GNU Linux-libre 5.15.137-gnu
[releases.git] / fs / f2fs / compress.c
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 ret:%d, "
316                                         "expected:%lu\n", KERN_ERR,
317                                         F2FS_I_SB(dic->inode)->sb->s_id, ret,
318                                         PAGE_SIZE << dic->log_cluster_size);
319                 return -EIO;
320         }
321         return 0;
322 }
323
324 static const struct f2fs_compress_ops f2fs_lz4_ops = {
325         .init_compress_ctx      = lz4_init_compress_ctx,
326         .destroy_compress_ctx   = lz4_destroy_compress_ctx,
327         .compress_pages         = lz4_compress_pages,
328         .decompress_pages       = lz4_decompress_pages,
329 };
330 #endif
331
332 #ifdef CONFIG_F2FS_FS_ZSTD
333 #define F2FS_ZSTD_DEFAULT_CLEVEL        1
334
335 static int zstd_init_compress_ctx(struct compress_ctx *cc)
336 {
337         ZSTD_parameters params;
338         ZSTD_CStream *stream;
339         void *workspace;
340         unsigned int workspace_size;
341         unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
342                                                 COMPRESS_LEVEL_OFFSET;
343
344         if (!level)
345                 level = F2FS_ZSTD_DEFAULT_CLEVEL;
346
347         params = ZSTD_getParams(level, cc->rlen, 0);
348         workspace_size = ZSTD_CStreamWorkspaceBound(params.cParams);
349
350         workspace = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
351                                         workspace_size, GFP_NOFS);
352         if (!workspace)
353                 return -ENOMEM;
354
355         stream = ZSTD_initCStream(params, 0, workspace, workspace_size);
356         if (!stream) {
357                 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initCStream failed\n",
358                                 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
359                                 __func__);
360                 kvfree(workspace);
361                 return -EIO;
362         }
363
364         cc->private = workspace;
365         cc->private2 = stream;
366
367         cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
368         return 0;
369 }
370
371 static void zstd_destroy_compress_ctx(struct compress_ctx *cc)
372 {
373         kvfree(cc->private);
374         cc->private = NULL;
375         cc->private2 = NULL;
376 }
377
378 static int zstd_compress_pages(struct compress_ctx *cc)
379 {
380         ZSTD_CStream *stream = cc->private2;
381         ZSTD_inBuffer inbuf;
382         ZSTD_outBuffer outbuf;
383         int src_size = cc->rlen;
384         int dst_size = src_size - PAGE_SIZE - COMPRESS_HEADER_SIZE;
385         int ret;
386
387         inbuf.pos = 0;
388         inbuf.src = cc->rbuf;
389         inbuf.size = src_size;
390
391         outbuf.pos = 0;
392         outbuf.dst = cc->cbuf->cdata;
393         outbuf.size = dst_size;
394
395         ret = ZSTD_compressStream(stream, &outbuf, &inbuf);
396         if (ZSTD_isError(ret)) {
397                 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
398                                 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
399                                 __func__, ZSTD_getErrorCode(ret));
400                 return -EIO;
401         }
402
403         ret = ZSTD_endStream(stream, &outbuf);
404         if (ZSTD_isError(ret)) {
405                 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_endStream returned %d\n",
406                                 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
407                                 __func__, ZSTD_getErrorCode(ret));
408                 return -EIO;
409         }
410
411         /*
412          * there is compressed data remained in intermediate buffer due to
413          * no more space in cbuf.cdata
414          */
415         if (ret)
416                 return -EAGAIN;
417
418         cc->clen = outbuf.pos;
419         return 0;
420 }
421
422 static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
423 {
424         ZSTD_DStream *stream;
425         void *workspace;
426         unsigned int workspace_size;
427         unsigned int max_window_size =
428                         MAX_COMPRESS_WINDOW_SIZE(dic->log_cluster_size);
429
430         workspace_size = ZSTD_DStreamWorkspaceBound(max_window_size);
431
432         workspace = f2fs_kvmalloc(F2FS_I_SB(dic->inode),
433                                         workspace_size, GFP_NOFS);
434         if (!workspace)
435                 return -ENOMEM;
436
437         stream = ZSTD_initDStream(max_window_size, workspace, workspace_size);
438         if (!stream) {
439                 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initDStream failed\n",
440                                 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
441                                 __func__);
442                 kvfree(workspace);
443                 return -EIO;
444         }
445
446         dic->private = workspace;
447         dic->private2 = stream;
448
449         return 0;
450 }
451
452 static void zstd_destroy_decompress_ctx(struct decompress_io_ctx *dic)
453 {
454         kvfree(dic->private);
455         dic->private = NULL;
456         dic->private2 = NULL;
457 }
458
459 static int zstd_decompress_pages(struct decompress_io_ctx *dic)
460 {
461         ZSTD_DStream *stream = dic->private2;
462         ZSTD_inBuffer inbuf;
463         ZSTD_outBuffer outbuf;
464         int ret;
465
466         inbuf.pos = 0;
467         inbuf.src = dic->cbuf->cdata;
468         inbuf.size = dic->clen;
469
470         outbuf.pos = 0;
471         outbuf.dst = dic->rbuf;
472         outbuf.size = dic->rlen;
473
474         ret = ZSTD_decompressStream(stream, &outbuf, &inbuf);
475         if (ZSTD_isError(ret)) {
476                 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
477                                 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
478                                 __func__, ZSTD_getErrorCode(ret));
479                 return -EIO;
480         }
481
482         if (dic->rlen != outbuf.pos) {
483                 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD invalid rlen:%zu, "
484                                 "expected:%lu\n", KERN_ERR,
485                                 F2FS_I_SB(dic->inode)->sb->s_id,
486                                 __func__, dic->rlen,
487                                 PAGE_SIZE << dic->log_cluster_size);
488                 return -EIO;
489         }
490
491         return 0;
492 }
493
494 static const struct f2fs_compress_ops f2fs_zstd_ops = {
495         .init_compress_ctx      = zstd_init_compress_ctx,
496         .destroy_compress_ctx   = zstd_destroy_compress_ctx,
497         .compress_pages         = zstd_compress_pages,
498         .init_decompress_ctx    = zstd_init_decompress_ctx,
499         .destroy_decompress_ctx = zstd_destroy_decompress_ctx,
500         .decompress_pages       = zstd_decompress_pages,
501 };
502 #endif
503
504 #ifdef CONFIG_F2FS_FS_LZO
505 #ifdef CONFIG_F2FS_FS_LZORLE
506 static int lzorle_compress_pages(struct compress_ctx *cc)
507 {
508         int ret;
509
510         ret = lzorle1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
511                                         &cc->clen, cc->private);
512         if (ret != LZO_E_OK) {
