Migrate device control applications from Windows to Linux

Overcome migration challenges by understanding differences in Windows, Linux device control


If you develop device control applications on different platforms, you know that Windows and Linux have different ways of doing device control, and migrating applications from one to the other can be a pain. In this article, we analyze how device control works in both operating systems, examining everything from architecture to system calls and focusing on the differences. We also give you a migration sample (in C/C++) to demonstrate the migration in detail.

For the purposes of this article, "Windows" refers to Windows 2000 or later, and Microsoft Visual C++® version 6 or later should be installed. For Linux, the kernel should be based on 2.6, and GNU GCC should be installed.

Comparing architectures of device control

Ways to control devices differ between Windows and Linux.

The Windows device control architecture

In Windows, the I/O subsystem connects user applications with device drivers and defines the infrastructure to support device drivers. Device drivers provide an I/O interface to particular devices (see Figure 1).

Figure 1. The Windows device control architecture
The Windows device control architecture
The Windows device control architecture

In the process of device control, I/O operations are encapsulated into an IRP (I/O Request Packet). The I/O manager creates the IRP and sends it to the top of the stack. Device drivers then get the stack location of an IRP, which contains parameters for the I/O request. According to the requirement in IRP (such as create, read, write, devioctl, cleanup, or close), each driver does its work through hardware interfaces.

The Linux device control architecture

The device control architecture is a little different in Linux, with the main difference being that normal files, directories, devices, and sockets are all files—everything is a file in Linux. To visit the device, the Linux kernel maps the device operation call to the device driver via the file system. There is no I/O manager in Linux: all I/O requests go to the file system at the beginning (see Figure 2).

Figure 2. The Linux device control architecture
The Linux device control architecture
The Linux device control architecture

Comparing device file names and path names

From a development point of view, getting the device handle is a prerequisite to device control; however, because the device control architecture varies, how you get the device handle is a different story depending on whether you're using Windows or Linux.

Generally speaking, the device handle is determined by the name of a particular device driver.

On Windows, the file name of a device driver is different from that of a common file; it's usually called the device pathname instead. It has a fixed format like \.DeviceName. In C/C++ programming, this character string should be \\.\DeviceName. And in code, we make it \\\\.\\DeviceName. DeviceName should be the same as the device name defined in the corresponding device driver program.

Some device names are defined by Microsoft and will not be changed (as listed in Table 1).

Table 1. Device names on Windows (x = 0, 1, 2, etc.)
Floppy drive A: B:
Hard disk logic sub-area C: D: E: . . .
Physical drivePhysicalDrivex
Tape driveTapex
COM portCOMx

For example, we use device pathnames such as \\\\.\\PhysicalDrive1, \\\\.\\CdRom0, and \\\\.\\Tape0 in C/C++ programming. For details on other devices not in this general list, check the Related topics section later in this article.

Because the devices are described as files in Linux, you can find all of these device files in the directory ./dev. The device drivers in this directory include:

  • IDE (Integrated Drive Electronics) hard drives, such as /dev/hda and /dev/hdb
  • CD-ROM drives, some of which are IDE; others are CD-RW (CD read/write) drives emulated as SCSI (Small Computer Systems Interface) devices such as /dev/scd0
  • Serial ports, such as /dev/ttyS0 for COM1, /dev/ttyS1 for COM2, and so on
  • Pointing devices, including /dev/input/mice and others
  • Printers, such as /dev/lp0

Most common device files can be found according to the description above. For other device file names and detailed device information, try the command dmesg.

Comparing main system calls

Main system calls for device control include the following operations: open, close, I/O control, read/write, etc. See the Windows/Linux mapping shown in Table 2.

Table 2. Device control function mapping

Now let's dig deeper into three of the most common functions: create, close, and devioctl.

Opening and closing a device in Windows

In Windows, we're talking about CreateFile and CloseHandle. You use the function CreateFile to open a device. The function returns a handle that can be used to access the object as shown in Listing 1.

Listing 1. The CreateFile function in Windows
HANDLE CreateFile (LPCTSTR lpFileName,          //File name of the device 
                                                  (Device Pathname)
   DWORD dwDesiredAccess,                       //Access mode to the object (read, write, 
                                                  or both)
   DWORD dwShareMode,                           //Sharing mode of the object (read, 
                                                  write, both or none)
   LPSECURITY_ATTRIBUTES lpSecurityAttributes,  //Security attribute determining whether 
                                                  the returned handle can be inherited by 
                                                  child processes
   DWORD dwCreationDisposition,                 //Action taken on files that exist and 
                                                  do not exist
   DWORD dwFlagsAndAttributes,                  //File attributes and flags
   HANDLE hTemplateFile);                       //A handle to a template file

The parameter lpFileName is the device path name that has been specified earlier. Generally, to open a device, we can set dwDesiredAccess to 0 or GENERIC_READ|GENERIC_WRITE, dwShareMode to FILE_SHARE_READ|FILE_SHARE_WRITE, dwCreationDisposition to OPEN_EXISTING, dwFlagsAndAttributes and hTemplateFile to 0 or NULL. The returned handle will be used in further device control operations.

