1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2008 Oracle. All rights reserved. 4 */ 5 6 #include <linux/kernel.h> 7 #include <linux/bio.h> 8 #include <linux/file.h> 9 #include <linux/fs.h> 10 #include <linux/pagemap.h> 11 #include <linux/pagevec.h> 12 #include <linux/highmem.h> 13 #include <linux/kthread.h> 14 #include <linux/time.h> 15 #include <linux/init.h> 16 #include <linux/string.h> 17 #include <linux/backing-dev.h> 18 #include <linux/writeback.h> 19 #include <linux/psi.h> 20 #include <linux/slab.h> 21 #include <linux/sched/mm.h> 22 #include <linux/log2.h> 23 #include <linux/shrinker.h> 24 #include <crypto/hash.h> 25 #include "misc.h" 26 #include "ctree.h" 27 #include "fs.h" 28 #include "btrfs_inode.h" 29 #include "bio.h" 30 #include "ordered-data.h" 31 #include "compression.h" 32 #include "extent_io.h" 33 #include "extent_map.h" 34 #include "subpage.h" 35 #include "messages.h" 36 #include "super.h" 37 38 static struct bio_set btrfs_compressed_bioset; 39 40 static const char* const btrfs_compress_types[] = { "", "zlib", "lzo", "zstd" }; 41 42 const char* btrfs_compress_type2str(enum btrfs_compression_type type) 43 { 44 switch (type) { 45 case BTRFS_COMPRESS_ZLIB: 46 case BTRFS_COMPRESS_LZO: 47 case BTRFS_COMPRESS_ZSTD: 48 case BTRFS_COMPRESS_NONE: 49 return btrfs_compress_types[type]; 50 default: 51 break; 52 } 53 54 return NULL; 55 } 56 57 static inline struct compressed_bio *to_compressed_bio(struct btrfs_bio *bbio) 58 { 59 return container_of(bbio, struct compressed_bio, bbio); 60 } 61 62 static struct compressed_bio *alloc_compressed_bio(struct btrfs_inode *inode, 63 u64 start, blk_opf_t op, 64 btrfs_bio_end_io_t end_io) 65 { 66 struct btrfs_bio *bbio; 67 68 bbio = btrfs_bio(bio_alloc_bioset(NULL, BTRFS_MAX_COMPRESSED_PAGES, op, 69 GFP_NOFS, &btrfs_compressed_bioset)); 70 btrfs_bio_init(bbio, inode->root->fs_info, end_io, NULL); 71 bbio->inode = inode; 72 bbio->file_offset = start; 73 return to_compressed_bio(bbio); 74 } 75 76 bool btrfs_compress_is_valid_type(const char *str, size_t len) 77 { 78 int i; 79 80 for (i = 1; i < ARRAY_SIZE(btrfs_compress_types); i++) { 81 size_t comp_len = strlen(btrfs_compress_types[i]); 82 83 if (len < comp_len) 84 continue; 85 86 if (!strncmp(btrfs_compress_types[i], str, comp_len)) 87 return true; 88 } 89 return false; 90 } 91 92 static int compression_compress_pages(int type, struct list_head *ws, 93 struct btrfs_inode *inode, u64 start, 94 struct folio **folios, unsigned long *out_folios, 95 unsigned long *total_in, unsigned long *total_out) 96 { 97 switch (type) { 98 case BTRFS_COMPRESS_ZLIB: 99 return zlib_compress_folios(ws, inode, start, folios, 100 out_folios, total_in, total_out); 101 case BTRFS_COMPRESS_LZO: 102 return lzo_compress_folios(ws, inode, start, folios, 103 out_folios, total_in, total_out); 104 case BTRFS_COMPRESS_ZSTD: 105 return zstd_compress_folios(ws, inode, start, folios, 106 out_folios, total_in, total_out); 107 case BTRFS_COMPRESS_NONE: 108 default: 109 /* 110 * This can happen when compression races with remount setting 111 * it to 'no compress', while caller doesn't call 112 * inode_need_compress() to check if we really need to 113 * compress. 114 * 115 * Not a big deal, just need to inform caller that we 116 * haven't allocated any pages yet. 117 */ 118 *out_folios = 0; 119 return -E2BIG; 120 } 121 } 122 123 static int compression_decompress_bio(struct list_head *ws, 124 struct compressed_bio *cb) 125 { 126 switch (cb->compress_type) { 127 case BTRFS_COMPRESS_ZLIB: return zlib_decompress_bio(ws, cb); 128 case BTRFS_COMPRESS_LZO: return lzo_decompress_bio(ws, cb); 129 case BTRFS_COMPRESS_ZSTD: return zstd_decompress_bio(ws, cb); 130 case BTRFS_COMPRESS_NONE: 131 default: 132 /* 133 * This can't happen, the type is validated several times 134 * before we get here. 135 */ 136 BUG(); 137 } 138 } 139 140 static int compression_decompress(int type, struct list_head *ws, 141 const u8 *data_in, struct folio *dest_folio, 142 unsigned long dest_pgoff, size_t srclen, size_t destlen) 143 { 144 switch (type) { 145 case BTRFS_COMPRESS_ZLIB: return zlib_decompress(ws, data_in, dest_folio, 146 dest_pgoff, srclen, destlen); 147 case BTRFS_COMPRESS_LZO: return lzo_decompress(ws, data_in, dest_folio, 148 dest_pgoff, srclen, destlen); 149 case BTRFS_COMPRESS_ZSTD: return zstd_decompress(ws, data_in, dest_folio, 150 dest_pgoff, srclen, destlen); 151 case BTRFS_COMPRESS_NONE: 152 default: 153 /* 154 * This can't happen, the type is validated several times 155 * before we get here. 156 */ 157 BUG(); 158 } 159 } 160 161 static void btrfs_free_compressed_folios(struct compressed_bio *cb) 162 { 163 for (unsigned int i = 0; i < cb->nr_folios; i++) 164 btrfs_free_compr_folio(cb->compressed_folios[i]); 165 kfree(cb->compressed_folios); 166 } 167 168 static int btrfs_decompress_bio(struct compressed_bio *cb); 169 170 /* 171 * Global cache of last unused pages for compression/decompression. 172 */ 173 static struct btrfs_compr_pool { 174 struct shrinker *shrinker; 175 spinlock_t lock; 176 struct list_head list; 177 int count; 178 int thresh; 179 } compr_pool; 180 181 static unsigned long btrfs_compr_pool_count(struct shrinker *sh, struct shrink_control *sc) 182 { 183 int ret; 184 185 /* 186 * We must not read the values more than once if 'ret' gets expanded in 187 * the return statement so we don't accidentally return a negative 188 * number, even if the first condition finds it positive. 189 */ 190 ret = READ_ONCE(compr_pool.count) - READ_ONCE(compr_pool.thresh); 191 192 return ret > 0 ? ret : 0; 193 } 194 195 static unsigned long btrfs_compr_pool_scan(struct shrinker *sh, struct shrink_control *sc) 196 { 197 struct list_head remove; 198 struct list_head *tmp, *next; 199 int freed; 200 201 if (compr_pool.count == 0) 202 return SHRINK_STOP; 203 204 INIT_LIST_HEAD(&remove); 205 206 /* For now, just simply drain the whole list. */ 207 spin_lock(&compr_pool.lock); 208 list_splice_init(&compr_pool.list, &remove); 209 freed = compr_pool.count; 210 compr_pool.count = 0; 211 spin_unlock(&compr_pool.lock); 212 213 list_for_each_safe(tmp, next, &remove) { 214 struct page *page = list_entry(tmp, struct page, lru); 215 216 ASSERT(page_ref_count(page) == 1); 217 put_page(page); 218 } 219 220 return freed; 221 } 222 223 /* 224 * Common wrappers for page allocation from compression wrappers 225 */ 226 struct folio *btrfs_alloc_compr_folio(struct btrfs_fs_info *fs_info) 