1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2008 Oracle. All rights reserved.
4 */
5
6 #include <linux/kernel.h>
7 #include <linux/bio.h>
8 #include <linux/file.h>
9 #include <linux/fs.h>
10 #include <linux/pagemap.h>
11 #include <linux/pagevec.h>
12 #include <linux/highmem.h>
13 #include <linux/kthread.h>
14 #include <linux/time.h>
15 #include <linux/init.h>
16 #include <linux/string.h>
17 #include <linux/backing-dev.h>
18 #include <linux/writeback.h>
19 #include <linux/psi.h>
20 #include <linux/slab.h>
21 #include <linux/sched/mm.h>
22 #include <linux/log2.h>
23 #include <linux/shrinker.h>
24 #include <crypto/hash.h>
25 #include "misc.h"
26 #include "ctree.h"
27 #include "fs.h"
28 #include "btrfs_inode.h"
29 #include "bio.h"
30 #include "ordered-data.h"
31 #include "compression.h"
32 #include "extent_io.h"
33 #include "extent_map.h"
34 #include "subpage.h"
35 #include "messages.h"
36 #include "super.h"
37
38 static struct bio_set btrfs_compressed_bioset;
39
40 static const char* const btrfs_compress_types[] = { "", "zlib", "lzo", "zstd" };
41
btrfs_compress_type2str(enum btrfs_compression_type type)42 const char* btrfs_compress_type2str(enum btrfs_compression_type type)
43 {
44 switch (type) {
45 case BTRFS_COMPRESS_ZLIB:
46 case BTRFS_COMPRESS_LZO:
47 case BTRFS_COMPRESS_ZSTD:
48 case BTRFS_COMPRESS_NONE:
49 return btrfs_compress_types[type];
50 default:
51 break;
52 }
53
54 return NULL;
55 }
56
to_compressed_bio(struct btrfs_bio * bbio)57 static inline struct compressed_bio *to_compressed_bio(struct btrfs_bio *bbio)
58 {
59 return container_of(bbio, struct compressed_bio, bbio);
60 }
61
alloc_compressed_bio(struct btrfs_inode * inode,u64 start,blk_opf_t op,btrfs_bio_end_io_t end_io)62 static struct compressed_bio *alloc_compressed_bio(struct btrfs_inode *inode,
63 u64 start, blk_opf_t op,
64 btrfs_bio_end_io_t end_io)
65 {
66 struct btrfs_bio *bbio;
67
68 bbio = btrfs_bio(bio_alloc_bioset(NULL, BTRFS_MAX_COMPRESSED_PAGES, op,
69 GFP_NOFS, &btrfs_compressed_bioset));
70 btrfs_bio_init(bbio, inode->root->fs_info, end_io, NULL);
71 bbio->inode = inode;
72 bbio->file_offset = start;
73 return to_compressed_bio(bbio);
74 }
75
btrfs_compress_is_valid_type(const char * str,size_t len)76 bool btrfs_compress_is_valid_type(const char *str, size_t len)
77 {
78 int i;
79
80 for (i = 1; i < ARRAY_SIZE(btrfs_compress_types); i++) {
81 size_t comp_len = strlen(btrfs_compress_types[i]);
82
83 if (len < comp_len)
84 continue;
85
86 if (!strncmp(btrfs_compress_types[i], str, comp_len))
87 return true;
88 }
89 return false;
90 }
91
compression_compress_pages(int type,struct list_head * ws,struct address_space * mapping,u64 start,struct folio ** folios,unsigned long * out_folios,unsigned long * total_in,unsigned long * total_out)92 static int compression_compress_pages(int type, struct list_head *ws,
93 struct address_space *mapping, u64 start,
94 struct folio **folios, unsigned long *out_folios,
95 unsigned long *total_in, unsigned long *total_out)
96 {
97 switch (type) {
98 case BTRFS_COMPRESS_ZLIB:
99 return zlib_compress_folios(ws, mapping, start, folios,
100 out_folios, total_in, total_out);
101 case BTRFS_COMPRESS_LZO:
102 return lzo_compress_folios(ws, mapping, start, folios,
103 out_folios, total_in, total_out);
104 case BTRFS_COMPRESS_ZSTD:
105 return zstd_compress_folios(ws, mapping, start, folios,
106 out_folios, total_in, total_out);
107 case BTRFS_COMPRESS_NONE:
108 default:
109 /*
110 * This can happen when compression races with remount setting
111 * it to 'no compress', while caller doesn't call
112 * inode_need_compress() to check if we really need to
113 * compress.
114 *
115 * Not a big deal, just need to inform caller that we
116 * haven't allocated any pages yet.
117 */
118 *out_folios = 0;
119 return -E2BIG;
120 }
121 }
122
compression_decompress_bio(struct list_head * ws,struct compressed_bio * cb)123 static int compression_decompress_bio(struct list_head *ws,
124 struct compressed_bio *cb)
125 {
126 switch (cb->compress_type) {
127 case BTRFS_COMPRESS_ZLIB: return zlib_decompress_bio(ws, cb);
128 case BTRFS_COMPRESS_LZO: return lzo_decompress_bio(ws, cb);
129 case BTRFS_COMPRESS_ZSTD: return zstd_decompress_bio(ws, cb);
130 case BTRFS_COMPRESS_NONE:
131 default:
132 /*
133 * This can't happen, the type is validated several times
134 * before we get here.
135 */
136 BUG();
137 }
138 }
139
compression_decompress(int type,struct list_head * ws,const u8 * data_in,struct folio * dest_folio,unsigned long dest_pgoff,size_t srclen,size_t destlen)140 static int compression_decompress(int type, struct list_head *ws,
141 const u8 *data_in, struct folio *dest_folio,
142 unsigned long dest_pgoff, size_t srclen, size_t destlen)
143 {
144 switch (type) {
145 case BTRFS_COMPRESS_ZLIB: return zlib_decompress(ws, data_in, dest_folio,
146 dest_pgoff, srclen, destlen);
147 case BTRFS_COMPRESS_LZO: return lzo_decompress(ws, data_in, dest_folio,
148 dest_pgoff, srclen, destlen);
149 case BTRFS_COMPRESS_ZSTD: return zstd_decompress(ws, data_in, dest_folio,
150 dest_pgoff, srclen, destlen);
151 case BTRFS_COMPRESS_NONE:
152 default:
153 /*
154 * This can't happen, the type is validated several times
155 * before we get here.
156 */
157 BUG();
158 }
159 }
160
btrfs_free_compressed_folios(struct compressed_bio * cb)161 static void btrfs_free_compressed_folios(struct compressed_bio *cb)
162 {
163 for (unsigned int i = 0; i < cb->nr_folios; i++)
164 btrfs_free_compr_folio(cb->compressed_folios[i]);
165 kfree(cb->compressed_folios);
166 }
167
168 static int btrfs_decompress_bio(struct compressed_bio *cb);
169
170 /*
171 * Global cache of last unused pages for compression/decompression.
172 */
173 static struct btrfs_compr_pool {
174 struct shrinker *shrinker;
175 spinlock_t lock;
176 struct list_head list;
177 int count;
178 int thresh;
179 } compr_pool;
180
btrfs_compr_pool_count(struct shrinker * sh,struct shrink_control * sc)181 static unsigned long btrfs_compr_pool_count(struct shrinker *sh, struct shrink_control *sc)
182 {
183 int ret;
184
185 /*
186 * We must not read the values more than once if 'ret' gets expanded in
187 * the return statement so we don't accidentally return a negative
188 * number, even if the first condition finds it positive.
