SgAsmGenericSection *
SgAsmGenericFile::get_section_by_name(const std::string &name, char sep/*or NUL*/, size_t *nfound/*optional*/) const
{
SgAsmGenericSectionPtrList possible = get_sections_by_name(name, sep);
if (nfound) *nfound = possible.size();
return possible.size()==1 ? possible[0] : NULL;
}
SgAsmGenericSection *
SgAsmGenericFile::get_section_by_offset(rose_addr_t offset, rose_addr_t size, size_t *nfound) const
{
SgAsmGenericSectionPtrList possible = get_sections_by_offset(offset, size);
if (nfound) *nfound = possible.size();
return possible.size()==1 ? possible[0] : NULL;
}
SgAsmGenericSection *
SgAsmGenericFile::get_section_by_id(int id, size_t *nfound/*optional*/) const
{
SgAsmGenericSectionPtrList possible = get_sections_by_id(id);
if (nfound) *nfound = possible.size();
return possible.size()==1 ? possible[0] : NULL;
}
SgAsmGenericSection *
SgAsmGenericFile::get_section_by_va(rose_addr_t va, size_t *nfound/*optional*/) const
{
SgAsmGenericSectionPtrList possible = get_sections_by_va(va);
if (nfound) *nfound = possible.size();
return possible.size()==1 ? possible[0] : NULL;
}
/** Returns the section having a preferred or actual mapping that includes the specified virtual address. If @p use_preferred
* is set, then the condition is evaluated by looking at the section's preferred mapping, otherwise the actual mapping is
* used. If an actual mapping is used, the specified virtual address must be part of the actual mapped section, not merely in
* the memory region that was also mapped to satisfy alignment constraints. If there are no sections or multiple sections
* satisfying this condition then a null pointer is returned. */
SgAsmGenericSection *
SgAsmGenericHeader::get_section_by_va(rose_addr_t va, bool use_preferred, size_t *nfound/*optional*/) const
{
SgAsmGenericSectionPtrList possible = get_sections_by_va(va, use_preferred);
if (nfound) *nfound = possible.size();
return possible.size()==1 ? possible[0] : NULL;
}
/** Write the section table section back to disk */
void
SgAsmElfSectionTable::unparse(std::ostream &f) const
{
SgAsmElfFileHeader *fhdr = dynamic_cast<SgAsmElfFileHeader*>(get_header());
ROSE_ASSERT(fhdr!=NULL);
ByteOrder::Endianness sex = fhdr->get_sex();
SgAsmGenericSectionPtrList sections = fhdr->get_sectab_sections();
/* Write the sections first */
for (size_t i=0; i<sections.size(); i++)
sections[i]->unparse(f);
unparse_holes(f);
/* Calculate sizes. The ELF File Header should have been updated in reallocate() prior to unparsing. */
size_t ent_size, struct_size, opt_size, nentries;
calculate_sizes(&ent_size, &struct_size, &opt_size, &nentries);
ROSE_ASSERT(fhdr->get_shextrasz()==opt_size);
ROSE_ASSERT(fhdr->get_e_shnum()==nentries);
/* Write the section table entries */
for (size_t i=0; i<sections.size(); ++i) {
SgAsmElfSection *section = dynamic_cast<SgAsmElfSection*>(sections[i]);
ROSE_ASSERT(section!=NULL);
SgAsmElfSectionTableEntry *shdr = section->get_section_entry();
ROSE_ASSERT(shdr!=NULL);
ROSE_ASSERT(shdr->get_sh_offset()==section->get_offset());/*section table entry should have been updated in reallocate()*/
int id = section->get_id();
ROSE_ASSERT(id>=0 && (size_t)id<nentries);
SgAsmElfSectionTableEntry::Elf32SectionTableEntry_disk disk32;
SgAsmElfSectionTableEntry::Elf64SectionTableEntry_disk disk64;
void *disk = NULL;
if (4==fhdr->get_word_size()) {
disk = shdr->encode(sex, &disk32);
} else if (8==fhdr->get_word_size()) {
disk = shdr->encode(sex, &disk64);
} else {
ROSE_ASSERT(!"invalid word size");
}
/* The disk struct */
rose_addr_t spos = write(f, id*ent_size, struct_size, disk);
if (shdr->get_extra().size() > 0) {
ROSE_ASSERT(shdr->get_extra().size()<=opt_size);
write(f, spos, shdr->get_extra());
}
}
}
/* Pre-unparsing updates */
bool
SgAsmPESectionTable::reallocate()
{
bool reallocated = false;
/* Resize based on section having largest ID */
SgAsmPEFileHeader *fhdr = dynamic_cast<SgAsmPEFileHeader*>(get_header());
ROSE_ASSERT(fhdr != NULL);
SgAsmGenericSectionPtrList sections = fhdr->get_sections()->get_sections();
int max_id = 0;
for (size_t i=0; i<sections.size(); i++) {
max_id = std::max(max_id, sections[i]->get_id());
}
size_t nsections = max_id; /*PE section IDs are 1-origin*/
size_t need = nsections * sizeof(SgAsmPESectionTableEntry::PESectionTableEntry_disk);
if (need < get_size()) {
if (is_mapped()) {
ROSE_ASSERT(get_mapped_size()==get_size());
set_mapped_size(need);
}
set_size(need);
reallocated = true;
} else if (need > get_size()) {
get_file()->shift_extend(this, 0, need-get_size(), SgAsmGenericFile::ADDRSP_ALL, SgAsmGenericFile::ELASTIC_HOLE);
reallocated = true;
}
return reallocated;
}
SgAsmGenericSectionPtrList
SgAsmGenericFile::get_mapped_sections() const
{
SgAsmGenericSectionPtrList retval;
SgAsmGenericSectionPtrList all = get_sections(true);
for (size_t i=0; i<all.size(); i++) {
if (all[i]->is_mapped())
retval.push_back(all[i]);
}
return retval;
}
/** Get the list of sections defined in the ELF Segment Table */
SgAsmGenericSectionPtrList
SgAsmElfFileHeader::get_segtab_sections()
{
SgAsmGenericSectionPtrList retval;
SgAsmGenericSectionPtrList sections = get_sections()->get_sections();
for (size_t i=0; i<sections.size(); i++) {
SgAsmElfSection *elfsec = dynamic_cast<SgAsmElfSection*>(sections[i]);
if (elfsec && elfsec->get_segment_entry()!=NULL)
retval.push_back(elfsec);
}
return retval;
}
/** Returns info about the size of the entries based on information already available. Any or all arguments may be null
* pointers if the caller is not interested in the value. */
rose_addr_t
SgAsmElfSectionTable::calculate_sizes(size_t *entsize, size_t *required, size_t *optional, size_t *entcount) const
