diff options
Diffstat (limited to 'third_party/rust/wasm-smith')
| -rw-r--r-- | third_party/rust/wasm-smith/.cargo-checksum.json | 2 | ||||
| -rw-r--r-- | third_party/rust/wasm-smith/Cargo.toml | 18 | ||||
| -rw-r--r-- | third_party/rust/wasm-smith/src/component.rs | 53 | ||||
| -rw-r--r-- | third_party/rust/wasm-smith/src/config.rs | 108 | ||||
| -rw-r--r-- | third_party/rust/wasm-smith/src/core.rs | 806 | ||||
| -rw-r--r-- | third_party/rust/wasm-smith/src/core/code_builder.rs | 141 | ||||
| -rw-r--r-- | third_party/rust/wasm-smith/src/core/code_builder/no_traps.rs | 2 | ||||
| -rw-r--r-- | third_party/rust/wasm-smith/src/core/encode.rs | 41 | ||||
| -rw-r--r-- | third_party/rust/wasm-smith/src/core/terminate.rs | 30 | ||||
| -rw-r--r-- | third_party/rust/wasm-smith/src/lib.rs | 7 | ||||
| -rw-r--r-- | third_party/rust/wasm-smith/tests/available_imports.rs | 46 | ||||
| -rw-r--r-- | third_party/rust/wasm-smith/tests/common/mod.rs | 43 | ||||
| -rw-r--r-- | third_party/rust/wasm-smith/tests/core.rs | 3 | ||||
| -rw-r--r-- | third_party/rust/wasm-smith/tests/exports.rs | 147 |
14 files changed, 1025 insertions, 422 deletions
diff --git a/third_party/rust/wasm-smith/.cargo-checksum.json b/third_party/rust/wasm-smith/.cargo-checksum.json index 092ef268b6..d61edb195b 100644 --- a/third_party/rust/wasm-smith/.cargo-checksum.json +++ b/third_party/rust/wasm-smith/.cargo-checksum.json @@ -1 +1 @@ -{"files":{"Cargo.toml":"b36ff9dd80b356c2fc8e62ec88a8e6433a387d4a44b71bd4b56106da793a1558","LICENSE":"268872b9816f90fd8e85db5a28d33f8150ebb8dd016653fb39ef1f94f2686bc5","README.md":"e7b2a96cf39834aa19aed7cad8658e33445cdec7c4957658ba39e3b82cc4254b","benches/corpus.rs":"2df29556be0799f0cb1f32c8d0ae5ba0c4b9815cf4d59a8b71744d926c0693a0","src/component.rs":"f605082d1e1ed2f5a277e9f7c1a0f36842fe5bc10d9ca816fbd301a380946617","src/component/encode.rs":"5985e3f412595e07372a0787ec3db6c3018840df2bcfc638ab8798fd74b22778","src/config.rs":"9e52b815c79e1b65af21b23c8148ca9330201bfff657626352d5055e5be77abb","src/core.rs":"a5e13de2fb7c97eb12f2c1bc439f617b8afe6c509868fcb92db5e2aa9f252e65","src/core/code_builder.rs":"2e9d1eb09711f3e4ec4334d6a3952d9b5aed827778c14a7e01fb9676640ac561","src/core/code_builder/no_traps.rs":"e595dbde06551f5f8b23e03cfac0634beacab08e6243c66d6ffda95079212b24","src/core/encode.rs":"76ac812803e16cebf852f6d1265a0bd9ef16661e87f77a15f3ba997e75b351ce","src/core/terminate.rs":"d24af5206a13aee7d6a6ea900ccdf088c09d053c36026cf1607cc38c972b3ba9","src/lib.rs":"3a5521a9d4f1f38d01206339b1997e2943d6129be7d434457d59d019e2937456","tests/available_imports.rs":"3157294aa496c9e571398bdbd907539350f1015368cf47e45ecf822473df1f4e","tests/component.rs":"44684e990e832590a0b348ce33fadf39ee55dec52fdfa9353570edcbf65325ee","tests/core.rs":"1f581629a235399127b525c3aa56f1fef5da54b7887824d7c40169642cb5adb1"},"package":"51803e2c31e82f93ec1796f781f9855844ff1203f993409eb74d1d17130928a7"}
\ No newline at end of file +{"files":{"Cargo.toml":"b911aa53dc1f5c0cb04a77e15b43c31cee98893c7bcb98b6d73405aeca6b447c","LICENSE":"268872b9816f90fd8e85db5a28d33f8150ebb8dd016653fb39ef1f94f2686bc5","README.md":"e7b2a96cf39834aa19aed7cad8658e33445cdec7c4957658ba39e3b82cc4254b","benches/corpus.rs":"2df29556be0799f0cb1f32c8d0ae5ba0c4b9815cf4d59a8b71744d926c0693a0","src/component.rs":"253cf8d8abf785e4b5a5c614561a85b032e7f7bdec366f83d9db0702f73d2122","src/component/encode.rs":"5985e3f412595e07372a0787ec3db6c3018840df2bcfc638ab8798fd74b22778","src/config.rs":"e75d93c0a6a5ba07ff3ec1164a9b58601fadc0a0899149e14a77e8f0ab4595e4","src/core.rs":"35d3854c8caa1f61494b2f8e2b5e150d67cf88582243feeeb198e2aa93247339","src/core/code_builder.rs":"a8b27f652c8ae4b964e05933df313016ab7ffc297ca76005faca3da2792e3afc","src/core/code_builder/no_traps.rs":"2f36c548a32a6a003e6bb22a8fe7c75517b9be00c5a299212836a7e2431a8693","src/core/encode.rs":"77e1395c78ecf1b688a3c515a58be6efe1aa58d450a20a50d69598c54f7162de","src/core/terminate.rs":"c7c5f5618d393e5d9e44ae5e9bbde4b81bfa1ca3d2111d1eea875e2490f0f64b","src/lib.rs":"d393ae9daba211ba68a48dbdee545e0570cb8298e4c62319b8c8a8b9355cd6e3","tests/available_imports.rs":"6d4c1b111f35fefce55b668474ac9ba25edf7fc3e3decb88696b644aec13d603","tests/common/mod.rs":"9d0712a0b3e30030934b04785fb8632576bdcc31921ee8f524e0b2c903dee28d","tests/component.rs":"44684e990e832590a0b348ce33fadf39ee55dec52fdfa9353570edcbf65325ee","tests/core.rs":"408a50da4438be4e5343b116aa80f5ce344ba9508fad413cc5c32cd0aa2164fb","tests/exports.rs":"dcac4e18f277cbdb821f4ad7eda0a900bd73924e0818cf1b408700c9b7bba789"},"package":"61ff53a54a853f174b0df74cdb1553f1451e7bcdc23b26b1379f664ee1913d1a"}
\ No newline at end of file diff --git a/third_party/rust/wasm-smith/Cargo.toml b/third_party/rust/wasm-smith/Cargo.toml index bebb2807d5..3847409e33 100644 --- a/third_party/rust/wasm-smith/Cargo.toml +++ b/third_party/rust/wasm-smith/Cargo.toml @@ -12,7 +12,7 @@ [package] edition = "2021" name = "wasm-smith" -version = "0.15.0" +version = "0.201.0" authors = ["Nick Fitzgerald <fitzgen@gmail.com>"] exclude = ["/benches/corpus"] description = "A WebAssembly test case generator" @@ -33,7 +33,6 @@ harness = false [dependencies.anyhow] version = "1.0.58" -optional = true [dependencies.arbitrary] version = "1.1.0" @@ -62,14 +61,14 @@ version = "1.0.166" optional = true [dependencies.wasm-encoder] -version = "0.40.0" +version = "0.201.0" [dependencies.wasmparser] -version = "0.120.0" +version = "0.201.0" optional = true [dependencies.wat] -version = "1.0.84" +version = "1.201.0" optional = true [dev-dependencies.criterion] @@ -80,11 +79,10 @@ version = "0.8.4" features = ["small_rng"] [dev-dependencies.wasmparser] -version = "0.120.0" +version = "0.201.0" [features] _internal_cli = [ - "anyhow", "clap", "flagset/serde", "serde", @@ -99,3 +97,9 @@ wasmparser = [ [target."cfg(not(target_family = \"wasm\"))".dev-dependencies.libfuzzer-sys] version = "0.4.0" + +[lints.clippy] +all = "allow" + +[lints.rust] +unsafe_code = "deny" diff --git a/third_party/rust/wasm-smith/src/component.rs b/third_party/rust/wasm-smith/src/component.rs index e18373b4f4..7a85b2bf10 100644 --- a/third_party/rust/wasm-smith/src/component.rs +++ b/third_party/rust/wasm-smith/src/component.rs @@ -7,12 +7,13 @@ use crate::{arbitrary_loop, Config}; use arbitrary::{Arbitrary, Result, Unstructured}; use std::collections::BTreeMap; -use std::convert::TryFrom; use std::{ collections::{HashMap, HashSet}, rc::Rc, }; -use wasm_encoder::{ComponentTypeRef, ComponentValType, PrimitiveValType, TypeBounds, ValType}; +use wasm_encoder::{ + ComponentTypeRef, ComponentValType, HeapType, PrimitiveValType, RefType, TypeBounds, ValType, +}; mod encode; @@ -540,7 +541,7 @@ impl ComponentBuilder { } let ty = match u.int_in_range::<u8>(0..=1)? { - 0 => CoreType::Func(crate::core::arbitrary_func_type( + 0 => CoreType::Func(arbitrary_func_type( u, &self.config, &self.core_valtypes, @@ -819,7 +820,7 @@ impl ComponentBuilder { // Type definition. 2 => { - let ty = crate::core::arbitrary_func_type( + let ty = arbitrary_func_type( u, &self.config, &self.core_valtypes, @@ -955,7 +956,7 @@ impl ComponentBuilder { } fn arbitrary_core_table_type(&self, u: &mut Unstructured) -> Result<crate::core::TableType> { - crate::core::arbitrary_table_type(u, &self.config) + crate::core::arbitrary_table_type(u, &self.config, None) } fn arbitrary_core_memory_type(&self, u: &mut Unstructured) -> Result<crate::core::MemoryType> { @@ -2176,3 +2177,45 @@ struct CoreInstanceSection {} struct CoreTypeSection { types: Vec<Rc<CoreType>>, } + +fn arbitrary_func_type( + u: &mut Unstructured, + config: &Config, + valtypes: &[ValType], + max_results: Option<usize>, + type_ref_limit: u32, +) -> Result<Rc<crate::core::FuncType>> { + let mut params = vec![]; + let mut results = vec![]; + arbitrary_loop(u, 0, 20, |u| { + params.push(arbitrary_valtype(u, config, valtypes, type_ref_limit)?); + Ok(true) + })?; + arbitrary_loop(u, 0, max_results.unwrap_or(20), |u| { + results.push(arbitrary_valtype(u, config, valtypes, type_ref_limit)?); + Ok(true) + })?; + Ok(Rc::new(crate::core::FuncType { params, results })) +} + +fn arbitrary_valtype( + u: &mut Unstructured, + config: &Config, + valtypes: &[ValType], + type_ref_limit: u32, +) -> Result<ValType> { + if config.gc_enabled && type_ref_limit > 0 && u.ratio(1, 20)? { + Ok(ValType::Ref(RefType { + // TODO: For now, only create allow nullable reference + // types. Eventually we should support non-nullable reference types, + // but this means that we will also need to recognize when it is + // impossible to create an instance of the reference (eg `(ref + // nofunc)` has no instances, and self-referential types that + // contain a non-null self-reference are also impossible to create). + nullable: true, + heap_type: HeapType::Concrete(u.int_in_range(0..