#rcu #reader #locking #update #read #primitive #concurrency

read-copy-update

Provides a read-copy-update locking primitive

1 unstable release

0.1.0 Jan 6, 2024

#1339 in Data structures

MIT license

20KB
457 lines

Provides a read-copy-update locking primitive

RCU locks are a popular concurrency primitive inside the linux kernel. They support multiple concurrent readers accessing a value updated by a single writer. RCU locks are useful for publishing new configuration values to multiple readers or as a lightweight garbage collection mechanism.

Example

To create a new RCU lock, call Rcu::new with an initial value. Then create a new Reader by calling Rcu::reader, this provides a handle which allows accessing a value by calling Reader::read.

# use read_copy_update::Rcu;
# use std::{thread, time::Duration};
let mut rcu = Rcu::new(10);

// This thread can read the value, new updates will be published without requiring locks.
let mut rdr = rcu.reader();
thread::spawn(move || {
    println!("{}", *rdr.read()); // prints '10'
    thread::sleep(Duration::from_millis(1));
    println!("{}", *rdr.read()); // prints '20'
});

// The value can be updated on the main thread, independently of other readers.
rcu.write(20);

A common use-case is sharing readers in thread-local storage. With the thread-local feature flag enabled Reader instances can be associated to a thread allowing lock free reads and publishing across thread-locals.

# use read_copy_update::{Rcu, ThreadLocal};
# use std::{thread, time::Duration};
let mut rcu = Rcu::new(10);
let tls = ThreadLocal::new(&rcu);

thread::scope(|s| {
    s.spawn(|| {
        let val = tls.get_or_init();
        println!("{}", *val);
    });
    s.spawn(|| {
        let val = tls.get_or_init();
        println!("{}", *val);
    });
});

// The value can be updated on the main thread, independently of other readers.
rcu.write(20);

Design

Internally, new values are allocated on the heap and associated with an epoch. Readers pull in the pointer on read() calls and increment a per-reader epoch counter. Reclamation is done based on ensuring that all active readers have advanced their epoch past the value to be reclaimed. Because of this design, nested read() calls will not pick up the latest value, only once all references are dropped will the value be refreshed.

Dependencies

~0–27MB
~333K SLoC