Merge pull request #17284 from ziglang/elf-tests

elf: link against musl libc, add ELF test harness, dynamically allocate misc SHF_ALLOC sections
This commit is contained in:
Jakub Konka 2023-09-27 07:39:58 +02:00 committed by GitHub
commit 7a43f45908
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 302 additions and 33 deletions

View File

@ -465,9 +465,10 @@ pub const File = struct {
.Exe => {},
}
switch (base.tag) {
.coff, .elf, .macho, .plan9, .wasm => if (base.file) |f| {
.elf => if (base.file) |f| {
if (build_options.only_c) unreachable;
if (base.intermediary_basename != null) {
const use_lld = build_options.have_llvm and base.options.use_lld;
if (base.intermediary_basename != null and use_lld) {
// The file we have open is not the final file that we want to
// make executable, so we don't have to close it.
return;
@ -480,6 +481,22 @@ pub const File = struct {
.linux => std.os.ptrace(std.os.linux.PTRACE.DETACH, pid, 0, 0) catch |err| {
log.warn("ptrace failure: {s}", .{@errorName(err)});
},
else => return error.HotSwapUnavailableOnHostOperatingSystem,
}
}
},
.coff, .macho, .plan9, .wasm => if (base.file) |f| {
if (build_options.only_c) unreachable;
if (base.intermediary_basename != null) {
// The file we have open is not the final file that we want to
// make executable, so we don't have to close it.
return;
}
f.close();
base.file = null;
if (base.child_pid) |pid| {
switch (builtin.os.tag) {
.macos => base.cast(MachO).?.ptraceDetach(pid) catch |err| {
log.warn("detaching failed with error: {s}", .{@errorName(err)});
},

