fs/btrfs/zlib.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  * Based on jffs2 zlib code:
  19  * Copyright © 2001-2007 Red Hat, Inc.
  20  * Created by David Woodhouse <dwmw2@infradead.org>
  21  */
  22
  23 #include <linux/kernel.h>
  24 #include <linux/slab.h>
  25 #include <linux/zlib.h>
  26 #include <linux/zutil.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/init.h>
  29 #include <linux/err.h>
  30 #include <linux/sched.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/bio.h>
  33 #include "compression.h"
  34
  35 /* Plan: call deflate() with avail_in == *sourcelen,
  36         avail_out = *dstlen - 12 and flush == Z_FINISH.
  37         If it doesn't manage to finish, call it again with
  38         avail_in == 0 and avail_out set to the remaining 12
  39         bytes for it to clean up.
  40    Q: Is 12 bytes sufficient?
  41 */
  42 #define STREAM_END_SPACE 12
  43
  44 struct workspace {
  45         z_stream inf_strm;
  46         z_stream def_strm;
  47         char *buf;
  48         struct list_head list;
  49 };
  50
  51 static LIST_HEAD(idle_workspace);
  52 static DEFINE_SPINLOCK(workspace_lock);
  53 static unsigned long num_workspace;
  54 static atomic_t alloc_workspace = ATOMIC_INIT(0);
  55 static DECLARE_WAIT_QUEUE_HEAD(workspace_wait);
  56
  57 /*
  58  * this finds an available zlib workspace or allocates a new one
  59  * NULL or an ERR_PTR is returned if things go bad.
  60  */
  61 static struct workspace *find_zlib_workspace(void)
  62 {
  63         struct workspace *workspace;
  64         int ret;
  65         int cpus = num_online_cpus();
  66
  67 again:
  68         spin_lock(&workspace_lock);
  69         if (!list_empty(&idle_workspace)) {
  70                 workspace = list_entry(idle_workspace.next, struct workspace,
  71                                        list);
  72                 list_del(&workspace->list);
  73                 num_workspace--;
  74                 spin_unlock(&workspace_lock);
  75                 return workspace;
  76
  77         }
  78         spin_unlock(&workspace_lock);
  79         if (atomic_read(&alloc_workspace) > cpus) {
  80                 DEFINE_WAIT(wait);
  81                 prepare_to_wait(&workspace_wait, &wait, TASK_UNINTERRUPTIBLE);
  82                 if (atomic_read(&alloc_workspace) > cpus)
  83                         schedule();
  84                 finish_wait(&workspace_wait, &wait);
  85                 goto again;
  86         }
  87         atomic_inc(&alloc_workspace);
  88         workspace = kzalloc(sizeof(*workspace), GFP_NOFS);
  89         if (!workspace) {
  90                 ret = -ENOMEM;
  91                 goto fail;
  92         }
  93
  94         workspace->def_strm.workspace = vmalloc(zlib_deflate_workspacesize());
  95         if (!workspace->def_strm.workspace) {
  96                 ret = -ENOMEM;
  97                 goto fail;
  98         }
  99         workspace->inf_strm.workspace = vmalloc(zlib_inflate_workspacesize());
 100         if (!workspace->inf_strm.workspace) {
 101                 ret = -ENOMEM;
 102                 goto fail_inflate;
 103         }
 104         workspace->buf = kmalloc(PAGE_CACHE_SIZE, GFP_NOFS);
 105         if (!workspace->buf) {
 106                 ret = -ENOMEM;
 107                 goto fail_kmalloc;
 108         }
 109         return workspace;
 110
 111 fail_kmalloc:
 112         vfree(workspace->inf_strm.workspace);
 113 fail_inflate:
 114         vfree(workspace->def_strm.workspace);
 115 fail:
 116         kfree(workspace);
 117         atomic_dec(&alloc_workspace);
 118         wake_up(&workspace_wait);
 119         return ERR_PTR(ret);
 120 }
 121
 122 /*
 123  * put a workspace struct back on the list or free it if we have enough
 124  * idle ones sitting around
 125  */
 126 static int free_workspace(struct workspace *workspace)
 127 {
 128         spin_lock(&workspace_lock);
 129         if (num_workspace < num_online_cpus()) {
 130                 list_add_tail(&workspace->list, &idle_workspace);
 131                 num_workspace++;
 132                 spin_unlock(&workspace_lock);
 133                 if (waitqueue_active(&workspace_wait))
 134                         wake_up(&workspace_wait);
 135                 return 0;
 136         }
 137         spin_unlock(&workspace_lock);
 138         vfree(workspace->def_strm.workspace);
 139         vfree(workspace->inf_strm.workspace);
 140         kfree(workspace->buf);
 141         kfree(workspace);
 142
 143         atomic_dec(&alloc_workspace);
 144         if (waitqueue_active(&workspace_wait))
 145                 wake_up(&workspace_wait);
 146         return 0;
 147 }
 148
 149 /*
 150  * cleanup function for module exit
 151  */
 152 static void free_workspaces(void)
 153 {
 154         struct workspace *workspace;
 155         while (!list_empty(&idle_workspace)) {
 156                 workspace = list_entry(idle_workspace.next, struct workspace,
 157                                        list);
 158                 list_del(&workspace->list);
 159                 vfree(workspace->def_strm.workspace);
 160                 vfree(workspace->inf_strm.workspace);
 161                 kfree(workspace->buf);
 162                 kfree(workspace);
 163                 atomic_dec(&alloc_workspace);
 164         }
 165 }
 166
 167 /*
 168  * given an address space and start/len, compress the bytes.
