xref: /linux/fs/btrfs/lzo.c (revision 87c9c16317882dd6dbbc07e349bc3223e14f3244)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2008 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/kernel.h>
7 #include <linux/slab.h>
8 #include <linux/mm.h>
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>
15 #include <linux/refcount.h>
16 #include "compression.h"
17 
18 #define LZO_LEN	4
19 
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, followed 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 
57 struct workspace {
58 	void *mem;
59 	void *buf;	/* where decompressed data goes */
60 	void *cbuf;	/* where compressed data goes */
61 	struct list_head list;
62 };
63 
64 static struct workspace_manager wsm;
65 
66 void lzo_free_workspace(struct list_head *ws)
67 {
68 	struct workspace *workspace = list_entry(ws, struct workspace, list);
69 
70 	kvfree(workspace->buf);
71 	kvfree(workspace->cbuf);
72 	kvfree(workspace->mem);
73 	kfree(workspace);
74 }
75 
76 struct list_head *lzo_alloc_workspace(unsigned int level)
77 {
78 	struct workspace *workspace;
79 
80 	workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
81 	if (!workspace)
82 		return ERR_PTR(-ENOMEM);
83 
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);
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 
106 static inline size_t read_compress_length(const char *buf)
107 {
108 	__le32 dlen;
109 
110 	memcpy(&dlen, buf, LZO_LEN);
111 	return le32_to_cpu(dlen);
112 }
113 
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)
117 {
118 	struct workspace *workspace = list_entry(ws, struct workspace, list);
119 	int ret = 0;
120 	char *data_in;
121 	char *cpage_out, *sizes_ptr;
122 	int nr_pages = 0;
123 	struct page *in_page = NULL;
124 	struct page *out_page = NULL;
125 	unsigned long bytes_left;
126 	unsigned long len = *total_out;
127 	unsigned long nr_dest_pages = *out_pages;
128 	const unsigned long max_out = nr_dest_pages * PAGE_SIZE;
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 
142 	in_page = find_get_page(mapping, start >> PAGE_SHIFT);
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;
159 	pg_bytes_left = PAGE_SIZE - LZO_LEN;
160 
161 	/* compress at most one page of data each time */
162 	in_len = min(len, PAGE_SIZE);
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) {
167 			pr_debug("BTRFS: lzo in loop returned %d\n",
168 			       ret);
169 			ret = -EIO;
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;
215 					ret = -E2BIG;
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 
227 				pg_bytes_left = PAGE_SIZE;
228 				out_offset = 0;
229 			}
230 		}
231 
232 		/* we're making it bigger, give up */
233 		if (tot_in > 8192 && tot_in < tot_out) {
234 			ret = -E2BIG;
235 			goto out;
236 		}
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);
247 		put_page(in_page);
248 
249 		start += PAGE_SIZE;
250 		in_page = find_get_page(mapping, start >> PAGE_SHIFT);
251 		data_in = kmap(in_page);
252 		in_len = min(bytes_left, PAGE_SIZE);
253 	}
254 
255 	if (tot_out >= tot_in) {
256 		ret = -E2BIG;
257 		goto out;
258 	}
259 
260 	/* store the size of all chunks of compressed data */
261 	sizes_ptr = kmap_local_page(pages[0]);
262 	write_compress_length(sizes_ptr, tot_out);
263 	kunmap_local(sizes_ptr);
264 
265 	ret = 0;
266 	*total_out = tot_out;
267 	*total_in = tot_in;
268 out:
269 	*out_pages = nr_pages;
270 	if (out_page)
271 		kunmap(out_page);
272 
273 	if (in_page) {
274 		kunmap(in_page);
275 		put_page(in_page);
276 	}
277 
278 	return ret;
279 }
280 
281 int lzo_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
282 {
283 	struct workspace *workspace = list_entry(ws, struct workspace, list);
284 	int ret = 0, ret2;
285 	char *data_in;
286 	unsigned long page_in_index = 0;
287 	size_t srclen = cb->compressed_len;
288 	unsigned long total_pages_in = DIV_ROUND_UP(srclen, PAGE_SIZE);
289 	unsigned long buf_start;
290 	unsigned long buf_offset = 0;
291 	unsigned long bytes;
292 	unsigned long working_bytes;
293 	size_t in_len;
294 	size_t out_len;
295 	const size_t max_segment_len = lzo1x_worst_compress(PAGE_SIZE);
296 	unsigned long in_offset;
297 	unsigned long in_page_bytes_left;
298 	unsigned long tot_in;
299 	unsigned long tot_out;
300 	unsigned long tot_len;
301 	char *buf;
302 	bool may_late_unmap, need_unmap;
303 	struct page **pages_in = cb->compressed_pages;
304 	u64 disk_start = cb->start;
305 	struct bio *orig_bio = cb->orig_bio;
306 
307 	data_in = kmap(pages_in[0]);
308 	tot_len = read_compress_length(data_in);
309 	/*
310 	 * Compressed data header check.
311 	 *
312 	 * The real compressed size can't exceed the maximum extent length, and
313 	 * all pages should be used (whole unused page with just the segment
314 	 * header is not possible).  If this happens it means the compressed
315 	 * extent is corrupted.
316 	 */
317 	if (tot_len > min_t(size_t, BTRFS_MAX_COMPRESSED, srclen) ||
318 	    tot_len < srclen - PAGE_SIZE) {
319 		ret = -EUCLEAN;
320 		goto done;
321 	}
322 
323 	tot_in = LZO_LEN;
324 	in_offset = LZO_LEN;
325 	in_page_bytes_left = PAGE_SIZE - LZO_LEN;
326 
327 	tot_out = 0;
328 
329 	while (tot_in < tot_len) {
330 		in_len = read_compress_length(data_in + in_offset);
331 		in_page_bytes_left -= LZO_LEN;
332 		in_offset += LZO_LEN;
333 		tot_in += LZO_LEN;
334 
335 		/*
336 		 * Segment header check.
