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/highmem.h>
12 #include <linux/kthread.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/slab.h>
19 #include <linux/sched/mm.h>
20 #include <linux/log2.h>
21 #include <crypto/hash.h>
22 #include "misc.h"
23 #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"
29 #include "compression.h"
30 #include "extent_io.h"
31 #include "extent_map.h"
32 #include "subpage.h"
33 #include "zoned.h"
34
35 static const char* const btrfs_compress_types[] = { "", "zlib", "lzo", "zstd" };
36
btrfs_compress_type2str(enum btrfs_compression_type type)37 const 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];
45 default:
46 break;
47 }
48
49 return NULL;
50 }
51
btrfs_compress_is_valid_type(const char * str,size_t len)52 bool 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
compression_compress_pages(int type,struct list_head * ws,struct address_space * mapping,u64 start,struct page ** pages,unsigned long * out_pages,unsigned long * total_in,unsigned long * total_out)68 static 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 /*
86 * 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.
93 */
94 *out_pages = 0;
95 return -E2BIG;
96 }
97 }
98
compression_decompress_bio(struct list_head * ws,struct compressed_bio * cb)99 static int compression_decompress_bio(struct list_head *ws,
100 struct compressed_bio *cb)
101 {
102 switch (cb->compress_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
compression_decompress(int type,struct list_head * ws,unsigned char * data_in,struct page * dest_page,unsigned long start_byte,size_t srclen,size_t destlen)116 static 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
137 static int btrfs_decompress_bio(struct compressed_bio *cb);
138
compressed_bio_size(struct btrfs_fs_info * fs_info,unsigned long disk_size)139 static inline int compressed_bio_size(struct btrfs_fs_info *fs_info,
140 unsigned long disk_size)
141 {
142 return sizeof(struct compressed_bio) +
143 (DIV_ROUND_UP(disk_size, fs_info->sectorsize)) * fs_info->csum_size;
144 }
145
check_compressed_csum(struct btrfs_inode * inode,struct bio * bio,u64 disk_start)146 static int check_compressed_csum(struct btrfs_inode *inode, struct bio *bio,
147 u64 disk_start)
148 {
149 struct btrfs_fs_info *fs_info = inode->root->fs_info;
150 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
151 const u32 csum_size = fs_info->csum_size;
152 const u32 sectorsize = fs_info->sectorsize;
153 struct page *page;
154 unsigned int i;
155 char *kaddr;
156 u8 csum[BTRFS_CSUM_SIZE];
157 struct compressed_bio *cb = bio->bi_private;
158 u8 *cb_sum = cb->sums;
159
160 if ((inode->flags & BTRFS_INODE_NODATASUM) ||
161 test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state))
162 return 0;
163
164 shash->tfm = fs_info->csum_shash;
165
166 for (i = 0; i < cb->nr_pages; i++) {
167 u32 pg_offset;
168 u32 bytes_left = PAGE_SIZE;
169 page = cb->compressed_pages[i];
170
171 /* Determine the remaining bytes inside the page first */
172 if (i == cb->nr_pages - 1)
173 bytes_left = cb->compressed_len - i * PAGE_SIZE;
174
175 /* Hash through the page sector by sector */
176 for (pg_offset = 0; pg_offset < bytes_left;
177 pg_offset += sectorsize) {
178 kaddr = kmap_atomic(page);
179 crypto_shash_digest(shash, kaddr + pg_offset,
180 sectorsize, csum);
181 kunmap_atomic(kaddr);
182
183 if (memcmp(&csum, cb_sum, csum_size) != 0) {
184 btrfs_print_data_csum_error(inode, disk_start,
185 csum, cb_sum, cb->mirror_num);
186 if (btrfs_bio(bio)->device)
187 btrfs_dev_stat_inc_and_print(
188 btrfs_bio(bio)->device,
189 BTRFS_DEV_STAT_CORRUPTION_ERRS);
190 return -EIO;
191 }
192 cb_sum += csum_size;
193 disk_start += sectorsize;
194 }
195 }
196 return 0;
197 }
198
199 /*
200 * Reduce bio and io accounting for a compressed_bio with its corresponding bio.
201 *
202 * Return true if there is no pending bio nor io.
203 * Return false otherwise.
204 */
dec_and_test_compressed_bio(struct compressed_bio * cb,struct bio * bio)205 static bool dec_and_test_compressed_bio(struct compressed_bio *cb, struct bio *bio)
206 {
207 struct btrfs_fs_info *fs_info = btrfs_sb(cb->inode->i_sb);
208 unsigned int bi_size = 0;
209 bool last_io = false;
210 struct bio_vec *bvec;
211 struct bvec_iter_all iter_all;
212
213 /*
214 * At endio time, bi_iter.bi_size doesn't represent the real bio size.
215 * Thus here we have to iterate through all segments to grab correct
216 * bio size.
217 */
218 bio_for_each_segment_all(bvec, bio, iter_all)
219 bi_size += bvec->bv_len;
220
221 if (bio->bi_status)
222 cb->status = bio->bi_status;
223
224 ASSERT(bi_size && bi_size <= cb->compressed_len);
225 last_io = refcount_sub_and_test(bi_size >> fs_info->sectorsize_bits,
226 &cb->pending_sectors);
227 /*
228 * Here we must wake up the possible error handler after all other
229 * operations on @cb finished, or we can race with
230 * finish_compressed_bio_*() which may free @cb.
231 */
232 wake_up_var(cb);
233
234 return last_io;
235 }
236
finish_compressed_bio_read(struct compressed_bio * cb)237 static void finish_compressed_bio_read(struct compressed_bio *cb)
238 {
239 unsigned int index;
240 struct page *page;
241
242 /* Release the compressed pages */
243 for (index = 0; index < cb->nr_pages; index++) {
244 page = cb->compressed_pages[index];
245 page->mapping = NULL;
246 put_page(page);
247 }
248
249 /* Do io completion on the original bio */
250 if (cb->status != BLK_STS_OK) {
251 cb->orig_bio->bi_status = cb->status;
252 bio_endio(cb->orig_bio);
253 } else {
254 struct bio_vec *bvec;
255 struct bvec_iter_all iter_all;
256
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(cb->orig_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
279 /* 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 */
end_compressed_bio_read(struct bio * bio)289 static void end_compressed_bio_read(struct bio *bio)
290 {
291 struct compressed_bio *cb = bio->bi_private;
292 struct inode *inode;
293 unsigned int mirror = btrfs_bio(bio)->mirror_num;
294 int ret = 0;
295
296 if (!dec_and_test_compressed_bio(cb, bio))
297 goto out;
298
299 /*
300 * Record the correct mirror_num in cb->orig_bio so that
301 * read-repair can work properly.
302 */
303 btrfs_bio(cb->orig_bio)->mirror_num = mirror;
304 cb->mirror_num = mirror;
305
306 /*
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->status != BLK_STS_OK)
311 goto csum_failed;
312
313 inode = cb->inode;
314 ret = check_compressed_csum(BTRFS_I(inode), bio,
315 bio->bi_iter.bi_sector << 9);
316 if (ret)
317 goto csum_failed;
318
319 /* ok, we're the last bio for this extent, lets start
320 * the decompression.
321 */
322 ret = btrfs_decompress_bio(cb);
323
324 csum_failed:
325 if (ret)
326 cb->status = errno_to_blk_status(ret);
327 finish_compressed_bio_read(cb);
328 out:
329 bio_put(bio);
330 }
331
332 /*
333 * Clear the writeback bits on all of the file
334 * pages for a compressed write
335 */
end_compressed_writeback(struct inode * inode,const struct compressed_bio * cb)336 static noinline void end_compressed_writeback(struct inode *inode,
337 const struct compressed_bio *cb)
338 {
339 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
340 unsigned long index = cb->start >> PAGE_SHIFT;
341 unsigned long end_index = (cb->start + cb->len - 1) >> PAGE_SHIFT;
342 struct page *pages[16];
343 unsigned long nr_pages = end_index - index + 1;
344 const int errno = blk_status_to_errno(cb->status);
345 int i;
346 int ret;
347
348 if (errno)
349 mapping_set_error(inode->i_mapping, errno);
350
351 while (nr_pages > 0) {
352 ret = find_get_pages_contig(inode->i_mapping, index,
353 min_t(unsigned long,
354 nr_pages, ARRAY_SIZE(pages)), pages);
355 if (ret == 0) {
356 nr_pages -= 1;
357 index += 1;
358 continue;
359 }
360 for (i = 0; i < ret; i++) {
361 if (errno)
362 SetPageError(pages[i]);
363 btrfs_page_clamp_clear_writeback(fs_info, pages[i],
364 cb->start, cb->len);
365 put_page(pages[i]);
366 }
367 nr_pages -= ret;
368 index += ret;
369 }
370 /* the inode may be gone now */
371 }
372
finish_compressed_bio_write(struct compressed_bio * cb)373 static void finish_compressed_bio_write(struct compressed_bio *cb)
374 {
375 struct inode *inode = cb->inode;
376 unsigned int index;
377
378 /*
379 * Ok, we're the last bio for this extent, step one is to call back
380 * into the FS and do all the end_io operations.