 169  *
 170  * pages are allocated to hold the compressed result and stored
 171  * in 'pages'
 172  *
 173  * out_pages is used to return the number of pages allocated.  There
 174  * may be pages allocated even if we return an error
 175  *
 176  * total_in is used to return the number of bytes actually read.  It
 177  * may be smaller then len if we had to exit early because we
 178  * ran out of room in the pages array or because we cross the
 179  * max_out threshold.
 180  *
 181  * total_out is used to return the total number of compressed bytes
 182  *
 183  * max_out tells us the max number of bytes that we're allowed to
 184  * stuff into pages
 185  */
 186 int btrfs_zlib_compress_pages(struct address_space *mapping,
 187                               u64 start, unsigned long len,
 188                               struct page **pages,
 189                               unsigned long nr_dest_pages,
 190                               unsigned long *out_pages,
 191                               unsigned long *total_in,
 192                               unsigned long *total_out,
 193                               unsigned long max_out)
 194 {
 195         int ret;
 196         struct workspace *workspace;
 197         char *data_in;
 198         char *cpage_out;
 199         int nr_pages = 0;
 200         struct page *in_page = NULL;
 201         struct page *out_page = NULL;
 202         int out_written = 0;
 203         int in_read = 0;
 204         unsigned long bytes_left;
 205
 206         *out_pages = 0;
 207         *total_out = 0;
 208         *total_in = 0;
 209
 210         workspace = find_zlib_workspace();
 211         if (!workspace)
 212                 return -1;
 213
 214         if (Z_OK != zlib_deflateInit(&workspace->def_strm, 3)) {
 215                 printk(KERN_WARNING "deflateInit failed\n");
 216                 ret = -1;
 217                 goto out;
 218         }
 219
 220         workspace->def_strm.total_in = 0;
 221         workspace->def_strm.total_out = 0;
 222
 223         in_page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
 224         data_in = kmap(in_page);
 225
 226         out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
 227         cpage_out = kmap(out_page);
 228         pages[0] = out_page;
 229         nr_pages = 1;
 230
 231         workspace->def_strm.next_in = data_in;
 232         workspace->def_strm.next_out = cpage_out;
 233         workspace->def_strm.avail_out = PAGE_CACHE_SIZE;
 234         workspace->def_strm.avail_in = min(len, PAGE_CACHE_SIZE);
 235
 236         out_written = 0;
 237         in_read = 0;
 238
 239         while (workspace->def_strm.total_in < len) {
 240                 ret = zlib_deflate(&workspace->def_strm, Z_SYNC_FLUSH);
 241                 if (ret != Z_OK) {
 242                         printk(KERN_DEBUG "btrfs deflate in loop returned %d\n",
 243                                ret);
 244                         zlib_deflateEnd(&workspace->def_strm);
 245                         ret = -1;
 246                         goto out;
 247                 }
 248
 249                 /* we're making it bigger, give up */
 250                 if (workspace->def_strm.total_in > 8192 &&
 251                     workspace->def_strm.total_in <
 252                     workspace->def_strm.total_out) {
 253                         ret = -1;
 254                         goto out;
 255                 }
 256                 /* we need another page for writing out.  Test this
 257                  * before the total_in so we will pull in a new page for
 258                  * the stream end if required
 259                  */
 260                 if (workspace->def_strm.avail_out == 0) {
 261                         kunmap(out_page);
 262                         if (nr_pages == nr_dest_pages) {
 263                                 out_page = NULL;
 264                                 ret = -1;
 265                                 goto out;
 266                         }
 267                         out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
 268                         cpage_out = kmap(out_page);
 269                         pages[nr_pages] = out_page;
 270                         nr_pages++;
 271                         workspace->def_strm.avail_out = PAGE_CACHE_SIZE;
 272                         workspace->def_strm.next_out = cpage_out;
 273                 }
 274                 /* we're all done */
 275                 if (workspace->def_strm.total_in >= len)
 276                         break;
 277
 278                 /* we've read in a full page, get a new one */
 279                 if (workspace->def_strm.avail_in == 0) {
 280                         if (workspace->def_strm.total_out > max_out)
 281                                 break;
 282
 283                         bytes_left = len - workspace->def_strm.total_in;
 284                         kunmap(in_page);
 285                         page_cache_release(in_page);
 286
 287                         start += PAGE_CACHE_SIZE;
 288                         in_page = find_get_page(mapping,
 289                                                 start >> PAGE_CACHE_SHIFT);
