]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
a6fa6fae LZ |
2 | /* |
3 | * Copyright (C) 2008 Oracle. All rights reserved. | |
a6fa6fae LZ |
4 | */ |
5 | ||
6 | #include <linux/kernel.h> | |
7 | #include <linux/slab.h> | |
6acafd1e | 8 | #include <linux/mm.h> |
a6fa6fae LZ |
9 | #include <linux/init.h> |
10 | #include <linux/err.h> | |
11 | #include <linux/sched.h> | |
12 | #include <linux/pagemap.h> | |
13 | #include <linux/bio.h> | |
14 | #include <linux/lzo.h> | |
e1ddce71 | 15 | #include <linux/refcount.h> |
a6fa6fae LZ |
16 | #include "compression.h" |
17 | ||
18 | #define LZO_LEN 4 | |
19 | ||
2a1f7c0c QW |
20 | /* |
21 | * Btrfs LZO compression format | |
22 | * | |
23 | * Regular and inlined LZO compressed data extents consist of: | |
24 | * | |
25 | * 1. Header | |
26 | * Fixed size. LZO_LEN (4) bytes long, LE32. | |
27 | * Records the total size (including the header) of compressed data. | |
28 | * | |
29 | * 2. Segment(s) | |
52042d8e | 30 | * Variable size. Each segment includes one segment header, followed by data |
2a1f7c0c QW |
31 | * payload. |
32 | * One regular LZO compressed extent can have one or more segments. | |
33 | * For inlined LZO compressed extent, only one segment is allowed. | |
34 | * One segment represents at most one page of uncompressed data. | |
35 | * | |
36 | * 2.1 Segment header | |
37 | * Fixed size. LZO_LEN (4) bytes long, LE32. | |
38 | * Records the total size of the segment (not including the header). | |
39 | * Segment header never crosses page boundary, thus it's possible to | |
40 | * have at most 3 padding zeros at the end of the page. | |
41 | * | |
42 | * 2.2 Data Payload | |
43 | * Variable size. Size up limit should be lzo1x_worst_compress(PAGE_SIZE) | |
44 | * which is 4419 for a 4KiB page. | |
45 | * | |
46 | * Example: | |
47 | * Page 1: | |
48 | * 0 0x2 0x4 0x6 0x8 0xa 0xc 0xe 0x10 | |
49 | * 0x0000 | Header | SegHdr 01 | Data payload 01 ... | | |
50 | * ... | |
51 | * 0x0ff0 | SegHdr N | Data payload N ... |00| | |
52 | * ^^ padding zeros | |
53 | * Page 2: | |
54 | * 0x1000 | SegHdr N+1| Data payload N+1 ... | | |
55 | */ | |
56 | ||
a6fa6fae LZ |
57 | struct workspace { |
58 | void *mem; | |
3fb40375 JL |
59 | void *buf; /* where decompressed data goes */ |
60 | void *cbuf; /* where compressed data goes */ | |
a6fa6fae LZ |
61 | struct list_head list; |
62 | }; | |
63 | ||
92ee5530 DZ |
64 | static struct workspace_manager wsm; |
65 | ||
d20f395f | 66 | void lzo_free_workspace(struct list_head *ws) |
a6fa6fae LZ |
67 | { |
68 | struct workspace *workspace = list_entry(ws, struct workspace, list); | |
69 | ||
6acafd1e DS |
70 | kvfree(workspace->buf); |
71 | kvfree(workspace->cbuf); | |
72 | kvfree(workspace->mem); | |
a6fa6fae LZ |
73 | kfree(workspace); |
74 | } | |
75 | ||
d20f395f | 76 | struct list_head *lzo_alloc_workspace(unsigned int level) |
a6fa6fae LZ |
77 | { |
78 | struct workspace *workspace; | |
79 | ||
389a6cfc | 80 | workspace = kzalloc(sizeof(*workspace), GFP_KERNEL); |
a6fa6fae LZ |
81 | if (!workspace) |
82 | return ERR_PTR(-ENOMEM); | |
83 | ||
6acafd1e DS |
84 | workspace->mem = kvmalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL); |
85 | workspace->buf = kvmalloc(lzo1x_worst_compress(PAGE_SIZE), GFP_KERNEL); | |
86 | workspace->cbuf = kvmalloc(lzo1x_worst_compress(PAGE_SIZE), GFP_KERNEL); | |
a6fa6fae LZ |
87 | if (!workspace->mem || !workspace->buf || !workspace->cbuf) |
