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 "btrfs_inode.h"
22 #include "compression.h"
23 #include "fs.h"
24 #include "subpage.h"
25
26 /* workspace buffer size for s390 zlib hardware support */
27 #define ZLIB_DFLTCC_BUF_SIZE (4 * PAGE_SIZE)
28
29 struct workspace {
30 z_stream strm;
31 char *buf;
32 unsigned int buf_size;
33 struct list_head list;
34 int level;
35 };
36
37 static struct workspace_manager wsm;
38
zlib_get_workspace(unsigned int level)39 struct list_head *zlib_get_workspace(unsigned int level)
40 {
41 struct list_head *ws = btrfs_get_workspace(BTRFS_COMPRESS_ZLIB, level);
42 struct workspace *workspace = list_entry(ws, struct workspace, list);
43
44 workspace->level = level;
45
46 return ws;
47 }
48
zlib_free_workspace(struct list_head * ws)49 void zlib_free_workspace(struct list_head *ws)
50 {
51 struct workspace *workspace = list_entry(ws, struct workspace, list);
52
53 kvfree(workspace->strm.workspace);
54 kfree(workspace->buf);
55 kfree(workspace);
56 }
57
zlib_alloc_workspace(unsigned int level)58 struct list_head *zlib_alloc_workspace(unsigned int level)
59 {
60 struct workspace *workspace;
61 int workspacesize;
62
63 workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
64 if (!workspace)
65 return ERR_PTR(-ENOMEM);
66
67 workspacesize = max(zlib_deflate_workspacesize(MAX_WBITS, MAX_MEM_LEVEL),
68 zlib_inflate_workspacesize());
69 workspace->strm.workspace = kvzalloc(workspacesize, GFP_KERNEL | __GFP_NOWARN);
70 workspace->level = level;
71 workspace->buf = NULL;
72 /*
73 * In case of s390 zlib hardware support, allocate lager workspace
74 * buffer. If allocator fails, fall back to a single page buffer.
75 */
76 if (zlib_deflate_dfltcc_enabled()) {
77 workspace->buf = kmalloc(ZLIB_DFLTCC_BUF_SIZE,
78 __GFP_NOMEMALLOC | __GFP_NORETRY |
79 __GFP_NOWARN | GFP_NOIO);
80 workspace->buf_size = ZLIB_DFLTCC_BUF_SIZE;
81 }
82 if (!workspace->buf) {
83 workspace->buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
84 workspace->buf_size = PAGE_SIZE;
85 }
86 if (!workspace->strm.workspace || !workspace->buf)
87 goto fail;
88
89 INIT_LIST_HEAD(&workspace->list);
90
91 return &workspace->list;
92 fail:
93 zlib_free_workspace(&workspace->list);
94 return ERR_PTR(-ENOMEM);
95 }
96
zlib_compress_folios(struct list_head * ws,struct address_space * mapping,u64 start,struct folio ** folios,unsigned long * out_folios,unsigned long * total_in,unsigned long * total_out)97 int zlib_compress_folios(struct list_head *ws, struct address_space *mapping,
98 u64 start, struct folio **folios, unsigned long *out_folios,
99 unsigned long *total_in, unsigned long *total_out)
100 {
101 struct workspace *workspace = list_entry(ws, struct workspace, list);
102 int ret;
103 char *data_in = NULL;
104 char *cfolio_out;
105 int nr_folios = 0;
106 struct folio *in_folio = NULL;
107 struct folio *out_folio = NULL;
108 unsigned long bytes_left;
109 unsigned int in_buf_folios;
110 unsigned long len = *total_out;
111 unsigned long nr_dest_folios = *out_folios;
112 const unsigned long max_out = nr_dest_folios * PAGE_SIZE;
113 const u64 orig_end = start + len;
114
115 *out_folios = 0;
116 *total_out = 0;
117 *total_in = 0;
118
119 ret = zlib_deflateInit(&workspace->strm, workspace->level);
120 if (unlikely(ret != Z_OK)) {
121 struct btrfs_inode *inode = BTRFS_I(mapping->host);
122
123 btrfs_err(inode->root->fs_info,
124 "zlib compression init failed, error %d root %llu inode %llu offset %llu",
125 ret, btrfs_root_id(inode->root), btrfs_ino(inode), start);
126 ret = -EIO;
127 goto out;
128 }
129
130 workspace->strm.total_in = 0;
131 workspace->strm.total_out = 0;
132
133 out_folio = btrfs_alloc_compr_folio();
134 if (out_folio == NULL) {
135 ret = -ENOMEM;
136 goto out;
137 }
138 cfolio_out = folio_address(out_folio);
139 folios[0] = out_folio;
140 nr_folios = 1;
141
142 workspace->strm.next_in = workspace->buf;
143 workspace->strm.avail_in = 0;
144 workspace->strm.next_out = cfolio_out;
145 workspace->strm.avail_out = PAGE_SIZE;
146
147 while (workspace->strm.total_in < len) {
148 /*
149 * Get next input pages and copy the contents to
150 * the workspace buffer if required.
