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/mm.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 <linux/refcount.h> 34 #include "compression.h" 35 36 struct workspace { 37 z_stream strm; 38 char *buf; 39 struct list_head list; 40 int level; 41 }; 42 43 static void zlib_free_workspace(struct list_head *ws) 44 { 45 struct workspace *workspace = list_entry(ws, struct workspace, list); 46 47 kvfree(workspace->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 int workspacesize; 56 57 workspace = kzalloc(sizeof(*workspace), GFP_KERNEL); 58 if (!workspace) 59 return ERR_PTR(-ENOMEM); 60 61 workspacesize = max(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL), 62 zlib_inflate_workspacesize()); 63 workspace->strm.workspace = kvmalloc(workspacesize, GFP_KERNEL); 64 workspace->buf = kmalloc(PAGE_SIZE, GFP_KERNEL); 65 if (!workspace->strm.workspace || !workspace->buf) 66 goto fail; 67 68 INIT_LIST_HEAD(&workspace->list); 69 70 return &workspace->list; 71 fail: 72 zlib_free_workspace(&workspace->list); 73 return ERR_PTR(-ENOMEM); 74 } 75 76 static int zlib_compress_pages(struct list_head *ws, 77 struct address_space *mapping, 78 u64 start, 79 struct page **pages, 80 unsigned long *out_pages, 81 unsigned long *total_in, 82 unsigned long *total_out) 83 { 84 struct workspace *workspace = list_entry(ws, struct workspace, list); 85 int ret; 86 char *data_in; 87 char *cpage_out; 88 int nr_pages = 0; 89 struct page *in_page = NULL; 90 struct page *out_page = NULL; 91 unsigned long bytes_left; 92 unsigned long len = *total_out; 93 unsigned long nr_dest_pages = *out_pages; 94 const unsigned long max_out = nr_dest_pages * PAGE_SIZE; 95 96 *out_pages = 0; 97 *total_out = 0; 98 *total_in = 0; 99 100 if (Z_OK != zlib_deflateInit(&workspace->strm, workspace->level)) { 101 pr_warn("BTRFS: deflateInit failed\n"); 102 ret = -EIO; 103 goto out; 104 } 105 106 workspace->strm.total_in = 0; 107 workspace->strm.total_out = 0; 108 109 in_page = find_get_page(mapping, start >> PAGE_SHIFT); 110 data_in = kmap(in_page); 111 112 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); 113 if (out_page == NULL) { 114 ret = -ENOMEM; 115 goto out; 116 } 117 cpage_out = kmap(out_page); 118 pages[0] = out_page; 119 nr_pages = 1; 120 121 workspace->strm.next_in = data_in; 122 workspace->strm.next_out = cpage_out; 123 workspace->strm.avail_out = PAGE_SIZE; 124 workspace->strm.avail_in = min(len, PAGE_SIZE); 125 126 while (workspace->strm.total_in < len) { 127 ret = zlib_deflate(&workspace->strm, Z_SYNC_FLUSH); 128 if (ret != Z_OK) { 129 pr_debug("BTRFS: deflate in loop returned %d\n", 130 ret); 131 zlib_deflateEnd(&workspace->strm); 132 ret = -EIO; 133 goto out; 134 } 135 136 /* we're making it bigger, give up */ 137 if (workspace->strm.total_in > 8192 && 138 workspace->strm.total_in < 139 workspace->strm.total_out) { 140 ret = -E2BIG; 141 goto out; 142 } 143 /* we need another page for writing out. Test this 144 * before the total_in so we will pull in a new page for 145 * the stream end if required 146 */ 147 if (workspace->strm.avail_out == 0) { 148 kunmap(out_page); 149 if (nr_pages == nr_dest_pages) { 150 out_page = NULL; 151 ret = -E2BIG; 152 goto out; 153 } 154 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); 155 if (out_page == NULL) { 156 ret = -ENOMEM; 157 goto out; 158 } 159 cpage_out = kmap(out_page); 160 pages[nr_pages] = out_page; 161 nr_pages++; 162 workspace->strm.avail_out = PAGE_SIZE; 163 workspace->strm.next_out = cpage_out; 164 } 165 /* we're all done */ 166 if (workspace->strm.total_in >= len) 167 break; 168 169 /* we've read in a full page, get a new one */ 170 if (workspace->strm.avail_in == 0) { 171 if (workspace->strm.total_out > max_out) 172 break; 173 174 bytes_left = len - workspace->strm.total_in; 175 kunmap(in_page); 176 put_page(in_page); 177 178 start += PAGE_SIZE; 179 in_page = find_get_page(mapping, 180 start >> PAGE_SHIFT); 181 data_in = kmap(in_page); 182 workspace->strm.avail_in = min(bytes_left, 183 PAGE_SIZE); 184 workspace->strm.next_in = data_in; 185 } 186 } 187 workspace->strm.avail_in = 0; 188 ret = zlib_deflate(&workspace->strm, Z_FINISH); 189 zlib_deflateEnd(&workspace->strm); 190 191 if (ret != Z_STREAM_END) { 192 ret = -EIO; 193 goto out; 194 } 195 196 if (workspace->strm.total_out >= workspace->strm.total_in) { 197 ret = -E2BIG; 198 goto out; 199 } 200 201 ret = 0; 202 *total_out = workspace->strm.total_out; 203 *total_in = workspace->strm.total_in; 204 out: 205 *out_pages = nr_pages; 206 if (out_page) 207 kunmap(out_page); 208 209 if (in_page) { 210 kunmap(in_page); 211 put_page(in_page); 212 } 213 return ret; 214 } 215 216 static int zlib_decompress_bio(struct list_head *ws, struct compressed_bio *cb) 217 { 218 struct workspace *workspace = list_entry(ws, struct workspace, list); 219 int ret = 0, ret2; 220 int wbits = MAX_WBITS; 221 char *data_in; 222 size_t total_out = 0; 223 unsigned long page_in_index = 0; 224 size_t srclen = cb->compressed_len; 225 unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE); 226 unsigned long buf_start; 227 struct page **pages_in = cb->compressed_pages; 228 u64 disk_start = cb->start; 229 struct bio *orig_bio = cb->orig_bio; 230 231 data_in = kmap(pages_in[page_in_index]); 232 workspace->strm.next_in = data_in; 233 workspace->strm.avail_in = min_t(size_t, srclen, PAGE_SIZE); 234 workspace->strm.total_in = 0; 235 236 workspace->strm.total_out = 0; 237 workspace->strm.next_out = workspace->buf; 238 workspace->strm.avail_out = PAGE_SIZE; 239 240 /* If it's deflate, and it's got no preset dictionary, then 241 we can tell zlib to skip the adler32 check. */ 242 if (srclen > 2 && !(data_in[1] & PRESET_DICT) && 243 ((data_in[0] & 0x0f) == Z_DEFLATED) && 244 !(((data_in[0]<<8) + data_in[1]) % 31)) { 245 246 wbits = -((data_in[0] >> 4) + 8); 247 workspace->strm.next_in += 2; 248 workspace->strm.avail_in -= 2; 249 } 250 251 if (Z_OK != zlib_inflateInit2(&workspace->strm, wbits)) { 252 pr_warn("BTRFS: inflateInit failed\n"); 253 kunmap(pages_in[page_in_index]); 254 return -EIO; 255 } 256 while (workspace->strm.total_in < srclen) { 257 ret = zlib_inflate(&workspace->strm, Z_NO_FLUSH); 258 if (ret != Z_OK && ret != Z_STREAM_END) 259 break; 260 261 buf_start = total_out; 262 total_out = workspace->strm.total_out; 263 264 /* we didn't make progress in this inflate call, we're done */ 265 if (buf_start == total_out) 266 break; 267 268 ret2 = btrfs_decompress_buf2page(workspace->buf, buf_start, 269 total_out, disk_start, 270 orig_bio); 271 if (ret2 == 0) { 272 ret = 0; 273 goto done; 274 } 275 276 workspace->strm.next_out = workspace->buf; 277 workspace->strm.avail_out = PAGE_SIZE; 278 279 if (workspace->strm.avail_in == 0) { 280 unsigned long tmp; 281 kunmap(pages_in[page_in_index]); 282 page_in_index++; 283 if (page_in_index >= total_pages_in) { 284 data_in = NULL; 285 break; 286 } 287 data_in = kmap(pages_in[page_in_index]); 288 workspace->strm.next_in = data_in; 289 tmp = srclen - workspace->strm.total_in; 290 workspace->strm.avail_in = min(tmp, 291 PAGE_SIZE); 292 } 293 } 294 if (ret != Z_STREAM_END) 