88 | goto fail; | |
89 | ||
90 | INIT_LIST_HEAD(&workspace->list); | |
91 | ||
92 | return &workspace->list; | |
93 | fail: | |
94 | lzo_free_workspace(&workspace->list); | |
95 | return ERR_PTR(-ENOMEM); | |
96 | } | |
97 | ||
98 | static inline void write_compress_length(char *buf, size_t len) | |
99 | { | |
100 | __le32 dlen; | |
101 | ||
102 | dlen = cpu_to_le32(len); | |
103 | memcpy(buf, &dlen, LZO_LEN); | |
104 | } | |
105 | ||
14a3357b | 106 | static inline size_t read_compress_length(const char *buf) |
a6fa6fae LZ |
107 | { |
108 | __le32 dlen; | |
109 | ||
110 | memcpy(&dlen, buf, LZO_LEN); | |
111 | return le32_to_cpu(dlen); | |
112 | } | |
113 | ||
c4bf665a DS |
114 | int lzo_compress_pages(struct list_head *ws, struct address_space *mapping, |
115 | u64 start, struct page **pages, unsigned long *out_pages, | |
116 | unsigned long *total_in, unsigned long *total_out) | |
a6fa6fae LZ |
117 | { |
118 | struct workspace *workspace = list_entry(ws, struct workspace, list); | |
119 | int ret = 0; | |
120 | char *data_in; | |
121 | char *cpage_out; | |
122 | int nr_pages = 0; | |
123 | struct page *in_page = NULL; | |
124 | struct page *out_page = NULL; | |
125 | unsigned long bytes_left; | |
38c31464 | 126 | unsigned long len = *total_out; |
4d3a800e | 127 | unsigned long nr_dest_pages = *out_pages; |
e5d74902 | 128 | const unsigned long max_out = nr_dest_pages * PAGE_SIZE; |
a6fa6fae LZ |
129 | size_t in_len; |
130 | size_t out_len; | |
131 | char *buf; | |
132 | unsigned long tot_in = 0; | |
133 | unsigned long tot_out = 0; | |
134 | unsigned long pg_bytes_left; | |
135 | unsigned long out_offset; | |
136 | unsigned long bytes; | |
137 | ||
138 | *out_pages = 0; | |
139 | *total_out = 0; | |
140 | *total_in = 0; | |
141 | ||
09cbfeaf | 142 | in_page = find_get_page(mapping, start >> PAGE_SHIFT); |
a6fa6fae LZ |
143 | data_in = kmap(in_page); |
144 | ||
145 | /* | |
146 | * store the size of all chunks of compressed data in | |
147 | * the first 4 bytes | |
148 | */ | |
149 | out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
150 | if (out_page == NULL) { | |
151 | ret = -ENOMEM; | |
152 | goto out; | |
153 | } | |
154 | cpage_out = kmap(out_page); | |
155 | out_offset = LZO_LEN; | |
156 | tot_out = LZO_LEN; | |
157 | pages[0] = out_page; | |
158 | nr_pages = 1; | |
09cbfeaf | 159 | pg_bytes_left = PAGE_SIZE - LZO_LEN; |
a6fa6fae LZ |
160 | |
161 | /* compress at most one page of data each time */ | |
09cbfeaf | 162 | in_len = min(len, PAGE_SIZE); |
a6fa6fae LZ |
163 | while (tot_in < len) { |
164 | ret = lzo1x_1_compress(data_in, in_len, workspace->cbuf, | |
165 | &out_len, workspace->mem); | |
166 | if (ret != LZO_E_OK) { | |
036b0217 | 167 | pr_debug("BTRFS: lzo in loop returned %d\n", |
a6fa6fae | 168 | ret); |
60e1975a | 169 | ret = -EIO; |
a6fa6fae LZ |
170 | goto out; |
171 | } | |
172 | ||
173 | /* store the size of this chunk of compressed data */ | |
174 | write_compress_length(cpage_out + out_offset, out_len); | |
175 | tot_out += LZO_LEN; | |
176 | out_offset += LZO_LEN; | |
177 | pg_bytes_left -= LZO_LEN; | |
178 | ||
179 | tot_in += in_len; | |
180 | tot_out += out_len; | |
181 | ||
182 | /* copy bytes from the working buffer into the pages */ | |
183 | buf = workspace->cbuf; | |
184 | while (out_len) { | |
185 | bytes = min_t(unsigned long, pg_bytes_left, out_len); | |
186 | ||
187 | memcpy(cpage_out + out_offset, buf, bytes); | |
188 | ||
