feat: Support DWARF debug info in PE files

CHANGELOG.md

@@ -7,18 +7,21 @@
 - `PortablePdbDebugSession` now returns files referenced in the Portable PDB file. ([#729](https://github.com/getsentry/symbolic/pull/729))
 - `PortablePdbDebugSession` now returns source files embedded in the Portable PDB file. ([#734](https://github.com/getsentry/symbolic/pull/734))
 - Implement `symbolic_common::AsSelf` `for SourceMapCache` ([#742](https://github.com/getsentry/symbolic/pull/742))
+- Debug information can now be retrieved from PE's with DWARF debug info. ([#744](https://github.com/getsentry/symbolic/pull/744))
 
 **Breaking changes**:
 
 - Demangling functionality is removed from C and Python bindings. ([#730](https://github.com/getsentry/symbolic/pull/730))
 - The fields of `FileInfo` and the `compilation_dir` field on `FileEntry` are now private. ([#729](https://github.com/getsentry/symbolic/pull/729))
 - `PortablePdbDebugSession` now has a lifetime parameter. ([#729](https://github.com/getsentry/symbolic/pull/729))
+- `PeDebugSession` placeholder has been removed. ([#744](https://github.com/getsentry/symbolic/pull/744))
 
 **Thank you**:
 
 Features, fixes and improvements in this release have been contributed by:
 
 - [@vaind](https://github.com/vaind)
+- [@casept](https://github.com/casept)
 
 ## 10.2.1
 

README.md

@@ -18,7 +18,7 @@ Symbolic provides the following functionality:
 - Symbolication based on custom cache files (symcache)
 - Symbol cache file generators from:
   - Mach, ELF and PE symbol tables
-  - Mach and ELF embedded DWARF data
+  - Mach, ELF and PE embedded DWARF data
   - PDB CodeView debug information
   - Breakpad symbol files
 - Demangling support

symbolic-debuginfo/src/dwarf.rs

@@ -1,11 +1,13 @@
 //! Support for DWARF debugging information, common to ELF and MachO.
+//! In rare cases, PE's may contain it as well.
 //!
-//! The central element of this module is the [`Dwarf`] trait, which is implemented by [`ElfObject`]
-//! and [`MachObject`]. The dwarf debug session object can be obtained via getters on those types.
+//! The central element of this module is the [`Dwarf`] trait, which is implemented by [`ElfObject`],
+//! [`MachObject`] and [`PeObject`]. The dwarf debug session object can be obtained via getters on those types.
 //!
 //! [`Dwarf`]: trait.Dwarf.html
 //! [`ElfObject`]: ../elf/struct.ElfObject.html
 //! [`MachObject`]: ../macho/struct.MachObject.html
+//! [`PeObject`]: ../pe/struct.PeObject.html
 
 use std::borrow::Cow;
 use std::collections::BTreeSet;

