| 1 | /* |
| 2 | Copyright (C) Andrew Tridgell 1996 |
| 3 | Copyright (C) Paul Mackerras 1996 |
| 4 | |
| 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. |
| 9 | |
| 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. |
| 14 | |
| 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; |
| 24 | extern int do_progress; |
| 25 | |
| 26 | typedef unsigned short tag; |
| 27 | |
| 28 | #define TABLESIZE (1<<16) |
| 29 | #define NULL_TAG ((size_t)-1) |
| 30 | |
| 31 | static int false_alarms; |
| 32 | static int tag_hits; |
| 33 | static int matches; |
| 34 | static int64 data_transfer; |
| 35 | |
| 36 | static int total_false_alarms; |
| 37 | static int total_tag_hits; |
| 38 | static int total_matches; |
| 39 | |
| 40 | extern struct stats stats; |
| 41 | |
| 42 | struct target { |
| 43 | tag t; |
| 44 | size_t i; |
| 45 | }; |
| 46 | |
| 47 | static struct target *targets; |
| 48 | |
| 49 | static size_t *tag_table; |
| 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 | { |
| 56 | return (int)t1->t - (int)t2->t; |
| 57 | } |
| 58 | |
| 59 | |
| 60 | static void build_hash_table(struct sum_struct *s) |
| 61 | { |
| 62 | size_t i; |
| 63 | |
| 64 | if (!tag_table) |
| 65 | tag_table = new_array(size_t, TABLESIZE); |
| 66 | |
| 67 | targets = new_array(struct target, s->count); |
| 68 | if (!tag_table || !targets) |
| 69 | out_of_memory("build_hash_table"); |
| 70 | |
| 71 | for (i = 0; i < s->count; i++) { |
| 72 | targets[i].i = i; |
| 73 | targets[i].t = gettag(s->sums[i].sum1); |
| 74 | } |
| 75 | |
| 76 | qsort(targets,s->count,sizeof(targets[0]),(int (*)())compare_targets); |
| 77 | |
| 78 | for (i = 0; i < TABLESIZE; i++) |
| 79 | tag_table[i] = NULL_TAG; |
| 80 | |
| 81 | for (i = s->count; i-- > 0; ) |
| 82 | tag_table[targets[i].t] = i; |
| 83 | } |
| 84 | |
| 85 | |
| 86 | static OFF_T last_match; |
| 87 | |
| 88 | |
| 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 | **/ |
| 101 | static void matched(int f,struct sum_struct *s,struct map_struct *buf, |
| 102 | OFF_T offset,int i) |
| 103 | { |
| 104 | OFF_T n = offset - last_match; |
| 105 | OFF_T j; |
| 106 | |
| 107 | if (verbose > 2 && i >= 0) |
| 108 | rprintf(FINFO,"match at %.0f last_match=%.0f j=%d len=%u n=%.0f\n", |
| 109 | (double)offset,(double)last_match,i,s->sums[i].len,(double)n); |
| 110 | |
| 111 | send_token(f,i,buf,last_match,n,i<0?0:s->sums[i].len); |
| 112 | data_transfer += n; |
| 113 | |
| 114 | if (i >= 0) { |
| 115 | stats.matched_data += s->sums[i].len; |
| 116 | n += s->sums[i].len; |
| 117 | } |
| 118 | |
| 119 | for (j = 0; j < n; j += CHUNK_SIZE) { |
| 120 | int n1 = MIN(CHUNK_SIZE,n-j); |
| 121 | sum_update(map_ptr(buf,last_match+j,n1),n1); |
| 122 | } |
| 123 | |
| 124 | |
| 125 | if (i >= 0) |
| 126 | last_match = offset + s->sums[i].len; |
| 127 | else |
| 128 | last_match = offset; |
| 129 | |
| 130 | if (buf && do_progress) { |
| 131 | show_progress(last_match, buf->file_size); |
| 132 | |
| 133 | if (i == -1) |
| 134 | end_progress(buf->file_size); |
