2 * Block matching used by the file-transfer code.
4 * Copyright (C) 1996 Andrew Tridgell
5 * Copyright (C) 1996 Paul Mackerras
6 * Copyright (C) 2003, 2004, 2005, 2006 Wayne Davison
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License along
19 * with this program; if not, write to the Free Software Foundation, Inc.,
20 * 51 Franklin Street - Fifth Floor, Boston, MA 02110-1301, USA.
26 extern int do_progress;
27 extern int checksum_seed;
28 extern int append_mode;
30 int updating_basis_file;
32 static int false_alarms;
35 static int64 data_transfer;
37 static int total_false_alarms;
38 static int total_hash_hits;
39 static int total_matches;
41 extern struct stats stats;
43 #define TABLESIZE (1<<16)
45 static int32 *hash_table;
47 #define SUM2HASH2(s1,s2) (((s1) + (s2)) & 0xFFFF)
48 #define SUM2HASH(sum) SUM2HASH2((sum)&0xFFFF,(sum)>>16)
50 static void build_hash_table(struct sum_struct *s)
55 hash_table = new_array(int32, TABLESIZE);
57 out_of_memory("build_hash_table");
60 memset(hash_table, 0xFF, TABLESIZE * sizeof hash_table[0]);
62 for (i = 0; i < s->count; i++) {
63 uint32 t = SUM2HASH(s->sums[i].sum1);
64 s->sums[i].chain = hash_table[t];
70 static OFF_T last_match;
74 * Transmit a literal and/or match token.
76 * This delightfully-named function is called either when we find a
77 * match and need to transmit all the unmatched data leading up to it,
78 * or when we get bored of accumulating literal data and just need to
79 * transmit it. As a result of this second case, it is called even if
80 * we have not matched at all!
82 * @param i If >0, the number of a matched token. If 0, indicates we
83 * have only literal data.
85 static void matched(int f, struct sum_struct *s, struct map_struct *buf,
86 OFF_T offset, int32 i)
88 int32 n = offset - last_match; /* max value: block_size (int32) */
91 if (verbose > 2 && i >= 0) {
93 "match at %.0f last_match=%.0f j=%d len=%ld n=%ld\n",
94 (double)offset, (double)last_match, i,
95 (long)s->sums[i].len, (long)n);
98 send_token(f, i, buf, last_match, n, i < 0 ? 0 : s->sums[i].len);
102 stats.matched_data += s->sums[i].len;
106 for (j = 0; j < n; j += CHUNK_SIZE) {
107 int32 n1 = MIN(CHUNK_SIZE, n - j);
108 sum_update(map_ptr(buf, last_match + j, n1), n1);
112 last_match = offset + s->sums[i].len;
116 if (buf && do_progress)
117 show_progress(last_match, buf->file_size);
121 static void hash_search(int f,struct sum_struct *s,
122 struct map_struct *buf, OFF_T len)
124 OFF_T offset, end, backup;
126 char sum2[SUM_LENGTH];
131 /* want_i is used to encourage adjacent matches, allowing the RLL
132 * coding of the output to work more efficiently. */
136 rprintf(FINFO, "hash search b=%ld len=%.0f\n",
137 (long)s->blength, (double)len);
140 k = (int32)MIN(len, (OFF_T)s->blength);
142 map = (schar *)map_ptr(buf, 0, k);
144 sum = get_checksum1((char *)map, k);
148 rprintf(FINFO, "sum=%.8x k=%ld\n", sum, (long)k);
152 end = len + 1 - s->sums[s->count-1].len;
155 rprintf(FINFO, "hash search s->blength=%ld len=%.0f count=%.0f\n",
156 (long)s->blength, (double)len, (double)s->count);
164 rprintf(FINFO, "offset=%.0f sum=%04x%04x\n",
165 (double)offset, s2 & 0xFFFF, s1 & 0xFFFF);
168 i = hash_table[SUM2HASH2(s1,s2)];
172 sum = (s1 & 0xffff) | (s2 << 16);
177 if (sum != s->sums[i].sum1)
180 /* also make sure the two blocks are the same length */
181 l = (int32)MIN((OFF_T)s->blength, len-offset);
182 if (l != s->sums[i].len)
185 /* in-place: ensure chunk's offset is either >= our
186 * offset or that the data didn't move. */
187 if (updating_basis_file && s->sums[i].offset < offset
188 && !(s->sums[i].flags & SUMFLG_SAME_OFFSET))
193 "potential match at %.0f i=%ld sum=%08x\n",
194 (double)offset, (long)i, sum);
198 map = (schar *)map_ptr(buf,offset,l);
199 get_checksum2((char *)map,l,sum2);
203 if (memcmp(sum2,s->sums[i].sum2,s->s2length) != 0) {
208 /* When updating in-place, the best possible match is
