Commit | Line | Data |
---|---|---|
1e4f48d6 | 1 | /* |
c627d613 AT |
2 | Copyright (C) Andrew Tridgell 1996 |
3 | Copyright (C) Paul Mackerras 1996 | |
1e4f48d6 | 4 | |
c627d613 AT |
5 | This program is free software; you can redistribute it and/or modify |
6 | it under the terms of the GNU General Public License as published by | |
7 | the Free Software Foundation; either version 2 of the License, or | |
8 | (at your option) any later version. | |
1e4f48d6 | 9 | |
c627d613 AT |
10 | This program is distributed in the hope that it will be useful, |
11 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | GNU General Public License for more details. | |
1e4f48d6 | 14 | |
c627d613 AT |
15 | You should have received a copy of the GNU General Public License |
16 | along with this program; if not, write to the Free Software | |
17 | Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
18 | */ | |
19 | ||
20 | #include "rsync.h" | |
21 | ||
22 | extern int verbose; | |
23 | extern int am_server; | |
16417f8b | 24 | extern int do_progress; |
c627d613 AT |
25 | |
26 | typedef unsigned short tag; | |
27 | ||
28 | #define TABLESIZE (1<<16) | |
0e36d9da | 29 | #define NULL_TAG ((size_t)-1) |
c627d613 AT |
30 | |
31 | static int false_alarms; | |
32 | static int tag_hits; | |
33 | static int matches; | |
a800434a | 34 | static int64 data_transfer; |
c627d613 | 35 | |
3a6a366f AT |
36 | static int total_false_alarms; |
37 | static int total_tag_hits; | |
38 | static int total_matches; | |
c627d613 | 39 | |
a800434a | 40 | extern struct stats stats; |
c627d613 AT |
41 | |
42 | struct target { | |
1e4f48d6 | 43 | tag t; |
0e36d9da | 44 | size_t i; |
c627d613 AT |
45 | }; |
46 | ||
3a6a366f | 47 | static struct target *targets; |
c627d613 | 48 | |
0e36d9da | 49 | static size_t *tag_table; |
c627d613 AT |
50 | |
51 | #define gettag2(s1,s2) (((s1) + (s2)) & 0xFFFF) | |
52 | #define gettag(sum) gettag2((sum)&0xFFFF,(sum)>>16) | |
53 | ||
54 | static int compare_targets(struct target *t1,struct target *t2) | |
55 | { | |
1e4f48d6 | 56 | return (int)t1->t - (int)t2->t; |
c627d613 AT |
57 | } |
58 | ||
59 | ||
60 | static void build_hash_table(struct sum_struct *s) | |
61 | { | |
0e36d9da | 62 | size_t i; |
c627d613 | 63 | |
1e4f48d6 | 64 | if (!tag_table) |
0e36d9da | 65 | tag_table = new_array(size_t, TABLESIZE); |
c627d613 | 66 | |
58cadc86 | 67 | targets = new_array(struct target, s->count); |
1e4f48d6 WD |
68 | if (!tag_table || !targets) |
69 | out_of_memory("build_hash_table"); | |
c627d613 | 70 | |
0e36d9da | 71 | for (i = 0; i < s->count; i++) { |
1e4f48d6 WD |
72 | targets[i].i = i; |
73 | targets[i].t = gettag(s->sums[i].sum1); | |
74 | } | |
c627d613 | 75 | |
1e4f48d6 | 76 | qsort(targets,s->count,sizeof(targets[0]),(int (*)())compare_targets); |
c627d613 | 77 | |
1e4f48d6 WD |
78 | for (i = 0; i < TABLESIZE; i++) |
79 | tag_table[i] = NULL_TAG; | |
c627d613 | 80 | |
0e36d9da | 81 | for (i = s->count; i-- > 0; ) |
1e4f48d6 | 82 | tag_table[targets[i].t] = i; |
c627d613 AT |
83 | } |
84 | ||
85 | ||
bcacc18b | 86 | static OFF_T last_match; |
c627d613 AT |
87 | |
88 | ||
f5f95a38 MP |
89 | /** |
90 | * Transmit a literal and/or match token. | |
91 | * | |
92 | * This delightfully-named function is called either when we find a | |
93 | * match and need to transmit all the unmatched data leading up to it, | |
