From f1576ef14c5956cdab742aaf31ec4c672b54252b Mon Sep 17 00:00:00 2001 From: Andrew Kelley Date: Sun, 20 Jul 2025 16:43:47 -0700 Subject: [PATCH] objcopy: delete most of it this code is not up to zig project standards tracked by #24522 oh, and fix not adjusting buffer seek position in std.fs.File.Reader --- lib/compiler/objcopy.zig | 969 ++--------------------- lib/std/elf.zig | 275 +++---- lib/std/fs/File.zig | 22 +- test/standalone/stack_iterator/build.zig | 113 +-- tools/gen_stubs.zig | 3 +- 5 files changed, 240 insertions(+), 1142 deletions(-) diff --git a/lib/compiler/objcopy.zig b/lib/compiler/objcopy.zig index 52ffe208f6..5908f8b73d 100644 --- a/lib/compiler/objcopy.zig +++ b/lib/compiler/objcopy.zig @@ -13,6 +13,9 @@ const Server = std.zig.Server; var stdin_buffer: [1024]u8 = undefined; var stdout_buffer: [1024]u8 = undefined; +var input_buffer: [1024]u8 = undefined; +var output_buffer: [1024]u8 = undefined; + pub fn main() !void { var arena_instance = std.heap.ArenaAllocator.init(std.heap.page_allocator); defer arena_instance.deinit(); @@ -145,13 +148,16 @@ fn cmdObjCopy(gpa: Allocator, arena: Allocator, args: []const []const u8) !void const input = opt_input orelse fatal("expected input parameter", .{}); const output = opt_output orelse fatal("expected output parameter", .{}); - var in_file = fs.cwd().openFile(input, .{}) catch |err| - fatal("unable to open '{s}': {s}", .{ input, @errorName(err) }); - defer in_file.close(); + const input_file = fs.cwd().openFile(input, .{}) catch |err| fatal("failed to open {s}: {t}", .{ input, err }); + defer input_file.close(); - const elf_hdr = std.elf.Header.read(in_file) catch |err| switch (err) { - error.InvalidElfMagic => fatal("not an ELF file: '{s}'", .{input}), - else => fatal("unable to read '{s}': {s}", .{ input, @errorName(err) }), + const stat = input_file.stat() catch |err| fatal("failed to stat {s}: {t}", .{ input, err }); + + var in: File.Reader = .initSize(input_file, &input_buffer, stat.size); + + const elf_hdr = std.elf.Header.read(&in.interface) catch |err| switch (err) { + error.ReadFailed => fatal("unable to read {s}: {t}", .{ input, in.err.? }), + else => |e| fatal("invalid elf file: {t}", .{e}), }; const in_ofmt = .elf; @@ -168,16 +174,12 @@ fn cmdObjCopy(gpa: Allocator, arena: Allocator, args: []const []const u8) !void } }; - const mode = mode: { - if (out_fmt != .elf or only_keep_debug) - break :mode fs.File.default_mode; - if (in_file.stat()) |stat| - break :mode stat.mode - else |_| - break :mode fs.File.default_mode; - }; - var out_file = try fs.cwd().createFile(output, .{ .mode = mode }); - defer out_file.close(); + const mode = if (out_fmt != .elf or only_keep_debug) fs.File.default_mode else stat.mode; + + var output_file = try fs.cwd().createFile(output, .{ .mode = mode }); + defer output_file.close(); + + var out = output_file.writer(&output_buffer); switch (out_fmt) { .hex, .raw => { @@ -192,7 +194,7 @@ fn cmdObjCopy(gpa: Allocator, arena: Allocator, args: []const []const u8) !void if (set_section_flags != null) fatal("zig objcopy: ELF to RAW or HEX copying does not support --set_section_flags", .{}); - try emitElf(arena, in_file, out_file, elf_hdr, .{ + try emitElf(arena, &in, &out, elf_hdr, .{ .ofmt = out_fmt, .only_section = only_section, .pad_to = pad_to, @@ -208,22 +210,13 @@ fn cmdObjCopy(gpa: Allocator, arena: Allocator, args: []const []const u8) !void if (pad_to) |_| fatal("zig objcopy: ELF to ELF copying does not support --pad-to", .{}); - try stripElf(arena, in_file, out_file, elf_hdr, .{ - .strip_debug = strip_debug, - .strip_all = strip_all, - .only_keep_debug = only_keep_debug, - .add_debuglink = opt_add_debuglink, - .extract_to = opt_extract, - .compress_debug = compress_debug_sections, - .add_section = add_section, - .set_section_alignment = set_section_alignment, - .set_section_flags = set_section_flags, - }); - return std.process.cleanExit(); + fatal("unimplemented", .{}); }, else => fatal("unsupported output object format: {s}", .{@tagName(out_fmt)}), } + try out.end(); + if (listen) { var stdin_reader = fs.File.stdin().reader(&stdin_buffer); var stdout_writer = fs.File.stdout().writer(&stdout_buffer); @@ -304,12 +297,12 @@ const SetSectionFlags = struct { fn emitElf( arena: Allocator, - in_file: File, - out_file: File, + in: *File.Reader, + out: *File.Writer, elf_hdr: elf.Header, options: EmitRawElfOptions, ) !void { - var binary_elf_output = try BinaryElfOutput.parse(arena, in_file, elf_hdr); + var binary_elf_output = try BinaryElfOutput.parse(arena, in, elf_hdr); defer binary_elf_output.deinit(); if (options.ofmt == .elf) { @@ -328,8 +321,8 @@ fn emitElf( continue; } - try writeBinaryElfSection(in_file, out_file, section); - try padFile(out_file, options.pad_to); + try writeBinaryElfSection(in, out, section); + try padFile(out, options.pad_to); return; } }, @@ -342,10 +335,10 @@ fn emitElf( switch (options.ofmt) { .raw => { for (binary_elf_output.sections.items) |section| { - try out_file.seekTo(section.binaryOffset); - try writeBinaryElfSection(in_file, out_file, section); + try out.seekTo(section.binaryOffset); + try writeBinaryElfSection(in, out, section); } - try padFile(out_file, options.pad_to); + try padFile(out, options.pad_to); }, .hex => { if (binary_elf_output.segments.items.len == 0) return; @@ -353,15 +346,15 @@ fn emitElf( return error.InvalidHexfileAddressRange; } - var hex_writer = HexWriter{ .out_file = out_file }; + var hex_writer = HexWriter{ .out = out }; for (binary_elf_output.segments.items) |segment| { - try hex_writer.writeSegment(segment, in_file); + try hex_writer.writeSegment(segment, in); } if (options.pad_to) |_| { // Padding to a size in hex files isn't applicable return error.InvalidArgument; } - try hex_writer.writeEOF(); + try hex_writer.writeEof(); }, else => unreachable, } @@ -399,7 +392,7 @@ const BinaryElfOutput = struct { self.segments.deinit(self.allocator); } - pub fn parse(allocator: Allocator, elf_file: File, elf_hdr: elf.Header) !Self { + pub fn parse(allocator: Allocator, in: *File.Reader, elf_hdr: elf.Header) !Self { var self: Self = .{ .segments = .{}, .sections = .{}, @@ -412,7 +405,7 @@ const BinaryElfOutput = struct { self.shstrtab = blk: { if (elf_hdr.shstrndx >= elf_hdr.shnum) break :blk null; - var section_headers = elf_hdr.section_header_iterator(&elf_file); + var section_headers = elf_hdr.iterateSectionHeaders(in); var section_counter: usize = 0; while (section_counter < elf_hdr.shstrndx) : (section_counter += 1) { @@ -421,18 +414,13 @@ const BinaryElfOutput = struct { const shstrtab_shdr = (try section_headers.next()).?; - const buffer = try allocator.alloc(u8, @intCast(shstrtab_shdr.sh_size)); - errdefer allocator.free(buffer); - - const num_read = try elf_file.preadAll(buffer, shstrtab_shdr.sh_offset); - if (num_read != buffer.len) return error.EndOfStream; - - break :blk buffer; + try in.seekTo(shstrtab_shdr.sh_offset); + break :blk try in.interface.readAlloc(allocator, shstrtab_shdr.sh_size); }; errdefer if (self.shstrtab) |shstrtab| allocator.free(shstrtab); - var section_headers = elf_hdr.section_header_iterator(&elf_file); + var section_headers = elf_hdr.iterateSectionHeaders(in); while (try section_headers.next()) |section| { if (sectionValidForOutput(section)) { const newSection = try allocator.create(BinaryElfSection); @@ -451,7 +439,7 @@ const BinaryElfOutput = struct { } } - var program_headers = elf_hdr.program_header_iterator(&elf_file); + var program_headers = elf_hdr.iterateProgramHeaders(in); while (try program_headers.next()) |phdr| { if (phdr.p_type == elf.PT_LOAD) { const newSegment = try allocator.create(BinaryElfSegment); @@ -539,19 +527,17 @@ const BinaryElfOutput = struct { } }; -fn writeBinaryElfSection(elf_file: File, out_file: File, section: *BinaryElfSection) !void { - try out_file.writeFileAll(elf_file, .