1 /* 2 * Copyright (C) 2008 Oracle. All rights reserved. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public 6 * License v2 as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public 14 * License along with this program; if not, write to the 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 16 * Boston, MA 021110-1307, USA. 17 */ 18 19 #include <linux/kernel.h> 20 #include <linux/bio.h> 21 #include <linux/buffer_head.h> 22 #include <linux/file.h> 23 #include <linux/fs.h> 24 #include <linux/pagemap.h> 25 #include <linux/highmem.h> 26 #include <linux/time.h> 27 #include <linux/init.h> 28 #include <linux/string.h> 29 #include <linux/backing-dev.h> 30 #include <linux/mpage.h> 31 #include <linux/swap.h> 32 #include <linux/writeback.h> 33 #include <linux/bit_spinlock.h> 34 #include <linux/slab.h> 35 #include "compat.h" 36 #include "ctree.h" 37 #include "disk-io.h" 38 #include "transaction.h" 39 #include "btrfs_inode.h" 40 #include "volumes.h" 41 #include "ordered-data.h" 42 #include "compression.h" 43 #include "extent_io.h" 44 #include "extent_map.h" 45 46 struct compressed_bio { 47 /* number of bios pending for this compressed extent */ 48 atomic_t pending_bios; 49 50 /* the pages with the compressed data on them */ 51 struct page **compressed_pages; 52 53 /* inode that owns this data */ 54 struct inode *inode; 55 56 /* starting offset in the inode for our pages */ 57 u64 start; 58 59 /* number of bytes in the inode we're working on */ 60 unsigned long len; 61 62 /* number of bytes on disk */ 63 unsigned long compressed_len; 64 65 /* the compression algorithm for this bio */ 66 int compress_type; 67 68 /* number of compressed pages in the array */ 69 unsigned long nr_pages; 70 71 /* IO errors */ 72 int errors; 73 int mirror_num; 74 75 /* for reads, this is the bio we are copying the data into */ 76 struct bio *orig_bio; 77 78 /* 79 * the start of a variable length array of checksums only 80 * used by reads 81 */ 82 u32 sums; 83 }; 84 85 static inline int compressed_bio_size(struct btrfs_root *root, 86 unsigned long disk_size) 87 { 88 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy); 89 90 return sizeof(struct compressed_bio) + 91 ((disk_size + root->sectorsize - 1) / root->sectorsize) * 92 csum_size; 93 } 94 95 static struct bio *compressed_bio_alloc(struct block_device *bdev, 96 u64 first_byte, gfp_t gfp_flags) 97 { 98 int nr_vecs; 99 100 nr_vecs = bio_get_nr_vecs(bdev); 101 return btrfs_bio_alloc(bdev, first_byte >> 9, nr_vecs, gfp_flags); 102 } 103 104 static int check_compressed_csum(struct inode *inode, 105 struct compressed_bio *cb, 106 u64 disk_start) 107 { 108 int ret; 109 struct btrfs_root *root = BTRFS_I(inode)->root; 110 struct page *page; 111 unsigned long i; 112 char *kaddr; 113 u32 csum; 114 u32 *cb_sum = &cb->sums; 115 116 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) 117 return 0; 118 119 for (i = 0; i < cb->nr_pages; i++) { 120 page = cb->compressed_pages[i]; 121 csum = ~(u32)0; 122 123 kaddr = kmap_atomic(page); 124 csum = btrfs_csum_data(root, kaddr, csum, PAGE_CACHE_SIZE); 125 btrfs_csum_final(csum, (char *)&csum); 126 kunmap_atomic(kaddr); 127 128 if (csum != *cb_sum) { 129 printk(KERN_INFO "btrfs csum failed ino %llu " 130 "extent %llu csum %u " 131 "wanted %u mirror %d\n", 132 (unsigned long long)btrfs_ino(inode), 133 (unsigned long long)disk_start, 134 csum, *cb_sum, cb->mirror_num); 135 ret = -EIO; 136 goto fail; 137 } 138 cb_sum++; 139 140 } 141 ret = 0; 142 fail: 143 return ret; 144 } 145 146 /* when we finish reading compressed pages from the disk, we 147 * decompress them and then run the bio end_io routines on the 148 * decompressed pages (in the inode address space). 