Import xz-5.0.4.
[dragonfly.git] / contrib / xz / src / liblzma / simple / simple_coder.c
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                 const lzma_ret ret = coder->next.code(
39                                 coder->next.coder, allocator,
40                                 in, in_pos, in_size,
41                                 out, out_pos, out_size, action);
42
43                 if (ret == LZMA_STREAM_END) {
44                         assert(!coder->is_encoder
45                                         || action == LZMA_FINISH);
46                         coder->end_was_reached = true;
47
48                 } else if (ret != LZMA_OK) {
49                         return ret;
50                 }
51         }
52
53         return LZMA_OK;
54 }
55
56
57 static size_t
58 call_filter(lzma_coder *coder, uint8_t *buffer, size_t size)
59 {
60         const size_t filtered = coder->filter(coder->simple,
61                         coder->now_pos, coder->is_encoder,
62                         buffer, size);
63         coder->now_pos += filtered;
64         return filtered;
65 }
66
67
68 static lzma_ret
69 simple_code(lzma_coder *coder, lzma_allocator *allocator,
70                 const uint8_t *restrict in, size_t *restrict in_pos,
71                 size_t in_size, uint8_t *restrict out,
72                 size_t *restrict out_pos, size_t out_size, lzma_action action)
73 {
74         // TODO: Add partial support for LZMA_SYNC_FLUSH. We can support it
75         // in cases when the filter is able to filter everything. With most
76         // simple filters it can be done at offset that is a multiple of 2,
77         // 4, or 16. With x86 filter, it needs good luck, and thus cannot
78         // be made to work predictably.
79         if (action == LZMA_SYNC_FLUSH)
80                 return LZMA_OPTIONS_ERROR;
81
82         // Flush already filtered data from coder->buffer[] to out[].
83         if (coder->pos < coder->filtered) {
84                 lzma_bufcpy(coder->buffer, &coder->pos, coder->filtered,
85                                 out, out_pos, out_size);
86
87                 // If we couldn't flush all the filtered data, return to
88                 // application immediately.
89                 if (coder->pos < coder->filtered)
90                         return LZMA_OK;
91
92                 if (coder->end_was_reached) {
93                         assert(coder->filtered == coder->size);
94                         return LZMA_STREAM_END;
95                 }
96         }
97
98         // If we get here, there is no filtered data left in the buffer.
99         coder->filtered = 0;
100
101         assert(!coder->end_was_reached);
102
103         // If there is more output space left than there is unfiltered data
104         // in coder->buffer[], flush coder->buffer[] to out[], and copy/code
105         // more data to out[] hopefully filling it completely. Then filter
106         // the data in out[]. This step is where most of the data gets
107         // filtered if the buffer sizes used by the application are reasonable.
108         const size_t out_avail = out_size - *out_pos;
109         const size_t buf_avail = coder->size - coder->pos;
110         if (out_avail > buf_avail || buf_avail == 0) {
111                 // Store the old position so that we know from which byte
112                 // to start filtering.
113                 const size_t out_start = *out_pos;
114
115                 // Flush data from coder->buffer[] to out[], but don't reset
116                 // coder->pos and coder->size yet. This way the coder can be
117                 // restarted if the next filter in the chain returns e.g.
118                 // LZMA_MEM_ERROR.
119                 memcpy(out + *out_pos, coder->buffer + coder->pos, buf_avail);
120                 *out_pos += buf_avail;
121
122                 // Copy/Encode/Decode more data to out[].
123                 {
124                         const lzma_ret ret = copy_or_code(coder, allocator,
125                                         in, in_pos, in_size,
126                                         out, out_pos, out_size, action);
127                         assert(ret != LZMA_STREAM_END);
128                         if (ret != LZMA_OK)
129                                 return ret;
130                 }
131
132                 // Filter out[].
133                 const size_t size = *out_pos - out_start;
134                 const size_t filtered = call_filter(
135                                 coder, out + out_start, size);
136
137                 const size_t unfiltered = size - filtered;
138                 assert(unfiltered <= coder->allocated / 2);
139
140                 // Now we can update coder->pos and coder->size, because
141                 // the next coder in the chain (if any) was successful.
142                 coder->pos = 0;
143                 coder->size = unfiltered;
144
145                 if (coder->end_was_reached) {
146                         // The last byte has been copied to out[] already.
147                         // They are left as is.
148                         coder->size = 0;
149
150                 } else if (unfiltered > 0) {
151                         // There is unfiltered data left in out[]. Copy it to
152                         // coder->buffer[] and rewind *out_pos appropriately.