{ - .in_offset = section.elfOffset, - .in_len = section.fileSize, - }); +fn writeBinaryElfSection(in: *File.Reader, out: *File.Writer, section: *BinaryElfSection) !void { + try in.seekTo(section.elfOffset); + _ = try out.interface.sendFileAll(in, .limited(section.fileSize)); } const HexWriter = struct { prev_addr: ?u32 = null, - out_file: File, + out: *File.Writer, /// Max data bytes per line of output - const MAX_PAYLOAD_LEN: u8 = 16; + const max_payload_len: u8 = 16; fn addressParts(address: u16) [2]u8 { const msb: u8 = @truncate(address >> 8); @@ -627,13 +613,13 @@ const HexWriter = struct { return (sum ^ 0xFF) +% 1; } - fn write(self: Record, file: File) File.WriteError!void { + fn write(self: Record, out: *File.Writer) !void { const linesep = "\r\n"; // colon, (length, address, type, payload, checksum) as hex, CRLF - const BUFSIZE = 1 + (1 + 2 + 1 + MAX_PAYLOAD_LEN + 1) * 2 + linesep.len; + const BUFSIZE = 1 + (1 + 2 + 1 + max_payload_len + 1) * 2 + linesep.len; var outbuf: [BUFSIZE]u8 = undefined; const payload_bytes = self.getPayloadBytes(); - assert(payload_bytes.len <= MAX_PAYLOAD_LEN); + assert(payload_bytes.len <= max_payload_len); const line = try std.fmt.bufPrint(&outbuf, ":{0X:0>2}{1X:0>4}{2X:0>2}{3X}{4X:0>2}" ++ linesep, .{ @as(u8, @intCast(payload_bytes.len)), @@ -642,38 +628,37 @@ const HexWriter = struct { payload_bytes, self.checksum(), }); - try file.writeAll(line); + try out.interface.writeAll(line); } }; - pub fn writeSegment(self: *HexWriter, segment: *const BinaryElfSegment, elf_file: File) !void { - var buf: [MAX_PAYLOAD_LEN]u8 = undefined; + pub fn writeSegment(self: *HexWriter, segment: *const BinaryElfSegment, in: *File.Reader) !void { + var buf: [max_payload_len]u8 = undefined; var bytes_read: usize = 0; while (bytes_read < segment.fileSize) { const row_address: u32 = @intCast(segment.physicalAddress + bytes_read); const remaining = segment.fileSize - bytes_read; - const to_read: usize = @intCast(@min(remaining, MAX_PAYLOAD_LEN)); - const did_read = try elf_file.preadAll(buf[0..to_read], segment.elfOffset + bytes_read); - if (did_read < to_read) return error.UnexpectedEOF; + const dest = buf[0..@min(remaining, max_payload_len)]; + try in.seekTo(segment.elfOffset + bytes_read); + try in.interface.readSliceAll(dest); + try self.writeDataRow(row_address, dest); - try self.writeDataRow(row_address, buf[0..did_read]); - - bytes_read += did_read; + bytes_read += dest.len; } } - fn writeDataRow(self: *HexWriter, address: u32, data: []const u8) File.WriteError!void { + fn writeDataRow(self: *HexWriter, address: u32, data: []const u8) !void { const record = Record.Data(address, data); if (address > 0xFFFF and (self.prev_addr == null or record.address != self.prev_addr.?)) { - try Record.Address(address).write(self.out_file); + try Record.Address(address).write(self.out); } - try record.write(self.out_file); + try record.write(self.out); self.prev_addr = @intCast(record.address + data.len); } - fn writeEOF(self: HexWriter) File.WriteError!void { - try Record.EOF().write(self.out_file); + fn writeEof(self: HexWriter) !void { + try Record.EOF().write(self.out); } }; @@ -686,9 +671,9 @@ fn containsValidAddressRange(segments: []*BinaryElfSegment) bool { return true; } -fn padFile(f: File, opt_size: ?u64) !void { +fn padFile(out: *File.Writer, opt_size: ?u64) !void { const size = opt_size orelse return; - try f.setEndPos(size); + try out.file.setEndPos(size); } test "HexWriter.Record.Address has correct payload and checksum" { @@ -732,836 +717,6 @@ test "containsValidAddressRange" { try std.testing.expect(containsValidAddressRange(&buf)); } -// ------------- -// ELF to ELF stripping - -const StripElfOptions = struct { - extract_to: ?[]const u8 = null, - add_debuglink: ?[]const u8 = null, - strip_all: bool = false, - strip_debug: bool = false, - only_keep_debug: bool = false, - compress_debug: bool = false, - add_section: ?AddSection, - set_section_alignment: ?SetSectionAlignment, - set_section_flags: ?SetSectionFlags, -}; - -fn stripElf( - allocator: Allocator, - in_file: File, - out_file: File, - elf_hdr: elf.Header, - options: StripElfOptions, -) !void { - const Filter = ElfFileHelper.Filter; - const DebugLink = ElfFileHelper.DebugLink; - - const filter: Filter = filter: { - if (options.only_keep_debug) break :filter .debug; - if (options.strip_all) break :filter .program; - if (options.strip_debug) break :filter .program_and_symbols; - break :filter .all; - }; - - const filter_complement: ?Filter = blk: { - if (options.extract_to) |_| { - break :blk switch (filter) { - .program => .debug_and_symbols, - .debug => .program_and_symbols, - .program_and_symbols => .debug, - .debug_and_symbols => .program, - .all => fatal("zig objcopy: nothing to extract", .{}), - }; - } else { - break :blk null; - } - }; - const debuglink_path = path: { - if (options.add_debuglink) |path| break :path path; - if (options.extract_to) |path| break :path path; - break :path null; - }; - - switch (elf_hdr.is_64) { - inline else => |is_64| { - var elf_file = try ElfFile(is_64).parse(allocator, in_file, elf_hdr); - defer elf_file.deinit(); - - if (options.add_section) |user_section| { - for (elf_file.sections) |section| { - if (std.mem.eql(u8, section.name, user_section.section_name)) { - fatal("zig objcopy: unable to add section '{s}'. Section already exists in input", .{user_section.section_name}); - } - } - } - - if (filter_complement) |flt| { - // write the .dbg file and close it, so it can be read back to compute the debuglink checksum. - const path = options.extract_to.?; - const dbg_file = std.fs.cwd().createFile(path, .{}) catch |err| { - fatal("zig objcopy: unable to create '{s}': {s}", .{ path, @errorName(err) }); - }; - defer dbg_file.close(); - - try elf_file.emit(allocator, dbg_file, in_file, .{ .section_filter = flt, .compress_debug = options.compress_debug }); - } - - const debuglink: ?DebugLink = if (debuglink_path) |path| ElfFileHelper.createDebugLink(path) else null; - try elf_file.emit(allocator, out_file, in_file, .