Freigeben über


HOWTO: ISAPI Filter logging request URL and headers based on User-Agent

A user recently reported on the microsoft.public.inetserver.iis NNTP newsgroup seeing MSNBot causing lots of 406s while indexing static files (like HTML) on his IIS web server. Alas, there is no built in feature in IIS to take a snapshot of a request which match certain criteria (in this case, a request whose User-Agent indicates MSNBot, and we want all the request headers, especially the Accept: header). So, it motivated me to whip together the following ISAPI Filter to do the job.

Question:

I've been having this problem with the MSNbot for several months.

Most requests that it sneds end up with IIS sending back a 406 error.

I've read that a 406 is usually caused by the request headers missing an accept-type header.

Is there anything I can do on IIS v5 (Win2K) to correct this problem with these 406 errors? I think it's really hurting my MSN search engine placement

Answer:

Unfortunately, there is no built-in functionality in IIS that can take a snapshot of the request given some criteria. However, IIS extensibility allows this functionality to be added, as the following ISAPI Filter illustrates.

After installing this filter (either globally to apply to all requests [restart IIS required] or per-site [no restart required]), this filter basically:

  • Looks at the User-Agent: request header and see if it starts with "msnbot/"
  • If the request is from a client claiming to be the MSNBot, then record the requested URL+Querystring as well as all headers of the request into a log file which is per-PID. Otherwise, do nothing
  • The name of the logfile defaults to: C:\inetpub\wwwroot\LogFile.<PID>.log

Thread-safe Logging

In addition to the "usual" sorts of ISAPI Filter behavior, this filter also illustrates a correct way to do something as common as logging while accounting for one of the potential breaking changes in ISAPI Filter behavior from IIS5 to IIS6: ISAPI Filters are no longer singletons on the system.

As always, an ISAPI Filter must log data to a single shared resource (log file) in a thread-safe manner. However, an ISAPI Filter on IIS6 must also be aware that in IIS6 Worker Process Isolation Mode, multiple w3wp.exe can simultaneously run the ISAPI Filter DLL - thus, any process-wide synchronization mechanisms need to be scaled across processes.

  1. For thread-safety - a critical section controls access to process-wide shared resource like the log file to prevent corruption of it's content by multiple threads running through a given ISAPI Filter DLL. Remember, the ISAPI Filter API may be synchronous, but IIS is multi-threaded so multiple threads could be executing in a synchronous manner through the same Filter module in the same process.
  2. For cross-process safety - the log filename includes the PID of the process that is running the ISAPI Filter DLL so that threads running through ISAPI Filter DLLs loaded in DIFFERENT processes are prevented from simultaneously accessing the same log file. Critical sections will not be able to control access because they are per-process, and global Mutex is too heavy handed because multiple threads in multiple processes in multiple application pools should be able to independently and simultaneously log data.

Enjoy.

//David

 #include <windows.h>
#include <httpfilt.h>
#include <stdio.h>

#define DEFAULT_BUFFER_SIZE         1024
#define MAX_BUFFER_SIZE             4096
#define LOG_FILENAME                "C:\\inetpub\\wwwroot\\LogFile"

HANDLE                              g_hLogFile = INVALID_HANDLE_VALUE;
CRITICAL_SECTION                    g_LogFileLock;


DWORD
OnPreprocHeaders(
    HTTP_FILTER_CONTEXT *           pfc,
    HTTP_FILTER_PREPROC_HEADERS *   pPPH
);

BOOL
WINAPI
GetFilterVersion(
    HTTP_FILTER_VERSION *           pVer
    )
/*++

Purpose:

    Required entry point for ISAPI filters.  This function
    is called when the server initially loads this DLL.

