fsys 0.6.0

Adaptive file and directory IO for Rust — fast, hardware-aware, multi-strategy.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
//! [`AlignedBufferPool`] — per-handle pool of reusable aligned
//! allocations for Direct IO.
//!
//! ## Design
//!
//! Lock-free fast path on top of
//! [`crossbeam_queue::ArrayQueue`] for buffer return + reuse. The
//! slow path (genuinely empty pool, all `capacity` buffers leased)
//! takes a [`std::sync::Mutex`] + [`std::sync::Condvar`] so the
//! caller can `wait` for a returned buffer without spinning.
//!
//! Allocations are lazy: `AlignedBufferPool::new` does not allocate
//! buffer memory. The first `capacity` `lease()` calls allocate
//! fresh `AlignedBuffer`s; subsequent calls reuse from the queue.
//! Idle handles cost zero buffer memory beyond the
//! `Arc<PoolInner>`'s metadata.

use std::alloc::{self, Layout};
use std::ptr::NonNull;
use std::sync::atomic::{AtomicUsize, Ordering};
use std::sync::{Arc, Condvar, Mutex};

use crossbeam_queue::ArrayQueue;

use super::aligned::{AlignedBuffer, RawAllocation};
use crate::Error;
use crate::Result;

/// Per-handle aligned buffer pool.
///
/// Cheap to construct — `new` only sets up the `Arc<PoolInner>`.
/// Buffer allocations happen lazily on the first `capacity`
/// `lease()` calls. Cloning the pool is `O(1)` Arc clone — the
/// dispatcher and caller threads can hold cheap clones to lease
/// concurrently.
#[derive(Clone)]
pub(crate) struct AlignedBufferPool {
    inner: Arc<PoolInner>,
}

impl AlignedBufferPool {
    /// Constructs a new pool.
    ///
    /// `capacity` — total buffers the pool can have outstanding.
    /// `block_size` — bytes per buffer. Must be a non-zero multiple
    /// of `block_align`.
    /// `block_align` — alignment in bytes. Must be a power of two and
    /// non-zero.
    ///
    /// # Errors
    ///
    /// Returns [`Error::AlignmentRequired`] when `block_align` is not
    /// a power of two, when `block_size == 0`, or when `block_size`
    /// is not a multiple of `block_align`. These violations make
    /// `Layout::from_size_align` fail and would render every
    /// allocation invalid.
    pub(crate) fn new(capacity: usize, block_size: usize, block_align: usize) -> Result<Self> {
        if block_align == 0 || !block_align.is_power_of_two() {
            return Err(Error::AlignmentRequired {
                detail: "buffer pool block_align must be a non-zero power of two",
            });
        }
        if block_size == 0 || block_size % block_align != 0 {
            return Err(Error::AlignmentRequired {
                detail: "buffer pool block_size must be a non-zero multiple of block_align",
            });
        }
        // Validate the layout produces successfully — catches
        // overflow / zero-size combinations that the above checks
        // theoretically miss.
        let block_layout = Layout::from_size_align(block_size, block_align).map_err(|_| {
            Error::AlignmentRequired {
                detail: "buffer pool layout rejected by Layout::from_size_align",
            }
        })?;

        // ArrayQueue panics on capacity 0; treat that as an error.
        if capacity == 0 {
            return Err(Error::AlignmentRequired {
                detail: "buffer pool capacity must be > 0",
            });
        }

        Ok(Self {
            inner: Arc::new(PoolInner {
                free: ArrayQueue::new(capacity),
                allocated: AtomicUsize::new(0),
                capacity,
                block_layout,
                wait_lock: Mutex::new(()),
                wait_cv: Condvar::new(),
            }),
        })
    }

    /// Leases a buffer from the pool. Blocks if the pool is
    /// exhausted (per checkpoint B' R-2' lock-in) until a buffer is
    /// returned.
    pub(crate) fn lease(&self) -> AlignedBuffer {
        self.inner.lease()
    }

    /// Returns the configured block size in bytes.
    #[allow(dead_code)] // surfaced via Handle integration in F+G
    pub(crate) fn block_size(&self) -> usize {
        self.inner.block_layout.size()
    }

    /// Returns the configured block alignment in bytes.
    #[allow(dead_code)] // surfaced via Handle integration in F+G
    pub(crate) fn block_align(&self) -> usize {
        self.inner.block_layout.align()
    }

    /// Returns the configured pool capacity (max outstanding buffers).
    #[allow(dead_code)] // surfaced via Handle integration in F+G
    pub(crate) fn capacity(&self) -> usize {
        self.inner.capacity
    }
}