{ - .section_filter = filter, - .debuglink = debuglink, - .compress_debug = options.compress_debug, - .add_section = options.add_section, - .set_section_alignment = options.set_section_alignment, - .set_section_flags = options.set_section_flags, - }); - }, - } -} - -// note: this is "a minimal effort implementation" -// It doesn't support all possibile elf files: some sections type may need fixups, the program header may need fix up, ... -// It was written for a specific use case (strip debug info to a sperate file, for linux 64-bits executables built with `zig` or `zig c++` ) -// It moves and reoders the sections as little as possible to avoid having to do fixups. -// TODO: support non-native endianess - -fn ElfFile(comptime is_64: bool) type { - const Elf_Ehdr = if (is_64) elf.Elf64_Ehdr else elf.Elf32_Ehdr; - const Elf_Phdr = if (is_64) elf.Elf64_Phdr else elf.Elf32_Phdr; - const Elf_Shdr = if (is_64) elf.Elf64_Shdr else elf.Elf32_Shdr; - const Elf_Chdr = if (is_64) elf.Elf64_Chdr else elf.Elf32_Chdr; - const Elf_Sym = if (is_64) elf.Elf64_Sym else elf.Elf32_Sym; - const Elf_OffSize = if (is_64) elf.Elf64_Off else elf.Elf32_Off; - - return struct { - raw_elf_header: Elf_Ehdr, - program_segments: []const Elf_Phdr, - sections: []const Section, - arena: std.heap.ArenaAllocator, - - const SectionCategory = ElfFileHelper.SectionCategory; - const section_memory_align: std.mem.Alignment = .of(Elf_Sym); // most restrictive of what we may load in memory - const Section = struct { - section: Elf_Shdr, - name: []const u8 = "", - segment: ?*const Elf_Phdr = null, // if the section is used by a program segment (there can be more than one) - payload: ?[]align(section_memory_align.toByteUnits()) const u8 = null, // if we need the data in memory - category: SectionCategory = .none, // should the section be kept in the exe or stripped to the debug database, or both. - }; - - const Self = @This(); - - pub fn parse(gpa: Allocator, in_file: File, header: elf.Header) !Self { - var arena = std.heap.ArenaAllocator.init(gpa); - errdefer arena.deinit(); - const allocator = arena.allocator(); - - var raw_header: Elf_Ehdr = undefined; - { - const bytes_read = try in_file.preadAll(std.mem.asBytes(&raw_header), 0); - if (bytes_read < @sizeOf(Elf_Ehdr)) - return error.TRUNCATED_ELF; - } - - // program header: list of segments - const program_segments = blk: { - if (@sizeOf(Elf_Phdr) != header.phentsize) - fatal("zig objcopy: unsupported ELF file, unexpected phentsize ({d})", .{header.phentsize}); - - const program_header = try allocator.alloc(Elf_Phdr, header.phnum); - const bytes_read = try in_file.preadAll(std.mem.sliceAsBytes(program_header), header.phoff); - if (bytes_read < @sizeOf(Elf_Phdr) * header.phnum) - return error.TRUNCATED_ELF; - break :blk program_header; - }; - - // section header - const sections = blk: { - if (@sizeOf(Elf_Shdr) != header.shentsize) - fatal("zig objcopy: unsupported ELF file, unexpected shentsize ({d})", .{header.shentsize}); - - const section_header = try allocator.alloc(Section, header.shnum); - - const raw_section_header = try allocator.alloc(Elf_Shdr, header.shnum); - defer allocator.free(raw_section_header); - const bytes_read = try in_file.preadAll(std.mem.sliceAsBytes(raw_section_header), header.shoff); - if (bytes_read < @sizeOf(Elf_Phdr) * header.shnum) - return error.TRUNCATED_ELF; - - for (section_header, raw_section_header) |*section, hdr| { - section.* = .{ .section = hdr }; - } - break :blk section_header; - }; - - // load data to memory for some sections: - // string tables for access - // sections than need modifications when other sections move. - for (sections, 0..) |*section, idx| { - const need_data = switch (section.section.sh_type) { - elf.DT_VERSYM => true, - elf.SHT_SYMTAB, elf.SHT_DYNSYM => true, - else => false, - }; - const need_strings = (idx == header.shstrndx); - - if (need_data or need_strings) { - const buffer = try allocator.alignedAlloc(u8, section_memory_align, @intCast(section.section.sh_size)); - const bytes_read = try in_file.preadAll(buffer, section.section.sh_offset); - if (bytes_read != section.section.sh_size) return error.TRUNCATED_ELF; - section.payload = buffer; - } - } - - // fill-in sections info: - // resolve the name - // find if a program segment uses the section - // categorize sections usage (used by program segments, debug datadase, common metadata, symbol table) - for (sections) |*section| { - section.segment = for (program_segments) |*seg| { - if (sectionWithinSegment(section.section, seg.*)) break seg; - } else null; - - if (section.section.sh_name != 0 and header.shstrndx != elf.SHN_UNDEF) - section.name = std.mem.span(@as([*:0]const u8, @ptrCast(§ions[header.shstrndx].payload.?[section.section.sh_name]))); - - const category_from_program: SectionCategory = if (section.segment != null) .exe else .debug; - section.category = switch (section.section.sh_type) { - elf.SHT_NOTE => .common, - elf.SHT_SYMTAB => .symbols, // "strip all" vs "strip only debug" - elf.SHT_DYNSYM => .exe, - elf.SHT_PROGBITS => cat: { - if (std.mem.eql(u8, section.name, ".comment")) break :cat .exe; - if (std.mem.eql(u8, section.name, ".gnu_debuglink")) break :cat .none; - break :cat category_from_program; - }, - elf.SHT_LOPROC...elf.SHT_HIPROC => .common, // don't strip unknown sections - elf.SHT_LOUSER...elf.SHT_HIUSER => .common, // don't strip unknown sections - else => category_from_program, - }; - } - - sections[0].category = .common; // mandatory null section - if (header.shstrndx != elf.SHN_UNDEF) - sections[header.shstrndx].category = .common; // string table for the headers - - // recursively propagate section categories to their linked sections, so that they are kept together - var dirty: u1 = 1; - while (dirty != 0) { - dirty = 0; - - for (sections) |*section| { - if (section.section.sh_link != elf.SHN_UNDEF) - dirty |= ElfFileHelper.propagateCategory(§ions[section.section.sh_link].category, section.category); - if ((section.section.sh_flags & elf.SHF_INFO_LINK) != 0 and section.section.sh_info != elf.SHN_UNDEF) - dirty |= ElfFileHelper.propagateCategory(§ions[section.section.sh_info].category, section.category); - } - } - - return Self{ - .arena = arena, - .raw_elf_header = raw_header, - .program_segments = program_segments, - .sections = sections, - }; - } - - pub fn deinit(self: *Self) void { - self.arena.deinit(); - } - - const Filter = ElfFileHelper.Filter; - const DebugLink = ElfFileHelper.DebugLink; - const EmitElfOptions = struct { - section_filter: Filter = .all, - debuglink: ?DebugLink = null, - compress_debug: bool = false, - add_section: ?AddSection = null, - set_section_alignment: ?SetSectionAlignment = null, - set_section_flags: ?SetSectionFlags = null, - }; - fn emit(self: *const Self, gpa: Allocator, out_file: File, in_file: File, options: EmitElfOptions) !void { - var arena = std.heap.ArenaAllocator.init(gpa); - defer arena.deinit(); - const allocator = arena.allocator(); - - // when emitting the stripped exe: - // - unused sections are removed - // when emitting the debug file: - // - all sections are kept, but some are emptied and their types is changed to SHT_NOBITS - // the program header is kept unchanged. (`strip` does update it, but `eu-strip` does not, and it still works) - - const Update = struct { - action: ElfFileHelper.Action, - - // remap the indexs after omitting the filtered sections - remap_idx: u16, - - // optionally overrides the payload from the source file - payload: ?[]align(section_memory_align.toByteUnits()) const u8 = null, - section: ?Elf_Shdr = null, - }; - const sections_update = try allocator.alloc(Update, self.sections.len); - const new_shnum = blk: { - var next_idx: u16 = 0; - for (self.sections, sections_update) |section, *update| { - const action = ElfFileHelper.selectAction(section.category, options.section_filter); - const remap_idx = idx: { - if (action == .strip) break :idx elf.SHN_UNDEF; - next_idx += 1; - break :idx next_idx - 1; - }; - update.