fs/btrfs/compression.c

   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 "ctree.h"
  36 #include "disk-io.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "volumes.h"
  40 #include "ordered-data.h"
  41 #include "compression.h"
  42 #include "extent_io.h"
  43 #include "extent_map.h"
  44
  45 struct compressed_bio {
  46         /* number of bios pending for this compressed extent */
  47         atomic_t pending_bios;
  48
  49         /* the pages with the compressed data on them */
  50         struct page **compressed_pages;
  51
  52         /* inode that owns this data */
  53         struct inode *inode;
  54
  55         /* starting offset in the inode for our pages */
  56         u64 start;
  57
  58         /* number of bytes in the inode we're working on */
  59         unsigned long len;
  60
  61         /* number of bytes on disk */
  62         unsigned long compressed_len;
  63
  64         /* the compression algorithm for this bio */
  65         int compress_type;
  66
  67         /* number of compressed pages in the array */
  68         unsigned long nr_pages;
  69
  70         /* IO errors */
  71         int errors;
  72         int mirror_num;
  73
  74         /* for reads, this is the bio we are copying the data into */
  75         struct bio *orig_bio;
  76
  77         /*
  78          * the start of a variable length array of checksums only
  79          * used by reads
  80          */
  81         u32 sums;
  82 };
  83
  84 static int btrfs_decompress_biovec(int type, struct page **pages_in,
  85                                    u64 disk_start, struct bio_vec *bvec,
  86                                    int vcnt, size_t srclen);
  87
  88 static inline int compressed_bio_size(struct btrfs_root *root,
  89                                       unsigned long disk_size)
  90 {
  91         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
  92
  93         return sizeof(struct compressed_bio) +
  94                 ((disk_size + root->sectorsize - 1) / root->sectorsize) *
  95                 csum_size;
  96 }
  97
  98 static struct bio *compressed_bio_alloc(struct block_device *bdev,
  99                                         u64 first_byte, gfp_t gfp_flags)
 100 {
 101         int nr_vecs;
 102
 103         nr_vecs = bio_get_nr_vecs(bdev);
 104         return btrfs_bio_alloc(bdev, first_byte >> 9, nr_vecs, gfp_flags);
 105 }
 106
 107 static int check_compressed_csum(struct inode *inode,
 108                                  struct compressed_bio *cb,
 109                                  u64 disk_start)
 110 {
 111         int ret;
 112         struct page *page;
 113         unsigned long i;
 114         char *kaddr;
 115         u32 csum;
 116         u32 *cb_sum = &cb->sums;
 117
 118         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
 119                 return 0;
 120
 121         for (i = 0; i < cb->nr_pages; i++) {
 122                 page = cb->compressed_pages[i];
 123                 csum = ~(u32)0;
 124
 125                 kaddr = kmap_atomic(page);
 126                 csum = btrfs_csum_data(kaddr, csum, PAGE_CACHE_SIZE);
 127                 btrfs_csum_final(csum, (char *)&csum);
 128                 kunmap_atomic(kaddr);
 129
 130                 if (csum != *cb_sum) {
 131                         btrfs_info(BTRFS_I(inode)->root->fs_info,
 132                            "csum failed ino %llu extent %llu csum %u wanted %u mirror %d",
 133                            btrfs_ino(inode), disk_start, csum, *cb_sum,
 134                            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,
 175                                     (u64)bio->bi_iter.bi_sector << 9);
 176         if (ret)
 177                 goto csum_failed;
 178
 179         /* ok, we're the last bio for this extent, lets start
 180          * the decompression.
