[Zucchini] Create elf disassembler.
Creates Disassembler that recognises and parses ELF format. For now, it only supports Intel architeture. Support for Arm will be added in follow-up CLs.
Change-Id: Ibdcf113b573f22844b6a1611c5ff6df46829b9b3
Reviewed-on: https://chromium-review.googlesource.com/1136841
Commit-Queue: Etienne Pierre-Doray <etiennep@chromium.org>
Reviewed-by: Greg Thompson <grt@chromium.org>
Reviewed-by: Samuel Huang <huangs@chromium.org>
Cr-Commit-Position: refs/heads/master@{#582233}
NOKEYCHECK=True
GitOrigin-RevId: 3c64e078fea9f23e44939c25ca02cf05b72b2c40
diff --git a/BUILD.gn b/BUILD.gn
index 4e289e3..b15da9c 100644
--- a/BUILD.gn
+++ b/BUILD.gn
@@ -11,12 +11,14 @@
# Disable DEX on Windows Official Builds.
_enable_dex = !(is_win && is_official_build)
+ _enable_elf = true
_enable_win = true
# Disable ZTF (Zucchini Text Format) on official builds it is for testing only.
_enable_ztf = !is_official_build
flags = [
"ENABLE_DEX=$_enable_dex",
+ "ENABLE_ELF=$_enable_elf",
"ENABLE_WIN=$_enable_win",
"ENABLE_ZTF=$_enable_ztf",
]
@@ -42,6 +44,8 @@
"disassembler.h",
"disassembler_dex.cc",
"disassembler_dex.h",
+ "disassembler_elf.cc",
+ "disassembler_elf.h",
"disassembler_no_op.cc",
"disassembler_no_op.h",
"disassembler_win32.cc",
@@ -158,6 +162,7 @@
"buffer_view_unittest.cc",
"crc32_unittest.cc",
"disassembler_dex_unittest.cc",
+ "disassembler_elf_unittest.cc",
"disassembler_ztf_unittest.cc",
"element_detection_unittest.cc",
"encoded_view_unittest.cc",
diff --git a/abs32_utils.cc b/abs32_utils.cc
index ab4372c..b3fd9f4 100644
--- a/abs32_utils.cc
+++ b/abs32_utils.cc
@@ -136,10 +136,12 @@
for (auto unit = abs32_rva_extractor_.GetNext(); unit.has_value();
unit = abs32_rva_extractor_.GetNext()) {
offset_t location = unit->location;
- offset_t target = target_rva_to_offset_.Convert(unit->target_rva);
- if (target == kInvalidOffset)
- continue;
- return Reference{location, target};
+ // |target| will not be dereferenced, so we don't worry about it
+ // exceeding |image_.size()| (in fact, there are valid cases where it
+ // does).
+ offset_t unsafe_target = target_rva_to_offset_.Convert(unit->target_rva);
+ if (unsafe_target < kOffsetBound)
+ return Reference{location, unsafe_target};
}
return base::nullopt;
}
diff --git a/disassembler.cc b/disassembler.cc
index 012e5ef..b6a2748 100644
--- a/disassembler.cc
+++ b/disassembler.cc
@@ -8,6 +8,12 @@
namespace zucchini {
+/******** EmptyReferenceReader ********/
+
+base::Optional<Reference> EmptyReferenceReader::GetNext() {
+ return base::nullopt;
+}
+
/******** ReferenceGroup ********/
std::unique_ptr<ReferenceReader> ReferenceGroup::GetReader(
diff --git a/disassembler.h b/disassembler.h
index 4f528ed..ab1dafb 100644
--- a/disassembler.h
+++ b/disassembler.h
@@ -17,6 +17,12 @@
namespace zucchini {
+// A vacuous ReferenceReader that produces no references.
+class EmptyReferenceReader : public ReferenceReader {
+ public:
+ base::Optional<Reference> GetNext() override;
+};
+
// Disassembler needs to be declared before ReferenceGroup because the latter
// contains member pointers based on the former, and we use a compiler flag,
// -fcomplete-member-pointers, which enforces that member pointer base types are
diff --git a/disassembler_elf.cc b/disassembler_elf.cc
new file mode 100644
index 0000000..75690e9
--- /dev/null
+++ b/disassembler_elf.cc
@@ -0,0 +1,424 @@
+// Copyright 2018 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "components/zucchini/disassembler_elf.h"
+
+#include <stddef.h>
+
+#include <algorithm>
+#include <utility>
+
+#include "base/logging.h"
+#include "base/numerics/safe_conversions.h"
+#include "components/zucchini/abs32_utils.h"
+#include "components/zucchini/algorithm.h"
+#include "components/zucchini/buffer_source.h"
+
+namespace zucchini {
+
+namespace {
+
+// Determines whether |section| is a reloc section.
