xref: /freebsd/contrib/xz/src/liblzma/common/block_buffer_encoder.c (revision 8ddb146abcdf061be9f2c0db7e391697dafad85c)
1 ///////////////////////////////////////////////////////////////////////////////
2 //
3 /// \file       block_buffer_encoder.c
4 /// \brief      Single-call .xz Block encoder
5 //
6 //  Author:     Lasse Collin
7 //
8 //  This file has been put into the public domain.
9 //  You can do whatever you want with this file.
10 //
11 ///////////////////////////////////////////////////////////////////////////////
12 
13 #include "block_buffer_encoder.h"
14 #include "block_encoder.h"
15 #include "filter_encoder.h"
16 #include "lzma2_encoder.h"
17 #include "check.h"
18 
19 
20 /// Estimate the maximum size of the Block Header and Check fields for
21 /// a Block that uses LZMA2 uncompressed chunks. We could use
22 /// lzma_block_header_size() but this is simpler.
23 ///
24 /// Block Header Size + Block Flags + Compressed Size
25 /// + Uncompressed Size + Filter Flags for LZMA2 + CRC32 + Check
26 /// and round up to the next multiple of four to take Header Padding
27 /// into account.
28 #define HEADERS_BOUND ((1 + 1 + 2 * LZMA_VLI_BYTES_MAX + 3 + 4 \
29 		+ LZMA_CHECK_SIZE_MAX + 3) & ~3)
30 
31 
32 static uint64_t
33 lzma2_bound(uint64_t uncompressed_size)
34 {
35 	// Prevent integer overflow in overhead calculation.
36 	if (uncompressed_size > COMPRESSED_SIZE_MAX)
37 		return 0;
38 
39 	// Calculate the exact overhead of the LZMA2 headers: Round
40 	// uncompressed_size up to the next multiple of LZMA2_CHUNK_MAX,
41 	// multiply by the size of per-chunk header, and add one byte for
42 	// the end marker.
43 	const uint64_t overhead = ((uncompressed_size + LZMA2_CHUNK_MAX - 1)
44 				/ LZMA2_CHUNK_MAX)
45 			* LZMA2_HEADER_UNCOMPRESSED + 1;
46 
47 	// Catch the possible integer overflow.
48 	if (COMPRESSED_SIZE_MAX - overhead < uncompressed_size)
49 		return 0;
50 
51 	return uncompressed_size + overhead;
52 }
53 
54 
55 extern uint64_t
56 lzma_block_buffer_bound64(uint64_t uncompressed_size)
57 {
58 	// If the data doesn't compress, we always use uncompressed
59 	// LZMA2 chunks.
60 	uint64_t lzma2_size = lzma2_bound(uncompressed_size);
61 	if (lzma2_size == 0)
62 		return 0;
63 
64 	// Take Block Padding into account.
65 	lzma2_size = (lzma2_size + 3) & ~UINT64_C(3);
66 
67 	// No risk of integer overflow because lzma2_bound() already takes
68 	// into account the size of the headers in the Block.
69 	return HEADERS_BOUND + lzma2_size;
70 }
71 
72 
73 extern LZMA_API(size_t)
74 lzma_block_buffer_bound(size_t uncompressed_size)
75 {
76 	uint64_t ret = lzma_block_buffer_bound64(uncompressed_size);
77 
78 #if SIZE_MAX < UINT64_MAX
79 	// Catch the possible integer overflow on 32-bit systems.
80 	if (ret > SIZE_MAX)
81 		return 0;
82 #endif
83 
84 	return ret;
85 }
86 
87 
88 static lzma_ret
89 block_encode_uncompressed(lzma_block *block, const uint8_t *in, size_t in_size,
90 		uint8_t *out, size_t *out_pos, size_t out_size)
91 {
92 	// Use LZMA2 uncompressed chunks. We wouldn't need a dictionary at
93 	// all, but LZMA2 always requires a dictionary, so use the minimum
94 	// value to minimize memory usage of the decoder.
95 	lzma_options_lzma lzma2 = {
96 		.dict_size = LZMA_DICT_SIZE_MIN,
97 	};
98 
99 	lzma_filter filters[2];
100 	filters[0].id = LZMA_FILTER_LZMA2;
101 	filters[0].options = &lzma2;
102 	filters[1].id = LZMA_VLI_UNKNOWN;
103 
104 	// Set the above filter options to *block temporarily so that we can
105 	// encode the Block Header.
