xref: /freebsd/contrib/xz/src/liblzma/common/block_buffer_encoder.c (revision 59c8e88e72633afbc47a4ace0d2170d00d51f7dc)
1 // SPDX-License-Identifier: 0BSD
2 
3 ///////////////////////////////////////////////////////////////////////////////
4 //
5 /// \file       block_buffer_encoder.c
6 /// \brief      Single-call .xz Block encoder
7 //
8 //  Author:     Lasse Collin
9 //
10 ///////////////////////////////////////////////////////////////////////////////
11 
12 #include "block_buffer_encoder.h"
13 #include "block_encoder.h"
14 #include "filter_encoder.h"
15 #include "lzma2_encoder.h"
16 #include "check.h"
17 
18 
19 /// Estimate the maximum size of the Block Header and Check fields for
20 /// a Block that uses LZMA2 uncompressed chunks. We could use
21 /// lzma_block_header_size() but this is simpler.
22 ///
23 /// Block Header Size + Block Flags + Compressed Size
24 /// + Uncompressed Size + Filter Flags for LZMA2 + CRC32 + Check
25 /// and round up to the next multiple of four to take Header Padding
26 /// into account.
27 #define HEADERS_BOUND ((1 + 1 + 2 * LZMA_VLI_BYTES_MAX + 3 + 4 \
28 		+ LZMA_CHECK_SIZE_MAX + 3) & ~3)
29 
30 
31 static uint64_t
32 lzma2_bound(uint64_t uncompressed_size)
33 {
34 	// Prevent integer overflow in overhead calculation.
35 	if (uncompressed_size > COMPRESSED_SIZE_MAX)
36 		return 0;
37 
38 	// Calculate the exact overhead of the LZMA2 headers: Round
39 	// uncompressed_size up to the next multiple of LZMA2_CHUNK_MAX,
40 	// multiply by the size of per-chunk header, and add one byte for
41 	// the end marker.
42 	const uint64_t overhead = ((uncompressed_size + LZMA2_CHUNK_MAX - 1)
43 				/ LZMA2_CHUNK_MAX)
44 			* LZMA2_HEADER_UNCOMPRESSED + 1;
45 
46 	// Catch the possible integer overflow.
47 	if (COMPRESSED_SIZE_MAX - overhead < uncompressed_size)
48 		return 0;
49 
50 	return uncompressed_size + overhead;
51 }
52 
53 
54 extern uint64_t
55 lzma_block_buffer_bound64(uint64_t uncompressed_size)
56 {
57 	// If the data doesn't compress, we always use uncompressed
58 	// LZMA2 chunks.
59 	uint64_t lzma2_size = lzma2_bound(uncompressed_size);
60 	if (lzma2_size == 0)
61 		return 0;
62 
63 	// Take Block Padding into account.
64 	lzma2_size = (lzma2_size + 3) & ~UINT64_C(3);
65 
66 	// No risk of integer overflow because lzma2_bound() already takes
67 	// into account the size of the headers in the Block.
68 	return HEADERS_BOUND + lzma2_size;
69 }
70 
71 
72 extern LZMA_API(size_t)
73 lzma_block_buffer_bound(size_t uncompressed_size)
74 {
75 	uint64_t ret = lzma_block_buffer_bound64(uncompressed_size);
76 
77 #if SIZE_MAX < UINT64_MAX
78 	// Catch the possible integer overflow on 32-bit systems.
79 	if (ret > SIZE_MAX)
80 		return 0;
81 #endif
82 
83 	return ret;
84 }
85 
86 
87 static lzma_ret
88 block_encode_uncompressed(lzma_block *block, const uint8_t *in, size_t in_size,
89 		uint8_t *out, size_t *out_pos, size_t out_size)
90 {
91 	// Use LZMA2 uncompressed chunks. We wouldn't need a dictionary at
92 	// all, but LZMA2 always requires a dictionary, so use the minimum
93 	// value to minimize memory usage of the decoder.
94 	lzma_options_lzma lzma2 = {
95 		.dict_size = LZMA_DICT_SIZE_MIN,
96 	};
97 
98 	lzma_filter filters[2];
99 	filters[0].id = LZMA_FILTER_LZMA2;
100 	filters[0].options = &lzma2;
101 	filters[1].id = LZMA_VLI_UNKNOWN;
102 
103 	// Set the above filter options to *block temporarily so that we can
104 	// encode the Block Header.
105 	lzma_filter *filters_orig = block->filters;
106 	block->filters = filters;
107 
108 	if (lzma_block_header_size(block) != LZMA_OK) {
109 		block->filters = filters_orig;
110 		return LZMA_PROG_ERROR;
111 	}
112 
113 	// Check that there's enough output space. The caller has already
114 	// set block->compressed_size to what lzma2_bound() has returned,
115 	// so we can reuse that value. We know that compressed_size is a
116 	// known valid VLI and header_size is a small value so their sum
117 	// will never overflow.
