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
zlib_get_workspace(unsigned int level)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
zlib_free_workspace(struct list_head * ws)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
zlib_alloc_workspace(unsigned int level)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
zlib_compress_pages(struct list_head * ws,struct address_space * mapping,u64 start,struct page ** pages,unsigned long * out_pages,unsigned long * total_in,unsigned long * total_out)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 | __GFP_HIGHMEM);
125 if (out_page == NULL) {
126 ret = -ENOMEM;
127 goto out;
128 }
129 cpage_out = kmap(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 kunmap(in_page);
153 put_page(in_page);
154 }
155 in_page = find_get_page(mapping,
156 start >> PAGE_SHIFT);
157 data_in = kmap(in_page);
158 memcpy(workspace->buf + i * PAGE_SIZE,
159 data_in, PAGE_SIZE);
160 start += PAGE_SIZE;
161 }
162 workspace->strm.next_in = workspace->buf;
163 } else {
164 if (in_page) {
165 kunmap(in_page);
166 put_page(in_page);
167 }
168 in_page = find_get_page(mapping,
169 start >> PAGE_SHIFT);
170 data_in = kmap(in_page);
171 start += PAGE_SIZE;
172 workspace->strm.next_in = data_in;
173 }
174 workspace->strm.avail_in = min(bytes_left,
175 (unsigned long) workspace->buf_size);
176 }
177
178 ret = zlib_deflate(&workspace->strm, Z_SYNC_FLUSH);
179 if (ret != Z_OK) {
180 pr_debug("BTRFS: deflate in loop returned %d\n",
181 ret);
182 zlib_deflateEnd(&workspace->strm);
183 ret = -EIO;
184 goto out;
185 }
186
187 /* we're making it bigger, give up */
188 if (workspace->strm.total_in > 8192 &&
189 workspace->strm.total_in <
190 workspace->strm.total_out) {
191 ret = -E2BIG;
192 goto out;
193 }
194 /* we need another page for writing out. Test this
195 * before the total_in so we will pull in a new page for
196 * the stream end if required
197 */
198 if (workspace->strm.avail_out == 0) {
199 kunmap(out_page);
200 if (nr_pages == nr_dest_pages) {
201 out_page = NULL;
202 ret = -E2BIG;
203 goto out;
204 }
205 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
206 if (out_page == NULL) {
207 ret = -ENOMEM;
208 goto out;
209 }
210 cpage_out = kmap(out_page);
211 pages[nr_pages] = out_page;
212 nr_pages++;
213 workspace->strm.avail_out = PAGE_SIZE;
214 workspace->strm.next_out = cpage_out;
215 }
216 /* we're all done */
217 if (workspace->strm.total_in >= len)
218 break;
219 if (workspace->strm.total_out > max_out)
220 break;
221 }
222 workspace->strm.avail_in = 0;
223 /*
224 * Call deflate with Z_FINISH flush parameter providing more output
225 * space but no more input data, until it returns with Z_STREAM_END.
226 */
227 while (ret != Z_STREAM_END) {
228 ret = zlib_deflate(&workspace->strm, Z_FINISH);
229 if (ret == Z_STREAM_END)
230 break;
231 if (ret != Z_OK && ret != Z_BUF_ERROR) {
232 zlib_deflateEnd(&workspace->strm);
233 ret = -EIO;
234 goto out;
235 } else if (workspace->strm.avail_out == 0) {
236 /* get another page for the stream end */
237 kunmap(out_page);
238 if (nr_pages == nr_dest_pages) {
239 out_page = NULL;
240 ret = -E2BIG;
241 goto out;
242 }
243 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
244 if (out_page == NULL) {
245 ret = -ENOMEM;
246 goto out;
247 }
248 cpage_out = kmap(out_page);
249 pages[nr_pages] = out_page;
250 nr_pages++;
251 workspace->strm.avail_out = PAGE_SIZE;
252 workspace->strm.next_out = cpage_out;
253 }
254 }
255 zlib_deflateEnd(&workspace->strm);
256
257 if (workspace->strm.total_out >= workspace->strm.total_in) {
258 ret = -E2BIG;
259 goto out;
260 }
261
262 ret = 0;
263 *total_out = workspace->strm.total_out;
264 *total_in = workspace->strm.total_in;
265 out:
266 *out_pages = nr_pages;
267 if (out_page)
268 kunmap(out_page);
269
270 if (in_page) {
271 kunmap(in_page);
272 put_page(in_page);
273 }
274 return ret;
275 }
276
zlib_decompress_bio(struct list_head * ws,struct compressed_bio * cb)277 int zlib_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
278 {
279 struct workspace *workspace = list_entry(ws, struct workspace, list);
280 int ret = 0, ret2;
281 int wbits = MAX_WBITS;
282 char *data_in;
283 size_t total_out = 0;
284 unsigned long page_in_index = 0;
285 size_t srclen = cb->compressed_len;
286 unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE);
287 unsigned long buf_start;
288 struct page **pages_in = cb->compressed_pages;
289 u64 disk_start = cb->start;
290 struct bio *orig_bio = cb->orig_bio;
291
292 data_in = kmap(pages_in[page_in_index]);
293 workspace->strm.next_in = data_in;
294 workspace->strm.avail_in = min_t(size_t, srclen, PAGE_SIZE);
295 workspace->strm.total_in = 0;
296
297 workspace->strm.total_out = 0;
298 workspace->strm.next_out = workspace->buf;
299 workspace->strm.avail_out = workspace->buf_size;
