zld: add prelim way of linking dylibs

The support is minimalistic in the sense that we only support actual
dylib files and not stubs/tbds yet, and we also don't support re-exports
just yet.
This commit is contained in:
Jakub Konka 2021-05-16 16:32:27 +02:00
parent 35c694d614
commit 138cecc028
6 changed files with 386 additions and 54 deletions

View File

@ -568,6 +568,7 @@ set(ZIG_STAGE2_SOURCES
"${CMAKE_SOURCE_DIR}/src/link/MachO/Archive.zig"
"${CMAKE_SOURCE_DIR}/src/link/MachO/CodeSignature.zig"
"${CMAKE_SOURCE_DIR}/src/link/MachO/DebugSymbols.zig"
"${CMAKE_SOURCE_DIR}/src/link/MachO/Dylib.zig"
"${CMAKE_SOURCE_DIR}/src/link/MachO/Object.zig"
"${CMAKE_SOURCE_DIR}/src/link/MachO/Symbol.zig"
"${CMAKE_SOURCE_DIR}/src/link/MachO/Trie.zig"

View File

@ -548,7 +548,7 @@ fn linkWithLLD(self: *MachO, comp: *Compilation) !void {
const is_dyn_lib = self.base.options.link_mode == .Dynamic and is_lib;
const is_exe_or_dyn_lib = is_dyn_lib or self.base.options.output_mode == .Exe;
const target = self.base.options.target;
const stack_size = self.base.options.stack_size_override orelse 16777216;
const stack_size = self.base.options.stack_size_override orelse 0;
const allow_shlib_undefined = self.base.options.allow_shlib_undefined orelse !self.base.options.is_native_os;
const id_symlink_basename = "lld.id";
@ -675,22 +675,114 @@ fn linkWithLLD(self: *MachO, comp: *Compilation) !void {
zld.deinit();
}
zld.arch = target.cpu.arch;
zld.stack_size = stack_size;
var input_files = std.ArrayList([]const u8).init(self.base.allocator);
defer input_files.deinit();
// Positional arguments to the linker such as object files.
try input_files.appendSlice(self.base.options.objects);
// Positional arguments to the linker such as object files and static archives.
var positionals = std.ArrayList([]const u8).init(self.base.allocator);
defer positionals.deinit();
try positionals.appendSlice(self.base.options.objects);
for (comp.c_object_table.items()) |entry| {
try input_files.append(entry.key.status.success.object_path);
try positionals.append(entry.key.status.success.object_path);
}
if (module_obj_path) |p| {
try input_files.append(p);
try positionals.append(p);
}
try input_files.append(comp.compiler_rt_static_lib.?.full_object_path);
try positionals.append(comp.compiler_rt_static_lib.?.full_object_path);
// libc++ dep
if (self.base.options.link_libcpp) {
try input_files.append(comp.libcxxabi_static_lib.?.full_object_path);
try input_files.append(comp.libcxx_static_lib.?.full_object_path);
try positionals.append(comp.libcxxabi_static_lib.?.full_object_path);
try positionals.append(comp.libcxx_static_lib.?.full_object_path);
}
if (self.base.options.is_native_os) {}
// Shared libraries.
var shared_libs = std.ArrayList([]const u8).init(self.base.allocator);
defer {
for (shared_libs.items) |sh| {
self.base.allocator.free(sh);
}
shared_libs.deinit();
}
