diff options
| author | Jakub Konka <kubkon@jakubkonka.com> | 2023-02-06 13:23:03 +0100 |
|---|---|---|
| committer | Jakub Konka <kubkon@jakubkonka.com> | 2023-02-06 13:23:08 +0100 |
| commit | b32f5ee93283f7794c611ebd4a1fbc579b78d8ab (patch) | |
| tree | 06d60c8a6011711279aa4ee971c7da68f42c9ca8 | |
| parent | a5b34a61ab61882bf55d87e4cbc8186215ecf320 (diff) | |
| download | zig-b32f5ee93283f7794c611ebd4a1fbc579b78d8ab.tar.gz zig-b32f5ee93283f7794c611ebd4a1fbc579b78d8ab.zip | |
macho: downgrade alignment requirements for symtab in object files
Parse and sort relocations by address descending.
| -rw-r--r-- | src/link/MachO/Object.zig | 120 | ||||
| -rw-r--r-- | src/link/MachO/UnwindInfo.zig | 8 | ||||
| -rw-r--r-- | src/link/MachO/ZldAtom.zig | 31 | ||||
| -rw-r--r-- | src/link/MachO/dead_strip.zig | 7 | ||||
| -rw-r--r-- | src/link/MachO/eh_frame.zig | 11 | ||||
| -rw-r--r-- | src/link/MachO/zld.zig | 2 |
6 files changed, 104 insertions, 75 deletions
diff --git a/src/link/MachO/Object.zig b/src/link/MachO/Object.zig index 13219084b6..771af9b8b3 100644 --- a/src/link/MachO/Object.zig +++ b/src/link/MachO/Object.zig @@ -60,14 +60,20 @@ globals_lookup: []i64 = undefined, /// Can be undefined as set together with in_symtab. relocs_lookup: []RelocEntry = undefined, +/// All relocations sorted and flatened. +relocations: std.ArrayListUnmanaged(macho.relocation_info) = .{}, +/// Beginning index to the relocations array for each input section +/// defined within this Object file. +section_relocs_lookup: std.ArrayListUnmanaged(u32) = .{}, + atoms: std.ArrayListUnmanaged(AtomIndex) = .{}, exec_atoms: std.ArrayListUnmanaged(AtomIndex) = .{}, -eh_frame_sect: ?macho.section_64 = null, +eh_frame_sect_id: ?u8 = null, eh_frame_relocs_lookup: std.AutoArrayHashMapUnmanaged(u32, Record) = .{}, eh_frame_records_lookup: std.AutoArrayHashMapUnmanaged(AtomIndex, u32) = .{}, -unwind_info_sect: ?macho.section_64 = null, +unwind_info_sect_id: ?u8 = null, unwind_relocs_lookup: []Record = undefined, unwind_records_lookup: std.AutoHashMapUnmanaged(AtomIndex, u32) = .{}, @@ -100,6 +106,8 @@ pub fn deinit(self: *Object, gpa: Allocator) void { gpa.free(self.unwind_relocs_lookup); } self.unwind_records_lookup.deinit(gpa); + self.relocations.deinit(gpa); + self.section_relocs_lookup.deinit(gpa); } pub fn parse(self: *Object, allocator: Allocator, cpu_arch: std.Target.Cpu.Arch) !void { @@ -137,15 +145,18 @@ pub fn parse(self: *Object, allocator: Allocator, cpu_arch: std.Target.Cpu.Arch) .buffer = self.contents[@sizeOf(macho.mach_header_64)..][0..self.header.sizeofcmds], }; const nsects = self.getSourceSections().len; + + // Prepopulate relocations per section lookup table. + try self.section_relocs_lookup.resize(allocator, nsects); + mem.set(u32, self.section_relocs_lookup.items, 0); + + // Parse symtab. const symtab = while (it.next()) |cmd| switch (cmd.cmd()) { .SYMTAB => break cmd.cast(macho.symtab_command).?, else => {}, } else return; - self.in_symtab = @ptrCast( - [*]const macho.nlist_64, - @alignCast(@alignOf(macho.nlist_64), &self.contents[symtab.symoff]), - )[0..symtab.nsyms]; + self.in_symtab = @ptrCast([*]align(1) const macho.nlist_64, self.contents.ptr + symtab.symoff)[0..symtab.nsyms]; self.in_strtab = self.contents[symtab.stroff..][0..symtab.strsize]; self.symtab = try allocator.alloc(macho.nlist_64, self.in_symtab.?.len + nsects); @@ -212,10 +223,10 @@ pub fn parse(self: *Object, allocator: Allocator, cpu_arch: std.Target.Cpu.Arch) } // Parse __TEXT,__eh_frame header if one exists - self.eh_frame_sect = self.getSourceSectionByName("__TEXT", "__eh_frame"); + self.eh_frame_sect_id = self.getSourceSectionIndexByName("__TEXT", "__eh_frame"); // Parse __LD,__compact_unwind header if one exists - self.unwind_info_sect = self.getSourceSectionByName("__LD", "__compact_unwind"); + self.unwind_info_sect_id = self.getSourceSectionIndexByName("__LD", "__compact_unwind"); if (self.hasUnwindRecords()) { self.unwind_relocs_lookup = try allocator.alloc(Record, self.getUnwindRecords().len); mem.set(Record, self.unwind_relocs_lookup, .{ @@ -452,6 +463,8 @@ pub fn splitRegularSections(self: *Object, zld: *Zld, object_id: u32) !void { zld.sections.items(.header)[out_sect_id].sectName(), }); + try self.parseRelocs(gpa, section.id); + const cpu_arch = zld.options.target.cpu.arch; const sect_loc = filterSymbolsBySection(symtab[sect_sym_index..], sect_id + 1); const sect_start_index = sect_sym_index + sect_loc.index; @@ -623,25 +636,36 @@ fn filterRelocs( return .{ .start = @intCast(u32, start), .len = @intCast(u32, len) }; } +/// Parse all relocs for the input section, and sort in descending order. +/// Previously, I have wrongly assumed the compilers output relocations for each +/// section in a sorted manner which is simply not true. +fn parseRelocs(self: *Object, gpa: Allocator, sect_id: u8) !void { + const section = self.getSourceSection(sect_id); + const start = @intCast(u32, self.relocations.items.len); + if (self.getSourceRelocs(section)) |relocs| { + try self.relocations.ensureUnusedCapacity(gpa, relocs.len); + self.relocations.appendUnalignedSliceAssumeCapacity(relocs); + std.sort.sort(macho.relocation_info, self.relocations.items[start..], {}, relocGreaterThan); + } + self.section_relocs_lookup.items[sect_id] = start; +} + fn cacheRelocs(self: *Object, zld: *Zld, atom_index: AtomIndex) !void { const atom = zld.getAtom(atom_index); - const source_sect = if (self.getSourceSymbol(atom.sym_index)) |source_sym| blk: { - const source_sect = self.getSourceSection(source_sym.n_sect - 1); - assert(!source_sect.isZerofill()); - break :blk source_sect; + const source_sect_id = if (self.getSourceSymbol(atom.sym_index)) |source_sym| blk: { + break :blk source_sym.n_sect - 1; } else blk: { // If there was no matching symbol present in the source symtab, this means // we are dealing with either an entire section, or part of it, but also // starting at the beginning. const nbase = @intCast(u32, self.in_symtab.?.len); - const sect_id = @intCast(u16, atom.sym_index - nbase); - const source_sect = self.getSourceSection(sect_id); - assert(!source_sect.isZerofill()); - break :blk source_sect; + const sect_id = @intCast(u8, atom.sym_index - nbase); + break :blk sect_id; }; - - const relocs = self.getRelocs(source_sect); + const source_sect = self.getSourceSection(source_sect_id); + assert(!source_sect.isZerofill()); + const relocs = self.getRelocs(source_sect_id); self.relocs_lookup[atom.sym_index] = if (self.getSourceSymbol(atom.sym_index)) |source_sym| blk: { const offset = source_sym.n_value - source_sect.addr; @@ -649,8 +673,14 @@ fn cacheRelocs(self: *Object, zld: *Zld, atom_index: AtomIndex) !void { } else filterRelocs(relocs, 0, atom.size); } +fn relocGreaterThan(ctx: void, lhs: macho.relocation_info, rhs: macho.relocation_info) bool { + _ = ctx; + return lhs.r_address > rhs.r_address; +} + fn parseEhFrameSection(self: *Object, zld: *Zld, object_id: u32) !void { - const sect = self.eh_frame_sect orelse return; + const sect_id = self.eh_frame_sect_id orelse return; + const sect = self.getSourceSection(sect_id); log.debug("parsing __TEXT,__eh_frame section", .