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.
22 extern int csum_length;
27 typedef unsigned short tag;
29 #define TABLESIZE (1<<16)
32 static int false_alarms;
35 static int64 data_transfer;
37 static int total_false_alarms;
38 static int total_tag_hits;
39 static int total_matches;
41 extern struct stats stats;
48 static struct target *targets;
50 static int *tag_table;
52 #define gettag2(s1,s2) (((s1) + (s2)) & 0xFFFF)
53 #define gettag(sum) gettag2((sum)&0xFFFF,(sum)>>16)
55 static int compare_targets(struct target *t1,struct target *t2)
57 return((int)t1->t - (int)t2->t);
61 static void build_hash_table(struct sum_struct *s)
66 tag_table = (int *)malloc(sizeof(tag_table[0])*TABLESIZE);
68 targets = (struct target *)malloc(sizeof(targets[0])*s->count);
69 if (!tag_table || !targets)
70 out_of_memory("build_hash_table");
72 for (i=0;i<(int) s->count;i++) {
74 targets[i].t = gettag(s->sums[i].sum1);
77 qsort(targets,s->count,sizeof(targets[0]),(int (*)())compare_targets);
79 for (i=0;i<TABLESIZE;i++)
80 tag_table[i] = NULL_TAG;
82 for (i=s->count-1;i>=0;i--) {
83 tag_table[targets[i].t] = i;
88 static OFF_T last_match;
92 * Transmit a literal and/or match token.
94 * This delightfully-named function is called either when we find a
95 * match and need to transmit all the unmatched data leading up to it,
96 * or when we get bored of accumulating literal data and just need to
97 * transmit it. As a result of this second case, it is called even if
98 * we have not matched at all!
100 * @param i If >0, the number of a matched token. If 0, indicates we
101 * have only literal data.
103 static void matched(int f,struct sum_struct *s,struct map_struct *buf,
106 OFF_T n = offset - last_match;
109 if (verbose > 2 && i >= 0)
110 rprintf(FINFO,"match at %.0f last_match=%.0f j=%d len=%d n=%.0f\n",
111 (double)offset,(double)last_match,i,s->sums[i].len,(double)n);
113 send_token(f,i,buf,last_match,n,i<0?0:s->sums[i].len);
117 stats.matched_data += s->sums[i].len;
121 for (j=0;j<n;j+=CHUNK_SIZE) {
122 int n1 = MIN(CHUNK_SIZE,n-j);
123 sum_update(map_ptr(buf,last_match+j,n1),n1);
128 last_match = offset + s->sums[i].len;
133 show_progress(last_match, buf->file_size);
135 if (i == -1) end_progress(buf->file_size);
140 static void hash_search(int f,struct sum_struct *s,
141 struct map_struct *buf,OFF_T len)
145 char sum2[SUM_LENGTH];
149 /* last_i is used to encourage adjacent matches, allowing the RLL coding of the
150 output to work more efficiently */
154 rprintf(FINFO,"hash search b=%ld len=%.0f\n",
155 (long) s->n, (double)len);
157 /* cast is to make s->n signed; it should always be reasonably
159 k = MIN(len, (OFF_T) s->n);
161 map = (schar *)map_ptr(buf,0,k);
163 sum = get_checksum1((char *)map, k);
167 rprintf(FINFO, "sum=%.8x k=%d\n", sum, k);
171 end = len + 1 - s->sums[s->count-1].len;
174 rprintf(FINFO, "hash search s->n=%ld len=%.0f count=%ld\n",
175 (long) s->n, (double) len, (long) s->count);
178 tag t = gettag2(s1,s2);
183 rprintf(FINFO,"offset=%.0f sum=%08x\n",(double)offset,sum);
189 sum = (s1 & 0xffff) | (s2 << 16);
191 for (; j < (int) s->count && targets[j].t == t; j++) {
192 int l, i = targets[j].i;
194 if (sum != s->sums[i].sum1) continue;