+template <class Traits>
+bool IsRelocSection(const typename Traits::Elf_Shdr& section) {
+ if (section.sh_size == 0)
+ return false;
+ if (section.sh_type == elf::SHT_REL) {
+ // Also validate |section.sh_entsize|, which gets used later.
+ return section.sh_entsize == sizeof(typename Traits::Elf_Rel);
+ }
+ if (section.sh_type == elf::SHT_RELA)
+ return section.sh_entsize == sizeof(typename Traits::Elf_Rela);
+ return false;
+}
+
+// Determines whether |section| is a section with executable code.
+template <class Traits>
+bool IsExecSection(const typename Traits::Elf_Shdr& section) {
+ return (section.sh_flags & elf::SHF_EXECINSTR) != 0;
+}
+
+} // namespace
+
+/******** ELF32Traits ********/
+
+// static
+constexpr Bitness Elf32Traits::kBitness;
+constexpr elf::FileClass Elf32Traits::kIdentificationClass;
+
+/******** ELF32IntelTraits ********/
+
+// static
+constexpr ExecutableType Elf32IntelTraits::kExeType;
+const char Elf32IntelTraits::kExeTypeString[] = "ELF x86";
+constexpr elf::MachineArchitecture Elf32IntelTraits::kMachineValue;
+constexpr uint32_t Elf32IntelTraits::kRelType;
+
+/******** ELF64Traits ********/
+
+// static
+constexpr Bitness Elf64Traits::kBitness;
+constexpr elf::FileClass Elf64Traits::kIdentificationClass;
+
+/******** ELF64IntelTraits ********/
+
+// static
+constexpr ExecutableType Elf64IntelTraits::kExeType;
+const char Elf64IntelTraits::kExeTypeString[] = "ELF x64";
+constexpr elf::MachineArchitecture Elf64IntelTraits::kMachineValue;
+constexpr uint32_t Elf64IntelTraits::kRelType;
+
+/******** DisassemblerElf ********/
+
+// static.
+template <class Traits>
+bool DisassemblerElf<Traits>::QuickDetect(ConstBufferView image) {
+ BufferSource source(image);
+
+ // Do not consume the bytes for the magic value, as they are part of the
+ // header.
+ if (!source.CheckNextBytes({0x7F, 'E', 'L', 'F'}))
+ return false;
+
+ auto* header = source.GetPointer<typename Traits::Elf_Ehdr>();
+ if (!header)
+ return false;
+
+ if (header->e_ident[elf::EI_CLASS] != Traits::kIdentificationClass)
+ return false;
+
+ if (header->e_ident[elf::EI_DATA] != 1) // Only ELFDATA2LSB is supported.
+ return false;
+
+ if (header->e_type != elf::ET_EXEC && header->e_type != elf::ET_DYN)
+ return false;
+
+ if (header->e_version != 1 || header->e_ident[elf::EI_VERSION] != 1)
+ return false;
+
+ if (header->e_machine != supported_architecture())
+ return false;
+
+ if (header->e_shentsize != sizeof(typename Traits::Elf_Shdr))
+ return false;
+
+ return true;
+}
+
+template <class Traits>
+DisassemblerElf<Traits>::~DisassemblerElf() = default;
+
+template <class Traits>
+ExecutableType DisassemblerElf<Traits>::GetExeType() const {
+ return Traits::kExeType;
+}
+
+template <class Traits>
+std::string DisassemblerElf<Traits>::GetExeTypeString() const {
+ return Traits::kExeTypeString;
+}
+
+// |num_equivalence_iterations_| = 2 for reloc -> abs32.
+template <class Traits>
+DisassemblerElf<Traits>::DisassemblerElf() : Disassembler(2) {}
+
+template <class Traits>
+bool DisassemblerElf<Traits>::Parse(ConstBufferView image) {
+ image_ = image;
+ if (!ParseHeader())
+ return false;
+ ParseSections();
+ return true;
+}
+
+template <class Traits>
+std::unique_ptr<ReferenceReader> DisassemblerElf<Traits>::MakeReadRelocs(
+ offset_t lo,
+ offset_t hi) {
+ DCHECK_LE(lo, hi);
+ DCHECK_LE(hi, image_.size());
+
+ if (reloc_section_dims_.empty())
+ return std::make_unique<EmptyReferenceReader>();
+
+ return std::make_unique<RelocReaderElf>(
+ image_, Traits::kBitness, reloc_section_dims_,
+ supported_relocation_type(), lo, hi, translator_);
+}
+
+template <class Traits>
+std::unique_ptr<ReferenceWriter> DisassemblerElf<Traits>::MakeWriteRelocs(
+ MutableBufferView image) {
+ return std::make_unique<RelocWriterElf>(image, Traits::kBitness, translator_);
+}
+
+template <class Traits>
+std::unique_ptr<ReferenceReader> DisassemblerElf<Traits>::MakeReadAbs32(
+ offset_t lo,
+ offset_t hi) {
+ Abs32RvaExtractorWin32 abs_rva_extractor(image_, {Traits::kBitness, 0},
+ abs32_locations_, lo, hi);
+ return std::make_unique<Abs32ReaderWin32>(std::move(abs_rva_extractor),
+ translator_);
+}
+
+template <class Traits>
+std::unique_ptr<ReferenceWriter> DisassemblerElf<Traits>::MakeWriteAbs32(
+ MutableBufferView image) {
+ return std::make_unique<Abs32WriterWin32>(
+ image, AbsoluteAddress(Traits::kBitness, 0), translator_);
+}
+
+template <class Traits>
+bool DisassemblerElf<Traits>::ParseHeader() {
+ BufferSource source(image_);
+
+ // Ensures |header_| is valid later on.
