// object.cc -- support for an object file for linking in gold // Copyright (C) 2006-2019 Free Software Foundation, Inc. // Written by Ian Lance Taylor . // This file is part of gold. // This program is free software; you can redistribute it and/or modify // it under the terms of the GNU General Public License as published by // the Free Software Foundation; either version 3 of the License, or // (at your option) any later version. // This program is distributed in the hope that it will be useful, // but WITHOUT ANY WARRANTY; without even the implied warranty of // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the // GNU General Public License for more details. // You should have received a copy of the GNU General Public License // along with this program; if not, write to the Free Software // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston, // MA 02110-1301, USA. #include "gold.h" #include #include #include #include "demangle.h" #include "libiberty.h" #include "gc.h" #include "target-select.h" #include "dwarf_reader.h" #include "layout.h" #include "output.h" #include "symtab.h" #include "cref.h" #include "reloc.h" #include "object.h" #include "dynobj.h" #include "plugin.h" #include "compressed_output.h" #include "incremental.h" #include "merge.h" namespace gold { // Struct Read_symbols_data. // Destroy any remaining File_view objects and buffers of decompressed // sections. Read_symbols_data::~Read_symbols_data() { if (this->section_headers != NULL) delete this->section_headers; if (this->section_names != NULL) delete this->section_names; if (this->symbols != NULL) delete this->symbols; if (this->symbol_names != NULL) delete this->symbol_names; if (this->versym != NULL) delete this->versym; if (this->verdef != NULL) delete this->verdef; if (this->verneed != NULL) delete this->verneed; } // Class Xindex. // Initialize the symtab_xindex_ array. Find the SHT_SYMTAB_SHNDX // section and read it in. SYMTAB_SHNDX is the index of the symbol // table we care about. template void Xindex::initialize_symtab_xindex(Object* object, unsigned int symtab_shndx) { if (!this->symtab_xindex_.empty()) return; gold_assert(symtab_shndx != 0); // Look through the sections in reverse order, on the theory that it // is more likely to be near the end than the beginning. unsigned int i = object->shnum(); while (i > 0) { --i; if (object->section_type(i) == elfcpp::SHT_SYMTAB_SHNDX && this->adjust_shndx(object->section_link(i)) == symtab_shndx) { this->read_symtab_xindex(object, i, NULL); return; } } object->error(_("missing SHT_SYMTAB_SHNDX section")); } // Read in the symtab_xindex_ array, given the section index of the // SHT_SYMTAB_SHNDX section. If PSHDRS is not NULL, it points at the // section headers. template void Xindex::read_symtab_xindex(Object* object, unsigned int xindex_shndx, const unsigned char* pshdrs) { section_size_type bytecount; const unsigned char* contents; if (pshdrs == NULL) contents = object->section_contents(xindex_shndx, &bytecount, false); else { const unsigned char* p = (pshdrs + (xindex_shndx * elfcpp::Elf_sizes::shdr_size)); typename elfcpp::Shdr shdr(p); bytecount = convert_to_section_size_type(shdr.get_sh_size()); contents = object->get_view(shdr.get_sh_offset(), bytecount, true, false); } gold_assert(this->symtab_xindex_.empty()); this->symtab_xindex_.reserve(bytecount / 4); for (section_size_type i = 0; i < bytecount; i += 4) { unsigned int shndx = elfcpp::Swap<32, big_endian>::readval(contents + i); // We preadjust the section indexes we save. this->symtab_xindex_.push_back(this->adjust_shndx(shndx)); } } // Symbol symndx has a section of SHN_XINDEX; return the real section // index. unsigned int Xindex::sym_xindex_to_shndx(Object* object, unsigned int symndx) { if (symndx >= this->symtab_xindex_.size()) { object->error(_("symbol %u out of range for SHT_SYMTAB_SHNDX section"), symndx); return elfcpp::SHN_UNDEF; } unsigned int shndx = this->symtab_xindex_[symndx]; if (shndx < elfcpp::SHN_LORESERVE || shndx >= object->shnum()) { object->error(_("extended index for symbol %u out of range: %u"), symndx, shndx); return elfcpp::SHN_UNDEF; } return shndx; } // Class Object. // Report an error for this object file. This is used by the // elfcpp::Elf_file interface, and also called by the Object code // itself. void Object::error(const char* format, ...) const { va_list args; va_start(args, format); char* buf = NULL; if (vasprintf(&buf, format, args) < 0) gold_nomem(); va_end(args); gold_error(_("%s: %s"), this->name().c_str(), buf); free(buf); } // Return a view of the contents of a section. const unsigned char* Object::section_contents(unsigned int shndx, section_size_type* plen, bool cache) { return this->do_section_contents(shndx, plen, cache); } // Read the section data into SD. This is code common to Sized_relobj_file // and Sized_dynobj, so we put it into Object. template void Object::read_section_data(elfcpp::Elf_file* elf_file, Read_symbols_data* sd) { const int shdr_size = elfcpp::Elf_sizes::shdr_size; // Read the section headers. const off_t shoff = elf_file->shoff(); const unsigned int shnum = this->shnum(); sd->section_headers = this->get_lasting_view(shoff, shnum * shdr_size, true, true); // Read the section names. const unsigned char* pshdrs = sd->section_headers->data(); const unsigned char* pshdrnames = pshdrs + elf_file->shstrndx() * shdr_size; typename elfcpp::Shdr shdrnames(pshdrnames); if (shdrnames.get_sh_type() != elfcpp::SHT_STRTAB) this->error(_("section name section has wrong type: %u"), static_cast(shdrnames.get_sh_type())); sd->section_names_size = convert_to_section_size_type(shdrnames.get_sh_size()); sd->section_names = this->get_lasting_view(shdrnames.get_sh_offset(), sd->section_names_size, false, false); } // If NAME is the name of a special .gnu.warning section, arrange for // the warning to be issued. SHNDX is the section index. Return // whether it is a warning section. bool Object::handle_gnu_warning_section(const char* name, unsigned int shndx, Symbol_table* symtab) { const char warn_prefix[] = ".gnu.warning."; const int warn_prefix_len = sizeof warn_prefix - 1; if (strncmp(name, warn_prefix, warn_prefix_len) == 0) { // Read the section contents to get the warning text. It would // be nicer if we only did this if we have to actually issue a // warning. Unfortunately, warnings are issued as we relocate // sections. That means that we can not lock the object then, // as we might try to issue the same warning multiple times // simultaneously. section_size_type len; const unsigned char* contents = this->section_contents(shndx, &len, false); if (len == 0) { const char* warning = name + warn_prefix_len; contents = reinterpret_cast(warning); len = strlen(warning); } std::string warning(reinterpret_cast(contents), len); symtab->add_warning(name + warn_prefix_len, this, warning); return true; } return false; } // If NAME is the name of the special section which indicates that // this object was compiled with -fsplit-stack, mark it accordingly. bool Object::handle_split_stack_section(const char* name) { if (strcmp(name, ".note.GNU-split-stack") == 0) { this->uses_split_stack_ = true; return true; } if (strcmp(name, ".note.GNU-no-split-stack") == 0) { this->has_no_split_stack_ = true; return true; } return false; } // Class Relobj template void Relobj::initialize_input_to_output_map(unsigned int shndx, typename elfcpp::Elf_types::Elf_Addr starting_address, Unordered_map::Elf_Addr>* output_addresses) const { Object_merge_map *map = this->object_merge_map_; map->initialize_input_to_output_map(shndx, starting_address, output_addresses); } void Relobj::add_merge_mapping(Output_section_data *output_data, unsigned int shndx, section_offset_type offset, section_size_type length, section_offset_type output_offset) { Object_merge_map* object_merge_map = this->get_or_create_merge_map(); object_merge_map->add_mapping(output_data, shndx, offset, length, output_offset); } bool Relobj::merge_output_offset(unsigned int shndx, section_offset_type offset, section_offset_type *poutput) const { Object_merge_map* object_merge_map = this->object_merge_map_; if (object_merge_map == NULL) return false; return object_merge_map->get_output_offset(shndx, offset, poutput); } const Output_section_data* Relobj::find_merge_section(unsigned int shndx) const { Object_merge_map* object_merge_map = this->object_merge_map_; if (object_merge_map == NULL) return NULL; return object_merge_map->find_merge_section(shndx); } // To copy the symbols data read from the file to a local data structure. // This function is called from do_layout only while doing garbage // collection. void Relobj::copy_symbols_data(Symbols_data* gc_sd, Read_symbols_data* sd, unsigned int section_header_size) { gc_sd->section_headers_data = new unsigned char[(section_header_size)]; memcpy(gc_sd->section_headers_data, sd->section_headers->data(), section_header_size); gc_sd->section_names_data = new unsigned char[sd->section_names_size]; memcpy(gc_sd->section_names_data, sd->section_names->data(), sd->section_names_size); gc_sd->section_names_size = sd->section_names_size; if (sd->symbols != NULL) { gc_sd->symbols_data = new unsigned char[sd->symbols_size]; memcpy(gc_sd->symbols_data, sd->symbols->data(), sd->symbols_size); } else { gc_sd->symbols_data = NULL; } gc_sd->symbols_size = sd->symbols_size; gc_sd->external_symbols_offset = sd->external_symbols_offset; if (sd->symbol_names != NULL) { gc_sd->symbol_names_data = new unsigned char[sd->symbol_names_size]; memcpy(gc_sd->symbol_names_data, sd->symbol_names->data(), sd->symbol_names_size); } else { gc_sd->symbol_names_data = NULL; } gc_sd->symbol_names_size = sd->symbol_names_size; } // This function determines if a particular section name must be included // in the link. This is used during garbage collection to determine the // roots of the worklist. bool Relobj::is_section_name_included(const char* name) { if (is_prefix_of(".ctors", name) || is_prefix_of(".dtors", name) || is_prefix_of(".note", name) || is_prefix_of(".init", name) || is_prefix_of(".fini", name) || is_prefix_of(".gcc_except_table", name) || is_prefix_of(".jcr", name) || is_prefix_of(".preinit_array", name) || (is_prefix_of(".text", name) && strstr(name, "personality")) || (is_prefix_of(".data", name) && strstr(name, "personality")) || (is_prefix_of(".sdata", name) && strstr(name, "personality")) || (is_prefix_of(".gnu.linkonce.d", name) && strstr(name, "personality")) || (is_prefix_of(".rodata", name) && strstr(name, "nptl_version"))) { return true; } return false; } // Finalize the incremental relocation information. Allocates a block // of relocation entries for each symbol, and sets the reloc_bases_ // array to point to the first entry in each block. If CLEAR_COUNTS // is TRUE, also clear the per-symbol relocation counters. void Relobj::finalize_incremental_relocs(Layout* layout, bool clear_counts) { unsigned int nsyms = this->get_global_symbols()->size(); this->reloc_bases_ = new unsigned int[nsyms]; gold_assert(this->reloc_bases_ != NULL); gold_assert(layout->incremental_inputs() != NULL); unsigned int rindex = layout->incremental_inputs()->get_reloc_count(); for (unsigned int i = 0; i < nsyms; ++i) { this->reloc_bases_[i] = rindex; rindex += this->reloc_counts_[i]; if (clear_counts) this->reloc_counts_[i] = 0; } layout->incremental_inputs()->set_reloc_count(rindex); } Object_merge_map* Relobj::get_or_create_merge_map() { if (!this->object_merge_map_) this->object_merge_map_ = new Object_merge_map(); return this->object_merge_map_; } // Class Sized_relobj. // Iterate over local symbols, calling a visitor class V for each GOT offset // associated with a local symbol. template void Sized_relobj::do_for_all_local_got_entries( Got_offset_list::Visitor* v) const { unsigned int nsyms = this->local_symbol_count(); for (unsigned int i = 0; i < nsyms; i++) { Local_got_entry_key key(i, 0); Local_got_offsets::const_iterator p = this->local_got_offsets_.find(key); if (p != this->local_got_offsets_.end()) { const Got_offset_list* got_offsets = p->second; got_offsets->for_all_got_offsets(v); } } } // Get the address of an output section. template uint64_t Sized_relobj::do_output_section_address( unsigned int shndx) { // If the input file is linked as --just-symbols, the output // section address is the