151 */
152 if (workspace->strm.avail_in == 0) {
153 bytes_left = len - workspace->strm.total_in;
154 in_buf_folios = min(DIV_ROUND_UP(bytes_left, PAGE_SIZE),
155 workspace->buf_size / PAGE_SIZE);
156 if (in_buf_folios > 1) {
157 int i;
158
159 /* S390 hardware acceleration path, not subpage. */
160 ASSERT(!btrfs_is_subpage(
161 inode_to_fs_info(mapping->host),
162 mapping));
163 for (i = 0; i < in_buf_folios; i++) {
164 if (data_in) {
165 kunmap_local(data_in);
166 folio_put(in_folio);
167 data_in = NULL;
168 }
169 ret = btrfs_compress_filemap_get_folio(mapping,
170 start, &in_folio);
171 if (ret < 0)
172 goto out;
173 data_in = kmap_local_folio(in_folio, 0);
174 copy_page(workspace->buf + i * PAGE_SIZE,
175 data_in);
176 start += PAGE_SIZE;
177 workspace->strm.avail_in =
178 (in_buf_folios << PAGE_SHIFT);
179 }
180 workspace->strm.next_in = workspace->buf;
181 } else {
182 unsigned int pg_off;
183 unsigned int cur_len;
184
185 if (data_in) {
186 kunmap_local(data_in);
187 folio_put(in_folio);
188 data_in = NULL;
189 }
190 ret = btrfs_compress_filemap_get_folio(mapping,
191 start, &in_folio);
192 if (ret < 0)
193 goto out;
194 pg_off = offset_in_page(start);
195 cur_len = btrfs_calc_input_length(orig_end, start);
196 data_in = kmap_local_folio(in_folio, pg_off);
197 start += PAGE_SIZE;
198 workspace->strm.next_in = data_in;
199 workspace->strm.avail_in = cur_len;
200 }
201 }
202
203 ret = zlib_deflate(&workspace->strm, Z_SYNC_FLUSH);
204 if (unlikely(ret != Z_OK)) {
205 struct btrfs_inode *inode = BTRFS_I(mapping->host);
206
207 btrfs_warn(inode->root->fs_info,
208 "zlib compression failed, error %d root %llu inode %llu offset %llu",
209 ret, btrfs_root_id(inode->root), btrfs_ino(inode),
210 start);
211 zlib_deflateEnd(&workspace->strm);
212 ret = -EIO;
213 goto out;
214 }
215
216 /* we're making it bigger, give up */
217 if (workspace->strm.total_in > 8192 &&
218 workspace->strm.total_in <
219 workspace->strm.total_out) {
220 ret = -E2BIG;
221 goto out;
222 }
223 /* we need another page for writing out. Test this
224 * before the total_in so we will pull in a new page for
225 * the stream end if required
226 */
227 if (workspace->strm.avail_out == 0) {
228 if (nr_folios == nr_dest_folios) {
229 ret = -E2BIG;
230 goto out;
231 }
232 out_folio = btrfs_alloc_compr_folio();
233 if (out_folio == NULL) {
234 ret = -ENOMEM;
235 goto out;
236 }
237 cfolio_out = folio_address(out_folio);
238 folios[nr_folios] = out_folio;
239 nr_folios++;
240 workspace->strm.avail_out = PAGE_SIZE;
241 workspace->strm.next_out = cfolio_out;
242 }
243 /* we're all done */
244 if (workspace->strm.total_in >= len)
245 break;
246 if (workspace->strm.total_out > max_out)
247 break;
248 }
249 workspace->strm.avail_in = 0;
250 /*
251 * Call deflate with Z_FINISH flush parameter providing more output
252 * space but no more input data, until it returns with Z_STREAM_END.