209 * one with an identical offset, so we prefer that over
210 * the following want_i optimization. */
211 if (updating_basis_file) {
213 for (i2 = i; i2 >= 0; i2 = s->sums[i2].chain) {
214 if (s->sums[i2].offset != offset)
217 if (sum != s->sums[i2].sum1)
219 if (memcmp(sum2, s->sums[i2].sum2,
224 /* This chunk was at the same offset on
225 * both the sender and the receiver. */
226 s->sums[i].flags |= SUMFLG_SAME_OFFSET;
231 /* we've found a match, but now check to see
232 * if want_i can hint at a better match. */
233 if (i != want_i && want_i < s->count
234 && (!updating_basis_file || s->sums[want_i].offset >= offset
235 || s->sums[want_i].flags & SUMFLG_SAME_OFFSET)
236 && sum == s->sums[want_i].sum1
237 && memcmp(sum2, s->sums[want_i].sum2, s->s2length) == 0) {
238 /* we've found an adjacent match - the RLL coder
245 matched(f,s,buf,offset,i);
246 offset += s->sums[i].len - 1;
247 k = (int32)MIN((OFF_T)s->blength, len-offset);
248 map = (schar *)map_ptr(buf, offset, k);
249 sum = get_checksum1((char *)map, k);
254 } while ((i = s->sums[i].chain) >= 0);
257 backup = offset - last_match;
258 /* We sometimes read 1 byte prior to last_match... */
262 /* Trim off the first byte from the checksum */
263 more = offset + k < len;
264 map = (schar *)map_ptr(buf, offset - backup, k + more + backup)
266 s1 -= map[0] + CHAR_OFFSET;
267 s2 -= k * (map[0]+CHAR_OFFSET);
269 /* Add on the next byte (if there is one) to the checksum */
271 s1 += map[k] + CHAR_OFFSET;
276 /* By matching early we avoid re-reading the
277 data 3 times in the case where a token
278 match comes a long way after last
279 match. The 3 reads are caused by the
280 running match, the checksum update and the
282 if (backup >= s->blength+CHUNK_SIZE && end-offset > CHUNK_SIZE)
283 matched(f, s, buf, offset - s->blength, -2);
284 } while (++offset < end);
286 matched(f, s, buf, len, -1);
287 map_ptr(buf, len-1, 1);
292 * Scan through a origin file, looking for sections that match
293 * checksums from the generator, and transmit either literal or token
296 * Also calculates the MD4 checksum of the whole file, using the md
297 * accumulator. This is transmitted with the file as protection
298 * against corruption on the wire.
300 * @param s Checksums received from the generator. If <tt>s->count ==
301 * 0</tt>, then there are actually no checksums for this file.
303 * @param len Length of the file to send.
305 void match_sums(int f, struct sum_struct *s, struct map_struct *buf, OFF_T len)
307 char file_sum[MD4_SUM_LENGTH];
315 sum_init(checksum_seed);
319 for (j = CHUNK_SIZE; j < s->flength; j += CHUNK_SIZE) {
320 if (buf && do_progress)
321 show_progress(last_match, buf->file_size);
322 sum_update(map_ptr(buf, last_match, CHUNK_SIZE),
326 if (last_match < s->flength) {
327 int32 len = s->flength - last_match;
328 if (buf && do_progress)
329 show_progress(last_match, buf->file_size);
330 sum_update(map_ptr(buf, last_match, len), len);
331 last_match = s->flength;
336 if (len > 0 && s->count > 0) {
340 rprintf(FINFO,"built hash table\n");
342 hash_search(f,s,buf,len);
345 rprintf(FINFO,"done hash search\n");
348 /* by doing this in pieces we avoid too many seeks */
349 for (j = last_match + CHUNK_SIZE; j < len; j += CHUNK_SIZE)
350 matched(f, s, buf, j, -2);
351 matched(f, s, buf, len, -1);
355 /* If we had a read error, send a bad checksum. */
356 if (buf && buf->status != 0)
360 rprintf(FINFO,"sending file_sum\n");
361 write_buf(f,file_sum,MD4_SUM_LENGTH);
364 rprintf(FINFO, "false_alarms=%d hash_hits=%d matches=%d\n",
365 false_alarms, hash_hits, matches);
367 total_hash_hits += hash_hits;
368 total_false_alarms += false_alarms;
369 total_matches += matches;
370 stats.literal_data += data_transfer;
373 void match_report(void)
379 "total: matches=%d hash_hits=%d false_alarms=%d data=%.0f\n",
380 total_matches, total_hash_hits, total_false_alarms,
381 (double)stats.literal_data);