+ if (!QuickDetect(image_))
+ return false;
+
+ header_ = source.GetPointer<typename Traits::Elf_Ehdr>();
+
+ sections_count_ = header_->e_shnum;
+ source = std::move(BufferSource(image_).Skip(header_->e_shoff));
+ sections_ = source.GetArray<typename Traits::Elf_Shdr>(sections_count_);
+ if (!sections_)
+ return false;
+ offset_t section_table_end =
+ base::checked_cast<offset_t>(source.begin() - image_.begin());
+
+ segments_count_ = header_->e_phnum;
+ source = std::move(BufferSource(image_).Skip(header_->e_phoff));
+ segments_ = source.GetArray<typename Traits::Elf_Phdr>(segments_count_);
+ if (!segments_)
+ return false;
+ offset_t segment_table_end =
+ base::checked_cast<offset_t>(source.begin() - image_.begin());
+
+ // Check string section -- even though we've stopped using them.
+ elf::Elf32_Half string_section_id = header_->e_shstrndx;
+ if (string_section_id >= sections_count_)
+ return false;
+ size_t section_names_size = sections_[string_section_id].sh_size;
+ if (section_names_size > 0) {
+ // If nonempty, then last byte of string section must be null.
+ const char* section_names = nullptr;
+ source = std::move(
+ BufferSource(image_).Skip(sections_[string_section_id].sh_offset));
+ section_names = source.GetArray<char>(section_names_size);
+ if (!section_names || section_names[section_names_size - 1] != '\0')
+ return false;
+ }
+
+ // Establish bound on encountered offsets.
+ offset_t offset_bound = std::max(section_table_end, segment_table_end);
+
+ // Visit each section, validate, and add address translation data to |units|.
+ std::vector<AddressTranslator::Unit> units;
+ units.reserve(sections_count_);
+
+ for (int i = 0; i < sections_count_; ++i) {
+ const typename Traits::Elf_Shdr* section = §ions_[i];
+
+ // Skip empty sections. These don't affect |offset_bound|, and don't
+ // contribute to RVA-offset mapping.
+ if (section->sh_size == 0)
+ continue;
+
+ // Be lax with RVAs: Assume they fit in int32_t, even for 64-bit. If
+ // assumption fails, simply skip the section with warning.
+ if (!RangeIsBounded(section->sh_addr, section->sh_size, kRvaBound) ||
+ !RangeIsBounded(section->sh_offset, section->sh_size, kOffsetBound)) {
+ LOG(WARNING) << "Section " << i << " does not fit in int32_t.";
+ continue;
+ }
+
+ // Extract dimensions to 32-bit integers to facilitate conversion. Range of
+ // values was ensured above when checking that the section is bounded.
+ uint32_t sh_size = base::checked_cast<uint32_t>(section->sh_size);
+ offset_t sh_offset = base::checked_cast<offset_t>(section->sh_offset);
+ rva_t sh_addr = base::checked_cast<rva_t>(section->sh_addr);
+
+ // Update |offset_bound|.
+ if (section->sh_type != elf::SHT_NOBITS) {
+ // Be strict with offsets: Any size overflow invalidates the file.
+ if (!image_.covers({sh_offset, sh_size}))
+ return false;
+
+ offset_t section_end = sh_offset + sh_size;
+ offset_bound = std::max(offset_bound, section_end);
+ }
+
+ // Compute mappings to translate between RVA and offset. As a heuristic,
+ // sections with RVA == 0 (i.e., |sh_addr == 0|) are ignored because these
+ // tend to be duplicates (which cause problems during lookup), and tend to
+ // be uninteresting.
+ if (section->sh_addr > 0) {
+ // Add |section| data for offset-RVA translation.