Arguments:

    pVer - Points to the filter version info structure

Returns:

    TRUE on successful initialization
    FALSE on initialization failure

--*/
{
    CHAR                            szLogFile[ MAX_PATH ];

    //
    // Initialize the logging critical section
    //
    InitializeCriticalSection( &g_LogFileLock );
    //
    // TODO: Choose a location where IIS process identity can write to
    //
    sprintf( szLogFile,
             "%s.%d.log",
             LOG_FILENAME,
             GetCurrentProcessId() );
    //
    // Initialize the logfile name
    //
    g_hLogFile = CreateFile(
        szLogFile,
        GENERIC_WRITE,
        FILE_SHARE_READ | FILE_SHARE_WRITE,
        NULL,
        OPEN_ALWAYS,
        FILE_ATTRIBUTE_NORMAL,
        NULL
        );
    if ( g_hLogFile == INVALID_HANDLE_VALUE )
    {
        //
        // Failed to open log file due to GetLastError(). Bail.
        //
        return FALSE;
    }

    pVer->dwFilterVersion = HTTP_FILTER_REVISION;
    lstrcpyn( pVer->lpszFilterDesc,
             "ISAPI Filter to log request headers based on User-Agent",
             SF_MAX_FILTER_DESC_LEN );
    pVer->dwFlags =
        SF_NOTIFY_ORDER_HIGH |
        SF_NOTIFY_PREPROC_HEADERS
        ;

    return TRUE;
}

DWORD
WINAPI
HttpFilterProc(
    IN HTTP_FILTER_CONTEXT *        pfc,
    DWORD                           dwNotificationType,
    LPVOID                          pvNotification
    )
/*++

Purpose:

    Required filter notification entry point.  This function is called
    whenever one of the events (as registered in GetFilterVersion) occurs.

Arguments:

    pfc              - A pointer to the filter context for this notification
    NotificationType - The type of notification
    pvNotification   - A pointer to the notification data

Returns:

    One of the following valid filter return codes:
    - SF_STATUS_REQ_FINISHED
    - SF_STATUS_REQ_FINISHED_KEEP_CONN
    - SF_STATUS_REQ_NEXT_NOTIFICATION
    - SF_STATUS_REQ_HANDLED_NOTIFICATION
    - SF_STATUS_REQ_ERROR
    - SF_STATUS_REQ_READ_NEXT

--*/
{
    switch ( dwNotificationType )
    {
    case SF_NOTIFY_PREPROC_HEADERS:
        return OnPreprocHeaders(
            pfc,
            (HTTP_FILTER_PREPROC_HEADERS *) pvNotification );
    }

    return SF_STATUS_REQ_NEXT_NOTIFICATION;
}

DWORD
OnPreprocHeaders(
    HTTP_FILTER_CONTEXT *           pfc,
    HTTP_FILTER_PREPROC_HEADERS *   pPPH
)
{
    DWORD                           dwRet = SF_STATUS_REQ_NEXT_NOTIFICATION;
    BOOL                            fRet = FALSE;
    CHAR                            pBuf[ DEFAULT_BUFFER_SIZE ];
    CHAR                            pBuf2[ DEFAULT_BUFFER_SIZE ];
    CHAR *                          pszBuf = pBuf;
    CHAR *                          pszBuf2 = pBuf2;
    DWORD                           cbBuf = DEFAULT_BUFFER_SIZE;
    DWORD                           cbBuf2 = DEFAULT_BUFFER_SIZE;
    CHAR                            szUserAgent[] = "User-Agent:";
    CHAR                            szUrl[] = "url";
    CHAR                            szMSNBot[] = "msnbot/";
    CHAR                            cbMSNBot = 7;

    SetLastError( NO_ERROR );

    if ( pfc == NULL ||
         pPPH == NULL )
    {
        SetLastError( ERROR_INVALID_PARAMETER );
        goto Finished;
    }

    fRet = pPPH->GetHeader( pfc, szUserAgent, pszBuf, &cbBuf );
    if ( fRet == FALSE )
    {
        if ( GetLastError() == ERROR_INVALID_INDEX )
        {
            //
            // User-Agent was not given
            // Since we are looking for a particular user-agent,
            // not finding on is ok.
            //
            OutputDebugString( "Nothing to do.\n" );
            SetLastError( NO_ERROR );
            goto Finished;
        }
        else if ( GetLastError() == ERROR_INSUFFICIENT_BUFFER &&
                  cbBuf < MAX_BUFFER_SIZE )
        {
            pszBuf = new CHAR[ cbBuf ];
            if ( pszBuf == NULL )
            {
                SetLastError( ERROR_NOT_ENOUGH_MEMORY );
                goto Finished;
            }

            fRet = pPPH->GetHeader( pfc, szUserAgent, pszBuf, &cbBuf );
            if ( fRet == FALSE )
            {
                goto Finished;
            }

