// fileread.cc -- read files for gold // Copyright 2006, 2007, 2008, 2009, 2010 Free Software Foundation, Inc. // Written by Ian Lance Taylor . // This file is part of gold. // This program is free software; you can redistribute it and/or modify // it under the terms of the GNU General Public License as published by // the Free Software Foundation; either version 3 of the License, or // (at your option) any later version. // This program is distributed in the hope that it will be useful, // but WITHOUT ANY WARRANTY; without even the implied warranty of // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the // GNU General Public License for more details. // You should have received a copy of the GNU General Public License // along with this program; if not, write to the Free Software // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston, // MA 02110-1301, USA. #include "gold.h" #include #include #include #include #include #ifdef HAVE_READV #include #endif #include #include "filenames.h" #include "debug.h" #include "parameters.h" #include "options.h" #include "dirsearch.h" #include "target.h" #include "binary.h" #include "descriptors.h" #include "gold-threads.h" #include "fileread.h" #ifndef HAVE_READV struct iovec { void* iov_base; size_t iov_len; }; ssize_t readv(int, const iovec*, int) { gold_unreachable(); } #endif namespace gold { // Class File_read::View. File_read::View::~View() { gold_assert(!this->is_locked()); switch (this->data_ownership_) { case DATA_ALLOCATED_ARRAY: delete[] this->data_; break; case DATA_MMAPPED: if (::munmap(const_cast(this->data_), this->size_) != 0) gold_warning(_("munmap failed: %s"), strerror(errno)); File_read::current_mapped_bytes -= this->size_; break; case DATA_NOT_OWNED: break; default: gold_unreachable(); } } void File_read::View::lock() { ++this->lock_count_; } void File_read::View::unlock() { gold_assert(this->lock_count_ > 0); --this->lock_count_; } bool File_read::View::is_locked() { return this->lock_count_ > 0; } // Class File_read. // A lock for the File_read static variables. static Lock* file_counts_lock = NULL; static Initialize_lock file_counts_initialize_lock(&file_counts_lock); // The File_read static variables. unsigned long long File_read::total_mapped_bytes; unsigned long long File_read::current_mapped_bytes; unsigned long long File_read::maximum_mapped_bytes; File_read::~File_read() { gold_assert(this->token_.is_writable()); if (this->is_descriptor_opened_) { release_descriptor(this->descriptor_, true); this->descriptor_ = -1; this->is_descriptor_opened_ = false; } this->name_.clear(); this->clear_views(CLEAR_VIEWS_ALL); } // Open the file. bool File_read::open(const Task* task, const std::string& name) { gold_assert(this->token_.is_writable() && this->descriptor_ < 0 && !this->is_descriptor_opened_ && this->name_.empty()); this->name_ = name; this->descriptor_ = open_descriptor(-1, this->name_.c_str(), O_RDONLY); if (this->descriptor_ >= 0) { this->is_descriptor_opened_ = true; struct stat s; if (::fstat(this->descriptor_, &s) < 0) gold_error(_("%s: fstat failed: %s"), this->name_.c_str(), strerror(errno)); this->size_ = s.st_size; gold_debug(DEBUG_FILES, "Attempt to open %s succeeded", this->name_.c_str()); this->token_.add_writer(task); } return this->descriptor_ >= 0; } // Open the file with the contents in memory. bool File_read::open(const Task* task, const std::string& name, const unsigned char* contents, off_t size) { gold_assert(this->token_.is_writable() && this->descriptor_ < 0 && !this->is_descriptor_opened_ && this->name_.empty()); this->name_ = name; this->whole_file_view_ = new View(0, size, contents, 0, false, View::DATA_NOT_OWNED); this->add_view(this->whole_file_view_); this->size_ = size; this->token_.add_writer(task); return true; } // Reopen a descriptor if necessary. void File_read::reopen_descriptor() { if (!this->is_descriptor_opened_) { this->descriptor_ = open_descriptor(this->descriptor_, this->name_.c_str(), O_RDONLY); if (this->descriptor_ < 0) gold_fatal(_("could not reopen file %s"), this->name_.c_str()); this->is_descriptor_opened_ = true; } } // Release the file. This is called when we are done with the file in // a Task. void File_read::release() { gold_assert(this->is_locked()); if (!parameters->options_valid() || parameters->options().stats()) { file_counts_initialize_lock.initialize(); Hold_optional_lock hl(file_counts_lock); File_read::total_mapped_bytes += this->mapped_bytes_; File_read::current_mapped_bytes += this->mapped_bytes_; if (File_read::current_mapped_bytes > File_read::maximum_mapped_bytes) File_read::maximum_mapped_bytes = File_read::current_mapped_bytes; } this->mapped_bytes_ = 0; // Only clear views if there is only one attached object. Otherwise // we waste time trying to clear cached archive views. Similarly // for releasing the descriptor. if (this->object_count_ <= 1) { this->clear_views(CLEAR_VIEWS_NORMAL); if (this->is_descriptor_opened_) { release_descriptor(this->descriptor_, false); this->is_descriptor_opened_ = false; } } this->released_ = true; } // Lock the file. void File_read::lock(const Task* task) { gold_assert(this->released_); this->token_.add_writer(task); this->released_ = false; } // Unlock the file. void File_read::unlock(const Task* task) { this->release(); this->token_.remove_writer(task); } // Return whether the file is locked. bool File_read::is_locked() const { if (!this->token_.is_writable()) return true; // The file is not locked, so it should have been released. gold_assert(this->released_); return false; } // See if we have a view which covers the file starting at START for // SIZE bytes. Return a pointer to the View if found, NULL if not. // If BYTESHIFT is not -1U, the returned View must have the specified // byte shift; otherwise, it may have any byte shift. If VSHIFTED is // not NULL, this sets *VSHIFTED to a view which would have worked if // not for the requested BYTESHIFT. inline File_read::View* File_read::find_view(off_t start, section_size_type size, unsigned int byteshift, File_read::View** vshifted) const { if (vshifted != NULL) *vshifted = NULL; // If we have the whole file mmapped, and the alignment is right, // we can return it. if (this->whole_file_view_) if (byteshift == -1U || byteshift == 0) return this->whole_file_view_; off_t page = File_read::page_offset(start); unsigned int bszero = 0; Views::const_iterator p = this->views_.upper_bound(std::make_pair(page - 1, bszero)); while (p != this->views_.end() && p->first.first <= page) { if (p->second->start() <= start && (p->second->start() + static_cast(p->second->size()) >= start + static_cast(size))) { if (byteshift == -1U || byteshift == p->second->byteshift()) { p->second->set_accessed(); return p->second; } if (vshifted != NULL && *vshifted == NULL) *vshifted = p->second; } ++p; } return NULL; } // Read SIZE bytes from the file starting at offset START. Read into // the buffer at P. void File_read::do_read(off_t start, section_size_type size, void* p) { ssize_t bytes; if (this->whole_file_view_ != NULL) { bytes = this->size_ - start; if (static_cast(bytes) >= size) { memcpy(p, this->whole_file_view_->data() + start, size); return; } } else { this->reopen_descriptor(); bytes = ::pread(this->descriptor_, p, size, start); if (static_cast(bytes) == size) return; if (bytes < 0) { gold_fatal(_("%s: pread failed: %s"), this->filename().c_str(), strerror(errno)); return; } } gold_fatal(_("%s: file too short: read only %lld of %lld bytes at %lld"), this->filename().c_str(), static_cast(bytes), static_cast(size), static_cast(start)); } // Read data from the file. void File_read::read(off_t start, section_size_type size, void* p) { const File_read::View* pv = this->find_view(start, size, -1U, NULL); if (pv != NULL) { memcpy(p, pv->data() + (start - pv->start() + pv->byteshift()), size); return; } this->do_read(start, size, p); } // Add a new view. There may already be an existing view at this // offset. If there is, the new view will