Anforderungen  |   Konzepte  |   Entwurf  |   Entwicklung  |   Qualitätssicherung  |   Lebenszyklus  |   Steuerung
 
 
 
 


Quelle  rwlock.rs   Sprache: unbekannt

 
// Copyright 2016 Amanieu d'Antras
//
// Licensed under the Apache License, Version 2.0, <LICENSE-APACHE or
// http://apache.org/licenses/LICENSE-2.0> or the MIT license <LICENSE-MIT or
// http://opensource.org/licenses/MIT>, at your option. This file may not be
// copied, modified, or distributed except according to those terms.

use crate::raw_rwlock::RawRwLock;
use lock_api;

/// A reader-writer lock
///
/// This type of lock allows a number of readers or at most one writer at any
/// point in time. The write portion of this lock typically allows modification
/// of the underlying data (exclusive access) and the read portion of this lock
/// typically allows for read-only access (shared access).
///
/// This lock uses a task-fair locking policy which avoids both reader and
/// writer starvation. This means that readers trying to acquire the lock will
/// block even if the lock is unlocked when there are writers waiting to acquire
/// the lock. Because of this, attempts to recursively acquire a read lock
/// within a single thread may result in a deadlock.
///
/// The type parameter `T` represents the data that this lock protects. It is
/// required that `T` satisfies `Send` to be shared across threads and `Sync` to
/// allow concurrent access through readers. The RAII guards returned from the
/// locking methods implement `Deref` (and `DerefMut` for the `write` methods)
/// to allow access to the contained of the lock.
///
/// # Fairness
///
/// A typical unfair lock can often end up in a situation where a single thread
/// quickly acquires and releases the same lock in succession, which can starve
/// other threads waiting to acquire the rwlock. While this improves throughput
/// because it doesn't force a context switch when a thread tries to re-acquire
/// a rwlock it has just released, this can starve other threads.
///
/// This rwlock uses [eventual fairness](https://trac.webkit.org/changeset/203350)
/// to ensure that the lock will be fair on average without sacrificing
/// throughput. This is done by forcing a fair unlock on average every 0.5ms,
/// which will force the lock to go to the next thread waiting for the rwlock.
///
/// Additionally, any critical section longer than 1ms will always use a fair
/// unlock, which has a negligible impact on throughput considering the length
/// of the critical section.
///
/// You can also force a fair unlock by calling `RwLockReadGuard::unlock_fair`
/// or `RwLockWriteGuard::unlock_fair` when unlocking a mutex instead of simply
/// dropping the guard.
///
/// # Differences from the standard library `RwLock`
///
/// - Supports atomically downgrading a write lock into a read lock.
/// - Task-fair locking policy instead of an unspecified platform default.
/// - No poisoning, the lock is released normally on panic.
/// - Only requires 1 word of space, whereas the standard library boxes the
///   `RwLock` due to platform limitations.
/// - Can be statically constructed.
/// - Does not require any drop glue when dropped.
/// - Inline fast path for the uncontended case.
/// - Efficient handling of micro-contention using adaptive spinning.
/// - Allows raw locking & unlocking without a guard.
/// - Supports eventual fairness so that the rwlock is fair on average.
/// - Optionally allows making the rwlock fair by calling
///   `RwLockReadGuard::unlock_fair` and `RwLockWriteGuard::unlock_fair`.
///
/// # Examples
///
/// ```
/// use parking_lot::RwLock;
///
/// let lock = RwLock::new(5);
///
/// // many reader locks can be held at once
/// {
///     let r1 = lock.read();
///     let r2 = lock.read();
///     assert_eq!(*r1, 5);
///     assert_eq!(*r2, 5);
/// } // read locks are dropped at this point
///
/// // only one write lock may be held, however
/// {
///     let mut w = lock.write();
///     *w += 1;
///     assert_eq!(*w, 6);
/// } // write lock is dropped here
/// ```
pub type RwLock<T> = lock_api::RwLock<RawRwLock, T>;