196 /* also make sure the two blocks are the same length */
197 l = MIN(s->n,len-offset);
198 if (l != s->sums[i].len) continue;
201 rprintf(FINFO,"potential match at %.0f target=%d %d sum=%08x\n",
202 (double)offset,j,i,sum);
205 map = (schar *)map_ptr(buf,offset,l);
206 get_checksum2((char *)map,l,sum2);
210 if (memcmp(sum2,s->sums[i].sum2,csum_length) != 0) {
215 /* we've found a match, but now check to see
216 if last_i can hint at a better match */
217 for (j++; j < (int) s->count && targets[j].t == t; j++) {
218 int i2 = targets[j].i;
219 if (i2 == last_i + 1) {
220 if (sum != s->sums[i2].sum1) break;
221 if (memcmp(sum2,s->sums[i2].sum2,csum_length) != 0) break;
222 /* we've found an adjacent match - the RLL coder
231 matched(f,s,buf,offset,i);
232 offset += s->sums[i].len - 1;
233 k = MIN((len-offset), s->n);
234 map = (schar *)map_ptr(buf,offset,k);
235 sum = get_checksum1((char *)map, k);
243 /* Trim off the first byte from the checksum */
244 map = (schar *)map_ptr(buf,offset,k+1);
245 s1 -= map[0] + CHAR_OFFSET;
246 s2 -= k * (map[0]+CHAR_OFFSET);
248 /* Add on the next byte (if there is one) to the checksum */
249 if (k < (len-offset)) {
250 s1 += (map[k]+CHAR_OFFSET);
256 /* By matching early we avoid re-reading the
257 data 3 times in the case where a token
258 match comes a long way after last
259 match. The 3 reads are caused by the
260 running match, the checksum update and the
262 if (offset > last_match &&
263 offset-last_match >= CHUNK_SIZE+s->n &&
264 (end-offset > CHUNK_SIZE)) {
265 matched(f,s,buf,offset - s->n, -2);
267 } while (++offset < end);
269 matched(f,s,buf,len,-1);
270 map_ptr(buf,len-1,1);
275 * Scan through a origin file, looking for sections that match
276 * checksums from the generator, and transmit either literal or token
279 * Also calculates the MD4 checksum of the whole file, using the md
280 * accumulator. This is transmitted with the file as protection
281 * against corruption on the wire.
283 * @param s Checksums received from the generator. If <tt>s->count ==
284 * 0</tt>, then there are actually no checksums for this file.
286 * @param len Length of the file to send.
288 void match_sums(int f, struct sum_struct *s, struct map_struct *buf, OFF_T len)
290 char file_sum[MD4_SUM_LENGTH];
291 extern int write_batch; /* dw */
301 if (len > 0 && s->count>0) {
305 rprintf(FINFO,"built hash table\n");
307 hash_search(f,s,buf,len);
310 rprintf(FINFO,"done hash search\n");
313 /* by doing this in pieces we avoid too many seeks */
314 for (j=0;j<(len-CHUNK_SIZE);j+=CHUNK_SIZE) {
315 int n1 = MIN(CHUNK_SIZE,(len-CHUNK_SIZE)-j);
316 matched(f,s,buf,j+n1,-2);
318 matched(f,s,buf,len,-1);
324 rprintf(FINFO,"sending file_sum\n");
325 write_buf(f,file_sum,MD4_SUM_LENGTH);
326 if (write_batch) /* dw */
327 write_batch_delta_file(file_sum, MD4_SUM_LENGTH);
335 rprintf(FINFO, "false_alarms=%d tag_hits=%d matches=%d\n",
336 false_alarms, tag_hits, matches);
338 total_tag_hits += tag_hits;
339 total_false_alarms += false_alarms;
340 total_matches += matches;
341 stats.literal_data += data_transfer;
344 void match_report(void)
350 "total: matches=%d tag_hits=%d false_alarms=%d data=%.0f\n",
351 total_matches,total_tag_hits,
353 (double)stats.literal_data);