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.
23 extern int do_progress;
24 extern int checksum_seed;
25 extern int append_mode;
27 int updating_basis_file;
29 static int false_alarms;
32 static int64 data_transfer;
34 static int total_false_alarms;
35 static int total_hash_hits;
36 static int total_matches;
38 extern struct stats stats;
40 #define TABLESIZE (1<<16)
42 static int32 *hash_table;
44 #define SUM2HASH2(s1,s2) (((s1) + (s2)) & 0xFFFF)
45 #define SUM2HASH(sum) SUM2HASH2((sum)&0xFFFF,(sum)>>16)
47 static void build_hash_table(struct sum_struct *s)
52 hash_table = new_array(int32, TABLESIZE);
54 out_of_memory("build_hash_table");
57 memset(hash_table, 0xFF, TABLESIZE * sizeof hash_table[0]);
59 for (i = 0; i < s->count; i++) {
60 uint32 t = SUM2HASH(s->sums[i].sum1);
61 s->sums[i].chain = hash_table[t];
67 static OFF_T last_match;
71 * Transmit a literal and/or match token.
73 * This delightfully-named function is called either when we find a
74 * match and need to transmit all the unmatched data leading up to it,
75 * or when we get bored of accumulating literal data and just need to
76 * transmit it. As a result of this second case, it is called even if
77 * we have not matched at all!
79 * @param i If >0, the number of a matched token. If 0, indicates we
80 * have only literal data.
82 static void matched(int f, struct sum_struct *s, struct map_struct *buf,
83 OFF_T offset, int32 i)
85 int32 n = offset - last_match; /* max value: block_size (int32) */
88 if (verbose > 2 && i >= 0) {
90 "match at %.0f last_match=%.0f j=%d len=%ld n=%ld\n",
91 (double)offset, (double)last_match, i,
92 (long)s->sums[i].len, (long)n);
95 send_token(f, i, buf, last_match, n, i < 0 ? 0 : s->sums[i].len);
99 stats.matched_data += s->sums[i].len;
103 for (j = 0; j < n; j += CHUNK_SIZE) {
104 int32 n1 = MIN(CHUNK_SIZE, n - j);
105 sum_update(map_ptr(buf, last_match + j, n1), n1);
109 last_match = offset + s->sums[i].len;
113 if (buf && do_progress)
114 show_progress(last_match, buf->file_size);
118 static void hash_search(int f,struct sum_struct *s,
119 struct map_struct *buf, OFF_T len)
121 OFF_T offset, end, backup;
123 char sum2[SUM_LENGTH];
128 /* want_i is used to encourage adjacent matches, allowing the RLL
129 * coding of the output to work more efficiently. */
133 rprintf(FINFO, "hash search b=%ld len=%.0f\n",
134 (long)s->blength, (double)len);
137 k = (int32)MIN(len, (OFF_T)s->blength);
139 map = (schar *)map_ptr(buf, 0, k);
141 sum = get_checksum1((char *)map, k);
145 rprintf(FINFO, "sum=%.8x k=%ld\n", sum, (long)k);
149 end = len + 1 - s->sums[s->count-1].len;
152 rprintf(FINFO, "hash search s->blength=%ld len=%.0f count=%.0f\n",
153 (long)s->blength, (double)len, (double)s->count);
161 rprintf(FINFO, "offset=%.0f sum=%04x%04x\n",
162 (double)offset, s1 & 0xFFFF, s2 & 0xFFFF);
165 i = hash_table[SUM2HASH2(s1,s2)];
169 sum = (s1 & 0xffff) | (s2 << 16);
174 if (sum != s->sums[i].sum1)
177 /* also make sure the two blocks are the same length */
178 l = (int32)MIN((OFF_T)s->blength, len-offset);
179 if (l != s->sums[i].len)
182 /* in-place: ensure chunk's offset is either >= our
183 * offset or that the data didn't move. */
184 if (updating_basis_file && s->sums[i].offset < offset
185 && !(s->sums[i].flags & SUMFLG_SAME_OFFSET))
190 "potential match at %.0f i=%ld sum=%08x\n",
191 (double)offset, (long)i, sum);
195 map = (schar *)map_ptr(buf,offset,l);
196 get_checksum2((char *)map,l,sum2);
200 if (memcmp(sum2,s->sums[i].sum2,s->s2length) != 0) {
205 /* When updating in-place, the best possible match is
206 * one with an identical offset, so we prefer that over
207 * the following want_i optimization. */
208 if (updating_basis_file) {
210 for (i2 = i; i2 >= 0; i2 = s->sums[i2].chain) {