300
301 /* If it's deflate, and it's got no preset dictionary, then
302 we can tell zlib to skip the adler32 check. */
303 if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
304 ((data_in[0] & 0x0f) == Z_DEFLATED) &&
305 !(((data_in[0]<<8) + data_in[1]) % 31)) {
306
307 wbits = -((data_in[0] >> 4) + 8);
308 workspace->strm.next_in += 2;
309 workspace->strm.avail_in -= 2;
310 }
311
312 if (Z_OK != zlib_inflateInit2(&workspace->strm, wbits)) {
313 pr_warn("BTRFS: inflateInit failed\n");
314 kunmap(pages_in[page_in_index]);
315 return -EIO;
316 }
317 while (workspace->strm.total_in < srclen) {
318 ret = zlib_inflate(&workspace->strm, Z_NO_FLUSH);
319 if (ret != Z_OK && ret != Z_STREAM_END)
320 break;
321
322 buf_start = total_out;
323 total_out = workspace->strm.total_out;
324
325 /* we didn't make progress in this inflate call, we're done */
326 if (buf_start == total_out)
327 break;
328
329 ret2 = btrfs_decompress_buf2page(workspace->buf, buf_start,
330 total_out, disk_start,
331 orig_bio);
332 if (ret2 == 0) {
333 ret = 0;
334 goto done;
335 }
336
337 workspace->strm.next_out = workspace->buf;
338 workspace->strm.avail_out = workspace->buf_size;
339
340 if (workspace->strm.avail_in == 0) {
341 unsigned long tmp;
342 kunmap(pages_in[page_in_index]);
343 page_in_index++;
344 if (page_in_index >= total_pages_in) {
345 data_in = NULL;
346 break;
347 }
348 data_in = kmap(pages_in[page_in_index]);
349 workspace->strm.next_in = data_in;
350 tmp = srclen - workspace->strm.total_in;
351 workspace->strm.avail_in = min(tmp,
352 PAGE_SIZE);
353 }
354 }
355 if (ret != Z_STREAM_END)
356 ret = -EIO;
357 else
358 ret = 0;
359 done:
360 zlib_inflateEnd(&workspace->strm);
361 if (data_in)
362 kunmap(pages_in[page_in_index]);
363 if (!ret)
364 zero_fill_bio(orig_bio);
365 return ret;
366 }
367
zlib_decompress(struct list_head * ws,unsigned char * data_in,struct page * dest_page,unsigned long start_byte,size_t srclen,size_t destlen)368 int zlib_decompress(struct list_head *ws, unsigned char *data_in,
369 struct page *dest_page, unsigned long start_byte, size_t srclen,
370 size_t destlen)
371 {
372 struct workspace *workspace = list_entry(ws, struct workspace, list);
373 int ret = 0;
374 int wbits = MAX_WBITS;
375 unsigned long bytes_left;
376 unsigned long total_out = 0;
377 unsigned long pg_offset = 0;
378
379 destlen = min_t(unsigned long, destlen, PAGE_SIZE);
380 bytes_left = destlen;
381
382 workspace->strm.next_in = data_in;
383 workspace->strm.avail_in = srclen;
384 workspace->strm.total_in = 0;
385
386 workspace->strm.next_out = workspace->buf;
387 workspace->strm.avail_out = workspace->buf_size;
388 workspace->strm.total_out = 0;
389 /* If it's deflate, and it's got no preset dictionary, then
390 we can tell zlib to skip the adler32 check. */
391 if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
392 ((data_in[0] & 0x0f) == Z_DEFLATED) &&
393 !(((data_in[0]<<8) + data_in[1]) % 31)) {
394
395 wbits = -((data_in[0] >> 4) + 8);
396 workspace->strm.next_in += 2;
397 workspace->strm.avail_in -= 2;
398 }
399
400 if (Z_OK != zlib_inflateInit2(&workspace->strm, wbits)) {
401 pr_warn("BTRFS: inflateInit failed\n");
402 return -EIO;
403 }
404
405 while (bytes_left > 0) {
406 unsigned long buf_start;
407 unsigned long buf_offset;
408 unsigned long bytes;
409
410 ret = zlib_inflate(&workspace->strm, Z_NO_FLUSH);
411 if (ret != Z_OK && ret != Z_STREAM_END)
412 break;
413
414 buf_start = total_out;
415 total_out = workspace->strm.total_out;
416
417 if (total_out == buf_start) {
418 ret = -EIO;
419 break;
420 }
421
422 if (total_out <= start_byte)
423 goto next;
424
425 if (total_out > start_byte && buf_start < start_byte)
426 buf_offset = start_byte - buf_start;
427 else
428 buf_offset = 0;
429
430 bytes = min(PAGE_SIZE - pg_offset,
431 PAGE_SIZE - (buf_offset % PAGE_SIZE));
432 bytes = min(bytes, bytes_left);
433
434 memcpy_to_page(dest_page, pg_offset,
435 workspace->buf + buf_offset, bytes);
436
437 pg_offset += bytes;
438 bytes_left -= bytes;
439 next:
440 workspace->strm.next_out = workspace->buf;
441 workspace->strm.avail_out = workspace->buf_size;
442 }
443
444 if (ret != Z_STREAM_END && bytes_left != 0)
445 ret = -EIO;
446 else
447 ret = 0;
448
449 zlib_inflateEnd(&workspace->strm);
450
451 /*
452 * this should only happen if zlib returned fewer bytes than we
453 * expected. btrfs_get_block is responsible for zeroing from the
454 * end of the inline extent (destlen) to the end of the page
455 */
456 if (pg_offset < destlen) {
457 memzero_page(dest_page, pg_offset, destlen - pg_offset);
458 }
459 return ret;
460 }
461
462 const struct btrfs_compress_op btrfs_zlib_compress = {
463 .workspace_manager = &wsm,
464 .max_level = 9,
465 .default_level = BTRFS_ZLIB_DEFAULT_LEVEL,
466 };
467