To close a device, use the function CloseHandle. Set the parameter hObject as the handle returned when the device is open: BOOL WINAPI CloseHandle (HANDLE hObject);.

Opening and closing a device in Linux

In Linux, we're talking about open and close. As we mentioned earlier, opening a device is just like opening a common file. Listing 2 shows how we use open to get a device handle.

Listing 2. The open function in Linux
int open (const char *pathname,
       int flags, 
       mode_t mode);

The file descriptor returned by a successful call will be the lowest-numbered file descriptor not currently open for the process. If failed, -1 is returned. The file descriptor is used as the device handle.

The parameter flags must include one of these: O_RDONLY, O_WRONLY, or O_RDWR. Other flags are optional. The argument mode specifies the permissions to use in case a new file is created.

The function close closes a device on Linux, just like closing a file: int close(int fd);.

DeviceIoControl in Windows

Device control (DeviceIoControl in Windows and ioctl in Linux) is the most common function used for device control, fulfilling such tasks as accessing devices, getting information, sending orders, and exchanging data. Listing 3 illustrates DeviceIoControl:

Listing 3. The DeviceIoControl function in Windows
BOOL DeviceIoControl (HANDLE hDevice,
      DWORD dwIoControlCode,
      LPVOID lpInBuffer,
      DWORD nInBufferSize,
      LPVOID lpOutBuffer,
      DWORD nOutBufferSize,
      LPDWORD lpBytesReturned,
      LPOVERLAPPED lpOverlapped);

This system call sends control code and other data to a specified device. The corresponding device drivers work according to what the control code dwIoControlCode tells them to do. For example, we can use IOCTL_DISK_GET_DRIVE_GEOMETRY to get structure parameters from physical drives (type of medium, number of cylinders, track number on each cylinder, number of sectors on each track, etc.). You can find all the control code definitions, header files, and other detailed information on the MSDN Web site (see Related topics for links).

Whether the input/output buffers are required and what their structure and size is depends on the device and operation the actual ioctl process relates to. They are also determined by dwIoControlCode specified in the call.

If the pointer for an overlapped operation is set to NULL, DeviceIoControl will work in a blocked (synchronous) way. Otherwise, it will work asynchronously.

ioctl in Linux

In Linux, you use ioctlint ioctl(int fildes, int request, /* arg */ ...);—to send the control information to a specified device. The first parameter fildes is the open file descriptor returned from the function open(), referring to a specific device.

Unlike the corresponding system call DeviceIOControl, the input parameter list in ioctl is not fixed. It depends on what kind of request the ioctl performs and what is specified by the parameter request, just like dwIoControlCode in Windows DeviceIOControl. However, during migration, you need to pay attention when choosing the correct request parameter because dwIoControlCode in DeviceIOControl and request in ioctl are not of the same value, and there is no explicit mapping list for dwIoControlCode/request. Normally, you choose the value of a parameter request by looking into its definition in its header file. All the definitions of the control codes are in /usr/include/{asm,linux}/*.h.

The parameter arg is left to deliver detailed command information needed by the specific device to do its required work. The data type of arg depends on the particular control request. We can use this argument to both deliver detailed commands and receive returned data.

Migration sample

Let's look at an example of the migration process from Windows to Linux. This example involves reading the SMART log from the main IDE hard drive on a personal computer.

Step 1. Identify the type of device

As we discussed, each device on Linux is regarded as a file. The first step is to figure out the file name of the device on Linux. Only by using this file name can we get the device handle that is required for device control.

In this example, the object is an IDE hard drive. In Linux, it is described as /dev/hda, /dev/hdb, etc. The device path name of the hard disk we are going to migrate in this example is \\\\.\\PhysicalDrive0. /dev/hda is the corresponding file name of the device on Linux.

Step 2. Change the include headers

We must change the #include header files to their Linux forms (see Table 3):

Table 3. #include header files
#include <windows.h> #include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <devioctl.h> #include <sys/ioctl.h>
#include <ntddscsi.h> #include <linux/hdreg.h>

windows.h is used for functions opening and closing the device (CreateFile and CloseHandle). Accordingly, header files needed for open() and close() on Linux should be included. They are sys/types.h, sys/stat.h, and fcntl.h.

devioctl.h in Windows is for function DeviceIoControl, and we change it to sys/ioctl.h to make sure the function ioctl can work.

In ntddscsi.h (this is the header file from DDK), a set of control codes are defined for device controlling. Because this sample deals only with the IDE hard drive, we just need to add linux/hdreg.h into the Linux program.

In other situations, make sure all the header files with definitions of required control codes are included. For example, to access a CD-ROM rather than a hard drive, include linux/cdrom.h instead.