{
SgAsmElfFileHeader *fhdr = dynamic_cast<SgAsmElfFileHeader*>(get_header());
ROSE_ASSERT(fhdr!=NULL);
size_t struct_size = 0;
size_t extra_size = fhdr->get_shextrasz();
size_t entry_size = 0;
size_t nentries = 0;
/* Size of required part of each entry */
if (4==fhdr->get_word_size()) {
struct_size = sizeof(SgAsmElfSectionTableEntry::Elf32SectionTableEntry_disk);
} else if (8==fhdr->get_word_size()) {
struct_size = sizeof(SgAsmElfSectionTableEntry::Elf64SectionTableEntry_disk);
} else {
throw FormatError("bad ELF word size");
}
/* Entire entry should be at least large enough for the required part. */
entry_size = struct_size;
/* Size of optional parts. If we've parsed the table then use the largest optional part, otherwise assume the entry from
* the ELF File Header is correct. */
SgAsmGenericSectionPtrList sections = fhdr->get_sections()->get_sections();
for (size_t i=0; i<sections.size(); i++) {
SgAsmElfSection *elfsec = dynamic_cast<SgAsmElfSection*>(sections[i]);
if (elfsec && elfsec->get_section_entry()) {
ROSE_ASSERT(elfsec->get_id()>=0);
nentries = std::max(nentries, (size_t)elfsec->get_id()+1);
extra_size = std::max(extra_size, elfsec->get_section_entry()->get_extra().size());
}
}
/* Total number of entries. Either we haven't parsed the section table yet (nor created the sections it defines) or we
* have. In the former case we use the setting from the ELF File Header. Otherwise the table has to be large enough to
* store the section with the largest ID (ID also serves as the index into the ELF Section Table). */
if (0==nentries)
nentries = fhdr->get_e_shnum();
/* Return values */
if (entsize)
*entsize = entry_size;
if (required)
*required = struct_size;
if (optional)
*optional = extra_size;
if (entcount)
*entcount = nentries;
return entry_size * nentries;
}
/** Class method that prints info about offsets into known sections */
void
SgAsmGenericSection::dump_containing_sections(FILE *f, const std::string &prefix, rose_rva_t rva,
const SgAsmGenericSectionPtrList &slist)
{
for (size_t i=0; i<slist.size(); i++) {
SgAsmGenericSection *s = slist[i];
if (s->is_mapped() && rva>=s->get_mapped_preferred_rva() && rva<s->get_mapped_preferred_rva()+s->get_mapped_size()) {
rose_addr_t offset = rva - s->get_mapped_preferred_rva();
fprintf(f, "%-*s is 0x%08"PRIx64" (%"PRIu64") bytes into section [%d] \"%s\"\n",
DUMP_FIELD_WIDTH, prefix.c_str(), offset, offset, s->get_id(), s->get_name()->get_string(true).c_str());
}
}
}
/* 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;
}
/* Writes the section table back to disk. */
void
SgAsmPESectionTable::unparse(std::ostream &f) const
{
SgAsmPEFileHeader *fhdr = dynamic_cast<SgAsmPEFileHeader*>(get_header());
ROSE_ASSERT(fhdr != NULL);
SgAsmGenericSectionPtrList sections = fhdr->get_sections()->get_sections();
for (size_t i = 0; i < sections.size(); i++) {
if (sections[i]->get_id()>=0) {
SgAsmPESection *section = isSgAsmPESection(sections[i]);
ROSE_ASSERT(section!=NULL);
/* Write the table entry */
ROSE_ASSERT(section->get_id() > 0); /*ID's are 1-origin in PE*/
size_t slot = section->get_id() - 1;
SgAsmPESectionTableEntry *shdr = section->get_section_entry();
SgAsmPESectionTableEntry::PESectionTableEntry_disk disk;
shdr->encode(&disk);
write(f, slot*sizeof(disk), sizeof disk, &disk);
}
}
}
void
SgAsmGenericFile::shift_extend(SgAsmGenericSection *s, rose_addr_t sa, rose_addr_t sn, AddressSpace space, Elasticity elasticity)
{
ROSE_ASSERT(s!=NULL);
ROSE_ASSERT(s->get_file()==this);
ROSE_ASSERT((space & (ADDRSP_FILE|ADDRSP_MEMORY)) != 0);
const bool debug = false;
static size_t ncalls=0;
char p[256];
if (debug) {
const char *space_s="unknown";
if (space & ADDRSP_FILE) {
space_s = "file";
} else if (space & ADDRSP_MEMORY) {
space_s = "memory";
}
sprintf(p, "SgAsmGenericFile::shift_extend[%" PRIuPTR "]: ", ncalls++);
fprintf(stderr, "%s -- START --\n", p);
fprintf(stderr, "%s S = [%d] \"%s\"\n", p, s->get_id(), s->get_name()->get_string(true).c_str());
fprintf(stderr, "%s %s Sa=0x%08" PRIx64 " (%" PRIu64 "), Sn=0x%08" PRIx64 " (%" PRIu64 ")\n",
p, space_s, sa, sa, sn, sn);
fprintf(stderr, "%s elasticity = %s\n", p, (ELASTIC_NONE==elasticity ? "none" :
ELASTIC_UNREF==elasticity ? "unref" :
ELASTIC_HOLE==elasticity ? "unref+holes" :
"unknown"));
}
/* No-op case */
if (0==sa && 0==sn) {
if (debug) {
fprintf(stderr, "%s No change necessary.\n", p);
fprintf(stderr, "%s -- END --\n", p);
}
return;
}
bool filespace = (space & ADDRSP_FILE)!=0;
bool memspace = (space & ADDRSP_MEMORY)!=0;
rose_addr_t align=1, aligned_sa, aligned_sasn;
SgAsmGenericSectionPtrList neighbors, villagers;
ExtentMap amap; /* address mappings for all extents */
Extent sp;
/* Get a list of all sections that may need to be adjusted. */
SgAsmGenericSectionPtrList all;
switch (elasticity) {
case ELASTIC_NONE:
case ELASTIC_UNREF:
all = filespace ? get_sections() : get_mapped_sections();
break;
case ELASTIC_HOLE:
all = filespace ? get_sections(false) : get_mapped_sections();
break;
}
if (debug) {
fprintf(stderr, "%s Following sections are in 'all' set:\n", p);
for (size_t i=0; i<all.size(); i++) {
Extent ep;
if (filespace) {
ep = all[i]->get_file_extent();
} else {
ROSE_ASSERT(all[i]->is_mapped());
ep = all[i]->get_mapped_preferred_extent();
}
fprintf(stderr, "%s 0x%08" PRIx64 " 0x%08" PRIx64 " 0x%08" PRIx64 " [%d] \"%s\"\n",
p, ep.relaxed_first(), ep.size(), ep.relaxed_first()+ep.size(), all[i]->get_id(),
all[i]->get_name()->get_string(true).c_str());
}
}
for (size_t pass=0; pass<2; pass++) {
if (debug) {
fprintf(stderr, "%s -- %s --\n",
p, 0==pass?"FIRST PASS":"******");
}
/* S offset and size in file or memory address space */
if (filespace) {
