fs/btrfs/lzo.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/slab.h>
  21 #include <linux/vmalloc.h>
  22 #include <linux/init.h>
  23 #include <linux/err.h>
  24 #include <linux/sched.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/bio.h>
  27 #include <linux/lzo.h>
  28 #include "compression.h"
  29
  30 #define LZO_LEN 4
  31
  32 struct workspace {
  33         void *mem;
  34         void *buf;      /* where decompressed data goes */
  35         void *cbuf;     /* where compressed data goes */
  36         struct list_head list;
  37 };
  38
  39 static void lzo_free_workspace(struct list_head *ws)
  40 {
  41         struct workspace *workspace = list_entry(ws, struct workspace, list);
  42
  43         vfree(workspace->buf);
  44         vfree(workspace->cbuf);
  45         vfree(workspace->mem);
  46         kfree(workspace);
  47 }
  48
  49 static struct list_head *lzo_alloc_workspace(void)
  50 {
  51         struct workspace *workspace;
  52
  53         workspace = kzalloc(sizeof(*workspace), GFP_NOFS);
  54         if (!workspace)
  55                 return ERR_PTR(-ENOMEM);
  56
  57         workspace->mem = vmalloc(LZO1X_MEM_COMPRESS);
  58         workspace->buf = vmalloc(lzo1x_worst_compress(PAGE_CACHE_SIZE));
  59         workspace->cbuf = vmalloc(lzo1x_worst_compress(PAGE_CACHE_SIZE));
  60         if (!workspace->mem || !workspace->buf || !workspace->cbuf)
  61                 goto fail;
  62
  63         INIT_LIST_HEAD(&workspace->list);
  64
  65         return &workspace->list;
  66 fail:
  67         lzo_free_workspace(&workspace->list);
  68         return ERR_PTR(-ENOMEM);
  69 }
  70
  71 static inline void write_compress_length(char *buf, size_t len)
  72 {
  73         __le32 dlen;
  74
  75         dlen = cpu_to_le32(len);
  76         memcpy(buf, &dlen, LZO_LEN);
  77 }
  78
  79 static inline size_t read_compress_length(char *buf)
  80 {
  81         __le32 dlen;
  82
  83         memcpy(&dlen, buf, LZO_LEN);
  84         return le32_to_cpu(dlen);
  85 }
  86
  87 static int lzo_compress_pages(struct list_head *ws,
  88                               struct address_space *mapping,
  89                               u64 start, unsigned long len,
  90                               struct page **pages,
  91                               unsigned long nr_dest_pages,
  92                               unsigned long *out_pages,
  93                               unsigned long *total_in,
  94                               unsigned long *total_out,
  95                               unsigned long max_out)
  96 {
  97         struct workspace *workspace = list_entry(ws, struct workspace, list);
  98         int ret = 0;
  99         char *data_in;
 100         char *cpage_out;
 101         int nr_pages = 0;
 102         struct page *in_page = NULL;
 103         struct page *out_page = NULL;
 104         unsigned long bytes_left;
 105
 106         size_t in_len;
 107         size_t out_len;
 108         char *buf;
 109         unsigned long tot_in = 0;
 110         unsigned long tot_out = 0;
 111         unsigned long pg_bytes_left;
 112         unsigned long out_offset;
 113         unsigned long bytes;
 114
 115         *out_pages = 0;
 116         *total_out = 0;
 117         *total_in = 0;
 118
 119         in_page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
 120         data_in = kmap(in_page);
 121
 122         /*
 123          * store the size of all chunks of compressed data in
 124          * the first 4 bytes
 125          */
 126         out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
 127         if (out_page == NULL) {
 128                 ret = -ENOMEM;
 129                 goto out;
 130         }
 131         cpage_out = kmap(out_page);
 132         out_offset = LZO_LEN;
 133         tot_out = LZO_LEN;
 134         pages[0] = out_page;
 135         nr_pages = 1;
 136         pg_bytes_left = PAGE_CACHE_SIZE - LZO_LEN;
 137
 138         /* compress at most one page of data each time */
 139         in_len = min(len, PAGE_CACHE_SIZE);
 140         while (tot_in < len) {
 141                 ret = lzo1x_1_compress(data_in, in_len, workspace->cbuf,
 142                                        &out_len, workspace->mem);
 143                 if (ret != LZO_E_OK) {
 144                         printk(KERN_DEBUG "BTRFS: deflate in loop returned %d\n",
 145                                ret);
 146                         ret = -EIO;
 147                         goto out;
 148                 }
 149
 150                 /* store the size of this chunk of compressed data */
 151                 write_compress_length(cpage_out + out_offset, out_len);
 152                 tot_out += LZO_LEN;
 153                 out_offset += LZO_LEN;
 154                 pg_bytes_left -= LZO_LEN;
 155
 156                 tot_in += in_len;
 157                 tot_out += out_len;
 158
 159                 /* copy bytes from the working buffer into the pages */
 160                 buf = workspace->cbuf;
 161                 while (out_len) {
 162                         bytes = min_t(unsigned long, pg_bytes_left, out_len);
 163
 164                         memcpy(cpage_out + out_offset, buf, bytes);
 165
 166                         out_len -= bytes;
 167                         pg_bytes_left -= bytes;
 168                         buf += bytes;
 169                         out_offset += bytes;
 170
 171                         /*
 172                          * we need another page for writing out.
