/** Reads content of a section and returns it as a container. The returned container will always have exactly @p size byte.
* If @p size bytes are not available in this section at the specified offset then the container will be zero padded. */
SgUnsignedCharList
SgAsmGenericSection::read_content_local_ucl(rose_addr_t rel_offset, rose_addr_t size)
{
SgUnsignedCharList retval;
unsigned char *buf = new unsigned char[size];
read_content_local(rel_offset, buf, size, false); /*zero pads; never throws*/
for (size_t i=0; i<size; i++)
retval.push_back(buf[i]);
delete[] buf;
return retval;
}
/** Initializes this ELF Symbol Section by parsing a file. */
SgAsmElfSymbolSection *
SgAsmElfSymbolSection::parse()
{
SgAsmElfSection::parse();
SgAsmElfFileHeader *fhdr = get_elf_header();
ROSE_ASSERT(fhdr!=NULL);
SgAsmElfSectionTableEntry *shdr = get_section_entry();
ROSE_ASSERT(shdr!=NULL);
SgAsmElfStringSection *strsec = dynamic_cast<SgAsmElfStringSection*>(get_linked_section());
ROSE_ASSERT(strsec!=NULL);
size_t entry_size, struct_size, extra_size, nentries;
calculate_sizes(&entry_size, &struct_size, &extra_size, &nentries);
ROSE_ASSERT(entry_size==shdr->get_sh_entsize());
/* Parse each entry */
for (size_t i=0; i<nentries; i++) {
SgAsmElfSymbol *entry=0;
if (4==fhdr->get_word_size()) {
entry = new SgAsmElfSymbol(this); /*adds symbol to this symbol table*/
SgAsmElfSymbol::Elf32SymbolEntry_disk disk;
read_content_local(i*entry_size, &disk, struct_size);
entry->parse(fhdr->get_sex(), &disk);
} else if (8==fhdr->get_word_size()) {
entry = new SgAsmElfSymbol(this); /*adds symbol to this symbol table*/
SgAsmElfSymbol::Elf64SymbolEntry_disk disk;
read_content_local(i*entry_size, &disk, struct_size);
entry->parse(fhdr->get_sex(), &disk);
} else {
throw FormatError("unsupported ELF word size");
}
if (extra_size>0)
entry->get_extra() = read_content_local_ucl(i*entry_size+struct_size, extra_size);
}
return this;
}
/** Extract an unsigned LEB128 value and adjust @p rel_offset according to how many bytes it occupied. If @p strict is set
* (the default) and the end of the section is reached then throw an SgAsmExecutableFileFormat::ShortRead exception. Upon
* return, the @p rel_offset will be adjusted to point to the first byte after the LEB128 value. */
uint64_t
SgAsmGenericSection::read_content_local_uleb128(rose_addr_t *rel_offset, bool strict)
{
int shift=0;
uint64_t retval=0;
while (1) {
unsigned char byte;
read_content_local(*rel_offset, &byte, 1, strict);
*rel_offset += 1;
ROSE_ASSERT(shift<64);
retval |= (byte & 0x7f) << shift;
shift += 7;
if (0==(byte & 0x80))
break;
}
return retval;
}
/* Parser */
SgAsmPESectionTable*
SgAsmPESectionTable::parse()
{
SgAsmGenericSection::parse();
SgAsmPEFileHeader *fhdr = dynamic_cast<SgAsmPEFileHeader*>(get_header());
ROSE_ASSERT(fhdr!=NULL);
/* Parse section table and construct section objects, but do not parse the sections yet. */
SgAsmGenericSectionPtrList pending;
const size_t entsize = sizeof(SgAsmPESectionTableEntry::PESectionTableEntry_disk);
for (size_t i=0; i<fhdr->get_e_nsections(); i++) {
SgAsmPESectionTableEntry::PESectionTableEntry_disk disk;
if (entsize!=read_content_local(i * entsize, &disk, entsize, false))
fprintf(stderr, "SgAsmPESectionTable::parse: warning: section table entry %" PRIuPTR " at file offset 0x%08"PRIx64
" extends beyond end of defined section table.\n",
i, get_offset()+i*entsize);
SgAsmPESectionTableEntry *entry = new SgAsmPESectionTableEntry(&disk);
SgAsmPESection *section = NULL;
if (entry->get_name() == ".idata") {
section = new SgAsmPEImportSection(fhdr);
} else {
section = new SgAsmPESection(fhdr);
}
section->init_from_section_table(entry, i+1);
pending.push_back(section);
}
/* Build the memory mapping like the real loader would do. This is the same code used by
* SgAsmExecutableFileFormat::parseBinaryFormat() except we're doing it here early because we need it in the rest of the
* PE parser. */
ROSE_ASSERT(NULL==fhdr->get_loader_map());
BinaryLoader *loader = BinaryLoader::lookup(fhdr); /*no need to clone; we're not changing any settings*/
ROSE_ASSERT(loader!=NULL);
MemoryMap *loader_map = new MemoryMap;
loader->remap(loader_map, fhdr);
fhdr->set_loader_map(loader_map);
/* Parse each section after the loader map is created */
for (size_t i=0; i<pending.size(); i++)
pending[i]->parse();
return this;
}
/* Adds the RVA/Size pairs to the end of the PE file header */
void
SgAsmPEFileHeader::add_rvasize_pairs()
{
rose_addr_t pairs_offset = get_size();
rose_addr_t pairs_size = p_e_num_rvasize_pairs * sizeof(SgAsmPERVASizePair::RVASizePair_disk);
SgAsmPERVASizePair::RVASizePair_disk pairs_disk;
ROSE_ASSERT(p_rvasize_pairs != NULL);
ROSE_ASSERT(p_rvasize_pairs->get_pairs().size()==0);
p_rvasize_pairs->set_isModified(true);
extend(pairs_size);
for (size_t i = 0; i < p_e_num_rvasize_pairs; i++, pairs_offset += sizeof pairs_disk) {
if (sizeof(pairs_disk)!=read_content_local(pairs_offset, &pairs_disk, sizeof pairs_disk, false))
fprintf(stderr, "SgAsmPEFileHeader::add_rvasize_pairs: warning: RVA/Size pair %zu at file offset 0x%08"PRIx64
" extends beyond the end of file (assuming 0/0)\n", i, get_offset()+pairs_offset);
