xref: /linux/fs/btrfs/lzo.c (revision db10cb9b)
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 "messages.h"
17 #include "compression.h"
18 #include "ctree.h"
19 #include "super.h"
20 #include "btrfs_inode.h"
21 
22 #define LZO_LEN	4
23 
24 /*
25  * Btrfs LZO compression format
26  *
27  * Regular and inlined LZO compressed data extents consist of:
28  *
29  * 1.  Header
30  *     Fixed size. LZO_LEN (4) bytes long, LE32.
31  *     Records the total size (including the header) of compressed data.
32  *
33  * 2.  Segment(s)
34  *     Variable size. Each segment includes one segment header, followed by data
35  *     payload.
36  *     One regular LZO compressed extent can have one or more segments.
37  *     For inlined LZO compressed extent, only one segment is allowed.
38  *     One segment represents at most one sector of uncompressed data.
39  *
40  * 2.1 Segment header
41  *     Fixed size. LZO_LEN (4) bytes long, LE32.
42  *     Records the total size of the segment (not including the header).
43  *     Segment header never crosses sector boundary, thus it's possible to
44  *     have at most 3 padding zeros at the end of the sector.
45  *
46  * 2.2 Data Payload
47  *     Variable size. Size up limit should be lzo1x_worst_compress(sectorsize)
48  *     which is 4419 for a 4KiB sectorsize.
49  *
50  * Example with 4K sectorsize:
51  * Page 1:
52  *          0     0x2   0x4   0x6   0x8   0xa   0xc   0xe     0x10
53  * 0x0000   |  Header   | SegHdr 01 | Data payload 01 ...     |
54  * ...
55  * 0x0ff0   | SegHdr  N | Data payload  N     ...          |00|
56  *                                                          ^^ padding zeros
57  * Page 2:
58  * 0x1000   | SegHdr N+1| Data payload N+1 ...                |
59  */
60 
61 #define WORKSPACE_BUF_LENGTH	(lzo1x_worst_compress(PAGE_SIZE))
62 #define WORKSPACE_CBUF_LENGTH	(lzo1x_worst_compress(PAGE_SIZE))
63 
64 struct workspace {
65 	void *mem;
66 	void *buf;	/* where decompressed data goes */
67 	void *cbuf;	/* where compressed data goes */
68 	struct list_head list;
69 };
70 
71 static struct workspace_manager wsm;
72 
73 void lzo_free_workspace(struct list_head *ws)
74 {
75 	struct workspace *workspace = list_entry(ws, struct workspace, list);
76 
77 	kvfree(workspace->buf);
78 	kvfree(workspace->cbuf);
79 	kvfree(workspace->mem);
80 	kfree(workspace);
81 }
82 
83 struct list_head *lzo_alloc_workspace(unsigned int level)
84 {
85 	struct workspace *workspace;
86 
87 	workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
88 	if (!workspace)
89 		return ERR_PTR(-ENOMEM);
90 
91 	workspace->mem = kvmalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL | __GFP_NOWARN);
92 	workspace->buf = kvmalloc(WORKSPACE_BUF_LENGTH, GFP_KERNEL | __GFP_NOWARN);
93 	workspace->cbuf = kvmalloc(WORKSPACE_CBUF_LENGTH, GFP_KERNEL | __GFP_NOWARN);
94 	if (!workspace->mem || !workspace->buf || !workspace->cbuf)
95 		goto fail;
96 
97 	INIT_LIST_HEAD(&workspace->list);
98 
99 	return &workspace->list;
100 fail:
101 	lzo_free_workspace(&workspace->list);
102 	return ERR_PTR(-ENOMEM);
103 }
104 
105 static inline void write_compress_length(char *buf, size_t len)
106 {
107 	__le32 dlen;
108 
109 	dlen = cpu_to_le32(len);
110 	memcpy(buf, &dlen, LZO_LEN);
111 }
112 
113 static inline size_t read_compress_length(const char *buf)
114 {
115 	__le32 dlen;
116 
117 	memcpy(&dlen, buf, LZO_LEN);
118 	return le32_to_cpu(dlen);
119 }
120 
121 /*
122  * Will do:
123  *
124  * - Write a segment header into the destination
125  * - Copy the compressed buffer into the destination
126  * - Make sure we have enough space in the last sector to fit a segment header
127  *   If not, we will pad at most (LZO_LEN (4)) - 1 bytes of zeros.
128  *
129  * Will allocate new pages when needed.