381 */
382 btrfs_writepage_endio_finish_ordered(BTRFS_I(inode), NULL,
383 cb->start, cb->start + cb->len - 1,
384 cb->status == BLK_STS_OK);
385
386 if (cb->writeback)
387 end_compressed_writeback(inode, cb);
388 /* Note, our inode could be gone now */
389
390 /*
391 * Release the compressed pages, these came from alloc_page and
392 * are not attached to the inode at all
393 */
394 for (index = 0; index < cb->nr_pages; index++) {
395 struct page *page = cb->compressed_pages[index];
396
397 page->mapping = NULL;
398 put_page(page);
399 }
400
401 /* Finally free the cb struct */
402 kfree(cb->compressed_pages);
403 kfree(cb);
404 }
405
406 /*
407 * Do the cleanup once all the compressed pages hit the disk. This will clear
408 * writeback on the file pages and free the compressed pages.
409 *
410 * This also calls the writeback end hooks for the file pages so that metadata
411 * and checksums can be updated in the file.
412 */
end_compressed_bio_write(struct bio * bio)413 static void end_compressed_bio_write(struct bio *bio)
414 {
415 struct compressed_bio *cb = bio->bi_private;
416
417 if (!dec_and_test_compressed_bio(cb, bio))
418 goto out;
419
420 btrfs_record_physical_zoned(cb->inode, cb->start, bio);
421
422 finish_compressed_bio_write(cb);
423 out:
424 bio_put(bio);
425 }
426
submit_compressed_bio(struct btrfs_fs_info * fs_info,struct bio * bio,int mirror_num)427 static blk_status_t submit_compressed_bio(struct btrfs_fs_info *fs_info,
428 struct bio *bio, int mirror_num)
429 {
430 blk_status_t ret;
431
432 ASSERT(bio->bi_iter.bi_size);
433 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
434 if (ret)
435 return ret;
436 ret = btrfs_map_bio(fs_info, bio, mirror_num);
437 return ret;
438 }
439
440 /*
441 * Allocate a compressed_bio, which will be used to read/write on-disk
442 * (aka, compressed) * data.
443 *
444 * @cb: The compressed_bio structure, which records all the needed
445 * information to bind the compressed data to the uncompressed
446 * page cache.
447 * @disk_byten: The logical bytenr where the compressed data will be read
448 * from or written to.
449 * @endio_func: The endio function to call after the IO for compressed data
450 * is finished.
451 * @next_stripe_start: Return value of logical bytenr of where next stripe starts.
452 * Let the caller know to only fill the bio up to the stripe
453 * boundary.
454 */
455
456
alloc_compressed_bio(struct compressed_bio * cb,u64 disk_bytenr,unsigned int opf,bio_end_io_t endio_func,u64 * next_stripe_start)457 static struct bio *alloc_compressed_bio(struct compressed_bio *cb, u64 disk_bytenr,
458 unsigned int opf, bio_end_io_t endio_func,
459 u64 *next_stripe_start)
460 {
461 struct btrfs_fs_info *fs_info = btrfs_sb(cb->inode->i_sb);
462 struct btrfs_io_geometry geom;
463 struct extent_map *em;
464 struct bio *bio;
465 int ret;
466
467 bio = btrfs_bio_alloc(BIO_MAX_VECS);
468
469 bio->bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
470 bio->bi_opf = opf;
471 bio->bi_private = cb;
472 bio->bi_end_io = endio_func;
473
474 em = btrfs_get_chunk_map(fs_info, disk_bytenr, fs_info->sectorsize);
475 if (IS_ERR(em)) {
476 bio_put(bio);
477 return ERR_CAST(em);
478 }
479
480 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
481 bio_set_dev(bio, em->map_lookup->stripes[0].dev->bdev);
482
483 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio), disk_bytenr, &geom);
484 free_extent_map(em);
485 if (ret < 0) {
486 bio_put(bio);
487 return ERR_PTR(ret);
488 }
489 *next_stripe_start = disk_bytenr + geom.len;
490
491 return bio;
492 }
493
494 /*
495 * worker function to build and submit bios for previously compressed pages.
496 * The corresponding pages in the inode should be marked for writeback
497 * and the compressed pages should have a reference on them for dropping
498 * when the IO is complete.
499 *
500 * This also checksums the file bytes and gets things ready for
501 * the end io hooks.
502 */
btrfs_submit_compressed_write(struct btrfs_inode * inode,u64 start,unsigned int len,u64 disk_start,unsigned int compressed_len,struct page ** compressed_pages,unsigned int nr_pages,unsigned int write_flags,struct cgroup_subsys_state * blkcg_css,bool writeback)503 blk_status_t btrfs_submit_compressed_write(struct btrfs_inode *inode, u64 start,
504 unsigned int len, u64 disk_start,
505 unsigned int compressed_len,
506 struct page **compressed_pages,
507 unsigned int nr_pages,
508 unsigned int write_flags,
509 struct cgroup_subsys_state *blkcg_css,
510 bool writeback)
511 {
512 struct btrfs_fs_info *fs_info = inode->root->fs_info;
513 struct bio *bio = NULL;
514 struct compressed_bio *cb;
515 u64 cur_disk_bytenr = disk_start;
516 u64 next_stripe_start;
517 blk_status_t ret;
518 int skip_sum = inode->flags & BTRFS_INODE_NODATASUM;
519 const bool use_append = btrfs_use_zone_append(inode, disk_start);
520 const unsigned int bio_op = use_append ? REQ_OP_ZONE_APPEND : REQ_OP_WRITE;
521
522 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
523 IS_ALIGNED(len, fs_info->sectorsize));
524 cb = kmalloc(compressed_bio_size(fs_info, compressed_len), GFP_NOFS);
525 if (!cb)
526 return BLK_STS_RESOURCE;
527 refcount_set(&cb->pending_sectors, compressed_len >> fs_info->sectorsize_bits);
528 cb->status = BLK_STS_OK;
529 cb->inode = &inode->vfs_inode;
530 cb->start = start;
531 cb->len = len;
532 cb->mirror_num = 0;
533 cb->compressed_pages = compressed_pages;
534 cb->compressed_len = compressed_len;
535 cb->writeback = writeback;
536 cb->orig_bio = NULL;
537 cb->nr_pages = nr_pages;
538
539 if (blkcg_css)
540 kthread_associate_blkcg(blkcg_css);
541
542 while (cur_disk_bytenr < disk_start + compressed_len) {
543 u64 offset = cur_disk_bytenr - disk_start;
544 unsigned int index = offset >> PAGE_SHIFT;
545 unsigned int real_size;
546 unsigned int added;
547 struct page *page = compressed_pages[index];
548 bool submit = false;
549
550 /* Allocate new bio if submitted or not yet allocated */
551 if (!bio) {
552 bio = alloc_compressed_bio(cb, cur_disk_bytenr,
553 bio_op | write_flags, end_compressed_bio_write,
554 &next_stripe_start);
555 if (IS_ERR(bio)) {
556 ret = errno_to_blk_status(PTR_ERR(bio));
557 bio = NULL;
558 goto finish_cb;
559 }
560 if (blkcg_css)
561 bio->bi_opf |= REQ_CGROUP_PUNT;
562 }
563 /*
564 * We should never reach next_stripe_start start as we will
565 * submit comp_bio when reach the boundary immediately.