513                 printk_ratelimited("%sF2FS-fs (%s): lzo-rle compress failed, ret:%d\n",
514                                 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
515                 return -EIO;
516         }
517         return 0;
518 }
519
520 static const struct f2fs_compress_ops f2fs_lzorle_ops = {
521         .init_compress_ctx      = lzo_init_compress_ctx,
522         .destroy_compress_ctx   = lzo_destroy_compress_ctx,
523         .compress_pages         = lzorle_compress_pages,
524         .decompress_pages       = lzo_decompress_pages,
525 };
526 #endif
527 #endif
528
529 static const struct f2fs_compress_ops *f2fs_cops[COMPRESS_MAX] = {
530 #ifdef CONFIG_F2FS_FS_LZO
531         &f2fs_lzo_ops,
532 #else
533         NULL,
534 #endif
535 #ifdef CONFIG_F2FS_FS_LZ4
536         &f2fs_lz4_ops,
537 #else
538         NULL,
539 #endif
540 #ifdef CONFIG_F2FS_FS_ZSTD
541         &f2fs_zstd_ops,
542 #else
543         NULL,
544 #endif
545 #if defined(CONFIG_F2FS_FS_LZO) && defined(CONFIG_F2FS_FS_LZORLE)
546         &f2fs_lzorle_ops,
547 #else
548         NULL,
549 #endif
550 };
551
552 bool f2fs_is_compress_backend_ready(struct inode *inode)
553 {
554         if (!f2fs_compressed_file(inode))
555                 return true;
556         return f2fs_cops[F2FS_I(inode)->i_compress_algorithm];
557 }
558
559 static mempool_t *compress_page_pool;
560 static int num_compress_pages = 512;
561 module_param(num_compress_pages, uint, 0444);
562 MODULE_PARM_DESC(num_compress_pages,
563                 "Number of intermediate compress pages to preallocate");
564
565 int f2fs_init_compress_mempool(void)
566 {
567         compress_page_pool = mempool_create_page_pool(num_compress_pages, 0);
568         if (!compress_page_pool)
569                 return -ENOMEM;
570
571         return 0;
572 }
573
574 void f2fs_destroy_compress_mempool(void)
575 {
576         mempool_destroy(compress_page_pool);
577 }
578
579 static struct page *f2fs_compress_alloc_page(void)
580 {
581         struct page *page;
582
583         page = mempool_alloc(compress_page_pool, GFP_NOFS);
584         lock_page(page);
585
586         return page;
587 }
588
589 static void f2fs_compress_free_page(struct page *page)
590 {
591         if (!page)
592                 return;
593         detach_page_private(page);
594         page->mapping = NULL;
595         unlock_page(page);
596         mempool_free(page, compress_page_pool);
597 }
598
599 #define MAX_VMAP_RETRIES        3
600
601 static void *f2fs_vmap(struct page **pages, unsigned int count)
602 {
603         int i;
604         void *buf = NULL;
605
606         for (i = 0; i < MAX_VMAP_RETRIES; i++) {
607                 buf = vm_map_ram(pages, count, -1);
608                 if (buf)
609                         break;
610                 vm_unmap_aliases();
611         }
612         return buf;
613 }
614
615 static int f2fs_compress_pages(struct compress_ctx *cc)
616 {
617         struct f2fs_inode_info *fi = F2FS_I(cc->inode);
618         const struct f2fs_compress_ops *cops =
619                                 f2fs_cops[fi->i_compress_algorithm];
620         unsigned int max_len, new_nr_cpages;
621         struct page **new_cpages;
622         u32 chksum = 0;
623         int i, ret;
624
625         trace_f2fs_compress_pages_start(cc->inode, cc->cluster_idx,
626                                 cc->cluster_size, fi->i_compress_algorithm);
627
628         if (cops->init_compress_ctx) {
629                 ret = cops->init_compress_ctx(cc);
630                 if (ret)
631                         goto out;
632         }
633
634         max_len = COMPRESS_HEADER_SIZE + cc->clen;
635         cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE);
636
637         cc->cpages = page_array_alloc(cc->inode, cc->nr_cpages);
638         if (!cc->cpages) {
639                 ret = -ENOMEM;
640                 goto destroy_compress_ctx;
641         }
642
643         for (i = 0; i < cc->nr_cpages; i++) {
644                 cc->cpages[i] = f2fs_compress_alloc_page();
645                 if (!cc->cpages[i]) {
646                         ret = -ENOMEM;
647                         goto out_free_cpages;
648                 }
649         }
650
651         cc->rbuf = f2fs_vmap(cc->rpages, cc->cluster_size);
652         if (!cc->rbuf) {
653                 ret = -ENOMEM;
654                 goto out_free_cpages;
655         }
656
657         cc->cbuf = f2fs_vmap(cc->cpages, cc->nr_cpages);
658         if (!cc->cbuf) {
659                 ret = -ENOMEM;
660                 goto out_vunmap_rbuf;
661         }
662
663         ret = cops->compress_pages(cc);
664         if (ret)
665                 goto out_vunmap_cbuf;
666
667         max_len = PAGE_SIZE * (cc->cluster_size - 1) - COMPRESS_HEADER_SIZE;
668
669         if (cc->clen > max_len) {
670                 ret = -EAGAIN;
671                 goto out_vunmap_cbuf;
672         }
673
674         cc->cbuf->clen = cpu_to_le32(cc->clen);
675
676         if (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)
677                 chksum = f2fs_crc32(F2FS_I_SB(cc->inode),
678                                         cc->cbuf->cdata, cc->clen);
679         cc->cbuf->chksum = cpu_to_le32(chksum);
680
681         for (i = 0; i < COMPRESS_DATA_RESERVED_SIZE; i++)
682                 cc->cbuf->reserved[i] = cpu_to_le32(0);
683
684         new_nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE);
685
686         /* Now we're going to cut unnecessary tail pages */
687         new_cpages = page_array_alloc(cc->inode, new_nr_cpages);
688         if (!new_cpages) {
689                 ret = -ENOMEM;
690                 goto out_vunmap_cbuf;
691         }
692
693         /* zero out any unused part of the last page */
694         memset(&cc->cbuf->cdata[cc->clen], 0,
695                         (new_nr_cpages * PAGE_SIZE) -
696                         (cc->clen + COMPRESS_HEADER_SIZE));
697
698         vm_unmap_ram(cc->cbuf, cc->nr_cpages);
699         vm_unmap_ram(cc->rbuf, cc->cluster_size);
700
701         for (i = 0; i < cc->nr_cpages; i++) {
702                 if (i < new_nr_cpages) {
703                         new_cpages[i] = cc->cpages[i];
704                         continue;
705                 }
706                 f2fs_compress_free_page(cc->cpages[i]);
707                 cc->cpages[i] = NULL;
708         }
709
710         if (cops->destroy_compress_ctx)
711                 cops->destroy_compress_ctx(cc);
712
713         page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
714         cc->cpages = new_cpages;
715         cc->nr_cpages = new_nr_cpages;
716
717         trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
718                                                         cc->clen, ret);
719         return 0;
720
721 out_vunmap_cbuf:
722         vm_unmap_ram(cc->cbuf, cc->nr_cpages);
723 out_vunmap_rbuf:
724         vm_unmap_ram(cc->rbuf, cc->cluster_size);
725 out_free_cpages:
726         for (i = 0; i < cc->nr_cpages; i++) {
727                 if (cc->cpages[i])
728                         f2fs_compress_free_page(cc->cpages[i]);
729         }
730         page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
731         cc->cpages = NULL;
732 destroy_compress_ctx:
733         if (cops->destroy_compress_ctx)
734                 cops->destroy_compress_ctx(cc);
735 out:
736         trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