227 { 228 struct folio *folio = NULL; 229 230 /* For bs > ps cases, no cached folio pool for now. */ 231 if (fs_info->block_min_order) 232 goto alloc; 233 234 spin_lock(&compr_pool.lock); 235 if (compr_pool.count > 0) { 236 folio = list_first_entry(&compr_pool.list, struct folio, lru); 237 list_del_init(&folio->lru); 238 compr_pool.count--; 239 } 240 spin_unlock(&compr_pool.lock); 241 242 if (folio) 243 return folio; 244 245 alloc: 246 return folio_alloc(GFP_NOFS, fs_info->block_min_order); 247 } 248 249 void btrfs_free_compr_folio(struct folio *folio) 250 { 251 bool do_free = false; 252 253 /* The folio is from bs > ps fs, no cached pool for now. */ 254 if (folio_order(folio)) 255 goto free; 256 257 spin_lock(&compr_pool.lock); 258 if (compr_pool.count > compr_pool.thresh) { 259 do_free = true; 260 } else { 261 list_add(&folio->lru, &compr_pool.list); 262 compr_pool.count++; 263 } 264 spin_unlock(&compr_pool.lock); 265 266 if (!do_free) 267 return; 268 269 free: 270 ASSERT(folio_ref_count(folio) == 1); 271 folio_put(folio); 272 } 273 274 static void end_bbio_compressed_read(struct btrfs_bio *bbio) 275 { 276 struct compressed_bio *cb = to_compressed_bio(bbio); 277 blk_status_t status = bbio->bio.bi_status; 278 279 if (!status) 280 status = errno_to_blk_status(btrfs_decompress_bio(cb)); 281 282 btrfs_free_compressed_folios(cb); 283 btrfs_bio_end_io(cb->orig_bbio, status); 284 bio_put(&bbio->bio); 285 } 286 287 /* 288 * Clear the writeback bits on all of the file 289 * pages for a compressed write 290 */ 291 static noinline void end_compressed_writeback(const struct compressed_bio *cb) 292 { 293 struct inode *inode = &cb->bbio.inode->vfs_inode; 294 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode); 295 pgoff_t index = cb->start >> PAGE_SHIFT; 296 const pgoff_t end_index = (cb->start + cb->len - 1) >> PAGE_SHIFT; 297 struct folio_batch fbatch; 298 int i; 299 int ret; 300 301 ret = blk_status_to_errno(cb->bbio.bio.bi_status); 302 if (ret) 303 mapping_set_error(inode->i_mapping, ret); 304 305 folio_batch_init(&fbatch); 306 while (index <= end_index) { 307 ret = filemap_get_folios(inode->i_mapping, &index, end_index, 308 &fbatch); 309 310 if (ret == 0) 311 return; 312 313 for (i = 0; i < ret; i++) { 314 struct folio *folio = fbatch.folios[i]; 315 316 btrfs_folio_clamp_clear_writeback(fs_info, folio, 317 cb->start, cb->len); 318 } 319 folio_batch_release(&fbatch); 320 } 321 /* the inode may be gone now */ 322 } 323 324 static void btrfs_finish_compressed_write_work(struct work_struct *work) 325 { 326 struct compressed_bio *cb = 327 container_of(work, struct compressed_bio, write_end_work); 328 329 btrfs_finish_ordered_extent(cb->bbio.ordered, NULL, cb->start, cb->len, 330 cb->bbio.bio.bi_status == BLK_STS_OK); 331 332 if (cb->writeback) 333 end_compressed_writeback(cb); 334 /* Note, our inode could be gone now */ 335 336 btrfs_free_compressed_folios(cb); 337 bio_put(&cb->bbio.bio); 338 } 339 340 /* 341 * Do the cleanup once all the compressed pages hit the disk. This will clear 342 * writeback on the file pages and free the compressed pages. 343 * 344 * This also calls the writeback end hooks for the file pages so that metadata 345 * and checksums can be updated in the file. 346 */ 347 static void end_bbio_compressed_write(struct btrfs_bio *bbio) 348 { 349 struct compressed_bio *cb = to_compressed_bio(bbio); 350 struct btrfs_fs_info *fs_info = bbio->inode->root->fs_info; 351 352 queue_work(fs_info->compressed_write_workers, &cb->write_end_work); 353 } 354 355 static void btrfs_add_compressed_bio_folios(struct compressed_bio *cb) 356 { 357 struct btrfs_fs_info *fs_info = cb->bbio.fs_info; 358 struct bio *bio = &cb->bbio.bio; 359 u32 offset = 0; 360 361 while (offset < cb->compressed_len) { 362 struct folio *folio; 363 int ret; 364 u32 len = min_t(u32, cb->compressed_len - offset, 365 btrfs_min_folio_size(fs_info)); 366 367 folio = cb->compressed_folios[offset >> (PAGE_SHIFT + fs_info->block_min_order)]; 368 /* Maximum compressed extent is smaller than bio size limit. */ 369 ret = bio_add_folio(bio, folio, len, 0); 370 ASSERT(ret); 371 offset += len; 372 } 373 } 374 375 /* 376 * worker function to build and submit bios for previously compressed pages. 377 * The corresponding pages in the inode should be marked for writeback 378 * and the compressed pages should have a reference on them for dropping 379 * when the IO is complete. 380 * 381 * This also checksums the file bytes and gets things ready for 382 * the end io hooks. 383 */ 384 void btrfs_submit_compressed_write(struct btrfs_ordered_extent *ordered, 385 struct folio **compressed_folios, 386 unsigned int nr_folios, 387 blk_opf_t write_flags, 388 bool writeback) 389 { 390 struct btrfs_inode *inode = ordered->inode; 391 struct btrfs_fs_info *fs_info = inode->root->fs_info; 392 struct compressed_bio *cb; 393 394 ASSERT(IS_ALIGNED(ordered->file_offset, fs_info->sectorsize)); 395 ASSERT(IS_ALIGNED(ordered->num_bytes, fs_info->sectorsize)); 396 397 cb = alloc_compressed_bio(inode, ordered->file_offset, 398 REQ_OP_WRITE | write_flags, 399 end_bbio_compressed_write); 400 cb->start = ordered->file_offset; 401 cb->len = ordered->num_bytes; 402 cb->compressed_folios = compressed_folios; 403 cb->compressed_len = ordered->disk_num_bytes; 404 cb->writeback = writeback; 405 INIT_WORK(&cb->write_end_work, btrfs_finish_compressed_write_work); 406 cb->nr_folios = nr_folios; 407 cb->bbio.bio.bi_iter.bi_sector = ordered->disk_bytenr >> SECTOR_SHIFT; 408 cb->bbio.ordered = ordered; 409 btrfs_add_compressed_bio_folios(cb); 410 411 btrfs_submit_bbio(&cb->bbio, 0); 412 } 413 414 /* 415 * Add extra pages in the same compressed file extent so that we don't need to 416 * re-read the same extent again and again. 417 * 418 * NOTE: this won't work well for subpage, as for subpage read, we lock the 419 * full page then submit bio for each compressed/regular extents. 