Loading...
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
22static struct kmem_cache *cic_entry_slab;
23static struct kmem_cache *dic_entry_slab;
24
25static void *page_array_alloc(struct inode *inode, int nr)
26{
27 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
28 unsigned int size = sizeof(struct page *) * nr;
29
30 if (likely(size <= sbi->page_array_slab_size))
31 return kmem_cache_zalloc(sbi->page_array_slab, GFP_NOFS);
32 return f2fs_kzalloc(sbi, size, GFP_NOFS);
33}
34
35static void page_array_free(struct inode *inode, void *pages, int nr)
36{
37 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
38 unsigned int size = sizeof(struct page *) * nr;
39
40 if (!pages)
41 return;
42
43 if (likely(size <= sbi->page_array_slab_size))
44 kmem_cache_free(sbi->page_array_slab, pages);
45 else
46 kfree(pages);
47}
48
49struct f2fs_compress_ops {
50 int (*init_compress_ctx)(struct compress_ctx *cc);
51 void (*destroy_compress_ctx)(struct compress_ctx *cc);
52 int (*compress_pages)(struct compress_ctx *cc);
53 int (*init_decompress_ctx)(struct decompress_io_ctx *dic);
54 void (*destroy_decompress_ctx)(struct decompress_io_ctx *dic);
55 int (*decompress_pages)(struct decompress_io_ctx *dic);
56};
57
58static unsigned int offset_in_cluster(struct compress_ctx *cc, pgoff_t index)
59{
60 return index & (cc->cluster_size - 1);
61}
62
63static pgoff_t cluster_idx(struct compress_ctx *cc, pgoff_t index)
64{
65 return index >> cc->log_cluster_size;
66}
67
68static pgoff_t start_idx_of_cluster(struct compress_ctx *cc)
69{
70 return cc->cluster_idx << cc->log_cluster_size;
71}
72
73bool f2fs_is_compressed_page(struct page *page)
74{
75 if (!PagePrivate(page))
76 return false;
77 if (!page_private(page))
78 return false;
79 if (page_private_nonpointer(page))
80 return false;
81
82 f2fs_bug_on(F2FS_M_SB(page->mapping),
83 *((u32 *)page_private(page)) != F2FS_COMPRESSED_PAGE_MAGIC);
84 return true;
85}
86
87static void f2fs_set_compressed_page(struct page *page,
88 struct inode *inode, pgoff_t index, void *data)
89{
90 attach_page_private(page, (void *)data);
91
92 /* i_crypto_info and iv index */
93 page->index = index;
94 page->mapping = inode->i_mapping;
95}
96
97static void f2fs_drop_rpages(struct compress_ctx *cc, int len, bool unlock)
98{
99 int i;
100
101 for (i = 0; i < len; i++) {
102 if (!cc->rpages[i])
103 continue;
104 if (unlock)
105 unlock_page(cc->rpages[i]);
106 else
107 put_page(cc->rpages[i]);
108 }
109}
110
111static void f2fs_put_rpages(struct compress_ctx *cc)
112{
113 f2fs_drop_rpages(cc, cc->cluster_size, false);
114}
115
116static void f2fs_unlock_rpages(struct compress_ctx *cc, int len)
117{
118 f2fs_drop_rpages(cc, len, true);
119}
120
121static void f2fs_put_rpages_wbc(struct compress_ctx *cc,
122 struct writeback_control *wbc, bool redirty, int unlock)
123{
124 unsigned int i;
125
126 for (i = 0; i < cc->cluster_size; i++) {
127 if (!cc->rpages[i])
128 continue;
129 if (redirty)
130 redirty_page_for_writepage(wbc, cc->rpages[i]);
131 f2fs_put_page(cc->rpages[i], unlock);
132 }
133}
134
135struct page *f2fs_compress_control_page(struct page *page)
136{
137 return ((struct compress_io_ctx *)page_private(page))->rpages[0];
138}
139
140int f2fs_init_compress_ctx(struct compress_ctx *cc)
141{
142 if (cc->rpages)
143 return 0;
144
145 cc->rpages = page_array_alloc(cc->inode, cc->cluster_size);
146 return cc->rpages ? 0 : -ENOMEM;
147}
148
149void f2fs_destroy_compress_ctx(struct compress_ctx *cc, bool reuse)
150{
151 page_array_free(cc->inode, cc->rpages, cc->cluster_size);
152 cc->rpages = NULL;
153 cc->nr_rpages = 0;
154 cc->nr_cpages = 0;
155 if (!reuse)
156 cc->cluster_idx = NULL_CLUSTER;
157}
158
159void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct page *page)
160{
161 unsigned int cluster_ofs;
162
163 if (!f2fs_cluster_can_merge_page(cc, page->index))
164 f2fs_bug_on(F2FS_I_SB(cc->inode), 1);
165
166 cluster_ofs = offset_in_cluster(cc, page->index);
167 cc->rpages[cluster_ofs] = page;
168 cc->nr_rpages++;
169 cc->cluster_idx = cluster_idx(cc, page->index);
170}
171
172#ifdef CONFIG_F2FS_FS_LZO
173static int lzo_init_compress_ctx(struct compress_ctx *cc)
174{
175 cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
176 LZO1X_MEM_COMPRESS, GFP_NOFS);
177 if (!cc->private)
178 return -ENOMEM;
179
180 cc->clen = lzo1x_worst_compress(PAGE_SIZE << cc->log_cluster_size);
181 return 0;
182}
183
184static void lzo_destroy_compress_ctx(struct compress_ctx *cc)
185{
186 kvfree(cc->private);
187 cc->private = NULL;
188}
189
190static int lzo_compress_pages(struct compress_ctx *cc)
191{
192 int ret;
193
194 ret = lzo1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
195 &cc->clen, cc->private);
196 if (ret != LZO_E_OK) {
197 printk_ratelimited("%sF2FS-fs (%s): lzo compress failed, ret:%d\n",
198 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
199 return -EIO;
200 }
201 return 0;
202}
203
204static int lzo_decompress_pages(struct decompress_io_ctx *dic)
205{
206 int ret;
207
208 ret = lzo1x_decompress_safe(dic->cbuf->cdata, dic->clen,
209 dic->rbuf, &dic->rlen);
210 if (ret != LZO_E_OK) {
211 printk_ratelimited("%sF2FS-fs (%s): lzo decompress failed, ret:%d\n",
212 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
213 return -EIO;
214 }
215
216 if (dic->rlen != PAGE_SIZE << dic->log_cluster_size) {
217 printk_ratelimited("%sF2FS-fs (%s): lzo invalid rlen:%zu, "
218 "expected:%lu\n", KERN_ERR,
219 F2FS_I_SB(dic->inode)->sb->s_id,
220 dic->rlen,
221 PAGE_SIZE << dic->log_cluster_size);
222 return -EIO;
223 }
224 return 0;
225}
226
227static const struct f2fs_compress_ops f2fs_lzo_ops = {
228 .init_compress_ctx = lzo_init_compress_ctx,
229 .destroy_compress_ctx = lzo_destroy_compress_ctx,
230 .compress_pages = lzo_compress_pages,
231 .decompress_pages = lzo_decompress_pages,
232};
233#endif
234
235#ifdef CONFIG_F2FS_FS_LZ4
236static int lz4_init_compress_ctx(struct compress_ctx *cc)
237{
238 unsigned int size = LZ4_MEM_COMPRESS;
239
240#ifdef CONFIG_F2FS_FS_LZ4HC
241 if (F2FS_I(cc->inode)->i_compress_flag >> COMPRESS_LEVEL_OFFSET)
242 size = LZ4HC_MEM_COMPRESS;
243#endif
244
245 cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), size, GFP_NOFS);
246 if (!cc->private)
247 return -ENOMEM;
248
249 /*
250 * we do not change cc->clen to LZ4_compressBound(inputsize) to
251 * adapt worst compress case, because lz4 compressor can handle
252 * output budget properly.
253 */
254 cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
255 return 0;
256}
257
258static void lz4_destroy_compress_ctx(struct compress_ctx *cc)
259{
260 kvfree(cc->private);
261 cc->private = NULL;
262}
263
264#ifdef CONFIG_F2FS_FS_LZ4HC
265static int lz4hc_compress_pages(struct compress_ctx *cc)
266{
267 unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
268 COMPRESS_LEVEL_OFFSET;
269 int len;
270
271 if (level)
272 len = LZ4_compress_HC(cc->rbuf, cc->cbuf->cdata, cc->rlen,
273 cc->clen, level, cc->private);
274 else
275 len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
276 cc->clen, cc->private);
277 if (!len)
278 return -EAGAIN;
279
280 cc->clen = len;
281 return 0;
282}
283#endif
284
285static int lz4_compress_pages(struct compress_ctx *cc)
286{
287 int len;
288
289#ifdef CONFIG_F2FS_FS_LZ4HC
290 return lz4hc_compress_pages(cc);
291#endif
292 len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
293 cc->clen, cc->private);
294 if (!len)
295 return -EAGAIN;
296
297 cc->clen = len;
298 return 0;
299}
300
301static int lz4_decompress_pages(struct decompress_io_ctx *dic)
302{
303 int ret;
304
305 ret = LZ4_decompress_safe(dic->cbuf->cdata, dic->rbuf,
306 dic->clen, dic->rlen);
307 if (ret < 0) {
308 printk_ratelimited("%sF2FS-fs (%s): lz4 decompress failed, ret:%d\n",
309 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
310 return -EIO;
311 }
312
313 if (ret != PAGE_SIZE << dic->log_cluster_size) {
314 printk_ratelimited("%sF2FS-fs (%s): lz4 invalid rlen:%zu, "
315 "expected:%lu\n", KERN_ERR,
316 F2FS_I_SB(dic->inode)->sb->s_id,
317 dic->rlen,
318 PAGE_SIZE << dic->log_cluster_size);
319 return -EIO;
320 }
321 return 0;
322}
323
324static 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
335static 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
371static void zstd_destroy_compress_ctx(struct compress_ctx *cc)
372{
373 kvfree(cc->private);
374 cc->private = NULL;
375 cc->private2 = NULL;
376}
377
378static 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
422static 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
452static 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
459static 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
494static 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
506static 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
520static 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
529static 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
552bool 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
559static mempool_t *compress_page_pool;
560static int num_compress_pages = 512;
561module_param(num_compress_pages, uint, 0444);
562MODULE_PARM_DESC(num_compress_pages,
563 "Number of intermediate compress pages to preallocate");
564
565int 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
574void f2fs_destroy_compress_mempool(void)
575{
576 mempool_destroy(compress_page_pool);
577}
578
579static 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
589static 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
601static 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
615static 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
721out_vunmap_cbuf:
722 vm_unmap_ram(cc->cbuf, cc->nr_cpages);
723out_vunmap_rbuf:
724 vm_unmap_ram(cc->rbuf, cc->cluster_size);
725out_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;
732destroy_compress_ctx:
733 if (cops->destroy_compress_ctx)
734 cops->destroy_compress_ctx(cc);
735out:
736 trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
737 cc->clen, ret);
738 return ret;
739}
740
741void 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
821out_vunmap_cbuf:
822 vm_unmap_ram(dic->cbuf, dic->nr_cpages);
823out_vunmap_rbuf:
824 vm_unmap_ram(dic->rbuf, dic->cluster_size);
825out_destroy_decompress_ctx:
826 if (cops->destroy_decompress_ctx)
827 cops->destroy_decompress_ctx(dic);
828out_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 */
840void 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
859static 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
866bool f2fs_cluster_is_empty(struct compress_ctx *cc)
867{
868 return cc->nr_rpages == 0;
869}
870
871static bool f2fs_cluster_is_full(struct compress_ctx *cc)
872{
873 return cc->cluster_size == cc->nr_rpages;
874}
875
876bool 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
883static 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
901static int __f2fs_cluster_blocks(struct inode *inode,
902 unsigned int cluster_idx, bool compr)
903{
904 struct dnode_of_data dn;
905 unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
906 unsigned int start_idx = cluster_idx <<
907 F2FS_I(inode)->i_log_cluster_size;
908 int ret;
909
910 set_new_dnode(&dn, inode, NULL, NULL, 0);
911 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
912 if (ret) {
913 if (ret == -ENOENT)
914 ret = 0;
915 goto fail;
916 }
917
918 if (dn.data_blkaddr == COMPRESS_ADDR) {
919 int i;
920
921 ret = 1;
922 for (i = 1; i < cluster_size; i++) {
923 block_t blkaddr;
924
925 blkaddr = data_blkaddr(dn.inode,
926 dn.node_page, dn.ofs_in_node + i);
927 if (compr) {
928 if (__is_valid_data_blkaddr(blkaddr))
929 ret++;
930 } else {
931 if (blkaddr != NULL_ADDR)
932 ret++;
933 }
934 }
935
936 f2fs_bug_on(F2FS_I_SB(inode),
937 !compr && ret != cluster_size &&
938 !is_inode_flag_set(inode, FI_COMPRESS_RELEASED));
939 }
940fail:
941 f2fs_put_dnode(&dn);
942 return ret;
943}
944
945/* return # of compressed blocks in compressed cluster */
946static int f2fs_compressed_blocks(struct compress_ctx *cc)
947{
948 return __f2fs_cluster_blocks(cc->inode, cc->cluster_idx, true);
949}
950
951/* return # of valid blocks in compressed cluster */
952int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index)
953{
954 return __f2fs_cluster_blocks(inode,
955 index >> F2FS_I(inode)->i_log_cluster_size,
956 false);
957}
958
959static bool cluster_may_compress(struct compress_ctx *cc)
960{
961 if (!f2fs_need_compress_data(cc->inode))
962 return false;
963 if (f2fs_is_atomic_file(cc->inode))
964 return false;
965 if (!f2fs_cluster_is_full(cc))
966 return false;
967 if (unlikely(f2fs_cp_error(F2FS_I_SB(cc->inode))))
968 return false;
969 return !cluster_has_invalid_data(cc);