 173                          *
 174                          * Note if there's less than 4 bytes left, we just
 175                          * skip to a new page.
 176                          */
 177                         if ((out_len == 0 && pg_bytes_left < LZO_LEN) ||
 178                             pg_bytes_left == 0) {
 179                                 if (pg_bytes_left) {
 180                                         memset(cpage_out + out_offset, 0,
 181                                                pg_bytes_left);
 182                                         tot_out += pg_bytes_left;
 183                                 }
 184
 185                                 /* we're done, don't allocate new page */
 186                                 if (out_len == 0 && tot_in >= len)
 187                                         break;
 188
 189                                 kunmap(out_page);
 190                                 if (nr_pages == nr_dest_pages) {
 191                                         out_page = NULL;
 192                                         ret = -E2BIG;
 193                                         goto out;
 194                                 }
 195
 196                                 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
 197                                 if (out_page == NULL) {
 198                                         ret = -ENOMEM;
 199                                         goto out;
 200                                 }
 201                                 cpage_out = kmap(out_page);
 202                                 pages[nr_pages++] = out_page;
 203
 204                                 pg_bytes_left = PAGE_CACHE_SIZE;
 205                                 out_offset = 0;
 206                         }
 207                 }
 208
 209                 /* we're making it bigger, give up */
 210                 if (tot_in > 8192 && tot_in < tot_out) {
 211                         ret = -E2BIG;
 212                         goto out;
 213                 }
 214
 215                 /* we're all done */
 216                 if (tot_in >= len)
 217                         break;
 218
 219                 if (tot_out > max_out)
 220                         break;
 221
 222                 bytes_left = len - tot_in;
 223                 kunmap(in_page);
 224                 page_cache_release(in_page);
 225
 226                 start += PAGE_CACHE_SIZE;
 227                 in_page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
 228                 data_in = kmap(in_page);
 229                 in_len = min(bytes_left, PAGE_CACHE_SIZE);
 230         }
 231
 232         if (tot_out > tot_in)
 233                 goto out;
 234
 235         /* store the size of all chunks of compressed data */
 236         cpage_out = kmap(pages[0]);
 237         write_compress_length(cpage_out, tot_out);
 238
 239         kunmap(pages[0]);
 240
 241         ret = 0;
 242         *total_out = tot_out;
 243         *total_in = tot_in;
 244 out:
 245         *out_pages = nr_pages;
 246         if (out_page)
 247                 kunmap(out_page);
 248
 249         if (in_page) {
 250                 kunmap(in_page);
 251                 page_cache_release(in_page);
 252         }
 253
 254         return ret;
 255 }
 256
 257 static int lzo_decompress_biovec(struct list_head *ws,
 258                                  struct page **pages_in,
 259                                  u64 disk_start,
 260                                  struct bio_vec *bvec,
 261                                  int vcnt,
 262                                  size_t srclen)
 263 {
 264         struct workspace *workspace = list_entry(ws, struct workspace, list);
 265         int ret = 0, ret2;
 266         char *data_in;
 267         unsigned long page_in_index = 0;
 268         unsigned long page_out_index = 0;
 269         unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_CACHE_SIZE);
 270         unsigned long buf_start;
 271         unsigned long buf_offset = 0;
 272         unsigned long bytes;
 273         unsigned long working_bytes;
 274         unsigned long pg_offset;
 275
 276         size_t in_len;
 277         size_t out_len;
 278         unsigned long in_offset;
 279         unsigned long in_page_bytes_left;
 280         unsigned long tot_in;
 281         unsigned long tot_out;
 282         unsigned long tot_len;
 283         char *buf;
 284         bool may_late_unmap, need_unmap;
 285
 286         data_in = kmap(pages_in[0]);
 287         tot_len = read_compress_length(data_in);
 288
 289         tot_in = LZO_LEN;
 290         in_offset = LZO_LEN;
 291         tot_len = min_t(size_t, srclen, tot_len);
 292         in_page_bytes_left = PAGE_CACHE_SIZE - LZO_LEN;
 293
 294         tot_out = 0;
 295         pg_offset = 0;
 296
 297         while (tot_in < tot_len) {
 298                 in_len = read_compress_length(data_in + in_offset);
 299                 in_page_bytes_left -= LZO_LEN;
 300                 in_offset += LZO_LEN;
 301                 tot_in += LZO_LEN;
 302
 303                 tot_in += in_len;
 304                 working_bytes = in_len;
 305                 may_late_unmap = need_unmap = false;
 306
 307                 /* fast path: avoid using the working buffer */
 308                 if (in_page_bytes_left >= in_len) {
 309                         buf = data_in + in_offset;
 310                         bytes = in_len;