737                                                         cc->clen, ret);
738         return ret;
739 }
740
741 void f2fs_decompress_cluster(struct decompress_io_ctx *dic)
742 {
743         struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
744         struct f2fs_inode_info *fi = F2FS_I(dic->inode);
745         const struct f2fs_compress_ops *cops =
746                         f2fs_cops[fi->i_compress_algorithm];
747         int ret;
748         int i;
749
750         trace_f2fs_decompress_pages_start(dic->inode, dic->cluster_idx,
751                                 dic->cluster_size, fi->i_compress_algorithm);
752
753         if (dic->failed) {
754                 ret = -EIO;
755                 goto out_end_io;
756         }
757
758         dic->tpages = page_array_alloc(dic->inode, dic->cluster_size);
759         if (!dic->tpages) {
760                 ret = -ENOMEM;
761                 goto out_end_io;
762         }
763
764         for (i = 0; i < dic->cluster_size; i++) {
765                 if (dic->rpages[i]) {
766                         dic->tpages[i] = dic->rpages[i];
767                         continue;
768                 }
769
770                 dic->tpages[i] = f2fs_compress_alloc_page();
771                 if (!dic->tpages[i]) {
772                         ret = -ENOMEM;
773                         goto out_end_io;
774                 }
775         }
776
777         if (cops->init_decompress_ctx) {
778                 ret = cops->init_decompress_ctx(dic);
779                 if (ret)
780                         goto out_end_io;
781         }
782
783         dic->rbuf = f2fs_vmap(dic->tpages, dic->cluster_size);
784         if (!dic->rbuf) {
785                 ret = -ENOMEM;
786                 goto out_destroy_decompress_ctx;
787         }
788
789         dic->cbuf = f2fs_vmap(dic->cpages, dic->nr_cpages);
790         if (!dic->cbuf) {
791                 ret = -ENOMEM;
792                 goto out_vunmap_rbuf;
793         }
794
795         dic->clen = le32_to_cpu(dic->cbuf->clen);
796         dic->rlen = PAGE_SIZE << dic->log_cluster_size;
797
798         if (dic->clen > PAGE_SIZE * dic->nr_cpages - COMPRESS_HEADER_SIZE) {
799                 ret = -EFSCORRUPTED;
800                 goto out_vunmap_cbuf;
801         }
802
803         ret = cops->decompress_pages(dic);
804
805         if (!ret && (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)) {
806                 u32 provided = le32_to_cpu(dic->cbuf->chksum);
807                 u32 calculated = f2fs_crc32(sbi, dic->cbuf->cdata, dic->clen);
808
809                 if (provided != calculated) {
810                         if (!is_inode_flag_set(dic->inode, FI_COMPRESS_CORRUPT)) {
811                                 set_inode_flag(dic->inode, FI_COMPRESS_CORRUPT);
812                                 printk_ratelimited(
813                                         "%sF2FS-fs (%s): checksum invalid, nid = %lu, %x vs %x",
814                                         KERN_INFO, sbi->sb->s_id, dic->inode->i_ino,
815                                         provided, calculated);
816                         }
817                         set_sbi_flag(sbi, SBI_NEED_FSCK);
818                 }
819         }
820
821 out_vunmap_cbuf:
822         vm_unmap_ram(dic->cbuf, dic->nr_cpages);
823 out_vunmap_rbuf:
824         vm_unmap_ram(dic->rbuf, dic->cluster_size);
825 out_destroy_decompress_ctx:
826         if (cops->destroy_decompress_ctx)
827                 cops->destroy_decompress_ctx(dic);
828 out_end_io:
829         trace_f2fs_decompress_pages_end(dic->inode, dic->cluster_idx,
830                                                         dic->clen, ret);
831         f2fs_decompress_end_io(dic, ret);
832 }
833
834 /*
835  * This is called when a page of a compressed cluster has been read from disk
836  * (or failed to be read from disk).  It checks whether this page was the last
837  * page being waited on in the cluster, and if so, it decompresses the cluster
838  * (or in the case of a failure, cleans up without actually decompressing).
839  */
840 void f2fs_end_read_compressed_page(struct page *page, bool failed,
841                                                 block_t blkaddr)
842 {
843         struct decompress_io_ctx *dic =
844                         (struct decompress_io_ctx *)page_private(page);
845         struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
846
847         dec_page_count(sbi, F2FS_RD_DATA);
848
849         if (failed)
850                 WRITE_ONCE(dic->failed, true);
851         else if (blkaddr)
852                 f2fs_cache_compressed_page(sbi, page,
853                                         dic->inode->i_ino, blkaddr);
854
855         if (atomic_dec_and_test(&dic->remaining_pages))
856                 f2fs_decompress_cluster(dic);
857 }
858
859 static bool is_page_in_cluster(struct compress_ctx *cc, pgoff_t index)
860 {
861         if (cc->cluster_idx == NULL_CLUSTER)
862                 return true;
863         return cc->cluster_idx == cluster_idx(cc, index);
864 }
865
866 bool f2fs_cluster_is_empty(struct compress_ctx *cc)
867 {
868         return cc->nr_rpages == 0;
869 }
870
871 static bool f2fs_cluster_is_full(struct compress_ctx *cc)
872 {
873         return cc->cluster_size == cc->nr_rpages;
874 }
875
876 bool f2fs_cluster_can_merge_page(struct compress_ctx *cc, pgoff_t index)
877 {
878         if (f2fs_cluster_is_empty(cc))
879                 return true;
880         return is_page_in_cluster(cc, index);
881 }
882
883 static bool cluster_has_invalid_data(struct compress_ctx *cc)
884 {
885         loff_t i_size = i_size_read(cc->inode);
886         unsigned nr_pages = DIV_ROUND_UP(i_size, PAGE_SIZE);
887         int i;
888
889         for (i = 0; i < cc->cluster_size; i++) {
890                 struct page *page = cc->rpages[i];
891
892                 f2fs_bug_on(F2FS_I_SB(cc->inode), !page);
893
894                 /* beyond EOF */
895                 if (page->index >= nr_pages)
896                         return true;
897         }
898         return false;
899 }
900
901 bool f2fs_sanity_check_cluster(struct dnode_of_data *dn)
902 {
903         struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
904         unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size;
905         bool compressed = dn->data_blkaddr == COMPRESS_ADDR;
906         int cluster_end = 0;
907         int i;
908         char *reason = "";
909
910         if (!compressed)
911                 return false;
912
913         /* [..., COMPR_ADDR, ...] */
914         if (dn->ofs_in_node % cluster_size) {
915                 reason = "[*|C|*|*]";
916                 goto out;
917         }
918
919         for (i = 1; i < cluster_size; i++) {
920                 block_t blkaddr = data_blkaddr(dn->inode, dn->node_page,
921                                                         dn->ofs_in_node + i);
922
923                 /* [COMPR_ADDR, ..., COMPR_ADDR] */
924                 if (blkaddr == COMPRESS_ADDR) {
925                         reason = "[C|*|C|*]";
926                         goto out;
927                 }
928                 if (compressed) {
929                         if (!__is_valid_data_blkaddr(blkaddr)) {
930                                 if (!cluster_end)
931                                         cluster_end = i;
