xref: /freebsd/contrib/xz/src/liblzma/lz/lz_decoder.c (revision 8be96e101f2691b80ff9562b72f874da82e735aa)
1 ///////////////////////////////////////////////////////////////////////////////
2 //
3 /// \file       lz_decoder.c
4 /// \brief      LZ out window
5 ///
6 //  Authors:    Igor Pavlov
7 //              Lasse Collin
8 //
9 //  This file has been put into the public domain.
10 //  You can do whatever you want with this file.
11 //
12 ///////////////////////////////////////////////////////////////////////////////
13 
14 // liblzma supports multiple LZ77-based filters. The LZ part is shared
15 // between these filters. The LZ code takes care of dictionary handling
16 // and passing the data between filters in the chain. The filter-specific
17 // part decodes from the input buffer to the dictionary.
18 
19 
20 #include "lz_decoder.h"
21 
22 
23 struct lzma_coder_s {
24 	/// Dictionary (history buffer)
25 	lzma_dict dict;
26 
27 	/// The actual LZ-based decoder e.g. LZMA
28 	lzma_lz_decoder lz;
29 
30 	/// Next filter in the chain, if any. Note that LZMA and LZMA2 are
31 	/// only allowed as the last filter, but the long-range filter in
32 	/// future can be in the middle of the chain.
33 	lzma_next_coder next;
34 
35 	/// True if the next filter in the chain has returned LZMA_STREAM_END.
36 	bool next_finished;
37 
38 	/// True if the LZ decoder (e.g. LZMA) has detected end of payload
39 	/// marker. This may become true before next_finished becomes true.
40 	bool this_finished;
41 
42 	/// Temporary buffer needed when the LZ-based filter is not the last
43 	/// filter in the chain. The output of the next filter is first
44 	/// decoded into buffer[], which is then used as input for the actual
45 	/// LZ-based decoder.
46 	struct {
47 		size_t pos;
48 		size_t size;
49 		uint8_t buffer[LZMA_BUFFER_SIZE];
50 	} temp;
51 };
52 
53 
54 static void
55 lz_decoder_reset(lzma_coder *coder)
56 {
57 	coder->dict.pos = 0;
58 	coder->dict.full = 0;
59 	coder->dict.buf[coder->dict.size - 1] = '\0';
60 	coder->dict.need_reset = false;
61 	return;
62 }
63 
64 
65 static lzma_ret
66 decode_buffer(lzma_coder *coder,
67 		const uint8_t *restrict in, size_t *restrict in_pos,
68 		size_t in_size, uint8_t *restrict out,
69 		size_t *restrict out_pos, size_t out_size)
70 {
71 	while (true) {
72 		// Wrap the dictionary if needed.
73 		if (coder->dict.pos == coder->dict.size)
74 			coder->dict.pos = 0;
75 
76 		// Store the current dictionary position. It is needed to know
77 		// where to start copying to the out[] buffer.
78 		const size_t dict_start = coder->dict.pos;
79 
80 		// Calculate how much we allow coder->lz.code() to decode.
81 		// It must not decode past the end of the dictionary
82 		// buffer, and we don't want it to decode more than is
83 		// actually needed to fill the out[] buffer.
84 		coder->dict.limit = coder->dict.pos + MIN(out_size - *out_pos,
85 				coder->dict.size - coder->dict.pos);
86 
87 		// Call the coder->lz.code() to do the actual decoding.
88 		const lzma_ret ret = coder->lz.code(
89 				coder->lz.coder, &coder->dict,
90 				in, in_pos, in_size);
91 
92 		// Copy the decoded data from the dictionary to the out[]
93 		// buffer.
94 		const size_t copy_size = coder->dict.pos - dict_start;
95 		assert(copy_size <= out_size - *out_pos);
96 		memcpy(out + *out_pos, coder->dict.buf + dict_start,
97 				copy_size);
98 		*out_pos += copy_size;
99 
100 		// Reset the dictionary if so requested by coder->lz.code().
