mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / btrfs / compression.c
blobccd9c709375e7fb7146453e1c2592edc804c791b
1 /*
2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/slab.h>
35 #include <linux/sched/mm.h>
36 #include "ctree.h"
37 #include "disk-io.h"
38 #include "transaction.h"
39 #include "btrfs_inode.h"
40 #include "volumes.h"
41 #include "ordered-data.h"
42 #include "compression.h"
43 #include "extent_io.h"
44 #include "extent_map.h"
46 static const char* const btrfs_compress_types[] = { "", "zlib", "lzo", "zstd" };
48 const char* btrfs_compress_type2str(enum btrfs_compression_type type)
50 switch (type) {
51 case BTRFS_COMPRESS_ZLIB:
52 case BTRFS_COMPRESS_LZO:
53 case BTRFS_COMPRESS_ZSTD:
54 case BTRFS_COMPRESS_NONE:
55 return btrfs_compress_types[type];
58 return NULL;
61 bool btrfs_compress_is_valid_type(const char *str, size_t len)
63 int i;
65 for (i = 1; i < ARRAY_SIZE(btrfs_compress_types); i++) {
66 size_t comp_len = strlen(btrfs_compress_types[i]);
68 if (len < comp_len)
69 continue;
71 if (!strncmp(btrfs_compress_types[i], str, comp_len))
72 return true;
74 return false;
77 static int btrfs_decompress_bio(struct compressed_bio *cb);
79 static inline int compressed_bio_size(struct btrfs_fs_info *fs_info,
80 unsigned long disk_size)
82 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
84 return sizeof(struct compressed_bio) +
85 (DIV_ROUND_UP(disk_size, fs_info->sectorsize)) * csum_size;
88 static int check_compressed_csum(struct btrfs_inode *inode,
89 struct compressed_bio *cb,
90 u64 disk_start)
92 int ret;
93 struct page *page;
94 unsigned long i;
95 char *kaddr;
96 u32 csum;
97 u32 *cb_sum = &cb->sums;
99 if (inode->flags & BTRFS_INODE_NODATASUM)
100 return 0;
102 for (i = 0; i < cb->nr_pages; i++) {
103 page = cb->compressed_pages[i];
104 csum = ~(u32)0;
106 kaddr = kmap_atomic(page);
107 csum = btrfs_csum_data(kaddr, csum, PAGE_SIZE);
108 btrfs_csum_final(csum, (u8 *)&csum);
109 kunmap_atomic(kaddr);
111 if (csum != *cb_sum) {
112 btrfs_print_data_csum_error(inode, disk_start, csum,
113 *cb_sum, cb->mirror_num);
114 ret = -EIO;
115 goto fail;
117 cb_sum++;
120 ret = 0;
121 fail:
122 return ret;
125 /* when we finish reading compressed pages from the disk, we
126 * decompress them and then run the bio end_io routines on the
127 * decompressed pages (in the inode address space).
129 * This allows the checksumming and other IO error handling routines
130 * to work normally
132 * The compressed pages are freed here, and it must be run
133 * in process context
135 static void end_compressed_bio_read(struct bio *bio)
137 struct compressed_bio *cb = bio->bi_private;
138 struct inode *inode;
139 struct page *page;
140 unsigned long index;
141 unsigned int mirror = btrfs_io_bio(bio)->mirror_num;
142 int ret = 0;
144 if (bio->bi_status)
145 cb->errors = 1;
147 /* if there are more bios still pending for this compressed
148 * extent, just exit
150 if (!refcount_dec_and_test(&cb->pending_bios))
151 goto out;
154 * Record the correct mirror_num in cb->orig_bio so that
155 * read-repair can work properly.
157 ASSERT(btrfs_io_bio(cb->orig_bio));
158 btrfs_io_bio(cb->orig_bio)->mirror_num = mirror;
159 cb->mirror_num = mirror;
162 * Some IO in this cb have failed, just skip checksum as there
163 * is no way it could be correct.
165 if (cb->errors == 1)
166 goto csum_failed;
168 inode = cb->inode;
169 ret = check_compressed_csum(BTRFS_I(inode), cb,
170 (u64)bio->bi_iter.bi_sector << 9);
171 if (ret)
172 goto csum_failed;
174 /* ok, we're the last bio for this extent, lets start
175 * the decompression.
