How can I perform a CopyFile, but also flush the file buffers before the destination handle is closed?

Raymond Chen

A customer wanted to copy a file, and the CopyFile function seemed to do the trick. However, they also wanted to perform a Flush­File­Buffers after the copy completed, but there’s a problem: The Flush­File­Buffers function requires a handle to the file, and the CopyFile function doesn’t give you a handle to the file.

One option is to replace the Copy­File with a manual Read­File/Write­File loop, and perform the Flush­File­Buffers on the destination handle before closing it. However, this means that you give up all the features of Copy­File, like managing alternate data streams, extended attributes, security attributes, and all that other stuff.

Fortunately, there is a way to hook into the file copy operation.

Windows XP added the function Copy­File­Ex which lets you pass a callback function that is called to report on the progress of the file copy, and that callback function is our foot in the door. Among other things, the callback function is given a reason for the callback, as well as the file handles being used for both the source and destination of the copy.

What we can do is detect that the file copy has started copying a new stream. When that happens, we flush the file buffers of the old stream, and then duplicate the new destination handle so we can flush that guy when it finishes.

// WARNING! All error checking has been deleted
// for expository purposes!

BOOL CopyFileWithFlush(
    LPCWSTR sourceFileName,
    LPCWSTR destinationFileName,
    BOOL failIfExists)
{
    HANDLE mostRecentStream = nullptr;

    auto callback = [](
        auto totalSize,
        auto totalBytes,
        auto streamSize,
        auto streamBytes,
        auto streamId,
        auto reason,
        auto sourceHandle,
        auto destHandle,
        auto refdata
    ) -> DWORD
    {
        auto& mostRecentStream = *reinterpret_cast<HANDLE*>(refdata);
        if (reason == CALLBACK_STREAM_SWITCH) {
            if (mostRecentStream) {
                FlushFileBuffers(mostRecentStream);
                CloseHandle(mostRecentStream);
            }
            DuplicateHandle(
                GetCurrentProcess(), destHandle,
                GetCurrentProcess(), &mostRecentStream,
                0, false, DUPLICATE_SAME_ACCESS);
        }
        return PROGRESS_CONTINUE;
    };

    auto result = CopyFileEx(
        sourceFileName,
        destinationFileName,
        callback,
        &mostRecentStream,
        nullptr,
        failIfExists ? COPY_FILE_FAIL_IF_EXISTS : 0);

    if (mostRecentStream) {
        if (result) {
            FlushFileBuffers(mostRecentStream);
        }
        CloseHandle(mostRecentStream);
    }

    return result;
}

This is a lot of code, but it’s actually not doing much.

We take action when we are told that the copy has started a new stream. First, we flush out the old stream, if any. We then close the old stream handle and duplicate the new destination handle, so that we can flush this new stream when it is finished.

If the copy completes successfully, we perform one last flush to flush out the last stream.

The nonsense with having to remember the most recent stream is a workaround for the fact that Copy­File­Ex does not notify you when it finishes the old stream. It only notifies you when it starts a new stream, so you have to infer that the old stream ended either when a new stream starts or when the entire operation completes.

If you can assume a minimum supported operating system of Windows 8, then you can use the newer Copy­File2 function, which calls you back when a stream finishes:

// WARNING! All error checking has been deleted
// for expository purposes!

HRESULT CopyFileWithFlush(
    LPCWSTR sourceFileName,
    LPCWSTR destinationFileName,
    BOOL failIfExists)
{
    COPYFILE2_EXTENDED_PARAMETERS params{ sizeof(params) };
    params.dwCopyFlags = failIfExists ? COPY_FILE_FAIL_IF_EXISTS : 0;
    params.pProgressRoutine = [](auto message, auto) -> COPYFILE2_MESSAGE_ACTION
    {
        if (message->Type == COPYFILE2_CALLBACK_STREAM_FINISHED)
        {
            FlushFileBuffers(message->Info.StreamFinished.hDestinationFile);
        }
        return COPYFILE2_PROGRESS_CONTINUE;
    };

    return CopyFile2(sourceFileName, destinationFileName, &params);
}

If the callback is telling us that we just finished copying a stream, we sneak in and flush the buffers associated with that handle.

This technique can be extended to cover other operations you want to perform during the file copy. You can use the file handles of the source and destination to perform additional operations before everything gets closed.

7 comments

Discussion is closed. Login to edit/delete existing comments.

  • Gunnar Dalsnes 0

    CopyFile v1
    CopyFileEx v2
    CopyFile2 v3
    Maybe it should been CopyFile3?
    Does this mean Ex is old style and a numer is new style?

    • Dmitry 0

      You start counting with a wrong number. It’s v0, v1 and v2. Their development ecosystem is derived from C and its descendants, so stuff not obvious to a normal human, like this one, is not to be surprised at 🙂

  • skSdnW 2

    “Windows XP added the function Copy­File­Ex” except that it was actually added way back in NT 4. You cannot trust the online documentation!

    • Georg Rottensteiner 2

      This. So much this.

      I don’t know who had the brainfart to go over ALL the MSDN documentation and rip out the actual availability parts and replace them with the earliest Windows version still in active support.

      • Dmitry 1

        I triple that! Destructive work someone has definitely been paid for.

      • Nick 0

        Oceania has always been at war with Eastasia.

      • Jan RingoÅ¡ 0

        And it’s especially infuriating when they got it wrong.

Feedback usabilla icon