xref: /freebsd/contrib/xz/src/liblzma/lz/lz_decoder.c (revision 40a8ac8f62b535d30349faf28cf47106b7041b83)
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
85 				+ my_min(out_size - *out_pos,
86 					coder->dict.size - coder->dict.pos);
87 
88 		// Call the coder->lz.code() to do the actual decoding.
89 		const lzma_ret ret = coder->lz.code(
90 				coder->lz.coder, &coder->dict,
91 				in, in_pos, in_size);
92 
93 		// Copy the decoded data from the dictionary to the out[]
94 		// buffer.
95 		const size_t copy_size = coder->dict.pos - dict_start;
96 		assert(copy_size <= out_size - *out_pos);
97 		memcpy(out + *out_pos, coder->dict.buf + dict_start,
98 				copy_size);
99 		*out_pos += copy_size;
100 
101 		// Reset the dictionary if so requested by coder->lz.code().
102 		if (coder->dict.need_reset) {
103 			lz_decoder_reset(coder);
104 
105 			// Since we reset dictionary, we don't check if
106 			// dictionary became full.
107 			if (ret != LZMA_OK || *out_pos == out_size)
108 				return ret;
109 		} else {
110 			// Return if everything got decoded or an error
111 			// occurred, or if there's no more data to decode.
112 			//
113 			// Note that detecting if there's something to decode
114 			// is done by looking if dictionary become full
115 			// instead of looking if *in_pos == in_size. This
116 			// is because it is possible that all the input was
117 			// consumed already but some data is pending to be
118 			// written to the dictionary.
119 			if (ret != LZMA_OK || *out_pos == out_size
120 					|| coder->dict.pos < coder->dict.size)
121 				return ret;
122 		}
123 	}
124 }
125 
126 
127 static lzma_ret
128 lz_decode(lzma_coder *coder,
129 		lzma_allocator *allocator lzma_attribute((__unused__)),
130 		const uint8_t *restrict in, size_t *restrict in_pos,
131 		size_t in_size, uint8_t *restrict out,
132 		size_t *restrict out_pos, size_t out_size,
133 		lzma_action action)
134 {
135 	if (coder->next.code == NULL)
136 		return decode_buffer(coder, in, in_pos, in_size,
137 				out, out_pos, out_size);
138 
139 	// We aren't the last coder in the chain, we need to decode
140 	// our input to a temporary buffer.
141 	while (*out_pos < out_size) {
142 		// Fill the temporary buffer if it is empty.
143 		if (!coder->next_finished
144 				&& coder->temp.pos == coder->temp.size) {
145 			coder->temp.pos = 0;
146 			coder->temp.size = 0;
147 
148 			const lzma_ret ret = coder->next.code(
149 					coder->next.coder,
150 					allocator, in, in_pos, in_size,
151 					coder->temp.buffer, &coder->temp.size,
152 					LZMA_BUFFER_SIZE, action);
153 
154 			if (ret == LZMA_STREAM_END)
155 				coder->next_finished = true;
156 			else if (ret != LZMA_OK || coder->temp.size == 0)
157 				return ret;
158 		}
159 
160 		if (coder->this_finished) {
161 			if (coder->temp.size != 0)
162 				return LZMA_DATA_ERROR;
163 
164 			if (coder->next_finished)
165 				return LZMA_STREAM_END;
166 
167 			return LZMA_OK;
168 		}
169 
170 		const lzma_ret ret = decode_buffer(coder, coder->temp.buffer,
171 				&coder->temp.pos, coder->temp.size,
172 				out, out_pos, out_size);
173 
174 		if (ret == LZMA_STREAM_END)
175 			coder->this_finished = true;
176 		else if (ret != LZMA_OK)
177 			return ret;
178 		else if (coder->next_finished && *out_pos < out_size)
179 			return LZMA_DATA_ERROR;
180 	}
181 
182 	return LZMA_OK;
183 }
184 
185 
186 static void
187 lz_decoder_end(lzma_coder *coder, lzma_allocator *allocator)
188 {
189 	lzma_next_end(&coder->next, allocator);
190 	lzma_free(coder->dict.buf, allocator);
191 
192 	if (coder->lz.end != NULL)
193 		coder->lz.end(coder->lz.coder, allocator);
194 	else
195 		lzma_free(coder->lz.coder, allocator);
196 
197 	lzma_free(coder, allocator);
198 	return;
199 }
200 
201 
202 extern lzma_ret
203 lzma_lz_decoder_init(lzma_next_coder *next, lzma_allocator *allocator,
204 		const lzma_filter_info *filters,
205 		lzma_ret (*lz_init)(lzma_lz_decoder *lz,
206 			lzma_allocator *allocator, const void *options,
207 			lzma_lz_options *lz_options))
208 {
209 	// Allocate the base structure if it isn't already allocated.