149 * 150 * This allows the checksumming and other IO error handling routines 151 * to work normally 152 * 153 * The compressed pages are freed here, and it must be run 154 * in process context 155 */ 156 static void end_compressed_bio_read(struct bio *bio, int err) 157 { 158 struct compressed_bio *cb = bio->bi_private; 159 struct inode *inode; 160 struct page *page; 161 unsigned long index; 162 int ret; 163 164 if (err) 165 cb->errors = 1; 166 167 /* if there are more bios still pending for this compressed 168 * extent, just exit 169 */ 170 if (!atomic_dec_and_test(&cb->pending_bios)) 171 goto out; 172 173 inode = cb->inode; 174 ret = check_compressed_csum(inode, cb, (u64)bio->bi_sector << 9); 175 if (ret) 176 goto csum_failed; 177 178 /* ok, we're the last bio for this extent, lets start 179 * the decompression. 180 */ 181 ret = btrfs_decompress_biovec(cb->compress_type, 182 cb->compressed_pages, 183 cb->start, 184 cb->orig_bio->bi_io_vec, 185 cb->orig_bio->bi_vcnt, 186 cb->compressed_len); 187 csum_failed: 188 if (ret) 189 cb->errors = 1; 190 191 /* release the compressed pages */ 192 index = 0; 193 for (index = 0; index < cb->nr_pages; index++) { 194 page = cb->compressed_pages[index]; 195 page->mapping = NULL; 196 page_cache_release(page); 197 } 198 199 /* do io completion on the original bio */ 200 if (cb->errors) { 201 bio_io_error(cb->orig_bio); 202 } else { 203 int bio_index = 0; 204 struct bio_vec *bvec = cb->orig_bio->bi_io_vec; 205 206 /* 207 * we have verified the checksum already, set page 208 * checked so the end_io handlers know about it 209 */ 210 while (bio_index < cb->orig_bio->bi_vcnt) { 211 SetPageChecked(bvec->bv_page); 212 bvec++; 213 bio_index++; 214 } 215 bio_endio(cb->orig_bio, 0); 216 } 217 218 /* finally free the cb struct */ 219 kfree(cb->compressed_pages); 220 kfree(cb); 221 out: 222 bio_put(bio); 223 } 224 225 /* 226 * Clear the writeback bits on all of the file 227 * pages for a compressed write 228 */ 229 static noinline void end_compressed_writeback(struct inode *inode, u64 start, 230 unsigned long ram_size) 231 { 232 unsigned long index = start >> PAGE_CACHE_SHIFT; 233 unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT; 234 struct page *pages[16]; 235 unsigned long nr_pages = end_index - index + 1; 236 int i; 237 int ret; 238 239 while (nr_pages > 0) { 240 ret = find_get_pages_contig(inode->i_mapping, index, 241 min_t(unsigned long, 242 nr_pages, ARRAY_SIZE(pages)), pages); 243 if (ret == 0) { 244 nr_pages -= 1; 245 index += 1; 246 continue; 247 } 248 for (i = 0; i < ret; i++) { 249 end_page_writeback(pages[i]); 250 page_cache_release(pages[i]); 251 } 252 nr_pages -= ret; 253 index += ret; 254 } 255 /* the inode may be gone now */ 256 } 257 258 /* 259 * do the cleanup once all the compressed pages hit the disk. 260 * This will clear writeback on the file pages and free the compressed 261 * pages. 262 * 263 * This also calls the writeback end hooks for the file pages so that 264 * metadata and checksums can be updated in the file. 265 */ 266 static void end_compressed_bio_write(struct bio *bio, int err) 267 { 268 struct extent_io_tree *tree; 269 struct compressed_bio *cb = bio->bi_private; 270 struct inode *inode; 271 struct page *page; 272 unsigned long index; 273 274 if (err) 275 cb->errors = 1; 276 277 /* if there are more bios still pending for this compressed 278 * extent, just exit 279 */ 280 if (!atomic_dec_and_test(&cb->pending_bios)) 281 goto out; 282 283 /* ok, we're the last bio for this extent, step one is to 284 * call back into the FS and do all the end_io operations 285 */ 286 inode = cb->inode; 287 tree = &BTRFS_I(inode)->io_tree; 288 cb->compressed_pages[0]->mapping = cb->inode->i_mapping; 289 tree->ops->writepage_end_io_hook(cb->compressed_pages[0], 290 cb->start, 291 cb->start + cb->len - 1, 292 NULL, 1); 293 cb->compressed_pages[0]->mapping = NULL; 294 295 end_compressed_writeback(inode, cb->start, cb->len); 296 /* note, our inode could be gone now */ 297 298 /* 299 * release the compressed pages, these came from alloc_page and 300 * are not attached to the inode at all 301 */ 302 index = 0; 303 for (index = 0; index < cb->nr_pages; index++) { 304 page = cb->compressed_pages[index]; 305 page->mapping = NULL; 306 page_cache_release(page); 307 } 308 309 /* finally free the cb struct */ 310 kfree(cb->compressed_pages); 311 kfree(cb); 312 out: 313 bio_put(bio); 314 } 315 316 /* 317 * worker function to build and submit bios for previously compressed pages. 318 * The corresponding pages in the inode should be marked for writeback 319 * and the compressed pages should have a reference on them for dropping 320 * when the IO is complete. 321 * 322 * This also checksums the file bytes and gets things ready for 323 * the end io hooks. 324 */ 325 int btrfs_submit_compressed_write(struct inode *inode, u64 start, 326 unsigned long len, u64 disk_start, 327 unsigned long compressed_len, 328 struct page **compressed_pages, 329 unsigned long nr_pages) 330 { 331 struct bio *bio = NULL; 332 struct btrfs_root *root = BTRFS_I(inode)->root; 333 struct compressed_bio *cb; 334 unsigned long bytes_left; 335 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 336 int pg_index = 0; 337 struct page *page; 338 u64 first_byte = disk_start; 339 struct block_device *bdev; 340 int ret; 341 int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; 342 343 WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1)); 344 cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); 345 if (!cb) 346 return -ENOMEM; 347 atomic_set(&cb->pending_bios, 0); 348 cb->errors = 0; 349 cb->inode = inode; 350 cb->start = start; 351 cb->len = len; 352 cb->mirror_num = 0; 353 cb->compressed_pages = compressed_pages; 354 cb->compressed_len = compressed_len; 355 cb->orig_bio = NULL; 356 cb->nr_pages = nr_pages; 357 358 bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 359 360 bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); 361 if(!bio) { 362 kfree(cb); 363 return -ENOMEM; 364 } 365 bio->bi_private = cb; 366 bio->bi_end_io = end_compressed_bio_write; 367 atomic_inc(&cb->pending_bios); 368 369 /* create and submit bios for the compressed pages */ 370 bytes_left = compressed_len; 371 for (pg_index = 0; pg_index < cb->nr_pages; pg_index++) { 372 page = compressed_pages[pg_index]; 373 page->mapping = inode->i_mapping; 374 if (bio->bi_size) 375 ret = io_tree->ops->merge_bio_hook(page, 0, 376 PAGE_CACHE_SIZE, 377 bio, 0); 378 else 379 ret = 0; 380 381 page->mapping = NULL; 382 if (ret || bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < 383 PAGE_CACHE_SIZE) { 384 bio_get(bio); 385 386 /* 387 * inc the count before we submit the bio so 388 * we know the end IO handler won't happen before 389 * we inc the count. Otherwise, the cb might get 390 * freed before we're done setting it up 391 */ 392 atomic_inc(&cb->pending_bios); 393 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); 394 BUG_ON(ret); /* -ENOMEM */ 395 396 if (!skip_sum) { 397 ret = btrfs_csum_one_bio(root, inode, bio, 398 start, 1); 399 BUG_ON(ret); /* -ENOMEM */ 400 } 401 402 ret = btrfs_map_bio(root, WRITE, bio, 0, 1); 403 BUG_ON(ret); /* -ENOMEM */ 404 405 bio_put(bio); 406 407 bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); 408 BUG_ON(!bio); 409 