420 * 421 * This means, if we have several sectors in the same page points to the same 422 * on-disk compressed data, we will re-read the same extent many times and 423 * this function can only help for the next page. 424 */ 425 static noinline int add_ra_bio_pages(struct inode *inode, 426 u64 compressed_end, 427 struct compressed_bio *cb, 428 int *memstall, unsigned long *pflags) 429 { 430 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode); 431 pgoff_t end_index; 432 struct bio *orig_bio = &cb->orig_bbio->bio; 433 u64 cur = cb->orig_bbio->file_offset + orig_bio->bi_iter.bi_size; 434 u64 isize = i_size_read(inode); 435 int ret; 436 struct folio *folio; 437 struct extent_map *em; 438 struct address_space *mapping = inode->i_mapping; 439 struct extent_map_tree *em_tree; 440 struct extent_io_tree *tree; 441 int sectors_missed = 0; 442 443 em_tree = &BTRFS_I(inode)->extent_tree; 444 tree = &BTRFS_I(inode)->io_tree; 445 446 if (isize == 0) 447 return 0; 448 449 /* 450 * For current subpage support, we only support 64K page size, 451 * which means maximum compressed extent size (128K) is just 2x page 452 * size. 453 * This makes readahead less effective, so here disable readahead for 454 * subpage for now, until full compressed write is supported. 455 */ 456 if (fs_info->sectorsize < PAGE_SIZE) 457 return 0; 458 459 /* For bs > ps cases, we don't support readahead for compressed folios for now. */ 460 if (fs_info->block_min_order) 461 return 0; 462 463 end_index = (i_size_read(inode) - 1) >> PAGE_SHIFT; 464 465 while (cur < compressed_end) { 466 pgoff_t page_end; 467 pgoff_t pg_index = cur >> PAGE_SHIFT; 468 u32 add_size; 469 470 if (pg_index > end_index) 471 break; 472 473 folio = filemap_get_folio(mapping, pg_index); 474 if (!IS_ERR(folio)) { 475 u64 folio_sz = folio_size(folio); 476 u64 offset = offset_in_folio(folio, cur); 477 478 folio_put(folio); 479 sectors_missed += (folio_sz - offset) >> 480 fs_info->sectorsize_bits; 481 482 /* Beyond threshold, no need to continue */ 483 if (sectors_missed > 4) 484 break; 485 486 /* 487 * Jump to next page start as we already have page for 488 * current offset. 489 */ 490 cur += (folio_sz - offset); 491 continue; 492 } 493 494 folio = filemap_alloc_folio(mapping_gfp_constraint(mapping, 495 ~__GFP_FS), 0); 496 if (!folio) 497 break; 498 499 if (filemap_add_folio(mapping, folio, pg_index, GFP_NOFS)) { 500 /* There is already a page, skip to page end */ 501 cur += folio_size(folio); 502 folio_put(folio); 503 continue; 504 } 505 506 if (!*memstall && folio_test_workingset(folio)) { 507 psi_memstall_enter(pflags); 508 *memstall = 1; 509 } 510 511 ret = set_folio_extent_mapped(folio); 512 if (ret < 0) { 513 folio_unlock(folio); 514 folio_put(folio); 515 break; 516 } 517 518 page_end = (pg_index << PAGE_SHIFT) + folio_size(folio) - 1; 519 btrfs_lock_extent(tree, cur, page_end, NULL); 520 read_lock(&em_tree->lock); 521 em = btrfs_lookup_extent_mapping(em_tree, cur, page_end + 1 - cur); 522 read_unlock(&em_tree->lock); 523 524 /* 525 * At this point, we have a locked page in the page cache for 526 * these bytes in the file. But, we have to make sure they map 527 * to this compressed extent on disk. 528 */ 529 if (!em || cur < em->start || 530 (cur + fs_info->sectorsize > btrfs_extent_map_end(em)) || 531 (btrfs_extent_map_block_start(em) >> SECTOR_SHIFT) != 532 orig_bio->bi_iter.bi_sector) { 533 btrfs_free_extent_map(em); 534 btrfs_unlock_extent(tree, cur, page_end, NULL); 535 folio_unlock(folio); 536 folio_put(folio); 537 break; 538 } 539 add_size = min(em->start + em->len, page_end + 1) - cur; 540 btrfs_free_extent_map(em); 541 btrfs_unlock_extent(tree, cur, page_end, NULL); 542 543 if (folio_contains(folio, end_index)) { 544 size_t zero_offset = offset_in_folio(folio, isize); 545 546 if (zero_offset) { 547 int zeros; 548 zeros = folio_size(folio) - zero_offset; 549 folio_zero_range(folio, zero_offset, zeros); 550 } 551 } 552 553 if (!bio_add_folio(orig_bio, folio, add_size, 554 offset_in_folio(folio, cur))) { 555 folio_unlock(folio); 556 folio_put(folio); 557 break; 558 } 559 /* 560 * If it's subpage, we also need to increase its 561 * subpage::readers number, as at endio we will decrease 562 * subpage::readers and to unlock the page. 563 */ 564 if (fs_info->sectorsize < PAGE_SIZE) 565 btrfs_folio_set_lock(fs_info, folio, cur, add_size); 566 folio_put(folio); 567 cur += add_size; 568 } 569 return 0; 570 } 571 572 /* 573 * for a compressed read, the bio we get passed has all the inode pages 574 * in it. We don't actually do IO on those pages but allocate new ones 575 * to hold the compressed pages on disk. 576 * 577 * bio->bi_iter.bi_sector points to the compressed extent on disk 578 * bio->bi_io_vec points to all of the inode pages 579 * 580 * After the compressed pages are read, we copy the bytes into the 581 * bio we were passed and then call the bio end_io calls 582 */ 583 void btrfs_submit_compressed_read(struct btrfs_bio *bbio) 584 { 585 struct btrfs_inode *inode = bbio->inode; 586 struct btrfs_fs_info *fs_info = inode->root->fs_info; 587 struct extent_map_tree *em_tree = &inode->extent_tree; 588 struct compressed_bio *cb; 589 unsigned int compressed_len; 590 u64 file_offset = bbio->file_offset; 591 u64 em_len; 592 u64 em_start; 593 struct extent_map *em; 594 unsigned long pflags; 595 int memstall = 0; 596 blk_status_t status; 597 int ret; 598 599 /* we need the actual starting offset of this extent in the file */ 600 read_lock(&em_tree->lock); 601 em = btrfs_lookup_extent_mapping(em_tree, file_offset, fs_info->sectorsize); 602 read_unlock(&em_tree->lock); 603 if (!em) { 604 status = BLK_STS_IOERR; 605 goto out; 606 } 607 608 ASSERT(btrfs_extent_map_is_compressed(em)); 609 compressed_len = em->disk_num_bytes; 610 611 cb = alloc_compressed_bio(inode, file_offset, REQ_OP_READ, 612 end_bbio_compressed_read); 613 614 cb->start = em->start - em->offset; 615 em_len = em->len; 616 em_start = em->start; 617 618 cb->len = bbio->bio.bi_iter.bi_size; 619 cb->compressed_len = compressed_len; 620 cb->compress_type = btrfs_extent_map_compression(em); 621 cb->orig_bbio = bbio; 622 cb->bbio.csum_search_commit_root = bbio->csum_search_commit_root; 623 624 btrfs_free_extent_map(em); 625 626 cb->nr_folios = DIV_ROUND_UP(compressed_len, btrfs_min_folio_size(fs_info)); 627 cb->compressed_folios = kcalloc(cb->nr_folios, sizeof(struct folio *), GFP_NOFS); 628 if (!cb->compressed_folios) { 629 status = BLK_STS_RESOURCE; 630 goto out_free_bio; 631 } 632 633 ret = btrfs_alloc_folio_array(cb->nr_folios, fs_info->block_min_order, 634 cb->compressed_folios); 635 if (ret) { 636 status = BLK_STS_RESOURCE; 637 goto out_free_compressed_pages; 638 } 639 640 add_ra_bio_pages(&inode->vfs_inode, em_start + em_len, cb, &memstall, 641 &pflags); 642 643 /* include any pages we added in add_ra-bio_pages */ 644 cb->len = bbio->bio.bi_iter.bi_size; 645 cb->bbio.bio.bi_iter.bi_sector = bbio->bio.bi_iter.bi_sector; 646 btrfs_add_compressed_bio_folios(cb); 647 648 if (memstall) 649 psi_memstall_leave(&pflags); 650 651 btrfs_submit_bbio(&cb->bbio, 0); 652 return; 653 654 out_free_compressed_pages: 655 kfree(cb->compressed_folios); 656 out_free_bio: 657 bio_put(&cb->bbio.bio); 658 out: 659 btrfs_bio_end_io(bbio, status); 660 } 661 662 /* 663 * Heuristic uses systematic sampling to collect data from the input data 664 * range, the logic can be tuned by the following constants: 665 * 666 * @SAMPLING_READ_SIZE - how many bytes will be copied from for each sample 667 * @SAMPLING_INTERVAL - range from which the sampled data can be collected 668 */ 669 #define SAMPLING_READ_SIZE (16) 670 #define SAMPLING_INTERVAL (256) 671 672 /* 673 * For statistical analysis of the input data we consider bytes that form a 674 * Galois Field of 256 objects. Each object has an attribute count, ie. how 675 * many times the object appeared in the sample. 