Mark autogenerated splits & rework ObjInfo::add_split

This commit is contained in:
Luke Street 2023-08-03 19:16:47 -04:00
parent 43856f1b79
commit 4ee63abea4
6 changed files with 255 additions and 37 deletions

View File

@ -202,6 +202,7 @@ pub fn apply_signatures(obj: &mut ObjInfo) -> Result<()> {
{ {
apply_signature(obj, entry, &signature)?; apply_signature(obj, entry, &signature)?;
} }
for &(name, sig_str) in SIGNATURES { for &(name, sig_str) in SIGNATURES {
if let Some((_, symbol)) = obj.symbols.by_name(name)? { if let Some((_, symbol)) = obj.symbols.by_name(name)? {
let addr = symbol.address as u32; let addr = symbol.address as u32;
@ -210,6 +211,7 @@ pub fn apply_signatures(obj: &mut ObjInfo) -> Result<()> {
} }
} }
} }
if let Some((_, symbol)) = obj.symbols.by_name("__init_user")? { if let Some((_, symbol)) = obj.symbols.by_name("__init_user")? {
// __init_user can be overridden, but we can still look for __init_cpp from it // __init_user can be overridden, but we can still look for __init_cpp from it
let mut analyzer = AnalyzerState::default(); let mut analyzer = AnalyzerState::default();
@ -225,6 +227,7 @@ pub fn apply_signatures(obj: &mut ObjInfo) -> Result<()> {
} }
} }
} }
if let Some((_, symbol)) = obj.symbols.by_name("_ctors")? { if let Some((_, symbol)) = obj.symbols.by_name("_ctors")? {
// First entry of ctors is __init_cpp_exceptions // First entry of ctors is __init_cpp_exceptions
let section = obj.section_at(symbol.address as u32)?; let section = obj.section_at(symbol.address as u32)?;
@ -260,30 +263,32 @@ pub fn apply_signatures(obj: &mut ObjInfo) -> Result<()> {
end: address as u32 + 4, end: address as u32 + 4,
align: None, align: None,
common: false, common: false,
}); autogenerated: true,
})?;
} }
} }
} }
} }
if let Some((_, symbol)) = obj.symbols.by_name("_dtors")? { if let Some((_, symbol)) = obj.symbols.by_name("_dtors")? {
let section = obj.section_at(symbol.address as u32)?; let section = obj.section_at(symbol.address as u32)?;
let address = symbol.address; let address = symbol.address;
let section_address = section.address; let section_address = section.address;
let section_index = section.index; let section_index = section.index;
// First entry of dtors is __destroy_global_chain // First entry of dtors is __destroy_global_chain
let target = read_u32(&section.data, address as u32, section_address as u32) let dgc_target = read_u32(&section.data, address as u32, section_address as u32)
.ok_or_else(|| anyhow!("Failed to read _dtors data"))?; .ok_or_else(|| anyhow!("Failed to read _dtors data"))?;
let target2 = read_u32(&section.data, address as u32 + 4, section_address as u32) let fce_target = read_u32(&section.data, address as u32 + 4, section_address as u32)
.ok_or_else(|| anyhow!("Failed to read _dtors data"))?; .ok_or_else(|| anyhow!("Failed to read _dtors data"))?;
let mut target_ok = false; let mut found_dgc = false;
let mut target2_ok = false; let mut found_fce = false;
if target != 0 { if dgc_target != 0 {
if let Some(signature) = check_signatures_str( if let Some(signature) = check_signatures_str(
obj, obj,
target, dgc_target,
include_str!("../../assets/signatures/__destroy_global_chain.yml"), include_str!("../../assets/signatures/__destroy_global_chain.yml"),
)? { )? {
apply_signature(obj, target, &signature)?; apply_signature(obj, dgc_target, &signature)?;
obj.add_symbol( obj.add_symbol(
ObjSymbol { ObjSymbol {
name: "__destroy_global_chain_reference".to_string(), name: "__destroy_global_chain_reference".to_string(),
@ -299,17 +304,22 @@ pub fn apply_signatures(obj: &mut ObjInfo) -> Result<()> {
}, },
true, true,
)?; )?;
target_ok = true; found_dgc = true;
} else {
log::warn!(
"Failed to match __destroy_global_chain signature ({:#010X})",
dgc_target
);
} }
} }
// Second entry of dtors is __fini_cpp_exceptions // Second entry of dtors is __fini_cpp_exceptions
if target2 != 0 { if fce_target != 0 {
if let Some(signature) = check_signatures_str( if let Some(signature) = check_signatures_str(
obj, obj,
target2, fce_target,
include_str!("../../assets/signatures/__fini_cpp_exceptions.yml"), include_str!("../../assets/signatures/__fini_cpp_exceptions.yml"),
)? { )? {
apply_signature(obj, target2, &signature)?; apply_signature(obj, fce_target, &signature)?;
obj.add_symbol( obj.add_symbol(
ObjSymbol { ObjSymbol {
name: "__fini_cpp_exceptions_reference".to_string(), name: "__fini_cpp_exceptions_reference".to_string(),
@ -325,19 +335,27 @@ pub fn apply_signatures(obj: &mut ObjInfo) -> Result<()> {
}, },
true, true,
)?; )?;
target2_ok = true; found_fce = true;
} }
} }
if target_ok && target2_ok && obj.split_for(address as u32).is_none() { if found_dgc {
let mut end = address as u32 + 4;
if found_fce {
end += 4;
}
if obj.split_for(address as u32).is_none() {
obj.add_split(address as u32, ObjSplit { obj.add_split(address as u32, ObjSplit {
unit: "__init_cpp_exceptions.cpp".to_string(), unit: "__init_cpp_exceptions.cpp".to_string(),
end: address as u32 + 8, end,
align: None, align: None,
common: false, common: false,
}); autogenerated: true,
})?;
} }
} }
}
Ok(()) Ok(())
} }