* = Update{ .action = action, .remap_idx = remap_idx }; - } - - if (options.debuglink != null) - next_idx += 1; - - if (options.add_section != null) { - next_idx += 1; - } - - break :blk next_idx; - }; - - // add a ".gnu_debuglink" to the string table if needed - const debuglink_name: u32 = blk: { - if (options.debuglink == null) break :blk elf.SHN_UNDEF; - if (self.raw_elf_header.e_shstrndx == elf.SHN_UNDEF) - fatal("zig objcopy: no strtab, cannot add the debuglink section", .{}); // TODO add the section if needed? - - const strtab = &self.sections[self.raw_elf_header.e_shstrndx]; - const update = §ions_update[self.raw_elf_header.e_shstrndx]; - - const name: []const u8 = ".gnu_debuglink"; - const new_offset: u32 = @intCast(strtab.payload.?.len); - const buf = try allocator.alignedAlloc(u8, section_memory_align, new_offset + name.len + 1); - @memcpy(buf[0..new_offset], strtab.payload.?); - @memcpy(buf[new_offset..][0..name.len], name); - buf[new_offset + name.len] = 0; - - assert(update.action == .keep); - update.payload = buf; - - break :blk new_offset; - }; - - // add user section to the string table if needed - const user_section_name: u32 = blk: { - if (options.add_section == null) break :blk elf.SHN_UNDEF; - if (self.raw_elf_header.e_shstrndx == elf.SHN_UNDEF) - fatal("zig objcopy: no strtab, cannot add the user section", .{}); // TODO add the section if needed? - - const strtab = &self.sections[self.raw_elf_header.e_shstrndx]; - const update = §ions_update[self.raw_elf_header.e_shstrndx]; - - const name = options.add_section.?.section_name; - const new_offset: u32 = @intCast(strtab.payload.?.len); - const buf = try allocator.alignedAlloc(u8, section_memory_align, new_offset + name.len + 1); - @memcpy(buf[0..new_offset], strtab.payload.?); - @memcpy(buf[new_offset..][0..name.len], name); - buf[new_offset + name.len] = 0; - - assert(update.action == .keep); - update.payload = buf; - - break :blk new_offset; - }; - - // maybe compress .debug sections - if (options.compress_debug) { - for (self.sections[1..], sections_update[1..]) |section, *update| { - if (update.action != .keep) continue; - if (!std.mem.startsWith(u8, section.name, ".debug_")) continue; - if ((section.section.sh_flags & elf.SHF_COMPRESSED) != 0) continue; // already compressed - - const chdr = Elf_Chdr{ - .ch_type = elf.COMPRESS.ZLIB, - .ch_size = section.section.sh_size, - .ch_addralign = section.section.sh_addralign, - }; - - const compressed_payload = try ElfFileHelper.tryCompressSection(allocator, in_file, section.section.sh_offset, section.section.sh_size, std.mem.asBytes(&chdr)); - if (compressed_payload) |payload| { - update.payload = payload; - update.section = section.section; - update.section.?.sh_addralign = @alignOf(Elf_Chdr); - update.section.?.sh_size = @intCast(payload.len); - update.section.?.sh_flags |= elf.SHF_COMPRESSED; - } - } - } - - var cmdbuf = std.ArrayList(ElfFileHelper.WriteCmd).init(allocator); - defer cmdbuf.deinit(); - try cmdbuf.ensureUnusedCapacity(3 + new_shnum); - var eof_offset: Elf_OffSize = 0; // track the end of the data written so far. - - // build the updated headers - // nb: updated_elf_header will be updated before the actual write - var updated_elf_header = self.raw_elf_header; - if (updated_elf_header.e_shstrndx != elf.SHN_UNDEF) - updated_elf_header.e_shstrndx = sections_update[updated_elf_header.e_shstrndx].remap_idx; - cmdbuf.appendAssumeCapacity(.{ .write_data = .{ .data = std.mem.asBytes(&updated_elf_header), .out_offset = 0 } }); - eof_offset = @sizeOf(Elf_Ehdr); - - // program header as-is. - // nb: for only-debug files, removing it appears to work, but is invalid by ELF specifcation. - { - assert(updated_elf_header.e_phoff == @sizeOf(Elf_Ehdr)); - const data = std.mem.sliceAsBytes(self.program_segments); - assert(data.len == @as(usize, updated_elf_header.e_phentsize) * updated_elf_header.e_phnum); - cmdbuf.appendAssumeCapacity(.{ .write_data = .{ .data = data, .out_offset = updated_elf_header.e_phoff } }); - eof_offset = updated_elf_header.e_phoff + @as(Elf_OffSize, @intCast(data.len)); - } - - // update sections and queue payload writes - const updated_section_header = blk: { - const dest_sections = try allocator.alloc(Elf_Shdr, new_shnum); - - { - // the ELF format doesn't specify the order for all sections. - // this code only supports when they are in increasing file order. - var offset: u64 = eof_offset; - for (self.sections[1..]) |section| { - if (section.section.sh_type == elf.SHT_NOBITS) - continue; - if (section.section.sh_offset < offset) { - fatal("zig objcopy: unsupported ELF file", .{}); - } - offset = section.section.sh_offset; - } - } - - dest_sections[0] = self.sections[0].section; - - var dest_section_idx: u32 = 1; - for (self.sections[1..], sections_update[1..]) |section, update| { - if (update.action == .strip) continue; - assert(update.remap_idx == dest_section_idx); - - const src = if (update.section) |*s| s else §ion.section; - const dest = &dest_sections[dest_section_idx]; - const payload = if (update.payload) |data| data else section.payload; - dest_section_idx += 1; - - dest.* = src.*; - - if (src.sh_link != elf.SHN_UNDEF) - dest.sh_link = sections_update[src.sh_link].remap_idx; - if ((src.sh_flags & elf.SHF_INFO_LINK) != 0 and src.sh_info != elf.SHN_UNDEF) - dest.sh_info = sections_update[src.sh_info].remap_idx; - - if (payload) |data| - dest.sh_size = @intCast(data.len); - - const addralign = if (src.sh_addralign == 0 or dest.sh_type == elf.SHT_NOBITS) 1 else src.sh_addralign; - dest.sh_offset = std.mem.alignForward(Elf_OffSize, eof_offset, addralign); - if (src.sh_offset != dest.sh_offset and section.segment != null and update.action != .empty and dest.sh_type != elf.SHT_NOTE and dest.sh_type != elf.SHT_NOBITS) { - if (src.sh_offset > dest.sh_offset) { - dest.sh_offset = src.sh_offset; // add padding to avoid modifing the program segments - } else { - fatal("zig objcopy: cannot adjust program segments", .{}); - } - } - assert(dest.sh_addr % addralign == dest.sh_offset % addralign); - - if (update.action == .empty) - dest.sh_type = elf.SHT_NOBITS; - - if (dest.sh_type != elf.SHT_NOBITS) { - if (payload) |src_data| { - // update sections payload and write - const dest_data = switch (src.sh_type) { - elf.DT_VERSYM => dst_data: { - const data = try allocator.alignedAlloc(u8, section_memory_align, src_data.len); - @memcpy(data, src_data); - - const defs = @as([*]elf.Verdef, @ptrCast(data))[0 .. @as(usize, @intCast(src.sh_size)) / @sizeOf(elf.Verdef)]; - for (defs) |*def| switch (def.ndx) { - .LOCAL, .GLOBAL => {}, - else => def.ndx = @enumFromInt(sections_update[src.sh_info].remap_idx), - }; - - break :dst_data data; - }, - elf.SHT_SYMTAB, elf.SHT_DYNSYM => dst_data: { - const data = try allocator.alignedAlloc(u8, section_memory_align, src_data.len); - @memcpy(data, src_data); - - const syms = @as([*]Elf_Sym, @ptrCast(data))[0 .. @as(usize, @intCast(src.sh_size)) / @sizeOf(Elf_Sym)]; - for (syms) |*sym| { - if (sym.st_shndx != elf.SHN_UNDEF and sym.st_shndx < elf.SHN_LORESERVE) - sym.st_shndx = sections_update[sym.st_shndx].remap_idx; - } - - break :dst_data data; - }, - else => src_data, - }; - - assert(dest_data.len == dest.sh_size); - cmdbuf.appendAssumeCapacity(.