189 */
190 ret = READ_ONCE(compr_pool.count) - READ_ONCE(compr_pool.thresh);
191
192 return ret > 0 ? ret : 0;
193 }
194
btrfs_compr_pool_scan(struct shrinker * sh,struct shrink_control * sc)195 static unsigned long btrfs_compr_pool_scan(struct shrinker *sh, struct shrink_control *sc)
196 {
197 struct list_head remove;
198 struct list_head *tmp, *next;
199 int freed;
200
201 if (compr_pool.count == 0)
202 return SHRINK_STOP;
203
204 INIT_LIST_HEAD(&remove);
205
206 /* For now, just simply drain the whole list. */
207 spin_lock(&compr_pool.lock);
208 list_splice_init(&compr_pool.list, &remove);
209 freed = compr_pool.count;
210 compr_pool.count = 0;
211 spin_unlock(&compr_pool.lock);
212
213 list_for_each_safe(tmp, next, &remove) {
214 struct page *page = list_entry(tmp, struct page, lru);
215
216 ASSERT(page_ref_count(page) == 1);
217 put_page(page);
218 }
219
220 return freed;
221 }
222
223 /*
224 * Common wrappers for page allocation from compression wrappers
225 */
btrfs_alloc_compr_folio(void)226 struct folio *btrfs_alloc_compr_folio(void)
227 {
228 struct folio *folio = NULL;
229
230 spin_lock(&compr_pool.lock);
231 if (compr_pool.count > 0) {
232 folio = list_first_entry(&compr_pool.list, struct folio, lru);
233 list_del_init(&folio->lru);
234 compr_pool.count--;
235 }
236 spin_unlock(&compr_pool.lock);
237
238 if (folio)
239 return folio;
240
241 return folio_alloc(GFP_NOFS, 0);
242 }
243
btrfs_free_compr_folio(struct folio * folio)244 void btrfs_free_compr_folio(struct folio *folio)
245 {
246 bool do_free = false;
247
248 spin_lock(&compr_pool.lock);
249 if (compr_pool.count > compr_pool.thresh) {
250 do_free = true;
251 } else {
252 list_add(&folio->lru, &compr_pool.list);
253 compr_pool.count++;
254 }
255 spin_unlock(&compr_pool.lock);
256
257 if (!do_free)
258 return;
259
260 ASSERT(folio_ref_count(folio) == 1);
261 folio_put(folio);
262 }
263
end_bbio_compressed_read(struct btrfs_bio * bbio)264 static void end_bbio_compressed_read(struct btrfs_bio *bbio)
265 {
266 struct compressed_bio *cb = to_compressed_bio(bbio);
267 blk_status_t status = bbio->bio.bi_status;
268
269 if (!status)
270 status = errno_to_blk_status(btrfs_decompress_bio(cb));
271
272 btrfs_free_compressed_folios(cb);
273 btrfs_bio_end_io(cb->orig_bbio, status);
274 bio_put(&bbio->bio);
275 }
276
277 /*
278 * Clear the writeback bits on all of the file
279 * pages for a compressed write
280 */
end_compressed_writeback(const struct compressed_bio * cb)281 static noinline void end_compressed_writeback(const struct compressed_bio *cb)
282 {
283 struct inode *inode = &cb->bbio.inode->vfs_inode;
284 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
285 unsigned long index = cb->start >> PAGE_SHIFT;
286 unsigned long end_index = (cb->start + cb->len - 1) >> PAGE_SHIFT;
287 struct folio_batch fbatch;
288 const int error = blk_status_to_errno(cb->bbio.bio.bi_status);
289 int i;
290 int ret;
291
292 if (error)
293 mapping_set_error(inode->i_mapping, error);
294
295 folio_batch_init(&fbatch);
296 while (index <= end_index) {
297 ret = filemap_get_folios(inode->i_mapping, &index, end_index,
298 &fbatch);
299
300 if (ret == 0)
301 return;
302
303 for (i = 0; i < ret; i++) {
304 struct folio *folio = fbatch.folios[i];
305
306 btrfs_folio_clamp_clear_writeback(fs_info, folio,
307 cb->start, cb->len);
308 }
309 folio_batch_release(&fbatch);
310 }
311 /* the inode may be gone now */
312 }
313
btrfs_finish_compressed_write_work(struct work_struct * work)314 static void btrfs_finish_compressed_write_work(struct work_struct *work)
315 {
316 struct compressed_bio *cb =
317 container_of(work, struct compressed_bio, write_end_work);
318
319 btrfs_finish_ordered_extent(cb->bbio.ordered, NULL, cb->start, cb->len,
320 cb->bbio.bio.bi_status == BLK_STS_OK);
321
322 if (cb->writeback)
323 end_compressed_writeback(cb);
324 /* Note, our inode could be gone now */
325
326 btrfs_free_compressed_folios(cb);
327 bio_put(&cb->bbio.bio);
328 }
329
330 /*
331 * Do the cleanup once all the compressed pages hit the disk. This will clear
332 * writeback on the file pages and free the compressed pages.
333 *
334 * This also calls the writeback end hooks for the file pages so that metadata
335 * and checksums can be updated in the file.
336 */
end_bbio_compressed_write(struct btrfs_bio * bbio)337 static void end_bbio_compressed_write(struct btrfs_bio *bbio)
338 {
339 struct compressed_bio *cb = to_compressed_bio(bbio);
340 struct btrfs_fs_info *fs_info = bbio->inode->root->fs_info;
341
342 queue_work(fs_info->compressed_write_workers, &cb->write_end_work);
343 }
344
btrfs_add_compressed_bio_folios(struct compressed_bio * cb)345 static void btrfs_add_compressed_bio_folios(struct compressed_bio *cb)
346 {
347 struct bio *bio = &cb->bbio.bio;
348 u32 offset = 0;
349
350 while (offset < cb->compressed_len) {
351 int ret;
352 u32 len = min_t(u32, cb->compressed_len - offset, PAGE_SIZE);
353
354 /* Maximum compressed extent is smaller than bio size limit. */
355 ret = bio_add_folio(bio, cb->compressed_folios[offset >> PAGE_SHIFT],
356 len, 0);
357 ASSERT(ret);
358 offset += len;
359 }
360 }
361
362 /*
363 * worker function to build and submit bios for previously compressed pages.
364 * The corresponding pages in the inode should be marked for writeback
365 * and the compressed pages should have a reference on them for dropping
366 * when the IO is complete.
367 *
368 * This also checksums the file bytes and gets things ready for
369 * the end io hooks.
370 */
btrfs_submit_compressed_write(struct btrfs_ordered_extent * ordered,struct folio ** compressed_folios,unsigned int nr_folios,blk_opf_t write_flags,bool writeback)371 void btrfs_submit_compressed_write(struct btrfs_ordered_extent *ordered,
372 struct folio **compressed_folios,
373 unsigned int nr_folios,
374 blk_opf_t write_flags,
375 bool writeback)
376 {
377 struct btrfs_inode *inode = ordered->inode;
378 struct btrfs_fs_info *fs_info = inode->root->fs_info;
379 struct compressed_bio *cb;
380
381 ASSERT(IS_ALIGNED(ordered->file_offset, fs_info->sectorsize));
382 ASSERT(IS_ALIGNED(ordered->num_bytes, fs_info->sectorsize));
383
384 cb = alloc_compressed_bio(inode, ordered->file_offset,
385 REQ_OP_WRITE | write_flags,
386 end_bbio_compressed_write);
387 cb->start = ordered->file_offset;
388 cb->len = ordered->num_bytes;
389 cb->compressed_folios = compressed_folios;
390 cb->compressed_len = ordered->disk_num_bytes;
391 cb->writeback = writeback;
392 INIT_WORK(&cb->write_end_work, btrfs_finish_compressed_write_work);
393 cb->nr_folios = nr_folios;
394 cb->bbio.bio.bi_iter.bi_sector = ordered->disk_bytenr >> SECTOR_SHIFT;
395 cb->bbio.ordered = ordered;
396 btrfs_add_compressed_bio_folios(cb);
397
398 btrfs_submit_bbio(&cb->bbio, 0);
399 }
400
401 /*
402 * Add extra pages in the same compressed file extent so that we don't need to
403 * re-read the same extent again and again.