566 */
567 ASSERT(cur_disk_bytenr != next_stripe_start);
568
569 /*
570 * We have various limits on the real read size:
571 * - stripe boundary
572 * - page boundary
573 * - compressed length boundary
574 */
575 real_size = min_t(u64, U32_MAX, next_stripe_start - cur_disk_bytenr);
576 real_size = min_t(u64, real_size, PAGE_SIZE - offset_in_page(offset));
577 real_size = min_t(u64, real_size, compressed_len - offset);
578 ASSERT(IS_ALIGNED(real_size, fs_info->sectorsize));
579
580 if (use_append)
581 added = bio_add_zone_append_page(bio, page, real_size,
582 offset_in_page(offset));
583 else
584 added = bio_add_page(bio, page, real_size,
585 offset_in_page(offset));
586 /* Reached zoned boundary */
587 if (added == 0)
588 submit = true;
589
590 cur_disk_bytenr += added;
591 /* Reached stripe boundary */
592 if (cur_disk_bytenr == next_stripe_start)
593 submit = true;
594
595 /* Finished the range */
596 if (cur_disk_bytenr == disk_start + compressed_len)
597 submit = true;
598
599 if (submit) {
600 if (!skip_sum) {
601 ret = btrfs_csum_one_bio(inode, bio, start, true);
602 if (ret)
603 goto finish_cb;
604 }
605
606 ret = submit_compressed_bio(fs_info, bio, 0);
607 if (ret)
608 goto finish_cb;
609 bio = NULL;
610 }
611 cond_resched();
612 }
613 if (blkcg_css)
614 kthread_associate_blkcg(NULL);
615
616 return 0;
617
618 finish_cb:
619 if (blkcg_css)
620 kthread_associate_blkcg(NULL);
621
622 if (bio) {
623 bio->bi_status = ret;
624 bio_endio(bio);
625 }
626 /* Last byte of @cb is submitted, endio will free @cb */
627 if (cur_disk_bytenr == disk_start + compressed_len)
628 return ret;
629
630 wait_var_event(cb, refcount_read(&cb->pending_sectors) ==
631 (disk_start + compressed_len - cur_disk_bytenr) >>
632 fs_info->sectorsize_bits);
633 /*
634 * Even with previous bio ended, we should still have io not yet
635 * submitted, thus need to finish manually.
636 */
637 ASSERT(refcount_read(&cb->pending_sectors));
638 /* Now we are the only one referring @cb, can finish it safely. */
639 finish_compressed_bio_write(cb);
640 return ret;
641 }
642
bio_end_offset(struct bio * bio)643 static u64 bio_end_offset(struct bio *bio)
644 {
645 struct bio_vec *last = bio_last_bvec_all(bio);
646
647 return page_offset(last->bv_page) + last->bv_len + last->bv_offset;
648 }
649
650 /*
651 * Add extra pages in the same compressed file extent so that we don't need to
652 * re-read the same extent again and again.
653 *
654 * NOTE: this won't work well for subpage, as for subpage read, we lock the
655 * full page then submit bio for each compressed/regular extents.
656 *
657 * This means, if we have several sectors in the same page points to the same
658 * on-disk compressed data, we will re-read the same extent many times and
659 * this function can only help for the next page.
660 */
add_ra_bio_pages(struct inode * inode,u64 compressed_end,struct compressed_bio * cb)661 static noinline int add_ra_bio_pages(struct inode *inode,
662 u64 compressed_end,
663 struct compressed_bio *cb)
664 {
665 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
666 unsigned long end_index;
667 u64 cur = bio_end_offset(cb->orig_bio);
668 u64 isize = i_size_read(inode);
669 int ret;
670 struct page *page;
671 struct extent_map *em;
672 struct address_space *mapping = inode->i_mapping;
673 struct extent_map_tree *em_tree;
674 struct extent_io_tree *tree;
675 int sectors_missed = 0;
676
677 em_tree = &BTRFS_I(inode)->extent_tree;
678 tree = &BTRFS_I(inode)->io_tree;
679
680 if (isize == 0)
681 return 0;
682
683 /*
684 * For current subpage support, we only support 64K page size,
685 * which means maximum compressed extent size (128K) is just 2x page
686 * size.
687 * This makes readahead less effective, so here disable readahead for
688 * subpage for now, until full compressed write is supported.
689 */
690 if (btrfs_sb(inode->i_sb)->sectorsize < PAGE_SIZE)
691 return 0;
692
693 end_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
694
695 while (cur < compressed_end) {
696 u64 page_end;
697 u64 pg_index = cur >> PAGE_SHIFT;
698 u32 add_size;
699
700 if (pg_index > end_index)
701 break;
702
703 page = xa_load(&mapping->i_pages, pg_index);
704 if (page && !xa_is_value(page)) {
705 sectors_missed += (PAGE_SIZE - offset_in_page(cur)) >>
706 fs_info->sectorsize_bits;
707
708 /* Beyond threshold, no need to continue */
709 if (sectors_missed > 4)
710 break;
711
712 /*
713 * Jump to next page start as we already have page for
714 * current offset.
715 */
716 cur = (pg_index << PAGE_SHIFT) + PAGE_SIZE;
717 continue;
718 }
719
720 page = __page_cache_alloc(mapping_gfp_constraint(mapping,
721 ~__GFP_FS));
722 if (!page)
723 break;
724
725 if (add_to_page_cache_lru(page, mapping, pg_index, GFP_NOFS)) {
726 put_page(page);
727 /* There is already a page, skip to page end */
728 cur = (pg_index << PAGE_SHIFT) + PAGE_SIZE;
729 continue;
730 }
731
732 ret = set_page_extent_mapped(page);
733 if (ret < 0) {
734 unlock_page(page);
735 put_page(page);
736 break;
737 }
738
739 page_end = (pg_index << PAGE_SHIFT) + PAGE_SIZE - 1;
740 lock_extent(tree, cur, page_end);
741 read_lock(&em_tree->lock);
742 em = lookup_extent_mapping(em_tree, cur, page_end + 1 - cur);
743 read_unlock(&em_tree->lock);
744
745 /*
746 * At this point, we have a locked page in the page cache for
747 * these bytes in the file. But, we have to make sure they map
748 * to this compressed extent on disk.
749 */
750 if (!em || cur < em->start ||
751 (cur + fs_info->sectorsize > extent_map_end(em)) ||
752 (em->block_start >> 9) != cb->orig_bio->bi_iter.bi_sector) {
753 free_extent_map(em);
754 unlock_extent(tree, cur, page_end);
755 unlock_page(page);
756 put_page(page);
757 break;
758 }
759 free_extent_map(em);
760
761 if (page->index == end_index) {
762 size_t zero_offset = offset_in_page(isize);
763
764 if (zero_offset) {
765 int zeros;
766 zeros = PAGE_SIZE - zero_offset;
767 memzero_page(page, zero_offset, zeros);
768 flush_dcache_page(page);
769 }
770 }
771
772 add_size = min(em->start + em->len, page_end + 1) - cur;
773 ret = bio_add_page(cb->orig_bio, page, add_size, offset_in_page(cur));
774 if (ret != add_size) {
775 unlock_extent(tree, cur, page_end);
776 unlock_page(page);
777 put_page(page);
778 break;
779 }
780 /*
781 * If it's subpage, we also need to increase its
782 * subpage::readers number, as at endio we will decrease
783 * subpage::readers and to unlock the page.
784 */
785 if (fs_info->sectorsize < PAGE_SIZE)
786 btrfs_subpage_start_reader(fs_info, page, cur, add_size);
787 put_page(page);
788 cur += add_size;
789 }
790 return 0;
791 }
792
793 /*
794 * for a compressed read, the bio we get passed has all the inode pages
795 * in it. We don't actually do IO on those pages but allocate new ones
796 * to hold the compressed pages on disk.