            OutputDebugString( "Found User-Agent header.\n" );
         }
         else
         {
             goto Finished;
         }
    }

    //
    // At this point, pszBuf points to the value of User-Agent header.
    // Figure out what to do.
    //
    OutputDebugString( pszBuf );
    OutputDebugString( "\n" );

    //
    // If the User-Agent matches the MSNBot, log request headers
    //
    if ( _strnicmp( pszBuf, szMSNBot, cbMSNBot ) == 0 )
    {
        fRet = pfc->GetServerVariable( pfc, "ALL_RAW", pszBuf2, &cbBuf2 );
        if ( fRet == FALSE )
        {
            if ( GetLastError() == ERROR_INSUFFICIENT_BUFFER &&
                 cbBuf2 < MAX_BUFFER_SIZE )
            {
                pszBuf2 = new CHAR[ cbBuf2 ];
                if ( pszBuf2 == NULL )
                {
                    SetLastError( ERROR_NOT_ENOUGH_MEMORY );
                    goto Finished;
                }

                fRet = pfc->GetServerVariable( pfc,
                                               "ALL_RAW",
                                               pszBuf2,
                                               &cbBuf2 );
                if ( fRet == FALSE )
                {
                    goto Finished;
                }

                OutputDebugString( "Read ALL_RAW\n" );
            }
            else
            {
                goto Finished;
            }
        }

        //
        // At this point, pszBuf2 points to the raw request headers
        // But first, retrieve the URL of this request by reusing pszBuf.
        // Then append it to disk, thread-safe.
        //
        if ( pszBuf != pBuf )
        {
            delete pszBuf;
            pszBuf = pBuf;
        }

        fRet = pPPH->GetHeader( pfc, szUrl, pszBuf, &cbBuf );
        if ( fRet == FALSE )
        {
            if ( GetLastError() == ERROR_INSUFFICIENT_BUFFER &&
                 cbBuf < MAX_BUFFER_SIZE )
            {
                pszBuf = new CHAR[ cbBuf ];
                if ( pszBuf == NULL )
                {
                    SetLastError( ERROR_NOT_ENOUGH_MEMORY );
                    goto Finished;
                }

                fRet = pPPH->GetHeader( pfc, szUrl, pszBuf, &cbBuf );
                if ( fRet == FALSE )
                {
                    goto Finished;
                }

                OutputDebugString( "Found URL.\n" );
             }
             else
             {
                 goto Finished;
             }
        }

        //
        // At this point, pszBuf points to the raw URL...
        //
        OutputDebugString( "URL = " );
        OutputDebugString( pszBuf );
        OutputDebugString( "\n" );
        OutputDebugString( "ALL_RAW =\n" );
        OutputDebugString( pszBuf2 );

        //
        // Write the URL and raw headers to disk
        //
        if ( g_hLogFile != INVALID_HANDLE_VALUE )
        {
            EnterCriticalSection( &g_LogFileLock );
            SetFilePointer( g_hLogFile, 0, NULL, FILE_END );
            cbBuf--;
            fRet = WriteFile( g_hLogFile, pszBuf, cbBuf, &cbBuf, NULL );
            cbBuf = 1;
            fRet = WriteFile( g_hLogFile, "\n", cbBuf, &cbBuf, NULL );
            cbBuf2--;
            fRet = WriteFile( g_hLogFile, pszBuf2, cbBuf2, &cbBuf2, NULL );
            LeaveCriticalSection( &g_LogFileLock );
        }
    }

    SetLastError( NO_ERROR );

Finished:

    if ( pszBuf != pBuf )
    {
        delete pszBuf;
    }

    if ( GetLastError() != NO_ERROR )
    {
        OutputDebugString( "Error!\n" );
        dwRet = SF_STATUS_REQ_ERROR;
    }

    return dwRet;
}

BOOL
WINAPI
TerminateFilter(
    DWORD                           dwFlags
)
/*++

  Optional filter entry point.  This function is called by the server
  before this DLL is unloaded.