Step 3. Revise functions and parameters

Now let's see the code in detail. Listing 4 shows the detailed command information.

Listing 4. Command details
unsigned char cmdBuff[7];
cmdBuff[0] = SMART_READ_LOG;  // Used for specifying SMART "commands"
cmdBuff[1] = 1;               // IDE sector count register
cmdBuff[2] = 1;               // IDE sector number register
cmdBuff[3] = SMART_CYL_LOW;   // IDE low order cylinder value
cmdBuff[4] = SMART_CYL_HI;    // IDE high order cylinder value
cmdBuff[5] = 0xA0 | (((Dev->Id-1) & 1) * 16); // IDE drive/head register
cmdBuff[6] = SMART_CMD;       // Actual IDE command

The command information is from the ATA command specification. Because no changes are needed to transplant the code to Linux, no further analysis is needed.

The code shown in Listing 5 opens the main hard drive on Windows.

Listing 5. Opening the main hard drive on Windows
HANDLE devHandle = CreateFile("\\\\.\\PhysicalDrive0",           //pathname
                             GENERIC_WRITE|GENERIC_READ,         //Access Mode
                             FILE_SHARE_READ|FILE_SHARE_WRITE,   //Sharing Mode

Remember from the section on opening and closing that we need two parameters (file path name and access mode to the device) to open a device on Linux. According to the previous original code, the first should be /dev/hda, and the second O_RDONLY|O_NONBLOCK. The changed code looks like this: HANDLE devHandle = open("/dev/hda", O_RDONLY | O_NONBLOCK);. Accordingly, change CloseHandle(devHandle); to close(devHandle);.

The main part is about how to use ioctl to get access to the particular device and the information we want. The original Windows code is shown in Listing 6:

Listing 6. Source of DeviceIoControl on Windows
typedef struct _Buffer{
       UCHAR   req[8];              // Detailed command information other than 
                                       control code
       ULONG   DataBufferSize;      // Size of Data Buffer, here is 512
       UCHAR   DataBuffer[512];     // Data Buffer
} Buffer;

Buffer regBuffer;
memcpy(regBuffer.req, cmdBuff, 7);  //req[7] is reserved for future use. Must be zero.
regBuffer.DataBufferSize = 512;
unsigned int size = 512+12;         // Size of regBuffer
                                    // 8 for req, 4 for DataBufferSize, 512 for data
DWORD bytesRet = 0;                 // Number of bytes returned
int retval;                         // Returned value

retval = DeviceIoControl(devHandle,
                         IOCTL_IDE_PASS_THROUGH,  //Control code
                         regBuffer, // Input Buffer, including detailed command
                         regBuffer, // Output Buffer, use the same buffer here
                         &bytesRet, NULL);
if (!retval)
	cout<<"DeviceIoControl failed."<<endl;
memcpy(data, retBuffer.DataBuffer, 512);

DeviceIoControl requires more parameters than ioctl does. The device handle is the first parameter on both platforms, returned from CreateFile/open() for Linux. But control code on Windows and requests on Linux are defined in such a different way that there is no fixed rule to find the mapping relations between these two parameters, as we discussed earlier. IOCTL_IDE_PASS_THROUGH is defined in header file ntddscsi.h as CTL_CODE (IOCTL_SCSI_BASE, 0x040a, METHOD_BUFFERED, FILE_READ_ACCESS | FILE_WRITE_ACCESS). By looking at the definitions in header file /usr/include/linux/hdreg.h, we choose the corresponding control code HDIO_DRIVE_CMD for Linux.

In addition, detailed command information is needed for the device to fulfill a specific task. The command is included in a buffer being exchanged in the process along with the space left for returned data. We use the same buffer to both send the command and get log information we need. In Linux, the space for data buffer size can be removed; not all eight bytes are required. In this sample, only four bytes of the command are used.

The corresponding code in Linux (Listing 7) seems much simpler because the structure and function arguments are simpler than in Windows.

Listing 7. Source of ioctl on Linux
int retval;
unsigned char req[4+512]; // Enough for returned data and the 4 byte detailed 
                             command information
req[0]= cmdBuff[6];       // Consider the requirement in this sample, only 4 bytes 
                             are used
req[1]= cmdBuff[2];
req[2]= cmdBuff[0];
req[3]= cmdBuff[1];

retval = ioctl(devHandle, HDIO_DRIVE_CMD, &req);
	cout<<"ioctl failed."<<endl;
memcpy(data, &req[4], 512);

Step 4. Testing in the Linux environment

After revising header files, functions, and parameters, the program is ready for a Linux run. Now the task is to compile it on a Linux platform and correct any remaining syntax errors. Some additional changes may be needed according to the edition of Linux and compiling environment.

Downloadable resources

Related topics


Sign in or register to add and subscribe to comments.

ArticleTitle=Migrate device control applications from Windows to Linux