797 *
798 * bio->bi_iter.bi_sector points to the compressed extent on disk
799 * bio->bi_io_vec points to all of the inode pages
800 *
801 * After the compressed pages are read, we copy the bytes into the
802 * bio we were passed and then call the bio end_io calls
803 */
btrfs_submit_compressed_read(struct inode * inode,struct bio * bio,int mirror_num)804 void btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
805 int mirror_num)
806 {
807 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
808 struct extent_map_tree *em_tree;
809 struct compressed_bio *cb;
810 unsigned int compressed_len;
811 struct bio *comp_bio = NULL;
812 const u64 disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
813 u64 cur_disk_byte = disk_bytenr;
814 u64 next_stripe_start;
815 u64 file_offset;
816 u64 em_len;
817 u64 em_start;
818 struct extent_map *em;
819 blk_status_t ret;
820 int ret2;
821 int i;
822 u8 *sums;
823
824 em_tree = &BTRFS_I(inode)->extent_tree;
825
826 file_offset = bio_first_bvec_all(bio)->bv_offset +
827 page_offset(bio_first_page_all(bio));
828
829 /* we need the actual starting offset of this extent in the file */
830 read_lock(&em_tree->lock);
831 em = lookup_extent_mapping(em_tree, file_offset, fs_info->sectorsize);
832 read_unlock(&em_tree->lock);
833 if (!em) {
834 ret = BLK_STS_IOERR;
835 goto out;
836 }
837
838 ASSERT(em->compress_type != BTRFS_COMPRESS_NONE);
839 compressed_len = em->block_len;
840 cb = kmalloc(compressed_bio_size(fs_info, compressed_len), GFP_NOFS);
841 if (!cb) {
842 ret = BLK_STS_RESOURCE;
843 goto out;
844 }
845
846 refcount_set(&cb->pending_sectors, compressed_len >> fs_info->sectorsize_bits);
847 cb->status = BLK_STS_OK;
848 cb->inode = inode;
849 cb->mirror_num = mirror_num;
850 sums = cb->sums;
851
852 cb->start = em->orig_start;
853 em_len = em->len;
854 em_start = em->start;
855
856 cb->len = bio->bi_iter.bi_size;
857 cb->compressed_len = compressed_len;
858 cb->compress_type = em->compress_type;
859 cb->orig_bio = bio;
860
861 free_extent_map(em);
862 em = NULL;
863
864 cb->nr_pages = DIV_ROUND_UP(compressed_len, PAGE_SIZE);
865 cb->compressed_pages = kcalloc(cb->nr_pages, sizeof(struct page *), GFP_NOFS);
866 if (!cb->compressed_pages) {
867 ret = BLK_STS_RESOURCE;
868 goto fail;
869 }
870
871 ret2 = btrfs_alloc_page_array(cb->nr_pages, cb->compressed_pages);
872 if (ret2) {
873 ret = BLK_STS_RESOURCE;
874 goto fail;
875 }
876
877 add_ra_bio_pages(inode, em_start + em_len, cb);
878
879 /* include any pages we added in add_ra-bio_pages */
880 cb->len = bio->bi_iter.bi_size;
881
882 while (cur_disk_byte < disk_bytenr + compressed_len) {
883 u64 offset = cur_disk_byte - disk_bytenr;
884 unsigned int index = offset >> PAGE_SHIFT;
885 unsigned int real_size;
886 unsigned int added;
887 struct page *page = cb->compressed_pages[index];
888 bool submit = false;
889
890 /* Allocate new bio if submitted or not yet allocated */
891 if (!comp_bio) {
892 comp_bio = alloc_compressed_bio(cb, cur_disk_byte,
893 REQ_OP_READ, end_compressed_bio_read,
894 &next_stripe_start);
895 if (IS_ERR(comp_bio)) {
896 ret = errno_to_blk_status(PTR_ERR(comp_bio));
897 comp_bio = NULL;
898 goto finish_cb;
899 }
900 }
901 /*
902 * We should never reach next_stripe_start start as we will
903 * submit comp_bio when reach the boundary immediately.
904 */
905 ASSERT(cur_disk_byte != next_stripe_start);
906 /*
907 * We have various limit on the real read size:
908 * - stripe boundary
909 * - page boundary
910 * - compressed length boundary
911 */
912 real_size = min_t(u64, U32_MAX, next_stripe_start - cur_disk_byte);
913 real_size = min_t(u64, real_size, PAGE_SIZE - offset_in_page(offset));
914 real_size = min_t(u64, real_size, compressed_len - offset);
915 ASSERT(IS_ALIGNED(real_size, fs_info->sectorsize));
916
917 added = bio_add_page(comp_bio, page, real_size, offset_in_page(offset));
918 /*
919 * Maximum compressed extent is smaller than bio size limit,
920 * thus bio_add_page() should always success.
921 */
922 ASSERT(added == real_size);
923 cur_disk_byte += added;
924
925 /* Reached stripe boundary, need to submit */
926 if (cur_disk_byte == next_stripe_start)
927 submit = true;
928
929 /* Has finished the range, need to submit */
930 if (cur_disk_byte == disk_bytenr + compressed_len)
931 submit = true;
932
933 if (submit) {
934 unsigned int nr_sectors;
935
936 ret = btrfs_lookup_bio_sums(inode, comp_bio, sums);
937 if (ret)
938 goto finish_cb;
939
940 nr_sectors = DIV_ROUND_UP(comp_bio->bi_iter.bi_size,
941 fs_info->sectorsize);
942 sums += fs_info->csum_size * nr_sectors;
943
944 ret = submit_compressed_bio(fs_info, comp_bio, mirror_num);
945 if (ret)
946 goto finish_cb;
947 comp_bio = NULL;
948 }
949 }
950 return;
951
952 fail:
953 if (cb->compressed_pages) {
954 for (i = 0; i < cb->nr_pages; i++) {
955 if (cb->compressed_pages[i])
956 __free_page(cb->compressed_pages[i]);
957 }
958 }
959
960 kfree(cb->compressed_pages);
961 kfree(cb);
962 out:
963 free_extent_map(em);
964 bio->bi_status = ret;
965 bio_endio(bio);
966 return;
967 finish_cb:
968 if (comp_bio) {
969 comp_bio->bi_status = ret;
970 bio_endio(comp_bio);
971 }
972 /* All bytes of @cb is submitted, endio will free @cb */
973 if (cur_disk_byte == disk_bytenr + compressed_len)
974 return;
975
976 wait_var_event(cb, refcount_read(&cb->pending_sectors) ==
977 (disk_bytenr + compressed_len - cur_disk_byte) >>
978 fs_info->sectorsize_bits);
979 /*
980 * Even with previous bio ended, we should still have io not yet
981 * submitted, thus need to finish @cb manually.
982 */
983 ASSERT(refcount_read(&cb->pending_sectors));
984 /* Now we are the only one referring @cb, can finish it safely. */
985 finish_compressed_bio_read(cb);
986 }
987
988 /*
989 * Heuristic uses systematic sampling to collect data from the input data
990 * range, the logic can be tuned by the following constants:
991 *
992 * @SAMPLING_READ_SIZE - how many bytes will be copied from for each sample
993 * @SAMPLING_INTERVAL - range from which the sampled data can be collected
994 */
995 #define SAMPLING_READ_SIZE (16)
996 #define SAMPLING_INTERVAL (256)
997
998 /*
999 * For statistical analysis of the input data we consider bytes that form a
1000 * Galois Field of 256 objects. Each object has an attribute count, ie. how
1001 * many times the object appeared in the sample.
1002 */
1003 #define BUCKET_SIZE (256)
1004
1005 /*
1006 * The size of the sample is based on a statistical sampling rule of thumb.
1007 * The common way is to perform sampling tests as long as the number of
1008 * elements in each cell is at least 5.
1009 *
1010 * Instead of 5, we choose 32 to obtain more accurate results.
1011 * If the data contain the maximum number of symbols, which is 256, we obtain a
1012 * sample size bound by 8192.
1013 *
1014 * For a sample of at most 8KB of data per data range: 16 consecutive bytes
1015 * from up to 512 locations.
