xref: /openbmc/linux/fs/f2fs/compress.c (revision 324105775c1982aacbd2972b7024d8cc06474abe)
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 f2fs_kmem_cache_alloc(sbi->page_array_slab,
32 					GFP_F2FS_ZERO, false, F2FS_I_SB(inode));
33 	return f2fs_kzalloc(sbi, size, GFP_NOFS);
34 }
35 
36 static void page_array_free(struct inode *inode, void *pages, int nr)
37 {
38 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
39 	unsigned int size = sizeof(struct page *) * nr;
40 
41 	if (!pages)
42 		return;
43 
44 	if (likely(size <= sbi->page_array_slab_size))
45 		kmem_cache_free(sbi->page_array_slab, pages);
46 	else
47 		kfree(pages);
48 }
49 
50 struct f2fs_compress_ops {
51 	int (*init_compress_ctx)(struct compress_ctx *cc);
52 	void (*destroy_compress_ctx)(struct compress_ctx *cc);
53 	int (*compress_pages)(struct compress_ctx *cc);
54 	int (*init_decompress_ctx)(struct decompress_io_ctx *dic);
55 	void (*destroy_decompress_ctx)(struct decompress_io_ctx *dic);
56 	int (*decompress_pages)(struct decompress_io_ctx *dic);
57 };
58 
59 static unsigned int offset_in_cluster(struct compress_ctx *cc, pgoff_t index)
60 {
61 	return index & (cc->cluster_size - 1);
62 }
63 
64 static pgoff_t cluster_idx(struct compress_ctx *cc, pgoff_t index)
65 {
66 	return index >> cc->log_cluster_size;
67 }
68 
69 static pgoff_t start_idx_of_cluster(struct compress_ctx *cc)
70 {
71 	return cc->cluster_idx << cc->log_cluster_size;
72 }
73 
74 bool f2fs_is_compressed_page(struct page *page)
75 {
76 	if (!PagePrivate(page))
77 		return false;
78 	if (!page_private(page))
79 		return false;
80 	if (page_private_nonpointer(page))
81 		return false;
82 
83 	f2fs_bug_on(F2FS_M_SB(page->mapping),
84 		*((u32 *)page_private(page)) != F2FS_COMPRESSED_PAGE_MAGIC);
85 	return true;
86 }
87 
88 static void f2fs_set_compressed_page(struct page *page,
89 		struct inode *inode, pgoff_t index, void *data)
90 {
91 	attach_page_private(page, (void *)data);
92 
93 	/* i_crypto_info and iv index */
94 	page->index = index;
95 	page->mapping = inode->i_mapping;
96 }
97 
98 static void f2fs_drop_rpages(struct compress_ctx *cc, int len, bool unlock)
99 {
100 	int i;
101 
102 	for (i = 0; i < len; i++) {
103 		if (!cc->rpages[i])
104 			continue;
105 		if (unlock)
106 			unlock_page(cc->rpages[i]);
107 		else
108 			put_page(cc->rpages[i]);
109 	}
110 }
111 
112 static void f2fs_put_rpages(struct compress_ctx *cc)
113 {
114 	f2fs_drop_rpages(cc, cc->cluster_size, false);
115 }
116 
117 static void f2fs_unlock_rpages(struct compress_ctx *cc, int len)
118 {
119 	f2fs_drop_rpages(cc, len, true);
120 }
121 
122 static void f2fs_put_rpages_wbc(struct compress_ctx *cc,
123 		struct writeback_control *wbc, bool redirty, int unlock)
124 {
125 	unsigned int i;
126 
127 	for (i = 0; i < cc->cluster_size; i++) {
128 		if (!cc->rpages[i])
129 			continue;
130 		if (redirty)
131 			redirty_page_for_writepage(wbc, cc->rpages[i]);
132 		f2fs_put_page(cc->rpages[i], unlock);
133 	}
134 }
135 
136 struct page *f2fs_compress_control_page(struct page *page)
137 {
138 	return ((struct compress_io_ctx *)page_private(page))->rpages[0];
139 }
140 
141 int f2fs_init_compress_ctx(struct compress_ctx *cc)
142 {
143 	if (cc->rpages)
144 		return 0;
145 
146 	cc->rpages = page_array_alloc(cc->inode, cc->cluster_size);
147 	return cc->rpages ? 0 : -ENOMEM;
148 }
149 
150 void f2fs_destroy_compress_ctx(struct compress_ctx *cc, bool reuse)
151 {
152 	page_array_free(cc->inode, cc->rpages, cc->cluster_size);
153 	cc->rpages = NULL;
154 	cc->nr_rpages = 0;
155 	cc->nr_cpages = 0;
156 	if (!reuse)
157 		cc->cluster_idx = NULL_CLUSTER;
158 }
159 
160 void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct page *page)
161 {
162 	unsigned int cluster_ofs;
163 
164 	if (!f2fs_cluster_can_merge_page(cc, page->index))
165 		f2fs_bug_on(F2FS_I_SB(cc->inode), 1);
166 
167 	cluster_ofs = offset_in_cluster(cc, page->index);
168 	cc->rpages[cluster_ofs] = page;
169 	cc->nr_rpages++;
170 	cc->cluster_idx = cluster_idx(cc, page->index);
171 }
172 
173 #ifdef CONFIG_F2FS_FS_LZO
174 static int lzo_init_compress_ctx(struct compress_ctx *cc)
175 {
176 	cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
177 				LZO1X_MEM_COMPRESS, GFP_NOFS);
178 	if (!cc->private)
179 		return -ENOMEM;
180 
181 	cc->clen = lzo1x_worst_compress(PAGE_SIZE << cc->log_cluster_size);
182 	return 0;
183 }
184 
185 static void lzo_destroy_compress_ctx(struct compress_ctx *cc)
186 {
187 	kvfree(cc->private);
188 	cc->private = NULL;
189 }
190 
191 static int lzo_compress_pages(struct compress_ctx *cc)
192 {
193 	int ret;
194 
195 	ret = lzo1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
196 					&cc->clen, cc->private);
197 	if (ret != LZO_E_OK) {
198 		printk_ratelimited("%sF2FS-fs (%s): lzo compress failed, ret:%d\n",
199 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
200 		return -EIO;
201 	}
202 	return 0;
203 }
204 
205 static int lzo_decompress_pages(struct decompress_io_ctx *dic)
206 {
207 	int ret;
208 
209 	ret = lzo1x_decompress_safe(dic->cbuf->cdata, dic->clen,
210 						dic->rbuf, &dic->rlen);
211 	if (ret != LZO_E_OK) {
212 		printk_ratelimited("%sF2FS-fs (%s): lzo decompress failed, ret:%d\n",
213 				KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
214 		return -EIO;
215 	}
216 
217 	if (dic->rlen != PAGE_SIZE << dic->log_cluster_size) {
218 		printk_ratelimited("%sF2FS-fs (%s): lzo invalid rlen:%zu, "
219 					"expected:%lu\n", KERN_ERR,
220 					F2FS_I_SB(dic->inode)->sb->s_id,
221 					dic->rlen,
222 					PAGE_SIZE << dic->log_cluster_size);
223 		return -EIO;
224 	}
225 	return 0;
226 }
227 
228 static const struct f2fs_compress_ops f2fs_lzo_ops = {
229 	.init_compress_ctx	= lzo_init_compress_ctx,
230 	.destroy_compress_ctx	= lzo_destroy_compress_ctx,
231 	.compress_pages		= lzo_compress_pages,
232 	.decompress_pages	= lzo_decompress_pages,
233 };
234 #endif
235 
236 #ifdef CONFIG_F2FS_FS_LZ4
237 static int lz4_init_compress_ctx(struct compress_ctx *cc)
238 {
239 	unsigned int size = LZ4_MEM_COMPRESS;
240 
241 #ifdef CONFIG_F2FS_FS_LZ4HC
242 	if (F2FS_I(cc->inode)->i_compress_flag >> COMPRESS_LEVEL_OFFSET)
243 		size = LZ4HC_MEM_COMPRESS;
244 #endif
245 
246 	cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), size, GFP_NOFS);
247 	if (!cc->private)
248 		return -ENOMEM;
249 
250 	/*
251 	 * we do not change cc->clen to LZ4_compressBound(inputsize) to
252 	 * adapt worst compress case, because lz4 compressor can handle
253 	 * output budget properly.