210 	if (next->coder == NULL) {
211 		next->coder = lzma_alloc(sizeof(lzma_coder), allocator);
212 		if (next->coder == NULL)
213 			return LZMA_MEM_ERROR;
214 
215 		next->code = &lz_decode;
216 		next->end = &lz_decoder_end;
217 
218 		next->coder->dict.buf = NULL;
219 		next->coder->dict.size = 0;
220 		next->coder->lz = LZMA_LZ_DECODER_INIT;
221 		next->coder->next = LZMA_NEXT_CODER_INIT;
222 	}
223 
224 	// Allocate and initialize the LZ-based decoder. It will also give
225 	// us the dictionary size.
226 	lzma_lz_options lz_options;
227 	return_if_error(lz_init(&next->coder->lz, allocator,
228 			filters[0].options, &lz_options));
229 
230 	// If the dictionary size is very small, increase it to 4096 bytes.
231 	// This is to prevent constant wrapping of the dictionary, which
232 	// would slow things down. The downside is that since we don't check
233 	// separately for the real dictionary size, we may happily accept
234 	// corrupt files.
235 	if (lz_options.dict_size < 4096)
236 		lz_options.dict_size = 4096;
237 
238 	// Make dictionary size a multipe of 16. Some LZ-based decoders like
239 	// LZMA use the lowest bits lzma_dict.pos to know the alignment of the
240 	// data. Aligned buffer is also good when memcpying from the
241 	// dictionary to the output buffer, since applications are
242 	// recommended to give aligned buffers to liblzma.
243 	//
244 	// Avoid integer overflow.
245 	if (lz_options.dict_size > SIZE_MAX - 15)
246 		return LZMA_MEM_ERROR;
247 
248 	lz_options.dict_size = (lz_options.dict_size + 15) & ~((size_t)(15));
249 
250 	// Allocate and initialize the dictionary.
251 	if (next->coder->dict.size != lz_options.dict_size) {
252 		lzma_free(next->coder->dict.buf, allocator);
253 		next->coder->dict.buf
254 				= lzma_alloc(lz_options.dict_size, allocator);
255 		if (next->coder->dict.buf == NULL)
256 			return LZMA_MEM_ERROR;
257 
258 		next->coder->dict.size = lz_options.dict_size;
259 	}
260 
261 	lz_decoder_reset(next->coder);
262 
263 	// Use the preset dictionary if it was given to us.
264 	if (lz_options.preset_dict != NULL
265 			&& lz_options.preset_dict_size > 0) {
266 		// If the preset dictionary is bigger than the actual
267 		// dictionary, copy only the tail.
268 		const size_t copy_size = my_min(lz_options.preset_dict_size,
269 				lz_options.dict_size);
270 		const size_t offset = lz_options.preset_dict_size - copy_size;
271 		memcpy(next->coder->dict.buf, lz_options.preset_dict + offset,
272 				copy_size);
273 		next->coder->dict.pos = copy_size;
274 		next->coder->dict.full = copy_size;
275 	}
276 
277 	// Miscellaneous initializations
278 	next->coder->next_finished = false;
279 	next->coder->this_finished = false;
280 	next->coder->temp.pos = 0;
281 	next->coder->temp.size = 0;
282 
283 	// Initialize the next filter in the chain, if any.
284 	return lzma_next_filter_init(&next->coder->next, allocator,
285 			filters + 1);
286 }
287 
288 
289 extern uint64_t
290 lzma_lz_decoder_memusage(size_t dictionary_size)
291 {
292 	return sizeof(lzma_coder) + (uint64_t)(dictionary_size);
293 }
294 
295 
296 extern void
297 lzma_lz_decoder_uncompressed(lzma_coder *coder, lzma_vli uncompressed_size)
298 {
299 	coder->lz.set_uncompressed(coder->lz.coder, uncompressed_size);
300 }
301