2008-08-06 Cary Coutant <ccoutant@google.com>

* archive.cc (Archive::total_archives, Archive::total_members)
	(Archive::total_members_loaded): New variables.
	(Archive::setup): Add parameter.  Add option to preread
	archive symbols.
	(Archive::read_armap): Add counter.
	(Archive::get_file_and_offset): New function.
	(Archive::get_elf_object_for_member): New function.
	(Archive::read_all_symbols): New function.
	(Archive::read_symbols): New function.
	(Archive::add_symbols): Add counters.
	(Archive::include_all_members): Use armap to find members if it's
	already built.
	(Archive::include_member): Skip reading symbols if already read.
	Factored code into Archive::get_file_and_offset and
	Archive::get_elf_object_for_member.  Changed call to
	Mapfile::report_include_archive_member.
	(Archive::print_stats): New function.
	* archive.h: Declare Object and Read_symbols_data classes.
	(Archive::Archive): Add initializers for new members.
	(Archive::setup): Add parameter.
	(Archive::print_stats): New function.
	(Archive::total_archives, Archive::total_members)
	(Archive::total_members_loaded): New variables.
	(Archive::get_file_and_offset): New function.
	(Archive::get_elf_object_for_member): New function.
	(Archive::read_all_symbols): New function.
	(Archive::read_symbols): New function.
	(Archive::Archive_member): New class.
	(Archive::members_): New member.
	(Archive::num_members_): New member.
	* main.cc: Include archive.h.
	(main): Call Archive::print_stats.
	* mapfile.cc (Mapfile::report_include_archive_member): Delete
	archive parameter; member_name is now the fully-decorated name.
	* mapfile.h (Mapfile::report_include_archive_member): Likewise.
	* options.h: (General_options): Add --preread-archive-symbols option.
	* readsyms.cc (Read_symbols::do_read_symbols): Change call to
	Archive::setup.
This commit is contained in:
Cary Coutant 2008-08-07 17:02:11 +00:00
parent ac0cb9be4e
commit ac45a351cf
7 changed files with 391 additions and 209 deletions