 290                         data_in = kmap(in_page);
 291                         workspace->def_strm.avail_in = min(bytes_left,
 292                                                            PAGE_CACHE_SIZE);
 293                         workspace->def_strm.next_in = data_in;
 294                 }
 295         }
 296         workspace->def_strm.avail_in = 0;
 297         ret = zlib_deflate(&workspace->def_strm, Z_FINISH);
 298         zlib_deflateEnd(&workspace->def_strm);
 299
 300         if (ret != Z_STREAM_END) {
 301                 ret = -1;
 302                 goto out;
 303         }
 304
 305         if (workspace->def_strm.total_out >= workspace->def_strm.total_in) {
 306                 ret = -1;
 307                 goto out;
 308         }
 309
 310         ret = 0;
 311         *total_out = workspace->def_strm.total_out;
 312         *total_in = workspace->def_strm.total_in;
 313 out:
 314         *out_pages = nr_pages;
 315         if (out_page)
 316                 kunmap(out_page);
 317
 318         if (in_page) {
 319                 kunmap(in_page);
 320                 page_cache_release(in_page);
 321         }
 322         free_workspace(workspace);
 323         return ret;
 324 }
 325
 326 /*
 327  * pages_in is an array of pages with compressed data.
 328  *
 329  * disk_start is the starting logical offset of this array in the file
 330  *
 331  * bvec is a bio_vec of pages from the file that we want to decompress into
 332  *
 333  * vcnt is the count of pages in the biovec
 334  *
 335  * srclen is the number of bytes in pages_in
 336  *
 337  * The basic idea is that we have a bio that was created by readpages.
 338  * The pages in the bio are for the uncompressed data, and they may not
 339  * be contiguous.  They all correspond to the range of bytes covered by
 340  * the compressed extent.
 341  */
 342 int btrfs_zlib_decompress_biovec(struct page **pages_in,
 343                               u64 disk_start,
 344                               struct bio_vec *bvec,
 345                               int vcnt,
 346                               size_t srclen)
 347 {
 348         int ret = 0;
 349         int wbits = MAX_WBITS;
 350         struct workspace *workspace;
 351         char *data_in;
 352         size_t total_out = 0;
 353         unsigned long page_bytes_left;
 354         unsigned long page_in_index = 0;
 355         unsigned long page_out_index = 0;
 356         struct page *page_out;
 357         unsigned long total_pages_in = (srclen + PAGE_CACHE_SIZE - 1) /
 358                                         PAGE_CACHE_SIZE;
 359         unsigned long buf_start;
 360         unsigned long buf_offset;
 361         unsigned long bytes;
 362         unsigned long working_bytes;
 363         unsigned long pg_offset;
 364         unsigned long start_byte;
 365         unsigned long current_buf_start;
 366         char *kaddr;
 367
 368         workspace = find_zlib_workspace();
 369         if (!workspace)
 370                 return -ENOMEM;
 371
 372         data_in = kmap(pages_in[page_in_index]);
 373         workspace->inf_strm.next_in = data_in;
 374         workspace->inf_strm.avail_in = min_t(size_t, srclen, PAGE_CACHE_SIZE);
 375         workspace->inf_strm.total_in = 0;
 376
 377         workspace->inf_strm.total_out = 0;
 378         workspace->inf_strm.next_out = workspace->buf;
 379         workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
 380         page_out = bvec[page_out_index].bv_page;
 381         page_bytes_left = PAGE_CACHE_SIZE;
 382         pg_offset = 0;
 383
 384         /* If it's deflate, and it's got no preset dictionary, then
 385            we can tell zlib to skip the adler32 check. */
 386         if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
 387             ((data_in[0] & 0x0f) == Z_DEFLATED) &&
 388             !(((data_in[0]<<8) + data_in[1]) % 31)) {
 389
 390                 wbits = -((data_in[0] >> 4) + 8);
 391                 workspace->inf_strm.next_in += 2;
 392                 workspace->inf_strm.avail_in -= 2;
 393         }
 394
 395         if (Z_OK != zlib_inflateInit2(&workspace->inf_strm, wbits)) {
 396                 printk(KERN_WARNING "inflateInit failed\n");
 397                 ret = -1;
 398                 goto out;
 399         }
 400         while (workspace->inf_strm.total_in < srclen) {
 401                 ret = zlib_inflate(&workspace->inf_strm, Z_NO_FLUSH);
 402                 if (ret != Z_OK && ret != Z_STREAM_END)
 403                         break;
 404                 /*
 405                  * buf start is the byte offset we're of the start of
 406                  * our workspace buffer
 407                  */
 408                 buf_start = total_out;
 409
 410                 /* total_out is the last byte of the workspace buffer */
 411                 total_out = workspace->inf_strm.total_out;
 412
 413                 working_bytes = total_out - buf_start;
 414
 415                 /*
 416                  * start byte is the first byte of the page we're currently
 417                  * copying into relative to the start of the compressed data.
