]>
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) | |
30 | * Variable size. Each segment includes one segment header, followd by data | |
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 | ||
64 | static void lzo_free_workspace(struct list_head *ws) | |
65 | { | |
66 | struct workspace *workspace = list_entry(ws, struct workspace, list); | |
67 | ||
6acafd1e DS |
68 | kvfree(workspace->buf); |
69 | kvfree(workspace->cbuf); | |
70 | kvfree(workspace->mem); | |
a6fa6fae LZ |
71 | kfree(workspace); |
72 | } | |
73 | ||
74 | static struct list_head *lzo_alloc_workspace(void) | |
75 | { | |
76 | struct workspace *workspace; | |
77 | ||
389a6cfc | 78 | workspace = kzalloc(sizeof(*workspace), GFP_KERNEL); |
a6fa6fae LZ |
79 | if (!workspace) |
80 | return ERR_PTR(-ENOMEM); | |
81 | ||
6acafd1e DS |
82 | workspace->mem = kvmalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL); |
83 | workspace->buf = kvmalloc(lzo1x_worst_compress(PAGE_SIZE), GFP_KERNEL); | |
84 | workspace->cbuf = kvmalloc(lzo1x_worst_compress(PAGE_SIZE), GFP_KERNEL); | |
a6fa6fae LZ |
85 | if (!workspace->mem || !workspace->buf || !workspace->cbuf) |
86 | goto fail; | |
87 | ||
88 | INIT_LIST_HEAD(&workspace->list); | |
89 | ||
90 | return &workspace->list; | |
91 | fail: | |
92 | lzo_free_workspace(&workspace->list); | |
93 | return ERR_PTR(-ENOMEM); | |
94 | } | |
95 | ||
96 | static inline void write_compress_length(char *buf, size_t len) | |
97 | { | |
98 | __le32 dlen; | |
99 | ||
100 | dlen = cpu_to_le32(len); | |
101 | memcpy(buf, &dlen, LZO_LEN); | |
102 | } | |
103 | ||
14a3357b | 104 | static inline size_t read_compress_length(const char *buf) |
a6fa6fae LZ |
105 | { |
106 | __le32 dlen; | |
107 | ||
108 | memcpy(&dlen, buf, LZO_LEN); | |
109 | return le32_to_cpu(dlen); | |
110 | } | |
111 | ||
112 | static int lzo_compress_pages(struct list_head *ws, | |
113 | struct address_space *mapping, | |
38c31464 | 114 | u64 start, |
a6fa6fae | 115 | struct page **pages, |
a6fa6fae LZ |
116 | unsigned long *out_pages, |
117 | unsigned long *total_in, | |
e5d74902 | 118 | unsigned long *total_out) |
a6fa6fae LZ |
119 | { |
120 | struct workspace *workspace = list_entry(ws, struct workspace, list); | |
121 | int ret = 0; | |
122 | char *data_in; | |
123 | char *cpage_out; | |
124 | int nr_pages = 0; | |
125 | struct page *in_page = NULL; | |
126 | struct page *out_page = NULL; | |
127 | unsigned long bytes_left; | |
38c31464 | 128 | unsigned long len = *total_out; |
4d3a800e | 129 | unsigned long nr_dest_pages = *out_pages; |
e5d74902 | 130 | const unsigned long max_out = nr_dest_pages * PAGE_SIZE; |
a6fa6fae LZ |
131 | size_t in_len; |
132 | size_t out_len; | |
133 | char *buf; | |
134 | unsigned long tot_in = 0; | |
135 | unsigned long tot_out = 0; | |
136 | unsigned long pg_bytes_left; | |
137 | unsigned long out_offset; | |
138 | unsigned long bytes; | |
139 | ||
140 | *out_pages = 0; | |
141 | *total_out = 0; | |
142 | *total_in = 0; | |
143 | ||
09cbfeaf | 144 | in_page = find_get_page(mapping, start >> PAGE_SHIFT); |
