xref: /freebsd/contrib/xz/src/liblzma/lzma/lzma2_encoder.c (revision 3b35e7ee8de9b0260149a2b77e87a2b9c7a36244)
1 // SPDX-License-Identifier: 0BSD
2 
3 ///////////////////////////////////////////////////////////////////////////////
4 //
5 /// \file       lzma2_encoder.c
6 /// \brief      LZMA2 encoder
7 ///
8 //  Authors:    Igor Pavlov
9 //              Lasse Collin
10 //
11 ///////////////////////////////////////////////////////////////////////////////
12 
13 #include "lz_encoder.h"
14 #include "lzma_encoder.h"
15 #include "fastpos.h"
16 #include "lzma2_encoder.h"
17 
18 
19 typedef struct {
20 	enum {
21 		SEQ_INIT,
22 		SEQ_LZMA_ENCODE,
23 		SEQ_LZMA_COPY,
24 		SEQ_UNCOMPRESSED_HEADER,
25 		SEQ_UNCOMPRESSED_COPY,
26 	} sequence;
27 
28 	/// LZMA encoder
29 	void *lzma;
30 
31 	/// LZMA options currently in use.
32 	lzma_options_lzma opt_cur;
33 
34 	bool need_properties;
35 	bool need_state_reset;
36 	bool need_dictionary_reset;
37 
38 	/// Uncompressed size of a chunk
39 	size_t uncompressed_size;
40 
41 	/// Compressed size of a chunk (excluding headers); this is also used
42 	/// to indicate the end of buf[] in SEQ_LZMA_COPY.
43 	size_t compressed_size;
44 
45 	/// Read position in buf[]
46 	size_t buf_pos;
47 
48 	/// Buffer to hold the chunk header and LZMA compressed data
49 	uint8_t buf[LZMA2_HEADER_MAX + LZMA2_CHUNK_MAX];
50 } lzma_lzma2_coder;
51 
52 
53 static void
54 lzma2_header_lzma(lzma_lzma2_coder *coder)
55 {
56 	assert(coder->uncompressed_size > 0);
57 	assert(coder->uncompressed_size <= LZMA2_UNCOMPRESSED_MAX);
58 	assert(coder->compressed_size > 0);
59 	assert(coder->compressed_size <= LZMA2_CHUNK_MAX);
60 
61 	size_t pos;
62 
63 	if (coder->need_properties) {
64 		pos = 0;
65 
66 		if (coder->need_dictionary_reset)
67 			coder->buf[pos] = 0x80 + (3 << 5);
68 		else
69 			coder->buf[pos] = 0x80 + (2 << 5);
70 	} else {
71 		pos = 1;
72 
73 		if (coder->need_state_reset)
74 			coder->buf[pos] = 0x80 + (1 << 5);
75 		else
76 			coder->buf[pos] = 0x80;
77 	}
78 
79 	// Set the start position for copying.
80 	coder->buf_pos = pos;
81 
82 	// Uncompressed size
83 	size_t size = coder->uncompressed_size - 1;
84 	coder->buf[pos++] += size >> 16;
85 	coder->buf[pos++] = (size >> 8) & 0xFF;
86 	coder->buf[pos++] = size & 0xFF;
87 
88 	// Compressed size
89 	size = coder->compressed_size - 1;
90 	coder->buf[pos++] = size >> 8;
91 	coder->buf[pos++] = size & 0xFF;
92 
93 	// Properties, if needed
94 	if (coder->need_properties)
95 		lzma_lzma_lclppb_encode(&coder->opt_cur, coder->buf + pos);
96 
97 	coder->need_properties = false;
98 	coder->need_state_reset = false;
99 	coder->need_dictionary_reset = false;
100 
101 	// The copying code uses coder->compressed_size to indicate the end
102 	// of coder->buf[], so we need add the maximum size of the header here.