sp = s->get_file_extent();
} else if (!memspace || !s->is_mapped()) {
return; /*nothing to do*/
} else {
sp = s->get_mapped_preferred_extent();
}
/* Build address map */
for (size_t i=0; i<all.size(); i++) {
if (filespace) {
amap.insert(all[i]->get_file_extent());
} else {
ROSE_ASSERT(all[i]->is_mapped());
amap.insert(all[i]->get_mapped_preferred_extent());
}
}
if (debug) {
fprintf(stderr, "%s Address map:\n", p);
amap.dump_extents(stderr, (std::string(p)+" ").c_str(), "amap");
fprintf(stderr, "%s Extent of S:\n", p);
fprintf(stderr, "%s start=0x%08" PRIx64 " size=0x%08" PRIx64 " end=0x%08" PRIx64 "\n",
p, sp.relaxed_first(), sp.size(), sp.relaxed_first()+sp.size());
}
/* Neighborhood (nhs) of S is a single extent. However, if S is zero size then nhs might be empty. The neighborhood of
* S is S plus all sections that overlap with S and all sections that are right-contiguous with S. */
ExtentMap nhs_map;
for (ExtentMap::iterator amapi=amap.begin(); amapi!=amap.end(); ++amapi) {
if (amapi->first.relaxed_first() <= sp.relaxed_first()+sp.size() &&
amapi->first.relaxed_first()+amapi->first.size() > sp.relaxed_first())
nhs_map.insert(amapi->first, amapi->second);
}
if (debug) {
fprintf(stderr, "%s Neighborhood of S:\n", p);
nhs_map.dump_extents(stderr, (std::string(p)+" ").c_str(), "nhs_map");
}
Extent nhs;
if (nhs_map.size()>0) {
assert(nhs_map.nranges()==1);
nhs = nhs_map.begin()->first;
} else {
nhs = sp;
}
/* What sections are in the neighborhood (including S), and right of the neighborhood? */
neighbors.clear(); /*sections in neighborhood*/
neighbors.push_back(s);
villagers.clear(); /*sections right of neighborhood*/
if (debug)
fprintf(stderr, "%s Ignoring left (L) sections:\n", p);
for (size_t i=0; i<all.size(); i++) {
SgAsmGenericSection *a = all[i];
if (a==s) continue; /*already pushed onto neighbors*/
Extent ap;
if (filespace) {
ap = a->get_file_extent();
} else if (!a->is_mapped()) {
continue;
} else {
ap = a->get_mapped_preferred_extent();
}
switch (ExtentMap::category(ap, nhs)) {
case 'L':
if (debug)
fprintf(stderr, "%s L 0x%08" PRIx64 " 0x%08" PRIx64 " 0x%08" PRIx64 " [%d] \"%s\"\n",
p, ap.relaxed_first(), ap.size(), ap.relaxed_first()+ap.size(),
a->get_id(), a->get_name()->get_string(true).c_str());
break;
case 'R':
if (ap.relaxed_first()==nhs.relaxed_first()+nhs.size() && 0==ap.size()) {
/* Empty sections immediately right of the neighborhood of S should actually be considered part of the
* neighborhood rather than right of it. */
neighbors.push_back(a);
} else if (elasticity!=ELASTIC_NONE) {
/* If holes are elastic then treat things right of the hole as being part of the right village; otherwise
* add those sections to the neighborhood of S even though they fall outside 'nhs' (it's OK because this
* partitioning of sections is the only thing we use 'nhs' for anyway. */
villagers.push_back(a);
} else if ('L'==ExtentMap::category(ap, sp)) {
/*ignore sections left of S*/
} else {
neighbors.push_back(a);
}
break;
default:
if ('L'!=ExtentMap::category(ap, sp)) /*ignore sections left of S*/
neighbors.push_back(a);
break;
}
}
if (debug) {
fprintf(stderr, "%s Neighbors:\n", p);
for (size_t i=0; i<neighbors.size(); i++) {
SgAsmGenericSection *a = neighbors[i];
Extent ap = filespace ? a->get_file_extent() : a->get_mapped_preferred_extent();
rose_addr_t align = filespace ? a->get_file_alignment() : a->get_mapped_alignment();
char cat = ExtentMap::category(ap, sp);
fprintf(stderr, "%s %c %c0x%08" PRIx64 " 0x%08" PRIx64 " 0x%08" PRIx64,
p, cat, 0==ap.relaxed_first() % (align?align:1) ? ' ' : '!',
ap.relaxed_first(), ap.size(), ap.relaxed_first()+ap.size());
if (strchr("RICE", cat)) {
fprintf(stderr, " align=0x%08" PRIx64, align);
} else {
fputs(" ", stderr);
}
fprintf(stderr, " [%2d] \"%s\"\n", a->get_id(), a->get_name()->get_string(true).c_str());
}
if (villagers.size()>0) fprintf(stderr, "%s Villagers:\n", p);
for (size_t i=0; i<villagers.size(); i++) {
SgAsmGenericSection *a = villagers[i];
Extent ap = filespace ? a->get_file_extent() : a->get_mapped_preferred_extent();
rose_addr_t align = filespace ? a->get_file_alignment() : a->get_mapped_alignment();
fprintf(stderr, "%s %c %c0x%08" PRIx64 " 0x%08" PRIx64 " 0x%08" PRIx64,
p, ExtentMap::category(ap, sp), /*cat should always be R*/
0==ap.relaxed_first() % (align?align:1) ? ' ' : '!',
ap.relaxed_first(), ap.size(), ap.relaxed_first()+ap.size());
fputs(" ", stderr);
fprintf(stderr, " [%2d] \"%s\"\n", a->get_id(), a->get_name()->get_string(true).c_str());
}
}
/* Adjust Sa to satisfy all alignment constraints in neighborhood(S) for sections that will move (cats R, I, C, and E). */
align = 1;
for (size_t i=0; i<neighbors.size(); i++) {
SgAsmGenericSection *a = neighbors[i];
Extent ap = filespace ? a->get_file_extent() : a->get_mapped_preferred_extent();
if (strchr("RICE", ExtentMap::category(ap, sp))) {
rose_addr_t x = filespace ? a->get_file_alignment() : a->get_mapped_alignment();
#if BOOST_VERSION < 106900
align = boost::math::lcm(align, x?x:1); // deprecated in boost-1.69.0
#else
align = boost::integer::lcm(align, x?x:1); // not present before boost-1.60.0
#endif
}
}
aligned_sa = (sa/align + (sa%align?1:0))*align;
aligned_sasn = ((sa+sn)/align + ((sa+sn)%align?1:0))*align;
if (debug) {
fprintf(stderr, "%s Alignment LCM = 0x%08" PRIx64 " (%" PRIu64 ")\n", p, align, align);
fprintf(stderr, "%s Aligned Sa = 0x%08" PRIx64 " (%" PRIu64 ")\n", p, aligned_sa, aligned_sa);
fprintf(stderr, "%s Aligned Sa+Sn = 0x%08" PRIx64 " (%" PRIu64 ")\n", p, aligned_sasn, aligned_sasn);
}
/* Are there any sections to the right of neighborhood(S)? If so, find the one with the lowest start address and use