+ units.push_back({sh_offset, sh_size, sh_addr, sh_size});
+ }
+ }
+
+ // Initialize |translator_| for offset-RVA translations. Any inconsistency
+ // (e.g., 2 offsets correspond to the same RVA) would invalidate the ELF file.
+ if (translator_.Initialize(std::move(units)) != AddressTranslator::kSuccess)
+ return false;
+
+ // Visits |segments_| to get better estimate on |offset_bound|.
+ for (const typename Traits::Elf_Phdr* segment = segments_;
+ segment != segments_ + segments_count_; ++segment) {
+ if (!RangeIsBounded(segment->p_offset, segment->p_filesz, kOffsetBound))
+ return false;
+ offset_t segment_end = segment->p_offset + segment->p_filesz;
+ offset_bound = std::max(offset_bound, segment_end);
+ }
+
+ if (offset_bound > image_.size())
+ return false;
+ image_.shrink(offset_bound);
+
+ return true;
+}
+
+template <class Traits>
+void DisassemblerElf<Traits>::ExtractInterestingSectionHeaders() {
+ DCHECK(reloc_section_dims_.empty());
+ DCHECK(exec_headers_.empty());
+ for (elf::Elf32_Half i = 0; i < sections_count_; ++i) {
+ const typename Traits::Elf_Shdr* section = sections_ + i;
+ if (IsRelocSection<Traits>(*section))
+ reloc_section_dims_.emplace_back(*section);
+ else if (IsExecSection<Traits>(*section))
+ exec_headers_.push_back(section);
+ }
+ auto comp = [](const typename Traits::Elf_Shdr* a,
+ const typename Traits::Elf_Shdr* b) {
+ return a->sh_offset < b->sh_offset;
+ };
+ std::sort(reloc_section_dims_.begin(), reloc_section_dims_.end());
+ std::sort(exec_headers_.begin(), exec_headers_.end(), comp);
+}
+
+template <class Traits>
+void DisassemblerElf<Traits>::GetAbs32FromRelocSections() {
+ constexpr int kAbs32Width = 4;
+ DCHECK(abs32_locations_.empty());
+ auto relocs = MakeReadRelocs(0, offset_t(size()));
+ for (auto ref = relocs->GetNext(); ref; ref = relocs->GetNext()) {
+ // Reject null targets and targets outside |image_|. Note that here we
+ // assume abs32 targets are never "fake offsets".
+ if (ref->target > 0 && image_.covers({ref->target, kAbs32Width}))
+ abs32_locations_.push_back(ref->target);
+ }
+ abs32_locations_.shrink_to_fit();
+ std::sort(abs32_locations_.begin(), abs32_locations_.end());
+
+ // Abs32 reference bodies must not overlap. If found, simply remove them.
+ size_t num_removed =
+ RemoveOverlappingAbs32Locations(Traits::kBitness, &abs32_locations_);
+ if (num_removed) {
+ LOG(WARNING) << "Warning: Found and removed " << num_removed
+ << " abs32 locations with overlapping bodies.";
+ }
+}
+
+template <class Traits>
+void DisassemblerElf<Traits>::GetRel32FromCodeSections() {
+ for (const typename Traits::Elf_Shdr* section : exec_headers_)
+ ParseExecSection(*section);
+ PostProcessRel32();
+}
+
+template <class Traits>
+void DisassemblerElf<Traits>::ParseSections() {
+ ExtractInterestingSectionHeaders();
+ GetAbs32FromRelocSections();
+ GetRel32FromCodeSections();
+}
+
+/******** DisassemblerElfIntel ********/
+
+template <class Traits>
+DisassemblerElfIntel<Traits>::DisassemblerElfIntel() = default;
+
+template <class Traits>
+DisassemblerElfIntel<Traits>::~DisassemblerElfIntel() = default;
+
+template <class Traits>
+std::vector<ReferenceGroup> DisassemblerElfIntel<Traits>::MakeReferenceGroups()
+ const {
+ return {{ReferenceTypeTraits{4, TypeTag(kReloc), PoolTag(kReloc)},
+ &DisassemblerElfIntel<Traits>::MakeReadRelocs,
+ &DisassemblerElfIntel<Traits>::MakeWriteRelocs},
+ {ReferenceTypeTraits{4, TypeTag(kAbs32), PoolTag(kAbs32)},
+ &DisassemblerElfIntel<Traits>::MakeReadAbs32,
+ &DisassemblerElfIntel<Traits>::MakeWriteAbs32},
+ {ReferenceTypeTraits{4, TypeTag(kRel32), PoolTag(kRel32)},
