Pileus Git - ~andy/linux/blob - 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 struct workspace {
  36         z_stream inf_strm;
  37         z_stream def_strm;
  38         char *buf;
  39         struct list_head list;
  40 };
  41
  42 static void zlib_free_workspace(struct list_head *ws)
  43 {
  44         struct workspace *workspace = list_entry(ws, struct workspace, list);
  45
  46         vfree(workspace->def_strm.workspace);
  47         vfree(workspace->inf_strm.workspace);
  48         kfree(workspace->buf);
  49         kfree(workspace);
  50 }
  51
  52 static struct list_head *zlib_alloc_workspace(void)
  53 {
  54         struct workspace *workspace;
  55
  56         workspace = kzalloc(sizeof(*workspace), GFP_NOFS);
  57         if (!workspace)
  58                 return ERR_PTR(-ENOMEM);
  59
  60         workspace->def_strm.workspace = vmalloc(zlib_deflate_workspacesize());
  61         workspace->inf_strm.workspace = vmalloc(zlib_inflate_workspacesize());
  62         workspace->buf = kmalloc(PAGE_CACHE_SIZE, GFP_NOFS);
  63         if (!workspace->def_strm.workspace ||
  64             !workspace->inf_strm.workspace || !workspace->buf)
  65                 goto fail;
  66
  67         INIT_LIST_HEAD(&workspace->list);
  68
  69         return &workspace->list;
  70 fail:
  71         zlib_free_workspace(&workspace->list);
  72         return ERR_PTR(-ENOMEM);
  73 }
  74
  75 static int zlib_compress_pages(struct list_head *ws,
  76                                struct address_space *mapping,
  77                                u64 start, unsigned long len,
  78                                struct page **pages,
  79                                unsigned long nr_dest_pages,
  80                                unsigned long *out_pages,
  81                                unsigned long *total_in,
  82                                unsigned long *total_out,
  83                                unsigned long max_out)
  84 {
  85         struct workspace *workspace = list_entry(ws, struct workspace, list);
  86         int ret;
  87         char *data_in;
  88         char *cpage_out;
  89         int nr_pages = 0;
  90         struct page *in_page = NULL;
  91         struct page *out_page = NULL;
  92         unsigned long bytes_left;
  93
  94         *out_pages = 0;
  95         *total_out = 0;
  96         *total_in = 0;
  97
  98         if (Z_OK != zlib_deflateInit(&workspace->def_strm, 3)) {
  99                 printk(KERN_WARNING "deflateInit failed\n");
 100                 ret = -1;
 101                 goto out;
 102         }
 103
 104         workspace->def_strm.total_in = 0;
 105         workspace->def_strm.total_out = 0;
 106
 107         in_page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
 108         data_in = kmap(in_page);
 109
 110         out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
 111         if (out_page == NULL) {
 112                 ret = -1;
 113                 goto out;
 114         }
 115         cpage_out = kmap(out_page);
 116         pages[0] = out_page;
 117         nr_pages = 1;
 118
 119         workspace->def_strm.next_in = data_in;
 120         workspace->def_strm.next_out = cpage_out;
 121         workspace->def_strm.avail_out = PAGE_CACHE_SIZE;
 122         workspace->def_strm.avail_in = min(len, PAGE_CACHE_SIZE);
 123
 124         while (workspace->def_strm.total_in < len) {
 125                 ret = zlib_deflate(&workspace->def_strm, Z_SYNC_FLUSH);
 126                 if (ret != Z_OK) {
 127                         printk(KERN_DEBUG "btrfs deflate in loop returned %d\n",
 128                                ret);
 129                         zlib_deflateEnd(&workspace->def_strm);
 130                         ret = -1;
 131                         goto out;
 132                 }
 133
 134                 /* we're making it bigger, give up */
 135                 if (workspace->def_strm.total_in > 8192 &&
 136                     workspace->def_strm.total_in <
 137                     workspace->def_strm.total_out) {
 138                         ret = -1;
 139                         goto out;
 140                 }
 141                 /* we need another page for writing out.  Test this
 142                  * before the total_in so we will pull in a new page for
 143                  * the stream end if required
 144                  */
 145                 if (workspace->def_strm.avail_out == 0) {
 146                         kunmap(out_page);
 147                         if (nr_pages == nr_dest_pages) {
 148                                 out_page = NULL;
 149                                 ret = -1;
 150                                 goto out;
 151                         }
 152                         out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
 153                         if (out_page == NULL) {
 154                                 ret = -1;
 155                                 goto out;
 156                         }
 157                         cpage_out = kmap(out_page);
 158                         pages[nr_pages] = out_page;
 159                         nr_pages++;
 160                         workspace->def_strm.avail_out = PAGE_CACHE_SIZE;