var search_lib_names = std.ArrayList([]const u8).init(self.base.allocator);
defer search_lib_names.deinit();
const system_libs = self.base.options.system_libs.items();
for (system_libs) |entry| {
const link_lib = entry.key;
// By this time, we depend on these libs being dynamically linked libraries and not static libraries
// (the check for that needs to be earlier), but they could be full paths to .dylib files, in which
// case we want to avoid prepending "-l".
if (Compilation.classifyFileExt(link_lib) == .shared_library) {
const path = try self.base.allocator.dupe(u8, link_lib);
try shared_libs.append(path);
continue;
}
try search_lib_names.append(link_lib);
}
for (search_lib_names.items) |l_name| {
// TODO text-based API, or .tbd files.
const l_name_ext = try std.fmt.allocPrint(self.base.allocator, "lib{s}.dylib", .{l_name});
defer self.base.allocator.free(l_name_ext);
var found = false;
if (self.base.options.syslibroot) |syslibroot| {
for (self.base.options.lib_dirs) |lib_dir| {
const path = try fs.path.join(self.base.allocator, &[_][]const u8{
syslibroot,
lib_dir,
l_name_ext,
});
const tmp = fs.cwd().openFile(path, .{}) catch |err| switch (err) {
error.FileNotFound => {
self.base.allocator.free(path);
continue;
},
else => |e| return e,
};
defer tmp.close();
try shared_libs.append(path);
found = true;
break;
}
}
for (self.base.options.lib_dirs) |lib_dir| {
const path = try fs.path.join(self.base.allocator, &[_][]const u8{ lib_dir, l_name_ext });
const tmp = fs.cwd().openFile(path, .{}) catch |err| switch (err) {
error.FileNotFound => {
self.base.allocator.free(path);
continue;
},
else => |e| return e,
};
defer tmp.close();
try shared_libs.append(path);
found = true;
break;
}
if (!found) {
log.warn("library '-l{s}' not found", .{l_name});
log.warn("searched paths:", .{});
if (self.base.options.syslibroot) |syslibroot| {
for (self.base.options.lib_dirs) |lib_dir| {
log.warn(" {s}/{s}", .{ syslibroot, lib_dir });
}
}
for (self.base.options.lib_dirs) |lib_dir| {
log.warn(" {s}/", .{lib_dir});
}
}
}
if (self.base.options.verbose_link) {
@ -700,17 +792,28 @@ fn linkWithLLD(self: *MachO, comp: *Compilation) !void {
try argv.append("zig");
try argv.append("ld");
try argv.appendSlice(input_files.items);
if (self.base.options.syslibroot) |syslibroot| {
try argv.append("-syslibroot");
try argv.append(syslibroot);
}
try argv.appendSlice(positionals.items);
try argv.append("-o");
try argv.append(full_out_path);
for (search_lib_names.items) |l_name| {
try argv.append(try std.fmt.allocPrint(self.base.allocator, "-l{s}", .{l_name}));
}
for (self.base.options.lib_dirs) |lib_dir| {
try argv.append(try std.fmt.allocPrint(self.base.allocator, "-L{s}", .{lib_dir}));
}
Compilation.dump_argv(argv.items);
}
try zld.link(input_files.items, full_out_path, .{
.stack_size = self.base.options.stack_size_override,
});
try zld.link(positionals.items, shared_libs.items, full_out_path);
break :outer;
}