254 	 */
255 	cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
256 	return 0;
257 }
258 
259 static void lz4_destroy_compress_ctx(struct compress_ctx *cc)
260 {
261 	kvfree(cc->private);
262 	cc->private = NULL;
263 }
264 
265 #ifdef CONFIG_F2FS_FS_LZ4HC
266 static int lz4hc_compress_pages(struct compress_ctx *cc)
267 {
268 	unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
269 						COMPRESS_LEVEL_OFFSET;
270 	int len;
271 
272 	if (level)
273 		len = LZ4_compress_HC(cc->rbuf, cc->cbuf->cdata, cc->rlen,
274 					cc->clen, level, cc->private);
275 	else
276 		len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
277 						cc->clen, cc->private);
278 	if (!len)
279 		return -EAGAIN;
280 
281 	cc->clen = len;
282 	return 0;
283 }
284 #endif
285 
286 static int lz4_compress_pages(struct compress_ctx *cc)
287 {
288 	int len;
289 
290 #ifdef CONFIG_F2FS_FS_LZ4HC
291 	return lz4hc_compress_pages(cc);
292 #endif
293 	len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
294 						cc->clen, cc->private);
295 	if (!len)
296 		return -EAGAIN;
297 
298 	cc->clen = len;
299 	return 0;
300 }
301 
302 static int lz4_decompress_pages(struct decompress_io_ctx *dic)
303 {
304 	int ret;
305 
306 	ret = LZ4_decompress_safe(dic->cbuf->cdata, dic->rbuf,
307 						dic->clen, dic->rlen);
308 	if (ret < 0) {
309 		printk_ratelimited("%sF2FS-fs (%s): lz4 decompress failed, ret:%d\n",
310 				KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
311 		return -EIO;
312 	}
313 
314 	if (ret != PAGE_SIZE << dic->log_cluster_size) {
315 		printk_ratelimited("%sF2FS-fs (%s): lz4 invalid rlen:%zu, "
316 					"expected:%lu\n", KERN_ERR,
317 					F2FS_I_SB(dic->inode)->sb->s_id,
318 					dic->rlen,
319 					PAGE_SIZE << dic->log_cluster_size);
320 		return -EIO;
321 	}
322 	return 0;
323 }
324 
325 static const struct f2fs_compress_ops f2fs_lz4_ops = {
326 	.init_compress_ctx	= lz4_init_compress_ctx,
327 	.destroy_compress_ctx	= lz4_destroy_compress_ctx,
328 	.compress_pages		= lz4_compress_pages,
329 	.decompress_pages	= lz4_decompress_pages,
330 };
331 #endif
332 
333 #ifdef CONFIG_F2FS_FS_ZSTD
334 #define F2FS_ZSTD_DEFAULT_CLEVEL	1
335 
336 static int zstd_init_compress_ctx(struct compress_ctx *cc)
337 {
338 	ZSTD_parameters params;
339 	ZSTD_CStream *stream;
340 	void *workspace;
341 	unsigned int workspace_size;
342 	unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
343 						COMPRESS_LEVEL_OFFSET;
344 
345 	if (!level)
346 		level = F2FS_ZSTD_DEFAULT_CLEVEL;
347 
348 	params = ZSTD_getParams(level, cc->rlen, 0);
349 	workspace_size = ZSTD_CStreamWorkspaceBound(params.cParams);
350 
351 	workspace = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
352 					workspace_size, GFP_NOFS);
353 	if (!workspace)
354 		return -ENOMEM;
355 
356 	stream = ZSTD_initCStream(params, 0, workspace, workspace_size);
357 	if (!stream) {
358 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initCStream failed\n",
359 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
360 				__func__);
361 		kvfree(workspace);
362 		return -EIO;
363 	}
364 
365 	cc->private = workspace;
366 	cc->private2 = stream;
367 
368 	cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
369 	return 0;
370 }
371 
372 static void zstd_destroy_compress_ctx(struct compress_ctx *cc)
373 {
374 	kvfree(cc->private);
375 	cc->private = NULL;
376 	cc->private2 = NULL;
377 }
378 
379 static int zstd_compress_pages(struct compress_ctx *cc)
380 {
381 	ZSTD_CStream *stream = cc->private2;
382 	ZSTD_inBuffer inbuf;
383 	ZSTD_outBuffer outbuf;
384 	int src_size = cc->rlen;
385 	int dst_size = src_size - PAGE_SIZE - COMPRESS_HEADER_SIZE;
386 	int ret;
387 
388 	inbuf.pos = 0;
389 	inbuf.src = cc->rbuf;
390 	inbuf.size = src_size;
391 
392 	outbuf.pos = 0;
393 	outbuf.dst = cc->cbuf->cdata;
394 	outbuf.size = dst_size;
395 
396 	ret = ZSTD_compressStream(stream, &outbuf, &inbuf);
397 	if (ZSTD_isError(ret)) {
398 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
399 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
400 				__func__, ZSTD_getErrorCode(ret));
401 		return -EIO;
402 	}
403 
404 	ret = ZSTD_endStream(stream, &outbuf);
405 	if (ZSTD_isError(ret)) {
406 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_endStream returned %d\n",
407 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
408 				__func__, ZSTD_getErrorCode(ret));
409 		return -EIO;
410 	}
411 
412 	/*
413 	 * there is compressed data remained in intermediate buffer due to
414 	 * no more space in cbuf.cdata
415 	 */
416 	if (ret)
417 		return -EAGAIN;
418 
419 	cc->clen = outbuf.pos;
420 	return 0;
421 }
422 
423 static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
424 {
425 	ZSTD_DStream *stream;
426 	void *workspace;
427 	unsigned int workspace_size;
428 	unsigned int max_window_size =
429 			MAX_COMPRESS_WINDOW_SIZE(dic->log_cluster_size);
430 
431 	workspace_size = ZSTD_DStreamWorkspaceBound(max_window_size);
432 
433 	workspace = f2fs_kvmalloc(F2FS_I_SB(dic->inode),
434 					workspace_size, GFP_NOFS);
435 	if (!workspace)
436 		return -ENOMEM;
437 
438 	stream = ZSTD_initDStream(max_window_size, workspace, workspace_size);
439 	if (!stream) {
440 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initDStream failed\n",
441 				KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
442 				__func__);
443 		kvfree(workspace);
444 		return -EIO;
445 	}
446 
447 	dic->private = workspace;
448 	dic->private2 = stream;
449 
450 	return 0;
451 }
452 
453 static void zstd_destroy_decompress_ctx(struct decompress_io_ctx *dic)
454 {
455 	kvfree(dic->private);
456 	dic->private = NULL;
457 	dic->private2 = NULL;
458 }
459 
460 static int zstd_decompress_pages(struct decompress_io_ctx *dic)
461 {
462 	ZSTD_DStream *stream = dic->private2;
463 	ZSTD_inBuffer inbuf;
464 	ZSTD_outBuffer outbuf;
465 	int ret;
466 
467 	inbuf.pos = 0;
468 	inbuf.src = dic->cbuf->cdata;
469 	inbuf.size = dic->clen;
470 
471 	outbuf.pos = 0;
472 	outbuf.dst = dic->rbuf;
473 	outbuf.size = dic->rlen;
474 
475 	ret = ZSTD_decompressStream(stream, &outbuf, &inbuf);
476 	if (ZSTD_isError(ret)) {
477 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
478 				KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
479 				__func__, ZSTD_getErrorCode(ret));
480 		return -EIO;
481 	}
482 
483 	if (dic->rlen != outbuf.pos) {
484 		printk_ratelimited("%sF2FS-fs (%s): %s ZSTD invalid rlen:%zu, "
485 				"expected:%lu\n", KERN_ERR,
486 				F2FS_I_SB(dic->inode)->sb->s_id,
