如何在大型稀疏文件上创建快速高效的文件流写入

Question

我有一个应用程序可以在多个段中写入大文件。我使用 FileStream.Seek 来定位每个文件。看来，当我在稀疏文件中的较深位置调用 FileStream.Write 时，写入会触发对所有先前字节的“回填”操作（写入 0），这很慢。

有没有更有效的方法来处理这种情况？

下面的代码演示了这个问题。在我的机器上，初始写入大约需要 370 毫秒。

    public void WriteToStream()
    {
        DateTime dt;
        using (FileStream fs = File.Create("C:\\testfile.file"))
        {   
            fs.SetLength(1024 * 1024 * 100);
            fs.Seek(-1, SeekOrigin.End);
            dt = DateTime.Now;
            fs.WriteByte(255);              
        }

        Console.WriteLine(@"WRITE MS: " + DateTime.Now.Subtract(dt).TotalMilliseconds.ToString());
    }

Answer 1

NTFS 确实支持Sparse Files，但是如果不调用一些本机方法，在 .net 中就无法做到这一点。

将文件标记为稀疏文件并不难，只要知道一旦文件被标记为稀疏文件，它就永远无法转换回非稀疏文件，除非将整个文件复制到一个新的非稀疏文件中文件。

用法示例

class Program
{
    [DllImport("Kernel32.dll", SetLastError = true, CharSet = CharSet.Auto)]
    private static extern bool DeviceIoControl(
        SafeFileHandle hDevice,
        int dwIoControlCode,
        IntPtr InBuffer,
        int nInBufferSize,
        IntPtr OutBuffer,
        int nOutBufferSize,
        ref int pBytesReturned,
        [In] ref NativeOverlapped lpOverlapped
    );

    static void MarkAsSparseFile(SafeFileHandle fileHandle)
    {
        int bytesReturned = 0;
        NativeOverlapped lpOverlapped = new NativeOverlapped();
        bool result =
            DeviceIoControl(
                fileHandle,
                590020, //FSCTL_SET_SPARSE,
                IntPtr.Zero,
                0,
                IntPtr.Zero,
                0,
                ref bytesReturned,
                ref lpOverlapped);
        if(result == false)
            throw new Win32Exception();
    }

    static void Main()
    {
        //Use stopwatch when benchmarking, not DateTime
        Stopwatch stopwatch = new Stopwatch();

        stopwatch.Start();
        using (FileStream fs = File.Create(@"e:\Test\test.dat"))
        {
            MarkAsSparseFile(fs.SafeFileHandle);

            fs.SetLength(1024 * 1024 * 100);
            fs.Seek(-1, SeekOrigin.End);
            fs.WriteByte(255);
        }
        stopwatch.Stop();

        //Returns 2 for sparse files and 1127 for non sparse
        Console.WriteLine(@"WRITE MS: " + stopwatch.ElapsedMilliseconds); 
    }
}

---

一旦一个文件被标记为稀疏，它现在的行为就像你一样，除了它在 cmets 中的行为。您无需写入字节即可将文件标记为设定的大小。

static void Main()
{
    string filename = @"e:\Test\test.dat";

    using (FileStream fs = new FileStream(filename, FileMode.Create))
    {
        MarkAsSparseFile(fs.SafeFileHandle);

        fs.SetLength(1024 * 1024 * 25);
    }
}

Answer 2

下面是一些使用稀疏文件的代码：

using System;
using System.ComponentModel;
using System.IO;
using System.Runtime.InteropServices;
using System.Text;
using System.Threading;

using Microsoft.Win32.SafeHandles;

public static class SparseFiles
{
    private const int FILE_SUPPORTS_SPARSE_FILES = 64;

    private const int FSCTL_SET_SPARSE = 0x000900c4;

    private const int FSCTL_SET_ZERO_DATA = 0x000980c8;

    public static void MakeSparse(this FileStream fileStream)
    {
        var bytesReturned = 0;
        var lpOverlapped = new NativeOverlapped();
        var result = DeviceIoControl(
            fileStream.SafeFileHandle, 
            FSCTL_SET_SPARSE, 
            IntPtr.Zero, 
            0, 
            IntPtr.Zero, 
            0, 
            ref bytesReturned, 
            ref lpOverlapped);

