xref: /openbmc/linux/fs/btrfs/zlib.c (revision 0ad53fe3)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2008 Oracle.  All rights reserved.
4  *
5  * Based on jffs2 zlib code:
6  * Copyright © 2001-2007 Red Hat, Inc.
7  * Created by David Woodhouse <dwmw2@infradead.org>
8  */
9 
10 #include <linux/kernel.h>
11 #include <linux/slab.h>
12 #include <linux/zlib.h>
13 #include <linux/zutil.h>
14 #include <linux/mm.h>
15 #include <linux/init.h>
16 #include <linux/err.h>
17 #include <linux/sched.h>
18 #include <linux/pagemap.h>
19 #include <linux/bio.h>
20 #include <linux/refcount.h>
21 #include "compression.h"
22 
23 /* workspace buffer size for s390 zlib hardware support */
24 #define ZLIB_DFLTCC_BUF_SIZE    (4 * PAGE_SIZE)
25 
26 struct workspace {
27 	z_stream strm;
28 	char *buf;
29 	unsigned int buf_size;
30 	struct list_head list;
31 	int level;
32 };
33 
34 static struct workspace_manager wsm;
35 
36 struct list_head *zlib_get_workspace(unsigned int level)
37 {
38 	struct list_head *ws = btrfs_get_workspace(BTRFS_COMPRESS_ZLIB, level);
39 	struct workspace *workspace = list_entry(ws, struct workspace, list);
40 
41 	workspace->level = level;
42 
43 	return ws;
44 }
45 
46 void zlib_free_workspace(struct list_head *ws)
47 {
48 	struct workspace *workspace = list_entry(ws, struct workspace, list);
49 
50 	kvfree(workspace->strm.workspace);
51 	kfree(workspace->buf);
52 	kfree(workspace);
53 }
54 
55 struct list_head *zlib_alloc_workspace(unsigned int level)
56 {
57 	struct workspace *workspace;
58 	int workspacesize;
59 
60 	workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
61 	if (!workspace)
62 		return ERR_PTR(-ENOMEM);
63 
64 	workspacesize = max(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL),
65 			zlib_inflate_workspacesize());
66 	workspace->strm.workspace = kvmalloc(workspacesize, GFP_KERNEL);
67 	workspace->level = level;
68 	workspace->buf = NULL;
69 	/*
70 	 * In case of s390 zlib hardware support, allocate lager workspace
71 	 * buffer. If allocator fails, fall back to a single page buffer.
72 	 */
73 	if (zlib_deflate_dfltcc_enabled()) {
74 		workspace->buf = kmalloc(ZLIB_DFLTCC_BUF_SIZE,
75 					 __GFP_NOMEMALLOC | __GFP_NORETRY |
76 					 __GFP_NOWARN | GFP_NOIO);
77 		workspace->buf_size = ZLIB_DFLTCC_BUF_SIZE;
78 	}
79 	if (!workspace->buf) {
80 		workspace->buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
81 		workspace->buf_size = PAGE_SIZE;
82 	}
83 	if (!workspace->strm.workspace || !workspace->buf)
84 		goto fail;
85 
86 	INIT_LIST_HEAD(&workspace->list);
87 
88 	return &workspace->list;
89 fail:
90 	zlib_free_workspace(&workspace->list);
91 	return ERR_PTR(-ENOMEM);
92 }
93 
94 int zlib_compress_pages(struct list_head *ws, struct address_space *mapping,
95 		u64 start, struct page **pages, unsigned long *out_pages,
96 		unsigned long *total_in, unsigned long *total_out)
97 {
98 	struct workspace *workspace = list_entry(ws, struct workspace, list);
99 	int ret;
100 	char *data_in;
101 	char *cpage_out;
102 	int nr_pages = 0;
103 	struct page *in_page = NULL;
104 	struct page *out_page = NULL;
105 	unsigned long bytes_left;
106 	unsigned int in_buf_pages;
107 	unsigned long len = *total_out;
108 	unsigned long nr_dest_pages = *out_pages;
109 	const unsigned long max_out = nr_dest_pages * PAGE_SIZE;
110 
111 	*out_pages = 0;
112 	*total_out = 0;
113 	*total_in = 0;
114 
115 	if (Z_OK != zlib_deflateInit(&workspace->strm, workspace->level)) {
116 		pr_warn("BTRFS: deflateInit failed\n");
117 		ret = -EIO;
118 		goto out;
119 	}
120 
121 	workspace->strm.total_in = 0;
122 	workspace->strm.total_out = 0;
123 
124 	out_page = alloc_page(GFP_NOFS);
125 	if (out_page == NULL) {
126 		ret = -ENOMEM;
127 		goto out;
128 	}
129 	cpage_out = page_address(out_page);
130 	pages[0] = out_page;
131 	nr_pages = 1;
132 
133 	workspace->strm.next_in = workspace->buf;
134 	workspace->strm.avail_in = 0;
135 	workspace->strm.next_out = cpage_out;
136 	workspace->strm.avail_out = PAGE_SIZE;
137 
138 	while (workspace->strm.total_in < len) {
139 		/*
140 		 * Get next input pages and copy the contents to
141 		 * the workspace buffer if required.
