/** 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;
}
/** 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;
}
/** 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;
}
/** 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;
}
