URL
https://opencores.org/ocsvn/open8_urisc/open8_urisc/trunk
Subversion Repositories open8_urisc
[/] [open8_urisc/] [trunk/] [gnu/] [binutils/] [gold/] [layout.cc] - Rev 60
Go to most recent revision | Compare with Previous | Blame | View Log
// layout.cc -- lay out output file sections for gold // Copyright 2006, 2007, 2008, 2009, 2010, 2011 Free Software Foundation, Inc. // Written by Ian Lance Taylor <iant@google.com>. // 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 <cerrno> #include <cstring> #include <algorithm> #include <iostream> #include <fstream> #include <utility> #include <fcntl.h> #include <fnmatch.h> #include <unistd.h> #include "libiberty.h" #include "md5.h" #include "sha1.h" #include "parameters.h" #include "options.h" #include "mapfile.h" #include "script.h" #include "script-sections.h" #include "output.h" #include "symtab.h" #include "dynobj.h" #include "ehframe.h" #include "compressed_output.h" #include "reduced_debug_output.h" #include "reloc.h" #include "descriptors.h" #include "plugin.h" #include "incremental.h" #include "layout.h" namespace gold { // Class Free_list. // The total number of free lists used. unsigned int Free_list::num_lists = 0; // The total number of free list nodes used. unsigned int Free_list::num_nodes = 0; // The total number of calls to Free_list::remove. unsigned int Free_list::num_removes = 0; // The total number of nodes visited during calls to Free_list::remove. unsigned int Free_list::num_remove_visits = 0; // The total number of calls to Free_list::allocate. unsigned int Free_list::num_allocates = 0; // The total number of nodes visited during calls to Free_list::allocate. unsigned int Free_list::num_allocate_visits = 0; // Initialize the free list. Creates a single free list node that // describes the entire region of length LEN. If EXTEND is true, // allocate() is allowed to extend the region beyond its initial // length. void Free_list::init(off_t len, bool extend) { this->list_.push_front(Free_list_node(0, len)); this->last_remove_ = this->list_.begin(); this->extend_ = extend; this->length_ = len; ++Free_list::num_lists; ++Free_list::num_nodes; } // Remove a chunk from the free list. Because we start with a single // node that covers the entire section, and remove chunks from it one // at a time, we do not need to coalesce chunks or handle cases that // span more than one free node. We expect to remove chunks from the // free list in order, and we expect to have only a few chunks of free // space left (corresponding to files that have changed since the last // incremental link), so a simple linear list should provide sufficient // performance. void Free_list::remove(off_t start, off_t end) { if (start == end) return; gold_assert(start < end); ++Free_list::num_removes; Iterator p = this->last_remove_; if (p->start_ > start) p = this->list_.begin(); for (; p != this->list_.end(); ++p) { ++Free_list::num_remove_visits; // Find a node that wholly contains the indicated region. if (p->start_ <= start && p->end_ >= end) { // Case 1: the indicated region spans the whole node. // Add some fuzz to avoid creating tiny free chunks. if (p->start_ + 3 >= start && p->end_ <= end + 3) p = this->list_.erase(p); // Case 2: remove a chunk from the start of the node. else if (p->start_ + 3 >= start) p->start_ = end; // Case 3: remove a chunk from the end of the node. else if (p->end_ <= end + 3) p->end_ = start; // Case 4: remove a chunk from the middle, and split // the node into two. else { Free_list_node newnode(p->start_, start); p->start_ = end; this->list_.insert(p, newnode); ++Free_list::num_nodes; } this->last_remove_ = p; return; } } // Did not find a node containing the given chunk. This could happen // because a small chunk was already removed due to the fuzz. gold_debug(DEBUG_INCREMENTAL, "Free_list::remove(%d,%d) not found", static_cast<int>(start), static_cast<int>(end)); } // Allocate a chunk of size LEN from the free list. Returns -1ULL // if a sufficiently large chunk of free space is not found. // We use a simple first-fit algorithm. off_t Free_list::allocate(off_t len, uint64_t align, off_t minoff) { gold_debug(DEBUG_INCREMENTAL, "Free_list::allocate(%08lx, %d, %08lx)", static_cast<long>(len), static_cast<int>(align), static_cast<long>(minoff)); if (len == 0) return align_address(minoff, align); ++Free_list::num_allocates; for (Iterator p = this->list_.begin(); p != this->list_.end(); ++p) { ++Free_list::num_allocate_visits; off_t start = p->start_ > minoff ? p->start_ : minoff; start = align_address(start, align); off_t end = start + len; if (end <= p->end_) { if (p->start_ + 3 >= start && p->end_ <= end + 3) this->list_.erase(p); else if (p->start_ + 3 >= start) p->start_ = end; else if (p->end_ <= end + 3) p->end_ = start; else { Free_list_node newnode(p->start_, start); p->start_ = end; this->list_.insert(p, newnode); ++Free_list::num_nodes; } return start; } } return -1; } // Dump the free list (for debugging). void Free_list::dump() { gold_info("Free list:\n start end length\n"); for (Iterator p = this->list_.begin(); p != this->list_.end(); ++p) gold_info(" %08lx %08lx %08lx", static_cast<long>(p->start_), static_cast<long>(p->end_), static_cast<long>(p->end_ - p->start_)); } // Print the statistics for the free lists. void Free_list::print_stats() { fprintf(stderr, _("%s: total free lists: %u\n"), program_name, Free_list::num_lists); fprintf(stderr, _("%s: total free list nodes: %u\n"), program_name, Free_list::num_nodes); fprintf(stderr, _("%s: calls to Free_list::remove: %u\n"), program_name, Free_list::num_removes); fprintf(stderr, _("%s: nodes visited: %u\n"), program_name, Free_list::num_remove_visits); fprintf(stderr, _("%s: calls to Free_list::allocate: %u\n"), program_name, Free_list::num_allocates); fprintf(stderr, _("%s: nodes visited: %u\n"), program_name, Free_list::num_allocate_visits); } // Layout::Relaxation_debug_check methods. // Check that sections and special data are in reset states. // We do not save states for Output_sections and special Output_data. // So we check that they have not assigned any addresses or offsets. // clean_up_after_relaxation simply resets their addresses and offsets. void Layout::Relaxation_debug_check::check_output_data_for_reset_values( const Layout::Section_list& sections, const Layout::Data_list& special_outputs) { for(Layout::Section_list::const_iterator p = sections.begin(); p != sections.end(); ++p) gold_assert((*p)->address_and_file_offset_have_reset_values()); for(Layout::Data_list::const_iterator p = special_outputs.begin(); p != special_outputs.end(); ++p) gold_assert((*p)->address_and_file_offset_have_reset_values()); } // Save information of SECTIONS for checking later. void Layout::Relaxation_debug_check::read_sections( const Layout::Section_list& sections) { for(Layout::Section_list::const_iterator p = sections.begin(); p != sections.end(); ++p) { Output_section* os = *p; Section_info info; info.output_section = os; info.address = os->is_address_valid() ? os->address() : 0; info.data_size = os->is_data_size_valid() ? os->data_size() : -1; info.offset = os->is_offset_valid()? os->offset() : -1 ; this->section_infos_.push_back(info); } } // Verify SECTIONS using previously recorded information. void Layout::Relaxation_debug_check::verify_sections( const Layout::Section_list& sections) { size_t i = 0; for(Layout::Section_list::const_iterator p = sections.begin(); p != sections.end(); ++p, ++i) { Output_section* os = *p; uint64_t address = os->is_address_valid() ? os->address() : 0; off_t data_size = os->is_data_size_valid() ? os->data_size() : -1; off_t offset = os->is_offset_valid()? os->offset() : -1 ; if (i >= this->section_infos_.size()) { gold_fatal("Section_info of %s missing.\n", os->name()); } const Section_info& info = this->section_infos_[i]; if (os != info.output_section) gold_fatal("Section order changed. Expecting %s but see %s\n", info.output_section->name(), os->name()); if (address != info.address || data_size != info.data_size || offset != info.offset) gold_fatal("Section %s changed.\n", os->name()); } } // Layout_task_runner methods. // Lay out the sections. This is called after all the input objects // have been read. void Layout_task_runner::run(Workqueue* workqueue, const Task* task) { Layout* layout = this->layout_; off_t file_size = layout->finalize(this->input_objects_, this->symtab_, this->target_, task); // Now we know the final size of the output file and we know where // each piece of information goes. if (this->mapfile_ != NULL) { this->mapfile_->print_discarded_sections(this->input_objects_); layout->print_to_mapfile(this->mapfile_); } Output_file* of; if (layout->incremental_base() == NULL) { of = new Output_file(parameters->options().output_file_name()); if (this->options_.oformat_enum() != General_options::OBJECT_FORMAT_ELF) of->set_is_temporary(); of->open(file_size); } else { of = layout->incremental_base()->output_file(); // Apply the incremental relocations for symbols whose values // have changed. We do this before we resize the file and start // writing anything else to it, so that we can read the old // incremental information from the file before (possibly) // overwriting it. if (parameters->incremental_update()) layout->incremental_base()->apply_incremental_relocs(this->symtab_, this->layout_, of); of->resize(file_size); } // Queue up the final set of tasks. gold::queue_final_tasks(this->options_, this->input_objects_, this->symtab_, layout, workqueue, of); } // Layout methods. Layout::Layout(int number_of_input_files, Script_options* script_options) : number_of_input_files_(number_of_input_files), script_options_(script_options), namepool_(), sympool_(), dynpool_(), signatures_(), section_name_map_(), segment_list_(), section_list_(), unattached_section_list_(), special_output_list_(), section_headers_(NULL), tls_segment_(NULL), relro_segment_(NULL), increase_relro_(0), symtab_section_(NULL), symtab_xindex_(NULL), dynsym_section_(NULL), dynsym_xindex_(NULL), dynamic_section_(NULL), dynamic_symbol_(NULL), dynamic_data_(NULL), eh_frame_section_(NULL), eh_frame_data_(NULL), added_eh_frame_data_(false), eh_frame_hdr_section_(NULL), build_id_note_(NULL), debug_abbrev_(NULL), debug_info_(NULL), group_signatures_(), output_file_size_(-1), have_added_input_section_(false), sections_are_attached_(false), input_requires_executable_stack_(false), input_with_gnu_stack_note_(false), input_without_gnu_stack_note_(false), has_static_tls_(false), any_postprocessing_sections_(false), resized_signatures_(false), have_stabstr_section_(false), incremental_inputs_(NULL), record_output_section_data_from_script_(false), script_output_section_data_list_(), segment_states_(NULL), relaxation_debug_check_(NULL), incremental_base_(NULL), free_list_() { // Make space for more than enough segments for a typical file. // This is just for efficiency--it's OK if we wind up needing more. this->segment_list_.reserve(12); // We expect two unattached Output_data objects: the file header and // the segment headers. this->special_output_list_.reserve(2); // Initialize structure needed for an incremental build. if (parameters->incremental()) this->incremental_inputs_ = new Incremental_inputs; // The section name pool is worth optimizing in all cases, because // it is small, but there are often overlaps due to .rel sections. this->namepool_.set_optimize(); } // For incremental links, record the base file to be modified. void Layout::set_incremental_base(Incremental_binary* base) { this->incremental_base_ = base; this->free_list_.init(base->output_file()->filesize(), true); } // Hash a key we use to look up an output section mapping. size_t Layout::Hash_key::operator()(const Layout::Key& k) const { return k.first + k.second.first + k.second.second; } // Returns whether the given section is in the list of // debug-sections-used-by-some-version-of-gdb. Currently, // we've checked versions of gdb up to and including 6.7.1. static const char* gdb_sections[] = { ".debug_abbrev", // ".debug_aranges", // not used by gdb as of 6.7.1 ".debug_frame", ".debug_info", ".debug_types", ".debug_line", ".debug_loc", ".debug_macinfo", // ".debug_pubnames", // not used by gdb as of 6.7.1 ".debug_ranges", ".debug_str", }; static const char* lines_only_debug_sections[] = { ".debug_abbrev", // ".debug_aranges", // not used by gdb as of 6.7.1 // ".debug_frame", ".debug_info", // ".debug_types", ".debug_line", // ".debug_loc", // ".debug_macinfo", // ".debug_pubnames", // not used by gdb as of 6.7.1 // ".debug_ranges", ".debug_str", }; static inline bool is_gdb_debug_section(const char* str) { // We can do this faster: binary search or a hashtable. But why bother? for (size_t i = 0; i < sizeof(gdb_sections)/sizeof(*gdb_sections); ++i) if (strcmp(str, gdb_sections[i]) == 0) return true; return false; } static inline bool is_lines_only_debug_section(const char* str) { // We can do this faster: binary search or a hashtable. But why bother? for (size_t i = 0; i < sizeof(lines_only_debug_sections)/sizeof(*lines_only_debug_sections); ++i) if (strcmp(str, lines_only_debug_sections[i]) == 0) return true; return false; } // Sometimes we compress sections. This is typically done for // sections that are not part of normal program execution (such as // .debug_* sections), and where the readers of these sections know // how to deal with compressed sections. This routine doesn't say for // certain whether we'll compress -- it depends on commandline options // as well -- just whether this section is a candidate for compression. // (The Output_compressed_section class decides whether to compress // a given section, and picks the name of the compressed section.) static bool is_compressible_debug_section(const char* secname) { return (is_prefix_of(".debug", secname)); } // We may see compressed debug sections in input files. Return TRUE // if this is the name of a compressed debug section. bool is_compressed_debug_section(const char* secname) { return (is_prefix_of(".zdebug", secname)); } // Whether to include this section in the link. template<int size, bool big_endian> bool Layout::include_section(Sized_relobj_file<size, big_endian>*, const char* name, const elfcpp::Shdr<size, big_endian>& shdr) { if (shdr.get_sh_flags() & elfcpp::SHF_EXCLUDE) return false; switch (shdr.get_sh_type()) { case elfcpp::SHT_NULL: case elfcpp::SHT_SYMTAB: case elfcpp::SHT_DYNSYM: case elfcpp::SHT_HASH: case elfcpp::SHT_DYNAMIC: case elfcpp::SHT_SYMTAB_SHNDX: return false; case elfcpp::SHT_STRTAB: // Discard the sections which have special meanings in the ELF // ABI. Keep others (e.g., .stabstr). We could also do this by // checking the sh_link fields of the appropriate sections. return (strcmp(name, ".dynstr") != 0 && strcmp(name, ".strtab") != 0 && strcmp(name, ".shstrtab") != 0); case elfcpp::SHT_RELA: case elfcpp::SHT_REL: case elfcpp::SHT_GROUP: // If we are emitting relocations these should be handled // elsewhere. gold_assert(!parameters->options().relocatable() && !parameters->options().emit_relocs()); return false; case elfcpp::SHT_PROGBITS: if (parameters->options().strip_debug() && (shdr.get_sh_flags() & elfcpp::SHF_ALLOC) == 0) { if (is_debug_info_section(name)) return false; } if (parameters->options().strip_debug_non_line() && (shdr.get_sh_flags() & elfcpp::SHF_ALLOC) == 0) { // Debugging sections can only be recognized by name. if (is_prefix_of(".debug", name) && !is_lines_only_debug_section(name)) return false; } if (parameters->options().strip_debug_gdb() && (shdr.get_sh_flags() & elfcpp::SHF_ALLOC) == 0) { // Debugging sections can only be recognized by name. if (is_prefix_of(".debug", name) && !is_gdb_debug_section(name)) return false; } if (parameters->options().strip_lto_sections() && !parameters->options().relocatable() && (shdr.get_sh_flags() & elfcpp::SHF_ALLOC) == 0) { // Ignore LTO sections containing intermediate code. if (is_prefix_of(".gnu.lto_", name)) return false; } // The GNU linker strips .gnu_debuglink sections, so we do too. // This is a feature used to keep debugging information in // separate files. if (strcmp(name, ".gnu_debuglink") == 0) return false; return true; default: return true; } } // Return an output section named NAME, or NULL if there is none. Output_section* Layout::find_output_section(const char* name) const { for (Section_list::const_iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) if (strcmp((*p)->name(), name) == 0) return *p; return NULL; } // Return an output segment of type TYPE, with segment flags SET set // and segment flags CLEAR clear. Return NULL if there is none. Output_segment* Layout::find_output_segment(elfcpp::PT type, elfcpp::Elf_Word set, elfcpp::Elf_Word clear) const { for (Segment_list::const_iterator p = this->segment_list_.begin(); p != this->segment_list_.end(); ++p) if (static_cast<elfcpp::PT>((*p)->type()) == type && ((*p)->flags() & set) == set && ((*p)->flags() & clear) == 0) return *p; return NULL; } // Return the output section to use for section NAME with type TYPE // and section flags FLAGS. NAME must be canonicalized in the string // pool, and NAME_KEY is the key. IS_INTERP is true if this is the // .interp section. IS_DYNAMIC_LINKER_SECTION is true if this section // is used by the dynamic linker. IS_RELRO is true for a relro // section. IS_LAST_RELRO is true for the last relro section. // IS_FIRST_NON_RELRO is true for the first non-relro