/// Creates a new instance of an `RwLock<T>` which is unlocked.
///
/// This allows creating a `RwLock<T>` in a constant context on stable Rust.
pub const fn const_rwlock<T>(val: T) -> RwLock<T> {
    RwLock::const_new(<RawRwLock as lock_api::RawRwLock>::INIT, val)
}

/// RAII structure used to release the shared read access of a lock when
/// dropped.
pub type RwLockReadGuard<'a, T> = lock_api::RwLockReadGuard<'a, RawRwLock, T>;

/// RAII structure used to release the exclusive write access of a lock when
/// dropped.
pub type RwLockWriteGuard<'a, T> = lock_api::RwLockWriteGuard<'a, RawRwLock, T>;

/// An RAII read lock guard returned by `RwLockReadGuard::map`, which can point to a
/// subfield of the protected data.
///
/// The main difference between `MappedRwLockReadGuard` and `RwLockReadGuard` is that the
/// former doesn't support temporarily unlocking and re-locking, since that
/// could introduce soundness issues if the locked object is modified by another
/// thread.
pub type MappedRwLockReadGuard<'a, T> = lock_api::MappedRwLockReadGuard<'a, RawRwLock, T>;

/// An RAII write lock guard returned by `RwLockWriteGuard::map`, which can point to a
/// subfield of the protected data.
///
/// The main difference between `MappedRwLockWriteGuard` and `RwLockWriteGuard` is that the
/// former doesn't support temporarily unlocking and re-locking, since that
/// could introduce soundness issues if the locked object is modified by another
/// thread.
pub type MappedRwLockWriteGuard<'a, T> = lock_api::MappedRwLockWriteGuard<'a, RawRwLock, T>;

/// RAII structure used to release the upgradable read access of a lock when
/// dropped.
pub type RwLockUpgradableReadGuard<'a, T> = lock_api::RwLockUpgradableReadGuard<'a, RawRwLock, T>;

#[cfg(test)]
mod tests {
    use crate::{RwLock, RwLockUpgradableReadGuard, RwLockWriteGuard};
    use rand::Rng;
    use std::sync::atomic::{AtomicUsize, Ordering};
    use std::sync::mpsc::channel;
    use std::sync::Arc;
    use std::thread;
    use std::time::Duration;

    #[cfg(feature = "serde")]
    use bincode::{deserialize, serialize};

    #[derive(Eq, PartialEq, Debug)]
    struct NonCopy(i32);

    #[test]
    fn smoke() {
        let l = RwLock::new(());
        drop(l.read());
        drop(l.write());
        drop(l.upgradable_read());
        drop((l.read(), l.read()));
        drop((l.read(), l.upgradable_read()));
        drop(l.write());
    }

    #[test]
    fn frob() {
        const N: u32 = 10;
        const M: u32 = 1000;

        let r = Arc::new(RwLock::new(()));

        let (tx, rx) = channel::<()>();
        for _ in 0..N {
            let tx = tx.clone();
            let r = r.clone();
            thread::spawn(move || {
                let mut rng = rand::thread_rng();
                for _ in 0..M {
                    if rng.gen_bool(1.0 / N as f64) {
                        drop(r.write());
                    } else {
                        drop(r.read());
                    }
                }
                drop(tx);
            });
        }
        drop(tx);
        let _ = rx.recv();
    }

    #[test]
    fn test_rw_arc_no_poison_wr() {
        let arc = Arc::new(RwLock::new(1));
        let arc2 = arc.clone();
        let _: Result<(), _> = thread::spawn(move || {
            let _lock = arc2.write();
            panic!();
        })
        .join();
        let lock = arc.read();
        assert_eq!(*lock, 1);
    }

    #[test]
    fn test_rw_arc_no_poison_ww() {
        let arc = Arc::new(RwLock::new(1));
        let arc2 = arc.clone();
        let _: Result<(), _> = thread::spawn(move || {
            let _lock = arc2.write();
            panic!();
        })
        .join();
        let lock = arc.write();
        assert_eq!(*lock, 1);
    }

