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;
26 extern int append_mode;
28 int updating_basis_file;
30 typedef unsigned short tag;
32 #define TABLESIZE (1<<16)
35 static int false_alarms;
38 static int64 data_transfer;
40 static int total_false_alarms;
41 static int total_tag_hits;
42 static int total_matches;
44 extern struct stats stats;
51 static struct target *targets;
53 static int32 *tag_table;
55 #define gettag2(s1,s2) (((s1) + (s2)) & 0xFFFF)
56 #define gettag(sum) gettag2((sum)&0xFFFF,(sum)>>16)
58 static int compare_targets(struct target *t1,struct target *t2)
60 return (int)t1->t - (int)t2->t;
64 static void build_hash_table(struct sum_struct *s)
69 tag_table = new_array(int32, TABLESIZE);
71 targets = new_array(struct target, s->count);
72 if (!tag_table || !targets)
73 out_of_memory("build_hash_table");
75 for (i = 0; i < s->count; i++) {
77 targets[i].t = gettag(s->sums[i].sum1);
80 qsort(targets,s->count,sizeof(targets[0]),(int (*)())compare_targets);
82 for (i = 0; i < TABLESIZE; i++)
83 tag_table[i] = NULL_TAG;
85 for (i = s->count; i-- > 0; )
86 tag_table[targets[i].t] = i;
90 static OFF_T last_match;
94 * Transmit a literal and/or match token.
96 * This delightfully-named function is called either when we find a
97 * match and need to transmit all the unmatched data leading up to it,
98 * or when we get bored of accumulating literal data and just need to
99 * transmit it. As a result of this second case, it is called even if
100 * we have not matched at all!
102 * @param i If >0, the number of a matched token. If 0, indicates we
103 * have only literal data.
105 static void matched(int f, struct sum_struct *s, struct map_struct *buf,
106 OFF_T offset, int32 i)
108 int32 n = offset - last_match; /* max value: block_size (int32) */
111 if (verbose > 2 && i >= 0) {
113 "match at %.0f last_match=%.0f j=%d len=%ld n=%ld\n",
114 (double)offset, (double)last_match, i,
115 (long)s->sums[i].len, (long)n);
118 send_token(f, i, buf, last_match, n, i < 0 ? 0 : s->sums[i].len);
122 stats.matched_data += s->sums[i].len;
126 for (j = 0; j < n; j += CHUNK_SIZE) {
127 int32 n1 = MIN(CHUNK_SIZE, n - j);
128 sum_update(map_ptr(buf, last_match + j, n1), n1);
133 last_match = offset + s->sums[i].len;
137 if (buf && do_progress)
138 show_progress(last_match, 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;
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=%ld len=%.0f\n",
158 (long)s->blength, (double)len);
161 k = (int32)MIN(len, (OFF_T)s->blength);
163 map = (schar *)map_ptr(buf, 0, k);
165 sum = get_checksum1((char *)map, k);
169 rprintf(FINFO, "sum=%.8x k=%ld\n", sum, (long)k);
173 end = len + 1 - s->sums[s->count-1].len;
176 rprintf(FINFO, "hash search s->blength=%ld len=%.0f count=%.0f\n",
177 (long)s->blength, (double)len, (double)s->count);
181 tag t = gettag2(s1,s2);
183 int32 j = tag_table[t];
186 rprintf(FINFO,"offset=%.0f sum=%08x\n",(double)offset,sum);
191 sum = (s1 & 0xffff) | (s2 << 16);
194 int32 l, i = targets[j].i;
196 if (sum != s->sums[i].sum1)
199 /* also make sure the two blocks are the same length */
200 l = (int32)MIN((OFF_T)s->blength, len-offset);
201 if (l != s->sums[i].len)
204 /* in-place: ensure chunk's offset is either >= our
205 * offset or that the data didn't move. */
206 if (updating_basis_file && s->sums[i].offset < offset
207 && !(s->sums[i].flags & SUMFLG_SAME_OFFSET))
211 rprintf(FINFO,"potential match at %.0f target=%.0f %.0f sum=%08x\n",
212 (double)offset,(double)j,(double)i,sum);
215 map = (schar *)map_ptr(buf,offset,l);
216 get_checksum2((char *)map,l,sum2);
220 if (memcmp(sum2,s->sums[i].sum2,s->s2length) != 0) {
225 /* When updating in-place, the best possible match is
226 * one with an identical offset, so we prefer that over
227 * the following want_i optimization. */
