diff options
Diffstat (limited to 'fs/btrfs/compression.c')
| -rw-r--r-- | fs/btrfs/compression.c | 2140 |
1 files changed, 1385 insertions, 755 deletions
diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c index b189bd1e7a3e..6b3357287b42 100644 --- a/fs/btrfs/compression.c +++ b/fs/btrfs/compression.c @@ -1,319 +1,360 @@ +// SPDX-License-Identifier: GPL-2.0 /* * Copyright (C) 2008 Oracle. All rights reserved. - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public - * License v2 as published by the Free Software Foundation. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * General Public License for more details. - * - * You should have received a copy of the GNU General Public - * License along with this program; if not, write to the - * Free Software Foundation, Inc., 59 Temple Place - Suite 330, - * Boston, MA 021110-1307, USA. */ #include <linux/kernel.h> #include <linux/bio.h> -#include <linux/buffer_head.h> #include <linux/file.h> #include <linux/fs.h> #include <linux/pagemap.h> +#include <linux/pagevec.h> #include <linux/highmem.h> +#include <linux/kthread.h> #include <linux/time.h> #include <linux/init.h> #include <linux/string.h> #include <linux/backing-dev.h> -#include <linux/mpage.h> -#include <linux/swap.h> #include <linux/writeback.h> -#include <linux/bit_spinlock.h> +#include <linux/psi.h> #include <linux/slab.h> -#include "compat.h" +#include <linux/sched/mm.h> +#include <linux/log2.h> +#include <linux/shrinker.h> +#include <crypto/hash.h> +#include "misc.h" #include "ctree.h" -#include "disk-io.h" -#include "transaction.h" +#include "fs.h" #include "btrfs_inode.h" -#include "volumes.h" +#include "bio.h" #include "ordered-data.h" #include "compression.h" #include "extent_io.h" #include "extent_map.h" +#include "subpage.h" +#include "messages.h" +#include "super.h" -struct compressed_bio { - /* number of bios pending for this compressed extent */ - atomic_t pending_bios; - - /* the pages with the compressed data on them */ - struct page **compressed_pages; +static struct bio_set btrfs_compressed_bioset; - /* inode that owns this data */ - struct inode *inode; +static const char* const btrfs_compress_types[] = { "", "zlib", "lzo", "zstd" }; - /* starting offset in the inode for our pages */ - u64 start; +const char* btrfs_compress_type2str(enum btrfs_compression_type type) +{ + switch (type) { + case BTRFS_COMPRESS_ZLIB: + case BTRFS_COMPRESS_LZO: + case BTRFS_COMPRESS_ZSTD: + case BTRFS_COMPRESS_NONE: + return btrfs_compress_types[type]; + default: + break; + } - /* number of bytes in the inode we're working on */ - unsigned long len; + return NULL; +} - /* number of bytes on disk */ - unsigned long compressed_len; +static inline struct compressed_bio *to_compressed_bio(struct btrfs_bio *bbio) +{ + return container_of(bbio, struct compressed_bio, bbio); +} - /* the compression algorithm for this bio */ - int compress_type; +static struct compressed_bio *alloc_compressed_bio(struct btrfs_inode *inode, + u64 start, blk_opf_t op, + btrfs_bio_end_io_t end_io) +{ + struct btrfs_bio *bbio; - /* number of compressed pages in the array */ - unsigned long nr_pages; + bbio = btrfs_bio(bio_alloc_bioset(NULL, BTRFS_MAX_COMPRESSED_PAGES, op, + GFP_NOFS, &btrfs_compressed_bioset)); + btrfs_bio_init(bbio, inode, start, end_io, NULL); + return to_compressed_bio(bbio); +} - /* IO errors */ - int errors; - int mirror_num; +bool btrfs_compress_is_valid_type(const char *str, size_t len) +{ + int i; - /* for reads, this is the bio we are copying the data into */ - struct bio *orig_bio; + for (i = 1; i < ARRAY_SIZE(btrfs_compress_types); i++) { + size_t comp_len = strlen(btrfs_compress_types[i]); - /* - * the start of a variable length array of checksums only - * used by reads - */ - u32 sums; -}; + if (len < comp_len) + continue; -static int btrfs_decompress_biovec(int type, struct page **pages_in, - u64 disk_start, struct bio_vec *bvec, - int vcnt, size_t srclen); + if (!strncmp(btrfs_compress_types[i], str, comp_len)) + return true; + } + return false; +} -static inline int compressed_bio_size(struct btrfs_root *root, - unsigned long disk_size) +static int compression_compress_pages(int type, struct list_head *ws, + struct btrfs_inode *inode, u64 start, + struct folio **folios, unsigned long *out_folios, + unsigned long *total_in, unsigned long *total_out) { - u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy); + switch (type) { + case BTRFS_COMPRESS_ZLIB: + return zlib_compress_folios(ws, inode, start, folios, + out_folios, total_in, total_out); + case BTRFS_COMPRESS_LZO: + return lzo_compress_folios(ws, inode, start, folios, + out_folios, total_in, total_out); + case BTRFS_COMPRESS_ZSTD: + return zstd_compress_folios(ws, inode, start, folios, + out_folios, total_in, total_out); + case BTRFS_COMPRESS_NONE: + default: + /* + * This can happen when compression races with remount setting + * it to 'no compress', while caller doesn't call + * inode_need_compress() to check if we really need to + * compress. + * + * Not a big deal, just need to inform caller that we + * haven't allocated any pages yet. + */ + *out_folios = 0; + return -E2BIG; + } +} - return sizeof(struct compressed_bio) + - ((disk_size + root->sectorsize - 1) / root->sectorsize) * - csum_size; +static int compression_decompress_bio(struct list_head *ws, + struct compressed_bio *cb) +{ + switch (cb->compress_type) { + case BTRFS_COMPRESS_ZLIB: return zlib_decompress_bio(ws, cb); + case BTRFS_COMPRESS_LZO: return lzo_decompress_bio(ws, cb); + case BTRFS_COMPRESS_ZSTD: return zstd_decompress_bio(ws, cb); + case BTRFS_COMPRESS_NONE: + default: + /* + * This can't happen, the type is validated several times + * before we get here. + */ + BUG(); + } } -static struct bio *compressed_bio_alloc(struct block_device *bdev, - u64 first_byte, gfp_t gfp_flags) +static int 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) { - int nr_vecs; + switch (type) { + case BTRFS_COMPRESS_ZLIB: return zlib_decompress(ws, data_in, dest_folio, + dest_pgoff, srclen, destlen); + case BTRFS_COMPRESS_LZO: return lzo_decompress(ws, data_in, dest_folio, + dest_pgoff, srclen, destlen); + case BTRFS_COMPRESS_ZSTD: return zstd_decompress(ws, data_in, dest_folio, + dest_pgoff, srclen, destlen); + case BTRFS_COMPRESS_NONE: + default: + /* + * This can't happen, the type is validated several times + * before we get here. + */ + BUG(); + } +} - nr_vecs = bio_get_nr_vecs(bdev); - return btrfs_bio_alloc(bdev, first_byte >> 9, nr_vecs, gfp_flags); +static void btrfs_free_compressed_folios(struct compressed_bio *cb) +{ + for (unsigned int i = 0; i < cb->nr_folios; i++) + btrfs_free_compr_folio(cb->compressed_folios[i]); + kfree(cb->compressed_folios); } -static int check_compressed_csum(struct inode *inode, - struct compressed_bio *cb, - u64 disk_start) +static int btrfs_decompress_bio(struct compressed_bio *cb); + +/* + * Global cache of last unused pages for compression/decompression. + */ +static struct btrfs_compr_pool { + struct shrinker *shrinker; + spinlock_t lock; + struct list_head list; + int count; + int thresh; +} compr_pool; + +static unsigned long btrfs_compr_pool_count(struct shrinker *sh, struct shrink_control *sc) { int ret; - struct page *page; - unsigned long i; - char *kaddr; - u32 csum; - u32 *cb_sum = &cb->sums; - if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) - return 0; + /* + * We must not read the values more than once if 'ret' gets expanded in + * the return statement so we don't accidentally return a negative + * number, even if the first condition finds it positive. + */ + ret = READ_ONCE(compr_pool.count) - READ_ONCE(compr_pool.thresh); - for (i = 0; i < cb->nr_pages; i++) { - page = cb->compressed_pages[i]; - csum = ~(u32)0; - - kaddr = kmap_atomic(page); - csum = btrfs_csum_data(kaddr, csum, PAGE_CACHE_SIZE); - btrfs_csum_final(csum, (char *)&csum); - kunmap_atomic(kaddr); - - if (csum != *cb_sum) { - printk(KERN_INFO "btrfs csum failed ino %llu " - "extent %llu csum %u " - "wanted %u mirror %d\n", - (unsigned long long)btrfs_ino(inode), - (unsigned long long)disk_start, - csum, *cb_sum, cb->mirror_num); - ret = -EIO; - goto fail; - } - cb_sum++; + return ret > 0 ? ret : 0; +} + +static unsigned long btrfs_compr_pool_scan(struct shrinker *sh, struct shrink_control *sc) +{ + LIST_HEAD(remove); + struct list_head *tmp, *next; + int freed; + + if (compr_pool.count == 0) + return SHRINK_STOP; + + /* For now, just simply drain the whole list. */ + spin_lock(&compr_pool.lock); + list_splice_init(&compr_pool.list, &remove); + freed = compr_pool.count; + compr_pool.count = 0; + spin_unlock(&compr_pool.lock); + list_for_each_safe(tmp, next, &remove) { + struct page *page = list_entry(tmp, struct