{ .write_data = .{ .data = dest_data, .out_offset = dest.sh_offset } }); - eof_offset = dest.sh_offset + dest.sh_size; - } else { - // direct contents copy - cmdbuf.appendAssumeCapacity(.{ .copy_range = .{ .in_offset = src.sh_offset, .len = dest.sh_size, .out_offset = dest.sh_offset } }); - eof_offset = dest.sh_offset + dest.sh_size; - } - } else { - // account for alignment padding even in empty sections to keep logical section order - eof_offset = dest.sh_offset; - } - } - - // add a ".gnu_debuglink" section - if (options.debuglink) |link| { - const payload = payload: { - const crc_offset = std.mem.alignForward(usize, link.name.len + 1, 4); - const buf = try allocator.alignedAlloc(u8, .@"4", crc_offset + 4); - @memcpy(buf[0..link.name.len], link.name); - @memset(buf[link.name.len..crc_offset], 0); - @memcpy(buf[crc_offset..], std.mem.asBytes(&link.crc32)); - break :payload buf; - }; - - dest_sections[dest_section_idx] = Elf_Shdr{ - .sh_name = debuglink_name, - .sh_type = elf.SHT_PROGBITS, - .sh_flags = 0, - .sh_addr = 0, - .sh_offset = eof_offset, - .sh_size = @intCast(payload.len), - .sh_link = elf.SHN_UNDEF, - .sh_info = elf.SHN_UNDEF, - .sh_addralign = 4, - .sh_entsize = 0, - }; - dest_section_idx += 1; - - cmdbuf.appendAssumeCapacity(.{ .write_data = .{ .data = payload, .out_offset = eof_offset } }); - eof_offset += @as(Elf_OffSize, @intCast(payload.len)); - } - - // --add-section - if (options.add_section) |add_section| { - var section_file = fs.cwd().openFile(add_section.file_path, .{}) catch |err| - fatal("unable to open '{s}': {s}", .{ add_section.file_path, @errorName(err) }); - defer section_file.close(); - - const payload = try section_file.readToEndAlloc(arena.allocator(), std.math.maxInt(usize)); - - dest_sections[dest_section_idx] = Elf_Shdr{ - .sh_name = user_section_name, - .sh_type = elf.SHT_PROGBITS, - .sh_flags = 0, - .sh_addr = 0, - .sh_offset = eof_offset, - .sh_size = @intCast(payload.len), - .sh_link = elf.SHN_UNDEF, - .sh_info = elf.SHN_UNDEF, - .sh_addralign = 4, - .sh_entsize = 0, - }; - dest_section_idx += 1; - - cmdbuf.appendAssumeCapacity(.{ .write_data = .{ .data = payload, .out_offset = eof_offset } }); - eof_offset += @as(Elf_OffSize, @intCast(payload.len)); - } - - assert(dest_section_idx == new_shnum); - break :blk dest_sections; - }; - - // --set-section-alignment: overwrite alignment - if (options.set_section_alignment) |set_align| { - if (self.raw_elf_header.e_shstrndx == elf.SHN_UNDEF) - fatal("zig objcopy: no strtab, cannot add the user section", .{}); // TODO add the section if needed? - - const strtab = §ions_update[self.raw_elf_header.e_shstrndx]; - for (updated_section_header) |*section| { - const section_name = std.mem.span(@as([*:0]const u8, @ptrCast(&strtab.payload.?[section.sh_name]))); - if (std.mem.eql(u8, section_name, set_align.section_name)) { - section.sh_addralign = set_align.alignment; - break; - } - } else std.log.warn("Skipping --set-section-alignment. Section '{s}' not found", .{set_align.section_name}); - } - - // --set-section-flags: overwrite flags - if (options.set_section_flags) |set_flags| { - if (self.raw_elf_header.e_shstrndx == elf.SHN_UNDEF) - fatal("zig objcopy: no strtab, cannot add the user section", .{}); // TODO add the section if needed? - - const strtab = §ions_update[self.raw_elf_header.e_shstrndx]; - for (updated_section_header) |*section| { - const section_name = std.mem.span(@as([*:0]const u8, @ptrCast(&strtab.payload.?[section.sh_name]))); - if (std.mem.eql(u8, section_name, set_flags.section_name)) { - section.sh_flags = std.elf.SHF_WRITE; // default is writable cleared by "readonly" - const f = set_flags.flags; - - // Supporting a subset of GNU and LLVM objcopy for ELF only - // GNU: - // alloc: add SHF_ALLOC - // contents: if section is SHT_NOBITS, set SHT_PROGBITS, otherwise do nothing - // load: if section is SHT_NOBITS, set SHT_PROGBITS, otherwise do nothing (same as contents) - // noload: not ELF relevant - // readonly: clear default SHF_WRITE flag - // code: add SHF_EXECINSTR - // data: not ELF relevant - // rom: ignored - // exclude: add SHF_EXCLUDE - // share: not ELF relevant - // debug: not ELF relevant - // large: add SHF_X86_64_LARGE. Fatal error if target is not x86_64 - if (f.alloc) section.sh_flags |= std.elf.SHF_ALLOC; - if (f.contents or f.load) { - if (section.sh_type == std.elf.SHT_NOBITS) section.sh_type = std.elf.SHT_PROGBITS; - } - if (f.readonly) section.sh_flags &= ~@as(@TypeOf(section.sh_type), std.elf.SHF_WRITE); - if (f.code) section.sh_flags |= std.elf.SHF_EXECINSTR; - if (f.exclude) section.sh_flags |= std.elf.SHF_EXCLUDE; - if (f.large) { - if (updated_elf_header.e_machine != std.elf.EM.X86_64) - fatal("zig objcopy: 'large' section flag is only supported on x86_64 targets", .{}); - section.sh_flags |= std.elf.SHF_X86_64_LARGE; - } - - // LLVM: - // merge: add SHF_MERGE - // strings: add SHF_STRINGS - if (f.merge) section.sh_flags |= std.elf.SHF_MERGE; - if (f.strings) section.sh_flags |= std.elf.SHF_STRINGS; - break; - } - } else std.log.warn("Skipping --set-section-flags. Section '{s}' not found", .{set_flags.section_name}); - } - - // write the section header at the tail - { - const offset = std.mem.alignForward(Elf_OffSize, eof_offset, @alignOf(Elf_Shdr)); - - const data = std.mem.sliceAsBytes(updated_section_header); - assert(data.len == @as(usize, updated_elf_header.e_shentsize) * new_shnum); - updated_elf_header.e_shoff = offset; - updated_elf_header.e_shnum = new_shnum; - - cmdbuf.appendAssumeCapacity(.{ .write_data = .{ .data = data, .out_offset = updated_elf_header.e_shoff } }); - } - - try ElfFileHelper.write(allocator, out_file, in_file, cmdbuf.items); - } - - fn sectionWithinSegment(section: Elf_Shdr, segment: Elf_Phdr) bool { - const file_size = if (section.sh_type == elf.SHT_NOBITS) 0 else section.sh_size; - return segment.p_offset <= section.sh_offset and (segment.p_offset + segment.p_filesz) >= (section.sh_offset + file_size); - } - }; -} - -const ElfFileHelper = struct { - const DebugLink = struct { name: []const u8, crc32: u32 }; - const Filter = enum { all, program, debug, program_and_symbols, debug_and_symbols }; - - const SectionCategory = enum { common, exe, debug, symbols, none }; - fn propagateCategory(cur: *SectionCategory, new: SectionCategory) u1 { - const cat: SectionCategory = switch (cur.*) { - .none => new, - .common => .common, - .debug => switch (new) { - .none, .debug => .debug, - else => new, - }, - .exe => switch (new) { - .common => .common, - .none, .debug, .exe => .exe, - .symbols => .exe, - }, - .symbols => switch (new) { - .none, .common, .debug, .exe => unreachable, - .symbols => .symbols, - }, - }; - - if (cur.* != cat) { - cur.