View File

@ -254,7 +254,7 @@ pub fn createEmpty(gpa: Allocator, options: link.Options) !*Elf {
.default_sym_version = default_sym_version,
};
const use_llvm = options.use_llvm;
if (use_llvm) {
if (use_llvm and options.module != null) {
self.llvm_object = try LlvmObject.create(gpa, options);
}
@ -409,16 +409,24 @@ fn findFreeSpace(self: *Elf, object_size: u64, min_alignment: u64) u64 {
}
const AllocateSegmentOpts = struct {
addr: u64, // TODO find free VM space
size: u64,
alignment: u64,
addr: ?u64 = null, // TODO find free VM space
flags: u32 = elf.PF_R,
};
fn allocateSegment(self: *Elf, opts: AllocateSegmentOpts) error{OutOfMemory}!u16 {
pub fn allocateSegment(self: *Elf, opts: AllocateSegmentOpts) error{OutOfMemory}!u16 {
const index = @as(u16, @intCast(self.phdrs.items.len));
try self.phdrs.ensureUnusedCapacity(self.base.allocator, 1);
const off = self.findFreeSpace(opts.size, opts.alignment);
// Memory is always allocated in sequence.
// TODO is this correct? Or should we implement something similar to `findFreeSpace`?
// How would that impact HCS?
const addr = opts.addr orelse blk: {
assert(self.phdr_table_load_index != null);
const phdr = &self.phdrs.items[index - 1];
break :blk mem.alignForward(u64, phdr.p_vaddr + phdr.p_memsz, opts.alignment);
};
log.debug("allocating phdr({d})({c}{c}{c}) from 0x{x} to 0x{x} (0x{x} - 0x{x})", .{
index,
if (opts.flags & elf.PF_R != 0) @as(u8, 'R') else '_',
@ -426,15 +434,15 @@ fn allocateSegment(self: *Elf, opts: AllocateSegmentOpts) error{OutOfMemory}!u16
if (opts.flags & elf.PF_X != 0) @as(u8, 'X') else '_',
off,
off + opts.size,
opts.addr,
opts.addr + opts.size,
addr,
addr + opts.size,
});
self.phdrs.appendAssumeCapacity(.{
.p_type = elf.PT_LOAD,
.p_offset = off,
.p_filesz = opts.size,
.p_vaddr = opts.addr,
.p_paddr = opts.addr,
.p_vaddr = addr,
.p_paddr = addr,
.p_memsz = opts.size,
.p_align = opts.alignment,
.p_flags = opts.flags,
@ -446,12 +454,12 @@ fn allocateSegment(self: *Elf, opts: AllocateSegmentOpts) error{OutOfMemory}!u16
const AllocateAllocSectionOpts = struct {
name: [:0]const u8,
phdr_index: u16,
alignment: u16 = 1,
flags: u16 = elf.SHF_ALLOC,
alignment: u64 = 1,
flags: u64 = elf.SHF_ALLOC,
type: u32 = elf.SHT_PROGBITS,
};
fn allocateAllocSection(self: *Elf, opts: AllocateAllocSectionOpts) error{OutOfMemory}!u16 {
pub fn allocateAllocSection(self: *Elf, opts: AllocateAllocSectionOpts) error{OutOfMemory}!u16 {
const gpa = self.base.allocator;
const phdr = &self.phdrs.items[opts.phdr_index];
const index = @as(u16, @intCast(self.shdrs.items.len));
@ -622,6 +630,7 @@ pub fn populateMissingMetadata(self: *Elf) !void {
});
const phdr = &self.phdrs.items[self.phdr_load_zerofill_index.?];
phdr.p_offset = self.phdrs.items[self.phdr_load_rw_index.?].p_offset; // .bss overlaps .data
phdr.p_memsz = 1024;
}
if (self.shstrtab_section_index == null) {
@ -965,6 +974,7 @@ pub fn flushModule(self: *Elf, comp: *Compilation, prog_node: *std.Progress.Node
defer arena_allocator.deinit();
const arena = arena_allocator.allocator();
const target = self.base.options.target;
const directory = self.base.options.emit.?.directory; // Just an alias to make it shorter to type.
const full_out_path = try directory.join(arena, &[_][]const u8{self.base.options.emit.?.sub_path});
@ -993,19 +1003,77 @@ pub fn flushModule(self: *Elf, comp: *Compilation, prog_node: *std.Progress.Node
try positionals.append(.{ .path = key.status.success.object_path });
}
// csu prelude
var csu = try CsuObjects.init(arena, self.base.options, comp);
if (csu.crt0) |v| try positionals.append(.{ .path = v });
if (csu.crti) |v| try positionals.append(.{ .path = v });