1016 */
1017 #define MAX_SAMPLE_SIZE (BTRFS_MAX_UNCOMPRESSED * \
1018 SAMPLING_READ_SIZE / SAMPLING_INTERVAL)
1019
1020 struct bucket_item {
1021 u32 count;
1022 };
1023
1024 struct heuristic_ws {
1025 /* Partial copy of input data */
1026 u8 *sample;
1027 u32 sample_size;
1028 /* Buckets store counters for each byte value */
1029 struct bucket_item *bucket;
1030 /* Sorting buffer */
1031 struct bucket_item *bucket_b;
1032 struct list_head list;
1033 };
1034
1035 static struct workspace_manager heuristic_wsm;
1036
free_heuristic_ws(struct list_head * ws)1037 static void free_heuristic_ws(struct list_head *ws)
1038 {
1039 struct heuristic_ws *workspace;
1040
1041 workspace = list_entry(ws, struct heuristic_ws, list);
1042
1043 kvfree(workspace->sample);
1044 kfree(workspace->bucket);
1045 kfree(workspace->bucket_b);
1046 kfree(workspace);
1047 }
1048
alloc_heuristic_ws(unsigned int level)1049 static struct list_head *alloc_heuristic_ws(unsigned int level)
1050 {
1051 struct heuristic_ws *ws;
1052
1053 ws = kzalloc(sizeof(*ws), GFP_KERNEL);
1054 if (!ws)
1055 return ERR_PTR(-ENOMEM);
1056
1057 ws->sample = kvmalloc(MAX_SAMPLE_SIZE, GFP_KERNEL);
1058 if (!ws->sample)
1059 goto fail;
1060
1061 ws->bucket = kcalloc(BUCKET_SIZE, sizeof(*ws->bucket), GFP_KERNEL);
1062 if (!ws->bucket)
1063 goto fail;
1064
1065 ws->bucket_b = kcalloc(BUCKET_SIZE, sizeof(*ws->bucket_b), GFP_KERNEL);
1066 if (!ws->bucket_b)
1067 goto fail;
1068
1069 INIT_LIST_HEAD(&ws->list);
1070 return &ws->list;
1071 fail:
1072 free_heuristic_ws(&ws->list);
1073 return ERR_PTR(-ENOMEM);
1074 }
1075
1076 const struct btrfs_compress_op btrfs_heuristic_compress = {
1077 .workspace_manager = &heuristic_wsm,
1078 };
1079
1080 static const struct btrfs_compress_op * const btrfs_compress_op[] = {
1081 /* The heuristic is represented as compression type 0 */
1082 &btrfs_heuristic_compress,
1083 &btrfs_zlib_compress,
1084 &btrfs_lzo_compress,
1085 &btrfs_zstd_compress,
1086 };
1087
alloc_workspace(int type,unsigned int level)1088 static struct list_head *alloc_workspace(int type, unsigned int level)
1089 {
1090 switch (type) {
1091 case BTRFS_COMPRESS_NONE: return alloc_heuristic_ws(level);
1092 case BTRFS_COMPRESS_ZLIB: return zlib_alloc_workspace(level);
1093 case BTRFS_COMPRESS_LZO: return lzo_alloc_workspace(level);
1094 case BTRFS_COMPRESS_ZSTD: return zstd_alloc_workspace(level);
1095 default:
1096 /*
1097 * This can't happen, the type is validated several times
1098 * before we get here.
1099 */
1100 BUG();
1101 }
1102 }
1103
free_workspace(int type,struct list_head * ws)1104 static void free_workspace(int type, struct list_head *ws)
1105 {
1106 switch (type) {
1107 case BTRFS_COMPRESS_NONE: return free_heuristic_ws(ws);
1108 case BTRFS_COMPRESS_ZLIB: return zlib_free_workspace(ws);
1109 case BTRFS_COMPRESS_LZO: return lzo_free_workspace(ws);
1110 case BTRFS_COMPRESS_ZSTD: return zstd_free_workspace(ws);
1111 default:
1112 /*
1113 * This can't happen, the type is validated several times
1114 * before we get here.
1115 */
1116 BUG();
1117 }
1118 }
1119
btrfs_init_workspace_manager(int type)1120 static void btrfs_init_workspace_manager(int type)
1121 {
1122 struct workspace_manager *wsm;
1123 struct list_head *workspace;
1124
1125 wsm = btrfs_compress_op[type]->workspace_manager;
1126 INIT_LIST_HEAD(&wsm->idle_ws);
1127 spin_lock_init(&wsm->ws_lock);
1128 atomic_set(&wsm->total_ws, 0);
1129 init_waitqueue_head(&wsm->ws_wait);
1130
1131 /*
1132 * Preallocate one workspace for each compression type so we can
1133 * guarantee forward progress in the worst case
1134 */
1135 workspace = alloc_workspace(type, 0);
1136 if (IS_ERR(workspace)) {
1137 pr_warn(
1138 "BTRFS: cannot preallocate compression workspace, will try later\n");
1139 } else {
1140 atomic_set(&wsm->total_ws, 1);
1141 wsm->free_ws = 1;
1142 list_add(workspace, &wsm->idle_ws);
1143 }
1144 }
1145
btrfs_cleanup_workspace_manager(int type)1146 static void btrfs_cleanup_workspace_manager(int type)
1147 {
1148 struct workspace_manager *wsman;
1149 struct list_head *ws;
1150
1151 wsman = btrfs_compress_op[type]->workspace_manager;
1152 while (!list_empty(&wsman->idle_ws)) {
1153 ws = wsman->idle_ws.next;
1154 list_del(ws);
1155 free_workspace(type, ws);
1156 atomic_dec(&wsman->total_ws);
1157 }
1158 }
1159
1160 /*
1161 * This finds an available workspace or allocates a new one.
1162 * If it's not possible to allocate a new one, waits until there's one.
1163 * Preallocation makes a forward progress guarantees and we do not return
1164 * errors.
1165 */
btrfs_get_workspace(int type,unsigned int level)1166 struct list_head *btrfs_get_workspace(int type, unsigned int level)
1167 {
1168 struct workspace_manager *wsm;
1169 struct list_head *workspace;
1170 int cpus = num_online_cpus();
1171 unsigned nofs_flag;
1172 struct list_head *idle_ws;
1173 spinlock_t *ws_lock;
1174 atomic_t *total_ws;
1175 wait_queue_head_t *ws_wait;
1176 int *free_ws;
1177
1178 wsm = btrfs_compress_op[type]->workspace_manager;
1179 idle_ws = &wsm->idle_ws;
1180 ws_lock = &wsm->ws_lock;
1181 total_ws = &wsm->total_ws;
1182 ws_wait = &wsm->ws_wait;
1183 free_ws = &wsm->free_ws;
1184
1185 again:
1186 spin_lock(ws_lock);
1187 if (!list_empty(idle_ws)) {
1188 workspace = idle_ws->next;
1189 list_del(workspace);
1190 (*free_ws)--;
1191 spin_unlock(ws_lock);
1192 return workspace;
1193
1194 }
1195 if (atomic_read(total_ws) > cpus) {
1196 DEFINE_WAIT(wait);
1197
1198 spin_unlock(ws_lock);
1199 prepare_to_wait(ws_wait, &wait, TASK_UNINTERRUPTIBLE);
1200 if (atomic_read(total_ws) > cpus && !*free_ws)
1201 schedule();
1202 finish_wait(ws_wait, &wait);
1203 goto again;
1204 }
1205 atomic_inc(total_ws);
1206 spin_unlock(ws_lock);
1207
1208 /*
1209 * Allocation helpers call vmalloc that can't use GFP_NOFS, so we have
1210 * to turn it off here because we might get called from the restricted
1211 * context of btrfs_compress_bio/btrfs_compress_pages
1212 */
1213 nofs_flag = memalloc_nofs_save();
1214 workspace = alloc_workspace(type, level);
1215 memalloc_nofs_restore(nofs_flag);
1216
1217 if (IS_ERR(workspace)) {
1218 atomic_dec(total_ws);
1219 wake_up(ws_wait);
1220
1221 /*
1222 * Do not return the error but go back to waiting. There's a
1223 * workspace preallocated for each type and the compression
1224 * time is bounded so we get to a workspace eventually. This
1225 * makes our caller's life easier.
1226 *
1227 * To prevent silent and low-probability deadlocks (when the
1228 * initial preallocation fails), check if there are any
1229 * workspaces at all.
1230 */
1231 if (atomic_read(total_ws) == 0) {
1232 static DEFINE_RATELIMIT_STATE(_rs,
1233 /* once per minute */ 60 * HZ,
1234 /* no burst */ 1);
1235
1236 if (__ratelimit(&_rs)) {
1237 pr_warn("BTRFS: no compression workspaces, low memory, retrying\n");
1238 }
1239 }
1240 goto again;
1241 }
1242 return workspace;
1243 }
1244
get_workspace(int type,int level)1245 static struct list_head *get_workspace(int type, int level)
1246 {
1247 switch (type) {
1248 case BTRFS_COMPRESS_NONE: return btrfs_get_workspace(type, level);
1249 case BTRFS_COMPRESS_ZLIB: return zlib_get_workspace(level);
1250 case BTRFS_COMPRESS_LZO: return btrfs_get_workspace(type, level);
1251 case BTRFS_COMPRESS_ZSTD: return zstd_get_workspace(level);
1252 default:
1253 /*
1254 * This can't happen, the type is validated several times
1255 * before we get here.