* that to define the size of the hole right of neighborhood(S). */
if (0==villagers.size()) break;
SgAsmGenericSection *after_hole = NULL;
Extent hp(0, 0);
for (size_t i=0; i<villagers.size(); i++) {
SgAsmGenericSection *a = villagers[i];
Extent ap = filespace ? a->get_file_extent() : a->get_mapped_preferred_extent();
if (!after_hole || ap.relaxed_first()<hp.relaxed_first()) {
after_hole = a;
hp = ap;
}
}
ROSE_ASSERT(after_hole);
ROSE_ASSERT(hp.relaxed_first() > nhs.relaxed_first()+nhs.size());
rose_addr_t hole_size = hp.relaxed_first() - (nhs.relaxed_first()+nhs.size());
if (debug) {
fprintf(stderr, "%s hole size = 0x%08" PRIx64 " (%" PRIu64 "); need 0x%08" PRIx64 " (%" PRIu64 "); %s\n",
p, hole_size, hole_size, aligned_sasn, aligned_sasn,
hole_size>=aligned_sasn ? "large enough" : "not large enough");
}
if (hole_size >= aligned_sasn) break;
rose_addr_t need_more = aligned_sasn - hole_size;
/* Hole is not large enough. We need to recursively move things that are right of our neighborhood, then recompute the
* all-sections address map and neighborhood(S). */
ROSE_ASSERT(0==pass); /*logic problem since the recursive call should have enlarged the hole enough*/
if (debug) {
fprintf(stderr, "%s Calling recursively to increase hole size by 0x%08" PRIx64 " (%" PRIu64 ") bytes\n",
p, need_more, need_more);
}
shift_extend(after_hole, need_more, 0, space, elasticity);
if (debug) fprintf(stderr, "%s Returned from recursive call\n", p);
}
/* Consider sections that are in the same neighborhood as S */
if (debug) fprintf(stderr, "%s -- ADJUSTING --\n", p);
bool resized_mem = false;
for (size_t i=0; i<neighbors.size(); i++) {
SgAsmGenericSection *a = neighbors[i];
Extent ap = filespace ? a->get_file_extent() : a->get_mapped_preferred_extent();
switch (ExtentMap::category(ap, sp)) {
case 'L':
break;
case 'R':
if (filespace) {
a->set_offset(a->get_offset()+aligned_sasn);
} else {
a->set_mapped_preferred_rva(a->get_mapped_preferred_rva()+aligned_sasn);
}
break;
case 'C': /*including S itself*/
case 'E':
if (filespace) {
a->set_offset(a->get_offset()+aligned_sa);
a->set_size(a->get_size()+sn);
if (memspace && !resized_mem && a->is_mapped()) {
shift_extend(a, 0, sn, ADDRSP_MEMORY, elasticity);
resized_mem = true;
}
} else {
a->set_mapped_preferred_rva(a->get_mapped_preferred_rva()+aligned_sa);
a->set_mapped_size(a->get_mapped_size()+sn);
}
break;
case 'O':
if (ap.relaxed_first()==sp.relaxed_first()) {
if (filespace) {
a->set_offset(a->get_offset()+aligned_sa);
a->set_size(a->get_size()+sn);
} else {
a->set_mapped_preferred_rva(a->get_mapped_preferred_rva()+aligned_sa);
a->set_mapped_size(a->get_mapped_size()+sn);
}
} else {
if (filespace) {
a->set_size(a->get_size()+aligned_sasn);
if (memspace && !resized_mem && a->is_mapped()) {
shift_extend(a, 0, aligned_sasn, ADDRSP_MEMORY, elasticity);
resized_mem = true;
}
} else {
a->set_mapped_size(a->get_mapped_size()+aligned_sasn);
}
}
break;
case 'I':
if (filespace) {
a->set_offset(a->get_offset()+aligned_sa);
} else {
a->set_mapped_preferred_rva(a->get_mapped_preferred_rva()+aligned_sa);
}
break;
case 'B':
if (filespace) {
a->set_size(a->get_size()+sn);
if (memspace && !resized_mem && a->is_mapped()) {
shift_extend(a, 0, sn, ADDRSP_MEMORY, elasticity);
resized_mem = true;
}
} else {
a->set_mapped_size(a->get_size()+sn);
}
break;
default:
ROSE_ASSERT(!"invalid extent category");
break;
}
if (debug) {
const char *space_name = filespace ? "file" : "mem";
rose_addr_t x = filespace ? a->get_file_alignment() : a->get_mapped_alignment();
fprintf(stderr, "%s %4s-%c %c0x%08" PRIx64 " 0x%08" PRIx64 " 0x%08" PRIx64,
p, space_name, ExtentMap::category(ap, sp),
0==ap.relaxed_first()%(x?x:1)?' ':'!',
ap.relaxed_first(), ap.size(), ap.relaxed_first()+ap.size());
Extent newap = filespace ? a->get_file_extent() : a->get_mapped_preferred_extent();
fprintf(stderr, " -> %c0x%08" PRIx64 " 0x%08" PRIx64 " 0x%08" PRIx64,
0==newap.relaxed_first()%(x?x:1)?' ':'!',
newap.relaxed_first(), newap.size(), newap.relaxed_first()+newap.size());
fprintf(stderr, " [%2d] \"%s\"\n", a->get_id(), a->get_name()->get_string(true).c_str());
}
}
if (debug) fprintf(stderr, "%s -- END --\n", p);
}
void
SgAsmGenericFile::dump(FILE *f) const
{
fprintf(f, "Encoding: %s\n", get_data_converter() ? escapeString(get_data_converter()->name()).c_str() : "none");
SgAsmGenericSectionPtrList sections = get_sections();
if (sections.size()==0) {
fprintf(f, "No sections defined for file.\n");
return;
}
/* Sort sections by offset (lowest to highest), then size (largest to smallest but zero-sized entries first) */
for (size_t i = 1; i < sections.size(); i++) {
for (size_t j=0; j<i; j++) {
if (sections[j]->get_offset() == sections[i]->get_offset()) {
rose_addr_t size_i = sections[i]->get_size();
if (0==size_i) size_i = ~(rose_addr_t)0;
rose_addr_t size_j = sections[j]->get_size();
if (0==size_j) size_j = ~(rose_addr_t)0;
if (size_j < size_i) {
SgAsmGenericSection *x = sections[j];
sections[j] = sections[i];
sections[i] = x;
}
} else if (sections[j]->get_offset() > sections[i]->get_offset()) {
SgAsmGenericSection *x = sections[j];
sections[j] = sections[i];
sections[i] = x;
}
}
}
/* Print results */
fprintf(f, "File sections:\n");
fprintf(f, " Flg File-Addr File-Size File-End Base-VA Start-RVA Virt-Size End-RVA Perm ID Name\n");
fprintf(f, " --- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---- --- -----------------\n");
rose_addr_t high_water = 0;
for (size_t i=0; i<sections.size(); i++) {
SgAsmGenericSection *section = sections[i];
/* Does section overlap with any other (before or after)? */
char overlap[4] = " "; /* status characters: overlap prior, overlap subsequent, hole */
