Commit 10aaf2983d
src/link/MachO/uuid.zig
@@ -0,0 +1,46 @@
+const std = @import("std");
+const fs = std.fs;
+const mem = std.mem;
+
+const Allocator = mem.Allocator;
+const Compilation = @import("../../Compilation.zig");
+const Md5 = std.crypto.hash.Md5;
+const Hasher = @import("hasher.zig").ParallelHasher;
+
+/// Somewhat random chunk size for MD5 hash calculation.
+pub const chunk_size = 0x4000;
+
+/// Calculates Md5 hash of each chunk in parallel and then hashes all Md5 hashes to produce
+/// the final digest.
+/// While this is NOT a correct MD5 hash of the contents, this methodology is used by LLVM/LLD
+/// and we will use it too as it seems accepted by Apple OSes.
+/// TODO LLD also hashes the output filename to disambiguate between same builds with different
+/// output files. Should we also do that?
+pub fn calcUuid(comp: *const Compilation, file: fs.File, file_size: u64, out: *[Md5.digest_length]u8) !void {
+ const total_hashes = mem.alignForward(u64, file_size, chunk_size) / chunk_size;
+
+ const hashes = try comp.gpa.alloc([Md5.digest_length]u8, total_hashes);
+ defer comp.gpa.free(hashes);
+
+ var hasher = Hasher(Md5){};
+ try hasher.hash(comp.gpa, comp.thread_pool, file, hashes, .{
+ .chunk_size = chunk_size,
+ .max_file_size = file_size,
+ });
+
+ const final_buffer = try comp.gpa.alloc(u8, total_hashes * Md5.digest_length);
+ defer comp.gpa.free(final_buffer);
+
+ for (hashes, 0..) |hash, i| {
+ mem.copy(u8, final_buffer[i * Md5.digest_length ..][0..Md5.digest_length], &hash);
+ }
+
+ Md5.hash(final_buffer, out, .{});
+ conform(out);
+}
+
+inline fn conform(out: *[Md5.digest_length]u8) void {
+ // LC_UUID uuids should conform to RFC 4122 UUID version 4 & UUID version 5 formats
+ out[6] = (out[6] & 0x0F) | (3 << 4);
+ out[8] = (out[8] & 0x3F) | 0x80;
+}
src/link/MachO/zld.zig
@@ -9,14 +9,15 @@ const math = std.math;
const mem = std.mem;
const aarch64 = @import("../../arch/aarch64/bits.zig");
+const calcUuid = @import("uuid.zig").calcUuid;
const dead_strip = @import("dead_strip.zig");
const eh_frame = @import("eh_frame.zig");
const fat = @import("fat.zig");
const link = @import("../../link.zig");
const load_commands = @import("load_commands.zig");
+const stub_helpers = @import("stubs.zig");
const thunks = @import("thunks.zig");
const trace = @import("../../tracy.zig").trace;
-const stub_helpers = @import("stubs.zig");
const Allocator = mem.Allocator;
const Archive = @import("Archive.zig");
@@ -2575,150 +2576,12 @@ pub const Zld = struct {
self.dysymtab_cmd.nindirectsyms = nindirectsyms;
}
- fn writeUuid(self: *Zld, comp: *const Compilation, args: struct {
- linkedit_cmd_offset: u32,
- symtab_cmd_offset: u32,
- uuid_cmd_offset: u32,
- codesig_cmd_offset: ?u32,
- }) !void {
- _ = comp;
- switch (self.options.optimize_mode) {
- .Debug => {
- // In Debug we don't really care about reproducibility, so put in a random value
- // and be done with it.
- std.crypto.random.bytes(&self.uuid_cmd.uuid);
- Md5.hash(&self.uuid_cmd.uuid, &self.uuid_cmd.uuid, .{});
- conformUuid(&self.uuid_cmd.uuid);
- },
- else => {
- // We set the max file size to the actual strtab buffer length to exclude any strtab padding.
- const max_file_end = @intCast(u32, self.symtab_cmd.stroff + self.strtab.buffer.items.len);
-
- const FileSubsection = struct {
- start: u32,
- end: u32,
- };
-
- var subsections: [5]FileSubsection = undefined;
- var count: usize = 0;
-
- // Exclude LINKEDIT segment command as it contains file size that includes stabs contribution
- // and code signature.
- subsections[count] = .{
- .start = 0,
- .end = args.linkedit_cmd_offset,
- };
- count += 1;
-
- // Exclude SYMTAB and DYSYMTAB commands for the same reason.
- subsections[count] = .{
- .start = subsections[count - 1].end + @sizeOf(macho.segment_command_64),
- .end = args.symtab_cmd_offset,
- };
- count += 1;
-
- // Exclude CODE_SIGNATURE command (if present).