142 		 */
143 		if (workspace->strm.avail_in == 0) {
144 			bytes_left = len - workspace->strm.total_in;
145 			in_buf_pages = min(DIV_ROUND_UP(bytes_left, PAGE_SIZE),
146 					   workspace->buf_size / PAGE_SIZE);
147 			if (in_buf_pages > 1) {
148 				int i;
149 
150 				for (i = 0; i < in_buf_pages; i++) {
151 					if (in_page)
152 						put_page(in_page);
153 					in_page = find_get_page(mapping,
154 								start >> PAGE_SHIFT);
155 					data_in = page_address(in_page);
156 					memcpy(workspace->buf + i * PAGE_SIZE,
157 					       data_in, PAGE_SIZE);
158 					start += PAGE_SIZE;
159 				}
160 				workspace->strm.next_in = workspace->buf;
161 			} else {
162 				if (in_page)
163 					put_page(in_page);
164 				in_page = find_get_page(mapping,
165 							start >> PAGE_SHIFT);
166 				data_in = page_address(in_page);
167 				start += PAGE_SIZE;
168 				workspace->strm.next_in = data_in;
169 			}
170 			workspace->strm.avail_in = min(bytes_left,
171 						       (unsigned long) workspace->buf_size);
172 		}
173 
174 		ret = zlib_deflate(&workspace->strm, Z_SYNC_FLUSH);
175 		if (ret != Z_OK) {
176 			pr_debug("BTRFS: deflate in loop returned %d\n",
177 			       ret);
178 			zlib_deflateEnd(&workspace->strm);
179 			ret = -EIO;
180 			goto out;
181 		}
182 
183 		/* we're making it bigger, give up */
184 		if (workspace->strm.total_in > 8192 &&
185 		    workspace->strm.total_in <
186 		    workspace->strm.total_out) {
187 			ret = -E2BIG;
188 			goto out;
189 		}
190 		/* we need another page for writing out.  Test this
191 		 * before the total_in so we will pull in a new page for
192 		 * the stream end if required
193 		 */
194 		if (workspace->strm.avail_out == 0) {
195 			if (nr_pages == nr_dest_pages) {
196 				out_page = NULL;
197 				ret = -E2BIG;
198 				goto out;
199 			}
200 			out_page = alloc_page(GFP_NOFS);
201 			if (out_page == NULL) {
202 				ret = -ENOMEM;
203 				goto out;
204 			}
205 			cpage_out = page_address(out_page);
206 			pages[nr_pages] = out_page;
207 			nr_pages++;
208 			workspace->strm.avail_out = PAGE_SIZE;
209 			workspace->strm.next_out = cpage_out;
210 		}
211 		/* we're all done */
212 		if (workspace->strm.total_in >= len)
213 			break;
214 		if (workspace->strm.total_out > max_out)
215 			break;
216 	}
217 	workspace->strm.avail_in = 0;
218 	/*
219 	 * Call deflate with Z_FINISH flush parameter providing more output
220 	 * space but no more input data, until it returns with Z_STREAM_END.