1256 */
1257 BUG();
1258 }
1259 }
1260
1261 /*
1262 * put a workspace struct back on the list or free it if we have enough
1263 * idle ones sitting around
1264 */
btrfs_put_workspace(int type,struct list_head * ws)1265 void btrfs_put_workspace(int type, struct list_head *ws)
1266 {
1267 struct workspace_manager *wsm;
1268 struct list_head *idle_ws;
1269 spinlock_t *ws_lock;
1270 atomic_t *total_ws;
1271 wait_queue_head_t *ws_wait;
1272 int *free_ws;
1273
1274 wsm = btrfs_compress_op[type]->workspace_manager;
1275 idle_ws = &wsm->idle_ws;
1276 ws_lock = &wsm->ws_lock;
1277 total_ws = &wsm->total_ws;
1278 ws_wait = &wsm->ws_wait;
1279 free_ws = &wsm->free_ws;
1280
1281 spin_lock(ws_lock);
1282 if (*free_ws <= num_online_cpus()) {
1283 list_add(ws, idle_ws);
1284 (*free_ws)++;
1285 spin_unlock(ws_lock);
1286 goto wake;
1287 }
1288 spin_unlock(ws_lock);
1289
1290 free_workspace(type, ws);
1291 atomic_dec(total_ws);
1292 wake:
1293 cond_wake_up(ws_wait);
1294 }
1295
put_workspace(int type,struct list_head * ws)1296 static void put_workspace(int type, struct list_head *ws)
1297 {
1298 switch (type) {
1299 case BTRFS_COMPRESS_NONE: return btrfs_put_workspace(type, ws);
1300 case BTRFS_COMPRESS_ZLIB: return btrfs_put_workspace(type, ws);
1301 case BTRFS_COMPRESS_LZO: return btrfs_put_workspace(type, ws);
1302 case BTRFS_COMPRESS_ZSTD: return zstd_put_workspace(ws);
1303 default:
1304 /*
1305 * This can't happen, the type is validated several times
1306 * before we get here.
1307 */
1308 BUG();
1309 }
1310 }
1311
1312 /*
1313 * Adjust @level according to the limits of the compression algorithm or
1314 * fallback to default
1315 */
btrfs_compress_set_level(int type,unsigned level)1316 static unsigned int btrfs_compress_set_level(int type, unsigned level)
1317 {
1318 const struct btrfs_compress_op *ops = btrfs_compress_op[type];
1319
1320 if (level == 0)
1321 level = ops->default_level;
1322 else
1323 level = min(level, ops->max_level);
1324
1325 return level;
1326 }
1327
1328 /*
1329 * Given an address space and start and length, compress the bytes into @pages
1330 * that are allocated on demand.
1331 *
1332 * @type_level is encoded algorithm and level, where level 0 means whatever
1333 * default the algorithm chooses and is opaque here;
1334 * - compression algo are 0-3
1335 * - the level are bits 4-7
1336 *
1337 * @out_pages is an in/out parameter, holds maximum number of pages to allocate
1338 * and returns number of actually allocated pages
1339 *
1340 * @total_in is used to return the number of bytes actually read. It
1341 * may be smaller than the input length if we had to exit early because we
1342 * ran out of room in the pages array or because we cross the
1343 * max_out threshold.
1344 *
1345 * @total_out is an in/out parameter, must be set to the input length and will
1346 * be also used to return the total number of compressed bytes
1347 */
btrfs_compress_pages(unsigned int type_level,struct address_space * mapping,u64 start,struct page ** pages,unsigned long * out_pages,unsigned long * total_in,unsigned long * total_out)1348 int btrfs_compress_pages(unsigned int type_level, struct address_space *mapping,
1349 u64 start, struct page **pages,
1350 unsigned long *out_pages,
1351 unsigned long *total_in,
1352 unsigned long *total_out)
1353 {
1354 int type = btrfs_compress_type(type_level);
1355 int level = btrfs_compress_level(type_level);
1356 struct list_head *workspace;
1357 int ret;
1358
1359 level = btrfs_compress_set_level(type, level);
1360 workspace = get_workspace(type, level);
1361 ret = compression_compress_pages(type, workspace, mapping, start, pages,
1362 out_pages, total_in, total_out);
1363 put_workspace(type, workspace);
1364 return ret;
1365 }
1366
btrfs_decompress_bio(struct compressed_bio * cb)1367 static int btrfs_decompress_bio(struct compressed_bio *cb)
1368 {
1369 struct list_head *workspace;
1370 int ret;
1371 int type = cb->compress_type;
1372
1373 workspace = get_workspace(type, 0);
1374 ret = compression_decompress_bio(workspace, cb);
1375 put_workspace(type, workspace);
1376
1377 return ret;
1378 }
1379
1380 /*
1381 * a less complex decompression routine. Our compressed data fits in a
1382 * single page, and we want to read a single page out of it.
1383 * start_byte tells us the offset into the compressed data we're interested in
1384 */
btrfs_decompress(int type,unsigned char * data_in,struct page * dest_page,unsigned long start_byte,size_t srclen,size_t destlen)1385 int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page,
1386 unsigned long start_byte, size_t srclen, size_t destlen)
1387 {
1388 struct list_head *workspace;
1389 int ret;
1390
1391 workspace = get_workspace(type, 0);
1392 ret = compression_decompress(type, workspace, data_in, dest_page,
1393 start_byte, srclen, destlen);
1394 put_workspace(type, workspace);
1395
1396 return ret;
1397 }
1398
btrfs_init_compress(void)1399 void __init btrfs_init_compress(void)
1400 {
1401 btrfs_init_workspace_manager(BTRFS_COMPRESS_NONE);
1402 btrfs_init_workspace_manager(BTRFS_COMPRESS_ZLIB);
1403 btrfs_init_workspace_manager(BTRFS_COMPRESS_LZO);
1404 zstd_init_workspace_manager();
1405 }
1406
btrfs_exit_compress(void)1407 void __cold btrfs_exit_compress(void)
1408 {
1409 btrfs_cleanup_workspace_manager(BTRFS_COMPRESS_NONE);
1410 btrfs_cleanup_workspace_manager(BTRFS_COMPRESS_ZLIB);
1411 btrfs_cleanup_workspace_manager(BTRFS_COMPRESS_LZO);
1412 zstd_cleanup_workspace_manager();
1413 }
1414
1415 /*
1416 * Copy decompressed data from working buffer to pages.
1417 *
1418 * @buf: The decompressed data buffer
1419 * @buf_len: The decompressed data length
1420 * @decompressed: Number of bytes that are already decompressed inside the
1421 * compressed extent
1422 * @cb: The compressed extent descriptor
1423 * @orig_bio: The original bio that the caller wants to read for
1424 *
1425 * An easier to understand graph is like below:
1426 *
1427 * |<- orig_bio ->| |<- orig_bio->|
1428 * |<------- full decompressed extent ----->|
1429 * |<----------- @cb range ---->|
1430 * | |<-- @buf_len -->|
1431 * |<--- @decompressed --->|
1432 *
1433 * Note that, @cb can be a subpage of the full decompressed extent, but
1434 * @cb->start always has the same as the orig_file_offset value of the full
1435 * decompressed extent.
1436 *
1437 * When reading compressed extent, we have to read the full compressed extent,
1438 * while @orig_bio may only want part of the range.
1439 * Thus this function will ensure only data covered by @orig_bio will be copied
1440 * to.
1441 *
1442 * Return 0 if we have copied all needed contents for @orig_bio.
1443 * Return >0 if we need continue decompress.
1444 */
btrfs_decompress_buf2page(const char * buf,u32 buf_len,struct compressed_bio * cb,u32 decompressed)1445 int btrfs_decompress_buf2page(const char *buf, u32 buf_len,
1446 struct compressed_bio *cb, u32 decompressed)
1447 {
1448 struct bio *orig_bio = cb->orig_bio;
1449 /* Offset inside the full decompressed extent */
1450 u32 cur_offset;
1451
1452 cur_offset = decompressed;
1453 /* The main loop to do the copy */
1454 while (cur_offset < decompressed + buf_len) {
1455 struct bio_vec bvec;
1456 size_t copy_len;
1457 u32 copy_start;
1458 /* Offset inside the full decompressed extent */
1459 u32 bvec_offset;
1460
1461 bvec = bio_iter_iovec(orig_bio, orig_bio->bi_iter);
1462 /*
1463 * cb->start may underflow, but subtracting that value can still
1464 * give us correct offset inside the full decompressed extent.