177 ret = btrfs_decompress_bio(cb);
179 csum_failed:
180 if (ret)
181 cb->errors = 1;
183 /* release the compressed pages */
184 index = 0;
185 for (index = 0; index < cb->nr_pages; index++) {
186 page = cb->compressed_pages[index];
187 page->mapping = NULL;
188 put_page(page);
191 /* do io completion on the original bio */
192 if (cb->errors) {
193 bio_io_error(cb->orig_bio);
194 } else {
195 int i;
196 struct bio_vec *bvec;
199 * we have verified the checksum already, set page
200 * checked so the end_io handlers know about it
202 ASSERT(!bio_flagged(bio, BIO_CLONED));
203 bio_for_each_segment_all(bvec, cb->orig_bio, i)
204 SetPageChecked(bvec->bv_page);
206 bio_endio(cb->orig_bio);
209 /* finally free the cb struct */
210 kfree(cb->compressed_pages);
211 kfree(cb);
212 out:
213 bio_put(bio);
217 * Clear the writeback bits on all of the file
218 * pages for a compressed write
220 static noinline void end_compressed_writeback(struct inode *inode,
221 const struct compressed_bio *cb)
223 unsigned long index = cb->start >> PAGE_SHIFT;
224 unsigned long end_index = (cb->start + cb->len - 1) >> PAGE_SHIFT;
225 struct page *pages[16];
226 unsigned long nr_pages = end_index - index + 1;
227 int i;
228 int ret;
230 if (cb->errors)
231 mapping_set_error(inode->i_mapping, -EIO);
233 while (nr_pages > 0) {
234 ret = find_get_pages_contig(inode->i_mapping, index,
235 min_t(unsigned long,
236 nr_pages, ARRAY_SIZE(pages)), pages);
237 if (ret == 0) {
238 nr_pages -= 1;
239 index += 1;
240 continue;
242 for (i = 0; i < ret; i++) {
243 if (cb->errors)
244 SetPageError(pages[i]);
245 end_page_writeback(pages[i]);
246 put_page(pages[i]);
248 nr_pages -= ret;
249 index += ret;
251 /* the inode may be gone now */
255 * do the cleanup once all the compressed pages hit the disk.
256 * This will clear writeback on the file pages and free the compressed
257 * pages.
259 * This also calls the writeback end hooks for the file pages so that
260 * metadata and checksums can be updated in the file.
262 static void end_compressed_bio_write(struct bio *bio)
264 struct extent_io_tree *tree;
265 struct compressed_bio *cb = bio->bi_private;
266 struct inode *inode;
267 struct page *page;
268 unsigned long index;
270 if (bio->bi_status)
271 cb->errors = 1;
273 /* if there are more bios still pending for this compressed
274 * extent, just exit
276 if (!refcount_dec_and_test(&cb->pending_bios))
277 goto out;
279 /* ok, we're the last bio for this extent, step one is to
280 * call back into the FS and do all the end_io operations
282 inode = cb->inode;
283 tree = &BTRFS_I(inode)->io_tree;
284 cb->compressed_pages[0]->mapping = cb->inode->i_mapping;
285 tree->ops->writepage_end_io_hook(cb->compressed_pages[0],
286 cb->start,
287 cb->start + cb->len - 1,
288 NULL,
289 bio->bi_status ? 0 : 1);
290 cb->compressed_pages[0]->mapping = NULL;
292 end_compressed_writeback(inode, cb);
293 /* note, our inode could be gone now */
296 * release the compressed pages, these came from alloc_page and
297 * are not attached to the inode at all
299 index = 0;
300 for (index = 0; index < cb->nr_pages; index++) {
301 page = cb->compressed_pages[index];
302 page->mapping = NULL;
303 put_page(page);
306 /* finally free the cb struct */
307 kfree(cb->compressed_pages);
308 kfree(cb);
309 out:
310 bio_put(bio);
314 * worker function to build and submit bios for previously compressed pages.
315 * The corresponding pages in the inode should be marked for writeback
316 * and the compressed pages should have a reference on them for dropping
317 * when the IO is complete.
319 * This also checksums the file bytes and gets things ready for
320 * the end io hooks.