View File

@ -63,6 +63,11 @@ struct Archive::Archive_header
char ar_fmag[2];
};
// Class Archive static variables.
unsigned int Archive::total_archives;
unsigned int Archive::total_members;
unsigned int Archive::total_members_loaded;
// Archive methods.
const char Archive::armag[sarmag] =
@ -81,7 +86,7 @@ const char Archive::arfmag[2] = { '`', '\n' };
// table.
void
Archive::setup()
Archive::setup(Input_objects* input_objects)
{
// We need to ignore empty archives.
if (this->input_file_->file().filesize() == sarmag)
@ -117,6 +122,13 @@ Archive::setup()
const char* px = reinterpret_cast<const char*>(p);
this->extended_names_.assign(px, extended_size);
}
bool preread_syms = (parameters->options().threads()
&& parameters->options().preread_archive_symbols());
#ifndef ENABLE_THREADS
preread_syms = false;
#endif
if (preread_syms)
this->read_all_symbols(input_objects);
}
// Unlock any nested archives.
@ -137,6 +149,12 @@ Archive::unlock_nested_archives()
void
Archive::read_armap(off_t start, section_size_type size)
{
// To count the total number of archive members, we'll just count
// the number of times the file offset changes. Since most archives
// group the symbols in the armap by object, this ought to give us
// an accurate count.
off_t last_seen_offset = -1;
// Read in the entire armap.
const unsigned char* p = this->get_view(start, size, true, false);
@ -160,6 +178,11 @@ Archive::read_armap(off_t start, section_size_type size)
this->armap_[i].file_offset = elfcpp::Swap<32, true>::readval(pword);
name_offset += strlen(pnames + name_offset) + 1;
++pword;
if (this->armap_[i].file_offset != last_seen_offset)
{
last_seen_offset = this->armap_[i].file_offset;
++this->num_members_;
}
}
if (static_cast<section_size_type>(name_offset) > names_size)
@ -278,93 +301,6 @@ Archive::interpret_header(const Archive_header* hdr, off_t off,
return member_size;
}
// Select members from the archive and add them to the link. We walk
// through the elements in the archive map, and look each one up in
// the symbol table. If it exists as a strong undefined symbol, we
// pull in the corresponding element. We have to do this in a loop,
// since pulling in one element may create new undefined symbols which
// may be satisfied by other objects in the archive.
void
Archive::add_symbols(Symbol_table* symtab, Layout* layout,
Input_objects* input_objects, Mapfile* mapfile)
{
if (this->input_file_->options().whole_archive())
return this->include_all_members(symtab, layout, input_objects,
mapfile);
input_objects->archive_start(this);
const size_t armap_size = this->armap_.size();
// This is a quick optimization, since we usually see many symbols
// in a row with the same offset. last_seen_offset holds the last
// offset we saw that was present in the seen_offsets_ set.
off_t last_seen_offset = -1;
// Track which symbols in the symbol table we've already found to be
// defined.
bool added_new_object;
do
{
added_new_object = false;
for (size_t i = 0; i < armap_size; ++i)
{
if (this->armap_checked_[i])
continue;
if (this->armap_[i].file_offset == last_seen_offset)
{
this->armap_checked_[i] = true;
continue;
}
if (this->seen_offsets_.find(this->armap_[i].file_offset)
!= this->seen_offsets_.end())
{
this->armap_checked_[i] = true;
last_seen_offset = this->armap_[i].file_offset;
continue;
}
const char* sym_name = (this->armap_names_.data()
+ this->armap_[i].name_offset);
Symbol* sym = symtab->lookup(sym_name);
if (sym == NULL)
{
// Check whether the symbol was named in a -u option.
if (!parameters->options().is_undefined(sym_name))
continue;
}
else if (!sym->is_undefined())
{
this->armap_checked_[i] = true;
continue;
}
else if (sym->binding() == elfcpp::STB_WEAK)
continue;
// We want to include this object in the link.
last_seen_offset = this->armap_[i].file_offset;
this->seen_offsets_.insert(last_seen_offset);
this->armap_checked_[i] = true;
std::string why;
if (sym == NULL)
{
why = "-u ";
why += sym_name;
}
this->include_member(symtab, layout, input_objects,
last_seen_offset, mapfile, sym, why.c_str());
added_new_object = true;
}
}
while (added_new_object);
input_objects->archive_stop(this);
}
// An archive member iterator.
class Archive::const_iterator
@ -494,6 +430,247 @@ Archive::end()
return Archive::const_iterator(this, this->input_file_->file().filesize());
}
// Get the file and offset for an archive member, which may be an
// external member of a thin archive. Set *INPUT_FILE to the
// file containing the actual member, *MEMOFF to the offset
// within that file (0 if not a nested archive), and *MEMBER_NAME
// to the name of the archive member. Return TRUE on success.
bool
Archive::get_file_and_offset(off_t off, Input_objects* input_objects,
Input_file** input_file, off_t* memoff,
std::string* member_name)
{
off_t nested_off;
this->read_header(off, false, member_name, &nested_off);
*input_file = this->input_file_;
*memoff = off + static_cast<off_t>(sizeof(Archive_header));
if (!this->is_thin_archive_)
return true;
// Adjust a relative pathname so that it is relative
// to the directory containing the archive.
if (!IS_ABSOLUTE_PATH(member_name->c_str()))
{
const char* arch_path = this->name().c_str();
const char* basename = lbasename(arch_path);
if (basename > arch_path)
member_name->replace(0, 0,
this->name().substr(0, basename - arch_path));
