Skip to content

Commit

Permalink
Auto merge of #3054 - Vanille-N:spurious-fail, r=RalfJung
Browse files Browse the repository at this point in the history
Issue discovered in TB: spurious reads are not (yet) possible in a concurrent setting

We discovered a week ago that in general, the current model of TB does not allow spurious reads because although reads provably never invalidate other reads, they migh invalidate writes.

Consider the code
```rs
fn f1(x: &u8) {}
fn f2(y: &mut u8) -> &mut u8 { &mut *y }

let mut data = 0;
let _ = thread::spawn(|| {
    f1(&mut data)
};
let _ = thread::spawn(|| {
    let y = f2(&mut data);
    *y = 42;
});
```
of which one possible interleaving is
```rs
1: retag x (&, protect) // x: [P]Frozen
2: retag y (&mut, protect) // y: [P]Reserved, x: [P]Frozen
1: return f1 // x: [P]Frozen -> Frozen, y: [P]Reserved
2: return f2 // x: Frozen, y: [P]Reserved -> Reserved
2: write y // x: Disabled, y: Active
```
that does not have UB.

Assume enough barriers to force this specific interleaving, and consider that the compiler could choose to insert a spurious read throug `x` during the call to `f1` which would produce
```rs
1: retag x (&, protect) // x: [P]Frozen
2: retag y (&mut, protect) // y: [P]Reserved, x: [P]Frozen
1: spurious read x // x:  [P]Frozen, y: [P]Reserved -> [P]Frozen
1: return f1 // x: [P]Frozen -> Frozen, y: [P]Frozen
2: return f2 // x: Frozen, y: [P]Frozen -> Frozen
2: write y // UB
```
Thus the target of the optimization (with a spurious read) has UB when the source did not.

This is bad.

SB is not affected because the code would be UB as early as `retag y`, this happens because we're trying to be a bit more subtle than that, and because the effects of a foreign read on a protected `&mut` bleed outside of the boundaries of the protector. Fortunately we have a fix planned, but in the meantime here are some `#[should_panic]` exhaustive tests to illustrate the issue.

The error message printed by the `#[should_panic]` tests flags the present issue in slightly more general terms: it says that the sequence `retag x (&, protect); retag y (&mut, protect);` produces the configuration `C_source := x: [P]Frozen, x: [P]Reserved`, and that inserting a spurious read through `x` turns it into `C_target := x: [P]Frozen, y: [P]Reserved`.
It then says that `C_source` is distinguishable from `C_target`, which means that there exists a sequence of instructions applied to both that triggers UB in `C_target` but not in `C_source`.
It happens that one such sequence is `1: return f1; 2: return f2; 2: write y;` as shown above, but it is not the only one, as for example the interleaving `1: return f1; 2: write y;` is also problematic.
  • Loading branch information
bors committed Sep 19, 2023
2 parents 13510ac + 69272a8 commit 401d897
Show file tree
Hide file tree
Showing 6 changed files with 917 additions and 146 deletions.
111 changes: 111 additions & 0 deletions src/borrow_tracker/tree_borrows/exhaustive.rs
Original file line number Diff line number Diff line change
@@ -0,0 +1,111 @@
//! Exhaustive testing utilities.
//! (These are used in Tree Borrows `#[test]`s for thorough verification
//! of the behavior of the state machine of permissions,
//! but the contents of this file are extremely generic)
#![cfg(test)]

pub trait Exhaustive: Sized {
fn exhaustive() -> Box<dyn Iterator<Item = Self>>;
}

macro_rules! precondition {
($cond:expr) => {
if !$cond {
continue;
}
};
}
pub(crate) use precondition;

// Trivial impls of `Exhaustive` for the standard types with 0, 1 and 2 elements respectively.

impl Exhaustive for ! {
fn exhaustive() -> Box<dyn Iterator<Item = Self>> {
Box::new(std::iter::empty())
}
}

impl Exhaustive for () {
fn exhaustive() -> Box<dyn Iterator<Item = Self>> {
Box::new(std::iter::once(()))
}
}

impl Exhaustive for bool {
fn exhaustive() -> Box<dyn Iterator<Item = Self>> {
Box::new(vec![true, false].into_iter())
}
}