970}
971
972static void set_cluster_writeback(struct compress_ctx *cc)
973{
974 int i;
975
976 for (i = 0; i < cc->cluster_size; i++) {
977 if (cc->rpages[i])
978 set_page_writeback(cc->rpages[i]);
979 }
980}
981
982static void set_cluster_dirty(struct compress_ctx *cc)
983{
984 int i;
985
986 for (i = 0; i < cc->cluster_size; i++)
987 if (cc->rpages[i])
988 set_page_dirty(cc->rpages[i]);
989}
990
991static int prepare_compress_overwrite(struct compress_ctx *cc,
992 struct page **pagep, pgoff_t index, void **fsdata)
993{
994 struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
995 struct address_space *mapping = cc->inode->i_mapping;
996 struct page *page;
997 sector_t last_block_in_bio;
998 unsigned fgp_flag = FGP_LOCK | FGP_WRITE | FGP_CREAT;
999 pgoff_t start_idx = start_idx_of_cluster(cc);
1000 int i, ret;
1001
1002retry:
1003 ret = f2fs_is_compressed_cluster(cc->inode, start_idx);
1004 if (ret <= 0)
1005 return ret;
1006
1007 ret = f2fs_init_compress_ctx(cc);
1008 if (ret)
1009 return ret;
1010
1011 /* keep page reference to avoid page reclaim */
1012 for (i = 0; i < cc->cluster_size; i++) {
1013 page = f2fs_pagecache_get_page(mapping, start_idx + i,
1014 fgp_flag, GFP_NOFS);
1015 if (!page) {
1016 ret = -ENOMEM;
1017 goto unlock_pages;
1018 }
1019
1020 if (PageUptodate(page))
1021 f2fs_put_page(page, 1);
1022 else
1023 f2fs_compress_ctx_add_page(cc, page);
1024 }
1025
1026 if (!f2fs_cluster_is_empty(cc)) {
1027 struct bio *bio = NULL;
1028
1029 ret = f2fs_read_multi_pages(cc, &bio, cc->cluster_size,
1030 &last_block_in_bio, false, true);
1031 f2fs_put_rpages(cc);
1032 f2fs_destroy_compress_ctx(cc, true);
1033 if (ret)
1034 goto out;
1035 if (bio)
1036 f2fs_submit_bio(sbi, bio, DATA);
1037
1038 ret = f2fs_init_compress_ctx(cc);
1039 if (ret)
1040 goto out;
1041 }
1042
1043 for (i = 0; i < cc->cluster_size; i++) {
1044 f2fs_bug_on(sbi, cc->rpages[i]);
1045
1046 page = find_lock_page(mapping, start_idx + i);
1047 if (!page) {
1048 /* page can be truncated */
1049 goto release_and_retry;
1050 }
1051
1052 f2fs_wait_on_page_writeback(page, DATA, true, true);
1053 f2fs_compress_ctx_add_page(cc, page);
1054
1055 if (!PageUptodate(page)) {
1056release_and_retry:
1057 f2fs_put_rpages(cc);
1058 f2fs_unlock_rpages(cc, i + 1);
1059 f2fs_destroy_compress_ctx(cc, true);
1060 goto retry;
1061 }
1062 }
1063
1064 if (likely(!ret)) {
1065 *fsdata = cc->rpages;
1066 *pagep = cc->rpages[offset_in_cluster(cc, index)];
1067 return cc->cluster_size;
1068 }
1069
1070unlock_pages:
1071 f2fs_put_rpages(cc);
1072 f2fs_unlock_rpages(cc, i);
1073 f2fs_destroy_compress_ctx(cc, true);
1074out:
1075 return ret;
1076}
1077
1078int f2fs_prepare_compress_overwrite(struct inode *inode,
1079 struct page **pagep, pgoff_t index, void **fsdata)
1080{
1081 struct compress_ctx cc = {
1082 .inode = inode,
1083 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1084 .cluster_size = F2FS_I(inode)->i_cluster_size,
1085 .cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
1086 .rpages = NULL,
1087 .nr_rpages = 0,
1088 };
1089
1090 return prepare_compress_overwrite(&cc, pagep, index, fsdata);
1091}
1092
1093bool f2fs_compress_write_end(struct inode *inode, void *fsdata,
1094 pgoff_t index, unsigned copied)
1095
1096{
1097 struct compress_ctx cc = {
1098 .inode = inode,
1099 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1100 .cluster_size = F2FS_I(inode)->i_cluster_size,
1101 .rpages = fsdata,
1102 };
1103 bool first_index = (index == cc.rpages[0]->index);
1104
1105 if (copied)
1106 set_cluster_dirty(&cc);
1107
1108 f2fs_put_rpages_wbc(&cc, NULL, false, 1);
1109 f2fs_destroy_compress_ctx(&cc, false);
1110
1111 return first_index;
1112}
1113
1114int f2fs_truncate_partial_cluster(struct inode *inode, u64 from, bool lock)
1115{
1116 void *fsdata = NULL;
1117 struct page *pagep;
1118 int log_cluster_size = F2FS_I(inode)->i_log_cluster_size;
1119 pgoff_t start_idx = from >> (PAGE_SHIFT + log_cluster_size) <<
1120 log_cluster_size;
1121 int err;
1122
1123 err = f2fs_is_compressed_cluster(inode, start_idx);
1124 if (err < 0)
1125 return err;
1126
1127 /* truncate normal cluster */
1128 if (!err)
1129 return f2fs_do_truncate_blocks(inode, from, lock);
1130
1131 /* truncate compressed cluster */
1132 err = f2fs_prepare_compress_overwrite(inode, &pagep,
1133 start_idx, &fsdata);
1134
1135 /* should not be a normal cluster */
1136 f2fs_bug_on(F2FS_I_SB(inode), err == 0);
1137
1138 if (err <= 0)
1139 return err;
1140
1141 if (err > 0) {
1142 struct page **rpages = fsdata;
1143 int cluster_size = F2FS_I(inode)->i_cluster_size;
1144 int i;
1145
1146 for (i = cluster_size - 1; i >= 0; i--) {
1147 loff_t start = rpages[i]->index << PAGE_SHIFT;
1148
1149 if (from <= start) {
1150 zero_user_segment(rpages[i], 0, PAGE_SIZE);
1151 } else {
1152 zero_user_segment(rpages[i], from - start,
1153 PAGE_SIZE);
1154 break;
1155 }
1156 }
1157
1158 f2fs_compress_write_end(inode, fsdata, start_idx, true);
1159 }
1160 return 0;
1161}
1162
1163static int f2fs_write_compressed_pages(struct compress_ctx *cc,
1164 int *submitted,
1165 struct writeback_control *wbc,
1166 enum iostat_type io_type)
1167{
1168 struct inode *inode = cc->inode;
1169 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1170 struct f2fs_inode_info *fi = F2FS_I(inode);
1171 struct f2fs_io_info fio = {
1172 .sbi = sbi,
1173 .ino = cc->inode->i_ino,
1174 .type = DATA,
1175 .op = REQ_OP_WRITE,
1176 .op_flags = wbc_to_write_flags(wbc),
1177 .old_blkaddr = NEW_ADDR,
1178 .page = NULL,
1179 .encrypted_page = NULL,
1180 .compressed_page = NULL,
1181 .submitted = false,
1182 .io_type = io_type,
1183 .io_wbc = wbc,
1184 .encrypted = fscrypt_inode_uses_fs_layer_crypto(cc->inode),
1185 };
1186 struct dnode_of_data dn;
1187 struct node_info ni;
1188 struct compress_io_ctx *cic;
1189 pgoff_t start_idx = start_idx_of_cluster(cc);
1190 unsigned int last_index = cc->cluster_size - 1;
1191 loff_t psize;
1192 int i, err;
1193
1194 /* we should bypass data pages to proceed the kworkder jobs */
1195 if (unlikely(f2fs_cp_error(sbi))) {
1196 mapping_set_error(cc->rpages[0]->mapping, -EIO);
1197 goto out_free;
1198 }
1199
1200 if (IS_NOQUOTA(inode)) {
1201 /*
1202 * We need to wait for node_write to avoid block allocation during
1203 * checkpoint. This can only happen to quota writes which can cause
1204 * the below discard race condition.
1205 */
1206 down_read(&sbi->node_write);
1207 } else if (!f2fs_trylock_op(sbi)) {
1208 goto out_free;
1209 }
1210
1211 set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1212
1213 err = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
1214 if (err)
1215 goto out_unlock_op;
1216
1217 for (i = 0; i < cc->cluster_size; i++) {
1218 if (data_blkaddr(dn.inode, dn.node_page,
1219 dn.ofs_in_node + i) == NULL_ADDR)
1220 goto out_put_dnode;
1221 }
1222
1223 psize = (loff_t)(cc->rpages[last_index]->index + 1) << PAGE_SHIFT;
1224
1225 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1226 if (err)
1227 goto out_put_dnode;
1228
1229 fio.version = ni.version;
1230
1231 cic = kmem_cache_zalloc(cic_entry_slab, GFP_NOFS);
1232 if (!cic)
1233 goto out_put_dnode;
1234
1235 cic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1236 cic->inode = inode;
1237 atomic_set(&cic->pending_pages, cc->nr_cpages);
1238 cic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1239 if (!cic->rpages)
1240 goto out_put_cic;
1241
1242 cic->nr_rpages = cc->cluster_size;
1243
1244 for (i = 0; i < cc->nr_cpages; i++) {
1245 f2fs_set_compressed_page(cc->cpages[i], inode,
1246 cc->rpages[i + 1]->index, cic);
1247 fio.compressed_page = cc->cpages[i];
1248
1249 fio.old_blkaddr = data_blkaddr(dn.inode, dn.node_page,
1250 dn.ofs_in_node + i + 1);
1251
1252 /* wait for GCed page writeback via META_MAPPING */
1253 f2fs_wait_on_block_writeback(inode, fio.old_blkaddr);
1254
1255 if (fio.encrypted) {
1256 fio.page = cc->rpages[i + 1];
1257 err = f2fs_encrypt_one_page(&fio);
1258 if (err)
1259 goto out_destroy_crypt;
1260 cc->cpages[i] = fio.encrypted_page;
1261 }
1262 }
1263
1264 set_cluster_writeback(cc);
1265
1266 for (i = 0; i < cc->cluster_size; i++)
1267 cic->rpages[i] = cc->rpages[i];
1268
1269 for (i = 0; i < cc->cluster_size; i++, dn.ofs_in_node++) {
1270 block_t blkaddr;
1271
1272 blkaddr = f2fs_data_blkaddr(&dn);
1273 fio.page = cc->rpages[i];
1274 fio.old_blkaddr = blkaddr;
1275
1276 /* cluster header */
1277 if (i == 0) {
1278 if (blkaddr == COMPRESS_ADDR)
1279 fio.compr_blocks++;
1280 if (__is_valid_data_blkaddr(blkaddr))
1281 f2fs_invalidate_blocks(sbi, blkaddr);
1282 f2fs_update_data_blkaddr(&dn, COMPRESS_ADDR);
1283 goto unlock_continue;
1284 }
1285
1286 if (fio.compr_blocks && __is_valid_data_blkaddr(blkaddr))
1287 fio.compr_blocks++;
1288
1289 if (i > cc->nr_cpages) {
1290 if (__is_valid_data_blkaddr(blkaddr)) {
1291 f2fs_invalidate_blocks(sbi, blkaddr);
1292 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
1293 }
1294 goto unlock_continue;
1295 }
1296
1297 f2fs_bug_on(fio.sbi, blkaddr == NULL_ADDR);
1298
1299 if (fio.encrypted)
1300 fio.encrypted_page = cc->cpages[i - 1];
1301 else
1302 fio.compressed_page = cc->cpages[i - 1];
1303
1304 cc->cpages[i - 1] = NULL;
1305 f2fs_outplace_write_data(&dn, &fio);
1306 (*submitted)++;
1307unlock_continue:
1308 inode_dec_dirty_pages(cc->inode);
1309 unlock_page(fio.page);
1310 }
1311
1312 if (fio.compr_blocks)
1313 f2fs_i_compr_blocks_update(inode, fio.compr_blocks - 1, false);
1314 f2fs_i_compr_blocks_update(inode, cc->nr_cpages, true);
1315 add_compr_block_stat(inode, cc->nr_cpages);
1316
1317 set_inode_flag(cc->inode, FI_APPEND_WRITE);
1318 if (cc->cluster_idx == 0)
1319 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1320
1321 f2fs_put_dnode(&dn);
1322 if (IS_NOQUOTA(inode))
1323 up_read(&sbi->node_write);
1324 else
1325 f2fs_unlock_op(sbi);
1326
1327 spin_lock(&fi->i_size_lock);
1328 if (fi->last_disk_size < psize)
1329 fi->last_disk_size = psize;
1330 spin_unlock(&fi->i_size_lock);
1331
1332 f2fs_put_rpages(cc);
1333 page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1334 cc->cpages = NULL;
1335 f2fs_destroy_compress_ctx(cc, false);
1336 return 0;
1337
1338out_destroy_crypt:
1339 page_array_free(cc->inode, cic->rpages, cc->cluster_size);
1340
1341 for (--i; i >= 0; i--)
1342 fscrypt_finalize_bounce_page(&cc->cpages[i]);
1343out_put_cic:
1344 kmem_cache_free(cic_entry_slab, cic);
1345out_put_dnode:
1346 f2fs_put_dnode(&dn);
1347out_unlock_op:
1348 if (IS_NOQUOTA(inode))
1349 up_read(&sbi->node_write);
1350 else
1351 f2fs_unlock_op(sbi);
1352out_free:
1353 for (i = 0; i < cc->nr_cpages; i++) {
1354 if (!cc->cpages[i])
1355 continue;
1356 f2fs_compress_free_page(cc->cpages[i]);
1357 cc->cpages[i] = NULL;
1358 }
1359 page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1360 cc->cpages = NULL;
1361 return -EAGAIN;
1362}
1363
1364void f2fs_compress_write_end_io(struct bio *bio, struct page *page)
1365{
1366 struct f2fs_sb_info *sbi = bio->bi_private;
1367 struct compress_io_ctx *cic =
1368 (struct compress_io_ctx *)page_private(page);
1369 int i;
1370
1371 if (unlikely(bio->bi_status))
1372 mapping_set_error(cic->inode->i_mapping, -EIO);
1373
1374 f2fs_compress_free_page(page);
1375
1376 dec_page_count(sbi, F2FS_WB_DATA);
1377
1378 if (atomic_dec_return(&cic->pending_pages))
1379 return;
1380
1381 for (i = 0; i < cic->nr_rpages; i++) {
1382 WARN_ON(!cic->rpages[i]);
1383 clear_page_private_gcing(cic->rpages[i]);
1384 end_page_writeback(cic->rpages[i]);
1385 }
1386
1387 page_array_free(cic->inode, cic->rpages, cic->nr_rpages);
1388 kmem_cache_free(cic_entry_slab, cic);
1389}
1390
1391static int f2fs_write_raw_pages(struct compress_ctx *cc,
1392 int *submitted,
1393 struct writeback_control *wbc,
1394 enum iostat_type io_type)
1395{
1396 struct address_space *mapping = cc->inode->i_mapping;
1397 int _submitted, compr_blocks, ret;
1398 int i = -1, err = 0;
1399
1400 compr_blocks = f2fs_compressed_blocks(cc);
1401 if (compr_blocks < 0) {
1402 err = compr_blocks;
1403 goto out_err;
1404 }
1405
1406 for (i = 0; i < cc->cluster_size; i++) {
1407 if (!cc->rpages[i])
1408 continue;
1409retry_write:
1410 if (cc->rpages[i]->mapping != mapping) {
1411 unlock_page(cc->rpages[i]);
1412 continue;
1413 }
1414
1415 BUG_ON(!PageLocked(cc->rpages[i]));
1416
1417 ret = f2fs_write_single_data_page(cc->rpages[i], &_submitted,
1418 NULL, NULL, wbc, io_type,
1419 compr_blocks, false);
1420 if (ret) {
1421 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1422 unlock_page(cc->rpages[i]);
1423 ret = 0;
1424 } else if (ret == -EAGAIN) {
1425 /*
1426 * for quota file, just redirty left pages to
1427 * avoid deadlock caused by cluster update race
1428 * from foreground operation.