bio->bi_private = cb; 410 bio->bi_end_io = end_compressed_bio_write; 411 bio_add_page(bio, page, PAGE_CACHE_SIZE, 0); 412 } 413 if (bytes_left < PAGE_CACHE_SIZE) { 414 printk("bytes left %lu compress len %lu nr %lu\n", 415 bytes_left, cb->compressed_len, cb->nr_pages); 416 } 417 bytes_left -= PAGE_CACHE_SIZE; 418 first_byte += PAGE_CACHE_SIZE; 419 cond_resched(); 420 } 421 bio_get(bio); 422 423 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); 424 BUG_ON(ret); /* -ENOMEM */ 425 426 if (!skip_sum) { 427 ret = btrfs_csum_one_bio(root, inode, bio, start, 1); 428 BUG_ON(ret); /* -ENOMEM */ 429 } 430 431 ret = btrfs_map_bio(root, WRITE, bio, 0, 1); 432 BUG_ON(ret); /* -ENOMEM */ 433 434 bio_put(bio); 435 return 0; 436 } 437 438 static noinline int add_ra_bio_pages(struct inode *inode, 439 u64 compressed_end, 440 struct compressed_bio *cb) 441 { 442 unsigned long end_index; 443 unsigned long pg_index; 444 u64 last_offset; 445 u64 isize = i_size_read(inode); 446 int ret; 447 struct page *page; 448 unsigned long nr_pages = 0; 449 struct extent_map *em; 450 struct address_space *mapping = inode->i_mapping; 451 struct extent_map_tree *em_tree; 452 struct extent_io_tree *tree; 453 u64 end; 454 int misses = 0; 455 456 page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page; 457 last_offset = (page_offset(page) + PAGE_CACHE_SIZE); 458 em_tree = &BTRFS_I(inode)->extent_tree; 459 tree = &BTRFS_I(inode)->io_tree; 460 461 if (isize == 0) 462 return 0; 463 464 end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; 465 466 while (last_offset < compressed_end) { 467 pg_index = last_offset >> PAGE_CACHE_SHIFT; 468 469 if (pg_index > end_index) 470 break; 471 472 rcu_read_lock(); 473 page = radix_tree_lookup(&mapping->page_tree, pg_index); 474 rcu_read_unlock(); 475 if (page) { 476 misses++; 477 if (misses > 4) 478 break; 479 goto next; 480 } 481 482 page = __page_cache_alloc(mapping_gfp_mask(mapping) & 483 ~__GFP_FS); 484 if (!page) 485 break; 486 487 if (add_to_page_cache_lru(page, mapping, pg_index, 488 GFP_NOFS)) { 489 page_cache_release(page); 490 goto next; 491 } 492 493 end = last_offset + PAGE_CACHE_SIZE - 1; 494 /* 495 * at this point, we have a locked page in the page cache 496 * for these bytes in the file. But, we have to make 497 * sure they map to this compressed extent on disk. 498 */ 499 set_page_extent_mapped(page); 500 lock_extent(tree, last_offset, end); 501 read_lock(&em_tree->lock); 502 em = lookup_extent_mapping(em_tree, last_offset, 503 PAGE_CACHE_SIZE); 504 read_unlock(&em_tree->lock); 505 506 if (!em || last_offset < em->start || 507 (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) || 508 (em->block_start >> 9) != cb->orig_bio->bi_sector) { 509 free_extent_map(em); 510 unlock_extent(tree, last_offset, end); 511 unlock_page(page); 512 page_cache_release(page); 513 break; 514 } 515 free_extent_map(em); 516 517 if (page->index == end_index) { 518 char *userpage; 519 size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1); 520 521 if (zero_offset) { 522 int zeros; 523 zeros = PAGE_CACHE_SIZE - zero_offset; 524 userpage = kmap_atomic(page); 525 memset(userpage + zero_offset, 0, zeros); 526 flush_dcache_page(page); 527 kunmap_atomic(userpage); 528 } 529 } 530 531 ret = bio_add_page(cb->orig_bio, page, 532 PAGE_CACHE_SIZE, 0); 533 534 if (ret == PAGE_CACHE_SIZE) { 535 nr_pages++; 536 page_cache_release(page); 537 } else { 538 unlock_extent(tree, last_offset, end); 539 unlock_page(page); 540 page_cache_release(page); 541 break; 542 } 543 next: 544 last_offset += PAGE_CACHE_SIZE; 545 } 546 return 0; 547 } 548 549 /* 550 * for a compressed read, the bio we get passed has all the