153                         *out_pos -= unfiltered;
154                         memcpy(coder->buffer, out + *out_pos, unfiltered);
155                 }
156         } else if (coder->pos > 0) {
157                 memmove(coder->buffer, coder->buffer + coder->pos, buf_avail);
158                 coder->size -= coder->pos;
159                 coder->pos = 0;
160         }
161
162         assert(coder->pos == 0);
163
164         // If coder->buffer[] isn't empty, try to fill it by copying/decoding
165         // more data. Then filter coder->buffer[] and copy the successfully
166         // filtered data to out[]. It is probable, that some filtered and
167         // unfiltered data will be left to coder->buffer[].
168         if (coder->size > 0) {
169                 {
170                         const lzma_ret ret = copy_or_code(coder, allocator,
171                                         in, in_pos, in_size,
172                                         coder->buffer, &coder->size,
173                                         coder->allocated, action);
174                         assert(ret != LZMA_STREAM_END);
175                         if (ret != LZMA_OK)
176                                 return ret;
177                 }
178
179                 coder->filtered = call_filter(
180                                 coder, coder->buffer, coder->size);
181
182                 // Everything is considered to be filtered if coder->buffer[]
183                 // contains the last bytes of the data.
184                 if (coder->end_was_reached)
185                         coder->filtered = coder->size;
186
187                 // Flush as much as possible.
188                 lzma_bufcpy(coder->buffer, &coder->pos, coder->filtered,
189                                 out, out_pos, out_size);
190         }
191
192         // Check if we got everything done.
193         if (coder->end_was_reached && coder->pos == coder->size)
194                 return LZMA_STREAM_END;
195
196         return LZMA_OK;
197 }
198
199
200 static void
201 simple_coder_end(lzma_coder *coder, lzma_allocator *allocator)
202 {
203         lzma_next_end(&coder->next, allocator);
204         lzma_free(coder->simple, allocator);
205         lzma_free(coder, allocator);
206         return;
207 }
208
209
210 static lzma_ret
211 simple_coder_update(lzma_coder *coder, lzma_allocator *allocator,
212                 const lzma_filter *filters_null lzma_attribute((__unused__)),
213                 const lzma_filter *reversed_filters)
214 {
215         // No update support, just call the next filter in the chain.
216         return lzma_next_filter_update(
217                         &coder->next, allocator, reversed_filters + 1);
218 }
219
220
221 extern lzma_ret
222 lzma_simple_coder_init(lzma_next_coder *next, lzma_allocator *allocator,
223                 const lzma_filter_info *filters,
224                 size_t (*filter)(lzma_simple *simple, uint32_t now_pos,
225                         bool is_encoder, uint8_t *buffer, size_t size),
226                 size_t simple_size, size_t unfiltered_max,
227                 uint32_t alignment, bool is_encoder)
228 {
229         // Allocate memory for the lzma_coder structure if needed.
230         if (next->coder == NULL) {
231                 // Here we allocate space also for the temporary buffer. We
232                 // need twice the size of unfiltered_max, because then it
233                 // is always possible to filter at least unfiltered_max bytes
234                 // more data in coder->buffer[] if it can be filled completely.
235                 next->coder = lzma_alloc(sizeof(lzma_coder)
236                                 + 2 * unfiltered_max, allocator);
237                 if (next->coder == NULL)
238                         return LZMA_MEM_ERROR;
239
240                 next->code = &simple_code;
241                 next->end = &simple_coder_end;
242                 next->update = &simple_coder_update;
243
244                 next->coder->next = LZMA_NEXT_CODER_INIT;
245                 next->coder->filter = filter;
246                 next->coder->allocated = 2 * unfiltered_max;
247
248                 // Allocate memory for filter-specific data structure.
249                 if (simple_size > 0) {
250                         next->coder->simple = lzma_alloc(
251                                         simple_size, allocator);
252                         if (next->coder->simple == NULL)
253                                 return LZMA_MEM_ERROR;
254                 } else {
255                         next->coder->simple = NULL;
256                 }
257         }
258
259         if (filters[0].options != NULL) {
260                 const lzma_options_bcj *simple = filters[0].options;
261                 next->coder->now_pos = simple->start_offset;
262                 if (next->coder->now_pos & (alignment - 1))
263                         return LZMA_OPTIONS_ERROR;
264         } else {
265                 next->coder->now_pos = 0;
266         }
267
268         // Reset variables.
269         next->coder->is_encoder = is_encoder;
270         next->coder->end_was_reached = false;
271         next->coder->pos = 0;
272         next->coder->filtered = 0;
273         next->coder->size = 0;
274
275         return lzma_next_filter_init(
276                         &next->coder->next, allocator, filters + 1);
277 }