p_rvasize_pairs->get_pairs().push_back(new SgAsmPERVASizePair(p_rvasize_pairs, &pairs_disk));
}
}
/** Reads a string from the file. The string begins at the specified file offset relative to the start of this section and
* continues until the first NUL byte or the end of section is reached. However, if @p strict is set (the default) and we
* reach the end-of-section then an SgAsmExecutableFileFormat::ShortRead exception is thrown. */
std::string
SgAsmGenericSection::read_content_local_str(rose_addr_t rel_offset, bool strict)
{
static char *buf=NULL;
static size_t nalloc=0;
size_t nused=0;
while (1) {
if (nused >= nalloc) {
nalloc = std::max((size_t)32, 2*nalloc);
buf = (char*)realloc(buf, nalloc);
ROSE_ASSERT(buf!=NULL);
}
unsigned char byte;
read_content_local(rel_offset+nused, &byte, 1, strict);
if (!byte)
return std::string(buf, nused);
buf[nused++] = byte;
}
}
/** Initialize the header with information parsed from the file and construct and parse everything that's reachable from the
* header. The PE File Header should have been constructed such that SgAsmPEFileHeader::ctor() was called. */
SgAsmPEFileHeader*
SgAsmPEFileHeader::parse()
{
SgAsmGenericHeader::parse();
/* Read header, zero padding if the file isn't large enough */
PEFileHeader_disk fh;
if (sizeof(fh)>get_size())
extend(sizeof(fh)-get_size());
if (sizeof(fh)!=read_content_local(0, &fh, sizeof fh, false))
fprintf(stderr, "SgAsmPEFileHeader::parse: warning: short read of PE header at byte 0x%08"PRIx64"\n", get_offset());
/* Check magic number before getting too far */
if (fh.e_magic[0]!='P' || fh.e_magic[1]!='E' || fh.e_magic[2]!='\0' || fh.e_magic[3]!='\0')
throw FormatError("Bad PE magic number");
/* Decode COFF file header */
p_e_cpu_type = le_to_host(fh.e_cpu_type);
p_e_nsections = le_to_host(fh.e_nsections);
p_e_time = le_to_host(fh.e_time);
p_e_coff_symtab = le_to_host(fh.e_coff_symtab);
p_e_coff_nsyms = le_to_host(fh.e_coff_nsyms);
p_e_nt_hdr_size = le_to_host(fh.e_nt_hdr_size);
p_e_flags = le_to_host(fh.e_flags);
/* Read the "Optional Header" (optional in the sense that not all files have one, but required for an executable), the
* size of which is stored in the e_nt_hdr_size of the main PE file header. According to
* http://www.phreedom.org/solar/code/tinype the Windows loader honors the e_nt_hdr_size even when set to smaller than the
* smallest possible documented size of the optional header. Also it's possible for the optional header to extend beyond
* the end of the file, in which case that part should be read as zero. */
PE32OptHeader_disk oh32;
rose_addr_t need32 = sizeof(PEFileHeader_disk) + std::min(p_e_nt_hdr_size, (rose_addr_t)(sizeof oh32));
if (need32>get_size())
extend(need32-get_size());
if (sizeof(oh32)!=read_content_local(sizeof fh, &oh32, sizeof oh32, false))
fprintf(stderr, "SgAsmPEFileHeader::parse: warning: short read of PE Optional Header at byte 0x%08"PRIx64"\n",
get_offset() + sizeof(fh));
p_e_opt_magic = le_to_host(oh32.e_opt_magic);
/* File format changes from ctor() */
p_exec_format->set_purpose(p_e_flags & HF_PROGRAM ? PURPOSE_EXECUTABLE : PURPOSE_LIBRARY);
p_exec_format->set_word_size(0x010b==p_e_opt_magic? 4 : 8);
/* Decode the optional header. */
rose_addr_t entry_rva;
if (4==p_exec_format->get_word_size()) {
p_e_lmajor = le_to_host(oh32.e_lmajor);
p_e_lminor = le_to_host(oh32.e_lminor);
p_e_code_size = le_to_host(oh32.e_code_size);
p_e_data_size = le_to_host(oh32.e_data_size);
p_e_bss_size = le_to_host(oh32.e_bss_size);
entry_rva = le_to_host(oh32.e_entrypoint_rva);
p_e_code_rva = le_to_host(oh32.e_code_rva);
p_e_data_rva = le_to_host(oh32.e_data_rva);
p_base_va = le_to_host(oh32.e_image_base);
p_e_section_align = le_to_host(oh32.e_section_align);
p_e_file_align = le_to_host(oh32.e_file_align);
p_e_os_major = le_to_host(oh32.e_os_major);
p_e_os_minor = le_to_host(oh32.e_os_minor);
p_e_user_major = le_to_host(oh32.e_user_major);
p_e_user_minor = le_to_host(oh32.e_user_minor);
p_e_subsys_major = le_to_host(oh32.e_subsys_major);
p_e_subsys_minor = le_to_host(oh32.e_subsys_minor);
p_e_reserved9 = le_to_host(oh32.e_reserved9);
p_e_image_size = le_to_host(oh32.e_image_size);
p_e_header_size = le_to_host(oh32.e_header_size);
p_e_file_checksum = le_to_host(oh32.e_file_checksum);
p_e_subsystem = le_to_host(oh32.e_subsystem);
p_e_dll_flags = le_to_host(oh32.e_dll_flags);
p_e_stack_reserve_size = le_to_host(oh32.e_stack_reserve_size);
p_e_stack_commit_size = le_to_host(oh32.e_stack_commit_size);
p_e_heap_reserve_size = le_to_host(oh32.e_heap_reserve_size);
p_e_heap_commit_size = le_to_host(oh32.e_heap_commit_size);
p_e_loader_flags = le_to_host(oh32.e_loader_flags);
p_e_num_rvasize_pairs = le_to_host(oh32.e_num_rvasize_pairs);
} else if (8==p_exec_format->get_word_size()) {
/* We guessed wrong. This is a 64-bit header, not 32-bit. */
PE64OptHeader_disk oh64;
rose_addr_t need64 = sizeof(PEFileHeader_disk) + std::min(p_e_nt_hdr_size, (rose_addr_t)(sizeof oh64));
if (need64>get_size())
extend(need64-get_size());
if (sizeof(oh64)!=read_content_local(sizeof fh, &oh64, sizeof oh64))
