xref: /freebsd/contrib/xz/src/liblzma/lz/lz_decoder.c (revision af23369a6deaaeb612ab266eb88b8bb8d560c322)
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 typedef struct {
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 } lzma_coder;
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. Do it conditionally because out can be NULL
95 		// (in which case copy_size is always 0). Calling memcpy()
96 		// with a null-pointer is undefined even if the third
97 		// argument is 0.
98 		const size_t copy_size = coder->dict.pos - dict_start;
99 		assert(copy_size <= out_size - *out_pos);
100 
101 		if (copy_size > 0)
102 			memcpy(out + *out_pos, coder->dict.buf + dict_start,
103 					copy_size);
104 
105 		*out_pos += copy_size;
106 
107 		// Reset the dictionary if so requested by coder->lz.code().
108 		if (coder->dict.need_reset) {
109 			lz_decoder_reset(coder);
110 
111 			// Since we reset dictionary, we don't check if
112 			// dictionary became full.
113 			if (ret != LZMA_OK || *out_pos == out_size)
114 				return ret;
115 		} else {
116 			// Return if everything got decoded or an error
117 			// occurred, or if there's no more data to decode.
118 			//
119 			// Note that detecting if there's something to decode
120 			// is done by looking if dictionary become full
121 			// instead of looking if *in_pos == in_size. This
122 			// is because it is possible that all the input was
123 			// consumed already but some data is pending to be
124 			// written to the dictionary.
125 			if (ret != LZMA_OK || *out_pos == out_size
126 					|| coder->dict.pos < coder->dict.size)
127 				return ret;
128 		}
129 	}
130 }
131 
132 
133 static lzma_ret
134 lz_decode(void *coder_ptr, const lzma_allocator *allocator,
135 		const uint8_t *restrict in, size_t *restrict in_pos,
136 		size_t in_size, uint8_t *restrict out,
137 		size_t *restrict out_pos, size_t out_size,
138 		lzma_action action)
139 {
140 	lzma_coder *coder = coder_ptr;
141 
142 	if (coder->next.code == NULL)
143 		return decode_buffer(coder, in, in_pos, in_size,
144 				out, out_pos, out_size);
145 
146 	// We aren't the last coder in the chain, we need to decode
147 	// our input to a temporary buffer.
148 	while (*out_pos < out_size) {
149 		// Fill the temporary buffer if it is empty.
150 		if (!coder->next_finished
151 				&& coder->temp.pos == coder->temp.size) {
152 			coder->temp.pos = 0;
153 			coder->temp.size = 0;
154 
155 			const lzma_ret ret = coder->next.code(
156 					coder->next.coder,
157 					allocator, in, in_pos, in_size,
158 					coder->temp.buffer, &coder->temp.size,
159 					LZMA_BUFFER_SIZE, action);
160 
161 			if (ret == LZMA_STREAM_END)
162 				coder->next_finished = true;
163 			else if (ret != LZMA_OK || coder->temp.size == 0)
164 				return ret;
165 		}
166 
167 		if (coder->this_finished) {
168 			if (coder->temp.size != 0)
169 				return LZMA_DATA_ERROR;
170 
171 			if (coder->next_finished)
172 				return LZMA_STREAM_END;
173 
174 			return LZMA_OK;
175 		}
176 
177 		const lzma_ret ret = decode_buffer(coder, coder->temp.buffer,
178 				&coder->temp.pos, coder->temp.size,
179 				out, out_pos, out_size);
180 
181 		if (ret == LZMA_STREAM_END)
182 			coder->this_finished = true;
183 		else if (ret != LZMA_OK)
184 			return ret;
185 		else if (coder->next_finished && *out_pos < out_size)
186 			return LZMA_DATA_ERROR;
187 	}
188 
189 	return LZMA_OK;
190 }
191 
192 
193 static void
194 lz_decoder_end(void *coder_ptr, const lzma_allocator *allocator)
195 {
196 	lzma_coder *coder = coder_ptr;
197 
198 	lzma_next_end(&coder->next, allocator);
199 	lzma_free(coder->dict.buf, allocator);
200 
201 	if (coder->lz.end != NULL)
202 		coder->lz.end(coder->lz.coder, allocator);
203 	else
204 		lzma_free(coder->lz.coder, allocator);
205 
206 	lzma_free(coder, allocator);
207 	return;
208 }
209 
210 
211 extern lzma_ret
212 lzma_lz_decoder_init(lzma_next_coder *next, const lzma_allocator *allocator,
213 		const lzma_filter_info *filters,
214 		lzma_ret (*lz_init)(lzma_lz_decoder *lz,
215 			const lzma_allocator *allocator,
216 			lzma_vli id, const void *options,
217 			lzma_lz_options *lz_options))
218 {
219 	// Allocate the base structure if it isn't already allocated.
