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 extern int remote_version;
29 typedef unsigned short tag;
31 #define TABLESIZE (1<<16)
34 static int false_alarms;
37 static int64 data_transfer;
39 static int total_false_alarms;
40 static int total_tag_hits;
41 static int total_matches;
43 extern struct stats stats;
50 static struct target *targets;
52 static int *tag_table;
54 #define gettag2(s1,s2) (((s1) + (s2)) & 0xFFFF)
55 #define gettag(sum) gettag2((sum)&0xFFFF,(sum)>>16)
57 static int compare_targets(struct target *t1,struct target *t2)
59 return((int)t1->t - (int)t2->t);
63 static void build_hash_table(struct sum_struct *s)
68 tag_table = (int *)malloc(sizeof(tag_table[0])*TABLESIZE);
70 targets = (struct target *)malloc(sizeof(targets[0])*s->count);
71 if (!tag_table || !targets)
72 out_of_memory("build_hash_table");
74 for (i=0;i<s->count;i++) {
76 targets[i].t = gettag(s->sums[i].sum1);
79 qsort(targets,s->count,sizeof(targets[0]),(int (*)())compare_targets);
81 for (i=0;i<TABLESIZE;i++)
82 tag_table[i] = NULL_TAG;
84 for (i=s->count-1;i>=0;i--) {
85 tag_table[targets[i].t] = i;
90 static OFF_T last_match;
93 static void matched(int f,struct sum_struct *s,struct map_struct *buf,
96 OFF_T n = offset - last_match;
99 if (verbose > 2 && i >= 0)
100 rprintf(FINFO,"match at %.0f last_match=%.0f j=%d len=%d n=%.0f\n",
101 (double)offset,(double)last_match,i,s->sums[i].len,(double)n);
103 send_token(f,i,buf,last_match,n,i<0?0:s->sums[i].len);
107 stats.matched_data += s->sums[i].len;
111 for (j=0;j<n;j+=CHUNK_SIZE) {
112 int n1 = MIN(CHUNK_SIZE,n-j);
113 sum_update(map_ptr(buf,last_match+j,n1),n1);
118 last_match = offset + s->sums[i].len;
123 show_progress(last_match, buf->file_size);
125 if (i == -1) end_progress(buf->file_size);
130 static void hash_search(int f,struct sum_struct *s,
131 struct map_struct *buf,OFF_T len)
135 char sum2[SUM_LENGTH];
139 /* last_i is used to encourage adjacent matches, allowing the RLL coding of the
140 output to work more efficiently */
144 rprintf(FINFO,"hash search b=%d len=%.0f\n",s->n,(double)len);
148 map = (schar *)map_ptr(buf,0,k);
150 sum = get_checksum1((char *)map, k);
154 rprintf(FINFO, "sum=%.8x k=%d\n", sum, k);
158 end = len + 1 - s->sums[s->count-1].len;
161 rprintf(FINFO,"hash search s->n=%d len=%.0f count=%d\n",
162 s->n,(double)len,s->count);
165 tag t = gettag2(s1,s2);
170 rprintf(FINFO,"offset=%.0f sum=%08x\n",(double)offset,sum);
176 sum = (s1 & 0xffff) | (s2 << 16);
178 for (; j<s->count && targets[j].t == t; j++) {
179 int l, i = targets[j].i;
181 if (sum != s->sums[i].sum1) continue;
183 /* also make sure the two blocks are the same length */
184 l = MIN(s->n,len-offset);
185 if (l != s->sums[i].len) continue;
188 rprintf(FINFO,"potential match at %.0f target=%d %d sum=%08x\n",
189 (double)offset,j,i,sum);
192 map = (schar *)map_ptr(buf,offset,l);
193 get_checksum2((char *)map,l,sum2);
197 if (memcmp(sum2,s->sums[i].sum2,csum_length) != 0) {
202 /* we've found a match, but now check to see
203 if last_i can hint at a better match */
204 for (j++; j<s->count && targets[j].t == t; j++) {
205 int i2 = targets[j].i;
206 if (i2 == last_i + 1) {
207 if (sum != s->sums[i2].sum1) break;
208 if (memcmp(sum2,s->sums[i2].sum2,csum_length) != 0) break;
209 /* we've found an adjacent match - the RLL coder
218 matched(f,s,buf,offset,i);
219 offset += s->sums[i].len - 1;
220 k = MIN((len-offset), s->n);
221 map = (schar *)map_ptr(buf,offset,k);
222 sum = get_checksum1((char *)map, k);
230 /* Trim off the first byte from the checksum */
231 map = (schar *)map_ptr(buf,offset,k+1);
232 s1 -= map[0] + CHAR_OFFSET;
233 s2 -= k * (map[0]+CHAR_OFFSET);
235 /* Add on the next byte (if there is one) to the checksum */
236 if (k < (len-offset)) {
237 s1 += (map[k]+CHAR_OFFSET);
243 /* By matching early we avoid re-reading the
244 data 3 times in the case where a token
245 match comes a long way after last
246 match. The 3 reads are caused by the
247 running match, the checksum update and the
249 if (offset-last_match >= CHUNK_SIZE+s->n &&
250 (end-offset > CHUNK_SIZE)) {
251 matched(f,s,buf,offset - s->n, -2);
253 } while (++offset < end);
255 matched(f,s,buf,len,-1);
256 map_ptr(buf,len-1,1);
260 void match_sums(int f,struct sum_struct *s,struct map_struct *buf,OFF_T len)
262 char file_sum[MD4_SUM_LENGTH];
272 if (len > 0 && s->count>0) {
276 rprintf(FINFO,"built hash table\n");
278 hash_search(f,s,buf,len);
281 rprintf(FINFO,"done hash search\n");
284 /* by doing this in pieces we avoid too many seeks */
285 for (j=0;j<(len-CHUNK_SIZE);j+=CHUNK_SIZE) {
286 int n1 = MIN(CHUNK_SIZE,(len-CHUNK_SIZE)-j);
287 matched(f,s,buf,j+n1,-2);
289 matched(f,s,buf,len,-1);
294 if (remote_version >= 14) {
296 rprintf(FINFO,"sending file_sum\n");
297 write_buf(f,file_sum,MD4_SUM_LENGTH);
306 rprintf(FINFO, "false_alarms=%d tag_hits=%d matches=%d\n",
307 false_alarms, tag_hits, matches);
309 total_tag_hits += tag_hits;
310 total_false_alarms += false_alarms;
311 total_matches += matches;
312 stats.literal_data += data_transfer;
315 void match_report(void)
321 "total: matches=%d tag_hits=%d false_alarms=%d data=%.0f\n",
322 total_matches,total_tag_hits,
324 (double)stats.literal_data);