=type_ref_limit - 1)?), + })) + } else { + Ok(*u.choose(valtypes)?) + } +} diff --git a/third_party/rust/wasm-smith/src/config.rs b/third_party/rust/wasm-smith/src/config.rs index b183530e4c..43dbecceb8 100644 --- a/third_party/rust/wasm-smith/src/config.rs +++ b/third_party/rust/wasm-smith/src/config.rs @@ -20,13 +20,13 @@ macro_rules! define_config { /// Defaults to `None` which means that any arbitrary import can be /// generated. /// - /// To only allow specific imports, override this method to return a - /// WebAssembly module which describes the imports allowed. + /// To only allow specific imports, set this field to a WebAssembly + /// module which describes the imports allowed. /// /// Note that [`Self::min_imports`] is ignored when /// `available_imports` are enabled. /// - /// The returned value must be a valid binary encoding of a + /// The provided value must be a valid binary encoding of a /// WebAssembly module. `wasm-smith` will panic if the module cannot /// be parsed. /// @@ -49,6 +49,51 @@ macro_rules! define_config { /// ``` pub available_imports: Option<Vec<u8>>, + /// If provided, the generated module will have exports with exactly + /// the same names and types as those in the provided WebAssembly + /// module. The implementation (e.g. function bodies, global + /// initializers) of each export in the generated module will be + /// random and unrelated to the implementation in the provided + /// module. Only globals and functions are supported. + /// + /// + /// Defaults to `None` which means arbitrary exports will be + /// generated. + /// + /// To specify which exports the generated modules should have, set + /// this field to a WebAssembly module which describes the desired + /// exports. To generate modules with varying exports that meet some + /// constraints, consider randomly generating the value for this + /// field. + /// + /// The provided value must be a valid binary encoding of a + /// WebAssembly module. `wasm-smith` will panic if the module cannot + /// be parsed. + /// + /// # Module Limits + /// + /// All types, functions, globals, and exports that are needed to + /// provide the required exports will be generated, even if it + /// causes the resulting module to exceed the limits defined in + /// [`Self::max_type_size`], [`Self::max_types`], + /// [`Self::max_funcs`], [`Self::max_globals`], or + /// [`Self::max_exports`]. + /// + /// # Example + /// + /// As for [`Self::available_imports`], the `wat` crate can be used + /// to provide an human-readable description of the desired exports: + /// + /// ```rust + /// Some(wat::parse_str(r#" + /// (module + /// (func (export "foo") (param i32) (result i64) unreachable) + /// (global (export "bar") f32 f32.const 0) + /// ) + /// "#)); + /// ``` + pub exports: Option<Vec<u8>>, + $( $(#[$field_attr])* pub $field: $field_ty, @@ -59,6 +104,7 @@ macro_rules! define_config { fn default() -> Config { Config { available_imports: None, + exports: None, $( $field: $default, @@ -82,12 +128,44 @@ macro_rules! define_config { /// Note that [`Self::min_imports`] is ignored when /// `available_imports` are enabled. /// - /// The returned value must be a valid binary encoding of a + /// The provided value must be a valid binary encoding of a /// WebAssembly module. `wasm-smith` will panic if the module cannot /// be parsed. #[cfg_attr(feature = "clap", clap(long))] available_imports: Option<std::path::PathBuf>, + /// If provided, the generated module will have exports with exactly + /// the same names and types as those in the provided WebAssembly + /// module. The implementation (e.g. function bodies, global + /// initializers) of each export in the generated module will be + /// random and unrelated to the implementation in the provided + /// module. Only globals and functions are supported. + /// + /// Defaults to `None` which means arbitrary exports will be + /// generated. + /// + /// To specify which exports the generated modules should have, set + /// this field to a WebAssembly module which describes the desired + /// exports. To generate modules with varying exports that meet some + /// constraints, consider randomly generating the value for this + /// field. + /// + /// The provided value must be a valid binary encoding of a + /// WebAssembly module. `wasm-smith` will panic if the module cannot + /// be parsed. + /// + /// # Module Limits + /// + /// All types, functions, globals, and exports that are needed to + /// provide the required exports will be generated, even if it + /// causes the resulting module to exceed the limits defined in + /// [`Self::max_type_size`], [`Self::max_types`], + /// [`Self::max_funcs`], [`Self::max_globals`], or + /// [`Self::max_exports`]. + /// + #[cfg_attr(feature = "clap", clap(long))] + exports: Option<std::path::PathBuf>, + $( $(#[$field_attr])* #[cfg_attr(feature = "clap", clap(long))] @@ -100,6 +178,7 @@ macro_rules! define_config { pub fn or(self, other: Self) -> Self { Self { available_imports: self.available_imports.or(other.available_imports), + exports: self.exports.or(other.exports), $( $field: self.$field.or(other.$field), @@ -121,6 +200,13 @@ macro_rules! define_config { } else { None }, + exports: if let Some(file) = config + .exports + .as_ref() { + Some(wat::parse_file(file)?) + } else { + None + }, $( $field: config.$field.unwrap_or(default.$field), @@ -481,6 +567,18 @@ define_config! { /// /// Defaults to `false`. pub threads_enabled: bool = false, + + /// Indicates whether wasm-smith is allowed to generate invalid function + /// bodies. + /// + /// When enabled this option will enable taking raw bytes from the input + /// byte stream and using them as a wasm function body. This means that + /// the output module is not guaranteed to be valid but can help tickle + /// various parts of validation/compilation in some circumstances as + /// well. + /// + /// Defaults to `false`. + pub allow_invalid_funcs: bool = false, } } @@ -611,12 +709,14 @@ impl<'a> Arbitrary<'a> for Config { max_type_size: 1000, canonicalize_nans: false, available_imports: None, + exports: None, threads_enabled: false, export_everything: false, disallow_traps: false, tail_call_enabled: false, gc_enabled: false, generate_custom_sections: false, + allow_invalid_funcs: false, }) } } diff --git a/third_party/rust/wasm-smith/src/core.rs b/third_party/rust/wasm-smith/src/core.rs index 6a2836ae0e..194d2101ae 100644 --- a/third_party/rust/wasm-smith/src/core.rs +++ b/third_party/rust/wasm-smith/src/core.rs @@ -9,7 +9,8 @@ use arbitrary::{Arbitrary, Result, Unstructured}; use code_builder::CodeBuilderAllocations; use flagset::{flags, FlagSet}; use std::collections::{HashMap, HashSet}; -use std::convert::TryFrom; +use std::fmt; +use std::mem; use std::ops::Range; use std::rc::Rc; use std::str::{self, FromStr}; @@ -42,7 +43,6 @@ type Instruction = wasm_encoder::Instruction<'static>; /// To configure the shape of generated module, create a /// [`Config`][crate::Config] and then call [`Module::new`][crate::Module::new] /// with it. -#[derive(Debug)] pub struct Module { config: Config, duplicate_imports_behavior: DuplicateImportsBehavior, @@ -95,9 +95,8 @@ pub struct Module { /// aliased). num_defined_funcs: usize, - /// The number of tables defined in this module (not imported or - /// aliased). - num_defined_tables: usize, + /// Initialization expressions for all defined tables in this module. + defined_tables: Vec<Option<ConstExpr>>, /// The number of memories defined in this module (not imported or /// aliased). @@ -105,7 +104,7 @@ pub struct Module { /// The indexes and initialization expressions of globals defined in this /// module. - defined_globals: Vec<(u32, GlobalInitExpr)>, + defined_globals: Vec<(u32, ConstExpr)>, /// All tags available to this module, sorted by their index. The list /// entry is the type of each tag. @@ -140,6 +139,13 @@ pub struct Module { /// Names currently exported from this module. export_names: HashSet<String>, + + /// Reusable buffer in `self.arbitrary_const_expr` to amortize the cost of + /// allocation. + const_expr_choices: Vec<Box<dyn Fn(&mut Unstructured, ValType) -> Result<ConstExpr>>>, + + /// What the maximum type index that can be referenced is. + max_type_limit: MaxTypeLimit, } impl<'a> Arbitrary<'a> for Module { @@ -148,12 +154,27 @@ impl<'a> Arbitrary<'a> for Module { } } +impl fmt::Debug for Module { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("Module") + .field("config", &self.config) + .field(&"...", &"...") + .finish() + } +} + #[derive(Debug, Clone, Copy, PartialEq, Eq)] pub(crate) enum DuplicateImportsBehavior { Allowed, Disallowed, } +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +enum MaxTypeLimit { + ModuleTypes, + Num(u32), +} + impl Module { /// Returns a reference to the internal configuration. pub fn config(&self) -> &Config { @@ -172,7 +193,7 @@ impl Module { duplicate_imports_behavior: DuplicateImportsBehavior, ) -> Result<Self> { let mut module = Module::empty(config, duplicate_imports_behavior); - module.build(u, false)?; + module.build(u)?; Ok(module) } @@ -194,7 +215,7 @@ impl Module { num_imports: 0, num_defined_tags: 0, num_defined_funcs: 0, - num_defined_tables: 0, + defined_tables: Vec::new(), num_defined_memories: 0, defined_globals: Vec::new(), tags: Vec::new(), @@ -209,34 +230,12 @@ impl Module { data: Vec::new(), type_size: 0, export_names: HashSet::new(), + const_expr_choices: Vec::new(), + max_type_limit: MaxTypeLimit::ModuleTypes, } } } -/// Same as [`Module`], but may be invalid. -/// -/// This module generates function bodies differnetly than `Module` to try to -/// better explore wasm decoders and such. -#[derive(Debug)] -pub struct MaybeInvalidModule { - module: Module, -} - -impl MaybeInvalidModule { - /// Encode this Wasm module into bytes. - pub fn to_bytes(&self) -> Vec<u8> { - self.module.to_bytes() - } -} - -impl<'a> Arbitrary<'a> for MaybeInvalidModule { - fn arbitrary(u: &mut Unstructured<'a>) -> Result<Self> { - let mut module = Module::empty(Config::default(), DuplicateImportsBehavior::Allowed); - module.build(u, true)?; - Ok(MaybeInvalidModule { module }) - } -} - #[derive(Clone, Debug, PartialEq, Eq, Hash)] pub(crate) struct RecGroup { pub(crate) types: Vec<SubType>, @@ -357,7 +356,7 @@ enum ElementKind { #[derive(Debug)] enum Elements { Functions(Vec<u32>), - Expressions(Vec<Option<u32>>), + Expressions(Vec<ConstExpr>), } #[derive(Debug)] @@ -391,14 +390,8 @@ pub(crate) enum Offset { Global(u32), } -#[derive(Debug)] -pub(crate) enum GlobalInitExpr { - FuncRef(u32), - ConstExpr(ConstExpr), -} - impl Module { - fn build(&mut self, u: &mut Unstructured, allow_invalid: bool) -> Result<()> { + fn build(&mut self, u: &mut Unstructured) -> Result<()> { self.valtypes = configured_valtypes(&self.config); // We attempt to figure out our available imports *before* creating the types section here, @@ -413,7 +406,6 @@ impl Module { self.arbitrary_imports(u)?; } - self.should_encode_types = !self.types.is_empty() || u.arbitrary()?; self.should_encode_imports = !self.imports.is_empty() || u.arbitrary()?; self.arbitrary_tags(u)?; @@ -421,11 +413,14 @@ impl Module { self.arbitrary_tables(u)?; self.arbitrary_memories(u)?; self.arbitrary_globals(u)?; - self.arbitrary_exports(u)?; + if !self.required_exports(u)? { + self.arbitrary_exports(u)?; + }; + self.should_encode_types = !self.types.is_empty() || u.arbitrary()?; self.arbitrary_start(u)?; self.arbitrary_elems(u)?; self.arbitrary_data(u)?; - self.arbitrary_code(u, allow_invalid)?; + self.arbitrary_code(u)?; Ok(()) } @@ -562,6 +557,8 @@ impl Module { fn arbitrary_rec_group(&mut self, u: &mut Unstructured) -> Result<()> { let rec_group_start = self.types.len(); + assert!(matches!(self.max_type_limit, MaxTypeLimit::ModuleTypes)); + if self.config.gc_enabled { // With small probability, clone an existing rec group. if self.clonable_rec_groups().next().is_some() && u.ratio(1, u8::MAX)? { @@ -572,16 +569,20 @@ impl Module { let max_rec_group_size = self.config.max_types - self.types.len(); let rec_group_size = u.int_in_range(0..=max_rec_group_size)?; let type_ref_limit = u32::try_from(self.types.len() + rec_group_size).unwrap(); + self.max_type_limit = MaxTypeLimit::Num(type_ref_limit); for _ in 0..rec_group_size { - let ty = self.arbitrary_sub_type(u, type_ref_limit)?; + let ty = self.arbitrary_sub_type(u)?; self.add_type(ty); } } else { let type_ref_limit = u32::try_from(self.types.len()).unwrap(); - let ty = self.arbitrary_sub_type(u, type_ref_limit)?; + self.max_type_limit = MaxTypeLimit::Num(type_ref_limit); + let ty = self.arbitrary_sub_type(u)?; self.add_type(ty); } + self.max_type_limit = MaxTypeLimit::ModuleTypes; + self.rec_groups.push(rec_group_start..self.types.len()); Ok(()) } @@ -615,40 +616,27 @@ impl Module { Ok(()) } - fn arbitrary_sub_type( - &mut self, - u: &mut Unstructured, - // NB: Types can be referenced up to (but not including) - // `type_ref_limit`. It is an exclusive bound to avoid an option: to - // disallow any reference to a concrete type (e.g. `(ref $type)`) you - // can simply pass `0` here. - type_ref_limit: u32, - ) -> Result<SubType> { + fn arbitrary_sub_type(&mut self, u: &mut Unstructured) -> Result<SubType> { if !self.config.gc_enabled { - debug_assert_eq!(type_ref_limit, u32::try_from(self.types.len()).unwrap()); return Ok(SubType { is_final: true, supertype: None, - composite_type: CompositeType::Func(self.arbitrary_func_type(u, type_ref_limit)?), + composite_type: CompositeType::Func(self.arbitrary_func_type(u)?), }); } if !self.can_subtype.is_empty() && u.ratio(1, 32_u8)? { - self.arbitrary_sub_type_of_super_type(u, type_ref_limit) + self.arbitrary_sub_type_of_super_type(u) } else { Ok(SubType { is_final: u.arbitrary()?, supertype: None, - composite_type: self.arbitrary_composite_type(u, type_ref_limit)?, + composite_type: self.arbitrary_composite_type(u)?, }) } } - fn arbitrary_sub_type_of_super_type( - &mut self, - u: &mut Unstructured, - type_ref_limit: u32, - ) -> Result<SubType> { + fn arbitrary_sub_type_of_super_type(&mut self, u: &mut Unstructured) -> Result<SubType> { let supertype = *u.choose(&self.can_subtype)?; let mut composite_type = self.types[usize::try_from(supertype).unwrap()] .composite_type @@ -661,7 +649,7 @@ impl Module { *f = self.arbitrary_matching_func_type(u, f)?; } CompositeType::Struct(s) => { - *s = self.arbitrary_matching_struct_type(u, s, type_ref_limit)?; + *s = self.arbitrary_matching_struct_type(u, s)?; } } Ok(SubType { @@ -675,7 +663,6 @@ impl Module { &mut self, u: &mut Unstructured, ty: &StructType, - type_ref_limit: u32, ) -> Result<StructType> { let len_extra_fields = u.int_in_range(0..=5)?; let mut fields = Vec::with_capacity(ty.fields.len() + len_extra_fields); @@ -683,7 +670,7 @@ impl Module { fields.push(self.arbitrary_matching_field_type(u, *field)?); } for _ in 0..len_extra_fields { - fields.push(self.arbitrary_field_type(u, type_ref_limit)?); + fields.push(self.arbitrary_field_type(u)?); } Ok(StructType { fields: fields.into_boxed_slice(), @@ -915,71 +902,52 @@ impl Module { Ok(*u.choose(&choices)?) } - fn arbitrary_composite_type( - &mut self, - u: &mut Unstructured, - type_ref_limit: u32, - ) -> Result<CompositeType> { + fn arbitrary_composite_type(&mut self, u: &mut Unstructured) -> Result<CompositeType> { if !self.config.gc_enabled { - return Ok(CompositeType::Func( - self.arbitrary_func_type(u, type_ref_limit)?, - )); + return Ok(CompositeType::Func(self.arbitrary_func_type(u)?)); } match u.int_in_range(0..=2)? { 0 => Ok(CompositeType::Array(ArrayType( - self.arbitrary_field_type(u, type_ref_limit)?, + self.arbitrary_field_type(u)?, ))), - 1 => Ok(CompositeType::Func( - self.arbitrary_func_type(u, type_ref_limit)?, - )), - 2 => Ok(CompositeType::Struct( - self.arbitrary_struct_type(u, type_ref_limit)?, - )), + 1 => Ok(CompositeType::Func(self.arbitrary_func_type(u)?)), + 2 => Ok(CompositeType::Struct(self.arbitrary_struct_type(u)?)), _ => unreachable!(), } } - fn arbitrary_struct_type( - &mut self, - u: &mut Unstructured, - type_ref_limit: u32, - ) -> Result<StructType> { + fn arbitrary_struct_type(&mut self, u: &mut Unstructured) -> Result<StructType> { let len = u.int_in_range(0..=20)?; let mut fields = Vec::with_capacity(len); for _ in 0..len { - fields.push(self.arbitrary_field_type(u, type_ref_limit)?); + fields.push(self.arbitrary_field_type(u)?); } Ok(StructType { fields: fields.into_boxed_slice(), }) } - fn arbitrary_field_type( - &mut self, - u: &mut Unstructured, - type_ref_limit: u32, - ) -> Result<FieldType> { + fn arbitrary_field_type(&mut self, u: &mut Unstructured) -> Result<FieldType> { Ok(FieldType { - element_type: self.arbitrary_storage_type(u, type_ref_limit)?, + element_type: self.arbitrary_storage_type(u)?, mutable: u.arbitrary()?, }) } - fn arbitrary_storage_type( - &mut self, - u: &mut Unstructured, - type_ref_limit: u32, - ) -> Result<StorageType> { + fn arbitrary_storage_type(&mut self, u: &mut Unstructured) -> Result<StorageType> { match u.int_in_range(0..=2)? { 0 => Ok(StorageType::I8), 1 => Ok(StorageType::I16), - 2 => Ok(StorageType::Val(self.arbitrary_valtype(u, type_ref_limit)?)), + 2 => Ok(StorageType::Val(self.arbitrary_valtype(u)?)), _ => unreachable!(), } } fn arbitrary_ref_type(&self, u: &mut Unstructured) -> Result<RefType> { + if !self.config.reference_types_enabled { + return Ok(RefType::FUNCREF); + } Ok(RefType { nullable: true, heap_type: self.arbitrary_heap_type(u)?, @@ -989,9 +957,13 @@ impl Module { fn arbitrary_heap_type(&self, u: &mut Unstructured) -> Result<HeapType> { assert!(self.config.reference_types_enabled); - if self.config.gc_enabled && !self.types.is_empty() && u.arbitrary()? { - let type_ref_limit = u32::try_from(self.types.len()).unwrap(); - let idx = u.int_in_range(0..=type_ref_limit)?; + let concrete_type_limit = match self.max_type_limit { + MaxTypeLimit::Num(n) => n, + MaxTypeLimit::ModuleTypes => u32::try_from(self.types.len()).unwrap(), + }; + + if self.config.gc_enabled && concrete_type_limit > 0 && u.arbitrary()? { + let idx = u.int_in_range(0..=concrete_type_limit - 1)?; return Ok(HeapType::Concrete(idx)); } @@ -1018,22 +990,24 @@ impl Module { u.choose(&choices).copied() } - fn arbitrary_func_type( - &mut self, - u: &mut Unstructured, - type_ref_limit: u32, - ) -> Result<Rc<FuncType>> { - arbitrary_func_type( - u, - &self.config, - &self.valtypes, - if !self.config.multi_value_enabled { - Some(1) - } else { - None - }, - type_ref_limit, - ) + fn arbitrary_func_type(&mut self, u: &mut Unstructured) -> Result<Rc<FuncType>> { + let mut params = vec![]; + let mut results = vec![]; + let max_params = 20; + arbitrary_loop(u, 0, max_params, |u| { + params.push(self.arbitrary_valtype(u)?); + Ok(true) + })?; + let max_results = if self.config.multi_value_enabled { + max_params + } else { + 1 + }; + arbitrary_loop(u, 0, max_results, |u| { + results.push(self.arbitrary_valtype(u)?); + Ok(true) + })?; + Ok(Rc::new(FuncType { params, results })) } fn can_add_local_or_import_tag(&self) -> bool { @@ -1097,7 +1071,7 @@ impl Module { } if self.can_add_local_or_import_table() { choices.push(|u, m| { - let ty = arbitrary_table_type(u, m.config())?; + let ty = arbitrary_table_type(u, m.config(), Some(m))?; Ok(EntityType::Table(ty)) }); } @@ -1411,13 +1385,42 @@ impl Module { .filter(move |i| self.func_type(*i).results.is_empty()) } - fn arbitrary_valtype(&self, u: &mut Unstructured, type_ref_limit: u32) -> Result<ValType> { - arbitrary_valtype(u, &self.config, &self.valtypes, type_ref_limit) + fn arbitrary_valtype(&self, u: &mut Unstructured) -> Result<ValType> { + #[derive(Arbitrary)] + enum