487 				__func__, dic->rlen,
488 				PAGE_SIZE << dic->log_cluster_size);
489 		return -EIO;
490 	}
491 
492 	return 0;
493 }
494 
495 static const struct f2fs_compress_ops f2fs_zstd_ops = {
496 	.init_compress_ctx	= zstd_init_compress_ctx,
497 	.destroy_compress_ctx	= zstd_destroy_compress_ctx,
498 	.compress_pages		= zstd_compress_pages,
499 	.init_decompress_ctx	= zstd_init_decompress_ctx,
500 	.destroy_decompress_ctx	= zstd_destroy_decompress_ctx,
501 	.decompress_pages	= zstd_decompress_pages,
502 };
503 #endif
504 
505 #ifdef CONFIG_F2FS_FS_LZO
506 #ifdef CONFIG_F2FS_FS_LZORLE
507 static int lzorle_compress_pages(struct compress_ctx *cc)
508 {
509 	int ret;
510 
511 	ret = lzorle1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
512 					&cc->clen, cc->private);
513 	if (ret != LZO_E_OK) {
514 		printk_ratelimited("%sF2FS-fs (%s): lzo-rle compress failed, ret:%d\n",
515 				KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
516 		return -EIO;
517 	}
518 	return 0;
519 }
520 
521 static const struct f2fs_compress_ops f2fs_lzorle_ops = {
522 	.init_compress_ctx	= lzo_init_compress_ctx,
523 	.destroy_compress_ctx	= lzo_destroy_compress_ctx,
524 	.compress_pages		= lzorle_compress_pages,
525 	.decompress_pages	= lzo_decompress_pages,
526 };
527 #endif
528 #endif
529 
530 static const struct f2fs_compress_ops *f2fs_cops[COMPRESS_MAX] = {
531 #ifdef CONFIG_F2FS_FS_LZO
532 	&f2fs_lzo_ops,
533 #else
534 	NULL,
535 #endif
536 #ifdef CONFIG_F2FS_FS_LZ4
537 	&f2fs_lz4_ops,
538 #else
539 	NULL,
540 #endif
541 #ifdef CONFIG_F2FS_FS_ZSTD
542 	&f2fs_zstd_ops,
543 #else
544 	NULL,
545 #endif
546 #if defined(CONFIG_F2FS_FS_LZO) && defined(CONFIG_F2FS_FS_LZORLE)
547 	&f2fs_lzorle_ops,
548 #else
549 	NULL,
550 #endif
551 };
552 
553 bool f2fs_is_compress_backend_ready(struct inode *inode)
554 {
555 	if (!f2fs_compressed_file(inode))
556 		return true;
557 	return f2fs_cops[F2FS_I(inode)->i_compress_algorithm];
558 }
559 
560 static mempool_t *compress_page_pool;
561 static int num_compress_pages = 512;
562 module_param(num_compress_pages, uint, 0444);
563 MODULE_PARM_DESC(num_compress_pages,
564 		"Number of intermediate compress pages to preallocate");
565 
566 int f2fs_init_compress_mempool(void)
567 {
568 	compress_page_pool = mempool_create_page_pool(num_compress_pages, 0);
569 	if (!compress_page_pool)
570 		return -ENOMEM;
571 
572 	return 0;
573 }
574 
575 void f2fs_destroy_compress_mempool(void)
576 {
577 	mempool_destroy(compress_page_pool);
578 }
579 
580 static struct page *f2fs_compress_alloc_page(void)
581 {
582 	struct page *page;
583 
584 	page = mempool_alloc(compress_page_pool, GFP_NOFS);
585 	lock_page(page);
586 
587 	return page;
588 }
589 
590 static void f2fs_compress_free_page(struct page *page)
591 {
592 	if (!page)
593 		return;
594 	detach_page_private(page);
595 	page->mapping = NULL;
596 	unlock_page(page);
597 	mempool_free(page, compress_page_pool);
598 }
599 
600 #define MAX_VMAP_RETRIES	3
601 
602 static void *f2fs_vmap(struct page **pages, unsigned int count)
603 {
604 	int i;
605 	void *buf = NULL;
606 
607 	for (i = 0; i < MAX_VMAP_RETRIES; i++) {
608 		buf = vm_map_ram(pages, count, -1);
609 		if (buf)
610 			break;
611 		vm_unmap_aliases();
612 	}
613 	return buf;
614 }
615 
616 static int f2fs_compress_pages(struct compress_ctx *cc)
617 {
618 	struct f2fs_inode_info *fi = F2FS_I(cc->inode);
619 	const struct f2fs_compress_ops *cops =
620 				f2fs_cops[fi->i_compress_algorithm];
621 	unsigned int max_len, new_nr_cpages;
622 	struct page **new_cpages;
623 	u32 chksum = 0;
624 	int i, ret;
625 
626 	trace_f2fs_compress_pages_start(cc->inode, cc->cluster_idx,
627 				cc->cluster_size, fi->i_compress_algorithm);
628 
629 	if (cops->init_compress_ctx) {
630 		ret = cops->init_compress_ctx(cc);
631 		if (ret)
632 			goto out;
633 	}
634 
635 	max_len = COMPRESS_HEADER_SIZE + cc->clen;
636 	cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE);
637 
638 	cc->cpages = page_array_alloc(cc->inode, cc->nr_cpages);
639 	if (!cc->cpages) {
640 		ret = -ENOMEM;
641 		goto destroy_compress_ctx;
642 	}
643 
644 	for (i = 0; i < cc->nr_cpages; i++) {
645 		cc->cpages[i] = f2fs_compress_alloc_page();
646 		if (!cc->cpages[i]) {
647 			ret = -ENOMEM;
648 			goto out_free_cpages;
649 		}
650 	}
651 
652 	cc->rbuf = f2fs_vmap(cc->rpages, cc->cluster_size);
653 	if (!cc->rbuf) {
654 		ret = -ENOMEM;
655 		goto out_free_cpages;
656 	}
657 
658 	cc->cbuf = f2fs_vmap(cc->cpages, cc->nr_cpages);
659 	if (!cc->cbuf) {
660 		ret = -ENOMEM;
661 		goto out_vunmap_rbuf;
662 	}
663 
664 	ret = cops->compress_pages(cc);
665 	if (ret)
666 		goto out_vunmap_cbuf;
667 
668 	max_len = PAGE_SIZE * (cc->cluster_size - 1) - COMPRESS_HEADER_SIZE;
669 
670 	if (cc->clen > max_len) {
671 		ret = -EAGAIN;
672 		goto out_vunmap_cbuf;
673 	}
674 
675 	cc->cbuf->clen = cpu_to_le32(cc->clen);
676 
677 	if (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)
678 		chksum = f2fs_crc32(F2FS_I_SB(cc->inode),
679 					cc->cbuf->cdata, cc->clen);
680 	cc->cbuf->chksum = cpu_to_le32(chksum);
681 
682 	for (i = 0; i < COMPRESS_DATA_RESERVED_SIZE; i++)
683 		cc->cbuf->reserved[i] = cpu_to_le32(0);
684 
685 	new_nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE);
686 
687 	/* Now we're going to cut unnecessary tail pages */
688 	new_cpages = page_array_alloc(cc->inode, new_nr_cpages);
689 	if (!new_cpages) {
690 		ret = -ENOMEM;
691 		goto out_vunmap_cbuf;
692 	}
693 
694 	/* zero out any unused part of the last page */
695 	memset(&cc->cbuf->cdata[cc->clen], 0,
696 			(new_nr_cpages * PAGE_SIZE) -
697 			(cc->clen + COMPRESS_HEADER_SIZE));
698 
699 	vm_unmap_ram(cc->cbuf, cc->nr_cpages);
700 	vm_unmap_ram(cc->rbuf, cc->cluster_size);
701 
702 	for (i = 0; i < cc->nr_cpages; i++) {
703 		if (i < new_nr_cpages) {
704 			new_cpages[i] = cc->cpages[i];
705 			continue;
706 		}
707 		f2fs_compress_free_page(cc->cpages[i]);
708 		cc->cpages[i] = NULL;
709 	}
710 
711 	if (cops->destroy_compress_ctx)
712 		cops->destroy_compress_ctx(cc);
713 
714 	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
715 	cc->cpages = new_cpages;
716 	cc->nr_cpages = new_nr_cpages;
717 
718 	trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
719 							cc->clen, ret);
720 	return 0;
721 
722 out_vunmap_cbuf:
723 	vm_unmap_ram(cc->cbuf, cc->nr_cpages);
724 out_vunmap_rbuf:
725 	vm_unmap_ram(cc->rbuf, cc->cluster_size);
726 out_free_cpages:
727 	for (i = 0; i < cc->nr_cpages; i++) {