View File

@ -2,8 +2,8 @@ pub mod signatures;
pub mod split; pub mod split;
use std::{ use std::{
cmp::min, cmp::{max, min},
collections::{btree_map, BTreeMap, HashMap}, collections::{btree_map, BTreeMap, BTreeSet, HashMap},
hash::{Hash, Hasher}, hash::{Hash, Hasher},
ops::{Range, RangeBounds}, ops::{Range, RangeBounds},
}; };
@ -144,7 +144,10 @@ pub struct ObjSplit {
pub unit: String, pub unit: String,
pub end: u32, pub end: u32,
pub align: Option<u32>, pub align: Option<u32>,
/// Common BSS
pub common: bool, pub common: bool,
/// Generated, replaceable by user
pub autogenerated: bool,
} }
type SymbolIndex = usize; type SymbolIndex = usize;
@ -580,10 +583,186 @@ impl ObjInfo {
self.splits.range(range).flat_map(|(addr, v)| v.iter().map(move |u| (*addr, u))) self.splits.range(range).flat_map(|(addr, v)| v.iter().map(move |u| (*addr, u)))
} }
pub fn add_split(&mut self, address: u32, split: ObjSplit) { pub fn split_for_unit(
log::debug!("Adding split @ {:#010X}: {:?}", address, split); &self,
// TODO merge with preceding split if possible unit: &str,
section: &ObjSection,
) -> Result<Option<(u32, &ObjSplit)>> {
let mut result = None::<(u32, &ObjSplit)>;
for (addr, split) in self
.splits_for_range(section.address as u32..(section.address + section.size) as u32)
.filter(|(_, split)| split.unit == unit)
{
ensure!(
result.is_none(),
"Multiple splits for unit {} in section {}: {:#010X}, {:#010X}",
unit,
section.name,
result.unwrap().0,
addr
);
result = Some((addr, split));
}
Ok(result)
}
pub fn add_split(&mut self, address: u32, split: ObjSplit) -> Result<()> {
let section = self.section_at(address)?;
let section_start = section.address as u32;
let section_end = (section.address + section.size) as u32;
ensure!(
split.end == 0 || split.end <= section_end,
"Split {} {:#010X}-{:#010X} is outside section {} {:#010X}-{:#010X}",
split.unit,
address,
split.end,
section.name,
section_start,
section_end
);
if let Some((existing_addr, existing_split)) = self.split_for_unit(&split.unit, section)? {
let new_start = min(existing_addr, address);
let new_end = max(existing_split.end, split.end);
// TODO use highest alignment?
let new_align = match (split.align, existing_split.align) {
(Some(a), Some(b)) if a == b => Some(a),
(Some(a), Some(b)) => {
bail!(
"Conflicting alignment for split {} {} {:#010X}-{:#010X}: {:#X} != {:#X}",
split.unit,
section.name,
existing_addr,
existing_split.end,
a,
b
);
}
(Some(a), _) => Some(a),
(_, Some(b)) => Some(b),
_ => None,
};
// TODO don't merge if common flag is different?
ensure!(
split.common == existing_split.common,
"Conflicting common flag for split {} {} {:#010X}-{:#010X} ({}) and {:#010X}-{:#010X} ({})",
split.unit,
