xref: /linux/fs/btrfs/zlib.c (revision 905e46acd3272d04566fec49afbd7ad9e2ed9ae3)
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,
77 			       struct page **pages,
78 			       unsigned long *out_pages,
79 			       unsigned long *total_in,
80 			       unsigned long *total_out)
81 {
82 	struct workspace *workspace = list_entry(ws, struct workspace, list);
83 	int ret;
84 	char *data_in;
85 	char *cpage_out;
86 	int nr_pages = 0;
87 	struct page *in_page = NULL;
88 	struct page *out_page = NULL;
89 	unsigned long bytes_left;
90 	unsigned long len = *total_out;
91 	unsigned long nr_dest_pages = *out_pages;
92 	const unsigned long max_out = nr_dest_pages * PAGE_SIZE;
93 
94 	*out_pages = 0;
95 	*total_out = 0;
96 	*total_in = 0;
97 
98 	if (Z_OK != zlib_deflateInit(&workspace->strm, 3)) {
99 		pr_warn("BTRFS: deflateInit failed\n");
100 		ret = -EIO;
101 		goto out;
102 	}
103 
104 	workspace->strm.total_in = 0;
105 	workspace->strm.total_out = 0;
106 
107 	in_page = find_get_page(mapping, start >> PAGE_SHIFT);
108 	data_in = kmap(in_page);
109 
110 	out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
111 	if (out_page == NULL) {
112 		ret = -ENOMEM;
113 		goto out;
114 	}
115 	cpage_out = kmap(out_page);
116 	pages[0] = out_page;
117 	nr_pages = 1;
118 
119 	workspace->strm.next_in = data_in;
120 	workspace->strm.next_out = cpage_out;
121 	workspace->strm.avail_out = PAGE_SIZE;
122 	workspace->strm.avail_in = min(len, PAGE_SIZE);
123 
124 	while (workspace->strm.total_in < len) {
125 		ret = zlib_deflate(&workspace->strm, Z_SYNC_FLUSH);
126 		if (ret != Z_OK) {
127 			pr_debug("BTRFS: deflate in loop returned %d\n",
128 			       ret);
129 			zlib_deflateEnd(&workspace->strm);
130 			ret = -EIO;
131 			goto out;
132 		}
133 
134 		/* we're making it bigger, give up */
135 		if (workspace->strm.total_in > 8192 &&
136 		    workspace->strm.total_in <
137 		    workspace->strm.total_out) {
138 			ret = -E2BIG;
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->strm.avail_out == 0) {
146 			kunmap(out_page);
147 			if (nr_pages == nr_dest_pages) {
148 				out_page = NULL;
149 				ret = -E2BIG;
150 				goto out;
151 			}
152 			out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
153 			if (out_page == NULL) {
154 				ret = -ENOMEM;
155 				goto out;
156 			}
157 			cpage_out = kmap(out_page);
158 			pages[nr_pages] = out_page;
159 			nr_pages++;
160 			workspace->strm.avail_out = PAGE_SIZE;
161 			workspace->strm.next_out = cpage_out;
162 		}
163 		/* we're all done */
164 		if (workspace->strm.total_in >= len)
165 			break;
166 
167 		/* we've read in a full page, get a new one */
168 		if (workspace->strm.avail_in == 0) {
169 			if (workspace->strm.total_out > max_out)
170 				break;
171 
172 			bytes_left = len - workspace->strm.total_in;
173 			kunmap(in_page);
174 			put_page(in_page);
175 
176 			start += PAGE_SIZE;
177 			in_page = find_get_page(mapping,
178 						start >> PAGE_SHIFT);
179 			data_in = kmap(in_page);
180 			workspace->strm.avail_in = min(bytes_left,
181 							   PAGE_SIZE);
182 			workspace->strm.next_in = data_in;
183 		}
184 	}