section. Output_section* Layout::get_output_section(const char* name, Stringpool::Key name_key, elfcpp::Elf_Word type, elfcpp::Elf_Xword flags, Output_section_order order, bool is_relro) { elfcpp::Elf_Xword lookup_flags = flags; // Ignoring SHF_WRITE and SHF_EXECINSTR here means that we combine // read-write with read-only sections. Some other ELF linkers do // not do this. FIXME: Perhaps there should be an option // controlling this. lookup_flags &= ~(elfcpp::SHF_WRITE | elfcpp::SHF_EXECINSTR); const Key key(name_key, std::make_pair(type, lookup_flags)); const std::pair<Key, Output_section*> v(key, NULL); std::pair<Section_name_map::iterator, bool> ins( this->section_name_map_.insert(v)); if (!ins.second) return ins.first->second; else { // This is the first time we've seen this name/type/flags // combination. For compatibility with the GNU linker, we // combine sections with contents and zero flags with sections // with non-zero flags. This is a workaround for cases where // assembler code forgets to set section flags. FIXME: Perhaps // there should be an option to control this. Output_section* os = NULL; if (type == elfcpp::SHT_PROGBITS) { if (flags == 0) { Output_section* same_name = this->find_output_section(name); if (same_name != NULL && same_name->type() == elfcpp::SHT_PROGBITS && (same_name->flags() & elfcpp::SHF_TLS) == 0) os = same_name; } else if ((flags & elfcpp::SHF_TLS) == 0) { elfcpp::Elf_Xword zero_flags = 0; const Key zero_key(name_key, std::make_pair(type, zero_flags)); Section_name_map::iterator p = this->section_name_map_.find(zero_key); if (p != this->section_name_map_.end()) os = p->second; } } if (os == NULL) os = this->make_output_section(name, type, flags, order, is_relro); ins.first->second = os; return os; } } // Pick the output section to use for section NAME, in input file // RELOBJ, with type TYPE and flags FLAGS. RELOBJ may be NULL for a // linker created section. IS_INPUT_SECTION is true if we are // choosing an output section for an input section found in a input // file. IS_INTERP is true if this is the .interp section. // IS_DYNAMIC_LINKER_SECTION is true if this section is used by the // dynamic linker. IS_RELRO is true for a relro section. // IS_LAST_RELRO is true for the last relro section. // IS_FIRST_NON_RELRO is true for the first non-relro section. This // will return NULL if the input section should be discarded. Output_section* Layout::choose_output_section(const Relobj* relobj, const char* name, elfcpp::Elf_Word type, elfcpp::Elf_Xword flags, bool is_input_section, Output_section_order order, bool is_relro) { // We should not see any input sections after we have attached // sections to segments. gold_assert(!is_input_section || !this->sections_are_attached_); // Some flags in the input section should not be automatically // copied to the output section. flags &= ~ (elfcpp::SHF_INFO_LINK | elfcpp::SHF_GROUP | elfcpp::SHF_MERGE | elfcpp::SHF_STRINGS); // We only clear the SHF_LINK_ORDER flag in for // a non-relocatable link. if (!parameters->options().relocatable()) flags &= ~elfcpp::SHF_LINK_ORDER; if (this->script_options_->saw_sections_clause()) { // We are using a SECTIONS clause, so the output section is // chosen based only on the name. Script_sections* ss = this->script_options_->script_sections(); const char* file_name = relobj == NULL ? NULL : relobj->name().c_str(); Output_section** output_section_slot; Script_sections::Section_type script_section_type; const char* orig_name = name; name = ss->output_section_name(file_name, name, &output_section_slot, &script_section_type); if (name == NULL) { gold_debug(DEBUG_SCRIPT, _("Unable to create output section '%s' " "because it is not allowed by the " "SECTIONS clause of the linker script"), orig_name); // The SECTIONS clause says to discard this input section. return NULL; } // We can only handle script section types ST_NONE and ST_NOLOAD. switch (script_section_type) { case Script_sections::ST_NONE: break; case Script_sections::ST_NOLOAD: flags &= elfcpp::SHF_ALLOC; break; default: gold_unreachable(); } // If this is an orphan section--one not mentioned in the linker // script--then OUTPUT_SECTION_SLOT will be NULL, and we do the // default processing below. if (output_section_slot != NULL) { if (*output_section_slot != NULL) { (*output_section_slot)->update_flags_for_input_section(flags); return *output_section_slot; } // We don't put sections found in the linker script into // SECTION_NAME_MAP_. That keeps us from getting confused // if an orphan section is mapped to a section with the same // name as one in the linker script. name = this->namepool_.add(name, false, NULL); Output_section* os = this->make_output_section(name, type, flags, order, is_relro); os->set_found_in_sections_clause(); // Special handling for NOLOAD sections. if (script_section_type == Script_sections::ST_NOLOAD) { os->set_is_noload(); // The constructor of Output_section sets addresses of non-ALLOC // sections to 0 by default. We don't want that for NOLOAD // sections even if they have no SHF_ALLOC flag. if ((os->flags() & elfcpp::SHF_ALLOC) == 0 && os->is_address_valid()) { gold_assert(os->address() == 0 && !os->is_offset_valid() && !os->is_data_size_valid()); os->reset_address_and_file_offset(); } } *output_section_slot = os; return os; } } // FIXME: Handle SHF_OS_NONCONFORMING somewhere. size_t len = strlen(name); char* uncompressed_name = NULL; // Compressed debug sections should be mapped to the corresponding // uncompressed section. if (is_compressed_debug_section(name)) { uncompressed_name = new char[len]; uncompressed_name[0] = '.'; gold_assert(name[0] == '.' && name[1] == 'z'); strncpy(&uncompressed_name[1], &name[2], len - 2); uncompressed_name[len - 1] = '\0'; len -= 1; name = uncompressed_name; } // Turn NAME from the name of the input section into the name of the // output section. if (is_input_section && !this->script_options_->saw_sections_clause() && !parameters->options().relocatable()) name = Layout::output_section_name(name, &len); Stringpool::Key name_key; name = this->namepool_.add_with_length(name, len, true, &name_key); if (uncompressed_name != NULL) delete[] uncompressed_name; // Find or make the output section. The output section is selected // based on the section name, type, and flags. return this->get_output_section(name, name_key, type, flags, order, is_relro); } // For incremental links, record the initial fixed layout of a section // from the base file, and return a pointer to the Output_section. template<int size, bool big_endian> Output_section* Layout::init_fixed_output_section(const char* name, elfcpp::Shdr<size, big_endian>& shdr) { unsigned int sh_type = shdr.get_sh_type(); // We preserve the layout of PROGBITS, NOBITS, and NOTE sections. // All others will be created from scratch and reallocated. if (sh_type != elfcpp::SHT_PROGBITS && sh_type != elfcpp::SHT_NOBITS && sh_type != elfcpp::SHT_NOTE) return NULL; typename elfcpp::Elf_types<size>::Elf_Addr sh_addr = shdr.get_sh_addr(); typename elfcpp::Elf_types<size>::Elf_Off sh_offset = shdr.get_sh_offset(); typename elfcpp::Elf_types<size>::Elf_WXword sh_size = shdr.get_sh_size(); typename elfcpp::Elf_types<size>::Elf_WXword sh_flags = shdr.get_sh_flags(); typename elfcpp::Elf_types<size>::Elf_WXword sh_addralign = shdr.get_sh_addralign(); // Make the output section. Stringpool::Key name_key; name = this->namepool_.add(name, true, &name_key); Output_section* os = this->get_output_section(name, name_key, sh_type, sh_flags, ORDER_INVALID, false); os->set_fixed_layout(sh_addr, sh_offset, sh_size, sh_addralign); if (sh_type != elfcpp::SHT_NOBITS) this->free_list_.remove(sh_offset, sh_offset + sh_size); return os; } // Return the output section to use for input section SHNDX, with name // NAME, with header HEADER, from object OBJECT. RELOC_SHNDX is the // index of a relocation section which applies to this section, or 0 // if none, or -1U if more than one. RELOC_TYPE is the type of the // relocation section if there is one. Set *OFF to the offset of this // input section without the output section. Return NULL if the // section should be discarded. Set *OFF to -1 if the section // contents should not be written directly to the output file, but // will instead receive special handling. template<int size, bool big_endian> Output_section* Layout::layout(Sized_relobj_file<size, big_endian>* object, unsigned int shndx, const char* name, const elfcpp::Shdr<size, big_endian>& shdr, unsigned int reloc_shndx, unsigned int, off_t* off) { *off = 0; if (!this->include_section(object, name, shdr)) return NULL; Output_section* os; // Sometimes .init_array*, .preinit_array* and .fini_array* do not have // correct section types. Force them here. elfcpp::Elf_Word sh_type = shdr.get_sh_type(); if (sh_type == elfcpp::SHT_PROGBITS) { static const char init_array_prefix[] = ".init_array"; static const char preinit_array_prefix[] = ".preinit_array"; static const char fini_array_prefix[] = ".fini_array"; static size_t init_array_prefix_size = sizeof(init_array_prefix) - 1; static size_t preinit_array_prefix_size = sizeof(preinit_array_prefix) - 1; static size_t fini_array_prefix_size = sizeof(fini_array_prefix) - 1; if (strncmp(name, init_array_prefix, init_array_prefix_size) == 0) sh_type = elfcpp::SHT_INIT_ARRAY; else if (strncmp(name, preinit_array_prefix, preinit_array_prefix_size) == 0) sh_type = elfcpp::SHT_PREINIT_ARRAY; else if (strncmp(name, fini_array_prefix, fini_array_prefix_size) == 0) sh_type = elfcpp::SHT_FINI_ARRAY; } // In a relocatable link a grouped section must not be combined with // any other sections. if (parameters->options().relocatable() && (shdr.get_sh_flags() & elfcpp::SHF_GROUP) != 0) { name = this->namepool_.add(name, true, NULL); os = this->make_output_section(name, sh_type, shdr.get_sh_flags(), ORDER_INVALID, false); } else { os = this->choose_output_section(object, name, sh_type, shdr.get_sh_flags(), true, ORDER_INVALID, false); if (os == NULL) return NULL; } // By default the GNU linker sorts input sections whose names match // .ctor.*, .dtor.*, .init_array.*, or .fini_array.*. The sections // are sorted by name. This is used to implement constructor // priority ordering. We are compatible. if (!this->script_options_->saw_sections_clause() && (is_prefix_of(".ctors.", name) || is_prefix_of(".dtors.", name) || is_prefix_of(".init_array.", name) || is_prefix_of(".fini_array.", name))) os->set_must_sort_attached_input_sections(); // FIXME: Handle SHF_LINK_ORDER somewhere. elfcpp::Elf_Xword orig_flags = os->flags(); *off = os->add_input_section(this, object, shndx, name, shdr, reloc_shndx, this->script_options_->saw_sections_clause()); // If the flags changed, we may have to change the order. if ((orig_flags & elfcpp::SHF_ALLOC) != 0) { orig_flags &= (elfcpp::SHF_WRITE | elfcpp::SHF_EXECINSTR); elfcpp::Elf_Xword new_flags = os->flags() & (elfcpp::SHF_WRITE | elfcpp::SHF_EXECINSTR); if (orig_flags != new_flags) os->set_order(this->default_section_order(os, false)); } this->have_added_input_section_ = true; return os; } // Handle a relocation section when doing a relocatable link. template<int size, bool big_endian> Output_section* Layout::layout_reloc(Sized_relobj_file<size, big_endian>* object, unsigned int, const elfcpp::Shdr<size, big_endian>& shdr, Output_section* data_section, Relocatable_relocs* rr) { gold_assert(parameters->options().relocatable() || parameters->options().emit_relocs()); int sh_type = shdr.get_sh_type(); std::string name; if (sh_type == elfcpp::SHT_REL) name = ".rel"; else if (sh_type == elfcpp::SHT_RELA) name = ".rela"; else gold_unreachable(); name += data_section->name(); // In a relocatable link relocs for a grouped section must not be // combined with other reloc sections. Output_section* os; if (!parameters->options().relocatable() || (data_section->flags() & elfcpp::SHF_GROUP) == 0) os = this->choose_output_section(object, name.c_str(), sh_type, shdr.get_sh_flags(), false, ORDER_INVALID, false); else { const char* n = this->namepool_.add(name.c_str(), true, NULL); os = this->make_output_section(n, sh_type, shdr.get_sh_flags(), ORDER_INVALID, false); } os->set_should_link_to_symtab(); os->set_info_section(data_section); Output_section_data* posd; if (sh_type == elfcpp::SHT_REL) { os->set_entsize(elfcpp::Elf_sizes<size>::rel_size); posd = new Output_relocatable_relocs<elfcpp::SHT_REL, size, big_endian>(rr); } else if (sh_type == elfcpp::SHT_RELA) { os->set_entsize(elfcpp::Elf_sizes<size>::rela_size); posd = new Output_relocatable_relocs<elfcpp::SHT_RELA, size, big_endian>(rr); } else gold_unreachable(); os->add_output_section_data(posd); rr->set_output_data(posd); return os; } // Handle a group section when doing a relocatable link. template<int size, bool big_endian> void Layout::layout_group(Symbol_table* symtab, Sized_relobj_file<size, big_endian>* object, unsigned int, const char* group_section_name, const char* signature, const elfcpp::Shdr<size, big_endian>& shdr, elfcpp::Elf_Word flags, std::vector<unsigned int>* shndxes) { gold_assert(parameters->options().relocatable()); gold_assert(shdr.get_sh_type() == elfcpp::SHT_GROUP); group_section_name = this->namepool_.add(group_section_name, true, NULL); Output_section* os = this->make_output_section(group_section_name, elfcpp::SHT_GROUP, shdr.get_sh_flags(), ORDER_INVALID, false); // We need to find a symbol with the signature in the symbol table. // If we don't find one now, we need to look again later. Symbol* sym = symtab->lookup(signature, NULL); if (sym != NULL) os->set_info_symndx(sym); else { // Reserve some space to minimize reallocations. if (this->group_signatures_.empty()) this->group_signatures_.reserve(this->number_of_input_files_ * 16); // We will wind up using a symbol whose name is the signature. // So just put the signature in the symbol name pool to save it. signature = symtab->canonicalize_name(signature); this->group_signatures_.push_back(Group_signature(os, signature)); } os->set_should_link_to_symtab(); os->set_entsize(4); section_size_type entry_count = convert_to_section_size_type(shdr.get_sh_size() / 4); Output_section_data* posd = new Output_data_group<size, big_endian>(object, entry_count, flags, shndxes); os->add_output_section_data(posd); } // Special GNU handling of sections name .eh_frame. They will // normally hold exception frame data as defined by the C++ ABI // (http://codesourcery.com/cxx-abi/). template<int size, bool big_endian> Output_section* Layout::layout_eh_frame(Sized_relobj_file<size, big_endian>* object, const unsigned char* symbols, off_t symbols_size, const unsigned char* symbol_names, off_t symbol_names_size, unsigned int shndx, const elfcpp::Shdr<size, big_endian>& shdr, unsigned int reloc_shndx, unsigned int reloc_type, off_t* off) { gold_assert(shdr.get_sh_type() == elfcpp::SHT_PROGBITS); gold_assert((shdr.get_sh_flags() & elfcpp::SHF_ALLOC) != 0); const char* const name = ".eh_frame"; Output_section* os = this->choose_output_section(object, name, elfcpp::SHT_PROGBITS, elfcpp::SHF_ALLOC, false, ORDER_EHFRAME, false); if (os == NULL) return NULL; if (this->eh_frame_section_ == NULL) { this->eh_frame_section_ = os; this->eh_frame_data_ = new Eh_frame(); // For incremental linking, we do not optimize .eh_frame sections // or create a .eh_frame_hdr section. if (parameters->options().eh_frame_hdr() && !parameters->incremental()) { Output_section* hdr_os = this->choose_output_section(NULL, ".eh_frame_hdr", elfcpp::SHT_PROGBITS, elfcpp::SHF_ALLOC, false, ORDER_EHFRAME, false); if (hdr_os != NULL) { Eh_frame_hdr* hdr_posd = new Eh_frame_hdr(os, this->eh_frame_data_); hdr_os->add_output_section_data(hdr_posd); hdr_os->set_after_input_sections(); if (!this->script_options_->saw_phdrs_clause()) { Output_segment* hdr_oseg; hdr_oseg = this->make_output_segment(elfcpp::PT_GNU_EH_FRAME, elfcpp::PF_R); hdr_oseg->add_output_section_to_nonload(hdr_os, elfcpp::PF_R); } this->eh_frame_data_->set_eh_frame_hdr(hdr_posd); } } } gold_assert(this->eh_frame_section_ == os); if (!parameters->incremental() && this->eh_frame_data_->add_ehframe_input_section(object, symbols, symbols_size, symbol_names, symbol_names_size, shndx, reloc_shndx, reloc_type)) { os->update_flags_for_input_section(shdr.get_sh_flags()); // A writable .eh_frame section is a RELRO section. if ((shdr.get_sh_flags() & elfcpp::SHF_WRITE) != 0) os->set_is_relro(); // We found a .eh_frame section we are going to optimize, so now // we can add the set of optimized sections to the output // section. We need to postpone adding this until we've found a // section we can optimize so that the .eh_frame section in // crtbegin.o winds up at the start of the output section. if (!this->added_eh_frame_data_) { os->add_output_section_data(this->eh_frame_data_); this->added_eh_frame_data_ = true; } *off = -1; } else { // We couldn't handle this .eh_frame section for some reason. // Add it as a normal section. bool saw_sections_clause = this->script_options_->saw_sections_clause(); *off = os->add_input_section(this, object, shndx, name, shdr, reloc_shndx, saw_sections_clause); this->have_added_input_section_ = true; } return os; } // Add POSD to an output section using NAME, TYPE, and FLAGS. Return // the output section. Output_section* Layout::add_output_section_data(const char* name, elfcpp::Elf_Word type, elfcpp::Elf_Xword flags, Output_section_data* posd, Output_section_order order, bool