322 blk_status_t btrfs_submit_compressed_write(struct inode *inode, u64 start,
323 unsigned long len, u64 disk_start,
324 unsigned long compressed_len,
325 struct page **compressed_pages,
326 unsigned long nr_pages)
328 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
329 struct bio *bio = NULL;
330 struct compressed_bio *cb;
331 unsigned long bytes_left;
332 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
333 int pg_index = 0;
334 struct page *page;
335 u64 first_byte = disk_start;
336 struct block_device *bdev;
337 blk_status_t ret;
338 int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
340 WARN_ON(start & ((u64)PAGE_SIZE - 1));
341 cb = kmalloc(compressed_bio_size(fs_info, compressed_len), GFP_NOFS);
342 if (!cb)
343 return BLK_STS_RESOURCE;
344 refcount_set(&cb->pending_bios, 0);
345 cb->errors = 0;
346 cb->inode = inode;
347 cb->start = start;
348 cb->len = len;
349 cb->mirror_num = 0;
350 cb->compressed_pages = compressed_pages;
351 cb->compressed_len = compressed_len;
352 cb->orig_bio = NULL;
353 cb->nr_pages = nr_pages;
355 bdev = fs_info->fs_devices->latest_bdev;
357 bio = btrfs_bio_alloc(bdev, first_byte);
358 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
359 bio->bi_private = cb;
360 bio->bi_end_io = end_compressed_bio_write;
361 refcount_set(&cb->pending_bios, 1);
363 /* create and submit bios for the compressed pages */
364 bytes_left = compressed_len;
365 for (pg_index = 0; pg_index < cb->nr_pages; pg_index++) {
366 int submit = 0;
368 page = compressed_pages[pg_index];
369 page->mapping = inode->i_mapping;
370 if (bio->bi_iter.bi_size)
371 submit = io_tree->ops->merge_bio_hook(page, 0,
372 PAGE_SIZE,
373 bio, 0);
375 page->mapping = NULL;
376 if (submit || bio_add_page(bio, page, PAGE_SIZE, 0) <
377 PAGE_SIZE) {
378 bio_get(bio);
381 * inc the count before we submit the bio so
382 * we know the end IO handler won't happen before
383 * we inc the count. Otherwise, the cb might get
384 * freed before we're done setting it up
386 refcount_inc(&cb->pending_bios);
387 ret = btrfs_bio_wq_end_io(fs_info, bio,
388 BTRFS_WQ_ENDIO_DATA);
389 BUG_ON(ret); /* -ENOMEM */
391 if (!skip_sum) {
392 ret = btrfs_csum_one_bio(inode, bio, start, 1);
393 BUG_ON(ret); /* -ENOMEM */
396 ret = btrfs_map_bio(fs_info, bio, 0, 1);
397 if (ret) {
398 bio->bi_status = ret;
399 bio_endio(bio);
402 bio_put(bio);
404 bio = btrfs_bio_alloc(bdev, first_byte);
405 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
406 bio->bi_private = cb;
407 bio->bi_end_io = end_compressed_bio_write;
408 bio_add_page(bio, page, PAGE_SIZE, 0);
410 if (bytes_left < PAGE_SIZE) {
411 btrfs_info(fs_info,
412 "bytes left %lu compress len %lu nr %lu",
413 bytes_left, cb->compressed_len, cb->nr_pages);
415 bytes_left -= PAGE_SIZE;
416 first_byte += PAGE_SIZE;
417 cond_resched();
419 bio_get(bio);
421 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
422 BUG_ON(ret); /* -ENOMEM */
424 if (!skip_sum) {
425 ret = btrfs_csum_one_bio(inode, bio, start, 1);
426 BUG_ON(ret); /* -ENOMEM */
429 ret = btrfs_map_bio(fs_info, bio, 0, 1);
430 if (ret) {
431 bio->bi_status = ret;
432 bio_endio(bio);
435 bio_put(bio);
436 return 0;
439 static u64 bio_end_offset(struct bio *bio)
441 struct bio_vec *last = &bio->bi_io_vec[bio->bi_vcnt - 1];
443 return page_offset(last->bv_page) + last->bv_len + last->bv_offset;
446 static noinline int add_ra_bio_pages(struct inode *inode,
447 u64 compressed_end,
448 struct compressed_bio *cb)