fprintf(stderr, "SgAsmPEFileHeader::parse: warning: short read of PE Optional Header at byte 0x%08"PRIx64"\n",
get_offset() + sizeof(fh));
p_e_lmajor = le_to_host(oh64.e_lmajor);
p_e_lminor = le_to_host(oh64.e_lminor);
p_e_code_size = le_to_host(oh64.e_code_size);
p_e_data_size = le_to_host(oh64.e_data_size);
p_e_bss_size = le_to_host(oh64.e_bss_size);
entry_rva = le_to_host(oh64.e_entrypoint_rva);
p_e_code_rva = le_to_host(oh64.e_code_rva);
// p_e_data_rva = le_to_host(oh.e_data_rva); /* not in PE32+ */
p_base_va = le_to_host(oh64.e_image_base);
p_e_section_align = le_to_host(oh64.e_section_align);
p_e_file_align = le_to_host(oh64.e_file_align);
p_e_os_major = le_to_host(oh64.e_os_major);
p_e_os_minor = le_to_host(oh64.e_os_minor);
p_e_user_major = le_to_host(oh64.e_user_major);
p_e_user_minor = le_to_host(oh64.e_user_minor);
p_e_subsys_major = le_to_host(oh64.e_subsys_major);
p_e_subsys_minor = le_to_host(oh64.e_subsys_minor);
p_e_reserved9 = le_to_host(oh64.e_reserved9);
p_e_image_size = le_to_host(oh64.e_image_size);
p_e_header_size = le_to_host(oh64.e_header_size);
p_e_file_checksum = le_to_host(oh64.e_file_checksum);
p_e_subsystem = le_to_host(oh64.e_subsystem);
p_e_dll_flags = le_to_host(oh64.e_dll_flags);
p_e_stack_reserve_size = le_to_host(oh64.e_stack_reserve_size);
p_e_stack_commit_size = le_to_host(oh64.e_stack_commit_size);
p_e_heap_reserve_size = le_to_host(oh64.e_heap_reserve_size);
p_e_heap_commit_size = le_to_host(oh64.e_heap_commit_size);
p_e_loader_flags = le_to_host(oh64.e_loader_flags);
p_e_num_rvasize_pairs = le_to_host(oh64.e_num_rvasize_pairs);
} else {
throw FormatError("unrecognized Windows PE optional header magic number");
}
/* Magic number */
p_magic.clear();
for (size_t i = 0; i < sizeof(fh.e_magic); ++i)
p_magic.push_back(fh.e_magic[i]);
/* File format */
ROSE_ASSERT(p_e_lmajor <= 0xffff && p_e_lminor <= 0xffff);
p_exec_format->set_version((p_e_lmajor << 16) | p_e_lminor);
p_exec_format->set_is_current_version(true); /*FIXME*/
/* Target architecture */
switch (p_e_cpu_type) {
case 0x0000:
set_isa(ISA_UNSPECIFIED);
break;
case 0x014c:
set_isa(ISA_IA32_386);
break;
case 0x014d:
set_isa(ISA_IA32_486);
break;
case 0x014e:
set_isa(ISA_IA32_Pentium);
break;
case 0x0162:
set_isa(ISA_MIPS_MarkI); /* R2000, R3000 */
break;
case 0x0163:
set_isa(ISA_MIPS_MarkII); /* R6000 */
break;
case 0x0166:
set_isa(ISA_MIPS_MarkIII); /* R4000 */
break;
case 0x01a2: /*Hitachi SH3*/
case 0x01a3: /*Hitachi SH3 with FPU*/
case 0x01a6: /*Hitachi SH4*/
case 0x01a8: /*Hitachi SH5*/
set_isa(ISA_Hitachi_SH);
break;
case 0x01c0:
set_isa(ISA_ARM_Family);
break;
case 0x01d3:
set_isa(ISA_Matsushita_AM33);
break;
case 0x01f0: /*w/o FPU*/
case 0x01f1: /*with FPU*/
set_isa(ISA_PowerPC);
break;
case 0x0200:
set_isa(ISA_IA64_Family);
break;
case 0x0266:
set_isa(ISA_MIPS_16);
break;
case 0x0366:
set_isa(ISA_MIPS_FPU);
break;
case 0x0466:
set_isa(ISA_MIPS_16FPU);
break;
case 0x0ebc:
set_isa(ISA_EFI_ByteCode);
break;
case 0x8664:
set_isa(ISA_X8664_Family);
break;
case 0x9041:
set_isa(ISA_Mitsubishi_M32R);
break;
default:
fprintf(stderr, "SgAsmPEFileHeader::parse: warning: unrecognized e_cputype = 0x%x (%u)\n", p_e_cpu_type, p_e_cpu_type);
set_isa(ISA_OTHER);
break;
}
/* The NT loader normally maps this file header at the header's base virtual address. */
set_mapped_preferred_rva(0);
set_mapped_actual_va(0); /* will be assigned by BinaryLoader */
set_mapped_size(p_e_header_size);
set_mapped_alignment(0);
set_mapped_rperm(true);
set_mapped_wperm(false);
set_mapped_xperm(false);
/* Entry point. We will eventually bind the entry point to a particular section (in SgAsmPEFileHeader::parse) so that if
* sections are rearranged, extended, etc. the entry point will be updated automatically. */
add_entry_rva(entry_rva);
/* The PE File Header has a fixed-size component followed by some number of RVA/Size pairs. The add_rvasize_pairs() will
* extend the header and parse the RVA/Size pairs. */
if (get_e_num_rvasize_pairs() > 1000) {
fprintf(stderr, "warning: PE File Header contains an unreasonable number of Rva/Size pairs. Limiting to 1000.\n");
set_e_num_rvasize_pairs(1000);
}
add_rvasize_pairs();
/* Construct the section table and its sections (non-synthesized sections). The specification says that the section table
* comes after the optional (NT) header, which in turn comes after the fixed part of the PE header. The size of the
* optional header is indicated in the fixed header. */
rose_addr_t secttab_offset = get_offset() + sizeof(PEFileHeader_disk) + get_e_nt_hdr_size();
rose_addr_t secttab_size = get_e_nsections() * sizeof(SgAsmPESectionTableEntry::PESectionTableEntry_disk);
SgAsmPESectionTable *secttab = new SgAsmPESectionTable(this);
secttab->set_offset(secttab_offset);
secttab->set_size(secttab_size);
secttab->parse();
set_section_table(secttab);
/* Parse the COFF symbol table */
if (get_e_coff_symtab() && get_e_coff_nsyms()) {
SgAsmCoffSymbolTable *symtab = new SgAsmCoffSymbolTable(this);
symtab->set_offset(get_e_coff_symtab());
symtab->parse();
set_coff_symtab(symtab);
}
/* Associate RVAs with particular sections so that if a section's mapping is changed the RVA gets adjusted automatically. */