- if (args.codesig_cmd_offset) |offset| {
- subsections[count] = .{
- .start = subsections[count - 1].end + @sizeOf(macho.symtab_command) + @sizeOf(macho.dysymtab_command),
- .end = offset,
- };
- count += 1;
- }
-
- if (!self.options.strip) {
- // Exclude region comprising all symbol stabs.
- const nlocals = self.dysymtab_cmd.nlocalsym;
-
- const locals = try self.gpa.alloc(macho.nlist_64, nlocals);
- defer self.gpa.free(locals);
-
- const locals_buf = @ptrCast([*]u8, locals.ptr)[0 .. @sizeOf(macho.nlist_64) * nlocals];
- const amt = try self.file.preadAll(locals_buf, self.symtab_cmd.symoff);
- if (amt != locals_buf.len) return error.InputOutput;
-
- const istab: usize = for (locals, 0..) |local, i| {
- if (local.stab()) break i;
- } else locals.len;
- const nstabs = locals.len - istab;
-
- if (nstabs == 0) {
- subsections[count] = .{
- .start = subsections[count - 1].end + if (args.codesig_cmd_offset == null)
- @as(u32, @sizeOf(macho.symtab_command) + @sizeOf(macho.dysymtab_command))
- else
- @sizeOf(macho.linkedit_data_command),
- .end = max_file_end,
- };
- count += 1;
- } else {
- // Exclude a subsection of the strtab with names of the stabs.
- // We do not care about anything succeeding strtab as it is the code signature data which is
- // not part of the UUID calculation anyway.
- const stab_stroff = locals[istab].n_strx;
-
- subsections[count] = .{
- .start = subsections[count - 1].end + if (args.codesig_cmd_offset == null)
- @as(u32, @sizeOf(macho.symtab_command) + @sizeOf(macho.dysymtab_command))
- else
- @sizeOf(macho.linkedit_data_command),
- .end = @intCast(u32, self.symtab_cmd.symoff + istab * @sizeOf(macho.nlist_64)),
- };
- count += 1;
-
- subsections[count] = .{
- .start = subsections[count - 1].end + @intCast(u32, nstabs * @sizeOf(macho.nlist_64)),
- .end = self.symtab_cmd.stroff + stab_stroff,
- };
- count += 1;
- }
- } else {
- subsections[count] = .{
- .start = subsections[count - 1].end + if (args.codesig_cmd_offset == null)
- @as(u32, @sizeOf(macho.symtab_command) + @sizeOf(macho.dysymtab_command))
- else
- @sizeOf(macho.linkedit_data_command),
- .end = max_file_end,
- };
- count += 1;
- }
-
- const chunk_size = 0x4000;
-
- var hasher = Md5.init(.{});
- var buffer: [chunk_size]u8 = undefined;
-
- for (subsections[0..count]) |cut| {
- const size = cut.end - cut.start;
- const num_chunks = mem.alignForward(usize, size, chunk_size) / chunk_size;
-
- var i: usize = 0;
- while (i < num_chunks) : (i += 1) {
- const fstart = cut.start + i * chunk_size;
- const fsize = if (fstart + chunk_size > cut.end)
- cut.end - fstart
- else
- chunk_size;
- const amt = try self.file.preadAll(buffer[0..fsize], fstart);
- if (amt != fsize) return error.InputOutput;
-
- hasher.update(buffer[0..fsize]);
- }
- }
-
- hasher.final(&self.uuid_cmd.uuid);
- conformUuid(&self.uuid_cmd.uuid);
- },
- }
-
- const in_file = args.uuid_cmd_offset + @sizeOf(macho.load_command);
- try self.file.pwriteAll(&self.uuid_cmd.uuid, in_file);
- }
-
- inline fn conformUuid(out: *[Md5.digest_length]u8) void {
- // LC_UUID uuids should conform to RFC 4122 UUID version 4 & UUID version 5 formats
- out[6] = (out[6] & 0x0F) | (3 << 4);
- out[8] = (out[8] & 0x3F) | 0x80;
+ fn writeUuid(self: *Zld, comp: *const Compilation, uuid_cmd_offset: u32) !void {
+ const seg = self.getLinkeditSegmentPtr();
+ const file_size = seg.fileoff + seg.filesize;
+ try calcUuid(comp, self.file, file_size, &self.uuid_cmd.uuid);
+ const offset = uuid_cmd_offset + @sizeOf(macho.load_command);
+ try self.file.pwriteAll(&self.uuid_cmd.uuid, offset);
}
fn writeCodeSignaturePadding(self: *Zld, code_sig: *CodeSignature) !void {
@@ -4041,16 +3904,11 @@ pub fn linkWithZld(macho_file: *MachO, comp: *Compilation, prog_node: *std.Progr
const lc_writer = lc_buffer.writer();