symbolic-debuginfo/src/object.rs

@@ -331,7 +331,7 @@ impl<'data> Object<'data> {
                 .map_err(ObjectError::transparent),
             Object::Pe(ref o) => o
                 .debug_session()
-                .map(ObjectDebugSession::Pe)
+                .map(ObjectDebugSession::Dwarf)
                 .map_err(ObjectError::transparent),
             Object::SourceBundle(ref o) => o
                 .debug_session()
@@ -446,7 +446,6 @@ pub enum ObjectDebugSession<'d> {
     Breakpad(BreakpadDebugSession<'d>),
     Dwarf(DwarfDebugSession<'d>),
     Pdb(PdbDebugSession<'d>),
-    Pe(PeDebugSession<'d>),
     SourceBundle(SourceBundleDebugSession<'d>),
     PortablePdb(PortablePdbDebugSession<'d>),
 }
@@ -464,7 +463,6 @@ impl<'d> ObjectDebugSession<'d> {
             ObjectDebugSession::Breakpad(ref s) => ObjectFunctionIterator::Breakpad(s.functions()),
             ObjectDebugSession::Dwarf(ref s) => ObjectFunctionIterator::Dwarf(s.functions()),
             ObjectDebugSession::Pdb(ref s) => ObjectFunctionIterator::Pdb(s.functions()),
-            ObjectDebugSession::Pe(ref s) => ObjectFunctionIterator::Pe(s.functions()),
             ObjectDebugSession::SourceBundle(ref s) => {
                 ObjectFunctionIterator::SourceBundle(s.functions())
             }
@@ -480,7 +478,6 @@ impl<'d> ObjectDebugSession<'d> {
             ObjectDebugSession::Breakpad(ref s) => ObjectFileIterator::Breakpad(s.files()),
             ObjectDebugSession::Dwarf(ref s) => ObjectFileIterator::Dwarf(s.files()),
             ObjectDebugSession::Pdb(ref s) => ObjectFileIterator::Pdb(s.files()),
-            ObjectDebugSession::Pe(ref s) => ObjectFileIterator::Pe(s.files()),
             ObjectDebugSession::SourceBundle(ref s) => ObjectFileIterator::SourceBundle(s.files()),
             ObjectDebugSession::PortablePdb(ref s) => ObjectFileIterator::PortablePdb(s.files()),
         }
@@ -500,9 +497,6 @@ impl<'d> ObjectDebugSession<'d> {
             ObjectDebugSession::Pdb(ref s) => {
                 s.source_by_path(path).map_err(ObjectError::transparent)
             }
-            ObjectDebugSession::Pe(ref s) => {
-                s.source_by_path(path).map_err(ObjectError::transparent)
-            }
             ObjectDebugSession::SourceBundle(ref s) => {
                 s.source_by_path(path).map_err(ObjectError::transparent)
             }
@@ -537,7 +531,6 @@ pub enum ObjectFunctionIterator<'s> {
     Breakpad(BreakpadFunctionIterator<'s>),
     Dwarf(DwarfFunctionIterator<'s>),
     Pdb(PdbFunctionIterator<'s>),
-    Pe(PeFunctionIterator<'s>),
     SourceBundle(SourceBundleFunctionIterator<'s>),
     PortablePdb(PortablePdbFunctionIterator<'s>),
 }
@@ -556,9 +549,6 @@ impl<'s> Iterator for ObjectFunctionIterator<'s> {
             ObjectFunctionIterator::Pdb(ref mut i) => {
                 Some(i.next()?.map_err(ObjectError::transparent))
             }
-            ObjectFunctionIterator::Pe(ref mut i) => {
-                Some(i.next()?.map_err(ObjectError::transparent))
-            }
             ObjectFunctionIterator::SourceBundle(ref mut i) => {
                 Some(i.next()?.map_err(ObjectError::transparent))
             }
@@ -576,7 +566,6 @@ pub enum ObjectFileIterator<'s> {
     Breakpad(BreakpadFileIterator<'s>),
     Dwarf(DwarfFileIterator<'s>),
     Pdb(PdbFileIterator<'s>),
-    Pe(PeFileIterator<'s>),
     SourceBundle(SourceBundleFileIterator<'s>),
     PortablePdb(PortablePdbFileIterator<'s>),
 }
@@ -593,7 +582,6 @@ impl<'s> Iterator for ObjectFileIterator<'s> {
                 Some(i.next()?.map_err(ObjectError::transparent))
             }
             ObjectFileIterator::Pdb(ref mut i) => Some(i.next()?.map_err(ObjectError::transparent)),
-            ObjectFileIterator::Pe(ref mut i) => Some(i.next()?.map_err(ObjectError::transparent)),
             ObjectFileIterator::SourceBundle(ref mut i) => {
                 Some(i.next()?.map_err(ObjectError::transparent))
             }

symbolic-debuginfo/src/pe.rs

@@ -3,14 +3,15 @@
 use std::borrow::Cow;
 use std::error::Error;
 use std::fmt;
-use std::marker::PhantomData;
 
