blob: 32da97c3c19db41fc3bd0222afcd11d60cabbc06 [file] [log] [blame]
David Sterbac1d7c512018-04-03 19:23:33 +02001// SPDX-License-Identifier: GPL-2.0
Chris Masonc8b97812008-10-29 14:49:59 -04002/*
3 * Copyright (C) 2008 Oracle. All rights reserved.
Chris Masonc8b97812008-10-29 14:49:59 -04004 */
5
6#include <linux/kernel.h>
7#include <linux/bio.h>
Chris Masonc8b97812008-10-29 14:49:59 -04008#include <linux/file.h>
9#include <linux/fs.h>
10#include <linux/pagemap.h>
11#include <linux/highmem.h>
Christoph Hellwige41d12f2021-09-20 14:33:13 +020012#include <linux/kthread.h>
Chris Masonc8b97812008-10-29 14:49:59 -040013#include <linux/time.h>
14#include <linux/init.h>
15#include <linux/string.h>
Chris Masonc8b97812008-10-29 14:49:59 -040016#include <linux/backing-dev.h>
Chris Masonc8b97812008-10-29 14:49:59 -040017#include <linux/writeback.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090018#include <linux/slab.h>
David Sterbafe308532017-05-31 17:14:56 +020019#include <linux/sched/mm.h>
Timofey Titovets19562432017-10-08 16:11:59 +030020#include <linux/log2.h>
Johannes Thumshirnd5178572019-06-03 16:58:57 +020021#include <crypto/hash.h>
David Sterba602cbe92019-08-21 18:48:25 +020022#include "misc.h"
Chris Masonc8b97812008-10-29 14:49:59 -040023#include "ctree.h"
24#include "disk-io.h"
25#include "transaction.h"
26#include "btrfs_inode.h"
27#include "volumes.h"
28#include "ordered-data.h"
Chris Masonc8b97812008-10-29 14:49:59 -040029#include "compression.h"
30#include "extent_io.h"
31#include "extent_map.h"
Qu Wenruo6a404912021-09-27 15:21:47 +080032#include "subpage.h"
Johannes Thumshirn764c7c92021-05-19 00:40:28 +090033#include "zoned.h"
Chris Masonc8b97812008-10-29 14:49:59 -040034
David Sterbae128f9c2017-10-31 17:24:26 +010035static const char* const btrfs_compress_types[] = { "", "zlib", "lzo", "zstd" };
36
37const char* btrfs_compress_type2str(enum btrfs_compression_type type)
38{
39 switch (type) {
40 case BTRFS_COMPRESS_ZLIB:
41 case BTRFS_COMPRESS_LZO:
42 case BTRFS_COMPRESS_ZSTD:
43 case BTRFS_COMPRESS_NONE:
44 return btrfs_compress_types[type];
Chengguang Xuce96b7f2019-10-10 15:59:57 +080045 default:
46 break;
David Sterbae128f9c2017-10-31 17:24:26 +010047 }
48
49 return NULL;
50}
51
Johannes Thumshirnaa53e3b2019-06-06 12:07:15 +020052bool btrfs_compress_is_valid_type(const char *str, size_t len)
53{
54 int i;
55
56 for (i = 1; i < ARRAY_SIZE(btrfs_compress_types); i++) {
57 size_t comp_len = strlen(btrfs_compress_types[i]);
58
59 if (len < comp_len)
60 continue;
61
62 if (!strncmp(btrfs_compress_types[i], str, comp_len))
63 return true;
64 }
65 return false;
66}
67
David Sterba1e4eb742019-10-02 00:06:15 +020068static int compression_compress_pages(int type, struct list_head *ws,
69 struct address_space *mapping, u64 start, struct page **pages,
70 unsigned long *out_pages, unsigned long *total_in,
71 unsigned long *total_out)
72{
73 switch (type) {
74 case BTRFS_COMPRESS_ZLIB:
75 return zlib_compress_pages(ws, mapping, start, pages,
76 out_pages, total_in, total_out);
77 case BTRFS_COMPRESS_LZO:
78 return lzo_compress_pages(ws, mapping, start, pages,
79 out_pages, total_in, total_out);
80 case BTRFS_COMPRESS_ZSTD:
81 return zstd_compress_pages(ws, mapping, start, pages,
82 out_pages, total_in, total_out);
83 case BTRFS_COMPRESS_NONE:
84 default:
85 /*
Qu Wenruo1d8ba9e2020-08-04 15:25:47 +080086 * This can happen when compression races with remount setting
87 * it to 'no compress', while caller doesn't call
88 * inode_need_compress() to check if we really need to
89 * compress.
90 *
91 * Not a big deal, just need to inform caller that we
92 * haven't allocated any pages yet.
David Sterba1e4eb742019-10-02 00:06:15 +020093 */
Qu Wenruo1d8ba9e2020-08-04 15:25:47 +080094 *out_pages = 0;
David Sterba1e4eb742019-10-02 00:06:15 +020095 return -E2BIG;
96 }
97}
98
99static int compression_decompress_bio(int type, struct list_head *ws,
100 struct compressed_bio *cb)
101{
102 switch (type) {
103 case BTRFS_COMPRESS_ZLIB: return zlib_decompress_bio(ws, cb);
104 case BTRFS_COMPRESS_LZO: return lzo_decompress_bio(ws, cb);
105 case BTRFS_COMPRESS_ZSTD: return zstd_decompress_bio(ws, cb);
106 case BTRFS_COMPRESS_NONE:
107 default:
108 /*
109 * This can't happen, the type is validated several times
110 * before we get here.
111 */
112 BUG();
113 }
114}
115
116static int compression_decompress(int type, struct list_head *ws,
117 unsigned char *data_in, struct page *dest_page,
118 unsigned long start_byte, size_t srclen, size_t destlen)
119{
120 switch (type) {
121 case BTRFS_COMPRESS_ZLIB: return zlib_decompress(ws, data_in, dest_page,
122 start_byte, srclen, destlen);
123 case BTRFS_COMPRESS_LZO: return lzo_decompress(ws, data_in, dest_page,
124 start_byte, srclen, destlen);
125 case BTRFS_COMPRESS_ZSTD: return zstd_decompress(ws, data_in, dest_page,
126 start_byte, srclen, destlen);
127 case BTRFS_COMPRESS_NONE:
128 default:
129 /*
130 * This can't happen, the type is validated several times
131 * before we get here.
132 */
133 BUG();
134 }
135}
136
Anand Jain8140dc32017-05-26 15:44:58 +0800137static int btrfs_decompress_bio(struct compressed_bio *cb);
Eric Sandeen48a3b632013-04-25 20:41:01 +0000138
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400139static inline int compressed_bio_size(struct btrfs_fs_info *fs_info,
Chris Masond20f7042008-12-08 16:58:54 -0500140 unsigned long disk_size)
141{
Chris Masond20f7042008-12-08 16:58:54 -0500142 return sizeof(struct compressed_bio) +
David Sterba713cebf2020-06-30 18:04:02 +0200143 (DIV_ROUND_UP(disk_size, fs_info->sectorsize)) * fs_info->csum_size;
Chris Masond20f7042008-12-08 16:58:54 -0500144}
145
Nikolay Borisov5a9472f2020-07-02 15:23:34 +0300146static int check_compressed_csum(struct btrfs_inode *inode, struct bio *bio,
Chris Masond20f7042008-12-08 16:58:54 -0500147 u64 disk_start)
148{
Johannes Thumshirn10fe6ca2019-05-22 10:19:02 +0200149 struct btrfs_fs_info *fs_info = inode->root->fs_info;
Johannes Thumshirnd5178572019-06-03 16:58:57 +0200150 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
David Sterba223486c2020-07-02 11:27:30 +0200151 const u32 csum_size = fs_info->csum_size;
Qu Wenruo04d4ba4c2021-02-04 15:03:24 +0800152 const u32 sectorsize = fs_info->sectorsize;
Chris Masond20f7042008-12-08 16:58:54 -0500153 struct page *page;
Anand Jain1d08ce582021-05-29 17:48:33 +0800154 unsigned int i;
Chris Masond20f7042008-12-08 16:58:54 -0500155 char *kaddr;
Johannes Thumshirnd5178572019-06-03 16:58:57 +0200156 u8 csum[BTRFS_CSUM_SIZE];
Nikolay Borisov5a9472f2020-07-02 15:23:34 +0300157 struct compressed_bio *cb = bio->bi_private;
Johannes Thumshirn10fe6ca2019-05-22 10:19:02 +0200158 u8 *cb_sum = cb->sums;
Chris Masond20f7042008-12-08 16:58:54 -0500159
Josef Bacik42437a62020-10-16 11:29:18 -0400160 if (!fs_info->csum_root || (inode->flags & BTRFS_INODE_NODATASUM))
Chris Masond20f7042008-12-08 16:58:54 -0500161 return 0;
162
Johannes Thumshirnd5178572019-06-03 16:58:57 +0200163 shash->tfm = fs_info->csum_shash;
164
Chris Masond20f7042008-12-08 16:58:54 -0500165 for (i = 0; i < cb->nr_pages; i++) {
Qu Wenruo04d4ba4c2021-02-04 15:03:24 +0800166 u32 pg_offset;
167 u32 bytes_left = PAGE_SIZE;
Chris Masond20f7042008-12-08 16:58:54 -0500168 page = cb->compressed_pages[i];
Chris Masond20f7042008-12-08 16:58:54 -0500169
Qu Wenruo04d4ba4c2021-02-04 15:03:24 +0800170 /* Determine the remaining bytes inside the page first */
171 if (i == cb->nr_pages - 1)
172 bytes_left = cb->compressed_len - i * PAGE_SIZE;
Chris Masond20f7042008-12-08 16:58:54 -0500173
Qu Wenruo04d4ba4c2021-02-04 15:03:24 +0800174 /* Hash through the page sector by sector */
175 for (pg_offset = 0; pg_offset < bytes_left;
176 pg_offset += sectorsize) {
David Sterba3a60f652021-10-27 10:39:03 +0200177 kaddr = kmap_atomic(page);
Qu Wenruo04d4ba4c2021-02-04 15:03:24 +0800178 crypto_shash_digest(shash, kaddr + pg_offset,
179 sectorsize, csum);
David Sterba3a60f652021-10-27 10:39:03 +0200180 kunmap_atomic(kaddr);
Qu Wenruo04d4ba4c2021-02-04 15:03:24 +0800181
182 if (memcmp(&csum, cb_sum, csum_size) != 0) {
183 btrfs_print_data_csum_error(inode, disk_start,
184 csum, cb_sum, cb->mirror_num);
Qu Wenruoc3a3b192021-09-15 15:17:18 +0800185 if (btrfs_bio(bio)->device)
Qu Wenruo04d4ba4c2021-02-04 15:03:24 +0800186 btrfs_dev_stat_inc_and_print(
Qu Wenruoc3a3b192021-09-15 15:17:18 +0800187 btrfs_bio(bio)->device,
Qu Wenruo04d4ba4c2021-02-04 15:03:24 +0800188 BTRFS_DEV_STAT_CORRUPTION_ERRS);
189 return -EIO;
190 }
191 cb_sum += csum_size;
192 disk_start += sectorsize;
Chris Masond20f7042008-12-08 16:58:54 -0500193 }
Chris Masond20f7042008-12-08 16:58:54 -0500194 }
Nikolay Borisov93c4c032020-07-02 16:46:47 +0300195 return 0;
Chris Masond20f7042008-12-08 16:58:54 -0500196}
197
Qu Wenruo6ec97652021-09-27 15:21:48 +0800198/*
199 * Reduce bio and io accounting for a compressed_bio with its corresponding bio.
200 *
201 * Return true if there is no pending bio nor io.
202 * Return false otherwise.
203 */
204static bool dec_and_test_compressed_bio(struct compressed_bio *cb, struct bio *bio)
205{
206 struct btrfs_fs_info *fs_info = btrfs_sb(cb->inode->i_sb);
207 unsigned int bi_size = 0;
208 bool last_io = false;
209 struct bio_vec *bvec;
210 struct bvec_iter_all iter_all;
211
212 /*
213 * At endio time, bi_iter.bi_size doesn't represent the real bio size.
214 * Thus here we have to iterate through all segments to grab correct
215 * bio size.
216 */
217 bio_for_each_segment_all(bvec, bio, iter_all)
218 bi_size += bvec->bv_len;
219
220 if (bio->bi_status)
221 cb->errors = 1;
222
223 ASSERT(bi_size && bi_size <= cb->compressed_len);
224 last_io = refcount_sub_and_test(bi_size >> fs_info->sectorsize_bits,
225 &cb->pending_sectors);
Qu Wenruo86ccbb42021-09-27 15:21:50 +0800226 /*
227 * Here we must wake up the possible error handler after all other
228 * operations on @cb finished, or we can race with
229 * finish_compressed_bio_*() which may free @cb.
