diff-delta: bound hash list length to avoid O(m*n) behavior

The diff-delta code can exhibit O(m*n) behavior with some patological
data set where most hash entries end up in the same hash bucket.

To prevent this, a limit is imposed to the number of entries that can
exist in the same hash bucket.

Because of the above the code is a tiny bit more expensive on average,
even if some small optimizations were added as well to atenuate the
overhead. But the problematic samples used to diagnoze the issue are now
orders of magnitude less expensive to process with only a slight loss in
compression.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>

1 files changed, 71 insertions, 30 deletions

diff --git a/diff-delta.c b/diff-delta.c
index 2ed5984..aaee7be 100644
--- a/diff-delta.c
+++ b/diff-delta.c
@@ -40,17 +40,18 @@ struct index {
 
 static struct index ** delta_index(const unsigned char *buf,
 				   unsigned long bufsize,
+				   unsigned long trg_bufsize,
 				   unsigned int *hash_shift)
 {
-	unsigned int hsize, hshift, entries, blksize, i;
+	unsigned int i, hsize, hshift, hlimit, entries, *hash_count;
 	const unsigned char *data;
 	struct index *entry, **hash;
 	void *mem;
 
 	/* determine index hash size */
-	entries = (bufsize + BLK_SIZE - 1) / BLK_SIZE;
+	entries = bufsize  / BLK_SIZE;
 	hsize = entries / 4;
-	for (i = 4; (1 << i) < hsize && i < 16; i++);
+	for (i = 4; (1 << i) < hsize && i < 31; i++);
 	hsize = 1 << i;
 	hshift = 32 - i;
 	*hash_shift = hshift;
@@ -63,20 +64,62 @@ static struct index ** delta_index(const unsigned char *buf,
 	entry = mem + hsize * sizeof(*hash);
 	memset(hash, 0, hsize * sizeof(*hash));
 
-	/* then populate it */
+	/* allocate an array to count hash entries */
+	hash_count = calloc(hsize, sizeof(*hash_count));
+	if (!hash_count) {
+		free(hash);
+		return NULL;
+	}
+
+	/* then populate the index */
 	data = buf + entries * BLK_SIZE - BLK_SIZE;
-	blksize = bufsize - (data - buf);
 	while (data >= buf) {
-		unsigned int val = adler32(0, data, blksize);
+		unsigned int val = adler32(0, data, BLK_SIZE);
 		i = HASH(val, hshift);
 		entry->ptr = data;
 		entry->val = val;
 		entry->next = hash[i];
 		hash[i] = entry++;
-		blksize = BLK_SIZE;
+		hash_count[i]++;
 		data -= BLK_SIZE;
  	}
 
+	/*
+	 * Determine a limit on the number of entries in the same hash
+	 * bucket.  This guard us against patological data sets causing
+	 * really bad hash distribution with most entries in the same hash
+	 * bucket that would bring us to O(m*n) computing costs (m and n
+	 * corresponding to reference and target buffer sizes).
+	 *
+	 * The more the target buffer is large, the more it is important to
+	 * have small entry lists for each hash buckets.  With such a limit
+	 * the cost is bounded to something more like O(m+n).
+	 */
+	hlimit = (1 << 26) / trg_bufsize;
+	if (hlimit < 4*BLK_SIZE)
+		hlimit = 4*BLK_SIZE;
+
+	/*
+	 * Now make sure none of the hash buckets has more entries than
+	 * we're willing to test.  Otherwise we cull the entry list
+	 * uniformly to still preserve a good repartition across
+	 * the reference buffer.
+	 */
+	for (i = 0; i < hsize; i++) {
+		if (hash_count[i] < hlimit)
+			continue;
+		entry = hash[i];
+		do {
+			struct index *keep = entry;
+			int skip = hash_count[i] / hlimit / 2;
+			do {
+				entry = entry->next;
+			} while(--skip && entry);
+			keep->next = entry;
+		} while(entry);
+	}
+	free(hash_count);
+
 	return hash;
 }
 
@@ -100,7 +143,7 @@ void *diff_delta(void *from_buf, unsigned long from_size,
 
 	if (!from_size || !to_size)
 		return NULL;
-	hash = delta_index(from_buf, from_size, &hash_shift);
+	hash = delta_index(from_buf, from_size, to_size, &hash_shift);
 	if (!hash)
 		return NULL;
 
@@ -141,29 +184,27 @@ void *diff_delta(void *from_buf, unsigned long from_size,
 
 	while (data < top) {
 		unsigned int moff = 0, msize = 0;
-		unsigned int blksize = MIN(top - data, BLK_SIZE);
-		unsigned int val = adler32(0, data, blksize);
-		i = HASH(val, hash_shift);
-		for (entry = hash[i]; entry; entry = entry->next) {
-			const unsigned char *ref = entry->ptr;
-			const unsigned char *src = data;
-			unsigned int ref_size = ref_top - ref;
-			if (entry->val != val)
-				continue;
-			if (ref_size > top - src)
-				ref_size = top - src;
-			while (ref_size && *src++ == *ref) {
-				ref++;
-				ref_size--;
-			}
-			ref_size = ref - entry->ptr;
-			if (ref_size > msize) {
-				/* this is our best match so far */
-				moff = entry->ptr - ref_data;
-				msize = ref_size;
-				if (msize >= 0x10000) {
-					msize = 0x10000;
+		if (data + BLK_SIZE <= top) {
+			unsigned int val = adler32(0, data, BLK_SIZE);
+			i = HASH(val, hash_shift);
+			for (entry = hash[i]; entry; entry = entry->next) {
+				const unsigned char *ref = entry->ptr;
+				const unsigned char *src = data;
+				unsigned int ref_size = ref_top - ref;
+				if (entry->val != val)
+					continue;
+				if (ref_size > top - src)
+					ref_size = top - src;
+				if (ref_size > 0x10000)
+					ref_size = 0x10000;
+				if (ref_size <= msize)
 					break;
+				while (ref_size-- && *src++ == *ref)
+					ref++;
+				if (msize < ref - entry->ptr) {
+					/* this is our best match so far */
+					msize = ref - entry->ptr;
+					moff = entry->ptr - ref_data;
 				}
 			}
 		}