450 unsigned long end_index;
451 unsigned long pg_index;
452 u64 last_offset;
453 u64 isize = i_size_read(inode);
454 int ret;
455 struct page *page;
456 unsigned long nr_pages = 0;
457 struct extent_map *em;
458 struct address_space *mapping = inode->i_mapping;
459 struct extent_map_tree *em_tree;
460 struct extent_io_tree *tree;
461 u64 end;
462 int misses = 0;
464 last_offset = bio_end_offset(cb->orig_bio);
465 em_tree = &BTRFS_I(inode)->extent_tree;
466 tree = &BTRFS_I(inode)->io_tree;
468 if (isize == 0)
469 return 0;
471 end_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
473 while (last_offset < compressed_end) {
474 pg_index = last_offset >> PAGE_SHIFT;
476 if (pg_index > end_index)
477 break;
479 rcu_read_lock();
480 page = radix_tree_lookup(&mapping->page_tree, pg_index);
481 rcu_read_unlock();
482 if (page && !radix_tree_exceptional_entry(page)) {
483 misses++;
484 if (misses > 4)
485 break;
486 goto next;
489 page = __page_cache_alloc(mapping_gfp_constraint(mapping,
490 ~__GFP_FS));
491 if (!page)
492 break;
494 if (add_to_page_cache_lru(page, mapping, pg_index, GFP_NOFS)) {
495 put_page(page);
496 goto next;
499 end = last_offset + PAGE_SIZE - 1;
501 * at this point, we have a locked page in the page cache
502 * for these bytes in the file. But, we have to make
503 * sure they map to this compressed extent on disk.
505 set_page_extent_mapped(page);
506 lock_extent(tree, last_offset, end);
507 read_lock(&em_tree->lock);
508 em = lookup_extent_mapping(em_tree, last_offset,
509 PAGE_SIZE);
510 read_unlock(&em_tree->lock);
512 if (!em || last_offset < em->start ||
513 (last_offset + PAGE_SIZE > extent_map_end(em)) ||
514 (em->block_start >> 9) != cb->orig_bio->bi_iter.bi_sector) {
515 free_extent_map(em);
516 unlock_extent(tree, last_offset, end);
517 unlock_page(page);
518 put_page(page);
519 break;
521 free_extent_map(em);
523 if (page->index == end_index) {
524 char *userpage;
525 size_t zero_offset = isize & (PAGE_SIZE - 1);
527 if (zero_offset) {
528 int zeros;
529 zeros = PAGE_SIZE - zero_offset;
530 userpage = kmap_atomic(page);
531 memset(userpage + zero_offset, 0, zeros);
532 flush_dcache_page(page);
533 kunmap_atomic(userpage);
537 ret = bio_add_page(cb->orig_bio, page,
538 PAGE_SIZE, 0);
540 if (ret == PAGE_SIZE) {
541 nr_pages++;
542 put_page(page);
543 } else {
544 unlock_extent(tree, last_offset, end);
545 unlock_page(page);
546 put_page(page);
547 break;
549 next:
550 last_offset += PAGE_SIZE;
552 return 0;
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.
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
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
566 blk_status_t btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
567 int mirror_num, unsigned long bio_flags)
569 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
570 struct extent_io_tree *tree;
571 struct extent_map_tree *em_tree;
572 struct compressed_bio *cb;
573 unsigned long compressed_len;
574 unsigned long nr_pages;
575 unsigned long pg_index;
576 struct page *page;
577 struct block_device *bdev;
578 struct bio *comp_bio;
579 u64 cur_disk_byte = (u64)bio->bi_iter.bi_sector << 9;
580 u64 em_len;
581 u64 em_start;
582 struct extent_map *em;
583 blk_status_t ret = BLK_STS_RESOURCE;
584 int faili = 0;
585 u32 *sums;
587 tree = &BTRFS_I(inode)->io_tree;
588 em_tree = &BTRFS_I(inode)->extent_tree;
590 /* we need the actual starting offset of this extent in the file */
591 read_lock(&em_tree->lock);