728 		if (cc->cpages[i])
729 			f2fs_compress_free_page(cc->cpages[i]);
730 	}
731 	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
732 	cc->cpages = NULL;
733 destroy_compress_ctx:
734 	if (cops->destroy_compress_ctx)
735 		cops->destroy_compress_ctx(cc);
736 out:
737 	trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
738 							cc->clen, ret);
739 	return ret;
740 }
741 
742 void f2fs_decompress_cluster(struct decompress_io_ctx *dic)
743 {
744 	struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
745 	struct f2fs_inode_info *fi = F2FS_I(dic->inode);
746 	const struct f2fs_compress_ops *cops =
747 			f2fs_cops[fi->i_compress_algorithm];
748 	int ret;
749 	int i;
750 
751 	trace_f2fs_decompress_pages_start(dic->inode, dic->cluster_idx,
752 				dic->cluster_size, fi->i_compress_algorithm);
753 
754 	if (dic->failed) {
755 		ret = -EIO;
756 		goto out_end_io;
757 	}
758 
759 	dic->tpages = page_array_alloc(dic->inode, dic->cluster_size);
760 	if (!dic->tpages) {
761 		ret = -ENOMEM;
762 		goto out_end_io;
763 	}
764 
765 	for (i = 0; i < dic->cluster_size; i++) {
766 		if (dic->rpages[i]) {
767 			dic->tpages[i] = dic->rpages[i];
768 			continue;
769 		}
770 
771 		dic->tpages[i] = f2fs_compress_alloc_page();
772 		if (!dic->tpages[i]) {
773 			ret = -ENOMEM;
774 			goto out_end_io;
775 		}
776 	}
777 
778 	if (cops->init_decompress_ctx) {
779 		ret = cops->init_decompress_ctx(dic);
780 		if (ret)
781 			goto out_end_io;
782 	}
783 
784 	dic->rbuf = f2fs_vmap(dic->tpages, dic->cluster_size);
785 	if (!dic->rbuf) {
786 		ret = -ENOMEM;
787 		goto out_destroy_decompress_ctx;
788 	}
789 
790 	dic->cbuf = f2fs_vmap(dic->cpages, dic->nr_cpages);
791 	if (!dic->cbuf) {
792 		ret = -ENOMEM;
793 		goto out_vunmap_rbuf;
794 	}
795 
796 	dic->clen = le32_to_cpu(dic->cbuf->clen);
797 	dic->rlen = PAGE_SIZE << dic->log_cluster_size;
798 
799 	if (dic->clen > PAGE_SIZE * dic->nr_cpages - COMPRESS_HEADER_SIZE) {
800 		ret = -EFSCORRUPTED;
801 		goto out_vunmap_cbuf;
802 	}
803 
804 	ret = cops->decompress_pages(dic);
805 
806 	if (!ret && (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)) {
807 		u32 provided = le32_to_cpu(dic->cbuf->chksum);
808 		u32 calculated = f2fs_crc32(sbi, dic->cbuf->cdata, dic->clen);
809 
810 		if (provided != calculated) {
811 			if (!is_inode_flag_set(dic->inode, FI_COMPRESS_CORRUPT)) {
812 				set_inode_flag(dic->inode, FI_COMPRESS_CORRUPT);
813 				printk_ratelimited(
814 					"%sF2FS-fs (%s): checksum invalid, nid = %lu, %x vs %x",
815 					KERN_INFO, sbi->sb->s_id, dic->inode->i_ino,
816 					provided, calculated);
817 			}
818 			set_sbi_flag(sbi, SBI_NEED_FSCK);
819 		}
820 	}
821 
822 out_vunmap_cbuf:
823 	vm_unmap_ram(dic->cbuf, dic->nr_cpages);
824 out_vunmap_rbuf:
825 	vm_unmap_ram(dic->rbuf, dic->cluster_size);
826 out_destroy_decompress_ctx:
827 	if (cops->destroy_decompress_ctx)
828 		cops->destroy_decompress_ctx(dic);
829 out_end_io:
830 	trace_f2fs_decompress_pages_end(dic->inode, dic->cluster_idx,
831 							dic->clen, ret);
832 	f2fs_decompress_end_io(dic, ret);
833 }
834 
835 /*
836  * This is called when a page of a compressed cluster has been read from disk
837  * (or failed to be read from disk).  It checks whether this page was the last
838  * page being waited on in the cluster, and if so, it decompresses the cluster
839  * (or in the case of a failure, cleans up without actually decompressing).
840  */
841 void f2fs_end_read_compressed_page(struct page *page, bool failed,
842 						block_t blkaddr)
843 {
844 	struct decompress_io_ctx *dic =
845 			(struct decompress_io_ctx *)page_private(page);
846 	struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
847 
848 	dec_page_count(sbi, F2FS_RD_DATA);
849 
850 	if (failed)
851 		WRITE_ONCE(dic->failed, true);
852 	else if (blkaddr)
853 		f2fs_cache_compressed_page(sbi, page,
854 					dic->inode->i_ino, blkaddr);
855 
856 	if (atomic_dec_and_test(&dic->remaining_pages))
857 		f2fs_decompress_cluster(dic);
858 }
859 
860 static bool is_page_in_cluster(struct compress_ctx *cc, pgoff_t index)
861 {
862 	if (cc->cluster_idx == NULL_CLUSTER)
863 		return true;
864 	return cc->cluster_idx == cluster_idx(cc, index);
865 }
866 
867 bool f2fs_cluster_is_empty(struct compress_ctx *cc)
868 {
869 	return cc->nr_rpages == 0;
870 }
871 
872 static bool f2fs_cluster_is_full(struct compress_ctx *cc)
873 {
874 	return cc->cluster_size == cc->nr_rpages;
875 }
876 
877 bool f2fs_cluster_can_merge_page(struct compress_ctx *cc, pgoff_t index)
878 {
879 	if (f2fs_cluster_is_empty(cc))
880 		return true;
881 	return is_page_in_cluster(cc, index);
882 }
883 
884 static bool cluster_has_invalid_data(struct compress_ctx *cc)
885 {
886 	loff_t i_size = i_size_read(cc->inode);
887 	unsigned nr_pages = DIV_ROUND_UP(i_size, PAGE_SIZE);
888 	int i;
889 
890 	for (i = 0; i < cc->cluster_size; i++) {
891 		struct page *page = cc->rpages[i];
892 
893 		f2fs_bug_on(F2FS_I_SB(cc->inode), !page);
894 
895 		/* beyond EOF */
896 		if (page->index >= nr_pages)
897 			return true;
898 	}
899 	return false;
900 }
901 
902 static int __f2fs_cluster_blocks(struct inode *inode,
903 				unsigned int cluster_idx, bool compr)
904 {
905 	struct dnode_of_data dn;
906 	unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
907 	unsigned int start_idx = cluster_idx <<
908 				F2FS_I(inode)->i_log_cluster_size;
909 	int ret;
910 
911 	set_new_dnode(&dn, inode, NULL, NULL, 0);
912 	ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
913 	if (ret) {
914 		if (ret == -ENOENT)
915 			ret = 0;
916 		goto fail;
917 	}
918 
919 	if (dn.data_blkaddr == COMPRESS_ADDR) {
920 		int i;
921 
922 		ret = 1;
923 		for (i = 1; i < cluster_size; i++) {
924 			block_t blkaddr;
925 
926 			blkaddr = data_blkaddr(dn.inode,
927 					dn.node_page, dn.ofs_in_node + i);
928 			if (compr) {
929 				if (__is_valid_data_blkaddr(blkaddr))
930 					ret++;
931 			} else {
932 				if (blkaddr != NULL_ADDR)
933 					ret++;
934 			}
935 		}
936 
937 		f2fs_bug_on(F2FS_I_SB(inode),
938 			!compr && ret != cluster_size &&
939 			!is_inode_flag_set(inode, FI_COMPRESS_RELEASED));
940 	}
941 fail:
942 	f2fs_put_dnode(&dn);
943 	return ret;
944 }
945 
946 /* return # of compressed blocks in compressed cluster */
947 static int f2fs_compressed_blocks(struct compress_ctx *cc)
948 {
949 	return __f2fs_cluster_blocks(cc->inode, cc->cluster_idx, true);
950 }
951 
952 /* return # of valid blocks in compressed cluster */