ROSE_ASSERT(get_entry_rvas().size()==1);
get_entry_rvas()[0].bind(this);
set_e_code_rva(get_e_code_rva().bind(this));
set_e_data_rva(get_e_data_rva().bind(this));
/* Turn header-specified tables (RVA/Size pairs) into generic sections */
create_table_sections();
return this;
}
/** Parses an ELF Segment (Program Header) Table and constructs and parses all segments reachable from the table. The section
* is extended as necessary based on the number of entries and teh size of each entry. */
SgAsmElfSegmentTable *
SgAsmElfSegmentTable::parse()
{
SgAsmGenericSection::parse();
SgAsmElfFileHeader *fhdr = dynamic_cast<SgAsmElfFileHeader*>(get_header());
ROSE_ASSERT(fhdr!=NULL);
ByteOrder sex = fhdr->get_sex();
size_t ent_size, struct_size, opt_size, nentries;
calculate_sizes(&ent_size, &struct_size, &opt_size, &nentries);
ROSE_ASSERT(opt_size==fhdr->get_phextrasz() && nentries==fhdr->get_e_phnum());
/* If the current size is very small (0 or 1 byte) then we're coming straight from the constructor and the parsing should
* also extend this section to hold all the entries. Otherwise the caller must have assigned a specific size for a good
* reason and we should leave that alone, reading zeros if the entries extend beyond the defined size. */
if (get_size()<=1 && get_size()<nentries*ent_size)
extend(nentries*ent_size - get_size());
rose_addr_t offset=0; /* w.r.t. the beginning of this section */
for (size_t i=0; i<nentries; i++, offset+=ent_size) {
/* Read/decode the segment header */
SgAsmElfSegmentTableEntry *shdr = NULL;
if (4==fhdr->get_word_size()) {
SgAsmElfSegmentTableEntry::Elf32SegmentTableEntry_disk disk;
read_content_local(offset, &disk, struct_size);
shdr = new SgAsmElfSegmentTableEntry(sex, &disk);
} else {
SgAsmElfSegmentTableEntry::Elf64SegmentTableEntry_disk disk;
read_content_local(offset, &disk, struct_size);
shdr = new SgAsmElfSegmentTableEntry(sex, &disk);
}
shdr->set_index(i);
if (opt_size>0)
shdr->get_extra() = read_content_local_ucl(offset+struct_size, opt_size);
/* Null segments are just unused slots in the table; no real section to create */
if (SgAsmElfSegmentTableEntry::PT_NULL == shdr->get_type())
continue;
/* Create SgAsmElfSection objects for each ELF Segment. However, if the ELF Segment Table describes a segment
* that's the same offset and size as a section from the Elf Section Table (and the memory mappings are
* consistent) then use the preexisting section instead of creating a new one. */
SgAsmElfSection *s = NULL;
SgAsmGenericSectionPtrList possible = fhdr->get_file()->get_sections_by_offset(shdr->get_offset(), shdr->get_filesz());
for (size_t j=0; !s && j<possible.size(); j++) {
if (possible[j]->get_offset()!=shdr->get_offset() || possible[j]->get_size()!=shdr->get_filesz())
continue; /*different file extent*/
if (possible[j]->is_mapped()) {
if (possible[j]->get_mapped_preferred_rva()!=shdr->get_vaddr() ||
possible[j]->get_mapped_size()!=shdr->get_memsz())
continue; /*different mapped address or size*/
unsigned section_perms = (possible[j]->get_mapped_rperm() ? 0x01 : 0x00) |
(possible[j]->get_mapped_wperm() ? 0x02 : 0x00) |
(possible[j]->get_mapped_xperm() ? 0x04 : 0x00);
unsigned segment_perms = (shdr->get_flags() & SgAsmElfSegmentTableEntry::PF_RPERM ? 0x01 : 0x00) |
(shdr->get_flags() & SgAsmElfSegmentTableEntry::PF_WPERM ? 0x02 : 0x00) |
(shdr->get_flags() & SgAsmElfSegmentTableEntry::PF_XPERM ? 0x04 : 0x00);
if (section_perms != segment_perms)
continue; /*different mapped permissions*/
}
/* Found a match. Set memory mapping params only. */
s = dynamic_cast<SgAsmElfSection*>(possible[j]);
if (!s) continue; /*potential match was not from the ELF Section or Segment table*/
if (s->get_segment_entry()) continue; /*potential match is assigned to some other segment table entry*/
s->init_from_segment_table(shdr, true); /*true=>set memory mapping params only*/
}
/* Create a new segment if no matching section was found. */
if (!s) {
if (SgAsmElfSegmentTableEntry::PT_NOTE == shdr->get_type()) {
s = new SgAsmElfNoteSection(fhdr);
} else {
s = new SgAsmElfSection(fhdr);
}
s->init_from_segment_table(shdr);
s->parse();
}
}
return this;
}
/** Initialize by parsing a file. */
SgAsmElfEHFrameSection *
SgAsmElfEHFrameSection::parse()
{
SgAsmElfSection::parse();
SgAsmElfFileHeader *fhdr = get_elf_header();
ROSE_ASSERT(fhdr!=NULL);
rose_addr_t record_offset=0;
std::map<rose_addr_t, SgAsmElfEHFrameEntryCI*> cies;
while (record_offset<get_size()) {
rose_addr_t at = record_offset;
unsigned char u8_disk;
uint32_t u32_disk;
uint64_t u64_disk;
/* Length or extended length */
rose_addr_t length_field_size = 4; /*number of bytes not counted in length*/
read_content_local(at, &u32_disk, 4); at += 4;
rose_addr_t record_size = disk_to_host(fhdr->get_sex(), u32_disk);
if (record_size==0xffffffff) {
read_content_local(at, &u64_disk, 8); at += 8;
record_size = disk_to_host(fhdr->get_sex(), u64_disk);
length_field_size += 8; /*FIXME: it's not entirely clear whether ExtendedLength includes this field*/
}
if (0==record_size)
break;
/* Backward offset to CIE record, or zero if this is a CIE record. */
read_content_local(at, &u32_disk, 4); at += 4;
