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 | ||
66 | static void lzo_init_workspace_manager(void) | |
67 | { | |
68 | btrfs_init_workspace_manager(&wsm, &btrfs_lzo_compress); | |
69 | } | |
70 | ||
71 | static void lzo_cleanup_workspace_manager(void) | |
72 | { | |
73 | btrfs_cleanup_workspace_manager(&wsm); | |
74 | } | |
75 | ||
76 | static struct list_head *lzo_get_workspace(void) | |
77 | { | |
78 | return btrfs_get_workspace(&wsm); | |
79 | } | |
80 | ||
81 | static void lzo_put_workspace(struct list_head *ws) | |
82 | { | |
83 | btrfs_put_workspace(&wsm, ws); | |
84 | } | |
85 | ||
a6fa6fae LZ |
86 | static void lzo_free_workspace(struct list_head *ws) |
87 | { | |
88 | struct workspace *workspace = list_entry(ws, struct workspace, list); | |
89 | ||
6acafd1e DS |
90 | kvfree(workspace->buf); |
91 | kvfree(workspace->cbuf); | |
92 | kvfree(workspace->mem); | |
a6fa6fae LZ |
93 | kfree(workspace); |
94 | } | |
95 | ||
96 | static struct list_head *lzo_alloc_workspace(void) | |
97 | { | |
98 | struct workspace *workspace; | |
99 | ||
389a6cfc | 100 | workspace = kzalloc(sizeof(*workspace), GFP_KERNEL); |
a6fa6fae LZ |
101 | if (!workspace) |
102 | return ERR_PTR(-ENOMEM); | |
103 | ||
6acafd1e DS |
104 | workspace->mem = kvmalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL); |
105 | workspace->buf = kvmalloc(lzo1x_worst_compress(PAGE_SIZE), GFP_KERNEL); | |
106 | workspace->cbuf = kvmalloc(lzo1x_worst_compress(PAGE_SIZE), GFP_KERNEL); | |
a6fa6fae LZ |
107 | if (!workspace->mem || !workspace->buf || !workspace->cbuf) |
108 | goto fail; | |
109 | ||
110 | INIT_LIST_HEAD(&workspace->list); | |
111 | ||
112 | return &workspace->list; | |
113 | fail: | |
114 | lzo_free_workspace(&workspace->list); | |
115 | return ERR_PTR(-ENOMEM); | |
116 | } | |
117 | ||
118 | static inline void write_compress_length(char *buf, size_t len) | |
119 | { | |
120 | __le32 dlen; | |
121 | ||
122 | dlen = cpu_to_le32(len); | |
123 | memcpy(buf, &dlen, LZO_LEN); | |
124 | } | |
125 | ||
14a3357b | 126 | static inline size_t read_compress_length(const char *buf) |
a6fa6fae LZ |
127 | { |
128 | __le32 dlen; | |
129 | ||
130 | memcpy(&dlen, buf, LZO_LEN); | |
131 | return le32_to_cpu(dlen); | |
132 | } | |
133 | ||
134 | static int lzo_compress_pages(struct list_head *ws, | |
135 | struct address_space *mapping, | |
38c31464 | 136 | u64 start, |
a6fa6fae | 137 | struct page **pages, |
a6fa6fae LZ |
138 | unsigned long *out_pages, |
139 | unsigned long *total_in, | |
e5d74902 | 140 | unsigned long *total_out) |
a6fa6fae LZ |
141 | { |
142 | struct workspace *workspace = list_entry(ws, struct workspace, list); | |
143 | int ret = 0; | |
144 | char *data_in; | |
145 | char *cpage_out; | |
146 | int nr_pages = 0; | |
147 | struct page *in_page = NULL; | |
148 | struct page *out_page = NULL; | |
149 | unsigned long bytes_left; | |
38c31464 | 150 | unsigned long len = *total_out; |
4d3a800e | 151 | unsigned long nr_dest_pages = *out_pages; |
e5d74902 | 152 | const unsigned long max_out = nr_dest_pages * PAGE_SIZE; |
a6fa6fae LZ |
153 | size_t in_len; |
154 | size_t out_len; | |
155 | char *buf; | |