106 	lzma_filter *filters_orig = block->filters;
107 	block->filters = filters;
108 
109 	if (lzma_block_header_size(block) != LZMA_OK) {
110 		block->filters = filters_orig;
111 		return LZMA_PROG_ERROR;
112 	}
113 
114 	// Check that there's enough output space. The caller has already
115 	// set block->compressed_size to what lzma2_bound() has returned,
116 	// so we can reuse that value. We know that compressed_size is a
117 	// known valid VLI and header_size is a small value so their sum
118 	// will never overflow.
119 	assert(block->compressed_size == lzma2_bound(in_size));
120 	if (out_size - *out_pos
121 			< block->header_size + block->compressed_size) {
122 		block->filters = filters_orig;
123 		return LZMA_BUF_ERROR;
124 	}
125 
126 	if (lzma_block_header_encode(block, out + *out_pos) != LZMA_OK) {
127 		block->filters = filters_orig;
128 		return LZMA_PROG_ERROR;
129 	}
130 
131 	block->filters = filters_orig;
132 	*out_pos += block->header_size;
133 
134 	// Encode the data using LZMA2 uncompressed chunks.
135 	size_t in_pos = 0;
136 	uint8_t control = 0x01; // Dictionary reset
137 
138 	while (in_pos < in_size) {
139 		// Control byte: Indicate uncompressed chunk, of which
140 		// the first resets the dictionary.
141 		out[(*out_pos)++] = control;
142 		control = 0x02; // No dictionary reset
143 
144 		// Size of the uncompressed chunk
145 		const size_t copy_size
146 				= my_min(in_size - in_pos, LZMA2_CHUNK_MAX);
147 		out[(*out_pos)++] = (copy_size - 1) >> 8;
148 		out[(*out_pos)++] = (copy_size - 1) & 0xFF;
149 
150 		// The actual data
151 		assert(*out_pos + copy_size <= out_size);
152 		memcpy(out + *out_pos, in + in_pos, copy_size);
153 
154 		in_pos += copy_size;
155 		*out_pos += copy_size;
156 	}
157 
158 	// End marker
159 	out[(*out_pos)++] = 0x00;
160 	assert(*out_pos <= out_size);
161 
162 	return LZMA_OK;
163 }
164 
165 
166 static lzma_ret
167 block_encode_normal(lzma_block *block, const lzma_allocator *allocator,
168 		const uint8_t *in, size_t in_size,
169 		uint8_t *out, size_t *out_pos, size_t out_size)
170 {
171 	// Find out the size of the Block Header.
172 	return_if_error(lzma_block_header_size(block));
173 
174 	// Reserve space for the Block Header and skip it for now.
175 	if (out_size - *out_pos <= block->header_size)
176 		return LZMA_BUF_ERROR;
177 
178 	const size_t out_start = *out_pos;
179 	*out_pos += block->header_size;
180 
181 	// Limit out_size so that we stop encoding if the output would grow
182 	// bigger than what uncompressed Block would be.
183 	if (out_size - *out_pos > block->compressed_size)
184 		out_size = *out_pos + block->compressed_size;
185 
186 	// TODO: In many common cases this could be optimized to use
187 	// significantly less memory.
188 	lzma_next_coder raw_encoder = LZMA_NEXT_CODER_INIT;
189 	lzma_ret ret = lzma_raw_encoder_init(
190 			&raw_encoder, allocator, block->filters);
191 
192 	if (ret == LZMA_OK) {
193 		size_t in_pos = 0;
194 		ret = raw_encoder.code(raw_encoder.coder, allocator,
195 				in, &in_pos, in_size, out, out_pos, out_size,
196 				LZMA_FINISH);
197 	}
198 
199 	// NOTE: This needs to be run even if lzma_raw_encoder_init() failed.
200 	lzma_next_end(&raw_encoder, allocator);
201 
202 	if (ret == LZMA_STREAM_END) {
203 		// Compression was successful. Write the Block Header.
204 		block->compressed_size
205 				= *out_pos - (out_start + block->header_size);
206 		ret = lzma_block_header_encode(block, out + out_start);
207 		if (ret != LZMA_OK)
208 			ret = LZMA_PROG_ERROR;
209 
210 	} else if (ret == LZMA_OK) {
211 		// Output buffer became full.
212 		ret = LZMA_BUF_ERROR;
213 	}
214 
215 	// Reset *out_pos if something went wrong.
216 	if (ret != LZMA_OK)
217 		*out_pos = out_start;
218 
219 	return ret;
220 }
221 
222 
223 static lzma_ret
224 block_buffer_encode(lzma_block *block, const lzma_allocator *allocator,
225 		const uint8_t *in, size_t in_size,
226 		uint8_t *out, size_t *out_pos, size_t out_size,
227 		bool try_to_compress)
228 {
229 	// Validate the arguments.