676 */ 677 #define BUCKET_SIZE (256) 678 679 /* 680 * The size of the sample is based on a statistical sampling rule of thumb. 681 * The common way is to perform sampling tests as long as the number of 682 * elements in each cell is at least 5. 683 * 684 * Instead of 5, we choose 32 to obtain more accurate results. 685 * If the data contain the maximum number of symbols, which is 256, we obtain a 686 * sample size bound by 8192. 687 * 688 * For a sample of at most 8KB of data per data range: 16 consecutive bytes 689 * from up to 512 locations. 690 */ 691 #define MAX_SAMPLE_SIZE (BTRFS_MAX_UNCOMPRESSED * \ 692 SAMPLING_READ_SIZE / SAMPLING_INTERVAL) 693 694 struct bucket_item { 695 u32 count; 696 }; 697 698 struct heuristic_ws { 699 /* Partial copy of input data */ 700 u8 *sample; 701 u32 sample_size; 702 /* Buckets store counters for each byte value */ 703 struct bucket_item *bucket; 704 /* Sorting buffer */ 705 struct bucket_item *bucket_b; 706 struct list_head list; 707 }; 708 709 static void free_heuristic_ws(struct list_head *ws) 710 { 711 struct heuristic_ws *workspace; 712 713 workspace = list_entry(ws, struct heuristic_ws, list); 714 715 kvfree(workspace->sample); 716 kfree(workspace->bucket); 717 kfree(workspace->bucket_b); 718 kfree(workspace); 719 } 720 721 static struct list_head *alloc_heuristic_ws(struct btrfs_fs_info *fs_info) 722 { 723 struct heuristic_ws *ws; 724 725 ws = kzalloc(sizeof(*ws), GFP_KERNEL); 726 if (!ws) 727 return ERR_PTR(-ENOMEM); 728 729 ws->sample = kvmalloc(MAX_SAMPLE_SIZE, GFP_KERNEL); 730 if (!ws->sample) 731 goto fail; 732 733 ws->bucket = kcalloc(BUCKET_SIZE, sizeof(*ws->bucket), GFP_KERNEL); 734 if (!ws->bucket) 735 goto fail; 736 737 ws->bucket_b = kcalloc(BUCKET_SIZE, sizeof(*ws->bucket_b), GFP_KERNEL); 738 if (!ws->bucket_b) 739 goto fail; 740 741 INIT_LIST_HEAD(&ws->list); 742 return &ws->list; 743 fail: 744 free_heuristic_ws(&ws->list); 745 return ERR_PTR(-ENOMEM); 746 } 747 748 const struct btrfs_compress_levels btrfs_heuristic_compress = { 0 }; 749 750 static const struct btrfs_compress_levels * const btrfs_compress_levels[] = { 751 /* The heuristic is represented as compression type 0 */ 752 &btrfs_heuristic_compress, 753 &btrfs_zlib_compress, 754 &btrfs_lzo_compress, 755 &btrfs_zstd_compress, 756 }; 757 758 static struct list_head *alloc_workspace(struct btrfs_fs_info *fs_info, int type, int level) 759 { 760 switch (type) { 761 case BTRFS_COMPRESS_NONE: return alloc_heuristic_ws(fs_info); 762 case BTRFS_COMPRESS_ZLIB: return zlib_alloc_workspace(fs_info, level); 763 case BTRFS_COMPRESS_LZO: return lzo_alloc_workspace(fs_info); 764 case BTRFS_COMPRESS_ZSTD: return zstd_alloc_workspace(fs_info, level); 765 default: 766 /* 767 * This can't happen, the type is validated several times 768 * before we get here. 769 */ 770 BUG(); 771 } 772 } 773 774 static void free_workspace(int type, struct list_head *ws) 775 { 776 switch (type) { 777 case BTRFS_COMPRESS_NONE: return free_heuristic_ws(ws); 778 case BTRFS_COMPRESS_ZLIB: return zlib_free_workspace(ws); 779 case BTRFS_COMPRESS_LZO: return lzo_free_workspace(ws); 780 case BTRFS_COMPRESS_ZSTD: return zstd_free_workspace(ws); 781 default: 782 /* 783 * This can't happen, the type is validated several times 784 * before we get here. 785 */ 786 BUG(); 787 } 788 } 789 790 static int alloc_workspace_manager(struct btrfs_fs_info *fs_info, 791 enum btrfs_compression_type type) 792 { 793 struct workspace_manager *gwsm; 794 struct list_head *workspace; 795 796 ASSERT(fs_info->compr_wsm[type] == NULL); 797 gwsm = kzalloc(sizeof(*gwsm), GFP_KERNEL); 798 if (!gwsm) 799 return -ENOMEM; 800 801 INIT_LIST_HEAD(&gwsm->idle_ws); 802 spin_lock_init(&gwsm->ws_lock); 803 atomic_set(&gwsm->total_ws, 0); 804 init_waitqueue_head(&gwsm->ws_wait); 805 fs_info->compr_wsm[type] = gwsm; 806 807 /* 808 * Preallocate one workspace for each compression type so we can 809 * guarantee forward progress in the worst case 810 */ 811 workspace = alloc_workspace(fs_info, type, 0); 812 if (IS_ERR(workspace)) { 813 btrfs_warn(fs_info, 814 "cannot preallocate compression workspace for %s, will try later", 815 btrfs_compress_type2str(type)); 816 } else { 817 atomic_set(&gwsm->total_ws, 1); 818 gwsm->free_ws = 1; 819 list_add(workspace, &gwsm->idle_ws); 820 } 821 return 0; 822 } 823 824 static void free_workspace_manager(struct btrfs_fs_info *fs_info, 825 enum btrfs_compression_type type) 826 { 827 struct list_head *ws; 828 struct workspace_manager *gwsm = fs_info->compr_wsm[type]; 829 830 /* ZSTD uses its own workspace manager, should enter here. */ 831 ASSERT(type != BTRFS_COMPRESS_ZSTD && type < BTRFS_NR_COMPRESS_TYPES); 832 if (!gwsm) 833 return; 834 fs_info->compr_wsm[type] = NULL; 835 while (!list_empty(&gwsm->idle_ws)) { 836 ws = gwsm->idle_ws.next; 837 list_del(ws); 838 free_workspace(type, ws); 839 atomic_dec(&gwsm->total_ws); 840 } 841 kfree(gwsm); 842 } 843 844 /* 845 * This finds an available workspace or allocates a new one. 846 * If it's not possible to allocate a new one, waits until there's one. 847 * Preallocation makes a forward progress guarantees and we do not return 848 * errors. 