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_tag_hits;
36 static int total_matches;
38 extern struct stats stats;
40 static uint32 tablesize;
41 static int32 *sum_table;
43 #define gettag2(s1,s2) gettag((s1) + ((s2)<<16))
44 #define gettag(sum) ((sum)%tablesize)
46 static void build_hash_table(struct sum_struct *s)
49 uint32 prior_size = tablesize;
51 /* Dynamically calculate the hash table size so that the hash load
52 * for big files is about 80%. This number must be odd or s2 will
53 * not be able to span the entire set. */
54 tablesize = (uint32)(s->count/8) * 10 + 11;
55 if (tablesize < 65537)
56 tablesize = 65537; /* a prime number */
57 if (tablesize != prior_size) {
59 sum_table = new_array(int32, tablesize);
61 out_of_memory("build_hash_table");
64 memset(sum_table, 0xFF, tablesize * sizeof sum_table[0]);
66 for (i = 0; i < s->count; i++) {
67 uint32 t = gettag(s->sums[i].sum1);
68 s->sums[i].chain = sum_table[t];
74 static OFF_T last_match;
78 * Transmit a literal and/or match token.
80 * This delightfully-named function is called either when we find a
81 * match and need to transmit all the unmatched data leading up to it,
82 * or when we get bored of accumulating literal data and just need to
83 * transmit it. As a result of this second case, it is called even if
84 * we have not matched at all!
86 * @param i If >0, the number of a matched token. If 0, indicates we
87 * have only literal data.
89 static void matched(int f, struct sum_struct *s, struct map_struct *buf,
90 OFF_T offset, int32 i)
92 int32 n = offset - last_match; /* max value: block_size (int32) */
95 if (verbose > 2 && i >= 0) {
97 "match at %.0f last_match=%.0f j=%d len=%ld n=%ld\n",
98 (double)offset, (double)last_match, i,
99 (long)s->sums[i].len, (long)n);
102 send_token(f, i, buf, last_match, n, i < 0 ? 0 : s->sums[i].len);
106 stats.matched_data += s->sums[i].len;
110 for (j = 0; j < n; j += CHUNK_SIZE) {
111 int32 n1 = MIN(CHUNK_SIZE, n - j);
112 sum_update(map_ptr(buf, last_match + j, n1), n1);
116 last_match = offset + s->sums[i].len;
120 if (buf && do_progress)
121 show_progress(last_match, buf->file_size);
125 static void hash_search(int f,struct sum_struct *s,
126 struct map_struct *buf, OFF_T len)
128 OFF_T offset, end, backup;
130 char sum2[SUM_LENGTH];
135 /* want_i is used to encourage adjacent matches, allowing the RLL
136 * coding of the output to work more efficiently. */
140 rprintf(FINFO, "hash search b=%ld len=%.0f\n",
141 (long)s->blength, (double)len);
144 k = (int32)MIN(len, (OFF_T)s->blength);
146 map = (schar *)map_ptr(buf, 0, k);
148 sum = get_checksum1((char *)map, k);
152 rprintf(FINFO, "sum=%.8x k=%ld\n", sum, (long)k);
156 end = len + 1 - s->sums[s->count-1].len;
159 rprintf(FINFO, "hash search s->blength=%ld len=%.0f count=%.0f\n",
160 (long)s->blength, (double)len, (double)s->count);
164 uint32 t = gettag2(s1,s2);
169 rprintf(FINFO,"offset=%.0f sum=%08x\n",(double)offset,sum);
171 sum = (s1 & 0xffff) | (s2 << 16);
173 for (i = sum_table[t]; i >= 0; i = s->sums[i].chain) {
176 if (sum != s->sums[i].sum1)
179 /* also make sure the two blocks are the same length */
180 l = (int32)MIN((OFF_T)s->blength, len-offset);
181 if (l != s->sums[i].len)
184 /* in-place: ensure chunk's offset is either >= our
185 * offset or that the data didn't move. */
186 if (updating_basis_file && s->sums[i].offset < offset
187 && !(s->sums[i].flags & SUMFLG_SAME_OFFSET))
192 "potential match at %.0f i=%ld sum=%08x\n",
193 (double)offset, (long)i, sum);
197 map = (schar *)map_ptr(buf,offset,l);
198 get_checksum2((char *)map,l,sum2);
202 if (memcmp(sum2,s->sums[i].sum2,s->s2length) != 0) {
207 /* When updating in-place, the best possible match is
208 * one with an identical offset, so we prefer that over
209 * the following want_i optimization. */
210 if (updating_basis_file) {
212 for (i2 = i; i2 >= 0; i2 = s->sums[i2].chain) {
213 if (s->sums[i2].offset != offset)
216 if (sum != s->sums[i2].sum1)
218 if (memcmp(sum2, s->sums[i2].sum2,
223 /* This chunk was at the same offset on
224 * both the sender and the receiver. */
225 s->sums[i].flags |= SUMFLG_SAME_OFFSET;
230 /* we've found a match, but now check to see
231 * if want_i can hint at a better match. */
232 if (i != want_i && want_i < s->count
233 && (!updating_basis_file || s->sums[want_i].offset >= offset
234 || s->sums[want_i].flags & SUMFLG_SAME_OFFSET)
235 && sum == s->sums[want_i].sum1
236 && memcmp(sum2, s->sums[want_i].sum2, s->s2length) == 0) {
237 /* we've found an adjacent match - the RLL coder
244 matched(f,s,buf,offset,i);
245 offset += s->sums[i].len - 1;
246 k = (int32)MIN((OFF_T)s->blength, len-offset);
247 map = (schar *)map_ptr(buf, offset, k);
248 sum = get_checksum1((char *)map, k);
255 backup = 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[MD4_SUM_LENGTH];
313 sum_init(checksum_seed);
317 for (j = CHUNK_SIZE; j < s->flength; j += CHUNK_SIZE) {
318 if (buf && do_progress)
319 show_progress(last_match, buf->file_size);
320 sum_update(map_ptr(buf, last_match, CHUNK_SIZE),
324 if (last_match < s->flength) {
325 int32 len = s->flength - last_match;
326 if (buf && do_progress)
327 show_progress(last_match, buf->file_size);
328 sum_update(map_ptr(buf, last_match, len), len);
329 last_match = s->flength;
334 if (len > 0 && s->count > 0) {
338 rprintf(FINFO,"built hash table\n");
340 hash_search(f,s,buf,len);
343 rprintf(FINFO,"done hash search\n");
346 /* by doing this in pieces we avoid too many seeks */
347 for (j = last_match + CHUNK_SIZE; j < len; j += CHUNK_SIZE)
348 matched(f, s, buf, j, -2);
349 matched(f, s, buf, len, -1);
353 /* If we had a read error, send a bad checksum. */
354 if (buf && buf->status != 0)
358 rprintf(FINFO,"sending file_sum\n");
359 write_buf(f,file_sum,MD4_SUM_LENGTH);
362 rprintf(FINFO, "false_alarms=%d tag_hits=%d matches=%d\n",
363 false_alarms, tag_hits, matches);
365 total_tag_hits += tag_hits;
366 total_false_alarms += false_alarms;
367 total_matches += matches;
368 stats.literal_data += data_transfer;
371 void match_report(void)
377 "total: matches=%d tag_hits=%d false_alarms=%d data=%.0f\n",
378 total_matches,total_tag_hits,
380 (double)stats.literal_data);