is_relro) { Output_section* os = this->choose_output_section(NULL, name, type, flags, false, order, is_relro); if (os != NULL) os->add_output_section_data(posd); return os; } // Map section flags to segment flags. elfcpp::Elf_Word Layout::section_flags_to_segment(elfcpp::Elf_Xword flags) { elfcpp::Elf_Word ret = elfcpp::PF_R; if ((flags & elfcpp::SHF_WRITE) != 0) ret |= elfcpp::PF_W; if ((flags & elfcpp::SHF_EXECINSTR) != 0) ret |= elfcpp::PF_X; return ret; } // Make a new Output_section, and attach it to segments as // appropriate. ORDER is the order in which this section should // appear in the output segment. IS_RELRO is true if this is a relro // (read-only after relocations) section. Output_section* Layout::make_output_section(const char* name, elfcpp::Elf_Word type, elfcpp::Elf_Xword flags, Output_section_order order, bool is_relro) { Output_section* os; if ((flags & elfcpp::SHF_ALLOC) == 0 && strcmp(parameters->options().compress_debug_sections(), "none") != 0 && is_compressible_debug_section(name)) os = new Output_compressed_section(¶meters->options(), name, type, flags); else if ((flags & elfcpp::SHF_ALLOC) == 0 && parameters->options().strip_debug_non_line() && strcmp(".debug_abbrev", name) == 0) { os = this->debug_abbrev_ = new Output_reduced_debug_abbrev_section( name, type, flags); if (this->debug_info_) this->debug_info_->set_abbreviations(this->debug_abbrev_); } else if ((flags & elfcpp::SHF_ALLOC) == 0 && parameters->options().strip_debug_non_line() && strcmp(".debug_info", name) == 0) { os = this->debug_info_ = new Output_reduced_debug_info_section( name, type, flags); if (this->debug_abbrev_) this->debug_info_->set_abbreviations(this->debug_abbrev_); } else { // FIXME: const_cast is ugly. Target* target = const_cast<Target*>(¶meters->target()); os = target->make_output_section(name, type, flags); } // With -z relro, we have to recognize the special sections by name. // There is no other way. bool is_relro_local = false; if (!this->script_options_->saw_sections_clause() && parameters->options().relro() && type == elfcpp::SHT_PROGBITS && (flags & elfcpp::SHF_ALLOC) != 0 && (flags & elfcpp::SHF_WRITE) != 0) { if (strcmp(name, ".data.rel.ro") == 0) is_relro = true; else if (strcmp(name, ".data.rel.ro.local") == 0) { is_relro = true; is_relro_local = true; } else if (type == elfcpp::SHT_INIT_ARRAY || type == elfcpp::SHT_FINI_ARRAY || type == elfcpp::SHT_PREINIT_ARRAY) is_relro = true; else if (strcmp(name, ".ctors") == 0 || strcmp(name, ".dtors") == 0 || strcmp(name, ".jcr") == 0) is_relro = true; } if (is_relro) os->set_is_relro(); if (order == ORDER_INVALID && (flags & elfcpp::SHF_ALLOC) != 0) order = this->default_section_order(os, is_relro_local); os->set_order(order); parameters->target().new_output_section(os); this->section_list_.push_back(os); // The GNU linker by default sorts some sections by priority, so we // do the same. We need to know that this might happen before we // attach any input sections. if (!this->script_options_->saw_sections_clause() && (strcmp(name, ".ctors") == 0 || strcmp(name, ".dtors") == 0 || strcmp(name, ".init_array") == 0 || strcmp(name, ".fini_array") == 0)) os->set_may_sort_attached_input_sections(); // Check for .stab*str sections, as .stab* sections need to link to // them. if (type == elfcpp::SHT_STRTAB && !this->have_stabstr_section_ && strncmp(name, ".stab", 5) == 0 && strcmp(name + strlen(name) - 3, "str") == 0) this->have_stabstr_section_ = true; // If we have already attached the sections to segments, then we // need to attach this one now. This happens for sections created // directly by the linker. if (this->sections_are_attached_) this->attach_section_to_segment(os); return os; } // Return the default order in which a section should be placed in an // output segment. This function captures a lot of the ideas in // ld/scripttempl/elf.sc in the GNU linker. Note that the order of a // linker created section is normally set when the section is created; // this function is used for input sections. Output_section_order Layout::default_section_order(Output_section* os, bool is_relro_local) { gold_assert((os->flags() & elfcpp::SHF_ALLOC) != 0); bool is_write = (os->flags() & elfcpp::SHF_WRITE) != 0; bool is_execinstr = (os->flags() & elfcpp::SHF_EXECINSTR) != 0; bool is_bss = false; switch (os->type()) { default: case elfcpp::SHT_PROGBITS: break; case elfcpp::SHT_NOBITS: is_bss = true; break; case elfcpp::SHT_RELA: case elfcpp::SHT_REL: if (!is_write) return ORDER_DYNAMIC_RELOCS; break; case elfcpp::SHT_HASH: case elfcpp::SHT_DYNAMIC: case elfcpp::SHT_SHLIB: case elfcpp::SHT_DYNSYM: case elfcpp::SHT_GNU_HASH: case elfcpp::SHT_GNU_verdef: case elfcpp::SHT_GNU_verneed: case elfcpp::SHT_GNU_versym: if (!is_write) return ORDER_DYNAMIC_LINKER; break; case elfcpp::SHT_NOTE: return is_write ? ORDER_RW_NOTE : ORDER_RO_NOTE; } if ((os->flags() & elfcpp::SHF_TLS) != 0) return is_bss ? ORDER_TLS_BSS : ORDER_TLS_DATA; if (!is_bss && !is_write) { if (is_execinstr) { if (strcmp(os->name(), ".init") == 0) return ORDER_INIT; else if (strcmp(os->name(), ".fini") == 0) return ORDER_FINI; } return is_execinstr ? ORDER_TEXT : ORDER_READONLY; } if (os->is_relro()) return is_relro_local ? ORDER_RELRO_LOCAL : ORDER_RELRO; if (os->is_small_section()) return is_bss ? ORDER_SMALL_BSS : ORDER_SMALL_DATA; if (os->is_large_section()) return is_bss ? ORDER_LARGE_BSS : ORDER_LARGE_DATA; return is_bss ? ORDER_BSS : ORDER_DATA; } // Attach output sections to segments. This is called after we have // seen all the input sections. void Layout::attach_sections_to_segments() { for (Section_list::iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) this->attach_section_to_segment(*p); this->sections_are_attached_ = true; } // Attach an output section to a segment. void Layout::attach_section_to_segment(Output_section* os) { if ((os->flags() & elfcpp::SHF_ALLOC) == 0) this->unattached_section_list_.push_back(os); else this->attach_allocated_section_to_segment(os); } // Attach an allocated output section to a segment. void Layout::attach_allocated_section_to_segment(Output_section* os) { elfcpp::Elf_Xword flags = os->flags(); gold_assert((flags & elfcpp::SHF_ALLOC) != 0); if (parameters->options().relocatable()) return; // If we have a SECTIONS clause, we can't handle the attachment to // segments until after we've seen all the sections. if (this->script_options_->saw_sections_clause()) return; gold_assert(!this->script_options_->saw_phdrs_clause()); // This output section goes into a PT_LOAD segment. elfcpp::Elf_Word seg_flags = Layout::section_flags_to_segment(flags); // Check for --section-start. uint64_t addr; bool is_address_set = parameters->options().section_start(os->name(), &addr); // In general the only thing we really care about for PT_LOAD // segments is whether or not they are writable or executable, // so that is how we search for them. // Large data sections also go into their own PT_LOAD segment. // People who need segments sorted on some other basis will // have to use a linker script. Segment_list::const_iterator p; for (p = this->segment_list_.begin(); p != this->segment_list_.end(); ++p) { if ((*p)->type() != elfcpp::PT_LOAD) continue; if (!parameters->options().omagic() && ((*p)->flags() & elfcpp::PF_W) != (seg_flags & elfcpp::PF_W)) continue; if (parameters->options().rosegment() && ((*p)->flags() & elfcpp::PF_X) != (seg_flags & elfcpp::PF_X)) continue; // If -Tbss was specified, we need to separate the data and BSS // segments. if (parameters->options().user_set_Tbss()) { if ((os->type() == elfcpp::SHT_NOBITS) == (*p)->has_any_data_sections()) continue; } if (os->is_large_data_section() && !(*p)->is_large_data_segment()) continue; if (is_address_set) { if ((*p)->are_addresses_set()) continue; (*p)->add_initial_output_data(os); (*p)->update_flags_for_output_section(seg_flags); (*p)->set_addresses(addr, addr); break; } (*p)->add_output_section_to_load(this, os, seg_flags); break; } if (p == this->segment_list_.end()) { Output_segment* oseg = this->make_output_segment(elfcpp::PT_LOAD, seg_flags); if (os->is_large_data_section()) oseg->set_is_large_data_segment(); oseg->add_output_section_to_load(this, os, seg_flags); if (is_address_set) oseg->set_addresses(addr, addr); } // If we see a loadable SHT_NOTE section, we create a PT_NOTE // segment. if (os->type() == elfcpp::SHT_NOTE) { // See if we already have an equivalent PT_NOTE segment. for (p = this->segment_list_.begin(); p != segment_list_.end(); ++p) { if ((*p)->type() == elfcpp::PT_NOTE && (((*p)->flags() & elfcpp::PF_W) == (seg_flags & elfcpp::PF_W))) { (*p)->add_output_section_to_nonload(os, seg_flags); break; } } if (p == this->segment_list_.end()) { Output_segment* oseg = this->make_output_segment(elfcpp::PT_NOTE, seg_flags); oseg->add_output_section_to_nonload(os, seg_flags); } } // If we see a loadable SHF_TLS section, we create a PT_TLS // segment. There can only be one such segment. if ((flags & elfcpp::SHF_TLS) != 0) { if (this->tls_segment_ == NULL) this->make_output_segment(elfcpp::PT_TLS, seg_flags); this->tls_segment_->add_output_section_to_nonload(os, seg_flags); } // If -z relro is in effect, and we see a relro section, we create a // PT_GNU_RELRO segment. There can only be one such segment. if (os->is_relro() && parameters->options().relro()) { gold_assert(seg_flags == (elfcpp::PF_R | elfcpp::PF_W)); if (this->relro_segment_ == NULL) this->make_output_segment(elfcpp::PT_GNU_RELRO, seg_flags); this->relro_segment_->add_output_section_to_nonload(os, seg_flags); } } // Make an output section for a script. Output_section* Layout::make_output_section_for_script( const char* name, Script_sections::Section_type section_type) { name = this->namepool_.add(name, false, NULL); elfcpp::Elf_Xword sh_flags = elfcpp::SHF_ALLOC; if (section_type == Script_sections::ST_NOLOAD) sh_flags = 0; Output_section* os = this->make_output_section(name, elfcpp::SHT_PROGBITS, sh_flags, ORDER_INVALID, false); os->set_found_in_sections_clause(); if (section_type == Script_sections::ST_NOLOAD) os->set_is_noload(); return os; } // Return the number of segments we expect to see. size_t Layout::expected_segment_count() const { size_t ret = this->segment_list_.size(); // If we didn't see a SECTIONS clause in a linker script, we should // already have the complete list of segments. Otherwise we ask the // SECTIONS clause how many segments it expects, and add in the ones // we already have (PT_GNU_STACK, PT_GNU_EH_FRAME, etc.) if (!this->script_options_->saw_sections_clause()) return ret; else { const Script_sections* ss = this->script_options_->script_sections(); return ret + ss->expected_segment_count(this); } } // Handle the .note.GNU-stack section at layout time. SEEN_GNU_STACK // is whether we saw a .note.GNU-stack section in the object file. // GNU_STACK_FLAGS is the section flags. The flags give the // protection required for stack memory. We record this in an // executable as a PT_GNU_STACK segment. If an object file does not // have a .note.GNU-stack segment, we must assume that it is an old // object. On some targets that will force an executable stack. void Layout::layout_gnu_stack(bool seen_gnu_stack, uint64_t gnu_stack_flags, const Object* obj) { if (!seen_gnu_stack) { this->input_without_gnu_stack_note_ = true; if (parameters->options().warn_execstack() && parameters->target().is_default_stack_executable()) gold_warning(_("%s: missing .note.GNU-stack section" " implies executable stack"), obj->name().c_str()); } else { this->input_with_gnu_stack_note_ = true; if ((gnu_stack_flags & elfcpp::SHF_EXECINSTR) != 0) { this->input_requires_executable_stack_ = true; if (parameters->options().warn_execstack() || parameters->options().is_stack_executable()) gold_warning(_("%s: requires executable stack"), obj->name().c_str()); } } } // Create automatic note sections. void Layout::create_notes() { this->create_gold_note(); this->create_executable_stack_info(); this->create_build_id(); } // Create the dynamic sections which are needed before we read the // relocs. void Layout::create_initial_dynamic_sections(Symbol_table* symtab) { if (parameters->doing_static_link()) return; this->dynamic_section_ = this->choose_output_section(NULL, ".dynamic", elfcpp::SHT_DYNAMIC, (elfcpp::SHF_ALLOC | elfcpp::SHF_WRITE), false, ORDER_RELRO, true); this->dynamic_symbol_ = symtab->define_in_output_data("_DYNAMIC", NULL, Symbol_table::PREDEFINED, this->dynamic_section_, 0, 0, elfcpp::STT_OBJECT, elfcpp::STB_LOCAL, elfcpp::STV_HIDDEN, 0, false, false); this->dynamic_data_ = new Output_data_dynamic(&this->dynpool_); this->dynamic_section_->add_output_section_data(this->dynamic_data_); } // For each output section whose name can be represented as C symbol, // define __start and __stop symbols for the section. This is a GNU // extension. void Layout::define_section_symbols(Symbol_table* symtab) { for (Section_list::const_iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) { const char* const name = (*p)->name(); if (is_cident(name)) { const std::string name_string(name); const std::string start_name(cident_section_start_prefix + name_string); const std::string stop_name(cident_section_stop_prefix + name_string); symtab->define_in_output_data(start_name.c_str(), NULL, // version Symbol_table::PREDEFINED, *p, 0, // value 0, // symsize elfcpp::STT_NOTYPE, elfcpp::STB_GLOBAL, elfcpp::STV_DEFAULT, 0, // nonvis false, // offset_is_from_end true); // only_if_ref symtab->define_in_output_data(stop_name.c_str(), NULL, // version Symbol_table::PREDEFINED, *p, 0, // value 0, // symsize elfcpp::STT_NOTYPE, elfcpp::STB_GLOBAL, elfcpp::STV_DEFAULT, 0, // nonvis true, // offset_is_from_end true); // only_if_ref } } } // Define symbols for group signatures. void Layout::define_group_signatures(Symbol_table* symtab) { for (Group_signatures::iterator p = this->group_signatures_.begin(); p != this->group_signatures_.end(); ++p) { Symbol* sym = symtab->lookup(p->signature, NULL); if (sym != NULL) p->section->set_info_symndx(sym); else { // Force the name of the group section to the group // signature, and use the group's section symbol as the // signature symbol. if (strcmp(p->section->name(), p->signature) != 0) { const char* name = this->namepool_.add(p->signature, true, NULL); p->section->set_name(name); } p->section->set_needs_symtab_index(); p->section->set_info_section_symndx(p->section); } } this->group_signatures_.clear(); } // Find the first read-only PT_LOAD segment, creating one if // necessary. Output_segment* Layout::find_first_load_seg() { Output_segment* best = NULL; for (Segment_list::const_iterator p = this->segment_list_.begin(); p != this->segment_list_.end(); ++p) { if ((*p)->type() == elfcpp::PT_LOAD && ((*p)->flags() & elfcpp::PF_R) != 0 && (parameters->options().omagic() || ((*p)->flags() & elfcpp::PF_W) == 0)) { if (best == NULL || this->segment_precedes(*p, best)) best = *p; } } if (best != NULL) return best; gold_assert(!this->script_options_->saw_phdrs_clause()); Output_segment* load_seg = this->make_output_segment(elfcpp::PT_LOAD, elfcpp::PF_R); return load_seg; } // Save states of all current output segments. Store saved states // in SEGMENT_STATES. void Layout::save_segments(Segment_states* segment_states) { for (Segment_list::const_iterator p = this->segment_list_.begin(); p != this->segment_list_.end(); ++p) { Output_segment* segment = *p; // Shallow copy. Output_segment* copy = new Output_segment(*segment); (*segment_states)[segment] = copy; } } // Restore states of output segments and delete any segment not found in // SEGMENT_STATES. void Layout::restore_segments(const Segment_states* segment_states) { // Go through the segment list and remove any segment added in the // relaxation loop. this->tls_segment_ = NULL; this->relro_segment_ = NULL; Segment_list::iterator list_iter = this->segment_list_.begin(); while (list_iter != this->segment_list_.end()) { Output_segment* segment = *list_iter; Segment_states::const_iterator states_iter = segment_states->find(segment); if (states_iter != segment_states->end()) { const Output_segment* copy = states_iter->second; // Shallow copy to restore states. *segment = *copy; // Also fix up TLS and RELRO segment pointers as appropriate. if (segment->type() == elfcpp::PT_TLS) this->tls_segment_ = segment; else if (segment->type() == elfcpp::PT_GNU_RELRO) this->relro_segment_ = segment; ++list_iter; } else { list_iter = this->segment_list_.erase(list_iter); // This is a segment created during section layout. It should be // safe to remove it since we should have removed all pointers to it. delete segment; } } } // Clean up after relaxation so that sections can be laid out again. void Layout::clean_up_after_relaxation() { // Restore the segments to point state just prior to the relaxation loop. Script_sections* script_section = this->script_options_->script_sections(); script_section->release_segments(); this->restore_segments(this->segment_states_); // Reset section addresses and file offsets for (Section_list::iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) { (*p)->restore_states(); // If an input section changes size because of relaxation, // we need to adjust the section offsets of all input sections. // after such a section. if ((*p)->section_offsets_need_adjustment()) (*p)->adjust_section_offsets(); (*p)->reset_address_and_file_offset(); } // Reset special output object address and file offsets. for (Data_list::iterator p = this->special_output_list_.begin(); p != this->special_output_list_.end(); ++p) (*p)->reset_address_and_file_offset(); // A linker script may have created some output section data objects. // They are useless now. for (Output_section_data_list::const_iterator p = this->script_output_section_data_list_.begin(); p != this->script_output_section_data_list_.end(); ++p) delete *p; this->script_output_section_data_list_.clear(); } // Prepare for relaxation. void Layout::prepare_for_relaxation() { // Create an relaxation debug check if in debugging mode. if (is_debugging_enabled(DEBUG_RELAXATION)) this->relaxation_debug_check_ = new Relaxation_debug_check(); // Save segment states. this->segment_states_ = new Segment_states(); this->save_segments(this->segment_states_); for(Section_list::const_iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) (*p)->save_states(); if (is_debugging_enabled(DEBUG_RELAXATION)) this->relaxation_debug_check_->check_output_data_for_reset_values( this->section_list_, this->special_output_list_); // Also enable recording of output section data from scripts. this->record_output_section_data_from_script_ = true; } // Relaxation loop body: If target has no relaxation, this runs only once // Otherwise, the target relaxation hook is called at the end of // each iteration. If the hook returns true, it means re-layout of // section is required. // // The number of segments created by a linking script without a PHDRS // clause may be affected by section sizes and alignments. There is // a remote chance that relaxation causes different number of PT_LOAD // segments are created and sections are attached to different segments. // Therefore, we always throw away all segments created during section // layout. In order to be able to restart the section layout, we keep // a copy of the segment list right before the relaxation loop and use // that to restore the segments. // // PASS is the current relaxation pass number. // SYMTAB is a symbol table. // PLOAD_SEG is the address of a pointer for the load segment. // PHDR_SEG is a pointer to the PHDR segment. // SEGMENT_HEADERS points to the output segment header. // FILE_HEADER points to the output file header. // PSHNDX is the address to store the output section index. off_t inline Layout::relaxation_loop_body( int pass, Target* target, Symbol_table* symtab, Output_segment** pload_seg, Output_segment* phdr_seg, Output_segment_headers* segment_headers, Output_file_header* file_header, unsigned int* pshndx) { // If this is not the first iteration, we need to clean up after // relaxation so that we can lay out the sections again. if (pass != 0) this->clean_up_after_relaxation(); // If there is a SECTIONS clause, put all the input sections into // the required order. Output_segment* load_seg; if (this->script_options_->saw_sections_clause()) load_seg = this->set_section_addresses_from_script(symtab); else if (parameters->options().relocatable()) load_seg = NULL; else load_seg = this->find_first_load_seg(); if (parameters->options().oformat_enum() != General_options::OBJECT_FORMAT_ELF) load_seg = NULL; // If the user set the address of the text segment, that may not be // compatible with putting the segment headers and file headers into // that segment. if (parameters->options().user_set_Ttext()) load_seg = NULL; gold_assert(phdr_seg == NULL || load_seg != NULL || this->script_options_->saw_sections_clause()); // If the address of the load segment we found has been set by // --section-start rather than by a script, then adjust the VMA and // LMA downward if possible to include the file and section headers. uint64_t header_gap = 0; if (load_seg != NULL && load_seg->are_addresses_set() && !this->script_options_->saw_sections_clause() && !parameters->options().relocatable()) { file_header->finalize_data_size(); segment_headers->finalize_data_size(); size_t sizeof_headers = (file_header->data_size() + segment_headers->data_size()); const uint64_t abi_pagesize = target->abi_pagesize(); uint64_t hdr_paddr = load_seg->paddr() - sizeof_headers; hdr_paddr &= ~(abi_pagesize - 1); uint64_t subtract = load_seg->paddr() - hdr_paddr; if (load_seg->paddr() < subtract || load_seg->vaddr() < subtract) load_seg = NULL; else { load_seg->set_addresses(load_seg->vaddr() - subtract, load_seg->paddr() - subtract); header_gap = subtract - sizeof_headers; } } // Lay out the segment headers. if (!parameters->options().relocatable()) { gold_assert(segment_headers != NULL); if (header_gap != 0 && load_seg != NULL) { Output_data_zero_fill* z = new Output_data_zero_fill(header_gap, 1); load_seg->add_initial_output_data(z); } if (load_seg != NULL) load_seg->add_initial_output_data(segment_headers); if (phdr_seg != NULL) phdr_seg->add_initial_output_data(segment_headers); } // Lay out the file header. if (load_seg != NULL) load_seg->add_initial_output_data(file_header); if (this->script_options_->saw_phdrs_clause() && !parameters->options().relocatable()) { // Support use of FILEHDRS and PHDRS attachments in a PHDRS // clause in a linker script. Script_sections* ss = this->script_options_->script_sections(); ss->put_headers_in_phdrs(file_header, segment_headers); } // We set the output section indexes in set_segment_offsets and // set_section_indexes. *pshndx = 1; // Set the file offsets of all the segments, and all the sections // they contain. off_t off; if (!parameters->options().relocatable()) off = this->set_segment_offsets(target, load_seg, pshndx); else off = this->set_relocatable_section_offsets(file_header, pshndx); // Verify that the dummy relaxation does not change anything. if (is_debugging_enabled(DEBUG_RELAXATION)) { if (pass == 0) this->relaxation_debug_check_->read_sections(this->section_list_); else this->relaxation_debug_check_->verify_sections(this->section_list_); } *pload_seg = load_seg; return off; } // Search the list of patterns and find the postion of the given section // name in the output section. If the section name matches a glob // pattern and a non-glob name, then the non-glob position takes // precedence. Return 0 if no match is found. unsigned int Layout::find_section_order_index(const std::string& section_name) { Unordered_map<std::string, unsigned int>::iterator map_it; map_it = this->input_section_position_.find(section_name); if (map_it != this->input_section_position_.end()) return map_it->second; // Absolute match failed. Linear search the glob patterns. std::vector<std::string>::iterator it; for (it = this->input_section_glob_.begin(); it != this->input_section_glob_.end(); ++it) { if (fnmatch((*it).c_str(), section_name.c_str(), FNM_NOESCAPE) == 0) { map_it = this->input_section_position_.find(*it); gold_assert(map_it != this->input_section_position_.end()); return map_it->second; } } return 0; } // Read the sequence of input sections from the file specified with // --section-ordering-file. void Layout::read_layout_from_file() { const char* filename = parameters->options().section_ordering_file(); std::ifstream in; std::string line; in.open(filename); if (!in) gold_fatal(_("unable to open --section-ordering-file file %s: %s"), filename, strerror(errno)); std::getline(in, line); // this chops off the trailing \n, if any unsigned int position = 1; while (in) { if (!line.empty() && line[line.length() - 1] == '\r') // Windows line.resize(line.length() - 1); // Ignore comments, beginning with '#' if (line[0] == '#') { std::getline(in, line); continue; } this->input_section_position_[line] = position; // Store all glob patterns in a vector. if (is_wildcard_string(line.c_str())) this->input_section_glob_.push_back(line); position++; std::getline(in, line); } } // Finalize the layout. When this is called, we have created all the // output sections and all the output segments which are based on // input sections. We have several things to do, and we have to do // them in the right order, so that we get the right results correctly // and efficiently. // 1) Finalize the list of output segments and create the segment // table header. // 2) Finalize the dynamic symbol table and associated sections. // 3) Determine the final file offset of all the output segments. // 4) Determine the final file offset of all the SHF_ALLOC output // sections. // 5) Create the symbol table sections and the section name table // section. // 6) Finalize the symbol table: set symbol values to their final // value and make a final determination of which symbols are going // into the output symbol table. // 7) Create the section table header. // 8) Determine the final file offset of all the output sections which // are not SHF_ALLOC, including the section table header. // 9) Finalize the ELF file header. // This function returns the size of the output file. off_t Layout::finalize(const Input_objects* input_objects, Symbol_table* symtab, Target* target, const Task* task) { target->finalize_sections(this, input_objects, symtab); this->count_local_symbols(task, input_objects); this->link_stabs_sections(); Output_segment* phdr_seg = NULL; if (!parameters->options().relocatable() && !parameters->doing_static_link()) { // There was a dynamic object in the link. We need to create // some information for the dynamic linker. // Create the PT_PHDR segment which will hold the program // headers. if (!this->script_options_->saw_phdrs_clause()) phdr_seg = this->make_output_segment(elfcpp::PT_PHDR, elfcpp::PF_R); // Create the dynamic symbol table, including the hash table. Output_section* dynstr; std::vector<Symbol*> dynamic_symbols; unsigned int local_dynamic_count; Versions versions(*this->script_options()->version_script_info(), &this->dynpool_); this->create_dynamic_symtab(input_objects, symtab, &dynstr, &local_dynamic_count, &dynamic_symbols, &versions); // Create the .interp section to hold the name of the // interpreter, and put it in a PT_INTERP segment. if (!parameters->options().shared()) this->create_interp(target); // Finish the .dynamic section to hold the dynamic data, and put // it in a PT_DYNAMIC segment. this->finish_dynamic_section(input_objects, symtab); // We should have added everything we need to the dynamic string // table. this->dynpool_.set_string_offsets(); // Create the version sections. We can't do this until the // dynamic string table is complete. this->create_version_sections(&versions, symtab, local_dynamic_count, dynamic_symbols, dynstr); // Set the size of the _DYNAMIC symbol. We can't do this until // after we call create_version_sections. this->set_dynamic_symbol_size(symtab); } // Create segment headers. Output_segment_headers* segment_headers = (parameters->options().relocatable() ? NULL : new Output_segment_headers(this->segment_list_)); // Lay out the file header. Output_file_header* file_header = new Output_file_header(target, symtab, segment_headers); this->special_output_list_.push_back(file_header); if (segment_headers != NULL) this->special_output_list_.push_back(segment_headers); // Find approriate places for orphan output sections if we are using // a linker script. if (this->script_options_->saw_sections_clause()) this->place_orphan_sections_in_script(); Output_segment* load_seg; off_t off; unsigned int shndx; int pass = 0; // Take a snapshot of the section layout as needed. if (target->may_relax()) this->prepare_for_relaxation(); // Run the relaxation loop to lay out sections. do { off = this->relaxation_loop_body(pass, target, symtab, &load_seg, phdr_seg, segment_headers, file_header, &shndx); pass++; } while (target->may_relax() && target->relax(pass, input_objects, symtab, this, task)); // Set the file offsets of all the non-data sections we've seen so // far which don't have to wait for the input sections. We need // this in order to finalize local symbols in non-allocated // sections. off = this->set_section_offsets(off, BEFORE_INPUT_SECTIONS_PASS); // Set the section indexes of all unallocated sections seen so far, // in case any of them are somehow referenced by a symbol. shndx = this->set_section_indexes(shndx); // Create the symbol table sections. this->create_symtab_sections(input_objects, symtab, shndx, &off); if (!parameters->doing_static_link()) this->assign_local_dynsym_offsets(input_objects); // Process any symbol assignments from a linker script. This must // be called after the symbol table has been finalized. this->script_options_->finalize_symbols(symtab, this); // Create the incremental inputs sections. if (this->incremental_inputs_) { this->incremental_inputs_->finalize(); this->create_incremental_info_sections(symtab); } // Create the .shstrtab section. Output_section* shstrtab_section = this->create_shstrtab(); // Set the file offsets of the rest of the non-data sections which // don't have to wait for the input sections. off = this->set_section_offsets(off, BEFORE_INPUT_SECTIONS_PASS); // Now that all sections have been created, set the section indexes // for any sections which haven't been done yet. shndx = this->set_section_indexes(shndx); // Create the section table header. this->create_shdrs(shstrtab_section, &off); // If there are no sections which require postprocessing, we can // handle the section names now, and avoid a resize later. if (!this->any_postprocessing_sections_) { off = this->set_section_offsets(off, POSTPROCESSING_SECTIONS_PASS); off = this->set_section_offsets(off, STRTAB_AFTER_POSTPROCESSING_SECTIONS_PASS); } file_header->set_section_info(this->section_headers_, shstrtab_section); // Now we know exactly where everything goes in the output file // (except for non-allocated sections which require postprocessing). Output_data::layout_complete(); this->output_file_size_ = off; return off; } // Create a note header following the format defined in the ELF ABI. // NAME is the name, NOTE_TYPE is the type, SECTION_NAME is the name // of the section to create, DESCSZ is the size of the descriptor. // ALLOCATE is true if the section should be allocated in memory. // This returns the new note section. It sets *TRAILING_PADDING to // the number of trailing zero bytes required. Output_section* Layout::create_note(const char* name, int note_type, const char* section_name, size_t descsz, bool allocate, size_t* trailing_padding) { // Authorities all agree that the values in a .note field should // be aligned on 4-byte boundaries for 32-bit binaries. However, // they differ on what the alignment is for 64-bit binaries. // The GABI says unambiguously they take 8-byte alignment: // http://sco.com/developers/gabi/latest/ch5.pheader.html#note_section // Other documentation says alignment should always be 4 bytes: // http://www.netbsd.org/docs/kernel/elf-notes.html#note-format // GNU ld and GNU readelf both support the latter (at least as of // version 2.16.91), and glibc always generates the latter for // .note.ABI-tag (as of version 1.6), so that's the one we go with // here. #ifdef GABI_FORMAT_FOR_DOTNOTE_SECTION // This is not defined by default. const int size = parameters->target().get_size(); #else const int size = 32; #endif // The contents of the .note section. size_t namesz = strlen(name) + 1; size_t aligned_namesz = align_address(namesz, size / 8); size_t aligned_descsz = align_address(descsz, size / 8); size_t notehdrsz = 3 * (size / 8) + aligned_namesz; unsigned char* buffer = new unsigned char[notehdrsz]; memset(buffer, 0, notehdrsz); bool is_big_endian = parameters->target().is_big_endian(); if (size == 32) { if (!is_big_endian) { elfcpp::Swap<32, false>::writeval(buffer, namesz); elfcpp::Swap<32, false>::writeval(buffer + 4, descsz); elfcpp::Swap<32, false>::writeval(buffer + 8, note_type); } else { elfcpp::Swap<32, true>::writeval(buffer, namesz); elfcpp::Swap<32, true>::writeval(buffer + 4, descsz); elfcpp::Swap<32, true>::writeval(buffer + 8, note_type); } } else if (size == 64) { if (!is_big_endian) { elfcpp::Swap<64, false>::writeval(buffer, namesz); elfcpp::Swap<64, false>::writeval(buffer + 8, descsz); elfcpp::Swap<64, false>::writeval(buffer + 16, note_type); } else { elfcpp::Swap<64, true>::writeval(buffer, namesz); elfcpp::Swap<64, true>::writeval(buffer + 8, descsz); elfcpp::Swap<64, true>::writeval(buffer + 16, note_type); } } else gold_unreachable(); memcpy(buffer + 3 * (size / 8), name, namesz); elfcpp::Elf_Xword flags = 0; Output_section_order order = ORDER_INVALID; if (allocate) { flags = elfcpp::SHF_ALLOC; order = ORDER_RO_NOTE; } Output_section* os = this->choose_output_section(NULL, section_name, elfcpp::SHT_NOTE, flags, false, order, false); if (os == NULL) return NULL; Output_section_data* posd = new Output_data_const_buffer(buffer, notehdrsz, size / 8, "** note header"); os->add_output_section_data(posd); *trailing_padding = aligned_descsz - descsz; return os; } // For an executable or shared library, create a note to record the // version of gold used to create the binary. void Layout::create_gold_note() { if (parameters->options().relocatable() || parameters->incremental_update()) return; std::string desc = std::string("gold ") + gold::get_version_string(); size_t trailing_padding; Output_section* os = this->create_note("GNU", elfcpp::NT_GNU_GOLD_VERSION, ".note.gnu.gold-version", desc.size(), false, &trailing_padding); if (os == NULL) return; Output_section_data* posd = new Output_data_const(desc, 4); os->add_output_section_data(posd); if (trailing_padding > 0) { posd = new Output_data_zero_fill(trailing_padding, 0); os->add_output_section_data(posd); } } // Record