953 int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index)
954 {
955 	return __f2fs_cluster_blocks(inode,
956 		index >> F2FS_I(inode)->i_log_cluster_size,
957 		false);
958 }
959 
960 static bool cluster_may_compress(struct compress_ctx *cc)
961 {
962 	if (!f2fs_need_compress_data(cc->inode))
963 		return false;
964 	if (f2fs_is_atomic_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_has_invalid_data(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 	sector_t last_block_in_bio;
999 	unsigned fgp_flag = FGP_LOCK | FGP_WRITE | FGP_CREAT;
1000 	pgoff_t start_idx = start_idx_of_cluster(cc);
1001 	int i, ret;
1002 
1003 retry:
1004 	ret = f2fs_is_compressed_cluster(cc->inode, start_idx);
1005 	if (ret <= 0)
1006 		return ret;
1007 
1008 	ret = f2fs_init_compress_ctx(cc);
1009 	if (ret)
1010 		return ret;
1011 
1012 	/* keep page reference to avoid page reclaim */
1013 	for (i = 0; i < cc->cluster_size; i++) {
1014 		page = f2fs_pagecache_get_page(mapping, start_idx + i,
1015 							fgp_flag, GFP_NOFS);
1016 		if (!page) {
1017 			ret = -ENOMEM;
1018 			goto unlock_pages;
1019 		}
1020 
1021 		if (PageUptodate(page))
1022 			f2fs_put_page(page, 1);
1023 		else
1024 			f2fs_compress_ctx_add_page(cc, page);
1025 	}
1026 
1027 	if (!f2fs_cluster_is_empty(cc)) {
1028 		struct bio *bio = NULL;
1029 
1030 		ret = f2fs_read_multi_pages(cc, &bio, cc->cluster_size,
1031 					&last_block_in_bio, false, true);
1032 		f2fs_put_rpages(cc);
1033 		f2fs_destroy_compress_ctx(cc, true);
1034 		if (ret)
1035 			goto out;
1036 		if (bio)
1037 			f2fs_submit_bio(sbi, bio, DATA);
1038 
1039 		ret = f2fs_init_compress_ctx(cc);
1040 		if (ret)
1041 			goto out;
1042 	}
1043 
1044 	for (i = 0; i < cc->cluster_size; i++) {
1045 		f2fs_bug_on(sbi, cc->rpages[i]);
1046 
1047 		page = find_lock_page(mapping, start_idx + i);
1048 		if (!page) {
1049 			/* page can be truncated */
1050 			goto release_and_retry;
1051 		}
1052 
1053 		f2fs_wait_on_page_writeback(page, DATA, true, true);
1054 		f2fs_compress_ctx_add_page(cc, page);
1055 
1056 		if (!PageUptodate(page)) {
1057 release_and_retry:
1058 			f2fs_put_rpages(cc);
1059 			f2fs_unlock_rpages(cc, i + 1);
1060 			f2fs_destroy_compress_ctx(cc, true);
1061 			goto retry;
1062 		}
1063 	}
1064 
1065 	if (likely(!ret)) {
1066 		*fsdata = cc->rpages;
1067 		*pagep = cc->rpages[offset_in_cluster(cc, index)];
1068 		return cc->cluster_size;
1069 	}
1070 
1071 unlock_pages:
1072 	f2fs_put_rpages(cc);
1073 	f2fs_unlock_rpages(cc, i);
1074 	f2fs_destroy_compress_ctx(cc, true);
1075 out:
1076 	return ret;
1077 }
1078 
1079 int f2fs_prepare_compress_overwrite(struct inode *inode,
1080 		struct page **pagep, pgoff_t index, void **fsdata)
1081 {
1082 	struct compress_ctx cc = {
1083 		.inode = inode,
1084 		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1085 		.cluster_size = F2FS_I(inode)->i_cluster_size,
1086 		.cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
1087 		.rpages = NULL,
1088 		.nr_rpages = 0,
1089 	};
1090 
1091 	return prepare_compress_overwrite(&cc, pagep, index, fsdata);
1092 }
1093 
1094 bool f2fs_compress_write_end(struct inode *inode, void *fsdata,
1095 					pgoff_t index, unsigned copied)
1096 
1097 {
1098 	struct compress_ctx cc = {
1099 		.inode = inode,
1100 		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1101 		.cluster_size = F2FS_I(inode)->i_cluster_size,
1102 		.rpages = fsdata,
1103 	};
1104 	bool first_index = (index == cc.rpages[0]->index);
1105 
1106 	if (copied)
1107 		set_cluster_dirty(&cc);
1108 
1109 	f2fs_put_rpages_wbc(&cc, NULL, false, 1);
1110 	f2fs_destroy_compress_ctx(&cc, false);
1111 
1112 	return first_index;
1113 }
1114 
1115 int f2fs_truncate_partial_cluster(struct inode *inode, u64 from, bool lock)
1116 {
1117 	void *fsdata = NULL;
1118 	struct page *pagep;
1119 	int log_cluster_size = F2FS_I(inode)->i_log_cluster_size;
1120 	pgoff_t start_idx = from >> (PAGE_SHIFT + log_cluster_size) <<
1121 							log_cluster_size;
1122 	int err;
1123 
1124 	err = f2fs_is_compressed_cluster(inode, start_idx);
1125 	if (err < 0)
1126 		return err;
1127 
1128 	/* truncate normal cluster */
1129 	if (!err)
1130 		return f2fs_do_truncate_blocks(inode, from, lock);
1131 
1132 	/* truncate compressed cluster */
1133 	err = f2fs_prepare_compress_overwrite(inode, &pagep,
1134 						start_idx, &fsdata);
1135 
1136 	/* should not be a normal cluster */
1137 	f2fs_bug_on(F2FS_I_SB(inode), err == 0);
1138 
1139 	if (err <= 0)
1140 		return err;
1141 
1142 	if (err > 0) {
1143 		struct page **rpages = fsdata;
1144 		int cluster_size = F2FS_I(inode)->i_cluster_size;
1145 		int i;
1146 
1147 		for (i = cluster_size - 1; i >= 0; i--) {
1148 			loff_t start = rpages[i]->index << PAGE_SHIFT;
1149 
1150 			if (from <= start) {
1151 				zero_user_segment(rpages[i], 0, PAGE_SIZE);
1152 			} else {
1153 				zero_user_segment(rpages[i], from - start,
1154 								PAGE_SIZE);
1155 				break;
1156 			}
1157 		}
1158 
1159 		f2fs_compress_write_end(inode, fsdata, start_idx, true);
1160 	}
1161 	return 0;
1162 }
1163 
1164 static int f2fs_write_compressed_pages(struct compress_ctx *cc,
1165 					int *submitted,
1166 					struct writeback_control *wbc,
1167 					enum iostat_type io_type)
1168 {
1169 	struct inode *inode = cc->inode;
1170 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1171 	struct f2fs_inode_info *fi = F2FS_I(inode);
1172 	struct f2fs_io_info fio = {
1173 		.sbi = sbi,
1174 		.ino = cc->inode->i_ino,
1175 		.type = DATA,
1176 		.op = REQ_OP_WRITE,
1177 		.op_flags = wbc_to_write_flags(wbc),
1178 		.old_blkaddr = NEW_ADDR,
1179 		.page = NULL,
1180 		.encrypted_page = NULL,
1181 		.compressed_page = NULL,
1182 		.submitted = false,
1183 		.io_type = io_type,
1184 		.io_wbc = wbc,
1185 		.encrypted = fscrypt_inode_uses_fs_layer_crypto(cc->inode),
1186 	};
1187 	struct dnode_of_data dn;
1188 	struct node_info ni;
1189 	struct compress_io_ctx *cic;
1190 	pgoff_t start_idx = start_idx_of_cluster(cc);
1191 	unsigned int last_index = cc->cluster_size - 1;
1192 	loff_t psize;
1193 	int i, err;
1194 
1195 	/* we should bypass data pages to proceed the kworkder jobs */
1196 	if (unlikely(f2fs_cp_error(sbi))) {
1197 		mapping_set_error(cc->rpages[0]->mapping, -EIO);
1198 		goto out_free;
1199 	}
1200 
1201 	if (IS_NOQUOTA(inode)) {
1202 		/*
1203 		 * We need to wait for node_write to avoid block allocation during
1204 		 * checkpoint. This can only happen to quota writes which can cause
1205 		 * the below discard race condition.
