fail gracefully on large inputs

1 files changed, 9 insertions, 5 deletions

diff --git a/fuzzy.c b/fuzzy.c
index d359b30..88af850 100644
--- a/fuzzy.c
+++ b/fuzzy.c
@@ -187,16 +187,18 @@ static int ssdeep_engine_step(struct ssdeep_context *self, unsigned char c) {
 			 * automatically false for all further bs. I.e. if
 			 * h === -1 (mod 2*bs) then h === -1 (mod bs). */
 			break;
-		/* We don't need bs until next iteration. Update now, so we
-		 * don't forget about it. */
-		bs *= 2;
 		/* We have hit a reset point. We now emit hashes which are
 		 * based on all characters in the piece of the message between
 		 * the last reset point and this one */
-		if(unlikely(0 == bh->dlen)) /* Can only happen 30 times. */
+		if(unlikely(0 == bh->dlen && bs < UINT32_MAX / 2)) {
+			/* Can only happen 30 times. */
 			/* First step for this blocksize. Clone next. */
 			if(blockhash_fork(bh) < 0)
 				return -1;
+		}
+		/* We don't need bs until next iteration. Update now, so we
+		 * don't forget about it. */
+		bs *= 2;
 		if(bh->dlen < SPAMSUM_LENGTH - 1) {
 			/* We can have a problem with the tail overflowing. The
 			 * easiest way to cope with this is to only reset the
@@ -241,7 +243,9 @@ static int ssdeep_digest(const struct ssdeep_context *self,
 	/* Initial blocksize guess. */
 	while((size_t)bs * SPAMSUM_LENGTH < self->total_size) {
 		bs *= 2;
-		assert(bh->next != NULL);
+		if(bh->next == NULL)
+			/* The input exceeds data types. */
+			return -1;
 		bh = bh->next;
 	}
 	/* Adapt blocksize guess to actual digest length. */