be larger, and should // replace the old view. void File_read::add_view(File_read::View* v) { std::pair ins = this->views_.insert(std::make_pair(std::make_pair(v->start(), v->byteshift()), v)); if (ins.second) return; // There was an existing view at this offset. It must not be large // enough. We can't delete it here, since something might be using // it; we put it on a list to be deleted when the file is unlocked. File_read::View* vold = ins.first->second; gold_assert(vold->size() < v->size()); if (vold->should_cache()) { v->set_cache(); vold->clear_cache(); } this->saved_views_.push_back(vold); ins.first->second = v; } // Make a new view with a specified byteshift, reading the data from // the file. File_read::View* File_read::make_view(off_t start, section_size_type size, unsigned int byteshift, bool cache) { gold_assert(size > 0); // Check that start and end of the view are within the file. if (start > this->size_ || (static_cast(size) > static_cast(this->size_ - start))) gold_fatal(_("%s: attempt to map %lld bytes at offset %lld exceeds " "size of file; the file may be corrupt"), this->filename().c_str(), static_cast(size), static_cast(start)); off_t poff = File_read::page_offset(start); section_size_type psize = File_read::pages(size + (start - poff)); if (poff + static_cast(psize) >= this->size_) { psize = this->size_ - poff; gold_assert(psize >= size); } File_read::View* v; if (byteshift != 0) { unsigned char* p = new unsigned char[psize + byteshift]; memset(p, 0, byteshift); this->do_read(poff, psize, p + byteshift); v = new File_read::View(poff, psize, p, byteshift, cache, View::DATA_ALLOCATED_ARRAY); } else { this->reopen_descriptor(); void* p = ::mmap(NULL, psize, PROT_READ, MAP_PRIVATE, this->descriptor_, poff); if (p == MAP_FAILED) gold_fatal(_("%s: mmap offset %lld size %lld failed: %s"), this->filename().c_str(), static_cast(poff), static_cast(psize), strerror(errno)); this->mapped_bytes_ += psize; const unsigned char* pbytes = static_cast(p); v = new File_read::View(poff, psize, pbytes, 0, cache, View::DATA_MMAPPED); } this->add_view(v); return v; } // Find a View or make a new one, shifted as required by the file // offset OFFSET and ALIGNED. File_read::View* File_read::find_or_make_view(off_t offset, off_t start, section_size_type size, bool aligned, bool cache) { unsigned int byteshift; if (offset == 0) byteshift = 0; else { unsigned int target_size = (!parameters->target_valid() ? 64 : parameters->target().get_size()); byteshift = offset & ((target_size / 8) - 1); // Set BYTESHIFT to the number of dummy bytes which must be // inserted before the data in order for this data to be // aligned. if (byteshift != 0) byteshift = (target_size / 8) - byteshift; } // If --map-whole-files is set, make sure we have a // whole file view. Options may not yet be ready, e.g., // when reading a version script. We then default to // --no-map-whole-files. if (this->whole_file_view_ == NULL && parameters->options_valid() && parameters->options().map_whole_files()) this->whole_file_view_ = this->make_view(0, this->size_, 0, cache); // Try to find a View with the required BYTESHIFT. File_read::View* vshifted; File_read::View* v = this->find_view(offset + start, size, aligned ? byteshift : -1U, &vshifted); if (v != NULL) { if (cache) v->set_cache(); return v; } // If VSHIFTED is not NULL, then it has the data we need, but with // the wrong byteshift. v = vshifted; if (v != NULL) { gold_assert(aligned); unsigned char* pbytes = new unsigned char[v->size() + byteshift]; memset(pbytes, 0, byteshift); memcpy(pbytes + byteshift, v->data() + v->byteshift(), v->size()); File_read::View* shifted_view = new File_read::View(v->start(), v->size(), pbytes, byteshift, cache, View::DATA_ALLOCATED_ARRAY); this->add_view(shifted_view); return shifted_view; } // Make a new view. If we don't need an aligned view, use a // byteshift of 0, so that we can use mmap. return this->make_view(offset + start, size, aligned ? byteshift : 0, cache); } // Get