230 */
231 wake_up_var(cb);
232
Qu Wenruo6ec97652021-09-27 15:21:48 +0800233 return last_io;
234}
235
Qu Wenruo86ccbb42021-09-27 15:21:50 +0800236static void finish_compressed_bio_read(struct compressed_bio *cb, struct bio *bio)
237{
238 unsigned int index;
239 struct page *page;
240
241 /* Release the compressed pages */
242 for (index = 0; index < cb->nr_pages; index++) {
243 page = cb->compressed_pages[index];
244 page->mapping = NULL;
245 put_page(page);
246 }
247
248 /* Do io completion on the original bio */
249 if (cb->errors) {
250 bio_io_error(cb->orig_bio);
251 } else {
252 struct bio_vec *bvec;
253 struct bvec_iter_all iter_all;
254
255 ASSERT(bio);
256 ASSERT(!bio->bi_status);
257 /*
258 * We have verified the checksum already, set page checked so
259 * the end_io handlers know about it
260 */
261 ASSERT(!bio_flagged(bio, BIO_CLONED));
262 bio_for_each_segment_all(bvec, cb->orig_bio, iter_all) {
263 u64 bvec_start = page_offset(bvec->bv_page) +
264 bvec->bv_offset;
265
266 btrfs_page_set_checked(btrfs_sb(cb->inode->i_sb),
267 bvec->bv_page, bvec_start,
268 bvec->bv_len);
269 }
270
271 bio_endio(cb->orig_bio);
272 }
273
274 /* Finally free the cb struct */
275 kfree(cb->compressed_pages);
276 kfree(cb);
277}
278
Chris Masonc8b97812008-10-29 14:49:59 -0400279/* when we finish reading compressed pages from the disk, we
280 * decompress them and then run the bio end_io routines on the
281 * decompressed pages (in the inode address space).
282 *
283 * This allows the checksumming and other IO error handling routines
284 * to work normally
285 *
286 * The compressed pages are freed here, and it must be run
287 * in process context
288 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200289static void end_compressed_bio_read(struct bio *bio)
Chris Masonc8b97812008-10-29 14:49:59 -0400290{
Chris Masonc8b97812008-10-29 14:49:59 -0400291 struct compressed_bio *cb = bio->bi_private;
292 struct inode *inode;
Qu Wenruoc3a3b192021-09-15 15:17:18 +0800293 unsigned int mirror = btrfs_bio(bio)->mirror_num;
Liu Boe6311f22017-09-20 17:50:19 -0600294 int ret = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400295
Qu Wenruo6ec97652021-09-27 15:21:48 +0800296 if (!dec_and_test_compressed_bio(cb, bio))
Chris Masonc8b97812008-10-29 14:49:59 -0400297 goto out;
298
Liu Bocf1167d2017-09-20 17:50:18 -0600299 /*
300 * Record the correct mirror_num in cb->orig_bio so that
301 * read-repair can work properly.
302 */
Qu Wenruoc3a3b192021-09-15 15:17:18 +0800303 btrfs_bio(cb->orig_bio)->mirror_num = mirror;
Liu Bocf1167d2017-09-20 17:50:18 -0600304 cb->mirror_num = mirror;
305
Liu Boe6311f22017-09-20 17:50:19 -0600306 /*
307 * Some IO in this cb have failed, just skip checksum as there
308 * is no way it could be correct.
309 */
310 if (cb->errors == 1)
311 goto csum_failed;
312
Chris Masond20f7042008-12-08 16:58:54 -0500313 inode = cb->inode;
Nikolay Borisov5a9472f2020-07-02 15:23:34 +0300314 ret = check_compressed_csum(BTRFS_I(inode), bio,
David Sterba1201b582020-11-26 15:41:27 +0100315 bio->bi_iter.bi_sector << 9);
Chris Masond20f7042008-12-08 16:58:54 -0500316 if (ret)
317 goto csum_failed;
318
Chris Masonc8b97812008-10-29 14:49:59 -0400319 /* ok, we're the last bio for this extent, lets start
320 * the decompression.
321 */
Anand Jain8140dc32017-05-26 15:44:58 +0800322 ret = btrfs_decompress_bio(cb);
323
Chris Masond20f7042008-12-08 16:58:54 -0500324csum_failed:
Chris Masonc8b97812008-10-29 14:49:59 -0400325 if (ret)
326 cb->errors = 1;
Qu Wenruo86ccbb42021-09-27 15:21:50 +0800327 finish_compressed_bio_read(cb, bio);
Chris Masonc8b97812008-10-29 14:49:59 -0400328out:
329 bio_put(bio);
330}
331
332/*
333 * Clear the writeback bits on all of the file
334 * pages for a compressed write
335 */
Filipe Manana7bdcefc2014-10-07 01:48:26 +0100336static noinline void end_compressed_writeback(struct inode *inode,
337 const struct compressed_bio *cb)
Chris Masonc8b97812008-10-29 14:49:59 -0400338{
Qu Wenruo741ec652021-09-27 15:22:01 +0800339 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300340 unsigned long index = cb->start >> PAGE_SHIFT;
341 unsigned long end_index = (cb->start + cb->len - 1) >> PAGE_SHIFT;
Chris Masonc8b97812008-10-29 14:49:59 -0400342 struct page *pages[16];
343 unsigned long nr_pages = end_index - index + 1;
344 int i;
345 int ret;
346
Filipe Manana7bdcefc2014-10-07 01:48:26 +0100347 if (cb->errors)
348 mapping_set_error(inode->i_mapping, -EIO);
349
Chris Masond3977122009-01-05 21:25:51 -0500350 while (nr_pages > 0) {
Chris Masonc8b97812008-10-29 14:49:59 -0400351 ret = find_get_pages_contig(inode->i_mapping, index,
Chris Mason5b050f02008-11-11 09:34:41 -0500352 min_t(unsigned long,
353 nr_pages, ARRAY_SIZE(pages)), pages);
Chris Masonc8b97812008-10-29 14:49:59 -0400354 if (ret == 0) {
355 nr_pages -= 1;
356 index += 1;
357 continue;
358 }
359 for (i = 0; i < ret; i++) {
Filipe Manana7bdcefc2014-10-07 01:48:26 +0100360 if (cb->errors)
361 SetPageError(pages[i]);
Qu Wenruo741ec652021-09-27 15:22:01 +0800362 btrfs_page_clamp_clear_writeback(fs_info, pages[i],
363 cb->start, cb->len);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300364 put_page(pages[i]);
Chris Masonc8b97812008-10-29 14:49:59 -0400365 }
366 nr_pages -= ret;
367 index += ret;
368 }
369 /* the inode may be gone now */
Chris Masonc8b97812008-10-29 14:49:59 -0400370}
371
Qu Wenruo6853c642021-09-27 15:21:51 +0800372static void finish_compressed_bio_write(struct compressed_bio *cb)
Chris Masonc8b97812008-10-29 14:49:59 -0400373{
Qu Wenruo6853c642021-09-27 15:21:51 +0800374 struct inode *inode = cb->inode;
Anand Jain1d08ce582021-05-29 17:48:33 +0800375 unsigned int index;
Chris Masonc8b97812008-10-29 14:49:59 -0400376
Qu Wenruo6853c642021-09-27 15:21:51 +0800377 /*
378 * Ok, we're the last bio for this extent, step one is to call back
379 * into the FS and do all the end_io operations.
Chris Masonc8b97812008-10-29 14:49:59 -0400380 */
Qu Wenruo38a39ac72021-04-08 20:32:27 +0800381 btrfs_writepage_endio_finish_ordered(BTRFS_I(inode), NULL,
Nikolay Borisovc6297322018-11-08 10:18:08 +0200382 cb->start, cb->start + cb->len - 1,
Goldwyn Rodrigues240246f2021-07-09 11:29:22 -0500383 !cb->errors);
Chris Masonc8b97812008-10-29 14:49:59 -0400384
Filipe Manana7bdcefc2014-10-07 01:48:26 +0100385 end_compressed_writeback(inode, cb);
Qu Wenruo6853c642021-09-27 15:21:51 +0800386 /* Note, our inode could be gone now */
Chris Masonc8b97812008-10-29 14:49:59 -0400387
388 /*
Qu Wenruo6853c642021-09-27 15:21:51 +0800389 * Release the compressed pages, these came from alloc_page and
Chris Masonc8b97812008-10-29 14:49:59 -0400390 * are not attached to the inode at all
391 */
Chris Masonc8b97812008-10-29 14:49:59 -0400392 for (index = 0; index < cb->nr_pages; index++) {
Qu Wenruo6853c642021-09-27 15:21:51 +0800393 struct page *page = cb->compressed_pages[index];
394
Chris Masonc8b97812008-10-29 14:49:59 -0400395 page->mapping = NULL;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300396 put_page(page);
Chris Masonc8b97812008-10-29 14:49:59 -0400397 }
398
Qu Wenruo6853c642021-09-27 15:21:51 +0800399 /* Finally free the cb struct */
Chris Masonc8b97812008-10-29 14:49:59 -0400400 kfree(cb->compressed_pages);
401 kfree(cb);
Qu Wenruo6853c642021-09-27 15:21:51 +0800402}
403
404/*
405 * Do the cleanup once all the compressed pages hit the disk. This will clear
406 * writeback on the file pages and free the compressed pages.
407 *
408 * This also calls the writeback end hooks for the file pages so that metadata
409 * and checksums can be updated in the file.
410 */
411static void end_compressed_bio_write(struct bio *bio)
412{
413 struct compressed_bio *cb = bio->bi_private;
414
415 if (!dec_and_test_compressed_bio(cb, bio))
416 goto out;
417
418 btrfs_record_physical_zoned(cb->inode, cb->start, bio);
419
420 finish_compressed_bio_write(cb);
Chris Masonc8b97812008-10-29 14:49:59 -0400421out:
422 bio_put(bio);
423}
424
Qu Wenruo2d4e0b82021-09-27 15:21:52 +0800425static blk_status_t submit_compressed_bio(struct btrfs_fs_info *fs_info,
426 struct compressed_bio *cb,
427 struct bio *bio, int mirror_num)
428{
429 blk_status_t ret;
430
431 ASSERT(bio->bi_iter.bi_size);
Qu Wenruo2d4e0b82021-09-27 15:21:52 +0800432 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
433 if (ret)
434 return ret;
435 ret = btrfs_map_bio(fs_info, bio, mirror_num);
436 return ret;
437}
438
Chris Masonc8b97812008-10-29 14:49:59 -0400439/*
Qu Wenruof472c28f2021-09-27 15:21:54 +0800440 * Allocate a compressed_bio, which will be used to read/write on-disk
441 * (aka, compressed) * data.
442 *
443 * @cb: The compressed_bio structure, which records all the needed
444 * information to bind the compressed data to the uncompressed
445 * page cache.
446 * @disk_byten: The logical bytenr where the compressed data will be read
447 * from or written to.
448 * @endio_func: The endio function to call after the IO for compressed data
449 * is finished.
450 * @next_stripe_start: Return value of logical bytenr of where next stripe starts.
451 * Let the caller know to only fill the bio up to the stripe
452 * boundary.
Qu Wenruo22c306f2021-09-27 15:21:53 +0800453 */
Qu Wenruof472c28f2021-09-27 15:21:54 +0800454
455
Qu Wenruo22c306f2021-09-27 15:21:53 +0800456static struct bio *alloc_compressed_bio(struct compressed_bio *cb, u64 disk_bytenr,
Qu Wenruof472c28f2021-09-27 15:21:54 +0800457 unsigned int opf, bio_end_io_t endio_func,
458 u64 *next_stripe_start)
Qu Wenruo22c306f2021-09-27 15:21:53 +0800459{
Qu Wenruof472c28f2021-09-27 15:21:54 +0800460 struct btrfs_fs_info *fs_info = btrfs_sb(cb->inode->i_sb);
461 struct btrfs_io_geometry geom;
462 struct extent_map *em;
Qu Wenruo22c306f2021-09-27 15:21:53 +0800463 struct bio *bio;
Qu Wenruof472c28f2021-09-27 15:21:54 +0800464 int ret;
Qu Wenruo22c306f2021-09-27 15:21:53 +0800465
466 bio = btrfs_bio_alloc(BIO_MAX_VECS);
467
468 bio->bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
469 bio->bi_opf = opf;
470 bio->bi_private = cb;
471 bio->bi_end_io = endio_func;
472
Qu Wenruof472c28f2021-09-27 15:21:54 +0800473 em = btrfs_get_chunk_map(fs_info, disk_bytenr, fs_info->sectorsize);
474 if (IS_ERR(em)) {
475 bio_put(bio);
476 return ERR_CAST(em);
Qu Wenruo22c306f2021-09-27 15:21:53 +0800477 }
Qu Wenruof472c28f2021-09-27 15:21:54 +0800478
479 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
480 bio_set_dev(bio, em->map_lookup->stripes[0].dev->bdev);
481
482 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio), disk_bytenr, &geom);
483 free_extent_map(em);
484 if (ret < 0) {
485 bio_put(bio);
486 return ERR_PTR(ret);
487 }
488 *next_stripe_start = disk_bytenr + geom.len;
489
Qu Wenruo22c306f2021-09-27 15:21:53 +0800490 return bio;
491}
492
Chris Masonc8b97812008-10-29 14:49:59 -0400493/*
494 * worker function to build and submit bios for previously compressed pages.
495 * The corresponding pages in the inode should be marked for writeback
496 * and the compressed pages should have a reference on them for dropping
497 * when the IO is complete.