+ &DisassemblerElfIntel<Traits>::MakeReadRel32,
+ &DisassemblerElfIntel<Traits>::MakeWriteRel32}};
+}
+
+template <class Traits>
+void DisassemblerElfIntel<Traits>::ParseExecSection(
+ const typename Traits::Elf_Shdr& section) {
+ ConstBufferView& image_ = this->image_;
+ auto& abs32_locations_ = this->abs32_locations_;
+
+ std::ptrdiff_t from_offset_to_rva = section.sh_addr - section.sh_offset;
+ rva_t start_rva = section.sh_addr;
+ rva_t end_rva = start_rva + section.sh_size;
+
+ AddressTranslator::RvaToOffsetCache target_rva_checker(this->translator_);
+
+ ConstBufferView region(image_.begin() + section.sh_offset, section.sh_size);
+ Abs32GapFinder gap_finder(image_, region, abs32_locations_, 4);
+ std::unique_ptr<Rel32FinderIntel> finder =
+ std::make_unique<typename Traits::Rel32FinderUse>(image_);
+ for (auto gap = gap_finder.GetNext(); gap.has_value();
+ gap = gap_finder.GetNext()) {
+ finder->Reset(gap.value());
+ for (auto rel32 = finder->GetNext(); rel32.has_value();
+ rel32 = finder->GetNext()) {
+ offset_t rel32_offset = offset_t(rel32->location - image_.begin());
+ rva_t rel32_rva = rva_t(rel32_offset + from_offset_to_rva);
+ rva_t target_rva = rel32_rva + 4 + image_.read<uint32_t>(rel32_offset);
+ if (target_rva_checker.IsValid(target_rva) &&
+ (rel32->can_point_outside_section ||
+ (start_rva <= target_rva && target_rva < end_rva))) {
+ finder->Accept();
+ rel32_locations_.push_back(rel32_offset);
+ }
+ }
+ }
+}
+
+template <class Traits>
+void DisassemblerElfIntel<Traits>::PostProcessRel32() {
+ rel32_locations_.shrink_to_fit();
+ std::sort(rel32_locations_.begin(), rel32_locations_.end());
+}
+
+template <class Traits>
+std::unique_ptr<ReferenceReader> DisassemblerElfIntel<Traits>::MakeReadRel32(
+ offset_t lo,
+ offset_t hi) {
+ return std::make_unique<Rel32ReaderX86>(this->image_, lo, hi,
+ &rel32_locations_, this->translator_);
+}
+
+template <class Traits>
+std::unique_ptr<ReferenceWriter> DisassemblerElfIntel<Traits>::MakeWriteRel32(
+ MutableBufferView image) {
+ return std::make_unique<Rel32WriterX86>(image, this->translator_);
+}
+
+// Explicit instantiation for supported classes.
+template class DisassemblerElfIntel<Elf32IntelTraits>;
+template class DisassemblerElfIntel<Elf64IntelTraits>;
+template bool DisassemblerElf<Elf32IntelTraits>::QuickDetect(
+ ConstBufferView image);
+template bool DisassemblerElf<Elf64IntelTraits>::QuickDetect(
+ ConstBufferView image);
+
+} // namespace zucchini
diff --git a/disassembler_elf.h b/disassembler_elf.h
new file mode 100644
index 0000000..a5b0a72
--- /dev/null
+++ b/disassembler_elf.h
@@ -0,0 +1,189 @@
+// Copyright 2018 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef COMPONENTS_ZUCCHINI_DISASSEMBLER_ELF_H_
+#define COMPONENTS_ZUCCHINI_DISASSEMBLER_ELF_H_
+
+#include <stdint.h>
+
+#include <memory>
+#include <string>
+#include <vector>
+
+#include "base/macros.h"
+#include "components/zucchini/address_translator.h"
+#include "components/zucchini/buffer_view.h"
+#include "components/zucchini/disassembler.h"
+#include "components/zucchini/image_utils.h"
+#include "components/zucchini/rel32_finder.h"
+#include "components/zucchini/rel32_utils.h"
+#include "components/zucchini/reloc_elf.h"
+#include "components/zucchini/type_elf.h"
+
+namespace zucchini {
+
+struct Elf32Traits {
+ static constexpr Bitness kBitness = kBit32;
+ static constexpr elf::FileClass kIdentificationClass = elf::ELFCLASS32;
+ using Elf_Shdr = elf::Elf32_Shdr;
+ using Elf_Phdr = elf::Elf32_Phdr;
+ using Elf_Ehdr = elf::Elf32_Ehdr;
+ using Elf_Rel = elf::Elf32_Rel;
+ using Elf_Rela = elf::Elf32_Rela;
+};
+
+// Architecture-specific definitions.