592 em = lookup_extent_mapping(em_tree,
593 page_offset(bio->bi_io_vec->bv_page),
594 PAGE_SIZE);
595 read_unlock(&em_tree->lock);
596 if (!em)
597 return BLK_STS_IOERR;
599 compressed_len = em->block_len;
600 cb = kmalloc(compressed_bio_size(fs_info, compressed_len), GFP_NOFS);
601 if (!cb)
602 goto out;
604 refcount_set(&cb->pending_bios, 0);
605 cb->errors = 0;
606 cb->inode = inode;
607 cb->mirror_num = mirror_num;
608 sums = &cb->sums;
610 cb->start = em->orig_start;
611 em_len = em->len;
612 em_start = em->start;
614 free_extent_map(em);
615 em = NULL;
617 cb->len = bio->bi_iter.bi_size;
618 cb->compressed_len = compressed_len;
619 cb->compress_type = extent_compress_type(bio_flags);
620 cb->orig_bio = bio;
622 nr_pages = DIV_ROUND_UP(compressed_len, PAGE_SIZE);
623 cb->compressed_pages = kcalloc(nr_pages, sizeof(struct page *),
624 GFP_NOFS);
625 if (!cb->compressed_pages)
626 goto fail1;
628 bdev = fs_info->fs_devices->latest_bdev;
630 for (pg_index = 0; pg_index < nr_pages; pg_index++) {
631 cb->compressed_pages[pg_index] = alloc_page(GFP_NOFS |
632 __GFP_HIGHMEM);
633 if (!cb->compressed_pages[pg_index]) {
634 faili = pg_index - 1;
635 ret = BLK_STS_RESOURCE;
636 goto fail2;
639 faili = nr_pages - 1;
640 cb->nr_pages = nr_pages;
642 add_ra_bio_pages(inode, em_start + em_len, cb);
644 /* include any pages we added in add_ra-bio_pages */
645 cb->len = bio->bi_iter.bi_size;
647 comp_bio = btrfs_bio_alloc(bdev, cur_disk_byte);
648 bio_set_op_attrs (comp_bio, REQ_OP_READ, 0);
649 comp_bio->bi_private = cb;
650 comp_bio->bi_end_io = end_compressed_bio_read;
651 refcount_set(&cb->pending_bios, 1);
653 for (pg_index = 0; pg_index < nr_pages; pg_index++) {
654 int submit = 0;
656 page = cb->compressed_pages[pg_index];
657 page->mapping = inode->i_mapping;
658 page->index = em_start >> PAGE_SHIFT;
660 if (comp_bio->bi_iter.bi_size)
661 submit = tree->ops->merge_bio_hook(page, 0,
662 PAGE_SIZE,
663 comp_bio, 0);
665 page->mapping = NULL;
666 if (submit || bio_add_page(comp_bio, page, PAGE_SIZE, 0) <
667 PAGE_SIZE) {
668 bio_get(comp_bio);
670 ret = btrfs_bio_wq_end_io(fs_info, comp_bio,
671 BTRFS_WQ_ENDIO_DATA);
672 BUG_ON(ret); /* -ENOMEM */
675 * inc the count before we submit the bio so
676 * we know the end IO handler won't happen before
677 * we inc the count. Otherwise, the cb might get
678 * freed before we're done setting it up
680 refcount_inc(&cb->pending_bios);
682 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
683 ret = btrfs_lookup_bio_sums(inode, comp_bio,
684 sums);
685 BUG_ON(ret); /* -ENOMEM */
687 sums += DIV_ROUND_UP(comp_bio->bi_iter.bi_size,
688 fs_info->sectorsize);
690 ret = btrfs_map_bio(fs_info, comp_bio, mirror_num, 0);
691 if (ret) {
692 comp_bio->bi_status = ret;
693 bio_endio(comp_bio);
696 bio_put(comp_bio);
698 comp_bio = btrfs_bio_alloc(bdev, cur_disk_byte);
699 bio_set_op_attrs(comp_bio, REQ_OP_READ, 0);
700 comp_bio->bi_private = cb;
701 comp_bio->bi_end_io = end_compressed_bio_read;
703 bio_add_page(comp_bio, page, PAGE_SIZE, 0);
705 cur_disk_byte += PAGE_SIZE;
707 bio_get(comp_bio);
709 ret = btrfs_bio_wq_end_io(fs_info, comp_bio, BTRFS_WQ_ENDIO_DATA);
710 BUG_ON(ret); /* -ENOMEM */
712 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
713 ret = btrfs_lookup_bio_sums(inode, comp_bio, sums);
714 BUG_ON(ret); /* -ENOMEM */
717 ret = btrfs_map_bio(fs_info, comp_bio, mirror_num, 0);
718 if (ret) {
719 comp_bio->bi_status = ret;
720 bio_endio(comp_bio);