156 | unsigned long tot_in = 0; | |
157 | unsigned long tot_out = 0; | |
158 | unsigned long pg_bytes_left; | |
159 | unsigned long out_offset; | |
160 | unsigned long bytes; | |
161 | ||
162 | *out_pages = 0; | |
163 | *total_out = 0; | |
164 | *total_in = 0; | |
165 | ||
09cbfeaf | 166 | in_page = find_get_page(mapping, start >> PAGE_SHIFT); |
a6fa6fae LZ |
167 | data_in = kmap(in_page); |
168 | ||
169 | /* | |
170 | * store the size of all chunks of compressed data in | |
171 | * the first 4 bytes | |
172 | */ | |
173 | out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
174 | if (out_page == NULL) { | |
175 | ret = -ENOMEM; | |
176 | goto out; | |
177 | } | |
178 | cpage_out = kmap(out_page); | |
179 | out_offset = LZO_LEN; | |
180 | tot_out = LZO_LEN; | |
181 | pages[0] = out_page; | |
182 | nr_pages = 1; | |
09cbfeaf | 183 | pg_bytes_left = PAGE_SIZE - LZO_LEN; |
a6fa6fae LZ |
184 | |
185 | /* compress at most one page of data each time */ | |
09cbfeaf | 186 | in_len = min(len, PAGE_SIZE); |
a6fa6fae LZ |
187 | while (tot_in < len) { |
188 | ret = lzo1x_1_compress(data_in, in_len, workspace->cbuf, | |
189 | &out_len, workspace->mem); | |
190 | if (ret != LZO_E_OK) { | |
036b0217 | 191 | pr_debug("BTRFS: lzo in loop returned %d\n", |
a6fa6fae | 192 | ret); |
60e1975a | 193 | ret = -EIO; |
a6fa6fae LZ |
194 | goto out; |
195 | } | |
196 | ||
197 | /* store the size of this chunk of compressed data */ | |
198 | write_compress_length(cpage_out + out_offset, out_len); | |
199 | tot_out += LZO_LEN; | |
200 | out_offset += LZO_LEN; | |
201 | pg_bytes_left -= LZO_LEN; | |
202 | ||
203 | tot_in += in_len; | |
204 | tot_out += out_len; | |
205 | ||
206 | /* copy bytes from the working buffer into the pages */ | |
207 | buf = workspace->cbuf; | |
208 | while (out_len) { | |
209 | bytes = min_t(unsigned long, pg_bytes_left, out_len); | |
210 | ||
211 | memcpy(cpage_out + out_offset, buf, bytes); | |
212 | ||
213 | out_len -= bytes; | |
214 | pg_bytes_left -= bytes; | |
215 | buf += bytes; | |
216 | out_offset += bytes; | |
217 | ||
218 | /* | |
219 | * we need another page for writing out. | |
220 | * | |
221 | * Note if there's less than 4 bytes left, we just | |
222 | * skip to a new page. | |
223 | */ | |
224 | if ((out_len == 0 && pg_bytes_left < LZO_LEN) || | |
225 | pg_bytes_left == 0) { | |
226 | if (pg_bytes_left) { | |
227 | memset(cpage_out + out_offset, 0, | |
228 | pg_bytes_left); | |
229 | tot_out += pg_bytes_left; | |
230 | } | |
231 | ||
232 | /* we're done, don't allocate new page */ | |
233 | if (out_len == 0 && tot_in >= len) | |
234 | break; | |
235 | ||
236 | kunmap(out_page); | |
237 | if (nr_pages == nr_dest_pages) { | |
238 | out_page = NULL; | |
60e1975a | 239 | ret = -E2BIG; |
a6fa6fae LZ |
240 | goto out; |
241 | } | |
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 | ||
09cbfeaf | 251 | pg_bytes_left = PAGE_SIZE; |
a6fa6fae LZ |
252 | out_offset = 0; |
253 | } | |
254 | } | |
255 | ||
256 | /* we're making it bigger, give up */ | |
59516f60 | 257 | if (tot_in > 8192 && tot_in < tot_out) { |
60e1975a | 258 | ret = -E2BIG; |
a6fa6fae | 259 | goto out; |
59516f60 | 260 | } |
