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;
27 extern int make_backups;
29 typedef unsigned short tag;
31 #define TABLESIZE (1<<16)
32 #define NULL_TAG ((size_t)-1)
34 static int false_alarms;
37 static int64 data_transfer;
39 static int total_false_alarms;
40 static int total_tag_hits;
41 static int total_matches;
43 extern struct stats stats;
50 static struct target *targets;
52 static size_t *tag_table;
54 #define gettag2(s1,s2) (((s1) + (s2)) & 0xFFFF)
55 #define gettag(sum) gettag2((sum)&0xFFFF,(sum)>>16)
57 static int compare_targets(struct target *t1,struct target *t2)
59 return (int)t1->t - (int)t2->t;
63 static void build_hash_table(struct sum_struct *s)
68 tag_table = new_array(size_t, TABLESIZE);
70 targets = new_array(struct target, s->count);
71 if (!tag_table || !targets)
72 out_of_memory("build_hash_table");
74 for (i = 0; i < s->count; i++) {
76 targets[i].t = gettag(s->sums[i].sum1);
79 qsort(targets,s->count,sizeof(targets[0]),(int (*)())compare_targets);
81 for (i = 0; i < TABLESIZE; i++)
82 tag_table[i] = NULL_TAG;
84 for (i = s->count; i-- > 0; )
85 tag_table[targets[i].t] = i;
89 static OFF_T last_match;
93 * Transmit a literal and/or match token.
95 * This delightfully-named function is called either when we find a
96 * match and need to transmit all the unmatched data leading up to it,
97 * or when we get bored of accumulating literal data and just need to
98 * transmit it. As a result of this second case, it is called even if
99 * we have not matched at all!
101 * @param i If >0, the number of a matched token. If 0, indicates we
102 * have only literal data.
104 static void matched(int f,struct sum_struct *s,struct map_struct *buf,
107 OFF_T n = offset - last_match;
110 if (verbose > 2 && i >= 0)
111 rprintf(FINFO,"match at %.0f last_match=%.0f j=%d len=%u n=%.0f\n",
112 (double)offset,(double)last_match,i,s->sums[i].len,(double)n);
114 send_token(f,i,buf,last_match,n,i<0?0:s->sums[i].len);
118 stats.matched_data += s->sums[i].len;
122 for (j = 0; j < n; j += CHUNK_SIZE) {
123 int n1 = MIN(CHUNK_SIZE,n-j);
124 sum_update(map_ptr(buf,last_match+j,n1),n1);
129 last_match = offset + s->sums[i].len;
133 if (buf && do_progress) {
134 show_progress(last_match, buf->file_size);
137 end_progress(buf->file_size);
142 static void hash_search(int f,struct sum_struct *s,
143 struct map_struct *buf, OFF_T len)
145 OFF_T offset, end, backup;
148 char sum2[SUM_LENGTH];
153 /* want_i is used to encourage adjacent matches, allowing the RLL
154 * coding of the output to work more efficiently. */
158 rprintf(FINFO,"hash search b=%u len=%.0f\n",
159 s->blength, (double)len);
162 k = MIN(len, s->blength);
164 map = (schar *)map_ptr(buf, 0, k);
166 sum = get_checksum1((char *)map, k);
170 rprintf(FINFO, "sum=%.8x k=%u\n", sum, k);
174 end = len + 1 - s->sums[s->count-1].len;
177 rprintf(FINFO, "hash search s->blength=%u len=%.0f count=%.0f\n",
178 s->blength, (double)len, (double)s->count);
182 tag t = gettag2(s1,s2);
184 size_t j = tag_table[t];
187 rprintf(FINFO,"offset=%.0f sum=%08x\n",(double)offset,sum);
192 sum = (s1 & 0xffff) | (s2 << 16);
196 size_t i = targets[j].i;
198 if (sum != s->sums[i].sum1)
201 /* also make sure the two blocks are the same length */
202 l = MIN((OFF_T)s->blength, len-offset);
203 if (l != s->sums[i].len)
206 /* inplace: ensure chunk's offset is either >= our
207 * offset or that the data didn't move. */
208 if (inplace && !make_backups && s->sums[i].offset < offset
209 && !(s->sums[i].flags & SUMFLG_SAME_OFFSET))
213 rprintf(FINFO,"potential match at %.0f target=%.0f %.0f sum=%08x\n",
214 (double)offset,(double)j,(double)i,sum);
217 map = (schar *)map_ptr(buf,offset,l);