/// Internal pool state, held inside an `Arc` and shared between the
/// pool handle and every outstanding [`AlignedBuffer`] (so leased
/// buffers can return to the queue on Drop even if the
/// [`AlignedBufferPool`] handle has been moved/cloned).
pub(super) struct PoolInner {
    free: ArrayQueue<RawAllocation>,
    allocated: AtomicUsize,
    capacity: usize,
    block_layout: Layout,
    /// Companion to `wait_cv`. Held only on the slow path (genuinely
    /// empty pool); the lock-free fast path bypasses it entirely.
    wait_lock: Mutex<()>,
    wait_cv: Condvar,
}

impl PoolInner {
    fn lease(self: &Arc<Self>) -> AlignedBuffer {
        loop {
            // Fast path — try the free queue.
            if let Some(raw) = self.free.pop() {
                return AlignedBuffer::new(raw, Arc::clone(self));
            }

            // Slow path: try to allocate fresh if under capacity.
            let allocated = self.allocated.load(Ordering::Acquire);
            if allocated < self.capacity {
                if self
                    .allocated
                    .compare_exchange(
                        allocated,
                        allocated + 1,
                        Ordering::AcqRel,
                        Ordering::Acquire,
                    )
                    .is_ok()
                {
                    let raw = match self.alloc_fresh() {
                        Some(r) => r,
                        None => {
                            // Allocator failed. Roll back our reservation
                            // so future leases can retry.
                            let _ = self.allocated.fetch_sub(1, Ordering::AcqRel);
                            // Wait for someone else to return a buffer.
                            self.wait_for_return();
                            continue;
                        }
                    };
                    return AlignedBuffer::new(raw, Arc::clone(self));
                }
                // CAS race lost — another thread took our slot. Retry
                // the loop to see if a buffer became available.
                continue;
            }

            // Pool fully allocated and queue empty: wait for return.
            self.wait_for_return();
            // After being woken, loop re-checks both fast and slow
            // paths.
        }
    }

    fn wait_for_return(&self) {
        // Take the lock and re-check before waiting to avoid lost
        // wakeups: a buffer may have been returned between our last
        // check and the lock acquisition.
        let guard = match self.wait_lock.lock() {
            Ok(g) => g,
            Err(p) => p.into_inner(),
        };
        if !self.free.is_empty() || self.allocated.load(Ordering::Acquire) < self.capacity {
            return;
        }
        // Bind the lock guard so the `let_underscore_lock` lint is
        // satisfied; the guard drops at the end of this scope which
        // releases the mutex (the wait happens *atomically* on the
        // Condvar regardless).
        let _woken_guard = self.wait_cv.wait(guard);
    }

    fn alloc_fresh(&self) -> Option<RawAllocation> {
        // SAFETY: `block_layout` was validated at pool construction
        // (non-zero size, power-of-two alignment); the allocator
        // contract for `alloc_zeroed` accepts any non-zero-size
        // Layout and returns either a valid pointer or null.
        let ptr = unsafe { alloc::alloc_zeroed(self.block_layout) };
        let nn = NonNull::new(ptr)?;
        Some(RawAllocation {
            ptr: nn,
            layout: self.block_layout,
        })
    }

    /// Returns a leased allocation to the free queue. Called from
    /// [`AlignedBuffer::drop`].
    pub(super) fn return_allocation(&self, raw: RawAllocation) {
        // ArrayQueue::push returns `Err` only if the queue is at
        // capacity. By construction, the queue's capacity equals
        // `self.capacity` and we never have more than that many
        // buffers outstanding, so push always succeeds. Defensive:
        // if it ever fails, deallocate immediately rather than leak.
        if let Err(raw) = self.free.push(raw) {
            // SAFETY: `raw.ptr` was returned by `alloc_zeroed` with
            // `raw.layout` and is currently owned by us (Drop just
            // moved it here). Releasing it now is sound; we also
            // decrement `allocated` so the pool can re-allocate
            // later.
            unsafe {
                alloc::dealloc(raw.ptr.as_ptr(), raw.layout);
            }
            let _ = self.allocated.fetch_sub(1, Ordering::AcqRel);
        }
        // Wake any thread blocked in `wait_for_return`. notify_one
        // is fine because at most one waiter can take this single
        // returned buffer.
        self.wait_cv.notify_one();
    }
}

impl Drop for PoolInner {
    fn drop(&mut self) {
        // Deallocate any buffers still sitting in the free queue. In
        // steady state, the pool drops only after the Handle drops,
        // and at that point all outstanding leases have returned to
        // the queue (or are still alive holding their `Arc<PoolInner>`,
        // which keeps `PoolInner` alive — so `Drop` only runs once
        // truly the last reference is gone).
        while let Some(raw) = self.free.pop() {
            // SAFETY: each `raw` was returned by `alloc_zeroed` with
            // `raw.layout`; we own it exclusively (it just came out
            // of the queue, which had the only reference); freeing
            // is sound.
            unsafe {
                alloc::dealloc(raw.ptr.as_ptr(), raw.layout);
            }
        }
    }
}

// Compile-time assertion: AlignedBufferPool must be Send + Sync so
// it can live inside a Handle that is shared across threads.
const _: () = {
    fn assert_send<T: Send>() {}
    fn assert_sync<T: Sync>() {}
    fn check() {
        assert_send::<AlignedBufferPool>();
        assert_sync::<AlignedBufferPool>();
        assert_send::<AlignedBuffer>();
    }
    let _ = check;
};

