Support bss sections

lib/caotral/binary/elf/section.rb

@@ -22,13 +22,18 @@ def build
         end
 
         def layout!(offset:)
-          size = build.bytesize
+          size = file_size
           align = [@header.addralign.to_i, 1].max
           offset = (offset + align - 1) / align * align
-          @header.set!(offset:, size:)
-
-          offset + size
+          if header.nobits?
+            header.set!(offset:)
+          else
+            header.set!(offset:, size:)
+          end
+          offset + file_size
         end
+        def file_size = header.nobits? ? 0 : build.bytesize
+        def memory_size = header.nobits? ? header.size : build.bytesize
       end
     end
   end

lib/caotral/binary/elf/section_header.rb

@@ -56,6 +56,7 @@ def addralign = @addralign.pack("C*").unpack1("Q<")
         def allocated? = (@flags.pack("C*").unpack1("Q<") & SHF[:ALLOC]) != 0
         def writable? = (@flags.pack("C*").unpack1("Q<") & SHF[:WRITE]) != 0
         def execinstr? = (@flags.pack("C*").unpack1("Q<") & SHF[:EXECINSTR]) != 0
+        def nobits? = type == :nobits
 
         private def bytes = [@name, @type, @flags, @addr, @offset, @size, @link, @info, @addralign, @entsize]
       end

lib/caotral/linker/builder.rb

@@ -88,6 +88,12 @@ def build
           section_name: ".strtab",
           header: Caotral::Binary::ELF::SectionHeader.new
         )
+        bss_section = Caotral::Binary::ELF::Section.new(
+          body: nil,
+          section_name: ".bss",
+          header: Caotral::Binary::ELF::SectionHeader.new
+                    .set!(type: SHT[:nobits], flags: SHF[:ALLOC] | SHF[:WRITE], addralign: 8)
+        )
         symtab_section = Caotral::Binary::ELF::Section.new(
           body: [],
           section_name: ".symtab",
@@ -119,14 +125,19 @@ def build
         rel_sections = []
         rel_texts = []
         pending_relative_relocations = []
+        dynamic_symbol_pairs = []
         elf.header = elf_obj.header.dup
         strtab_names = []
         text_offsets = {}
         rodata_offsets = {}
+        bss_offsets = {}
         data_offsets = {}
         got_plt_offsets = {}
         text_offset = 0
         rodata_offset = 0
+        bss_offset = 0
+        bss_present = false
+        bss_addralign = 1
         data_offset = 0
         got_plt_offset = 24
         sym_by_elf = Hash.new { |h, k| h[k] = [] }
@@ -154,13 +165,23 @@ def build
             data_offsets[elf_obj.object_id] = data_offset
             data_offset += data.body.bytesize
           end
+          bss = elf_obj.find_by_name(".bss")
+          unless bss.nil?
+            bss_present = true
+            alignment = [bss.header.addralign, 1].max
+            bss_offset = align_up(bss_offset, alignment)
+            bss_addralign = [bss_addralign, alignment].compact.max
+            bss_offsets[elf_obj.object_id] = bss_offset
+            bss_offset += bss.memory_size
+          end
           strtab = elf_obj.find_by_name(".strtab")
           strtab.body.names.split("\0").each { |name| strtab_names << name } unless strtab.nil?
           symtab = elf_obj.find_by_name(".symtab")
           base_index = nil
           unless symtab.nil?
             base_index = symtab_section.body.size
             text_index = elf_obj.sections.index(text) unless text.nil?
+            bss_index = elf_obj.sections.index(bss) unless bss.nil?
             data_index = elf_obj.sections.index(data) unless data.nil?
             rodata_index = elf_obj.sections.index(rodata) unless rodata.nil?
 
@@ -169,6 +190,7 @@ def build
               name, info, other, shndx, value, size = st.build.unpack("L<CCS<Q<Q<")
               sym_by_elf[elf_obj] << sym
               value += text_offsets.fetch(elf_obj.object_id, 0) if shndx == text_index
+              value += bss_offsets.fetch(elf_obj.object_id, 0) if shndx == bss_index
               value += data_offsets.fetch(elf_obj.object_id, 0) if shndx == data_index
               value += rodata_offsets.fetch(elf_obj.object_id, 0) if shndx == rodata_index
               sym.set!(name:, info:, other:, shndx:, value:, size:)
@@ -278,7 +300,6 @@ def build
           sections << plt_section
           sections << got_plt_section
         end
-        sections << strtab_section
         text_index = sections.index(text_section)
         symtab_section.body.each do |sym|
           next if sym.shndx == 0
@@ -297,12 +318,10 @@ def build
         symtab_section.header.set!(
           type: 2,
           flags: 0,
-          link: elf.sections.index(strtab_section),
           info: local_count,
           addralign: 8,
           entsize: 24
         )
-
         sections += build_pie_sections if @pie
         if dynamic?
           @needed.each { |lib| dynstr.body.names += lib + "\0" if dynstr.body.offset_of(lib).nil? }
@@ -326,6 +345,7 @@ def build
               name = dynstr.body.offset_of(copy_sym.name_string)
             end
             copy_sym.name_string = sym.name_string
+            dynamic_symbol_pairs << [copy_sym, sym]
             dynsym.body << copy_sym.set!(name:, shndx:, value: sym.value)
           end
           hash = Caotral::Binary::ELF::Section::Hash.new(nchain: dynsym.body.size)
@@ -344,7 +364,9 @@ def build
 