+struct Elf32IntelTraits : public Elf32Traits {
+ static constexpr ExecutableType kExeType = kExeTypeElfX86;
+ static const char kExeTypeString[];
+ static constexpr elf::MachineArchitecture kMachineValue = elf::EM_386;
+ static constexpr uint32_t kRelType = elf::R_386_RELATIVE;
+ using Rel32FinderUse = Rel32FinderX86;
+};
+
+struct Elf64Traits {
+ static constexpr Bitness kBitness = kBit64;
+ static constexpr elf::FileClass kIdentificationClass = elf::ELFCLASS64;
+ using Elf_Shdr = elf::Elf64_Shdr;
+ using Elf_Phdr = elf::Elf64_Phdr;
+ using Elf_Ehdr = elf::Elf64_Ehdr;
+ using Elf_Rel = elf::Elf64_Rel;
+ using Elf_Rela = elf::Elf64_Rela;
+};
+
+// Architecture-specific definitions.
+struct Elf64IntelTraits : public Elf64Traits {
+ static constexpr ExecutableType kExeType = kExeTypeElfX64;
+ static const char kExeTypeString[];
+ static constexpr elf::MachineArchitecture kMachineValue = elf::EM_X86_64;
+ static constexpr uint32_t kRelType = elf::R_X86_64_RELATIVE;
+ using Rel32FinderUse = Rel32FinderX64;
+};
+
+// Disassembler for ELF.
+template <class Traits>
+class DisassemblerElf : public Disassembler {
+ public:
+ using HeaderVector = std::vector<const typename Traits::Elf_Shdr*>;
+
+ // Applies quick checks to determine whether |image| *may* point to the start
+ // of an executable. Returns true iff the check passes.
+ static bool QuickDetect(ConstBufferView image);
+
+ ~DisassemblerElf() override;
+
+ // Disassembler:
+ ExecutableType GetExeType() const override;
+ std::string GetExeTypeString() const override;
+ std::vector<ReferenceGroup> MakeReferenceGroups() const override = 0;
+
+ // Find/Receive functions that are common among different architectures.
+ std::unique_ptr<ReferenceReader> MakeReadRelocs(offset_t lo, offset_t hi);
+ std::unique_ptr<ReferenceWriter> MakeWriteRelocs(MutableBufferView image);
+ std::unique_ptr<ReferenceReader> MakeReadAbs32(offset_t lo, offset_t hi);
+ std::unique_ptr<ReferenceWriter> MakeWriteAbs32(MutableBufferView image);
+
+ const AddressTranslator& translator() const { return translator_; }
+
+ protected:
+ friend Disassembler;
+
+ DisassemblerElf();
+
+ bool Parse(ConstBufferView image) override;
+
+ // Returns the supported Elf_Ehdr::e_machine enum.
+ static constexpr elf::MachineArchitecture supported_architecture() {
+ return Traits::kMachineValue;
+ }
+
+ // Returns the type to look for in the reloc section.
+ static constexpr uint32_t supported_relocation_type() {
+ return Traits::kRelType;
+ }
+
+ // Performs architecture-specific parsing of an executable section, to extract
+ // rel32 references.
+ virtual void ParseExecSection(const typename Traits::Elf_Shdr& section) = 0;
+
+ // Processes rel32 data after they are extracted from executable sections.
+ virtual void PostProcessRel32() = 0;
+
+ // The parsing routines below return true on success, and false on failure.
+
+ // Parses ELF header and section headers, and performs basic validation.
+ bool ParseHeader();
+
+ // Extracts and stores section headers that we need.
+ void ExtractInterestingSectionHeaders();
+
+ // Parsing functions that extract references from various sections.
+ void GetAbs32FromRelocSections();
+ void GetRel32FromCodeSections();
+ void ParseSections();
+
+ // Main ELF header.
+ const typename Traits::Elf_Ehdr* header_ = nullptr;
+
+ // Section header table, ordered by section id.
+ elf::Elf32_Half sections_count_ = 0;
+ const typename Traits::Elf_Shdr* sections_ = nullptr;
+
+ // Program header table.
+ elf::Elf32_Half segments_count_ = 0;
+ const typename Traits::Elf_Phdr* segments_ = nullptr;
+
+ // Translator between offsets and RVAs.
+ AddressTranslator translator_;
+
+ // Identity translator for abs32 translation.
+ AddressTranslator identity_translator_;
+
+ // Extracted relocation section dimensions data, sorted by file offsets.
+ std::vector<SectionDimensionsElf> reloc_section_dims_;
+
+ // Headers of executable sections, sorted by file offsets of the data each
+ // header points to.
+ std::vector<const typename Traits::Elf_Shdr*> exec_headers_;
+
+ // Sorted file offsets of abs32 locations.
+ std::vector<offset_t> abs32_locations_;
+
+ private:
+ DISALLOW_COPY_AND_ASSIGN(DisassemblerElf);
+};
+
+// Disassembler for ELF with Intel architectures.