498 *
499 * This also checksums the file bytes and gets things ready for
500 * the end io hooks.
501 */
Nikolay Borisovc7ee1812020-06-03 08:55:16 +0300502blk_status_t btrfs_submit_compressed_write(struct btrfs_inode *inode, u64 start,
Anand Jain65b53552021-05-29 17:48:35 +0800503 unsigned int len, u64 disk_start,
504 unsigned int compressed_len,
Chris Masonc8b97812008-10-29 14:49:59 -0400505 struct page **compressed_pages,
Anand Jain65b53552021-05-29 17:48:35 +0800506 unsigned int nr_pages,
Chris Masonec39f762019-07-10 12:28:17 -0700507 unsigned int write_flags,
508 struct cgroup_subsys_state *blkcg_css)
Chris Masonc8b97812008-10-29 14:49:59 -0400509{
Nikolay Borisovc7ee1812020-06-03 08:55:16 +0300510 struct btrfs_fs_info *fs_info = inode->root->fs_info;
Chris Masonc8b97812008-10-29 14:49:59 -0400511 struct bio *bio = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -0400512 struct compressed_bio *cb;
Qu Wenruo91507242021-09-27 15:21:55 +0800513 u64 cur_disk_bytenr = disk_start;
Qu Wenruof472c28f2021-09-27 15:21:54 +0800514 u64 next_stripe_start;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200515 blk_status_t ret;
Nikolay Borisovc7ee1812020-06-03 08:55:16 +0300516 int skip_sum = inode->flags & BTRFS_INODE_NODATASUM;
Johannes Thumshirn764c7c92021-05-19 00:40:28 +0900517 const bool use_append = btrfs_use_zone_append(inode, disk_start);
518 const unsigned int bio_op = use_append ? REQ_OP_ZONE_APPEND : REQ_OP_WRITE;
Chris Masonc8b97812008-10-29 14:49:59 -0400519
Qu Wenruobbbff012021-09-27 15:22:00 +0800520 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
521 IS_ALIGNED(len, fs_info->sectorsize));
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400522 cb = kmalloc(compressed_bio_size(fs_info, compressed_len), GFP_NOFS);
Yoshinori Sanodac97e52011-02-15 12:01:42 +0000523 if (!cb)
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200524 return BLK_STS_RESOURCE;
Qu Wenruo6ec97652021-09-27 15:21:48 +0800525 refcount_set(&cb->pending_sectors, compressed_len >> fs_info->sectorsize_bits);
Chris Masonc8b97812008-10-29 14:49:59 -0400526 cb->errors = 0;
Nikolay Borisovc7ee1812020-06-03 08:55:16 +0300527 cb->inode = &inode->vfs_inode;
Chris Masonc8b97812008-10-29 14:49:59 -0400528 cb->start = start;
529 cb->len = len;
Chris Masond20f7042008-12-08 16:58:54 -0500530 cb->mirror_num = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400531 cb->compressed_pages = compressed_pages;
532 cb->compressed_len = compressed_len;
533 cb->orig_bio = NULL;
534 cb->nr_pages = nr_pages;
535
Qu Wenruo91507242021-09-27 15:21:55 +0800536 while (cur_disk_bytenr < disk_start + compressed_len) {
537 u64 offset = cur_disk_bytenr - disk_start;
538 unsigned int index = offset >> PAGE_SHIFT;
539 unsigned int real_size;
540 unsigned int added;
541 struct page *page = compressed_pages[index];
542 bool submit = false;
Chris Masonec39f762019-07-10 12:28:17 -0700543
Qu Wenruo91507242021-09-27 15:21:55 +0800544 /* Allocate new bio if submitted or not yet allocated */
545 if (!bio) {
546 bio = alloc_compressed_bio(cb, cur_disk_bytenr,
547 bio_op | write_flags, end_compressed_bio_write,
548 &next_stripe_start);
549 if (IS_ERR(bio)) {
550 ret = errno_to_blk_status(PTR_ERR(bio));
551 bio = NULL;
552 goto finish_cb;
553 }
Johannes Thumshirn764c7c92021-05-19 00:40:28 +0900554 }
Qu Wenruo91507242021-09-27 15:21:55 +0800555 /*
556 * We should never reach next_stripe_start start as we will
557 * submit comp_bio when reach the boundary immediately.
558 */
559 ASSERT(cur_disk_bytenr != next_stripe_start);
Chris Masonc8b97812008-10-29 14:49:59 -0400560
Qu Wenruo4c80a972021-05-25 13:52:43 +0800561 /*
Qu Wenruo91507242021-09-27 15:21:55 +0800562 * We have various limits on the real read size:
563 * - stripe boundary
564 * - page boundary
565 * - compressed length boundary
Qu Wenruo4c80a972021-05-25 13:52:43 +0800566 */
Qu Wenruo91507242021-09-27 15:21:55 +0800567 real_size = min_t(u64, U32_MAX, next_stripe_start - cur_disk_bytenr);
568 real_size = min_t(u64, real_size, PAGE_SIZE - offset_in_page(offset));
569 real_size = min_t(u64, real_size, compressed_len - offset);
570 ASSERT(IS_ALIGNED(real_size, fs_info->sectorsize));
Johannes Thumshirn764c7c92021-05-19 00:40:28 +0900571
Qu Wenruo91507242021-09-27 15:21:55 +0800572 if (use_append)
573 added = bio_add_zone_append_page(bio, page, real_size,
574 offset_in_page(offset));
575 else
576 added = bio_add_page(bio, page, real_size,
577 offset_in_page(offset));
578 /* Reached zoned boundary */
579 if (added == 0)
580 submit = true;
Chris Masonc8b97812008-10-29 14:49:59 -0400581
Qu Wenruo91507242021-09-27 15:21:55 +0800582 cur_disk_bytenr += added;
583 /* Reached stripe boundary */
584 if (cur_disk_bytenr == next_stripe_start)
585 submit = true;
586
587 /* Finished the range */
588 if (cur_disk_bytenr == disk_start + compressed_len)
589 submit = true;
590
591 if (submit) {
Li Zefane55179b2011-07-14 03:16:47 +0000592 if (!skip_sum) {
Nikolay Borisovc7ee1812020-06-03 08:55:16 +0300593 ret = btrfs_csum_one_bio(inode, bio, start, 1);
Qu Wenruo6853c642021-09-27 15:21:51 +0800594 if (ret)
595 goto finish_cb;
Li Zefane55179b2011-07-14 03:16:47 +0000596 }
Chris Masond20f7042008-12-08 16:58:54 -0500597
Qu Wenruo2d4e0b82021-09-27 15:21:52 +0800598 ret = submit_compressed_bio(fs_info, cb, bio, 0);
Qu Wenruo6853c642021-09-27 15:21:51 +0800599 if (ret)
600 goto finish_cb;
Qu Wenruo91507242021-09-27 15:21:55 +0800601 bio = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -0400602 }
Chris Mason771ed682008-11-06 22:02:51 -0500603 cond_resched();
Chris Masonc8b97812008-10-29 14:49:59 -0400604 }
Dennis Zhou46bcff2b2019-12-11 15:20:15 -0800605 if (blkcg_css)
606 kthread_associate_blkcg(NULL);
607
Chris Masonc8b97812008-10-29 14:49:59 -0400608 return 0;
Qu Wenruo6853c642021-09-27 15:21:51 +0800609
610finish_cb:
611 if (bio) {
612 bio->bi_status = ret;
613 bio_endio(bio);
614 }
Qu Wenruo91507242021-09-27 15:21:55 +0800615 /* Last byte of @cb is submitted, endio will free @cb */
616 if (cur_disk_bytenr == disk_start + compressed_len)
617 return ret;
Qu Wenruo6853c642021-09-27 15:21:51 +0800618
Qu Wenruo91507242021-09-27 15:21:55 +0800619 wait_var_event(cb, refcount_read(&cb->pending_sectors) ==
620 (disk_start + compressed_len - cur_disk_bytenr) >>
621 fs_info->sectorsize_bits);
Qu Wenruo6853c642021-09-27 15:21:51 +0800622 /*
623 * Even with previous bio ended, we should still have io not yet
624 * submitted, thus need to finish manually.
625 */
626 ASSERT(refcount_read(&cb->pending_sectors));
627 /* Now we are the only one referring @cb, can finish it safely. */
628 finish_compressed_bio_write(cb);
629 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400630}
631
Christoph Hellwig2a4d0c92016-11-25 09:07:51 +0100632static u64 bio_end_offset(struct bio *bio)
633{
Ming Leic45a8f22017-12-18 20:22:05 +0800634 struct bio_vec *last = bio_last_bvec_all(bio);
Christoph Hellwig2a4d0c92016-11-25 09:07:51 +0100635
636 return page_offset(last->bv_page) + last->bv_len + last->bv_offset;
637}
638
Qu Wenruo6a404912021-09-27 15:21:47 +0800639/*
640 * Add extra pages in the same compressed file extent so that we don't need to
641 * re-read the same extent again and again.
642 *
643 * NOTE: this won't work well for subpage, as for subpage read, we lock the
644 * full page then submit bio for each compressed/regular extents.
645 *
646 * This means, if we have several sectors in the same page points to the same
647 * on-disk compressed data, we will re-read the same extent many times and
648 * this function can only help for the next page.
649 */
Chris Mason771ed682008-11-06 22:02:51 -0500650static noinline int add_ra_bio_pages(struct inode *inode,
651 u64 compressed_end,
652 struct compressed_bio *cb)
653{
Qu Wenruo6a404912021-09-27 15:21:47 +0800654 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Mason771ed682008-11-06 22:02:51 -0500655 unsigned long end_index;
Qu Wenruo6a404912021-09-27 15:21:47 +0800656 u64 cur = bio_end_offset(cb->orig_bio);
Chris Mason771ed682008-11-06 22:02:51 -0500657 u64 isize = i_size_read(inode);
658 int ret;
659 struct page *page;
Chris Mason771ed682008-11-06 22:02:51 -0500660 struct extent_map *em;
661 struct address_space *mapping = inode->i_mapping;
Chris Mason771ed682008-11-06 22:02:51 -0500662 struct extent_map_tree *em_tree;
663 struct extent_io_tree *tree;
Qu Wenruo6a404912021-09-27 15:21:47 +0800664 int sectors_missed = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500665
Chris Mason771ed682008-11-06 22:02:51 -0500666 em_tree = &BTRFS_I(inode)->extent_tree;
667 tree = &BTRFS_I(inode)->io_tree;
668
669 if (isize == 0)
670 return 0;
671
Qu Wenruoca62e852021-07-26 14:34:52 +0800672 /*
673 * For current subpage support, we only support 64K page size,
674 * which means maximum compressed extent size (128K) is just 2x page
675 * size.
676 * This makes readahead less effective, so here disable readahead for
677 * subpage for now, until full compressed write is supported.
678 */
679 if (btrfs_sb(inode->i_sb)->sectorsize < PAGE_SIZE)
680 return 0;
681
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300682 end_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
Chris Mason771ed682008-11-06 22:02:51 -0500683
Qu Wenruo6a404912021-09-27 15:21:47 +0800684 while (cur < compressed_end) {
685 u64 page_end;
686 u64 pg_index = cur >> PAGE_SHIFT;
687 u32 add_size;
Chris Mason771ed682008-11-06 22:02:51 -0500688
David Sterba306e16c2011-04-19 14:29:38 +0200689 if (pg_index > end_index)
Chris Mason771ed682008-11-06 22:02:51 -0500690 break;
691
Matthew Wilcox0a943c62017-12-04 10:37:22 -0500692 page = xa_load(&mapping->i_pages, pg_index);
Matthew Wilcox3159f942017-11-03 13:30:42 -0400693 if (page && !xa_is_value(page)) {
Qu Wenruo6a404912021-09-27 15:21:47 +0800694 sectors_missed += (PAGE_SIZE - offset_in_page(cur)) >>
695 fs_info->sectorsize_bits;
696
697 /* Beyond threshold, no need to continue */
698 if (sectors_missed > 4)
Chris Mason771ed682008-11-06 22:02:51 -0500699 break;
Qu Wenruo6a404912021-09-27 15:21:47 +0800700
701 /*
702 * Jump to next page start as we already have page for
703 * current offset.
704 */
705 cur = (pg_index << PAGE_SHIFT) + PAGE_SIZE;
706 continue;
Chris Mason771ed682008-11-06 22:02:51 -0500707 }
708
Michal Hockoc62d2552015-11-06 16:28:49 -0800709 page = __page_cache_alloc(mapping_gfp_constraint(mapping,
710 ~__GFP_FS));
Chris Mason771ed682008-11-06 22:02:51 -0500711 if (!page)
712 break;
713
Michal Hockoc62d2552015-11-06 16:28:49 -0800714 if (add_to_page_cache_lru(page, mapping, pg_index, GFP_NOFS)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300715 put_page(page);
Qu Wenruo6a404912021-09-27 15:21:47 +0800716 /* There is already a page, skip to page end */
717 cur = (pg_index << PAGE_SHIFT) + PAGE_SIZE;
718 continue;
Chris Mason771ed682008-11-06 22:02:51 -0500719 }
720
Qu Wenruo32443de2021-01-26 16:34:00 +0800721 ret = set_page_extent_mapped(page);
722 if (ret < 0) {
723 unlock_page(page);
724 put_page(page);
725 break;
726 }
727
Qu Wenruo6a404912021-09-27 15:21:47 +0800728 page_end = (pg_index << PAGE_SHIFT) + PAGE_SIZE - 1;
729 lock_extent(tree, cur, page_end);
Chris Mason890871b2009-09-02 16:24:52 -0400730 read_lock(&em_tree->lock);
Qu Wenruo6a404912021-09-27 15:21:47 +0800731 em = lookup_extent_mapping(em_tree, cur, page_end + 1 - cur);
Chris Mason890871b2009-09-02 16:24:52 -0400732 read_unlock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500733
Qu Wenruo6a404912021-09-27 15:21:47 +0800734 /*
735 * At this point, we have a locked page in the page cache for
736 * these bytes in the file. But, we have to make sure they map
737 * to this compressed extent on disk.