    #[test]
    fn test_rw_arc_no_poison_rr() {
        let arc = Arc::new(RwLock::new(1));
        let arc2 = arc.clone();
        let _: Result<(), _> = thread::spawn(move || {
            let _lock = arc2.read();
            panic!();
        })
        .join();
        let lock = arc.read();
        assert_eq!(*lock, 1);
    }

    #[test]
    fn test_rw_arc_no_poison_rw() {
        let arc = Arc::new(RwLock::new(1));
        let arc2 = arc.clone();
        let _: Result<(), _> = thread::spawn(move || {
            let _lock = arc2.read();
            panic!()
        })
        .join();
        let lock = arc.write();
        assert_eq!(*lock, 1);
    }

    #[test]
    fn test_ruw_arc() {
        let arc = Arc::new(RwLock::new(0));
        let arc2 = arc.clone();
        let (tx, rx) = channel();

        thread::spawn(move || {
            for _ in 0..10 {
                let mut lock = arc2.write();
                let tmp = *lock;
                *lock = -1;
                thread::yield_now();
                *lock = tmp + 1;
            }
            tx.send(()).unwrap();
        });

        let mut children = Vec::new();

        // Upgradable readers try to catch the writer in the act and also
        // try to touch the value
        for _ in 0..5 {
            let arc3 = arc.clone();
            children.push(thread::spawn(move || {
                let lock = arc3.upgradable_read();
                let tmp = *lock;
                assert!(tmp >= 0);
                thread::yield_now();
                let mut lock = RwLockUpgradableReadGuard::upgrade(lock);
                assert_eq!(tmp, *lock);
                *lock = -1;
                thread::yield_now();
                *lock = tmp + 1;
            }));
        }

        // Readers try to catch the writers in the act
        for _ in 0..5 {
            let arc4 = arc.clone();
            children.push(thread::spawn(move || {
                let lock = arc4.read();
                assert!(*lock >= 0);
            }));
        }

        // Wait for children to pass their asserts
        for r in children {
            assert!(r.join().is_ok());
        }

        // Wait for writer to finish
        rx.recv().unwrap();
        let lock = arc.read();
        assert_eq!(*lock, 15);
    }

    #[test]
    fn test_rw_arc() {
        let arc = Arc::new(RwLock::new(0));
        let arc2 = arc.clone();
        let (tx, rx) = channel();

        thread::spawn(move || {
            let mut lock = arc2.write();
            for _ in 0..10 {
                let tmp = *lock;
                *lock = -1;
                thread::yield_now();
                *lock = tmp + 1;
            }
            tx.send(()).unwrap();
        });

        // Readers try to catch the writer in the act
        let mut children = Vec::new();
        for _ in 0..5 {
            let arc3 = arc.clone();
            children.push(thread::spawn(move || {
                let lock = arc3.read();
                assert!(*lock >= 0);
            }));
        }

        // Wait for children to pass their asserts
        for r in children {
            assert!(r.join().is_ok());
        }

        // Wait for writer to finish
        rx.recv().unwrap();
        let lock = arc.read();
        assert_eq!(*lock, 10);
    }

    #[test]
    fn test_rw_arc_access_in_unwind() {
        let arc = Arc::new(RwLock::new(1));
        let arc2 = arc.clone();
        let _ = thread::spawn(move || {
            struct Unwinder {
                i: Arc<RwLock<isize>>,
            }
            impl Drop for Unwinder {
                fn drop(&mut self) {
                    let mut lock = self.i.write();
                    *lock += 1;
                }
            }
            let _u = Unwinder { i: arc2 };
            panic!();
        })
        .join();
        let lock = arc.read();
        assert_eq!(*lock, 2);
    }