404 *
405 * NOTE: this won't work well for subpage, as for subpage read, we lock the
406 * full page then submit bio for each compressed/regular extents.
407 *
408 * This means, if we have several sectors in the same page points to the same
409 * on-disk compressed data, we will re-read the same extent many times and
410 * this function can only help for the next page.
411 */
add_ra_bio_pages(struct inode * inode,u64 compressed_end,struct compressed_bio * cb,int * memstall,unsigned long * pflags)412 static noinline int add_ra_bio_pages(struct inode *inode,
413 u64 compressed_end,
414 struct compressed_bio *cb,
415 int *memstall, unsigned long *pflags)
416 {
417 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
418 unsigned long end_index;
419 struct bio *orig_bio = &cb->orig_bbio->bio;
420 u64 cur = cb->orig_bbio->file_offset + orig_bio->bi_iter.bi_size;
421 u64 isize = i_size_read(inode);
422 int ret;
423 struct folio *folio;
424 struct extent_map *em;
425 struct address_space *mapping = inode->i_mapping;
426 struct extent_map_tree *em_tree;
427 struct extent_io_tree *tree;
428 int sectors_missed = 0;
429
430 em_tree = &BTRFS_I(inode)->extent_tree;
431 tree = &BTRFS_I(inode)->io_tree;
432
433 if (isize == 0)
434 return 0;
435
436 /*
437 * For current subpage support, we only support 64K page size,
438 * which means maximum compressed extent size (128K) is just 2x page
439 * size.
440 * This makes readahead less effective, so here disable readahead for
441 * subpage for now, until full compressed write is supported.
442 */
443 if (fs_info->sectorsize < PAGE_SIZE)
444 return 0;
445
446 end_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
447
448 while (cur < compressed_end) {
449 u64 page_end;
450 u64 pg_index = cur >> PAGE_SHIFT;
451 u32 add_size;
452
453 if (pg_index > end_index)
454 break;
455
456 folio = __filemap_get_folio(mapping, pg_index, 0, 0);
457 if (!IS_ERR(folio)) {
458 u64 folio_sz = folio_size(folio);
459 u64 offset = offset_in_folio(folio, cur);
460
461 folio_put(folio);
462 sectors_missed += (folio_sz - offset) >>
463 fs_info->sectorsize_bits;
464
465 /* Beyond threshold, no need to continue */
466 if (sectors_missed > 4)
467 break;
468
469 /*
470 * Jump to next page start as we already have page for
471 * current offset.
472 */
473 cur += (folio_sz - offset);
474 continue;
475 }
476
477 folio = filemap_alloc_folio(mapping_gfp_constraint(mapping,
478 ~__GFP_FS), 0);
479 if (!folio)
480 break;
481
482 if (filemap_add_folio(mapping, folio, pg_index, GFP_NOFS)) {
483 /* There is already a page, skip to page end */
484 cur += folio_size(folio);
485 folio_put(folio);
486 continue;
487 }
488
489 if (!*memstall && folio_test_workingset(folio)) {
490 psi_memstall_enter(pflags);
491 *memstall = 1;
492 }
493
494 ret = set_folio_extent_mapped(folio);
495 if (ret < 0) {
496 folio_unlock(folio);
497 folio_put(folio);
498 break;
499 }
500
501 page_end = (pg_index << PAGE_SHIFT) + folio_size(folio) - 1;
502 lock_extent(tree, cur, page_end, NULL);
503 read_lock(&em_tree->lock);
504 em = lookup_extent_mapping(em_tree, cur, page_end + 1 - cur);
505 read_unlock(&em_tree->lock);
506
507 /*
508 * At this point, we have a locked page in the page cache for
509 * these bytes in the file. But, we have to make sure they map
510 * to this compressed extent on disk.
511 */
512 if (!em || cur < em->start ||
513 (cur + fs_info->sectorsize > extent_map_end(em)) ||
514 (extent_map_block_start(em) >> SECTOR_SHIFT) !=
515 orig_bio->bi_iter.bi_sector) {
516 free_extent_map(em);
517 unlock_extent(tree, cur, page_end, NULL);
518 folio_unlock(folio);
519 folio_put(folio);
520 break;
521 }
522 add_size = min(em->start + em->len, page_end + 1) - cur;
523 free_extent_map(em);
524 unlock_extent(tree, cur, page_end, NULL);
525
526 if (folio->index == end_index) {
527 size_t zero_offset = offset_in_folio(folio, isize);
528
529 if (zero_offset) {
530 int zeros;
531 zeros = folio_size(folio) - zero_offset;
532 folio_zero_range(folio, zero_offset, zeros);
533 }
534 }
535
536 if (!bio_add_folio(orig_bio, folio, add_size,
537 offset_in_folio(folio, cur))) {
538 folio_unlock(folio);
539 folio_put(folio);
540 break;
541 }
542 /*
543 * If it's subpage, we also need to increase its
544 * subpage::readers number, as at endio we will decrease
545 * subpage::readers and to unlock the page.
546 */
547 if (fs_info->sectorsize < PAGE_SIZE)
548 btrfs_folio_set_lock(fs_info, folio, cur, add_size);
549 folio_put(folio);
550 cur += add_size;
551 }
552 return 0;
553 }
554
555 /*
556 * for a compressed read, the bio we get passed has all the inode pages
557 * in it. We don't actually do IO on those pages but allocate new ones
558 * to hold the compressed pages on disk.
559 *
560 * bio->bi_iter.bi_sector points to the compressed extent on disk
561 * bio->bi_io_vec points to all of the inode pages
562 *
563 * After the compressed pages are read, we copy the bytes into the
564 * bio we were passed and then call the bio end_io calls
565 */
btrfs_submit_compressed_read(struct btrfs_bio * bbio)566 void btrfs_submit_compressed_read(struct btrfs_bio *bbio)
567 {
568 struct btrfs_inode *inode = bbio->inode;
569 struct btrfs_fs_info *fs_info = inode->root->fs_info;
570 struct extent_map_tree *em_tree = &inode->extent_tree;
571 struct compressed_bio *cb;
572 unsigned int compressed_len;
573 u64 file_offset = bbio->file_offset;
574 u64 em_len;
575 u64 em_start;
576 struct extent_map *em;
577 unsigned long pflags;
578 int memstall = 0;
579 blk_status_t ret;
580 int ret2;
581
582 /* we need the actual starting offset of this extent in the file */
583 read_lock(&em_tree->lock);
584 em = lookup_extent_mapping(em_tree, file_offset, fs_info->sectorsize);
585 read_unlock(&em_tree->lock);
586 if (!em) {
587 ret = BLK_STS_IOERR;
588 goto out;
589 }
590
591 ASSERT(extent_map_is_compressed(em));
592 compressed_len = em->disk_num_bytes;
593
594 cb = alloc_compressed_bio(inode, file_offset, REQ_OP_READ,
595 end_bbio_compressed_read);
596
597 cb->start = em->start - em->offset;
598 em_len = em->len;
599 em_start = em->start;
600
601 cb->len = bbio->bio.bi_iter.bi_size;
602 cb->compressed_len = compressed_len;
603 cb->compress_type = extent_map_compression(em);
604 cb->orig_bbio = bbio;
605
606 free_extent_map(em);
607
608 cb->nr_folios = DIV_ROUND_UP(compressed_len, PAGE_SIZE);
609 cb->compressed_folios = kcalloc(cb->nr_folios, sizeof(struct folio *), GFP_NOFS);
610 if (!cb->compressed_folios) {
611 ret = BLK_STS_RESOURCE;
612 goto out_free_bio;
613 }
614
615 ret2 = btrfs_alloc_folio_array(cb->nr_folios, cb->compressed_folios);
616 if (ret2) {
617 ret = BLK_STS_RESOURCE;
618 goto out_free_compressed_pages;
619 }
620
621 add_ra_bio_pages(&inode->vfs_inode, em_start + em_len, cb, &memstall,
622 &pflags);
623
624 /* include any pages we added in add_ra-bio_pages */
625 cb->len = bbio->bio.bi_iter.bi_size;
626 cb->bbio.bio.bi_iter.bi_sector = bbio->bio.bi_iter.bi_sector;
627 btrfs_add_compressed_bio_folios(cb);
628
629 if (memstall)
630 psi_memstall_leave(&pflags);
631
632 btrfs_submit_bbio(&cb->bbio, 0);
633 return;
634
635 out_free_compressed_pages:
636 kfree(cb->compressed_folios);
637 out_free_bio:
638 bio_put(&cb->bbio.bio);
639 out:
640 btrfs_bio_end_io(bbio, ret);
641 }
642
643 /*
644 * Heuristic uses systematic sampling to collect data from the input data
645 * range, the logic can be tuned by the following constants:
646 *
647 * @SAMPLING_READ_SIZE - how many bytes will be copied from for each sample
648 * @SAMPLING_INTERVAL - range from which the sampled data can be collected
649 */
650 #define SAMPLING_READ_SIZE (16)
651 #define SAMPLING_INTERVAL (256)
652
653 /*
654 * For statistical analysis of the input data we consider bytes that form a
655 * Galois Field of 256 objects. Each object has an attribute count, ie. how
656 * many times the object appeared in the sample.