1206 		 */
1207 		down_read(&sbi->node_write);
1208 	} else if (!f2fs_trylock_op(sbi)) {
1209 		goto out_free;
1210 	}
1211 
1212 	set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1213 
1214 	err = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
1215 	if (err)
1216 		goto out_unlock_op;
1217 
1218 	for (i = 0; i < cc->cluster_size; i++) {
1219 		if (data_blkaddr(dn.inode, dn.node_page,
1220 					dn.ofs_in_node + i) == NULL_ADDR)
1221 			goto out_put_dnode;
1222 	}
1223 
1224 	psize = (loff_t)(cc->rpages[last_index]->index + 1) << PAGE_SHIFT;
1225 
1226 	err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1227 	if (err)
1228 		goto out_put_dnode;
1229 
1230 	fio.version = ni.version;
1231 
1232 	cic = f2fs_kmem_cache_alloc(cic_entry_slab, GFP_F2FS_ZERO, false, sbi);
1233 	if (!cic)
1234 		goto out_put_dnode;
1235 
1236 	cic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1237 	cic->inode = inode;
1238 	atomic_set(&cic->pending_pages, cc->nr_cpages);
1239 	cic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1240 	if (!cic->rpages)
1241 		goto out_put_cic;
1242 
1243 	cic->nr_rpages = cc->cluster_size;
1244 
1245 	for (i = 0; i < cc->nr_cpages; i++) {
1246 		f2fs_set_compressed_page(cc->cpages[i], inode,
1247 					cc->rpages[i + 1]->index, cic);
1248 		fio.compressed_page = cc->cpages[i];
1249 
1250 		fio.old_blkaddr = data_blkaddr(dn.inode, dn.node_page,
1251 						dn.ofs_in_node + i + 1);
1252 
1253 		/* wait for GCed page writeback via META_MAPPING */
1254 		f2fs_wait_on_block_writeback(inode, fio.old_blkaddr);
1255 
1256 		if (fio.encrypted) {
1257 			fio.page = cc->rpages[i + 1];
1258 			err = f2fs_encrypt_one_page(&fio);
1259 			if (err)
1260 				goto out_destroy_crypt;
1261 			cc->cpages[i] = fio.encrypted_page;
1262 		}
1263 	}
1264 
1265 	set_cluster_writeback(cc);
1266 
1267 	for (i = 0; i < cc->cluster_size; i++)
1268 		cic->rpages[i] = cc->rpages[i];
1269 
1270 	for (i = 0; i < cc->cluster_size; i++, dn.ofs_in_node++) {
1271 		block_t blkaddr;
1272 
1273 		blkaddr = f2fs_data_blkaddr(&dn);
1274 		fio.page = cc->rpages[i];
1275 		fio.old_blkaddr = blkaddr;
1276 
1277 		/* cluster header */
1278 		if (i == 0) {
1279 			if (blkaddr == COMPRESS_ADDR)
1280 				fio.compr_blocks++;
1281 			if (__is_valid_data_blkaddr(blkaddr))
1282 				f2fs_invalidate_blocks(sbi, blkaddr);
1283 			f2fs_update_data_blkaddr(&dn, COMPRESS_ADDR);
1284 			goto unlock_continue;
1285 		}
1286 
1287 		if (fio.compr_blocks && __is_valid_data_blkaddr(blkaddr))
1288 			fio.compr_blocks++;
1289 
1290 		if (i > cc->nr_cpages) {
1291 			if (__is_valid_data_blkaddr(blkaddr)) {
1292 				f2fs_invalidate_blocks(sbi, blkaddr);
1293 				f2fs_update_data_blkaddr(&dn, NEW_ADDR);
1294 			}
1295 			goto unlock_continue;
1296 		}
1297 
1298 		f2fs_bug_on(fio.sbi, blkaddr == NULL_ADDR);
1299 
1300 		if (fio.encrypted)
1301 			fio.encrypted_page = cc->cpages[i - 1];
1302 		else
1303 			fio.compressed_page = cc->cpages[i - 1];
1304 
1305 		cc->cpages[i - 1] = NULL;
1306 		f2fs_outplace_write_data(&dn, &fio);
1307 		(*submitted)++;
1308 unlock_continue:
1309 		inode_dec_dirty_pages(cc->inode);
1310 		unlock_page(fio.page);
1311 	}
1312 
1313 	if (fio.compr_blocks)
1314 		f2fs_i_compr_blocks_update(inode, fio.compr_blocks - 1, false);
1315 	f2fs_i_compr_blocks_update(inode, cc->nr_cpages, true);
1316 	add_compr_block_stat(inode, cc->nr_cpages);
1317 
1318 	set_inode_flag(cc->inode, FI_APPEND_WRITE);
1319 	if (cc->cluster_idx == 0)
1320 		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1321 
1322 	f2fs_put_dnode(&dn);
1323 	if (IS_NOQUOTA(inode))
1324 		up_read(&sbi->node_write);
1325 	else
1326 		f2fs_unlock_op(sbi);
1327 
1328 	spin_lock(&fi->i_size_lock);
1329 	if (fi->last_disk_size < psize)
1330 		fi->last_disk_size = psize;
1331 	spin_unlock(&fi->i_size_lock);
1332 
1333 	f2fs_put_rpages(cc);
1334 	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1335 	cc->cpages = NULL;
1336 	f2fs_destroy_compress_ctx(cc, false);
1337 	return 0;
1338 
1339 out_destroy_crypt:
1340 	page_array_free(cc->inode, cic->rpages, cc->cluster_size);
1341 
1342 	for (--i; i >= 0; i--)
1343 		fscrypt_finalize_bounce_page(&cc->cpages[i]);
1344 	for (i = 0; i < cc->nr_cpages; i++) {
1345 		if (!cc->cpages[i])
1346 			continue;
1347 		f2fs_compress_free_page(cc->cpages[i]);
1348 		cc->cpages[i] = NULL;
1349 	}
1350 out_put_cic:
1351 	kmem_cache_free(cic_entry_slab, cic);
1352 out_put_dnode:
1353 	f2fs_put_dnode(&dn);
1354 out_unlock_op:
1355 	if (IS_NOQUOTA(inode))
1356 		up_read(&sbi->node_write);
1357 	else
1358 		f2fs_unlock_op(sbi);
1359 out_free:
1360 	page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1361 	cc->cpages = NULL;
1362 	return -EAGAIN;
1363 }
1364 
1365 void f2fs_compress_write_end_io(struct bio *bio, struct page *page)
1366 {
1367 	struct f2fs_sb_info *sbi = bio->bi_private;
1368 	struct compress_io_ctx *cic =
1369 			(struct compress_io_ctx *)page_private(page);
1370 	int i;
1371 
1372 	if (unlikely(bio->bi_status))
1373 		mapping_set_error(cic->inode->i_mapping, -EIO);
1374 
1375 	f2fs_compress_free_page(page);
1376 
1377 	dec_page_count(sbi, F2FS_WB_DATA);
1378 
1379 	if (atomic_dec_return(&cic->pending_pages))
1380 		return;
1381 
1382 	for (i = 0; i < cic->nr_rpages; i++) {
1383 		WARN_ON(!cic->rpages[i]);
1384 		clear_page_private_gcing(cic->rpages[i]);
1385 		end_page_writeback(cic->rpages[i]);
1386 	}
1387 
1388 	page_array_free(cic->inode, cic->rpages, cic->nr_rpages);
1389 	kmem_cache_free(cic_entry_slab, cic);
1390 }
1391 
1392 static int f2fs_write_raw_pages(struct compress_ctx *cc,
1393 					int *submitted,
1394 					struct writeback_control *wbc,
1395 					enum iostat_type io_type)
1396 {
1397 	struct address_space *mapping = cc->inode->i_mapping;
1398 	int _submitted, compr_blocks, ret;
1399 	int i = -1, err = 0;
1400 
1401 	compr_blocks = f2fs_compressed_blocks(cc);
1402 	if (compr_blocks < 0) {
1403 		err = compr_blocks;
1404 		goto out_err;
1405 	}
1406 
1407 	for (i = 0; i < cc->cluster_size; i++) {
1408 		if (!cc->rpages[i])
1409 			continue;
1410 retry_write:
1411 		if (cc->rpages[i]->mapping != mapping) {
1412 			unlock_page(cc->rpages[i]);
1413 			continue;
1414 		}
1415 
1416 		BUG_ON(!PageLocked(cc->rpages[i]));
1417 
1418 		ret = f2fs_write_single_data_page(cc->rpages[i], &_submitted,
1419 						NULL, NULL, wbc, io_type,
1420 						compr_blocks, false);
1421 		if (ret) {
1422 			if (ret == AOP_WRITEPAGE_ACTIVATE) {
1423 				unlock_page(cc->rpages[i]);
1424 				ret = 0;
1425 			} else if (ret == -EAGAIN) {
1426 				/*
1427 				 * for quota file, just redirty left pages to
1428 				 * avoid deadlock caused by cluster update race
1429 				 * from foreground operation.