#[cfg(test)]
mod tests {
    use super::*;
    use std::sync::Arc;

    #[test]
    fn test_pool_new_validates_alignment_power_of_two() {
        let r = AlignedBufferPool::new(4, 4096, 7);
        assert!(r.is_err(), "non-power-of-two alignment must error");
    }

    #[test]
    fn test_pool_new_validates_size_multiple_of_align() {
        let r = AlignedBufferPool::new(4, 4097, 512);
        assert!(r.is_err(), "size not a multiple of align must error");
    }

    #[test]
    fn test_pool_new_validates_zero_capacity() {
        let r = AlignedBufferPool::new(0, 4096, 512);
        assert!(r.is_err(), "zero capacity must error");
    }

    #[test]
    fn test_pool_new_validates_zero_block_size() {
        let r = AlignedBufferPool::new(4, 0, 512);
        assert!(r.is_err(), "zero block size must error");
    }

    #[test]
    fn test_pool_new_succeeds_with_valid_inputs() {
        let p = AlignedBufferPool::new(64, 4096, 512).expect("valid pool");
        assert_eq!(p.capacity(), 64);
        assert_eq!(p.block_size(), 4096);
        assert_eq!(p.block_align(), 512);
    }

    #[test]
    fn test_lease_returns_aligned_buffer_with_correct_size() {
        let p = AlignedBufferPool::new(4, 4096, 512).expect("pool");
        let b = p.lease();
        assert_eq!(b.len(), 4096);
        assert_eq!(b.align(), 512);
        assert_eq!((b.as_ptr() as usize) % 512, 0);
    }

    #[test]
    fn test_lease_returns_zero_initialised_buffer() {
        let p = AlignedBufferPool::new(4, 4096, 512).expect("pool");
        let b = p.lease();
        assert!(b.as_slice().iter().all(|&x| x == 0));
    }

    #[test]
    fn test_drop_returns_buffer_to_pool() {
        let p = AlignedBufferPool::new(2, 4096, 512).expect("pool");
        // Lease two buffers (allocated count = 2).
        let b1 = p.lease();
        let b2 = p.lease();
        // Capture pointers before drop.
        let p1 = b1.as_ptr() as usize;
        let p2 = b2.as_ptr() as usize;
        drop(b1);
        drop(b2);
        // Subsequent leases must reuse those allocations.
        let b3 = p.lease();
        let b4 = p.lease();
        let p3 = b3.as_ptr() as usize;
        let p4 = b4.as_ptr() as usize;
        // Either order acceptable — pool is a queue, returns FIFO.
        let returned = [p3, p4];
        assert!(returned.contains(&p1), "buf 1 should be reused");
        assert!(returned.contains(&p2), "buf 2 should be reused");
    }

    #[test]
    fn test_concurrent_lease_and_return() {
        let pool = AlignedBufferPool::new(4, 4096, 512).expect("pool");
        let pool_arc = Arc::new(pool);
        let n_threads = 8;
        let leases_per_thread = 16;
        let mut handles = Vec::new();
        for _ in 0..n_threads {
            let p = Arc::clone(&pool_arc);
            handles.push(std::thread::spawn(move || {
                for _ in 0..leases_per_thread {
                    let mut b = p.lease();
                    // Touch the buffer to exercise the pointer.
                    b.as_mut_slice()[0] = 0x42;
                    // Buffer drops at end of iteration → returns to pool.
                }
            }));
        }
        for h in handles {
            h.join().expect("thread");
        }
        // After all threads complete, the pool should still be functional.
        let _ = pool_arc.lease();
    }

    #[test]
    fn test_lazy_allocation_no_buffers_until_first_lease() {
        let p = AlignedBufferPool::new(64, 4096, 512).expect("pool");
        // Before any lease, allocated count must be 0.
        assert_eq!(p.inner.allocated.load(Ordering::Acquire), 0);
        let _b = p.lease();
        // After one lease, allocated == 1.
        assert_eq!(p.inner.allocated.load(Ordering::Acquire), 1);
    }

    #[test]
    fn test_lease_blocks_when_capacity_exhausted_then_succeeds_after_return() {
        // capacity = 1: once leased, the next lease must wait until
        // the first one is returned.
        let pool = AlignedBufferPool::new(1, 4096, 512).expect("pool");
        let pool_arc = Arc::new(pool);

        let b1 = pool_arc.lease();
        let p1_ptr = b1.as_ptr() as usize;

        let p_clone = Arc::clone(&pool_arc);
        let handle = std::thread::spawn(move || {
            // This must block because the only buffer is held.
            let b2 = p_clone.lease();
            b2.as_ptr() as usize
        });

        // Give the spawned thread a moment to reach `wait_for_return`.
        std::thread::sleep(std::time::Duration::from_millis(50));
        // Drop the held buffer — wakes the waiter.
        drop(b1);

        let p2_ptr = handle.join().expect("thread");
        // Same allocation reused.
        assert_eq!(p1_ptr, p2_ptr);
    }
}