    #[test]
    fn test_rwlock_unsized() {
        let rw: &RwLock<[i32]> = &RwLock::new([1, 2, 3]);
        {
            let b = &mut *rw.write();
            b[0] = 4;
            b[2] = 5;
        }
        let comp: &[i32] = &[4, 2, 5];
        assert_eq!(&*rw.read(), comp);
    }

    #[test]
    fn test_rwlock_try_read() {
        let lock = RwLock::new(0isize);
        {
            let read_guard = lock.read();

            let read_result = lock.try_read();
            assert!(
                read_result.is_some(),
                "try_read should succeed while read_guard is in scope"
            );

            drop(read_guard);
        }
        {
            let upgrade_guard = lock.upgradable_read();

            let read_result = lock.try_read();
            assert!(
                read_result.is_some(),
                "try_read should succeed while upgrade_guard is in scope"
            );

            drop(upgrade_guard);
        }
        {
            let write_guard = lock.write();

            let read_result = lock.try_read();
            assert!(
                read_result.is_none(),
                "try_read should fail while write_guard is in scope"
            );

            drop(write_guard);
        }
    }

    #[test]
    fn test_rwlock_try_write() {
        let lock = RwLock::new(0isize);
        {
            let read_guard = lock.read();

            let write_result = lock.try_write();
            assert!(
                write_result.is_none(),
                "try_write should fail while read_guard is in scope"
            );
            assert!(lock.is_locked());
            assert!(!lock.is_locked_exclusive());

            drop(read_guard);
        }
        {
            let upgrade_guard = lock.upgradable_read();

            let write_result = lock.try_write();
            assert!(
                write_result.is_none(),
                "try_write should fail while upgrade_guard is in scope"
            );
            assert!(lock.is_locked());
            assert!(!lock.is_locked_exclusive());

            drop(upgrade_guard);
        }
        {
            let write_guard = lock.write();

            let write_result = lock.try_write();
            assert!(
                write_result.is_none(),
                "try_write should fail while write_guard is in scope"
            );
            assert!(lock.is_locked());
            assert!(lock.is_locked_exclusive());

            drop(write_guard);
        }
    }

    #[test]
    fn test_rwlock_try_upgrade() {
        let lock = RwLock::new(0isize);
        {
            let read_guard = lock.read();

            let upgrade_result = lock.try_upgradable_read();
            assert!(
                upgrade_result.is_some(),
                "try_upgradable_read should succeed while read_guard is in scope"
            );

            drop(read_guard);
        }
        {
            let upgrade_guard = lock.upgradable_read();

            let upgrade_result = lock.try_upgradable_read();
            assert!(
                upgrade_result.is_none(),
                "try_upgradable_read should fail while upgrade_guard is in scope"
            );

            drop(upgrade_guard);
        }
        {
            let write_guard = lock.write();

            let upgrade_result = lock.try_upgradable_read();
            assert!(
                upgrade_result.is_none(),
                "try_upgradable should fail while write_guard is in scope"
            );

            drop(write_guard);
        }
    }

    #[test]
    fn test_into_inner() {
        let m = RwLock::new(NonCopy(10));
        assert_eq!(m.into_inner(), NonCopy(10));
    }

    #[test]
    fn test_into_inner_drop() {
        struct Foo(Arc<AtomicUsize>);
        impl Drop for Foo {
            fn drop(&mut self) {
                self.0.fetch_add(1, Ordering::SeqCst);
            }
        }
        let num_drops = Arc::new(AtomicUsize::new(0));
        let m = RwLock::new(Foo(num_drops.clone()));
        assert_eq!(num_drops.load(Ordering::SeqCst), 0);
        {
            let _inner = m.into_inner();
            assert_eq!(num_drops.load(Ordering::SeqCst), 0);
        }
        assert_eq!(num_drops.load(Ordering::SeqCst), 1);
    }

    #[test]
    fn test_get_mut() {
        let mut m = RwLock::new(NonCopy(10));
        *m.get_mut() = NonCopy(20);
        assert_eq!(m.into_inner(), NonCopy(20));
    }