103 	coder->compressed_size += LZMA2_HEADER_MAX;
104 
105 	return;
106 }
107 
108 
109 static void
110 lzma2_header_uncompressed(lzma_lzma2_coder *coder)
111 {
112 	assert(coder->uncompressed_size > 0);
113 	assert(coder->uncompressed_size <= LZMA2_CHUNK_MAX);
114 
115 	// If this is the first chunk, we need to include dictionary
116 	// reset indicator.
117 	if (coder->need_dictionary_reset)
118 		coder->buf[0] = 1;
119 	else
120 		coder->buf[0] = 2;
121 
122 	coder->need_dictionary_reset = false;
123 
124 	// "Compressed" size
125 	coder->buf[1] = (coder->uncompressed_size - 1) >> 8;
126 	coder->buf[2] = (coder->uncompressed_size - 1) & 0xFF;
127 
128 	// Set the start position for copying.
129 	coder->buf_pos = 0;
130 	return;
131 }
132 
133 
134 static lzma_ret
135 lzma2_encode(void *coder_ptr, lzma_mf *restrict mf,
136 		uint8_t *restrict out, size_t *restrict out_pos,
137 		size_t out_size)
138 {
139 	lzma_lzma2_coder *restrict coder = coder_ptr;
140 
141 	while (*out_pos < out_size)
142 	switch (coder->sequence) {
143 	case SEQ_INIT:
144 		// If there's no input left and we are flushing or finishing,
145 		// don't start a new chunk.
146 		if (mf_unencoded(mf) == 0) {
147 			// Write end of payload marker if finishing.
148 			if (mf->action == LZMA_FINISH)
149 				out[(*out_pos)++] = 0;
150 
151 			return mf->action == LZMA_RUN
152 					? LZMA_OK : LZMA_STREAM_END;
153 		}
154 
155 		if (coder->need_state_reset)
156 			return_if_error(lzma_lzma_encoder_reset(
157 					coder->lzma, &coder->opt_cur));
158 
159 		coder->uncompressed_size = 0;
160 		coder->compressed_size = 0;
161 		coder->sequence = SEQ_LZMA_ENCODE;
162 
163 	// Fall through
164 
165 	case SEQ_LZMA_ENCODE: {
166 		// Calculate how much more uncompressed data this chunk
167 		// could accept.
168 		const uint32_t left = LZMA2_UNCOMPRESSED_MAX
169 				- coder->uncompressed_size;
170 		uint32_t limit;
171 
172 		if (left < mf->match_len_max) {
173 			// Must flush immediately since the next LZMA symbol
174 			// could make the uncompressed size of the chunk too
175 			// big.
176 			limit = 0;
177 		} else {
178 			// Calculate maximum read_limit that is OK from point
179 			// of view of LZMA2 chunk size.
180 			limit = mf->read_pos - mf->read_ahead
181 					+ left - mf->match_len_max;
182 		}
183 
184 		// Save the start position so that we can update
185 		// coder->uncompressed_size.
186 		const uint32_t read_start = mf->read_pos - mf->read_ahead;
187 
188 		// Call the LZMA encoder until the chunk is finished.
189 		const lzma_ret ret = lzma_lzma_encode(coder->lzma, mf,
190 				coder->buf + LZMA2_HEADER_MAX,
191 				&coder->compressed_size,
192 				LZMA2_CHUNK_MAX, limit);
193 
194 		coder->uncompressed_size += mf->read_pos - mf->read_ahead
195 				- read_start;
196 
197 		assert(coder->compressed_size <= LZMA2_CHUNK_MAX);
198 		assert(coder->uncompressed_size <= LZMA2_UNCOMPRESSED_MAX);
199 
200 		if (ret != LZMA_STREAM_END)
201 			return LZMA_OK;
202 
203 		// See if the chunk compressed. If it didn't, we encode it
204 		// as uncompressed chunk. This saves a few bytes of space
205 		// and makes decoding faster.