130  */
131 static int copy_compressed_data_to_page(char *compressed_data,
132 					size_t compressed_size,
133 					struct page **out_pages,
134 					unsigned long max_nr_page,
135 					u32 *cur_out,
136 					const u32 sectorsize)
137 {
138 	u32 sector_bytes_left;
139 	u32 orig_out;
140 	struct page *cur_page;
141 	char *kaddr;
142 
143 	if ((*cur_out / PAGE_SIZE) >= max_nr_page)
144 		return -E2BIG;
145 
146 	/*
147 	 * We never allow a segment header crossing sector boundary, previous
148 	 * run should ensure we have enough space left inside the sector.
149 	 */
150 	ASSERT((*cur_out / sectorsize) == (*cur_out + LZO_LEN - 1) / sectorsize);
151 
152 	cur_page = out_pages[*cur_out / PAGE_SIZE];
153 	/* Allocate a new page */
154 	if (!cur_page) {
155 		cur_page = alloc_page(GFP_NOFS);
156 		if (!cur_page)
157 			return -ENOMEM;
158 		out_pages[*cur_out / PAGE_SIZE] = cur_page;
159 	}
160 
161 	kaddr = kmap_local_page(cur_page);
162 	write_compress_length(kaddr + offset_in_page(*cur_out),
163 			      compressed_size);
164 	*cur_out += LZO_LEN;
165 
166 	orig_out = *cur_out;
167 
168 	/* Copy compressed data */
169 	while (*cur_out - orig_out < compressed_size) {
170 		u32 copy_len = min_t(u32, sectorsize - *cur_out % sectorsize,
171 				     orig_out + compressed_size - *cur_out);
172 
173 		kunmap_local(kaddr);
174 
175 		if ((*cur_out / PAGE_SIZE) >= max_nr_page)
176 			return -E2BIG;
177 
178 		cur_page = out_pages[*cur_out / PAGE_SIZE];
179 		/* Allocate a new page */
180 		if (!cur_page) {
181 			cur_page = alloc_page(GFP_NOFS);
182 			if (!cur_page)
183 				return -ENOMEM;
184 			out_pages[*cur_out / PAGE_SIZE] = cur_page;
185 		}
186 		kaddr = kmap_local_page(cur_page);
187 
188 		memcpy(kaddr + offset_in_page(*cur_out),
189 		       compressed_data + *cur_out - orig_out, copy_len);
190 
191 		*cur_out += copy_len;
192 	}
193 
194 	/*
195 	 * Check if we can fit the next segment header into the remaining space
196 	 * of the sector.
197 	 */
198 	sector_bytes_left = round_up(*cur_out, sectorsize) - *cur_out;
199 	if (sector_bytes_left >= LZO_LEN || sector_bytes_left == 0)
200 		goto out;
201 
202 	/* The remaining size is not enough, pad it with zeros */
203 	memset(kaddr + offset_in_page(*cur_out), 0,
204 	       sector_bytes_left);
205 	*cur_out += sector_bytes_left;
206 
207 out:
208 	kunmap_local(kaddr);
209 	return 0;
210 }
211 
212 int lzo_compress_pages(struct list_head *ws, struct address_space *mapping,
213 		u64 start, struct page **pages, unsigned long *out_pages,
214 		unsigned long *total_in, unsigned long *total_out)
215 {
216 	struct workspace *workspace = list_entry(ws, struct workspace, list);
217 	const u32 sectorsize = btrfs_sb(mapping->host->i_sb)->sectorsize;
218 	struct page *page_in = NULL;
219 	char *sizes_ptr;
220 	const unsigned long max_nr_page = *out_pages;
221 	int ret = 0;
222 	/* Points to the file offset of input data */
223 	u64 cur_in = start;
224 	/* Points to the current output byte */
225 	u32 cur_out = 0;
226 	u32 len = *total_out;
227 
228 	ASSERT(max_nr_page > 0);
229 	*out_pages = 0;
230 	*total_out = 0;
231 	*total_in = 0;
232 
233 	/*
234 	 * Skip the header for now, we will later come back and write the total
235 	 * compressed size
236 	 */
237 	cur_out += LZO_LEN;
238 	while (cur_in < start + len) {
239 		char *data_in;
240 		const u32 sectorsize_mask = sectorsize - 1;
241 		u32 sector_off = (cur_in - start) & sectorsize_mask;
242 		u32 in_len;
243 		size_t out_len;
244 
245 		/* Get the input page first */
246 		if (!page_in) {
247 			page_in = find_get_page(mapping, cur_in >> PAGE_SHIFT);
248 			ASSERT(page_in);
249 		}
250 
251 		/* Compress at most one sector of data each time */
252 		in_len = min_t(u32, start + len - cur_in, sectorsize - sector_off);
253 		ASSERT(in_len);
254 		data_in = kmap_local_page(page_in);
255 		ret = lzo1x_1_compress(data_in +
256 				       offset_in_page(cur_in), in_len,
257 				       workspace->cbuf, &out_len,
258 				       workspace->mem);
259 		kunmap_local(data_in);
260 		if (ret < 0) {
261 			pr_debug("BTRFS: lzo in loop returned %d\n", ret);
262 			ret = -EIO;
263 			goto out;
264 		}
265 
266 		ret = copy_compressed_data_to_page(workspace->cbuf, out_len,
267 						   pages, max_nr_page,
268 						   &cur_out, sectorsize);
269 		if (ret < 0)
270 			goto out;
271 
272 		cur_in += in_len;
273 
274 		/*
275 		 * Check if we're making it bigger after two sectors.  And if
276 		 * it is so, give up.
