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 int updating_basis_file;
29 typedef unsigned short tag;
31 #define TABLESIZE (1<<16)
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 int32 *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(int32, 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,
105 OFF_T offset, int32 i)
107 int32 n = offset - last_match; /* max value: block_size (int32) */
110 if (verbose > 2 && i >= 0) {
112 "match at %.0f last_match=%.0f j=%d len=%ld n=%ld\n",
113 (double)offset, (double)last_match, i,
114 (long)s->sums[i].len, (long)n);
117 send_token(f, i, buf, last_match, n, i < 0 ? 0 : s->sums[i].len);
121 stats.matched_data += s->sums[i].len;
125 for (j = 0; j < n; j += CHUNK_SIZE) {
126 int32 n1 = MIN(CHUNK_SIZE, n - j);
127 sum_update(map_ptr(buf, last_match + j, n1), n1);
132 last_match = offset + s->sums[i].len;
136 if (buf && do_progress) {
137 show_progress(last_match, buf->file_size);
140 end_progress(buf->file_size);
145 static void hash_search(int f,struct sum_struct *s,
146 struct map_struct *buf, OFF_T len)
148 OFF_T offset, end, backup;
150 char sum2[SUM_LENGTH];
155 /* want_i is used to encourage adjacent matches, allowing the RLL
156 * coding of the output to work more efficiently. */
160 rprintf(FINFO, "hash search b=%ld len=%.0f\n",
161 (long)s->blength, (double)len);
164 k = (int32)MIN(len, (OFF_T)s->blength);
166 map = (schar *)map_ptr(buf, 0, k);
168 sum = get_checksum1((char *)map, k);
172 rprintf(FINFO, "sum=%.8x k=%ld\n", sum, (long)k);
176 end = len + 1 - s->sums[s->count-1].len;
179 rprintf(FINFO, "hash search s->blength=%ld len=%.0f count=%.0f\n",
180 (long)s->blength, (double)len, (double)s->count);
184 tag t = gettag2(s1,s2);
186 int32 j = tag_table[t];
189 rprintf(FINFO,"offset=%.0f sum=%08x\n",(double)offset,sum);
194 sum = (s1 & 0xffff) | (s2 << 16);
197 int32 l, i = targets[j].i;
199 if (sum != s->sums[i].sum1)
202 /* also make sure the two blocks are the same length */
203 l = (int32)MIN((OFF_T)s->blength, len-offset);
204 if (l != s->sums[i].len)
207 /* in-place: ensure chunk's offset is either >= our
208 * offset or that the data didn't move. */
209 if (updating_basis_file && s->sums[i].offset < offset
210 && !(s->sums[i].flags & SUMFLG_SAME_OFFSET))
214 rprintf(FINFO,"potential match at %.0f target=%.0f %.0f sum=%08x\n",
215 (double)offset,(double)j,(double)i,sum);
218 map = (schar *)map_ptr(buf,offset,l);
219 get_checksum2((char *)map,l,sum2);
223 if (memcmp(sum2,s->sums[i].sum2,s->s2length) != 0) {
228 /* When updating in-place, the best possible match is
229 * one with an identical offset, so we prefer that over
230 * the following want_i optimization. */
231 if (updating_basis_file) {
233 int32 i2 = targets[j].i;
234 if (s->sums[i2].offset != offset)
237 if (sum != s->sums[i2].sum1)
239 if (memcmp(sum2, s->sums[i2].sum2,
244 /* This chunk was at the same offset on
245 * both the sender and the receiver. */
246 s->sums[i].flags |= SUMFLG_SAME_OFFSET;
248 } while (++j < s->count && targets[j].t == t);
251 /* we've found a match, but now check to see
252 * if want_i can hint at a better match. */
253 if (i != want_i && want_i < s->count
254 && (!updating_basis_file || s->sums[want_i].offset >= offset
255 || s->sums[want_i].flags & SUMFLG_SAME_OFFSET)
256 && sum == s->sums[want_i].sum1
257 && memcmp(sum2, s->sums[want_i].sum2, s->s2length) == 0) {
258 /* we've found an adjacent match - the RLL coder
265 matched(f,s,buf,offset,i);
266 offset += s->sums[i].len - 1;
267 k = (int32)MIN((OFF_T)s->blength, len-offset);
268 map = (schar *)map_ptr(buf, offset, k);
269 sum = get_checksum1((char *)map, k);
274 } while (++j < s->count && targets[j].t == t);
277 backup = offset - last_match;
278 /* We sometimes read 1 byte prior to last_match... */
282 /* Trim off the first byte from the checksum */
283 more = offset + k < len;
284 map = (schar *)map_ptr(buf, offset - backup, k + more + backup)
286 s1 -= map[0] + CHAR_OFFSET;
287 s2 -= k * (map[0]+CHAR_OFFSET);
289 /* Add on the next byte (if there is one) to the checksum */
291 s1 += map[k] + CHAR_OFFSET;
296 /* By matching early we avoid re-reading the
297 data 3 times in the case where a token
298 match comes a long way after last
299 match. The 3 reads are caused by the
300 running match, the checksum update and the
302 if (backup >= s->blength+CHUNK_SIZE && end-offset > CHUNK_SIZE)
303 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 = CHUNK_SIZE; j < len; j += CHUNK_SIZE)
351 matched(f, s, buf, j, -2);
352 matched(f, s, buf, len, -1);
356 /* If we had a read error, send a bad checksum. */
357 if (buf && buf->status != 0)
361 rprintf(FINFO,"sending file_sum\n");
362 write_buf(f,file_sum,MD4_SUM_LENGTH);
370 rprintf(FINFO, "false_alarms=%d tag_hits=%d matches=%d\n",
371 false_alarms, tag_hits, matches);
373 total_tag_hits += tag_hits;
374 total_false_alarms += false_alarms;
375 total_matches += matches;
376 stats.literal_data += data_transfer;
379 void match_report(void)
385 "total: matches=%d tag_hits=%d false_alarms=%d data=%.0f\n",
386 total_matches,total_tag_hits,
388 (double)stats.literal_data);