 418                  */
 419                 start_byte = page_offset(page_out) - disk_start;
 420
 421                 if (working_bytes == 0) {
 422                         /* we didn't make progress in this inflate
 423                          * call, we're done
 424                          */
 425                         if (ret != Z_STREAM_END)
 426                                 ret = -1;
 427                         break;
 428                 }
 429
 430                 /* we haven't yet hit data corresponding to this page */
 431                 if (total_out <= start_byte)
 432                         goto next;
 433
 434                 /*
 435                  * the start of the data we care about is offset into
 436                  * the middle of our working buffer
 437                  */
 438                 if (total_out > start_byte && buf_start < start_byte) {
 439                         buf_offset = start_byte - buf_start;
 440                         working_bytes -= buf_offset;
 441                 } else {
 442                         buf_offset = 0;
 443                 }
 444                 current_buf_start = buf_start;
 445
 446                 /* copy bytes from the working buffer into the pages */
 447                 while (working_bytes > 0) {
 448                         bytes = min(PAGE_CACHE_SIZE - pg_offset,
 449                                     PAGE_CACHE_SIZE - buf_offset);
 450                         bytes = min(bytes, working_bytes);
 451                         kaddr = kmap_atomic(page_out, KM_USER0);
 452                         memcpy(kaddr + pg_offset, workspace->buf + buf_offset,
 453                                bytes);
 454                         kunmap_atomic(kaddr, KM_USER0);
 455                         flush_dcache_page(page_out);
 456
 457                         pg_offset += bytes;
 458                         page_bytes_left -= bytes;
 459                         buf_offset += bytes;
 460                         working_bytes -= bytes;
 461                         current_buf_start += bytes;
 462
 463                         /* check if we need to pick another page */
 464                         if (page_bytes_left == 0) {
 465                                 page_out_index++;
 466                                 if (page_out_index >= vcnt) {
 467                                         ret = 0;
 468                                         goto done;
 469                                 }
 470
 471                                 page_out = bvec[page_out_index].bv_page;
 472                                 pg_offset = 0;
 473                                 page_bytes_left = PAGE_CACHE_SIZE;
 474                                 start_byte = page_offset(page_out) - disk_start;
 475
 476                                 /*
 477                                  * make sure our new page is covered by this
 478                                  * working buffer
 479                                  */
 480                                 if (total_out <= start_byte)
 481                                         goto next;
 482
 483                                 /* the next page in the biovec might not
 484                                  * be adjacent to the last page, but it
 485                                  * might still be found inside this working
 486                                  * buffer.  bump our offset pointer
 487                                  */
 488                                 if (total_out > start_byte &&
 489                                     current_buf_start < start_byte) {
 490                                         buf_offset = start_byte - buf_start;
 491                                         working_bytes = total_out - start_byte;
 492                                         current_buf_start = buf_start +
 493                                                 buf_offset;
 494                                 }
 495                         }
 496                 }
 497 next:
 498                 workspace->inf_strm.next_out = workspace->buf;
 499                 workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
 500
 501                 if (workspace->inf_strm.avail_in == 0) {
 502                         unsigned long tmp;
 503                         kunmap(pages_in[page_in_index]);
 504                         page_in_index++;
 505                         if (page_in_index >= total_pages_in) {
 506                                 data_in = NULL;
 507                                 break;
 508                         }
 509                         data_in = kmap(pages_in[page_in_index]);
 510                         workspace->inf_strm.next_in = data_in;
 511                         tmp = srclen - workspace->inf_strm.total_in;
 512                         workspace->inf_strm.avail_in = min(tmp,
 513                                                            PAGE_CACHE_SIZE);
 514                 }
 515         }
 516         if (ret != Z_STREAM_END)
 517                 ret = -1;
 518         else
 519                 ret = 0;
 520 done:
 521         zlib_inflateEnd(&workspace->inf_strm);
 522         if (data_in)
 523                 kunmap(pages_in[page_in_index]);
 524 out:
 525         free_workspace(workspace);
 526         return ret;
 527 }
 528
 529 /*
 530  * a less complex decompression routine.  Our compressed data fits in a
 531  * single page, and we want to read a single page out of it.