rose_addr_t cie_back_offset = disk_to_host(fhdr->get_sex(), u32_disk);
if (0==cie_back_offset) {
/* This is a CIE record */
SgAsmElfEHFrameEntryCI *cie = new SgAsmElfEHFrameEntryCI(this);
cies[record_offset] = cie;
/* Version */
uint8_t cie_version;
read_content_local(at++, &cie_version, 1);
cie->set_version(cie_version);
/* Augmentation String */
std::string astr = read_content_local_str(at);
at += astr.size() + 1;
cie->set_augmentation_string(astr);
/* Alignment factors */
cie->set_code_alignment_factor(read_content_local_uleb128(&at));
cie->set_data_alignment_factor(read_content_local_sleb128(&at));
/* Augmentation data length. This is apparently the length of the data described by the Augmentation String plus
* the Initial Instructions plus any padding. [RPM 2009-01-15] */
cie->set_augmentation_data_length(read_content_local_uleb128(&at));
/* Augmentation data. The format of the augmentation data in the CIE record is determined by reading the
* characters of the augmentation string. */
if (!astr.empty() && astr[0]=='z') {
for (size_t i=1; i<astr.size(); i++) {
if ('L'==astr[i]) {
read_content_local(at++, &u8_disk, 1);
cie->set_lsda_encoding(u8_disk);
} else if ('P'==astr[i]) {
/* The first byte is an encoding method which describes the following bytes, which are the address of
* a Personality Routine Handler. There appears to be very little documentation about these fields. */
read_content_local(at++, &u8_disk, 1);
cie->set_prh_encoding(u8_disk);
switch (cie->get_prh_encoding()) {
case 0x05: /* See Ubuntu 32bit /usr/bin/aptitude */
case 0x06: /* See second CIE record for Gentoo-Amd64 /usr/bin/addftinfo */
case 0x07: /* See first CIE record for Gentoo-Amd64 /usr/bin/addftinfo */
read_content_local(at++, &u8_disk, 1); /* not sure what this is; arg for __gxx_personality_v0? */
cie->set_prh_arg(u8_disk);
read_content_local(at, &u32_disk, 4); at+=4; /* address of <__gxx_personality_v0@plt> */
cie->set_prh_addr(ByteOrder::le_to_host(u32_disk));
break;
case 0x09: /* *.o file generated by gcc-4.0.x */
/* FIXME: Cannot find any info about this entry. Fix SgAsmElfEHFrameSection::parse() if we
* ever figure this out. [RPM 2009-09-29] */
/*fallthrough*/
default: {
if (++nwarnings<=WARNING_LIMIT) {
fprintf(stderr, "%s:%u: warning: ELF CIE 0x%08"PRIx64" has unknown PRH encoding 0x%02x\n",
__FILE__, __LINE__, get_offset()+record_offset, cie->get_prh_encoding());
if (WARNING_LIMIT==nwarnings)
fprintf(stderr, " (additional frame warnings will be suppressed)\n");
}
break;
}
}
} else if ('R'==astr[i]) {
read_content_local(at++, &u8_disk, 1);
cie->set_addr_encoding(u8_disk);
} else if ('S'==astr[i]) {
/* See http://lkml.indiana.edu/hypermail/linux/kernel/0602.3/1144.html and GCC PR #26208*/
cie->set_sig_frame(true);
} else {
/* Some stuff we don't handle yet. Warn about it and don't read anything. */
if (++nwarnings<=WARNING_LIMIT) {
fprintf(stderr, "%s:%u: warning: ELF CIE 0x%08"PRIx64" has invalid augmentation string \"%s\"\n",
__FILE__, __LINE__, get_offset()+record_offset, escapeString(astr).c_str());
if (WARNING_LIMIT==nwarnings)
fprintf(stderr, " (additional frame warnings will be suppressed)\n");
}
}
}
}
/* Initial instructions. These are apparently included in the augmentation_data_length. The final instructions can
* be zero padding (no-op instructions) to bring the record up to a multiple of the word size. */
rose_addr_t init_insn_size = (length_field_size + record_size) - (at - record_offset);
cie->get_instructions() = read_content_local_ucl(at, init_insn_size);
ROSE_ASSERT(cie->get_instructions().size()==init_insn_size);
} else {
/* This is a FDE record */
rose_addr_t cie_offset = record_offset + length_field_size - cie_back_offset;
assert(cies.find(cie_offset)!=cies.end());
SgAsmElfEHFrameEntryCI *cie = cies[cie_offset];
SgAsmElfEHFrameEntryFD *fde = new SgAsmElfEHFrameEntryFD(cie);
/* PC Begin (begin_rva) and size */
switch (cie->get_addr_encoding()) {
case -1: /* No address encoding specified */
case 0x01:
case 0x03:
case 0x1b: /* Address doesn't look valid (e.g., 0xfffd74e8) but still four bytes [RPM 2008-01-16]*/
{
read_content_local(at, &u32_disk, 4); at+=4;
fde->set_begin_rva(ByteOrder::le_to_host(u32_disk));
read_content_local(at, &u32_disk, 4); at+=4;
fde->set_size(ByteOrder::le_to_host(u32_disk));
break;
}
default:
fprintf(stderr, "%s:%u: ELF CIE 0x%08"PRIx64", FDE 0x%08"PRIx64": unknown address encoding: 0x%02x\n",
__FILE__, __LINE__, get_offset()+cie_offset, get_offset()+record_offset, cie->get_addr_encoding());
abort();
}
/* Augmentation Data */
std::string astring = cie->get_augmentation_string();
if (astring.size()>0 && astring[0]=='z') {
rose_addr_t aug_length = read_content_local_uleb128(&at);
fde->get_augmentation_data() = read_content_local_ucl(at, aug_length);
at += aug_length;
ROSE_ASSERT(fde->get_augmentation_data().size()==aug_length);
}
/* Call frame instructions */
rose_addr_t cf_insn_size = (length_field_size + record_size) - (at - record_offset);
fde->get_instructions() = read_content_local_ucl(at, cf_insn_size);
ROSE_ASSERT(fde->get_instructions().size()==cf_insn_size);
}