253 */
254 while (ret != Z_STREAM_END) {
255 ret = zlib_deflate(&workspace->strm, Z_FINISH);
256 if (ret == Z_STREAM_END)
257 break;
258 if (ret != Z_OK && ret != Z_BUF_ERROR) {
259 zlib_deflateEnd(&workspace->strm);
260 ret = -EIO;
261 goto out;
262 } else if (workspace->strm.avail_out == 0) {
263 /* Get another folio for the stream end. */
264 if (nr_folios == nr_dest_folios) {
265 ret = -E2BIG;
266 goto out;
267 }
268 out_folio = btrfs_alloc_compr_folio();
269 if (out_folio == NULL) {
270 ret = -ENOMEM;
271 goto out;
272 }
273 cfolio_out = folio_address(out_folio);
274 folios[nr_folios] = out_folio;
275 nr_folios++;
276 workspace->strm.avail_out = PAGE_SIZE;
277 workspace->strm.next_out = cfolio_out;
278 }
279 }
280 zlib_deflateEnd(&workspace->strm);
281
282 if (workspace->strm.total_out >= workspace->strm.total_in) {
283 ret = -E2BIG;
284 goto out;
285 }
286
287 ret = 0;
288 *total_out = workspace->strm.total_out;
289 *total_in = workspace->strm.total_in;
290 out:
291 *out_folios = nr_folios;
292 if (data_in) {
293 kunmap_local(data_in);
294 folio_put(in_folio);
295 }
296
297 return ret;
298 }
299
zlib_decompress_bio(struct list_head * ws,struct compressed_bio * cb)300 int zlib_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
301 {
302 struct workspace *workspace = list_entry(ws, struct workspace, list);
303 int ret = 0, ret2;
304 int wbits = MAX_WBITS;
305 char *data_in;
306 size_t total_out = 0;
307 unsigned long folio_in_index = 0;
308 size_t srclen = cb->compressed_len;
309 unsigned long total_folios_in = DIV_ROUND_UP(srclen, PAGE_SIZE);
310 unsigned long buf_start;
311 struct folio **folios_in = cb->compressed_folios;
312
313 data_in = kmap_local_folio(folios_in[folio_in_index], 0);
314 workspace->strm.next_in = data_in;
315 workspace->strm.avail_in = min_t(size_t, srclen, PAGE_SIZE);
316 workspace->strm.total_in = 0;
317
318 workspace->strm.total_out = 0;
319 workspace->strm.next_out = workspace->buf;
320 workspace->strm.avail_out = workspace->buf_size;
321
322 /* If it's deflate, and it's got no preset dictionary, then
323 we can tell zlib to skip the adler32 check. */
324 if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
325 ((data_in[0] & 0x0f) == Z_DEFLATED) &&
326 !(((data_in[0]<<8) + data_in[1]) % 31)) {
327
328 wbits = -((data_in[0] >> 4) + 8);
329 workspace->strm.next_in += 2;
330 workspace->strm.avail_in -= 2;
331 }
332
333 ret = zlib_inflateInit2(&workspace->strm, wbits);
334 if (unlikely(ret != Z_OK)) {
335 struct btrfs_inode *inode = cb->bbio.inode;
336
337 kunmap_local(data_in);
338 btrfs_err(inode->root->fs_info,
339 "zlib decompression init failed, error %d root %llu inode %llu offset %llu",
340 ret, btrfs_root_id(inode->root), btrfs_ino(inode), cb->start);
341 return -EIO;
342 }
343 while (workspace->strm.total_in < srclen) {
344 ret = zlib_inflate(&workspace->strm, Z_NO_FLUSH);
345 if (ret != Z_OK && ret != Z_STREAM_END)
346 break;
347
348 buf_start = total_out;
349 total_out = workspace->strm.total_out;
350
351 /* we didn't make progress in this inflate call, we're done */
352 if (buf_start == total_out)
353 break;
354
355 ret2 = btrfs_decompress_buf2page(workspace->buf,
356 total_out - buf_start, cb, buf_start);
357 if (ret2 == 0) {
358 ret = 0;
359 goto done;
360 }
361
362 workspace->strm.next_out = workspace->buf;
363 workspace->strm.avail_out = workspace->buf_size;
364