 181          */
 182         ret = btrfs_decompress_biovec(cb->compress_type,
 183                                       cb->compressed_pages,
 184                                       cb->start,
 185                                       cb->orig_bio->bi_io_vec,
 186                                       cb->orig_bio->bi_vcnt,
 187                                       cb->compressed_len);
 188 csum_failed:
 189         if (ret)
 190                 cb->errors = 1;
 191
 192         /* release the compressed pages */
 193         index = 0;
 194         for (index = 0; index < cb->nr_pages; index++) {
 195                 page = cb->compressed_pages[index];
 196                 page->mapping = NULL;
 197                 page_cache_release(page);
 198         }
 199
 200         /* do io completion on the original bio */
 201         if (cb->errors) {
 202                 bio_io_error(cb->orig_bio);
 203         } else {
 204                 int i;
 205                 struct bio_vec *bvec;
 206
 207                 /*
 208                  * we have verified the checksum already, set page
 209                  * checked so the end_io handlers know about it
 210                  */
 211                 bio_for_each_segment_all(bvec, cb->orig_bio, i)
 212                         SetPageChecked(bvec->bv_page);
 213
 214                 bio_endio(cb->orig_bio, 0);
 215         }
 216
 217         /* finally free the cb struct */
 218         kfree(cb->compressed_pages);
 219         kfree(cb);
 220 out:
 221         bio_put(bio);
 222 }
 223
 224 /*
 225  * Clear the writeback bits on all of the file
 226  * pages for a compressed write
 227  */
 228 static noinline void end_compressed_writeback(struct inode *inode, u64 start,
 229                                               unsigned long ram_size)
 230 {
 231         unsigned long index = start >> PAGE_CACHE_SHIFT;
 232         unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT;
 233         struct page *pages[16];
 234         unsigned long nr_pages = end_index - index + 1;
 235         int i;
 236         int ret;
 237
 238         while (nr_pages > 0) {
 239                 ret = find_get_pages_contig(inode->i_mapping, index,
 240                                      min_t(unsigned long,
 241                                      nr_pages, ARRAY_SIZE(pages)), pages);
 242                 if (ret == 0) {
 243                         nr_pages -= 1;
 244                         index += 1;
 245                         continue;
 246                 }
 247                 for (i = 0; i < ret; i++) {
 248                         end_page_writeback(pages[i]);
 249                         page_cache_release(pages[i]);
 250                 }
 251                 nr_pages -= ret;
 252                 index += ret;
 253         }
 254         /* the inode may be gone now */
 255 }
 256
 257 /*
 258  * do the cleanup once all the compressed pages hit the disk.
 259  * This will clear writeback on the file pages and free the compressed
 260  * pages.
 261  *
 262  * This also calls the writeback end hooks for the file pages so that
 263  * metadata and checksums can be updated in the file.
 264  */
 265 static void end_compressed_bio_write(struct bio *bio, int err)
 266 {
 267         struct extent_io_tree *tree;
 268         struct compressed_bio *cb = bio->bi_private;
 269         struct inode *inode;
 270         struct page *page;
 271         unsigned long index;
 272
 273         if (err)
 274                 cb->errors = 1;
 275
 276         /* if there are more bios still pending for this compressed
 277          * extent, just exit
 278          */
 279         if (!atomic_dec_and_test(&cb->pending_bios))
 280                 goto out;
 281
 282         /* ok, we're the last bio for this extent, step one is to
 283          * call back into the FS and do all the end_io operations
 284          */
 285         inode = cb->inode;
 286         tree = &BTRFS_I(inode)->io_tree;
 287         cb->compressed_pages[0]->mapping = cb->inode->i_mapping;
 288         tree->ops->writepage_end_io_hook(cb->compressed_pages[0],
 289                                          cb->start,
 290                                          cb->start + cb->len - 1,
 291                                          NULL, 1);
 292         cb->compressed_pages[0]->mapping = NULL;
 293
 294         end_compressed_writeback(inode, cb->start, cb->len);
 295         /* note, our inode could be gone now */
 296
 297         /*
 298          * release the compressed pages, these came from alloc_page and
 299          * are not attached to the inode at all
 300          */
 301         index = 0;
 302         for (index = 0; index < cb->nr_pages; index++) {
 303                 page = cb->compressed_pages[index];
 304                 page->mapping = NULL;
 305                 page_cache_release(page);
 306         }
 307
 308         /* finally free the cb struct */
 309         kfree(cb->compressed_pages);
 310         kfree(cb);
 311 out:
 312         bio_put(bio);
 313 }
 314
 315 /*
 316  * worker function to build and submit bios for previously compressed pages.