section.name,
existing_addr,
existing_split.end,
existing_split.common,
address,
split.end,
split.common
);
// Only set autogenerated flag if both splits are autogenerated
let new_autogenerated = split.autogenerated && existing_split.autogenerated;
// If the new split is contained within the existing split, do nothing
if new_start >= existing_addr && new_end <= existing_split.end {
log::debug!(
"Split {} {} {:#010X}-{:#010X} already covers {:#010X}-{:#010X}",
split.unit,
section.name,
existing_addr,
existing_split.end,
address,
split.end
);
return Ok(());
}
log::debug!(
"Extending split {} {} {:#010X}-{:#010X} to include {:#010X}-{:#010X}: {:#010X}-{:#010X}",
split.unit,
section.name,
existing_addr,
existing_split.end,
address,
split.end,
new_start,
new_end
);
// Check if new split overlaps any existing splits
let mut to_remove = BTreeSet::new();
let mut to_rename = BTreeSet::new();
for (existing_addr, existing_split) in self.splits_for_range(new_start..new_end) {
// TODO the logic in this method should be reworked, this is a hack
if split.autogenerated && !existing_split.autogenerated {
log::debug!(
"-> Found existing split {} {} {:#010X}-{:#010X} (not autogenerated)",
existing_split.unit,
section.name,
existing_addr,
existing_split.end
);
return Ok(());
}
ensure!(
existing_split.autogenerated || existing_split.unit == split.unit,
"New split {} {} {:#010X}-{:#010X} overlaps existing split {} {:#010X}-{:#010X}",
split.unit,
section.name,
new_start,
new_end,
existing_split.unit,
existing_addr,
existing_split.end
);
log::debug!(
"-> Replacing existing split {} {} {:#010X}-{:#010X}",
existing_split.unit,
section.name,
existing_addr,
existing_split.end
);
to_remove.insert(existing_addr);
if existing_split.unit != split.unit {
to_rename.insert(existing_split.unit.clone());
}
}
// Remove overlapping splits
for addr in to_remove {
self.splits.remove(&addr);
}
// Rename any units that were overwritten
// TODO this should also merge with existing splits
for unit in to_rename {
for (existing_addr, existing) in self
.splits
.iter_mut()
.flat_map(|(addr, v)| v.iter_mut().map(move |u| (addr, u)))
.filter(|(_, split)| split.unit == unit)
{
log::debug!(
"-> Renaming {} {:#010X}-{:#010X} to {}",
existing.unit,
existing_addr,
existing.end,
split.unit
);
existing.unit = split.unit.clone();
}
}
self.add_split(new_start, ObjSplit {
unit: split.unit,
end: new_end,
align: new_align,
common: split.common,
autogenerated: new_autogenerated,
})?;
return Ok(());
}
log::debug!("Adding split @ {} {:#010X}: {:?}", section.name, address, split);
self.splits.entry(address).or_default().push(split); self.splits.entry(address).or_default().push(split);
Ok(())
} }
} }