295 ret = -EIO; 296 else 297 ret = 0; 298 done: 299 zlib_inflateEnd(&workspace->strm); 300 if (data_in) 301 kunmap(pages_in[page_in_index]); 302 if (!ret) 303 zero_fill_bio(orig_bio); 304 return ret; 305 } 306 307 static int zlib_decompress(struct list_head *ws, unsigned char *data_in, 308 struct page *dest_page, 309 unsigned long start_byte, 310 size_t srclen, size_t destlen) 311 { 312 struct workspace *workspace = list_entry(ws, struct workspace, list); 313 int ret = 0; 314 int wbits = MAX_WBITS; 315 unsigned long bytes_left; 316 unsigned long total_out = 0; 317 unsigned long pg_offset = 0; 318 char *kaddr; 319 320 destlen = min_t(unsigned long, destlen, PAGE_SIZE); 321 bytes_left = destlen; 322 323 workspace->strm.next_in = data_in; 324 workspace->strm.avail_in = srclen; 325 workspace->strm.total_in = 0; 326 327 workspace->strm.next_out = workspace->buf; 328 workspace->strm.avail_out = PAGE_SIZE; 329 workspace->strm.total_out = 0; 330 /* If it's deflate, and it's got no preset dictionary, then 331 we can tell zlib to skip the adler32 check. */ 332 if (srclen > 2 && !(data_in[1] & PRESET_DICT) && 333 ((data_in[0] & 0x0f) == Z_DEFLATED) && 334 !(((data_in[0]<<8) + data_in[1]) % 31)) { 335 336 wbits = -((data_in[0] >> 4) + 8); 337 workspace->strm.next_in += 2; 338 workspace->strm.avail_in -= 2; 339 } 340 341 if (Z_OK != zlib_inflateInit2(&workspace->strm, wbits)) { 342 pr_warn("BTRFS: inflateInit failed\n"); 343 return -EIO; 344 } 345 346 while (bytes_left > 0) { 347 unsigned long buf_start; 348 unsigned long buf_offset; 349 unsigned long bytes; 350 351 ret = zlib_inflate(&workspace->strm, Z_NO_FLUSH); 352 if (ret != Z_OK && ret != Z_STREAM_END) 353 break; 354 355 buf_start = total_out; 356 total_out = workspace->strm.total_out; 357 358 if (total_out == buf_start) { 359 ret = -EIO; 360 break; 361 } 362 363 if (total_out <= start_byte) 364 goto next; 365 366 if (total_out > start_byte && buf_start < start_byte) 367 buf_offset = start_byte - buf_start; 368 else 369 buf_offset = 0; 370 371 bytes = min(PAGE_SIZE - pg_offset, 372 PAGE_SIZE - buf_offset); 373 bytes = min(bytes, bytes_left); 374 375 kaddr = kmap_atomic(dest_page); 376 memcpy(kaddr + pg_offset, workspace->buf + buf_offset, bytes); 377 kunmap_atomic(kaddr); 378 379 pg_offset += bytes; 380 bytes_left -= bytes; 381 next: 382 workspace->strm.next_out = workspace->buf; 383 workspace->strm.avail_out = PAGE_SIZE; 384 } 385 386 if (ret != Z_STREAM_END && bytes_left != 0) 387 ret = -EIO; 388 else 389 ret = 0; 390 391 zlib_inflateEnd(&workspace->strm); 392 393 /* 394 * this should only happen if zlib returned fewer bytes than we 395 * expected. btrfs_get_block is responsible for zeroing from the 396 * end of the inline extent (destlen) to the end of the page 397 */ 398 if (pg_offset < destlen) { 399 kaddr = kmap_atomic(dest_page); 400 memset(kaddr + pg_offset, 0, destlen - pg_offset); 401 kunmap_atomic(kaddr); 402 } 403 return ret; 404 } 405 406 static void zlib_set_level(struct list_head *ws, unsigned int type) 407 { 408 struct workspace *workspace = list_entry(ws, struct workspace, list); 409 unsigned level = (type & 0xF0) >> 4; 410 411 if (level > 9) 412 level = 9; 413 414 workspace->level = level > 0 ? level : 3; 415 } 416 417 const struct btrfs_compress_op btrfs_zlib_compress = { 418 .alloc_workspace = zlib_alloc_workspace, 419 .free_workspace = zlib_free_workspace, 420 .compress_pages = zlib_compress_pages, 421 .decompress_bio = zlib_decompress_bio, 422 .decompress = zlib_decompress, 423 .set_level = zlib_set_level, 424 }; 425