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 }; 41 42 static void zlib_free_workspace(struct list_head *ws) 43 { 44 struct workspace *workspace = list_entry(ws, struct workspace, list); 45 46 kvfree(workspace->strm.workspace); 47 kfree(workspace->buf); 48 kfree(workspace); 49 } 50 51 static struct list_head *zlib_alloc_workspace(void) 52 { 53 struct workspace *workspace; 54 int workspacesize; 55 56 workspace = kzalloc(sizeof(*workspace), GFP_KERNEL); 57 if (!workspace) 58 return ERR_PTR(-ENOMEM); 59 60 workspacesize = max(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL), 61 zlib_inflate_workspacesize()); 62 workspace->strm.workspace = kvmalloc(workspacesize, GFP_KERNEL); 63 workspace->buf = kmalloc(PAGE_SIZE, GFP_KERNEL); 64 if (!workspace->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, 78 struct page **pages, 79 unsigned long *out_pages, 80 unsigned long *total_in, 81 unsigned long *total_out) 82 { 83 struct workspace *workspace = list_entry(ws, struct workspace, list); 84 int ret; 85 char *data_in; 86 char *cpage_out; 87 int nr_pages = 0; 88 struct page *in_page = NULL; 89 struct page *out_page = NULL; 90 unsigned long bytes_left; 91 unsigned long len = *total_out; 92 unsigned long nr_dest_pages = *out_pages; 93 const unsigned long max_out = nr_dest_pages * PAGE_SIZE; 94 95 *out_pages = 0; 96 *total_out = 0; 97 *total_in = 0; 98 99 if (Z_OK != zlib_deflateInit(&workspace->strm, 3)) { 100 pr_warn("BTRFS: deflateInit failed\n"); 101 ret = -EIO; 102 goto out; 103 } 104 105 workspace->strm.total_in = 0; 106 workspace->strm.total_out = 0; 107 108 in_page = find_get_page(mapping, start >> PAGE_SHIFT); 109 data_in = kmap(in_page); 110 111 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); 112 if (out_page == NULL) { 113 ret = -ENOMEM; 114 goto out; 115 } 116 cpage_out = kmap(out_page); 117 pages[0] = out_page; 118 nr_pages = 1; 119 120 workspace->strm.next_in = data_in; 121 workspace->strm.next_out = cpage_out; 122 workspace->strm.avail_out = PAGE_SIZE; 123 workspace->strm.avail_in = min(len, PAGE_SIZE); 124 125 while (workspace->strm.total_in < len) { 126 ret = zlib_deflate(&workspace->strm, Z_SYNC_FLUSH); 127 if (ret != Z_OK) { 128 pr_debug("BTRFS: deflate in loop returned %d\n", 129 ret); 130 zlib_deflateEnd(&workspace->strm); 131 ret = -EIO; 132 goto out; 133 } 134 135 /* we're making it bigger, give up */ 136 if (workspace->strm.total_in > 8192 && 137 workspace->strm.total_in < 138 workspace->strm.total_out) { 139 ret = -E2BIG; 140 goto out; 141 } 142 /* we need another page for writing out. Test this 143 * before the total_in so we will pull in a new page for 144 * the stream end if required 145 */ 146 if (workspace->strm.avail_out == 0) { 147 kunmap(out_page); 148 if (nr_pages == nr_dest_pages) { 149 out_page = NULL; 150 ret = -E2BIG; 151 goto out; 152 } 153 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); 154 if (out_page == NULL) { 155 ret = -ENOMEM; 156 goto out; 157 } 158 cpage_out = kmap(out_page); 159 pages[nr_pages] = out_page; 160 nr_pages++; 161 workspace->strm.avail_out = PAGE_SIZE; 162 workspace->strm.next_out = cpage_out; 163 } 164 /* we're all done */ 165 if (workspace->strm.total_in >= len) 166 break; 167 168 /* we've read in a full page, get a new one */ 169 if (workspace->strm.avail_in == 0) { 170 if (workspace->strm.total_out > max_out) 171 break; 172 173 bytes_left = len - workspace->strm.total_in; 174 kunmap(in_page); 175 put_page(in_page); 176 177 start += PAGE_SIZE; 178 in_page = find_get_page(mapping, 179 start >> PAGE_SHIFT); 180 data_in = kmap(in_page); 181 workspace->strm.avail_in = min(bytes_left, 182 PAGE_SIZE); 183 workspace->strm.next_in = data_in; 184 } 185 } 186 workspace->strm.avail_in = 0; 187 ret = zlib_deflate(&workspace->strm, Z_FINISH); 188 zlib_deflateEnd(&workspace->strm); 189 190 if (ret != Z_STREAM_END) { 191 ret = -EIO; 192 goto out; 193 } 194 195 if (workspace->strm.total_out >= workspace->strm.total_in) { 196 ret = -E2BIG; 197 goto out; 198 } 199 200 ret = 0; 201 *total_out = workspace->strm.total_out; 202 *total_in = workspace->strm.total_in; 203 out: 204 *out_pages = nr_pages; 205 if (out_page) 206 kunmap(out_page); 207 208 if (in_page) { 209 kunmap(in_page); 210 put_page(in_page); 211 } 212 return ret; 213 } 214 215 static int zlib_decompress_bio(struct list_head *ws, struct compressed_bio *cb) 216 { 217 struct workspace *workspace = list_entry(ws, struct workspace, list); 218 int ret = 0, ret2; 219 int wbits = MAX_WBITS; 220 char *data_in; 221 size_t total_out = 0; 222 unsigned long page_in_index = 0; 223 size_t srclen = cb->compressed_len; 224 unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE); 225 unsigned long buf_start; 226 struct page **pages_in = cb->compressed_pages; 227 u64 disk_start = cb->start; 228 struct bio *orig_bio = cb->orig_bio; 229 230 data_in = kmap(pages_in[page_in_index]); 231 workspace->strm.next_in = data_in; 232 workspace->strm.avail_in = min_t(size_t, srclen, PAGE_SIZE); 233 workspace->strm.total_in = 0; 234 235 workspace->strm.total_out = 0; 236 workspace->strm.next_out = workspace->buf; 237 workspace->strm.avail_out = PAGE_SIZE; 238 239 /* If it's deflate, and it's got no preset dictionary, then 240 we can tell zlib to skip the adler32 check. */ 241 if (srclen > 2 && !(data_in[1] & PRESET_DICT) && 242 ((data_in[0] & 0x0f) == Z_DEFLATED) && 243 !(((data_in[0]<<8) + data_in[1]) % 31)) { 244 245 wbits = -((data_in[0] >> 4) + 8); 246 workspace->strm.next_in += 2; 247 workspace->strm.avail_in -= 2; 248 } 249 250 if (Z_OK != zlib_inflateInit2(&workspace->strm, wbits)) { 251 pr_warn("BTRFS: inflateInit failed\n"); 252 kunmap(pages_in[page_in_index]); 253 return -EIO; 254 } 255 while (workspace->strm.total_in < srclen) { 256 ret = zlib_inflate(&workspace->strm, Z_NO_FLUSH); 257 if (ret != Z_OK && ret != Z_STREAM_END) 258 break; 259 260 buf_start = total_out; 261 total_out = workspace->strm.total_out; 262 263 /* we didn't make progress in this inflate call, we're done */ 264 if (buf_start == total_out) 265 break; 266 267 ret2 = btrfs_decompress_buf2page(workspace->buf, buf_start, 268 total_out, disk_start, 269 orig_bio); 270 if (ret2 == 0) { 271 ret = 0; 272 goto done; 273 } 274 275 workspace->strm.next_out = workspace->buf; 276 workspace->strm.avail_out = PAGE_SIZE; 277 278 if (workspace->strm.avail_in == 0) { 279 unsigned long tmp; 280 kunmap(pages_in[page_in_index]); 281 page_in_index++; 282 if (page_in_index >= total_pages_in) { 283 data_in = NULL; 284 break; 285 } 286 data_in = kmap(pages_in[page_in_index]); 287 workspace->strm.next_in = data_in; 288 tmp = srclen - workspace->strm.total_in; 289 