228 if (updating_basis_file) {
230 int32 i2 = targets[j].i;
231 if (s->sums[i2].offset != offset)
234 if (sum != s->sums[i2].sum1)
236 if (memcmp(sum2, s->sums[i2].sum2,
241 /* This chunk was at the same offset on
242 * both the sender and the receiver. */
243 s->sums[i].flags |= SUMFLG_SAME_OFFSET;
245 } while (++j < s->count && targets[j].t == t);
248 /* we've found a match, but now check to see
249 * if want_i can hint at a better match. */
250 if (i != want_i && want_i < s->count
251 && (!updating_basis_file || s->sums[want_i].offset >= offset
252 || s->sums[want_i].flags & SUMFLG_SAME_OFFSET)
253 && sum == s->sums[want_i].sum1
254 && memcmp(sum2, s->sums[want_i].sum2, s->s2length) == 0) {
255 /* we've found an adjacent match - the RLL coder
262 matched(f,s,buf,offset,i);
263 offset += s->sums[i].len - 1;
264 k = (int32)MIN((OFF_T)s->blength, len-offset);
265 map = (schar *)map_ptr(buf, offset, k);
266 sum = get_checksum1((char *)map, k);
271 } while (++j < s->count && targets[j].t == t);
274 backup = offset - last_match;
275 /* We sometimes read 1 byte prior to last_match... */
279 /* Trim off the first byte from the checksum */
280 more = offset + k < len;
281 map = (schar *)map_ptr(buf, offset - backup, k + more + backup)
283 s1 -= map[0] + CHAR_OFFSET;
284 s2 -= k * (map[0]+CHAR_OFFSET);
286 /* Add on the next byte (if there is one) to the checksum */
288 s1 += map[k] + CHAR_OFFSET;
293 /* By matching early we avoid re-reading the
294 data 3 times in the case where a token
295 match comes a long way after last
296 match. The 3 reads are caused by the
297 running match, the checksum update and the
299 if (backup >= s->blength+CHUNK_SIZE && end-offset > CHUNK_SIZE)
300 matched(f, s, buf, offset - s->blength, -2);
301 } while (++offset < end);
303 matched(f, s, buf, len, -1);
304 map_ptr(buf, len-1, 1);
309 * Scan through a origin file, looking for sections that match
310 * checksums from the generator, and transmit either literal or token
313 * Also calculates the MD4 checksum of the whole file, using the md
314 * accumulator. This is transmitted with the file as protection
315 * against corruption on the wire.
317 * @param s Checksums received from the generator. If <tt>s->count ==
318 * 0</tt>, then there are actually no checksums for this file.
320 * @param len Length of the file to send.
322 void match_sums(int f, struct sum_struct *s, struct map_struct *buf, OFF_T len)
324 char file_sum[MD4_SUM_LENGTH];
332 sum_init(checksum_seed);
336 for (j = CHUNK_SIZE; j < s->flength; j += CHUNK_SIZE) {
337 if (buf && do_progress)
338 show_progress(last_match, buf->file_size);
339 sum_update(map_ptr(buf, last_match, CHUNK_SIZE),
343 if (last_match < s->flength) {
344 int32 len = s->flength - last_match;
345 if (buf && do_progress)
346 show_progress(last_match, buf->file_size);
347 sum_update(map_ptr(buf, last_match, len), len);
348 last_match = s->flength;
353 if (len > 0 && s->count > 0) {
357 rprintf(FINFO,"built hash table\n");
359 hash_search(f,s,buf,len);
362 rprintf(FINFO,"done hash search\n");
365 /* by doing this in pieces we avoid too many seeks */
366 for (j = last_match + CHUNK_SIZE; j < len; j += CHUNK_SIZE)
367 matched(f, s, buf, j, -2);
368 matched(f, s, buf, len, -1);
372 /* If we had a read error, send a bad checksum. */
373 if (buf && buf->status != 0)
377 rprintf(FINFO,"sending file_sum\n");
378 write_buf(f,file_sum,MD4_SUM_LENGTH);
386 rprintf(FINFO, "false_alarms=%d tag_hits=%d matches=%d\n",
387 false_alarms, tag_hits, matches);
389 total_tag_hits += tag_hits;
390 total_false_alarms += false_alarms;
391 total_matches += matches;
392 stats.literal_data += data_transfer;
395 void match_report(void)
401 "total: matches=%d tag_hits=%d false_alarms=%d data=%.0f\n",
402 total_matches,total_tag_hits,
404 (double)stats.literal_data);