365 if (workspace->strm.avail_in == 0) {
366 unsigned long tmp;
367 kunmap_local(data_in);
368 folio_in_index++;
369 if (folio_in_index >= total_folios_in) {
370 data_in = NULL;
371 break;
372 }
373 data_in = kmap_local_folio(folios_in[folio_in_index], 0);
374 workspace->strm.next_in = data_in;
375 tmp = srclen - workspace->strm.total_in;
376 workspace->strm.avail_in = min(tmp, PAGE_SIZE);
377 }
378 }
379 if (unlikely(ret != Z_STREAM_END)) {
380 btrfs_err(cb->bbio.inode->root->fs_info,
381 "zlib decompression failed, error %d root %llu inode %llu offset %llu",
382 ret, btrfs_root_id(cb->bbio.inode->root),
383 btrfs_ino(cb->bbio.inode), cb->start);
384 ret = -EIO;
385 } else {
386 ret = 0;
387 }
388 done:
389 zlib_inflateEnd(&workspace->strm);
390 if (data_in)
391 kunmap_local(data_in);
392 return ret;
393 }
394
zlib_decompress(struct list_head * ws,const u8 * data_in,struct folio * dest_folio,unsigned long dest_pgoff,size_t srclen,size_t destlen)395 int zlib_decompress(struct list_head *ws, const u8 *data_in,
396 struct folio *dest_folio, unsigned long dest_pgoff, size_t srclen,
397 size_t destlen)
398 {
399 struct workspace *workspace = list_entry(ws, struct workspace, list);
400 int ret = 0;
401 int wbits = MAX_WBITS;
402 unsigned long to_copy;
403
404 workspace->strm.next_in = data_in;
405 workspace->strm.avail_in = srclen;
406 workspace->strm.total_in = 0;
407
408 workspace->strm.next_out = workspace->buf;
409 workspace->strm.avail_out = workspace->buf_size;
410 workspace->strm.total_out = 0;
411 /* If it's deflate, and it's got no preset dictionary, then
412 we can tell zlib to skip the adler32 check. */
413 if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
414 ((data_in[0] & 0x0f) == Z_DEFLATED) &&
415 !(((data_in[0]<<8) + data_in[1]) % 31)) {
416
417 wbits = -((data_in[0] >> 4) + 8);
418 workspace->strm.next_in += 2;
419 workspace->strm.avail_in -= 2;
420 }
421
422 ret = zlib_inflateInit2(&workspace->strm, wbits);
423 if (unlikely(ret != Z_OK)) {
424 struct btrfs_inode *inode = folio_to_inode(dest_folio);
425
426 btrfs_err(inode->root->fs_info,
427 "zlib decompression init failed, error %d root %llu inode %llu offset %llu",
428 ret, btrfs_root_id(inode->root), btrfs_ino(inode),
429 folio_pos(dest_folio));
430 return -EIO;
431 }
432
433 /*
434 * Everything (in/out buf) should be at most one sector, there should
435 * be no need to switch any input/output buffer.
436 */
437 ret = zlib_inflate(&workspace->strm, Z_FINISH);
438 to_copy = min(workspace->strm.total_out, destlen);
439 if (ret != Z_STREAM_END)
440 goto out;
441
442 memcpy_to_folio(dest_folio, dest_pgoff, workspace->buf, to_copy);
443
444 out:
445 if (unlikely(to_copy != destlen)) {
446 struct btrfs_inode *inode = folio_to_inode(dest_folio);
447
448 btrfs_err(inode->root->fs_info,
449 "zlib decompression failed, error %d root %llu inode %llu offset %llu decompressed %lu expected %zu",
450 ret, btrfs_root_id(inode->root), btrfs_ino(inode),
451 folio_pos(dest_folio), to_copy, destlen);
452 ret = -EIO;
453 } else {
454 ret = 0;
455 }
456
457 zlib_inflateEnd(&workspace->strm);
458
459 if (unlikely(to_copy < destlen))
460 folio_zero_range(dest_folio, dest_pgoff + to_copy, destlen - to_copy);
461 return ret;
462 }
463
464 const struct btrfs_compress_op btrfs_zlib_compress = {
465 .workspace_manager = &wsm,
466 .max_level = 9,
467 .default_level = BTRFS_ZLIB_DEFAULT_LEVEL,
468 };
469