ValTypeClass { + I32, + I64, + F32, + F64, + V128, + Ref, + } + + match u.arbitrary::<ValTypeClass>()? { + ValTypeClass::I32 => Ok(ValType::I32), + ValTypeClass::I64 => Ok(ValType::I64), + ValTypeClass::F32 => Ok(ValType::F32), + ValTypeClass::F64 => Ok(ValType::F64), + ValTypeClass::V128 => { + if self.config.simd_enabled { + Ok(ValType::V128) + } else { + Ok(ValType::I32) + } + } + ValTypeClass::Ref => { + if self.config.reference_types_enabled { + Ok(ValType::Ref(self.arbitrary_ref_type(u)?)) + } else { + Ok(ValType::I32) + } + } + } } fn arbitrary_global_type(&self, u: &mut Unstructured) -> Result<GlobalType> { Ok(GlobalType { - val_type: self.arbitrary_valtype(u, u32::try_from(self.types.len()).unwrap())?, + val_type: self.arbitrary_valtype(u)?, mutable: u.arbitrary()?, }) } @@ -1470,14 +1473,38 @@ impl Module { if !self.can_add_local_or_import_table() { return Ok(false); } - self.num_defined_tables += 1; - let ty = arbitrary_table_type(u, self.config())?; + let ty = arbitrary_table_type(u, self.config(), Some(self))?; + let init = self.arbitrary_table_init(u, ty.element_type)?; + self.defined_tables.push(init); self.tables.push(ty); Ok(true) }, ) } + /// Generates an arbitrary table initialization expression for a table whose + /// element type is `ty`. + /// + /// Table initialization expressions were added by the GC proposal to + /// initialize non-nullable tables. + fn arbitrary_table_init( + &mut self, + u: &mut Unstructured, + ty: RefType, + ) -> Result<Option<ConstExpr>> { + if !self.config.gc_enabled { + assert!(ty.nullable); + return Ok(None); + } + // Even with the GC proposal an initialization expression is not + // required if the element type is nullable. + if ty.nullable && u.arbitrary()? { + return Ok(None); + } + let expr = self.arbitrary_const_expr(ValType::Ref(ty), u)?; + Ok(Some(expr)) + } + fn arbitrary_memories(&mut self, u: &mut Unstructured) -> Result<()> { arbitrary_loop( u, @@ -1494,53 +1521,260 @@ impl Module { ) } - fn arbitrary_globals(&mut self, u: &mut Unstructured) -> Result<()> { - let mut choices: Vec<Box<dyn Fn(&mut Unstructured, ValType) -> Result<GlobalInitExpr>>> = - vec![]; - let num_imported_globals = self.globals.len(); + /// Add a new global of the given type and return its global index. + fn add_arbitrary_global_of_type( + &mut self, + ty: GlobalType, + u: &mut Unstructured, + ) -> Result<u32> { + let expr = self.arbitrary_const_expr(ty.val_type, u)?; + let global_idx = self.globals.len() as u32; + self.globals.push(ty); + self.defined_globals.push((global_idx, expr)); + Ok(global_idx) + } + + /// Generates an arbitrary constant expression of the type `ty`. + fn arbitrary_const_expr(&mut self, ty: ValType, u: &mut Unstructured) -> Result<ConstExpr> { + let mut choices = mem::take(&mut self.const_expr_choices); + choices.clear(); + let num_funcs = self.funcs.len() as u32; + + // MVP wasm can `global.get` any immutable imported global in a + // constant expression, and the GC proposal enables this for all + // globals, so make all matching globals a candidate. + for i in self.globals_for_const_expr(ty) { + choices.push(Box::new(move |_, _| Ok(ConstExpr::global_get(i)))); + } + + // Another option for all types is to have an actual value of each type. + // Change `ty` to any valid subtype of `ty` and then generate a matching + // type of that value. + let ty = self.arbitrary_matching_val_type(u, ty)?; + match ty { + ValType::I32 => choices.push(Box::new(|u, _| Ok(ConstExpr::i32_const(u.arbitrary()?)))), + ValType::I64 => choices.push(Box::new(|u, _| Ok(ConstExpr::i64_const(u.arbitrary()?)))), + ValType::F32 => choices.push(Box::new(|u, _| Ok(ConstExpr::f32_const(u.arbitrary()?)))), + ValType::F64 => choices.push(Box::new(|u, _| Ok(ConstExpr::f64_const(u.arbitrary()?)))), + ValType::V128 => { + choices.push(Box::new(|u, _| Ok(ConstExpr::v128_const(u.arbitrary()?)))) + } + ValType::Ref(ty) => { + if ty.nullable { + choices.push(Box::new(move |_, _| Ok(ConstExpr::ref_null(ty.heap_type)))); + } + + match ty.heap_type { + HeapType::Func if num_funcs > 0 => { + choices.push(Box::new(move |u, _| { + let func = u.int_in_range(0..=num_funcs - 1)?; + Ok(ConstExpr::ref_func(func)) + })); + } + + HeapType::Concrete(ty) => { + for (i, fty) in self.funcs.iter().map(|(t, _)| *t).enumerate() { + if ty != fty { + continue; + } + choices.push(Box::new(move |_, _| Ok(ConstExpr::ref_func(i as u32)))); + } + } + + // TODO: fill out more GC types e.g `array.new` and + // `struct.new` + _ => {} + } + } + } + + let f = u.choose(&choices)?; + let ret = f(u, ty); + self.const_expr_choices = choices; + ret + } + + fn arbitrary_globals(&mut self, u: &mut Unstructured) -> Result<()> { arbitrary_loop(u, self.config.min_globals, self.config.max_globals, |u| { if !self.can_add_local_or_import_global() { return Ok(false); } let ty = self.arbitrary_global_type(u)?; + self.add_arbitrary_global_of_type(ty, u)?; - choices.clear(); - let num_funcs = self.funcs.len() as u32; - choices.push(Box::new(move |u, ty| { - Ok(GlobalInitExpr::ConstExpr(match ty { - ValType::I32 => ConstExpr::i32_const(u.arbitrary()?), - ValType::I64 => ConstExpr::i64_const(u.arbitrary()?), - ValType::F32 => ConstExpr::f32_const(u.arbitrary()?), - ValType::F64 => ConstExpr::f64_const(u.arbitrary()?), - ValType::V128 => ConstExpr::v128_const(u.arbitrary()?), - ValType::Ref(ty) => { - assert!(ty.nullable); - if ty.heap_type == HeapType::Func && num_funcs > 0 && u.arbitrary()? { - let func = u.int_in_range(0..=num_funcs - 1)?; - return Ok(GlobalInitExpr::FuncRef(func)); + Ok(true) + }) + } + + fn required_exports(&mut self, u: &mut Unstructured) -> Result<bool> { + let example_module = if let Some(wasm) = self.config.exports.clone() { + wasm + } else { + return Ok(false); + }; + + #[cfg(feature = "wasmparser")] + { + self._required_exports(u, &example_module)?; + Ok(true) + } + #[cfg(not(feature = "wasmparser"))] + { + let _ = (example_module, u); + panic!("support for `exports` was disabled at compile time"); + } + } + + #[cfg(feature = "wasmparser")] + fn _required_exports(&mut self, u: &mut Unstructured, example_module: &[u8]) -> Result<()> { + fn convert_heap_type(ty: &wasmparser::HeapType) -> HeapType { + match ty { + wasmparser::HeapType::Concrete(_) => { + panic!("Unable to handle concrete types in exports") + } + wasmparser::HeapType::Func => HeapType::Func, + wasmparser::HeapType::Extern => HeapType::Extern, + wasmparser::HeapType::Any => HeapType::Any, + wasmparser::HeapType::None => HeapType::None, + wasmparser::HeapType::NoExtern => HeapType::NoExtern, + wasmparser::HeapType::NoFunc => HeapType::NoFunc, + wasmparser::HeapType::Eq => HeapType::Eq, + wasmparser::HeapType::Struct => HeapType::Struct, + wasmparser::HeapType::Array => HeapType::Array, + wasmparser::HeapType::I31 => HeapType::I31, + wasmparser::HeapType::Exn => HeapType::Exn, + } + } + + fn convert_val_type(ty: &wasmparser::ValType) -> ValType { + match ty { + wasmparser::ValType::I32 => ValType::I32, + wasmparser::ValType::I64 => ValType::I64, + wasmparser::ValType::F32 => ValType::F32, + wasmparser::ValType::F64 => ValType::F64, + wasmparser::ValType::V128 => ValType::V128, + wasmparser::ValType::Ref(r) => ValType::Ref(RefType { + nullable: r.is_nullable(), + heap_type: convert_heap_type(&r.heap_type()), + }), + } + } + + fn convert_export_kind(kind: &wasmparser::ExternalKind) -> ExportKind { + match kind { + wasmparser::ExternalKind::Func => ExportKind::Func, + wasmparser::ExternalKind::Table => ExportKind::Table, + wasmparser::ExternalKind::Memory => ExportKind::Memory, + wasmparser::ExternalKind::Global => ExportKind::Global, + wasmparser::ExternalKind::Tag => ExportKind::Tag, + } + } + + let mut required_exports: Vec<wasmparser::Export> = vec![]; + let mut validator = wasmparser::Validator::new(); + let exports_types = validator + .validate_all(&example_module) + .expect("Failed to validate `exports` Wasm"); + for payload in wasmparser::Parser::new(0).parse_all(&example_module) { + match payload.expect("Failed to read `exports` Wasm") { + wasmparser::Payload::ExportSection(export_reader) => { + required_exports = export_reader + .into_iter() + .collect::<Result<_, _>>() + .expect("Failed to read `exports` export section"); + } + _ => {} + } + } + + // For each export, add necessary prerequisites to the module. + for export in required_exports { + let new_index = match exports_types + .entity_type_from_export(&export) + .unwrap_or_else(|| { + panic!( + "Unable to get type from export {:?} in `exports` Wasm", + export, + ) + }) { + // For functions, add the type and a function with that type. + wasmparser::types::EntityType::Func(id) => { + let subtype = exports_types.get(id).unwrap_or_else(|| { + panic!( + "Unable to get subtype for function {:?} in `exports` Wasm", + id + ) + }); + match &subtype.composite_type { + wasmparser::CompositeType::Func(func_type) => { + assert!( + subtype.is_final, + "Subtype {:?} from `exports` Wasm is not final", + subtype + ); + assert!( + subtype.supertype_idx.is_none(), + "Subtype {:?} from `exports` Wasm has non-empty supertype", + subtype + ); + let new_type = Rc::new(FuncType { + params: func_type + .params() + .into_iter() + .map(convert_val_type) + .collect(), + results: func_type + .results() + .into_iter() + .map(convert_val_type) + .collect(), + }); + self.rec_groups.push(self.types.len()..self.types.len() + 1); + let type_index = self.add_type(SubType { + is_final: true, + supertype: None, + composite_type: CompositeType::Func(Rc::clone(&new_type)), + }); + let func_index = self.funcs.len() as u32; + self.funcs.push((type_index, new_type)); + self.num_defined_funcs += 1; + func_index } - ConstExpr::ref_null(ty.heap_type) + _ => panic!