849 */ 850 struct list_head *btrfs_get_workspace(struct btrfs_fs_info *fs_info, int type, int level) 851 { 852 struct workspace_manager *wsm = fs_info->compr_wsm[type]; 853 struct list_head *workspace; 854 int cpus = num_online_cpus(); 855 unsigned nofs_flag; 856 struct list_head *idle_ws; 857 spinlock_t *ws_lock; 858 atomic_t *total_ws; 859 wait_queue_head_t *ws_wait; 860 int *free_ws; 861 862 ASSERT(wsm); 863 idle_ws = &wsm->idle_ws; 864 ws_lock = &wsm->ws_lock; 865 total_ws = &wsm->total_ws; 866 ws_wait = &wsm->ws_wait; 867 free_ws = &wsm->free_ws; 868 869 again: 870 spin_lock(ws_lock); 871 if (!list_empty(idle_ws)) { 872 workspace = idle_ws->next; 873 list_del(workspace); 874 (*free_ws)--; 875 spin_unlock(ws_lock); 876 return workspace; 877 878 } 879 if (atomic_read(total_ws) > cpus) { 880 DEFINE_WAIT(wait); 881 882 spin_unlock(ws_lock); 883 prepare_to_wait(ws_wait, &wait, TASK_UNINTERRUPTIBLE); 884 if (atomic_read(total_ws) > cpus && !*free_ws) 885 schedule(); 886 finish_wait(ws_wait, &wait); 887 goto again; 888 } 889 atomic_inc(total_ws); 890 spin_unlock(ws_lock); 891 892 /* 893 * Allocation helpers call vmalloc that can't use GFP_NOFS, so we have 894 * to turn it off here because we might get called from the restricted 895 * context of btrfs_compress_bio/btrfs_compress_pages 896 */ 897 nofs_flag = memalloc_nofs_save(); 898 workspace = alloc_workspace(fs_info, type, level); 899 memalloc_nofs_restore(nofs_flag); 900 901 if (IS_ERR(workspace)) { 902 atomic_dec(total_ws); 903 wake_up(ws_wait); 904 905 /* 906 * Do not return the error but go back to waiting. There's a 907 * workspace preallocated for each type and the compression 908 * time is bounded so we get to a workspace eventually. This 909 * makes our caller's life easier. 910 * 911 * To prevent silent and low-probability deadlocks (when the 912 * initial preallocation fails), check if there are any 913 * workspaces at all. 914 */ 915 if (atomic_read(total_ws) == 0) { 916 static DEFINE_RATELIMIT_STATE(_rs, 917 /* once per minute */ 60 * HZ, 918 /* no burst */ 1); 919 920 if (__ratelimit(&_rs)) 921 btrfs_warn(fs_info, 922 "no compression workspaces, low memory, retrying"); 923 } 924 goto again; 925 } 926 return workspace; 927 } 928 929 static struct list_head *get_workspace(struct btrfs_fs_info *fs_info, int type, int level) 930 { 931 switch (type) { 932 case BTRFS_COMPRESS_NONE: return btrfs_get_workspace(fs_info, type, level); 933 case BTRFS_COMPRESS_ZLIB: return zlib_get_workspace(fs_info, level); 934 case BTRFS_COMPRESS_LZO: return btrfs_get_workspace(fs_info, type, level); 935 case BTRFS_COMPRESS_ZSTD: return zstd_get_workspace(fs_info, level); 936 default: 937 /* 938 * This can't happen, the type is validated several times 939 * before we get here. 940 */ 941 BUG(); 942 } 943 } 944 945 /* 946 * put a workspace struct back on the list or free it if we have enough 947 * idle ones sitting around 948 */ 949 void btrfs_put_workspace(struct btrfs_fs_info *fs_info, int type, struct list_head *ws) 950 { 951 struct workspace_manager *gwsm = fs_info->compr_wsm[type]; 952 struct list_head *idle_ws; 953 spinlock_t *ws_lock; 954 atomic_t *total_ws; 955 wait_queue_head_t *ws_wait; 956 int *free_ws; 957 958 ASSERT(gwsm); 959 idle_ws = &gwsm->idle_ws; 960 ws_lock = &gwsm->ws_lock; 961 total_ws = &gwsm->total_ws; 962 ws_wait = &gwsm->ws_wait; 963 free_ws = &gwsm->free_ws; 964 965 spin_lock(ws_lock); 966 if (*free_ws <= num_online_cpus()) { 967 list_add(ws, idle_ws); 968 (*free_ws)++; 969 spin_unlock(ws_lock); 970 goto wake; 971 } 972 spin_unlock(ws_lock); 973 974 free_workspace(type, ws); 975 atomic_dec(total_ws); 976 wake: 977 cond_wake_up(ws_wait); 978 } 979 980 static void put_workspace(struct btrfs_fs_info *fs_info, int type, struct list_head *ws) 981 { 982 switch (type) { 983 case BTRFS_COMPRESS_NONE: return btrfs_put_workspace(fs_info, type, ws); 984 case BTRFS_COMPRESS_ZLIB: return btrfs_put_workspace(fs_info, type, ws); 985 case BTRFS_COMPRESS_LZO: return btrfs_put_workspace(fs_info, type, ws); 986 case BTRFS_COMPRESS_ZSTD: return zstd_put_workspace(fs_info, ws); 987 default: 988 /* 989 * This can't happen, the type is validated several times 990 * before we get here. 991 */ 992 BUG(); 993 } 994 } 995 996 /* 997 * Adjust @level according to the limits of the compression algorithm or 998 * fallback to default 999 */ 1000 static int btrfs_compress_set_level(unsigned int type, int level) 1001 { 1002 const struct btrfs_compress_levels *levels = btrfs_compress_levels[type]; 1003 1004 if (level == 0) 1005 level = levels->default_level; 1006 else 1007 level = clamp(level, levels->min_level, levels->max_level); 1008 1009 return level; 1010 } 1011 1012 /* 1013 * Check whether the @level is within the valid range for the given type. 1014 */ 1015 bool btrfs_compress_level_valid(unsigned int type, int level) 1016 { 1017 const struct btrfs_compress_levels *levels = btrfs_compress_levels[type]; 1018 1019 return levels->min_level <= level && level <= levels->max_level; 1020 } 1021 1022 /* Wrapper around find_get_page(), with extra error message. */ 1023 int btrfs_compress_filemap_get_folio(struct address_space *mapping, u64 start, 1024 struct folio **in_folio_ret) 1025 { 1026 struct folio *in_folio; 1027 1028 /* 1029 * The compressed write path should have the folio locked already, thus 1030 * we only need to grab one reference. 1031 */ 1032 in_folio = filemap_get_folio(mapping, start >> PAGE_SHIFT); 1033 if (IS_ERR(in_folio)) { 1034 struct btrfs_inode *inode = BTRFS_I(mapping->host); 1035 1036 btrfs_crit(inode->root->fs_info, 1037 "failed to get page cache, root %lld ino %llu file offset %llu", 1038 btrfs_root_id(inode->root), btrfs_ino(inode), start); 1039 return -ENOENT; 1040 } 1041 *in_folio_ret = in_folio; 1042 return 0; 1043 } 1044 1045 /* 1046 * Given an address space and start and length, compress the bytes into @pages 1047 * that are allocated on demand. 1048 * 1049 * @type_level is encoded algorithm and level, where level 0 means whatever 1050 * default the algorithm chooses and is opaque here; 1051 * - compression algo are 0-3 1052 * - the level are bits 4-7 1053 * 1054 * @out_folios is an in/out parameter, holds maximum number of folios to allocate 1055 * and returns number of actually allocated folios 1056 * 1057 * @total_in is used to return the number of bytes actually read. It 1058 * may be smaller than the input length if we had to exit early because we 1059 * ran out of room in the folios array or because we cross the 1060 * max_out threshold. 