2 * Block matching used by the file-transfer code.
4 * Copyright (C) 1996 Andrew Tridgell
5 * Copyright (C) 1996 Paul Mackerras
6 * Copyright (C) 2003-2007 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 version 3 as
10 * published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License along
18 * with this program; if not, visit the http://fsf.org website.
24 extern int do_progress;
25 extern int checksum_seed;
26 extern int append_mode;
28 int updating_basis_file;
30 static int false_alarms;
33 static int64 data_transfer;
35 static int total_false_alarms;
36 static int total_hash_hits;
37 static int total_matches;
39 extern struct stats stats;
41 #define TABLESIZE (1<<16)
43 static int32 *hash_table;
45 #define SUM2HASH2(s1,s2) (((s1) + (s2)) & 0xFFFF)
46 #define SUM2HASH(sum) SUM2HASH2((sum)&0xFFFF,(sum)>>16)
48 static void build_hash_table(struct sum_struct *s)
53 hash_table = new_array(int32, TABLESIZE);
55 out_of_memory("build_hash_table");
58 memset(hash_table, 0xFF, TABLESIZE * sizeof hash_table[0]);
60 for (i = 0; i < s->count; i++) {
61 uint32 t = SUM2HASH(s->sums[i].sum1);
62 s->sums[i].chain = hash_table[t];
68 static OFF_T last_match;
72 * Transmit a literal and/or match token.
74 * This delightfully-named function is called either when we find a
75 * match and need to transmit all the unmatched data leading up to it,
76 * or when we get bored of accumulating literal data and just need to
77 * transmit it. As a result of this second case, it is called even if
78 * we have not matched at all!
80 * @param i If >0, the number of a matched token. If 0, indicates we
81 * have only literal data.
83 static void matched(int f, struct sum_struct *s, struct map_struct *buf,
84 OFF_T offset, int32 i)
86 int32 n = (int32)(offset - last_match); /* max value: block_size (int32) */
89 if (verbose > 2 && i >= 0) {
91 "match at %.0f last_match=%.0f j=%d len=%ld n=%ld\n",
92 (double)offset, (double)last_match, i,
93 (long)s->sums[i].len, (long)n);
96 send_token(f, i, buf, last_match, n, i < 0 ? 0 : s->sums[i].len);
100 stats.matched_data += s->sums[i].len;
104 for (j = 0; j < n; j += CHUNK_SIZE) {
105 int32 n1 = MIN(CHUNK_SIZE, n - j);
106 sum_update(map_ptr(buf, last_match + j, n1), n1);
110 last_match = offset + s->sums[i].len;
114 if (buf && do_progress)
115 show_progress(last_match, buf->file_size);
119 static void hash_search(int f,struct sum_struct *s,
120 struct map_struct *buf, OFF_T len)
123 int32 k, want_i, backup;
124 char sum2[SUM_LENGTH];
129 /* want_i is used to encourage adjacent matches, allowing the RLL
130 * coding of the output to work more efficiently. */
134 rprintf(FINFO, "hash search b=%ld len=%.0f\n",
135 (long)s->blength, (double)len);
138 k = (int32)MIN(len, (OFF_T)s->blength);
140 map = (schar *)map_ptr(buf, 0, k);
142 sum = get_checksum1((char *)map, k);
146 rprintf(FINFO, "sum=%.8x k=%ld\n", sum, (long)k);
150 end = len + 1 - s->sums[s->count-1].len;
153 rprintf(FINFO, "hash search s->blength=%ld len=%.0f count=%.0f\n",
154 (long)s->blength, (double)len, (double)s->count);
162 rprintf(FINFO, "offset=%.0f sum=%04x%04x\n",
163 (double)offset, s2 & 0xFFFF, s1 & 0xFFFF);
166 i = hash_table[SUM2HASH2(s1,s2)];
170 sum = (s1 & 0xffff) | (s2 << 16);
175 if (sum != s->sums[i].sum1)
178 /* also make sure the two blocks are the same length */
179 l = (int32)MIN((OFF_T)s->blength, len-offset);
180 if (l != s->sums[i].len)
183 /* in-place: ensure chunk's offset is either >= our
184 * offset or that the data didn't move. */
185 if (updating_basis_file && s->sums[i].offset < offset
186 && !(s->sums[i].flags & SUMFLG_SAME_OFFSET))
191 "potential match at %.0f i=%ld sum=%08x\n",
192 (double)offset, (long)i, sum);
196 map = (schar *)map_ptr(buf,offset,l);
197 get_checksum2((char *)map,l,sum2);
201 if (memcmp(sum2,s->sums[i].sum2,s->s2length) != 0) {
206 /* When updating in-place, the best possible match is
207 * one with an identical offset, so we prefer that over