206 		if (coder->compressed_size >= coder->uncompressed_size) {
207 			coder->uncompressed_size += mf->read_ahead;
208 			assert(coder->uncompressed_size
209 					<= LZMA2_UNCOMPRESSED_MAX);
210 			mf->read_ahead = 0;
211 			lzma2_header_uncompressed(coder);
212 			coder->need_state_reset = true;
213 			coder->sequence = SEQ_UNCOMPRESSED_HEADER;
214 			break;
215 		}
216 
217 		// The chunk did compress at least by one byte, so we store
218 		// the chunk as LZMA.
219 		lzma2_header_lzma(coder);
220 
221 		coder->sequence = SEQ_LZMA_COPY;
222 	}
223 
224 	// Fall through
225 
226 	case SEQ_LZMA_COPY:
227 		// Copy the compressed chunk along its headers to the
228 		// output buffer.
229 		lzma_bufcpy(coder->buf, &coder->buf_pos,
230 				coder->compressed_size,
231 				out, out_pos, out_size);
232 		if (coder->buf_pos != coder->compressed_size)
233 			return LZMA_OK;
234 
235 		coder->sequence = SEQ_INIT;
236 		break;
237 
238 	case SEQ_UNCOMPRESSED_HEADER:
239 		// Copy the three-byte header to indicate uncompressed chunk.
240 		lzma_bufcpy(coder->buf, &coder->buf_pos,
241 				LZMA2_HEADER_UNCOMPRESSED,
242 				out, out_pos, out_size);
243 		if (coder->buf_pos != LZMA2_HEADER_UNCOMPRESSED)
244 			return LZMA_OK;
245 
246 		coder->sequence = SEQ_UNCOMPRESSED_COPY;
247 
248 	// Fall through
249 
250 	case SEQ_UNCOMPRESSED_COPY:
251 		// Copy the uncompressed data as is from the dictionary
252 		// to the output buffer.
253 		mf_read(mf, out, out_pos, out_size, &coder->uncompressed_size);
254 		if (coder->uncompressed_size != 0)
255 			return LZMA_OK;
256 
257 		coder->sequence = SEQ_INIT;
258 		break;
259 	}
260 
261 	return LZMA_OK;
262 }
263 
264 
265 static void
266 lzma2_encoder_end(void *coder_ptr, const lzma_allocator *allocator)
267 {
268 	lzma_lzma2_coder *coder = coder_ptr;
269 	lzma_free(coder->lzma, allocator);
270 	lzma_free(coder, allocator);
271 	return;
272 }
273 
274 
275 static lzma_ret
276 lzma2_encoder_options_update(void *coder_ptr, const lzma_filter *filter)
277 {
278 	lzma_lzma2_coder *coder = coder_ptr;
279 
280 	// New options can be set only when there is no incomplete chunk.
281 	// This is the case at the beginning of the raw stream and right
282 	// after LZMA_SYNC_FLUSH.
283 	if (filter->options == NULL || coder->sequence != SEQ_INIT)
284 		return LZMA_PROG_ERROR;
285 
286 	// Look if there are new options. At least for now,
287 	// only lc/lp/pb can be changed.
288 	const lzma_options_lzma *opt = filter->options;
289 	if (coder->opt_cur.lc != opt->lc || coder->opt_cur.lp != opt->lp
290 			|| coder->opt_cur.pb != opt->pb) {
291 		// Validate the options.
292 		if (opt->lc > LZMA_LCLP_MAX || opt->lp > LZMA_LCLP_MAX
293 				|| opt->lc + opt->lp > LZMA_LCLP_MAX
294 				|| opt->pb > LZMA_PB_MAX)
295 			return LZMA_OPTIONS_ERROR;
296 
297 		// The new options will be used when the encoder starts
298 		// a new LZMA2 chunk.