738 */
739 if (!em || cur < em->start ||
740 (cur + fs_info->sectorsize > extent_map_end(em)) ||
Kent Overstreet4f024f32013-10-11 15:44:27 -0700741 (em->block_start >> 9) != cb->orig_bio->bi_iter.bi_sector) {
Chris Mason771ed682008-11-06 22:02:51 -0500742 free_extent_map(em);
Qu Wenruo6a404912021-09-27 15:21:47 +0800743 unlock_extent(tree, cur, page_end);
Chris Mason771ed682008-11-06 22:02:51 -0500744 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300745 put_page(page);
Chris Mason771ed682008-11-06 22:02:51 -0500746 break;
747 }
748 free_extent_map(em);
749
750 if (page->index == end_index) {
Johannes Thumshirn70730172018-12-05 15:23:03 +0100751 size_t zero_offset = offset_in_page(isize);
Chris Mason771ed682008-11-06 22:02:51 -0500752
753 if (zero_offset) {
754 int zeros;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300755 zeros = PAGE_SIZE - zero_offset;
Ira Weinyd048b9c2021-05-04 18:40:07 -0700756 memzero_page(page, zero_offset, zeros);
Chris Mason771ed682008-11-06 22:02:51 -0500757 flush_dcache_page(page);
Chris Mason771ed682008-11-06 22:02:51 -0500758 }
759 }
760
Qu Wenruo6a404912021-09-27 15:21:47 +0800761 add_size = min(em->start + em->len, page_end + 1) - cur;
762 ret = bio_add_page(cb->orig_bio, page, add_size, offset_in_page(cur));
763 if (ret != add_size) {
764 unlock_extent(tree, cur, page_end);
Chris Mason771ed682008-11-06 22:02:51 -0500765 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300766 put_page(page);
Chris Mason771ed682008-11-06 22:02:51 -0500767 break;
768 }
Qu Wenruo6a404912021-09-27 15:21:47 +0800769 /*
770 * If it's subpage, we also need to increase its
771 * subpage::readers number, as at endio we will decrease
772 * subpage::readers and to unlock the page.
773 */
774 if (fs_info->sectorsize < PAGE_SIZE)
775 btrfs_subpage_start_reader(fs_info, page, cur, add_size);
776 put_page(page);
777 cur += add_size;
Chris Mason771ed682008-11-06 22:02:51 -0500778 }
Chris Mason771ed682008-11-06 22:02:51 -0500779 return 0;
780}
781
Chris Masonc8b97812008-10-29 14:49:59 -0400782/*
783 * for a compressed read, the bio we get passed has all the inode pages
784 * in it. We don't actually do IO on those pages but allocate new ones
785 * to hold the compressed pages on disk.
786 *
Kent Overstreet4f024f32013-10-11 15:44:27 -0700787 * bio->bi_iter.bi_sector points to the compressed extent on disk
Chris Masonc8b97812008-10-29 14:49:59 -0400788 * bio->bi_io_vec points to all of the inode pages
Chris Masonc8b97812008-10-29 14:49:59 -0400789 *
790 * After the compressed pages are read, we copy the bytes into the
791 * bio we were passed and then call the bio end_io calls
792 */
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200793blk_status_t btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
Chris Masonc8b97812008-10-29 14:49:59 -0400794 int mirror_num, unsigned long bio_flags)
795{
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400796 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
Chris Masonc8b97812008-10-29 14:49:59 -0400797 struct extent_map_tree *em_tree;
798 struct compressed_bio *cb;
Anand Jain356b4a22021-05-29 17:48:34 +0800799 unsigned int compressed_len;
800 unsigned int nr_pages;
801 unsigned int pg_index;
Qu Wenruof472c28f2021-09-27 15:21:54 +0800802 struct bio *comp_bio = NULL;
803 const u64 disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
804 u64 cur_disk_byte = disk_bytenr;
805 u64 next_stripe_start;
Qu Wenruo557023e2021-07-05 10:00:56 +0800806 u64 file_offset;
Chris Masone04ca622008-11-10 11:44:58 -0500807 u64 em_len;
808 u64 em_start;
Chris Masonc8b97812008-10-29 14:49:59 -0400809 struct extent_map *em;
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200810 blk_status_t ret = BLK_STS_RESOURCE;
Josef Bacik15e3004a2012-10-05 13:39:50 -0400811 int faili = 0;
Johannes Thumshirn10fe6ca2019-05-22 10:19:02 +0200812 u8 *sums;
Chris Masonc8b97812008-10-29 14:49:59 -0400813
Chris Masonc8b97812008-10-29 14:49:59 -0400814 em_tree = &BTRFS_I(inode)->extent_tree;
815
Qu Wenruo557023e2021-07-05 10:00:56 +0800816 file_offset = bio_first_bvec_all(bio)->bv_offset +
817 page_offset(bio_first_page_all(bio));
818
Chris Masonc8b97812008-10-29 14:49:59 -0400819 /* we need the actual starting offset of this extent in the file */
Chris Mason890871b2009-09-02 16:24:52 -0400820 read_lock(&em_tree->lock);
Qu Wenruo557023e2021-07-05 10:00:56 +0800821 em = lookup_extent_mapping(em_tree, file_offset, fs_info->sectorsize);
Chris Mason890871b2009-09-02 16:24:52 -0400822 read_unlock(&em_tree->lock);
Tsutomu Itoh285190d2012-02-16 16:23:58 +0900823 if (!em)
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200824 return BLK_STS_IOERR;
Chris Masonc8b97812008-10-29 14:49:59 -0400825
Qu Wenruo557023e2021-07-05 10:00:56 +0800826 ASSERT(em->compress_type != BTRFS_COMPRESS_NONE);
Chris Masond20f7042008-12-08 16:58:54 -0500827 compressed_len = em->block_len;
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400828 cb = kmalloc(compressed_bio_size(fs_info, compressed_len), GFP_NOFS);
liubo6b82ce82011-01-26 06:21:39 +0000829 if (!cb)
830 goto out;
831
Qu Wenruo6ec97652021-09-27 15:21:48 +0800832 refcount_set(&cb->pending_sectors, compressed_len >> fs_info->sectorsize_bits);
Chris Masonc8b97812008-10-29 14:49:59 -0400833 cb->errors = 0;
834 cb->inode = inode;
Chris Masond20f7042008-12-08 16:58:54 -0500835 cb->mirror_num = mirror_num;
Johannes Thumshirn10fe6ca2019-05-22 10:19:02 +0200836 sums = cb->sums;
Chris Masonc8b97812008-10-29 14:49:59 -0400837
Yan Zhengff5b7ee2008-11-10 07:34:43 -0500838 cb->start = em->orig_start;
Chris Masone04ca622008-11-10 11:44:58 -0500839 em_len = em->len;
840 em_start = em->start;
Chris Masond20f7042008-12-08 16:58:54 -0500841
Chris Masonc8b97812008-10-29 14:49:59 -0400842 free_extent_map(em);
Chris Masone04ca622008-11-10 11:44:58 -0500843 em = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -0400844
Christoph Hellwig81381052016-11-25 09:07:50 +0100845 cb->len = bio->bi_iter.bi_size;
Chris Masonc8b97812008-10-29 14:49:59 -0400846 cb->compressed_len = compressed_len;
Li Zefan261507a02010-12-17 14:21:50 +0800847 cb->compress_type = extent_compress_type(bio_flags);
Chris Masonc8b97812008-10-29 14:49:59 -0400848 cb->orig_bio = bio;
849
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300850 nr_pages = DIV_ROUND_UP(compressed_len, PAGE_SIZE);
David Sterba31e818f2015-02-20 18:00:26 +0100851 cb->compressed_pages = kcalloc(nr_pages, sizeof(struct page *),
Chris Masonc8b97812008-10-29 14:49:59 -0400852 GFP_NOFS);
liubo6b82ce82011-01-26 06:21:39 +0000853 if (!cb->compressed_pages)
854 goto fail1;
855
David Sterba306e16c2011-04-19 14:29:38 +0200856 for (pg_index = 0; pg_index < nr_pages; pg_index++) {
David Sterbab0ee5e12021-06-14 22:22:22 +0200857 cb->compressed_pages[pg_index] = alloc_page(GFP_NOFS);
Josef Bacik15e3004a2012-10-05 13:39:50 -0400858 if (!cb->compressed_pages[pg_index]) {
859 faili = pg_index - 1;
Dan Carpenter0e9350d2017-06-19 13:55:37 +0300860 ret = BLK_STS_RESOURCE;
liubo6b82ce82011-01-26 06:21:39 +0000861 goto fail2;
Josef Bacik15e3004a2012-10-05 13:39:50 -0400862 }
Chris Masonc8b97812008-10-29 14:49:59 -0400863 }
Josef Bacik15e3004a2012-10-05 13:39:50 -0400864 faili = nr_pages - 1;
Chris Masonc8b97812008-10-29 14:49:59 -0400865 cb->nr_pages = nr_pages;
866
Filipe Manana7f042a82016-01-27 19:17:20 +0000867 add_ra_bio_pages(inode, em_start + em_len, cb);
Chris Mason771ed682008-11-06 22:02:51 -0500868
Chris Mason771ed682008-11-06 22:02:51 -0500869 /* include any pages we added in add_ra-bio_pages */
Christoph Hellwig81381052016-11-25 09:07:50 +0100870 cb->len = bio->bi_iter.bi_size;
Chris Mason771ed682008-11-06 22:02:51 -0500871
Qu Wenruof472c28f2021-09-27 15:21:54 +0800872 while (cur_disk_byte < disk_bytenr + compressed_len) {
873 u64 offset = cur_disk_byte - disk_bytenr;
874 unsigned int index = offset >> PAGE_SHIFT;
875 unsigned int real_size;
876 unsigned int added;
877 struct page *page = cb->compressed_pages[index];
878 bool submit = false;
Chris Masonc8b97812008-10-29 14:49:59 -0400879
Qu Wenruof472c28f2021-09-27 15:21:54 +0800880 /* Allocate new bio if submitted or not yet allocated */
881 if (!comp_bio) {
882 comp_bio = alloc_compressed_bio(cb, cur_disk_byte,
883 REQ_OP_READ, end_compressed_bio_read,
884 &next_stripe_start);
885 if (IS_ERR(comp_bio)) {
886 ret = errno_to_blk_status(PTR_ERR(comp_bio));
887 comp_bio = NULL;
888 goto finish_cb;
889 }
890 }
Qu Wenruobe6a1362021-02-04 15:03:23 +0800891 /*
Qu Wenruof472c28f2021-09-27 15:21:54 +0800892 * We should never reach next_stripe_start start as we will
893 * submit comp_bio when reach the boundary immediately.
Qu Wenruobe6a1362021-02-04 15:03:23 +0800894 */
Qu Wenruof472c28f2021-09-27 15:21:54 +0800895 ASSERT(cur_disk_byte != next_stripe_start);
896 /*
897 * We have various limit on the real read size:
898 * - stripe boundary
899 * - page boundary
900 * - compressed length boundary
901 */
902 real_size = min_t(u64, U32_MAX, next_stripe_start - cur_disk_byte);
903 real_size = min_t(u64, real_size, PAGE_SIZE - offset_in_page(offset));
904 real_size = min_t(u64, real_size, compressed_len - offset);
905 ASSERT(IS_ALIGNED(real_size, fs_info->sectorsize));
Qu Wenruobe6a1362021-02-04 15:03:23 +0800906
Qu Wenruof472c28f2021-09-27 15:21:54 +0800907 added = bio_add_page(comp_bio, page, real_size, offset_in_page(offset));
908 /*
909 * Maximum compressed extent is smaller than bio size limit,
910 * thus bio_add_page() should always success.