221 	 */
222 	while (ret != Z_STREAM_END) {
223 		ret = zlib_deflate(&workspace->strm, Z_FINISH);
224 		if (ret == Z_STREAM_END)
225 			break;
226 		if (ret != Z_OK && ret != Z_BUF_ERROR) {
227 			zlib_deflateEnd(&workspace->strm);
228 			ret = -EIO;
229 			goto out;
230 		} else if (workspace->strm.avail_out == 0) {
231 			/* get another page for the stream end */
232 			if (nr_pages == nr_dest_pages) {
233 				out_page = NULL;
234 				ret = -E2BIG;
235 				goto out;
236 			}
237 			out_page = alloc_page(GFP_NOFS);
238 			if (out_page == NULL) {
239 				ret = -ENOMEM;
240 				goto out;
241 			}
242 			cpage_out = page_address(out_page);
243 			pages[nr_pages] = out_page;
244 			nr_pages++;
245 			workspace->strm.avail_out = PAGE_SIZE;
246 			workspace->strm.next_out = cpage_out;
247 		}
248 	}
249 	zlib_deflateEnd(&workspace->strm);
250 
251 	if (workspace->strm.total_out >= workspace->strm.total_in) {
252 		ret = -E2BIG;
253 		goto out;
254 	}
255 
256 	ret = 0;
257 	*total_out = workspace->strm.total_out;
258 	*total_in = workspace->strm.total_in;
259 out:
260 	*out_pages = nr_pages;
261 	if (in_page)
262 		put_page(in_page);
263 	return ret;
264 }
265 
266 int zlib_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
267 {
268 	struct workspace *workspace = list_entry(ws, struct workspace, list);
269 	int ret = 0, ret2;
270 	int wbits = MAX_WBITS;
271 	char *data_in;
272 	size_t total_out = 0;
273 	unsigned long page_in_index = 0;
274 	size_t srclen = cb->compressed_len;
275 	unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE);
276 	unsigned long buf_start;
277 	struct page **pages_in = cb->compressed_pages;
278 
279 	data_in = page_address(pages_in[page_in_index]);
280 	workspace->strm.next_in = data_in;
281 	workspace->strm.avail_in = min_t(size_t, srclen, PAGE_SIZE);
282 	workspace->strm.total_in = 0;
283 
284 	workspace->strm.total_out = 0;
285 	workspace->strm.next_out = workspace->buf;
286 	workspace->strm.avail_out = workspace->buf_size;
287 
288 	/* If it's deflate, and it's got no preset dictionary, then
289 	   we can tell zlib to skip the adler32 check. */
290 	if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
291 	    ((data_in[0] & 0x0f) == Z_DEFLATED) &&
292 	    !(((data_in[0]<<8) + data_in[1]) % 31)) {
293 
294 		wbits = -((data_in[0] >> 4) + 8);
295 		workspace->strm.next_in += 2;
296 		workspace->strm.avail_in -= 2;
297 	}
298 
299 	if (Z_OK != zlib_inflateInit2(&workspace->strm, wbits)) {
300 		pr_warn("BTRFS: inflateInit failed\n");
301 		return -EIO;
302 	}
303 	while (workspace->strm.total_in < srclen) {
304 		ret = zlib_inflate(&workspace->strm, Z_NO_FLUSH);
305 		if (ret != Z_OK && ret != Z_STREAM_END)
306 			break;
307 
308 		buf_start = total_out;
309 		total_out = workspace->strm.total_out;
310 
311 		/* we didn't make progress in this inflate call, we're done */
312 		if (buf_start == total_out)
313 			break;
314 
315 		ret2 = btrfs_decompress_buf2page(workspace->buf,
316 				total_out - buf_start, cb, buf_start);
317 		if (ret2 == 0) {
318 			ret = 0;
319 			goto done;
320 		}
321 
322 		workspace->strm.next_out = workspace->buf;
323 		workspace->strm.avail_out = workspace->buf_size;
324 
325 		if (workspace->strm.avail_in == 0) {
326 			unsigned long tmp;
327 
328 			page_in_index++;
329 			if (page_in_index >= total_pages_in) {
330 				data_in = NULL;
331 				break;
332 			}
