diff --git a/src/compress.h b/src/compress.h
index c8f2389f2..2b7c84cd2 100644
--- a/src/compress.h
+++ b/src/compress.h
@@ -7,7 +7,7 @@
 
 namespace maxcso {
 
-static const char *VERSION = "1.4.3";
+static const char *VERSION = "1.4.4";
 
 struct Task;
 
diff --git a/src/input.cpp b/src/input.cpp
index 2048423af..bd14e4867 100644
--- a/src/input.cpp
+++ b/src/input.cpp
@@ -79,7 +79,7 @@ void Input::DetectFormat() {
 					++csoBlockShift_;
 				}
 
-				const uint32_t sectors = static_cast<uint32_t>(size_ >> csoBlockShift_);
+				const uint32_t sectors = static_cast<uint32_t>(SizeAligned() >> csoBlockShift_);
 				csoIndex_ = new uint32_t[sectors + 1];
 				const unsigned int bytes = (sectors + 1) * sizeof(uint32_t);
 				const uv_buf_t buf = uv_buf_init(reinterpret_cast<char *>(csoIndex_), bytes);
@@ -383,4 +383,8 @@ bool Input::DecompressSectorLZ4(uint8_t *dst, const uint8_t *src, int dstSize, s
 	return true;
 }
 
+inline int64_t Input::SizeAligned() {
+	return size_ + csoBlockSize_ - 1;
+}
+
 };
\ No newline at end of file
diff --git a/src/input.h b/src/input.h
index 65b690688..17b49a398 100644
--- a/src/input.h
+++ b/src/input.h
@@ -24,6 +24,7 @@ class Input {
 	void SetupCache(uint32_t minSize);
 	void ReadSector();
 	void EnqueueDecompressSector(uint8_t *src, uint32_t len, uint32_t offset, bool isLZ4);
+	inline int64_t SizeAligned();
 
 	enum FileType {
 		UNKNOWN,
diff --git a/src/output.cpp b/src/output.cpp
index 8c65e984d..00ea0b4eb 100644
--- a/src/output.cpp
+++ b/src/output.cpp
@@ -44,7 +44,7 @@ void Output::SetFile(uv_file file, int64_t srcSize, uint32_t blockSize, CSOForma
 		++blockShift_;
 	}
 
-	const uint32_t sectors = static_cast<uint32_t>(srcSize >> blockShift_);
+	const uint32_t sectors = static_cast<uint32_t>((srcSize + blockSize_ - 1) >> blockShift_);
 	// Start after the header and index, which we'll fill in later.
 	index_ = new uint32_t[sectors + 1];
 	// Start after the end of the index data and header.
@@ -120,7 +120,7 @@ void Output::Enqueue(int64_t pos, uint8_t *buffer) {
 	if (blockSize_ != SECTOR_SIZE && pos + SECTOR_SIZE >= srcSize_) {
 		// Our src may not be aligned to the blockSize_, so this sector might never wake up.
 		// So let's send in some padding if needed.
-		const int64_t paddedSize = (srcSize_ + blockSize_ - 1) & ~static_cast<int64_t>(blockSize_ - 1);
+		const int64_t paddedSize = SrcSizeAligned() & ~static_cast<int64_t>(blockSize_ - 1);
 		for (int64_t padPos = srcSize_; padPos < paddedSize; padPos += SECTOR_SIZE) {
 			// Sector takes ownership, so we need a new one each time.
 			uint8_t *padBuffer = pool.Alloc();
@@ -220,7 +220,7 @@ void Output::HandleReadySector(Sector *sector) {
 	// If we're working on the last sectors, then the index is ready to write.
 	if (nextPos >= srcSize_) {
 		// Update the final index entry.
-		const int32_t s = static_cast<int32_t>(srcSize_ >> blockShift_);
+		const int32_t s = static_cast<int32_t>(SrcSizeAligned() >> blockShift_);
 		index_[s] = static_cast<int32_t>(dstPos >> indexShift_);
 
 		state_ |= STATE_INDEX_READY;
@@ -304,7 +304,7 @@ void Output::Flush() {
 	header->unused[0] = 0;
 	header->unused[1] = 0;
 
-	const uint32_t sectors = static_cast<uint32_t>(srcSize_ >> blockShift_);
+	const uint32_t sectors = static_cast<uint32_t>(SrcSizeAligned() >> blockShift_);
 
 	uv_buf_t bufs[2];
 	bufs[0] = uv_buf_init(reinterpret_cast<char *>(header), sizeof(CSOHeader));
@@ -328,4 +328,8 @@ void Output::CheckFinish() {
 	}
 }
 
+inline int64_t Output::SrcSizeAligned() {
+	return srcSize_ + blockSize_ - 1;
+}
+
 };
diff --git a/src/output.h b/src/output.h
index 55a2b446e..f27eced37 100644
--- a/src/output.h
+++ b/src/output.h
@@ -31,6 +31,7 @@ class Output {
 	int32_t Align(int64_t &pos);
 	void HandleReadySector(Sector *sector);
 	bool ShouldCompress(int64_t pos, uint8_t *buffer);
+	inline int64_t SrcSizeAligned();
 
 	enum State {
 		STATE_INIT = 0x00,