transmutability: ensure_sufficient_stack when answering query

This commit is contained in:
Tomasz Miąsko
2025-04-30 11:40:36 +02:00
parent 88a86794b9
commit 0138df1f3d
@@ -1,3 +1,4 @@
use rustc_data_structures::stack::ensure_sufficient_stack;
use tracing::{debug, instrument, trace};
pub(crate) mod query_context;
@@ -149,130 +150,137 @@ fn answer_memo(
if let Some(answer) = cache.get(&(src_state, dst_state)) {
answer.clone()
} else {
debug!(?src_state, ?dst_state);
debug!(src = ?self.src);
debug!(dst = ?self.dst);
debug!(
src_transitions_len = self.src.transitions.len(),
dst_transitions_len = self.dst.transitions.len()
);
let answer = if dst_state == self.dst.accept {
// truncation: `size_of(Src) >= size_of(Dst)`
//
// Why is truncation OK to do? Because even though the Src is bigger, all we care about
// is whether we have enough data for the Dst to be valid in accordance with what its
// type dictates.
// For example, in a u8 to `()` transmutation, we have enough data available from the u8
// to transmute it to a `()` (though in this case does `()` really need any data to
// begin with? It doesn't). Same thing with u8 to fieldless struct.
// Now then, why is something like u8 to bool not allowed? That is not because the bool
// is smaller in size, but rather because those 2 bits that we are re-interpreting from
// the u8 could introduce invalid states for the bool type.
//
// So, if it's possible to transmute to a smaller Dst by truncating, and we can guarantee
// that none of the actually-used data can introduce an invalid state for Dst's type, we
// are able to safely transmute, even with truncation.
Answer::Yes
} else if src_state == self.src.accept {
// extension: `size_of(Src) <= size_of(Dst)`
if let Some(dst_state_prime) = self.dst.get_uninit_edge_dst(dst_state) {
self.answer_memo(cache, src_state, dst_state_prime)
} else {
Answer::No(Reason::DstIsTooBig)
}
} else {
let src_quantifier = if self.assume.validity {
// if the compiler may assume that the programmer is doing additional validity checks,
// (e.g.: that `src != 3u8` when the destination type is `bool`)
// then there must exist at least one transition out of `src_state` such that the transmute is viable...
Quantifier::ThereExists
} else {
// if the compiler cannot assume that the programmer is doing additional validity checks,
// then for all transitions out of `src_state`, such that the transmute is viable...
// then there must exist at least one transition out of `dst_state` such that the transmute is viable...
Quantifier::ForAll
};
let bytes_answer = src_quantifier.apply(
union(self.src.bytes_from(src_state), self.dst.bytes_from(dst_state))
.filter_map(|(_range, (src_state_prime, dst_state_prime))| {
match (src_state_prime, dst_state_prime) {
// No matching transitions in `src`. Skip.
(None, _) => None,
// No matching transitions in `dst`. Fail.
(Some(_), None) => Some(Answer::No(Reason::DstIsBitIncompatible)),
// Matching transitions. Continue with successor states.
(Some(src_state_prime), Some(dst_state_prime)) => {
Some(self.answer_memo(cache, src_state_prime, dst_state_prime))
}
}
}),
);
// The below early returns reflect how this code would behave:
// if self.assume.validity {
// or(bytes_answer, refs_answer)
// } else {
// and(bytes_answer, refs_answer)
// }
// ...if `refs_answer` was computed lazily. The below early
// returns can be deleted without impacting the correctness of
// the algorithm; only its performance.
debug!(?bytes_answer);
match bytes_answer {
Answer::No(_) if !self.assume.validity => return bytes_answer,
Answer::Yes if self.assume.validity => return bytes_answer,
_ => {}
};
let refs_answer = src_quantifier.apply(
// for each reference transition out of `src_state`...
self.src.refs_from(src_state).map(|(src_ref, src_state_prime)| {
// ...there exists a reference transition out of `dst_state`...
Quantifier::ThereExists.apply(self.dst.refs_from(dst_state).map(
|(dst_ref, dst_state_prime)| {
if !src_ref.is_mutable() && dst_ref.is_mutable() {
Answer::No(Reason::DstIsMoreUnique)
} else if !self.assume.alignment
&& src_ref.min_align() < dst_ref.min_align()
{
Answer::No(Reason::DstHasStricterAlignment {
src_min_align: src_ref.min_align(),
dst_min_align: dst_ref.min_align(),
})
} else if dst_ref.size() > src_ref.size() {
Answer::No(Reason::DstRefIsTooBig {
src: src_ref,
dst: dst_ref,
})
} else {
// ...such that `src` is transmutable into `dst`, if
// `src_ref` is transmutability into `dst_ref`.