1430 				 */
1431 				if (IS_NOQUOTA(cc->inode)) {
1432 					err = 0;
1433 					goto out_err;
1434 				}
1435 				ret = 0;
1436 				cond_resched();
1437 				congestion_wait(BLK_RW_ASYNC,
1438 						DEFAULT_IO_TIMEOUT);
1439 				lock_page(cc->rpages[i]);
1440 
1441 				if (!PageDirty(cc->rpages[i])) {
1442 					unlock_page(cc->rpages[i]);
1443 					continue;
1444 				}
1445 
1446 				clear_page_dirty_for_io(cc->rpages[i]);
1447 				goto retry_write;
1448 			}
1449 			err = ret;
1450 			goto out_err;
1451 		}
1452 
1453 		*submitted += _submitted;
1454 	}
1455 
1456 	f2fs_balance_fs(F2FS_M_SB(mapping), true);
1457 
1458 	return 0;
1459 out_err:
1460 	for (++i; i < cc->cluster_size; i++) {
1461 		if (!cc->rpages[i])
1462 			continue;
1463 		redirty_page_for_writepage(wbc, cc->rpages[i]);
1464 		unlock_page(cc->rpages[i]);
1465 	}
1466 	return err;
1467 }
1468 
1469 int f2fs_write_multi_pages(struct compress_ctx *cc,
1470 					int *submitted,
1471 					struct writeback_control *wbc,
1472 					enum iostat_type io_type)
1473 {
1474 	int err;
1475 
1476 	*submitted = 0;
1477 	if (cluster_may_compress(cc)) {
1478 		err = f2fs_compress_pages(cc);
1479 		if (err == -EAGAIN) {
1480 			goto write;
1481 		} else if (err) {
1482 			f2fs_put_rpages_wbc(cc, wbc, true, 1);
1483 			goto destroy_out;
1484 		}
1485 
1486 		err = f2fs_write_compressed_pages(cc, submitted,
1487 							wbc, io_type);
1488 		if (!err)
1489 			return 0;
1490 		f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN);
1491 	}
1492 write:
1493 	f2fs_bug_on(F2FS_I_SB(cc->inode), *submitted);
1494 
1495 	err = f2fs_write_raw_pages(cc, submitted, wbc, io_type);
1496 	f2fs_put_rpages_wbc(cc, wbc, false, 0);
1497 destroy_out:
1498 	f2fs_destroy_compress_ctx(cc, false);
1499 	return err;
1500 }
1501 
1502 static void f2fs_free_dic(struct decompress_io_ctx *dic);
1503 
1504 struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
1505 {
1506 	struct decompress_io_ctx *dic;
1507 	pgoff_t start_idx = start_idx_of_cluster(cc);
1508 	int i;
1509 
1510 	dic = f2fs_kmem_cache_alloc(dic_entry_slab, GFP_F2FS_ZERO,
1511 					false, F2FS_I_SB(cc->inode));
1512 	if (!dic)
1513 		return ERR_PTR(-ENOMEM);
1514 
1515 	dic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1516 	if (!dic->rpages) {
1517 		kmem_cache_free(dic_entry_slab, dic);
1518 		return ERR_PTR(-ENOMEM);
1519 	}
1520 
1521 	dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1522 	dic->inode = cc->inode;
1523 	atomic_set(&dic->remaining_pages, cc->nr_cpages);
1524 	dic->cluster_idx = cc->cluster_idx;
1525 	dic->cluster_size = cc->cluster_size;
1526 	dic->log_cluster_size = cc->log_cluster_size;
1527 	dic->nr_cpages = cc->nr_cpages;
1528 	refcount_set(&dic->refcnt, 1);
1529 	dic->failed = false;
1530 	dic->need_verity = f2fs_need_verity(cc->inode, start_idx);
1531 
1532 	for (i = 0; i < dic->cluster_size; i++)
1533 		dic->rpages[i] = cc->rpages[i];
1534 	dic->nr_rpages = cc->cluster_size;
1535 
1536 	dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages);
1537 	if (!dic->cpages)
1538 		goto out_free;
1539 
1540 	for (i = 0; i < dic->nr_cpages; i++) {
1541 		struct page *page;
1542 
1543 		page = f2fs_compress_alloc_page();
1544 		if (!page)
1545 			goto out_free;
1546 
1547 		f2fs_set_compressed_page(page, cc->inode,
1548 					start_idx + i + 1, dic);
1549 		dic->cpages[i] = page;
1550 	}
1551 
1552 	return dic;
1553 
1554 out_free:
1555 	f2fs_free_dic(dic);
1556 	return ERR_PTR(-ENOMEM);
1557 }
1558 
1559 static void f2fs_free_dic(struct decompress_io_ctx *dic)
1560 {
1561 	int i;
1562 
1563 	if (dic->tpages) {
1564 		for (i = 0; i < dic->cluster_size; i++) {
1565 			if (dic->rpages[i])
1566 				continue;
1567 			if (!dic->tpages[i])
1568 				continue;
1569 			f2fs_compress_free_page(dic->tpages[i]);
1570 		}
1571 		page_array_free(dic->inode, dic->tpages, dic->cluster_size);
1572 	}
1573 
1574 	if (dic->cpages) {
1575 		for (i = 0; i < dic->nr_cpages; i++) {
1576 			if (!dic->cpages[i])
1577 				continue;
1578 			f2fs_compress_free_page(dic->cpages[i]);
1579 		}
1580 		page_array_free(dic->inode, dic->cpages, dic->nr_cpages);
1581 	}
1582 
1583 	page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
1584 	kmem_cache_free(dic_entry_slab, dic);
1585 }
1586 
1587 static void f2fs_put_dic(struct decompress_io_ctx *dic)
1588 {
1589 	if (refcount_dec_and_test(&dic->refcnt))
1590 		f2fs_free_dic(dic);
1591 }
1592 
1593 /*
1594  * Update and unlock the cluster's pagecache pages, and release the reference to
1595  * the decompress_io_ctx that was being held for I/O completion.
1596  */
1597 static void __f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1598 {
1599 	int i;
1600 
1601 	for (i = 0; i < dic->cluster_size; i++) {
1602 		struct page *rpage = dic->rpages[i];
1603 
1604 		if (!rpage)
1605 			continue;
1606 
1607 		/* PG_error was set if verity failed. */
1608 		if (failed || PageError(rpage)) {
1609 			ClearPageUptodate(rpage);
1610 			/* will re-read again later */
1611 			ClearPageError(rpage);
1612 		} else {
1613 			SetPageUptodate(rpage);
1614 		}
1615 		unlock_page(rpage);
1616 	}
1617 
1618 	f2fs_put_dic(dic);
1619 }
1620 
1621 static void f2fs_verify_cluster(struct work_struct *work)
1622 {
1623 	struct decompress_io_ctx *dic =
1624 		container_of(work, struct decompress_io_ctx, verity_work);
1625 	int i;
1626 
1627 	/* Verify the cluster's decompressed pages with fs-verity. */
1628 	for (i = 0; i < dic->cluster_size; i++) {
1629 		struct page *rpage = dic->rpages[i];
1630 
1631 		if (rpage && !fsverity_verify_page(rpage))
1632 			SetPageError(rpage);
1633 	}
1634 
1635 	__f2fs_decompress_end_io(dic, false);
1636 }
1637 
1638 /*
1639  * This is called when a compressed cluster has been decompressed
1640  * (or failed to be read and/or decompressed).
1641  */
1642 void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1643 {
1644 	if (!failed && dic->need_verity) {
1645 		/*
1646 		 * Note that to avoid deadlocks, the verity work can't be done
1647 		 * on the decompression workqueue.  This is because verifying
1648 		 * the data pages can involve reading metadata pages from the
1649 		 * file, and these metadata pages may be compressed.
1650 		 */
1651 		INIT_WORK(&dic->verity_work, f2fs_verify_cluster);
1652 		fsverity_enqueue_verify_work(&dic->verity_work);
1653 	} else {
1654 		__f2fs_decompress_end_io(dic, failed);
1655 	}
1656 }
1657 
1658 /*
1659  * Put a reference to a compressed page's decompress_io_ctx.
1660  *
1661  * This is called when the page is no longer needed and can be freed.
1662  */
1663 void f2fs_put_page_dic(struct page *page)
1664 {
1665 	struct decompress_io_ctx *dic =
1666 			(struct decompress_io_ctx *)page_private(page);
1667 
1668 	f2fs_put_dic(dic);
1669 }
1670 
1671 /*
1672  * check whether cluster blocks are contiguous, and add extent cache entry
1673  * only if cluster blocks are logically and physically contiguous.