for (size_t j=0; overlap[0]==' ' && j<i; j++) {
if (sections[j]->get_offset()+sections[j]->get_size() > section->get_offset()) {
overlap[0] = '<';
}
}
for (size_t j=i+1; overlap[1]==' ' && j<sections.size(); j++) {
if (section->get_offset()+section->get_size() > sections[j]->get_offset()) {
overlap[1] = '>';
}
}
/* Is there a hole before section[i]? */
if (high_water < section->get_offset()) {
overlap[2] = 'H'; /* truly unaccounted region of the file */
} else if (i>0 && sections[i-1]->get_offset()+sections[i-1]->get_size() < section->get_offset()) {
overlap[2] = 'h'; /* unaccounted only if overlaps are not allowed */
}
high_water = std::max(high_water, section->get_offset() + section->get_size());
fprintf(f, " %3s", overlap);
/* File addresses */
fprintf(f, "%c0x%08" PRIx64 " 0x%08" PRIx64 " 0x%08" PRIx64,
section->get_file_alignment()==0 || section->get_offset()%section->get_file_alignment()==0?' ':'!',
section->get_offset(), section->get_size(), section->get_offset()+section->get_size());
/* Mapped addresses */
if (section->is_mapped()) {
fprintf(f, " %c0x%08" PRIx64 " 0x%08" PRIx64 " 0x%08" PRIx64 " 0x%08" PRIx64,
(section->get_mapped_alignment()==0 ||
section->get_mapped_preferred_rva()%section->get_mapped_alignment()==0?' ':'!'),
section->get_base_va(), section->get_mapped_preferred_rva(), section->get_mapped_size(),
section->get_mapped_preferred_rva()+section->get_mapped_size());
} else {
fprintf(f, " %*s", 4*11, "");
}
/* Permissions */
if (section->is_mapped()) {
fprintf(f, " %c%c%c ",
section->get_mapped_rperm()?'r':'-',
section->get_mapped_wperm()?'w':'-',
section->get_mapped_xperm()?'x':'-');
} else {
fputs(" ", f);
}
/* Section ID, name */
if (section->get_id()>=0) {
fprintf(f, " %3d", section->get_id());
} else {
fputs(" ", f);
}
fprintf(f, " %s\n", section->get_name()->get_string(true).c_str());
}
char overlap[4] = " ";
if (high_water < get_current_size()) {
overlap[2] = 'H';
} else if (sections.back()->get_offset() + sections.back()->get_size() < get_current_size()) {
overlap[2] = 'h';
}
fprintf(f, " %3s 0x%08" PRIx64 "%*s EOF", overlap, get_current_size(), 76, "");
if (get_current_size()!=p_data.size())
fprintf(f, " (original EOF was 0x%08zx)", p_data.size());
if (get_truncate_zeros())
fputs(" [ztrunc]", f);
fputc('\n', f);
fprintf(f, " --- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---- --- -----------------\n");
/* Show what part of the file has not been referenced */
AddressIntervalSet holes = get_unreferenced_extents();
if (holes.size()>0) {
fprintf(f, "These parts of the file have not been referenced during parsing:\n");
BOOST_FOREACH (const AddressInterval &interval, holes.intervals()) {
std::ostringstream ss;
using namespace StringUtility;
ss <<" " <<toHex(interval.least()) <<" + " <<toHex(interval.size()) <<" = " <<toHex(interval.greatest()+1) <<"\n";
fputs(ss.str().c_str(), f);
}
}
/** 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;
}
int
main()
{
/* The SgAsmGenericFile is the root of a tree describing a binary file (executable, shared lib, object, core dump).
* Usually this node appears in a much larger AST containing instructions contexts and other information, but for the
* purposes of this example, we'll just be generating the ELF Container and not any instructions. This root will
* eventually be passed to unparse() to generate the executable file. */
SgAsmGenericFile *ef = new SgAsmGenericFile;
/* The ELF File Header is the first thing in the file, always at offset zero. The constructors generally take arguments to
* describe the new object's relationship with existing objects, i.e., its location in the AST. In this case, the
* SgAsmElfFileHeader is a child of the SgAsmGenericFile (file headers are always children of their respective file).
*
* Section constructors (SgAsmElfFileHeader derives from SgAsmGenericSection) always place the new section at the current
* end-of-file and give it an initial size of one byte. This ensures that each section has a unique starting address,
* which is important when file memory is actually allocated for a section and other sections need to be moved out of the
* way--the allocator needs to know the relative positions of the sections in order to correctly relocate them.
*
* If we were parsing an ELF file we would usually use ROSE's frontend() method. However, one can also parse the file by
* first constructing the SgAsmElfFileHeader and then invoking its parse() method, which parses the ELF File Header and
* everything that can be reached from that header.
*
* We use 0x400000 as the virtual address of the main LOAD segment, which occupies the first part of the file, up through
* the end of the .text section (see below). ELF Headers don't actually store a base address, so instead of assigning one
* to the SgAsmElfFileHeader we'll leave the headers's base address at zero and add base_va explicitly whenever we need
* it. */
SgAsmElfFileHeader *fhdr = new SgAsmElfFileHeader(ef);
fhdr->get_exec_format()->set_word_size(8); /* default is 32-bit; we want 64-bit */
fhdr->set_isa(SgAsmExecutableFileFormat::ISA_X8664_Family); /* instruction set architecture; default is ISA_IA32_386 */
rose_addr_t base_va = 0x400000; /* base virtual address */
/* ELF executable files always have an ELF Segment Table (also called the ELF Program Header Table) usually appears
* immediately after the ELF File Header. The ELF Segment Table describes contiguous regions of the file that should be
* memory mapped by the loader. ELF Segments don't have names--names are imparted to the segment by virtue of a segment
* also being described by the ELF Section Table which we'll create later.