723 bio_put(comp_bio);
724 return 0;
726 fail2:
727 while (faili >= 0) {
728 __free_page(cb->compressed_pages[faili]);
729 faili--;
732 kfree(cb->compressed_pages);
733 fail1:
734 kfree(cb);
735 out:
736 free_extent_map(em);
737 return ret;
740 static struct {
741 struct list_head idle_ws;
742 spinlock_t ws_lock;
743 /* Number of free workspaces */
744 int free_ws;
745 /* Total number of allocated workspaces */
746 atomic_t total_ws;
747 /* Waiters for a free workspace */
748 wait_queue_head_t ws_wait;
749 } btrfs_comp_ws[BTRFS_COMPRESS_TYPES];
751 static const struct btrfs_compress_op * const btrfs_compress_op[] = {
752 &btrfs_zlib_compress,
753 &btrfs_lzo_compress,
754 &btrfs_zstd_compress,
757 void __init btrfs_init_compress(void)
759 int i;
761 for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) {
762 struct list_head *workspace;
764 INIT_LIST_HEAD(&btrfs_comp_ws[i].idle_ws);
765 spin_lock_init(&btrfs_comp_ws[i].ws_lock);
766 atomic_set(&btrfs_comp_ws[i].total_ws, 0);
767 init_waitqueue_head(&btrfs_comp_ws[i].ws_wait);
770 * Preallocate one workspace for each compression type so
771 * we can guarantee forward progress in the worst case
773 workspace = btrfs_compress_op[i]->alloc_workspace();
774 if (IS_ERR(workspace)) {
775 pr_warn("BTRFS: cannot preallocate compression workspace, will try later\n");
776 } else {
777 atomic_set(&btrfs_comp_ws[i].total_ws, 1);
778 btrfs_comp_ws[i].free_ws = 1;
779 list_add(workspace, &btrfs_comp_ws[i].idle_ws);
785 * This finds an available workspace or allocates a new one.
786 * If it's not possible to allocate a new one, waits until there's one.
787 * Preallocation makes a forward progress guarantees and we do not return
788 * errors.
790 static struct list_head *find_workspace(int type)
792 struct list_head *workspace;
793 int cpus = num_online_cpus();
794 int idx = type - 1;
795 unsigned nofs_flag;
797 struct list_head *idle_ws = &btrfs_comp_ws[idx].idle_ws;
798 spinlock_t *ws_lock = &btrfs_comp_ws[idx].ws_lock;
799 atomic_t *total_ws = &btrfs_comp_ws[idx].total_ws;
800 wait_queue_head_t *ws_wait = &btrfs_comp_ws[idx].ws_wait;
801 int *free_ws = &btrfs_comp_ws[idx].free_ws;
802 again:
803 spin_lock(ws_lock);
804 if (!list_empty(idle_ws)) {
805 workspace = idle_ws->next;
806 list_del(workspace);
807 (*free_ws)--;
808 spin_unlock(ws_lock);
809 return workspace;
812 if (atomic_read(total_ws) > cpus) {
813 DEFINE_WAIT(wait);
815 spin_unlock(ws_lock);
816 prepare_to_wait(ws_wait, &wait, TASK_UNINTERRUPTIBLE);
817 if (atomic_read(total_ws) > cpus && !*free_ws)
818 schedule();
819 finish_wait(ws_wait, &wait);
820 goto again;
822 atomic_inc(total_ws);
823 spin_unlock(ws_lock);
826 * Allocation helpers call vmalloc that can't use GFP_NOFS, so we have
827 * to turn it off here because we might get called from the restricted
828 * context of btrfs_compress_bio/btrfs_compress_pages
830 nofs_flag = memalloc_nofs_save();
831 workspace = btrfs_compress_op[idx]->alloc_workspace();
832 memalloc_nofs_restore(nofs_flag);
834 if (IS_ERR(workspace)) {
835 atomic_dec(total_ws);
836 wake_up(ws_wait);
839 * Do not return the error but go back to waiting. There's a
840 * workspace preallocated for each type and the compression
841 * time is bounded so we get to a workspace eventually. This
842 * makes our caller's life easier.
844 * To prevent silent and low-probability deadlocks (when the
845 * initial preallocation fails), check if there are any
846 * workspaces at all.