+use gimli::RunTimeEndian;
 use goblin::pe;
 use thiserror::Error;
 
 use symbolic_common::{Arch, AsSelf, CodeId, DebugId, Uuid};
 
 use crate::base::*;
+use crate::dwarf::*;
 use crate::Parse;
 
 pub use goblin::pe::exception::*;
@@ -59,7 +60,8 @@ fn is_pe_stub(pe: &pe::PE<'_>) -> bool {
 /// container, [`PdbObject`]. The PE file contains a reference to the PDB and vice versa to verify
 /// that the files belong together.
 ///
-/// While in rare instances, PE files might contain debug information, this case is not supported.
+/// In rare instances, PE files might contain debug information.
+/// This is supported for DWARF debug information.
 ///
 /// [`PdbObject`]: ../pdb/struct.PdbObject.html
 pub struct PeObject<'data> {
@@ -108,7 +110,7 @@ impl<'data> PeObject<'data> {
 
     /// The debug information identifier of this PE.
     ///
-    /// Since debug information is stored in an external
+    /// Since debug information is usually stored in an external
     /// [`PdbObject`](crate::pdb::PdbObject), this identifier actually refers to the
     /// PDB. While strictly the filename of the PDB would also be necessary fully resolve
     /// it, in most instances the GUID and age contained in this identifier are sufficient.
@@ -194,9 +196,10 @@ impl<'data> PeObject<'data> {
 
     /// Determines whether this object contains debug information.
     ///
-    /// This is always `false`, as debug information is not supported for PE files.
+    /// Not usually the case, except for PE's generated by some alternative toolchains
+    /// which contain DWARF debug info.
     pub fn has_debug_info(&self) -> bool {
-        false
+        self.section(".debug_info").is_some()
     }
 
     /// Determines whether this object contains embedded source.
@@ -209,9 +212,21 @@ impl<'data> PeObject<'data> {
         false
     }
 
-    /// Constructs a no-op debugging session.
-    pub fn debug_session(&self) -> Result<PeDebugSession<'data>, PeError> {
-        Ok(PeDebugSession { _ph: PhantomData })
+    /// Constructs a debugging session.
+    ///
+    /// A debugging session loads certain information from the object file and creates caches for
+    /// efficient access to various records in the debug information. Since this can be quite a
+    /// costly process, try to reuse the debugging session as long as possible.
+    ///
+    /// PE files usually don't have embedded debugging information,
+    /// but some toolchains (e.g. MinGW) generate DWARF debug info.
+    ///
+    /// Constructing this session will also work if the object does not contain debugging
+    /// information, in which case the session will be a no-op. This can be checked via
+    /// [`has_debug_info`](struct.PeObject.html#method.has_debug_info).
+    pub fn debug_session(&self) -> Result<DwarfDebugSession<'data>, DwarfError> {
+        let symbols = self.symbol_map();
+        DwarfDebugSession::parse(self, symbols, self.load_address() as i64, self.kind())
     }
 
     /// Determines whether this object contains stack unwinding information.
@@ -229,6 +244,17 @@ impl<'data> PeObject<'data> {
         &self.pe.sections
     }
 
+    /// Returns the `SectionTable` for the section with this name, if present.
+    pub fn section(&self, name: &str) -> Option<SectionTable> {
+        for s in &self.pe.sections {
+            let sect_name = s.name();
+            if sect_name.is_ok() && sect_name.unwrap() == name {
+                return Some(s.clone());
+            }
+        }
+        None
+    }
+
     /// Returns exception data containing unwind information.
     pub fn exception_data(&self) -> Option<&ExceptionData<'_>> {
         if self.is_stub {
@@ -277,8 +303,8 @@ impl<'data> Parse<'data> for PeObject<'data> {
 }
 
 impl<'data: 'object, 'object> ObjectLike<'data, 'object> for PeObject<'data> {
-    type Error = PeError;
-    type Session = PeDebugSession<'data>;
+    type Error = DwarfError;
+    type Session = DwarfDebugSession<'data>;
     type SymbolIterator = PeSymbolIterator<'data, 'object>;
 
     fn file_format(&self) -> FileFormat {
@@ -357,54 +383,26 @@ impl<'data, 'object> Iterator for PeSymbolIterator<'data, 'object> {
     }
 }
 