657 */
658 #define BUCKET_SIZE (256)
659
660 /*
661 * The size of the sample is based on a statistical sampling rule of thumb.
662 * The common way is to perform sampling tests as long as the number of
663 * elements in each cell is at least 5.
664 *
665 * Instead of 5, we choose 32 to obtain more accurate results.
666 * If the data contain the maximum number of symbols, which is 256, we obtain a
667 * sample size bound by 8192.
668 *
669 * For a sample of at most 8KB of data per data range: 16 consecutive bytes
670 * from up to 512 locations.
671 */
672 #define MAX_SAMPLE_SIZE (BTRFS_MAX_UNCOMPRESSED * \
673 SAMPLING_READ_SIZE / SAMPLING_INTERVAL)
674
675 struct bucket_item {
676 u32 count;
677 };
678
679 struct heuristic_ws {
680 /* Partial copy of input data */
681 u8 *sample;
682 u32 sample_size;
683 /* Buckets store counters for each byte value */
684 struct bucket_item *bucket;
685 /* Sorting buffer */
686 struct bucket_item *bucket_b;
687 struct list_head list;
688 };
689
690 static struct workspace_manager heuristic_wsm;
691
free_heuristic_ws(struct list_head * ws)692 static void free_heuristic_ws(struct list_head *ws)
693 {
694 struct heuristic_ws *workspace;
695
696 workspace = list_entry(ws, struct heuristic_ws, list);
697
698 kvfree(workspace->sample);
699 kfree(workspace->bucket);
700 kfree(workspace->bucket_b);
701 kfree(workspace);
702 }
703
alloc_heuristic_ws(unsigned int level)704 static struct list_head *alloc_heuristic_ws(unsigned int level)
705 {
706 struct heuristic_ws *ws;
707
708 ws = kzalloc(sizeof(*ws), GFP_KERNEL);
709 if (!ws)
710 return ERR_PTR(-ENOMEM);
711
712 ws->sample = kvmalloc(MAX_SAMPLE_SIZE, GFP_KERNEL);
713 if (!ws->sample)
714 goto fail;
715
716 ws->bucket = kcalloc(BUCKET_SIZE, sizeof(*ws->bucket), GFP_KERNEL);
717 if (!ws->bucket)
718 goto fail;
719
720 ws->bucket_b = kcalloc(BUCKET_SIZE, sizeof(*ws->bucket_b), GFP_KERNEL);
721 if (!ws->bucket_b)
722 goto fail;
723
724 INIT_LIST_HEAD(&ws->list);
725 return &ws->list;
726 fail:
727 free_heuristic_ws(&ws->list);
728 return ERR_PTR(-ENOMEM);
729 }
730
731 const struct btrfs_compress_op btrfs_heuristic_compress = {
732 .workspace_manager = &heuristic_wsm,
733 };
734
735 static const struct btrfs_compress_op * const btrfs_compress_op[] = {
736 /* The heuristic is represented as compression type 0 */
737 &btrfs_heuristic_compress,
738 &btrfs_zlib_compress,
739 &btrfs_lzo_compress,
740 &btrfs_zstd_compress,
741 };
742
alloc_workspace(int type,unsigned int level)743 static struct list_head *alloc_workspace(int type, unsigned int level)
744 {
745 switch (type) {
746 case BTRFS_COMPRESS_NONE: return alloc_heuristic_ws(level);
747 case BTRFS_COMPRESS_ZLIB: return zlib_alloc_workspace(level);
748 case BTRFS_COMPRESS_LZO: return lzo_alloc_workspace(level);
749 case BTRFS_COMPRESS_ZSTD: return zstd_alloc_workspace(level);
750 default:
751 /*
752 * This can't happen, the type is validated several times
753 * before we get here.
754 */
755 BUG();
756 }
757 }
758
free_workspace(int type,struct list_head * ws)759 static void free_workspace(int type, struct list_head *ws)
760 {
761 switch (type) {
762 case BTRFS_COMPRESS_NONE: return free_heuristic_ws(ws);
763 case BTRFS_COMPRESS_ZLIB: return zlib_free_workspace(ws);
764 case BTRFS_COMPRESS_LZO: return lzo_free_workspace(ws);
765 case BTRFS_COMPRESS_ZSTD: return zstd_free_workspace(ws);
766 default:
767 /*
768 * This can't happen, the type is validated several times
769 * before we get here.
770 */
771 BUG();
772 }
773 }
774
btrfs_init_workspace_manager(int type)775 static void btrfs_init_workspace_manager(int type)
776 {
777 struct workspace_manager *wsm;
778 struct list_head *workspace;
779
780 wsm = btrfs_compress_op[type]->workspace_manager;
781 INIT_LIST_HEAD(&wsm->idle_ws);
782 spin_lock_init(&wsm->ws_lock);
783 atomic_set(&wsm->total_ws, 0);
784 init_waitqueue_head(&wsm->ws_wait);
785
786 /*
787 * Preallocate one workspace for each compression type so we can
788 * guarantee forward progress in the worst case
789 */
790 workspace = alloc_workspace(type, 0);
791 if (IS_ERR(workspace)) {
792 pr_warn(
793 "BTRFS: cannot preallocate compression workspace, will try later\n");
794 } else {
795 atomic_set(&wsm->total_ws, 1);
796 wsm->free_ws = 1;
797 list_add(workspace, &wsm->idle_ws);
798 }
799 }
800
btrfs_cleanup_workspace_manager(int type)801 static void btrfs_cleanup_workspace_manager(int type)
802 {
803 struct workspace_manager *wsman;
804 struct list_head *ws;
805
806 wsman = btrfs_compress_op[type]->workspace_manager;
807 while (!list_empty(&wsman->idle_ws)) {
808 ws = wsman->idle_ws.next;
809 list_del(ws);
810 free_workspace(type, ws);
811 atomic_dec(&wsman->total_ws);
812 }
813 }
814
815 /*
816 * This finds an available workspace or allocates a new one.
