/* * Block matching used by the file-transfer code. * * Copyright (C) 1996 Andrew Tridgell * Copyright (C) 1996 Paul Mackerras * Copyright (C) 2003-2023 Wayne Davison * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 3 of the License, or * (at your option) any later version. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License along * with this program; if not, visit the http://fsf.org website. */ #include "rsync.h"#include "inums.h" extern int checksum_seed;extern int append_mode; extern struct name_num_item *xfer_sum_nni;extern int xfer_sum_len; int updating_basis_file;char sender_file_sum[MAX_DIGEST_LEN]; static int false_alarms;static int hash_hits;static int matches;static int64 data_transfer; static int total_false_alarms;static int total_hash_hits;static int total_matches; extern struct stats stats; #define TRADITIONAL_TABLESIZE (1<<16) /* The maximum number of same-weak-checksum candidates we will compare * against at a single file offset before giving up and rolling forward a * byte.  A weak checksum that collides thousands of times (very common in * disk/VM images, which contain large runs of identical blocks) would * otherwise turn hash_search()'s inner loop into an O(file_size * * chain_length) scan, pegging a CPU at 100% for hours with no apparent * progress (issue #217). * * Concretely, a synthetic 40000-block basis whose blocks all share one weak * checksum took ~18.4s to sync a 60KB source on a modern x86_64 box before * this cap and ~0.7s after it -- and the unbounded cost grows with the * square of the file size, which is what produced the multi-hour "hangs" * reported against real multi-GB images. * * Capping the per-offset work keeps the search bounded; any block we skip * over is simply sent as literal data, so the result is always correct -- * only the transfer size is (slightly) affected.  This is purely a * sender-side search limit: it changes no checksum, emitted byte, or * protocol field, so a capped sender interoperates with any receiver. */#ifndef MAX_CHAIN_LEN#define MAX_CHAIN_LEN 1024#endif static uint32 tablesize;static int32 *hash_table; #define SUM2HASH2(s1,s2) (((s1) + (s2)) & 0xFFFF)#define SUM2HASH(sum) SUM2HASH2((sum)&0xFFFF,(sum)>>16) #define BIG_SUM2HASH(sum) ((sum)%tablesize) static void build_hash_table(struct sum_struct *s){	static uint32 alloc_size;	int32 i; 	/* Dynamically calculate the hash table size so that the hash load	 * for big files is about 80%.  A number greater than the traditional	 * size must be odd or s2 will not be able to span the entire set. */	tablesize = (uint32)(s->count/8) * 10 + 11;	if (tablesize < TRADITIONAL_TABLESIZE)		tablesize = TRADITIONAL_TABLESIZE;	if (tablesize > alloc_size || tablesize < alloc_size - 16*1024) {		if (hash_table)			free(hash_table);		hash_table = new_array(int32, tablesize);		alloc_size = tablesize;	} 	memset(hash_table, 0xFF, tablesize * sizeof hash_table[0]); 	if (tablesize == TRADITIONAL_TABLESIZE) {		for (i = 0; i < s->count; i++) {			uint32 t = SUM2HASH(s->sums[i].sum1);			s->sums[i].chain = hash_table[t];			hash_table[t] = i;		}	} else {		for (i = 0; i < s->count; i++) {			uint32 t = BIG_SUM2HASH(s->sums[i].sum1);			s->sums[i].chain = hash_table[t];			hash_table[t] = i;		}	}}  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! * * If i >= 0, the number of a matched token.  If < 0, indicates we have * only literal data.  A -1 will send a 0-token-int too, and a -2 sends * only literal data, w/o any token-int. */static void matched(int f, struct sum_struct *s, struct map_struct *buf, OFF_T offset, int32 i){	int32 n = (int32)(offset - last_match); /* max value: block_size (int32) */	int32 j; 	if (DEBUG_GTE(DELTASUM, 2) && i >= 0) {		rprintf(FINFO,			"match at %s last_match=%s j=%d len=%ld n=%ld\n",			big_num(offset), big_num(last_match), i,			(long)s->sums[i].len, (long)n);	} 	send_token(f, i, buf, last_match, n, i < 0 ? 