185 	workspace->strm.avail_in = 0;
186 	ret = zlib_deflate(&workspace->strm, Z_FINISH);
187 	zlib_deflateEnd(&workspace->strm);
188 
189 	if (ret != Z_STREAM_END) {
190 		ret = -EIO;
191 		goto out;
192 	}
193 
194 	if (workspace->strm.total_out >= workspace->strm.total_in) {
195 		ret = -E2BIG;
196 		goto out;
197 	}
198 
199 	ret = 0;
200 	*total_out = workspace->strm.total_out;
201 	*total_in = workspace->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 		put_page(in_page);
210 	}
211 	return ret;
212 }
213 
214 static int zlib_decompress_bio(struct list_head *ws, struct page **pages_in,
215 				  u64 disk_start,
216 				  struct bio *orig_bio,
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 total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE);
226 	unsigned long buf_start;
227 
228 	data_in = kmap(pages_in[page_in_index]);
229 	workspace->strm.next_in = data_in;
230 	workspace->strm.avail_in = min_t(size_t, srclen, PAGE_SIZE);
231 	workspace->strm.total_in = 0;
232 
233 	workspace->strm.total_out = 0;
234 	workspace->strm.next_out = workspace->buf;
235 	workspace->strm.avail_out = PAGE_SIZE;
236 
237 	/* If it's deflate, and it's got no preset dictionary, then
238 	   we can tell zlib to skip the adler32 check. */
239 	if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
240 	    ((data_in[0] & 0x0f) == Z_DEFLATED) &&
241 	    !(((data_in[0]<<8) + data_in[1]) % 31)) {
242 
243 		wbits = -((data_in[0] >> 4) + 8);
244 		workspace->strm.next_in += 2;
245 		workspace->strm.avail_in -= 2;
246 	}
247 
248 	if (Z_OK != zlib_inflateInit2(&workspace->strm, wbits)) {
249 		pr_warn("BTRFS: inflateInit failed\n");
250 		kunmap(pages_in[page_in_index]);
251 		return -EIO;
252 	}
253 	while (workspace->strm.total_in < srclen) {
254 		ret = zlib_inflate(&workspace->strm, Z_NO_FLUSH);
255 		if (ret != Z_OK && ret != Z_STREAM_END)
256 			break;
257 
258 		buf_start = total_out;
259 		total_out = workspace->strm.total_out;
260 
261 		/* we didn't make progress in this inflate call, we're done */
262 		if (buf_start == total_out)
263 			break;
264 
265 		ret2 = btrfs_decompress_buf2page(workspace->buf, buf_start,
266 						 total_out, disk_start,
267 						 orig_bio);
268 		if (ret2 == 0) {
269 			ret = 0;
270 			goto done;
271 		}
272 
273 		workspace->strm.next_out = workspace->buf;
274 		workspace->strm.avail_out = PAGE_SIZE;
275 
276 		if (workspace->strm.avail_in == 0) {
277 			unsigned long tmp;
278 			kunmap(pages_in[page_in_index]);
279 			page_in_index++;
280 			if (page_in_index >= total_pages_in) {
281 				data_in = NULL;
282 				break;
283 			}
284 			data_in = kmap(pages_in[page_in_index]);
285 			workspace->strm.next_in = data_in;
286 			tmp = srclen - workspace->strm.total_in;
287 			workspace->strm.avail_in = min(tmp,
288 							   PAGE_SIZE);
289 		}
290 	}
291 	if (ret != Z_STREAM_END)
292 		ret = -EIO;
293 	else
294 		ret = 0;
295 done:
296 	zlib_inflateEnd(&workspace->strm);
297 	if (data_in)
298 		kunmap(pages_in[page_in_index]);
299 	if (!ret)
300 		zero_fill_bio(orig_bio);
301 	return ret;
302 }
303 
304 static int zlib_decompress(struct list_head *ws, unsigned char *data_in,