+template <class Traits>
+class DisassemblerElfIntel : public DisassemblerElf<Traits> {
+ public:
+ enum ReferenceType : uint8_t { kReloc, kAbs32, kRel32, kTypeCount };
+
+ DisassemblerElfIntel();
+ ~DisassemblerElfIntel() override;
+
+ // Disassembler:
+ std::vector<ReferenceGroup> MakeReferenceGroups() const override;
+
+ // DisassemblerElf:
+ void ParseExecSection(const typename Traits::Elf_Shdr& section) override;
+ void PostProcessRel32() override;
+
+ // Specialized Find/Receive functions.
+ std::unique_ptr<ReferenceReader> MakeReadRel32(offset_t lo, offset_t hi);
+ std::unique_ptr<ReferenceWriter> MakeWriteRel32(MutableBufferView image);
+
+ private:
+ // Sorted file offsets of rel32 locations.
+ std::vector<offset_t> rel32_locations_;
+
+ DISALLOW_COPY_AND_ASSIGN(DisassemblerElfIntel);
+};
+
+using DisassemblerElfX86 = DisassemblerElfIntel<Elf32IntelTraits>;
+using DisassemblerElfX64 = DisassemblerElfIntel<Elf64IntelTraits>;
+
+} // namespace zucchini
+
+#endif // COMPONENTS_ZUCCHINI_DISASSEMBLER_ELF_H_
diff --git a/disassembler_elf_unittest.cc b/disassembler_elf_unittest.cc
new file mode 100644
index 0000000..2db8635
--- /dev/null
+++ b/disassembler_elf_unittest.cc
@@ -0,0 +1,94 @@
+// Copyright 2018 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "components/zucchini/disassembler_elf.h"
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include <algorithm>
+#include <random>
+#include <vector>
+
+#include "components/zucchini/test_utils.h"
+#include "components/zucchini/type_elf.h"
+#include "testing/gtest/include/gtest/gtest.h"
+
+namespace zucchini {
+
+TEST(DisassemblerElfTest, QuickDetect) {
+ std::vector<uint8_t> image_data;
+ ConstBufferView image;
+
+ // Empty.
+ EXPECT_FALSE(DisassemblerElfX86::QuickDetect(image));
+ EXPECT_FALSE(DisassemblerElfX64::QuickDetect(image));
+
+ // Unrelated.
+ image_data = ParseHexString("DE AD");
+ image = {image_data.data(), image_data.size()};
+ EXPECT_FALSE(DisassemblerElfX86::QuickDetect(image));
+ EXPECT_FALSE(DisassemblerElfX64::QuickDetect(image));
+
+ // Only Magic.
+ image_data = ParseHexString("7F 45 4C 46");
+ image = {image_data.data(), image_data.size()};
+ EXPECT_FALSE(DisassemblerElfX86::QuickDetect(image));
+ EXPECT_FALSE(DisassemblerElfX64::QuickDetect(image));
+
+ // Only identification.
+ image_data =
+ ParseHexString("7F 45 4C 46 01 01 01 00 00 00 00 00 00 00 00 00");
+ image = {image_data.data(), image_data.size()};
+ EXPECT_FALSE(DisassemblerElfX86::QuickDetect(image));
+ EXPECT_FALSE(DisassemblerElfX64::QuickDetect(image));
+
+ // Large enough, filled with zeros.
+ image_data.assign(sizeof(elf::Elf32_Ehdr), 0);
+ image = {image_data.data(), image_data.size()};
+ EXPECT_FALSE(DisassemblerElfX86::QuickDetect(image));
+ EXPECT_FALSE(DisassemblerElfX64::QuickDetect(image));
+
+ // Random.
+ std::random_device rd;
+ std::mt19937 gen{rd()};
+ std::generate(image_data.begin(), image_data.end(), gen);
+ image = {image_data.data(), image_data.size()};
+ EXPECT_FALSE(DisassemblerElfX86::QuickDetect(image));
+ EXPECT_FALSE(DisassemblerElfX64::QuickDetect(image));
+
+ // Typical x86 elf header.
+ {
+ elf::Elf32_Ehdr header = {};
+ auto e_ident =
+ ParseHexString("7F 45 4C 46 01 01 01 00 00 00 00 00 00 00 00 00");
+ std::copy(e_ident.begin(), e_ident.end(), header.e_ident);
+ header.e_type = elf::ET_EXEC;
+ header.e_machine = elf::EM_386;
+ header.e_version = 1;
+ header.e_shentsize = sizeof(elf::Elf32_Shdr);
+ ConstBufferView image(reinterpret_cast<const uint8_t*>(&header),
+ sizeof(header));
+ EXPECT_TRUE(DisassemblerElfX86::QuickDetect(image));
+ EXPECT_FALSE(DisassemblerElfX64::QuickDetect(image));
+ }
+
+ // Typical x64 elf header.