230 	if (block == NULL || (in == NULL && in_size != 0) || out == NULL
231 			|| out_pos == NULL || *out_pos > out_size)
232 		return LZMA_PROG_ERROR;
233 
234 	// The contents of the structure may depend on the version so
235 	// check the version before validating the contents of *block.
236 	if (block->version > 1)
237 		return LZMA_OPTIONS_ERROR;
238 
239 	if ((unsigned int)(block->check) > LZMA_CHECK_ID_MAX
240 			|| (try_to_compress && block->filters == NULL))
241 		return LZMA_PROG_ERROR;
242 
243 	if (!lzma_check_is_supported(block->check))
244 		return LZMA_UNSUPPORTED_CHECK;
245 
246 	// Size of a Block has to be a multiple of four, so limit the size
247 	// here already. This way we don't need to check it again when adding
248 	// Block Padding.
249 	out_size -= (out_size - *out_pos) & 3;
250 
251 	// Get the size of the Check field.
252 	const size_t check_size = lzma_check_size(block->check);
253 	assert(check_size != UINT32_MAX);
254 
255 	// Reserve space for the Check field.
256 	if (out_size - *out_pos <= check_size)
257 		return LZMA_BUF_ERROR;
258 
259 	out_size -= check_size;
260 
261 	// Initialize block->uncompressed_size and calculate the worst-case
262 	// value for block->compressed_size.
263 	block->uncompressed_size = in_size;
264 	block->compressed_size = lzma2_bound(in_size);
265 	if (block->compressed_size == 0)
266 		return LZMA_DATA_ERROR;
267 
268 	// Do the actual compression.
269 	lzma_ret ret = LZMA_BUF_ERROR;
270 	if (try_to_compress)
271 		ret = block_encode_normal(block, allocator,
272 				in, in_size, out, out_pos, out_size);
273 
274 	if (ret != LZMA_OK) {
275 		// If the error was something else than output buffer
276 		// becoming full, return the error now.
277 		if (ret != LZMA_BUF_ERROR)
278 			return ret;
279 
280 		// The data was uncompressible (at least with the options
281 		// given to us) or the output buffer was too small. Use the
282 		// uncompressed chunks of LZMA2 to wrap the data into a valid
283 		// Block. If we haven't been given enough output space, even
284 		// this may fail.
285 		return_if_error(block_encode_uncompressed(block, in, in_size,
286 				out, out_pos, out_size));
287 	}
288 
289 	assert(*out_pos <= out_size);
290 
291 	// Block Padding. No buffer overflow here, because we already adjusted
292 	// out_size so that (out_size - out_start) is a multiple of four.
293 	// Thus, if the buffer is full, the loop body can never run.
294 	for (size_t i = (size_t)(block->compressed_size); i & 3; ++i) {
295 		assert(*out_pos < out_size);
296 		out[(*out_pos)++] = 0x00;
297 	}
298 
299 	// If there's no Check field, we are done now.
300 	if (check_size > 0) {
301 		// Calculate the integrity check. We reserved space for
302 		// the Check field earlier so we don't need to check for
303 		// available output space here.
304 		lzma_check_state check;
305 		lzma_check_init(&check, block->check);
306 		lzma_check_update(&check, block->check, in, in_size);
307 		lzma_check_finish(&check, block->check);
308 
309 		memcpy(block->raw_check, check.buffer.u8, check_size);
310 		memcpy(out + *out_pos, check.buffer.u8, check_size);
311 		*out_pos += check_size;
312 	}
313 
314 	return LZMA_OK;
315 }
316 
317 
318 extern LZMA_API(lzma_ret)
319 lzma_block_buffer_encode(lzma_block *block, const lzma_allocator *allocator,
320 		const uint8_t *in, size_t in_size,
321 		uint8_t *out, size_t *out_pos, size_t out_size)
322 {
323 	return block_buffer_encode(block, allocator,
324 			in, in_size, out, out_pos, out_size, true);
325 }
326 
327 
328 extern LZMA_API(lzma_ret)
329 lzma_block_uncomp_encode(lzma_block *block,
330 		const uint8_t *in, size_t in_size,
331 		uint8_t *out, size_t *out_pos, size_t out_size)
332 {
333 	// It won't allocate any memory from heap so no need
334 	// for lzma_allocator.
335 	return block_buffer_encode(block, NULL,
336 			in, in_size, out, out_pos, out_size, false);
337 }
338