1061 * 1062 * @total_out is an in/out parameter, must be set to the input length and will 1063 * be also used to return the total number of compressed bytes 1064 */ 1065 int btrfs_compress_folios(unsigned int type, int level, struct btrfs_inode *inode, 1066 u64 start, struct folio **folios, unsigned long *out_folios, 1067 unsigned long *total_in, unsigned long *total_out) 1068 { 1069 struct btrfs_fs_info *fs_info = inode->root->fs_info; 1070 const unsigned long orig_len = *total_out; 1071 struct list_head *workspace; 1072 int ret; 1073 1074 level = btrfs_compress_set_level(type, level); 1075 workspace = get_workspace(fs_info, type, level); 1076 ret = compression_compress_pages(type, workspace, inode, start, folios, 1077 out_folios, total_in, total_out); 1078 /* The total read-in bytes should be no larger than the input. */ 1079 ASSERT(*total_in <= orig_len); 1080 put_workspace(fs_info, type, workspace); 1081 return ret; 1082 } 1083 1084 static int btrfs_decompress_bio(struct compressed_bio *cb) 1085 { 1086 struct btrfs_fs_info *fs_info = cb_to_fs_info(cb); 1087 struct list_head *workspace; 1088 int ret; 1089 int type = cb->compress_type; 1090 1091 workspace = get_workspace(fs_info, type, 0); 1092 ret = compression_decompress_bio(workspace, cb); 1093 put_workspace(fs_info, type, workspace); 1094 1095 if (!ret) 1096 zero_fill_bio(&cb->orig_bbio->bio); 1097 return ret; 1098 } 1099 1100 /* 1101 * a less complex decompression routine. Our compressed data fits in a 1102 * single page, and we want to read a single page out of it. 1103 * start_byte tells us the offset into the compressed data we're interested in 1104 */ 1105 int btrfs_decompress(int type, const u8 *data_in, struct folio *dest_folio, 1106 unsigned long dest_pgoff, size_t srclen, size_t destlen) 1107 { 1108 struct btrfs_fs_info *fs_info = folio_to_fs_info(dest_folio); 1109 struct list_head *workspace; 1110 const u32 sectorsize = fs_info->sectorsize; 1111 int ret; 1112 1113 /* 1114 * The full destination folio range should not exceed the folio size. 1115 * And the @destlen should not exceed sectorsize, as this is only called for 1116 * inline file extents, which should not exceed sectorsize. 1117 */ 1118 ASSERT(dest_pgoff + destlen <= folio_size(dest_folio) && destlen <= sectorsize); 1119 1120 workspace = get_workspace(fs_info, type, 0); 1121 ret = compression_decompress(type, workspace, data_in, dest_folio, 1122 dest_pgoff, srclen, destlen); 1123 put_workspace(fs_info, type, workspace); 1124 1125 return ret; 1126 } 1127 1128 int btrfs_alloc_compress_wsm(struct btrfs_fs_info *fs_info) 1129 { 1130 int ret; 1131 1132 ret = alloc_workspace_manager(fs_info, BTRFS_COMPRESS_NONE); 1133 if (ret < 0) 1134 goto error; 1135 ret = alloc_workspace_manager(fs_info, BTRFS_COMPRESS_ZLIB); 1136 if (ret < 0) 1137 goto error; 1138 ret = alloc_workspace_manager(fs_info, BTRFS_COMPRESS_LZO); 1139 if (ret < 0) 1140 goto error; 1141 ret = zstd_alloc_workspace_manager(fs_info); 1142 if (ret < 0) 1143 goto error; 1144 return 0; 1145 error: 1146 btrfs_free_compress_wsm(fs_info); 1147 return ret; 1148 } 1149 1150 void btrfs_free_compress_wsm(struct btrfs_fs_info *fs_info) 1151 { 1152 free_workspace_manager(fs_info, BTRFS_COMPRESS_NONE); 1153 free_workspace_manager(fs_info, BTRFS_COMPRESS_ZLIB); 1154 free_workspace_manager(fs_info, BTRFS_COMPRESS_LZO); 1155 zstd_free_workspace_manager(fs_info); 1156 } 1157 1158 int __init btrfs_init_compress(void) 1159 { 1160 if (bioset_init(&btrfs_compressed_bioset, BIO_POOL_SIZE, 1161 offsetof(struct compressed_bio, bbio.bio), 1162 BIOSET_NEED_BVECS)) 1163 return -ENOMEM; 1164 1165 compr_pool.shrinker = shrinker_alloc(SHRINKER_NONSLAB, "btrfs-compr-pages"); 1166 if (!compr_pool.shrinker) 1167 return -ENOMEM; 1168 1169 spin_lock_init(&compr_pool.lock); 1170 INIT_LIST_HEAD(&compr_pool.list); 1171 compr_pool.count = 0; 1172 /* 128K / 4K = 32, for 8 threads is 256 pages. */ 1173 compr_pool.thresh = BTRFS_MAX_COMPRESSED / PAGE_SIZE * 8; 1174 compr_pool.shrinker->count_objects = btrfs_compr_pool_count; 1175 compr_pool.shrinker->scan_objects = btrfs_compr_pool_scan; 1176 compr_pool.shrinker->batch = 32; 1177 compr_pool.shrinker->seeks = DEFAULT_SEEKS; 1178 shrinker_register(compr_pool.shrinker); 1179 1180 return 0; 1181 } 1182 1183 void __cold btrfs_exit_compress(void) 1184 { 1185 /* For now scan drains all pages and does not touch the parameters. */ 1186 btrfs_compr_pool_scan(NULL, NULL); 1187 shrinker_free(compr_pool.shrinker); 1188 1189 bioset_exit(&btrfs_compressed_bioset); 1190 } 1191 1192 /* 1193 * The bvec is a single page bvec from a bio that contains folios from a filemap. 1194 * 1195 * Since the folio may be a large one, and if the bv_page is not a head page of 1196 * a large folio, then page->index is unreliable. 1197 * 1198 * Thus we need this helper to grab the proper file offset. 1199 */ 1200 static u64 file_offset_from_bvec(const struct bio_vec *bvec) 1201 { 1202 const struct page *page = bvec->bv_page; 1203 const struct folio *folio = page_folio(page); 1204 1205 return (page_pgoff(folio, page) << PAGE_SHIFT) + bvec->bv_offset; 1206 } 1207 1208 /* 1209 * Copy decompressed data from working buffer to pages. 1210 * 1211 * @buf: The decompressed data buffer 1212 * @buf_len: The decompressed data length 1213 * @decompressed: Number of bytes that are already decompressed inside the 1214 * compressed extent 1215 * @cb: The compressed extent descriptor 1216 * @orig_bio: The original bio that the caller wants to read for 1217 * 1218 * An easier to understand graph is like below: 1219 * 1220 * |<- orig_bio ->| |<- orig_bio->| 1221 * |<------- full decompressed extent ----->| 1222 * |<----------- @cb range ---->| 1223 * | |<-- @buf_len -->| 1224 * |<--- @decompressed --->| 1225 * 1226 * Note that, @cb can be a subpage of the full decompressed extent, but 1227 * @cb->start always has the same as the orig_file_offset value of the full 1228 * decompressed extent. 1229 * 1230 * When reading compressed extent, we have to read the full compressed extent, 1231 * while @orig_bio may only want part of the range. 1232 * Thus this function will ensure only data covered by @orig_bio will be copied 1233 * to. 1234 * 1235 * Return 0 if we have copied all needed contents for @orig_bio. 1236 * Return >0 if we need continue decompress. 