a6fa6fae LZ |
145 | data_in = kmap(in_page); |
146 | ||
147 | /* | |
148 | * store the size of all chunks of compressed data in | |
149 | * the first 4 bytes | |
150 | */ | |
151 | out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
152 | if (out_page == NULL) { | |
153 | ret = -ENOMEM; | |
154 | goto out; | |
155 | } | |
156 | cpage_out = kmap(out_page); | |
157 | out_offset = LZO_LEN; | |
158 | tot_out = LZO_LEN; | |
159 | pages[0] = out_page; | |
160 | nr_pages = 1; | |
09cbfeaf | 161 | pg_bytes_left = PAGE_SIZE - LZO_LEN; |
a6fa6fae LZ |
162 | |
163 | /* compress at most one page of data each time */ | |
09cbfeaf | 164 | in_len = min(len, PAGE_SIZE); |
a6fa6fae LZ |
165 | while (tot_in < len) { |
166 | ret = lzo1x_1_compress(data_in, in_len, workspace->cbuf, | |
167 | &out_len, workspace->mem); | |
168 | if (ret != LZO_E_OK) { | |
036b0217 | 169 | pr_debug("BTRFS: lzo in loop returned %d\n", |
a6fa6fae | 170 | ret); |
60e1975a | 171 | ret = -EIO; |
a6fa6fae LZ |
172 | goto out; |
173 | } | |
174 | ||
175 | /* store the size of this chunk of compressed data */ | |
176 | write_compress_length(cpage_out + out_offset, out_len); | |
177 | tot_out += LZO_LEN; | |
178 | out_offset += LZO_LEN; | |
179 | pg_bytes_left -= LZO_LEN; | |
180 | ||
181 | tot_in += in_len; | |
182 | tot_out += out_len; | |
183 | ||
184 | /* copy bytes from the working buffer into the pages */ | |
185 | buf = workspace->cbuf; | |
186 | while (out_len) { | |
187 | bytes = min_t(unsigned long, pg_bytes_left, out_len); | |
188 | ||
189 | memcpy(cpage_out + out_offset, buf, bytes); | |
190 | ||
191 | out_len -= bytes; | |
192 | pg_bytes_left -= bytes; | |
193 | buf += bytes; | |
194 | out_offset += bytes; | |
195 | ||
196 | /* | |
197 | * we need another page for writing out. | |
198 | * | |
199 | * Note if there's less than 4 bytes left, we just | |
200 | * skip to a new page. | |
201 | */ | |
202 | if ((out_len == 0 && pg_bytes_left < LZO_LEN) || | |
203 | pg_bytes_left == 0) { | |
204 | if (pg_bytes_left) { | |
205 | memset(cpage_out + out_offset, 0, | |
206 | pg_bytes_left); | |
207 | tot_out += pg_bytes_left; | |
208 | } | |
209 | ||
210 | /* we're done, don't allocate new page */ | |
211 | if (out_len == 0 && tot_in >= len) | |
212 | break; | |
213 | ||
214 | kunmap(out_page); | |
215 | if (nr_pages == nr_dest_pages) { | |
216 | out_page = NULL; | |
60e1975a | 217 | ret = -E2BIG; |
a6fa6fae LZ |
218 | goto out; |
219 | } | |
220 | ||
221 | out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
222 | if (out_page == NULL) { | |
223 | ret = -ENOMEM; | |
224 | goto out; | |
225 | } | |
226 | cpage_out = kmap(out_page); | |
227 | pages[nr_pages++] = out_page; | |
228 | ||
09cbfeaf | 229 | pg_bytes_left = PAGE_SIZE; |
a6fa6fae LZ |
230 | out_offset = 0; |
231 | } | |
232 | } | |
233 | ||
234 | /* we're making it bigger, give up */ | |
59516f60 | 235 | if (tot_in > 8192 && tot_in < tot_out) { |
60e1975a | 236 | ret = -E2BIG; |
a6fa6fae | 237 | goto out; |
59516f60 | 238 | } |
a6fa6fae LZ |
239 | |
240 | /* we're all done */ | |