        if (!result)
        {
            throw new Win32Exception();
        }
    }

    public static void SetSparseRange(this FileStream fileStream, long fileOffset, long length)
    {
        var fzd = new FILE_ZERO_DATA_INFORMATION();
        fzd.FileOffset = fileOffset;
        fzd.BeyondFinalZero = fileOffset + length;
        var lpOverlapped = new NativeOverlapped();
        var dwTemp = 0;

        var result = DeviceIoControl(
            fileStream.SafeFileHandle, 
            FSCTL_SET_ZERO_DATA, 
            ref fzd, 
            Marshal.SizeOf(typeof(FILE_ZERO_DATA_INFORMATION)), 
            IntPtr.Zero, 
            0, 
            ref dwTemp, 
            ref lpOverlapped);
        if (!result)
        {
            throw new Win32Exception();
        }
    }

    public static bool SupportedOnVolume(string filename)
    {
        var targetVolume = Path.GetPathRoot(filename);
        var fileSystemName = new StringBuilder(300);
        var volumeName = new StringBuilder(300);
        uint lpFileSystemFlags;
        uint lpVolumeSerialNumber;
        uint lpMaxComponentLength;

        var result = GetVolumeInformationW(
            targetVolume, 
            volumeName, 
            (uint)volumeName.Capacity, 
            out lpVolumeSerialNumber, 
            out lpMaxComponentLength, 
            out lpFileSystemFlags, 
            fileSystemName, 
            (uint)fileSystemName.Capacity);
        if (!result)
        {
            throw new Win32Exception();
        }

        return (lpFileSystemFlags & FILE_SUPPORTS_SPARSE_FILES) == FILE_SUPPORTS_SPARSE_FILES;
    }

    [DllImport("Kernel32.dll", SetLastError = true, CharSet = CharSet.Auto)]
    [return: MarshalAs(UnmanagedType.Bool)]
    private static extern bool DeviceIoControl(
        SafeFileHandle hDevice, 
        int dwIoControlCode, 
        IntPtr InBuffer, 
        int nInBufferSize, 
        IntPtr OutBuffer, 
        int nOutBufferSize, 
        ref int pBytesReturned, 
        [In] ref NativeOverlapped lpOverlapped);

    [DllImport("Kernel32.dll", SetLastError = true, CharSet = CharSet.Auto)]
    [return: MarshalAs(UnmanagedType.Bool)]
    private static extern bool DeviceIoControl(
        SafeFileHandle hDevice, 
        int dwIoControlCode, 
        ref FILE_ZERO_DATA_INFORMATION InBuffer, 
        int nInBufferSize, 
        IntPtr OutBuffer, 
        int nOutBufferSize, 
        ref int pBytesReturned, 
        [In] ref NativeOverlapped lpOverlapped);

    [DllImport("kernel32.dll", EntryPoint = "GetVolumeInformationW")]
    [return: MarshalAs(UnmanagedType.Bool)]
    private static extern bool GetVolumeInformationW(
        [In] [MarshalAs(UnmanagedType.LPWStr)] string lpRootPathName, 
        [Out] [MarshalAs(UnmanagedType.LPWStr)] StringBuilder lpVolumeNameBuffer, 
        uint nVolumeNameSize, 
        out uint lpVolumeSerialNumber, 
        out uint lpMaximumComponentLength, 
        out uint lpFileSystemFlags, 
        [Out] [MarshalAs(UnmanagedType.LPWStr)] StringBuilder lpFileSystemNameBuffer, 
        uint nFileSystemNameSize);

    [StructLayout(LayoutKind.Sequential)]
    private struct FILE_ZERO_DATA_INFORMATION
    {
        public long FileOffset;

        public long BeyondFinalZero;
    }
}

以及测试上述类的示例代码。

class Program
{
    static void Main(string[] args)
    {
        using (var fileStream = new FileStream("test", FileMode.Create, FileAccess.ReadWrite, FileShare.None))
        {
            fileStream.SetLength(1024 * 1024 * 128);
            fileStream.MakeSparse();
            fileStream.SetSparseRange(0, fileStream.Length);
        }
    }
}

希望对你有帮助