911 */
912 ASSERT(added == real_size);
913 cur_disk_byte += added;
Chris Masond20f7042008-12-08 16:58:54 -0500914
Qu Wenruof472c28f2021-09-27 15:21:54 +0800915 /* Reached stripe boundary, need to submit */
916 if (cur_disk_byte == next_stripe_start)
917 submit = true;
Chris Masonc8b97812008-10-29 14:49:59 -0400918
Qu Wenruof472c28f2021-09-27 15:21:54 +0800919 /* Has finished the range, need to submit */
920 if (cur_disk_byte == disk_bytenr + compressed_len)
921 submit = true;
922
923 if (submit) {
Johannes Thumshirn10fe6ca2019-05-22 10:19:02 +0200924 unsigned int nr_sectors;
925
Qu Wenruo62751932020-12-02 14:48:06 +0800926 ret = btrfs_lookup_bio_sums(inode, comp_bio, sums);
Qu Wenruo86ccbb42021-09-27 15:21:50 +0800927 if (ret)
928 goto finish_cb;
Johannes Thumshirn10fe6ca2019-05-22 10:19:02 +0200929
930 nr_sectors = DIV_ROUND_UP(comp_bio->bi_iter.bi_size,
931 fs_info->sectorsize);
David Sterba713cebf2020-06-30 18:04:02 +0200932 sums += fs_info->csum_size * nr_sectors;
Chris Masond20f7042008-12-08 16:58:54 -0500933
Qu Wenruo2d4e0b82021-09-27 15:21:52 +0800934 ret = submit_compressed_bio(fs_info, cb, comp_bio, mirror_num);
Qu Wenruo86ccbb42021-09-27 15:21:50 +0800935 if (ret)
936 goto finish_cb;
Qu Wenruof472c28f2021-09-27 15:21:54 +0800937 comp_bio = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -0400938 }
Chris Masonc8b97812008-10-29 14:49:59 -0400939 }
Chris Masonc8b97812008-10-29 14:49:59 -0400940 return 0;
liubo6b82ce82011-01-26 06:21:39 +0000941
942fail2:
Josef Bacik15e3004a2012-10-05 13:39:50 -0400943 while (faili >= 0) {
944 __free_page(cb->compressed_pages[faili]);
945 faili--;
946 }
liubo6b82ce82011-01-26 06:21:39 +0000947
948 kfree(cb->compressed_pages);
949fail1:
950 kfree(cb);
951out:
952 free_extent_map(em);
953 return ret;
Qu Wenruo86ccbb42021-09-27 15:21:50 +0800954finish_cb:
955 if (comp_bio) {
956 comp_bio->bi_status = ret;
957 bio_endio(comp_bio);
958 }
Qu Wenruof472c28f2021-09-27 15:21:54 +0800959 /* All bytes of @cb is submitted, endio will free @cb */
960 if (cur_disk_byte == disk_bytenr + compressed_len)
961 return ret;
962
963 wait_var_event(cb, refcount_read(&cb->pending_sectors) ==
964 (disk_bytenr + compressed_len - cur_disk_byte) >>
965 fs_info->sectorsize_bits);
Qu Wenruo86ccbb42021-09-27 15:21:50 +0800966 /*
967 * Even with previous bio ended, we should still have io not yet
968 * submitted, thus need to finish @cb manually.
969 */
970 ASSERT(refcount_read(&cb->pending_sectors));
971 /* Now we are the only one referring @cb, can finish it safely. */
972 finish_compressed_bio_read(cb, NULL);
973 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400974}
Li Zefan261507a02010-12-17 14:21:50 +0800975
Timofey Titovets17b5a6c2017-09-28 17:33:37 +0300976/*
977 * Heuristic uses systematic sampling to collect data from the input data
978 * range, the logic can be tuned by the following constants:
979 *
980 * @SAMPLING_READ_SIZE - how many bytes will be copied from for each sample
981 * @SAMPLING_INTERVAL - range from which the sampled data can be collected
982 */
983#define SAMPLING_READ_SIZE (16)
984#define SAMPLING_INTERVAL (256)
985
986/*
987 * For statistical analysis of the input data we consider bytes that form a
988 * Galois Field of 256 objects. Each object has an attribute count, ie. how
989 * many times the object appeared in the sample.
990 */
991#define BUCKET_SIZE (256)
992
993/*
994 * The size of the sample is based on a statistical sampling rule of thumb.
995 * The common way is to perform sampling tests as long as the number of
996 * elements in each cell is at least 5.
997 *
998 * Instead of 5, we choose 32 to obtain more accurate results.
999 * If the data contain the maximum number of symbols, which is 256, we obtain a
1000 * sample size bound by 8192.
1001 *
1002 * For a sample of at most 8KB of data per data range: 16 consecutive bytes
1003 * from up to 512 locations.
1004 */
1005#define MAX_SAMPLE_SIZE (BTRFS_MAX_UNCOMPRESSED * \
1006 SAMPLING_READ_SIZE / SAMPLING_INTERVAL)
1007
1008struct bucket_item {
1009 u32 count;
1010};
Timofey Titovets4e439a02017-09-28 17:33:36 +03001011
1012struct heuristic_ws {
Timofey Titovets17b5a6c2017-09-28 17:33:37 +03001013 /* Partial copy of input data */
1014 u8 *sample;
Timofey Titovetsa440d482017-09-28 17:33:38 +03001015 u32 sample_size;
Timofey Titovets17b5a6c2017-09-28 17:33:37 +03001016 /* Buckets store counters for each byte value */
1017 struct bucket_item *bucket;
Timofey Titovets440c8402017-12-04 00:30:33 +03001018 /* Sorting buffer */
1019 struct bucket_item *bucket_b;
Timofey Titovets4e439a02017-09-28 17:33:36 +03001020 struct list_head list;
1021};
1022
Dennis Zhou92ee55302019-02-04 15:20:03 -05001023static struct workspace_manager heuristic_wsm;
1024
Timofey Titovets4e439a02017-09-28 17:33:36 +03001025static void free_heuristic_ws(struct list_head *ws)
1026{
1027 struct heuristic_ws *workspace;
1028
1029 workspace = list_entry(ws, struct heuristic_ws, list);
1030
Timofey Titovets17b5a6c2017-09-28 17:33:37 +03001031 kvfree(workspace->sample);
1032 kfree(workspace->bucket);
Timofey Titovets440c8402017-12-04 00:30:33 +03001033 kfree(workspace->bucket_b);
Timofey Titovets4e439a02017-09-28 17:33:36 +03001034 kfree(workspace);
1035}
1036
Dennis Zhou7bf49942019-02-04 15:20:04 -05001037static struct list_head *alloc_heuristic_ws(unsigned int level)
Timofey Titovets4e439a02017-09-28 17:33:36 +03001038{
1039 struct heuristic_ws *ws;
1040
1041 ws = kzalloc(sizeof(*ws), GFP_KERNEL);
1042 if (!ws)
1043 return ERR_PTR(-ENOMEM);
1044
Timofey Titovets17b5a6c2017-09-28 17:33:37 +03001045 ws->sample = kvmalloc(MAX_SAMPLE_SIZE, GFP_KERNEL);
1046 if (!ws->sample)
1047 goto fail;
Timofey Titovets4e439a02017-09-28 17:33:36 +03001048
Timofey Titovets17b5a6c2017-09-28 17:33:37 +03001049 ws->bucket = kcalloc(BUCKET_SIZE, sizeof(*ws->bucket), GFP_KERNEL);
1050 if (!ws->bucket)
1051 goto fail;
1052
Timofey Titovets440c8402017-12-04 00:30:33 +03001053 ws->bucket_b = kcalloc(BUCKET_SIZE, sizeof(*ws->bucket_b), GFP_KERNEL);
1054 if (!ws->bucket_b)
1055 goto fail;
1056
Timofey Titovets17b5a6c2017-09-28 17:33:37 +03001057 INIT_LIST_HEAD(&ws->list);
Timofey Titovets4e439a02017-09-28 17:33:36 +03001058 return &ws->list;
Timofey Titovets17b5a6c2017-09-28 17:33:37 +03001059fail:
1060 free_heuristic_ws(&ws->list);
1061 return ERR_PTR(-ENOMEM);
Timofey Titovets4e439a02017-09-28 17:33:36 +03001062}
1063
Dennis Zhouca4ac362019-02-04 15:19:59 -05001064const struct btrfs_compress_op btrfs_heuristic_compress = {
David Sterbabe9510452019-10-02 00:53:31 +02001065 .workspace_manager = &heuristic_wsm,
Dennis Zhouca4ac362019-02-04 15:19:59 -05001066};
1067
David Sterbae8c9f182015-01-02 18:23:10 +01001068static const struct btrfs_compress_op * const btrfs_compress_op[] = {
Dennis Zhouca4ac362019-02-04 15:19:59 -05001069 /* The heuristic is represented as compression type 0 */
1070 &btrfs_heuristic_compress,
Li Zefan261507a02010-12-17 14:21:50 +08001071 &btrfs_zlib_compress,
Li Zefana6fa6fa2010-10-25 15:12:26 +08001072 &btrfs_lzo_compress,
Nick Terrell5c1aab12017-08-09 19:39:02 -07001073 &btrfs_zstd_compress,
Li Zefan261507a02010-12-17 14:21:50 +08001074};
1075
David Sterbac778df12019-10-04 02:47:39 +02001076static struct list_head *alloc_workspace(int type, unsigned int level)
1077{
1078 switch (type) {
1079 case BTRFS_COMPRESS_NONE: return alloc_heuristic_ws(level);
1080 case BTRFS_COMPRESS_ZLIB: return zlib_alloc_workspace(level);
1081 case BTRFS_COMPRESS_LZO: return lzo_alloc_workspace(level);
1082 case BTRFS_COMPRESS_ZSTD: return zstd_alloc_workspace(level);
1083 default:
1084 /*
1085 * This can't happen, the type is validated several times
1086 * before we get here.
1087 */
1088 BUG();
1089 }
1090}
1091
David Sterba1e002352019-10-04 02:57:22 +02001092static void free_workspace(int type, struct list_head *ws)
1093{
1094 switch (type) {
1095 case BTRFS_COMPRESS_NONE: return free_heuristic_ws(ws);
1096 case BTRFS_COMPRESS_ZLIB: return zlib_free_workspace(ws);
1097 case BTRFS_COMPRESS_LZO: return lzo_free_workspace(ws);
1098 case BTRFS_COMPRESS_ZSTD: return zstd_free_workspace(ws);
1099 default:
1100 /*
1101 * This can't happen, the type is validated several times
1102 * before we get here.
1103 */
1104 BUG();
1105 }
1106}
1107
David Sterbad5517032019-10-02 01:08:03 +02001108static void btrfs_init_workspace_manager(int type)
Li Zefan261507a02010-12-17 14:21:50 +08001109{
David Sterba0cf25212019-10-04 03:09:55 +02001110 struct workspace_manager *wsm;
Timofey Titovets4e439a02017-09-28 17:33:36 +03001111 struct list_head *workspace;
Li Zefan261507a02010-12-17 14:21:50 +08001112
David Sterba0cf25212019-10-04 03:09:55 +02001113 wsm = btrfs_compress_op[type]->workspace_manager;
Dennis Zhou92ee55302019-02-04 15:20:03 -05001114 INIT_LIST_HEAD(&wsm->idle_ws);
1115 spin_lock_init(&wsm->ws_lock);
1116 atomic_set(&wsm->total_ws, 0);
1117 init_waitqueue_head(&wsm->ws_wait);
David Sterbaf77dd0d2016-04-27 02:55:15 +02001118
Dennis Zhou1666eda2019-02-04 15:20:01 -05001119 /*
1120 * Preallocate one workspace for each compression type so we can
1121 * guarantee forward progress in the worst case
1122 */
David Sterbac778df12019-10-04 02:47:39 +02001123 workspace = alloc_workspace(type, 0);
Dennis Zhou1666eda2019-02-04 15:20:01 -05001124 if (IS_ERR(workspace)) {
1125 pr_warn(
1126 "BTRFS: cannot preallocate compression workspace, will try later\n");
1127 } else {
Dennis Zhou92ee55302019-02-04 15:20:03 -05001128 atomic_set(&wsm->total_ws, 1);
1129 wsm->free_ws = 1;
1130 list_add(workspace, &wsm->idle_ws);
Dennis Zhou1666eda2019-02-04 15:20:01 -05001131 }
1132}
1133
David Sterba25103072019-10-02 01:08:03 +02001134static void btrfs_cleanup_workspace_manager(int type)
Dennis Zhou1666eda2019-02-04 15:20:01 -05001135{
David Sterba2dba7142019-10-04 01:40:58 +02001136 struct workspace_manager *wsman;
Dennis Zhou1666eda2019-02-04 15:20:01 -05001137 struct list_head *ws;
1138
David Sterba2dba7142019-10-04 01:40:58 +02001139 wsman = btrfs_compress_op[type]->workspace_manager;
Dennis Zhou1666eda2019-02-04 15:20:01 -05001140 while (!list_empty(&wsman->idle_ws)) {
1141 ws = wsman->idle_ws.next;
1142 list_del(ws);
David Sterba1e002352019-10-04 02:57:22 +02001143 free_workspace(type, ws);
Dennis Zhou1666eda2019-02-04 15:20:01 -05001144 atomic_dec(&wsman->total_ws);
Li Zefan261507a02010-12-17 14:21:50 +08001145 }
Li Zefan261507a02010-12-17 14:21:50 +08001146}
1147
1148/*
David Sterbae721e492016-04-27 02:41:17 +02001149 * This finds an available workspace or allocates a new one.
1150 * If it's not possible to allocate a new one, waits until there's one.
1151 * Preallocation makes a forward progress guarantees and we do not return
1152 * errors.