record_offset += length_field_size + record_size;
}
return this;
}
/** Initialize this header with information parsed from the file and construct and parse everything that's reachable from the
* header. Since the size of the ELF File Header is determined by the contents of the ELF File Header as stored in the file,
* the size of the ELF File Header will be adjusted upward if necessary. The ELF File Header should have been constructed
* such that SgAsmElfFileHeader::ctor() was called. */
SgAsmElfFileHeader*
SgAsmElfFileHeader::parse()
{
SgAsmGenericHeader::parse();
/* Read 32-bit header for now. Might need to re-read as 64-bit later. */
Elf32FileHeader_disk disk32;
if (sizeof(disk32)>get_size())
extend(sizeof(disk32)-get_size());
read_content_local(0, &disk32, sizeof disk32, false); /*zero pad if we read EOF*/
/* Check magic number early */
if (disk32.e_ident_magic[0]!=0x7f || disk32.e_ident_magic[1]!='E' ||
disk32.e_ident_magic[2]!='L' || disk32.e_ident_magic[3]!='F')
throw FormatError("Bad ELF magic number");
/* File byte order should be 1 or 2. However, we've seen at least one example that left the byte order at zero, implying
* that it was the native order. We don't have the luxury of decoding the file on the native machine, so in that case we
* try to infer the byte order by looking at one of the other multi-byte fields of the file. */
ByteOrder::Endianness sex;
if (1 == disk32.e_ident_data_encoding) {
sex = ByteOrder::ORDER_LSB;
} else if (2==disk32.e_ident_data_encoding) {
sex = ByteOrder::ORDER_MSB;
} else if ((disk32.e_type & 0xff00)==0xff00) {
/* One of the 0xffxx processor-specific flags in native order */
if ((disk32.e_type & 0x00ff)==0xff)
throw FormatError("invalid ELF header byte order"); /*ambiguous*/
sex = ByteOrder::host_order();
} else if ((disk32.e_type & 0x00ff)==0x00ff) {
/* One of the 0xffxx processor specific orders in reverse native order */
sex = ByteOrder::host_order()==ByteOrder::ORDER_LSB ? ByteOrder::ORDER_MSB : ByteOrder::ORDER_LSB;
} else if ((disk32.e_type & 0xff00)==0) {
/* One of the low-valued file types in native order */
if ((disk32.e_type & 0x00ff)==0)
throw FormatError("invalid ELF header byte order"); /*ambiguous*/
sex = ByteOrder::host_order();
} else if ((disk32.e_type & 0x00ff)==0) {
/* One of the low-valued file types in reverse native order */
sex = ByteOrder::host_order() == ByteOrder::ORDER_LSB ? ByteOrder::ORDER_MSB : ByteOrder::ORDER_LSB;
} else {
/* Ambiguous order */
throw FormatError("invalid ELF header byte order");
}
ROSE_ASSERT(p_exec_format != NULL);
p_exec_format->set_sex(sex);
p_e_ident_data_encoding = disk32.e_ident_data_encoding; /*save original value*/
/* Decode header to native format */
rose_rva_t entry_rva, sectab_rva, segtab_rva;
if (1 == disk32.e_ident_file_class) {
p_exec_format->set_word_size(4);
p_e_ident_padding.clear();
for (size_t i=0; i<sizeof(disk32.e_ident_padding); i++)
p_e_ident_padding.push_back(disk32.e_ident_padding[i]);
p_e_ident_file_class = ByteOrder::disk_to_host(sex, disk32.e_ident_file_class);
p_e_ident_file_version = ByteOrder::disk_to_host(sex, disk32.e_ident_file_version);
p_e_type = ByteOrder::disk_to_host(sex, disk32.e_type);
p_e_machine = ByteOrder::disk_to_host(sex, disk32.e_machine);
p_exec_format->set_version(ByteOrder::disk_to_host(sex, disk32.e_version));
entry_rva = ByteOrder::disk_to_host(sex, disk32.e_entry);
segtab_rva = ByteOrder::disk_to_host(sex, disk32.e_phoff);
sectab_rva = ByteOrder::disk_to_host(sex, disk32.e_shoff);
p_e_flags = ByteOrder::disk_to_host(sex, disk32.e_flags);
p_e_ehsize = ByteOrder::disk_to_host(sex, disk32.e_ehsize);
p_e_phnum = ByteOrder::disk_to_host(sex, disk32.e_phnum);
if (p_e_phnum>0) {
p_phextrasz = ByteOrder::disk_to_host(sex, disk32.e_phentsize);
ROSE_ASSERT(p_phextrasz>=sizeof(SgAsmElfSegmentTableEntry::Elf32SegmentTableEntry_disk));
p_phextrasz -= sizeof(SgAsmElfSegmentTableEntry::Elf32SegmentTableEntry_disk);
} else {
p_phextrasz = 0;
}
p_e_shnum = ByteOrder::disk_to_host(sex, disk32.e_shnum);
if (p_e_shnum>0) {
p_shextrasz = ByteOrder::disk_to_host(sex, disk32.e_shentsize);
ROSE_ASSERT(p_shextrasz>=sizeof(SgAsmElfSectionTableEntry::Elf32SectionTableEntry_disk));
p_shextrasz -= sizeof(SgAsmElfSectionTableEntry::Elf32SectionTableEntry_disk);
} else {
p_shextrasz = 0;
}
p_e_shstrndx = ByteOrder::disk_to_host(sex, disk32.e_shstrndx);
} else if (2 == disk32.e_ident_file_class) {
/* We guessed wrong. This is a 64-bit header, not 32-bit. */
p_exec_format->set_word_size(8);
Elf64FileHeader_disk disk64;
if (sizeof(disk64)>get_size())
extend(sizeof(disk64)-get_size());
read_content_local(0, &disk64, sizeof disk64, false); /*zero pad at EOF*/
p_e_ident_padding.clear();
for (size_t i=0; i<sizeof(disk64.e_ident_padding); i++)
p_e_ident_padding.push_back(disk64.e_ident_padding[i]);
p_e_ident_file_class = ByteOrder::disk_to_host(sex, disk64.e_ident_file_class);
p_e_ident_file_version = ByteOrder::disk_to_host(sex, disk64.e_ident_file_version);
p_e_type = ByteOrder::disk_to_host(sex, disk64.e_type);
p_e_machine = ByteOrder::disk_to_host(sex, disk64.e_machine);
p_exec_format->set_version(ByteOrder::disk_to_host(sex, disk64.e_version));