1429 */
1430 if (IS_NOQUOTA(cc->inode)) {
1431 err = 0;
1432 goto out_err;
1433 }
1434 ret = 0;
1435 cond_resched();
1436 congestion_wait(BLK_RW_ASYNC,
1437 DEFAULT_IO_TIMEOUT);
1438 lock_page(cc->rpages[i]);
1439
1440 if (!PageDirty(cc->rpages[i])) {
1441 unlock_page(cc->rpages[i]);
1442 continue;
1443 }
1444
1445 clear_page_dirty_for_io(cc->rpages[i]);
1446 goto retry_write;
1447 }
1448 err = ret;
1449 goto out_err;
1450 }
1451
1452 *submitted += _submitted;
1453 }
1454
1455 f2fs_balance_fs(F2FS_M_SB(mapping), true);
1456
1457 return 0;
1458out_err:
1459 for (++i; i < cc->cluster_size; i++) {
1460 if (!cc->rpages[i])
1461 continue;
1462 redirty_page_for_writepage(wbc, cc->rpages[i]);
1463 unlock_page(cc->rpages[i]);
1464 }
1465 return err;
1466}
1467
1468int f2fs_write_multi_pages(struct compress_ctx *cc,
1469 int *submitted,
1470 struct writeback_control *wbc,
1471 enum iostat_type io_type)
1472{
1473 int err;
1474
1475 *submitted = 0;
1476 if (cluster_may_compress(cc)) {
1477 err = f2fs_compress_pages(cc);
1478 if (err == -EAGAIN) {
1479 goto write;
1480 } else if (err) {
1481 f2fs_put_rpages_wbc(cc, wbc, true, 1);
1482 goto destroy_out;
1483 }
1484
1485 err = f2fs_write_compressed_pages(cc, submitted,
1486 wbc, io_type);
1487 if (!err)
1488 return 0;
1489 f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN);
1490 }
1491write:
1492 f2fs_bug_on(F2FS_I_SB(cc->inode), *submitted);
1493
1494 err = f2fs_write_raw_pages(cc, submitted, wbc, io_type);
1495 f2fs_put_rpages_wbc(cc, wbc, false, 0);
1496destroy_out:
1497 f2fs_destroy_compress_ctx(cc, false);
1498 return err;
1499}
1500
1501static void f2fs_free_dic(struct decompress_io_ctx *dic);
1502
1503struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
1504{
1505 struct decompress_io_ctx *dic;
1506 pgoff_t start_idx = start_idx_of_cluster(cc);
1507 int i;
1508
1509 dic = kmem_cache_zalloc(dic_entry_slab, GFP_NOFS);
1510 if (!dic)
1511 return ERR_PTR(-ENOMEM);
1512
1513 dic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1514 if (!dic->rpages) {
1515 kmem_cache_free(dic_entry_slab, dic);
1516 return ERR_PTR(-ENOMEM);
1517 }
1518
1519 dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1520 dic->inode = cc->inode;
1521 atomic_set(&dic->remaining_pages, cc->nr_cpages);
1522 dic->cluster_idx = cc->cluster_idx;
1523 dic->cluster_size = cc->cluster_size;
1524 dic->log_cluster_size = cc->log_cluster_size;
1525 dic->nr_cpages = cc->nr_cpages;
1526 refcount_set(&dic->refcnt, 1);
1527 dic->failed = false;
1528 dic->need_verity = f2fs_need_verity(cc->inode, start_idx);
1529
1530 for (i = 0; i < dic->cluster_size; i++)
1531 dic->rpages[i] = cc->rpages[i];
1532 dic->nr_rpages = cc->cluster_size;
1533
1534 dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages);
1535 if (!dic->cpages)
1536 goto out_free;
1537
1538 for (i = 0; i < dic->nr_cpages; i++) {
1539 struct page *page;
1540
1541 page = f2fs_compress_alloc_page();
1542 if (!page)
1543 goto out_free;
1544
1545 f2fs_set_compressed_page(page, cc->inode,
1546 start_idx + i + 1, dic);
1547 dic->cpages[i] = page;
1548 }
1549
1550 return dic;
1551
1552out_free:
1553 f2fs_free_dic(dic);
1554 return ERR_PTR(-ENOMEM);
1555}
1556
1557static void f2fs_free_dic(struct decompress_io_ctx *dic)
1558{
1559 int i;
1560
1561 if (dic->tpages) {
1562 for (i = 0; i < dic->cluster_size; i++) {
1563 if (dic->rpages[i])
1564 continue;
1565 if (!dic->tpages[i])
1566 continue;
1567 f2fs_compress_free_page(dic->tpages[i]);
1568 }
1569 page_array_free(dic->inode, dic->tpages, dic->cluster_size);
1570 }
1571
1572 if (dic->cpages) {
1573 for (i = 0; i < dic->nr_cpages; i++) {
1574 if (!dic->cpages[i])
1575 continue;
1576 f2fs_compress_free_page(dic->cpages[i]);
1577 }
1578 page_array_free(dic->inode, dic->cpages, dic->nr_cpages);
1579 }
1580
1581 page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
1582 kmem_cache_free(dic_entry_slab, dic);
1583}
1584
1585static void f2fs_put_dic(struct decompress_io_ctx *dic)
1586{
1587 if (refcount_dec_and_test(&dic->refcnt))
1588 f2fs_free_dic(dic);
1589}
1590
1591/*
1592 * Update and unlock the cluster's pagecache pages, and release the reference to
1593 * the decompress_io_ctx that was being held for I/O completion.
1594 */
1595static void __f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1596{
1597 int i;
1598
1599 for (i = 0; i < dic->cluster_size; i++) {
1600 struct page *rpage = dic->rpages[i];
1601
1602 if (!rpage)
1603 continue;
1604
1605 /* PG_error was set if verity failed. */
1606 if (failed || PageError(rpage)) {
1607 ClearPageUptodate(rpage);
1608 /* will re-read again later */
1609 ClearPageError(rpage);
1610 } else {
1611 SetPageUptodate(rpage);
1612 }
1613 unlock_page(rpage);
1614 }
1615
1616 f2fs_put_dic(dic);
1617}
1618
1619static void f2fs_verify_cluster(struct work_struct *work)
1620{
1621 struct decompress_io_ctx *dic =
1622 container_of(work, struct decompress_io_ctx, verity_work);
1623 int i;
1624
1625 /* Verify the cluster's decompressed pages with fs-verity. */
1626 for (i = 0; i < dic->cluster_size; i++) {
1627 struct page *rpage = dic->rpages[i];
1628
1629 if (rpage && !fsverity_verify_page(rpage))
1630 SetPageError(rpage);
1631 }
1632
1633 __f2fs_decompress_end_io(dic, false);
1634}
1635
1636/*
1637 * This is called when a compressed cluster has been decompressed
1638 * (or failed to be read and/or decompressed).
1639 */
1640void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1641{
1642 if (!failed && dic->need_verity) {
1643 /*
1644 * Note that to avoid deadlocks, the verity work can't be done
1645 * on the decompression workqueue. This is because verifying
1646 * the data pages can involve reading metadata pages from the
1647 * file, and these metadata pages may be compressed.
1648 */
1649 INIT_WORK(&dic->verity_work, f2fs_verify_cluster);
1650 fsverity_enqueue_verify_work(&dic->verity_work);
1651 } else {
1652 __f2fs_decompress_end_io(dic, failed);
1653 }
1654}
1655
1656/*
1657 * Put a reference to a compressed page's decompress_io_ctx.
1658 *
1659 * This is called when the page is no longer needed and can be freed.
1660 */
1661void f2fs_put_page_dic(struct page *page)
1662{
1663 struct decompress_io_ctx *dic =
1664 (struct decompress_io_ctx *)page_private(page);
1665
1666 f2fs_put_dic(dic);
1667}
1668
1669const struct address_space_operations f2fs_compress_aops = {
1670 .releasepage = f2fs_release_page,
1671 .invalidatepage = f2fs_invalidate_page,
1672};
1673
1674struct address_space *COMPRESS_MAPPING(struct f2fs_sb_info *sbi)
1675{
1676 return sbi->compress_inode->i_mapping;
1677}
1678
1679void f2fs_invalidate_compress_page(struct f2fs_sb_info *sbi, block_t blkaddr)
1680{
1681 if (!sbi->compress_inode)
1682 return;
1683 invalidate_mapping_pages(COMPRESS_MAPPING(sbi), blkaddr, blkaddr);
1684}
1685
1686void f2fs_cache_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
1687 nid_t ino, block_t blkaddr)
1688{
1689 struct page *cpage;
1690 int ret;
1691
1692 if (!test_opt(sbi, COMPRESS_CACHE))
1693 return;
1694
1695 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
1696 return;
1697
1698 if (!f2fs_available_free_memory(sbi, COMPRESS_PAGE))
1699 return;
1700
1701 cpage = find_get_page(COMPRESS_MAPPING(sbi), blkaddr);
1702 if (cpage) {
1703 f2fs_put_page(cpage, 0);
1704 return;
1705 }
1706
1707 cpage = alloc_page(__GFP_NOWARN | __GFP_IO);
1708 if (!cpage)
1709 return;
1710
1711 ret = add_to_page_cache_lru(cpage, COMPRESS_MAPPING(sbi),
1712 blkaddr, GFP_NOFS);
1713 if (ret) {
1714 f2fs_put_page(cpage, 0);
1715 return;
1716 }
1717
1718 set_page_private_data(cpage, ino);
1719
1720 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
1721 goto out;
1722
1723 memcpy(page_address(cpage), page_address(page), PAGE_SIZE);
1724 SetPageUptodate(cpage);
1725out:
1726 f2fs_put_page(cpage, 1);
1727}
1728
1729bool f2fs_load_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
1730 block_t blkaddr)
1731{
1732 struct page *cpage;
1733 bool hitted = false;
1734
1735 if (!test_opt(sbi, COMPRESS_CACHE))
1736 return false;
1737
1738 cpage = f2fs_pagecache_get_page(COMPRESS_MAPPING(sbi),
1739 blkaddr, FGP_LOCK | FGP_NOWAIT, GFP_NOFS);
1740 if (cpage) {
1741 if (PageUptodate(cpage)) {
1742 atomic_inc(&sbi->compress_page_hit);
1743 memcpy(page_address(page),
1744 page_address(cpage), PAGE_SIZE);
1745 hitted = true;
1746 }
1747 f2fs_put_page(cpage, 1);
1748 }
1749
1750 return hitted;
1751}
1752
1753void f2fs_invalidate_compress_pages(struct f2fs_sb_info *sbi, nid_t ino)
1754{
1755 struct address_space *mapping = sbi->compress_inode->i_mapping;
1756 struct pagevec pvec;
1757 pgoff_t index = 0;
1758 pgoff_t end = MAX_BLKADDR(sbi);
1759
1760 if (!mapping->nrpages)
1761 return;
1762
1763 pagevec_init(&pvec);
1764
1765 do {
1766 unsigned int nr_pages;
1767 int i;
1768
1769 nr_pages = pagevec_lookup_range(&pvec, mapping,
1770 &index, end - 1);
1771 if (!nr_pages)
1772 break;
1773
1774 for (i = 0; i < nr_pages; i++) {
1775 struct page *page = pvec.pages[i];
1776
1777 if (page->index > end)
1778 break;
1779
1780 lock_page(page);
1781 if (page->mapping != mapping) {
1782 unlock_page(page);
1783 continue;
1784 }
1785
1786 if (ino != get_page_private_data(page)) {
1787 unlock_page(page);
1788 continue;
1789 }
1790
1791 generic_error_remove_page(mapping, page);
1792 unlock_page(page);
1793 }
1794 pagevec_release(&pvec);
1795 cond_resched();
1796 } while (index < end);
1797}
1798
1799int f2fs_init_compress_inode(struct f2fs_sb_info *sbi)
1800{
1801 struct inode *inode;
1802
1803 if (!test_opt(sbi, COMPRESS_CACHE))
1804 return 0;
1805
1806 inode = f2fs_iget(sbi->sb, F2FS_COMPRESS_INO(sbi));
1807 if (IS_ERR(inode))
1808 return PTR_ERR(inode);
1809 sbi->compress_inode = inode;
1810
1811 sbi->compress_percent = COMPRESS_PERCENT;
1812 sbi->compress_watermark = COMPRESS_WATERMARK;
1813
1814 atomic_set(&sbi->compress_page_hit, 0);
1815
1816 return 0;
1817}
1818
1819void f2fs_destroy_compress_inode(struct f2fs_sb_info *sbi)
1820{
1821 if (!sbi->compress_inode)
1822 return;
1823 iput(sbi->compress_inode);
1824 sbi->compress_inode = NULL;
1825}
1826
1827int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi)
1828{
1829 dev_t dev = sbi->sb->s_bdev->bd_dev;
1830 char slab_name[32];
1831
1832 sprintf(slab_name, "f2fs_page_array_entry-%u:%u", MAJOR(dev), MINOR(dev));
1833
1834 sbi->page_array_slab_size = sizeof(struct page *) <<
1835 F2FS_OPTION(sbi).compress_log_size;
1836
1837 sbi->page_array_slab = f2fs_kmem_cache_create(slab_name,
1838 sbi->page_array_slab_size);
1839 if (!sbi->page_array_slab)
1840 return -ENOMEM;
1841 return 0;
1842}
1843
1844void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi)
1845{
1846 kmem_cache_destroy(sbi->page_array_slab);
1847}
1848
1849static int __init f2fs_init_cic_cache(void)
1850{
1851 cic_entry_slab = f2fs_kmem_cache_create("f2fs_cic_entry",
1852 sizeof(struct compress_io_ctx));
1853 if (!cic_entry_slab)
1854 return -ENOMEM;
1855 return 0;
1856}
1857
1858static void f2fs_destroy_cic_cache(void)
1859{
1860 kmem_cache_destroy(cic_entry_slab);
1861}
1862
1863static int __init f2fs_init_dic_cache(void)
1864{
1865 dic_entry_slab = f2fs_kmem_cache_create("f2fs_dic_entry",
1866 sizeof(struct decompress_io_ctx));
1867 if (!dic_entry_slab)
1868 return -ENOMEM;
1869 return 0;
1870}
1871