 317  * The corresponding pages in the inode should be marked for writeback
 318  * and the compressed pages should have a reference on them for dropping
 319  * when the IO is complete.
 320  *
 321  * This also checksums the file bytes and gets things ready for
 322  * the end io hooks.
 323  */
 324 int btrfs_submit_compressed_write(struct inode *inode, u64 start,
 325                                  unsigned long len, u64 disk_start,
 326                                  unsigned long compressed_len,
 327                                  struct page **compressed_pages,
 328                                  unsigned long nr_pages)
 329 {
 330         struct bio *bio = NULL;
 331         struct btrfs_root *root = BTRFS_I(inode)->root;
 332         struct compressed_bio *cb;
 333         unsigned long bytes_left;
 334         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 335         int pg_index = 0;
 336         struct page *page;
 337         u64 first_byte = disk_start;
 338         struct block_device *bdev;
 339         int ret;
 340         int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
 341
 342         WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1));
 343         cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
 344         if (!cb)
 345                 return -ENOMEM;
 346         atomic_set(&cb->pending_bios, 0);
 347         cb->errors = 0;
 348         cb->inode = inode;
 349         cb->start = start;
 350         cb->len = len;
 351         cb->mirror_num = 0;
 352         cb->compressed_pages = compressed_pages;
 353         cb->compressed_len = compressed_len;
 354         cb->orig_bio = NULL;
 355         cb->nr_pages = nr_pages;
 356
 357         bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 358
 359         bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
 360         if (!bio) {
 361                 kfree(cb);
 362                 return -ENOMEM;
 363         }
 364         bio->bi_private = cb;
 365         bio->bi_end_io = end_compressed_bio_write;
 366         atomic_inc(&cb->pending_bios);
 367
 368         /* create and submit bios for the compressed pages */
 369         bytes_left = compressed_len;
 370         for (pg_index = 0; pg_index < cb->nr_pages; pg_index++) {
 371                 page = compressed_pages[pg_index];
 372                 page->mapping = inode->i_mapping;
 373                 if (bio->bi_iter.bi_size)
 374                         ret = io_tree->ops->merge_bio_hook(WRITE, page, 0,
 375                                                            PAGE_CACHE_SIZE,
 376                                                            bio, 0);
 377                 else
 378                         ret = 0;
 379
 380                 page->mapping = NULL;
 381                 if (ret || bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) <
 382                     PAGE_CACHE_SIZE) {
 383                         bio_get(bio);
 384
 385                         /*
 386                          * inc the count before we submit the bio so
 387                          * we know the end IO handler won't happen before
 388                          * we inc the count.  Otherwise, the cb might get
 389                          * freed before we're done setting it up
 390                          */
 391                         atomic_inc(&cb->pending_bios);
 392                         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
 393                         BUG_ON(ret); /* -ENOMEM */
 394
 395                         if (!skip_sum) {
 396                                 ret = btrfs_csum_one_bio(root, inode, bio,
 397                                                          start, 1);
 398                                 BUG_ON(ret); /* -ENOMEM */
 399                         }
 400
 401                         ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
 402                         BUG_ON(ret); /* -ENOMEM */
 403
 404                         bio_put(bio);
 405
 406                         bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
 407                         BUG_ON(!bio);
 408                         bio->bi_private = cb;
 409                         bio->bi_end_io = end_compressed_bio_write;
 410                         bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
 411                 }
 412                 if (bytes_left < PAGE_CACHE_SIZE) {
 413                         btrfs_info(BTRFS_I(inode)->root->fs_info,
 414                                         "bytes left %lu compress len %lu nr %lu",
 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_iter.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_iter.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_iter.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         /* In the parent-locked case, we only locked the range we are
 640          * interested in.  In all other cases, we can opportunistically
 641          * cache decompressed data that goes beyond the requested range. */
 642         if (!(bio_flags & EXTENT_BIO_PARENT_LOCKED))