1465 */
1466 bvec_offset = page_offset(bvec.bv_page) + bvec.bv_offset - cb->start;
1467
1468 /* Haven't reached the bvec range, exit */
1469 if (decompressed + buf_len <= bvec_offset)
1470 return 1;
1471
1472 copy_start = max(cur_offset, bvec_offset);
1473 copy_len = min(bvec_offset + bvec.bv_len,
1474 decompressed + buf_len) - copy_start;
1475 ASSERT(copy_len);
1476
1477 /*
1478 * Extra range check to ensure we didn't go beyond
1479 * @buf + @buf_len.
1480 */
1481 ASSERT(copy_start - decompressed < buf_len);
1482 memcpy_to_page(bvec.bv_page, bvec.bv_offset,
1483 buf + copy_start - decompressed, copy_len);
1484 flush_dcache_page(bvec.bv_page);
1485 cur_offset += copy_len;
1486
1487 bio_advance(orig_bio, copy_len);
1488 /* Finished the bio */
1489 if (!orig_bio->bi_iter.bi_size)
1490 return 0;
1491 }
1492 return 1;
1493 }
1494
1495 /*
1496 * Shannon Entropy calculation
1497 *
1498 * Pure byte distribution analysis fails to determine compressibility of data.
1499 * Try calculating entropy to estimate the average minimum number of bits
1500 * needed to encode the sampled data.
1501 *
1502 * For convenience, return the percentage of needed bits, instead of amount of
1503 * bits directly.
1504 *
1505 * @ENTROPY_LVL_ACEPTABLE - below that threshold, sample has low byte entropy
1506 * and can be compressible with high probability
1507 *
1508 * @ENTROPY_LVL_HIGH - data are not compressible with high probability
1509 *
1510 * Use of ilog2() decreases precision, we lower the LVL to 5 to compensate.
1511 */
1512 #define ENTROPY_LVL_ACEPTABLE (65)
1513 #define ENTROPY_LVL_HIGH (80)
1514
1515 /*
1516 * For increasead precision in shannon_entropy calculation,
1517 * let's do pow(n, M) to save more digits after comma:
1518 *
1519 * - maximum int bit length is 64
1520 * - ilog2(MAX_SAMPLE_SIZE) -> 13
1521 * - 13 * 4 = 52 < 64 -> M = 4
1522 *
1523 * So use pow(n, 4).
1524 */
ilog2_w(u64 n)1525 static inline u32 ilog2_w(u64 n)
1526 {
1527 return ilog2(n * n * n * n);
1528 }
1529
shannon_entropy(struct heuristic_ws * ws)1530 static u32 shannon_entropy(struct heuristic_ws *ws)
1531 {
1532 const u32 entropy_max = 8 * ilog2_w(2);
1533 u32 entropy_sum = 0;
1534 u32 p, p_base, sz_base;
1535 u32 i;
1536
1537 sz_base = ilog2_w(ws->sample_size);
1538 for (i = 0; i < BUCKET_SIZE && ws->bucket[i].count > 0; i++) {
1539 p = ws->bucket[i].count;
1540 p_base = ilog2_w(p);
1541 entropy_sum += p * (sz_base - p_base);
1542 }
1543
1544 entropy_sum /= ws->sample_size;
1545 return entropy_sum * 100 / entropy_max;
1546 }
1547
1548 #define RADIX_BASE 4U
1549 #define COUNTERS_SIZE (1U << RADIX_BASE)
1550
get4bits(u64 num,int shift)1551 static u8 get4bits(u64 num, int shift) {
1552 u8 low4bits;
1553
1554 num >>= shift;
1555 /* Reverse order */
1556 low4bits = (COUNTERS_SIZE - 1) - (num % COUNTERS_SIZE);
1557 return low4bits;
1558 }
1559
1560 /*
1561 * Use 4 bits as radix base
1562 * Use 16 u32 counters for calculating new position in buf array
1563 *
1564 * @array - array that will be sorted
1565 * @array_buf - buffer array to store sorting results
1566 * must be equal in size to @array
1567 * @num - array size
1568 */
radix_sort(struct bucket_item * array,struct bucket_item * array_buf,int num)1569 static void radix_sort(struct bucket_item *array, struct bucket_item *array_buf,
1570 int num)
1571 {
1572 u64 max_num;
1573 u64 buf_num;
1574 u32 counters[COUNTERS_SIZE];
1575 u32 new_addr;
1576 u32 addr;
1577 int bitlen;
1578 int shift;
1579 int i;
1580
1581 /*
1582 * Try avoid useless loop iterations for small numbers stored in big
1583 * counters. Example: 48 33 4 ... in 64bit array
1584 */
1585 max_num = array[0].count;
1586 for (i = 1; i < num; i++) {
1587 buf_num = array[i].count;
1588 if (buf_num > max_num)
1589 max_num = buf_num;
1590 }
1591
1592 buf_num = ilog2(max_num);
1593 bitlen = ALIGN(buf_num, RADIX_BASE * 2);
1594
1595 shift = 0;
1596 while (shift < bitlen) {
1597 memset(counters, 0, sizeof(counters));
1598
1599 for (i = 0; i < num; i++) {
1600 buf_num = array[i].count;
1601 addr = get4bits(buf_num, shift);
1602 counters[addr]++;
1603 }
1604
1605 for (i = 1; i < COUNTERS_SIZE; i++)
1606 counters[i] += counters[i - 1];
1607
1608 for (i = num - 1; i >= 0; i--) {
1609 buf_num = array[i].count;
1610 addr = get4bits(buf_num, shift);
1611 counters[addr]--;
1612 new_addr = counters[addr];
1613 array_buf[new_addr] = array[i];
1614 }
1615
1616 shift += RADIX_BASE;
1617
1618 /*
1619 * Normal radix expects to move data from a temporary array, to
1620 * the main one. But that requires some CPU time. Avoid that
1621 * by doing another sort iteration to original array instead of
1622 * memcpy()
1623 */
1624 memset(counters, 0, sizeof(counters));
1625
1626 for (i = 0; i < num; i ++) {
1627 buf_num = array_buf[i].count;
1628 addr = get4bits(buf_num, shift);
1629 counters[addr]++;
1630 }
1631
1632 for (i = 1; i < COUNTERS_SIZE; i++)
1633 counters[i] += counters[i - 1];
1634
1635 for (i = num - 1; i >= 0; i--) {
1636 buf_num = array_buf[i].count;
1637 addr = get4bits(buf_num, shift);
1638 counters[addr]--;
1639 new_addr = counters[addr];
1640 array[new_addr] = array_buf[i];
1641 }
1642
1643 shift += RADIX_BASE;
1644 }
1645 }
1646
1647 /*
1648 * Size of the core byte set - how many bytes cover 90% of the sample
1649 *
1650 * There are several types of structured binary data that use nearly all byte
1651 * values. The distribution can be uniform and counts in all buckets will be
1652 * nearly the same (eg. encrypted data). Unlikely to be compressible.
1653 *
1654 * Other possibility is normal (Gaussian) distribution, where the data could
1655 * be potentially compressible, but we have to take a few more steps to decide
1656 * how much.
1657 *
1658 * @BYTE_CORE_SET_LOW - main part of byte values repeated frequently,
1659 * compression algo can easy fix that
1660 * @BYTE_CORE_SET_HIGH - data have uniform distribution and with high
1661 * probability is not compressible
1662 */
1663 #define BYTE_CORE_SET_LOW (64)
1664 #define BYTE_CORE_SET_HIGH (200)
1665
byte_core_set_size(struct heuristic_ws * ws)1666 static int byte_core_set_size(struct heuristic_ws *ws)
1667 {
1668 u32 i;
1669 u32 coreset_sum = 0;
1670 const u32 core_set_threshold = ws->sample_size * 90 / 100;
1671 struct bucket_item *bucket = ws->bucket;
1672
1673 /* Sort in reverse order */
1674 radix_sort(ws->bucket, ws->bucket_b, BUCKET_SIZE);
1675
1676 for (i = 0; i < BYTE_CORE_SET_LOW; i++)
1677 coreset_sum += bucket[i].count;
1678
1679 if (coreset_sum > core_set_threshold)
1680 return i;
1681
1682 for (; i < BYTE_CORE_SET_HIGH && bucket[i].count > 0; i++) {
1683 coreset_sum += bucket[i].count;
1684 if (coreset_sum > core_set_threshold)
1685 break;
1686 }
1687
1688 return i;
1689 }
1690
1691 /*
1692 * Count byte values in buckets.