0 : s->sums[i].len);	data_transfer += n; 	if (i >= 0) {		stats.matched_data += s->sums[i].len;		n += s->sums[i].len;	} 	for (j = 0; j < n; j += CHUNK_SIZE) {		int32 n1 = MIN(CHUNK_SIZE, n - j);		sum_update(map_ptr(buf, last_match + j, n1), n1);	} 	if (i >= 0)		last_match = offset + s->sums[i].len;	else		last_match = offset; 	if (buf && INFO_GTE(PROGRESS, 1))		show_progress(last_match, buf->file_size);}  static void hash_search(int f,struct sum_struct *s,			struct map_struct *buf, OFF_T len){	OFF_T offset, aligned_offset, end;	int32 k, want_i, aligned_i, backup;	char sum2[MAX_DIGEST_LEN];	uint32 s1, s2, sum;	int more;	schar *map; 	// prevent possible memory leaks	memset(sum2, 0, sizeof sum2); 	/* want_i is used to encourage adjacent matches, allowing the RLL	 * coding of the output to work more efficiently. */	want_i = 0; 	if (DEBUG_GTE(DELTASUM, 2)) {		rprintf(FINFO, "hash search b=%ld len=%s\n",			(long)s->blength, big_num(len));	} 	k = (int32)MIN(len, (OFF_T)s->blength); 	map = (schar *)map_ptr(buf, 0, k); 	sum = get_checksum1((char *)map, k);	s1 = sum & 0xFFFF;	s2 = sum >> 16;	if (DEBUG_GTE(DELTASUM, 3))		rprintf(FINFO, "sum=%.8x k=%ld\n", sum, (long)k); 	offset = aligned_offset = aligned_i = 0; 	end = len + 1 - s->sums[s->count-1].len; 	if (DEBUG_GTE(DELTASUM, 3)) {		rprintf(FINFO, "hash search s->blength=%ld len=%s count=%s\n",			(long)s->blength, big_num(len), big_num(s->count));	} 	do {		int done_csum2 = 0;		uint32 hash_entry;		int32 i, *prev;		int32 chain_len = 0; 		if (DEBUG_GTE(DELTASUM, 4)) {			rprintf(FINFO, "offset=%s sum=%04x%04x\n",				big_num(offset), s2 & 0xFFFF, s1 & 0xFFFF);		} 		if (tablesize == TRADITIONAL_TABLESIZE) {			hash_entry = SUM2HASH2(s1,s2);			if ((i = hash_table[hash_entry]) < 0)				goto null_hash;			sum = (s1 & 0xffff) | (s2 << 16);		} else {			sum = (s1 & 0xffff) | (s2 << 16);			hash_entry = BIG_SUM2HASH(sum);			if ((i = hash_table[hash_entry]) < 0)				goto null_hash;		}		prev = &hash_table[hash_entry]; 		hash_hits++;		do {			int32 l; 			/* When updating in-place, the chunk's offset must be			 * either >= our offset or identical data at that offset.			 * Remove any bypassed entries that we can never use. */			if (updating_basis_file && s->sums[i].offset < offset			 && !(s->sums[i].flags & SUMFLG_SAME_OFFSET)) {				*prev = s->sums[i].chain;				continue;			}			prev = &s->sums[i].chain; 			if (sum != s->sums[i].sum1)				continue; 			/* Bound the work spent on a single pathological hash			 * bucket.  If this weak checksum matches more than			 * MAX_CHAIN_LEN records, stop scanning and treat this			 * offset as a non-match (issue #217).  The skipped data			 * is sent literally, never corrupted. */			if (++chain_len > MAX_CHAIN_LEN)				break; 			/* also make sure the two blocks are the same length */			l = (int32)MIN((OFF_T)s->blength, len-offset);			if (l != s->sums[i].len)				continue; 			if (DEBUG_GTE(DELTASUM, 3)) {				rprintf(FINFO,					"potential match at %s i=%ld sum=%08x\n",					big_num(offset), (long)i, sum);			} 			if (!done_csum2) {				map = (schar *)map_ptr(buf,offset,l);				get_checksum2((char *)map,l,sum2);				done_csum2 = 1;			} 			if (memcmp(sum2, sum2_at(s, i), s->s2length) != 0) {				false_alarms++;				continue;			} 			/* When updating in-place, the best possible match is			 * one with an identical offset, so we prefer that over			 * the adjacent want_i optimization. */			if (updating_basis_file) {				/* All the generator's chunks start at blength boundaries. */				while (aligned_offset < offset) {					aligned_offset += s->blength;					aligned_i++;				}				if ((offset == aligned_offset				  || (sum == 0 && l == s->blength && aligned_offset + l <= len))				 && aligned_i < s->count) {					if (i != aligned_i) {						if (sum != s->sums[aligned_i].sum1						 || l != s->sums[aligned_i].len						 || memcmp(sum2, sum2_at(s, aligned_i), s->s2length) != 0)							goto check_want_i;						i = aligned_i;					}					if (offset != aligned_offset) {						/* We've matched some zeros in a spot that is also zeros						 * further along in the basis file, if we find zeros ahead						 * in the sender's file, we'll output enough literal data						 * to re-align with the basis file, and get back to seeking						 * instead of writing. */						backup = (int32)(aligned_offset - last_match);						if (backup < 0)							backup = 0;						map = (schar *)map_ptr(buf, aligned_offset - backup, l + backup)						    + backup;						sum = get_checksum1((char *)map, l);						if (sum != s->sums[i].sum1)							goto check_want_i;						get_checksum2((char *)map, l, sum2);						if (memcmp(sum2, sum2_at(s, i), s->s2length) != 0)							goto check_want_i;						/* OK, we have a re-alignment match.  