and(
Answer::If(Condition::IfTransmutable {
src: src_ref,
dst: dst_ref,
}),
self.answer_memo(cache, src_state_prime, dst_state_prime),
)
}
},
))
}),
);
if self.assume.validity {
or(bytes_answer, refs_answer)
} else {
and(bytes_answer, refs_answer)
}
};
let answer = ensure_sufficient_stack(|| self.answer_impl(cache, src_state, dst_state));
if let Some(..) = cache.insert((src_state, dst_state), answer.clone()) {
panic!("failed to correctly cache transmutability")
}
answer
}
}
fn answer_impl(
&self,
cache: &mut Map<(dfa::State, dfa::State), Answer<<C as QueryContext>::Ref>>,
src_state: dfa::State,
dst_state: dfa::State,
) -> Answer<<C as QueryContext>::Ref> {
debug!(?src_state, ?dst_state);
debug!(src = ?self.src);
debug!(dst = ?self.dst);
debug!(
src_transitions_len = self.src.transitions.len(),
dst_transitions_len = self.dst.transitions.len()
);
if dst_state == self.dst.accept {
// truncation: `size_of(Src) >= size_of(Dst)`
//
// Why is truncation OK to do? Because even though the Src is bigger, all we care about
// is whether we have enough data for the Dst to be valid in accordance with what its
// type dictates.
// For example, in a u8 to `()` transmutation, we have enough data available from the u8
// to transmute it to a `()` (though in this case does `()` really need any data to
// begin with? It doesn't). Same thing with u8 to fieldless struct.
// Now then, why is something like u8 to bool not allowed? That is not because the bool
// is smaller in size, but rather because those 2 bits that we are re-interpreting from
// the u8 could introduce invalid states for the bool type.
//
// So, if it's possible to transmute to a smaller Dst by truncating, and we can guarantee
// that none of the actually-used data can introduce an invalid state for Dst's type, we
// are able to safely transmute, even with truncation.
Answer::Yes
} else if src_state == self.src.accept {
// extension: `size_of(Src) <= size_of(Dst)`
if let Some(dst_state_prime) = self.dst.get_uninit_edge_dst(dst_state) {
self.answer_memo(cache, src_state, dst_state_prime)
} else {
Answer::No(Reason::DstIsTooBig)
}
} else {
let src_quantifier = if self.assume.validity {
// if the compiler may assume that the programmer is doing additional validity checks,
// (e.g.: that `src != 3u8` when the destination type is `bool`)
// then there must exist at least one transition out of `src_state` such that the transmute is viable...
Quantifier::ThereExists
} else {
// if the compiler cannot assume that the programmer is doing additional validity checks,
// then for all transitions out of `src_state`, such that the transmute is viable...
// then there must exist at least one transition out of `dst_state` such that the transmute is viable...
Quantifier::ForAll
};
let bytes_answer = src_quantifier.apply(
union(self.src.bytes_from(src_state), self.dst.bytes_from(dst_state)).filter_map(
|(_range, (src_state_prime, dst_state_prime))| {
match (src_state_prime, dst_state_prime) {
// No matching transitions in `src`. Skip.
(None, _) => None,
// No matching transitions in `dst`. Fail.
(Some(_), None) => Some(Answer::No(Reason::DstIsBitIncompatible)),
// Matching transitions. Continue with successor states.
(Some(src_state_prime), Some(dst_state_prime)) => {
Some(self.answer_memo(cache, src_state_prime, dst_state_prime))
}
}
},
),
);
// The below early returns reflect how this code would behave:
// if self.assume.validity {
// or(bytes_answer, refs_answer)
// } else {
// and(bytes_answer, refs_answer)
// }
// ...if `refs_answer` was computed lazily. The below early
// returns can be deleted without impacting the correctness of
// the algorithm; only its performance.
debug!(?bytes_answer);
match bytes_answer {
Answer::No(_) if !self.assume.validity => return bytes_answer,
Answer::Yes if self.assume.validity => return bytes_answer,
_ => {}
};
let refs_answer = src_quantifier.apply(
// for each reference transition out of `src_state`...
self.src.refs_from(src_state).map(|(src_ref, src_state_prime)| {
// ...there exists a reference transition out of `dst_state`...
Quantifier::ThereExists.apply(self.dst.refs_from(dst_state).map(
|(dst_ref, dst_state_prime)| {
if !src_ref.is_mutable() && dst_ref.is_mutable() {
Answer::No(Reason::DstIsMoreUnique)
} else if !self.assume.alignment
&& src_ref.min_align() < dst_ref.min_align()
{
Answer::No(Reason::DstHasStricterAlignment {
src_min_align: src_ref.min_align(),
dst_min_align: dst_ref.min_align(),
})
} else if dst_ref.size() > src_ref.size() {
Answer::No(Reason::DstRefIsTooBig { src: src_ref, dst: dst_ref })
} else {
// ...such that `src` is transmutable into `dst`, if
// `src_ref` is transmutability into `dst_ref`.
and(
Answer::If(Condition::IfTransmutable {
src: src_ref,
dst: dst_ref,
}),
self.answer_memo(cache, src_state_prime, dst_state_prime),
)
}
},
))
}),
);
if self.assume.validity {
or(bytes_answer, refs_answer)
} else {
and(bytes_answer, refs_answer)
}
}
}
}
fn and<R>(lhs: Answer<R>, rhs: Answer<R>) -> Answer<R>