137
src/link/MachO/Dylib.zig Normal file
View File

@ -0,0 +1,137 @@
const Dylib = @This();
const std = @import("std");
const fs = std.fs;
const log = std.log.scoped(.dylib);
const macho = std.macho;
const mem = std.mem;
const Allocator = mem.Allocator;
const Symbol = @import("Symbol.zig");
usingnamespace @import("commands.zig");
allocator: *Allocator,
arch: ?std.Target.Cpu.Arch = null,
header: ?macho.mach_header_64 = null,
file: ?fs.File = null,
name: ?[]const u8 = null,
ordinal: ?u16 = null,
load_commands: std.ArrayListUnmanaged(LoadCommand) = .{},
symtab_cmd_index: ?u16 = null,
dysymtab_cmd_index: ?u16 = null,
symbols: std.StringArrayHashMapUnmanaged(*Symbol) = .{},
pub fn init(allocator: *Allocator) Dylib {
return .{ .allocator = allocator };
}
pub fn deinit(self: *Dylib) void {
for (self.load_commands.items) |*lc| {
lc.deinit(self.allocator);
}
self.load_commands.deinit(self.allocator);
for (self.symbols.items()) |entry| {
entry.value.deinit(self.allocator);
self.allocator.destroy(entry.value);
}
self.symbols.deinit(self.allocator);
if (self.name) |name| {
self.allocator.free(name);
}
}
pub fn closeFile(self: Dylib) void {
if (self.file) |file| {
file.close();
}
}
pub fn parse(self: *Dylib) !void {
log.warn("parsing shared library '{s}'", .{self.name.?});
var reader = self.file.?.reader();
self.header = try reader.readStruct(macho.mach_header_64);
if (self.header.?.filetype != macho.MH_DYLIB) {
log.err("invalid filetype: expected 0x{x}, found 0x{x}", .{ macho.MH_DYLIB, self.header.?.filetype });
return error.MalformedDylib;
}
const this_arch: std.Target.Cpu.Arch = switch (self.header.?.cputype) {
macho.CPU_TYPE_ARM64 => .aarch64,
macho.CPU_TYPE_X86_64 => .x86_64,
else => |value| {
log.err("unsupported cpu architecture 0x{x}", .{value});
return error.UnsupportedCpuArchitecture;
},
};
if (this_arch != self.arch.?) {
log.err("mismatched cpu architecture: expected {s}, found {s}", .{ self.arch.?, this_arch });
return error.MismatchedCpuArchitecture;
}
try self.readLoadCommands(reader);
try self.parseSymbols();
}
pub fn readLoadCommands(self: *Dylib, reader: anytype) !void {
try self.load_commands.ensureCapacity(self.allocator, self.header.?.ncmds);
var i: u16 = 0;
while (i < self.header.?.ncmds) : (i += 1) {
var cmd = try LoadCommand.read(self.allocator, reader);
switch (cmd.cmd()) {
macho.LC_SYMTAB => {
self.symtab_cmd_index = i;
},
macho.LC_DYSYMTAB => {
self.dysymtab_cmd_index = i;
},
else => {
log.debug("Unknown load command detected: 0x{x}.", .{cmd.cmd()});
},
}
self.load_commands.appendAssumeCapacity(cmd);
}
}
pub fn parseSymbols(self: *Dylib) !void {
const index = self.symtab_cmd_index orelse return;
const symtab_cmd = self.load_commands.items[index].Symtab;
var symtab = try self.allocator.alloc(u8, @sizeOf(macho.nlist_64) * symtab_cmd.nsyms);
defer self.allocator.free(symtab);
_ = try self.file.?.preadAll(symtab, symtab_cmd.symoff);
const slice = @alignCast(@alignOf(macho.nlist_64), mem.bytesAsSlice(macho.nlist_64, symtab));
var strtab = try self.allocator.alloc(u8, symtab_cmd.strsize);
defer self.allocator.free(strtab);
_ = try self.file.?.preadAll(strtab, symtab_cmd.stroff);
for (slice) |sym| {
const sym_name = mem.spanZ(@ptrCast([*:0]const u8, strtab.ptr + sym.n_strx));
if (!(Symbol.isSect(sym) and Symbol.isExt(sym))) continue;
const name = try self.allocator.dupe(u8, sym_name);
const proxy = try self.allocator.create(Symbol.Proxy);
errdefer self.allocator.destroy(proxy);
proxy.* = .{
.base = .{
.@"type" = .proxy,
.name = name,
},
.dylib = self,
};
try self.symbols.putNoClobber(self.allocator, name, &proxy.base);
}
}

View File

@ -5,6 +5,7 @@ const macho = std.macho;
const mem = std.mem;
const Allocator = mem.Allocator;
const Dylib = @import("Dylib.zig");
const Object = @import("Object.zig");
pub const Type = enum {
@ -43,7 +44,7 @@ pub const Regular = struct {
/// Whether the symbol is a weak ref.
weak_ref: bool,
/// File where to locate this symbol.
/// Object file where to locate this symbol.
file: *Object,
/// Debug stab if defined.
@ -78,8 +79,8 @@ pub const Regular = struct {
pub const Proxy = struct {
base: Symbol,
/// Dylib ordinal.
dylib: u16,
/// Dylib where to locate this symbol.
dylib: ?*Dylib = null,
pub const base_type: Symbol.Type = .proxy;
};