input section address. if (this->just_symbols()) return this->section_address(shndx); const Output_section* os = this->do_output_section(shndx); gold_assert(os != NULL); return os->address(); } // Class Sized_relobj_file. template Sized_relobj_file::Sized_relobj_file( const std::string& name, Input_file* input_file, off_t offset, const elfcpp::Ehdr& ehdr) : Sized_relobj(name, input_file, offset), elf_file_(this, ehdr), symtab_shndx_(-1U), local_symbol_count_(0), output_local_symbol_count_(0), output_local_dynsym_count_(0), symbols_(), defined_count_(0), local_symbol_offset_(0), local_dynsym_offset_(0), local_values_(), local_plt_offsets_(), kept_comdat_sections_(), has_eh_frame_(false), is_deferred_layout_(false), deferred_layout_(), deferred_layout_relocs_(), output_views_(NULL) { this->e_type_ = ehdr.get_e_type(); } template Sized_relobj_file::~Sized_relobj_file() { } // Set up an object file based on the file header. This sets up the // section information. template void Sized_relobj_file::do_setup() { const unsigned int shnum = this->elf_file_.shnum(); this->set_shnum(shnum); } // Find the SHT_SYMTAB section, given the section headers. The ELF // standard says that maybe in the future there can be more than one // SHT_SYMTAB section. Until somebody figures out how that could // work, we assume there is only one. template void Sized_relobj_file::find_symtab(const unsigned char* pshdrs) { const unsigned int shnum = this->shnum(); this->symtab_shndx_ = 0; if (shnum > 0) { // Look through the sections in reverse order, since gas tends // to put the symbol table at the end. const unsigned char* p = pshdrs + shnum * This::shdr_size; unsigned int i = shnum; unsigned int xindex_shndx = 0; unsigned int xindex_link = 0; while (i > 0) { --i; p -= This::shdr_size; typename This::Shdr shdr(p); if (shdr.get_sh_type() == elfcpp::SHT_SYMTAB) { this->symtab_shndx_ = i; if (xindex_shndx > 0 && xindex_link == i) { Xindex* xindex = new Xindex(this->elf_file_.large_shndx_offset()); xindex->read_symtab_xindex(this, xindex_shndx, pshdrs); this->set_xindex(xindex); } break; } // Try to pick up the SHT_SYMTAB_SHNDX section, if there is // one. This will work if it follows the SHT_SYMTAB // section. if (shdr.get_sh_type() == elfcpp::SHT_SYMTAB_SHNDX) { xindex_shndx = i; xindex_link = this->adjust_shndx(shdr.get_sh_link()); } } } } // Return the Xindex structure to use for object with lots of // sections. template Xindex* Sized_relobj_file::do_initialize_xindex() { gold_assert(this->symtab_shndx_ != -1U); Xindex* xindex = new Xindex(this->elf_file_.large_shndx_offset()); xindex->initialize_symtab_xindex(this, this->symtab_shndx_); return xindex; } // Return whether SHDR has the right type and flags to be a GNU // .eh_frame section. template bool Sized_relobj_file::check_eh_frame_flags( const elfcpp::Shdr* shdr) const { elfcpp::Elf_Word sh_type = shdr->get_sh_type(); return ((sh_type == elfcpp::SHT_PROGBITS || sh_type == parameters->target().unwind_section_type()) && (shdr->get_sh_flags() & elfcpp::SHF_ALLOC) != 0); } // Find the section header with the given name. template const unsigned char* Object::find_shdr( const unsigned char* pshdrs, const char* name, const char* names, section_size_type names_size, const unsigned char* hdr) const { const int shdr_size = elfcpp::Elf_sizes::shdr_size; const unsigned int shnum = this->shnum(); const unsigned char* hdr_end = pshdrs + shdr_size * shnum; size_t sh_name = 0; while (1) { if (hdr) { // We found HDR last time we were called, continue looking. typename elfcpp::Shdr shdr(hdr); sh_name = shdr.get_sh_name(); } else { // Look for the next occurrence of NAME in NAMES. // The fact that .shstrtab produced by current GNU tools is // string merged means we shouldn't have both .not.foo and // .foo in .shstrtab, and multiple .foo sections should all // have the same sh_name. However, this is not guaranteed // by the ELF spec and not all ELF object file producers may // be so clever. size_t len = strlen(name) + 1; const char *p = sh_name ? names + sh_name + len : names; p = reinterpret_cast(memmem(p, names_size - (p - names), name, len)); if (p == NULL) return NULL; sh_name = p - names; hdr = pshdrs; if (sh_name == 0) return hdr; } hdr += shdr_size; while (hdr < hdr_end) { typename elfcpp::Shdr shdr(hdr); if (shdr.get_sh_name() == sh_name) return hdr; hdr += shdr_size; } hdr = NULL; if (sh_name == 0) return hdr; } } // Return whether there is a GNU .eh_frame section, given the section // headers and the section names. template bool Sized_relobj_file::find_eh_frame( const unsigned char* pshdrs, const char* names, section_size_type names_size) const { const unsigned char* s = NULL; while (1) { s = this->template find_shdr(pshdrs, ".eh_frame", names, names_size, s); if (s == NULL) return false; typename This::Shdr shdr(s); if (this->check_eh_frame_flags(&shdr)) return true; } } // Return TRUE if this is a section whose contents will be needed in the // Add_symbols task. This function is only called for sections that have // already passed the test in is_compressed_debug_section() and the debug // section name prefix, ".debug"/".zdebug", has been skipped. static bool need_decompressed_section(const char* name) { if (*name++ != '_') return false; #ifdef ENABLE_THREADS // Decompressing these sections now will help only if we're // multithreaded. if (parameters->options().threads()) { // We will need .zdebug_str if this is not an incremental link // (i.e., we are processing string merge sections) or if we need // to build a gdb index. if ((!parameters->incremental() || parameters->options().gdb_index()) && strcmp(name, "str") == 0) return true; // We will need these other sections when building a gdb index. if (parameters->options().gdb_index() && (strcmp(name, "info") == 0 || strcmp(name, "types") == 0 || strcmp(name, "pubnames") == 0 || strcmp(name, "pubtypes") == 0 || strcmp(name, "ranges") == 0 || strcmp(name, "abbrev") == 0)) return true; } #endif // Even when single-threaded, we will need .zdebug_str if this is // not an incremental link and we are building a gdb index. // Otherwise, we would decompress the section twice: once for // string merge processing, and once for building the gdb index. if (!parameters->incremental() && parameters->options().gdb_index() && strcmp(name, "str") == 0) return true; return false; } // Build a table for any compressed debug sections, mapping each section index // to the uncompressed size and (if needed) the decompressed contents. template Compressed_section_map* build_compressed_section_map( const unsigned char* pshdrs, unsigned int shnum, const char* names, section_size_type names_size, Object* obj, bool decompress_if_needed) { Compressed_section_map* uncompressed_map = new Compressed_section_map(); const unsigned int shdr_size = elfcpp::Elf_sizes::shdr_size; const unsigned char* p = pshdrs + shdr_size; for (unsigned int i = 1; i < shnum; ++i, p += shdr_size) { typename elfcpp::Shdr shdr(p); if (shdr.get_sh_type() == elfcpp::SHT_PROGBITS && (shdr.get_sh_flags() & elfcpp::SHF_ALLOC) == 0) { if (shdr.get_sh_name() >= names_size) { obj->error(_("bad section name offset for section %u: %lu"), i, static_cast(shdr.get_sh_name())); continue; } const char* name = names + shdr.get_sh_name(); bool is_compressed = ((shdr.get_sh_flags() & elfcpp::SHF_COMPRESSED) != 0); bool is_zcompressed = (!is_compressed && is_compressed_debug_section(name)); if (is_zcompressed || is_compressed) { section_size_type len; const unsigned char* contents = obj->section_contents(i, &len, false); uint64_t uncompressed_size; Compressed_section_info info; if (is_zcompressed) { // Skip over the ".zdebug" prefix. name += 7; uncompressed_size = get_uncompressed_size(contents, len); info.addralign = shdr.get_sh_addralign(); } else { // Skip over the ".debug" prefix. name += 6; elfcpp::Chdr chdr(contents); uncompressed_size = chdr.get_ch_size(); info.addralign = chdr.get_ch_addralign(); } info.size = convert_to_section_size_type(uncompressed_size); info.flag = shdr.get_sh_flags(); info.contents = NULL; if (uncompressed_size != -1ULL) { unsigned char* uncompressed_data = NULL; if (decompress_if_needed && need_decompressed_section(name)) { uncompressed_data = new unsigned char[uncompressed_size]; if (decompress_input_section(contents, len, uncompressed_data, uncompressed_size, size, big_endian, shdr.get_sh_flags())) info.contents = uncompressed_data; else delete[] uncompressed_data; } (*uncompressed_map)[i] = info; } } } } return uncompressed_map; } // Stash away info for a number of special sections. // Return true if any of the sections found require local symbols to be read. template bool Sized_relobj_file::do_find_special_sections( Read_symbols_data* sd) { const unsigned char* const pshdrs = sd->section_headers->data(); const unsigned char* namesu = sd->section_names->data(); const char* names = reinterpret_cast(namesu); if (this->find_eh_frame(pshdrs, names, sd->section_names_size)) this->has_eh_frame_ = true; Compressed_section_map* compressed_sections = build_compressed_section_map( pshdrs, this->shnum(), names, sd->section_names_size, this, true); if (compressed_sections != NULL) this->set_compressed_sections(compressed_sections); return (this->has_eh_frame_ || (!parameters->options().relocatable() && parameters->options().gdb_index() && (memmem(names, sd->section_names_size, "debug_info", 11) != NULL || memmem(names, sd->section_names_size, "debug_types", 12) != NULL))); } // Read the sections and symbols from an object file. template void Sized_relobj_file::do_read_symbols(Read_symbols_data* sd) { this->base_read_symbols(sd); } // Read the sections and symbols from an object file. This is common // code for all target-specific overrides of do_read_symbols(). template void Sized_relobj_file::base_read_symbols(Read_symbols_data* sd) { this->read_section_data(&this->elf_file_, sd); const unsigned char* const pshdrs = sd->section_headers->data(); this->find_symtab(pshdrs); bool need_local_symbols = this->do_find_special_sections(sd); sd->symbols = NULL; sd->symbols_size = 0; sd->external_symbols_offset = 0; sd->symbol_names = NULL; sd->symbol_names_size = 0; if (this->symtab_shndx_ == 0) { // No symbol table. Weird but legal. return; } // Get the symbol table section header. typename This::Shdr symtabshdr(pshdrs + this->symtab_shndx_ * This::shdr_size); gold_assert(symtabshdr.get_sh_type() == elfcpp::SHT_SYMTAB); // If this object has a .eh_frame section, or if building a .gdb_index // section and there is debug info, we need all the symbols. // Otherwise we only need the external symbols. While it would be // simpler to just always read all the symbols, I've seen object // files with well over 2000 local symbols, which for a 64-bit // object file format is over 5 pages that we don't need to read // now. const int sym_size = This::sym_size; const unsigned int loccount = symtabshdr.get_sh_info(); this->local_symbol_count_ = loccount; this->local_values_.resize(loccount); section_offset_type locsize = loccount * sym_size; off_t dataoff = symtabshdr.get_sh_offset(); section_size_type datasize = convert_to_section_size_type(symtabshdr.get_sh_size()); off_t extoff = dataoff + locsize; section_size_type extsize = datasize - locsize; off_t readoff = need_local_symbols ? dataoff : extoff; section_size_type readsize = need_local_symbols ? datasize : extsize; if (readsize == 0) { // No external symbols. Also weird but also legal. return; } File_view* fvsymtab = this->get_lasting_view(readoff, readsize, true, false); // Read the section header for the symbol names. unsigned int strtab_shndx = this->adjust_shndx(symtabshdr.get_sh_link()); if (strtab_shndx >= this->shnum()) { this->error(_("invalid symbol table name index: %u"), strtab_shndx); return; } typename This::Shdr strtabshdr(pshdrs + strtab_shndx * This::shdr_size); if (strtabshdr.get_sh_type() != elfcpp::SHT_STRTAB) { this->error(_("symbol table name section has wrong type: %u"), static_cast(strtabshdr.get_sh_type())); return; } // Read the symbol names. File_view* fvstrtab = this->get_lasting_view(strtabshdr.get_sh_offset(), strtabshdr.get_sh_size(), false, true); sd->symbols = fvsymtab; sd->symbols_size = readsize; sd->external_symbols_offset = need_local_symbols ? locsize : 0; sd->symbol_names = fvstrtab; sd->symbol_names_size = convert_to_section_size_type(strtabshdr.get_sh_size()); } // Return the section index of symbol SYM. Set *VALUE to its value in // the object file. Set *IS_ORDINARY if this is an ordinary section // index, not a special code between SHN_LORESERVE and SHN_HIRESERVE. // Note that for a symbol which is not defined in this object file, // this will set *VALUE to 0 and return SHN_UNDEF; it will not return // the final value of the symbol in the link. template unsigned int Sized_relobj_file::symbol_section_and_value(unsigned int sym, Address* value, bool* is_ordinary) { section_size_type symbols_size; const unsigned char* symbols = this->section_contents(this->symtab_shndx_, &symbols_size, false); const size_t count = symbols_size / This::sym_size; gold_assert(sym < count); elfcpp::Sym elfsym(symbols + sym * This::sym_size); *value = elfsym.get_st_value(); return this->adjust_sym_shndx(sym, elfsym.get_st_shndx(), is_ordinary); } // Return whether to include a section group in the link. LAYOUT is // used to keep track of which section groups we have already seen. // INDEX is the index of the section group and SHDR is the section // header. If we do not want to include this group, we set bits in // OMIT for each section which should be discarded. template bool Sized_relobj_file::include_section_group( Symbol_table* symtab, Layout* layout, unsigned int index, const char* name, const unsigned char* shdrs, const char* section_names, section_size_type section_names_size, std::vector* omit) { // Read the section contents. typename This::Shdr shdr(shdrs + index * This::shdr_size); const unsigned char* pcon = this->get_view(shdr.get_sh_offset(), shdr.get_sh_size(), true, false); const elfcpp::Elf_Word* pword = reinterpret_cast(pcon); // The first word contains flags. We only care about COMDAT section // groups. Other section groups are always included in the link // just like ordinary sections. elfcpp::Elf_Word flags = elfcpp::Swap<32, big_endian>::readval(pword); // Look up the group signature, which is the name of a symbol. ELF // uses a symbol name because some group signatures are long, and // the name is generally already in the symbol table, so it makes // sense to put the long string just once in .strtab rather than in // both .strtab and .shstrtab. // Get the appropriate symbol table header (this will normally be // the single SHT_SYMTAB section, but in principle it need not be). const unsigned int link = this->adjust_shndx(shdr.get_sh_link()); typename This::Shdr symshdr(this, this->elf_file_.section_header(link)); // Read the symbol table entry. unsigned int symndx = shdr.get_sh_info(); if (symndx >= symshdr.get_sh_size() / This::sym_size) { this->error(_("section group %u info %u out of range"), index, symndx); return false; } off_t symoff = symshdr.get_sh_offset() + symndx * This::sym_size; const unsigned char* psym = this->get_view(symoff, This::sym_size, true, false); elfcpp::Sym sym(psym); // Read the symbol table names. section_size_type symnamelen; const unsigned char* psymnamesu; psymnamesu = this->section_contents(this->adjust_shndx(symshdr.get_sh_link()), &symnamelen, true); const char* psymnames = reinterpret_cast(psymnamesu); // Get the section group signature. if (sym.get_st_name() >= symnamelen) { this->error(_("symbol %u name offset %u out of range"), symndx, sym.get_st_name()); return false; } std::string signature(psymnames + sym.get_st_name()); // It seems that some versions of gas will create a section group // associated with a section symbol, and then fail to give a name to // the section symbol. In such a case, use the name of the section. if (signature[0] == '\0' && sym.get_st_type() == elfcpp::STT_SECTION) { bool is_ordinary; unsigned int sym_shndx = this->adjust_sym_shndx(symndx, sym.get_st_shndx(), &is_ordinary); if (!is_ordinary || sym_shndx >= this->shnum()) { this->error(_("symbol %u invalid section index %u"), symndx, sym_shndx); return false; } typename This::Shdr member_shdr(shdrs + sym_shndx * This::shdr_size); if (member_shdr.get_sh_name() < section_names_size) signature = section_names + member_shdr.get_sh_name(); } // Record this section group in the layout, and see whether we've already // seen one with the same signature. bool include_group; bool is_comdat; Kept_section* kept_section = NULL; if ((flags & elfcpp::GRP_COMDAT) == 0) { include_group = true; is_comdat = false; } else { include_group = layout->find_or_add_kept_section(signature, this, index, true, true, &kept_section); is_comdat = true; } if (is_comdat && include_group) { Incremental_inputs* incremental_inputs = layout->incremental_inputs(); if (incremental_inputs != NULL) incremental_inputs->report_comdat_group(this, signature.c_str()); } size_t count = shdr.get_sh_size() / sizeof(elfcpp::Elf_Word); std::vector shndxes; bool relocate_group = include_group && parameters->options().relocatable(); if (relocate_group) shndxes.reserve(count - 1); for (size_t i = 1; i < count; ++i) { elfcpp::Elf_Word shndx = this->adjust_shndx(elfcpp::Swap<32, big_endian>::readval(pword + i)); if (relocate_group) shndxes.push_back(shndx); if (shndx >= this->shnum()) { this->error(_("section %u in section group %u out of range"), shndx, index); continue; } // Check for an earlier section number, since we're going to get // it wrong--we may have already decided to include the section. if (shndx < index) this->error(_("invalid section group %u refers to earlier section %u"), index, shndx); // Get the name of the member section. typename This::Shdr member_shdr(shdrs + shndx * This::shdr_size); if (member_shdr.get_sh_name() >= section_names_size) { // This is an error, but it will be diagnosed eventually // in do_layout, so we don't need to do anything here but // ignore it. continue; } std::string mname(section_names + member_shdr.get_sh_name()); if (include_group) { if (is_comdat) kept_section->add_comdat_section(mname, shndx, member_shdr.get_sh_size()); } else { (*omit)[shndx] = true; // Store a mapping from this section to the Kept_section // information for the group. This mapping is used for // relocation processing and diagnostics. // If the kept section is a linkonce section, we don't // bother with it unless the comdat group contains just // a single section, making it easy to match up. if (is_comdat && (kept_section->is_comdat() || count == 2)) this->set_kept_comdat_section(shndx, true, symndx, member_shdr.get_sh_size(), kept_section); } } if (relocate_group) layout->layout_group(symtab, this, index, name, signature.c_str(), shdr, flags, &shndxes); return include_group; } // Whether to include a linkonce section in the link. NAME is the // name of the section and SHDR is the section header. // Linkonce sections are a GNU extension implemented in the original // GNU linker before section groups were defined. The semantics are // that we only include one linkonce section with a given name. The // name of a linkonce section is normally .gnu.linkonce.T.SYMNAME, // where T is the type of section and SYMNAME is the name of a symbol. // In an attempt to make linkonce sections interact well with section // groups, we try to identify SYMNAME and use it like a section group // signature. We want to block section groups with that signature, // but not other linkonce sections with that signature. We also use // the full name of the linkonce section as a normal section group // signature. template bool Sized_relobj_file::include_linkonce_section( Layout* layout, unsigned int index, const char* name, const elfcpp::Shdr& shdr) { typename elfcpp::Elf_types::Elf_WXword sh_size = shdr.get_sh_size(); // In general the symbol name we want will be the string following // the last '.'. However, we have to handle the case of // .gnu.linkonce.t.__i686.get_pc_thunk.bx, which was generated by // some versions of gcc. So we use a heuristic: if the name starts // with ".gnu.linkonce.t.", we use everything after that. Otherwise // we look for the last '.'. We can't always simply skip // ".gnu.linkonce.X", because we have to deal with cases like // ".gnu.linkonce.d.rel.ro.local". const char* const linkonce_t = ".gnu.linkonce.t."; const char* symname; if (strncmp(name, linkonce_t, strlen(linkonce_t)) == 0) symname = name + strlen(linkonce_t); else symname = strrchr(name, '.') + 1; std::string sig1(symname); std::string sig2(name); Kept_section* kept1; Kept_section* kept2; bool include1 = layout->find_or_add_kept_section(sig1, this, index, false, false, &kept1); bool include2 = layout->find_or_add_kept_section(sig2, this, index, false, true, &kept2); if (!include2) { // We are not including this section because we already saw the // name of the section as a signature. This normally implies // that the kept section is another linkonce section. If it is // the same size, record it as the section which corresponds to // this one. if (kept2->object() != NULL && !kept2->is_comdat()) this->set_kept_comdat_section(index, false, 0, sh_size, kept2); } else if (!include1) { // The section is being discarded on the basis of its symbol // name. This means that the corresponding kept section was // part of a comdat group, and it will be difficult to identify // the specific section within that group that corresponds to // this linkonce section. We'll handle the simple case where // the group has only one member section. Otherwise, it's not // worth the effort. if (kept1->object() != NULL && kept1->is_comdat()) this->set_kept_comdat_section(index, false, 0, sh_size, kept1); } else { kept1->set_linkonce_size(sh_size); kept2->set_linkonce_size(sh_size); } return include1 && include2; } // Layout an input section. template inline void Sized_relobj_file::layout_section( Layout* layout, unsigned int shndx, const char* name, const typename This::Shdr& shdr, unsigned int sh_type, unsigned int reloc_shndx, unsigned int reloc_type) { off_t offset; Output_section* os = layout->layout(this, shndx, name, shdr, sh_type, reloc_shndx, reloc_type, &offset); this->output_sections()[shndx] = os; if (offset == -1) this->section_offsets()[shndx] = invalid_address; else this->section_offsets()[shndx] = convert_types(offset); // If this section requires special handling, and if there are // relocs that apply to it, then we must do the special handling // before we apply the relocs. if (offset == -1 && reloc_shndx != 0) this->set_relocs_must_follow_section_writes(); } // Layout an input .eh_frame section. template void Sized_relobj_file::layout_eh_frame_section( Layout* layout, const unsigned char* symbols_data, section_size_type symbols_size, const unsigned char* symbol_names_data, section_size_type symbol_names_size, unsigned int shndx, const typename This::Shdr& shdr, unsigned int reloc_shndx, unsigned int reloc_type) { gold_assert(this->has_eh_frame_); off_t offset; Output_section* os = layout->layout_eh_frame(this, symbols_data, symbols_size, symbol_names_data, symbol_names_size, shndx, shdr, reloc_shndx, reloc_type, &offset); this->output_sections()[shndx] = os; if (os == NULL || offset == -1) this->section_offsets()[shndx] = invalid_address; else this->section_offsets()[shndx] = convert_types(offset); // If this section requires special handling, and if there are // relocs that aply to it, then we must do the special handling // before we apply the relocs. if (os != NULL && offset == -1 && reloc_shndx != 0) this->set_relocs_must_follow_section_writes(); } // Layout an input .note.gnu.property section. // This note section has an *extremely* non-standard layout. // The gABI spec says that ELF-64 files should have 8-byte fields and // 8-byte alignment in the note section, but the Gnu tools generally // use 4-byte fields and 4-byte alignment (see the comment for // Layout::create_note). This section uses 4-byte fields (i.e., // namesz, descsz, and type are always 4 bytes), the name field is // padded to a multiple of 4 bytes, but the desc field is padded // to a multiple of 4 or 8 bytes, depending on the ELF class. // The individual properties within the desc field always use // 4-byte pr_type and pr_datasz fields, but pr_data is padded to // a multiple of 4 or 8 bytes, depending on the ELF class. template void Sized_relobj_file::layout_gnu_property_section( Layout* layout, unsigned int shndx) { section_size_type contents_len; const unsigned char* pcontents = this->section_contents(shndx, &contents_len, false); const unsigned char* pcontents_end = pcontents + contents_len; // Loop over all the notes in this section. while (pcontents < pcontents_end) { if (pcontents + 16 > pcontents_end) { gold_warning(_("%s: corrupt .note.gnu.property section " "(note too short)"), this->name().c_str()); return; } size_t namesz = elfcpp::Swap<32, big_endian>::readval(pcontents); size_t descsz = elfcpp::Swap<32, big_endian>::readval(pcontents + 4); unsigned int ntype = elfcpp::Swap<32, big_endian>::readval(pcontents + 8); const unsigned char* pname = pcontents + 12; if (namesz != 4 || strcmp(reinterpret_cast(pname), "GNU") != 0) { gold_warning(_("%s: corrupt .note.gnu.property section " "(name is not 'GNU')"), this->name().c_str()); return; } if (ntype != elfcpp::NT_GNU_PROPERTY_TYPE_0) { gold_warning(_("%s: unsupported note type %d " "in .note.gnu.property section"), this->name().c_str(), ntype); return; } size_t aligned_namesz = align_address(namesz, 4); const unsigned char* pdesc = pname + aligned_namesz; if (pdesc + descsz > pcontents + contents_len) { gold_warning(_("%s: corrupt .note.gnu.property section"), this->name().c_str()); return; } const unsigned char* pprop = pdesc; // Loop over the program properties in this note. while (pprop < pdesc + descsz) { if (pprop + 8 > pdesc + descsz) { gold_warning(_("%s: corrupt .note.gnu.property section"), this->name().c_str()); return; } unsigned int pr_type = elfcpp::Swap<32, big_endian>::readval(pprop); size_t pr_datasz = elfcpp::Swap<32, big_endian>::readval(pprop + 4); pprop += 8; if (pprop + pr_datasz > pdesc + descsz) { gold_warning(_("%s: corrupt .note.gnu.property section"), this->name().c_str()); return; } layout->layout_gnu_property(ntype, pr_type, pr_datasz, pprop, this); pprop += align_address(pr_datasz, size / 8); } pcontents = pdesc + align_address(descsz, size / 8); } } // This a copy of lto_section defined in GCC (lto-streamer.h) struct lto_section { int16_t major_version; int16_t minor_version; unsigned char slim_object; /* Flags is a private field that is not defined publicly. */ uint16_t flags; }; // Lay out the input sections. We walk through the sections and check // whether they should be included in the link. If they should, we // pass them to the Layout object, which will return an output section // and an offset. // This function is called twice sometimes, two passes, when mapping // of input sections to output sections must be delayed. // This is true for the following : // * Garbage collection (--gc-sections): Some input sections will be // discarded and hence the assignment must wait until the second pass. // In the first pass, it is for setting up some sections as roots to // a work-list for --gc-sections and to do comdat processing. // * Identical Code Folding (--icf=): Some input sections // will be folded and hence the assignment must wait. // * Using plugins to map some sections to unique segments: Mapping // some sections to unique segments requires mapping them to unique // output sections too. This can be done via plugins now and this // information is not available in the first pass. template void Sized_relobj_file::do_layout(Symbol_table* symtab, Layout* layout, Read_symbols_data* sd) { const unsigned int unwind_section_type = parameters->target().unwind_section_type(); const unsigned int shnum = this->shnum(); /* Should this function be called twice? */ bool is_two_pass = (parameters->options().gc_sections() || parameters->options().icf_enabled() || layout->is_unique_segment_for_sections_specified()); /* Only one of is_pass_one and is_pass_two is true. Both are false when a two-pass approach is not needed. */ bool is_pass_one = false; bool is_pass_two = false; Symbols_data* gc_sd = NULL; /* Check if do_layout needs to be two-pass. If so, find out which pass should happen. In the first pass, the data in sd is saved to be used later in the second pass. */ if (is_two_pass) { gc_sd = this->get_symbols_data(); if (gc_sd == NULL) { gold_assert(sd != NULL); is_pass_one = true; } else { if (parameters->options().gc_sections()) gold_assert(symtab->gc()->is_worklist_ready()); if (parameters->options().icf_enabled()) gold_assert(symtab->icf()->is_icf_ready()); is_pass_two = true; } } if (shnum == 0) return; if (is_pass_one) { // During garbage collection save the symbols data to use it when // re-entering this function. gc_sd = new Symbols_data; this->copy_symbols_data(gc_sd, sd, This::shdr_size * shnum); this->set_symbols_data(gc_sd); } const unsigned char* section_headers_data = NULL; section_size_type section_names_size; const unsigned char* symbols_data = NULL; section_size_type symbols_size; const unsigned char* symbol_names_data = NULL; section_size_type symbol_names_size; if (is_two_pass) { section_headers_data = gc_sd->section_headers_data; section_names_size = gc_sd->section_names_size; symbols_data = gc_sd->symbols_data; symbols_size = gc_sd->symbols_size; symbol_names_data = gc_sd->symbol_names_data; symbol_names_size = gc_sd->symbol_names_size; } else { section_headers_data = sd->section_headers->data(); section_names_size = sd->section_names_size; if (sd->symbols != NULL) symbols_data = sd->symbols->data(); symbols_size = sd->symbols_size; if (sd->symbol_names != NULL) symbol_names_data = sd->symbol_names->data(); symbol_names_size = sd->symbol_names_size; } // Get the section headers. const unsigned char* shdrs = section_headers_data; const unsigned char* pshdrs; // Get the section names. const unsigned char* pnamesu = (is_two_pass ? gc_sd->section_names_data : sd->section_names->data()); const char* pnames = reinterpret_cast(pnamesu); // If any input files have been claimed by plugins, we need to defer // actual layout until the replacement files have arrived. const bool should_defer_layout = (parameters->options().has_plugins() && parameters->options().plugins()->should_defer_layout()); unsigned int num_sections_to_defer = 0; // For each section, record the index of the reloc section if any. // Use 0 to mean that there is no reloc section, -1U to mean that // there is more than one. std::vector reloc_shndx(shnum, 0); std::vector reloc_type(shnum, elfcpp::SHT_NULL); // Skip the first, dummy, section. pshdrs = shdrs + This::shdr_size; for (unsigned int i = 1; i < shnum; ++i, pshdrs += This::shdr_size) { typename This::Shdr shdr(pshdrs); // Count the number of sections whose layout will be deferred. if (should_defer_layout && (shdr.get_sh_flags() & elfcpp::SHF_ALLOC)) ++num_sections_to_defer; unsigned int sh_type = shdr.get_sh_type(); if (sh_type == elfcpp::SHT_REL || sh_type == elfcpp::SHT_RELA) { unsigned int target_shndx = this->adjust_shndx(shdr.get_sh_info()); if (target_shndx == 0 || target_shndx >= shnum) { this->error(_("relocation section %u has bad info %u"), i, target_shndx); continue; } if (reloc_shndx[target_shndx] != 0) reloc_shndx[target_shndx] = -1U; else { reloc_shndx[target_shndx] = i; reloc_type[target_shndx] = sh_type; } } } Output_sections& out_sections(this->output_sections()); std::vector
& out_section_offsets(this->section_offsets()); if (!is_pass_two) { out_sections.resize(shnum); out_section_offsets.resize(shnum); } // If we are only linking for symbols, then there is nothing else to // do here. if (this->input_file()->just_symbols()) { if (!is_pass_two) { delete sd->section_headers; sd->section_headers = NULL; delete sd->section_names; sd->section_names = NULL; } return; } if (num_sections_to_defer > 0) { parameters->options().plugins()->add_deferred_layout_object(this); this->deferred_layout_.reserve(num_sections_to_defer); this->is_deferred_layout_ = true; } // Whether we've seen a .note.GNU-stack section. bool seen_gnu_stack = false; // The flags of a .note.GNU-stack section. uint64_t gnu_stack_flags = 0; // Keep track of which sections to omit. std::vector omit(shnum, false); // Keep track of reloc sections when emitting relocations. const bool relocatable = parameters->options().relocatable(); const bool emit_relocs = (relocatable || parameters->options().emit_relocs()); std::vector reloc_sections; // Keep track of .eh_frame sections. std::vector eh_frame_sections; // Keep track of .debug_info and .debug_types sections. std::vector debug_info_sections; std::vector debug_types_sections; // Skip the first, dummy, section. pshdrs = shdrs + This::shdr_size; for (unsigned int i = 1; i < shnum; ++i, pshdrs += This::shdr_size) { typename This::Shdr shdr(pshdrs); const unsigned int sh_name = shdr.get_sh_name(); unsigned int sh_type = shdr.get_sh_type(); if (sh_name >= section_names_size) { this->error(_("bad section name offset for section %u: %lu"), i, static_cast(sh_name)); return; } const char* name = pnames + sh_name; if (!is_pass_two) { if (this->handle_gnu_warning_section(name, i, symtab)) { if (!relocatable && !parameters->options().shared()) omit[i] = true; } // The .note.GNU-stack section is special. It gives the // protection flags that this object file requires for the stack // in memory. if (strcmp(name, ".note.GNU-stack") == 0) { seen_gnu_stack = true; gnu_stack_flags |= shdr.get_sh_flags(); omit[i] = true; } // The .note.GNU-split-stack section is also special. It // indicates that the object was compiled with // -fsplit-stack. if (this->handle_split_stack_section(name)) { if (!relocatable && !parameters->options().shared()) omit[i] = true; } // Skip attributes section. if (parameters->target().is_attributes_section(name)) { omit[i] = true; } // Handle .note.gnu.property sections. if (sh_type == elfcpp::SHT_NOTE && strcmp(name, ".note.gnu.property") == 0) { this->layout_gnu_property_section(layout, i); omit[i] = true; } bool discard = omit[i]; if (!discard) { if (sh_type == elfcpp::SHT_GROUP) { if (!this->include_section_group(symtab, layout, i, name, shdrs, pnames, section_names_size, &omit)) discard = true; } else if ((shdr.get_sh_flags() & elfcpp::SHF_GROUP) == 0 && Layout::is_linkonce(name)) { if (!this->include_linkonce_section(layout, i, name, shdr)) discard = true; } } // Add the section to the incremental inputs layout. Incremental_inputs* incremental_inputs = layout->incremental_inputs(); if (incremental_inputs != NULL && !discard && can_incremental_update(sh_type)) { off_t sh_size = shdr.get_sh_size(); section_size_type uncompressed_size; if (this->section_is_compressed(i, &uncompressed_size)) sh_size = uncompressed_size; incremental_inputs->report_input_section(this, i, name, sh_size); } if (discard) { // Do not include this section in the link. out_sections[i] = NULL; out_section_offsets[i] = invalid_address; continue; } } if (is_pass_one && parameters->options().gc_sections()) { if (this->is_section_name_included(name) || layout->keep_input_section (this, name) || sh_type == elfcpp::SHT_INIT_ARRAY || sh_type == elfcpp::SHT_FINI_ARRAY) { symtab->gc()->worklist().push_back(Section_id(this, i)); } // If the section name XXX can be represented as a C identifier // it cannot be discarded if there are references to // __start_XXX and __stop_XXX symbols. These need to be // specially handled. if (is_cident(name)) { symtab->gc()->add_cident_section(name, Section_id(this, i)); } } // When doing a relocatable link we are going to copy input // reloc sections into the output. We only want to copy the // ones associated with sections which are not being discarded. // However, we don't know that yet for all sections. So save // reloc sections and process them later. Garbage collection is // not triggered when relocatable code is desired. if (emit_relocs && (sh_type == elfcpp::SHT_REL || sh_type == elfcpp::SHT_RELA)) { reloc_sections.push_back(i); continue; } if (relocatable && sh_type == elfcpp::SHT_GROUP) continue; // The .eh_frame section is special. It holds exception frame // information that we need to read in order to generate the // exception frame header. We process these after all the other // sections so that the exception frame reader can reliably // determine which sections are being discarded, and discard the // corresponding information. if (this->check_eh_frame_flags(&shdr) && strcmp(name, ".eh_frame") == 0) { // If the target has a special unwind section type, let's // canonicalize it here. sh_type = unwind_section_type; if (!relocatable) { if (is_pass_one) { if (this->is_deferred_layout()) out_sections[i] = reinterpret_cast(2); else