220 	lzma_coder *coder = next->coder;
221 	if (coder == NULL) {
222 		coder = lzma_alloc(sizeof(lzma_coder), allocator);
223 		if (coder == NULL)
224 			return LZMA_MEM_ERROR;
225 
226 		next->coder = coder;
227 		next->code = &lz_decode;
228 		next->end = &lz_decoder_end;
229 
230 		coder->dict.buf = NULL;
231 		coder->dict.size = 0;
232 		coder->lz = LZMA_LZ_DECODER_INIT;
233 		coder->next = LZMA_NEXT_CODER_INIT;
234 	}
235 
236 	// Allocate and initialize the LZ-based decoder. It will also give
237 	// us the dictionary size.
238 	lzma_lz_options lz_options;
239 	return_if_error(lz_init(&coder->lz, allocator,
240 			filters[0].id, filters[0].options, &lz_options));
241 
242 	// If the dictionary size is very small, increase it to 4096 bytes.
243 	// This is to prevent constant wrapping of the dictionary, which
244 	// would slow things down. The downside is that since we don't check
245 	// separately for the real dictionary size, we may happily accept
246 	// corrupt files.
247 	if (lz_options.dict_size < 4096)
248 		lz_options.dict_size = 4096;
249 
250 	// Make dictionary size a multiple of 16. Some LZ-based decoders like
251 	// LZMA use the lowest bits lzma_dict.pos to know the alignment of the
252 	// data. Aligned buffer is also good when memcpying from the
253 	// dictionary to the output buffer, since applications are
254 	// recommended to give aligned buffers to liblzma.
255 	//
256 	// Avoid integer overflow.
257 	if (lz_options.dict_size > SIZE_MAX - 15)
258 		return LZMA_MEM_ERROR;
259 
260 	lz_options.dict_size = (lz_options.dict_size + 15) & ~((size_t)(15));
261 
262 	// Allocate and initialize the dictionary.
263 	if (coder->dict.size != lz_options.dict_size) {
264 		lzma_free(coder->dict.buf, allocator);
265 		coder->dict.buf
266 				= lzma_alloc(lz_options.dict_size, allocator);
267 		if (coder->dict.buf == NULL)
268 			return LZMA_MEM_ERROR;
269 
270 		coder->dict.size = lz_options.dict_size;
271 	}
272 
273 	lz_decoder_reset(next->coder);
274 
275 	// Use the preset dictionary if it was given to us.
276 	if (lz_options.preset_dict != NULL
277 			&& lz_options.preset_dict_size > 0) {
278 		// If the preset dictionary is bigger than the actual
279 		// dictionary, copy only the tail.
280 		const size_t copy_size = my_min(lz_options.preset_dict_size,
281 				lz_options.dict_size);
282 		const size_t offset = lz_options.preset_dict_size - copy_size;
283 		memcpy(coder->dict.buf, lz_options.preset_dict + offset,
284 				copy_size);
285 		coder->dict.pos = copy_size;
286 		coder->dict.full = copy_size;
287 	}
288 
289 	// Miscellaneous initializations
290 	coder->next_finished = false;
291 	coder->this_finished = false;
292 	coder->temp.pos = 0;
293 	coder->temp.size = 0;
294 
295 	// Initialize the next filter in the chain, if any.
296 	return lzma_next_filter_init(&coder->next, allocator, filters + 1);
297 }
298 
299 
300 extern uint64_t
301 lzma_lz_decoder_memusage(size_t dictionary_size)
302 {
303 	return sizeof(lzma_coder) + (uint64_t)(dictionary_size);
304 }
305