305 			   struct page *dest_page,
306 			   unsigned long start_byte,
307 			   size_t srclen, size_t destlen)
308 {
309 	struct workspace *workspace = list_entry(ws, struct workspace, list);
310 	int ret = 0;
311 	int wbits = MAX_WBITS;
312 	unsigned long bytes_left;
313 	unsigned long total_out = 0;
314 	unsigned long pg_offset = 0;
315 	char *kaddr;
316 
317 	destlen = min_t(unsigned long, destlen, PAGE_SIZE);
318 	bytes_left = destlen;
319 
320 	workspace->strm.next_in = data_in;
321 	workspace->strm.avail_in = srclen;
322 	workspace->strm.total_in = 0;
323 
324 	workspace->strm.next_out = workspace->buf;
325 	workspace->strm.avail_out = PAGE_SIZE;
326 	workspace->strm.total_out = 0;
327 	/* If it's deflate, and it's got no preset dictionary, then
328 	   we can tell zlib to skip the adler32 check. */
329 	if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
330 	    ((data_in[0] & 0x0f) == Z_DEFLATED) &&
331 	    !(((data_in[0]<<8) + data_in[1]) % 31)) {
332 
333 		wbits = -((data_in[0] >> 4) + 8);
334 		workspace->strm.next_in += 2;
335 		workspace->strm.avail_in -= 2;
336 	}
337 
338 	if (Z_OK != zlib_inflateInit2(&workspace->strm, wbits)) {
339 		pr_warn("BTRFS: inflateInit failed\n");
340 		return -EIO;
341 	}
342 
343 	while (bytes_left > 0) {
344 		unsigned long buf_start;
345 		unsigned long buf_offset;
346 		unsigned long bytes;
347 
348 		ret = zlib_inflate(&workspace->strm, Z_NO_FLUSH);
349 		if (ret != Z_OK && ret != Z_STREAM_END)
350 			break;
351 
352 		buf_start = total_out;
353 		total_out = workspace->strm.total_out;
354 
355 		if (total_out == buf_start) {
356 			ret = -EIO;
357 			break;
358 		}
359 
360 		if (total_out <= start_byte)
361 			goto next;
362 
363 		if (total_out > start_byte && buf_start < start_byte)
364 			buf_offset = start_byte - buf_start;
365 		else
366 			buf_offset = 0;
367 
368 		bytes = min(PAGE_SIZE - pg_offset,
369 			    PAGE_SIZE - buf_offset);
370 		bytes = min(bytes, bytes_left);
371 
372 		kaddr = kmap_atomic(dest_page);
373 		memcpy(kaddr + pg_offset, workspace->buf + buf_offset, bytes);
374 		kunmap_atomic(kaddr);
375 
376 		pg_offset += bytes;
377 		bytes_left -= bytes;
378 next:
379 		workspace->strm.next_out = workspace->buf;
380 		workspace->strm.avail_out = PAGE_SIZE;
381 	}
382 
383 	if (ret != Z_STREAM_END && bytes_left != 0)
384 		ret = -EIO;
385 	else
386 		ret = 0;
387 
388 	zlib_inflateEnd(&workspace->strm);
389 
390 	/*
391 	 * this should only happen if zlib returned fewer bytes than we
392 	 * expected.  btrfs_get_block is responsible for zeroing from the
393 	 * end of the inline extent (destlen) to the end of the page
394 	 */
395 	if (pg_offset < destlen) {
396 		kaddr = kmap_atomic(dest_page);
397 		memset(kaddr + pg_offset, 0, destlen - pg_offset);
398 		kunmap_atomic(kaddr);
399 	}
400 	return ret;
401 }
402 
403 const struct btrfs_compress_op btrfs_zlib_compress = {
404 	.alloc_workspace	= zlib_alloc_workspace,
405 	.free_workspace		= zlib_free_workspace,
406 	.compress_pages		= zlib_compress_pages,
407 	.decompress_bio		= zlib_decompress_bio,
408 	.decompress		= zlib_decompress,
409 };
410