out_sections[i] = reinterpret_cast(1); out_section_offsets[i] = invalid_address; } else if (this->is_deferred_layout()) { out_sections[i] = reinterpret_cast(2); out_section_offsets[i] = invalid_address; this->deferred_layout_.push_back( Deferred_layout(i, name, sh_type, pshdrs, reloc_shndx[i], reloc_type[i])); } else eh_frame_sections.push_back(i); continue; } } if (is_pass_two && parameters->options().gc_sections()) { // This is executed during the second pass of garbage // collection. do_layout has been called before and some // sections have been already discarded. Simply ignore // such sections this time around. if (out_sections[i] == NULL) { gold_assert(out_section_offsets[i] == invalid_address); continue; } if (((shdr.get_sh_flags() & elfcpp::SHF_ALLOC) != 0) && symtab->gc()->is_section_garbage(this, i)) { if (parameters->options().print_gc_sections()) gold_info(_("%s: removing unused section from '%s'" " in file '%s'"), program_name, this->section_name(i).c_str(), this->name().c_str()); out_sections[i] = NULL; out_section_offsets[i] = invalid_address; continue; } } if (is_pass_two && parameters->options().icf_enabled()) { if (out_sections[i] == NULL) { gold_assert(out_section_offsets[i] == invalid_address); continue; } if (((shdr.get_sh_flags() & elfcpp::SHF_ALLOC) != 0) && symtab->icf()->is_section_folded(this, i)) { if (parameters->options().print_icf_sections()) { Section_id folded = symtab->icf()->get_folded_section(this, i); Relobj* folded_obj = reinterpret_cast(folded.first); gold_info(_("%s: ICF folding section '%s' in file '%s' " "into '%s' in file '%s'"), program_name, this->section_name(i).c_str(), this->name().c_str(), folded_obj->section_name(folded.second).c_str(), folded_obj->name().c_str()); } out_sections[i] = NULL; out_section_offsets[i] = invalid_address; continue; } } // Defer layout here if input files are claimed by plugins. When gc // is turned on this function is called twice; we only want to do this // on the first pass. if (!is_pass_two && this->is_deferred_layout() && (shdr.get_sh_flags() & elfcpp::SHF_ALLOC)) { this->deferred_layout_.push_back(Deferred_layout(i, name, sh_type, pshdrs, reloc_shndx[i], reloc_type[i])); // Put dummy values here; real values will be supplied by // do_layout_deferred_sections. out_sections[i] = reinterpret_cast(2); out_section_offsets[i] = invalid_address; continue; } // During gc_pass_two if a section that was previously deferred is // found, do not layout the section as layout_deferred_sections will // do it later from gold.cc. if (is_pass_two && (out_sections[i] == reinterpret_cast(2))) continue; if (is_pass_one) { // This is during garbage collection. The out_sections are // assigned in the second call to this function. out_sections[i] = reinterpret_cast(1); out_section_offsets[i] = invalid_address; } else { // When garbage collection is switched on the actual layout // only happens in the second call. this->layout_section(layout, i, name, shdr, sh_type, reloc_shndx[i], reloc_type[i]); // When generating a .gdb_index section, we do additional // processing of .debug_info and .debug_types sections after all // the other sections for the same reason as above. if (!relocatable && parameters->options().gdb_index() && !(shdr.get_sh_flags() & elfcpp::SHF_ALLOC)) { if (strcmp(name, ".debug_info") == 0 || strcmp(name, ".zdebug_info") == 0) debug_info_sections.push_back(i); else if (strcmp(name, ".debug_types") == 0 || strcmp(name, ".zdebug_types") == 0) debug_types_sections.push_back(i); } } /* GCC uses .gnu.lto_.lto. as a LTO bytecode information section. */ const char *lto_section_name = ".gnu.lto_.lto."; if (strncmp (name, lto_section_name, strlen (lto_section_name)) == 0) { section_size_type contents_len; const unsigned char* pcontents = this->section_contents(i, &contents_len, false); struct lto_section lsection = *(const lto_section*)pcontents; if (lsection.slim_object) gold_info(_("%s: plugin needed to handle lto object"), this->name().c_str()); } } if (!is_pass_two) { layout->merge_gnu_properties(this); layout->layout_gnu_stack(seen_gnu_stack, gnu_stack_flags, this); } // Handle the .eh_frame sections after the other sections. gold_assert(!is_pass_one || eh_frame_sections.empty()); for (std::vector::const_iterator p = eh_frame_sections.begin(); p != eh_frame_sections.end(); ++p) { unsigned int i = *p; const unsigned char* pshdr; pshdr = section_headers_data + i * This::shdr_size; typename This::Shdr shdr(pshdr); this->layout_eh_frame_section(layout, symbols_data, symbols_size, symbol_names_data, symbol_names_size, i, shdr, reloc_shndx[i], reloc_type[i]); } // When doing a relocatable link handle the reloc sections at the // end. Garbage collection and Identical Code Folding is not // turned on for relocatable code. if (emit_relocs) this->size_relocatable_relocs(); gold_assert(!is_two_pass || reloc_sections.empty()); for (std::vector::const_iterator p = reloc_sections.begin(); p != reloc_sections.end(); ++p) { unsigned int i = *p; const unsigned char* pshdr; pshdr = section_headers_data + i * This::shdr_size; typename This::Shdr shdr(pshdr); unsigned int data_shndx = this->adjust_shndx(shdr.get_sh_info()); if (data_shndx >= shnum) { // We already warned about this above. continue; } Output_section* data_section = out_sections[data_shndx]; if (data_section == reinterpret_cast(2)) { if (is_pass_two) continue; // The layout for the data section was deferred, so we need // to defer the relocation section, too. const char* name = pnames + shdr.get_sh_name(); this->deferred_layout_relocs_.push_back( Deferred_layout(i, name, shdr.get_sh_type(), pshdr, 0, elfcpp::SHT_NULL)); out_sections[i] = reinterpret_cast(2); out_section_offsets[i] = invalid_address; continue; } if (data_section == NULL) { out_sections[i] = NULL; out_section_offsets[i] = invalid_address; continue; } Relocatable_relocs* rr = new Relocatable_relocs(); this->set_relocatable_relocs(i, rr); Output_section* os = layout->layout_reloc(this, i, shdr, data_section, rr); out_sections[i] = os; out_section_offsets[i] = invalid_address; } // When building a .gdb_index section, scan the .debug_info and // .debug_types sections. gold_assert(!is_pass_one || (debug_info_sections.empty() && debug_types_sections.empty())); for (std::vector::const_iterator p = debug_info_sections.begin(); p != debug_info_sections.end(); ++p) { unsigned int i = *p; layout->add_to_gdb_index(false, this, symbols_data, symbols_size, i, reloc_shndx[i], reloc_type[i]); } for (std::vector::const_iterator p = debug_types_sections.begin(); p != debug_types_sections.end(); ++p) { unsigned int i = *p; layout->add_to_gdb_index(true, this, symbols_data, symbols_size, i, reloc_shndx[i], reloc_type[i]); } if (is_pass_two) { delete[] gc_sd->section_headers_data; delete[] gc_sd->section_names_data; delete[] gc_sd->symbols_data; delete[] gc_sd->symbol_names_data; this->set_symbols_data(NULL); } else { delete sd->section_headers; sd->section_headers = NULL; delete sd->section_names; sd->section_names = NULL; } } // Layout sections whose layout was deferred while waiting for // input files from a plugin. template void Sized_relobj_file::do_layout_deferred_sections(Layout* layout) { typename std::vector::iterator deferred; for (deferred = this->deferred_layout_.begin(); deferred != this->deferred_layout_.end(); ++deferred) { typename This::Shdr shdr(deferred->shdr_data_); if (!parameters->options().relocatable() && deferred->name_ == ".eh_frame" && this->check_eh_frame_flags(&shdr)) { // Checking is_section_included is not reliable for // .eh_frame sections, because they do not have an output // section. This is not a problem normally because we call // layout_eh_frame_section unconditionally, but when // deferring sections that is not true. We don't want to // keep all .eh_frame sections because that will cause us to // keep all sections that they refer to, which is the wrong // way around. Instead, the eh_frame code will discard // .eh_frame sections that refer to discarded sections. // Reading the symbols again here may be slow. Read_symbols_data sd; this->base_read_symbols(&sd); this->layout_eh_frame_section(layout, sd.symbols->data(), sd.symbols_size, sd.symbol_names->data(), sd.symbol_names_size, deferred->shndx_, shdr, deferred->reloc_shndx_, deferred->reloc_type_); continue; } // If the section is not included, it is because the garbage collector // decided it is not needed. Avoid reverting that decision. if (!this->is_section_included(deferred->shndx_)) continue; this->layout_section(layout, deferred->shndx_, deferred->name_.c_str(), shdr, shdr.get_sh_type(), deferred->reloc_shndx_, deferred->reloc_type_); } this->deferred_layout_.clear(); // Now handle the deferred relocation sections. Output_sections& out_sections(this->output_sections()); std::vector
& out_section_offsets(this->section_offsets()); for (deferred = this->deferred_layout_relocs_.begin(); deferred != this->deferred_layout_relocs_.end(); ++deferred) { unsigned int shndx = deferred->shndx_; typename This::Shdr shdr(deferred->shdr_data_); unsigned int data_shndx = this->adjust_shndx(shdr.get_sh_info()); Output_section* data_section = out_sections[data_shndx]; if (data_section == NULL) { out_sections[shndx] = NULL; out_section_offsets[shndx] = invalid_address; continue; } Relocatable_relocs* rr = new Relocatable_relocs(); this->set_relocatable_relocs(shndx, rr); Output_section* os = layout->layout_reloc(this, shndx, shdr, data_section, rr); out_sections[shndx] = os; out_section_offsets[shndx] = invalid_address; } } // Add the symbols to the symbol table. template void Sized_relobj_file::do_add_symbols(Symbol_table* symtab, Read_symbols_data* sd, Layout* layout) { if (sd->symbols == NULL) { gold_assert(sd->symbol_names == NULL); return; } const int sym_size = This::sym_size; size_t symcount = ((sd->symbols_size - sd->external_symbols_offset) / sym_size); if (symcount * sym_size != sd->symbols_size - sd->external_symbols_offset) { this->error(_("size of symbols is not multiple of symbol size")); return; } this->symbols_.resize(symcount); if (layout->is_lto_slim_object ()) gold_info(_("%s: plugin needed to handle lto object"), this->name().c_str()); const char* sym_names = reinterpret_cast(sd->symbol_names->data()); symtab->add_from_relobj(this, sd->symbols->data() + sd->external_symbols_offset, symcount, this->local_symbol_count_, sym_names, sd->symbol_names_size, &this->symbols_, &this->defined_count_); delete sd->symbols; sd->symbols = NULL; delete sd->symbol_names; sd->symbol_names = NULL; } // Find out if this object, that is a member of a lib group, should be included // in the link. We check every symbol defined by this object. If the symbol // table has a strong undefined reference to that symbol, we have to include // the object. template Archive::Should_include Sized_relobj_file::do_should_include_member( Symbol_table* symtab, Layout* layout, Read_symbols_data* sd, std::string* why) { char* tmpbuf = NULL; size_t tmpbuflen = 0; const char* sym_names = reinterpret_cast(sd->symbol_names->data()); const unsigned char* syms = sd->symbols->data() + sd->external_symbols_offset; const int sym_size = elfcpp::Elf_sizes::sym_size; size_t symcount = ((sd->symbols_size - sd->external_symbols_offset) / sym_size); const unsigned char* p = syms; for (size_t i = 0; i < symcount; ++i, p += sym_size) { elfcpp::Sym sym(p); unsigned int st_shndx = sym.get_st_shndx(); if (st_shndx == elfcpp::SHN_UNDEF) continue; unsigned int st_name = sym.get_st_name(); const char* name = sym_names + st_name; Symbol* symbol; Archive::Should_include t = Archive::should_include_member(symtab, layout, name, &symbol, why, &tmpbuf, &tmpbuflen); if (t == Archive::SHOULD_INCLUDE_YES) { if (tmpbuf != NULL) free(tmpbuf); return t; } } if (tmpbuf != NULL) free(tmpbuf); return Archive::SHOULD_INCLUDE_UNKNOWN; } // Iterate over global defined symbols, calling a visitor class V for each. template void Sized_relobj_file::do_for_all_global_symbols( Read_symbols_data* sd, Library_base::Symbol_visitor_base* v) { const char* sym_names = reinterpret_cast(sd->symbol_names->data()); const unsigned char* syms = sd->symbols->data() + sd->external_symbols_offset; const int sym_size = elfcpp::Elf_sizes::sym_size; size_t symcount = ((sd->symbols_size - sd->external_symbols_offset) / sym_size); const unsigned char* p = syms; for (size_t i = 0; i < symcount; ++i, p += sym_size) { elfcpp::Sym sym(p); if (sym.get_st_shndx() != elfcpp::SHN_UNDEF) v->visit(sym_names + sym.get_st_name()); } } // Return whether the local symbol SYMNDX has a PLT offset. template bool Sized_relobj_file::local_has_plt_offset( unsigned int symndx) const { typename Local_plt_offsets::const_iterator p = this->local_plt_offsets_.find(symndx); return p != this->local_plt_offsets_.end(); } // Get the PLT offset of a local symbol. template unsigned int Sized_relobj_file::do_local_plt_offset( unsigned int symndx) const { typename Local_plt_offsets::const_iterator p = this->local_plt_offsets_.find(symndx); gold_assert(p != this->local_plt_offsets_.end()); return p->second; } // Set the PLT offset of a local symbol. template void Sized_relobj_file::set_local_plt_offset( unsigned int symndx, unsigned int plt_offset) { std::pair ins = this->local_plt_offsets_.insert(std::make_pair(symndx, plt_offset)); gold_assert(ins.second); } // First pass over the local symbols. Here we add their names to // *POOL and *DYNPOOL, and we store the symbol value in // THIS->LOCAL_VALUES_. This function is always called from a // singleton thread. This is followed by a call to // finalize_local_symbols. template void Sized_relobj_file::do_count_local_symbols(Stringpool* pool, Stringpool* dynpool) { gold_assert(this->symtab_shndx_ != -1U); if (this->symtab_shndx_ == 0) { // This object has no symbols. Weird but legal. return; } // Read the symbol table section header. const unsigned int symtab_shndx = this->symtab_shndx_; typename This::Shdr symtabshdr(this, this->elf_file_.section_header(symtab_shndx)); gold_assert(symtabshdr.get_sh_type() == elfcpp::SHT_SYMTAB); // Read the local symbols. const int sym_size = This::sym_size; const unsigned int loccount = this->local_symbol_count_; gold_assert(loccount == symtabshdr.get_sh_info()); off_t locsize = loccount * sym_size; const unsigned char* psyms = this->get_view(symtabshdr.get_sh_offset(), locsize, true, true); // Read the symbol names. const unsigned int strtab_shndx = this->adjust_shndx(symtabshdr.get_sh_link()); section_size_type strtab_size; const unsigned char* pnamesu = this->section_contents(strtab_shndx, &strtab_size, true); const char* pnames = reinterpret_cast(pnamesu); // Loop over the local symbols. const Output_sections& out_sections(this->output_sections()); std::vector
& out_section_offsets(this->section_offsets()); unsigned int shnum = this->shnum(); unsigned int count = 0; unsigned int dyncount = 0; // Skip the first, dummy, symbol. psyms += sym_size; bool strip_all = parameters->options().strip_all(); bool discard_all = parameters->options().discard_all(); bool discard_locals = parameters->options().discard_locals(); bool discard_sec_merge = parameters->options().discard_sec_merge(); for (unsigned int i = 1; i < loccount; ++i, psyms += sym_size) { elfcpp::Sym sym(psyms); Symbol_value& lv(this->local_values_[i]); bool is_ordinary; unsigned int shndx = this->adjust_sym_shndx(i, sym.get_st_shndx(), &is_ordinary); lv.set_input_shndx(shndx, is_ordinary); if (sym.get_st_type() == elfcpp::STT_SECTION) lv.set_is_section_symbol(); else if (sym.get_st_type() == elfcpp::STT_TLS) lv.set_is_tls_symbol(); else if (sym.get_st_type() == elfcpp::STT_GNU_IFUNC) lv.set_is_ifunc_symbol(); // Save the input symbol value for use in do_finalize_local_symbols(). lv.set_input_value(sym.get_st_value()); // Decide whether this symbol should go into the output file. if (is_ordinary && shndx < shnum && (out_sections[shndx] == NULL || (out_sections[shndx]->order() == ORDER_EHFRAME && out_section_offsets[shndx] == invalid_address))) { // This is either a discarded section or an optimized .eh_frame // section. lv.set_no_output_symtab_entry(); gold_assert(!lv.needs_output_dynsym_entry()); continue; } if (sym.get_st_type() == elfcpp::STT_SECTION || !this->adjust_local_symbol(&lv)) { lv.set_no_output_symtab_entry(); gold_assert(!lv.needs_output_dynsym_entry()); continue; } if (sym.get_st_name() >= strtab_size) { this->error(_("local symbol %u section name out of range: %u >= %u"), i, sym.get_st_name(), static_cast(strtab_size)); lv.set_no_output_symtab_entry(); continue; } const char* name = pnames + sym.get_st_name(); // If needed, add the symbol to the dynamic symbol table string pool. if (lv.needs_output_dynsym_entry()) { dynpool->add(name, true, NULL); ++dyncount; } if (strip_all || (discard_all && lv.may_be_discarded_from_output_symtab())) { lv.set_no_output_symtab_entry(); continue; } // By default, discard temporary local symbols in merge sections. // If --discard-locals option is used, discard all temporary local // symbols. These symbols start with system-specific local label // prefixes, typically .L for ELF system. We want to be compatible // with GNU ld so here we essentially use the same check in // bfd_is_local_label(). The code is different because we already // know that: // // - the symbol is local and thus cannot have global or weak binding. // - the symbol is not a section symbol. // - the symbol has a name. // // We do not discard a symbol if it needs a dynamic symbol entry. if ((discard_locals || (discard_sec_merge && is_ordinary && out_section_offsets[shndx] == invalid_address)) && sym.get_st_type() != elfcpp::STT_FILE && !lv.needs_output_dynsym_entry() && lv.may_be_discarded_from_output_symtab() && parameters->target().is_local_label_name(name)) { lv.set_no_output_symtab_entry(); continue; } // Discard the local symbol if -retain_symbols_file is specified // and the local symbol is not in that file. if (!parameters->options().should_retain_symbol(name)) { lv.set_no_output_symtab_entry(); continue; } // Add the symbol to the symbol table string pool. pool->add(name, true, NULL); ++count; } this->output_local_symbol_count_ = count; this->output_local_dynsym_count_ = dyncount; } // Compute the final value of a local symbol. template typename Sized_relobj_file::Compute_final_local_value_status Sized_relobj_file::compute_final_local_value_internal( unsigned int r_sym, const Symbol_value* lv_in, Symbol_value* lv_out, bool relocatable, const Output_sections& out_sections, const std::vector
& out_offsets, const Symbol_table* symtab) { // We are going to overwrite *LV_OUT, if it has a merged symbol value, // we may have a memory leak. gold_assert(lv_out->has_output_value()); bool is_ordinary; unsigned int shndx = lv_in->input_shndx(&is_ordinary); // Set the output symbol value. if (!is_ordinary) { if (shndx == elfcpp::SHN_ABS || Symbol::is_common_shndx(shndx)) lv_out->set_output_value(lv_in->input_value()); else { this->error(_("unknown section index %u for local symbol %u"), shndx, r_sym); lv_out->set_output_value(0); return This::CFLV_ERROR; } } else { if (shndx >= this->shnum()) { this->error(_("local symbol %u section index %u out of range"), r_sym, shndx); lv_out->set_output_value(0); return This::CFLV_ERROR; } Output_section* os = out_sections[shndx]; Address secoffset = out_offsets[shndx]; if (symtab->is_section_folded(this, shndx)) { gold_assert(os == NULL && secoffset == invalid_address); // Get the os of the section it is folded onto. Section_id folded = symtab->icf()->get_folded_section(this, shndx); gold_assert(folded.first != NULL); Sized_relobj_file* folded_obj = reinterpret_cast *>(folded.first); os = folded_obj->output_section(folded.second); gold_assert(os != NULL); secoffset = folded_obj->get_output_section_offset(folded.second); // This could be a relaxed input section. if (secoffset == invalid_address) { const Output_relaxed_input_section* relaxed_section = os->find_relaxed_input_section(folded_obj, folded.second); gold_assert(relaxed_section != NULL); secoffset = relaxed_section->address() - os->address(); } } if (os == NULL) { // This local symbol belongs to a section we are discarding. // In some cases when applying relocations later, we will // attempt to match it to the corresponding kept section, // so we leave the input value unchanged here. return This::CFLV_DISCARDED; } else if (secoffset == invalid_address) { uint64_t start; // This is a SHF_MERGE section or one which otherwise // requires special handling. if (os->order() == ORDER_EHFRAME) { // This local symbol belongs to a discarded or optimized // .eh_frame section. Just treat it like the case in which // os == NULL above. gold_assert(this->has_eh_frame_); return This::CFLV_DISCARDED; } else if (!lv_in->is_section_symbol()) { // This is not a section symbol. We can determine // the final value now. uint64_t value = os->output_address(this, shndx, lv_in->input_value()); if (relocatable) value -= os->address(); lv_out->set_output_value(value); } else if (!os->find_starting_output_address(this, shndx, &start)) { // This is a section symbol, but apparently not one in a // merged section. First check to see if this is a relaxed // input section. If so, use its address. Otherwise just // use the start of the output section. This happens with // relocatable links when the input object has section // symbols for arbitrary non-merge sections. const Output_section_data* posd = os->find_relaxed_input_section(this, shndx); if (posd != NULL) { uint64_t value = posd->address(); if (relocatable) value -= os->address(); lv_out->set_output_value(value); } else lv_out->set_output_value(os->address()); } else { // We have to consider the addend to determine the // value to use in a relocation. START is the start // of this input section. If we are doing a relocatable // link, use offset from start output section instead of // address. Address adjusted_start = relocatable ? start - os->address() : start; Merged_symbol_value* msv = new Merged_symbol_value(lv_in->input_value(), adjusted_start); lv_out->set_merged_symbol_value(msv); } } else if (lv_in->is_tls_symbol() || (lv_in->is_section_symbol() && (os->flags() & elfcpp::SHF_TLS))) lv_out->set_output_value(os->tls_offset() + secoffset + lv_in->input_value()); else lv_out->set_output_value((relocatable ? 0 : os->address()) + secoffset + lv_in->input_value()); } return This::CFLV_OK; } // Compute final local symbol value. R_SYM is the index of a local // symbol in symbol table. LV points to a symbol value, which is // expected to hold the input value and to be over-written by the // final value. SYMTAB points to a symbol table. Some targets may want // to know would-be-finalized local symbol values in relaxation. // Hence we provide this method. Since this method updates *LV, a // callee should make a copy of the original local symbol value and // use the copy instead of modifying an object's local symbols before // everything is finalized. The caller should also free up any allocated // memory in the return value in *LV. template typename Sized_relobj_file::Compute_final_local_value_status Sized_relobj_file::compute_final_local_value( unsigned int r_sym, const Symbol_value* lv_in, Symbol_value* lv_out, const Symbol_table* symtab) { // This is just a wrapper of compute_final_local_value_internal. const bool relocatable = parameters->options().relocatable(); const Output_sections& out_sections(this->output_sections()); const std::vector
& out_offsets(this->section_offsets()); return this->compute_final_local_value_internal(r_sym, lv_in, lv_out, relocatable, out_sections, out_offsets, symtab); } // Finalize the local symbols. Here we set the final value in // THIS->LOCAL_VALUES_ and set their output symbol table indexes. // This function is always called from a singleton thread. The actual // output of the local symbols will occur in a separate task. template unsigned int Sized_relobj_file::do_finalize_local_symbols( unsigned int index, off_t off, Symbol_table* symtab) { gold_assert(off == static_cast(align_address(off, size >> 3))); const unsigned int loccount = this->local_symbol_count_; this->local_symbol_offset_ = off; const bool relocatable = parameters->options().relocatable(); const Output_sections& out_sections(this->output_sections()); const std::vector