    #[test]
    fn test_rwlockguard_sync() {
        fn sync<T: Sync>(_: T) {}

        let rwlock = RwLock::new(());
        sync(rwlock.read());
        sync(rwlock.write());
    }

    #[test]
    fn test_rwlock_downgrade() {
        let x = Arc::new(RwLock::new(0));
        let mut handles = Vec::new();
        for _ in 0..8 {
            let x = x.clone();
            handles.push(thread::spawn(move || {
                for _ in 0..100 {
                    let mut writer = x.write();
                    *writer += 1;
                    let cur_val = *writer;
                    let reader = RwLockWriteGuard::downgrade(writer);
                    assert_eq!(cur_val, *reader);
                }
            }));
        }
        for handle in handles {
            handle.join().unwrap()
        }
        assert_eq!(*x.read(), 800);
    }

    #[test]
    fn test_rwlock_recursive() {
        let arc = Arc::new(RwLock::new(1));
        let arc2 = arc.clone();
        let lock1 = arc.read();
        let t = thread::spawn(move || {
            let _lock = arc2.write();
        });

        if cfg!(not(all(target_env = "sgx", target_vendor = "fortanix"))) {
            thread::sleep(Duration::from_millis(100));
        } else {
            // FIXME: https://github.com/fortanix/rust-sgx/issues/31
            for _ in 0..100 {
                thread::yield_now();
            }
        }

        // A normal read would block here since there is a pending writer
        let lock2 = arc.read_recursive();

        // Unblock the thread and join it.
        drop(lock1);
        drop(lock2);
        t.join().unwrap();
    }

    #[test]
    fn test_rwlock_debug() {
        let x = RwLock::new(vec![0u8, 10]);

        assert_eq!(format!("{:?}", x), "RwLock { data: [0, 10] }");
        let _lock = x.write();
        assert_eq!(format!("{:?}", x), "RwLock { data: <locked> }");
    }

    #[test]
    fn test_clone() {
        let rwlock = RwLock::new(Arc::new(1));
        let a = rwlock.read_recursive();
        let b = a.clone();
        assert_eq!(Arc::strong_count(&b), 2);
    }

    #[cfg(feature = "serde")]
    #[test]
    fn test_serde() {
        let contents: Vec<u8> = vec![0, 1, 2];
        let mutex = RwLock::new(contents.clone());

        let serialized = serialize(&mutex).unwrap();
        let deserialized: RwLock<Vec<u8>> = deserialize(&serialized).unwrap();

        assert_eq!(*(mutex.read()), *(deserialized.read()));
        assert_eq!(contents, *(deserialized.read()));
    }

    #[test]
    fn test_issue_203() {
        struct Bar(RwLock<()>);

        impl Drop for Bar {
            fn drop(&mut self) {
                let _n = self.0.write();
            }
        }

        thread_local! {
            static B: Bar = Bar(RwLock::new(()));
        }

        thread::spawn(|| {
            B.with(|_| ());

            let a = RwLock::new(());
            let _a = a.read();
        })
        .join()
        .unwrap();
    }

    #[test]
    fn test_rw_write_is_locked() {
        let lock = RwLock::new(0isize);
        {
            let _read_guard = lock.read();

            assert!(lock.is_locked());
            assert!(!lock.is_locked_exclusive());
        }

        {
            let _write_guard = lock.write();

            assert!(lock.is_locked());
            assert!(lock.is_locked_exclusive());
        }
    }
}

[ Dauer der Verarbeitung: 0.32 Sekunden  (vorverarbeitet)  ]

                                                                                                                                                                                                                                                                                                                                                                                                     


Neuigkeiten

     Aktuelles
     Motto des Tages

Software

     Produkte
     Quellcodebibliothek

Aktivitäten

     Artikel über Sicherheit
     Anleitung zur Aktivierung von SSL

Muße

     Gedichte
     Musik
     Bilder

Jenseits des Üblichen ....
    

Besucherstatistik

Besucherstatistik

Monitoring

Montastic status badge