if (csu.crtbegin) |v| try positionals.append(.{ .path = v });
for (positionals.items) |obj| {
const in_file = try std.fs.cwd().openFile(obj.path, .{});
defer in_file.close();
var parse_ctx: ParseErrorCtx = .{ .detected_cpu_arch = undefined };
self.parsePositional(in_file, obj.path, obj.must_link, &parse_ctx) catch |err|
try self.handleAndReportParseError(obj.path, err, &parse_ctx);
}
var system_libs = std.ArrayList(SystemLib).init(arena);
// libc dep
self.error_flags.missing_libc = false;
if (self.base.options.link_libc) {
if (self.base.options.libc_installation != null) {
@panic("TODO explicit libc_installation");
} else if (target.isGnuLibC()) {
try system_libs.ensureUnusedCapacity(glibc.libs.len + 1);
for (glibc.libs) |lib| {
const lib_path = try std.fmt.allocPrint(arena, "{s}{c}lib{s}.so.{d}", .{
comp.glibc_so_files.?.dir_path, fs.path.sep, lib.name, lib.sover,
});
system_libs.appendAssumeCapacity(.{ .path = lib_path });
}
system_libs.appendAssumeCapacity(.{
.path = try comp.get_libc_crt_file(arena, "libc_nonshared.a"),
});
} else if (target.isMusl()) {
const path = try comp.get_libc_crt_file(arena, switch (self.base.options.link_mode) {
.Static => "libc.a",
.Dynamic => "libc.so",
});
try system_libs.append(.{ .path = path });
} else {
self.error_flags.missing_libc = true;
}
}
for (system_libs.items) |lib| {
const in_file = try std.fs.cwd().openFile(lib.path, .{});
defer in_file.close();
var parse_ctx: ParseErrorCtx = .{ .detected_cpu_arch = undefined };
self.parseLibrary(in_file, lib, false, &parse_ctx) catch |err|
try self.handleAndReportParseError(lib.path, err, &parse_ctx);
}
// Finally, as the last input objects we add compiler_rt and CSU postlude (if any).
positionals.clearRetainingCapacity();
// compiler-rt. Since compiler_rt exports symbols like `memset`, it needs
// to be after the shared libraries, so they are picked up from the shared
// libraries, not libcompiler_rt.
const compiler_rt_path: ?[]const u8 = blk: {
if (comp.compiler_rt_lib) |x| break :blk x.full_object_path;
if (comp.compiler_rt_obj) |x| break :blk x.full_object_path;
break :blk null;
};
if (compiler_rt_path) |path| {
try positionals.append(.{ .path = path });
}
if (compiler_rt_path) |path| try positionals.append(.{ .path = path });
// csu postlude
if (csu.crtend) |v| try positionals.append(.{ .path = v });
if (csu.crtn) |v| try positionals.append(.{ .path = v });
for (positionals.items) |obj| {
const in_file = try std.fs.cwd().openFile(obj.path, .{});
defer in_file.close();
var parse_ctx: ParseErrorCtx = .{ .detected_cpu_arch = undefined };
self.parsePositional(in_file, obj.path, obj.must_link, &parse_ctx) catch |err|
try self.handleAndReportParseError(obj.path, err, &parse_ctx);
@ -1347,28 +1415,22 @@ fn parsePositional(
if (Object.isObject(in_file)) {
try self.parseObject(in_file, path, ctx);
} else {
try self.parseLibrary(in_file, path, .{
.path = null,
.needed = false,
.weak = false,
}, must_link, ctx);
try self.parseLibrary(in_file, .{ .path = path }, must_link, ctx);
}
}
fn parseLibrary(
self: *Elf,
in_file: std.fs.File,
path: []const u8,
lib: link.SystemLib,
lib: SystemLib,
must_link: bool,
ctx: *ParseErrorCtx,
) ParseError!void {
const tracy = trace(@src());
defer tracy.end();
_ = lib;
if (Archive.isArchive(in_file)) {
try self.parseArchive(in_file, path, must_link, ctx);
try self.parseArchive(in_file, lib.path, must_link, ctx);
} else return error.UnknownFileType;
}
@ -4150,6 +4212,11 @@ pub const null_sym = elf.Elf64_Sym{
.st_size = 0,
};
const SystemLib = struct {
needed: bool = false,
path: []const u8,
};
const std = @import("std");
const build_options = @import("build_options");
const builtin = @import("builtin");