1872static void f2fs_destroy_dic_cache(void)
1873{
1874 kmem_cache_destroy(dic_entry_slab);
1875}
1876
1877int __init f2fs_init_compress_cache(void)
1878{
1879 int err;
1880
1881 err = f2fs_init_cic_cache();
1882 if (err)
1883 goto out;
1884 err = f2fs_init_dic_cache();
1885 if (err)
1886 goto free_cic;
1887 return 0;
1888free_cic:
1889 f2fs_destroy_cic_cache();
1890out:
1891 return -ENOMEM;
1892}
1893
1894void f2fs_destroy_compress_cache(void)
1895{
1896 f2fs_destroy_dic_cache();
1897 f2fs_destroy_cic_cache();
1898}
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/moduleparam.h>
11#include <linux/writeback.h>
12#include <linux/backing-dev.h>
13#include <linux/lzo.h>
14#include <linux/lz4.h>
15#include <linux/zstd.h>
16#include <linux/pagevec.h>
17
18#include "f2fs.h"
19#include "node.h"
20#include "segment.h"
21#include <trace/events/f2fs.h>
22
23static struct kmem_cache *cic_entry_slab;
24static struct kmem_cache *dic_entry_slab;
25
26static void *page_array_alloc(struct inode *inode, int nr)
27{
28 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
29 unsigned int size = sizeof(struct page *) * nr;
30
31 if (likely(size <= sbi->page_array_slab_size))
32 return f2fs_kmem_cache_alloc(sbi->page_array_slab,
33 GFP_F2FS_ZERO, false, F2FS_I_SB(inode));
34 return f2fs_kzalloc(sbi, size, GFP_NOFS);
35}
36
37static void page_array_free(struct inode *inode, void *pages, int nr)
38{
39 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
40 unsigned int size = sizeof(struct page *) * nr;
41
42 if (!pages)
43 return;
44
45 if (likely(size <= sbi->page_array_slab_size))
46 kmem_cache_free(sbi->page_array_slab, pages);
47 else
48 kfree(pages);
49}
50
51struct f2fs_compress_ops {
52 int (*init_compress_ctx)(struct compress_ctx *cc);
53 void (*destroy_compress_ctx)(struct compress_ctx *cc);
54 int (*compress_pages)(struct compress_ctx *cc);
55 int (*init_decompress_ctx)(struct decompress_io_ctx *dic);
56 void (*destroy_decompress_ctx)(struct decompress_io_ctx *dic);
57 int (*decompress_pages)(struct decompress_io_ctx *dic);
58 bool (*is_level_valid)(int level);
59};
60
61static unsigned int offset_in_cluster(struct compress_ctx *cc, pgoff_t index)
62{
63 return index & (cc->cluster_size - 1);
64}
65
66static pgoff_t cluster_idx(struct compress_ctx *cc, pgoff_t index)
67{
68 return index >> cc->log_cluster_size;
69}
70
71static pgoff_t start_idx_of_cluster(struct compress_ctx *cc)
72{
73 return cc->cluster_idx << cc->log_cluster_size;
74}
75
76bool f2fs_is_compressed_page(struct page *page)
77{
78 if (!PagePrivate(page))
79 return false;
80 if (!page_private(page))
81 return false;
82 if (page_private_nonpointer(page))
83 return false;
84
85 f2fs_bug_on(F2FS_M_SB(page->mapping),
86 *((u32 *)page_private(page)) != F2FS_COMPRESSED_PAGE_MAGIC);
87 return true;
88}
89
90static void f2fs_set_compressed_page(struct page *page,
91 struct inode *inode, pgoff_t index, void *data)
92{
93 struct folio *folio = page_folio(page);
94
95 folio_attach_private(folio, (void *)data);
96
97 /* i_crypto_info and iv index */
98 folio->index = index;
99 folio->mapping = inode->i_mapping;
100}
101
102static void f2fs_drop_rpages(struct compress_ctx *cc, int len, bool unlock)
103{
104 int i;
105
106 for (i = 0; i < len; i++) {
107 if (!cc->rpages[i])
108 continue;
109 if (unlock)
110 unlock_page(cc->rpages[i]);
111 else
112 put_page(cc->rpages[i]);
113 }
114}
115
116static void f2fs_put_rpages(struct compress_ctx *cc)
117{
118 f2fs_drop_rpages(cc, cc->cluster_size, false);
119}
120
121static void f2fs_unlock_rpages(struct compress_ctx *cc, int len)
122{
123 f2fs_drop_rpages(cc, len, true);
124}
125
126static void f2fs_put_rpages_wbc(struct compress_ctx *cc,
127 struct writeback_control *wbc, bool redirty, int unlock)
128{
129 unsigned int i;
130
131 for (i = 0; i < cc->cluster_size; i++) {
132 if (!cc->rpages[i])
133 continue;
134 if (redirty)
135 redirty_page_for_writepage(wbc, cc->rpages[i]);
136 f2fs_put_page(cc->rpages[i], unlock);
137 }
138}
139
140struct page *f2fs_compress_control_page(struct page *page)
141{
142 return ((struct compress_io_ctx *)page_private(page))->rpages[0];
143}
144
145int f2fs_init_compress_ctx(struct compress_ctx *cc)
146{
147 if (cc->rpages)
148 return 0;
149
150 cc->rpages = page_array_alloc(cc->inode, cc->cluster_size);
151 return cc->rpages ? 0 : -ENOMEM;
152}
153
154void f2fs_destroy_compress_ctx(struct compress_ctx *cc, bool reuse)
155{
156 page_array_free(cc->inode, cc->rpages, cc->cluster_size);
157 cc->rpages = NULL;
158 cc->nr_rpages = 0;
159 cc->nr_cpages = 0;
160 cc->valid_nr_cpages = 0;
161 if (!reuse)
162 cc->cluster_idx = NULL_CLUSTER;
163}
164
165void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct folio *folio)
166{
167 unsigned int cluster_ofs;
168
169 if (!f2fs_cluster_can_merge_page(cc, folio->index))
170 f2fs_bug_on(F2FS_I_SB(cc->inode), 1);
171
172 cluster_ofs = offset_in_cluster(cc, folio->index);
173 cc->rpages[cluster_ofs] = folio_page(folio, 0);
174 cc->nr_rpages++;
175 cc->cluster_idx = cluster_idx(cc, folio->index);
176}
177
178#ifdef CONFIG_F2FS_FS_LZO
179static int lzo_init_compress_ctx(struct compress_ctx *cc)
180{
181 cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
182 LZO1X_MEM_COMPRESS, GFP_NOFS);
183 if (!cc->private)
184 return -ENOMEM;
185
186 cc->clen = lzo1x_worst_compress(PAGE_SIZE << cc->log_cluster_size);
187 return 0;
188}
189
190static void lzo_destroy_compress_ctx(struct compress_ctx *cc)
191{
192 kvfree(cc->private);
193 cc->private = NULL;
194}
195
196static int lzo_compress_pages(struct compress_ctx *cc)
197{
198 int ret;
199
200 ret = lzo1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
201 &cc->clen, cc->private);
202 if (ret != LZO_E_OK) {
203 f2fs_err_ratelimited(F2FS_I_SB(cc->inode),
204 "lzo compress failed, ret:%d", ret);
205 return -EIO;
206 }
207 return 0;
208}
209
210static int lzo_decompress_pages(struct decompress_io_ctx *dic)
211{
212 int ret;
213
214 ret = lzo1x_decompress_safe(dic->cbuf->cdata, dic->clen,
215 dic->rbuf, &dic->rlen);
216 if (ret != LZO_E_OK) {
217 f2fs_err_ratelimited(F2FS_I_SB(dic->inode),
218 "lzo decompress failed, ret:%d", ret);
219 return -EIO;
220 }
221
222 if (dic->rlen != PAGE_SIZE << dic->log_cluster_size) {
223 f2fs_err_ratelimited(F2FS_I_SB(dic->inode),
224 "lzo invalid rlen:%zu, expected:%lu",
225 dic->rlen, PAGE_SIZE << dic->log_cluster_size);
226 return -EIO;
227 }
228 return 0;
229}
230
231static const struct f2fs_compress_ops f2fs_lzo_ops = {
232 .init_compress_ctx = lzo_init_compress_ctx,
233 .destroy_compress_ctx = lzo_destroy_compress_ctx,
234 .compress_pages = lzo_compress_pages,
235 .decompress_pages = lzo_decompress_pages,
236};
237#endif
238
239#ifdef CONFIG_F2FS_FS_LZ4
240static int lz4_init_compress_ctx(struct compress_ctx *cc)
241{
242 unsigned int size = LZ4_MEM_COMPRESS;
243
244#ifdef CONFIG_F2FS_FS_LZ4HC
245 if (F2FS_I(cc->inode)->i_compress_level)
246 size = LZ4HC_MEM_COMPRESS;
247#endif
248
249 cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), size, GFP_NOFS);
250 if (!cc->private)
251 return -ENOMEM;
252
253 /*
254 * we do not change cc->clen to LZ4_compressBound(inputsize) to
255 * adapt worst compress case, because lz4 compressor can handle
256 * output budget properly.
257 */
258 cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
259 return 0;
260}
261
262static void lz4_destroy_compress_ctx(struct compress_ctx *cc)
263{
264 kvfree(cc->private);
265 cc->private = NULL;
266}
267
268static int lz4_compress_pages(struct compress_ctx *cc)
269{
270 int len = -EINVAL;
271 unsigned char level = F2FS_I(cc->inode)->i_compress_level;
272
273 if (!level)
274 len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
275 cc->clen, cc->private);
276#ifdef CONFIG_F2FS_FS_LZ4HC
277 else
278 len = LZ4_compress_HC(cc->rbuf, cc->cbuf->cdata, cc->rlen,
279 cc->clen, level, cc->private);
280#endif
281 if (len < 0)
282 return len;
283 if (!len)
284 return -EAGAIN;
285
286 cc->clen = len;
287 return 0;
288}
289
290static int lz4_decompress_pages(struct decompress_io_ctx *dic)
291{
292 int ret;
293
294 ret = LZ4_decompress_safe(dic->cbuf->cdata, dic->rbuf,
295 dic->clen, dic->rlen);
296 if (ret < 0) {
297 f2fs_err_ratelimited(F2FS_I_SB(dic->inode),
298 "lz4 decompress failed, ret:%d", ret);
299 return -EIO;
300 }
301
302 if (ret != PAGE_SIZE << dic->log_cluster_size) {
303 f2fs_err_ratelimited(F2FS_I_SB(dic->inode),
304 "lz4 invalid ret:%d, expected:%lu",
305 ret, PAGE_SIZE << dic->log_cluster_size);
306 return -EIO;
307 }
308 return 0;
309}
310
311static bool lz4_is_level_valid(int lvl)
312{
313#ifdef CONFIG_F2FS_FS_LZ4HC
314 return !lvl || (lvl >= LZ4HC_MIN_CLEVEL && lvl <= LZ4HC_MAX_CLEVEL);
315#else
316 return lvl == 0;
317#endif
318}
319
320static const struct f2fs_compress_ops f2fs_lz4_ops = {
321 .init_compress_ctx = lz4_init_compress_ctx,
322 .destroy_compress_ctx = lz4_destroy_compress_ctx,
323 .compress_pages = lz4_compress_pages,
324 .decompress_pages = lz4_decompress_pages,
325 .is_level_valid = lz4_is_level_valid,
326};
327#endif
328
329#ifdef CONFIG_F2FS_FS_ZSTD
330static int zstd_init_compress_ctx(struct compress_ctx *cc)
331{
332 zstd_parameters params;
333 zstd_cstream *stream;
334 void *workspace;
335 unsigned int workspace_size;
336 unsigned char level = F2FS_I(cc->inode)->i_compress_level;
337
338 /* Need to remain this for backward compatibility */
339 if (!level)
340 level = F2FS_ZSTD_DEFAULT_CLEVEL;
341
342 params = zstd_get_params(level, cc->rlen);
343 workspace_size = zstd_cstream_workspace_bound(¶ms.cParams);
344
345 workspace = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
346 workspace_size, GFP_NOFS);
347 if (!workspace)
348 return -ENOMEM;
349
350 stream = zstd_init_cstream(¶ms, 0, workspace, workspace_size);
351 if (!stream) {
352 f2fs_err_ratelimited(F2FS_I_SB(cc->inode),
353 "%s zstd_init_cstream failed", __func__);
354 kvfree(workspace);
355 return -EIO;
356 }
357
358 cc->private = workspace;
359 cc->private2 = stream;
360
361 cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
362 return 0;
363}
364
365static void zstd_destroy_compress_ctx(struct compress_ctx *cc)
366{
367 kvfree(cc->private);
368 cc->private = NULL;
369 cc->private2 = NULL;
370}
371
372static int zstd_compress_pages(struct compress_ctx *cc)
373{
374 zstd_cstream *stream = cc->private2;
375 zstd_in_buffer inbuf;
376 zstd_out_buffer outbuf;
377 int src_size = cc->rlen;
378 int dst_size = src_size - PAGE_SIZE - COMPRESS_HEADER_SIZE;
379 int ret;
380
381 inbuf.pos = 0;
382 inbuf.src = cc->rbuf;
383 inbuf.size = src_size;
384
385 outbuf.pos = 0;
386 outbuf.dst = cc->cbuf->cdata;
387 outbuf.size = dst_size;
388
389 ret = zstd_compress_stream(stream, &outbuf, &inbuf);
390 if (zstd_is_error(ret)) {
391 f2fs_err_ratelimited(F2FS_I_SB(cc->inode),
392 "%s zstd_compress_stream failed, ret: %d",
393 __func__, zstd_get_error_code(ret));
394 return -EIO;
395 }
396
397 ret = zstd_end_stream(stream, &outbuf);
398 if (zstd_is_error(ret)) {
399 f2fs_err_ratelimited(F2FS_I_SB(cc->inode),
400 "%s zstd_end_stream returned %d",
401 __func__, zstd_get_error_code(ret));
402 return -EIO;
403 }
404
405 /*
406 * there is compressed data remained in intermediate buffer due to
407 * no more space in cbuf.cdata
408 */
409 if (ret)
410 return -EAGAIN;