333 			data_in = page_address(pages_in[page_in_index]);
334 			workspace->strm.next_in = data_in;
335 			tmp = srclen - workspace->strm.total_in;
336 			workspace->strm.avail_in = min(tmp, PAGE_SIZE);
337 		}
338 	}
339 	if (ret != Z_STREAM_END)
340 		ret = -EIO;
341 	else
342 		ret = 0;
343 done:
344 	zlib_inflateEnd(&workspace->strm);
345 	if (!ret)
346 		zero_fill_bio(cb->orig_bio);
347 	return ret;
348 }
349 
350 int zlib_decompress(struct list_head *ws, unsigned char *data_in,
351 		struct page *dest_page, unsigned long start_byte, size_t srclen,
352 		size_t destlen)
353 {
354 	struct workspace *workspace = list_entry(ws, struct workspace, list);
355 	int ret = 0;
356 	int wbits = MAX_WBITS;
357 	unsigned long bytes_left;
358 	unsigned long total_out = 0;
359 	unsigned long pg_offset = 0;
360 
361 	destlen = min_t(unsigned long, destlen, PAGE_SIZE);
362 	bytes_left = destlen;
363 
364 	workspace->strm.next_in = data_in;
365 	workspace->strm.avail_in = srclen;
366 	workspace->strm.total_in = 0;
367 
368 	workspace->strm.next_out = workspace->buf;
369 	workspace->strm.avail_out = workspace->buf_size;
370 	workspace->strm.total_out = 0;
371 	/* If it's deflate, and it's got no preset dictionary, then
372 	   we can tell zlib to skip the adler32 check. */
373 	if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
374 	    ((data_in[0] & 0x0f) == Z_DEFLATED) &&
375 	    !(((data_in[0]<<8) + data_in[1]) % 31)) {
376 
377 		wbits = -((data_in[0] >> 4) + 8);
378 		workspace->strm.next_in += 2;
379 		workspace->strm.avail_in -= 2;
380 	}
381 
382 	if (Z_OK != zlib_inflateInit2(&workspace->strm, wbits)) {
383 		pr_warn("BTRFS: inflateInit failed\n");
384 		return -EIO;
385 	}
386 
387 	while (bytes_left > 0) {
388 		unsigned long buf_start;
389 		unsigned long buf_offset;
390 		unsigned long bytes;
391 
392 		ret = zlib_inflate(&workspace->strm, Z_NO_FLUSH);
393 		if (ret != Z_OK && ret != Z_STREAM_END)
394 			break;
395 
396 		buf_start = total_out;
397 		total_out = workspace->strm.total_out;
398 
399 		if (total_out == buf_start) {
400 			ret = -EIO;
401 			break;
402 		}
403 
404 		if (total_out <= start_byte)
405 			goto next;
406 
407 		if (total_out > start_byte && buf_start < start_byte)
408 			buf_offset = start_byte - buf_start;
409 		else
410 			buf_offset = 0;
411 
412 		bytes = min(PAGE_SIZE - pg_offset,
413 			    PAGE_SIZE - (buf_offset % PAGE_SIZE));
414 		bytes = min(bytes, bytes_left);
415 
416 		memcpy_to_page(dest_page, pg_offset,
417 			       workspace->buf + buf_offset, bytes);
418 
419 		pg_offset += bytes;
420 		bytes_left -= bytes;
421 next:
422 		workspace->strm.next_out = workspace->buf;
423 		workspace->strm.avail_out = workspace->buf_size;
424 	}
425 
426 	if (ret != Z_STREAM_END && bytes_left != 0)
427 		ret = -EIO;
428 	else
429 		ret = 0;
430 
431 	zlib_inflateEnd(&workspace->strm);
432 
433 	/*
434 	 * this should only happen if zlib returned fewer bytes than we
435 	 * expected.  btrfs_get_block is responsible for zeroing from the
436 	 * end of the inline extent (destlen) to the end of the page
437 	 */
438 	if (pg_offset < destlen) {
439 		memzero_page(dest_page, pg_offset, destlen - pg_offset);
440 	}
441 	return ret;
442 }
443 
444 const struct btrfs_compress_op btrfs_zlib_compress = {
445 	.workspace_manager	= &wsm,
446 	.max_level		= 9,
447 	.default_level		= BTRFS_ZLIB_DEFAULT_LEVEL,
448 };
449