miri/shims/unix/linux_like/

sync.rs

use crate::concurrency::sync::{FutexRef, SyncObj};
use crate::shims::sig::check_min_vararg_count;
use crate::*;

struct LinuxFutex {
    futex: FutexRef,
}

impl SyncObj for LinuxFutex {}

/// Implementation of the SYS_futex syscall.
/// `args` is the arguments *including* the syscall number.
pub fn futex<'tcx>(
    ecx: &mut MiriInterpCx<'tcx>,
    varargs: &[OpTy<'tcx>],
    dest: &MPlaceTy<'tcx>,
) -> InterpResult<'tcx> {
    let [addr, op, val] = check_min_vararg_count("`syscall(SYS_futex, ...)`", varargs)?;

    // See <https://man7.org/linux/man-pages/man2/futex.2.html> for docs.
    // The first three arguments (after the syscall number itself) are the same to all futex operations:
    //     (uint32_t *addr, int op, uint32_t val).
    // We checked above that these definitely exist.
    let addr = ecx.read_pointer(addr)?;
    let op = ecx.read_scalar(op)?.to_i32()?;
    let val = ecx.read_scalar(val)?.to_u32()?;

    // This is a vararg function so we have to bring our own type for this pointer.
    let addr = ecx.ptr_to_mplace(addr, ecx.machine.layouts.i32);

    let futex_private = ecx.eval_libc_i32("FUTEX_PRIVATE_FLAG");
    let futex_wait = ecx.eval_libc_i32("FUTEX_WAIT");
    let futex_wait_bitset = ecx.eval_libc_i32("FUTEX_WAIT_BITSET");
    let futex_wake = ecx.eval_libc_i32("FUTEX_WAKE");
    let futex_wake_bitset = ecx.eval_libc_i32("FUTEX_WAKE_BITSET");
    let futex_realtime = ecx.eval_libc_i32("FUTEX_CLOCK_REALTIME");

    // FUTEX_PRIVATE enables an optimization that stops it from working across processes.
    // Miri doesn't support that anyway, so we ignore that flag.
    match op & !futex_private {
        // FUTEX_WAIT: (int *addr, int op = FUTEX_WAIT, int val, const timespec *timeout)
        // Blocks the thread if *addr still equals val. Wakes up when FUTEX_WAKE is called on the same address,
        // or *timeout expires. `timeout == null` for an infinite timeout.
        //
        // FUTEX_WAIT_BITSET: (int *addr, int op = FUTEX_WAIT_BITSET, int val, const timespec *timeout, int *_ignored, unsigned int bitset)
        // This is identical to FUTEX_WAIT, except:
        //  - The timeout is absolute rather than relative.
        //  - You can specify the bitset to selecting what WAKE operations to respond to.
        op if op & !futex_realtime == futex_wait || op & !futex_realtime == futex_wait_bitset => {
            let wait_bitset = op & !futex_realtime == futex_wait_bitset;

            let (timeout, bitset) = if wait_bitset {
                let [_, _, _, timeout, uaddr2, bitset] = check_min_vararg_count(
                    "`syscall(SYS_futex, FUTEX_WAIT_BITSET, ...)`",
                    varargs,
                )?;
                let _timeout = ecx.read_pointer(timeout)?;
                let _uaddr2 = ecx.read_pointer(uaddr2)?;
                (timeout, ecx.read_scalar(bitset)?.to_u32()?)
            } else {
                let [_, _, _, timeout] =
                    check_min_vararg_count("`syscall(SYS_futex, FUTEX_WAIT, ...)`", varargs)?;
                (timeout, u32::MAX)
            };

            if bitset == 0 {
                return ecx.set_last_error_and_return(LibcError("EINVAL"), dest);
            }