}
if (nested_off > 0)
{
// This is a member of a nested archive. Open the containing
// archive if we don't already have it open, then do a recursive
// call to include the member from that archive.
Archive* arch;
Nested_archive_table::const_iterator p =
this->nested_archives_.find(*member_name);
if (p != this->nested_archives_.end())
arch = p->second;
else
{
Input_file_argument* input_file_arg =
new Input_file_argument(member_name->c_str(), false, "", false,
parameters->options());
*input_file = new Input_file(input_file_arg);
if (!(*input_file)->open(parameters->options(), *this->dirpath_,
this->task_))
return false;
arch = new Archive(*member_name, *input_file, false, this->dirpath_,
this->task_);
arch->setup(input_objects);
std::pair<Nested_archive_table::iterator, bool> ins =
this->nested_archives_.insert(std::make_pair(*member_name, arch));
gold_assert(ins.second);
}
return arch->get_file_and_offset(nested_off, input_objects,
input_file, memoff, member_name);
}
// This is an external member of a thin archive. Open the
// file as a regular relocatable object file.
Input_file_argument* input_file_arg =
new Input_file_argument(member_name->c_str(), false, "", false,
this->input_file_->options());
*input_file = new Input_file(input_file_arg);
if (!(*input_file)->open(parameters->options(), *this->dirpath_,
this->task_))
return false;
*memoff = 0;
return true;
}
// Return an ELF object for the member at offset OFF. Set *MEMBER_NAME to
// the name of the member.
Object*
Archive::get_elf_object_for_member(off_t off, Input_objects* input_objects)
{
std::string member_name;
Input_file* input_file;
off_t memoff;
if (!this->get_file_and_offset(off, input_objects, &input_file, &memoff,
&member_name))
return NULL;
off_t filesize = input_file->file().filesize();
int read_size = elfcpp::Elf_sizes<64>::ehdr_size;
if (filesize - memoff < read_size)
read_size = filesize - memoff;
if (read_size < 4)
{
gold_error(_("%s: member at %zu is not an ELF object"),
this->name().c_str(), static_cast<size_t>(off));
return NULL;
}
const unsigned char* ehdr = input_file->file().get_view(memoff, 0, read_size,
true, false);
static unsigned char elfmagic[4] =
{
elfcpp::ELFMAG0, elfcpp::ELFMAG1,
elfcpp::ELFMAG2, elfcpp::ELFMAG3
};
if (memcmp(ehdr, elfmagic, 4) != 0)
{
gold_error(_("%s: member at %zu is not an ELF object"),
this->name().c_str(), static_cast<size_t>(off));
return NULL;
}
return make_elf_object((std::string(this->input_file_->filename())
+ "(" + member_name + ")"),
input_file, memoff, ehdr, read_size);
}
// Read the symbols from all the archive members in the link.
void
Archive::read_all_symbols(Input_objects* input_objects)
{
for (Archive::const_iterator p = this->begin();
p != this->end();
++p)
this->read_symbols(input_objects, p->off);
}
// Read the symbols from an archive member in the link. OFF is the file
// offset of the member header.
void
Archive::read_symbols(Input_objects* input_objects, off_t off)
{
Object* obj = this->get_elf_object_for_member(off, input_objects);
if (obj == NULL)
return;
Read_symbols_data* sd = new Read_symbols_data;
obj->read_symbols(sd);
Archive_member member(obj, sd);
this->members_[off] = member;
}
// Select members from the archive and add them to the link. We walk
// through the elements in the archive map, and look each one up in
// the symbol table. If it exists as a strong undefined symbol, we
// pull in the corresponding element. We have to do this in a loop,
// since pulling in one element may create new undefined symbols which
// may be satisfied by other objects in the archive.
void
Archive::add_symbols(Symbol_table* symtab, Layout* layout,
Input_objects* input_objects, Mapfile* mapfile)
{
++Archive::total_archives;
if (this->input_file_->options().whole_archive())
return this->include_all_members(symtab, layout, input_objects,
mapfile);
Archive::total_members += this->num_members_;
input_objects->archive_start(this);
const size_t armap_size = this->armap_.size();
// This is a quick optimization, since we usually see many symbols
// in a row with the same offset. last_seen_offset holds the last
// offset we saw that was present in the seen_offsets_ set.
off_t last_seen_offset = -1;
// Track which symbols in the symbol table we've already found to be
// defined.
bool added_new_object;
do
{
added_new_object = false;
for (size_t i = 0; i < armap_size; ++i)
{
if (this->armap_checked_[i])
continue;
if (this->armap_[i].file_offset == last_seen_offset)
{
this->armap_checked_[i] = true;
continue;
}
if (this->seen_offsets_.find(this->armap_[i].file_offset)
!= this->seen_offsets_.end())
{
this->armap_checked_[i] = true;
last_seen_offset = this->armap_[i].file_offset;
continue;
}
const char* sym_name = (this->armap_names_.data()
+ this->armap_[i].name_offset);
Symbol* sym = symtab->lookup(sym_name);
if (sym == NULL)
{
// Check whether the symbol was named in a -u option.
if (!parameters->options().is_undefined(sym_name))
continue;
}
else if (!sym->is_undefined())
{
this->armap_checked_[i] = true;
continue;
}
else if (sym->binding() == elfcpp::STB_WEAK)
continue;
// We want to include this object in the link.
last_seen_offset = this->armap_[i].file_offset;
this->seen_offsets_.insert(last_seen_offset);
this->armap_checked_[i] = true;
std::string why;
if (sym == NULL)
{