189 | out_len -= bytes; | |
190 | pg_bytes_left -= bytes; | |
191 | buf += bytes; | |
192 | out_offset += bytes; | |
193 | ||
194 | /* | |
195 | * we need another page for writing out. | |
196 | * | |
197 | * Note if there's less than 4 bytes left, we just | |
198 | * skip to a new page. | |
199 | */ | |
200 | if ((out_len == 0 && pg_bytes_left < LZO_LEN) || | |
201 | pg_bytes_left == 0) { | |
202 | if (pg_bytes_left) { | |
203 | memset(cpage_out + out_offset, 0, | |
204 | pg_bytes_left); | |
205 | tot_out += pg_bytes_left; | |
206 | } | |
207 | ||
208 | /* we're done, don't allocate new page */ | |
209 | if (out_len == 0 && tot_in >= len) | |
210 | break; | |
211 | ||
212 | kunmap(out_page); | |
213 | if (nr_pages == nr_dest_pages) { | |
214 | out_page = NULL; | |
60e1975a | 215 | ret = -E2BIG; |
a6fa6fae LZ |
216 | goto out; |
217 | } | |
218 | ||
219 | out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
220 | if (out_page == NULL) { | |
221 | ret = -ENOMEM; | |
222 | goto out; | |
223 | } | |
224 | cpage_out = kmap(out_page); | |
225 | pages[nr_pages++] = out_page; | |
226 | ||
09cbfeaf | 227 | pg_bytes_left = PAGE_SIZE; |
a6fa6fae LZ |
228 | out_offset = 0; |
229 | } | |
230 | } | |
231 | ||
232 | /* we're making it bigger, give up */ | |
59516f60 | 233 | if (tot_in > 8192 && tot_in < tot_out) { |
60e1975a | 234 | ret = -E2BIG; |
a6fa6fae | 235 | goto out; |
59516f60 | 236 | } |
a6fa6fae LZ |
237 | |
238 | /* we're all done */ | |
239 | if (tot_in >= len) | |
240 | break; | |
241 | ||
242 | if (tot_out > max_out) | |
243 | break; | |
244 | ||
245 | bytes_left = len - tot_in; | |
246 | kunmap(in_page); | |
09cbfeaf | 247 | put_page(in_page); |
a6fa6fae | 248 | |
09cbfeaf KS |
249 | start += PAGE_SIZE; |
250 | in_page = find_get_page(mapping, start >> PAGE_SHIFT); | |
a6fa6fae | 251 | data_in = kmap(in_page); |
09cbfeaf | 252 | in_len = min(bytes_left, PAGE_SIZE); |
a6fa6fae LZ |
253 | } |
254 | ||
1e9d7291 TT |
255 | if (tot_out >= tot_in) { |
256 | ret = -E2BIG; | |
a6fa6fae | 257 | goto out; |
1e9d7291 | 258 | } |
a6fa6fae LZ |
259 | |
260 | /* store the size of all chunks of compressed data */ | |
261 | cpage_out = kmap(pages[0]); | |
262 | write_compress_length(cpage_out, tot_out); | |
263 | ||
264 | kunmap(pages[0]); | |
265 | ||
266 | ret = 0; | |
267 | *total_out = tot_out; | |
268 | *total_in = tot_in; | |
269 | out: | |
270 | *out_pages = nr_pages; | |
271 | if (out_page) | |
272 | kunmap(out_page); | |
273 | ||
274 | if (in_page) { | |
275 | kunmap(in_page); | |
09cbfeaf | 276 | put_page(in_page); |
a6fa6fae LZ |
277 | } |
278 | ||
279 | return ret; | |
280 | } | |
281 | ||
c4bf665a | 282 | int lzo_decompress_bio(struct list_head *ws, struct compressed_bio *cb) |
a6fa6fae LZ |
283 | { |
284 | struct workspace *workspace = list_entry(ws, struct workspace, list); | |
3a39c18d | 285 | int ret = 0, ret2; |
a6fa6fae | 286 | char *data_in; |
a6fa6fae | 287 | unsigned long page_in_index = 0; |
e1ddce71 | 288 | size_t srclen = cb->compressed_len; |
09cbfeaf | 289 | unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE); |
a6fa6fae LZ |
290 | unsigned long buf_start; |
291 | unsigned long buf_offset = 0; | |
292 | unsigned long bytes; | |
293 | unsigned long working_bytes; | |
a6fa6fae LZ |
294 | size_t in_len; |
295 | size_t out_len; | |
314bfa47 | 296 | const size_t max_segment_len = lzo1x_worst_compress(PAGE_SIZE); |
a6fa6fae LZ |
297 | unsigned long in_offset; |