218 get_checksum2((char *)map,l,sum2);
222 if (memcmp(sum2,s->sums[i].sum2,s->s2length) != 0) {
227 /* If inplace is enabled, the best possible match is
228 * one with an identical offset, so we prefer that over
229 * the following want_i optimization. */
230 if (inplace && !make_backups) {
232 size_t i2 = targets[j].i;
233 if (s->sums[i2].offset != offset)
236 if (sum != s->sums[i2].sum1)
238 if (memcmp(sum2, s->sums[i2].sum2,
243 /* This chunk was at the same offset on
244 * both the sender and the receiver. */
245 s->sums[i].flags |= SUMFLG_SAME_OFFSET;
247 } while (++j < s->count && targets[j].t == t);
250 /* we've found a match, but now check to see
251 * if want_i can hint at a better match. */
252 if (i != want_i && want_i < s->count
253 && (!inplace || make_backups || s->sums[want_i].offset >= offset
254 || s->sums[want_i].flags & SUMFLG_SAME_OFFSET)
255 && sum == s->sums[want_i].sum1
256 && memcmp(sum2, s->sums[want_i].sum2, s->s2length) == 0) {
257 /* we've found an adjacent match - the RLL coder
264 matched(f,s,buf,offset,i);
265 offset += s->sums[i].len - 1;
266 k = MIN(s->blength, len-offset);
267 map = (schar *)map_ptr(buf, offset, k);
268 sum = get_checksum1((char *)map, k);
273 } while (++j < s->count && targets[j].t == t);
276 backup = offset - last_match;
277 /* We sometimes read 1 byte prior to last_match... */
281 /* Trim off the first byte from the checksum */
282 more = offset + k < len;
283 map = (schar *)map_ptr(buf, offset - backup, k + more + backup)
285 s1 -= map[0] + CHAR_OFFSET;
286 s2 -= k * (map[0]+CHAR_OFFSET);
288 /* Add on the next byte (if there is one) to the checksum */
290 s1 += map[k] + CHAR_OFFSET;
295 /* By matching early we avoid re-reading the
296 data 3 times in the case where a token
297 match comes a long way after last
298 match. The 3 reads are caused by the
299 running match, the checksum update and the
301 if (backup >= CHUNK_SIZE + s->blength
302 && end - offset > CHUNK_SIZE) {
303 matched(f,s,buf,offset - s->blength, -2);
305 } while (++offset < end);
307 matched(f,s,buf,len,-1);
308 map_ptr(buf,len-1,1);
313 * Scan through a origin file, looking for sections that match
314 * checksums from the generator, and transmit either literal or token
317 * Also calculates the MD4 checksum of the whole file, using the md
318 * accumulator. This is transmitted with the file as protection
319 * against corruption on the wire.
321 * @param s Checksums received from the generator. If <tt>s->count ==
322 * 0</tt>, then there are actually no checksums for this file.
324 * @param len Length of the file to send.
326 void match_sums(int f, struct sum_struct *s, struct map_struct *buf, OFF_T len)
328 char file_sum[MD4_SUM_LENGTH];
336 sum_init(checksum_seed);
338 if (len > 0 && s->count>0) {
342 rprintf(FINFO,"built hash table\n");
344 hash_search(f,s,buf,len);
347 rprintf(FINFO,"done hash search\n");
350 /* by doing this in pieces we avoid too many seeks */
351 for (j = 0; j < len-CHUNK_SIZE; j += CHUNK_SIZE) {
352 int n1 = MIN(CHUNK_SIZE,(len-CHUNK_SIZE)-j);
353 matched(f,s,buf,j+n1,-2);
355 matched(f,s,buf,len,-1);
359 /* If we had a read error, send a bad checksum. */
360 if (buf && buf->status != 0)
364 rprintf(FINFO,"sending file_sum\n");
365 write_buf(f,file_sum,MD4_SUM_LENGTH);
373 rprintf(FINFO, "false_alarms=%d tag_hits=%d matches=%d\n",
374 false_alarms, tag_hits, matches);
376 total_tag_hits += tag_hits;
377 total_false_alarms += false_alarms;
378 total_matches += matches;
379 stats.literal_data += data_transfer;
382 void match_report(void)
388 "total: matches=%d tag_hits=%d false_alarms=%d data=%.0f\n",
389 total_matches,total_tag_hits,
391 (double)stats.literal_data);