page, lru); + + ASSERT(page_ref_count(page) == 1); + put_page(page); } - ret = 0; -fail: - return ret; + + return freed; } -/* when we finish reading compressed pages from the disk, we - * decompress them and then run the bio end_io routines on the - * decompressed pages (in the inode address space). - * - * This allows the checksumming and other IO error handling routines - * to work normally - * - * The compressed pages are freed here, and it must be run - * in process context +/* + * Common wrappers for page allocation from compression wrappers */ -static void end_compressed_bio_read(struct bio *bio, int err) +struct folio *btrfs_alloc_compr_folio(struct btrfs_fs_info *fs_info) { - struct compressed_bio *cb = bio->bi_private; - struct inode *inode; - struct page *page; - unsigned long index; - int ret; + struct folio *folio = NULL; - if (err) - cb->errors = 1; + /* For bs > ps cases, no cached folio pool for now. */ + if (fs_info->block_min_order) + goto alloc; - /* if there are more bios still pending for this compressed - * extent, just exit - */ - if (!atomic_dec_and_test(&cb->pending_bios)) - goto out; + spin_lock(&compr_pool.lock); + if (compr_pool.count > 0) { + folio = list_first_entry(&compr_pool.list, struct folio, lru); + list_del_init(&folio->lru); + compr_pool.count--; + } + spin_unlock(&compr_pool.lock); - inode = cb->inode; - ret = check_compressed_csum(inode, cb, (u64)bio->bi_sector << 9); - if (ret) - goto csum_failed; + if (folio) + return folio; - /* ok, we're the last bio for this extent, lets start - * the decompression. - */ - ret = btrfs_decompress_biovec(cb->compress_type, - cb->compressed_pages, - cb->start, - cb->orig_bio->bi_io_vec, - cb->orig_bio->bi_vcnt, - cb->compressed_len); -csum_failed: - if (ret) - cb->errors = 1; - - /* release the compressed pages */ - index = 0; - for (index = 0; index < cb->nr_pages; index++) { - page = cb->compressed_pages[index]; - page->mapping = NULL; - page_cache_release(page); - } +alloc: + return folio_alloc(GFP_NOFS, fs_info->block_min_order); +} - /* do io completion on the original bio */ - if (cb->errors) { - bio_io_error(cb->orig_bio); - } else { - int bio_index = 0; - struct bio_vec *bvec = cb->orig_bio->bi_io_vec; +void btrfs_free_compr_folio(struct folio *folio) +{ + bool do_free = false; - /* - * we have verified the checksum already, set page - * checked so the end_io handlers know about it - */ - while (bio_index < cb->orig_bio->bi_vcnt) { - SetPageChecked(bvec->bv_page); - bvec++; - bio_index++; - } - bio_endio(cb->orig_bio, 0); + /* The folio is from bs > ps fs, no cached pool for now. */ + if (folio_order(folio)) + goto free; + + spin_lock(&compr_pool.lock); + if (compr_pool.count > compr_pool.thresh) { + do_free = true; + } else { + list_add(&folio->lru, &compr_pool.list); + compr_pool.count++; } + spin_unlock(&compr_pool.lock); - /* finally free the cb struct */ - kfree(cb->compressed_pages); - kfree(cb); -out: - bio_put(bio); + if (!do_free) + return; + +free: + ASSERT(folio_ref_count(folio) == 1); + folio_put(folio); +} + +static void end_bbio_compressed_read(struct btrfs_bio *bbio) +{ + struct compressed_bio *cb = to_compressed_bio(bbio); + blk_status_t status = bbio->bio.bi_status; + + if (!status) + status = errno_to_blk_status(btrfs_decompress_bio(cb)); + + btrfs_free_compressed_folios(cb); + btrfs_bio_end_io(cb->orig_bbio, status); + bio_put(&bbio->bio); } /* * Clear the writeback bits on all of the file * pages for a compressed write */ -static noinline void end_compressed_writeback(struct inode *inode, u64 start, - unsigned long ram_size) +static noinline void end_compressed_writeback(const struct compressed_bio *cb) { - unsigned long index = start >> PAGE_CACHE_SHIFT; - unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT; - struct page *pages[16]; - unsigned long nr_pages = end_index - index + 1; + struct inode *inode = &cb->bbio.inode->vfs_inode; + struct btrfs_fs_info *fs_info = inode_to_fs_info(inode); + pgoff_t index = cb->start >> PAGE_SHIFT; + const pgoff_t end_index = (cb->start + cb->len - 1) >> PAGE_SHIFT; + struct folio_batch fbatch; int i; int ret; - while (nr_pages > 0) { - ret = find_get_pages_contig(inode->i_mapping, index, - min_t(unsigned long, - nr_pages, ARRAY_SIZE(pages)), pages); - if (ret == 0) { - nr_pages -= 1; - index += 1; - continue; - } + ret = blk_status_to_errno(cb->bbio.bio.bi_status); + if (ret) + mapping_set_error(inode->i_mapping, ret); + + folio_batch_init(&fbatch); + while (index <= end_index) { + ret = filemap_get_folios(inode->i_mapping, &index, end_index, + &fbatch); + + if (ret == 0) + return; + for (i = 0; i < ret; i++) { - end_page_writeback(pages[i]); - page_cache_release(pages[i]); + struct folio *folio = fbatch.folios[i]; + + btrfs_folio_clamp_clear_writeback(fs_info, folio, + cb->start, cb->len); } - nr_pages -= ret; - index += ret; + folio_batch_release(&fbatch); } /* the inode may be gone now */ } /* - * do the cleanup once all the compressed pages hit the disk. - * This will clear writeback on the file pages and free the compressed - * pages. + * Do the cleanup once all the compressed pages hit the disk. This will clear + * writeback on the file pages and free the compressed pages. * - * This also calls the writeback end hooks for the file pages so that - * metadata and checksums can be updated in the file. + * This also calls the writeback end hooks for the file pages so that metadata + * and checksums can be updated in the file. */ -static void end_compressed_bio_write(struct bio *bio, int err) +static void end_bbio_compressed_write(struct btrfs_bio *bbio) { - struct extent_io_tree *tree; - struct compressed_bio *cb = bio->bi_private; - struct inode *inode; - struct page *page; - unsigned long index; - - if (err) - cb->errors = 1; - - /* if there are more bios still pending for this compressed - * extent, just exit - */ - if (!atomic_dec_and_test(&cb->pending_bios)) - goto out; + struct compressed_bio *cb = to_compressed_bio(bbio); - /* ok, we're the last bio for this extent, step one is to - * call back into the FS and do all the end_io operations - */ - inode = cb->inode; - tree = &BTRFS_I(inode)->io_tree; - cb->compressed_pages[0]->mapping = cb->inode->i_mapping; - tree->ops->writepage_end_io_hook(cb->compressed_pages[0], - cb->start, - cb->start + cb->len - 1, - NULL, 1); - cb->compressed_pages[0]->mapping = NULL; + btrfs_finish_ordered_extent(cb->bbio.ordered, NULL, cb->start, cb->len, + cb->bbio.bio.bi_status == BLK_STS_OK); - end_compressed_writeback(inode, cb->start, cb->len); - /* note, our inode could be gone now */ + if (cb->writeback) + end_compressed_writeback(cb); + /* Note, our inode could be gone now. */ + btrfs_free_compressed_folios(cb); + bio_put(&cb->bbio.bio); +} - /* - * release the compressed pages, these came from alloc_page and - * are not attached to the inode at all - */ - index = 0; - for (index = 0; index < cb->nr_pages; index++) { - page = cb->compressed_pages[index]; - page->mapping = NULL; - page_cache_release(page); +static void btrfs_add_compressed_bio_folios(struct compressed_bio *cb) +{ + struct bio *bio = &cb->bbio.bio; + u32 offset = 0; + unsigned int findex = 0; + + while (offset < cb->compressed_len) { + struct folio *folio = cb->compressed_folios[findex]; + u32 len = min_t(u32, cb->compressed_len - offset, folio_size(folio)); + int ret; + + /* Maximum compressed extent is smaller than bio size limit. */ + ret = bio_add_folio(bio, folio, len, 0); + ASSERT(ret); + offset += len; + findex++; } - - /* finally free the cb struct */ - kfree(cb->compressed_pages); - kfree(cb); -out: - bio_put(bio); } /* @@ -325,226 +366,189 @@ out: * This also checksums the file bytes and gets things ready for * the end io hooks. */ -int btrfs_submit_compressed_write(struct inode *inode, u64 start, - unsigned long len, u64 disk_start, - unsigned long compressed_len, - struct page **compressed_pages, - unsigned long nr_pages) +void btrfs_submit_compressed_write(struct btrfs_ordered_extent *ordered, + struct folio **compressed_folios, + unsigned int nr_folios, + blk_opf_t write_flags, + bool writeback) { - struct bio *bio = NULL; - struct btrfs_root *root = BTRFS_I(inode)->root; + struct btrfs_inode *inode = ordered->inode; + struct btrfs_fs_info *fs_info = inode->root->fs_info; struct compressed_bio *cb; - unsigned long bytes_left; - struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; - int pg_index = 0; - struct page *page; - u64 first_byte = disk_start; - struct block_device *bdev; - int ret; - int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; - - WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1)); - cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); - if (!cb) - return -ENOMEM; - atomic_set(&cb->pending_bios, 0); - cb->errors = 0; - cb->inode = inode; - cb->start = start; - cb->len = len; - cb->mirror_num = 0; - cb->compressed_pages = compressed_pages; - cb->compressed_len = compressed_len; - cb->orig_bio = NULL; - cb->nr_pages = nr_pages; - - bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; - - bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); - if(!bio) { - kfree(cb); - return -ENOMEM; - } - bio->bi_private = cb; - bio->bi_end_io = end_compressed_bio_write; - atomic_inc(&cb->pending_bios); - - /* create and submit bios for the compressed pages */ - bytes_left = compressed_len; - for (pg_index = 0; pg_index < cb->nr_pages; pg_index++) { - page = compressed_pages[pg_index]; - page->mapping = inode->i_mapping; - if (bio->bi_size) - ret = io_tree->ops->merge_bio_hook(WRITE, page, 0, - PAGE_CACHE_SIZE, - bio, 0); - else - ret = 0; - - page->mapping = NULL; - if (ret || bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < - PAGE_CACHE_SIZE) { - bio_get(bio); - - /* - * inc the count before we submit the bio so - * we know the end IO handler won't happen before - * we inc the count. Otherwise, the cb might get - * freed before we're done setting it up - */ - atomic_inc(&cb->pending_bios); - ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); - BUG_ON(ret); /* -ENOMEM */ - - if (!skip_sum) { - ret = btrfs_csum_one_bio(root, inode, bio, - start, 1); - BUG_ON(ret); /* -ENOMEM */ - } - - ret = btrfs_map_bio(root, WRITE, bio, 0, 1); - BUG_ON(ret); /* -ENOMEM */ - - bio_put(bio); - - bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); - BUG_ON(!bio); - bio->bi_private = cb; - bio->bi_end_io = end_compressed_bio_write; - bio_add_page(bio, page, PAGE_CACHE_SIZE, 0); - } - if (bytes_left < PAGE_CACHE_SIZE) { - printk("bytes left %lu compress len %lu nr %lu\n", - bytes_left, cb->compressed_len, cb->nr_pages); - } - bytes_left -= PAGE_CACHE_SIZE; - first_byte += PAGE_CACHE_SIZE; - cond_resched(); - } - bio_get(bio); - - ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); - BUG_ON(ret); /* -ENOMEM */ - if (!skip_sum) { - ret = btrfs_csum_one_bio(root, inode, bio, start, 1); - BUG_ON(ret); /* -ENOMEM */ - } - - ret = btrfs_map_bio(root, WRITE, bio, 0, 1); - BUG_ON(ret); /* -ENOMEM */ - - bio_put(bio); - return 0; + ASSERT(IS_ALIGNED(ordered->file_offset, fs_info->sectorsize)); + ASSERT(IS_ALIGNED(ordered->num_bytes, fs_info->sectorsize)); + + cb = alloc_compressed_bio(inode, ordered->file_offset, + REQ_OP_WRITE | write_flags, + end_bbio_compressed_write); + cb->start = ordered->file_offset; + cb->len = ordered->num_bytes; + cb->compressed_folios = compressed_folios; + cb->compressed_len = ordered->disk_num_bytes; + cb->writeback = writeback; + cb->nr_folios = nr_folios; + cb->bbio.bio.bi_iter.bi_sector = ordered->disk_bytenr >> SECTOR_SHIFT; + cb->bbio.ordered = ordered; + btrfs_add_compressed_bio_folios(cb); + + btrfs_submit_bbio(&cb->bbio, 0); } +/* + * Add extra pages in the same compressed file extent so that we don't need to + * re-read the same extent again and again. + * + * NOTE: this won't work well for subpage, as for subpage read, we lock the + * full page then submit bio for each compressed/regular extents. + * + * This means, if we have several sectors in the same page points to the same + * on-disk compressed data, we will re-read the same extent many times and + * this function can only help for the next page. + */ static noinline int add_ra_bio_pages(struct inode *inode, u64 compressed_end, - struct compressed_bio *cb) + struct compressed_bio *cb, + int *memstall, unsigned long *pflags) { - unsigned long end_index; - unsigned long pg_index; - u64 last_offset; + struct btrfs_fs_info *fs_info = inode_to_fs_info(inode); + pgoff_t end_index; + struct bio *orig_bio = &cb->orig_bbio->bio; + u64 cur = cb->orig_bbio->file_offset + orig_bio->bi_iter.bi_size; u64 isize = i_size_read(inode); int ret; - struct page *page; - unsigned long nr_pages = 0; + struct folio *folio; struct extent_map *em; struct address_space *mapping = inode->i_mapping; struct extent_map_tree *em_tree; struct extent_io_tree *tree; - u64 end; - int misses = 0; + int sectors_missed = 0; - page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page; - last_offset = (page_offset(page) + PAGE_CACHE_SIZE); em_tree = &BTRFS_I(inode)->extent_tree; tree = &BTRFS_I(inode)->io_tree; if (isize == 0) return 0; - end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; + /* + * For current subpage support, we only support 64K page size, + * which means maximum compressed extent size (128K) is just 2x page + * size. + * This makes readahead less effective, so here disable readahead for + * subpage for now, until full compressed write is supported. + */ + if (fs_info->sectorsize < PAGE_SIZE) + return 0; + + /* For bs > ps cases, we don't support readahead for compressed folios for now. */ + if (fs_info->block_min_order) + return 0; + + end_index = (i_size_read(inode) - 1) >> PAGE_SHIFT; - while (last_offset < compressed_end) { - pg_index = last_offset >> PAGE_CACHE_SHIFT; + while (cur < compressed_end) { + pgoff_t page_end; + pgoff_t pg_index = cur >> PAGE_SHIFT; + u32 add_size; if (pg_index > end_index) break; - rcu_read_lock(); - page = radix_tree_lookup(&mapping->page_tree, pg_index); - rcu_read_unlock(); - if (page) { - misses++; - if (misses > 4) + folio = filemap_get_folio(mapping, pg_index); + if (!IS_ERR(folio)) { + u64 folio_sz = folio_size(folio); + u64 offset = offset_in_folio(folio, cur); + + folio_put(folio); + sectors_missed += (folio_sz - offset) >> + fs_info->sectorsize_bits; + + /* Beyond threshold, no need to continue */ + if (sectors_missed > 4) break; - goto next; + + /* + * Jump to next page start as we already have page for + * current offset. + */ + cur += (folio_sz - offset); + continue; } - page = __page_cache_alloc(mapping_gfp_mask(mapping) & - ~__GFP_FS); - if (!page) + folio = filemap_alloc_folio(mapping_gfp_constraint(mapping, ~__GFP_FS), + 0, NULL); + if (!folio) break; - if (add_to_page_cache_lru(page, mapping, pg_index, - GFP_NOFS)) { - page_cache_release(page); - goto next; + if (filemap_add_folio(mapping, folio, pg_index, GFP_NOFS)) { + /* There is already a page, skip to page end */ + cur += folio_size(folio); + folio_put(folio); + continue; + } + + if (!*memstall && folio_test_workingset(folio)) { + psi_memstall_enter(pflags); + *memstall = 1; } - end = last_offset + PAGE_CACHE_SIZE - 1; - /* - * at this point, we have a locked page in the page cache - * for these bytes in the file. But, we have to make - * sure they map to this compressed extent on disk. - */ - set_page_extent_mapped(page); - lock_extent(tree, last_offset, end); + ret = set_folio_extent_mapped(folio); + if (ret < 0) { + folio_unlock(folio); + folio_put(folio); + break; + } + + page_end = (pg_index << PAGE_SHIFT) + folio_size(folio) - 1; + btrfs_lock_extent(tree, cur, page_end, NULL); read_lock(&em_tree->lock); - em = lookup_extent_mapping(em_tree, last_offset, - PAGE_CACHE_SIZE); + em = btrfs_lookup_extent_mapping(em_tree, cur, page_end + 1 - cur); read_unlock(&em_tree->lock); - if (!em || last_offset < em->start || - (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) || - (em->block_start >> 9) != cb->orig_bio->bi_sector) { - free_extent_map(em); - unlock_extent(tree, last_offset, end); - unlock_page(page); - page_cache_release(page); + /* + * At this point, we have a locked page in the page cache for + * these bytes in the file. But, we have to make sure they map + * to this compressed extent on disk. + */ + if (!em || cur < em->start || + (cur + fs_info->sectorsize > btrfs_extent_map_end(em)) || + (btrfs_extent_map_block_start(em) >> SECTOR_SHIFT) != + orig_bio->bi_iter.bi_sector) { + btrfs_free_extent_map(em); + btrfs_unlock_extent(tree, cur, page_end, NULL); + folio_unlock(folio); + folio_put(folio); break; } - free_extent_map(em); + add_size = min(em->start + em->len, page_end + 1) - cur; + btrfs_free_extent_map(em); + btrfs_unlock_extent(tree, cur, page_end, NULL); - if (page->index == end_index) { - char *userpage; - size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1); + if (folio_contains(folio, end_index)) { + size_t zero_offset = offset_in_folio(folio, isize); if (zero_offset) { int zeros; - zeros = PAGE_CACHE_SIZE - zero_offset; - userpage = kmap_atomic(page); - memset(userpage + zero_offset, 0, zeros); - flush_dcache_page(page); - kunmap_atomic(userpage); + zeros = folio_size(folio) - zero_offset; + folio_zero_range(folio, zero_offset, zeros); } } - ret = bio_add_page(cb->orig_bio, page, - PAGE_CACHE_SIZE, 0); - - if (ret == PAGE_CACHE_SIZE) { - nr_pages++; - page_cache_release(page); - } else { - unlock_extent(tree, last_offset, end); - unlock_page(page); - page_cache_release(page); + if (!bio_add_folio(orig_bio, folio, add_size, + offset_in_folio(folio, cur))) { + folio_unlock(folio); + folio_put(folio); break; } -next: - last_offset += PAGE_CACHE_SIZE; + /* + * If it's subpage, we also need to increase its + * subpage::readers number, as at endio we will decrease + * subpage::readers and to unlock the page. + */ + if (fs_info->sectorsize < PAGE_SIZE) + btrfs_folio_set_lock(fs_info, folio, cur, add_size); + folio_put(folio); + cur += add_size; } return 0; } @@ -554,499 +558,1125 @@ next: * in it. We don't actually do IO on those pages but allocate new ones * to hold the compressed pages on disk. * - * bio->bi_sector points to the compressed extent on disk + * bio->bi_iter.bi_sector points to the compressed extent on disk * bio->bi_io_vec points to all of the inode pages - * bio->bi_vcnt is a count of pages * * After the compressed pages are read, we copy the bytes into the * bio we were passed and then call the bio end_io calls */ -int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio, - int mirror_num, unsigned long bio_flags) +void btrfs_submit_compressed_read(struct btrfs_bio *bbio) { - struct extent_io_tree *tree; - struct extent_map_tree *em_tree; + struct btrfs_inode *inode = bbio->inode; + struct btrfs_fs_info *fs_info = inode->root->fs_info; + struct extent_map_tree *em_tree = &inode->extent_tree; struct compressed_bio *cb; - struct btrfs_root *root = BTRFS_I(inode)->root; - unsigned long uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE; - unsigned long compressed_len; - unsigned long nr_pages; - unsigned long pg_index; - struct page *page; - struct block_device *bdev; - struct bio *comp_bio; - u64 cur_disk_byte = (u64)bio->bi_sector << 9; + unsigned int compressed_len; + u64 file_offset = bbio->file_offset; u64 em_len; u64 em_start; struct extent_map *em; - int ret = -ENOMEM; - int faili = 0; - u32 *sums; - - tree = &BTRFS_I(inode)->io_tree; - em_tree = &BTRFS_I(inode)->extent_tree; + unsigned long pflags; + int memstall = 0; + blk_status_t status; + int ret; /* we need the actual starting offset of this extent in the file */ read_lock(&em_tree->lock); - em = lookup_extent_mapping(em_tree, - page_offset(bio->bi_io_vec->bv_page), - PAGE_CACHE_SIZE); + em = btrfs_lookup_extent_mapping(em_tree, file_offset, fs_info->sectorsize); read_unlock(&em_tree->lock); - if (!em) - return -EIO; - - compressed_len = em->block_len; - cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); - if (!cb) + if (!em) { + status = BLK_STS_IOERR; goto out; + } + + ASSERT(btrfs_extent_map_is_compressed(em)); + compressed_len = em->disk_num_bytes; - atomic_set(&cb->pending_bios, 0); - cb->errors = 0; - cb->inode = inode; - cb->mirror_num = mirror_num; - sums = &cb->sums; + cb = alloc_compressed_bio(inode, file_offset, REQ_OP_READ, + end_bbio_compressed_read); - cb->start = em->orig_start; + cb->start = em->start - em->offset; em_len = em->len; em_start = em->start; - free_extent_map(em); - em = NULL; - - cb->len = uncompressed_len; + cb->len = bbio->bio.bi_iter.bi_size; cb->compressed_len = compressed_len; - cb->compress_type = extent_compress_type(bio_flags); - cb->orig_bio = bio; - - nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) / - PAGE_CACHE_SIZE; - cb->compressed_pages = kzalloc(sizeof(struct page *) * nr_pages, - GFP_NOFS); - if (!cb->compressed_pages) - goto fail1; - - bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; - - for (pg_index = 0; pg_index < nr_pages; pg_index++) { - cb->compressed_pages[pg_index] = alloc_page(GFP_NOFS | - __GFP_HIGHMEM); - if (!cb->compressed_pages[pg_index]) { - faili = pg_index - 1; - ret = -ENOMEM; - goto fail2; - } + cb->compress_type = btrfs_extent_map_compression(em); + cb->orig_bbio = bbio; + cb->bbio.csum_search_commit_root = bbio->csum_search_commit_root; + + btrfs_free_extent_map(em); + + cb->nr_folios = DIV_ROUND_UP(compressed_len, btrfs_min_folio_size(fs_info)); + cb->compressed_folios = kcalloc(cb->nr_folios, sizeof(struct folio *), GFP_NOFS); + if (!cb->compressed_folios) { + status = BLK_STS_RESOURCE; + goto out_free_bio; } - faili = nr_pages - 1; - cb->nr_pages = nr_pages; - add_ra_bio_pages(inode, em_start + em_len, cb); + ret = btrfs_alloc_folio_array(cb->nr_folios, fs_info->block_min_order, + cb->compressed_folios); + if (ret) { + status = BLK_STS_RESOURCE; + goto out_free_compressed_pages; + } + + add_ra_bio_pages(&inode->vfs_inode, em_start + em_len, cb, &memstall, + &pflags); /* include any pages we added in add_ra-bio_pages */ - uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE; - cb->len = uncompressed_len; - - comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS); - if (!comp_bio) - goto fail2; - comp_bio->bi_private = cb; - comp_bio->bi_end_io = end_compressed_bio_read; - atomic_inc(&cb->pending_bios); - - for (pg_index = 0; pg_index < nr_pages; pg_index++) { - page = cb->compressed_pages[pg_index]; - page->mapping = inode->i_mapping; - page->index = em_start >> PAGE_CACHE_SHIFT; - - if (comp_bio->bi_size) - ret = tree->ops->merge_bio_hook(READ, page, 0, - PAGE_CACHE_SIZE, - comp_bio, 0); - else - ret = 0; - - page->mapping = NULL; - if (ret || bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) < - PAGE_CACHE_SIZE) { - bio_get(comp_bio); - - ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0); - BUG_ON(ret); /* -ENOMEM */ + cb->len = bbio->bio.bi_iter.bi_size; + cb->bbio.bio.bi_iter.bi_sector = bbio->bio.bi_iter.bi_sector; + btrfs_add_compressed_bio_folios(cb); - /* - * inc the count before we submit the bio so - * we know the end IO handler won't happen before - * we inc the count. Otherwise, the cb might get - * freed before we're done setting it up - */ - atomic_inc(&cb->pending_bios); + if (memstall) + psi_memstall_leave(&pflags); - if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { - ret = btrfs_lookup_bio_sums(root, inode, - comp_bio, sums); - BUG_ON(ret); /* -ENOMEM */ - } - sums += (comp_bio->bi_size + root->sectorsize - 1) / - root->sectorsize; + btrfs_submit_bbio(&cb->bbio, 0); + return; + +out_free_compressed_pages: + kfree(cb->compressed_folios); +out_free_bio: + bio_put(&cb->bbio.bio); +out: + btrfs_bio_end_io(bbio, status); +} + +/* + * Heuristic uses systematic sampling to collect data from the input data + * range, the logic can be tuned by the following constants: + * + * @SAMPLING_READ_SIZE - how many bytes will be copied from for each sample + * @SAMPLING_INTERVAL - range from which the sampled data can be collected + */ +#define SAMPLING_READ_SIZE (16) +#define SAMPLING_INTERVAL (256) - ret = btrfs_map_bio(root, READ, comp_bio, - mirror_num, 0); - if (ret) - bio_endio(comp_bio, ret); +/* + * For statistical analysis of the input data we consider bytes that form a + * Galois Field of 256 objects. Each object has an attribute count, ie. how + * many times the object appeared in the sample. + */ +#define BUCKET_SIZE (256) - bio_put(comp_bio); +/* + * The size of the sample is based on a statistical sampling rule of thumb. + * The common way is to perform sampling tests as long as the number of + * elements in each cell is at least 5. + * + * Instead of 5, we choose 32 to obtain more accurate results. + * If the data contain the maximum number of symbols, which is 256, we obtain a + * sample size bound by 8192. + * + * For a sample of at most 8KB of data per data range: 16 consecutive bytes + * from up to 512 locations. + */ +#define MAX_SAMPLE_SIZE (BTRFS_MAX_UNCOMPRESSED * \ + SAMPLING_READ_SIZE / SAMPLING_INTERVAL) - comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, - GFP_NOFS); - BUG_ON(!comp_bio); - comp_bio->bi_private = cb; - comp_bio->bi_end_io = end_compressed_bio_read; +struct bucket_item { + u32 count; +}; - bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0); - } - cur_disk_byte += PAGE_CACHE_SIZE; - } - bio_get(comp_bio); +struct heuristic_ws { + /* Partial copy of input data */ + u8 *sample; + u32 sample_size; + /* Buckets store counters for each byte value */ + struct bucket_item *bucket; + /* Sorting buffer */ + struct bucket_item *bucket_b; + struct list_head list; +}; - ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0); - BUG_ON(ret); /* -ENOMEM */ +static void free_heuristic_ws(struct list_head *ws) +{ + struct heuristic_ws *workspace; - if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { - ret = btrfs_lookup_bio_sums(root, inode, comp_bio, sums); - BUG_ON(ret); /* -ENOMEM */ - } + workspace = list_entry(ws, struct heuristic_ws, list); - ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0); - if (ret) - bio_endio(comp_bio, ret); + kvfree(workspace->sample); + kfree(workspace->bucket); + kfree(workspace->bucket_b); + kfree(workspace); +} - bio_put(comp_bio); - return 