298 | unsigned long in_page_bytes_left; | |
299 | unsigned long tot_in; | |
300 | unsigned long tot_out; | |
301 | unsigned long tot_len; | |
302 | char *buf; | |
ca9b688c | 303 | bool may_late_unmap, need_unmap; |
e1ddce71 AJ |
304 | struct page **pages_in = cb->compressed_pages; |
305 | u64 disk_start = cb->start; | |
306 | struct bio *orig_bio = cb->orig_bio; | |
a6fa6fae LZ |
307 | |
308 | data_in = kmap(pages_in[0]); | |
309 | tot_len = read_compress_length(data_in); | |
314bfa47 QW |
310 | /* |
311 | * Compressed data header check. | |
312 | * | |
313 | * The real compressed size can't exceed the maximum extent length, and | |
314 | * all pages should be used (whole unused page with just the segment | |
315 | * header is not possible). If this happens it means the compressed | |
316 | * extent is corrupted. | |
317 | */ | |
318 | if (tot_len > min_t(size_t, BTRFS_MAX_COMPRESSED, srclen) || | |
319 | tot_len < srclen - PAGE_SIZE) { | |
320 | ret = -EUCLEAN; | |
321 | goto done; | |
322 | } | |
a6fa6fae LZ |
323 | |
324 | tot_in = LZO_LEN; | |
325 | in_offset = LZO_LEN; | |
09cbfeaf | 326 | in_page_bytes_left = PAGE_SIZE - LZO_LEN; |
a6fa6fae LZ |
327 | |
328 | tot_out = 0; | |
a6fa6fae LZ |
329 | |
330 | while (tot_in < tot_len) { | |
331 | in_len = read_compress_length(data_in + in_offset); | |
332 | in_page_bytes_left -= LZO_LEN; | |
333 | in_offset += LZO_LEN; | |
334 | tot_in += LZO_LEN; | |
335 | ||
314bfa47 QW |
336 | /* |
337 | * Segment header check. | |
338 | * | |
339 | * The segment length must not exceed the maximum LZO | |
340 | * compression size, nor the total compressed size. | |
341 | */ | |
342 | if (in_len > max_segment_len || tot_in + in_len > tot_len) { | |
343 | ret = -EUCLEAN; | |
344 | goto done; | |
345 | } | |
346 | ||
a6fa6fae LZ |
347 | tot_in += in_len; |
348 | working_bytes = in_len; | |
ca9b688c | 349 | may_late_unmap = need_unmap = false; |
a6fa6fae LZ |
350 | |
351 | /* fast path: avoid using the working buffer */ | |
352 | if (in_page_bytes_left >= in_len) { | |
353 | buf = data_in + in_offset; | |
354 | bytes = in_len; | |
ca9b688c | 355 | may_late_unmap = true; |
a6fa6fae LZ |
356 | goto cont; |
357 | } | |
358 | ||
359 | /* copy bytes from the pages into the working buffer */ | |
360 | buf = workspace->cbuf; | |
361 | buf_offset = 0; | |
362 | while (working_bytes) { | |
363 | bytes = min(working_bytes, in_page_bytes_left); | |
364 | ||
365 | memcpy(buf + buf_offset, data_in + in_offset, bytes); | |
366 | buf_offset += bytes; | |
367 | cont: | |
368 | working_bytes -= bytes; | |
369 | in_page_bytes_left -= bytes; | |
370 | in_offset += bytes; | |
371 | ||
372 | /* check if we need to pick another page */ | |
373 | if ((working_bytes == 0 && in_page_bytes_left < LZO_LEN) | |
374 | || in_page_bytes_left == 0) { | |
375 | tot_in += in_page_bytes_left; | |
376 | ||
377 | if (working_bytes == 0 && tot_in >= tot_len) | |
378 | break; | |
379 | ||
ca9b688c | 380 | if (page_in_index + 1 >= total_pages_in) { |
60e1975a | 381 | ret = -EIO; |
a6fa6fae LZ |
382 | goto done; |
383 | } | |
ca9b688c LZ |
384 | |
385 | if (may_late_unmap) | |
386 | need_unmap = true; | |
387 | else | |
388 | kunmap(pages_in[page_in_index]); | |
389 | ||
390 | data_in = kmap(pages_in[++page_in_index]); | |
a6fa6fae | 391 | |
09cbfeaf | 392 | in_page_bytes_left = PAGE_SIZE; |
a6fa6fae LZ |
393 | in_offset = 0; |
394 | } | |
395 | } | |
396 | ||