 161                         workspace->def_strm.next_out = cpage_out;
 162                 }
 163                 /* we're all done */
 164                 if (workspace->def_strm.total_in >= len)
 165                         break;
 166
 167                 /* we've read in a full page, get a new one */
 168                 if (workspace->def_strm.avail_in == 0) {
 169                         if (workspace->def_strm.total_out > max_out)
 170                                 break;
 171
 172                         bytes_left = len - workspace->def_strm.total_in;
 173                         kunmap(in_page);
 174                         page_cache_release(in_page);
 175
 176                         start += PAGE_CACHE_SIZE;
 177                         in_page = find_get_page(mapping,
 178                                                 start >> PAGE_CACHE_SHIFT);
 179                         data_in = kmap(in_page);
 180                         workspace->def_strm.avail_in = min(bytes_left,
 181                                                            PAGE_CACHE_SIZE);
 182                         workspace->def_strm.next_in = data_in;
 183                 }
 184         }
 185         workspace->def_strm.avail_in = 0;
 186         ret = zlib_deflate(&workspace->def_strm, Z_FINISH);
 187         zlib_deflateEnd(&workspace->def_strm);
 188
 189         if (ret != Z_STREAM_END) {
 190                 ret = -1;
 191                 goto out;
 192         }
 193
 194         if (workspace->def_strm.total_out >= workspace->def_strm.total_in) {
 195                 ret = -1;
 196                 goto out;
 197         }
 198
 199         ret = 0;
 200         *total_out = workspace->def_strm.total_out;
 201         *total_in = workspace->def_strm.total_in;
 202 out:
 203         *out_pages = nr_pages;
 204         if (out_page)
 205                 kunmap(out_page);
 206
 207         if (in_page) {
 208                 kunmap(in_page);
 209                 page_cache_release(in_page);
 210         }
 211         return ret;
 212 }
 213
 214 static int zlib_decompress_biovec(struct list_head *ws, struct page **pages_in,
 215                                   u64 disk_start,
 216                                   struct bio_vec *bvec,
 217                                   int vcnt,
 218                                   size_t srclen)
 219 {
 220         struct workspace *workspace = list_entry(ws, struct workspace, list);
 221         int ret = 0;
 222         int wbits = MAX_WBITS;
 223         char *data_in;
 224         size_t total_out = 0;
 225         unsigned long page_bytes_left;
 226         unsigned long page_in_index = 0;
 227         unsigned long page_out_index = 0;
 228         struct page *page_out;
 229         unsigned long total_pages_in = (srclen + PAGE_CACHE_SIZE - 1) /
 230                                         PAGE_CACHE_SIZE;
 231         unsigned long buf_start;
 232         unsigned long buf_offset;
 233         unsigned long bytes;
 234         unsigned long working_bytes;
 235         unsigned long pg_offset;
 236         unsigned long start_byte;
 237         unsigned long current_buf_start;
 238         char *kaddr;
 239
 240         data_in = kmap(pages_in[page_in_index]);
 241         workspace->inf_strm.next_in = data_in;
 242         workspace->inf_strm.avail_in = min_t(size_t, srclen, PAGE_CACHE_SIZE);
 243         workspace->inf_strm.total_in = 0;
 244
 245         workspace->inf_strm.total_out = 0;
 246         workspace->inf_strm.next_out = workspace->buf;
 247         workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
 248         page_out = bvec[page_out_index].bv_page;
 249         page_bytes_left = PAGE_CACHE_SIZE;
 250         pg_offset = 0;
 251
 252         /* If it's deflate, and it's got no preset dictionary, then
 253            we can tell zlib to skip the adler32 check. */
 254         if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
 255             ((data_in[0] & 0x0f) == Z_DEFLATED) &&
 256             !(((data_in[0]<<8) + data_in[1]) % 31)) {
 257
 258                 wbits = -((data_in[0] >> 4) + 8);
 259                 workspace->inf_strm.next_in += 2;
 260                 workspace->inf_strm.avail_in -= 2;
 261         }
 262
 263         if (Z_OK != zlib_inflateInit2(&workspace->inf_strm, wbits)) {
 264                 printk(KERN_WARNING "inflateInit failed\n");
 265                 return -1;
 266         }
 267         while (workspace->inf_strm.total_in < srclen) {
 268                 ret = zlib_inflate(&workspace->inf_strm, Z_NO_FLUSH);
 269                 if (ret != Z_OK && ret != Z_STREAM_END)
 270                         break;
 271                 /*
 272                  * buf start is the byte offset we're of the start of
 273                  * our workspace buffer
 274                  */
 275                 buf_start = total_out;
 276
 277                 /* total_out is the last byte of the workspace buffer */
 278                 total_out = workspace->inf_strm.total_out;
 279
 280                 working_bytes = total_out - buf_start;
 281
 282                 /*
 283                  * start byte is the first byte of the page we're currently
 284                  * copying into relative to the start of the compressed data.