94 | * or when we get bored of accumulating literal data and just need to | |
95 | * transmit it. As a result of this second case, it is called even if | |
96 | * we have not matched at all! | |
97 | * | |
98 | * @param i If >0, the number of a matched token. If 0, indicates we | |
99 | * have only literal data. | |
100 | **/ | |
c6e7fcb4 | 101 | static void matched(int f,struct sum_struct *s,struct map_struct *buf, |
bcacc18b | 102 | OFF_T offset,int i) |
c627d613 | 103 | { |
bcacc18b | 104 | OFF_T n = offset - last_match; |
1309d90d | 105 | OFF_T j; |
9e31c482 | 106 | |
0d0e2e93 | 107 | if (verbose > 2 && i >= 0) |
da9d12f5 | 108 | rprintf(FINFO,"match at %.0f last_match=%.0f j=%d len=%u n=%.0f\n", |
5f808dfb | 109 | (double)offset,(double)last_match,i,s->sums[i].len,(double)n); |
c627d613 | 110 | |
45f133b9 | 111 | send_token(f,i,buf,last_match,n,i<0?0:s->sums[i].len); |
e7ebc36c | 112 | data_transfer += n; |
70d794dc | 113 | |
a800434a AT |
114 | if (i >= 0) { |
115 | stats.matched_data += s->sums[i].len; | |
e7ebc36c | 116 | n += s->sums[i].len; |
a800434a | 117 | } |
1e4f48d6 WD |
118 | |
119 | for (j = 0; j < n; j += CHUNK_SIZE) { | |
e7ebc36c AT |
120 | int n1 = MIN(CHUNK_SIZE,n-j); |
121 | sum_update(map_ptr(buf,last_match+j,n1),n1); | |
122 | } | |
9e31c482 | 123 | |
9e31c482 | 124 | |
0d0e2e93 | 125 | if (i >= 0) |
e7ebc36c | 126 | last_match = offset + s->sums[i].len; |
0d0e2e93 AT |
127 | else |
128 | last_match = offset; | |
eb86d661 | 129 | |
16417f8b | 130 | if (buf && do_progress) { |
4440b8aa | 131 | show_progress(last_match, buf->file_size); |
eb86d661 | 132 | |
0e36d9da WD |
133 | if (i == -1) |
134 | end_progress(buf->file_size); | |
5e12ce11 | 135 | } |
9e31c482 AT |
136 | } |
137 | ||
138 | ||
c6e7fcb4 | 139 | static void hash_search(int f,struct sum_struct *s, |
da9d12f5 | 140 | struct map_struct *buf, OFF_T len) |
c627d613 | 141 | { |
5f808dfb | 142 | OFF_T offset, end; |
0e36d9da WD |
143 | unsigned int k; |
144 | size_t last_i; | |
e7ebc36c | 145 | char sum2[SUM_LENGTH]; |
1e4f48d6 | 146 | uint32 s1, s2, sum; |
debb4505 | 147 | schar *map; |
e7ebc36c | 148 | |
923fa978 AT |
149 | /* last_i is used to encourage adjacent matches, allowing the RLL coding of the |
150 | output to work more efficiently */ | |
0e36d9da | 151 | last_i = (size_t)-1; |
923fa978 | 152 | |
0e36d9da WD |
153 | if (verbose > 2) { |
154 | rprintf(FINFO,"hash search b=%u len=%.0f\n", | |
155 | s->blength, (double)len); | |
156 | } | |
e7ebc36c | 157 | |
0e36d9da | 158 | k = MIN(len, s->blength); |
1e4f48d6 | 159 | |
0e36d9da | 160 | map = (schar *)map_ptr(buf, 0, k); |
1e4f48d6 | 161 | |
e7ebc36c AT |
162 | sum = get_checksum1((char *)map, k); |
163 | s1 = sum & 0xFFFF; | |
164 | s2 = sum >> 16; | |
165 | if (verbose > 3) | |
0e36d9da | 166 | rprintf(FINFO, "sum=%.8x k=%u\n", sum, k); |
1e4f48d6 | 167 | |
e7ebc36c | 168 | offset = 0; |
1e4f48d6 | 169 | |
e7ebc36c | 170 | end = len + 1 - s->sums[s->count-1].len; |
1e4f48d6 | 171 | |
0e36d9da WD |
172 | if (verbose > 3) { |
173 | rprintf(FINFO, "hash search s->blength=%u len=%.0f count=%.0f\n", | |
174 | s->blength, (double)len, (double)s->count); | |
175 | } | |
1e4f48d6 | 176 | |
e7ebc36c AT |
177 | do { |
178 | tag t = gettag2(s1,s2); | |
179 | int done_csum2 = 0; | |
