2 Copyright (C) Andrew Tridgell 1996
3 Copyright (C) Paul Mackerras 1996
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.
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.
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.
24 extern int do_progress;
25 extern int checksum_seed;
28 typedef unsigned short tag;
30 #define TABLESIZE (1<<16)
31 #define NULL_TAG ((size_t)-1)
33 static int false_alarms;
36 static int64 data_transfer;
38 static int total_false_alarms;
39 static int total_tag_hits;
40 static int total_matches;
42 extern struct stats stats;
49 static struct target *targets;
51 static size_t *tag_table;
53 #define gettag2(s1,s2) (((s1) + (s2)) & 0xFFFF)
54 #define gettag(sum) gettag2((sum)&0xFFFF,(sum)>>16)
56 static int compare_targets(struct target *t1,struct target *t2)
58 return (int)t1->t - (int)t2->t;
62 static void build_hash_table(struct sum_struct *s)
67 tag_table = new_array(size_t, TABLESIZE);
69 targets = new_array(struct target, s->count);
70 if (!tag_table || !targets)
71 out_of_memory("build_hash_table");
73 for (i = 0; i < s->count; i++) {
75 targets[i].t = gettag(s->sums[i].sum1);
78 qsort(targets,s->count,sizeof(targets[0]),(int (*)())compare_targets);
80 for (i = 0; i < TABLESIZE; i++)
81 tag_table[i] = NULL_TAG;
83 for (i = s->count; i-- > 0; )
84 tag_table[targets[i].t] = i;
88 static OFF_T last_match;
92 * Transmit a literal and/or match token.
94 * This delightfully-named function is called either when we find a
95 * match and need to transmit all the unmatched data leading up to it,
96 * or when we get bored of accumulating literal data and just need to
97 * transmit it. As a result of this second case, it is called even if
98 * we have not matched at all!
100 * @param i If >0, the number of a matched token. If 0, indicates we
101 * have only literal data.
103 static void matched(int f,struct sum_struct *s,struct map_struct *buf,
106 OFF_T n = offset - last_match;
109 if (verbose > 2 && i >= 0)
110 rprintf(FINFO,"match at %.0f last_match=%.0f j=%d len=%u n=%.0f\n",
111 (double)offset,(double)last_match,i,s->sums[i].len,(double)n);
113 send_token(f,i,buf,last_match,n,i<0?0:s->sums[i].len);
117 stats.matched_data += s->sums[i].len;
121 for (j = 0; j < n; j += CHUNK_SIZE) {
122 int n1 = MIN(CHUNK_SIZE,n-j);
123 sum_update(map_ptr(buf,last_match+j,n1),n1);
128 last_match = offset + s->sums[i].len;
132 if (buf && do_progress) {
133 show_progress(last_match, buf->file_size);
136 end_progress(buf->file_size);
141 static void hash_search(int f,struct sum_struct *s,
142 struct map_struct *buf, OFF_T len)
144 OFF_T offset, end, backup;
147 char sum2[SUM_LENGTH];
152 /* want_i is used to encourage adjacent matches, allowing the RLL
153 * coding of the output to work more efficiently. */
157 rprintf(FINFO,"hash search b=%u len=%.0f\n",
158 s->blength, (double)len);
161 k = MIN(len, s->blength);
163 map = (schar *)map_ptr(buf, 0, k);
165 sum = get_checksum1((char *)map, k);
169 rprintf(FINFO, "sum=%.8x k=%u\n", sum, k);
173 end = len + 1 - s->sums[s->count-1].len;
176 rprintf(FINFO, "hash search s->blength=%u len=%.0f count=%.0f\n",
177 s->blength, (double)len, (double)s->count);
181 tag t = gettag2(s1,s2);
183 size_t j = tag_table[t];
186 rprintf(FINFO,"offset=%.0f sum=%08x\n",(double)offset,sum);
191 sum = (s1 & 0xffff) | (s2 << 16);
195 size_t i = targets[j].i;
197 if (sum != s->sums[i].sum1)
200 /* also make sure the two blocks are the same length */
201 l = MIN((OFF_T)s->blength, len-offset);
202 if (l != s->sums[i].len)
205 /* inplace: ensure chunk's offset is either >= our
206 * offset or that the data didn't move. */
207 if (inplace && s->sums[i].offset < offset
208 && !(s->sums[i].flags & SUMFLG_SAME_OFFSET))