a6fa6fae LZ |
261 | |
262 | /* we're all done */ | |
263 | if (tot_in >= len) | |
264 | break; | |
265 | ||
266 | if (tot_out > max_out) | |
267 | break; | |
268 | ||
269 | bytes_left = len - tot_in; | |
270 | kunmap(in_page); | |
09cbfeaf | 271 | put_page(in_page); |
a6fa6fae | 272 | |
09cbfeaf KS |
273 | start += PAGE_SIZE; |
274 | in_page = find_get_page(mapping, start >> PAGE_SHIFT); | |
a6fa6fae | 275 | data_in = kmap(in_page); |
09cbfeaf | 276 | in_len = min(bytes_left, PAGE_SIZE); |
a6fa6fae LZ |
277 | } |
278 | ||
1e9d7291 TT |
279 | if (tot_out >= tot_in) { |
280 | ret = -E2BIG; | |
a6fa6fae | 281 | goto out; |
1e9d7291 | 282 | } |
a6fa6fae LZ |
283 | |
284 | /* store the size of all chunks of compressed data */ | |
285 | cpage_out = kmap(pages[0]); | |
286 | write_compress_length(cpage_out, tot_out); | |
287 | ||
288 | kunmap(pages[0]); | |
289 | ||
290 | ret = 0; | |
291 | *total_out = tot_out; | |
292 | *total_in = tot_in; | |
293 | out: | |
294 | *out_pages = nr_pages; | |
295 | if (out_page) | |
296 | kunmap(out_page); | |
297 | ||
298 | if (in_page) { | |
299 | kunmap(in_page); | |
09cbfeaf | 300 | put_page(in_page); |
a6fa6fae LZ |
301 | } |
302 | ||
303 | return ret; | |
304 | } | |
305 | ||
e1ddce71 | 306 | static int lzo_decompress_bio(struct list_head *ws, struct compressed_bio *cb) |
a6fa6fae LZ |
307 | { |
308 | struct workspace *workspace = list_entry(ws, struct workspace, list); | |
3a39c18d | 309 | int ret = 0, ret2; |
a6fa6fae | 310 | char *data_in; |
a6fa6fae | 311 | unsigned long page_in_index = 0; |
e1ddce71 | 312 | size_t srclen = cb->compressed_len; |
09cbfeaf | 313 | unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE); |
a6fa6fae LZ |
314 | unsigned long buf_start; |
315 | unsigned long buf_offset = 0; | |
316 | unsigned long bytes; | |
317 | unsigned long working_bytes; | |
a6fa6fae LZ |
318 | size_t in_len; |
319 | size_t out_len; | |
314bfa47 | 320 | const size_t max_segment_len = lzo1x_worst_compress(PAGE_SIZE); |
a6fa6fae LZ |
321 | unsigned long in_offset; |
322 | unsigned long in_page_bytes_left; | |
323 | unsigned long tot_in; | |
324 | unsigned long tot_out; | |
325 | unsigned long tot_len; | |
326 | char *buf; | |
ca9b688c | 327 | bool may_late_unmap, need_unmap; |
e1ddce71 AJ |
328 | struct page **pages_in = cb->compressed_pages; |
329 | u64 disk_start = cb->start; | |
330 | struct bio *orig_bio = cb->orig_bio; | |
a6fa6fae LZ |
331 | |
332 | data_in = kmap(pages_in[0]); | |
333 | tot_len = read_compress_length(data_in); | |
314bfa47 QW |
334 | /* |
335 | * Compressed data header check. | |
336 | * | |
337 | * The real compressed size can't exceed the maximum extent length, and | |
338 | * all pages should be used (whole unused page with just the segment | |
339 | * header is not possible). If this happens it means the compressed | |
340 | * extent is corrupted. | |
341 | */ | |
342 | if (tot_len > min_t(size_t, BTRFS_MAX_COMPRESSED, srclen) || | |
343 | tot_len < srclen - PAGE_SIZE) { | |
344 | ret = -EUCLEAN; | |
345 | goto done; | |
346 | } | |
a6fa6fae LZ |
347 | |
348 | tot_in = LZO_LEN; | |
349 | in_offset = LZO_LEN; | |