101 		if (coder->dict.need_reset) {
102 			lz_decoder_reset(coder);
103 
104 			// Since we reset dictionary, we don't check if
105 			// dictionary became full.
106 			if (ret != LZMA_OK || *out_pos == out_size)
107 				return ret;
108 		} else {
109 			// Return if everything got decoded or an error
110 			// occurred, or if there's no more data to decode.
111 			//
112 			// Note that detecting if there's something to decode
113 			// is done by looking if dictionary become full
114 			// instead of looking if *in_pos == in_size. This
115 			// is because it is possible that all the input was
116 			// consumed already but some data is pending to be
117 			// written to the dictionary.
118 			if (ret != LZMA_OK || *out_pos == out_size
119 					|| coder->dict.pos < coder->dict.size)
120 				return ret;
121 		}
122 	}
123 }
124 
125 
126 static lzma_ret
127 lz_decode(lzma_coder *coder,
128 		lzma_allocator *allocator lzma_attribute((unused)),
129 		const uint8_t *restrict in, size_t *restrict in_pos,
130 		size_t in_size, uint8_t *restrict out,
131 		size_t *restrict out_pos, size_t out_size,
132 		lzma_action action)
133 {
134 	if (coder->next.code == NULL)
135 		return decode_buffer(coder, in, in_pos, in_size,
136 				out, out_pos, out_size);
137 
138 	// We aren't the last coder in the chain, we need to decode
139 	// our input to a temporary buffer.
140 	while (*out_pos < out_size) {
141 		// Fill the temporary buffer if it is empty.
142 		if (!coder->next_finished
143 				&& coder->temp.pos == coder->temp.size) {
144 			coder->temp.pos = 0;
145 			coder->temp.size = 0;
146 
147 			const lzma_ret ret = coder->next.code(
148 					coder->next.coder,
149 					allocator, in, in_pos, in_size,
150 					coder->temp.buffer, &coder->temp.size,
151 					LZMA_BUFFER_SIZE, action);
152 
153 			if (ret == LZMA_STREAM_END)
154 				coder->next_finished = true;
155 			else if (ret != LZMA_OK || coder->temp.size == 0)
156 				return ret;
157 		}
158 
159 		if (coder->this_finished) {
160 			if (coder->temp.size != 0)
161 				return LZMA_DATA_ERROR;
162 
163 			if (coder->next_finished)
164 				return LZMA_STREAM_END;
165 
166 			return LZMA_OK;
167 		}
168 
169 		const lzma_ret ret = decode_buffer(coder, coder->temp.buffer,
170 				&coder->temp.pos, coder->temp.size,
171 				out, out_pos, out_size);
172 
173 		if (ret == LZMA_STREAM_END)
174 			coder->this_finished = true;
175 		else if (ret != LZMA_OK)
176 			return ret;
177 		else if (coder->next_finished && *out_pos < out_size)
178 			return LZMA_DATA_ERROR;
179 	}
180 
181 	return LZMA_OK;
182 }
183 
184 
185 static void
186 lz_decoder_end(lzma_coder *coder, lzma_allocator *allocator)
187 {
188 	lzma_next_end(&coder->next, allocator);
189 	lzma_free(coder->dict.buf, allocator);
190 
191 	if (coder->lz.end != NULL)
192 		coder->lz.end(coder->lz.coder, allocator);
193 	else
194 		lzma_free(coder->lz.coder, allocator);
195 
196 	lzma_free(coder, allocator);
197 	return;
198 }
199 
200 
201 extern lzma_ret
202 lzma_lz_decoder_init(lzma_next_coder *next, lzma_allocator *allocator,
203 		const lzma_filter_info *filters,
204 		lzma_ret (*lz_init)(lzma_lz_decoder *lz,
205 			lzma_allocator *allocator, const void *options,
206 			lzma_lz_options *lz_options))
207 {
208 	// Allocate the base structure if it isn't already allocated.