848 if (atomic_read(total_ws) == 0) {
849 static DEFINE_RATELIMIT_STATE(_rs,
850 /* once per minute */ 60 * HZ,
851 /* no burst */ 1);
853 if (__ratelimit(&_rs)) {
854 pr_warn("BTRFS: no compression workspaces, low memory, retrying\n");
857 goto again;
859 return workspace;
863 * put a workspace struct back on the list or free it if we have enough
864 * idle ones sitting around
866 static void free_workspace(int type, struct list_head *workspace)
868 int idx = type - 1;
869 struct list_head *idle_ws = &btrfs_comp_ws[idx].idle_ws;
870 spinlock_t *ws_lock = &btrfs_comp_ws[idx].ws_lock;
871 atomic_t *total_ws = &btrfs_comp_ws[idx].total_ws;
872 wait_queue_head_t *ws_wait = &btrfs_comp_ws[idx].ws_wait;
873 int *free_ws = &btrfs_comp_ws[idx].free_ws;
875 spin_lock(ws_lock);
876 if (*free_ws <= num_online_cpus()) {
877 list_add(workspace, idle_ws);
878 (*free_ws)++;
879 spin_unlock(ws_lock);
880 goto wake;
882 spin_unlock(ws_lock);
884 btrfs_compress_op[idx]->free_workspace(workspace);
885 atomic_dec(total_ws);
886 wake:
888 * Make sure counter is updated before we wake up waiters.
890 smp_mb();
891 if (waitqueue_active(ws_wait))
892 wake_up(ws_wait);
896 * cleanup function for module exit
898 static void free_workspaces(void)
900 struct list_head *workspace;
901 int i;
903 for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) {
904 while (!list_empty(&btrfs_comp_ws[i].idle_ws)) {
905 workspace = btrfs_comp_ws[i].idle_ws.next;
906 list_del(workspace);
907 btrfs_compress_op[i]->free_workspace(workspace);
908 atomic_dec(&btrfs_comp_ws[i].total_ws);
914 * Given an address space and start and length, compress the bytes into @pages
915 * that are allocated on demand.
917 * @out_pages is an in/out parameter, holds maximum number of pages to allocate
918 * and returns number of actually allocated pages
920 * @total_in is used to return the number of bytes actually read. It
921 * may be smaller than the input length if we had to exit early because we
922 * ran out of room in the pages array or because we cross the
923 * max_out threshold.
925 * @total_out is an in/out parameter, must be set to the input length and will
926 * be also used to return the total number of compressed bytes
928 * @max_out tells us the max number of bytes that we're allowed to
929 * stuff into pages
931 int btrfs_compress_pages(int type, struct address_space *mapping,
932 u64 start, struct page **pages,
933 unsigned long *out_pages,
934 unsigned long *total_in,
935 unsigned long *total_out)
937 struct list_head *workspace;
938 int ret;
940 workspace = find_workspace(type);
942 ret = btrfs_compress_op[type-1]->compress_pages(workspace, mapping,
943 start, pages,
944 out_pages,
945 total_in, total_out);
946 free_workspace(type, workspace);
947 return ret;
951 * pages_in is an array of pages with compressed data.
953 * disk_start is the starting logical offset of this array in the file
955 * orig_bio contains the pages from the file that we want to decompress into
957 * srclen is the number of bytes in pages_in
959 * The basic idea is that we have a bio that was created by readpages.
960 * The pages in the bio are for the uncompressed data, and they may not
961 * be contiguous. They all correspond to the range of bytes covered by
962 * the compressed extent.
964 static int btrfs_decompress_bio(struct compressed_bio *cb)
966 struct list_head *workspace;
967 int ret;
968 int type = cb->compress_type;
970 workspace = find_workspace(type);
971 ret = btrfs_compress_op[type - 1]->decompress_bio(workspace, cb);
972 free_workspace(type, workspace);
974 return ret;
978 * a less complex decompression routine. Our compressed data fits in a
979 * single page, and we want to read a single page out of it.