*
* Note that an SgAsmGenericSection (which represents both ELF Segments and ELF Sections) does not have the ELF Segment
* Table or ELF Section Table as a parent in the AST, but rather has the ELF File Header as a parent. This allows a single
* SgAsmGenericSection to represent an ELF Segment, an ELF Section, both, or neither. It also allows us to define ELF
* Segments and ELF Sections without needing to first define the ELF Segment Table or ELF Section Table, which is
* particularly useful when one wants one of those tables to appear at the end of the file. */
SgAsmElfSegmentTable *segtab = new SgAsmElfSegmentTable(fhdr);
#if 1
/* Create the .text section to hold machine instructions. We'll eventually add it to the ELF Section Table (which will be
* created at the end of the file) and store its name in an ELF String Section associated with the ELF Section Table. The
* fact that we haven't created the string table yet doesn't prevent us from giving the section a name.
*
* A note about names: Names are represented by SgAsmGenericString objects. The SgAsmGenericString class is virtual and
* subclasses SgAsmBasicString and SgAsmStoredString represent strings that have no file storage and strings that are
* stored in the file, respectively. The stored-string classes are further specialized to handle string tables for
* various file formats and understand how to map the string into a file, how to allocate and reallocate space, etc. In
* general, you want to change the string characters (with set_string()) stored in an SgAsmGenericString that's already
* present in the AST.
*
* The section constructors and reallocators don't worry about alignment issues--they always allocate from the next
* available byte. However, instructions typically need to satisfy some alignment constraints. We can easily adjust the
* file offset chosen by the constructor, but we also need to tell the reallocator about the alignment constraint.
*
* NOTE: There's a slight nuance of the reallocator that must be accounted for if you want to be very precise about where
* things are located in the file. When the reallocator shifts sections to make room for a section that needs to
* grow, it shifts all affected sections by a uniform amount. This preserves overlaps (like when a segment overlaps
* with a bunch of sections) but it also means that the alignment constraint we'll set for ".text" will cause the
* ELF Segment Table constructed above at a lower offset to move in tandem with ".text" when the ELF File Header (at
* an even lower address) needs to grow to its final size. This isn't a problem--you just end up with a couple unused
* bytes between the header and the segment table that will show up as a "hole" when subsequently parsed. If you're
* picky, the correct solution is to give the header a chance to allocate its file space before we construct the
* ".text" segment.
*
* In order to write the instructions out during the unparse phase, we create a new class from SgAsmElfSection and
* override the virtual unparse() method. The reallocate() method is called just before unparsing and we augment that in
* order to set an ELF-specific bit in the ELF Section Table flags (alternatively, we could have set that bit after the
* add_section() call below that adds the .text section to the ELF Section Table, but we do it here for demonstration).
* The calculate_sizes() function is used by SgAsmElfSection::reallocate() to figure out how much space is needed for the
* section.
*
* The instructions correspond to the assembly:
* .section .text
* .global _start
* _start:
* movl $1, %eax # _exit(86)
* movl $86, %ebx
* int $0x80 */
class TextSection : public SgAsmElfSection {
public:
TextSection(SgAsmElfFileHeader *fhdr, size_t ins_size, const unsigned char *ins_bytes)
: SgAsmElfSection(fhdr), ins_size(ins_size), ins_bytes(ins_bytes)
{}
virtual rose_addr_t calculate_sizes(size_t *entsize, size_t *required, size_t *optional, size_t *entcount) const {
if (entsize) *entsize = 1; /* an "entry" is one byte of instruction */
if (required) *required = 1; /* each "entry" is also stored in one byte of the file */
if (optional) *optional = 0; /* there is no extra data per instruction byte */
if (entcount) *entcount = ins_size; /* number of "entries" is the total instruction bytes */
return ins_size; /* return value is section size required */
}
virtual bool reallocate() {
bool retval = SgAsmElfSection::reallocate(); /* returns true if size or position of any section changed */
SgAsmElfSectionTableEntry *ste = get_section_entry();
ste->set_sh_flags(ste->get_sh_flags() | SgAsmElfSectionTableEntry::SHF_ALLOC); /* set the SHF_ALLOC bit */
return retval;
}
virtual void unparse(std::ostream &f) const {
write(f, 0, ins_size, ins_bytes); /* Write the instructions at offset zero in section */
}
size_t ins_size;
const unsigned char *ins_bytes;
};
ef->reallocate(); /* Give existing sections a chance to allocate file space */
static const unsigned char instructions[] = {0xb8, 0x01, 0x00, 0x00, 0x00, 0xbb, 0x56, 0x00, 0x00, 0x00, 0xcd, 0x80};
SgAsmElfSection *text = new TextSection(fhdr, NELMTS(instructions), instructions);
text->set_purpose(SgAsmGenericSection::SP_PROGRAM); /* Program-supplied text/data/etc. */
text->set_file_alignment(4); /* Tell reallocator about alignment constraint */
text->set_mapped_alignment(4); /* Alignment constraint for memory mapping */
text->set_mapped_preferred_rva(base_va+text->get_offset()); /* Mapped address is based on file offset */
text->set_mapped_size(text->get_size()); /* Mapped size is same as file size */
text->set_mapped_rperm(true); /* Readable */
text->set_mapped_wperm(false); /* Not writable */
text->set_mapped_xperm(true); /* Executable */
text->get_name()->set_string(".text"); /* Give section the standard name */
text->align(); /* Because we changed alignment constraints */
#else
/* Another way to do the same thing without a temporary class. */
ef->reallocate();
static const unsigned char instructions[] = {0xb8, 0x01, 0x00, 0x00, 0x00, 0xbb, 0x56, 0x00, 0x00, 0x00, 0xcd, 0x80};
SgAsmElfSection *text = new SgAsmElfSection(fhdr);
text->set_purpose(SgAsmGenericSection::SP_PROGRAM); /* Program-supplied text/data/etc. */
text->set_file_alignment(4); /* Tell reallocator about alignment constraint */
text->set_size(sizeof instructions); /* Give section a specific size */
text->set_mapped_alignment(4); /* Alignment constraint for memory mapping */
text->set_mapped_preferred_rva(base_va+text->get_offset()); /* Mapped address is based on file offset */
text->set_mapped_size(text->get_size()); /* Mapped size is same as file size */
text->set_mapped_rperm(true); /* Readable */
text->set_mapped_wperm(false); /* Not writable */
text->set_mapped_xperm(true); /* Executable */
text->get_name()->set_string(".text"); /* Give section the standard name */
text->align(); /* Because we changed alignment constraints */
unsigned char *content = text->writable_content(sizeof instructions);
memcpy(content, instructions, sizeof instructions);