// Some container impls for `Exhaustive`.

impl<T> Exhaustive for Option<T>
where
T: Exhaustive + 'static,
{
fn exhaustive() -> Box<dyn Iterator<Item = Self>> {
Box::new(std::iter::once(None).chain(T::exhaustive().map(Some)))
}
}

impl<T1, T2> Exhaustive for (T1, T2)
where
T1: Exhaustive + Clone + 'static,
T2: Exhaustive + 'static,
{
fn exhaustive() -> Box<dyn Iterator<Item = Self>> {
Box::new(T1::exhaustive().flat_map(|t1| T2::exhaustive().map(move |t2| (t1.clone(), t2))))
}
}

impl<T> Exhaustive for [T; 1]
where
T: Exhaustive + 'static,
{
fn exhaustive() -> Box<dyn Iterator<Item = Self>> {
Box::new(T::exhaustive().map(|t| [t]))
}
}

impl<T> Exhaustive for [T; 2]
where
T: Exhaustive + Clone + 'static,
{
fn exhaustive() -> Box<dyn Iterator<Item = Self>> {
Box::new(T::exhaustive().flat_map(|t1| T::exhaustive().map(move |t2| [t1.clone(), t2])))
}
}

impl<T> Exhaustive for [T; 3]
where
T: Exhaustive + Clone + 'static,
{
fn exhaustive() -> Box<dyn Iterator<Item = Self>> {
Box::new(
<[T; 2]>::exhaustive()
.flat_map(|[t1, t2]| T::exhaustive().map(move |t3| [t1.clone(), t2.clone(), t3])),
)
}
}

impl<T> Exhaustive for [T; 4]
where
T: Exhaustive + Clone + 'static,
{
fn exhaustive() -> Box<dyn Iterator<Item = Self>> {
Box::new(<[T; 2]>::exhaustive().flat_map(|[t1, t2]| {
<[T; 2]>::exhaustive().map(move |[t3, t4]| [t1.clone(), t2.clone(), t3, t4])
}))
}
}

impl<T> Exhaustive for [T; 5]
where
T: Exhaustive + Clone + 'static,
{
fn exhaustive() -> Box<dyn Iterator<Item = Self>> {
Box::new(<[T; 2]>::exhaustive().flat_map(|[t1, t2]| {
<[T; 3]>::exhaustive().map(move |[t3, t4, t5]| [t1.clone(), t2.clone(), t3, t4, t5])
}))
}
}
10 changes: 9 additions & 1 deletion src/borrow_tracker/tree_borrows/mod.rs
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,10 @@ pub mod diagnostics;
mod perms;
mod tree;
mod unimap;

#[cfg(test)]
mod exhaustive;

use perms::Permission;
pub use tree::Tree;

Expand Down Expand Up @@ -271,6 +275,10 @@ trait EvalContextPrivExt<'mir: 'ecx, 'tcx: 'mir, 'ecx>: crate::MiriInterpCxExt<'
diagnostics::AccessCause::Reborrow,
)?;
// Record the parent-child pair in the tree.
// FIXME: We should eventually ensure that the following `assert` holds, because
// some "exhaustive" tests consider only the initial configurations that satisfy it.
// The culprit is `Permission::new_active` in `tb_protect_place`.
//assert!(new_perm.initial_state.is_initial());
tree_borrows.new_child(orig_tag, new_tag, new_perm.initial_state, range, span)?;
drop(tree_borrows);

Expand All @@ -283,7 +291,7 @@ trait EvalContextPrivExt<'mir: 'ecx, 'tcx: 'mir, 'ecx>: crate::MiriInterpCxExt<'
// interleaving, but wether UB happens can depend on whether a write occurs in the
// future...
let is_write = new_perm.initial_state.is_active()
|| (new_perm.initial_state.is_reserved() && new_perm.protector.is_some());
|| (new_perm.initial_state.is_reserved(None) && new_perm.protector.is_some());
if is_write {
// Need to get mutable access to alloc_extra.
// (Cannot always do this as we can do read-only reborrowing on read-only allocations.)
Expand Down
Loading

0 comments on commit 401d897

Please sign in to comment.