inode pages 551 * in it. We don't actually do IO on those pages but allocate new ones 552 * to hold the compressed pages on disk. 553 * 554 * bio->bi_sector points to the compressed extent on disk 555 * bio->bi_io_vec points to all of the inode pages 556 * bio->bi_vcnt is a count of pages 557 * 558 * After the compressed pages are read, we copy the bytes into the 559 * bio we were passed and then call the bio end_io calls 560 */ 561 int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio, 562 int mirror_num, unsigned long bio_flags) 563 { 564 struct extent_io_tree *tree; 565 struct extent_map_tree *em_tree; 566 struct compressed_bio *cb; 567 struct btrfs_root *root = BTRFS_I(inode)->root; 568 unsigned long uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE; 569 unsigned long compressed_len; 570 unsigned long nr_pages; 571 unsigned long pg_index; 572 struct page *page; 573 struct block_device *bdev; 574 struct bio *comp_bio; 575 u64 cur_disk_byte = (u64)bio->bi_sector << 9; 576 u64 em_len; 577 u64 em_start; 578 struct extent_map *em; 579 int ret = -ENOMEM; 580 int faili = 0; 581 u32 *sums; 582 583 tree = &BTRFS_I(inode)->io_tree; 584 em_tree = &BTRFS_I(inode)->extent_tree; 585 586 /* we need the actual starting offset of this extent in the file */ 587 read_lock(&em_tree->lock); 588 em = lookup_extent_mapping(em_tree, 589 page_offset(bio->bi_io_vec->bv_page), 590 PAGE_CACHE_SIZE); 591 read_unlock(&em_tree->lock); 592 if (!em) 593 return -EIO; 594 595 compressed_len = em->block_len; 596 cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); 597 if (!cb) 598 goto out; 599 600 atomic_set(&cb->pending_bios, 0); 601 cb->errors = 0; 602 cb->inode = inode; 603 cb->mirror_num = mirror_num; 604 sums = &cb->sums; 605 606 cb->start = em->orig_start; 607 em_len = em->len; 608 em_start = em->start; 609 610 free_extent_map(em); 611 em = NULL; 612 613 cb->len = uncompressed_len; 614 cb->compressed_len = compressed_len; 615 cb->compress_type = extent_compress_type(bio_flags); 616 cb->orig_bio = bio; 617 618 nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) / 619 PAGE_CACHE_SIZE; 620 cb->compressed_pages = kzalloc(sizeof(struct page *) * nr_pages, 621 GFP_NOFS); 622 if (!cb->compressed_pages) 623 goto fail1; 624 625 bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 626 627 for (pg_index = 0; pg_index < nr_pages; pg_index++) { 628 cb->compressed_pages[pg_index] = alloc_page(GFP_NOFS | 629 __GFP_HIGHMEM); 630 if (!cb->compressed_pages[pg_index]) { 631 faili = pg_index - 1; 632 ret = -ENOMEM; 633 goto fail2; 634 } 635 } 636 faili = nr_pages - 1; 637 cb->nr_pages = nr_pages; 638 639 add_ra_bio_pages(inode, em_start + em_len, cb); 640 641 /* include any pages we added in add_ra-bio_pages */ 642 uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE; 643 cb->len = uncompressed_len; 644 645 comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS); 646 if (!comp_bio) 647 goto fail2; 648 comp_bio->bi_private = cb; 649 comp_bio->bi_end_io = end_compressed_bio_read; 650 atomic_inc(&cb->pending_bios); 651 652 for (pg_index = 0; pg_index < nr_pages; pg_index++) { 653 page = cb->compressed_pages[pg_index]; 654 page->mapping = inode->i_mapping; 655 page->index = em_start >> PAGE_CACHE_SHIFT; 656 657 if (comp_bio->bi_size) 658 ret = tree->ops->merge_bio_hook(page, 0, 659 PAGE_CACHE_SIZE, 660 comp_bio, 0); 661 else 662 ret = 0; 663 664 page->mapping = NULL; 665 if (ret || bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) < 666 PAGE_CACHE_SIZE) { 667 bio_get(comp_bio); 668 669 ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0); 670 BUG_ON(ret); /* -ENOMEM */ 671 672 /* 673 * inc the count