try zld.writeSegmentHeaders(lc_writer);
- const linkedit_cmd_offset = @sizeOf(macho.mach_header_64) + @intCast(u32, lc_buffer.items.len - @sizeOf(macho.segment_command_64));
-
try lc_writer.writeStruct(zld.dyld_info_cmd);
try lc_writer.writeStruct(zld.function_starts_cmd);
try lc_writer.writeStruct(zld.data_in_code_cmd);
-
- const symtab_cmd_offset = @sizeOf(macho.mach_header_64) + @intCast(u32, lc_buffer.items.len);
try lc_writer.writeStruct(zld.symtab_cmd);
try lc_writer.writeStruct(zld.dysymtab_cmd);
-
try load_commands.writeDylinkerLC(lc_writer);
if (zld.options.output_mode == .Exe) {
@@ -4088,22 +3946,14 @@ pub fn linkWithZld(macho_file: *MachO, comp: *Compilation, prog_node: *std.Progr
try load_commands.writeLoadDylibLCs(zld.dylibs.items, zld.referenced_dylibs.keys(), lc_writer);
- var codesig_cmd_offset: ?u32 = null;
if (requires_codesig) {
- codesig_cmd_offset = @sizeOf(macho.mach_header_64) + @intCast(u32, lc_buffer.items.len);
try lc_writer.writeStruct(zld.codesig_cmd);
}
const ncmds = load_commands.calcNumOfLCs(lc_buffer.items);
try zld.file.pwriteAll(lc_buffer.items, @sizeOf(macho.mach_header_64));
try zld.writeHeader(ncmds, @intCast(u32, lc_buffer.items.len));
-
- try zld.writeUuid(comp, .{
- .linkedit_cmd_offset = linkedit_cmd_offset,
- .symtab_cmd_offset = symtab_cmd_offset,
- .uuid_cmd_offset = uuid_cmd_offset,
- .codesig_cmd_offset = codesig_cmd_offset,
- });
+ try zld.writeUuid(comp, uuid_cmd_offset);
if (codesig) |*csig| {
try zld.writeCodeSignature(comp, csig); // code signing always comes last
src/link/MachO.zig
@@ -13,6 +13,7 @@ const mem = std.mem;
const meta = std.meta;
const aarch64 = @import("../arch/aarch64/bits.zig");
+const calcUuid = @import("MachO/uuid.zig").calcUuid;
const codegen = @import("../codegen.zig");
const dead_strip = @import("MachO/dead_strip.zig");
const fat = @import("MachO/fat.zig");
@@ -756,11 +757,7 @@ pub fn flushModule(self: *MachO, comp: *Compilation, prog_node: *std.Progress.No
});
try load_commands.writeBuildVersionLC(&self.base.options, lc_writer);
- if (self.cold_start) {
- std.crypto.random.bytes(&self.uuid_cmd.uuid);
- Md5.hash(&self.uuid_cmd.uuid, &self.uuid_cmd.uuid, .{});
- conformUuid(&self.uuid_cmd.uuid);
- }
+ const uuid_cmd_offset = @sizeOf(macho.mach_header_64) + @intCast(u32, lc_buffer.items.len);
try lc_writer.writeStruct(self.uuid_cmd);
try load_commands.writeLoadDylibLCs(self.dylibs.items, self.referenced_dylibs.keys(), lc_writer);
@@ -769,10 +766,10 @@ pub fn flushModule(self: *MachO, comp: *Compilation, prog_node: *std.Progress.No
try lc_writer.writeStruct(self.codesig_cmd);
}
- try self.base.file.?.pwriteAll(lc_buffer.items, @sizeOf(macho.mach_header_64));
-
const ncmds = load_commands.calcNumOfLCs(lc_buffer.items);
+ try self.base.file.?.pwriteAll(lc_buffer.items, @sizeOf(macho.mach_header_64));
try self.writeHeader(ncmds, @intCast(u32, lc_buffer.items.len));
+ try self.writeUuid(comp, uuid_cmd_offset);
if (codesig) |*csig| {
try self.writeCodeSignature(comp, csig); // code signing always comes last
@@ -3510,6 +3507,14 @@ fn writeDysymtab(self: *MachO, ctx: SymtabCtx) !void {
self.dysymtab_cmd.nindirectsyms = nindirectsyms;
}
+fn writeUuid(self: *MachO, comp: *const Compilation, uuid_cmd_offset: u32) !void {
+ const seg = self.getLinkeditSegmentPtr();
+ const file_size = seg.fileoff + seg.filesize;
+ try calcUuid(comp, self.base.file.?, file_size, &self.uuid_cmd.uuid);
+ const offset = uuid_cmd_offset + @sizeOf(macho.load_command);
+ try self.base.file.?.pwriteAll(&self.uuid_cmd.uuid, offset);
+}
+
fn writeCodeSignaturePadding(self: *MachO, code_sig: *CodeSignature) !void {
const seg = self.getLinkeditSegmentPtr();
// Code signature data has to be 16-bytes aligned for Apple tools to recognize the file