 643                 add_ra_bio_pages(inode, em_start + em_len, cb);
 644
 645         /* include any pages we added in add_ra-bio_pages */
 646         uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
 647         cb->len = uncompressed_len;
 648
 649         comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS);
 650         if (!comp_bio)
 651                 goto fail2;
 652         comp_bio->bi_private = cb;
 653         comp_bio->bi_end_io = end_compressed_bio_read;
 654         atomic_inc(&cb->pending_bios);
 655
 656         for (pg_index = 0; pg_index < nr_pages; pg_index++) {
 657                 page = cb->compressed_pages[pg_index];
 658                 page->mapping = inode->i_mapping;
 659                 page->index = em_start >> PAGE_CACHE_SHIFT;
 660
 661                 if (comp_bio->bi_iter.bi_size)
 662                         ret = tree->ops->merge_bio_hook(READ, page, 0,
 663                                                         PAGE_CACHE_SIZE,
 664                                                         comp_bio, 0);
 665                 else
 666                         ret = 0;
 667
 668                 page->mapping = NULL;
 669                 if (ret || bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) <
 670                     PAGE_CACHE_SIZE) {
 671                         bio_get(comp_bio);
 672
 673                         ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
 674                         BUG_ON(ret); /* -ENOMEM */
 675
 676                         /*
 677                          * inc the count before we submit the bio so
 678                          * we know the end IO handler won't happen before
 679                          * we inc the count.  Otherwise, the cb might get
 680                          * freed before we're done setting it up
 681                          */
 682                         atomic_inc(&cb->pending_bios);
 683
 684                         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
 685                                 ret = btrfs_lookup_bio_sums(root, inode,
 686                                                         comp_bio, sums);
 687                                 BUG_ON(ret); /* -ENOMEM */
 688                         }
 689                         sums += (comp_bio->bi_iter.bi_size +
 690                                  root->sectorsize - 1) / root->sectorsize;
 691
 692                         ret = btrfs_map_bio(root, READ, comp_bio,
 693                                             mirror_num, 0);
 694                         if (ret)
 695                                 bio_endio(comp_bio, ret);
 696
 697                         bio_put(comp_bio);
 698
 699                         comp_bio = compressed_bio_alloc(bdev, cur_disk_byte,
 700                                                         GFP_NOFS);
 701                         BUG_ON(!comp_bio);
 702                         comp_bio->bi_private = cb;
 703                         comp_bio->bi_end_io = end_compressed_bio_read;
 704
 705                         bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0);
 706                 }
 707                 cur_disk_byte += PAGE_CACHE_SIZE;
 708         }
 709         bio_get(comp_bio);
 710
 711         ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
 712         BUG_ON(ret); /* -ENOMEM */
 713
 714         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
 715                 ret = btrfs_lookup_bio_sums(root, inode, comp_bio, sums);
 716                 BUG_ON(ret); /* -ENOMEM */
 717         }
 718
 719         ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0);
 720         if (ret)
 721                 bio_endio(comp_bio, ret);
 722
 723         bio_put(comp_bio);
 724         return 0;
 725
 726 fail2:
 727         while (faili >= 0) {
 728                 __free_page(cb->compressed_pages[faili]);
 729                 faili--;
 730         }
 731
 732         kfree(cb->compressed_pages);
 733 fail1:
 734         kfree(cb);
 735 out:
 736         free_extent_map(em);
 737         return ret;
 738 }
 739
 740 static struct list_head comp_idle_workspace[BTRFS_COMPRESS_TYPES];
 741 static spinlock_t comp_workspace_lock[BTRFS_COMPRESS_TYPES];
 742 static int comp_num_workspace[BTRFS_COMPRESS_TYPES];
 743 static atomic_t comp_alloc_workspace[BTRFS_COMPRESS_TYPES];
 744 static wait_queue_head_t comp_workspace_wait[BTRFS_COMPRESS_TYPES];
 745
 746 static struct btrfs_compress_op *btrfs_compress_op[] = {
 747         &btrfs_zlib_compress,
 748         &btrfs_lzo_compress,
 749 };
 750
 751 void __init btrfs_init_compress(void)
 752 {
 753         int i;
 754
 755         for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) {
 756                 INIT_LIST_HEAD(&comp_idle_workspace[i]);
 757                 spin_lock_init(&comp_workspace_lock[i]);
 758                 atomic_set(&comp_alloc_workspace[i], 0);
 759                 init_waitqueue_head(&comp_workspace_wait[i]);
 760         }
 761 }
 762
 763 /*
 764  * this finds an available workspace or allocates a new one
 765  * ERR_PTR is returned if things go bad.
