
1491 {
1492 	int err;
1493 
1494 	*submitted = 0;
1495 	if (cluster_may_compress(cc)) {
1496 		err = f2fs_compress_pages(cc);
1497 		if (err == -EAGAIN) {
1498 			goto write;
1499 		} else if (err) {
1500 			f2fs_put_rpages_wbc(cc, wbc, true, 1);
1501 			goto destroy_out;
1502 		}
1503 
1504 		err = f2fs_write_compressed_pages(cc, submitted,
1505 							wbc, io_type);
1506 		if (!err)
1507 			return 0;
1508 		f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN);
1509 	}
1510 write:
1511 	f2fs_bug_on(F2FS_I_SB(cc->inode), *submitted);
1512 
1513 	err = f2fs_write_raw_pages(cc, submitted, wbc, io_type);
1514 	f2fs_put_rpages_wbc(cc, wbc, false, 0);
1515 destroy_out:
1516 	f2fs_destroy_compress_ctx(cc, false);
1517 	return err;
1518 }
1519 
1520 static void f2fs_free_dic(struct decompress_io_ctx *dic);
1521 
1522 struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
1523 {
1524 	struct decompress_io_ctx *dic;
1525 	pgoff_t start_idx = start_idx_of_cluster(cc);
1526 	int i;
1527 
1528 	dic = kmem_cache_zalloc(dic_entry_slab, GFP_NOFS);
1529 	if (!dic)
1530 		return ERR_PTR(-ENOMEM);
1531 
1532 	dic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1533 	if (!dic->rpages) {
1534 		kmem_cache_free(dic_entry_slab, dic);
1535 		return ERR_PTR(-ENOMEM);
1536 	}
1537 
1538 	dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1539 	dic->inode = cc->inode;
1540 	atomic_set(&dic->remaining_pages, cc->nr_cpages);
1541 	dic->cluster_idx = cc->cluster_idx;
1542 	dic->cluster_size = cc->cluster_size;
1543 	dic->log_cluster_size = cc->log_cluster_size;
1544 	dic->nr_cpages = cc->nr_cpages;
1545 	refcount_set(&dic->refcnt, 1);
1546 	dic->failed = false;
1547 	dic->need_verity = f2fs_need_verity(cc->inode, start_idx);
1548 
1549 	for (i = 0; i < dic->cluster_size; i++)
1550 		dic->rpages[i] = cc->rpages[i];
1551 	dic->nr_rpages = cc->cluster_size;
1552 
1553 	dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages);
1554 	if (!dic->cpages)
1555 		goto out_free;
1556 
1557 	for (i = 0; i < dic->nr_cpages; i++) {
1558 		struct page *page;
1559 
1560 		page = f2fs_compress_alloc_page();
1561 		if (!page)
1562 			goto out_free;
1563 
1564 		f2fs_set_compressed_page(page, cc->inode,
1565 					start_idx + i + 1, dic);
1566 		dic->cpages[i] = page;
1567 	}
1568 
1569 	return dic;
1570 
1571 out_free:
1572 	f2fs_free_dic(dic);
1573 	return ERR_PTR(-ENOMEM);
1574 }
1575 
1576 static void f2fs_free_dic(struct decompress_io_ctx *dic)
1577 {
1578 	int i;
1579 
1580 	if (dic->tpages) {
1581 		for (i = 0; i < dic->cluster_size; i++) {
1582 			if (dic->rpages[i])
1583 				continue;
1584 			if (!dic->tpages[i])
1585 				continue;
1586 			f2fs_compress_free_page(dic->tpages[i]);
1587 		}
1588 		page_array_free(dic->inode, dic->tpages, dic->cluster_size);
1589 	}
1590 
1591 	if (dic->cpages) {
1592 		for (i = 0; i < dic->nr_cpages; i++) {
1593 			if (!dic->cpages[i])
1594 				continue;
1595 			f2fs_compress_free_page(dic->cpages[i]);
1596 		}
1597 		page_array_free(dic->inode, dic->cpages, dic->nr_cpages);
1598 	}
1599 
1600 	page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
1601 	kmem_cache_free(dic_entry_slab, dic);
1602 }
1603 
1604 static void f2fs_put_dic(struct decompress_io_ctx *dic)
1605 {
1606 	if (refcount_dec_and_test(&dic->refcnt))
1607 		f2fs_free_dic(dic);
1608 }
1609 
1610 /*
1611  * Update and unlock the cluster's pagecache pages, and release the reference to
1612  * the decompress_io_ctx that was being held for I/O completion.