whether the stack should be executable. This can be set // from the command line using the -z execstack or -z noexecstack // options. Otherwise, if any input file has a .note.GNU-stack // section with the SHF_EXECINSTR flag set, the stack should be // executable. Otherwise, if at least one input file a // .note.GNU-stack section, and some input file has no .note.GNU-stack // section, we use the target default for whether the stack should be // executable. Otherwise, we don't generate a stack note. When // generating a object file, we create a .note.GNU-stack section with // the appropriate marking. When generating an executable or shared // library, we create a PT_GNU_STACK segment. void Layout::create_executable_stack_info() { bool is_stack_executable; if (parameters->options().is_execstack_set()) is_stack_executable = parameters->options().is_stack_executable(); else if (!this->input_with_gnu_stack_note_) return; else { if (this->input_requires_executable_stack_) is_stack_executable = true; else if (this->input_without_gnu_stack_note_) is_stack_executable = parameters->target().is_default_stack_executable(); else is_stack_executable = false; } if (parameters->options().relocatable()) { const char* name = this->namepool_.add(".note.GNU-stack", false, NULL); elfcpp::Elf_Xword flags = 0; if (is_stack_executable) flags |= elfcpp::SHF_EXECINSTR; this->make_output_section(name, elfcpp::SHT_PROGBITS, flags, ORDER_INVALID, false); } else { if (this->script_options_->saw_phdrs_clause()) return; int flags = elfcpp::PF_R | elfcpp::PF_W; if (is_stack_executable) flags |= elfcpp::PF_X; this->make_output_segment(elfcpp::PT_GNU_STACK, flags); } } // If --build-id was used, set up the build ID note. void Layout::create_build_id() { if (!parameters->options().user_set_build_id()) return; const char* style = parameters->options().build_id(); if (strcmp(style, "none") == 0) return; // Set DESCSZ to the size of the note descriptor. When possible, // set DESC to the note descriptor contents. size_t descsz; std::string desc; if (strcmp(style, "md5") == 0) descsz = 128 / 8; else if (strcmp(style, "sha1") == 0) descsz = 160 / 8; else if (strcmp(style, "uuid") == 0) { const size_t uuidsz = 128 / 8; char buffer[uuidsz]; memset(buffer, 0, uuidsz); int descriptor = open_descriptor(-1, "/dev/urandom", O_RDONLY); if (descriptor < 0) gold_error(_("--build-id=uuid failed: could not open /dev/urandom: %s"), strerror(errno)); else { ssize_t got = ::read(descriptor, buffer, uuidsz); release_descriptor(descriptor, true); if (got < 0) gold_error(_("/dev/urandom: read failed: %s"), strerror(errno)); else if (static_cast<size_t>(got) != uuidsz) gold_error(_("/dev/urandom: expected %zu bytes, got %zd bytes"), uuidsz, got); } desc.assign(buffer, uuidsz); descsz = uuidsz; } else if (strncmp(style, "0x", 2) == 0) { hex_init(); const char* p = style + 2; while (*p != '\0') { if (hex_p(p[0]) && hex_p(p[1])) { char c = (hex_value(p[0]) << 4) | hex_value(p[1]); desc += c; p += 2; } else if (*p == '-' || *p == ':') ++p; else gold_fatal(_("--build-id argument '%s' not a valid hex number"), style); } descsz = desc.size(); } else gold_fatal(_("unrecognized --build-id argument '%s'"), style); // Create the note. size_t trailing_padding; Output_section* os = this->create_note("GNU", elfcpp::NT_GNU_BUILD_ID, ".note.gnu.build-id", descsz, true, &trailing_padding); if (os == NULL) return; if (!desc.empty()) { // We know the value already, so we fill it in now. gold_assert(desc.size() == descsz); Output_section_data* posd = new Output_data_const(desc, 4); os->add_output_section_data(posd); if (trailing_padding != 0) { posd = new Output_data_zero_fill(trailing_padding, 0); os->add_output_section_data(posd); } } else { // We need to compute a checksum after we have completed the // link. gold_assert(trailing_padding == 0); this->build_id_note_ = new Output_data_zero_fill(descsz, 4); os->add_output_section_data(this->build_id_note_); } } // If we have both .stabXX and .stabXXstr sections, then the sh_link // field of the former should point to the latter. I'm not sure who // started this, but the GNU linker does it, and some tools depend // upon it. void Layout::link_stabs_sections() { if (!this->have_stabstr_section_) return; for (Section_list::iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) { if ((*p)->type() != elfcpp::SHT_STRTAB) continue; const char* name = (*p)->name(); if (strncmp(name, ".stab", 5) != 0) continue; size_t len = strlen(name); if (strcmp(name + len - 3, "str") != 0) continue; std::string stab_name(name, len - 3); Output_section* stab_sec; stab_sec = this->find_output_section(stab_name.c_str()); if (stab_sec != NULL) stab_sec->set_link_section(*p); } } // Create .gnu_incremental_inputs and related sections needed // for the next run of incremental linking to check what has changed. void Layout::create_incremental_info_sections(Symbol_table* symtab) { Incremental_inputs* incr = this->incremental_inputs_; gold_assert(incr != NULL); // Create the .gnu_incremental_inputs, _symtab, and _relocs input sections. incr->create_data_sections(symtab); // Add the .gnu_incremental_inputs section. const char* incremental_inputs_name = this->namepool_.add(".gnu_incremental_inputs", false, NULL); Output_section* incremental_inputs_os = this->make_output_section(incremental_inputs_name, elfcpp::SHT_GNU_INCREMENTAL_INPUTS, 0, ORDER_INVALID, false); incremental_inputs_os->add_output_section_data(incr->inputs_section()); // Add the .gnu_incremental_symtab section. const char* incremental_symtab_name = this->namepool_.add(".gnu_incremental_symtab", false, NULL); Output_section* incremental_symtab_os = this->make_output_section(incremental_symtab_name, elfcpp::SHT_GNU_INCREMENTAL_SYMTAB, 0, ORDER_INVALID, false); incremental_symtab_os->add_output_section_data(incr->symtab_section()); incremental_symtab_os->set_entsize(4); // Add the .gnu_incremental_relocs section. const char* incremental_relocs_name = this->namepool_.add(".gnu_incremental_relocs", false, NULL); Output_section* incremental_relocs_os = this->make_output_section(incremental_relocs_name, elfcpp::SHT_GNU_INCREMENTAL_RELOCS, 0, ORDER_INVALID, false); incremental_relocs_os->add_output_section_data(incr->relocs_section()); incremental_relocs_os->set_entsize(incr->relocs_entsize()); // Add the .gnu_incremental_got_plt section. const char* incremental_got_plt_name = this->namepool_.add(".gnu_incremental_got_plt", false, NULL); Output_section* incremental_got_plt_os = this->make_output_section(incremental_got_plt_name, elfcpp::SHT_GNU_INCREMENTAL_GOT_PLT, 0, ORDER_INVALID, false); incremental_got_plt_os->add_output_section_data(incr->got_plt_section()); // Add the .gnu_incremental_strtab section. const char* incremental_strtab_name = this->namepool_.add(".gnu_incremental_strtab", false, NULL); Output_section* incremental_strtab_os = this->make_output_section(incremental_strtab_name, elfcpp::SHT_STRTAB, 0, ORDER_INVALID, false); Output_data_strtab* strtab_data = new Output_data_strtab(incr->get_stringpool()); incremental_strtab_os->add_output_section_data(strtab_data); incremental_inputs_os->set_after_input_sections(); incremental_symtab_os->set_after_input_sections(); incremental_relocs_os->set_after_input_sections(); incremental_got_plt_os->set_after_input_sections(); incremental_inputs_os->set_link_section(incremental_strtab_os); incremental_symtab_os->set_link_section(incremental_inputs_os); incremental_relocs_os->set_link_section(incremental_inputs_os); incremental_got_plt_os->set_link_section(incremental_inputs_os); } // Return whether SEG1 should be before SEG2 in the output file. This // is based entirely on the segment type and flags. When this is // called the segment addresses has normally not yet been set. bool Layout::segment_precedes(const Output_segment* seg1, const Output_segment* seg2) { elfcpp::Elf_Word type1 = seg1->type(); elfcpp::Elf_Word type2 = seg2->type(); // The single PT_PHDR segment is required to precede any loadable // segment. We simply make it always first. if (type1 == elfcpp::PT_PHDR) { gold_assert(type2 != elfcpp::PT_PHDR); return true; } if (type2 == elfcpp::PT_PHDR) return false; // The single PT_INTERP segment is required to precede any loadable // segment. We simply make it always second. if (type1 == elfcpp::PT_INTERP) { gold_assert(type2 != elfcpp::PT_INTERP); return true; } if (type2 == elfcpp::PT_INTERP) return false; // We then put PT_LOAD segments before any other segments. if (type1 == elfcpp::PT_LOAD && type2 != elfcpp::PT_LOAD) return true; if (type2 == elfcpp::PT_LOAD && type1 != elfcpp::PT_LOAD) return false; // We put the PT_TLS segment last except for the PT_GNU_RELRO // segment, because that is where the dynamic linker expects to find // it (this is just for efficiency; other positions would also work // correctly). if (type1 == elfcpp::PT_TLS && type2 != elfcpp::PT_TLS && type2 != elfcpp::PT_GNU_RELRO) return false; if (type2 == elfcpp::PT_TLS && type1 != elfcpp::PT_TLS && type1 != elfcpp::PT_GNU_RELRO) return true; // We put the PT_GNU_RELRO segment last, because that is where the // dynamic linker expects to find it (as with PT_TLS, this is just // for efficiency). if (type1 == elfcpp::PT_GNU_RELRO && type2 != elfcpp::PT_GNU_RELRO) return false; if (type2 == elfcpp::PT_GNU_RELRO && type1 != elfcpp::PT_GNU_RELRO) return true; const elfcpp::Elf_Word flags1 = seg1->flags(); const elfcpp::Elf_Word flags2 = seg2->flags(); // The order of non-PT_LOAD segments is unimportant. We simply sort // by the numeric segment type and flags values. There should not // be more than one segment with the same type and flags. if (type1 != elfcpp::PT_LOAD) { if (type1 != type2) return type1 < type2; gold_assert(flags1 != flags2); return flags1 < flags2; } // If the addresses are set already, sort by load address. if (seg1->are_addresses_set()) { if (!seg2->are_addresses_set()) return true; unsigned int section_count1 = seg1->output_section_count(); unsigned int section_count2 = seg2->output_section_count(); if (section_count1 == 0 && section_count2 > 0) return true; if (section_count1 > 0 && section_count2 == 0) return false; uint64_t paddr1 = (seg1->are_addresses_set() ? seg1->paddr() : seg1->first_section_load_address()); uint64_t paddr2 = (seg2->are_addresses_set() ? seg2->paddr() : seg2->first_section_load_address()); if (paddr1 != paddr2) return paddr1 < paddr2; } else if (seg2->are_addresses_set()) return false; // A segment which holds large data comes after a segment which does // not hold large data. if (seg1->is_large_data_segment()) { if (!seg2->is_large_data_segment()) return false; } else if (seg2->is_large_data_segment()) return true; // Otherwise, we sort PT_LOAD segments based on the flags. Readonly // segments come before writable segments. Then writable segments // with data come before writable segments without data. Then // executable segments come before non-executable segments. Then // the unlikely case of a non-readable segment comes before the // normal case of a readable segment. If there are multiple // segments with the same type and flags, we require that the // address be set, and we sort by virtual address and then physical // address. if ((flags1 & elfcpp::PF_W) != (flags2 & elfcpp::PF_W)) return (flags1 & elfcpp::PF_W) == 0; if ((flags1 & elfcpp::PF_W) != 0 && seg1->has_any_data_sections() != seg2->has_any_data_sections()) return seg1->has_any_data_sections(); if ((flags1 & elfcpp::PF_X) != (flags2 & elfcpp::PF_X)) return (flags1 & elfcpp::PF_X) != 0; if ((flags1 & elfcpp::PF_R) != (flags2 & elfcpp::PF_R)) return (flags1 & elfcpp::PF_R) == 0; // We shouldn't get here--we shouldn't create segments which we // can't distinguish. gold_unreachable(); } // Increase OFF so that it is congruent to ADDR modulo ABI_PAGESIZE. static off_t align_file_offset(off_t off, uint64_t addr, uint64_t abi_pagesize) { uint64_t unsigned_off = off; uint64_t aligned_off = ((unsigned_off & ~(abi_pagesize - 1)) | (addr & (abi_pagesize - 1))); if (aligned_off < unsigned_off) aligned_off += abi_pagesize; return aligned_off; } // Set the file offsets of all the segments, and all the sections they // contain. They have all been created. LOAD_SEG must be be laid out // first. Return the offset of the data to follow. off_t Layout::set_segment_offsets(const Target* target, Output_segment* load_seg, unsigned int* pshndx) { // Sort them into the final order. std::sort(this->segment_list_.begin(), this->segment_list_.end(), Layout::Compare_segments()); // Find the PT_LOAD segments, and set their addresses and offsets // and their section's addresses and offsets. uint64_t addr; if (parameters->options().user_set_Ttext()) addr = parameters->options().Ttext(); else if (parameters->options().output_is_position_independent()) addr = 0; else addr = target->default_text_segment_address(); off_t off = 0; // If LOAD_SEG is NULL, then the file header and segment headers // will not be loadable. But they still need to be at offset 0 in // the file. Set their offsets now. if (load_seg == NULL) { for (Data_list::iterator p = this->special_output_list_.begin(); p != this->special_output_list_.end(); ++p) { off = align_address(off, (*p)->addralign()); (*p)->set_address_and_file_offset(0, off); off += (*p)->data_size(); } } unsigned int increase_relro = this->increase_relro_; if (this->script_options_->saw_sections_clause()) increase_relro = 0; const bool check_sections = parameters->options().check_sections(); Output_segment* last_load_segment = NULL; for (Segment_list::iterator p = this->segment_list_.begin(); p != this->segment_list_.end(); ++p) { if ((*p)->type() == elfcpp::PT_LOAD) { if (load_seg != NULL && load_seg != *p) gold_unreachable(); load_seg = NULL; bool are_addresses_set = (*p)->are_addresses_set(); if (are_addresses_set) { // When it comes to setting file offsets, we care about // the physical address. addr = (*p)->paddr(); } else if (parameters->options().user_set_Tdata() && ((*p)->flags() & elfcpp::PF_W) != 0 && (!parameters->options().user_set_Tbss() || (*p)->has_any_data_sections())) { addr = parameters->options().Tdata(); are_addresses_set = true; } else if (parameters->options().user_set_Tbss() && ((*p)->flags() & elfcpp::PF_W) != 0 && !(*p)->has_any_data_sections()) { addr = parameters->options().Tbss(); are_addresses_set = true; } uint64_t orig_addr = addr; uint64_t orig_off = off; uint64_t aligned_addr = 0; uint64_t abi_pagesize = target->abi_pagesize(); uint64_t common_pagesize = target->common_pagesize(); if (!parameters->options().nmagic() && !parameters->options().omagic()) (*p)->set_minimum_p_align(common_pagesize); if (!are_addresses_set) { // Skip the address forward one page, maintaining the same // position within the page. This lets us store both segments // overlapping on a single page in the file, but the loader will // put them on different pages in memory. We will revisit this // decision once we know the size of the segment. addr = align_address(addr, (*p)->maximum_alignment()); aligned_addr = addr; if ((addr & (abi_pagesize - 1)) != 0) addr = addr + abi_pagesize; off = orig_off + ((addr - orig_addr) & (abi_pagesize - 1)); } if (!parameters->options().nmagic() && !parameters->options().omagic()) off = align_file_offset(off, addr, abi_pagesize); else if (load_seg == NULL) { // This is -N or -n with a section script which prevents // us from using a load segment. We need to ensure that // the file offset is aligned to the alignment of the // segment. This is because the linker script // implicitly assumed a zero offset. If we don't align // here, then the alignment of the sections in the // linker script may not match the alignment of the // sections in the set_section_addresses call below, // causing an error about dot moving backward. off = align_address(off, (*p)->maximum_alignment()); } unsigned int shndx_hold = *pshndx; bool has_relro = false; uint64_t new_addr = (*p)->set_section_addresses(this, false, addr, &increase_relro, &has_relro, &off, pshndx); // Now that we know the size of this segment, we may be able // to save a page in memory, at the cost of wasting some // file space, by instead aligning to the start of a new // page. Here we use the real machine page size rather than // the ABI mandated page size. If the segment has been // aligned so that the relro data ends at a page boundary, // we do not try to realign it. if (!are_addresses_set && !has_relro && aligned_addr != addr && !parameters->incremental_update()) { uint64_t first_off = (common_pagesize - (aligned_addr & (common_pagesize - 1))); uint64_t last_off = new_addr & (common_pagesize - 1); if (first_off > 0 && last_off > 0 && ((aligned_addr & ~ (common_pagesize - 1)) != (new_addr & ~ (common_pagesize - 1))) && first_off + last_off <= common_pagesize) { *pshndx = shndx_hold; addr = align_address(aligned_addr, common_pagesize); addr = align_address(addr, (*p)->maximum_alignment()); off = orig_off + ((addr - orig_addr) & (abi_pagesize - 1)); off = align_file_offset(off, addr, abi_pagesize); increase_relro = this->increase_relro_; if (this->script_options_->saw_sections_clause()) increase_relro = 0; has_relro = false; new_addr = (*p)->set_section_addresses(this, true, addr, &increase_relro, &has_relro, &off, pshndx); } } addr = new_addr; // Implement --check-sections. We know that the segments // are sorted by LMA. if (check_sections && last_load_segment != NULL) { gold_assert(last_load_segment->paddr() <= (*p)->paddr()); if (last_load_segment->paddr() + last_load_segment->memsz() > (*p)->paddr()) { unsigned long long lb1 = last_load_segment->paddr(); unsigned long long le1 = lb1 + last_load_segment->memsz(); unsigned long long lb2 = (*p)->paddr(); unsigned long long le2 = lb2 + (*p)->memsz(); gold_error(_("load segment overlap [0x%llx -> 0x%llx] and " "[0x%llx -> 0x%llx]"), lb1, le1, lb2, le2); } } last_load_segment = *p; } } // Handle the non-PT_LOAD segments, setting their offsets from their // section's offsets. for (Segment_list::iterator p = this->segment_list_.begin(); p != this->segment_list_.end(); ++p) { if ((*p)->type() != elfcpp::PT_LOAD) (*p)->set_offset((*p)->type() == elfcpp::PT_GNU_RELRO ? increase_relro : 0); } // Set the TLS offsets for each section in the PT_TLS segment. if (this->tls_segment_ != NULL) this->tls_segment_->set_tls_offsets(); return off; } // Set the offsets of all the allocated sections when doing a // relocatable link. This does the same jobs as set_segment_offsets, // only for a relocatable link. off_t Layout::set_relocatable_section_offsets(Output_data* file_header, unsigned int* pshndx) { off_t off = 0; file_header->set_address_and_file_offset(0, 0); off += file_header->data_size(); for (Section_list::iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) { // We skip unallocated sections here, except that group sections // have to come first. if (((*p)->flags() & elfcpp::SHF_ALLOC) == 0 && (*p)->type() != elfcpp::SHT_GROUP) continue; off = align_address(off, (*p)->addralign()); // The linker script might have set the address. if (!