 766  */
 767 static struct list_head *find_workspace(int type)
 768 {
 769         struct list_head *workspace;
 770         int cpus = num_online_cpus();
 771         int idx = type - 1;
 772
 773         struct list_head *idle_workspace        = &comp_idle_workspace[idx];
 774         spinlock_t *workspace_lock              = &comp_workspace_lock[idx];
 775         atomic_t *alloc_workspace               = &comp_alloc_workspace[idx];
 776         wait_queue_head_t *workspace_wait       = &comp_workspace_wait[idx];
 777         int *num_workspace                      = &comp_num_workspace[idx];
 778 again:
 779         spin_lock(workspace_lock);
 780         if (!list_empty(idle_workspace)) {
 781                 workspace = idle_workspace->next;
 782                 list_del(workspace);
 783                 (*num_workspace)--;
 784                 spin_unlock(workspace_lock);
 785                 return workspace;
 786
 787         }
 788         if (atomic_read(alloc_workspace) > cpus) {
 789                 DEFINE_WAIT(wait);
 790
 791                 spin_unlock(workspace_lock);
 792                 prepare_to_wait(workspace_wait, &wait, TASK_UNINTERRUPTIBLE);
 793                 if (atomic_read(alloc_workspace) > cpus && !*num_workspace)
 794                         schedule();
 795                 finish_wait(workspace_wait, &wait);
 796                 goto again;
 797         }
 798         atomic_inc(alloc_workspace);
 799         spin_unlock(workspace_lock);
 800
 801         workspace = btrfs_compress_op[idx]->alloc_workspace();
 802         if (IS_ERR(workspace)) {
 803                 atomic_dec(alloc_workspace);
 804                 wake_up(workspace_wait);
 805         }
 806         return workspace;
 807 }
 808
 809 /*
 810  * put a workspace struct back on the list or free it if we have enough
 811  * idle ones sitting around
 812  */
 813 static void free_workspace(int type, struct list_head *workspace)
 814 {
 815         int idx = type - 1;
 816         struct list_head *idle_workspace        = &comp_idle_workspace[idx];
 817         spinlock_t *workspace_lock              = &comp_workspace_lock[idx];
 818         atomic_t *alloc_workspace               = &comp_alloc_workspace[idx];
 819         wait_queue_head_t *workspace_wait       = &comp_workspace_wait[idx];
 820         int *num_workspace                      = &comp_num_workspace[idx];
 821
 822         spin_lock(workspace_lock);
 823         if (*num_workspace < num_online_cpus()) {
 824                 list_add_tail(workspace, idle_workspace);
 825                 (*num_workspace)++;
 826                 spin_unlock(workspace_lock);
 827                 goto wake;
 828         }
 829         spin_unlock(workspace_lock);
 830
 831         btrfs_compress_op[idx]->free_workspace(workspace);
 832         atomic_dec(alloc_workspace);
 833 wake:
 834         smp_mb();
 835         if (waitqueue_active(workspace_wait))
 836                 wake_up(workspace_wait);
 837 }
 838
 839 /*
 840  * cleanup function for module exit
 841  */
 842 static void free_workspaces(void)
 843 {
 844         struct list_head *workspace;
 845         int i;
 846
 847         for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) {
 848                 while (!list_empty(&comp_idle_workspace[i])) {
 849                         workspace = comp_idle_workspace[i].next;
 850                         list_del(workspace);
 851                         btrfs_compress_op[i]->free_workspace(workspace);
 852                         atomic_dec(&comp_alloc_workspace[i]);
 853                 }
 854         }
 855 }
 856
 857 /*
 858  * given an address space and start/len, compress the bytes.