before we submit the bio so 674 * we know the end IO handler won't happen before 675 * we inc the count. Otherwise, the cb might get 676 * freed before we're done setting it up 677 */ 678 atomic_inc(&cb->pending_bios); 679 680 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { 681 ret = btrfs_lookup_bio_sums(root, inode, 682 comp_bio, sums); 683 BUG_ON(ret); /* -ENOMEM */ 684 } 685 sums += (comp_bio->bi_size + root->sectorsize - 1) / 686 root->sectorsize; 687 688 ret = btrfs_map_bio(root, READ, comp_bio, 689 mirror_num, 0); 690 if (ret) 691 bio_endio(comp_bio, ret); 692 693 bio_put(comp_bio); 694 695 comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, 696 GFP_NOFS); 697 BUG_ON(!comp_bio); 698 comp_bio->bi_private = cb; 699 comp_bio->bi_end_io = end_compressed_bio_read; 700 701 bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0); 702 } 703 cur_disk_byte += PAGE_CACHE_SIZE; 704 } 705 bio_get(comp_bio); 706 707 ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0); 708 BUG_ON(ret); /* -ENOMEM */ 709 710 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { 711 ret = btrfs_lookup_bio_sums(root, inode, comp_bio, sums); 712 BUG_ON(ret); /* -ENOMEM */ 713 } 714 715 ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0); 716 if (ret) 717 bio_endio(comp_bio, ret); 718 719 bio_put(comp_bio); 720 return 0; 721 722 fail2: 723 while (faili >= 0) { 724 __free_page(cb->compressed_pages[faili]); 725 faili--; 726 } 727 728 kfree(cb->compressed_pages); 729 fail1: 730 kfree(cb); 731 out: 732 free_extent_map(em); 733 return ret; 734 } 735 736 static struct list_head comp_idle_workspace[BTRFS_COMPRESS_TYPES]; 737 static spinlock_t comp_workspace_lock[BTRFS_COMPRESS_TYPES]; 738 static int comp_num_workspace[BTRFS_COMPRESS_TYPES]; 739 static atomic_t comp_alloc_workspace[BTRFS_COMPRESS_TYPES]; 740 static wait_queue_head_t comp_workspace_wait[BTRFS_COMPRESS_TYPES]; 741 742 struct btrfs_compress_op *btrfs_compress_op[] = { 743 &btrfs_zlib_compress, 744 &btrfs_lzo_compress, 745 }; 746 747 void __init btrfs_init_compress(void) 748 { 749 int i; 750 751 for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { 752 INIT_LIST_HEAD(&comp_idle_workspace[i]); 753 spin_lock_init(&comp_workspace_lock[i]); 754 atomic_set(&comp_alloc_workspace[i], 0); 755 init_waitqueue_head(&comp_workspace_wait[i]); 756 } 757 } 758 759 /* 760 * this finds an available workspace or allocates a new one 761 * ERR_PTR is returned if things go bad. 762 */ 763 static struct list_head *find_workspace(int type) 764 { 765 struct list_head *workspace; 766 int cpus = num_online_cpus(); 767 int idx = type - 1; 768 769 struct list_head *idle_workspace = &comp_idle_workspace[idx]; 770 spinlock_t *workspace_lock = &comp_workspace_lock[idx]; 771 atomic_t *alloc_workspace = &comp_alloc_workspace[idx]; 772 wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx]; 773 int *num_workspace = &comp_num_workspace[idx]; 774 again: 775 spin_lock(workspace_lock); 776 if (!list_empty(idle_workspace)) { 777 workspace = idle_workspace->next; 778 list_del(workspace); 779 (*num_workspace)--; 780 spin_unlock(workspace_lock); 781 return workspace; 782 783 } 784 if (atomic_read(alloc_workspace) > cpus) { 785 DEFINE_WAIT(wait); 786 787 spin_unlock(workspace_lock); 788 prepare_to_wait(workspace_wait, &wait, TASK_UNINTERRUPTIBLE); 789 if (atomic_read(alloc_workspace) > cpus && !