Bump the offset						 * forward to the new match point. */						offset = aligned_offset;					}					/* This identical chunk is in the same spot in the old and new file. */					s->sums[i].flags |= SUMFLG_SAME_OFFSET;					want_i = i;				}			} 		  check_want_i:			/* we've found a match, but now check to see			 * if want_i can hint at a better match. */			if (i != want_i && want_i < s->count			 && (!updating_basis_file || s->sums[want_i].offset >= offset			  || s->sums[want_i].flags & SUMFLG_SAME_OFFSET)			 && sum == s->sums[want_i].sum1			 && l == s->sums[want_i].len			 && memcmp(sum2, sum2_at(s, want_i), s->s2length) == 0) {				/* we've found an adjacent match - the RLL coder				 * will be happy */				i = want_i;			}			want_i = i + 1; 			matched(f,s,buf,offset,i);			offset += s->sums[i].len - 1;			k = (int32)MIN((OFF_T)s->blength, len-offset);			map = (schar *)map_ptr(buf, offset, k);			sum = get_checksum1((char *)map, k);			s1 = sum & 0xFFFF;			s2 = sum >> 16;			matches++;			break;		} while ((i = s->sums[i].chain) >= 0); 	  null_hash:		backup = (int32)(offset - last_match);		/* We sometimes read 1 byte prior to last_match... */		if (backup < 0)			backup = 0; 		/* Trim off the first byte from the checksum */		more = offset + k < len;		map = (schar *)map_ptr(buf, offset - backup, k + more + backup) + backup;		s1 -= map[0] + CHAR_OFFSET;		s2 -= k * (map[0]+CHAR_OFFSET); 		/* Add on the next byte (if there is one) to the checksum */		if (more) {			s1 += map[k] + CHAR_OFFSET;			s2 += s1;		} 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 (backup >= s->blength+CHUNK_SIZE && end-offset > CHUNK_SIZE)			matched(f, s, buf, offset - s->blength, -2);	} while (++offset < end); 	matched(f, s, buf, len, -1);	map_ptr(buf, len-1, 1);}  /** * 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 <tt>s->count == * 0</tt>, 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){	last_match = 0;	false_alarms = 0;	hash_hits = 0;	matches = 0;	data_transfer = 0; 	sum_init(xfer_sum_nni, checksum_seed); 	if (append_mode > 0) {		if (s->flength > len) {			/* A hostile or confused peer can claim a verified-prefix			 * length that exceeds what we have on disk -- including			 * for an empty local file, where buf is NULL and the			 * map_ptr() calls below would dereference it.  Clamp to			 * what we can actually read. */			s->flength = len;		}		if (append_mode == 2) {			OFF_T j = 0;			for (j = CHUNK_SIZE; j < s->flength; j += CHUNK_SIZE) {				if (buf && INFO_GTE(PROGRESS, 1))					show_progress(last_match, buf->file_size);				sum_update(map_ptr(buf, last_match, CHUNK_SIZE),					   CHUNK_SIZE);				last_match = j;			}			if (last_match < s->flength) {				int32 n = (int32)(s->flength - last_match);				if (buf && INFO_GTE(PROGRESS, 1))					show_progress(last_match, buf->file_size);				sum_update(map_ptr(buf, last_match, n), n);			}		}		last_match = s->flength;		s->count = 0;	} 	if (len > 0 && s->count > 0) {		build_hash_table(s); 		if (DEBUG_GTE(DELTASUM, 2))			rprintf(FINFO,"built hash table\n"); 		hash_search(f, s, buf, len); 		if (DEBUG_GTE(DELTASUM, 2))			rprintf(FINFO,"done hash search\n");	} else {		OFF_T j;		/* by doing this in pieces we avoid too many seeks */		for (j = last_match + CHUNK_SIZE; j < len; j += CHUNK_SIZE)			matched(f, s, buf, j, -2);		matched(f, s, buf, len, -1);	} 	sum_end(sender_file_sum); 	/* If we had a read error, send a bad checksum.  We use all bits	 * off as long as the checksum doesn't happen to be that, in	 * which case we turn the last 0 bit into a 1. */	if (buf && buf->status != 0) {		int i;		for (i = 0; i < xfer_sum_len && sender_file_sum[i] == 0; i++) {}		memset(sender_file_sum, 0, xfer_sum_len);		if (i == xfer_sum_len)			sender_file_sum[i-1]++;	} 	if (DEBUG_GTE(DELTASUM, 2))		rprintf(FINFO,"sending file_sum\n");	write_buf(f, sender_file_sum, xfer_sum_len); 	if (DEBUG_GTE(DELTASUM, 2)) {		rprintf(FINFO, "false_alarms=%d hash_hits=%d matches=%d\n",			false_alarms, hash_hits, matches);	} 	total_hash_hits += hash_hits;	total_false_alarms += false_alarms;	total_matches += matches;	stats.literal_data += data_transfer;} void match_report(void){	if (!DEBUG_GTE(DELTASUM, 1))		return; 	rprintf(FINFO,		"total: matches=%d  hash_hits=%d  false_alarms=%d data=%s\n",		total_matches, total_hash_hits, total_false_alarms,		big_num(stats.literal_data));}