#endif
/* Set the main entry point to be the first virtual address in the text section. A rose_rva_t address is associated
* with a section (.text in this case) so that if the starting address of the section changes (such as via reallocate())
* then the rose_rva_t address will also change so as to continue to refer to the same byte of the section. */
rose_rva_t entry_rva(text->get_mapped_preferred_rva(), text);
fhdr->add_entry_rva(entry_rva);
/* We now add an entry to the ELF Segment Table. ELF Segments often overlap multiple sections, and in this case we're
* going to create a segment that overlaps with the ELF File Header, the ELF Segment Table, and the ".text" section. The
* functions that reallocate space for sections (since most are born with a size of one byte) understand that things may
* overlap and make appropriate adjustments. In this case all we need to do is set the starting offset and size. */
SgAsmElfSection *seg1 = new SgAsmElfSection(fhdr); /* ELF Segments are represented by SgAsmElfSection */
seg1->get_name()->set_string("LOAD"); /* Segment names aren't saved (but useful for debugging) */
seg1->set_offset(0); /* Starts at beginning of file */
seg1->set_size(text->get_offset() + text->get_size()); /* Extends to end of .text section */
seg1->set_mapped_preferred_rva(base_va); /* Typically mapped by loader to this memory address */
seg1->set_mapped_size(seg1->get_size()); /* Make mapped size match size in the file */
seg1->set_mapped_rperm(true); /* Readable */
seg1->set_mapped_wperm(false); /* Not writable */
seg1->set_mapped_xperm(true); /* Executable */
segtab->add_section(seg1); /* Add definition to ELF Segment Table */
/* Add an empty PAX segment as further demonstration of adding an ELF Segment. We need to tell the loader that this
* section holds PAX Flags. There's no interface (currently) to do this at a generic level, so we make an adjustment
* directly to the ELF Segment Table entry. */
SgAsmElfSection *pax = new SgAsmElfSection(fhdr);
pax->get_name()->set_string("PAX Flags"); /* Name just for debugging */
pax->set_offset(0); /* Override address to be at zero rather than EOF */
pax->set_size(0); /* Override size to be zero rather than one byte */
segtab->add_section(pax); /* Add definition to ELF Segment Table */
pax->get_segment_entry()->set_type(SgAsmElfSegmentTableEntry::PT_PAX_FLAGS);
/* In order to give sections names there has to be an ELF String Section where the names are stored. An ELF String Section
* is a section itself and will appear in the ELF Section Table. The name of this string section will be stored in a
* string section, possibly this same string section. */
SgAsmElfStringSection *shstrtab = new SgAsmElfStringSection(fhdr);
shstrtab->get_name()->set_string(".shstrtab");
/* Most files have an ELF Section Table that describes the various sections in the file (the code, data, dynamic linking,
* symbol tables, string tables, etc). The ELF Section Table is a child of the ELF File Header, just as all other sections
* reachable from the ELF File Header are also children of the header. The class SgAsmGenericSection is used to represent
* all types of contiguous file regions, whether they're ELF Sections, ELF Segments, PE Objects, etc.
*
* The first entry of an ELF Section Table should always be an all-zero entry. It is not necessary to explicitly create
* the zero entry since the unparse() method will do so. When parsing an ELF Section Table, the parser will read the
* all-zero entry and actually create an empty, no-name, section with ID==0 (this is to handle cases that violate the ELF
* specification).
*
* Section names are stored in an ELF String Section that's associated with the ELF Section Table. We make that
* association by adding the ELF String Section to the ELF Section Table. The first string table that's added will be the
* one that stores section names. */
SgAsmElfSectionTable *sectab = new SgAsmElfSectionTable(fhdr);
sectab->add_section(text); /* Add the .text section */
sectab->add_section(shstrtab); /* Add the string table to store section names. */
#if 1 /* This stuff is here for demonstration. It's not necessary for the generated a.out to function properly. */
/* Create a symbol table. Symbol tables should always be added to the ELF Section Table, and thus have a name. ELF Symbol
* Tables need to know the section that serves as storage for the symbol names, an ELF String Section. The string table
* for the section name (set above) does not need to be the same as the string table used for symbol names, although we
* use the same one here just for brevity. */
SgAsmElfSymbolSection *symtab = new SgAsmElfSymbolSection(fhdr, shstrtab);
symtab->get_name()->set_string(".dynsym-test");
symtab->set_is_dynamic(true); /* This is a symbol table for dynamic linking */
sectab->add_section(symtab); /* Make the entry in the ELF Section Table */
/* Add some symbols to the symbol table. Symbols are born with an empty name pointing into the symbol string table and the
* names should be adjusted by calling get_name()->set_string("xxx"); there is no need to construct new SgAsmStoredString
* objects and assign them with set_name(). */
SgAsmElfSymbol *symbol;
symbol = new SgAsmElfSymbol(symtab); /* New symbol is a child of the symbol table */
symbol->set_value(0xaabbccdd); /* Arbitrary symbol value, often an RVA */
symbol->set_size(512); /* Arbitrary size of the thing to which the symbol refers */
symbol->get_name()->set_string("symbolA"); /* Symbol name */
symbol = new SgAsmElfSymbol(symtab);
symbol->set_value(0x11223344);
symbol->set_size(4);
symbol->get_name()->set_string("symbolB");
/* Create a .dynamic section. Dynamic sections are generally added to both the section table and the segment table and are
* memory mapped with read/write permission. They need an ELF String Table in order to store the library names. In real
* executables the string table is usually separate from the others, but we'll just use the one we already created. */
SgAsmElfDynamicSection *dynamic = new SgAsmElfDynamicSection(fhdr, shstrtab);
dynamic->get_name()->set_string(".dynamic-test"); /* Give section a name */
dynamic->set_mapped_preferred_rva(0x800000); /* RVA where loader will map section */
dynamic->set_mapped_size(dynamic->get_size()); /* Make mapped size same as file size */
dynamic->set_mapped_rperm(true); /* Readable */
dynamic->set_mapped_wperm(true); /* Writable */
dynamic->set_mapped_xperm(false); /* Not executable */
sectab->add_section(dynamic); /* Make an ELF Section Table entry */
segtab->add_section(dynamic); /* Make an ELF Segment Table entry */
/* Create some entries for the .dynamic section. Entries are born with a null name rather than an empty string since most
* entries don't have a string value. So an entry that requires a string value (like DT_NEEDED) will need to have a new
* string constructed. You don't have to worry about placing the new string in a particular string table since