entry_rva = ByteOrder::disk_to_host(sex, disk64.e_entry);
segtab_rva = ByteOrder::disk_to_host(sex, disk64.e_phoff);
sectab_rva = ByteOrder::disk_to_host(sex, disk64.e_shoff);
p_e_flags = ByteOrder::disk_to_host(sex, disk64.e_flags);
p_e_ehsize = ByteOrder::disk_to_host(sex, disk64.e_ehsize);
p_e_phnum = ByteOrder::disk_to_host(sex, disk64.e_phnum);
if (p_e_phnum>0) {
p_phextrasz = ByteOrder::disk_to_host(sex, disk64.e_phentsize);
ROSE_ASSERT(p_phextrasz>=sizeof(SgAsmElfSegmentTableEntry::Elf64SegmentTableEntry_disk));
p_phextrasz -= sizeof(SgAsmElfSegmentTableEntry::Elf64SegmentTableEntry_disk);
} else {
p_phextrasz = 0;
}
p_e_shnum = ByteOrder::disk_to_host(sex, disk64.e_shnum);
if (p_e_shnum>0) {
p_shextrasz = ByteOrder::disk_to_host(sex, disk64.e_shentsize);
ROSE_ASSERT(p_shextrasz>=sizeof(SgAsmElfSectionTableEntry::Elf64SectionTableEntry_disk));
p_shextrasz -= sizeof(SgAsmElfSectionTableEntry::Elf64SectionTableEntry_disk);
} else {
p_shextrasz = 0;
}
p_e_shstrndx = ByteOrder::disk_to_host(sex, disk64.e_shstrndx);
} else {
throw FormatError("invalid ELF header file class");
}
/* Magic number. disk32 and disk64 have header bytes at same offset */
p_magic.clear();
for (size_t i=0; i<sizeof(disk32.e_ident_magic); i++)
p_magic.push_back(disk32.e_ident_magic[i]);
/* File format */
p_exec_format->set_family(FAMILY_ELF);
switch (p_e_type) {
case 0:
p_exec_format->set_purpose(PURPOSE_UNSPECIFIED);
break;
case 1:
case 3:
p_exec_format->set_purpose(PURPOSE_LIBRARY);
break;
case 2:
p_exec_format->set_purpose(PURPOSE_EXECUTABLE);
break;
case 4:
p_exec_format->set_purpose(PURPOSE_CORE_DUMP);
break;
default:
if (p_e_type >= 0xff00 && p_e_type <= 0xffff) {
p_exec_format->set_purpose(PURPOSE_PROC_SPECIFIC);
} else {
p_exec_format->set_purpose(PURPOSE_OTHER);
}
break;
}
p_exec_format->set_is_current_version(1 == p_exec_format->get_version());
p_exec_format->set_abi(ABI_UNSPECIFIED); /* ELF specifies a target architecture rather than an ABI */
p_exec_format->set_abi_version(0);
/* Target architecture */
set_isa(machine_to_isa(p_e_machine));
/* Read the optional section and segment tables and the sections to which they point. An empty section or segment table is
* treated as if it doesn't exist. This seems to be compatible with the loader since the 45-bit "tiny" ELF executable
* stores a zero in the e_shnum member and a completely invalid value in the e_shoff member. */
if (sectab_rva>0 && get_e_shnum()>0) {
SgAsmElfSectionTable *tab = new SgAsmElfSectionTable(this);
tab->set_offset(sectab_rva.get_rva());
tab->parse();
}
if (segtab_rva>0 && get_e_phnum()>0) {
SgAsmElfSegmentTable *tab = new SgAsmElfSegmentTable(this);
tab->set_offset(segtab_rva.get_rva());
tab->parse();
}
/* Associate the entry point with a particular section. */
entry_rva.bind(this);
add_entry_rva(entry_rva);
return this;
}
/** Parses an ELF Section Table and constructs and parses all sections reachable from the table. The section is extended as
* necessary based on the number of entries and the size of each entry. */
SgAsmElfSectionTable *
SgAsmElfSectionTable::parse()
{
SgAsmGenericSection::parse();
SgAsmElfFileHeader *fhdr = dynamic_cast<SgAsmElfFileHeader*>(get_header());
ROSE_ASSERT(fhdr!=NULL);
ByteOrder sex = fhdr->get_sex();
size_t ent_size, struct_size, opt_size, nentries;
calculate_sizes(&ent_size, &struct_size, &opt_size, &nentries);
ROSE_ASSERT(opt_size==fhdr->get_shextrasz() && nentries==fhdr->get_e_shnum());
/* If the current size is very small (0 or 1 byte) then we're coming straight from the constructor and the parsing should
* also extend this section to hold all the entries. Otherwise the caller must have assigned a specific size for a good
* reason and we should leave that alone, reading zeros if the entries extend beyond the defined size. */
if (get_size()<=1 && get_size()<nentries*ent_size)
extend(nentries*ent_size - get_size());
/* Read all the section headers. */
std::vector<SgAsmElfSectionTableEntry*> entries;
rose_addr_t offset = 0;
for (size_t i=0; i<nentries; i++, offset+=ent_size) {
SgAsmElfSectionTableEntry *shdr = NULL;
if (4 == fhdr->get_word_size()) {
SgAsmElfSectionTableEntry::Elf32SectionTableEntry_disk disk;
read_content_local(offset, &disk, struct_size);
shdr = new SgAsmElfSectionTableEntry(sex, &disk);
} else {
SgAsmElfSectionTableEntry::Elf64SectionTableEntry_disk disk;
read_content_local(offset, &disk, struct_size);
shdr = new SgAsmElfSectionTableEntry(sex, &disk);
}
if (opt_size>0)
shdr->get_extra() = read_content_local_ucl(offset+struct_size, opt_size);
entries.push_back(shdr);
}
/* This vector keeps track of which sections have already been parsed. We could get the same information by calling
* fhdr->get_section_by_id() and passing the entry number since entry numbers and IDs are one and the same in ELF. However,
* this is a bit easier. */
std::vector<SgAsmElfSection*> is_parsed;
is_parsed.resize(entries.size(), NULL);
/* All sections implicitly depend on the section string table for their names. */
SgAsmElfStringSection *section_name_strings=NULL;
if (fhdr->get_e_shstrndx() > 0) {
SgAsmElfSectionTableEntry *entry = entries[fhdr->get_e_shstrndx()];
