| 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 strm; |
| 37 | char *buf; |
| 38 | struct list_head list; |
| 39 | }; |
| 40 | |
| 41 | static void zlib_free_workspace(struct list_head *ws) |
| 42 | { |
| 43 | struct workspace *workspace = list_entry(ws, struct workspace, list); |
| 44 | |
| 45 | vfree(workspace->strm.workspace); |
| 46 | kfree(workspace->buf); |
| 47 | kfree(workspace); |
| 48 | } |
| 49 | |
| 50 | static struct list_head *zlib_alloc_workspace(void) |
| 51 | { |
| 52 | struct workspace *workspace; |
| 53 | int workspacesize; |
| 54 | |
| 55 | workspace = kzalloc(sizeof(*workspace), GFP_NOFS); |
| 56 | if (!workspace) |
| 57 | return ERR_PTR(-ENOMEM); |
| 58 | |
| 59 | workspacesize = max(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL), |
| 60 | zlib_inflate_workspacesize()); |
| 61 | workspace->strm.workspace = vmalloc(workspacesize); |
| 62 | workspace->buf = kmalloc(PAGE_SIZE, GFP_NOFS); |
| 63 | if (!workspace->strm.workspace || !workspace->buf) |
| 64 | goto fail; |
| 65 | |
| 66 | INIT_LIST_HEAD(&workspace->list); |
| 67 | |
| 68 | return &workspace->list; |
| 69 | fail: |
| 70 | zlib_free_workspace(&workspace->list); |
| 71 | return ERR_PTR(-ENOMEM); |
| 72 | } |
| 73 | |
| 74 | static int zlib_compress_pages(struct list_head *ws, |
| 75 | struct address_space *mapping, |
| 76 | u64 start, unsigned long len, |
| 77 | struct page **pages, |
| 78 | unsigned long nr_dest_pages, |
| 79 | unsigned long *out_pages, |
| 80 | unsigned long *total_in, |
| 81 | unsigned long *total_out, |
| 82 | unsigned long max_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 | |
| 93 | *out_pages = 0; |
| 94 | *total_out = 0; |
| 95 | *total_in = 0; |
| 96 | |
| 97 | if (Z_OK != zlib_deflateInit(&workspace->strm, 3)) { |
| 98 | printk(KERN_WARNING "BTRFS: deflateInit failed\n"); |
| 99 | ret = -EIO; |
| 100 | goto out; |
| 101 | } |
| 102 | |
| 103 | workspace->strm.total_in = 0; |
| 104 | workspace->strm.total_out = 0; |
| 105 | |
| 106 | in_page = find_get_page(mapping, start >> PAGE_SHIFT); |
| 107 | data_in = kmap(in_page); |
| 108 | |
| 109 | out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); |
| 110 | if (out_page == NULL) { |
| 111 | ret = -ENOMEM; |
| 112 | goto out; |
| 113 | } |
| 114 | cpage_out = kmap(out_page); |
| 115 | pages[0] = out_page; |
| 116 | nr_pages = 1; |
| 117 | |
| 118 | workspace->strm.next_in = data_in; |
| 119 | workspace->strm.next_out = cpage_out; |
| 120 | workspace->strm.avail_out = PAGE_SIZE; |
| 121 | workspace->strm.avail_in = min(len, PAGE_SIZE); |
| 122 | |
| 123 | while (workspace->strm.total_in < len) { |
| 124 | ret = zlib_deflate(&workspace->strm, Z_SYNC_FLUSH); |
| 125 | if (ret != Z_OK) { |
| 126 | printk(KERN_DEBUG "BTRFS: deflate in loop returned %d\n", |
| 127 | ret); |
| 128 | zlib_deflateEnd(&workspace->strm); |
| 129 | ret = -EIO; |
| 130 | goto out; |
| 131 | } |
| 132 | |
| 133 | /* we're making it bigger, give up */ |
| 134 | if (workspace->strm.total_in > 8192 && |
| 135 | workspace->strm.total_in < |
| 136 | workspace->strm.total_out) { |
| 137 | ret = -E2BIG; |
| 138 | goto out; |
| 139 | } |
| 140 | /* we need another page for writing out. Test this |
| 141 | * before the total_in so we will pull in a new page for |
| 142 | * the stream end if required |
| 143 | */ |
| 144 | if (workspace->strm.avail_out == 0) { |
| 145 | kunmap(out_page); |
| 146 | if (nr_pages == nr_dest_pages) { |