0e36d9da | 180 | size_t j = tag_table[t]; |
1e4f48d6 | 181 | |
e7ebc36c | 182 | if (verbose > 4) |
5f808dfb | 183 | rprintf(FINFO,"offset=%.0f sum=%08x\n",(double)offset,sum); |
1e4f48d6 | 184 | |
0e36d9da | 185 | if (j == NULL_TAG) |
e7ebc36c | 186 | goto null_tag; |
e7ebc36c AT |
187 | |
188 | sum = (s1 & 0xffff) | (s2 << 16); | |
189 | tag_hits++; | |
0e36d9da | 190 | for (; j < s->count && targets[j].t == t; j++) { |
da9d12f5 | 191 | unsigned int l; |
0e36d9da | 192 | size_t i = targets[j].i; |
1e4f48d6 | 193 | |
0e36d9da WD |
194 | if (sum != s->sums[i].sum1) |
195 | continue; | |
1e4f48d6 | 196 | |
536541d5 | 197 | /* also make sure the two blocks are the same length */ |
c95dcb39 | 198 | l = MIN((OFF_T)s->blength, len-offset); |
0e36d9da WD |
199 | if (l != s->sums[i].len) |
200 | continue; | |
536541d5 | 201 | |
e7ebc36c | 202 | if (verbose > 3) |
0e36d9da WD |
203 | rprintf(FINFO,"potential match at %.0f target=%.0f %.0f sum=%08x\n", |
204 | (double)offset,(double)j,(double)i,sum); | |
1e4f48d6 | 205 | |
e7ebc36c | 206 | if (!done_csum2) { |
debb4505 | 207 | map = (schar *)map_ptr(buf,offset,l); |
e7ebc36c AT |
208 | get_checksum2((char *)map,l,sum2); |
209 | done_csum2 = 1; | |
210 | } | |
1e4f48d6 | 211 | |
fc0257c9 | 212 | if (memcmp(sum2,s->sums[i].sum2,s->s2length) != 0) { |
e7ebc36c AT |
213 | false_alarms++; |
214 | continue; | |
215 | } | |
923fa978 AT |
216 | |
217 | /* we've found a match, but now check to see | |
1e4f48d6 | 218 | * if last_i can hint at a better match */ |
0e36d9da WD |
219 | for (j++; j < s->count && targets[j].t == t; j++) { |
220 | size_t i2 = targets[j].i; | |
923fa978 | 221 | if (i2 == last_i + 1) { |
0e36d9da WD |
222 | if (sum != s->sums[i2].sum1) |
223 | break; | |
224 | if (memcmp(sum2,s->sums[i2].sum2,s->s2length) != 0) | |
225 | break; | |
1e4f48d6 WD |
226 | /* we've found an adjacent match - the RLL coder |
227 | * will be happy */ | |
923fa978 AT |
228 | i = i2; |
229 | break; | |
230 | } | |
231 | } | |
232 | ||
233 | last_i = i; | |
1e4f48d6 | 234 | |
e7ebc36c AT |
235 | matched(f,s,buf,offset,i); |
236 | offset += s->sums[i].len - 1; | |
0e36d9da WD |
237 | k = MIN(s->blength, len-offset); |
238 | map = (schar *)map_ptr(buf, offset, k); | |
e7ebc36c AT |
239 | sum = get_checksum1((char *)map, k); |
240 | s1 = sum & 0xFFFF; | |
241 | s2 = sum >> 16; | |
242 | matches++; | |
243 | break; | |
244 | } | |
1e4f48d6 | 245 | |
e7ebc36c AT |
246 | null_tag: |
247 | /* Trim off the first byte from the checksum */ | |
0e36d9da | 248 | map = (schar *)map_ptr(buf, offset, k+1); |
e7ebc36c AT |
249 | s1 -= map[0] + CHAR_OFFSET; |
250 | s2 -= k * (map[0]+CHAR_OFFSET); | |
1e4f48d6 | 251 | |
e7ebc36c AT |
252 | /* Add on the next byte (if there is one) to the checksum */ |
253 | if (k < (len-offset)) { | |
254 | s1 += (map[k]+CHAR_OFFSET); | |
255 | s2 += s1; | |
0e36d9da | 256 | } else |
e7ebc36c | 257 | --k; |
45f133b9 | 258 | |
5914bf15 PM |
259 | /* By matching early we avoid re-reading the |
260 | data 3 times in the case where a token | |
261 | match comes a long way after last | |
262 | match. The 3 reads are caused by the | |
263 | running match, the checksum update and the | |
264 | literal send. */ | |
1e4f48d6 WD |
265 | if (offset > last_match |
266 | && offset-last_match >= CHUNK_SIZE+s->blength | |
267 | && end-offset > CHUNK_SIZE) { | |