0; +static struct list_head *alloc_heuristic_ws(struct btrfs_fs_info *fs_info) +{ + struct heuristic_ws *ws; -fail2: - while (faili >= 0) { - __free_page(cb->compressed_pages[faili]); - faili--; - } + ws = kzalloc(sizeof(*ws), GFP_KERNEL); + if (!ws) + return ERR_PTR(-ENOMEM); - kfree(cb->compressed_pages); -fail1: - kfree(cb); -out: - free_extent_map(em); - return ret; + ws->sample = kvmalloc(MAX_SAMPLE_SIZE, GFP_KERNEL); + if (!ws->sample) + goto fail; + + ws->bucket = kcalloc(BUCKET_SIZE, sizeof(*ws->bucket), GFP_KERNEL); + if (!ws->bucket) + goto fail; + + ws->bucket_b = kcalloc(BUCKET_SIZE, sizeof(*ws->bucket_b), GFP_KERNEL); + if (!ws->bucket_b) + goto fail; + + INIT_LIST_HEAD(&ws->list); + return &ws->list; +fail: + free_heuristic_ws(&ws->list); + return ERR_PTR(-ENOMEM); } -static struct list_head comp_idle_workspace[BTRFS_COMPRESS_TYPES]; -static spinlock_t comp_workspace_lock[BTRFS_COMPRESS_TYPES]; -static int comp_num_workspace[BTRFS_COMPRESS_TYPES]; -static atomic_t comp_alloc_workspace[BTRFS_COMPRESS_TYPES]; -static wait_queue_head_t comp_workspace_wait[BTRFS_COMPRESS_TYPES]; +const struct btrfs_compress_levels btrfs_heuristic_compress = { 0 }; -static struct btrfs_compress_op *btrfs_compress_op[] = { +static const struct btrfs_compress_levels * const btrfs_compress_levels[] = { + /* The heuristic is represented as compression type 0 */ + &btrfs_heuristic_compress, &btrfs_zlib_compress, &btrfs_lzo_compress, + &btrfs_zstd_compress, }; -void __init btrfs_init_compress(void) +static struct list_head *alloc_workspace(struct btrfs_fs_info *fs_info, int type, int level) { - int i; + switch (type) { + case BTRFS_COMPRESS_NONE: return alloc_heuristic_ws(fs_info); + case BTRFS_COMPRESS_ZLIB: return zlib_alloc_workspace(fs_info, level); + case BTRFS_COMPRESS_LZO: return lzo_alloc_workspace(fs_info); + case BTRFS_COMPRESS_ZSTD: return zstd_alloc_workspace(fs_info, level); + default: + /* + * This can't happen, the type is validated several times + * before we get here. + */ + BUG(); + } +} - for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { - INIT_LIST_HEAD(&comp_idle_workspace[i]); - spin_lock_init(&comp_workspace_lock[i]); - atomic_set(&comp_alloc_workspace[i], 0); - init_waitqueue_head(&comp_workspace_wait[i]); +static void free_workspace(int type, struct list_head *ws) +{ + switch (type) { + case BTRFS_COMPRESS_NONE: return free_heuristic_ws(ws); + case BTRFS_COMPRESS_ZLIB: return zlib_free_workspace(ws); + case BTRFS_COMPRESS_LZO: return lzo_free_workspace(ws); + case BTRFS_COMPRESS_ZSTD: return zstd_free_workspace(ws); + default: + /* + * This can't happen, the type is validated several times + * before we get here. + */ + BUG(); } } +static int alloc_workspace_manager(struct btrfs_fs_info *fs_info, + enum btrfs_compression_type type) +{ + struct workspace_manager *gwsm; + struct list_head *workspace; + + ASSERT(fs_info->compr_wsm[type] == NULL); + gwsm = kzalloc(sizeof(*gwsm), GFP_KERNEL); + if (!gwsm) + return -ENOMEM; + + INIT_LIST_HEAD(&gwsm->idle_ws); + spin_lock_init(&gwsm->ws_lock); + atomic_set(&gwsm->total_ws, 0); + init_waitqueue_head(&gwsm->ws_wait); + fs_info->compr_wsm[type] = gwsm; + + /* + * Preallocate one workspace for each compression type so we can + * guarantee forward progress in the worst case + */ + workspace = alloc_workspace(fs_info, type, 0); + if (IS_ERR(workspace)) { + btrfs_warn(fs_info, + "cannot preallocate compression workspace for %s, will try later", + btrfs_compress_type2str(type)); + } else { + atomic_set(&gwsm->total_ws, 1); + gwsm->free_ws = 1; + list_add(workspace, &gwsm->idle_ws); + } + return 0; +} + +static void free_workspace_manager(struct btrfs_fs_info *fs_info, + enum btrfs_compression_type type) +{ + struct list_head *ws; + struct workspace_manager *gwsm = fs_info->compr_wsm[type]; + + /* ZSTD uses its own workspace manager, should enter here. */ + ASSERT(type != BTRFS_COMPRESS_ZSTD && type < BTRFS_NR_COMPRESS_TYPES); + if (!gwsm) + return; + fs_info->compr_wsm[type] = NULL; + while (!list_empty(&gwsm->idle_ws)) { + ws = gwsm->idle_ws.next; + list_del(ws); + free_workspace(type, ws); + atomic_dec(&gwsm->total_ws); + } + kfree(gwsm); +} + /* - * this finds an available workspace or allocates a new one - * ERR_PTR is returned if things go bad. + * This finds an available workspace or allocates a new one. + * If it's not possible to allocate a new one, waits until there's one. + * Preallocation makes a forward progress guarantees and we do not return + * errors. */ -static struct list_head *find_workspace(int type) +struct list_head *btrfs_get_workspace(struct btrfs_fs_info *fs_info, int type, int level) { + struct workspace_manager *wsm = fs_info->compr_wsm[type]; struct list_head *workspace; int cpus = num_online_cpus(); - int idx = type - 1; + unsigned nofs_flag; + struct list_head *idle_ws; + spinlock_t *ws_lock; + atomic_t *total_ws; + wait_queue_head_t *ws_wait; + int *free_ws; + + ASSERT(wsm); + idle_ws = &wsm->idle_ws; + ws_lock = &wsm->ws_lock; + total_ws = &wsm->total_ws; + ws_wait = &wsm->ws_wait; + free_ws = &wsm->free_ws; - struct list_head *idle_workspace = &comp_idle_workspace[idx]; - spinlock_t *workspace_lock = &comp_workspace_lock[idx]; - atomic_t *alloc_workspace = &comp_alloc_workspace[idx]; - wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx]; - int *num_workspace = &comp_num_workspace[idx]; again: - spin_lock(workspace_lock); - if (!list_empty(idle_workspace)) { - workspace = idle_workspace->next; + spin_lock(ws_lock); + if (!list_empty(idle_ws)) { + workspace = idle_ws->next; list_del(workspace); - (*num_workspace)--; - spin_unlock(workspace_lock); + (*free_ws)--; + spin_unlock(ws_lock); return workspace; } - if (atomic_read(alloc_workspace) > cpus) { + if (atomic_read(total_ws) > cpus) { DEFINE_WAIT(wait); - spin_unlock(workspace_lock); - prepare_to_wait(workspace_wait, &wait, TASK_UNINTERRUPTIBLE); - if (atomic_read(alloc_workspace) > cpus && !*num_workspace) + spin_unlock(ws_lock); + prepare_to_wait(ws_wait, &wait, TASK_UNINTERRUPTIBLE); + if (atomic_read(total_ws) > cpus && !*free_ws) schedule(); - finish_wait(workspace_wait, &wait); + finish_wait(ws_wait, &wait); goto again; } - atomic_inc(alloc_workspace); - spin_unlock(workspace_lock); + atomic_inc(total_ws); + spin_unlock(ws_lock); + + /* + * Allocation helpers call vmalloc that can't use GFP_NOFS, so we have + * to turn it off here because we might get called from the restricted + * context of btrfs_compress_bio/btrfs_compress_pages + */ + nofs_flag = memalloc_nofs_save(); + workspace = alloc_workspace(fs_info, type, level); + memalloc_nofs_restore(nofs_flag); - workspace = btrfs_compress_op[idx]->alloc_workspace(); if (IS_ERR(workspace)) { - atomic_dec(alloc_workspace); - wake_up(workspace_wait); + atomic_dec(total_ws); + wake_up(ws_wait); + + /* + * Do not return the error but go back to waiting. There's a + * workspace preallocated for each type and the compression + * time is bounded so we get to a workspace eventually. This + * makes our caller's life easier. + * + * To prevent silent and low-probability deadlocks (when the + * initial preallocation fails), check if there are any + * workspaces at all. + */ + if (atomic_read(total_ws) == 0) { + static DEFINE_RATELIMIT_STATE(_rs, + /* once per minute */ 60 * HZ, + /* no burst */ 1); + + if (__ratelimit(&_rs)) + btrfs_warn(fs_info, + "no compression workspaces, low memory, retrying"); + } + goto again; } return workspace; } +static struct list_head *get_workspace(struct btrfs_fs_info *fs_info, int type, int level) +{ + switch (type) { + case BTRFS_COMPRESS_NONE: return btrfs_get_workspace(fs_info, type, level); + case BTRFS_COMPRESS_ZLIB: return zlib_get_workspace(fs_info, level); + case BTRFS_COMPRESS_LZO: return btrfs_get_workspace(fs_info, type, level); + case BTRFS_COMPRESS_ZSTD: return zstd_get_workspace(fs_info, level); + default: + /* + * This can't happen, the type is validated several times + * before we get here. + */ + BUG(); + } +} + /* * put a workspace struct back on the list or free it if we have enough * idle ones sitting around */ -static void free_workspace(int type, struct list_head *workspace) +void btrfs_put_workspace(struct btrfs_fs_info *fs_info, int type, struct list_head *ws) { - int idx = type - 1; - struct list_head *idle_workspace = &comp_idle_workspace[idx]; - spinlock_t *workspace_lock = &comp_workspace_lock[idx]; - atomic_t *alloc_workspace = &comp_alloc_workspace[idx]; - wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx]; - int *num_workspace = &comp_num_workspace[idx]; - - spin_lock(workspace_lock); - if (*num_workspace < num_online_cpus()) { - list_add_tail(workspace, idle_workspace); - (*num_workspace)++; - spin_unlock(workspace_lock); + struct