09cbfeaf | 350 | in_page_bytes_left = PAGE_SIZE - LZO_LEN; |
a6fa6fae LZ |
351 | |
352 | tot_out = 0; | |
a6fa6fae LZ |
353 | |
354 | while (tot_in < tot_len) { | |
355 | in_len = read_compress_length(data_in + in_offset); | |
356 | in_page_bytes_left -= LZO_LEN; | |
357 | in_offset += LZO_LEN; | |
358 | tot_in += LZO_LEN; | |
359 | ||
314bfa47 QW |
360 | /* |
361 | * Segment header check. | |
362 | * | |
363 | * The segment length must not exceed the maximum LZO | |
364 | * compression size, nor the total compressed size. | |
365 | */ | |
366 | if (in_len > max_segment_len || tot_in + in_len > tot_len) { | |
367 | ret = -EUCLEAN; | |
368 | goto done; | |
369 | } | |
370 | ||
a6fa6fae LZ |
371 | tot_in += in_len; |
372 | working_bytes = in_len; | |
ca9b688c | 373 | may_late_unmap = need_unmap = false; |
a6fa6fae LZ |
374 | |
375 | /* fast path: avoid using the working buffer */ | |
376 | if (in_page_bytes_left >= in_len) { | |
377 | buf = data_in + in_offset; | |
378 | bytes = in_len; | |
ca9b688c | 379 | may_late_unmap = true; |
a6fa6fae LZ |
380 | goto cont; |
381 | } | |
382 | ||
383 | /* copy bytes from the pages into the working buffer */ | |
384 | buf = workspace->cbuf; | |
385 | buf_offset = 0; | |
386 | while (working_bytes) { | |
387 | bytes = min(working_bytes, in_page_bytes_left); | |
388 | ||
389 | memcpy(buf + buf_offset, data_in + in_offset, bytes); | |
390 | buf_offset += bytes; | |
391 | cont: | |
392 | working_bytes -= bytes; | |
393 | in_page_bytes_left -= bytes; | |
394 | in_offset += bytes; | |
395 | ||
396 | /* check if we need to pick another page */ | |
397 | if ((working_bytes == 0 && in_page_bytes_left < LZO_LEN) | |
398 | || in_page_bytes_left == 0) { | |
399 | tot_in += in_page_bytes_left; | |
400 | ||
401 | if (working_bytes == 0 && tot_in >= tot_len) | |
402 | break; | |
403 | ||
ca9b688c | 404 | if (page_in_index + 1 >= total_pages_in) { |
60e1975a | 405 | ret = -EIO; |
a6fa6fae LZ |
406 | goto done; |
407 | } | |
ca9b688c LZ |
408 | |
409 | if (may_late_unmap) | |
410 | need_unmap = true; | |
411 | else | |
412 | kunmap(pages_in[page_in_index]); | |
413 | ||
414 | data_in = kmap(pages_in[++page_in_index]); | |
a6fa6fae | 415 | |
09cbfeaf | 416 | in_page_bytes_left = PAGE_SIZE; |
a6fa6fae LZ |
417 | in_offset = 0; |
418 | } | |
419 | } | |
420 | ||
314bfa47 | 421 | out_len = max_segment_len; |
a6fa6fae LZ |
422 | ret = lzo1x_decompress_safe(buf, in_len, workspace->buf, |
423 | &out_len); | |
ca9b688c LZ |
424 | if (need_unmap) |
425 | kunmap(pages_in[page_in_index - 1]); | |
a6fa6fae | 426 | if (ret != LZO_E_OK) { |
62e85577 | 427 | pr_warn("BTRFS: decompress failed\n"); |
60e1975a | 428 | ret = -EIO; |
a6fa6fae LZ |
429 | break; |
430 | } | |
431 | ||
a6fa6fae | 432 | buf_start = tot_out; |
a6fa6fae LZ |
433 | tot_out += out_len; |
434 | ||
3a39c18d | 435 | ret2 = btrfs_decompress_buf2page(workspace->buf, buf_start, |
974b1adc | 436 | tot_out, disk_start, orig_bio); |
3a39c18d | 437 | if (ret2 == 0) |
a6fa6fae | 438 | break; |
a6fa6fae LZ |
439 | } |
440 | done: | |
ca9b688c | 441 | kunmap(pages_in[page_in_index]); |
2f19cad9 | 442 | if (!ret) |
974b1adc | 443 | zero_fill_bio(orig_bio); |