209 	if (next->coder == NULL) {
210 		next->coder = lzma_alloc(sizeof(lzma_coder), allocator);
211 		if (next->coder == NULL)
212 			return LZMA_MEM_ERROR;
213 
214 		next->code = &lz_decode;
215 		next->end = &lz_decoder_end;
216 
217 		next->coder->dict.buf = NULL;
218 		next->coder->dict.size = 0;
219 		next->coder->lz = LZMA_LZ_DECODER_INIT;
220 		next->coder->next = LZMA_NEXT_CODER_INIT;
221 	}
222 
223 	// Allocate and initialize the LZ-based decoder. It will also give
224 	// us the dictionary size.
225 	lzma_lz_options lz_options;
226 	return_if_error(lz_init(&next->coder->lz, allocator,
227 			filters[0].options, &lz_options));
228 
229 	// If the dictionary size is very small, increase it to 4096 bytes.
230 	// This is to prevent constant wrapping of the dictionary, which
231 	// would slow things down. The downside is that since we don't check
232 	// separately for the real dictionary size, we may happily accept
233 	// corrupt files.
234 	if (lz_options.dict_size < 4096)
235 		lz_options.dict_size = 4096;
236 
237 	// Make dictionary size a multipe of 16. Some LZ-based decoders like
238 	// LZMA use the lowest bits lzma_dict.pos to know the alignment of the
239 	// data. Aligned buffer is also good when memcpying from the
240 	// dictionary to the output buffer, since applications are
241 	// recommended to give aligned buffers to liblzma.
242 	//
243 	// Avoid integer overflow.
244 	if (lz_options.dict_size > SIZE_MAX - 15)
245 		return LZMA_MEM_ERROR;
246 
247 	lz_options.dict_size = (lz_options.dict_size + 15) & ~((size_t)(15));
248 
249 	// Allocate and initialize the dictionary.
250 	if (next->coder->dict.size != lz_options.dict_size) {
251 		lzma_free(next->coder->dict.buf, allocator);
252 		next->coder->dict.buf
253 				= lzma_alloc(lz_options.dict_size, allocator);
254 		if (next->coder->dict.buf == NULL)
255 			return LZMA_MEM_ERROR;
256 
257 		next->coder->dict.size = lz_options.dict_size;
258 	}
259 
260 	lz_decoder_reset(next->coder);
261 
262 	// Use the preset dictionary if it was given to us.
263 	if (lz_options.preset_dict != NULL
264 			&& lz_options.preset_dict_size > 0) {
265 		// If the preset dictionary is bigger than the actual
266 		// dictionary, copy only the tail.
267 		const size_t copy_size = MIN(lz_options.preset_dict_size,
268 				lz_options.dict_size);
269 		const size_t offset = lz_options.preset_dict_size - copy_size;
270 		memcpy(next->coder->dict.buf, lz_options.preset_dict + offset,
271 				copy_size);
272 		next->coder->dict.pos = copy_size;
273 		next->coder->dict.full = copy_size;
274 	}
275 
276 	// Miscellaneous initializations
277 	next->coder->next_finished = false;
278 	next->coder->this_finished = false;
279 	next->coder->temp.pos = 0;
280 	next->coder->temp.size = 0;
281 
282 	// Initialize the next filter in the chain, if any.
283 	return lzma_next_filter_init(&next->coder->next, allocator,
284 			filters + 1);
285 }
286 
287 
288 extern uint64_t
289 lzma_lz_decoder_memusage(size_t dictionary_size)
290 {
291 	return sizeof(lzma_coder) + (uint64_t)(dictionary_size);
292 }
293 
294 
295 extern void
296 lzma_lz_decoder_uncompressed(lzma_coder *coder, lzma_vli uncompressed_size)
297 {
298 	coder->lz.set_uncompressed(coder->lz.coder, uncompressed_size);
299 }
300