a view into the file. const unsigned char* File_read::get_view(off_t offset, off_t start, section_size_type size, bool aligned, bool cache) { File_read::View* pv = this->find_or_make_view(offset, start, size, aligned, cache); return pv->data() + (offset + start - pv->start() + pv->byteshift()); } File_view* File_read::get_lasting_view(off_t offset, off_t start, section_size_type size, bool aligned, bool cache) { File_read::View* pv = this->find_or_make_view(offset, start, size, aligned, cache); pv->lock(); return new File_view(*this, pv, (pv->data() + (offset + start - pv->start() + pv->byteshift()))); } // Use readv to read COUNT entries from RM starting at START. BASE // must be added to all file offsets in RM. void File_read::do_readv(off_t base, const Read_multiple& rm, size_t start, size_t count) { unsigned char discard[File_read::page_size]; iovec iov[File_read::max_readv_entries * 2]; size_t iov_index = 0; off_t first_offset = rm[start].file_offset; off_t last_offset = first_offset; ssize_t want = 0; for (size_t i = 0; i < count; ++i) { const Read_multiple_entry& i_entry(rm[start + i]); if (i_entry.file_offset > last_offset) { size_t skip = i_entry.file_offset - last_offset; gold_assert(skip <= sizeof discard); iov[iov_index].iov_base = discard; iov[iov_index].iov_len = skip; ++iov_index; want += skip; } iov[iov_index].iov_base = i_entry.buffer; iov[iov_index].iov_len = i_entry.size; ++iov_index; want += i_entry.size; last_offset = i_entry.file_offset + i_entry.size; } this->reopen_descriptor(); gold_assert(iov_index < sizeof iov / sizeof iov[0]); if (::lseek(this->descriptor_, base + first_offset, SEEK_SET) < 0) gold_fatal(_("%s: lseek failed: %s"), this->filename().c_str(), strerror(errno)); ssize_t got = ::readv(this->descriptor_, iov, iov_index); if (got < 0) gold_fatal(_("%s: readv failed: %s"), this->filename().c_str(), strerror(errno)); if (got != want) gold_fatal(_("%s: file too short: read only %zd of %zd bytes at %lld"), this->filename().c_str(), got, want, static_cast(base + first_offset)); } // Read several pieces of data from the file. void File_read::read_multiple(off_t base, const Read_multiple& rm) { size_t count = rm.size(); size_t i = 0; while (i < count) { // Find up to MAX_READV_ENTRIES consecutive entries which are // less than one page apart. const Read_multiple_entry& i_entry(rm[i]); off_t i_off = i_entry.file_offset; off_t end_off = i_off + i_entry.size; size_t j; for (j = i + 1; j < count; ++j) { if (j - i >= File_read::max_readv_entries) break; const Read_multiple_entry& j_entry(rm[j]); off_t j_off = j_entry.file_offset; gold_assert(j_off >= end_off); off_t j_end_off = j_off + j_entry.size; if (j_end_off - end_off >= File_read::page_size) break; end_off = j_end_off; } if (j == i + 1) this->read(base + i_off, i_entry.size, i_entry.buffer); else { File_read::View* view = this->find_view(base + i_off, end_off - i_off, -1U, NULL); if (view == NULL) this->do_readv(base, rm, i, j - i); else { const unsigned char* v = (view->data() + (base + i_off - view->start() + view->byteshift())); for (size_t k = i; k < j; ++k) { const Read_multiple_entry& k_entry(rm[k]); gold_assert((convert_to_section_size_type(k_entry.file_offset - i_off) + k_entry.size) <= convert_to_section_size_type(end_off - i_off)); memcpy(k_entry.buffer, v + (k_entry.file_offset - i_off), k_entry.size); } } } i = j; } } // Mark all views as no longer cached. void File_read::clear_view_cache_marks() { // Just ignore this if there are multiple objects associated with // the file. Otherwise we will wind up uncaching and freeing some // views for other objects. if (this->object_count_ > 1) return; for (Views::iterator p = this->views_.begin(); p != this->views_.end(); ++p) p->second->clear_cache(); for (Saved_views::iterator p = this->saved_views_.begin(); p != this->saved_views_.end(); ++p) (*p)->clear_cache(); } // Remove all the file views. For a file which has multiple // associated objects (i.e., an archive), we keep accessed