1237 */ 1238 int btrfs_decompress_buf2page(const char *buf, u32 buf_len, 1239 struct compressed_bio *cb, u32 decompressed) 1240 { 1241 struct bio *orig_bio = &cb->orig_bbio->bio; 1242 /* Offset inside the full decompressed extent */ 1243 u32 cur_offset; 1244 1245 cur_offset = decompressed; 1246 /* The main loop to do the copy */ 1247 while (cur_offset < decompressed + buf_len) { 1248 struct bio_vec bvec; 1249 size_t copy_len; 1250 u32 copy_start; 1251 /* Offset inside the full decompressed extent */ 1252 u32 bvec_offset; 1253 void *kaddr; 1254 1255 bvec = bio_iter_iovec(orig_bio, orig_bio->bi_iter); 1256 /* 1257 * cb->start may underflow, but subtracting that value can still 1258 * give us correct offset inside the full decompressed extent. 1259 */ 1260 bvec_offset = file_offset_from_bvec(&bvec) - cb->start; 1261 1262 /* Haven't reached the bvec range, exit */ 1263 if (decompressed + buf_len <= bvec_offset) 1264 return 1; 1265 1266 copy_start = max(cur_offset, bvec_offset); 1267 copy_len = min(bvec_offset + bvec.bv_len, 1268 decompressed + buf_len) - copy_start; 1269 ASSERT(copy_len); 1270 1271 /* 1272 * Extra range check to ensure we didn't go beyond 1273 * @buf + @buf_len. 1274 */ 1275 ASSERT(copy_start - decompressed < buf_len); 1276 1277 kaddr = bvec_kmap_local(&bvec); 1278 memcpy(kaddr, buf + copy_start - decompressed, copy_len); 1279 kunmap_local(kaddr); 1280 1281 cur_offset += copy_len; 1282 bio_advance(orig_bio, copy_len); 1283 /* Finished the bio */ 1284 if (!orig_bio->bi_iter.bi_size) 1285 return 0; 1286 } 1287 return 1; 1288 } 1289 1290 /* 1291 * Shannon Entropy calculation 1292 * 1293 * Pure byte distribution analysis fails to determine compressibility of data. 1294 * Try calculating entropy to estimate the average minimum number of bits 1295 * needed to encode the sampled data. 1296 * 1297 * For convenience, return the percentage of needed bits, instead of amount of 1298 * bits directly. 1299 * 1300 * @ENTROPY_LVL_ACEPTABLE - below that threshold, sample has low byte entropy 1301 * and can be compressible with high probability 1302 * 1303 * @ENTROPY_LVL_HIGH - data are not compressible with high probability 1304 * 1305 * Use of ilog2() decreases precision, we lower the LVL to 5 to compensate. 1306 */ 1307 #define ENTROPY_LVL_ACEPTABLE (65) 1308 #define ENTROPY_LVL_HIGH (80) 1309 1310 /* 1311 * For increased precision in shannon_entropy calculation, 1312 * let's do pow(n, M) to save more digits after comma: 1313 * 1314 * - maximum int bit length is 64 1315 * - ilog2(MAX_SAMPLE_SIZE) -> 13 1316 * - 13 * 4 = 52 < 64 -> M = 4 1317 * 1318 * So use pow(n, 4). 1319 */ 1320 static inline u32 ilog2_w(u64 n) 1321 { 1322 return ilog2(n * n * n * n); 1323 } 1324 1325 static u32 shannon_entropy(struct heuristic_ws *ws) 1326 { 1327 const u32 entropy_max = 8 * ilog2_w(2); 1328 u32 entropy_sum = 0; 1329 u32 p, p_base, sz_base; 1330 u32 i; 1331 1332 sz_base = ilog2_w(ws->sample_size); 1333 for (i = 0; i < BUCKET_SIZE && ws->bucket[i].count > 0; i++) { 1334 p = ws->bucket[i].count; 1335 p_base = ilog2_w(p); 1336 entropy_sum += p * (sz_base - p_base); 1337 } 1338 1339 entropy_sum /= ws->sample_size; 1340 return entropy_sum * 100 / entropy_max; 1341 } 1342 1343 #define RADIX_BASE 4U 1344 #define COUNTERS_SIZE (1U << RADIX_BASE) 1345 1346 static u8 get4bits(u64 num, int shift) { 1347 u8 low4bits; 1348 1349 num >>= shift; 1350 /* Reverse order */ 1351 low4bits = (COUNTERS_SIZE - 1) - (num % COUNTERS_SIZE); 1352 return low4bits; 1353 } 1354 1355 /* 1356 * Use 4 bits as radix base 1357 * Use 16 u32 counters for calculating new position in buf array 1358 * 1359 * @array - array that will be sorted 1360 * @array_buf - buffer array to store sorting results 1361 * must be equal in size to @array 1362 * @num - array size 1363 */ 1364 static void radix_sort(struct bucket_item *array, struct bucket_item *array_buf, 1365 int num) 1366 { 1367 u64 max_num; 1368 u64 buf_num; 1369 u32 counters[COUNTERS_SIZE]; 1370 u32 new_addr; 1371 u32 addr; 1372 int bitlen; 1373 int shift; 1374 int i; 1375 1376 /* 1377 * Try avoid useless loop iterations for small numbers stored in big 1378 * counters. Example: 48 33 4 ... in 64bit array 1379 */ 1380 max_num = array[0].count; 1381 for (i = 1; i < num; i++) { 1382 buf_num = array[i].count; 1383 if (buf_num > max_num) 1384 max_num = buf_num; 1385 } 1386 1387 buf_num = ilog2(max_num); 1388 bitlen = ALIGN(buf_num, RADIX_BASE * 2); 1389 1390 shift = 0; 1391 while (shift < bitlen) { 1392 memset(counters, 0, sizeof(counters)); 1393 1394 for (i = 0; i < num; i++) { 1395 buf_num = array[i].count; 1396 addr = get4bits(buf_num, shift); 1397 counters[addr]++; 1398 } 1399 1400 for (i = 1; i < COUNTERS_SIZE; i++) 1401 counters[i] += counters[i - 1]; 1402 1403 for (i = num - 1; i >= 0; i--) { 1404 buf_num = array[i].count; 1405 addr = get4bits(buf_num, shift); 1406 counters[addr]--; 1407 new_addr = counters[addr]; 1408 array_buf[new_addr] = array[i]; 1409 } 1410 1411 shift += RADIX_BASE; 1412 1413 /* 1414 * Normal radix expects to move data from a temporary array, to 1415 * the main one. But that requires some CPU time. Avoid that 1416 * by doing another sort iteration to original array instead of 1417 * memcpy() 1418 */ 1419 memset(counters, 0, sizeof(counters)); 1420 1421 for (i = 0; i < num; i ++) { 1422 buf_num = array_buf[i].count; 1423 addr = get4bits(buf_num, shift); 1424 counters[addr]++; 1425 } 1426 1427 for (i = 1; i < COUNTERS_SIZE; i++) 1428 counters[i] += counters[i - 1]; 1429 1430 for (i = num - 1; i >= 0; i--) { 1431 buf_num = array_buf[i].count; 1432 addr = get4bits(buf_num, shift); 1433 counters[addr]--; 1434 new_addr = counters[addr]; 1435 array[new_addr] = array_buf[i]; 1436 } 1437 1438 shift += RADIX_BASE; 1439 } 1440 } 1441 1442 /* 1443 * Size of the core byte set - how many bytes cover 90% of the sample 1444 * 1445 * There are several types of structured binary data that use nearly all byte 1446 * values. The distribution can be uniform and counts in all buckets will be 1447 * nearly the same (eg. encrypted data). Unlikely to be compressible. 1448 * 1449 * Other possibility is normal (Gaussian) distribution, where the data could 1450 * be potentially compressible, but we have to take a few more steps to decide 1451 * how much. 1452 * 1453 * @BYTE_CORE_SET_LOW - main part of byte values repeated frequently, 1454 * compression algo can easy fix that 1455 * @BYTE_CORE_SET_HIGH - data have uniform distribution and with high 1456 * probability is not compressible 1457 */ 1458 #define BYTE_CORE_SET_LOW (64) 1459 #define BYTE_CORE_SET_HIGH (200) 1460 1461 static int byte_core_set_size(struct heuristic_ws *ws) 1462 { 1463 u32 i; 1464 u32 coreset_sum = 0; 1465 const u32 core_set_threshold = ws->sample_size * 90 / 100; 1466 struct bucket_item *bucket = ws->bucket; 1467 1468 /* Sort in reverse order */ 1469 radix_sort(ws->bucket, ws->bucket_b, BUCKET_SIZE); 1470 1471 for (i = 0; i < BYTE_CORE_SET_LOW; i++) 1472 coreset_sum += bucket[i].count; 1473 1474 if (coreset_sum > core_set_threshold) 1475 return i; 1476 1477 for (; i < BYTE_CORE_SET_HIGH && bucket[i].count > 0; i++) { 1478 coreset_sum += bucket[i].count; 1479 if (coreset_sum > core_set_threshold) 1480 break; 1481 } 1482 1483 return i; 1484 } 1485 1486 /* 1487 * Count byte values in buckets. 