411
412 cc->clen = outbuf.pos;
413 return 0;
414}
415
416static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
417{
418 zstd_dstream *stream;
419 void *workspace;
420 unsigned int workspace_size;
421 unsigned int max_window_size =
422 MAX_COMPRESS_WINDOW_SIZE(dic->log_cluster_size);
423
424 workspace_size = zstd_dstream_workspace_bound(max_window_size);
425
426 workspace = f2fs_kvmalloc(F2FS_I_SB(dic->inode),
427 workspace_size, GFP_NOFS);
428 if (!workspace)
429 return -ENOMEM;
430
431 stream = zstd_init_dstream(max_window_size, workspace, workspace_size);
432 if (!stream) {
433 f2fs_err_ratelimited(F2FS_I_SB(dic->inode),
434 "%s zstd_init_dstream failed", __func__);
435 kvfree(workspace);
436 return -EIO;
437 }
438
439 dic->private = workspace;
440 dic->private2 = stream;
441
442 return 0;
443}
444
445static void zstd_destroy_decompress_ctx(struct decompress_io_ctx *dic)
446{
447 kvfree(dic->private);
448 dic->private = NULL;
449 dic->private2 = NULL;
450}
451
452static int zstd_decompress_pages(struct decompress_io_ctx *dic)
453{
454 zstd_dstream *stream = dic->private2;
455 zstd_in_buffer inbuf;
456 zstd_out_buffer outbuf;
457 int ret;
458
459 inbuf.pos = 0;
460 inbuf.src = dic->cbuf->cdata;
461 inbuf.size = dic->clen;
462
463 outbuf.pos = 0;
464 outbuf.dst = dic->rbuf;
465 outbuf.size = dic->rlen;
466
467 ret = zstd_decompress_stream(stream, &outbuf, &inbuf);
468 if (zstd_is_error(ret)) {
469 f2fs_err_ratelimited(F2FS_I_SB(dic->inode),
470 "%s zstd_decompress_stream failed, ret: %d",
471 __func__, zstd_get_error_code(ret));
472 return -EIO;
473 }
474
475 if (dic->rlen != outbuf.pos) {
476 f2fs_err_ratelimited(F2FS_I_SB(dic->inode),
477 "%s ZSTD invalid rlen:%zu, expected:%lu",
478 __func__, dic->rlen,
479 PAGE_SIZE << dic->log_cluster_size);
480 return -EIO;
481 }
482
483 return 0;
484}
485
486static bool zstd_is_level_valid(int lvl)
487{
488 return lvl >= zstd_min_clevel() && lvl <= zstd_max_clevel();
489}
490
491static const struct f2fs_compress_ops f2fs_zstd_ops = {
492 .init_compress_ctx = zstd_init_compress_ctx,
493 .destroy_compress_ctx = zstd_destroy_compress_ctx,
494 .compress_pages = zstd_compress_pages,
495 .init_decompress_ctx = zstd_init_decompress_ctx,
496 .destroy_decompress_ctx = zstd_destroy_decompress_ctx,
497 .decompress_pages = zstd_decompress_pages,
498 .is_level_valid = zstd_is_level_valid,
499};
500#endif
501
502#ifdef CONFIG_F2FS_FS_LZO
503#ifdef CONFIG_F2FS_FS_LZORLE
504static int lzorle_compress_pages(struct compress_ctx *cc)
505{
506 int ret;
507
508 ret = lzorle1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
509 &cc->clen, cc->private);
510 if (ret != LZO_E_OK) {
511 f2fs_err_ratelimited(F2FS_I_SB(cc->inode),
512 "lzo-rle compress failed, ret:%d", ret);
513 return -EIO;
514 }
515 return 0;
516}
517
518static const struct f2fs_compress_ops f2fs_lzorle_ops = {
519 .init_compress_ctx = lzo_init_compress_ctx,
520 .destroy_compress_ctx = lzo_destroy_compress_ctx,
521 .compress_pages = lzorle_compress_pages,
522 .decompress_pages = lzo_decompress_pages,
523};
524#endif
525#endif
526
527static const struct f2fs_compress_ops *f2fs_cops[COMPRESS_MAX] = {
528#ifdef CONFIG_F2FS_FS_LZO
529 &f2fs_lzo_ops,
530#else
531 NULL,
532#endif
533#ifdef CONFIG_F2FS_FS_LZ4
534 &f2fs_lz4_ops,
535#else
536 NULL,
537#endif
538#ifdef CONFIG_F2FS_FS_ZSTD
539 &f2fs_zstd_ops,
540#else
541 NULL,
542#endif
543#if defined(CONFIG_F2FS_FS_LZO) && defined(CONFIG_F2FS_FS_LZORLE)
544 &f2fs_lzorle_ops,
545#else
546 NULL,
547#endif
548};
549
550bool f2fs_is_compress_backend_ready(struct inode *inode)
551{
552 if (!f2fs_compressed_file(inode))
553 return true;
554 return f2fs_cops[F2FS_I(inode)->i_compress_algorithm];
555}
556
557bool f2fs_is_compress_level_valid(int alg, int lvl)
558{
559 const struct f2fs_compress_ops *cops = f2fs_cops[alg];
560
561 if (cops->is_level_valid)
562 return cops->is_level_valid(lvl);
563
564 return lvl == 0;
565}
566
567static mempool_t *compress_page_pool;
568static int num_compress_pages = 512;
569module_param(num_compress_pages, uint, 0444);
570MODULE_PARM_DESC(num_compress_pages,
571 "Number of intermediate compress pages to preallocate");
572
573int __init f2fs_init_compress_mempool(void)
574{
575 compress_page_pool = mempool_create_page_pool(num_compress_pages, 0);
576 return compress_page_pool ? 0 : -ENOMEM;
577}
578
579void f2fs_destroy_compress_mempool(void)
580{
581 mempool_destroy(compress_page_pool);
582}
583
584static struct page *f2fs_compress_alloc_page(void)
585{
586 struct page *page;
587
588 page = mempool_alloc(compress_page_pool, GFP_NOFS);
589 lock_page(page);
590
591 return page;
592}
593
594static void f2fs_compress_free_page(struct page *page)
595{
596 if (!page)
597 return;
598 detach_page_private(page);
599 page->mapping = NULL;
600 unlock_page(page);
601 mempool_free(page, compress_page_pool);
602}
603
604#define MAX_VMAP_RETRIES 3
605
606static void *f2fs_vmap(struct page **pages, unsigned int count)
607{
608 int i;
609 void *buf = NULL;
610
611 for (i = 0; i < MAX_VMAP_RETRIES; i++) {
612 buf = vm_map_ram(pages, count, -1);
613 if (buf)
614 break;
615 vm_unmap_aliases();
616 }
617 return buf;
618}
619
620static int f2fs_compress_pages(struct compress_ctx *cc)
621{
622 struct f2fs_inode_info *fi = F2FS_I(cc->inode);
623 const struct f2fs_compress_ops *cops =
624 f2fs_cops[fi->i_compress_algorithm];
625 unsigned int max_len, new_nr_cpages;
626 u32 chksum = 0;
627 int i, ret;
628
629 trace_f2fs_compress_pages_start(cc->inode, cc->cluster_idx,
630 cc->cluster_size, fi->i_compress_algorithm);
631
632 if (cops->init_compress_ctx) {
633 ret = cops->init_compress_ctx(cc);
634 if (ret)
635 goto out;
636 }
637
638 max_len = COMPRESS_HEADER_SIZE + cc->clen;
639 cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE);
640 cc->valid_nr_cpages = cc->nr_cpages;
641
642 cc->cpages = page_array_alloc(cc->inode, cc->nr_cpages);
643 if (!cc->cpages) {
644 ret = -ENOMEM;
645 goto destroy_compress_ctx;
646 }
647
648 for (i = 0; i < cc->nr_cpages; i++)
649 cc->cpages[i] = f2fs_compress_alloc_page();
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 & BIT(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 /* zero out any unused part of the last page */
687 memset(&cc->cbuf->cdata[cc->clen], 0,
688 (new_nr_cpages * PAGE_SIZE) -
689 (cc->clen + COMPRESS_HEADER_SIZE));
690
691 vm_unmap_ram(cc->cbuf, cc->nr_cpages);
692 vm_unmap_ram(cc->rbuf, cc->cluster_size);
693
694 for (i = new_nr_cpages; i < cc->nr_cpages; i++) {
695 f2fs_compress_free_page(cc->cpages[i]);
696 cc->cpages[i] = NULL;
697 }
698
699 if (cops->destroy_compress_ctx)
700 cops->destroy_compress_ctx(cc);
701
702 cc->valid_nr_cpages = new_nr_cpages;
703
704 trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
705 cc->clen, ret);
706 return 0;
707
708out_vunmap_cbuf:
709 vm_unmap_ram(cc->cbuf, cc->nr_cpages);
710out_vunmap_rbuf:
711 vm_unmap_ram(cc->rbuf, cc->cluster_size);
712out_free_cpages:
713 for (i = 0; i < cc->nr_cpages; i++) {
714 if (cc->cpages[i])
715 f2fs_compress_free_page(cc->cpages[i]);
716 }
717 page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
718 cc->cpages = NULL;
719destroy_compress_ctx:
720 if (cops->destroy_compress_ctx)
721 cops->destroy_compress_ctx(cc);
722out:
723 trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
724 cc->clen, ret);
725 return ret;
726}
727
728static int f2fs_prepare_decomp_mem(struct decompress_io_ctx *dic,
729 bool pre_alloc);
730static void f2fs_release_decomp_mem(struct decompress_io_ctx *dic,
731 bool bypass_destroy_callback, bool pre_alloc);
732
733void f2fs_decompress_cluster(struct decompress_io_ctx *dic, bool in_task)
734{
735 struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
736 struct f2fs_inode_info *fi = F2FS_I(dic->inode);
737 const struct f2fs_compress_ops *cops =
738 f2fs_cops[fi->i_compress_algorithm];
739 bool bypass_callback = false;
740 int ret;
741
742 trace_f2fs_decompress_pages_start(dic->inode, dic->cluster_idx,
743 dic->cluster_size, fi->i_compress_algorithm);
744
745 if (dic->failed) {
746 ret = -EIO;
747 goto out_end_io;
748 }
749
750 ret = f2fs_prepare_decomp_mem(dic, false);
751 if (ret) {
752 bypass_callback = true;
753 goto out_release;
754 }
755
756 dic->clen = le32_to_cpu(dic->cbuf->clen);
757 dic->rlen = PAGE_SIZE << dic->log_cluster_size;
758
759 if (dic->clen > PAGE_SIZE * dic->nr_cpages - COMPRESS_HEADER_SIZE) {
760 ret = -EFSCORRUPTED;
761
762 /* Avoid f2fs_commit_super in irq context */
763 if (!in_task)
764 f2fs_handle_error_async(sbi, ERROR_FAIL_DECOMPRESSION);
765 else
766 f2fs_handle_error(sbi, ERROR_FAIL_DECOMPRESSION);
767 goto out_release;
768 }
769
770 ret = cops->decompress_pages(dic);
771
772 if (!ret && (fi->i_compress_flag & BIT(COMPRESS_CHKSUM))) {
773 u32 provided = le32_to_cpu(dic->cbuf->chksum);
774 u32 calculated = f2fs_crc32(sbi, dic->cbuf->cdata, dic->clen);
775
776 if (provided != calculated) {
777 if (!is_inode_flag_set(dic->inode, FI_COMPRESS_CORRUPT)) {
778 set_inode_flag(dic->inode, FI_COMPRESS_CORRUPT);
779 f2fs_info_ratelimited(sbi,
780 "checksum invalid, nid = %lu, %x vs %x",
781 dic->inode->i_ino,
782 provided, calculated);
783 }
784 set_sbi_flag(sbi, SBI_NEED_FSCK);
785 }
786 }
787
788out_release:
789 f2fs_release_decomp_mem(dic, bypass_callback, false);
790
791out_end_io:
792 trace_f2fs_decompress_pages_end(dic->inode, dic->cluster_idx,
793 dic->clen, ret);
794 f2fs_decompress_end_io(dic, ret, in_task);
795}
796
797/*
798 * This is called when a page of a compressed cluster has been read from disk
799 * (or failed to be read from disk). It checks whether this page was the last
800 * page being waited on in the cluster, and if so, it decompresses the cluster
801 * (or in the case of a failure, cleans up without actually decompressing).
802 */
803void f2fs_end_read_compressed_page(struct page *page, bool failed,
804 block_t blkaddr, bool in_task)
805{
806 struct decompress_io_ctx *dic =
807 (struct decompress_io_ctx *)page_private(page);
808 struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
809
810 dec_page_count(sbi, F2FS_RD_DATA);
811
812 if (failed)
813 WRITE_ONCE(dic->failed, true);
814 else if (blkaddr && in_task)
815 f2fs_cache_compressed_page(sbi, page,
816 dic->inode->i_ino, blkaddr);
817
818 if (atomic_dec_and_test(&dic->remaining_pages))
819 f2fs_decompress_cluster(dic, in_task);
820}
821
822static bool is_page_in_cluster(struct compress_ctx *cc, pgoff_t index)
823{
824 if (cc->cluster_idx == NULL_CLUSTER)
825 return true;
826 return cc->cluster_idx == cluster_idx(cc, index);
827}
828
829bool f2fs_cluster_is_empty(struct compress_ctx *cc)
830{
831 return cc->nr_rpages == 0;
832}
833
834static bool f2fs_cluster_is_full(struct compress_ctx *cc)
835{
836 return cc->cluster_size == cc->nr_rpages;
837}
838
839bool f2fs_cluster_can_merge_page(struct compress_ctx *cc, pgoff_t index)
840{
841 if (f2fs_cluster_is_empty(cc))
842 return true;
843 return is_page_in_cluster(cc, index);
844}
845
846bool f2fs_all_cluster_page_ready(struct compress_ctx *cc, struct page **pages,
847 int index, int nr_pages, bool uptodate)
848{
849 unsigned long pgidx = pages[index]->index;
850 int i = uptodate ? 0 : 1;
851
852 /*
853 * when uptodate set to true, try to check all pages in cluster is
854 * uptodate or not.