118 	assert(block->compressed_size == lzma2_bound(in_size));
119 	if (out_size - *out_pos
120 			< block->header_size + block->compressed_size) {
121 		block->filters = filters_orig;
122 		return LZMA_BUF_ERROR;
123 	}
124 
125 	if (lzma_block_header_encode(block, out + *out_pos) != LZMA_OK) {
126 		block->filters = filters_orig;
127 		return LZMA_PROG_ERROR;
128 	}
129 
130 	block->filters = filters_orig;
131 	*out_pos += block->header_size;
132 
133 	// Encode the data using LZMA2 uncompressed chunks.
134 	size_t in_pos = 0;
135 	uint8_t control = 0x01; // Dictionary reset
136 
137 	while (in_pos < in_size) {
138 		// Control byte: Indicate uncompressed chunk, of which
139 		// the first resets the dictionary.
140 		out[(*out_pos)++] = control;
141 		control = 0x02; // No dictionary reset
142 
143 		// Size of the uncompressed chunk
144 		const size_t copy_size
145 				= my_min(in_size - in_pos, LZMA2_CHUNK_MAX);
146 		out[(*out_pos)++] = (copy_size - 1) >> 8;
147 		out[(*out_pos)++] = (copy_size - 1) & 0xFF;
148 
149 		// The actual data
150 		assert(*out_pos + copy_size <= out_size);
151 		memcpy(out + *out_pos, in + in_pos, copy_size);
152 
153 		in_pos += copy_size;
154 		*out_pos += copy_size;
155 	}
156 
157 	// End marker
158 	out[(*out_pos)++] = 0x00;
159 	assert(*out_pos <= out_size);
160 
161 	return LZMA_OK;
162 }
163 
164 
165 static lzma_ret
166 block_encode_normal(lzma_block *block, const lzma_allocator *allocator,
167 		const uint8_t *in, size_t in_size,
168 		uint8_t *out, size_t *out_pos, size_t out_size)
169 {
170 	// Find out the size of the Block Header.
171 	return_if_error(lzma_block_header_size(block));
172 
173 	// Reserve space for the Block Header and skip it for now.
174 	if (out_size - *out_pos <= block->header_size)
175 		return LZMA_BUF_ERROR;
176 
177 	const size_t out_start = *out_pos;
178 	*out_pos += block->header_size;
179 
180 	// Limit out_size so that we stop encoding if the output would grow
181 	// bigger than what uncompressed Block would be.
182 	if (out_size - *out_pos > block->compressed_size)
183 		out_size = *out_pos + block->compressed_size;
184 
185 	// TODO: In many common cases this could be optimized to use
186 	// significantly less memory.
187 	lzma_next_coder raw_encoder = LZMA_NEXT_CODER_INIT;
188 	lzma_ret ret = lzma_raw_encoder_init(
189 			&raw_encoder, allocator, block->filters);
190 
191 	if (ret == LZMA_OK) {
192 		size_t in_pos = 0;
193 		ret = raw_encoder.code(raw_encoder.coder, allocator,
194 				in, &in_pos, in_size, out, out_pos, out_size,
195 				LZMA_FINISH);
196 	}
197 
198 	// NOTE: This needs to be run even if lzma_raw_encoder_init() failed.
199 	lzma_next_end(&raw_encoder, allocator);
200 
201 	if (ret == LZMA_STREAM_END) {
202 		// Compression was successful. Write the Block Header.
203 		block->compressed_size
204 				= *out_pos - (out_start + block->header_size);
205 		ret = lzma_block_header_encode(block, out + out_start);
206 		if (ret != LZMA_OK)
207 			ret = LZMA_PROG_ERROR;
208 
209 	} else if (ret == LZMA_OK) {
210 		// Output buffer became full.
211 		ret = LZMA_BUF_ERROR;
212 	}
213 
214 	// Reset *out_pos if something went wrong.
215 	if (ret != LZMA_OK)
216 		*out_pos = out_start;
217 
218 	return ret;
219 }
220 
221 
222 static lzma_ret
223 block_buffer_encode(lzma_block *block, const lzma_allocator *allocator,
224 		const uint8_t *in, size_t in_size,
225 		uint8_t *out, size_t *out_pos, size_t out_size,
226 		bool try_to_compress)
227 {
228 	// Validate the arguments.
229 	if (block == NULL || (in == NULL && in_size != 0) || out == NULL
230 			|| out_pos == NULL || *out_pos > out_size)
231 		return LZMA_PROG_ERROR;
232 
233 	// The contents of the structure may depend on the version so
234 	// check the version before validating the contents of *block.
235 	if (block->version > 1)
236 		return LZMA_OPTIONS_ERROR;
237 
238 	if ((unsigned int)(block->check) > LZMA_CHECK_ID_MAX
239 			|| (try_to_compress && block->filters == NULL))
240 		return LZMA_PROG_ERROR;
241 
242 	if (!lzma_check_is_supported(block->check))
243 		return LZMA_UNSUPPORTED_CHECK;
244 
245 	// Size of a Block has to be a multiple of four, so limit the size
246 	// here already. This way we don't need to check it again when adding
247 	// Block Padding.