932                                 continue;
933                         }
934                         /* [COMPR_ADDR, NULL_ADDR or NEW_ADDR, valid_blkaddr] */
935                         if (cluster_end) {
936                                 reason = "[C|N|N|V]";
937                                 goto out;
938                         }
939                 }
940         }
941         return false;
942 out:
943         f2fs_warn(sbi, "access invalid cluster, ino:%lu, nid:%u, ofs_in_node:%u, reason:%s",
944                         dn->inode->i_ino, dn->nid, dn->ofs_in_node, reason);
945         set_sbi_flag(sbi, SBI_NEED_FSCK);
946         return true;
947 }
948
949 static int __f2fs_cluster_blocks(struct inode *inode,
950                                 unsigned int cluster_idx, bool compr)
951 {
952         struct dnode_of_data dn;
953         unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
954         unsigned int start_idx = cluster_idx <<
955                                 F2FS_I(inode)->i_log_cluster_size;
956         int ret;
957
958         set_new_dnode(&dn, inode, NULL, NULL, 0);
959         ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
960         if (ret) {
961                 if (ret == -ENOENT)
962                         ret = 0;
963                 goto fail;
964         }
965
966         if (f2fs_sanity_check_cluster(&dn)) {
967                 ret = -EFSCORRUPTED;
968                 goto fail;
969         }
970
971         if (dn.data_blkaddr == COMPRESS_ADDR) {
972                 int i;
973
974                 ret = 1;
975                 for (i = 1; i < cluster_size; i++) {
976                         block_t blkaddr;
977
978                         blkaddr = data_blkaddr(dn.inode,
979                                         dn.node_page, dn.ofs_in_node + i);
980                         if (compr) {
981                                 if (__is_valid_data_blkaddr(blkaddr))
982                                         ret++;
983                         } else {
984                                 if (blkaddr != NULL_ADDR)
985                                         ret++;
986                         }
987                 }
988
989                 f2fs_bug_on(F2FS_I_SB(inode),
990                         !compr && ret != cluster_size &&
991                         !is_inode_flag_set(inode, FI_COMPRESS_RELEASED));
992         }
993 fail:
994         f2fs_put_dnode(&dn);
995         return ret;
996 }
997
998 /* return # of compressed blocks in compressed cluster */
999 static int f2fs_compressed_blocks(struct compress_ctx *cc)
1000 {
1001         return __f2fs_cluster_blocks(cc->inode, cc->cluster_idx, true);
1002 }
1003
1004 /* return # of valid blocks in compressed cluster */
1005 int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index)
1006 {
1007         return __f2fs_cluster_blocks(inode,
1008                 index >> F2FS_I(inode)->i_log_cluster_size,
1009                 false);
1010 }
1011
1012 static bool cluster_may_compress(struct compress_ctx *cc)
1013 {
1014         if (!f2fs_need_compress_data(cc->inode))
1015                 return false;
1016         if (f2fs_is_atomic_file(cc->inode))
1017                 return false;
1018         if (!f2fs_cluster_is_full(cc))
1019                 return false;
1020         if (unlikely(f2fs_cp_error(F2FS_I_SB(cc->inode))))
1021                 return false;
1022         return !cluster_has_invalid_data(cc);
1023 }
1024
1025 static void set_cluster_writeback(struct compress_ctx *cc)
1026 {
1027         int i;
1028
1029         for (i = 0; i < cc->cluster_size; i++) {
1030                 if (cc->rpages[i])
1031                         set_page_writeback(cc->rpages[i]);
1032         }
1033 }
1034
1035 static void set_cluster_dirty(struct compress_ctx *cc)
1036 {
1037         int i;
1038
1039         for (i = 0; i < cc->cluster_size; i++)
1040                 if (cc->rpages[i])
1041                         set_page_dirty(cc->rpages[i]);
1042 }
1043
1044 static int prepare_compress_overwrite(struct compress_ctx *cc,
1045                 struct page **pagep, pgoff_t index, void **fsdata)
1046 {
1047         struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
1048         struct address_space *mapping = cc->inode->i_mapping;
1049         struct page *page;
1050         sector_t last_block_in_bio;
1051         unsigned fgp_flag = FGP_LOCK | FGP_WRITE | FGP_CREAT;
1052         pgoff_t start_idx = start_idx_of_cluster(cc);
1053         int i, ret;
1054
1055 retry:
1056         ret = f2fs_is_compressed_cluster(cc->inode, start_idx);
1057         if (ret <= 0)
1058                 return ret;
1059
1060         ret = f2fs_init_compress_ctx(cc);
1061         if (ret)
1062                 return ret;
1063
1064         /* keep page reference to avoid page reclaim */
1065         for (i = 0; i < cc->cluster_size; i++) {
1066                 page = f2fs_pagecache_get_page(mapping, start_idx + i,
1067                                                         fgp_flag, GFP_NOFS);
1068                 if (!page) {
1069                         ret = -ENOMEM;
1070                         goto unlock_pages;
1071                 }
1072
1073                 if (PageUptodate(page))
1074                         f2fs_put_page(page, 1);
1075                 else
1076                         f2fs_compress_ctx_add_page(cc, page);
1077         }
1078
1079         if (!f2fs_cluster_is_empty(cc)) {
1080                 struct bio *bio = NULL;
1081
1082                 ret = f2fs_read_multi_pages(cc, &bio, cc->cluster_size,
1083                                         &last_block_in_bio, false, true);
1084                 f2fs_put_rpages(cc);
1085                 f2fs_destroy_compress_ctx(cc, true);
1086                 if (ret)
1087                         goto out;
1088                 if (bio)
1089                         f2fs_submit_bio(sbi, bio, DATA);
1090
1091                 ret = f2fs_init_compress_ctx(cc);
1092                 if (ret)
1093                         goto out;
1094         }
1095
1096         for (i = 0; i < cc->cluster_size; i++) {
1097                 f2fs_bug_on(sbi, cc->rpages[i]);
1098
1099                 page = find_lock_page(mapping, start_idx + i);
1100                 if (!page) {
1101                         /* page can be truncated */
1102                         goto release_and_retry;
1103                 }
1104
1105                 f2fs_wait_on_page_writeback(page, DATA, true, true);
1106                 f2fs_compress_ctx_add_page(cc, page);
1107
1108                 if (!PageUptodate(page)) {
1109 release_and_retry:
1110                         f2fs_put_rpages(cc);
1111                         f2fs_unlock_rpages(cc, i + 1);
1112                         f2fs_destroy_compress_ctx(cc, true);
1113                         goto retry;
1114                 }
1115         }
1116
1117         if (likely(!ret)) {
1118                 *fsdata = cc->rpages;
1119                 *pagep = cc->rpages[offset_in_cluster(cc, index)];
1120                 return cc->cluster_size;
1121         }
1122
1123 unlock_pages:
1124         f2fs_put_rpages(cc);
1125         f2fs_unlock_rpages(cc, i);
1126         f2fs_destroy_compress_ctx(cc, true);
1127 out:
1128         return ret;
1129 }
1130
1131 int f2fs_prepare_compress_overwrite(struct inode *inode,
1132                 struct page **pagep, pgoff_t index, void **fsdata)
1133 {
