Synthesize sections from section map (fixes #153)

Author: vvuk

src/dbi.rs

@@ -107,6 +107,19 @@ impl<'s> DebugInformation<'s> {
         let contributions_buf = buf.take(self.header.section_contribution_size as usize)?;
         DBISectionContributionIter::parse(contributions_buf.into())
     }
+
+    /// Returns an iterator that can traverse the section map in sequential order. Also known as the "OMF Segment map".
+    pub fn section_map(&self) -> Result<DBISectionMapIter<'_>> {
+        let mut buf = self.stream.parse_buffer();
+        // drop the header, modules list, and section contributions list
+        let offset = self.header_len
+            + self.header.module_list_size as usize
+            + self.header.section_contribution_size as usize;
+
+        buf.take(offset)?;
+        let section_map_buf = buf.take(self.header.section_map_size as usize)?;
+        DBISectionMapIter::parse(section_map_buf.into())
+    }
 }
 
 /// The version of the PDB format.
@@ -584,6 +597,86 @@ impl<'c> FallibleIterator for DBISectionContributionIter<'c> {
     }
 }
 
+/// See https://github.com/google/syzygy/blob/8164b24ebde9c5649c9a09e88a7fc0b0fcbd1bc5/syzygy/pdb/pdb_data.h#L172
+/// Also see https://www.virtualbox.org/browser/vbox/trunk/include/iprt/formats/codeview.h?rev=93115#L272
+/// This is also known as OMF Segment Map. In the OMF SegmentMap structure, flags and section_type
+/// are a single 16-bit value.
+#[derive(Debug, Copy, Clone)]
+pub struct DBISectionMapItem {
+    /// flags: 0x1 read, 0x2 write, 0x4 execute, 0x8 32-bit
+    pub flags: u8,
+    /// section_type: 0x1 = SEL, 0x2 = ABS, 0x10 = GROUP
+    pub section_type: u8,
+    /// Overlay number
+    pub overlay: u16,
+    /// group index, 0 if not relevant
+    pub group: u16,
+    /// Technically "frame" in OMF SegmentMap, which is complicated
+    pub section_number: u16,
+    /// Index into name table, or 0xffff
+    pub seg_name_index: u16,
+    /// Index into name table, or 0xffff
+    pub class_name_index: u16,
+    /// RVA offset of this section
+    pub rva_offset: u32,
+    /// Length of this section
+    pub section_length: u32,
+}
+
+impl DBISectionMapItem {
+    fn parse(buf: &mut ParseBuffer<'_>) -> Result<Self> {
+        Ok(Self {
+            flags: buf.parse_u8()?,
+            section_type: buf.parse_u8()?,
+            overlay: buf.parse_u16()?,
+            group: buf.parse_u16()?,
+            section_number: buf.parse_u16()?,
+            seg_name_index: buf.parse_u16()?,
+            class_name_index: buf.parse_u16()?,
+            rva_offset: buf.parse_u32()?,
+            section_length: buf.parse_u32()?,
+        })
+    }
+}
+
+/// A `DBISectionMapIter` iterates over the section map in the DBI section, producing `DBISectionMap`s.
+#[derive(Debug)]
+pub struct DBISectionMapIter<'c> {
+    /// The section count.
+    pub sec_count: u16,
+    /// The logical section count. Typically equals sec_count, if no groups are in use. (?)
+    pub sec_count_log: u16,
+    buf: ParseBuffer<'c>,
+}
+
+impl<'c> DBISectionMapIter<'c> {
+    fn parse(mut buf: ParseBuffer<'c>) -> Result<Self> {
+        let sec_count = buf.parse_u16()?;
+        let sec_count_log = buf.parse_u16()?;
+
+        Ok(Self {
+            buf,
+            sec_count,
+            sec_count_log,
+        })
+    }
+}
+
+impl<'c> FallibleIterator for DBISectionMapIter<'c> {
+    type Item = DBISectionMapItem;
+    type Error = Error;
+
+    fn next(&mut self) -> result::Result<Option<Self::Item>, Self::Error> {
+        // see if we're at EOF
+        if self.buf.is_empty() {
+            return Ok(None);
+        }
+
+        let segmap = Self::Item::parse(&mut self.buf)?;
+        Ok(Some(segmap))
+    }
+}
+
 /// A `DbgDataHdr`, which contains a series of (optional) MSF stream numbers.
 #[derive(Debug, Copy, Clone)]
 #[allow(dead_code)] // reason = "unused fields added for completeness"

src/pdb.rs

@@ -5,7 +5,8 @@
 // http://opensource.org/licenses/MIT>, at your option. This file may not be
 // copied, modified, or distributed except according to those terms.
 