-/// Debug session for PE objects.
-///
-/// Since debug information in PE containers is not supported, this session consists of NoOps and
-/// always returns empty results.
-#[derive(Debug)]
-pub struct PeDebugSession<'data> {
-    _ph: PhantomData<&'data ()>,
-}
-
-impl<'data> PeDebugSession<'data> {
-    /// Returns an iterator over all functions in this debug file.
-    pub fn functions(&self) -> PeFunctionIterator<'_> {
-        std::iter::empty()
-    }
-
-    /// Returns an iterator over all source files referenced by this debug file.
-    pub fn files(&self) -> PeFileIterator<'_> {
-        std::iter::empty()
-    }
-
-    /// Looks up a file's source contents by its full canonicalized path.
-    ///
-    /// The given path must be canonicalized.
-    pub fn source_by_path(&self, _path: &str) -> Result<Option<Cow<'_, str>>, PeError> {
-        Ok(None)
+impl<'data> Dwarf<'data> for PeObject<'data> {
+    fn endianity(&self) -> RunTimeEndian {
+        // According to https://reverseengineering.stackexchange.com/questions/17922/determining-endianness-of-pe-files-windows-on-arm,
+        // the only known platform running PE's with big-endian code is the Xbox360. Probably not worth handling.
+        RunTimeEndian::Little
+    }
+
+    fn raw_section(&self, name: &str) -> Option<DwarfSection<'data>> {
+        // Name is given without leading "."
+        let sect = self.section(&format!(".{}", name))?;
+        let start = sect.pointer_to_raw_data as usize;
+        let end = start + (sect.virtual_size as usize);
+        let dwarf_data: &'data [u8] = self.data.get(start..end)?;
+        let dwarf_sect = DwarfSection {
+            // TODO: What about 64-bit PE+? Still 32 bit?
+            address: u64::from(sect.virtual_address),
+            data: Cow::from(dwarf_data),
+            offset: u64::from(sect.pointer_to_raw_data),
+            align: 4096, // TODO: Does goblin expose this? For now, assume 4K page size
+        };
+        Some(dwarf_sect)
     }
 }
-
-impl<'session> DebugSession<'session> for PeDebugSession<'_> {
-    type Error = PeError;
-    type FunctionIterator = PeFunctionIterator<'session>;
-    type FileIterator = PeFileIterator<'session>;
-
-    fn functions(&'session self) -> Self::FunctionIterator {
-        self.functions()
-    }
-
-    fn files(&'session self) -> Self::FileIterator {
-        self.files()
-    }
-
-    fn source_by_path(&self, path: &str) -> Result<Option<Cow<'_, str>>, Self::Error> {
-        self.source_by_path(path)
-    }
-}
-
-/// An iterator over functions in a PE file.
-pub type PeFunctionIterator<'s> = std::iter::Empty<Result<Function<'s>, PeError>>;
-
-/// An iterator over source files in a PE file.
-pub type PeFileIterator<'s> = std::iter::Empty<Result<FileEntry<'s>, PeError>>;

symbolic-debuginfo/tests/snapshots/test_objects__pe_dwarf_files.snap

@@ -0,0 +1,15 @@
+---
+source: symbolic-debuginfo/tests/test_objects.rs
+expression: "FilesDebug(&files[..10])"
+---
+/mxe/tmp-gcc-x86_64-w64-mingw32.static/gcc-11.3.0.build_/mingw-w64-v10.0.0/mingw-w64-crt/crt/crtexe.c
+/mxe/tmp-gcc-x86_64-w64-mingw32.static/gcc-11.3.0.build_/mingw-w64-v10.0.0/mingw-w64-crt/crt/crtexe.c
+/mxe/usr/x86_64-w64-mingw32.static/include/winnt.h
+/mxe/usr/x86_64-w64-mingw32.static/include/psdk_inc/intrin-impl.h
+/mxe/usr/x86_64-w64-mingw32.static/include/corecrt.h
+/mxe/usr/x86_64-w64-mingw32.static/include/minwindef.h
+/mxe/usr/x86_64-w64-mingw32.static/include/basetsd.h
+/mxe/usr/x86_64-w64-mingw32.static/include/stdlib.h
+/mxe/usr/x86_64-w64-mingw32.static/include/errhandlingapi.h
+/mxe/usr/x86_64-w64-mingw32.static/include/processthreadsapi.h
+