817 * If it's not possible to allocate a new one, waits until there's one.
818 * Preallocation makes a forward progress guarantees and we do not return
819 * errors.
820 */
btrfs_get_workspace(int type,unsigned int level)821 struct list_head *btrfs_get_workspace(int type, unsigned int level)
822 {
823 struct workspace_manager *wsm;
824 struct list_head *workspace;
825 int cpus = num_online_cpus();
826 unsigned nofs_flag;
827 struct list_head *idle_ws;
828 spinlock_t *ws_lock;
829 atomic_t *total_ws;
830 wait_queue_head_t *ws_wait;
831 int *free_ws;
832
833 wsm = btrfs_compress_op[type]->workspace_manager;
834 idle_ws = &wsm->idle_ws;
835 ws_lock = &wsm->ws_lock;
836 total_ws = &wsm->total_ws;
837 ws_wait = &wsm->ws_wait;
838 free_ws = &wsm->free_ws;
839
840 again:
841 spin_lock(ws_lock);
842 if (!list_empty(idle_ws)) {
843 workspace = idle_ws->next;
844 list_del(workspace);
845 (*free_ws)--;
846 spin_unlock(ws_lock);
847 return workspace;
848
849 }
850 if (atomic_read(total_ws) > cpus) {
851 DEFINE_WAIT(wait);
852
853 spin_unlock(ws_lock);
854 prepare_to_wait(ws_wait, &wait, TASK_UNINTERRUPTIBLE);
855 if (atomic_read(total_ws) > cpus && !*free_ws)
856 schedule();
857 finish_wait(ws_wait, &wait);
858 goto again;
859 }
860 atomic_inc(total_ws);
861 spin_unlock(ws_lock);
862
863 /*
864 * Allocation helpers call vmalloc that can't use GFP_NOFS, so we have
865 * to turn it off here because we might get called from the restricted
866 * context of btrfs_compress_bio/btrfs_compress_pages
867 */
868 nofs_flag = memalloc_nofs_save();
869 workspace = alloc_workspace(type, level);
870 memalloc_nofs_restore(nofs_flag);
871
872 if (IS_ERR(workspace)) {
873 atomic_dec(total_ws);
874 wake_up(ws_wait);
875
876 /*
877 * Do not return the error but go back to waiting. There's a
878 * workspace preallocated for each type and the compression
879 * time is bounded so we get to a workspace eventually. This
880 * makes our caller's life easier.
881 *
882 * To prevent silent and low-probability deadlocks (when the
883 * initial preallocation fails), check if there are any
884 * workspaces at all.
885 */
886 if (atomic_read(total_ws) == 0) {
887 static DEFINE_RATELIMIT_STATE(_rs,
888 /* once per minute */ 60 * HZ,
889 /* no burst */ 1);
890
891 if (__ratelimit(&_rs)) {
892 pr_warn("BTRFS: no compression workspaces, low memory, retrying\n");
893 }
894 }
895 goto again;
896 }
897 return workspace;
898 }
899
get_workspace(int type,int level)900 static struct list_head *get_workspace(int type, int level)
901 {
902 switch (type) {
903 case BTRFS_COMPRESS_NONE: return btrfs_get_workspace(type, level);
904 case BTRFS_COMPRESS_ZLIB: return zlib_get_workspace(level);
905 case BTRFS_COMPRESS_LZO: return btrfs_get_workspace(type, level);
906 case BTRFS_COMPRESS_ZSTD: return zstd_get_workspace(level);
907 default:
908 /*
909 * This can't happen, the type is validated several times
910 * before we get here.
911 */
912 BUG();
913 }
914 }
915
916 /*
917 * put a workspace struct back on the list or free it if we have enough
918 * idle ones sitting around
919 */
btrfs_put_workspace(int type,struct list_head * ws)920 void btrfs_put_workspace(int type, struct list_head *ws)
921 {
922 struct workspace_manager *wsm;
923 struct list_head *idle_ws;
924 spinlock_t *ws_lock;
925 atomic_t *total_ws;
926 wait_queue_head_t *ws_wait;
927 int *free_ws;
928
929 wsm = btrfs_compress_op[type]->workspace_manager;
930 idle_ws = &wsm->idle_ws;
931 ws_lock = &wsm->ws_lock;
932 total_ws = &wsm->total_ws;
933 ws_wait = &wsm->ws_wait;
934 free_ws = &wsm->free_ws;
935
936 spin_lock(ws_lock);
937 if (*free_ws <= num_online_cpus()) {
938 list_add(ws, idle_ws);
939 (*free_ws)++;
940 spin_unlock(ws_lock);
941 goto wake;
942 }
943 spin_unlock(ws_lock);
944
945 free_workspace(type, ws);
946 atomic_dec(total_ws);
947 wake:
948 cond_wake_up(ws_wait);
949 }
950
put_workspace(int type,struct list_head * ws)951 static void put_workspace(int type, struct list_head *ws)
952 {
953 switch (type) {
954 case BTRFS_COMPRESS_NONE: return btrfs_put_workspace(type, ws);
955 case BTRFS_COMPRESS_ZLIB: return btrfs_put_workspace(type, ws);
956 case BTRFS_COMPRESS_LZO: return btrfs_put_workspace(type, ws);
957 case BTRFS_COMPRESS_ZSTD: return zstd_put_workspace(ws);
958 default:
959 /*
960 * This can't happen, the type is validated several times
961 * before we get here.
962 */
963 BUG();
964 }
965 }
966
967 /*
968 * Adjust @level according to the limits of the compression algorithm or
969 * fallback to default
970 */
btrfs_compress_set_level(int type,unsigned level)971 static unsigned int btrfs_compress_set_level(int type, unsigned level)
972 {
973 const struct btrfs_compress_op *ops = btrfs_compress_op[type];
974
975 if (level == 0)
976 level = ops->default_level;
977 else
978 level = min(level, ops->max_level);
979
980 return level;
981 }
982
983 /* Wrapper around find_get_page(), with extra error message. */
btrfs_compress_filemap_get_folio(struct address_space * mapping,u64 start,struct folio ** in_folio_ret)984 int btrfs_compress_filemap_get_folio(struct address_space *mapping, u64 start,
985 struct folio **in_folio_ret)
986 {
987 struct folio *in_folio;
988
989 /*
990 * The compressed write path should have the folio locked already, thus
991 * we only need to grab one reference.
992 */
993 in_folio = filemap_get_folio(mapping, start >> PAGE_SHIFT);
994 if (IS_ERR(in_folio)) {
995 struct btrfs_inode *inode = BTRFS_I(mapping->host);
996
997 btrfs_crit(inode->root->fs_info,
998 "failed to get page cache, root %lld ino %llu file offset %llu",
999 btrfs_root_id(inode->root), btrfs_ino(inode), start);
1000 return -ENOENT;
1001 }
1002 *in_folio_ret = in_folio;
1003 return 0;
1004 }
1005
1006 /*
1007 * Given an address space and start and length, compress the bytes into @pages
1008 * that are allocated on demand.
1009 *
1010 * @type_level is encoded algorithm and level, where level 0 means whatever
1011 * default the algorithm chooses and is opaque here;
1012 * - compression algo are 0-3
1013 * - the level are bits 4-7
1014 *
1015 * @out_pages is an in/out parameter, holds maximum number of pages to allocate
1016 * and returns number of actually allocated pages
1017 *
1018 * @total_in is used to return the number of bytes actually read. It
1019 * may be smaller than the input length if we had to exit early because we
1020 * ran out of room in the pages array or because we cross the
1021 * max_out threshold.