1134         struct compress_ctx cc = {
1135                 .inode = inode,
1136                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1137                 .cluster_size = F2FS_I(inode)->i_cluster_size,
1138                 .cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
1139                 .rpages = NULL,
1140                 .nr_rpages = 0,
1141         };
1142
1143         return prepare_compress_overwrite(&cc, pagep, index, fsdata);
1144 }
1145
1146 bool f2fs_compress_write_end(struct inode *inode, void *fsdata,
1147                                         pgoff_t index, unsigned copied)
1148
1149 {
1150         struct compress_ctx cc = {
1151                 .inode = inode,
1152                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1153                 .cluster_size = F2FS_I(inode)->i_cluster_size,
1154                 .rpages = fsdata,
1155         };
1156         bool first_index = (index == cc.rpages[0]->index);
1157
1158         if (copied)
1159                 set_cluster_dirty(&cc);
1160
1161         f2fs_put_rpages_wbc(&cc, NULL, false, 1);
1162         f2fs_destroy_compress_ctx(&cc, false);
1163
1164         return first_index;
1165 }
1166
1167 int f2fs_truncate_partial_cluster(struct inode *inode, u64 from, bool lock)
1168 {
1169         void *fsdata = NULL;
1170         struct page *pagep;
1171         int log_cluster_size = F2FS_I(inode)->i_log_cluster_size;
1172         pgoff_t start_idx = from >> (PAGE_SHIFT + log_cluster_size) <<
1173                                                         log_cluster_size;
1174         int err;
1175
1176         err = f2fs_is_compressed_cluster(inode, start_idx);
1177         if (err < 0)
1178                 return err;
1179
1180         /* truncate normal cluster */
1181         if (!err)
1182                 return f2fs_do_truncate_blocks(inode, from, lock);
1183
1184         /* truncate compressed cluster */
1185         err = f2fs_prepare_compress_overwrite(inode, &pagep,
1186                                                 start_idx, &fsdata);
1187
1188         /* should not be a normal cluster */
1189         f2fs_bug_on(F2FS_I_SB(inode), err == 0);
1190
1191         if (err <= 0)
1192                 return err;
1193
1194         if (err > 0) {
1195                 struct page **rpages = fsdata;
1196                 int cluster_size = F2FS_I(inode)->i_cluster_size;
1197                 int i;
1198
1199                 for (i = cluster_size - 1; i >= 0; i--) {
1200                         loff_t start = rpages[i]->index << PAGE_SHIFT;
1201
1202                         if (from <= start) {
1203                                 zero_user_segment(rpages[i], 0, PAGE_SIZE);
1204                         } else {
1205                                 zero_user_segment(rpages[i], from - start,
1206                                                                 PAGE_SIZE);
1207                                 break;
1208                         }
1209                 }
1210
1211                 f2fs_compress_write_end(inode, fsdata, start_idx, true);
1212         }
1213         return 0;
1214 }
1215
1216 static int f2fs_write_compressed_pages(struct compress_ctx *cc,
1217                                         int *submitted,
1218                                         struct writeback_control *wbc,
1219                                         enum iostat_type io_type)
1220 {
1221         struct inode *inode = cc->inode;
1222         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1223         struct f2fs_inode_info *fi = F2FS_I(inode);
1224         struct f2fs_io_info fio = {
1225                 .sbi = sbi,
1226                 .ino = cc->inode->i_ino,
1227                 .type = DATA,
1228                 .op = REQ_OP_WRITE,
1229                 .op_flags = wbc_to_write_flags(wbc),
1230                 .old_blkaddr = NEW_ADDR,
1231                 .page = NULL,
1232                 .encrypted_page = NULL,
1233                 .compressed_page = NULL,
1234                 .submitted = false,
1235                 .io_type = io_type,
1236                 .io_wbc = wbc,
1237                 .encrypted = fscrypt_inode_uses_fs_layer_crypto(cc->inode),
1238         };
1239         struct dnode_of_data dn;
1240         struct node_info ni;
1241         struct compress_io_ctx *cic;
1242         pgoff_t start_idx = start_idx_of_cluster(cc);
1243         unsigned int last_index = cc->cluster_size - 1;
1244         loff_t psize;
1245         int i, err;
1246
1247         /* we should bypass data pages to proceed the kworkder jobs */
1248         if (unlikely(f2fs_cp_error(sbi))) {
1249                 mapping_set_error(cc->rpages[0]->mapping, -EIO);
1250                 goto out_free;
1251         }
1252
1253         if (IS_NOQUOTA(inode)) {
1254                 /*
1255                  * We need to wait for node_write to avoid block allocation during
1256                  * checkpoint. This can only happen to quota writes which can cause
1257                  * the below discard race condition.
1258                  */
1259                 down_read(&sbi->node_write);
1260         } else if (!f2fs_trylock_op(sbi)) {
1261                 goto out_free;
1262         }
1263
1264         set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1265
1266         err = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
1267         if (err)
1268                 goto out_unlock_op;
1269
1270         for (i = 0; i < cc->cluster_size; i++) {
1271                 if (data_blkaddr(dn.inode, dn.node_page,
1272                                         dn.ofs_in_node + i) == NULL_ADDR)
1273                         goto out_put_dnode;
1274         }
1275
1276         psize = (loff_t)(cc->rpages[last_index]->index + 1) << PAGE_SHIFT;
1277
1278         err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1279         if (err)
1280                 goto out_put_dnode;
1281
1282         fio.version = ni.version;
1283
1284         cic = f2fs_kmem_cache_alloc(cic_entry_slab, GFP_F2FS_ZERO, false, sbi);
1285         if (!cic)
1286                 goto out_put_dnode;
1287
1288         cic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1289         cic->inode = inode;
1290         atomic_set(&cic->pending_pages, cc->nr_cpages);
1291         cic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1292         if (!cic->rpages)
1293                 goto out_put_cic;
1294
1295         cic->nr_rpages = cc->cluster_size;
1296
1297         for (i = 0; i < cc->nr_cpages; i++) {
1298                 f2fs_set_compressed_page(cc->cpages[i], inode,
1299                                         cc->rpages[i + 1]->index, cic);
1300                 fio.compressed_page = cc->cpages[i];
1301
1302                 fio.old_blkaddr = data_blkaddr(dn.inode, dn.node_page,
1303                                                 dn.ofs_in_node + i + 1);
1304
1305                 /* wait for GCed page writeback via META_MAPPING */
1306                 f2fs_wait_on_block_writeback(inode, fio.old_blkaddr);
1307
1308                 if (fio.encrypted) {
1309                         fio.page = cc->rpages[i + 1];
1310                         err = f2fs_encrypt_one_page(&fio);
1311                         if (err)
1312                                 goto out_destroy_crypt;
1313                         cc->cpages[i] = fio.encrypted_page;