 859  *
 860  * pages are allocated to hold the compressed result and stored
 861  * in 'pages'
 862  *
 863  * out_pages is used to return the number of pages allocated.  There
 864  * may be pages allocated even if we return an error
 865  *
 866  * total_in is used to return the number of bytes actually read.  It
 867  * may be smaller then len if we had to exit early because we
 868  * ran out of room in the pages array or because we cross the
 869  * max_out threshold.
 870  *
 871  * total_out is used to return the total number of compressed bytes
 872  *
 873  * max_out tells us the max number of bytes that we're allowed to
 874  * stuff into pages
 875  */
 876 int btrfs_compress_pages(int type, struct address_space *mapping,
 877                          u64 start, unsigned long len,
 878                          struct page **pages,
 879                          unsigned long nr_dest_pages,
 880                          unsigned long *out_pages,
 881                          unsigned long *total_in,
 882                          unsigned long *total_out,
 883                          unsigned long max_out)
 884 {
 885         struct list_head *workspace;
 886         int ret;
 887
 888         workspace = find_workspace(type);
 889         if (IS_ERR(workspace))
 890                 return -1;
 891
 892         ret = btrfs_compress_op[type-1]->compress_pages(workspace, mapping,
 893                                                       start, len, pages,
 894                                                       nr_dest_pages, out_pages,
 895                                                       total_in, total_out,
 896                                                       max_out);
 897         free_workspace(type, workspace);
 898         return ret;
 899 }
 900
 901 /*
 902  * pages_in is an array of pages with compressed data.
 903  *
 904  * disk_start is the starting logical offset of this array in the file
 905  *
 906  * bvec is a bio_vec of pages from the file that we want to decompress into
 907  *
 908  * vcnt is the count of pages in the biovec
 909  *
 910  * srclen is the number of bytes in pages_in
 911  *
 912  * The basic idea is that we have a bio that was created by readpages.
 913  * The pages in the bio are for the uncompressed data, and they may not
 914  * be contiguous.  They all correspond to the range of bytes covered by
 915  * the compressed extent.
 916  */
 917 static int btrfs_decompress_biovec(int type, struct page **pages_in,
 918                                    u64 disk_start, struct bio_vec *bvec,
 919                                    int vcnt, size_t srclen)
 920 {
 921         struct list_head *workspace;
 922         int ret;
 923
 924         workspace = find_workspace(type);
 925         if (IS_ERR(workspace))
 926                 return -ENOMEM;
 927
 928         ret = btrfs_compress_op[type-1]->decompress_biovec(workspace, pages_in,
 929                                                          disk_start,
 930                                                          bvec, vcnt, srclen);
 931         free_workspace(type, workspace);
 932         return ret;
 933 }
 934
 935 /*
 936  * a less complex decompression routine.  Our compressed data fits in a
 937  * single page, and we want to read a single page out of it.