| 135 | } |
| 136 | } |
| 137 | |
| 138 | |
| 139 | static void hash_search(int f,struct sum_struct *s, |
| 140 | struct map_struct *buf, OFF_T len) |
| 141 | { |
| 142 | OFF_T offset, end; |
| 143 | unsigned int k; |
| 144 | size_t last_i; |
| 145 | char sum2[SUM_LENGTH]; |
| 146 | uint32 s1, s2, sum; |
| 147 | schar *map; |
| 148 | |
| 149 | /* last_i is used to encourage adjacent matches, allowing the RLL coding of the |
| 150 | output to work more efficiently */ |
| 151 | last_i = (size_t)-1; |
| 152 | |
| 153 | if (verbose > 2) { |
| 154 | rprintf(FINFO,"hash search b=%u len=%.0f\n", |
| 155 | s->blength, (double)len); |
| 156 | } |
| 157 | |
| 158 | k = MIN(len, s->blength); |
| 159 | |
| 160 | map = (schar *)map_ptr(buf, 0, k); |
| 161 | |
| 162 | sum = get_checksum1((char *)map, k); |
| 163 | s1 = sum & 0xFFFF; |
| 164 | s2 = sum >> 16; |
| 165 | if (verbose > 3) |
| 166 | rprintf(FINFO, "sum=%.8x k=%u\n", sum, k); |
| 167 | |
| 168 | offset = 0; |
| 169 | |
| 170 | end = len + 1 - s->sums[s->count-1].len; |
| 171 | |
| 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 | } |
| 176 | |
| 177 | do { |
| 178 | tag t = gettag2(s1,s2); |
| 179 | int done_csum2 = 0; |
| 180 | size_t j = tag_table[t]; |
| 181 | |
| 182 | if (verbose > 4) |
| 183 | rprintf(FINFO,"offset=%.0f sum=%08x\n",(double)offset,sum); |
| 184 | |
| 185 | if (j == NULL_TAG) |
| 186 | goto null_tag; |
| 187 | |
| 188 | sum = (s1 & 0xffff) | (s2 << 16); |
| 189 | tag_hits++; |
| 190 | for (; j < s->count && targets[j].t == t; j++) { |
| 191 | unsigned int l; |
| 192 | size_t i = targets[j].i; |
| 193 | |
| 194 | if (sum != s->sums[i].sum1) |
| 195 | continue; |
| 196 | |
| 197 | /* also make sure the two blocks are the same length */ |
| 198 | l = MIN((OFF_T)s->blength, len-offset); |
| 199 | if (l != s->sums[i].len) |
| 200 | continue; |
| 201 | |
| 202 | if (verbose > 3) |
| 203 | rprintf(FINFO,"potential match at %.0f target=%.0f %.0f sum=%08x\n", |
| 204 | (double)offset,(double)j,(double)i,sum); |
| 205 | |
| 206 | if (!done_csum2) { |
| 207 | map = (schar *)map_ptr(buf,offset,l); |
| 208 | get_checksum2((char *)map,l,sum2); |
| 209 | done_csum2 = 1; |
| 210 | } |
| 211 | |
| 212 | if (memcmp(sum2,s->sums[i].sum2,s->s2length) != 0) { |
| 213 | false_alarms++; |
| 214 | continue; |
| 215 | } |
| 216 | |
| 217 | /* we've found a match, but now check to see |
| 218 | * if last_i can hint at a better match */ |
| 219 | for (j++; j < s->count && targets[j].t == t; j++) { |
| 220 | size_t i2 = targets[j].i; |
| 221 | if (i2 == last_i + 1) { |
| 222 | if (sum != s->sums[i2].sum1) |
| 223 | break; |
| 224 | if (memcmp(sum2,s->sums[i2].sum2,s->s2length) != 0) |
| 225 | break; |
| 226 | /* we've found an adjacent match - the RLL coder |
| 227 | * will be happy */ |
| 228 | i = i2; |
| 229 | break; |
| 230 | } |
| 231 | } |
| 232 | |
| 233 | last_i = i; |
| 234 | |
| 235 | matched(f,s,buf,offset,i); |
| 236 | offset += s->sums[i].len - 1; |
| 237 | k = MIN(s->blength, len-offset); |