* = cat; - return 1; - } else { - return 0; - } - } - - const Action = enum { keep, strip, empty }; - fn selectAction(category: SectionCategory, filter: Filter) Action { - if (category == .none) return .strip; - return switch (filter) { - .all => switch (category) { - .none => .strip, - else => .keep, - }, - .program => switch (category) { - .common, .exe => .keep, - else => .strip, - }, - .program_and_symbols => switch (category) { - .common, .exe, .symbols => .keep, - else => .strip, - }, - .debug => switch (category) { - .exe, .symbols => .empty, - .none => .strip, - else => .keep, - }, - .debug_and_symbols => switch (category) { - .exe => .empty, - .none => .strip, - else => .keep, - }, - }; - } - - const WriteCmd = union(enum) { - copy_range: struct { in_offset: u64, len: u64, out_offset: u64 }, - write_data: struct { data: []const u8, out_offset: u64 }, - }; - fn write(allocator: Allocator, out_file: File, in_file: File, cmds: []const WriteCmd) !void { - // consolidate holes between writes: - // by coping original padding data from in_file (by fusing contiguous ranges) - // by writing zeroes otherwise - const zeroes = [1]u8{0} ** 4096; - var consolidated = std.ArrayList(WriteCmd).init(allocator); - defer consolidated.deinit(); - try consolidated.ensureUnusedCapacity(cmds.len * 2); - var offset: u64 = 0; - var fused_cmd: ?WriteCmd = null; - for (cmds) |cmd| { - switch (cmd) { - .write_data => |data| { - assert(data.out_offset >= offset); - if (fused_cmd) |prev| { - consolidated.appendAssumeCapacity(prev); - fused_cmd = null; - } - if (data.out_offset > offset) { - consolidated.appendAssumeCapacity(.{ .write_data = .{ .data = zeroes[0..@intCast(data.out_offset - offset)], .out_offset = offset } }); - } - consolidated.appendAssumeCapacity(cmd); - offset = data.out_offset + data.data.len; - }, - .copy_range => |range| { - assert(range.out_offset >= offset); - if (fused_cmd) |prev| { - if (range.in_offset >= prev.copy_range.in_offset + prev.copy_range.len and (range.out_offset - prev.copy_range.out_offset == range.in_offset - prev.copy_range.in_offset)) { - fused_cmd = .{ .copy_range = .{ - .in_offset = prev.copy_range.in_offset, - .out_offset = prev.copy_range.out_offset, - .len = (range.out_offset + range.len) - prev.copy_range.out_offset, - } }; - } else { - consolidated.appendAssumeCapacity(prev); - if (range.out_offset > offset) { - consolidated.appendAssumeCapacity(.{ .write_data = .{ .data = zeroes[0..@intCast(range.out_offset - offset)], .out_offset = offset } }); - } - fused_cmd = cmd; - } - } else { - fused_cmd = cmd; - } - offset = range.out_offset + range.len; - }, - } - } - if (fused_cmd) |cmd| { - consolidated.appendAssumeCapacity(cmd); - } - - // write the output file - for (consolidated.items) |cmd| { - switch (cmd) { - .write_data => |data| { - var iovec = [_]std.posix.iovec_const{.{ .base = data.data.ptr, .len = data.data.len }}; - try out_file.pwritevAll(&iovec, data.out_offset); - }, - .copy_range => |range| { - const copied_bytes = try in_file.copyRangeAll(range.in_offset, out_file, range.out_offset, range.len); - if (copied_bytes < range.len) return error.TRUNCATED_ELF; - }, - } - } - } - - fn tryCompressSection(allocator: Allocator, in_file: File, offset: u64, size: u64, prefix: []const u8) !?[]align(8) const u8 { - if (size < prefix.len) return null; - - try in_file.seekTo(offset); - var section_reader = std.io.limitedReader(in_file.deprecatedReader(), size); - - // allocate as large as decompressed data. if the compression doesn't fit, keep the data uncompressed. - const compressed_data = try allocator.alignedAlloc(u8, .@"8", @intCast(size)); - var compressed_stream = std.io.fixedBufferStream(compressed_data); - - try compressed_stream.writer().writeAll(prefix); - - { - var compressor = try std.compress.zlib.compressor(compressed_stream.writer(), .{}); - - var buf: [8000]u8 = undefined; - while (true) { - const bytes_read = try section_reader.read(&buf); - if (bytes_read == 0) break; - const bytes_written = compressor.write(buf[0..bytes_read]) catch |err| switch (err) { - error.NoSpaceLeft => { - allocator.free(compressed_data); - return null; - }, - else => return err, - }; - std.debug.assert(bytes_written == bytes_read); - } - compressor.finish() catch |err| switch (err) { - error.NoSpaceLeft => { - allocator.free(compressed_data); - return null; - }, - else => return err, - }; - } - - const compressed_len: usize = @intCast(compressed_stream.getPos() catch unreachable); - const data = allocator.realloc(compressed_data, compressed_len) catch compressed_data; - return data[0..compressed_len]; - } - - fn createDebugLink(path: []const u8) DebugLink { - const file = std.fs.cwd().openFile(path, .{}) catch |err| { - fatal("zig objcopy: could not open `{s}`: {s}\n", .{ path, @errorName(err) }); - }; - defer file.close(); - - const crc = ElfFileHelper.computeFileCrc(file) catch |err| { - fatal("zig objcopy: could not read `{s}`: {s}\n", .{ path, @errorName(err) }); - }; - return .{ - .name = std.fs.path.basename(path), - .crc32 = crc, - }; - } - - fn computeFileCrc(file: File) !u32 { - var buf: [8000]u8 = undefined; - - try file.seekTo(0); - var hasher = std.hash.Crc32.init(); - while (true) { - const bytes_read = try file.read(&buf); - if (bytes_read == 0) break; - hasher.update(buf[0..bytes_read]); - } - return hasher.final(); - } -}; - const SectionFlags = packed struct { alloc: bool = false, contents: bool = false, diff --git a/lib/std/elf.zig b/lib/std/elf.zig index 4e15cd3a09..47b3add84f 100644 --- a/lib/std/elf.zig +++ b/lib/std/elf.zig @@ -482,6 +482,7 @@ pub const Header = struct { is_64: bool, endian: std.builtin.Endian, os_abi: OSABI, + /// The meaning of this value depends on `os_abi`. abi_version: u8, type: ET, machine: EM, @@ -494,205 +495,135 @@ pub const Header = struct { shnum: u16, shstrndx: u16, - pub fn program_header_iterator(self: Header, parse_source: anytype) ProgramHeaderIterator(@TypeOf(parse_source)) { - return ProgramHeaderIterator(@TypeOf(parse_source)){ - .elf_header = self, - .parse_source = parse_source, + pub fn iterateProgramHeaders(h: Header, file_reader: *std.fs.File.Reader) ProgramHeaderIterator { + return .{ + .elf_header = h, + .file_reader = file_reader, }; } - pub fn section_header_iterator(self: Header, parse_source: anytype) SectionHeaderIterator(@TypeOf(parse_source)) { - return SectionHeaderIterator(@TypeOf(parse_source)){ - .elf_header = self, - .parse_source = parse_source, + pub fn iterateSectionHeaders(h: Header, file_reader: *std.fs.File.Reader) SectionHeaderIterator { + return .{ + .elf_header = h, + .file_reader = file_reader, }; } - pub fn read(parse_source: anytype) !Header { - var hdr_buf: [@sizeOf(Elf64_Ehdr)]u8 align(@alignOf(Elf64_Ehdr)) = undefined; - try parse_source.seekableStream().seekTo(0); - try parse_source.deprecatedReader().readNoEof(&hdr_buf); - return Header.parse(&hdr_buf); - } + pub const ReadError = std.Io.Reader.Error || error{ + InvalidElfMagic, + InvalidElfVersion, + InvalidElfClass, + InvalidElfEndian, + }; - pub fn parse(hdr_buf: *align(@alignOf(Elf64_Ehdr)) const [@sizeOf(Elf64_Ehdr)]u8) !Header { - const hdr32 = @as(*const Elf32_Ehdr, @ptrCast(hdr_buf)); - const hdr64 = @as(*const Elf64_Ehdr, @ptrCast(hdr_buf)); - if (!mem.eql(u8, hdr32.e_ident[0..4], MAGIC)) return error.InvalidElfMagic; - if (hdr32.e_ident[EI_VERSION] != 1) return error.InvalidElfVersion; + pub fn read(r: *std.Io.Reader) ReadError!Header { + const buf = try r.peek(@sizeOf(Elf64_Ehdr)); - const is_64 = switch (hdr32.e_ident[EI_CLASS]) { - ELFCLASS32 => false, - ELFCLASS64 => true, - else => return error.InvalidElfClass, - }; + if (!mem.eql(u8, buf[0..4], MAGIC)) return error.InvalidElfMagic; + if (buf[EI_VERSION] != 1) return error.InvalidElfVersion; - const endian: std.builtin.Endian = switch (hdr32.e_ident[EI_DATA]) { + const endian: std.builtin.Endian = switch (buf[EI_DATA]) { ELFDATA2LSB => .little, ELFDATA2MSB => .big, else => return error.InvalidElfEndian, }; - const need_bswap = endian != native_endian; + return switch (buf[EI_CLASS]) { + ELFCLASS32 => .init(try r.takeStruct(Elf32_Ehdr, endian), endian), + ELFCLASS64 => .init(try r.takeStruct(Elf64_Ehdr, endian), endian), + else => return error.InvalidElfClass, + }; + } + + pub fn init(hdr: anytype, endian: std.builtin.Endian) Header { // Converting integers to exhaustive enums using `@enumFromInt` could cause a panic. comptime assert(!