855 */
856 if (uptodate && (pgidx % cc->cluster_size))
857 return false;
858
859 if (nr_pages - index < cc->cluster_size)
860 return false;
861
862 for (; i < cc->cluster_size; i++) {
863 if (pages[index + i]->index != pgidx + i)
864 return false;
865 if (uptodate && !PageUptodate(pages[index + i]))
866 return false;
867 }
868
869 return true;
870}
871
872static bool cluster_has_invalid_data(struct compress_ctx *cc)
873{
874 loff_t i_size = i_size_read(cc->inode);
875 unsigned nr_pages = DIV_ROUND_UP(i_size, PAGE_SIZE);
876 int i;
877
878 for (i = 0; i < cc->cluster_size; i++) {
879 struct page *page = cc->rpages[i];
880
881 f2fs_bug_on(F2FS_I_SB(cc->inode), !page);
882
883 /* beyond EOF */
884 if (page_folio(page)->index >= nr_pages)
885 return true;
886 }
887 return false;
888}
889
890bool f2fs_sanity_check_cluster(struct dnode_of_data *dn)
891{
892#ifdef CONFIG_F2FS_CHECK_FS
893 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
894 unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size;
895 int cluster_end = 0;
896 unsigned int count;
897 int i;
898 char *reason = "";
899
900 if (dn->data_blkaddr != COMPRESS_ADDR)
901 return false;
902
903 /* [..., COMPR_ADDR, ...] */
904 if (dn->ofs_in_node % cluster_size) {
905 reason = "[*|C|*|*]";
906 goto out;
907 }
908
909 for (i = 1, count = 1; i < cluster_size; i++, count++) {
910 block_t blkaddr = data_blkaddr(dn->inode, dn->node_page,
911 dn->ofs_in_node + i);
912
913 /* [COMPR_ADDR, ..., COMPR_ADDR] */
914 if (blkaddr == COMPRESS_ADDR) {
915 reason = "[C|*|C|*]";
916 goto out;
917 }
918 if (!__is_valid_data_blkaddr(blkaddr)) {
919 if (!cluster_end)
920 cluster_end = i;
921 continue;
922 }
923 /* [COMPR_ADDR, NULL_ADDR or NEW_ADDR, valid_blkaddr] */
924 if (cluster_end) {
925 reason = "[C|N|N|V]";
926 goto out;
927 }
928 }
929
930 f2fs_bug_on(F2FS_I_SB(dn->inode), count != cluster_size &&
931 !is_inode_flag_set(dn->inode, FI_COMPRESS_RELEASED));
932
933 return false;
934out:
935 f2fs_warn(sbi, "access invalid cluster, ino:%lu, nid:%u, ofs_in_node:%u, reason:%s",
936 dn->inode->i_ino, dn->nid, dn->ofs_in_node, reason);
937 set_sbi_flag(sbi, SBI_NEED_FSCK);
938 return true;
939#else
940 return false;
941#endif
942}
943
944static int __f2fs_get_cluster_blocks(struct inode *inode,
945 struct dnode_of_data *dn)
946{
947 unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
948 int count, i;
949
950 for (i = 0, count = 0; i < cluster_size; i++) {
951 block_t blkaddr = data_blkaddr(dn->inode, dn->node_page,
952 dn->ofs_in_node + i);
953
954 if (__is_valid_data_blkaddr(blkaddr))
955 count++;
956 }
957
958 return count;
959}
960
961static int __f2fs_cluster_blocks(struct inode *inode, unsigned int cluster_idx,
962 enum cluster_check_type type)
963{
964 struct dnode_of_data dn;
965 unsigned int start_idx = cluster_idx <<
966 F2FS_I(inode)->i_log_cluster_size;
967 int ret;
968
969 set_new_dnode(&dn, inode, NULL, NULL, 0);
970 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
971 if (ret) {
972 if (ret == -ENOENT)
973 ret = 0;
974 goto fail;
975 }
976
977 if (f2fs_sanity_check_cluster(&dn)) {
978 ret = -EFSCORRUPTED;
979 goto fail;
980 }
981
982 if (dn.data_blkaddr == COMPRESS_ADDR) {
983 if (type == CLUSTER_COMPR_BLKS)
984 ret = 1 + __f2fs_get_cluster_blocks(inode, &dn);
985 else if (type == CLUSTER_IS_COMPR)
986 ret = 1;
987 } else if (type == CLUSTER_RAW_BLKS) {
988 ret = __f2fs_get_cluster_blocks(inode, &dn);
989 }
990fail:
991 f2fs_put_dnode(&dn);
992 return ret;
993}
994
995/* return # of compressed blocks in compressed cluster */
996static int f2fs_compressed_blocks(struct compress_ctx *cc)
997{
998 return __f2fs_cluster_blocks(cc->inode, cc->cluster_idx,
999 CLUSTER_COMPR_BLKS);
1000}
1001
1002/* return # of raw blocks in non-compressed cluster */
1003static int f2fs_decompressed_blocks(struct inode *inode,
1004 unsigned int cluster_idx)
1005{
1006 return __f2fs_cluster_blocks(inode, cluster_idx,
1007 CLUSTER_RAW_BLKS);
1008}
1009
1010/* return whether cluster is compressed one or not */
1011int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index)
1012{
1013 return __f2fs_cluster_blocks(inode,
1014 index >> F2FS_I(inode)->i_log_cluster_size,
1015 CLUSTER_IS_COMPR);
1016}
1017
1018/* return whether cluster contains non raw blocks or not */
1019bool f2fs_is_sparse_cluster(struct inode *inode, pgoff_t index)
1020{
1021 unsigned int cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size;
1022
1023 return f2fs_decompressed_blocks(inode, cluster_idx) !=
1024 F2FS_I(inode)->i_cluster_size;
1025}
1026
1027static bool cluster_may_compress(struct compress_ctx *cc)
1028{
1029 if (!f2fs_need_compress_data(cc->inode))
1030 return false;
1031 if (f2fs_is_atomic_file(cc->inode))
1032 return false;
1033 if (!f2fs_cluster_is_full(cc))
1034 return false;
1035 if (unlikely(f2fs_cp_error(F2FS_I_SB(cc->inode))))
1036 return false;
1037 return !cluster_has_invalid_data(cc);
1038}
1039
1040static void set_cluster_writeback(struct compress_ctx *cc)
1041{
1042 int i;
1043
1044 for (i = 0; i < cc->cluster_size; i++) {
1045 if (cc->rpages[i])
1046 set_page_writeback(cc->rpages[i]);
1047 }
1048}
1049
1050static void cancel_cluster_writeback(struct compress_ctx *cc,
1051 struct compress_io_ctx *cic, int submitted)
1052{
1053 int i;
1054
1055 /* Wait for submitted IOs. */
1056 if (submitted > 1) {
1057 f2fs_submit_merged_write(F2FS_I_SB(cc->inode), DATA);
1058 while (atomic_read(&cic->pending_pages) !=
1059 (cc->valid_nr_cpages - submitted + 1))
1060 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1061 }
1062
1063 /* Cancel writeback and stay locked. */
1064 for (i = 0; i < cc->cluster_size; i++) {
1065 if (i < submitted) {
1066 inode_inc_dirty_pages(cc->inode);
1067 lock_page(cc->rpages[i]);
1068 }
1069 clear_page_private_gcing(cc->rpages[i]);
1070 if (folio_test_writeback(page_folio(cc->rpages[i])))
1071 end_page_writeback(cc->rpages[i]);
1072 }
1073}
1074
1075static void set_cluster_dirty(struct compress_ctx *cc)
1076{
1077 int i;
1078
1079 for (i = 0; i < cc->cluster_size; i++)
1080 if (cc->rpages[i]) {
1081 set_page_dirty(cc->rpages[i]);
1082 set_page_private_gcing(cc->rpages[i]);
1083 }
1084}
1085
1086static int prepare_compress_overwrite(struct compress_ctx *cc,
1087 struct page **pagep, pgoff_t index, void **fsdata)
1088{
1089 struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
1090 struct address_space *mapping = cc->inode->i_mapping;
1091 struct page *page;
1092 sector_t last_block_in_bio;
1093 fgf_t fgp_flag = FGP_LOCK | FGP_WRITE | FGP_CREAT;
1094 pgoff_t start_idx = start_idx_of_cluster(cc);
1095 int i, ret;
1096
1097retry:
1098 ret = f2fs_is_compressed_cluster(cc->inode, start_idx);
1099 if (ret <= 0)
1100 return ret;
1101
1102 ret = f2fs_init_compress_ctx(cc);
1103 if (ret)
1104 return ret;
1105
1106 /* keep page reference to avoid page reclaim */
1107 for (i = 0; i < cc->cluster_size; i++) {
1108 page = f2fs_pagecache_get_page(mapping, start_idx + i,
1109 fgp_flag, GFP_NOFS);
1110 if (!page) {
1111 ret = -ENOMEM;
1112 goto unlock_pages;
1113 }
1114
1115 if (PageUptodate(page))
1116 f2fs_put_page(page, 1);
1117 else
1118 f2fs_compress_ctx_add_page(cc, page_folio(page));
1119 }
1120
1121 if (!f2fs_cluster_is_empty(cc)) {
1122 struct bio *bio = NULL;
1123
1124 ret = f2fs_read_multi_pages(cc, &bio, cc->cluster_size,
1125 &last_block_in_bio, NULL, true);
1126 f2fs_put_rpages(cc);
1127 f2fs_destroy_compress_ctx(cc, true);
1128 if (ret)
1129 goto out;
1130 if (bio)
1131 f2fs_submit_read_bio(sbi, bio, DATA);
1132
1133 ret = f2fs_init_compress_ctx(cc);
1134 if (ret)
1135 goto out;
1136 }
1137
1138 for (i = 0; i < cc->cluster_size; i++) {
1139 f2fs_bug_on(sbi, cc->rpages[i]);
1140
1141 page = find_lock_page(mapping, start_idx + i);
1142 if (!page) {
1143 /* page can be truncated */
1144 goto release_and_retry;
1145 }
1146
1147 f2fs_wait_on_page_writeback(page, DATA, true, true);
1148 f2fs_compress_ctx_add_page(cc, page_folio(page));
1149
1150 if (!PageUptodate(page)) {
1151release_and_retry:
1152 f2fs_put_rpages(cc);
1153 f2fs_unlock_rpages(cc, i + 1);
1154 f2fs_destroy_compress_ctx(cc, true);
1155 goto retry;
1156 }
1157 }
1158
1159 if (likely(!ret)) {
1160 *fsdata = cc->rpages;
1161 *pagep = cc->rpages[offset_in_cluster(cc, index)];
1162 return cc->cluster_size;
1163 }
1164
1165unlock_pages:
1166 f2fs_put_rpages(cc);
1167 f2fs_unlock_rpages(cc, i);
1168 f2fs_destroy_compress_ctx(cc, true);
1169out:
1170 return ret;
1171}
1172
1173int f2fs_prepare_compress_overwrite(struct inode *inode,
1174 struct page **pagep, pgoff_t index, void **fsdata)
1175{
1176 struct compress_ctx cc = {
1177 .inode = inode,
1178 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1179 .cluster_size = F2FS_I(inode)->i_cluster_size,
1180 .cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
1181 .rpages = NULL,
1182 .nr_rpages = 0,
1183 };
1184
1185 return prepare_compress_overwrite(&cc, pagep, index, fsdata);
1186}
1187
1188bool f2fs_compress_write_end(struct inode *inode, void *fsdata,
1189 pgoff_t index, unsigned copied)
1190
1191{
1192 struct compress_ctx cc = {
1193 .inode = inode,
1194 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1195 .cluster_size = F2FS_I(inode)->i_cluster_size,
1196 .rpages = fsdata,
1197 };
1198 bool first_index = (index == cc.rpages[0]->index);
1199
1200 if (copied)
1201 set_cluster_dirty(&cc);
1202
1203 f2fs_put_rpages_wbc(&cc, NULL, false, 1);
1204 f2fs_destroy_compress_ctx(&cc, false);
1205
1206 return first_index;
1207}
1208
1209int f2fs_truncate_partial_cluster(struct inode *inode, u64 from, bool lock)
1210{
1211 void *fsdata = NULL;
1212 struct page *pagep;
1213 int log_cluster_size = F2FS_I(inode)->i_log_cluster_size;
1214 pgoff_t start_idx = from >> (PAGE_SHIFT + log_cluster_size) <<
1215 log_cluster_size;
1216 int err;
1217
1218 err = f2fs_is_compressed_cluster(inode, start_idx);
1219 if (err < 0)
1220 return err;
1221
1222 /* truncate normal cluster */
1223 if (!err)
1224 return f2fs_do_truncate_blocks(inode, from, lock);
1225
1226 /* truncate compressed cluster */
1227 err = f2fs_prepare_compress_overwrite(inode, &pagep,
1228 start_idx, &fsdata);
1229
1230 /* should not be a normal cluster */
1231 f2fs_bug_on(F2FS_I_SB(inode), err == 0);
1232
1233 if (err <= 0)
1234 return err;
1235
1236 if (err > 0) {
1237 struct page **rpages = fsdata;
1238 int cluster_size = F2FS_I(inode)->i_cluster_size;
1239 int i;
1240
1241 for (i = cluster_size - 1; i >= 0; i--) {
1242 loff_t start = rpages[i]->index << PAGE_SHIFT;
1243
1244 if (from <= start) {
1245 zero_user_segment(rpages[i], 0, PAGE_SIZE);
1246 } else {
1247 zero_user_segment(rpages[i], from - start,
1248 PAGE_SIZE);
1249 break;
1250 }
1251 }
1252
1253 f2fs_compress_write_end(inode, fsdata, start_idx, true);
1254 }
1255 return 0;
1256}
1257
1258static int f2fs_write_compressed_pages(struct compress_ctx *cc,
1259 int *submitted,
1260 struct writeback_control *wbc,
1261 enum iostat_type io_type)
1262{
1263 struct inode *inode = cc->inode;
1264 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1265 struct f2fs_inode_info *fi = F2FS_I(inode);
1266 struct f2fs_io_info fio = {
1267 .sbi = sbi,
1268 .ino = cc->inode->i_ino,
1269 .type = DATA,
1270 .op = REQ_OP_WRITE,
1271 .op_flags = wbc_to_write_flags(wbc),
1272 .old_blkaddr = NEW_ADDR,
1273 .page = NULL,
1274 .encrypted_page = NULL,
1275 .compressed_page = NULL,
1276 .io_type = io_type,
1277 .io_wbc = wbc,
1278 .encrypted = fscrypt_inode_uses_fs_layer_crypto(cc->inode) ?