(*p)->is_address_valid()) (*p)->set_address(0); (*p)->set_file_offset(off); (*p)->finalize_data_size(); off += (*p)->data_size(); (*p)->set_out_shndx(*pshndx); ++*pshndx; } return off; } // Set the file offset of all the sections not associated with a // segment. off_t Layout::set_section_offsets(off_t off, Layout::Section_offset_pass pass) { off_t startoff = off; off_t maxoff = off; for (Section_list::iterator p = this->unattached_section_list_.begin(); p != this->unattached_section_list_.end(); ++p) { // The symtab section is handled in create_symtab_sections. if (*p == this->symtab_section_) continue; // If we've already set the data size, don't set it again. if ((*p)->is_offset_valid() && (*p)->is_data_size_valid()) continue; if (pass == BEFORE_INPUT_SECTIONS_PASS && (*p)->requires_postprocessing()) { (*p)->create_postprocessing_buffer(); this->any_postprocessing_sections_ = true; } if (pass == BEFORE_INPUT_SECTIONS_PASS && (*p)->after_input_sections()) continue; else if (pass == POSTPROCESSING_SECTIONS_PASS && (!(*p)->after_input_sections() || (*p)->type() == elfcpp::SHT_STRTAB)) continue; else if (pass == STRTAB_AFTER_POSTPROCESSING_SECTIONS_PASS && (!(*p)->after_input_sections() || (*p)->type() != elfcpp::SHT_STRTAB)) continue; if (!parameters->incremental_update()) { off = align_address(off, (*p)->addralign()); (*p)->set_file_offset(off); (*p)->finalize_data_size(); } else { // Incremental update: allocate file space from free list. (*p)->pre_finalize_data_size(); off_t current_size = (*p)->current_data_size(); off = this->allocate(current_size, (*p)->addralign(), startoff); if (off == -1) { if (is_debugging_enabled(DEBUG_INCREMENTAL)) this->free_list_.dump(); gold_assert((*p)->output_section() != NULL); gold_fatal(_("out of patch space for section %s; " "relink with --incremental-full"), (*p)->output_section()->name()); } (*p)->set_file_offset(off); (*p)->finalize_data_size(); if ((*p)->data_size() > current_size) { gold_assert((*p)->output_section() != NULL); gold_fatal(_("%s: section changed size; " "relink with --incremental-full"), (*p)->output_section()->name()); } gold_debug(DEBUG_INCREMENTAL, "set_section_offsets: %08lx %08lx %s", static_cast<long>(off), static_cast<long>((*p)->data_size()), ((*p)->output_section() != NULL ? (*p)->output_section()->name() : "(special)")); } off += (*p)->data_size(); if (off > maxoff) maxoff = off; // At this point the name must be set. if (pass != STRTAB_AFTER_POSTPROCESSING_SECTIONS_PASS) this->namepool_.add((*p)->name(), false, NULL); } return maxoff; } // Set the section indexes of all the sections not associated with a // segment. unsigned int Layout::set_section_indexes(unsigned int shndx) { for (Section_list::iterator p = this->unattached_section_list_.begin(); p != this->unattached_section_list_.end(); ++p) { if (!(*p)->has_out_shndx()) { (*p)->set_out_shndx(shndx); ++shndx; } } return shndx; } // Set the section addresses according to the linker script. This is // only called when we see a SECTIONS clause. This returns the // program segment which should hold the file header and segment // headers, if any. It will return NULL if they should not be in a // segment. Output_segment* Layout::set_section_addresses_from_script(Symbol_table* symtab) { Script_sections* ss = this->script_options_->script_sections(); gold_assert(ss->saw_sections_clause()); return this->script_options_->set_section_addresses(symtab, this); } // Place the orphan sections in the linker script. void Layout::place_orphan_sections_in_script() { Script_sections* ss = this->script_options_->script_sections(); gold_assert(ss->saw_sections_clause()); // Place each orphaned output section in the script. for (Section_list::iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) { if (!(*p)->found_in_sections_clause()) ss->place_orphan(*p); } } // Count the local symbols in the regular symbol table and the dynamic // symbol table, and build the respective string pools. void Layout::count_local_symbols(const Task* task, const Input_objects* input_objects) { // First, figure out an upper bound on the number of symbols we'll // be inserting into each pool. This helps us create the pools with // the right size, to avoid unnecessary hashtable resizing. unsigned int symbol_count = 0; for (Input_objects::Relobj_iterator p = input_objects->relobj_begin(); p != input_objects->relobj_end(); ++p) symbol_count += (*p)->local_symbol_count(); // Go from "upper bound" to "estimate." We overcount for two // reasons: we double-count symbols that occur in more than one // object file, and we count symbols that are dropped from the // output. Add it all together and assume we overcount by 100%. symbol_count /= 2; // We assume all symbols will go into both the sympool and dynpool. this->sympool_.reserve(symbol_count); this->dynpool_.reserve(symbol_count); for (Input_objects::Relobj_iterator p = input_objects->relobj_begin(); p != input_objects->relobj_end(); ++p) { Task_lock_obj<Object> tlo(task, *p); (*p)->count_local_symbols(&this->sympool_, &this->dynpool_); } } // Create the symbol table sections. Here we also set the final // values of the symbols. At this point all the loadable sections are // fully laid out. SHNUM is the number of sections so far. void Layout::create_symtab_sections(const Input_objects* input_objects, Symbol_table* symtab, unsigned int shnum, off_t* poff) { int symsize; unsigned int align; if (parameters->target().get_size() == 32) { symsize = elfcpp::Elf_sizes<32>::sym_size; align = 4; } else if (parameters->target().get_size() == 64) { symsize = elfcpp::Elf_sizes<64>::sym_size; align = 8; } else gold_unreachable(); // Compute file offsets relative to the start of the symtab section. off_t off = 0; // Save space for the dummy symbol at the start of the section. We // never bother to write this out--it will just be left as zero. off += symsize; unsigned int local_symbol_index = 1; // Add STT_SECTION symbols for each Output section which needs one. for (Section_list::iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) { if (!(*p)->needs_symtab_index()) (*p)->set_symtab_index(-1U); else { (*p)->set_symtab_index(local_symbol_index); ++local_symbol_index; off += symsize; } } for (Input_objects::Relobj_iterator p = input_objects->relobj_begin(); p != input_objects->relobj_end(); ++p) { unsigned int index = (*p)->finalize_local_symbols(local_symbol_index, off, symtab); off += (index - local_symbol_index) * symsize; local_symbol_index = index; } unsigned int local_symcount = local_symbol_index; gold_assert(static_cast<off_t>(local_symcount * symsize) == off); off_t dynoff; size_t dyn_global_index; size_t dyncount; if (this->dynsym_section_ == NULL) { dynoff = 0; dyn_global_index = 0; dyncount = 0; } else { dyn_global_index = this->dynsym_section_->info(); off_t locsize = dyn_global_index * this->dynsym_section_->entsize(); dynoff = this->dynsym_section_->offset() + locsize; dyncount = (this->dynsym_section_->data_size() - locsize) / symsize; gold_assert(static_cast<off_t>(dyncount * symsize) == this->dynsym_section_->data_size() - locsize); } off_t global_off = off; off = symtab->finalize(off, dynoff, dyn_global_index, dyncount, &this->sympool_, &local_symcount); if (!parameters->options().strip_all()) { this->sympool_.set_string_offsets(); const char* symtab_name = this->namepool_.add(".symtab", false, NULL); Output_section* osymtab = this->make_output_section(symtab_name, elfcpp::SHT_SYMTAB, 0, ORDER_INVALID, false); this->symtab_section_ = osymtab; Output_section_data* pos = new Output_data_fixed_space(off, align, "** symtab"); osymtab->add_output_section_data(pos); // We generate a .symtab_shndx section if we have more than // SHN_LORESERVE sections. Technically it is possible that we // don't need one, because it is possible that there are no // symbols in any of sections with indexes larger than // SHN_LORESERVE. That is probably unusual, though, and it is // easier to always create one than to compute section indexes // twice (once here, once when writing out the symbols). if (shnum >= elfcpp::SHN_LORESERVE) { const char* symtab_xindex_name = this->namepool_.add(".symtab_shndx", false, NULL); Output_section* osymtab_xindex = this->make_output_section(symtab_xindex_name, elfcpp::SHT_SYMTAB_SHNDX, 0, ORDER_INVALID, false); size_t symcount = off / symsize; this->symtab_xindex_ = new Output_symtab_xindex(symcount); osymtab_xindex->add_output_section_data(this->symtab_xindex_); osymtab_xindex->set_link_section(osymtab); osymtab_xindex->set_addralign(4); osymtab_xindex->set_entsize(4); osymtab_xindex->set_after_input_sections(); // This tells the driver code to wait until the symbol table // has written out before writing out the postprocessing // sections, including the .symtab_shndx section. this->any_postprocessing_sections_ = true; } const char* strtab_name = this->namepool_.add(".strtab", false, NULL); Output_section* ostrtab = this->make_output_section(strtab_name, elfcpp::SHT_STRTAB, 0, ORDER_INVALID, false); Output_section_data* pstr = new Output_data_strtab(&this->sympool_); ostrtab->add_output_section_data(pstr); off_t symtab_off; if (!parameters->incremental_update()) symtab_off = align_address(*poff, align); else { symtab_off = this->allocate(off, align, *poff); if (off == -1) gold_fatal(_("out of patch space for symbol table; " "relink with --incremental-full")); gold_debug(DEBUG_INCREMENTAL, "create_symtab_sections: %08lx %08lx .symtab", static_cast<long>(symtab_off), static_cast<long>(off)); } symtab->set_file_offset(symtab_off + global_off); osymtab->set_file_offset(symtab_off); osymtab->finalize_data_size(); osymtab->set_link_section(ostrtab); osymtab->set_info(local_symcount); osymtab->set_entsize(symsize); if (symtab_off + off > *poff) *poff = symtab_off + off; } } // Create the .shstrtab section, which holds the names of the // sections. At the time this is called, we have created all the // output sections except .shstrtab itself. Output_section* Layout::create_shstrtab() { // FIXME: We don't need to create a .shstrtab section if we are // stripping everything. const char* name = this->namepool_.add(".shstrtab", false, NULL); Output_section* os = this->make_output_section(name, elfcpp::SHT_STRTAB, 0, ORDER_INVALID, false); if (strcmp(parameters->options().compress_debug_sections(), "none") != 0) { // We can't write out this section until we've set all the // section names, and we don't set the names of compressed // output sections until relocations are complete. FIXME: With // the current names we use, this is unnecessary. os->set_after_input_sections(); } Output_section_data* posd = new Output_data_strtab(&this->namepool_); os->add_output_section_data(posd); return os; } // Create the section headers. SIZE is 32 or 64. OFF is the file // offset. void Layout::create_shdrs(const Output_section* shstrtab_section, off_t* poff) { Output_section_headers* oshdrs; oshdrs = new Output_section_headers(this, &this->segment_list_, &this->section_list_, &this->unattached_section_list_, &this->namepool_, shstrtab_section); off_t off; if (!parameters->incremental_update()) off = align_address(*poff, oshdrs->addralign()); else { oshdrs->pre_finalize_data_size(); off = this->allocate(oshdrs->data_size(), oshdrs->addralign(), *poff); if (off == -1) gold_fatal(_("out of patch space for section header table; " "relink with --incremental-full")); gold_debug(DEBUG_INCREMENTAL, "create_shdrs: %08lx %08lx (section header table)", static_cast<long>(off), static_cast<long>(off + oshdrs->data_size())); } oshdrs->set_address_and_file_offset(0, off); off += oshdrs->data_size(); if (off > *poff) *poff = off; this->section_headers_ = oshdrs; } // Count the allocated sections. size_t Layout::allocated_output_section_count() const { size_t section_count = 0; for (Segment_list::const_iterator p = this->segment_list_.begin(); p != this->segment_list_.end(); ++p) section_count += (*p)->output_section_count(); return section_count; } // Create the dynamic symbol table. void Layout::create_dynamic_symtab(const Input_objects* input_objects, Symbol_table* symtab, Output_section** pdynstr, unsigned int* plocal_dynamic_count, std::vector<Symbol*>* pdynamic_symbols, Versions* pversions) { // Count all the symbols in the dynamic symbol table, and set the // dynamic symbol indexes. // Skip symbol 0, which is always all zeroes. unsigned int index = 1; // Add STT_SECTION symbols for each Output section which needs one. for (Section_list::iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) { if (!(*p)->needs_dynsym_index()) (*p)->set_dynsym_index(-1U); else { (*p)->set_dynsym_index(index); ++index; } } // Count the local symbols that need to go in the dynamic symbol table, // and set the dynamic symbol indexes. for (Input_objects::Relobj_iterator p = input_objects->relobj_begin(); p != input_objects->relobj_end(); ++p) { unsigned int new_index = (*p)->set_local_dynsym_indexes(index); index = new_index; } unsigned int local_symcount = index; *plocal_dynamic_count = local_symcount; index = symtab->set_dynsym_indexes(index, pdynamic_symbols, &this->dynpool_, pversions); int symsize; unsigned int align; const int size = parameters->target().get_size(); if (size == 32) { symsize = elfcpp::Elf_sizes<32>::sym_size; align = 4; } else if (size == 64) { symsize = elfcpp::Elf_sizes<64>::sym_size; align = 8; } else gold_unreachable(); // Create the dynamic symbol table section. Output_section* dynsym = this->choose_output_section(NULL, ".dynsym", elfcpp::SHT_DYNSYM, elfcpp::SHF_ALLOC, false, ORDER_DYNAMIC_LINKER, false); Output_section_data* odata = new Output_data_fixed_space(index * symsize, align, "** dynsym"); dynsym->add_output_section_data(odata); dynsym->set_info(local_symcount); dynsym->set_entsize(symsize); dynsym->set_addralign(align); this->dynsym_section_ = dynsym; Output_data_dynamic* const odyn = this->dynamic_data_; odyn->add_section_address(elfcpp::DT_SYMTAB, dynsym); odyn->add_constant(elfcpp::DT_SYMENT, symsize); // If there are more than SHN_LORESERVE allocated sections, we // create a .dynsym_shndx section. It is possible that we don't // need one, because it is possible that there are no dynamic // symbols in any of the sections with indexes larger than // SHN_LORESERVE. This is probably unusual, though, and at this // time we don't know the actual section indexes so it is // inconvenient to check. if (this->allocated_output_section_count() >= elfcpp::SHN_LORESERVE) { Output_section* dynsym_xindex = this->choose_output_section(NULL, ".dynsym_shndx", elfcpp::SHT_SYMTAB_SHNDX, elfcpp::SHF_ALLOC, false, ORDER_DYNAMIC_LINKER, false); this->dynsym_xindex_ = new Output_symtab_xindex(index); dynsym_xindex->add_output_section_data(this->dynsym_xindex_); dynsym_xindex->set_link_section(dynsym); dynsym_xindex->set_addralign(4); dynsym_xindex->set_entsize(4); dynsym_xindex->set_after_input_sections(); // This tells the driver code to wait until the symbol table has // written out before writing out the postprocessing sections, // including the .dynsym_shndx section. this->any_postprocessing_sections_ = true; } // Create the dynamic string table section. Output_section* dynstr = this->choose_output_section(NULL, ".dynstr", elfcpp::SHT_STRTAB, elfcpp::SHF_ALLOC, false, ORDER_DYNAMIC_LINKER, false); Output_section_data* strdata = new Output_data_strtab(&this->dynpool_); dynstr->add_output_section_data(strdata); dynsym->set_link_section(dynstr); this->dynamic_section_->set_link_section(dynstr); odyn->add_section_address(elfcpp::DT_STRTAB, dynstr); odyn->add_section_size(elfcpp::DT_STRSZ, dynstr); *pdynstr = dynstr; // Create the hash tables. if (strcmp(parameters->options().hash_style(), "sysv") == 0 || strcmp(parameters->options().hash_style(), "both") == 0) { unsigned char* phash; unsigned int