read/macho: fix data for segments in dyld caches (#673)

Use the correct subcache for operations on the segment itself.
Individual section data was already using the correct subcache.

This also does the change for section relocations, but they shouldn't
be present in the dyld cache.

Author: philipc

Diff for: src/read/macho/file.rs

@@ -44,7 +44,7 @@ where
     pub(super) header_offset: u64,
     pub(super) header: &'data Mach,
     pub(super) segments: Vec<MachOSegmentInternal<'data, Mach, R>>,
-    pub(super) sections: Vec<MachOSectionInternal<'data, Mach>>,
+    pub(super) sections: Vec<MachOSectionInternal<'data, Mach, R>>,
     pub(super) symbols: SymbolTable<'data, Mach, R>,
 }
 
@@ -65,11 +65,10 @@ where
         if let Ok(mut commands) = header.load_commands(endian, data, 0) {
             while let Ok(Some(command)) = commands.next() {
                 if let Some((segment, section_data)) = Mach::Segment::from_command(command)? {
-                    let segment_index = segments.len();
                     segments.push(MachOSegmentInternal { segment, data });
                     for section in segment.sections(endian, section_data)? {
                         let index = SectionIndex(sections.len() + 1);
-                        sections.push(MachOSectionInternal::parse(index, segment_index, section));
+                        sections.push(MachOSectionInternal::parse(index, section, data));
                     }
                 } else if let Some(symtab) = command.symtab()? {
                     symbols = symtab.symbols(endian, data)?;
@@ -110,6 +109,7 @@ where
                 if let Some((segment, section_data)) = Mach::Segment::from_command(command)? {
                     // Each segment can be stored in a different subcache. Get the segment's
                     // address and look it up in the cache mappings, to find the correct cache data.
+                    // This was observed for the arm64e __LINKEDIT segment in macOS 12.0.1.
                     let addr = segment.vmaddr(endian).into();
                     let (data, _offset) = image
                         .cache
@@ -118,12 +118,11 @@ where
                     if segment.name() == macho::SEG_LINKEDIT.as_bytes() {
                         linkedit_data = Some(data);
                     }
-                    let segment_index = segments.len();
                     segments.push(MachOSegmentInternal { segment, data });
 
                     for section in segment.sections(endian, section_data)? {
                         let index = SectionIndex(sections.len() + 1);
-                        sections.push(MachOSectionInternal::parse(index, segment_index, section));
+                        sections.push(MachOSectionInternal::parse(index, section, data));
                     }
                 } else if let Some(st) = command.symtab()? {
                     symtab = Some(st);
@@ -154,23 +153,14 @@ where
     pub(super) fn section_internal(
         &self,
         index: SectionIndex,
-    ) -> Result<&MachOSectionInternal<'data, Mach>> {
+    ) -> Result<&MachOSectionInternal<'data, Mach, R>> {
         index
             .0
             .checked_sub(1)
             .and_then(|index| self.sections.get(index))
             .read_error("Invalid Mach-O section index")
     }
 
-    pub(super) fn segment_internal(
-        &self,
-        index: usize,
-    ) -> Result<&MachOSegmentInternal<'data, Mach, R>> {
-        self.segments
-            .get(index)
-            .read_error("Invalid Mach-O segment index")
-    }
-
     /// Returns the endianness.
     pub fn endian(&self) -> Mach::Endian {
         self.endian

Diff for: src/read/macho/section.rs

@@ -25,7 +25,7 @@ where
     R: ReadRef<'data>,
 {
     pub(super) file: &'file MachOFile<'data, Mach, R>,
-    pub(super) iter: slice::Iter<'file, MachOSectionInternal<'data, Mach>>,
+    pub(super) iter: slice::Iter<'file, MachOSectionInternal<'data, Mach, R>>,
 }
 
 impl<'data, 'file, Mach, R> fmt::Debug for MachOSectionIterator<'data, 'file, Mach, R>
@@ -71,7 +71,7 @@ where
     R: ReadRef<'data>,
 {
     pub(super) file: &'file MachOFile<'data, Mach, R>,
-    pub(super) internal: MachOSectionInternal<'data, Mach>,
+    pub(super) internal: MachOSectionInternal<'data, Mach, R>,
 }
 
 impl<'data, 'file, Mach, R> MachOSection<'data, 'file, Mach, R>
@@ -80,11 +80,9 @@ where
     R: ReadRef<'data>,
 {
     fn bytes(&self) -> Result<&'data [u8]> {
-        let segment_index = self.internal.segment_index;
-        let segment = self.file.segment_internal(segment_index)?;
         self.internal
             .section
-            .data(self.file.endian, segment.data)
+            .data(self.file.endian, self.internal.data)
             .read_error("Invalid Mach-O section size or offset")
     }
 }
@@ -193,7 +191,7 @@ where
             relocations: self
                 .internal
                 .section
-                .relocations(self.file.endian, self.file.data)
+                .relocations(self.file.endian, self.internal.data)
                 .unwrap_or(&[])
                 .iter(),
         }
@@ -211,19 +209,19 @@ where
 }
 
 #[derive(Debug, Clone, Copy)]
-pub(super) struct MachOSectionInternal<'data, Mach: MachHeader> {
+pub(super) struct MachOSectionInternal<'data, Mach: MachHeader, R: ReadRef<'data>> {
     pub index: SectionIndex,
-    pub segment_index: usize,
     pub kind: SectionKind,
     pub section: &'data Mach::Section,
+    /// The data for the file that contains the section data.
+    ///
+    /// This is required for dyld caches, where this may be a different subcache
+    /// from the file containing the Mach-O load commands.
+    pub data: R,
 }
 
-impl<'data, Mach: MachHeader> MachOSectionInternal<'data, Mach> {
-    pub(super) fn parse(
-        index: SectionIndex,
-        segment_index: usize,
-        section: &'data Mach::Section,
-    ) -> Self {
+impl<'data, Mach: MachHeader, R: ReadRef<'data>> MachOSectionInternal<'data, Mach, R> {
+    pub(super) fn parse(index: SectionIndex, section: &'data Mach::Section, data: R) -> Self {
         // TODO: we don't validate flags, should we?
         let kind = match (section.segment_name(), section.name()) {
             (b"__TEXT", b"__text") => SectionKind::Text,
@@ -246,9 +244,9 @@ impl<'data, Mach: MachHeader> MachOSectionInternal<'data, Mach> {
         };
         MachOSectionInternal {
             index,
-            segment_index,
             kind,
             section,
+            data,
         }
     }
 }

Diff for: src/read/macho/segment.rs

@@ -69,7 +69,7 @@ where
     fn bytes(&self) -> Result<&'data [u8]> {
         self.internal
             .segment
-            .data(self.file.endian, self.file.data)
+            .data(self.file.endian, self.internal.data)
             .read_error("Invalid Mach-O segment size or offset")
     }
 }
@@ -149,8 +149,12 @@ where
 
 #[derive(Debug, Clone, Copy)]
 pub(super) struct MachOSegmentInternal<'data, Mach: MachHeader, R: ReadRef<'data>> {
-    pub data: R,
     pub segment: &'data Mach::Segment,
+    /// The data for the file that contains the segment data.
+    ///
+    /// This is required for dyld caches, where this may be a different subcache
+    /// from the file containing the Mach-O load commands.
+    pub data: R,
 }
 
 /// A trait for generic access to [`macho::SegmentCommand32`] and [`macho::SegmentCommand64`].