Linux Kernel Architecture

Chapter 6: Device Drivers

slices are allocated for each of the queues, and a round robin algorithm is used to process the

queue. This ensures that the I/O bandwidth is shared in a fair manner between the different

queues. If the number of queues is bigger than or equal to the number of processes doing simul-

taneous I/O, this implies that the bandwidth is also fairly distributed between the processes.

Some practical problems (like multiple processes being mapped to identical queues, varying

request sizes, different I/O priorities, etc.) make the distribution not completely fair, but basi-

cally, the method achieves its goal to a good extent.

The deadline scheduler was the default scheduler almost up to the end of development of 2.5 but was

replaced with the anticipatory scheduler until 2.6.17. From 2.6.18 onward, the Completely Fair Queuing

scheduler is the default choice.

For reasons of space, I shall not deal with the implementation details of each scheduler. It should be

noted, however, that the deadline scheduler is much less complicated than the anticipatory scheduler but

delivers practically identical performance in most situations.

6.5.9 Implementation of Ioctls

Ioctls permit the use of special device-specific functions that cannot be accessed by normal read and write

operations. This form of support is implemented by means of theioctlsystem call that can be used with

regular files (detailed descriptions of how it is used are provided in the many system programming

manuals).

As expected, the system call is implemented insys_ioctl, but the main work is done invfs_ioctl.

Figure 6-19 shows the associated code flow diagram.

Process standard ioctls

Block devices: blkdef_ioctl

Yes

No

vfs_ioctl

file_ioctl

file->f_op->ioctl

Regular file?

Figure 6-19: Code flow diagram forsys_ioctl.

The desired ioctl is specified by means of a passed constant; typically, symbolic pre-processor constants

are used for this purpose.

Two situations must be distinguished once the kernel has checked whether one of the standard ioctls has

been applied (these are available for all files in the system regardless of type); for example, whether the

file descriptor is to be closed whenexecis executed (see Chapter 2).