View File

@ -136,7 +136,7 @@ fn initAtoms(self: *Object, elf_file: *Elf) !void {
try self.comdat_groups.append(elf_file.base.allocator, comdat_group_index);
},
elf.SHT_SYMTAB_SHNDX => @panic("TODO"),
elf.SHT_SYMTAB_SHNDX => @panic("TODO SHT_SYMTAB_SHNDX"),
elf.SHT_NULL,
elf.SHT_REL,
@ -166,14 +166,20 @@ fn initAtoms(self: *Object, elf_file: *Elf) !void {
};
}
fn addAtom(self: *Object, shdr: elf.Elf64_Shdr, shndx: u16, name: [:0]const u8, elf_file: *Elf) !void {
fn addAtom(
self: *Object,
shdr: elf.Elf64_Shdr,
shndx: u16,
name: [:0]const u8,
elf_file: *Elf,
) error{ OutOfMemory, Overflow }!void {
const atom_index = try elf_file.addAtom();
const atom = elf_file.atom(atom_index).?;
atom.atom_index = atom_index;
atom.name_offset = try elf_file.strtab.insert(elf_file.base.allocator, name);
atom.file_index = self.index;
atom.input_section_index = shndx;
atom.output_section_index = self.getOutputSectionIndex(elf_file, shdr);
atom.output_section_index = try self.getOutputSectionIndex(elf_file, shdr);
atom.alive = true;
self.atoms.items[shndx] = atom_index;
@ -188,7 +194,7 @@ fn addAtom(self: *Object, shdr: elf.Elf64_Shdr, shndx: u16, name: [:0]const u8,
}
}
fn getOutputSectionIndex(self: *Object, elf_file: *Elf, shdr: elf.Elf64_Shdr) u16 {
fn getOutputSectionIndex(self: *Object, elf_file: *Elf, shdr: elf.Elf64_Shdr) error{OutOfMemory}!u16 {
const name = blk: {
const name = self.strings.getAssumeExists(shdr.sh_name);
// if (shdr.sh_flags & elf.SHF_MERGE != 0) break :blk name;
@ -223,10 +229,32 @@ fn getOutputSectionIndex(self: *Object, elf_file: *Elf, shdr: elf.Elf64_Shdr) u1
else => flags,
};
};
_ = flags;
const out_shndx = elf_file.sectionByName(name) orelse {
log.err("{}: output section {s} not found", .{ self.fmtPath(), name });
@panic("TODO: missing output section!");
const out_shndx = elf_file.sectionByName(name) orelse blk: {
const is_alloc = flags & elf.SHF_ALLOC != 0;
const is_write = flags & elf.SHF_WRITE != 0;
const is_exec = flags & elf.SHF_EXECINSTR != 0;
const is_tls = flags & elf.SHF_TLS != 0;
if (!is_alloc or is_tls) {
log.err("{}: output section {s} not found", .{ self.fmtPath(), name });
@panic("TODO: missing output section!");
}
var phdr_flags: u32 = elf.PF_R;
if (is_write) phdr_flags |= elf.PF_W;
if (is_exec) phdr_flags |= elf.PF_X;
const phdr_index = try elf_file.allocateSegment(.{
.size = Elf.padToIdeal(shdr.sh_size),
.alignment = if (is_tls) shdr.sh_addralign else elf_file.page_size,
.flags = phdr_flags,
});
const shndx = try elf_file.allocateAllocSection(.{
.name = name,
.phdr_index = phdr_index,
.alignment = shdr.sh_addralign,
.flags = flags,
.type = @"type",
});
try elf_file.last_atom_and_free_list_table.putNoClobber(elf_file.base.allocator, shndx, .{});
break :blk shndx;
};
return out_shndx;
}

View File

@ -29,6 +29,12 @@ pub const cases = [_]Case{
.import = @import("link/glibc_compat/build.zig"),
},
// Elf Cases
.{
.build_root = "test/link",
.import = @import("link/elf.zig"),
},
// WASM Cases
// https://github.com/ziglang/zig/issues/16938
//.{