 938  * start_byte tells us the offset into the compressed data we're interested in
 939  */
 940 int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page,
 941                      unsigned long start_byte, size_t srclen, size_t destlen)
 942 {
 943         struct list_head *workspace;
 944         int ret;
 945
 946         workspace = find_workspace(type);
 947         if (IS_ERR(workspace))
 948                 return -ENOMEM;
 949
 950         ret = btrfs_compress_op[type-1]->decompress(workspace, data_in,
 951                                                   dest_page, start_byte,
 952                                                   srclen, destlen);
 953
 954         free_workspace(type, workspace);
 955         return ret;
 956 }
 957
 958 void btrfs_exit_compress(void)
 959 {
 960         free_workspaces();
 961 }
 962
 963 /*
 964  * Copy uncompressed data from working buffer to pages.
 965  *
 966  * buf_start is the byte offset we're of the start of our workspace buffer.
 967  *
 968  * total_out is the last byte of the buffer
 969  */
 970 int btrfs_decompress_buf2page(char *buf, unsigned long buf_start,
 971                               unsigned long total_out, u64 disk_start,
 972                               struct bio_vec *bvec, int vcnt,
 973                               unsigned long *pg_index,
 974                               unsigned long *pg_offset)
 975 {
 976         unsigned long buf_offset;
 977         unsigned long current_buf_start;
 978         unsigned long start_byte;
 979         unsigned long working_bytes = total_out - buf_start;
 980         unsigned long bytes;
 981         char *kaddr;
 982         struct page *page_out = bvec[*pg_index].bv_page;
 983
 984         /*
 985          * start byte is the first byte of the page we're currently
 986          * copying into relative to the start of the compressed data.
 987          */
 988         start_byte = page_offset(page_out) - disk_start;
 989
 990         /* we haven't yet hit data corresponding to this page */
 991         if (total_out <= start_byte)
 992                 return 1;
 993
 994         /*
 995          * the start of the data we care about is offset into
 996          * the middle of our working buffer
 997          */
 998         if (total_out > start_byte && buf_start < start_byte) {
 999                 buf_offset = start_byte - buf_start;
1000                 working_bytes -= buf_offset;
1001         } else {
1002                 buf_offset = 0;
1003         }
1004         current_buf_start = buf_start;
1005
1006         /* copy bytes from the working buffer into the pages */
1007         while (working_bytes > 0) {
1008                 bytes = min(PAGE_CACHE_SIZE - *pg_offset,
1009                             PAGE_CACHE_SIZE - buf_offset);
1010                 bytes = min(bytes, working_bytes);
1011                 kaddr = kmap_atomic(page_out);
1012                 memcpy(kaddr + *pg_offset, buf + buf_offset, bytes);
1013                 if (*pg_index == (vcnt - 1) && *pg_offset == 0)
1014                         memset(kaddr + bytes, 0, PAGE_CACHE_SIZE - bytes);
1015                 kunmap_atomic(kaddr);
1016                 flush_dcache_page(page_out);
1017
1018                 *pg_offset += bytes;
1019                 buf_offset += bytes;
1020                 working_bytes -= bytes;
1021                 current_buf_start += bytes;
1022
1023                 /* check if we need to pick another page */
1024                 if (*pg_offset == PAGE_CACHE_SIZE) {
1025                         (*pg_index)++;
1026                         if (*pg_index >= vcnt)
1027                                 return 0;
1028
1029                         page_out = bvec[*pg_index].bv_page;
1030                         *pg_offset = 0;
1031                         start_byte = page_offset(page_out) - disk_start;
1032
1033                         /*
1034                          * make sure our new page is covered by this
1035                          * working buffer
1036                          */
1037                         if (total_out <= start_byte)
1038                                 return 1;
1039
1040                         /*
1041                          * the next page in the biovec might not be adjacent
1042                          * to the last page, but it might still be found
1043                          * inside this working buffer. bump our offset pointer
1044                          */
1045                         if (total_out > start_byte &&
1046                             current_buf_start < start_byte) {
1047                                 buf_offset = start_byte - buf_start;
1048                                 working_bytes = total_out - start_byte;
1049                                 current_buf_start = buf_start + buf_offset;
1050                         }
1051                 }
1052         }
1053
1054         return 1;
1055 }