{}); @@ -660,7 +690,8 @@ fn parseEhFrameSection(self: *Object, zld: *Zld, object_id: u32) !void { const gpa = zld.gpa; const cpu_arch = zld.options.target.cpu.arch; - const relocs = self.getRelocs(sect); + try self.parseRelocs(gpa, sect_id); + const relocs = self.getRelocs(sect_id); var it = self.getEhFrameRecordsIterator(); var record_count: u32 = 0; @@ -728,12 +759,12 @@ fn parseEhFrameSection(self: *Object, zld: *Zld, object_id: u32) !void { } fn parseUnwindInfo(self: *Object, zld: *Zld, object_id: u32) !void { - const sect = self.unwind_info_sect orelse { + const sect_id = self.unwind_info_sect_id orelse { // If it so happens that the object had `__eh_frame` section defined but no `__compact_unwind`, // we will try fully synthesising unwind info records to somewhat match Apple ld's // approach. However, we will only synthesise DWARF records and nothing more. For this reason, // we still create the output `__TEXT,__unwind_info` section. - if (self.eh_frame_sect != null) { + if (self.hasEhFrameRecords()) { if (zld.getSectionByName("__TEXT", "__unwind_info") == null) { _ = try zld.initSection("__TEXT", "__unwind_info", .{}); } @@ -758,15 +789,15 @@ fn parseUnwindInfo(self: *Object, zld: *Zld, object_id: u32) !void { if (UnwindInfo.UnwindEncoding.isDwarf(record.compactUnwindEncoding, cpu_arch)) break true; } else false; - if (needs_eh_frame) { - if (self.eh_frame_sect == null) { - log.err("missing __TEXT,__eh_frame section", .{}); - log.err(" in object {s}", .{self.name}); - return error.MissingSection; - } + if (needs_eh_frame and !self.hasEhFrameRecords()) { + log.err("missing __TEXT,__eh_frame section", .{}); + log.err(" in object {s}", .{self.name}); + return error.MissingSection; } - const relocs = self.getRelocs(sect); + try self.parseRelocs(gpa, sect_id); + const relocs = self.getRelocs(sect_id); + for (unwind_records) |record, record_id| { const offset = record_id * @sizeOf(macho.compact_unwind_entry); const rel_pos = filterRelocs( @@ -806,25 +837,23 @@ pub fn getSourceSymbol(self: Object, index: u32) ?macho.nlist_64 { return symtab[mapped_index]; } -pub fn getSourceSection(self: Object, index: u16) macho.section_64 { +pub fn getSourceSection(self: Object, index: u8) macho.section_64 { const sections = self.getSourceSections(); assert(index < sections.len); return sections[index]; } pub fn getSourceSectionByName(self: Object, segname: []const u8, sectname: []const u8) ?macho.section_64 { + const index = self.getSourceSectionIndexByName(segname, sectname) orelse return null; const sections = self.getSourceSections(); - for (sections) |sect| { - if (mem.eql(u8, segname, sect.segName()) and mem.eql(u8, sectname, sect.sectName())) - return sect; - } else return null; + return sections[index]; } pub fn getSourceSectionIndexByName(self: Object, segname: []const u8, sectname: []const u8) ?u8 { const sections = self.getSourceSections(); for (sections) |sect, i| { if (mem.eql(u8, segname, sect.segName()) and mem.eql(u8, sectname, sect.sectName())) - return @intCast(u8, i + 1); + return @intCast(u8, i); } else return null; } @@ -914,11 +943,18 @@ pub fn getSectionAliasSymbolPtr(self: *Object, sect_id: u8) *macho.nlist_64 { return &self.symtab[self.getSectionAliasSymbolIndex(sect_id)]; } -pub fn getRelocs(self: Object, sect: macho.section_64) []align(1) const macho.relocation_info { - if (sect.nreloc == 0) return &[0]macho.relocation_info{}; +fn getSourceRelocs(self: Object, sect: macho.section_64) ?