  Arguments:

    dwFlags - No flags have been defined at this time

  Returns:

    Always returns TRUE;

--*/
{
    if ( g_hLogFile != INVALID_HANDLE_VALUE )
    {
        CloseHandle( g_hLogFile );
        g_hLogFile = INVALID_HANDLE_VALUE;
    }

    DeleteCriticalSection( &g_LogFileLock );

    return TRUE;
}

Comments

  • Anonymous
    May 12, 2006
    Hello!

    I know nothing about ISAPI... and I'm pretty new to IIS.
    But I want to do something I think is similar to this.

    I want to redirect users to a specific web site according to the User_Agent (platform - windows/mac).

    For example, I want every user to use the same URL to access my web site.  However, if the user is using a MacIntosh, I want him to be redirected to another web site.  if the user is using a PC (windows), I don't want to change anything.

    Is that possible?
    If so, is it easy to achieve?

    I would greatly appreciate if you could give me a hand on this.

    Thanks!
    (... and great site!)

  • Anonymous
    May 12, 2006
    The comment has been removed

  • Anonymous
    May 15, 2006
    Hello and thanks for your answer!

    My understanding is that a ISAPI filter is executed before the user access my Web Site... and the ASP is executed at the Web Site level (or virtual directory). Am I right?

    My Web Site answers requests from every host+domain name. Users can type whatever URL exists.

    I was thinking about a ISAPI filter to check his browser/platform and then rewrite the URL if neccessary (to access another Web Site).

    Using ASP, I would have to call a .aspx file, right? So this ASP won't be executed if the user type a URL like this "http://www.domain.com/dir/appl?open". The ASP file will be run only if the user type the URL to it in a browser... which nobody will do.

    I want to trap EVERY request (URL) set to IIS and redirect them to another Web Site if the platform is a MacIntosh.  If it's possible and simplier to use ASP, I'm willing to use this.  However, I might be missing something (I don't know ASP yet), but I don't see how I could use ASP in this context? (having a kind of filter to all requests using ASP)

    Thanks again!
    Pascal

  • Anonymous
    July 04, 2006
    Hi

    What does the "IN" do here

    HttpFilterProc(IN HTTP_FILTER_CONTEXT

    Thanks!

  • Anonymous
    July 04, 2006
    David - "IN" is a standard NT #define which C-preprocesses away to nothing.

    It is useful to decorate code to describe whether a parameter is used for INput or OUTput. It can also be used by static source-code analysis tools, amongst other things.

    //David

  • Anonymous
    July 12, 2006
    Can anybody help me, I'm not a C++ programmer, and either a ISAPI speciallist... but I need an ISAPI program to remove some characters in "post/get" methods.

    I'll appreciate your help.

    Thanks.

  • Anonymous
    July 19, 2007
    Our www.dfs.com site is running at IIS 6.0 server.  We see out Website performance monitoring System (www.keynote.com) hits are logged into our IIS log files which causes increased usage of page views in Web Analytics application. We would like to filter out these hits from our log files.  Could you please let me know how it can be done?  

  • Anonymous
    July 21, 2007
    Binod - You can use a tool like LogParser to filter through IIS logfiles. Otherwise, you should configure the Web Analytics application to filter these things out. But in no case should IIS be filtering request logging. That is not performant nor secure. //David

  • Anonymous
    July 24, 2007
    David, We can use the command QGREP <A href="http://www.ss64.com/nt/qgrep.html">QGREP windows resource kit</a> to filter such entries

  • Anonymous
    October 26, 2007
    Hi David, I just copied your code and tried it on my computer. It seems that ISAPI filter can't create a file. The error code is 5 -- ERROR_ACCESS_DENIED. Could you please tell me how to configure the ISAPI filter to allow its access? My testing environment is IIS 6.0 on Window 2003 Enterprise. Thanks // maone