section_name_strings = new SgAsmElfStringSection(fhdr);
section_name_strings->init_from_section_table(entry, section_name_strings, fhdr->get_e_shstrndx());
section_name_strings->parse();
is_parsed[fhdr->get_e_shstrndx()] = section_name_strings;
}
/* Read all the sections. Some sections depend on other sections, so we read them in such an order that all dependencies
* are satisfied first. */
while (1) {
bool try_again=false;
for (size_t i=0; i<entries.size(); i++) {
SgAsmElfSectionTableEntry *entry = entries[i];
ROSE_ASSERT(entry->get_sh_link()<entries.size());
/* Some sections might reference another section through the sh_link member. */
bool need_linked = entry->get_sh_link() > 0;
ROSE_ASSERT(!need_linked || entry->get_sh_link()<entries.size());
SgAsmElfSection *linked = need_linked ? is_parsed[entry->get_sh_link()] : NULL;
/* Relocation sections might have a second linked section stored in sh_info. */
bool need_info_linked = (entry->get_sh_type() == SgAsmElfSectionTableEntry::SHT_REL ||
entry->get_sh_type() == SgAsmElfSectionTableEntry::SHT_RELA) &&
entry->get_sh_info() > 0;
ROSE_ASSERT(!need_info_linked || entry->get_sh_info()<entries.size());
SgAsmElfSection *info_linked = need_info_linked ? is_parsed[entry->get_sh_info()] : NULL;
if (is_parsed[i]) {
/* This section has already been parsed. */
} else if ((need_linked && !linked) || (need_info_linked && !info_linked)) {
/* Don't parse this section yet because it depends on something that's not parsed yet. */
try_again = true;
} else {
switch (entry->get_sh_type()) {
case SgAsmElfSectionTableEntry::SHT_NULL:
/* Null entry. We still create the section just to hold the section header. */
is_parsed[i] = new SgAsmElfSection(fhdr);
break;
case SgAsmElfSectionTableEntry::SHT_NOBITS:
/* These types of sections don't occupy any file space (e.g., BSS) */
is_parsed[i] = new SgAsmElfSection(fhdr);
break;
case SgAsmElfSectionTableEntry::SHT_DYNAMIC: {
SgAsmElfStringSection *strsec = dynamic_cast<SgAsmElfStringSection*>(linked);
ROSE_ASSERT(strsec);
is_parsed[i] = new SgAsmElfDynamicSection(fhdr, strsec);
break;
}
case SgAsmElfSectionTableEntry::SHT_DYNSYM: {
SgAsmElfStringSection *strsec = dynamic_cast<SgAsmElfStringSection*>(linked);
ROSE_ASSERT(strsec);
SgAsmElfSymbolSection *symsec = new SgAsmElfSymbolSection(fhdr, strsec);
symsec->set_is_dynamic(true);
is_parsed[i] = symsec;
break;
}
case SgAsmElfSectionTableEntry::SHT_SYMTAB: {
SgAsmElfStringSection *strsec = dynamic_cast<SgAsmElfStringSection*>(linked);
ROSE_ASSERT(strsec);
SgAsmElfSymbolSection *symsec = new SgAsmElfSymbolSection(fhdr, strsec);
symsec->set_is_dynamic(false);
is_parsed[i] = symsec;
break;
}
case SgAsmElfSectionTableEntry::SHT_STRTAB:
is_parsed[i] = new SgAsmElfStringSection(fhdr);
break;
case SgAsmElfSectionTableEntry::SHT_REL: {
SgAsmElfSymbolSection *symbols = dynamic_cast<SgAsmElfSymbolSection*>(linked);
SgAsmElfRelocSection *relocsec = new SgAsmElfRelocSection(fhdr, symbols, info_linked);
relocsec->set_uses_addend(false);
is_parsed[i] = relocsec;
break;
}
case SgAsmElfSectionTableEntry::SHT_RELA: {
SgAsmElfSymbolSection *symbols = dynamic_cast<SgAsmElfSymbolSection*>(linked);
SgAsmElfRelocSection *relocsec = new SgAsmElfRelocSection(fhdr, symbols, info_linked);
relocsec->set_uses_addend(true);
is_parsed[i] = relocsec;
break;
}
case SgAsmElfSectionTableEntry::SHT_PROGBITS: {
if (!section_name_strings) {
fprintf(stderr, "SgAsmElfSectionTable::parse(): no string table for section table\n");
is_parsed[i] = new SgAsmElfSection(fhdr);
} else {
std::string section_name = section_name_strings->read_content_local_str(entry->get_sh_name());
if (section_name == ".eh_frame") {
is_parsed[i] = new SgAsmElfEHFrameSection(fhdr);
} else {
is_parsed[i] = new SgAsmElfSection(fhdr);
}
}
break;
}
case SgAsmElfSectionTableEntry::SHT_GNU_versym: {
is_parsed[i] = new SgAsmElfSymverSection(fhdr);
break;
}
case SgAsmElfSectionTableEntry::SHT_GNU_verdef: {
SgAsmElfStringSection *strsec = dynamic_cast<SgAsmElfStringSection*>(linked);
ROSE_ASSERT(strsec);
is_parsed[i] = new SgAsmElfSymverDefinedSection(fhdr,strsec);
break;
}
case SgAsmElfSectionTableEntry::SHT_GNU_verneed: {
SgAsmElfStringSection *strsec = dynamic_cast<SgAsmElfStringSection*>(linked);
ROSE_ASSERT(strsec);
is_parsed[i] = new SgAsmElfSymverNeededSection(fhdr,strsec);
break;
}
default:
is_parsed[i] = new SgAsmElfSection(fhdr);
break;
}
is_parsed[i]->init_from_section_table(entry, section_name_strings, i);
is_parsed[i]->parse();
}
}
if (!try_again)
break;
}
/* Initialize links between sections */
for (size_t i = 0; i < entries.size(); i++) {
SgAsmElfSectionTableEntry *shdr = entries[i];
if (shdr->get_sh_link() > 0) {
SgAsmElfSection *source = isSgAsmElfSection(fhdr->get_file()->get_section_by_id(i));
SgAsmElfSection *target = isSgAsmElfSection(fhdr->get_file()->get_section_by_id(shdr->get_sh_link()));
assert(source); /* because we created it above */
source->set_linked_section(target);
}
}
/* Finish parsing sections now that we have basic info for all the sections. */
for (size_t i=0; i<is_parsed.size(); i++)
is_parsed[i]->finish_parsing();
return this;
}