1314                 }
1315         }
1316
1317         set_cluster_writeback(cc);
1318
1319         for (i = 0; i < cc->cluster_size; i++)
1320                 cic->rpages[i] = cc->rpages[i];
1321
1322         for (i = 0; i < cc->cluster_size; i++, dn.ofs_in_node++) {
1323                 block_t blkaddr;
1324
1325                 blkaddr = f2fs_data_blkaddr(&dn);
1326                 fio.page = cc->rpages[i];
1327                 fio.old_blkaddr = blkaddr;
1328
1329                 /* cluster header */
1330                 if (i == 0) {
1331                         if (blkaddr == COMPRESS_ADDR)
1332                                 fio.compr_blocks++;
1333                         if (__is_valid_data_blkaddr(blkaddr))
1334                                 f2fs_invalidate_blocks(sbi, blkaddr);
1335                         f2fs_update_data_blkaddr(&dn, COMPRESS_ADDR);
1336                         goto unlock_continue;
1337                 }
1338
1339                 if (fio.compr_blocks && __is_valid_data_blkaddr(blkaddr))
1340                         fio.compr_blocks++;
1341
1342                 if (i > cc->nr_cpages) {
1343                         if (__is_valid_data_blkaddr(blkaddr)) {
1344                                 f2fs_invalidate_blocks(sbi, blkaddr);
1345                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
1346                         }
1347                         goto unlock_continue;
1348                 }
1349
1350                 f2fs_bug_on(fio.sbi, blkaddr == NULL_ADDR);
1351
1352                 if (fio.encrypted)
1353                         fio.encrypted_page = cc->cpages[i - 1];
1354                 else
1355                         fio.compressed_page = cc->cpages[i - 1];
1356
1357                 cc->cpages[i - 1] = NULL;
1358                 f2fs_outplace_write_data(&dn, &fio);
1359                 (*submitted)++;
1360 unlock_continue:
1361                 inode_dec_dirty_pages(cc->inode);
1362                 unlock_page(fio.page);
1363         }
1364
1365         if (fio.compr_blocks)
1366                 f2fs_i_compr_blocks_update(inode, fio.compr_blocks - 1, false);
1367         f2fs_i_compr_blocks_update(inode, cc->nr_cpages, true);
1368         add_compr_block_stat(inode, cc->nr_cpages);
1369
1370         set_inode_flag(cc->inode, FI_APPEND_WRITE);
1371         if (cc->cluster_idx == 0)
1372                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1373
1374         f2fs_put_dnode(&dn);
1375         if (IS_NOQUOTA(inode))
1376                 up_read(&sbi->node_write);
1377         else
1378                 f2fs_unlock_op(sbi);
1379
1380         spin_lock(&fi->i_size_lock);
1381         if (fi->last_disk_size < psize)
1382                 fi->last_disk_size = psize;
1383         spin_unlock(&fi->i_size_lock);
1384
1385         f2fs_put_rpages(cc);
1386         page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1387         cc->cpages = NULL;
1388         f2fs_destroy_compress_ctx(cc, false);
1389         return 0;
1390
1391 out_destroy_crypt:
1392         page_array_free(cc->inode, cic->rpages, cc->cluster_size);
1393
1394         for (--i; i >= 0; i--)
1395                 fscrypt_finalize_bounce_page(&cc->cpages[i]);
1396 out_put_cic:
1397         kmem_cache_free(cic_entry_slab, cic);
1398 out_put_dnode:
1399         f2fs_put_dnode(&dn);
1400 out_unlock_op:
1401         if (IS_NOQUOTA(inode))
1402                 up_read(&sbi->node_write);
1403         else
1404                 f2fs_unlock_op(sbi);
1405 out_free:
1406         for (i = 0; i < cc->nr_cpages; i++) {
1407                 if (!cc->cpages[i])
1408                         continue;
1409                 f2fs_compress_free_page(cc->cpages[i]);
1410                 cc->cpages[i] = NULL;
1411         }
1412         page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1413         cc->cpages = NULL;
1414         return -EAGAIN;
1415 }
1416
1417 void f2fs_compress_write_end_io(struct bio *bio, struct page *page)
1418 {
1419         struct f2fs_sb_info *sbi = bio->bi_private;
1420         struct compress_io_ctx *cic =
1421                         (struct compress_io_ctx *)page_private(page);
1422         int i;
1423
1424         if (unlikely(bio->bi_status))
1425                 mapping_set_error(cic->inode->i_mapping, -EIO);
1426
1427         f2fs_compress_free_page(page);
1428
1429         dec_page_count(sbi, F2FS_WB_DATA);
1430
1431         if (atomic_dec_return(&cic->pending_pages))
1432                 return;
1433
1434         for (i = 0; i < cic->nr_rpages; i++) {
1435                 WARN_ON(!cic->rpages[i]);
1436                 clear_page_private_gcing(cic->rpages[i]);
1437                 end_page_writeback(cic->rpages[i]);
1438         }
1439
1440         page_array_free(cic->inode, cic->rpages, cic->nr_rpages);
1441         kmem_cache_free(cic_entry_slab, cic);
1442 }
1443
1444 static int f2fs_write_raw_pages(struct compress_ctx *cc,
1445                                         int *submitted,
1446                                         struct writeback_control *wbc,
1447                                         enum iostat_type io_type)
1448 {
1449         struct address_space *mapping = cc->inode->i_mapping;
1450         int _submitted, compr_blocks, ret, i;
1451
1452         compr_blocks = f2fs_compressed_blocks(cc);
1453
1454         for (i = 0; i < cc->cluster_size; i++) {
1455                 if (!cc->rpages[i])
1456                         continue;
1457
1458                 redirty_page_for_writepage(wbc, cc->rpages[i]);
1459                 unlock_page(cc->rpages[i]);
1460         }
1461
1462         if (compr_blocks < 0)
1463                 return compr_blocks;
1464
1465         for (i = 0; i < cc->cluster_size; i++) {
1466                 if (!cc->rpages[i])
1467                         continue;
1468 retry_write:
1469                 lock_page(cc->rpages[i]);
1470
1471                 if (cc->rpages[i]->mapping != mapping) {
1472 continue_unlock:
1473                         unlock_page(cc->rpages[i]);
1474                         continue;
1475                 }
1476
1477                 if (!PageDirty(cc->rpages[i]))
1478                         goto continue_unlock;
1479
1480                 if (PageWriteback(cc->rpages[i])) {
1481                         if (wbc->sync_mode == WB_SYNC_NONE)
1482                                 goto continue_unlock;
1483                         f2fs_wait_on_page_writeback(cc->rpages[i], DATA, true, true);
1484                 }
1485
1486                 if (!clear_page_dirty_for_io(cc->rpages[i]))
1487                         goto continue_unlock;
1488
1489                 ret = f2fs_write_single_data_page(cc->rpages[i], &_submitted,
1490                                                 NULL, NULL, wbc, io_type,
1491                                                 compr_blocks, false);
1492                 if (ret) {
1493                         if (ret == AOP_WRITEPAGE_ACTIVATE) {
1494                                 unlock_page(cc->rpages[i]);
1495                                 ret = 0;
1496                         } else if (ret == -EAGAIN) {
1497                                 /*
1498                                  * for quota file, just redirty left pages to
1499                                  * avoid deadlock caused by cluster update race
1500                                  * from foreground operation.