a6fa6fae LZ |
444 | return ret; |
445 | } | |
446 | ||
447 | static int lzo_decompress(struct list_head *ws, unsigned char *data_in, | |
448 | struct page *dest_page, | |
449 | unsigned long start_byte, | |
450 | size_t srclen, size_t destlen) | |
451 | { | |
452 | struct workspace *workspace = list_entry(ws, struct workspace, list); | |
453 | size_t in_len; | |
454 | size_t out_len; | |
de885e3e | 455 | size_t max_segment_len = lzo1x_worst_compress(PAGE_SIZE); |
a6fa6fae LZ |
456 | int ret = 0; |
457 | char *kaddr; | |
458 | unsigned long bytes; | |
459 | ||
de885e3e QW |
460 | if (srclen < LZO_LEN || srclen > max_segment_len + LZO_LEN * 2) |
461 | return -EUCLEAN; | |
a6fa6fae | 462 | |
de885e3e QW |
463 | in_len = read_compress_length(data_in); |
464 | if (in_len != srclen) | |
465 | return -EUCLEAN; | |
a6fa6fae LZ |
466 | data_in += LZO_LEN; |
467 | ||
468 | in_len = read_compress_length(data_in); | |
de885e3e QW |
469 | if (in_len != srclen - LZO_LEN * 2) { |
470 | ret = -EUCLEAN; | |
471 | goto out; | |
472 | } | |
a6fa6fae LZ |
473 | data_in += LZO_LEN; |
474 | ||
09cbfeaf | 475 | out_len = PAGE_SIZE; |
a6fa6fae LZ |
476 | ret = lzo1x_decompress_safe(data_in, in_len, workspace->buf, &out_len); |
477 | if (ret != LZO_E_OK) { | |
62e85577 | 478 | pr_warn("BTRFS: decompress failed!\n"); |
60e1975a | 479 | ret = -EIO; |
a6fa6fae LZ |
480 | goto out; |
481 | } | |
482 | ||
483 | if (out_len < start_byte) { | |
60e1975a | 484 | ret = -EIO; |
a6fa6fae LZ |
485 | goto out; |
486 | } | |
487 | ||
2f19cad9 CM |
488 | /* |
489 | * the caller is already checking against PAGE_SIZE, but lets | |
490 | * move this check closer to the memcpy/memset | |
491 | */ | |
492 | destlen = min_t(unsigned long, destlen, PAGE_SIZE); | |
a6fa6fae LZ |
493 | bytes = min_t(unsigned long, destlen, out_len - start_byte); |
494 | ||
7ac687d9 | 495 | kaddr = kmap_atomic(dest_page); |
a6fa6fae | 496 | memcpy(kaddr, workspace->buf + start_byte, bytes); |
2f19cad9 CM |
497 | |
498 | /* | |
499 | * btrfs_getblock is doing a zero on the tail of the page too, | |
500 | * but this will cover anything missing from the decompressed | |
501 | * data. | |
502 | */ | |
503 | if (bytes < destlen) | |
504 | memset(kaddr+bytes, 0, destlen-bytes); | |
7ac687d9 | 505 | kunmap_atomic(kaddr); |
a6fa6fae LZ |
506 | out: |
507 | return ret; | |
508 | } | |
509 | ||
f51d2b59 DS |
510 | static void lzo_set_level(struct list_head *ws, unsigned int type) |
511 | { | |
512 | } | |
513 | ||
e8c9f186 | 514 | const struct btrfs_compress_op btrfs_lzo_compress = { |
92ee5530 DZ |
515 | .init_workspace_manager = lzo_init_workspace_manager, |
516 | .cleanup_workspace_manager = lzo_cleanup_workspace_manager, | |
517 | .get_workspace = lzo_get_workspace, | |
518 | .put_workspace = lzo_put_workspace, | |
a6fa6fae LZ |
519 | .alloc_workspace = lzo_alloc_workspace, |
520 | .free_workspace = lzo_free_workspace, | |
521 | .compress_pages = lzo_compress_pages, | |
974b1adc | 522 | .decompress_bio = lzo_decompress_bio, |
a6fa6fae | 523 | .decompress = lzo_decompress, |
f51d2b59 | 524 | .set_level = lzo_set_level, |
a6fa6fae | 525 | }; |