241 | if (tot_in >= len) | |
242 | break; | |
243 | ||
244 | if (tot_out > max_out) | |
245 | break; | |
246 | ||
247 | bytes_left = len - tot_in; | |
248 | kunmap(in_page); | |
09cbfeaf | 249 | put_page(in_page); |
a6fa6fae | 250 | |
09cbfeaf KS |
251 | start += PAGE_SIZE; |
252 | in_page = find_get_page(mapping, start >> PAGE_SHIFT); | |
a6fa6fae | 253 | data_in = kmap(in_page); |
09cbfeaf | 254 | in_len = min(bytes_left, PAGE_SIZE); |
a6fa6fae LZ |
255 | } |
256 | ||
1e9d7291 TT |
257 | if (tot_out >= tot_in) { |
258 | ret = -E2BIG; | |
a6fa6fae | 259 | goto out; |
1e9d7291 | 260 | } |
a6fa6fae LZ |
261 | |
262 | /* store the size of all chunks of compressed data */ | |
263 | cpage_out = kmap(pages[0]); | |
264 | write_compress_length(cpage_out, tot_out); | |
265 | ||
266 | kunmap(pages[0]); | |
267 | ||
268 | ret = 0; | |
269 | *total_out = tot_out; | |
270 | *total_in = tot_in; | |
271 | out: | |
272 | *out_pages = nr_pages; | |
273 | if (out_page) | |
274 | kunmap(out_page); | |
275 | ||
276 | if (in_page) { | |
277 | kunmap(in_page); | |
09cbfeaf | 278 | put_page(in_page); |
a6fa6fae LZ |
279 | } |
280 | ||
281 | return ret; | |
282 | } | |
283 | ||
e1ddce71 | 284 | static int lzo_decompress_bio(struct list_head *ws, struct compressed_bio *cb) |
a6fa6fae LZ |
285 | { |
286 | struct workspace *workspace = list_entry(ws, struct workspace, list); | |
3a39c18d | 287 | int ret = 0, ret2; |
a6fa6fae | 288 | char *data_in; |
a6fa6fae | 289 | unsigned long page_in_index = 0; |
e1ddce71 | 290 | size_t srclen = cb->compressed_len; |
09cbfeaf | 291 | unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE); |
a6fa6fae LZ |
292 | unsigned long buf_start; |
293 | unsigned long buf_offset = 0; | |
294 | unsigned long bytes; | |
295 | unsigned long working_bytes; | |
a6fa6fae LZ |
296 | size_t in_len; |
297 | size_t out_len; | |
314bfa47 | 298 | const size_t max_segment_len = lzo1x_worst_compress(PAGE_SIZE); |
a6fa6fae LZ |
299 | unsigned long in_offset; |
300 | unsigned long in_page_bytes_left; | |
301 | unsigned long tot_in; | |
302 | unsigned long tot_out; | |
303 | unsigned long tot_len; | |
304 | char *buf; | |
ca9b688c | 305 | bool may_late_unmap, need_unmap; |
e1ddce71 AJ |
306 | struct page **pages_in = cb->compressed_pages; |
307 | u64 disk_start = cb->start; | |
308 | struct bio *orig_bio = cb->orig_bio; | |
a6fa6fae LZ |
309 | |
310 | data_in = kmap(pages_in[0]); | |
311 | tot_len = read_compress_length(data_in); | |
314bfa47 QW |
312 | /* |
313 | * Compressed data header check. | |
314 | * | |
315 | * The real compressed size can't exceed the maximum extent length, and | |
316 | * all pages should be used (whole unused page with just the segment | |
317 | * header is not possible). If this happens it means the compressed | |
318 | * extent is corrupted. | |
319 | */ | |
320 | if (tot_len > min_t(size_t, BTRFS_MAX_COMPRESSED, srclen) || | |
321 | tot_len < srclen - PAGE_SIZE) { | |
322 | ret = -EUCLEAN; | |
323 | goto done; | |
324 | } | |
a6fa6fae LZ |
325 | |
326 | tot_in = LZO_LEN; | |