why = "-u ";
why += sym_name;
}
this->include_member(symtab, layout, input_objects,
last_seen_offset, mapfile, sym, why.c_str());
added_new_object = true;
}
}
while (added_new_object);
input_objects->archive_stop(this);
}
// Include all the archive members in the link. This is for --whole-archive.
void
@ -502,11 +679,29 @@ Archive::include_all_members(Symbol_table* symtab, Layout* layout,
{
input_objects->archive_start(this);
for (Archive::const_iterator p = this->begin();
p != this->end();
++p)
this->include_member(symtab, layout, input_objects, p->off,
mapfile, NULL, "--whole-archive");
if (this->members_.size() > 0)
{
std::map<off_t, Archive_member>::const_iterator p;
for (p = this->members_.begin();
p != this->members_.end();
++p)
{
this->include_member(symtab, layout, input_objects, p->first,
mapfile, NULL, "--whole-archive");
++Archive::total_members;
}
}
else
{
for (Archive::const_iterator p = this->begin();
p != this->end();
++p)
{
this->include_member(symtab, layout, input_objects, p->off,
mapfile, NULL, "--whole-archive");
++Archive::total_members;
}
}
input_objects->archive_stop(this);
}
@ -533,106 +728,30 @@ Archive::include_member(Symbol_table* symtab, Layout* layout,
Input_objects* input_objects, off_t off,
Mapfile* mapfile, Symbol* sym, const char* why)
{
std::string n;
off_t nested_off;
this->read_header(off, false, &n, &nested_off);
++Archive::total_members_loaded;
std::map<off_t, Archive_member>::const_iterator p = this->members_.find(off);
if (p != this->members_.end())
{
Object *obj = p->second.obj_;
Read_symbols_data *sd = p->second.sd_;
if (mapfile != NULL)
mapfile->report_include_archive_member(obj->name(), sym, why);
if (input_objects->add_object(obj))
{
obj->layout(symtab, layout, sd);
obj->add_symbols(symtab, sd);
}
delete sd;
return;
}
Object* obj = this->get_elf_object_for_member(off, input_objects);
if (obj == NULL)
return;
if (mapfile != NULL)
mapfile->report_include_archive_member(this, n, sym, why);
Input_file* input_file;
off_t memoff;
if (!this->is_thin_archive_)
{
input_file = this->input_file_;
memoff = off + static_cast<off_t>(sizeof(Archive_header));
}
else
{
// Adjust a relative pathname so that it is relative
// to the directory containing the archive.
if (!IS_ABSOLUTE_PATH(n.c_str()))
{
const char *arch_path = this->name().c_str();
const char *basename = lbasename(arch_path);
if (basename > arch_path)
n.replace(0, 0, this->name().substr(0, basename - arch_path));
}
if (nested_off > 0)
{
// This is a member of a nested archive. Open the containing
// archive if we don't already have it open, then do a recursive
// call to include the member from that archive.
Archive* arch;
Nested_archive_table::const_iterator p =
this->nested_archives_.find(n);
if (p != this->nested_archives_.end())
arch = p->second;
else
{
Input_file_argument* input_file_arg =
new Input_file_argument(n.c_str(), false, "", false,
parameters->options());
input_file = new Input_file(input_file_arg);
if (!input_file->open(parameters->options(), *this->dirpath_,
this->task_))
return;
arch = new Archive(n, input_file, false, this->dirpath_,
this->task_);
arch->setup();
std::pair<Nested_archive_table::iterator, bool> ins =
this->nested_archives_.insert(std::make_pair(n, arch));
gold_assert(ins.second);
}
arch->include_member(symtab, layout, input_objects, nested_off,
NULL, NULL, NULL);
return;
}
// This is an external member of a thin archive. Open the
// file as a regular relocatable object file.
Input_file_argument* input_file_arg =
new Input_file_argument(n.c_str(), false, "", false,
this->input_file_->options());
input_file = new Input_file(input_file_arg);
if (!input_file->open(parameters->options(), *this->dirpath_,
this->task_))
{
return;
}
memoff = 0;
}
off_t filesize = input_file->file().filesize();
int read_size = elfcpp::Elf_sizes<64>::ehdr_size;
if (filesize - memoff < read_size)
read_size = filesize - memoff;
if (read_size < 4)
{
gold_error(_("%s: member at %zu is not an ELF object"),
this->name().c_str(), static_cast<size_t>(off));
return;
}
const unsigned char* ehdr = input_file->file().get_view(memoff, 0, read_size,
true, false);
static unsigned char elfmagic[4] =
{
elfcpp::ELFMAG0, elfcpp::ELFMAG1,
elfcpp::ELFMAG2, elfcpp::ELFMAG3
};
if (memcmp(ehdr, elfmagic, 4) != 0)
{
gold_error(_("%s: member at %zu is not an ELF object"),
this->name().c_str(), static_cast<size_t>(off));
return;
}
Object* obj = make_elf_object((std::string(this->input_file_->filename())
+ "(" + n + ")"),
input_file, memoff, ehdr, read_size);
mapfile->report_include_archive_member(obj->name(), sym, why);
if (input_objects->add_object(obj))
{
@ -646,12 +765,19 @@ Archive::include_member(Symbol_table* symtab, Layout* layout,
// FIXME: We need to close the descriptor here.
delete obj;
}
}
if (this->is_thin_archive_)
{
// Opening the file locked it. Unlock it now.
input_file->file().unlock(this->task_);
}
// Print statistical information to stderr. This is used for --stats.
void
Archive::print_stats()
{
fprintf(stderr, _("%s: archive libraries: %u\n"),
program_name, Archive::total_archives);
fprintf(stderr, _("%s: total archive members: %u\n"),
program_name, Archive::total_members);
fprintf(stderr, _("%s: loaded archive members: %u\n"),
program_name, Archive::total_members_loaded);
}
// Add_archive_symbols methods.