+ {
+ elf::Elf64_Ehdr header = {};
+ auto e_ident =
+ ParseHexString("7F 45 4C 46 02 01 01 00 00 00 00 00 00 00 00 00");
+ std::copy(e_ident.begin(), e_ident.end(), header.e_ident);
+ header.e_type = elf::ET_EXEC;
+ header.e_machine = elf::EM_X86_64;
+ header.e_version = 1;
+ header.e_shentsize = sizeof(elf::Elf64_Shdr);
+ ConstBufferView image(reinterpret_cast<const uint8_t*>(&header),
+ sizeof(header));
+ EXPECT_FALSE(DisassemblerElfX86::QuickDetect(image));
+ EXPECT_TRUE(DisassemblerElfX64::QuickDetect(image));
+ }
+}
+
+} // namespace zucchini
diff --git a/element_detection.cc b/element_detection.cc
index 6b31f61..fa5adf4 100644
--- a/element_detection.cc
+++ b/element_detection.cc
@@ -19,6 +19,10 @@
#include "components/zucchini/disassembler_win32.h"
#endif // BUILDFLAG(ENABLE_WIN)
+#if BUILDFLAG(ENABLE_ELF)
+#include "components/zucchini/disassembler_elf.h"
+#endif // BUILDFLAG(ENABLE_ELF)
+
#if BUILDFLAG(ENABLE_ZTF)
#include "components/zucchini/disassembler_ztf.h"
#endif // BUILDFLAG(ENABLE_ZTF)
@@ -50,6 +54,20 @@
}
#endif // BUILDFLAG(ENABLE_WIN)
+#if BUILDFLAG(ENABLE_ELF)
+ if (DisassemblerElfX86::QuickDetect(image)) {
+ auto disasm = Disassembler::Make<DisassemblerElfX86>(image);
+ if (disasm && disasm->size() >= kMinProgramSize)
+ return disasm;
+ }
+
+ if (DisassemblerElfX64::QuickDetect(image)) {
+ auto disasm = Disassembler::Make<DisassemblerElfX64>(image);
+ if (disasm && disasm->size() >= kMinProgramSize)
+ return disasm;
+ }
+#endif // BUILDFLAG(ENABLE_ELF)
+
#if BUILDFLAG(ENABLE_DEX)
if (DisassemblerDex::QuickDetect(image)) {
auto disasm = Disassembler::Make<DisassemblerDex>(image);
@@ -79,6 +97,12 @@
case kExeTypeWin32X64:
return Disassembler::Make<DisassemblerWin32X64>(image);
#endif // BUILDFLAG(ENABLE_WIN)
+#if BUILDFLAG(ENABLE_ELF)
+ case kExeTypeElfX86:
+ return Disassembler::Make<DisassemblerElfX86>(image);
+ case kExeTypeElfX64:
+ return Disassembler::Make<DisassemblerElfX64>(image);
+#endif // BUILDFLAG(ENABLE_ELF)
#if BUILDFLAG(ENABLE_DEX)
case kExeTypeDex:
return Disassembler::Make<DisassemblerDex>(image);
diff --git a/type_elf.h b/type_elf.h
index 99dcba2..2bbe6dc 100644
--- a/type_elf.h
+++ b/type_elf.h
@@ -67,8 +67,30 @@
Elf64_Half e_shstrndx;
};
+// Identification Indexes in header->e_ident.
+enum IdentificationIndex {
+ EI_MAG0 = 0, // File identification.
+ EI_MAG1 = 1, // File identification.
+ EI_MAG2 = 2, // File identification.
+ EI_MAG3 = 3, // File identification.
+ EI_CLASS = 4, // File class.
+ EI_DATA = 5, // Data encoding.
+ EI_VERSION = 6, // File version.
+ EI_OSABI = 7, // Operating system/ABI identification.
+ EI_ABIVERSION = 8, // ABI version.
+ EI_PAD = 9, // Start of padding bytes.
+ EI_NIDENT = 16 // Size of e_ident[].
+};
+
+// Values for header->e_ident[EI_CLASS].
+enum FileClass {
+ ELFCLASSNONE = 0, // Invalid class.
+ ELFCLASS32 = 1, // 32-bit objects.
+ ELFCLASS64 = 2 // 64-bit objects.
+};
+
// Values for header->e_type.
-enum e_type_values {
+enum FileType {
ET_NONE = 0, // No file type
ET_REL = 1, // Relocatable file
ET_EXEC = 2, // Executable file
@@ -79,7 +101,7 @@
};
// Values for header->e_machine.
-enum e_machine_values {
+enum MachineArchitecture {
EM_NONE = 0, // No machine.
EM_386 = 3, // Intel Architecture.
EM_ARM = 40, // ARM Architecture.