151
test/link/elf.zig Normal file
View File

@ -0,0 +1,151 @@
//! Here we test our ELF linker for correctness and functionality.
//! Currently, we support linking x86_64 Linux, but in the future we
//! will progressively relax those to exercise more combinations.
pub fn build(b: *Build) void {
const elf_step = b.step("test-elf", "Run ELF tests");
b.default_step = elf_step;
const musl_target = CrossTarget{
.cpu_arch = .x86_64, // TODO relax this once ELF linker is able to handle other archs
.os_tag = .linux,
.abi = .musl,
};
// Exercise linker with self-hosted backend (no LLVM)
elf_step.dependOn(testLinkingZig(b, .{ .use_llvm = false }));
// Exercise linker with LLVM backend
elf_step.dependOn(testEmptyObject(b, .{ .target = musl_target }));
elf_step.dependOn(testLinkingC(b, .{ .target = musl_target }));
elf_step.dependOn(testLinkingZig(b, .{}));
}
fn testEmptyObject(b: *Build, opts: Options) *Step {
const test_step = addTestStep(b, "empty-object", opts);
const exe = addExecutable(b, opts);
addCSourceBytes(exe, "int main() { return 0; }");
addCSourceBytes(exe, "");
exe.is_linking_libc = true;
const run = addRunArtifact(exe);
run.expectExitCode(0);
test_step.dependOn(&run.step);
return test_step;
}
fn testLinkingC(b: *Build, opts: Options) *Step {
const test_step = addTestStep(b, "linking-c-static", opts);
const exe = addExecutable(b, opts);
addCSourceBytes(exe,
\\#include <stdio.h>
\\int main() {
\\ printf("Hello World!\n");
\\ return 0;
\\}
);
exe.is_linking_libc = true;
const run = addRunArtifact(exe);
run.expectStdOutEqual("Hello World!\n");
test_step.dependOn(&run.step);
const check = exe.checkObject();
check.checkStart();
check.checkExact("header");
check.checkExact("type EXEC");
check.checkStart();
check.checkExact("section headers");
check.checkNotPresent("name .dynamic");
test_step.dependOn(&check.step);
return test_step;
}
fn testLinkingZig(b: *Build, opts: Options) *Step {
const test_step = addTestStep(b, "linking-zig-static", opts);
const exe = addExecutable(b, opts);
addZigSourceBytes(exe,
\\pub fn main() void {
\\ @import("std").debug.print("Hello World!\n", .{});
\\}
);
const run = addRunArtifact(exe);
run.expectStdErrEqual("Hello World!\n");
test_step.dependOn(&run.step);
const check = exe.checkObject();
check.checkStart();
check.checkExact("header");
check.checkExact("type EXEC");
check.checkStart();
check.checkExact("section headers");
check.checkNotPresent("name .dynamic");
test_step.dependOn(&check.step);
return test_step;
}
const Options = struct {
target: CrossTarget = .{ .cpu_arch = .x86_64, .os_tag = .linux },
optimize: std.builtin.OptimizeMode = .Debug,
use_llvm: bool = true,
};
fn addTestStep(b: *Build, comptime prefix: []const u8, opts: Options) *Step {
const target = opts.target.zigTriple(b.allocator) catch @panic("OOM");
const optimize = @tagName(opts.optimize);
const use_llvm = if (opts.use_llvm) "llvm" else "no-llvm";
const name = std.fmt.allocPrint(b.allocator, "test-elf-" ++ prefix ++ "-{s}-{s}-{s}", .{
target,
optimize,
use_llvm,
}) catch @panic("OOM");
return b.step(name, "");
}
fn addExecutable(b: *Build, opts: Options) *Compile {
return b.addExecutable(.{
.name = "test",
.target = opts.target,
.optimize = opts.optimize,
.single_threaded = true, // TODO temp until we teach linker how to handle TLS
.use_llvm = opts.use_llvm,
.use_lld = false,
});
}
fn addRunArtifact(comp: *Compile) *Run {
const b = comp.step.owner;
const run = b.addRunArtifact(comp);
run.skip_foreign_checks = true;
return run;
}
fn addZigSourceBytes(comp: *Compile, bytes: []const u8) void {
const b = comp.step.owner;
const file = WriteFile.create(b).add("a.zig", bytes);
file.addStepDependencies(&comp.step);
comp.root_src = file;
}
fn addCSourceBytes(comp: *Compile, bytes: []const u8) void {
const b = comp.step.owner;
const file = WriteFile.create(b).add("a.c", bytes);
comp.addCSourceFile(.{ .file = file, .flags = &.{} });
}
const std = @import("std");
const Build = std.Build;
const Compile = Step.Compile;
const CrossTarget = std.zig.CrossTarget;
const LazyPath = Build.LazyPath;
const Run = Step.Run;
const Step = Build.Step;
const WriteFile = Step.WriteFile;