1022 *
1023 * @total_out is an in/out parameter, must be set to the input length and will
1024 * be also used to return the total number of compressed bytes
1025 */
btrfs_compress_folios(unsigned int type_level,struct address_space * mapping,u64 start,struct folio ** folios,unsigned long * out_folios,unsigned long * total_in,unsigned long * total_out)1026 int btrfs_compress_folios(unsigned int type_level, struct address_space *mapping,
1027 u64 start, struct folio **folios, unsigned long *out_folios,
1028 unsigned long *total_in, unsigned long *total_out)
1029 {
1030 int type = btrfs_compress_type(type_level);
1031 int level = btrfs_compress_level(type_level);
1032 struct list_head *workspace;
1033 int ret;
1034
1035 level = btrfs_compress_set_level(type, level);
1036 workspace = get_workspace(type, level);
1037 ret = compression_compress_pages(type, workspace, mapping, start, folios,
1038 out_folios, total_in, total_out);
1039 put_workspace(type, workspace);
1040 return ret;
1041 }
1042
btrfs_decompress_bio(struct compressed_bio * cb)1043 static int btrfs_decompress_bio(struct compressed_bio *cb)
1044 {
1045 struct list_head *workspace;
1046 int ret;
1047 int type = cb->compress_type;
1048
1049 workspace = get_workspace(type, 0);
1050 ret = compression_decompress_bio(workspace, cb);
1051 put_workspace(type, workspace);
1052
1053 if (!ret)
1054 zero_fill_bio(&cb->orig_bbio->bio);
1055 return ret;
1056 }
1057
1058 /*
1059 * a less complex decompression routine. Our compressed data fits in a
1060 * single page, and we want to read a single page out of it.
1061 * start_byte tells us the offset into the compressed data we're interested in
1062 */
btrfs_decompress(int type,const u8 * data_in,struct folio * dest_folio,unsigned long dest_pgoff,size_t srclen,size_t destlen)1063 int btrfs_decompress(int type, const u8 *data_in, struct folio *dest_folio,
1064 unsigned long dest_pgoff, size_t srclen, size_t destlen)
1065 {
1066 struct btrfs_fs_info *fs_info = folio_to_fs_info(dest_folio);
1067 struct list_head *workspace;
1068 const u32 sectorsize = fs_info->sectorsize;
1069 int ret;
1070
1071 /*
1072 * The full destination page range should not exceed the page size.
1073 * And the @destlen should not exceed sectorsize, as this is only called for
1074 * inline file extents, which should not exceed sectorsize.
1075 */
1076 ASSERT(dest_pgoff + destlen <= PAGE_SIZE && destlen <= sectorsize);
1077
1078 workspace = get_workspace(type, 0);
1079 ret = compression_decompress(type, workspace, data_in, dest_folio,
1080 dest_pgoff, srclen, destlen);
1081 put_workspace(type, workspace);
1082
1083 return ret;
1084 }
1085
btrfs_init_compress(void)1086 int __init btrfs_init_compress(void)
1087 {
1088 if (bioset_init(&btrfs_compressed_bioset, BIO_POOL_SIZE,
1089 offsetof(struct compressed_bio, bbio.bio),
1090 BIOSET_NEED_BVECS))
1091 return -ENOMEM;
1092
1093 compr_pool.shrinker = shrinker_alloc(SHRINKER_NONSLAB, "btrfs-compr-pages");
1094 if (!compr_pool.shrinker)
1095 return -ENOMEM;
1096
1097 btrfs_init_workspace_manager(BTRFS_COMPRESS_NONE);
1098 btrfs_init_workspace_manager(BTRFS_COMPRESS_ZLIB);
1099 btrfs_init_workspace_manager(BTRFS_COMPRESS_LZO);
1100 zstd_init_workspace_manager();
1101
1102 spin_lock_init(&compr_pool.lock);
1103 INIT_LIST_HEAD(&compr_pool.list);
1104 compr_pool.count = 0;
1105 /* 128K / 4K = 32, for 8 threads is 256 pages. */
1106 compr_pool.thresh = BTRFS_MAX_COMPRESSED / PAGE_SIZE * 8;
1107 compr_pool.shrinker->count_objects = btrfs_compr_pool_count;
1108 compr_pool.shrinker->scan_objects = btrfs_compr_pool_scan;
1109 compr_pool.shrinker->batch = 32;
1110 compr_pool.shrinker->seeks = DEFAULT_SEEKS;
1111 shrinker_register(compr_pool.shrinker);
1112
1113 return 0;
1114 }
1115
btrfs_exit_compress(void)1116 void __cold btrfs_exit_compress(void)
1117 {
1118 /* For now scan drains all pages and does not touch the parameters. */
1119 btrfs_compr_pool_scan(NULL, NULL);
1120 shrinker_free(compr_pool.shrinker);
1121
1122 btrfs_cleanup_workspace_manager(BTRFS_COMPRESS_NONE);
1123 btrfs_cleanup_workspace_manager(BTRFS_COMPRESS_ZLIB);
1124 btrfs_cleanup_workspace_manager(BTRFS_COMPRESS_LZO);
1125 zstd_cleanup_workspace_manager();
1126 bioset_exit(&btrfs_compressed_bioset);
1127 }
1128
1129 /*
1130 * Copy decompressed data from working buffer to pages.
1131 *
1132 * @buf: The decompressed data buffer
1133 * @buf_len: The decompressed data length
1134 * @decompressed: Number of bytes that are already decompressed inside the
1135 * compressed extent
1136 * @cb: The compressed extent descriptor
1137 * @orig_bio: The original bio that the caller wants to read for
1138 *
1139 * An easier to understand graph is like below:
1140 *
1141 * |<- orig_bio ->| |<- orig_bio->|
1142 * |<------- full decompressed extent ----->|
1143 * |<----------- @cb range ---->|
1144 * | |<-- @buf_len -->|
1145 * |<--- @decompressed --->|
1146 *
1147 * Note that, @cb can be a subpage of the full decompressed extent, but
1148 * @cb->start always has the same as the orig_file_offset value of the full
1149 * decompressed extent.
1150 *
1151 * When reading compressed extent, we have to read the full compressed extent,
1152 * while @orig_bio may only want part of the range.
1153 * Thus this function will ensure only data covered by @orig_bio will be copied
1154 * to.
1155 *
1156 * Return 0 if we have copied all needed contents for @orig_bio.
1157 * Return >0 if we need continue decompress.
1158 */
btrfs_decompress_buf2page(const char * buf,u32 buf_len,struct compressed_bio * cb,u32 decompressed)1159 int btrfs_decompress_buf2page(const char *buf, u32 buf_len,
1160 struct compressed_bio *cb, u32 decompressed)
1161 {
1162 struct bio *orig_bio = &cb->orig_bbio->bio;
1163 /* Offset inside the full decompressed extent */
1164 u32 cur_offset;
1165
1166 cur_offset = decompressed;
1167 /* The main loop to do the copy */
1168 while (cur_offset < decompressed + buf_len) {
1169 struct bio_vec bvec;
1170 size_t copy_len;
1171 u32 copy_start;
1172 /* Offset inside the full decompressed extent */
1173 u32 bvec_offset;
1174
1175 bvec = bio_iter_iovec(orig_bio, orig_bio->bi_iter);
1176 /*
1177 * cb->start may underflow, but subtracting that value can still
1178 * give us correct offset inside the full decompressed extent.
1179 */
1180 bvec_offset = page_offset(bvec.bv_page) + bvec.bv_offset - cb->start;
1181
1182 /* Haven't reached the bvec range, exit */
1183 if (decompressed + buf_len <= bvec_offset)
1184 return 1;
1185
1186 copy_start = max(cur_offset, bvec_offset);
1187 copy_len = min(bvec_offset + bvec.bv_len,
1188 decompressed + buf_len) - copy_start;
1189 ASSERT(copy_len);
1190
1191 /*
1192 * Extra range check to ensure we didn't go beyond
1193 * @buf + @buf_len.