1501                                  */
1502                                 if (IS_NOQUOTA(cc->inode))
1503                                         return 0;
1504                                 ret = 0;
1505                                 cond_resched();
1506                                 congestion_wait(BLK_RW_ASYNC,
1507                                                 DEFAULT_IO_TIMEOUT);
1508                                 goto retry_write;
1509                         }
1510                         return ret;
1511                 }
1512
1513                 *submitted += _submitted;
1514         }
1515
1516         f2fs_balance_fs(F2FS_M_SB(mapping), true);
1517
1518         return 0;
1519 }
1520
1521 int f2fs_write_multi_pages(struct compress_ctx *cc,
1522                                         int *submitted,
1523                                         struct writeback_control *wbc,
1524                                         enum iostat_type io_type)
1525 {
1526         int err;
1527
1528         *submitted = 0;
1529         if (cluster_may_compress(cc)) {
1530                 err = f2fs_compress_pages(cc);
1531                 if (err == -EAGAIN) {
1532                         add_compr_block_stat(cc->inode, cc->cluster_size);
1533                         goto write;
1534                 } else if (err) {
1535                         f2fs_put_rpages_wbc(cc, wbc, true, 1);
1536                         goto destroy_out;
1537                 }
1538
1539                 err = f2fs_write_compressed_pages(cc, submitted,
1540                                                         wbc, io_type);
1541                 if (!err)
1542                         return 0;
1543                 f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN);
1544         }
1545 write:
1546         f2fs_bug_on(F2FS_I_SB(cc->inode), *submitted);
1547
1548         err = f2fs_write_raw_pages(cc, submitted, wbc, io_type);
1549         f2fs_put_rpages_wbc(cc, wbc, false, 0);
1550 destroy_out:
1551         f2fs_destroy_compress_ctx(cc, false);
1552         return err;
1553 }
1554
1555 static void f2fs_free_dic(struct decompress_io_ctx *dic);
1556
1557 struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
1558 {
1559         struct decompress_io_ctx *dic;
1560         pgoff_t start_idx = start_idx_of_cluster(cc);
1561         int i;
1562
1563         dic = f2fs_kmem_cache_alloc(dic_entry_slab, GFP_F2FS_ZERO,
1564                                         false, F2FS_I_SB(cc->inode));
1565         if (!dic)
1566                 return ERR_PTR(-ENOMEM);
1567
1568         dic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1569         if (!dic->rpages) {
1570                 kmem_cache_free(dic_entry_slab, dic);
1571                 return ERR_PTR(-ENOMEM);
1572         }
1573
1574         dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1575         dic->inode = cc->inode;
1576         atomic_set(&dic->remaining_pages, cc->nr_cpages);
1577         dic->cluster_idx = cc->cluster_idx;
1578         dic->cluster_size = cc->cluster_size;
1579         dic->log_cluster_size = cc->log_cluster_size;
1580         dic->nr_cpages = cc->nr_cpages;
1581         refcount_set(&dic->refcnt, 1);
1582         dic->failed = false;
1583         dic->need_verity = f2fs_need_verity(cc->inode, start_idx);
1584
1585         for (i = 0; i < dic->cluster_size; i++)
1586                 dic->rpages[i] = cc->rpages[i];
1587         dic->nr_rpages = cc->cluster_size;
1588
1589         dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages);
1590         if (!dic->cpages)
1591                 goto out_free;
1592
1593         for (i = 0; i < dic->nr_cpages; i++) {
1594                 struct page *page;
1595
1596                 page = f2fs_compress_alloc_page();
1597                 if (!page)
1598                         goto out_free;
1599
1600                 f2fs_set_compressed_page(page, cc->inode,
1601                                         start_idx + i + 1, dic);
1602                 dic->cpages[i] = page;
1603         }
1604
1605         return dic;
1606
1607 out_free:
1608         f2fs_free_dic(dic);
1609         return ERR_PTR(-ENOMEM);
1610 }
1611
1612 static void f2fs_free_dic(struct decompress_io_ctx *dic)
1613 {
1614         int i;
1615
1616         if (dic->tpages) {
1617                 for (i = 0; i < dic->cluster_size; i++) {
1618                         if (dic->rpages[i])
1619                                 continue;
1620                         if (!dic->tpages[i])
1621                                 continue;
1622                         f2fs_compress_free_page(dic->tpages[i]);
1623                 }
1624                 page_array_free(dic->inode, dic->tpages, dic->cluster_size);
1625         }
1626
1627         if (dic->cpages) {
1628                 for (i = 0; i < dic->nr_cpages; i++) {
1629                         if (!dic->cpages[i])
1630                                 continue;
1631                         f2fs_compress_free_page(dic->cpages[i]);
1632                 }
1633                 page_array_free(dic->inode, dic->cpages, dic->nr_cpages);
1634         }
1635
1636         page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
1637         kmem_cache_free(dic_entry_slab, dic);
1638 }
1639
1640 static void f2fs_put_dic(struct decompress_io_ctx *dic)
1641 {
1642         if (refcount_dec_and_test(&dic->refcnt))
1643                 f2fs_free_dic(dic);
1644 }
1645
1646 /*
1647  * Update and unlock the cluster's pagecache pages, and release the reference to
1648  * the decompress_io_ctx that was being held for I/O completion.
1649  */
1650 static void __f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1651 {
1652         int i;
1653
1654         for (i = 0; i < dic->cluster_size; i++) {
1655                 struct page *rpage = dic->rpages[i];
1656
1657                 if (!rpage)
1658                         continue;
1659
1660                 /* PG_error was set if verity failed. */
1661                 if (failed || PageError(rpage)) {
1662                         ClearPageUptodate(rpage);
1663                         /* will re-read again later */
1664                         ClearPageError(rpage);
1665                 } else {
1666                         SetPageUptodate(rpage);
1667                 }
1668                 unlock_page(rpage);
1669         }
1670
1671         f2fs_put_dic(dic);
1672 }
1673
1674 static void f2fs_verify_cluster(struct work_struct *work)
1675 {
1676         struct decompress_io_ctx *dic =
1677                 container_of(work, struct decompress_io_ctx, verity_work);
1678         int i;
1679
1680         /* Verify the cluster's decompressed pages with fs-verity. */
1681         for (i = 0; i < dic->cluster_size; i++) {
1682                 struct page *rpage = dic->rpages[i];
1683
1684                 if (rpage && !fsverity_verify_page(rpage))
1685                         SetPageError(rpage);
1686         }
1687
1688         __f2fs_decompress_end_io(dic, false);
1689 }
1690
1691 /*
1692  * This is called when a compressed cluster has been decompressed
1693  * (or failed to be read and/or decompressed).
1694  */
1695 void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1696 {
1697         if (!failed && dic->need_verity) {
1698                 /*
1699                  * Note that to avoid deadlocks, the verity work can't be done
1700                  * on the decompression workqueue.  This is because verifying
1701                  * the data pages can involve reading metadata pages from the
1702                  * file, and these metadata pages may be compressed.
1703                  */
1704                 INIT_WORK(&dic->verity_work, f2fs_verify_cluster);
1705                 fsverity_enqueue_verify_work(&dic->verity_work);
1706         } else {
1707                 __f2fs_decompress_end_io(dic, failed);
1708         }
1709 }
1710
1711 /*
1712  * Put a reference to a compressed page's decompress_io_ctx.
1713  *
1714  * This is called when the page is no longer needed and can be freed.
1715  */
1716 void f2fs_put_page_dic(struct page *page)
1717 {
1718         struct decompress_io_ctx *dic =
1719                         (struct decompress_io_ctx *)page_private(page);
1720
1721         f2fs_put_dic(dic);
1722 }
1723
1724 /*
1725  * check whether cluster blocks are contiguous, and add extent cache entry
1726  * only if cluster blocks are logically and physically contiguous.