327 | in_offset = LZO_LEN; | |
09cbfeaf | 328 | in_page_bytes_left = PAGE_SIZE - LZO_LEN; |
a6fa6fae LZ |
329 | |
330 | tot_out = 0; | |
a6fa6fae LZ |
331 | |
332 | while (tot_in < tot_len) { | |
333 | in_len = read_compress_length(data_in + in_offset); | |
334 | in_page_bytes_left -= LZO_LEN; | |
335 | in_offset += LZO_LEN; | |
336 | tot_in += LZO_LEN; | |
337 | ||
314bfa47 QW |
338 | /* |
339 | * Segment header check. | |
340 | * | |
341 | * The segment length must not exceed the maximum LZO | |
342 | * compression size, nor the total compressed size. | |
343 | */ | |
344 | if (in_len > max_segment_len || tot_in + in_len > tot_len) { | |
345 | ret = -EUCLEAN; | |
346 | goto done; | |
347 | } | |
348 | ||
a6fa6fae LZ |
349 | tot_in += in_len; |
350 | working_bytes = in_len; | |
ca9b688c | 351 | may_late_unmap = need_unmap = false; |
a6fa6fae LZ |
352 | |
353 | /* fast path: avoid using the working buffer */ | |
354 | if (in_page_bytes_left >= in_len) { | |
355 | buf = data_in + in_offset; | |
356 | bytes = in_len; | |
ca9b688c | 357 | may_late_unmap = true; |
a6fa6fae LZ |
358 | goto cont; |
359 | } | |
360 | ||
361 | /* copy bytes from the pages into the working buffer */ | |
362 | buf = workspace->cbuf; | |
363 | buf_offset = 0; | |
364 | while (working_bytes) { | |
365 | bytes = min(working_bytes, in_page_bytes_left); | |
366 | ||
367 | memcpy(buf + buf_offset, data_in + in_offset, bytes); | |
368 | buf_offset += bytes; | |
369 | cont: | |
370 | working_bytes -= bytes; | |
371 | in_page_bytes_left -= bytes; | |
372 | in_offset += bytes; | |
373 | ||
374 | /* check if we need to pick another page */ | |
375 | if ((working_bytes == 0 && in_page_bytes_left < LZO_LEN) | |
376 | || in_page_bytes_left == 0) { | |
377 | tot_in += in_page_bytes_left; | |
378 | ||
379 | if (working_bytes == 0 && tot_in >= tot_len) | |
380 | break; | |
381 | ||
ca9b688c | 382 | if (page_in_index + 1 >= total_pages_in) { |
60e1975a | 383 | ret = -EIO; |
a6fa6fae LZ |
384 | goto done; |
385 | } | |
ca9b688c LZ |
386 | |
387 | if (may_late_unmap) | |
388 | need_unmap = true; | |
389 | else | |
390 | kunmap(pages_in[page_in_index]); | |
391 | ||
392 | data_in = kmap(pages_in[++page_in_index]); | |
a6fa6fae | 393 | |
09cbfeaf | 394 | in_page_bytes_left = PAGE_SIZE; |
a6fa6fae LZ |
395 | in_offset = 0; |
396 | } | |
397 | } | |
398 | ||
314bfa47 | 399 | out_len = max_segment_len; |
a6fa6fae LZ |
400 | ret = lzo1x_decompress_safe(buf, in_len, workspace->buf, |
401 | &out_len); | |
ca9b688c LZ |
402 | if (need_unmap) |
403 | kunmap(pages_in[page_in_index - 1]); | |
a6fa6fae | 404 | if (ret != LZO_E_OK) { |
62e85577 | 405 | pr_warn("BTRFS: decompress failed\n"); |
60e1975a | 406 | ret = -EIO; |
a6fa6fae LZ |
407 | break; |
408 | } | |
409 | ||
a6fa6fae | 410 | buf_start = tot_out; |
a6fa6fae LZ |
411 | tot_out += out_len; |
412 | ||
3a39c18d | 413 | ret2 = btrfs_decompress_buf2page(workspace->buf, buf_start, |
974b1adc | 414 | tot_out, disk_start, orig_bio); |
3a39c18d | 415 | if (ret2 == 0) |
a6fa6fae | 416 | break; |