View File

@ -15,6 +15,7 @@ const reloc = @import("reloc.zig");
const Allocator = mem.Allocator;
const Archive = @import("Archive.zig");
const CodeSignature = @import("CodeSignature.zig");
const Dylib = @import("Dylib.zig");
const Object = @import("Object.zig");
const Symbol = @import("Symbol.zig");
const Trie = @import("Trie.zig");
@ -35,6 +36,7 @@ stack_size: u64 = 0,
objects: std.ArrayListUnmanaged(*Object) = .{},
archives: std.ArrayListUnmanaged(*Archive) = .{},
dylibs: std.ArrayListUnmanaged(*Dylib) = .{},
load_commands: std.ArrayListUnmanaged(LoadCommand) = .{},
@ -151,10 +153,18 @@ pub fn deinit(self: *Zld) void {
}
self.archives.deinit(self.allocator);
for (self.dylibs.items) |dylib| {
dylib.deinit();
self.allocator.destroy(dylib);
}
self.dylibs.deinit(self.allocator);
self.mappings.deinit(self.allocator);
self.unhandled_sections.deinit(self.allocator);
self.globals.deinit(self.allocator);
self.imports.deinit(self.allocator);
self.unresolved.deinit(self.allocator);
self.strtab.deinit(self.allocator);
{
@ -176,11 +186,7 @@ pub fn closeFiles(self: Zld) void {
if (self.file) |f| f.close();
}
const LinkArgs = struct {
stack_size: ?u64 = null,
};
pub fn link(self: *Zld, files: []const []const u8, out_path: []const u8, args: LinkArgs) !void {
pub fn link(self: *Zld, files: []const []const u8, shared_libs: []const []const u8, out_path: []const u8) !void {
if (files.len == 0) return error.NoInputFiles;
if (out_path.len == 0) return error.EmptyOutputPath;
@ -214,10 +220,10 @@ pub fn link(self: *Zld, files: []const []const u8, out_path: []const u8, args: L
.read = true,
.mode = if (std.Target.current.os.tag == .windows) 0 else 0o777,
});
self.stack_size = args.stack_size orelse 0;
try self.populateMetadata();
try self.parseInputFiles(files);
try self.parseDylibs(shared_libs);
try self.resolveSymbols();
try self.resolveStubsAndGotEntries();
try self.updateMetadata();
@ -315,6 +321,48 @@ fn parseInputFiles(self: *Zld, files: []const []const u8) !void {
}
}
fn parseDylibs(self: *Zld, shared_libs: []const []const u8) !void {
for (shared_libs) |lib| {
const dylib = try self.allocator.create(Dylib);
errdefer self.allocator.destroy(dylib);
dylib.* = Dylib.init(self.allocator);
dylib.arch = self.arch.?;
dylib.name = try self.allocator.dupe(u8, lib);
dylib.file = try fs.cwd().openFile(lib, .{});
const ordinal = @intCast(u16, self.dylibs.items.len);
dylib.ordinal = ordinal + 2; // TODO +2 since 1 is reserved for libSystem
// TODO Defer parsing of the dylibs until they are actually needed
try dylib.parse();
try self.dylibs.append(self.allocator, dylib);
// Add LC_LOAD_DYLIB command
const cmdsize = @intCast(u32, mem.alignForwardGeneric(
u64,
@sizeOf(macho.dylib_command) + dylib.name.?.len,
@sizeOf(u64),
));
// TODO Read the min version from the dylib itself.
const min_version = 0x0;
var dylib_cmd = emptyGenericCommandWithData(macho.dylib_command{
.cmd = macho.LC_LOAD_DYLIB,
.cmdsize = cmdsize,
.dylib = .{
.name = @sizeOf(macho.dylib_command),
.timestamp = 2, // TODO parse from the dylib.
.current_version = min_version,
.compatibility_version = min_version,
},
});
dylib_cmd.data = try self.allocator.alloc(u8, cmdsize - dylib_cmd.inner.dylib.name);
mem.set(u8, dylib_cmd.data, 0);
mem.copy(u8, dylib_cmd.data, dylib.name.?);
try self.load_commands.append(self.allocator, .{ .Dylib = dylib_cmd });
}
}
fn mapAndUpdateSections(
self: *Zld,
object_id: u16,
@ -1398,35 +1446,51 @@ fn resolveSymbols(self: *Zld) !void {
// Third pass, resolve symbols in dynamic libraries.
// TODO Implement libSystem as a hard-coded library, or ship with
// a libSystem.B.tbd definition file?
