xref: /freebsd/contrib/xz/src/liblzma/simple/simple_coder.c (revision 7aa383846770374466b1dcb2cefd71bde9acf463)
1 ///////////////////////////////////////////////////////////////////////////////
2 //
3 /// \file       simple_coder.c
4 /// \brief      Wrapper for simple filters
5 ///
6 /// Simple filters don't change the size of the data i.e. number of bytes
7 /// in equals the number of bytes out.
8 //
9 //  Author:     Lasse Collin
10 //
11 //  This file has been put into the public domain.
12 //  You can do whatever you want with this file.
13 //
14 ///////////////////////////////////////////////////////////////////////////////
15 
16 #include "simple_private.h"
17 
18 
19 /// Copied or encodes/decodes more data to out[].
20 static lzma_ret
21 copy_or_code(lzma_coder *coder, lzma_allocator *allocator,
22 		const uint8_t *restrict in, size_t *restrict in_pos,
23 		size_t in_size, uint8_t *restrict out,
24 		size_t *restrict out_pos, size_t out_size, lzma_action action)
25 {
26 	assert(!coder->end_was_reached);
27 
28 	if (coder->next.code == NULL) {
29 		lzma_bufcpy(in, in_pos, in_size, out, out_pos, out_size);
30 
31 		// Check if end of stream was reached.
32 		if (coder->is_encoder && action == LZMA_FINISH
33 				&& *in_pos == in_size)
34 			coder->end_was_reached = true;
35 
36 	} else {
37 		// Call the next coder in the chain to provide us some data.
38 		// We don't care about uncompressed_size here, because
39 		// the next filter in the chain will do it for us (since
40 		// we don't change the size of the data).
41 		const lzma_ret ret = coder->next.code(
42 				coder->next.coder, allocator,
43 				in, in_pos, in_size,
44 				out, out_pos, out_size, action);
45 
46 		if (ret == LZMA_STREAM_END) {
47 			assert(!coder->is_encoder
48 					|| action == LZMA_FINISH);
49 			coder->end_was_reached = true;
50 
51 		} else if (ret != LZMA_OK) {
52 			return ret;
53 		}
54 	}
55 
56 	return LZMA_OK;
57 }
58 
59 
60 static size_t
61 call_filter(lzma_coder *coder, uint8_t *buffer, size_t size)
62 {
63 	const size_t filtered = coder->filter(coder->simple,
64 			coder->now_pos, coder->is_encoder,
65 			buffer, size);
66 	coder->now_pos += filtered;
67 	return filtered;
68 }
69 
70 
71 static lzma_ret
72 simple_code(lzma_coder *coder, lzma_allocator *allocator,
73 		const uint8_t *restrict in, size_t *restrict in_pos,
74 		size_t in_size, uint8_t *restrict out,
75 		size_t *restrict out_pos, size_t out_size, lzma_action action)
76 {
77 	// TODO: Add partial support for LZMA_SYNC_FLUSH. We can support it
78 	// in cases when the filter is able to filter everything. With most
79 	// simple filters it can be done at offset that is a multiple of 2,
80 	// 4, or 16. With x86 filter, it needs good luck, and thus cannot
81 	// be made to work predictably.
82 	if (action == LZMA_SYNC_FLUSH)
83 		return LZMA_OPTIONS_ERROR;
84 
85 	// Flush already filtered data from coder->buffer[] to out[].
86 	if (coder->pos < coder->filtered) {
87 		lzma_bufcpy(coder->buffer, &coder->pos, coder->filtered,
88 				out, out_pos, out_size);
89 
90 		// If we couldn't flush all the filtered data, return to
91 		// application immediately.
92 		if (coder->pos < coder->filtered)
93 			return LZMA_OK;
94 
95 		if (coder->end_was_reached) {
96 			assert(coder->filtered == coder->size);
97 			return LZMA_STREAM_END;
98 		}
99 	}
100 
101 	// If we get here, there is no filtered data left in the buffer.
102 	coder->filtered = 0;
103 
104 	assert(!coder->end_was_reached);
105 
106 	// If there is more output space left than there is unfiltered data
107 	// in coder->buffer[], flush coder->buffer[] to out[], and copy/code
108 	// more data to out[] hopefully filling it completely. Then filter
109 	// the data in out[]. This step is where most of the data gets
110 	// filtered if the buffer sizes used by the application are reasonable.
111 	const size_t out_avail = out_size - *out_pos;
112 	const size_t buf_avail = coder->size - coder->pos;
113 	if (out_avail > buf_avail) {
114 		// Store the old position so that we know from which byte
115 		// to start filtering.
116 		const size_t out_start = *out_pos;
117 
118 		// Flush data from coder->buffer[] to out[], but don't reset
119 		// coder->pos and coder->size yet. This way the coder can be
120 		// restarted if the next filter in the chain returns e.g.
121 		// LZMA_MEM_ERROR.
122 		memcpy(out + *out_pos, coder->buffer + coder->pos, buf_avail);
123 		*out_pos += buf_avail;
124 
125 		// Copy/Encode/Decode more data to out[].
126 		{
127 			const lzma_ret ret = copy_or_code(coder, allocator,
128 					in, in_pos, in_size,
129 					out, out_pos, out_size, action);
130 			assert(ret != LZMA_STREAM_END);
131 			if (ret != LZMA_OK)
132 				return ret;
133 		}
134 
135 		// Filter out[].