View File

@ -8,8 +8,8 @@ use itertools::Itertools;
use petgraph::{graph::NodeIndex, Graph}; use petgraph::{graph::NodeIndex, Graph};
use crate::obj::{ use crate::obj::{
ObjArchitecture, ObjInfo, ObjKind, ObjReloc, ObjSection, ObjSectionKind, ObjSplit, ObjSymbol, ObjArchitecture, ObjInfo, ObjKind, ObjReloc, ObjRelocKind, ObjSection, ObjSectionKind,
ObjSymbolFlagSet, ObjSymbolFlags, ObjSymbolKind, ObjSplit, ObjSymbol, ObjSymbolFlagSet, ObjSymbolFlags, ObjSymbolKind, SymbolIndex,
}; };
/// Create splits for function pointers in the given section. /// Create splits for function pointers in the given section.
@ -68,6 +68,7 @@ fn split_ctors_dtors(obj: &mut ObjInfo, section_start: u32, section_end: u32) ->
end: current_address + 4, end: current_address + 4,
align: None, align: None,
common: false, common: false,
autogenerated: true,
}); });
} }
if function_split.is_none() { if function_split.is_none() {
@ -77,6 +78,7 @@ fn split_ctors_dtors(obj: &mut ObjInfo, section_start: u32, section_end: u32) ->
end: function_addr + function_symbol.size as u32, end: function_addr + function_symbol.size as u32,
align: None, align: None,
common: false, common: false,
autogenerated: true,
}); });
} }
} }
@ -85,7 +87,7 @@ fn split_ctors_dtors(obj: &mut ObjInfo, section_start: u32, section_end: u32) ->
} }
for (addr, split) in new_splits { for (addr, split) in new_splits {
obj.add_split(addr, split); obj.add_split(addr, split)?;
} }
Ok(()) Ok(())
@ -203,30 +205,40 @@ fn split_extabindex(obj: &mut ObjInfo, section_index: usize, section_start: u32)
log::debug!("Adding splits to unit {}", unit); log::debug!("Adding splits to unit {}", unit);
if extabindex_split.is_none() { if extabindex_split.is_none() {
log::debug!("Adding split for extabindex entry @ {:#010X}", current_address); let end = current_address + 12;
log::debug!(
"Adding split for extabindex entry @ {:#010X}-{:#010X}",
current_address,
end
);
new_splits.insert(current_address, ObjSplit { new_splits.insert(current_address, ObjSplit {
unit: unit.clone(), unit: unit.clone(),
end: current_address + 12, end,
align: None, align: None,
common: false, common: false,
autogenerated: true,
}); });
} }
if extab_split.is_none() { if extab_split.is_none() {
log::debug!("Adding split for extab @ {:#010X}", extab_addr); let end = extab_addr + extab_symbol.size as u32;
log::debug!("Adding split for extab @ {:#010X}-{:#010X}", extab_addr, end);
new_splits.insert(extab_addr, ObjSplit { new_splits.insert(extab_addr, ObjSplit {
unit: unit.clone(), unit: unit.clone(),
end: extab_addr + extab_symbol.size as u32, end,
align: None, align: None,
common: false, common: false,
autogenerated: true,
}); });
} }
if function_split.is_none() { if function_split.is_none() {
log::debug!("Adding split for function @ {:#010X}", function_addr); let end = function_addr + function_symbol.size as u32;
log::debug!("Adding split for function @ {:#010X}-{:#010X}", function_addr, end);
new_splits.insert(function_addr, ObjSplit { new_splits.insert(function_addr, ObjSplit {
unit, unit,
end: function_addr + function_symbol.size as u32, end,
align: None, align: None,
common: false, common: false,
autogenerated: true,
}); });
} }
} }
@ -235,7 +247,7 @@ fn split_extabindex(obj: &mut ObjInfo, section_index: usize, section_start: u32)
} }
for (addr, split) in new_splits { for (addr, split) in new_splits {
obj.add_split(addr, split); obj.add_split(addr, split)?;
} }
Ok(()) Ok(())
@ -310,6 +322,7 @@ fn create_gap_splits(obj: &mut ObjInfo) -> Result<()> {
end: new_split_end, end: new_split_end,
align: None, align: None,
common: false, common: false,
autogenerated: true,
}); });
current_address = new_split_end; current_address = new_split_end;
continue; continue;
@ -330,7 +343,7 @@ fn create_gap_splits(obj: &mut ObjInfo) -> Result<()> {
// Add new splits // Add new splits
for (addr, split) in new_splits { for (addr, split) in new_splits {
obj.add_split(addr, split); obj.add_split(addr, split)?;
} }
Ok(()) Ok(())

View File

@ -347,7 +347,13 @@ pub fn apply_splits<R: BufRead>(r: R, obj: &mut ObjInfo) -> Result<()> {
bail!("Section {} defined outside of unit", name); bail!("Section {} defined outside of unit", name);
} }
(SplitState::Unit(unit), SplitLine::Section { name, start, end, align, common }) => { (SplitState::Unit(unit), SplitLine::Section { name, start, end, align, common }) => {
obj.splits.nested_push(start, ObjSplit { unit: unit.clone(), end, align, common }); obj.splits.nested_push(start, ObjSplit {
unit: unit.clone(),
end,
align,
common,
autogenerated: false,
});
obj.named_sections.insert(start, name); obj.named_sections.insert(start, name);
} }
_ => {} _ => {}

View File

@ -267,6 +267,7 @@ pub fn process_elf<P: AsRef<Path>>(path: P) -> Result<ObjInfo> {
end: 0, // TODO end: 0, // TODO
align: None, align: None,
common: false, // TODO common: false, // TODO
autogenerated: false,
}); });
} }
} }

View File

@ -709,6 +709,7 @@ pub fn apply_map(result: &MapInfo, obj: &mut ObjInfo) -> Result<()> {
end: 0, // TODO? end: 0, // TODO?
align: None, align: None,
common: false, // TODO? common: false, // TODO?
autogenerated: false,
}); });
} }
section_order.push((section.clone(), units)); section_order.push((section.clone(), units));