211 if (s->sums[i2].offset != offset)
214 if (sum != s->sums[i2].sum1)
216 if (memcmp(sum2, s->sums[i2].sum2,
221 /* This chunk was at the same offset on
222 * both the sender and the receiver. */
223 s->sums[i].flags |= SUMFLG_SAME_OFFSET;
228 /* we've found a match, but now check to see
229 * if want_i can hint at a better match. */
230 if (i != want_i && want_i < s->count
231 && (!updating_basis_file || s->sums[want_i].offset >= offset
232 || s->sums[want_i].flags & SUMFLG_SAME_OFFSET)
233 && sum == s->sums[want_i].sum1
234 && memcmp(sum2, s->sums[want_i].sum2, s->s2length) == 0) {
235 /* we've found an adjacent match - the RLL coder
242 matched(f,s,buf,offset,i);
243 offset += s->sums[i].len - 1;
244 k = (int32)MIN((OFF_T)s->blength, len-offset);
245 map = (schar *)map_ptr(buf, offset, k);
246 sum = get_checksum1((char *)map, k);
251 } while ((i = s->sums[i].chain) >= 0);
254 backup = offset - last_match;
255 /* We sometimes read 1 byte prior to last_match... */
259 /* Trim off the first byte from the checksum */
260 more = offset + k < len;
261 map = (schar *)map_ptr(buf, offset - backup, k + more + backup)
263 s1 -= map[0] + CHAR_OFFSET;
264 s2 -= k * (map[0]+CHAR_OFFSET);
266 /* Add on the next byte (if there is one) to the checksum */
268 s1 += map[k] + CHAR_OFFSET;
273 /* By matching early we avoid re-reading the
274 data 3 times in the case where a token
275 match comes a long way after last
276 match. The 3 reads are caused by the
277 running match, the checksum update and the
279 if (backup >= s->blength+CHUNK_SIZE && end-offset > CHUNK_SIZE)
280 matched(f, s, buf, offset - s->blength, -2);
281 } while (++offset < end);
283 matched(f, s, buf, len, -1);
284 map_ptr(buf, len-1, 1);
289 * Scan through a origin file, looking for sections that match
290 * checksums from the generator, and transmit either literal or token
293 * Also calculates the MD4 checksum of the whole file, using the md
294 * accumulator. This is transmitted with the file as protection
295 * against corruption on the wire.
297 * @param s Checksums received from the generator. If <tt>s->count ==
298 * 0</tt>, then there are actually no checksums for this file.
300 * @param len Length of the file to send.
302 void match_sums(int f, struct sum_struct *s, struct map_struct *buf, OFF_T len)
304 char file_sum[MD4_SUM_LENGTH];
312 sum_init(checksum_seed);
316 for (j = CHUNK_SIZE; j < s->flength; j += CHUNK_SIZE) {
317 if (buf && do_progress)
318 show_progress(last_match, buf->file_size);
319 sum_update(map_ptr(buf, last_match, CHUNK_SIZE),
323 if (last_match < s->flength) {
324 int32 len = s->flength - last_match;
325 if (buf && do_progress)
326 show_progress(last_match, buf->file_size);
327 sum_update(map_ptr(buf, last_match, len), len);
328 last_match = s->flength;
333 if (len > 0 && s->count > 0) {
337 rprintf(FINFO,"built hash table\n");
339 hash_search(f,s,buf,len);
342 rprintf(FINFO,"done hash search\n");
345 /* by doing this in pieces we avoid too many seeks */
346 for (j = last_match + CHUNK_SIZE; j < len; j += CHUNK_SIZE)
347 matched(f, s, buf, j, -2);
348 matched(f, s, buf, len, -1);
352 /* If we had a read error, send a bad checksum. */
353 if (buf && buf->status != 0)
357 rprintf(FINFO,"sending file_sum\n");
358 write_buf(f,file_sum,MD4_SUM_LENGTH);
361 rprintf(FINFO, "false_alarms=%d hash_hits=%d matches=%d\n",
362 false_alarms, hash_hits, matches);
364 total_hash_hits += hash_hits;
365 total_false_alarms += false_alarms;
366 total_matches += matches;
367 stats.literal_data += data_transfer;
370 void match_report(void)
376 "total: matches=%d hash_hits=%d false_alarms=%d data=%.0f\n",
377 total_matches, total_hash_hits, total_false_alarms,
378 (double)stats.literal_data);