Li Zefan261507a02010-12-17 14:21:50 +08001153 */
David Sterba5907a9b2019-10-04 02:50:28 +02001154struct list_head *btrfs_get_workspace(int type, unsigned int level)
Li Zefan261507a02010-12-17 14:21:50 +08001155{
David Sterba5907a9b2019-10-04 02:50:28 +02001156 struct workspace_manager *wsm;
Li Zefan261507a02010-12-17 14:21:50 +08001157 struct list_head *workspace;
1158 int cpus = num_online_cpus();
David Sterbafe308532017-05-31 17:14:56 +02001159 unsigned nofs_flag;
Timofey Titovets4e439a02017-09-28 17:33:36 +03001160 struct list_head *idle_ws;
1161 spinlock_t *ws_lock;
1162 atomic_t *total_ws;
1163 wait_queue_head_t *ws_wait;
1164 int *free_ws;
Li Zefan261507a02010-12-17 14:21:50 +08001165
David Sterba5907a9b2019-10-04 02:50:28 +02001166 wsm = btrfs_compress_op[type]->workspace_manager;
Dennis Zhou92ee55302019-02-04 15:20:03 -05001167 idle_ws = &wsm->idle_ws;
1168 ws_lock = &wsm->ws_lock;
1169 total_ws = &wsm->total_ws;
1170 ws_wait = &wsm->ws_wait;
1171 free_ws = &wsm->free_ws;
Timofey Titovets4e439a02017-09-28 17:33:36 +03001172
Li Zefan261507a02010-12-17 14:21:50 +08001173again:
Byongho Leed9187642015-10-14 14:05:24 +09001174 spin_lock(ws_lock);
1175 if (!list_empty(idle_ws)) {
1176 workspace = idle_ws->next;
Li Zefan261507a02010-12-17 14:21:50 +08001177 list_del(workspace);
David Sterba6ac10a62016-04-27 02:15:15 +02001178 (*free_ws)--;
Byongho Leed9187642015-10-14 14:05:24 +09001179 spin_unlock(ws_lock);
Li Zefan261507a02010-12-17 14:21:50 +08001180 return workspace;
1181
1182 }
David Sterba6ac10a62016-04-27 02:15:15 +02001183 if (atomic_read(total_ws) > cpus) {
Li Zefan261507a02010-12-17 14:21:50 +08001184 DEFINE_WAIT(wait);
1185
Byongho Leed9187642015-10-14 14:05:24 +09001186 spin_unlock(ws_lock);
1187 prepare_to_wait(ws_wait, &wait, TASK_UNINTERRUPTIBLE);
David Sterba6ac10a62016-04-27 02:15:15 +02001188 if (atomic_read(total_ws) > cpus && !*free_ws)
Li Zefan261507a02010-12-17 14:21:50 +08001189 schedule();
Byongho Leed9187642015-10-14 14:05:24 +09001190 finish_wait(ws_wait, &wait);
Li Zefan261507a02010-12-17 14:21:50 +08001191 goto again;
1192 }
David Sterba6ac10a62016-04-27 02:15:15 +02001193 atomic_inc(total_ws);
Byongho Leed9187642015-10-14 14:05:24 +09001194 spin_unlock(ws_lock);
Li Zefan261507a02010-12-17 14:21:50 +08001195
David Sterbafe308532017-05-31 17:14:56 +02001196 /*
1197 * Allocation helpers call vmalloc that can't use GFP_NOFS, so we have
1198 * to turn it off here because we might get called from the restricted
1199 * context of btrfs_compress_bio/btrfs_compress_pages
1200 */
1201 nofs_flag = memalloc_nofs_save();
David Sterbac778df12019-10-04 02:47:39 +02001202 workspace = alloc_workspace(type, level);
David Sterbafe308532017-05-31 17:14:56 +02001203 memalloc_nofs_restore(nofs_flag);
1204
Li Zefan261507a02010-12-17 14:21:50 +08001205 if (IS_ERR(workspace)) {
David Sterba6ac10a62016-04-27 02:15:15 +02001206 atomic_dec(total_ws);
Byongho Leed9187642015-10-14 14:05:24 +09001207 wake_up(ws_wait);
David Sterbae721e492016-04-27 02:41:17 +02001208
1209 /*
1210 * Do not return the error but go back to waiting. There's a
1211 * workspace preallocated for each type and the compression
1212 * time is bounded so we get to a workspace eventually. This
1213 * makes our caller's life easier.
David Sterba523567162016-04-27 03:07:39 +02001214 *
1215 * To prevent silent and low-probability deadlocks (when the
1216 * initial preallocation fails), check if there are any
1217 * workspaces at all.
David Sterbae721e492016-04-27 02:41:17 +02001218 */
David Sterba523567162016-04-27 03:07:39 +02001219 if (atomic_read(total_ws) == 0) {
1220 static DEFINE_RATELIMIT_STATE(_rs,
1221 /* once per minute */ 60 * HZ,
1222 /* no burst */ 1);
1223
1224 if (__ratelimit(&_rs)) {
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04001225 pr_warn("BTRFS: no compression workspaces, low memory, retrying\n");
David Sterba523567162016-04-27 03:07:39 +02001226 }
1227 }
David Sterbae721e492016-04-27 02:41:17 +02001228 goto again;
Li Zefan261507a02010-12-17 14:21:50 +08001229 }
1230 return workspace;
1231}
1232
Dennis Zhou7bf49942019-02-04 15:20:04 -05001233static struct list_head *get_workspace(int type, int level)
Dennis Zhou929f4ba2019-02-04 15:20:02 -05001234{
David Sterba6a0d1272019-10-04 02:36:16 +02001235 switch (type) {
David Sterba5907a9b2019-10-04 02:50:28 +02001236 case BTRFS_COMPRESS_NONE: return btrfs_get_workspace(type, level);
David Sterba6a0d1272019-10-04 02:36:16 +02001237 case BTRFS_COMPRESS_ZLIB: return zlib_get_workspace(level);
David Sterba5907a9b2019-10-04 02:50:28 +02001238 case BTRFS_COMPRESS_LZO: return btrfs_get_workspace(type, level);
David Sterba6a0d1272019-10-04 02:36:16 +02001239 case BTRFS_COMPRESS_ZSTD: return zstd_get_workspace(level);
1240 default:
1241 /*
1242 * This can't happen, the type is validated several times
1243 * before we get here.
1244 */
1245 BUG();
1246 }
Dennis Zhou929f4ba2019-02-04 15:20:02 -05001247}
1248
Li Zefan261507a02010-12-17 14:21:50 +08001249/*
1250 * put a workspace struct back on the list or free it if we have enough
1251 * idle ones sitting around
1252 */
David Sterbaa3bbd2a2019-10-04 02:50:28 +02001253void btrfs_put_workspace(int type, struct list_head *ws)
Li Zefan261507a02010-12-17 14:21:50 +08001254{
David Sterbaa3bbd2a2019-10-04 02:50:28 +02001255 struct workspace_manager *wsm;
Timofey Titovets4e439a02017-09-28 17:33:36 +03001256 struct list_head *idle_ws;
1257 spinlock_t *ws_lock;
1258 atomic_t *total_ws;
1259 wait_queue_head_t *ws_wait;
1260 int *free_ws;
1261
David Sterbaa3bbd2a2019-10-04 02:50:28 +02001262 wsm = btrfs_compress_op[type]->workspace_manager;
Dennis Zhou92ee55302019-02-04 15:20:03 -05001263 idle_ws = &wsm->idle_ws;
1264 ws_lock = &wsm->ws_lock;
1265 total_ws = &wsm->total_ws;
1266 ws_wait = &wsm->ws_wait;
1267 free_ws = &wsm->free_ws;
Li Zefan261507a02010-12-17 14:21:50 +08001268
Byongho Leed9187642015-10-14 14:05:24 +09001269 spin_lock(ws_lock);
Nick Terrell26b28dc2017-06-29 10:57:26 -07001270 if (*free_ws <= num_online_cpus()) {
Dennis Zhou929f4ba2019-02-04 15:20:02 -05001271 list_add(ws, idle_ws);
David Sterba6ac10a62016-04-27 02:15:15 +02001272 (*free_ws)++;
Byongho Leed9187642015-10-14 14:05:24 +09001273 spin_unlock(ws_lock);
Li Zefan261507a02010-12-17 14:21:50 +08001274 goto wake;
1275 }
Byongho Leed9187642015-10-14 14:05:24 +09001276 spin_unlock(ws_lock);
Li Zefan261507a02010-12-17 14:21:50 +08001277
David Sterba1e002352019-10-04 02:57:22 +02001278 free_workspace(type, ws);
David Sterba6ac10a62016-04-27 02:15:15 +02001279 atomic_dec(total_ws);
Li Zefan261507a02010-12-17 14:21:50 +08001280wake:
David Sterba093258e2018-02-26 16:15:17 +01001281 cond_wake_up(ws_wait);
Li Zefan261507a02010-12-17 14:21:50 +08001282}
1283
Dennis Zhou929f4ba2019-02-04 15:20:02 -05001284static void put_workspace(int type, struct list_head *ws)
1285{
David Sterbabd3a5282019-10-04 02:42:03 +02001286 switch (type) {
David Sterbaa3bbd2a2019-10-04 02:50:28 +02001287 case BTRFS_COMPRESS_NONE: return btrfs_put_workspace(type, ws);
1288 case BTRFS_COMPRESS_ZLIB: return btrfs_put_workspace(type, ws);
1289 case BTRFS_COMPRESS_LZO: return btrfs_put_workspace(type, ws);
David Sterbabd3a5282019-10-04 02:42:03 +02001290 case BTRFS_COMPRESS_ZSTD: return zstd_put_workspace(ws);
1291 default:
1292 /*
1293 * This can't happen, the type is validated several times
1294 * before we get here.
1295 */
1296 BUG();
1297 }
Dennis Zhou929f4ba2019-02-04 15:20:02 -05001298}
1299
Li Zefan261507a02010-12-17 14:21:50 +08001300/*
Anand Jainadbab642020-05-11 22:37:51 -07001301 * Adjust @level according to the limits of the compression algorithm or
1302 * fallback to default
1303 */
1304static unsigned int btrfs_compress_set_level(int type, unsigned level)
1305{
1306 const struct btrfs_compress_op *ops = btrfs_compress_op[type];
1307
1308 if (level == 0)
1309 level = ops->default_level;
1310 else
1311 level = min(level, ops->max_level);
1312
1313 return level;
1314}
1315
1316/*
David Sterba38c31462017-02-14 19:04:07 +01001317 * Given an address space and start and length, compress the bytes into @pages
1318 * that are allocated on demand.
Li Zefan261507a02010-12-17 14:21:50 +08001319 *
David Sterbaf51d2b52017-09-15 17:36:57 +02001320 * @type_level is encoded algorithm and level, where level 0 means whatever
1321 * default the algorithm chooses and is opaque here;
1322 * - compression algo are 0-3
1323 * - the level are bits 4-7
1324 *
David Sterba4d3a8002017-02-14 19:04:07 +01001325 * @out_pages is an in/out parameter, holds maximum number of pages to allocate
1326 * and returns number of actually allocated pages
Li Zefan261507a02010-12-17 14:21:50 +08001327 *
David Sterba38c31462017-02-14 19:04:07 +01001328 * @total_in is used to return the number of bytes actually read. It
1329 * may be smaller than the input length if we had to exit early because we
Li Zefan261507a02010-12-17 14:21:50 +08001330 * ran out of room in the pages array or because we cross the
1331 * max_out threshold.
1332 *
David Sterba38c31462017-02-14 19:04:07 +01001333 * @total_out is an in/out parameter, must be set to the input length and will
1334 * be also used to return the total number of compressed bytes
Li Zefan261507a02010-12-17 14:21:50 +08001335 */
David Sterbaf51d2b52017-09-15 17:36:57 +02001336int btrfs_compress_pages(unsigned int type_level, struct address_space *mapping,
David Sterba38c31462017-02-14 19:04:07 +01001337 u64 start, struct page **pages,
Li Zefan261507a02010-12-17 14:21:50 +08001338 unsigned long *out_pages,
1339 unsigned long *total_in,
David Sterbae5d74902017-02-14 19:45:05 +01001340 unsigned long *total_out)
Li Zefan261507a02010-12-17 14:21:50 +08001341{
Dennis Zhou19727082019-02-04 15:19:57 -05001342 int type = btrfs_compress_type(type_level);
Dennis Zhou7bf49942019-02-04 15:20:04 -05001343 int level = btrfs_compress_level(type_level);
Li Zefan261507a02010-12-17 14:21:50 +08001344 struct list_head *workspace;
1345 int ret;
1346
David Sterbab0c1fe12019-08-09 16:49:06 +02001347 level = btrfs_compress_set_level(type, level);
Dennis Zhou7bf49942019-02-04 15:20:04 -05001348 workspace = get_workspace(type, level);
David Sterba1e4eb742019-10-02 00:06:15 +02001349 ret = compression_compress_pages(type, workspace, mapping, start, pages,
1350 out_pages, total_in, total_out);
Dennis Zhou929f4ba2019-02-04 15:20:02 -05001351 put_workspace(type, workspace);
Li Zefan261507a02010-12-17 14:21:50 +08001352 return ret;
1353}
1354
Anand Jain8140dc32017-05-26 15:44:58 +08001355static int btrfs_decompress_bio(struct compressed_bio *cb)
Li Zefan261507a02010-12-17 14:21:50 +08001356{
1357 struct list_head *workspace;
1358 int ret;
Anand Jain8140dc32017-05-26 15:44:58 +08001359 int type = cb->compress_type;
Li Zefan261507a02010-12-17 14:21:50 +08001360
Dennis Zhou7bf49942019-02-04 15:20:04 -05001361 workspace = get_workspace(type, 0);
David Sterba1e4eb742019-10-02 00:06:15 +02001362 ret = compression_decompress_bio(type, workspace, cb);
Dennis Zhou929f4ba2019-02-04 15:20:02 -05001363 put_workspace(type, workspace);
Anand Jaine1ddce72017-05-26 15:44:59 +08001364
Li Zefan261507a02010-12-17 14:21:50 +08001365 return ret;
1366}
1367
1368/*
1369 * a less complex decompression routine. Our compressed data fits in a
1370 * single page, and we want to read a single page out of it.