@typeInfo(OSABI).@"enum".is_exhaustive); - const os_abi: OSABI = @enumFromInt(hdr32.e_ident[EI_OSABI]); - - // The meaning of this value depends on `os_abi` so just make it available as `u8`. - const abi_version = hdr32.e_ident[EI_ABIVERSION]; - - const @"type" = if (need_bswap) blk: { - comptime assert(!@typeInfo(ET).@"enum".is_exhaustive); - const value = @intFromEnum(hdr32.e_type); - break :blk @as(ET, @enumFromInt(@byteSwap(value))); - } else hdr32.e_type; - - const machine = if (need_bswap) blk: { - comptime assert(!@typeInfo(EM).@"enum".is_exhaustive); - const value = @intFromEnum(hdr32.e_machine); - break :blk @as(EM, @enumFromInt(@byteSwap(value))); - } else hdr32.e_machine; - - return @as(Header, .{ - .is_64 = is_64, + return .{ + .is_64 = switch (@TypeOf(hdr)) { + Elf32_Ehdr => false, + Elf64_Ehdr => true, + else => @compileError("bad type"), + }, .endian = endian, - .os_abi = os_abi, - .abi_version = abi_version, - .type = @"type", - .machine = machine, - .entry = int(is_64, need_bswap, hdr32.e_entry, hdr64.e_entry), - .phoff = int(is_64, need_bswap, hdr32.e_phoff, hdr64.e_phoff), - .shoff = int(is_64, need_bswap, hdr32.e_shoff, hdr64.e_shoff), - .phentsize = int(is_64, need_bswap, hdr32.e_phentsize, hdr64.e_phentsize), - .phnum = int(is_64, need_bswap, hdr32.e_phnum, hdr64.e_phnum), - .shentsize = int(is_64, need_bswap, hdr32.e_shentsize, hdr64.e_shentsize), - .shnum = int(is_64, need_bswap, hdr32.e_shnum, hdr64.e_shnum), - .shstrndx = int(is_64, need_bswap, hdr32.e_shstrndx, hdr64.e_shstrndx), - }); + .os_abi = @enumFromInt(hdr.e_ident[EI_OSABI]), + .abi_version = hdr.e_ident[EI_ABIVERSION], + .type = hdr.e_type, + .machine = hdr.e_machine, + .entry = hdr.e_entry, + .phoff = hdr.e_phoff, + .shoff = hdr.e_shoff, + .phentsize = hdr.e_phentsize, + .phnum = hdr.e_phnum, + .shentsize = hdr.e_shentsize, + .shnum = hdr.e_shnum, + .shstrndx = hdr.e_shstrndx, + }; } }; -pub fn ProgramHeaderIterator(comptime ParseSource: anytype) type { - return struct { - elf_header: Header, - parse_source: ParseSource, - index: usize = 0, +pub const ProgramHeaderIterator = struct { + elf_header: Header, + file_reader: *std.fs.File.Reader, + index: usize = 0, - pub fn next(self: *@This()) !?Elf64_Phdr { - if (self.index >= self.elf_header.phnum) return null; - defer self.index += 1; + pub fn next(it: *ProgramHeaderIterator) !?Elf64_Phdr { + if (it.index >= it.elf_header.phnum) return null; + defer it.index += 1; - if (self.elf_header.is_64) { - var phdr: Elf64_Phdr = undefined; - const offset = self.elf_header.phoff + @sizeOf(@TypeOf(phdr)) * self.index; - try self.parse_source.seekableStream().seekTo(offset); - try self.parse_source.deprecatedReader().readNoEof(mem.asBytes(&phdr)); - - // ELF endianness matches native endianness. - if (self.elf_header.endian == native_endian) return phdr; - - // Convert fields to native endianness. - mem.byteSwapAllFields(Elf64_Phdr, &phdr); - return phdr; - } - - var phdr: Elf32_Phdr = undefined; - const offset = self.elf_header.phoff + @sizeOf(@TypeOf(phdr)) * self.index; - try self.parse_source.seekableStream().seekTo(offset); - try self.parse_source.deprecatedReader().readNoEof(mem.asBytes(&phdr)); - - // ELF endianness does NOT match native endianness. - if (self.elf_header.endian != native_endian) { - // Convert fields to native endianness. - mem.byteSwapAllFields(Elf32_Phdr, &phdr); - } - - // Convert 32-bit header to 64-bit. - return Elf64_Phdr{ - .p_type = phdr.p_type, - .p_offset = phdr.p_offset, - .p_vaddr = phdr.p_vaddr, - .p_paddr = phdr.p_paddr, - .p_filesz = phdr.p_filesz, - .p_memsz = phdr.p_memsz, - .p_flags = phdr.p_flags, - .p_align = phdr.p_align, - }; + if (it.elf_header.is_64) { + const offset = it.elf_header.phoff + @sizeOf(Elf64_Phdr) * it.index; + try it.file_reader.seekTo(offset); + const phdr = try it.file_reader.interface.takeStruct(Elf64_Phdr, it.elf_header.endian); + return phdr; } - }; -} -pub fn SectionHeaderIterator(comptime ParseSource: anytype) type { - return struct { - elf_header: Header, - parse_source: ParseSource, - index: usize = 0, - - pub fn next(self: *@This()) !?Elf64_Shdr { - if (self.index >= self.elf_header.shnum) return null; - defer self.index += 1; - - if (self.elf_header.is_64) { - var shdr: Elf64_Shdr = undefined; - const offset = self.elf_header.shoff + @sizeOf(@TypeOf(shdr)) * self.index; - try self.parse_source.seekableStream().seekTo(offset); - try self.parse_source.deprecatedReader().readNoEof(mem.asBytes(&shdr)); - - // ELF endianness matches native endianness. - if (self.elf_header.endian == native_endian) return shdr; - - // Convert fields to native endianness. - mem.byteSwapAllFields(Elf64_Shdr, &shdr); - return shdr; - } - - var shdr: Elf32_Shdr = undefined; - const offset = self.elf_header.shoff + @sizeOf(@TypeOf(shdr)) * self.index; - try self.parse_source.seekableStream().seekTo(offset); - try self.parse_source.deprecatedReader().readNoEof(mem.asBytes(&shdr)); - - // ELF endianness does NOT match native endianness. - if (self.elf_header.endian != native_endian) { - // Convert fields to native endianness. - mem.byteSwapAllFields(Elf32_Shdr, &shdr); - } - - // Convert 32-bit header to 64-bit. - return Elf64_Shdr{ - .sh_name = shdr.sh_name, - .sh_type = shdr.sh_type, - .sh_flags = shdr.sh_flags, - .sh_addr = shdr.sh_addr, - .sh_offset = shdr.sh_offset, - .sh_size = shdr.sh_size, - .sh_link = shdr.sh_link, - .sh_info = shdr.sh_info, - .sh_addralign = shdr.sh_addralign, - .sh_entsize = shdr.sh_entsize, - }; - } - }; -} - -fn int(is_64: bool, need_bswap: bool, int_32: anytype, int_64: anytype) @TypeOf(int_64) { - if (is_64) { - if (need_bswap) { - return @byteSwap(int_64); - } else { - return int_64; - } - } else { - return int32(need_bswap, int_32, @TypeOf(int_64)); + const offset = it.elf_header.phoff + @sizeOf(Elf32_Phdr) * it.index; + try it.file_reader.seekTo(offset); + const phdr = try it.file_reader.interface.takeStruct(Elf32_Phdr, it.elf_header.endian); + return .{ + .p_type = phdr.p_type, + .p_offset = phdr.p_offset, + .p_vaddr = phdr.p_vaddr, + .p_paddr = phdr.p_paddr, + .p_filesz = phdr.p_filesz, + .p_memsz = phdr.p_memsz, + .p_flags = phdr.p_flags, + .p_align = phdr.p_align, + }; } -} +}; -fn int32(need_bswap: bool, int_32: anytype, comptime Int64: anytype) Int64 { - if (need_bswap) { - return @byteSwap(int_32); - } else { - return int_32; +pub const SectionHeaderIterator = struct { + elf_header: Header, + file_reader: *std.fs.File.Reader, + index: usize = 0, + + pub fn next(it: *SectionHeaderIterator) !?Elf64_Shdr { + if (it.index >= it.elf_header.shnum) return null; + defer it.index += 1; + + if (it.elf_header.is_64) { + try it.file_reader.seekTo(it.elf_header.shoff + @sizeOf(Elf64_Shdr) * it.index); + const shdr = try it.file_reader.interface.takeStruct(Elf64_Shdr, it.elf_header.endian); + return shdr; + } + + try it.file_reader.seekTo(it.elf_header.shoff + @sizeOf(Elf32_Shdr) * it.index); + const shdr = try it.file_reader.interface.takeStruct(Elf32_Shdr, it.elf_header.endian); + return .