136 		const size_t size = *out_pos - out_start;
137 		const size_t filtered = call_filter(
138 				coder, out + out_start, size);
139 
140 		const size_t unfiltered = size - filtered;
141 		assert(unfiltered <= coder->allocated / 2);
142 
143 		// Now we can update coder->pos and coder->size, because
144 		// the next coder in the chain (if any) was successful.
145 		coder->pos = 0;
146 		coder->size = unfiltered;
147 
148 		if (coder->end_was_reached) {
149 			// The last byte has been copied to out[] already.
150 			// They are left as is.
151 			coder->size = 0;
152 
153 		} else if (unfiltered > 0) {
154 			// There is unfiltered data left in out[]. Copy it to
155 			// coder->buffer[] and rewind *out_pos appropriately.
156 			*out_pos -= unfiltered;
157 			memcpy(coder->buffer, out + *out_pos, unfiltered);
158 		}
159 	} else if (coder->pos > 0) {
160 		memmove(coder->buffer, coder->buffer + coder->pos, buf_avail);
161 		coder->size -= coder->pos;
162 		coder->pos = 0;
163 	}
164 
165 	assert(coder->pos == 0);
166 
167 	// If coder->buffer[] isn't empty, try to fill it by copying/decoding
168 	// more data. Then filter coder->buffer[] and copy the successfully
169 	// filtered data to out[]. It is probable, that some filtered and
170 	// unfiltered data will be left to coder->buffer[].
171 	if (coder->size > 0) {
172 		{
173 			const lzma_ret ret = copy_or_code(coder, allocator,
174 					in, in_pos, in_size,
175 					coder->buffer, &coder->size,
176 					coder->allocated, action);
177 			assert(ret != LZMA_STREAM_END);
178 			if (ret != LZMA_OK)
179 				return ret;
180 		}
181 
182 		coder->filtered = call_filter(
183 				coder, coder->buffer, coder->size);
184 
185 		// Everything is considered to be filtered if coder->buffer[]
186 		// contains the last bytes of the data.
187 		if (coder->end_was_reached)
188 			coder->filtered = coder->size;
189 
190 		// Flush as much as possible.
191 		lzma_bufcpy(coder->buffer, &coder->pos, coder->filtered,
192 				out, out_pos, out_size);
193 	}
194 
195 	// Check if we got everything done.
196 	if (coder->end_was_reached && coder->pos == coder->size)
197 		return LZMA_STREAM_END;
198 
199 	return LZMA_OK;
200 }
201 
202 
203 static void
204 simple_coder_end(lzma_coder *coder, lzma_allocator *allocator)
205 {
206 	lzma_next_end(&coder->next, allocator);
207 	lzma_free(coder->simple, allocator);
208 	lzma_free(coder, allocator);
209 	return;
210 }
211 
212 
213 static lzma_ret
214 simple_coder_update(lzma_coder *coder, lzma_allocator *allocator,
215 		const lzma_filter *filters_null lzma_attribute((unused)),
216 		const lzma_filter *reversed_filters)
217 {
218 	// No update support, just call the next filter in the chain.
219 	return lzma_next_filter_update(
220 			&coder->next, allocator, reversed_filters + 1);
221 }
222 
223 
224 extern lzma_ret
225 lzma_simple_coder_init(lzma_next_coder *next, lzma_allocator *allocator,
226 		const lzma_filter_info *filters,
227 		size_t (*filter)(lzma_simple *simple, uint32_t now_pos,
228 			bool is_encoder, uint8_t *buffer, size_t size),
229 		size_t simple_size, size_t unfiltered_max,
230 		uint32_t alignment, bool is_encoder)
231 {
232 	// Allocate memory for the lzma_coder structure if needed.
233 	if (next->coder == NULL) {
234 		// Here we allocate space also for the temporary buffer. We
235 		// need twice the size of unfiltered_max, because then it
236 		// is always possible to filter at least unfiltered_max bytes
237 		// more data in coder->buffer[] if it can be filled completely.
238 		next->coder = lzma_alloc(sizeof(lzma_coder)
239 				+ 2 * unfiltered_max, allocator);
240 		if (next->coder == NULL)
241 			return LZMA_MEM_ERROR;
242 
243 		next->code = &simple_code;
244 		next->end = &simple_coder_end;
245 		next->update = &simple_coder_update;
246 
247 		next->coder->next = LZMA_NEXT_CODER_INIT;
248 		next->coder->filter = filter;
249 		next->coder->allocated = 2 * unfiltered_max;
250 
251 		// Allocate memory for filter-specific data structure.
252 		if (simple_size > 0) {
253 			next->coder->simple = lzma_alloc(
254 					simple_size, allocator);
255 			if (next->coder->simple == NULL)
256 				return LZMA_MEM_ERROR;
257 		} else {
258 			next->coder->simple = NULL;
259 		}
260 	}
261 
262 	if (filters[0].options != NULL) {
263 		const lzma_options_bcj *simple = filters[0].options;
264 		next->coder->now_pos = simple->start_offset;
265 		if (next->coder->now_pos & (alignment - 1))
266 			return LZMA_OPTIONS_ERROR;
267 	} else {
268 		next->coder->now_pos = 0;
269 	}
270 
271 	// Reset variables.
272 	next->coder->is_encoder = is_encoder;
273 	next->coder->end_was_reached = false;
274 	next->coder->pos = 0;
275 	next->coder->filtered = 0;
276 	next->coder->size = 0;
277 
278 	return lzma_next_filter_init(
279 			&next->coder->next, allocator, filters + 1);
280 }
281