           sections << dynamic_section
         end
+        sections << bss_section if bss_present
         sections << symtab_section
+        sections << strtab_section
 
         rel_sections.each { |s| sections << s.dup }
 
@@ -357,6 +379,8 @@ def build
 
         sections << shstrtab_section
 
+        bss_section.header.set!(size: bss_offset, addralign: bss_addralign)
+
         shstrtab_section_names = [*sections.map(&:section_name), "\0"].join("\0")
         shstrtab_section.body.names = shstrtab_section_names
 
@@ -376,6 +400,8 @@ def build
           end
         end
 
+        dynamic_symbol_pairs.each { |copy_sym, sym| copy_sym.set!(shndx: sym.shndx) }
+
         defined_global_index = {}
         symtab_section.body.each_with_index do |sym, index|
           next unless sym.bind == SYMTAB_BIND[:globals]
@@ -535,6 +561,7 @@ def ref_index(sections, section_name)
 
       def rel_type(section) = section.section_name&.start_with?(".rela.") ? 4 : 9
       def rel_entsize(section) = section.section_name&.start_with?(".rela.") ? 24 : 16
+      def align_up(val, align) = (val + align - 1) / align * align
 
       def dynamic? = @shared || @pie
     end

lib/caotral/linker/layout.rb

@@ -118,15 +118,23 @@ def finalize_program_headers!
           when :LOAD
             allocated_sections = @load_sections[ph.flags]
 
-            section_end = allocated_sections.map { |s| s.header.offset + s.header.size }.max
+            memory_end = allocated_sections.map { |s| s.header.offset + s.memory_size }.max
+            file_end = allocated_sections
+                         .reject { |s| s.header.nobits? }
+                         .map { |s| s.header.offset + s.file_size }.max
             segment_start = ph.flags == :R ? 0 : allocated_sections.map { |s| s.header.offset }.min
-            segment_end = ph.flags == :R ? [header_end, section_end].compact.max : section_end
+            if ph.flags == :R
+              memory_end = [header_end, memory_end].compact.max
+              file_end = [header_end, file_end].compact.max
+            end
+            memory_end ||= segment_start
+            file_end ||= segment_start
 
             offset = segment_start
             vaddr = @load_bias + offset
             paddr = vaddr
-            filesz = segment_end - segment_start
-            memsz = filesz
+            filesz = file_end - segment_start
+            memsz = memory_end - segment_start
             flags = PF[ph.flags]
             align = LOAD_ALIGNMENT
           when :INTERP

sample/assembler/bss.s

@@ -0,0 +1,16 @@
+.section .bss
+.balign 8
+.globl value
+.type value, @object
+.size value, 8
+value:
+  .zero 8
+
+.text
+.globl _start
+.type _start, @function
+_start:
+  movq $42, value(%rip)
+  movq value(%rip), %rdi
+  movq $60, %rax
+  syscall

sample/assembler/bss_initialize_zero.s

@@ -0,0 +1,27 @@
+.section .bss
+.balign 8
+.zero 24
+
+.globl value
+.type value, @object
+.size value, 8
+value:
+  .zero 8
+
+.text
+.globl _start
+.type _start, @function
+_start:
+  cmpq $0, value(%rip)
+  jne .Lnot_zero
+
+  movq $42, value(%rip)
+  movq value(%rip), %rdi
+  jmp .Lexit
+
+.Lnot_zero:
+  movq $1, %rdi
+
+.Lexit:
+  movq $60, %rax
+  syscall

sample/assembler/empty_bss.s

@@ -0,0 +1,15 @@
+.section .bss
+.balign 8
+.globl value
+.type value, @object
+.size value, 0
+value:
+
+.text
+.globl _start
+.type _start, @function
+_start:
+  movq $42, value(%rip)
+  movq value(%rip), %rdi
+  movq $60, %rax
+  syscall

sample/assembler/two_bss_sections.s

@@ -0,0 +1,7 @@
+.section .bss
+.balign 16
+.globl aligned_value
+.type aligned_value, @object
+.size aligned_value, 16
+aligned_value:
+  .zero 16

sig/caotral/binary/elf/section.rbs

@@ -8,4 +8,7 @@ class Caotral::Binary::ELF::Section
     body: untyped,
     section_name: String?
   ) -> void
+
+  def file_size: () -> Integer
+  def memory_size: () -> Integer
 end

sig/caotral/binary/elf/section_header.rbs

@@ -26,4 +26,5 @@ class Caotral::Binary::ELF::SectionHeader
   def addr: () -> Integer
   def writable?: () -> bool
   def execinstr?: () -> bool
+  def nobits?: () -> bool
 end