views // around until next time, in the hopes that they will be useful for // the next object. void File_read::clear_views(Clear_views_mode mode) { bool keep_files_mapped = (parameters->options_valid() && parameters->options().keep_files_mapped()); Views::iterator p = this->views_.begin(); while (p != this->views_.end()) { bool should_delete; if (p->second->is_locked() || p->second->is_permanent_view()) should_delete = false; else if (mode == CLEAR_VIEWS_ALL) should_delete = true; else if (p->second->should_cache() && keep_files_mapped) should_delete = false; else if (this->object_count_ > 1 && p->second->accessed() && mode != CLEAR_VIEWS_ARCHIVE) should_delete = false; else should_delete = true; if (should_delete) { if (p->second == this->whole_file_view_) this->whole_file_view_ = NULL; delete p->second; // map::erase invalidates only the iterator to the deleted // element. Views::iterator pe = p; ++p; this->views_.erase(pe); } else { p->second->clear_accessed(); ++p; } } Saved_views::iterator q = this->saved_views_.begin(); while (q != this->saved_views_.end()) { if (!(*q)->is_locked()) { delete *q; q = this->saved_views_.erase(q); } else { gold_assert(mode != CLEAR_VIEWS_ALL); ++q; } } } // Print statistical information to stderr. This is used for --stats. void File_read::print_stats() { fprintf(stderr, _("%s: total bytes mapped for read: %llu\n"), program_name, File_read::total_mapped_bytes); fprintf(stderr, _("%s: maximum bytes mapped for read at one time: %llu\n"), program_name, File_read::maximum_mapped_bytes); } // Class File_view. File_view::~File_view() { gold_assert(this->file_.is_locked()); this->view_->unlock(); } // Class Input_file. // Create a file for testing. Input_file::Input_file(const Task* task, const char* name, const unsigned char* contents, off_t size) : file_() { this->input_argument_ = new Input_file_argument(name, Input_file_argument::INPUT_FILE_TYPE_FILE, "", false, Position_dependent_options()); bool ok = this->file_.open(task, name, contents, size); gold_assert(ok); } // Return the position dependent options in force for this file. const Position_dependent_options& Input_file::options() const { return this->input_argument_->options(); } // Return the name given by the user. For -lc this will return "c". const char* Input_file::name() const { return this->input_argument_->name(); } // Return whether this file is in a system directory. bool Input_file::is_in_system_directory() const { if (this->is_in_sysroot()) return true; return parameters->options().is_in_system_directory(this->filename()); } // Return whether we are only reading symbols. bool Input_file::just_symbols() const { return this->input_argument_->just_symbols(); } // Return whether this is a file that we will search for in the list // of directories. bool Input_file::will_search_for() const { return (!IS_ABSOLUTE_PATH(this->input_argument_->name()) && (this->input_argument_->is_lib() || this->input_argument_->is_searched_file() || this->input_argument_->extra_search_path() != NULL)); } // Return the file last modification time. Calls gold_fatal if the stat // system call failed. Timespec File_read::get_mtime() { struct stat file_stat; this->reopen_descriptor(); if (fstat(this->descriptor_, &file_stat) < 0) gold_fatal(_("%s: stat failed: %s"), this->name_.c_str(), strerror(errno)); #ifdef HAVE_STAT_ST_MTIM return Timespec(file_stat.st_mtim.tv_sec, file_stat.st_mtim.tv_nsec); #else return Timespec(file_stat.st_mtime, 0); #endif } // Try to find a file in the extra search dirs. Returns true on success. bool Input_file::try_extra_search_path(int* pindex, const Input_file_argument* input_argument, std::string filename, std::string* found_name, std::string* namep) { if (input_argument->extra_search_path() == NULL) return false; std::string name = input_argument->extra_search_path(); if (!IS_DIR_SEPARATOR(name[name.length() - 1])) name += '/'; name += filename; struct stat dummy_stat; if (*pindex > 0 || ::stat(name.c_str(), &dummy_stat) < 