View File

@ -38,6 +38,8 @@ class Input_objects;
class Input_group;
class Layout;
class Symbol_table;
class Object;
class Read_symbols_data;
// This class represents an archive--generally a libNAME.a file.
// Archives have a symbol table and a list of objects.
@ -48,8 +50,9 @@ class Archive
Archive(const std::string& name, Input_file* input_file,
bool is_thin_archive, Dirsearch* dirpath, Task* task)
: name_(name), input_file_(input_file), armap_(), armap_names_(),
extended_names_(), armap_checked_(), seen_offsets_(),
is_thin_archive_(is_thin_archive), dirpath_(dirpath), task_(task)
extended_names_(), armap_checked_(), seen_offsets_(), members_(),
is_thin_archive_(is_thin_archive), nested_archives_(),
dirpath_(dirpath), task_(task), num_members_(0)
{ }
// The length of the magic string at the start of an archive.
@ -76,7 +79,7 @@ class Archive
// Set up the archive: read the symbol map.
void
setup();
setup(Input_objects*);
// Get a reference to the underlying file.
File_read&
@ -131,6 +134,10 @@ class Archive
void
add_symbols(Symbol_table*, Layout*, Input_objects*, Mapfile*);
// Dump statistical information to stderr.
static void
print_stats();
// Return the number of members in the archive.
size_t
count_members();
@ -141,6 +148,13 @@ class Archive
struct Archive_header;
// Total number of archives seen.
static unsigned int total_archives;
// Total number of archive members seen.
static unsigned int total_members;
// Number of archive members loaded.
static unsigned int total_members_loaded;
// Get a view into the underlying file.
const unsigned char*
get_view(off_t start, section_size_type size, bool aligned, bool cache)
@ -162,6 +176,30 @@ class Archive
interpret_header(const Archive_header* hdr, off_t off, std::string* pname,
off_t* nested_off) const;
// Get the file and offset for an archive member, which may be an
// external member of a thin archive. Set *INPUT_FILE to the
// file containing the actual member, *MEMOFF to the offset
// within that file (0 if not a nested archive), and *MEMBER_NAME
// to the name of the archive member. Return TRUE on success.
bool
get_file_and_offset(off_t off, Input_objects* input_objects,
Input_file** input_file, off_t* memoff,
std::string* member_name);
// Return an ELF object for the member at offset OFF. Set *MEMBER_NAME to
// the name of the member.
Object*
get_elf_object_for_member(off_t off, Input_objects* input_objects);
// Read the symbols from all the archive members in the link.
void
read_all_symbols(Input_objects* input_objects);
// Read the symbols from an archive member in the link. OFF is the file
// offset of the member header.
void
read_symbols(Input_objects* input_objects, off_t off);
// Include all the archive members in the link.
void
include_all_members(Symbol_table*, Layout*, Input_objects*, Mapfile*);
@ -191,6 +229,21 @@ class Archive
off_t file_offset;
};
// An entry in the archive map of offsets to members.
struct Archive_member
{
Archive_member()
: obj_(NULL), sd_(NULL)
{ }
Archive_member(Object* obj, Read_symbols_data* sd)
: obj_(obj), sd_(sd)
{ }
// The object file.
Object* obj_;
// The data to pass from read_symbols() to add_symbols().
Read_symbols_data* sd_;
};
// A simple hash code for off_t values.
class Seen_hash
{
@ -217,6 +270,8 @@ class Archive
std::vector<bool> armap_checked_;
// Track which elements have been included by offset.
Unordered_set<off_t, Seen_hash> seen_offsets_;
// Table of objects whose symbols have been pre-read.
std::map<off_t, Archive_member> members_;
// True if this is a thin archive.
const bool is_thin_archive_;
// Table of nested archives, indexed by filename.
@ -225,6 +280,8 @@ class Archive
Dirsearch* dirpath_;
// The task reading this archive.
Task *task_;
// Number of members in this archive;
unsigned int num_members_;
};
// This class is used to read an archive and pick out the desired

