Short note: principles of I/O software

The goals of I/O software are described by the following principles.

• Device independence: programs should be able to access any I/O device without knowing it in advance. A program reading a file should work whether the file is on a floppy, hard disk or USB stick.
• Uniform naming: a file or device name is just a string or integer and does not depend on the device (e.g. /dev/sda in UNIX, all mounted in one tree).
• Error handling: errors should be handled as close to the hardware as possible. The controller tries to fix it first, then the driver (e.g. retry a read); upper layers hear about it only if the lower layers fail.
• Synchronous (blocking) vs asynchronous (interrupt-driven) transfers: the hardware is interrupt driven, but user programs find blocking calls easier to write, so the OS makes interrupt-driven I/O look blocking.
• Buffering: data from a device cannot always be stored directly at the final destination (e.g. network packets), so buffers are used for checking, decoupling speed differences and packet size matching.
• Sharable vs dedicated devices: disks can be shared by many users; printers and tape drives are dedicated. The OS must handle both without causing deadlocks.

These principles are achieved by organising the I/O software in layers: interrupt handlers, device drivers, device-independent software and user-level software.