0) return false; *found_name = filename; *namep = name; return true; } // Find the actual file. // If the filename is not absolute, we assume it is in the current // directory *except* when: // A) input_argument_->is_lib() is true; // B) input_argument_->is_searched_file() is true; or // C) input_argument_->extra_search_path() is not empty. // In each, we look in extra_search_path + library_path to find // the file location, rather than the current directory. bool Input_file::find_file(const Dirsearch& dirpath, int* pindex, const Input_file_argument* input_argument, bool* is_in_sysroot, std::string* found_name, std::string* namep) { std::string name; // Case 1: name is an absolute file, just try to open it // Case 2: name is relative but is_lib is false, is_searched_file is false, // and extra_search_path is empty if (IS_ABSOLUTE_PATH(input_argument->name()) || (!input_argument->is_lib() && !input_argument->is_searched_file() && input_argument->extra_search_path() == NULL)) { name = input_argument->name(); *found_name = name; *namep = name; return true; } // Case 3: is_lib is true or is_searched_file is true else if (input_argument->is_lib() || input_argument->is_searched_file()) { std::string n1, n2; if (input_argument->is_lib()) { n1 = "lib"; n1 += input_argument->name(); if (parameters->options().is_static() || !input_argument->options().Bdynamic()) n1 += ".a"; else { n2 = n1 + ".a"; n1 += ".so"; } } else n1 = input_argument->name(); if (Input_file::try_extra_search_path(pindex, input_argument, n1, found_name, namep)) return true; if (!n2.empty() && Input_file::try_extra_search_path(pindex, input_argument, n2, found_name, namep)) return true; // It is not in the extra_search_path. name = dirpath.find(n1, n2, is_in_sysroot, pindex); if (name.empty()) { gold_error(_("cannot find %s%s"), input_argument->is_lib() ? "-l" : "", input_argument->name()); return false; } if (n2.empty() || name[name.length() - 1] == 'o') *found_name = n1; else *found_name = n2; *namep = name; return true; } // Case 4: extra_search_path is not empty else { gold_assert(input_argument->extra_search_path() != NULL); if (try_extra_search_path(pindex, input_argument, input_argument->name(), found_name, namep)) return true; // extra_search_path failed, so check the normal search-path. int index = *pindex; if (index > 0) --index; name = dirpath.find(input_argument->name(), "", is_in_sysroot, &index); if (name.empty()) { gold_error(_("cannot find %s"), input_argument->name()); return false; } *pindex = index + 1; return true; } } // Open the file. bool Input_file::open(const Dirsearch& dirpath, const Task* task, int *pindex) { std::string name; if (!Input_file::find_file(dirpath, pindex, this->input_argument_, &this->is_in_sysroot_, &this->found_name_, &name)) return false; // Now that we've figured out where the file lives, try to open it. General_options::Object_format format = this->input_argument_->options().format_enum(); bool ok; if (format == General_options::OBJECT_FORMAT_ELF) { ok = this->file_.open(task, name); this->format_ = FORMAT_ELF; } else { gold_assert(format == General_options::OBJECT_FORMAT_BINARY); ok = this->open_binary(task, name); this->format_ = FORMAT_BINARY; } if (!ok) { gold_error(_("cannot open %s: %s"), name.c_str(), strerror(errno)); this->format_ = FORMAT_NONE; return false; } return true; } // Open a file for --format binary. bool Input_file::open_binary(const Task* task, const std::string& name) { // In order to open a binary file, we need machine code, size, and // endianness. We may not have a valid target at this point, in // which case we use the default target. parameters_force_valid_target(); const Target& target(parameters->target()); Binary_to_elf binary_to_elf(target.machine_code(), target.get_size(), target.is_big_endian(), name); if (!binary_to_elf.convert(task)) return false; return this->file_.open(task, name, binary_to_elf.converted_data_leak(), binary_to_elf.converted_size()); } } // End namespace gold.