1727  */
1728 unsigned int f2fs_cluster_blocks_are_contiguous(struct dnode_of_data *dn)
1729 {
1730         bool compressed = f2fs_data_blkaddr(dn) == COMPRESS_ADDR;
1731         int i = compressed ? 1 : 0;
1732         block_t first_blkaddr = data_blkaddr(dn->inode, dn->node_page,
1733                                                 dn->ofs_in_node + i);
1734
1735         for (i += 1; i < F2FS_I(dn->inode)->i_cluster_size; i++) {
1736                 block_t blkaddr = data_blkaddr(dn->inode, dn->node_page,
1737                                                 dn->ofs_in_node + i);
1738
1739                 if (!__is_valid_data_blkaddr(blkaddr))
1740                         break;
1741                 if (first_blkaddr + i - (compressed ? 1 : 0) != blkaddr)
1742                         return 0;
1743         }
1744
1745         return compressed ? i - 1 : i;
1746 }
1747
1748 const struct address_space_operations f2fs_compress_aops = {
1749         .releasepage = f2fs_release_page,
1750         .invalidatepage = f2fs_invalidate_page,
1751 };
1752
1753 struct address_space *COMPRESS_MAPPING(struct f2fs_sb_info *sbi)
1754 {
1755         return sbi->compress_inode->i_mapping;
1756 }
1757
1758 void f2fs_invalidate_compress_page(struct f2fs_sb_info *sbi, block_t blkaddr)
1759 {
1760         if (!sbi->compress_inode)
1761                 return;
1762         invalidate_mapping_pages(COMPRESS_MAPPING(sbi), blkaddr, blkaddr);
1763 }
1764
1765 void f2fs_cache_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
1766                                                 nid_t ino, block_t blkaddr)
1767 {
1768         struct page *cpage;
1769         int ret;
1770
1771         if (!test_opt(sbi, COMPRESS_CACHE))
1772                 return;
1773
1774         if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
1775                 return;
1776
1777         if (!f2fs_available_free_memory(sbi, COMPRESS_PAGE))
1778                 return;
1779
1780         cpage = find_get_page(COMPRESS_MAPPING(sbi), blkaddr);
1781         if (cpage) {
1782                 f2fs_put_page(cpage, 0);
1783                 return;
1784         }
1785
1786         cpage = alloc_page(__GFP_NOWARN | __GFP_IO);
1787         if (!cpage)
1788                 return;
1789
1790         ret = add_to_page_cache_lru(cpage, COMPRESS_MAPPING(sbi),
1791                                                 blkaddr, GFP_NOFS);
1792         if (ret) {
1793                 f2fs_put_page(cpage, 0);
1794                 return;
1795         }
1796
1797         set_page_private_data(cpage, ino);
1798
1799         if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
1800                 goto out;
1801
1802         memcpy(page_address(cpage), page_address(page), PAGE_SIZE);
1803         SetPageUptodate(cpage);
1804 out:
1805         f2fs_put_page(cpage, 1);
1806 }
1807
1808 bool f2fs_load_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
1809                                                                 block_t blkaddr)
1810 {
1811         struct page *cpage;
1812         bool hitted = false;
1813
1814         if (!test_opt(sbi, COMPRESS_CACHE))
1815                 return false;
1816
1817         cpage = f2fs_pagecache_get_page(COMPRESS_MAPPING(sbi),
1818                                 blkaddr, FGP_LOCK | FGP_NOWAIT, GFP_NOFS);
1819         if (cpage) {
1820                 if (PageUptodate(cpage)) {
1821                         atomic_inc(&sbi->compress_page_hit);
1822                         memcpy(page_address(page),
1823                                 page_address(cpage), PAGE_SIZE);
1824                         hitted = true;
1825                 }
1826                 f2fs_put_page(cpage, 1);
1827         }
1828
1829         return hitted;
1830 }
1831
1832 void f2fs_invalidate_compress_pages(struct f2fs_sb_info *sbi, nid_t ino)
1833 {
1834         struct address_space *mapping = sbi->compress_inode->i_mapping;
1835         struct pagevec pvec;
1836         pgoff_t index = 0;
1837         pgoff_t end = MAX_BLKADDR(sbi);
1838
1839         if (!mapping->nrpages)
1840                 return;
1841
1842         pagevec_init(&pvec);
1843
1844         do {
1845                 unsigned int nr_pages;
1846                 int i;
1847
1848                 nr_pages = pagevec_lookup_range(&pvec, mapping,
1849                                                 &index, end - 1);
1850                 if (!nr_pages)
1851                         break;
1852
1853                 for (i = 0; i < nr_pages; i++) {
1854                         struct page *page = pvec.pages[i];
1855
1856                         if (page->index > end)
1857                                 break;
1858
1859                         lock_page(page);
1860                         if (page->mapping != mapping) {
1861                                 unlock_page(page);
1862                                 continue;
1863                         }
1864
1865                         if (ino != get_page_private_data(page)) {
1866                                 unlock_page(page);
1867                                 continue;
1868                         }
1869
1870                         generic_error_remove_page(mapping, page);
1871                         unlock_page(page);
1872                 }
1873                 pagevec_release(&pvec);
1874                 cond_resched();
1875         } while (index < end);
1876 }
1877
1878 int f2fs_init_compress_inode(struct f2fs_sb_info *sbi)
1879 {
1880         struct inode *inode;
1881
1882         if (!test_opt(sbi, COMPRESS_CACHE))
1883                 return 0;
1884
1885         inode = f2fs_iget(sbi->sb, F2FS_COMPRESS_INO(sbi));
1886         if (IS_ERR(inode))
1887                 return PTR_ERR(inode);
1888         sbi->compress_inode = inode;
1889
1890         sbi->compress_percent = COMPRESS_PERCENT;
1891         sbi->compress_watermark = COMPRESS_WATERMARK;
1892
1893         atomic_set(&sbi->compress_page_hit, 0);
1894
1895         return 0;
1896 }
1897
1898 void f2fs_destroy_compress_inode(struct f2fs_sb_info *sbi)
1899 {
1900         if (!sbi->compress_inode)
1901                 return;
1902         iput(sbi->compress_inode);
1903         sbi->compress_inode = NULL;
1904 }
1905
1906 int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi)
1907 {
1908         dev_t dev = sbi->sb->s_bdev->bd_dev;
1909         char slab_name[32];
1910
1911         sprintf(slab_name, "f2fs_page_array_entry-%u:%u", MAJOR(dev), MINOR(dev));
1912
1913         sbi->page_array_slab_size = sizeof(struct page *) <<
1914                                         F2FS_OPTION(sbi).compress_log_size;
1915
1916         sbi->page_array_slab = f2fs_kmem_cache_create(slab_name,
1917                                         sbi->page_array_slab_size);
1918         if (!sbi->page_array_slab)
1919                 return -ENOMEM;
1920         return 0;
1921 }
1922
1923 void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi)
1924 {
1925         kmem_cache_destroy(sbi->page_array_slab);
1926 }
1927
1928 static int __init f2fs_init_cic_cache(void)
1929 {
1930         cic_entry_slab = f2fs_kmem_cache_create("f2fs_cic_entry",
1931                                         sizeof(struct compress_io_ctx));
1932         if (!cic_entry_slab)
1933                 return -ENOMEM;
1934         return 0;
1935 }
1936
1937 static void f2fs_destroy_cic_cache(void)
1938 {
1939         kmem_cache_destroy(cic_entry_slab);
1940 }
1941
1942 static int __init f2fs_init_dic_cache(void)
1943 {
1944         dic_entry_slab = f2fs_kmem_cache_create("f2fs_dic_entry",
1945                                         sizeof(struct decompress_io_ctx));
1946         if (!dic_entry_slab)
1947                 return -ENOMEM;
1948         return 0;
1949 }
1950
1951 static void f2fs_destroy_dic_cache(void)
1952 {
1953         kmem_cache_destroy(dic_entry_slab);
1954 }
1955
1956 int __init f2fs_init_compress_cache(void)
1957 {
1958         int err;
1959
1960         err = f2fs_init_cic_cache();
1961         if (err)
1962                 goto out;
1963         err = f2fs_init_dic_cache();
1964         if (err)
1965                 goto free_cic;
1966         return 0;
1967 free_cic:
1968         f2fs_destroy_cic_cache();
1969 out:
1970         return -ENOMEM;
1971 }
1972
1973 void f2fs_destroy_compress_cache(void)
1974 {
1975         f2fs_destroy_dic_cache();
1976         f2fs_destroy_cic_cache();
1977 }