 285                  */
 286                 start_byte = page_offset(page_out) - disk_start;
 287
 288                 if (working_bytes == 0) {
 289                         /* we didn't make progress in this inflate
 290                          * call, we're done
 291                          */
 292                         if (ret != Z_STREAM_END)
 293                                 ret = -1;
 294                         break;
 295                 }
 296
 297                 /* we haven't yet hit data corresponding to this page */
 298                 if (total_out <= start_byte)
 299                         goto next;
 300
 301                 /*
 302                  * the start of the data we care about is offset into
 303                  * the middle of our working buffer
 304                  */
 305                 if (total_out > start_byte && buf_start < start_byte) {
 306                         buf_offset = start_byte - buf_start;
 307                         working_bytes -= buf_offset;
 308                 } else {
 309                         buf_offset = 0;
 310                 }
 311                 current_buf_start = buf_start;
 312
 313                 /* copy bytes from the working buffer into the pages */
 314                 while (working_bytes > 0) {
 315                         bytes = min(PAGE_CACHE_SIZE - pg_offset,
 316                                     PAGE_CACHE_SIZE - buf_offset);
 317                         bytes = min(bytes, working_bytes);
 318                         kaddr = kmap_atomic(page_out, KM_USER0);
 319                         memcpy(kaddr + pg_offset, workspace->buf + buf_offset,
 320                                bytes);
 321                         kunmap_atomic(kaddr, KM_USER0);
 322                         flush_dcache_page(page_out);
 323
 324                         pg_offset += bytes;
 325                         page_bytes_left -= bytes;
 326                         buf_offset += bytes;
 327                         working_bytes -= bytes;
 328                         current_buf_start += bytes;
 329
 330                         /* check if we need to pick another page */
 331                         if (page_bytes_left == 0) {
 332                                 page_out_index++;
 333                                 if (page_out_index >= vcnt) {
 334                                         ret = 0;
 335                                         goto done;
 336                                 }
 337
 338                                 page_out = bvec[page_out_index].bv_page;
 339                                 pg_offset = 0;
 340                                 page_bytes_left = PAGE_CACHE_SIZE;
 341                                 start_byte = page_offset(page_out) - disk_start;
 342
 343                                 /*
 344                                  * make sure our new page is covered by this
 345                                  * working buffer
 346                                  */
 347                                 if (total_out <= start_byte)
 348                                         goto next;
 349
 350                                 /* the next page in the biovec might not
 351                                  * be adjacent to the last page, but it
 352                                  * might still be found inside this working
 353                                  * buffer.  bump our offset pointer
 354                                  */
 355                                 if (total_out > start_byte &&
 356                                     current_buf_start < start_byte) {
 357                                         buf_offset = start_byte - buf_start;
 358                                         working_bytes = total_out - start_byte;
 359                                         current_buf_start = buf_start +
 360                                                 buf_offset;
 361                                 }
 362                         }
 363                 }
 364 next:
 365                 workspace->inf_strm.next_out = workspace->buf;
 366                 workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
 367
 368                 if (workspace->inf_strm.avail_in == 0) {
 369                         unsigned long tmp;
 370                         kunmap(pages_in[page_in_index]);
 371                         page_in_index++;
 372                         if (page_in_index >= total_pages_in) {
 373                                 data_in = NULL;