1371 * start_byte tells us the offset into the compressed data we're interested in
1372 */
1373int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page,
1374 unsigned long start_byte, size_t srclen, size_t destlen)
1375{
1376 struct list_head *workspace;
1377 int ret;
1378
Dennis Zhou7bf49942019-02-04 15:20:04 -05001379 workspace = get_workspace(type, 0);
David Sterba1e4eb742019-10-02 00:06:15 +02001380 ret = compression_decompress(type, workspace, data_in, dest_page,
1381 start_byte, srclen, destlen);
Dennis Zhou929f4ba2019-02-04 15:20:02 -05001382 put_workspace(type, workspace);
Dennis Zhou7bf49942019-02-04 15:20:04 -05001383
Li Zefan261507a02010-12-17 14:21:50 +08001384 return ret;
1385}
1386
Dennis Zhou1666eda2019-02-04 15:20:01 -05001387void __init btrfs_init_compress(void)
1388{
David Sterbad5517032019-10-02 01:08:03 +02001389 btrfs_init_workspace_manager(BTRFS_COMPRESS_NONE);
1390 btrfs_init_workspace_manager(BTRFS_COMPRESS_ZLIB);
1391 btrfs_init_workspace_manager(BTRFS_COMPRESS_LZO);
1392 zstd_init_workspace_manager();
Dennis Zhou1666eda2019-02-04 15:20:01 -05001393}
1394
David Sterbae67c7182018-02-19 17:24:18 +01001395void __cold btrfs_exit_compress(void)
Li Zefan261507a02010-12-17 14:21:50 +08001396{
David Sterba25103072019-10-02 01:08:03 +02001397 btrfs_cleanup_workspace_manager(BTRFS_COMPRESS_NONE);
1398 btrfs_cleanup_workspace_manager(BTRFS_COMPRESS_ZLIB);
1399 btrfs_cleanup_workspace_manager(BTRFS_COMPRESS_LZO);
1400 zstd_cleanup_workspace_manager();
Li Zefan261507a02010-12-17 14:21:50 +08001401}
Li Zefan3a39c182010-11-08 15:22:19 +08001402
1403/*
Qu Wenruo1c3dc172021-07-05 10:00:58 +08001404 * Copy decompressed data from working buffer to pages.
Li Zefan3a39c182010-11-08 15:22:19 +08001405 *
Qu Wenruo1c3dc172021-07-05 10:00:58 +08001406 * @buf: The decompressed data buffer
1407 * @buf_len: The decompressed data length
1408 * @decompressed: Number of bytes that are already decompressed inside the
1409 * compressed extent
1410 * @cb: The compressed extent descriptor
1411 * @orig_bio: The original bio that the caller wants to read for
Li Zefan3a39c182010-11-08 15:22:19 +08001412 *
Qu Wenruo1c3dc172021-07-05 10:00:58 +08001413 * An easier to understand graph is like below:
1414 *
1415 * |<- orig_bio ->| |<- orig_bio->|
1416 * |<------- full decompressed extent ----->|
1417 * |<----------- @cb range ---->|
1418 * | |<-- @buf_len -->|
1419 * |<--- @decompressed --->|
1420 *
1421 * Note that, @cb can be a subpage of the full decompressed extent, but
1422 * @cb->start always has the same as the orig_file_offset value of the full
1423 * decompressed extent.
1424 *
1425 * When reading compressed extent, we have to read the full compressed extent,
1426 * while @orig_bio may only want part of the range.
1427 * Thus this function will ensure only data covered by @orig_bio will be copied
1428 * to.
1429 *
1430 * Return 0 if we have copied all needed contents for @orig_bio.
1431 * Return >0 if we need continue decompress.
Li Zefan3a39c182010-11-08 15:22:19 +08001432 */
Qu Wenruo1c3dc172021-07-05 10:00:58 +08001433int btrfs_decompress_buf2page(const char *buf, u32 buf_len,
1434 struct compressed_bio *cb, u32 decompressed)
Li Zefan3a39c182010-11-08 15:22:19 +08001435{
Qu Wenruo1c3dc172021-07-05 10:00:58 +08001436 struct bio *orig_bio = cb->orig_bio;
1437 /* Offset inside the full decompressed extent */
1438 u32 cur_offset;
Li Zefan3a39c182010-11-08 15:22:19 +08001439
Qu Wenruo1c3dc172021-07-05 10:00:58 +08001440 cur_offset = decompressed;
1441 /* The main loop to do the copy */
1442 while (cur_offset < decompressed + buf_len) {
1443 struct bio_vec bvec;
1444 size_t copy_len;
1445 u32 copy_start;
1446 /* Offset inside the full decompressed extent */
1447 u32 bvec_offset;
Li Zefan3a39c182010-11-08 15:22:19 +08001448
Qu Wenruo1c3dc172021-07-05 10:00:58 +08001449 bvec = bio_iter_iovec(orig_bio, orig_bio->bi_iter);
1450 /*
1451 * cb->start may underflow, but subtracting that value can still
1452 * give us correct offset inside the full decompressed extent.
1453 */
1454 bvec_offset = page_offset(bvec.bv_page) + bvec.bv_offset - cb->start;
Li Zefan3a39c182010-11-08 15:22:19 +08001455
Qu Wenruo1c3dc172021-07-05 10:00:58 +08001456 /* Haven't reached the bvec range, exit */
1457 if (decompressed + buf_len <= bvec_offset)
1458 return 1;
Li Zefan3a39c182010-11-08 15:22:19 +08001459
Qu Wenruo1c3dc172021-07-05 10:00:58 +08001460 copy_start = max(cur_offset, bvec_offset);
1461 copy_len = min(bvec_offset + bvec.bv_len,
1462 decompressed + buf_len) - copy_start;
1463 ASSERT(copy_len);
Li Zefan3a39c182010-11-08 15:22:19 +08001464
Christoph Hellwig974b1ad2016-11-25 09:07:46 +01001465 /*
Qu Wenruo1c3dc172021-07-05 10:00:58 +08001466 * Extra range check to ensure we didn't go beyond
1467 * @buf + @buf_len.
Christoph Hellwig974b1ad2016-11-25 09:07:46 +01001468 */
Qu Wenruo1c3dc172021-07-05 10:00:58 +08001469 ASSERT(copy_start - decompressed < buf_len);
1470 memcpy_to_page(bvec.bv_page, bvec.bv_offset,
1471 buf + copy_start - decompressed, copy_len);
1472 flush_dcache_page(bvec.bv_page);
1473 cur_offset += copy_len;
Li Zefan3a39c182010-11-08 15:22:19 +08001474
Qu Wenruo1c3dc172021-07-05 10:00:58 +08001475 bio_advance(orig_bio, copy_len);
1476 /* Finished the bio */
1477 if (!orig_bio->bi_iter.bi_size)
1478 return 0;
Li Zefan3a39c182010-11-08 15:22:19 +08001479 }
Li Zefan3a39c182010-11-08 15:22:19 +08001480 return 1;
1481}
Timofey Titovetsc2fcdcd2017-07-17 16:52:58 +03001482
Timofey Titovets19562432017-10-08 16:11:59 +03001483/*
1484 * Shannon Entropy calculation
1485 *
Andrea Gelmini52042d82018-11-28 12:05:13 +01001486 * Pure byte distribution analysis fails to determine compressibility of data.
Timofey Titovets19562432017-10-08 16:11:59 +03001487 * Try calculating entropy to estimate the average minimum number of bits
1488 * needed to encode the sampled data.
1489 *
1490 * For convenience, return the percentage of needed bits, instead of amount of
1491 * bits directly.
1492 *
1493 * @ENTROPY_LVL_ACEPTABLE - below that threshold, sample has low byte entropy
1494 * and can be compressible with high probability
1495 *
1496 * @ENTROPY_LVL_HIGH - data are not compressible with high probability
1497 *
1498 * Use of ilog2() decreases precision, we lower the LVL to 5 to compensate.
1499 */
1500#define ENTROPY_LVL_ACEPTABLE (65)
1501#define ENTROPY_LVL_HIGH (80)
1502
1503/*
1504 * For increasead precision in shannon_entropy calculation,
1505 * let's do pow(n, M) to save more digits after comma:
1506 *
1507 * - maximum int bit length is 64
1508 * - ilog2(MAX_SAMPLE_SIZE) -> 13
1509 * - 13 * 4 = 52 < 64 -> M = 4
1510 *
1511 * So use pow(n, 4).
1512 */
1513static inline u32 ilog2_w(u64 n)
1514{
1515 return ilog2(n * n * n * n);
1516}
1517
1518static u32 shannon_entropy(struct heuristic_ws *ws)
1519{
1520 const u32 entropy_max = 8 * ilog2_w(2);
1521 u32 entropy_sum = 0;
1522 u32 p, p_base, sz_base;
1523 u32 i;
1524
1525 sz_base = ilog2_w(ws->sample_size);
1526 for (i = 0; i < BUCKET_SIZE && ws->bucket[i].count > 0; i++) {
1527 p = ws->bucket[i].count;
1528 p_base = ilog2_w(p);
1529 entropy_sum += p * (sz_base - p_base);
1530 }
1531
1532 entropy_sum /= ws->sample_size;
1533 return entropy_sum * 100 / entropy_max;
1534}
1535
Timofey Titovets440c8402017-12-04 00:30:33 +03001536#define RADIX_BASE 4U
1537#define COUNTERS_SIZE (1U << RADIX_BASE)
Timofey Titovets858177d2017-09-28 17:33:41 +03001538
Timofey Titovets440c8402017-12-04 00:30:33 +03001539static u8 get4bits(u64 num, int shift) {
1540 u8 low4bits;
1541
1542 num >>= shift;
1543 /* Reverse order */
1544 low4bits = (COUNTERS_SIZE - 1) - (num % COUNTERS_SIZE);
1545 return low4bits;
1546}
1547
Timofey Titovets440c8402017-12-04 00:30:33 +03001548/*
1549 * Use 4 bits as radix base
Andrea Gelmini52042d82018-11-28 12:05:13 +01001550 * Use 16 u32 counters for calculating new position in buf array
Timofey Titovets440c8402017-12-04 00:30:33 +03001551 *
1552 * @array - array that will be sorted
1553 * @array_buf - buffer array to store sorting results
1554 * must be equal in size to @array
1555 * @num - array size
Timofey Titovets440c8402017-12-04 00:30:33 +03001556 */
David Sterba23ae8c62017-12-12 20:35:02 +01001557static void radix_sort(struct bucket_item *array, struct bucket_item *array_buf,
David Sterba36243c92017-12-12 20:35:02 +01001558 int num)
Timofey Titovets440c8402017-12-04 00:30:33 +03001559{
1560 u64 max_num;
1561 u64 buf_num;
1562 u32 counters[COUNTERS_SIZE];
1563 u32 new_addr;
1564 u32 addr;
1565 int bitlen;
1566 int shift;
1567 int i;
1568
1569 /*
1570 * Try avoid useless loop iterations for small numbers stored in big
1571 * counters. Example: 48 33 4 ... in 64bit array
1572 */
David Sterba23ae8c62017-12-12 20:35:02 +01001573 max_num = array[0].count;
Timofey Titovets440c8402017-12-04 00:30:33 +03001574 for (i = 1; i < num; i++) {
David Sterba23ae8c62017-12-12 20:35:02 +01001575 buf_num = array[i].count;
Timofey Titovets440c8402017-12-04 00:30:33 +03001576 if (buf_num > max_num)
1577 max_num = buf_num;
1578 }
1579
1580 buf_num = ilog2(max_num);
1581 bitlen = ALIGN(buf_num, RADIX_BASE * 2);
1582
1583 shift = 0;
1584 while (shift < bitlen) {
1585 memset(counters, 0, sizeof(counters));
1586
1587 for (i = 0; i < num; i++) {
David Sterba23ae8c62017-12-12 20:35:02 +01001588 buf_num = array[i].count;
Timofey Titovets440c8402017-12-04 00:30:33 +03001589 addr = get4bits(buf_num, shift);
1590 counters[addr]++;
1591 }
1592
1593 for (i = 1; i < COUNTERS_SIZE; i++)
1594 counters[i] += counters[i - 1];
1595
1596 for (i = num - 1; i >= 0; i--) {
David Sterba23ae8c62017-12-12 20:35:02 +01001597 buf_num = array[i].count;
Timofey Titovets440c8402017-12-04 00:30:33 +03001598 addr = get4bits(buf_num, shift);
1599 counters[addr]--;
1600 new_addr = counters[addr];
David Sterba7add17b2017-12-12 20:35:02 +01001601 array_buf[new_addr] = array[i];
Timofey Titovets440c8402017-12-04 00:30:33 +03001602 }
1603
1604 shift += RADIX_BASE;
1605
1606 /*
1607 * Normal radix expects to move data from a temporary array, to
1608 * the main one. But that requires some CPU time. Avoid that
1609 * by doing another sort iteration to original array instead of
1610 * memcpy()
1611 */
1612 memset(counters, 0, sizeof(counters));
1613
1614 for (i = 0; i < num; i ++) {
David Sterba23ae8c62017-12-12 20:35:02 +01001615 buf_num = array_buf[i].count;
Timofey Titovets440c8402017-12-04 00:30:33 +03001616 addr = get4bits(buf_num, shift);
1617 counters[addr]++;
1618 }
1619
1620 for (i = 1; i < COUNTERS_SIZE; i++)
1621 counters[i] += counters[i - 1];
1622
1623 for (i = num - 1; i >= 0; i--) {
David Sterba23ae8c62017-12-12 20:35:02 +01001624 buf_num = array_buf[i].count;
Timofey Titovets440c8402017-12-04 00:30:33 +03001625 addr = get4bits(buf_num, shift);
1626 counters[addr]--;
1627 new_addr = counters[addr];
David Sterba7add17b2017-12-12 20:35:02 +01001628 array[new_addr] = array_buf[i];
Timofey Titovets440c8402017-12-04 00:30:33 +03001629 }
1630
1631 shift += RADIX_BASE;
1632 }
Timofey Titovets858177d2017-09-28 17:33:41 +03001633}
1634
1635/*
1636 * Size of the core byte set - how many bytes cover 90% of the sample
1637 *
1638 * There are several types of structured binary data that use nearly all byte
1639 * values. The distribution can be uniform and counts in all buckets will be
1640 * nearly the same (eg. encrypted data). Unlikely to be compressible.