workspace_manager *gwsm = fs_info->compr_wsm[type]; + struct list_head *idle_ws; + spinlock_t *ws_lock; + atomic_t *total_ws; + wait_queue_head_t *ws_wait; + int *free_ws; + + ASSERT(gwsm); + idle_ws = &gwsm->idle_ws; + ws_lock = &gwsm->ws_lock; + total_ws = &gwsm->total_ws; + ws_wait = &gwsm->ws_wait; + free_ws = &gwsm->free_ws; + + spin_lock(ws_lock); + if (*free_ws <= num_online_cpus()) { + list_add(ws, idle_ws); + (*free_ws)++; + spin_unlock(ws_lock); goto wake; } - spin_unlock(workspace_lock); + spin_unlock(ws_lock); - btrfs_compress_op[idx]->free_workspace(workspace); - atomic_dec(alloc_workspace); + free_workspace(type, ws); + atomic_dec(total_ws); wake: - smp_mb(); - if (waitqueue_active(workspace_wait)) - wake_up(workspace_wait); + cond_wake_up(ws_wait); +} + +static void put_workspace(struct btrfs_fs_info *fs_info, int type, struct list_head *ws) +{ + switch (type) { + case BTRFS_COMPRESS_NONE: return btrfs_put_workspace(fs_info, type, ws); + case BTRFS_COMPRESS_ZLIB: return btrfs_put_workspace(fs_info, type, ws); + case BTRFS_COMPRESS_LZO: return btrfs_put_workspace(fs_info, type, ws); + case BTRFS_COMPRESS_ZSTD: return zstd_put_workspace(fs_info, ws); + default: + /* + * This can't happen, the type is validated several times + * before we get here. + */ + BUG(); + } } /* - * cleanup function for module exit + * Adjust @level according to the limits of the compression algorithm or + * fallback to default */ -static void free_workspaces(void) +static int btrfs_compress_set_level(unsigned int type, int level) { - struct list_head *workspace; - int i; + const struct btrfs_compress_levels *levels = btrfs_compress_levels[type]; - for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { - while (!list_empty(&comp_idle_workspace[i])) { - workspace = comp_idle_workspace[i].next; - list_del(workspace); - btrfs_compress_op[i]->free_workspace(workspace); - atomic_dec(&comp_alloc_workspace[i]); - } + if (level == 0) + level = levels->default_level; + else + level = clamp(level, levels->min_level, levels->max_level); + + return level; +} + +/* + * Check whether the @level is within the valid range for the given type. + */ +bool btrfs_compress_level_valid(unsigned int type, int level) +{ + const struct btrfs_compress_levels *levels = btrfs_compress_levels[type]; + + return levels->min_level <= level && level <= levels->max_level; +} + +/* Wrapper around find_get_page(), with extra error message. */ +int btrfs_compress_filemap_get_folio(struct address_space *mapping, u64 start, + struct folio **in_folio_ret) +{ + struct folio *in_folio; + + /* + * The compressed write path should have the folio locked already, thus + * we only need to grab one reference. + */ + in_folio = filemap_get_folio(mapping, start >> PAGE_SHIFT); + if (IS_ERR(in_folio)) { + struct btrfs_inode *inode = BTRFS_I(mapping->host); + + btrfs_crit(inode->root->fs_info, + "failed to get page cache, root %lld ino %llu file offset %llu", + btrfs_root_id(inode->root), btrfs_ino(inode), start); + return -ENOENT; } + *in_folio_ret = in_folio; + return 0; } /* - * given an address space and start/len, compress the bytes. + * Given an address space and start and length, compress the bytes into @pages + * that are allocated on demand. * - * pages are allocated to hold the compressed result and stored - * in 'pages' + * @type_level is encoded algorithm and level, where level 0 means whatever + * default the algorithm chooses and is opaque here; + * - compression algo are 0-3 + * - the level are bits 4-7 * - * out_pages is used to return the number of pages allocated. There - * may be pages allocated even if we return an error + * @out_folios is an in/out parameter, holds maximum number of folios to allocate + * and returns number of actually allocated folios * - * total_in is used to return the number of bytes actually read. It - * may be smaller then len if we had to exit early because we - * ran out of room in the pages array or because we cross the + * @total_in is used to return the number of bytes actually read. It + * may be smaller than the input length if we had to exit early because we + * ran out of room in the folios array or because we cross the * max_out threshold. * - * total_out is used to return the total number of compressed bytes - * - * max_out tells us the max number of bytes that we're allowed to - * stuff into pages + * @total_out is an in/out parameter, must be set to the input length and will + * be also used to return the total number of compressed bytes */ -int btrfs_compress_pages(int type, struct address_space *mapping, - u64 start, unsigned long len, - struct page **pages, - unsigned long nr_dest_pages, - unsigned long *out_pages, - unsigned long *total_in, - unsigned long *total_out, - unsigned long max_out) +int btrfs_compress_folios(unsigned int type, int level, struct btrfs_inode *inode, + u64 start, struct folio **folios, unsigned long *out_folios, + unsigned long *total_in, unsigned long *total_out) { + struct btrfs_fs_info *fs_info = inode->root->fs_info; + const unsigned long orig_len = *total_out; struct list_head *workspace; int ret; - workspace = find_workspace(type); - if (IS_ERR(workspace)) - return -1; - - ret = btrfs_compress_op[type-1]->compress_pages(workspace, mapping, - start, len, pages, - nr_dest_pages, out_pages, - total_in, total_out, - max_out); - free_workspace(type, workspace); + level = btrfs_compress_set_level(type, level); + workspace = get_workspace(fs_info, type, level); + ret = compression_compress_pages(type, workspace, inode, start, folios, + out_folios, total_in, total_out); + /* The total read-in bytes should be no larger than the input. */ + ASSERT(*total_in <= orig_len); + put_workspace(fs_info, type, workspace); return ret; } -/* - * pages_in is an array of pages with compressed data. - * - * disk_start is the starting logical offset of this array in the file - * - * bvec is a bio_vec of pages from the file that we want to decompress into - * - * vcnt is the count of pages in the biovec - * - * srclen is the number of bytes in pages_in - * - * The basic idea is that we have a bio that was created by readpages. - * The pages in the bio are for the uncompressed data, and they may not - * be contiguous. They all correspond to the range of bytes covered by - * the compressed extent. - */ -static int btrfs_decompress_biovec(int type, struct page **pages_in, - u64 disk_start, struct bio_vec *bvec, - int vcnt, size_t srclen) +static int btrfs_decompress_bio(struct compressed_bio *cb) { + struct btrfs_fs_info *fs_info = cb_to_fs_info(cb); struct list_head *workspace; int ret; + int type = cb->compress_type; - workspace = find_workspace(type); - if (IS_ERR(workspace)) - return -ENOMEM; + workspace = get_workspace(fs_info, type, 0); + ret = compression_decompress_bio(workspace, cb); + put_workspace(fs_info, type, workspace); - ret = btrfs_compress_op[type-1]->decompress_biovec(workspace, pages_in, - disk_start, - bvec, vcnt, srclen); - free_workspace(type, workspace); + if (!ret) + zero_fill_bio(&cb->orig_bbio->bio); return ret; } /* * a less complex decompression routine. Our compressed data fits in a * single page, and we want to read a single page out of it. - * start_byte tells us the offset into the compressed data we're interested in + * dest_pgoff tells us the offset into the destination folio where we write the + * decompressed data. */ -int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page, - unsigned long start_byte, size_t srclen, size_t destlen) +int btrfs_decompress(int type, const u8 *data_in, struct folio *dest_folio, + unsigned long dest_pgoff, size_t srclen, size_t destlen) { + struct btrfs_fs_info *fs_info = folio_to_fs_info(dest_folio); struct list_head *workspace; + const u32 sectorsize = fs_info->sectorsize; int ret; - workspace = find_workspace(type); - if (IS_ERR(workspace)) - return -ENOMEM; + /* + * The full destination folio range should not exceed the folio size. + * And the @destlen should not exceed sectorsize, as this is only called for + * inline file extents, which should not exceed sectorsize. + */ + ASSERT(dest_pgoff + destlen <= folio_size(dest_folio) && destlen <= sectorsize); - ret = btrfs_compress_op[type-1]->decompress(workspace, data_in, - dest_page, start_byte, - srclen, destlen); + workspace = get_workspace(fs_info, type, 0); + ret = compression_decompress(type, workspace, data_in, dest_folio, + dest_pgoff, srclen, destlen); + put_workspace(fs_info, type, workspace); - free_workspace(type, workspace); return ret; } -void btrfs_exit_compress(void) +int btrfs_alloc_compress_wsm(struct btrfs_fs_info *fs_info) +{ + int ret; + + ret = alloc_workspace_manager(fs_info, BTRFS_COMPRESS_NONE); + if (ret < 0) + goto error; + ret = alloc_workspace_manager(fs_info, BTRFS_COMPRESS_ZLIB); + if (ret < 0) + goto error; + ret = alloc_workspace_manager(fs_info, BTRFS_COMPRESS_LZO); + if (ret < 0) + goto error; + ret = zstd_alloc_workspace_manager(fs_info); + if (ret < 0) + goto error; + return 0; +error: + btrfs_free_compress_wsm(fs_info); + return ret; +} + +void btrfs_free_compress_wsm(struct btrfs_fs_info *fs_info) +{ + free_workspace_manager(fs_info, BTRFS_COMPRESS_NONE); + free_workspace_manager(fs_info, BTRFS_COMPRESS_ZLIB); + free_workspace_manager(fs_info, BTRFS_COMPRESS_LZO); + zstd_free_workspace_manager(fs_info); +} + +int __init btrfs_init_compress(void) +{ + if (bioset_init(&btrfs_compressed_bioset, BIO_POOL_SIZE, + offsetof(struct compressed_bio, bbio.bio), + BIOSET_NEED_BVECS)) + return -ENOMEM; + + compr_pool.shrinker = shrinker_alloc(SHRINKER_NONSLAB, "btrfs-compr-pages"); + if (!compr_pool.shrinker) + return -ENOMEM; + + spin_lock_init(&compr_pool.lock); + INIT_LIST_HEAD(&compr_pool.list); + compr_pool.count = 0; + /* 128K / 4K = 32, for 8 threads is 256 pages. */ + compr_pool.thresh = BTRFS_MAX_COMPRESSED / PAGE_SIZE * 8; + compr_pool.shrinker->count_objects = btrfs_compr_pool_count; + compr_pool.shrinker->scan_objects = btrfs_compr_pool_scan; + compr_pool.shrinker->batch = 32; + compr_pool.shrinker->seeks = DEFAULT_SEEKS; + shrinker_register(compr_pool.shrinker); + + return 0; +} + +void __cold btrfs_exit_compress(void) +{ + /* For now scan drains all pages and does not touch the parameters. */ + btrfs_compr_pool_scan(NULL, NULL); + shrinker_free(compr_pool.shrinker); + + bioset_exit(&btrfs_compressed_bioset); +} + +/* + * The bvec is a single page bvec from a bio that contains folios from a filemap. + * + * Since the folio may be a large one, and if the bv_page is not a head page of + * a large folio, then page->index is unreliable. + * + * Thus we need this helper to grab the proper file offset. + */ +static u64 file_offset_from_bvec(const struct bio_vec *bvec) +{ + const struct page *page = bvec->bv_page; + const struct folio *folio = page_folio(page); + + return (page_pgoff(folio, page) << PAGE_SHIFT) + bvec->bv_offset; +} + +/* + * Copy decompressed data from working buffer to pages. + * + * @buf: The decompressed data buffer + * @buf_len: The decompressed data length + * @decompressed: Number of bytes that are already decompressed inside the + * compressed extent + * @cb: The compressed extent descriptor + * @orig_bio: The original bio that the caller wants to read for + * + * An easier to understand graph is like below: + * + * |<- orig_bio ->| |<- orig_bio->| + * |<------- full decompressed extent ----->| + * |<----------- @cb range ---->| + * | |<-- @buf_len -->| + * |<--- @decompressed --->| + * + * Note that, @cb can be a subpage of the full decompressed extent, but + * @cb->start always has the same as the orig_file_offset value of the full + * decompressed extent. + * + * When reading compressed extent, we have to read the full compressed extent, + * while @orig_bio may only want part of the range. + * Thus this function will ensure only data covered by @orig_bio will be copied + * to. + * + * Return 0 if we have copied all needed contents for @orig_bio. + * Return >0 if we need continue decompress. + */ +int btrfs_decompress_buf2page(const char *buf, u32 buf_len, + struct compressed_bio *cb, u32 decompressed) +{ + struct bio *orig_bio = &cb->orig_bbio->bio; + /* Offset inside the full decompressed extent */ + u32 cur_offset; + + cur_offset = decompressed; + /* The main loop to do the copy */ + while (cur_offset < decompressed + buf_len) { + struct bio_vec bvec; + size_t copy_len; + u32 copy_start; + /* Offset inside the full decompressed extent */ + u32 bvec_offset; + void *kaddr; + + bvec = bio_iter_iovec(orig_bio, orig_bio->bi_iter); + /* + * cb->start may underflow, but subtracting that value can still + * give us correct offset inside the full decompressed extent. + */ + bvec_offset = file_offset_from_bvec(&bvec) - cb->start; + + /* Haven't reached the bvec range, exit */ + if (decompressed + buf_len <= bvec_offset) + return 1; + + copy_start = max(cur_offset, bvec_offset); + copy_len = min(bvec_offset + bvec.bv_len, + decompressed + buf_len) - copy_start; + ASSERT(copy_len); + + /* + * Extra range check to ensure we didn't go beyond + * @buf + @buf_len. + */ + ASSERT(copy_start - decompressed < buf_len); + + kaddr = bvec_kmap_local(&bvec); + memcpy(kaddr, buf + copy_start - decompressed, copy_len); + kunmap_local(kaddr); + + cur_offset += copy_len; + bio_advance(orig_bio, copy_len); + /* Finished the bio */ + if (!orig_bio->bi_iter.bi_size) + return 0; + } + return 1; +} + +/* + * Shannon Entropy calculation + * + * Pure byte distribution analysis fails to determine compressibility of data. + * Try calculating entropy to estimate the average minimum number of bits + * needed to encode the sampled data. + * + * For convenience, return the percentage of needed bits, instead of amount of + * bits directly. + * + * @ENTROPY_LVL_ACEPTABLE - below that threshold, sample has low byte entropy + * and can be compressible with high probability + * + * @ENTROPY_LVL_HIGH - data are not compressible with high probability + * + * Use of ilog2() decreases precision, we lower the LVL to 5 to compensate. + */ +#define ENTROPY_LVL_ACEPTABLE (65) +#define ENTROPY_LVL_HIGH (80) + +/* + * For increased precision in shannon_entropy calculation, + * let's do pow(n, M) to save more digits after comma: + * + * - maximum int bit length is 64 + * - ilog2(MAX_SAMPLE_SIZE) -> 13 + * - 13 * 4 = 52 < 64 -> M = 4 + * + * So use pow(n, 4). + */ +static inline u32 ilog2_w(u64 n) +{ + return ilog2(n * n * n * n); +} + +static u32 shannon_entropy(struct heuristic_ws *ws) +{ + const u32 entropy_max = 8 * ilog2_w(2); + u32 entropy_sum = 0; + u32 p, p_base, sz_base; + u32 i; + + sz_base = ilog2_w(ws->sample_size); + for (i = 0; i < BUCKET_SIZE && ws->bucket[i].count > 0; i++) { + p = ws->bucket[i].count; + p_base = ilog2_w(p); + entropy_sum += p * (sz_base - p_base); + } + + entropy_sum /= ws->sample_size; + return entropy_sum * 100 / entropy_max; +} + +#define RADIX_BASE 4U +#define COUNTERS_SIZE (1U << RADIX_BASE) + +static u8 get4bits(u64 num, int shift) { + u8 low4bits; + + num >>= shift; + /* Reverse order */ + low4bits = (COUNTERS_SIZE - 1) - (num % COUNTERS_SIZE); + return low4bits; +} + +/* + * Use 4 bits as radix base + * Use 16 u32 counters for calculating new position in buf array + * + * @array - array that will be sorted + * @array_buf - buffer array to store sorting results + * must be equal in size to @array + * @num - array size + */ +static void radix_sort(struct bucket_item *array, struct bucket_item *array_buf, + int num) { - free_workspaces(); + u64 max_num; + u64 buf_num; + u32 counters[COUNTERS_SIZE]; + u32 new_addr; + u32 addr; + int bitlen; + int shift; + int i; + + /* + * Try avoid useless loop iterations for small numbers stored in big + * counters. Example: 48 33 4 ... in 64bit array + */ + max_num = array[0].count; + for (i = 1; i < num; i++) { + buf_num = array[i].count; + if (buf_num > max_num) + max_num = buf_num; + } + + buf_num = ilog2(max_num); + bitlen = ALIGN(buf_num, RADIX_BASE * 2); + + shift = 0; + while (shift < bitlen) { + memset(counters, 0, sizeof(counters)); + + for (i = 0; i < num; i++) { + buf_num = array[i].count; + addr = get4bits(buf_num, shift); + counters[addr]++; + } + + for (i = 1; i < COUNTERS_SIZE; i++) + counters[i] += counters[i - 1]; + + for (i = num - 1; i >= 0; i--) { + buf_num = array[i].count; + addr = get4bits(buf_num, shift); + counters[addr]--; + new_addr = counters[addr]; + array_buf[new_addr] = array[i]; + } + + shift += RADIX_BASE; + + /* + * Normal radix expects to move data from a temporary array, to + * the main one. But that requires some CPU time. Avoid that + * by doing another sort iteration to original array instead of + * memcpy() + */ + memset(counters, 0, sizeof(counters)); + + for (i = 0; i < num; i ++) { + buf_num = array_buf[i].count; + addr = get4bits(buf_num, shift); + counters[addr]++; + } + + for (i = 1; i < COUNTERS_SIZE; i++) + counters[i] += counters[i - 1]; + + for (i = num - 1; i >= 0; i--) { + buf_num = array_buf[i].count; + addr = get4bits(buf_num, shift); + counters[addr]--; + new_addr = counters[addr]; + array[new_addr] = array_buf[i]; + } + + shift += RADIX_BASE; + } } /* - * Copy uncompressed data from working buffer to pages. + * Size of the core byte set - how many bytes cover 90% of the sample + * + * There are several types of structured binary data that use nearly all byte + * values. The distribution can be uniform and counts in