1194 */
1195 ASSERT(copy_start - decompressed < buf_len);
1196 memcpy_to_page(bvec.bv_page, bvec.bv_offset,
1197 buf + copy_start - decompressed, copy_len);
1198 cur_offset += copy_len;
1199
1200 bio_advance(orig_bio, copy_len);
1201 /* Finished the bio */
1202 if (!orig_bio->bi_iter.bi_size)
1203 return 0;
1204 }
1205 return 1;
1206 }
1207
1208 /*
1209 * Shannon Entropy calculation
1210 *
1211 * Pure byte distribution analysis fails to determine compressibility of data.
1212 * Try calculating entropy to estimate the average minimum number of bits
1213 * needed to encode the sampled data.
1214 *
1215 * For convenience, return the percentage of needed bits, instead of amount of
1216 * bits directly.
1217 *
1218 * @ENTROPY_LVL_ACEPTABLE - below that threshold, sample has low byte entropy
1219 * and can be compressible with high probability
1220 *
1221 * @ENTROPY_LVL_HIGH - data are not compressible with high probability
1222 *
1223 * Use of ilog2() decreases precision, we lower the LVL to 5 to compensate.
1224 */
1225 #define ENTROPY_LVL_ACEPTABLE (65)
1226 #define ENTROPY_LVL_HIGH (80)
1227
1228 /*
1229 * For increasead precision in shannon_entropy calculation,
1230 * let's do pow(n, M) to save more digits after comma:
1231 *
1232 * - maximum int bit length is 64
1233 * - ilog2(MAX_SAMPLE_SIZE) -> 13
1234 * - 13 * 4 = 52 < 64 -> M = 4
1235 *
1236 * So use pow(n, 4).
1237 */
ilog2_w(u64 n)1238 static inline u32 ilog2_w(u64 n)
1239 {
1240 return ilog2(n * n * n * n);
1241 }
1242
shannon_entropy(struct heuristic_ws * ws)1243 static u32 shannon_entropy(struct heuristic_ws *ws)
1244 {
1245 const u32 entropy_max = 8 * ilog2_w(2);
1246 u32 entropy_sum = 0;
1247 u32 p, p_base, sz_base;
1248 u32 i;
1249
1250 sz_base = ilog2_w(ws->sample_size);
1251 for (i = 0; i < BUCKET_SIZE && ws->bucket[i].count > 0; i++) {
1252 p = ws->bucket[i].count;
1253 p_base = ilog2_w(p);
1254 entropy_sum += p * (sz_base - p_base);
1255 }
1256
1257 entropy_sum /= ws->sample_size;
1258 return entropy_sum * 100 / entropy_max;
1259 }
1260
1261 #define RADIX_BASE 4U
1262 #define COUNTERS_SIZE (1U << RADIX_BASE)
1263
get4bits(u64 num,int shift)1264 static u8 get4bits(u64 num, int shift) {
1265 u8 low4bits;
1266
1267 num >>= shift;
1268 /* Reverse order */
1269 low4bits = (COUNTERS_SIZE - 1) - (num % COUNTERS_SIZE);
1270 return low4bits;
1271 }
1272
1273 /*
1274 * Use 4 bits as radix base
1275 * Use 16 u32 counters for calculating new position in buf array
1276 *
1277 * @array - array that will be sorted
1278 * @array_buf - buffer array to store sorting results
1279 * must be equal in size to @array
1280 * @num - array size
1281 */
radix_sort(struct bucket_item * array,struct bucket_item * array_buf,int num)1282 static void radix_sort(struct bucket_item *array, struct bucket_item *array_buf,
1283 int num)
1284 {
1285 u64 max_num;
1286 u64 buf_num;
1287 u32 counters[COUNTERS_SIZE];
1288 u32 new_addr;
1289 u32 addr;
1290 int bitlen;
1291 int shift;
1292 int i;
1293
1294 /*
1295 * Try avoid useless loop iterations for small numbers stored in big
1296 * counters. Example: 48 33 4 ... in 64bit array
1297 */
1298 max_num = array[0].count;
1299 for (i = 1; i < num; i++) {
1300 buf_num = array[i].count;
1301 if (buf_num > max_num)
1302 max_num = buf_num;
1303 }
1304
1305 buf_num = ilog2(max_num);
1306 bitlen = ALIGN(buf_num, RADIX_BASE * 2);
1307
1308 shift = 0;
1309 while (shift < bitlen) {
1310 memset(counters, 0, sizeof(counters));
1311
1312 for (i = 0; i < num; i++) {
1313 buf_num = array[i].count;
1314 addr = get4bits(buf_num, shift);
1315 counters[addr]++;
1316 }
1317
1318 for (i = 1; i < COUNTERS_SIZE; i++)
1319 counters[i] += counters[i - 1];
1320
1321 for (i = num - 1; i >= 0; i--) {
1322 buf_num = array[i].count;
1323 addr = get4bits(buf_num, shift);
1324 counters[addr]--;
1325 new_addr = counters[addr];
1326 array_buf[new_addr] = array[i];
1327 }
1328
1329 shift += RADIX_BASE;
1330
1331 /*
1332 * Normal radix expects to move data from a temporary array, to
1333 * the main one. But that requires some CPU time. Avoid that
1334 * by doing another sort iteration to original array instead of
1335 * memcpy()
1336 */
1337 memset(counters, 0, sizeof(counters));
1338
1339 for (i = 0; i < num; i ++) {
1340 buf_num = array_buf[i].count;
1341 addr = get4bits(buf_num, shift);
1342 counters[addr]++;
1343 }
1344
1345 for (i = 1; i < COUNTERS_SIZE; i++)
1346 counters[i] += counters[i - 1];
1347
1348 for (i = num - 1; i >= 0; i--) {
1349 buf_num = array_buf[i].count;
1350 addr = get4bits(buf_num, shift);
1351 counters[addr]--;
1352 new_addr = counters[addr];
1353 array[new_addr] = array_buf[i];
1354 }
1355
1356 shift += RADIX_BASE;
1357 }
1358 }
1359
1360 /*
1361 * Size of the core byte set - how many bytes cover 90% of the sample
1362 *
1363 * There are several types of structured binary data that use nearly all byte
1364 * values. The distribution can be uniform and counts in all buckets will be
1365 * nearly the same (eg. encrypted data). Unlikely to be compressible.
1366 *
1367 * Other possibility is normal (Gaussian) distribution, where the data could
1368 * be potentially compressible, but we have to take a few more steps to decide
1369 * how much.
1370 *
1371 * @BYTE_CORE_SET_LOW - main part of byte values repeated frequently,
1372 * compression algo can easy fix that
1373 * @BYTE_CORE_SET_HIGH - data have uniform distribution and with high
1374 * probability is not compressible
1375 */
1376 #define BYTE_CORE_SET_LOW (64)
1377 #define BYTE_CORE_SET_HIGH (200)
1378
byte_core_set_size(struct heuristic_ws * ws)1379 static int byte_core_set_size(struct heuristic_ws *ws)
1380 {
1381 u32 i;
1382 u32 coreset_sum = 0;
1383 const u32 core_set_threshold = ws->sample_size * 90 / 100;
1384 struct bucket_item *bucket = ws->bucket;
1385
1386 /* Sort in reverse order */
1387 radix_sort(ws->bucket, ws->bucket_b, BUCKET_SIZE);
1388
1389 for (i = 0; i < BYTE_CORE_SET_LOW; i++)
1390 coreset_sum += bucket[i].count;
1391
1392 if (coreset_sum > core_set_threshold)
1393 return i;
1394
1395 for (; i < BYTE_CORE_SET_HIGH && bucket[i].count > 0; i++) {
1396 coreset_sum += bucket[i].count;
1397 if (coreset_sum > core_set_threshold)
1398 break;
1399 }
1400
1401 return i;
1402 }
1403
1404 /*
1405 * Count byte values in buckets.