[]align(1) const macho.relocation_info { + if (sect.nreloc == 0) return null; return @ptrCast([*]align(1) const macho.relocation_info, self.contents.ptr + sect.reloff)[0..sect.nreloc]; } +pub fn getRelocs(self: Object, sect_id: u8) []const macho.relocation_info { + const sect = self.getSourceSection(sect_id); + const start = self.section_relocs_lookup.items[sect_id]; + const len = sect.nreloc; + return self.relocations.items[start..][0..len]; +} + pub fn getSymbolName(self: Object, index: u32) []const u8 { const strtab = self.in_strtab.?; const sym = self.symtab[index]; @@ -976,22 +1012,24 @@ pub fn getAtomIndexForSymbol(self: Object, sym_index: u32) ?AtomIndex { } pub fn hasUnwindRecords(self: Object) bool { - return self.unwind_info_sect != null; + return self.unwind_info_sect_id != null; } pub fn getUnwindRecords(self: Object) []align(1) const macho.compact_unwind_entry { - const sect = self.unwind_info_sect orelse return &[0]macho.compact_unwind_entry{}; + const sect_id = self.unwind_info_sect_id orelse return &[0]macho.compact_unwind_entry{}; + const sect = self.getSourceSection(sect_id); const data = self.getSectionContents(sect); const num_entries = @divExact(data.len, @sizeOf(macho.compact_unwind_entry)); return @ptrCast([*]align(1) const macho.compact_unwind_entry, data)[0..num_entries]; } pub fn hasEhFrameRecords(self: Object) bool { - return self.eh_frame_sect != null; + return self.eh_frame_sect_id != null; } pub fn getEhFrameRecordsIterator(self: Object) eh_frame.Iterator { - const sect = self.eh_frame_sect orelse return .{ .data = &[0]u8{} }; + const sect_id = self.eh_frame_sect_id orelse return .{ .data = &[0]u8{} }; + const sect = self.getSourceSection(sect_id); const data = self.getSectionContents(sect); return .{ .data = data }; } diff --git a/src/link/MachO/UnwindInfo.zig b/src/link/MachO/UnwindInfo.zig index 07091af08a..25bd667610 100644 --- a/src/link/MachO/UnwindInfo.zig +++ b/src/link/MachO/UnwindInfo.zig @@ -703,15 +703,11 @@ pub fn parseRelocTarget( } else return sym_loc; } -fn getRelocs( - zld: *Zld, - object_id: u32, - record_id: usize, -) []align(1) const macho.relocation_info { +fn getRelocs(zld: *Zld, object_id: u32, record_id: usize) []const macho.relocation_info { const object = &zld.objects.items[object_id]; assert(object.hasUnwindRecords()); const rel_pos = object.unwind_relocs_lookup[record_id].reloc; - const relocs = object.getRelocs(object.unwind_info_sect.?); + const relocs = object.getRelocs(object.unwind_info_sect_id.?); return relocs[rel_pos.start..][0..rel_pos.len]; } diff --git a/src/link/MachO/ZldAtom.zig b/src/link/MachO/ZldAtom.zig index bb8b7cffb3..e3d5f62a12 100644 --- a/src/link/MachO/ZldAtom.zig +++ b/src/link/MachO/ZldAtom.zig @@ -143,7 +143,7 @@ pub fn calcInnerSymbolOffset(zld: *Zld, atom_index: AtomIndex, sym_index: u32) u sym.n_value else blk: { const nbase = @intCast(u32, object.in_symtab.?.len); - const sect_id = @intCast(u16, atom.sym_index - nbase); + const sect_id = @intCast(u8, atom.sym_index - nbase); const source_sect = object.getSourceSection(sect_id); break :blk source_sect.addr; }; @@ -180,7 +180,7 @@ pub fn getRelocContext(zld: *Zld, atom_index: AtomIndex) RelocContext { }; } const nbase = @intCast(u32, object.in_symtab.?.len); - const sect_id = @intCast(u16, atom.sym_index - nbase); + const sect_id = @intCast(u8, atom.sym_index - nbase); const source_sect = object.getSourceSection(sect_id); return .