1613  */
1614 static void __f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1615 {
1616 	int i;
1617 
1618 	for (i = 0; i < dic->cluster_size; i++) {
1619 		struct page *rpage = dic->rpages[i];
1620 
1621 		if (!rpage)
1622 			continue;
1623 
1624 		/* PG_error was set if verity failed. */
1625 		if (failed || PageError(rpage)) {
1626 			ClearPageUptodate(rpage);
1627 			/* will re-read again later */
1628 			ClearPageError(rpage);
1629 		} else {
1630 			SetPageUptodate(rpage);
1631 		}
1632 		unlock_page(rpage);
1633 	}
1634 
1635 	f2fs_put_dic(dic);
1636 }
1637 
1638 static void f2fs_verify_cluster(struct work_struct *work)
1639 {
1640 	struct decompress_io_ctx *dic =
1641 		container_of(work, struct decompress_io_ctx, verity_work);
1642 	int i;
1643 
1644 	/* Verify the cluster's decompressed pages with fs-verity. */
1645 	for (i = 0; i < dic->cluster_size; i++) {
1646 		struct page *rpage = dic->rpages[i];
1647 
1648 		if (rpage && !fsverity_verify_page(rpage))
1649 			SetPageError(rpage);
1650 	}
1651 
1652 	__f2fs_decompress_end_io(dic, false);
1653 }
1654 
1655 /*
1656  * This is called when a compressed cluster has been decompressed
1657  * (or failed to be read and/or decompressed).
1658  */
1659 void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1660 {
1661 	if (!failed && dic->need_verity) {
1662 		/*
1663 		 * Note that to avoid deadlocks, the verity work can't be done
1664 		 * on the decompression workqueue.  This is because verifying
1665 		 * the data pages can involve reading metadata pages from the
1666 		 * file, and these metadata pages may be compressed.
1667 		 */
1668 		INIT_WORK(&dic->verity_work, f2fs_verify_cluster);
1669 		fsverity_enqueue_verify_work(&dic->verity_work);
1670 	} else {
1671 		__f2fs_decompress_end_io(dic, failed);
1672 	}
1673 }
1674 
1675 /*
1676  * Put a reference to a compressed page's decompress_io_ctx.
1677  *
1678  * This is called when the page is no longer needed and can be freed.
1679  */
1680 void f2fs_put_page_dic(struct page *page)
1681 {
1682 	struct decompress_io_ctx *dic =
1683 			(struct decompress_io_ctx *)page_private(page);
1684 
1685 	f2fs_put_dic(dic);
1686 }
1687 
1688 int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi)
1689 {
1690 	dev_t dev = sbi->sb->s_bdev->bd_dev;
1691 	char slab_name[32];
1692 
1693 	sprintf(slab_name, "f2fs_page_array_entry-%u:%u", MAJOR(dev), MINOR(dev));
1694 
1695 	sbi->page_array_slab_size = sizeof(struct page *) <<
1696 					F2FS_OPTION(sbi).compress_log_size;
1697 
1698 	sbi->page_array_slab = f2fs_kmem_cache_create(slab_name,
1699 					sbi->page_array_slab_size);
1700 	if (!sbi->page_array_slab)
1701 		return -ENOMEM;
1702 	return 0;
1703 }
1704 
1705 void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi)
1706 {
1707 	kmem_cache_destroy(sbi->page_array_slab);
1708 }
1709 
1710 static int __init f2fs_init_cic_cache(void)
1711 {
1712 	cic_entry_slab = f2fs_kmem_cache_create("f2fs_cic_entry",
1713 					sizeof(struct compress_io_ctx));
1714 	if (!cic_entry_slab)
1715 		return -ENOMEM;
1716 	return 0;
1717 }
1718 
1719 static void f2fs_destroy_cic_cache(void)
1720 {
1721 	kmem_cache_destroy(cic_entry_slab);
1722 }
1723 
1724 static int __init f2fs_init_dic_cache(void)
1725 {
1726 	dic_entry_slab = f2fs_kmem_cache_create("f2fs_dic_entry",
1727 					sizeof(struct decompress_io_ctx));
1728 	if (!dic_entry_slab)
1729 		return -ENOMEM;
1730 	return 0;
1731 }
1732 
1733 static void f2fs_destroy_dic_cache(void)
1734 {
1735 	kmem_cache_destroy(dic_entry_slab);
1736 }
1737 
1738 int __init f2fs_init_compress_cache(void)
1739 {
1740 	int err;
1741 
1742 	err = f2fs_init_cic_cache();
1743 	if (err)
1744 		goto out;
1745 	err = f2fs_init_dic_cache();
1746 	if (err)
1747 		goto free_cic;
1748 	return 0;
1749 free_cic:
1750 	f2fs_destroy_cic_cache();
1751 out:
1752 	return -ENOMEM;
1753 }
1754 
1755 void f2fs_destroy_compress_cache(void)
1756 {
1757 	f2fs_destroy_dic_cache();
1758 	f2fs_destroy_cic_cache();
1759 }
1760