wasmparser: Implement no-hash-maps crate feature & support (#1521)

* move map.rs into its own folder

* add no-hash-maps crate feature to wasmparser

Disabled by default.

* rename Hash{Map,Set} type aliases to just Map,Set

* switch to BTree{Map,Set} if no-hash-maps is enabled

This introduced a lot of compile errors in case no-hash-maps is actually enabled since many keys are not Ord.

* use BTree{Map,Set} type aliases with no-hash-maps

Some methods are not shared between Hash{Map,Set} and BTree{Map,Set} such as reserve and the Key type has different trait bounds (Eq + Hash) vs (Eq + Ord) which can be confusing when using map and set in wasmparser during development.

Thus we actually do not want to use type aliases but proper types that wraps both Hash{Map,Set} and BTree{Map,Set} and provide a unified API.

The same is true for Index{Map,Set} but we will do this in another commit.

* fix intradoc links

* cfg instead of comment out

This is just a workaround until the new Map/Set type with unified API is done.

* move random state definition into their own file

I want to keep the base mod.rs as clean as possible.

* rename map module to collections

* rename random_state.rs -> hash.rs and make mod pub

* restructure collections module

* fix docs

* add Entry type def to map data structures

* update Set docs

* turn the Set type alias into an opaque type

Currently it only supports mostly the minimal API needed to make the wasmparser crate work.

* add Clone derive for Set

* turn the map type into an opaque type

Currently it only mostly supports the minimal API needed to make the wasmparser crate work again.

* fix workspace uses of wasparser collection types

* apply rustfmt

* add Iterator API for Set

* add Iterator API to Map

* simplify type defs and cfg guards

* remove unnecessary cfg guard

* simplify type aliases

* add IterMut for Map

* improve doc links

* add FromIterator impls for Map and Set

They are used in IndexMap so for symmetry I add them here too.

* add Map::reserve

symmetry: this method is used by both Set and IndexMap

* add IntoIterator API for Map and Set

symmetry: used on IndexMap

* add Keys, Values and ValuesMut iterators for Map

* improve docs of iter_mut

* add doc link

* add Map::contains_key

* add doc link

* add Index trait impl for Map

* add Extent trait impl for Map and Set

* update collections module docs

* add doc comment

* move Ord trait bound into impl

* apply rustfmt

* implement Ord and PartialOrd for KebabString

* fix errors with upcoming IndexMap type usage

* add IndexMap wrapper type

This is still missing the custom IndexMap implementation when `no-hash-maps` is enabled.

* add missing Index<usize> impl to IndexMap

* add IndexMap::swap_remove

used by IndexSet within wasmparser. We add this to keep APIs as mirrored as possible.

* add doc link in doc comment for Set

* add Set::replace

Used by IndexSet in wasmparser. Keep this for API mirroring.

* add basic set methods: Set::{is_disjoint, is_subset, is_superset}

* PartialOrd and Ord impls for some types that were still missing them

This is for IndexSet opaque type usage to work.

* add opaque IndexSet type

This does not yet support the custom implementation that avoid usage of hash maps internally.

* improve collections module docs

* improve docs of [Index]{Hash,Set} types

* apply rustfmt

* remove OccupiedEntry::[swap_]remove

We removed this API because it was unused in wasmparser and implementing it on the custom BTreeMap based IndexMap is very cumbersome to do.

* add BTreeMap based IndexMap backend

This required us to add some Clone trait bounds to parts of the IndexMap API.

* properly mirrior internal insert_full signature

* add opaque IndexSet type

* add where bounds to FusedIterator impls

* remove unused import

* add forwarding size_hint impls

* fix bug in detail::IndexMap::is_empty

* add GitHub CI job for no-hash-maps crate feature testing

* simplify API surface of detail::IndexSet

* base IndexSet impl on IndexMap<T, ()>

* complete wasmparser::collections::Map shared API surface

* complete wasmparser::collections::Set shared API surface

* fix no-hash-maps compile error

* fix bug in IndexMap::swap_remove_full impl

For the no-hash-maps custom impl.

* properly fix swap_remove_full and apply fmt

* fix doc links

* always compile custom IndexMap implementation

This is friendlier for maintenance and developers.

* add tests for custom IndexMap implementation

Author: Robbepop

.github/workflows/main.yml

@@ -56,6 +56,18 @@ jobs:
         name: bins-${{ matrix.build }}
         path: dist
 
+  test-no-hash-maps:
+    name: Test (no-hash-maps)
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: true
+      - name: Install Rust
+        run: rustup update stable --no-self-update && rustup default stable
+      - name: Test (no-hash-maps)
+        run: cargo test --workspace --locked --features no-hash-maps
+
   test:
     name: Test
     runs-on: ${{ matrix.os }}

crates/wasm-compose/src/encoding.rs

@@ -730,7 +730,10 @@ impl<'a> TypeEncoder<'a> {
         index
     }
 
-    fn flags(encodable: &mut Encodable, names: &wasmparser::map::IndexSet<KebabString>) -> u32 {
+    fn flags(
+        encodable: &mut Encodable,
+        names: &wasmparser::collections::IndexSet<KebabString>,
+    ) -> u32 {
         let index = encodable.type_count();
         encodable
             .ty()
@@ -739,7 +742,10 @@ impl<'a> TypeEncoder<'a> {
         index
     }
 
-    fn enum_type(encodable: &mut Encodable, cases: &wasmparser::map::IndexSet<KebabString>) -> u32 {
+    fn enum_type(
+        encodable: &mut Encodable,
+        cases: &wasmparser::collections::IndexSet<KebabString>,
+    ) -> u32 {
         let index = encodable.type_count();
         encodable
             .ty()

crates/wasmparser/Cargo.toml

@@ -45,6 +45,12 @@ default = ['std', 'validate']
 # library's source of randomness for seeding hash maps.
 std = ['indexmap/std']
 
+# Tells the wasmparser crate to avoid using hash based maps and sets.
+# 
+# Some embedded environments cannot provide a random source which is required
+# to properly initialize hashmap based data structures for resilience against
+# malious actors that control their inputs.
+no-hash-maps = []
 # A feature that enables validating WebAssembly files. This is enabled by
 # default but not required if you're only parsing a file, for example, as
 # opposed to validating all of its contents.

crates/wasmparser/src/map.rs crates/wasmparser/src/collections/hash.rscrates/wasmparser/src/map.rs renamed to crates/wasmparser/src/collections/hash.rs

@@ -1,24 +1,7 @@
-//! Type aliases for maps used by `wasmparser`
-//!
-//! This module contains type aliases used for [`HashMap`], [`HashSet`],
-//! [`IndexMap`], and [`IndexSet`]. Note that these differ from upstream types
-//! in the `indexmap` crate and the standard library due to customization of the
-//! hash algorithm type parameter.
+//! Utilities for hashmap initialization based on random sources.
 
 use core::hash::{BuildHasher, Hasher};
 
-/// Wasmparser-specific type alias for an ordered map.
-pub type IndexMap<K, V> = indexmap::IndexMap<K, V, RandomState>;
-
-/// Wasmparser-specific type alias for an ordered set.
-pub type IndexSet<K> = indexmap::IndexSet<K, RandomState>;
-
-/// Wasmparser-specific type alias for hash map.
-pub type HashMap<K, V> = hashbrown::HashMap<K, V, RandomState>;
-
-/// Wasmparser-specific type alias for hash set.
-pub type HashSet<K> = hashbrown::HashSet<K, RandomState>;
-
 /// Wasmparser's hashing state stored per-map.
 ///
 /// This is DoS-resistant when the `std` feature is activated and still somewhat