299 		coder->opt_cur.lc = opt->lc;
300 		coder->opt_cur.lp = opt->lp;
301 		coder->opt_cur.pb = opt->pb;
302 		coder->need_properties = true;
303 		coder->need_state_reset = true;
304 	}
305 
306 	return LZMA_OK;
307 }
308 
309 
310 static lzma_ret
311 lzma2_encoder_init(lzma_lz_encoder *lz, const lzma_allocator *allocator,
312 		lzma_vli id lzma_attribute((__unused__)), const void *options,
313 		lzma_lz_options *lz_options)
314 {
315 	if (options == NULL)
316 		return LZMA_PROG_ERROR;
317 
318 	lzma_lzma2_coder *coder = lz->coder;
319 	if (coder == NULL) {
320 		coder = lzma_alloc(sizeof(lzma_lzma2_coder), allocator);
321 		if (coder == NULL)
322 			return LZMA_MEM_ERROR;
323 
324 		lz->coder = coder;
325 		lz->code = &lzma2_encode;
326 		lz->end = &lzma2_encoder_end;
327 		lz->options_update = &lzma2_encoder_options_update;
328 
329 		coder->lzma = NULL;
330 	}
331 
332 	coder->opt_cur = *(const lzma_options_lzma *)(options);
333 
334 	coder->sequence = SEQ_INIT;
335 	coder->need_properties = true;
336 	coder->need_state_reset = false;
337 	coder->need_dictionary_reset
338 			= coder->opt_cur.preset_dict == NULL
339 			|| coder->opt_cur.preset_dict_size == 0;
340 
341 	// Initialize LZMA encoder
342 	return_if_error(lzma_lzma_encoder_create(&coder->lzma, allocator,
343 			LZMA_FILTER_LZMA2, &coder->opt_cur, lz_options));
344 
345 	// Make sure that we will always have enough history available in
346 	// case we need to use uncompressed chunks. They are used when the
347 	// compressed size of a chunk is not smaller than the uncompressed
348 	// size, so we need to have at least LZMA2_COMPRESSED_MAX bytes
349 	// history available.
350 	if (lz_options->before_size + lz_options->dict_size < LZMA2_CHUNK_MAX)
351 		lz_options->before_size
352 				= LZMA2_CHUNK_MAX - lz_options->dict_size;
353 
354 	return LZMA_OK;
355 }
356 
357 
358 extern lzma_ret
359 lzma_lzma2_encoder_init(lzma_next_coder *next, const lzma_allocator *allocator,
360 		const lzma_filter_info *filters)
361 {
362 	return lzma_lz_encoder_init(
363 			next, allocator, filters, &lzma2_encoder_init);
364 }
365 
366 
367 extern uint64_t
368 lzma_lzma2_encoder_memusage(const void *options)
369 {
370 	const uint64_t lzma_mem = lzma_lzma_encoder_memusage(options);
371 	if (lzma_mem == UINT64_MAX)
372 		return UINT64_MAX;
373 
374 	return sizeof(lzma_lzma2_coder) + lzma_mem;
375 }
376 
377 
378 extern lzma_ret
379 lzma_lzma2_props_encode(const void *options, uint8_t *out)
380 {
381 	if (options == NULL)
382 		return LZMA_PROG_ERROR;
383 
384 	const lzma_options_lzma *const opt = options;
385 	uint32_t d = my_max(opt->dict_size, LZMA_DICT_SIZE_MIN);
386 
387 	// Round up to the next 2^n - 1 or 2^n + 2^(n - 1) - 1 depending
388 	// on which one is the next:
389 	--d;
390 	d |= d >> 2;
391 	d |= d >> 3;
392 	d |= d >> 4;
393 	d |= d >> 8;
394 	d |= d >> 16;
395 
396 	// Get the highest two bits using the proper encoding:
397 	if (d == UINT32_MAX)
398 		out[0] = 40;
399 	else
400 		out[0] = get_dist_slot(d + 1) - 24;
401 
402 	return LZMA_OK;
403 }
404 
405 
406 extern uint64_t
407 lzma_lzma2_block_size(const void *options)
408 {
409 	const lzma_options_lzma *const opt = options;
410 
411 	if (!IS_ENC_DICT_SIZE_VALID(opt->dict_size))
412 		return UINT64_MAX;
413 
414 	// Use at least 1 MiB to keep compression ratio better.
415 	return my_max((uint64_t)(opt->dict_size) * 3, UINT64_C(1) << 20);
416 }
417