-use crate::common::*;
+use fallible_iterator::FallibleIterator;
+
 use crate::dbi::{DBIExtraStreams, DBIHeader, DebugInformation, Module};
 use crate::framedata::FrameTable;
 use crate::modi::ModuleInfo;
@@ -17,6 +18,7 @@ use crate::source::Source;
 use crate::strings::StringTable;
 use crate::symbol::SymbolTable;
 use crate::tpi::{IdInformation, TypeInformation};
+use crate::{common::*, SectionCharacteristics};
 
 // Some streams have a fixed stream index.
 // http://llvm.org/docs/PDB/index.html
@@ -242,7 +244,7 @@ impl<'s, S: Source<'s> + 's> PDB<'s, S> {
         let index = self.extra_streams()?.section_headers;
         let stream = match self.raw_stream(index)? {
             Some(stream) => stream,
-            None => return Ok(None),
+            None => return self.maybe_synthesize_section(),
         };
 
         let mut buf = stream.parse_buffer();
@@ -254,6 +256,67 @@ impl<'s, S: Source<'s> + 's> PDB<'s, S> {
         Ok(Some(headers))
     }
 
+    // If there are no section_headers in the file, attempt to synthesize sections
+    // based on the section map. This seems to be necessary to handle NGEN-generated PDB
+    // files (.ni.pdb from Crossgen2).
+    fn maybe_synthesize_section(&mut self) -> Result<Option<Vec<ImageSectionHeader>>> {
+        // If we have OMAP From data, I don't believe we can do this, because the RVAs
+        // won't map. But I'm not 100% sure of that, be conservative.
+        if self.omap_from_src()?.is_some() {
+            return Ok(None);
+        }
+
+        let debug_info = self.debug_information()?;
+        let sec_map = debug_info.section_map()?;
+        if sec_map.sec_count != sec_map.sec_count_log {
+            return Ok(None);
+        }
+        let sec_map = sec_map.collect::<Vec<_>>()?;
+
+        let mut rva = 0x1000u32; // in the absence of explicit section data, this starts at 0x1000
+        let sections = sec_map.into_iter()
+        .filter(|sm| {
+            // the section with a bogus section length also doesn't have any rwx flags,
+            // and has section_type == 2
+            sm.section_type == 1 && // "SEL" section, not ABS (0x2) or GROUP (0x10)
+            sm.section_length != u32::MAX // shouldn't happen, but just in case
+        })
+        .map(|sm| {
+            let mut characteristics = 0u32;
+            if sm.flags & 0x1 != 0 { // R
+                characteristics |= 0x40000000; // IMAGE_SCN_MEM_READ
+            }
+            if sm.flags & 0x2 != 0 { // W
+                characteristics |= 0x80000000; // IMAGE_SCN_MEM_WRITE
+            }
+            if sm.flags & 0x4 != 0 { // X
+                characteristics |= 0x20000000; // IMAGE_SCN_MEM_EXECUTE
+                characteristics |= 0x20; // IMAGE_SCN_CNT_CODE
+            }
+
+            if sm.rva_offset != 0 {
+                eprintln!("pdb: synthesizing section with rva_offset != 0, might not be correct! {:?}", sm);
+            }
+
+            let this_rva = rva + sm.rva_offset;
+            rva = this_rva + sm.section_length;
+            ImageSectionHeader {
+                name: [0; 8],
+                virtual_size: sm.section_length,
+                virtual_address: this_rva,
+                size_of_raw_data: sm.section_length,
+                pointer_to_raw_data: 0,
+                pointer_to_relocations: 0,
+                pointer_to_line_numbers: 0,
+                number_of_relocations: 0,
+                number_of_line_numbers: 0,
+                characteristics: SectionCharacteristics(characteristics),
+            }
+        }).collect::<Vec<_>>();
+
+        Ok(Some(sections))
+    }
+
     /// Retrieve the global frame data table.
     ///
     /// This table describes the stack frame layout for functions from all modules in the PDB. Not