337 		 *
338 		 * The segment length must not exceed the maximum LZO
339 		 * compression size, nor the total compressed size.
340 		 */
341 		if (in_len > max_segment_len || tot_in + in_len > tot_len) {
342 			ret = -EUCLEAN;
343 			goto done;
344 		}
345 
346 		tot_in += in_len;
347 		working_bytes = in_len;
348 		may_late_unmap = need_unmap = false;
349 
350 		/* fast path: avoid using the working buffer */
351 		if (in_page_bytes_left >= in_len) {
352 			buf = data_in + in_offset;
353 			bytes = in_len;
354 			may_late_unmap = true;
355 			goto cont;
356 		}
357 
358 		/* copy bytes from the pages into the working buffer */
359 		buf = workspace->cbuf;
360 		buf_offset = 0;
361 		while (working_bytes) {
362 			bytes = min(working_bytes, in_page_bytes_left);
363 
364 			memcpy(buf + buf_offset, data_in + in_offset, bytes);
365 			buf_offset += bytes;
366 cont:
367 			working_bytes -= bytes;
368 			in_page_bytes_left -= bytes;
369 			in_offset += bytes;
370 
371 			/* check if we need to pick another page */
372 			if ((working_bytes == 0 && in_page_bytes_left < LZO_LEN)
373 			    || in_page_bytes_left == 0) {
374 				tot_in += in_page_bytes_left;
375 
376 				if (working_bytes == 0 && tot_in >= tot_len)
377 					break;
378 
379 				if (page_in_index + 1 >= total_pages_in) {
380 					ret = -EIO;
381 					goto done;
382 				}
383 
384 				if (may_late_unmap)
385 					need_unmap = true;
386 				else
387 					kunmap(pages_in[page_in_index]);
388 
389 				data_in = kmap(pages_in[++page_in_index]);
390 
391 				in_page_bytes_left = PAGE_SIZE;
392 				in_offset = 0;
393 			}
394 		}
395 
396 		out_len = max_segment_len;
397 		ret = lzo1x_decompress_safe(buf, in_len, workspace->buf,
398 					    &out_len);
399 		if (need_unmap)
400 			kunmap(pages_in[page_in_index - 1]);
401 		if (ret != LZO_E_OK) {
402 			pr_warn("BTRFS: decompress failed\n");
403 			ret = -EIO;
404 			break;
405 		}
406 
407 		buf_start = tot_out;
408 		tot_out += out_len;
409 
410 		ret2 = btrfs_decompress_buf2page(workspace->buf, buf_start,
411 						 tot_out, disk_start, orig_bio);
412 		if (ret2 == 0)
413 			break;
414 	}
415 done:
416 	kunmap(pages_in[page_in_index]);
417 	if (!ret)
418 		zero_fill_bio(orig_bio);
419 	return ret;
420 }
421 
422 int lzo_decompress(struct list_head *ws, unsigned char *data_in,
423 		struct page *dest_page, unsigned long start_byte, size_t srclen,
424 		size_t destlen)
425 {
426 	struct workspace *workspace = list_entry(ws, struct workspace, list);
427 	size_t in_len;
428 	size_t out_len;
429 	size_t max_segment_len = lzo1x_worst_compress(PAGE_SIZE);
430 	int ret = 0;
431 	char *kaddr;
432 	unsigned long bytes;
433 
434 	if (srclen < LZO_LEN || srclen > max_segment_len + LZO_LEN * 2)
435 		return -EUCLEAN;
436 
437 	in_len = read_compress_length(data_in);
438 	if (in_len != srclen)
439 		return -EUCLEAN;
440 	data_in += LZO_LEN;
441 
442 	in_len = read_compress_length(data_in);
443 	if (in_len != srclen - LZO_LEN * 2) {
444 		ret = -EUCLEAN;
445 		goto out;
446 	}
447 	data_in += LZO_LEN;
448 
449 	out_len = PAGE_SIZE;
450 	ret = lzo1x_decompress_safe(data_in, in_len, workspace->buf, &out_len);
451 	if (ret != LZO_E_OK) {
452 		pr_warn("BTRFS: decompress failed!\n");
453 		ret = -EIO;
454 		goto out;
455 	}
456 
457 	if (out_len < start_byte) {
458 		ret = -EIO;
459 		goto out;
460 	}
461 
462 	/*
463 	 * the caller is already checking against PAGE_SIZE, but lets
464 	 * move this check closer to the memcpy/memset
465 	 */
466 	destlen = min_t(unsigned long, destlen, PAGE_SIZE);
467 	bytes = min_t(unsigned long, destlen, out_len - start_byte);
468 
469 	kaddr = kmap_local_page(dest_page);
470 	memcpy(kaddr, workspace->buf + start_byte, bytes);
471 
472 	/*
473 	 * btrfs_getblock is doing a zero on the tail of the page too,
474 	 * but this will cover anything missing from the decompressed
475 	 * data.
476 	 */
477 	if (bytes < destlen)
478 		memset(kaddr+bytes, 0, destlen-bytes);
479 	kunmap_local(kaddr);
480 out:
481 	return ret;
482 }
483 
484 const struct btrfs_compress_op btrfs_lzo_compress = {
485 	.workspace_manager	= &wsm,
486 	.max_level		= 1,
487 	.default_level		= 1,
488 };
489