hashlen; Dynobj::create_elf_hash_table(*pdynamic_symbols, local_symcount, &phash, &hashlen); Output_section* hashsec = this->choose_output_section(NULL, ".hash", elfcpp::SHT_HASH, elfcpp::SHF_ALLOC, false, ORDER_DYNAMIC_LINKER, false); Output_section_data* hashdata = new Output_data_const_buffer(phash, hashlen, align, "** hash"); hashsec->add_output_section_data(hashdata); hashsec->set_link_section(dynsym); hashsec->set_entsize(4); odyn->add_section_address(elfcpp::DT_HASH, hashsec); } if (strcmp(parameters->options().hash_style(), "gnu") == 0 || strcmp(parameters->options().hash_style(), "both") == 0) { unsigned char* phash; unsigned int hashlen; Dynobj::create_gnu_hash_table(*pdynamic_symbols, local_symcount, &phash, &hashlen); Output_section* hashsec = this->choose_output_section(NULL, ".gnu.hash", elfcpp::SHT_GNU_HASH, elfcpp::SHF_ALLOC, false, ORDER_DYNAMIC_LINKER, false); Output_section_data* hashdata = new Output_data_const_buffer(phash, hashlen, align, "** hash"); hashsec->add_output_section_data(hashdata); hashsec->set_link_section(dynsym); // For a 64-bit target, the entries in .gnu.hash do not have a // uniform size, so we only set the entry size for a 32-bit // target. if (parameters->target().get_size() == 32) hashsec->set_entsize(4); odyn->add_section_address(elfcpp::DT_GNU_HASH, hashsec); } } // Assign offsets to each local portion of the dynamic symbol table. void Layout::assign_local_dynsym_offsets(const Input_objects* input_objects) { Output_section* dynsym = this->dynsym_section_; gold_assert(dynsym != NULL); off_t off = dynsym->offset(); // Skip the dummy symbol at the start of the section. off += dynsym->entsize(); for (Input_objects::Relobj_iterator p = input_objects->relobj_begin(); p != input_objects->relobj_end(); ++p) { unsigned int count = (*p)->set_local_dynsym_offset(off); off += count * dynsym->entsize(); } } // Create the version sections. void Layout::create_version_sections(const Versions* versions, const Symbol_table* symtab, unsigned int local_symcount, const std::vector<Symbol*>& dynamic_symbols, const Output_section* dynstr) { if (!versions->any_defs() && !versions->any_needs()) return; switch (parameters->size_and_endianness()) { #ifdef HAVE_TARGET_32_LITTLE case Parameters::TARGET_32_LITTLE: this->sized_create_version_sections<32, false>(versions, symtab, local_symcount, dynamic_symbols, dynstr); break; #endif #ifdef HAVE_TARGET_32_BIG case Parameters::TARGET_32_BIG: this->sized_create_version_sections<32, true>(versions, symtab, local_symcount, dynamic_symbols, dynstr); break; #endif #ifdef HAVE_TARGET_64_LITTLE case Parameters::TARGET_64_LITTLE: this->sized_create_version_sections<64, false>(versions, symtab, local_symcount, dynamic_symbols, dynstr); break; #endif #ifdef HAVE_TARGET_64_BIG case Parameters::TARGET_64_BIG: this->sized_create_version_sections<64, true>(versions, symtab, local_symcount, dynamic_symbols, dynstr); break; #endif default: gold_unreachable(); } } // Create the version sections, sized version. template<int size, bool big_endian> void Layout::sized_create_version_sections( const Versions* versions, const Symbol_table* symtab, unsigned int local_symcount, const std::vector<Symbol*>& dynamic_symbols, const Output_section* dynstr) { Output_section* vsec = this->choose_output_section(NULL, ".gnu.version", elfcpp::SHT_GNU_versym, elfcpp::SHF_ALLOC, false, ORDER_DYNAMIC_LINKER, false); unsigned char* vbuf; unsigned int vsize; versions->symbol_section_contents<size, big_endian>(symtab, &this->dynpool_, local_symcount, dynamic_symbols, &vbuf, &vsize); Output_section_data* vdata = new Output_data_const_buffer(vbuf, vsize, 2, "** versions"); vsec->add_output_section_data(vdata); vsec->set_entsize(2); vsec->set_link_section(this->dynsym_section_); Output_data_dynamic* const odyn = this->dynamic_data_; odyn->add_section_address(elfcpp::DT_VERSYM, vsec); if (versions->any_defs()) { Output_section* vdsec; vdsec= this->choose_output_section(NULL, ".gnu.version_d", elfcpp::SHT_GNU_verdef, elfcpp::SHF_ALLOC, false, ORDER_DYNAMIC_LINKER, false); unsigned char* vdbuf; unsigned int vdsize; unsigned int vdentries; versions->def_section_contents<size, big_endian>(&this->dynpool_, &vdbuf, &vdsize, &vdentries); Output_section_data* vddata = new Output_data_const_buffer(vdbuf, vdsize, 4, "** version defs"); vdsec->add_output_section_data(vddata); vdsec->set_link_section(dynstr); vdsec->set_info(vdentries); odyn->add_section_address(elfcpp::DT_VERDEF, vdsec); odyn->add_constant(elfcpp::DT_VERDEFNUM, vdentries); } if (versions->any_needs()) { Output_section* vnsec; vnsec = this->choose_output_section(NULL, ".gnu.version_r", elfcpp::SHT_GNU_verneed, elfcpp::SHF_ALLOC, false, ORDER_DYNAMIC_LINKER, false); unsigned char* vnbuf; unsigned int vnsize; unsigned int vnentries; versions->need_section_contents<size, big_endian>(&this->dynpool_, &vnbuf, &vnsize, &vnentries); Output_section_data* vndata = new Output_data_const_buffer(vnbuf, vnsize, 4, "** version refs"); vnsec->add_output_section_data(vndata); vnsec->set_link_section(dynstr); vnsec->set_info(vnentries); odyn->add_section_address(elfcpp::DT_VERNEED, vnsec); odyn->add_constant(elfcpp::DT_VERNEEDNUM, vnentries); } } // Create the .interp section and PT_INTERP segment. void Layout::create_interp(const Target* target) { const char* interp = parameters->options().dynamic_linker(); if (interp == NULL) { interp = target->dynamic_linker(); gold_assert(interp != NULL); } size_t len = strlen(interp) + 1; Output_section_data* odata = new Output_data_const(interp, len, 1); Output_section* osec = this->choose_output_section(NULL, ".interp", elfcpp::SHT_PROGBITS, elfcpp::SHF_ALLOC, false, ORDER_INTERP, false); osec->add_output_section_data(odata); if (!this->script_options_->saw_phdrs_clause()) { Output_segment* oseg = this->make_output_segment(elfcpp::PT_INTERP, elfcpp::PF_R); oseg->add_output_section_to_nonload(osec, elfcpp::PF_R); } } // Add dynamic tags for the PLT and the dynamic relocs. This is // called by the target-specific code. This does nothing if not doing // a dynamic link. // USE_REL is true for REL relocs rather than RELA relocs. // If PLT_GOT is not NULL, then DT_PLTGOT points to it. // If PLT_REL is not NULL, it is used for DT_PLTRELSZ, and DT_JMPREL, // and we also set DT_PLTREL. We use PLT_REL's output section, since // some targets have multiple reloc sections in PLT_REL. // If DYN_REL is not NULL, it is used for DT_REL/DT_RELA, // DT_RELSZ/DT_RELASZ, DT_RELENT/DT_RELAENT. // If ADD_DEBUG is true, we add a DT_DEBUG entry when generating an // executable. void Layout::add_target_dynamic_tags(bool use_rel, const Output_data* plt_got, const Output_data* plt_rel, const Output_data_reloc_generic* dyn_rel, bool add_debug, bool dynrel_includes_plt) { Output_data_dynamic* odyn = this->dynamic_data_; if (odyn == NULL) return; if (plt_got != NULL && plt_got->output_section() != NULL) odyn->add_section_address(elfcpp::DT_PLTGOT, plt_got); if (plt_rel != NULL && plt_rel->output_section() != NULL) { odyn->add_section_size(elfcpp::DT_PLTRELSZ, plt_rel->output_section()); odyn->add_section_address(elfcpp::DT_JMPREL, plt_rel->output_section()); odyn->add_constant(elfcpp::DT_PLTREL, use_rel ? elfcpp::DT_REL : elfcpp::DT_RELA); } if (dyn_rel != NULL && dyn_rel->output_section() != NULL) { odyn->add_section_address(use_rel ? elfcpp::DT_REL : elfcpp::DT_RELA, dyn_rel); if (plt_rel != NULL && dynrel_includes_plt) odyn->add_section_size(use_rel ? elfcpp::DT_RELSZ : elfcpp::DT_RELASZ, dyn_rel, plt_rel); else odyn->add_section_size(use_rel ? elfcpp::DT_RELSZ : elfcpp::DT_RELASZ, dyn_rel); const int size = parameters->target().get_size(); elfcpp::DT rel_tag; int rel_size; if (use_rel) { rel_tag = elfcpp::DT_RELENT; if (size == 32) rel_size = Reloc_types<elfcpp::SHT_REL, 32, false>::reloc_size; else if (size == 64) rel_size = Reloc_types<elfcpp::SHT_REL, 64, false>::reloc_size; else gold_unreachable(); } else { rel_tag = elfcpp::DT_RELAENT; if (size == 32) rel_size = Reloc_types<elfcpp::SHT_RELA, 32, false>::reloc_size; else if (size == 64) rel_size = Reloc_types<elfcpp::SHT_RELA, 64, false>::reloc_size; else gold_unreachable(); } odyn->add_constant(rel_tag, rel_size); if (parameters->options().combreloc()) { size_t c = dyn_rel->relative_reloc_count(); if (c > 0) odyn->add_constant((use_rel ? elfcpp::DT_RELCOUNT : elfcpp::DT_RELACOUNT), c); } } if (add_debug && !parameters->options().shared()) { // The value of the DT_DEBUG tag is filled in by the dynamic // linker at run time, and used by the debugger. odyn->add_constant(elfcpp::DT_DEBUG, 0); } } // Finish the .dynamic section and PT_DYNAMIC segment. void Layout::finish_dynamic_section(const Input_objects* input_objects, const Symbol_table* symtab) { if (!this->script_options_->saw_phdrs_clause()) { Output_segment* oseg = this->make_output_segment(elfcpp::PT_DYNAMIC, (elfcpp::PF_R | elfcpp::PF_W)); oseg->add_output_section_to_nonload(this->dynamic_section_, elfcpp::PF_R | elfcpp::PF_W); } Output_data_dynamic* const odyn = this->dynamic_data_; for (Input_objects::Dynobj_iterator p = input_objects->dynobj_begin(); p != input_objects->dynobj_end(); ++p) { if (!(*p)->is_needed() && (*p)->as_needed()) { // This dynamic object was linked with --as-needed, but it // is not needed. continue; } odyn->add_string(elfcpp::DT_NEEDED, (*p)->soname()); } if (parameters->options().shared()) { const char* soname = parameters->options().soname(); if (soname != NULL) odyn->add_string(elfcpp::DT_SONAME, soname); } Symbol* sym = symtab->lookup(parameters->options().init()); if (sym != NULL && sym->is_defined() && !sym->is_from_dynobj()) odyn->add_symbol(elfcpp::DT_INIT, sym); sym = symtab->lookup(parameters->options().fini()); if (sym != NULL && sym->is_defined() && !sym->is_from_dynobj()) odyn->add_symbol(elfcpp::DT_FINI, sym); // Look for .init_array, .preinit_array and .fini_array by checking // section types. for(Layout::Section_list::const_iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) switch((*p)->type()) { case elfcpp::SHT_FINI_ARRAY: odyn->add_section_address(elfcpp::DT_FINI_ARRAY, *p); odyn->add_section_size(elfcpp::DT_FINI_ARRAYSZ, *p); break; case elfcpp::SHT_INIT_ARRAY: odyn->add_section_address(elfcpp::DT_INIT_ARRAY, *p); odyn->add_section_size(elfcpp::DT_INIT_ARRAYSZ, *p); break; case elfcpp::SHT_PREINIT_ARRAY: odyn->add_section_address(elfcpp::DT_PREINIT_ARRAY, *p); odyn->add_section_size(elfcpp::DT_PREINIT_ARRAYSZ, *p); break; default: break; } // Add a DT_RPATH entry if needed. const General_options::Dir_list& rpath(parameters->options().rpath()); if (!rpath.empty()) { std::string rpath_val; for (General_options::Dir_list::const_iterator p = rpath.begin(); p != rpath.end(); ++p) { if (rpath_val.empty()) rpath_val = p->name(); else { // Eliminate duplicates. General_options::Dir_list::const_iterator q; for (q = rpath.begin(); q != p; ++q) if (q->name() == p->name()) break; if (q == p) { rpath_val += ':'; rpath_val += p->name(); } } } odyn->add_string(elfcpp::DT_RPATH, rpath_val); if (parameters->options().enable_new_dtags()) odyn->add_string(elfcpp::DT_RUNPATH, rpath_val); } // Look for text segments that have dynamic relocations. bool have_textrel = false; if (!this->script_options_->saw_sections_clause()) { for (Segment_list::const_iterator p = this->segment_list_.begin(); p != this->segment_list_.end(); ++p) { if (((*p)->flags() & elfcpp::PF_W) == 0 && (*p)->has_dynamic_reloc()) { have_textrel = true; break; } } } else { // We don't know the section -> segment mapping, so we are // conservative and just look for readonly sections with // relocations. If those sections wind up in writable segments, // then we have created an unnecessary DT_TEXTREL entry. for (Section_list::const_iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) { if (((*p)->flags() & elfcpp::SHF_ALLOC) != 0 && ((*p)->flags() & elfcpp::SHF_WRITE) == 0 && ((*p)->has_dynamic_reloc())) { have_textrel = true; break; } } } // Add a DT_FLAGS entry. We add it even if no flags are set so that // post-link tools can easily modify these flags if desired. unsigned int flags = 0; if (have_textrel) { // Add a DT_TEXTREL for compatibility with older loaders. odyn->add_constant(elfcpp::DT_TEXTREL, 0); flags |= elfcpp::DF_TEXTREL; if (parameters->options().text()) gold_error(_("read-only segment has dynamic relocations")); else if (parameters->options().warn_shared_textrel() && parameters->options().shared()) gold_warning(_("shared library text segment is not shareable")); } if (parameters->options().shared() && this->has_static_tls()) flags |= elfcpp::DF_STATIC_TLS; if (parameters->options().origin()) flags |= elfcpp::DF_ORIGIN; if (parameters->options().Bsymbolic()) { flags |= elfcpp::DF_SYMBOLIC; // Add DT_SYMBOLIC for compatibility with older loaders. odyn->add_constant(elfcpp::DT_SYMBOLIC, 0); } if (parameters->options().now()) flags |= elfcpp::DF_BIND_NOW; odyn->add_constant(elfcpp::DT_FLAGS, flags); flags = 0; if (parameters->options().initfirst()) flags |= elfcpp::DF_1_INITFIRST; if (parameters->options().interpose()) flags |= elfcpp::DF_1_INTERPOSE; if (parameters->options().loadfltr()) flags |= elfcpp::DF_1_LOADFLTR; if (parameters->options().nodefaultlib()) flags |= elfcpp::DF_1_NODEFLIB; if (parameters->options().nodelete()) flags |= elfcpp::DF_1_NODELETE; if (parameters->options().nodlopen()) flags |= elfcpp::DF_1_NOOPEN; if (parameters->options().nodump()) flags |= elfcpp::DF_1_NODUMP; if (!parameters->options().shared()) flags &= ~(elfcpp::DF_1_INITFIRST | elfcpp::DF_1_NODELETE | elfcpp::DF_1_NOOPEN); if (parameters->options().origin()) flags |= elfcpp::DF_1_ORIGIN; if (parameters->options().now()) flags |= elfcpp::DF_1_NOW; if (flags) odyn->add_constant(elfcpp::DT_FLAGS_1, flags); } // Set the size of the _DYNAMIC symbol table to be the size of the // dynamic data. void Layout::set_dynamic_symbol_size(const Symbol_table* symtab) { Output_data_dynamic* const odyn = this->dynamic_data_; odyn->finalize_data_size(); off_t data_size = odyn->data_size(); const int size = parameters->target().get_size(); if (size == 32) symtab->get_sized_symbol<32>(this->dynamic_symbol_)->set_symsize(data_size); else if (size == 64) symtab->get_sized_symbol<64>(this->dynamic_symbol_)->set_symsize(data_size); else gold_unreachable(); } // The mapping of input section name prefixes to output section names. // In some cases one prefix is itself a prefix of another prefix; in // such a case the longer prefix must come first. These prefixes are // based on the GNU linker default ELF linker script. #define MAPPING_INIT(f, t) { f, sizeof(f) - 1, t, sizeof(t) - 1 } const Layout::Section_name_mapping Layout::section_name_mapping[] = { MAPPING_INIT(".text.", ".text"), MAPPING_INIT(".ctors.", ".ctors"), MAPPING_INIT(".dtors.", ".dtors"), MAPPING_INIT(".rodata.", ".rodata"), MAPPING_INIT(".data.rel.ro.local", ".data.rel.ro.local"), MAPPING_INIT(".data.rel.ro", ".data.rel.ro"), MAPPING_INIT(".data.", ".data"), MAPPING_INIT(".bss.", ".bss"), MAPPING_INIT(".tdata.", ".tdata"), MAPPING_INIT(".tbss.", ".tbss"), MAPPING_INIT(".init_array.", ".init_array"), MAPPING_INIT(".fini_array.", ".fini_array"), MAPPING_INIT(".sdata.", ".sdata"), MAPPING_INIT(".sbss.", ".sbss"), // FIXME: In the GNU linker, .sbss2 and .sdata2 are handled // differently depending on whether it is creating a shared library. MAPPING_INIT(".sdata2.", ".sdata"), MAPPING_INIT(".sbss2.", ".sbss"), MAPPING_INIT(".lrodata.", ".lrodata"), MAPPING_INIT(".ldata.", ".ldata"), MAPPING_INIT(".lbss.", ".lbss"), MAPPING_INIT(".gcc_except_table.", ".gcc_except_table"), MAPPING_INIT(".gnu.linkonce.d.rel.ro.local.", ".data.rel.ro.local"), MAPPING_INIT(".gnu.linkonce.d.rel.ro.", ".data.rel.ro"), MAPPING_INIT(".gnu.linkonce.t.", ".text"), MAPPING_INIT(".gnu.linkonce.r.", ".rodata"), MAPPING_INIT(".gnu.linkonce.d.", ".data"), MAPPING_INIT(".gnu.linkonce.b.", ".bss"), MAPPING_INIT(".gnu.linkonce.s.", ".sdata"), MAPPING_INIT(".gnu.linkonce.sb.", ".sbss"), MAPPING_INIT(".gnu.linkonce.s2.", ".sdata"), MAPPING_INIT(".gnu.linkonce.sb2.", ".sbss"), MAPPING_INIT(".gnu.linkonce.wi.", ".debug_info"), MAPPING_INIT(".gnu.linkonce.td.", ".tdata"), MAPPING_INIT(".gnu.linkonce.tb.", ".tbss"), MAPPING_INIT(".gnu.linkonce.lr.", ".lrodata"), MAPPING_INIT(".gnu.linkonce.l.", ".ldata"), MAPPING_INIT(".gnu.linkonce.lb.", ".lbss"), MAPPING_INIT(".ARM.extab", ".ARM.extab"), MAPPING_INIT(".gnu.linkonce.armextab.", ".ARM.extab"), MAPPING_INIT(".ARM.exidx", ".ARM.exidx"), MAPPING_INIT(".gnu.linkonce.armexidx.", ".ARM.exidx"), }; #undef MAPPING_INIT const int Layout::section_name_mapping_count = (sizeof(Layout::section_name_mapping) / sizeof(Layout::section_name_mapping[0])); // Choose the output section name to use given an input section name. // Set *PLEN to the length of the name. *PLEN is initialized to the // length of NAME. const char* Layout::output_section_name(const char* name, size_t* plen) { // gcc 4.3 generates the following sorts of section names when it // needs a section name specific to a function: // .text.FN // .rodata.FN // .sdata2.FN // .data.FN // .data.rel.FN // .data.rel.local.FN // .data.rel.ro.FN // .data.rel.ro.local.FN // .sdata.FN // .bss.FN // .sbss.FN // .tdata.FN // .tbss.FN // The GNU linker maps all of those to the part before the .FN, // except that .data.rel.local.FN is mapped to .data, and // .data.rel.ro.local.FN is mapped to .data.rel.ro. The sections // beginning with .data.rel.ro.local are grouped together. // For an anonymous namespace, the string FN can contain a '.'