( + "Unable to handle type {:?} from `exports` Wasm", + subtype.composite_type + ), } - })) - })); - - for (i, g) in self.globals[..num_imported_globals].iter().enumerate() { - if !g.mutable && g.val_type == ty.val_type { - choices.push(Box::new(move |_, _| { - Ok(GlobalInitExpr::ConstExpr(ConstExpr::global_get(i as u32))) - })); } - } + // For globals, add a new global. + wasmparser::types::EntityType::Global(global_type) => self + .add_arbitrary_global_of_type( + GlobalType { + val_type: convert_val_type(&global_type.content_type), + mutable: global_type.mutable, + }, + u, + )?, + wasmparser::types::EntityType::Table(_) + | wasmparser::types::EntityType::Memory(_) + | wasmparser::types::EntityType::Tag(_) => { + panic!( + "Config `exports` has an export of type {:?} which cannot yet be handled.", + export.kind + ) + } + }; + self.exports.push(( + export.name.to_string(), + convert_export_kind(&export.kind), + new_index, + )); + self.export_names.insert(export.name.to_string()); + } - let f = u.choose(&choices)?; - let expr = f(u, ty.val_type)?; - let global_idx = self.globals.len() as u32; - self.globals.push(ty); - self.defined_globals.push((global_idx, expr)); - Ok(true) - }) + Ok(()) } fn arbitrary_exports(&mut self, u: &mut Unstructured) -> Result<()> { @@ -1644,24 +1878,17 @@ impl Module { } fn arbitrary_elems(&mut self, u: &mut Unstructured) -> Result<()> { - let func_max = self.funcs.len() as u32; - // Create a helper closure to choose an arbitrary offset. let mut offset_global_choices = vec![]; if !self.config.disallow_traps { - for (i, g) in self.globals[..self.globals.len() - self.defined_globals.len()] - .iter() - .enumerate() - { - if !g.mutable && g.val_type == ValType::I32 { - offset_global_choices.push(i as u32); - } + for i in self.globals_for_const_expr(ValType::I32) { + offset_global_choices.push(i); } } + let disallow_traps = self.config.disallow_traps; let arbitrary_active_elem = |u: &mut Unstructured, min_mem_size: u32, table: Option<u32>, - disallow_traps: bool, table_ty: &TableType| { let (offset, max_size_hint) = if !offset_global_choices.is_empty() && u.arbitrary()? { let g = u.choose(&offset_global_choices)?; @@ -1686,11 +1913,20 @@ impl Module { Ok((ElementKind::Active { table, offset }, max_size_hint)) }; + // Generate a list of candidates for "kinds" of elements segments. For + // example we can have an active segment for any existing table or + // passive/declared segments if the right wasm features are enabled. type GenElemSegment<'a> = dyn Fn(&mut Unstructured) -> Result<(ElementKind, Option<u32>)> + 'a; - let mut funcrefs: Vec<Box<GenElemSegment>> = Vec::new(); - let mut externrefs: Vec<Box<GenElemSegment>> = Vec::new(); - let disallow_traps = self.config.disallow_traps; + let mut choices: Vec<Box<GenElemSegment>> = Vec::new(); + + // Bulk memory enables passive/declared segments, and note that the + // types used are selected later. + if self.config.bulk_memory_enabled { + choices.push(Box::new(|_| Ok((ElementKind::Passive, None)))); + choices.push(Box::new(|_| Ok((ElementKind::Declared, None)))); + } + for (i, ty) in self.tables.iter().enumerate() { // If this table starts with no capacity then any non-empty element // segment placed onto it will immediately trap, which isn't too @@ -1700,93 +1936,100 @@ impl Module { continue; } - let dst = if ty.element_type == RefType::FUNCREF { - &mut funcrefs - } else { - &mut externrefs - }; let minimum = ty.minimum; // If the first table is a funcref table then it's a candidate for // the MVP encoding of element segments. + let ty = *ty; if i == 0 && ty.element_type == RefType::FUNCREF { - dst.push(Box::new(move |u| { - arbitrary_active_elem(u, minimum, None, disallow_traps, ty) + choices.push(Box::new(move |u| { + arbitrary_active_elem(u, minimum, None, &ty) })); } if self.config.bulk_memory_enabled { let idx = Some(i as u32); - dst.push(Box::new(move |u| { - arbitrary_active_elem(u, minimum, idx, disallow_traps, ty) + choices.push(Box::new(move |u| { + arbitrary_active_elem(u, minimum, idx, &ty) })); } } - // Bulk memory enables passive/declared segments for funcrefs, and - // reference types additionally enables the segments for externrefs. - if self.config.bulk_memory_enabled { - funcrefs.push(Box::new(|_| Ok((ElementKind::Passive, None)))); - funcrefs.push(Box::new(|_| Ok((ElementKind::Declared, None)))); - if self.config.reference_types_enabled { - externrefs.push(Box::new(|_| Ok((ElementKind::Passive, None)))); - externrefs.push(Box::new(|_| Ok((ElementKind::Declared, None)))); - } - } - - let mut choices = Vec::new(); - if !funcrefs.is_empty() { - choices.push((&funcrefs, RefType::FUNCREF)); - } - if !externrefs.is_empty() { - choices.push((&externrefs, RefType::EXTERNREF)); - } - if choices.is_empty() { return Ok(()); } + arbitrary_loop( u, self.config.min_element_segments, self.config.max_element_segments, |u| { - // Choose whether to generate a segment whose elements are initialized via - // expressions, or one whose elements are initialized via function indices. - let (kind_candidates, ty) = *u.choose(&choices)?; - - // Select a kind for this segment now that we know the number of - // items the segment will hold. - let (kind, max_size_hint) = u.choose(kind_candidates)?(u)?; + // Pick a kind of element segment to generate which will also + // give us a hint of the maximum size, if any. + let (kind, max_size_hint) = u.choose(&choices)?(u)?; let max = max_size_hint .map(|i| usize::try_from(i).unwrap()) .unwrap_or_else(|| self.config.max_elements); - // Pick whether we're going to use expression elements or - // indices. Note that externrefs must use expressions, - // and functions without reference types must use indices. - let items = if ty == RefType::EXTERNREF - || (self.config.reference_types_enabled && u.arbitrary()?) + // Infer, from the kind of segment, the type of the element + // segment. Passive/declared segments can be declared with any + // reference type, but active segments must match their table. + let ty = match kind { + ElementKind::Passive | ElementKind::Declared => self.arbitrary_ref_type(u)?, + ElementKind::Active { table, .. } => { + let idx = table.unwrap_or(0); + self.arbitrary_matching_ref_type(u, self.tables[idx as usize].element_type)? + } + }; + + // The `Elements::Functions` encoding is only possible when the + // element type is a `funcref` because the binary format can't + // allow encoding any other type in that form. + let can_use_function_list = ty == RefType::FUNCREF; + if !self.config.reference_types_enabled { + assert!(can_use_function_list); + } + + // If a function list is possible then build up a list of + // functions that can be selected from. + let mut func_candidates = Vec::new(); + if can_use_function_list { + match ty.heap_type { + HeapType::Func => { + func_candidates.extend(0..self.funcs.len() as u32); + } + HeapType::Concrete(ty) => { + for (i, (fty, _)) in self.funcs.iter().enumerate() { + if *fty == ty { + func_candidates.push(i as u32); + } + } + } + _ => {} + } + } + + // And finally actually generate the arbitrary elements of this + // element segment. Function indices are used if they're either + // forced or allowed, and otherwise expressions are used + // instead. + let items = if !self.config.reference_types_enabled + || (can_use_function_list && u.arbitrary()?) { let mut init = vec![]; - arbitrary_loop(u, self.config.min_elements, max, |u| { - init.push( - if ty == RefType::EXTERNREF || func_max == 0 || u.arbitrary()? { - None - } else { - Some(u.int_in_range(0..=func_max - 1)?) - }, - ); - Ok(true) - })?; - Elements::Expressions(init) - } else { - let mut init = vec![]; - if func_max > 0 { + if func_candidates.len() > 0 { arbitrary_loop(u, self.config.min_elements, max, |u| { - let func_idx = u.int_in_range(0..