try self.imports.ensureCapacity(self.allocator, self.unresolved.count());
var unresolved = std.ArrayList(*Symbol).init(self.allocator);
defer unresolved.deinit();
try unresolved.ensureCapacity(self.unresolved.count());
for (self.unresolved.items()) |entry| {
const proxy = try self.allocator.create(Symbol.Proxy);
errdefer self.allocator.destroy(proxy);
proxy.* = .{
.base = .{
.@"type" = .proxy,
.name = try self.allocator.dupe(u8, entry.key),
},
.dylib = 0,
};
self.imports.putAssumeCapacityNoClobber(proxy.base.name, &proxy.base);
entry.value.alias = &proxy.base;
unresolved.appendAssumeCapacity(entry.value);
}
self.unresolved.clearAndFree(self.allocator);
// If there are any undefs left, flag an error.
if (self.unresolved.count() > 0) {
for (self.unresolved.items()) |entry| {
log.err("undefined reference to symbol '{s}'", .{entry.key});
log.err(" | referenced in {s}", .{
entry.value.cast(Symbol.Unresolved).?.file.name.?,
});
var has_undefined = false;
while (unresolved.popOrNull()) |undef| {
var found = false;
for (self.dylibs.items) |dylib| {
const proxy = dylib.symbols.get(undef.name) orelse continue;
try self.imports.putNoClobber(self.allocator, proxy.name, proxy);
undef.alias = proxy;
found = true;
}
if (!found) {
// TODO we currently hardcode all unresolved symbols to libSystem
const proxy = try self.allocator.create(Symbol.Proxy);
errdefer self.allocator.destroy(proxy);
proxy.* = .{
.base = .{
.@"type" = .proxy,
.name = try self.allocator.dupe(u8, undef.name),
},
.dylib = null, // TODO null means libSystem
};
try self.imports.putNoClobber(self.allocator, proxy.base.name, &proxy.base);
undef.alias = &proxy.base;
// log.err("undefined reference to symbol '{s}'", .{undef.name});
// log.err(" | referenced in {s}", .{
// undef.cast(Symbol.Unresolved).?.file.name.?,
// });
// has_undefined = true;
}
return error.UndefinedSymbolReference;
}
if (has_undefined) return error.UndefinedSymbolReference;
// Finally put dyld_stub_binder as an Import
const dyld_stub_binder = try self.allocator.create(Symbol.Proxy);
errdefer self.allocator.destroy(dyld_stub_binder);
@ -1436,7 +1500,7 @@ fn resolveSymbols(self: *Zld) !void {
.@"type" = .proxy,
.name = try self.allocator.dupe(u8, "dyld_stub_binder"),
},
.dylib = 0,
.dylib = null, // TODO null means libSystem
};
try self.imports.putNoClobber(
@ -2303,7 +2367,10 @@ fn writeBindInfoTable(self: *Zld) !void {
for (self.got_entries.items) |sym| {
if (sym.cast(Symbol.Proxy)) |proxy| {
const dylib_ordinal = proxy.dylib + 1;
const dylib_ordinal = ordinal: {
const dylib = proxy.dylib orelse break :ordinal 1; // TODO embedded libSystem
break :ordinal dylib.ordinal.?;
};
try pointers.append(.{
.offset = base_offset + proxy.base.got_index.? * @sizeOf(u64),
.segment_id = segment_id,
@ -2322,7 +2389,10 @@ fn writeBindInfoTable(self: *Zld) !void {
const sym = self.imports.get("__tlv_bootstrap") orelse unreachable;
const proxy = sym.cast(Symbol.Proxy) orelse unreachable;
const dylib_ordinal = proxy.dylib + 1;
const dylib_ordinal = ordinal: {
const dylib = proxy.dylib orelse break :ordinal 1; // TODO embedded libSystem
break :ordinal dylib.ordinal.?;
};
try pointers.append(.{
.offset = base_offset,
@ -2364,7 +2434,11 @@ fn writeLazyBindInfoTable(self: *Zld) !void {
for (self.stubs.items) |sym| {
const proxy = sym.cast(Symbol.Proxy) orelse unreachable;
const dylib_ordinal = proxy.dylib + 1;
const dylib_ordinal = ordinal: {
const dylib = proxy.dylib orelse break :ordinal 1; // TODO embedded libSystem
break :ordinal dylib.ordinal.?;
};
pointers.appendAssumeCapacity(.{
.offset = base_offset + sym.stubs_index.? * @sizeOf(u64),
.segment_id = segment_id,