277 		 */
278 		if (cur_in - start > sectorsize * 2 && cur_in - start < cur_out) {
279 			ret = -E2BIG;
280 			goto out;
281 		}
282 
283 		/* Check if we have reached page boundary */
284 		if (PAGE_ALIGNED(cur_in)) {
285 			put_page(page_in);
286 			page_in = NULL;
287 		}
288 	}
289 
290 	/* Store the size of all chunks of compressed data */
291 	sizes_ptr = kmap_local_page(pages[0]);
292 	write_compress_length(sizes_ptr, cur_out);
293 	kunmap_local(sizes_ptr);
294 
295 	ret = 0;
296 	*total_out = cur_out;
297 	*total_in = cur_in - start;
298 out:
299 	if (page_in)
300 		put_page(page_in);
301 	*out_pages = DIV_ROUND_UP(cur_out, PAGE_SIZE);
302 	return ret;
303 }
304 
305 /*
306  * Copy the compressed segment payload into @dest.
307  *
308  * For the payload there will be no padding, just need to do page switching.
309  */
310 static void copy_compressed_segment(struct compressed_bio *cb,
311 				    char *dest, u32 len, u32 *cur_in)
312 {
313 	u32 orig_in = *cur_in;
314 
315 	while (*cur_in < orig_in + len) {
316 		struct page *cur_page;
317 		u32 copy_len = min_t(u32, PAGE_SIZE - offset_in_page(*cur_in),
318 					  orig_in + len - *cur_in);
319 
320 		ASSERT(copy_len);
321 		cur_page = cb->compressed_pages[*cur_in / PAGE_SIZE];
322 
323 		memcpy_from_page(dest + *cur_in - orig_in, cur_page,
324 				 offset_in_page(*cur_in), copy_len);
325 
326 		*cur_in += copy_len;
327 	}
328 }
329 
330 int lzo_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
331 {
332 	struct workspace *workspace = list_entry(ws, struct workspace, list);
333 	const struct btrfs_fs_info *fs_info = cb->bbio.inode->root->fs_info;
334 	const u32 sectorsize = fs_info->sectorsize;
335 	char *kaddr;
336 	int ret;
337 	/* Compressed data length, can be unaligned */
338 	u32 len_in;
339 	/* Offset inside the compressed data */
340 	u32 cur_in = 0;
341 	/* Bytes decompressed so far */
342 	u32 cur_out = 0;
343 
344 	kaddr = kmap_local_page(cb->compressed_pages[0]);
345 	len_in = read_compress_length(kaddr);
346 	kunmap_local(kaddr);
347 	cur_in += LZO_LEN;
348 
349 	/*
350 	 * LZO header length check
351 	 *
352 	 * The total length should not exceed the maximum extent length,
353 	 * and all sectors should be used.
354 	 * If this happens, it means the compressed extent is corrupted.
355 	 */
356 	if (len_in > min_t(size_t, BTRFS_MAX_COMPRESSED, cb->compressed_len) ||
357 	    round_up(len_in, sectorsize) < cb->compressed_len) {
358 		btrfs_err(fs_info,
359 			"invalid lzo header, lzo len %u compressed len %u",
360 			len_in, cb->compressed_len);
361 		return -EUCLEAN;
362 	}
363 
364 	/* Go through each lzo segment */
365 	while (cur_in < len_in) {
366 		struct page *cur_page;
367 		/* Length of the compressed segment */
368 		u32 seg_len;
369 		u32 sector_bytes_left;
370 		size_t out_len = lzo1x_worst_compress(sectorsize);
371 
372 		/*
373 		 * We should always have enough space for one segment header
374 		 * inside current sector.