all buckets will be + * nearly the same (eg. encrypted data). Unlikely to be compressible. * - * buf_start is the byte offset we're of the start of our workspace buffer. + * Other possibility is normal (Gaussian) distribution, where the data could + * be potentially compressible, but we have to take a few more steps to decide + * how much. * - * total_out is the last byte of the buffer + * @BYTE_CORE_SET_LOW - main part of byte values repeated frequently, + * compression algo can easy fix that + * @BYTE_CORE_SET_HIGH - data have uniform distribution and with high + * probability is not compressible */ -int btrfs_decompress_buf2page(char *buf, unsigned long buf_start, - unsigned long total_out, u64 disk_start, - struct bio_vec *bvec, int vcnt, - unsigned long *pg_index, - unsigned long *pg_offset) +#define BYTE_CORE_SET_LOW (64) +#define BYTE_CORE_SET_HIGH (200) + +static int byte_core_set_size(struct heuristic_ws *ws) +{ + u32 i; + u32 coreset_sum = 0; + const u32 core_set_threshold = ws->sample_size * 90 / 100; + struct bucket_item *bucket = ws->bucket; + + /* Sort in reverse order */ + radix_sort(ws->bucket, ws->bucket_b, BUCKET_SIZE); + + for (i = 0; i < BYTE_CORE_SET_LOW; i++) + coreset_sum += bucket[i].count; + + if (coreset_sum > core_set_threshold) + return i; + + for (; i < BYTE_CORE_SET_HIGH && bucket[i].count > 0; i++) { + coreset_sum += bucket[i].count; + if (coreset_sum > core_set_threshold) + break; + } + + return i; +} + +/* + * Count byte values in buckets. + * This heuristic can detect textual data (configs, xml, json, html, etc). + * Because in most text-like data byte set is restricted to limited number of + * possible characters, and that restriction in most cases makes data easy to + * compress. + * + * @BYTE_SET_THRESHOLD - consider all data within this byte set size: + * less - compressible + * more - need additional analysis + */ +#define BYTE_SET_THRESHOLD (64) + +static u32 byte_set_size(const struct heuristic_ws *ws) +{ + u32 i; + u32 byte_set_size = 0; + + for (i = 0; i < BYTE_SET_THRESHOLD; i++) { + if (ws->bucket[i].count > 0) + byte_set_size++; + } + + /* + * Continue collecting count of byte values in buckets. If the byte + * set size is bigger then the threshold, it's pointless to continue, + * the detection technique would fail for this type of data. + */ + for (; i < BUCKET_SIZE; i++) { + if (ws->bucket[i].count > 0) { + byte_set_size++; + if (byte_set_size > BYTE_SET_THRESHOLD) + return byte_set_size; + } + } + + return byte_set_size; +} + +static bool sample_repeated_patterns(struct heuristic_ws *ws) +{ + const u32 half_of_sample = ws->sample_size / 2; + const u8 *data = ws->sample; + + return memcmp(&data[0], &data[half_of_sample], half_of_sample) == 0; +} + +static void heuristic_collect_sample(struct inode *inode, u64 start, u64 end, + struct heuristic_ws *ws) { - unsigned long buf_offset; - unsigned long current_buf_start; - unsigned long start_byte; - unsigned long working_bytes = total_out - buf_start; - unsigned long bytes; - char *kaddr; - struct page *page_out = bvec[*pg_index].bv_page; + struct page *page; + pgoff_t index, index_end; + u32 i, curr_sample_pos; + u8 *in_data; /* - * start byte is the first byte of the page we're currently - * copying into relative to the start of the compressed data. + * Compression handles the input data by chunks of 128KiB + * (defined by BTRFS_MAX_UNCOMPRESSED) + * + * We do the same for the heuristic and loop over the whole range. + * + * MAX_SAMPLE_SIZE - calculated under assumption that heuristic will + * process no more than BTRFS_MAX_UNCOMPRESSED at a time. */ - start_byte = page_offset(page_out) - disk_start; + if (end - start > BTRFS_MAX_UNCOMPRESSED) + end = start + BTRFS_MAX_UNCOMPRESSED; + + index = start >> PAGE_SHIFT; + index_end = end >> PAGE_SHIFT; + + /* Don't miss unaligned end */ + if (!PAGE_ALIGNED(end)) + index_end++; + + curr_sample_pos = 0; + while (index < index_end) { + page = find_get_page(inode->i_mapping, index); + in_data = kmap_local_page(page); + /* Handle case where the start is not aligned to PAGE_SIZE */ + i = start % PAGE_SIZE; + while (i < PAGE_SIZE - SAMPLING_READ_SIZE) { + /* Don't sample any garbage from the last page */ + if (start > end - SAMPLING_READ_SIZE) + break; + memcpy(&ws->sample[curr_sample_pos], &in_data[i], + SAMPLING_READ_SIZE); + i += SAMPLING_INTERVAL; + start += SAMPLING_INTERVAL; + curr_sample_pos += SAMPLING_READ_SIZE; + } + kunmap_local(in_data); + put_page(page); + + index++; + } + + ws->sample_size = curr_sample_pos; +} + +/* + * Compression heuristic. + * + * The following types of analysis can be performed: + * - detect mostly zero data + * - detect data with low "byte set" size (text, etc) + * - detect data with low/high "core byte" set + * + * Return non-zero if the compression should be done, 0 otherwise. + */ +int btrfs_compress_heuristic(struct btrfs_inode *inode, u64 start, u64 end) +{ + struct btrfs_fs_info *fs_info = inode->root->fs_info; + struct list_head *ws_list = get_workspace(fs_info, 0, 0); + struct heuristic_ws *ws; + u32 i; + u8 byte; + int ret = 0; + + ws = list_entry(ws_list, struct heuristic_ws, list); + + heuristic_collect_sample(&inode->vfs_inode, start, end, ws); + + if (sample_repeated_patterns(ws)) { + ret = 1; + goto out; + } + + memset(ws->bucket, 0, sizeof(*ws->bucket)*BUCKET_SIZE); + + for (i = 0; i < ws->sample_size; i++) { + byte = ws->sample[i]; + ws->bucket[byte].count++; + } + + i = byte_set_size(ws); + if (i < BYTE_SET_THRESHOLD) { + ret = 2; + goto out; + } + + i = byte_core_set_size(ws); + if (i <= BYTE_CORE_SET_LOW) { + ret = 3; + goto out; + } + + if (i >= BYTE_CORE_SET_HIGH) { + ret = 0; + goto out; + } - /* we haven't yet hit data corresponding to this page */ - if (total_out <= start_byte) - return 1; + i = shannon_entropy(ws); + if (i <= ENTROPY_LVL_ACEPTABLE) { + ret = 4; + goto out; + } /* - * the start of the data we care about is offset into - * the middle of our working buffer + * For the levels below ENTROPY_LVL_HIGH, additional analysis would be + * needed to give green light to compression. + * + * For now just assume that compression at that level is not worth the + * resources because: + * + * 1. it is possible to defrag the data later + * + * 2. the data would turn out to be hardly compressible, eg. 150 byte + * values, every bucket has counter at level ~54. The heuristic would + * be confused. This can happen when data have some internal repeated + * patterns like "abbacbbc...". This can be detected by analyzing + * pairs of bytes, which is too costly. */ - if (total_out > start_byte && buf_start < start_byte) { - buf_offset = start_byte - buf_start; - working_bytes -= buf_offset; + if (i < ENTROPY_LVL_HIGH) { + ret = 5; + goto out; } else { - buf_offset = 0; + ret = 0; + goto out; } - current_buf_start = buf_start; - - /* copy bytes from the working buffer into the pages */ - while (working_bytes > 0) { - bytes = min(PAGE_CACHE_SIZE - *pg_offset, - PAGE_CACHE_SIZE - buf_offset); - bytes = min(bytes, working_bytes); - kaddr = kmap_atomic(page_out); - memcpy(kaddr + *pg_offset, buf + buf_offset, bytes); - kunmap_atomic(kaddr); - flush_dcache_page(page_out); - - *pg_offset += bytes; - buf_offset += bytes; - working_bytes -= bytes; - current_buf_start += bytes; - - /* check if we need to pick another page */ - if (*pg_offset == PAGE_CACHE_SIZE) { - (*pg_index)++; - if (*pg_index >= vcnt) - return 0; - - page_out = bvec[*pg_index].bv_page; - *pg_offset = 0; - start_byte = page_offset(page_out) - disk_start; - /* - * make sure our new page is covered by this - * working buffer - */ - if (total_out <= start_byte) - return 1; +out: + put_workspace(fs_info, 0, ws_list); + return ret; +} - /* - * the next page in the biovec might not be adjacent - * to the last page, but it might still be found - * inside this working buffer. bump our offset pointer - */ - if (total_out > start_byte && - current_buf_start < start_byte) { - buf_offset = start_byte - buf_start; - working_bytes = total_out - start_byte; - current_buf_start = buf_start + buf_offset; - } - } +/* + * Convert the compression suffix (eg. after "zlib" starting with ":") to level. + * + * If the resulting level exceeds the algo's supported levels, it will be clamped. + * + * Return <0 if no valid string can be found. + * Return 0 if everything is fine. + */ +int btrfs_compress_str2level(unsigned int type, const char *str, int *level_ret) +{ + int level = 0; + int ret; + + if (!type) { + *level_ret = btrfs_compress_set_level(type, level); + return 0; } - return 1; + if (str[0] == ':') { + ret = kstrtoint(str + 1, 10, &level); + if (ret) + return ret; + } + + *level_ret = btrfs_compress_set_level(type, level); + return 0; } |