View File

@ -39,6 +39,7 @@
#include "dirsearch.h"
#include "workqueue.h"
#include "object.h"
#include "archive.h"
#include "symtab.h"
#include "layout.h"
@ -228,6 +229,7 @@ main(int argc, char** argv)
program_name, m.arena);
#endif
File_read::print_stats();
Archive::print_stats();
fprintf(stderr, _("%s: output file size: %lld bytes\n"),
program_name, static_cast<long long>(layout.output_file_size()));
symtab.print_stats();

View File

@ -105,8 +105,7 @@ Mapfile::advance_to_column(size_t from, size_t to)
// Report about including a member from an archive.
void
Mapfile::report_include_archive_member(const Archive* archive,
const std::string& member_name,
Mapfile::report_include_archive_member(const std::string& member_name,
const Symbol* sym, const char* why)
{
// We print a header before the list of archive members, mainly for
@ -118,13 +117,9 @@ Mapfile::report_include_archive_member(const Archive* archive,
this->printed_archive_header_ = true;
}
fprintf(this->map_file_, "%s(%s)", archive->file().filename().c_str(),
member_name.c_str());
fprintf(this->map_file_, "%s", member_name.c_str());
size_t len = (archive->file().filename().length()
+ member_name.length()
+ 2);
this->advance_to_column(len, 30);
this->advance_to_column(member_name.length(), 30);
if (sym == NULL)
fprintf(this->map_file_, "%s", why);

View File

@ -57,7 +57,7 @@ class Mapfile
// Report that we are including a member from an archive. This is
// called by the archive reading code.
void
report_include_archive_member(const Archive*, const std::string& member_name,
report_include_archive_member(const std::string& member_name,
const Symbol* sym, const char* why);
// Report allocating a common symbol.

View File

@ -689,6 +689,8 @@ class General_options
DEFINE_string(oformat, options::EXACTLY_TWO_DASHES, '\0', "elf",
N_("Set output format"), N_("[binary]"));
DEFINE_bool(preread_archive_symbols, options::TWO_DASHES, '\0', false,
N_("Preread archive symbols when multi-threaded"), NULL);
DEFINE_string(print_symbol_counts, options::TWO_DASHES, '\0', NULL,
N_("Print symbols defined and used for each input"),
N_("FILENAME"));

View File

@ -212,7 +212,7 @@ Read_symbols::do_read_symbols(Workqueue* workqueue)
Archive* arch = new Archive(this->input_argument_->file().name(),
input_file, is_thin_archive,
this->dirpath_, this);
arch->setup();
arch->setup(this->input_objects_);
// Unlock the archive so it can be used in the next task.
arch->unlock(this);