208 * the following want_i optimization. */
209 if (updating_basis_file) {
211 for (i2 = i; i2 >= 0; i2 = s->sums[i2].chain) {
212 if (s->sums[i2].offset != offset)
215 if (sum != s->sums[i2].sum1)
217 if (memcmp(sum2, s->sums[i2].sum2,
222 /* This chunk was at the same offset on
223 * both the sender and the receiver. */
224 s->sums[i].flags |= SUMFLG_SAME_OFFSET;
229 /* we've found a match, but now check to see
230 * if want_i can hint at a better match. */
231 if (i != want_i && want_i < s->count
232 && (!updating_basis_file || s->sums[want_i].offset >= offset
233 || s->sums[want_i].flags & SUMFLG_SAME_OFFSET)
234 && sum == s->sums[want_i].sum1
235 && memcmp(sum2, s->sums[want_i].sum2, s->s2length) == 0) {
236 /* we've found an adjacent match - the RLL coder
243 matched(f,s,buf,offset,i);
244 offset += s->sums[i].len - 1;
245 k = (int32)MIN((OFF_T)s->blength, len-offset);
246 map = (schar *)map_ptr(buf, offset, k);
247 sum = get_checksum1((char *)map, k);
252 } while ((i = s->sums[i].chain) >= 0);
255 backup = (int32)(offset - last_match);
256 /* We sometimes read 1 byte prior to last_match... */
260 /* Trim off the first byte from the checksum */
261 more = offset + k < len;
262 map = (schar *)map_ptr(buf, offset - backup, k + more + backup)
264 s1 -= map[0] + CHAR_OFFSET;
265 s2 -= k * (map[0]+CHAR_OFFSET);
267 /* Add on the next byte (if there is one) to the checksum */
269 s1 += map[k] + CHAR_OFFSET;
274 /* By matching early we avoid re-reading the
275 data 3 times in the case where a token
276 match comes a long way after last
277 match. The 3 reads are caused by the
278 running match, the checksum update and the
280 if (backup >= s->blength+CHUNK_SIZE && end-offset > CHUNK_SIZE)
281 matched(f, s, buf, offset - s->blength, -2);
282 } while (++offset < end);
284 matched(f, s, buf, len, -1);
285 map_ptr(buf, len-1, 1);
290 * Scan through a origin file, looking for sections that match
291 * checksums from the generator, and transmit either literal or token
294 * Also calculates the MD4 checksum of the whole file, using the md
295 * accumulator. This is transmitted with the file as protection
296 * against corruption on the wire.
298 * @param s Checksums received from the generator. If <tt>s->count ==
299 * 0</tt>, then there are actually no checksums for this file.
301 * @param len Length of the file to send.
303 void match_sums(int f, struct sum_struct *s, struct map_struct *buf, OFF_T len)
305 char file_sum[MAX_DIGEST_LEN];
314 sum_init(checksum_seed);
316 if (append_mode > 0) {
318 for (j = CHUNK_SIZE; j < s->flength; j += CHUNK_SIZE) {
319 if (buf && do_progress)
320 show_progress(last_match, buf->file_size);
321 sum_update(map_ptr(buf, last_match, CHUNK_SIZE),
325 if (last_match < s->flength) {
326 int32 n = (int32)(s->flength - last_match);
327 if (buf && do_progress)
328 show_progress(last_match, buf->file_size);
329 sum_update(map_ptr(buf, last_match, n), n);
330 last_match = s->flength;
335 if (len > 0 && s->count > 0) {
339 rprintf(FINFO,"built hash table\n");
341 hash_search(f, s, buf, len);
344 rprintf(FINFO,"done hash search\n");
347 /* by doing this in pieces we avoid too many seeks */
348 for (j = last_match + CHUNK_SIZE; j < len; j += CHUNK_SIZE)
349 matched(f, s, buf, j, -2);
350 matched(f, s, buf, len, -1);
353 sum_len = sum_end(file_sum);
354 /* If we had a read error, send a bad checksum. */
355 if (buf && buf->status != 0)
359 rprintf(FINFO,"sending file_sum\n");
360 write_buf(f, file_sum, sum_len);
363 rprintf(FINFO, "false_alarms=%d hash_hits=%d matches=%d\n",
364 false_alarms, hash_hits, matches);
366 total_hash_hits += hash_hits;
367 total_false_alarms += false_alarms;
368 total_matches += matches;
369 stats.literal_data += data_transfer;
372 void match_report(void)
378 "total: matches=%d hash_hits=%d false_alarms=%d data=%.0f\n",
379 total_matches, total_hash_hits, total_false_alarms,
380 (double)stats.literal_data);