{ + .sh_name = shdr.sh_name, + .sh_type = shdr.sh_type, + .sh_flags = shdr.sh_flags, + .sh_addr = shdr.sh_addr, + .sh_offset = shdr.sh_offset, + .sh_size = shdr.sh_size, + .sh_link = shdr.sh_link, + .sh_info = shdr.sh_info, + .sh_addralign = shdr.sh_addralign, + .sh_entsize = shdr.sh_entsize, + }; } -} +}; pub const ELFCLASSNONE = 0; pub const ELFCLASS32 = 1; diff --git a/lib/std/fs/File.zig b/lib/std/fs/File.zig index 50f2a30876..85264dd4db 100644 --- a/lib/std/fs/File.zig +++ b/lib/std/fs/File.zig @@ -1228,14 +1228,12 @@ pub const Reader = struct { pub fn seekBy(r: *Reader, offset: i64) Reader.SeekError!void { switch (r.mode) { .positional, .positional_reading => { - // TODO: make += operator allow any integer types - r.pos = @intCast(@as(i64, @intCast(r.pos)) + offset); + setPosAdjustingBuffer(r, @intCast(@as(i64, @intCast(r.pos)) + offset)); }, .streaming, .streaming_reading => { const seek_err = r.seek_err orelse e: { if (posix.lseek_CUR(r.file.handle, offset)) |_| { - // TODO: make += operator allow any integer types - r.pos = @intCast(@as(i64, @intCast(r.pos)) + offset); + setPosAdjustingBuffer(r, @intCast(@as(i64, @intCast(r.pos)) + offset)); return; } else |err| { r.seek_err = err; @@ -1251,6 +1249,8 @@ pub const Reader = struct { r.pos += n; remaining -= n; } + r.interface.seek = 0; + r.interface.end = 0; }, .failure => return r.seek_err.?, } @@ -1259,7 +1259,7 @@ pub const Reader = struct { pub fn seekTo(r: *Reader, offset: u64) Reader.SeekError!void { switch (r.mode) { .positional, .positional_reading => { - r.pos = offset; + setPosAdjustingBuffer(r, offset); }, .streaming, .streaming_reading => { if (offset >= r.pos) return Reader.seekBy(r, @intCast(offset - r.pos)); @@ -1268,12 +1268,22 @@ pub const Reader = struct { r.seek_err = err; return err; }; - r.pos = offset; + setPosAdjustingBuffer(r, offset); }, .failure => return r.seek_err.?, } } + fn setPosAdjustingBuffer(r: *Reader, offset: u64) void { + if (offset < r.pos or offset >= r.pos + r.interface.bufferedLen()) { + r.interface.seek = 0; + r.interface.end = 0; + } else { + r.interface.seek += @intCast(offset - r.pos); + } + r.pos = offset; + } + /// Number of slices to store on the stack, when trying to send as many byte /// vectors through the underlying read calls as possible. const max_buffers_len = 16; diff --git a/test/standalone/stack_iterator/build.zig b/test/standalone/stack_iterator/build.zig index a036a64ab7..878859312b 100644 --- a/test/standalone/stack_iterator/build.zig +++ b/test/standalone/stack_iterator/build.zig @@ -65,69 +65,70 @@ pub fn build(b: *std.Build) void { test_step.dependOn(&run_cmd.step); } - // Unwinding through a C shared library without a frame pointer (libc) - // - // getcontext version: libc - // - // Unwind info type: - // - ELF: DWARF .eh_frame + .debug_frame - // - MachO: __unwind_info encodings: - // - x86_64: STACK_IMMD, STACK_IND - // - aarch64: FRAMELESS, DWARF - { - const c_shared_lib = b.addLibrary(.{ - .linkage = .dynamic, - .name = "c_shared_lib", - .root_module = b.createModule(.{ - .root_source_file = null, - .target = target, - .optimize = optimize, - .link_libc = true, - .strip = false, - }), - }); + // https://github.com/ziglang/zig/issues/24522 + //// Unwinding through a C shared library without a frame pointer (libc) + //// + //// getcontext version: libc + //// + //// Unwind info type: + //// - ELF: DWARF .eh_frame + .debug_frame + //// - MachO: __unwind_info encodings: + //// - x86_64: STACK_IMMD, STACK_IND + //// - aarch64: FRAMELESS, DWARF + //{ + // const c_shared_lib = b.addLibrary(.{ + // .linkage = .dynamic, + // .name = "c_shared_lib", + // .root_module = b.createModule(.{ + // .root_source_file = null, + // .target = target, + // .optimize = optimize, + // .link_libc = true, + // .strip = false, + // }), + // }); - if (target.result.os.tag == .windows) - c_shared_lib.root_module.addCMacro("LIB_API", "__declspec(dllexport)"); + // if (target.result.os.tag == .windows) + // c_shared_lib.root_module.addCMacro("LIB_API", "__declspec(dllexport)"); - c_shared_lib.root_module.addCSourceFile(.{ - .file = b.path("shared_lib.c"), - .flags = &.{"-fomit-frame-pointer"}, - }); + // c_shared_lib.root_module.addCSourceFile(.{ + // .file = b.path("shared_lib.c"), + // .flags = &.{"-fomit-frame-pointer"}, + // }); - const exe = b.addExecutable(.{ - .name = "shared_lib_unwind", - .root_module = b.createModule(.{ - .root_source_file = b.path("shared_lib_unwind.zig"), - .target = target, - .optimize = optimize, - .unwind_tables = if (target.result.os.tag.isDarwin()) .async else null, - .omit_frame_pointer = true, - }), - // zig objcopy doesn't support incremental binaries - .use_llvm = true, - }); + // const exe = b.addExecutable(.{ + // .name = "shared_lib_unwind", + // .root_module = b.createModule(.{ + // .root_source_file = b.path("shared_lib_unwind.zig"), + // .target = target, + // .optimize = optimize, + // .unwind_tables = if (target.result.os.tag.isDarwin()) .async else null, + // .omit_frame_pointer = true, + // }), + // // zig objcopy doesn't support incremental binaries + // .use_llvm = true, + // }); - exe.linkLibrary(c_shared_lib); + // exe.linkLibrary(c_shared_lib); - const run_cmd = b.addRunArtifact(exe); - test_step.dependOn(&run_cmd.step); + // const run_cmd = b.addRunArtifact(exe); + // test_step.dependOn(&run_cmd.step); - // Separate debug info ELF file - if (target.result.ofmt == .elf) { - const filename = b.fmt("{s}_stripped", .{exe.out_filename}); - const stripped_exe = b.addObjCopy(exe.getEmittedBin(), .{ - .basename = filename, // set the name for the debuglink - .compress_debug = true, - .strip = .debug, - .extract_to_separate_file = true, - }); + // // Separate debug info ELF file + // if (target.result.ofmt == .elf) { + // const filename = b.fmt("{s}_stripped", .{exe.out_filename}); + // const stripped_exe = b.addObjCopy(exe.getEmittedBin(), .{ + // .basename = filename, // set the name for the debuglink + // .compress_debug = true, + // .strip = .debug, + // .extract_to_separate_file = true, + // }); - const run_stripped = std.Build.Step.Run.create(b, b.fmt("run {s}", .{filename})); - run_stripped.addFileArg(stripped_exe.getOutput()); - test_step.dependOn(&run_stripped.step); - } - } + // const run_stripped = std.Build.Step.Run.create(b, b.fmt("run {s}", .{filename})); + // run_stripped.addFileArg(stripped_exe.getOutput()); + // test_step.dependOn(&run_stripped.step); + // } + //} // Unwinding without libc/posix // diff --git a/tools/gen_stubs.zig b/tools/gen_stubs.zig index ed60d7e67d..9fdd63eda1 100644 --- a/tools/gen_stubs.zig +++ b/tools/gen_stubs.zig @@ -310,7 +310,8 @@ pub fn main() !void { build_all_path, libc_so_path, @errorName(err), }); }; - const header = try elf.Header.parse(elf_bytes[0..@sizeOf(elf.Elf64_Ehdr)]); + var stream: std.Io.Reader = .fixed(elf_bytes); + const header = try elf.Header.read(&stream); const parse: Parse = .{ .arena = arena,