View File

@ -38,7 +38,9 @@ pub const LoadCommand = union(enum) {
macho.LC_SEGMENT_64 => LoadCommand{
.Segment = try SegmentCommand.read(allocator, stream.reader()),
},
macho.LC_DYLD_INFO, macho.LC_DYLD_INFO_ONLY => LoadCommand{
macho.LC_DYLD_INFO,
macho.LC_DYLD_INFO_ONLY,
=> LoadCommand{
.DyldInfoOnly = try stream.reader().readStruct(macho.dyld_info_command),
},
macho.LC_SYMTAB => LoadCommand{
@ -47,16 +49,27 @@ pub const LoadCommand = union(enum) {
macho.LC_DYSYMTAB => LoadCommand{
.Dysymtab = try stream.reader().readStruct(macho.dysymtab_command),
},
macho.LC_ID_DYLINKER, macho.LC_LOAD_DYLINKER, macho.LC_DYLD_ENVIRONMENT => LoadCommand{
macho.LC_ID_DYLINKER,
macho.LC_LOAD_DYLINKER,
macho.LC_DYLD_ENVIRONMENT,
=> LoadCommand{
.Dylinker = try GenericCommandWithData(macho.dylinker_command).read(allocator, stream.reader()),
},
macho.LC_ID_DYLIB, macho.LC_LOAD_WEAK_DYLIB, macho.LC_LOAD_DYLIB, macho.LC_REEXPORT_DYLIB => LoadCommand{
macho.LC_ID_DYLIB,
macho.LC_LOAD_WEAK_DYLIB,
macho.LC_LOAD_DYLIB,
macho.LC_REEXPORT_DYLIB,
=> LoadCommand{
.Dylib = try GenericCommandWithData(macho.dylib_command).read(allocator, stream.reader()),
},
macho.LC_MAIN => LoadCommand{
.Main = try stream.reader().readStruct(macho.entry_point_command),
},
macho.LC_VERSION_MIN_MACOSX, macho.LC_VERSION_MIN_IPHONEOS, macho.LC_VERSION_MIN_WATCHOS, macho.LC_VERSION_MIN_TVOS => LoadCommand{
macho.LC_VERSION_MIN_MACOSX,
macho.LC_VERSION_MIN_IPHONEOS,
macho.LC_VERSION_MIN_WATCHOS,
macho.LC_VERSION_MIN_TVOS,
=> LoadCommand{
.VersionMin = try stream.reader().readStruct(macho.version_min_command),
},
macho.LC_SOURCE_VERSION => LoadCommand{
@ -65,7 +78,10 @@ pub const LoadCommand = union(enum) {
macho.LC_UUID => LoadCommand{
.Uuid = try stream.reader().readStruct(macho.uuid_command),
},
macho.LC_FUNCTION_STARTS, macho.LC_DATA_IN_CODE, macho.LC_CODE_SIGNATURE => LoadCommand{
macho.LC_FUNCTION_STARTS,
macho.LC_DATA_IN_CODE,
macho.LC_CODE_SIGNATURE,
=> LoadCommand{
.LinkeditData = try stream.reader().readStruct(macho.linkedit_data_command),
},
else => LoadCommand{