What are file descriptors and how does I/O buffering work?

Deep dive

Descriptors are OS-managed references:

• Each process has a descriptor table.
• Descriptors refer to kernel objects (files, sockets, pipes).
• Buffering reduces syscall overhead and enables readahead.
• fsync/flush forces data to disk when durability matters.

Examples

Buffered write flow:

write() -> user buffer -> kernel buffer -> disk later
fsync() -> force disk commit

Common pitfalls

• Forgetting to close descriptors (leaks).
• Assuming buffered writes are durable.
• Mixing buffered and unbuffered I/O incorrectly.

Interview follow-ups

• What happens on a partial write?
• How do pipes use file descriptors?
• When do you need fsync or O_DIRECT?