314bfa47 | 397 | out_len = max_segment_len; |
a6fa6fae LZ |
398 | ret = lzo1x_decompress_safe(buf, in_len, workspace->buf, |
399 | &out_len); | |
ca9b688c LZ |
400 | if (need_unmap) |
401 | kunmap(pages_in[page_in_index - 1]); | |
a6fa6fae | 402 | if (ret != LZO_E_OK) { |
62e85577 | 403 | pr_warn("BTRFS: decompress failed\n"); |
60e1975a | 404 | ret = -EIO; |
a6fa6fae LZ |
405 | break; |
406 | } | |
407 | ||
a6fa6fae | 408 | buf_start = tot_out; |
a6fa6fae LZ |
409 | tot_out += out_len; |
410 | ||
3a39c18d | 411 | ret2 = btrfs_decompress_buf2page(workspace->buf, buf_start, |
974b1adc | 412 | tot_out, disk_start, orig_bio); |
3a39c18d | 413 | if (ret2 == 0) |
a6fa6fae | 414 | break; |
a6fa6fae LZ |
415 | } |
416 | done: | |
ca9b688c | 417 | kunmap(pages_in[page_in_index]); |
2f19cad9 | 418 | if (!ret) |
974b1adc | 419 | zero_fill_bio(orig_bio); |
a6fa6fae LZ |
420 | return ret; |
421 | } | |
422 | ||
c4bf665a DS |
423 | int lzo_decompress(struct list_head *ws, unsigned char *data_in, |
424 | struct page *dest_page, unsigned long start_byte, size_t srclen, | |
425 | size_t destlen) | |
a6fa6fae LZ |
426 | { |
427 | struct workspace *workspace = list_entry(ws, struct workspace, list); | |
428 | size_t in_len; | |
429 | size_t out_len; | |
de885e3e | 430 | size_t max_segment_len = lzo1x_worst_compress(PAGE_SIZE); |
a6fa6fae LZ |
431 | int ret = 0; |
432 | char *kaddr; | |
433 | unsigned long bytes; | |
434 | ||
de885e3e QW |
435 | if (srclen < LZO_LEN || srclen > max_segment_len + LZO_LEN * 2) |
436 | return -EUCLEAN; | |
a6fa6fae | 437 | |
de885e3e QW |
438 | in_len = read_compress_length(data_in); |
439 | if (in_len != srclen) | |
440 | return -EUCLEAN; | |
a6fa6fae LZ |
441 | data_in += LZO_LEN; |
442 | ||
443 | in_len = read_compress_length(data_in); | |
de885e3e QW |
444 | if (in_len != srclen - LZO_LEN * 2) { |
445 | ret = -EUCLEAN; | |
446 | goto out; | |
447 | } | |
a6fa6fae LZ |
448 | data_in += LZO_LEN; |
449 | ||
09cbfeaf | 450 | out_len = PAGE_SIZE; |
a6fa6fae LZ |
451 | ret = lzo1x_decompress_safe(data_in, in_len, workspace->buf, &out_len); |
452 | if (ret != LZO_E_OK) { | |
62e85577 | 453 | pr_warn("BTRFS: decompress failed!\n"); |
60e1975a | 454 | ret = -EIO; |
a6fa6fae LZ |
455 | goto out; |
456 | } | |
457 | ||
458 | if (out_len < start_byte) { | |
60e1975a | 459 | ret = -EIO; |
a6fa6fae LZ |
460 | goto out; |
461 | } | |
462 | ||
2f19cad9 CM |
463 | /* |
464 | * the caller is already checking against PAGE_SIZE, but lets | |
465 | * move this check closer to the memcpy/memset | |
466 | */ | |
467 | destlen = min_t(unsigned long, destlen, PAGE_SIZE); | |
a6fa6fae LZ |
468 | bytes = min_t(unsigned long, destlen, out_len - start_byte); |
469 | ||
7ac687d9 | 470 | kaddr = kmap_atomic(dest_page); |
a6fa6fae | 471 | memcpy(kaddr, workspace->buf + start_byte, bytes); |
2f19cad9 CM |
472 | |
473 | /* | |
474 | * btrfs_getblock is doing a zero on the tail of the page too, | |
475 | * but this will cover anything missing from the decompressed | |
476 | * data. | |
477 | */ | |
478 | if (bytes < destlen) | |
479 | memset(kaddr+bytes, 0, destlen-bytes); | |
7ac687d9 | 480 | kunmap_atomic(kaddr); |
a6fa6fae LZ |
481 | out: |
482 | return ret; | |
483 | } | |
484 | ||
e8c9f186 | 485 | const struct btrfs_compress_op btrfs_lzo_compress = { |
be951045 | 486 | .workspace_manager = &wsm, |
e18333a7 DS |
487 | .max_level = 1, |
488 | .default_level = 1, | |
a6fa6fae | 489 | }; |