=func_max - 1)?; + let func_idx = *u.choose(&func_candidates)?; init.push(func_idx); Ok(true) })?; } Elements::Functions(init) + } else { + let mut init = vec![]; + arbitrary_loop(u, self.config.min_elements, max, |u| { + init.push(self.arbitrary_const_expr(ValType::Ref(ty), u)?); + Ok(true) + })?; + Elements::Expressions(init) }; self.elems.push(ElementSegment { kind, ty, items }); @@ -1795,11 +2038,11 @@ impl Module { ) } - fn arbitrary_code(&mut self, u: &mut Unstructured, allow_invalid: bool) -> Result<()> { + fn arbitrary_code(&mut self, u: &mut Unstructured) -> Result<()> { self.code.reserve(self.num_defined_funcs); - let mut allocs = CodeBuilderAllocations::new(self); + let mut allocs = CodeBuilderAllocations::new(self, self.config.exports.is_some()); for (_, ty) in self.funcs[self.funcs.len() - self.num_defined_funcs..].iter() { - let body = self.arbitrary_func_body(u, ty, &mut allocs, allow_invalid)?; + let body = self.arbitrary_func_body(u, ty, &mut allocs)?; self.code.push(body); } allocs.finish(u, self)?; @@ -1811,11 +2054,10 @@ impl Module { u: &mut Unstructured, ty: &FuncType, allocs: &mut CodeBuilderAllocations, - allow_invalid: bool, ) -> Result<Code> { let mut locals = self.arbitrary_locals(u)?; let builder = allocs.builder(ty, &mut locals); - let instructions = if allow_invalid && u.arbitrary().unwrap_or(false) { + let instructions = if self.config.allow_invalid_funcs && u.arbitrary().unwrap_or(false) { Instructions::Arbitrary(arbitrary_vec_u8(u)?) } else { Instructions::Generated(builder.arbitrary(u, self)?) @@ -1830,7 +2072,7 @@ impl Module { fn arbitrary_locals(&self, u: &mut Unstructured) -> Result<Vec<ValType>> { let mut ret = Vec::new(); arbitrary_loop(u, 0, 100, |u| { - ret.push(self.arbitrary_valtype(u, u32::try_from(self.types.len()).unwrap())?); + ret.push(self.arbitrary_valtype(u)?); Ok(true) })?; Ok(ret) @@ -1865,18 +2107,11 @@ impl Module { )) })); if !self.config.disallow_traps { - for (i, g) in self.globals[..self.globals.len() - self.defined_globals.len()] - .iter() - .enumerate() - { - if g.mutable { - continue; - } - if g.val_type == ValType::I32 { - choices32.push(Box::new(move |_, _, _| Ok(Offset::Global(i as u32)))); - } else if g.val_type == ValType::I64 { - choices64.push(Box::new(move |_, _, _| Ok(Offset::Global(i as u32)))); - } + for i in self.globals_for_const_expr(ValType::I32) { + choices32.push(Box::new(move |_, _, _| Ok(Offset::Global(i)))); + } + for i in self.globals_for_const_expr(ValType::I64) { + choices64.push(Box::new(move |_, _, _| Ok(Offset::Global(i)))); } } @@ -1962,6 +2197,33 @@ impl Module { } } } + + /// Returns an iterator of all globals which can be used in constant + /// expressions for a value of type `ty` specified. + fn globals_for_const_expr(&self, ty: ValType) -> impl Iterator<Item = u32> + '_ { + // Before the GC proposal only imported globals could be referenced, but + // the GC proposal relaxed this feature to allow any global. + let num_imported_globals = self.globals.len() - self.defined_globals.len(); + let max_global = if self.config.gc_enabled { + self.globals.len() + } else { + num_imported_globals + }; + + self.globals[..max_global] + .iter() + .enumerate() + .filter_map(move |(i, g)| { + // Mutable globals cannot participate in constant expressions, + // but otherwise so long as the global is a subtype of the + // desired type it's a candidate. + if !g.mutable && self.val_type_is_sub_type(g.val_type, ty) { + Some(i as u32) + } else { + None + } + }) + } } pub(crate) fn arbitrary_limits32( @@ -2044,49 +2306,11 @@ pub(crate) fn configured_valtypes(config: &Config) -> Vec<ValType> { valtypes } -pub(crate) fn arbitrary_func_type( - u: &mut Unstructured, - config: &Config, - valtypes: &[ValType], - max_results: Option<usize>, - type_ref_limit: u32, -) -> Result<Rc<FuncType>> { - let mut params = vec![]; - let mut results = vec![]; - arbitrary_loop(u, 0, 20, |u| { - params.push(arbitrary_valtype(u, config, valtypes, type_ref_limit)?); - Ok(true) - })?; - arbitrary_loop(u, 0, max_results.unwrap_or(20), |u| { - results.push(arbitrary_valtype(u, config, valtypes, type_ref_limit)?); - Ok(true) - })?; - Ok(Rc::new(FuncType { params, results })) -} - -fn arbitrary_valtype( +pub(crate) fn arbitrary_table_type( u: &mut Unstructured, config: &Config, - valtypes: &[ValType], - type_ref_limit: u32, -) -> Result<ValType> { - if config.gc_enabled && type_ref_limit > 0 && u.ratio(1, 20)? { - Ok(ValType::Ref(RefType { - // TODO: For now, only create allow nullable reference - // types. Eventually we should support non-nullable reference types, - // but this means that we will also need to recognize when it is - // impossible to create an instance of the reference (eg `(ref - // nofunc)` has no instances, and self-referential types that - // contain a non-null self-reference are also impossible to create). - nullable: true, - heap_type: HeapType::Concrete(u.int_in_range(0..=type_ref_limit - 1)?), - })) - } else { - Ok(*u.choose(valtypes)?) - } -} - -pub(crate) fn arbitrary_table_type(u: &mut Unstructured, config: &Config) -> Result<TableType> { + module: Option<&Module>, +) -> Result<TableType> { // We don't want to generate tables that are too large on average, so // keep the "inbounds" limit here a bit smaller. let max_inbounds = 10_000; @@ -2102,12 +2326,12 @@ pub(crate) fn arbitrary_table_type(u: &mut Unstructured, config: &Config) -> Res if config.disallow_traps { assert!(minimum > 0); } + let element_type = match module { + Some(module) => module.arbitrary_ref_type(u)?, + None => RefType::FUNCREF, + }; Ok(TableType { - element_type: if config.reference_types_enabled { - *u.choose(&[RefType::FUNCREF, RefType::EXTERNREF])? - } else { - RefType::FUNCREF - }, + element_type, minimum, maximum, }) diff --git a/third_party/rust/wasm-smith/src/core/code_builder.rs b/third_party/rust/wasm-smith/src/core/code_builder.rs index 1bc09008ad..a55c5aafda 100644 --- a/third_party/rust/wasm-smith/src/core/code_builder.rs +++ b/third_party/rust/wasm-smith/src/core/code_builder.rs @@ -1,11 +1,10 @@ use super::{ - CompositeType, Elements, FuncType, GlobalInitExpr, Instruction, InstructionKind::*, - InstructionKinds, Module, ValType, + CompositeType, Elements, FuncType, Instruction, InstructionKind::*, InstructionKinds, Module, + ValType, }; use crate::{unique_string, MemoryOffsetChoices}; use arbitrary::{Result, Unstructured}; use std::collections::{BTreeMap, BTreeSet}; -use std::convert::TryFrom; use std::rc::Rc; use wasm_encoder::{ ArrayType, BlockType, Catch, ConstExpr, ExportKind, FieldType, GlobalType, HeapType, MemArg, @@ -648,6 +647,10 @@ pub(crate) struct CodeBuilderAllocations { global_dropped_f32: Option<u32>, global_dropped_f64: Option<u32>, global_dropped_v128: Option<u32>, + + // Indicates that additional exports cannot be generated. This will be true + // if the `Config` specifies exactly which exports should be present. + disallow_exporting: bool, } pub(crate) struct CodeBuilder<'a> { @@ -704,7 +707,7 @@ enum Float { } impl CodeBuilderAllocations { - pub(crate) fn new(module: &Module) -> Self { + pub(crate) fn new(module: &Module, disallow_exporting: bool) -> Self { let mut mutable_globals = BTreeMap::new(); for (i, global) in module.globals.iter().enumerate() { if global.mutable { @@ -741,14 +744,14 @@ impl CodeBuilderAllocations { let mut referenced_functions = BTreeSet::new(); for (_, expr) in module.defined_globals.iter() { - if let GlobalInitExpr::FuncRef(i) = *expr { + if let Some(i) = expr.get_ref_func() { referenced_functions.insert(i); } } for g in module.elems.iter() { match &g.items { Elements::Expressions(e) => { - let iter = e.iter().filter_map(|i| *i); + let iter = e.iter().filter_map(|e| e.get_ref_func()); referenced_functions.extend(iter); } Elements::Functions(e) => { @@ -769,6 +772,42 @@ impl CodeBuilderAllocations { } } + let mut global_dropped_i32 = None; + let mut global_dropped_i64 = None; + let mut global_dropped_f32 = None; + let mut global_dropped_f64 = None; + let mut global_dropped_v128 = None; + + // If we can't export additional globals, try to use existing exported + // mutable globals for dropped values. + if disallow_exporting { + for (_, kind, index) in module.exports.iter() { + if *kind == ExportKind::Global { + let ty = module.globals[*index as usize]; + if ty.mutable { + match ty.val_type { + ValType::I32 => { + if global_dropped_i32.is_none() { + global_dropped_i32 = Some(*index) + } else { + global_dropped_f32 = Some(*index) + } + } + ValType::I64 => { + if global_dropped_i64.is_none() { + global_dropped_i64 = Some(*index) + } else { + global_dropped_f64 = Some(*index) + } + } + ValType::V128 => global_dropped_v128 = Some(*index), + _ => {} + } + } + } + } + } + CodeBuilderAllocations { controls: Vec::with_capacity(4), operands: Vec::with_capacity(16), @@ -782,13 +821,14 @@ impl CodeBuilderAllocations { memory32, memory64, - global_dropped_i32: None, - global_dropped_i64: None, - global_dropped_f32: None, - global_dropped_f64: None, - global_dropped_v128: None, + global_dropped_i32, + global_dropped_i64, + global_dropped_f32, + global_dropped_f64, + global_dropped_v128, globals_cnt: module.globals.len() as u32, new_globals: Vec::new(), + disallow_exporting, } } @@ -822,18 +862,18 @@ impl CodeBuilderAllocations { pub fn finish(self, u: &mut Unstructured<'_>, module: &mut Module) -> arbitrary::Result<()> { // Any globals injected as part of dropping operands on the stack get // injected into the module here. Each global is then exported, most of - // the time, to ensure it's part of the "image" of this module available - // for differential execution for example. + // the time (if additional exports are allowed), to ensure it's part of + // the "image" of this module available for differential execution for + // example. for (ty, init) in self.new_globals { let global_idx = module.globals.len() as u32; module.globals.push(GlobalType { val_type: ty, mutable: true, }); - let init = GlobalInitExpr::ConstExpr(init); module.defined_globals.push((global_idx, init)); - if u.ratio(1, 100).unwrap_or(false) { + if self.disallow_exporting || u.ratio(1, 100).unwrap_or(false) { continue; } @@ -922,6 +962,33 @@ impl CodeBuilder<'_> { self.allocs.operands.push(ty); } + fn pop_label_types(&mut self, module: &Module, target: u32) { + let target = usize::try_from(target).unwrap(); + let control = &self.allocs.controls[self.allocs.controls.len() - 1 - target]; + debug_assert!