fc0257c9 | 268 | matched(f,s,buf,offset - s->blength, -2); |
45f133b9 | 269 | } |
e7ebc36c | 270 | } while (++offset < end); |
1e4f48d6 | 271 | |
e7ebc36c AT |
272 | matched(f,s,buf,len,-1); |
273 | map_ptr(buf,len-1,1); | |
c627d613 AT |
274 | } |
275 | ||
276 | ||
f5f95a38 MP |
277 | /** |
278 | * Scan through a origin file, looking for sections that match | |
279 | * checksums from the generator, and transmit either literal or token | |
280 | * data. | |
281 | * | |
538ba24f MP |
282 | * Also calculates the MD4 checksum of the whole file, using the md |
283 | * accumulator. This is transmitted with the file as protection | |
284 | * against corruption on the wire. | |
285 | * | |
f5f95a38 MP |
286 | * @param s Checksums received from the generator. If <tt>s->count == |
287 | * 0</tt>, then there are actually no checksums for this file. | |
288 | * | |
289 | * @param len Length of the file to send. | |
290 | **/ | |
291 | void match_sums(int f, struct sum_struct *s, struct map_struct *buf, OFF_T len) | |
c627d613 | 292 | { |
e7ebc36c | 293 | char file_sum[MD4_SUM_LENGTH]; |
64c3523a | 294 | extern int write_batch; |
e7ebc36c AT |
295 | |
296 | last_match = 0; | |
297 | false_alarms = 0; | |
298 | tag_hits = 0; | |
1e4f48d6 WD |
299 | matches = 0; |
300 | data_transfer = 0; | |
e7ebc36c AT |
301 | |
302 | sum_init(); | |
303 | ||
304 | if (len > 0 && s->count>0) { | |
305 | build_hash_table(s); | |
1e4f48d6 WD |
306 | |
307 | if (verbose > 2) | |
9486289c | 308 | rprintf(FINFO,"built hash table\n"); |
1e4f48d6 | 309 | |
e7ebc36c | 310 | hash_search(f,s,buf,len); |
1e4f48d6 WD |
311 | |
312 | if (verbose > 2) | |
9486289c | 313 | rprintf(FINFO,"done hash search\n"); |
e7ebc36c | 314 | } else { |
eb86d661 AT |
315 | OFF_T j; |
316 | /* by doing this in pieces we avoid too many seeks */ | |
1e4f48d6 | 317 | for (j = 0; j < len-CHUNK_SIZE; j += CHUNK_SIZE) { |
eb86d661 AT |
318 | int n1 = MIN(CHUNK_SIZE,(len-CHUNK_SIZE)-j); |
319 | matched(f,s,buf,j+n1,-2); | |
320 | } | |
e7ebc36c AT |
321 | matched(f,s,buf,len,-1); |
322 | } | |
9e31c482 | 323 | |
e7ebc36c | 324 | sum_end(file_sum); |
c627d613 | 325 | |
bc63ae3f S |
326 | if (verbose > 2) |
327 | rprintf(FINFO,"sending file_sum\n"); | |
328 | write_buf(f,file_sum,MD4_SUM_LENGTH); | |
64c3523a | 329 | if (write_batch) |
bc63ae3f | 330 | write_batch_delta_file(file_sum, MD4_SUM_LENGTH); |
c627d613 | 331 | |
e7ebc36c AT |
332 | if (targets) { |
333 | free(targets); | |
334 | targets=NULL; | |
335 | } | |
1e4f48d6 | 336 | |
e7ebc36c | 337 | if (verbose > 2) |
9486289c | 338 | rprintf(FINFO, "false_alarms=%d tag_hits=%d matches=%d\n", |
e7ebc36c | 339 | false_alarms, tag_hits, matches); |
1e4f48d6 | 340 | |
e7ebc36c AT |
341 | total_tag_hits += tag_hits; |
342 | total_false_alarms += false_alarms; | |
343 | total_matches += matches; | |
a800434a | 344 | stats.literal_data += data_transfer; |
c627d613 AT |
345 | } |
346 | ||
347 | void match_report(void) | |
348 | { | |
e7ebc36c AT |
349 | if (verbose <= 1) |
350 | return; | |
c627d613 | 351 | |
9486289c | 352 | rprintf(FINFO, |
a800434a | 353 | "total: matches=%d tag_hits=%d false_alarms=%d data=%.0f\n", |
e7ebc36c | 354 | total_matches,total_tag_hits, |
a800434a AT |
355 | total_false_alarms, |
356 | (double)stats.literal_data); | |
c627d613 | 357 | } |