& out_offsets(this->section_offsets()); for (unsigned int i = 1; i < loccount; ++i) { Symbol_value* lv = &this->local_values_[i]; Compute_final_local_value_status cflv_status = this->compute_final_local_value_internal(i, lv, lv, relocatable, out_sections, out_offsets, symtab); switch (cflv_status) { case CFLV_OK: if (!lv->is_output_symtab_index_set()) { lv->set_output_symtab_index(index); ++index; } break; case CFLV_DISCARDED: case CFLV_ERROR: // Do nothing. break; default: gold_unreachable(); } } return index; } // Set the output dynamic symbol table indexes for the local variables. template unsigned int Sized_relobj_file::do_set_local_dynsym_indexes( unsigned int index) { const unsigned int loccount = this->local_symbol_count_; for (unsigned int i = 1; i < loccount; ++i) { Symbol_value& lv(this->local_values_[i]); if (lv.needs_output_dynsym_entry()) { lv.set_output_dynsym_index(index); ++index; } } return index; } // Set the offset where local dynamic symbol information will be stored. // Returns the count of local symbols contributed to the symbol table by // this object. template unsigned int Sized_relobj_file::do_set_local_dynsym_offset(off_t off) { gold_assert(off == static_cast(align_address(off, size >> 3))); this->local_dynsym_offset_ = off; return this->output_local_dynsym_count_; } // If Symbols_data is not NULL get the section flags from here otherwise // get it from the file. template uint64_t Sized_relobj_file::do_section_flags(unsigned int shndx) { Symbols_data* sd = this->get_symbols_data(); if (sd != NULL) { const unsigned char* pshdrs = sd->section_headers_data + This::shdr_size * shndx; typename This::Shdr shdr(pshdrs); return shdr.get_sh_flags(); } // If sd is NULL, read the section header from the file. return this->elf_file_.section_flags(shndx); } // Get the section's ent size from Symbols_data. Called by get_section_contents // in icf.cc template uint64_t Sized_relobj_file::do_section_entsize(unsigned int shndx) { Symbols_data* sd = this->get_symbols_data(); gold_assert(sd != NULL); const unsigned char* pshdrs = sd->section_headers_data + This::shdr_size * shndx; typename This::Shdr shdr(pshdrs); return shdr.get_sh_entsize(); } // Write out the local symbols. template void Sized_relobj_file::write_local_symbols( Output_file* of, const Stringpool* sympool, const Stringpool* dynpool, Output_symtab_xindex* symtab_xindex, Output_symtab_xindex* dynsym_xindex, off_t symtab_off) { const bool strip_all = parameters->options().strip_all(); if (strip_all) { if (this->output_local_dynsym_count_ == 0) return; this->output_local_symbol_count_ = 0; } gold_assert(this->symtab_shndx_ != -1U); if (this->symtab_shndx_ == 0) { // This object has no symbols. Weird but legal. return; } // Read the symbol table section header. const unsigned int symtab_shndx = this->symtab_shndx_; typename This::Shdr symtabshdr(this, this->elf_file_.section_header(symtab_shndx)); gold_assert(symtabshdr.get_sh_type() == elfcpp::SHT_SYMTAB); const unsigned int loccount = this->local_symbol_count_; gold_assert(loccount == symtabshdr.get_sh_info()); // Read the local symbols. const int sym_size = This::sym_size; off_t locsize = loccount * sym_size; const unsigned char* psyms = this->get_view(symtabshdr.get_sh_offset(), locsize, true, false); // Read the symbol names. const unsigned int strtab_shndx = this->adjust_shndx(symtabshdr.get_sh_link()); section_size_type strtab_size; const unsigned char* pnamesu = this->section_contents(strtab_shndx, &strtab_size, false); const char* pnames = reinterpret_cast(pnamesu); // Get views into the output file for the portions of the symbol table // and the dynamic symbol table that we will be writing. off_t output_size = this->output_local_symbol_count_ * sym_size; unsigned char* oview = NULL; if (output_size > 0) oview = of->get_output_view(symtab_off + this->local_symbol_offset_, output_size); off_t dyn_output_size = this->output_local_dynsym_count_ * sym_size; unsigned char* dyn_oview = NULL; if (dyn_output_size > 0) dyn_oview = of->get_output_view(this->local_dynsym_offset_, dyn_output_size); const Output_sections& out_sections(this->output_sections()); gold_assert(this->local_values_.size() == loccount); unsigned char* ov = oview; unsigned char* dyn_ov = dyn_oview; psyms += sym_size; for (unsigned int i = 1; i < loccount; ++i, psyms += sym_size) { elfcpp::Sym isym(psyms); Symbol_value& lv(this->local_values_[i]); bool is_ordinary; unsigned int st_shndx = this->adjust_sym_shndx(i, isym.get_st_shndx(), &is_ordinary); if (is_ordinary) { gold_assert(st_shndx < out_sections.size()); if (out_sections[st_shndx] == NULL) continue; st_shndx = out_sections[st_shndx]->out_shndx(); if (st_shndx >= elfcpp::SHN_LORESERVE) { if (lv.has_output_symtab_entry()) symtab_xindex->add(lv.output_symtab_index(), st_shndx); if (lv.has_output_dynsym_entry()) dynsym_xindex->add(lv.output_dynsym_index(), st_shndx); st_shndx = elfcpp::SHN_XINDEX; } } // Write the symbol to the output symbol table. if (lv.has_output_symtab_entry()) { elfcpp::Sym_write osym(ov); gold_assert(isym.get_st_name() < strtab_size); const char* name = pnames + isym.get_st_name(); osym.put_st_name(sympool->get_offset(name)); osym.put_st_value(lv.value(this, 0)); osym.put_st_size(isym.get_st_size()); osym.put_st_info(isym.get_st_info()); osym.put_st_other(isym.get_st_other()); osym.put_st_shndx(st_shndx); ov += sym_size; } // Write the symbol to the output dynamic symbol table. if (lv.has_output_dynsym_entry()) { gold_assert(dyn_ov < dyn_oview + dyn_output_size); elfcpp::Sym_write osym(dyn_ov); gold_assert(isym.get_st_name() < strtab_size); const char* name = pnames + isym.get_st_name(); osym.put_st_name(dynpool->get_offset(name)); osym.put_st_value(lv.value(this, 0)); osym.put_st_size(isym.get_st_size()); osym.put_st_info(isym.get_st_info()); osym.put_st_other(isym.get_st_other()); osym.put_st_shndx(st_shndx); dyn_ov += sym_size; } } if (output_size > 0) { gold_assert(ov - oview == output_size); of->write_output_view(symtab_off + this->local_symbol_offset_, output_size, oview); } if (dyn_output_size > 0) { gold_assert(dyn_ov - dyn_oview == dyn_output_size); of->write_output_view(this->local_dynsym_offset_, dyn_output_size, dyn_oview); } } // Set *INFO to symbolic information about the offset OFFSET in the // section SHNDX. Return true if we found something, false if we // found nothing. template bool Sized_relobj_file::get_symbol_location_info( unsigned int shndx, off_t offset, Symbol_location_info* info) { if (this->symtab_shndx_ == 0) return false; section_size_type symbols_size; const unsigned char* symbols = this->section_contents(this->symtab_shndx_, &symbols_size, false); unsigned int symbol_names_shndx = this->adjust_shndx(this->section_link(this->symtab_shndx_)); section_size_type names_size; const unsigned char* symbol_names_u = this->section_contents(symbol_names_shndx, &names_size, false); const char* symbol_names = reinterpret_cast(symbol_names_u); const int sym_size = This::sym_size; const size_t count = symbols_size / sym_size; const unsigned char* p = symbols; for (size_t i = 0; i < count; ++i, p += sym_size) { elfcpp::Sym sym(p); if (sym.get_st_type() == elfcpp::STT_FILE) { if (sym.get_st_name() >= names_size) info->source_file = "(invalid)"; else info->source_file = symbol_names + sym.get_st_name(); continue; } bool is_ordinary; unsigned int st_shndx = this->adjust_sym_shndx(i, sym.get_st_shndx(), &is_ordinary); if (is_ordinary && st_shndx == shndx && static_cast(sym.get_st_value()) <= offset && (static_cast(sym.get_st_value() + sym.get_st_size()) > offset)) { info->enclosing_symbol_type = sym.get_st_type(); if (sym.get_st_name() > names_size) info->enclosing_symbol_name = "(invalid)"; else { info->enclosing_symbol_name = symbol_names + sym.get_st_name(); if (parameters->options().do_demangle()) { char* demangled_name = cplus_demangle( info->enclosing_symbol_name.c_str(), DMGL_ANSI | DMGL_PARAMS); if (demangled_name != NULL) { info->enclosing_symbol_name.assign(demangled_name); free(demangled_name); } } } return true; } } return false; } // Look for a kept section corresponding to the given discarded section, // and return its output address. This is used only for relocations in // debugging sections. If we can't find the kept section, return 0. template typename Sized_relobj_file::Address Sized_relobj_file::map_to_kept_section( unsigned int shndx, std::string& section_name, bool* pfound) const { Kept_section* kept_section; bool is_comdat; uint64_t sh_size; unsigned int symndx; bool found = false; if (this->get_kept_comdat_section(shndx, &is_comdat, &symndx, &sh_size, &kept_section)) { Relobj* kept_object = kept_section->object(); unsigned int kept_shndx = 0; if (!kept_section->is_comdat()) { // The kept section is a linkonce section. if (sh_size == kept_section->linkonce_size()) found = true; } else { if (is_comdat) { // Find the corresponding kept section. // Since we're using this mapping for relocation processing, // we don't want to match sections unless they have the same // size. uint64_t kept_size = 0; if (kept_section->find_comdat_section(section_name, &kept_shndx, &kept_size)) { if (sh_size == kept_size) found = true; } } else { uint64_t kept_size = 0; if (kept_section->find_single_comdat_section(&kept_shndx, &kept_size) && sh_size == kept_size) found = true; } } if (found) { Sized_relobj_file* kept_relobj = static_cast*>(kept_object); Output_section* os = kept_relobj->output_section(kept_shndx); Address offset = kept_relobj->get_output_section_offset(kept_shndx); if (os != NULL && offset != invalid_address) { *pfound = true; return os->address() + offset; } } } *pfound = false; return 0; } // Look for a kept section corresponding to the given discarded section, // and return its object file. template Relobj* Sized_relobj_file::find_kept_section_object( unsigned int shndx, unsigned int *symndx_p) const { Kept_section* kept_section; bool is_comdat; uint64_t sh_size; if (this->get_kept_comdat_section(shndx, &is_comdat, symndx_p, &sh_size, &kept_section)) return kept_section->object(); return NULL; } // Return the name of symbol SYMNDX. template const char* Sized_relobj_file::get_symbol_name(unsigned int symndx) { if (this->symtab_shndx_ == 0) return NULL; section_size_type symbols_size; const unsigned char* symbols = this->section_contents(this->symtab_shndx_, &symbols_size, false); unsigned int symbol_names_shndx = this->adjust_shndx(this->section_link(this->symtab_shndx_)); section_size_type names_size; const unsigned char* symbol_names_u = this->section_contents(symbol_names_shndx, &names_size, false); const char* symbol_names = reinterpret_cast(symbol_names_u); const unsigned char* p = symbols + symndx * This::sym_size; if (p >= symbols + symbols_size) return NULL; elfcpp::Sym sym(p); return symbol_names + sym.get_st_name(); } // Get symbol counts. template void Sized_relobj_file::do_get_global_symbol_counts( const Symbol_table*, size_t* defined, size_t* used) const { *defined = this->defined_count_; size_t count = 0; for (typename Symbols::const_iterator p = this->symbols_.begin(); p != this->symbols_.end(); ++p) if (*p != NULL && (*p)->source() == Symbol::FROM_OBJECT && (*p)->object() == this && (*p)->is_defined()) ++count; *used = count; } // Return a view of the decompressed contents of a section. Set *PLEN // to the size. Set *IS_NEW to true if the contents need to be freed // by the caller. const unsigned char* Object::decompressed_section_contents( unsigned int shndx, section_size_type* plen, bool* is_new, uint64_t* palign) { section_size_type buffer_size; const unsigned char* buffer = this->do_section_contents(shndx, &buffer_size, false); if (this->compressed_sections_ == NULL) { *plen = buffer_size; *is_new = false; return buffer; } Compressed_section_map::const_iterator p = this->compressed_sections_->find(shndx); if (p == this->compressed_sections_->end()) { *plen = buffer_size; *is_new = false; return buffer; } section_size_type uncompressed_size = p->second.size; if (p->second.contents != NULL) { *plen = uncompressed_size; *is_new = false; if (palign != NULL) *palign = p->second.addralign; return p->second.contents; } unsigned char* uncompressed_data = new unsigned char[uncompressed_size]; if (!decompress_input_section(buffer, buffer_size, uncompressed_data, uncompressed_size, elfsize(), is_big_endian(), p->second.flag)) this->error(_("could not decompress section %s"), this->do_section_name(shndx).c_str()); // We could cache the results in p->second.contents and store // false in *IS_NEW, but build_compressed_section_map() would // have done so if it had expected it to be profitable. If // we reach this point, we expect to need