            let timeout = ecx.deref_pointer_as(timeout, ecx.libc_ty_layout("timespec"))?;
            let timeout = if ecx.ptr_is_null(timeout.ptr())? {
                None
            } else {
                let duration = match ecx.read_timespec(&timeout)? {
                    Some(duration) => duration,
                    None => {
                        return ecx.set_last_error_and_return(LibcError("EINVAL"), dest);
                    }
                };
                let timeout_clock = if op & futex_realtime == futex_realtime {
                    ecx.check_no_isolation(
                        "`futex` syscall with `op=FUTEX_WAIT` and non-null timeout with `FUTEX_CLOCK_REALTIME`",
                    )?;
                    TimeoutClock::RealTime
                } else {
                    TimeoutClock::Monotonic
                };
                let timeout_anchor = if wait_bitset {
                    // FUTEX_WAIT_BITSET uses an absolute timestamp.
                    TimeoutAnchor::Absolute
                } else {
                    // FUTEX_WAIT uses a relative timestamp.
                    TimeoutAnchor::Relative
                };
                Some((timeout_clock, timeout_anchor, duration))
            };
            // There may be a concurrent thread changing the value of addr
            // and then invoking the FUTEX_WAKE syscall. It is critical that the
            // effects of this and the other thread are correctly observed,
            // otherwise we will deadlock.
            //
            // There are two scenarios to consider, depending on whether WAIT or WAKE goes first:
            // 1. If we (FUTEX_WAIT) execute first, we'll push ourselves into the waiters queue and
            //    go to sleep. They (FUTEX_WAKE) will see us in the queue and wake us up. It doesn't
            //    matter how the addr write is ordered.
            // 2. If they (FUTEX_WAKE) execute first, that means the addr write is also before us
            //    (FUTEX_WAIT). It is crucial that we observe addr's new value. If we see an
            //    outdated value that happens to equal the expected val, then we'll put ourselves to
            //    sleep with no one to wake us up, so we end up with a deadlock. This is prevented
            //    by having a SeqCst fence inside FUTEX_WAKE syscall, and another SeqCst fence here
            //    in FUTEX_WAIT. The atomic read on addr after the SeqCst fence is guaranteed not to
            //    see any value older than the addr write immediately before calling FUTEX_WAKE.
            //    We'll see futex_val != val and return without sleeping.
            //
            //    Note that the fences do not create any happens-before relationship.
            //    The read sees the write immediately before the fence not because
            //    one happens after the other, but is instead due to a guarantee unique
            //    to SeqCst fences that restricts what an atomic read placed AFTER the
            //    fence can see. The read still has to be atomic, otherwise it's a data
            //    race. This guarantee cannot be achieved with acquire-release fences
            //    since they only talk about reads placed BEFORE a fence - and places
            //    no restrictions on what the read itself can see, only that there is
            //    a happens-before between the fences IF the read happens to see the
            //    right value. This is useless to us, since we need the read itself
            //    to see an up-to-date value.
            //
            // The above case distinction is valid since both FUTEX_WAIT and FUTEX_WAKE
            // contain a SeqCst fence, therefore inducing a total order between the operations.
            // It is also critical that the fence, the atomic load, and the comparison in FUTEX_WAIT
            // altogether happen atomically. If the other thread's fence in FUTEX_WAKE
            // gets interleaved after our fence, then we lose the guarantee on the
            // atomic load being up-to-date; if the other thread's write on addr and FUTEX_WAKE
            // call are interleaved after the load but before the comparison, then we get a TOCTOU
            // race condition, and go to sleep thinking the other thread will wake us up,
            // even though they have already finished.
            //
            // Thankfully, preemptions cannot happen inside a Miri shim, so we do not need to
            // do anything special to guarantee fence-load-comparison atomicity.
            ecx.atomic_fence(AtomicFenceOrd::SeqCst)?;
            // Read an `i32` through the pointer, regardless of any wrapper types.
            // It's not uncommon for `addr` to be passed as another type than `*mut i32`, such as `*const AtomicI32`.
            // We do an acquire read -- it only seems reasonable that if we observe a value here, we
            // actually establish an ordering with that value.
            let futex_val = ecx.read_scalar_atomic(&addr, AtomicReadOrd::Acquire)?.to_u32()?;
            if val == futex_val {
                // The value still matches, so we block the thread and make it wait for FUTEX_WAKE.

                // This cannot fail since we already did an atomic acquire read on that pointer.
                // Acquire reads are only allowed on mutable memory.
                let futex_ref = ecx
                    .get_sync_or_init(addr.ptr(), |_| LinuxFutex { futex: Default::default() })
                    .unwrap()
                    .futex
                    .clone();

                let dest = dest.clone();
                ecx.futex_wait(
                    futex_ref,
                    bitset,
                    timeout,
                    callback!(
                        @capture<'tcx> {
                            dest: MPlaceTy<'tcx>,
                        }
                        |ecx, unblock: UnblockKind| match unblock {
                            UnblockKind::Ready => {
                                ecx.write_int(0, &dest)
                            }
                            UnblockKind::TimedOut => {
                                ecx.set_last_error_and_return(LibcError("ETIMEDOUT"), &dest)
                            }
                        }
                    ),
                );
            } else {
                // The futex value doesn't match the expected value, so we return failure
                // right away without sleeping: -1 and errno set to EAGAIN.
                return ecx.set_last_error_and_return(LibcError("EAGAIN"), dest);
            }
        }
        // FUTEX_WAKE: (int *addr, int op = FUTEX_WAKE, int val)
        // Wakes at most `val` threads waiting on the futex at `addr`.
        // Returns the amount of threads woken up.
        // Does not access the futex value at *addr.
        // FUTEX_WAKE_BITSET: (int *addr, int op = FUTEX_WAKE, int val, const timespect *_unused, int *_unused, unsigned int bitset)
        // Same as FUTEX_WAKE, but allows you to specify a bitset to select which threads to wake up.
        op if op == futex_wake || op == futex_wake_bitset => {
            let Some(futex_ref) =
                ecx.get_sync_or_init(addr.ptr(), |_| LinuxFutex { futex: Default::default() })
            else {
                // No AllocId, or no live allocation at that AllocId.
                // Return an error code. (That seems nicer than silently doing something non-intuitive.)
                // This means that if an address gets reused by a new allocation,
                // we'll use an independent futex queue for this... that seems acceptable.
                return ecx.set_last_error_and_return(LibcError("EFAULT"), dest);
            };
            let futex_ref = futex_ref.futex.clone();

            let bitset = if op == futex_wake_bitset {
                let [_, _, _, timeout, uaddr2, bitset] = check_min_vararg_count(
                    "`syscall(SYS_futex, FUTEX_WAKE_BITSET, ...)`",
                    varargs,
                )?;
                let _timeout = ecx.read_pointer(timeout)?;
                let _uaddr2 = ecx.read_pointer(uaddr2)?;
                ecx.read_scalar(bitset)?.to_u32()?
            } else {
                u32::MAX
            };
            if bitset == 0 {
                return ecx.set_last_error_and_return(LibcError("EINVAL"), dest);
            }
            // Together with the SeqCst fence in futex_wait, this makes sure that futex_wait
            // will see the latest value on addr which could be changed by our caller
            // before doing the syscall.
            ecx.atomic_fence(AtomicFenceOrd::SeqCst)?;
            let woken = ecx.futex_wake(&futex_ref, bitset, val.try_into().unwrap())?;
            ecx.write_scalar(Scalar::from_target_isize(woken.try_into().unwrap(), ecx), dest)?;
        }
        op => throw_unsup_format!("Miri does not support `futex` syscall with op={}", op),
    }

    interp_ok(())
}