*num_workspace) 790 schedule(); 791 finish_wait(workspace_wait, &wait); 792 goto again; 793 } 794 atomic_inc(alloc_workspace); 795 spin_unlock(workspace_lock); 796 797 workspace = btrfs_compress_op[idx]->alloc_workspace(); 798 if (IS_ERR(workspace)) { 799 atomic_dec(alloc_workspace); 800 wake_up(workspace_wait); 801 } 802 return workspace; 803 } 804 805 /* 806 * put a workspace struct back on the list or free it if we have enough 807 * idle ones sitting around 808 */ 809 static void free_workspace(int type, struct list_head *workspace) 810 { 811 int idx = type - 1; 812 struct list_head *idle_workspace = &comp_idle_workspace[idx]; 813 spinlock_t *workspace_lock = &comp_workspace_lock[idx]; 814 atomic_t *alloc_workspace = &comp_alloc_workspace[idx]; 815 wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx]; 816 int *num_workspace = &comp_num_workspace[idx]; 817 818 spin_lock(workspace_lock); 819 if (*num_workspace < num_online_cpus()) { 820 list_add_tail(workspace, idle_workspace); 821 (*num_workspace)++; 822 spin_unlock(workspace_lock); 823 goto wake; 824 } 825 spin_unlock(workspace_lock); 826 827 btrfs_compress_op[idx]->free_workspace(workspace); 828 atomic_dec(alloc_workspace); 829 wake: 830 smp_mb(); 831 if (waitqueue_active(workspace_wait)) 832 wake_up(workspace_wait); 833 } 834 835 /* 836 * cleanup function for module exit 837 */ 838 static void free_workspaces(void) 839 { 840 struct list_head *workspace; 841 int i; 842 843 for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { 844 while (!list_empty(&comp_idle_workspace[i])) { 845 workspace = comp_idle_workspace[i].next; 846 list_del(workspace); 847 btrfs_compress_op[i]->free_workspace(workspace); 848 atomic_dec(&comp_alloc_workspace[i]); 849 } 850 } 851 } 852 853 /* 854 * given an address space and start/len, compress the bytes. 855 * 856 * pages are allocated to hold the compressed result and stored 857 * in 'pages' 858 * 859 * out_pages is used to return the number of pages allocated. There 860 * may be pages allocated even if we return an error 861 * 862 * total_in is used to return the number of bytes actually read. It 863 * may be smaller then len if we had to exit early because we 864 * ran out of room in the pages array or because we cross the 865 * max_out threshold. 866 * 867 * total_out is used to return the total number of compressed bytes 868 * 869 * max_out tells us the max number of bytes that we're allowed to 870 * stuff into pages 871 */ 872 int btrfs_compress_pages(int type, struct address_space *mapping, 873 u64 start, unsigned long len, 874 struct page **pages, 875 unsigned long nr_dest_pages, 876 unsigned long *out_pages, 877 unsigned long *total_in, 878 unsigned long *total_out, 879 unsigned long max_out) 880 { 881 struct list_head *workspace; 882 int ret; 883 884 workspace = find_workspace(type); 885 if (IS_ERR(workspace)) 886 return -1; 887 888 ret = btrfs_compress_op[type-1]->compress_pages(workspace, mapping, 889 start, len, pages, 890 nr_dest_pages, out_pages, 891 total_in, total_out, 892 max_out); 893 free_workspace(type, workspace); 894 return ret; 895 } 896 897 /* 898 * pages_in is an array of pages with compressed data. 899 * 900 * disk_start is the starting logical offset of this array in the file 901 * 902 * bvec is a bio_vec of pages from the file that we want to decompress into 903 * 904 * vcnt is the count of pages in the biovec 905 * 906 * srclen is the number of bytes in pages_in 907 * 908 * The basic idea is that we have a bio that was created by readpages. 909 * The pages in the bio are for the uncompressed data, and they may not 910 * be contiguous. They all correspond to the range of bytes covered by 911 * the compressed extent. 