1488 * This heuristic can detect textual data (configs, xml, json, html, etc). 1489 * Because in most text-like data byte set is restricted to limited number of 1490 * possible characters, and that restriction in most cases makes data easy to 1491 * compress. 1492 * 1493 * @BYTE_SET_THRESHOLD - consider all data within this byte set size: 1494 * less - compressible 1495 * more - need additional analysis 1496 */ 1497 #define BYTE_SET_THRESHOLD (64) 1498 1499 static u32 byte_set_size(const struct heuristic_ws *ws) 1500 { 1501 u32 i; 1502 u32 byte_set_size = 0; 1503 1504 for (i = 0; i < BYTE_SET_THRESHOLD; i++) { 1505 if (ws->bucket[i].count > 0) 1506 byte_set_size++; 1507 } 1508 1509 /* 1510 * Continue collecting count of byte values in buckets. If the byte 1511 * set size is bigger then the threshold, it's pointless to continue, 1512 * the detection technique would fail for this type of data. 1513 */ 1514 for (; i < BUCKET_SIZE; i++) { 1515 if (ws->bucket[i].count > 0) { 1516 byte_set_size++; 1517 if (byte_set_size > BYTE_SET_THRESHOLD) 1518 return byte_set_size; 1519 } 1520 } 1521 1522 return byte_set_size; 1523 } 1524 1525 static bool sample_repeated_patterns(struct heuristic_ws *ws) 1526 { 1527 const u32 half_of_sample = ws->sample_size / 2; 1528 const u8 *data = ws->sample; 1529 1530 return memcmp(&data[0], &data[half_of_sample], half_of_sample) == 0; 1531 } 1532 1533 static void heuristic_collect_sample(struct inode *inode, u64 start, u64 end, 1534 struct heuristic_ws *ws) 1535 { 1536 struct page *page; 1537 pgoff_t index, index_end; 1538 u32 i, curr_sample_pos; 1539 u8 *in_data; 1540 1541 /* 1542 * Compression handles the input data by chunks of 128KiB 1543 * (defined by BTRFS_MAX_UNCOMPRESSED) 1544 * 1545 * We do the same for the heuristic and loop over the whole range. 1546 * 1547 * MAX_SAMPLE_SIZE - calculated under assumption that heuristic will 1548 * process no more than BTRFS_MAX_UNCOMPRESSED at a time. 1549 */ 1550 if (end - start > BTRFS_MAX_UNCOMPRESSED) 1551 end = start + BTRFS_MAX_UNCOMPRESSED; 1552 1553 index = start >> PAGE_SHIFT; 1554 index_end = end >> PAGE_SHIFT; 1555 1556 /* Don't miss unaligned end */ 1557 if (!PAGE_ALIGNED(end)) 1558 index_end++; 1559 1560 curr_sample_pos = 0; 1561 while (index < index_end) { 1562 page = find_get_page(inode->i_mapping, index); 1563 in_data = kmap_local_page(page); 1564 /* Handle case where the start is not aligned to PAGE_SIZE */ 1565 i = start % PAGE_SIZE; 1566 while (i < PAGE_SIZE - SAMPLING_READ_SIZE) { 1567 /* Don't sample any garbage from the last page */ 1568 if (start > end - SAMPLING_READ_SIZE) 1569 break; 1570 memcpy(&ws->sample[curr_sample_pos], &in_data[i], 1571 SAMPLING_READ_SIZE); 1572 i += SAMPLING_INTERVAL; 1573 start += SAMPLING_INTERVAL; 1574 curr_sample_pos += SAMPLING_READ_SIZE; 1575 } 1576 kunmap_local(in_data); 1577 put_page(page); 1578 1579 index++; 1580 } 1581 1582 ws->sample_size = curr_sample_pos; 1583 } 1584 1585 /* 1586 * Compression heuristic. 1587 * 1588 * The following types of analysis can be performed: 1589 * - detect mostly zero data 1590 * - detect data with low "byte set" size (text, etc) 1591 * - detect data with low/high "core byte" set 1592 * 1593 * Return non-zero if the compression should be done, 0 otherwise. 1594 */ 1595 int btrfs_compress_heuristic(struct btrfs_inode *inode, u64 start, u64 end) 1596 { 1597 struct btrfs_fs_info *fs_info = inode->root->fs_info; 1598 struct list_head *ws_list = get_workspace(fs_info, 0, 0); 1599 struct heuristic_ws *ws; 1600 u32 i; 1601 u8 byte; 1602 int ret = 0; 1603 1604 ws = list_entry(ws_list, struct heuristic_ws, list); 1605 1606 heuristic_collect_sample(&inode->vfs_inode, start, end, ws); 1607 1608 if (sample_repeated_patterns(ws)) { 1609 ret = 1; 1610 goto out; 1611 } 1612 1613 memset(ws->bucket, 0, sizeof(*ws->bucket)*BUCKET_SIZE); 1614 1615 for (i = 0; i < ws->sample_size; i++) { 1616 byte = ws->sample[i]; 1617 ws->bucket[byte].count++; 1618 } 1619 1620 i = byte_set_size(ws); 1621 if (i < BYTE_SET_THRESHOLD) { 1622 ret = 2; 1623 goto out; 1624 } 1625 1626 i = byte_core_set_size(ws); 1627 if (i <= BYTE_CORE_SET_LOW) { 1628 ret = 3; 1629 goto out; 1630 } 1631 1632 if (i >= BYTE_CORE_SET_HIGH) { 1633 ret = 0; 1634 goto out; 1635 } 1636 1637 i = shannon_entropy(ws); 1638 if (i <= ENTROPY_LVL_ACEPTABLE) { 1639 ret = 4; 1640 goto out; 1641 } 1642 1643 /* 1644 * For the levels below ENTROPY_LVL_HIGH, additional analysis would be 1645 * needed to give green light to compression. 1646 * 1647 * For now just assume that compression at that level is not worth the 1648 * resources because: 1649 * 1650 * 1. it is possible to defrag the data later 1651 * 1652 * 2. the data would turn out to be hardly compressible, eg. 150 byte 1653 * values, every bucket has counter at level ~54. The heuristic would 1654 * be confused. This can happen when data have some internal repeated 1655 * patterns like "abbacbbc...". This can be detected by analyzing 1656 * pairs of bytes, which is too costly. 1657 */ 1658 if (i < ENTROPY_LVL_HIGH) { 1659 ret = 5; 1660 goto out; 1661 } else { 1662 ret = 0; 1663 goto out; 1664 } 1665 1666 out: 1667 put_workspace(fs_info, 0, ws_list); 1668 return ret; 1669 } 1670 1671 /* 1672 * Convert the compression suffix (eg. after "zlib" starting with ":") to level. 1673 * 1674 * If the resulting level exceeds the algo's supported levels, it will be clamped. 1675 * 1676 * Return <0 if no valid string can be found. 1677 * Return 0 if everything is fine. 1678 */ 1679 int btrfs_compress_str2level(unsigned int type, const char *str, int *level_ret) 1680 { 1681 int level = 0; 1682 int ret; 1683 1684 if (!type) { 1685 *level_ret = btrfs_compress_set_level(type, level); 1686 return 0; 1687 } 1688 1689 if (str[0] == ':') { 1690 ret = kstrtoint(str + 1, 10, &level); 1691 if (ret) 1692 return ret; 1693 } 1694 1695 *level_ret = btrfs_compress_set_level(type, level); 1696 return 0; 1697 } 1698