the contents only // once in this pass. *plen = uncompressed_size; *is_new = true; if (palign != NULL) *palign = p->second.addralign; return uncompressed_data; } // Discard any buffers of uncompressed sections. This is done // at the end of the Add_symbols task. void Object::discard_decompressed_sections() { if (this->compressed_sections_ == NULL) return; for (Compressed_section_map::iterator p = this->compressed_sections_->begin(); p != this->compressed_sections_->end(); ++p) { if (p->second.contents != NULL) { delete[] p->second.contents; p->second.contents = NULL; } } } // Input_objects methods. // Add a regular relocatable object to the list. Return false if this // object should be ignored. bool Input_objects::add_object(Object* obj) { // Print the filename if the -t/--trace option is selected. if (parameters->options().trace()) gold_info("%s", obj->name().c_str()); if (!obj->is_dynamic()) this->relobj_list_.push_back(static_cast(obj)); else { // See if this is a duplicate SONAME. Dynobj* dynobj = static_cast(obj); const char* soname = dynobj->soname(); Unordered_map::value_type val(soname, obj); std::pair::iterator, bool> ins = this->sonames_.insert(val); if (!ins.second) { // We have already seen a dynamic object with this soname. // If any instances of this object on the command line have // the --no-as-needed flag, make sure the one we keep is // marked so. if (!obj->as_needed()) { gold_assert(ins.first->second != NULL); ins.first->second->clear_as_needed(); } return false; } this->dynobj_list_.push_back(dynobj); } // Add this object to the cross-referencer if requested. if (parameters->options().user_set_print_symbol_counts() || parameters->options().cref()) { if (this->cref_ == NULL) this->cref_ = new Cref(); this->cref_->add_object(obj); } return true; } // For each dynamic object, record whether we've seen all of its // explicit dependencies. void Input_objects::check_dynamic_dependencies() const { bool issued_copy_dt_needed_error = false; for (Dynobj_list::const_iterator p = this->dynobj_list_.begin(); p != this->dynobj_list_.end(); ++p) { const Dynobj::Needed& needed((*p)->needed()); bool found_all = true; Dynobj::Needed::const_iterator pneeded; for (pneeded = needed.begin(); pneeded != needed.end(); ++pneeded) { if (this->sonames_.find(*pneeded) == this->sonames_.end()) { found_all = false; break; } } (*p)->set_has_unknown_needed_entries(!found_all); // --copy-dt-needed-entries aka --add-needed is a GNU ld option // that gold does not support. However, they cause no trouble // unless there is a DT_NEEDED entry that we don't know about; // warn only in that case. if (!found_all && !issued_copy_dt_needed_error && (parameters->options().copy_dt_needed_entries() || parameters->options().add_needed())) { const char* optname; if (parameters->options().copy_dt_needed_entries()) optname = "--copy-dt-needed-entries"; else optname = "--add-needed"; gold_error(_("%s is not supported but is required for %s in %s"), optname, (*pneeded).c_str(), (*p)->name().c_str()); issued_copy_dt_needed_error = true; } } } // Start processing an archive. void Input_objects::archive_start(Archive* archive) { if (parameters->options().user_set_print_symbol_counts() || parameters->options().cref()) { if (this->cref_ == NULL) this->cref_ = new Cref(); this->cref_->add_archive_start(archive); } } // Stop processing an archive. void Input_objects::archive_stop(Archive* archive) { if (parameters->options().user_set_print_symbol_counts() || parameters->options().cref()) this->cref_->add_archive_stop(archive); } // Print symbol counts void Input_objects::print_symbol_counts(const Symbol_table* symtab) const { if (parameters->options().user_set_print_symbol_counts() && this->cref_ != NULL) this->cref_->print_symbol_counts(symtab); } // Print a cross reference table. void Input_objects::print_cref(const Symbol_table* symtab, FILE* f) const { if (parameters->options().cref() && this->cref_ != NULL) this->cref_->print_cref(symtab, f); } // Relocate_info methods. // Return a string describing the location of a relocation when file // and lineno information is not available. This is only used in // error messages. template std::string Relocate_info::location(size_t, off_t offset) const { Sized_dwarf_line_info line_info(this->object); std::string ret = line_info.addr2line(this->data_shndx, offset, NULL); if (!ret.empty()) return ret; ret = this->object->name(); Symbol_location_info info; if (this->object->get_symbol_location_info(this->data_shndx, offset, &info)) { if (!info.source_file.empty()) { ret += ":"; ret += info.source_file; } ret += ":"; if (info.enclosing_symbol_type == elfcpp::STT_FUNC) ret += _("function "); ret += info.enclosing_symbol_name; return ret; } ret += "("; ret += this->object->section_name(this->data_shndx); char buf[100]; snprintf(buf, sizeof buf, "+0x%lx)", static_cast(offset)); ret += buf; return ret; } } // End namespace gold. namespace { using namespace gold; // Read an ELF file with the header and return the appropriate // instance of Object. template Object* make_elf_sized_object(const std::string& name, Input_file* input_file, off_t offset, const elfcpp::Ehdr& ehdr, bool* punconfigured) { Target* target = select_target(input_file, offset, ehdr.get_e_machine(), size, big_endian, ehdr.get_e_ident()[elfcpp::EI_OSABI], ehdr.get_e_ident()[elfcpp::EI_ABIVERSION]); if (target == NULL) gold_fatal(_("%s: unsupported ELF machine number %d"), name.c_str(), ehdr.get_e_machine()); if (!parameters->target_valid()) set_parameters_target(target); else if (target != ¶meters->target()) { if (punconfigured != NULL) *punconfigured = true; else gold_error(_("%s: incompatible target"), name.c_str()); return NULL; } return target->make_elf_object(name, input_file, offset, ehdr); } } // End anonymous namespace. namespace gold { // Return whether INPUT_FILE is an ELF object. bool is_elf_object(Input_file* input_file, off_t offset, const unsigned char** start, int* read_size) { off_t filesize = input_file->file().filesize(); int want = elfcpp::Elf_recognizer::max_header_size; if (filesize - offset < want) want = filesize - offset; const unsigned char* p = input_file->file().get_view(offset, 0, want, true, false); *start = p; *read_size = want; return elfcpp::Elf_recognizer::is_elf_file(p, want); } // Read an ELF file and return the appropriate instance of Object. Object* make_elf_object(const std::string& name, Input_file* input_file, off_t offset, const unsigned char* p, section_offset_type bytes, bool* punconfigured) { if (punconfigured != NULL) *punconfigured = false; std::string error; bool big_endian = false; int size = 0; if (!elfcpp::Elf_recognizer::is_valid_header(p, bytes, &size, &big_endian, &error)) { gold_error(_("%s: %s"), name.c_str(), error.c_str()); return NULL; } if (size == 32) { if (big_endian) { #ifdef HAVE_TARGET_32_BIG elfcpp::Ehdr<32, true> ehdr(p); return make_elf_sized_object<32, true>(name, input_file, offset, ehdr, punconfigured); #else if (punconfigured != NULL) *punconfigured = true; else gold_error(_("%s: not configured to support " "32-bit big-endian object"), name.c_str()); return NULL; #endif } else { #ifdef HAVE_TARGET_32_LITTLE elfcpp::Ehdr<32, false> ehdr(p); return make_elf_sized_object<32, false>(name, input_file, offset, ehdr, punconfigured); #else if (punconfigured != NULL) *punconfigured = true; else gold_error(_("%s: not configured to support " "32-bit little-endian object"), name.c_str()); return NULL; #endif } } else if (size == 64) { if (big_endian) { #ifdef HAVE_TARGET_64_BIG elfcpp::Ehdr<64, true> ehdr(p); return make_elf_sized_object<64, true>(name, input_file, offset, ehdr, punconfigured); #else if (punconfigured != NULL) *punconfigured = true; else gold_error(_("%s: not configured to support " "64-bit big-endian object"), name.c_str()); return NULL; #endif } else { #ifdef HAVE_TARGET_64_LITTLE elfcpp::Ehdr<64, false> ehdr(p); return make_elf_sized_object<64, false>(name, input_file, offset, ehdr, punconfigured); #else if (punconfigured != NULL) *punconfigured = true; else gold_error(_("%s: not configured to support " "64-bit little-endian object"), name.c_str()); return NULL; #endif } } else gold_unreachable(); } // Instantiate the templates we need. #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG) template void Relobj::initialize_input_to_output_map<64>(unsigned int shndx, elfcpp::Elf_types<64>::Elf_Addr starting_address, Unordered_map::Elf_Addr>* output_addresses) const; #endif #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG) template void Relobj::initialize_input_to_output_map<32>(unsigned int shndx, elfcpp::Elf_types<32>::Elf_Addr starting_address, Unordered_map::Elf_Addr>* output_addresses) const; #endif #ifdef HAVE_TARGET_32_LITTLE template void Object::read_section_data<32, false>(elfcpp::Elf_file<32, false, Object>*, Read_symbols_data*); template const unsigned char* Object::find_shdr<32,false>(const unsigned char*, const char*, const char*, section_size_type, const unsigned char*) const; #endif #ifdef HAVE_TARGET_32_BIG template void Object::read_section_data<32, true>(elfcpp::Elf_file<32, true, Object>*, Read_symbols_data*); template const unsigned char* Object::find_shdr<32,true>(const unsigned char*, const char*, const char*, section_size_type, const unsigned char*) const; #endif #ifdef HAVE_TARGET_64_LITTLE template void Object::read_section_data<64, false>(elfcpp::Elf_file<64, false, Object>*, Read_symbols_data*); template const unsigned char* Object::find_shdr<64,false>(const unsigned char*, const char*, const char*, section_size_type, const unsigned char*) const; #endif #ifdef HAVE_TARGET_64_BIG template void Object::read_section_data<64, true>(elfcpp::Elf_file<64, true, Object>*, Read_symbols_data*); template const unsigned char* Object::find_shdr<64,true>(const unsigned char*, const char*, const char*, section_size_type, const unsigned char*) const; #endif #ifdef HAVE_TARGET_32_LITTLE template class Sized_relobj<32, false>; template class Sized_relobj_file<32, false>; #endif #ifdef HAVE_TARGET_32_BIG template class Sized_relobj<32, true>; template class Sized_relobj_file<32, true>; #endif #ifdef HAVE_TARGET_64_LITTLE template class Sized_relobj<64, false>; template class Sized_relobj_file<64, false>; #endif #ifdef HAVE_TARGET_64_BIG template class Sized_relobj<64, true>; template class Sized_relobj_file<64, true>; #endif #ifdef HAVE_TARGET_32_LITTLE template struct Relocate_info<32, false>; #endif #ifdef HAVE_TARGET_32_BIG template struct Relocate_info<32, true>; #endif #ifdef HAVE_TARGET_64_LITTLE template struct Relocate_info<64, false>; #endif #ifdef HAVE_TARGET_64_BIG template struct Relocate_info<64, true>; #endif #ifdef HAVE_TARGET_32_LITTLE template void Xindex::initialize_symtab_xindex<32, false>(Object*, unsigned int); template void Xindex::read_symtab_xindex<32, false>(Object*, unsigned int, const unsigned char*); #endif #ifdef HAVE_TARGET_32_BIG template void Xindex::initialize_symtab_xindex<32, true>(Object*, unsigned int); template void Xindex::read_symtab_xindex<32, true>(Object*, unsigned int, const unsigned char*); #endif #ifdef HAVE_TARGET_64_LITTLE template void Xindex::initialize_symtab_xindex<64, false>(Object*, unsigned int); template void Xindex::read_symtab_xindex<64, false>(Object*, unsigned int, const unsigned char*); #endif #ifdef HAVE_TARGET_64_BIG template void Xindex::initialize_symtab_xindex<64, true>(Object*, unsigned int); template void Xindex::read_symtab_xindex<64, true>(Object*, unsigned int, const unsigned char*); #endif #ifdef HAVE_TARGET_32_LITTLE template Compressed_section_map* build_compressed_section_map<32, false>(const unsigned char*, unsigned int, const char*, section_size_type, Object*, bool); #endif #ifdef HAVE_TARGET_32_BIG template Compressed_section_map* build_compressed_section_map<32, true>(const unsigned char*, unsigned int, const char*, section_size_type, Object*, bool); #endif #ifdef HAVE_TARGET_64_LITTLE template Compressed_section_map* build_compressed_section_map<64, false>(const unsigned char*, unsigned int, const char*, section_size_type, Object*, bool); #endif #ifdef HAVE_TARGET_64_BIG template Compressed_section_map* build_compressed_section_map<64, true>(const unsigned char*, unsigned int, const char*, section_size_type, Object*, bool); #endif } // End namespace gold.