 532  * start_byte tells us the offset into the compressed data we're interested in
 533  */
 534 int btrfs_zlib_decompress(unsigned char *data_in,
 535                           struct page *dest_page,
 536                           unsigned long start_byte,
 537                           size_t srclen, size_t destlen)
 538 {
 539         int ret = 0;
 540         int wbits = MAX_WBITS;
 541         struct workspace *workspace;
 542         unsigned long bytes_left = destlen;
 543         unsigned long total_out = 0;
 544         char *kaddr;
 545
 546         if (destlen > PAGE_CACHE_SIZE)
 547                 return -ENOMEM;
 548
 549         workspace = find_zlib_workspace();
 550         if (!workspace)
 551                 return -ENOMEM;
 552
 553         workspace->inf_strm.next_in = data_in;
 554         workspace->inf_strm.avail_in = srclen;
 555         workspace->inf_strm.total_in = 0;
 556
 557         workspace->inf_strm.next_out = workspace->buf;
 558         workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
 559         workspace->inf_strm.total_out = 0;
 560         /* If it's deflate, and it's got no preset dictionary, then
 561            we can tell zlib to skip the adler32 check. */
 562         if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
 563             ((data_in[0] & 0x0f) == Z_DEFLATED) &&
 564             !(((data_in[0]<<8) + data_in[1]) % 31)) {
 565
 566                 wbits = -((data_in[0] >> 4) + 8);
 567                 workspace->inf_strm.next_in += 2;
 568                 workspace->inf_strm.avail_in -= 2;
 569         }
 570
 571         if (Z_OK != zlib_inflateInit2(&workspace->inf_strm, wbits)) {
 572                 printk(KERN_WARNING "inflateInit failed\n");
 573                 ret = -1;
 574                 goto out;
 575         }
 576
 577         while (bytes_left > 0) {
 578                 unsigned long buf_start;
 579                 unsigned long buf_offset;
 580                 unsigned long bytes;
 581                 unsigned long pg_offset = 0;
 582
 583                 ret = zlib_inflate(&workspace->inf_strm, Z_NO_FLUSH);
 584                 if (ret != Z_OK && ret != Z_STREAM_END)
 585                         break;
 586
 587                 buf_start = total_out;
 588                 total_out = workspace->inf_strm.total_out;
 589
 590                 if (total_out == buf_start) {
 591                         ret = -1;
 592                         break;
 593                 }
 594
 595                 if (total_out <= start_byte)
 596                         goto next;
 597
 598                 if (total_out > start_byte && buf_start < start_byte)
 599                         buf_offset = start_byte - buf_start;
 600                 else
 601                         buf_offset = 0;
 602
 603                 bytes = min(PAGE_CACHE_SIZE - pg_offset,
 604                             PAGE_CACHE_SIZE - buf_offset);
 605                 bytes = min(bytes, bytes_left);
 606
 607                 kaddr = kmap_atomic(dest_page, KM_USER0);
 608                 memcpy(kaddr + pg_offset, workspace->buf + buf_offset, bytes);
 609                 kunmap_atomic(kaddr, KM_USER0);
 610
 611                 pg_offset += bytes;
 612                 bytes_left -= bytes;
 613 next:
 614                 workspace->inf_strm.next_out = workspace->buf;
 615                 workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
 616         }
 617
 618         if (ret != Z_STREAM_END && bytes_left != 0)
 619                 ret = -1;
 620         else
 621                 ret = 0;
 622
 623         zlib_inflateEnd(&workspace->inf_strm);
 624 out:
 625         free_workspace(workspace);
 626         return ret;
 627 }
 628
 629 void btrfs_zlib_exit(void)
 630 {
 631     free_workspaces();
 632 }