1406 * This heuristic can detect textual data (configs, xml, json, html, etc).
1407 * Because in most text-like data byte set is restricted to limited number of
1408 * possible characters, and that restriction in most cases makes data easy to
1409 * compress.
1410 *
1411 * @BYTE_SET_THRESHOLD - consider all data within this byte set size:
1412 * less - compressible
1413 * more - need additional analysis
1414 */
1415 #define BYTE_SET_THRESHOLD (64)
1416
byte_set_size(const struct heuristic_ws * ws)1417 static u32 byte_set_size(const struct heuristic_ws *ws)
1418 {
1419 u32 i;
1420 u32 byte_set_size = 0;
1421
1422 for (i = 0; i < BYTE_SET_THRESHOLD; i++) {
1423 if (ws->bucket[i].count > 0)
1424 byte_set_size++;
1425 }
1426
1427 /*
1428 * Continue collecting count of byte values in buckets. If the byte
1429 * set size is bigger then the threshold, it's pointless to continue,
1430 * the detection technique would fail for this type of data.
1431 */
1432 for (; i < BUCKET_SIZE; i++) {
1433 if (ws->bucket[i].count > 0) {
1434 byte_set_size++;
1435 if (byte_set_size > BYTE_SET_THRESHOLD)
1436 return byte_set_size;
1437 }
1438 }
1439
1440 return byte_set_size;
1441 }
1442
sample_repeated_patterns(struct heuristic_ws * ws)1443 static bool sample_repeated_patterns(struct heuristic_ws *ws)
1444 {
1445 const u32 half_of_sample = ws->sample_size / 2;
1446 const u8 *data = ws->sample;
1447
1448 return memcmp(&data[0], &data[half_of_sample], half_of_sample) == 0;
1449 }
1450
heuristic_collect_sample(struct inode * inode,u64 start,u64 end,struct heuristic_ws * ws)1451 static void heuristic_collect_sample(struct inode *inode, u64 start, u64 end,
1452 struct heuristic_ws *ws)
1453 {
1454 struct page *page;
1455 u64 index, index_end;
1456 u32 i, curr_sample_pos;
1457 u8 *in_data;
1458
1459 /*
1460 * Compression handles the input data by chunks of 128KiB
1461 * (defined by BTRFS_MAX_UNCOMPRESSED)
1462 *
1463 * We do the same for the heuristic and loop over the whole range.
1464 *
1465 * MAX_SAMPLE_SIZE - calculated under assumption that heuristic will
1466 * process no more than BTRFS_MAX_UNCOMPRESSED at a time.
1467 */
1468 if (end - start > BTRFS_MAX_UNCOMPRESSED)
1469 end = start + BTRFS_MAX_UNCOMPRESSED;
1470
1471 index = start >> PAGE_SHIFT;
1472 index_end = end >> PAGE_SHIFT;
1473
1474 /* Don't miss unaligned end */
1475 if (!PAGE_ALIGNED(end))
1476 index_end++;
1477
1478 curr_sample_pos = 0;
1479 while (index < index_end) {
1480 page = find_get_page(inode->i_mapping, index);
1481 in_data = kmap_local_page(page);
1482 /* Handle case where the start is not aligned to PAGE_SIZE */
1483 i = start % PAGE_SIZE;
1484 while (i < PAGE_SIZE - SAMPLING_READ_SIZE) {
1485 /* Don't sample any garbage from the last page */
1486 if (start > end - SAMPLING_READ_SIZE)
1487 break;
1488 memcpy(&ws->sample[curr_sample_pos], &in_data[i],
1489 SAMPLING_READ_SIZE);
1490 i += SAMPLING_INTERVAL;
1491 start += SAMPLING_INTERVAL;
1492 curr_sample_pos += SAMPLING_READ_SIZE;
1493 }
1494 kunmap_local(in_data);
1495 put_page(page);
1496
1497 index++;
1498 }
1499
1500 ws->sample_size = curr_sample_pos;
1501 }
1502
1503 /*
1504 * Compression heuristic.
1505 *
1506 * The following types of analysis can be performed:
1507 * - detect mostly zero data
1508 * - detect data with low "byte set" size (text, etc)
1509 * - detect data with low/high "core byte" set
1510 *
1511 * Return non-zero if the compression should be done, 0 otherwise.
1512 */
btrfs_compress_heuristic(struct btrfs_inode * inode,u64 start,u64 end)1513 int btrfs_compress_heuristic(struct btrfs_inode *inode, u64 start, u64 end)
1514 {
1515 struct list_head *ws_list = get_workspace(0, 0);
1516 struct heuristic_ws *ws;
1517 u32 i;
1518 u8 byte;
1519 int ret = 0;
1520
1521 ws = list_entry(ws_list, struct heuristic_ws, list);
1522
1523 heuristic_collect_sample(&inode->vfs_inode, start, end, ws);
1524
1525 if (sample_repeated_patterns(ws)) {
1526 ret = 1;
1527 goto out;
1528 }
1529
1530 memset(ws->bucket, 0, sizeof(*ws->bucket)*BUCKET_SIZE);
1531
1532 for (i = 0; i < ws->sample_size; i++) {
1533 byte = ws->sample[i];
1534 ws->bucket[byte].count++;
1535 }
1536
1537 i = byte_set_size(ws);
1538 if (i < BYTE_SET_THRESHOLD) {
1539 ret = 2;
1540 goto out;
1541 }
1542
1543 i = byte_core_set_size(ws);
1544 if (i <= BYTE_CORE_SET_LOW) {
1545 ret = 3;
1546 goto out;
1547 }
1548
1549 if (i >= BYTE_CORE_SET_HIGH) {
1550 ret = 0;
1551 goto out;
1552 }
1553
1554 i = shannon_entropy(ws);
1555 if (i <= ENTROPY_LVL_ACEPTABLE) {
1556 ret = 4;
1557 goto out;
1558 }
1559
1560 /*
1561 * For the levels below ENTROPY_LVL_HIGH, additional analysis would be
1562 * needed to give green light to compression.
1563 *
1564 * For now just assume that compression at that level is not worth the
1565 * resources because:
1566 *
1567 * 1. it is possible to defrag the data later
1568 *
1569 * 2. the data would turn out to be hardly compressible, eg. 150 byte
1570 * values, every bucket has counter at level ~54. The heuristic would
1571 * be confused. This can happen when data have some internal repeated
1572 * patterns like "abbacbbc...". This can be detected by analyzing
1573 * pairs of bytes, which is too costly.
1574 */
1575 if (i < ENTROPY_LVL_HIGH) {
1576 ret = 5;
1577 goto out;
1578 } else {
1579 ret = 0;
1580 goto out;
1581 }
1582
1583 out:
1584 put_workspace(0, ws_list);
1585 return ret;
1586 }
1587
1588 /*
1589 * Convert the compression suffix (eg. after "zlib" starting with ":") to
1590 * level, unrecognized string will set the default level
1591 */
btrfs_compress_str2level(unsigned int type,const char * str)1592 unsigned int btrfs_compress_str2level(unsigned int type, const char *str)
1593 {
1594 unsigned int level = 0;
1595 int ret;
1596
1597 if (!type)
1598 return 0;
1599
1600 if (str[0] == ':') {
1601 ret = kstrtouint(str + 1, 10, &level);
1602 if (ret)
1603 level = 0;
1604 }
1605
1606 level = btrfs_compress_set_level(type, level);
1607
1608 return level;
1609 }
1610