1279 1 : 0,
1280 };
1281 struct dnode_of_data dn;
1282 struct node_info ni;
1283 struct compress_io_ctx *cic;
1284 pgoff_t start_idx = start_idx_of_cluster(cc);
1285 unsigned int last_index = cc->cluster_size - 1;
1286 loff_t psize;
1287 int i, err;
1288 bool quota_inode = IS_NOQUOTA(inode);
1289
1290 /* we should bypass data pages to proceed the kworker jobs */
1291 if (unlikely(f2fs_cp_error(sbi))) {
1292 mapping_set_error(cc->rpages[0]->mapping, -EIO);
1293 goto out_free;
1294 }
1295
1296 if (quota_inode) {
1297 /*
1298 * We need to wait for node_write to avoid block allocation during
1299 * checkpoint. This can only happen to quota writes which can cause
1300 * the below discard race condition.
1301 */
1302 f2fs_down_read(&sbi->node_write);
1303 } else if (!f2fs_trylock_op(sbi)) {
1304 goto out_free;
1305 }
1306
1307 set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1308
1309 err = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
1310 if (err)
1311 goto out_unlock_op;
1312
1313 for (i = 0; i < cc->cluster_size; i++) {
1314 if (data_blkaddr(dn.inode, dn.node_page,
1315 dn.ofs_in_node + i) == NULL_ADDR)
1316 goto out_put_dnode;
1317 }
1318
1319 psize = (loff_t)(cc->rpages[last_index]->index + 1) << PAGE_SHIFT;
1320
1321 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false);
1322 if (err)
1323 goto out_put_dnode;
1324
1325 fio.version = ni.version;
1326
1327 cic = f2fs_kmem_cache_alloc(cic_entry_slab, GFP_F2FS_ZERO, false, sbi);
1328 if (!cic)
1329 goto out_put_dnode;
1330
1331 cic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1332 cic->inode = inode;
1333 atomic_set(&cic->pending_pages, cc->valid_nr_cpages);
1334 cic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1335 if (!cic->rpages)
1336 goto out_put_cic;
1337
1338 cic->nr_rpages = cc->cluster_size;
1339
1340 for (i = 0; i < cc->valid_nr_cpages; i++) {
1341 f2fs_set_compressed_page(cc->cpages[i], inode,
1342 cc->rpages[i + 1]->index, cic);
1343 fio.compressed_page = cc->cpages[i];
1344
1345 fio.old_blkaddr = data_blkaddr(dn.inode, dn.node_page,
1346 dn.ofs_in_node + i + 1);
1347
1348 /* wait for GCed page writeback via META_MAPPING */
1349 f2fs_wait_on_block_writeback(inode, fio.old_blkaddr);
1350
1351 if (fio.encrypted) {
1352 fio.page = cc->rpages[i + 1];
1353 err = f2fs_encrypt_one_page(&fio);
1354 if (err)
1355 goto out_destroy_crypt;
1356 cc->cpages[i] = fio.encrypted_page;
1357 }
1358 }
1359
1360 set_cluster_writeback(cc);
1361
1362 for (i = 0; i < cc->cluster_size; i++)
1363 cic->rpages[i] = cc->rpages[i];
1364
1365 for (i = 0; i < cc->cluster_size; i++, dn.ofs_in_node++) {
1366 block_t blkaddr;
1367
1368 blkaddr = f2fs_data_blkaddr(&dn);
1369 fio.page = cc->rpages[i];
1370 fio.old_blkaddr = blkaddr;
1371
1372 /* cluster header */
1373 if (i == 0) {
1374 if (blkaddr == COMPRESS_ADDR)
1375 fio.compr_blocks++;
1376 if (__is_valid_data_blkaddr(blkaddr))
1377 f2fs_invalidate_blocks(sbi, blkaddr);
1378 f2fs_update_data_blkaddr(&dn, COMPRESS_ADDR);
1379 goto unlock_continue;
1380 }
1381
1382 if (fio.compr_blocks && __is_valid_data_blkaddr(blkaddr))
1383 fio.compr_blocks++;
1384
1385 if (i > cc->valid_nr_cpages) {
1386 if (__is_valid_data_blkaddr(blkaddr)) {
1387 f2fs_invalidate_blocks(sbi, blkaddr);
1388 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
1389 }
1390 goto unlock_continue;
1391 }
1392
1393 f2fs_bug_on(fio.sbi, blkaddr == NULL_ADDR);
1394
1395 if (fio.encrypted)
1396 fio.encrypted_page = cc->cpages[i - 1];
1397 else
1398 fio.compressed_page = cc->cpages[i - 1];
1399
1400 cc->cpages[i - 1] = NULL;
1401 fio.submitted = 0;
1402 f2fs_outplace_write_data(&dn, &fio);
1403 if (unlikely(!fio.submitted)) {
1404 cancel_cluster_writeback(cc, cic, i);
1405
1406 /* To call fscrypt_finalize_bounce_page */
1407 i = cc->valid_nr_cpages;
1408 *submitted = 0;
1409 goto out_destroy_crypt;
1410 }
1411 (*submitted)++;
1412unlock_continue:
1413 inode_dec_dirty_pages(cc->inode);
1414 unlock_page(fio.page);
1415 }
1416
1417 if (fio.compr_blocks)
1418 f2fs_i_compr_blocks_update(inode, fio.compr_blocks - 1, false);
1419 f2fs_i_compr_blocks_update(inode, cc->valid_nr_cpages, true);
1420 add_compr_block_stat(inode, cc->valid_nr_cpages);
1421
1422 set_inode_flag(cc->inode, FI_APPEND_WRITE);
1423
1424 f2fs_put_dnode(&dn);
1425 if (quota_inode)
1426 f2fs_up_read(&sbi->node_write);
1427 else
1428 f2fs_unlock_op(sbi);
1429
1430 spin_lock(&fi->i_size_lock);
1431 if (fi->last_disk_size < psize)
1432 fi->last_disk_size = psize;
1433 spin_unlock(&fi->i_size_lock);
1434
1435 f2fs_put_rpages(cc);
1436 page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1437 cc->cpages = NULL;
1438 f2fs_destroy_compress_ctx(cc, false);
1439 return 0;
1440
1441out_destroy_crypt:
1442 page_array_free(cc->inode, cic->rpages, cc->cluster_size);
1443
1444 for (--i; i >= 0; i--) {
1445 if (!cc->cpages[i])
1446 continue;
1447 fscrypt_finalize_bounce_page(&cc->cpages[i]);
1448 }
1449out_put_cic:
1450 kmem_cache_free(cic_entry_slab, cic);
1451out_put_dnode:
1452 f2fs_put_dnode(&dn);
1453out_unlock_op:
1454 if (quota_inode)
1455 f2fs_up_read(&sbi->node_write);
1456 else
1457 f2fs_unlock_op(sbi);
1458out_free:
1459 for (i = 0; i < cc->valid_nr_cpages; i++) {
1460 f2fs_compress_free_page(cc->cpages[i]);
1461 cc->cpages[i] = NULL;
1462 }
1463 page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1464 cc->cpages = NULL;
1465 return -EAGAIN;
1466}
1467
1468void f2fs_compress_write_end_io(struct bio *bio, struct page *page)
1469{
1470 struct f2fs_sb_info *sbi = bio->bi_private;
1471 struct compress_io_ctx *cic =
1472 (struct compress_io_ctx *)page_private(page);
1473 enum count_type type = WB_DATA_TYPE(page,
1474 f2fs_is_compressed_page(page));
1475 int i;
1476
1477 if (unlikely(bio->bi_status))
1478 mapping_set_error(cic->inode->i_mapping, -EIO);
1479
1480 f2fs_compress_free_page(page);
1481
1482 dec_page_count(sbi, type);
1483
1484 if (atomic_dec_return(&cic->pending_pages))
1485 return;
1486
1487 for (i = 0; i < cic->nr_rpages; i++) {
1488 WARN_ON(!cic->rpages[i]);
1489 clear_page_private_gcing(cic->rpages[i]);
1490 end_page_writeback(cic->rpages[i]);
1491 }
1492
1493 page_array_free(cic->inode, cic->rpages, cic->nr_rpages);
1494 kmem_cache_free(cic_entry_slab, cic);
1495}
1496
1497static int f2fs_write_raw_pages(struct compress_ctx *cc,
1498 int *submitted_p,
1499 struct writeback_control *wbc,
1500 enum iostat_type io_type)
1501{
1502 struct address_space *mapping = cc->inode->i_mapping;
1503 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1504 int submitted, compr_blocks, i;
1505 int ret = 0;
1506
1507 compr_blocks = f2fs_compressed_blocks(cc);
1508
1509 for (i = 0; i < cc->cluster_size; i++) {
1510 if (!cc->rpages[i])
1511 continue;
1512
1513 redirty_page_for_writepage(wbc, cc->rpages[i]);
1514 unlock_page(cc->rpages[i]);
1515 }
1516
1517 if (compr_blocks < 0)
1518 return compr_blocks;
1519
1520 /* overwrite compressed cluster w/ normal cluster */
1521 if (compr_blocks > 0)
1522 f2fs_lock_op(sbi);
1523
1524 for (i = 0; i < cc->cluster_size; i++) {
1525 if (!cc->rpages[i])
1526 continue;
1527retry_write:
1528 lock_page(cc->rpages[i]);
1529
1530 if (cc->rpages[i]->mapping != mapping) {
1531continue_unlock:
1532 unlock_page(cc->rpages[i]);
1533 continue;
1534 }
1535
1536 if (!PageDirty(cc->rpages[i]))
1537 goto continue_unlock;
1538
1539 if (folio_test_writeback(page_folio(cc->rpages[i]))) {
1540 if (wbc->sync_mode == WB_SYNC_NONE)
1541 goto continue_unlock;
1542 f2fs_wait_on_page_writeback(cc->rpages[i], DATA, true, true);
1543 }
1544
1545 if (!clear_page_dirty_for_io(cc->rpages[i]))
1546 goto continue_unlock;
1547
1548 ret = f2fs_write_single_data_page(page_folio(cc->rpages[i]),
1549 &submitted,
1550 NULL, NULL, wbc, io_type,
1551 compr_blocks, false);
1552 if (ret) {
1553 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1554 unlock_page(cc->rpages[i]);
1555 ret = 0;
1556 } else if (ret == -EAGAIN) {
1557 ret = 0;
1558 /*
1559 * for quota file, just redirty left pages to
1560 * avoid deadlock caused by cluster update race
1561 * from foreground operation.