212 rprintf(FINFO,"potential match at %.0f target=%.0f %.0f sum=%08x\n",
213 (double)offset,(double)j,(double)i,sum);
216 map = (schar *)map_ptr(buf,offset,l);
217 get_checksum2((char *)map,l,sum2);
221 if (memcmp(sum2,s->sums[i].sum2,s->s2length) != 0) {
226 /* If inplace is enabled, the best possible match is
227 * one with an identical offset, so we prefer that over
228 * the following want_i optimization. */
231 size_t i2 = targets[j].i;
232 if (s->sums[i2].offset != offset)
235 if (sum != s->sums[i2].sum1)
237 if (memcmp(sum2, s->sums[i2].sum2,
242 /* This chunk was at the same offset on
243 * both the sender and the receiver. */
244 s->sums[i].flags |= SUMFLG_SAME_OFFSET;
246 } while (++j < s->count && targets[j].t == t);
249 /* we've found a match, but now check to see
250 * if want_i can hint at a better match. */
251 if (i != want_i && want_i < s->count
252 && (!inplace || s->sums[want_i].offset >= offset
253 || s->sums[want_i].flags & SUMFLG_SAME_OFFSET)
254 && sum == s->sums[want_i].sum1
255 && memcmp(sum2, s->sums[want_i].sum2, s->s2length) == 0) {
256 /* we've found an adjacent match - the RLL coder
263 matched(f,s,buf,offset,i);
264 offset += s->sums[i].len - 1;
265 k = MIN(s->blength, len-offset);
266 map = (schar *)map_ptr(buf, offset, k);
267 sum = get_checksum1((char *)map, k);
272 } while (++j < s->count && targets[j].t == t);
275 backup = offset - last_match;
276 /* We sometimes read 1 byte prior to last_match... */
280 /* Trim off the first byte from the checksum */
281 more = offset + k < len;
282 map = (schar *)map_ptr(buf, offset - backup, k + more + backup)
284 s1 -= map[0] + CHAR_OFFSET;
285 s2 -= k * (map[0]+CHAR_OFFSET);
287 /* Add on the next byte (if there is one) to the checksum */
289 s1 += map[k] + CHAR_OFFSET;
294 /* By matching early we avoid re-reading the
295 data 3 times in the case where a token
296 match comes a long way after last
297 match. The 3 reads are caused by the
298 running match, the checksum update and the
300 if (backup >= CHUNK_SIZE + s->blength
301 && end - offset > CHUNK_SIZE) {
302 matched(f,s,buf,offset - s->blength, -2);
304 } while (++offset < end);
306 matched(f,s,buf,len,-1);
307 map_ptr(buf,len-1,1);
312 * Scan through a origin file, looking for sections that match
313 * checksums from the generator, and transmit either literal or token
316 * Also calculates the MD4 checksum of the whole file, using the md
317 * accumulator. This is transmitted with the file as protection
318 * against corruption on the wire.
320 * @param s Checksums received from the generator. If <tt>s->count ==
321 * 0</tt>, then there are actually no checksums for this file.
323 * @param len Length of the file to send.
325 void match_sums(int f, struct sum_struct *s, struct map_struct *buf, OFF_T len)
327 char file_sum[MD4_SUM_LENGTH];
335 sum_init(checksum_seed);
337 if (len > 0 && s->count>0) {
341 rprintf(FINFO,"built hash table\n");
343 hash_search(f,s,buf,len);
346 rprintf(FINFO,"done hash search\n");
349 /* by doing this in pieces we avoid too many seeks */
350 for (j = 0; j < len-CHUNK_SIZE; j += CHUNK_SIZE) {
351 int n1 = MIN(CHUNK_SIZE,(len-CHUNK_SIZE)-j);
352 matched(f,s,buf,j+n1,-2);
354 matched(f,s,buf,len,-1);
358 /* If we had a read error, send a bad checksum. */
359 if (buf && buf->status != 0)
363 rprintf(FINFO,"sending file_sum\n");
364 write_buf(f,file_sum,MD4_SUM_LENGTH);
372 rprintf(FINFO, "false_alarms=%d tag_hits=%d matches=%d\n",
373 false_alarms, tag_hits, matches);
375 total_tag_hits += tag_hits;
376 total_false_alarms += false_alarms;
377 total_matches += matches;
378 stats.literal_data += data_transfer;
381 void match_report(void)
387 "total: matches=%d tag_hits=%d false_alarms=%d data=%.0f\n",
388 total_matches,total_tag_hits,
390 (double)stats.literal_data);