248 	out_size -= (out_size - *out_pos) & 3;
249 
250 	// Get the size of the Check field.
251 	const size_t check_size = lzma_check_size(block->check);
252 	assert(check_size != UINT32_MAX);
253 
254 	// Reserve space for the Check field.
255 	if (out_size - *out_pos <= check_size)
256 		return LZMA_BUF_ERROR;
257 
258 	out_size -= check_size;
259 
260 	// Initialize block->uncompressed_size and calculate the worst-case
261 	// value for block->compressed_size.
262 	block->uncompressed_size = in_size;
263 	block->compressed_size = lzma2_bound(in_size);
264 	if (block->compressed_size == 0)
265 		return LZMA_DATA_ERROR;
266 
267 	// Do the actual compression.
268 	lzma_ret ret = LZMA_BUF_ERROR;
269 	if (try_to_compress)
270 		ret = block_encode_normal(block, allocator,
271 				in, in_size, out, out_pos, out_size);
272 
273 	if (ret != LZMA_OK) {
274 		// If the error was something else than output buffer
275 		// becoming full, return the error now.
276 		if (ret != LZMA_BUF_ERROR)
277 			return ret;
278 
279 		// The data was incompressible (at least with the options
280 		// given to us) or the output buffer was too small. Use the
281 		// uncompressed chunks of LZMA2 to wrap the data into a valid
282 		// Block. If we haven't been given enough output space, even
283 		// this may fail.
284 		return_if_error(block_encode_uncompressed(block, in, in_size,
285 				out, out_pos, out_size));
286 	}
287 
288 	assert(*out_pos <= out_size);
289 
290 	// Block Padding. No buffer overflow here, because we already adjusted
291 	// out_size so that (out_size - out_start) is a multiple of four.
292 	// Thus, if the buffer is full, the loop body can never run.
293 	for (size_t i = (size_t)(block->compressed_size); i & 3; ++i) {
294 		assert(*out_pos < out_size);
295 		out[(*out_pos)++] = 0x00;
296 	}
297 
298 	// If there's no Check field, we are done now.
299 	if (check_size > 0) {
300 		// Calculate the integrity check. We reserved space for
301 		// the Check field earlier so we don't need to check for
302 		// available output space here.
303 		lzma_check_state check;
304 		lzma_check_init(&check, block->check);
305 		lzma_check_update(&check, block->check, in, in_size);
306 		lzma_check_finish(&check, block->check);
307 
308 		memcpy(block->raw_check, check.buffer.u8, check_size);
309 		memcpy(out + *out_pos, check.buffer.u8, check_size);
310 		*out_pos += check_size;
311 	}
312 
313 	return LZMA_OK;
314 }
315 
316 
317 extern LZMA_API(lzma_ret)
318 lzma_block_buffer_encode(lzma_block *block, const lzma_allocator *allocator,
319 		const uint8_t *in, size_t in_size,
320 		uint8_t *out, size_t *out_pos, size_t out_size)
321 {
322 	return block_buffer_encode(block, allocator,
323 			in, in_size, out, out_pos, out_size, true);
324 }
325 
326 
327 #ifdef HAVE_SYMBOL_VERSIONS_LINUX
328 // This is for compatibility with binaries linked against liblzma that
329 // has been patched with xz-5.2.2-compat-libs.patch from RHEL/CentOS 7.
330 LZMA_SYMVER_API("lzma_block_uncomp_encode@XZ_5.2.2",
331 	lzma_ret, lzma_block_uncomp_encode_522)(lzma_block *block,
332 		const uint8_t *in, size_t in_size,
333 		uint8_t *out, size_t *out_pos, size_t out_size)
334 		lzma_nothrow lzma_attr_warn_unused_result
335 		__attribute__((__alias__("lzma_block_uncomp_encode_52")));
336 
337 LZMA_SYMVER_API("lzma_block_uncomp_encode@@XZ_5.2",
338 	lzma_ret, lzma_block_uncomp_encode_52)(lzma_block *block,
339 		const uint8_t *in, size_t in_size,
340 		uint8_t *out, size_t *out_pos, size_t out_size)
341 		lzma_nothrow lzma_attr_warn_unused_result;
342 
343 #define lzma_block_uncomp_encode lzma_block_uncomp_encode_52
344 #endif
345 extern LZMA_API(lzma_ret)
346 lzma_block_uncomp_encode(lzma_block *block,
347 		const uint8_t *in, size_t in_size,
348 		uint8_t *out, size_t *out_pos, size_t out_size)
349 {
350 	// It won't allocate any memory from heap so no need
351 	// for lzma_allocator.
352 	return block_buffer_encode(block, NULL,
353 			in, in_size, out, out_pos, out_size, false);
354 }
355