1693 * This heuristic can detect textual data (configs, xml, json, html, etc).
1694 * Because in most text-like data byte set is restricted to limited number of
1695 * possible characters, and that restriction in most cases makes data easy to
1696 * compress.
1697 *
1698 * @BYTE_SET_THRESHOLD - consider all data within this byte set size:
1699 * less - compressible
1700 * more - need additional analysis
1701 */
1702 #define BYTE_SET_THRESHOLD (64)
1703
byte_set_size(const struct heuristic_ws * ws)1704 static u32 byte_set_size(const struct heuristic_ws *ws)
1705 {
1706 u32 i;
1707 u32 byte_set_size = 0;
1708
1709 for (i = 0; i < BYTE_SET_THRESHOLD; i++) {
1710 if (ws->bucket[i].count > 0)
1711 byte_set_size++;
1712 }
1713
1714 /*
1715 * Continue collecting count of byte values in buckets. If the byte
1716 * set size is bigger then the threshold, it's pointless to continue,
1717 * the detection technique would fail for this type of data.
1718 */
1719 for (; i < BUCKET_SIZE; i++) {
1720 if (ws->bucket[i].count > 0) {
1721 byte_set_size++;
1722 if (byte_set_size > BYTE_SET_THRESHOLD)
1723 return byte_set_size;
1724 }
1725 }
1726
1727 return byte_set_size;
1728 }
1729
sample_repeated_patterns(struct heuristic_ws * ws)1730 static bool sample_repeated_patterns(struct heuristic_ws *ws)
1731 {
1732 const u32 half_of_sample = ws->sample_size / 2;
1733 const u8 *data = ws->sample;
1734
1735 return memcmp(&data[0], &data[half_of_sample], half_of_sample) == 0;
1736 }
1737
heuristic_collect_sample(struct inode * inode,u64 start,u64 end,struct heuristic_ws * ws)1738 static void heuristic_collect_sample(struct inode *inode, u64 start, u64 end,
1739 struct heuristic_ws *ws)
1740 {
1741 struct page *page;
1742 u64 index, index_end;
1743 u32 i, curr_sample_pos;
1744 u8 *in_data;
1745
1746 /*
1747 * Compression handles the input data by chunks of 128KiB
1748 * (defined by BTRFS_MAX_UNCOMPRESSED)
1749 *
1750 * We do the same for the heuristic and loop over the whole range.
1751 *
1752 * MAX_SAMPLE_SIZE - calculated under assumption that heuristic will
1753 * process no more than BTRFS_MAX_UNCOMPRESSED at a time.
1754 */
1755 if (end - start > BTRFS_MAX_UNCOMPRESSED)
1756 end = start + BTRFS_MAX_UNCOMPRESSED;
1757
1758 index = start >> PAGE_SHIFT;
1759 index_end = end >> PAGE_SHIFT;
1760
1761 /* Don't miss unaligned end */
1762 if (!IS_ALIGNED(end, PAGE_SIZE))
1763 index_end++;
1764
1765 curr_sample_pos = 0;
1766 while (index < index_end) {
1767 page = find_get_page(inode->i_mapping, index);
1768 in_data = kmap_local_page(page);
1769 /* Handle case where the start is not aligned to PAGE_SIZE */
1770 i = start % PAGE_SIZE;
1771 while (i < PAGE_SIZE - SAMPLING_READ_SIZE) {
1772 /* Don't sample any garbage from the last page */
1773 if (start > end - SAMPLING_READ_SIZE)
1774 break;
1775 memcpy(&ws->sample[curr_sample_pos], &in_data[i],
1776 SAMPLING_READ_SIZE);
1777 i += SAMPLING_INTERVAL;
1778 start += SAMPLING_INTERVAL;
1779 curr_sample_pos += SAMPLING_READ_SIZE;
1780 }
1781 kunmap_local(in_data);
1782 put_page(page);
1783
1784 index++;
1785 }
1786
1787 ws->sample_size = curr_sample_pos;
1788 }
1789
1790 /*
1791 * Compression heuristic.
1792 *
1793 * For now is's a naive and optimistic 'return true', we'll extend the logic to
1794 * quickly (compared to direct compression) detect data characteristics
1795 * (compressible/uncompressible) to avoid wasting CPU time on uncompressible
1796 * data.
1797 *
1798 * The following types of analysis can be performed:
1799 * - detect mostly zero data
1800 * - detect data with low "byte set" size (text, etc)
1801 * - detect data with low/high "core byte" set
1802 *
1803 * Return non-zero if the compression should be done, 0 otherwise.
1804 */
btrfs_compress_heuristic(struct inode * inode,u64 start,u64 end)1805 int btrfs_compress_heuristic(struct inode *inode, u64 start, u64 end)
1806 {
1807 struct list_head *ws_list = get_workspace(0, 0);
1808 struct heuristic_ws *ws;
1809 u32 i;
1810 u8 byte;
1811 int ret = 0;
1812
1813 ws = list_entry(ws_list, struct heuristic_ws, list);
1814
1815 heuristic_collect_sample(inode, start, end, ws);
1816
1817 if (sample_repeated_patterns(ws)) {
1818 ret = 1;
1819 goto out;
1820 }
1821
1822 memset(ws->bucket, 0, sizeof(*ws->bucket)*BUCKET_SIZE);
1823
1824 for (i = 0; i < ws->sample_size; i++) {
1825 byte = ws->sample[i];
1826 ws->bucket[byte].count++;
1827 }
1828
1829 i = byte_set_size(ws);
1830 if (i < BYTE_SET_THRESHOLD) {
1831 ret = 2;
1832 goto out;
1833 }
1834
1835 i = byte_core_set_size(ws);
1836 if (i <= BYTE_CORE_SET_LOW) {
1837 ret = 3;
1838 goto out;
1839 }
1840
1841 if (i >= BYTE_CORE_SET_HIGH) {
1842 ret = 0;
1843 goto out;
1844 }
1845
1846 i = shannon_entropy(ws);
1847 if (i <= ENTROPY_LVL_ACEPTABLE) {
1848 ret = 4;
1849 goto out;
1850 }
1851
1852 /*
1853 * For the levels below ENTROPY_LVL_HIGH, additional analysis would be
1854 * needed to give green light to compression.
1855 *
1856 * For now just assume that compression at that level is not worth the
1857 * resources because:
1858 *
1859 * 1. it is possible to defrag the data later
1860 *
1861 * 2. the data would turn out to be hardly compressible, eg. 150 byte
1862 * values, every bucket has counter at level ~54. The heuristic would
1863 * be confused. This can happen when data have some internal repeated
1864 * patterns like "abbacbbc...". This can be detected by analyzing
1865 * pairs of bytes, which is too costly.
1866 */
1867 if (i < ENTROPY_LVL_HIGH) {
1868 ret = 5;
1869 goto out;
1870 } else {
1871 ret = 0;
1872 goto out;
1873 }
1874
1875 out:
1876 put_workspace(0, ws_list);
1877 return ret;
1878 }
1879
1880 /*
1881 * Convert the compression suffix (eg. after "zlib" starting with ":") to
1882 * level, unrecognized string will set the default level
1883 */
btrfs_compress_str2level(unsigned int type,const char * str)1884 unsigned int btrfs_compress_str2level(unsigned int type, const char *str)
1885 {
1886 unsigned int level = 0;
1887 int ret;
1888
1889 if (!type)
1890 return 0;
1891
1892 if (str[0] == ':') {
1893 ret = kstrtouint(str + 1, 10, &level);
1894 if (ret)
1895 level = 0;
1896 }
1897
1898 level = btrfs_compress_set_level(type, level);
1899
1900 return level;
1901 }
1902