1674  */
1675 unsigned int f2fs_cluster_blocks_are_contiguous(struct dnode_of_data *dn)
1676 {
1677 	bool compressed = f2fs_data_blkaddr(dn) == COMPRESS_ADDR;
1678 	int i = compressed ? 1 : 0;
1679 	block_t first_blkaddr = data_blkaddr(dn->inode, dn->node_page,
1680 						dn->ofs_in_node + i);
1681 
1682 	for (i += 1; i < F2FS_I(dn->inode)->i_cluster_size; i++) {
1683 		block_t blkaddr = data_blkaddr(dn->inode, dn->node_page,
1684 						dn->ofs_in_node + i);
1685 
1686 		if (!__is_valid_data_blkaddr(blkaddr))
1687 			break;
1688 		if (first_blkaddr + i - (compressed ? 1 : 0) != blkaddr)
1689 			return 0;
1690 	}
1691 
1692 	return compressed ? i - 1 : i;
1693 }
1694 
1695 const struct address_space_operations f2fs_compress_aops = {
1696 	.releasepage = f2fs_release_page,
1697 	.invalidatepage = f2fs_invalidate_page,
1698 };
1699 
1700 struct address_space *COMPRESS_MAPPING(struct f2fs_sb_info *sbi)
1701 {
1702 	return sbi->compress_inode->i_mapping;
1703 }
1704 
1705 void f2fs_invalidate_compress_page(struct f2fs_sb_info *sbi, block_t blkaddr)
1706 {
1707 	if (!sbi->compress_inode)
1708 		return;
1709 	invalidate_mapping_pages(COMPRESS_MAPPING(sbi), blkaddr, blkaddr);
1710 }
1711 
1712 void f2fs_cache_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
1713 						nid_t ino, block_t blkaddr)
1714 {
1715 	struct page *cpage;
1716 	int ret;
1717 
1718 	if (!test_opt(sbi, COMPRESS_CACHE))
1719 		return;
1720 
1721 	if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
1722 		return;
1723 
1724 	if (!f2fs_available_free_memory(sbi, COMPRESS_PAGE))
1725 		return;
1726 
1727 	cpage = find_get_page(COMPRESS_MAPPING(sbi), blkaddr);
1728 	if (cpage) {
1729 		f2fs_put_page(cpage, 0);
1730 		return;
1731 	}
1732 
1733 	cpage = alloc_page(__GFP_NOWARN | __GFP_IO);
1734 	if (!cpage)
1735 		return;
1736 
1737 	ret = add_to_page_cache_lru(cpage, COMPRESS_MAPPING(sbi),
1738 						blkaddr, GFP_NOFS);
1739 	if (ret) {
1740 		f2fs_put_page(cpage, 0);
1741 		return;
1742 	}
1743 
1744 	set_page_private_data(cpage, ino);
1745 
1746 	if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
1747 		goto out;
1748 
1749 	memcpy(page_address(cpage), page_address(page), PAGE_SIZE);
1750 	SetPageUptodate(cpage);
1751 out:
1752 	f2fs_put_page(cpage, 1);
1753 }
1754 
1755 bool f2fs_load_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
1756 								block_t blkaddr)
1757 {
1758 	struct page *cpage;
1759 	bool hitted = false;
1760 
1761 	if (!test_opt(sbi, COMPRESS_CACHE))
1762 		return false;
1763 
1764 	cpage = f2fs_pagecache_get_page(COMPRESS_MAPPING(sbi),
1765 				blkaddr, FGP_LOCK | FGP_NOWAIT, GFP_NOFS);
1766 	if (cpage) {
1767 		if (PageUptodate(cpage)) {
1768 			atomic_inc(&sbi->compress_page_hit);
1769 			memcpy(page_address(page),
1770 				page_address(cpage), PAGE_SIZE);
1771 			hitted = true;
1772 		}
1773 		f2fs_put_page(cpage, 1);
1774 	}
1775 
1776 	return hitted;
1777 }
1778 
1779 void f2fs_invalidate_compress_pages(struct f2fs_sb_info *sbi, nid_t ino)
1780 {
1781 	struct address_space *mapping = sbi->compress_inode->i_mapping;
1782 	struct pagevec pvec;
1783 	pgoff_t index = 0;
1784 	pgoff_t end = MAX_BLKADDR(sbi);
1785 
1786 	if (!mapping->nrpages)
1787 		return;
1788 
1789 	pagevec_init(&pvec);
1790 
1791 	do {
1792 		unsigned int nr_pages;
1793 		int i;
1794 
1795 		nr_pages = pagevec_lookup_range(&pvec, mapping,
1796 						&index, end - 1);
1797 		if (!nr_pages)
1798 			break;
1799 
1800 		for (i = 0; i < nr_pages; i++) {
1801 			struct page *page = pvec.pages[i];
1802 
1803 			if (page->index > end)
1804 				break;
1805 
1806 			lock_page(page);
1807 			if (page->mapping != mapping) {
1808 				unlock_page(page);
1809 				continue;
1810 			}
1811 
1812 			if (ino != get_page_private_data(page)) {
1813 				unlock_page(page);
1814 				continue;
1815 			}
1816 
1817 			generic_error_remove_page(mapping, page);
1818 			unlock_page(page);
1819 		}
1820 		pagevec_release(&pvec);
1821 		cond_resched();
1822 	} while (index < end);
1823 }
1824 
1825 int f2fs_init_compress_inode(struct f2fs_sb_info *sbi)
1826 {
1827 	struct inode *inode;
1828 
1829 	if (!test_opt(sbi, COMPRESS_CACHE))
1830 		return 0;
1831 
1832 	inode = f2fs_iget(sbi->sb, F2FS_COMPRESS_INO(sbi));
1833 	if (IS_ERR(inode))
1834 		return PTR_ERR(inode);
1835 	sbi->compress_inode = inode;
1836 
1837 	sbi->compress_percent = COMPRESS_PERCENT;
1838 	sbi->compress_watermark = COMPRESS_WATERMARK;
1839 
1840 	atomic_set(&sbi->compress_page_hit, 0);
1841 
1842 	return 0;
1843 }
1844 
1845 void f2fs_destroy_compress_inode(struct f2fs_sb_info *sbi)
1846 {
1847 	if (!sbi->compress_inode)
1848 		return;
1849 	iput(sbi->compress_inode);
1850 	sbi->compress_inode = NULL;
1851 }
1852 
1853 int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi)
1854 {
1855 	dev_t dev = sbi->sb->s_bdev->bd_dev;
1856 	char slab_name[32];
1857 
1858 	sprintf(slab_name, "f2fs_page_array_entry-%u:%u", MAJOR(dev), MINOR(dev));
1859 
1860 	sbi->page_array_slab_size = sizeof(struct page *) <<
1861 					F2FS_OPTION(sbi).compress_log_size;
1862 
1863 	sbi->page_array_slab = f2fs_kmem_cache_create(slab_name,
1864 					sbi->page_array_slab_size);
1865 	if (!sbi->page_array_slab)
1866 		return -ENOMEM;
1867 	return 0;
1868 }
1869 
1870 void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi)
1871 {
1872 	kmem_cache_destroy(sbi->page_array_slab);
1873 }
1874 
1875 static int __init f2fs_init_cic_cache(void)
1876 {
1877 	cic_entry_slab = f2fs_kmem_cache_create("f2fs_cic_entry",
1878 					sizeof(struct compress_io_ctx));
1879 	if (!cic_entry_slab)
1880 		return -ENOMEM;
1881 	return 0;
1882 }
1883 
1884 static void f2fs_destroy_cic_cache(void)
1885 {
1886 	kmem_cache_destroy(cic_entry_slab);
1887 }
1888 
1889 static int __init f2fs_init_dic_cache(void)
1890 {
1891 	dic_entry_slab = f2fs_kmem_cache_create("f2fs_dic_entry",
1892 					sizeof(struct decompress_io_ctx));
1893 	if (!dic_entry_slab)
1894 		return -ENOMEM;
1895 	return 0;
1896 }
1897 
1898 static void f2fs_destroy_dic_cache(void)
1899 {
1900 	kmem_cache_destroy(dic_entry_slab);
1901 }
1902 
1903 int __init f2fs_init_compress_cache(void)
1904 {
1905 	int err;
1906 
1907 	err = f2fs_init_cic_cache();
1908 	if (err)
1909 		goto out;
1910 	err = f2fs_init_dic_cache();
1911 	if (err)
1912 		goto free_cic;
1913 	return 0;
1914 free_cic:
1915 	f2fs_destroy_cic_cache();
1916 out:
1917 	return -ENOMEM;
1918 }
1919 
1920 void f2fs_destroy_compress_cache(void)
1921 {
1922 	f2fs_destroy_dic_cache();
1923 	f2fs_destroy_cic_cache();
1924 }
1925