| 147 | out_page = NULL; |
| 148 | ret = -E2BIG; |
| 149 | goto out; |
| 150 | } |
| 151 | out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); |
| 152 | if (out_page == NULL) { |
| 153 | ret = -ENOMEM; |
| 154 | goto out; |
| 155 | } |
| 156 | cpage_out = kmap(out_page); |
| 157 | pages[nr_pages] = out_page; |
| 158 | nr_pages++; |
| 159 | workspace->strm.avail_out = PAGE_SIZE; |
| 160 | workspace->strm.next_out = cpage_out; |
| 161 | } |
| 162 | /* we're all done */ |
| 163 | if (workspace->strm.total_in >= len) |
| 164 | break; |
| 165 | |
| 166 | /* we've read in a full page, get a new one */ |
| 167 | if (workspace->strm.avail_in == 0) { |
| 168 | if (workspace->strm.total_out > max_out) |
| 169 | break; |
| 170 | |
| 171 | bytes_left = len - workspace->strm.total_in; |
| 172 | kunmap(in_page); |
| 173 | put_page(in_page); |
| 174 | |
| 175 | start += PAGE_SIZE; |
| 176 | in_page = find_get_page(mapping, |
| 177 | start >> PAGE_SHIFT); |
| 178 | data_in = kmap(in_page); |
| 179 | workspace->strm.avail_in = min(bytes_left, |
| 180 | PAGE_SIZE); |
| 181 | workspace->strm.next_in = data_in; |
| 182 | } |
| 183 | } |
| 184 | workspace->strm.avail_in = 0; |
| 185 | ret = zlib_deflate(&workspace->strm, Z_FINISH); |
| 186 | zlib_deflateEnd(&workspace->strm); |
| 187 | |
| 188 | if (ret != Z_STREAM_END) { |
| 189 | ret = -EIO; |
| 190 | goto out; |
| 191 | } |
| 192 | |
| 193 | if (workspace->strm.total_out >= workspace->strm.total_in) { |
| 194 | ret = -E2BIG; |
| 195 | goto out; |
| 196 | } |
| 197 | |
| 198 | ret = 0; |
| 199 | *total_out = workspace->strm.total_out; |
| 200 | *total_in = workspace->strm.total_in; |
| 201 | out: |
| 202 | *out_pages = nr_pages; |
| 203 | if (out_page) |
| 204 | kunmap(out_page); |
| 205 | |
| 206 | if (in_page) { |
| 207 | kunmap(in_page); |
| 208 | put_page(in_page); |
| 209 | } |
| 210 | return ret; |
| 211 | } |
| 212 | |
| 213 | static int zlib_decompress_biovec(struct list_head *ws, struct page **pages_in, |
| 214 | u64 disk_start, |
| 215 | struct bio_vec *bvec, |
| 216 | int vcnt, |
| 217 | size_t srclen) |
| 218 | { |
| 219 | struct workspace *workspace = list_entry(ws, struct workspace, list); |
| 220 | int ret = 0, ret2; |
| 221 | int wbits = MAX_WBITS; |
| 222 | char *data_in; |
| 223 | size_t total_out = 0; |
| 224 | unsigned long page_in_index = 0; |
| 225 | unsigned long page_out_index = 0; |
| 226 | unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE); |
| 227 | unsigned long buf_start; |
| 228 | unsigned long pg_offset; |
| 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 | pg_offset = 0; |
| 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 | printk(KERN_WARNING "BTRFS: inflateInit failed\n"); |
| 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 | bvec, vcnt, |
| 270 | &page_out_index, &pg_offset); |
| 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 | btrfs_clear_biovec_end(bvec, vcnt, page_out_index, pg_offset); |
| 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 | printk(KERN_WARNING "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 | const struct btrfs_compress_op btrfs_zlib_compress = { |
| 407 | .alloc_workspace = zlib_alloc_workspace, |
| 408 | .free_workspace = zlib_free_workspace, |
| 409 | .compress_pages = zlib_compress_pages, |
| 410 | .decompress_biovec = zlib_decompress_biovec, |
| 411 | .decompress = zlib_decompress, |
| 412 | }; |