912 */ 913 int btrfs_decompress_biovec(int type, struct page **pages_in, u64 disk_start, 914 struct bio_vec *bvec, int vcnt, size_t srclen) 915 { 916 struct list_head *workspace; 917 int ret; 918 919 workspace = find_workspace(type); 920 if (IS_ERR(workspace)) 921 return -ENOMEM; 922 923 ret = btrfs_compress_op[type-1]->decompress_biovec(workspace, pages_in, 924 disk_start, 925 bvec, vcnt, srclen); 926 free_workspace(type, workspace); 927 return ret; 928 } 929 930 /* 931 * a less complex decompression routine. Our compressed data fits in a 932 * single page, and we want to read a single page out of it. 933 * start_byte tells us the offset into the compressed data we're interested in 934 */ 935 int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page, 936 unsigned long start_byte, size_t srclen, size_t destlen) 937 { 938 struct list_head *workspace; 939 int ret; 940 941 workspace = find_workspace(type); 942 if (IS_ERR(workspace)) 943 return -ENOMEM; 944 945 ret = btrfs_compress_op[type-1]->decompress(workspace, data_in, 946 dest_page, start_byte, 947 srclen, destlen); 948 949 free_workspace(type, workspace); 950 return ret; 951 } 952 953 void btrfs_exit_compress(void) 954 { 955 free_workspaces(); 956 } 957 958 /* 959 * Copy uncompressed data from working buffer to pages. 960 * 961 * buf_start is the byte offset we're of the start of our workspace buffer. 962 * 963 * total_out is the last byte of the buffer 964 */ 965 int btrfs_decompress_buf2page(char *buf, unsigned long buf_start, 966 unsigned long total_out, u64 disk_start, 967 struct bio_vec *bvec, int vcnt, 968 unsigned long *pg_index, 969 unsigned long *pg_offset) 970 { 971 unsigned long buf_offset; 972 unsigned long current_buf_start; 973 unsigned long start_byte; 974 unsigned long working_bytes = total_out - buf_start; 975 unsigned long bytes; 976 char *kaddr; 977 struct page *page_out = bvec[*pg_index].bv_page; 978 979 /* 980 * start byte is the first byte of the page we're currently 981 * copying into relative to the start of the compressed data. 982 */ 983 start_byte = page_offset(page_out) - disk_start; 984 985 /* we haven't yet hit data corresponding to this page */ 986 if (total_out <= start_byte) 987 return 1; 988 989 /* 990 * the start of the data we care about is offset into 991 * the middle of our working buffer 992 */ 993 if (total_out > start_byte && buf_start < start_byte) { 994 buf_offset = start_byte - buf_start; 995 working_bytes -= buf_offset; 996 } else { 997 buf_offset = 0; 998 } 999 current_buf_start = buf_start; 1000 1001 /* copy bytes from the working buffer into the pages */ 1002 while (working_bytes > 0) { 1003 bytes = min(PAGE_CACHE_SIZE - *pg_offset, 1004 PAGE_CACHE_SIZE - buf_offset); 1005 bytes = min(bytes, working_bytes); 1006 kaddr = kmap_atomic(page_out); 1007 memcpy(kaddr + *pg_offset, buf + buf_offset, bytes); 1008 kunmap_atomic(kaddr); 1009 flush_dcache_page(page_out); 1010 1011 *pg_offset += bytes; 1012 buf_offset += bytes; 1013 working_bytes -= bytes; 1014 current_buf_start += bytes; 1015 1016 /* check if we need to pick another page */ 1017 if (*pg_offset == PAGE_CACHE_SIZE) { 1018 (*pg_index)++; 1019 if (*pg_index >= vcnt) 1020 return 0; 1021 1022 page_out = bvec[*pg_index].bv_page; 1023 *pg_offset = 0; 1024 start_byte = page_offset(page_out) - disk_start; 1025 1026 /* 1027 * make sure our new page is covered by this 1028 * working buffer 1029 */ 1030 if (total_out <= start_byte) 1031 return 1; 1032 1033 /* 1034 * the next page in the biovec might not be adjacent 1035 * to the last page, but it might still be found 1036 * inside this working buffer. bump our offset pointer 1037 */ 1038 if (total_out > start_byte && 1039 current_buf_start < start_byte) { 1040 buf_offset = start_byte - buf_start; 1041 working_bytes = total_out - start_byte; 1042 current_buf_start = buf_start + buf_offset; 1043 } 1044 } 1045 } 1046 1047 return 1; 1048 } 1049