| 238 | map = (schar *)map_ptr(buf, offset, k); |
| 239 | sum = get_checksum1((char *)map, k); |
| 240 | s1 = sum & 0xFFFF; |
| 241 | s2 = sum >> 16; |
| 242 | matches++; |
| 243 | break; |
| 244 | } |
| 245 | |
| 246 | null_tag: |
| 247 | /* Trim off the first byte from the checksum */ |
| 248 | map = (schar *)map_ptr(buf, offset, k+1); |
| 249 | s1 -= map[0] + CHAR_OFFSET; |
| 250 | s2 -= k * (map[0]+CHAR_OFFSET); |
| 251 | |
| 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; |
| 256 | } else |
| 257 | --k; |
| 258 | |
| 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. */ |
| 265 | if (offset > last_match |
| 266 | && offset-last_match >= CHUNK_SIZE+s->blength |
| 267 | && end-offset > CHUNK_SIZE) { |
| 268 | matched(f,s,buf,offset - s->blength, -2); |
| 269 | } |
| 270 | } while (++offset < end); |
| 271 | |
| 272 | matched(f,s,buf,len,-1); |
| 273 | map_ptr(buf,len-1,1); |
| 274 | } |
| 275 | |
| 276 | |
| 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 | * |
| 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 | * |
| 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) |
| 292 | { |
| 293 | char file_sum[MD4_SUM_LENGTH]; |
| 294 | extern int write_batch; |
| 295 | |
| 296 | last_match = 0; |
| 297 | false_alarms = 0; |
| 298 | tag_hits = 0; |
| 299 | matches = 0; |
| 300 | data_transfer = 0; |
| 301 | |
| 302 | sum_init(); |
| 303 | |
| 304 | if (len > 0 && s->count>0) { |
| 305 | build_hash_table(s); |
| 306 | |
| 307 | if (verbose > 2) |
| 308 | rprintf(FINFO,"built hash table\n"); |
| 309 | |
| 310 | hash_search(f,s,buf,len); |
| 311 | |
| 312 | if (verbose > 2) |
| 313 | rprintf(FINFO,"done hash search\n"); |
| 314 | } else { |
| 315 | OFF_T j; |
| 316 | /* by doing this in pieces we avoid too many seeks */ |
| 317 | for (j = 0; j < len-CHUNK_SIZE; j += CHUNK_SIZE) { |
| 318 | int n1 = MIN(CHUNK_SIZE,(len-CHUNK_SIZE)-j); |
| 319 | matched(f,s,buf,j+n1,-2); |
| 320 | } |
| 321 | matched(f,s,buf,len,-1); |
| 322 | } |
| 323 | |
| 324 | sum_end(file_sum); |
| 325 | |
| 326 | if (verbose > 2) |
| 327 | rprintf(FINFO,"sending file_sum\n"); |
| 328 | write_buf(f,file_sum,MD4_SUM_LENGTH); |
| 329 | if (write_batch) |
| 330 | write_batch_delta_file(file_sum, MD4_SUM_LENGTH); |
| 331 | |
| 332 | if (targets) { |
| 333 | free(targets); |
| 334 | targets=NULL; |
| 335 | } |
| 336 | |
| 337 | if (verbose > 2) |
| 338 | rprintf(FINFO, "false_alarms=%d tag_hits=%d matches=%d\n", |
| 339 | false_alarms, tag_hits, matches); |
| 340 | |
| 341 | total_tag_hits += tag_hits; |
| 342 | total_false_alarms += false_alarms; |
| 343 | total_matches += matches; |
| 344 | stats.literal_data += data_transfer; |
| 345 | } |
| 346 | |
| 347 | void match_report(void) |
| 348 | { |
| 349 | if (verbose <= 1) |
| 350 | return; |
| 351 | |
| 352 | rprintf(FINFO, |
| 353 | "total: matches=%d tag_hits=%d false_alarms=%d data=%.0f\n", |
| 354 | total_matches,total_tag_hits, |
| 355 | total_false_alarms, |
| 356 | (double)stats.literal_data); |
| 357 | } |