{ .base_addr = source_sect.addr, @@ -724,7 +724,7 @@ fn resolveRelocsArm64( if (rel.r_extern == 0) { const base_addr = if (target.sym_index > object.source_address_lookup.len) - @intCast(i64, object.getSourceSection(@intCast(u16, rel.r_symbolnum - 1)).addr) + @intCast(i64, object.getSourceSection(@intCast(u8, rel.r_symbolnum - 1)).addr) else object.source_address_lookup[target.sym_index]; ptr_addend -= base_addr; @@ -861,7 +861,7 @@ fn resolveRelocsX86( if (rel.r_extern == 0) { const base_addr = if (target.sym_index > object.source_address_lookup.len) - @intCast(i64, object.getSourceSection(@intCast(u16, rel.r_symbolnum - 1)).addr) + @intCast(i64, object.getSourceSection(@intCast(u8, rel.r_symbolnum - 1)).addr) else object.source_address_lookup[target.sym_index]; addend += @intCast(i32, @intCast(i64, context.base_addr) + rel.r_address + 4 - @@ -884,7 +884,7 @@ fn resolveRelocsX86( if (rel.r_extern == 0) { const base_addr = if (target.sym_index > object.source_address_lookup.len) - @intCast(i64, object.getSourceSection(@intCast(u16, rel.r_symbolnum - 1)).addr) + @intCast(i64, object.getSourceSection(@intCast(u8, rel.r_symbolnum - 1)).addr) else object.source_address_lookup[target.sym_index]; addend -= base_addr; @@ -928,7 +928,7 @@ pub fn getAtomCode(zld: *Zld, atom_index: AtomIndex) []const u8 { // we are dealing with either an entire section, or part of it, but also // starting at the beginning. const nbase = @intCast(u32, object.in_symtab.?.len); - const sect_id = @intCast(u16, atom.sym_index - nbase); + const sect_id = @intCast(u8, atom.sym_index - nbase); const source_sect = object.getSourceSection(sect_id); assert(!source_sect.isZerofill()); const code = object.getSectionContents(source_sect); @@ -943,28 +943,25 @@ pub fn getAtomCode(zld: *Zld, atom_index: AtomIndex) []const u8 { return code[offset..][0..code_len]; } -pub fn getAtomRelocs(zld: *Zld, atom_index: AtomIndex) []align(1) const macho.relocation_info { +pub fn getAtomRelocs(zld: *Zld, atom_index: AtomIndex) []const macho.relocation_info { const atom = zld.getAtom(atom_index); assert(atom.getFile() != null); // Synthetic atom shouldn't need to unique for relocs. const object = zld.objects.items[atom.getFile().?]; const cache = object.relocs_lookup[atom.sym_index]; - const source_sect = if (object.getSourceSymbol(atom.sym_index)) |source_sym| blk: { - const source_sect = object.getSourceSection(source_sym.n_sect - 1); - assert(!source_sect.isZerofill()); - break :blk source_sect; + const source_sect_id = if (object.getSourceSymbol(atom.sym_index)) |source_sym| blk: { + break :blk source_sym.n_sect - 1; } else blk: { // If there was no matching symbol present in the source symtab, this means // we are dealing with either an entire section, or part of it, but also // starting at the beginning. const nbase = @intCast(u32, object.in_symtab.?.len); - const sect_id = @intCast(u16, atom.sym_index - nbase); - const source_sect = object.getSourceSection(sect_id); - assert(!source_sect.isZerofill()); - break :blk source_sect; + const sect_id = @intCast(u8, atom.sym_index - nbase); + break :blk sect_id; }; - - const relocs = object.getRelocs(source_sect); + const source_sect = object.getSourceSection(source_sect_id); + assert(!source_sect.isZerofill()); + const relocs = object.getRelocs(source_sect_id); return relocs[cache.start..][0..cache.len]; } diff --git a/src/link/MachO/dead_strip.zig b/src/link/MachO/dead_strip.zig index 1f8def96f3..6ba70acbfd 100644 --- a/src/link/MachO/dead_strip.zig +++ b/src/link/MachO/dead_strip.zig @@ -88,7 +88,7 @@ fn collectRoots(zld: *Zld, roots: *AtomTable) !void { source_sym.n_sect - 1 else sect_id: { const nbase = @intCast(u32, object.in_symtab.?.len); - const sect_id = @intCast(u16, atom.sym_index - nbase); + const sect_id = @intCast(u8, atom.sym_index - nbase); break :sect_id sect_id; }; const source_sect = object.getSourceSection(sect_id); @@ -223,7 +223,7 @@ fn mark(zld: *Zld, roots: AtomTable, alive: *AtomTable) !void { source_sym.n_sect - 1 else blk: { const nbase = @intCast(u32, object.in_symtab.?.len); - const sect_id = @intCast(u16, atom.sym_index - nbase); + const sect_id = @intCast(u8, atom.sym_index - nbase); break :blk sect_id; }; const source_sect = object.getSourceSection(sect_id); @@ -350,8 +350,9 @@ fn markEhFrameRecord(zld: *Zld, object_id: u32, atom_index: AtomIndex, alive: *A } }, .x86_64 => { + const sect = object.getSourceSection(object.eh_frame_sect_id.?); const lsda_ptr = try fde.getLsdaPointer(cie, .{ - .base_addr = object.eh_frame_sect.?.addr, + .base_addr = sect.addr, .base_offset = fde_offset, }); if (lsda_ptr) |lsda_address| { diff --git a/src/link/MachO/eh_frame.zig b/src/link/MachO/eh_frame.zig index 8345a8e4f3..3867b15a96 100644 --- a/src/link/MachO/eh_frame.zig +++ b/src/link/MachO/eh_frame.zig @@ -171,8 +171,9 @@ pub fn write(zld: *Zld, unwind_info: *UnwindInfo) !void { const cie_record = eh_records.get( eh_frame_offset + 4 - fde_record.getCiePointer(), ).?; + const eh_frame_sect = object.getSourceSection(object.eh_frame_sect_id.?); const source_lsda_ptr = try fde_record.getLsdaPointer(cie_record, .{ - .base_addr = object.eh_frame_sect.?.addr, + .base_addr = eh_frame_sect.addr, .base_offset = fde_record_offset, }); if (source_lsda_ptr) |ptr| { @@ -552,16 +553,12 @@ pub fn EhFrameRecord(comptime is_mutable: bool) type { }; } -pub fn getRelocs( - zld: *Zld, - object_id: u32, - source_offset: u32, -) []align(1) const macho.relocation_info { +pub fn getRelocs(zld: *Zld, object_id: u32, source_offset: u32) []const macho.relocation_info { const object = &zld.objects.items[object_id]; assert(object.hasEhFrameRecords()); const urel = object.eh_frame_relocs_lookup.get(source_offset) orelse return &[0]macho.relocation_info{}; - const all_relocs = object.getRelocs(object.eh_frame_sect.?); + const all_relocs = object.getRelocs(object.eh_frame_sect_id.?); return all_relocs[urel.reloc.start..][0..urel.reloc.len]; } diff --git a/src/link/MachO/zld.zig b/src/link/MachO/zld.zig index 81fae399ef..8619163ef2 100644 --- a/src/link/MachO/zld.zig +++ b/src/link/MachO/zld.zig @@ -2406,7 +2406,7 @@ pub const Zld = struct { source_sym.n_value else blk: { const nbase = @intCast(u32, object.in_symtab.?.len); - const source_sect_id = @intCast(u16, atom.sym_index - nbase); + const source_sect_id = @intCast(u8, atom.sym_index - nbase); break :blk object.getSourceSection(source_sect_id).addr; }; const filtered_dice = filterDataInCode(dice, source_addr, source_addr + atom.size); |