980 * start_byte tells us the offset into the compressed data we're interested in
982 int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page,
983 unsigned long start_byte, size_t srclen, size_t destlen)
985 struct list_head *workspace;
986 int ret;
988 workspace = find_workspace(type);
990 ret = btrfs_compress_op[type-1]->decompress(workspace, data_in,
991 dest_page, start_byte,
992 srclen, destlen);
994 free_workspace(type, workspace);
995 return ret;
998 void btrfs_exit_compress(void)
1000 free_workspaces();
1004 * Copy uncompressed data from working buffer to pages.
1006 * buf_start is the byte offset we're of the start of our workspace buffer.
1008 * total_out is the last byte of the buffer
1010 int btrfs_decompress_buf2page(const char *buf, unsigned long buf_start,
1011 unsigned long total_out, u64 disk_start,
1012 struct bio *bio)
1014 unsigned long buf_offset;
1015 unsigned long current_buf_start;
1016 unsigned long start_byte;
1017 unsigned long prev_start_byte;
1018 unsigned long working_bytes = total_out - buf_start;
1019 unsigned long bytes;
1020 char *kaddr;
1021 struct bio_vec bvec = bio_iter_iovec(bio, bio->bi_iter);
1024 * start byte is the first byte of the page we're currently
1025 * copying into relative to the start of the compressed data.
1027 start_byte = page_offset(bvec.bv_page) - disk_start;
1029 /* we haven't yet hit data corresponding to this page */
1030 if (total_out <= start_byte)
1031 return 1;
1034 * the start of the data we care about is offset into
1035 * the middle of our working buffer
1037 if (total_out > start_byte && buf_start < start_byte) {
1038 buf_offset = start_byte - buf_start;
1039 working_bytes -= buf_offset;
1040 } else {
1041 buf_offset = 0;
1043 current_buf_start = buf_start;
1045 /* copy bytes from the working buffer into the pages */
1046 while (working_bytes > 0) {
1047 bytes = min_t(unsigned long, bvec.bv_len,
1048 PAGE_SIZE - buf_offset);
1049 bytes = min(bytes, working_bytes);
1051 kaddr = kmap_atomic(bvec.bv_page);
1052 memcpy(kaddr + bvec.bv_offset, buf + buf_offset, bytes);
1053 kunmap_atomic(kaddr);
1054 flush_dcache_page(bvec.bv_page);
1056 buf_offset += bytes;
1057 working_bytes -= bytes;
1058 current_buf_start += bytes;
1060 /* check if we need to pick another page */
1061 bio_advance(bio, bytes);
1062 if (!bio->bi_iter.bi_size)
1063 return 0;
1064 bvec = bio_iter_iovec(bio, bio->bi_iter);
1065 prev_start_byte = start_byte;
1066 start_byte = page_offset(bvec.bv_page) - disk_start;
1069 * We need to make sure we're only adjusting
1070 * our offset into compression working buffer when
1071 * we're switching pages. Otherwise we can incorrectly
1072 * keep copying when we were actually done.
1074 if (start_byte != prev_start_byte) {
1076 * make sure our new page is covered by this
1077 * working buffer
1079 if (total_out <= start_byte)
1080 return 1;
1083 * the next page in the biovec might not be adjacent
1084 * to the last page, but it might still be found
1085 * inside this working buffer. bump our offset pointer
1087 if (total_out > start_byte &&
1088 current_buf_start < start_byte) {
1089 buf_offset = start_byte - buf_start;
1090 working_bytes = total_out - start_byte;
1091 current_buf_start = buf_start + buf_offset;
1096 return 1;
1100 * Compression heuristic.
1102 * For now is's a naive and optimistic 'return true', we'll extend the logic to
1103 * quickly (compared to direct compression) detect data characteristics
1104 * (compressible/uncompressible) to avoid wasting CPU time on uncompressible
1105 * data.
1107 * The following types of analysis can be performed:
1108 * - detect mostly zero data
1109 * - detect data with low "byte set" size (text, etc)
1110 * - detect data with low/high "core byte" set
1112 * Return non-zero if the compression should be done, 0 otherwise.
1114 int btrfs_compress_heuristic(struct inode *inode, u64 start, u64 end)
1116 u64 index = start >> PAGE_SHIFT;
1117 u64 end_index = end >> PAGE_SHIFT;
1118 struct page *page;
1119 int ret = 1;
1121 while (index <= end_index) {
1122 page = find_get_page(inode->i_mapping, index);
1123 kmap(page);
1124 kunmap(page);
1125 put_page(page);
1126 index++;
1129 return ret;