* reallocate() will eventually take care of that, so you can even use the very simple SgAsmBasicString interface. */
SgAsmElfDynamicEntry *dynent;
/* Add some DT_NULL entries */
dynent = new SgAsmElfDynamicEntry(dynamic);
dynent = new SgAsmElfDynamicEntry(dynamic);
dynent = new SgAsmElfDynamicEntry(dynamic);
/* Add a library name */
dynent = new SgAsmElfDynamicEntry(dynamic);
dynent->set_d_tag(SgAsmElfDynamicEntry::DT_NEEDED);
dynent->set_name(new SgAsmBasicString("testlib"));
/* Create a relocation table. Relocation tables depend on a symbol table and, indirectly, a string table. */
SgAsmElfRelocSection *reladyn = new SgAsmElfRelocSection(fhdr, symtab,NULL);
reladyn->get_name()->set_string(".rela.dyn-test");
reladyn->set_uses_addend(true); /* This is a "rela" table rather than a "rel" table */
sectab->add_section(reladyn); /* Make an ELF Section Table entry */
/* Add a relocation entry to the relocation table. We create just one here, but in practice you would create as many as
* you need. Even REL tabels have a set_r_addend() method, but you should leave the addend at zero for them. The
* relocation type determines how many bits of the offset are significant (and other details) and are described in the
* system <elf.h> file (we may eventually change this interface to unify the various file formats). */
SgAsmElfRelocEntry *relent = new SgAsmElfRelocEntry(reladyn);
relent->set_r_offset(0x080495ac); /* An arbitrary offset */
relent->set_r_addend(0xaa); /* Arbitrary value to add to relocation address */
relent->set_sym(symtab->index_of(symbol)); /* Index of symbol in its symbol table */
relent->set_type(SgAsmElfRelocEntry::R_386_JMP_SLOT); /* */
/* We can add notes to an executable */
SgAsmElfNoteSection *notes = new SgAsmElfNoteSection(fhdr);
notes->set_name(new SgAsmBasicString("ELF Notes"));
notes->set_mapped_preferred_rva(0x900000); /* RVA where loader will map section */
notes->set_mapped_size(notes->get_size()); /* Make mapped size same as file size */
notes->set_mapped_rperm(true); /* Readable */
notes->set_mapped_wperm(true); /* Writable */
notes->set_mapped_xperm(false); /* Not executable */
SgAsmElfNoteEntry *note = new SgAsmElfNoteEntry(notes);
note->set_name(new SgAsmBasicString("ROSE"));
const char *payload = "This is a test note.";
for (size_t i=0; payload[i]; i++) {
note->get_payload().push_back(payload[i]);
}
segtab->add_section(notes);
/* Add a Symbol Version section. This is a gnu extension to elf to support versioning of individual symbols
the size of the symver section should match the size of the dynamic symbol section (nominally .dynsym).
there should also be an entry in the .dynamic section [DT_VERDEFNUM] that matches this same number.
*/
SgAsmElfSymverSection* symver = new SgAsmElfSymverSection(fhdr);
symver->get_name()->set_string(".gnu-version-test"); /* Give section a name */
symver->set_mapped_preferred_rva(0xA00000); /* RVA where loader will map section */
symver->set_mapped_size(symver->get_size()); /* Make mapped size same as file size */
symver->set_mapped_rperm(true); /* Readable */
symver->set_mapped_wperm(false); /* Writable */
symver->set_mapped_xperm(false); /* Not executable */
sectab->add_section(symver); /* Make an ELF Section Table entry */
for(size_t i=0; i < symtab->get_symbols()->get_symbols().size(); ++i){
SgAsmElfSymverEntry* symver_entry = new SgAsmElfSymverEntry(symver);
symver_entry->set_value(i);
}
SgAsmElfSymverDefinedSection* symver_def = new SgAsmElfSymverDefinedSection(fhdr,shstrtab);
symver_def->get_name()->set_string(".gnu.version_d-test");
symver_def->set_mapped_preferred_rva(0xB00000); /* RVA where loader will map section */
symver_def->set_mapped_size(symver_def->get_size()); /* Make mapped size same as file size */
symver_def->set_mapped_rperm(true); /* Readable */
symver_def->set_mapped_wperm(false); /* Writable */
symver_def->set_mapped_xperm(false); /* Not executable */
sectab->add_section(symver_def); /* Make an ELF Section Table entry */
/** normal case - one Aux structure*/
SgAsmElfSymverDefinedEntry* symdefEntry1 = new SgAsmElfSymverDefinedEntry(symver_def);
symdefEntry1->set_version(1);
symdefEntry1->set_hash(0xdeadbeef);// need to implement real hash support?
symdefEntry1->set_flags(0x0);
SgAsmElfSymverDefinedAux* symdefAux1a = new SgAsmElfSymverDefinedAux(symdefEntry1, symver_def);
symdefAux1a->get_name()->set_string("version-1");
SgAsmElfSymverDefinedEntry* symdefEntry2 = new SgAsmElfSymverDefinedEntry(symver_def);
new SgAsmElfSymverDefinedAux(symdefEntry2, symver_def);
new SgAsmElfSymverDefinedAux(symdefEntry2, symver_def);
/* Very strange case, but technically legal - zero Aux */
new SgAsmElfSymverDefinedEntry(symver_def);
SgAsmElfSymverNeededSection* symver_need = new SgAsmElfSymverNeededSection(fhdr,shstrtab);
symver_need->get_name()->set_string(".gnu.version_d-test");
symver_need->set_mapped_preferred_rva(0xC00000); /* RVA where loader will map section */
symver_need->set_mapped_size(symver_need->get_size()); /* Make mapped size same as file size */
symver_need->set_mapped_rperm(true); /* Readable */
symver_need->set_mapped_wperm(false); /* Writable */
symver_need->set_mapped_xperm(false); /* Not executable */
sectab->add_section(symver_need); /* Make an ELF Section Table entry */
#endif
/* Some of the sections we created above have default sizes of one byte because there's no way to determine their true
* size until we're all done creating sections. The SgAsmGenericFile::reallocate() traverses the AST and allocates the
* actual file storage for each section, adjusting other sections to make room and updating various tables and
* file-based data structures. You can call this function at any time, but it must be called before unparse(). */
ef->reallocate();
/* The reallocate() has a shortcoming [as of 2008-12-19] in that it might not correctly update memory mappings in the case
* where the mapping for a section is inferred from the mapping of a containing segment when the section is shifted in the
* file to make room for an earlier section that's also in the segment.
*
* We can work around that here by explicitly setting the memory mapping to be relative to the segment. However, we'll
* need to call reallocate() again because besides moving things around, reallocate() also updates some ELF-specific data
* structures from values stored in the more generic structures. */
text->set_mapped_preferred_rva(seg1->get_mapped_preferred_rva()+(text->get_offset()-seg1->get_offset()));
ef->reallocate(); /*won't resize or move things this time since we didn't modify much since the last call to reallocate()*/
/* Show some results. This is largely the same output as would be found in the *.dump file produced by many of the other
* binary analysis tools. */
ef->dump(stdout);
SgAsmGenericSectionPtrList all = ef->get_sections(true);
for (size_t i=0; i<all.size(); i++) {
fprintf(stdout, "Section %zu:\n", i);
all[i]->dump(stdout, " ", -1);
}
/* Unparse the AST to generate an executable. */
std::ofstream f("a.out");
ef->unparse(f);
return 0;
}