a6fa6fae LZ |
417 | } |
418 | done: | |
ca9b688c | 419 | kunmap(pages_in[page_in_index]); |
2f19cad9 | 420 | if (!ret) |
974b1adc | 421 | zero_fill_bio(orig_bio); |
a6fa6fae LZ |
422 | return ret; |
423 | } | |
424 | ||
425 | static int lzo_decompress(struct list_head *ws, unsigned char *data_in, | |
426 | struct page *dest_page, | |
427 | unsigned long start_byte, | |
428 | size_t srclen, size_t destlen) | |
429 | { | |
430 | struct workspace *workspace = list_entry(ws, struct workspace, list); | |
431 | size_t in_len; | |
432 | size_t out_len; | |
de885e3e | 433 | size_t max_segment_len = lzo1x_worst_compress(PAGE_SIZE); |
a6fa6fae LZ |
434 | int ret = 0; |
435 | char *kaddr; | |
436 | unsigned long bytes; | |
437 | ||
de885e3e QW |
438 | if (srclen < LZO_LEN || srclen > max_segment_len + LZO_LEN * 2) |
439 | return -EUCLEAN; | |
a6fa6fae | 440 | |
de885e3e QW |
441 | in_len = read_compress_length(data_in); |
442 | if (in_len != srclen) | |
443 | return -EUCLEAN; | |
a6fa6fae LZ |
444 | data_in += LZO_LEN; |
445 | ||
446 | in_len = read_compress_length(data_in); | |
de885e3e QW |
447 | if (in_len != srclen - LZO_LEN * 2) { |
448 | ret = -EUCLEAN; | |
449 | goto out; | |
450 | } | |
a6fa6fae LZ |
451 | data_in += LZO_LEN; |
452 | ||
09cbfeaf | 453 | out_len = PAGE_SIZE; |
a6fa6fae LZ |
454 | ret = lzo1x_decompress_safe(data_in, in_len, workspace->buf, &out_len); |
455 | if (ret != LZO_E_OK) { | |
62e85577 | 456 | pr_warn("BTRFS: decompress failed!\n"); |
60e1975a | 457 | ret = -EIO; |
a6fa6fae LZ |
458 | goto out; |
459 | } | |
460 | ||
461 | if (out_len < start_byte) { | |
60e1975a | 462 | ret = -EIO; |
a6fa6fae LZ |
463 | goto out; |
464 | } | |
465 | ||
2f19cad9 CM |
466 | /* |
467 | * the caller is already checking against PAGE_SIZE, but lets | |
468 | * move this check closer to the memcpy/memset | |
469 | */ | |
470 | destlen = min_t(unsigned long, destlen, PAGE_SIZE); | |
a6fa6fae LZ |
471 | bytes = min_t(unsigned long, destlen, out_len - start_byte); |
472 | ||
7ac687d9 | 473 | kaddr = kmap_atomic(dest_page); |
a6fa6fae | 474 | memcpy(kaddr, workspace->buf + start_byte, bytes); |
2f19cad9 CM |
475 | |
476 | /* | |
477 | * btrfs_getblock is doing a zero on the tail of the page too, | |
478 | * but this will cover anything missing from the decompressed | |
479 | * data. | |
480 | */ | |
481 | if (bytes < destlen) | |
482 | memset(kaddr+bytes, 0, destlen-bytes); | |
7ac687d9 | 483 | kunmap_atomic(kaddr); |
a6fa6fae LZ |
484 | out: |
485 | return ret; | |
486 | } | |
487 | ||
f51d2b59 DS |
488 | static void lzo_set_level(struct list_head *ws, unsigned int type) |
489 | { | |
490 | } | |
491 | ||
e8c9f186 | 492 | const struct btrfs_compress_op btrfs_lzo_compress = { |
a6fa6fae LZ |
493 | .alloc_workspace = lzo_alloc_workspace, |
494 | .free_workspace = lzo_free_workspace, | |
495 | .compress_pages = lzo_compress_pages, | |
974b1adc | 496 | .decompress_bio = lzo_decompress_bio, |
a6fa6fae | 497 | .decompress = lzo_decompress, |
f51d2b59 | 498 | .set_level = lzo_set_level, |
a6fa6fae | 499 | }; |