 374                                 break;
 375                         }
 376                         data_in = kmap(pages_in[page_in_index]);
 377                         workspace->inf_strm.next_in = data_in;
 378                         tmp = srclen - workspace->inf_strm.total_in;
 379                         workspace->inf_strm.avail_in = min(tmp,
 380                                                            PAGE_CACHE_SIZE);
 381                 }
 382         }
 383         if (ret != Z_STREAM_END)
 384                 ret = -1;
 385         else
 386                 ret = 0;
 387 done:
 388         zlib_inflateEnd(&workspace->inf_strm);
 389         if (data_in)
 390                 kunmap(pages_in[page_in_index]);
 391         return ret;
 392 }
 393
 394 static int zlib_decompress(struct list_head *ws, unsigned char *data_in,
 395                            struct page *dest_page,
 396                            unsigned long start_byte,
 397                            size_t srclen, size_t destlen)
 398 {
 399         struct workspace *workspace = list_entry(ws, struct workspace, list);
 400         int ret = 0;
 401         int wbits = MAX_WBITS;
 402         unsigned long bytes_left = destlen;
 403         unsigned long total_out = 0;
 404         char *kaddr;
 405
 406         workspace->inf_strm.next_in = data_in;
 407         workspace->inf_strm.avail_in = srclen;
 408         workspace->inf_strm.total_in = 0;
 409
 410         workspace->inf_strm.next_out = workspace->buf;
 411         workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
 412         workspace->inf_strm.total_out = 0;
 413         /* If it's deflate, and it's got no preset dictionary, then
 414            we can tell zlib to skip the adler32 check. */
 415         if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
 416             ((data_in[0] & 0x0f) == Z_DEFLATED) &&
 417             !(((data_in[0]<<8) + data_in[1]) % 31)) {
 418
 419                 wbits = -((data_in[0] >> 4) + 8);
 420                 workspace->inf_strm.next_in += 2;
 421                 workspace->inf_strm.avail_in -= 2;
 422         }
 423
 424         if (Z_OK != zlib_inflateInit2(&workspace->inf_strm, wbits)) {
 425                 printk(KERN_WARNING "inflateInit failed\n");
 426                 return -1;
 427         }
 428
 429         while (bytes_left > 0) {
 430                 unsigned long buf_start;
 431                 unsigned long buf_offset;
 432                 unsigned long bytes;
 433                 unsigned long pg_offset = 0;
 434
 435                 ret = zlib_inflate(&workspace->inf_strm, Z_NO_FLUSH);
 436                 if (ret != Z_OK && ret != Z_STREAM_END)
 437                         break;
 438
 439                 buf_start = total_out;
 440                 total_out = workspace->inf_strm.total_out;
 441
 442                 if (total_out == buf_start) {
 443                         ret = -1;
 444                         break;
 445                 }
 446
 447                 if (total_out <= start_byte)
 448                         goto next;
 449
 450                 if (total_out > start_byte && buf_start < start_byte)
 451                         buf_offset = start_byte - buf_start;
 452                 else
 453                         buf_offset = 0;
 454
 455                 bytes = min(PAGE_CACHE_SIZE - pg_offset,
 456                             PAGE_CACHE_SIZE - buf_offset);
 457                 bytes = min(bytes, bytes_left);
 458
 459                 kaddr = kmap_atomic(dest_page, KM_USER0);
 460                 memcpy(kaddr + pg_offset, workspace->buf + buf_offset, bytes);
 461                 kunmap_atomic(kaddr, KM_USER0);
 462
 463                 pg_offset += bytes;
 464                 bytes_left -= bytes;
 465 next:
 466                 workspace->inf_strm.next_out = workspace->buf;
 467                 workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
 468         }
 469
 470         if (ret != Z_STREAM_END && bytes_left != 0)
 471                 ret = -1;
 472         else
 473                 ret = 0;
 474
 475         zlib_inflateEnd(&workspace->inf_strm);
 476         return ret;
 477 }
 478
 479 struct btrfs_compress_op btrfs_zlib_compress = {
 480         .alloc_workspace        = zlib_alloc_workspace,
 481         .free_workspace         = zlib_free_workspace,
 482         .compress_pages         = zlib_compress_pages,
 483         .decompress_biovec      = zlib_decompress_biovec,
 484         .decompress             = zlib_decompress,
 485 };