 311                         may_late_unmap = true;
 312                         goto cont;
 313                 }
 314
 315                 /* copy bytes from the pages into the working buffer */
 316                 buf = workspace->cbuf;
 317                 buf_offset = 0;
 318                 while (working_bytes) {
 319                         bytes = min(working_bytes, in_page_bytes_left);
 320
 321                         memcpy(buf + buf_offset, data_in + in_offset, bytes);
 322                         buf_offset += bytes;
 323 cont:
 324                         working_bytes -= bytes;
 325                         in_page_bytes_left -= bytes;
 326                         in_offset += bytes;
 327
 328                         /* check if we need to pick another page */
 329                         if ((working_bytes == 0 && in_page_bytes_left < LZO_LEN)
 330                             || in_page_bytes_left == 0) {
 331                                 tot_in += in_page_bytes_left;
 332
 333                                 if (working_bytes == 0 && tot_in >= tot_len)
 334                                         break;
 335
 336                                 if (page_in_index + 1 >= total_pages_in) {
 337                                         ret = -EIO;
 338                                         goto done;
 339                                 }
 340
 341                                 if (may_late_unmap)
 342                                         need_unmap = true;
 343                                 else
 344                                         kunmap(pages_in[page_in_index]);
 345
 346                                 data_in = kmap(pages_in[++page_in_index]);
 347
 348                                 in_page_bytes_left = PAGE_CACHE_SIZE;
 349                                 in_offset = 0;
 350                         }
 351                 }
 352
 353                 out_len = lzo1x_worst_compress(PAGE_CACHE_SIZE);
 354                 ret = lzo1x_decompress_safe(buf, in_len, workspace->buf,
 355                                             &out_len);
 356                 if (need_unmap)
 357                         kunmap(pages_in[page_in_index - 1]);
 358                 if (ret != LZO_E_OK) {
 359                         printk(KERN_WARNING "BTRFS: decompress failed\n");
 360                         ret = -EIO;
 361                         break;
 362                 }
 363
 364                 buf_start = tot_out;
 365                 tot_out += out_len;
 366
 367                 ret2 = btrfs_decompress_buf2page(workspace->buf, buf_start,
 368                                                  tot_out, disk_start,
 369                                                  bvec, vcnt,
 370                                                  &page_out_index, &pg_offset);
 371                 if (ret2 == 0)
 372                         break;
 373         }
 374 done:
 375         kunmap(pages_in[page_in_index]);
 376         if (!ret)
 377                 btrfs_clear_biovec_end(bvec, vcnt, page_out_index, pg_offset);
 378         return ret;
 379 }
 380
 381 static int lzo_decompress(struct list_head *ws, unsigned char *data_in,
 382                           struct page *dest_page,
 383                           unsigned long start_byte,
 384                           size_t srclen, size_t destlen)
 385 {
 386         struct workspace *workspace = list_entry(ws, struct workspace, list);
 387         size_t in_len;
 388         size_t out_len;
 389         size_t tot_len;
 390         int ret = 0;
 391         char *kaddr;
 392         unsigned long bytes;
 393
 394         BUG_ON(srclen < LZO_LEN);
 395
 396         tot_len = read_compress_length(data_in);
 397         data_in += LZO_LEN;
 398
 399         in_len = read_compress_length(data_in);
 400         data_in += LZO_LEN;
 401
 402         out_len = PAGE_CACHE_SIZE;
 403         ret = lzo1x_decompress_safe(data_in, in_len, workspace->buf, &out_len);
 404         if (ret != LZO_E_OK) {
 405                 printk(KERN_WARNING "BTRFS: decompress failed!\n");
 406                 ret = -EIO;
 407                 goto out;
 408         }
 409
 410         if (out_len < start_byte) {
 411                 ret = -EIO;
 412                 goto out;
 413         }
 414
 415         /*
 416          * the caller is already checking against PAGE_SIZE, but lets
 417          * move this check closer to the memcpy/memset
 418          */
 419         destlen = min_t(unsigned long, destlen, PAGE_SIZE);
 420         bytes = min_t(unsigned long, destlen, out_len - start_byte);
 421
 422         kaddr = kmap_atomic(dest_page);
 423         memcpy(kaddr, workspace->buf + start_byte, bytes);
 424
 425         /*
 426          * btrfs_getblock is doing a zero on the tail of the page too,
 427          * but this will cover anything missing from the decompressed
 428          * data.
 429          */
 430         if (bytes < destlen)
 431                 memset(kaddr+bytes, 0, destlen-bytes);
 432         kunmap_atomic(kaddr);
 433 out:
 434         return ret;
 435 }
 436
 437 const struct btrfs_compress_op btrfs_lzo_compress = {
 438         .alloc_workspace        = lzo_alloc_workspace,
 439         .free_workspace         = lzo_free_workspace,
 440         .compress_pages         = lzo_compress_pages,
 441         .decompress_biovec      = lzo_decompress_biovec,
 442         .decompress             = lzo_decompress,
 443 };