. // Also of interest: .rodata.strN.N, .rodata.cstN, both of which the // GNU linker maps to .rodata. // The .data.rel.ro sections are used with -z relro. The sections // are recognized by name. We use the same names that the GNU // linker does for these sections. // It is hard to handle this in a principled way, so we don't even // try. We use a table of mappings. If the input section name is // not found in the table, we simply use it as the output section // name. const Section_name_mapping* psnm = section_name_mapping; for (int i = 0; i < section_name_mapping_count; ++i, ++psnm) { if (strncmp(name, psnm->from, psnm->fromlen) == 0) { *plen = psnm->tolen; return psnm->to; } } return name; } // Check if a comdat group or .gnu.linkonce section with the given // NAME is selected for the link. If there is already a section, // *KEPT_SECTION is set to point to the existing section and the // function returns false. Otherwise, OBJECT, SHNDX, IS_COMDAT, and // IS_GROUP_NAME are recorded for this NAME in the layout object, // *KEPT_SECTION is set to the internal copy and the function returns // true. bool Layout::find_or_add_kept_section(const std::string& name, Relobj* object, unsigned int shndx, bool is_comdat, bool is_group_name, Kept_section** kept_section) { // It's normal to see a couple of entries here, for the x86 thunk // sections. If we see more than a few, we're linking a C++ // program, and we resize to get more space to minimize rehashing. if (this->signatures_.size() > 4 && !this->resized_signatures_) { reserve_unordered_map(&this->signatures_, this->number_of_input_files_ * 64); this->resized_signatures_ = true; } Kept_section candidate; std::pair<Signatures::iterator, bool> ins = this->signatures_.insert(std::make_pair(name, candidate)); if (kept_section != NULL) *kept_section = &ins.first->second; if (ins.second) { // This is the first time we've seen this signature. ins.first->second.set_object(object); ins.first->second.set_shndx(shndx); if (is_comdat) ins.first->second.set_is_comdat(); if (is_group_name) ins.first->second.set_is_group_name(); return true; } // We have already seen this signature. if (ins.first->second.is_group_name()) { // We've already seen a real section group with this signature. // If the kept group is from a plugin object, and we're in the // replacement phase, accept the new one as a replacement. if (ins.first->second.object() == NULL && parameters->options().plugins()->in_replacement_phase()) { ins.first->second.set_object(object); ins.first->second.set_shndx(shndx); return true; } return false; } else if (is_group_name) { // This is a real section group, and we've already seen a // linkonce section with this signature. Record that we've seen // a section group, and don't include this section group. ins.first->second.set_is_group_name(); return false; } else { // We've already seen a linkonce section and this is a linkonce // section. These don't block each other--this may be the same // symbol name with different section types. return true; } } // Store the allocated sections into the section list. void Layout::get_allocated_sections(Section_list* section_list) const { for (Section_list::const_iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) if (((*p)->flags() & elfcpp::SHF_ALLOC) != 0) section_list->push_back(*p); } // Create an output segment. Output_segment* Layout::make_output_segment(elfcpp::Elf_Word type, elfcpp::Elf_Word flags) { gold_assert(!parameters->options().relocatable()); Output_segment* oseg = new Output_segment(type, flags); this->segment_list_.push_back(oseg); if (type == elfcpp::PT_TLS) this->tls_segment_ = oseg; else if (type == elfcpp::PT_GNU_RELRO) this->relro_segment_ = oseg; return oseg; } // Return the file offset of the normal symbol table. off_t Layout::symtab_section_offset() const { if (this->symtab_section_ != NULL) return this->symtab_section_->offset(); return 0; } // Write out the Output_sections. Most won't have anything to write, // since most of the data will come from input sections which are // handled elsewhere. But some Output_sections do have Output_data. void Layout::write_output_sections(Output_file* of) const { for (Section_list::const_iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) { if (!(*p)->after_input_sections()) (*p)->write(of); } } // Write out data not associated with a section or the symbol table. void Layout::write_data(const Symbol_table* symtab, Output_file* of) const { if (!parameters->options().strip_all()) { const Output_section* symtab_section = this->symtab_section_; for (Section_list::const_iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) { if ((*p)->needs_symtab_index()) { gold_assert(symtab_section != NULL); unsigned int index = (*p)->symtab_index(); gold_assert(index > 0 && index != -1U); off_t off = (symtab_section->offset() + index * symtab_section->entsize()); symtab->write_section_symbol(*p, this->symtab_xindex_, of, off); } } } const Output_section* dynsym_section = this->dynsym_section_; for (Section_list::const_iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) { if ((*p)->needs_dynsym_index()) { gold_assert(dynsym_section != NULL); unsigned int index = (*p)->dynsym_index(); gold_assert(index > 0 && index != -1U); off_t off = (dynsym_section->offset() + index * dynsym_section->entsize()); symtab->write_section_symbol(*p, this->dynsym_xindex_, of, off); } } // Write out the Output_data which are not in an Output_section. for (Data_list::const_iterator p = this->special_output_list_.begin(); p != this->special_output_list_.end(); ++p) (*p)->write(of); } // Write out the Output_sections which can only be written after the // input sections are complete. void Layout::write_sections_after_input_sections(Output_file* of) { // Determine the final section offsets, and thus the final output // file size. Note we finalize the .shstrab last, to allow the // after_input_section sections to modify their section-names before // writing. if (this->any_postprocessing_sections_) { off_t off = this->output_file_size_; off = this->set_section_offsets(off, POSTPROCESSING_SECTIONS_PASS); // Now that we've finalized the names, we can finalize the shstrab. off = this->set_section_offsets(off, STRTAB_AFTER_POSTPROCESSING_SECTIONS_PASS); if (off > this->output_file_size_) { of->resize(off); this->output_file_size_ = off; } } for (Section_list::const_iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) { if ((*p)->after_input_sections()) (*p)->write(of); } this->section_headers_->write(of); } // If the build ID requires computing a checksum, do so here, and // write it out. We compute a checksum over the entire file because // that is simplest. void Layout::write_build_id(Output_file* of) const { if (this->build_id_note_ == NULL) return; const unsigned char* iv = of->get_input_view(0, this->output_file_size_); unsigned char* ov = of->get_output_view(this->build_id_note_->offset(), this->build_id_note_->data_size()); const char* style = parameters->options().build_id(); if (strcmp(style, "sha1") == 0) { sha1_ctx ctx; sha1_init_ctx(&ctx); sha1_process_bytes(iv, this->output_file_size_, &ctx); sha1_finish_ctx(&ctx, ov); } else if (strcmp(style, "md5") == 0) { md5_ctx ctx; md5_init_ctx(&ctx); md5_process_bytes(iv, this->output_file_size_, &ctx); md5_finish_ctx(&ctx, ov); } else gold_unreachable(); of->write_output_view(this->build_id_note_->offset(), this->build_id_note_->data_size(), ov); of->free_input_view(0, this->output_file_size_, iv); } // Write out a binary file. This is called after the link is // complete. IN is the temporary output file we used to generate the // ELF code. We simply walk through the segments, read them from // their file offset in IN, and write them to their load address in // the output file. FIXME: with a bit more work, we could support // S-records and/or Intel hex format here. void Layout::write_binary(Output_file* in) const { gold_assert(parameters->options().oformat_enum() == General_options::OBJECT_FORMAT_BINARY); // Get the size of the binary file. uint64_t max_load_address = 0; for (Segment_list::const_iterator p = this->segment_list_.begin(); p != this->segment_list_.end(); ++p) { if ((*p)->type() == elfcpp::PT_LOAD && (*p)->filesz() > 0) { uint64_t max_paddr = (*p)->paddr() + (*p)->filesz(); if (max_paddr > max_load_address) max_load_address = max_paddr; } } Output_file out(parameters->options().output_file_name()); out.open(max_load_address); for (Segment_list::const_iterator p = this->segment_list_.begin(); p != this->segment_list_.end(); ++p) { if ((*p)->type() == elfcpp::PT_LOAD && (*p)->filesz() > 0) { const unsigned char* vin = in->get_input_view((*p)->offset(), (*p)->filesz()); unsigned char* vout = out.get_output_view((*p)->paddr(), (*p)->filesz()); memcpy(vout, vin, (*p)->filesz()); out.write_output_view((*p)->paddr(), (*p)->filesz(), vout); in->free_input_view((*p)->offset(), (*p)->filesz(), vin); } } out.close(); } // Print the output sections to the map file. void Layout::print_to_mapfile(Mapfile* mapfile) const { for (Segment_list::const_iterator p = this->segment_list_.begin(); p != this->segment_list_.end(); ++p) (*p)->print_sections_to_mapfile(mapfile); } // Print statistical information to stderr. This is used for --stats. void Layout::print_stats() const { this->namepool_.print_stats("section name pool"); this->sympool_.print_stats("output symbol name pool"); this->dynpool_.print_stats("dynamic name pool"); for (Section_list::const_iterator p = this->section_list_.begin(); p != this->section_list_.end(); ++p) (*p)->print_merge_stats(); } // Write_sections_task methods. // We can always run this task. Task_token* Write_sections_task::is_runnable() { return NULL; } // We need to unlock both OUTPUT_SECTIONS_BLOCKER and FINAL_BLOCKER // when finished. void Write_sections_task::locks(Task_locker* tl) { tl->add(this, this->output_sections_blocker_); tl->add(this, this->final_blocker_); } // Run the task--write out the data. void Write_sections_task::run(Workqueue*) { this->layout_->write_output_sections(this->of_); } // Write_data_task methods. // We can always run this task. Task_token* Write_data_task::is_runnable() { return NULL; } // We need to unlock FINAL_BLOCKER when finished. void Write_data_task::locks(Task_locker* tl) { tl->add(this, this->final_blocker_); } // Run the task--write out the data. void Write_data_task::run(Workqueue*) { this->layout_->write_data(this->symtab_, this->of_); } // Write_symbols_task methods. // We can always run this task. Task_token* Write_symbols_task::is_runnable() { return NULL; } // We need to unlock FINAL_BLOCKER when finished. void Write_symbols_task::locks(Task_locker* tl) { tl->add(this, this->final_blocker_); } // Run the task--write out the symbols. void Write_symbols_task::run(Workqueue*) { this->symtab_->write_globals(this->sympool_, this->dynpool_, this->layout_->symtab_xindex(), this->layout_->dynsym_xindex(), this->of_); } // Write_after_input_sections_task methods. // We can only run this task after the input sections have completed. Task_token* Write_after_input_sections_task::is_runnable() { if (this->input_sections_blocker_->is_blocked()) return this->input_sections_blocker_; return NULL; } // We need to unlock FINAL_BLOCKER when finished. void Write_after_input_sections_task::locks(Task_locker* tl) { tl->add(this, this->final_blocker_); } // Run the task. void Write_after_input_sections_task::run(Workqueue*) { this->layout_->write_sections_after_input_sections(this->of_); } // Close_task_runner methods. // Run the task--close the file. void Close_task_runner::run(Workqueue*, const Task*) { // If we need to compute a checksum for the BUILD if, we do so here. this->layout_->write_build_id(this->of_); // If we've been asked to create a binary file, we do so here. if (this->options_->oformat_enum() != General_options::OBJECT_FORMAT_ELF) this->layout_->write_binary(this->of_); this->of_->close(); } // Instantiate the templates we need. We could use the configure // script to restrict this to only the ones for implemented targets. #ifdef HAVE_TARGET_32_LITTLE template Output_section* Layout::init_fixed_output_section<32, false>( const char* name, elfcpp::Shdr<32, false>& shdr); #endif #ifdef HAVE_TARGET_32_BIG template Output_section* Layout::init_fixed_output_section<32, true>( const char* name, elfcpp::Shdr<32, true>& shdr); #endif #ifdef HAVE_TARGET_64_LITTLE template Output_section* Layout::init_fixed_output_section<64, false>( const char* name, elfcpp::Shdr<64, false>& shdr); #endif #ifdef HAVE_TARGET_64_BIG template Output_section* Layout::init_fixed_output_section<64, true>( const char* name, elfcpp::Shdr<64, true>& shdr); #endif #ifdef HAVE_TARGET_32_LITTLE template Output_section* Layout::layout<32, false>(Sized_relobj_file<32, false>* object, unsigned int shndx, const char* name, const elfcpp::Shdr<32, false>& shdr, unsigned int, unsigned int, off_t*); #endif #ifdef HAVE_TARGET_32_BIG template Output_section* Layout::layout<32, true>(Sized_relobj_file<32, true>* object, unsigned int shndx, const char* name, const elfcpp::Shdr<32, true>& shdr, unsigned int, unsigned int, off_t*); #endif #ifdef HAVE_TARGET_64_LITTLE template Output_section* Layout::layout<64, false>(Sized_relobj_file<64, false>* object, unsigned int shndx, const char* name, const elfcpp::Shdr<64, false>& shdr, unsigned int, unsigned int, off_t*); #endif #ifdef HAVE_TARGET_64_BIG template Output_section* Layout::layout<64, true>(Sized_relobj_file<64, true>* object, unsigned int shndx, const char* name, const elfcpp::Shdr<64, true>& shdr, unsigned int, unsigned int, off_t*); #endif #ifdef HAVE_TARGET_32_LITTLE template Output_section* Layout::layout_reloc<32, false>(Sized_relobj_file<32, false>* object, unsigned int reloc_shndx, const elfcpp::Shdr<32, false>& shdr, Output_section* data_section, Relocatable_relocs* rr); #endif #ifdef HAVE_TARGET_32_BIG template Output_section* Layout::layout_reloc<32, true>(Sized_relobj_file<32, true>* object, unsigned int reloc_shndx, const elfcpp::Shdr<32, true>& shdr, Output_section* data_section, Relocatable_relocs* rr); #endif #ifdef HAVE_TARGET_64_LITTLE template Output_section* Layout::layout_reloc<64, false>(Sized_relobj_file<64, false>* object, unsigned int reloc_shndx, const elfcpp::Shdr<64, false>& shdr, Output_section* data_section, Relocatable_relocs* rr); #endif #ifdef HAVE_TARGET_64_BIG template Output_section* Layout::layout_reloc<64, true>(Sized_relobj_file<64, true>* object, unsigned int reloc_shndx, const elfcpp::Shdr<64, true>& shdr, Output_section* data_section, Relocatable_relocs* rr); #endif #ifdef HAVE_TARGET_32_LITTLE template void Layout::layout_group<32, false>(Symbol_table* symtab, Sized_relobj_file<32, false>* object, unsigned int, const char* group_section_name, const char* signature, const elfcpp::Shdr<32, false>& shdr, elfcpp::Elf_Word flags, std::vector<unsigned int>* shndxes); #endif #ifdef HAVE_TARGET_32_BIG template void Layout::layout_group<32, true>(Symbol_table* symtab, Sized_relobj_file<32, true>* object, unsigned int, const char* group_section_name, const char* signature, const elfcpp::Shdr<32, true>& shdr, elfcpp::Elf_Word flags, std::vector<unsigned int>* shndxes); #endif #ifdef HAVE_TARGET_64_LITTLE template void Layout::layout_group<64, false>(Symbol_table* symtab, Sized_relobj_file<64, false>* object, unsigned int, const char* group_section_name, const char* signature, const elfcpp::Shdr<64, false>& shdr, elfcpp::Elf_Word flags, std::vector<unsigned int>* shndxes); #endif #ifdef HAVE_TARGET_64_BIG template void Layout::layout_group<64, true>(Symbol_table* symtab, Sized_relobj_file<64, true>* object, unsigned int, const char* group_section_name, const char* signature, const elfcpp::Shdr<64, true>& shdr, elfcpp::Elf_Word flags, std::vector<unsigned int>* shndxes); #endif #ifdef HAVE_TARGET_32_LITTLE template Output_section* Layout::layout_eh_frame<32, false>(Sized_relobj_file<32, false>* object, const unsigned char* symbols, off_t symbols_size, const unsigned char* symbol_names, off_t symbol_names_size, unsigned int shndx, const elfcpp::Shdr<32, false>& shdr, unsigned int reloc_shndx, unsigned int reloc_type, off_t* off); #endif #ifdef HAVE_TARGET_32_BIG template Output_section* Layout::layout_eh_frame<32, true>(Sized_relobj_file<32, true>* object, const unsigned char* symbols, off_t symbols_size, const unsigned char* symbol_names, off_t symbol_names_size, unsigned int shndx, const elfcpp::Shdr<32, true>& shdr, unsigned int reloc_shndx, unsigned int reloc_type, off_t* off); #endif #ifdef HAVE_TARGET_64_LITTLE template Output_section* Layout::layout_eh_frame<64, false>(Sized_relobj_file<64, false>* object, const unsigned char* symbols, off_t symbols_size, const unsigned char* symbol_names, off_t symbol_names_size, unsigned int shndx, const elfcpp::Shdr<64, false>& shdr, unsigned int reloc_shndx, unsigned int reloc_type, off_t* off); #endif #ifdef HAVE_TARGET_64_BIG template Output_section* Layout::layout_eh_frame<64, true>(Sized_relobj_file<64, true>* object, const unsigned char* symbols, off_t symbols_size, const unsigned char* symbol_names, off_t symbol_names_size, unsigned int shndx, const elfcpp::Shdr<64, true>& shdr, unsigned int reloc_shndx, unsigned int reloc_type, off_t* off); #endif } // End namespace gold.
Go to most recent revision | Compare with Previous | Blame | View Log