375 		 */
376 		ASSERT(cur_in / sectorsize ==
377 		       (cur_in + LZO_LEN - 1) / sectorsize);
378 		cur_page = cb->compressed_pages[cur_in / PAGE_SIZE];
379 		ASSERT(cur_page);
380 		kaddr = kmap_local_page(cur_page);
381 		seg_len = read_compress_length(kaddr + offset_in_page(cur_in));
382 		kunmap_local(kaddr);
383 		cur_in += LZO_LEN;
384 
385 		if (seg_len > WORKSPACE_CBUF_LENGTH) {
386 			/*
387 			 * seg_len shouldn't be larger than we have allocated
388 			 * for workspace->cbuf
389 			 */
390 			btrfs_err(fs_info, "unexpectedly large lzo segment len %u",
391 					seg_len);
392 			return -EIO;
393 		}
394 
395 		/* Copy the compressed segment payload into workspace */
396 		copy_compressed_segment(cb, workspace->cbuf, seg_len, &cur_in);
397 
398 		/* Decompress the data */
399 		ret = lzo1x_decompress_safe(workspace->cbuf, seg_len,
400 					    workspace->buf, &out_len);
401 		if (ret != LZO_E_OK) {
402 			btrfs_err(fs_info, "failed to decompress");
403 			return -EIO;
404 		}
405 
406 		/* Copy the data into inode pages */
407 		ret = btrfs_decompress_buf2page(workspace->buf, out_len, cb, cur_out);
408 		cur_out += out_len;
409 
410 		/* All data read, exit */
411 		if (ret == 0)
412 			return 0;
413 		ret = 0;
414 
415 		/* Check if the sector has enough space for a segment header */
416 		sector_bytes_left = sectorsize - (cur_in % sectorsize);
417 		if (sector_bytes_left >= LZO_LEN)
418 			continue;
419 
420 		/* Skip the padding zeros */
421 		cur_in += sector_bytes_left;
422 	}
423 
424 	return 0;
425 }
426 
427 int lzo_decompress(struct list_head *ws, const u8 *data_in,
428 		struct page *dest_page, unsigned long start_byte, size_t srclen,
429 		size_t destlen)
430 {
431 	struct workspace *workspace = list_entry(ws, struct workspace, list);
432 	size_t in_len;
433 	size_t out_len;
434 	size_t max_segment_len = WORKSPACE_BUF_LENGTH;
435 	int ret = 0;
436 	char *kaddr;
437 	unsigned long bytes;
438 
439 	if (srclen < LZO_LEN || srclen > max_segment_len + LZO_LEN * 2)
440 		return -EUCLEAN;
441 
442 	in_len = read_compress_length(data_in);
443 	if (in_len != srclen)
444 		return -EUCLEAN;
445 	data_in += LZO_LEN;
446 
447 	in_len = read_compress_length(data_in);
448 	if (in_len != srclen - LZO_LEN * 2) {
449 		ret = -EUCLEAN;
450 		goto out;
451 	}
452 	data_in += LZO_LEN;
453 
454 	out_len = PAGE_SIZE;
455 	ret = lzo1x_decompress_safe(data_in, in_len, workspace->buf, &out_len);
456 	if (ret != LZO_E_OK) {
457 		pr_warn("BTRFS: decompress failed!\n");
458 		ret = -EIO;
459 		goto out;
460 	}
461 
462 	if (out_len < start_byte) {
463 		ret = -EIO;
464 		goto out;
465 	}
466 
467 	/*
468 	 * the caller is already checking against PAGE_SIZE, but lets
469 	 * move this check closer to the memcpy/memset
470 	 */
471 	destlen = min_t(unsigned long, destlen, PAGE_SIZE);
472 	bytes = min_t(unsigned long, destlen, out_len - start_byte);
473 
474 	kaddr = kmap_local_page(dest_page);
475 	memcpy(kaddr, workspace->buf + start_byte, bytes);
476 
477 	/*
478 	 * btrfs_getblock is doing a zero on the tail of the page too,
479 	 * but this will cover anything missing from the decompressed
480 	 * data.
481 	 */
482 	if (bytes < destlen)
483 		memset(kaddr+bytes, 0, destlen-bytes);
484 	kunmap_local(kaddr);
485 out:
486 	return ret;
487 }
488 
489 const struct btrfs_compress_op btrfs_lzo_compress = {
490 	.workspace_manager	= &wsm,
491 	.max_level		= 1,
492 	.default_level		= 1,
493 };
494