(self.label_types_on_stack(module, control)); + self.allocs + .operands + .truncate(self.allocs.operands.len() - control.label_types().len()); + } + + fn push_label_types(&mut self, target: u32) { + let target = usize::try_from(target).unwrap(); + let control = &self.allocs.controls[self.allocs.controls.len() - 1 - target]; + self.allocs + .operands + .extend(control.label_types().iter().copied().map(Some)); + } + + /// Pop the target label types, and then push them again. + /// + /// This is not a no-op due to subtyping: if we have a `T <: U` on the + /// stack, and the target label's type is `[U]`, then this will erase the + /// information about `T` and subsequent operations may only operate on `U`. + fn pop_push_label_types(&mut self, module: &Module, target: u32) { + self.pop_label_types(module, target); + self.push_label_types(target) + } + fn label_types_on_stack(&self, module: &Module, to_check: &Control) -> bool { self.types_on_stack(module, to_check.label_types()) } @@ -1087,12 +1154,7 @@ impl CodeBuilder<'_> { fn arbitrary_block_type(&self, u: &mut Unstructured, module: &Module) -> Result<BlockType> { let mut options: Vec<Box<dyn Fn(&mut Unstructured) -> Result<BlockType>>> = vec![ Box::new(|_| Ok(BlockType::Empty)), - Box::new(|u| { - Ok(BlockType::Result(module.arbitrary_valtype( - u, - u32::try_from(module.types.len()).unwrap(), - )?)) - }), + Box::new(|u| Ok(BlockType::Result(module.arbitrary_valtype(u)?))), ]; if module.config.multi_value_enabled { for (i, ty) in module.func_types() { @@ -1710,10 +1772,9 @@ fn br( .filter(|(_, l)| builder.label_types_on_stack(module, l)) .nth(i) .unwrap(); - let control = &builder.allocs.controls[builder.allocs.controls.len() - 1 - target]; - let tys = control.label_types().to_vec(); - builder.pop_operands(module, &tys); - instructions.push(Instruction::Br(target as u32)); + let target = u32::try_from(target).unwrap(); + builder.pop_label_types(module, target); + instructions.push(Instruction::Br(target)); Ok(()) } @@ -1757,7 +1818,9 @@ fn br_if( .filter(|(_, l)| builder.label_types_on_stack(module, l)) .nth(i) .unwrap(); - instructions.push(Instruction::BrIf(target as u32)); + let target = u32::try_from(target).unwrap(); + builder.pop_push_label_types(module, target); + instructions.push(Instruction::BrIf(target)); Ok(()) } @@ -2203,13 +2266,15 @@ fn br_on_null( .filter(|(_, l)| builder.label_types_on_stack(module, l)) .nth(i) .unwrap(); + let target = u32::try_from(target).unwrap(); + builder.pop_push_label_types(module, target); builder.push_operands(&[ValType::Ref(RefType { nullable: false, heap_type, })]); - instructions.push(Instruction::BrOnNull(u32::try_from(target).unwrap())); + instructions.push(Instruction::BrOnNull(target)); Ok(()) } @@ -2270,9 +2335,11 @@ fn br_on_non_null( .filter(|(_, l)| is_valid_br_on_non_null_control(module, l, builder)) .nth(i) .unwrap(); + let target = u32::try_from(target).unwrap(); + builder.pop_push_label_types(module, target); builder.pop_ref_type(); - instructions.push(Instruction::BrOnNonNull(u32::try_from(target).unwrap())); + instructions.push(Instruction::BrOnNonNull(target)); Ok(()) } @@ -2354,6 +2421,7 @@ fn br_on_cast( .unwrap(); let relative_depth = u32::try_from(relative_depth).unwrap(); + let num_label_types = control.label_types().len(); let to_ref_type = match control.label_types().last() { Some(ValType::Ref(r)) => *r, _ => unreachable!(), @@ -2363,6 +2431,15 @@ fn br_on_cast( let from_ref_type = from_ref_type.unwrap_or(to_ref_type); let from_ref_type = module.arbitrary_super_type_of_ref_type(u, from_ref_type)?; + // Do `pop_push_label_types` but without its debug assert that the types are + // on the stack, since we know that we have a `from_ref_type` but the label + // requires a `to_ref_type`. + for _ in 0..num_label_types { + builder.pop_operand(); + } + builder.push_label_types(relative_depth); + + // Replace the label's `to_ref_type` with the type difference. builder.pop_operand(); builder.push_operands(&[ValType::Ref(ref_type_difference( from_ref_type, @@ -2433,7 +2510,7 @@ fn br_on_cast_fail( debug_assert!(n > 0); let i = u.int_in_range(0..=n - 1)?; - let (target, control) = builder + let (relative_depth, control) = builder .allocs .controls .iter() @@ -2442,6 +2519,7 @@ fn br_on_cast_fail( .filter(|(_, l)| is_valid_br_on_cast_fail_control(module, builder, l, from_ref_type)) .nth(i) .unwrap(); + let relative_depth = u32::try_from(relative_depth).unwrap(); let from_ref_type = from_ref_type.unwrap_or_else(|| match control.label_types().last().unwrap() { @@ -2450,13 +2528,16 @@ fn br_on_cast_fail( }); let to_ref_type = module.arbitrary_matching_ref_type(u, from_ref_type)?; + // Pop-push the label types and then replace its last reference type with + // our `to_ref_type`. + builder.pop_push_label_types(module, relative_depth); builder.pop_operand(); builder.push_operand(Some(ValType::Ref(to_ref_type))); instructions.push(Instruction::BrOnCastFail { from_ref_type, to_ref_type, - relative_depth: u32::try_from(target).unwrap(), + relative_depth, }); Ok(()) } diff --git a/third_party/rust/wasm-smith/src/core/code_builder/no_traps.rs b/third_party/rust/wasm-smith/src/core/code_builder/no_traps.rs index a1232b6922..4f8b31ca9f 100644 --- a/third_party/rust/wasm-smith/src/core/code_builder/no_traps.rs +++ b/third_party/rust/wasm-smith/src/core/code_builder/no_traps.rs @@ -1,5 +1,5 @@ use crate::core::*; -use wasm_encoder::{BlockType, Instruction, ValType}; +use wasm_encoder::Instruction; use super::CodeBuilder; diff --git a/third_party/rust/wasm-smith/src/core/encode.rs b/third_party/rust/wasm-smith/src/core/encode.rs index 7c781dc731..5a661509a6 100644 --- a/third_party/rust/wasm-smith/src/core/encode.rs +++ b/third_party/rust/wasm-smith/src/core/encode.rs @@ -1,5 +1,4 @@ use super::*; -use std::convert::TryFrom; impl Module { /// Encode this Wasm module into bytes. @@ -122,12 +121,22 @@ impl Module { } fn encode_tables(&self, module: &mut wasm_encoder::Module) { - if self.num_defined_tables == 0 { + if self.defined_tables.is_empty() { return; } let mut tables = wasm_encoder::TableSection::new(); - for t in self.tables[self.tables.len() - self.num_defined_tables..].iter() { - tables.table(*t); + for (t, init) in self.tables[self.tables.len() - self.defined_tables.len()..] + .iter() + .zip(&self.defined_tables) + { + match init { + Some(init) => { + tables.table_with_init(*t, init); + } + None => { + tables.table(*t); + } + } } module.section(&tables); } @@ -150,10 +159,7 @@ impl Module { let mut globals = wasm_encoder::GlobalSection::new(); for (idx, expr) in &self.defined_globals { let ty = &self.globals[*idx as usize]; - match expr { - GlobalInitExpr::ConstExpr(expr) => globals.global(*ty, expr), - GlobalInitExpr::FuncRef(func) => globals.global(*ty, &ConstExpr::ref_func(*func)), - }; + globals.global(*ty, expr); } module.section(&globals); } @@ -180,24 +186,13 @@ impl Module { return; } let mut elems = wasm_encoder::ElementSection::new(); - let mut exps = vec![]; for el in &self.elems { let elements = match &el.items { - Elements::Expressions(es) => { - exps.clear(); - exps.extend(es.iter().map(|e| { - // TODO(nagisa): generate global.get of imported ref globals too. - match e { - Some(i) => match el.ty { - RefType::FUNCREF => wasm_encoder::ConstExpr::ref_func(*i), - _ => unreachable!(), - }, - None => wasm_encoder::ConstExpr::ref_null(el.ty.heap_type), - } - })); - wasm_encoder::Elements::Expressions(el.ty, &exps) + Elements::Expressions(es) => wasm_encoder::Elements::Expressions(el.ty, es), + Elements::Functions(fs) => { + assert_eq!(el.ty, RefType::FUNCREF); + wasm_encoder::Elements::Functions(fs) } - Elements::Functions(fs) => wasm_encoder::Elements::Functions(fs), }; match &el.kind { ElementKind::Active { table, offset } => { diff --git a/third_party/rust/wasm-smith/src/core/terminate.rs b/third_party/rust/wasm-smith/src/core/terminate.rs index adcfeed54f..7983c35be6 100644 --- a/third_party/rust/wasm-smith/src/core/terminate.rs +++ b/third_party/rust/wasm-smith/src/core/terminate.rs @@ -1,6 +1,5 @@ use super::*; -use std::mem; -use wasm_encoder::BlockType; +use anyhow::{bail, Result}; impl Module { /// Ensure that all of this Wasm module's functions will terminate when @@ -12,16 +11,21 @@ impl Module { /// /// The index of the fuel global is returned, so that you may control how /// much fuel the module is given. - pub fn ensure_termination(&mut self, default_fuel: u32) -> u32 { + /// + /// # Errors + /// + /// Returns an error if any function body was generated with + /// possibly-invalid bytes rather than being generated by wasm-smith. In + /// such a situation this pass does not parse the input bytes and inject + /// instructions, instead it returns an error. + pub fn ensure_termination(&mut self, default_fuel: u32) -> Result<u32> { let fuel_global = self.globals.len() as u32; self.globals.push(GlobalType { val_type: ValType::I32, mutable: true, }); - self.defined_globals.push(( - fuel_global, - GlobalInitExpr::ConstExpr(ConstExpr::i32_const(default_fuel as i32)), - )); + self.defined_globals + .push((fuel_global, ConstExpr::i32_const(default_fuel as i32))); for code in &mut self.code { let check_fuel = |insts: &mut Vec<Instruction>| { @@ -41,10 +45,12 @@ impl Module { let instrs = match &mut code.instructions { Instructions::Generated(list) => list, - // only present on modules contained within - // `MaybeInvalidModule`, which doesn't expose its internal - // `Module`. - Instructions::Arbitrary(_) => unreachable!(), + Instructions::Arbitrary(_) => { + bail!