1562 */
1563 if (IS_NOQUOTA(cc->inode))
1564 goto out;
1565 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1566 goto retry_write;
1567 }
1568 goto out;
1569 }
1570
1571 *submitted_p += submitted;
1572 }
1573
1574out:
1575 if (compr_blocks > 0)
1576 f2fs_unlock_op(sbi);
1577
1578 f2fs_balance_fs(sbi, true);
1579 return ret;
1580}
1581
1582int f2fs_write_multi_pages(struct compress_ctx *cc,
1583 int *submitted,
1584 struct writeback_control *wbc,
1585 enum iostat_type io_type)
1586{
1587 int err;
1588
1589 *submitted = 0;
1590 if (cluster_may_compress(cc)) {
1591 err = f2fs_compress_pages(cc);
1592 if (err == -EAGAIN) {
1593 add_compr_block_stat(cc->inode, cc->cluster_size);
1594 goto write;
1595 } else if (err) {
1596 f2fs_put_rpages_wbc(cc, wbc, true, 1);
1597 goto destroy_out;
1598 }
1599
1600 err = f2fs_write_compressed_pages(cc, submitted,
1601 wbc, io_type);
1602 if (!err)
1603 return 0;
1604 f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN);
1605 }
1606write:
1607 f2fs_bug_on(F2FS_I_SB(cc->inode), *submitted);
1608
1609 err = f2fs_write_raw_pages(cc, submitted, wbc, io_type);
1610 f2fs_put_rpages_wbc(cc, wbc, false, 0);
1611destroy_out:
1612 f2fs_destroy_compress_ctx(cc, false);
1613 return err;
1614}
1615
1616static inline bool allow_memalloc_for_decomp(struct f2fs_sb_info *sbi,
1617 bool pre_alloc)
1618{
1619 return pre_alloc ^ f2fs_low_mem_mode(sbi);
1620}
1621
1622static int f2fs_prepare_decomp_mem(struct decompress_io_ctx *dic,
1623 bool pre_alloc)
1624{
1625 const struct f2fs_compress_ops *cops =
1626 f2fs_cops[F2FS_I(dic->inode)->i_compress_algorithm];
1627 int i;
1628
1629 if (!allow_memalloc_for_decomp(F2FS_I_SB(dic->inode), pre_alloc))
1630 return 0;
1631
1632 dic->tpages = page_array_alloc(dic->inode, dic->cluster_size);
1633 if (!dic->tpages)
1634 return -ENOMEM;
1635
1636 for (i = 0; i < dic->cluster_size; i++) {
1637 if (dic->rpages[i]) {
1638 dic->tpages[i] = dic->rpages[i];
1639 continue;
1640 }
1641
1642 dic->tpages[i] = f2fs_compress_alloc_page();
1643 }
1644
1645 dic->rbuf = f2fs_vmap(dic->tpages, dic->cluster_size);
1646 if (!dic->rbuf)
1647 return -ENOMEM;
1648
1649 dic->cbuf = f2fs_vmap(dic->cpages, dic->nr_cpages);
1650 if (!dic->cbuf)
1651 return -ENOMEM;
1652
1653 if (cops->init_decompress_ctx)
1654 return cops->init_decompress_ctx(dic);
1655
1656 return 0;
1657}
1658
1659static void f2fs_release_decomp_mem(struct decompress_io_ctx *dic,
1660 bool bypass_destroy_callback, bool pre_alloc)
1661{
1662 const struct f2fs_compress_ops *cops =
1663 f2fs_cops[F2FS_I(dic->inode)->i_compress_algorithm];
1664
1665 if (!allow_memalloc_for_decomp(F2FS_I_SB(dic->inode), pre_alloc))
1666 return;
1667
1668 if (!bypass_destroy_callback && cops->destroy_decompress_ctx)
1669 cops->destroy_decompress_ctx(dic);
1670
1671 if (dic->cbuf)
1672 vm_unmap_ram(dic->cbuf, dic->nr_cpages);
1673
1674 if (dic->rbuf)
1675 vm_unmap_ram(dic->rbuf, dic->cluster_size);
1676}
1677
1678static void f2fs_free_dic(struct decompress_io_ctx *dic,
1679 bool bypass_destroy_callback);
1680
1681struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
1682{
1683 struct decompress_io_ctx *dic;
1684 pgoff_t start_idx = start_idx_of_cluster(cc);
1685 struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
1686 int i, ret;
1687
1688 dic = f2fs_kmem_cache_alloc(dic_entry_slab, GFP_F2FS_ZERO, false, sbi);
1689 if (!dic)
1690 return ERR_PTR(-ENOMEM);
1691
1692 dic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1693 if (!dic->rpages) {
1694 kmem_cache_free(dic_entry_slab, dic);
1695 return ERR_PTR(-ENOMEM);
1696 }
1697
1698 dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1699 dic->inode = cc->inode;
1700 atomic_set(&dic->remaining_pages, cc->nr_cpages);
1701 dic->cluster_idx = cc->cluster_idx;
1702 dic->cluster_size = cc->cluster_size;
1703 dic->log_cluster_size = cc->log_cluster_size;
1704 dic->nr_cpages = cc->nr_cpages;
1705 refcount_set(&dic->refcnt, 1);
1706 dic->failed = false;
1707 dic->need_verity = f2fs_need_verity(cc->inode, start_idx);
1708
1709 for (i = 0; i < dic->cluster_size; i++)
1710 dic->rpages[i] = cc->rpages[i];
1711 dic->nr_rpages = cc->cluster_size;
1712
1713 dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages);
1714 if (!dic->cpages) {
1715 ret = -ENOMEM;
1716 goto out_free;
1717 }
1718
1719 for (i = 0; i < dic->nr_cpages; i++) {
1720 struct page *page;
1721
1722 page = f2fs_compress_alloc_page();
1723 f2fs_set_compressed_page(page, cc->inode,
1724 start_idx + i + 1, dic);
1725 dic->cpages[i] = page;
1726 }
1727
1728 ret = f2fs_prepare_decomp_mem(dic, true);
1729 if (ret)
1730 goto out_free;
1731
1732 return dic;
1733
1734out_free:
1735 f2fs_free_dic(dic, true);
1736 return ERR_PTR(ret);
1737}
1738
1739static void f2fs_free_dic(struct decompress_io_ctx *dic,
1740 bool bypass_destroy_callback)
1741{
1742 int i;
1743
1744 f2fs_release_decomp_mem(dic, bypass_destroy_callback, true);
1745
1746 if (dic->tpages) {
1747 for (i = 0; i < dic->cluster_size; i++) {
1748 if (dic->rpages[i])
1749 continue;
1750 if (!dic->tpages[i])
1751 continue;
1752 f2fs_compress_free_page(dic->tpages[i]);
1753 }
1754 page_array_free(dic->inode, dic->tpages, dic->cluster_size);
1755 }
1756
1757 if (dic->cpages) {
1758 for (i = 0; i < dic->nr_cpages; i++) {
1759 if (!dic->cpages[i])
1760 continue;
1761 f2fs_compress_free_page(dic->cpages[i]);
1762 }
1763 page_array_free(dic->inode, dic->cpages, dic->nr_cpages);
1764 }
1765
1766 page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
1767 kmem_cache_free(dic_entry_slab, dic);
1768}
1769
1770static void f2fs_late_free_dic(struct work_struct *work)
1771{
1772 struct decompress_io_ctx *dic =
1773 container_of(work, struct decompress_io_ctx, free_work);
1774
1775 f2fs_free_dic(dic, false);
1776}
1777
1778static void f2fs_put_dic(struct decompress_io_ctx *dic, bool in_task)
1779{
1780 if (refcount_dec_and_test(&dic->refcnt)) {
1781 if (in_task) {
1782 f2fs_free_dic(dic, false);
1783 } else {
1784 INIT_WORK(&dic->free_work, f2fs_late_free_dic);
1785 queue_work(F2FS_I_SB(dic->inode)->post_read_wq,
1786 &dic->free_work);
1787 }
1788 }
1789}
1790
1791static void f2fs_verify_cluster(struct work_struct *work)
1792{
1793 struct decompress_io_ctx *dic =
1794 container_of(work, struct decompress_io_ctx, verity_work);
1795 int i;
1796
1797 /* Verify, update, and unlock the decompressed pages. */
1798 for (i = 0; i < dic->cluster_size; i++) {
1799 struct page *rpage = dic->rpages[i];
1800
1801 if (!rpage)
1802 continue;
1803
1804 if (fsverity_verify_page(rpage))
1805 SetPageUptodate(rpage);
1806 else
1807 ClearPageUptodate(rpage);
1808 unlock_page(rpage);
1809 }
1810
1811 f2fs_put_dic(dic, true);
1812}
1813
1814/*
1815 * This is called when a compressed cluster has been decompressed
1816 * (or failed to be read and/or decompressed).
1817 */
1818void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed,
1819 bool in_task)
1820{
1821 int i;
1822
1823 if (!failed && dic->need_verity) {
1824 /*
1825 * Note that to avoid deadlocks, the verity work can't be done
1826 * on the decompression workqueue. This is because verifying
1827 * the data pages can involve reading metadata pages from the
1828 * file, and these metadata pages may be compressed.
1829 */
1830 INIT_WORK(&dic->verity_work, f2fs_verify_cluster);
1831 fsverity_enqueue_verify_work(&dic->verity_work);
1832 return;
1833 }
1834
1835 /* Update and unlock the cluster's pagecache pages. */
1836 for (i = 0; i < dic->cluster_size; i++) {
1837 struct page *rpage = dic->rpages[i];
1838
1839 if (!rpage)
1840 continue;
1841
1842 if (failed)
1843 ClearPageUptodate(rpage);
1844 else
1845 SetPageUptodate(rpage);
1846 unlock_page(rpage);
1847 }
1848
1849 /*
1850 * Release the reference to the decompress_io_ctx that was being held
1851 * for I/O completion.
1852 */
1853 f2fs_put_dic(dic, in_task);
1854}
1855
1856/*
1857 * Put a reference to a compressed page's decompress_io_ctx.
1858 *
1859 * This is called when the page is no longer needed and can be freed.
1860 */
1861void f2fs_put_page_dic(struct page *page, bool in_task)
1862{
1863 struct decompress_io_ctx *dic =
1864 (struct decompress_io_ctx *)page_private(page);
1865
1866 f2fs_put_dic(dic, in_task);
1867}
1868
1869/*
1870 * check whether cluster blocks are contiguous, and add extent cache entry
1871 * only if cluster blocks are logically and physically contiguous.
1872 */
1873unsigned int f2fs_cluster_blocks_are_contiguous(struct dnode_of_data *dn,
1874 unsigned int ofs_in_node)
1875{
1876 bool compressed = data_blkaddr(dn->inode, dn->node_page,
1877 ofs_in_node) == COMPRESS_ADDR;
1878 int i = compressed ? 1 : 0;
1879 block_t first_blkaddr = data_blkaddr(dn->inode, dn->node_page,
1880 ofs_in_node + i);
1881
1882 for (i += 1; i < F2FS_I(dn->inode)->i_cluster_size; i++) {
1883 block_t blkaddr = data_blkaddr(dn->inode, dn->node_page,
1884 ofs_in_node + i);
1885
1886 if (!__is_valid_data_blkaddr(blkaddr))
1887 break;
1888 if (first_blkaddr + i - (compressed ? 1 : 0) != blkaddr)
1889 return 0;
1890 }
1891
1892 return compressed ? i - 1 : i;
1893}
1894
1895const struct address_space_operations f2fs_compress_aops = {
1896 .release_folio = f2fs_release_folio,
1897 .invalidate_folio = f2fs_invalidate_folio,
1898 .migrate_folio = filemap_migrate_folio,
1899};
1900
1901struct address_space *COMPRESS_MAPPING(struct f2fs_sb_info *sbi)
1902{
1903 return sbi->compress_inode->i_mapping;
1904}
1905
1906void f2fs_invalidate_compress_page(struct f2fs_sb_info *sbi, block_t blkaddr)
1907{
1908 if (!sbi->compress_inode)
1909 return;
1910 invalidate_mapping_pages(COMPRESS_MAPPING(sbi), blkaddr, blkaddr);
1911}
1912
1913void f2fs_cache_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
1914 nid_t ino, block_t blkaddr)
1915{
1916 struct page *cpage;
1917 int ret;
1918
1919 if (!test_opt(sbi, COMPRESS_CACHE))
1920 return;
1921
1922 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
1923 return;
1924
1925 if (!f2fs_available_free_memory(sbi, COMPRESS_PAGE))
1926 return;
1927
1928 cpage = find_get_page(COMPRESS_MAPPING(sbi), blkaddr);
1929 if (cpage) {
1930 f2fs_put_page(cpage, 0);
1931 return;
1932 }
1933
1934 cpage = alloc_page(__GFP_NOWARN | __GFP_IO);
1935 if (!cpage)
1936 return;
1937
1938 ret = add_to_page_cache_lru(cpage, COMPRESS_MAPPING(sbi),
1939 blkaddr, GFP_NOFS);
1940 if (ret) {
1941 f2fs_put_page(cpage, 0);
1942 return;
1943 }
1944
1945 set_page_private_data(cpage, ino);
1946
1947 memcpy(page_address(cpage), page_address(page), PAGE_SIZE);
1948 SetPageUptodate(cpage);
1949 f2fs_put_page(cpage, 1);
1950}
1951
1952bool f2fs_load_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
1953 block_t blkaddr)
1954{
1955 struct page *cpage;
1956 bool hitted = false;
1957
1958 if (!test_opt(sbi, COMPRESS_CACHE))
1959 return false;
1960
1961 cpage = f2fs_pagecache_get_page(COMPRESS_MAPPING(sbi),
1962 blkaddr, FGP_LOCK | FGP_NOWAIT, GFP_NOFS);
1963 if (cpage) {
1964 if (PageUptodate(cpage)) {
1965 atomic_inc(&sbi->compress_page_hit);
1966 memcpy(page_address(page),
1967 page_address(cpage), PAGE_SIZE);
1968 hitted = true;
1969 }
1970 f2fs_put_page(cpage, 1);
1971 }
1972
1973 return hitted;
1974}
1975
1976void f2fs_invalidate_compress_pages(struct f2fs_sb_info *sbi, nid_t ino)
1977{
1978 struct address_space *mapping = COMPRESS_MAPPING(sbi);
1979 struct folio_batch fbatch;
1980 pgoff_t index = 0;
1981 pgoff_t end = MAX_BLKADDR(sbi);
1982
1983 if (!mapping->nrpages)
1984 return;
1985
1986 folio_batch_init(&fbatch);
1987
1988 do {
1989 unsigned int nr, i;
1990
1991 nr = filemap_get_folios(mapping, &index, end - 1, &fbatch);
1992 if (!nr)
1993 break;
1994
1995 for (i = 0; i < nr; i++) {
1996 struct folio *folio = fbatch.folios[i];
1997
1998 folio_lock(folio);
1999 if (folio->mapping != mapping) {
2000 folio_unlock(folio);
2001 continue;
2002 }
2003
2004 if (ino != get_page_private_data(&folio->page)) {
2005 folio_unlock(folio);
2006 continue;
2007 }
2008
2009 generic_error_remove_folio(mapping, folio);
2010 folio_unlock(folio);
2011 }
2012 folio_batch_release(&fbatch);
2013 cond_resched();
2014 } while (index < end);
2015}
2016
2017int f2fs_init_compress_inode(struct f2fs_sb_info *sbi)
2018{
2019 struct inode *inode;
2020
2021 if (!test_opt(sbi, COMPRESS_CACHE))
2022 return 0;
2023
2024 inode = f2fs_iget(sbi->sb, F2FS_COMPRESS_INO(sbi));
2025 if (IS_ERR(inode))
2026 return PTR_ERR(inode);
2027 sbi->compress_inode = inode;
2028
2029 sbi->compress_percent = COMPRESS_PERCENT;
2030 sbi->compress_watermark = COMPRESS_WATERMARK;
2031
2032 atomic_set(&sbi->compress_page_hit, 0);
2033
2034 return 0;
2035}
2036
2037void f2fs_destroy_compress_inode(struct f2fs_sb_info *sbi)
2038{
2039 if (!sbi->compress_inode)
2040 return;
2041 iput(sbi->compress_inode);
2042 sbi->compress_inode = NULL;
2043}
2044
2045int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi)
2046{
2047 dev_t dev = sbi->sb->s_bdev->bd_dev;
2048 char slab_name[35];
2049
2050 if (!f2fs_sb_has_compression(sbi))
2051 return 0;
2052
2053 sprintf(slab_name, "f2fs_page_array_entry-%u:%u", MAJOR(dev), MINOR(dev));
2054
2055 sbi->page_array_slab_size = sizeof(struct page *) <<
2056 F2FS_OPTION(sbi).compress_log_size;
2057
2058 sbi->page_array_slab = f2fs_kmem_cache_create(slab_name,
2059 sbi->page_array_slab_size);
2060 return sbi->page_array_slab ? 0 : -ENOMEM;
2061}
2062
2063void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi)
2064{
2065 kmem_cache_destroy(sbi->page_array_slab);
2066}
2067
2068int __init f2fs_init_compress_cache(void)
2069{
2070 cic_entry_slab = f2fs_kmem_cache_create("f2fs_cic_entry",
2071 sizeof(struct compress_io_ctx));
2072 if (!cic_entry_slab)
2073 return -ENOMEM;
2074 dic_entry_slab = f2fs_kmem_cache_create("f2fs_dic_entry",
2075 sizeof(struct decompress_io_ctx));
2076 if (!dic_entry_slab)
2077 goto free_cic;
2078 return 0;
2079free_cic:
2080 kmem_cache_destroy(cic_entry_slab);
2081 return -ENOMEM;
2082}
2083
2084void f2fs_destroy_compress_cache(void)
2085{
2086 kmem_cache_destroy(dic_entry_slab);
2087 kmem_cache_destroy(cic_entry_slab);
2088}