1641 *
1642 * Other possibility is normal (Gaussian) distribution, where the data could
1643 * be potentially compressible, but we have to take a few more steps to decide
1644 * how much.
1645 *
1646 * @BYTE_CORE_SET_LOW - main part of byte values repeated frequently,
1647 * compression algo can easy fix that
1648 * @BYTE_CORE_SET_HIGH - data have uniform distribution and with high
1649 * probability is not compressible
1650 */
1651#define BYTE_CORE_SET_LOW (64)
1652#define BYTE_CORE_SET_HIGH (200)
1653
1654static int byte_core_set_size(struct heuristic_ws *ws)
1655{
1656 u32 i;
1657 u32 coreset_sum = 0;
1658 const u32 core_set_threshold = ws->sample_size * 90 / 100;
1659 struct bucket_item *bucket = ws->bucket;
1660
1661 /* Sort in reverse order */
David Sterba36243c92017-12-12 20:35:02 +01001662 radix_sort(ws->bucket, ws->bucket_b, BUCKET_SIZE);
Timofey Titovets858177d2017-09-28 17:33:41 +03001663
1664 for (i = 0; i < BYTE_CORE_SET_LOW; i++)
1665 coreset_sum += bucket[i].count;
1666
1667 if (coreset_sum > core_set_threshold)
1668 return i;
1669
1670 for (; i < BYTE_CORE_SET_HIGH && bucket[i].count > 0; i++) {
1671 coreset_sum += bucket[i].count;
1672 if (coreset_sum > core_set_threshold)
1673 break;
1674 }
1675
1676 return i;
1677}
1678
Timofey Titovetsa288e922017-09-28 17:33:40 +03001679/*
1680 * Count byte values in buckets.
1681 * This heuristic can detect textual data (configs, xml, json, html, etc).
1682 * Because in most text-like data byte set is restricted to limited number of
1683 * possible characters, and that restriction in most cases makes data easy to
1684 * compress.
1685 *
1686 * @BYTE_SET_THRESHOLD - consider all data within this byte set size:
1687 * less - compressible
1688 * more - need additional analysis
1689 */
1690#define BYTE_SET_THRESHOLD (64)
1691
1692static u32 byte_set_size(const struct heuristic_ws *ws)
1693{
1694 u32 i;
1695 u32 byte_set_size = 0;
1696
1697 for (i = 0; i < BYTE_SET_THRESHOLD; i++) {
1698 if (ws->bucket[i].count > 0)
1699 byte_set_size++;
1700 }
1701
1702 /*
1703 * Continue collecting count of byte values in buckets. If the byte
1704 * set size is bigger then the threshold, it's pointless to continue,
1705 * the detection technique would fail for this type of data.
1706 */
1707 for (; i < BUCKET_SIZE; i++) {
1708 if (ws->bucket[i].count > 0) {
1709 byte_set_size++;
1710 if (byte_set_size > BYTE_SET_THRESHOLD)
1711 return byte_set_size;
1712 }
1713 }
1714
1715 return byte_set_size;
1716}
1717
Timofey Titovets1fe4f6f2017-09-28 17:33:39 +03001718static bool sample_repeated_patterns(struct heuristic_ws *ws)
1719{
1720 const u32 half_of_sample = ws->sample_size / 2;
1721 const u8 *data = ws->sample;
1722
1723 return memcmp(&data[0], &data[half_of_sample], half_of_sample) == 0;
1724}
1725
Timofey Titovetsa440d482017-09-28 17:33:38 +03001726static void heuristic_collect_sample(struct inode *inode, u64 start, u64 end,
1727 struct heuristic_ws *ws)
1728{
1729 struct page *page;
1730 u64 index, index_end;
1731 u32 i, curr_sample_pos;
1732 u8 *in_data;
1733
1734 /*
1735 * Compression handles the input data by chunks of 128KiB
1736 * (defined by BTRFS_MAX_UNCOMPRESSED)
1737 *
1738 * We do the same for the heuristic and loop over the whole range.
1739 *
1740 * MAX_SAMPLE_SIZE - calculated under assumption that heuristic will
1741 * process no more than BTRFS_MAX_UNCOMPRESSED at a time.
1742 */
1743 if (end - start > BTRFS_MAX_UNCOMPRESSED)
1744 end = start + BTRFS_MAX_UNCOMPRESSED;
1745
1746 index = start >> PAGE_SHIFT;
1747 index_end = end >> PAGE_SHIFT;
1748
1749 /* Don't miss unaligned end */
1750 if (!IS_ALIGNED(end, PAGE_SIZE))
1751 index_end++;
1752
1753 curr_sample_pos = 0;
1754 while (index < index_end) {
1755 page = find_get_page(inode->i_mapping, index);
Ira Weiny58c1a352021-02-16 18:48:23 -08001756 in_data = kmap_local_page(page);
Timofey Titovetsa440d482017-09-28 17:33:38 +03001757 /* Handle case where the start is not aligned to PAGE_SIZE */
1758 i = start % PAGE_SIZE;
1759 while (i < PAGE_SIZE - SAMPLING_READ_SIZE) {
1760 /* Don't sample any garbage from the last page */
1761 if (start > end - SAMPLING_READ_SIZE)
1762 break;
1763 memcpy(&ws->sample[curr_sample_pos], &in_data[i],
1764 SAMPLING_READ_SIZE);
1765 i += SAMPLING_INTERVAL;
1766 start += SAMPLING_INTERVAL;
1767 curr_sample_pos += SAMPLING_READ_SIZE;
1768 }
Ira Weiny58c1a352021-02-16 18:48:23 -08001769 kunmap_local(in_data);
Timofey Titovetsa440d482017-09-28 17:33:38 +03001770 put_page(page);
1771
1772 index++;
1773 }
1774
1775 ws->sample_size = curr_sample_pos;
1776}
1777
Timofey Titovetsc2fcdcd2017-07-17 16:52:58 +03001778/*
1779 * Compression heuristic.
1780 *
1781 * For now is's a naive and optimistic 'return true', we'll extend the logic to
1782 * quickly (compared to direct compression) detect data characteristics
1783 * (compressible/uncompressible) to avoid wasting CPU time on uncompressible
1784 * data.
1785 *
1786 * The following types of analysis can be performed:
1787 * - detect mostly zero data
1788 * - detect data with low "byte set" size (text, etc)
1789 * - detect data with low/high "core byte" set
1790 *
1791 * Return non-zero if the compression should be done, 0 otherwise.
1792 */
1793int btrfs_compress_heuristic(struct inode *inode, u64 start, u64 end)
1794{
Dennis Zhou7bf49942019-02-04 15:20:04 -05001795 struct list_head *ws_list = get_workspace(0, 0);
Timofey Titovets4e439a02017-09-28 17:33:36 +03001796 struct heuristic_ws *ws;
Timofey Titovetsa440d482017-09-28 17:33:38 +03001797 u32 i;
1798 u8 byte;
Timofey Titovets19562432017-10-08 16:11:59 +03001799 int ret = 0;
Timofey Titovetsc2fcdcd2017-07-17 16:52:58 +03001800
Timofey Titovets4e439a02017-09-28 17:33:36 +03001801 ws = list_entry(ws_list, struct heuristic_ws, list);
1802
Timofey Titovetsa440d482017-09-28 17:33:38 +03001803 heuristic_collect_sample(inode, start, end, ws);
1804
Timofey Titovets1fe4f6f2017-09-28 17:33:39 +03001805 if (sample_repeated_patterns(ws)) {
1806 ret = 1;
1807 goto out;
1808 }
1809
Timofey Titovetsa440d482017-09-28 17:33:38 +03001810 memset(ws->bucket, 0, sizeof(*ws->bucket)*BUCKET_SIZE);
1811
1812 for (i = 0; i < ws->sample_size; i++) {
1813 byte = ws->sample[i];
1814 ws->bucket[byte].count++;
Timofey Titovetsc2fcdcd2017-07-17 16:52:58 +03001815 }
1816
Timofey Titovetsa288e922017-09-28 17:33:40 +03001817 i = byte_set_size(ws);
1818 if (i < BYTE_SET_THRESHOLD) {
1819 ret = 2;
1820 goto out;
1821 }
1822
Timofey Titovets858177d2017-09-28 17:33:41 +03001823 i = byte_core_set_size(ws);
1824 if (i <= BYTE_CORE_SET_LOW) {
1825 ret = 3;
1826 goto out;
1827 }
1828
1829 if (i >= BYTE_CORE_SET_HIGH) {
1830 ret = 0;
1831 goto out;
1832 }
1833
Timofey Titovets19562432017-10-08 16:11:59 +03001834 i = shannon_entropy(ws);
1835 if (i <= ENTROPY_LVL_ACEPTABLE) {
1836 ret = 4;
1837 goto out;
1838 }
1839
1840 /*
1841 * For the levels below ENTROPY_LVL_HIGH, additional analysis would be
1842 * needed to give green light to compression.
1843 *
1844 * For now just assume that compression at that level is not worth the
1845 * resources because:
1846 *
1847 * 1. it is possible to defrag the data later
1848 *
1849 * 2. the data would turn out to be hardly compressible, eg. 150 byte
1850 * values, every bucket has counter at level ~54. The heuristic would
1851 * be confused. This can happen when data have some internal repeated
1852 * patterns like "abbacbbc...". This can be detected by analyzing
1853 * pairs of bytes, which is too costly.
1854 */
1855 if (i < ENTROPY_LVL_HIGH) {
1856 ret = 5;
1857 goto out;
1858 } else {
1859 ret = 0;
1860 goto out;
1861 }
1862
Timofey Titovets1fe4f6f2017-09-28 17:33:39 +03001863out:
Dennis Zhou929f4ba2019-02-04 15:20:02 -05001864 put_workspace(0, ws_list);
Timofey Titovetsc2fcdcd2017-07-17 16:52:58 +03001865 return ret;
1866}
David Sterbaf51d2b52017-09-15 17:36:57 +02001867
Dennis Zhoud0ab62c2019-02-04 15:20:05 -05001868/*
1869 * Convert the compression suffix (eg. after "zlib" starting with ":") to
1870 * level, unrecognized string will set the default level
1871 */
1872unsigned int btrfs_compress_str2level(unsigned int type, const char *str)
David Sterbaf51d2b52017-09-15 17:36:57 +02001873{
Dennis Zhoud0ab62c2019-02-04 15:20:05 -05001874 unsigned int level = 0;
1875 int ret;
1876
1877 if (!type)
David Sterbaf51d2b52017-09-15 17:36:57 +02001878 return 0;
1879
Dennis Zhoud0ab62c2019-02-04 15:20:05 -05001880 if (str[0] == ':') {
1881 ret = kstrtouint(str + 1, 10, &level);
1882 if (ret)
1883 level = 0;
1884 }
David Sterbaf51d2b52017-09-15 17:36:57 +02001885
David Sterbab0c1fe12019-08-09 16:49:06 +02001886 level = btrfs_compress_set_level(type, level);
1887
1888 return level;
1889}