( + "failed to ensure that a function generated due to it \ + containing arbitrary instructions" + ) + } }; let mut new_insts = Vec::with_capacity(instrs.len() * 2); @@ -65,6 +71,6 @@ impl Module { *instrs = new_insts; } - fuel_global + Ok(fuel_global) } } diff --git a/third_party/rust/wasm-smith/src/lib.rs b/third_party/rust/wasm-smith/src/lib.rs index 8d17290473..b985791aae 100644 --- a/third_party/rust/wasm-smith/src/lib.rs +++ b/third_party/rust/wasm-smith/src/lib.rs @@ -58,7 +58,7 @@ mod component; mod config; mod core; -pub use crate::core::{InstructionKind, InstructionKinds, MaybeInvalidModule, Module}; +pub use crate::core::{InstructionKind, InstructionKinds, Module}; use arbitrary::{Result, Unstructured}; pub use component::Component; pub use config::{Config, MemoryOffsetChoices}; @@ -108,10 +108,7 @@ pub(crate) fn limited_str<'a>(max_size: usize, u: &mut Unstructured<'a>) -> Resu Err(e) => { let i = e.valid_up_to(); let valid = u.bytes(i).unwrap(); - let s = unsafe { - debug_assert!(str::from_utf8(valid).is_ok()); - str::from_utf8_unchecked(valid) - }; + let s = str::from_utf8(valid).unwrap(); Ok(s) } } diff --git a/third_party/rust/wasm-smith/tests/available_imports.rs b/third_party/rust/wasm-smith/tests/available_imports.rs index 2ccc4c1cea..27e4852da9 100644 --- a/third_party/rust/wasm-smith/tests/available_imports.rs +++ b/third_party/rust/wasm-smith/tests/available_imports.rs @@ -4,8 +4,11 @@ use arbitrary::{Arbitrary, Unstructured}; use rand::{rngs::SmallRng, RngCore, SeedableRng}; use std::collections::HashMap; use wasm_smith::{Config, Module}; +use wasmparser::Validator; use wasmparser::{Parser, TypeRef, ValType}; -use wasmparser::{Validator, WasmFeatures}; + +mod common; +use common::{parser_features_from_config, validate}; #[test] fn smoke_test_imports_config() { @@ -159,44 +162,3 @@ fn import_config( ); (config, available) } - -fn parser_features_from_config(config: &Config) -> WasmFeatures { - WasmFeatures { - mutable_global: true, - saturating_float_to_int: config.saturating_float_to_int_enabled, - sign_extension: config.sign_extension_ops_enabled, - reference_types: config.reference_types_enabled, - multi_value: config.multi_value_enabled, - bulk_memory: config.bulk_memory_enabled, - simd: config.simd_enabled, - relaxed_simd: config.relaxed_simd_enabled, - multi_memory: config.max_memories > 1, - exceptions: config.exceptions_enabled, - memory64: config.memory64_enabled, - tail_call: config.tail_call_enabled, - function_references: config.gc_enabled, - gc: config.gc_enabled, - - threads: false, - floats: true, - extended_const: false, - component_model: false, - memory_control: false, - component_model_values: false, - component_model_nested_names: false, - } -} - -fn validate(validator: &mut Validator, bytes: &[u8]) { - let err = match validator.validate_all(bytes) { - Ok(_) => return, - Err(e) => e, - }; - eprintln!("Writing Wasm to `test.wasm`"); - drop(std::fs::write("test.wasm", &bytes)); - if let Ok(text) = wasmprinter::print_bytes(bytes) { - eprintln!("Writing WAT to `test.wat`"); - drop(std::fs::write("test.wat", &text)); - } - panic!("wasm failed to validate: {}", err); -} diff --git a/third_party/rust/wasm-smith/tests/common/mod.rs b/third_party/rust/wasm-smith/tests/common/mod.rs new file mode 100644 index 0000000000..cc24eed278 --- /dev/null +++ b/third_party/rust/wasm-smith/tests/common/mod.rs @@ -0,0 +1,43 @@ +use wasm_smith::Config; +use wasmparser::{types::Types, Validator, WasmFeatures}; + +pub fn parser_features_from_config(config: &Config) -> WasmFeatures { + WasmFeatures { + mutable_global: true, + saturating_float_to_int: config.saturating_float_to_int_enabled, + sign_extension: config.sign_extension_ops_enabled, + reference_types: config.reference_types_enabled, + multi_value: config.multi_value_enabled, + bulk_memory: config.bulk_memory_enabled, + simd: config.simd_enabled, + relaxed_simd: config.relaxed_simd_enabled, + multi_memory: config.max_memories > 1, + exceptions: config.exceptions_enabled, + memory64: config.memory64_enabled, + tail_call: config.tail_call_enabled, + function_references: config.gc_enabled, + gc: config.gc_enabled, + + threads: false, + floats: true, + extended_const: false, + component_model: false, + memory_control: false, + component_model_values: false, + component_model_nested_names: false, + } +} + +pub fn validate(validator: &mut Validator, bytes: &[u8]) -> Types { + let err = match validator.validate_all(bytes) { + Ok(types) => return types, + Err(e) => e, + }; + eprintln!("Writing Wasm to `test.wasm`"); + drop(std::fs::write("test.wasm", &bytes)); + if let Ok(text) = wasmprinter::print_bytes(bytes) { + eprintln!("Writing WAT to `test.wat`"); + drop(std::fs::write("test.wat", &text)); + } + panic!("wasm failed to validate: {}", err); +} diff --git a/third_party/rust/wasm-smith/tests/core.rs b/third_party/rust/wasm-smith/tests/core.rs index 5816f8e38b..7286952dc6 100644 --- a/third_party/rust/wasm-smith/tests/core.rs +++ b/third_party/rust/wasm-smith/tests/core.rs @@ -27,7 +27,7 @@ fn smoke_test_ensure_termination() { rng.fill_bytes(&mut buf); let u = Unstructured::new(&buf); if let Ok(mut module) = Module::arbitrary_take_rest(u) { - module.ensure_termination(10); + module.ensure_termination(10).unwrap(); let wasm_bytes = module.to_bytes(); let mut validator = Validator::new_with_features(wasm_features()); @@ -128,6 +128,7 @@ fn smoke_test_wasm_gc() { let mut u = Unstructured::new(&buf); let config = Config { gc_enabled: true, + reference_types_enabled: true, ..Config::default() }; if let Ok(module) = Module::new(config, &mut u) { diff --git a/third_party/rust/wasm-smith/tests/exports.rs b/third_party/rust/wasm-smith/tests/exports.rs new file mode 100644 index 0000000000..ff1dac0cbe --- /dev/null +++ b/third_party/rust/wasm-smith/tests/exports.rs @@ -0,0 +1,147 @@ +use arbitrary::{Arbitrary, Unstructured}; +use rand::{rngs::SmallRng, RngCore, SeedableRng}; +use wasm_smith::{Config, Module}; +use wasmparser::{ + types::EntityType, CompositeType, FuncType, GlobalType, Parser, Validator, WasmFeatures, +}; + +mod common; +use common::{parser_features_from_config, validate}; + +#[derive(Debug, PartialEq)] +enum ExportType { + Func(FuncType), + Global(GlobalType), +} + +#[test] +fn smoke_test_single_export() { + let test = r#" + (module + (func (export "foo") (param i32) (result i64) + unreachable + ) + ) + "#; + smoke_test_exports(test, 11) +} + +#[test] +fn smoke_test_multiple_exports() { + let test = r#" + (module + (func (export "a") (param i32) (result i64) + unreachable + ) + (func (export "b") + unreachable + ) + (func (export "c") + unreachable + ) + ) + "#; + smoke_test_exports(test, 12) +} + +#[test] +fn smoke_test_exported_global() { + let test = r#" + (module + (func (export "a") (param i32 i32 f32 f64) (result f32) + unreachable + ) + (global (export "glob") f64 f64.const 0) + ) + "#; + smoke_test_exports(test, 20) +} + +#[test] +fn smoke_test_export_with_imports() { + let test = r#" + (module + (import "" "foo" (func (param i32))) + (import "" "bar" (global (mut f32))) + (func (param i64) unreachable) + (global i32 (i32.const 0)) + (export "a" (func 0)) + (export "b" (global 0)) + (export "c" (func 1)) + (export "d" (global 1)) + ) + "#; + smoke_test_exports(test, 21) +} + +#[test] +fn smoke_test_with_mutable_global_exports() { + let test = r#" + (module + (global (export "1i32") (mut i32) (i32.const 0)) + (global (export "2i32") (mut i32) (i32.const 0)) + (global (export "1i64") (mut i64) (i64.const 0)) + (global (export "2i64") (mut i64) (i64.const 0)) + (global (export "3i32") (mut i32) (i32.const 0)) + (global (export "3i64") (mut i64) (i64.const 0)) + (global (export "4i32") i32 (i32.const 0)) + (global (export "4i64") i64 (i64.const 0)) + )"#; + smoke_test_exports(test, 22) +} + +fn get_func_and_global_exports(features: WasmFeatures, module: &[u8]) -> Vec<(String, ExportType)> { + let mut validator = Validator::new_with_features(features); + let types = validate(&mut validator, module); + let mut exports = vec![]; + + for payload in Parser::new(0).parse_all(module) { + let payload = payload.unwrap(); + if let wasmparser::Payload::ExportSection(rdr) = payload { + for export in rdr { + let export = export.unwrap(); + match types.entity_type_from_export(&export).unwrap() { + EntityType::Func(core_id) => { + let sub_type = types.get(core_id).expect("Failed to lookup core id"); + assert!(sub_type.is_final); + assert!(sub_type.supertype_idx.is_none()); + let CompositeType::Func(func_type) = &sub_type.composite_type else { + panic!("Expected Func CompositeType, but found {:?}", sub_type); + }; + exports + .push((export.name.to_string(), ExportType::Func(func_type.clone()))); + } + EntityType::Global(global_type) => { + exports.push((export.name.to_string(), ExportType::Global(global_type))) + } + other => { + panic!("Unexpected entity type {:?}", other) + } + } + } + } + } + exports +} + +fn smoke_test_exports(exports_test_case: &str, seed: u64) { + let mut rng = SmallRng::seed_from_u64(seed); + let mut buf = vec![0; 512]; + let wasm = wat::parse_str(exports_test_case).unwrap(); + let expected_exports = get_func_and_global_exports(WasmFeatures::default(), &wasm); + + for _ in 0..1024 { + rng.fill_bytes(&mut buf); + let mut u = Unstructured::new(&buf); + + let mut config = Config::arbitrary(&mut u).expect("arbitrary config"); + config.exports = Some(wasm.clone()); + + let features = parser_features_from_config(&config); + let module = Module::new(config, &mut u).unwrap(); + let wasm_bytes = module.to_bytes(); + + let generated_exports = get_func_and_global_exports(features, &wasm_bytes); + assert_eq!(expected_exports, generated_exports); + } +} |