workspace->strm.avail_in = min(tmp, 290 PAGE_SIZE); 291 } 292 } 293 if (ret != Z_STREAM_END) 294 ret = -EIO; 295 else 296 ret = 0; 297 done: 298 zlib_inflateEnd(&workspace->strm); 299 if (data_in) 300 kunmap(pages_in[page_in_index]); 301 if (!ret) 302 zero_fill_bio(orig_bio); 303 return ret; 304 } 305 306 static int zlib_decompress(struct list_head *ws, unsigned char *data_in, 307 struct page *dest_page, 308 unsigned long start_byte, 309 size_t srclen, size_t destlen) 310 { 311 struct workspace *workspace = list_entry(ws, struct workspace, list); 312 int ret = 0; 313 int wbits = MAX_WBITS; 314 unsigned long bytes_left; 315 unsigned long total_out = 0; 316 unsigned long pg_offset = 0; 317 char *kaddr; 318 319 destlen = min_t(unsigned long, destlen, PAGE_SIZE); 320 bytes_left = destlen; 321 322 workspace->strm.next_in = data_in; 323 workspace->strm.avail_in = srclen; 324 workspace->strm.total_in = 0; 325 326 workspace->strm.next_out = workspace->buf; 327 workspace->strm.avail_out = PAGE_SIZE; 328 workspace->strm.total_out = 0; 329 /* If it's deflate, and it's got no preset dictionary, then 330 we can tell zlib to skip the adler32 check. */ 331 if (srclen > 2 && !(data_in[1] & PRESET_DICT) && 332 ((data_in[0] & 0x0f) == Z_DEFLATED) && 333 !(((data_in[0]<<8) + data_in[1]) % 31)) { 334 335 wbits = -((data_in[0] >> 4) + 8); 336 workspace->strm.next_in += 2; 337 workspace->strm.avail_in -= 2; 338 } 339 340 if (Z_OK != zlib_inflateInit2(&workspace->strm, wbits)) { 341 pr_warn("BTRFS: inflateInit failed\n"); 342 return -EIO; 343 } 344 345 while (bytes_left > 0) { 346 unsigned long buf_start; 347 unsigned long buf_offset; 348 unsigned long bytes; 349 350 ret = zlib_inflate(&workspace->strm, Z_NO_FLUSH); 351 if (ret != Z_OK && ret != Z_STREAM_END) 352 break; 353 354 buf_start = total_out; 355 total_out = workspace->strm.total_out; 356 357 if (total_out == buf_start) { 358 ret = -EIO; 359 break; 360 } 361 362 if (total_out <= start_byte) 363 goto next; 364 365 if (total_out > start_byte && buf_start < start_byte) 366 buf_offset = start_byte - buf_start; 367 else 368 buf_offset = 0; 369 370 bytes = min(PAGE_SIZE - pg_offset, 371 PAGE_SIZE - buf_offset); 372 bytes = min(bytes, bytes_left); 373 374 kaddr = kmap_atomic(dest_page); 375 memcpy(kaddr + pg_offset, workspace->buf + buf_offset, bytes); 376 kunmap_atomic(kaddr); 377 378 pg_offset += bytes; 379 bytes_left -= bytes; 380 next: 381 workspace->strm.next_out = workspace->buf; 382 workspace->strm.avail_out = PAGE_SIZE; 383 } 384 385 if (ret != Z_STREAM_END && bytes_left != 0) 386 ret = -EIO; 387 else 388 ret = 0; 389 390 zlib_inflateEnd(&workspace->strm); 391 392 /* 393 * this should only happen if zlib returned fewer bytes than we 394 * expected. btrfs_get_block is responsible for zeroing from the 395 * end of the inline extent (destlen) to the end of the page 396 */ 397 if (pg_offset < destlen) { 398 kaddr = kmap_atomic(dest_page); 399 memset(kaddr + pg_offset, 0, destlen - pg_offset); 400 kunmap_atomic(kaddr); 401 } 402 return ret; 403 } 404 405 const struct btrfs_compress_op btrfs_zlib_compress = { 406 .alloc_workspace = zlib_alloc_workspace, 407 .free_workspace = zlib_free_workspace, 408 .compress_pages = zlib_compress_pages, 409 .decompress_bio = zlib_decompress_bio, 410 .decompress = zlib_decompress, 411 }; 412