X-Git-Url: https://mattmccutchen.net/rsync/rsync.git/blobdiff_plain/debb45054a3c4d4da3faac372e993b594e5a0df7..0f9c48b1d23fbe958544094e97ba8ad2a7d8b8f4:/match.c diff --git a/match.c b/match.c index 3689af4f..292c537e 100644 --- a/match.c +++ b/match.c @@ -29,18 +29,18 @@ extern int remote_version; typedef unsigned short tag; #define TABLESIZE (1<<16) -#define NULL_TAG ((tag)-1) +#define NULL_TAG (-1) static int false_alarms; static int tag_hits; static int matches; -static int data_transfer; +static int64 data_transfer; static int total_false_alarms; static int total_tag_hits; static int total_matches; -static int64 total_data_transfer; +extern struct stats stats; struct target { tag t; @@ -49,7 +49,7 @@ struct target { static struct target *targets; -static tag *tag_table; +static int *tag_table; #define gettag2(s1,s2) (((s1) + (s2)) & 0xFFFF) #define gettag(sum) gettag2((sum)&0xFFFF,(sum)>>16) @@ -65,13 +65,13 @@ static void build_hash_table(struct sum_struct *s) int i; if (!tag_table) - tag_table = (tag *)malloc(sizeof(tag)*TABLESIZE); + tag_table = (int *)malloc(sizeof(tag_table[0])*TABLESIZE); targets = (struct target *)malloc(sizeof(targets[0])*s->count); if (!tag_table || !targets) out_of_memory("build_hash_table"); - for (i=0;icount;i++) { + for (i=0;i<(int) s->count;i++) { targets[i].i = i; targets[i].t = gettag(s->sums[i].sum1); } @@ -90,24 +90,35 @@ static void build_hash_table(struct sum_struct *s) static OFF_T last_match; +/** + * Transmit a literal and/or match token. + * + * This delightfully-named function is called either when we find a + * match and need to transmit all the unmatched data leading up to it, + * or when we get bored of accumulating literal data and just need to + * transmit it. As a result of this second case, it is called even if + * we have not matched at all! + * + * @param i If >0, the number of a matched token. If 0, indicates we + * have only literal data. + **/ static void matched(int f,struct sum_struct *s,struct map_struct *buf, OFF_T offset,int i) { OFF_T n = offset - last_match; - int j; + OFF_T j; if (verbose > 2 && i >= 0) - fprintf(FINFO,"match at %d last_match=%d j=%d len=%d n=%d\n", - (int)offset,(int)last_match,i,(int)s->sums[i].len,(int)n); + rprintf(FINFO,"match at %.0f last_match=%.0f j=%d len=%d n=%.0f\n", + (double)offset,(double)last_match,i,s->sums[i].len,(double)n); - send_token(f,i,buf,last_match,n,i==-1?0:s->sums[i].len); + send_token(f,i,buf,last_match,n,i<0?0:s->sums[i].len); data_transfer += n; - if (n > 0) - write_flush(f); - - if (i >= 0) + if (i >= 0) { + stats.matched_data += s->sums[i].len; n += s->sums[i].len; + } for (j=0;jsums[i].len; else last_match = offset; + + if (buf) { + show_progress(last_match, buf->file_size); + + if (i == -1) end_progress(buf->file_size); + } } static void hash_search(int f,struct sum_struct *s, struct map_struct *buf,OFF_T len) { - OFF_T offset; - int j,k; - int end; + OFF_T offset, end; + int j,k, last_i; char sum2[SUM_LENGTH]; uint32 s1, s2, sum; schar *map; + /* last_i is used to encourage adjacent matches, allowing the RLL coding of the + output to work more efficiently */ + last_i = -1; + if (verbose > 2) - fprintf(FINFO,"hash search b=%d len=%d\n",s->n,(int)len); + rprintf(FINFO,"hash search b=%d len=%.0f\n",s->n,(double)len); k = MIN(len, s->n); @@ -143,15 +163,15 @@ static void hash_search(int f,struct sum_struct *s, s1 = sum & 0xFFFF; s2 = sum >> 16; if (verbose > 3) - fprintf(FINFO, "sum=%.8x k=%d\n", sum, k); + rprintf(FINFO, "sum=%.8x k=%d\n", sum, k); offset = 0; end = len + 1 - s->sums[s->count-1].len; if (verbose > 3) - fprintf(FINFO,"hash search s->n=%d len=%d count=%d\n", - s->n,(int)len,s->count); + rprintf(FINFO,"hash search s->n=%d len=%.0f count=%d\n", + s->n,(double)len,s->count); do { tag t = gettag2(s1,s2); @@ -159,7 +179,7 @@ static void hash_search(int f,struct sum_struct *s, j = tag_table[t]; if (verbose > 4) - fprintf(FINFO,"offset=%d sum=%08x\n",(int)offset,sum); + rprintf(FINFO,"offset=%.0f sum=%08x\n",(double)offset,sum); if (j == NULL_TAG) { goto null_tag; @@ -167,17 +187,20 @@ static void hash_search(int f,struct sum_struct *s, sum = (s1 & 0xffff) | (s2 << 16); tag_hits++; - for (; jcount && targets[j].t == t; j++) { - int i = targets[j].i; + for (; j < (int) s->count && targets[j].t == t; j++) { + int l, i = targets[j].i; if (sum != s->sums[i].sum1) continue; + /* also make sure the two blocks are the same length */ + l = MIN(s->n,len-offset); + if (l != s->sums[i].len) continue; + if (verbose > 3) - fprintf(FINFO,"potential match at %d target=%d %d sum=%08x\n", - (int)offset,j,i,sum); + rprintf(FINFO,"potential match at %.0f target=%d %d sum=%08x\n", + (double)offset,j,i,sum); if (!done_csum2) { - int l = MIN(s->n,len-offset); map = (schar *)map_ptr(buf,offset,l); get_checksum2((char *)map,l,sum2); done_csum2 = 1; @@ -187,6 +210,22 @@ static void hash_search(int f,struct sum_struct *s, false_alarms++; continue; } + + /* we've found a match, but now check to see + if last_i can hint at a better match */ + for (j++; j < (int) s->count && targets[j].t == t; j++) { + int i2 = targets[j].i; + if (i2 == last_i + 1) { + if (sum != s->sums[i2].sum1) break; + if (memcmp(sum2,s->sums[i2].sum2,csum_length) != 0) break; + /* we've found an adjacent match - the RLL coder + will be happy */ + i = i2; + break; + } + } + + last_i = i; matched(f,s,buf,offset,i); offset += s->sums[i].len - 1; @@ -212,7 +251,18 @@ static void hash_search(int f,struct sum_struct *s, } else { --k; } - + + /* By matching early we avoid re-reading the + data 3 times in the case where a token + match comes a long way after last + match. The 3 reads are caused by the + running match, the checksum update and the + literal send. */ + if (offset > last_match && + offset-last_match >= CHUNK_SIZE+s->n && + (end-offset > CHUNK_SIZE)) { + matched(f,s,buf,offset - s->n, -2); + } } while (++offset < end); matched(f,s,buf,len,-1); @@ -220,9 +270,24 @@ static void hash_search(int f,struct sum_struct *s, } -void match_sums(int f,struct sum_struct *s,struct map_struct *buf,OFF_T len) +/** + * Scan through a origin file, looking for sections that match + * checksums from the generator, and transmit either literal or token + * data. + * + * Also calculates the MD4 checksum of the whole file, using the md + * accumulator. This is transmitted with the file as protection + * against corruption on the wire. + * + * @param s Checksums received from the generator. If s->count == + * 0, then there are actually no checksums for this file. + * + * @param len Length of the file to send. + **/ +void match_sums(int f, struct sum_struct *s, struct map_struct *buf, OFF_T len) { char file_sum[MD4_SUM_LENGTH]; + extern int write_batch; /* dw */ last_match = 0; false_alarms = 0; @@ -236,13 +301,19 @@ void match_sums(int f,struct sum_struct *s,struct map_struct *buf,OFF_T len) build_hash_table(s); if (verbose > 2) - fprintf(FINFO,"built hash table\n"); + rprintf(FINFO,"built hash table\n"); hash_search(f,s,buf,len); if (verbose > 2) - fprintf(FINFO,"done hash search\n"); + rprintf(FINFO,"done hash search\n"); } else { + OFF_T j; + /* by doing this in pieces we avoid too many seeks */ + for (j=0;j<(len-CHUNK_SIZE);j+=CHUNK_SIZE) { + int n1 = MIN(CHUNK_SIZE,(len-CHUNK_SIZE)-j); + matched(f,s,buf,j+n1,-2); + } matched(f,s,buf,len,-1); } @@ -250,8 +321,10 @@ void match_sums(int f,struct sum_struct *s,struct map_struct *buf,OFF_T len) if (remote_version >= 14) { if (verbose > 2) - fprintf(FINFO,"sending file_sum\n"); + rprintf(FINFO,"sending file_sum\n"); write_buf(f,file_sum,MD4_SUM_LENGTH); + if (write_batch) /* dw */ + write_batch_delta_file(file_sum, MD4_SUM_LENGTH); } if (targets) { @@ -260,13 +333,13 @@ void match_sums(int f,struct sum_struct *s,struct map_struct *buf,OFF_T len) } if (verbose > 2) - fprintf(FINFO, "false_alarms=%d tag_hits=%d matches=%d\n", + rprintf(FINFO, "false_alarms=%d tag_hits=%d matches=%d\n", false_alarms, tag_hits, matches); total_tag_hits += tag_hits; total_false_alarms += false_alarms; total_matches += matches; - total_data_transfer += data_transfer; + stats.literal_data += data_transfer; } void match_report(void) @@ -274,8 +347,9 @@ void match_report(void) if (verbose <= 1) return; - fprintf(FINFO, - "total: matches=%d tag_hits=%d false_alarms=%d data=%ld\n", + rprintf(FINFO, + "total: matches=%d tag_hits=%d false_alarms=%d data=%.0f\n", total_matches,total_tag_hits, - total_false_alarms,(long)total_data_transfer); + total_false_alarms, + (double)stats.literal_data); }